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"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
54 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static int ext4_split_extent(handle_t *handle,
61 struct ext4_ext_path *path,
62 struct ext4_map_blocks *map,
66 static int ext4_split_extent_at(handle_t *handle,
68 struct ext4_ext_path *path,
73 static int ext4_ext_truncate_extend_restart(handle_t *handle,
79 if (!ext4_handle_valid(handle))
81 if (handle->h_buffer_credits > needed)
83 err = ext4_journal_extend(handle, needed);
86 err = ext4_truncate_restart_trans(handle, inode, needed);
98 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
99 struct ext4_ext_path *path)
102 /* path points to block */
103 return ext4_journal_get_write_access(handle, path->p_bh);
105 /* path points to leaf/index in inode body */
106 /* we use in-core data, no need to protect them */
116 #define ext4_ext_dirty(handle, inode, path) \
117 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
118 static int __ext4_ext_dirty(const char *where, unsigned int line,
119 handle_t *handle, struct inode *inode,
120 struct ext4_ext_path *path)
124 /* path points to block */
125 err = __ext4_handle_dirty_metadata(where, line, handle,
128 /* path points to leaf/index in inode body */
129 err = ext4_mark_inode_dirty(handle, inode);
134 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
135 struct ext4_ext_path *path,
139 int depth = path->p_depth;
140 struct ext4_extent *ex;
143 * Try to predict block placement assuming that we are
144 * filling in a file which will eventually be
145 * non-sparse --- i.e., in the case of libbfd writing
146 * an ELF object sections out-of-order but in a way
147 * the eventually results in a contiguous object or
148 * executable file, or some database extending a table
149 * space file. However, this is actually somewhat
150 * non-ideal if we are writing a sparse file such as
151 * qemu or KVM writing a raw image file that is going
152 * to stay fairly sparse, since it will end up
153 * fragmenting the file system's free space. Maybe we
154 * should have some hueristics or some way to allow
155 * userspace to pass a hint to file system,
156 * especially if the latter case turns out to be
159 ex = path[depth].p_ext;
161 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
162 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
164 if (block > ext_block)
165 return ext_pblk + (block - ext_block);
167 return ext_pblk - (ext_block - block);
170 /* it looks like index is empty;
171 * try to find starting block from index itself */
172 if (path[depth].p_bh)
173 return path[depth].p_bh->b_blocknr;
176 /* OK. use inode's group */
177 return ext4_inode_to_goal_block(inode);
181 * Allocation for a meta data block
184 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
185 struct ext4_ext_path *path,
186 struct ext4_extent *ex, int *err, unsigned int flags)
188 ext4_fsblk_t goal, newblock;
190 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
191 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
196 static inline int ext4_ext_space_block(struct inode *inode, int check)
200 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
201 / sizeof(struct ext4_extent);
202 #ifdef AGGRESSIVE_TEST
203 if (!check && size > 6)
209 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
213 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
214 / sizeof(struct ext4_extent_idx);
215 #ifdef AGGRESSIVE_TEST
216 if (!check && size > 5)
222 static inline int ext4_ext_space_root(struct inode *inode, int check)
226 size = sizeof(EXT4_I(inode)->i_data);
227 size -= sizeof(struct ext4_extent_header);
228 size /= sizeof(struct ext4_extent);
229 #ifdef AGGRESSIVE_TEST
230 if (!check && size > 3)
236 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
240 size = sizeof(EXT4_I(inode)->i_data);
241 size -= sizeof(struct ext4_extent_header);
242 size /= sizeof(struct ext4_extent_idx);
243 #ifdef AGGRESSIVE_TEST
244 if (!check && size > 4)
251 * Calculate the number of metadata blocks needed
252 * to allocate @blocks
253 * Worse case is one block per extent
255 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
257 struct ext4_inode_info *ei = EXT4_I(inode);
260 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
261 / sizeof(struct ext4_extent_idx));
264 * If the new delayed allocation block is contiguous with the
265 * previous da block, it can share index blocks with the
266 * previous block, so we only need to allocate a new index
267 * block every idxs leaf blocks. At ldxs**2 blocks, we need
268 * an additional index block, and at ldxs**3 blocks, yet
269 * another index blocks.
271 if (ei->i_da_metadata_calc_len &&
272 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
275 if ((ei->i_da_metadata_calc_len % idxs) == 0)
277 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
279 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
281 ei->i_da_metadata_calc_len = 0;
283 ei->i_da_metadata_calc_len++;
284 ei->i_da_metadata_calc_last_lblock++;
289 * In the worst case we need a new set of index blocks at
290 * every level of the inode's extent tree.
292 ei->i_da_metadata_calc_len = 1;
293 ei->i_da_metadata_calc_last_lblock = lblock;
294 return ext_depth(inode) + 1;
298 ext4_ext_max_entries(struct inode *inode, int depth)
302 if (depth == ext_depth(inode)) {
304 max = ext4_ext_space_root(inode, 1);
306 max = ext4_ext_space_root_idx(inode, 1);
309 max = ext4_ext_space_block(inode, 1);
311 max = ext4_ext_space_block_idx(inode, 1);
317 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
319 ext4_fsblk_t block = ext4_ext_pblock(ext);
320 int len = ext4_ext_get_actual_len(ext);
324 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
327 static int ext4_valid_extent_idx(struct inode *inode,
328 struct ext4_extent_idx *ext_idx)
330 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
332 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
335 static int ext4_valid_extent_entries(struct inode *inode,
336 struct ext4_extent_header *eh,
339 unsigned short entries;
340 if (eh->eh_entries == 0)
343 entries = le16_to_cpu(eh->eh_entries);
347 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
349 if (!ext4_valid_extent(inode, ext))
355 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
357 if (!ext4_valid_extent_idx(inode, ext_idx))
366 static int __ext4_ext_check(const char *function, unsigned int line,
367 struct inode *inode, struct ext4_extent_header *eh,
370 const char *error_msg;
373 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
374 error_msg = "invalid magic";
377 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
378 error_msg = "unexpected eh_depth";
381 if (unlikely(eh->eh_max == 0)) {
382 error_msg = "invalid eh_max";
385 max = ext4_ext_max_entries(inode, depth);
386 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
387 error_msg = "too large eh_max";
390 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
391 error_msg = "invalid eh_entries";
394 if (!ext4_valid_extent_entries(inode, eh, depth)) {
395 error_msg = "invalid extent entries";
401 ext4_error_inode(inode, function, line, 0,
402 "bad header/extent: %s - magic %x, "
403 "entries %u, max %u(%u), depth %u(%u)",
404 error_msg, le16_to_cpu(eh->eh_magic),
405 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
406 max, le16_to_cpu(eh->eh_depth), depth);
411 #define ext4_ext_check(inode, eh, depth) \
412 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
414 int ext4_ext_check_inode(struct inode *inode)
416 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
420 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
422 int k, l = path->p_depth;
425 for (k = 0; k <= l; k++, path++) {
427 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
428 ext4_idx_pblock(path->p_idx));
429 } else if (path->p_ext) {
430 ext_debug(" %d:[%d]%d:%llu ",
431 le32_to_cpu(path->p_ext->ee_block),
432 ext4_ext_is_uninitialized(path->p_ext),
433 ext4_ext_get_actual_len(path->p_ext),
434 ext4_ext_pblock(path->p_ext));
441 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
443 int depth = ext_depth(inode);
444 struct ext4_extent_header *eh;
445 struct ext4_extent *ex;
451 eh = path[depth].p_hdr;
452 ex = EXT_FIRST_EXTENT(eh);
454 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
456 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
457 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
458 ext4_ext_is_uninitialized(ex),
459 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
464 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
465 ext4_fsblk_t newblock, int level)
467 int depth = ext_depth(inode);
468 struct ext4_extent *ex;
470 if (depth != level) {
471 struct ext4_extent_idx *idx;
472 idx = path[level].p_idx;
473 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
474 ext_debug("%d: move %d:%llu in new index %llu\n", level,
475 le32_to_cpu(idx->ei_block),
476 ext4_idx_pblock(idx),
484 ex = path[depth].p_ext;
485 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
486 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
487 le32_to_cpu(ex->ee_block),
489 ext4_ext_is_uninitialized(ex),
490 ext4_ext_get_actual_len(ex),
497 #define ext4_ext_show_path(inode, path)
498 #define ext4_ext_show_leaf(inode, path)
499 #define ext4_ext_show_move(inode, path, newblock, level)
502 void ext4_ext_drop_refs(struct ext4_ext_path *path)
504 int depth = path->p_depth;
507 for (i = 0; i <= depth; i++, path++)
515 * ext4_ext_binsearch_idx:
516 * binary search for the closest index of the given block
517 * the header must be checked before calling this
520 ext4_ext_binsearch_idx(struct inode *inode,
521 struct ext4_ext_path *path, ext4_lblk_t block)
523 struct ext4_extent_header *eh = path->p_hdr;
524 struct ext4_extent_idx *r, *l, *m;
527 ext_debug("binsearch for %u(idx): ", block);
529 l = EXT_FIRST_INDEX(eh) + 1;
530 r = EXT_LAST_INDEX(eh);
533 if (block < le32_to_cpu(m->ei_block))
537 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
538 m, le32_to_cpu(m->ei_block),
539 r, le32_to_cpu(r->ei_block));
543 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
544 ext4_idx_pblock(path->p_idx));
546 #ifdef CHECK_BINSEARCH
548 struct ext4_extent_idx *chix, *ix;
551 chix = ix = EXT_FIRST_INDEX(eh);
552 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
554 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
555 printk(KERN_DEBUG "k=%d, ix=0x%p, "
557 ix, EXT_FIRST_INDEX(eh));
558 printk(KERN_DEBUG "%u <= %u\n",
559 le32_to_cpu(ix->ei_block),
560 le32_to_cpu(ix[-1].ei_block));
562 BUG_ON(k && le32_to_cpu(ix->ei_block)
563 <= le32_to_cpu(ix[-1].ei_block));
564 if (block < le32_to_cpu(ix->ei_block))
568 BUG_ON(chix != path->p_idx);
575 * ext4_ext_binsearch:
576 * binary search for closest extent of the given block
577 * the header must be checked before calling this
580 ext4_ext_binsearch(struct inode *inode,
581 struct ext4_ext_path *path, ext4_lblk_t block)
583 struct ext4_extent_header *eh = path->p_hdr;
584 struct ext4_extent *r, *l, *m;
586 if (eh->eh_entries == 0) {
588 * this leaf is empty:
589 * we get such a leaf in split/add case
594 ext_debug("binsearch for %u: ", block);
596 l = EXT_FIRST_EXTENT(eh) + 1;
597 r = EXT_LAST_EXTENT(eh);
601 if (block < le32_to_cpu(m->ee_block))
605 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
606 m, le32_to_cpu(m->ee_block),
607 r, le32_to_cpu(r->ee_block));
611 ext_debug(" -> %d:%llu:[%d]%d ",
612 le32_to_cpu(path->p_ext->ee_block),
613 ext4_ext_pblock(path->p_ext),
614 ext4_ext_is_uninitialized(path->p_ext),
615 ext4_ext_get_actual_len(path->p_ext));
617 #ifdef CHECK_BINSEARCH
619 struct ext4_extent *chex, *ex;
622 chex = ex = EXT_FIRST_EXTENT(eh);
623 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
624 BUG_ON(k && le32_to_cpu(ex->ee_block)
625 <= le32_to_cpu(ex[-1].ee_block));
626 if (block < le32_to_cpu(ex->ee_block))
630 BUG_ON(chex != path->p_ext);
636 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
638 struct ext4_extent_header *eh;
640 eh = ext_inode_hdr(inode);
643 eh->eh_magic = EXT4_EXT_MAGIC;
644 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
645 ext4_mark_inode_dirty(handle, inode);
646 ext4_ext_invalidate_cache(inode);
650 struct ext4_ext_path *
651 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
652 struct ext4_ext_path *path)
654 struct ext4_extent_header *eh;
655 struct buffer_head *bh;
656 short int depth, i, ppos = 0, alloc = 0;
659 eh = ext_inode_hdr(inode);
660 depth = ext_depth(inode);
662 /* account possible depth increase */
664 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
667 return ERR_PTR(-ENOMEM);
674 /* walk through the tree */
676 int need_to_validate = 0;
678 ext_debug("depth %d: num %d, max %d\n",
679 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
681 ext4_ext_binsearch_idx(inode, path + ppos, block);
682 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
683 path[ppos].p_depth = i;
684 path[ppos].p_ext = NULL;
686 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
691 if (!bh_uptodate_or_lock(bh)) {
692 trace_ext4_ext_load_extent(inode, block,
694 ret = bh_submit_read(bh);
699 /* validate the extent entries */
700 need_to_validate = 1;
702 eh = ext_block_hdr(bh);
704 if (unlikely(ppos > depth)) {
706 EXT4_ERROR_INODE(inode,
707 "ppos %d > depth %d", ppos, depth);
711 path[ppos].p_bh = bh;
712 path[ppos].p_hdr = eh;
715 ret = need_to_validate ? ext4_ext_check(inode, eh, i) : 0;
720 path[ppos].p_depth = i;
721 path[ppos].p_ext = NULL;
722 path[ppos].p_idx = NULL;
725 ext4_ext_binsearch(inode, path + ppos, block);
726 /* if not an empty leaf */
727 if (path[ppos].p_ext)
728 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
730 ext4_ext_show_path(inode, path);
735 ext4_ext_drop_refs(path);
742 * ext4_ext_insert_index:
743 * insert new index [@logical;@ptr] into the block at @curp;
744 * check where to insert: before @curp or after @curp
746 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
747 struct ext4_ext_path *curp,
748 int logical, ext4_fsblk_t ptr)
750 struct ext4_extent_idx *ix;
753 err = ext4_ext_get_access(handle, inode, curp);
757 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
758 EXT4_ERROR_INODE(inode,
759 "logical %d == ei_block %d!",
760 logical, le32_to_cpu(curp->p_idx->ei_block));
764 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
765 >= le16_to_cpu(curp->p_hdr->eh_max))) {
766 EXT4_ERROR_INODE(inode,
767 "eh_entries %d >= eh_max %d!",
768 le16_to_cpu(curp->p_hdr->eh_entries),
769 le16_to_cpu(curp->p_hdr->eh_max));
773 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
775 ext_debug("insert new index %d after: %llu\n", logical, ptr);
776 ix = curp->p_idx + 1;
779 ext_debug("insert new index %d before: %llu\n", logical, ptr);
783 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
786 ext_debug("insert new index %d: "
787 "move %d indices from 0x%p to 0x%p\n",
788 logical, len, ix, ix + 1);
789 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
792 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
793 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
797 ix->ei_block = cpu_to_le32(logical);
798 ext4_idx_store_pblock(ix, ptr);
799 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
801 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
802 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
806 err = ext4_ext_dirty(handle, inode, curp);
807 ext4_std_error(inode->i_sb, err);
814 * inserts new subtree into the path, using free index entry
816 * - allocates all needed blocks (new leaf and all intermediate index blocks)
817 * - makes decision where to split
818 * - moves remaining extents and index entries (right to the split point)
819 * into the newly allocated blocks
820 * - initializes subtree
822 static int ext4_ext_split(handle_t *handle, struct inode *inode,
824 struct ext4_ext_path *path,
825 struct ext4_extent *newext, int at)
827 struct buffer_head *bh = NULL;
828 int depth = ext_depth(inode);
829 struct ext4_extent_header *neh;
830 struct ext4_extent_idx *fidx;
832 ext4_fsblk_t newblock, oldblock;
834 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
837 /* make decision: where to split? */
838 /* FIXME: now decision is simplest: at current extent */
840 /* if current leaf will be split, then we should use
841 * border from split point */
842 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
843 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
846 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
847 border = path[depth].p_ext[1].ee_block;
848 ext_debug("leaf will be split."
849 " next leaf starts at %d\n",
850 le32_to_cpu(border));
852 border = newext->ee_block;
853 ext_debug("leaf will be added."
854 " next leaf starts at %d\n",
855 le32_to_cpu(border));
859 * If error occurs, then we break processing
860 * and mark filesystem read-only. index won't
861 * be inserted and tree will be in consistent
862 * state. Next mount will repair buffers too.
866 * Get array to track all allocated blocks.
867 * We need this to handle errors and free blocks
870 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
874 /* allocate all needed blocks */
875 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
876 for (a = 0; a < depth - at; a++) {
877 newblock = ext4_ext_new_meta_block(handle, inode, path,
878 newext, &err, flags);
881 ablocks[a] = newblock;
884 /* initialize new leaf */
885 newblock = ablocks[--a];
886 if (unlikely(newblock == 0)) {
887 EXT4_ERROR_INODE(inode, "newblock == 0!");
891 bh = sb_getblk(inode->i_sb, newblock);
898 err = ext4_journal_get_create_access(handle, bh);
902 neh = ext_block_hdr(bh);
904 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
905 neh->eh_magic = EXT4_EXT_MAGIC;
908 /* move remainder of path[depth] to the new leaf */
909 if (unlikely(path[depth].p_hdr->eh_entries !=
910 path[depth].p_hdr->eh_max)) {
911 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
912 path[depth].p_hdr->eh_entries,
913 path[depth].p_hdr->eh_max);
917 /* start copy from next extent */
918 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
919 ext4_ext_show_move(inode, path, newblock, depth);
921 struct ext4_extent *ex;
922 ex = EXT_FIRST_EXTENT(neh);
923 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
924 le16_add_cpu(&neh->eh_entries, m);
927 set_buffer_uptodate(bh);
930 err = ext4_handle_dirty_metadata(handle, inode, bh);
936 /* correct old leaf */
938 err = ext4_ext_get_access(handle, inode, path + depth);
941 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
942 err = ext4_ext_dirty(handle, inode, path + depth);
948 /* create intermediate indexes */
950 if (unlikely(k < 0)) {
951 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
956 ext_debug("create %d intermediate indices\n", k);
957 /* insert new index into current index block */
958 /* current depth stored in i var */
962 newblock = ablocks[--a];
963 bh = sb_getblk(inode->i_sb, newblock);
970 err = ext4_journal_get_create_access(handle, bh);
974 neh = ext_block_hdr(bh);
975 neh->eh_entries = cpu_to_le16(1);
976 neh->eh_magic = EXT4_EXT_MAGIC;
977 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
978 neh->eh_depth = cpu_to_le16(depth - i);
979 fidx = EXT_FIRST_INDEX(neh);
980 fidx->ei_block = border;
981 ext4_idx_store_pblock(fidx, oldblock);
983 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
984 i, newblock, le32_to_cpu(border), oldblock);
986 /* move remainder of path[i] to the new index block */
987 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
988 EXT_LAST_INDEX(path[i].p_hdr))) {
989 EXT4_ERROR_INODE(inode,
990 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
991 le32_to_cpu(path[i].p_ext->ee_block));
995 /* start copy indexes */
996 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
997 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
998 EXT_MAX_INDEX(path[i].p_hdr));
999 ext4_ext_show_move(inode, path, newblock, i);
1001 memmove(++fidx, path[i].p_idx,
1002 sizeof(struct ext4_extent_idx) * m);
1003 le16_add_cpu(&neh->eh_entries, m);
1005 set_buffer_uptodate(bh);
1008 err = ext4_handle_dirty_metadata(handle, inode, bh);
1014 /* correct old index */
1016 err = ext4_ext_get_access(handle, inode, path + i);
1019 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1020 err = ext4_ext_dirty(handle, inode, path + i);
1028 /* insert new index */
1029 err = ext4_ext_insert_index(handle, inode, path + at,
1030 le32_to_cpu(border), newblock);
1034 if (buffer_locked(bh))
1040 /* free all allocated blocks in error case */
1041 for (i = 0; i < depth; i++) {
1044 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1045 EXT4_FREE_BLOCKS_METADATA);
1054 * ext4_ext_grow_indepth:
1055 * implements tree growing procedure:
1056 * - allocates new block
1057 * - moves top-level data (index block or leaf) into the new block
1058 * - initializes new top-level, creating index that points to the
1059 * just created block
1061 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1063 struct ext4_extent *newext)
1065 struct ext4_extent_header *neh;
1066 struct buffer_head *bh;
1067 ext4_fsblk_t newblock;
1070 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1071 newext, &err, flags);
1075 bh = sb_getblk(inode->i_sb, newblock);
1080 err = ext4_journal_get_create_access(handle, bh);
1086 /* move top-level index/leaf into new block */
1087 memmove(bh->b_data, EXT4_I(inode)->i_data,
1088 sizeof(EXT4_I(inode)->i_data));
1090 /* set size of new block */
1091 neh = ext_block_hdr(bh);
1092 /* old root could have indexes or leaves
1093 * so calculate e_max right way */
1094 if (ext_depth(inode))
1095 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1097 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1098 neh->eh_magic = EXT4_EXT_MAGIC;
1099 set_buffer_uptodate(bh);
1102 err = ext4_handle_dirty_metadata(handle, inode, bh);
1106 /* Update top-level index: num,max,pointer */
1107 neh = ext_inode_hdr(inode);
1108 neh->eh_entries = cpu_to_le16(1);
1109 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1110 if (neh->eh_depth == 0) {
1111 /* Root extent block becomes index block */
1112 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1113 EXT_FIRST_INDEX(neh)->ei_block =
1114 EXT_FIRST_EXTENT(neh)->ee_block;
1116 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1117 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1118 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1119 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1121 neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
1122 ext4_mark_inode_dirty(handle, inode);
1130 * ext4_ext_create_new_leaf:
1131 * finds empty index and adds new leaf.
1132 * if no free index is found, then it requests in-depth growing.
1134 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1136 struct ext4_ext_path *path,
1137 struct ext4_extent *newext)
1139 struct ext4_ext_path *curp;
1140 int depth, i, err = 0;
1143 i = depth = ext_depth(inode);
1145 /* walk up to the tree and look for free index entry */
1146 curp = path + depth;
1147 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1152 /* we use already allocated block for index block,
1153 * so subsequent data blocks should be contiguous */
1154 if (EXT_HAS_FREE_INDEX(curp)) {
1155 /* if we found index with free entry, then use that
1156 * entry: create all needed subtree and add new leaf */
1157 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1162 ext4_ext_drop_refs(path);
1163 path = ext4_ext_find_extent(inode,
1164 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1167 err = PTR_ERR(path);
1169 /* tree is full, time to grow in depth */
1170 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
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 ? le32_to_cpu(ix->ei_block) : 0,
1243 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1244 le32_to_cpu(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 largest 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 path[depth].p_ext !=
1404 EXT_LAST_EXTENT(path[depth].p_hdr))
1405 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1408 if (path[depth].p_idx !=
1409 EXT_LAST_INDEX(path[depth].p_hdr))
1410 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1415 return EXT_MAX_BLOCKS;
1419 * ext4_ext_next_leaf_block:
1420 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1422 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1426 BUG_ON(path == NULL);
1427 depth = path->p_depth;
1429 /* zero-tree has no leaf blocks at all */
1431 return EXT_MAX_BLOCKS;
1433 /* go to index block */
1436 while (depth >= 0) {
1437 if (path[depth].p_idx !=
1438 EXT_LAST_INDEX(path[depth].p_hdr))
1439 return (ext4_lblk_t)
1440 le32_to_cpu(path[depth].p_idx[1].ei_block);
1444 return EXT_MAX_BLOCKS;
1448 * ext4_ext_correct_indexes:
1449 * if leaf gets modified and modified extent is first in the leaf,
1450 * then we have to correct all indexes above.
1451 * TODO: do we need to correct tree in all cases?
1453 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1454 struct ext4_ext_path *path)
1456 struct ext4_extent_header *eh;
1457 int depth = ext_depth(inode);
1458 struct ext4_extent *ex;
1462 eh = path[depth].p_hdr;
1463 ex = path[depth].p_ext;
1465 if (unlikely(ex == NULL || eh == NULL)) {
1466 EXT4_ERROR_INODE(inode,
1467 "ex %p == NULL or eh %p == NULL", ex, eh);
1472 /* there is no tree at all */
1476 if (ex != EXT_FIRST_EXTENT(eh)) {
1477 /* we correct tree if first leaf got modified only */
1482 * TODO: we need correction if border is smaller than current one
1485 border = path[depth].p_ext->ee_block;
1486 err = ext4_ext_get_access(handle, inode, path + k);
1489 path[k].p_idx->ei_block = border;
1490 err = ext4_ext_dirty(handle, inode, path + k);
1495 /* change all left-side indexes */
1496 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1498 err = ext4_ext_get_access(handle, inode, path + k);
1501 path[k].p_idx->ei_block = border;
1502 err = ext4_ext_dirty(handle, inode, path + k);
1511 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1512 struct ext4_extent *ex2)
1514 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1517 * Make sure that either both extents are uninitialized, or
1520 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1523 if (ext4_ext_is_uninitialized(ex1))
1524 max_len = EXT_UNINIT_MAX_LEN;
1526 max_len = EXT_INIT_MAX_LEN;
1528 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1529 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1531 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1532 le32_to_cpu(ex2->ee_block))
1536 * To allow future support for preallocated extents to be added
1537 * as an RO_COMPAT feature, refuse to merge to extents if
1538 * this can result in the top bit of ee_len being set.
1540 if (ext1_ee_len + ext2_ee_len > max_len)
1542 #ifdef AGGRESSIVE_TEST
1543 if (ext1_ee_len >= 4)
1547 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1553 * This function tries to merge the "ex" extent to the next extent in the tree.
1554 * It always tries to merge towards right. If you want to merge towards
1555 * left, pass "ex - 1" as argument instead of "ex".
1556 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1557 * 1 if they got merged.
1559 static int ext4_ext_try_to_merge_right(struct inode *inode,
1560 struct ext4_ext_path *path,
1561 struct ext4_extent *ex)
1563 struct ext4_extent_header *eh;
1564 unsigned int depth, len;
1566 int uninitialized = 0;
1568 depth = ext_depth(inode);
1569 BUG_ON(path[depth].p_hdr == NULL);
1570 eh = path[depth].p_hdr;
1572 while (ex < EXT_LAST_EXTENT(eh)) {
1573 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1575 /* merge with next extent! */
1576 if (ext4_ext_is_uninitialized(ex))
1578 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1579 + ext4_ext_get_actual_len(ex + 1));
1581 ext4_ext_mark_uninitialized(ex);
1583 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1584 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1585 * sizeof(struct ext4_extent);
1586 memmove(ex + 1, ex + 2, len);
1588 le16_add_cpu(&eh->eh_entries, -1);
1590 WARN_ON(eh->eh_entries == 0);
1591 if (!eh->eh_entries)
1592 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1599 * This function tries to merge the @ex extent to neighbours in the tree.
1600 * return 1 if merge left else 0.
1602 static int ext4_ext_try_to_merge(struct inode *inode,
1603 struct ext4_ext_path *path,
1604 struct ext4_extent *ex) {
1605 struct ext4_extent_header *eh;
1610 depth = ext_depth(inode);
1611 BUG_ON(path[depth].p_hdr == NULL);
1612 eh = path[depth].p_hdr;
1614 if (ex > EXT_FIRST_EXTENT(eh))
1615 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1618 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1624 * check if a portion of the "newext" extent overlaps with an
1627 * If there is an overlap discovered, it updates the length of the newext
1628 * such that there will be no overlap, and then returns 1.
1629 * If there is no overlap found, it returns 0.
1631 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1632 struct inode *inode,
1633 struct ext4_extent *newext,
1634 struct ext4_ext_path *path)
1637 unsigned int depth, len1;
1638 unsigned int ret = 0;
1640 b1 = le32_to_cpu(newext->ee_block);
1641 len1 = ext4_ext_get_actual_len(newext);
1642 depth = ext_depth(inode);
1643 if (!path[depth].p_ext)
1645 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1646 b2 &= ~(sbi->s_cluster_ratio - 1);
1649 * get the next allocated block if the extent in the path
1650 * is before the requested block(s)
1653 b2 = ext4_ext_next_allocated_block(path);
1654 if (b2 == EXT_MAX_BLOCKS)
1656 b2 &= ~(sbi->s_cluster_ratio - 1);
1659 /* check for wrap through zero on extent logical start block*/
1660 if (b1 + len1 < b1) {
1661 len1 = EXT_MAX_BLOCKS - b1;
1662 newext->ee_len = cpu_to_le16(len1);
1666 /* check for overlap */
1667 if (b1 + len1 > b2) {
1668 newext->ee_len = cpu_to_le16(b2 - b1);
1676 * ext4_ext_insert_extent:
1677 * tries to merge requsted extent into the existing extent or
1678 * inserts requested extent as new one into the tree,
1679 * creating new leaf in the no-space case.
1681 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1682 struct ext4_ext_path *path,
1683 struct ext4_extent *newext, int flag)
1685 struct ext4_extent_header *eh;
1686 struct ext4_extent *ex, *fex;
1687 struct ext4_extent *nearex; /* nearest extent */
1688 struct ext4_ext_path *npath = NULL;
1689 int depth, len, err;
1691 unsigned uninitialized = 0;
1694 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1695 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1698 depth = ext_depth(inode);
1699 ex = path[depth].p_ext;
1700 if (unlikely(path[depth].p_hdr == NULL)) {
1701 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1705 /* try to insert block into found extent and return */
1706 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1707 && ext4_can_extents_be_merged(inode, ex, newext)) {
1708 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1709 ext4_ext_is_uninitialized(newext),
1710 ext4_ext_get_actual_len(newext),
1711 le32_to_cpu(ex->ee_block),
1712 ext4_ext_is_uninitialized(ex),
1713 ext4_ext_get_actual_len(ex),
1714 ext4_ext_pblock(ex));
1715 err = ext4_ext_get_access(handle, inode, path + depth);
1720 * ext4_can_extents_be_merged should have checked that either
1721 * both extents are uninitialized, or both aren't. Thus we
1722 * need to check only one of them here.
1724 if (ext4_ext_is_uninitialized(ex))
1726 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1727 + ext4_ext_get_actual_len(newext));
1729 ext4_ext_mark_uninitialized(ex);
1730 eh = path[depth].p_hdr;
1735 depth = ext_depth(inode);
1736 eh = path[depth].p_hdr;
1737 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1740 /* probably next leaf has space for us? */
1741 fex = EXT_LAST_EXTENT(eh);
1742 next = EXT_MAX_BLOCKS;
1743 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1744 next = ext4_ext_next_leaf_block(path);
1745 if (next != EXT_MAX_BLOCKS) {
1746 ext_debug("next leaf block - %u\n", next);
1747 BUG_ON(npath != NULL);
1748 npath = ext4_ext_find_extent(inode, next, NULL);
1750 return PTR_ERR(npath);
1751 BUG_ON(npath->p_depth != path->p_depth);
1752 eh = npath[depth].p_hdr;
1753 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1754 ext_debug("next leaf isn't full(%d)\n",
1755 le16_to_cpu(eh->eh_entries));
1759 ext_debug("next leaf has no free space(%d,%d)\n",
1760 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1764 * There is no free space in the found leaf.
1765 * We're gonna add a new leaf in the tree.
1767 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1768 flags = EXT4_MB_USE_ROOT_BLOCKS;
1769 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1772 depth = ext_depth(inode);
1773 eh = path[depth].p_hdr;
1776 nearex = path[depth].p_ext;
1778 err = ext4_ext_get_access(handle, inode, path + depth);
1783 /* there is no extent in this leaf, create first one */
1784 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1785 le32_to_cpu(newext->ee_block),
1786 ext4_ext_pblock(newext),
1787 ext4_ext_is_uninitialized(newext),
1788 ext4_ext_get_actual_len(newext));
1789 nearex = EXT_FIRST_EXTENT(eh);
1791 if (le32_to_cpu(newext->ee_block)
1792 > le32_to_cpu(nearex->ee_block)) {
1794 ext_debug("insert %u:%llu:[%d]%d before: "
1796 le32_to_cpu(newext->ee_block),
1797 ext4_ext_pblock(newext),
1798 ext4_ext_is_uninitialized(newext),
1799 ext4_ext_get_actual_len(newext),
1804 BUG_ON(newext->ee_block == nearex->ee_block);
1805 ext_debug("insert %u:%llu:[%d]%d after: "
1807 le32_to_cpu(newext->ee_block),
1808 ext4_ext_pblock(newext),
1809 ext4_ext_is_uninitialized(newext),
1810 ext4_ext_get_actual_len(newext),
1813 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1815 ext_debug("insert %u:%llu:[%d]%d: "
1816 "move %d extents from 0x%p to 0x%p\n",
1817 le32_to_cpu(newext->ee_block),
1818 ext4_ext_pblock(newext),
1819 ext4_ext_is_uninitialized(newext),
1820 ext4_ext_get_actual_len(newext),
1821 len, nearex, nearex + 1);
1822 memmove(nearex + 1, nearex,
1823 len * sizeof(struct ext4_extent));
1827 le16_add_cpu(&eh->eh_entries, 1);
1828 path[depth].p_ext = nearex;
1829 nearex->ee_block = newext->ee_block;
1830 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1831 nearex->ee_len = newext->ee_len;
1834 /* try to merge extents to the right */
1835 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1836 ext4_ext_try_to_merge(inode, path, nearex);
1838 /* try to merge extents to the left */
1840 /* time to correct all indexes above */
1841 err = ext4_ext_correct_indexes(handle, inode, path);
1845 err = ext4_ext_dirty(handle, inode, path + depth);
1849 ext4_ext_drop_refs(npath);
1852 ext4_ext_invalidate_cache(inode);
1856 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1857 ext4_lblk_t num, ext_prepare_callback func,
1860 struct ext4_ext_path *path = NULL;
1861 struct ext4_ext_cache cbex;
1862 struct ext4_extent *ex;
1863 ext4_lblk_t next, start = 0, end = 0;
1864 ext4_lblk_t last = block + num;
1865 int depth, exists, err = 0;
1867 BUG_ON(func == NULL);
1868 BUG_ON(inode == NULL);
1870 while (block < last && block != EXT_MAX_BLOCKS) {
1872 /* find extent for this block */
1873 down_read(&EXT4_I(inode)->i_data_sem);
1874 path = ext4_ext_find_extent(inode, block, path);
1875 up_read(&EXT4_I(inode)->i_data_sem);
1877 err = PTR_ERR(path);
1882 depth = ext_depth(inode);
1883 if (unlikely(path[depth].p_hdr == NULL)) {
1884 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1888 ex = path[depth].p_ext;
1889 next = ext4_ext_next_allocated_block(path);
1893 /* there is no extent yet, so try to allocate
1894 * all requested space */
1897 } else if (le32_to_cpu(ex->ee_block) > block) {
1898 /* need to allocate space before found extent */
1900 end = le32_to_cpu(ex->ee_block);
1901 if (block + num < end)
1903 } else if (block >= le32_to_cpu(ex->ee_block)
1904 + ext4_ext_get_actual_len(ex)) {
1905 /* need to allocate space after found extent */
1910 } else if (block >= le32_to_cpu(ex->ee_block)) {
1912 * some part of requested space is covered
1916 end = le32_to_cpu(ex->ee_block)
1917 + ext4_ext_get_actual_len(ex);
1918 if (block + num < end)
1924 BUG_ON(end <= start);
1927 cbex.ec_block = start;
1928 cbex.ec_len = end - start;
1931 cbex.ec_block = le32_to_cpu(ex->ee_block);
1932 cbex.ec_len = ext4_ext_get_actual_len(ex);
1933 cbex.ec_start = ext4_ext_pblock(ex);
1936 if (unlikely(cbex.ec_len == 0)) {
1937 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1941 err = func(inode, next, &cbex, ex, cbdata);
1942 ext4_ext_drop_refs(path);
1947 if (err == EXT_REPEAT)
1949 else if (err == EXT_BREAK) {
1954 if (ext_depth(inode) != depth) {
1955 /* depth was changed. we have to realloc path */
1960 block = cbex.ec_block + cbex.ec_len;
1964 ext4_ext_drop_refs(path);
1972 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1973 __u32 len, ext4_fsblk_t start)
1975 struct ext4_ext_cache *cex;
1977 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1978 trace_ext4_ext_put_in_cache(inode, block, len, start);
1979 cex = &EXT4_I(inode)->i_cached_extent;
1980 cex->ec_block = block;
1982 cex->ec_start = start;
1983 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1987 * ext4_ext_put_gap_in_cache:
1988 * calculate boundaries of the gap that the requested block fits into
1989 * and cache this gap
1992 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1995 int depth = ext_depth(inode);
1998 struct ext4_extent *ex;
2000 ex = path[depth].p_ext;
2002 /* there is no extent yet, so gap is [0;-] */
2004 len = EXT_MAX_BLOCKS;
2005 ext_debug("cache gap(whole file):");
2006 } else if (block < le32_to_cpu(ex->ee_block)) {
2008 len = le32_to_cpu(ex->ee_block) - block;
2009 ext_debug("cache gap(before): %u [%u:%u]",
2011 le32_to_cpu(ex->ee_block),
2012 ext4_ext_get_actual_len(ex));
2013 } else if (block >= le32_to_cpu(ex->ee_block)
2014 + ext4_ext_get_actual_len(ex)) {
2016 lblock = le32_to_cpu(ex->ee_block)
2017 + ext4_ext_get_actual_len(ex);
2019 next = ext4_ext_next_allocated_block(path);
2020 ext_debug("cache gap(after): [%u:%u] %u",
2021 le32_to_cpu(ex->ee_block),
2022 ext4_ext_get_actual_len(ex),
2024 BUG_ON(next == lblock);
2025 len = next - lblock;
2031 ext_debug(" -> %u:%lu\n", lblock, len);
2032 ext4_ext_put_in_cache(inode, lblock, len, 0);
2036 * ext4_ext_check_cache()
2037 * Checks to see if the given block is in the cache.
2038 * If it is, the cached extent is stored in the given
2039 * cache extent pointer. If the cached extent is a hole,
2040 * this routine should be used instead of
2041 * ext4_ext_in_cache if the calling function needs to
2042 * know the size of the hole.
2044 * @inode: The files inode
2045 * @block: The block to look for in the cache
2046 * @ex: Pointer where the cached extent will be stored
2047 * if it contains block
2049 * Return 0 if cache is invalid; 1 if the cache is valid
2051 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2052 struct ext4_ext_cache *ex){
2053 struct ext4_ext_cache *cex;
2054 struct ext4_sb_info *sbi;
2058 * We borrow i_block_reservation_lock to protect i_cached_extent
2060 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2061 cex = &EXT4_I(inode)->i_cached_extent;
2062 sbi = EXT4_SB(inode->i_sb);
2064 /* has cache valid data? */
2065 if (cex->ec_len == 0)
2068 if (in_range(block, cex->ec_block, cex->ec_len)) {
2069 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2070 ext_debug("%u cached by %u:%u:%llu\n",
2072 cex->ec_block, cex->ec_len, cex->ec_start);
2076 trace_ext4_ext_in_cache(inode, block, ret);
2077 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2082 * ext4_ext_in_cache()
2083 * Checks to see if the given block is in the cache.
2084 * If it is, the cached extent is stored in the given
2087 * @inode: The files inode
2088 * @block: The block to look for in the cache
2089 * @ex: Pointer where the cached extent will be stored
2090 * if it contains block
2092 * Return 0 if cache is invalid; 1 if the cache is valid
2095 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2096 struct ext4_extent *ex)
2098 struct ext4_ext_cache cex;
2101 if (ext4_ext_check_cache(inode, block, &cex)) {
2102 ex->ee_block = cpu_to_le32(cex.ec_block);
2103 ext4_ext_store_pblock(ex, cex.ec_start);
2104 ex->ee_len = cpu_to_le16(cex.ec_len);
2114 * removes index from the index block.
2116 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2117 struct ext4_ext_path *path, int depth)
2122 /* free index block */
2124 path = path + depth;
2125 leaf = ext4_idx_pblock(path->p_idx);
2126 if (unlikely(path->p_hdr->eh_entries == 0)) {
2127 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2130 err = ext4_ext_get_access(handle, inode, path);
2134 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2135 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2136 len *= sizeof(struct ext4_extent_idx);
2137 memmove(path->p_idx, path->p_idx + 1, len);
2140 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2141 err = ext4_ext_dirty(handle, inode, path);
2144 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2145 trace_ext4_ext_rm_idx(inode, leaf);
2147 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2148 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2150 while (--depth >= 0) {
2151 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2154 err = ext4_ext_get_access(handle, inode, path);
2157 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2158 err = ext4_ext_dirty(handle, inode, path);
2166 * ext4_ext_calc_credits_for_single_extent:
2167 * This routine returns max. credits that needed to insert an extent
2168 * to the extent tree.
2169 * When pass the actual path, the caller should calculate credits
2172 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2173 struct ext4_ext_path *path)
2176 int depth = ext_depth(inode);
2179 /* probably there is space in leaf? */
2180 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2181 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2184 * There are some space in the leaf tree, no
2185 * need to account for leaf block credit
2187 * bitmaps and block group descriptor blocks
2188 * and other metadata blocks still need to be
2191 /* 1 bitmap, 1 block group descriptor */
2192 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2197 return ext4_chunk_trans_blocks(inode, nrblocks);
2201 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2203 * if nrblocks are fit in a single extent (chunk flag is 1), then
2204 * in the worse case, each tree level index/leaf need to be changed
2205 * if the tree split due to insert a new extent, then the old tree
2206 * index/leaf need to be updated too
2208 * If the nrblocks are discontiguous, they could cause
2209 * the whole tree split more than once, but this is really rare.
2211 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2214 int depth = ext_depth(inode);
2224 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2225 struct ext4_extent *ex,
2226 ext4_fsblk_t *partial_cluster,
2227 ext4_lblk_t from, ext4_lblk_t to)
2229 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2230 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2232 int flags = EXT4_FREE_BLOCKS_FORGET;
2234 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2235 flags |= EXT4_FREE_BLOCKS_METADATA;
2237 * For bigalloc file systems, we never free a partial cluster
2238 * at the beginning of the extent. Instead, we make a note
2239 * that we tried freeing the cluster, and check to see if we
2240 * need to free it on a subsequent call to ext4_remove_blocks,
2241 * or at the end of the ext4_truncate() operation.
2243 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2245 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2247 * If we have a partial cluster, and it's different from the
2248 * cluster of the last block, we need to explicitly free the
2249 * partial cluster here.
2251 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2252 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2253 ext4_free_blocks(handle, inode, NULL,
2254 EXT4_C2B(sbi, *partial_cluster),
2255 sbi->s_cluster_ratio, flags);
2256 *partial_cluster = 0;
2259 #ifdef EXTENTS_STATS
2261 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2262 spin_lock(&sbi->s_ext_stats_lock);
2263 sbi->s_ext_blocks += ee_len;
2264 sbi->s_ext_extents++;
2265 if (ee_len < sbi->s_ext_min)
2266 sbi->s_ext_min = ee_len;
2267 if (ee_len > sbi->s_ext_max)
2268 sbi->s_ext_max = ee_len;
2269 if (ext_depth(inode) > sbi->s_depth_max)
2270 sbi->s_depth_max = ext_depth(inode);
2271 spin_unlock(&sbi->s_ext_stats_lock);
2274 if (from >= le32_to_cpu(ex->ee_block)
2275 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2279 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2280 pblk = ext4_ext_pblock(ex) + ee_len - num;
2281 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2282 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2284 * If the block range to be freed didn't start at the
2285 * beginning of a cluster, and we removed the entire
2286 * extent, save the partial cluster here, since we
2287 * might need to delete if we determine that the
2288 * truncate operation has removed all of the blocks in
2291 if (pblk & (sbi->s_cluster_ratio - 1) &&
2293 *partial_cluster = EXT4_B2C(sbi, pblk);
2295 *partial_cluster = 0;
2296 } else if (from == le32_to_cpu(ex->ee_block)
2297 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2303 start = ext4_ext_pblock(ex);
2305 ext_debug("free first %u blocks starting %llu\n", num, start);
2306 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2309 printk(KERN_INFO "strange request: removal(2) "
2310 "%u-%u from %u:%u\n",
2311 from, to, le32_to_cpu(ex->ee_block), ee_len);
2318 * ext4_ext_rm_leaf() Removes the extents associated with the
2319 * blocks appearing between "start" and "end", and splits the extents
2320 * if "start" and "end" appear in the same extent
2322 * @handle: The journal handle
2323 * @inode: The files inode
2324 * @path: The path to the leaf
2325 * @start: The first block to remove
2326 * @end: The last block to remove
2329 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2330 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2331 ext4_lblk_t start, ext4_lblk_t end)
2333 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2334 int err = 0, correct_index = 0;
2335 int depth = ext_depth(inode), credits;
2336 struct ext4_extent_header *eh;
2339 ext4_lblk_t ex_ee_block;
2340 unsigned short ex_ee_len;
2341 unsigned uninitialized = 0;
2342 struct ext4_extent *ex;
2344 /* the header must be checked already in ext4_ext_remove_space() */
2345 ext_debug("truncate since %u in leaf to %u\n", start, end);
2346 if (!path[depth].p_hdr)
2347 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2348 eh = path[depth].p_hdr;
2349 if (unlikely(path[depth].p_hdr == NULL)) {
2350 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2353 /* find where to start removing */
2354 ex = EXT_LAST_EXTENT(eh);
2356 ex_ee_block = le32_to_cpu(ex->ee_block);
2357 ex_ee_len = ext4_ext_get_actual_len(ex);
2359 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2361 while (ex >= EXT_FIRST_EXTENT(eh) &&
2362 ex_ee_block + ex_ee_len > start) {
2364 if (ext4_ext_is_uninitialized(ex))
2369 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2370 uninitialized, ex_ee_len);
2371 path[depth].p_ext = ex;
2373 a = ex_ee_block > start ? ex_ee_block : start;
2374 b = ex_ee_block+ex_ee_len - 1 < end ?
2375 ex_ee_block+ex_ee_len - 1 : end;
2377 ext_debug(" border %u:%u\n", a, b);
2379 /* If this extent is beyond the end of the hole, skip it */
2380 if (end < ex_ee_block) {
2382 ex_ee_block = le32_to_cpu(ex->ee_block);
2383 ex_ee_len = ext4_ext_get_actual_len(ex);
2385 } else if (b != ex_ee_block + ex_ee_len - 1) {
2386 EXT4_ERROR_INODE(inode," bad truncate %u:%u\n",
2390 } else if (a != ex_ee_block) {
2391 /* remove tail of the extent */
2392 num = a - ex_ee_block;
2394 /* remove whole extent: excellent! */
2398 * 3 for leaf, sb, and inode plus 2 (bmap and group
2399 * descriptor) for each block group; assume two block
2400 * groups plus ex_ee_len/blocks_per_block_group for
2403 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2404 if (ex == EXT_FIRST_EXTENT(eh)) {
2406 credits += (ext_depth(inode)) + 1;
2408 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2410 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2414 err = ext4_ext_get_access(handle, inode, path + depth);
2418 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2424 /* this extent is removed; mark slot entirely unused */
2425 ext4_ext_store_pblock(ex, 0);
2427 ex->ee_len = cpu_to_le16(num);
2429 * Do not mark uninitialized if all the blocks in the
2430 * extent have been removed.
2432 if (uninitialized && num)
2433 ext4_ext_mark_uninitialized(ex);
2435 * If the extent was completely released,
2436 * we need to remove it from the leaf
2439 if (end != EXT_MAX_BLOCKS - 1) {
2441 * For hole punching, we need to scoot all the
2442 * extents up when an extent is removed so that
2443 * we dont have blank extents in the middle
2445 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2446 sizeof(struct ext4_extent));
2448 /* Now get rid of the one at the end */
2449 memset(EXT_LAST_EXTENT(eh), 0,
2450 sizeof(struct ext4_extent));
2452 le16_add_cpu(&eh->eh_entries, -1);
2454 *partial_cluster = 0;
2456 err = ext4_ext_dirty(handle, inode, path + depth);
2460 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2461 ext4_ext_pblock(ex));
2463 ex_ee_block = le32_to_cpu(ex->ee_block);
2464 ex_ee_len = ext4_ext_get_actual_len(ex);
2467 if (correct_index && eh->eh_entries)
2468 err = ext4_ext_correct_indexes(handle, inode, path);
2471 * If there is still a entry in the leaf node, check to see if
2472 * it references the partial cluster. This is the only place
2473 * where it could; if it doesn't, we can free the cluster.
2475 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2476 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2477 *partial_cluster)) {
2478 int flags = EXT4_FREE_BLOCKS_FORGET;
2480 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2481 flags |= EXT4_FREE_BLOCKS_METADATA;
2483 ext4_free_blocks(handle, inode, NULL,
2484 EXT4_C2B(sbi, *partial_cluster),
2485 sbi->s_cluster_ratio, flags);
2486 *partial_cluster = 0;
2489 /* if this leaf is free, then we should
2490 * remove it from index block above */
2491 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2492 err = ext4_ext_rm_idx(handle, inode, path, depth);
2499 * ext4_ext_more_to_rm:
2500 * returns 1 if current index has to be freed (even partial)
2503 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2505 BUG_ON(path->p_idx == NULL);
2507 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2511 * if truncate on deeper level happened, it wasn't partial,
2512 * so we have to consider current index for truncation
2514 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2519 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2522 struct super_block *sb = inode->i_sb;
2523 int depth = ext_depth(inode);
2524 struct ext4_ext_path *path = NULL;
2525 ext4_fsblk_t partial_cluster = 0;
2529 ext_debug("truncate since %u to %u\n", start, end);
2531 /* probably first extent we're gonna free will be last in block */
2532 handle = ext4_journal_start(inode, depth + 1);
2534 return PTR_ERR(handle);
2537 ext4_ext_invalidate_cache(inode);
2539 trace_ext4_ext_remove_space(inode, start, depth);
2542 * Check if we are removing extents inside the extent tree. If that
2543 * is the case, we are going to punch a hole inside the extent tree
2544 * so we have to check whether we need to split the extent covering
2545 * the last block to remove so we can easily remove the part of it
2546 * in ext4_ext_rm_leaf().
2548 if (end < EXT_MAX_BLOCKS - 1) {
2549 struct ext4_extent *ex;
2550 ext4_lblk_t ee_block;
2552 /* find extent for this block */
2553 path = ext4_ext_find_extent(inode, end, NULL);
2555 ext4_journal_stop(handle);
2556 return PTR_ERR(path);
2558 depth = ext_depth(inode);
2559 ex = path[depth].p_ext;
2561 ext4_ext_drop_refs(path);
2567 ee_block = le32_to_cpu(ex->ee_block);
2570 * See if the last block is inside the extent, if so split
2571 * the extent at 'end' block so we can easily remove the
2572 * tail of the first part of the split extent in
2573 * ext4_ext_rm_leaf().
2575 if (end >= ee_block &&
2576 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2579 if (ext4_ext_is_uninitialized(ex))
2580 split_flag = EXT4_EXT_MARK_UNINIT1 |
2581 EXT4_EXT_MARK_UNINIT2;
2584 * Split the extent in two so that 'end' is the last
2585 * block in the first new extent
2587 err = ext4_split_extent_at(handle, inode, path,
2588 end + 1, split_flag,
2589 EXT4_GET_BLOCKS_PRE_IO |
2590 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2599 * We start scanning from right side, freeing all the blocks
2600 * after i_size and walking into the tree depth-wise.
2602 depth = ext_depth(inode);
2607 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2609 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2612 ext4_journal_stop(handle);
2615 path[0].p_depth = depth;
2616 path[0].p_hdr = ext_inode_hdr(inode);
2618 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2625 while (i >= 0 && err == 0) {
2627 /* this is leaf block */
2628 err = ext4_ext_rm_leaf(handle, inode, path,
2629 &partial_cluster, start,
2631 /* root level has p_bh == NULL, brelse() eats this */
2632 brelse(path[i].p_bh);
2633 path[i].p_bh = NULL;
2638 /* this is index block */
2639 if (!path[i].p_hdr) {
2640 ext_debug("initialize header\n");
2641 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2644 if (!path[i].p_idx) {
2645 /* this level hasn't been touched yet */
2646 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2647 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2648 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2650 le16_to_cpu(path[i].p_hdr->eh_entries));
2652 /* we were already here, see at next index */
2656 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2657 i, EXT_FIRST_INDEX(path[i].p_hdr),
2659 if (ext4_ext_more_to_rm(path + i)) {
2660 struct buffer_head *bh;
2661 /* go to the next level */
2662 ext_debug("move to level %d (block %llu)\n",
2663 i + 1, ext4_idx_pblock(path[i].p_idx));
2664 memset(path + i + 1, 0, sizeof(*path));
2665 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2667 /* should we reset i_size? */
2671 if (WARN_ON(i + 1 > depth)) {
2675 if (ext4_ext_check(inode, ext_block_hdr(bh),
2680 path[i + 1].p_bh = bh;
2682 /* save actual number of indexes since this
2683 * number is changed at the next iteration */
2684 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2687 /* we finished processing this index, go up */
2688 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2689 /* index is empty, remove it;
2690 * handle must be already prepared by the
2691 * truncatei_leaf() */
2692 err = ext4_ext_rm_idx(handle, inode, path, i);
2694 /* root level has p_bh == NULL, brelse() eats this */
2695 brelse(path[i].p_bh);
2696 path[i].p_bh = NULL;
2698 ext_debug("return to level %d\n", i);
2702 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2703 path->p_hdr->eh_entries);
2705 /* If we still have something in the partial cluster and we have removed
2706 * even the first extent, then we should free the blocks in the partial
2707 * cluster as well. */
2708 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2709 int flags = EXT4_FREE_BLOCKS_FORGET;
2711 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2712 flags |= EXT4_FREE_BLOCKS_METADATA;
2714 ext4_free_blocks(handle, inode, NULL,
2715 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2716 EXT4_SB(sb)->s_cluster_ratio, flags);
2717 partial_cluster = 0;
2720 /* TODO: flexible tree reduction should be here */
2721 if (path->p_hdr->eh_entries == 0) {
2723 * truncate to zero freed all the tree,
2724 * so we need to correct eh_depth
2726 err = ext4_ext_get_access(handle, inode, path);
2728 ext_inode_hdr(inode)->eh_depth = 0;
2729 ext_inode_hdr(inode)->eh_max =
2730 cpu_to_le16(ext4_ext_space_root(inode, 0));
2731 err = ext4_ext_dirty(handle, inode, path);
2735 ext4_ext_drop_refs(path);
2737 if (err == -EAGAIN) {
2741 ext4_journal_stop(handle);
2747 * called at mount time
2749 void ext4_ext_init(struct super_block *sb)
2752 * possible initialization would be here
2755 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2756 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2757 printk(KERN_INFO "EXT4-fs: file extents enabled");
2758 #ifdef AGGRESSIVE_TEST
2759 printk(", aggressive tests");
2761 #ifdef CHECK_BINSEARCH
2762 printk(", check binsearch");
2764 #ifdef EXTENTS_STATS
2769 #ifdef EXTENTS_STATS
2770 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2771 EXT4_SB(sb)->s_ext_min = 1 << 30;
2772 EXT4_SB(sb)->s_ext_max = 0;
2778 * called at umount time
2780 void ext4_ext_release(struct super_block *sb)
2782 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2785 #ifdef EXTENTS_STATS
2786 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2787 struct ext4_sb_info *sbi = EXT4_SB(sb);
2788 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2789 sbi->s_ext_blocks, sbi->s_ext_extents,
2790 sbi->s_ext_blocks / sbi->s_ext_extents);
2791 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2792 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2797 /* FIXME!! we need to try to merge to left or right after zero-out */
2798 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2800 ext4_fsblk_t ee_pblock;
2801 unsigned int ee_len;
2804 ee_len = ext4_ext_get_actual_len(ex);
2805 ee_pblock = ext4_ext_pblock(ex);
2807 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2815 * ext4_split_extent_at() splits an extent at given block.
2817 * @handle: the journal handle
2818 * @inode: the file inode
2819 * @path: the path to the extent
2820 * @split: the logical block where the extent is splitted.
2821 * @split_flags: indicates if the extent could be zeroout if split fails, and
2822 * the states(init or uninit) of new extents.
2823 * @flags: flags used to insert new extent to extent tree.
2826 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2827 * of which are deterimined by split_flag.
2829 * There are two cases:
2830 * a> the extent are splitted into two extent.
2831 * b> split is not needed, and just mark the extent.
2833 * return 0 on success.
2835 static int ext4_split_extent_at(handle_t *handle,
2836 struct inode *inode,
2837 struct ext4_ext_path *path,
2842 ext4_fsblk_t newblock;
2843 ext4_lblk_t ee_block;
2844 struct ext4_extent *ex, newex, orig_ex;
2845 struct ext4_extent *ex2 = NULL;
2846 unsigned int ee_len, depth;
2849 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2850 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2852 ext_debug("ext4_split_extents_at: inode %lu, logical"
2853 "block %llu\n", inode->i_ino, (unsigned long long)split);
2855 ext4_ext_show_leaf(inode, path);
2857 depth = ext_depth(inode);
2858 ex = path[depth].p_ext;
2859 ee_block = le32_to_cpu(ex->ee_block);
2860 ee_len = ext4_ext_get_actual_len(ex);
2861 newblock = split - ee_block + ext4_ext_pblock(ex);
2863 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2865 err = ext4_ext_get_access(handle, inode, path + depth);
2869 if (split == ee_block) {
2871 * case b: block @split is the block that the extent begins with
2872 * then we just change the state of the extent, and splitting
2875 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2876 ext4_ext_mark_uninitialized(ex);
2878 ext4_ext_mark_initialized(ex);
2880 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2881 ext4_ext_try_to_merge(inode, path, ex);
2883 err = ext4_ext_dirty(handle, inode, path + depth);
2888 memcpy(&orig_ex, ex, sizeof(orig_ex));
2889 ex->ee_len = cpu_to_le16(split - ee_block);
2890 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2891 ext4_ext_mark_uninitialized(ex);
2894 * path may lead to new leaf, not to original leaf any more
2895 * after ext4_ext_insert_extent() returns,
2897 err = ext4_ext_dirty(handle, inode, path + depth);
2899 goto fix_extent_len;
2902 ex2->ee_block = cpu_to_le32(split);
2903 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2904 ext4_ext_store_pblock(ex2, newblock);
2905 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2906 ext4_ext_mark_uninitialized(ex2);
2908 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2909 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2910 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
2911 if (split_flag & EXT4_EXT_DATA_VALID1)
2912 err = ext4_ext_zeroout(inode, ex2);
2914 err = ext4_ext_zeroout(inode, ex);
2916 err = ext4_ext_zeroout(inode, &orig_ex);
2919 goto fix_extent_len;
2920 /* update the extent length and mark as initialized */
2921 ex->ee_len = cpu_to_le16(ee_len);
2922 ext4_ext_try_to_merge(inode, path, ex);
2923 err = ext4_ext_dirty(handle, inode, path + depth);
2926 goto fix_extent_len;
2929 ext4_ext_show_leaf(inode, path);
2933 ex->ee_len = orig_ex.ee_len;
2934 ext4_ext_dirty(handle, inode, path + depth);
2939 * ext4_split_extents() splits an extent and mark extent which is covered
2940 * by @map as split_flags indicates
2942 * It may result in splitting the extent into multiple extents (upto three)
2943 * There are three possibilities:
2944 * a> There is no split required
2945 * b> Splits in two extents: Split is happening at either end of the extent
2946 * c> Splits in three extents: Somone is splitting in middle of the extent
2949 static int ext4_split_extent(handle_t *handle,
2950 struct inode *inode,
2951 struct ext4_ext_path *path,
2952 struct ext4_map_blocks *map,
2956 ext4_lblk_t ee_block;
2957 struct ext4_extent *ex;
2958 unsigned int ee_len, depth;
2961 int split_flag1, flags1;
2963 depth = ext_depth(inode);
2964 ex = path[depth].p_ext;
2965 ee_block = le32_to_cpu(ex->ee_block);
2966 ee_len = ext4_ext_get_actual_len(ex);
2967 uninitialized = ext4_ext_is_uninitialized(ex);
2969 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2970 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
2971 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2973 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2974 EXT4_EXT_MARK_UNINIT2;
2975 if (split_flag & EXT4_EXT_DATA_VALID2)
2976 split_flag1 |= EXT4_EXT_DATA_VALID1;
2977 err = ext4_split_extent_at(handle, inode, path,
2978 map->m_lblk + map->m_len, split_flag1, flags1);
2983 ext4_ext_drop_refs(path);
2984 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2986 return PTR_ERR(path);
2988 if (map->m_lblk >= ee_block) {
2989 split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT |
2990 EXT4_EXT_DATA_VALID2);
2992 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2993 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2994 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2995 err = ext4_split_extent_at(handle, inode, path,
2996 map->m_lblk, split_flag1, flags);
3001 ext4_ext_show_leaf(inode, path);
3003 return err ? err : map->m_len;
3006 #define EXT4_EXT_ZERO_LEN 7
3008 * This function is called by ext4_ext_map_blocks() if someone tries to write
3009 * to an uninitialized extent. It may result in splitting the uninitialized
3010 * extent into multiple extents (up to three - one initialized and two
3012 * There are three possibilities:
3013 * a> There is no split required: Entire extent should be initialized
3014 * b> Splits in two extents: Write is happening at either end of the extent
3015 * c> Splits in three extents: Somone is writing in middle of the extent
3018 * - The extent pointed to by 'path' is uninitialized.
3019 * - The extent pointed to by 'path' contains a superset
3020 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3022 * Post-conditions on success:
3023 * - the returned value is the number of blocks beyond map->l_lblk
3024 * that are allocated and initialized.
3025 * It is guaranteed to be >= map->m_len.
3027 static int ext4_ext_convert_to_initialized(handle_t *handle,
3028 struct inode *inode,
3029 struct ext4_map_blocks *map,
3030 struct ext4_ext_path *path)
3032 struct ext4_extent_header *eh;
3033 struct ext4_map_blocks split_map;
3034 struct ext4_extent zero_ex;
3035 struct ext4_extent *ex;
3036 ext4_lblk_t ee_block, eof_block;
3037 unsigned int ee_len, depth;
3042 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3043 "block %llu, max_blocks %u\n", inode->i_ino,
3044 (unsigned long long)map->m_lblk, map->m_len);
3046 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3047 inode->i_sb->s_blocksize_bits;
3048 if (eof_block < map->m_lblk + map->m_len)
3049 eof_block = map->m_lblk + map->m_len;
3051 depth = ext_depth(inode);
3052 eh = path[depth].p_hdr;
3053 ex = path[depth].p_ext;
3054 ee_block = le32_to_cpu(ex->ee_block);
3055 ee_len = ext4_ext_get_actual_len(ex);
3056 allocated = ee_len - (map->m_lblk - ee_block);
3058 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3060 /* Pre-conditions */
3061 BUG_ON(!ext4_ext_is_uninitialized(ex));
3062 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3065 * Attempt to transfer newly initialized blocks from the currently
3066 * uninitialized extent to its left neighbor. This is much cheaper
3067 * than an insertion followed by a merge as those involve costly
3068 * memmove() calls. This is the common case in steady state for
3069 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3072 * Limitations of the current logic:
3073 * - L1: we only deal with writes at the start of the extent.
3074 * The approach could be extended to writes at the end
3075 * of the extent but this scenario was deemed less common.
3076 * - L2: we do not deal with writes covering the whole extent.
3077 * This would require removing the extent if the transfer
3079 * - L3: we only attempt to merge with an extent stored in the
3080 * same extent tree node.
3082 if ((map->m_lblk == ee_block) && /*L1*/
3083 (map->m_len < ee_len) && /*L2*/
3084 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
3085 struct ext4_extent *prev_ex;
3086 ext4_lblk_t prev_lblk;
3087 ext4_fsblk_t prev_pblk, ee_pblk;
3088 unsigned int prev_len, write_len;
3091 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3092 prev_len = ext4_ext_get_actual_len(prev_ex);
3093 prev_pblk = ext4_ext_pblock(prev_ex);
3094 ee_pblk = ext4_ext_pblock(ex);
3095 write_len = map->m_len;
3098 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3099 * upon those conditions:
3100 * - C1: prev_ex is initialized,
3101 * - C2: prev_ex is logically abutting ex,
3102 * - C3: prev_ex is physically abutting ex,
3103 * - C4: prev_ex can receive the additional blocks without
3104 * overflowing the (initialized) length limit.
3106 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3107 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3108 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3109 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3110 err = ext4_ext_get_access(handle, inode, path + depth);
3114 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3117 /* Shift the start of ex by 'write_len' blocks */
3118 ex->ee_block = cpu_to_le32(ee_block + write_len);
3119 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3120 ex->ee_len = cpu_to_le16(ee_len - write_len);
3121 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3123 /* Extend prev_ex by 'write_len' blocks */
3124 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3126 /* Mark the block containing both extents as dirty */
3127 ext4_ext_dirty(handle, inode, path + depth);
3129 /* Update path to point to the right extent */
3130 path[depth].p_ext = prev_ex;
3132 /* Result: number of initialized blocks past m_lblk */
3133 allocated = write_len;
3138 WARN_ON(map->m_lblk < ee_block);
3140 * It is safe to convert extent to initialized via explicit
3141 * zeroout only if extent is fully insde i_size or new_size.
3143 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3145 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3146 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3147 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3148 err = ext4_ext_zeroout(inode, ex);
3152 err = ext4_ext_get_access(handle, inode, path + depth);
3155 ext4_ext_mark_initialized(ex);
3156 ext4_ext_try_to_merge(inode, path, ex);
3157 err = ext4_ext_dirty(handle, inode, path + depth);
3163 * 1. split the extent into three extents.
3164 * 2. split the extent into two extents, zeroout the first half.
3165 * 3. split the extent into two extents, zeroout the second half.
3166 * 4. split the extent into two extents with out zeroout.
3168 split_map.m_lblk = map->m_lblk;
3169 split_map.m_len = map->m_len;
3171 if (allocated > map->m_len) {
3172 if (allocated <= EXT4_EXT_ZERO_LEN &&
3173 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3176 cpu_to_le32(map->m_lblk);
3177 zero_ex.ee_len = cpu_to_le16(allocated);
3178 ext4_ext_store_pblock(&zero_ex,
3179 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3180 err = ext4_ext_zeroout(inode, &zero_ex);
3183 split_map.m_lblk = map->m_lblk;
3184 split_map.m_len = allocated;
3185 } else if ((map->m_lblk - ee_block + map->m_len <
3186 EXT4_EXT_ZERO_LEN) &&
3187 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3189 if (map->m_lblk != ee_block) {
3190 zero_ex.ee_block = ex->ee_block;
3191 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3193 ext4_ext_store_pblock(&zero_ex,
3194 ext4_ext_pblock(ex));
3195 err = ext4_ext_zeroout(inode, &zero_ex);
3200 split_map.m_lblk = ee_block;
3201 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3202 allocated = map->m_len;
3206 allocated = ext4_split_extent(handle, inode, path,
3207 &split_map, split_flag, 0);
3212 return err ? err : allocated;
3216 * This function is called by ext4_ext_map_blocks() from
3217 * ext4_get_blocks_dio_write() when DIO to write
3218 * to an uninitialized extent.
3220 * Writing to an uninitialized extent may result in splitting the uninitialized
3221 * extent into multiple /initialized uninitialized extents (up to three)
3222 * There are three possibilities:
3223 * a> There is no split required: Entire extent should be uninitialized
3224 * b> Splits in two extents: Write is happening at either end of the extent
3225 * c> Splits in three extents: Somone is writing in middle of the extent
3227 * One of more index blocks maybe needed if the extent tree grow after
3228 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3229 * complete, we need to split the uninitialized extent before DIO submit
3230 * the IO. The uninitialized extent called at this time will be split
3231 * into three uninitialized extent(at most). After IO complete, the part
3232 * being filled will be convert to initialized by the end_io callback function
3233 * via ext4_convert_unwritten_extents().
3235 * Returns the size of uninitialized extent to be written on success.
3237 static int ext4_split_unwritten_extents(handle_t *handle,
3238 struct inode *inode,
3239 struct ext4_map_blocks *map,
3240 struct ext4_ext_path *path,
3243 ext4_lblk_t eof_block;
3244 ext4_lblk_t ee_block;
3245 struct ext4_extent *ex;
3246 unsigned int ee_len;
3247 int split_flag = 0, depth;
3249 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3250 "block %llu, max_blocks %u\n", inode->i_ino,
3251 (unsigned long long)map->m_lblk, map->m_len);
3253 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3254 inode->i_sb->s_blocksize_bits;
3255 if (eof_block < map->m_lblk + map->m_len)
3256 eof_block = map->m_lblk + map->m_len;
3258 * It is safe to convert extent to initialized via explicit
3259 * zeroout only if extent is fully insde i_size or new_size.
3261 depth = ext_depth(inode);
3262 ex = path[depth].p_ext;
3263 ee_block = le32_to_cpu(ex->ee_block);
3264 ee_len = ext4_ext_get_actual_len(ex);
3266 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3267 split_flag |= EXT4_EXT_MARK_UNINIT2;
3268 if (flags & EXT4_GET_BLOCKS_CONVERT)
3269 split_flag |= EXT4_EXT_DATA_VALID2;
3270 flags |= EXT4_GET_BLOCKS_PRE_IO;
3271 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3274 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3275 struct inode *inode,
3276 struct ext4_map_blocks *map,
3277 struct ext4_ext_path *path)
3279 struct ext4_extent *ex;
3280 ext4_lblk_t ee_block;
3281 unsigned int ee_len;
3285 depth = ext_depth(inode);
3286 ex = path[depth].p_ext;
3287 ee_block = le32_to_cpu(ex->ee_block);
3288 ee_len = ext4_ext_get_actual_len(ex);
3290 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3291 "block %llu, max_blocks %u\n", inode->i_ino,
3292 (unsigned long long)ee_block, ee_len);
3294 /* If extent is larger than requested then split is required */
3295 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3296 err = ext4_split_unwritten_extents(handle, inode, map, path,
3297 EXT4_GET_BLOCKS_CONVERT);
3300 ext4_ext_drop_refs(path);
3301 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3303 err = PTR_ERR(path);
3306 depth = ext_depth(inode);
3307 ex = path[depth].p_ext;
3310 err = ext4_ext_get_access(handle, inode, path + depth);
3313 /* first mark the extent as initialized */
3314 ext4_ext_mark_initialized(ex);
3316 /* note: ext4_ext_correct_indexes() isn't needed here because
3317 * borders are not changed
3319 ext4_ext_try_to_merge(inode, path, ex);
3321 /* Mark modified extent as dirty */
3322 err = ext4_ext_dirty(handle, inode, path + depth);
3324 ext4_ext_show_leaf(inode, path);
3328 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3329 sector_t block, int count)
3332 for (i = 0; i < count; i++)
3333 unmap_underlying_metadata(bdev, block + i);
3337 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3339 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3341 struct ext4_ext_path *path,
3345 struct ext4_extent_header *eh;
3346 struct ext4_extent *last_ex;
3348 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3351 depth = ext_depth(inode);
3352 eh = path[depth].p_hdr;
3354 if (unlikely(!eh->eh_entries)) {
3355 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3356 "EOFBLOCKS_FL set");
3359 last_ex = EXT_LAST_EXTENT(eh);
3361 * We should clear the EOFBLOCKS_FL flag if we are writing the
3362 * last block in the last extent in the file. We test this by
3363 * first checking to see if the caller to
3364 * ext4_ext_get_blocks() was interested in the last block (or
3365 * a block beyond the last block) in the current extent. If
3366 * this turns out to be false, we can bail out from this
3367 * function immediately.
3369 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3370 ext4_ext_get_actual_len(last_ex))
3373 * If the caller does appear to be planning to write at or
3374 * beyond the end of the current extent, we then test to see
3375 * if the current extent is the last extent in the file, by
3376 * checking to make sure it was reached via the rightmost node
3377 * at each level of the tree.
3379 for (i = depth-1; i >= 0; i--)
3380 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3382 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3383 return ext4_mark_inode_dirty(handle, inode);
3387 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3389 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3390 * whether there are any buffers marked for delayed allocation. It returns '1'
3391 * on the first delalloc'ed buffer head found. If no buffer head in the given
3392 * range is marked for delalloc, it returns 0.
3393 * lblk_start should always be <= lblk_end.
3394 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3395 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3396 * block sooner). This is useful when blocks are truncated sequentially from
3397 * lblk_start towards lblk_end.
3399 static int ext4_find_delalloc_range(struct inode *inode,
3400 ext4_lblk_t lblk_start,
3401 ext4_lblk_t lblk_end,
3402 int search_hint_reverse)
3404 struct address_space *mapping = inode->i_mapping;
3405 struct buffer_head *head, *bh = NULL;
3407 ext4_lblk_t i, pg_lblk;
3410 /* reverse search wont work if fs block size is less than page size */
3411 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3412 search_hint_reverse = 0;
3414 if (search_hint_reverse)
3419 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3421 while ((i >= lblk_start) && (i <= lblk_end)) {
3422 page = find_get_page(mapping, index);
3426 if (!page_has_buffers(page))
3429 head = page_buffers(page);
3434 pg_lblk = index << (PAGE_CACHE_SHIFT -
3437 if (unlikely(pg_lblk < lblk_start)) {
3439 * This is possible when fs block size is less
3440 * than page size and our cluster starts/ends in
3441 * middle of the page. So we need to skip the
3442 * initial few blocks till we reach the 'lblk'
3448 /* Check if the buffer is delayed allocated and that it
3449 * is not yet mapped. (when da-buffers are mapped during
3450 * their writeout, their da_mapped bit is set.)
3452 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3453 page_cache_release(page);
3454 trace_ext4_find_delalloc_range(inode,
3455 lblk_start, lblk_end,
3456 search_hint_reverse,
3460 if (search_hint_reverse)
3464 } while ((i >= lblk_start) && (i <= lblk_end) &&
3465 ((bh = bh->b_this_page) != head));
3468 page_cache_release(page);
3470 * Move to next page. 'i' will be the first lblk in the next
3473 if (search_hint_reverse)
3477 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3480 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3481 search_hint_reverse, 0, 0);
3485 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3486 int search_hint_reverse)
3488 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3489 ext4_lblk_t lblk_start, lblk_end;
3490 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3491 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3493 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3494 search_hint_reverse);
3498 * Determines how many complete clusters (out of those specified by the 'map')
3499 * are under delalloc and were reserved quota for.
3500 * This function is called when we are writing out the blocks that were
3501 * originally written with their allocation delayed, but then the space was
3502 * allocated using fallocate() before the delayed allocation could be resolved.
3503 * The cases to look for are:
3504 * ('=' indicated delayed allocated blocks
3505 * '-' indicates non-delayed allocated blocks)
3506 * (a) partial clusters towards beginning and/or end outside of allocated range
3507 * are not delalloc'ed.
3509 * |----c---=|====c====|====c====|===-c----|
3510 * |++++++ allocated ++++++|
3511 * ==> 4 complete clusters in above example
3513 * (b) partial cluster (outside of allocated range) towards either end is
3514 * marked for delayed allocation. In this case, we will exclude that
3517 * |----====c========|========c========|
3518 * |++++++ allocated ++++++|
3519 * ==> 1 complete clusters in above example
3522 * |================c================|
3523 * |++++++ allocated ++++++|
3524 * ==> 0 complete clusters in above example
3526 * The ext4_da_update_reserve_space will be called only if we
3527 * determine here that there were some "entire" clusters that span
3528 * this 'allocated' range.
3529 * In the non-bigalloc case, this function will just end up returning num_blks
3530 * without ever calling ext4_find_delalloc_range.
3533 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3534 unsigned int num_blks)
3536 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3537 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3538 ext4_lblk_t lblk_from, lblk_to, c_offset;
3539 unsigned int allocated_clusters = 0;
3541 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3542 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3544 /* max possible clusters for this allocation */
3545 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3547 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3549 /* Check towards left side */
3550 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3552 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3553 lblk_to = lblk_from + c_offset - 1;
3555 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3556 allocated_clusters--;
3559 /* Now check towards right. */
3560 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3561 if (allocated_clusters && c_offset) {
3562 lblk_from = lblk_start + num_blks;
3563 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3565 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3566 allocated_clusters--;
3569 return allocated_clusters;
3573 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3574 struct ext4_map_blocks *map,
3575 struct ext4_ext_path *path, int flags,
3576 unsigned int allocated, ext4_fsblk_t newblock)
3580 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3582 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3583 "block %llu, max_blocks %u, flags %d, allocated %u",
3584 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3586 ext4_ext_show_leaf(inode, path);
3588 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3591 /* get_block() before submit the IO, split the extent */
3592 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3593 ret = ext4_split_unwritten_extents(handle, inode, map,
3596 * Flag the inode(non aio case) or end_io struct (aio case)
3597 * that this IO needs to conversion to written when IO is
3601 ext4_set_io_unwritten_flag(inode, io);
3603 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3604 if (ext4_should_dioread_nolock(inode))
3605 map->m_flags |= EXT4_MAP_UNINIT;
3608 /* IO end_io complete, convert the filled extent to written */
3609 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3610 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3613 ext4_update_inode_fsync_trans(handle, inode, 1);
3614 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3620 /* buffered IO case */
3622 * repeat fallocate creation request
3623 * we already have an unwritten extent
3625 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3628 /* buffered READ or buffered write_begin() lookup */
3629 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3631 * We have blocks reserved already. We
3632 * return allocated blocks so that delalloc
3633 * won't do block reservation for us. But
3634 * the buffer head will be unmapped so that
3635 * a read from the block returns 0s.
3637 map->m_flags |= EXT4_MAP_UNWRITTEN;
3641 /* buffered write, writepage time, convert*/
3642 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3644 ext4_update_inode_fsync_trans(handle, inode, 1);
3651 map->m_flags |= EXT4_MAP_NEW;
3653 * if we allocated more blocks than requested
3654 * we need to make sure we unmap the extra block
3655 * allocated. The actual needed block will get
3656 * unmapped later when we find the buffer_head marked
3659 if (allocated > map->m_len) {
3660 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3661 newblock + map->m_len,
3662 allocated - map->m_len);
3663 allocated = map->m_len;
3667 * If we have done fallocate with the offset that is already
3668 * delayed allocated, we would have block reservation
3669 * and quota reservation done in the delayed write path.
3670 * But fallocate would have already updated quota and block
3671 * count for this offset. So cancel these reservation
3673 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3674 unsigned int reserved_clusters;
3675 reserved_clusters = get_reserved_cluster_alloc(inode,
3676 map->m_lblk, map->m_len);
3677 if (reserved_clusters)
3678 ext4_da_update_reserve_space(inode,
3684 map->m_flags |= EXT4_MAP_MAPPED;
3685 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3686 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3692 if (allocated > map->m_len)
3693 allocated = map->m_len;
3694 ext4_ext_show_leaf(inode, path);
3695 map->m_pblk = newblock;
3696 map->m_len = allocated;
3699 ext4_ext_drop_refs(path);
3702 return err ? err : allocated;
3706 * get_implied_cluster_alloc - check to see if the requested
3707 * allocation (in the map structure) overlaps with a cluster already
3708 * allocated in an extent.
3709 * @sb The filesystem superblock structure
3710 * @map The requested lblk->pblk mapping
3711 * @ex The extent structure which might contain an implied
3712 * cluster allocation
3714 * This function is called by ext4_ext_map_blocks() after we failed to
3715 * find blocks that were already in the inode's extent tree. Hence,
3716 * we know that the beginning of the requested region cannot overlap
3717 * the extent from the inode's extent tree. There are three cases we
3718 * want to catch. The first is this case:
3720 * |--- cluster # N--|
3721 * |--- extent ---| |---- requested region ---|
3724 * The second case that we need to test for is this one:
3726 * |--------- cluster # N ----------------|
3727 * |--- requested region --| |------- extent ----|
3728 * |=======================|
3730 * The third case is when the requested region lies between two extents
3731 * within the same cluster:
3732 * |------------- cluster # N-------------|
3733 * |----- ex -----| |---- ex_right ----|
3734 * |------ requested region ------|
3735 * |================|
3737 * In each of the above cases, we need to set the map->m_pblk and
3738 * map->m_len so it corresponds to the return the extent labelled as
3739 * "|====|" from cluster #N, since it is already in use for data in
3740 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3741 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3742 * as a new "allocated" block region. Otherwise, we will return 0 and
3743 * ext4_ext_map_blocks() will then allocate one or more new clusters
3744 * by calling ext4_mb_new_blocks().
3746 static int get_implied_cluster_alloc(struct super_block *sb,
3747 struct ext4_map_blocks *map,
3748 struct ext4_extent *ex,
3749 struct ext4_ext_path *path)
3751 struct ext4_sb_info *sbi = EXT4_SB(sb);
3752 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3753 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3754 ext4_lblk_t rr_cluster_start, rr_cluster_end;
3755 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3756 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3757 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3759 /* The extent passed in that we are trying to match */
3760 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3761 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3763 /* The requested region passed into ext4_map_blocks() */
3764 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3765 rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3767 if ((rr_cluster_start == ex_cluster_end) ||
3768 (rr_cluster_start == ex_cluster_start)) {
3769 if (rr_cluster_start == ex_cluster_end)
3770 ee_start += ee_len - 1;
3771 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3773 map->m_len = min(map->m_len,
3774 (unsigned) sbi->s_cluster_ratio - c_offset);
3776 * Check for and handle this case:
3778 * |--------- cluster # N-------------|
3779 * |------- extent ----|
3780 * |--- requested region ---|
3784 if (map->m_lblk < ee_block)
3785 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3788 * Check for the case where there is already another allocated
3789 * block to the right of 'ex' but before the end of the cluster.
3791 * |------------- cluster # N-------------|
3792 * |----- ex -----| |---- ex_right ----|
3793 * |------ requested region ------|
3794 * |================|
3796 if (map->m_lblk > ee_block) {
3797 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3798 map->m_len = min(map->m_len, next - map->m_lblk);
3801 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3805 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3811 * Block allocation/map/preallocation routine for extents based files
3814 * Need to be called with
3815 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3816 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3818 * return > 0, number of of blocks already mapped/allocated
3819 * if create == 0 and these are pre-allocated blocks
3820 * buffer head is unmapped
3821 * otherwise blocks are mapped
3823 * return = 0, if plain look up failed (blocks have not been allocated)
3824 * buffer head is unmapped
3826 * return < 0, error case.
3828 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3829 struct ext4_map_blocks *map, int flags)
3831 struct ext4_ext_path *path = NULL;
3832 struct ext4_extent newex, *ex, *ex2;
3833 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3834 ext4_fsblk_t newblock = 0;
3835 int free_on_err = 0, err = 0, depth, ret;
3836 unsigned int allocated = 0, offset = 0;
3837 unsigned int allocated_clusters = 0;
3838 unsigned int punched_out = 0;
3839 unsigned int result = 0;
3840 struct ext4_allocation_request ar;
3841 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3842 ext4_lblk_t cluster_offset;
3844 ext_debug("blocks %u/%u requested for inode %lu\n",
3845 map->m_lblk, map->m_len, inode->i_ino);
3846 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3848 /* check in cache */
3849 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3850 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3851 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3852 if ((sbi->s_cluster_ratio > 1) &&
3853 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3854 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3856 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3858 * block isn't allocated yet and
3859 * user doesn't want to allocate it
3863 /* we should allocate requested block */
3865 /* block is already allocated */
3866 if (sbi->s_cluster_ratio > 1)
3867 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3868 newblock = map->m_lblk
3869 - le32_to_cpu(newex.ee_block)
3870 + ext4_ext_pblock(&newex);
3871 /* number of remaining blocks in the extent */
3872 allocated = ext4_ext_get_actual_len(&newex) -
3873 (map->m_lblk - le32_to_cpu(newex.ee_block));
3878 /* find extent for this block */
3879 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3881 err = PTR_ERR(path);
3886 depth = ext_depth(inode);
3889 * consistent leaf must not be empty;
3890 * this situation is possible, though, _during_ tree modification;
3891 * this is why assert can't be put in ext4_ext_find_extent()
3893 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3894 EXT4_ERROR_INODE(inode, "bad extent address "
3895 "lblock: %lu, depth: %d pblock %lld",
3896 (unsigned long) map->m_lblk, depth,
3897 path[depth].p_block);
3902 ex = path[depth].p_ext;
3904 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3905 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3906 unsigned short ee_len;
3909 * Uninitialized extents are treated as holes, except that
3910 * we split out initialized portions during a write.
3912 ee_len = ext4_ext_get_actual_len(ex);
3914 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3916 /* if found extent covers block, simply return it */
3917 if (in_range(map->m_lblk, ee_block, ee_len)) {
3918 struct ext4_map_blocks punch_map;
3919 ext4_fsblk_t partial_cluster = 0;
3921 newblock = map->m_lblk - ee_block + ee_start;
3922 /* number of remaining blocks in the extent */
3923 allocated = ee_len - (map->m_lblk - ee_block);
3924 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3925 ee_block, ee_len, newblock);
3927 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3929 * Do not put uninitialized extent
3932 if (!ext4_ext_is_uninitialized(ex)) {
3933 ext4_ext_put_in_cache(inode, ee_block,
3937 ret = ext4_ext_handle_uninitialized_extents(
3938 handle, inode, map, path, flags,
3939 allocated, newblock);
3944 * Punch out the map length, but only to the
3947 punched_out = allocated < map->m_len ?
3948 allocated : map->m_len;
3951 * Sense extents need to be converted to
3952 * uninitialized, they must fit in an
3953 * uninitialized extent
3955 if (punched_out > EXT_UNINIT_MAX_LEN)
3956 punched_out = EXT_UNINIT_MAX_LEN;
3958 punch_map.m_lblk = map->m_lblk;
3959 punch_map.m_pblk = newblock;
3960 punch_map.m_len = punched_out;
3961 punch_map.m_flags = 0;
3963 /* Check to see if the extent needs to be split */
3964 if (punch_map.m_len != ee_len ||
3965 punch_map.m_lblk != ee_block) {
3967 ret = ext4_split_extent(handle, inode,
3968 path, &punch_map, 0,
3969 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3970 EXT4_GET_BLOCKS_PRE_IO);
3977 * find extent for the block at
3978 * the start of the hole
3980 ext4_ext_drop_refs(path);
3983 path = ext4_ext_find_extent(inode,
3986 err = PTR_ERR(path);
3991 depth = ext_depth(inode);
3992 ex = path[depth].p_ext;
3993 ee_len = ext4_ext_get_actual_len(ex);
3994 ee_block = le32_to_cpu(ex->ee_block);
3995 ee_start = ext4_ext_pblock(ex);
3999 ext4_ext_mark_uninitialized(ex);
4001 ext4_ext_invalidate_cache(inode);
4003 err = ext4_ext_rm_leaf(handle, inode, path,
4004 &partial_cluster, map->m_lblk,
4005 map->m_lblk + punched_out);
4007 if (!err && path->p_hdr->eh_entries == 0) {
4009 * Punch hole freed all of this sub tree,
4010 * so we need to correct eh_depth
4012 err = ext4_ext_get_access(handle, inode, path);
4014 ext_inode_hdr(inode)->eh_depth = 0;
4015 ext_inode_hdr(inode)->eh_max =
4016 cpu_to_le16(ext4_ext_space_root(
4019 err = ext4_ext_dirty(
4020 handle, inode, path);
4028 if ((sbi->s_cluster_ratio > 1) &&
4029 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
4030 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4033 * requested block isn't allocated yet;
4034 * we couldn't try to create block if create flag is zero
4036 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4038 * put just found gap into cache to speed up
4039 * subsequent requests
4041 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4046 * Okay, we need to do block allocation.
4048 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4049 newex.ee_block = cpu_to_le32(map->m_lblk);
4050 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4053 * If we are doing bigalloc, check to see if the extent returned
4054 * by ext4_ext_find_extent() implies a cluster we can use.
4056 if (cluster_offset && ex &&
4057 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4058 ar.len = allocated = map->m_len;
4059 newblock = map->m_pblk;
4060 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4061 goto got_allocated_blocks;
4064 /* find neighbour allocated blocks */
4065 ar.lleft = map->m_lblk;
4066 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4069 ar.lright = map->m_lblk;
4071 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4075 /* Check if the extent after searching to the right implies a
4076 * cluster we can use. */
4077 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4078 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4079 ar.len = allocated = map->m_len;
4080 newblock = map->m_pblk;
4081 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4082 goto got_allocated_blocks;
4086 * See if request is beyond maximum number of blocks we can have in
4087 * a single extent. For an initialized extent this limit is
4088 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4089 * EXT_UNINIT_MAX_LEN.
4091 if (map->m_len > EXT_INIT_MAX_LEN &&
4092 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4093 map->m_len = EXT_INIT_MAX_LEN;
4094 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4095 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4096 map->m_len = EXT_UNINIT_MAX_LEN;
4098 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4099 newex.ee_len = cpu_to_le16(map->m_len);
4100 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4102 allocated = ext4_ext_get_actual_len(&newex);
4104 allocated = map->m_len;
4106 /* allocate new block */
4108 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4109 ar.logical = map->m_lblk;
4111 * We calculate the offset from the beginning of the cluster
4112 * for the logical block number, since when we allocate a
4113 * physical cluster, the physical block should start at the
4114 * same offset from the beginning of the cluster. This is
4115 * needed so that future calls to get_implied_cluster_alloc()
4118 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4119 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4121 ar.logical -= offset;
4122 if (S_ISREG(inode->i_mode))
4123 ar.flags = EXT4_MB_HINT_DATA;
4125 /* disable in-core preallocation for non-regular files */
4127 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4128 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4129 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4132 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4133 ar.goal, newblock, allocated);
4135 allocated_clusters = ar.len;
4136 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4137 if (ar.len > allocated)
4140 got_allocated_blocks:
4141 /* try to insert new extent into found leaf and return */
4142 ext4_ext_store_pblock(&newex, newblock + offset);
4143 newex.ee_len = cpu_to_le16(ar.len);
4144 /* Mark uninitialized */
4145 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4146 ext4_ext_mark_uninitialized(&newex);
4148 * io_end structure was created for every IO write to an
4149 * uninitialized extent. To avoid unnecessary conversion,
4150 * here we flag the IO that really needs the conversion.
4151 * For non asycn direct IO case, flag the inode state
4152 * that we need to perform conversion when IO is done.
4154 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4156 ext4_set_io_unwritten_flag(inode, io);
4158 ext4_set_inode_state(inode,
4159 EXT4_STATE_DIO_UNWRITTEN);
4161 if (ext4_should_dioread_nolock(inode))
4162 map->m_flags |= EXT4_MAP_UNINIT;
4166 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4167 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4170 err = ext4_ext_insert_extent(handle, inode, path,
4172 if (err && free_on_err) {
4173 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4174 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4175 /* free data blocks we just allocated */
4176 /* not a good idea to call discard here directly,
4177 * but otherwise we'd need to call it every free() */
4178 ext4_discard_preallocations(inode);
4179 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4180 ext4_ext_get_actual_len(&newex), fb_flags);
4184 /* previous routine could use block we allocated */
4185 newblock = ext4_ext_pblock(&newex);
4186 allocated = ext4_ext_get_actual_len(&newex);
4187 if (allocated > map->m_len)
4188 allocated = map->m_len;
4189 map->m_flags |= EXT4_MAP_NEW;
4192 * Update reserved blocks/metadata blocks after successful
4193 * block allocation which had been deferred till now.
4195 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4196 unsigned int reserved_clusters;
4198 * Check how many clusters we had reserved this allocated range
4200 reserved_clusters = get_reserved_cluster_alloc(inode,
4201 map->m_lblk, allocated);
4202 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4203 if (reserved_clusters) {
4205 * We have clusters reserved for this range.
4206 * But since we are not doing actual allocation
4207 * and are simply using blocks from previously
4208 * allocated cluster, we should release the
4209 * reservation and not claim quota.
4211 ext4_da_update_reserve_space(inode,
4212 reserved_clusters, 0);
4215 BUG_ON(allocated_clusters < reserved_clusters);
4216 /* We will claim quota for all newly allocated blocks.*/
4217 ext4_da_update_reserve_space(inode, allocated_clusters,
4219 if (reserved_clusters < allocated_clusters) {
4220 struct ext4_inode_info *ei = EXT4_I(inode);
4221 int reservation = allocated_clusters -
4224 * It seems we claimed few clusters outside of
4225 * the range of this allocation. We should give
4226 * it back to the reservation pool. This can
4227 * happen in the following case:
4229 * * Suppose s_cluster_ratio is 4 (i.e., each
4230 * cluster has 4 blocks. Thus, the clusters
4231 * are [0-3],[4-7],[8-11]...
4232 * * First comes delayed allocation write for
4233 * logical blocks 10 & 11. Since there were no
4234 * previous delayed allocated blocks in the
4235 * range [8-11], we would reserve 1 cluster
4237 * * Next comes write for logical blocks 3 to 8.
4238 * In this case, we will reserve 2 clusters
4239 * (for [0-3] and [4-7]; and not for [8-11] as
4240 * that range has a delayed allocated blocks.
4241 * Thus total reserved clusters now becomes 3.
4242 * * Now, during the delayed allocation writeout
4243 * time, we will first write blocks [3-8] and
4244 * allocate 3 clusters for writing these
4245 * blocks. Also, we would claim all these
4246 * three clusters above.
4247 * * Now when we come here to writeout the
4248 * blocks [10-11], we would expect to claim
4249 * the reservation of 1 cluster we had made
4250 * (and we would claim it since there are no
4251 * more delayed allocated blocks in the range
4252 * [8-11]. But our reserved cluster count had
4253 * already gone to 0.
4255 * Thus, at the step 4 above when we determine
4256 * that there are still some unwritten delayed
4257 * allocated blocks outside of our current
4258 * block range, we should increment the
4259 * reserved clusters count so that when the
4260 * remaining blocks finally gets written, we
4263 dquot_reserve_block(inode,
4264 EXT4_C2B(sbi, reservation));
4265 spin_lock(&ei->i_block_reservation_lock);
4266 ei->i_reserved_data_blocks += reservation;
4267 spin_unlock(&ei->i_block_reservation_lock);
4273 * Cache the extent and update transaction to commit on fdatasync only
4274 * when it is _not_ an uninitialized extent.
4276 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4277 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4278 ext4_update_inode_fsync_trans(handle, inode, 1);
4280 ext4_update_inode_fsync_trans(handle, inode, 0);
4282 if (allocated > map->m_len)
4283 allocated = map->m_len;
4284 ext4_ext_show_leaf(inode, path);
4285 map->m_flags |= EXT4_MAP_MAPPED;
4286 map->m_pblk = newblock;
4287 map->m_len = allocated;
4290 ext4_ext_drop_refs(path);
4293 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4294 punched_out : allocated;
4296 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4297 newblock, map->m_len, err ? err : result);
4299 return err ? err : result;
4302 void ext4_ext_truncate(struct inode *inode)
4304 struct address_space *mapping = inode->i_mapping;
4305 struct super_block *sb = inode->i_sb;
4306 ext4_lblk_t last_block;
4312 * finish any pending end_io work so we won't run the risk of
4313 * converting any truncated blocks to initialized later
4315 ext4_flush_completed_IO(inode);
4318 * probably first extent we're gonna free will be last in block
4320 err = ext4_writepage_trans_blocks(inode);
4321 handle = ext4_journal_start(inode, err);
4325 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4326 page_len = PAGE_CACHE_SIZE -
4327 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4329 err = ext4_discard_partial_page_buffers(handle,
4330 mapping, inode->i_size, page_len, 0);
4336 if (ext4_orphan_add(handle, inode))
4339 down_write(&EXT4_I(inode)->i_data_sem);
4340 ext4_ext_invalidate_cache(inode);
4342 ext4_discard_preallocations(inode);
4345 * TODO: optimization is possible here.
4346 * Probably we need not scan at all,
4347 * because page truncation is enough.
4350 /* we have to know where to truncate from in crash case */
4351 EXT4_I(inode)->i_disksize = inode->i_size;
4352 ext4_mark_inode_dirty(handle, inode);
4354 last_block = (inode->i_size + sb->s_blocksize - 1)
4355 >> EXT4_BLOCK_SIZE_BITS(sb);
4356 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4358 /* In a multi-transaction truncate, we only make the final
4359 * transaction synchronous.
4362 ext4_handle_sync(handle);
4364 up_write(&EXT4_I(inode)->i_data_sem);
4368 * If this was a simple ftruncate() and the file will remain alive,
4369 * then we need to clear up the orphan record which we created above.
4370 * However, if this was a real unlink then we were called by
4371 * ext4_delete_inode(), and we allow that function to clean up the
4372 * orphan info for us.
4375 ext4_orphan_del(handle, inode);
4377 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4378 ext4_mark_inode_dirty(handle, inode);
4379 ext4_journal_stop(handle);
4382 static void ext4_falloc_update_inode(struct inode *inode,
4383 int mode, loff_t new_size, int update_ctime)
4385 struct timespec now;
4388 now = current_fs_time(inode->i_sb);
4389 if (!timespec_equal(&inode->i_ctime, &now))
4390 inode->i_ctime = now;
4393 * Update only when preallocation was requested beyond
4396 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4397 if (new_size > i_size_read(inode))
4398 i_size_write(inode, new_size);
4399 if (new_size > EXT4_I(inode)->i_disksize)
4400 ext4_update_i_disksize(inode, new_size);
4403 * Mark that we allocate beyond EOF so the subsequent truncate
4404 * can proceed even if the new size is the same as i_size.
4406 if (new_size > i_size_read(inode))
4407 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4413 * preallocate space for a file. This implements ext4's fallocate file
4414 * operation, which gets called from sys_fallocate system call.
4415 * For block-mapped files, posix_fallocate should fall back to the method
4416 * of writing zeroes to the required new blocks (the same behavior which is
4417 * expected for file systems which do not support fallocate() system call).
4419 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4421 struct inode *inode = file->f_path.dentry->d_inode;
4424 unsigned int max_blocks;
4429 struct ext4_map_blocks map;
4430 unsigned int credits, blkbits = inode->i_blkbits;
4433 * currently supporting (pre)allocate mode for extent-based
4436 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4439 /* Return error if mode is not supported */
4440 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4443 if (mode & FALLOC_FL_PUNCH_HOLE)
4444 return ext4_punch_hole(file, offset, len);
4446 trace_ext4_fallocate_enter(inode, offset, len, mode);
4447 map.m_lblk = offset >> blkbits;
4449 * We can't just convert len to max_blocks because
4450 * If blocksize = 4096 offset = 3072 and len = 2048
4452 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4455 * credits to insert 1 extent into extent tree
4457 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4458 mutex_lock(&inode->i_mutex);
4459 ret = inode_newsize_ok(inode, (len + offset));
4461 mutex_unlock(&inode->i_mutex);
4462 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4465 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4466 if (mode & FALLOC_FL_KEEP_SIZE)
4467 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4469 * Don't normalize the request if it can fit in one extent so
4470 * that it doesn't get unnecessarily split into multiple
4473 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4474 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4476 while (ret >= 0 && ret < max_blocks) {
4477 map.m_lblk = map.m_lblk + ret;
4478 map.m_len = max_blocks = max_blocks - ret;
4479 handle = ext4_journal_start(inode, credits);
4480 if (IS_ERR(handle)) {
4481 ret = PTR_ERR(handle);
4484 ret = ext4_map_blocks(handle, inode, &map, flags);
4488 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4489 "returned error inode#%lu, block=%u, "
4490 "max_blocks=%u", __func__,
4491 inode->i_ino, map.m_lblk, max_blocks);
4493 ext4_mark_inode_dirty(handle, inode);
4494 ret2 = ext4_journal_stop(handle);
4497 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4498 blkbits) >> blkbits))
4499 new_size = offset + len;
4501 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4503 ext4_falloc_update_inode(inode, mode, new_size,
4504 (map.m_flags & EXT4_MAP_NEW));
4505 ext4_mark_inode_dirty(handle, inode);
4506 ret2 = ext4_journal_stop(handle);
4510 if (ret == -ENOSPC &&
4511 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4515 mutex_unlock(&inode->i_mutex);
4516 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4517 ret > 0 ? ret2 : ret);
4518 return ret > 0 ? ret2 : ret;
4522 * This function convert a range of blocks to written extents
4523 * The caller of this function will pass the start offset and the size.
4524 * all unwritten extents within this range will be converted to
4527 * This function is called from the direct IO end io call back
4528 * function, to convert the fallocated extents after IO is completed.
4529 * Returns 0 on success.
4531 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4535 unsigned int max_blocks;
4538 struct ext4_map_blocks map;
4539 unsigned int credits, blkbits = inode->i_blkbits;
4541 map.m_lblk = offset >> blkbits;
4543 * We can't just convert len to max_blocks because
4544 * If blocksize = 4096 offset = 3072 and len = 2048
4546 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4549 * credits to insert 1 extent into extent tree
4551 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4552 while (ret >= 0 && ret < max_blocks) {
4554 map.m_len = (max_blocks -= ret);
4555 handle = ext4_journal_start(inode, credits);
4556 if (IS_ERR(handle)) {
4557 ret = PTR_ERR(handle);
4560 ret = ext4_map_blocks(handle, inode, &map,
4561 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4564 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4565 "returned error inode#%lu, block=%u, "
4566 "max_blocks=%u", __func__,
4567 inode->i_ino, map.m_lblk, map.m_len);
4569 ext4_mark_inode_dirty(handle, inode);
4570 ret2 = ext4_journal_stop(handle);
4571 if (ret <= 0 || ret2 )
4574 return ret > 0 ? ret2 : ret;
4578 * Callback function called for each extent to gather FIEMAP information.
4580 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4581 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4589 struct fiemap_extent_info *fieinfo = data;
4590 unsigned char blksize_bits;
4592 blksize_bits = inode->i_sb->s_blocksize_bits;
4593 logical = (__u64)newex->ec_block << blksize_bits;
4595 if (newex->ec_start == 0) {
4597 * No extent in extent-tree contains block @newex->ec_start,
4598 * then the block may stay in 1)a hole or 2)delayed-extent.
4600 * Holes or delayed-extents are processed as follows.
4601 * 1. lookup dirty pages with specified range in pagecache.
4602 * If no page is got, then there is no delayed-extent and
4603 * return with EXT_CONTINUE.
4604 * 2. find the 1st mapped buffer,
4605 * 3. check if the mapped buffer is both in the request range
4606 * and a delayed buffer. If not, there is no delayed-extent,
4608 * 4. a delayed-extent is found, the extent will be collected.
4610 ext4_lblk_t end = 0;
4611 pgoff_t last_offset;
4614 pgoff_t start_index = 0;
4615 struct page **pages = NULL;
4616 struct buffer_head *bh = NULL;
4617 struct buffer_head *head = NULL;
4618 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4620 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4624 offset = logical >> PAGE_SHIFT;
4626 last_offset = offset;
4628 ret = find_get_pages_tag(inode->i_mapping, &offset,
4629 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4631 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4632 /* First time, try to find a mapped buffer. */
4635 for (index = 0; index < ret; index++)
4636 page_cache_release(pages[index]);
4639 return EXT_CONTINUE;
4644 /* Try to find the 1st mapped buffer. */
4645 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4647 if (!page_has_buffers(pages[index]))
4649 head = page_buffers(pages[index]);
4656 if (end >= newex->ec_block +
4658 /* The buffer is out of
4659 * the request range.
4663 if (buffer_mapped(bh) &&
4664 end >= newex->ec_block) {
4665 start_index = index - 1;
4666 /* get the 1st mapped buffer. */
4667 goto found_mapped_buffer;
4670 bh = bh->b_this_page;
4672 } while (bh != head);
4674 /* No mapped buffer in the range found in this page,
4675 * We need to look up next page.
4678 /* There is no page left, but we need to limit
4681 newex->ec_len = end - newex->ec_block;
4686 /*Find contiguous delayed buffers. */
4687 if (ret > 0 && pages[0]->index == last_offset)
4688 head = page_buffers(pages[0]);
4694 found_mapped_buffer:
4695 if (bh != NULL && buffer_delay(bh)) {
4696 /* 1st or contiguous delayed buffer found. */
4697 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4699 * 1st delayed buffer found, record
4700 * the start of extent.
4702 flags |= FIEMAP_EXTENT_DELALLOC;
4703 newex->ec_block = end;
4704 logical = (__u64)end << blksize_bits;
4706 /* Find contiguous delayed buffers. */
4708 if (!buffer_delay(bh))
4709 goto found_delayed_extent;
4710 bh = bh->b_this_page;
4712 } while (bh != head);
4714 for (; index < ret; index++) {
4715 if (!page_has_buffers(pages[index])) {
4719 head = page_buffers(pages[index]);
4725 if (pages[index]->index !=
4726 pages[start_index]->index + index
4728 /* Blocks are not contiguous. */
4734 if (!buffer_delay(bh))
4735 /* Delayed-extent ends. */
4736 goto found_delayed_extent;
4737 bh = bh->b_this_page;
4739 } while (bh != head);
4741 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4745 found_delayed_extent:
4746 newex->ec_len = min(end - newex->ec_block,
4747 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4748 if (ret == nr_pages && bh != NULL &&
4749 newex->ec_len < EXT_INIT_MAX_LEN &&
4751 /* Have not collected an extent and continue. */
4752 for (index = 0; index < ret; index++)
4753 page_cache_release(pages[index]);
4757 for (index = 0; index < ret; index++)
4758 page_cache_release(pages[index]);
4762 physical = (__u64)newex->ec_start << blksize_bits;
4763 length = (__u64)newex->ec_len << blksize_bits;
4765 if (ex && ext4_ext_is_uninitialized(ex))
4766 flags |= FIEMAP_EXTENT_UNWRITTEN;
4768 if (next == EXT_MAX_BLOCKS)
4769 flags |= FIEMAP_EXTENT_LAST;
4771 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4777 return EXT_CONTINUE;
4779 /* fiemap flags we can handle specified here */
4780 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4782 static int ext4_xattr_fiemap(struct inode *inode,
4783 struct fiemap_extent_info *fieinfo)
4787 __u32 flags = FIEMAP_EXTENT_LAST;
4788 int blockbits = inode->i_sb->s_blocksize_bits;
4792 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4793 struct ext4_iloc iloc;
4794 int offset; /* offset of xattr in inode */
4796 error = ext4_get_inode_loc(inode, &iloc);
4799 physical = iloc.bh->b_blocknr << blockbits;
4800 offset = EXT4_GOOD_OLD_INODE_SIZE +
4801 EXT4_I(inode)->i_extra_isize;
4803 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4804 flags |= FIEMAP_EXTENT_DATA_INLINE;
4806 } else { /* external block */
4807 physical = EXT4_I(inode)->i_file_acl << blockbits;
4808 length = inode->i_sb->s_blocksize;
4812 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4814 return (error < 0 ? error : 0);
4818 * ext4_ext_punch_hole
4820 * Punches a hole of "length" bytes in a file starting
4823 * @inode: The inode of the file to punch a hole in
4824 * @offset: The starting byte offset of the hole
4825 * @length: The length of the hole
4827 * Returns the number of blocks removed or negative on err
4829 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4831 struct inode *inode = file->f_path.dentry->d_inode;
4832 struct super_block *sb = inode->i_sb;
4833 ext4_lblk_t first_block, stop_block;
4834 struct address_space *mapping = inode->i_mapping;
4836 loff_t first_page, last_page, page_len;
4837 loff_t first_page_offset, last_page_offset;
4838 int credits, err = 0;
4840 /* No need to punch hole beyond i_size */
4841 if (offset >= inode->i_size)
4845 * If the hole extends beyond i_size, set the hole
4846 * to end after the page that contains i_size
4848 if (offset + length > inode->i_size) {
4849 length = inode->i_size +
4850 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4854 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4855 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4857 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4858 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4861 * Write out all dirty pages to avoid race conditions
4862 * Then release them.
4864 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4865 err = filemap_write_and_wait_range(mapping,
4866 offset, offset + length - 1);
4872 /* Now release the pages */
4873 if (last_page_offset > first_page_offset) {
4874 truncate_inode_pages_range(mapping, first_page_offset,
4875 last_page_offset-1);
4878 /* finish any pending end_io work */
4879 ext4_flush_completed_IO(inode);
4881 credits = ext4_writepage_trans_blocks(inode);
4882 handle = ext4_journal_start(inode, credits);
4884 return PTR_ERR(handle);
4886 err = ext4_orphan_add(handle, inode);
4891 * Now we need to zero out the non-page-aligned data in the
4892 * pages at the start and tail of the hole, and unmap the buffer
4893 * heads for the block aligned regions of the page that were
4894 * completely zeroed.
4896 if (first_page > last_page) {
4898 * If the file space being truncated is contained within a page
4899 * just zero out and unmap the middle of that page
4901 err = ext4_discard_partial_page_buffers(handle,
4902 mapping, offset, length, 0);
4908 * zero out and unmap the partial page that contains
4909 * the start of the hole
4911 page_len = first_page_offset - offset;
4913 err = ext4_discard_partial_page_buffers(handle, mapping,
4914 offset, page_len, 0);
4920 * zero out and unmap the partial page that contains
4921 * the end of the hole
4923 page_len = offset + length - last_page_offset;
4925 err = ext4_discard_partial_page_buffers(handle, mapping,
4926 last_page_offset, page_len, 0);
4933 * If i_size is contained in the last page, we need to
4934 * unmap and zero the partial page after i_size
4936 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4937 inode->i_size % PAGE_CACHE_SIZE != 0) {
4939 page_len = PAGE_CACHE_SIZE -
4940 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4943 err = ext4_discard_partial_page_buffers(handle,
4944 mapping, inode->i_size, page_len, 0);
4951 first_block = (offset + sb->s_blocksize - 1) >>
4952 EXT4_BLOCK_SIZE_BITS(sb);
4953 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4955 /* If there are no blocks to remove, return now */
4956 if (first_block >= stop_block)
4959 down_write(&EXT4_I(inode)->i_data_sem);
4960 ext4_ext_invalidate_cache(inode);
4961 ext4_discard_preallocations(inode);
4963 err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
4965 ext4_ext_invalidate_cache(inode);
4966 ext4_discard_preallocations(inode);
4969 ext4_handle_sync(handle);
4971 up_write(&EXT4_I(inode)->i_data_sem);
4974 ext4_orphan_del(handle, inode);
4975 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4976 ext4_mark_inode_dirty(handle, inode);
4977 ext4_journal_stop(handle);
4980 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4981 __u64 start, __u64 len)
4983 ext4_lblk_t start_blk;
4986 /* fallback to generic here if not in extents fmt */
4987 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4988 return generic_block_fiemap(inode, fieinfo, start, len,
4991 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4994 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4995 error = ext4_xattr_fiemap(inode, fieinfo);
4997 ext4_lblk_t len_blks;
5000 start_blk = start >> inode->i_sb->s_blocksize_bits;
5001 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5002 if (last_blk >= EXT_MAX_BLOCKS)
5003 last_blk = EXT_MAX_BLOCKS-1;
5004 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5007 * Walk the extent tree gathering extent information.
5008 * ext4_ext_fiemap_cb will push extents back to user.
5010 error = ext4_ext_walk_space(inode, start_blk, len_blks,
5011 ext4_ext_fiemap_cb, fieinfo);