2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
49 static int ext4_split_extent(handle_t *handle,
51 struct ext4_ext_path *path,
52 struct ext4_map_blocks *map,
56 static int ext4_ext_truncate_extend_restart(handle_t *handle,
62 if (!ext4_handle_valid(handle))
64 if (handle->h_buffer_credits > needed)
66 err = ext4_journal_extend(handle, needed);
69 err = ext4_truncate_restart_trans(handle, inode, needed);
81 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
82 struct ext4_ext_path *path)
85 /* path points to block */
86 return ext4_journal_get_write_access(handle, path->p_bh);
88 /* path points to leaf/index in inode body */
89 /* we use in-core data, no need to protect them */
99 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
100 struct ext4_ext_path *path)
104 /* path points to block */
105 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
107 /* path points to leaf/index in inode body */
108 err = ext4_mark_inode_dirty(handle, inode);
113 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
114 struct ext4_ext_path *path,
120 struct ext4_extent *ex;
121 depth = path->p_depth;
124 * Try to predict block placement assuming that we are
125 * filling in a file which will eventually be
126 * non-sparse --- i.e., in the case of libbfd writing
127 * an ELF object sections out-of-order but in a way
128 * the eventually results in a contiguous object or
129 * executable file, or some database extending a table
130 * space file. However, this is actually somewhat
131 * non-ideal if we are writing a sparse file such as
132 * qemu or KVM writing a raw image file that is going
133 * to stay fairly sparse, since it will end up
134 * fragmenting the file system's free space. Maybe we
135 * should have some hueristics or some way to allow
136 * userspace to pass a hint to file system,
137 * especially if the latter case turns out to be
140 ex = path[depth].p_ext;
142 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
143 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
145 if (block > ext_block)
146 return ext_pblk + (block - ext_block);
148 return ext_pblk - (ext_block - block);
151 /* it looks like index is empty;
152 * try to find starting block from index itself */
153 if (path[depth].p_bh)
154 return path[depth].p_bh->b_blocknr;
157 /* OK. use inode's group */
158 return ext4_inode_to_goal_block(inode);
162 * Allocation for a meta data block
165 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
166 struct ext4_ext_path *path,
167 struct ext4_extent *ex, int *err, unsigned int flags)
169 ext4_fsblk_t goal, newblock;
171 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
172 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
177 static inline int ext4_ext_space_block(struct inode *inode, int check)
181 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
182 / sizeof(struct ext4_extent);
184 #ifdef AGGRESSIVE_TEST
192 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
196 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
197 / sizeof(struct ext4_extent_idx);
199 #ifdef AGGRESSIVE_TEST
207 static inline int ext4_ext_space_root(struct inode *inode, int check)
211 size = sizeof(EXT4_I(inode)->i_data);
212 size -= sizeof(struct ext4_extent_header);
213 size /= sizeof(struct ext4_extent);
215 #ifdef AGGRESSIVE_TEST
223 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
227 size = sizeof(EXT4_I(inode)->i_data);
228 size -= sizeof(struct ext4_extent_header);
229 size /= sizeof(struct ext4_extent_idx);
231 #ifdef AGGRESSIVE_TEST
240 * Calculate the number of metadata blocks needed
241 * to allocate @blocks
242 * Worse case is one block per extent
244 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
246 struct ext4_inode_info *ei = EXT4_I(inode);
249 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
250 / sizeof(struct ext4_extent_idx));
253 * If the new delayed allocation block is contiguous with the
254 * previous da block, it can share index blocks with the
255 * previous block, so we only need to allocate a new index
256 * block every idxs leaf blocks. At ldxs**2 blocks, we need
257 * an additional index block, and at ldxs**3 blocks, yet
258 * another index blocks.
260 if (ei->i_da_metadata_calc_len &&
261 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
262 if ((ei->i_da_metadata_calc_len % idxs) == 0)
264 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
266 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
268 ei->i_da_metadata_calc_len = 0;
270 ei->i_da_metadata_calc_len++;
271 ei->i_da_metadata_calc_last_lblock++;
276 * In the worst case we need a new set of index blocks at
277 * every level of the inode's extent tree.
279 ei->i_da_metadata_calc_len = 1;
280 ei->i_da_metadata_calc_last_lblock = lblock;
281 return ext_depth(inode) + 1;
285 ext4_ext_max_entries(struct inode *inode, int depth)
289 if (depth == ext_depth(inode)) {
291 max = ext4_ext_space_root(inode, 1);
293 max = ext4_ext_space_root_idx(inode, 1);
296 max = ext4_ext_space_block(inode, 1);
298 max = ext4_ext_space_block_idx(inode, 1);
304 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
306 ext4_fsblk_t block = ext4_ext_pblock(ext);
307 int len = ext4_ext_get_actual_len(ext);
309 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
312 static int ext4_valid_extent_idx(struct inode *inode,
313 struct ext4_extent_idx *ext_idx)
315 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
317 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
320 static int ext4_valid_extent_entries(struct inode *inode,
321 struct ext4_extent_header *eh,
324 struct ext4_extent *ext;
325 struct ext4_extent_idx *ext_idx;
326 unsigned short entries;
327 if (eh->eh_entries == 0)
330 entries = le16_to_cpu(eh->eh_entries);
334 ext = EXT_FIRST_EXTENT(eh);
336 if (!ext4_valid_extent(inode, ext))
342 ext_idx = EXT_FIRST_INDEX(eh);
344 if (!ext4_valid_extent_idx(inode, ext_idx))
353 static int __ext4_ext_check(const char *function, unsigned int line,
354 struct inode *inode, struct ext4_extent_header *eh,
357 const char *error_msg;
360 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
361 error_msg = "invalid magic";
364 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
365 error_msg = "unexpected eh_depth";
368 if (unlikely(eh->eh_max == 0)) {
369 error_msg = "invalid eh_max";
372 max = ext4_ext_max_entries(inode, depth);
373 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
374 error_msg = "too large eh_max";
377 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
378 error_msg = "invalid eh_entries";
381 if (!ext4_valid_extent_entries(inode, eh, depth)) {
382 error_msg = "invalid extent entries";
388 ext4_error_inode(inode, function, line, 0,
389 "bad header/extent: %s - magic %x, "
390 "entries %u, max %u(%u), depth %u(%u)",
391 error_msg, le16_to_cpu(eh->eh_magic),
392 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
393 max, le16_to_cpu(eh->eh_depth), depth);
398 #define ext4_ext_check(inode, eh, depth) \
399 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
401 int ext4_ext_check_inode(struct inode *inode)
403 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
407 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
409 int k, l = path->p_depth;
412 for (k = 0; k <= l; k++, path++) {
414 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
415 ext4_idx_pblock(path->p_idx));
416 } else if (path->p_ext) {
417 ext_debug(" %d:[%d]%d:%llu ",
418 le32_to_cpu(path->p_ext->ee_block),
419 ext4_ext_is_uninitialized(path->p_ext),
420 ext4_ext_get_actual_len(path->p_ext),
421 ext4_ext_pblock(path->p_ext));
428 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
430 int depth = ext_depth(inode);
431 struct ext4_extent_header *eh;
432 struct ext4_extent *ex;
438 eh = path[depth].p_hdr;
439 ex = EXT_FIRST_EXTENT(eh);
441 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
443 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
444 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
445 ext4_ext_is_uninitialized(ex),
446 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
451 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
452 ext4_fsblk_t newblock, int level)
454 int depth = ext_depth(inode);
455 struct ext4_extent *ex;
457 if (depth != level) {
458 struct ext4_extent_idx *idx;
459 idx = path[level].p_idx;
460 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
461 ext_debug("%d: move %d:%llu in new index %llu\n", level,
462 le32_to_cpu(idx->ei_block),
463 ext4_idx_pblock(idx),
471 ex = path[depth].p_ext;
472 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
473 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
474 le32_to_cpu(ex->ee_block),
476 ext4_ext_is_uninitialized(ex),
477 ext4_ext_get_actual_len(ex),
484 #define ext4_ext_show_path(inode, path)
485 #define ext4_ext_show_leaf(inode, path)
486 #define ext4_ext_show_move(inode, path, newblock, level)
489 void ext4_ext_drop_refs(struct ext4_ext_path *path)
491 int depth = path->p_depth;
494 for (i = 0; i <= depth; i++, path++)
502 * ext4_ext_binsearch_idx:
503 * binary search for the closest index of the given block
504 * the header must be checked before calling this
507 ext4_ext_binsearch_idx(struct inode *inode,
508 struct ext4_ext_path *path, ext4_lblk_t block)
510 struct ext4_extent_header *eh = path->p_hdr;
511 struct ext4_extent_idx *r, *l, *m;
514 ext_debug("binsearch for %u(idx): ", block);
516 l = EXT_FIRST_INDEX(eh) + 1;
517 r = EXT_LAST_INDEX(eh);
520 if (block < le32_to_cpu(m->ei_block))
524 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
525 m, le32_to_cpu(m->ei_block),
526 r, le32_to_cpu(r->ei_block));
530 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
531 ext4_idx_pblock(path->p_idx));
533 #ifdef CHECK_BINSEARCH
535 struct ext4_extent_idx *chix, *ix;
538 chix = ix = EXT_FIRST_INDEX(eh);
539 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
541 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
542 printk(KERN_DEBUG "k=%d, ix=0x%p, "
544 ix, EXT_FIRST_INDEX(eh));
545 printk(KERN_DEBUG "%u <= %u\n",
546 le32_to_cpu(ix->ei_block),
547 le32_to_cpu(ix[-1].ei_block));
549 BUG_ON(k && le32_to_cpu(ix->ei_block)
550 <= le32_to_cpu(ix[-1].ei_block));
551 if (block < le32_to_cpu(ix->ei_block))
555 BUG_ON(chix != path->p_idx);
562 * ext4_ext_binsearch:
563 * binary search for closest extent of the given block
564 * the header must be checked before calling this
567 ext4_ext_binsearch(struct inode *inode,
568 struct ext4_ext_path *path, ext4_lblk_t block)
570 struct ext4_extent_header *eh = path->p_hdr;
571 struct ext4_extent *r, *l, *m;
573 if (eh->eh_entries == 0) {
575 * this leaf is empty:
576 * we get such a leaf in split/add case
581 ext_debug("binsearch for %u: ", block);
583 l = EXT_FIRST_EXTENT(eh) + 1;
584 r = EXT_LAST_EXTENT(eh);
588 if (block < le32_to_cpu(m->ee_block))
592 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
593 m, le32_to_cpu(m->ee_block),
594 r, le32_to_cpu(r->ee_block));
598 ext_debug(" -> %d:%llu:[%d]%d ",
599 le32_to_cpu(path->p_ext->ee_block),
600 ext4_ext_pblock(path->p_ext),
601 ext4_ext_is_uninitialized(path->p_ext),
602 ext4_ext_get_actual_len(path->p_ext));
604 #ifdef CHECK_BINSEARCH
606 struct ext4_extent *chex, *ex;
609 chex = ex = EXT_FIRST_EXTENT(eh);
610 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
611 BUG_ON(k && le32_to_cpu(ex->ee_block)
612 <= le32_to_cpu(ex[-1].ee_block));
613 if (block < le32_to_cpu(ex->ee_block))
617 BUG_ON(chex != path->p_ext);
623 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
625 struct ext4_extent_header *eh;
627 eh = ext_inode_hdr(inode);
630 eh->eh_magic = EXT4_EXT_MAGIC;
631 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
632 ext4_mark_inode_dirty(handle, inode);
633 ext4_ext_invalidate_cache(inode);
637 struct ext4_ext_path *
638 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
639 struct ext4_ext_path *path)
641 struct ext4_extent_header *eh;
642 struct buffer_head *bh;
643 short int depth, i, ppos = 0, alloc = 0;
645 eh = ext_inode_hdr(inode);
646 depth = ext_depth(inode);
648 /* account possible depth increase */
650 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
653 return ERR_PTR(-ENOMEM);
660 /* walk through the tree */
662 int need_to_validate = 0;
664 ext_debug("depth %d: num %d, max %d\n",
665 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
667 ext4_ext_binsearch_idx(inode, path + ppos, block);
668 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
669 path[ppos].p_depth = i;
670 path[ppos].p_ext = NULL;
672 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
675 if (!bh_uptodate_or_lock(bh)) {
676 trace_ext4_ext_load_extent(inode, block,
678 if (bh_submit_read(bh) < 0) {
682 /* validate the extent entries */
683 need_to_validate = 1;
685 eh = ext_block_hdr(bh);
687 if (unlikely(ppos > depth)) {
689 EXT4_ERROR_INODE(inode,
690 "ppos %d > depth %d", ppos, depth);
693 path[ppos].p_bh = bh;
694 path[ppos].p_hdr = eh;
697 if (need_to_validate && ext4_ext_check(inode, eh, i))
701 path[ppos].p_depth = i;
702 path[ppos].p_ext = NULL;
703 path[ppos].p_idx = NULL;
706 ext4_ext_binsearch(inode, path + ppos, block);
707 /* if not an empty leaf */
708 if (path[ppos].p_ext)
709 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
711 ext4_ext_show_path(inode, path);
716 ext4_ext_drop_refs(path);
719 return ERR_PTR(-EIO);
723 * ext4_ext_insert_index:
724 * insert new index [@logical;@ptr] into the block at @curp;
725 * check where to insert: before @curp or after @curp
727 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
728 struct ext4_ext_path *curp,
729 int logical, ext4_fsblk_t ptr)
731 struct ext4_extent_idx *ix;
734 err = ext4_ext_get_access(handle, inode, curp);
738 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
739 EXT4_ERROR_INODE(inode,
740 "logical %d == ei_block %d!",
741 logical, le32_to_cpu(curp->p_idx->ei_block));
745 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
746 >= le16_to_cpu(curp->p_hdr->eh_max))) {
747 EXT4_ERROR_INODE(inode,
748 "eh_entries %d >= eh_max %d!",
749 le16_to_cpu(curp->p_hdr->eh_entries),
750 le16_to_cpu(curp->p_hdr->eh_max));
754 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
755 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
757 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
758 len = (len - 1) * sizeof(struct ext4_extent_idx);
759 len = len < 0 ? 0 : len;
760 ext_debug("insert new index %d after: %llu. "
761 "move %d from 0x%p to 0x%p\n",
763 (curp->p_idx + 1), (curp->p_idx + 2));
764 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
766 ix = curp->p_idx + 1;
769 len = len * sizeof(struct ext4_extent_idx);
770 len = len < 0 ? 0 : len;
771 ext_debug("insert new index %d before: %llu. "
772 "move %d from 0x%p to 0x%p\n",
774 curp->p_idx, (curp->p_idx + 1));
775 memmove(curp->p_idx + 1, curp->p_idx, len);
779 ix->ei_block = cpu_to_le32(logical);
780 ext4_idx_store_pblock(ix, ptr);
781 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
783 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
784 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
788 err = ext4_ext_dirty(handle, inode, curp);
789 ext4_std_error(inode->i_sb, err);
796 * inserts new subtree into the path, using free index entry
798 * - allocates all needed blocks (new leaf and all intermediate index blocks)
799 * - makes decision where to split
800 * - moves remaining extents and index entries (right to the split point)
801 * into the newly allocated blocks
802 * - initializes subtree
804 static int ext4_ext_split(handle_t *handle, struct inode *inode,
806 struct ext4_ext_path *path,
807 struct ext4_extent *newext, int at)
809 struct buffer_head *bh = NULL;
810 int depth = ext_depth(inode);
811 struct ext4_extent_header *neh;
812 struct ext4_extent_idx *fidx;
814 ext4_fsblk_t newblock, oldblock;
816 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
819 /* make decision: where to split? */
820 /* FIXME: now decision is simplest: at current extent */
822 /* if current leaf will be split, then we should use
823 * border from split point */
824 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
825 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
828 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
829 border = path[depth].p_ext[1].ee_block;
830 ext_debug("leaf will be split."
831 " next leaf starts at %d\n",
832 le32_to_cpu(border));
834 border = newext->ee_block;
835 ext_debug("leaf will be added."
836 " next leaf starts at %d\n",
837 le32_to_cpu(border));
841 * If error occurs, then we break processing
842 * and mark filesystem read-only. index won't
843 * be inserted and tree will be in consistent
844 * state. Next mount will repair buffers too.
848 * Get array to track all allocated blocks.
849 * We need this to handle errors and free blocks
852 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
856 /* allocate all needed blocks */
857 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
858 for (a = 0; a < depth - at; a++) {
859 newblock = ext4_ext_new_meta_block(handle, inode, path,
860 newext, &err, flags);
863 ablocks[a] = newblock;
866 /* initialize new leaf */
867 newblock = ablocks[--a];
868 if (unlikely(newblock == 0)) {
869 EXT4_ERROR_INODE(inode, "newblock == 0!");
873 bh = sb_getblk(inode->i_sb, newblock);
880 err = ext4_journal_get_create_access(handle, bh);
884 neh = ext_block_hdr(bh);
886 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
887 neh->eh_magic = EXT4_EXT_MAGIC;
890 /* move remainder of path[depth] to the new leaf */
891 if (unlikely(path[depth].p_hdr->eh_entries !=
892 path[depth].p_hdr->eh_max)) {
893 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
894 path[depth].p_hdr->eh_entries,
895 path[depth].p_hdr->eh_max);
899 /* start copy from next extent */
900 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
901 ext4_ext_show_move(inode, path, newblock, depth);
903 struct ext4_extent *ex;
904 ex = EXT_FIRST_EXTENT(neh);
905 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
906 le16_add_cpu(&neh->eh_entries, m);
909 set_buffer_uptodate(bh);
912 err = ext4_handle_dirty_metadata(handle, inode, bh);
918 /* correct old leaf */
920 err = ext4_ext_get_access(handle, inode, path + depth);
923 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
924 err = ext4_ext_dirty(handle, inode, path + depth);
930 /* create intermediate indexes */
932 if (unlikely(k < 0)) {
933 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
938 ext_debug("create %d intermediate indices\n", k);
939 /* insert new index into current index block */
940 /* current depth stored in i var */
944 newblock = ablocks[--a];
945 bh = sb_getblk(inode->i_sb, newblock);
952 err = ext4_journal_get_create_access(handle, bh);
956 neh = ext_block_hdr(bh);
957 neh->eh_entries = cpu_to_le16(1);
958 neh->eh_magic = EXT4_EXT_MAGIC;
959 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
960 neh->eh_depth = cpu_to_le16(depth - i);
961 fidx = EXT_FIRST_INDEX(neh);
962 fidx->ei_block = border;
963 ext4_idx_store_pblock(fidx, oldblock);
965 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
966 i, newblock, le32_to_cpu(border), oldblock);
968 /* move remainder of path[i] to the new index block */
969 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
970 EXT_LAST_INDEX(path[i].p_hdr))) {
971 EXT4_ERROR_INODE(inode,
972 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
973 le32_to_cpu(path[i].p_ext->ee_block));
977 /* start copy indexes */
978 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
979 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
980 EXT_MAX_INDEX(path[i].p_hdr));
981 ext4_ext_show_move(inode, path, newblock, i);
983 memmove(++fidx, path[i].p_idx,
984 sizeof(struct ext4_extent_idx) * m);
985 le16_add_cpu(&neh->eh_entries, m);
987 set_buffer_uptodate(bh);
990 err = ext4_handle_dirty_metadata(handle, inode, bh);
996 /* correct old index */
998 err = ext4_ext_get_access(handle, inode, path + i);
1001 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1002 err = ext4_ext_dirty(handle, inode, path + i);
1010 /* insert new index */
1011 err = ext4_ext_insert_index(handle, inode, path + at,
1012 le32_to_cpu(border), newblock);
1016 if (buffer_locked(bh))
1022 /* free all allocated blocks in error case */
1023 for (i = 0; i < depth; i++) {
1026 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1027 EXT4_FREE_BLOCKS_METADATA);
1036 * ext4_ext_grow_indepth:
1037 * implements tree growing procedure:
1038 * - allocates new block
1039 * - moves top-level data (index block or leaf) into the new block
1040 * - initializes new top-level, creating index that points to the
1041 * just created block
1043 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1045 struct ext4_ext_path *path,
1046 struct ext4_extent *newext)
1048 struct ext4_ext_path *curp = path;
1049 struct ext4_extent_header *neh;
1050 struct buffer_head *bh;
1051 ext4_fsblk_t newblock;
1054 newblock = ext4_ext_new_meta_block(handle, inode, path,
1055 newext, &err, flags);
1059 bh = sb_getblk(inode->i_sb, newblock);
1062 ext4_std_error(inode->i_sb, err);
1067 err = ext4_journal_get_create_access(handle, bh);
1073 /* move top-level index/leaf into new block */
1074 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1076 /* set size of new block */
1077 neh = ext_block_hdr(bh);
1078 /* old root could have indexes or leaves
1079 * so calculate e_max right way */
1080 if (ext_depth(inode))
1081 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1083 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1084 neh->eh_magic = EXT4_EXT_MAGIC;
1085 set_buffer_uptodate(bh);
1088 err = ext4_handle_dirty_metadata(handle, inode, bh);
1092 /* create index in new top-level index: num,max,pointer */
1093 err = ext4_ext_get_access(handle, inode, curp);
1097 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1098 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1099 curp->p_hdr->eh_entries = cpu_to_le16(1);
1100 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1102 if (path[0].p_hdr->eh_depth)
1103 curp->p_idx->ei_block =
1104 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1106 curp->p_idx->ei_block =
1107 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1108 ext4_idx_store_pblock(curp->p_idx, newblock);
1110 neh = ext_inode_hdr(inode);
1111 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1112 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1113 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1114 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1116 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1117 err = ext4_ext_dirty(handle, inode, curp);
1125 * ext4_ext_create_new_leaf:
1126 * finds empty index and adds new leaf.
1127 * if no free index is found, then it requests in-depth growing.
1129 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1131 struct ext4_ext_path *path,
1132 struct ext4_extent *newext)
1134 struct ext4_ext_path *curp;
1135 int depth, i, err = 0;
1138 i = depth = ext_depth(inode);
1140 /* walk up to the tree and look for free index entry */
1141 curp = path + depth;
1142 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1147 /* we use already allocated block for index block,
1148 * so subsequent data blocks should be contiguous */
1149 if (EXT_HAS_FREE_INDEX(curp)) {
1150 /* if we found index with free entry, then use that
1151 * entry: create all needed subtree and add new leaf */
1152 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1157 ext4_ext_drop_refs(path);
1158 path = ext4_ext_find_extent(inode,
1159 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1162 err = PTR_ERR(path);
1164 /* tree is full, time to grow in depth */
1165 err = ext4_ext_grow_indepth(handle, inode, flags,
1171 ext4_ext_drop_refs(path);
1172 path = ext4_ext_find_extent(inode,
1173 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1176 err = PTR_ERR(path);
1181 * only first (depth 0 -> 1) produces free space;
1182 * in all other cases we have to split the grown tree
1184 depth = ext_depth(inode);
1185 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1186 /* now we need to split */
1196 * search the closest allocated block to the left for *logical
1197 * and returns it at @logical + it's physical address at @phys
1198 * if *logical is the smallest allocated block, the function
1199 * returns 0 at @phys
1200 * return value contains 0 (success) or error code
1202 static int ext4_ext_search_left(struct inode *inode,
1203 struct ext4_ext_path *path,
1204 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1206 struct ext4_extent_idx *ix;
1207 struct ext4_extent *ex;
1210 if (unlikely(path == NULL)) {
1211 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1214 depth = path->p_depth;
1217 if (depth == 0 && path->p_ext == NULL)
1220 /* usually extent in the path covers blocks smaller
1221 * then *logical, but it can be that extent is the
1222 * first one in the file */
1224 ex = path[depth].p_ext;
1225 ee_len = ext4_ext_get_actual_len(ex);
1226 if (*logical < le32_to_cpu(ex->ee_block)) {
1227 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1228 EXT4_ERROR_INODE(inode,
1229 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1230 *logical, le32_to_cpu(ex->ee_block));
1233 while (--depth >= 0) {
1234 ix = path[depth].p_idx;
1235 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1236 EXT4_ERROR_INODE(inode,
1237 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1238 ix != NULL ? ix->ei_block : 0,
1239 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1240 EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1248 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1249 EXT4_ERROR_INODE(inode,
1250 "logical %d < ee_block %d + ee_len %d!",
1251 *logical, le32_to_cpu(ex->ee_block), ee_len);
1255 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1256 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1261 * search the closest allocated block to the right for *logical
1262 * and returns it at @logical + it's physical address at @phys
1263 * if *logical is the smallest allocated block, the function
1264 * returns 0 at @phys
1265 * return value contains 0 (success) or error code
1267 static int ext4_ext_search_right(struct inode *inode,
1268 struct ext4_ext_path *path,
1269 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1271 struct buffer_head *bh = NULL;
1272 struct ext4_extent_header *eh;
1273 struct ext4_extent_idx *ix;
1274 struct ext4_extent *ex;
1276 int depth; /* Note, NOT eh_depth; depth from top of tree */
1279 if (unlikely(path == NULL)) {
1280 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1283 depth = path->p_depth;
1286 if (depth == 0 && path->p_ext == NULL)
1289 /* usually extent in the path covers blocks smaller
1290 * then *logical, but it can be that extent is the
1291 * first one in the file */
1293 ex = path[depth].p_ext;
1294 ee_len = ext4_ext_get_actual_len(ex);
1295 if (*logical < le32_to_cpu(ex->ee_block)) {
1296 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1297 EXT4_ERROR_INODE(inode,
1298 "first_extent(path[%d].p_hdr) != ex",
1302 while (--depth >= 0) {
1303 ix = path[depth].p_idx;
1304 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1305 EXT4_ERROR_INODE(inode,
1306 "ix != EXT_FIRST_INDEX *logical %d!",
1311 *logical = le32_to_cpu(ex->ee_block);
1312 *phys = ext4_ext_pblock(ex);
1316 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1317 EXT4_ERROR_INODE(inode,
1318 "logical %d < ee_block %d + ee_len %d!",
1319 *logical, le32_to_cpu(ex->ee_block), ee_len);
1323 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1324 /* next allocated block in this leaf */
1326 *logical = le32_to_cpu(ex->ee_block);
1327 *phys = ext4_ext_pblock(ex);
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);
1371 *logical = le32_to_cpu(ex->ee_block);
1372 *phys = ext4_ext_pblock(ex);
1378 * ext4_ext_next_allocated_block:
1379 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1380 * NOTE: it considers block number from index entry as
1381 * allocated block. Thus, index entries have to be consistent
1385 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1389 BUG_ON(path == NULL);
1390 depth = path->p_depth;
1392 if (depth == 0 && path->p_ext == NULL)
1393 return EXT_MAX_BLOCKS;
1395 while (depth >= 0) {
1396 if (depth == path->p_depth) {
1398 if (path[depth].p_ext !=
1399 EXT_LAST_EXTENT(path[depth].p_hdr))
1400 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1403 if (path[depth].p_idx !=
1404 EXT_LAST_INDEX(path[depth].p_hdr))
1405 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1410 return EXT_MAX_BLOCKS;
1414 * ext4_ext_next_leaf_block:
1415 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1417 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1421 BUG_ON(path == NULL);
1422 depth = path->p_depth;
1424 /* zero-tree has no leaf blocks at all */
1426 return EXT_MAX_BLOCKS;
1428 /* go to index block */
1431 while (depth >= 0) {
1432 if (path[depth].p_idx !=
1433 EXT_LAST_INDEX(path[depth].p_hdr))
1434 return (ext4_lblk_t)
1435 le32_to_cpu(path[depth].p_idx[1].ei_block);
1439 return EXT_MAX_BLOCKS;
1443 * ext4_ext_correct_indexes:
1444 * if leaf gets modified and modified extent is first in the leaf,
1445 * then we have to correct all indexes above.
1446 * TODO: do we need to correct tree in all cases?
1448 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1449 struct ext4_ext_path *path)
1451 struct ext4_extent_header *eh;
1452 int depth = ext_depth(inode);
1453 struct ext4_extent *ex;
1457 eh = path[depth].p_hdr;
1458 ex = path[depth].p_ext;
1460 if (unlikely(ex == NULL || eh == NULL)) {
1461 EXT4_ERROR_INODE(inode,
1462 "ex %p == NULL or eh %p == NULL", ex, eh);
1467 /* there is no tree at all */
1471 if (ex != EXT_FIRST_EXTENT(eh)) {
1472 /* we correct tree if first leaf got modified only */
1477 * TODO: we need correction if border is smaller than current one
1480 border = path[depth].p_ext->ee_block;
1481 err = ext4_ext_get_access(handle, inode, path + k);
1484 path[k].p_idx->ei_block = border;
1485 err = ext4_ext_dirty(handle, inode, path + k);
1490 /* change all left-side indexes */
1491 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1493 err = ext4_ext_get_access(handle, inode, path + k);
1496 path[k].p_idx->ei_block = border;
1497 err = ext4_ext_dirty(handle, inode, path + k);
1506 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1507 struct ext4_extent *ex2)
1509 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1512 * Make sure that either both extents are uninitialized, or
1515 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1518 if (ext4_ext_is_uninitialized(ex1))
1519 max_len = EXT_UNINIT_MAX_LEN;
1521 max_len = EXT_INIT_MAX_LEN;
1523 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1524 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1526 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1527 le32_to_cpu(ex2->ee_block))
1531 * To allow future support for preallocated extents to be added
1532 * as an RO_COMPAT feature, refuse to merge to extents if
1533 * this can result in the top bit of ee_len being set.
1535 if (ext1_ee_len + ext2_ee_len > max_len)
1537 #ifdef AGGRESSIVE_TEST
1538 if (ext1_ee_len >= 4)
1542 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1548 * This function tries to merge the "ex" extent to the next extent in the tree.
1549 * It always tries to merge towards right. If you want to merge towards
1550 * left, pass "ex - 1" as argument instead of "ex".
1551 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1552 * 1 if they got merged.
1554 static int ext4_ext_try_to_merge_right(struct inode *inode,
1555 struct ext4_ext_path *path,
1556 struct ext4_extent *ex)
1558 struct ext4_extent_header *eh;
1559 unsigned int depth, len;
1561 int uninitialized = 0;
1563 depth = ext_depth(inode);
1564 BUG_ON(path[depth].p_hdr == NULL);
1565 eh = path[depth].p_hdr;
1567 while (ex < EXT_LAST_EXTENT(eh)) {
1568 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1570 /* merge with next extent! */
1571 if (ext4_ext_is_uninitialized(ex))
1573 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1574 + ext4_ext_get_actual_len(ex + 1));
1576 ext4_ext_mark_uninitialized(ex);
1578 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1579 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1580 * sizeof(struct ext4_extent);
1581 memmove(ex + 1, ex + 2, len);
1583 le16_add_cpu(&eh->eh_entries, -1);
1585 WARN_ON(eh->eh_entries == 0);
1586 if (!eh->eh_entries)
1587 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1594 * This function tries to merge the @ex extent to neighbours in the tree.
1595 * return 1 if merge left else 0.
1597 static int ext4_ext_try_to_merge(struct inode *inode,
1598 struct ext4_ext_path *path,
1599 struct ext4_extent *ex) {
1600 struct ext4_extent_header *eh;
1605 depth = ext_depth(inode);
1606 BUG_ON(path[depth].p_hdr == NULL);
1607 eh = path[depth].p_hdr;
1609 if (ex > EXT_FIRST_EXTENT(eh))
1610 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1613 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1619 * check if a portion of the "newext" extent overlaps with an
1622 * If there is an overlap discovered, it updates the length of the newext
1623 * such that there will be no overlap, and then returns 1.
1624 * If there is no overlap found, it returns 0.
1626 static unsigned int ext4_ext_check_overlap(struct inode *inode,
1627 struct ext4_extent *newext,
1628 struct ext4_ext_path *path)
1631 unsigned int depth, len1;
1632 unsigned int ret = 0;
1634 b1 = le32_to_cpu(newext->ee_block);
1635 len1 = ext4_ext_get_actual_len(newext);
1636 depth = ext_depth(inode);
1637 if (!path[depth].p_ext)
1639 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1642 * get the next allocated block if the extent in the path
1643 * is before the requested block(s)
1646 b2 = ext4_ext_next_allocated_block(path);
1647 if (b2 == EXT_MAX_BLOCKS)
1651 /* check for wrap through zero on extent logical start block*/
1652 if (b1 + len1 < b1) {
1653 len1 = EXT_MAX_BLOCKS - b1;
1654 newext->ee_len = cpu_to_le16(len1);
1658 /* check for overlap */
1659 if (b1 + len1 > b2) {
1660 newext->ee_len = cpu_to_le16(b2 - b1);
1668 * ext4_ext_insert_extent:
1669 * tries to merge requsted extent into the existing extent or
1670 * inserts requested extent as new one into the tree,
1671 * creating new leaf in the no-space case.
1673 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1674 struct ext4_ext_path *path,
1675 struct ext4_extent *newext, int flag)
1677 struct ext4_extent_header *eh;
1678 struct ext4_extent *ex, *fex;
1679 struct ext4_extent *nearex; /* nearest extent */
1680 struct ext4_ext_path *npath = NULL;
1681 int depth, len, err;
1683 unsigned uninitialized = 0;
1686 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1687 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1690 depth = ext_depth(inode);
1691 ex = path[depth].p_ext;
1692 if (unlikely(path[depth].p_hdr == NULL)) {
1693 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1697 /* try to insert block into found extent and return */
1698 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1699 && ext4_can_extents_be_merged(inode, ex, newext)) {
1700 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1701 ext4_ext_is_uninitialized(newext),
1702 ext4_ext_get_actual_len(newext),
1703 le32_to_cpu(ex->ee_block),
1704 ext4_ext_is_uninitialized(ex),
1705 ext4_ext_get_actual_len(ex),
1706 ext4_ext_pblock(ex));
1707 err = ext4_ext_get_access(handle, inode, path + depth);
1712 * ext4_can_extents_be_merged should have checked that either
1713 * both extents are uninitialized, or both aren't. Thus we
1714 * need to check only one of them here.
1716 if (ext4_ext_is_uninitialized(ex))
1718 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1719 + ext4_ext_get_actual_len(newext));
1721 ext4_ext_mark_uninitialized(ex);
1722 eh = path[depth].p_hdr;
1727 depth = ext_depth(inode);
1728 eh = path[depth].p_hdr;
1729 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1732 /* probably next leaf has space for us? */
1733 fex = EXT_LAST_EXTENT(eh);
1734 next = EXT_MAX_BLOCKS;
1735 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1736 next = ext4_ext_next_leaf_block(path);
1737 if (next != EXT_MAX_BLOCKS) {
1738 ext_debug("next leaf block - %d\n", next);
1739 BUG_ON(npath != NULL);
1740 npath = ext4_ext_find_extent(inode, next, NULL);
1742 return PTR_ERR(npath);
1743 BUG_ON(npath->p_depth != path->p_depth);
1744 eh = npath[depth].p_hdr;
1745 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1746 ext_debug("next leaf isn't full(%d)\n",
1747 le16_to_cpu(eh->eh_entries));
1751 ext_debug("next leaf has no free space(%d,%d)\n",
1752 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1756 * There is no free space in the found leaf.
1757 * We're gonna add a new leaf in the tree.
1759 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1760 flags = EXT4_MB_USE_ROOT_BLOCKS;
1761 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1764 depth = ext_depth(inode);
1765 eh = path[depth].p_hdr;
1768 nearex = path[depth].p_ext;
1770 err = ext4_ext_get_access(handle, inode, path + depth);
1775 /* there is no extent in this leaf, create first one */
1776 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1777 le32_to_cpu(newext->ee_block),
1778 ext4_ext_pblock(newext),
1779 ext4_ext_is_uninitialized(newext),
1780 ext4_ext_get_actual_len(newext));
1781 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1782 } else if (le32_to_cpu(newext->ee_block)
1783 > le32_to_cpu(nearex->ee_block)) {
1784 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1785 if (nearex != EXT_LAST_EXTENT(eh)) {
1786 len = EXT_MAX_EXTENT(eh) - nearex;
1787 len = (len - 1) * sizeof(struct ext4_extent);
1788 len = len < 0 ? 0 : len;
1789 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1790 "move %d from 0x%p to 0x%p\n",
1791 le32_to_cpu(newext->ee_block),
1792 ext4_ext_pblock(newext),
1793 ext4_ext_is_uninitialized(newext),
1794 ext4_ext_get_actual_len(newext),
1795 nearex, len, nearex + 1, nearex + 2);
1796 memmove(nearex + 2, nearex + 1, len);
1798 path[depth].p_ext = nearex + 1;
1800 BUG_ON(newext->ee_block == nearex->ee_block);
1801 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1802 len = len < 0 ? 0 : len;
1803 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1804 "move %d from 0x%p to 0x%p\n",
1805 le32_to_cpu(newext->ee_block),
1806 ext4_ext_pblock(newext),
1807 ext4_ext_is_uninitialized(newext),
1808 ext4_ext_get_actual_len(newext),
1809 nearex, len, nearex + 1, nearex + 2);
1810 memmove(nearex + 1, nearex, len);
1811 path[depth].p_ext = nearex;
1814 le16_add_cpu(&eh->eh_entries, 1);
1815 nearex = path[depth].p_ext;
1816 nearex->ee_block = newext->ee_block;
1817 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1818 nearex->ee_len = newext->ee_len;
1821 /* try to merge extents to the right */
1822 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1823 ext4_ext_try_to_merge(inode, path, nearex);
1825 /* try to merge extents to the left */
1827 /* time to correct all indexes above */
1828 err = ext4_ext_correct_indexes(handle, inode, path);
1832 err = ext4_ext_dirty(handle, inode, path + depth);
1836 ext4_ext_drop_refs(npath);
1839 ext4_ext_invalidate_cache(inode);
1843 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1844 ext4_lblk_t num, ext_prepare_callback func,
1847 struct ext4_ext_path *path = NULL;
1848 struct ext4_ext_cache cbex;
1849 struct ext4_extent *ex;
1850 ext4_lblk_t next, start = 0, end = 0;
1851 ext4_lblk_t last = block + num;
1852 int depth, exists, err = 0;
1854 BUG_ON(func == NULL);
1855 BUG_ON(inode == NULL);
1857 while (block < last && block != EXT_MAX_BLOCKS) {
1859 /* find extent for this block */
1860 down_read(&EXT4_I(inode)->i_data_sem);
1861 path = ext4_ext_find_extent(inode, block, path);
1862 up_read(&EXT4_I(inode)->i_data_sem);
1864 err = PTR_ERR(path);
1869 depth = ext_depth(inode);
1870 if (unlikely(path[depth].p_hdr == NULL)) {
1871 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1875 ex = path[depth].p_ext;
1876 next = ext4_ext_next_allocated_block(path);
1880 /* there is no extent yet, so try to allocate
1881 * all requested space */
1884 } else if (le32_to_cpu(ex->ee_block) > block) {
1885 /* need to allocate space before found extent */
1887 end = le32_to_cpu(ex->ee_block);
1888 if (block + num < end)
1890 } else if (block >= le32_to_cpu(ex->ee_block)
1891 + ext4_ext_get_actual_len(ex)) {
1892 /* need to allocate space after found extent */
1897 } else if (block >= le32_to_cpu(ex->ee_block)) {
1899 * some part of requested space is covered
1903 end = le32_to_cpu(ex->ee_block)
1904 + ext4_ext_get_actual_len(ex);
1905 if (block + num < end)
1911 BUG_ON(end <= start);
1914 cbex.ec_block = start;
1915 cbex.ec_len = end - start;
1918 cbex.ec_block = le32_to_cpu(ex->ee_block);
1919 cbex.ec_len = ext4_ext_get_actual_len(ex);
1920 cbex.ec_start = ext4_ext_pblock(ex);
1923 if (unlikely(cbex.ec_len == 0)) {
1924 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1928 err = func(inode, next, &cbex, ex, cbdata);
1929 ext4_ext_drop_refs(path);
1934 if (err == EXT_REPEAT)
1936 else if (err == EXT_BREAK) {
1941 if (ext_depth(inode) != depth) {
1942 /* depth was changed. we have to realloc path */
1947 block = cbex.ec_block + cbex.ec_len;
1951 ext4_ext_drop_refs(path);
1959 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1960 __u32 len, ext4_fsblk_t start)
1962 struct ext4_ext_cache *cex;
1964 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1965 cex = &EXT4_I(inode)->i_cached_extent;
1966 cex->ec_block = block;
1968 cex->ec_start = start;
1969 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1973 * ext4_ext_put_gap_in_cache:
1974 * calculate boundaries of the gap that the requested block fits into
1975 * and cache this gap
1978 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1981 int depth = ext_depth(inode);
1984 struct ext4_extent *ex;
1986 ex = path[depth].p_ext;
1988 /* there is no extent yet, so gap is [0;-] */
1990 len = EXT_MAX_BLOCKS;
1991 ext_debug("cache gap(whole file):");
1992 } else if (block < le32_to_cpu(ex->ee_block)) {
1994 len = le32_to_cpu(ex->ee_block) - block;
1995 ext_debug("cache gap(before): %u [%u:%u]",
1997 le32_to_cpu(ex->ee_block),
1998 ext4_ext_get_actual_len(ex));
1999 } else if (block >= le32_to_cpu(ex->ee_block)
2000 + ext4_ext_get_actual_len(ex)) {
2002 lblock = le32_to_cpu(ex->ee_block)
2003 + ext4_ext_get_actual_len(ex);
2005 next = ext4_ext_next_allocated_block(path);
2006 ext_debug("cache gap(after): [%u:%u] %u",
2007 le32_to_cpu(ex->ee_block),
2008 ext4_ext_get_actual_len(ex),
2010 BUG_ON(next == lblock);
2011 len = next - lblock;
2017 ext_debug(" -> %u:%lu\n", lblock, len);
2018 ext4_ext_put_in_cache(inode, lblock, len, 0);
2022 * ext4_ext_in_cache()
2023 * Checks to see if the given block is in the cache.
2024 * If it is, the cached extent is stored in the given
2025 * cache extent pointer. If the cached extent is a hole,
2026 * this routine should be used instead of
2027 * ext4_ext_in_cache if the calling function needs to
2028 * know the size of the hole.
2030 * @inode: The files inode
2031 * @block: The block to look for in the cache
2032 * @ex: Pointer where the cached extent will be stored
2033 * if it contains block
2035 * Return 0 if cache is invalid; 1 if the cache is valid
2037 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2038 struct ext4_ext_cache *ex){
2039 struct ext4_ext_cache *cex;
2040 struct ext4_sb_info *sbi;
2044 * We borrow i_block_reservation_lock to protect i_cached_extent
2046 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2047 cex = &EXT4_I(inode)->i_cached_extent;
2048 sbi = EXT4_SB(inode->i_sb);
2050 /* has cache valid data? */
2051 if (cex->ec_len == 0)
2054 if (in_range(block, cex->ec_block, cex->ec_len)) {
2055 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2056 ext_debug("%u cached by %u:%u:%llu\n",
2058 cex->ec_block, cex->ec_len, cex->ec_start);
2063 sbi->extent_cache_misses++;
2065 sbi->extent_cache_hits++;
2066 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2071 * ext4_ext_in_cache()
2072 * Checks to see if the given block is in the cache.
2073 * If it is, the cached extent is stored in the given
2076 * @inode: The files inode
2077 * @block: The block to look for in the cache
2078 * @ex: Pointer where the cached extent will be stored
2079 * if it contains block
2081 * Return 0 if cache is invalid; 1 if the cache is valid
2084 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2085 struct ext4_extent *ex)
2087 struct ext4_ext_cache cex;
2090 if (ext4_ext_check_cache(inode, block, &cex)) {
2091 ex->ee_block = cpu_to_le32(cex.ec_block);
2092 ext4_ext_store_pblock(ex, cex.ec_start);
2093 ex->ee_len = cpu_to_le16(cex.ec_len);
2103 * removes index from the index block.
2104 * It's used in truncate case only, thus all requests are for
2105 * last index in the block only.
2107 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2108 struct ext4_ext_path *path)
2113 /* free index block */
2115 leaf = ext4_idx_pblock(path->p_idx);
2116 if (unlikely(path->p_hdr->eh_entries == 0)) {
2117 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2120 err = ext4_ext_get_access(handle, inode, path);
2123 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2124 err = ext4_ext_dirty(handle, inode, path);
2127 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2128 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2129 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2134 * ext4_ext_calc_credits_for_single_extent:
2135 * This routine returns max. credits that needed to insert an extent
2136 * to the extent tree.
2137 * When pass the actual path, the caller should calculate credits
2140 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2141 struct ext4_ext_path *path)
2144 int depth = ext_depth(inode);
2147 /* probably there is space in leaf? */
2148 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2149 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2152 * There are some space in the leaf tree, no
2153 * need to account for leaf block credit
2155 * bitmaps and block group descriptor blocks
2156 * and other metadat blocks still need to be
2159 /* 1 bitmap, 1 block group descriptor */
2160 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2165 return ext4_chunk_trans_blocks(inode, nrblocks);
2169 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2171 * if nrblocks are fit in a single extent (chunk flag is 1), then
2172 * in the worse case, each tree level index/leaf need to be changed
2173 * if the tree split due to insert a new extent, then the old tree
2174 * index/leaf need to be updated too
2176 * If the nrblocks are discontiguous, they could cause
2177 * the whole tree split more than once, but this is really rare.
2179 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2182 int depth = ext_depth(inode);
2192 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2193 struct ext4_extent *ex,
2194 ext4_lblk_t from, ext4_lblk_t to)
2196 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2197 int flags = EXT4_FREE_BLOCKS_FORGET;
2199 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2200 flags |= EXT4_FREE_BLOCKS_METADATA;
2201 #ifdef EXTENTS_STATS
2203 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2204 spin_lock(&sbi->s_ext_stats_lock);
2205 sbi->s_ext_blocks += ee_len;
2206 sbi->s_ext_extents++;
2207 if (ee_len < sbi->s_ext_min)
2208 sbi->s_ext_min = ee_len;
2209 if (ee_len > sbi->s_ext_max)
2210 sbi->s_ext_max = ee_len;
2211 if (ext_depth(inode) > sbi->s_depth_max)
2212 sbi->s_depth_max = ext_depth(inode);
2213 spin_unlock(&sbi->s_ext_stats_lock);
2216 if (from >= le32_to_cpu(ex->ee_block)
2217 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2222 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2223 start = ext4_ext_pblock(ex) + ee_len - num;
2224 ext_debug("free last %u blocks starting %llu\n", num, start);
2225 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2226 } else if (from == le32_to_cpu(ex->ee_block)
2227 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2233 start = ext4_ext_pblock(ex);
2235 ext_debug("free first %u blocks starting %llu\n", num, start);
2236 ext4_free_blocks(handle, inode, 0, start, num, flags);
2239 printk(KERN_INFO "strange request: removal(2) "
2240 "%u-%u from %u:%u\n",
2241 from, to, le32_to_cpu(ex->ee_block), ee_len);
2248 * ext4_ext_rm_leaf() Removes the extents associated with the
2249 * blocks appearing between "start" and "end", and splits the extents
2250 * if "start" and "end" appear in the same extent
2252 * @handle: The journal handle
2253 * @inode: The files inode
2254 * @path: The path to the leaf
2255 * @start: The first block to remove
2256 * @end: The last block to remove
2259 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2260 struct ext4_ext_path *path, ext4_lblk_t start,
2263 int err = 0, correct_index = 0;
2264 int depth = ext_depth(inode), credits;
2265 struct ext4_extent_header *eh;
2266 ext4_lblk_t a, b, block;
2268 ext4_lblk_t ex_ee_block;
2269 unsigned short ex_ee_len;
2270 unsigned uninitialized = 0;
2271 struct ext4_extent *ex;
2272 struct ext4_map_blocks map;
2274 /* the header must be checked already in ext4_ext_remove_space() */
2275 ext_debug("truncate since %u in leaf\n", start);
2276 if (!path[depth].p_hdr)
2277 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2278 eh = path[depth].p_hdr;
2279 if (unlikely(path[depth].p_hdr == NULL)) {
2280 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2283 /* find where to start removing */
2284 ex = EXT_LAST_EXTENT(eh);
2286 ex_ee_block = le32_to_cpu(ex->ee_block);
2287 ex_ee_len = ext4_ext_get_actual_len(ex);
2289 while (ex >= EXT_FIRST_EXTENT(eh) &&
2290 ex_ee_block + ex_ee_len > start) {
2292 if (ext4_ext_is_uninitialized(ex))
2297 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2298 uninitialized, ex_ee_len);
2299 path[depth].p_ext = ex;
2301 a = ex_ee_block > start ? ex_ee_block : start;
2302 b = ex_ee_block+ex_ee_len - 1 < end ?
2303 ex_ee_block+ex_ee_len - 1 : end;
2305 ext_debug(" border %u:%u\n", a, b);
2307 /* If this extent is beyond the end of the hole, skip it */
2308 if (end <= ex_ee_block) {
2310 ex_ee_block = le32_to_cpu(ex->ee_block);
2311 ex_ee_len = ext4_ext_get_actual_len(ex);
2313 } else if (a != ex_ee_block &&
2314 b != ex_ee_block + ex_ee_len - 1) {
2316 * If this is a truncate, then this condition should
2317 * never happen because at least one of the end points
2318 * needs to be on the edge of the extent.
2320 if (end == EXT_MAX_BLOCKS - 1) {
2321 ext_debug(" bad truncate %u:%u\n",
2329 * else this is a hole punch, so the extent needs to
2330 * be split since neither edge of the hole is on the
2334 map.m_pblk = ext4_ext_pblock(ex);
2335 map.m_lblk = ex_ee_block;
2336 map.m_len = b - ex_ee_block;
2338 err = ext4_split_extent(handle,
2339 inode, path, &map, 0,
2340 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2341 EXT4_GET_BLOCKS_PRE_IO);
2346 ex_ee_len = ext4_ext_get_actual_len(ex);
2348 b = ex_ee_block+ex_ee_len - 1 < end ?
2349 ex_ee_block+ex_ee_len - 1 : end;
2351 /* Then remove tail of this extent */
2352 block = ex_ee_block;
2355 } else if (a != ex_ee_block) {
2356 /* remove tail of the extent */
2357 block = ex_ee_block;
2359 } else if (b != ex_ee_block + ex_ee_len - 1) {
2360 /* remove head of the extent */
2362 num = ex_ee_block + ex_ee_len - b;
2365 * If this is a truncate, this condition
2366 * should never happen
2368 if (end == EXT_MAX_BLOCKS - 1) {
2369 ext_debug(" bad truncate %u:%u\n",
2375 /* remove whole extent: excellent! */
2376 block = ex_ee_block;
2378 if (a != ex_ee_block) {
2379 ext_debug(" bad truncate %u:%u\n",
2385 if (b != ex_ee_block + ex_ee_len - 1) {
2386 ext_debug(" bad truncate %u:%u\n",
2394 * 3 for leaf, sb, and inode plus 2 (bmap and group
2395 * descriptor) for each block group; assume two block
2396 * groups plus ex_ee_len/blocks_per_block_group for
2399 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2400 if (ex == EXT_FIRST_EXTENT(eh)) {
2402 credits += (ext_depth(inode)) + 1;
2404 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2406 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2410 err = ext4_ext_get_access(handle, inode, path + depth);
2414 err = ext4_remove_blocks(handle, inode, ex, a, b);
2419 /* this extent is removed; mark slot entirely unused */
2420 ext4_ext_store_pblock(ex, 0);
2421 } else if (block != ex_ee_block) {
2423 * If this was a head removal, then we need to update
2424 * the physical block since it is now at a different
2427 ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
2430 ex->ee_block = cpu_to_le32(block);
2431 ex->ee_len = cpu_to_le16(num);
2433 * Do not mark uninitialized if all the blocks in the
2434 * extent have been removed.
2436 if (uninitialized && num)
2437 ext4_ext_mark_uninitialized(ex);
2439 err = ext4_ext_dirty(handle, inode, path + depth);
2444 * If the extent was completely released,
2445 * we need to remove it from the leaf
2448 if (end != EXT_MAX_BLOCKS - 1) {
2450 * For hole punching, we need to scoot all the
2451 * extents up when an extent is removed so that
2452 * we dont have blank extents in the middle
2454 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2455 sizeof(struct ext4_extent));
2457 /* Now get rid of the one at the end */
2458 memset(EXT_LAST_EXTENT(eh), 0,
2459 sizeof(struct ext4_extent));
2461 le16_add_cpu(&eh->eh_entries, -1);
2464 ext_debug("new extent: %u:%u:%llu\n", block, num,
2465 ext4_ext_pblock(ex));
2467 ex_ee_block = le32_to_cpu(ex->ee_block);
2468 ex_ee_len = ext4_ext_get_actual_len(ex);
2471 if (correct_index && eh->eh_entries)
2472 err = ext4_ext_correct_indexes(handle, inode, path);
2474 /* if this leaf is free, then we should
2475 * remove it from index block above */
2476 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2477 err = ext4_ext_rm_idx(handle, inode, path + depth);
2484 * ext4_ext_more_to_rm:
2485 * returns 1 if current index has to be freed (even partial)
2488 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2490 BUG_ON(path->p_idx == NULL);
2492 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2496 * if truncate on deeper level happened, it wasn't partial,
2497 * so we have to consider current index for truncation
2499 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2504 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2506 struct super_block *sb = inode->i_sb;
2507 int depth = ext_depth(inode);
2508 struct ext4_ext_path *path;
2512 ext_debug("truncate since %u\n", start);
2514 /* probably first extent we're gonna free will be last in block */
2515 handle = ext4_journal_start(inode, depth + 1);
2517 return PTR_ERR(handle);
2520 ext4_ext_invalidate_cache(inode);
2523 * We start scanning from right side, freeing all the blocks
2524 * after i_size and walking into the tree depth-wise.
2526 depth = ext_depth(inode);
2527 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2529 ext4_journal_stop(handle);
2532 path[0].p_depth = depth;
2533 path[0].p_hdr = ext_inode_hdr(inode);
2534 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2540 while (i >= 0 && err == 0) {
2542 /* this is leaf block */
2543 err = ext4_ext_rm_leaf(handle, inode, path,
2544 start, EXT_MAX_BLOCKS - 1);
2545 /* root level has p_bh == NULL, brelse() eats this */
2546 brelse(path[i].p_bh);
2547 path[i].p_bh = NULL;
2552 /* this is index block */
2553 if (!path[i].p_hdr) {
2554 ext_debug("initialize header\n");
2555 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2558 if (!path[i].p_idx) {
2559 /* this level hasn't been touched yet */
2560 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2561 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2562 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2564 le16_to_cpu(path[i].p_hdr->eh_entries));
2566 /* we were already here, see at next index */
2570 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2571 i, EXT_FIRST_INDEX(path[i].p_hdr),
2573 if (ext4_ext_more_to_rm(path + i)) {
2574 struct buffer_head *bh;
2575 /* go to the next level */
2576 ext_debug("move to level %d (block %llu)\n",
2577 i + 1, ext4_idx_pblock(path[i].p_idx));
2578 memset(path + i + 1, 0, sizeof(*path));
2579 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2581 /* should we reset i_size? */
2585 if (WARN_ON(i + 1 > depth)) {
2589 if (ext4_ext_check(inode, ext_block_hdr(bh),
2594 path[i + 1].p_bh = bh;
2596 /* save actual number of indexes since this
2597 * number is changed at the next iteration */
2598 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2601 /* we finished processing this index, go up */
2602 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2603 /* index is empty, remove it;
2604 * handle must be already prepared by the
2605 * truncatei_leaf() */
2606 err = ext4_ext_rm_idx(handle, inode, path + i);
2608 /* root level has p_bh == NULL, brelse() eats this */
2609 brelse(path[i].p_bh);
2610 path[i].p_bh = NULL;
2612 ext_debug("return to level %d\n", i);
2616 /* TODO: flexible tree reduction should be here */
2617 if (path->p_hdr->eh_entries == 0) {
2619 * truncate to zero freed all the tree,
2620 * so we need to correct eh_depth
2622 err = ext4_ext_get_access(handle, inode, path);
2624 ext_inode_hdr(inode)->eh_depth = 0;
2625 ext_inode_hdr(inode)->eh_max =
2626 cpu_to_le16(ext4_ext_space_root(inode, 0));
2627 err = ext4_ext_dirty(handle, inode, path);
2631 ext4_ext_drop_refs(path);
2635 ext4_journal_stop(handle);
2641 * called at mount time
2643 void ext4_ext_init(struct super_block *sb)
2646 * possible initialization would be here
2649 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2650 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2651 printk(KERN_INFO "EXT4-fs: file extents enabled");
2652 #ifdef AGGRESSIVE_TEST
2653 printk(", aggressive tests");
2655 #ifdef CHECK_BINSEARCH
2656 printk(", check binsearch");
2658 #ifdef EXTENTS_STATS
2663 #ifdef EXTENTS_STATS
2664 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2665 EXT4_SB(sb)->s_ext_min = 1 << 30;
2666 EXT4_SB(sb)->s_ext_max = 0;
2672 * called at umount time
2674 void ext4_ext_release(struct super_block *sb)
2676 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2679 #ifdef EXTENTS_STATS
2680 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2681 struct ext4_sb_info *sbi = EXT4_SB(sb);
2682 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2683 sbi->s_ext_blocks, sbi->s_ext_extents,
2684 sbi->s_ext_blocks / sbi->s_ext_extents);
2685 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2686 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2691 /* FIXME!! we need to try to merge to left or right after zero-out */
2692 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2694 ext4_fsblk_t ee_pblock;
2695 unsigned int ee_len;
2698 ee_len = ext4_ext_get_actual_len(ex);
2699 ee_pblock = ext4_ext_pblock(ex);
2701 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2709 * used by extent splitting.
2711 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2713 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2714 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2717 * ext4_split_extent_at() splits an extent at given block.
2719 * @handle: the journal handle
2720 * @inode: the file inode
2721 * @path: the path to the extent
2722 * @split: the logical block where the extent is splitted.
2723 * @split_flags: indicates if the extent could be zeroout if split fails, and
2724 * the states(init or uninit) of new extents.
2725 * @flags: flags used to insert new extent to extent tree.
2728 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2729 * of which are deterimined by split_flag.
2731 * There are two cases:
2732 * a> the extent are splitted into two extent.
2733 * b> split is not needed, and just mark the extent.
2735 * return 0 on success.
2737 static int ext4_split_extent_at(handle_t *handle,
2738 struct inode *inode,
2739 struct ext4_ext_path *path,
2744 ext4_fsblk_t newblock;
2745 ext4_lblk_t ee_block;
2746 struct ext4_extent *ex, newex, orig_ex;
2747 struct ext4_extent *ex2 = NULL;
2748 unsigned int ee_len, depth;
2751 ext_debug("ext4_split_extents_at: inode %lu, logical"
2752 "block %llu\n", inode->i_ino, (unsigned long long)split);
2754 ext4_ext_show_leaf(inode, path);
2756 depth = ext_depth(inode);
2757 ex = path[depth].p_ext;
2758 ee_block = le32_to_cpu(ex->ee_block);
2759 ee_len = ext4_ext_get_actual_len(ex);
2760 newblock = split - ee_block + ext4_ext_pblock(ex);
2762 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2764 err = ext4_ext_get_access(handle, inode, path + depth);
2768 if (split == ee_block) {
2770 * case b: block @split is the block that the extent begins with
2771 * then we just change the state of the extent, and splitting
2774 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2775 ext4_ext_mark_uninitialized(ex);
2777 ext4_ext_mark_initialized(ex);
2779 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2780 ext4_ext_try_to_merge(inode, path, ex);
2782 err = ext4_ext_dirty(handle, inode, path + depth);
2787 memcpy(&orig_ex, ex, sizeof(orig_ex));
2788 ex->ee_len = cpu_to_le16(split - ee_block);
2789 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2790 ext4_ext_mark_uninitialized(ex);
2793 * path may lead to new leaf, not to original leaf any more
2794 * after ext4_ext_insert_extent() returns,
2796 err = ext4_ext_dirty(handle, inode, path + depth);
2798 goto fix_extent_len;
2801 ex2->ee_block = cpu_to_le32(split);
2802 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2803 ext4_ext_store_pblock(ex2, newblock);
2804 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2805 ext4_ext_mark_uninitialized(ex2);
2807 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2808 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2809 err = ext4_ext_zeroout(inode, &orig_ex);
2811 goto fix_extent_len;
2812 /* update the extent length and mark as initialized */
2813 ex->ee_len = cpu_to_le32(ee_len);
2814 ext4_ext_try_to_merge(inode, path, ex);
2815 err = ext4_ext_dirty(handle, inode, path + depth);
2818 goto fix_extent_len;
2821 ext4_ext_show_leaf(inode, path);
2825 ex->ee_len = orig_ex.ee_len;
2826 ext4_ext_dirty(handle, inode, path + depth);
2831 * ext4_split_extents() splits an extent and mark extent which is covered
2832 * by @map as split_flags indicates
2834 * It may result in splitting the extent into multiple extents (upto three)
2835 * There are three possibilities:
2836 * a> There is no split required
2837 * b> Splits in two extents: Split is happening at either end of the extent
2838 * c> Splits in three extents: Somone is splitting in middle of the extent
2841 static int ext4_split_extent(handle_t *handle,
2842 struct inode *inode,
2843 struct ext4_ext_path *path,
2844 struct ext4_map_blocks *map,
2848 ext4_lblk_t ee_block;
2849 struct ext4_extent *ex;
2850 unsigned int ee_len, depth;
2853 int split_flag1, flags1;
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 uninitialized = ext4_ext_is_uninitialized(ex);
2861 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2862 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2863 EXT4_EXT_MAY_ZEROOUT : 0;
2864 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2866 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2867 EXT4_EXT_MARK_UNINIT2;
2868 err = ext4_split_extent_at(handle, inode, path,
2869 map->m_lblk + map->m_len, split_flag1, flags1);
2874 ext4_ext_drop_refs(path);
2875 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2877 return PTR_ERR(path);
2879 if (map->m_lblk >= ee_block) {
2880 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2881 EXT4_EXT_MAY_ZEROOUT : 0;
2883 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2884 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2885 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2886 err = ext4_split_extent_at(handle, inode, path,
2887 map->m_lblk, split_flag1, flags);
2892 ext4_ext_show_leaf(inode, path);
2894 return err ? err : map->m_len;
2897 #define EXT4_EXT_ZERO_LEN 7
2899 * This function is called by ext4_ext_map_blocks() if someone tries to write
2900 * to an uninitialized extent. It may result in splitting the uninitialized
2901 * extent into multiple extents (up to three - one initialized and two
2903 * There are three possibilities:
2904 * a> There is no split required: Entire extent should be initialized
2905 * b> Splits in two extents: Write is happening at either end of the extent
2906 * c> Splits in three extents: Somone is writing in middle of the extent
2908 static int ext4_ext_convert_to_initialized(handle_t *handle,
2909 struct inode *inode,
2910 struct ext4_map_blocks *map,
2911 struct ext4_ext_path *path)
2913 struct ext4_map_blocks split_map;
2914 struct ext4_extent zero_ex;
2915 struct ext4_extent *ex;
2916 ext4_lblk_t ee_block, eof_block;
2917 unsigned int allocated, ee_len, depth;
2921 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2922 "block %llu, max_blocks %u\n", inode->i_ino,
2923 (unsigned long long)map->m_lblk, map->m_len);
2925 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2926 inode->i_sb->s_blocksize_bits;
2927 if (eof_block < map->m_lblk + map->m_len)
2928 eof_block = map->m_lblk + map->m_len;
2930 depth = ext_depth(inode);
2931 ex = path[depth].p_ext;
2932 ee_block = le32_to_cpu(ex->ee_block);
2933 ee_len = ext4_ext_get_actual_len(ex);
2934 allocated = ee_len - (map->m_lblk - ee_block);
2936 WARN_ON(map->m_lblk < ee_block);
2938 * It is safe to convert extent to initialized via explicit
2939 * zeroout only if extent is fully insde i_size or new_size.
2941 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
2943 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2944 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
2945 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2946 err = ext4_ext_zeroout(inode, ex);
2950 err = ext4_ext_get_access(handle, inode, path + depth);
2953 ext4_ext_mark_initialized(ex);
2954 ext4_ext_try_to_merge(inode, path, ex);
2955 err = ext4_ext_dirty(handle, inode, path + depth);
2961 * 1. split the extent into three extents.
2962 * 2. split the extent into two extents, zeroout the first half.
2963 * 3. split the extent into two extents, zeroout the second half.
2964 * 4. split the extent into two extents with out zeroout.
2966 split_map.m_lblk = map->m_lblk;
2967 split_map.m_len = map->m_len;
2969 if (allocated > map->m_len) {
2970 if (allocated <= EXT4_EXT_ZERO_LEN &&
2971 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2974 cpu_to_le32(map->m_lblk);
2975 zero_ex.ee_len = cpu_to_le16(allocated);
2976 ext4_ext_store_pblock(&zero_ex,
2977 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
2978 err = ext4_ext_zeroout(inode, &zero_ex);
2981 split_map.m_lblk = map->m_lblk;
2982 split_map.m_len = allocated;
2983 } else if ((map->m_lblk - ee_block + map->m_len <
2984 EXT4_EXT_ZERO_LEN) &&
2985 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2987 if (map->m_lblk != ee_block) {
2988 zero_ex.ee_block = ex->ee_block;
2989 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
2991 ext4_ext_store_pblock(&zero_ex,
2992 ext4_ext_pblock(ex));
2993 err = ext4_ext_zeroout(inode, &zero_ex);
2998 split_map.m_lblk = ee_block;
2999 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3000 allocated = map->m_len;
3004 allocated = ext4_split_extent(handle, inode, path,
3005 &split_map, split_flag, 0);
3010 return err ? err : allocated;
3014 * This function is called by ext4_ext_map_blocks() from
3015 * ext4_get_blocks_dio_write() when DIO to write
3016 * to an uninitialized extent.
3018 * Writing to an uninitialized extent may result in splitting the uninitialized
3019 * extent into multiple /initialized uninitialized extents (up to three)
3020 * There are three possibilities:
3021 * a> There is no split required: Entire extent should be uninitialized
3022 * b> Splits in two extents: Write is happening at either end of the extent
3023 * c> Splits in three extents: Somone is writing in middle of the extent
3025 * One of more index blocks maybe needed if the extent tree grow after
3026 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3027 * complete, we need to split the uninitialized extent before DIO submit
3028 * the IO. The uninitialized extent called at this time will be split
3029 * into three uninitialized extent(at most). After IO complete, the part
3030 * being filled will be convert to initialized by the end_io callback function
3031 * via ext4_convert_unwritten_extents().
3033 * Returns the size of uninitialized extent to be written on success.
3035 static int ext4_split_unwritten_extents(handle_t *handle,
3036 struct inode *inode,
3037 struct ext4_map_blocks *map,
3038 struct ext4_ext_path *path,
3041 ext4_lblk_t eof_block;
3042 ext4_lblk_t ee_block;
3043 struct ext4_extent *ex;
3044 unsigned int ee_len;
3045 int split_flag = 0, depth;
3047 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3048 "block %llu, max_blocks %u\n", inode->i_ino,
3049 (unsigned long long)map->m_lblk, map->m_len);
3051 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3052 inode->i_sb->s_blocksize_bits;
3053 if (eof_block < map->m_lblk + map->m_len)
3054 eof_block = map->m_lblk + map->m_len;
3056 * It is safe to convert extent to initialized via explicit
3057 * zeroout only if extent is fully insde i_size or new_size.
3059 depth = ext_depth(inode);
3060 ex = path[depth].p_ext;
3061 ee_block = le32_to_cpu(ex->ee_block);
3062 ee_len = ext4_ext_get_actual_len(ex);
3064 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3065 split_flag |= EXT4_EXT_MARK_UNINIT2;
3067 flags |= EXT4_GET_BLOCKS_PRE_IO;
3068 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3071 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3072 struct inode *inode,
3073 struct ext4_ext_path *path)
3075 struct ext4_extent *ex;
3079 depth = ext_depth(inode);
3080 ex = path[depth].p_ext;
3082 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3083 "block %llu, max_blocks %u\n", inode->i_ino,
3084 (unsigned long long)le32_to_cpu(ex->ee_block),
3085 ext4_ext_get_actual_len(ex));
3087 err = ext4_ext_get_access(handle, inode, path + depth);
3090 /* first mark the extent as initialized */
3091 ext4_ext_mark_initialized(ex);
3093 /* note: ext4_ext_correct_indexes() isn't needed here because
3094 * borders are not changed
3096 ext4_ext_try_to_merge(inode, path, ex);
3098 /* Mark modified extent as dirty */
3099 err = ext4_ext_dirty(handle, inode, path + depth);
3101 ext4_ext_show_leaf(inode, path);
3105 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3106 sector_t block, int count)
3109 for (i = 0; i < count; i++)
3110 unmap_underlying_metadata(bdev, block + i);
3114 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3116 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3118 struct ext4_ext_path *path,
3122 struct ext4_extent_header *eh;
3123 struct ext4_extent *last_ex;
3125 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3128 depth = ext_depth(inode);
3129 eh = path[depth].p_hdr;
3131 if (unlikely(!eh->eh_entries)) {
3132 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3133 "EOFBLOCKS_FL set");
3136 last_ex = EXT_LAST_EXTENT(eh);
3138 * We should clear the EOFBLOCKS_FL flag if we are writing the
3139 * last block in the last extent in the file. We test this by
3140 * first checking to see if the caller to
3141 * ext4_ext_get_blocks() was interested in the last block (or
3142 * a block beyond the last block) in the current extent. If
3143 * this turns out to be false, we can bail out from this
3144 * function immediately.
3146 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3147 ext4_ext_get_actual_len(last_ex))
3150 * If the caller does appear to be planning to write at or
3151 * beyond the end of the current extent, we then test to see
3152 * if the current extent is the last extent in the file, by
3153 * checking to make sure it was reached via the rightmost node
3154 * at each level of the tree.
3156 for (i = depth-1; i >= 0; i--)
3157 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3159 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3160 return ext4_mark_inode_dirty(handle, inode);
3164 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3165 struct ext4_map_blocks *map,
3166 struct ext4_ext_path *path, int flags,
3167 unsigned int allocated, ext4_fsblk_t newblock)
3171 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3173 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3174 "block %llu, max_blocks %u, flags %d, allocated %u",
3175 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3177 ext4_ext_show_leaf(inode, path);
3179 /* get_block() before submit the IO, split the extent */
3180 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3181 ret = ext4_split_unwritten_extents(handle, inode, map,
3184 * Flag the inode(non aio case) or end_io struct (aio case)
3185 * that this IO needs to conversion to written when IO is
3188 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3189 io->flag = EXT4_IO_END_UNWRITTEN;
3190 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3192 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3193 if (ext4_should_dioread_nolock(inode))
3194 map->m_flags |= EXT4_MAP_UNINIT;
3197 /* IO end_io complete, convert the filled extent to written */
3198 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3199 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3202 ext4_update_inode_fsync_trans(handle, inode, 1);
3203 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3209 /* buffered IO case */
3211 * repeat fallocate creation request
3212 * we already have an unwritten extent
3214 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3217 /* buffered READ or buffered write_begin() lookup */
3218 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3220 * We have blocks reserved already. We
3221 * return allocated blocks so that delalloc
3222 * won't do block reservation for us. But
3223 * the buffer head will be unmapped so that
3224 * a read from the block returns 0s.
3226 map->m_flags |= EXT4_MAP_UNWRITTEN;
3230 /* buffered write, writepage time, convert*/
3231 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3233 ext4_update_inode_fsync_trans(handle, inode, 1);
3234 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3246 map->m_flags |= EXT4_MAP_NEW;
3248 * if we allocated more blocks than requested
3249 * we need to make sure we unmap the extra block
3250 * allocated. The actual needed block will get
3251 * unmapped later when we find the buffer_head marked
3254 if (allocated > map->m_len) {
3255 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3256 newblock + map->m_len,
3257 allocated - map->m_len);
3258 allocated = map->m_len;
3262 * If we have done fallocate with the offset that is already
3263 * delayed allocated, we would have block reservation
3264 * and quota reservation done in the delayed write path.
3265 * But fallocate would have already updated quota and block
3266 * count for this offset. So cancel these reservation
3268 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3269 ext4_da_update_reserve_space(inode, allocated, 0);
3272 map->m_flags |= EXT4_MAP_MAPPED;
3274 if (allocated > map->m_len)
3275 allocated = map->m_len;
3276 ext4_ext_show_leaf(inode, path);
3277 map->m_pblk = newblock;
3278 map->m_len = allocated;
3281 ext4_ext_drop_refs(path);
3284 return err ? err : allocated;
3288 * Block allocation/map/preallocation routine for extents based files
3291 * Need to be called with
3292 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3293 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3295 * return > 0, number of of blocks already mapped/allocated
3296 * if create == 0 and these are pre-allocated blocks
3297 * buffer head is unmapped
3298 * otherwise blocks are mapped
3300 * return = 0, if plain look up failed (blocks have not been allocated)
3301 * buffer head is unmapped
3303 * return < 0, error case.
3305 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3306 struct ext4_map_blocks *map, int flags)
3308 struct ext4_ext_path *path = NULL;
3309 struct ext4_extent newex, *ex;
3310 ext4_fsblk_t newblock = 0;
3311 int err = 0, depth, ret;
3312 unsigned int allocated = 0;
3313 unsigned int punched_out = 0;
3314 unsigned int result = 0;
3315 struct ext4_allocation_request ar;
3316 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3317 struct ext4_map_blocks punch_map;
3319 ext_debug("blocks %u/%u requested for inode %lu\n",
3320 map->m_lblk, map->m_len, inode->i_ino);
3321 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3323 /* check in cache */
3324 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3325 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3326 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3327 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3329 * block isn't allocated yet and
3330 * user doesn't want to allocate it
3334 /* we should allocate requested block */
3336 /* block is already allocated */
3337 newblock = map->m_lblk
3338 - le32_to_cpu(newex.ee_block)
3339 + ext4_ext_pblock(&newex);
3340 /* number of remaining blocks in the extent */
3341 allocated = ext4_ext_get_actual_len(&newex) -
3342 (map->m_lblk - le32_to_cpu(newex.ee_block));
3347 /* find extent for this block */
3348 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3350 err = PTR_ERR(path);
3355 depth = ext_depth(inode);
3358 * consistent leaf must not be empty;
3359 * this situation is possible, though, _during_ tree modification;
3360 * this is why assert can't be put in ext4_ext_find_extent()
3362 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3363 EXT4_ERROR_INODE(inode, "bad extent address "
3364 "lblock: %lu, depth: %d pblock %lld",
3365 (unsigned long) map->m_lblk, depth,
3366 path[depth].p_block);
3371 ex = path[depth].p_ext;
3373 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3374 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3375 unsigned short ee_len;
3378 * Uninitialized extents are treated as holes, except that
3379 * we split out initialized portions during a write.
3381 ee_len = ext4_ext_get_actual_len(ex);
3382 /* if found extent covers block, simply return it */
3383 if (in_range(map->m_lblk, ee_block, ee_len)) {
3384 newblock = map->m_lblk - ee_block + ee_start;
3385 /* number of remaining blocks in the extent */
3386 allocated = ee_len - (map->m_lblk - ee_block);
3387 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3388 ee_block, ee_len, newblock);
3390 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3392 * Do not put uninitialized extent
3395 if (!ext4_ext_is_uninitialized(ex)) {
3396 ext4_ext_put_in_cache(inode, ee_block,
3400 ret = ext4_ext_handle_uninitialized_extents(
3401 handle, inode, map, path, flags,
3402 allocated, newblock);
3407 * Punch out the map length, but only to the
3410 punched_out = allocated < map->m_len ?
3411 allocated : map->m_len;
3414 * Sense extents need to be converted to
3415 * uninitialized, they must fit in an
3416 * uninitialized extent
3418 if (punched_out > EXT_UNINIT_MAX_LEN)
3419 punched_out = EXT_UNINIT_MAX_LEN;
3421 punch_map.m_lblk = map->m_lblk;
3422 punch_map.m_pblk = newblock;
3423 punch_map.m_len = punched_out;
3424 punch_map.m_flags = 0;
3426 /* Check to see if the extent needs to be split */
3427 if (punch_map.m_len != ee_len ||
3428 punch_map.m_lblk != ee_block) {
3430 ret = ext4_split_extent(handle, inode,
3431 path, &punch_map, 0,
3432 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3433 EXT4_GET_BLOCKS_PRE_IO);
3440 * find extent for the block at
3441 * the start of the hole
3443 ext4_ext_drop_refs(path);
3446 path = ext4_ext_find_extent(inode,
3449 err = PTR_ERR(path);
3454 depth = ext_depth(inode);
3455 ex = path[depth].p_ext;
3456 ee_len = ext4_ext_get_actual_len(ex);
3457 ee_block = le32_to_cpu(ex->ee_block);
3458 ee_start = ext4_ext_pblock(ex);
3462 ext4_ext_mark_uninitialized(ex);
3464 ext4_ext_invalidate_cache(inode);
3466 err = ext4_ext_rm_leaf(handle, inode, path,
3467 map->m_lblk, map->m_lblk + punched_out);
3469 if (!err && path->p_hdr->eh_entries == 0) {
3471 * Punch hole freed all of this sub tree,
3472 * so we need to correct eh_depth
3474 err = ext4_ext_get_access(handle, inode, path);
3476 ext_inode_hdr(inode)->eh_depth = 0;
3477 ext_inode_hdr(inode)->eh_max =
3478 cpu_to_le16(ext4_ext_space_root(
3481 err = ext4_ext_dirty(
3482 handle, inode, path);
3491 * requested block isn't allocated yet;
3492 * we couldn't try to create block if create flag is zero
3494 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3496 * put just found gap into cache to speed up
3497 * subsequent requests
3499 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3503 * Okay, we need to do block allocation.
3506 /* find neighbour allocated blocks */
3507 ar.lleft = map->m_lblk;
3508 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3511 ar.lright = map->m_lblk;
3512 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3517 * See if request is beyond maximum number of blocks we can have in
3518 * a single extent. For an initialized extent this limit is
3519 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3520 * EXT_UNINIT_MAX_LEN.
3522 if (map->m_len > EXT_INIT_MAX_LEN &&
3523 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3524 map->m_len = EXT_INIT_MAX_LEN;
3525 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3526 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3527 map->m_len = EXT_UNINIT_MAX_LEN;
3529 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3530 newex.ee_block = cpu_to_le32(map->m_lblk);
3531 newex.ee_len = cpu_to_le16(map->m_len);
3532 err = ext4_ext_check_overlap(inode, &newex, path);
3534 allocated = ext4_ext_get_actual_len(&newex);
3536 allocated = map->m_len;
3538 /* allocate new block */
3540 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3541 ar.logical = map->m_lblk;
3543 if (S_ISREG(inode->i_mode))
3544 ar.flags = EXT4_MB_HINT_DATA;
3546 /* disable in-core preallocation for non-regular files */
3548 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
3549 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
3550 newblock = ext4_mb_new_blocks(handle, &ar, &err);
3553 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3554 ar.goal, newblock, allocated);
3556 /* try to insert new extent into found leaf and return */
3557 ext4_ext_store_pblock(&newex, newblock);
3558 newex.ee_len = cpu_to_le16(ar.len);
3559 /* Mark uninitialized */
3560 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3561 ext4_ext_mark_uninitialized(&newex);
3563 * io_end structure was created for every IO write to an
3564 * uninitialized extent. To avoid unnecessary conversion,
3565 * here we flag the IO that really needs the conversion.
3566 * For non asycn direct IO case, flag the inode state
3567 * that we need to perform conversion when IO is done.
3569 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3570 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3571 io->flag = EXT4_IO_END_UNWRITTEN;
3572 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3574 ext4_set_inode_state(inode,
3575 EXT4_STATE_DIO_UNWRITTEN);
3577 if (ext4_should_dioread_nolock(inode))
3578 map->m_flags |= EXT4_MAP_UNINIT;
3581 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3583 err = ext4_ext_insert_extent(handle, inode, path,
3586 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
3587 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
3588 /* free data blocks we just allocated */
3589 /* not a good idea to call discard here directly,
3590 * but otherwise we'd need to call it every free() */
3591 ext4_discard_preallocations(inode);
3592 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3593 ext4_ext_get_actual_len(&newex), fb_flags);
3597 /* previous routine could use block we allocated */
3598 newblock = ext4_ext_pblock(&newex);
3599 allocated = ext4_ext_get_actual_len(&newex);
3600 if (allocated > map->m_len)
3601 allocated = map->m_len;
3602 map->m_flags |= EXT4_MAP_NEW;
3605 * Update reserved blocks/metadata blocks after successful
3606 * block allocation which had been deferred till now.
3608 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3609 ext4_da_update_reserve_space(inode, allocated, 1);
3612 * Cache the extent and update transaction to commit on fdatasync only
3613 * when it is _not_ an uninitialized extent.
3615 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3616 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3617 ext4_update_inode_fsync_trans(handle, inode, 1);
3619 ext4_update_inode_fsync_trans(handle, inode, 0);
3621 if (allocated > map->m_len)
3622 allocated = map->m_len;
3623 ext4_ext_show_leaf(inode, path);
3624 map->m_flags |= EXT4_MAP_MAPPED;
3625 map->m_pblk = newblock;
3626 map->m_len = allocated;
3629 ext4_ext_drop_refs(path);
3632 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3633 newblock, map->m_len, err ? err : allocated);
3635 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
3636 punched_out : allocated;
3638 return err ? err : result;
3641 void ext4_ext_truncate(struct inode *inode)
3643 struct address_space *mapping = inode->i_mapping;
3644 struct super_block *sb = inode->i_sb;
3645 ext4_lblk_t last_block;
3650 * finish any pending end_io work so we won't run the risk of
3651 * converting any truncated blocks to initialized later
3653 ext4_flush_completed_IO(inode);
3656 * probably first extent we're gonna free will be last in block
3658 err = ext4_writepage_trans_blocks(inode);
3659 handle = ext4_journal_start(inode, err);
3663 if (inode->i_size & (sb->s_blocksize - 1))
3664 ext4_block_truncate_page(handle, mapping, inode->i_size);
3666 if (ext4_orphan_add(handle, inode))
3669 down_write(&EXT4_I(inode)->i_data_sem);
3670 ext4_ext_invalidate_cache(inode);
3672 ext4_discard_preallocations(inode);
3675 * TODO: optimization is possible here.
3676 * Probably we need not scan at all,
3677 * because page truncation is enough.
3680 /* we have to know where to truncate from in crash case */
3681 EXT4_I(inode)->i_disksize = inode->i_size;
3682 ext4_mark_inode_dirty(handle, inode);
3684 last_block = (inode->i_size + sb->s_blocksize - 1)
3685 >> EXT4_BLOCK_SIZE_BITS(sb);
3686 err = ext4_ext_remove_space(inode, last_block);
3688 /* In a multi-transaction truncate, we only make the final
3689 * transaction synchronous.
3692 ext4_handle_sync(handle);
3694 up_write(&EXT4_I(inode)->i_data_sem);
3698 * If this was a simple ftruncate() and the file will remain alive,
3699 * then we need to clear up the orphan record which we created above.
3700 * However, if this was a real unlink then we were called by
3701 * ext4_delete_inode(), and we allow that function to clean up the
3702 * orphan info for us.
3705 ext4_orphan_del(handle, inode);
3707 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3708 ext4_mark_inode_dirty(handle, inode);
3709 ext4_journal_stop(handle);
3712 static void ext4_falloc_update_inode(struct inode *inode,
3713 int mode, loff_t new_size, int update_ctime)
3715 struct timespec now;
3718 now = current_fs_time(inode->i_sb);
3719 if (!timespec_equal(&inode->i_ctime, &now))
3720 inode->i_ctime = now;
3723 * Update only when preallocation was requested beyond
3726 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3727 if (new_size > i_size_read(inode))
3728 i_size_write(inode, new_size);
3729 if (new_size > EXT4_I(inode)->i_disksize)
3730 ext4_update_i_disksize(inode, new_size);
3733 * Mark that we allocate beyond EOF so the subsequent truncate
3734 * can proceed even if the new size is the same as i_size.
3736 if (new_size > i_size_read(inode))
3737 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3743 * preallocate space for a file. This implements ext4's fallocate file
3744 * operation, which gets called from sys_fallocate system call.
3745 * For block-mapped files, posix_fallocate should fall back to the method
3746 * of writing zeroes to the required new blocks (the same behavior which is
3747 * expected for file systems which do not support fallocate() system call).
3749 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
3751 struct inode *inode = file->f_path.dentry->d_inode;
3754 unsigned int max_blocks;
3758 struct ext4_map_blocks map;
3759 unsigned int credits, blkbits = inode->i_blkbits;
3762 * currently supporting (pre)allocate mode for extent-based
3765 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3768 /* Return error if mode is not supported */
3769 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3772 if (mode & FALLOC_FL_PUNCH_HOLE)
3773 return ext4_punch_hole(file, offset, len);
3775 trace_ext4_fallocate_enter(inode, offset, len, mode);
3776 map.m_lblk = offset >> blkbits;
3778 * We can't just convert len to max_blocks because
3779 * If blocksize = 4096 offset = 3072 and len = 2048
3781 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3784 * credits to insert 1 extent into extent tree
3786 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3787 mutex_lock(&inode->i_mutex);
3788 ret = inode_newsize_ok(inode, (len + offset));
3790 mutex_unlock(&inode->i_mutex);
3791 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
3795 while (ret >= 0 && ret < max_blocks) {
3796 map.m_lblk = map.m_lblk + ret;
3797 map.m_len = max_blocks = max_blocks - ret;
3798 handle = ext4_journal_start(inode, credits);
3799 if (IS_ERR(handle)) {
3800 ret = PTR_ERR(handle);
3803 ret = ext4_map_blocks(handle, inode, &map,
3804 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
3805 EXT4_GET_BLOCKS_NO_NORMALIZE);
3809 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3810 "returned error inode#%lu, block=%u, "
3811 "max_blocks=%u", __func__,
3812 inode->i_ino, map.m_lblk, max_blocks);
3814 ext4_mark_inode_dirty(handle, inode);
3815 ret2 = ext4_journal_stop(handle);
3818 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3819 blkbits) >> blkbits))
3820 new_size = offset + len;
3822 new_size = (map.m_lblk + ret) << blkbits;
3824 ext4_falloc_update_inode(inode, mode, new_size,
3825 (map.m_flags & EXT4_MAP_NEW));
3826 ext4_mark_inode_dirty(handle, inode);
3827 ret2 = ext4_journal_stop(handle);
3831 if (ret == -ENOSPC &&
3832 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3836 mutex_unlock(&inode->i_mutex);
3837 trace_ext4_fallocate_exit(inode, offset, max_blocks,
3838 ret > 0 ? ret2 : ret);
3839 return ret > 0 ? ret2 : ret;
3843 * This function convert a range of blocks to written extents
3844 * The caller of this function will pass the start offset and the size.
3845 * all unwritten extents within this range will be converted to
3848 * This function is called from the direct IO end io call back
3849 * function, to convert the fallocated extents after IO is completed.
3850 * Returns 0 on success.
3852 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3856 unsigned int max_blocks;
3859 struct ext4_map_blocks map;
3860 unsigned int credits, blkbits = inode->i_blkbits;
3862 map.m_lblk = offset >> blkbits;
3864 * We can't just convert len to max_blocks because
3865 * If blocksize = 4096 offset = 3072 and len = 2048
3867 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3870 * credits to insert 1 extent into extent tree
3872 credits = ext4_chunk_trans_blocks(inode, max_blocks);
3873 while (ret >= 0 && ret < max_blocks) {
3875 map.m_len = (max_blocks -= ret);
3876 handle = ext4_journal_start(inode, credits);
3877 if (IS_ERR(handle)) {
3878 ret = PTR_ERR(handle);
3881 ret = ext4_map_blocks(handle, inode, &map,
3882 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3885 printk(KERN_ERR "%s: ext4_ext_map_blocks "
3886 "returned error inode#%lu, block=%u, "
3887 "max_blocks=%u", __func__,
3888 inode->i_ino, map.m_lblk, map.m_len);
3890 ext4_mark_inode_dirty(handle, inode);
3891 ret2 = ext4_journal_stop(handle);
3892 if (ret <= 0 || ret2 )
3895 return ret > 0 ? ret2 : ret;
3899 * Callback function called for each extent to gather FIEMAP information.
3901 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
3902 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3910 struct fiemap_extent_info *fieinfo = data;
3911 unsigned char blksize_bits;
3913 blksize_bits = inode->i_sb->s_blocksize_bits;
3914 logical = (__u64)newex->ec_block << blksize_bits;
3916 if (newex->ec_start == 0) {
3918 * No extent in extent-tree contains block @newex->ec_start,
3919 * then the block may stay in 1)a hole or 2)delayed-extent.
3921 * Holes or delayed-extents are processed as follows.
3922 * 1. lookup dirty pages with specified range in pagecache.
3923 * If no page is got, then there is no delayed-extent and
3924 * return with EXT_CONTINUE.
3925 * 2. find the 1st mapped buffer,
3926 * 3. check if the mapped buffer is both in the request range
3927 * and a delayed buffer. If not, there is no delayed-extent,
3929 * 4. a delayed-extent is found, the extent will be collected.
3931 ext4_lblk_t end = 0;
3932 pgoff_t last_offset;
3935 pgoff_t start_index = 0;
3936 struct page **pages = NULL;
3937 struct buffer_head *bh = NULL;
3938 struct buffer_head *head = NULL;
3939 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
3941 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
3945 offset = logical >> PAGE_SHIFT;
3947 last_offset = offset;
3949 ret = find_get_pages_tag(inode->i_mapping, &offset,
3950 PAGECACHE_TAG_DIRTY, nr_pages, pages);
3952 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
3953 /* First time, try to find a mapped buffer. */
3956 for (index = 0; index < ret; index++)
3957 page_cache_release(pages[index]);
3960 return EXT_CONTINUE;
3965 /* Try to find the 1st mapped buffer. */
3966 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
3968 if (!page_has_buffers(pages[index]))
3970 head = page_buffers(pages[index]);
3977 if (end >= newex->ec_block +
3979 /* The buffer is out of
3980 * the request range.
3984 if (buffer_mapped(bh) &&
3985 end >= newex->ec_block) {
3986 start_index = index - 1;
3987 /* get the 1st mapped buffer. */
3988 goto found_mapped_buffer;
3991 bh = bh->b_this_page;
3993 } while (bh != head);
3995 /* No mapped buffer in the range found in this page,
3996 * We need to look up next page.
3999 /* There is no page left, but we need to limit
4002 newex->ec_len = end - newex->ec_block;
4007 /*Find contiguous delayed buffers. */
4008 if (ret > 0 && pages[0]->index == last_offset)
4009 head = page_buffers(pages[0]);
4015 found_mapped_buffer:
4016 if (bh != NULL && buffer_delay(bh)) {
4017 /* 1st or contiguous delayed buffer found. */
4018 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4020 * 1st delayed buffer found, record
4021 * the start of extent.
4023 flags |= FIEMAP_EXTENT_DELALLOC;
4024 newex->ec_block = end;
4025 logical = (__u64)end << blksize_bits;
4027 /* Find contiguous delayed buffers. */
4029 if (!buffer_delay(bh))
4030 goto found_delayed_extent;
4031 bh = bh->b_this_page;
4033 } while (bh != head);
4035 for (; index < ret; index++) {
4036 if (!page_has_buffers(pages[index])) {
4040 head = page_buffers(pages[index]);
4046 if (pages[index]->index !=
4047 pages[start_index]->index + index
4049 /* Blocks are not contiguous. */
4055 if (!buffer_delay(bh))
4056 /* Delayed-extent ends. */
4057 goto found_delayed_extent;
4058 bh = bh->b_this_page;
4060 } while (bh != head);
4062 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4066 found_delayed_extent:
4067 newex->ec_len = min(end - newex->ec_block,
4068 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4069 if (ret == nr_pages && bh != NULL &&
4070 newex->ec_len < EXT_INIT_MAX_LEN &&
4072 /* Have not collected an extent and continue. */
4073 for (index = 0; index < ret; index++)
4074 page_cache_release(pages[index]);
4078 for (index = 0; index < ret; index++)
4079 page_cache_release(pages[index]);
4083 physical = (__u64)newex->ec_start << blksize_bits;
4084 length = (__u64)newex->ec_len << blksize_bits;
4086 if (ex && ext4_ext_is_uninitialized(ex))
4087 flags |= FIEMAP_EXTENT_UNWRITTEN;
4089 if (next == EXT_MAX_BLOCKS)
4090 flags |= FIEMAP_EXTENT_LAST;
4092 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4098 return EXT_CONTINUE;
4101 /* fiemap flags we can handle specified here */
4102 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4104 static int ext4_xattr_fiemap(struct inode *inode,
4105 struct fiemap_extent_info *fieinfo)
4109 __u32 flags = FIEMAP_EXTENT_LAST;
4110 int blockbits = inode->i_sb->s_blocksize_bits;
4114 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4115 struct ext4_iloc iloc;
4116 int offset; /* offset of xattr in inode */
4118 error = ext4_get_inode_loc(inode, &iloc);
4121 physical = iloc.bh->b_blocknr << blockbits;
4122 offset = EXT4_GOOD_OLD_INODE_SIZE +
4123 EXT4_I(inode)->i_extra_isize;
4125 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4126 flags |= FIEMAP_EXTENT_DATA_INLINE;
4128 } else { /* external block */
4129 physical = EXT4_I(inode)->i_file_acl << blockbits;
4130 length = inode->i_sb->s_blocksize;
4134 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4136 return (error < 0 ? error : 0);
4140 * ext4_ext_punch_hole
4142 * Punches a hole of "length" bytes in a file starting
4145 * @inode: The inode of the file to punch a hole in
4146 * @offset: The starting byte offset of the hole
4147 * @length: The length of the hole
4149 * Returns the number of blocks removed or negative on err
4151 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4153 struct inode *inode = file->f_path.dentry->d_inode;
4154 struct super_block *sb = inode->i_sb;
4155 struct ext4_ext_cache cache_ex;
4156 ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4157 struct address_space *mapping = inode->i_mapping;
4158 struct ext4_map_blocks map;
4160 loff_t first_block_offset, last_block_offset, block_len;
4161 loff_t first_page, last_page, first_page_offset, last_page_offset;
4162 int ret, credits, blocks_released, err = 0;
4164 first_block = (offset + sb->s_blocksize - 1) >>
4165 EXT4_BLOCK_SIZE_BITS(sb);
4166 last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4168 first_block_offset = first_block << EXT4_BLOCK_SIZE_BITS(sb);
4169 last_block_offset = last_block << EXT4_BLOCK_SIZE_BITS(sb);
4171 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4172 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4174 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4175 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4178 * Write out all dirty pages to avoid race conditions
4179 * Then release them.
4181 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4182 err = filemap_write_and_wait_range(mapping,
4183 first_page_offset == 0 ? 0 : first_page_offset-1,
4190 /* Now release the pages */
4191 if (last_page_offset > first_page_offset) {
4192 truncate_inode_pages_range(mapping, first_page_offset,
4193 last_page_offset-1);
4196 /* finish any pending end_io work */
4197 ext4_flush_completed_IO(inode);
4199 credits = ext4_writepage_trans_blocks(inode);
4200 handle = ext4_journal_start(inode, credits);
4202 return PTR_ERR(handle);
4204 err = ext4_orphan_add(handle, inode);
4209 * Now we need to zero out the un block aligned data.
4210 * If the file is smaller than a block, just
4211 * zero out the middle
4213 if (first_block > last_block)
4214 ext4_block_zero_page_range(handle, mapping, offset, length);
4216 /* zero out the head of the hole before the first block */
4217 block_len = first_block_offset - offset;
4219 ext4_block_zero_page_range(handle, mapping,
4222 /* zero out the tail of the hole after the last block */
4223 block_len = offset + length - last_block_offset;
4224 if (block_len > 0) {
4225 ext4_block_zero_page_range(handle, mapping,
4226 last_block_offset, block_len);
4230 /* If there are no blocks to remove, return now */
4231 if (first_block >= last_block)
4234 down_write(&EXT4_I(inode)->i_data_sem);
4235 ext4_ext_invalidate_cache(inode);
4236 ext4_discard_preallocations(inode);
4239 * Loop over all the blocks and identify blocks
4240 * that need to be punched out
4242 iblock = first_block;
4243 blocks_released = 0;
4244 while (iblock < last_block) {
4245 max_blocks = last_block - iblock;
4247 memset(&map, 0, sizeof(map));
4248 map.m_lblk = iblock;
4249 map.m_len = max_blocks;
4250 ret = ext4_ext_map_blocks(handle, inode, &map,
4251 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4254 blocks_released += ret;
4256 } else if (ret == 0) {
4258 * If map blocks could not find the block,
4259 * then it is in a hole. If the hole was
4260 * not already cached, then map blocks should
4261 * put it in the cache. So we can get the hole
4264 memset(&cache_ex, 0, sizeof(cache_ex));
4265 if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4266 !cache_ex.ec_start) {
4268 /* The hole is cached */
4269 num_blocks = cache_ex.ec_block +
4270 cache_ex.ec_len - iblock;
4273 /* The block could not be identified */
4278 /* Map blocks error */
4283 if (num_blocks == 0) {
4284 /* This condition should never happen */
4285 ext_debug("Block lookup failed");
4290 iblock += num_blocks;
4293 if (blocks_released > 0) {
4294 ext4_ext_invalidate_cache(inode);
4295 ext4_discard_preallocations(inode);
4299 ext4_handle_sync(handle);
4301 up_write(&EXT4_I(inode)->i_data_sem);
4304 ext4_orphan_del(handle, inode);
4305 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4306 ext4_mark_inode_dirty(handle, inode);
4307 ext4_journal_stop(handle);
4310 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4311 __u64 start, __u64 len)
4313 ext4_lblk_t start_blk;
4316 /* fallback to generic here if not in extents fmt */
4317 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4318 return generic_block_fiemap(inode, fieinfo, start, len,
4321 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4324 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4325 error = ext4_xattr_fiemap(inode, fieinfo);
4327 ext4_lblk_t len_blks;
4330 start_blk = start >> inode->i_sb->s_blocksize_bits;
4331 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4332 if (last_blk >= EXT_MAX_BLOCKS)
4333 last_blk = EXT_MAX_BLOCKS-1;
4334 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4337 * Walk the extent tree gathering extent information.
4338 * ext4_ext_fiemap_cb will push extents back to user.
4340 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4341 ext4_ext_fiemap_cb, fieinfo);