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 #include <trace/events/ext4.h>
49 static int ext4_split_extent(handle_t *handle,
51 struct ext4_ext_path *path,
52 struct ext4_map_blocks *map,
56 static int ext4_ext_truncate_extend_restart(handle_t *handle,
62 if (!ext4_handle_valid(handle))
64 if (handle->h_buffer_credits > needed)
66 err = ext4_journal_extend(handle, needed);
69 err = ext4_truncate_restart_trans(handle, inode, needed);
81 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
82 struct ext4_ext_path *path)
85 /* path points to block */
86 return ext4_journal_get_write_access(handle, path->p_bh);
88 /* path points to leaf/index in inode body */
89 /* we use in-core data, no need to protect them */
99 #define ext4_ext_dirty(handle, inode, path) \
100 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
101 static int __ext4_ext_dirty(const char *where, unsigned int line,
102 handle_t *handle, struct inode *inode,
103 struct ext4_ext_path *path)
107 /* path points to block */
108 err = __ext4_handle_dirty_metadata(where, line, handle,
111 /* path points to leaf/index in inode body */
112 err = ext4_mark_inode_dirty(handle, inode);
117 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
118 struct ext4_ext_path *path,
124 struct ext4_extent *ex;
125 depth = path->p_depth;
128 * Try to predict block placement assuming that we are
129 * filling in a file which will eventually be
130 * non-sparse --- i.e., in the case of libbfd writing
131 * an ELF object sections out-of-order but in a way
132 * the eventually results in a contiguous object or
133 * executable file, or some database extending a table
134 * space file. However, this is actually somewhat
135 * non-ideal if we are writing a sparse file such as
136 * qemu or KVM writing a raw image file that is going
137 * to stay fairly sparse, since it will end up
138 * fragmenting the file system's free space. Maybe we
139 * should have some hueristics or some way to allow
140 * userspace to pass a hint to file system,
141 * especially if the latter case turns out to be
144 ex = path[depth].p_ext;
146 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
147 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
149 if (block > ext_block)
150 return ext_pblk + (block - ext_block);
152 return ext_pblk - (ext_block - block);
155 /* it looks like index is empty;
156 * try to find starting block from index itself */
157 if (path[depth].p_bh)
158 return path[depth].p_bh->b_blocknr;
161 /* OK. use inode's group */
162 return ext4_inode_to_goal_block(inode);
166 * Allocation for a meta data block
169 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
170 struct ext4_ext_path *path,
171 struct ext4_extent *ex, int *err, unsigned int flags)
173 ext4_fsblk_t goal, newblock;
175 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
176 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
181 static inline int ext4_ext_space_block(struct inode *inode, int check)
185 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
186 / sizeof(struct ext4_extent);
188 #ifdef AGGRESSIVE_TEST
196 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
200 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
201 / sizeof(struct ext4_extent_idx);
203 #ifdef AGGRESSIVE_TEST
211 static inline int ext4_ext_space_root(struct inode *inode, int check)
215 size = sizeof(EXT4_I(inode)->i_data);
216 size -= sizeof(struct ext4_extent_header);
217 size /= sizeof(struct ext4_extent);
219 #ifdef AGGRESSIVE_TEST
227 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
231 size = sizeof(EXT4_I(inode)->i_data);
232 size -= sizeof(struct ext4_extent_header);
233 size /= sizeof(struct ext4_extent_idx);
235 #ifdef AGGRESSIVE_TEST
244 * Calculate the number of metadata blocks needed
245 * to allocate @blocks
246 * Worse case is one block per extent
248 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
250 struct ext4_inode_info *ei = EXT4_I(inode);
253 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
254 / sizeof(struct ext4_extent_idx));
257 * If the new delayed allocation block is contiguous with the
258 * previous da block, it can share index blocks with the
259 * previous block, so we only need to allocate a new index
260 * block every idxs leaf blocks. At ldxs**2 blocks, we need
261 * an additional index block, and at ldxs**3 blocks, yet
262 * another index blocks.
264 if (ei->i_da_metadata_calc_len &&
265 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
266 if ((ei->i_da_metadata_calc_len % idxs) == 0)
268 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
270 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
272 ei->i_da_metadata_calc_len = 0;
274 ei->i_da_metadata_calc_len++;
275 ei->i_da_metadata_calc_last_lblock++;
280 * In the worst case we need a new set of index blocks at
281 * every level of the inode's extent tree.
283 ei->i_da_metadata_calc_len = 1;
284 ei->i_da_metadata_calc_last_lblock = lblock;
285 return ext_depth(inode) + 1;
289 ext4_ext_max_entries(struct inode *inode, int depth)
293 if (depth == ext_depth(inode)) {
295 max = ext4_ext_space_root(inode, 1);
297 max = ext4_ext_space_root_idx(inode, 1);
300 max = ext4_ext_space_block(inode, 1);
302 max = ext4_ext_space_block_idx(inode, 1);
308 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
310 ext4_fsblk_t block = ext4_ext_pblock(ext);
311 int len = ext4_ext_get_actual_len(ext);
313 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
316 static int ext4_valid_extent_idx(struct inode *inode,
317 struct ext4_extent_idx *ext_idx)
319 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
321 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
324 static int ext4_valid_extent_entries(struct inode *inode,
325 struct ext4_extent_header *eh,
328 struct ext4_extent *ext;
329 struct ext4_extent_idx *ext_idx;
330 unsigned short entries;
331 if (eh->eh_entries == 0)
334 entries = le16_to_cpu(eh->eh_entries);
338 ext = EXT_FIRST_EXTENT(eh);
340 if (!ext4_valid_extent(inode, ext))
346 ext_idx = EXT_FIRST_INDEX(eh);
348 if (!ext4_valid_extent_idx(inode, ext_idx))
357 static int __ext4_ext_check(const char *function, unsigned int line,
358 struct inode *inode, struct ext4_extent_header *eh,
361 const char *error_msg;
364 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
365 error_msg = "invalid magic";
368 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
369 error_msg = "unexpected eh_depth";
372 if (unlikely(eh->eh_max == 0)) {
373 error_msg = "invalid eh_max";
376 max = ext4_ext_max_entries(inode, depth);
377 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
378 error_msg = "too large eh_max";
381 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
382 error_msg = "invalid eh_entries";
385 if (!ext4_valid_extent_entries(inode, eh, depth)) {
386 error_msg = "invalid extent entries";
392 ext4_error_inode(inode, function, line, 0,
393 "bad header/extent: %s - magic %x, "
394 "entries %u, max %u(%u), depth %u(%u)",
395 error_msg, le16_to_cpu(eh->eh_magic),
396 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
397 max, le16_to_cpu(eh->eh_depth), depth);
402 #define ext4_ext_check(inode, eh, depth) \
403 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
405 int ext4_ext_check_inode(struct inode *inode)
407 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
411 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
413 int k, l = path->p_depth;
416 for (k = 0; k <= l; k++, path++) {
418 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
419 ext4_idx_pblock(path->p_idx));
420 } else if (path->p_ext) {
421 ext_debug(" %d:[%d]%d:%llu ",
422 le32_to_cpu(path->p_ext->ee_block),
423 ext4_ext_is_uninitialized(path->p_ext),
424 ext4_ext_get_actual_len(path->p_ext),
425 ext4_ext_pblock(path->p_ext));
432 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
434 int depth = ext_depth(inode);
435 struct ext4_extent_header *eh;
436 struct ext4_extent *ex;
442 eh = path[depth].p_hdr;
443 ex = EXT_FIRST_EXTENT(eh);
445 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
447 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
448 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
449 ext4_ext_is_uninitialized(ex),
450 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
455 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
456 ext4_fsblk_t newblock, int level)
458 int depth = ext_depth(inode);
459 struct ext4_extent *ex;
461 if (depth != level) {
462 struct ext4_extent_idx *idx;
463 idx = path[level].p_idx;
464 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
465 ext_debug("%d: move %d:%llu in new index %llu\n", level,
466 le32_to_cpu(idx->ei_block),
467 ext4_idx_pblock(idx),
475 ex = path[depth].p_ext;
476 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
477 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
478 le32_to_cpu(ex->ee_block),
480 ext4_ext_is_uninitialized(ex),
481 ext4_ext_get_actual_len(ex),
488 #define ext4_ext_show_path(inode, path)
489 #define ext4_ext_show_leaf(inode, path)
490 #define ext4_ext_show_move(inode, path, newblock, level)
493 void ext4_ext_drop_refs(struct ext4_ext_path *path)
495 int depth = path->p_depth;
498 for (i = 0; i <= depth; i++, path++)
506 * ext4_ext_binsearch_idx:
507 * binary search for the closest index of the given block
508 * the header must be checked before calling this
511 ext4_ext_binsearch_idx(struct inode *inode,
512 struct ext4_ext_path *path, ext4_lblk_t block)
514 struct ext4_extent_header *eh = path->p_hdr;
515 struct ext4_extent_idx *r, *l, *m;
518 ext_debug("binsearch for %u(idx): ", block);
520 l = EXT_FIRST_INDEX(eh) + 1;
521 r = EXT_LAST_INDEX(eh);
524 if (block < le32_to_cpu(m->ei_block))
528 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
529 m, le32_to_cpu(m->ei_block),
530 r, le32_to_cpu(r->ei_block));
534 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
535 ext4_idx_pblock(path->p_idx));
537 #ifdef CHECK_BINSEARCH
539 struct ext4_extent_idx *chix, *ix;
542 chix = ix = EXT_FIRST_INDEX(eh);
543 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
545 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
546 printk(KERN_DEBUG "k=%d, ix=0x%p, "
548 ix, EXT_FIRST_INDEX(eh));
549 printk(KERN_DEBUG "%u <= %u\n",
550 le32_to_cpu(ix->ei_block),
551 le32_to_cpu(ix[-1].ei_block));
553 BUG_ON(k && le32_to_cpu(ix->ei_block)
554 <= le32_to_cpu(ix[-1].ei_block));
555 if (block < le32_to_cpu(ix->ei_block))
559 BUG_ON(chix != path->p_idx);
566 * ext4_ext_binsearch:
567 * binary search for closest extent of the given block
568 * the header must be checked before calling this
571 ext4_ext_binsearch(struct inode *inode,
572 struct ext4_ext_path *path, ext4_lblk_t block)
574 struct ext4_extent_header *eh = path->p_hdr;
575 struct ext4_extent *r, *l, *m;
577 if (eh->eh_entries == 0) {
579 * this leaf is empty:
580 * we get such a leaf in split/add case
585 ext_debug("binsearch for %u: ", block);
587 l = EXT_FIRST_EXTENT(eh) + 1;
588 r = EXT_LAST_EXTENT(eh);
592 if (block < le32_to_cpu(m->ee_block))
596 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
597 m, le32_to_cpu(m->ee_block),
598 r, le32_to_cpu(r->ee_block));
602 ext_debug(" -> %d:%llu:[%d]%d ",
603 le32_to_cpu(path->p_ext->ee_block),
604 ext4_ext_pblock(path->p_ext),
605 ext4_ext_is_uninitialized(path->p_ext),
606 ext4_ext_get_actual_len(path->p_ext));
608 #ifdef CHECK_BINSEARCH
610 struct ext4_extent *chex, *ex;
613 chex = ex = EXT_FIRST_EXTENT(eh);
614 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
615 BUG_ON(k && le32_to_cpu(ex->ee_block)
616 <= le32_to_cpu(ex[-1].ee_block));
617 if (block < le32_to_cpu(ex->ee_block))
621 BUG_ON(chex != path->p_ext);
627 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
629 struct ext4_extent_header *eh;
631 eh = ext_inode_hdr(inode);
634 eh->eh_magic = EXT4_EXT_MAGIC;
635 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
636 ext4_mark_inode_dirty(handle, inode);
637 ext4_ext_invalidate_cache(inode);
641 struct ext4_ext_path *
642 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
643 struct ext4_ext_path *path)
645 struct ext4_extent_header *eh;
646 struct buffer_head *bh;
647 short int depth, i, ppos = 0, alloc = 0;
649 eh = ext_inode_hdr(inode);
650 depth = ext_depth(inode);
652 /* account possible depth increase */
654 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
657 return ERR_PTR(-ENOMEM);
664 /* walk through the tree */
666 int need_to_validate = 0;
668 ext_debug("depth %d: num %d, max %d\n",
669 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
671 ext4_ext_binsearch_idx(inode, path + ppos, block);
672 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
673 path[ppos].p_depth = i;
674 path[ppos].p_ext = NULL;
676 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
679 if (!bh_uptodate_or_lock(bh)) {
680 trace_ext4_ext_load_extent(inode, block,
682 if (bh_submit_read(bh) < 0) {
686 /* validate the extent entries */
687 need_to_validate = 1;
689 eh = ext_block_hdr(bh);
691 if (unlikely(ppos > depth)) {
693 EXT4_ERROR_INODE(inode,
694 "ppos %d > depth %d", ppos, depth);
697 path[ppos].p_bh = bh;
698 path[ppos].p_hdr = eh;
701 if (need_to_validate && ext4_ext_check(inode, eh, i))
705 path[ppos].p_depth = i;
706 path[ppos].p_ext = NULL;
707 path[ppos].p_idx = NULL;
710 ext4_ext_binsearch(inode, path + ppos, block);
711 /* if not an empty leaf */
712 if (path[ppos].p_ext)
713 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
715 ext4_ext_show_path(inode, path);
720 ext4_ext_drop_refs(path);
723 return ERR_PTR(-EIO);
727 * ext4_ext_insert_index:
728 * insert new index [@logical;@ptr] into the block at @curp;
729 * check where to insert: before @curp or after @curp
731 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
732 struct ext4_ext_path *curp,
733 int logical, ext4_fsblk_t ptr)
735 struct ext4_extent_idx *ix;
738 err = ext4_ext_get_access(handle, inode, curp);
742 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
743 EXT4_ERROR_INODE(inode,
744 "logical %d == ei_block %d!",
745 logical, le32_to_cpu(curp->p_idx->ei_block));
749 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
750 >= le16_to_cpu(curp->p_hdr->eh_max))) {
751 EXT4_ERROR_INODE(inode,
752 "eh_entries %d >= eh_max %d!",
753 le16_to_cpu(curp->p_hdr->eh_entries),
754 le16_to_cpu(curp->p_hdr->eh_max));
758 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
759 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
761 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
762 len = (len - 1) * sizeof(struct ext4_extent_idx);
763 len = len < 0 ? 0 : len;
764 ext_debug("insert new index %d after: %llu. "
765 "move %d from 0x%p to 0x%p\n",
767 (curp->p_idx + 1), (curp->p_idx + 2));
768 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
770 ix = curp->p_idx + 1;
773 len = len * sizeof(struct ext4_extent_idx);
774 len = len < 0 ? 0 : len;
775 ext_debug("insert new index %d before: %llu. "
776 "move %d from 0x%p to 0x%p\n",
778 curp->p_idx, (curp->p_idx + 1));
779 memmove(curp->p_idx + 1, curp->p_idx, len);
783 ix->ei_block = cpu_to_le32(logical);
784 ext4_idx_store_pblock(ix, ptr);
785 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
787 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
788 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
792 err = ext4_ext_dirty(handle, inode, curp);
793 ext4_std_error(inode->i_sb, err);
800 * inserts new subtree into the path, using free index entry
802 * - allocates all needed blocks (new leaf and all intermediate index blocks)
803 * - makes decision where to split
804 * - moves remaining extents and index entries (right to the split point)
805 * into the newly allocated blocks
806 * - initializes subtree
808 static int ext4_ext_split(handle_t *handle, struct inode *inode,
810 struct ext4_ext_path *path,
811 struct ext4_extent *newext, int at)
813 struct buffer_head *bh = NULL;
814 int depth = ext_depth(inode);
815 struct ext4_extent_header *neh;
816 struct ext4_extent_idx *fidx;
818 ext4_fsblk_t newblock, oldblock;
820 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
823 /* make decision: where to split? */
824 /* FIXME: now decision is simplest: at current extent */
826 /* if current leaf will be split, then we should use
827 * border from split point */
828 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
829 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
832 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
833 border = path[depth].p_ext[1].ee_block;
834 ext_debug("leaf will be split."
835 " next leaf starts at %d\n",
836 le32_to_cpu(border));
838 border = newext->ee_block;
839 ext_debug("leaf will be added."
840 " next leaf starts at %d\n",
841 le32_to_cpu(border));
845 * If error occurs, then we break processing
846 * and mark filesystem read-only. index won't
847 * be inserted and tree will be in consistent
848 * state. Next mount will repair buffers too.
852 * Get array to track all allocated blocks.
853 * We need this to handle errors and free blocks
856 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
860 /* allocate all needed blocks */
861 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
862 for (a = 0; a < depth - at; a++) {
863 newblock = ext4_ext_new_meta_block(handle, inode, path,
864 newext, &err, flags);
867 ablocks[a] = newblock;
870 /* initialize new leaf */
871 newblock = ablocks[--a];
872 if (unlikely(newblock == 0)) {
873 EXT4_ERROR_INODE(inode, "newblock == 0!");
877 bh = sb_getblk(inode->i_sb, newblock);
884 err = ext4_journal_get_create_access(handle, bh);
888 neh = ext_block_hdr(bh);
890 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
891 neh->eh_magic = EXT4_EXT_MAGIC;
894 /* move remainder of path[depth] to the new leaf */
895 if (unlikely(path[depth].p_hdr->eh_entries !=
896 path[depth].p_hdr->eh_max)) {
897 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
898 path[depth].p_hdr->eh_entries,
899 path[depth].p_hdr->eh_max);
903 /* start copy from next extent */
904 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
905 ext4_ext_show_move(inode, path, newblock, depth);
907 struct ext4_extent *ex;
908 ex = EXT_FIRST_EXTENT(neh);
909 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
910 le16_add_cpu(&neh->eh_entries, m);
913 set_buffer_uptodate(bh);
916 err = ext4_handle_dirty_metadata(handle, inode, bh);
922 /* correct old leaf */
924 err = ext4_ext_get_access(handle, inode, path + depth);
927 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
928 err = ext4_ext_dirty(handle, inode, path + depth);
934 /* create intermediate indexes */
936 if (unlikely(k < 0)) {
937 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
942 ext_debug("create %d intermediate indices\n", k);
943 /* insert new index into current index block */
944 /* current depth stored in i var */
948 newblock = ablocks[--a];
949 bh = sb_getblk(inode->i_sb, newblock);
956 err = ext4_journal_get_create_access(handle, bh);
960 neh = ext_block_hdr(bh);
961 neh->eh_entries = cpu_to_le16(1);
962 neh->eh_magic = EXT4_EXT_MAGIC;
963 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
964 neh->eh_depth = cpu_to_le16(depth - i);
965 fidx = EXT_FIRST_INDEX(neh);
966 fidx->ei_block = border;
967 ext4_idx_store_pblock(fidx, oldblock);
969 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
970 i, newblock, le32_to_cpu(border), oldblock);
972 /* move remainder of path[i] to the new index block */
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 /* start copy indexes */
982 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
983 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
984 EXT_MAX_INDEX(path[i].p_hdr));
985 ext4_ext_show_move(inode, path, newblock, i);
987 memmove(++fidx, path[i].p_idx,
988 sizeof(struct ext4_extent_idx) * m);
989 le16_add_cpu(&neh->eh_entries, m);
991 set_buffer_uptodate(bh);
994 err = ext4_handle_dirty_metadata(handle, inode, bh);
1000 /* correct old index */
1002 err = ext4_ext_get_access(handle, inode, path + i);
1005 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1006 err = ext4_ext_dirty(handle, inode, path + i);
1014 /* insert new index */
1015 err = ext4_ext_insert_index(handle, inode, path + at,
1016 le32_to_cpu(border), newblock);
1020 if (buffer_locked(bh))
1026 /* free all allocated blocks in error case */
1027 for (i = 0; i < depth; i++) {
1030 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1031 EXT4_FREE_BLOCKS_METADATA);
1040 * ext4_ext_grow_indepth:
1041 * implements tree growing procedure:
1042 * - allocates new block
1043 * - moves top-level data (index block or leaf) into the new block
1044 * - initializes new top-level, creating index that points to the
1045 * just created block
1047 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1049 struct ext4_ext_path *path,
1050 struct ext4_extent *newext)
1052 struct ext4_ext_path *curp = path;
1053 struct ext4_extent_header *neh;
1054 struct buffer_head *bh;
1055 ext4_fsblk_t newblock;
1058 newblock = ext4_ext_new_meta_block(handle, inode, path,
1059 newext, &err, flags);
1063 bh = sb_getblk(inode->i_sb, newblock);
1066 ext4_std_error(inode->i_sb, err);
1071 err = ext4_journal_get_create_access(handle, bh);
1077 /* move top-level index/leaf into new block */
1078 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1080 /* set size of new block */
1081 neh = ext_block_hdr(bh);
1082 /* old root could have indexes or leaves
1083 * so calculate e_max right way */
1084 if (ext_depth(inode))
1085 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1087 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1088 neh->eh_magic = EXT4_EXT_MAGIC;
1089 set_buffer_uptodate(bh);
1092 err = ext4_handle_dirty_metadata(handle, inode, bh);
1096 /* create index in new top-level index: num,max,pointer */
1097 err = ext4_ext_get_access(handle, inode, curp);
1101 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1102 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1103 curp->p_hdr->eh_entries = cpu_to_le16(1);
1104 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1106 if (path[0].p_hdr->eh_depth)
1107 curp->p_idx->ei_block =
1108 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1110 curp->p_idx->ei_block =
1111 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1112 ext4_idx_store_pblock(curp->p_idx, newblock);
1114 neh = ext_inode_hdr(inode);
1115 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1116 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1117 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1118 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1120 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1121 err = ext4_ext_dirty(handle, inode, curp);
1129 * ext4_ext_create_new_leaf:
1130 * finds empty index and adds new leaf.
1131 * if no free index is found, then it requests in-depth growing.
1133 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1135 struct ext4_ext_path *path,
1136 struct ext4_extent *newext)
1138 struct ext4_ext_path *curp;
1139 int depth, i, err = 0;
1142 i = depth = ext_depth(inode);
1144 /* walk up to the tree and look for free index entry */
1145 curp = path + depth;
1146 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1151 /* we use already allocated block for index block,
1152 * so subsequent data blocks should be contiguous */
1153 if (EXT_HAS_FREE_INDEX(curp)) {
1154 /* if we found index with free entry, then use that
1155 * entry: create all needed subtree and add new leaf */
1156 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1161 ext4_ext_drop_refs(path);
1162 path = ext4_ext_find_extent(inode,
1163 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1166 err = PTR_ERR(path);
1168 /* tree is full, time to grow in depth */
1169 err = ext4_ext_grow_indepth(handle, inode, flags,
1175 ext4_ext_drop_refs(path);
1176 path = ext4_ext_find_extent(inode,
1177 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1180 err = PTR_ERR(path);
1185 * only first (depth 0 -> 1) produces free space;
1186 * in all other cases we have to split the grown tree
1188 depth = ext_depth(inode);
1189 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1190 /* now we need to split */
1200 * search the closest allocated block to the left for *logical
1201 * and returns it at @logical + it's physical address at @phys
1202 * if *logical is the smallest allocated block, the function
1203 * returns 0 at @phys
1204 * return value contains 0 (success) or error code
1206 static int ext4_ext_search_left(struct inode *inode,
1207 struct ext4_ext_path *path,
1208 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1210 struct ext4_extent_idx *ix;
1211 struct ext4_extent *ex;
1214 if (unlikely(path == NULL)) {
1215 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1218 depth = path->p_depth;
1221 if (depth == 0 && path->p_ext == NULL)
1224 /* usually extent in the path covers blocks smaller
1225 * then *logical, but it can be that extent is the
1226 * first one in the file */
1228 ex = path[depth].p_ext;
1229 ee_len = ext4_ext_get_actual_len(ex);
1230 if (*logical < le32_to_cpu(ex->ee_block)) {
1231 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1232 EXT4_ERROR_INODE(inode,
1233 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1234 *logical, le32_to_cpu(ex->ee_block));
1237 while (--depth >= 0) {
1238 ix = path[depth].p_idx;
1239 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1240 EXT4_ERROR_INODE(inode,
1241 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1242 ix != NULL ? ix->ei_block : 0,
1243 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1244 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1252 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1253 EXT4_ERROR_INODE(inode,
1254 "logical %d < ee_block %d + ee_len %d!",
1255 *logical, le32_to_cpu(ex->ee_block), ee_len);
1259 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1260 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1265 * search the closest allocated block to the right for *logical
1266 * and returns it at @logical + it's physical address at @phys
1267 * if *logical is the smallest allocated block, the function
1268 * returns 0 at @phys
1269 * return value contains 0 (success) or error code
1271 static int ext4_ext_search_right(struct inode *inode,
1272 struct ext4_ext_path *path,
1273 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1274 struct ext4_extent **ret_ex)
1276 struct buffer_head *bh = NULL;
1277 struct ext4_extent_header *eh;
1278 struct ext4_extent_idx *ix;
1279 struct ext4_extent *ex;
1281 int depth; /* Note, NOT eh_depth; depth from top of tree */
1284 if (unlikely(path == NULL)) {
1285 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1288 depth = path->p_depth;
1291 if (depth == 0 && path->p_ext == NULL)
1294 /* usually extent in the path covers blocks smaller
1295 * then *logical, but it can be that extent is the
1296 * first one in the file */
1298 ex = path[depth].p_ext;
1299 ee_len = ext4_ext_get_actual_len(ex);
1300 if (*logical < le32_to_cpu(ex->ee_block)) {
1301 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1302 EXT4_ERROR_INODE(inode,
1303 "first_extent(path[%d].p_hdr) != ex",
1307 while (--depth >= 0) {
1308 ix = path[depth].p_idx;
1309 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1310 EXT4_ERROR_INODE(inode,
1311 "ix != EXT_FIRST_INDEX *logical %d!",
1319 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1320 EXT4_ERROR_INODE(inode,
1321 "logical %d < ee_block %d + ee_len %d!",
1322 *logical, le32_to_cpu(ex->ee_block), ee_len);
1326 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1327 /* next allocated block in this leaf */
1332 /* go up and search for index to the right */
1333 while (--depth >= 0) {
1334 ix = path[depth].p_idx;
1335 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1339 /* we've gone up to the root and found no index to the right */
1343 /* we've found index to the right, let's
1344 * follow it and find the closest allocated
1345 * block to the right */
1347 block = ext4_idx_pblock(ix);
1348 while (++depth < path->p_depth) {
1349 bh = sb_bread(inode->i_sb, block);
1352 eh = ext_block_hdr(bh);
1353 /* subtract from p_depth to get proper eh_depth */
1354 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1358 ix = EXT_FIRST_INDEX(eh);
1359 block = ext4_idx_pblock(ix);
1363 bh = sb_bread(inode->i_sb, block);
1366 eh = ext_block_hdr(bh);
1367 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1371 ex = EXT_FIRST_EXTENT(eh);
1373 *logical = le32_to_cpu(ex->ee_block);
1374 *phys = ext4_ext_pblock(ex);
1382 * ext4_ext_next_allocated_block:
1383 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1384 * NOTE: it considers block number from index entry as
1385 * allocated block. Thus, index entries have to be consistent
1389 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1393 BUG_ON(path == NULL);
1394 depth = path->p_depth;
1396 if (depth == 0 && path->p_ext == NULL)
1397 return EXT_MAX_BLOCKS;
1399 while (depth >= 0) {
1400 if (depth == path->p_depth) {
1402 if (path[depth].p_ext !=
1403 EXT_LAST_EXTENT(path[depth].p_hdr))
1404 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1407 if (path[depth].p_idx !=
1408 EXT_LAST_INDEX(path[depth].p_hdr))
1409 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1414 return EXT_MAX_BLOCKS;
1418 * ext4_ext_next_leaf_block:
1419 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1421 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1425 BUG_ON(path == NULL);
1426 depth = path->p_depth;
1428 /* zero-tree has no leaf blocks at all */
1430 return EXT_MAX_BLOCKS;
1432 /* go to index block */
1435 while (depth >= 0) {
1436 if (path[depth].p_idx !=
1437 EXT_LAST_INDEX(path[depth].p_hdr))
1438 return (ext4_lblk_t)
1439 le32_to_cpu(path[depth].p_idx[1].ei_block);
1443 return EXT_MAX_BLOCKS;
1447 * ext4_ext_correct_indexes:
1448 * if leaf gets modified and modified extent is first in the leaf,
1449 * then we have to correct all indexes above.
1450 * TODO: do we need to correct tree in all cases?
1452 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1453 struct ext4_ext_path *path)
1455 struct ext4_extent_header *eh;
1456 int depth = ext_depth(inode);
1457 struct ext4_extent *ex;
1461 eh = path[depth].p_hdr;
1462 ex = path[depth].p_ext;
1464 if (unlikely(ex == NULL || eh == NULL)) {
1465 EXT4_ERROR_INODE(inode,
1466 "ex %p == NULL or eh %p == NULL", ex, eh);
1471 /* there is no tree at all */
1475 if (ex != EXT_FIRST_EXTENT(eh)) {
1476 /* we correct tree if first leaf got modified only */
1481 * TODO: we need correction if border is smaller than current one
1484 border = path[depth].p_ext->ee_block;
1485 err = ext4_ext_get_access(handle, inode, path + k);
1488 path[k].p_idx->ei_block = border;
1489 err = ext4_ext_dirty(handle, inode, path + k);
1494 /* change all left-side indexes */
1495 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1497 err = ext4_ext_get_access(handle, inode, path + k);
1500 path[k].p_idx->ei_block = border;
1501 err = ext4_ext_dirty(handle, inode, path + k);
1510 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1511 struct ext4_extent *ex2)
1513 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1516 * Make sure that either both extents are uninitialized, or
1519 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1522 if (ext4_ext_is_uninitialized(ex1))
1523 max_len = EXT_UNINIT_MAX_LEN;
1525 max_len = EXT_INIT_MAX_LEN;
1527 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1528 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1530 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1531 le32_to_cpu(ex2->ee_block))
1535 * To allow future support for preallocated extents to be added
1536 * as an RO_COMPAT feature, refuse to merge to extents if
1537 * this can result in the top bit of ee_len being set.
1539 if (ext1_ee_len + ext2_ee_len > max_len)
1541 #ifdef AGGRESSIVE_TEST
1542 if (ext1_ee_len >= 4)
1546 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1552 * This function tries to merge the "ex" extent to the next extent in the tree.
1553 * It always tries to merge towards right. If you want to merge towards
1554 * left, pass "ex - 1" as argument instead of "ex".
1555 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1556 * 1 if they got merged.
1558 static int ext4_ext_try_to_merge_right(struct inode *inode,
1559 struct ext4_ext_path *path,
1560 struct ext4_extent *ex)
1562 struct ext4_extent_header *eh;
1563 unsigned int depth, len;
1565 int uninitialized = 0;
1567 depth = ext_depth(inode);
1568 BUG_ON(path[depth].p_hdr == NULL);
1569 eh = path[depth].p_hdr;
1571 while (ex < EXT_LAST_EXTENT(eh)) {
1572 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1574 /* merge with next extent! */
1575 if (ext4_ext_is_uninitialized(ex))
1577 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1578 + ext4_ext_get_actual_len(ex + 1));
1580 ext4_ext_mark_uninitialized(ex);
1582 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1583 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1584 * sizeof(struct ext4_extent);
1585 memmove(ex + 1, ex + 2, len);
1587 le16_add_cpu(&eh->eh_entries, -1);
1589 WARN_ON(eh->eh_entries == 0);
1590 if (!eh->eh_entries)
1591 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1598 * This function tries to merge the @ex extent to neighbours in the tree.
1599 * return 1 if merge left else 0.
1601 static int ext4_ext_try_to_merge(struct inode *inode,
1602 struct ext4_ext_path *path,
1603 struct ext4_extent *ex) {
1604 struct ext4_extent_header *eh;
1609 depth = ext_depth(inode);
1610 BUG_ON(path[depth].p_hdr == NULL);
1611 eh = path[depth].p_hdr;
1613 if (ex > EXT_FIRST_EXTENT(eh))
1614 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1617 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1623 * check if a portion of the "newext" extent overlaps with an
1626 * If there is an overlap discovered, it updates the length of the newext
1627 * such that there will be no overlap, and then returns 1.
1628 * If there is no overlap found, it returns 0.
1630 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1631 struct inode *inode,
1632 struct ext4_extent *newext,
1633 struct ext4_ext_path *path)
1636 unsigned int depth, len1;
1637 unsigned int ret = 0;
1639 b1 = le32_to_cpu(newext->ee_block);
1640 len1 = ext4_ext_get_actual_len(newext);
1641 depth = ext_depth(inode);
1642 if (!path[depth].p_ext)
1644 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1645 b2 &= ~(sbi->s_cluster_ratio - 1);
1648 * get the next allocated block if the extent in the path
1649 * is before the requested block(s)
1652 b2 = ext4_ext_next_allocated_block(path);
1653 if (b2 == EXT_MAX_BLOCKS)
1655 b2 &= ~(sbi->s_cluster_ratio - 1);
1658 /* check for wrap through zero on extent logical start block*/
1659 if (b1 + len1 < b1) {
1660 len1 = EXT_MAX_BLOCKS - b1;
1661 newext->ee_len = cpu_to_le16(len1);
1665 /* check for overlap */
1666 if (b1 + len1 > b2) {
1667 newext->ee_len = cpu_to_le16(b2 - b1);
1675 * ext4_ext_insert_extent:
1676 * tries to merge requsted extent into the existing extent or
1677 * inserts requested extent as new one into the tree,
1678 * creating new leaf in the no-space case.
1680 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1681 struct ext4_ext_path *path,
1682 struct ext4_extent *newext, int flag)
1684 struct ext4_extent_header *eh;
1685 struct ext4_extent *ex, *fex;
1686 struct ext4_extent *nearex; /* nearest extent */
1687 struct ext4_ext_path *npath = NULL;
1688 int depth, len, err;
1690 unsigned uninitialized = 0;
1693 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1694 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1697 depth = ext_depth(inode);
1698 ex = path[depth].p_ext;
1699 if (unlikely(path[depth].p_hdr == NULL)) {
1700 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1704 /* try to insert block into found extent and return */
1705 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1706 && ext4_can_extents_be_merged(inode, ex, newext)) {
1707 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1708 ext4_ext_is_uninitialized(newext),
1709 ext4_ext_get_actual_len(newext),
1710 le32_to_cpu(ex->ee_block),
1711 ext4_ext_is_uninitialized(ex),
1712 ext4_ext_get_actual_len(ex),
1713 ext4_ext_pblock(ex));
1714 err = ext4_ext_get_access(handle, inode, path + depth);
1719 * ext4_can_extents_be_merged should have checked that either
1720 * both extents are uninitialized, or both aren't. Thus we
1721 * need to check only one of them here.
1723 if (ext4_ext_is_uninitialized(ex))
1725 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1726 + ext4_ext_get_actual_len(newext));
1728 ext4_ext_mark_uninitialized(ex);
1729 eh = path[depth].p_hdr;
1734 depth = ext_depth(inode);
1735 eh = path[depth].p_hdr;
1736 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1739 /* probably next leaf has space for us? */
1740 fex = EXT_LAST_EXTENT(eh);
1741 next = EXT_MAX_BLOCKS;
1742 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1743 next = ext4_ext_next_leaf_block(path);
1744 if (next != EXT_MAX_BLOCKS) {
1745 ext_debug("next leaf block - %d\n", next);
1746 BUG_ON(npath != NULL);
1747 npath = ext4_ext_find_extent(inode, next, NULL);
1749 return PTR_ERR(npath);
1750 BUG_ON(npath->p_depth != path->p_depth);
1751 eh = npath[depth].p_hdr;
1752 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1753 ext_debug("next leaf isn't full(%d)\n",
1754 le16_to_cpu(eh->eh_entries));
1758 ext_debug("next leaf has no free space(%d,%d)\n",
1759 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1763 * There is no free space in the found leaf.
1764 * We're gonna add a new leaf in the tree.
1766 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1767 flags = EXT4_MB_USE_ROOT_BLOCKS;
1768 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1771 depth = ext_depth(inode);
1772 eh = path[depth].p_hdr;
1775 nearex = path[depth].p_ext;
1777 err = ext4_ext_get_access(handle, inode, path + depth);
1782 /* there is no extent in this leaf, create first one */
1783 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1784 le32_to_cpu(newext->ee_block),
1785 ext4_ext_pblock(newext),
1786 ext4_ext_is_uninitialized(newext),
1787 ext4_ext_get_actual_len(newext));
1788 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1789 } else if (le32_to_cpu(newext->ee_block)
1790 > le32_to_cpu(nearex->ee_block)) {
1791 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1792 if (nearex != EXT_LAST_EXTENT(eh)) {
1793 len = EXT_MAX_EXTENT(eh) - nearex;
1794 len = (len - 1) * sizeof(struct ext4_extent);
1795 len = len < 0 ? 0 : len;
1796 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1797 "move %d from 0x%p to 0x%p\n",
1798 le32_to_cpu(newext->ee_block),
1799 ext4_ext_pblock(newext),
1800 ext4_ext_is_uninitialized(newext),
1801 ext4_ext_get_actual_len(newext),
1802 nearex, len, nearex + 1, nearex + 2);
1803 memmove(nearex + 2, nearex + 1, len);
1805 path[depth].p_ext = nearex + 1;
1807 BUG_ON(newext->ee_block == nearex->ee_block);
1808 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1809 len = len < 0 ? 0 : len;
1810 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1811 "move %d from 0x%p to 0x%p\n",
1812 le32_to_cpu(newext->ee_block),
1813 ext4_ext_pblock(newext),
1814 ext4_ext_is_uninitialized(newext),
1815 ext4_ext_get_actual_len(newext),
1816 nearex, len, nearex, nearex + 1);
1817 memmove(nearex + 1, nearex, len);
1818 path[depth].p_ext = nearex;
1821 le16_add_cpu(&eh->eh_entries, 1);
1822 nearex = path[depth].p_ext;
1823 nearex->ee_block = newext->ee_block;
1824 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1825 nearex->ee_len = newext->ee_len;
1828 /* try to merge extents to the right */
1829 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1830 ext4_ext_try_to_merge(inode, path, nearex);
1832 /* try to merge extents to the left */
1834 /* time to correct all indexes above */
1835 err = ext4_ext_correct_indexes(handle, inode, path);
1839 err = ext4_ext_dirty(handle, inode, path + depth);
1843 ext4_ext_drop_refs(npath);
1846 ext4_ext_invalidate_cache(inode);
1850 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1851 ext4_lblk_t num, ext_prepare_callback func,
1854 struct ext4_ext_path *path = NULL;
1855 struct ext4_ext_cache cbex;
1856 struct ext4_extent *ex;
1857 ext4_lblk_t next, start = 0, end = 0;
1858 ext4_lblk_t last = block + num;
1859 int depth, exists, err = 0;
1861 BUG_ON(func == NULL);
1862 BUG_ON(inode == NULL);
1864 while (block < last && block != EXT_MAX_BLOCKS) {
1866 /* find extent for this block */
1867 down_read(&EXT4_I(inode)->i_data_sem);
1868 path = ext4_ext_find_extent(inode, block, path);
1869 up_read(&EXT4_I(inode)->i_data_sem);
1871 err = PTR_ERR(path);
1876 depth = ext_depth(inode);
1877 if (unlikely(path[depth].p_hdr == NULL)) {
1878 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1882 ex = path[depth].p_ext;
1883 next = ext4_ext_next_allocated_block(path);
1887 /* there is no extent yet, so try to allocate
1888 * all requested space */
1891 } else if (le32_to_cpu(ex->ee_block) > block) {
1892 /* need to allocate space before found extent */
1894 end = le32_to_cpu(ex->ee_block);
1895 if (block + num < end)
1897 } else if (block >= le32_to_cpu(ex->ee_block)
1898 + ext4_ext_get_actual_len(ex)) {
1899 /* need to allocate space after found extent */
1904 } else if (block >= le32_to_cpu(ex->ee_block)) {
1906 * some part of requested space is covered
1910 end = le32_to_cpu(ex->ee_block)
1911 + ext4_ext_get_actual_len(ex);
1912 if (block + num < end)
1918 BUG_ON(end <= start);
1921 cbex.ec_block = start;
1922 cbex.ec_len = end - start;
1925 cbex.ec_block = le32_to_cpu(ex->ee_block);
1926 cbex.ec_len = ext4_ext_get_actual_len(ex);
1927 cbex.ec_start = ext4_ext_pblock(ex);
1930 if (unlikely(cbex.ec_len == 0)) {
1931 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1935 err = func(inode, next, &cbex, ex, cbdata);
1936 ext4_ext_drop_refs(path);
1941 if (err == EXT_REPEAT)
1943 else if (err == EXT_BREAK) {
1948 if (ext_depth(inode) != depth) {
1949 /* depth was changed. we have to realloc path */
1954 block = cbex.ec_block + cbex.ec_len;
1958 ext4_ext_drop_refs(path);
1966 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1967 __u32 len, ext4_fsblk_t start)
1969 struct ext4_ext_cache *cex;
1971 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1972 cex = &EXT4_I(inode)->i_cached_extent;
1973 cex->ec_block = block;
1975 cex->ec_start = start;
1976 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1980 * ext4_ext_put_gap_in_cache:
1981 * calculate boundaries of the gap that the requested block fits into
1982 * and cache this gap
1985 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1988 int depth = ext_depth(inode);
1991 struct ext4_extent *ex;
1993 ex = path[depth].p_ext;
1995 /* there is no extent yet, so gap is [0;-] */
1997 len = EXT_MAX_BLOCKS;
1998 ext_debug("cache gap(whole file):");
1999 } else if (block < le32_to_cpu(ex->ee_block)) {
2001 len = le32_to_cpu(ex->ee_block) - block;
2002 ext_debug("cache gap(before): %u [%u:%u]",
2004 le32_to_cpu(ex->ee_block),
2005 ext4_ext_get_actual_len(ex));
2006 } else if (block >= le32_to_cpu(ex->ee_block)
2007 + ext4_ext_get_actual_len(ex)) {
2009 lblock = le32_to_cpu(ex->ee_block)
2010 + ext4_ext_get_actual_len(ex);
2012 next = ext4_ext_next_allocated_block(path);
2013 ext_debug("cache gap(after): [%u:%u] %u",
2014 le32_to_cpu(ex->ee_block),
2015 ext4_ext_get_actual_len(ex),
2017 BUG_ON(next == lblock);
2018 len = next - lblock;
2024 ext_debug(" -> %u:%lu\n", lblock, len);
2025 ext4_ext_put_in_cache(inode, lblock, len, 0);
2029 * ext4_ext_check_cache()
2030 * Checks to see if the given block is in the cache.
2031 * If it is, the cached extent is stored in the given
2032 * cache extent pointer. If the cached extent is a hole,
2033 * this routine should be used instead of
2034 * ext4_ext_in_cache if the calling function needs to
2035 * know the size of the hole.
2037 * @inode: The files inode
2038 * @block: The block to look for in the cache
2039 * @ex: Pointer where the cached extent will be stored
2040 * if it contains block
2042 * Return 0 if cache is invalid; 1 if the cache is valid
2044 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2045 struct ext4_ext_cache *ex){
2046 struct ext4_ext_cache *cex;
2047 struct ext4_sb_info *sbi;
2051 * We borrow i_block_reservation_lock to protect i_cached_extent
2053 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2054 cex = &EXT4_I(inode)->i_cached_extent;
2055 sbi = EXT4_SB(inode->i_sb);
2057 /* has cache valid data? */
2058 if (cex->ec_len == 0)
2061 if (in_range(block, cex->ec_block, cex->ec_len)) {
2062 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2063 ext_debug("%u cached by %u:%u:%llu\n",
2065 cex->ec_block, cex->ec_len, cex->ec_start);
2070 sbi->extent_cache_misses++;
2072 sbi->extent_cache_hits++;
2073 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2078 * ext4_ext_in_cache()
2079 * Checks to see if the given block is in the cache.
2080 * If it is, the cached extent is stored in the given
2083 * @inode: The files inode
2084 * @block: The block to look for in the cache
2085 * @ex: Pointer where the cached extent will be stored
2086 * if it contains block
2088 * Return 0 if cache is invalid; 1 if the cache is valid
2091 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2092 struct ext4_extent *ex)
2094 struct ext4_ext_cache cex;
2097 if (ext4_ext_check_cache(inode, block, &cex)) {
2098 ex->ee_block = cpu_to_le32(cex.ec_block);
2099 ext4_ext_store_pblock(ex, cex.ec_start);
2100 ex->ee_len = cpu_to_le16(cex.ec_len);
2110 * removes index from the index block.
2112 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2113 struct ext4_ext_path *path)
2118 /* free index block */
2120 leaf = ext4_idx_pblock(path->p_idx);
2121 if (unlikely(path->p_hdr->eh_entries == 0)) {
2122 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2125 err = ext4_ext_get_access(handle, inode, path);
2129 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2130 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2131 len *= sizeof(struct ext4_extent_idx);
2132 memmove(path->p_idx, path->p_idx + 1, len);
2135 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2136 err = ext4_ext_dirty(handle, inode, path);
2139 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2140 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2141 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2146 * ext4_ext_calc_credits_for_single_extent:
2147 * This routine returns max. credits that needed to insert an extent
2148 * to the extent tree.
2149 * When pass the actual path, the caller should calculate credits
2152 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2153 struct ext4_ext_path *path)
2156 int depth = ext_depth(inode);
2159 /* probably there is space in leaf? */
2160 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2161 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2164 * There are some space in the leaf tree, no
2165 * need to account for leaf block credit
2167 * bitmaps and block group descriptor blocks
2168 * and other metadat blocks still need to be
2171 /* 1 bitmap, 1 block group descriptor */
2172 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2177 return ext4_chunk_trans_blocks(inode, nrblocks);
2181 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2183 * if nrblocks are fit in a single extent (chunk flag is 1), then
2184 * in the worse case, each tree level index/leaf need to be changed
2185 * if the tree split due to insert a new extent, then the old tree
2186 * index/leaf need to be updated too
2188 * If the nrblocks are discontiguous, they could cause
2189 * the whole tree split more than once, but this is really rare.
2191 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2194 int depth = ext_depth(inode);
2204 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2205 struct ext4_extent *ex,
2206 ext4_fsblk_t *partial_cluster,
2207 ext4_lblk_t from, ext4_lblk_t to)
2209 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2210 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2212 int flags = EXT4_FREE_BLOCKS_FORGET;
2214 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2215 flags |= EXT4_FREE_BLOCKS_METADATA;
2217 * For bigalloc file systems, we never free a partial cluster
2218 * at the beginning of the extent. Instead, we make a note
2219 * that we tried freeing the cluster, and check to see if we
2220 * need to free it on a subsequent call to ext4_remove_blocks,
2221 * or at the end of the ext4_truncate() operation.
2223 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2226 * If we have a partial cluster, and it's different from the
2227 * cluster of the last block, we need to explicitly free the
2228 * partial cluster here.
2230 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2231 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2232 ext4_free_blocks(handle, inode, NULL,
2233 EXT4_C2B(sbi, *partial_cluster),
2234 sbi->s_cluster_ratio, flags);
2235 *partial_cluster = 0;
2238 #ifdef EXTENTS_STATS
2240 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2241 spin_lock(&sbi->s_ext_stats_lock);
2242 sbi->s_ext_blocks += ee_len;
2243 sbi->s_ext_extents++;
2244 if (ee_len < sbi->s_ext_min)
2245 sbi->s_ext_min = ee_len;
2246 if (ee_len > sbi->s_ext_max)
2247 sbi->s_ext_max = ee_len;
2248 if (ext_depth(inode) > sbi->s_depth_max)
2249 sbi->s_depth_max = ext_depth(inode);
2250 spin_unlock(&sbi->s_ext_stats_lock);
2253 if (from >= le32_to_cpu(ex->ee_block)
2254 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2258 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2259 pblk = ext4_ext_pblock(ex) + ee_len - num;
2260 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2261 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2263 * If the block range to be freed didn't start at the
2264 * beginning of a cluster, and we removed the entire
2265 * extent, save the partial cluster here, since we
2266 * might need to delete if we determine that the
2267 * truncate operation has removed all of the blocks in
2270 if (pblk & (sbi->s_cluster_ratio - 1) &&
2272 *partial_cluster = EXT4_B2C(sbi, pblk);
2274 *partial_cluster = 0;
2275 } else if (from == le32_to_cpu(ex->ee_block)
2276 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2282 start = ext4_ext_pblock(ex);
2284 ext_debug("free first %u blocks starting %llu\n", num, start);
2285 ext4_free_blocks(handle, inode, 0, start, num, flags);
2288 printk(KERN_INFO "strange request: removal(2) "
2289 "%u-%u from %u:%u\n",
2290 from, to, le32_to_cpu(ex->ee_block), ee_len);
2297 * ext4_ext_rm_leaf() Removes the extents associated with the
2298 * blocks appearing between "start" and "end", and splits the extents
2299 * if "start" and "end" appear in the same extent
2301 * @handle: The journal handle
2302 * @inode: The files inode
2303 * @path: The path to the leaf
2304 * @start: The first block to remove
2305 * @end: The last block to remove
2308 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2309 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2310 ext4_lblk_t start, ext4_lblk_t end)
2312 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2313 int err = 0, correct_index = 0;
2314 int depth = ext_depth(inode), credits;
2315 struct ext4_extent_header *eh;
2316 ext4_lblk_t a, b, block;
2318 ext4_lblk_t ex_ee_block;
2319 unsigned short ex_ee_len;
2320 unsigned uninitialized = 0;
2321 struct ext4_extent *ex;
2322 struct ext4_map_blocks map;
2324 /* the header must be checked already in ext4_ext_remove_space() */
2325 ext_debug("truncate since %u in leaf\n", start);
2326 if (!path[depth].p_hdr)
2327 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2328 eh = path[depth].p_hdr;
2329 if (unlikely(path[depth].p_hdr == NULL)) {
2330 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2333 /* find where to start removing */
2334 ex = EXT_LAST_EXTENT(eh);
2336 ex_ee_block = le32_to_cpu(ex->ee_block);
2337 ex_ee_len = ext4_ext_get_actual_len(ex);
2339 while (ex >= EXT_FIRST_EXTENT(eh) &&
2340 ex_ee_block + ex_ee_len > start) {
2342 if (ext4_ext_is_uninitialized(ex))
2347 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2348 uninitialized, ex_ee_len);
2349 path[depth].p_ext = ex;
2351 a = ex_ee_block > start ? ex_ee_block : start;
2352 b = ex_ee_block+ex_ee_len - 1 < end ?
2353 ex_ee_block+ex_ee_len - 1 : end;
2355 ext_debug(" border %u:%u\n", a, b);
2357 /* If this extent is beyond the end of the hole, skip it */
2358 if (end <= ex_ee_block) {
2360 ex_ee_block = le32_to_cpu(ex->ee_block);
2361 ex_ee_len = ext4_ext_get_actual_len(ex);
2363 } else if (a != ex_ee_block &&
2364 b != ex_ee_block + ex_ee_len - 1) {
2366 * If this is a truncate, then this condition should
2367 * never happen because at least one of the end points
2368 * needs to be on the edge of the extent.
2370 if (end == EXT_MAX_BLOCKS - 1) {
2371 ext_debug(" bad truncate %u:%u\n",
2379 * else this is a hole punch, so the extent needs to
2380 * be split since neither edge of the hole is on the
2384 map.m_pblk = ext4_ext_pblock(ex);
2385 map.m_lblk = ex_ee_block;
2386 map.m_len = b - ex_ee_block;
2388 err = ext4_split_extent(handle,
2389 inode, path, &map, 0,
2390 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2391 EXT4_GET_BLOCKS_PRE_IO);
2396 ex_ee_len = ext4_ext_get_actual_len(ex);
2398 b = ex_ee_block+ex_ee_len - 1 < end ?
2399 ex_ee_block+ex_ee_len - 1 : end;
2401 /* Then remove tail of this extent */
2402 block = ex_ee_block;
2405 } else if (a != ex_ee_block) {
2406 /* remove tail of the extent */
2407 block = ex_ee_block;
2409 } else if (b != ex_ee_block + ex_ee_len - 1) {
2410 /* remove head of the extent */
2412 num = ex_ee_block + ex_ee_len - b;
2415 * If this is a truncate, this condition
2416 * should never happen
2418 if (end == EXT_MAX_BLOCKS - 1) {
2419 ext_debug(" bad truncate %u:%u\n",
2425 /* remove whole extent: excellent! */
2426 block = ex_ee_block;
2428 if (a != ex_ee_block) {
2429 ext_debug(" bad truncate %u:%u\n",
2435 if (b != ex_ee_block + ex_ee_len - 1) {
2436 ext_debug(" bad truncate %u:%u\n",
2444 * 3 for leaf, sb, and inode plus 2 (bmap and group
2445 * descriptor) for each block group; assume two block
2446 * groups plus ex_ee_len/blocks_per_block_group for
2449 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2450 if (ex == EXT_FIRST_EXTENT(eh)) {
2452 credits += (ext_depth(inode)) + 1;
2454 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2456 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2460 err = ext4_ext_get_access(handle, inode, path + depth);
2464 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2470 /* this extent is removed; mark slot entirely unused */
2471 ext4_ext_store_pblock(ex, 0);
2472 } else if (block != ex_ee_block) {
2474 * If this was a head removal, then we need to update
2475 * the physical block since it is now at a different
2478 ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
2481 ex->ee_block = cpu_to_le32(block);
2482 ex->ee_len = cpu_to_le16(num);
2484 * Do not mark uninitialized if all the blocks in the
2485 * extent have been removed.
2487 if (uninitialized && num)
2488 ext4_ext_mark_uninitialized(ex);
2490 err = ext4_ext_dirty(handle, inode, path + depth);
2495 * If the extent was completely released,
2496 * we need to remove it from the leaf
2499 if (end != EXT_MAX_BLOCKS - 1) {
2501 * For hole punching, we need to scoot all the
2502 * extents up when an extent is removed so that
2503 * we dont have blank extents in the middle
2505 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2506 sizeof(struct ext4_extent));
2508 /* Now get rid of the one at the end */
2509 memset(EXT_LAST_EXTENT(eh), 0,
2510 sizeof(struct ext4_extent));
2512 le16_add_cpu(&eh->eh_entries, -1);
2514 *partial_cluster = 0;
2516 ext_debug("new extent: %u:%u:%llu\n", block, num,
2517 ext4_ext_pblock(ex));
2519 ex_ee_block = le32_to_cpu(ex->ee_block);
2520 ex_ee_len = ext4_ext_get_actual_len(ex);
2523 if (correct_index && eh->eh_entries)
2524 err = ext4_ext_correct_indexes(handle, inode, path);
2527 * If there is still a entry in the leaf node, check to see if
2528 * it references the partial cluster. This is the only place
2529 * where it could; if it doesn't, we can free the cluster.
2531 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2532 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2533 *partial_cluster)) {
2534 int flags = EXT4_FREE_BLOCKS_FORGET;
2536 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2537 flags |= EXT4_FREE_BLOCKS_METADATA;
2539 ext4_free_blocks(handle, inode, NULL,
2540 EXT4_C2B(sbi, *partial_cluster),
2541 sbi->s_cluster_ratio, flags);
2542 *partial_cluster = 0;
2545 /* if this leaf is free, then we should
2546 * remove it from index block above */
2547 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2548 err = ext4_ext_rm_idx(handle, inode, path + depth);
2555 * ext4_ext_more_to_rm:
2556 * returns 1 if current index has to be freed (even partial)
2559 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2561 BUG_ON(path->p_idx == NULL);
2563 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2567 * if truncate on deeper level happened, it wasn't partial,
2568 * so we have to consider current index for truncation
2570 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2575 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2577 struct super_block *sb = inode->i_sb;
2578 int depth = ext_depth(inode);
2579 struct ext4_ext_path *path;
2580 ext4_fsblk_t partial_cluster = 0;
2584 ext_debug("truncate since %u\n", start);
2586 /* probably first extent we're gonna free will be last in block */
2587 handle = ext4_journal_start(inode, depth + 1);
2589 return PTR_ERR(handle);
2592 ext4_ext_invalidate_cache(inode);
2595 * We start scanning from right side, freeing all the blocks
2596 * after i_size and walking into the tree depth-wise.
2598 depth = ext_depth(inode);
2599 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2601 ext4_journal_stop(handle);
2604 path[0].p_depth = depth;
2605 path[0].p_hdr = ext_inode_hdr(inode);
2606 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2612 while (i >= 0 && err == 0) {
2614 /* this is leaf block */
2615 err = ext4_ext_rm_leaf(handle, inode, path,
2616 &partial_cluster, start,
2617 EXT_MAX_BLOCKS - 1);
2618 /* root level has p_bh == NULL, brelse() eats this */
2619 brelse(path[i].p_bh);
2620 path[i].p_bh = NULL;
2625 /* this is index block */
2626 if (!path[i].p_hdr) {
2627 ext_debug("initialize header\n");
2628 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2631 if (!path[i].p_idx) {
2632 /* this level hasn't been touched yet */
2633 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2634 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2635 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2637 le16_to_cpu(path[i].p_hdr->eh_entries));
2639 /* we were already here, see at next index */
2643 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2644 i, EXT_FIRST_INDEX(path[i].p_hdr),
2646 if (ext4_ext_more_to_rm(path + i)) {
2647 struct buffer_head *bh;
2648 /* go to the next level */
2649 ext_debug("move to level %d (block %llu)\n",
2650 i + 1, ext4_idx_pblock(path[i].p_idx));
2651 memset(path + i + 1, 0, sizeof(*path));
2652 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2654 /* should we reset i_size? */
2658 if (WARN_ON(i + 1 > depth)) {
2662 if (ext4_ext_check(inode, ext_block_hdr(bh),
2667 path[i + 1].p_bh = bh;
2669 /* save actual number of indexes since this
2670 * number is changed at the next iteration */
2671 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2674 /* we finished processing this index, go up */
2675 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2676 /* index is empty, remove it;
2677 * handle must be already prepared by the
2678 * truncatei_leaf() */
2679 err = ext4_ext_rm_idx(handle, inode, path + i);
2681 /* root level has p_bh == NULL, brelse() eats this */
2682 brelse(path[i].p_bh);
2683 path[i].p_bh = NULL;
2685 ext_debug("return to level %d\n", i);
2689 /* TODO: flexible tree reduction should be here */
2690 if (path->p_hdr->eh_entries == 0) {
2692 * truncate to zero freed all the tree,
2693 * so we need to correct eh_depth
2695 err = ext4_ext_get_access(handle, inode, path);
2697 ext_inode_hdr(inode)->eh_depth = 0;
2698 ext_inode_hdr(inode)->eh_max =
2699 cpu_to_le16(ext4_ext_space_root(inode, 0));
2700 err = ext4_ext_dirty(handle, inode, path);
2704 ext4_ext_drop_refs(path);
2708 ext4_journal_stop(handle);
2714 * called at mount time
2716 void ext4_ext_init(struct super_block *sb)
2719 * possible initialization would be here
2722 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2723 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2724 printk(KERN_INFO "EXT4-fs: file extents enabled");
2725 #ifdef AGGRESSIVE_TEST
2726 printk(", aggressive tests");
2728 #ifdef CHECK_BINSEARCH
2729 printk(", check binsearch");
2731 #ifdef EXTENTS_STATS
2736 #ifdef EXTENTS_STATS
2737 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2738 EXT4_SB(sb)->s_ext_min = 1 << 30;
2739 EXT4_SB(sb)->s_ext_max = 0;
2745 * called at umount time
2747 void ext4_ext_release(struct super_block *sb)
2749 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2752 #ifdef EXTENTS_STATS
2753 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2754 struct ext4_sb_info *sbi = EXT4_SB(sb);
2755 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2756 sbi->s_ext_blocks, sbi->s_ext_extents,
2757 sbi->s_ext_blocks / sbi->s_ext_extents);
2758 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2759 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2764 /* FIXME!! we need to try to merge to left or right after zero-out */
2765 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2767 ext4_fsblk_t ee_pblock;
2768 unsigned int ee_len;
2771 ee_len = ext4_ext_get_actual_len(ex);
2772 ee_pblock = ext4_ext_pblock(ex);
2774 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2782 * used by extent splitting.
2784 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2786 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2787 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2790 * ext4_split_extent_at() splits an extent at given block.
2792 * @handle: the journal handle
2793 * @inode: the file inode
2794 * @path: the path to the extent
2795 * @split: the logical block where the extent is splitted.
2796 * @split_flags: indicates if the extent could be zeroout if split fails, and
2797 * the states(init or uninit) of new extents.
2798 * @flags: flags used to insert new extent to extent tree.
2801 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2802 * of which are deterimined by split_flag.
2804 * There are two cases:
2805 * a> the extent are splitted into two extent.
2806 * b> split is not needed, and just mark the extent.
2808 * return 0 on success.
2810 static int ext4_split_extent_at(handle_t *handle,
2811 struct inode *inode,
2812 struct ext4_ext_path *path,
2817 ext4_fsblk_t newblock;
2818 ext4_lblk_t ee_block;
2819 struct ext4_extent *ex, newex, orig_ex;
2820 struct ext4_extent *ex2 = NULL;
2821 unsigned int ee_len, depth;
2824 ext_debug("ext4_split_extents_at: inode %lu, logical"
2825 "block %llu\n", inode->i_ino, (unsigned long long)split);
2827 ext4_ext_show_leaf(inode, path);
2829 depth = ext_depth(inode);
2830 ex = path[depth].p_ext;
2831 ee_block = le32_to_cpu(ex->ee_block);
2832 ee_len = ext4_ext_get_actual_len(ex);
2833 newblock = split - ee_block + ext4_ext_pblock(ex);
2835 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2837 err = ext4_ext_get_access(handle, inode, path + depth);
2841 if (split == ee_block) {
2843 * case b: block @split is the block that the extent begins with
2844 * then we just change the state of the extent, and splitting
2847 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2848 ext4_ext_mark_uninitialized(ex);
2850 ext4_ext_mark_initialized(ex);
2852 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2853 ext4_ext_try_to_merge(inode, path, ex);
2855 err = ext4_ext_dirty(handle, inode, path + depth);
2860 memcpy(&orig_ex, ex, sizeof(orig_ex));
2861 ex->ee_len = cpu_to_le16(split - ee_block);
2862 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2863 ext4_ext_mark_uninitialized(ex);
2866 * path may lead to new leaf, not to original leaf any more
2867 * after ext4_ext_insert_extent() returns,
2869 err = ext4_ext_dirty(handle, inode, path + depth);
2871 goto fix_extent_len;
2874 ex2->ee_block = cpu_to_le32(split);
2875 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2876 ext4_ext_store_pblock(ex2, newblock);
2877 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2878 ext4_ext_mark_uninitialized(ex2);
2880 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2881 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2882 err = ext4_ext_zeroout(inode, &orig_ex);
2884 goto fix_extent_len;
2885 /* update the extent length and mark as initialized */
2886 ex->ee_len = cpu_to_le32(ee_len);
2887 ext4_ext_try_to_merge(inode, path, ex);
2888 err = ext4_ext_dirty(handle, inode, path + depth);
2891 goto fix_extent_len;
2894 ext4_ext_show_leaf(inode, path);
2898 ex->ee_len = orig_ex.ee_len;
2899 ext4_ext_dirty(handle, inode, path + depth);
2904 * ext4_split_extents() splits an extent and mark extent which is covered
2905 * by @map as split_flags indicates
2907 * It may result in splitting the extent into multiple extents (upto three)
2908 * There are three possibilities:
2909 * a> There is no split required
2910 * b> Splits in two extents: Split is happening at either end of the extent
2911 * c> Splits in three extents: Somone is splitting in middle of the extent
2914 static int ext4_split_extent(handle_t *handle,
2915 struct inode *inode,
2916 struct ext4_ext_path *path,
2917 struct ext4_map_blocks *map,
2921 ext4_lblk_t ee_block;
2922 struct ext4_extent *ex;
2923 unsigned int ee_len, depth;
2926 int split_flag1, flags1;
2928 depth = ext_depth(inode);
2929 ex = path[depth].p_ext;
2930 ee_block = le32_to_cpu(ex->ee_block);
2931 ee_len = ext4_ext_get_actual_len(ex);
2932 uninitialized = ext4_ext_is_uninitialized(ex);
2934 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2935 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2936 EXT4_EXT_MAY_ZEROOUT : 0;
2937 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2939 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2940 EXT4_EXT_MARK_UNINIT2;
2941 err = ext4_split_extent_at(handle, inode, path,
2942 map->m_lblk + map->m_len, split_flag1, flags1);
2947 ext4_ext_drop_refs(path);
2948 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2950 return PTR_ERR(path);
2952 if (map->m_lblk >= ee_block) {
2953 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2954 EXT4_EXT_MAY_ZEROOUT : 0;
2956 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2957 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2958 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2959 err = ext4_split_extent_at(handle, inode, path,
2960 map->m_lblk, split_flag1, flags);
2965 ext4_ext_show_leaf(inode, path);
2967 return err ? err : map->m_len;
2970 #define EXT4_EXT_ZERO_LEN 7
2972 * This function is called by ext4_ext_map_blocks() if someone tries to write
2973 * to an uninitialized extent. It may result in splitting the uninitialized
2974 * extent into multiple extents (up to three - one initialized and two
2976 * There are three possibilities:
2977 * a> There is no split required: Entire extent should be initialized
2978 * b> Splits in two extents: Write is happening at either end of the extent
2979 * c> Splits in three extents: Somone is writing in middle of the extent
2981 static int ext4_ext_convert_to_initialized(handle_t *handle,
2982 struct inode *inode,
2983 struct ext4_map_blocks *map,
2984 struct ext4_ext_path *path)
2986 struct ext4_map_blocks split_map;
2987 struct ext4_extent zero_ex;
2988 struct ext4_extent *ex;
2989 ext4_lblk_t ee_block, eof_block;
2990 unsigned int allocated, ee_len, depth;
2994 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2995 "block %llu, max_blocks %u\n", inode->i_ino,
2996 (unsigned long long)map->m_lblk, map->m_len);
2998 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2999 inode->i_sb->s_blocksize_bits;
3000 if (eof_block < map->m_lblk + map->m_len)
3001 eof_block = map->m_lblk + map->m_len;
3003 depth = ext_depth(inode);
3004 ex = path[depth].p_ext;
3005 ee_block = le32_to_cpu(ex->ee_block);
3006 ee_len = ext4_ext_get_actual_len(ex);
3007 allocated = ee_len - (map->m_lblk - ee_block);
3009 WARN_ON(map->m_lblk < ee_block);
3011 * It is safe to convert extent to initialized via explicit
3012 * zeroout only if extent is fully insde i_size or new_size.
3014 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3016 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3017 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3018 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3019 err = ext4_ext_zeroout(inode, ex);
3023 err = ext4_ext_get_access(handle, inode, path + depth);
3026 ext4_ext_mark_initialized(ex);
3027 ext4_ext_try_to_merge(inode, path, ex);
3028 err = ext4_ext_dirty(handle, inode, path + depth);
3034 * 1. split the extent into three extents.
3035 * 2. split the extent into two extents, zeroout the first half.
3036 * 3. split the extent into two extents, zeroout the second half.
3037 * 4. split the extent into two extents with out zeroout.
3039 split_map.m_lblk = map->m_lblk;
3040 split_map.m_len = map->m_len;
3042 if (allocated > map->m_len) {
3043 if (allocated <= EXT4_EXT_ZERO_LEN &&
3044 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3047 cpu_to_le32(map->m_lblk);
3048 zero_ex.ee_len = cpu_to_le16(allocated);
3049 ext4_ext_store_pblock(&zero_ex,
3050 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3051 err = ext4_ext_zeroout(inode, &zero_ex);
3054 split_map.m_lblk = map->m_lblk;
3055 split_map.m_len = allocated;
3056 } else if ((map->m_lblk - ee_block + map->m_len <
3057 EXT4_EXT_ZERO_LEN) &&
3058 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3060 if (map->m_lblk != ee_block) {
3061 zero_ex.ee_block = ex->ee_block;
3062 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3064 ext4_ext_store_pblock(&zero_ex,
3065 ext4_ext_pblock(ex));
3066 err = ext4_ext_zeroout(inode, &zero_ex);
3071 split_map.m_lblk = ee_block;
3072 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3073 allocated = map->m_len;
3077 allocated = ext4_split_extent(handle, inode, path,
3078 &split_map, split_flag, 0);
3083 return err ? err : allocated;
3087 * This function is called by ext4_ext_map_blocks() from
3088 * ext4_get_blocks_dio_write() when DIO to write
3089 * to an uninitialized extent.
3091 * Writing to an uninitialized extent may result in splitting the uninitialized
3092 * extent into multiple /initialized uninitialized extents (up to three)
3093 * There are three possibilities:
3094 * a> There is no split required: Entire extent should be uninitialized
3095 * b> Splits in two extents: Write is happening at either end of the extent
3096 * c> Splits in three extents: Somone is writing in middle of the extent
3098 * One of more index blocks maybe needed if the extent tree grow after
3099 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3100 * complete, we need to split the uninitialized extent before DIO submit
3101 * the IO. The uninitialized extent called at this time will be split
3102 * into three uninitialized extent(at most). After IO complete, the part
3103 * being filled will be convert to initialized by the end_io callback function
3104 * via ext4_convert_unwritten_extents().
3106 * Returns the size of uninitialized extent to be written on success.
3108 static int ext4_split_unwritten_extents(handle_t *handle,
3109 struct inode *inode,
3110 struct ext4_map_blocks *map,
3111 struct ext4_ext_path *path,
3114 ext4_lblk_t eof_block;
3115 ext4_lblk_t ee_block;
3116 struct ext4_extent *ex;
3117 unsigned int ee_len;
3118 int split_flag = 0, depth;
3120 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3121 "block %llu, max_blocks %u\n", inode->i_ino,
3122 (unsigned long long)map->m_lblk, map->m_len);
3124 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3125 inode->i_sb->s_blocksize_bits;
3126 if (eof_block < map->m_lblk + map->m_len)
3127 eof_block = map->m_lblk + map->m_len;
3129 * It is safe to convert extent to initialized via explicit
3130 * zeroout only if extent is fully insde i_size or new_size.
3132 depth = ext_depth(inode);
3133 ex = path[depth].p_ext;
3134 ee_block = le32_to_cpu(ex->ee_block);
3135 ee_len = ext4_ext_get_actual_len(ex);
3137 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3138 split_flag |= EXT4_EXT_MARK_UNINIT2;
3140 flags |= EXT4_GET_BLOCKS_PRE_IO;
3141 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3144 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3145 struct inode *inode,
3146 struct ext4_ext_path *path)
3148 struct ext4_extent *ex;
3152 depth = ext_depth(inode);
3153 ex = path[depth].p_ext;
3155 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3156 "block %llu, max_blocks %u\n", inode->i_ino,
3157 (unsigned long long)le32_to_cpu(ex->ee_block),
3158 ext4_ext_get_actual_len(ex));
3160 err = ext4_ext_get_access(handle, inode, path + depth);
3163 /* first mark the extent as initialized */
3164 ext4_ext_mark_initialized(ex);
3166 /* note: ext4_ext_correct_indexes() isn't needed here because
3167 * borders are not changed
3169 ext4_ext_try_to_merge(inode, path, ex);
3171 /* Mark modified extent as dirty */
3172 err = ext4_ext_dirty(handle, inode, path + depth);
3174 ext4_ext_show_leaf(inode, path);
3178 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3179 sector_t block, int count)
3182 for (i = 0; i < count; i++)
3183 unmap_underlying_metadata(bdev, block + i);
3187 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3189 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3191 struct ext4_ext_path *path,
3195 struct ext4_extent_header *eh;
3196 struct ext4_extent *last_ex;
3198 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3201 depth = ext_depth(inode);
3202 eh = path[depth].p_hdr;
3204 if (unlikely(!eh->eh_entries)) {
3205 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3206 "EOFBLOCKS_FL set");
3209 last_ex = EXT_LAST_EXTENT(eh);
3211 * We should clear the EOFBLOCKS_FL flag if we are writing the
3212 * last block in the last extent in the file. We test this by
3213 * first checking to see if the caller to
3214 * ext4_ext_get_blocks() was interested in the last block (or
3215 * a block beyond the last block) in the current extent. If
3216 * this turns out to be false, we can bail out from this
3217 * function immediately.
3219 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3220 ext4_ext_get_actual_len(last_ex))
3223 * If the caller does appear to be planning to write at or
3224 * beyond the end of the current extent, we then test to see
3225 * if the current extent is the last extent in the file, by
3226 * checking to make sure it was reached via the rightmost node
3227 * at each level of the tree.
3229 for (i = depth-1; i >= 0; i--)
3230 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3232 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3233 return ext4_mark_inode_dirty(handle, inode);
3237 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3238 struct ext4_map_blocks *map,
3239 struct ext4_ext_path *path, int flags,
3240 unsigned int allocated, ext4_fsblk_t newblock)
3244 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3246 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3247 "block %llu, max_blocks %u, flags %d, allocated %u",
3248 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3250 ext4_ext_show_leaf(inode, path);
3252 /* get_block() before submit the IO, split the extent */
3253 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3254 ret = ext4_split_unwritten_extents(handle, inode, map,
3257 * Flag the inode(non aio case) or end_io struct (aio case)
3258 * that this IO needs to conversion to written when IO is
3261 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3262 io->flag = EXT4_IO_END_UNWRITTEN;
3263 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3265 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3266 if (ext4_should_dioread_nolock(inode))
3267 map->m_flags |= EXT4_MAP_UNINIT;
3270 /* IO end_io complete, convert the filled extent to written */
3271 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3272 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3275 ext4_update_inode_fsync_trans(handle, inode, 1);
3276 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3282 /* buffered IO case */
3284 * repeat fallocate creation request
3285 * we already have an unwritten extent
3287 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3290 /* buffered READ or buffered write_begin() lookup */
3291 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3293 * We have blocks reserved already. We
3294 * return allocated blocks so that delalloc
3295 * won't do block reservation for us. But
3296 * the buffer head will be unmapped so that
3297 * a read from the block returns 0s.
3299 map->m_flags |= EXT4_MAP_UNWRITTEN;
3303 /* buffered write, writepage time, convert*/
3304 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3306 ext4_update_inode_fsync_trans(handle, inode, 1);
3307 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3319 map->m_flags |= EXT4_MAP_NEW;
3321 * if we allocated more blocks than requested
3322 * we need to make sure we unmap the extra block
3323 * allocated. The actual needed block will get
3324 * unmapped later when we find the buffer_head marked
3327 if (allocated > map->m_len) {
3328 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3329 newblock + map->m_len,
3330 allocated - map->m_len);
3331 allocated = map->m_len;
3335 * If we have done fallocate with the offset that is already
3336 * delayed allocated, we would have block reservation
3337 * and quota reservation done in the delayed write path.
3338 * But fallocate would have already updated quota and block
3339 * count for this offset. So cancel these reservation
3341 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3342 ext4_da_update_reserve_space(inode, allocated, 0);
3345 map->m_flags |= EXT4_MAP_MAPPED;
3347 if (allocated > map->m_len)
3348 allocated = map->m_len;
3349 ext4_ext_show_leaf(inode, path);
3350 map->m_pblk = newblock;
3351 map->m_len = allocated;
3354 ext4_ext_drop_refs(path);
3357 return err ? err : allocated;
3361 * get_implied_cluster_alloc - check to see if the requested
3362 * allocation (in the map structure) overlaps with a cluster already
3363 * allocated in an extent.
3364 * @sbi The ext4-specific superblock structure
3365 * @map The requested lblk->pblk mapping
3366 * @ex The extent structure which might contain an implied
3367 * cluster allocation
3369 * This function is called by ext4_ext_map_blocks() after we failed to
3370 * find blocks that were already in the inode's extent tree. Hence,
3371 * we know that the beginning of the requested region cannot overlap
3372 * the extent from the inode's extent tree. There are three cases we
3373 * want to catch. The first is this case:
3375 * |--- cluster # N--|
3376 * |--- extent ---| |---- requested region ---|
3379 * The second case that we need to test for is this one:
3381 * |--------- cluster # N ----------------|
3382 * |--- requested region --| |------- extent ----|
3383 * |=======================|
3385 * The third case is when the requested region lies between two extents
3386 * within the same cluster:
3387 * |------------- cluster # N-------------|
3388 * |----- ex -----| |---- ex_right ----|
3389 * |------ requested region ------|
3390 * |================|
3392 * In each of the above cases, we need to set the map->m_pblk and
3393 * map->m_len so it corresponds to the return the extent labelled as
3394 * "|====|" from cluster #N, since it is already in use for data in
3395 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3396 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3397 * as a new "allocated" block region. Otherwise, we will return 0 and
3398 * ext4_ext_map_blocks() will then allocate one or more new clusters
3399 * by calling ext4_mb_new_blocks().
3401 static int get_implied_cluster_alloc(struct ext4_sb_info *sbi,
3402 struct ext4_map_blocks *map,
3403 struct ext4_extent *ex,
3404 struct ext4_ext_path *path)
3406 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3407 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3408 ext4_lblk_t rr_cluster_start, rr_cluster_end;
3409 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3410 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3411 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3413 /* The extent passed in that we are trying to match */
3414 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3415 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3417 /* The requested region passed into ext4_map_blocks() */
3418 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3419 rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3421 if ((rr_cluster_start == ex_cluster_end) ||
3422 (rr_cluster_start == ex_cluster_start)) {
3423 if (rr_cluster_start == ex_cluster_end)
3424 ee_start += ee_len - 1;
3425 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3427 map->m_len = min(map->m_len,
3428 (unsigned) sbi->s_cluster_ratio - c_offset);
3430 * Check for and handle this case:
3432 * |--------- cluster # N-------------|
3433 * |------- extent ----|
3434 * |--- requested region ---|
3438 if (map->m_lblk < ee_block)
3439 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3442 * Check for the case where there is already another allocated
3443 * block to the right of 'ex' but before the end of the cluster.
3445 * |------------- cluster # N-------------|
3446 * |----- ex -----| |---- ex_right ----|
3447 * |------ requested region ------|
3448 * |================|
3450 if (map->m_lblk > ee_block) {
3451 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3452 map->m_len = min(map->m_len, next - map->m_lblk);
3461 * Block allocation/map/preallocation routine for extents based files
3464 * Need to be called with
3465 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3466 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3468 * return > 0, number of of blocks already mapped/allocated
3469 * if create == 0 and these are pre-allocated blocks
3470 * buffer head is unmapped
3471 * otherwise blocks are mapped
3473 * return = 0, if plain look up failed (blocks have not been allocated)
3474 * buffer head is unmapped
3476 * return < 0, error case.
3478 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3479 struct ext4_map_blocks *map, int flags)
3481 struct ext4_ext_path *path = NULL;
3482 struct ext4_extent newex, *ex, *ex2;
3483 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3484 ext4_fsblk_t newblock = 0;
3485 int free_on_err = 0, err = 0, depth, ret;
3486 unsigned int allocated = 0, offset = 0;
3487 unsigned int punched_out = 0;
3488 unsigned int result = 0;
3489 struct ext4_allocation_request ar;
3490 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3491 ext4_lblk_t cluster_offset;
3492 struct ext4_map_blocks punch_map;
3494 ext_debug("blocks %u/%u requested for inode %lu\n",
3495 map->m_lblk, map->m_len, inode->i_ino);
3496 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3498 /* check in cache */
3499 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3500 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3501 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3502 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3504 * block isn't allocated yet and
3505 * user doesn't want to allocate it
3509 /* we should allocate requested block */
3511 /* block is already allocated */
3512 newblock = map->m_lblk
3513 - le32_to_cpu(newex.ee_block)
3514 + ext4_ext_pblock(&newex);
3515 /* number of remaining blocks in the extent */
3516 allocated = ext4_ext_get_actual_len(&newex) -
3517 (map->m_lblk - le32_to_cpu(newex.ee_block));
3522 /* find extent for this block */
3523 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3525 err = PTR_ERR(path);
3530 depth = ext_depth(inode);
3533 * consistent leaf must not be empty;
3534 * this situation is possible, though, _during_ tree modification;
3535 * this is why assert can't be put in ext4_ext_find_extent()
3537 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3538 EXT4_ERROR_INODE(inode, "bad extent address "
3539 "lblock: %lu, depth: %d pblock %lld",
3540 (unsigned long) map->m_lblk, depth,
3541 path[depth].p_block);
3546 ex = path[depth].p_ext;
3548 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3549 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3550 unsigned short ee_len;
3553 * Uninitialized extents are treated as holes, except that
3554 * we split out initialized portions during a write.
3556 ee_len = ext4_ext_get_actual_len(ex);
3557 /* if found extent covers block, simply return it */
3558 if (in_range(map->m_lblk, ee_block, ee_len)) {
3559 ext4_fsblk_t partial_cluster = 0;
3561 newblock = map->m_lblk - ee_block + ee_start;
3562 /* number of remaining blocks in the extent */
3563 allocated = ee_len - (map->m_lblk - ee_block);
3564 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3565 ee_block, ee_len, newblock);
3567 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3569 * Do not put uninitialized extent
3572 if (!ext4_ext_is_uninitialized(ex)) {
3573 ext4_ext_put_in_cache(inode, ee_block,
3577 ret = ext4_ext_handle_uninitialized_extents(
3578 handle, inode, map, path, flags,
3579 allocated, newblock);
3584 * Punch out the map length, but only to the
3587 punched_out = allocated < map->m_len ?
3588 allocated : map->m_len;
3591 * Sense extents need to be converted to
3592 * uninitialized, they must fit in an
3593 * uninitialized extent
3595 if (punched_out > EXT_UNINIT_MAX_LEN)
3596 punched_out = EXT_UNINIT_MAX_LEN;
3598 punch_map.m_lblk = map->m_lblk;
3599 punch_map.m_pblk = newblock;
3600 punch_map.m_len = punched_out;
3601 punch_map.m_flags = 0;
3603 /* Check to see if the extent needs to be split */
3604 if (punch_map.m_len != ee_len ||
3605 punch_map.m_lblk != ee_block) {
3607 ret = ext4_split_extent(handle, inode,
3608 path, &punch_map, 0,
3609 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3610 EXT4_GET_BLOCKS_PRE_IO);
3617 * find extent for the block at
3618 * the start of the hole
3620 ext4_ext_drop_refs(path);
3623 path = ext4_ext_find_extent(inode,
3626 err = PTR_ERR(path);
3631 depth = ext_depth(inode);
3632 ex = path[depth].p_ext;
3633 ee_len = ext4_ext_get_actual_len(ex);
3634 ee_block = le32_to_cpu(ex->ee_block);
3635 ee_start = ext4_ext_pblock(ex);
3639 ext4_ext_mark_uninitialized(ex);
3641 ext4_ext_invalidate_cache(inode);
3643 err = ext4_ext_rm_leaf(handle, inode, path,
3644 &partial_cluster, map->m_lblk,
3645 map->m_lblk + punched_out);
3647 if (!err && path->p_hdr->eh_entries == 0) {
3649 * Punch hole freed all of this sub tree,
3650 * so we need to correct eh_depth
3652 err = ext4_ext_get_access(handle, inode, path);
3654 ext_inode_hdr(inode)->eh_depth = 0;
3655 ext_inode_hdr(inode)->eh_max =
3656 cpu_to_le16(ext4_ext_space_root(
3659 err = ext4_ext_dirty(
3660 handle, inode, path);
3669 * requested block isn't allocated yet;
3670 * we couldn't try to create block if create flag is zero
3672 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3674 * put just found gap into cache to speed up
3675 * subsequent requests
3677 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3682 * Okay, we need to do block allocation.
3684 newex.ee_block = cpu_to_le32(map->m_lblk);
3685 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3688 * If we are doing bigalloc, check to see if the extent returned
3689 * by ext4_ext_find_extent() implies a cluster we can use.
3691 if (cluster_offset && ex &&
3692 get_implied_cluster_alloc(sbi, map, ex, path)) {
3693 ar.len = allocated = map->m_len;
3694 newblock = map->m_pblk;
3695 goto got_allocated_blocks;
3698 /* find neighbour allocated blocks */
3699 ar.lleft = map->m_lblk;
3700 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3703 ar.lright = map->m_lblk;
3705 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
3709 /* Check if the extent after searching to the right implies a
3710 * cluster we can use. */
3711 if ((sbi->s_cluster_ratio > 1) && ex2 &&
3712 get_implied_cluster_alloc(sbi, map, ex2, path)) {
3713 ar.len = allocated = map->m_len;
3714 newblock = map->m_pblk;
3715 goto got_allocated_blocks;
3719 * See if request is beyond maximum number of blocks we can have in
3720 * a single extent. For an initialized extent this limit is
3721 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3722 * EXT_UNINIT_MAX_LEN.
3724 if (map->m_len > EXT_INIT_MAX_LEN &&
3725 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3726 map->m_len = EXT_INIT_MAX_LEN;
3727 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3728 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3729 map->m_len = EXT_UNINIT_MAX_LEN;
3731 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3732 newex.ee_len = cpu_to_le16(map->m_len);
3733 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
3735 allocated = ext4_ext_get_actual_len(&newex);
3737 allocated = map->m_len;
3739 /* allocate new block */
3741 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3742 ar.logical = map->m_lblk;
3744 * We calculate the offset from the beginning of the cluster
3745 * for the logical block number, since when we allocate a
3746 * physical cluster, the physical block should start at the
3747 * same offset from the beginning of the cluster. This is
3748 * needed so that future calls to get_implied_cluster_alloc()
3751 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
3752 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
3754 ar.logical -= offset;
3755 if (S_ISREG(inode->i_mode))
3756 ar.flags = EXT4_MB_HINT_DATA;
3758 /* disable in-core preallocation for non-regular files */
3760 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
3761 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
3762 newblock = ext4_mb_new_blocks(handle, &ar, &err);
3765 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3766 ar.goal, newblock, allocated);
3768 ar.len = EXT4_C2B(sbi, ar.len) - offset;
3769 if (ar.len > allocated)
3772 got_allocated_blocks:
3773 /* try to insert new extent into found leaf and return */
3774 ext4_ext_store_pblock(&newex, newblock + offset);
3775 newex.ee_len = cpu_to_le16(ar.len);
3776 /* Mark uninitialized */
3777 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3778 ext4_ext_mark_uninitialized(&newex);
3780 * io_end structure was created for every IO write to an
3781 * uninitialized extent. To avoid unnecessary conversion,
3782 * here we flag the IO that really needs the conversion.
3783 * For non asycn direct IO case, flag the inode state
3784 * that we need to perform conversion when IO is done.
3786 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3787 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3788 io->flag = EXT4_IO_END_UNWRITTEN;
3789 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3791 ext4_set_inode_state(inode,
3792 EXT4_STATE_DIO_UNWRITTEN);
3794 if (ext4_should_dioread_nolock(inode))
3795 map->m_flags |= EXT4_MAP_UNINIT;
3798 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3800 err = ext4_ext_insert_extent(handle, inode, path,
3802 if (err && free_on_err) {
3803 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
3804 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
3805 /* free data blocks we just allocated */
3806 /* not a good idea to call discard here directly,
3807 * but otherwise we'd need to call it every free() */
3808 ext4_discard_preallocations(inode);
3809 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3810 ext4_ext_get_actual_len(&newex), fb_flags);
3814 /* previous routine could use block we allocated */
3815 newblock = ext4_ext_pblock(&newex);
3816 allocated = ext4_ext_get_actual_len(&newex);
3817 if (allocated > map->m_len)
3818 allocated = map->m_len;
3819 map->m_flags |= EXT4_MAP_NEW;
3822 * Update reserved blocks/metadata blocks after successful
3823 * block allocation which had been deferred till now.
3825 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3826 ext4_da_update_reserve_space(inode, allocated, 1);
3829 * Cache the extent and update transaction to commit on fdatasync only
3830 * when it is _not_ an uninitialized extent.
3832 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3833 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3834 ext4_update_inode_fsync_trans(handle, inode, 1);
3836 ext4_update_inode_fsync_trans(handle, inode, 0);
3838 if (allocated > map->m_len)
3839 allocated = map->m_len;
3840 ext4_ext_show_leaf(inode, path);
3841 map->m_flags |= EXT4_MAP_MAPPED;
3842 map->m_pblk = newblock;
3843 map->m_len = allocated;
3846 ext4_ext_drop_refs(path);
3849 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3850 newblock, map->m_len, err ? err : allocated);
3852 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
3853 punched_out : allocated;
3855 return err ? err : result;
3858 void ext4_ext_truncate(struct inode *inode)
3860 struct address_space *mapping = inode->i_mapping;
3861 struct super_block *sb = inode->i_sb;
3862 ext4_lblk_t last_block;
3868 * finish any pending end_io work so we won't run the risk of
3869 * converting any truncated blocks to initialized later
3871 ext4_flush_completed_IO(inode);
3874 * probably first extent we're gonna free will be last in block
3876 err = ext4_writepage_trans_blocks(inode);
3877 handle = ext4_journal_start(inode, err);
3881 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3882 page_len = PAGE_CACHE_SIZE -
3883 (inode->i_size & (PAGE_CACHE_SIZE - 1));
3885 err = ext4_discard_partial_page_buffers(handle,
3886 mapping, inode->i_size, page_len, 0);
3892 if (ext4_orphan_add(handle, inode))
3895 down_write(&EXT4_I(inode)->i_data_sem);
3896 ext4_ext_invalidate_cache(inode);
3898 ext4_discard_preallocations(inode);
3901 * TODO: optimization is possible here.
3902 * Probably we need not scan at all,
3903 * because page truncation is enough.
3906 /* we have to know where to truncate from in crash case */
3907 EXT4_I(inode)->i_disksize = inode->i_size;
3908 ext4_mark_inode_dirty(handle, inode);
3910 last_block = (inode->i_size + sb->s_blocksize - 1)
3911 >> EXT4_BLOCK_SIZE_BITS(sb);
3912 err = ext4_ext_remove_space(inode, last_block);
3914 /* In a multi-transaction truncate, we only make the final
3915 * transaction synchronous.
3918 ext4_handle_sync(handle);
3920 up_write(&EXT4_I(inode)->i_data_sem);
3924 * If this was a simple ftruncate() and the file will remain alive,
3925 * then we need to clear up the orphan record which we created above.
3926 * However, if this was a real unlink then we were called by
3927 * ext4_delete_inode(), and we allow that function to clean up the
3928 * orphan info for us.
3931 ext4_orphan_del(handle, inode);
3933 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3934 ext4_mark_inode_dirty(handle, inode);
3935 ext4_journal_stop(handle);
3938 static void ext4_falloc_update_inode(struct inode *inode,
3939 int mode, loff_t new_size, int update_ctime)
3941 struct timespec now;
3944 now = current_fs_time(inode->i_sb);
3945 if (!timespec_equal(&inode->i_ctime, &now))
3946 inode->i_ctime = now;
3949 * Update only when preallocation was requested beyond
3952 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3953 if (new_size > i_size_read(inode))
3954 i_size_write(inode, new_size);
3955 if (new_size > EXT4_I(inode)->i_disksize)
3956 ext4_update_i_disksize(inode, new_size);
3959 * Mark that we allocate beyond EOF so the subsequent truncate
3960 * can proceed even if the new size is the same as i_size.
3962 if (new_size > i_size_read(inode))
3963 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3969 * preallocate space for a file. This implements ext4's fallocate file
3970 * operation, which gets called from sys_fallocate system call.
3971 * For block-mapped files, posix_fallocate should fall back to the method
3972 * of writing zeroes to the required new blocks (the same behavior which is
3973 * expected for file systems which do not support fallocate() system call).
3975 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
3977 struct inode *inode = file->f_path.dentry->d_inode;
3980 unsigned int max_blocks;
3984 struct ext4_map_blocks map;
3985 unsigned int credits, blkbits = inode->i_blkbits;
3988 * currently supporting (pre)allocate mode for extent-based
3991 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3994 /* Return error if mode is not supported */
3995 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3998 if (mode & FALLOC_FL_PUNCH_HOLE)
3999 return ext4_punch_hole(file, offset, len);
4001 trace_ext4_fallocate_enter(inode, offset, len, mode);
4002 map.m_lblk = offset >> blkbits;
4004 * We can't just convert len to max_blocks because
4005 * If blocksize = 4096 offset = 3072 and len = 2048
4007 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4010 * credits to insert 1 extent into extent tree
4012 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4013 mutex_lock(&inode->i_mutex);
4014 ret = inode_newsize_ok(inode, (len + offset));
4016 mutex_unlock(&inode->i_mutex);
4017 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4021 while (ret >= 0 && ret < max_blocks) {
4022 map.m_lblk = map.m_lblk + ret;
4023 map.m_len = max_blocks = max_blocks - ret;
4024 handle = ext4_journal_start(inode, credits);
4025 if (IS_ERR(handle)) {
4026 ret = PTR_ERR(handle);
4029 ret = ext4_map_blocks(handle, inode, &map,
4030 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
4031 EXT4_GET_BLOCKS_NO_NORMALIZE);
4035 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4036 "returned error inode#%lu, block=%u, "
4037 "max_blocks=%u", __func__,
4038 inode->i_ino, map.m_lblk, max_blocks);
4040 ext4_mark_inode_dirty(handle, inode);
4041 ret2 = ext4_journal_stop(handle);
4044 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4045 blkbits) >> blkbits))
4046 new_size = offset + len;
4048 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4050 ext4_falloc_update_inode(inode, mode, new_size,
4051 (map.m_flags & EXT4_MAP_NEW));
4052 ext4_mark_inode_dirty(handle, inode);
4053 ret2 = ext4_journal_stop(handle);
4057 if (ret == -ENOSPC &&
4058 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4062 mutex_unlock(&inode->i_mutex);
4063 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4064 ret > 0 ? ret2 : ret);
4065 return ret > 0 ? ret2 : ret;
4069 * This function convert a range of blocks to written extents
4070 * The caller of this function will pass the start offset and the size.
4071 * all unwritten extents within this range will be converted to
4074 * This function is called from the direct IO end io call back
4075 * function, to convert the fallocated extents after IO is completed.
4076 * Returns 0 on success.
4078 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4082 unsigned int max_blocks;
4085 struct ext4_map_blocks map;
4086 unsigned int credits, blkbits = inode->i_blkbits;
4088 map.m_lblk = offset >> blkbits;
4090 * We can't just convert len to max_blocks because
4091 * If blocksize = 4096 offset = 3072 and len = 2048
4093 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4096 * credits to insert 1 extent into extent tree
4098 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4099 while (ret >= 0 && ret < max_blocks) {
4101 map.m_len = (max_blocks -= ret);
4102 handle = ext4_journal_start(inode, credits);
4103 if (IS_ERR(handle)) {
4104 ret = PTR_ERR(handle);
4107 ret = ext4_map_blocks(handle, inode, &map,
4108 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4111 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4112 "returned error inode#%lu, block=%u, "
4113 "max_blocks=%u", __func__,
4114 inode->i_ino, map.m_lblk, map.m_len);
4116 ext4_mark_inode_dirty(handle, inode);
4117 ret2 = ext4_journal_stop(handle);
4118 if (ret <= 0 || ret2 )
4121 return ret > 0 ? ret2 : ret;
4125 * Callback function called for each extent to gather FIEMAP information.
4127 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4128 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4136 struct fiemap_extent_info *fieinfo = data;
4137 unsigned char blksize_bits;
4139 blksize_bits = inode->i_sb->s_blocksize_bits;
4140 logical = (__u64)newex->ec_block << blksize_bits;
4142 if (newex->ec_start == 0) {
4144 * No extent in extent-tree contains block @newex->ec_start,
4145 * then the block may stay in 1)a hole or 2)delayed-extent.
4147 * Holes or delayed-extents are processed as follows.
4148 * 1. lookup dirty pages with specified range in pagecache.
4149 * If no page is got, then there is no delayed-extent and
4150 * return with EXT_CONTINUE.
4151 * 2. find the 1st mapped buffer,
4152 * 3. check if the mapped buffer is both in the request range
4153 * and a delayed buffer. If not, there is no delayed-extent,
4155 * 4. a delayed-extent is found, the extent will be collected.
4157 ext4_lblk_t end = 0;
4158 pgoff_t last_offset;
4161 pgoff_t start_index = 0;
4162 struct page **pages = NULL;
4163 struct buffer_head *bh = NULL;
4164 struct buffer_head *head = NULL;
4165 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4167 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4171 offset = logical >> PAGE_SHIFT;
4173 last_offset = offset;
4175 ret = find_get_pages_tag(inode->i_mapping, &offset,
4176 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4178 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4179 /* First time, try to find a mapped buffer. */
4182 for (index = 0; index < ret; index++)
4183 page_cache_release(pages[index]);
4186 return EXT_CONTINUE;
4191 /* Try to find the 1st mapped buffer. */
4192 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4194 if (!page_has_buffers(pages[index]))
4196 head = page_buffers(pages[index]);
4203 if (end >= newex->ec_block +
4205 /* The buffer is out of
4206 * the request range.
4210 if (buffer_mapped(bh) &&
4211 end >= newex->ec_block) {
4212 start_index = index - 1;
4213 /* get the 1st mapped buffer. */
4214 goto found_mapped_buffer;
4217 bh = bh->b_this_page;
4219 } while (bh != head);
4221 /* No mapped buffer in the range found in this page,
4222 * We need to look up next page.
4225 /* There is no page left, but we need to limit
4228 newex->ec_len = end - newex->ec_block;
4233 /*Find contiguous delayed buffers. */
4234 if (ret > 0 && pages[0]->index == last_offset)
4235 head = page_buffers(pages[0]);
4241 found_mapped_buffer:
4242 if (bh != NULL && buffer_delay(bh)) {
4243 /* 1st or contiguous delayed buffer found. */
4244 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4246 * 1st delayed buffer found, record
4247 * the start of extent.
4249 flags |= FIEMAP_EXTENT_DELALLOC;
4250 newex->ec_block = end;
4251 logical = (__u64)end << blksize_bits;
4253 /* Find contiguous delayed buffers. */
4255 if (!buffer_delay(bh))
4256 goto found_delayed_extent;
4257 bh = bh->b_this_page;
4259 } while (bh != head);
4261 for (; index < ret; index++) {
4262 if (!page_has_buffers(pages[index])) {
4266 head = page_buffers(pages[index]);
4272 if (pages[index]->index !=
4273 pages[start_index]->index + index
4275 /* Blocks are not contiguous. */
4281 if (!buffer_delay(bh))
4282 /* Delayed-extent ends. */
4283 goto found_delayed_extent;
4284 bh = bh->b_this_page;
4286 } while (bh != head);
4288 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4292 found_delayed_extent:
4293 newex->ec_len = min(end - newex->ec_block,
4294 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4295 if (ret == nr_pages && bh != NULL &&
4296 newex->ec_len < EXT_INIT_MAX_LEN &&
4298 /* Have not collected an extent and continue. */
4299 for (index = 0; index < ret; index++)
4300 page_cache_release(pages[index]);
4304 for (index = 0; index < ret; index++)
4305 page_cache_release(pages[index]);
4309 physical = (__u64)newex->ec_start << blksize_bits;
4310 length = (__u64)newex->ec_len << blksize_bits;
4312 if (ex && ext4_ext_is_uninitialized(ex))
4313 flags |= FIEMAP_EXTENT_UNWRITTEN;
4315 if (next == EXT_MAX_BLOCKS)
4316 flags |= FIEMAP_EXTENT_LAST;
4318 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4324 return EXT_CONTINUE;
4326 /* fiemap flags we can handle specified here */
4327 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4329 static int ext4_xattr_fiemap(struct inode *inode,
4330 struct fiemap_extent_info *fieinfo)
4334 __u32 flags = FIEMAP_EXTENT_LAST;
4335 int blockbits = inode->i_sb->s_blocksize_bits;
4339 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4340 struct ext4_iloc iloc;
4341 int offset; /* offset of xattr in inode */
4343 error = ext4_get_inode_loc(inode, &iloc);
4346 physical = iloc.bh->b_blocknr << blockbits;
4347 offset = EXT4_GOOD_OLD_INODE_SIZE +
4348 EXT4_I(inode)->i_extra_isize;
4350 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4351 flags |= FIEMAP_EXTENT_DATA_INLINE;
4353 } else { /* external block */
4354 physical = EXT4_I(inode)->i_file_acl << blockbits;
4355 length = inode->i_sb->s_blocksize;
4359 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4361 return (error < 0 ? error : 0);
4365 * ext4_ext_punch_hole
4367 * Punches a hole of "length" bytes in a file starting
4370 * @inode: The inode of the file to punch a hole in
4371 * @offset: The starting byte offset of the hole
4372 * @length: The length of the hole
4374 * Returns the number of blocks removed or negative on err
4376 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4378 struct inode *inode = file->f_path.dentry->d_inode;
4379 struct super_block *sb = inode->i_sb;
4380 struct ext4_ext_cache cache_ex;
4381 ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4382 struct address_space *mapping = inode->i_mapping;
4383 struct ext4_map_blocks map;
4385 loff_t first_page, last_page, page_len;
4386 loff_t first_page_offset, last_page_offset;
4387 int ret, credits, blocks_released, err = 0;
4389 /* No need to punch hole beyond i_size */
4390 if (offset >= inode->i_size)
4394 * If the hole extends beyond i_size, set the hole
4395 * to end after the page that contains i_size
4397 if (offset + length > inode->i_size) {
4398 length = inode->i_size +
4399 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4403 first_block = (offset + sb->s_blocksize - 1) >>
4404 EXT4_BLOCK_SIZE_BITS(sb);
4405 last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4407 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4408 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4410 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4411 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4414 * Write out all dirty pages to avoid race conditions
4415 * Then release them.
4417 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4418 err = filemap_write_and_wait_range(mapping,
4419 offset, offset + length - 1);
4425 /* Now release the pages */
4426 if (last_page_offset > first_page_offset) {
4427 truncate_inode_pages_range(mapping, first_page_offset,
4428 last_page_offset-1);
4431 /* finish any pending end_io work */
4432 ext4_flush_completed_IO(inode);
4434 credits = ext4_writepage_trans_blocks(inode);
4435 handle = ext4_journal_start(inode, credits);
4437 return PTR_ERR(handle);
4439 err = ext4_orphan_add(handle, inode);
4444 * Now we need to zero out the non-page-aligned data in the
4445 * pages at the start and tail of the hole, and unmap the buffer
4446 * heads for the block aligned regions of the page that were
4447 * completely zeroed.
4449 if (first_page > last_page) {
4451 * If the file space being truncated is contained within a page
4452 * just zero out and unmap the middle of that page
4454 err = ext4_discard_partial_page_buffers(handle,
4455 mapping, offset, length, 0);
4461 * zero out and unmap the partial page that contains
4462 * the start of the hole
4464 page_len = first_page_offset - offset;
4466 err = ext4_discard_partial_page_buffers(handle, mapping,
4467 offset, page_len, 0);
4473 * zero out and unmap the partial page that contains
4474 * the end of the hole
4476 page_len = offset + length - last_page_offset;
4478 err = ext4_discard_partial_page_buffers(handle, mapping,
4479 last_page_offset, page_len, 0);
4487 * If i_size is contained in the last page, we need to
4488 * unmap and zero the partial page after i_size
4490 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4491 inode->i_size % PAGE_CACHE_SIZE != 0) {
4493 page_len = PAGE_CACHE_SIZE -
4494 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4497 err = ext4_discard_partial_page_buffers(handle,
4498 mapping, inode->i_size, page_len, 0);
4505 /* If there are no blocks to remove, return now */
4506 if (first_block >= last_block)
4509 down_write(&EXT4_I(inode)->i_data_sem);
4510 ext4_ext_invalidate_cache(inode);
4511 ext4_discard_preallocations(inode);
4514 * Loop over all the blocks and identify blocks
4515 * that need to be punched out
4517 iblock = first_block;
4518 blocks_released = 0;
4519 while (iblock < last_block) {
4520 max_blocks = last_block - iblock;
4522 memset(&map, 0, sizeof(map));
4523 map.m_lblk = iblock;
4524 map.m_len = max_blocks;
4525 ret = ext4_ext_map_blocks(handle, inode, &map,
4526 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4529 blocks_released += ret;
4531 } else if (ret == 0) {
4533 * If map blocks could not find the block,
4534 * then it is in a hole. If the hole was
4535 * not already cached, then map blocks should
4536 * put it in the cache. So we can get the hole
4539 memset(&cache_ex, 0, sizeof(cache_ex));
4540 if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4541 !cache_ex.ec_start) {
4543 /* The hole is cached */
4544 num_blocks = cache_ex.ec_block +
4545 cache_ex.ec_len - iblock;
4548 /* The block could not be identified */
4553 /* Map blocks error */
4558 if (num_blocks == 0) {
4559 /* This condition should never happen */
4560 ext_debug("Block lookup failed");
4565 iblock += num_blocks;
4568 if (blocks_released > 0) {
4569 ext4_ext_invalidate_cache(inode);
4570 ext4_discard_preallocations(inode);
4574 ext4_handle_sync(handle);
4576 up_write(&EXT4_I(inode)->i_data_sem);
4579 ext4_orphan_del(handle, inode);
4580 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4581 ext4_mark_inode_dirty(handle, inode);
4582 ext4_journal_stop(handle);
4585 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4586 __u64 start, __u64 len)
4588 ext4_lblk_t start_blk;
4591 /* fallback to generic here if not in extents fmt */
4592 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4593 return generic_block_fiemap(inode, fieinfo, start, len,
4596 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4599 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4600 error = ext4_xattr_fiemap(inode, fieinfo);
4602 ext4_lblk_t len_blks;
4605 start_blk = start >> inode->i_sb->s_blocksize_bits;
4606 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4607 if (last_blk >= EXT_MAX_BLOCKS)
4608 last_blk = EXT_MAX_BLOCKS-1;
4609 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4612 * Walk the extent tree gathering extent information.
4613 * ext4_ext_fiemap_cb will push extents back to user.
4615 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4616 ext4_ext_fiemap_cb, fieinfo);