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>
48 static int ext4_split_extent(handle_t *handle,
50 struct ext4_ext_path *path,
51 struct ext4_map_blocks *map,
55 static int ext4_ext_truncate_extend_restart(handle_t *handle,
61 if (!ext4_handle_valid(handle))
63 if (handle->h_buffer_credits > needed)
65 err = ext4_journal_extend(handle, needed);
68 err = ext4_truncate_restart_trans(handle, inode, needed);
80 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
81 struct ext4_ext_path *path)
84 /* path points to block */
85 return ext4_journal_get_write_access(handle, path->p_bh);
87 /* path points to leaf/index in inode body */
88 /* we use in-core data, no need to protect them */
98 #define ext4_ext_dirty(handle, inode, path) \
99 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
100 static int __ext4_ext_dirty(const char *where, unsigned int line,
101 handle_t *handle, struct inode *inode,
102 struct ext4_ext_path *path)
106 /* path points to block */
107 err = __ext4_handle_dirty_metadata(where, line, handle,
110 /* path points to leaf/index in inode body */
111 err = ext4_mark_inode_dirty(handle, inode);
116 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
117 struct ext4_ext_path *path,
123 struct ext4_extent *ex;
124 depth = path->p_depth;
127 * Try to predict block placement assuming that we are
128 * filling in a file which will eventually be
129 * non-sparse --- i.e., in the case of libbfd writing
130 * an ELF object sections out-of-order but in a way
131 * the eventually results in a contiguous object or
132 * executable file, or some database extending a table
133 * space file. However, this is actually somewhat
134 * non-ideal if we are writing a sparse file such as
135 * qemu or KVM writing a raw image file that is going
136 * to stay fairly sparse, since it will end up
137 * fragmenting the file system's free space. Maybe we
138 * should have some hueristics or some way to allow
139 * userspace to pass a hint to file system,
140 * especially if the latter case turns out to be
143 ex = path[depth].p_ext;
145 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
146 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
148 if (block > ext_block)
149 return ext_pblk + (block - ext_block);
151 return ext_pblk - (ext_block - block);
154 /* it looks like index is empty;
155 * try to find starting block from index itself */
156 if (path[depth].p_bh)
157 return path[depth].p_bh->b_blocknr;
160 /* OK. use inode's group */
161 return ext4_inode_to_goal_block(inode);
165 * Allocation for a meta data block
168 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
169 struct ext4_ext_path *path,
170 struct ext4_extent *ex, int *err, unsigned int flags)
172 ext4_fsblk_t goal, newblock;
174 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
175 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
180 static inline int ext4_ext_space_block(struct inode *inode, int check)
184 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
185 / sizeof(struct ext4_extent);
187 #ifdef AGGRESSIVE_TEST
195 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
199 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
200 / sizeof(struct ext4_extent_idx);
202 #ifdef AGGRESSIVE_TEST
210 static inline int ext4_ext_space_root(struct inode *inode, int check)
214 size = sizeof(EXT4_I(inode)->i_data);
215 size -= sizeof(struct ext4_extent_header);
216 size /= sizeof(struct ext4_extent);
218 #ifdef AGGRESSIVE_TEST
226 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
230 size = sizeof(EXT4_I(inode)->i_data);
231 size -= sizeof(struct ext4_extent_header);
232 size /= sizeof(struct ext4_extent_idx);
234 #ifdef AGGRESSIVE_TEST
243 * Calculate the number of metadata blocks needed
244 * to allocate @blocks
245 * Worse case is one block per extent
247 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
249 struct ext4_inode_info *ei = EXT4_I(inode);
252 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
253 / sizeof(struct ext4_extent_idx));
256 * If the new delayed allocation block is contiguous with the
257 * previous da block, it can share index blocks with the
258 * previous block, so we only need to allocate a new index
259 * block every idxs leaf blocks. At ldxs**2 blocks, we need
260 * an additional index block, and at ldxs**3 blocks, yet
261 * another index blocks.
263 if (ei->i_da_metadata_calc_len &&
264 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
265 if ((ei->i_da_metadata_calc_len % idxs) == 0)
267 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
269 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
271 ei->i_da_metadata_calc_len = 0;
273 ei->i_da_metadata_calc_len++;
274 ei->i_da_metadata_calc_last_lblock++;
279 * In the worst case we need a new set of index blocks at
280 * every level of the inode's extent tree.
282 ei->i_da_metadata_calc_len = 1;
283 ei->i_da_metadata_calc_last_lblock = lblock;
284 return ext_depth(inode) + 1;
288 ext4_ext_max_entries(struct inode *inode, int depth)
292 if (depth == ext_depth(inode)) {
294 max = ext4_ext_space_root(inode, 1);
296 max = ext4_ext_space_root_idx(inode, 1);
299 max = ext4_ext_space_block(inode, 1);
301 max = ext4_ext_space_block_idx(inode, 1);
307 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
309 ext4_fsblk_t block = ext4_ext_pblock(ext);
310 int len = ext4_ext_get_actual_len(ext);
312 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
315 static int ext4_valid_extent_idx(struct inode *inode,
316 struct ext4_extent_idx *ext_idx)
318 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
320 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
323 static int ext4_valid_extent_entries(struct inode *inode,
324 struct ext4_extent_header *eh,
327 struct ext4_extent *ext;
328 struct ext4_extent_idx *ext_idx;
329 unsigned short entries;
330 if (eh->eh_entries == 0)
333 entries = le16_to_cpu(eh->eh_entries);
337 ext = EXT_FIRST_EXTENT(eh);
339 if (!ext4_valid_extent(inode, ext))
345 ext_idx = EXT_FIRST_INDEX(eh);
347 if (!ext4_valid_extent_idx(inode, ext_idx))
356 static int __ext4_ext_check(const char *function, unsigned int line,
357 struct inode *inode, struct ext4_extent_header *eh,
360 const char *error_msg;
363 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
364 error_msg = "invalid magic";
367 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
368 error_msg = "unexpected eh_depth";
371 if (unlikely(eh->eh_max == 0)) {
372 error_msg = "invalid eh_max";
375 max = ext4_ext_max_entries(inode, depth);
376 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
377 error_msg = "too large eh_max";
380 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
381 error_msg = "invalid eh_entries";
384 if (!ext4_valid_extent_entries(inode, eh, depth)) {
385 error_msg = "invalid extent entries";
391 ext4_error_inode(inode, function, line, 0,
392 "bad header/extent: %s - magic %x, "
393 "entries %u, max %u(%u), depth %u(%u)",
394 error_msg, le16_to_cpu(eh->eh_magic),
395 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
396 max, le16_to_cpu(eh->eh_depth), depth);
401 #define ext4_ext_check(inode, eh, depth) \
402 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
404 int ext4_ext_check_inode(struct inode *inode)
406 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
410 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
412 int k, l = path->p_depth;
415 for (k = 0; k <= l; k++, path++) {
417 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
418 ext4_idx_pblock(path->p_idx));
419 } else if (path->p_ext) {
420 ext_debug(" %d:[%d]%d:%llu ",
421 le32_to_cpu(path->p_ext->ee_block),
422 ext4_ext_is_uninitialized(path->p_ext),
423 ext4_ext_get_actual_len(path->p_ext),
424 ext4_ext_pblock(path->p_ext));
431 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
433 int depth = ext_depth(inode);
434 struct ext4_extent_header *eh;
435 struct ext4_extent *ex;
441 eh = path[depth].p_hdr;
442 ex = EXT_FIRST_EXTENT(eh);
444 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
446 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
447 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
448 ext4_ext_is_uninitialized(ex),
449 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
454 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
455 ext4_fsblk_t newblock, int level)
457 int depth = ext_depth(inode);
458 struct ext4_extent *ex;
460 if (depth != level) {
461 struct ext4_extent_idx *idx;
462 idx = path[level].p_idx;
463 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
464 ext_debug("%d: move %d:%llu in new index %llu\n", level,
465 le32_to_cpu(idx->ei_block),
466 ext4_idx_pblock(idx),
474 ex = path[depth].p_ext;
475 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
476 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
477 le32_to_cpu(ex->ee_block),
479 ext4_ext_is_uninitialized(ex),
480 ext4_ext_get_actual_len(ex),
487 #define ext4_ext_show_path(inode, path)
488 #define ext4_ext_show_leaf(inode, path)
489 #define ext4_ext_show_move(inode, path, newblock, level)
492 void ext4_ext_drop_refs(struct ext4_ext_path *path)
494 int depth = path->p_depth;
497 for (i = 0; i <= depth; i++, path++)
505 * ext4_ext_binsearch_idx:
506 * binary search for the closest index of the given block
507 * the header must be checked before calling this
510 ext4_ext_binsearch_idx(struct inode *inode,
511 struct ext4_ext_path *path, ext4_lblk_t block)
513 struct ext4_extent_header *eh = path->p_hdr;
514 struct ext4_extent_idx *r, *l, *m;
517 ext_debug("binsearch for %u(idx): ", block);
519 l = EXT_FIRST_INDEX(eh) + 1;
520 r = EXT_LAST_INDEX(eh);
523 if (block < le32_to_cpu(m->ei_block))
527 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
528 m, le32_to_cpu(m->ei_block),
529 r, le32_to_cpu(r->ei_block));
533 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
534 ext4_idx_pblock(path->p_idx));
536 #ifdef CHECK_BINSEARCH
538 struct ext4_extent_idx *chix, *ix;
541 chix = ix = EXT_FIRST_INDEX(eh);
542 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
544 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
545 printk(KERN_DEBUG "k=%d, ix=0x%p, "
547 ix, EXT_FIRST_INDEX(eh));
548 printk(KERN_DEBUG "%u <= %u\n",
549 le32_to_cpu(ix->ei_block),
550 le32_to_cpu(ix[-1].ei_block));
552 BUG_ON(k && le32_to_cpu(ix->ei_block)
553 <= le32_to_cpu(ix[-1].ei_block));
554 if (block < le32_to_cpu(ix->ei_block))
558 BUG_ON(chix != path->p_idx);
565 * ext4_ext_binsearch:
566 * binary search for closest extent of the given block
567 * the header must be checked before calling this
570 ext4_ext_binsearch(struct inode *inode,
571 struct ext4_ext_path *path, ext4_lblk_t block)
573 struct ext4_extent_header *eh = path->p_hdr;
574 struct ext4_extent *r, *l, *m;
576 if (eh->eh_entries == 0) {
578 * this leaf is empty:
579 * we get such a leaf in split/add case
584 ext_debug("binsearch for %u: ", block);
586 l = EXT_FIRST_EXTENT(eh) + 1;
587 r = EXT_LAST_EXTENT(eh);
591 if (block < le32_to_cpu(m->ee_block))
595 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
596 m, le32_to_cpu(m->ee_block),
597 r, le32_to_cpu(r->ee_block));
601 ext_debug(" -> %d:%llu:[%d]%d ",
602 le32_to_cpu(path->p_ext->ee_block),
603 ext4_ext_pblock(path->p_ext),
604 ext4_ext_is_uninitialized(path->p_ext),
605 ext4_ext_get_actual_len(path->p_ext));
607 #ifdef CHECK_BINSEARCH
609 struct ext4_extent *chex, *ex;
612 chex = ex = EXT_FIRST_EXTENT(eh);
613 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
614 BUG_ON(k && le32_to_cpu(ex->ee_block)
615 <= le32_to_cpu(ex[-1].ee_block));
616 if (block < le32_to_cpu(ex->ee_block))
620 BUG_ON(chex != path->p_ext);
626 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
628 struct ext4_extent_header *eh;
630 eh = ext_inode_hdr(inode);
633 eh->eh_magic = EXT4_EXT_MAGIC;
634 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
635 ext4_mark_inode_dirty(handle, inode);
636 ext4_ext_invalidate_cache(inode);
640 struct ext4_ext_path *
641 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
642 struct ext4_ext_path *path)
644 struct ext4_extent_header *eh;
645 struct buffer_head *bh;
646 short int depth, i, ppos = 0, alloc = 0;
648 eh = ext_inode_hdr(inode);
649 depth = ext_depth(inode);
651 /* account possible depth increase */
653 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
656 return ERR_PTR(-ENOMEM);
663 /* walk through the tree */
665 int need_to_validate = 0;
667 ext_debug("depth %d: num %d, max %d\n",
668 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
670 ext4_ext_binsearch_idx(inode, path + ppos, block);
671 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
672 path[ppos].p_depth = i;
673 path[ppos].p_ext = NULL;
675 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
678 if (!bh_uptodate_or_lock(bh)) {
679 trace_ext4_ext_load_extent(inode, block,
681 if (bh_submit_read(bh) < 0) {
685 /* validate the extent entries */
686 need_to_validate = 1;
688 eh = ext_block_hdr(bh);
690 if (unlikely(ppos > depth)) {
692 EXT4_ERROR_INODE(inode,
693 "ppos %d > depth %d", ppos, depth);
696 path[ppos].p_bh = bh;
697 path[ppos].p_hdr = eh;
700 if (need_to_validate && ext4_ext_check(inode, eh, i))
704 path[ppos].p_depth = i;
705 path[ppos].p_ext = NULL;
706 path[ppos].p_idx = NULL;
709 ext4_ext_binsearch(inode, path + ppos, block);
710 /* if not an empty leaf */
711 if (path[ppos].p_ext)
712 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
714 ext4_ext_show_path(inode, path);
719 ext4_ext_drop_refs(path);
722 return ERR_PTR(-EIO);
726 * ext4_ext_insert_index:
727 * insert new index [@logical;@ptr] into the block at @curp;
728 * check where to insert: before @curp or after @curp
730 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
731 struct ext4_ext_path *curp,
732 int logical, ext4_fsblk_t ptr)
734 struct ext4_extent_idx *ix;
737 err = ext4_ext_get_access(handle, inode, curp);
741 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
742 EXT4_ERROR_INODE(inode,
743 "logical %d == ei_block %d!",
744 logical, le32_to_cpu(curp->p_idx->ei_block));
748 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
749 >= le16_to_cpu(curp->p_hdr->eh_max))) {
750 EXT4_ERROR_INODE(inode,
751 "eh_entries %d >= eh_max %d!",
752 le16_to_cpu(curp->p_hdr->eh_entries),
753 le16_to_cpu(curp->p_hdr->eh_max));
757 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
758 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
760 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
761 len = (len - 1) * sizeof(struct ext4_extent_idx);
762 len = len < 0 ? 0 : len;
763 ext_debug("insert new index %d after: %llu. "
764 "move %d from 0x%p to 0x%p\n",
766 (curp->p_idx + 1), (curp->p_idx + 2));
767 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
769 ix = curp->p_idx + 1;
772 len = len * sizeof(struct ext4_extent_idx);
773 len = len < 0 ? 0 : len;
774 ext_debug("insert new index %d before: %llu. "
775 "move %d from 0x%p to 0x%p\n",
777 curp->p_idx, (curp->p_idx + 1));
778 memmove(curp->p_idx + 1, curp->p_idx, len);
782 ix->ei_block = cpu_to_le32(logical);
783 ext4_idx_store_pblock(ix, ptr);
784 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
786 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
787 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
791 err = ext4_ext_dirty(handle, inode, curp);
792 ext4_std_error(inode->i_sb, err);
799 * inserts new subtree into the path, using free index entry
801 * - allocates all needed blocks (new leaf and all intermediate index blocks)
802 * - makes decision where to split
803 * - moves remaining extents and index entries (right to the split point)
804 * into the newly allocated blocks
805 * - initializes subtree
807 static int ext4_ext_split(handle_t *handle, struct inode *inode,
809 struct ext4_ext_path *path,
810 struct ext4_extent *newext, int at)
812 struct buffer_head *bh = NULL;
813 int depth = ext_depth(inode);
814 struct ext4_extent_header *neh;
815 struct ext4_extent_idx *fidx;
817 ext4_fsblk_t newblock, oldblock;
819 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
822 /* make decision: where to split? */
823 /* FIXME: now decision is simplest: at current extent */
825 /* if current leaf will be split, then we should use
826 * border from split point */
827 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
828 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
831 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
832 border = path[depth].p_ext[1].ee_block;
833 ext_debug("leaf will be split."
834 " next leaf starts at %d\n",
835 le32_to_cpu(border));
837 border = newext->ee_block;
838 ext_debug("leaf will be added."
839 " next leaf starts at %d\n",
840 le32_to_cpu(border));
844 * If error occurs, then we break processing
845 * and mark filesystem read-only. index won't
846 * be inserted and tree will be in consistent
847 * state. Next mount will repair buffers too.
851 * Get array to track all allocated blocks.
852 * We need this to handle errors and free blocks
855 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
859 /* allocate all needed blocks */
860 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
861 for (a = 0; a < depth - at; a++) {
862 newblock = ext4_ext_new_meta_block(handle, inode, path,
863 newext, &err, flags);
866 ablocks[a] = newblock;
869 /* initialize new leaf */
870 newblock = ablocks[--a];
871 if (unlikely(newblock == 0)) {
872 EXT4_ERROR_INODE(inode, "newblock == 0!");
876 bh = sb_getblk(inode->i_sb, newblock);
883 err = ext4_journal_get_create_access(handle, bh);
887 neh = ext_block_hdr(bh);
889 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
890 neh->eh_magic = EXT4_EXT_MAGIC;
893 /* move remainder of path[depth] to the new leaf */
894 if (unlikely(path[depth].p_hdr->eh_entries !=
895 path[depth].p_hdr->eh_max)) {
896 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
897 path[depth].p_hdr->eh_entries,
898 path[depth].p_hdr->eh_max);
902 /* start copy from next extent */
903 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
904 ext4_ext_show_move(inode, path, newblock, depth);
906 struct ext4_extent *ex;
907 ex = EXT_FIRST_EXTENT(neh);
908 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
909 le16_add_cpu(&neh->eh_entries, m);
912 set_buffer_uptodate(bh);
915 err = ext4_handle_dirty_metadata(handle, inode, bh);
921 /* correct old leaf */
923 err = ext4_ext_get_access(handle, inode, path + depth);
926 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
927 err = ext4_ext_dirty(handle, inode, path + depth);
933 /* create intermediate indexes */
935 if (unlikely(k < 0)) {
936 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
941 ext_debug("create %d intermediate indices\n", k);
942 /* insert new index into current index block */
943 /* current depth stored in i var */
947 newblock = ablocks[--a];
948 bh = sb_getblk(inode->i_sb, newblock);
955 err = ext4_journal_get_create_access(handle, bh);
959 neh = ext_block_hdr(bh);
960 neh->eh_entries = cpu_to_le16(1);
961 neh->eh_magic = EXT4_EXT_MAGIC;
962 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
963 neh->eh_depth = cpu_to_le16(depth - i);
964 fidx = EXT_FIRST_INDEX(neh);
965 fidx->ei_block = border;
966 ext4_idx_store_pblock(fidx, oldblock);
968 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
969 i, newblock, le32_to_cpu(border), oldblock);
971 /* move remainder of path[i] to the new index block */
972 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
973 EXT_LAST_INDEX(path[i].p_hdr))) {
974 EXT4_ERROR_INODE(inode,
975 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
976 le32_to_cpu(path[i].p_ext->ee_block));
980 /* start copy indexes */
981 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
982 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
983 EXT_MAX_INDEX(path[i].p_hdr));
984 ext4_ext_show_move(inode, path, newblock, i);
986 memmove(++fidx, path[i].p_idx,
987 sizeof(struct ext4_extent_idx) * m);
988 le16_add_cpu(&neh->eh_entries, m);
990 set_buffer_uptodate(bh);
993 err = ext4_handle_dirty_metadata(handle, inode, bh);
999 /* correct old index */
1001 err = ext4_ext_get_access(handle, inode, path + i);
1004 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1005 err = ext4_ext_dirty(handle, inode, path + i);
1013 /* insert new index */
1014 err = ext4_ext_insert_index(handle, inode, path + at,
1015 le32_to_cpu(border), newblock);
1019 if (buffer_locked(bh))
1025 /* free all allocated blocks in error case */
1026 for (i = 0; i < depth; i++) {
1029 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1030 EXT4_FREE_BLOCKS_METADATA);
1039 * ext4_ext_grow_indepth:
1040 * implements tree growing procedure:
1041 * - allocates new block
1042 * - moves top-level data (index block or leaf) into the new block
1043 * - initializes new top-level, creating index that points to the
1044 * just created block
1046 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1048 struct ext4_ext_path *path,
1049 struct ext4_extent *newext)
1051 struct ext4_ext_path *curp = path;
1052 struct ext4_extent_header *neh;
1053 struct buffer_head *bh;
1054 ext4_fsblk_t newblock;
1057 newblock = ext4_ext_new_meta_block(handle, inode, path,
1058 newext, &err, flags);
1062 bh = sb_getblk(inode->i_sb, newblock);
1065 ext4_std_error(inode->i_sb, err);
1070 err = ext4_journal_get_create_access(handle, bh);
1076 /* move top-level index/leaf into new block */
1077 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1079 /* set size of new block */
1080 neh = ext_block_hdr(bh);
1081 /* old root could have indexes or leaves
1082 * so calculate e_max right way */
1083 if (ext_depth(inode))
1084 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1086 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1087 neh->eh_magic = EXT4_EXT_MAGIC;
1088 set_buffer_uptodate(bh);
1091 err = ext4_handle_dirty_metadata(handle, inode, bh);
1095 /* create index in new top-level index: num,max,pointer */
1096 err = ext4_ext_get_access(handle, inode, curp);
1100 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1101 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1102 curp->p_hdr->eh_entries = cpu_to_le16(1);
1103 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1105 if (path[0].p_hdr->eh_depth)
1106 curp->p_idx->ei_block =
1107 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1109 curp->p_idx->ei_block =
1110 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1111 ext4_idx_store_pblock(curp->p_idx, newblock);
1113 neh = ext_inode_hdr(inode);
1114 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1115 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1116 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1117 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1119 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1120 err = ext4_ext_dirty(handle, inode, curp);
1128 * ext4_ext_create_new_leaf:
1129 * finds empty index and adds new leaf.
1130 * if no free index is found, then it requests in-depth growing.
1132 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1134 struct ext4_ext_path *path,
1135 struct ext4_extent *newext)
1137 struct ext4_ext_path *curp;
1138 int depth, i, err = 0;
1141 i = depth = ext_depth(inode);
1143 /* walk up to the tree and look for free index entry */
1144 curp = path + depth;
1145 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1150 /* we use already allocated block for index block,
1151 * so subsequent data blocks should be contiguous */
1152 if (EXT_HAS_FREE_INDEX(curp)) {
1153 /* if we found index with free entry, then use that
1154 * entry: create all needed subtree and add new leaf */
1155 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1160 ext4_ext_drop_refs(path);
1161 path = ext4_ext_find_extent(inode,
1162 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1165 err = PTR_ERR(path);
1167 /* tree is full, time to grow in depth */
1168 err = ext4_ext_grow_indepth(handle, inode, flags,
1174 ext4_ext_drop_refs(path);
1175 path = ext4_ext_find_extent(inode,
1176 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1179 err = PTR_ERR(path);
1184 * only first (depth 0 -> 1) produces free space;
1185 * in all other cases we have to split the grown tree
1187 depth = ext_depth(inode);
1188 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1189 /* now we need to split */
1199 * search the closest allocated block to the left for *logical
1200 * and returns it at @logical + it's physical address at @phys
1201 * if *logical is the smallest allocated block, the function
1202 * returns 0 at @phys
1203 * return value contains 0 (success) or error code
1205 static int ext4_ext_search_left(struct inode *inode,
1206 struct ext4_ext_path *path,
1207 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1209 struct ext4_extent_idx *ix;
1210 struct ext4_extent *ex;
1213 if (unlikely(path == NULL)) {
1214 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1217 depth = path->p_depth;
1220 if (depth == 0 && path->p_ext == NULL)
1223 /* usually extent in the path covers blocks smaller
1224 * then *logical, but it can be that extent is the
1225 * first one in the file */
1227 ex = path[depth].p_ext;
1228 ee_len = ext4_ext_get_actual_len(ex);
1229 if (*logical < le32_to_cpu(ex->ee_block)) {
1230 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1231 EXT4_ERROR_INODE(inode,
1232 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1233 *logical, le32_to_cpu(ex->ee_block));
1236 while (--depth >= 0) {
1237 ix = path[depth].p_idx;
1238 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1239 EXT4_ERROR_INODE(inode,
1240 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1241 ix != NULL ? ix->ei_block : 0,
1242 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1243 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1251 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1252 EXT4_ERROR_INODE(inode,
1253 "logical %d < ee_block %d + ee_len %d!",
1254 *logical, le32_to_cpu(ex->ee_block), ee_len);
1258 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1259 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1264 * search the closest allocated block to the right for *logical
1265 * and returns it at @logical + it's physical address at @phys
1266 * if *logical is the smallest allocated block, the function
1267 * returns 0 at @phys
1268 * return value contains 0 (success) or error code
1270 static int ext4_ext_search_right(struct inode *inode,
1271 struct ext4_ext_path *path,
1272 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1273 struct ext4_extent **ret_ex)
1275 struct buffer_head *bh = NULL;
1276 struct ext4_extent_header *eh;
1277 struct ext4_extent_idx *ix;
1278 struct ext4_extent *ex;
1280 int depth; /* Note, NOT eh_depth; depth from top of tree */
1283 if (unlikely(path == NULL)) {
1284 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1287 depth = path->p_depth;
1290 if (depth == 0 && path->p_ext == NULL)
1293 /* usually extent in the path covers blocks smaller
1294 * then *logical, but it can be that extent is the
1295 * first one in the file */
1297 ex = path[depth].p_ext;
1298 ee_len = ext4_ext_get_actual_len(ex);
1299 if (*logical < le32_to_cpu(ex->ee_block)) {
1300 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1301 EXT4_ERROR_INODE(inode,
1302 "first_extent(path[%d].p_hdr) != ex",
1306 while (--depth >= 0) {
1307 ix = path[depth].p_idx;
1308 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1309 EXT4_ERROR_INODE(inode,
1310 "ix != EXT_FIRST_INDEX *logical %d!",
1318 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1319 EXT4_ERROR_INODE(inode,
1320 "logical %d < ee_block %d + ee_len %d!",
1321 *logical, le32_to_cpu(ex->ee_block), ee_len);
1325 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1326 /* next allocated block in this leaf */
1331 /* go up and search for index to the right */
1332 while (--depth >= 0) {
1333 ix = path[depth].p_idx;
1334 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1338 /* we've gone up to the root and found no index to the right */
1342 /* we've found index to the right, let's
1343 * follow it and find the closest allocated
1344 * block to the right */
1346 block = ext4_idx_pblock(ix);
1347 while (++depth < path->p_depth) {
1348 bh = sb_bread(inode->i_sb, block);
1351 eh = ext_block_hdr(bh);
1352 /* subtract from p_depth to get proper eh_depth */
1353 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1357 ix = EXT_FIRST_INDEX(eh);
1358 block = ext4_idx_pblock(ix);
1362 bh = sb_bread(inode->i_sb, block);
1365 eh = ext_block_hdr(bh);
1366 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1370 ex = EXT_FIRST_EXTENT(eh);
1372 *logical = le32_to_cpu(ex->ee_block);
1373 *phys = ext4_ext_pblock(ex);
1381 * ext4_ext_next_allocated_block:
1382 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1383 * NOTE: it considers block number from index entry as
1384 * allocated block. Thus, index entries have to be consistent
1388 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1392 BUG_ON(path == NULL);
1393 depth = path->p_depth;
1395 if (depth == 0 && path->p_ext == NULL)
1396 return EXT_MAX_BLOCKS;
1398 while (depth >= 0) {
1399 if (depth == path->p_depth) {
1401 if (path[depth].p_ext !=
1402 EXT_LAST_EXTENT(path[depth].p_hdr))
1403 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1406 if (path[depth].p_idx !=
1407 EXT_LAST_INDEX(path[depth].p_hdr))
1408 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1413 return EXT_MAX_BLOCKS;
1417 * ext4_ext_next_leaf_block:
1418 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1420 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1424 BUG_ON(path == NULL);
1425 depth = path->p_depth;
1427 /* zero-tree has no leaf blocks at all */
1429 return EXT_MAX_BLOCKS;
1431 /* go to index block */
1434 while (depth >= 0) {
1435 if (path[depth].p_idx !=
1436 EXT_LAST_INDEX(path[depth].p_hdr))
1437 return (ext4_lblk_t)
1438 le32_to_cpu(path[depth].p_idx[1].ei_block);
1442 return EXT_MAX_BLOCKS;
1446 * ext4_ext_correct_indexes:
1447 * if leaf gets modified and modified extent is first in the leaf,
1448 * then we have to correct all indexes above.
1449 * TODO: do we need to correct tree in all cases?
1451 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1452 struct ext4_ext_path *path)
1454 struct ext4_extent_header *eh;
1455 int depth = ext_depth(inode);
1456 struct ext4_extent *ex;
1460 eh = path[depth].p_hdr;
1461 ex = path[depth].p_ext;
1463 if (unlikely(ex == NULL || eh == NULL)) {
1464 EXT4_ERROR_INODE(inode,
1465 "ex %p == NULL or eh %p == NULL", ex, eh);
1470 /* there is no tree at all */
1474 if (ex != EXT_FIRST_EXTENT(eh)) {
1475 /* we correct tree if first leaf got modified only */
1480 * TODO: we need correction if border is smaller than current one
1483 border = path[depth].p_ext->ee_block;
1484 err = ext4_ext_get_access(handle, inode, path + k);
1487 path[k].p_idx->ei_block = border;
1488 err = ext4_ext_dirty(handle, inode, path + k);
1493 /* change all left-side indexes */
1494 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1496 err = ext4_ext_get_access(handle, inode, path + k);
1499 path[k].p_idx->ei_block = border;
1500 err = ext4_ext_dirty(handle, inode, path + k);
1509 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1510 struct ext4_extent *ex2)
1512 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1515 * Make sure that either both extents are uninitialized, or
1518 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1521 if (ext4_ext_is_uninitialized(ex1))
1522 max_len = EXT_UNINIT_MAX_LEN;
1524 max_len = EXT_INIT_MAX_LEN;
1526 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1527 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1529 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1530 le32_to_cpu(ex2->ee_block))
1534 * To allow future support for preallocated extents to be added
1535 * as an RO_COMPAT feature, refuse to merge to extents if
1536 * this can result in the top bit of ee_len being set.
1538 if (ext1_ee_len + ext2_ee_len > max_len)
1540 #ifdef AGGRESSIVE_TEST
1541 if (ext1_ee_len >= 4)
1545 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1551 * This function tries to merge the "ex" extent to the next extent in the tree.
1552 * It always tries to merge towards right. If you want to merge towards
1553 * left, pass "ex - 1" as argument instead of "ex".
1554 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1555 * 1 if they got merged.
1557 static int ext4_ext_try_to_merge_right(struct inode *inode,
1558 struct ext4_ext_path *path,
1559 struct ext4_extent *ex)
1561 struct ext4_extent_header *eh;
1562 unsigned int depth, len;
1564 int uninitialized = 0;
1566 depth = ext_depth(inode);
1567 BUG_ON(path[depth].p_hdr == NULL);
1568 eh = path[depth].p_hdr;
1570 while (ex < EXT_LAST_EXTENT(eh)) {
1571 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1573 /* merge with next extent! */
1574 if (ext4_ext_is_uninitialized(ex))
1576 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1577 + ext4_ext_get_actual_len(ex + 1));
1579 ext4_ext_mark_uninitialized(ex);
1581 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1582 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1583 * sizeof(struct ext4_extent);
1584 memmove(ex + 1, ex + 2, len);
1586 le16_add_cpu(&eh->eh_entries, -1);
1588 WARN_ON(eh->eh_entries == 0);
1589 if (!eh->eh_entries)
1590 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1597 * This function tries to merge the @ex extent to neighbours in the tree.
1598 * return 1 if merge left else 0.
1600 static int ext4_ext_try_to_merge(struct inode *inode,
1601 struct ext4_ext_path *path,
1602 struct ext4_extent *ex) {
1603 struct ext4_extent_header *eh;
1608 depth = ext_depth(inode);
1609 BUG_ON(path[depth].p_hdr == NULL);
1610 eh = path[depth].p_hdr;
1612 if (ex > EXT_FIRST_EXTENT(eh))
1613 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1616 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1622 * check if a portion of the "newext" extent overlaps with an
1625 * If there is an overlap discovered, it updates the length of the newext
1626 * such that there will be no overlap, and then returns 1.
1627 * If there is no overlap found, it returns 0.
1629 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1630 struct inode *inode,
1631 struct ext4_extent *newext,
1632 struct ext4_ext_path *path)
1635 unsigned int depth, len1;
1636 unsigned int ret = 0;
1638 b1 = le32_to_cpu(newext->ee_block);
1639 len1 = ext4_ext_get_actual_len(newext);
1640 depth = ext_depth(inode);
1641 if (!path[depth].p_ext)
1643 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1644 b2 &= ~(sbi->s_cluster_ratio - 1);
1647 * get the next allocated block if the extent in the path
1648 * is before the requested block(s)
1651 b2 = ext4_ext_next_allocated_block(path);
1652 if (b2 == EXT_MAX_BLOCKS)
1654 b2 &= ~(sbi->s_cluster_ratio - 1);
1657 /* check for wrap through zero on extent logical start block*/
1658 if (b1 + len1 < b1) {
1659 len1 = EXT_MAX_BLOCKS - b1;
1660 newext->ee_len = cpu_to_le16(len1);
1664 /* check for overlap */
1665 if (b1 + len1 > b2) {
1666 newext->ee_len = cpu_to_le16(b2 - b1);
1674 * ext4_ext_insert_extent:
1675 * tries to merge requsted extent into the existing extent or
1676 * inserts requested extent as new one into the tree,
1677 * creating new leaf in the no-space case.
1679 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1680 struct ext4_ext_path *path,
1681 struct ext4_extent *newext, int flag)
1683 struct ext4_extent_header *eh;
1684 struct ext4_extent *ex, *fex;
1685 struct ext4_extent *nearex; /* nearest extent */
1686 struct ext4_ext_path *npath = NULL;
1687 int depth, len, err;
1689 unsigned uninitialized = 0;
1692 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1693 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1696 depth = ext_depth(inode);
1697 ex = path[depth].p_ext;
1698 if (unlikely(path[depth].p_hdr == NULL)) {
1699 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1703 /* try to insert block into found extent and return */
1704 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1705 && ext4_can_extents_be_merged(inode, ex, newext)) {
1706 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1707 ext4_ext_is_uninitialized(newext),
1708 ext4_ext_get_actual_len(newext),
1709 le32_to_cpu(ex->ee_block),
1710 ext4_ext_is_uninitialized(ex),
1711 ext4_ext_get_actual_len(ex),
1712 ext4_ext_pblock(ex));
1713 err = ext4_ext_get_access(handle, inode, path + depth);
1718 * ext4_can_extents_be_merged should have checked that either
1719 * both extents are uninitialized, or both aren't. Thus we
1720 * need to check only one of them here.
1722 if (ext4_ext_is_uninitialized(ex))
1724 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1725 + ext4_ext_get_actual_len(newext));
1727 ext4_ext_mark_uninitialized(ex);
1728 eh = path[depth].p_hdr;
1733 depth = ext_depth(inode);
1734 eh = path[depth].p_hdr;
1735 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1738 /* probably next leaf has space for us? */
1739 fex = EXT_LAST_EXTENT(eh);
1740 next = EXT_MAX_BLOCKS;
1741 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1742 next = ext4_ext_next_leaf_block(path);
1743 if (next != EXT_MAX_BLOCKS) {
1744 ext_debug("next leaf block - %d\n", next);
1745 BUG_ON(npath != NULL);
1746 npath = ext4_ext_find_extent(inode, next, NULL);
1748 return PTR_ERR(npath);
1749 BUG_ON(npath->p_depth != path->p_depth);
1750 eh = npath[depth].p_hdr;
1751 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1752 ext_debug("next leaf isn't full(%d)\n",
1753 le16_to_cpu(eh->eh_entries));
1757 ext_debug("next leaf has no free space(%d,%d)\n",
1758 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1762 * There is no free space in the found leaf.
1763 * We're gonna add a new leaf in the tree.
1765 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1766 flags = EXT4_MB_USE_ROOT_BLOCKS;
1767 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1770 depth = ext_depth(inode);
1771 eh = path[depth].p_hdr;
1774 nearex = path[depth].p_ext;
1776 err = ext4_ext_get_access(handle, inode, path + depth);
1781 /* there is no extent in this leaf, create first one */
1782 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1783 le32_to_cpu(newext->ee_block),
1784 ext4_ext_pblock(newext),
1785 ext4_ext_is_uninitialized(newext),
1786 ext4_ext_get_actual_len(newext));
1787 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1788 } else if (le32_to_cpu(newext->ee_block)
1789 > le32_to_cpu(nearex->ee_block)) {
1790 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1791 if (nearex != EXT_LAST_EXTENT(eh)) {
1792 len = EXT_MAX_EXTENT(eh) - nearex;
1793 len = (len - 1) * sizeof(struct ext4_extent);
1794 len = len < 0 ? 0 : len;
1795 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1796 "move %d from 0x%p to 0x%p\n",
1797 le32_to_cpu(newext->ee_block),
1798 ext4_ext_pblock(newext),
1799 ext4_ext_is_uninitialized(newext),
1800 ext4_ext_get_actual_len(newext),
1801 nearex, len, nearex + 1, nearex + 2);
1802 memmove(nearex + 2, nearex + 1, len);
1804 path[depth].p_ext = nearex + 1;
1806 BUG_ON(newext->ee_block == nearex->ee_block);
1807 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1808 len = len < 0 ? 0 : len;
1809 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1810 "move %d from 0x%p to 0x%p\n",
1811 le32_to_cpu(newext->ee_block),
1812 ext4_ext_pblock(newext),
1813 ext4_ext_is_uninitialized(newext),
1814 ext4_ext_get_actual_len(newext),
1815 nearex, len, nearex, nearex + 1);
1816 memmove(nearex + 1, nearex, len);
1817 path[depth].p_ext = nearex;
1820 le16_add_cpu(&eh->eh_entries, 1);
1821 nearex = path[depth].p_ext;
1822 nearex->ee_block = newext->ee_block;
1823 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1824 nearex->ee_len = newext->ee_len;
1827 /* try to merge extents to the right */
1828 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1829 ext4_ext_try_to_merge(inode, path, nearex);
1831 /* try to merge extents to the left */
1833 /* time to correct all indexes above */
1834 err = ext4_ext_correct_indexes(handle, inode, path);
1838 err = ext4_ext_dirty(handle, inode, path + depth);
1842 ext4_ext_drop_refs(npath);
1845 ext4_ext_invalidate_cache(inode);
1849 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1850 ext4_lblk_t num, ext_prepare_callback func,
1853 struct ext4_ext_path *path = NULL;
1854 struct ext4_ext_cache cbex;
1855 struct ext4_extent *ex;
1856 ext4_lblk_t next, start = 0, end = 0;
1857 ext4_lblk_t last = block + num;
1858 int depth, exists, err = 0;
1860 BUG_ON(func == NULL);
1861 BUG_ON(inode == NULL);
1863 while (block < last && block != EXT_MAX_BLOCKS) {
1865 /* find extent for this block */
1866 down_read(&EXT4_I(inode)->i_data_sem);
1867 path = ext4_ext_find_extent(inode, block, path);
1868 up_read(&EXT4_I(inode)->i_data_sem);
1870 err = PTR_ERR(path);
1875 depth = ext_depth(inode);
1876 if (unlikely(path[depth].p_hdr == NULL)) {
1877 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1881 ex = path[depth].p_ext;
1882 next = ext4_ext_next_allocated_block(path);
1886 /* there is no extent yet, so try to allocate
1887 * all requested space */
1890 } else if (le32_to_cpu(ex->ee_block) > block) {
1891 /* need to allocate space before found extent */
1893 end = le32_to_cpu(ex->ee_block);
1894 if (block + num < end)
1896 } else if (block >= le32_to_cpu(ex->ee_block)
1897 + ext4_ext_get_actual_len(ex)) {
1898 /* need to allocate space after found extent */
1903 } else if (block >= le32_to_cpu(ex->ee_block)) {
1905 * some part of requested space is covered
1909 end = le32_to_cpu(ex->ee_block)
1910 + ext4_ext_get_actual_len(ex);
1911 if (block + num < end)
1917 BUG_ON(end <= start);
1920 cbex.ec_block = start;
1921 cbex.ec_len = end - start;
1924 cbex.ec_block = le32_to_cpu(ex->ee_block);
1925 cbex.ec_len = ext4_ext_get_actual_len(ex);
1926 cbex.ec_start = ext4_ext_pblock(ex);
1929 if (unlikely(cbex.ec_len == 0)) {
1930 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1934 err = func(inode, next, &cbex, ex, cbdata);
1935 ext4_ext_drop_refs(path);
1940 if (err == EXT_REPEAT)
1942 else if (err == EXT_BREAK) {
1947 if (ext_depth(inode) != depth) {
1948 /* depth was changed. we have to realloc path */
1953 block = cbex.ec_block + cbex.ec_len;
1957 ext4_ext_drop_refs(path);
1965 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1966 __u32 len, ext4_fsblk_t start)
1968 struct ext4_ext_cache *cex;
1970 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1971 trace_ext4_ext_put_in_cache(inode, block, len, start);
1972 cex = &EXT4_I(inode)->i_cached_extent;
1973 cex->ec_block = block;
1975 cex->ec_start = start;
1976 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1980 * ext4_ext_put_gap_in_cache:
1981 * calculate boundaries of the gap that the requested block fits into
1982 * and cache this gap
1985 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1988 int depth = ext_depth(inode);
1991 struct ext4_extent *ex;
1993 ex = path[depth].p_ext;
1995 /* there is no extent yet, so gap is [0;-] */
1997 len = EXT_MAX_BLOCKS;
1998 ext_debug("cache gap(whole file):");
1999 } else if (block < le32_to_cpu(ex->ee_block)) {
2001 len = le32_to_cpu(ex->ee_block) - block;
2002 ext_debug("cache gap(before): %u [%u:%u]",
2004 le32_to_cpu(ex->ee_block),
2005 ext4_ext_get_actual_len(ex));
2006 } else if (block >= le32_to_cpu(ex->ee_block)
2007 + ext4_ext_get_actual_len(ex)) {
2009 lblock = le32_to_cpu(ex->ee_block)
2010 + ext4_ext_get_actual_len(ex);
2012 next = ext4_ext_next_allocated_block(path);
2013 ext_debug("cache gap(after): [%u:%u] %u",
2014 le32_to_cpu(ex->ee_block),
2015 ext4_ext_get_actual_len(ex),
2017 BUG_ON(next == lblock);
2018 len = next - lblock;
2024 ext_debug(" -> %u:%lu\n", lblock, len);
2025 ext4_ext_put_in_cache(inode, lblock, len, 0);
2029 * ext4_ext_check_cache()
2030 * Checks to see if the given block is in the cache.
2031 * If it is, the cached extent is stored in the given
2032 * cache extent pointer. If the cached extent is a hole,
2033 * this routine should be used instead of
2034 * ext4_ext_in_cache if the calling function needs to
2035 * know the size of the hole.
2037 * @inode: The files inode
2038 * @block: The block to look for in the cache
2039 * @ex: Pointer where the cached extent will be stored
2040 * if it contains block
2042 * Return 0 if cache is invalid; 1 if the cache is valid
2044 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2045 struct ext4_ext_cache *ex){
2046 struct ext4_ext_cache *cex;
2047 struct ext4_sb_info *sbi;
2051 * We borrow i_block_reservation_lock to protect i_cached_extent
2053 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2054 cex = &EXT4_I(inode)->i_cached_extent;
2055 sbi = EXT4_SB(inode->i_sb);
2057 /* has cache valid data? */
2058 if (cex->ec_len == 0)
2061 if (in_range(block, cex->ec_block, cex->ec_len)) {
2062 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2063 ext_debug("%u cached by %u:%u:%llu\n",
2065 cex->ec_block, cex->ec_len, cex->ec_start);
2070 sbi->extent_cache_misses++;
2072 sbi->extent_cache_hits++;
2073 trace_ext4_ext_in_cache(inode, block, ret);
2074 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2079 * ext4_ext_in_cache()
2080 * Checks to see if the given block is in the cache.
2081 * If it is, the cached extent is stored in the given
2084 * @inode: The files inode
2085 * @block: The block to look for in the cache
2086 * @ex: Pointer where the cached extent will be stored
2087 * if it contains block
2089 * Return 0 if cache is invalid; 1 if the cache is valid
2092 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2093 struct ext4_extent *ex)
2095 struct ext4_ext_cache cex;
2098 if (ext4_ext_check_cache(inode, block, &cex)) {
2099 ex->ee_block = cpu_to_le32(cex.ec_block);
2100 ext4_ext_store_pblock(ex, cex.ec_start);
2101 ex->ee_len = cpu_to_le16(cex.ec_len);
2111 * removes index from the index block.
2113 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2114 struct ext4_ext_path *path)
2119 /* free index block */
2121 leaf = ext4_idx_pblock(path->p_idx);
2122 if (unlikely(path->p_hdr->eh_entries == 0)) {
2123 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2126 err = ext4_ext_get_access(handle, inode, path);
2130 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2131 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2132 len *= sizeof(struct ext4_extent_idx);
2133 memmove(path->p_idx, path->p_idx + 1, len);
2136 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2137 err = ext4_ext_dirty(handle, inode, path);
2140 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2141 trace_ext4_ext_rm_idx(inode, leaf);
2143 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2144 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2149 * ext4_ext_calc_credits_for_single_extent:
2150 * This routine returns max. credits that needed to insert an extent
2151 * to the extent tree.
2152 * When pass the actual path, the caller should calculate credits
2155 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2156 struct ext4_ext_path *path)
2159 int depth = ext_depth(inode);
2162 /* probably there is space in leaf? */
2163 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2164 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2167 * There are some space in the leaf tree, no
2168 * need to account for leaf block credit
2170 * bitmaps and block group descriptor blocks
2171 * and other metadat blocks still need to be
2174 /* 1 bitmap, 1 block group descriptor */
2175 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2180 return ext4_chunk_trans_blocks(inode, nrblocks);
2184 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2186 * if nrblocks are fit in a single extent (chunk flag is 1), then
2187 * in the worse case, each tree level index/leaf need to be changed
2188 * if the tree split due to insert a new extent, then the old tree
2189 * index/leaf need to be updated too
2191 * If the nrblocks are discontiguous, they could cause
2192 * the whole tree split more than once, but this is really rare.
2194 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2197 int depth = ext_depth(inode);
2207 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2208 struct ext4_extent *ex,
2209 ext4_fsblk_t *partial_cluster,
2210 ext4_lblk_t from, ext4_lblk_t to)
2212 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2213 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2215 int flags = EXT4_FREE_BLOCKS_FORGET;
2217 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2218 flags |= EXT4_FREE_BLOCKS_METADATA;
2220 * For bigalloc file systems, we never free a partial cluster
2221 * at the beginning of the extent. Instead, we make a note
2222 * that we tried freeing the cluster, and check to see if we
2223 * need to free it on a subsequent call to ext4_remove_blocks,
2224 * or at the end of the ext4_truncate() operation.
2226 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2228 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2230 * If we have a partial cluster, and it's different from the
2231 * cluster of the last block, we need to explicitly free the
2232 * partial cluster here.
2234 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2235 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2236 ext4_free_blocks(handle, inode, NULL,
2237 EXT4_C2B(sbi, *partial_cluster),
2238 sbi->s_cluster_ratio, flags);
2239 *partial_cluster = 0;
2242 #ifdef EXTENTS_STATS
2244 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2245 spin_lock(&sbi->s_ext_stats_lock);
2246 sbi->s_ext_blocks += ee_len;
2247 sbi->s_ext_extents++;
2248 if (ee_len < sbi->s_ext_min)
2249 sbi->s_ext_min = ee_len;
2250 if (ee_len > sbi->s_ext_max)
2251 sbi->s_ext_max = ee_len;
2252 if (ext_depth(inode) > sbi->s_depth_max)
2253 sbi->s_depth_max = ext_depth(inode);
2254 spin_unlock(&sbi->s_ext_stats_lock);
2257 if (from >= le32_to_cpu(ex->ee_block)
2258 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2262 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2263 pblk = ext4_ext_pblock(ex) + ee_len - num;
2264 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2265 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2267 * If the block range to be freed didn't start at the
2268 * beginning of a cluster, and we removed the entire
2269 * extent, save the partial cluster here, since we
2270 * might need to delete if we determine that the
2271 * truncate operation has removed all of the blocks in
2274 if (pblk & (sbi->s_cluster_ratio - 1) &&
2276 *partial_cluster = EXT4_B2C(sbi, pblk);
2278 *partial_cluster = 0;
2279 } else if (from == le32_to_cpu(ex->ee_block)
2280 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2286 start = ext4_ext_pblock(ex);
2288 ext_debug("free first %u blocks starting %llu\n", num, start);
2289 ext4_free_blocks(handle, inode, 0, start, num, flags);
2292 printk(KERN_INFO "strange request: removal(2) "
2293 "%u-%u from %u:%u\n",
2294 from, to, le32_to_cpu(ex->ee_block), ee_len);
2301 * ext4_ext_rm_leaf() Removes the extents associated with the
2302 * blocks appearing between "start" and "end", and splits the extents
2303 * if "start" and "end" appear in the same extent
2305 * @handle: The journal handle
2306 * @inode: The files inode
2307 * @path: The path to the leaf
2308 * @start: The first block to remove
2309 * @end: The last block to remove
2312 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2313 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2314 ext4_lblk_t start, ext4_lblk_t end)
2316 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2317 int err = 0, correct_index = 0;
2318 int depth = ext_depth(inode), credits;
2319 struct ext4_extent_header *eh;
2320 ext4_lblk_t a, b, block;
2322 ext4_lblk_t ex_ee_block;
2323 unsigned short ex_ee_len;
2324 unsigned uninitialized = 0;
2325 struct ext4_extent *ex;
2326 struct ext4_map_blocks map;
2328 /* the header must be checked already in ext4_ext_remove_space() */
2329 ext_debug("truncate since %u in leaf\n", start);
2330 if (!path[depth].p_hdr)
2331 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2332 eh = path[depth].p_hdr;
2333 if (unlikely(path[depth].p_hdr == NULL)) {
2334 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2337 /* find where to start removing */
2338 ex = EXT_LAST_EXTENT(eh);
2340 ex_ee_block = le32_to_cpu(ex->ee_block);
2341 ex_ee_len = ext4_ext_get_actual_len(ex);
2343 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2345 while (ex >= EXT_FIRST_EXTENT(eh) &&
2346 ex_ee_block + ex_ee_len > start) {
2348 if (ext4_ext_is_uninitialized(ex))
2353 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2354 uninitialized, ex_ee_len);
2355 path[depth].p_ext = ex;
2357 a = ex_ee_block > start ? ex_ee_block : start;
2358 b = ex_ee_block+ex_ee_len - 1 < end ?
2359 ex_ee_block+ex_ee_len - 1 : end;
2361 ext_debug(" border %u:%u\n", a, b);
2363 /* If this extent is beyond the end of the hole, skip it */
2364 if (end <= ex_ee_block) {
2366 ex_ee_block = le32_to_cpu(ex->ee_block);
2367 ex_ee_len = ext4_ext_get_actual_len(ex);
2369 } else if (a != ex_ee_block &&
2370 b != ex_ee_block + ex_ee_len - 1) {
2372 * If this is a truncate, then this condition should
2373 * never happen because at least one of the end points
2374 * needs to be on the edge of the extent.
2376 if (end == EXT_MAX_BLOCKS - 1) {
2377 ext_debug(" bad truncate %u:%u\n",
2385 * else this is a hole punch, so the extent needs to
2386 * be split since neither edge of the hole is on the
2390 map.m_pblk = ext4_ext_pblock(ex);
2391 map.m_lblk = ex_ee_block;
2392 map.m_len = b - ex_ee_block;
2394 err = ext4_split_extent(handle,
2395 inode, path, &map, 0,
2396 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2397 EXT4_GET_BLOCKS_PRE_IO);
2402 ex_ee_len = ext4_ext_get_actual_len(ex);
2404 b = ex_ee_block+ex_ee_len - 1 < end ?
2405 ex_ee_block+ex_ee_len - 1 : end;
2407 /* Then remove tail of this extent */
2408 block = ex_ee_block;
2411 } else if (a != ex_ee_block) {
2412 /* remove tail of the extent */
2413 block = ex_ee_block;
2415 } else if (b != ex_ee_block + ex_ee_len - 1) {
2416 /* remove head of the extent */
2418 num = ex_ee_block + ex_ee_len - b;
2421 * If this is a truncate, this condition
2422 * should never happen
2424 if (end == EXT_MAX_BLOCKS - 1) {
2425 ext_debug(" bad truncate %u:%u\n",
2431 /* remove whole extent: excellent! */
2432 block = ex_ee_block;
2434 if (a != ex_ee_block) {
2435 ext_debug(" bad truncate %u:%u\n",
2441 if (b != ex_ee_block + ex_ee_len - 1) {
2442 ext_debug(" bad truncate %u:%u\n",
2450 * 3 for leaf, sb, and inode plus 2 (bmap and group
2451 * descriptor) for each block group; assume two block
2452 * groups plus ex_ee_len/blocks_per_block_group for
2455 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2456 if (ex == EXT_FIRST_EXTENT(eh)) {
2458 credits += (ext_depth(inode)) + 1;
2460 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2462 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2466 err = ext4_ext_get_access(handle, inode, path + depth);
2470 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2476 /* this extent is removed; mark slot entirely unused */
2477 ext4_ext_store_pblock(ex, 0);
2478 } else if (block != ex_ee_block) {
2480 * If this was a head removal, then we need to update
2481 * the physical block since it is now at a different
2484 ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
2487 ex->ee_block = cpu_to_le32(block);
2488 ex->ee_len = cpu_to_le16(num);
2490 * Do not mark uninitialized if all the blocks in the
2491 * extent have been removed.
2493 if (uninitialized && num)
2494 ext4_ext_mark_uninitialized(ex);
2496 err = ext4_ext_dirty(handle, inode, path + depth);
2501 * If the extent was completely released,
2502 * we need to remove it from the leaf
2505 if (end != EXT_MAX_BLOCKS - 1) {
2507 * For hole punching, we need to scoot all the
2508 * extents up when an extent is removed so that
2509 * we dont have blank extents in the middle
2511 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2512 sizeof(struct ext4_extent));
2514 /* Now get rid of the one at the end */
2515 memset(EXT_LAST_EXTENT(eh), 0,
2516 sizeof(struct ext4_extent));
2518 le16_add_cpu(&eh->eh_entries, -1);
2520 *partial_cluster = 0;
2522 ext_debug("new extent: %u:%u:%llu\n", block, num,
2523 ext4_ext_pblock(ex));
2525 ex_ee_block = le32_to_cpu(ex->ee_block);
2526 ex_ee_len = ext4_ext_get_actual_len(ex);
2529 if (correct_index && eh->eh_entries)
2530 err = ext4_ext_correct_indexes(handle, inode, path);
2533 * If there is still a entry in the leaf node, check to see if
2534 * it references the partial cluster. This is the only place
2535 * where it could; if it doesn't, we can free the cluster.
2537 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2538 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2539 *partial_cluster)) {
2540 int flags = EXT4_FREE_BLOCKS_FORGET;
2542 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2543 flags |= EXT4_FREE_BLOCKS_METADATA;
2545 ext4_free_blocks(handle, inode, NULL,
2546 EXT4_C2B(sbi, *partial_cluster),
2547 sbi->s_cluster_ratio, flags);
2548 *partial_cluster = 0;
2551 /* if this leaf is free, then we should
2552 * remove it from index block above */
2553 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2554 err = ext4_ext_rm_idx(handle, inode, path + depth);
2561 * ext4_ext_more_to_rm:
2562 * returns 1 if current index has to be freed (even partial)
2565 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2567 BUG_ON(path->p_idx == NULL);
2569 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2573 * if truncate on deeper level happened, it wasn't partial,
2574 * so we have to consider current index for truncation
2576 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2581 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2583 struct super_block *sb = inode->i_sb;
2584 int depth = ext_depth(inode);
2585 struct ext4_ext_path *path;
2586 ext4_fsblk_t partial_cluster = 0;
2590 ext_debug("truncate since %u\n", start);
2592 /* probably first extent we're gonna free will be last in block */
2593 handle = ext4_journal_start(inode, depth + 1);
2595 return PTR_ERR(handle);
2598 ext4_ext_invalidate_cache(inode);
2600 trace_ext4_ext_remove_space(inode, start, depth);
2603 * We start scanning from right side, freeing all the blocks
2604 * after i_size and walking into the tree depth-wise.
2606 depth = ext_depth(inode);
2607 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2609 ext4_journal_stop(handle);
2612 path[0].p_depth = depth;
2613 path[0].p_hdr = ext_inode_hdr(inode);
2614 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2620 while (i >= 0 && err == 0) {
2622 /* this is leaf block */
2623 err = ext4_ext_rm_leaf(handle, inode, path,
2624 &partial_cluster, start,
2625 EXT_MAX_BLOCKS - 1);
2626 /* root level has p_bh == NULL, brelse() eats this */
2627 brelse(path[i].p_bh);
2628 path[i].p_bh = NULL;
2633 /* this is index block */
2634 if (!path[i].p_hdr) {
2635 ext_debug("initialize header\n");
2636 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2639 if (!path[i].p_idx) {
2640 /* this level hasn't been touched yet */
2641 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2642 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2643 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2645 le16_to_cpu(path[i].p_hdr->eh_entries));
2647 /* we were already here, see at next index */
2651 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2652 i, EXT_FIRST_INDEX(path[i].p_hdr),
2654 if (ext4_ext_more_to_rm(path + i)) {
2655 struct buffer_head *bh;
2656 /* go to the next level */
2657 ext_debug("move to level %d (block %llu)\n",
2658 i + 1, ext4_idx_pblock(path[i].p_idx));
2659 memset(path + i + 1, 0, sizeof(*path));
2660 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2662 /* should we reset i_size? */
2666 if (WARN_ON(i + 1 > depth)) {
2670 if (ext4_ext_check(inode, ext_block_hdr(bh),
2675 path[i + 1].p_bh = bh;
2677 /* save actual number of indexes since this
2678 * number is changed at the next iteration */
2679 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2682 /* we finished processing this index, go up */
2683 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2684 /* index is empty, remove it;
2685 * handle must be already prepared by the
2686 * truncatei_leaf() */
2687 err = ext4_ext_rm_idx(handle, inode, path + i);
2689 /* root level has p_bh == NULL, brelse() eats this */
2690 brelse(path[i].p_bh);
2691 path[i].p_bh = NULL;
2693 ext_debug("return to level %d\n", i);
2697 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2698 path->p_hdr->eh_entries);
2700 /* If we still have something in the partial cluster and we have removed
2701 * even the first extent, then we should free the blocks in the partial
2702 * cluster as well. */
2703 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2704 int flags = EXT4_FREE_BLOCKS_FORGET;
2706 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2707 flags |= EXT4_FREE_BLOCKS_METADATA;
2709 ext4_free_blocks(handle, inode, NULL,
2710 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2711 EXT4_SB(sb)->s_cluster_ratio, flags);
2712 partial_cluster = 0;
2715 /* TODO: flexible tree reduction should be here */
2716 if (path->p_hdr->eh_entries == 0) {
2718 * truncate to zero freed all the tree,
2719 * so we need to correct eh_depth
2721 err = ext4_ext_get_access(handle, inode, path);
2723 ext_inode_hdr(inode)->eh_depth = 0;
2724 ext_inode_hdr(inode)->eh_max =
2725 cpu_to_le16(ext4_ext_space_root(inode, 0));
2726 err = ext4_ext_dirty(handle, inode, path);
2730 ext4_ext_drop_refs(path);
2734 ext4_journal_stop(handle);
2740 * called at mount time
2742 void ext4_ext_init(struct super_block *sb)
2745 * possible initialization would be here
2748 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2749 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2750 printk(KERN_INFO "EXT4-fs: file extents enabled");
2751 #ifdef AGGRESSIVE_TEST
2752 printk(", aggressive tests");
2754 #ifdef CHECK_BINSEARCH
2755 printk(", check binsearch");
2757 #ifdef EXTENTS_STATS
2762 #ifdef EXTENTS_STATS
2763 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2764 EXT4_SB(sb)->s_ext_min = 1 << 30;
2765 EXT4_SB(sb)->s_ext_max = 0;
2771 * called at umount time
2773 void ext4_ext_release(struct super_block *sb)
2775 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2778 #ifdef EXTENTS_STATS
2779 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2780 struct ext4_sb_info *sbi = EXT4_SB(sb);
2781 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2782 sbi->s_ext_blocks, sbi->s_ext_extents,
2783 sbi->s_ext_blocks / sbi->s_ext_extents);
2784 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2785 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2790 /* FIXME!! we need to try to merge to left or right after zero-out */
2791 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2793 ext4_fsblk_t ee_pblock;
2794 unsigned int ee_len;
2797 ee_len = ext4_ext_get_actual_len(ex);
2798 ee_pblock = ext4_ext_pblock(ex);
2800 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2808 * used by extent splitting.
2810 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2812 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2813 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2816 * ext4_split_extent_at() splits an extent at given block.
2818 * @handle: the journal handle
2819 * @inode: the file inode
2820 * @path: the path to the extent
2821 * @split: the logical block where the extent is splitted.
2822 * @split_flags: indicates if the extent could be zeroout if split fails, and
2823 * the states(init or uninit) of new extents.
2824 * @flags: flags used to insert new extent to extent tree.
2827 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2828 * of which are deterimined by split_flag.
2830 * There are two cases:
2831 * a> the extent are splitted into two extent.
2832 * b> split is not needed, and just mark the extent.
2834 * return 0 on success.
2836 static int ext4_split_extent_at(handle_t *handle,
2837 struct inode *inode,
2838 struct ext4_ext_path *path,
2843 ext4_fsblk_t newblock;
2844 ext4_lblk_t ee_block;
2845 struct ext4_extent *ex, newex, orig_ex;
2846 struct ext4_extent *ex2 = NULL;
2847 unsigned int ee_len, depth;
2850 ext_debug("ext4_split_extents_at: inode %lu, logical"
2851 "block %llu\n", inode->i_ino, (unsigned long long)split);
2853 ext4_ext_show_leaf(inode, path);
2855 depth = ext_depth(inode);
2856 ex = path[depth].p_ext;
2857 ee_block = le32_to_cpu(ex->ee_block);
2858 ee_len = ext4_ext_get_actual_len(ex);
2859 newblock = split - ee_block + ext4_ext_pblock(ex);
2861 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2863 err = ext4_ext_get_access(handle, inode, path + depth);
2867 if (split == ee_block) {
2869 * case b: block @split is the block that the extent begins with
2870 * then we just change the state of the extent, and splitting
2873 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2874 ext4_ext_mark_uninitialized(ex);
2876 ext4_ext_mark_initialized(ex);
2878 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2879 ext4_ext_try_to_merge(inode, path, ex);
2881 err = ext4_ext_dirty(handle, inode, path + depth);
2886 memcpy(&orig_ex, ex, sizeof(orig_ex));
2887 ex->ee_len = cpu_to_le16(split - ee_block);
2888 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2889 ext4_ext_mark_uninitialized(ex);
2892 * path may lead to new leaf, not to original leaf any more
2893 * after ext4_ext_insert_extent() returns,
2895 err = ext4_ext_dirty(handle, inode, path + depth);
2897 goto fix_extent_len;
2900 ex2->ee_block = cpu_to_le32(split);
2901 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2902 ext4_ext_store_pblock(ex2, newblock);
2903 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2904 ext4_ext_mark_uninitialized(ex2);
2906 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2907 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2908 err = ext4_ext_zeroout(inode, &orig_ex);
2910 goto fix_extent_len;
2911 /* update the extent length and mark as initialized */
2912 ex->ee_len = cpu_to_le32(ee_len);
2913 ext4_ext_try_to_merge(inode, path, ex);
2914 err = ext4_ext_dirty(handle, inode, path + depth);
2917 goto fix_extent_len;
2920 ext4_ext_show_leaf(inode, path);
2924 ex->ee_len = orig_ex.ee_len;
2925 ext4_ext_dirty(handle, inode, path + depth);
2930 * ext4_split_extents() splits an extent and mark extent which is covered
2931 * by @map as split_flags indicates
2933 * It may result in splitting the extent into multiple extents (upto three)
2934 * There are three possibilities:
2935 * a> There is no split required
2936 * b> Splits in two extents: Split is happening at either end of the extent
2937 * c> Splits in three extents: Somone is splitting in middle of the extent
2940 static int ext4_split_extent(handle_t *handle,
2941 struct inode *inode,
2942 struct ext4_ext_path *path,
2943 struct ext4_map_blocks *map,
2947 ext4_lblk_t ee_block;
2948 struct ext4_extent *ex;
2949 unsigned int ee_len, depth;
2952 int split_flag1, flags1;
2954 depth = ext_depth(inode);
2955 ex = path[depth].p_ext;
2956 ee_block = le32_to_cpu(ex->ee_block);
2957 ee_len = ext4_ext_get_actual_len(ex);
2958 uninitialized = ext4_ext_is_uninitialized(ex);
2960 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2961 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2962 EXT4_EXT_MAY_ZEROOUT : 0;
2963 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2965 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2966 EXT4_EXT_MARK_UNINIT2;
2967 err = ext4_split_extent_at(handle, inode, path,
2968 map->m_lblk + map->m_len, split_flag1, flags1);
2973 ext4_ext_drop_refs(path);
2974 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2976 return PTR_ERR(path);
2978 if (map->m_lblk >= ee_block) {
2979 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2980 EXT4_EXT_MAY_ZEROOUT : 0;
2982 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2983 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2984 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2985 err = ext4_split_extent_at(handle, inode, path,
2986 map->m_lblk, split_flag1, flags);
2991 ext4_ext_show_leaf(inode, path);
2993 return err ? err : map->m_len;
2996 #define EXT4_EXT_ZERO_LEN 7
2998 * This function is called by ext4_ext_map_blocks() if someone tries to write
2999 * to an uninitialized extent. It may result in splitting the uninitialized
3000 * extent into multiple extents (up to three - one initialized and two
3002 * There are three possibilities:
3003 * a> There is no split required: Entire extent should be initialized
3004 * b> Splits in two extents: Write is happening at either end of the extent
3005 * c> Splits in three extents: Somone is writing in middle of the extent
3007 static int ext4_ext_convert_to_initialized(handle_t *handle,
3008 struct inode *inode,
3009 struct ext4_map_blocks *map,
3010 struct ext4_ext_path *path)
3012 struct ext4_map_blocks split_map;
3013 struct ext4_extent zero_ex;
3014 struct ext4_extent *ex;
3015 ext4_lblk_t ee_block, eof_block;
3016 unsigned int allocated, ee_len, depth;
3020 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3021 "block %llu, max_blocks %u\n", inode->i_ino,
3022 (unsigned long long)map->m_lblk, map->m_len);
3024 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3025 inode->i_sb->s_blocksize_bits;
3026 if (eof_block < map->m_lblk + map->m_len)
3027 eof_block = map->m_lblk + map->m_len;
3029 depth = ext_depth(inode);
3030 ex = path[depth].p_ext;
3031 ee_block = le32_to_cpu(ex->ee_block);
3032 ee_len = ext4_ext_get_actual_len(ex);
3033 allocated = ee_len - (map->m_lblk - ee_block);
3035 WARN_ON(map->m_lblk < ee_block);
3037 * It is safe to convert extent to initialized via explicit
3038 * zeroout only if extent is fully insde i_size or new_size.
3040 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3042 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3043 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3044 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3045 err = ext4_ext_zeroout(inode, ex);
3049 err = ext4_ext_get_access(handle, inode, path + depth);
3052 ext4_ext_mark_initialized(ex);
3053 ext4_ext_try_to_merge(inode, path, ex);
3054 err = ext4_ext_dirty(handle, inode, path + depth);
3060 * 1. split the extent into three extents.
3061 * 2. split the extent into two extents, zeroout the first half.
3062 * 3. split the extent into two extents, zeroout the second half.
3063 * 4. split the extent into two extents with out zeroout.
3065 split_map.m_lblk = map->m_lblk;
3066 split_map.m_len = map->m_len;
3068 if (allocated > map->m_len) {
3069 if (allocated <= EXT4_EXT_ZERO_LEN &&
3070 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3073 cpu_to_le32(map->m_lblk);
3074 zero_ex.ee_len = cpu_to_le16(allocated);
3075 ext4_ext_store_pblock(&zero_ex,
3076 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3077 err = ext4_ext_zeroout(inode, &zero_ex);
3080 split_map.m_lblk = map->m_lblk;
3081 split_map.m_len = allocated;
3082 } else if ((map->m_lblk - ee_block + map->m_len <
3083 EXT4_EXT_ZERO_LEN) &&
3084 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3086 if (map->m_lblk != ee_block) {
3087 zero_ex.ee_block = ex->ee_block;
3088 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3090 ext4_ext_store_pblock(&zero_ex,
3091 ext4_ext_pblock(ex));
3092 err = ext4_ext_zeroout(inode, &zero_ex);
3097 split_map.m_lblk = ee_block;
3098 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3099 allocated = map->m_len;
3103 allocated = ext4_split_extent(handle, inode, path,
3104 &split_map, split_flag, 0);
3109 return err ? err : allocated;
3113 * This function is called by ext4_ext_map_blocks() from
3114 * ext4_get_blocks_dio_write() when DIO to write
3115 * to an uninitialized extent.
3117 * Writing to an uninitialized extent may result in splitting the uninitialized
3118 * extent into multiple /initialized uninitialized extents (up to three)
3119 * There are three possibilities:
3120 * a> There is no split required: Entire extent should be uninitialized
3121 * b> Splits in two extents: Write is happening at either end of the extent
3122 * c> Splits in three extents: Somone is writing in middle of the extent
3124 * One of more index blocks maybe needed if the extent tree grow after
3125 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3126 * complete, we need to split the uninitialized extent before DIO submit
3127 * the IO. The uninitialized extent called at this time will be split
3128 * into three uninitialized extent(at most). After IO complete, the part
3129 * being filled will be convert to initialized by the end_io callback function
3130 * via ext4_convert_unwritten_extents().
3132 * Returns the size of uninitialized extent to be written on success.
3134 static int ext4_split_unwritten_extents(handle_t *handle,
3135 struct inode *inode,
3136 struct ext4_map_blocks *map,
3137 struct ext4_ext_path *path,
3140 ext4_lblk_t eof_block;
3141 ext4_lblk_t ee_block;
3142 struct ext4_extent *ex;
3143 unsigned int ee_len;
3144 int split_flag = 0, depth;
3146 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3147 "block %llu, max_blocks %u\n", inode->i_ino,
3148 (unsigned long long)map->m_lblk, map->m_len);
3150 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3151 inode->i_sb->s_blocksize_bits;
3152 if (eof_block < map->m_lblk + map->m_len)
3153 eof_block = map->m_lblk + map->m_len;
3155 * It is safe to convert extent to initialized via explicit
3156 * zeroout only if extent is fully insde i_size or new_size.
3158 depth = ext_depth(inode);
3159 ex = path[depth].p_ext;
3160 ee_block = le32_to_cpu(ex->ee_block);
3161 ee_len = ext4_ext_get_actual_len(ex);
3163 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3164 split_flag |= EXT4_EXT_MARK_UNINIT2;
3166 flags |= EXT4_GET_BLOCKS_PRE_IO;
3167 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3170 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3171 struct inode *inode,
3172 struct ext4_ext_path *path)
3174 struct ext4_extent *ex;
3178 depth = ext_depth(inode);
3179 ex = path[depth].p_ext;
3181 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3182 "block %llu, max_blocks %u\n", inode->i_ino,
3183 (unsigned long long)le32_to_cpu(ex->ee_block),
3184 ext4_ext_get_actual_len(ex));
3186 err = ext4_ext_get_access(handle, inode, path + depth);
3189 /* first mark the extent as initialized */
3190 ext4_ext_mark_initialized(ex);
3192 /* note: ext4_ext_correct_indexes() isn't needed here because
3193 * borders are not changed
3195 ext4_ext_try_to_merge(inode, path, ex);
3197 /* Mark modified extent as dirty */
3198 err = ext4_ext_dirty(handle, inode, path + depth);
3200 ext4_ext_show_leaf(inode, path);
3204 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3205 sector_t block, int count)
3208 for (i = 0; i < count; i++)
3209 unmap_underlying_metadata(bdev, block + i);
3213 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3215 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3217 struct ext4_ext_path *path,
3221 struct ext4_extent_header *eh;
3222 struct ext4_extent *last_ex;
3224 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3227 depth = ext_depth(inode);
3228 eh = path[depth].p_hdr;
3230 if (unlikely(!eh->eh_entries)) {
3231 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3232 "EOFBLOCKS_FL set");
3235 last_ex = EXT_LAST_EXTENT(eh);
3237 * We should clear the EOFBLOCKS_FL flag if we are writing the
3238 * last block in the last extent in the file. We test this by
3239 * first checking to see if the caller to
3240 * ext4_ext_get_blocks() was interested in the last block (or
3241 * a block beyond the last block) in the current extent. If
3242 * this turns out to be false, we can bail out from this
3243 * function immediately.
3245 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3246 ext4_ext_get_actual_len(last_ex))
3249 * If the caller does appear to be planning to write at or
3250 * beyond the end of the current extent, we then test to see
3251 * if the current extent is the last extent in the file, by
3252 * checking to make sure it was reached via the rightmost node
3253 * at each level of the tree.
3255 for (i = depth-1; i >= 0; i--)
3256 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3258 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3259 return ext4_mark_inode_dirty(handle, inode);
3263 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3265 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3266 * whether there are any buffers marked for delayed allocation. It returns '1'
3267 * on the first delalloc'ed buffer head found. If no buffer head in the given
3268 * range is marked for delalloc, it returns 0.
3269 * lblk_start should always be <= lblk_end.
3270 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3271 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3272 * block sooner). This is useful when blocks are truncated sequentially from
3273 * lblk_start towards lblk_end.
3275 static int ext4_find_delalloc_range(struct inode *inode,
3276 ext4_lblk_t lblk_start,
3277 ext4_lblk_t lblk_end,
3278 int search_hint_reverse)
3280 struct address_space *mapping = inode->i_mapping;
3281 struct buffer_head *head, *bh = NULL;
3283 ext4_lblk_t i, pg_lblk;
3286 /* reverse search wont work if fs block size is less than page size */
3287 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3288 search_hint_reverse = 0;
3290 if (search_hint_reverse)
3295 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3297 while ((i >= lblk_start) && (i <= lblk_end)) {
3298 page = find_get_page(mapping, index);
3302 if (!page_has_buffers(page))
3305 head = page_buffers(page);
3310 pg_lblk = index << (PAGE_CACHE_SHIFT -
3313 if (unlikely(pg_lblk < lblk_start)) {
3315 * This is possible when fs block size is less
3316 * than page size and our cluster starts/ends in
3317 * middle of the page. So we need to skip the
3318 * initial few blocks till we reach the 'lblk'
3324 /* Check if the buffer is delayed allocated and that it
3325 * is not yet mapped. (when da-buffers are mapped during
3326 * their writeout, their da_mapped bit is set.)
3328 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3329 page_cache_release(page);
3330 trace_ext4_find_delalloc_range(inode,
3331 lblk_start, lblk_end,
3332 search_hint_reverse,
3336 if (search_hint_reverse)
3340 } while ((i >= lblk_start) && (i <= lblk_end) &&
3341 ((bh = bh->b_this_page) != head));
3344 page_cache_release(page);
3346 * Move to next page. 'i' will be the first lblk in the next
3349 if (search_hint_reverse)
3353 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3356 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3357 search_hint_reverse, 0, 0);
3361 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3362 int search_hint_reverse)
3364 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3365 ext4_lblk_t lblk_start, lblk_end;
3366 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3367 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3369 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3370 search_hint_reverse);
3374 * Determines how many complete clusters (out of those specified by the 'map')
3375 * are under delalloc and were reserved quota for.
3376 * This function is called when we are writing out the blocks that were
3377 * originally written with their allocation delayed, but then the space was
3378 * allocated using fallocate() before the delayed allocation could be resolved.
3379 * The cases to look for are:
3380 * ('=' indicated delayed allocated blocks
3381 * '-' indicates non-delayed allocated blocks)
3382 * (a) partial clusters towards beginning and/or end outside of allocated range
3383 * are not delalloc'ed.
3385 * |----c---=|====c====|====c====|===-c----|
3386 * |++++++ allocated ++++++|
3387 * ==> 4 complete clusters in above example
3389 * (b) partial cluster (outside of allocated range) towards either end is
3390 * marked for delayed allocation. In this case, we will exclude that
3393 * |----====c========|========c========|
3394 * |++++++ allocated ++++++|
3395 * ==> 1 complete clusters in above example
3398 * |================c================|
3399 * |++++++ allocated ++++++|
3400 * ==> 0 complete clusters in above example
3402 * The ext4_da_update_reserve_space will be called only if we
3403 * determine here that there were some "entire" clusters that span
3404 * this 'allocated' range.
3405 * In the non-bigalloc case, this function will just end up returning num_blks
3406 * without ever calling ext4_find_delalloc_range.
3409 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3410 unsigned int num_blks)
3412 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3413 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3414 ext4_lblk_t lblk_from, lblk_to, c_offset;
3415 unsigned int allocated_clusters = 0;
3417 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3418 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3420 /* max possible clusters for this allocation */
3421 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3423 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3425 /* Check towards left side */
3426 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3428 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3429 lblk_to = lblk_from + c_offset - 1;
3431 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3432 allocated_clusters--;
3435 /* Now check towards right. */
3436 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3437 if (allocated_clusters && c_offset) {
3438 lblk_from = lblk_start + num_blks;
3439 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3441 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3442 allocated_clusters--;
3445 return allocated_clusters;
3449 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3450 struct ext4_map_blocks *map,
3451 struct ext4_ext_path *path, int flags,
3452 unsigned int allocated, ext4_fsblk_t newblock)
3456 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3458 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3459 "block %llu, max_blocks %u, flags %d, allocated %u",
3460 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3462 ext4_ext_show_leaf(inode, path);
3464 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3467 /* get_block() before submit the IO, split the extent */
3468 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3469 ret = ext4_split_unwritten_extents(handle, inode, map,
3472 * Flag the inode(non aio case) or end_io struct (aio case)
3473 * that this IO needs to conversion to written when IO is
3476 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3477 io->flag = EXT4_IO_END_UNWRITTEN;
3478 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3480 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3481 if (ext4_should_dioread_nolock(inode))
3482 map->m_flags |= EXT4_MAP_UNINIT;
3485 /* IO end_io complete, convert the filled extent to written */
3486 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3487 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3490 ext4_update_inode_fsync_trans(handle, inode, 1);
3491 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3497 /* buffered IO case */
3499 * repeat fallocate creation request
3500 * we already have an unwritten extent
3502 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3505 /* buffered READ or buffered write_begin() lookup */
3506 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3508 * We have blocks reserved already. We
3509 * return allocated blocks so that delalloc
3510 * won't do block reservation for us. But
3511 * the buffer head will be unmapped so that
3512 * a read from the block returns 0s.
3514 map->m_flags |= EXT4_MAP_UNWRITTEN;
3518 /* buffered write, writepage time, convert*/
3519 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3521 ext4_update_inode_fsync_trans(handle, inode, 1);
3522 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3534 map->m_flags |= EXT4_MAP_NEW;
3536 * if we allocated more blocks than requested
3537 * we need to make sure we unmap the extra block
3538 * allocated. The actual needed block will get
3539 * unmapped later when we find the buffer_head marked
3542 if (allocated > map->m_len) {
3543 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3544 newblock + map->m_len,
3545 allocated - map->m_len);
3546 allocated = map->m_len;
3550 * If we have done fallocate with the offset that is already
3551 * delayed allocated, we would have block reservation
3552 * and quota reservation done in the delayed write path.
3553 * But fallocate would have already updated quota and block
3554 * count for this offset. So cancel these reservation
3556 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3557 unsigned int reserved_clusters;
3558 reserved_clusters = get_reserved_cluster_alloc(inode,
3559 map->m_lblk, map->m_len);
3560 if (reserved_clusters)
3561 ext4_da_update_reserve_space(inode,
3567 map->m_flags |= EXT4_MAP_MAPPED;
3569 if (allocated > map->m_len)
3570 allocated = map->m_len;
3571 ext4_ext_show_leaf(inode, path);
3572 map->m_pblk = newblock;
3573 map->m_len = allocated;
3576 ext4_ext_drop_refs(path);
3579 return err ? err : allocated;
3583 * get_implied_cluster_alloc - check to see if the requested
3584 * allocation (in the map structure) overlaps with a cluster already
3585 * allocated in an extent.
3586 * @sb The filesystem superblock structure
3587 * @map The requested lblk->pblk mapping
3588 * @ex The extent structure which might contain an implied
3589 * cluster allocation
3591 * This function is called by ext4_ext_map_blocks() after we failed to
3592 * find blocks that were already in the inode's extent tree. Hence,
3593 * we know that the beginning of the requested region cannot overlap
3594 * the extent from the inode's extent tree. There are three cases we
3595 * want to catch. The first is this case:
3597 * |--- cluster # N--|
3598 * |--- extent ---| |---- requested region ---|
3601 * The second case that we need to test for is this one:
3603 * |--------- cluster # N ----------------|
3604 * |--- requested region --| |------- extent ----|
3605 * |=======================|
3607 * The third case is when the requested region lies between two extents
3608 * within the same cluster:
3609 * |------------- cluster # N-------------|
3610 * |----- ex -----| |---- ex_right ----|
3611 * |------ requested region ------|
3612 * |================|
3614 * In each of the above cases, we need to set the map->m_pblk and
3615 * map->m_len so it corresponds to the return the extent labelled as
3616 * "|====|" from cluster #N, since it is already in use for data in
3617 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3618 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3619 * as a new "allocated" block region. Otherwise, we will return 0 and
3620 * ext4_ext_map_blocks() will then allocate one or more new clusters
3621 * by calling ext4_mb_new_blocks().
3623 static int get_implied_cluster_alloc(struct super_block *sb,
3624 struct ext4_map_blocks *map,
3625 struct ext4_extent *ex,
3626 struct ext4_ext_path *path)
3628 struct ext4_sb_info *sbi = EXT4_SB(sb);
3629 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3630 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3631 ext4_lblk_t rr_cluster_start, rr_cluster_end;
3632 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3633 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3634 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3636 /* The extent passed in that we are trying to match */
3637 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3638 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3640 /* The requested region passed into ext4_map_blocks() */
3641 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3642 rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3644 if ((rr_cluster_start == ex_cluster_end) ||
3645 (rr_cluster_start == ex_cluster_start)) {
3646 if (rr_cluster_start == ex_cluster_end)
3647 ee_start += ee_len - 1;
3648 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3650 map->m_len = min(map->m_len,
3651 (unsigned) sbi->s_cluster_ratio - c_offset);
3653 * Check for and handle this case:
3655 * |--------- cluster # N-------------|
3656 * |------- extent ----|
3657 * |--- requested region ---|
3661 if (map->m_lblk < ee_block)
3662 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3665 * Check for the case where there is already another allocated
3666 * block to the right of 'ex' but before the end of the cluster.
3668 * |------------- cluster # N-------------|
3669 * |----- ex -----| |---- ex_right ----|
3670 * |------ requested region ------|
3671 * |================|
3673 if (map->m_lblk > ee_block) {
3674 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3675 map->m_len = min(map->m_len, next - map->m_lblk);
3678 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3682 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3688 * Block allocation/map/preallocation routine for extents based files
3691 * Need to be called with
3692 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3693 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3695 * return > 0, number of of blocks already mapped/allocated
3696 * if create == 0 and these are pre-allocated blocks
3697 * buffer head is unmapped
3698 * otherwise blocks are mapped
3700 * return = 0, if plain look up failed (blocks have not been allocated)
3701 * buffer head is unmapped
3703 * return < 0, error case.
3705 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3706 struct ext4_map_blocks *map, int flags)
3708 struct ext4_ext_path *path = NULL;
3709 struct ext4_extent newex, *ex, *ex2;
3710 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3711 ext4_fsblk_t newblock = 0;
3712 int free_on_err = 0, err = 0, depth, ret;
3713 unsigned int allocated = 0, offset = 0;
3714 unsigned int allocated_clusters = 0, reserved_clusters = 0;
3715 unsigned int punched_out = 0;
3716 unsigned int result = 0;
3717 struct ext4_allocation_request ar;
3718 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3719 ext4_lblk_t cluster_offset;
3720 struct ext4_map_blocks punch_map;
3722 ext_debug("blocks %u/%u requested for inode %lu\n",
3723 map->m_lblk, map->m_len, inode->i_ino);
3724 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3726 /* check in cache */
3727 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3728 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3729 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3730 if ((sbi->s_cluster_ratio > 1) &&
3731 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3732 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3734 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3736 * block isn't allocated yet and
3737 * user doesn't want to allocate it
3741 /* we should allocate requested block */
3743 /* block is already allocated */
3744 if (sbi->s_cluster_ratio > 1)
3745 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3746 newblock = map->m_lblk
3747 - le32_to_cpu(newex.ee_block)
3748 + ext4_ext_pblock(&newex);
3749 /* number of remaining blocks in the extent */
3750 allocated = ext4_ext_get_actual_len(&newex) -
3751 (map->m_lblk - le32_to_cpu(newex.ee_block));
3756 /* find extent for this block */
3757 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3759 err = PTR_ERR(path);
3764 depth = ext_depth(inode);
3767 * consistent leaf must not be empty;
3768 * this situation is possible, though, _during_ tree modification;
3769 * this is why assert can't be put in ext4_ext_find_extent()
3771 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3772 EXT4_ERROR_INODE(inode, "bad extent address "
3773 "lblock: %lu, depth: %d pblock %lld",
3774 (unsigned long) map->m_lblk, depth,
3775 path[depth].p_block);
3780 ex = path[depth].p_ext;
3782 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3783 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3784 unsigned short ee_len;
3787 * Uninitialized extents are treated as holes, except that
3788 * we split out initialized portions during a write.
3790 ee_len = ext4_ext_get_actual_len(ex);
3792 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3794 /* if found extent covers block, simply return it */
3795 if (in_range(map->m_lblk, ee_block, ee_len)) {
3796 ext4_fsblk_t partial_cluster = 0;
3798 newblock = map->m_lblk - ee_block + ee_start;
3799 /* number of remaining blocks in the extent */
3800 allocated = ee_len - (map->m_lblk - ee_block);
3801 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3802 ee_block, ee_len, newblock);
3804 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3806 * Do not put uninitialized extent
3809 if (!ext4_ext_is_uninitialized(ex)) {
3810 ext4_ext_put_in_cache(inode, ee_block,
3814 ret = ext4_ext_handle_uninitialized_extents(
3815 handle, inode, map, path, flags,
3816 allocated, newblock);
3821 * Punch out the map length, but only to the
3824 punched_out = allocated < map->m_len ?
3825 allocated : map->m_len;
3828 * Sense extents need to be converted to
3829 * uninitialized, they must fit in an
3830 * uninitialized extent
3832 if (punched_out > EXT_UNINIT_MAX_LEN)
3833 punched_out = EXT_UNINIT_MAX_LEN;
3835 punch_map.m_lblk = map->m_lblk;
3836 punch_map.m_pblk = newblock;
3837 punch_map.m_len = punched_out;
3838 punch_map.m_flags = 0;
3840 /* Check to see if the extent needs to be split */
3841 if (punch_map.m_len != ee_len ||
3842 punch_map.m_lblk != ee_block) {
3844 ret = ext4_split_extent(handle, inode,
3845 path, &punch_map, 0,
3846 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3847 EXT4_GET_BLOCKS_PRE_IO);
3854 * find extent for the block at
3855 * the start of the hole
3857 ext4_ext_drop_refs(path);
3860 path = ext4_ext_find_extent(inode,
3863 err = PTR_ERR(path);
3868 depth = ext_depth(inode);
3869 ex = path[depth].p_ext;
3870 ee_len = ext4_ext_get_actual_len(ex);
3871 ee_block = le32_to_cpu(ex->ee_block);
3872 ee_start = ext4_ext_pblock(ex);
3876 ext4_ext_mark_uninitialized(ex);
3878 ext4_ext_invalidate_cache(inode);
3880 err = ext4_ext_rm_leaf(handle, inode, path,
3881 &partial_cluster, map->m_lblk,
3882 map->m_lblk + punched_out);
3884 if (!err && path->p_hdr->eh_entries == 0) {
3886 * Punch hole freed all of this sub tree,
3887 * so we need to correct eh_depth
3889 err = ext4_ext_get_access(handle, inode, path);
3891 ext_inode_hdr(inode)->eh_depth = 0;
3892 ext_inode_hdr(inode)->eh_max =
3893 cpu_to_le16(ext4_ext_space_root(
3896 err = ext4_ext_dirty(
3897 handle, inode, path);
3905 if ((sbi->s_cluster_ratio > 1) &&
3906 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3907 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3910 * requested block isn't allocated yet;
3911 * we couldn't try to create block if create flag is zero
3913 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3915 * put just found gap into cache to speed up
3916 * subsequent requests
3918 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3923 * Okay, we need to do block allocation.
3925 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
3926 newex.ee_block = cpu_to_le32(map->m_lblk);
3927 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3930 * If we are doing bigalloc, check to see if the extent returned
3931 * by ext4_ext_find_extent() implies a cluster we can use.
3933 if (cluster_offset && ex &&
3934 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
3935 ar.len = allocated = map->m_len;
3936 newblock = map->m_pblk;
3937 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3938 goto got_allocated_blocks;
3941 /* find neighbour allocated blocks */
3942 ar.lleft = map->m_lblk;
3943 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3946 ar.lright = map->m_lblk;
3948 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
3952 /* Check if the extent after searching to the right implies a
3953 * cluster we can use. */
3954 if ((sbi->s_cluster_ratio > 1) && ex2 &&
3955 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
3956 ar.len = allocated = map->m_len;
3957 newblock = map->m_pblk;
3958 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3959 goto got_allocated_blocks;
3963 * See if request is beyond maximum number of blocks we can have in
3964 * a single extent. For an initialized extent this limit is
3965 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3966 * EXT_UNINIT_MAX_LEN.
3968 if (map->m_len > EXT_INIT_MAX_LEN &&
3969 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3970 map->m_len = EXT_INIT_MAX_LEN;
3971 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3972 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3973 map->m_len = EXT_UNINIT_MAX_LEN;
3975 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3976 newex.ee_len = cpu_to_le16(map->m_len);
3977 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
3979 allocated = ext4_ext_get_actual_len(&newex);
3981 allocated = map->m_len;
3983 /* allocate new block */
3985 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3986 ar.logical = map->m_lblk;
3988 * We calculate the offset from the beginning of the cluster
3989 * for the logical block number, since when we allocate a
3990 * physical cluster, the physical block should start at the
3991 * same offset from the beginning of the cluster. This is
3992 * needed so that future calls to get_implied_cluster_alloc()
3995 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
3996 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
3998 ar.logical -= offset;
3999 if (S_ISREG(inode->i_mode))
4000 ar.flags = EXT4_MB_HINT_DATA;
4002 /* disable in-core preallocation for non-regular files */
4004 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4005 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4006 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4009 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4010 ar.goal, newblock, allocated);
4012 allocated_clusters = ar.len;
4013 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4014 if (ar.len > allocated)
4017 got_allocated_blocks:
4018 /* try to insert new extent into found leaf and return */
4019 ext4_ext_store_pblock(&newex, newblock + offset);
4020 newex.ee_len = cpu_to_le16(ar.len);
4021 /* Mark uninitialized */
4022 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4023 ext4_ext_mark_uninitialized(&newex);
4025 * io_end structure was created for every IO write to an
4026 * uninitialized extent. To avoid unnecessary conversion,
4027 * here we flag the IO that really needs the conversion.
4028 * For non asycn direct IO case, flag the inode state
4029 * that we need to perform conversion when IO is done.
4031 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4032 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
4033 io->flag = EXT4_IO_END_UNWRITTEN;
4034 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
4036 ext4_set_inode_state(inode,
4037 EXT4_STATE_DIO_UNWRITTEN);
4039 if (ext4_should_dioread_nolock(inode))
4040 map->m_flags |= EXT4_MAP_UNINIT;
4043 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
4045 err = ext4_ext_insert_extent(handle, inode, path,
4047 if (err && free_on_err) {
4048 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4049 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4050 /* free data blocks we just allocated */
4051 /* not a good idea to call discard here directly,
4052 * but otherwise we'd need to call it every free() */
4053 ext4_discard_preallocations(inode);
4054 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4055 ext4_ext_get_actual_len(&newex), fb_flags);
4059 /* previous routine could use block we allocated */
4060 newblock = ext4_ext_pblock(&newex);
4061 allocated = ext4_ext_get_actual_len(&newex);
4062 if (allocated > map->m_len)
4063 allocated = map->m_len;
4064 map->m_flags |= EXT4_MAP_NEW;
4067 * Update reserved blocks/metadata blocks after successful
4068 * block allocation which had been deferred till now.
4070 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4072 * Check how many clusters we had reserved this allocted range.
4074 reserved_clusters = get_reserved_cluster_alloc(inode,
4075 map->m_lblk, allocated);
4076 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4077 if (reserved_clusters) {
4079 * We have clusters reserved for this range.
4080 * But since we are not doing actual allocation
4081 * and are simply using blocks from previously
4082 * allocated cluster, we should release the
4083 * reservation and not claim quota.
4085 ext4_da_update_reserve_space(inode,
4086 reserved_clusters, 0);
4089 BUG_ON(allocated_clusters < reserved_clusters);
4090 /* We will claim quota for all newly allocated blocks.*/
4091 ext4_da_update_reserve_space(inode, allocated_clusters,
4093 if (reserved_clusters < allocated_clusters) {
4094 struct ext4_inode_info *ei = EXT4_I(inode);
4095 int reservation = allocated_clusters -
4098 * It seems we claimed few clusters outside of
4099 * the range of this allocation. We should give
4100 * it back to the reservation pool. This can
4101 * happen in the following case:
4103 * * Suppose s_cluster_ratio is 4 (i.e., each
4104 * cluster has 4 blocks. Thus, the clusters
4105 * are [0-3],[4-7],[8-11]...
4106 * * First comes delayed allocation write for
4107 * logical blocks 10 & 11. Since there were no
4108 * previous delayed allocated blocks in the
4109 * range [8-11], we would reserve 1 cluster
4111 * * Next comes write for logical blocks 3 to 8.
4112 * In this case, we will reserve 2 clusters
4113 * (for [0-3] and [4-7]; and not for [8-11] as
4114 * that range has a delayed allocated blocks.
4115 * Thus total reserved clusters now becomes 3.
4116 * * Now, during the delayed allocation writeout
4117 * time, we will first write blocks [3-8] and
4118 * allocate 3 clusters for writing these
4119 * blocks. Also, we would claim all these
4120 * three clusters above.
4121 * * Now when we come here to writeout the
4122 * blocks [10-11], we would expect to claim
4123 * the reservation of 1 cluster we had made
4124 * (and we would claim it since there are no
4125 * more delayed allocated blocks in the range
4126 * [8-11]. But our reserved cluster count had
4127 * already gone to 0.
4129 * Thus, at the step 4 above when we determine
4130 * that there are still some unwritten delayed
4131 * allocated blocks outside of our current
4132 * block range, we should increment the
4133 * reserved clusters count so that when the
4134 * remaining blocks finally gets written, we
4137 dquot_reserve_block(inode,
4138 EXT4_C2B(sbi, reservation));
4139 spin_lock(&ei->i_block_reservation_lock);
4140 ei->i_reserved_data_blocks += reservation;
4141 spin_unlock(&ei->i_block_reservation_lock);
4147 * Cache the extent and update transaction to commit on fdatasync only
4148 * when it is _not_ an uninitialized extent.
4150 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4151 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4152 ext4_update_inode_fsync_trans(handle, inode, 1);
4154 ext4_update_inode_fsync_trans(handle, inode, 0);
4156 if (allocated > map->m_len)
4157 allocated = map->m_len;
4158 ext4_ext_show_leaf(inode, path);
4159 map->m_flags |= EXT4_MAP_MAPPED;
4160 map->m_pblk = newblock;
4161 map->m_len = allocated;
4164 ext4_ext_drop_refs(path);
4167 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4168 newblock, map->m_len, err ? err : allocated);
4170 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4171 punched_out : allocated;
4173 return err ? err : result;
4176 void ext4_ext_truncate(struct inode *inode)
4178 struct address_space *mapping = inode->i_mapping;
4179 struct super_block *sb = inode->i_sb;
4180 ext4_lblk_t last_block;
4186 * finish any pending end_io work so we won't run the risk of
4187 * converting any truncated blocks to initialized later
4189 ext4_flush_completed_IO(inode);
4192 * probably first extent we're gonna free will be last in block
4194 err = ext4_writepage_trans_blocks(inode);
4195 handle = ext4_journal_start(inode, err);
4199 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4200 page_len = PAGE_CACHE_SIZE -
4201 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4203 err = ext4_discard_partial_page_buffers(handle,
4204 mapping, inode->i_size, page_len, 0);
4210 if (ext4_orphan_add(handle, inode))
4213 down_write(&EXT4_I(inode)->i_data_sem);
4214 ext4_ext_invalidate_cache(inode);
4216 ext4_discard_preallocations(inode);
4219 * TODO: optimization is possible here.
4220 * Probably we need not scan at all,
4221 * because page truncation is enough.
4224 /* we have to know where to truncate from in crash case */
4225 EXT4_I(inode)->i_disksize = inode->i_size;
4226 ext4_mark_inode_dirty(handle, inode);
4228 last_block = (inode->i_size + sb->s_blocksize - 1)
4229 >> EXT4_BLOCK_SIZE_BITS(sb);
4230 err = ext4_ext_remove_space(inode, last_block);
4232 /* In a multi-transaction truncate, we only make the final
4233 * transaction synchronous.
4236 ext4_handle_sync(handle);
4238 up_write(&EXT4_I(inode)->i_data_sem);
4242 * If this was a simple ftruncate() and the file will remain alive,
4243 * then we need to clear up the orphan record which we created above.
4244 * However, if this was a real unlink then we were called by
4245 * ext4_delete_inode(), and we allow that function to clean up the
4246 * orphan info for us.
4249 ext4_orphan_del(handle, inode);
4251 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4252 ext4_mark_inode_dirty(handle, inode);
4253 ext4_journal_stop(handle);
4256 static void ext4_falloc_update_inode(struct inode *inode,
4257 int mode, loff_t new_size, int update_ctime)
4259 struct timespec now;
4262 now = current_fs_time(inode->i_sb);
4263 if (!timespec_equal(&inode->i_ctime, &now))
4264 inode->i_ctime = now;
4267 * Update only when preallocation was requested beyond
4270 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4271 if (new_size > i_size_read(inode))
4272 i_size_write(inode, new_size);
4273 if (new_size > EXT4_I(inode)->i_disksize)
4274 ext4_update_i_disksize(inode, new_size);
4277 * Mark that we allocate beyond EOF so the subsequent truncate
4278 * can proceed even if the new size is the same as i_size.
4280 if (new_size > i_size_read(inode))
4281 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4287 * preallocate space for a file. This implements ext4's fallocate file
4288 * operation, which gets called from sys_fallocate system call.
4289 * For block-mapped files, posix_fallocate should fall back to the method
4290 * of writing zeroes to the required new blocks (the same behavior which is
4291 * expected for file systems which do not support fallocate() system call).
4293 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4295 struct inode *inode = file->f_path.dentry->d_inode;
4298 unsigned int max_blocks;
4302 struct ext4_map_blocks map;
4303 unsigned int credits, blkbits = inode->i_blkbits;
4306 * currently supporting (pre)allocate mode for extent-based
4309 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4312 /* Return error if mode is not supported */
4313 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4316 if (mode & FALLOC_FL_PUNCH_HOLE)
4317 return ext4_punch_hole(file, offset, len);
4319 trace_ext4_fallocate_enter(inode, offset, len, mode);
4320 map.m_lblk = offset >> blkbits;
4322 * We can't just convert len to max_blocks because
4323 * If blocksize = 4096 offset = 3072 and len = 2048
4325 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4328 * credits to insert 1 extent into extent tree
4330 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4331 mutex_lock(&inode->i_mutex);
4332 ret = inode_newsize_ok(inode, (len + offset));
4334 mutex_unlock(&inode->i_mutex);
4335 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4339 while (ret >= 0 && ret < max_blocks) {
4340 map.m_lblk = map.m_lblk + ret;
4341 map.m_len = max_blocks = max_blocks - ret;
4342 handle = ext4_journal_start(inode, credits);
4343 if (IS_ERR(handle)) {
4344 ret = PTR_ERR(handle);
4347 ret = ext4_map_blocks(handle, inode, &map,
4348 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
4349 EXT4_GET_BLOCKS_NO_NORMALIZE);
4353 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4354 "returned error inode#%lu, block=%u, "
4355 "max_blocks=%u", __func__,
4356 inode->i_ino, map.m_lblk, max_blocks);
4358 ext4_mark_inode_dirty(handle, inode);
4359 ret2 = ext4_journal_stop(handle);
4362 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4363 blkbits) >> blkbits))
4364 new_size = offset + len;
4366 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4368 ext4_falloc_update_inode(inode, mode, new_size,
4369 (map.m_flags & EXT4_MAP_NEW));
4370 ext4_mark_inode_dirty(handle, inode);
4371 ret2 = ext4_journal_stop(handle);
4375 if (ret == -ENOSPC &&
4376 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4380 mutex_unlock(&inode->i_mutex);
4381 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4382 ret > 0 ? ret2 : ret);
4383 return ret > 0 ? ret2 : ret;
4387 * This function convert a range of blocks to written extents
4388 * The caller of this function will pass the start offset and the size.
4389 * all unwritten extents within this range will be converted to
4392 * This function is called from the direct IO end io call back
4393 * function, to convert the fallocated extents after IO is completed.
4394 * Returns 0 on success.
4396 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4400 unsigned int max_blocks;
4403 struct ext4_map_blocks map;
4404 unsigned int credits, blkbits = inode->i_blkbits;
4406 map.m_lblk = offset >> blkbits;
4408 * We can't just convert len to max_blocks because
4409 * If blocksize = 4096 offset = 3072 and len = 2048
4411 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4414 * credits to insert 1 extent into extent tree
4416 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4417 while (ret >= 0 && ret < max_blocks) {
4419 map.m_len = (max_blocks -= ret);
4420 handle = ext4_journal_start(inode, credits);
4421 if (IS_ERR(handle)) {
4422 ret = PTR_ERR(handle);
4425 ret = ext4_map_blocks(handle, inode, &map,
4426 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4429 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4430 "returned error inode#%lu, block=%u, "
4431 "max_blocks=%u", __func__,
4432 inode->i_ino, map.m_lblk, map.m_len);
4434 ext4_mark_inode_dirty(handle, inode);
4435 ret2 = ext4_journal_stop(handle);
4436 if (ret <= 0 || ret2 )
4439 return ret > 0 ? ret2 : ret;
4443 * Callback function called for each extent to gather FIEMAP information.
4445 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4446 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4454 struct fiemap_extent_info *fieinfo = data;
4455 unsigned char blksize_bits;
4457 blksize_bits = inode->i_sb->s_blocksize_bits;
4458 logical = (__u64)newex->ec_block << blksize_bits;
4460 if (newex->ec_start == 0) {
4462 * No extent in extent-tree contains block @newex->ec_start,
4463 * then the block may stay in 1)a hole or 2)delayed-extent.
4465 * Holes or delayed-extents are processed as follows.
4466 * 1. lookup dirty pages with specified range in pagecache.
4467 * If no page is got, then there is no delayed-extent and
4468 * return with EXT_CONTINUE.
4469 * 2. find the 1st mapped buffer,
4470 * 3. check if the mapped buffer is both in the request range
4471 * and a delayed buffer. If not, there is no delayed-extent,
4473 * 4. a delayed-extent is found, the extent will be collected.
4475 ext4_lblk_t end = 0;
4476 pgoff_t last_offset;
4479 pgoff_t start_index = 0;
4480 struct page **pages = NULL;
4481 struct buffer_head *bh = NULL;
4482 struct buffer_head *head = NULL;
4483 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4485 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4489 offset = logical >> PAGE_SHIFT;
4491 last_offset = offset;
4493 ret = find_get_pages_tag(inode->i_mapping, &offset,
4494 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4496 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4497 /* First time, try to find a mapped buffer. */
4500 for (index = 0; index < ret; index++)
4501 page_cache_release(pages[index]);
4504 return EXT_CONTINUE;
4509 /* Try to find the 1st mapped buffer. */
4510 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4512 if (!page_has_buffers(pages[index]))
4514 head = page_buffers(pages[index]);
4521 if (end >= newex->ec_block +
4523 /* The buffer is out of
4524 * the request range.
4528 if (buffer_mapped(bh) &&
4529 end >= newex->ec_block) {
4530 start_index = index - 1;
4531 /* get the 1st mapped buffer. */
4532 goto found_mapped_buffer;
4535 bh = bh->b_this_page;
4537 } while (bh != head);
4539 /* No mapped buffer in the range found in this page,
4540 * We need to look up next page.
4543 /* There is no page left, but we need to limit
4546 newex->ec_len = end - newex->ec_block;
4551 /*Find contiguous delayed buffers. */
4552 if (ret > 0 && pages[0]->index == last_offset)
4553 head = page_buffers(pages[0]);
4559 found_mapped_buffer:
4560 if (bh != NULL && buffer_delay(bh)) {
4561 /* 1st or contiguous delayed buffer found. */
4562 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4564 * 1st delayed buffer found, record
4565 * the start of extent.
4567 flags |= FIEMAP_EXTENT_DELALLOC;
4568 newex->ec_block = end;
4569 logical = (__u64)end << blksize_bits;
4571 /* Find contiguous delayed buffers. */
4573 if (!buffer_delay(bh))
4574 goto found_delayed_extent;
4575 bh = bh->b_this_page;
4577 } while (bh != head);
4579 for (; index < ret; index++) {
4580 if (!page_has_buffers(pages[index])) {
4584 head = page_buffers(pages[index]);
4590 if (pages[index]->index !=
4591 pages[start_index]->index + index
4593 /* Blocks are not contiguous. */
4599 if (!buffer_delay(bh))
4600 /* Delayed-extent ends. */
4601 goto found_delayed_extent;
4602 bh = bh->b_this_page;
4604 } while (bh != head);
4606 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4610 found_delayed_extent:
4611 newex->ec_len = min(end - newex->ec_block,
4612 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4613 if (ret == nr_pages && bh != NULL &&
4614 newex->ec_len < EXT_INIT_MAX_LEN &&
4616 /* Have not collected an extent and continue. */
4617 for (index = 0; index < ret; index++)
4618 page_cache_release(pages[index]);
4622 for (index = 0; index < ret; index++)
4623 page_cache_release(pages[index]);
4627 physical = (__u64)newex->ec_start << blksize_bits;
4628 length = (__u64)newex->ec_len << blksize_bits;
4630 if (ex && ext4_ext_is_uninitialized(ex))
4631 flags |= FIEMAP_EXTENT_UNWRITTEN;
4633 if (next == EXT_MAX_BLOCKS)
4634 flags |= FIEMAP_EXTENT_LAST;
4636 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4642 return EXT_CONTINUE;
4644 /* fiemap flags we can handle specified here */
4645 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4647 static int ext4_xattr_fiemap(struct inode *inode,
4648 struct fiemap_extent_info *fieinfo)
4652 __u32 flags = FIEMAP_EXTENT_LAST;
4653 int blockbits = inode->i_sb->s_blocksize_bits;
4657 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4658 struct ext4_iloc iloc;
4659 int offset; /* offset of xattr in inode */
4661 error = ext4_get_inode_loc(inode, &iloc);
4664 physical = iloc.bh->b_blocknr << blockbits;
4665 offset = EXT4_GOOD_OLD_INODE_SIZE +
4666 EXT4_I(inode)->i_extra_isize;
4668 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4669 flags |= FIEMAP_EXTENT_DATA_INLINE;
4671 } else { /* external block */
4672 physical = EXT4_I(inode)->i_file_acl << blockbits;
4673 length = inode->i_sb->s_blocksize;
4677 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4679 return (error < 0 ? error : 0);
4683 * ext4_ext_punch_hole
4685 * Punches a hole of "length" bytes in a file starting
4688 * @inode: The inode of the file to punch a hole in
4689 * @offset: The starting byte offset of the hole
4690 * @length: The length of the hole
4692 * Returns the number of blocks removed or negative on err
4694 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4696 struct inode *inode = file->f_path.dentry->d_inode;
4697 struct super_block *sb = inode->i_sb;
4698 struct ext4_ext_cache cache_ex;
4699 ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4700 struct address_space *mapping = inode->i_mapping;
4701 struct ext4_map_blocks map;
4703 loff_t first_page, last_page, page_len;
4704 loff_t first_page_offset, last_page_offset;
4705 int ret, credits, blocks_released, err = 0;
4707 /* No need to punch hole beyond i_size */
4708 if (offset >= inode->i_size)
4712 * If the hole extends beyond i_size, set the hole
4713 * to end after the page that contains i_size
4715 if (offset + length > inode->i_size) {
4716 length = inode->i_size +
4717 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4721 first_block = (offset + sb->s_blocksize - 1) >>
4722 EXT4_BLOCK_SIZE_BITS(sb);
4723 last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4725 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4726 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4728 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4729 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4732 * Write out all dirty pages to avoid race conditions
4733 * Then release them.
4735 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4736 err = filemap_write_and_wait_range(mapping,
4737 offset, offset + length - 1);
4743 /* Now release the pages */
4744 if (last_page_offset > first_page_offset) {
4745 truncate_inode_pages_range(mapping, first_page_offset,
4746 last_page_offset-1);
4749 /* finish any pending end_io work */
4750 ext4_flush_completed_IO(inode);
4752 credits = ext4_writepage_trans_blocks(inode);
4753 handle = ext4_journal_start(inode, credits);
4755 return PTR_ERR(handle);
4757 err = ext4_orphan_add(handle, inode);
4762 * Now we need to zero out the non-page-aligned data in the
4763 * pages at the start and tail of the hole, and unmap the buffer
4764 * heads for the block aligned regions of the page that were
4765 * completely zeroed.
4767 if (first_page > last_page) {
4769 * If the file space being truncated is contained within a page
4770 * just zero out and unmap the middle of that page
4772 err = ext4_discard_partial_page_buffers(handle,
4773 mapping, offset, length, 0);
4779 * zero out and unmap the partial page that contains
4780 * the start of the hole
4782 page_len = first_page_offset - offset;
4784 err = ext4_discard_partial_page_buffers(handle, mapping,
4785 offset, page_len, 0);
4791 * zero out and unmap the partial page that contains
4792 * the end of the hole
4794 page_len = offset + length - last_page_offset;
4796 err = ext4_discard_partial_page_buffers(handle, mapping,
4797 last_page_offset, page_len, 0);
4805 * If i_size is contained in the last page, we need to
4806 * unmap and zero the partial page after i_size
4808 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4809 inode->i_size % PAGE_CACHE_SIZE != 0) {
4811 page_len = PAGE_CACHE_SIZE -
4812 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4815 err = ext4_discard_partial_page_buffers(handle,
4816 mapping, inode->i_size, page_len, 0);
4823 /* If there are no blocks to remove, return now */
4824 if (first_block >= last_block)
4827 down_write(&EXT4_I(inode)->i_data_sem);
4828 ext4_ext_invalidate_cache(inode);
4829 ext4_discard_preallocations(inode);
4832 * Loop over all the blocks and identify blocks
4833 * that need to be punched out
4835 iblock = first_block;
4836 blocks_released = 0;
4837 while (iblock < last_block) {
4838 max_blocks = last_block - iblock;
4840 memset(&map, 0, sizeof(map));
4841 map.m_lblk = iblock;
4842 map.m_len = max_blocks;
4843 ret = ext4_ext_map_blocks(handle, inode, &map,
4844 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4847 blocks_released += ret;
4849 } else if (ret == 0) {
4851 * If map blocks could not find the block,
4852 * then it is in a hole. If the hole was
4853 * not already cached, then map blocks should
4854 * put it in the cache. So we can get the hole
4857 memset(&cache_ex, 0, sizeof(cache_ex));
4858 if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4859 !cache_ex.ec_start) {
4861 /* The hole is cached */
4862 num_blocks = cache_ex.ec_block +
4863 cache_ex.ec_len - iblock;
4866 /* The block could not be identified */
4871 /* Map blocks error */
4876 if (num_blocks == 0) {
4877 /* This condition should never happen */
4878 ext_debug("Block lookup failed");
4883 iblock += num_blocks;
4886 if (blocks_released > 0) {
4887 ext4_ext_invalidate_cache(inode);
4888 ext4_discard_preallocations(inode);
4892 ext4_handle_sync(handle);
4894 up_write(&EXT4_I(inode)->i_data_sem);
4897 ext4_orphan_del(handle, inode);
4898 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4899 ext4_mark_inode_dirty(handle, inode);
4900 ext4_journal_stop(handle);
4903 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4904 __u64 start, __u64 len)
4906 ext4_lblk_t start_blk;
4909 /* fallback to generic here if not in extents fmt */
4910 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4911 return generic_block_fiemap(inode, fieinfo, start, len,
4914 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4917 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4918 error = ext4_xattr_fiemap(inode, fieinfo);
4920 ext4_lblk_t len_blks;
4923 start_blk = start >> inode->i_sb->s_blocksize_bits;
4924 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4925 if (last_blk >= EXT_MAX_BLOCKS)
4926 last_blk = EXT_MAX_BLOCKS-1;
4927 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4930 * Walk the extent tree gathering extent information.
4931 * ext4_ext_fiemap_cb will push extents back to user.
4933 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4934 ext4_ext_fiemap_cb, fieinfo);