2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 static int ext4_split_extent(handle_t *handle,
50 struct ext4_ext_path *path,
51 struct ext4_map_blocks *map,
55 static int ext4_ext_truncate_extend_restart(handle_t *handle,
61 if (!ext4_handle_valid(handle))
63 if (handle->h_buffer_credits > needed)
65 err = ext4_journal_extend(handle, needed);
68 err = ext4_truncate_restart_trans(handle, inode, needed);
80 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
81 struct ext4_ext_path *path)
84 /* path points to block */
85 return ext4_journal_get_write_access(handle, path->p_bh);
87 /* path points to leaf/index in inode body */
88 /* we use in-core data, no need to protect them */
98 #define ext4_ext_dirty(handle, inode, path) \
99 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
100 static int __ext4_ext_dirty(const char *where, unsigned int line,
101 handle_t *handle, struct inode *inode,
102 struct ext4_ext_path *path)
106 /* path points to block */
107 err = __ext4_handle_dirty_metadata(where, line, handle,
110 /* path points to leaf/index in inode body */
111 err = ext4_mark_inode_dirty(handle, inode);
116 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
117 struct ext4_ext_path *path,
121 int depth = path->p_depth;
122 struct ext4_extent *ex;
125 * Try to predict block placement assuming that we are
126 * filling in a file which will eventually be
127 * non-sparse --- i.e., in the case of libbfd writing
128 * an ELF object sections out-of-order but in a way
129 * the eventually results in a contiguous object or
130 * executable file, or some database extending a table
131 * space file. However, this is actually somewhat
132 * non-ideal if we are writing a sparse file such as
133 * qemu or KVM writing a raw image file that is going
134 * to stay fairly sparse, since it will end up
135 * fragmenting the file system's free space. Maybe we
136 * should have some hueristics or some way to allow
137 * userspace to pass a hint to file system,
138 * especially if the latter case turns out to be
141 ex = path[depth].p_ext;
143 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
144 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
146 if (block > ext_block)
147 return ext_pblk + (block - ext_block);
149 return ext_pblk - (ext_block - block);
152 /* it looks like index is empty;
153 * try to find starting block from index itself */
154 if (path[depth].p_bh)
155 return path[depth].p_bh->b_blocknr;
158 /* OK. use inode's group */
159 return ext4_inode_to_goal_block(inode);
163 * Allocation for a meta data block
166 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
167 struct ext4_ext_path *path,
168 struct ext4_extent *ex, int *err, unsigned int flags)
170 ext4_fsblk_t goal, newblock;
172 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
173 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
178 static inline int ext4_ext_space_block(struct inode *inode, int check)
182 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
183 / sizeof(struct ext4_extent);
184 #ifdef AGGRESSIVE_TEST
185 if (!check && size > 6)
191 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
195 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
196 / sizeof(struct ext4_extent_idx);
197 #ifdef AGGRESSIVE_TEST
198 if (!check && size > 5)
204 static inline int ext4_ext_space_root(struct inode *inode, int check)
208 size = sizeof(EXT4_I(inode)->i_data);
209 size -= sizeof(struct ext4_extent_header);
210 size /= sizeof(struct ext4_extent);
211 #ifdef AGGRESSIVE_TEST
212 if (!check && size > 3)
218 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
222 size = sizeof(EXT4_I(inode)->i_data);
223 size -= sizeof(struct ext4_extent_header);
224 size /= sizeof(struct ext4_extent_idx);
225 #ifdef AGGRESSIVE_TEST
226 if (!check && size > 4)
233 * Calculate the number of metadata blocks needed
234 * to allocate @blocks
235 * Worse case is one block per extent
237 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
239 struct ext4_inode_info *ei = EXT4_I(inode);
242 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
243 / sizeof(struct ext4_extent_idx));
246 * If the new delayed allocation block is contiguous with the
247 * previous da block, it can share index blocks with the
248 * previous block, so we only need to allocate a new index
249 * block every idxs leaf blocks. At ldxs**2 blocks, we need
250 * an additional index block, and at ldxs**3 blocks, yet
251 * another index blocks.
253 if (ei->i_da_metadata_calc_len &&
254 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
257 if ((ei->i_da_metadata_calc_len % idxs) == 0)
259 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
261 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
263 ei->i_da_metadata_calc_len = 0;
265 ei->i_da_metadata_calc_len++;
266 ei->i_da_metadata_calc_last_lblock++;
271 * In the worst case we need a new set of index blocks at
272 * every level of the inode's extent tree.
274 ei->i_da_metadata_calc_len = 1;
275 ei->i_da_metadata_calc_last_lblock = lblock;
276 return ext_depth(inode) + 1;
280 ext4_ext_max_entries(struct inode *inode, int depth)
284 if (depth == ext_depth(inode)) {
286 max = ext4_ext_space_root(inode, 1);
288 max = ext4_ext_space_root_idx(inode, 1);
291 max = ext4_ext_space_block(inode, 1);
293 max = ext4_ext_space_block_idx(inode, 1);
299 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
301 ext4_fsblk_t block = ext4_ext_pblock(ext);
302 int len = ext4_ext_get_actual_len(ext);
306 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
309 static int ext4_valid_extent_idx(struct inode *inode,
310 struct ext4_extent_idx *ext_idx)
312 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
314 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
317 static int ext4_valid_extent_entries(struct inode *inode,
318 struct ext4_extent_header *eh,
321 unsigned short entries;
322 if (eh->eh_entries == 0)
325 entries = le16_to_cpu(eh->eh_entries);
329 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
331 if (!ext4_valid_extent(inode, ext))
337 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
339 if (!ext4_valid_extent_idx(inode, ext_idx))
348 static int __ext4_ext_check(const char *function, unsigned int line,
349 struct inode *inode, struct ext4_extent_header *eh,
352 const char *error_msg;
355 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
356 error_msg = "invalid magic";
359 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
360 error_msg = "unexpected eh_depth";
363 if (unlikely(eh->eh_max == 0)) {
364 error_msg = "invalid eh_max";
367 max = ext4_ext_max_entries(inode, depth);
368 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
369 error_msg = "too large eh_max";
372 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
373 error_msg = "invalid eh_entries";
376 if (!ext4_valid_extent_entries(inode, eh, depth)) {
377 error_msg = "invalid extent entries";
383 ext4_error_inode(inode, function, line, 0,
384 "bad header/extent: %s - magic %x, "
385 "entries %u, max %u(%u), depth %u(%u)",
386 error_msg, le16_to_cpu(eh->eh_magic),
387 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
388 max, le16_to_cpu(eh->eh_depth), depth);
393 #define ext4_ext_check(inode, eh, depth) \
394 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
396 int ext4_ext_check_inode(struct inode *inode)
398 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
402 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
404 int k, l = path->p_depth;
407 for (k = 0; k <= l; k++, path++) {
409 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
410 ext4_idx_pblock(path->p_idx));
411 } else if (path->p_ext) {
412 ext_debug(" %d:[%d]%d:%llu ",
413 le32_to_cpu(path->p_ext->ee_block),
414 ext4_ext_is_uninitialized(path->p_ext),
415 ext4_ext_get_actual_len(path->p_ext),
416 ext4_ext_pblock(path->p_ext));
423 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
425 int depth = ext_depth(inode);
426 struct ext4_extent_header *eh;
427 struct ext4_extent *ex;
433 eh = path[depth].p_hdr;
434 ex = EXT_FIRST_EXTENT(eh);
436 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
438 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
439 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
440 ext4_ext_is_uninitialized(ex),
441 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
446 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
447 ext4_fsblk_t newblock, int level)
449 int depth = ext_depth(inode);
450 struct ext4_extent *ex;
452 if (depth != level) {
453 struct ext4_extent_idx *idx;
454 idx = path[level].p_idx;
455 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
456 ext_debug("%d: move %d:%llu in new index %llu\n", level,
457 le32_to_cpu(idx->ei_block),
458 ext4_idx_pblock(idx),
466 ex = path[depth].p_ext;
467 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
468 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
469 le32_to_cpu(ex->ee_block),
471 ext4_ext_is_uninitialized(ex),
472 ext4_ext_get_actual_len(ex),
479 #define ext4_ext_show_path(inode, path)
480 #define ext4_ext_show_leaf(inode, path)
481 #define ext4_ext_show_move(inode, path, newblock, level)
484 void ext4_ext_drop_refs(struct ext4_ext_path *path)
486 int depth = path->p_depth;
489 for (i = 0; i <= depth; i++, path++)
497 * ext4_ext_binsearch_idx:
498 * binary search for the closest index of the given block
499 * the header must be checked before calling this
502 ext4_ext_binsearch_idx(struct inode *inode,
503 struct ext4_ext_path *path, ext4_lblk_t block)
505 struct ext4_extent_header *eh = path->p_hdr;
506 struct ext4_extent_idx *r, *l, *m;
509 ext_debug("binsearch for %u(idx): ", block);
511 l = EXT_FIRST_INDEX(eh) + 1;
512 r = EXT_LAST_INDEX(eh);
515 if (block < le32_to_cpu(m->ei_block))
519 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
520 m, le32_to_cpu(m->ei_block),
521 r, le32_to_cpu(r->ei_block));
525 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
526 ext4_idx_pblock(path->p_idx));
528 #ifdef CHECK_BINSEARCH
530 struct ext4_extent_idx *chix, *ix;
533 chix = ix = EXT_FIRST_INDEX(eh);
534 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
536 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
537 printk(KERN_DEBUG "k=%d, ix=0x%p, "
539 ix, EXT_FIRST_INDEX(eh));
540 printk(KERN_DEBUG "%u <= %u\n",
541 le32_to_cpu(ix->ei_block),
542 le32_to_cpu(ix[-1].ei_block));
544 BUG_ON(k && le32_to_cpu(ix->ei_block)
545 <= le32_to_cpu(ix[-1].ei_block));
546 if (block < le32_to_cpu(ix->ei_block))
550 BUG_ON(chix != path->p_idx);
557 * ext4_ext_binsearch:
558 * binary search for closest extent of the given block
559 * the header must be checked before calling this
562 ext4_ext_binsearch(struct inode *inode,
563 struct ext4_ext_path *path, ext4_lblk_t block)
565 struct ext4_extent_header *eh = path->p_hdr;
566 struct ext4_extent *r, *l, *m;
568 if (eh->eh_entries == 0) {
570 * this leaf is empty:
571 * we get such a leaf in split/add case
576 ext_debug("binsearch for %u: ", block);
578 l = EXT_FIRST_EXTENT(eh) + 1;
579 r = EXT_LAST_EXTENT(eh);
583 if (block < le32_to_cpu(m->ee_block))
587 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
588 m, le32_to_cpu(m->ee_block),
589 r, le32_to_cpu(r->ee_block));
593 ext_debug(" -> %d:%llu:[%d]%d ",
594 le32_to_cpu(path->p_ext->ee_block),
595 ext4_ext_pblock(path->p_ext),
596 ext4_ext_is_uninitialized(path->p_ext),
597 ext4_ext_get_actual_len(path->p_ext));
599 #ifdef CHECK_BINSEARCH
601 struct ext4_extent *chex, *ex;
604 chex = ex = EXT_FIRST_EXTENT(eh);
605 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
606 BUG_ON(k && le32_to_cpu(ex->ee_block)
607 <= le32_to_cpu(ex[-1].ee_block));
608 if (block < le32_to_cpu(ex->ee_block))
612 BUG_ON(chex != path->p_ext);
618 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
620 struct ext4_extent_header *eh;
622 eh = ext_inode_hdr(inode);
625 eh->eh_magic = EXT4_EXT_MAGIC;
626 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
627 ext4_mark_inode_dirty(handle, inode);
628 ext4_ext_invalidate_cache(inode);
632 struct ext4_ext_path *
633 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
634 struct ext4_ext_path *path)
636 struct ext4_extent_header *eh;
637 struct buffer_head *bh;
638 short int depth, i, ppos = 0, alloc = 0;
641 eh = ext_inode_hdr(inode);
642 depth = ext_depth(inode);
644 /* account possible depth increase */
646 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
649 return ERR_PTR(-ENOMEM);
656 /* walk through the tree */
658 int need_to_validate = 0;
660 ext_debug("depth %d: num %d, max %d\n",
661 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
663 ext4_ext_binsearch_idx(inode, path + ppos, block);
664 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
665 path[ppos].p_depth = i;
666 path[ppos].p_ext = NULL;
668 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
673 if (!bh_uptodate_or_lock(bh)) {
674 trace_ext4_ext_load_extent(inode, block,
676 ret = bh_submit_read(bh);
681 /* validate the extent entries */
682 need_to_validate = 1;
684 eh = ext_block_hdr(bh);
686 if (unlikely(ppos > depth)) {
688 EXT4_ERROR_INODE(inode,
689 "ppos %d > depth %d", ppos, depth);
693 path[ppos].p_bh = bh;
694 path[ppos].p_hdr = eh;
697 ret = need_to_validate ? ext4_ext_check(inode, eh, i) : 0;
702 path[ppos].p_depth = i;
703 path[ppos].p_ext = NULL;
704 path[ppos].p_idx = NULL;
707 ext4_ext_binsearch(inode, path + ppos, block);
708 /* if not an empty leaf */
709 if (path[ppos].p_ext)
710 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
712 ext4_ext_show_path(inode, path);
717 ext4_ext_drop_refs(path);
724 * ext4_ext_insert_index:
725 * insert new index [@logical;@ptr] into the block at @curp;
726 * check where to insert: before @curp or after @curp
728 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
729 struct ext4_ext_path *curp,
730 int logical, ext4_fsblk_t ptr)
732 struct ext4_extent_idx *ix;
735 err = ext4_ext_get_access(handle, inode, curp);
739 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
740 EXT4_ERROR_INODE(inode,
741 "logical %d == ei_block %d!",
742 logical, le32_to_cpu(curp->p_idx->ei_block));
746 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
747 >= le16_to_cpu(curp->p_hdr->eh_max))) {
748 EXT4_ERROR_INODE(inode,
749 "eh_entries %d >= eh_max %d!",
750 le16_to_cpu(curp->p_hdr->eh_entries),
751 le16_to_cpu(curp->p_hdr->eh_max));
755 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
757 ext_debug("insert new index %d after: %llu\n", logical, ptr);
758 ix = curp->p_idx + 1;
761 ext_debug("insert new index %d before: %llu\n", logical, ptr);
765 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
768 ext_debug("insert new index %d: "
769 "move %d indices from 0x%p to 0x%p\n",
770 logical, len, ix, ix + 1);
771 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
774 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
775 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
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_extent *newext)
1047 struct ext4_extent_header *neh;
1048 struct buffer_head *bh;
1049 ext4_fsblk_t newblock;
1052 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1053 newext, &err, flags);
1057 bh = sb_getblk(inode->i_sb, newblock);
1062 err = ext4_journal_get_create_access(handle, bh);
1068 /* move top-level index/leaf into new block */
1069 memmove(bh->b_data, EXT4_I(inode)->i_data,
1070 sizeof(EXT4_I(inode)->i_data));
1072 /* set size of new block */
1073 neh = ext_block_hdr(bh);
1074 /* old root could have indexes or leaves
1075 * so calculate e_max right way */
1076 if (ext_depth(inode))
1077 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1079 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1080 neh->eh_magic = EXT4_EXT_MAGIC;
1081 set_buffer_uptodate(bh);
1084 err = ext4_handle_dirty_metadata(handle, inode, bh);
1088 /* Update top-level index: num,max,pointer */
1089 neh = ext_inode_hdr(inode);
1090 neh->eh_entries = cpu_to_le16(1);
1091 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1092 if (neh->eh_depth == 0) {
1093 /* Root extent block becomes index block */
1094 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1095 EXT_FIRST_INDEX(neh)->ei_block =
1096 EXT_FIRST_EXTENT(neh)->ee_block;
1098 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1099 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1100 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1101 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1103 neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
1104 ext4_mark_inode_dirty(handle, inode);
1112 * ext4_ext_create_new_leaf:
1113 * finds empty index and adds new leaf.
1114 * if no free index is found, then it requests in-depth growing.
1116 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1118 struct ext4_ext_path *path,
1119 struct ext4_extent *newext)
1121 struct ext4_ext_path *curp;
1122 int depth, i, err = 0;
1125 i = depth = ext_depth(inode);
1127 /* walk up to the tree and look for free index entry */
1128 curp = path + depth;
1129 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1134 /* we use already allocated block for index block,
1135 * so subsequent data blocks should be contiguous */
1136 if (EXT_HAS_FREE_INDEX(curp)) {
1137 /* if we found index with free entry, then use that
1138 * entry: create all needed subtree and add new leaf */
1139 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1144 ext4_ext_drop_refs(path);
1145 path = ext4_ext_find_extent(inode,
1146 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1149 err = PTR_ERR(path);
1151 /* tree is full, time to grow in depth */
1152 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
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);
1167 * only first (depth 0 -> 1) produces free space;
1168 * in all other cases we have to split the grown tree
1170 depth = ext_depth(inode);
1171 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1172 /* now we need to split */
1182 * search the closest allocated block to the left for *logical
1183 * and returns it at @logical + it's physical address at @phys
1184 * if *logical is the smallest allocated block, the function
1185 * returns 0 at @phys
1186 * return value contains 0 (success) or error code
1188 static int ext4_ext_search_left(struct inode *inode,
1189 struct ext4_ext_path *path,
1190 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1192 struct ext4_extent_idx *ix;
1193 struct ext4_extent *ex;
1196 if (unlikely(path == NULL)) {
1197 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1200 depth = path->p_depth;
1203 if (depth == 0 && path->p_ext == NULL)
1206 /* usually extent in the path covers blocks smaller
1207 * then *logical, but it can be that extent is the
1208 * first one in the file */
1210 ex = path[depth].p_ext;
1211 ee_len = ext4_ext_get_actual_len(ex);
1212 if (*logical < le32_to_cpu(ex->ee_block)) {
1213 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1214 EXT4_ERROR_INODE(inode,
1215 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1216 *logical, le32_to_cpu(ex->ee_block));
1219 while (--depth >= 0) {
1220 ix = path[depth].p_idx;
1221 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1222 EXT4_ERROR_INODE(inode,
1223 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1224 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1225 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1226 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1234 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1235 EXT4_ERROR_INODE(inode,
1236 "logical %d < ee_block %d + ee_len %d!",
1237 *logical, le32_to_cpu(ex->ee_block), ee_len);
1241 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1242 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1247 * search the closest allocated block to the right for *logical
1248 * and returns it at @logical + it's physical address at @phys
1249 * if *logical is the largest allocated block, the function
1250 * returns 0 at @phys
1251 * return value contains 0 (success) or error code
1253 static int ext4_ext_search_right(struct inode *inode,
1254 struct ext4_ext_path *path,
1255 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1256 struct ext4_extent **ret_ex)
1258 struct buffer_head *bh = NULL;
1259 struct ext4_extent_header *eh;
1260 struct ext4_extent_idx *ix;
1261 struct ext4_extent *ex;
1263 int depth; /* Note, NOT eh_depth; depth from top of tree */
1266 if (unlikely(path == NULL)) {
1267 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1270 depth = path->p_depth;
1273 if (depth == 0 && path->p_ext == NULL)
1276 /* usually extent in the path covers blocks smaller
1277 * then *logical, but it can be that extent is the
1278 * first one in the file */
1280 ex = path[depth].p_ext;
1281 ee_len = ext4_ext_get_actual_len(ex);
1282 if (*logical < le32_to_cpu(ex->ee_block)) {
1283 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1284 EXT4_ERROR_INODE(inode,
1285 "first_extent(path[%d].p_hdr) != ex",
1289 while (--depth >= 0) {
1290 ix = path[depth].p_idx;
1291 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1292 EXT4_ERROR_INODE(inode,
1293 "ix != EXT_FIRST_INDEX *logical %d!",
1301 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1302 EXT4_ERROR_INODE(inode,
1303 "logical %d < ee_block %d + ee_len %d!",
1304 *logical, le32_to_cpu(ex->ee_block), ee_len);
1308 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1309 /* next allocated block in this leaf */
1314 /* go up and search for index to the right */
1315 while (--depth >= 0) {
1316 ix = path[depth].p_idx;
1317 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1321 /* we've gone up to the root and found no index to the right */
1325 /* we've found index to the right, let's
1326 * follow it and find the closest allocated
1327 * block to the right */
1329 block = ext4_idx_pblock(ix);
1330 while (++depth < path->p_depth) {
1331 bh = sb_bread(inode->i_sb, block);
1334 eh = ext_block_hdr(bh);
1335 /* subtract from p_depth to get proper eh_depth */
1336 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1340 ix = EXT_FIRST_INDEX(eh);
1341 block = ext4_idx_pblock(ix);
1345 bh = sb_bread(inode->i_sb, block);
1348 eh = ext_block_hdr(bh);
1349 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1353 ex = EXT_FIRST_EXTENT(eh);
1355 *logical = le32_to_cpu(ex->ee_block);
1356 *phys = ext4_ext_pblock(ex);
1364 * ext4_ext_next_allocated_block:
1365 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1366 * NOTE: it considers block number from index entry as
1367 * allocated block. Thus, index entries have to be consistent
1371 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1375 BUG_ON(path == NULL);
1376 depth = path->p_depth;
1378 if (depth == 0 && path->p_ext == NULL)
1379 return EXT_MAX_BLOCKS;
1381 while (depth >= 0) {
1382 if (depth == path->p_depth) {
1384 if (path[depth].p_ext &&
1385 path[depth].p_ext !=
1386 EXT_LAST_EXTENT(path[depth].p_hdr))
1387 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1390 if (path[depth].p_idx !=
1391 EXT_LAST_INDEX(path[depth].p_hdr))
1392 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1397 return EXT_MAX_BLOCKS;
1401 * ext4_ext_next_leaf_block:
1402 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1404 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1408 BUG_ON(path == NULL);
1409 depth = path->p_depth;
1411 /* zero-tree has no leaf blocks at all */
1413 return EXT_MAX_BLOCKS;
1415 /* go to index block */
1418 while (depth >= 0) {
1419 if (path[depth].p_idx !=
1420 EXT_LAST_INDEX(path[depth].p_hdr))
1421 return (ext4_lblk_t)
1422 le32_to_cpu(path[depth].p_idx[1].ei_block);
1426 return EXT_MAX_BLOCKS;
1430 * ext4_ext_correct_indexes:
1431 * if leaf gets modified and modified extent is first in the leaf,
1432 * then we have to correct all indexes above.
1433 * TODO: do we need to correct tree in all cases?
1435 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1436 struct ext4_ext_path *path)
1438 struct ext4_extent_header *eh;
1439 int depth = ext_depth(inode);
1440 struct ext4_extent *ex;
1444 eh = path[depth].p_hdr;
1445 ex = path[depth].p_ext;
1447 if (unlikely(ex == NULL || eh == NULL)) {
1448 EXT4_ERROR_INODE(inode,
1449 "ex %p == NULL or eh %p == NULL", ex, eh);
1454 /* there is no tree at all */
1458 if (ex != EXT_FIRST_EXTENT(eh)) {
1459 /* we correct tree if first leaf got modified only */
1464 * TODO: we need correction if border is smaller than current one
1467 border = path[depth].p_ext->ee_block;
1468 err = ext4_ext_get_access(handle, inode, path + k);
1471 path[k].p_idx->ei_block = border;
1472 err = ext4_ext_dirty(handle, inode, path + k);
1477 /* change all left-side indexes */
1478 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1480 err = ext4_ext_get_access(handle, inode, path + k);
1483 path[k].p_idx->ei_block = border;
1484 err = ext4_ext_dirty(handle, inode, path + k);
1493 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1494 struct ext4_extent *ex2)
1496 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1499 * Make sure that either both extents are uninitialized, or
1502 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1505 if (ext4_ext_is_uninitialized(ex1))
1506 max_len = EXT_UNINIT_MAX_LEN;
1508 max_len = EXT_INIT_MAX_LEN;
1510 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1511 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1513 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1514 le32_to_cpu(ex2->ee_block))
1518 * To allow future support for preallocated extents to be added
1519 * as an RO_COMPAT feature, refuse to merge to extents if
1520 * this can result in the top bit of ee_len being set.
1522 if (ext1_ee_len + ext2_ee_len > max_len)
1524 #ifdef AGGRESSIVE_TEST
1525 if (ext1_ee_len >= 4)
1529 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1535 * This function tries to merge the "ex" extent to the next extent in the tree.
1536 * It always tries to merge towards right. If you want to merge towards
1537 * left, pass "ex - 1" as argument instead of "ex".
1538 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1539 * 1 if they got merged.
1541 static int ext4_ext_try_to_merge_right(struct inode *inode,
1542 struct ext4_ext_path *path,
1543 struct ext4_extent *ex)
1545 struct ext4_extent_header *eh;
1546 unsigned int depth, len;
1548 int uninitialized = 0;
1550 depth = ext_depth(inode);
1551 BUG_ON(path[depth].p_hdr == NULL);
1552 eh = path[depth].p_hdr;
1554 while (ex < EXT_LAST_EXTENT(eh)) {
1555 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1557 /* merge with next extent! */
1558 if (ext4_ext_is_uninitialized(ex))
1560 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1561 + ext4_ext_get_actual_len(ex + 1));
1563 ext4_ext_mark_uninitialized(ex);
1565 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1566 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1567 * sizeof(struct ext4_extent);
1568 memmove(ex + 1, ex + 2, len);
1570 le16_add_cpu(&eh->eh_entries, -1);
1572 WARN_ON(eh->eh_entries == 0);
1573 if (!eh->eh_entries)
1574 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1581 * This function tries to merge the @ex extent to neighbours in the tree.
1582 * return 1 if merge left else 0.
1584 static int ext4_ext_try_to_merge(struct inode *inode,
1585 struct ext4_ext_path *path,
1586 struct ext4_extent *ex) {
1587 struct ext4_extent_header *eh;
1592 depth = ext_depth(inode);
1593 BUG_ON(path[depth].p_hdr == NULL);
1594 eh = path[depth].p_hdr;
1596 if (ex > EXT_FIRST_EXTENT(eh))
1597 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1600 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1606 * check if a portion of the "newext" extent overlaps with an
1609 * If there is an overlap discovered, it updates the length of the newext
1610 * such that there will be no overlap, and then returns 1.
1611 * If there is no overlap found, it returns 0.
1613 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1614 struct inode *inode,
1615 struct ext4_extent *newext,
1616 struct ext4_ext_path *path)
1619 unsigned int depth, len1;
1620 unsigned int ret = 0;
1622 b1 = le32_to_cpu(newext->ee_block);
1623 len1 = ext4_ext_get_actual_len(newext);
1624 depth = ext_depth(inode);
1625 if (!path[depth].p_ext)
1627 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1628 b2 &= ~(sbi->s_cluster_ratio - 1);
1631 * get the next allocated block if the extent in the path
1632 * is before the requested block(s)
1635 b2 = ext4_ext_next_allocated_block(path);
1636 if (b2 == EXT_MAX_BLOCKS)
1638 b2 &= ~(sbi->s_cluster_ratio - 1);
1641 /* check for wrap through zero on extent logical start block*/
1642 if (b1 + len1 < b1) {
1643 len1 = EXT_MAX_BLOCKS - b1;
1644 newext->ee_len = cpu_to_le16(len1);
1648 /* check for overlap */
1649 if (b1 + len1 > b2) {
1650 newext->ee_len = cpu_to_le16(b2 - b1);
1658 * ext4_ext_insert_extent:
1659 * tries to merge requsted extent into the existing extent or
1660 * inserts requested extent as new one into the tree,
1661 * creating new leaf in the no-space case.
1663 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1664 struct ext4_ext_path *path,
1665 struct ext4_extent *newext, int flag)
1667 struct ext4_extent_header *eh;
1668 struct ext4_extent *ex, *fex;
1669 struct ext4_extent *nearex; /* nearest extent */
1670 struct ext4_ext_path *npath = NULL;
1671 int depth, len, err;
1673 unsigned uninitialized = 0;
1676 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1677 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1680 depth = ext_depth(inode);
1681 ex = path[depth].p_ext;
1682 if (unlikely(path[depth].p_hdr == NULL)) {
1683 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1687 /* try to insert block into found extent and return */
1688 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1689 && ext4_can_extents_be_merged(inode, ex, newext)) {
1690 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1691 ext4_ext_is_uninitialized(newext),
1692 ext4_ext_get_actual_len(newext),
1693 le32_to_cpu(ex->ee_block),
1694 ext4_ext_is_uninitialized(ex),
1695 ext4_ext_get_actual_len(ex),
1696 ext4_ext_pblock(ex));
1697 err = ext4_ext_get_access(handle, inode, path + depth);
1702 * ext4_can_extents_be_merged should have checked that either
1703 * both extents are uninitialized, or both aren't. Thus we
1704 * need to check only one of them here.
1706 if (ext4_ext_is_uninitialized(ex))
1708 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1709 + ext4_ext_get_actual_len(newext));
1711 ext4_ext_mark_uninitialized(ex);
1712 eh = path[depth].p_hdr;
1717 depth = ext_depth(inode);
1718 eh = path[depth].p_hdr;
1719 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1722 /* probably next leaf has space for us? */
1723 fex = EXT_LAST_EXTENT(eh);
1724 next = EXT_MAX_BLOCKS;
1725 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1726 next = ext4_ext_next_leaf_block(path);
1727 if (next != EXT_MAX_BLOCKS) {
1728 ext_debug("next leaf block - %u\n", next);
1729 BUG_ON(npath != NULL);
1730 npath = ext4_ext_find_extent(inode, next, NULL);
1732 return PTR_ERR(npath);
1733 BUG_ON(npath->p_depth != path->p_depth);
1734 eh = npath[depth].p_hdr;
1735 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1736 ext_debug("next leaf isn't full(%d)\n",
1737 le16_to_cpu(eh->eh_entries));
1741 ext_debug("next leaf has no free space(%d,%d)\n",
1742 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1746 * There is no free space in the found leaf.
1747 * We're gonna add a new leaf in the tree.
1749 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1750 flags = EXT4_MB_USE_ROOT_BLOCKS;
1751 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1754 depth = ext_depth(inode);
1755 eh = path[depth].p_hdr;
1758 nearex = path[depth].p_ext;
1760 err = ext4_ext_get_access(handle, inode, path + depth);
1765 /* there is no extent in this leaf, create first one */
1766 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1767 le32_to_cpu(newext->ee_block),
1768 ext4_ext_pblock(newext),
1769 ext4_ext_is_uninitialized(newext),
1770 ext4_ext_get_actual_len(newext));
1771 nearex = EXT_FIRST_EXTENT(eh);
1773 if (le32_to_cpu(newext->ee_block)
1774 > le32_to_cpu(nearex->ee_block)) {
1776 ext_debug("insert %u:%llu:[%d]%d before: "
1778 le32_to_cpu(newext->ee_block),
1779 ext4_ext_pblock(newext),
1780 ext4_ext_is_uninitialized(newext),
1781 ext4_ext_get_actual_len(newext),
1786 BUG_ON(newext->ee_block == nearex->ee_block);
1787 ext_debug("insert %u:%llu:[%d]%d after: "
1789 le32_to_cpu(newext->ee_block),
1790 ext4_ext_pblock(newext),
1791 ext4_ext_is_uninitialized(newext),
1792 ext4_ext_get_actual_len(newext),
1795 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1797 ext_debug("insert %u:%llu:[%d]%d: "
1798 "move %d extents from 0x%p to 0x%p\n",
1799 le32_to_cpu(newext->ee_block),
1800 ext4_ext_pblock(newext),
1801 ext4_ext_is_uninitialized(newext),
1802 ext4_ext_get_actual_len(newext),
1803 len, nearex, nearex + 1);
1804 memmove(nearex + 1, nearex,
1805 len * sizeof(struct ext4_extent));
1809 le16_add_cpu(&eh->eh_entries, 1);
1810 path[depth].p_ext = nearex;
1811 nearex->ee_block = newext->ee_block;
1812 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1813 nearex->ee_len = newext->ee_len;
1816 /* try to merge extents to the right */
1817 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1818 ext4_ext_try_to_merge(inode, path, nearex);
1820 /* try to merge extents to the left */
1822 /* time to correct all indexes above */
1823 err = ext4_ext_correct_indexes(handle, inode, path);
1827 err = ext4_ext_dirty(handle, inode, path + depth);
1831 ext4_ext_drop_refs(npath);
1834 ext4_ext_invalidate_cache(inode);
1838 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1839 ext4_lblk_t num, ext_prepare_callback func,
1842 struct ext4_ext_path *path = NULL;
1843 struct ext4_ext_cache cbex;
1844 struct ext4_extent *ex;
1845 ext4_lblk_t next, start = 0, end = 0;
1846 ext4_lblk_t last = block + num;
1847 int depth, exists, err = 0;
1849 BUG_ON(func == NULL);
1850 BUG_ON(inode == NULL);
1852 while (block < last && block != EXT_MAX_BLOCKS) {
1854 /* find extent for this block */
1855 down_read(&EXT4_I(inode)->i_data_sem);
1856 path = ext4_ext_find_extent(inode, block, path);
1857 up_read(&EXT4_I(inode)->i_data_sem);
1859 err = PTR_ERR(path);
1864 depth = ext_depth(inode);
1865 if (unlikely(path[depth].p_hdr == NULL)) {
1866 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1870 ex = path[depth].p_ext;
1871 next = ext4_ext_next_allocated_block(path);
1875 /* there is no extent yet, so try to allocate
1876 * all requested space */
1879 } else if (le32_to_cpu(ex->ee_block) > block) {
1880 /* need to allocate space before found extent */
1882 end = le32_to_cpu(ex->ee_block);
1883 if (block + num < end)
1885 } else if (block >= le32_to_cpu(ex->ee_block)
1886 + ext4_ext_get_actual_len(ex)) {
1887 /* need to allocate space after found extent */
1892 } else if (block >= le32_to_cpu(ex->ee_block)) {
1894 * some part of requested space is covered
1898 end = le32_to_cpu(ex->ee_block)
1899 + ext4_ext_get_actual_len(ex);
1900 if (block + num < end)
1906 BUG_ON(end <= start);
1909 cbex.ec_block = start;
1910 cbex.ec_len = end - start;
1913 cbex.ec_block = le32_to_cpu(ex->ee_block);
1914 cbex.ec_len = ext4_ext_get_actual_len(ex);
1915 cbex.ec_start = ext4_ext_pblock(ex);
1918 if (unlikely(cbex.ec_len == 0)) {
1919 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1923 err = func(inode, next, &cbex, ex, cbdata);
1924 ext4_ext_drop_refs(path);
1929 if (err == EXT_REPEAT)
1931 else if (err == EXT_BREAK) {
1936 if (ext_depth(inode) != depth) {
1937 /* depth was changed. we have to realloc path */
1942 block = cbex.ec_block + cbex.ec_len;
1946 ext4_ext_drop_refs(path);
1954 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1955 __u32 len, ext4_fsblk_t start)
1957 struct ext4_ext_cache *cex;
1959 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1960 trace_ext4_ext_put_in_cache(inode, block, len, start);
1961 cex = &EXT4_I(inode)->i_cached_extent;
1962 cex->ec_block = block;
1964 cex->ec_start = start;
1965 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1969 * ext4_ext_put_gap_in_cache:
1970 * calculate boundaries of the gap that the requested block fits into
1971 * and cache this gap
1974 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1977 int depth = ext_depth(inode);
1980 struct ext4_extent *ex;
1982 ex = path[depth].p_ext;
1984 /* there is no extent yet, so gap is [0;-] */
1986 len = EXT_MAX_BLOCKS;
1987 ext_debug("cache gap(whole file):");
1988 } else if (block < le32_to_cpu(ex->ee_block)) {
1990 len = le32_to_cpu(ex->ee_block) - block;
1991 ext_debug("cache gap(before): %u [%u:%u]",
1993 le32_to_cpu(ex->ee_block),
1994 ext4_ext_get_actual_len(ex));
1995 } else if (block >= le32_to_cpu(ex->ee_block)
1996 + ext4_ext_get_actual_len(ex)) {
1998 lblock = le32_to_cpu(ex->ee_block)
1999 + ext4_ext_get_actual_len(ex);
2001 next = ext4_ext_next_allocated_block(path);
2002 ext_debug("cache gap(after): [%u:%u] %u",
2003 le32_to_cpu(ex->ee_block),
2004 ext4_ext_get_actual_len(ex),
2006 BUG_ON(next == lblock);
2007 len = next - lblock;
2013 ext_debug(" -> %u:%lu\n", lblock, len);
2014 ext4_ext_put_in_cache(inode, lblock, len, 0);
2018 * ext4_ext_check_cache()
2019 * Checks to see if the given block is in the cache.
2020 * If it is, the cached extent is stored in the given
2021 * cache extent pointer. If the cached extent is a hole,
2022 * this routine should be used instead of
2023 * ext4_ext_in_cache if the calling function needs to
2024 * know the size of the hole.
2026 * @inode: The files inode
2027 * @block: The block to look for in the cache
2028 * @ex: Pointer where the cached extent will be stored
2029 * if it contains block
2031 * Return 0 if cache is invalid; 1 if the cache is valid
2033 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2034 struct ext4_ext_cache *ex){
2035 struct ext4_ext_cache *cex;
2036 struct ext4_sb_info *sbi;
2040 * We borrow i_block_reservation_lock to protect i_cached_extent
2042 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2043 cex = &EXT4_I(inode)->i_cached_extent;
2044 sbi = EXT4_SB(inode->i_sb);
2046 /* has cache valid data? */
2047 if (cex->ec_len == 0)
2050 if (in_range(block, cex->ec_block, cex->ec_len)) {
2051 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2052 ext_debug("%u cached by %u:%u:%llu\n",
2054 cex->ec_block, cex->ec_len, cex->ec_start);
2058 trace_ext4_ext_in_cache(inode, block, ret);
2059 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2064 * ext4_ext_in_cache()
2065 * Checks to see if the given block is in the cache.
2066 * If it is, the cached extent is stored in the given
2069 * @inode: The files inode
2070 * @block: The block to look for in the cache
2071 * @ex: Pointer where the cached extent will be stored
2072 * if it contains block
2074 * Return 0 if cache is invalid; 1 if the cache is valid
2077 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2078 struct ext4_extent *ex)
2080 struct ext4_ext_cache cex;
2083 if (ext4_ext_check_cache(inode, block, &cex)) {
2084 ex->ee_block = cpu_to_le32(cex.ec_block);
2085 ext4_ext_store_pblock(ex, cex.ec_start);
2086 ex->ee_len = cpu_to_le16(cex.ec_len);
2096 * removes index from the index block.
2098 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2099 struct ext4_ext_path *path, int depth)
2104 /* free index block */
2106 path = path + depth;
2107 leaf = ext4_idx_pblock(path->p_idx);
2108 if (unlikely(path->p_hdr->eh_entries == 0)) {
2109 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2112 err = ext4_ext_get_access(handle, inode, path);
2116 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2117 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2118 len *= sizeof(struct ext4_extent_idx);
2119 memmove(path->p_idx, path->p_idx + 1, len);
2122 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2123 err = ext4_ext_dirty(handle, inode, path);
2126 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2127 trace_ext4_ext_rm_idx(inode, leaf);
2129 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2130 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2132 while (--depth >= 0) {
2133 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2136 err = ext4_ext_get_access(handle, inode, path);
2139 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2140 err = ext4_ext_dirty(handle, inode, path);
2148 * ext4_ext_calc_credits_for_single_extent:
2149 * This routine returns max. credits that needed to insert an extent
2150 * to the extent tree.
2151 * When pass the actual path, the caller should calculate credits
2154 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2155 struct ext4_ext_path *path)
2158 int depth = ext_depth(inode);
2161 /* probably there is space in leaf? */
2162 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2163 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2166 * There are some space in the leaf tree, no
2167 * need to account for leaf block credit
2169 * bitmaps and block group descriptor blocks
2170 * and other metadata blocks still need to be
2173 /* 1 bitmap, 1 block group descriptor */
2174 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2179 return ext4_chunk_trans_blocks(inode, nrblocks);
2183 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2185 * if nrblocks are fit in a single extent (chunk flag is 1), then
2186 * in the worse case, each tree level index/leaf need to be changed
2187 * if the tree split due to insert a new extent, then the old tree
2188 * index/leaf need to be updated too
2190 * If the nrblocks are discontiguous, they could cause
2191 * the whole tree split more than once, but this is really rare.
2193 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2196 int depth = ext_depth(inode);
2206 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2207 struct ext4_extent *ex,
2208 ext4_fsblk_t *partial_cluster,
2209 ext4_lblk_t from, ext4_lblk_t to)
2211 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2212 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2214 int flags = EXT4_FREE_BLOCKS_FORGET;
2216 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2217 flags |= EXT4_FREE_BLOCKS_METADATA;
2219 * For bigalloc file systems, we never free a partial cluster
2220 * at the beginning of the extent. Instead, we make a note
2221 * that we tried freeing the cluster, and check to see if we
2222 * need to free it on a subsequent call to ext4_remove_blocks,
2223 * or at the end of the ext4_truncate() operation.
2225 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2227 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2229 * If we have a partial cluster, and it's different from the
2230 * cluster of the last block, we need to explicitly free the
2231 * partial cluster here.
2233 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2234 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2235 ext4_free_blocks(handle, inode, NULL,
2236 EXT4_C2B(sbi, *partial_cluster),
2237 sbi->s_cluster_ratio, flags);
2238 *partial_cluster = 0;
2241 #ifdef EXTENTS_STATS
2243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2244 spin_lock(&sbi->s_ext_stats_lock);
2245 sbi->s_ext_blocks += ee_len;
2246 sbi->s_ext_extents++;
2247 if (ee_len < sbi->s_ext_min)
2248 sbi->s_ext_min = ee_len;
2249 if (ee_len > sbi->s_ext_max)
2250 sbi->s_ext_max = ee_len;
2251 if (ext_depth(inode) > sbi->s_depth_max)
2252 sbi->s_depth_max = ext_depth(inode);
2253 spin_unlock(&sbi->s_ext_stats_lock);
2256 if (from >= le32_to_cpu(ex->ee_block)
2257 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2261 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2262 pblk = ext4_ext_pblock(ex) + ee_len - num;
2263 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2264 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2266 * If the block range to be freed didn't start at the
2267 * beginning of a cluster, and we removed the entire
2268 * extent, save the partial cluster here, since we
2269 * might need to delete if we determine that the
2270 * truncate operation has removed all of the blocks in
2273 if (pblk & (sbi->s_cluster_ratio - 1) &&
2275 *partial_cluster = EXT4_B2C(sbi, pblk);
2277 *partial_cluster = 0;
2278 } else if (from == le32_to_cpu(ex->ee_block)
2279 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2285 start = ext4_ext_pblock(ex);
2287 ext_debug("free first %u blocks starting %llu\n", num, start);
2288 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2291 printk(KERN_INFO "strange request: removal(2) "
2292 "%u-%u from %u:%u\n",
2293 from, to, le32_to_cpu(ex->ee_block), ee_len);
2300 * ext4_ext_rm_leaf() Removes the extents associated with the
2301 * blocks appearing between "start" and "end", and splits the extents
2302 * if "start" and "end" appear in the same extent
2304 * @handle: The journal handle
2305 * @inode: The files inode
2306 * @path: The path to the leaf
2307 * @start: The first block to remove
2308 * @end: The last block to remove
2311 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2312 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2313 ext4_lblk_t start, ext4_lblk_t end)
2315 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2316 int err = 0, correct_index = 0;
2317 int depth = ext_depth(inode), credits;
2318 struct ext4_extent_header *eh;
2321 ext4_lblk_t ex_ee_block;
2322 unsigned short ex_ee_len;
2323 unsigned uninitialized = 0;
2324 struct ext4_extent *ex;
2326 /* the header must be checked already in ext4_ext_remove_space() */
2327 ext_debug("truncate since %u in leaf\n", start);
2328 if (!path[depth].p_hdr)
2329 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2330 eh = path[depth].p_hdr;
2331 if (unlikely(path[depth].p_hdr == NULL)) {
2332 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2335 /* find where to start removing */
2336 ex = EXT_LAST_EXTENT(eh);
2338 ex_ee_block = le32_to_cpu(ex->ee_block);
2339 ex_ee_len = ext4_ext_get_actual_len(ex);
2341 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2343 while (ex >= EXT_FIRST_EXTENT(eh) &&
2344 ex_ee_block + ex_ee_len > start) {
2346 if (ext4_ext_is_uninitialized(ex))
2351 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2352 uninitialized, ex_ee_len);
2353 path[depth].p_ext = ex;
2355 a = ex_ee_block > start ? ex_ee_block : start;
2356 b = ex_ee_block+ex_ee_len - 1 < end ?
2357 ex_ee_block+ex_ee_len - 1 : end;
2359 ext_debug(" border %u:%u\n", a, b);
2361 /* If this extent is beyond the end of the hole, skip it */
2362 if (end <= ex_ee_block) {
2364 ex_ee_block = le32_to_cpu(ex->ee_block);
2365 ex_ee_len = ext4_ext_get_actual_len(ex);
2367 } else if (b != ex_ee_block + ex_ee_len - 1) {
2368 EXT4_ERROR_INODE(inode," bad truncate %u:%u\n",
2372 } else if (a != ex_ee_block) {
2373 /* remove tail of the extent */
2374 num = a - ex_ee_block;
2376 /* remove whole extent: excellent! */
2380 * 3 for leaf, sb, and inode plus 2 (bmap and group
2381 * descriptor) for each block group; assume two block
2382 * groups plus ex_ee_len/blocks_per_block_group for
2385 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2386 if (ex == EXT_FIRST_EXTENT(eh)) {
2388 credits += (ext_depth(inode)) + 1;
2390 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2392 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2396 err = ext4_ext_get_access(handle, inode, path + depth);
2400 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2406 /* this extent is removed; mark slot entirely unused */
2407 ext4_ext_store_pblock(ex, 0);
2409 ex->ee_len = cpu_to_le16(num);
2411 * Do not mark uninitialized if all the blocks in the
2412 * extent have been removed.
2414 if (uninitialized && num)
2415 ext4_ext_mark_uninitialized(ex);
2417 * If the extent was completely released,
2418 * we need to remove it from the leaf
2421 if (end != EXT_MAX_BLOCKS - 1) {
2423 * For hole punching, we need to scoot all the
2424 * extents up when an extent is removed so that
2425 * we dont have blank extents in the middle
2427 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2428 sizeof(struct ext4_extent));
2430 /* Now get rid of the one at the end */
2431 memset(EXT_LAST_EXTENT(eh), 0,
2432 sizeof(struct ext4_extent));
2434 le16_add_cpu(&eh->eh_entries, -1);
2436 *partial_cluster = 0;
2438 err = ext4_ext_dirty(handle, inode, path + depth);
2442 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2443 ext4_ext_pblock(ex));
2445 ex_ee_block = le32_to_cpu(ex->ee_block);
2446 ex_ee_len = ext4_ext_get_actual_len(ex);
2449 if (correct_index && eh->eh_entries)
2450 err = ext4_ext_correct_indexes(handle, inode, path);
2453 * If there is still a entry in the leaf node, check to see if
2454 * it references the partial cluster. This is the only place
2455 * where it could; if it doesn't, we can free the cluster.
2457 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2458 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2459 *partial_cluster)) {
2460 int flags = EXT4_FREE_BLOCKS_FORGET;
2462 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2463 flags |= EXT4_FREE_BLOCKS_METADATA;
2465 ext4_free_blocks(handle, inode, NULL,
2466 EXT4_C2B(sbi, *partial_cluster),
2467 sbi->s_cluster_ratio, flags);
2468 *partial_cluster = 0;
2471 /* if this leaf is free, then we should
2472 * remove it from index block above */
2473 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2474 err = ext4_ext_rm_idx(handle, inode, path, depth);
2481 * ext4_ext_more_to_rm:
2482 * returns 1 if current index has to be freed (even partial)
2485 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2487 BUG_ON(path->p_idx == NULL);
2489 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2493 * if truncate on deeper level happened, it wasn't partial,
2494 * so we have to consider current index for truncation
2496 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2501 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2503 struct super_block *sb = inode->i_sb;
2504 int depth = ext_depth(inode);
2505 struct ext4_ext_path *path;
2506 ext4_fsblk_t partial_cluster = 0;
2510 ext_debug("truncate since %u\n", start);
2512 /* probably first extent we're gonna free will be last in block */
2513 handle = ext4_journal_start(inode, depth + 1);
2515 return PTR_ERR(handle);
2518 ext4_ext_invalidate_cache(inode);
2520 trace_ext4_ext_remove_space(inode, start, depth);
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 &partial_cluster, start,
2545 EXT_MAX_BLOCKS - 1);
2546 /* root level has p_bh == NULL, brelse() eats this */
2547 brelse(path[i].p_bh);
2548 path[i].p_bh = NULL;
2553 /* this is index block */
2554 if (!path[i].p_hdr) {
2555 ext_debug("initialize header\n");
2556 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2559 if (!path[i].p_idx) {
2560 /* this level hasn't been touched yet */
2561 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2562 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2563 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2565 le16_to_cpu(path[i].p_hdr->eh_entries));
2567 /* we were already here, see at next index */
2571 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2572 i, EXT_FIRST_INDEX(path[i].p_hdr),
2574 if (ext4_ext_more_to_rm(path + i)) {
2575 struct buffer_head *bh;
2576 /* go to the next level */
2577 ext_debug("move to level %d (block %llu)\n",
2578 i + 1, ext4_idx_pblock(path[i].p_idx));
2579 memset(path + i + 1, 0, sizeof(*path));
2580 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2582 /* should we reset i_size? */
2586 if (WARN_ON(i + 1 > depth)) {
2590 if (ext4_ext_check(inode, ext_block_hdr(bh),
2595 path[i + 1].p_bh = bh;
2597 /* save actual number of indexes since this
2598 * number is changed at the next iteration */
2599 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2602 /* we finished processing this index, go up */
2603 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2604 /* index is empty, remove it;
2605 * handle must be already prepared by the
2606 * truncatei_leaf() */
2607 err = ext4_ext_rm_idx(handle, inode, path, i);
2609 /* root level has p_bh == NULL, brelse() eats this */
2610 brelse(path[i].p_bh);
2611 path[i].p_bh = NULL;
2613 ext_debug("return to level %d\n", i);
2617 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2618 path->p_hdr->eh_entries);
2620 /* If we still have something in the partial cluster and we have removed
2621 * even the first extent, then we should free the blocks in the partial
2622 * cluster as well. */
2623 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2624 int flags = EXT4_FREE_BLOCKS_FORGET;
2626 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2627 flags |= EXT4_FREE_BLOCKS_METADATA;
2629 ext4_free_blocks(handle, inode, NULL,
2630 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2631 EXT4_SB(sb)->s_cluster_ratio, flags);
2632 partial_cluster = 0;
2635 /* TODO: flexible tree reduction should be here */
2636 if (path->p_hdr->eh_entries == 0) {
2638 * truncate to zero freed all the tree,
2639 * so we need to correct eh_depth
2641 err = ext4_ext_get_access(handle, inode, path);
2643 ext_inode_hdr(inode)->eh_depth = 0;
2644 ext_inode_hdr(inode)->eh_max =
2645 cpu_to_le16(ext4_ext_space_root(inode, 0));
2646 err = ext4_ext_dirty(handle, inode, path);
2650 ext4_ext_drop_refs(path);
2654 ext4_journal_stop(handle);
2660 * called at mount time
2662 void ext4_ext_init(struct super_block *sb)
2665 * possible initialization would be here
2668 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2669 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2670 printk(KERN_INFO "EXT4-fs: file extents enabled");
2671 #ifdef AGGRESSIVE_TEST
2672 printk(", aggressive tests");
2674 #ifdef CHECK_BINSEARCH
2675 printk(", check binsearch");
2677 #ifdef EXTENTS_STATS
2682 #ifdef EXTENTS_STATS
2683 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2684 EXT4_SB(sb)->s_ext_min = 1 << 30;
2685 EXT4_SB(sb)->s_ext_max = 0;
2691 * called at umount time
2693 void ext4_ext_release(struct super_block *sb)
2695 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2698 #ifdef EXTENTS_STATS
2699 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2700 struct ext4_sb_info *sbi = EXT4_SB(sb);
2701 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2702 sbi->s_ext_blocks, sbi->s_ext_extents,
2703 sbi->s_ext_blocks / sbi->s_ext_extents);
2704 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2705 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2710 /* FIXME!! we need to try to merge to left or right after zero-out */
2711 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2713 ext4_fsblk_t ee_pblock;
2714 unsigned int ee_len;
2717 ee_len = ext4_ext_get_actual_len(ex);
2718 ee_pblock = ext4_ext_pblock(ex);
2720 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2728 * used by extent splitting.
2730 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2732 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2733 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2735 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
2736 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
2739 * ext4_split_extent_at() splits an extent at given block.
2741 * @handle: the journal handle
2742 * @inode: the file inode
2743 * @path: the path to the extent
2744 * @split: the logical block where the extent is splitted.
2745 * @split_flags: indicates if the extent could be zeroout if split fails, and
2746 * the states(init or uninit) of new extents.
2747 * @flags: flags used to insert new extent to extent tree.
2750 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2751 * of which are deterimined by split_flag.
2753 * There are two cases:
2754 * a> the extent are splitted into two extent.
2755 * b> split is not needed, and just mark the extent.
2757 * return 0 on success.
2759 static int ext4_split_extent_at(handle_t *handle,
2760 struct inode *inode,
2761 struct ext4_ext_path *path,
2766 ext4_fsblk_t newblock;
2767 ext4_lblk_t ee_block;
2768 struct ext4_extent *ex, newex, orig_ex;
2769 struct ext4_extent *ex2 = NULL;
2770 unsigned int ee_len, depth;
2773 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2774 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2776 ext_debug("ext4_split_extents_at: inode %lu, logical"
2777 "block %llu\n", inode->i_ino, (unsigned long long)split);
2779 ext4_ext_show_leaf(inode, path);
2781 depth = ext_depth(inode);
2782 ex = path[depth].p_ext;
2783 ee_block = le32_to_cpu(ex->ee_block);
2784 ee_len = ext4_ext_get_actual_len(ex);
2785 newblock = split - ee_block + ext4_ext_pblock(ex);
2787 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2789 err = ext4_ext_get_access(handle, inode, path + depth);
2793 if (split == ee_block) {
2795 * case b: block @split is the block that the extent begins with
2796 * then we just change the state of the extent, and splitting
2799 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2800 ext4_ext_mark_uninitialized(ex);
2802 ext4_ext_mark_initialized(ex);
2804 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2805 ext4_ext_try_to_merge(inode, path, ex);
2807 err = ext4_ext_dirty(handle, inode, path + depth);
2812 memcpy(&orig_ex, ex, sizeof(orig_ex));
2813 ex->ee_len = cpu_to_le16(split - ee_block);
2814 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2815 ext4_ext_mark_uninitialized(ex);
2818 * path may lead to new leaf, not to original leaf any more
2819 * after ext4_ext_insert_extent() returns,
2821 err = ext4_ext_dirty(handle, inode, path + depth);
2823 goto fix_extent_len;
2826 ex2->ee_block = cpu_to_le32(split);
2827 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2828 ext4_ext_store_pblock(ex2, newblock);
2829 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2830 ext4_ext_mark_uninitialized(ex2);
2832 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2833 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2834 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
2835 if (split_flag & EXT4_EXT_DATA_VALID1)
2836 err = ext4_ext_zeroout(inode, ex2);
2838 err = ext4_ext_zeroout(inode, ex);
2840 err = ext4_ext_zeroout(inode, &orig_ex);
2843 goto fix_extent_len;
2844 /* update the extent length and mark as initialized */
2845 ex->ee_len = cpu_to_le16(ee_len);
2846 ext4_ext_try_to_merge(inode, path, ex);
2847 err = ext4_ext_dirty(handle, inode, path + depth);
2850 goto fix_extent_len;
2853 ext4_ext_show_leaf(inode, path);
2857 ex->ee_len = orig_ex.ee_len;
2858 ext4_ext_dirty(handle, inode, path + depth);
2863 * ext4_split_extents() splits an extent and mark extent which is covered
2864 * by @map as split_flags indicates
2866 * It may result in splitting the extent into multiple extents (upto three)
2867 * There are three possibilities:
2868 * a> There is no split required
2869 * b> Splits in two extents: Split is happening at either end of the extent
2870 * c> Splits in three extents: Somone is splitting in middle of the extent
2873 static int ext4_split_extent(handle_t *handle,
2874 struct inode *inode,
2875 struct ext4_ext_path *path,
2876 struct ext4_map_blocks *map,
2880 ext4_lblk_t ee_block;
2881 struct ext4_extent *ex;
2882 unsigned int ee_len, depth;
2885 int split_flag1, flags1;
2887 depth = ext_depth(inode);
2888 ex = path[depth].p_ext;
2889 ee_block = le32_to_cpu(ex->ee_block);
2890 ee_len = ext4_ext_get_actual_len(ex);
2891 uninitialized = ext4_ext_is_uninitialized(ex);
2893 if (map->m_lblk + map->m_len < ee_block + ee_len) {
2894 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
2895 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2897 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2898 EXT4_EXT_MARK_UNINIT2;
2899 if (split_flag & EXT4_EXT_DATA_VALID2)
2900 split_flag1 |= EXT4_EXT_DATA_VALID1;
2901 err = ext4_split_extent_at(handle, inode, path,
2902 map->m_lblk + map->m_len, split_flag1, flags1);
2907 ext4_ext_drop_refs(path);
2908 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2910 return PTR_ERR(path);
2912 if (map->m_lblk >= ee_block) {
2913 split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT |
2914 EXT4_EXT_DATA_VALID2);
2916 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2917 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2918 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2919 err = ext4_split_extent_at(handle, inode, path,
2920 map->m_lblk, split_flag1, flags);
2925 ext4_ext_show_leaf(inode, path);
2927 return err ? err : map->m_len;
2930 #define EXT4_EXT_ZERO_LEN 7
2932 * This function is called by ext4_ext_map_blocks() if someone tries to write
2933 * to an uninitialized extent. It may result in splitting the uninitialized
2934 * extent into multiple extents (up to three - one initialized and two
2936 * There are three possibilities:
2937 * a> There is no split required: Entire extent should be initialized
2938 * b> Splits in two extents: Write is happening at either end of the extent
2939 * c> Splits in three extents: Somone is writing in middle of the extent
2942 * - The extent pointed to by 'path' is uninitialized.
2943 * - The extent pointed to by 'path' contains a superset
2944 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
2946 * Post-conditions on success:
2947 * - the returned value is the number of blocks beyond map->l_lblk
2948 * that are allocated and initialized.
2949 * It is guaranteed to be >= map->m_len.
2951 static int ext4_ext_convert_to_initialized(handle_t *handle,
2952 struct inode *inode,
2953 struct ext4_map_blocks *map,
2954 struct ext4_ext_path *path)
2956 struct ext4_extent_header *eh;
2957 struct ext4_map_blocks split_map;
2958 struct ext4_extent zero_ex;
2959 struct ext4_extent *ex;
2960 ext4_lblk_t ee_block, eof_block;
2961 unsigned int ee_len, depth;
2966 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2967 "block %llu, max_blocks %u\n", inode->i_ino,
2968 (unsigned long long)map->m_lblk, map->m_len);
2970 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2971 inode->i_sb->s_blocksize_bits;
2972 if (eof_block < map->m_lblk + map->m_len)
2973 eof_block = map->m_lblk + map->m_len;
2975 depth = ext_depth(inode);
2976 eh = path[depth].p_hdr;
2977 ex = path[depth].p_ext;
2978 ee_block = le32_to_cpu(ex->ee_block);
2979 ee_len = ext4_ext_get_actual_len(ex);
2980 allocated = ee_len - (map->m_lblk - ee_block);
2982 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
2984 /* Pre-conditions */
2985 BUG_ON(!ext4_ext_is_uninitialized(ex));
2986 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
2989 * Attempt to transfer newly initialized blocks from the currently
2990 * uninitialized extent to its left neighbor. This is much cheaper
2991 * than an insertion followed by a merge as those involve costly
2992 * memmove() calls. This is the common case in steady state for
2993 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
2996 * Limitations of the current logic:
2997 * - L1: we only deal with writes at the start of the extent.
2998 * The approach could be extended to writes at the end
2999 * of the extent but this scenario was deemed less common.
3000 * - L2: we do not deal with writes covering the whole extent.
3001 * This would require removing the extent if the transfer
3003 * - L3: we only attempt to merge with an extent stored in the
3004 * same extent tree node.
3006 if ((map->m_lblk == ee_block) && /*L1*/
3007 (map->m_len < ee_len) && /*L2*/
3008 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
3009 struct ext4_extent *prev_ex;
3010 ext4_lblk_t prev_lblk;
3011 ext4_fsblk_t prev_pblk, ee_pblk;
3012 unsigned int prev_len, write_len;
3015 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3016 prev_len = ext4_ext_get_actual_len(prev_ex);
3017 prev_pblk = ext4_ext_pblock(prev_ex);
3018 ee_pblk = ext4_ext_pblock(ex);
3019 write_len = map->m_len;
3022 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3023 * upon those conditions:
3024 * - C1: prev_ex is initialized,
3025 * - C2: prev_ex is logically abutting ex,
3026 * - C3: prev_ex is physically abutting ex,
3027 * - C4: prev_ex can receive the additional blocks without
3028 * overflowing the (initialized) length limit.
3030 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3031 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3032 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3033 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3034 err = ext4_ext_get_access(handle, inode, path + depth);
3038 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3041 /* Shift the start of ex by 'write_len' blocks */
3042 ex->ee_block = cpu_to_le32(ee_block + write_len);
3043 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3044 ex->ee_len = cpu_to_le16(ee_len - write_len);
3045 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3047 /* Extend prev_ex by 'write_len' blocks */
3048 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3050 /* Mark the block containing both extents as dirty */
3051 ext4_ext_dirty(handle, inode, path + depth);
3053 /* Update path to point to the right extent */
3054 path[depth].p_ext = prev_ex;
3056 /* Result: number of initialized blocks past m_lblk */
3057 allocated = write_len;
3062 WARN_ON(map->m_lblk < ee_block);
3064 * It is safe to convert extent to initialized via explicit
3065 * zeroout only if extent is fully insde i_size or new_size.
3067 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3069 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3070 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3071 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3072 err = ext4_ext_zeroout(inode, ex);
3076 err = ext4_ext_get_access(handle, inode, path + depth);
3079 ext4_ext_mark_initialized(ex);
3080 ext4_ext_try_to_merge(inode, path, ex);
3081 err = ext4_ext_dirty(handle, inode, path + depth);
3087 * 1. split the extent into three extents.
3088 * 2. split the extent into two extents, zeroout the first half.
3089 * 3. split the extent into two extents, zeroout the second half.
3090 * 4. split the extent into two extents with out zeroout.
3092 split_map.m_lblk = map->m_lblk;
3093 split_map.m_len = map->m_len;
3095 if (allocated > map->m_len) {
3096 if (allocated <= EXT4_EXT_ZERO_LEN &&
3097 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3100 cpu_to_le32(map->m_lblk);
3101 zero_ex.ee_len = cpu_to_le16(allocated);
3102 ext4_ext_store_pblock(&zero_ex,
3103 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3104 err = ext4_ext_zeroout(inode, &zero_ex);
3107 split_map.m_lblk = map->m_lblk;
3108 split_map.m_len = allocated;
3109 } else if ((map->m_lblk - ee_block + map->m_len <
3110 EXT4_EXT_ZERO_LEN) &&
3111 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3113 if (map->m_lblk != ee_block) {
3114 zero_ex.ee_block = ex->ee_block;
3115 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3117 ext4_ext_store_pblock(&zero_ex,
3118 ext4_ext_pblock(ex));
3119 err = ext4_ext_zeroout(inode, &zero_ex);
3124 split_map.m_lblk = ee_block;
3125 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3126 allocated = map->m_len;
3130 allocated = ext4_split_extent(handle, inode, path,
3131 &split_map, split_flag, 0);
3136 return err ? err : allocated;
3140 * This function is called by ext4_ext_map_blocks() from
3141 * ext4_get_blocks_dio_write() when DIO to write
3142 * to an uninitialized extent.
3144 * Writing to an uninitialized extent may result in splitting the uninitialized
3145 * extent into multiple /initialized uninitialized extents (up to three)
3146 * There are three possibilities:
3147 * a> There is no split required: Entire extent should be uninitialized
3148 * b> Splits in two extents: Write is happening at either end of the extent
3149 * c> Splits in three extents: Somone is writing in middle of the extent
3151 * One of more index blocks maybe needed if the extent tree grow after
3152 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3153 * complete, we need to split the uninitialized extent before DIO submit
3154 * the IO. The uninitialized extent called at this time will be split
3155 * into three uninitialized extent(at most). After IO complete, the part
3156 * being filled will be convert to initialized by the end_io callback function
3157 * via ext4_convert_unwritten_extents().
3159 * Returns the size of uninitialized extent to be written on success.
3161 static int ext4_split_unwritten_extents(handle_t *handle,
3162 struct inode *inode,
3163 struct ext4_map_blocks *map,
3164 struct ext4_ext_path *path,
3167 ext4_lblk_t eof_block;
3168 ext4_lblk_t ee_block;
3169 struct ext4_extent *ex;
3170 unsigned int ee_len;
3171 int split_flag = 0, depth;
3173 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3174 "block %llu, max_blocks %u\n", inode->i_ino,
3175 (unsigned long long)map->m_lblk, map->m_len);
3177 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3178 inode->i_sb->s_blocksize_bits;
3179 if (eof_block < map->m_lblk + map->m_len)
3180 eof_block = map->m_lblk + map->m_len;
3182 * It is safe to convert extent to initialized via explicit
3183 * zeroout only if extent is fully insde i_size or new_size.
3185 depth = ext_depth(inode);
3186 ex = path[depth].p_ext;
3187 ee_block = le32_to_cpu(ex->ee_block);
3188 ee_len = ext4_ext_get_actual_len(ex);
3190 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3191 split_flag |= EXT4_EXT_MARK_UNINIT2;
3192 if (flags & EXT4_GET_BLOCKS_CONVERT)
3193 split_flag |= EXT4_EXT_DATA_VALID2;
3194 flags |= EXT4_GET_BLOCKS_PRE_IO;
3195 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3198 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3199 struct inode *inode,
3200 struct ext4_map_blocks *map,
3201 struct ext4_ext_path *path)
3203 struct ext4_extent *ex;
3204 ext4_lblk_t ee_block;
3205 unsigned int ee_len;
3209 depth = ext_depth(inode);
3210 ex = path[depth].p_ext;
3211 ee_block = le32_to_cpu(ex->ee_block);
3212 ee_len = ext4_ext_get_actual_len(ex);
3214 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3215 "block %llu, max_blocks %u\n", inode->i_ino,
3216 (unsigned long long)ee_block, ee_len);
3218 /* If extent is larger than requested then split is required */
3219 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3220 err = ext4_split_unwritten_extents(handle, inode, map, path,
3221 EXT4_GET_BLOCKS_CONVERT);
3224 ext4_ext_drop_refs(path);
3225 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3227 err = PTR_ERR(path);
3230 depth = ext_depth(inode);
3231 ex = path[depth].p_ext;
3234 err = ext4_ext_get_access(handle, inode, path + depth);
3237 /* first mark the extent as initialized */
3238 ext4_ext_mark_initialized(ex);
3240 /* note: ext4_ext_correct_indexes() isn't needed here because
3241 * borders are not changed
3243 ext4_ext_try_to_merge(inode, path, ex);
3245 /* Mark modified extent as dirty */
3246 err = ext4_ext_dirty(handle, inode, path + depth);
3248 ext4_ext_show_leaf(inode, path);
3252 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3253 sector_t block, int count)
3256 for (i = 0; i < count; i++)
3257 unmap_underlying_metadata(bdev, block + i);
3261 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3263 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3265 struct ext4_ext_path *path,
3269 struct ext4_extent_header *eh;
3270 struct ext4_extent *last_ex;
3272 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3275 depth = ext_depth(inode);
3276 eh = path[depth].p_hdr;
3278 if (unlikely(!eh->eh_entries)) {
3279 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3280 "EOFBLOCKS_FL set");
3283 last_ex = EXT_LAST_EXTENT(eh);
3285 * We should clear the EOFBLOCKS_FL flag if we are writing the
3286 * last block in the last extent in the file. We test this by
3287 * first checking to see if the caller to
3288 * ext4_ext_get_blocks() was interested in the last block (or
3289 * a block beyond the last block) in the current extent. If
3290 * this turns out to be false, we can bail out from this
3291 * function immediately.
3293 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3294 ext4_ext_get_actual_len(last_ex))
3297 * If the caller does appear to be planning to write at or
3298 * beyond the end of the current extent, we then test to see
3299 * if the current extent is the last extent in the file, by
3300 * checking to make sure it was reached via the rightmost node
3301 * at each level of the tree.
3303 for (i = depth-1; i >= 0; i--)
3304 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3306 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3307 return ext4_mark_inode_dirty(handle, inode);
3311 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3313 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3314 * whether there are any buffers marked for delayed allocation. It returns '1'
3315 * on the first delalloc'ed buffer head found. If no buffer head in the given
3316 * range is marked for delalloc, it returns 0.
3317 * lblk_start should always be <= lblk_end.
3318 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3319 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3320 * block sooner). This is useful when blocks are truncated sequentially from
3321 * lblk_start towards lblk_end.
3323 static int ext4_find_delalloc_range(struct inode *inode,
3324 ext4_lblk_t lblk_start,
3325 ext4_lblk_t lblk_end,
3326 int search_hint_reverse)
3328 struct address_space *mapping = inode->i_mapping;
3329 struct buffer_head *head, *bh = NULL;
3331 ext4_lblk_t i, pg_lblk;
3334 /* reverse search wont work if fs block size is less than page size */
3335 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3336 search_hint_reverse = 0;
3338 if (search_hint_reverse)
3343 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3345 while ((i >= lblk_start) && (i <= lblk_end)) {
3346 page = find_get_page(mapping, index);
3350 if (!page_has_buffers(page))
3353 head = page_buffers(page);
3358 pg_lblk = index << (PAGE_CACHE_SHIFT -
3361 if (unlikely(pg_lblk < lblk_start)) {
3363 * This is possible when fs block size is less
3364 * than page size and our cluster starts/ends in
3365 * middle of the page. So we need to skip the
3366 * initial few blocks till we reach the 'lblk'
3372 /* Check if the buffer is delayed allocated and that it
3373 * is not yet mapped. (when da-buffers are mapped during
3374 * their writeout, their da_mapped bit is set.)
3376 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3377 page_cache_release(page);
3378 trace_ext4_find_delalloc_range(inode,
3379 lblk_start, lblk_end,
3380 search_hint_reverse,
3384 if (search_hint_reverse)
3388 } while ((i >= lblk_start) && (i <= lblk_end) &&
3389 ((bh = bh->b_this_page) != head));
3392 page_cache_release(page);
3394 * Move to next page. 'i' will be the first lblk in the next
3397 if (search_hint_reverse)
3401 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3404 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3405 search_hint_reverse, 0, 0);
3409 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3410 int search_hint_reverse)
3412 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3413 ext4_lblk_t lblk_start, lblk_end;
3414 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3415 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3417 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3418 search_hint_reverse);
3422 * Determines how many complete clusters (out of those specified by the 'map')
3423 * are under delalloc and were reserved quota for.
3424 * This function is called when we are writing out the blocks that were
3425 * originally written with their allocation delayed, but then the space was
3426 * allocated using fallocate() before the delayed allocation could be resolved.
3427 * The cases to look for are:
3428 * ('=' indicated delayed allocated blocks
3429 * '-' indicates non-delayed allocated blocks)
3430 * (a) partial clusters towards beginning and/or end outside of allocated range
3431 * are not delalloc'ed.
3433 * |----c---=|====c====|====c====|===-c----|
3434 * |++++++ allocated ++++++|
3435 * ==> 4 complete clusters in above example
3437 * (b) partial cluster (outside of allocated range) towards either end is
3438 * marked for delayed allocation. In this case, we will exclude that
3441 * |----====c========|========c========|
3442 * |++++++ allocated ++++++|
3443 * ==> 1 complete clusters in above example
3446 * |================c================|
3447 * |++++++ allocated ++++++|
3448 * ==> 0 complete clusters in above example
3450 * The ext4_da_update_reserve_space will be called only if we
3451 * determine here that there were some "entire" clusters that span
3452 * this 'allocated' range.
3453 * In the non-bigalloc case, this function will just end up returning num_blks
3454 * without ever calling ext4_find_delalloc_range.
3457 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3458 unsigned int num_blks)
3460 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3461 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3462 ext4_lblk_t lblk_from, lblk_to, c_offset;
3463 unsigned int allocated_clusters = 0;
3465 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3466 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3468 /* max possible clusters for this allocation */
3469 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3471 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3473 /* Check towards left side */
3474 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3476 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3477 lblk_to = lblk_from + c_offset - 1;
3479 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3480 allocated_clusters--;
3483 /* Now check towards right. */
3484 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3485 if (allocated_clusters && c_offset) {
3486 lblk_from = lblk_start + num_blks;
3487 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3489 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3490 allocated_clusters--;
3493 return allocated_clusters;
3497 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3498 struct ext4_map_blocks *map,
3499 struct ext4_ext_path *path, int flags,
3500 unsigned int allocated, ext4_fsblk_t newblock)
3504 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3506 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3507 "block %llu, max_blocks %u, flags %d, allocated %u",
3508 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3510 ext4_ext_show_leaf(inode, path);
3512 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3515 /* get_block() before submit the IO, split the extent */
3516 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3517 ret = ext4_split_unwritten_extents(handle, inode, map,
3520 * Flag the inode(non aio case) or end_io struct (aio case)
3521 * that this IO needs to conversion to written when IO is
3525 ext4_set_io_unwritten_flag(inode, io);
3527 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3528 if (ext4_should_dioread_nolock(inode))
3529 map->m_flags |= EXT4_MAP_UNINIT;
3532 /* IO end_io complete, convert the filled extent to written */
3533 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3534 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3537 ext4_update_inode_fsync_trans(handle, inode, 1);
3538 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3544 /* buffered IO case */
3546 * repeat fallocate creation request
3547 * we already have an unwritten extent
3549 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3552 /* buffered READ or buffered write_begin() lookup */
3553 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3555 * We have blocks reserved already. We
3556 * return allocated blocks so that delalloc
3557 * won't do block reservation for us. But
3558 * the buffer head will be unmapped so that
3559 * a read from the block returns 0s.
3561 map->m_flags |= EXT4_MAP_UNWRITTEN;
3565 /* buffered write, writepage time, convert*/
3566 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3568 ext4_update_inode_fsync_trans(handle, inode, 1);
3575 map->m_flags |= EXT4_MAP_NEW;
3577 * if we allocated more blocks than requested
3578 * we need to make sure we unmap the extra block
3579 * allocated. The actual needed block will get
3580 * unmapped later when we find the buffer_head marked
3583 if (allocated > map->m_len) {
3584 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3585 newblock + map->m_len,
3586 allocated - map->m_len);
3587 allocated = map->m_len;
3591 * If we have done fallocate with the offset that is already
3592 * delayed allocated, we would have block reservation
3593 * and quota reservation done in the delayed write path.
3594 * But fallocate would have already updated quota and block
3595 * count for this offset. So cancel these reservation
3597 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3598 unsigned int reserved_clusters;
3599 reserved_clusters = get_reserved_cluster_alloc(inode,
3600 map->m_lblk, map->m_len);
3601 if (reserved_clusters)
3602 ext4_da_update_reserve_space(inode,
3608 map->m_flags |= EXT4_MAP_MAPPED;
3609 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3610 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3616 if (allocated > map->m_len)
3617 allocated = map->m_len;
3618 ext4_ext_show_leaf(inode, path);
3619 map->m_pblk = newblock;
3620 map->m_len = allocated;
3623 ext4_ext_drop_refs(path);
3626 return err ? err : allocated;
3630 * get_implied_cluster_alloc - check to see if the requested
3631 * allocation (in the map structure) overlaps with a cluster already
3632 * allocated in an extent.
3633 * @sb The filesystem superblock structure
3634 * @map The requested lblk->pblk mapping
3635 * @ex The extent structure which might contain an implied
3636 * cluster allocation
3638 * This function is called by ext4_ext_map_blocks() after we failed to
3639 * find blocks that were already in the inode's extent tree. Hence,
3640 * we know that the beginning of the requested region cannot overlap
3641 * the extent from the inode's extent tree. There are three cases we
3642 * want to catch. The first is this case:
3644 * |--- cluster # N--|
3645 * |--- extent ---| |---- requested region ---|
3648 * The second case that we need to test for is this one:
3650 * |--------- cluster # N ----------------|
3651 * |--- requested region --| |------- extent ----|
3652 * |=======================|
3654 * The third case is when the requested region lies between two extents
3655 * within the same cluster:
3656 * |------------- cluster # N-------------|
3657 * |----- ex -----| |---- ex_right ----|
3658 * |------ requested region ------|
3659 * |================|
3661 * In each of the above cases, we need to set the map->m_pblk and
3662 * map->m_len so it corresponds to the return the extent labelled as
3663 * "|====|" from cluster #N, since it is already in use for data in
3664 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3665 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3666 * as a new "allocated" block region. Otherwise, we will return 0 and
3667 * ext4_ext_map_blocks() will then allocate one or more new clusters
3668 * by calling ext4_mb_new_blocks().
3670 static int get_implied_cluster_alloc(struct super_block *sb,
3671 struct ext4_map_blocks *map,
3672 struct ext4_extent *ex,
3673 struct ext4_ext_path *path)
3675 struct ext4_sb_info *sbi = EXT4_SB(sb);
3676 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3677 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3678 ext4_lblk_t rr_cluster_start, rr_cluster_end;
3679 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3680 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3681 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3683 /* The extent passed in that we are trying to match */
3684 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3685 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3687 /* The requested region passed into ext4_map_blocks() */
3688 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3689 rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3691 if ((rr_cluster_start == ex_cluster_end) ||
3692 (rr_cluster_start == ex_cluster_start)) {
3693 if (rr_cluster_start == ex_cluster_end)
3694 ee_start += ee_len - 1;
3695 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3697 map->m_len = min(map->m_len,
3698 (unsigned) sbi->s_cluster_ratio - c_offset);
3700 * Check for and handle this case:
3702 * |--------- cluster # N-------------|
3703 * |------- extent ----|
3704 * |--- requested region ---|
3708 if (map->m_lblk < ee_block)
3709 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3712 * Check for the case where there is already another allocated
3713 * block to the right of 'ex' but before the end of the cluster.
3715 * |------------- cluster # N-------------|
3716 * |----- ex -----| |---- ex_right ----|
3717 * |------ requested region ------|
3718 * |================|
3720 if (map->m_lblk > ee_block) {
3721 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3722 map->m_len = min(map->m_len, next - map->m_lblk);
3725 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3729 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3735 * Block allocation/map/preallocation routine for extents based files
3738 * Need to be called with
3739 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3740 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3742 * return > 0, number of of blocks already mapped/allocated
3743 * if create == 0 and these are pre-allocated blocks
3744 * buffer head is unmapped
3745 * otherwise blocks are mapped
3747 * return = 0, if plain look up failed (blocks have not been allocated)
3748 * buffer head is unmapped
3750 * return < 0, error case.
3752 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3753 struct ext4_map_blocks *map, int flags)
3755 struct ext4_ext_path *path = NULL;
3756 struct ext4_extent newex, *ex, *ex2;
3757 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3758 ext4_fsblk_t newblock = 0;
3759 int free_on_err = 0, err = 0, depth, ret;
3760 unsigned int allocated = 0, offset = 0;
3761 unsigned int allocated_clusters = 0;
3762 unsigned int punched_out = 0;
3763 unsigned int result = 0;
3764 struct ext4_allocation_request ar;
3765 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3766 ext4_lblk_t cluster_offset;
3768 ext_debug("blocks %u/%u requested for inode %lu\n",
3769 map->m_lblk, map->m_len, inode->i_ino);
3770 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3772 /* check in cache */
3773 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3774 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3775 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3776 if ((sbi->s_cluster_ratio > 1) &&
3777 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3778 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3780 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3782 * block isn't allocated yet and
3783 * user doesn't want to allocate it
3787 /* we should allocate requested block */
3789 /* block is already allocated */
3790 if (sbi->s_cluster_ratio > 1)
3791 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3792 newblock = map->m_lblk
3793 - le32_to_cpu(newex.ee_block)
3794 + ext4_ext_pblock(&newex);
3795 /* number of remaining blocks in the extent */
3796 allocated = ext4_ext_get_actual_len(&newex) -
3797 (map->m_lblk - le32_to_cpu(newex.ee_block));
3802 /* find extent for this block */
3803 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3805 err = PTR_ERR(path);
3810 depth = ext_depth(inode);
3813 * consistent leaf must not be empty;
3814 * this situation is possible, though, _during_ tree modification;
3815 * this is why assert can't be put in ext4_ext_find_extent()
3817 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3818 EXT4_ERROR_INODE(inode, "bad extent address "
3819 "lblock: %lu, depth: %d pblock %lld",
3820 (unsigned long) map->m_lblk, depth,
3821 path[depth].p_block);
3826 ex = path[depth].p_ext;
3828 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3829 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3830 unsigned short ee_len;
3833 * Uninitialized extents are treated as holes, except that
3834 * we split out initialized portions during a write.
3836 ee_len = ext4_ext_get_actual_len(ex);
3838 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3840 /* if found extent covers block, simply return it */
3841 if (in_range(map->m_lblk, ee_block, ee_len)) {
3842 struct ext4_map_blocks punch_map;
3843 ext4_fsblk_t partial_cluster = 0;
3845 newblock = map->m_lblk - ee_block + ee_start;
3846 /* number of remaining blocks in the extent */
3847 allocated = ee_len - (map->m_lblk - ee_block);
3848 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3849 ee_block, ee_len, newblock);
3851 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3853 * Do not put uninitialized extent
3856 if (!ext4_ext_is_uninitialized(ex)) {
3857 ext4_ext_put_in_cache(inode, ee_block,
3861 ret = ext4_ext_handle_uninitialized_extents(
3862 handle, inode, map, path, flags,
3863 allocated, newblock);
3868 * Punch out the map length, but only to the
3871 punched_out = allocated < map->m_len ?
3872 allocated : map->m_len;
3875 * Sense extents need to be converted to
3876 * uninitialized, they must fit in an
3877 * uninitialized extent
3879 if (punched_out > EXT_UNINIT_MAX_LEN)
3880 punched_out = EXT_UNINIT_MAX_LEN;
3882 punch_map.m_lblk = map->m_lblk;
3883 punch_map.m_pblk = newblock;
3884 punch_map.m_len = punched_out;
3885 punch_map.m_flags = 0;
3887 /* Check to see if the extent needs to be split */
3888 if (punch_map.m_len != ee_len ||
3889 punch_map.m_lblk != ee_block) {
3891 ret = ext4_split_extent(handle, inode,
3892 path, &punch_map, 0,
3893 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3894 EXT4_GET_BLOCKS_PRE_IO);
3901 * find extent for the block at
3902 * the start of the hole
3904 ext4_ext_drop_refs(path);
3907 path = ext4_ext_find_extent(inode,
3910 err = PTR_ERR(path);
3915 depth = ext_depth(inode);
3916 ex = path[depth].p_ext;
3917 ee_len = ext4_ext_get_actual_len(ex);
3918 ee_block = le32_to_cpu(ex->ee_block);
3919 ee_start = ext4_ext_pblock(ex);
3923 ext4_ext_mark_uninitialized(ex);
3925 ext4_ext_invalidate_cache(inode);
3927 err = ext4_ext_rm_leaf(handle, inode, path,
3928 &partial_cluster, map->m_lblk,
3929 map->m_lblk + punched_out);
3931 if (!err && path->p_hdr->eh_entries == 0) {
3933 * Punch hole freed all of this sub tree,
3934 * so we need to correct eh_depth
3936 err = ext4_ext_get_access(handle, inode, path);
3938 ext_inode_hdr(inode)->eh_depth = 0;
3939 ext_inode_hdr(inode)->eh_max =
3940 cpu_to_le16(ext4_ext_space_root(
3943 err = ext4_ext_dirty(
3944 handle, inode, path);
3952 if ((sbi->s_cluster_ratio > 1) &&
3953 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3954 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3957 * requested block isn't allocated yet;
3958 * we couldn't try to create block if create flag is zero
3960 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3962 * put just found gap into cache to speed up
3963 * subsequent requests
3965 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3970 * Okay, we need to do block allocation.
3972 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
3973 newex.ee_block = cpu_to_le32(map->m_lblk);
3974 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3977 * If we are doing bigalloc, check to see if the extent returned
3978 * by ext4_ext_find_extent() implies a cluster we can use.
3980 if (cluster_offset && ex &&
3981 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
3982 ar.len = allocated = map->m_len;
3983 newblock = map->m_pblk;
3984 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3985 goto got_allocated_blocks;
3988 /* find neighbour allocated blocks */
3989 ar.lleft = map->m_lblk;
3990 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3993 ar.lright = map->m_lblk;
3995 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
3999 /* Check if the extent after searching to the right implies a
4000 * cluster we can use. */
4001 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4002 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4003 ar.len = allocated = map->m_len;
4004 newblock = map->m_pblk;
4005 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4006 goto got_allocated_blocks;
4010 * See if request is beyond maximum number of blocks we can have in
4011 * a single extent. For an initialized extent this limit is
4012 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4013 * EXT_UNINIT_MAX_LEN.
4015 if (map->m_len > EXT_INIT_MAX_LEN &&
4016 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4017 map->m_len = EXT_INIT_MAX_LEN;
4018 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4019 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4020 map->m_len = EXT_UNINIT_MAX_LEN;
4022 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4023 newex.ee_len = cpu_to_le16(map->m_len);
4024 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4026 allocated = ext4_ext_get_actual_len(&newex);
4028 allocated = map->m_len;
4030 /* allocate new block */
4032 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4033 ar.logical = map->m_lblk;
4035 * We calculate the offset from the beginning of the cluster
4036 * for the logical block number, since when we allocate a
4037 * physical cluster, the physical block should start at the
4038 * same offset from the beginning of the cluster. This is
4039 * needed so that future calls to get_implied_cluster_alloc()
4042 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4043 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4045 ar.logical -= offset;
4046 if (S_ISREG(inode->i_mode))
4047 ar.flags = EXT4_MB_HINT_DATA;
4049 /* disable in-core preallocation for non-regular files */
4051 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4052 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4053 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4056 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4057 ar.goal, newblock, allocated);
4059 allocated_clusters = ar.len;
4060 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4061 if (ar.len > allocated)
4064 got_allocated_blocks:
4065 /* try to insert new extent into found leaf and return */
4066 ext4_ext_store_pblock(&newex, newblock + offset);
4067 newex.ee_len = cpu_to_le16(ar.len);
4068 /* Mark uninitialized */
4069 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4070 ext4_ext_mark_uninitialized(&newex);
4072 * io_end structure was created for every IO write to an
4073 * uninitialized extent. To avoid unnecessary conversion,
4074 * here we flag the IO that really needs the conversion.
4075 * For non asycn direct IO case, flag the inode state
4076 * that we need to perform conversion when IO is done.
4078 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4080 ext4_set_io_unwritten_flag(inode, io);
4082 ext4_set_inode_state(inode,
4083 EXT4_STATE_DIO_UNWRITTEN);
4085 if (ext4_should_dioread_nolock(inode))
4086 map->m_flags |= EXT4_MAP_UNINIT;
4090 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4091 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4094 err = ext4_ext_insert_extent(handle, inode, path,
4096 if (err && free_on_err) {
4097 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4098 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4099 /* free data blocks we just allocated */
4100 /* not a good idea to call discard here directly,
4101 * but otherwise we'd need to call it every free() */
4102 ext4_discard_preallocations(inode);
4103 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4104 ext4_ext_get_actual_len(&newex), fb_flags);
4108 /* previous routine could use block we allocated */
4109 newblock = ext4_ext_pblock(&newex);
4110 allocated = ext4_ext_get_actual_len(&newex);
4111 if (allocated > map->m_len)
4112 allocated = map->m_len;
4113 map->m_flags |= EXT4_MAP_NEW;
4116 * Update reserved blocks/metadata blocks after successful
4117 * block allocation which had been deferred till now.
4119 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4120 unsigned int reserved_clusters;
4122 * Check how many clusters we had reserved this allocated range
4124 reserved_clusters = get_reserved_cluster_alloc(inode,
4125 map->m_lblk, allocated);
4126 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4127 if (reserved_clusters) {
4129 * We have clusters reserved for this range.
4130 * But since we are not doing actual allocation
4131 * and are simply using blocks from previously
4132 * allocated cluster, we should release the
4133 * reservation and not claim quota.
4135 ext4_da_update_reserve_space(inode,
4136 reserved_clusters, 0);
4139 BUG_ON(allocated_clusters < reserved_clusters);
4140 /* We will claim quota for all newly allocated blocks.*/
4141 ext4_da_update_reserve_space(inode, allocated_clusters,
4143 if (reserved_clusters < allocated_clusters) {
4144 struct ext4_inode_info *ei = EXT4_I(inode);
4145 int reservation = allocated_clusters -
4148 * It seems we claimed few clusters outside of
4149 * the range of this allocation. We should give
4150 * it back to the reservation pool. This can
4151 * happen in the following case:
4153 * * Suppose s_cluster_ratio is 4 (i.e., each
4154 * cluster has 4 blocks. Thus, the clusters
4155 * are [0-3],[4-7],[8-11]...
4156 * * First comes delayed allocation write for
4157 * logical blocks 10 & 11. Since there were no
4158 * previous delayed allocated blocks in the
4159 * range [8-11], we would reserve 1 cluster
4161 * * Next comes write for logical blocks 3 to 8.
4162 * In this case, we will reserve 2 clusters
4163 * (for [0-3] and [4-7]; and not for [8-11] as
4164 * that range has a delayed allocated blocks.
4165 * Thus total reserved clusters now becomes 3.
4166 * * Now, during the delayed allocation writeout
4167 * time, we will first write blocks [3-8] and
4168 * allocate 3 clusters for writing these
4169 * blocks. Also, we would claim all these
4170 * three clusters above.
4171 * * Now when we come here to writeout the
4172 * blocks [10-11], we would expect to claim
4173 * the reservation of 1 cluster we had made
4174 * (and we would claim it since there are no
4175 * more delayed allocated blocks in the range
4176 * [8-11]. But our reserved cluster count had
4177 * already gone to 0.
4179 * Thus, at the step 4 above when we determine
4180 * that there are still some unwritten delayed
4181 * allocated blocks outside of our current
4182 * block range, we should increment the
4183 * reserved clusters count so that when the
4184 * remaining blocks finally gets written, we
4187 dquot_reserve_block(inode,
4188 EXT4_C2B(sbi, reservation));
4189 spin_lock(&ei->i_block_reservation_lock);
4190 ei->i_reserved_data_blocks += reservation;
4191 spin_unlock(&ei->i_block_reservation_lock);
4197 * Cache the extent and update transaction to commit on fdatasync only
4198 * when it is _not_ an uninitialized extent.
4200 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4201 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4202 ext4_update_inode_fsync_trans(handle, inode, 1);
4204 ext4_update_inode_fsync_trans(handle, inode, 0);
4206 if (allocated > map->m_len)
4207 allocated = map->m_len;
4208 ext4_ext_show_leaf(inode, path);
4209 map->m_flags |= EXT4_MAP_MAPPED;
4210 map->m_pblk = newblock;
4211 map->m_len = allocated;
4214 ext4_ext_drop_refs(path);
4217 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4218 punched_out : allocated;
4220 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4221 newblock, map->m_len, err ? err : result);
4223 return err ? err : result;
4226 void ext4_ext_truncate(struct inode *inode)
4228 struct address_space *mapping = inode->i_mapping;
4229 struct super_block *sb = inode->i_sb;
4230 ext4_lblk_t last_block;
4236 * finish any pending end_io work so we won't run the risk of
4237 * converting any truncated blocks to initialized later
4239 ext4_flush_completed_IO(inode);
4242 * probably first extent we're gonna free will be last in block
4244 err = ext4_writepage_trans_blocks(inode);
4245 handle = ext4_journal_start(inode, err);
4249 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4250 page_len = PAGE_CACHE_SIZE -
4251 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4253 err = ext4_discard_partial_page_buffers(handle,
4254 mapping, inode->i_size, page_len, 0);
4260 if (ext4_orphan_add(handle, inode))
4263 down_write(&EXT4_I(inode)->i_data_sem);
4264 ext4_ext_invalidate_cache(inode);
4266 ext4_discard_preallocations(inode);
4269 * TODO: optimization is possible here.
4270 * Probably we need not scan at all,
4271 * because page truncation is enough.
4274 /* we have to know where to truncate from in crash case */
4275 EXT4_I(inode)->i_disksize = inode->i_size;
4276 ext4_mark_inode_dirty(handle, inode);
4278 last_block = (inode->i_size + sb->s_blocksize - 1)
4279 >> EXT4_BLOCK_SIZE_BITS(sb);
4280 err = ext4_ext_remove_space(inode, last_block);
4282 /* In a multi-transaction truncate, we only make the final
4283 * transaction synchronous.
4286 ext4_handle_sync(handle);
4288 up_write(&EXT4_I(inode)->i_data_sem);
4292 * If this was a simple ftruncate() and the file will remain alive,
4293 * then we need to clear up the orphan record which we created above.
4294 * However, if this was a real unlink then we were called by
4295 * ext4_delete_inode(), and we allow that function to clean up the
4296 * orphan info for us.
4299 ext4_orphan_del(handle, inode);
4301 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4302 ext4_mark_inode_dirty(handle, inode);
4303 ext4_journal_stop(handle);
4306 static void ext4_falloc_update_inode(struct inode *inode,
4307 int mode, loff_t new_size, int update_ctime)
4309 struct timespec now;
4312 now = current_fs_time(inode->i_sb);
4313 if (!timespec_equal(&inode->i_ctime, &now))
4314 inode->i_ctime = now;
4317 * Update only when preallocation was requested beyond
4320 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4321 if (new_size > i_size_read(inode))
4322 i_size_write(inode, new_size);
4323 if (new_size > EXT4_I(inode)->i_disksize)
4324 ext4_update_i_disksize(inode, new_size);
4327 * Mark that we allocate beyond EOF so the subsequent truncate
4328 * can proceed even if the new size is the same as i_size.
4330 if (new_size > i_size_read(inode))
4331 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4337 * preallocate space for a file. This implements ext4's fallocate file
4338 * operation, which gets called from sys_fallocate system call.
4339 * For block-mapped files, posix_fallocate should fall back to the method
4340 * of writing zeroes to the required new blocks (the same behavior which is
4341 * expected for file systems which do not support fallocate() system call).
4343 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4345 struct inode *inode = file->f_path.dentry->d_inode;
4348 unsigned int max_blocks;
4353 struct ext4_map_blocks map;
4354 unsigned int credits, blkbits = inode->i_blkbits;
4357 * currently supporting (pre)allocate mode for extent-based
4360 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4363 /* Return error if mode is not supported */
4364 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4367 if (mode & FALLOC_FL_PUNCH_HOLE)
4368 return ext4_punch_hole(file, offset, len);
4370 trace_ext4_fallocate_enter(inode, offset, len, mode);
4371 map.m_lblk = offset >> blkbits;
4373 * We can't just convert len to max_blocks because
4374 * If blocksize = 4096 offset = 3072 and len = 2048
4376 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4379 * credits to insert 1 extent into extent tree
4381 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4382 mutex_lock(&inode->i_mutex);
4383 ret = inode_newsize_ok(inode, (len + offset));
4385 mutex_unlock(&inode->i_mutex);
4386 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4389 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4390 if (mode & FALLOC_FL_KEEP_SIZE)
4391 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4393 * Don't normalize the request if it can fit in one extent so
4394 * that it doesn't get unnecessarily split into multiple
4397 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4398 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4400 while (ret >= 0 && ret < max_blocks) {
4401 map.m_lblk = map.m_lblk + ret;
4402 map.m_len = max_blocks = max_blocks - ret;
4403 handle = ext4_journal_start(inode, credits);
4404 if (IS_ERR(handle)) {
4405 ret = PTR_ERR(handle);
4408 ret = ext4_map_blocks(handle, inode, &map, flags);
4412 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4413 "returned error inode#%lu, block=%u, "
4414 "max_blocks=%u", __func__,
4415 inode->i_ino, map.m_lblk, max_blocks);
4417 ext4_mark_inode_dirty(handle, inode);
4418 ret2 = ext4_journal_stop(handle);
4421 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4422 blkbits) >> blkbits))
4423 new_size = offset + len;
4425 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4427 ext4_falloc_update_inode(inode, mode, new_size,
4428 (map.m_flags & EXT4_MAP_NEW));
4429 ext4_mark_inode_dirty(handle, inode);
4430 ret2 = ext4_journal_stop(handle);
4434 if (ret == -ENOSPC &&
4435 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4439 mutex_unlock(&inode->i_mutex);
4440 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4441 ret > 0 ? ret2 : ret);
4442 return ret > 0 ? ret2 : ret;
4446 * This function convert a range of blocks to written extents
4447 * The caller of this function will pass the start offset and the size.
4448 * all unwritten extents within this range will be converted to
4451 * This function is called from the direct IO end io call back
4452 * function, to convert the fallocated extents after IO is completed.
4453 * Returns 0 on success.
4455 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4459 unsigned int max_blocks;
4462 struct ext4_map_blocks map;
4463 unsigned int credits, blkbits = inode->i_blkbits;
4465 map.m_lblk = offset >> blkbits;
4467 * We can't just convert len to max_blocks because
4468 * If blocksize = 4096 offset = 3072 and len = 2048
4470 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4473 * credits to insert 1 extent into extent tree
4475 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4476 while (ret >= 0 && ret < max_blocks) {
4478 map.m_len = (max_blocks -= ret);
4479 handle = ext4_journal_start(inode, credits);
4480 if (IS_ERR(handle)) {
4481 ret = PTR_ERR(handle);
4484 ret = ext4_map_blocks(handle, inode, &map,
4485 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4488 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4489 "returned error inode#%lu, block=%u, "
4490 "max_blocks=%u", __func__,
4491 inode->i_ino, map.m_lblk, map.m_len);
4493 ext4_mark_inode_dirty(handle, inode);
4494 ret2 = ext4_journal_stop(handle);
4495 if (ret <= 0 || ret2 )
4498 return ret > 0 ? ret2 : ret;
4502 * Callback function called for each extent to gather FIEMAP information.
4504 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4505 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4513 struct fiemap_extent_info *fieinfo = data;
4514 unsigned char blksize_bits;
4516 blksize_bits = inode->i_sb->s_blocksize_bits;
4517 logical = (__u64)newex->ec_block << blksize_bits;
4519 if (newex->ec_start == 0) {
4521 * No extent in extent-tree contains block @newex->ec_start,
4522 * then the block may stay in 1)a hole or 2)delayed-extent.
4524 * Holes or delayed-extents are processed as follows.
4525 * 1. lookup dirty pages with specified range in pagecache.
4526 * If no page is got, then there is no delayed-extent and
4527 * return with EXT_CONTINUE.
4528 * 2. find the 1st mapped buffer,
4529 * 3. check if the mapped buffer is both in the request range
4530 * and a delayed buffer. If not, there is no delayed-extent,
4532 * 4. a delayed-extent is found, the extent will be collected.
4534 ext4_lblk_t end = 0;
4535 pgoff_t last_offset;
4538 pgoff_t start_index = 0;
4539 struct page **pages = NULL;
4540 struct buffer_head *bh = NULL;
4541 struct buffer_head *head = NULL;
4542 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4544 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4548 offset = logical >> PAGE_SHIFT;
4550 last_offset = offset;
4552 ret = find_get_pages_tag(inode->i_mapping, &offset,
4553 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4555 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4556 /* First time, try to find a mapped buffer. */
4559 for (index = 0; index < ret; index++)
4560 page_cache_release(pages[index]);
4563 return EXT_CONTINUE;
4568 /* Try to find the 1st mapped buffer. */
4569 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4571 if (!page_has_buffers(pages[index]))
4573 head = page_buffers(pages[index]);
4580 if (end >= newex->ec_block +
4582 /* The buffer is out of
4583 * the request range.
4587 if (buffer_mapped(bh) &&
4588 end >= newex->ec_block) {
4589 start_index = index - 1;
4590 /* get the 1st mapped buffer. */
4591 goto found_mapped_buffer;
4594 bh = bh->b_this_page;
4596 } while (bh != head);
4598 /* No mapped buffer in the range found in this page,
4599 * We need to look up next page.
4602 /* There is no page left, but we need to limit
4605 newex->ec_len = end - newex->ec_block;
4610 /*Find contiguous delayed buffers. */
4611 if (ret > 0 && pages[0]->index == last_offset)
4612 head = page_buffers(pages[0]);
4618 found_mapped_buffer:
4619 if (bh != NULL && buffer_delay(bh)) {
4620 /* 1st or contiguous delayed buffer found. */
4621 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4623 * 1st delayed buffer found, record
4624 * the start of extent.
4626 flags |= FIEMAP_EXTENT_DELALLOC;
4627 newex->ec_block = end;
4628 logical = (__u64)end << blksize_bits;
4630 /* Find contiguous delayed buffers. */
4632 if (!buffer_delay(bh))
4633 goto found_delayed_extent;
4634 bh = bh->b_this_page;
4636 } while (bh != head);
4638 for (; index < ret; index++) {
4639 if (!page_has_buffers(pages[index])) {
4643 head = page_buffers(pages[index]);
4649 if (pages[index]->index !=
4650 pages[start_index]->index + index
4652 /* Blocks are not contiguous. */
4658 if (!buffer_delay(bh))
4659 /* Delayed-extent ends. */
4660 goto found_delayed_extent;
4661 bh = bh->b_this_page;
4663 } while (bh != head);
4665 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4669 found_delayed_extent:
4670 newex->ec_len = min(end - newex->ec_block,
4671 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4672 if (ret == nr_pages && bh != NULL &&
4673 newex->ec_len < EXT_INIT_MAX_LEN &&
4675 /* Have not collected an extent and continue. */
4676 for (index = 0; index < ret; index++)
4677 page_cache_release(pages[index]);
4681 for (index = 0; index < ret; index++)
4682 page_cache_release(pages[index]);
4686 physical = (__u64)newex->ec_start << blksize_bits;
4687 length = (__u64)newex->ec_len << blksize_bits;
4689 if (ex && ext4_ext_is_uninitialized(ex))
4690 flags |= FIEMAP_EXTENT_UNWRITTEN;
4692 if (next == EXT_MAX_BLOCKS)
4693 flags |= FIEMAP_EXTENT_LAST;
4695 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4701 return EXT_CONTINUE;
4703 /* fiemap flags we can handle specified here */
4704 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4706 static int ext4_xattr_fiemap(struct inode *inode,
4707 struct fiemap_extent_info *fieinfo)
4711 __u32 flags = FIEMAP_EXTENT_LAST;
4712 int blockbits = inode->i_sb->s_blocksize_bits;
4716 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4717 struct ext4_iloc iloc;
4718 int offset; /* offset of xattr in inode */
4720 error = ext4_get_inode_loc(inode, &iloc);
4723 physical = iloc.bh->b_blocknr << blockbits;
4724 offset = EXT4_GOOD_OLD_INODE_SIZE +
4725 EXT4_I(inode)->i_extra_isize;
4727 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4728 flags |= FIEMAP_EXTENT_DATA_INLINE;
4730 } else { /* external block */
4731 physical = EXT4_I(inode)->i_file_acl << blockbits;
4732 length = inode->i_sb->s_blocksize;
4736 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4738 return (error < 0 ? error : 0);
4742 * ext4_ext_punch_hole
4744 * Punches a hole of "length" bytes in a file starting
4747 * @inode: The inode of the file to punch a hole in
4748 * @offset: The starting byte offset of the hole
4749 * @length: The length of the hole
4751 * Returns the number of blocks removed or negative on err
4753 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4755 struct inode *inode = file->f_path.dentry->d_inode;
4756 struct super_block *sb = inode->i_sb;
4757 struct ext4_ext_cache cache_ex;
4758 ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4759 struct address_space *mapping = inode->i_mapping;
4760 struct ext4_map_blocks map;
4762 loff_t first_page, last_page, page_len;
4763 loff_t first_page_offset, last_page_offset;
4764 int ret, credits, blocks_released, err = 0;
4766 /* No need to punch hole beyond i_size */
4767 if (offset >= inode->i_size)
4771 * If the hole extends beyond i_size, set the hole
4772 * to end after the page that contains i_size
4774 if (offset + length > inode->i_size) {
4775 length = inode->i_size +
4776 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4780 first_block = (offset + sb->s_blocksize - 1) >>
4781 EXT4_BLOCK_SIZE_BITS(sb);
4782 last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4784 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4785 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4787 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4788 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4791 * Write out all dirty pages to avoid race conditions
4792 * Then release them.
4794 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4795 err = filemap_write_and_wait_range(mapping,
4796 offset, offset + length - 1);
4802 /* Now release the pages */
4803 if (last_page_offset > first_page_offset) {
4804 truncate_inode_pages_range(mapping, first_page_offset,
4805 last_page_offset-1);
4808 /* finish any pending end_io work */
4809 ext4_flush_completed_IO(inode);
4811 credits = ext4_writepage_trans_blocks(inode);
4812 handle = ext4_journal_start(inode, credits);
4814 return PTR_ERR(handle);
4816 err = ext4_orphan_add(handle, inode);
4821 * Now we need to zero out the non-page-aligned data in the
4822 * pages at the start and tail of the hole, and unmap the buffer
4823 * heads for the block aligned regions of the page that were
4824 * completely zeroed.
4826 if (first_page > last_page) {
4828 * If the file space being truncated is contained within a page
4829 * just zero out and unmap the middle of that page
4831 err = ext4_discard_partial_page_buffers(handle,
4832 mapping, offset, length, 0);
4838 * zero out and unmap the partial page that contains
4839 * the start of the hole
4841 page_len = first_page_offset - offset;
4843 err = ext4_discard_partial_page_buffers(handle, mapping,
4844 offset, page_len, 0);
4850 * zero out and unmap the partial page that contains
4851 * the end of the hole
4853 page_len = offset + length - last_page_offset;
4855 err = ext4_discard_partial_page_buffers(handle, mapping,
4856 last_page_offset, page_len, 0);
4864 * If i_size is contained in the last page, we need to
4865 * unmap and zero the partial page after i_size
4867 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4868 inode->i_size % PAGE_CACHE_SIZE != 0) {
4870 page_len = PAGE_CACHE_SIZE -
4871 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4874 err = ext4_discard_partial_page_buffers(handle,
4875 mapping, inode->i_size, page_len, 0);
4882 /* If there are no blocks to remove, return now */
4883 if (first_block >= last_block)
4886 down_write(&EXT4_I(inode)->i_data_sem);
4887 ext4_ext_invalidate_cache(inode);
4888 ext4_discard_preallocations(inode);
4891 * Loop over all the blocks and identify blocks
4892 * that need to be punched out
4894 iblock = first_block;
4895 blocks_released = 0;
4896 while (iblock < last_block) {
4897 max_blocks = last_block - iblock;
4899 memset(&map, 0, sizeof(map));
4900 map.m_lblk = iblock;
4901 map.m_len = max_blocks;
4902 ret = ext4_ext_map_blocks(handle, inode, &map,
4903 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4906 blocks_released += ret;
4908 } else if (ret == 0) {
4910 * If map blocks could not find the block,
4911 * then it is in a hole. If the hole was
4912 * not already cached, then map blocks should
4913 * put it in the cache. So we can get the hole
4916 memset(&cache_ex, 0, sizeof(cache_ex));
4917 if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4918 !cache_ex.ec_start) {
4920 /* The hole is cached */
4921 num_blocks = cache_ex.ec_block +
4922 cache_ex.ec_len - iblock;
4925 /* The block could not be identified */
4930 /* Map blocks error */
4935 if (num_blocks == 0) {
4936 /* This condition should never happen */
4937 ext_debug("Block lookup failed");
4942 iblock += num_blocks;
4945 if (blocks_released > 0) {
4946 ext4_ext_invalidate_cache(inode);
4947 ext4_discard_preallocations(inode);
4951 ext4_handle_sync(handle);
4953 up_write(&EXT4_I(inode)->i_data_sem);
4956 ext4_orphan_del(handle, inode);
4957 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4958 ext4_mark_inode_dirty(handle, inode);
4959 ext4_journal_stop(handle);
4962 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4963 __u64 start, __u64 len)
4965 ext4_lblk_t start_blk;
4968 /* fallback to generic here if not in extents fmt */
4969 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4970 return generic_block_fiemap(inode, fieinfo, start, len,
4973 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4976 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4977 error = ext4_xattr_fiemap(inode, fieinfo);
4979 ext4_lblk_t len_blks;
4982 start_blk = start >> inode->i_sb->s_blocksize_bits;
4983 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4984 if (last_blk >= EXT_MAX_BLOCKS)
4985 last_blk = EXT_MAX_BLOCKS-1;
4986 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4989 * Walk the extent tree gathering extent information.
4990 * ext4_ext_fiemap_cb will push extents back to user.
4992 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4993 ext4_ext_fiemap_cb, fieinfo);