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);
321 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
322 ext4_lblk_t last = lblock + len - 1;
324 if (len == 0 || lblock > last)
326 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
329 static int ext4_valid_extent_idx(struct inode *inode,
330 struct ext4_extent_idx *ext_idx)
332 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
334 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
337 static int ext4_valid_extent_entries(struct inode *inode,
338 struct ext4_extent_header *eh,
341 unsigned short entries;
342 if (eh->eh_entries == 0)
345 entries = le16_to_cpu(eh->eh_entries);
349 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
350 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
351 ext4_fsblk_t pblock = 0;
352 ext4_lblk_t lblock = 0;
353 ext4_lblk_t prev = 0;
356 if (!ext4_valid_extent(inode, ext))
359 /* Check for overlapping extents */
360 lblock = le32_to_cpu(ext->ee_block);
361 len = ext4_ext_get_actual_len(ext);
362 if ((lblock <= prev) && prev) {
363 pblock = ext4_ext_pblock(ext);
364 es->s_last_error_block = cpu_to_le64(pblock);
369 prev = lblock + len - 1;
372 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
374 if (!ext4_valid_extent_idx(inode, ext_idx))
383 static int __ext4_ext_check(const char *function, unsigned int line,
384 struct inode *inode, struct ext4_extent_header *eh,
387 const char *error_msg;
390 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
391 error_msg = "invalid magic";
394 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
395 error_msg = "unexpected eh_depth";
398 if (unlikely(eh->eh_max == 0)) {
399 error_msg = "invalid eh_max";
402 max = ext4_ext_max_entries(inode, depth);
403 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
404 error_msg = "too large eh_max";
407 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
408 error_msg = "invalid eh_entries";
411 if (!ext4_valid_extent_entries(inode, eh, depth)) {
412 error_msg = "invalid extent entries";
418 ext4_error_inode(inode, function, line, 0,
419 "bad header/extent: %s - magic %x, "
420 "entries %u, max %u(%u), depth %u(%u)",
421 error_msg, le16_to_cpu(eh->eh_magic),
422 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
423 max, le16_to_cpu(eh->eh_depth), depth);
428 #define ext4_ext_check(inode, eh, depth) \
429 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
431 int ext4_ext_check_inode(struct inode *inode)
433 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
437 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
439 int k, l = path->p_depth;
442 for (k = 0; k <= l; k++, path++) {
444 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
445 ext4_idx_pblock(path->p_idx));
446 } else if (path->p_ext) {
447 ext_debug(" %d:[%d]%d:%llu ",
448 le32_to_cpu(path->p_ext->ee_block),
449 ext4_ext_is_uninitialized(path->p_ext),
450 ext4_ext_get_actual_len(path->p_ext),
451 ext4_ext_pblock(path->p_ext));
458 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
460 int depth = ext_depth(inode);
461 struct ext4_extent_header *eh;
462 struct ext4_extent *ex;
468 eh = path[depth].p_hdr;
469 ex = EXT_FIRST_EXTENT(eh);
471 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
473 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
474 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
475 ext4_ext_is_uninitialized(ex),
476 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
481 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
482 ext4_fsblk_t newblock, int level)
484 int depth = ext_depth(inode);
485 struct ext4_extent *ex;
487 if (depth != level) {
488 struct ext4_extent_idx *idx;
489 idx = path[level].p_idx;
490 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
491 ext_debug("%d: move %d:%llu in new index %llu\n", level,
492 le32_to_cpu(idx->ei_block),
493 ext4_idx_pblock(idx),
501 ex = path[depth].p_ext;
502 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
503 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
504 le32_to_cpu(ex->ee_block),
506 ext4_ext_is_uninitialized(ex),
507 ext4_ext_get_actual_len(ex),
514 #define ext4_ext_show_path(inode, path)
515 #define ext4_ext_show_leaf(inode, path)
516 #define ext4_ext_show_move(inode, path, newblock, level)
519 void ext4_ext_drop_refs(struct ext4_ext_path *path)
521 int depth = path->p_depth;
524 for (i = 0; i <= depth; i++, path++)
532 * ext4_ext_binsearch_idx:
533 * binary search for the closest index of the given block
534 * the header must be checked before calling this
537 ext4_ext_binsearch_idx(struct inode *inode,
538 struct ext4_ext_path *path, ext4_lblk_t block)
540 struct ext4_extent_header *eh = path->p_hdr;
541 struct ext4_extent_idx *r, *l, *m;
544 ext_debug("binsearch for %u(idx): ", block);
546 l = EXT_FIRST_INDEX(eh) + 1;
547 r = EXT_LAST_INDEX(eh);
550 if (block < le32_to_cpu(m->ei_block))
554 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
555 m, le32_to_cpu(m->ei_block),
556 r, le32_to_cpu(r->ei_block));
560 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
561 ext4_idx_pblock(path->p_idx));
563 #ifdef CHECK_BINSEARCH
565 struct ext4_extent_idx *chix, *ix;
568 chix = ix = EXT_FIRST_INDEX(eh);
569 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
571 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
572 printk(KERN_DEBUG "k=%d, ix=0x%p, "
574 ix, EXT_FIRST_INDEX(eh));
575 printk(KERN_DEBUG "%u <= %u\n",
576 le32_to_cpu(ix->ei_block),
577 le32_to_cpu(ix[-1].ei_block));
579 BUG_ON(k && le32_to_cpu(ix->ei_block)
580 <= le32_to_cpu(ix[-1].ei_block));
581 if (block < le32_to_cpu(ix->ei_block))
585 BUG_ON(chix != path->p_idx);
592 * ext4_ext_binsearch:
593 * binary search for closest extent of the given block
594 * the header must be checked before calling this
597 ext4_ext_binsearch(struct inode *inode,
598 struct ext4_ext_path *path, ext4_lblk_t block)
600 struct ext4_extent_header *eh = path->p_hdr;
601 struct ext4_extent *r, *l, *m;
603 if (eh->eh_entries == 0) {
605 * this leaf is empty:
606 * we get such a leaf in split/add case
611 ext_debug("binsearch for %u: ", block);
613 l = EXT_FIRST_EXTENT(eh) + 1;
614 r = EXT_LAST_EXTENT(eh);
618 if (block < le32_to_cpu(m->ee_block))
622 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
623 m, le32_to_cpu(m->ee_block),
624 r, le32_to_cpu(r->ee_block));
628 ext_debug(" -> %d:%llu:[%d]%d ",
629 le32_to_cpu(path->p_ext->ee_block),
630 ext4_ext_pblock(path->p_ext),
631 ext4_ext_is_uninitialized(path->p_ext),
632 ext4_ext_get_actual_len(path->p_ext));
634 #ifdef CHECK_BINSEARCH
636 struct ext4_extent *chex, *ex;
639 chex = ex = EXT_FIRST_EXTENT(eh);
640 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
641 BUG_ON(k && le32_to_cpu(ex->ee_block)
642 <= le32_to_cpu(ex[-1].ee_block));
643 if (block < le32_to_cpu(ex->ee_block))
647 BUG_ON(chex != path->p_ext);
653 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
655 struct ext4_extent_header *eh;
657 eh = ext_inode_hdr(inode);
660 eh->eh_magic = EXT4_EXT_MAGIC;
661 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
662 ext4_mark_inode_dirty(handle, inode);
663 ext4_ext_invalidate_cache(inode);
667 struct ext4_ext_path *
668 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
669 struct ext4_ext_path *path)
671 struct ext4_extent_header *eh;
672 struct buffer_head *bh;
673 short int depth, i, ppos = 0, alloc = 0;
676 eh = ext_inode_hdr(inode);
677 depth = ext_depth(inode);
679 /* account possible depth increase */
681 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
684 return ERR_PTR(-ENOMEM);
691 /* walk through the tree */
693 int need_to_validate = 0;
695 ext_debug("depth %d: num %d, max %d\n",
696 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
698 ext4_ext_binsearch_idx(inode, path + ppos, block);
699 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
700 path[ppos].p_depth = i;
701 path[ppos].p_ext = NULL;
703 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
708 if (!bh_uptodate_or_lock(bh)) {
709 trace_ext4_ext_load_extent(inode, block,
711 ret = bh_submit_read(bh);
716 /* validate the extent entries */
717 need_to_validate = 1;
719 eh = ext_block_hdr(bh);
721 if (unlikely(ppos > depth)) {
723 EXT4_ERROR_INODE(inode,
724 "ppos %d > depth %d", ppos, depth);
728 path[ppos].p_bh = bh;
729 path[ppos].p_hdr = eh;
732 ret = need_to_validate ? ext4_ext_check(inode, eh, i) : 0;
737 path[ppos].p_depth = i;
738 path[ppos].p_ext = NULL;
739 path[ppos].p_idx = NULL;
742 ext4_ext_binsearch(inode, path + ppos, block);
743 /* if not an empty leaf */
744 if (path[ppos].p_ext)
745 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
747 ext4_ext_show_path(inode, path);
752 ext4_ext_drop_refs(path);
759 * ext4_ext_insert_index:
760 * insert new index [@logical;@ptr] into the block at @curp;
761 * check where to insert: before @curp or after @curp
763 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
764 struct ext4_ext_path *curp,
765 int logical, ext4_fsblk_t ptr)
767 struct ext4_extent_idx *ix;
770 err = ext4_ext_get_access(handle, inode, curp);
774 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
775 EXT4_ERROR_INODE(inode,
776 "logical %d == ei_block %d!",
777 logical, le32_to_cpu(curp->p_idx->ei_block));
781 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
782 >= le16_to_cpu(curp->p_hdr->eh_max))) {
783 EXT4_ERROR_INODE(inode,
784 "eh_entries %d >= eh_max %d!",
785 le16_to_cpu(curp->p_hdr->eh_entries),
786 le16_to_cpu(curp->p_hdr->eh_max));
790 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
792 ext_debug("insert new index %d after: %llu\n", logical, ptr);
793 ix = curp->p_idx + 1;
796 ext_debug("insert new index %d before: %llu\n", logical, ptr);
800 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
803 ext_debug("insert new index %d: "
804 "move %d indices from 0x%p to 0x%p\n",
805 logical, len, ix, ix + 1);
806 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
809 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
810 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
814 ix->ei_block = cpu_to_le32(logical);
815 ext4_idx_store_pblock(ix, ptr);
816 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
818 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
819 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
823 err = ext4_ext_dirty(handle, inode, curp);
824 ext4_std_error(inode->i_sb, err);
831 * inserts new subtree into the path, using free index entry
833 * - allocates all needed blocks (new leaf and all intermediate index blocks)
834 * - makes decision where to split
835 * - moves remaining extents and index entries (right to the split point)
836 * into the newly allocated blocks
837 * - initializes subtree
839 static int ext4_ext_split(handle_t *handle, struct inode *inode,
841 struct ext4_ext_path *path,
842 struct ext4_extent *newext, int at)
844 struct buffer_head *bh = NULL;
845 int depth = ext_depth(inode);
846 struct ext4_extent_header *neh;
847 struct ext4_extent_idx *fidx;
849 ext4_fsblk_t newblock, oldblock;
851 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
854 /* make decision: where to split? */
855 /* FIXME: now decision is simplest: at current extent */
857 /* if current leaf will be split, then we should use
858 * border from split point */
859 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
860 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
863 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
864 border = path[depth].p_ext[1].ee_block;
865 ext_debug("leaf will be split."
866 " next leaf starts at %d\n",
867 le32_to_cpu(border));
869 border = newext->ee_block;
870 ext_debug("leaf will be added."
871 " next leaf starts at %d\n",
872 le32_to_cpu(border));
876 * If error occurs, then we break processing
877 * and mark filesystem read-only. index won't
878 * be inserted and tree will be in consistent
879 * state. Next mount will repair buffers too.
883 * Get array to track all allocated blocks.
884 * We need this to handle errors and free blocks
887 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
891 /* allocate all needed blocks */
892 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
893 for (a = 0; a < depth - at; a++) {
894 newblock = ext4_ext_new_meta_block(handle, inode, path,
895 newext, &err, flags);
898 ablocks[a] = newblock;
901 /* initialize new leaf */
902 newblock = ablocks[--a];
903 if (unlikely(newblock == 0)) {
904 EXT4_ERROR_INODE(inode, "newblock == 0!");
908 bh = sb_getblk(inode->i_sb, newblock);
915 err = ext4_journal_get_create_access(handle, bh);
919 neh = ext_block_hdr(bh);
921 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
922 neh->eh_magic = EXT4_EXT_MAGIC;
925 /* move remainder of path[depth] to the new leaf */
926 if (unlikely(path[depth].p_hdr->eh_entries !=
927 path[depth].p_hdr->eh_max)) {
928 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
929 path[depth].p_hdr->eh_entries,
930 path[depth].p_hdr->eh_max);
934 /* start copy from next extent */
935 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
936 ext4_ext_show_move(inode, path, newblock, depth);
938 struct ext4_extent *ex;
939 ex = EXT_FIRST_EXTENT(neh);
940 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
941 le16_add_cpu(&neh->eh_entries, m);
944 set_buffer_uptodate(bh);
947 err = ext4_handle_dirty_metadata(handle, inode, bh);
953 /* correct old leaf */
955 err = ext4_ext_get_access(handle, inode, path + depth);
958 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
959 err = ext4_ext_dirty(handle, inode, path + depth);
965 /* create intermediate indexes */
967 if (unlikely(k < 0)) {
968 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
973 ext_debug("create %d intermediate indices\n", k);
974 /* insert new index into current index block */
975 /* current depth stored in i var */
979 newblock = ablocks[--a];
980 bh = sb_getblk(inode->i_sb, newblock);
987 err = ext4_journal_get_create_access(handle, bh);
991 neh = ext_block_hdr(bh);
992 neh->eh_entries = cpu_to_le16(1);
993 neh->eh_magic = EXT4_EXT_MAGIC;
994 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
995 neh->eh_depth = cpu_to_le16(depth - i);
996 fidx = EXT_FIRST_INDEX(neh);
997 fidx->ei_block = border;
998 ext4_idx_store_pblock(fidx, oldblock);
1000 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1001 i, newblock, le32_to_cpu(border), oldblock);
1003 /* move remainder of path[i] to the new index block */
1004 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1005 EXT_LAST_INDEX(path[i].p_hdr))) {
1006 EXT4_ERROR_INODE(inode,
1007 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1008 le32_to_cpu(path[i].p_ext->ee_block));
1012 /* start copy indexes */
1013 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1014 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1015 EXT_MAX_INDEX(path[i].p_hdr));
1016 ext4_ext_show_move(inode, path, newblock, i);
1018 memmove(++fidx, path[i].p_idx,
1019 sizeof(struct ext4_extent_idx) * m);
1020 le16_add_cpu(&neh->eh_entries, m);
1022 set_buffer_uptodate(bh);
1025 err = ext4_handle_dirty_metadata(handle, inode, bh);
1031 /* correct old index */
1033 err = ext4_ext_get_access(handle, inode, path + i);
1036 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1037 err = ext4_ext_dirty(handle, inode, path + i);
1045 /* insert new index */
1046 err = ext4_ext_insert_index(handle, inode, path + at,
1047 le32_to_cpu(border), newblock);
1051 if (buffer_locked(bh))
1057 /* free all allocated blocks in error case */
1058 for (i = 0; i < depth; i++) {
1061 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1062 EXT4_FREE_BLOCKS_METADATA);
1071 * ext4_ext_grow_indepth:
1072 * implements tree growing procedure:
1073 * - allocates new block
1074 * - moves top-level data (index block or leaf) into the new block
1075 * - initializes new top-level, creating index that points to the
1076 * just created block
1078 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1080 struct ext4_extent *newext)
1082 struct ext4_extent_header *neh;
1083 struct buffer_head *bh;
1084 ext4_fsblk_t newblock;
1087 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1088 newext, &err, flags);
1092 bh = sb_getblk(inode->i_sb, newblock);
1097 err = ext4_journal_get_create_access(handle, bh);
1103 /* move top-level index/leaf into new block */
1104 memmove(bh->b_data, EXT4_I(inode)->i_data,
1105 sizeof(EXT4_I(inode)->i_data));
1107 /* set size of new block */
1108 neh = ext_block_hdr(bh);
1109 /* old root could have indexes or leaves
1110 * so calculate e_max right way */
1111 if (ext_depth(inode))
1112 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1114 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1115 neh->eh_magic = EXT4_EXT_MAGIC;
1116 set_buffer_uptodate(bh);
1119 err = ext4_handle_dirty_metadata(handle, inode, bh);
1123 /* Update top-level index: num,max,pointer */
1124 neh = ext_inode_hdr(inode);
1125 neh->eh_entries = cpu_to_le16(1);
1126 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1127 if (neh->eh_depth == 0) {
1128 /* Root extent block becomes index block */
1129 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1130 EXT_FIRST_INDEX(neh)->ei_block =
1131 EXT_FIRST_EXTENT(neh)->ee_block;
1133 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1134 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1135 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1136 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1138 neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
1139 ext4_mark_inode_dirty(handle, inode);
1147 * ext4_ext_create_new_leaf:
1148 * finds empty index and adds new leaf.
1149 * if no free index is found, then it requests in-depth growing.
1151 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1153 struct ext4_ext_path *path,
1154 struct ext4_extent *newext)
1156 struct ext4_ext_path *curp;
1157 int depth, i, err = 0;
1160 i = depth = ext_depth(inode);
1162 /* walk up to the tree and look for free index entry */
1163 curp = path + depth;
1164 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1169 /* we use already allocated block for index block,
1170 * so subsequent data blocks should be contiguous */
1171 if (EXT_HAS_FREE_INDEX(curp)) {
1172 /* if we found index with free entry, then use that
1173 * entry: create all needed subtree and add new leaf */
1174 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1179 ext4_ext_drop_refs(path);
1180 path = ext4_ext_find_extent(inode,
1181 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1184 err = PTR_ERR(path);
1186 /* tree is full, time to grow in depth */
1187 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1192 ext4_ext_drop_refs(path);
1193 path = ext4_ext_find_extent(inode,
1194 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1197 err = PTR_ERR(path);
1202 * only first (depth 0 -> 1) produces free space;
1203 * in all other cases we have to split the grown tree
1205 depth = ext_depth(inode);
1206 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1207 /* now we need to split */
1217 * search the closest allocated block to the left for *logical
1218 * and returns it at @logical + it's physical address at @phys
1219 * if *logical is the smallest allocated block, the function
1220 * returns 0 at @phys
1221 * return value contains 0 (success) or error code
1223 static int ext4_ext_search_left(struct inode *inode,
1224 struct ext4_ext_path *path,
1225 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1227 struct ext4_extent_idx *ix;
1228 struct ext4_extent *ex;
1231 if (unlikely(path == NULL)) {
1232 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1235 depth = path->p_depth;
1238 if (depth == 0 && path->p_ext == NULL)
1241 /* usually extent in the path covers blocks smaller
1242 * then *logical, but it can be that extent is the
1243 * first one in the file */
1245 ex = path[depth].p_ext;
1246 ee_len = ext4_ext_get_actual_len(ex);
1247 if (*logical < le32_to_cpu(ex->ee_block)) {
1248 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1249 EXT4_ERROR_INODE(inode,
1250 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1251 *logical, le32_to_cpu(ex->ee_block));
1254 while (--depth >= 0) {
1255 ix = path[depth].p_idx;
1256 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1257 EXT4_ERROR_INODE(inode,
1258 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1259 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1260 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1261 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1269 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1270 EXT4_ERROR_INODE(inode,
1271 "logical %d < ee_block %d + ee_len %d!",
1272 *logical, le32_to_cpu(ex->ee_block), ee_len);
1276 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1277 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1282 * search the closest allocated block to the right for *logical
1283 * and returns it at @logical + it's physical address at @phys
1284 * if *logical is the largest allocated block, the function
1285 * returns 0 at @phys
1286 * return value contains 0 (success) or error code
1288 static int ext4_ext_search_right(struct inode *inode,
1289 struct ext4_ext_path *path,
1290 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1291 struct ext4_extent **ret_ex)
1293 struct buffer_head *bh = NULL;
1294 struct ext4_extent_header *eh;
1295 struct ext4_extent_idx *ix;
1296 struct ext4_extent *ex;
1298 int depth; /* Note, NOT eh_depth; depth from top of tree */
1301 if (unlikely(path == NULL)) {
1302 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1305 depth = path->p_depth;
1308 if (depth == 0 && path->p_ext == NULL)
1311 /* usually extent in the path covers blocks smaller
1312 * then *logical, but it can be that extent is the
1313 * first one in the file */
1315 ex = path[depth].p_ext;
1316 ee_len = ext4_ext_get_actual_len(ex);
1317 if (*logical < le32_to_cpu(ex->ee_block)) {
1318 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1319 EXT4_ERROR_INODE(inode,
1320 "first_extent(path[%d].p_hdr) != ex",
1324 while (--depth >= 0) {
1325 ix = path[depth].p_idx;
1326 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1327 EXT4_ERROR_INODE(inode,
1328 "ix != EXT_FIRST_INDEX *logical %d!",
1336 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1337 EXT4_ERROR_INODE(inode,
1338 "logical %d < ee_block %d + ee_len %d!",
1339 *logical, le32_to_cpu(ex->ee_block), ee_len);
1343 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1344 /* next allocated block in this leaf */
1349 /* go up and search for index to the right */
1350 while (--depth >= 0) {
1351 ix = path[depth].p_idx;
1352 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1356 /* we've gone up to the root and found no index to the right */
1360 /* we've found index to the right, let's
1361 * follow it and find the closest allocated
1362 * block to the right */
1364 block = ext4_idx_pblock(ix);
1365 while (++depth < path->p_depth) {
1366 bh = sb_bread(inode->i_sb, block);
1369 eh = ext_block_hdr(bh);
1370 /* subtract from p_depth to get proper eh_depth */
1371 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1375 ix = EXT_FIRST_INDEX(eh);
1376 block = ext4_idx_pblock(ix);
1380 bh = sb_bread(inode->i_sb, block);
1383 eh = ext_block_hdr(bh);
1384 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1388 ex = EXT_FIRST_EXTENT(eh);
1390 *logical = le32_to_cpu(ex->ee_block);
1391 *phys = ext4_ext_pblock(ex);
1399 * ext4_ext_next_allocated_block:
1400 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1401 * NOTE: it considers block number from index entry as
1402 * allocated block. Thus, index entries have to be consistent
1406 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1410 BUG_ON(path == NULL);
1411 depth = path->p_depth;
1413 if (depth == 0 && path->p_ext == NULL)
1414 return EXT_MAX_BLOCKS;
1416 while (depth >= 0) {
1417 if (depth == path->p_depth) {
1419 if (path[depth].p_ext &&
1420 path[depth].p_ext !=
1421 EXT_LAST_EXTENT(path[depth].p_hdr))
1422 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1425 if (path[depth].p_idx !=
1426 EXT_LAST_INDEX(path[depth].p_hdr))
1427 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1432 return EXT_MAX_BLOCKS;
1436 * ext4_ext_next_leaf_block:
1437 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1439 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1443 BUG_ON(path == NULL);
1444 depth = path->p_depth;
1446 /* zero-tree has no leaf blocks at all */
1448 return EXT_MAX_BLOCKS;
1450 /* go to index block */
1453 while (depth >= 0) {
1454 if (path[depth].p_idx !=
1455 EXT_LAST_INDEX(path[depth].p_hdr))
1456 return (ext4_lblk_t)
1457 le32_to_cpu(path[depth].p_idx[1].ei_block);
1461 return EXT_MAX_BLOCKS;
1465 * ext4_ext_correct_indexes:
1466 * if leaf gets modified and modified extent is first in the leaf,
1467 * then we have to correct all indexes above.
1468 * TODO: do we need to correct tree in all cases?
1470 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1471 struct ext4_ext_path *path)
1473 struct ext4_extent_header *eh;
1474 int depth = ext_depth(inode);
1475 struct ext4_extent *ex;
1479 eh = path[depth].p_hdr;
1480 ex = path[depth].p_ext;
1482 if (unlikely(ex == NULL || eh == NULL)) {
1483 EXT4_ERROR_INODE(inode,
1484 "ex %p == NULL or eh %p == NULL", ex, eh);
1489 /* there is no tree at all */
1493 if (ex != EXT_FIRST_EXTENT(eh)) {
1494 /* we correct tree if first leaf got modified only */
1499 * TODO: we need correction if border is smaller than current one
1502 border = path[depth].p_ext->ee_block;
1503 err = ext4_ext_get_access(handle, inode, path + k);
1506 path[k].p_idx->ei_block = border;
1507 err = ext4_ext_dirty(handle, inode, path + k);
1512 /* change all left-side indexes */
1513 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1515 err = ext4_ext_get_access(handle, inode, path + k);
1518 path[k].p_idx->ei_block = border;
1519 err = ext4_ext_dirty(handle, inode, path + k);
1528 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1529 struct ext4_extent *ex2)
1531 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1534 * Make sure that either both extents are uninitialized, or
1537 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1540 if (ext4_ext_is_uninitialized(ex1))
1541 max_len = EXT_UNINIT_MAX_LEN;
1543 max_len = EXT_INIT_MAX_LEN;
1545 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1546 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1548 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1549 le32_to_cpu(ex2->ee_block))
1553 * To allow future support for preallocated extents to be added
1554 * as an RO_COMPAT feature, refuse to merge to extents if
1555 * this can result in the top bit of ee_len being set.
1557 if (ext1_ee_len + ext2_ee_len > max_len)
1559 #ifdef AGGRESSIVE_TEST
1560 if (ext1_ee_len >= 4)
1564 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1570 * This function tries to merge the "ex" extent to the next extent in the tree.
1571 * It always tries to merge towards right. If you want to merge towards
1572 * left, pass "ex - 1" as argument instead of "ex".
1573 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1574 * 1 if they got merged.
1576 static int ext4_ext_try_to_merge_right(struct inode *inode,
1577 struct ext4_ext_path *path,
1578 struct ext4_extent *ex)
1580 struct ext4_extent_header *eh;
1581 unsigned int depth, len;
1583 int uninitialized = 0;
1585 depth = ext_depth(inode);
1586 BUG_ON(path[depth].p_hdr == NULL);
1587 eh = path[depth].p_hdr;
1589 while (ex < EXT_LAST_EXTENT(eh)) {
1590 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1592 /* merge with next extent! */
1593 if (ext4_ext_is_uninitialized(ex))
1595 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1596 + ext4_ext_get_actual_len(ex + 1));
1598 ext4_ext_mark_uninitialized(ex);
1600 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1601 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1602 * sizeof(struct ext4_extent);
1603 memmove(ex + 1, ex + 2, len);
1605 le16_add_cpu(&eh->eh_entries, -1);
1607 WARN_ON(eh->eh_entries == 0);
1608 if (!eh->eh_entries)
1609 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1616 * This function tries to merge the @ex extent to neighbours in the tree.
1617 * return 1 if merge left else 0.
1619 static int ext4_ext_try_to_merge(struct inode *inode,
1620 struct ext4_ext_path *path,
1621 struct ext4_extent *ex) {
1622 struct ext4_extent_header *eh;
1627 depth = ext_depth(inode);
1628 BUG_ON(path[depth].p_hdr == NULL);
1629 eh = path[depth].p_hdr;
1631 if (ex > EXT_FIRST_EXTENT(eh))
1632 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1635 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1641 * check if a portion of the "newext" extent overlaps with an
1644 * If there is an overlap discovered, it updates the length of the newext
1645 * such that there will be no overlap, and then returns 1.
1646 * If there is no overlap found, it returns 0.
1648 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1649 struct inode *inode,
1650 struct ext4_extent *newext,
1651 struct ext4_ext_path *path)
1654 unsigned int depth, len1;
1655 unsigned int ret = 0;
1657 b1 = le32_to_cpu(newext->ee_block);
1658 len1 = ext4_ext_get_actual_len(newext);
1659 depth = ext_depth(inode);
1660 if (!path[depth].p_ext)
1662 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1665 * get the next allocated block if the extent in the path
1666 * is before the requested block(s)
1669 b2 = ext4_ext_next_allocated_block(path);
1670 if (b2 == EXT_MAX_BLOCKS)
1672 b2 = EXT4_LBLK_CMASK(sbi, b2);
1675 /* check for wrap through zero on extent logical start block*/
1676 if (b1 + len1 < b1) {
1677 len1 = EXT_MAX_BLOCKS - b1;
1678 newext->ee_len = cpu_to_le16(len1);
1682 /* check for overlap */
1683 if (b1 + len1 > b2) {
1684 newext->ee_len = cpu_to_le16(b2 - b1);
1692 * ext4_ext_insert_extent:
1693 * tries to merge requsted extent into the existing extent or
1694 * inserts requested extent as new one into the tree,
1695 * creating new leaf in the no-space case.
1697 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1698 struct ext4_ext_path *path,
1699 struct ext4_extent *newext, int flag)
1701 struct ext4_extent_header *eh;
1702 struct ext4_extent *ex, *fex;
1703 struct ext4_extent *nearex; /* nearest extent */
1704 struct ext4_ext_path *npath = NULL;
1705 int depth, len, err;
1707 unsigned uninitialized = 0;
1710 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1711 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1714 depth = ext_depth(inode);
1715 ex = path[depth].p_ext;
1716 if (unlikely(path[depth].p_hdr == NULL)) {
1717 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1721 /* try to insert block into found extent and return */
1722 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1723 && ext4_can_extents_be_merged(inode, ex, newext)) {
1724 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1725 ext4_ext_is_uninitialized(newext),
1726 ext4_ext_get_actual_len(newext),
1727 le32_to_cpu(ex->ee_block),
1728 ext4_ext_is_uninitialized(ex),
1729 ext4_ext_get_actual_len(ex),
1730 ext4_ext_pblock(ex));
1731 err = ext4_ext_get_access(handle, inode, path + depth);
1736 * ext4_can_extents_be_merged should have checked that either
1737 * both extents are uninitialized, or both aren't. Thus we
1738 * need to check only one of them here.
1740 if (ext4_ext_is_uninitialized(ex))
1742 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1743 + ext4_ext_get_actual_len(newext));
1745 ext4_ext_mark_uninitialized(ex);
1746 eh = path[depth].p_hdr;
1751 depth = ext_depth(inode);
1752 eh = path[depth].p_hdr;
1753 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1756 /* probably next leaf has space for us? */
1757 fex = EXT_LAST_EXTENT(eh);
1758 next = EXT_MAX_BLOCKS;
1759 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1760 next = ext4_ext_next_leaf_block(path);
1761 if (next != EXT_MAX_BLOCKS) {
1762 ext_debug("next leaf block - %u\n", next);
1763 BUG_ON(npath != NULL);
1764 npath = ext4_ext_find_extent(inode, next, NULL);
1766 return PTR_ERR(npath);
1767 BUG_ON(npath->p_depth != path->p_depth);
1768 eh = npath[depth].p_hdr;
1769 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1770 ext_debug("next leaf isn't full(%d)\n",
1771 le16_to_cpu(eh->eh_entries));
1775 ext_debug("next leaf has no free space(%d,%d)\n",
1776 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1780 * There is no free space in the found leaf.
1781 * We're gonna add a new leaf in the tree.
1783 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1784 flags = EXT4_MB_USE_ROOT_BLOCKS;
1785 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1788 depth = ext_depth(inode);
1789 eh = path[depth].p_hdr;
1792 nearex = path[depth].p_ext;
1794 err = ext4_ext_get_access(handle, inode, path + depth);
1799 /* there is no extent in this leaf, create first one */
1800 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1801 le32_to_cpu(newext->ee_block),
1802 ext4_ext_pblock(newext),
1803 ext4_ext_is_uninitialized(newext),
1804 ext4_ext_get_actual_len(newext));
1805 nearex = EXT_FIRST_EXTENT(eh);
1807 if (le32_to_cpu(newext->ee_block)
1808 > le32_to_cpu(nearex->ee_block)) {
1810 ext_debug("insert %u:%llu:[%d]%d before: "
1812 le32_to_cpu(newext->ee_block),
1813 ext4_ext_pblock(newext),
1814 ext4_ext_is_uninitialized(newext),
1815 ext4_ext_get_actual_len(newext),
1820 BUG_ON(newext->ee_block == nearex->ee_block);
1821 ext_debug("insert %u:%llu:[%d]%d after: "
1823 le32_to_cpu(newext->ee_block),
1824 ext4_ext_pblock(newext),
1825 ext4_ext_is_uninitialized(newext),
1826 ext4_ext_get_actual_len(newext),
1829 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1831 ext_debug("insert %u:%llu:[%d]%d: "
1832 "move %d extents from 0x%p to 0x%p\n",
1833 le32_to_cpu(newext->ee_block),
1834 ext4_ext_pblock(newext),
1835 ext4_ext_is_uninitialized(newext),
1836 ext4_ext_get_actual_len(newext),
1837 len, nearex, nearex + 1);
1838 memmove(nearex + 1, nearex,
1839 len * sizeof(struct ext4_extent));
1843 le16_add_cpu(&eh->eh_entries, 1);
1844 path[depth].p_ext = nearex;
1845 nearex->ee_block = newext->ee_block;
1846 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1847 nearex->ee_len = newext->ee_len;
1850 /* try to merge extents to the right */
1851 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1852 ext4_ext_try_to_merge(inode, path, nearex);
1854 /* try to merge extents to the left */
1856 /* time to correct all indexes above */
1857 err = ext4_ext_correct_indexes(handle, inode, path);
1861 err = ext4_ext_dirty(handle, inode, path + depth);
1865 ext4_ext_drop_refs(npath);
1868 ext4_ext_invalidate_cache(inode);
1872 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1873 ext4_lblk_t num, ext_prepare_callback func,
1876 struct ext4_ext_path *path = NULL;
1877 struct ext4_ext_cache cbex;
1878 struct ext4_extent *ex;
1879 ext4_lblk_t next, start = 0, end = 0;
1880 ext4_lblk_t last = block + num;
1881 int depth, exists, err = 0;
1883 BUG_ON(func == NULL);
1884 BUG_ON(inode == NULL);
1886 while (block < last && block != EXT_MAX_BLOCKS) {
1888 /* find extent for this block */
1889 down_read(&EXT4_I(inode)->i_data_sem);
1890 path = ext4_ext_find_extent(inode, block, path);
1891 up_read(&EXT4_I(inode)->i_data_sem);
1893 err = PTR_ERR(path);
1898 depth = ext_depth(inode);
1899 if (unlikely(path[depth].p_hdr == NULL)) {
1900 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1904 ex = path[depth].p_ext;
1905 next = ext4_ext_next_allocated_block(path);
1909 /* there is no extent yet, so try to allocate
1910 * all requested space */
1913 } else if (le32_to_cpu(ex->ee_block) > block) {
1914 /* need to allocate space before found extent */
1916 end = le32_to_cpu(ex->ee_block);
1917 if (block + num < end)
1919 } else if (block >= le32_to_cpu(ex->ee_block)
1920 + ext4_ext_get_actual_len(ex)) {
1921 /* need to allocate space after found extent */
1926 } else if (block >= le32_to_cpu(ex->ee_block)) {
1928 * some part of requested space is covered
1932 end = le32_to_cpu(ex->ee_block)
1933 + ext4_ext_get_actual_len(ex);
1934 if (block + num < end)
1940 BUG_ON(end <= start);
1943 cbex.ec_block = start;
1944 cbex.ec_len = end - start;
1947 cbex.ec_block = le32_to_cpu(ex->ee_block);
1948 cbex.ec_len = ext4_ext_get_actual_len(ex);
1949 cbex.ec_start = ext4_ext_pblock(ex);
1952 if (unlikely(cbex.ec_len == 0)) {
1953 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1957 err = func(inode, next, &cbex, ex, cbdata);
1958 ext4_ext_drop_refs(path);
1963 if (err == EXT_REPEAT)
1965 else if (err == EXT_BREAK) {
1970 if (ext_depth(inode) != depth) {
1971 /* depth was changed. we have to realloc path */
1976 block = cbex.ec_block + cbex.ec_len;
1980 ext4_ext_drop_refs(path);
1988 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1989 __u32 len, ext4_fsblk_t start)
1991 struct ext4_ext_cache *cex;
1993 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1994 trace_ext4_ext_put_in_cache(inode, block, len, start);
1995 cex = &EXT4_I(inode)->i_cached_extent;
1996 cex->ec_block = block;
1998 cex->ec_start = start;
1999 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2003 * ext4_ext_put_gap_in_cache:
2004 * calculate boundaries of the gap that the requested block fits into
2005 * and cache this gap
2008 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2011 int depth = ext_depth(inode);
2014 struct ext4_extent *ex;
2016 ex = path[depth].p_ext;
2018 /* there is no extent yet, so gap is [0;-] */
2020 len = EXT_MAX_BLOCKS;
2021 ext_debug("cache gap(whole file):");
2022 } else if (block < le32_to_cpu(ex->ee_block)) {
2024 len = le32_to_cpu(ex->ee_block) - block;
2025 ext_debug("cache gap(before): %u [%u:%u]",
2027 le32_to_cpu(ex->ee_block),
2028 ext4_ext_get_actual_len(ex));
2029 } else if (block >= le32_to_cpu(ex->ee_block)
2030 + ext4_ext_get_actual_len(ex)) {
2032 lblock = le32_to_cpu(ex->ee_block)
2033 + ext4_ext_get_actual_len(ex);
2035 next = ext4_ext_next_allocated_block(path);
2036 ext_debug("cache gap(after): [%u:%u] %u",
2037 le32_to_cpu(ex->ee_block),
2038 ext4_ext_get_actual_len(ex),
2040 BUG_ON(next == lblock);
2041 len = next - lblock;
2047 ext_debug(" -> %u:%lu\n", lblock, len);
2048 ext4_ext_put_in_cache(inode, lblock, len, 0);
2052 * ext4_ext_check_cache()
2053 * Checks to see if the given block is in the cache.
2054 * If it is, the cached extent is stored in the given
2055 * cache extent pointer. If the cached extent is a hole,
2056 * this routine should be used instead of
2057 * ext4_ext_in_cache if the calling function needs to
2058 * know the size of the hole.
2060 * @inode: The files inode
2061 * @block: The block to look for in the cache
2062 * @ex: Pointer where the cached extent will be stored
2063 * if it contains block
2065 * Return 0 if cache is invalid; 1 if the cache is valid
2067 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2068 struct ext4_ext_cache *ex){
2069 struct ext4_ext_cache *cex;
2070 struct ext4_sb_info *sbi;
2074 * We borrow i_block_reservation_lock to protect i_cached_extent
2076 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2077 cex = &EXT4_I(inode)->i_cached_extent;
2078 sbi = EXT4_SB(inode->i_sb);
2080 /* has cache valid data? */
2081 if (cex->ec_len == 0)
2084 if (in_range(block, cex->ec_block, cex->ec_len)) {
2085 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2086 ext_debug("%u cached by %u:%u:%llu\n",
2088 cex->ec_block, cex->ec_len, cex->ec_start);
2092 trace_ext4_ext_in_cache(inode, block, ret);
2093 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2098 * ext4_ext_in_cache()
2099 * Checks to see if the given block is in the cache.
2100 * If it is, the cached extent is stored in the given
2103 * @inode: The files inode
2104 * @block: The block to look for in the cache
2105 * @ex: Pointer where the cached extent will be stored
2106 * if it contains block
2108 * Return 0 if cache is invalid; 1 if the cache is valid
2111 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2112 struct ext4_extent *ex)
2114 struct ext4_ext_cache cex;
2117 if (ext4_ext_check_cache(inode, block, &cex)) {
2118 ex->ee_block = cpu_to_le32(cex.ec_block);
2119 ext4_ext_store_pblock(ex, cex.ec_start);
2120 ex->ee_len = cpu_to_le16(cex.ec_len);
2130 * removes index from the index block.
2132 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2133 struct ext4_ext_path *path, int depth)
2138 /* free index block */
2140 path = path + depth;
2141 leaf = ext4_idx_pblock(path->p_idx);
2142 if (unlikely(path->p_hdr->eh_entries == 0)) {
2143 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2146 err = ext4_ext_get_access(handle, inode, path);
2150 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2151 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2152 len *= sizeof(struct ext4_extent_idx);
2153 memmove(path->p_idx, path->p_idx + 1, len);
2156 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2157 err = ext4_ext_dirty(handle, inode, path);
2160 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2161 trace_ext4_ext_rm_idx(inode, leaf);
2163 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2164 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2166 while (--depth >= 0) {
2167 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2170 err = ext4_ext_get_access(handle, inode, path);
2173 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2174 err = ext4_ext_dirty(handle, inode, path);
2182 * ext4_ext_calc_credits_for_single_extent:
2183 * This routine returns max. credits that needed to insert an extent
2184 * to the extent tree.
2185 * When pass the actual path, the caller should calculate credits
2188 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2189 struct ext4_ext_path *path)
2192 int depth = ext_depth(inode);
2195 /* probably there is space in leaf? */
2196 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2197 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2200 * There are some space in the leaf tree, no
2201 * need to account for leaf block credit
2203 * bitmaps and block group descriptor blocks
2204 * and other metadata blocks still need to be
2207 /* 1 bitmap, 1 block group descriptor */
2208 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2213 return ext4_chunk_trans_blocks(inode, nrblocks);
2217 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2219 * if nrblocks are fit in a single extent (chunk flag is 1), then
2220 * in the worse case, each tree level index/leaf need to be changed
2221 * if the tree split due to insert a new extent, then the old tree
2222 * index/leaf need to be updated too
2224 * If the nrblocks are discontiguous, they could cause
2225 * the whole tree split more than once, but this is really rare.
2227 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2230 int depth = ext_depth(inode);
2240 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2241 struct ext4_extent *ex,
2242 ext4_fsblk_t *partial_cluster,
2243 ext4_lblk_t from, ext4_lblk_t to)
2245 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2246 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2248 int flags = EXT4_FREE_BLOCKS_FORGET;
2250 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2251 flags |= EXT4_FREE_BLOCKS_METADATA;
2253 * For bigalloc file systems, we never free a partial cluster
2254 * at the beginning of the extent. Instead, we make a note
2255 * that we tried freeing the cluster, and check to see if we
2256 * need to free it on a subsequent call to ext4_remove_blocks,
2257 * or at the end of the ext4_truncate() operation.
2259 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2261 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2263 * If we have a partial cluster, and it's different from the
2264 * cluster of the last block, we need to explicitly free the
2265 * partial cluster here.
2267 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2268 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2269 ext4_free_blocks(handle, inode, NULL,
2270 EXT4_C2B(sbi, *partial_cluster),
2271 sbi->s_cluster_ratio, flags);
2272 *partial_cluster = 0;
2275 #ifdef EXTENTS_STATS
2277 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2278 spin_lock(&sbi->s_ext_stats_lock);
2279 sbi->s_ext_blocks += ee_len;
2280 sbi->s_ext_extents++;
2281 if (ee_len < sbi->s_ext_min)
2282 sbi->s_ext_min = ee_len;
2283 if (ee_len > sbi->s_ext_max)
2284 sbi->s_ext_max = ee_len;
2285 if (ext_depth(inode) > sbi->s_depth_max)
2286 sbi->s_depth_max = ext_depth(inode);
2287 spin_unlock(&sbi->s_ext_stats_lock);
2290 if (from >= le32_to_cpu(ex->ee_block)
2291 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2295 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2296 pblk = ext4_ext_pblock(ex) + ee_len - num;
2297 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2298 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2300 * If the block range to be freed didn't start at the
2301 * beginning of a cluster, and we removed the entire
2302 * extent, save the partial cluster here, since we
2303 * might need to delete if we determine that the
2304 * truncate operation has removed all of the blocks in
2307 if (EXT4_PBLK_COFF(sbi, pblk) &&
2309 *partial_cluster = EXT4_B2C(sbi, pblk);
2311 *partial_cluster = 0;
2312 } else if (from == le32_to_cpu(ex->ee_block)
2313 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2319 start = ext4_ext_pblock(ex);
2321 ext_debug("free first %u blocks starting %llu\n", num, start);
2322 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2325 printk(KERN_INFO "strange request: removal(2) "
2326 "%u-%u from %u:%u\n",
2327 from, to, le32_to_cpu(ex->ee_block), ee_len);
2334 * ext4_ext_rm_leaf() Removes the extents associated with the
2335 * blocks appearing between "start" and "end", and splits the extents
2336 * if "start" and "end" appear in the same extent
2338 * @handle: The journal handle
2339 * @inode: The files inode
2340 * @path: The path to the leaf
2341 * @start: The first block to remove
2342 * @end: The last block to remove
2345 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2346 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2347 ext4_lblk_t start, ext4_lblk_t end)
2349 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2350 int err = 0, correct_index = 0;
2351 int depth = ext_depth(inode), credits;
2352 struct ext4_extent_header *eh;
2355 ext4_lblk_t ex_ee_block;
2356 unsigned short ex_ee_len;
2357 unsigned uninitialized = 0;
2358 struct ext4_extent *ex;
2360 /* the header must be checked already in ext4_ext_remove_space() */
2361 ext_debug("truncate since %u in leaf to %u\n", start, end);
2362 if (!path[depth].p_hdr)
2363 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2364 eh = path[depth].p_hdr;
2365 if (unlikely(path[depth].p_hdr == NULL)) {
2366 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2369 /* find where to start removing */
2370 ex = EXT_LAST_EXTENT(eh);
2372 ex_ee_block = le32_to_cpu(ex->ee_block);
2373 ex_ee_len = ext4_ext_get_actual_len(ex);
2376 * If we're starting with an extent other than the last one in the
2377 * node, we need to see if it shares a cluster with the extent to
2378 * the right (towards the end of the file). If its leftmost cluster
2379 * is this extent's rightmost cluster and it is not cluster aligned,
2380 * we'll mark it as a partial that is not to be deallocated.
2383 if (ex != EXT_LAST_EXTENT(eh)) {
2384 ext4_fsblk_t current_pblk, right_pblk;
2385 long long current_cluster, right_cluster;
2387 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2388 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2389 right_pblk = ext4_ext_pblock(ex + 1);
2390 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2391 if (current_cluster == right_cluster &&
2392 EXT4_PBLK_COFF(sbi, right_pblk))
2393 *partial_cluster = -right_cluster;
2396 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2398 while (ex >= EXT_FIRST_EXTENT(eh) &&
2399 ex_ee_block + ex_ee_len > start) {
2401 if (ext4_ext_is_uninitialized(ex))
2406 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2407 uninitialized, ex_ee_len);
2408 path[depth].p_ext = ex;
2410 a = ex_ee_block > start ? ex_ee_block : start;
2411 b = ex_ee_block+ex_ee_len - 1 < end ?
2412 ex_ee_block+ex_ee_len - 1 : end;
2414 ext_debug(" border %u:%u\n", a, b);
2416 /* If this extent is beyond the end of the hole, skip it */
2417 if (end < ex_ee_block) {
2419 ex_ee_block = le32_to_cpu(ex->ee_block);
2420 ex_ee_len = ext4_ext_get_actual_len(ex);
2422 } else if (b != ex_ee_block + ex_ee_len - 1) {
2423 EXT4_ERROR_INODE(inode," bad truncate %u:%u\n",
2427 } else if (a != ex_ee_block) {
2428 /* remove tail of the extent */
2429 num = a - ex_ee_block;
2431 /* remove whole extent: excellent! */
2435 * 3 for leaf, sb, and inode plus 2 (bmap and group
2436 * descriptor) for each block group; assume two block
2437 * groups plus ex_ee_len/blocks_per_block_group for
2440 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2441 if (ex == EXT_FIRST_EXTENT(eh)) {
2443 credits += (ext_depth(inode)) + 1;
2445 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2447 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2451 err = ext4_ext_get_access(handle, inode, path + depth);
2455 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2461 /* this extent is removed; mark slot entirely unused */
2462 ext4_ext_store_pblock(ex, 0);
2464 ex->ee_len = cpu_to_le16(num);
2466 * Do not mark uninitialized if all the blocks in the
2467 * extent have been removed.
2469 if (uninitialized && num)
2470 ext4_ext_mark_uninitialized(ex);
2472 * If the extent was completely released,
2473 * we need to remove it from the leaf
2476 if (end != EXT_MAX_BLOCKS - 1) {
2478 * For hole punching, we need to scoot all the
2479 * extents up when an extent is removed so that
2480 * we dont have blank extents in the middle
2482 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2483 sizeof(struct ext4_extent));
2485 /* Now get rid of the one at the end */
2486 memset(EXT_LAST_EXTENT(eh), 0,
2487 sizeof(struct ext4_extent));
2489 le16_add_cpu(&eh->eh_entries, -1);
2491 *partial_cluster = 0;
2493 err = ext4_ext_dirty(handle, inode, path + depth);
2497 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2498 ext4_ext_pblock(ex));
2500 ex_ee_block = le32_to_cpu(ex->ee_block);
2501 ex_ee_len = ext4_ext_get_actual_len(ex);
2504 if (correct_index && eh->eh_entries)
2505 err = ext4_ext_correct_indexes(handle, inode, path);
2508 * If there is still a entry in the leaf node, check to see if
2509 * it references the partial cluster. This is the only place
2510 * where it could; if it doesn't, we can free the cluster.
2512 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2513 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2514 *partial_cluster)) {
2515 int flags = EXT4_FREE_BLOCKS_FORGET;
2517 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2518 flags |= EXT4_FREE_BLOCKS_METADATA;
2520 ext4_free_blocks(handle, inode, NULL,
2521 EXT4_C2B(sbi, *partial_cluster),
2522 sbi->s_cluster_ratio, flags);
2523 *partial_cluster = 0;
2526 /* if this leaf is free, then we should
2527 * remove it from index block above */
2528 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2529 err = ext4_ext_rm_idx(handle, inode, path, depth);
2536 * ext4_ext_more_to_rm:
2537 * returns 1 if current index has to be freed (even partial)
2540 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2542 BUG_ON(path->p_idx == NULL);
2544 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2548 * if truncate on deeper level happened, it wasn't partial,
2549 * so we have to consider current index for truncation
2551 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2556 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2559 struct super_block *sb = inode->i_sb;
2560 int depth = ext_depth(inode);
2561 struct ext4_ext_path *path = NULL;
2562 ext4_fsblk_t partial_cluster = 0;
2566 ext_debug("truncate since %u to %u\n", start, end);
2568 /* probably first extent we're gonna free will be last in block */
2569 handle = ext4_journal_start(inode, depth + 1);
2571 return PTR_ERR(handle);
2574 ext4_ext_invalidate_cache(inode);
2576 trace_ext4_ext_remove_space(inode, start, depth);
2579 * Check if we are removing extents inside the extent tree. If that
2580 * is the case, we are going to punch a hole inside the extent tree
2581 * so we have to check whether we need to split the extent covering
2582 * the last block to remove so we can easily remove the part of it
2583 * in ext4_ext_rm_leaf().
2585 if (end < EXT_MAX_BLOCKS - 1) {
2586 struct ext4_extent *ex;
2587 ext4_lblk_t ee_block;
2589 /* find extent for this block */
2590 path = ext4_ext_find_extent(inode, end, NULL);
2592 ext4_journal_stop(handle);
2593 return PTR_ERR(path);
2595 depth = ext_depth(inode);
2596 ex = path[depth].p_ext;
2598 ext4_ext_drop_refs(path);
2604 ee_block = le32_to_cpu(ex->ee_block);
2607 * See if the last block is inside the extent, if so split
2608 * the extent at 'end' block so we can easily remove the
2609 * tail of the first part of the split extent in
2610 * ext4_ext_rm_leaf().
2612 if (end >= ee_block &&
2613 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2616 if (ext4_ext_is_uninitialized(ex))
2617 split_flag = EXT4_EXT_MARK_UNINIT1 |
2618 EXT4_EXT_MARK_UNINIT2;
2621 * Split the extent in two so that 'end' is the last
2622 * block in the first new extent
2624 err = ext4_split_extent_at(handle, inode, path,
2625 end + 1, split_flag,
2626 EXT4_GET_BLOCKS_PRE_IO |
2627 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2636 * We start scanning from right side, freeing all the blocks
2637 * after i_size and walking into the tree depth-wise.
2639 depth = ext_depth(inode);
2644 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2646 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2649 ext4_journal_stop(handle);
2652 path[0].p_depth = depth;
2653 path[0].p_hdr = ext_inode_hdr(inode);
2656 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2663 while (i >= 0 && err == 0) {
2665 /* this is leaf block */
2666 err = ext4_ext_rm_leaf(handle, inode, path,
2667 &partial_cluster, start,
2669 /* root level has p_bh == NULL, brelse() eats this */
2670 brelse(path[i].p_bh);
2671 path[i].p_bh = NULL;
2676 /* this is index block */
2677 if (!path[i].p_hdr) {
2678 ext_debug("initialize header\n");
2679 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2682 if (!path[i].p_idx) {
2683 /* this level hasn't been touched yet */
2684 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2685 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2686 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2688 le16_to_cpu(path[i].p_hdr->eh_entries));
2690 /* we were already here, see at next index */
2694 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2695 i, EXT_FIRST_INDEX(path[i].p_hdr),
2697 if (ext4_ext_more_to_rm(path + i)) {
2698 struct buffer_head *bh;
2699 /* go to the next level */
2700 ext_debug("move to level %d (block %llu)\n",
2701 i + 1, ext4_idx_pblock(path[i].p_idx));
2702 memset(path + i + 1, 0, sizeof(*path));
2703 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2705 /* should we reset i_size? */
2709 if (WARN_ON(i + 1 > depth)) {
2713 if (ext4_ext_check(inode, ext_block_hdr(bh),
2718 path[i + 1].p_bh = bh;
2720 /* save actual number of indexes since this
2721 * number is changed at the next iteration */
2722 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2725 /* we finished processing this index, go up */
2726 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2727 /* index is empty, remove it;
2728 * handle must be already prepared by the
2729 * truncatei_leaf() */
2730 err = ext4_ext_rm_idx(handle, inode, path, i);
2732 /* root level has p_bh == NULL, brelse() eats this */
2733 brelse(path[i].p_bh);
2734 path[i].p_bh = NULL;
2736 ext_debug("return to level %d\n", i);
2740 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2741 path->p_hdr->eh_entries);
2743 /* If we still have something in the partial cluster and we have removed
2744 * even the first extent, then we should free the blocks in the partial
2745 * cluster as well. */
2746 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2747 int flags = EXT4_FREE_BLOCKS_FORGET;
2749 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2750 flags |= EXT4_FREE_BLOCKS_METADATA;
2752 ext4_free_blocks(handle, inode, NULL,
2753 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2754 EXT4_SB(sb)->s_cluster_ratio, flags);
2755 partial_cluster = 0;
2758 /* TODO: flexible tree reduction should be here */
2759 if (path->p_hdr->eh_entries == 0) {
2761 * truncate to zero freed all the tree,
2762 * so we need to correct eh_depth
2764 err = ext4_ext_get_access(handle, inode, path);
2766 ext_inode_hdr(inode)->eh_depth = 0;
2767 ext_inode_hdr(inode)->eh_max =
2768 cpu_to_le16(ext4_ext_space_root(inode, 0));
2769 err = ext4_ext_dirty(handle, inode, path);
2773 ext4_ext_drop_refs(path);
2775 if (err == -EAGAIN) {
2779 ext4_journal_stop(handle);
2785 * called at mount time
2787 void ext4_ext_init(struct super_block *sb)
2790 * possible initialization would be here
2793 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2794 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2795 printk(KERN_INFO "EXT4-fs: file extents enabled");
2796 #ifdef AGGRESSIVE_TEST
2797 printk(", aggressive tests");
2799 #ifdef CHECK_BINSEARCH
2800 printk(", check binsearch");
2802 #ifdef EXTENTS_STATS
2807 #ifdef EXTENTS_STATS
2808 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2809 EXT4_SB(sb)->s_ext_min = 1 << 30;
2810 EXT4_SB(sb)->s_ext_max = 0;
2816 * called at umount time
2818 void ext4_ext_release(struct super_block *sb)
2820 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2823 #ifdef EXTENTS_STATS
2824 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2825 struct ext4_sb_info *sbi = EXT4_SB(sb);
2826 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2827 sbi->s_ext_blocks, sbi->s_ext_extents,
2828 sbi->s_ext_blocks / sbi->s_ext_extents);
2829 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2830 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2835 /* FIXME!! we need to try to merge to left or right after zero-out */
2836 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2838 ext4_fsblk_t ee_pblock;
2839 unsigned int ee_len;
2842 ee_len = ext4_ext_get_actual_len(ex);
2843 ee_pblock = ext4_ext_pblock(ex);
2845 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2853 * ext4_split_extent_at() splits an extent at given block.
2855 * @handle: the journal handle
2856 * @inode: the file inode
2857 * @path: the path to the extent
2858 * @split: the logical block where the extent is splitted.
2859 * @split_flags: indicates if the extent could be zeroout if split fails, and
2860 * the states(init or uninit) of new extents.
2861 * @flags: flags used to insert new extent to extent tree.
2864 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2865 * of which are deterimined by split_flag.
2867 * There are two cases:
2868 * a> the extent are splitted into two extent.
2869 * b> split is not needed, and just mark the extent.
2871 * return 0 on success.
2873 static int ext4_split_extent_at(handle_t *handle,
2874 struct inode *inode,
2875 struct ext4_ext_path *path,
2880 ext4_fsblk_t newblock;
2881 ext4_lblk_t ee_block;
2882 struct ext4_extent *ex, newex, orig_ex;
2883 struct ext4_extent *ex2 = NULL;
2884 unsigned int ee_len, depth;
2887 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2888 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2890 ext_debug("ext4_split_extents_at: inode %lu, logical"
2891 "block %llu\n", inode->i_ino, (unsigned long long)split);
2893 ext4_ext_show_leaf(inode, path);
2895 depth = ext_depth(inode);
2896 ex = path[depth].p_ext;
2897 ee_block = le32_to_cpu(ex->ee_block);
2898 ee_len = ext4_ext_get_actual_len(ex);
2899 newblock = split - ee_block + ext4_ext_pblock(ex);
2901 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2903 err = ext4_ext_get_access(handle, inode, path + depth);
2907 if (split == ee_block) {
2909 * case b: block @split is the block that the extent begins with
2910 * then we just change the state of the extent, and splitting
2913 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2914 ext4_ext_mark_uninitialized(ex);
2916 ext4_ext_mark_initialized(ex);
2918 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2919 ext4_ext_try_to_merge(inode, path, ex);
2921 err = ext4_ext_dirty(handle, inode, path + depth);
2926 memcpy(&orig_ex, ex, sizeof(orig_ex));
2927 ex->ee_len = cpu_to_le16(split - ee_block);
2928 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2929 ext4_ext_mark_uninitialized(ex);
2932 * path may lead to new leaf, not to original leaf any more
2933 * after ext4_ext_insert_extent() returns,
2935 err = ext4_ext_dirty(handle, inode, path + depth);
2937 goto fix_extent_len;
2940 ex2->ee_block = cpu_to_le32(split);
2941 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2942 ext4_ext_store_pblock(ex2, newblock);
2943 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2944 ext4_ext_mark_uninitialized(ex2);
2946 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2947 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2948 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
2949 if (split_flag & EXT4_EXT_DATA_VALID1)
2950 err = ext4_ext_zeroout(inode, ex2);
2952 err = ext4_ext_zeroout(inode, ex);
2954 err = ext4_ext_zeroout(inode, &orig_ex);
2957 goto fix_extent_len;
2958 /* update the extent length and mark as initialized */
2959 ex->ee_len = cpu_to_le16(ee_len);
2960 ext4_ext_try_to_merge(inode, path, ex);
2961 err = ext4_ext_dirty(handle, inode, path + depth);
2964 goto fix_extent_len;
2967 ext4_ext_show_leaf(inode, path);
2971 ex->ee_len = orig_ex.ee_len;
2972 ext4_ext_dirty(handle, inode, path + depth);
2977 * ext4_split_extents() splits an extent and mark extent which is covered
2978 * by @map as split_flags indicates
2980 * It may result in splitting the extent into multiple extents (upto three)
2981 * There are three possibilities:
2982 * a> There is no split required
2983 * b> Splits in two extents: Split is happening at either end of the extent
2984 * c> Splits in three extents: Somone is splitting in middle of the extent
2987 static int ext4_split_extent(handle_t *handle,
2988 struct inode *inode,
2989 struct ext4_ext_path *path,
2990 struct ext4_map_blocks *map,
2994 ext4_lblk_t ee_block;
2995 struct ext4_extent *ex;
2996 unsigned int ee_len, depth;
2999 int split_flag1, flags1;
3000 int allocated = map->m_len;
3002 depth = ext_depth(inode);
3003 ex = path[depth].p_ext;
3004 ee_block = le32_to_cpu(ex->ee_block);
3005 ee_len = ext4_ext_get_actual_len(ex);
3006 uninitialized = ext4_ext_is_uninitialized(ex);
3008 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3009 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3010 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3012 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3013 EXT4_EXT_MARK_UNINIT2;
3014 if (split_flag & EXT4_EXT_DATA_VALID2)
3015 split_flag1 |= EXT4_EXT_DATA_VALID1;
3016 err = ext4_split_extent_at(handle, inode, path,
3017 map->m_lblk + map->m_len, split_flag1, flags1);
3021 allocated = ee_len - (map->m_lblk - ee_block);
3024 ext4_ext_drop_refs(path);
3025 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3027 return PTR_ERR(path);
3029 if (map->m_lblk >= ee_block) {
3030 split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT |
3031 EXT4_EXT_DATA_VALID2);
3033 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3034 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3035 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3036 err = ext4_split_extent_at(handle, inode, path,
3037 map->m_lblk, split_flag1, flags);
3042 ext4_ext_show_leaf(inode, path);
3044 return err ? err : allocated;
3047 #define EXT4_EXT_ZERO_LEN 7
3049 * This function is called by ext4_ext_map_blocks() if someone tries to write
3050 * to an uninitialized extent. It may result in splitting the uninitialized
3051 * extent into multiple extents (up to three - one initialized and two
3053 * There are three possibilities:
3054 * a> There is no split required: Entire extent should be initialized
3055 * b> Splits in two extents: Write is happening at either end of the extent
3056 * c> Splits in three extents: Somone is writing in middle of the extent
3059 * - The extent pointed to by 'path' is uninitialized.
3060 * - The extent pointed to by 'path' contains a superset
3061 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3063 * Post-conditions on success:
3064 * - the returned value is the number of blocks beyond map->l_lblk
3065 * that are allocated and initialized.
3066 * It is guaranteed to be >= map->m_len.
3068 static int ext4_ext_convert_to_initialized(handle_t *handle,
3069 struct inode *inode,
3070 struct ext4_map_blocks *map,
3071 struct ext4_ext_path *path)
3073 struct ext4_extent_header *eh;
3074 struct ext4_map_blocks split_map;
3075 struct ext4_extent zero_ex;
3076 struct ext4_extent *ex;
3077 ext4_lblk_t ee_block, eof_block;
3078 unsigned int ee_len, depth;
3083 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3084 "block %llu, max_blocks %u\n", inode->i_ino,
3085 (unsigned long long)map->m_lblk, map->m_len);
3087 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3088 inode->i_sb->s_blocksize_bits;
3089 if (eof_block < map->m_lblk + map->m_len)
3090 eof_block = map->m_lblk + map->m_len;
3092 depth = ext_depth(inode);
3093 eh = path[depth].p_hdr;
3094 ex = path[depth].p_ext;
3095 ee_block = le32_to_cpu(ex->ee_block);
3096 ee_len = ext4_ext_get_actual_len(ex);
3097 allocated = ee_len - (map->m_lblk - ee_block);
3099 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3101 /* Pre-conditions */
3102 BUG_ON(!ext4_ext_is_uninitialized(ex));
3103 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3106 * Attempt to transfer newly initialized blocks from the currently
3107 * uninitialized extent to its left neighbor. This is much cheaper
3108 * than an insertion followed by a merge as those involve costly
3109 * memmove() calls. This is the common case in steady state for
3110 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3113 * Limitations of the current logic:
3114 * - L1: we only deal with writes at the start of the extent.
3115 * The approach could be extended to writes at the end
3116 * of the extent but this scenario was deemed less common.
3117 * - L2: we do not deal with writes covering the whole extent.
3118 * This would require removing the extent if the transfer
3120 * - L3: we only attempt to merge with an extent stored in the
3121 * same extent tree node.
3123 if ((map->m_lblk == ee_block) && /*L1*/
3124 (map->m_len < ee_len) && /*L2*/
3125 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
3126 struct ext4_extent *prev_ex;
3127 ext4_lblk_t prev_lblk;
3128 ext4_fsblk_t prev_pblk, ee_pblk;
3129 unsigned int prev_len, write_len;
3132 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3133 prev_len = ext4_ext_get_actual_len(prev_ex);
3134 prev_pblk = ext4_ext_pblock(prev_ex);
3135 ee_pblk = ext4_ext_pblock(ex);
3136 write_len = map->m_len;
3139 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3140 * upon those conditions:
3141 * - C1: prev_ex is initialized,
3142 * - C2: prev_ex is logically abutting ex,
3143 * - C3: prev_ex is physically abutting ex,
3144 * - C4: prev_ex can receive the additional blocks without
3145 * overflowing the (initialized) length limit.
3147 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3148 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3149 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3150 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3151 err = ext4_ext_get_access(handle, inode, path + depth);
3155 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3158 /* Shift the start of ex by 'write_len' blocks */
3159 ex->ee_block = cpu_to_le32(ee_block + write_len);
3160 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3161 ex->ee_len = cpu_to_le16(ee_len - write_len);
3162 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3164 /* Extend prev_ex by 'write_len' blocks */
3165 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3167 /* Mark the block containing both extents as dirty */
3168 ext4_ext_dirty(handle, inode, path + depth);
3170 /* Update path to point to the right extent */
3171 path[depth].p_ext = prev_ex;
3173 /* Result: number of initialized blocks past m_lblk */
3174 allocated = write_len;
3179 WARN_ON(map->m_lblk < ee_block);
3181 * It is safe to convert extent to initialized via explicit
3182 * zeroout only if extent is fully insde i_size or new_size.
3184 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3186 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3187 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3188 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3189 err = ext4_ext_zeroout(inode, ex);
3193 err = ext4_ext_get_access(handle, inode, path + depth);
3196 ext4_ext_mark_initialized(ex);
3197 ext4_ext_try_to_merge(inode, path, ex);
3198 err = ext4_ext_dirty(handle, inode, path + depth);
3204 * 1. split the extent into three extents.
3205 * 2. split the extent into two extents, zeroout the first half.
3206 * 3. split the extent into two extents, zeroout the second half.
3207 * 4. split the extent into two extents with out zeroout.
3209 split_map.m_lblk = map->m_lblk;
3210 split_map.m_len = map->m_len;
3212 if (allocated > map->m_len) {
3213 if (allocated <= EXT4_EXT_ZERO_LEN &&
3214 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3217 cpu_to_le32(map->m_lblk);
3218 zero_ex.ee_len = cpu_to_le16(allocated);
3219 ext4_ext_store_pblock(&zero_ex,
3220 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3221 err = ext4_ext_zeroout(inode, &zero_ex);
3224 split_map.m_lblk = map->m_lblk;
3225 split_map.m_len = allocated;
3226 } else if ((map->m_lblk - ee_block + map->m_len <
3227 EXT4_EXT_ZERO_LEN) &&
3228 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3230 if (map->m_lblk != ee_block) {
3231 zero_ex.ee_block = ex->ee_block;
3232 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3234 ext4_ext_store_pblock(&zero_ex,
3235 ext4_ext_pblock(ex));
3236 err = ext4_ext_zeroout(inode, &zero_ex);
3241 split_map.m_lblk = ee_block;
3242 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3243 allocated = map->m_len;
3247 allocated = ext4_split_extent(handle, inode, path,
3248 &split_map, split_flag, 0);
3253 return err ? err : allocated;
3257 * This function is called by ext4_ext_map_blocks() from
3258 * ext4_get_blocks_dio_write() when DIO to write
3259 * to an uninitialized extent.
3261 * Writing to an uninitialized extent may result in splitting the uninitialized
3262 * extent into multiple /initialized uninitialized extents (up to three)
3263 * There are three possibilities:
3264 * a> There is no split required: Entire extent should be uninitialized
3265 * b> Splits in two extents: Write is happening at either end of the extent
3266 * c> Splits in three extents: Somone is writing in middle of the extent
3268 * One of more index blocks maybe needed if the extent tree grow after
3269 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3270 * complete, we need to split the uninitialized extent before DIO submit
3271 * the IO. The uninitialized extent called at this time will be split
3272 * into three uninitialized extent(at most). After IO complete, the part
3273 * being filled will be convert to initialized by the end_io callback function
3274 * via ext4_convert_unwritten_extents().
3276 * Returns the size of uninitialized extent to be written on success.
3278 static int ext4_split_unwritten_extents(handle_t *handle,
3279 struct inode *inode,
3280 struct ext4_map_blocks *map,
3281 struct ext4_ext_path *path,
3284 ext4_lblk_t eof_block;
3285 ext4_lblk_t ee_block;
3286 struct ext4_extent *ex;
3287 unsigned int ee_len;
3288 int split_flag = 0, depth;
3290 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3291 "block %llu, max_blocks %u\n", inode->i_ino,
3292 (unsigned long long)map->m_lblk, map->m_len);
3294 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3295 inode->i_sb->s_blocksize_bits;
3296 if (eof_block < map->m_lblk + map->m_len)
3297 eof_block = map->m_lblk + map->m_len;
3299 * It is safe to convert extent to initialized via explicit
3300 * zeroout only if extent is fully insde i_size or new_size.
3302 depth = ext_depth(inode);
3303 ex = path[depth].p_ext;
3304 ee_block = le32_to_cpu(ex->ee_block);
3305 ee_len = ext4_ext_get_actual_len(ex);
3307 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3308 split_flag |= EXT4_EXT_MARK_UNINIT2;
3309 if (flags & EXT4_GET_BLOCKS_CONVERT)
3310 split_flag |= EXT4_EXT_DATA_VALID2;
3311 flags |= EXT4_GET_BLOCKS_PRE_IO;
3312 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3315 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3316 struct inode *inode,
3317 struct ext4_map_blocks *map,
3318 struct ext4_ext_path *path)
3320 struct ext4_extent *ex;
3321 ext4_lblk_t ee_block;
3322 unsigned int ee_len;
3326 depth = ext_depth(inode);
3327 ex = path[depth].p_ext;
3328 ee_block = le32_to_cpu(ex->ee_block);
3329 ee_len = ext4_ext_get_actual_len(ex);
3331 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3332 "block %llu, max_blocks %u\n", inode->i_ino,
3333 (unsigned long long)ee_block, ee_len);
3335 /* If extent is larger than requested then split is required */
3336 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3337 err = ext4_split_unwritten_extents(handle, inode, map, path,
3338 EXT4_GET_BLOCKS_CONVERT);
3341 ext4_ext_drop_refs(path);
3342 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3344 err = PTR_ERR(path);
3347 depth = ext_depth(inode);
3348 ex = path[depth].p_ext;
3351 err = ext4_ext_get_access(handle, inode, path + depth);
3354 /* first mark the extent as initialized */
3355 ext4_ext_mark_initialized(ex);
3357 /* note: ext4_ext_correct_indexes() isn't needed here because
3358 * borders are not changed
3360 ext4_ext_try_to_merge(inode, path, ex);
3362 /* Mark modified extent as dirty */
3363 err = ext4_ext_dirty(handle, inode, path + depth);
3365 ext4_ext_show_leaf(inode, path);
3369 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3370 sector_t block, int count)
3373 for (i = 0; i < count; i++)
3374 unmap_underlying_metadata(bdev, block + i);
3378 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3380 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3382 struct ext4_ext_path *path,
3386 struct ext4_extent_header *eh;
3387 struct ext4_extent *last_ex;
3389 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3392 depth = ext_depth(inode);
3393 eh = path[depth].p_hdr;
3395 if (unlikely(!eh->eh_entries)) {
3396 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3397 "EOFBLOCKS_FL set");
3400 last_ex = EXT_LAST_EXTENT(eh);
3402 * We should clear the EOFBLOCKS_FL flag if we are writing the
3403 * last block in the last extent in the file. We test this by
3404 * first checking to see if the caller to
3405 * ext4_ext_get_blocks() was interested in the last block (or
3406 * a block beyond the last block) in the current extent. If
3407 * this turns out to be false, we can bail out from this
3408 * function immediately.
3410 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3411 ext4_ext_get_actual_len(last_ex))
3414 * If the caller does appear to be planning to write at or
3415 * beyond the end of the current extent, we then test to see
3416 * if the current extent is the last extent in the file, by
3417 * checking to make sure it was reached via the rightmost node
3418 * at each level of the tree.
3420 for (i = depth-1; i >= 0; i--)
3421 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3423 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3424 return ext4_mark_inode_dirty(handle, inode);
3428 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3430 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3431 * whether there are any buffers marked for delayed allocation. It returns '1'
3432 * on the first delalloc'ed buffer head found. If no buffer head in the given
3433 * range is marked for delalloc, it returns 0.
3434 * lblk_start should always be <= lblk_end.
3435 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3436 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3437 * block sooner). This is useful when blocks are truncated sequentially from
3438 * lblk_start towards lblk_end.
3440 static int ext4_find_delalloc_range(struct inode *inode,
3441 ext4_lblk_t lblk_start,
3442 ext4_lblk_t lblk_end,
3443 int search_hint_reverse)
3445 struct address_space *mapping = inode->i_mapping;
3446 struct buffer_head *head, *bh = NULL;
3448 ext4_lblk_t i, pg_lblk;
3451 /* reverse search wont work if fs block size is less than page size */
3452 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3453 search_hint_reverse = 0;
3455 if (search_hint_reverse)
3460 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3462 while ((i >= lblk_start) && (i <= lblk_end)) {
3463 page = find_get_page(mapping, index);
3467 if (!page_has_buffers(page))
3470 head = page_buffers(page);
3475 pg_lblk = index << (PAGE_CACHE_SHIFT -
3478 if (unlikely(pg_lblk < lblk_start)) {
3480 * This is possible when fs block size is less
3481 * than page size and our cluster starts/ends in
3482 * middle of the page. So we need to skip the
3483 * initial few blocks till we reach the 'lblk'
3489 /* Check if the buffer is delayed allocated and that it
3490 * is not yet mapped. (when da-buffers are mapped during
3491 * their writeout, their da_mapped bit is set.)
3493 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3494 page_cache_release(page);
3495 trace_ext4_find_delalloc_range(inode,
3496 lblk_start, lblk_end,
3497 search_hint_reverse,
3501 if (search_hint_reverse)
3505 } while ((i >= lblk_start) && (i <= lblk_end) &&
3506 ((bh = bh->b_this_page) != head));
3509 page_cache_release(page);
3511 * Move to next page. 'i' will be the first lblk in the next
3514 if (search_hint_reverse)
3518 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3521 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3522 search_hint_reverse, 0, 0);
3526 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3527 int search_hint_reverse)
3529 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3530 ext4_lblk_t lblk_start, lblk_end;
3531 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3532 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3534 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3535 search_hint_reverse);
3539 * Determines how many complete clusters (out of those specified by the 'map')
3540 * are under delalloc and were reserved quota for.
3541 * This function is called when we are writing out the blocks that were
3542 * originally written with their allocation delayed, but then the space was
3543 * allocated using fallocate() before the delayed allocation could be resolved.
3544 * The cases to look for are:
3545 * ('=' indicated delayed allocated blocks
3546 * '-' indicates non-delayed allocated blocks)
3547 * (a) partial clusters towards beginning and/or end outside of allocated range
3548 * are not delalloc'ed.
3550 * |----c---=|====c====|====c====|===-c----|
3551 * |++++++ allocated ++++++|
3552 * ==> 4 complete clusters in above example
3554 * (b) partial cluster (outside of allocated range) towards either end is
3555 * marked for delayed allocation. In this case, we will exclude that
3558 * |----====c========|========c========|
3559 * |++++++ allocated ++++++|
3560 * ==> 1 complete clusters in above example
3563 * |================c================|
3564 * |++++++ allocated ++++++|
3565 * ==> 0 complete clusters in above example
3567 * The ext4_da_update_reserve_space will be called only if we
3568 * determine here that there were some "entire" clusters that span
3569 * this 'allocated' range.
3570 * In the non-bigalloc case, this function will just end up returning num_blks
3571 * without ever calling ext4_find_delalloc_range.
3574 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3575 unsigned int num_blks)
3577 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3578 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3579 ext4_lblk_t lblk_from, lblk_to, c_offset;
3580 unsigned int allocated_clusters = 0;
3582 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3583 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3585 /* max possible clusters for this allocation */
3586 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3588 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3590 /* Check towards left side */
3591 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3593 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3594 lblk_to = lblk_from + c_offset - 1;
3596 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3597 allocated_clusters--;
3600 /* Now check towards right. */
3601 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3602 if (allocated_clusters && c_offset) {
3603 lblk_from = lblk_start + num_blks;
3604 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3606 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3607 allocated_clusters--;
3610 return allocated_clusters;
3614 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3615 struct ext4_map_blocks *map,
3616 struct ext4_ext_path *path, int flags,
3617 unsigned int allocated, ext4_fsblk_t newblock)
3621 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3623 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3624 "block %llu, max_blocks %u, flags %d, allocated %u",
3625 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3627 ext4_ext_show_leaf(inode, path);
3629 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3632 /* get_block() before submit the IO, split the extent */
3633 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3634 ret = ext4_split_unwritten_extents(handle, inode, map,
3637 * Flag the inode(non aio case) or end_io struct (aio case)
3638 * that this IO needs to conversion to written when IO is
3642 ext4_set_io_unwritten_flag(inode, io);
3644 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3645 if (ext4_should_dioread_nolock(inode))
3646 map->m_flags |= EXT4_MAP_UNINIT;
3649 /* IO end_io complete, convert the filled extent to written */
3650 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3651 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3654 ext4_update_inode_fsync_trans(handle, inode, 1);
3655 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3661 /* buffered IO case */
3663 * repeat fallocate creation request
3664 * we already have an unwritten extent
3666 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3669 /* buffered READ or buffered write_begin() lookup */
3670 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3672 * We have blocks reserved already. We
3673 * return allocated blocks so that delalloc
3674 * won't do block reservation for us. But
3675 * the buffer head will be unmapped so that
3676 * a read from the block returns 0s.
3678 map->m_flags |= EXT4_MAP_UNWRITTEN;
3682 /* buffered write, writepage time, convert*/
3683 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3685 ext4_update_inode_fsync_trans(handle, inode, 1);
3692 map->m_flags |= EXT4_MAP_NEW;
3694 * if we allocated more blocks than requested
3695 * we need to make sure we unmap the extra block
3696 * allocated. The actual needed block will get
3697 * unmapped later when we find the buffer_head marked
3700 if (allocated > map->m_len) {
3701 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3702 newblock + map->m_len,
3703 allocated - map->m_len);
3704 allocated = map->m_len;
3706 map->m_len = allocated;
3709 * If we have done fallocate with the offset that is already
3710 * delayed allocated, we would have block reservation
3711 * and quota reservation done in the delayed write path.
3712 * But fallocate would have already updated quota and block
3713 * count for this offset. So cancel these reservation
3715 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3716 unsigned int reserved_clusters;
3717 reserved_clusters = get_reserved_cluster_alloc(inode,
3718 map->m_lblk, map->m_len);
3719 if (reserved_clusters)
3720 ext4_da_update_reserve_space(inode,
3726 map->m_flags |= EXT4_MAP_MAPPED;
3727 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3728 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3734 if (allocated > map->m_len)
3735 allocated = map->m_len;
3736 ext4_ext_show_leaf(inode, path);
3737 map->m_pblk = newblock;
3738 map->m_len = allocated;
3741 ext4_ext_drop_refs(path);
3744 return err ? err : allocated;
3748 * get_implied_cluster_alloc - check to see if the requested
3749 * allocation (in the map structure) overlaps with a cluster already
3750 * allocated in an extent.
3751 * @sb The filesystem superblock structure
3752 * @map The requested lblk->pblk mapping
3753 * @ex The extent structure which might contain an implied
3754 * cluster allocation
3756 * This function is called by ext4_ext_map_blocks() after we failed to
3757 * find blocks that were already in the inode's extent tree. Hence,
3758 * we know that the beginning of the requested region cannot overlap
3759 * the extent from the inode's extent tree. There are three cases we
3760 * want to catch. The first is this case:
3762 * |--- cluster # N--|
3763 * |--- extent ---| |---- requested region ---|
3766 * The second case that we need to test for is this one:
3768 * |--------- cluster # N ----------------|
3769 * |--- requested region --| |------- extent ----|
3770 * |=======================|
3772 * The third case is when the requested region lies between two extents
3773 * within the same cluster:
3774 * |------------- cluster # N-------------|
3775 * |----- ex -----| |---- ex_right ----|
3776 * |------ requested region ------|
3777 * |================|
3779 * In each of the above cases, we need to set the map->m_pblk and
3780 * map->m_len so it corresponds to the return the extent labelled as
3781 * "|====|" from cluster #N, since it is already in use for data in
3782 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3783 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3784 * as a new "allocated" block region. Otherwise, we will return 0 and
3785 * ext4_ext_map_blocks() will then allocate one or more new clusters
3786 * by calling ext4_mb_new_blocks().
3788 static int get_implied_cluster_alloc(struct super_block *sb,
3789 struct ext4_map_blocks *map,
3790 struct ext4_extent *ex,
3791 struct ext4_ext_path *path)
3793 struct ext4_sb_info *sbi = EXT4_SB(sb);
3794 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3795 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3796 ext4_lblk_t rr_cluster_start, rr_cluster_end;
3797 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3798 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3799 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3801 /* The extent passed in that we are trying to match */
3802 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3803 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3805 /* The requested region passed into ext4_map_blocks() */
3806 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3807 rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3809 if ((rr_cluster_start == ex_cluster_end) ||
3810 (rr_cluster_start == ex_cluster_start)) {
3811 if (rr_cluster_start == ex_cluster_end)
3812 ee_start += ee_len - 1;
3813 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
3814 map->m_len = min(map->m_len,
3815 (unsigned) sbi->s_cluster_ratio - c_offset);
3817 * Check for and handle this case:
3819 * |--------- cluster # N-------------|
3820 * |------- extent ----|
3821 * |--- requested region ---|
3825 if (map->m_lblk < ee_block)
3826 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3829 * Check for the case where there is already another allocated
3830 * block to the right of 'ex' but before the end of the cluster.
3832 * |------------- cluster # N-------------|
3833 * |----- ex -----| |---- ex_right ----|
3834 * |------ requested region ------|
3835 * |================|
3837 if (map->m_lblk > ee_block) {
3838 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3839 map->m_len = min(map->m_len, next - map->m_lblk);
3842 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3846 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3852 * Block allocation/map/preallocation routine for extents based files
3855 * Need to be called with
3856 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3857 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3859 * return > 0, number of of blocks already mapped/allocated
3860 * if create == 0 and these are pre-allocated blocks
3861 * buffer head is unmapped
3862 * otherwise blocks are mapped
3864 * return = 0, if plain look up failed (blocks have not been allocated)
3865 * buffer head is unmapped
3867 * return < 0, error case.
3869 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3870 struct ext4_map_blocks *map, int flags)
3872 struct ext4_ext_path *path = NULL;
3873 struct ext4_extent newex, *ex, *ex2;
3874 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3875 ext4_fsblk_t newblock = 0;
3876 int free_on_err = 0, err = 0, depth, ret;
3877 unsigned int allocated = 0, offset = 0;
3878 unsigned int allocated_clusters = 0;
3879 unsigned int punched_out = 0;
3880 unsigned int result = 0;
3881 struct ext4_allocation_request ar;
3882 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3883 ext4_lblk_t cluster_offset;
3885 ext_debug("blocks %u/%u requested for inode %lu\n",
3886 map->m_lblk, map->m_len, inode->i_ino);
3887 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3889 /* check in cache */
3890 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3891 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3892 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3893 if ((sbi->s_cluster_ratio > 1) &&
3894 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3895 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3897 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3899 * block isn't allocated yet and
3900 * user doesn't want to allocate it
3904 /* we should allocate requested block */
3906 /* block is already allocated */
3907 if (sbi->s_cluster_ratio > 1)
3908 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3909 newblock = map->m_lblk
3910 - le32_to_cpu(newex.ee_block)
3911 + ext4_ext_pblock(&newex);
3912 /* number of remaining blocks in the extent */
3913 allocated = ext4_ext_get_actual_len(&newex) -
3914 (map->m_lblk - le32_to_cpu(newex.ee_block));
3919 /* find extent for this block */
3920 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3922 err = PTR_ERR(path);
3927 depth = ext_depth(inode);
3930 * consistent leaf must not be empty;
3931 * this situation is possible, though, _during_ tree modification;
3932 * this is why assert can't be put in ext4_ext_find_extent()
3934 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3935 EXT4_ERROR_INODE(inode, "bad extent address "
3936 "lblock: %lu, depth: %d pblock %lld",
3937 (unsigned long) map->m_lblk, depth,
3938 path[depth].p_block);
3943 ex = path[depth].p_ext;
3945 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3946 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3947 unsigned short ee_len;
3950 * Uninitialized extents are treated as holes, except that
3951 * we split out initialized portions during a write.
3953 ee_len = ext4_ext_get_actual_len(ex);
3955 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3957 /* if found extent covers block, simply return it */
3958 if (in_range(map->m_lblk, ee_block, ee_len)) {
3959 struct ext4_map_blocks punch_map;
3960 ext4_fsblk_t partial_cluster = 0;
3962 newblock = map->m_lblk - ee_block + ee_start;
3963 /* number of remaining blocks in the extent */
3964 allocated = ee_len - (map->m_lblk - ee_block);
3965 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3966 ee_block, ee_len, newblock);
3968 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3970 * Do not put uninitialized extent
3973 if (!ext4_ext_is_uninitialized(ex)) {
3974 ext4_ext_put_in_cache(inode, ee_block,
3978 ret = ext4_ext_handle_uninitialized_extents(
3979 handle, inode, map, path, flags,
3980 allocated, newblock);
3985 * Punch out the map length, but only to the
3988 punched_out = allocated < map->m_len ?
3989 allocated : map->m_len;
3992 * Sense extents need to be converted to
3993 * uninitialized, they must fit in an
3994 * uninitialized extent
3996 if (punched_out > EXT_UNINIT_MAX_LEN)
3997 punched_out = EXT_UNINIT_MAX_LEN;
3999 punch_map.m_lblk = map->m_lblk;
4000 punch_map.m_pblk = newblock;
4001 punch_map.m_len = punched_out;
4002 punch_map.m_flags = 0;
4004 /* Check to see if the extent needs to be split */
4005 if (punch_map.m_len != ee_len ||
4006 punch_map.m_lblk != ee_block) {
4008 ret = ext4_split_extent(handle, inode,
4009 path, &punch_map, 0,
4010 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
4011 EXT4_GET_BLOCKS_PRE_IO);
4018 * find extent for the block at
4019 * the start of the hole
4021 ext4_ext_drop_refs(path);
4024 path = ext4_ext_find_extent(inode,
4027 err = PTR_ERR(path);
4032 depth = ext_depth(inode);
4033 ex = path[depth].p_ext;
4034 ee_len = ext4_ext_get_actual_len(ex);
4035 ee_block = le32_to_cpu(ex->ee_block);
4036 ee_start = ext4_ext_pblock(ex);
4040 ext4_ext_mark_uninitialized(ex);
4042 ext4_ext_invalidate_cache(inode);
4044 err = ext4_ext_rm_leaf(handle, inode, path,
4045 &partial_cluster, map->m_lblk,
4046 map->m_lblk + punched_out);
4048 if (!err && path->p_hdr->eh_entries == 0) {
4050 * Punch hole freed all of this sub tree,
4051 * so we need to correct eh_depth
4053 err = ext4_ext_get_access(handle, inode, path);
4055 ext_inode_hdr(inode)->eh_depth = 0;
4056 ext_inode_hdr(inode)->eh_max =
4057 cpu_to_le16(ext4_ext_space_root(
4060 err = ext4_ext_dirty(
4061 handle, inode, path);
4069 if ((sbi->s_cluster_ratio > 1) &&
4070 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
4071 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4074 * requested block isn't allocated yet;
4075 * we couldn't try to create block if create flag is zero
4077 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4079 * put just found gap into cache to speed up
4080 * subsequent requests
4082 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4087 * Okay, we need to do block allocation.
4089 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4090 newex.ee_block = cpu_to_le32(map->m_lblk);
4091 cluster_offset = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4094 * If we are doing bigalloc, check to see if the extent returned
4095 * by ext4_ext_find_extent() implies a cluster we can use.
4097 if (cluster_offset && ex &&
4098 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4099 ar.len = allocated = map->m_len;
4100 newblock = map->m_pblk;
4101 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4102 goto got_allocated_blocks;
4105 /* find neighbour allocated blocks */
4106 ar.lleft = map->m_lblk;
4107 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4110 ar.lright = map->m_lblk;
4112 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4116 /* Check if the extent after searching to the right implies a
4117 * cluster we can use. */
4118 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4119 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4120 ar.len = allocated = map->m_len;
4121 newblock = map->m_pblk;
4122 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4123 goto got_allocated_blocks;
4127 * See if request is beyond maximum number of blocks we can have in
4128 * a single extent. For an initialized extent this limit is
4129 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4130 * EXT_UNINIT_MAX_LEN.
4132 if (map->m_len > EXT_INIT_MAX_LEN &&
4133 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4134 map->m_len = EXT_INIT_MAX_LEN;
4135 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4136 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4137 map->m_len = EXT_UNINIT_MAX_LEN;
4139 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4140 newex.ee_len = cpu_to_le16(map->m_len);
4141 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4143 allocated = ext4_ext_get_actual_len(&newex);
4145 allocated = map->m_len;
4147 /* allocate new block */
4149 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4150 ar.logical = map->m_lblk;
4152 * We calculate the offset from the beginning of the cluster
4153 * for the logical block number, since when we allocate a
4154 * physical cluster, the physical block should start at the
4155 * same offset from the beginning of the cluster. This is
4156 * needed so that future calls to get_implied_cluster_alloc()
4159 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4160 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4162 ar.logical -= offset;
4163 if (S_ISREG(inode->i_mode))
4164 ar.flags = EXT4_MB_HINT_DATA;
4166 /* disable in-core preallocation for non-regular files */
4168 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4169 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4170 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4173 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4174 ar.goal, newblock, allocated);
4176 allocated_clusters = ar.len;
4177 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4178 if (ar.len > allocated)
4181 got_allocated_blocks:
4182 /* try to insert new extent into found leaf and return */
4183 ext4_ext_store_pblock(&newex, newblock + offset);
4184 newex.ee_len = cpu_to_le16(ar.len);
4185 /* Mark uninitialized */
4186 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4187 ext4_ext_mark_uninitialized(&newex);
4189 * io_end structure was created for every IO write to an
4190 * uninitialized extent. To avoid unnecessary conversion,
4191 * here we flag the IO that really needs the conversion.
4192 * For non asycn direct IO case, flag the inode state
4193 * that we need to perform conversion when IO is done.
4195 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4197 ext4_set_io_unwritten_flag(inode, io);
4199 ext4_set_inode_state(inode,
4200 EXT4_STATE_DIO_UNWRITTEN);
4202 if (ext4_should_dioread_nolock(inode))
4203 map->m_flags |= EXT4_MAP_UNINIT;
4207 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4208 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4211 err = ext4_ext_insert_extent(handle, inode, path,
4213 if (err && free_on_err) {
4214 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4215 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4216 /* free data blocks we just allocated */
4217 /* not a good idea to call discard here directly,
4218 * but otherwise we'd need to call it every free() */
4219 ext4_discard_preallocations(inode);
4220 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4221 ext4_ext_get_actual_len(&newex), fb_flags);
4225 /* previous routine could use block we allocated */
4226 newblock = ext4_ext_pblock(&newex);
4227 allocated = ext4_ext_get_actual_len(&newex);
4228 if (allocated > map->m_len)
4229 allocated = map->m_len;
4230 map->m_flags |= EXT4_MAP_NEW;
4233 * Update reserved blocks/metadata blocks after successful
4234 * block allocation which had been deferred till now.
4236 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4237 unsigned int reserved_clusters;
4239 * Check how many clusters we had reserved this allocated range
4241 reserved_clusters = get_reserved_cluster_alloc(inode,
4242 map->m_lblk, allocated);
4243 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4244 if (reserved_clusters) {
4246 * We have clusters reserved for this range.
4247 * But since we are not doing actual allocation
4248 * and are simply using blocks from previously
4249 * allocated cluster, we should release the
4250 * reservation and not claim quota.
4252 ext4_da_update_reserve_space(inode,
4253 reserved_clusters, 0);
4256 BUG_ON(allocated_clusters < reserved_clusters);
4257 /* We will claim quota for all newly allocated blocks.*/
4258 ext4_da_update_reserve_space(inode, allocated_clusters,
4260 if (reserved_clusters < allocated_clusters) {
4261 struct ext4_inode_info *ei = EXT4_I(inode);
4262 int reservation = allocated_clusters -
4265 * It seems we claimed few clusters outside of
4266 * the range of this allocation. We should give
4267 * it back to the reservation pool. This can
4268 * happen in the following case:
4270 * * Suppose s_cluster_ratio is 4 (i.e., each
4271 * cluster has 4 blocks. Thus, the clusters
4272 * are [0-3],[4-7],[8-11]...
4273 * * First comes delayed allocation write for
4274 * logical blocks 10 & 11. Since there were no
4275 * previous delayed allocated blocks in the
4276 * range [8-11], we would reserve 1 cluster
4278 * * Next comes write for logical blocks 3 to 8.
4279 * In this case, we will reserve 2 clusters
4280 * (for [0-3] and [4-7]; and not for [8-11] as
4281 * that range has a delayed allocated blocks.
4282 * Thus total reserved clusters now becomes 3.
4283 * * Now, during the delayed allocation writeout
4284 * time, we will first write blocks [3-8] and
4285 * allocate 3 clusters for writing these
4286 * blocks. Also, we would claim all these
4287 * three clusters above.
4288 * * Now when we come here to writeout the
4289 * blocks [10-11], we would expect to claim
4290 * the reservation of 1 cluster we had made
4291 * (and we would claim it since there are no
4292 * more delayed allocated blocks in the range
4293 * [8-11]. But our reserved cluster count had
4294 * already gone to 0.
4296 * Thus, at the step 4 above when we determine
4297 * that there are still some unwritten delayed
4298 * allocated blocks outside of our current
4299 * block range, we should increment the
4300 * reserved clusters count so that when the
4301 * remaining blocks finally gets written, we
4304 dquot_reserve_block(inode,
4305 EXT4_C2B(sbi, reservation));
4306 spin_lock(&ei->i_block_reservation_lock);
4307 ei->i_reserved_data_blocks += reservation;
4308 spin_unlock(&ei->i_block_reservation_lock);
4314 * Cache the extent and update transaction to commit on fdatasync only
4315 * when it is _not_ an uninitialized extent.
4317 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4318 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4319 ext4_update_inode_fsync_trans(handle, inode, 1);
4321 ext4_update_inode_fsync_trans(handle, inode, 0);
4323 if (allocated > map->m_len)
4324 allocated = map->m_len;
4325 ext4_ext_show_leaf(inode, path);
4326 map->m_flags |= EXT4_MAP_MAPPED;
4327 map->m_pblk = newblock;
4328 map->m_len = allocated;
4331 ext4_ext_drop_refs(path);
4334 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4335 punched_out : allocated;
4337 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4338 newblock, map->m_len, err ? err : result);
4340 return err ? err : result;
4343 void ext4_ext_truncate(struct inode *inode)
4345 struct address_space *mapping = inode->i_mapping;
4346 struct super_block *sb = inode->i_sb;
4347 ext4_lblk_t last_block;
4353 * finish any pending end_io work so we won't run the risk of
4354 * converting any truncated blocks to initialized later
4356 ext4_flush_completed_IO(inode);
4359 * probably first extent we're gonna free will be last in block
4361 err = ext4_writepage_trans_blocks(inode);
4362 handle = ext4_journal_start(inode, err);
4366 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4367 page_len = PAGE_CACHE_SIZE -
4368 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4370 err = ext4_discard_partial_page_buffers(handle,
4371 mapping, inode->i_size, page_len, 0);
4377 if (ext4_orphan_add(handle, inode))
4380 down_write(&EXT4_I(inode)->i_data_sem);
4381 ext4_ext_invalidate_cache(inode);
4383 ext4_discard_preallocations(inode);
4386 * TODO: optimization is possible here.
4387 * Probably we need not scan at all,
4388 * because page truncation is enough.
4391 /* we have to know where to truncate from in crash case */
4392 EXT4_I(inode)->i_disksize = inode->i_size;
4393 ext4_mark_inode_dirty(handle, inode);
4395 last_block = (inode->i_size + sb->s_blocksize - 1)
4396 >> EXT4_BLOCK_SIZE_BITS(sb);
4397 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4399 /* In a multi-transaction truncate, we only make the final
4400 * transaction synchronous.
4403 ext4_handle_sync(handle);
4405 up_write(&EXT4_I(inode)->i_data_sem);
4409 * If this was a simple ftruncate() and the file will remain alive,
4410 * then we need to clear up the orphan record which we created above.
4411 * However, if this was a real unlink then we were called by
4412 * ext4_delete_inode(), and we allow that function to clean up the
4413 * orphan info for us.
4416 ext4_orphan_del(handle, inode);
4418 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4419 ext4_mark_inode_dirty(handle, inode);
4420 ext4_journal_stop(handle);
4423 static void ext4_falloc_update_inode(struct inode *inode,
4424 int mode, loff_t new_size, int update_ctime)
4426 struct timespec now;
4429 now = current_fs_time(inode->i_sb);
4430 if (!timespec_equal(&inode->i_ctime, &now))
4431 inode->i_ctime = now;
4434 * Update only when preallocation was requested beyond
4437 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4438 if (new_size > i_size_read(inode))
4439 i_size_write(inode, new_size);
4440 if (new_size > EXT4_I(inode)->i_disksize)
4441 ext4_update_i_disksize(inode, new_size);
4444 * Mark that we allocate beyond EOF so the subsequent truncate
4445 * can proceed even if the new size is the same as i_size.
4447 if (new_size > i_size_read(inode))
4448 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4454 * preallocate space for a file. This implements ext4's fallocate file
4455 * operation, which gets called from sys_fallocate system call.
4456 * For block-mapped files, posix_fallocate should fall back to the method
4457 * of writing zeroes to the required new blocks (the same behavior which is
4458 * expected for file systems which do not support fallocate() system call).
4460 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4462 struct inode *inode = file->f_path.dentry->d_inode;
4465 unsigned int max_blocks;
4470 struct ext4_map_blocks map;
4471 unsigned int credits, blkbits = inode->i_blkbits;
4473 /* Return error if mode is not supported */
4474 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4477 if (mode & FALLOC_FL_PUNCH_HOLE)
4478 return ext4_punch_hole(file, offset, len);
4480 trace_ext4_fallocate_enter(inode, offset, len, mode);
4481 map.m_lblk = offset >> blkbits;
4483 * We can't just convert len to max_blocks because
4484 * If blocksize = 4096 offset = 3072 and len = 2048
4486 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4489 * credits to insert 1 extent into extent tree
4491 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4492 mutex_lock(&inode->i_mutex);
4495 * We only support preallocation for extent-based files only
4497 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4502 ret = inode_newsize_ok(inode, (len + offset));
4504 mutex_unlock(&inode->i_mutex);
4505 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4508 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4509 if (mode & FALLOC_FL_KEEP_SIZE)
4510 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4512 * Don't normalize the request if it can fit in one extent so
4513 * that it doesn't get unnecessarily split into multiple
4516 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4517 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4519 while (ret >= 0 && ret < max_blocks) {
4520 map.m_lblk = map.m_lblk + ret;
4521 map.m_len = max_blocks = max_blocks - ret;
4522 handle = ext4_journal_start(inode, credits);
4523 if (IS_ERR(handle)) {
4524 ret = PTR_ERR(handle);
4527 ret = ext4_map_blocks(handle, inode, &map, flags);
4531 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4532 "returned error inode#%lu, block=%u, "
4533 "max_blocks=%u", __func__,
4534 inode->i_ino, map.m_lblk, max_blocks);
4536 ext4_mark_inode_dirty(handle, inode);
4537 ret2 = ext4_journal_stop(handle);
4540 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4541 blkbits) >> blkbits))
4542 new_size = offset + len;
4544 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4546 ext4_falloc_update_inode(inode, mode, new_size,
4547 (map.m_flags & EXT4_MAP_NEW));
4548 ext4_mark_inode_dirty(handle, inode);
4549 ret2 = ext4_journal_stop(handle);
4553 if (ret == -ENOSPC &&
4554 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4559 mutex_unlock(&inode->i_mutex);
4560 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4561 ret > 0 ? ret2 : ret);
4562 return ret > 0 ? ret2 : ret;
4566 * This function convert a range of blocks to written extents
4567 * The caller of this function will pass the start offset and the size.
4568 * all unwritten extents within this range will be converted to
4571 * This function is called from the direct IO end io call back
4572 * function, to convert the fallocated extents after IO is completed.
4573 * Returns 0 on success.
4575 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4579 unsigned int max_blocks;
4582 struct ext4_map_blocks map;
4583 unsigned int credits, blkbits = inode->i_blkbits;
4585 map.m_lblk = offset >> blkbits;
4587 * We can't just convert len to max_blocks because
4588 * If blocksize = 4096 offset = 3072 and len = 2048
4590 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4593 * credits to insert 1 extent into extent tree
4595 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4596 while (ret >= 0 && ret < max_blocks) {
4598 map.m_len = (max_blocks -= ret);
4599 handle = ext4_journal_start(inode, credits);
4600 if (IS_ERR(handle)) {
4601 ret = PTR_ERR(handle);
4604 ret = ext4_map_blocks(handle, inode, &map,
4605 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4608 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4609 "returned error inode#%lu, block=%u, "
4610 "max_blocks=%u", __func__,
4611 inode->i_ino, map.m_lblk, map.m_len);
4613 ext4_mark_inode_dirty(handle, inode);
4614 ret2 = ext4_journal_stop(handle);
4615 if (ret <= 0 || ret2 )
4618 return ret > 0 ? ret2 : ret;
4622 * Callback function called for each extent to gather FIEMAP information.
4624 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4625 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4633 struct fiemap_extent_info *fieinfo = data;
4634 unsigned char blksize_bits;
4636 blksize_bits = inode->i_sb->s_blocksize_bits;
4637 logical = (__u64)newex->ec_block << blksize_bits;
4639 if (newex->ec_start == 0) {
4641 * No extent in extent-tree contains block @newex->ec_start,
4642 * then the block may stay in 1)a hole or 2)delayed-extent.
4644 * Holes or delayed-extents are processed as follows.
4645 * 1. lookup dirty pages with specified range in pagecache.
4646 * If no page is got, then there is no delayed-extent and
4647 * return with EXT_CONTINUE.
4648 * 2. find the 1st mapped buffer,
4649 * 3. check if the mapped buffer is both in the request range
4650 * and a delayed buffer. If not, there is no delayed-extent,
4652 * 4. a delayed-extent is found, the extent will be collected.
4654 ext4_lblk_t end = 0;
4655 pgoff_t last_offset;
4658 pgoff_t start_index = 0;
4659 struct page **pages = NULL;
4660 struct buffer_head *bh = NULL;
4661 struct buffer_head *head = NULL;
4662 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4664 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4668 offset = logical >> PAGE_SHIFT;
4670 last_offset = offset;
4672 ret = find_get_pages_tag(inode->i_mapping, &offset,
4673 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4675 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4676 /* First time, try to find a mapped buffer. */
4679 for (index = 0; index < ret; index++)
4680 page_cache_release(pages[index]);
4683 return EXT_CONTINUE;
4688 /* Try to find the 1st mapped buffer. */
4689 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4691 if (!page_has_buffers(pages[index]))
4693 head = page_buffers(pages[index]);
4700 if (end >= newex->ec_block +
4702 /* The buffer is out of
4703 * the request range.
4707 if (buffer_mapped(bh) &&
4708 end >= newex->ec_block) {
4709 start_index = index - 1;
4710 /* get the 1st mapped buffer. */
4711 goto found_mapped_buffer;
4714 bh = bh->b_this_page;
4716 } while (bh != head);
4718 /* No mapped buffer in the range found in this page,
4719 * We need to look up next page.
4722 /* There is no page left, but we need to limit
4725 newex->ec_len = end - newex->ec_block;
4730 /*Find contiguous delayed buffers. */
4731 if (ret > 0 && pages[0]->index == last_offset)
4732 head = page_buffers(pages[0]);
4738 found_mapped_buffer:
4739 if (bh != NULL && buffer_delay(bh)) {
4740 /* 1st or contiguous delayed buffer found. */
4741 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4743 * 1st delayed buffer found, record
4744 * the start of extent.
4746 flags |= FIEMAP_EXTENT_DELALLOC;
4747 newex->ec_block = end;
4748 logical = (__u64)end << blksize_bits;
4750 /* Find contiguous delayed buffers. */
4752 if (!buffer_delay(bh))
4753 goto found_delayed_extent;
4754 bh = bh->b_this_page;
4756 } while (bh != head);
4758 for (; index < ret; index++) {
4759 if (!page_has_buffers(pages[index])) {
4763 head = page_buffers(pages[index]);
4769 if (pages[index]->index !=
4770 pages[start_index]->index + index
4772 /* Blocks are not contiguous. */
4778 if (!buffer_delay(bh))
4779 /* Delayed-extent ends. */
4780 goto found_delayed_extent;
4781 bh = bh->b_this_page;
4783 } while (bh != head);
4785 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4789 found_delayed_extent:
4790 newex->ec_len = min(end - newex->ec_block,
4791 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4792 if (ret == nr_pages && bh != NULL &&
4793 newex->ec_len < EXT_INIT_MAX_LEN &&
4795 /* Have not collected an extent and continue. */
4796 for (index = 0; index < ret; index++)
4797 page_cache_release(pages[index]);
4801 for (index = 0; index < ret; index++)
4802 page_cache_release(pages[index]);
4806 physical = (__u64)newex->ec_start << blksize_bits;
4807 length = (__u64)newex->ec_len << blksize_bits;
4809 if (ex && ext4_ext_is_uninitialized(ex))
4810 flags |= FIEMAP_EXTENT_UNWRITTEN;
4812 if (next == EXT_MAX_BLOCKS)
4813 flags |= FIEMAP_EXTENT_LAST;
4815 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4821 return EXT_CONTINUE;
4823 /* fiemap flags we can handle specified here */
4824 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4826 static int ext4_xattr_fiemap(struct inode *inode,
4827 struct fiemap_extent_info *fieinfo)
4831 __u32 flags = FIEMAP_EXTENT_LAST;
4832 int blockbits = inode->i_sb->s_blocksize_bits;
4836 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4837 struct ext4_iloc iloc;
4838 int offset; /* offset of xattr in inode */
4840 error = ext4_get_inode_loc(inode, &iloc);
4843 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4844 offset = EXT4_GOOD_OLD_INODE_SIZE +
4845 EXT4_I(inode)->i_extra_isize;
4847 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4848 flags |= FIEMAP_EXTENT_DATA_INLINE;
4850 } else { /* external block */
4851 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4852 length = inode->i_sb->s_blocksize;
4856 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4858 return (error < 0 ? error : 0);
4862 * ext4_ext_punch_hole
4864 * Punches a hole of "length" bytes in a file starting
4867 * @inode: The inode of the file to punch a hole in
4868 * @offset: The starting byte offset of the hole
4869 * @length: The length of the hole
4871 * Returns the number of blocks removed or negative on err
4873 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4875 struct inode *inode = file->f_path.dentry->d_inode;
4876 struct super_block *sb = inode->i_sb;
4877 ext4_lblk_t first_block, stop_block;
4878 struct address_space *mapping = inode->i_mapping;
4880 loff_t first_page, last_page, page_len;
4881 loff_t first_page_offset, last_page_offset;
4882 int credits, err = 0;
4884 /* No need to punch hole beyond i_size */
4885 if (offset >= inode->i_size)
4889 * If the hole extends beyond i_size, set the hole
4890 * to end after the page that contains i_size
4892 if (offset + length > inode->i_size) {
4893 length = inode->i_size +
4894 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4898 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4899 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4901 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4902 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4905 * Write out all dirty pages to avoid race conditions
4906 * Then release them.
4908 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4909 err = filemap_write_and_wait_range(mapping,
4910 offset, offset + length - 1);
4916 /* Now release the pages */
4917 if (last_page_offset > first_page_offset) {
4918 truncate_inode_pages_range(mapping, first_page_offset,
4919 last_page_offset-1);
4922 /* finish any pending end_io work */
4923 ext4_flush_completed_IO(inode);
4925 credits = ext4_writepage_trans_blocks(inode);
4926 handle = ext4_journal_start(inode, credits);
4928 return PTR_ERR(handle);
4930 err = ext4_orphan_add(handle, inode);
4935 * Now we need to zero out the non-page-aligned data in the
4936 * pages at the start and tail of the hole, and unmap the buffer
4937 * heads for the block aligned regions of the page that were
4938 * completely zeroed.
4940 if (first_page > last_page) {
4942 * If the file space being truncated is contained within a page
4943 * just zero out and unmap the middle of that page
4945 err = ext4_discard_partial_page_buffers(handle,
4946 mapping, offset, length, 0);
4952 * zero out and unmap the partial page that contains
4953 * the start of the hole
4955 page_len = first_page_offset - offset;
4957 err = ext4_discard_partial_page_buffers(handle, mapping,
4958 offset, page_len, 0);
4964 * zero out and unmap the partial page that contains
4965 * the end of the hole
4967 page_len = offset + length - last_page_offset;
4969 err = ext4_discard_partial_page_buffers(handle, mapping,
4970 last_page_offset, page_len, 0);
4977 * If i_size is contained in the last page, we need to
4978 * unmap and zero the partial page after i_size
4980 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4981 inode->i_size % PAGE_CACHE_SIZE != 0) {
4983 page_len = PAGE_CACHE_SIZE -
4984 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4987 err = ext4_discard_partial_page_buffers(handle,
4988 mapping, inode->i_size, page_len, 0);
4995 first_block = (offset + sb->s_blocksize - 1) >>
4996 EXT4_BLOCK_SIZE_BITS(sb);
4997 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4999 /* If there are no blocks to remove, return now */
5000 if (first_block >= stop_block)
5003 down_write(&EXT4_I(inode)->i_data_sem);
5004 ext4_ext_invalidate_cache(inode);
5005 ext4_discard_preallocations(inode);
5007 err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
5009 ext4_ext_invalidate_cache(inode);
5010 ext4_discard_preallocations(inode);
5013 ext4_handle_sync(handle);
5015 up_write(&EXT4_I(inode)->i_data_sem);
5018 ext4_orphan_del(handle, inode);
5019 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5020 ext4_mark_inode_dirty(handle, inode);
5021 ext4_journal_stop(handle);
5024 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5025 __u64 start, __u64 len)
5027 ext4_lblk_t start_blk;
5030 /* fallback to generic here if not in extents fmt */
5031 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5032 return generic_block_fiemap(inode, fieinfo, start, len,
5035 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5038 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5039 error = ext4_xattr_fiemap(inode, fieinfo);
5041 ext4_lblk_t len_blks;
5044 start_blk = start >> inode->i_sb->s_blocksize_bits;
5045 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5046 if (last_blk >= EXT_MAX_BLOCKS)
5047 last_blk = EXT_MAX_BLOCKS-1;
5048 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5051 * Walk the extent tree gathering extent information.
5052 * ext4_ext_fiemap_cb will push extents back to user.
5054 error = ext4_ext_walk_space(inode, start_blk, len_blks,
5055 ext4_ext_fiemap_cb, fieinfo);