2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
54 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static int ext4_split_extent(handle_t *handle,
61 struct ext4_ext_path *path,
62 struct ext4_map_blocks *map,
66 static int ext4_split_extent_at(handle_t *handle,
68 struct ext4_ext_path *path,
73 static int ext4_ext_truncate_extend_restart(handle_t *handle,
79 if (!ext4_handle_valid(handle))
81 if (handle->h_buffer_credits > needed)
83 err = ext4_journal_extend(handle, needed);
86 err = ext4_truncate_restart_trans(handle, inode, needed);
98 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
99 struct ext4_ext_path *path)
102 /* path points to block */
103 return ext4_journal_get_write_access(handle, path->p_bh);
105 /* path points to leaf/index in inode body */
106 /* we use in-core data, no need to protect them */
116 #define ext4_ext_dirty(handle, inode, path) \
117 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
118 static int __ext4_ext_dirty(const char *where, unsigned int line,
119 handle_t *handle, struct inode *inode,
120 struct ext4_ext_path *path)
124 /* path points to block */
125 err = __ext4_handle_dirty_metadata(where, line, handle,
128 /* path points to leaf/index in inode body */
129 err = ext4_mark_inode_dirty(handle, inode);
134 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
135 struct ext4_ext_path *path,
139 int depth = path->p_depth;
140 struct ext4_extent *ex;
143 * Try to predict block placement assuming that we are
144 * filling in a file which will eventually be
145 * non-sparse --- i.e., in the case of libbfd writing
146 * an ELF object sections out-of-order but in a way
147 * the eventually results in a contiguous object or
148 * executable file, or some database extending a table
149 * space file. However, this is actually somewhat
150 * non-ideal if we are writing a sparse file such as
151 * qemu or KVM writing a raw image file that is going
152 * to stay fairly sparse, since it will end up
153 * fragmenting the file system's free space. Maybe we
154 * should have some hueristics or some way to allow
155 * userspace to pass a hint to file system,
156 * especially if the latter case turns out to be
159 ex = path[depth].p_ext;
161 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
162 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
164 if (block > ext_block)
165 return ext_pblk + (block - ext_block);
167 return ext_pblk - (ext_block - block);
170 /* it looks like index is empty;
171 * try to find starting block from index itself */
172 if (path[depth].p_bh)
173 return path[depth].p_bh->b_blocknr;
176 /* OK. use inode's group */
177 return ext4_inode_to_goal_block(inode);
181 * Allocation for a meta data block
184 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
185 struct ext4_ext_path *path,
186 struct ext4_extent *ex, int *err, unsigned int flags)
188 ext4_fsblk_t goal, newblock;
190 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
191 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
196 static inline int ext4_ext_space_block(struct inode *inode, int check)
200 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
201 / sizeof(struct ext4_extent);
202 #ifdef AGGRESSIVE_TEST
203 if (!check && size > 6)
209 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
213 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
214 / sizeof(struct ext4_extent_idx);
215 #ifdef AGGRESSIVE_TEST
216 if (!check && size > 5)
222 static inline int ext4_ext_space_root(struct inode *inode, int check)
226 size = sizeof(EXT4_I(inode)->i_data);
227 size -= sizeof(struct ext4_extent_header);
228 size /= sizeof(struct ext4_extent);
229 #ifdef AGGRESSIVE_TEST
230 if (!check && size > 3)
236 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
240 size = sizeof(EXT4_I(inode)->i_data);
241 size -= sizeof(struct ext4_extent_header);
242 size /= sizeof(struct ext4_extent_idx);
243 #ifdef AGGRESSIVE_TEST
244 if (!check && size > 4)
251 * Calculate the number of metadata blocks needed
252 * to allocate @blocks
253 * Worse case is one block per extent
255 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
257 struct ext4_inode_info *ei = EXT4_I(inode);
260 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
261 / sizeof(struct ext4_extent_idx));
264 * If the new delayed allocation block is contiguous with the
265 * previous da block, it can share index blocks with the
266 * previous block, so we only need to allocate a new index
267 * block every idxs leaf blocks. At ldxs**2 blocks, we need
268 * an additional index block, and at ldxs**3 blocks, yet
269 * another index blocks.
271 if (ei->i_da_metadata_calc_len &&
272 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
275 if ((ei->i_da_metadata_calc_len % idxs) == 0)
277 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
279 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
281 ei->i_da_metadata_calc_len = 0;
283 ei->i_da_metadata_calc_len++;
284 ei->i_da_metadata_calc_last_lblock++;
289 * In the worst case we need a new set of index blocks at
290 * every level of the inode's extent tree.
292 ei->i_da_metadata_calc_len = 1;
293 ei->i_da_metadata_calc_last_lblock = lblock;
294 return ext_depth(inode) + 1;
298 ext4_ext_max_entries(struct inode *inode, int depth)
302 if (depth == ext_depth(inode)) {
304 max = ext4_ext_space_root(inode, 1);
306 max = ext4_ext_space_root_idx(inode, 1);
309 max = ext4_ext_space_block(inode, 1);
311 max = ext4_ext_space_block_idx(inode, 1);
317 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
319 ext4_fsblk_t block = ext4_ext_pblock(ext);
320 int len = ext4_ext_get_actual_len(ext);
321 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
322 ext4_lblk_t last = lblock + len - 1;
324 if (len == 0 || lblock > last)
326 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
329 static int ext4_valid_extent_idx(struct inode *inode,
330 struct ext4_extent_idx *ext_idx)
332 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
334 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
337 static int ext4_valid_extent_entries(struct inode *inode,
338 struct ext4_extent_header *eh,
341 unsigned short entries;
342 if (eh->eh_entries == 0)
345 entries = le16_to_cpu(eh->eh_entries);
349 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
350 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
351 ext4_fsblk_t pblock = 0;
352 ext4_lblk_t lblock = 0;
353 ext4_lblk_t prev = 0;
356 if (!ext4_valid_extent(inode, ext))
359 /* Check for overlapping extents */
360 lblock = le32_to_cpu(ext->ee_block);
361 len = ext4_ext_get_actual_len(ext);
362 if ((lblock <= prev) && prev) {
363 pblock = ext4_ext_pblock(ext);
364 es->s_last_error_block = cpu_to_le64(pblock);
369 prev = lblock + len - 1;
372 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
374 if (!ext4_valid_extent_idx(inode, ext_idx))
383 static int __ext4_ext_check(const char *function, unsigned int line,
384 struct inode *inode, struct ext4_extent_header *eh,
387 const char *error_msg;
390 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
391 error_msg = "invalid magic";
394 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
395 error_msg = "unexpected eh_depth";
398 if (unlikely(eh->eh_max == 0)) {
399 error_msg = "invalid eh_max";
402 max = ext4_ext_max_entries(inode, depth);
403 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
404 error_msg = "too large eh_max";
407 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
408 error_msg = "invalid eh_entries";
411 if (!ext4_valid_extent_entries(inode, eh, depth)) {
412 error_msg = "invalid extent entries";
418 ext4_error_inode(inode, function, line, 0,
419 "bad header/extent: %s - magic %x, "
420 "entries %u, max %u(%u), depth %u(%u)",
421 error_msg, le16_to_cpu(eh->eh_magic),
422 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
423 max, le16_to_cpu(eh->eh_depth), depth);
428 #define ext4_ext_check(inode, eh, depth) \
429 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
431 int ext4_ext_check_inode(struct inode *inode)
433 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
437 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
439 int k, l = path->p_depth;
442 for (k = 0; k <= l; k++, path++) {
444 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
445 ext4_idx_pblock(path->p_idx));
446 } else if (path->p_ext) {
447 ext_debug(" %d:[%d]%d:%llu ",
448 le32_to_cpu(path->p_ext->ee_block),
449 ext4_ext_is_uninitialized(path->p_ext),
450 ext4_ext_get_actual_len(path->p_ext),
451 ext4_ext_pblock(path->p_ext));
458 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
460 int depth = ext_depth(inode);
461 struct ext4_extent_header *eh;
462 struct ext4_extent *ex;
468 eh = path[depth].p_hdr;
469 ex = EXT_FIRST_EXTENT(eh);
471 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
473 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
474 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
475 ext4_ext_is_uninitialized(ex),
476 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
481 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
482 ext4_fsblk_t newblock, int level)
484 int depth = ext_depth(inode);
485 struct ext4_extent *ex;
487 if (depth != level) {
488 struct ext4_extent_idx *idx;
489 idx = path[level].p_idx;
490 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
491 ext_debug("%d: move %d:%llu in new index %llu\n", level,
492 le32_to_cpu(idx->ei_block),
493 ext4_idx_pblock(idx),
501 ex = path[depth].p_ext;
502 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
503 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
504 le32_to_cpu(ex->ee_block),
506 ext4_ext_is_uninitialized(ex),
507 ext4_ext_get_actual_len(ex),
514 #define ext4_ext_show_path(inode, path)
515 #define ext4_ext_show_leaf(inode, path)
516 #define ext4_ext_show_move(inode, path, newblock, level)
519 void ext4_ext_drop_refs(struct ext4_ext_path *path)
521 int depth = path->p_depth;
524 for (i = 0; i <= depth; i++, path++)
532 * ext4_ext_binsearch_idx:
533 * binary search for the closest index of the given block
534 * the header must be checked before calling this
537 ext4_ext_binsearch_idx(struct inode *inode,
538 struct ext4_ext_path *path, ext4_lblk_t block)
540 struct ext4_extent_header *eh = path->p_hdr;
541 struct ext4_extent_idx *r, *l, *m;
544 ext_debug("binsearch for %u(idx): ", block);
546 l = EXT_FIRST_INDEX(eh) + 1;
547 r = EXT_LAST_INDEX(eh);
550 if (block < le32_to_cpu(m->ei_block))
554 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
555 m, le32_to_cpu(m->ei_block),
556 r, le32_to_cpu(r->ei_block));
560 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
561 ext4_idx_pblock(path->p_idx));
563 #ifdef CHECK_BINSEARCH
565 struct ext4_extent_idx *chix, *ix;
568 chix = ix = EXT_FIRST_INDEX(eh);
569 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
571 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
572 printk(KERN_DEBUG "k=%d, ix=0x%p, "
574 ix, EXT_FIRST_INDEX(eh));
575 printk(KERN_DEBUG "%u <= %u\n",
576 le32_to_cpu(ix->ei_block),
577 le32_to_cpu(ix[-1].ei_block));
579 BUG_ON(k && le32_to_cpu(ix->ei_block)
580 <= le32_to_cpu(ix[-1].ei_block));
581 if (block < le32_to_cpu(ix->ei_block))
585 BUG_ON(chix != path->p_idx);
592 * ext4_ext_binsearch:
593 * binary search for closest extent of the given block
594 * the header must be checked before calling this
597 ext4_ext_binsearch(struct inode *inode,
598 struct ext4_ext_path *path, ext4_lblk_t block)
600 struct ext4_extent_header *eh = path->p_hdr;
601 struct ext4_extent *r, *l, *m;
603 if (eh->eh_entries == 0) {
605 * this leaf is empty:
606 * we get such a leaf in split/add case
611 ext_debug("binsearch for %u: ", block);
613 l = EXT_FIRST_EXTENT(eh) + 1;
614 r = EXT_LAST_EXTENT(eh);
618 if (block < le32_to_cpu(m->ee_block))
622 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
623 m, le32_to_cpu(m->ee_block),
624 r, le32_to_cpu(r->ee_block));
628 ext_debug(" -> %d:%llu:[%d]%d ",
629 le32_to_cpu(path->p_ext->ee_block),
630 ext4_ext_pblock(path->p_ext),
631 ext4_ext_is_uninitialized(path->p_ext),
632 ext4_ext_get_actual_len(path->p_ext));
634 #ifdef CHECK_BINSEARCH
636 struct ext4_extent *chex, *ex;
639 chex = ex = EXT_FIRST_EXTENT(eh);
640 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
641 BUG_ON(k && le32_to_cpu(ex->ee_block)
642 <= le32_to_cpu(ex[-1].ee_block));
643 if (block < le32_to_cpu(ex->ee_block))
647 BUG_ON(chex != path->p_ext);
653 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
655 struct ext4_extent_header *eh;
657 eh = ext_inode_hdr(inode);
660 eh->eh_magic = EXT4_EXT_MAGIC;
661 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
662 ext4_mark_inode_dirty(handle, inode);
663 ext4_ext_invalidate_cache(inode);
667 struct ext4_ext_path *
668 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
669 struct ext4_ext_path *path)
671 struct ext4_extent_header *eh;
672 struct buffer_head *bh;
673 short int depth, i, ppos = 0, alloc = 0;
676 eh = ext_inode_hdr(inode);
677 depth = ext_depth(inode);
679 /* account possible depth increase */
681 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
684 return ERR_PTR(-ENOMEM);
691 /* walk through the tree */
693 int need_to_validate = 0;
695 ext_debug("depth %d: num %d, max %d\n",
696 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
698 ext4_ext_binsearch_idx(inode, path + ppos, block);
699 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
700 path[ppos].p_depth = i;
701 path[ppos].p_ext = NULL;
703 bh = sb_getblk_gfp(inode->i_sb, path[ppos].p_block,
704 __GFP_MOVABLE | GFP_NOFS);
709 if (!bh_uptodate_or_lock(bh)) {
710 trace_ext4_ext_load_extent(inode, block,
712 ret = bh_submit_read(bh);
717 /* validate the extent entries */
718 need_to_validate = 1;
720 eh = ext_block_hdr(bh);
722 if (unlikely(ppos > depth)) {
724 EXT4_ERROR_INODE(inode,
725 "ppos %d > depth %d", ppos, depth);
729 path[ppos].p_bh = bh;
730 path[ppos].p_hdr = eh;
733 ret = need_to_validate ? ext4_ext_check(inode, eh, i) : 0;
738 path[ppos].p_depth = i;
739 path[ppos].p_ext = NULL;
740 path[ppos].p_idx = NULL;
743 ext4_ext_binsearch(inode, path + ppos, block);
744 /* if not an empty leaf */
745 if (path[ppos].p_ext)
746 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
748 ext4_ext_show_path(inode, path);
753 ext4_ext_drop_refs(path);
760 * ext4_ext_insert_index:
761 * insert new index [@logical;@ptr] into the block at @curp;
762 * check where to insert: before @curp or after @curp
764 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
765 struct ext4_ext_path *curp,
766 int logical, ext4_fsblk_t ptr)
768 struct ext4_extent_idx *ix;
771 err = ext4_ext_get_access(handle, inode, curp);
775 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
776 EXT4_ERROR_INODE(inode,
777 "logical %d == ei_block %d!",
778 logical, le32_to_cpu(curp->p_idx->ei_block));
782 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
783 >= le16_to_cpu(curp->p_hdr->eh_max))) {
784 EXT4_ERROR_INODE(inode,
785 "eh_entries %d >= eh_max %d!",
786 le16_to_cpu(curp->p_hdr->eh_entries),
787 le16_to_cpu(curp->p_hdr->eh_max));
791 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
793 ext_debug("insert new index %d after: %llu\n", logical, ptr);
794 ix = curp->p_idx + 1;
797 ext_debug("insert new index %d before: %llu\n", logical, ptr);
801 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
804 ext_debug("insert new index %d: "
805 "move %d indices from 0x%p to 0x%p\n",
806 logical, len, ix, ix + 1);
807 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
810 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
811 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
815 ix->ei_block = cpu_to_le32(logical);
816 ext4_idx_store_pblock(ix, ptr);
817 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
819 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
820 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
824 err = ext4_ext_dirty(handle, inode, curp);
825 ext4_std_error(inode->i_sb, err);
832 * inserts new subtree into the path, using free index entry
834 * - allocates all needed blocks (new leaf and all intermediate index blocks)
835 * - makes decision where to split
836 * - moves remaining extents and index entries (right to the split point)
837 * into the newly allocated blocks
838 * - initializes subtree
840 static int ext4_ext_split(handle_t *handle, struct inode *inode,
842 struct ext4_ext_path *path,
843 struct ext4_extent *newext, int at)
845 struct buffer_head *bh = NULL;
846 int depth = ext_depth(inode);
847 struct ext4_extent_header *neh;
848 struct ext4_extent_idx *fidx;
850 ext4_fsblk_t newblock, oldblock;
852 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
855 /* make decision: where to split? */
856 /* FIXME: now decision is simplest: at current extent */
858 /* if current leaf will be split, then we should use
859 * border from split point */
860 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
861 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
864 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
865 border = path[depth].p_ext[1].ee_block;
866 ext_debug("leaf will be split."
867 " next leaf starts at %d\n",
868 le32_to_cpu(border));
870 border = newext->ee_block;
871 ext_debug("leaf will be added."
872 " next leaf starts at %d\n",
873 le32_to_cpu(border));
877 * If error occurs, then we break processing
878 * and mark filesystem read-only. index won't
879 * be inserted and tree will be in consistent
880 * state. Next mount will repair buffers too.
884 * Get array to track all allocated blocks.
885 * We need this to handle errors and free blocks
888 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
892 /* allocate all needed blocks */
893 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
894 for (a = 0; a < depth - at; a++) {
895 newblock = ext4_ext_new_meta_block(handle, inode, path,
896 newext, &err, flags);
899 ablocks[a] = newblock;
902 /* initialize new leaf */
903 newblock = ablocks[--a];
904 if (unlikely(newblock == 0)) {
905 EXT4_ERROR_INODE(inode, "newblock == 0!");
909 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
916 err = ext4_journal_get_create_access(handle, bh);
920 neh = ext_block_hdr(bh);
922 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
923 neh->eh_magic = EXT4_EXT_MAGIC;
926 /* move remainder of path[depth] to the new leaf */
927 if (unlikely(path[depth].p_hdr->eh_entries !=
928 path[depth].p_hdr->eh_max)) {
929 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
930 path[depth].p_hdr->eh_entries,
931 path[depth].p_hdr->eh_max);
935 /* start copy from next extent */
936 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
937 ext4_ext_show_move(inode, path, newblock, depth);
939 struct ext4_extent *ex;
940 ex = EXT_FIRST_EXTENT(neh);
941 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
942 le16_add_cpu(&neh->eh_entries, m);
945 set_buffer_uptodate(bh);
948 err = ext4_handle_dirty_metadata(handle, inode, bh);
954 /* correct old leaf */
956 err = ext4_ext_get_access(handle, inode, path + depth);
959 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
960 err = ext4_ext_dirty(handle, inode, path + depth);
966 /* create intermediate indexes */
968 if (unlikely(k < 0)) {
969 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
974 ext_debug("create %d intermediate indices\n", k);
975 /* insert new index into current index block */
976 /* current depth stored in i var */
980 newblock = ablocks[--a];
981 bh = sb_getblk(inode->i_sb, newblock);
988 err = ext4_journal_get_create_access(handle, bh);
992 neh = ext_block_hdr(bh);
993 neh->eh_entries = cpu_to_le16(1);
994 neh->eh_magic = EXT4_EXT_MAGIC;
995 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
996 neh->eh_depth = cpu_to_le16(depth - i);
997 fidx = EXT_FIRST_INDEX(neh);
998 fidx->ei_block = border;
999 ext4_idx_store_pblock(fidx, oldblock);
1001 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1002 i, newblock, le32_to_cpu(border), oldblock);
1004 /* move remainder of path[i] to the new index block */
1005 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1006 EXT_LAST_INDEX(path[i].p_hdr))) {
1007 EXT4_ERROR_INODE(inode,
1008 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1009 le32_to_cpu(path[i].p_ext->ee_block));
1013 /* start copy indexes */
1014 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1015 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1016 EXT_MAX_INDEX(path[i].p_hdr));
1017 ext4_ext_show_move(inode, path, newblock, i);
1019 memmove(++fidx, path[i].p_idx,
1020 sizeof(struct ext4_extent_idx) * m);
1021 le16_add_cpu(&neh->eh_entries, m);
1023 set_buffer_uptodate(bh);
1026 err = ext4_handle_dirty_metadata(handle, inode, bh);
1032 /* correct old index */
1034 err = ext4_ext_get_access(handle, inode, path + i);
1037 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1038 err = ext4_ext_dirty(handle, inode, path + i);
1046 /* insert new index */
1047 err = ext4_ext_insert_index(handle, inode, path + at,
1048 le32_to_cpu(border), newblock);
1052 if (buffer_locked(bh))
1058 /* free all allocated blocks in error case */
1059 for (i = 0; i < depth; i++) {
1062 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1063 EXT4_FREE_BLOCKS_METADATA);
1072 * ext4_ext_grow_indepth:
1073 * implements tree growing procedure:
1074 * - allocates new block
1075 * - moves top-level data (index block or leaf) into the new block
1076 * - initializes new top-level, creating index that points to the
1077 * just created block
1079 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1081 struct ext4_extent *newext)
1083 struct ext4_extent_header *neh;
1084 struct buffer_head *bh;
1085 ext4_fsblk_t newblock;
1088 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1089 newext, &err, flags);
1093 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1098 err = ext4_journal_get_create_access(handle, bh);
1104 /* move top-level index/leaf into new block */
1105 memmove(bh->b_data, EXT4_I(inode)->i_data,
1106 sizeof(EXT4_I(inode)->i_data));
1108 /* set size of new block */
1109 neh = ext_block_hdr(bh);
1110 /* old root could have indexes or leaves
1111 * so calculate e_max right way */
1112 if (ext_depth(inode))
1113 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1115 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1116 neh->eh_magic = EXT4_EXT_MAGIC;
1117 set_buffer_uptodate(bh);
1120 err = ext4_handle_dirty_metadata(handle, inode, bh);
1124 /* Update top-level index: num,max,pointer */
1125 neh = ext_inode_hdr(inode);
1126 neh->eh_entries = cpu_to_le16(1);
1127 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1128 if (neh->eh_depth == 0) {
1129 /* Root extent block becomes index block */
1130 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1131 EXT_FIRST_INDEX(neh)->ei_block =
1132 EXT_FIRST_EXTENT(neh)->ee_block;
1134 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1135 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1136 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1137 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1139 neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
1140 ext4_mark_inode_dirty(handle, inode);
1148 * ext4_ext_create_new_leaf:
1149 * finds empty index and adds new leaf.
1150 * if no free index is found, then it requests in-depth growing.
1152 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1154 struct ext4_ext_path *path,
1155 struct ext4_extent *newext)
1157 struct ext4_ext_path *curp;
1158 int depth, i, err = 0;
1161 i = depth = ext_depth(inode);
1163 /* walk up to the tree and look for free index entry */
1164 curp = path + depth;
1165 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1170 /* we use already allocated block for index block,
1171 * so subsequent data blocks should be contiguous */
1172 if (EXT_HAS_FREE_INDEX(curp)) {
1173 /* if we found index with free entry, then use that
1174 * entry: create all needed subtree and add new leaf */
1175 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1180 ext4_ext_drop_refs(path);
1181 path = ext4_ext_find_extent(inode,
1182 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1185 err = PTR_ERR(path);
1187 /* tree is full, time to grow in depth */
1188 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1193 ext4_ext_drop_refs(path);
1194 path = ext4_ext_find_extent(inode,
1195 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1198 err = PTR_ERR(path);
1203 * only first (depth 0 -> 1) produces free space;
1204 * in all other cases we have to split the grown tree
1206 depth = ext_depth(inode);
1207 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1208 /* now we need to split */
1218 * search the closest allocated block to the left for *logical
1219 * and returns it at @logical + it's physical address at @phys
1220 * if *logical is the smallest allocated block, the function
1221 * returns 0 at @phys
1222 * return value contains 0 (success) or error code
1224 static int ext4_ext_search_left(struct inode *inode,
1225 struct ext4_ext_path *path,
1226 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1228 struct ext4_extent_idx *ix;
1229 struct ext4_extent *ex;
1232 if (unlikely(path == NULL)) {
1233 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1236 depth = path->p_depth;
1239 if (depth == 0 && path->p_ext == NULL)
1242 /* usually extent in the path covers blocks smaller
1243 * then *logical, but it can be that extent is the
1244 * first one in the file */
1246 ex = path[depth].p_ext;
1247 ee_len = ext4_ext_get_actual_len(ex);
1248 if (*logical < le32_to_cpu(ex->ee_block)) {
1249 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1250 EXT4_ERROR_INODE(inode,
1251 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1252 *logical, le32_to_cpu(ex->ee_block));
1255 while (--depth >= 0) {
1256 ix = path[depth].p_idx;
1257 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1258 EXT4_ERROR_INODE(inode,
1259 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1260 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1261 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1262 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1270 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1271 EXT4_ERROR_INODE(inode,
1272 "logical %d < ee_block %d + ee_len %d!",
1273 *logical, le32_to_cpu(ex->ee_block), ee_len);
1277 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1278 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1283 * search the closest allocated block to the right for *logical
1284 * and returns it at @logical + it's physical address at @phys
1285 * if *logical is the largest allocated block, the function
1286 * returns 0 at @phys
1287 * return value contains 0 (success) or error code
1289 static int ext4_ext_search_right(struct inode *inode,
1290 struct ext4_ext_path *path,
1291 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1292 struct ext4_extent **ret_ex)
1294 struct buffer_head *bh = NULL;
1295 struct ext4_extent_header *eh;
1296 struct ext4_extent_idx *ix;
1297 struct ext4_extent *ex;
1299 int depth; /* Note, NOT eh_depth; depth from top of tree */
1302 if (unlikely(path == NULL)) {
1303 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1306 depth = path->p_depth;
1309 if (depth == 0 && path->p_ext == NULL)
1312 /* usually extent in the path covers blocks smaller
1313 * then *logical, but it can be that extent is the
1314 * first one in the file */
1316 ex = path[depth].p_ext;
1317 ee_len = ext4_ext_get_actual_len(ex);
1318 if (*logical < le32_to_cpu(ex->ee_block)) {
1319 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1320 EXT4_ERROR_INODE(inode,
1321 "first_extent(path[%d].p_hdr) != ex",
1325 while (--depth >= 0) {
1326 ix = path[depth].p_idx;
1327 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1328 EXT4_ERROR_INODE(inode,
1329 "ix != EXT_FIRST_INDEX *logical %d!",
1337 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1338 EXT4_ERROR_INODE(inode,
1339 "logical %d < ee_block %d + ee_len %d!",
1340 *logical, le32_to_cpu(ex->ee_block), ee_len);
1344 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1345 /* next allocated block in this leaf */
1350 /* go up and search for index to the right */
1351 while (--depth >= 0) {
1352 ix = path[depth].p_idx;
1353 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1357 /* we've gone up to the root and found no index to the right */
1361 /* we've found index to the right, let's
1362 * follow it and find the closest allocated
1363 * block to the right */
1365 block = ext4_idx_pblock(ix);
1366 while (++depth < path->p_depth) {
1367 bh = sb_bread(inode->i_sb, block);
1370 eh = ext_block_hdr(bh);
1371 /* subtract from p_depth to get proper eh_depth */
1372 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1376 ix = EXT_FIRST_INDEX(eh);
1377 block = ext4_idx_pblock(ix);
1381 bh = sb_bread(inode->i_sb, block);
1384 eh = ext_block_hdr(bh);
1385 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1389 ex = EXT_FIRST_EXTENT(eh);
1391 *logical = le32_to_cpu(ex->ee_block);
1392 *phys = ext4_ext_pblock(ex);
1400 * ext4_ext_next_allocated_block:
1401 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1402 * NOTE: it considers block number from index entry as
1403 * allocated block. Thus, index entries have to be consistent
1407 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1411 BUG_ON(path == NULL);
1412 depth = path->p_depth;
1414 if (depth == 0 && path->p_ext == NULL)
1415 return EXT_MAX_BLOCKS;
1417 while (depth >= 0) {
1418 if (depth == path->p_depth) {
1420 if (path[depth].p_ext &&
1421 path[depth].p_ext !=
1422 EXT_LAST_EXTENT(path[depth].p_hdr))
1423 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1426 if (path[depth].p_idx !=
1427 EXT_LAST_INDEX(path[depth].p_hdr))
1428 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1433 return EXT_MAX_BLOCKS;
1437 * ext4_ext_next_leaf_block:
1438 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1440 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1444 BUG_ON(path == NULL);
1445 depth = path->p_depth;
1447 /* zero-tree has no leaf blocks at all */
1449 return EXT_MAX_BLOCKS;
1451 /* go to index block */
1454 while (depth >= 0) {
1455 if (path[depth].p_idx !=
1456 EXT_LAST_INDEX(path[depth].p_hdr))
1457 return (ext4_lblk_t)
1458 le32_to_cpu(path[depth].p_idx[1].ei_block);
1462 return EXT_MAX_BLOCKS;
1466 * ext4_ext_correct_indexes:
1467 * if leaf gets modified and modified extent is first in the leaf,
1468 * then we have to correct all indexes above.
1469 * TODO: do we need to correct tree in all cases?
1471 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1472 struct ext4_ext_path *path)
1474 struct ext4_extent_header *eh;
1475 int depth = ext_depth(inode);
1476 struct ext4_extent *ex;
1480 eh = path[depth].p_hdr;
1481 ex = path[depth].p_ext;
1483 if (unlikely(ex == NULL || eh == NULL)) {
1484 EXT4_ERROR_INODE(inode,
1485 "ex %p == NULL or eh %p == NULL", ex, eh);
1490 /* there is no tree at all */
1494 if (ex != EXT_FIRST_EXTENT(eh)) {
1495 /* we correct tree if first leaf got modified only */
1500 * TODO: we need correction if border is smaller than current one
1503 border = path[depth].p_ext->ee_block;
1504 err = ext4_ext_get_access(handle, inode, path + k);
1507 path[k].p_idx->ei_block = border;
1508 err = ext4_ext_dirty(handle, inode, path + k);
1513 /* change all left-side indexes */
1514 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1516 err = ext4_ext_get_access(handle, inode, path + k);
1519 path[k].p_idx->ei_block = border;
1520 err = ext4_ext_dirty(handle, inode, path + k);
1529 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1530 struct ext4_extent *ex2)
1532 unsigned short ext1_ee_len, ext2_ee_len, max_len;
1535 * Make sure that either both extents are uninitialized, or
1538 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1541 if (ext4_ext_is_uninitialized(ex1))
1542 max_len = EXT_UNINIT_MAX_LEN;
1544 max_len = EXT_INIT_MAX_LEN;
1546 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1547 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1549 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1550 le32_to_cpu(ex2->ee_block))
1554 * To allow future support for preallocated extents to be added
1555 * as an RO_COMPAT feature, refuse to merge to extents if
1556 * this can result in the top bit of ee_len being set.
1558 if (ext1_ee_len + ext2_ee_len > max_len)
1560 #ifdef AGGRESSIVE_TEST
1561 if (ext1_ee_len >= 4)
1565 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1571 * This function tries to merge the "ex" extent to the next extent in the tree.
1572 * It always tries to merge towards right. If you want to merge towards
1573 * left, pass "ex - 1" as argument instead of "ex".
1574 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1575 * 1 if they got merged.
1577 static int ext4_ext_try_to_merge_right(struct inode *inode,
1578 struct ext4_ext_path *path,
1579 struct ext4_extent *ex)
1581 struct ext4_extent_header *eh;
1582 unsigned int depth, len;
1584 int uninitialized = 0;
1586 depth = ext_depth(inode);
1587 BUG_ON(path[depth].p_hdr == NULL);
1588 eh = path[depth].p_hdr;
1590 while (ex < EXT_LAST_EXTENT(eh)) {
1591 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1593 /* merge with next extent! */
1594 if (ext4_ext_is_uninitialized(ex))
1596 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1597 + ext4_ext_get_actual_len(ex + 1));
1599 ext4_ext_mark_uninitialized(ex);
1601 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1602 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1603 * sizeof(struct ext4_extent);
1604 memmove(ex + 1, ex + 2, len);
1606 le16_add_cpu(&eh->eh_entries, -1);
1608 WARN_ON(eh->eh_entries == 0);
1609 if (!eh->eh_entries)
1610 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1617 * This function tries to merge the @ex extent to neighbours in the tree.
1618 * return 1 if merge left else 0.
1620 static int ext4_ext_try_to_merge(struct inode *inode,
1621 struct ext4_ext_path *path,
1622 struct ext4_extent *ex) {
1623 struct ext4_extent_header *eh;
1628 depth = ext_depth(inode);
1629 BUG_ON(path[depth].p_hdr == NULL);
1630 eh = path[depth].p_hdr;
1632 if (ex > EXT_FIRST_EXTENT(eh))
1633 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1636 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1642 * check if a portion of the "newext" extent overlaps with an
1645 * If there is an overlap discovered, it updates the length of the newext
1646 * such that there will be no overlap, and then returns 1.
1647 * If there is no overlap found, it returns 0.
1649 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1650 struct inode *inode,
1651 struct ext4_extent *newext,
1652 struct ext4_ext_path *path)
1655 unsigned int depth, len1;
1656 unsigned int ret = 0;
1658 b1 = le32_to_cpu(newext->ee_block);
1659 len1 = ext4_ext_get_actual_len(newext);
1660 depth = ext_depth(inode);
1661 if (!path[depth].p_ext)
1663 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1666 * get the next allocated block if the extent in the path
1667 * is before the requested block(s)
1670 b2 = ext4_ext_next_allocated_block(path);
1671 if (b2 == EXT_MAX_BLOCKS)
1673 b2 = EXT4_LBLK_CMASK(sbi, b2);
1676 /* check for wrap through zero on extent logical start block*/
1677 if (b1 + len1 < b1) {
1678 len1 = EXT_MAX_BLOCKS - b1;
1679 newext->ee_len = cpu_to_le16(len1);
1683 /* check for overlap */
1684 if (b1 + len1 > b2) {
1685 newext->ee_len = cpu_to_le16(b2 - b1);
1693 * ext4_ext_insert_extent:
1694 * tries to merge requsted extent into the existing extent or
1695 * inserts requested extent as new one into the tree,
1696 * creating new leaf in the no-space case.
1698 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1699 struct ext4_ext_path *path,
1700 struct ext4_extent *newext, int flag)
1702 struct ext4_extent_header *eh;
1703 struct ext4_extent *ex, *fex;
1704 struct ext4_extent *nearex; /* nearest extent */
1705 struct ext4_ext_path *npath = NULL;
1706 int depth, len, err;
1708 unsigned uninitialized = 0;
1711 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1712 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1715 depth = ext_depth(inode);
1716 ex = path[depth].p_ext;
1717 if (unlikely(path[depth].p_hdr == NULL)) {
1718 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1722 /* try to insert block into found extent and return */
1723 if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1724 && ext4_can_extents_be_merged(inode, ex, newext)) {
1725 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1726 ext4_ext_is_uninitialized(newext),
1727 ext4_ext_get_actual_len(newext),
1728 le32_to_cpu(ex->ee_block),
1729 ext4_ext_is_uninitialized(ex),
1730 ext4_ext_get_actual_len(ex),
1731 ext4_ext_pblock(ex));
1732 err = ext4_ext_get_access(handle, inode, path + depth);
1737 * ext4_can_extents_be_merged should have checked that either
1738 * both extents are uninitialized, or both aren't. Thus we
1739 * need to check only one of them here.
1741 if (ext4_ext_is_uninitialized(ex))
1743 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1744 + ext4_ext_get_actual_len(newext));
1746 ext4_ext_mark_uninitialized(ex);
1747 eh = path[depth].p_hdr;
1752 depth = ext_depth(inode);
1753 eh = path[depth].p_hdr;
1754 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1757 /* probably next leaf has space for us? */
1758 fex = EXT_LAST_EXTENT(eh);
1759 next = EXT_MAX_BLOCKS;
1760 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1761 next = ext4_ext_next_leaf_block(path);
1762 if (next != EXT_MAX_BLOCKS) {
1763 ext_debug("next leaf block - %u\n", next);
1764 BUG_ON(npath != NULL);
1765 npath = ext4_ext_find_extent(inode, next, NULL);
1767 return PTR_ERR(npath);
1768 BUG_ON(npath->p_depth != path->p_depth);
1769 eh = npath[depth].p_hdr;
1770 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1771 ext_debug("next leaf isn't full(%d)\n",
1772 le16_to_cpu(eh->eh_entries));
1776 ext_debug("next leaf has no free space(%d,%d)\n",
1777 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1781 * There is no free space in the found leaf.
1782 * We're gonna add a new leaf in the tree.
1784 if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1785 flags = EXT4_MB_USE_ROOT_BLOCKS;
1786 err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1789 depth = ext_depth(inode);
1790 eh = path[depth].p_hdr;
1793 nearex = path[depth].p_ext;
1795 err = ext4_ext_get_access(handle, inode, path + depth);
1800 /* there is no extent in this leaf, create first one */
1801 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1802 le32_to_cpu(newext->ee_block),
1803 ext4_ext_pblock(newext),
1804 ext4_ext_is_uninitialized(newext),
1805 ext4_ext_get_actual_len(newext));
1806 nearex = EXT_FIRST_EXTENT(eh);
1808 if (le32_to_cpu(newext->ee_block)
1809 > le32_to_cpu(nearex->ee_block)) {
1811 ext_debug("insert %u:%llu:[%d]%d before: "
1813 le32_to_cpu(newext->ee_block),
1814 ext4_ext_pblock(newext),
1815 ext4_ext_is_uninitialized(newext),
1816 ext4_ext_get_actual_len(newext),
1821 BUG_ON(newext->ee_block == nearex->ee_block);
1822 ext_debug("insert %u:%llu:[%d]%d after: "
1824 le32_to_cpu(newext->ee_block),
1825 ext4_ext_pblock(newext),
1826 ext4_ext_is_uninitialized(newext),
1827 ext4_ext_get_actual_len(newext),
1830 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1832 ext_debug("insert %u:%llu:[%d]%d: "
1833 "move %d extents from 0x%p to 0x%p\n",
1834 le32_to_cpu(newext->ee_block),
1835 ext4_ext_pblock(newext),
1836 ext4_ext_is_uninitialized(newext),
1837 ext4_ext_get_actual_len(newext),
1838 len, nearex, nearex + 1);
1839 memmove(nearex + 1, nearex,
1840 len * sizeof(struct ext4_extent));
1844 le16_add_cpu(&eh->eh_entries, 1);
1845 path[depth].p_ext = nearex;
1846 nearex->ee_block = newext->ee_block;
1847 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1848 nearex->ee_len = newext->ee_len;
1851 /* try to merge extents to the right */
1852 if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1853 ext4_ext_try_to_merge(inode, path, nearex);
1855 /* try to merge extents to the left */
1857 /* time to correct all indexes above */
1858 err = ext4_ext_correct_indexes(handle, inode, path);
1862 err = ext4_ext_dirty(handle, inode, path + depth);
1866 ext4_ext_drop_refs(npath);
1869 ext4_ext_invalidate_cache(inode);
1873 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1874 ext4_lblk_t num, ext_prepare_callback func,
1877 struct ext4_ext_path *path = NULL;
1878 struct ext4_ext_cache cbex;
1879 struct ext4_extent *ex;
1880 ext4_lblk_t next, start = 0, end = 0;
1881 ext4_lblk_t last = block + num;
1882 int depth, exists, err = 0;
1884 BUG_ON(func == NULL);
1885 BUG_ON(inode == NULL);
1887 while (block < last && block != EXT_MAX_BLOCKS) {
1889 /* find extent for this block */
1890 down_read(&EXT4_I(inode)->i_data_sem);
1891 path = ext4_ext_find_extent(inode, block, path);
1892 up_read(&EXT4_I(inode)->i_data_sem);
1894 err = PTR_ERR(path);
1899 depth = ext_depth(inode);
1900 if (unlikely(path[depth].p_hdr == NULL)) {
1901 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1905 ex = path[depth].p_ext;
1906 next = ext4_ext_next_allocated_block(path);
1910 /* there is no extent yet, so try to allocate
1911 * all requested space */
1914 } else if (le32_to_cpu(ex->ee_block) > block) {
1915 /* need to allocate space before found extent */
1917 end = le32_to_cpu(ex->ee_block);
1918 if (block + num < end)
1920 } else if (block >= le32_to_cpu(ex->ee_block)
1921 + ext4_ext_get_actual_len(ex)) {
1922 /* need to allocate space after found extent */
1927 } else if (block >= le32_to_cpu(ex->ee_block)) {
1929 * some part of requested space is covered
1933 end = le32_to_cpu(ex->ee_block)
1934 + ext4_ext_get_actual_len(ex);
1935 if (block + num < end)
1941 BUG_ON(end <= start);
1944 cbex.ec_block = start;
1945 cbex.ec_len = end - start;
1948 cbex.ec_block = le32_to_cpu(ex->ee_block);
1949 cbex.ec_len = ext4_ext_get_actual_len(ex);
1950 cbex.ec_start = ext4_ext_pblock(ex);
1953 if (unlikely(cbex.ec_len == 0)) {
1954 EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1958 err = func(inode, next, &cbex, ex, cbdata);
1959 ext4_ext_drop_refs(path);
1964 if (err == EXT_REPEAT)
1966 else if (err == EXT_BREAK) {
1971 if (ext_depth(inode) != depth) {
1972 /* depth was changed. we have to realloc path */
1977 block = cbex.ec_block + cbex.ec_len;
1981 ext4_ext_drop_refs(path);
1989 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1990 __u32 len, ext4_fsblk_t start)
1992 struct ext4_ext_cache *cex;
1994 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1995 trace_ext4_ext_put_in_cache(inode, block, len, start);
1996 cex = &EXT4_I(inode)->i_cached_extent;
1997 cex->ec_block = block;
1999 cex->ec_start = start;
2000 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2004 * ext4_ext_put_gap_in_cache:
2005 * calculate boundaries of the gap that the requested block fits into
2006 * and cache this gap
2009 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2012 int depth = ext_depth(inode);
2015 struct ext4_extent *ex;
2017 ex = path[depth].p_ext;
2019 /* there is no extent yet, so gap is [0;-] */
2021 len = EXT_MAX_BLOCKS;
2022 ext_debug("cache gap(whole file):");
2023 } else if (block < le32_to_cpu(ex->ee_block)) {
2025 len = le32_to_cpu(ex->ee_block) - block;
2026 ext_debug("cache gap(before): %u [%u:%u]",
2028 le32_to_cpu(ex->ee_block),
2029 ext4_ext_get_actual_len(ex));
2030 } else if (block >= le32_to_cpu(ex->ee_block)
2031 + ext4_ext_get_actual_len(ex)) {
2033 lblock = le32_to_cpu(ex->ee_block)
2034 + ext4_ext_get_actual_len(ex);
2036 next = ext4_ext_next_allocated_block(path);
2037 ext_debug("cache gap(after): [%u:%u] %u",
2038 le32_to_cpu(ex->ee_block),
2039 ext4_ext_get_actual_len(ex),
2041 BUG_ON(next == lblock);
2042 len = next - lblock;
2048 ext_debug(" -> %u:%lu\n", lblock, len);
2049 ext4_ext_put_in_cache(inode, lblock, len, 0);
2053 * ext4_ext_check_cache()
2054 * Checks to see if the given block is in the cache.
2055 * If it is, the cached extent is stored in the given
2056 * cache extent pointer. If the cached extent is a hole,
2057 * this routine should be used instead of
2058 * ext4_ext_in_cache if the calling function needs to
2059 * know the size of the hole.
2061 * @inode: The files inode
2062 * @block: The block to look for in the cache
2063 * @ex: Pointer where the cached extent will be stored
2064 * if it contains block
2066 * Return 0 if cache is invalid; 1 if the cache is valid
2068 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2069 struct ext4_ext_cache *ex){
2070 struct ext4_ext_cache *cex;
2071 struct ext4_sb_info *sbi;
2075 * We borrow i_block_reservation_lock to protect i_cached_extent
2077 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2078 cex = &EXT4_I(inode)->i_cached_extent;
2079 sbi = EXT4_SB(inode->i_sb);
2081 /* has cache valid data? */
2082 if (cex->ec_len == 0)
2085 if (in_range(block, cex->ec_block, cex->ec_len)) {
2086 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2087 ext_debug("%u cached by %u:%u:%llu\n",
2089 cex->ec_block, cex->ec_len, cex->ec_start);
2093 trace_ext4_ext_in_cache(inode, block, ret);
2094 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2099 * ext4_ext_in_cache()
2100 * Checks to see if the given block is in the cache.
2101 * If it is, the cached extent is stored in the given
2104 * @inode: The files inode
2105 * @block: The block to look for in the cache
2106 * @ex: Pointer where the cached extent will be stored
2107 * if it contains block
2109 * Return 0 if cache is invalid; 1 if the cache is valid
2112 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2113 struct ext4_extent *ex)
2115 struct ext4_ext_cache cex;
2118 if (ext4_ext_check_cache(inode, block, &cex)) {
2119 ex->ee_block = cpu_to_le32(cex.ec_block);
2120 ext4_ext_store_pblock(ex, cex.ec_start);
2121 ex->ee_len = cpu_to_le16(cex.ec_len);
2131 * removes index from the index block.
2133 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2134 struct ext4_ext_path *path, int depth)
2139 /* free index block */
2141 path = path + depth;
2142 leaf = ext4_idx_pblock(path->p_idx);
2143 if (unlikely(path->p_hdr->eh_entries == 0)) {
2144 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2147 err = ext4_ext_get_access(handle, inode, path);
2151 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2152 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2153 len *= sizeof(struct ext4_extent_idx);
2154 memmove(path->p_idx, path->p_idx + 1, len);
2157 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2158 err = ext4_ext_dirty(handle, inode, path);
2161 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2162 trace_ext4_ext_rm_idx(inode, leaf);
2164 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2165 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2167 while (--depth >= 0) {
2168 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2171 err = ext4_ext_get_access(handle, inode, path);
2174 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2175 err = ext4_ext_dirty(handle, inode, path);
2183 * ext4_ext_calc_credits_for_single_extent:
2184 * This routine returns max. credits that needed to insert an extent
2185 * to the extent tree.
2186 * When pass the actual path, the caller should calculate credits
2189 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2190 struct ext4_ext_path *path)
2193 int depth = ext_depth(inode);
2196 /* probably there is space in leaf? */
2197 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2198 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2201 * There are some space in the leaf tree, no
2202 * need to account for leaf block credit
2204 * bitmaps and block group descriptor blocks
2205 * and other metadata blocks still need to be
2208 /* 1 bitmap, 1 block group descriptor */
2209 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2214 return ext4_chunk_trans_blocks(inode, nrblocks);
2218 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2220 * if nrblocks are fit in a single extent (chunk flag is 1), then
2221 * in the worse case, each tree level index/leaf need to be changed
2222 * if the tree split due to insert a new extent, then the old tree
2223 * index/leaf need to be updated too
2225 * If the nrblocks are discontiguous, they could cause
2226 * the whole tree split more than once, but this is really rare.
2228 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2231 int depth = ext_depth(inode);
2241 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2242 struct ext4_extent *ex,
2243 ext4_fsblk_t *partial_cluster,
2244 ext4_lblk_t from, ext4_lblk_t to)
2246 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2247 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2249 int flags = EXT4_FREE_BLOCKS_FORGET;
2251 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2252 flags |= EXT4_FREE_BLOCKS_METADATA;
2254 * For bigalloc file systems, we never free a partial cluster
2255 * at the beginning of the extent. Instead, we make a note
2256 * that we tried freeing the cluster, and check to see if we
2257 * need to free it on a subsequent call to ext4_remove_blocks,
2258 * or at the end of the ext4_truncate() operation.
2260 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2262 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2264 * If we have a partial cluster, and it's different from the
2265 * cluster of the last block, we need to explicitly free the
2266 * partial cluster here.
2268 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2269 if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2270 ext4_free_blocks(handle, inode, NULL,
2271 EXT4_C2B(sbi, *partial_cluster),
2272 sbi->s_cluster_ratio, flags);
2273 *partial_cluster = 0;
2276 #ifdef EXTENTS_STATS
2278 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2279 spin_lock(&sbi->s_ext_stats_lock);
2280 sbi->s_ext_blocks += ee_len;
2281 sbi->s_ext_extents++;
2282 if (ee_len < sbi->s_ext_min)
2283 sbi->s_ext_min = ee_len;
2284 if (ee_len > sbi->s_ext_max)
2285 sbi->s_ext_max = ee_len;
2286 if (ext_depth(inode) > sbi->s_depth_max)
2287 sbi->s_depth_max = ext_depth(inode);
2288 spin_unlock(&sbi->s_ext_stats_lock);
2291 if (from >= le32_to_cpu(ex->ee_block)
2292 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2296 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2297 pblk = ext4_ext_pblock(ex) + ee_len - num;
2298 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2299 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2301 * If the block range to be freed didn't start at the
2302 * beginning of a cluster, and we removed the entire
2303 * extent, save the partial cluster here, since we
2304 * might need to delete if we determine that the
2305 * truncate operation has removed all of the blocks in
2308 if (EXT4_PBLK_COFF(sbi, pblk) &&
2310 *partial_cluster = EXT4_B2C(sbi, pblk);
2312 *partial_cluster = 0;
2313 } else if (from == le32_to_cpu(ex->ee_block)
2314 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2320 start = ext4_ext_pblock(ex);
2322 ext_debug("free first %u blocks starting %llu\n", num, start);
2323 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2326 printk(KERN_INFO "strange request: removal(2) "
2327 "%u-%u from %u:%u\n",
2328 from, to, le32_to_cpu(ex->ee_block), ee_len);
2335 * ext4_ext_rm_leaf() Removes the extents associated with the
2336 * blocks appearing between "start" and "end", and splits the extents
2337 * if "start" and "end" appear in the same extent
2339 * @handle: The journal handle
2340 * @inode: The files inode
2341 * @path: The path to the leaf
2342 * @start: The first block to remove
2343 * @end: The last block to remove
2346 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2347 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2348 ext4_lblk_t start, ext4_lblk_t end)
2350 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2351 int err = 0, correct_index = 0;
2352 int depth = ext_depth(inode), credits;
2353 struct ext4_extent_header *eh;
2356 ext4_lblk_t ex_ee_block;
2357 unsigned short ex_ee_len;
2358 unsigned uninitialized = 0;
2359 struct ext4_extent *ex;
2361 /* the header must be checked already in ext4_ext_remove_space() */
2362 ext_debug("truncate since %u in leaf to %u\n", start, end);
2363 if (!path[depth].p_hdr)
2364 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2365 eh = path[depth].p_hdr;
2366 if (unlikely(path[depth].p_hdr == NULL)) {
2367 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2370 /* find where to start removing */
2371 ex = EXT_LAST_EXTENT(eh);
2373 ex_ee_block = le32_to_cpu(ex->ee_block);
2374 ex_ee_len = ext4_ext_get_actual_len(ex);
2377 * If we're starting with an extent other than the last one in the
2378 * node, we need to see if it shares a cluster with the extent to
2379 * the right (towards the end of the file). If its leftmost cluster
2380 * is this extent's rightmost cluster and it is not cluster aligned,
2381 * we'll mark it as a partial that is not to be deallocated.
2384 if (ex != EXT_LAST_EXTENT(eh)) {
2385 ext4_fsblk_t current_pblk, right_pblk;
2386 long long current_cluster, right_cluster;
2388 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2389 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2390 right_pblk = ext4_ext_pblock(ex + 1);
2391 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2392 if (current_cluster == right_cluster &&
2393 EXT4_PBLK_COFF(sbi, right_pblk))
2394 *partial_cluster = -right_cluster;
2397 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2399 while (ex >= EXT_FIRST_EXTENT(eh) &&
2400 ex_ee_block + ex_ee_len > start) {
2402 if (ext4_ext_is_uninitialized(ex))
2407 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2408 uninitialized, ex_ee_len);
2409 path[depth].p_ext = ex;
2411 a = ex_ee_block > start ? ex_ee_block : start;
2412 b = ex_ee_block+ex_ee_len - 1 < end ?
2413 ex_ee_block+ex_ee_len - 1 : end;
2415 ext_debug(" border %u:%u\n", a, b);
2417 /* If this extent is beyond the end of the hole, skip it */
2418 if (end < ex_ee_block) {
2420 ex_ee_block = le32_to_cpu(ex->ee_block);
2421 ex_ee_len = ext4_ext_get_actual_len(ex);
2423 } else if (b != ex_ee_block + ex_ee_len - 1) {
2424 EXT4_ERROR_INODE(inode," bad truncate %u:%u\n",
2428 } else if (a != ex_ee_block) {
2429 /* remove tail of the extent */
2430 num = a - ex_ee_block;
2432 /* remove whole extent: excellent! */
2436 * 3 for leaf, sb, and inode plus 2 (bmap and group
2437 * descriptor) for each block group; assume two block
2438 * groups plus ex_ee_len/blocks_per_block_group for
2441 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2442 if (ex == EXT_FIRST_EXTENT(eh)) {
2444 credits += (ext_depth(inode)) + 1;
2446 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2448 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2452 err = ext4_ext_get_access(handle, inode, path + depth);
2456 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2462 /* this extent is removed; mark slot entirely unused */
2463 ext4_ext_store_pblock(ex, 0);
2465 ex->ee_len = cpu_to_le16(num);
2467 * Do not mark uninitialized if all the blocks in the
2468 * extent have been removed.
2470 if (uninitialized && num)
2471 ext4_ext_mark_uninitialized(ex);
2473 * If the extent was completely released,
2474 * we need to remove it from the leaf
2477 if (end != EXT_MAX_BLOCKS - 1) {
2479 * For hole punching, we need to scoot all the
2480 * extents up when an extent is removed so that
2481 * we dont have blank extents in the middle
2483 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2484 sizeof(struct ext4_extent));
2486 /* Now get rid of the one at the end */
2487 memset(EXT_LAST_EXTENT(eh), 0,
2488 sizeof(struct ext4_extent));
2490 le16_add_cpu(&eh->eh_entries, -1);
2492 *partial_cluster = 0;
2494 err = ext4_ext_dirty(handle, inode, path + depth);
2498 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2499 ext4_ext_pblock(ex));
2501 ex_ee_block = le32_to_cpu(ex->ee_block);
2502 ex_ee_len = ext4_ext_get_actual_len(ex);
2505 if (correct_index && eh->eh_entries)
2506 err = ext4_ext_correct_indexes(handle, inode, path);
2509 * If there is still a entry in the leaf node, check to see if
2510 * it references the partial cluster. This is the only place
2511 * where it could; if it doesn't, we can free the cluster.
2513 if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2514 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2515 *partial_cluster)) {
2516 int flags = EXT4_FREE_BLOCKS_FORGET;
2518 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2519 flags |= EXT4_FREE_BLOCKS_METADATA;
2521 ext4_free_blocks(handle, inode, NULL,
2522 EXT4_C2B(sbi, *partial_cluster),
2523 sbi->s_cluster_ratio, flags);
2524 *partial_cluster = 0;
2527 /* if this leaf is free, then we should
2528 * remove it from index block above */
2529 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2530 err = ext4_ext_rm_idx(handle, inode, path, depth);
2537 * ext4_ext_more_to_rm:
2538 * returns 1 if current index has to be freed (even partial)
2541 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2543 BUG_ON(path->p_idx == NULL);
2545 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2549 * if truncate on deeper level happened, it wasn't partial,
2550 * so we have to consider current index for truncation
2552 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2557 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2560 struct super_block *sb = inode->i_sb;
2561 int depth = ext_depth(inode);
2562 struct ext4_ext_path *path = NULL;
2563 ext4_fsblk_t partial_cluster = 0;
2567 ext_debug("truncate since %u to %u\n", start, end);
2569 /* probably first extent we're gonna free will be last in block */
2570 handle = ext4_journal_start(inode, depth + 1);
2572 return PTR_ERR(handle);
2575 ext4_ext_invalidate_cache(inode);
2577 trace_ext4_ext_remove_space(inode, start, depth);
2580 * Check if we are removing extents inside the extent tree. If that
2581 * is the case, we are going to punch a hole inside the extent tree
2582 * so we have to check whether we need to split the extent covering
2583 * the last block to remove so we can easily remove the part of it
2584 * in ext4_ext_rm_leaf().
2586 if (end < EXT_MAX_BLOCKS - 1) {
2587 struct ext4_extent *ex;
2588 ext4_lblk_t ee_block;
2590 /* find extent for this block */
2591 path = ext4_ext_find_extent(inode, end, NULL);
2593 ext4_journal_stop(handle);
2594 return PTR_ERR(path);
2596 depth = ext_depth(inode);
2597 ex = path[depth].p_ext;
2599 ext4_ext_drop_refs(path);
2605 ee_block = le32_to_cpu(ex->ee_block);
2608 * See if the last block is inside the extent, if so split
2609 * the extent at 'end' block so we can easily remove the
2610 * tail of the first part of the split extent in
2611 * ext4_ext_rm_leaf().
2613 if (end >= ee_block &&
2614 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2617 if (ext4_ext_is_uninitialized(ex))
2618 split_flag = EXT4_EXT_MARK_UNINIT1 |
2619 EXT4_EXT_MARK_UNINIT2;
2622 * Split the extent in two so that 'end' is the last
2623 * block in the first new extent
2625 err = ext4_split_extent_at(handle, inode, path,
2626 end + 1, split_flag,
2627 EXT4_GET_BLOCKS_PRE_IO |
2628 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2637 * We start scanning from right side, freeing all the blocks
2638 * after i_size and walking into the tree depth-wise.
2640 depth = ext_depth(inode);
2645 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2647 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2650 ext4_journal_stop(handle);
2653 path[0].p_depth = depth;
2654 path[0].p_hdr = ext_inode_hdr(inode);
2657 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2664 while (i >= 0 && err == 0) {
2666 /* this is leaf block */
2667 err = ext4_ext_rm_leaf(handle, inode, path,
2668 &partial_cluster, start,
2670 /* root level has p_bh == NULL, brelse() eats this */
2671 brelse(path[i].p_bh);
2672 path[i].p_bh = NULL;
2677 /* this is index block */
2678 if (!path[i].p_hdr) {
2679 ext_debug("initialize header\n");
2680 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2683 if (!path[i].p_idx) {
2684 /* this level hasn't been touched yet */
2685 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2686 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2687 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2689 le16_to_cpu(path[i].p_hdr->eh_entries));
2691 /* we were already here, see at next index */
2695 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2696 i, EXT_FIRST_INDEX(path[i].p_hdr),
2698 if (ext4_ext_more_to_rm(path + i)) {
2699 struct buffer_head *bh;
2700 /* go to the next level */
2701 ext_debug("move to level %d (block %llu)\n",
2702 i + 1, ext4_idx_pblock(path[i].p_idx));
2703 memset(path + i + 1, 0, sizeof(*path));
2704 bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2706 /* should we reset i_size? */
2710 if (WARN_ON(i + 1 > depth)) {
2714 if (ext4_ext_check(inode, ext_block_hdr(bh),
2719 path[i + 1].p_bh = bh;
2721 /* save actual number of indexes since this
2722 * number is changed at the next iteration */
2723 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2726 /* we finished processing this index, go up */
2727 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2728 /* index is empty, remove it;
2729 * handle must be already prepared by the
2730 * truncatei_leaf() */
2731 err = ext4_ext_rm_idx(handle, inode, path, i);
2733 /* root level has p_bh == NULL, brelse() eats this */
2734 brelse(path[i].p_bh);
2735 path[i].p_bh = NULL;
2737 ext_debug("return to level %d\n", i);
2741 trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2742 path->p_hdr->eh_entries);
2744 /* If we still have something in the partial cluster and we have removed
2745 * even the first extent, then we should free the blocks in the partial
2746 * cluster as well. */
2747 if (partial_cluster && path->p_hdr->eh_entries == 0) {
2748 int flags = EXT4_FREE_BLOCKS_FORGET;
2750 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2751 flags |= EXT4_FREE_BLOCKS_METADATA;
2753 ext4_free_blocks(handle, inode, NULL,
2754 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2755 EXT4_SB(sb)->s_cluster_ratio, flags);
2756 partial_cluster = 0;
2759 /* TODO: flexible tree reduction should be here */
2760 if (path->p_hdr->eh_entries == 0) {
2762 * truncate to zero freed all the tree,
2763 * so we need to correct eh_depth
2765 err = ext4_ext_get_access(handle, inode, path);
2767 ext_inode_hdr(inode)->eh_depth = 0;
2768 ext_inode_hdr(inode)->eh_max =
2769 cpu_to_le16(ext4_ext_space_root(inode, 0));
2770 err = ext4_ext_dirty(handle, inode, path);
2774 ext4_ext_drop_refs(path);
2776 if (err == -EAGAIN) {
2780 ext4_journal_stop(handle);
2786 * called at mount time
2788 void ext4_ext_init(struct super_block *sb)
2791 * possible initialization would be here
2794 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2795 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2796 printk(KERN_INFO "EXT4-fs: file extents enabled");
2797 #ifdef AGGRESSIVE_TEST
2798 printk(", aggressive tests");
2800 #ifdef CHECK_BINSEARCH
2801 printk(", check binsearch");
2803 #ifdef EXTENTS_STATS
2808 #ifdef EXTENTS_STATS
2809 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2810 EXT4_SB(sb)->s_ext_min = 1 << 30;
2811 EXT4_SB(sb)->s_ext_max = 0;
2817 * called at umount time
2819 void ext4_ext_release(struct super_block *sb)
2821 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2824 #ifdef EXTENTS_STATS
2825 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2826 struct ext4_sb_info *sbi = EXT4_SB(sb);
2827 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2828 sbi->s_ext_blocks, sbi->s_ext_extents,
2829 sbi->s_ext_blocks / sbi->s_ext_extents);
2830 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2831 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2836 /* FIXME!! we need to try to merge to left or right after zero-out */
2837 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2839 ext4_fsblk_t ee_pblock;
2840 unsigned int ee_len;
2843 ee_len = ext4_ext_get_actual_len(ex);
2844 ee_pblock = ext4_ext_pblock(ex);
2846 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2854 * ext4_split_extent_at() splits an extent at given block.
2856 * @handle: the journal handle
2857 * @inode: the file inode
2858 * @path: the path to the extent
2859 * @split: the logical block where the extent is splitted.
2860 * @split_flags: indicates if the extent could be zeroout if split fails, and
2861 * the states(init or uninit) of new extents.
2862 * @flags: flags used to insert new extent to extent tree.
2865 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2866 * of which are deterimined by split_flag.
2868 * There are two cases:
2869 * a> the extent are splitted into two extent.
2870 * b> split is not needed, and just mark the extent.
2872 * return 0 on success.
2874 static int ext4_split_extent_at(handle_t *handle,
2875 struct inode *inode,
2876 struct ext4_ext_path *path,
2881 ext4_fsblk_t newblock;
2882 ext4_lblk_t ee_block;
2883 struct ext4_extent *ex, newex, orig_ex;
2884 struct ext4_extent *ex2 = NULL;
2885 unsigned int ee_len, depth;
2888 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2889 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2891 ext_debug("ext4_split_extents_at: inode %lu, logical"
2892 "block %llu\n", inode->i_ino, (unsigned long long)split);
2894 ext4_ext_show_leaf(inode, path);
2896 depth = ext_depth(inode);
2897 ex = path[depth].p_ext;
2898 ee_block = le32_to_cpu(ex->ee_block);
2899 ee_len = ext4_ext_get_actual_len(ex);
2900 newblock = split - ee_block + ext4_ext_pblock(ex);
2902 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2904 err = ext4_ext_get_access(handle, inode, path + depth);
2908 if (split == ee_block) {
2910 * case b: block @split is the block that the extent begins with
2911 * then we just change the state of the extent, and splitting
2914 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2915 ext4_ext_mark_uninitialized(ex);
2917 ext4_ext_mark_initialized(ex);
2919 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2920 ext4_ext_try_to_merge(inode, path, ex);
2922 err = ext4_ext_dirty(handle, inode, path + depth);
2927 memcpy(&orig_ex, ex, sizeof(orig_ex));
2928 ex->ee_len = cpu_to_le16(split - ee_block);
2929 if (split_flag & EXT4_EXT_MARK_UNINIT1)
2930 ext4_ext_mark_uninitialized(ex);
2933 * path may lead to new leaf, not to original leaf any more
2934 * after ext4_ext_insert_extent() returns,
2936 err = ext4_ext_dirty(handle, inode, path + depth);
2938 goto fix_extent_len;
2941 ex2->ee_block = cpu_to_le32(split);
2942 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
2943 ext4_ext_store_pblock(ex2, newblock);
2944 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2945 ext4_ext_mark_uninitialized(ex2);
2947 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2948 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2949 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
2950 if (split_flag & EXT4_EXT_DATA_VALID1)
2951 err = ext4_ext_zeroout(inode, ex2);
2953 err = ext4_ext_zeroout(inode, ex);
2955 err = ext4_ext_zeroout(inode, &orig_ex);
2958 goto fix_extent_len;
2959 /* update the extent length and mark as initialized */
2960 ex->ee_len = cpu_to_le16(ee_len);
2961 ext4_ext_try_to_merge(inode, path, ex);
2962 err = ext4_ext_dirty(handle, inode, path + depth);
2965 goto fix_extent_len;
2968 ext4_ext_show_leaf(inode, path);
2972 ex->ee_len = orig_ex.ee_len;
2973 ext4_ext_dirty(handle, inode, path + depth);
2978 * ext4_split_extents() splits an extent and mark extent which is covered
2979 * by @map as split_flags indicates
2981 * It may result in splitting the extent into multiple extents (upto three)
2982 * There are three possibilities:
2983 * a> There is no split required
2984 * b> Splits in two extents: Split is happening at either end of the extent
2985 * c> Splits in three extents: Somone is splitting in middle of the extent
2988 static int ext4_split_extent(handle_t *handle,
2989 struct inode *inode,
2990 struct ext4_ext_path *path,
2991 struct ext4_map_blocks *map,
2995 ext4_lblk_t ee_block;
2996 struct ext4_extent *ex;
2997 unsigned int ee_len, depth;
3000 int split_flag1, flags1;
3001 int allocated = map->m_len;
3003 depth = ext_depth(inode);
3004 ex = path[depth].p_ext;
3005 ee_block = le32_to_cpu(ex->ee_block);
3006 ee_len = ext4_ext_get_actual_len(ex);
3007 uninitialized = ext4_ext_is_uninitialized(ex);
3009 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3010 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3011 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3013 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3014 EXT4_EXT_MARK_UNINIT2;
3015 if (split_flag & EXT4_EXT_DATA_VALID2)
3016 split_flag1 |= EXT4_EXT_DATA_VALID1;
3017 err = ext4_split_extent_at(handle, inode, path,
3018 map->m_lblk + map->m_len, split_flag1, flags1);
3022 allocated = ee_len - (map->m_lblk - ee_block);
3025 ext4_ext_drop_refs(path);
3026 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3028 return PTR_ERR(path);
3030 if (map->m_lblk >= ee_block) {
3031 split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT |
3032 EXT4_EXT_DATA_VALID2);
3034 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3035 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3036 split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3037 err = ext4_split_extent_at(handle, inode, path,
3038 map->m_lblk, split_flag1, flags);
3043 ext4_ext_show_leaf(inode, path);
3045 return err ? err : allocated;
3048 #define EXT4_EXT_ZERO_LEN 7
3050 * This function is called by ext4_ext_map_blocks() if someone tries to write
3051 * to an uninitialized extent. It may result in splitting the uninitialized
3052 * extent into multiple extents (up to three - one initialized and two
3054 * There are three possibilities:
3055 * a> There is no split required: Entire extent should be initialized
3056 * b> Splits in two extents: Write is happening at either end of the extent
3057 * c> Splits in three extents: Somone is writing in middle of the extent
3060 * - The extent pointed to by 'path' is uninitialized.
3061 * - The extent pointed to by 'path' contains a superset
3062 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3064 * Post-conditions on success:
3065 * - the returned value is the number of blocks beyond map->l_lblk
3066 * that are allocated and initialized.
3067 * It is guaranteed to be >= map->m_len.
3069 static int ext4_ext_convert_to_initialized(handle_t *handle,
3070 struct inode *inode,
3071 struct ext4_map_blocks *map,
3072 struct ext4_ext_path *path)
3074 struct ext4_extent_header *eh;
3075 struct ext4_map_blocks split_map;
3076 struct ext4_extent zero_ex;
3077 struct ext4_extent *ex;
3078 ext4_lblk_t ee_block, eof_block;
3079 unsigned int ee_len, depth;
3084 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3085 "block %llu, max_blocks %u\n", inode->i_ino,
3086 (unsigned long long)map->m_lblk, map->m_len);
3088 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3089 inode->i_sb->s_blocksize_bits;
3090 if (eof_block < map->m_lblk + map->m_len)
3091 eof_block = map->m_lblk + map->m_len;
3093 depth = ext_depth(inode);
3094 eh = path[depth].p_hdr;
3095 ex = path[depth].p_ext;
3096 ee_block = le32_to_cpu(ex->ee_block);
3097 ee_len = ext4_ext_get_actual_len(ex);
3098 allocated = ee_len - (map->m_lblk - ee_block);
3100 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3102 /* Pre-conditions */
3103 BUG_ON(!ext4_ext_is_uninitialized(ex));
3104 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3107 * Attempt to transfer newly initialized blocks from the currently
3108 * uninitialized extent to its left neighbor. This is much cheaper
3109 * than an insertion followed by a merge as those involve costly
3110 * memmove() calls. This is the common case in steady state for
3111 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3114 * Limitations of the current logic:
3115 * - L1: we only deal with writes at the start of the extent.
3116 * The approach could be extended to writes at the end
3117 * of the extent but this scenario was deemed less common.
3118 * - L2: we do not deal with writes covering the whole extent.
3119 * This would require removing the extent if the transfer
3121 * - L3: we only attempt to merge with an extent stored in the
3122 * same extent tree node.
3124 if ((map->m_lblk == ee_block) && /*L1*/
3125 (map->m_len < ee_len) && /*L2*/
3126 (ex > EXT_FIRST_EXTENT(eh))) { /*L3*/
3127 struct ext4_extent *prev_ex;
3128 ext4_lblk_t prev_lblk;
3129 ext4_fsblk_t prev_pblk, ee_pblk;
3130 unsigned int prev_len, write_len;
3133 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3134 prev_len = ext4_ext_get_actual_len(prev_ex);
3135 prev_pblk = ext4_ext_pblock(prev_ex);
3136 ee_pblk = ext4_ext_pblock(ex);
3137 write_len = map->m_len;
3140 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3141 * upon those conditions:
3142 * - C1: prev_ex is initialized,
3143 * - C2: prev_ex is logically abutting ex,
3144 * - C3: prev_ex is physically abutting ex,
3145 * - C4: prev_ex can receive the additional blocks without
3146 * overflowing the (initialized) length limit.
3148 if ((!ext4_ext_is_uninitialized(prev_ex)) && /*C1*/
3149 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3150 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3151 (prev_len < (EXT_INIT_MAX_LEN - write_len))) { /*C4*/
3152 err = ext4_ext_get_access(handle, inode, path + depth);
3156 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3159 /* Shift the start of ex by 'write_len' blocks */
3160 ex->ee_block = cpu_to_le32(ee_block + write_len);
3161 ext4_ext_store_pblock(ex, ee_pblk + write_len);
3162 ex->ee_len = cpu_to_le16(ee_len - write_len);
3163 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3165 /* Extend prev_ex by 'write_len' blocks */
3166 prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3168 /* Mark the block containing both extents as dirty */
3169 ext4_ext_dirty(handle, inode, path + depth);
3171 /* Update path to point to the right extent */
3172 path[depth].p_ext = prev_ex;
3174 /* Result: number of initialized blocks past m_lblk */
3175 allocated = write_len;
3180 WARN_ON(map->m_lblk < ee_block);
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 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3187 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3188 if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3189 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3190 err = ext4_ext_zeroout(inode, ex);
3194 err = ext4_ext_get_access(handle, inode, path + depth);
3197 ext4_ext_mark_initialized(ex);
3198 ext4_ext_try_to_merge(inode, path, ex);
3199 err = ext4_ext_dirty(handle, inode, path + depth);
3205 * 1. split the extent into three extents.
3206 * 2. split the extent into two extents, zeroout the first half.
3207 * 3. split the extent into two extents, zeroout the second half.
3208 * 4. split the extent into two extents with out zeroout.
3210 split_map.m_lblk = map->m_lblk;
3211 split_map.m_len = map->m_len;
3213 if (allocated > map->m_len) {
3214 if (allocated <= EXT4_EXT_ZERO_LEN &&
3215 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3218 cpu_to_le32(map->m_lblk);
3219 zero_ex.ee_len = cpu_to_le16(allocated);
3220 ext4_ext_store_pblock(&zero_ex,
3221 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3222 err = ext4_ext_zeroout(inode, &zero_ex);
3225 split_map.m_lblk = map->m_lblk;
3226 split_map.m_len = allocated;
3227 } else if ((map->m_lblk - ee_block + map->m_len <
3228 EXT4_EXT_ZERO_LEN) &&
3229 (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3231 if (map->m_lblk != ee_block) {
3232 zero_ex.ee_block = ex->ee_block;
3233 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3235 ext4_ext_store_pblock(&zero_ex,
3236 ext4_ext_pblock(ex));
3237 err = ext4_ext_zeroout(inode, &zero_ex);
3242 split_map.m_lblk = ee_block;
3243 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3244 allocated = map->m_len;
3248 allocated = ext4_split_extent(handle, inode, path,
3249 &split_map, split_flag, 0);
3254 return err ? err : allocated;
3258 * This function is called by ext4_ext_map_blocks() from
3259 * ext4_get_blocks_dio_write() when DIO to write
3260 * to an uninitialized extent.
3262 * Writing to an uninitialized extent may result in splitting the uninitialized
3263 * extent into multiple /initialized uninitialized extents (up to three)
3264 * There are three possibilities:
3265 * a> There is no split required: Entire extent should be uninitialized
3266 * b> Splits in two extents: Write is happening at either end of the extent
3267 * c> Splits in three extents: Somone is writing in middle of the extent
3269 * One of more index blocks maybe needed if the extent tree grow after
3270 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3271 * complete, we need to split the uninitialized extent before DIO submit
3272 * the IO. The uninitialized extent called at this time will be split
3273 * into three uninitialized extent(at most). After IO complete, the part
3274 * being filled will be convert to initialized by the end_io callback function
3275 * via ext4_convert_unwritten_extents().
3277 * Returns the size of uninitialized extent to be written on success.
3279 static int ext4_split_unwritten_extents(handle_t *handle,
3280 struct inode *inode,
3281 struct ext4_map_blocks *map,
3282 struct ext4_ext_path *path,
3285 ext4_lblk_t eof_block;
3286 ext4_lblk_t ee_block;
3287 struct ext4_extent *ex;
3288 unsigned int ee_len;
3289 int split_flag = 0, depth;
3291 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3292 "block %llu, max_blocks %u\n", inode->i_ino,
3293 (unsigned long long)map->m_lblk, map->m_len);
3295 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3296 inode->i_sb->s_blocksize_bits;
3297 if (eof_block < map->m_lblk + map->m_len)
3298 eof_block = map->m_lblk + map->m_len;
3300 * It is safe to convert extent to initialized via explicit
3301 * zeroout only if extent is fully insde i_size or new_size.
3303 depth = ext_depth(inode);
3304 ex = path[depth].p_ext;
3305 ee_block = le32_to_cpu(ex->ee_block);
3306 ee_len = ext4_ext_get_actual_len(ex);
3308 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3309 split_flag |= EXT4_EXT_MARK_UNINIT2;
3310 if (flags & EXT4_GET_BLOCKS_CONVERT)
3311 split_flag |= EXT4_EXT_DATA_VALID2;
3312 flags |= EXT4_GET_BLOCKS_PRE_IO;
3313 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3316 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3317 struct inode *inode,
3318 struct ext4_map_blocks *map,
3319 struct ext4_ext_path *path)
3321 struct ext4_extent *ex;
3322 ext4_lblk_t ee_block;
3323 unsigned int ee_len;
3327 depth = ext_depth(inode);
3328 ex = path[depth].p_ext;
3329 ee_block = le32_to_cpu(ex->ee_block);
3330 ee_len = ext4_ext_get_actual_len(ex);
3332 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3333 "block %llu, max_blocks %u\n", inode->i_ino,
3334 (unsigned long long)ee_block, ee_len);
3336 /* If extent is larger than requested then split is required */
3337 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3338 err = ext4_split_unwritten_extents(handle, inode, map, path,
3339 EXT4_GET_BLOCKS_CONVERT);
3342 ext4_ext_drop_refs(path);
3343 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3345 err = PTR_ERR(path);
3348 depth = ext_depth(inode);
3349 ex = path[depth].p_ext;
3352 err = ext4_ext_get_access(handle, inode, path + depth);
3355 /* first mark the extent as initialized */
3356 ext4_ext_mark_initialized(ex);
3358 /* note: ext4_ext_correct_indexes() isn't needed here because
3359 * borders are not changed
3361 ext4_ext_try_to_merge(inode, path, ex);
3363 /* Mark modified extent as dirty */
3364 err = ext4_ext_dirty(handle, inode, path + depth);
3366 ext4_ext_show_leaf(inode, path);
3370 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3371 sector_t block, int count)
3374 for (i = 0; i < count; i++)
3375 unmap_underlying_metadata(bdev, block + i);
3379 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3381 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3383 struct ext4_ext_path *path,
3387 struct ext4_extent_header *eh;
3388 struct ext4_extent *last_ex;
3390 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3393 depth = ext_depth(inode);
3394 eh = path[depth].p_hdr;
3396 if (unlikely(!eh->eh_entries)) {
3397 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3398 "EOFBLOCKS_FL set");
3401 last_ex = EXT_LAST_EXTENT(eh);
3403 * We should clear the EOFBLOCKS_FL flag if we are writing the
3404 * last block in the last extent in the file. We test this by
3405 * first checking to see if the caller to
3406 * ext4_ext_get_blocks() was interested in the last block (or
3407 * a block beyond the last block) in the current extent. If
3408 * this turns out to be false, we can bail out from this
3409 * function immediately.
3411 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3412 ext4_ext_get_actual_len(last_ex))
3415 * If the caller does appear to be planning to write at or
3416 * beyond the end of the current extent, we then test to see
3417 * if the current extent is the last extent in the file, by
3418 * checking to make sure it was reached via the rightmost node
3419 * at each level of the tree.
3421 for (i = depth-1; i >= 0; i--)
3422 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3424 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3425 return ext4_mark_inode_dirty(handle, inode);
3429 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3431 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3432 * whether there are any buffers marked for delayed allocation. It returns '1'
3433 * on the first delalloc'ed buffer head found. If no buffer head in the given
3434 * range is marked for delalloc, it returns 0.
3435 * lblk_start should always be <= lblk_end.
3436 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3437 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3438 * block sooner). This is useful when blocks are truncated sequentially from
3439 * lblk_start towards lblk_end.
3441 static int ext4_find_delalloc_range(struct inode *inode,
3442 ext4_lblk_t lblk_start,
3443 ext4_lblk_t lblk_end,
3444 int search_hint_reverse)
3446 struct address_space *mapping = inode->i_mapping;
3447 struct buffer_head *head, *bh = NULL;
3449 ext4_lblk_t i, pg_lblk;
3452 /* reverse search wont work if fs block size is less than page size */
3453 if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3454 search_hint_reverse = 0;
3456 if (search_hint_reverse)
3461 index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3463 while ((i >= lblk_start) && (i <= lblk_end)) {
3464 page = find_get_page(mapping, index);
3468 if (!page_has_buffers(page))
3471 head = page_buffers(page);
3476 pg_lblk = index << (PAGE_CACHE_SHIFT -
3479 if (unlikely(pg_lblk < lblk_start)) {
3481 * This is possible when fs block size is less
3482 * than page size and our cluster starts/ends in
3483 * middle of the page. So we need to skip the
3484 * initial few blocks till we reach the 'lblk'
3490 /* Check if the buffer is delayed allocated and that it
3491 * is not yet mapped. (when da-buffers are mapped during
3492 * their writeout, their da_mapped bit is set.)
3494 if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3495 page_cache_release(page);
3496 trace_ext4_find_delalloc_range(inode,
3497 lblk_start, lblk_end,
3498 search_hint_reverse,
3502 if (search_hint_reverse)
3506 } while ((i >= lblk_start) && (i <= lblk_end) &&
3507 ((bh = bh->b_this_page) != head));
3510 page_cache_release(page);
3512 * Move to next page. 'i' will be the first lblk in the next
3515 if (search_hint_reverse)
3519 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3522 trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3523 search_hint_reverse, 0, 0);
3527 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3528 int search_hint_reverse)
3530 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3531 ext4_lblk_t lblk_start, lblk_end;
3532 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3533 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3535 return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3536 search_hint_reverse);
3540 * Determines how many complete clusters (out of those specified by the 'map')
3541 * are under delalloc and were reserved quota for.
3542 * This function is called when we are writing out the blocks that were
3543 * originally written with their allocation delayed, but then the space was
3544 * allocated using fallocate() before the delayed allocation could be resolved.
3545 * The cases to look for are:
3546 * ('=' indicated delayed allocated blocks
3547 * '-' indicates non-delayed allocated blocks)
3548 * (a) partial clusters towards beginning and/or end outside of allocated range
3549 * are not delalloc'ed.
3551 * |----c---=|====c====|====c====|===-c----|
3552 * |++++++ allocated ++++++|
3553 * ==> 4 complete clusters in above example
3555 * (b) partial cluster (outside of allocated range) towards either end is
3556 * marked for delayed allocation. In this case, we will exclude that
3559 * |----====c========|========c========|
3560 * |++++++ allocated ++++++|
3561 * ==> 1 complete clusters in above example
3564 * |================c================|
3565 * |++++++ allocated ++++++|
3566 * ==> 0 complete clusters in above example
3568 * The ext4_da_update_reserve_space will be called only if we
3569 * determine here that there were some "entire" clusters that span
3570 * this 'allocated' range.
3571 * In the non-bigalloc case, this function will just end up returning num_blks
3572 * without ever calling ext4_find_delalloc_range.
3575 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3576 unsigned int num_blks)
3578 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3579 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3580 ext4_lblk_t lblk_from, lblk_to, c_offset;
3581 unsigned int allocated_clusters = 0;
3583 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3584 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3586 /* max possible clusters for this allocation */
3587 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3589 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3591 /* Check towards left side */
3592 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3594 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3595 lblk_to = lblk_from + c_offset - 1;
3597 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3598 allocated_clusters--;
3601 /* Now check towards right. */
3602 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3603 if (allocated_clusters && c_offset) {
3604 lblk_from = lblk_start + num_blks;
3605 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3607 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3608 allocated_clusters--;
3611 return allocated_clusters;
3615 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3616 struct ext4_map_blocks *map,
3617 struct ext4_ext_path *path, int flags,
3618 unsigned int allocated, ext4_fsblk_t newblock)
3622 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3624 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3625 "block %llu, max_blocks %u, flags %d, allocated %u",
3626 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3628 ext4_ext_show_leaf(inode, path);
3630 trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3633 /* get_block() before submit the IO, split the extent */
3634 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3635 ret = ext4_split_unwritten_extents(handle, inode, map,
3638 * Flag the inode(non aio case) or end_io struct (aio case)
3639 * that this IO needs to conversion to written when IO is
3643 ext4_set_io_unwritten_flag(inode, io);
3645 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3646 if (ext4_should_dioread_nolock(inode))
3647 map->m_flags |= EXT4_MAP_UNINIT;
3650 /* IO end_io complete, convert the filled extent to written */
3651 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3652 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3655 ext4_update_inode_fsync_trans(handle, inode, 1);
3656 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3662 /* buffered IO case */
3664 * repeat fallocate creation request
3665 * we already have an unwritten extent
3667 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3670 /* buffered READ or buffered write_begin() lookup */
3671 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3673 * We have blocks reserved already. We
3674 * return allocated blocks so that delalloc
3675 * won't do block reservation for us. But
3676 * the buffer head will be unmapped so that
3677 * a read from the block returns 0s.
3679 map->m_flags |= EXT4_MAP_UNWRITTEN;
3683 /* buffered write, writepage time, convert*/
3684 ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3686 ext4_update_inode_fsync_trans(handle, inode, 1);
3693 map->m_flags |= EXT4_MAP_NEW;
3695 * if we allocated more blocks than requested
3696 * we need to make sure we unmap the extra block
3697 * allocated. The actual needed block will get
3698 * unmapped later when we find the buffer_head marked
3701 if (allocated > map->m_len) {
3702 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3703 newblock + map->m_len,
3704 allocated - map->m_len);
3705 allocated = map->m_len;
3707 map->m_len = allocated;
3710 * If we have done fallocate with the offset that is already
3711 * delayed allocated, we would have block reservation
3712 * and quota reservation done in the delayed write path.
3713 * But fallocate would have already updated quota and block
3714 * count for this offset. So cancel these reservation
3716 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3717 unsigned int reserved_clusters;
3718 reserved_clusters = get_reserved_cluster_alloc(inode,
3719 map->m_lblk, map->m_len);
3720 if (reserved_clusters)
3721 ext4_da_update_reserve_space(inode,
3727 map->m_flags |= EXT4_MAP_MAPPED;
3728 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3729 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3735 if (allocated > map->m_len)
3736 allocated = map->m_len;
3737 ext4_ext_show_leaf(inode, path);
3738 map->m_pblk = newblock;
3739 map->m_len = allocated;
3742 ext4_ext_drop_refs(path);
3745 return err ? err : allocated;
3749 * get_implied_cluster_alloc - check to see if the requested
3750 * allocation (in the map structure) overlaps with a cluster already
3751 * allocated in an extent.
3752 * @sb The filesystem superblock structure
3753 * @map The requested lblk->pblk mapping
3754 * @ex The extent structure which might contain an implied
3755 * cluster allocation
3757 * This function is called by ext4_ext_map_blocks() after we failed to
3758 * find blocks that were already in the inode's extent tree. Hence,
3759 * we know that the beginning of the requested region cannot overlap
3760 * the extent from the inode's extent tree. There are three cases we
3761 * want to catch. The first is this case:
3763 * |--- cluster # N--|
3764 * |--- extent ---| |---- requested region ---|
3767 * The second case that we need to test for is this one:
3769 * |--------- cluster # N ----------------|
3770 * |--- requested region --| |------- extent ----|
3771 * |=======================|
3773 * The third case is when the requested region lies between two extents
3774 * within the same cluster:
3775 * |------------- cluster # N-------------|
3776 * |----- ex -----| |---- ex_right ----|
3777 * |------ requested region ------|
3778 * |================|
3780 * In each of the above cases, we need to set the map->m_pblk and
3781 * map->m_len so it corresponds to the return the extent labelled as
3782 * "|====|" from cluster #N, since it is already in use for data in
3783 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3784 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3785 * as a new "allocated" block region. Otherwise, we will return 0 and
3786 * ext4_ext_map_blocks() will then allocate one or more new clusters
3787 * by calling ext4_mb_new_blocks().
3789 static int get_implied_cluster_alloc(struct super_block *sb,
3790 struct ext4_map_blocks *map,
3791 struct ext4_extent *ex,
3792 struct ext4_ext_path *path)
3794 struct ext4_sb_info *sbi = EXT4_SB(sb);
3795 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3796 ext4_lblk_t ex_cluster_start, ex_cluster_end;
3797 ext4_lblk_t rr_cluster_start, rr_cluster_end;
3798 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3799 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3800 unsigned short ee_len = ext4_ext_get_actual_len(ex);
3802 /* The extent passed in that we are trying to match */
3803 ex_cluster_start = EXT4_B2C(sbi, ee_block);
3804 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3806 /* The requested region passed into ext4_map_blocks() */
3807 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3808 rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3810 if ((rr_cluster_start == ex_cluster_end) ||
3811 (rr_cluster_start == ex_cluster_start)) {
3812 if (rr_cluster_start == ex_cluster_end)
3813 ee_start += ee_len - 1;
3814 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
3815 map->m_len = min(map->m_len,
3816 (unsigned) sbi->s_cluster_ratio - c_offset);
3818 * Check for and handle this case:
3820 * |--------- cluster # N-------------|
3821 * |------- extent ----|
3822 * |--- requested region ---|
3826 if (map->m_lblk < ee_block)
3827 map->m_len = min(map->m_len, ee_block - map->m_lblk);
3830 * Check for the case where there is already another allocated
3831 * block to the right of 'ex' but before the end of the cluster.
3833 * |------------- cluster # N-------------|
3834 * |----- ex -----| |---- ex_right ----|
3835 * |------ requested region ------|
3836 * |================|
3838 if (map->m_lblk > ee_block) {
3839 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3840 map->m_len = min(map->m_len, next - map->m_lblk);
3843 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3847 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3853 * Block allocation/map/preallocation routine for extents based files
3856 * Need to be called with
3857 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3858 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3860 * return > 0, number of of blocks already mapped/allocated
3861 * if create == 0 and these are pre-allocated blocks
3862 * buffer head is unmapped
3863 * otherwise blocks are mapped
3865 * return = 0, if plain look up failed (blocks have not been allocated)
3866 * buffer head is unmapped
3868 * return < 0, error case.
3870 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3871 struct ext4_map_blocks *map, int flags)
3873 struct ext4_ext_path *path = NULL;
3874 struct ext4_extent newex, *ex, *ex2;
3875 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3876 ext4_fsblk_t newblock = 0;
3877 int free_on_err = 0, err = 0, depth, ret;
3878 unsigned int allocated = 0, offset = 0;
3879 unsigned int allocated_clusters = 0;
3880 unsigned int punched_out = 0;
3881 unsigned int result = 0;
3882 struct ext4_allocation_request ar;
3883 ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3884 ext4_lblk_t cluster_offset;
3886 ext_debug("blocks %u/%u requested for inode %lu\n",
3887 map->m_lblk, map->m_len, inode->i_ino);
3888 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3890 /* check in cache */
3891 if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3892 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3893 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3894 if ((sbi->s_cluster_ratio > 1) &&
3895 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3896 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3898 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3900 * block isn't allocated yet and
3901 * user doesn't want to allocate it
3905 /* we should allocate requested block */
3907 /* block is already allocated */
3908 if (sbi->s_cluster_ratio > 1)
3909 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3910 newblock = map->m_lblk
3911 - le32_to_cpu(newex.ee_block)
3912 + ext4_ext_pblock(&newex);
3913 /* number of remaining blocks in the extent */
3914 allocated = ext4_ext_get_actual_len(&newex) -
3915 (map->m_lblk - le32_to_cpu(newex.ee_block));
3920 /* find extent for this block */
3921 path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3923 err = PTR_ERR(path);
3928 depth = ext_depth(inode);
3931 * consistent leaf must not be empty;
3932 * this situation is possible, though, _during_ tree modification;
3933 * this is why assert can't be put in ext4_ext_find_extent()
3935 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3936 EXT4_ERROR_INODE(inode, "bad extent address "
3937 "lblock: %lu, depth: %d pblock %lld",
3938 (unsigned long) map->m_lblk, depth,
3939 path[depth].p_block);
3944 ex = path[depth].p_ext;
3946 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3947 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3948 unsigned short ee_len;
3951 * Uninitialized extents are treated as holes, except that
3952 * we split out initialized portions during a write.
3954 ee_len = ext4_ext_get_actual_len(ex);
3956 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3958 /* if found extent covers block, simply return it */
3959 if (in_range(map->m_lblk, ee_block, ee_len)) {
3960 struct ext4_map_blocks punch_map;
3961 ext4_fsblk_t partial_cluster = 0;
3963 newblock = map->m_lblk - ee_block + ee_start;
3964 /* number of remaining blocks in the extent */
3965 allocated = ee_len - (map->m_lblk - ee_block);
3966 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3967 ee_block, ee_len, newblock);
3969 if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3971 * Do not put uninitialized extent
3974 if (!ext4_ext_is_uninitialized(ex)) {
3975 ext4_ext_put_in_cache(inode, ee_block,
3979 ret = ext4_ext_handle_uninitialized_extents(
3980 handle, inode, map, path, flags,
3981 allocated, newblock);
3986 * Punch out the map length, but only to the
3989 punched_out = allocated < map->m_len ?
3990 allocated : map->m_len;
3993 * Sense extents need to be converted to
3994 * uninitialized, they must fit in an
3995 * uninitialized extent
3997 if (punched_out > EXT_UNINIT_MAX_LEN)
3998 punched_out = EXT_UNINIT_MAX_LEN;
4000 punch_map.m_lblk = map->m_lblk;
4001 punch_map.m_pblk = newblock;
4002 punch_map.m_len = punched_out;
4003 punch_map.m_flags = 0;
4005 /* Check to see if the extent needs to be split */
4006 if (punch_map.m_len != ee_len ||
4007 punch_map.m_lblk != ee_block) {
4009 ret = ext4_split_extent(handle, inode,
4010 path, &punch_map, 0,
4011 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
4012 EXT4_GET_BLOCKS_PRE_IO);
4019 * find extent for the block at
4020 * the start of the hole
4022 ext4_ext_drop_refs(path);
4025 path = ext4_ext_find_extent(inode,
4028 err = PTR_ERR(path);
4033 depth = ext_depth(inode);
4034 ex = path[depth].p_ext;
4035 ee_len = ext4_ext_get_actual_len(ex);
4036 ee_block = le32_to_cpu(ex->ee_block);
4037 ee_start = ext4_ext_pblock(ex);
4041 ext4_ext_mark_uninitialized(ex);
4043 ext4_ext_invalidate_cache(inode);
4045 err = ext4_ext_rm_leaf(handle, inode, path,
4046 &partial_cluster, map->m_lblk,
4047 map->m_lblk + punched_out);
4049 if (!err && path->p_hdr->eh_entries == 0) {
4051 * Punch hole freed all of this sub tree,
4052 * so we need to correct eh_depth
4054 err = ext4_ext_get_access(handle, inode, path);
4056 ext_inode_hdr(inode)->eh_depth = 0;
4057 ext_inode_hdr(inode)->eh_max =
4058 cpu_to_le16(ext4_ext_space_root(
4061 err = ext4_ext_dirty(
4062 handle, inode, path);
4070 if ((sbi->s_cluster_ratio > 1) &&
4071 ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
4072 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4075 * requested block isn't allocated yet;
4076 * we couldn't try to create block if create flag is zero
4078 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4080 * put just found gap into cache to speed up
4081 * subsequent requests
4083 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4088 * Okay, we need to do block allocation.
4090 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4091 newex.ee_block = cpu_to_le32(map->m_lblk);
4092 cluster_offset = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4095 * If we are doing bigalloc, check to see if the extent returned
4096 * by ext4_ext_find_extent() implies a cluster we can use.
4098 if (cluster_offset && ex &&
4099 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4100 ar.len = allocated = map->m_len;
4101 newblock = map->m_pblk;
4102 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4103 goto got_allocated_blocks;
4106 /* find neighbour allocated blocks */
4107 ar.lleft = map->m_lblk;
4108 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4111 ar.lright = map->m_lblk;
4113 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4117 /* Check if the extent after searching to the right implies a
4118 * cluster we can use. */
4119 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4120 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4121 ar.len = allocated = map->m_len;
4122 newblock = map->m_pblk;
4123 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4124 goto got_allocated_blocks;
4128 * See if request is beyond maximum number of blocks we can have in
4129 * a single extent. For an initialized extent this limit is
4130 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4131 * EXT_UNINIT_MAX_LEN.
4133 if (map->m_len > EXT_INIT_MAX_LEN &&
4134 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4135 map->m_len = EXT_INIT_MAX_LEN;
4136 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4137 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4138 map->m_len = EXT_UNINIT_MAX_LEN;
4140 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4141 newex.ee_len = cpu_to_le16(map->m_len);
4142 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4144 allocated = ext4_ext_get_actual_len(&newex);
4146 allocated = map->m_len;
4148 /* allocate new block */
4150 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4151 ar.logical = map->m_lblk;
4153 * We calculate the offset from the beginning of the cluster
4154 * for the logical block number, since when we allocate a
4155 * physical cluster, the physical block should start at the
4156 * same offset from the beginning of the cluster. This is
4157 * needed so that future calls to get_implied_cluster_alloc()
4160 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4161 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4163 ar.logical -= offset;
4164 if (S_ISREG(inode->i_mode))
4165 ar.flags = EXT4_MB_HINT_DATA;
4167 /* disable in-core preallocation for non-regular files */
4169 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4170 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4171 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4174 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4175 ar.goal, newblock, allocated);
4177 allocated_clusters = ar.len;
4178 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4179 if (ar.len > allocated)
4182 got_allocated_blocks:
4183 /* try to insert new extent into found leaf and return */
4184 ext4_ext_store_pblock(&newex, newblock + offset);
4185 newex.ee_len = cpu_to_le16(ar.len);
4186 /* Mark uninitialized */
4187 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4188 ext4_ext_mark_uninitialized(&newex);
4190 * io_end structure was created for every IO write to an
4191 * uninitialized extent. To avoid unnecessary conversion,
4192 * here we flag the IO that really needs the conversion.
4193 * For non asycn direct IO case, flag the inode state
4194 * that we need to perform conversion when IO is done.
4196 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4198 ext4_set_io_unwritten_flag(inode, io);
4200 ext4_set_inode_state(inode,
4201 EXT4_STATE_DIO_UNWRITTEN);
4203 if (ext4_should_dioread_nolock(inode))
4204 map->m_flags |= EXT4_MAP_UNINIT;
4208 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4209 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4212 err = ext4_ext_insert_extent(handle, inode, path,
4214 if (err && free_on_err) {
4215 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4216 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4217 /* free data blocks we just allocated */
4218 /* not a good idea to call discard here directly,
4219 * but otherwise we'd need to call it every free() */
4220 ext4_discard_preallocations(inode);
4221 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4222 ext4_ext_get_actual_len(&newex), fb_flags);
4226 /* previous routine could use block we allocated */
4227 newblock = ext4_ext_pblock(&newex);
4228 allocated = ext4_ext_get_actual_len(&newex);
4229 if (allocated > map->m_len)
4230 allocated = map->m_len;
4231 map->m_flags |= EXT4_MAP_NEW;
4234 * Update reserved blocks/metadata blocks after successful
4235 * block allocation which had been deferred till now.
4237 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4238 unsigned int reserved_clusters;
4240 * Check how many clusters we had reserved this allocated range
4242 reserved_clusters = get_reserved_cluster_alloc(inode,
4243 map->m_lblk, allocated);
4244 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4245 if (reserved_clusters) {
4247 * We have clusters reserved for this range.
4248 * But since we are not doing actual allocation
4249 * and are simply using blocks from previously
4250 * allocated cluster, we should release the
4251 * reservation and not claim quota.
4253 ext4_da_update_reserve_space(inode,
4254 reserved_clusters, 0);
4257 BUG_ON(allocated_clusters < reserved_clusters);
4258 /* We will claim quota for all newly allocated blocks.*/
4259 ext4_da_update_reserve_space(inode, allocated_clusters,
4261 if (reserved_clusters < allocated_clusters) {
4262 struct ext4_inode_info *ei = EXT4_I(inode);
4263 int reservation = allocated_clusters -
4266 * It seems we claimed few clusters outside of
4267 * the range of this allocation. We should give
4268 * it back to the reservation pool. This can
4269 * happen in the following case:
4271 * * Suppose s_cluster_ratio is 4 (i.e., each
4272 * cluster has 4 blocks. Thus, the clusters
4273 * are [0-3],[4-7],[8-11]...
4274 * * First comes delayed allocation write for
4275 * logical blocks 10 & 11. Since there were no
4276 * previous delayed allocated blocks in the
4277 * range [8-11], we would reserve 1 cluster
4279 * * Next comes write for logical blocks 3 to 8.
4280 * In this case, we will reserve 2 clusters
4281 * (for [0-3] and [4-7]; and not for [8-11] as
4282 * that range has a delayed allocated blocks.
4283 * Thus total reserved clusters now becomes 3.
4284 * * Now, during the delayed allocation writeout
4285 * time, we will first write blocks [3-8] and
4286 * allocate 3 clusters for writing these
4287 * blocks. Also, we would claim all these
4288 * three clusters above.
4289 * * Now when we come here to writeout the
4290 * blocks [10-11], we would expect to claim
4291 * the reservation of 1 cluster we had made
4292 * (and we would claim it since there are no
4293 * more delayed allocated blocks in the range
4294 * [8-11]. But our reserved cluster count had
4295 * already gone to 0.
4297 * Thus, at the step 4 above when we determine
4298 * that there are still some unwritten delayed
4299 * allocated blocks outside of our current
4300 * block range, we should increment the
4301 * reserved clusters count so that when the
4302 * remaining blocks finally gets written, we
4305 dquot_reserve_block(inode,
4306 EXT4_C2B(sbi, reservation));
4307 spin_lock(&ei->i_block_reservation_lock);
4308 ei->i_reserved_data_blocks += reservation;
4309 spin_unlock(&ei->i_block_reservation_lock);
4315 * Cache the extent and update transaction to commit on fdatasync only
4316 * when it is _not_ an uninitialized extent.
4318 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4319 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4320 ext4_update_inode_fsync_trans(handle, inode, 1);
4322 ext4_update_inode_fsync_trans(handle, inode, 0);
4324 if (allocated > map->m_len)
4325 allocated = map->m_len;
4326 ext4_ext_show_leaf(inode, path);
4327 map->m_flags |= EXT4_MAP_MAPPED;
4328 map->m_pblk = newblock;
4329 map->m_len = allocated;
4332 ext4_ext_drop_refs(path);
4335 result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4336 punched_out : allocated;
4338 trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4339 newblock, map->m_len, err ? err : result);
4341 return err ? err : result;
4344 void ext4_ext_truncate(struct inode *inode)
4346 struct address_space *mapping = inode->i_mapping;
4347 struct super_block *sb = inode->i_sb;
4348 ext4_lblk_t last_block;
4354 * finish any pending end_io work so we won't run the risk of
4355 * converting any truncated blocks to initialized later
4357 ext4_flush_completed_IO(inode);
4360 * probably first extent we're gonna free will be last in block
4362 err = ext4_writepage_trans_blocks(inode);
4363 handle = ext4_journal_start(inode, err);
4367 if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4368 page_len = PAGE_CACHE_SIZE -
4369 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4371 err = ext4_discard_partial_page_buffers(handle,
4372 mapping, inode->i_size, page_len, 0);
4378 if (ext4_orphan_add(handle, inode))
4381 down_write(&EXT4_I(inode)->i_data_sem);
4382 ext4_ext_invalidate_cache(inode);
4384 ext4_discard_preallocations(inode);
4387 * TODO: optimization is possible here.
4388 * Probably we need not scan at all,
4389 * because page truncation is enough.
4392 /* we have to know where to truncate from in crash case */
4393 EXT4_I(inode)->i_disksize = inode->i_size;
4394 ext4_mark_inode_dirty(handle, inode);
4396 last_block = (inode->i_size + sb->s_blocksize - 1)
4397 >> EXT4_BLOCK_SIZE_BITS(sb);
4398 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4400 /* In a multi-transaction truncate, we only make the final
4401 * transaction synchronous.
4404 ext4_handle_sync(handle);
4406 up_write(&EXT4_I(inode)->i_data_sem);
4410 * If this was a simple ftruncate() and the file will remain alive,
4411 * then we need to clear up the orphan record which we created above.
4412 * However, if this was a real unlink then we were called by
4413 * ext4_delete_inode(), and we allow that function to clean up the
4414 * orphan info for us.
4417 ext4_orphan_del(handle, inode);
4419 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4420 ext4_mark_inode_dirty(handle, inode);
4421 ext4_journal_stop(handle);
4424 static void ext4_falloc_update_inode(struct inode *inode,
4425 int mode, loff_t new_size, int update_ctime)
4427 struct timespec now;
4430 now = current_fs_time(inode->i_sb);
4431 if (!timespec_equal(&inode->i_ctime, &now))
4432 inode->i_ctime = now;
4435 * Update only when preallocation was requested beyond
4438 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4439 if (new_size > i_size_read(inode))
4440 i_size_write(inode, new_size);
4441 if (new_size > EXT4_I(inode)->i_disksize)
4442 ext4_update_i_disksize(inode, new_size);
4445 * Mark that we allocate beyond EOF so the subsequent truncate
4446 * can proceed even if the new size is the same as i_size.
4448 if (new_size > i_size_read(inode))
4449 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4455 * preallocate space for a file. This implements ext4's fallocate file
4456 * operation, which gets called from sys_fallocate system call.
4457 * For block-mapped files, posix_fallocate should fall back to the method
4458 * of writing zeroes to the required new blocks (the same behavior which is
4459 * expected for file systems which do not support fallocate() system call).
4461 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4463 struct inode *inode = file->f_path.dentry->d_inode;
4466 unsigned int max_blocks;
4471 struct ext4_map_blocks map;
4472 unsigned int credits, blkbits = inode->i_blkbits;
4474 /* Return error if mode is not supported */
4475 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4478 if (mode & FALLOC_FL_PUNCH_HOLE)
4479 return ext4_punch_hole(file, offset, len);
4481 trace_ext4_fallocate_enter(inode, offset, len, mode);
4482 map.m_lblk = offset >> blkbits;
4484 * We can't just convert len to max_blocks because
4485 * If blocksize = 4096 offset = 3072 and len = 2048
4487 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4490 * credits to insert 1 extent into extent tree
4492 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4493 mutex_lock(&inode->i_mutex);
4496 * We only support preallocation for extent-based files only
4498 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4503 ret = inode_newsize_ok(inode, (len + offset));
4505 mutex_unlock(&inode->i_mutex);
4506 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4509 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4510 if (mode & FALLOC_FL_KEEP_SIZE)
4511 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4513 * Don't normalize the request if it can fit in one extent so
4514 * that it doesn't get unnecessarily split into multiple
4517 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4518 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4520 while (ret >= 0 && ret < max_blocks) {
4521 map.m_lblk = map.m_lblk + ret;
4522 map.m_len = max_blocks = max_blocks - ret;
4523 handle = ext4_journal_start(inode, credits);
4524 if (IS_ERR(handle)) {
4525 ret = PTR_ERR(handle);
4528 ret = ext4_map_blocks(handle, inode, &map, flags);
4532 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4533 "returned error inode#%lu, block=%u, "
4534 "max_blocks=%u", __func__,
4535 inode->i_ino, map.m_lblk, max_blocks);
4537 ext4_mark_inode_dirty(handle, inode);
4538 ret2 = ext4_journal_stop(handle);
4541 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4542 blkbits) >> blkbits))
4543 new_size = offset + len;
4545 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4547 ext4_falloc_update_inode(inode, mode, new_size,
4548 (map.m_flags & EXT4_MAP_NEW));
4549 ext4_mark_inode_dirty(handle, inode);
4550 ret2 = ext4_journal_stop(handle);
4554 if (ret == -ENOSPC &&
4555 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4560 mutex_unlock(&inode->i_mutex);
4561 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4562 ret > 0 ? ret2 : ret);
4563 return ret > 0 ? ret2 : ret;
4567 * This function convert a range of blocks to written extents
4568 * The caller of this function will pass the start offset and the size.
4569 * all unwritten extents within this range will be converted to
4572 * This function is called from the direct IO end io call back
4573 * function, to convert the fallocated extents after IO is completed.
4574 * Returns 0 on success.
4576 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4580 unsigned int max_blocks;
4583 struct ext4_map_blocks map;
4584 unsigned int credits, blkbits = inode->i_blkbits;
4586 map.m_lblk = offset >> blkbits;
4588 * We can't just convert len to max_blocks because
4589 * If blocksize = 4096 offset = 3072 and len = 2048
4591 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4594 * credits to insert 1 extent into extent tree
4596 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4597 while (ret >= 0 && ret < max_blocks) {
4599 map.m_len = (max_blocks -= ret);
4600 handle = ext4_journal_start(inode, credits);
4601 if (IS_ERR(handle)) {
4602 ret = PTR_ERR(handle);
4605 ret = ext4_map_blocks(handle, inode, &map,
4606 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4609 printk(KERN_ERR "%s: ext4_ext_map_blocks "
4610 "returned error inode#%lu, block=%u, "
4611 "max_blocks=%u", __func__,
4612 inode->i_ino, map.m_lblk, map.m_len);
4614 ext4_mark_inode_dirty(handle, inode);
4615 ret2 = ext4_journal_stop(handle);
4616 if (ret <= 0 || ret2 )
4619 return ret > 0 ? ret2 : ret;
4623 * Callback function called for each extent to gather FIEMAP information.
4625 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4626 struct ext4_ext_cache *newex, struct ext4_extent *ex,
4634 struct fiemap_extent_info *fieinfo = data;
4635 unsigned char blksize_bits;
4637 blksize_bits = inode->i_sb->s_blocksize_bits;
4638 logical = (__u64)newex->ec_block << blksize_bits;
4640 if (newex->ec_start == 0) {
4642 * No extent in extent-tree contains block @newex->ec_start,
4643 * then the block may stay in 1)a hole or 2)delayed-extent.
4645 * Holes or delayed-extents are processed as follows.
4646 * 1. lookup dirty pages with specified range in pagecache.
4647 * If no page is got, then there is no delayed-extent and
4648 * return with EXT_CONTINUE.
4649 * 2. find the 1st mapped buffer,
4650 * 3. check if the mapped buffer is both in the request range
4651 * and a delayed buffer. If not, there is no delayed-extent,
4653 * 4. a delayed-extent is found, the extent will be collected.
4655 ext4_lblk_t end = 0;
4656 pgoff_t last_offset;
4659 pgoff_t start_index = 0;
4660 struct page **pages = NULL;
4661 struct buffer_head *bh = NULL;
4662 struct buffer_head *head = NULL;
4663 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4665 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4669 offset = logical >> PAGE_SHIFT;
4671 last_offset = offset;
4673 ret = find_get_pages_tag(inode->i_mapping, &offset,
4674 PAGECACHE_TAG_DIRTY, nr_pages, pages);
4676 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4677 /* First time, try to find a mapped buffer. */
4680 for (index = 0; index < ret; index++)
4681 page_cache_release(pages[index]);
4684 return EXT_CONTINUE;
4689 /* Try to find the 1st mapped buffer. */
4690 end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4692 if (!page_has_buffers(pages[index]))
4694 head = page_buffers(pages[index]);
4701 if (end >= newex->ec_block +
4703 /* The buffer is out of
4704 * the request range.
4708 if (buffer_mapped(bh) &&
4709 end >= newex->ec_block) {
4710 start_index = index - 1;
4711 /* get the 1st mapped buffer. */
4712 goto found_mapped_buffer;
4715 bh = bh->b_this_page;
4717 } while (bh != head);
4719 /* No mapped buffer in the range found in this page,
4720 * We need to look up next page.
4723 /* There is no page left, but we need to limit
4726 newex->ec_len = end - newex->ec_block;
4731 /*Find contiguous delayed buffers. */
4732 if (ret > 0 && pages[0]->index == last_offset)
4733 head = page_buffers(pages[0]);
4739 found_mapped_buffer:
4740 if (bh != NULL && buffer_delay(bh)) {
4741 /* 1st or contiguous delayed buffer found. */
4742 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4744 * 1st delayed buffer found, record
4745 * the start of extent.
4747 flags |= FIEMAP_EXTENT_DELALLOC;
4748 newex->ec_block = end;
4749 logical = (__u64)end << blksize_bits;
4751 /* Find contiguous delayed buffers. */
4753 if (!buffer_delay(bh))
4754 goto found_delayed_extent;
4755 bh = bh->b_this_page;
4757 } while (bh != head);
4759 for (; index < ret; index++) {
4760 if (!page_has_buffers(pages[index])) {
4764 head = page_buffers(pages[index]);
4770 if (pages[index]->index !=
4771 pages[start_index]->index + index
4773 /* Blocks are not contiguous. */
4779 if (!buffer_delay(bh))
4780 /* Delayed-extent ends. */
4781 goto found_delayed_extent;
4782 bh = bh->b_this_page;
4784 } while (bh != head);
4786 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4790 found_delayed_extent:
4791 newex->ec_len = min(end - newex->ec_block,
4792 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4793 if (ret == nr_pages && bh != NULL &&
4794 newex->ec_len < EXT_INIT_MAX_LEN &&
4796 /* Have not collected an extent and continue. */
4797 for (index = 0; index < ret; index++)
4798 page_cache_release(pages[index]);
4802 for (index = 0; index < ret; index++)
4803 page_cache_release(pages[index]);
4807 physical = (__u64)newex->ec_start << blksize_bits;
4808 length = (__u64)newex->ec_len << blksize_bits;
4810 if (ex && ext4_ext_is_uninitialized(ex))
4811 flags |= FIEMAP_EXTENT_UNWRITTEN;
4813 if (next == EXT_MAX_BLOCKS)
4814 flags |= FIEMAP_EXTENT_LAST;
4816 ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4822 return EXT_CONTINUE;
4824 /* fiemap flags we can handle specified here */
4825 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4827 static int ext4_xattr_fiemap(struct inode *inode,
4828 struct fiemap_extent_info *fieinfo)
4832 __u32 flags = FIEMAP_EXTENT_LAST;
4833 int blockbits = inode->i_sb->s_blocksize_bits;
4837 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4838 struct ext4_iloc iloc;
4839 int offset; /* offset of xattr in inode */
4841 error = ext4_get_inode_loc(inode, &iloc);
4844 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4845 offset = EXT4_GOOD_OLD_INODE_SIZE +
4846 EXT4_I(inode)->i_extra_isize;
4848 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4849 flags |= FIEMAP_EXTENT_DATA_INLINE;
4851 } else { /* external block */
4852 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4853 length = inode->i_sb->s_blocksize;
4857 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4859 return (error < 0 ? error : 0);
4863 * ext4_ext_punch_hole
4865 * Punches a hole of "length" bytes in a file starting
4868 * @inode: The inode of the file to punch a hole in
4869 * @offset: The starting byte offset of the hole
4870 * @length: The length of the hole
4872 * Returns the number of blocks removed or negative on err
4874 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4876 struct inode *inode = file->f_path.dentry->d_inode;
4877 struct super_block *sb = inode->i_sb;
4878 ext4_lblk_t first_block, stop_block;
4879 struct address_space *mapping = inode->i_mapping;
4881 loff_t first_page, last_page, page_len;
4882 loff_t first_page_offset, last_page_offset;
4883 int credits, err = 0;
4885 /* No need to punch hole beyond i_size */
4886 if (offset >= inode->i_size)
4890 * If the hole extends beyond i_size, set the hole
4891 * to end after the page that contains i_size
4893 if (offset + length > inode->i_size) {
4894 length = inode->i_size +
4895 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4899 first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4900 last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4902 first_page_offset = first_page << PAGE_CACHE_SHIFT;
4903 last_page_offset = last_page << PAGE_CACHE_SHIFT;
4906 * Write out all dirty pages to avoid race conditions
4907 * Then release them.
4909 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4910 err = filemap_write_and_wait_range(mapping,
4911 offset, offset + length - 1);
4917 /* Now release the pages */
4918 if (last_page_offset > first_page_offset) {
4919 truncate_inode_pages_range(mapping, first_page_offset,
4920 last_page_offset-1);
4923 /* finish any pending end_io work */
4924 ext4_flush_completed_IO(inode);
4926 credits = ext4_writepage_trans_blocks(inode);
4927 handle = ext4_journal_start(inode, credits);
4929 return PTR_ERR(handle);
4931 err = ext4_orphan_add(handle, inode);
4936 * Now we need to zero out the non-page-aligned data in the
4937 * pages at the start and tail of the hole, and unmap the buffer
4938 * heads for the block aligned regions of the page that were
4939 * completely zeroed.
4941 if (first_page > last_page) {
4943 * If the file space being truncated is contained within a page
4944 * just zero out and unmap the middle of that page
4946 err = ext4_discard_partial_page_buffers(handle,
4947 mapping, offset, length, 0);
4953 * zero out and unmap the partial page that contains
4954 * the start of the hole
4956 page_len = first_page_offset - offset;
4958 err = ext4_discard_partial_page_buffers(handle, mapping,
4959 offset, page_len, 0);
4965 * zero out and unmap the partial page that contains
4966 * the end of the hole
4968 page_len = offset + length - last_page_offset;
4970 err = ext4_discard_partial_page_buffers(handle, mapping,
4971 last_page_offset, page_len, 0);
4978 * If i_size is contained in the last page, we need to
4979 * unmap and zero the partial page after i_size
4981 if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4982 inode->i_size % PAGE_CACHE_SIZE != 0) {
4984 page_len = PAGE_CACHE_SIZE -
4985 (inode->i_size & (PAGE_CACHE_SIZE - 1));
4988 err = ext4_discard_partial_page_buffers(handle,
4989 mapping, inode->i_size, page_len, 0);
4996 first_block = (offset + sb->s_blocksize - 1) >>
4997 EXT4_BLOCK_SIZE_BITS(sb);
4998 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
5000 /* If there are no blocks to remove, return now */
5001 if (first_block >= stop_block)
5004 down_write(&EXT4_I(inode)->i_data_sem);
5005 ext4_ext_invalidate_cache(inode);
5006 ext4_discard_preallocations(inode);
5008 err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
5010 ext4_ext_invalidate_cache(inode);
5011 ext4_discard_preallocations(inode);
5014 ext4_handle_sync(handle);
5016 up_write(&EXT4_I(inode)->i_data_sem);
5019 ext4_orphan_del(handle, inode);
5020 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5021 ext4_mark_inode_dirty(handle, inode);
5022 ext4_journal_stop(handle);
5025 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5026 __u64 start, __u64 len)
5028 ext4_lblk_t start_blk;
5031 /* fallback to generic here if not in extents fmt */
5032 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5033 return generic_block_fiemap(inode, fieinfo, start, len,
5036 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5039 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5040 error = ext4_xattr_fiemap(inode, fieinfo);
5042 ext4_lblk_t len_blks;
5045 start_blk = start >> inode->i_sb->s_blocksize_bits;
5046 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5047 if (last_blk >= EXT_MAX_BLOCKS)
5048 last_blk = EXT_MAX_BLOCKS-1;
5049 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5052 * Walk the extent tree gathering extent information.
5053 * ext4_ext_fiemap_cb will push extents back to user.
5055 error = ext4_ext_walk_space(inode, start_blk, len_blks,
5056 ext4_ext_fiemap_cb, fieinfo);