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;
121 * Try to predict block placement assuming that we are
122 * filling in a file which will eventually be
123 * non-sparse --- i.e., in the case of libbfd writing
124 * an ELF object sections out-of-order but in a way
125 * the eventually results in a contiguous object or
126 * executable file, or some database extending a table
127 * space file. However, this is actually somewhat
128 * non-ideal if we are writing a sparse file such as
129 * qemu or KVM writing a raw image file that is going
130 * to stay fairly sparse, since it will end up
131 * fragmenting the file system's free space. Maybe we
132 * should have some hueristics or some way to allow
133 * userspace to pass a hint to file system,
134 * especiially if the latter case turns out to be
137 ex = path[depth].p_ext;
139 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
140 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
142 if (block > ext_block)
143 return ext_pblk + (block - ext_block);
145 return ext_pblk - (ext_block - block);
148 /* it looks like index is empty;
149 * try to find starting block from index itself */
150 if (path[depth].p_bh)
151 return path[depth].p_bh->b_blocknr;
154 /* OK. use inode's group */
155 block_group = ei->i_block_group;
156 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
158 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
159 * block groups per flexgroup, reserve the first block
160 * group for directories and special files. Regular
161 * files will start at the second block group. This
162 * tends to speed up directory access and improves
165 block_group &= ~(flex_size-1);
166 if (S_ISREG(inode->i_mode))
169 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
170 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
173 * If we are doing delayed allocation, we don't need take
174 * colour into account.
176 if (test_opt(inode->i_sb, DELALLOC))
179 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
180 colour = (current->pid % 16) *
181 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
183 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
184 return bg_start + colour + block;
188 * Allocation for a meta data block
191 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
192 struct ext4_ext_path *path,
193 struct ext4_extent *ex, int *err)
195 ext4_fsblk_t goal, newblock;
197 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
198 newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
202 static inline int ext4_ext_space_block(struct inode *inode, int check)
206 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
207 / sizeof(struct ext4_extent);
209 #ifdef AGGRESSIVE_TEST
217 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
221 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
222 / sizeof(struct ext4_extent_idx);
224 #ifdef AGGRESSIVE_TEST
232 static inline int ext4_ext_space_root(struct inode *inode, int check)
236 size = sizeof(EXT4_I(inode)->i_data);
237 size -= sizeof(struct ext4_extent_header);
238 size /= sizeof(struct ext4_extent);
240 #ifdef AGGRESSIVE_TEST
248 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
252 size = sizeof(EXT4_I(inode)->i_data);
253 size -= sizeof(struct ext4_extent_header);
254 size /= sizeof(struct ext4_extent_idx);
256 #ifdef AGGRESSIVE_TEST
265 * Calculate the number of metadata blocks needed
266 * to allocate @blocks
267 * Worse case is one block per extent
269 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
271 struct ext4_inode_info *ei = EXT4_I(inode);
274 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
275 / sizeof(struct ext4_extent_idx));
278 * If the new delayed allocation block is contiguous with the
279 * previous da block, it can share index blocks with the
280 * previous block, so we only need to allocate a new index
281 * block every idxs leaf blocks. At ldxs**2 blocks, we need
282 * an additional index block, and at ldxs**3 blocks, yet
283 * another index blocks.
285 if (ei->i_da_metadata_calc_len &&
286 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
287 if ((ei->i_da_metadata_calc_len % idxs) == 0)
289 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
291 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
293 ei->i_da_metadata_calc_len = 0;
295 ei->i_da_metadata_calc_len++;
296 ei->i_da_metadata_calc_last_lblock++;
301 * In the worst case we need a new set of index blocks at
302 * every level of the inode's extent tree.
304 ei->i_da_metadata_calc_len = 1;
305 ei->i_da_metadata_calc_last_lblock = lblock;
306 return ext_depth(inode) + 1;
310 ext4_ext_max_entries(struct inode *inode, int depth)
314 if (depth == ext_depth(inode)) {
316 max = ext4_ext_space_root(inode, 1);
318 max = ext4_ext_space_root_idx(inode, 1);
321 max = ext4_ext_space_block(inode, 1);
323 max = ext4_ext_space_block_idx(inode, 1);
329 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
331 ext4_fsblk_t block = ext4_ext_pblock(ext);
332 int len = ext4_ext_get_actual_len(ext);
334 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
337 static int ext4_valid_extent_idx(struct inode *inode,
338 struct ext4_extent_idx *ext_idx)
340 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
342 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
345 static int ext4_valid_extent_entries(struct inode *inode,
346 struct ext4_extent_header *eh,
349 struct ext4_extent *ext;
350 struct ext4_extent_idx *ext_idx;
351 unsigned short entries;
352 if (eh->eh_entries == 0)
355 entries = le16_to_cpu(eh->eh_entries);
359 ext = EXT_FIRST_EXTENT(eh);
361 if (!ext4_valid_extent(inode, ext))
367 ext_idx = EXT_FIRST_INDEX(eh);
369 if (!ext4_valid_extent_idx(inode, ext_idx))
378 static int __ext4_ext_check(const char *function, unsigned int line,
379 struct inode *inode, struct ext4_extent_header *eh,
382 const char *error_msg;
385 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
386 error_msg = "invalid magic";
389 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
390 error_msg = "unexpected eh_depth";
393 if (unlikely(eh->eh_max == 0)) {
394 error_msg = "invalid eh_max";
397 max = ext4_ext_max_entries(inode, depth);
398 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
399 error_msg = "too large eh_max";
402 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
403 error_msg = "invalid eh_entries";
406 if (!ext4_valid_extent_entries(inode, eh, depth)) {
407 error_msg = "invalid extent entries";
413 ext4_error_inode(inode, function, line, 0,
414 "bad header/extent: %s - magic %x, "
415 "entries %u, max %u(%u), depth %u(%u)",
416 error_msg, le16_to_cpu(eh->eh_magic),
417 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
418 max, le16_to_cpu(eh->eh_depth), depth);
423 #define ext4_ext_check(inode, eh, depth) \
424 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
426 int ext4_ext_check_inode(struct inode *inode)
428 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
432 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
434 int k, l = path->p_depth;
437 for (k = 0; k <= l; k++, path++) {
439 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
440 ext4_idx_pblock(path->p_idx));
441 } else if (path->p_ext) {
442 ext_debug(" %d:[%d]%d:%llu ",
443 le32_to_cpu(path->p_ext->ee_block),
444 ext4_ext_is_uninitialized(path->p_ext),
445 ext4_ext_get_actual_len(path->p_ext),
446 ext4_ext_pblock(path->p_ext));
453 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
455 int depth = ext_depth(inode);
456 struct ext4_extent_header *eh;
457 struct ext4_extent *ex;
463 eh = path[depth].p_hdr;
464 ex = EXT_FIRST_EXTENT(eh);
466 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
468 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
469 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
470 ext4_ext_is_uninitialized(ex),
471 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
476 #define ext4_ext_show_path(inode, path)
477 #define ext4_ext_show_leaf(inode, path)
480 void ext4_ext_drop_refs(struct ext4_ext_path *path)
482 int depth = path->p_depth;
485 for (i = 0; i <= depth; i++, path++)
493 * ext4_ext_binsearch_idx:
494 * binary search for the closest index of the given block
495 * the header must be checked before calling this
498 ext4_ext_binsearch_idx(struct inode *inode,
499 struct ext4_ext_path *path, ext4_lblk_t block)
501 struct ext4_extent_header *eh = path->p_hdr;
502 struct ext4_extent_idx *r, *l, *m;
505 ext_debug("binsearch for %u(idx): ", block);
507 l = EXT_FIRST_INDEX(eh) + 1;
508 r = EXT_LAST_INDEX(eh);
511 if (block < le32_to_cpu(m->ei_block))
515 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
516 m, le32_to_cpu(m->ei_block),
517 r, le32_to_cpu(r->ei_block));
521 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
522 ext4_idx_pblock(path->p_idx));
524 #ifdef CHECK_BINSEARCH
526 struct ext4_extent_idx *chix, *ix;
529 chix = ix = EXT_FIRST_INDEX(eh);
530 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
532 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
533 printk(KERN_DEBUG "k=%d, ix=0x%p, "
535 ix, EXT_FIRST_INDEX(eh));
536 printk(KERN_DEBUG "%u <= %u\n",
537 le32_to_cpu(ix->ei_block),
538 le32_to_cpu(ix[-1].ei_block));
540 BUG_ON(k && le32_to_cpu(ix->ei_block)
541 <= le32_to_cpu(ix[-1].ei_block));
542 if (block < le32_to_cpu(ix->ei_block))
546 BUG_ON(chix != path->p_idx);
553 * ext4_ext_binsearch:
554 * binary search for closest extent of the given block
555 * the header must be checked before calling this
558 ext4_ext_binsearch(struct inode *inode,
559 struct ext4_ext_path *path, ext4_lblk_t block)
561 struct ext4_extent_header *eh = path->p_hdr;
562 struct ext4_extent *r, *l, *m;
564 if (eh->eh_entries == 0) {
566 * this leaf is empty:
567 * we get such a leaf in split/add case
572 ext_debug("binsearch for %u: ", block);
574 l = EXT_FIRST_EXTENT(eh) + 1;
575 r = EXT_LAST_EXTENT(eh);
579 if (block < le32_to_cpu(m->ee_block))
583 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
584 m, le32_to_cpu(m->ee_block),
585 r, le32_to_cpu(r->ee_block));
589 ext_debug(" -> %d:%llu:[%d]%d ",
590 le32_to_cpu(path->p_ext->ee_block),
591 ext4_ext_pblock(path->p_ext),
592 ext4_ext_is_uninitialized(path->p_ext),
593 ext4_ext_get_actual_len(path->p_ext));
595 #ifdef CHECK_BINSEARCH
597 struct ext4_extent *chex, *ex;
600 chex = ex = EXT_FIRST_EXTENT(eh);
601 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
602 BUG_ON(k && le32_to_cpu(ex->ee_block)
603 <= le32_to_cpu(ex[-1].ee_block));
604 if (block < le32_to_cpu(ex->ee_block))
608 BUG_ON(chex != path->p_ext);
614 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
616 struct ext4_extent_header *eh;
618 eh = ext_inode_hdr(inode);
621 eh->eh_magic = EXT4_EXT_MAGIC;
622 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
623 ext4_mark_inode_dirty(handle, inode);
624 ext4_ext_invalidate_cache(inode);
628 struct ext4_ext_path *
629 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
630 struct ext4_ext_path *path)
632 struct ext4_extent_header *eh;
633 struct buffer_head *bh;
634 short int depth, i, ppos = 0, alloc = 0;
636 eh = ext_inode_hdr(inode);
637 depth = ext_depth(inode);
639 /* account possible depth increase */
641 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
644 return ERR_PTR(-ENOMEM);
651 /* walk through the tree */
653 int need_to_validate = 0;
655 ext_debug("depth %d: num %d, max %d\n",
656 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
658 ext4_ext_binsearch_idx(inode, path + ppos, block);
659 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
660 path[ppos].p_depth = i;
661 path[ppos].p_ext = NULL;
663 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
666 if (!bh_uptodate_or_lock(bh)) {
667 if (bh_submit_read(bh) < 0) {
671 /* validate the extent entries */
672 need_to_validate = 1;
674 eh = ext_block_hdr(bh);
676 if (unlikely(ppos > depth)) {
678 EXT4_ERROR_INODE(inode,
679 "ppos %d > depth %d", ppos, depth);
682 path[ppos].p_bh = bh;
683 path[ppos].p_hdr = eh;
686 if (need_to_validate && ext4_ext_check(inode, eh, i))
690 path[ppos].p_depth = i;
691 path[ppos].p_ext = NULL;
692 path[ppos].p_idx = NULL;
695 ext4_ext_binsearch(inode, path + ppos, block);
696 /* if not an empty leaf */
697 if (path[ppos].p_ext)
698 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
700 ext4_ext_show_path(inode, path);
705 ext4_ext_drop_refs(path);
708 return ERR_PTR(-EIO);
712 * ext4_ext_insert_index:
713 * insert new index [@logical;@ptr] into the block at @curp;
714 * check where to insert: before @curp or after @curp
716 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
717 struct ext4_ext_path *curp,
718 int logical, ext4_fsblk_t ptr)
720 struct ext4_extent_idx *ix;
723 err = ext4_ext_get_access(handle, inode, curp);
727 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
728 EXT4_ERROR_INODE(inode,
729 "logical %d == ei_block %d!",
730 logical, le32_to_cpu(curp->p_idx->ei_block));
733 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
734 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
736 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
737 len = (len - 1) * sizeof(struct ext4_extent_idx);
738 len = len < 0 ? 0 : len;
739 ext_debug("insert new index %d after: %llu. "
740 "move %d from 0x%p to 0x%p\n",
742 (curp->p_idx + 1), (curp->p_idx + 2));
743 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
745 ix = curp->p_idx + 1;
748 len = len * sizeof(struct ext4_extent_idx);
749 len = len < 0 ? 0 : len;
750 ext_debug("insert new index %d before: %llu. "
751 "move %d from 0x%p to 0x%p\n",
753 curp->p_idx, (curp->p_idx + 1));
754 memmove(curp->p_idx + 1, curp->p_idx, len);
758 ix->ei_block = cpu_to_le32(logical);
759 ext4_idx_store_pblock(ix, ptr);
760 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
762 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
763 > le16_to_cpu(curp->p_hdr->eh_max))) {
764 EXT4_ERROR_INODE(inode,
765 "logical %d == ei_block %d!",
766 logical, le32_to_cpu(curp->p_idx->ei_block));
769 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
770 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
774 err = ext4_ext_dirty(handle, inode, curp);
775 ext4_std_error(inode->i_sb, err);
782 * inserts new subtree into the path, using free index entry
784 * - allocates all needed blocks (new leaf and all intermediate index blocks)
785 * - makes decision where to split
786 * - moves remaining extents and index entries (right to the split point)
787 * into the newly allocated blocks
788 * - initializes subtree
790 static int ext4_ext_split(handle_t *handle, struct inode *inode,
791 struct ext4_ext_path *path,
792 struct ext4_extent *newext, int at)
794 struct buffer_head *bh = NULL;
795 int depth = ext_depth(inode);
796 struct ext4_extent_header *neh;
797 struct ext4_extent_idx *fidx;
798 struct ext4_extent *ex;
800 ext4_fsblk_t newblock, oldblock;
802 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
805 /* make decision: where to split? */
806 /* FIXME: now decision is simplest: at current extent */
808 /* if current leaf will be split, then we should use
809 * border from split point */
810 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
811 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
814 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
815 border = path[depth].p_ext[1].ee_block;
816 ext_debug("leaf will be split."
817 " next leaf starts at %d\n",
818 le32_to_cpu(border));
820 border = newext->ee_block;
821 ext_debug("leaf will be added."
822 " next leaf starts at %d\n",
823 le32_to_cpu(border));
827 * If error occurs, then we break processing
828 * and mark filesystem read-only. index won't
829 * be inserted and tree will be in consistent
830 * state. Next mount will repair buffers too.
834 * Get array to track all allocated blocks.
835 * We need this to handle errors and free blocks
838 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
842 /* allocate all needed blocks */
843 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
844 for (a = 0; a < depth - at; a++) {
845 newblock = ext4_ext_new_meta_block(handle, inode, path,
849 ablocks[a] = newblock;
852 /* initialize new leaf */
853 newblock = ablocks[--a];
854 if (unlikely(newblock == 0)) {
855 EXT4_ERROR_INODE(inode, "newblock == 0!");
859 bh = sb_getblk(inode->i_sb, newblock);
866 err = ext4_journal_get_create_access(handle, bh);
870 neh = ext_block_hdr(bh);
872 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
873 neh->eh_magic = EXT4_EXT_MAGIC;
875 ex = EXT_FIRST_EXTENT(neh);
877 /* move remainder of path[depth] to the new leaf */
878 if (unlikely(path[depth].p_hdr->eh_entries !=
879 path[depth].p_hdr->eh_max)) {
880 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
881 path[depth].p_hdr->eh_entries,
882 path[depth].p_hdr->eh_max);
886 /* start copy from next extent */
887 /* TODO: we could do it by single memmove */
890 while (path[depth].p_ext <=
891 EXT_MAX_EXTENT(path[depth].p_hdr)) {
892 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
893 le32_to_cpu(path[depth].p_ext->ee_block),
894 ext4_ext_pblock(path[depth].p_ext),
895 ext4_ext_is_uninitialized(path[depth].p_ext),
896 ext4_ext_get_actual_len(path[depth].p_ext),
898 /*memmove(ex++, path[depth].p_ext++,
899 sizeof(struct ext4_extent));
905 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
906 le16_add_cpu(&neh->eh_entries, m);
909 set_buffer_uptodate(bh);
912 err = ext4_handle_dirty_metadata(handle, inode, bh);
918 /* correct old leaf */
920 err = ext4_ext_get_access(handle, inode, path + depth);
923 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
924 err = ext4_ext_dirty(handle, inode, path + depth);
930 /* create intermediate indexes */
932 if (unlikely(k < 0)) {
933 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
938 ext_debug("create %d intermediate indices\n", k);
939 /* insert new index into current index block */
940 /* current depth stored in i var */
944 newblock = ablocks[--a];
945 bh = sb_getblk(inode->i_sb, newblock);
952 err = ext4_journal_get_create_access(handle, bh);
956 neh = ext_block_hdr(bh);
957 neh->eh_entries = cpu_to_le16(1);
958 neh->eh_magic = EXT4_EXT_MAGIC;
959 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
960 neh->eh_depth = cpu_to_le16(depth - i);
961 fidx = EXT_FIRST_INDEX(neh);
962 fidx->ei_block = border;
963 ext4_idx_store_pblock(fidx, oldblock);
965 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
966 i, newblock, le32_to_cpu(border), oldblock);
971 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
972 EXT_MAX_INDEX(path[i].p_hdr));
973 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
974 EXT_LAST_INDEX(path[i].p_hdr))) {
975 EXT4_ERROR_INODE(inode,
976 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
977 le32_to_cpu(path[i].p_ext->ee_block));
981 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
982 ext_debug("%d: move %d:%llu in new index %llu\n", i,
983 le32_to_cpu(path[i].p_idx->ei_block),
984 ext4_idx_pblock(path[i].p_idx),
986 /*memmove(++fidx, path[i].p_idx++,
987 sizeof(struct ext4_extent_idx));
989 BUG_ON(neh->eh_entries > neh->eh_max);*/
994 memmove(++fidx, path[i].p_idx - m,
995 sizeof(struct ext4_extent_idx) * m);
996 le16_add_cpu(&neh->eh_entries, m);
998 set_buffer_uptodate(bh);
1001 err = ext4_handle_dirty_metadata(handle, inode, bh);
1007 /* correct old index */
1009 err = ext4_ext_get_access(handle, inode, path + i);
1012 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1013 err = ext4_ext_dirty(handle, inode, path + i);
1021 /* insert new index */
1022 err = ext4_ext_insert_index(handle, inode, path + at,
1023 le32_to_cpu(border), newblock);
1027 if (buffer_locked(bh))
1033 /* free all allocated blocks in error case */
1034 for (i = 0; i < depth; i++) {
1037 ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1038 EXT4_FREE_BLOCKS_METADATA);
1047 * ext4_ext_grow_indepth:
1048 * implements tree growing procedure:
1049 * - allocates new block
1050 * - moves top-level data (index block or leaf) into the new block
1051 * - initializes new top-level, creating index that points to the
1052 * just created block
1054 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1055 struct ext4_ext_path *path,
1056 struct ext4_extent *newext)
1058 struct ext4_ext_path *curp = path;
1059 struct ext4_extent_header *neh;
1060 struct buffer_head *bh;
1061 ext4_fsblk_t newblock;
1064 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1068 bh = sb_getblk(inode->i_sb, newblock);
1071 ext4_std_error(inode->i_sb, err);
1076 err = ext4_journal_get_create_access(handle, bh);
1082 /* move top-level index/leaf into new block */
1083 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1085 /* set size of new block */
1086 neh = ext_block_hdr(bh);
1087 /* old root could have indexes or leaves
1088 * so calculate e_max right way */
1089 if (ext_depth(inode))
1090 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1092 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1093 neh->eh_magic = EXT4_EXT_MAGIC;
1094 set_buffer_uptodate(bh);
1097 err = ext4_handle_dirty_metadata(handle, inode, bh);
1101 /* create index in new top-level index: num,max,pointer */
1102 err = ext4_ext_get_access(handle, inode, curp);
1106 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1107 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1108 curp->p_hdr->eh_entries = cpu_to_le16(1);
1109 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1111 if (path[0].p_hdr->eh_depth)
1112 curp->p_idx->ei_block =
1113 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1115 curp->p_idx->ei_block =
1116 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1117 ext4_idx_store_pblock(curp->p_idx, newblock);
1119 neh = ext_inode_hdr(inode);
1120 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1121 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1122 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1123 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1125 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1126 err = ext4_ext_dirty(handle, inode, curp);
1134 * ext4_ext_create_new_leaf:
1135 * finds empty index and adds new leaf.
1136 * if no free index is found, then it requests in-depth growing.
1138 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1139 struct ext4_ext_path *path,
1140 struct ext4_extent *newext)
1142 struct ext4_ext_path *curp;
1143 int depth, i, err = 0;
1146 i = depth = ext_depth(inode);
1148 /* walk up to the tree and look for free index entry */
1149 curp = path + depth;
1150 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1155 /* we use already allocated block for index block,
1156 * so subsequent data blocks should be contiguous */
1157 if (EXT_HAS_FREE_INDEX(curp)) {
1158 /* if we found index with free entry, then use that
1159 * entry: create all needed subtree and add new leaf */
1160 err = ext4_ext_split(handle, inode, path, newext, i);
1165 ext4_ext_drop_refs(path);
1166 path = ext4_ext_find_extent(inode,
1167 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1170 err = PTR_ERR(path);
1172 /* tree is full, time to grow in depth */
1173 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1178 ext4_ext_drop_refs(path);
1179 path = ext4_ext_find_extent(inode,
1180 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1183 err = PTR_ERR(path);
1188 * only first (depth 0 -> 1) produces free space;
1189 * in all other cases we have to split the grown tree
1191 depth = ext_depth(inode);
1192 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1193 /* now we need to split */
1203 * search the closest allocated block to the left for *logical
1204 * and returns it at @logical + it's physical address at @phys
1205 * if *logical is the smallest allocated block, the function
1206 * returns 0 at @phys
1207 * return value contains 0 (success) or error code
1209 static int ext4_ext_search_left(struct inode *inode,
1210 struct ext4_ext_path *path,
1211 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1213 struct ext4_extent_idx *ix;
1214 struct ext4_extent *ex;
1217 if (unlikely(path == NULL)) {
1218 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1221 depth = path->p_depth;
1224 if (depth == 0 && path->p_ext == NULL)
1227 /* usually extent in the path covers blocks smaller
1228 * then *logical, but it can be that extent is the
1229 * first one in the file */
1231 ex = path[depth].p_ext;
1232 ee_len = ext4_ext_get_actual_len(ex);
1233 if (*logical < le32_to_cpu(ex->ee_block)) {
1234 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1235 EXT4_ERROR_INODE(inode,
1236 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1237 *logical, le32_to_cpu(ex->ee_block));
1240 while (--depth >= 0) {
1241 ix = path[depth].p_idx;
1242 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1243 EXT4_ERROR_INODE(inode,
1244 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1245 ix != NULL ? ix->ei_block : 0,
1246 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1247 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1255 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1256 EXT4_ERROR_INODE(inode,
1257 "logical %d < ee_block %d + ee_len %d!",
1258 *logical, le32_to_cpu(ex->ee_block), ee_len);
1262 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1263 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1268 * search the closest allocated block to the right for *logical
1269 * and returns it at @logical + it's physical address at @phys
1270 * if *logical is the smallest allocated block, the function
1271 * returns 0 at @phys
1272 * return value contains 0 (success) or error code
1274 static int ext4_ext_search_right(struct inode *inode,
1275 struct ext4_ext_path *path,
1276 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1278 struct buffer_head *bh = NULL;
1279 struct ext4_extent_header *eh;
1280 struct ext4_extent_idx *ix;
1281 struct ext4_extent *ex;
1283 int depth; /* Note, NOT eh_depth; depth from top of tree */
1286 if (unlikely(path == NULL)) {
1287 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1290 depth = path->p_depth;
1293 if (depth == 0 && path->p_ext == NULL)
1296 /* usually extent in the path covers blocks smaller
1297 * then *logical, but it can be that extent is the
1298 * first one in the file */
1300 ex = path[depth].p_ext;
1301 ee_len = ext4_ext_get_actual_len(ex);
1302 if (*logical < le32_to_cpu(ex->ee_block)) {
1303 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1304 EXT4_ERROR_INODE(inode,
1305 "first_extent(path[%d].p_hdr) != ex",
1309 while (--depth >= 0) {
1310 ix = path[depth].p_idx;
1311 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1312 EXT4_ERROR_INODE(inode,
1313 "ix != EXT_FIRST_INDEX *logical %d!",
1318 *logical = le32_to_cpu(ex->ee_block);
1319 *phys = ext4_ext_pblock(ex);
1323 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1324 EXT4_ERROR_INODE(inode,
1325 "logical %d < ee_block %d + ee_len %d!",
1326 *logical, le32_to_cpu(ex->ee_block), ee_len);
1330 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1331 /* next allocated block in this leaf */
1333 *logical = le32_to_cpu(ex->ee_block);
1334 *phys = ext4_ext_pblock(ex);
1338 /* go up and search for index to the right */
1339 while (--depth >= 0) {
1340 ix = path[depth].p_idx;
1341 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1345 /* we've gone up to the root and found no index to the right */
1349 /* we've found index to the right, let's
1350 * follow it and find the closest allocated
1351 * block to the right */
1353 block = ext4_idx_pblock(ix);
1354 while (++depth < path->p_depth) {
1355 bh = sb_bread(inode->i_sb, block);
1358 eh = ext_block_hdr(bh);
1359 /* subtract from p_depth to get proper eh_depth */
1360 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1364 ix = EXT_FIRST_INDEX(eh);
1365 block = ext4_idx_pblock(ix);
1369 bh = sb_bread(inode->i_sb, block);
1372 eh = ext_block_hdr(bh);
1373 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1377 ex = EXT_FIRST_EXTENT(eh);
1378 *logical = le32_to_cpu(ex->ee_block);
1379 *phys = ext4_ext_pblock(ex);
1385 * ext4_ext_next_allocated_block:
1386 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1387 * NOTE: it considers block number from index entry as
1388 * allocated block. Thus, index entries have to be consistent
1392 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1396 BUG_ON(path == NULL);
1397 depth = path->p_depth;
1399 if (depth == 0 && path->p_ext == NULL)
1400 return EXT_MAX_BLOCK;
1402 while (depth >= 0) {
1403 if (depth == path->p_depth) {
1405 if (path[depth].p_ext !=
1406 EXT_LAST_EXTENT(path[depth].p_hdr))
1407 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1410 if (path[depth].p_idx !=
1411 EXT_LAST_INDEX(path[depth].p_hdr))
1412 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1417 return EXT_MAX_BLOCK;
1421 * ext4_ext_next_leaf_block:
1422 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1424 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1425 struct ext4_ext_path *path)
1429 BUG_ON(path == NULL);
1430 depth = path->p_depth;
1432 /* zero-tree has no leaf blocks at all */
1434 return EXT_MAX_BLOCK;
1436 /* go to index block */
1439 while (depth >= 0) {
1440 if (path[depth].p_idx !=
1441 EXT_LAST_INDEX(path[depth].p_hdr))
1442 return (ext4_lblk_t)
1443 le32_to_cpu(path[depth].p_idx[1].ei_block);
1447 return EXT_MAX_BLOCK;
1451 * ext4_ext_correct_indexes:
1452 * if leaf gets modified and modified extent is first in the leaf,
1453 * then we have to correct all indexes above.
1454 * TODO: do we need to correct tree in all cases?
1456 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1457 struct ext4_ext_path *path)
1459 struct ext4_extent_header *eh;
1460 int depth = ext_depth(inode);
1461 struct ext4_extent *ex;
1465 eh = path[depth].p_hdr;
1466 ex = path[depth].p_ext;
1468 if (unlikely(ex == NULL || eh == NULL)) {
1469 EXT4_ERROR_INODE(inode,
1470 "ex %p == NULL or eh %p == NULL", ex, eh);
1475 /* there is no tree at all */
1479 if (ex != EXT_FIRST_EXTENT(eh)) {
1480 /* we correct tree if first leaf got modified only */
1485 * TODO: we need correction if border is smaller than current one
1488 border = path[depth].p_ext->ee_block;
1489 err = ext4_ext_get_access(handle, inode, path + k);
1492 path[k].p_idx->ei_block = border;
1493 err = ext4_ext_dirty(handle, inode, path + k);
1498 /* change all left-side indexes */
1499 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1501 err = ext4_ext_get_access(handle, inode, path + k);
1504 path[k].p_idx->ei_block = border;
1505 err = ext4_ext_dirty(handle, inode, path + k);
1514 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1515 struct ext4_extent *ex2)
1517 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1520 * Make sure that either both extents are uninitialized, or
1523 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1526 if (ext4_ext_is_uninitialized(ex1))
1527 max_len = EXT_UNINIT_MAX_LEN;
1529 max_len = EXT_INIT_MAX_LEN;
1531 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1532 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1534 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1535 le32_to_cpu(ex2->ee_block))
1539 * To allow future support for preallocated extents to be added
1540 * as an RO_COMPAT feature, refuse to merge to extents if
1541 * this can result in the top bit of ee_len being set.
1543 if (ext1_ee_len + ext2_ee_len > max_len)
1545 #ifdef AGGRESSIVE_TEST
1546 if (ext1_ee_len >= 4)
1550 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1556 * This function tries to merge the "ex" extent to the next extent in the tree.
1557 * It always tries to merge towards right. If you want to merge towards
1558 * left, pass "ex - 1" as argument instead of "ex".
1559 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1560 * 1 if they got merged.
1562 static int ext4_ext_try_to_merge(struct inode *inode,
1563 struct ext4_ext_path *path,
1564 struct ext4_extent *ex)
1566 struct ext4_extent_header *eh;
1567 unsigned int depth, len;
1569 int uninitialized = 0;
1571 depth = ext_depth(inode);
1572 BUG_ON(path[depth].p_hdr == NULL);
1573 eh = path[depth].p_hdr;
1575 while (ex < EXT_LAST_EXTENT(eh)) {
1576 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1578 /* merge with next extent! */
1579 if (ext4_ext_is_uninitialized(ex))
1581 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1582 + ext4_ext_get_actual_len(ex + 1));
1584 ext4_ext_mark_uninitialized(ex);
1586 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1587 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1588 * sizeof(struct ext4_extent);
1589 memmove(ex + 1, ex + 2, len);
1591 le16_add_cpu(&eh->eh_entries, -1);
1593 WARN_ON(eh->eh_entries == 0);
1594 if (!eh->eh_entries)
1595 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1602 * check if a portion of the "newext" extent overlaps with an
1605 * If there is an overlap discovered, it updates the length of the newext
1606 * such that there will be no overlap, and then returns 1.
1607 * If there is no overlap found, it returns 0.
1609 static unsigned int ext4_ext_check_overlap(struct inode *inode,
1610 struct ext4_extent *newext,
1611 struct ext4_ext_path *path)
1614 unsigned int depth, len1;
1615 unsigned int ret = 0;
1617 b1 = le32_to_cpu(newext->ee_block);
1618 len1 = ext4_ext_get_actual_len(newext);
1619 depth = ext_depth(inode);
1620 if (!path[depth].p_ext)
1622 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1625 * get the next allocated block if the extent in the path
1626 * is before the requested block(s)
1629 b2 = ext4_ext_next_allocated_block(path);
1630 if (b2 == EXT_MAX_BLOCK)
1634 /* check for wrap through zero on extent logical start block*/
1635 if (b1 + len1 < b1) {
1636 len1 = EXT_MAX_BLOCK - b1;
1637 newext->ee_len = cpu_to_le16(len1);
1641 /* check for overlap */
1642 if (b1 + len1 > b2) {
1643 newext->ee_len = cpu_to_le16(b2 - b1);
1651 * ext4_ext_insert_extent:
1652 * tries to merge requsted extent into the existing extent or
1653 * inserts requested extent as new one into the tree,
1654 * creating new leaf in the no-space case.
1656 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1657 struct ext4_ext_path *path,
1658 struct ext4_extent *newext, int flag)
1660 struct ext4_extent_header *eh;
1661 struct ext4_extent *ex, *fex;
1662 struct ext4_extent *nearex; /* nearest extent */
1663 struct ext4_ext_path *npath = NULL;
1664 int depth, len, err;
1666 unsigned uninitialized = 0;
1668 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1669 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1672 depth = ext_depth(inode);
1673 ex = path[depth].p_ext;
1674 if (unlikely(path[depth].p_hdr == NULL)) {
1675 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1679 /* try to insert block into found extent and return */
1680 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1681 && ext4_can_extents_be_merged(inode, ex, newext)) {
1682 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1683 ext4_ext_is_uninitialized(newext),
1684 ext4_ext_get_actual_len(newext),
1685 le32_to_cpu(ex->ee_block),
1686 ext4_ext_is_uninitialized(ex),
1687 ext4_ext_get_actual_len(ex),
1688 ext4_ext_pblock(ex));
1689 err = ext4_ext_get_access(handle, inode, path + depth);
1694 * ext4_can_extents_be_merged should have checked that either
1695 * both extents are uninitialized, or both aren't. Thus we
1696 * need to check only one of them here.
1698 if (ext4_ext_is_uninitialized(ex))
1700 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1701 + ext4_ext_get_actual_len(newext));
1703 ext4_ext_mark_uninitialized(ex);
1704 eh = path[depth].p_hdr;
1710 depth = ext_depth(inode);
1711 eh = path[depth].p_hdr;
1712 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1715 /* probably next leaf has space for us? */
1716 fex = EXT_LAST_EXTENT(eh);
1717 next = ext4_ext_next_leaf_block(inode, path);
1718 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1719 && next != EXT_MAX_BLOCK) {
1720 ext_debug("next leaf block - %d\n", next);
1721 BUG_ON(npath != NULL);
1722 npath = ext4_ext_find_extent(inode, next, NULL);
1724 return PTR_ERR(npath);
1725 BUG_ON(npath->p_depth != path->p_depth);
1726 eh = npath[depth].p_hdr;
1727 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1728 ext_debug("next leaf isnt full(%d)\n",
1729 le16_to_cpu(eh->eh_entries));
1733 ext_debug("next leaf has no free space(%d,%d)\n",
1734 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1738 * There is no free space in the found leaf.
1739 * We're gonna add a new leaf in the tree.
1741 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1744 depth = ext_depth(inode);
1745 eh = path[depth].p_hdr;
1748 nearex = path[depth].p_ext;
1750 err = ext4_ext_get_access(handle, inode, path + depth);
1755 /* there is no extent in this leaf, create first one */
1756 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1757 le32_to_cpu(newext->ee_block),
1758 ext4_ext_pblock(newext),
1759 ext4_ext_is_uninitialized(newext),
1760 ext4_ext_get_actual_len(newext));
1761 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1762 } else if (le32_to_cpu(newext->ee_block)
1763 > le32_to_cpu(nearex->ee_block)) {
1764 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1765 if (nearex != EXT_LAST_EXTENT(eh)) {
1766 len = EXT_MAX_EXTENT(eh) - nearex;
1767 len = (len - 1) * sizeof(struct ext4_extent);
1768 len = len < 0 ? 0 : len;
1769 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1770 "move %d from 0x%p to 0x%p\n",
1771 le32_to_cpu(newext->ee_block),
1772 ext4_ext_pblock(newext),
1773 ext4_ext_is_uninitialized(newext),
1774 ext4_ext_get_actual_len(newext),
1775 nearex, len, nearex + 1, nearex + 2);
1776 memmove(nearex + 2, nearex + 1, len);
1778 path[depth].p_ext = nearex + 1;
1780 BUG_ON(newext->ee_block == nearex->ee_block);
1781 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1782 len = len < 0 ? 0 : len;
1783 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1784 "move %d from 0x%p to 0x%p\n",
1785 le32_to_cpu(newext->ee_block),
1786 ext4_ext_pblock(newext),
1787 ext4_ext_is_uninitialized(newext),
1788 ext4_ext_get_actual_len(newext),
1789 nearex, len, nearex + 1, nearex + 2);
1790 memmove(nearex + 1, nearex, len);
1791 path[depth].p_ext = nearex;
1794 le16_add_cpu(&eh->eh_entries, 1);
1795 nearex = path[depth].p_ext;
1796 nearex->ee_block = newext->ee_block;
1797 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1798 nearex->ee_len = newext->ee_len;
1801 /* try to merge extents to the right */
1802 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1803 ext4_ext_try_to_merge(inode, path, nearex);
1805 /* try to merge extents to the left */
1807 /* time to correct all indexes above */
1808 err = ext4_ext_correct_indexes(handle, inode, path);
1812 err = ext4_ext_dirty(handle, inode, path + depth);
1816 ext4_ext_drop_refs(npath);
1819 ext4_ext_invalidate_cache(inode);
1823 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1824 ext4_lblk_t num, ext_prepare_callback func,
1827 struct ext4_ext_path *path = NULL;
1828 struct ext4_ext_cache cbex;
1829 struct ext4_extent *ex;
1830 ext4_lblk_t next, start = 0, end = 0;
1831 ext4_lblk_t last = block + num;
1832 int depth, exists, err = 0;
1834 BUG_ON(func == NULL);
1835 BUG_ON(inode == NULL);
1837 while (block < last && block != EXT_MAX_BLOCK) {
1839 /* find extent for this block */
1840 down_read(&EXT4_I(inode)->i_data_sem);
1841 path = ext4_ext_find_extent(inode, block, path);
1842 up_read(&EXT4_I(inode)->i_data_sem);
1844 err = PTR_ERR(path);
1849 depth = ext_depth(inode);
1850 if (unlikely(path[depth].p_hdr == NULL)) {
1851 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1855 ex = path[depth].p_ext;
1856 next = ext4_ext_next_allocated_block(path);
1860 /* there is no extent yet, so try to allocate
1861 * all requested space */
1864 } else if (le32_to_cpu(ex->ee_block) > block) {
1865 /* need to allocate space before found extent */
1867 end = le32_to_cpu(ex->ee_block);
1868 if (block + num < end)
1870 } else if (block >= le32_to_cpu(ex->ee_block)
1871 + ext4_ext_get_actual_len(ex)) {
1872 /* need to allocate space after found extent */
1877 } else if (block >= le32_to_cpu(ex->ee_block)) {
1879 * some part of requested space is covered
1883 end = le32_to_cpu(ex->ee_block)
1884 + ext4_ext_get_actual_len(ex);
1885 if (block + num < end)
1891 BUG_ON(end <= start);
1894 cbex.ec_block = start;
1895 cbex.ec_len = end - start;
1897 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1899 cbex.ec_block = le32_to_cpu(ex->ee_block);
1900 cbex.ec_len = ext4_ext_get_actual_len(ex);
1901 cbex.ec_start = ext4_ext_pblock(ex);
1902 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1905 if (unlikely(cbex.ec_len == 0)) {
1906 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1910 err = func(inode, path, &cbex, ex, cbdata);
1911 ext4_ext_drop_refs(path);
1916 if (err == EXT_REPEAT)
1918 else if (err == EXT_BREAK) {
1923 if (ext_depth(inode) != depth) {
1924 /* depth was changed. we have to realloc path */
1929 block = cbex.ec_block + cbex.ec_len;
1933 ext4_ext_drop_refs(path);
1941 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1942 __u32 len, ext4_fsblk_t start, int type)
1944 struct ext4_ext_cache *cex;
1946 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1947 cex = &EXT4_I(inode)->i_cached_extent;
1948 cex->ec_type = type;
1949 cex->ec_block = block;
1951 cex->ec_start = start;
1952 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1956 * ext4_ext_put_gap_in_cache:
1957 * calculate boundaries of the gap that the requested block fits into
1958 * and cache this gap
1961 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1964 int depth = ext_depth(inode);
1967 struct ext4_extent *ex;
1969 ex = path[depth].p_ext;
1971 /* there is no extent yet, so gap is [0;-] */
1973 len = EXT_MAX_BLOCK;
1974 ext_debug("cache gap(whole file):");
1975 } else if (block < le32_to_cpu(ex->ee_block)) {
1977 len = le32_to_cpu(ex->ee_block) - block;
1978 ext_debug("cache gap(before): %u [%u:%u]",
1980 le32_to_cpu(ex->ee_block),
1981 ext4_ext_get_actual_len(ex));
1982 } else if (block >= le32_to_cpu(ex->ee_block)
1983 + ext4_ext_get_actual_len(ex)) {
1985 lblock = le32_to_cpu(ex->ee_block)
1986 + ext4_ext_get_actual_len(ex);
1988 next = ext4_ext_next_allocated_block(path);
1989 ext_debug("cache gap(after): [%u:%u] %u",
1990 le32_to_cpu(ex->ee_block),
1991 ext4_ext_get_actual_len(ex),
1993 BUG_ON(next == lblock);
1994 len = next - lblock;
2000 ext_debug(" -> %u:%lu\n", lblock, len);
2001 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
2005 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2006 struct ext4_extent *ex)
2008 struct ext4_ext_cache *cex;
2009 int ret = EXT4_EXT_CACHE_NO;
2012 * We borrow i_block_reservation_lock to protect i_cached_extent
2014 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2015 cex = &EXT4_I(inode)->i_cached_extent;
2017 /* has cache valid data? */
2018 if (cex->ec_type == EXT4_EXT_CACHE_NO)
2021 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
2022 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
2023 if (in_range(block, cex->ec_block, cex->ec_len)) {
2024 ex->ee_block = cpu_to_le32(cex->ec_block);
2025 ext4_ext_store_pblock(ex, cex->ec_start);
2026 ex->ee_len = cpu_to_le16(cex->ec_len);
2027 ext_debug("%u cached by %u:%u:%llu\n",
2029 cex->ec_block, cex->ec_len, cex->ec_start);
2033 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2039 * removes index from the index block.
2040 * It's used in truncate case only, thus all requests are for
2041 * last index in the block only.
2043 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2044 struct ext4_ext_path *path)
2049 /* free index block */
2051 leaf = ext4_idx_pblock(path->p_idx);
2052 if (unlikely(path->p_hdr->eh_entries == 0)) {
2053 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2056 err = ext4_ext_get_access(handle, inode, path);
2059 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2060 err = ext4_ext_dirty(handle, inode, path);
2063 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2064 ext4_free_blocks(handle, inode, 0, leaf, 1,
2065 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2070 * ext4_ext_calc_credits_for_single_extent:
2071 * This routine returns max. credits that needed to insert an extent
2072 * to the extent tree.
2073 * When pass the actual path, the caller should calculate credits
2076 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2077 struct ext4_ext_path *path)
2080 int depth = ext_depth(inode);
2083 /* probably there is space in leaf? */
2084 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2085 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2088 * There are some space in the leaf tree, no
2089 * need to account for leaf block credit
2091 * bitmaps and block group descriptor blocks
2092 * and other metadat blocks still need to be
2095 /* 1 bitmap, 1 block group descriptor */
2096 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2101 return ext4_chunk_trans_blocks(inode, nrblocks);
2105 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2107 * if nrblocks are fit in a single extent (chunk flag is 1), then
2108 * in the worse case, each tree level index/leaf need to be changed
2109 * if the tree split due to insert a new extent, then the old tree
2110 * index/leaf need to be updated too
2112 * If the nrblocks are discontiguous, they could cause
2113 * the whole tree split more than once, but this is really rare.
2115 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2118 int depth = ext_depth(inode);
2128 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2129 struct ext4_extent *ex,
2130 ext4_lblk_t from, ext4_lblk_t to)
2132 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2133 int flags = EXT4_FREE_BLOCKS_FORGET;
2135 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2136 flags |= EXT4_FREE_BLOCKS_METADATA;
2137 #ifdef EXTENTS_STATS
2139 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2140 spin_lock(&sbi->s_ext_stats_lock);
2141 sbi->s_ext_blocks += ee_len;
2142 sbi->s_ext_extents++;
2143 if (ee_len < sbi->s_ext_min)
2144 sbi->s_ext_min = ee_len;
2145 if (ee_len > sbi->s_ext_max)
2146 sbi->s_ext_max = ee_len;
2147 if (ext_depth(inode) > sbi->s_depth_max)
2148 sbi->s_depth_max = ext_depth(inode);
2149 spin_unlock(&sbi->s_ext_stats_lock);
2152 if (from >= le32_to_cpu(ex->ee_block)
2153 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2158 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2159 start = ext4_ext_pblock(ex) + ee_len - num;
2160 ext_debug("free last %u blocks starting %llu\n", num, start);
2161 ext4_free_blocks(handle, inode, 0, start, num, flags);
2162 } else if (from == le32_to_cpu(ex->ee_block)
2163 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2164 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2165 from, to, le32_to_cpu(ex->ee_block), ee_len);
2167 printk(KERN_INFO "strange request: removal(2) "
2168 "%u-%u from %u:%u\n",
2169 from, to, le32_to_cpu(ex->ee_block), ee_len);
2175 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2176 struct ext4_ext_path *path, ext4_lblk_t start)
2178 int err = 0, correct_index = 0;
2179 int depth = ext_depth(inode), credits;
2180 struct ext4_extent_header *eh;
2181 ext4_lblk_t a, b, block;
2183 ext4_lblk_t ex_ee_block;
2184 unsigned short ex_ee_len;
2185 unsigned uninitialized = 0;
2186 struct ext4_extent *ex;
2188 /* the header must be checked already in ext4_ext_remove_space() */
2189 ext_debug("truncate since %u in leaf\n", start);
2190 if (!path[depth].p_hdr)
2191 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2192 eh = path[depth].p_hdr;
2193 if (unlikely(path[depth].p_hdr == NULL)) {
2194 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2197 /* find where to start removing */
2198 ex = EXT_LAST_EXTENT(eh);
2200 ex_ee_block = le32_to_cpu(ex->ee_block);
2201 ex_ee_len = ext4_ext_get_actual_len(ex);
2203 while (ex >= EXT_FIRST_EXTENT(eh) &&
2204 ex_ee_block + ex_ee_len > start) {
2206 if (ext4_ext_is_uninitialized(ex))
2211 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2212 uninitialized, ex_ee_len);
2213 path[depth].p_ext = ex;
2215 a = ex_ee_block > start ? ex_ee_block : start;
2216 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2217 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2219 ext_debug(" border %u:%u\n", a, b);
2221 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2225 } else if (a != ex_ee_block) {
2226 /* remove tail of the extent */
2227 block = ex_ee_block;
2229 } else if (b != ex_ee_block + ex_ee_len - 1) {
2230 /* remove head of the extent */
2233 /* there is no "make a hole" API yet */
2236 /* remove whole extent: excellent! */
2237 block = ex_ee_block;
2239 BUG_ON(a != ex_ee_block);
2240 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2244 * 3 for leaf, sb, and inode plus 2 (bmap and group
2245 * descriptor) for each block group; assume two block
2246 * groups plus ex_ee_len/blocks_per_block_group for
2249 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2250 if (ex == EXT_FIRST_EXTENT(eh)) {
2252 credits += (ext_depth(inode)) + 1;
2254 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2256 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2260 err = ext4_ext_get_access(handle, inode, path + depth);
2264 err = ext4_remove_blocks(handle, inode, ex, a, b);
2269 /* this extent is removed; mark slot entirely unused */
2270 ext4_ext_store_pblock(ex, 0);
2271 le16_add_cpu(&eh->eh_entries, -1);
2274 ex->ee_block = cpu_to_le32(block);
2275 ex->ee_len = cpu_to_le16(num);
2277 * Do not mark uninitialized if all the blocks in the
2278 * extent have been removed.
2280 if (uninitialized && num)
2281 ext4_ext_mark_uninitialized(ex);
2283 err = ext4_ext_dirty(handle, inode, path + depth);
2287 ext_debug("new extent: %u:%u:%llu\n", block, num,
2288 ext4_ext_pblock(ex));
2290 ex_ee_block = le32_to_cpu(ex->ee_block);
2291 ex_ee_len = ext4_ext_get_actual_len(ex);
2294 if (correct_index && eh->eh_entries)
2295 err = ext4_ext_correct_indexes(handle, inode, path);
2297 /* if this leaf is free, then we should
2298 * remove it from index block above */
2299 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2300 err = ext4_ext_rm_idx(handle, inode, path + depth);
2307 * ext4_ext_more_to_rm:
2308 * returns 1 if current index has to be freed (even partial)
2311 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2313 BUG_ON(path->p_idx == NULL);
2315 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2319 * if truncate on deeper level happened, it wasn't partial,
2320 * so we have to consider current index for truncation
2322 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2327 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2329 struct super_block *sb = inode->i_sb;
2330 int depth = ext_depth(inode);
2331 struct ext4_ext_path *path;
2335 ext_debug("truncate since %u\n", start);
2337 /* probably first extent we're gonna free will be last in block */
2338 handle = ext4_journal_start(inode, depth + 1);
2340 return PTR_ERR(handle);
2343 ext4_ext_invalidate_cache(inode);
2346 * We start scanning from right side, freeing all the blocks
2347 * after i_size and walking into the tree depth-wise.
2349 depth = ext_depth(inode);
2350 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2352 ext4_journal_stop(handle);
2355 path[0].p_depth = depth;
2356 path[0].p_hdr = ext_inode_hdr(inode);
2357 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2363 while (i >= 0 && err == 0) {
2365 /* this is leaf block */
2366 err = ext4_ext_rm_leaf(handle, inode, path, start);
2367 /* root level has p_bh == NULL, brelse() eats this */
2368 brelse(path[i].p_bh);
2369 path[i].p_bh = NULL;
2374 /* this is index block */
2375 if (!path[i].p_hdr) {
2376 ext_debug("initialize header\n");
2377 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2380 if (!path[i].p_idx) {
2381 /* this level hasn't been touched yet */
2382 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2383 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2384 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2386 le16_to_cpu(path[i].p_hdr->eh_entries));
2388 /* we were already here, see at next index */
2392 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2393 i, EXT_FIRST_INDEX(path[i].p_hdr),
2395 if (ext4_ext_more_to_rm(path + i)) {
2396 struct buffer_head *bh;
2397 /* go to the next level */
2398 ext_debug("move to level %d (block %llu)\n",
2399 i + 1, ext4_idx_pblock(path[i].p_idx));
2400 memset(path + i + 1, 0, sizeof(*path));
2401 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2403 /* should we reset i_size? */
2407 if (WARN_ON(i + 1 > depth)) {
2411 if (ext4_ext_check(inode, ext_block_hdr(bh),
2416 path[i + 1].p_bh = bh;
2418 /* save actual number of indexes since this
2419 * number is changed at the next iteration */
2420 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2423 /* we finished processing this index, go up */
2424 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2425 /* index is empty, remove it;
2426 * handle must be already prepared by the
2427 * truncatei_leaf() */
2428 err = ext4_ext_rm_idx(handle, inode, path + i);
2430 /* root level has p_bh == NULL, brelse() eats this */
2431 brelse(path[i].p_bh);
2432 path[i].p_bh = NULL;
2434 ext_debug("return to level %d\n", i);
2438 /* TODO: flexible tree reduction should be here */
2439 if (path->p_hdr->eh_entries == 0) {
2441 * truncate to zero freed all the tree,
2442 * so we need to correct eh_depth
2444 err = ext4_ext_get_access(handle, inode, path);
2446 ext_inode_hdr(inode)->eh_depth = 0;
2447 ext_inode_hdr(inode)->eh_max =
2448 cpu_to_le16(ext4_ext_space_root(inode, 0));
2449 err = ext4_ext_dirty(handle, inode, path);
2453 ext4_ext_drop_refs(path);
2457 ext4_journal_stop(handle);
2463 * called at mount time
2465 void ext4_ext_init(struct super_block *sb)
2468 * possible initialization would be here
2471 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2472 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2473 printk(KERN_INFO "EXT4-fs: file extents enabled");
2474 #ifdef AGGRESSIVE_TEST
2475 printk(", aggressive tests");
2477 #ifdef CHECK_BINSEARCH
2478 printk(", check binsearch");
2480 #ifdef EXTENTS_STATS
2485 #ifdef EXTENTS_STATS
2486 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2487 EXT4_SB(sb)->s_ext_min = 1 << 30;
2488 EXT4_SB(sb)->s_ext_max = 0;
2494 * called at umount time
2496 void ext4_ext_release(struct super_block *sb)
2498 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2501 #ifdef EXTENTS_STATS
2502 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2503 struct ext4_sb_info *sbi = EXT4_SB(sb);
2504 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2505 sbi->s_ext_blocks, sbi->s_ext_extents,
2506 sbi->s_ext_blocks / sbi->s_ext_extents);
2507 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2508 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2513 /* FIXME!! we need to try to merge to left or right after zero-out */
2514 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2516 ext4_fsblk_t ee_pblock;
2517 unsigned int ee_len;
2520 ee_len = ext4_ext_get_actual_len(ex);
2521 ee_pblock = ext4_ext_pblock(ex);
2523 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2530 #define EXT4_EXT_ZERO_LEN 7
2532 * This function is called by ext4_ext_map_blocks() if someone tries to write
2533 * to an uninitialized extent. It may result in splitting the uninitialized
2534 * extent into multiple extents (upto three - one initialized and two
2536 * There are three possibilities:
2537 * a> There is no split required: Entire extent should be initialized
2538 * b> Splits in two extents: Write is happening at either end of the extent
2539 * c> Splits in three extents: Somone is writing in middle of the extent
2541 static int ext4_ext_convert_to_initialized(handle_t *handle,
2542 struct inode *inode,
2543 struct ext4_map_blocks *map,
2544 struct ext4_ext_path *path)
2546 struct ext4_extent *ex, newex, orig_ex;
2547 struct ext4_extent *ex1 = NULL;
2548 struct ext4_extent *ex2 = NULL;
2549 struct ext4_extent *ex3 = NULL;
2550 struct ext4_extent_header *eh;
2551 ext4_lblk_t ee_block, eof_block;
2552 unsigned int allocated, ee_len, depth;
2553 ext4_fsblk_t newblock;
2558 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2559 "block %llu, max_blocks %u\n", inode->i_ino,
2560 (unsigned long long)map->m_lblk, map->m_len);
2562 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2563 inode->i_sb->s_blocksize_bits;
2564 if (eof_block < map->m_lblk + map->m_len)
2565 eof_block = map->m_lblk + map->m_len;
2567 depth = ext_depth(inode);
2568 eh = path[depth].p_hdr;
2569 ex = path[depth].p_ext;
2570 ee_block = le32_to_cpu(ex->ee_block);
2571 ee_len = ext4_ext_get_actual_len(ex);
2572 allocated = ee_len - (map->m_lblk - ee_block);
2573 newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2576 orig_ex.ee_block = ex->ee_block;
2577 orig_ex.ee_len = cpu_to_le16(ee_len);
2578 ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2581 * It is safe to convert extent to initialized via explicit
2582 * zeroout only if extent is fully insde i_size or new_size.
2584 may_zeroout = ee_block + ee_len <= eof_block;
2586 err = ext4_ext_get_access(handle, inode, path + depth);
2589 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2590 if (ee_len <= 2*EXT4_EXT_ZERO_LEN && may_zeroout) {
2591 err = ext4_ext_zeroout(inode, &orig_ex);
2593 goto fix_extent_len;
2594 /* update the extent length and mark as initialized */
2595 ex->ee_block = orig_ex.ee_block;
2596 ex->ee_len = orig_ex.ee_len;
2597 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2598 ext4_ext_dirty(handle, inode, path + depth);
2599 /* zeroed the full extent */
2603 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2604 if (map->m_lblk > ee_block) {
2606 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2607 ext4_ext_mark_uninitialized(ex1);
2611 * for sanity, update the length of the ex2 extent before
2612 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2613 * overlap of blocks.
2615 if (!ex1 && allocated > map->m_len)
2616 ex2->ee_len = cpu_to_le16(map->m_len);
2617 /* ex3: to ee_block + ee_len : uninitialised */
2618 if (allocated > map->m_len) {
2619 unsigned int newdepth;
2620 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2621 if (allocated <= EXT4_EXT_ZERO_LEN && may_zeroout) {
2623 * map->m_lblk == ee_block is handled by the zerouout
2625 * Mark first half uninitialized.
2626 * Mark second half initialized and zero out the
2627 * initialized extent
2629 ex->ee_block = orig_ex.ee_block;
2630 ex->ee_len = cpu_to_le16(ee_len - allocated);
2631 ext4_ext_mark_uninitialized(ex);
2632 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2633 ext4_ext_dirty(handle, inode, path + depth);
2636 ex3->ee_block = cpu_to_le32(map->m_lblk);
2637 ext4_ext_store_pblock(ex3, newblock);
2638 ex3->ee_len = cpu_to_le16(allocated);
2639 err = ext4_ext_insert_extent(handle, inode, path,
2641 if (err == -ENOSPC) {
2642 err = ext4_ext_zeroout(inode, &orig_ex);
2644 goto fix_extent_len;
2645 ex->ee_block = orig_ex.ee_block;
2646 ex->ee_len = orig_ex.ee_len;
2647 ext4_ext_store_pblock(ex,
2648 ext4_ext_pblock(&orig_ex));
2649 ext4_ext_dirty(handle, inode, path + depth);
2650 /* blocks available from map->m_lblk */
2654 goto fix_extent_len;
2657 * We need to zero out the second half because
2658 * an fallocate request can update file size and
2659 * converting the second half to initialized extent
2660 * implies that we can leak some junk data to user
2663 err = ext4_ext_zeroout(inode, ex3);
2666 * We should actually mark the
2667 * second half as uninit and return error
2668 * Insert would have changed the extent
2670 depth = ext_depth(inode);
2671 ext4_ext_drop_refs(path);
2672 path = ext4_ext_find_extent(inode, map->m_lblk,
2675 err = PTR_ERR(path);
2678 /* get the second half extent details */
2679 ex = path[depth].p_ext;
2680 err = ext4_ext_get_access(handle, inode,
2684 ext4_ext_mark_uninitialized(ex);
2685 ext4_ext_dirty(handle, inode, path + depth);
2689 /* zeroed the second half */
2693 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2694 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2695 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2696 ext4_ext_mark_uninitialized(ex3);
2697 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2698 if (err == -ENOSPC && may_zeroout) {
2699 err = ext4_ext_zeroout(inode, &orig_ex);
2701 goto fix_extent_len;
2702 /* update the extent length and mark as initialized */
2703 ex->ee_block = orig_ex.ee_block;
2704 ex->ee_len = orig_ex.ee_len;
2705 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2706 ext4_ext_dirty(handle, inode, path + depth);
2707 /* zeroed the full extent */
2708 /* blocks available from map->m_lblk */
2712 goto fix_extent_len;
2714 * The depth, and hence eh & ex might change
2715 * as part of the insert above.
2717 newdepth = ext_depth(inode);
2719 * update the extent length after successful insert of the
2722 ee_len -= ext4_ext_get_actual_len(ex3);
2723 orig_ex.ee_len = cpu_to_le16(ee_len);
2724 may_zeroout = ee_block + ee_len <= eof_block;
2727 ext4_ext_drop_refs(path);
2728 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2730 err = PTR_ERR(path);
2733 eh = path[depth].p_hdr;
2734 ex = path[depth].p_ext;
2738 err = ext4_ext_get_access(handle, inode, path + depth);
2742 allocated = map->m_len;
2744 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2745 * to insert a extent in the middle zerout directly
2746 * otherwise give the extent a chance to merge to left
2748 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2749 map->m_lblk != ee_block && may_zeroout) {
2750 err = ext4_ext_zeroout(inode, &orig_ex);
2752 goto fix_extent_len;
2753 /* update the extent length and mark as initialized */
2754 ex->ee_block = orig_ex.ee_block;
2755 ex->ee_len = orig_ex.ee_len;
2756 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2757 ext4_ext_dirty(handle, inode, path + depth);
2758 /* zero out the first half */
2759 /* blocks available from map->m_lblk */
2764 * If there was a change of depth as part of the
2765 * insertion of ex3 above, we need to update the length
2766 * of the ex1 extent again here
2768 if (ex1 && ex1 != ex) {
2770 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2771 ext4_ext_mark_uninitialized(ex1);
2774 /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2775 ex2->ee_block = cpu_to_le32(map->m_lblk);
2776 ext4_ext_store_pblock(ex2, newblock);
2777 ex2->ee_len = cpu_to_le16(allocated);
2781 * New (initialized) extent starts from the first block
2782 * in the current extent. i.e., ex2 == ex
2783 * We have to see if it can be merged with the extent
2786 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2788 * To merge left, pass "ex2 - 1" to try_to_merge(),
2789 * since it merges towards right _only_.
2791 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2793 err = ext4_ext_correct_indexes(handle, inode, path);
2796 depth = ext_depth(inode);
2801 * Try to Merge towards right. This might be required
2802 * only when the whole extent is being written to.
2803 * i.e. ex2 == ex and ex3 == NULL.
2806 ret = ext4_ext_try_to_merge(inode, path, ex2);
2808 err = ext4_ext_correct_indexes(handle, inode, path);
2813 /* Mark modified extent as dirty */
2814 err = ext4_ext_dirty(handle, inode, path + depth);
2817 err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2818 if (err == -ENOSPC && may_zeroout) {
2819 err = ext4_ext_zeroout(inode, &orig_ex);
2821 goto fix_extent_len;
2822 /* update the extent length and mark as initialized */
2823 ex->ee_block = orig_ex.ee_block;
2824 ex->ee_len = orig_ex.ee_len;
2825 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2826 ext4_ext_dirty(handle, inode, path + depth);
2827 /* zero out the first half */
2830 goto fix_extent_len;
2832 ext4_ext_show_leaf(inode, path);
2833 return err ? err : allocated;
2836 ex->ee_block = orig_ex.ee_block;
2837 ex->ee_len = orig_ex.ee_len;
2838 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2839 ext4_ext_mark_uninitialized(ex);
2840 ext4_ext_dirty(handle, inode, path + depth);
2845 * This function is called by ext4_ext_map_blocks() from
2846 * ext4_get_blocks_dio_write() when DIO to write
2847 * to an uninitialized extent.
2849 * Writing to an uninitized extent may result in splitting the uninitialized
2850 * extent into multiple /intialized unintialized extents (up to three)
2851 * There are three possibilities:
2852 * a> There is no split required: Entire extent should be uninitialized
2853 * b> Splits in two extents: Write is happening at either end of the extent
2854 * c> Splits in three extents: Somone is writing in middle of the extent
2856 * One of more index blocks maybe needed if the extent tree grow after
2857 * the unintialized extent split. To prevent ENOSPC occur at the IO
2858 * complete, we need to split the uninitialized extent before DIO submit
2859 * the IO. The uninitialized extent called at this time will be split
2860 * into three uninitialized extent(at most). After IO complete, the part
2861 * being filled will be convert to initialized by the end_io callback function
2862 * via ext4_convert_unwritten_extents().
2864 * Returns the size of uninitialized extent to be written on success.
2866 static int ext4_split_unwritten_extents(handle_t *handle,
2867 struct inode *inode,
2868 struct ext4_map_blocks *map,
2869 struct ext4_ext_path *path,
2872 struct ext4_extent *ex, newex, orig_ex;
2873 struct ext4_extent *ex1 = NULL;
2874 struct ext4_extent *ex2 = NULL;
2875 struct ext4_extent *ex3 = NULL;
2876 ext4_lblk_t ee_block, eof_block;
2877 unsigned int allocated, ee_len, depth;
2878 ext4_fsblk_t newblock;
2882 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2883 "block %llu, max_blocks %u\n", inode->i_ino,
2884 (unsigned long long)map->m_lblk, map->m_len);
2886 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2887 inode->i_sb->s_blocksize_bits;
2888 if (eof_block < map->m_lblk + map->m_len)
2889 eof_block = map->m_lblk + map->m_len;
2891 depth = ext_depth(inode);
2892 ex = path[depth].p_ext;
2893 ee_block = le32_to_cpu(ex->ee_block);
2894 ee_len = ext4_ext_get_actual_len(ex);
2895 allocated = ee_len - (map->m_lblk - ee_block);
2896 newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2899 orig_ex.ee_block = ex->ee_block;
2900 orig_ex.ee_len = cpu_to_le16(ee_len);
2901 ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2904 * It is safe to convert extent to initialized via explicit
2905 * zeroout only if extent is fully insde i_size or new_size.
2907 may_zeroout = ee_block + ee_len <= eof_block;
2910 * If the uninitialized extent begins at the same logical
2911 * block where the write begins, and the write completely
2912 * covers the extent, then we don't need to split it.
2914 if ((map->m_lblk == ee_block) && (allocated <= map->m_len))
2917 err = ext4_ext_get_access(handle, inode, path + depth);
2920 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2921 if (map->m_lblk > ee_block) {
2923 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2924 ext4_ext_mark_uninitialized(ex1);
2928 * for sanity, update the length of the ex2 extent before
2929 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2930 * overlap of blocks.
2932 if (!ex1 && allocated > map->m_len)
2933 ex2->ee_len = cpu_to_le16(map->m_len);
2934 /* ex3: to ee_block + ee_len : uninitialised */
2935 if (allocated > map->m_len) {
2936 unsigned int newdepth;
2938 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2939 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2940 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2941 ext4_ext_mark_uninitialized(ex3);
2942 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2943 if (err == -ENOSPC && may_zeroout) {
2944 err = ext4_ext_zeroout(inode, &orig_ex);
2946 goto fix_extent_len;
2947 /* update the extent length and mark as initialized */
2948 ex->ee_block = orig_ex.ee_block;
2949 ex->ee_len = orig_ex.ee_len;
2950 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2951 ext4_ext_dirty(handle, inode, path + depth);
2952 /* zeroed the full extent */
2953 /* blocks available from map->m_lblk */
2957 goto fix_extent_len;
2959 * The depth, and hence eh & ex might change
2960 * as part of the insert above.
2962 newdepth = ext_depth(inode);
2964 * update the extent length after successful insert of the
2967 ee_len -= ext4_ext_get_actual_len(ex3);
2968 orig_ex.ee_len = cpu_to_le16(ee_len);
2969 may_zeroout = ee_block + ee_len <= eof_block;
2972 ext4_ext_drop_refs(path);
2973 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2975 err = PTR_ERR(path);
2978 ex = path[depth].p_ext;
2982 err = ext4_ext_get_access(handle, inode, path + depth);
2986 allocated = map->m_len;
2989 * If there was a change of depth as part of the
2990 * insertion of ex3 above, we need to update the length
2991 * of the ex1 extent again here
2993 if (ex1 && ex1 != ex) {
2995 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2996 ext4_ext_mark_uninitialized(ex1);
3000 * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
3001 * using direct I/O, uninitialised still.
3003 ex2->ee_block = cpu_to_le32(map->m_lblk);
3004 ext4_ext_store_pblock(ex2, newblock);
3005 ex2->ee_len = cpu_to_le16(allocated);
3006 ext4_ext_mark_uninitialized(ex2);
3009 /* Mark modified extent as dirty */
3010 err = ext4_ext_dirty(handle, inode, path + depth);
3011 ext_debug("out here\n");
3014 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3015 if (err == -ENOSPC && may_zeroout) {
3016 err = ext4_ext_zeroout(inode, &orig_ex);
3018 goto fix_extent_len;
3019 /* update the extent length and mark as initialized */
3020 ex->ee_block = orig_ex.ee_block;
3021 ex->ee_len = orig_ex.ee_len;
3022 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3023 ext4_ext_dirty(handle, inode, path + depth);
3024 /* zero out the first half */
3027 goto fix_extent_len;
3029 ext4_ext_show_leaf(inode, path);
3030 return err ? err : allocated;
3033 ex->ee_block = orig_ex.ee_block;
3034 ex->ee_len = orig_ex.ee_len;
3035 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3036 ext4_ext_mark_uninitialized(ex);
3037 ext4_ext_dirty(handle, inode, path + depth);
3040 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3041 struct inode *inode,
3042 struct ext4_ext_path *path)
3044 struct ext4_extent *ex;
3045 struct ext4_extent_header *eh;
3050 depth = ext_depth(inode);
3051 eh = path[depth].p_hdr;
3052 ex = path[depth].p_ext;
3054 err = ext4_ext_get_access(handle, inode, path + depth);
3057 /* first mark the extent as initialized */
3058 ext4_ext_mark_initialized(ex);
3061 * We have to see if it can be merged with the extent
3064 if (ex > EXT_FIRST_EXTENT(eh)) {
3066 * To merge left, pass "ex - 1" to try_to_merge(),
3067 * since it merges towards right _only_.
3069 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3071 err = ext4_ext_correct_indexes(handle, inode, path);
3074 depth = ext_depth(inode);
3079 * Try to Merge towards right.
3081 ret = ext4_ext_try_to_merge(inode, path, ex);
3083 err = ext4_ext_correct_indexes(handle, inode, path);
3086 depth = ext_depth(inode);
3088 /* Mark modified extent as dirty */
3089 err = ext4_ext_dirty(handle, inode, path + depth);
3091 ext4_ext_show_leaf(inode, path);
3095 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3096 sector_t block, int count)
3099 for (i = 0; i < count; i++)
3100 unmap_underlying_metadata(bdev, block + i);
3104 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3106 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3107 struct ext4_map_blocks *map,
3108 struct ext4_ext_path *path,
3112 struct ext4_extent_header *eh;
3113 struct ext4_extent *ex, *last_ex;
3115 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3118 depth = ext_depth(inode);
3119 eh = path[depth].p_hdr;
3120 ex = path[depth].p_ext;
3122 if (unlikely(!eh->eh_entries)) {
3123 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3124 "EOFBLOCKS_FL set");
3127 last_ex = EXT_LAST_EXTENT(eh);
3129 * We should clear the EOFBLOCKS_FL flag if we are writing the
3130 * last block in the last extent in the file. We test this by
3131 * first checking to see if the caller to
3132 * ext4_ext_get_blocks() was interested in the last block (or
3133 * a block beyond the last block) in the current extent. If
3134 * this turns out to be false, we can bail out from this
3135 * function immediately.
3137 if (map->m_lblk + len < le32_to_cpu(last_ex->ee_block) +
3138 ext4_ext_get_actual_len(last_ex))
3141 * If the caller does appear to be planning to write at or
3142 * beyond the end of the current extent, we then test to see
3143 * if the current extent is the last extent in the file, by
3144 * checking to make sure it was reached via the rightmost node
3145 * at each level of the tree.
3147 for (i = depth-1; i >= 0; i--)
3148 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3150 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3151 return ext4_mark_inode_dirty(handle, inode);
3155 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3156 struct ext4_map_blocks *map,
3157 struct ext4_ext_path *path, int flags,
3158 unsigned int allocated, ext4_fsblk_t newblock)
3162 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3164 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3165 "block %llu, max_blocks %u, flags %d, allocated %u",
3166 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3168 ext4_ext_show_leaf(inode, path);
3170 /* get_block() before submit the IO, split the extent */
3171 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3172 ret = ext4_split_unwritten_extents(handle, inode, map,
3175 * Flag the inode(non aio case) or end_io struct (aio case)
3176 * that this IO needs to convertion to written when IO is
3180 io->flag = EXT4_IO_END_UNWRITTEN;
3182 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3183 if (ext4_should_dioread_nolock(inode))
3184 map->m_flags |= EXT4_MAP_UNINIT;
3187 /* IO end_io complete, convert the filled extent to written */
3188 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3189 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3192 ext4_update_inode_fsync_trans(handle, inode, 1);
3193 err = check_eofblocks_fl(handle, inode, map, path,
3199 /* buffered IO case */
3201 * repeat fallocate creation request
3202 * we already have an unwritten extent
3204 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3207 /* buffered READ or buffered write_begin() lookup */
3208 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3210 * We have blocks reserved already. We
3211 * return allocated blocks so that delalloc
3212 * won't do block reservation for us. But
3213 * the buffer head will be unmapped so that
3214 * a read from the block returns 0s.
3216 map->m_flags |= EXT4_MAP_UNWRITTEN;
3220 /* buffered write, writepage time, convert*/
3221 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3223 ext4_update_inode_fsync_trans(handle, inode, 1);
3224 err = check_eofblocks_fl(handle, inode, map, path, map->m_len);
3235 map->m_flags |= EXT4_MAP_NEW;
3237 * if we allocated more blocks than requested
3238 * we need to make sure we unmap the extra block
3239 * allocated. The actual needed block will get
3240 * unmapped later when we find the buffer_head marked
3243 if (allocated > map->m_len) {
3244 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3245 newblock + map->m_len,
3246 allocated - map->m_len);
3247 allocated = map->m_len;
3251 * If we have done fallocate with the offset that is already
3252 * delayed allocated, we would have block reservation
3253 * and quota reservation done in the delayed write path.
3254 * But fallocate would have already updated quota and block
3255 * count for this offset. So cancel these reservation
3257 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3258 ext4_da_update_reserve_space(inode, allocated, 0);
3261 map->m_flags |= EXT4_MAP_MAPPED;
3263 if (allocated > map->m_len)
3264 allocated = map->m_len;
3265 ext4_ext_show_leaf(inode, path);
3266 map->m_pblk = newblock;
3267 map->m_len = allocated;
3270 ext4_ext_drop_refs(path);
3273 return err ? err : allocated;
3277 * Block allocation/map/preallocation routine for extents based files
3280 * Need to be called with
3281 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3282 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3284 * return > 0, number of of blocks already mapped/allocated
3285 * if create == 0 and these are pre-allocated blocks
3286 * buffer head is unmapped
3287 * otherwise blocks are mapped
3289 * return = 0, if plain look up failed (blocks have not been allocated)
3290 * buffer head is unmapped
3292 * return < 0, error case.
3294 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3295 struct ext4_map_blocks *map, int flags)
3297 struct ext4_ext_path *path = NULL;
3298 struct ext4_extent_header *eh;
3299 struct ext4_extent newex, *ex;
3300 ext4_fsblk_t newblock;
3301 int err = 0, depth, ret, cache_type;
3302 unsigned int allocated = 0;
3303 struct ext4_allocation_request ar;
3304 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3306 ext_debug("blocks %u/%u requested for inode %lu\n",
3307 map->m_lblk, map->m_len, inode->i_ino);
3309 /* check in cache */
3310 cache_type = ext4_ext_in_cache(inode, map->m_lblk, &newex);
3312 if (cache_type == EXT4_EXT_CACHE_GAP) {
3313 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3315 * block isn't allocated yet and
3316 * user doesn't want to allocate it
3320 /* we should allocate requested block */
3321 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3322 /* block is already allocated */
3323 newblock = map->m_lblk
3324 - le32_to_cpu(newex.ee_block)
3325 + ext4_ext_pblock(&newex);
3326 /* number of remaining blocks in the extent */
3327 allocated = ext4_ext_get_actual_len(&newex) -
3328 (map->m_lblk - le32_to_cpu(newex.ee_block));
3335 /* find extent for this block */
3336 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3338 err = PTR_ERR(path);
3343 depth = ext_depth(inode);
3346 * consistent leaf must not be empty;
3347 * this situation is possible, though, _during_ tree modification;
3348 * this is why assert can't be put in ext4_ext_find_extent()
3350 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3351 EXT4_ERROR_INODE(inode, "bad extent address "
3352 "lblock: %lu, depth: %d pblock %lld",
3353 (unsigned long) map->m_lblk, depth,
3354 path[depth].p_block);
3358 eh = path[depth].p_hdr;
3360 ex = path[depth].p_ext;
3362 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3363 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3364 unsigned short ee_len;
3367 * Uninitialized extents are treated as holes, except that
3368 * we split out initialized portions during a write.
3370 ee_len = ext4_ext_get_actual_len(ex);
3371 /* if found extent covers block, simply return it */
3372 if (in_range(map->m_lblk, ee_block, ee_len)) {
3373 newblock = map->m_lblk - ee_block + ee_start;
3374 /* number of remaining blocks in the extent */
3375 allocated = ee_len - (map->m_lblk - ee_block);
3376 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3377 ee_block, ee_len, newblock);
3379 /* Do not put uninitialized extent in the cache */
3380 if (!ext4_ext_is_uninitialized(ex)) {
3381 ext4_ext_put_in_cache(inode, ee_block,
3383 EXT4_EXT_CACHE_EXTENT);
3386 ret = ext4_ext_handle_uninitialized_extents(handle,
3387 inode, map, path, flags, allocated,
3394 * requested block isn't allocated yet;
3395 * we couldn't try to create block if create flag is zero
3397 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3399 * put just found gap into cache to speed up
3400 * subsequent requests
3402 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3406 * Okay, we need to do block allocation.
3409 /* find neighbour allocated blocks */
3410 ar.lleft = map->m_lblk;
3411 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3414 ar.lright = map->m_lblk;
3415 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3420 * See if request is beyond maximum number of blocks we can have in
3421 * a single extent. For an initialized extent this limit is
3422 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3423 * EXT_UNINIT_MAX_LEN.
3425 if (map->m_len > EXT_INIT_MAX_LEN &&
3426 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3427 map->m_len = EXT_INIT_MAX_LEN;
3428 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3429 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3430 map->m_len = EXT_UNINIT_MAX_LEN;
3432 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3433 newex.ee_block = cpu_to_le32(map->m_lblk);
3434 newex.ee_len = cpu_to_le16(map->m_len);
3435 err = ext4_ext_check_overlap(inode, &newex, path);
3437 allocated = ext4_ext_get_actual_len(&newex);
3439 allocated = map->m_len;
3441 /* allocate new block */
3443 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3444 ar.logical = map->m_lblk;
3446 if (S_ISREG(inode->i_mode))
3447 ar.flags = EXT4_MB_HINT_DATA;
3449 /* disable in-core preallocation for non-regular files */
3451 newblock = ext4_mb_new_blocks(handle, &ar, &err);
3454 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3455 ar.goal, newblock, allocated);
3457 /* try to insert new extent into found leaf and return */
3458 ext4_ext_store_pblock(&newex, newblock);
3459 newex.ee_len = cpu_to_le16(ar.len);
3460 /* Mark uninitialized */
3461 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3462 ext4_ext_mark_uninitialized(&newex);
3464 * io_end structure was created for every IO write to an
3465 * uninitialized extent. To avoid unecessary conversion,
3466 * here we flag the IO that really needs the conversion.
3467 * For non asycn direct IO case, flag the inode state
3468 * that we need to perform convertion when IO is done.
3470 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3472 io->flag = EXT4_IO_END_UNWRITTEN;
3474 ext4_set_inode_state(inode,
3475 EXT4_STATE_DIO_UNWRITTEN);
3477 if (ext4_should_dioread_nolock(inode))
3478 map->m_flags |= EXT4_MAP_UNINIT;
3481 err = check_eofblocks_fl(handle, inode, map, path, ar.len);
3485 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3487 /* free data blocks we just allocated */
3488 /* not a good idea to call discard here directly,
3489 * but otherwise we'd need to call it every free() */
3490 ext4_discard_preallocations(inode);
3491 ext4_free_blocks(handle, inode, 0, ext4_ext_pblock(&newex),
3492 ext4_ext_get_actual_len(&newex), 0);
3496 /* previous routine could use block we allocated */
3497 newblock = ext4_ext_pblock(&newex);
3498 allocated = ext4_ext_get_actual_len(&newex);
3499 if (allocated > map->m_len)
3500 allocated = map->m_len;
3501 map->m_flags |= EXT4_MAP_NEW;
3504 * Update reserved blocks/metadata blocks after successful
3505 * block allocation which had been deferred till now.
3507 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3508 ext4_da_update_reserve_space(inode, allocated, 1);
3511 * Cache the extent and update transaction to commit on fdatasync only
3512 * when it is _not_ an uninitialized extent.
3514 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3515 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock,
3516 EXT4_EXT_CACHE_EXTENT);
3517 ext4_update_inode_fsync_trans(handle, inode, 1);
3519 ext4_update_inode_fsync_trans(handle, inode, 0);
3521 if (allocated > map->m_len)
3522 allocated = map->m_len;
3523 ext4_ext_show_leaf(inode, path);
3524 map->m_flags |= EXT4_MAP_MAPPED;
3525 map->m_pblk = newblock;
3526 map->m_len = allocated;
3529 ext4_ext_drop_refs(path);
3532 return err ? err : allocated;
3535 void ext4_ext_truncate(struct inode *inode)
3537 struct address_space *mapping = inode->i_mapping;
3538 struct super_block *sb = inode->i_sb;
3539 ext4_lblk_t last_block;
3544 * probably first extent we're gonna free will be last in block
3546 err = ext4_writepage_trans_blocks(inode);
3547 handle = ext4_journal_start(inode, err);
3551 if (inode->i_size & (sb->s_blocksize - 1))
3552 ext4_block_truncate_page(handle, mapping, inode->i_size);
3554 if (ext4_orphan_add(handle, inode))
3557 down_write(&EXT4_I(inode)->i_data_sem);
3558 ext4_ext_invalidate_cache(inode);
3560 ext4_discard_preallocations(inode);
3563 * TODO: optimization is possible here.
3564 * Probably we need not scan at all,
3565 * because page truncation is enough.
3568 /* we have to know where to truncate from in crash case */
3569 EXT4_I(inode)->i_disksize = inode->i_size;
3570 ext4_mark_inode_dirty(handle, inode);
3572 last_block = (inode->i_size + sb->s_blocksize - 1)
3573 >> EXT4_BLOCK_SIZE_BITS(sb);
3574 err = ext4_ext_remove_space(inode, last_block);
3576 /* In a multi-transaction truncate, we only make the final
3577 * transaction synchronous.
3580 ext4_handle_sync(handle);
3583 up_write(&EXT4_I(inode)->i_data_sem);
3585 * If this was a simple ftruncate() and the file will remain alive,
3586 * then we need to clear up the orphan record which we created above.
3587 * However, if this was a real unlink then we were called by
3588 * ext4_delete_inode(), and we allow that function to clean up the
3589 * orphan info for us.
3592 ext4_orphan_del(handle, inode);
3594 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3595 ext4_mark_inode_dirty(handle, inode);
3596 ext4_journal_stop(handle);
3599 static void ext4_falloc_update_inode(struct inode *inode,
3600 int mode, loff_t new_size, int update_ctime)
3602 struct timespec now;
3605 now = current_fs_time(inode->i_sb);
3606 if (!timespec_equal(&inode->i_ctime, &now))
3607 inode->i_ctime = now;
3610 * Update only when preallocation was requested beyond
3613 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3614 if (new_size > i_size_read(inode))
3615 i_size_write(inode, new_size);
3616 if (new_size > EXT4_I(inode)->i_disksize)
3617 ext4_update_i_disksize(inode, new_size);
3620 * Mark that we allocate beyond EOF so the subsequent truncate
3621 * can proceed even if the new size is the same as i_size.
3623 if (new_size > i_size_read(inode))
3624 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3630 * preallocate space for a file. This implements ext4's fallocate inode
3631 * operation, which gets called from sys_fallocate system call.
3632 * For block-mapped files, posix_fallocate should fall back to the method
3633 * of writing zeroes to the required new blocks (the same behavior which is
3634 * expected for file systems which do not support fallocate() system call).
3636 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3640 unsigned int max_blocks;
3644 struct ext4_map_blocks map;
3645 unsigned int credits, blkbits = inode->i_blkbits;
3648 * currently supporting (pre)allocate mode for extent-based
3651 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3654 /* preallocation to directories is currently not supported */
3655 if (S_ISDIR(inode->i_mode))
3658 map.m_lblk = offset >> blkbits;
3660 * We can't just convert len to max_blocks because
3661 * If blocksize = 4096 offset = 3072 and len = 2048
3663 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3666 * credits to insert 1 extent into extent tree
3668 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3669 mutex_lock(&inode->i_mutex);
3670 ret = inode_newsize_ok(inode, (len + offset));
3672 mutex_unlock(&inode->i_mutex);
3676 while (ret >= 0 && ret < max_blocks) {
3677 map.m_lblk = map.m_lblk + ret;
3678 map.m_len = max_blocks = max_blocks - ret;
3679 handle = ext4_journal_start(inode, credits);
3680 if (IS_ERR(handle)) {
3681 ret = PTR_ERR(handle);
3684 ret = ext4_map_blocks(handle, inode, &map,
3685 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3689 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3690 "returned error inode#%lu, block=%u, "
3691 "max_blocks=%u", __func__,
3692 inode->i_ino, map.m_lblk, max_blocks);
3694 ext4_mark_inode_dirty(handle, inode);
3695 ret2 = ext4_journal_stop(handle);
3698 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3699 blkbits) >> blkbits))
3700 new_size = offset + len;
3702 new_size = (map.m_lblk + ret) << blkbits;
3704 ext4_falloc_update_inode(inode, mode, new_size,
3705 (map.m_flags & EXT4_MAP_NEW));
3706 ext4_mark_inode_dirty(handle, inode);
3707 ret2 = ext4_journal_stop(handle);
3711 if (ret == -ENOSPC &&
3712 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3716 mutex_unlock(&inode->i_mutex);
3717 return ret > 0 ? ret2 : ret;
3721 * This function convert a range of blocks to written extents
3722 * The caller of this function will pass the start offset and the size.
3723 * all unwritten extents within this range will be converted to
3726 * This function is called from the direct IO end io call back
3727 * function, to convert the fallocated extents after IO is completed.
3728 * Returns 0 on success.
3730 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3734 unsigned int max_blocks;
3737 struct ext4_map_blocks map;
3738 unsigned int credits, blkbits = inode->i_blkbits;
3740 map.m_lblk = offset >> blkbits;
3742 * We can't just convert len to max_blocks because
3743 * If blocksize = 4096 offset = 3072 and len = 2048
3745 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3748 * credits to insert 1 extent into extent tree
3750 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3751 while (ret >= 0 && ret < max_blocks) {
3753 map.m_len = (max_blocks -= ret);
3754 handle = ext4_journal_start(inode, credits);
3755 if (IS_ERR(handle)) {
3756 ret = PTR_ERR(handle);
3759 ret = ext4_map_blocks(handle, inode, &map,
3760 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3763 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3764 "returned error inode#%lu, block=%u, "
3765 "max_blocks=%u", __func__,
3766 inode->i_ino, map.m_lblk, map.m_len);
3768 ext4_mark_inode_dirty(handle, inode);
3769 ret2 = ext4_journal_stop(handle);
3770 if (ret <= 0 || ret2 )
3773 return ret > 0 ? ret2 : ret;
3776 * Callback function called for each extent to gather FIEMAP information.
3778 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3779 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3782 struct fiemap_extent_info *fieinfo = data;
3783 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3790 logical = (__u64)newex->ec_block << blksize_bits;
3792 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3795 struct buffer_head *bh = NULL;
3797 offset = logical >> PAGE_SHIFT;
3798 page = find_get_page(inode->i_mapping, offset);
3799 if (!page || !page_has_buffers(page))
3800 return EXT_CONTINUE;
3802 bh = page_buffers(page);
3805 return EXT_CONTINUE;
3807 if (buffer_delay(bh)) {
3808 flags |= FIEMAP_EXTENT_DELALLOC;
3809 page_cache_release(page);
3811 page_cache_release(page);
3812 return EXT_CONTINUE;
3816 physical = (__u64)newex->ec_start << blksize_bits;
3817 length = (__u64)newex->ec_len << blksize_bits;
3819 if (ex && ext4_ext_is_uninitialized(ex))
3820 flags |= FIEMAP_EXTENT_UNWRITTEN;
3823 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3825 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3826 * this also indicates no more allocated blocks.
3828 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3830 if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3831 newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3832 loff_t size = i_size_read(inode);
3833 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3835 flags |= FIEMAP_EXTENT_LAST;
3836 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3837 logical+length > size)
3838 length = (size - logical + bs - 1) & ~(bs-1);
3841 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3848 return EXT_CONTINUE;
3851 /* fiemap flags we can handle specified here */
3852 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3854 static int ext4_xattr_fiemap(struct inode *inode,
3855 struct fiemap_extent_info *fieinfo)
3859 __u32 flags = FIEMAP_EXTENT_LAST;
3860 int blockbits = inode->i_sb->s_blocksize_bits;
3864 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3865 struct ext4_iloc iloc;
3866 int offset; /* offset of xattr in inode */
3868 error = ext4_get_inode_loc(inode, &iloc);
3871 physical = iloc.bh->b_blocknr << blockbits;
3872 offset = EXT4_GOOD_OLD_INODE_SIZE +
3873 EXT4_I(inode)->i_extra_isize;
3875 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3876 flags |= FIEMAP_EXTENT_DATA_INLINE;
3878 } else { /* external block */
3879 physical = EXT4_I(inode)->i_file_acl << blockbits;
3880 length = inode->i_sb->s_blocksize;
3884 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3886 return (error < 0 ? error : 0);
3889 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3890 __u64 start, __u64 len)
3892 ext4_lblk_t start_blk;
3895 /* fallback to generic here if not in extents fmt */
3896 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3897 return generic_block_fiemap(inode, fieinfo, start, len,
3900 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3903 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3904 error = ext4_xattr_fiemap(inode, fieinfo);
3906 ext4_lblk_t len_blks;
3909 start_blk = start >> inode->i_sb->s_blocksize_bits;
3910 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
3911 if (last_blk >= EXT_MAX_BLOCK)
3912 last_blk = EXT_MAX_BLOCK-1;
3913 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
3916 * Walk the extent tree gathering extent information.
3917 * ext4_ext_fiemap_cb will push extents back to user.
3919 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3920 ext4_ext_fiemap_cb, fieinfo);