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
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <linux/falloc.h>
41 #include <asm/uaccess.h>
42 #include <linux/fiemap.h>
43 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
50 * used by extent splitting.
52 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
54 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
55 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
57 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
60 static __le32 ext4_extent_block_csum(struct inode *inode,
61 struct ext4_extent_header *eh)
63 struct ext4_inode_info *ei = EXT4_I(inode);
64 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
68 EXT4_EXTENT_TAIL_OFFSET(eh));
69 return cpu_to_le32(csum);
72 static int ext4_extent_block_csum_verify(struct inode *inode,
73 struct ext4_extent_header *eh)
75 struct ext4_extent_tail *et;
77 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
78 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
81 et = find_ext4_extent_tail(eh);
82 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
87 static void ext4_extent_block_csum_set(struct inode *inode,
88 struct ext4_extent_header *eh)
90 struct ext4_extent_tail *et;
92 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
93 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
96 et = find_ext4_extent_tail(eh);
97 et->et_checksum = ext4_extent_block_csum(inode, eh);
100 static int ext4_split_extent(handle_t *handle,
102 struct ext4_ext_path *path,
103 struct ext4_map_blocks *map,
107 static int ext4_split_extent_at(handle_t *handle,
109 struct ext4_ext_path *path,
114 static int ext4_find_delayed_extent(struct inode *inode,
115 struct extent_status *newes);
117 static int ext4_ext_truncate_extend_restart(handle_t *handle,
123 if (!ext4_handle_valid(handle))
125 if (handle->h_buffer_credits > needed)
127 err = ext4_journal_extend(handle, needed);
130 err = ext4_truncate_restart_trans(handle, inode, needed);
142 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
143 struct ext4_ext_path *path)
146 /* path points to block */
147 return ext4_journal_get_write_access(handle, path->p_bh);
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
160 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
161 struct inode *inode, struct ext4_ext_path *path)
165 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
166 /* path points to block */
167 err = __ext4_handle_dirty_metadata(where, line, handle,
170 /* path points to leaf/index in inode body */
171 err = ext4_mark_inode_dirty(handle, inode);
176 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
177 struct ext4_ext_path *path,
181 int depth = path->p_depth;
182 struct ext4_extent *ex;
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
198 * especially if the latter case turns out to be
201 ex = path[depth].p_ext;
203 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
204 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
206 if (block > ext_block)
207 return ext_pblk + (block - ext_block);
209 return ext_pblk - (ext_block - block);
212 /* it looks like index is empty;
213 * try to find starting block from index itself */
214 if (path[depth].p_bh)
215 return path[depth].p_bh->b_blocknr;
218 /* OK. use inode's group */
219 return ext4_inode_to_goal_block(inode);
223 * Allocation for a meta data block
226 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
227 struct ext4_ext_path *path,
228 struct ext4_extent *ex, int *err, unsigned int flags)
230 ext4_fsblk_t goal, newblock;
232 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
233 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
238 static inline int ext4_ext_space_block(struct inode *inode, int check)
242 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
243 / sizeof(struct ext4_extent);
244 #ifdef AGGRESSIVE_TEST
245 if (!check && size > 6)
251 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
255 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
256 / sizeof(struct ext4_extent_idx);
257 #ifdef AGGRESSIVE_TEST
258 if (!check && size > 5)
264 static inline int ext4_ext_space_root(struct inode *inode, int check)
268 size = sizeof(EXT4_I(inode)->i_data);
269 size -= sizeof(struct ext4_extent_header);
270 size /= sizeof(struct ext4_extent);
271 #ifdef AGGRESSIVE_TEST
272 if (!check && size > 3)
278 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent_idx);
285 #ifdef AGGRESSIVE_TEST
286 if (!check && size > 4)
293 * Calculate the number of metadata blocks needed
294 * to allocate @blocks
295 * Worse case is one block per extent
297 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
299 struct ext4_inode_info *ei = EXT4_I(inode);
302 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
303 / sizeof(struct ext4_extent_idx));
306 * If the new delayed allocation block is contiguous with the
307 * previous da block, it can share index blocks with the
308 * previous block, so we only need to allocate a new index
309 * block every idxs leaf blocks. At ldxs**2 blocks, we need
310 * an additional index block, and at ldxs**3 blocks, yet
311 * another index blocks.
313 if (ei->i_da_metadata_calc_len &&
314 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
317 if ((ei->i_da_metadata_calc_len % idxs) == 0)
319 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
321 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
323 ei->i_da_metadata_calc_len = 0;
325 ei->i_da_metadata_calc_len++;
326 ei->i_da_metadata_calc_last_lblock++;
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
334 ei->i_da_metadata_calc_len = 1;
335 ei->i_da_metadata_calc_last_lblock = lblock;
336 return ext_depth(inode) + 1;
340 ext4_ext_max_entries(struct inode *inode, int depth)
344 if (depth == ext_depth(inode)) {
346 max = ext4_ext_space_root(inode, 1);
348 max = ext4_ext_space_root_idx(inode, 1);
351 max = ext4_ext_space_block(inode, 1);
353 max = ext4_ext_space_block_idx(inode, 1);
359 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
361 ext4_fsblk_t block = ext4_ext_pblock(ext);
362 int len = ext4_ext_get_actual_len(ext);
363 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
364 ext4_lblk_t last = lblock + len - 1;
368 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
371 static int ext4_valid_extent_idx(struct inode *inode,
372 struct ext4_extent_idx *ext_idx)
374 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
376 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
379 static int ext4_valid_extent_entries(struct inode *inode,
380 struct ext4_extent_header *eh,
383 unsigned short entries;
384 if (eh->eh_entries == 0)
387 entries = le16_to_cpu(eh->eh_entries);
391 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
392 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
393 ext4_fsblk_t pblock = 0;
394 ext4_lblk_t lblock = 0;
395 ext4_lblk_t prev = 0;
398 if (!ext4_valid_extent(inode, ext))
401 /* Check for overlapping extents */
402 lblock = le32_to_cpu(ext->ee_block);
403 len = ext4_ext_get_actual_len(ext);
404 if ((lblock <= prev) && prev) {
405 pblock = ext4_ext_pblock(ext);
406 es->s_last_error_block = cpu_to_le64(pblock);
411 prev = lblock + len - 1;
414 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
416 if (!ext4_valid_extent_idx(inode, ext_idx))
425 static int __ext4_ext_check(const char *function, unsigned int line,
426 struct inode *inode, struct ext4_extent_header *eh,
427 int depth, ext4_fsblk_t pblk)
429 const char *error_msg;
432 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
433 error_msg = "invalid magic";
436 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
437 error_msg = "unexpected eh_depth";
440 if (unlikely(eh->eh_max == 0)) {
441 error_msg = "invalid eh_max";
444 max = ext4_ext_max_entries(inode, depth);
445 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
446 error_msg = "too large eh_max";
449 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
450 error_msg = "invalid eh_entries";
453 if (!ext4_valid_extent_entries(inode, eh, depth)) {
454 error_msg = "invalid extent entries";
457 /* Verify checksum on non-root extent tree nodes */
458 if (ext_depth(inode) != depth &&
459 !ext4_extent_block_csum_verify(inode, eh)) {
460 error_msg = "extent tree corrupted";
466 ext4_error_inode(inode, function, line, 0,
467 "pblk %llu bad header/extent: %s - magic %x, "
468 "entries %u, max %u(%u), depth %u(%u)",
469 (unsigned long long) pblk, error_msg,
470 le16_to_cpu(eh->eh_magic),
471 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
472 max, le16_to_cpu(eh->eh_depth), depth);
476 #define ext4_ext_check(inode, eh, depth, pblk) \
477 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
479 int ext4_ext_check_inode(struct inode *inode)
481 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
484 static struct buffer_head *
485 __read_extent_tree_block(const char *function, unsigned int line,
486 struct inode *inode, ext4_fsblk_t pblk, int depth,
489 struct buffer_head *bh;
492 bh = sb_getblk(inode->i_sb, pblk);
494 return ERR_PTR(-ENOMEM);
496 if (!bh_uptodate_or_lock(bh)) {
497 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
498 err = bh_submit_read(bh);
502 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
504 err = __ext4_ext_check(function, line, inode,
505 ext_block_hdr(bh), depth, pblk);
508 set_buffer_verified(bh);
510 * If this is a leaf block, cache all of its entries
512 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
513 struct ext4_extent_header *eh = ext_block_hdr(bh);
514 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
515 ext4_lblk_t prev = 0;
518 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
519 unsigned int status = EXTENT_STATUS_WRITTEN;
520 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
521 int len = ext4_ext_get_actual_len(ex);
523 if (prev && (prev != lblk))
524 ext4_es_cache_extent(inode, prev,
528 if (ext4_ext_is_uninitialized(ex))
529 status = EXTENT_STATUS_UNWRITTEN;
530 ext4_es_cache_extent(inode, lblk, len,
531 ext4_ext_pblock(ex), status);
542 #define read_extent_tree_block(inode, pblk, depth, flags) \
543 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
547 * This function is called to cache a file's extent information in the
550 int ext4_ext_precache(struct inode *inode)
552 struct ext4_inode_info *ei = EXT4_I(inode);
553 struct ext4_ext_path *path = NULL;
554 struct buffer_head *bh;
555 int i = 0, depth, ret = 0;
557 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
558 return 0; /* not an extent-mapped inode */
560 down_read(&ei->i_data_sem);
561 depth = ext_depth(inode);
563 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
566 up_read(&ei->i_data_sem);
570 /* Don't cache anything if there are no external extent blocks */
573 path[0].p_hdr = ext_inode_hdr(inode);
574 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
577 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
580 * If this is a leaf block or we've reached the end of
581 * the index block, go up
584 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
585 brelse(path[i].p_bh);
590 bh = read_extent_tree_block(inode,
591 ext4_idx_pblock(path[i].p_idx++),
593 EXT4_EX_FORCE_CACHE);
600 path[i].p_hdr = ext_block_hdr(bh);
601 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
603 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
605 up_read(&ei->i_data_sem);
606 ext4_ext_drop_refs(path);
612 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
614 int k, l = path->p_depth;
617 for (k = 0; k <= l; k++, path++) {
619 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
620 ext4_idx_pblock(path->p_idx));
621 } else if (path->p_ext) {
622 ext_debug(" %d:[%d]%d:%llu ",
623 le32_to_cpu(path->p_ext->ee_block),
624 ext4_ext_is_uninitialized(path->p_ext),
625 ext4_ext_get_actual_len(path->p_ext),
626 ext4_ext_pblock(path->p_ext));
633 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
635 int depth = ext_depth(inode);
636 struct ext4_extent_header *eh;
637 struct ext4_extent *ex;
643 eh = path[depth].p_hdr;
644 ex = EXT_FIRST_EXTENT(eh);
646 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
648 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
649 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
650 ext4_ext_is_uninitialized(ex),
651 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
656 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
657 ext4_fsblk_t newblock, int level)
659 int depth = ext_depth(inode);
660 struct ext4_extent *ex;
662 if (depth != level) {
663 struct ext4_extent_idx *idx;
664 idx = path[level].p_idx;
665 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
666 ext_debug("%d: move %d:%llu in new index %llu\n", level,
667 le32_to_cpu(idx->ei_block),
668 ext4_idx_pblock(idx),
676 ex = path[depth].p_ext;
677 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
678 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
679 le32_to_cpu(ex->ee_block),
681 ext4_ext_is_uninitialized(ex),
682 ext4_ext_get_actual_len(ex),
689 #define ext4_ext_show_path(inode, path)
690 #define ext4_ext_show_leaf(inode, path)
691 #define ext4_ext_show_move(inode, path, newblock, level)
694 void ext4_ext_drop_refs(struct ext4_ext_path *path)
696 int depth = path->p_depth;
699 for (i = 0; i <= depth; i++, path++)
707 * ext4_ext_binsearch_idx:
708 * binary search for the closest index of the given block
709 * the header must be checked before calling this
712 ext4_ext_binsearch_idx(struct inode *inode,
713 struct ext4_ext_path *path, ext4_lblk_t block)
715 struct ext4_extent_header *eh = path->p_hdr;
716 struct ext4_extent_idx *r, *l, *m;
719 ext_debug("binsearch for %u(idx): ", block);
721 l = EXT_FIRST_INDEX(eh) + 1;
722 r = EXT_LAST_INDEX(eh);
725 if (block < le32_to_cpu(m->ei_block))
729 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
730 m, le32_to_cpu(m->ei_block),
731 r, le32_to_cpu(r->ei_block));
735 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
736 ext4_idx_pblock(path->p_idx));
738 #ifdef CHECK_BINSEARCH
740 struct ext4_extent_idx *chix, *ix;
743 chix = ix = EXT_FIRST_INDEX(eh);
744 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
746 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
747 printk(KERN_DEBUG "k=%d, ix=0x%p, "
749 ix, EXT_FIRST_INDEX(eh));
750 printk(KERN_DEBUG "%u <= %u\n",
751 le32_to_cpu(ix->ei_block),
752 le32_to_cpu(ix[-1].ei_block));
754 BUG_ON(k && le32_to_cpu(ix->ei_block)
755 <= le32_to_cpu(ix[-1].ei_block));
756 if (block < le32_to_cpu(ix->ei_block))
760 BUG_ON(chix != path->p_idx);
767 * ext4_ext_binsearch:
768 * binary search for closest extent of the given block
769 * the header must be checked before calling this
772 ext4_ext_binsearch(struct inode *inode,
773 struct ext4_ext_path *path, ext4_lblk_t block)
775 struct ext4_extent_header *eh = path->p_hdr;
776 struct ext4_extent *r, *l, *m;
778 if (eh->eh_entries == 0) {
780 * this leaf is empty:
781 * we get such a leaf in split/add case
786 ext_debug("binsearch for %u: ", block);
788 l = EXT_FIRST_EXTENT(eh) + 1;
789 r = EXT_LAST_EXTENT(eh);
793 if (block < le32_to_cpu(m->ee_block))
797 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
798 m, le32_to_cpu(m->ee_block),
799 r, le32_to_cpu(r->ee_block));
803 ext_debug(" -> %d:%llu:[%d]%d ",
804 le32_to_cpu(path->p_ext->ee_block),
805 ext4_ext_pblock(path->p_ext),
806 ext4_ext_is_uninitialized(path->p_ext),
807 ext4_ext_get_actual_len(path->p_ext));
809 #ifdef CHECK_BINSEARCH
811 struct ext4_extent *chex, *ex;
814 chex = ex = EXT_FIRST_EXTENT(eh);
815 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
816 BUG_ON(k && le32_to_cpu(ex->ee_block)
817 <= le32_to_cpu(ex[-1].ee_block));
818 if (block < le32_to_cpu(ex->ee_block))
822 BUG_ON(chex != path->p_ext);
828 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
830 struct ext4_extent_header *eh;
832 eh = ext_inode_hdr(inode);
835 eh->eh_magic = EXT4_EXT_MAGIC;
836 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
837 ext4_mark_inode_dirty(handle, inode);
841 struct ext4_ext_path *
842 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
843 struct ext4_ext_path *path, int flags)
845 struct ext4_extent_header *eh;
846 struct buffer_head *bh;
847 short int depth, i, ppos = 0, alloc = 0;
850 eh = ext_inode_hdr(inode);
851 depth = ext_depth(inode);
853 /* account possible depth increase */
855 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
858 return ERR_PTR(-ENOMEM);
865 /* walk through the tree */
867 ext_debug("depth %d: num %d, max %d\n",
868 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
870 ext4_ext_binsearch_idx(inode, path + ppos, block);
871 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
872 path[ppos].p_depth = i;
873 path[ppos].p_ext = NULL;
875 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
882 eh = ext_block_hdr(bh);
884 if (unlikely(ppos > depth)) {
886 EXT4_ERROR_INODE(inode,
887 "ppos %d > depth %d", ppos, depth);
891 path[ppos].p_bh = bh;
892 path[ppos].p_hdr = eh;
895 path[ppos].p_depth = i;
896 path[ppos].p_ext = NULL;
897 path[ppos].p_idx = NULL;
900 ext4_ext_binsearch(inode, path + ppos, block);
901 /* if not an empty leaf */
902 if (path[ppos].p_ext)
903 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
905 ext4_ext_show_path(inode, path);
910 ext4_ext_drop_refs(path);
917 * ext4_ext_insert_index:
918 * insert new index [@logical;@ptr] into the block at @curp;
919 * check where to insert: before @curp or after @curp
921 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
922 struct ext4_ext_path *curp,
923 int logical, ext4_fsblk_t ptr)
925 struct ext4_extent_idx *ix;
928 err = ext4_ext_get_access(handle, inode, curp);
932 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
933 EXT4_ERROR_INODE(inode,
934 "logical %d == ei_block %d!",
935 logical, le32_to_cpu(curp->p_idx->ei_block));
939 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
940 >= le16_to_cpu(curp->p_hdr->eh_max))) {
941 EXT4_ERROR_INODE(inode,
942 "eh_entries %d >= eh_max %d!",
943 le16_to_cpu(curp->p_hdr->eh_entries),
944 le16_to_cpu(curp->p_hdr->eh_max));
948 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
950 ext_debug("insert new index %d after: %llu\n", logical, ptr);
951 ix = curp->p_idx + 1;
954 ext_debug("insert new index %d before: %llu\n", logical, ptr);
958 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
961 ext_debug("insert new index %d: "
962 "move %d indices from 0x%p to 0x%p\n",
963 logical, len, ix, ix + 1);
964 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
967 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
968 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
972 ix->ei_block = cpu_to_le32(logical);
973 ext4_idx_store_pblock(ix, ptr);
974 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
976 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
977 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
981 err = ext4_ext_dirty(handle, inode, curp);
982 ext4_std_error(inode->i_sb, err);
989 * inserts new subtree into the path, using free index entry
991 * - allocates all needed blocks (new leaf and all intermediate index blocks)
992 * - makes decision where to split
993 * - moves remaining extents and index entries (right to the split point)
994 * into the newly allocated blocks
995 * - initializes subtree
997 static int ext4_ext_split(handle_t *handle, struct inode *inode,
999 struct ext4_ext_path *path,
1000 struct ext4_extent *newext, int at)
1002 struct buffer_head *bh = NULL;
1003 int depth = ext_depth(inode);
1004 struct ext4_extent_header *neh;
1005 struct ext4_extent_idx *fidx;
1006 int i = at, k, m, a;
1007 ext4_fsblk_t newblock, oldblock;
1009 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1012 /* make decision: where to split? */
1013 /* FIXME: now decision is simplest: at current extent */
1015 /* if current leaf will be split, then we should use
1016 * border from split point */
1017 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1018 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1021 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1022 border = path[depth].p_ext[1].ee_block;
1023 ext_debug("leaf will be split."
1024 " next leaf starts at %d\n",
1025 le32_to_cpu(border));
1027 border = newext->ee_block;
1028 ext_debug("leaf will be added."
1029 " next leaf starts at %d\n",
1030 le32_to_cpu(border));
1034 * If error occurs, then we break processing
1035 * and mark filesystem read-only. index won't
1036 * be inserted and tree will be in consistent
1037 * state. Next mount will repair buffers too.
1041 * Get array to track all allocated blocks.
1042 * We need this to handle errors and free blocks
1045 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1049 /* allocate all needed blocks */
1050 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1051 for (a = 0; a < depth - at; a++) {
1052 newblock = ext4_ext_new_meta_block(handle, inode, path,
1053 newext, &err, flags);
1056 ablocks[a] = newblock;
1059 /* initialize new leaf */
1060 newblock = ablocks[--a];
1061 if (unlikely(newblock == 0)) {
1062 EXT4_ERROR_INODE(inode, "newblock == 0!");
1066 bh = sb_getblk(inode->i_sb, newblock);
1067 if (unlikely(!bh)) {
1073 err = ext4_journal_get_create_access(handle, bh);
1077 neh = ext_block_hdr(bh);
1078 neh->eh_entries = 0;
1079 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1080 neh->eh_magic = EXT4_EXT_MAGIC;
1083 /* move remainder of path[depth] to the new leaf */
1084 if (unlikely(path[depth].p_hdr->eh_entries !=
1085 path[depth].p_hdr->eh_max)) {
1086 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1087 path[depth].p_hdr->eh_entries,
1088 path[depth].p_hdr->eh_max);
1092 /* start copy from next extent */
1093 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1094 ext4_ext_show_move(inode, path, newblock, depth);
1096 struct ext4_extent *ex;
1097 ex = EXT_FIRST_EXTENT(neh);
1098 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1099 le16_add_cpu(&neh->eh_entries, m);
1102 ext4_extent_block_csum_set(inode, neh);
1103 set_buffer_uptodate(bh);
1106 err = ext4_handle_dirty_metadata(handle, inode, bh);
1112 /* correct old leaf */
1114 err = ext4_ext_get_access(handle, inode, path + depth);
1117 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1118 err = ext4_ext_dirty(handle, inode, path + depth);
1124 /* create intermediate indexes */
1126 if (unlikely(k < 0)) {
1127 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1132 ext_debug("create %d intermediate indices\n", k);
1133 /* insert new index into current index block */
1134 /* current depth stored in i var */
1137 oldblock = newblock;
1138 newblock = ablocks[--a];
1139 bh = sb_getblk(inode->i_sb, newblock);
1140 if (unlikely(!bh)) {
1146 err = ext4_journal_get_create_access(handle, bh);
1150 neh = ext_block_hdr(bh);
1151 neh->eh_entries = cpu_to_le16(1);
1152 neh->eh_magic = EXT4_EXT_MAGIC;
1153 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1154 neh->eh_depth = cpu_to_le16(depth - i);
1155 fidx = EXT_FIRST_INDEX(neh);
1156 fidx->ei_block = border;
1157 ext4_idx_store_pblock(fidx, oldblock);
1159 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1160 i, newblock, le32_to_cpu(border), oldblock);
1162 /* move remainder of path[i] to the new index block */
1163 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1164 EXT_LAST_INDEX(path[i].p_hdr))) {
1165 EXT4_ERROR_INODE(inode,
1166 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1167 le32_to_cpu(path[i].p_ext->ee_block));
1171 /* start copy indexes */
1172 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1173 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1174 EXT_MAX_INDEX(path[i].p_hdr));
1175 ext4_ext_show_move(inode, path, newblock, i);
1177 memmove(++fidx, path[i].p_idx,
1178 sizeof(struct ext4_extent_idx) * m);
1179 le16_add_cpu(&neh->eh_entries, m);
1181 ext4_extent_block_csum_set(inode, neh);
1182 set_buffer_uptodate(bh);
1185 err = ext4_handle_dirty_metadata(handle, inode, bh);
1191 /* correct old index */
1193 err = ext4_ext_get_access(handle, inode, path + i);
1196 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1197 err = ext4_ext_dirty(handle, inode, path + i);
1205 /* insert new index */
1206 err = ext4_ext_insert_index(handle, inode, path + at,
1207 le32_to_cpu(border), newblock);
1211 if (buffer_locked(bh))
1217 /* free all allocated blocks in error case */
1218 for (i = 0; i < depth; i++) {
1221 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1222 EXT4_FREE_BLOCKS_METADATA);
1231 * ext4_ext_grow_indepth:
1232 * implements tree growing procedure:
1233 * - allocates new block
1234 * - moves top-level data (index block or leaf) into the new block
1235 * - initializes new top-level, creating index that points to the
1236 * just created block
1238 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1240 struct ext4_extent *newext)
1242 struct ext4_extent_header *neh;
1243 struct buffer_head *bh;
1244 ext4_fsblk_t newblock;
1247 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1248 newext, &err, flags);
1252 bh = sb_getblk(inode->i_sb, newblock);
1257 err = ext4_journal_get_create_access(handle, bh);
1263 /* move top-level index/leaf into new block */
1264 memmove(bh->b_data, EXT4_I(inode)->i_data,
1265 sizeof(EXT4_I(inode)->i_data));
1267 /* set size of new block */
1268 neh = ext_block_hdr(bh);
1269 /* old root could have indexes or leaves
1270 * so calculate e_max right way */
1271 if (ext_depth(inode))
1272 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1274 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1275 neh->eh_magic = EXT4_EXT_MAGIC;
1276 ext4_extent_block_csum_set(inode, neh);
1277 set_buffer_uptodate(bh);
1280 err = ext4_handle_dirty_metadata(handle, inode, bh);
1284 /* Update top-level index: num,max,pointer */
1285 neh = ext_inode_hdr(inode);
1286 neh->eh_entries = cpu_to_le16(1);
1287 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1288 if (neh->eh_depth == 0) {
1289 /* Root extent block becomes index block */
1290 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1291 EXT_FIRST_INDEX(neh)->ei_block =
1292 EXT_FIRST_EXTENT(neh)->ee_block;
1294 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1295 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1296 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1297 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1299 le16_add_cpu(&neh->eh_depth, 1);
1300 ext4_mark_inode_dirty(handle, inode);
1308 * ext4_ext_create_new_leaf:
1309 * finds empty index and adds new leaf.
1310 * if no free index is found, then it requests in-depth growing.
1312 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1313 unsigned int mb_flags,
1314 unsigned int gb_flags,
1315 struct ext4_ext_path *path,
1316 struct ext4_extent *newext)
1318 struct ext4_ext_path *curp;
1319 int depth, i, err = 0;
1322 i = depth = ext_depth(inode);
1324 /* walk up to the tree and look for free index entry */
1325 curp = path + depth;
1326 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1331 /* we use already allocated block for index block,
1332 * so subsequent data blocks should be contiguous */
1333 if (EXT_HAS_FREE_INDEX(curp)) {
1334 /* if we found index with free entry, then use that
1335 * entry: create all needed subtree and add new leaf */
1336 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1341 ext4_ext_drop_refs(path);
1342 path = ext4_ext_find_extent(inode,
1343 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1346 err = PTR_ERR(path);
1348 /* tree is full, time to grow in depth */
1349 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1354 ext4_ext_drop_refs(path);
1355 path = ext4_ext_find_extent(inode,
1356 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1359 err = PTR_ERR(path);
1364 * only first (depth 0 -> 1) produces free space;
1365 * in all other cases we have to split the grown tree
1367 depth = ext_depth(inode);
1368 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1369 /* now we need to split */
1379 * search the closest allocated block to the left for *logical
1380 * and returns it at @logical + it's physical address at @phys
1381 * if *logical is the smallest allocated block, the function
1382 * returns 0 at @phys
1383 * return value contains 0 (success) or error code
1385 static int ext4_ext_search_left(struct inode *inode,
1386 struct ext4_ext_path *path,
1387 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1389 struct ext4_extent_idx *ix;
1390 struct ext4_extent *ex;
1393 if (unlikely(path == NULL)) {
1394 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1397 depth = path->p_depth;
1400 if (depth == 0 && path->p_ext == NULL)
1403 /* usually extent in the path covers blocks smaller
1404 * then *logical, but it can be that extent is the
1405 * first one in the file */
1407 ex = path[depth].p_ext;
1408 ee_len = ext4_ext_get_actual_len(ex);
1409 if (*logical < le32_to_cpu(ex->ee_block)) {
1410 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1411 EXT4_ERROR_INODE(inode,
1412 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1413 *logical, le32_to_cpu(ex->ee_block));
1416 while (--depth >= 0) {
1417 ix = path[depth].p_idx;
1418 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1419 EXT4_ERROR_INODE(inode,
1420 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1421 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1422 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1423 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1431 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1432 EXT4_ERROR_INODE(inode,
1433 "logical %d < ee_block %d + ee_len %d!",
1434 *logical, le32_to_cpu(ex->ee_block), ee_len);
1438 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1439 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1444 * search the closest allocated block to the right for *logical
1445 * and returns it at @logical + it's physical address at @phys
1446 * if *logical is the largest allocated block, the function
1447 * returns 0 at @phys
1448 * return value contains 0 (success) or error code
1450 static int ext4_ext_search_right(struct inode *inode,
1451 struct ext4_ext_path *path,
1452 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1453 struct ext4_extent **ret_ex)
1455 struct buffer_head *bh = NULL;
1456 struct ext4_extent_header *eh;
1457 struct ext4_extent_idx *ix;
1458 struct ext4_extent *ex;
1460 int depth; /* Note, NOT eh_depth; depth from top of tree */
1463 if (unlikely(path == NULL)) {
1464 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1467 depth = path->p_depth;
1470 if (depth == 0 && path->p_ext == NULL)
1473 /* usually extent in the path covers blocks smaller
1474 * then *logical, but it can be that extent is the
1475 * first one in the file */
1477 ex = path[depth].p_ext;
1478 ee_len = ext4_ext_get_actual_len(ex);
1479 if (*logical < le32_to_cpu(ex->ee_block)) {
1480 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1481 EXT4_ERROR_INODE(inode,
1482 "first_extent(path[%d].p_hdr) != ex",
1486 while (--depth >= 0) {
1487 ix = path[depth].p_idx;
1488 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1489 EXT4_ERROR_INODE(inode,
1490 "ix != EXT_FIRST_INDEX *logical %d!",
1498 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1499 EXT4_ERROR_INODE(inode,
1500 "logical %d < ee_block %d + ee_len %d!",
1501 *logical, le32_to_cpu(ex->ee_block), ee_len);
1505 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1506 /* next allocated block in this leaf */
1511 /* go up and search for index to the right */
1512 while (--depth >= 0) {
1513 ix = path[depth].p_idx;
1514 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1518 /* we've gone up to the root and found no index to the right */
1522 /* we've found index to the right, let's
1523 * follow it and find the closest allocated
1524 * block to the right */
1526 block = ext4_idx_pblock(ix);
1527 while (++depth < path->p_depth) {
1528 /* subtract from p_depth to get proper eh_depth */
1529 bh = read_extent_tree_block(inode, block,
1530 path->p_depth - depth, 0);
1533 eh = ext_block_hdr(bh);
1534 ix = EXT_FIRST_INDEX(eh);
1535 block = ext4_idx_pblock(ix);
1539 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1542 eh = ext_block_hdr(bh);
1543 ex = EXT_FIRST_EXTENT(eh);
1545 *logical = le32_to_cpu(ex->ee_block);
1546 *phys = ext4_ext_pblock(ex);
1554 * ext4_ext_next_allocated_block:
1555 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1556 * NOTE: it considers block number from index entry as
1557 * allocated block. Thus, index entries have to be consistent
1561 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1565 BUG_ON(path == NULL);
1566 depth = path->p_depth;
1568 if (depth == 0 && path->p_ext == NULL)
1569 return EXT_MAX_BLOCKS;
1571 while (depth >= 0) {
1572 if (depth == path->p_depth) {
1574 if (path[depth].p_ext &&
1575 path[depth].p_ext !=
1576 EXT_LAST_EXTENT(path[depth].p_hdr))
1577 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1580 if (path[depth].p_idx !=
1581 EXT_LAST_INDEX(path[depth].p_hdr))
1582 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1587 return EXT_MAX_BLOCKS;
1591 * ext4_ext_next_leaf_block:
1592 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1594 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1598 BUG_ON(path == NULL);
1599 depth = path->p_depth;
1601 /* zero-tree has no leaf blocks at all */
1603 return EXT_MAX_BLOCKS;
1605 /* go to index block */
1608 while (depth >= 0) {
1609 if (path[depth].p_idx !=
1610 EXT_LAST_INDEX(path[depth].p_hdr))
1611 return (ext4_lblk_t)
1612 le32_to_cpu(path[depth].p_idx[1].ei_block);
1616 return EXT_MAX_BLOCKS;
1620 * ext4_ext_correct_indexes:
1621 * if leaf gets modified and modified extent is first in the leaf,
1622 * then we have to correct all indexes above.
1623 * TODO: do we need to correct tree in all cases?
1625 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1626 struct ext4_ext_path *path)
1628 struct ext4_extent_header *eh;
1629 int depth = ext_depth(inode);
1630 struct ext4_extent *ex;
1634 eh = path[depth].p_hdr;
1635 ex = path[depth].p_ext;
1637 if (unlikely(ex == NULL || eh == NULL)) {
1638 EXT4_ERROR_INODE(inode,
1639 "ex %p == NULL or eh %p == NULL", ex, eh);
1644 /* there is no tree at all */
1648 if (ex != EXT_FIRST_EXTENT(eh)) {
1649 /* we correct tree if first leaf got modified only */
1654 * TODO: we need correction if border is smaller than current one
1657 border = path[depth].p_ext->ee_block;
1658 err = ext4_ext_get_access(handle, inode, path + k);
1661 path[k].p_idx->ei_block = border;
1662 err = ext4_ext_dirty(handle, inode, path + k);
1667 /* change all left-side indexes */
1668 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1670 err = ext4_ext_get_access(handle, inode, path + k);
1673 path[k].p_idx->ei_block = border;
1674 err = ext4_ext_dirty(handle, inode, path + k);
1683 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1684 struct ext4_extent *ex2)
1686 unsigned short ext1_ee_len, ext2_ee_len;
1689 * Make sure that both extents are initialized. We don't merge
1690 * uninitialized extents so that we can be sure that end_io code has
1691 * the extent that was written properly split out and conversion to
1692 * initialized is trivial.
1694 if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
1697 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1698 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1700 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1701 le32_to_cpu(ex2->ee_block))
1705 * To allow future support for preallocated extents to be added
1706 * as an RO_COMPAT feature, refuse to merge to extents if
1707 * this can result in the top bit of ee_len being set.
1709 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1711 #ifdef AGGRESSIVE_TEST
1712 if (ext1_ee_len >= 4)
1716 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1722 * This function tries to merge the "ex" extent to the next extent in the tree.
1723 * It always tries to merge towards right. If you want to merge towards
1724 * left, pass "ex - 1" as argument instead of "ex".
1725 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1726 * 1 if they got merged.
1728 static int ext4_ext_try_to_merge_right(struct inode *inode,
1729 struct ext4_ext_path *path,
1730 struct ext4_extent *ex)
1732 struct ext4_extent_header *eh;
1733 unsigned int depth, len;
1736 depth = ext_depth(inode);
1737 BUG_ON(path[depth].p_hdr == NULL);
1738 eh = path[depth].p_hdr;
1740 while (ex < EXT_LAST_EXTENT(eh)) {
1741 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1743 /* merge with next extent! */
1744 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1745 + ext4_ext_get_actual_len(ex + 1));
1747 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1748 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1749 * sizeof(struct ext4_extent);
1750 memmove(ex + 1, ex + 2, len);
1752 le16_add_cpu(&eh->eh_entries, -1);
1754 WARN_ON(eh->eh_entries == 0);
1755 if (!eh->eh_entries)
1756 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1763 * This function does a very simple check to see if we can collapse
1764 * an extent tree with a single extent tree leaf block into the inode.
1766 static void ext4_ext_try_to_merge_up(handle_t *handle,
1767 struct inode *inode,
1768 struct ext4_ext_path *path)
1771 unsigned max_root = ext4_ext_space_root(inode, 0);
1774 if ((path[0].p_depth != 1) ||
1775 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1776 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1780 * We need to modify the block allocation bitmap and the block
1781 * group descriptor to release the extent tree block. If we
1782 * can't get the journal credits, give up.
1784 if (ext4_journal_extend(handle, 2))
1788 * Copy the extent data up to the inode
1790 blk = ext4_idx_pblock(path[0].p_idx);
1791 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1792 sizeof(struct ext4_extent_idx);
1793 s += sizeof(struct ext4_extent_header);
1795 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1796 path[0].p_depth = 0;
1797 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1798 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1799 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1801 brelse(path[1].p_bh);
1802 ext4_free_blocks(handle, inode, NULL, blk, 1,
1803 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET |
1804 EXT4_FREE_BLOCKS_RESERVE);
1808 * This function tries to merge the @ex extent to neighbours in the tree.
1809 * return 1 if merge left else 0.
1811 static void ext4_ext_try_to_merge(handle_t *handle,
1812 struct inode *inode,
1813 struct ext4_ext_path *path,
1814 struct ext4_extent *ex) {
1815 struct ext4_extent_header *eh;
1819 depth = ext_depth(inode);
1820 BUG_ON(path[depth].p_hdr == NULL);
1821 eh = path[depth].p_hdr;
1823 if (ex > EXT_FIRST_EXTENT(eh))
1824 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1827 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1829 ext4_ext_try_to_merge_up(handle, inode, path);
1833 * check if a portion of the "newext" extent overlaps with an
1836 * If there is an overlap discovered, it updates the length of the newext
1837 * such that there will be no overlap, and then returns 1.
1838 * If there is no overlap found, it returns 0.
1840 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1841 struct inode *inode,
1842 struct ext4_extent *newext,
1843 struct ext4_ext_path *path)
1846 unsigned int depth, len1;
1847 unsigned int ret = 0;
1849 b1 = le32_to_cpu(newext->ee_block);
1850 len1 = ext4_ext_get_actual_len(newext);
1851 depth = ext_depth(inode);
1852 if (!path[depth].p_ext)
1854 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1855 b2 &= ~(sbi->s_cluster_ratio - 1);
1858 * get the next allocated block if the extent in the path
1859 * is before the requested block(s)
1862 b2 = ext4_ext_next_allocated_block(path);
1863 if (b2 == EXT_MAX_BLOCKS)
1865 b2 &= ~(sbi->s_cluster_ratio - 1);
1868 /* check for wrap through zero on extent logical start block*/
1869 if (b1 + len1 < b1) {
1870 len1 = EXT_MAX_BLOCKS - b1;
1871 newext->ee_len = cpu_to_le16(len1);
1875 /* check for overlap */
1876 if (b1 + len1 > b2) {
1877 newext->ee_len = cpu_to_le16(b2 - b1);
1885 * ext4_ext_insert_extent:
1886 * tries to merge requsted extent into the existing extent or
1887 * inserts requested extent as new one into the tree,
1888 * creating new leaf in the no-space case.
1890 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1891 struct ext4_ext_path *path,
1892 struct ext4_extent *newext, int gb_flags)
1894 struct ext4_extent_header *eh;
1895 struct ext4_extent *ex, *fex;
1896 struct ext4_extent *nearex; /* nearest extent */
1897 struct ext4_ext_path *npath = NULL;
1898 int depth, len, err;
1902 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1903 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1906 depth = ext_depth(inode);
1907 ex = path[depth].p_ext;
1908 eh = path[depth].p_hdr;
1909 if (unlikely(path[depth].p_hdr == NULL)) {
1910 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1914 /* try to insert block into found extent and return */
1915 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1918 * Try to see whether we should rather test the extent on
1919 * right from ex, or from the left of ex. This is because
1920 * ext4_ext_find_extent() can return either extent on the
1921 * left, or on the right from the searched position. This
1922 * will make merging more effective.
1924 if (ex < EXT_LAST_EXTENT(eh) &&
1925 (le32_to_cpu(ex->ee_block) +
1926 ext4_ext_get_actual_len(ex) <
1927 le32_to_cpu(newext->ee_block))) {
1930 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1931 (le32_to_cpu(newext->ee_block) +
1932 ext4_ext_get_actual_len(newext) <
1933 le32_to_cpu(ex->ee_block)))
1936 /* Try to append newex to the ex */
1937 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1938 ext_debug("append [%d]%d block to %u:[%d]%d"
1940 ext4_ext_is_uninitialized(newext),
1941 ext4_ext_get_actual_len(newext),
1942 le32_to_cpu(ex->ee_block),
1943 ext4_ext_is_uninitialized(ex),
1944 ext4_ext_get_actual_len(ex),
1945 ext4_ext_pblock(ex));
1946 err = ext4_ext_get_access(handle, inode,
1951 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1952 + ext4_ext_get_actual_len(newext));
1953 eh = path[depth].p_hdr;
1959 /* Try to prepend newex to the ex */
1960 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1961 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1963 le32_to_cpu(newext->ee_block),
1964 ext4_ext_is_uninitialized(newext),
1965 ext4_ext_get_actual_len(newext),
1966 le32_to_cpu(ex->ee_block),
1967 ext4_ext_is_uninitialized(ex),
1968 ext4_ext_get_actual_len(ex),
1969 ext4_ext_pblock(ex));
1970 err = ext4_ext_get_access(handle, inode,
1975 ex->ee_block = newext->ee_block;
1976 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1977 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1978 + ext4_ext_get_actual_len(newext));
1979 eh = path[depth].p_hdr;
1985 depth = ext_depth(inode);
1986 eh = path[depth].p_hdr;
1987 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1990 /* probably next leaf has space for us? */
1991 fex = EXT_LAST_EXTENT(eh);
1992 next = EXT_MAX_BLOCKS;
1993 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1994 next = ext4_ext_next_leaf_block(path);
1995 if (next != EXT_MAX_BLOCKS) {
1996 ext_debug("next leaf block - %u\n", next);
1997 BUG_ON(npath != NULL);
1998 npath = ext4_ext_find_extent(inode, next, NULL, 0);
2000 return PTR_ERR(npath);
2001 BUG_ON(npath->p_depth != path->p_depth);
2002 eh = npath[depth].p_hdr;
2003 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2004 ext_debug("next leaf isn't full(%d)\n",
2005 le16_to_cpu(eh->eh_entries));
2009 ext_debug("next leaf has no free space(%d,%d)\n",
2010 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2014 * There is no free space in the found leaf.
2015 * We're gonna add a new leaf in the tree.
2017 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2018 mb_flags = EXT4_MB_USE_RESERVED;
2019 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2023 depth = ext_depth(inode);
2024 eh = path[depth].p_hdr;
2027 nearex = path[depth].p_ext;
2029 err = ext4_ext_get_access(handle, inode, path + depth);
2034 /* there is no extent in this leaf, create first one */
2035 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2036 le32_to_cpu(newext->ee_block),
2037 ext4_ext_pblock(newext),
2038 ext4_ext_is_uninitialized(newext),
2039 ext4_ext_get_actual_len(newext));
2040 nearex = EXT_FIRST_EXTENT(eh);
2042 if (le32_to_cpu(newext->ee_block)
2043 > le32_to_cpu(nearex->ee_block)) {
2045 ext_debug("insert %u:%llu:[%d]%d before: "
2047 le32_to_cpu(newext->ee_block),
2048 ext4_ext_pblock(newext),
2049 ext4_ext_is_uninitialized(newext),
2050 ext4_ext_get_actual_len(newext),
2055 BUG_ON(newext->ee_block == nearex->ee_block);
2056 ext_debug("insert %u:%llu:[%d]%d after: "
2058 le32_to_cpu(newext->ee_block),
2059 ext4_ext_pblock(newext),
2060 ext4_ext_is_uninitialized(newext),
2061 ext4_ext_get_actual_len(newext),
2064 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2066 ext_debug("insert %u:%llu:[%d]%d: "
2067 "move %d extents from 0x%p to 0x%p\n",
2068 le32_to_cpu(newext->ee_block),
2069 ext4_ext_pblock(newext),
2070 ext4_ext_is_uninitialized(newext),
2071 ext4_ext_get_actual_len(newext),
2072 len, nearex, nearex + 1);
2073 memmove(nearex + 1, nearex,
2074 len * sizeof(struct ext4_extent));
2078 le16_add_cpu(&eh->eh_entries, 1);
2079 path[depth].p_ext = nearex;
2080 nearex->ee_block = newext->ee_block;
2081 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2082 nearex->ee_len = newext->ee_len;
2085 /* try to merge extents */
2086 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2087 ext4_ext_try_to_merge(handle, inode, path, nearex);
2090 /* time to correct all indexes above */
2091 err = ext4_ext_correct_indexes(handle, inode, path);
2095 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2099 ext4_ext_drop_refs(npath);
2105 static int ext4_fill_fiemap_extents(struct inode *inode,
2106 ext4_lblk_t block, ext4_lblk_t num,
2107 struct fiemap_extent_info *fieinfo)
2109 struct ext4_ext_path *path = NULL;
2110 struct ext4_extent *ex;
2111 struct extent_status es;
2112 ext4_lblk_t next, next_del, start = 0, end = 0;
2113 ext4_lblk_t last = block + num;
2114 int exists, depth = 0, err = 0;
2115 unsigned int flags = 0;
2116 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2118 while (block < last && block != EXT_MAX_BLOCKS) {
2120 /* find extent for this block */
2121 down_read(&EXT4_I(inode)->i_data_sem);
2123 if (path && ext_depth(inode) != depth) {
2124 /* depth was changed. we have to realloc path */
2129 path = ext4_ext_find_extent(inode, block, path, 0);
2131 up_read(&EXT4_I(inode)->i_data_sem);
2132 err = PTR_ERR(path);
2137 depth = ext_depth(inode);
2138 if (unlikely(path[depth].p_hdr == NULL)) {
2139 up_read(&EXT4_I(inode)->i_data_sem);
2140 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2144 ex = path[depth].p_ext;
2145 next = ext4_ext_next_allocated_block(path);
2146 ext4_ext_drop_refs(path);
2151 /* there is no extent yet, so try to allocate
2152 * all requested space */
2155 } else if (le32_to_cpu(ex->ee_block) > block) {
2156 /* need to allocate space before found extent */
2158 end = le32_to_cpu(ex->ee_block);
2159 if (block + num < end)
2161 } else if (block >= le32_to_cpu(ex->ee_block)
2162 + ext4_ext_get_actual_len(ex)) {
2163 /* need to allocate space after found extent */
2168 } else if (block >= le32_to_cpu(ex->ee_block)) {
2170 * some part of requested space is covered
2174 end = le32_to_cpu(ex->ee_block)
2175 + ext4_ext_get_actual_len(ex);
2176 if (block + num < end)
2182 BUG_ON(end <= start);
2186 es.es_len = end - start;
2189 es.es_lblk = le32_to_cpu(ex->ee_block);
2190 es.es_len = ext4_ext_get_actual_len(ex);
2191 es.es_pblk = ext4_ext_pblock(ex);
2192 if (ext4_ext_is_uninitialized(ex))
2193 flags |= FIEMAP_EXTENT_UNWRITTEN;
2197 * Find delayed extent and update es accordingly. We call
2198 * it even in !exists case to find out whether es is the
2199 * last existing extent or not.
2201 next_del = ext4_find_delayed_extent(inode, &es);
2202 if (!exists && next_del) {
2204 flags |= (FIEMAP_EXTENT_DELALLOC |
2205 FIEMAP_EXTENT_UNKNOWN);
2207 up_read(&EXT4_I(inode)->i_data_sem);
2209 if (unlikely(es.es_len == 0)) {
2210 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2216 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2217 * we need to check next == EXT_MAX_BLOCKS because it is
2218 * possible that an extent is with unwritten and delayed
2219 * status due to when an extent is delayed allocated and
2220 * is allocated by fallocate status tree will track both of
2223 * So we could return a unwritten and delayed extent, and
2224 * its block is equal to 'next'.
2226 if (next == next_del && next == EXT_MAX_BLOCKS) {
2227 flags |= FIEMAP_EXTENT_LAST;
2228 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2229 next != EXT_MAX_BLOCKS)) {
2230 EXT4_ERROR_INODE(inode,
2231 "next extent == %u, next "
2232 "delalloc extent = %u",
2240 err = fiemap_fill_next_extent(fieinfo,
2241 (__u64)es.es_lblk << blksize_bits,
2242 (__u64)es.es_pblk << blksize_bits,
2243 (__u64)es.es_len << blksize_bits,
2253 block = es.es_lblk + es.es_len;
2257 ext4_ext_drop_refs(path);
2265 * ext4_ext_put_gap_in_cache:
2266 * calculate boundaries of the gap that the requested block fits into
2267 * and cache this gap
2270 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2273 int depth = ext_depth(inode);
2274 unsigned long len = 0;
2275 ext4_lblk_t lblock = 0;
2276 struct ext4_extent *ex;
2278 ex = path[depth].p_ext;
2281 * there is no extent yet, so gap is [0;-] and we
2284 ext_debug("cache gap(whole file):");
2285 } else if (block < le32_to_cpu(ex->ee_block)) {
2287 len = le32_to_cpu(ex->ee_block) - block;
2288 ext_debug("cache gap(before): %u [%u:%u]",
2290 le32_to_cpu(ex->ee_block),
2291 ext4_ext_get_actual_len(ex));
2292 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2293 ext4_es_insert_extent(inode, lblock, len, ~0,
2294 EXTENT_STATUS_HOLE);
2295 } else if (block >= le32_to_cpu(ex->ee_block)
2296 + ext4_ext_get_actual_len(ex)) {
2298 lblock = le32_to_cpu(ex->ee_block)
2299 + ext4_ext_get_actual_len(ex);
2301 next = ext4_ext_next_allocated_block(path);
2302 ext_debug("cache gap(after): [%u:%u] %u",
2303 le32_to_cpu(ex->ee_block),
2304 ext4_ext_get_actual_len(ex),
2306 BUG_ON(next == lblock);
2307 len = next - lblock;
2308 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2309 ext4_es_insert_extent(inode, lblock, len, ~0,
2310 EXTENT_STATUS_HOLE);
2315 ext_debug(" -> %u:%lu\n", lblock, len);
2320 * removes index from the index block.
2322 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2323 struct ext4_ext_path *path, int depth)
2328 /* free index block */
2330 path = path + depth;
2331 leaf = ext4_idx_pblock(path->p_idx);
2332 if (unlikely(path->p_hdr->eh_entries == 0)) {
2333 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2336 err = ext4_ext_get_access(handle, inode, path);
2340 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2341 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2342 len *= sizeof(struct ext4_extent_idx);
2343 memmove(path->p_idx, path->p_idx + 1, len);
2346 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2347 err = ext4_ext_dirty(handle, inode, path);
2350 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2351 trace_ext4_ext_rm_idx(inode, leaf);
2353 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2354 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2356 while (--depth >= 0) {
2357 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2360 err = ext4_ext_get_access(handle, inode, path);
2363 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2364 err = ext4_ext_dirty(handle, inode, path);
2372 * ext4_ext_calc_credits_for_single_extent:
2373 * This routine returns max. credits that needed to insert an extent
2374 * to the extent tree.
2375 * When pass the actual path, the caller should calculate credits
2378 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2379 struct ext4_ext_path *path)
2382 int depth = ext_depth(inode);
2385 /* probably there is space in leaf? */
2386 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2387 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2390 * There are some space in the leaf tree, no
2391 * need to account for leaf block credit
2393 * bitmaps and block group descriptor blocks
2394 * and other metadata blocks still need to be
2397 /* 1 bitmap, 1 block group descriptor */
2398 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2403 return ext4_chunk_trans_blocks(inode, nrblocks);
2407 * How many index/leaf blocks need to change/allocate to add @extents extents?
2409 * If we add a single extent, then in the worse case, each tree level
2410 * index/leaf need to be changed in case of the tree split.
2412 * If more extents are inserted, they could cause the whole tree split more
2413 * than once, but this is really rare.
2415 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2420 /* If we are converting the inline data, only one is needed here. */
2421 if (ext4_has_inline_data(inode))
2424 depth = ext_depth(inode);
2434 static inline int get_default_free_blocks_flags(struct inode *inode)
2436 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2437 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2438 else if (ext4_should_journal_data(inode))
2439 return EXT4_FREE_BLOCKS_FORGET;
2443 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2444 struct ext4_extent *ex,
2445 long long *partial_cluster,
2446 ext4_lblk_t from, ext4_lblk_t to)
2448 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2449 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2451 int flags = get_default_free_blocks_flags(inode);
2454 * For bigalloc file systems, we never free a partial cluster
2455 * at the beginning of the extent. Instead, we make a note
2456 * that we tried freeing the cluster, and check to see if we
2457 * need to free it on a subsequent call to ext4_remove_blocks,
2458 * or at the end of the ext4_truncate() operation.
2460 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2462 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2464 * If we have a partial cluster, and it's different from the
2465 * cluster of the last block, we need to explicitly free the
2466 * partial cluster here.
2468 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2469 if ((*partial_cluster > 0) &&
2470 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2471 ext4_free_blocks(handle, inode, NULL,
2472 EXT4_C2B(sbi, *partial_cluster),
2473 sbi->s_cluster_ratio, flags);
2474 *partial_cluster = 0;
2477 #ifdef EXTENTS_STATS
2479 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2480 spin_lock(&sbi->s_ext_stats_lock);
2481 sbi->s_ext_blocks += ee_len;
2482 sbi->s_ext_extents++;
2483 if (ee_len < sbi->s_ext_min)
2484 sbi->s_ext_min = ee_len;
2485 if (ee_len > sbi->s_ext_max)
2486 sbi->s_ext_max = ee_len;
2487 if (ext_depth(inode) > sbi->s_depth_max)
2488 sbi->s_depth_max = ext_depth(inode);
2489 spin_unlock(&sbi->s_ext_stats_lock);
2492 if (from >= le32_to_cpu(ex->ee_block)
2493 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2496 unsigned int unaligned;
2498 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2499 pblk = ext4_ext_pblock(ex) + ee_len - num;
2501 * Usually we want to free partial cluster at the end of the
2502 * extent, except for the situation when the cluster is still
2503 * used by any other extent (partial_cluster is negative).
2505 if (*partial_cluster < 0 &&
2506 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2507 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2509 ext_debug("free last %u blocks starting %llu partial %lld\n",
2510 num, pblk, *partial_cluster);
2511 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2513 * If the block range to be freed didn't start at the
2514 * beginning of a cluster, and we removed the entire
2515 * extent and the cluster is not used by any other extent,
2516 * save the partial cluster here, since we might need to
2517 * delete if we determine that the truncate operation has
2518 * removed all of the blocks in the cluster.
2520 * On the other hand, if we did not manage to free the whole
2521 * extent, we have to mark the cluster as used (store negative
2522 * cluster number in partial_cluster).
2524 unaligned = pblk & (sbi->s_cluster_ratio - 1);
2525 if (unaligned && (ee_len == num) &&
2526 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2527 *partial_cluster = EXT4_B2C(sbi, pblk);
2529 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2530 else if (*partial_cluster > 0)
2531 *partial_cluster = 0;
2533 ext4_error(sbi->s_sb, "strange request: removal(2) "
2534 "%u-%u from %u:%u\n",
2535 from, to, le32_to_cpu(ex->ee_block), ee_len);
2541 * ext4_ext_rm_leaf() Removes the extents associated with the
2542 * blocks appearing between "start" and "end", and splits the extents
2543 * if "start" and "end" appear in the same extent
2545 * @handle: The journal handle
2546 * @inode: The files inode
2547 * @path: The path to the leaf
2548 * @partial_cluster: The cluster which we'll have to free if all extents
2549 * has been released from it. It gets negative in case
2550 * that the cluster is still used.
2551 * @start: The first block to remove
2552 * @end: The last block to remove
2555 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2556 struct ext4_ext_path *path,
2557 long long *partial_cluster,
2558 ext4_lblk_t start, ext4_lblk_t end)
2560 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2561 int err = 0, correct_index = 0;
2562 int depth = ext_depth(inode), credits;
2563 struct ext4_extent_header *eh;
2566 ext4_lblk_t ex_ee_block;
2567 unsigned short ex_ee_len;
2568 unsigned uninitialized = 0;
2569 struct ext4_extent *ex;
2572 /* the header must be checked already in ext4_ext_remove_space() */
2573 ext_debug("truncate since %u in leaf to %u\n", start, end);
2574 if (!path[depth].p_hdr)
2575 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2576 eh = path[depth].p_hdr;
2577 if (unlikely(path[depth].p_hdr == NULL)) {
2578 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2581 /* find where to start removing */
2582 ex = path[depth].p_ext;
2584 ex = EXT_LAST_EXTENT(eh);
2586 ex_ee_block = le32_to_cpu(ex->ee_block);
2587 ex_ee_len = ext4_ext_get_actual_len(ex);
2589 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2591 while (ex >= EXT_FIRST_EXTENT(eh) &&
2592 ex_ee_block + ex_ee_len > start) {
2594 if (ext4_ext_is_uninitialized(ex))
2599 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2600 uninitialized, ex_ee_len);
2601 path[depth].p_ext = ex;
2603 a = ex_ee_block > start ? ex_ee_block : start;
2604 b = ex_ee_block+ex_ee_len - 1 < end ?
2605 ex_ee_block+ex_ee_len - 1 : end;
2607 ext_debug(" border %u:%u\n", a, b);
2609 /* If this extent is beyond the end of the hole, skip it */
2610 if (end < ex_ee_block) {
2612 * We're going to skip this extent and move to another,
2613 * so if this extent is not cluster aligned we have
2614 * to mark the current cluster as used to avoid
2615 * accidentally freeing it later on
2617 pblk = ext4_ext_pblock(ex);
2618 if (pblk & (sbi->s_cluster_ratio - 1))
2620 -((long long)EXT4_B2C(sbi, pblk));
2622 ex_ee_block = le32_to_cpu(ex->ee_block);
2623 ex_ee_len = ext4_ext_get_actual_len(ex);
2625 } else if (b != ex_ee_block + ex_ee_len - 1) {
2626 EXT4_ERROR_INODE(inode,
2627 "can not handle truncate %u:%u "
2629 start, end, ex_ee_block,
2630 ex_ee_block + ex_ee_len - 1);
2633 } else if (a != ex_ee_block) {
2634 /* remove tail of the extent */
2635 num = a - ex_ee_block;
2637 /* remove whole extent: excellent! */
2641 * 3 for leaf, sb, and inode plus 2 (bmap and group
2642 * descriptor) for each block group; assume two block
2643 * groups plus ex_ee_len/blocks_per_block_group for
2646 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2647 if (ex == EXT_FIRST_EXTENT(eh)) {
2649 credits += (ext_depth(inode)) + 1;
2651 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2653 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2657 err = ext4_ext_get_access(handle, inode, path + depth);
2661 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2667 /* this extent is removed; mark slot entirely unused */
2668 ext4_ext_store_pblock(ex, 0);
2670 ex->ee_len = cpu_to_le16(num);
2672 * Do not mark uninitialized if all the blocks in the
2673 * extent have been removed.
2675 if (uninitialized && num)
2676 ext4_ext_mark_uninitialized(ex);
2678 * If the extent was completely released,
2679 * we need to remove it from the leaf
2682 if (end != EXT_MAX_BLOCKS - 1) {
2684 * For hole punching, we need to scoot all the
2685 * extents up when an extent is removed so that
2686 * we dont have blank extents in the middle
2688 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2689 sizeof(struct ext4_extent));
2691 /* Now get rid of the one at the end */
2692 memset(EXT_LAST_EXTENT(eh), 0,
2693 sizeof(struct ext4_extent));
2695 le16_add_cpu(&eh->eh_entries, -1);
2696 } else if (*partial_cluster > 0)
2697 *partial_cluster = 0;
2699 err = ext4_ext_dirty(handle, inode, path + depth);
2703 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2704 ext4_ext_pblock(ex));
2706 ex_ee_block = le32_to_cpu(ex->ee_block);
2707 ex_ee_len = ext4_ext_get_actual_len(ex);
2710 if (correct_index && eh->eh_entries)
2711 err = ext4_ext_correct_indexes(handle, inode, path);
2714 * Free the partial cluster only if the current extent does not
2715 * reference it. Otherwise we might free used cluster.
2717 if (*partial_cluster > 0 &&
2718 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2719 *partial_cluster)) {
2720 int flags = get_default_free_blocks_flags(inode);
2722 ext4_free_blocks(handle, inode, NULL,
2723 EXT4_C2B(sbi, *partial_cluster),
2724 sbi->s_cluster_ratio, flags);
2725 *partial_cluster = 0;
2728 /* if this leaf is free, then we should
2729 * remove it from index block above */
2730 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2731 err = ext4_ext_rm_idx(handle, inode, path, depth);
2738 * ext4_ext_more_to_rm:
2739 * returns 1 if current index has to be freed (even partial)
2742 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2744 BUG_ON(path->p_idx == NULL);
2746 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2750 * if truncate on deeper level happened, it wasn't partial,
2751 * so we have to consider current index for truncation
2753 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2758 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2761 struct super_block *sb = inode->i_sb;
2762 int depth = ext_depth(inode);
2763 struct ext4_ext_path *path = NULL;
2764 long long partial_cluster = 0;
2768 ext_debug("truncate since %u to %u\n", start, end);
2770 /* probably first extent we're gonna free will be last in block */
2771 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2773 return PTR_ERR(handle);
2776 trace_ext4_ext_remove_space(inode, start, end, depth);
2779 * Check if we are removing extents inside the extent tree. If that
2780 * is the case, we are going to punch a hole inside the extent tree
2781 * so we have to check whether we need to split the extent covering
2782 * the last block to remove so we can easily remove the part of it
2783 * in ext4_ext_rm_leaf().
2785 if (end < EXT_MAX_BLOCKS - 1) {
2786 struct ext4_extent *ex;
2787 ext4_lblk_t ee_block;
2789 /* find extent for this block */
2790 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2792 ext4_journal_stop(handle);
2793 return PTR_ERR(path);
2795 depth = ext_depth(inode);
2796 /* Leaf not may not exist only if inode has no blocks at all */
2797 ex = path[depth].p_ext;
2800 EXT4_ERROR_INODE(inode,
2801 "path[%d].p_hdr == NULL",
2808 ee_block = le32_to_cpu(ex->ee_block);
2811 * See if the last block is inside the extent, if so split
2812 * the extent at 'end' block so we can easily remove the
2813 * tail of the first part of the split extent in
2814 * ext4_ext_rm_leaf().
2816 if (end >= ee_block &&
2817 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2820 if (ext4_ext_is_uninitialized(ex))
2821 split_flag = EXT4_EXT_MARK_UNINIT1 |
2822 EXT4_EXT_MARK_UNINIT2;
2825 * Split the extent in two so that 'end' is the last
2826 * block in the first new extent. Also we should not
2827 * fail removing space due to ENOSPC so try to use
2828 * reserved block if that happens.
2830 err = ext4_split_extent_at(handle, inode, path,
2831 end + 1, split_flag,
2833 EXT4_GET_BLOCKS_PRE_IO |
2834 EXT4_GET_BLOCKS_METADATA_NOFAIL);
2841 * We start scanning from right side, freeing all the blocks
2842 * after i_size and walking into the tree depth-wise.
2844 depth = ext_depth(inode);
2849 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2851 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2854 ext4_journal_stop(handle);
2857 path[0].p_depth = depth;
2858 path[0].p_hdr = ext_inode_hdr(inode);
2861 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2868 while (i >= 0 && err == 0) {
2870 /* this is leaf block */
2871 err = ext4_ext_rm_leaf(handle, inode, path,
2872 &partial_cluster, start,
2874 /* root level has p_bh == NULL, brelse() eats this */
2875 brelse(path[i].p_bh);
2876 path[i].p_bh = NULL;
2881 /* this is index block */
2882 if (!path[i].p_hdr) {
2883 ext_debug("initialize header\n");
2884 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2887 if (!path[i].p_idx) {
2888 /* this level hasn't been touched yet */
2889 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2890 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2891 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2893 le16_to_cpu(path[i].p_hdr->eh_entries));
2895 /* we were already here, see at next index */
2899 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2900 i, EXT_FIRST_INDEX(path[i].p_hdr),
2902 if (ext4_ext_more_to_rm(path + i)) {
2903 struct buffer_head *bh;
2904 /* go to the next level */
2905 ext_debug("move to level %d (block %llu)\n",
2906 i + 1, ext4_idx_pblock(path[i].p_idx));
2907 memset(path + i + 1, 0, sizeof(*path));
2908 bh = read_extent_tree_block(inode,
2909 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2912 /* should we reset i_size? */
2916 /* Yield here to deal with large extent trees.
2917 * Should be a no-op if we did IO above. */
2919 if (WARN_ON(i + 1 > depth)) {
2923 path[i + 1].p_bh = bh;
2925 /* save actual number of indexes since this
2926 * number is changed at the next iteration */
2927 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2930 /* we finished processing this index, go up */
2931 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2932 /* index is empty, remove it;
2933 * handle must be already prepared by the
2934 * truncatei_leaf() */
2935 err = ext4_ext_rm_idx(handle, inode, path, i);
2937 /* root level has p_bh == NULL, brelse() eats this */
2938 brelse(path[i].p_bh);
2939 path[i].p_bh = NULL;
2941 ext_debug("return to level %d\n", i);
2945 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2946 partial_cluster, path->p_hdr->eh_entries);
2948 /* If we still have something in the partial cluster and we have removed
2949 * even the first extent, then we should free the blocks in the partial
2950 * cluster as well. */
2951 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2952 int flags = get_default_free_blocks_flags(inode);
2954 ext4_free_blocks(handle, inode, NULL,
2955 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2956 EXT4_SB(sb)->s_cluster_ratio, flags);
2957 partial_cluster = 0;
2960 /* TODO: flexible tree reduction should be here */
2961 if (path->p_hdr->eh_entries == 0) {
2963 * truncate to zero freed all the tree,
2964 * so we need to correct eh_depth
2966 err = ext4_ext_get_access(handle, inode, path);
2968 ext_inode_hdr(inode)->eh_depth = 0;
2969 ext_inode_hdr(inode)->eh_max =
2970 cpu_to_le16(ext4_ext_space_root(inode, 0));
2971 err = ext4_ext_dirty(handle, inode, path);
2975 ext4_ext_drop_refs(path);
2977 if (err == -EAGAIN) {
2981 ext4_journal_stop(handle);
2987 * called at mount time
2989 void ext4_ext_init(struct super_block *sb)
2992 * possible initialization would be here
2995 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2996 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2997 printk(KERN_INFO "EXT4-fs: file extents enabled"
2998 #ifdef AGGRESSIVE_TEST
2999 ", aggressive tests"
3001 #ifdef CHECK_BINSEARCH
3004 #ifdef EXTENTS_STATS
3009 #ifdef EXTENTS_STATS
3010 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3011 EXT4_SB(sb)->s_ext_min = 1 << 30;
3012 EXT4_SB(sb)->s_ext_max = 0;
3018 * called at umount time
3020 void ext4_ext_release(struct super_block *sb)
3022 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3025 #ifdef EXTENTS_STATS
3026 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3027 struct ext4_sb_info *sbi = EXT4_SB(sb);
3028 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3029 sbi->s_ext_blocks, sbi->s_ext_extents,
3030 sbi->s_ext_blocks / sbi->s_ext_extents);
3031 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3032 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3037 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3039 ext4_lblk_t ee_block;
3040 ext4_fsblk_t ee_pblock;
3041 unsigned int ee_len;
3043 ee_block = le32_to_cpu(ex->ee_block);
3044 ee_len = ext4_ext_get_actual_len(ex);
3045 ee_pblock = ext4_ext_pblock(ex);
3050 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3051 EXTENT_STATUS_WRITTEN);
3054 /* FIXME!! we need to try to merge to left or right after zero-out */
3055 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3057 ext4_fsblk_t ee_pblock;
3058 unsigned int ee_len;
3061 ee_len = ext4_ext_get_actual_len(ex);
3062 ee_pblock = ext4_ext_pblock(ex);
3064 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3072 * ext4_split_extent_at() splits an extent at given block.
3074 * @handle: the journal handle
3075 * @inode: the file inode
3076 * @path: the path to the extent
3077 * @split: the logical block where the extent is splitted.
3078 * @split_flags: indicates if the extent could be zeroout if split fails, and
3079 * the states(init or uninit) of new extents.
3080 * @flags: flags used to insert new extent to extent tree.
3083 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3084 * of which are deterimined by split_flag.
3086 * There are two cases:
3087 * a> the extent are splitted into two extent.
3088 * b> split is not needed, and just mark the extent.
3090 * return 0 on success.
3092 static int ext4_split_extent_at(handle_t *handle,
3093 struct inode *inode,
3094 struct ext4_ext_path *path,
3099 ext4_fsblk_t newblock;
3100 ext4_lblk_t ee_block;
3101 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3102 struct ext4_extent *ex2 = NULL;
3103 unsigned int ee_len, depth;
3106 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3107 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3109 ext_debug("ext4_split_extents_at: inode %lu, logical"
3110 "block %llu\n", inode->i_ino, (unsigned long long)split);
3112 ext4_ext_show_leaf(inode, path);
3114 depth = ext_depth(inode);
3115 ex = path[depth].p_ext;
3116 ee_block = le32_to_cpu(ex->ee_block);
3117 ee_len = ext4_ext_get_actual_len(ex);
3118 newblock = split - ee_block + ext4_ext_pblock(ex);
3120 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3121 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3122 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3123 EXT4_EXT_MARK_UNINIT1 |
3124 EXT4_EXT_MARK_UNINIT2));
3126 err = ext4_ext_get_access(handle, inode, path + depth);
3130 if (split == ee_block) {
3132 * case b: block @split is the block that the extent begins with
3133 * then we just change the state of the extent, and splitting
3136 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3137 ext4_ext_mark_uninitialized(ex);
3139 ext4_ext_mark_initialized(ex);
3141 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3142 ext4_ext_try_to_merge(handle, inode, path, ex);
3144 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3149 memcpy(&orig_ex, ex, sizeof(orig_ex));
3150 ex->ee_len = cpu_to_le16(split - ee_block);
3151 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3152 ext4_ext_mark_uninitialized(ex);
3155 * path may lead to new leaf, not to original leaf any more
3156 * after ext4_ext_insert_extent() returns,
3158 err = ext4_ext_dirty(handle, inode, path + depth);
3160 goto fix_extent_len;
3163 ex2->ee_block = cpu_to_le32(split);
3164 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3165 ext4_ext_store_pblock(ex2, newblock);
3166 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3167 ext4_ext_mark_uninitialized(ex2);
3169 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3170 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3171 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3172 if (split_flag & EXT4_EXT_DATA_VALID1) {
3173 err = ext4_ext_zeroout(inode, ex2);
3174 zero_ex.ee_block = ex2->ee_block;
3175 zero_ex.ee_len = cpu_to_le16(
3176 ext4_ext_get_actual_len(ex2));
3177 ext4_ext_store_pblock(&zero_ex,
3178 ext4_ext_pblock(ex2));
3180 err = ext4_ext_zeroout(inode, ex);
3181 zero_ex.ee_block = ex->ee_block;
3182 zero_ex.ee_len = cpu_to_le16(
3183 ext4_ext_get_actual_len(ex));
3184 ext4_ext_store_pblock(&zero_ex,
3185 ext4_ext_pblock(ex));
3188 err = ext4_ext_zeroout(inode, &orig_ex);
3189 zero_ex.ee_block = orig_ex.ee_block;
3190 zero_ex.ee_len = cpu_to_le16(
3191 ext4_ext_get_actual_len(&orig_ex));
3192 ext4_ext_store_pblock(&zero_ex,
3193 ext4_ext_pblock(&orig_ex));
3197 goto fix_extent_len;
3198 /* update the extent length and mark as initialized */
3199 ex->ee_len = cpu_to_le16(ee_len);
3200 ext4_ext_try_to_merge(handle, inode, path, ex);
3201 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3203 goto fix_extent_len;
3205 /* update extent status tree */
3206 err = ext4_zeroout_es(inode, &zero_ex);
3210 goto fix_extent_len;
3213 ext4_ext_show_leaf(inode, path);
3217 ex->ee_len = orig_ex.ee_len;
3218 ext4_ext_dirty(handle, inode, path + depth);
3223 * ext4_split_extents() splits an extent and mark extent which is covered
3224 * by @map as split_flags indicates
3226 * It may result in splitting the extent into multiple extents (up to three)
3227 * There are three possibilities:
3228 * a> There is no split required
3229 * b> Splits in two extents: Split is happening at either end of the extent
3230 * c> Splits in three extents: Somone is splitting in middle of the extent
3233 static int ext4_split_extent(handle_t *handle,
3234 struct inode *inode,
3235 struct ext4_ext_path *path,
3236 struct ext4_map_blocks *map,
3240 ext4_lblk_t ee_block;
3241 struct ext4_extent *ex;
3242 unsigned int ee_len, depth;
3245 int split_flag1, flags1;
3246 int allocated = map->m_len;
3248 depth = ext_depth(inode);
3249 ex = path[depth].p_ext;
3250 ee_block = le32_to_cpu(ex->ee_block);
3251 ee_len = ext4_ext_get_actual_len(ex);
3252 uninitialized = ext4_ext_is_uninitialized(ex);
3254 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3255 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3256 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3258 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3259 EXT4_EXT_MARK_UNINIT2;
3260 if (split_flag & EXT4_EXT_DATA_VALID2)
3261 split_flag1 |= EXT4_EXT_DATA_VALID1;
3262 err = ext4_split_extent_at(handle, inode, path,
3263 map->m_lblk + map->m_len, split_flag1, flags1);
3267 allocated = ee_len - (map->m_lblk - ee_block);
3270 * Update path is required because previous ext4_split_extent_at() may
3271 * result in split of original leaf or extent zeroout.
3273 ext4_ext_drop_refs(path);
3274 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3276 return PTR_ERR(path);
3277 depth = ext_depth(inode);
3278 ex = path[depth].p_ext;
3279 uninitialized = ext4_ext_is_uninitialized(ex);
3282 if (map->m_lblk >= ee_block) {
3283 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3284 if (uninitialized) {
3285 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3286 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3287 EXT4_EXT_MARK_UNINIT2);
3289 err = ext4_split_extent_at(handle, inode, path,
3290 map->m_lblk, split_flag1, flags);
3295 ext4_ext_show_leaf(inode, path);
3297 return err ? err : allocated;
3301 * This function is called by ext4_ext_map_blocks() if someone tries to write
3302 * to an uninitialized extent. It may result in splitting the uninitialized
3303 * extent into multiple extents (up to three - one initialized and two
3305 * There are three possibilities:
3306 * a> There is no split required: Entire extent should be initialized
3307 * b> Splits in two extents: Write is happening at either end of the extent
3308 * c> Splits in three extents: Somone is writing in middle of the extent
3311 * - The extent pointed to by 'path' is uninitialized.
3312 * - The extent pointed to by 'path' contains a superset
3313 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3315 * Post-conditions on success:
3316 * - the returned value is the number of blocks beyond map->l_lblk
3317 * that are allocated and initialized.
3318 * It is guaranteed to be >= map->m_len.
3320 static int ext4_ext_convert_to_initialized(handle_t *handle,
3321 struct inode *inode,
3322 struct ext4_map_blocks *map,
3323 struct ext4_ext_path *path,
3326 struct ext4_sb_info *sbi;
3327 struct ext4_extent_header *eh;
3328 struct ext4_map_blocks split_map;
3329 struct ext4_extent zero_ex;
3330 struct ext4_extent *ex, *abut_ex;
3331 ext4_lblk_t ee_block, eof_block;
3332 unsigned int ee_len, depth, map_len = map->m_len;
3333 int allocated = 0, max_zeroout = 0;
3337 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3338 "block %llu, max_blocks %u\n", inode->i_ino,
3339 (unsigned long long)map->m_lblk, map_len);
3341 sbi = EXT4_SB(inode->i_sb);
3342 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3343 inode->i_sb->s_blocksize_bits;
3344 if (eof_block < map->m_lblk + map_len)
3345 eof_block = map->m_lblk + map_len;
3347 depth = ext_depth(inode);
3348 eh = path[depth].p_hdr;
3349 ex = path[depth].p_ext;
3350 ee_block = le32_to_cpu(ex->ee_block);
3351 ee_len = ext4_ext_get_actual_len(ex);
3354 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3356 /* Pre-conditions */
3357 BUG_ON(!ext4_ext_is_uninitialized(ex));
3358 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3361 * Attempt to transfer newly initialized blocks from the currently
3362 * uninitialized extent to its neighbor. This is much cheaper
3363 * than an insertion followed by a merge as those involve costly
3364 * memmove() calls. Transferring to the left is the common case in
3365 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3366 * followed by append writes.
3368 * Limitations of the current logic:
3369 * - L1: we do not deal with writes covering the whole extent.
3370 * This would require removing the extent if the transfer
3372 * - L2: we only attempt to merge with an extent stored in the
3373 * same extent tree node.
3375 if ((map->m_lblk == ee_block) &&
3376 /* See if we can merge left */
3377 (map_len < ee_len) && /*L1*/
3378 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3379 ext4_lblk_t prev_lblk;
3380 ext4_fsblk_t prev_pblk, ee_pblk;
3381 unsigned int prev_len;
3384 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3385 prev_len = ext4_ext_get_actual_len(abut_ex);
3386 prev_pblk = ext4_ext_pblock(abut_ex);
3387 ee_pblk = ext4_ext_pblock(ex);
3390 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3391 * upon those conditions:
3392 * - C1: abut_ex is initialized,
3393 * - C2: abut_ex is logically abutting ex,
3394 * - C3: abut_ex is physically abutting ex,
3395 * - C4: abut_ex can receive the additional blocks without
3396 * overflowing the (initialized) length limit.
3398 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3399 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3400 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3401 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3402 err = ext4_ext_get_access(handle, inode, path + depth);
3406 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3409 /* Shift the start of ex by 'map_len' blocks */
3410 ex->ee_block = cpu_to_le32(ee_block + map_len);
3411 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3412 ex->ee_len = cpu_to_le16(ee_len - map_len);
3413 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3415 /* Extend abut_ex by 'map_len' blocks */
3416 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3418 /* Result: number of initialized blocks past m_lblk */
3419 allocated = map_len;
3421 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3422 (map_len < ee_len) && /*L1*/
3423 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3424 /* See if we can merge right */
3425 ext4_lblk_t next_lblk;
3426 ext4_fsblk_t next_pblk, ee_pblk;
3427 unsigned int next_len;
3430 next_lblk = le32_to_cpu(abut_ex->ee_block);
3431 next_len = ext4_ext_get_actual_len(abut_ex);
3432 next_pblk = ext4_ext_pblock(abut_ex);
3433 ee_pblk = ext4_ext_pblock(ex);
3436 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3437 * upon those conditions:
3438 * - C1: abut_ex is initialized,
3439 * - C2: abut_ex is logically abutting ex,
3440 * - C3: abut_ex is physically abutting ex,
3441 * - C4: abut_ex can receive the additional blocks without
3442 * overflowing the (initialized) length limit.
3444 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3445 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3446 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3447 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3448 err = ext4_ext_get_access(handle, inode, path + depth);
3452 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3455 /* Shift the start of abut_ex by 'map_len' blocks */
3456 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3457 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3458 ex->ee_len = cpu_to_le16(ee_len - map_len);
3459 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3461 /* Extend abut_ex by 'map_len' blocks */
3462 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3464 /* Result: number of initialized blocks past m_lblk */
3465 allocated = map_len;
3469 /* Mark the block containing both extents as dirty */
3470 ext4_ext_dirty(handle, inode, path + depth);
3472 /* Update path to point to the right extent */
3473 path[depth].p_ext = abut_ex;
3476 allocated = ee_len - (map->m_lblk - ee_block);
3478 WARN_ON(map->m_lblk < ee_block);
3480 * It is safe to convert extent to initialized via explicit
3481 * zeroout only if extent is fully insde i_size or new_size.
3483 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3485 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3486 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3487 (inode->i_sb->s_blocksize_bits - 10);
3489 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3490 if (max_zeroout && (ee_len <= max_zeroout)) {
3491 err = ext4_ext_zeroout(inode, ex);
3494 zero_ex.ee_block = ex->ee_block;
3495 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3496 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3498 err = ext4_ext_get_access(handle, inode, path + depth);
3501 ext4_ext_mark_initialized(ex);
3502 ext4_ext_try_to_merge(handle, inode, path, ex);
3503 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3509 * 1. split the extent into three extents.
3510 * 2. split the extent into two extents, zeroout the first half.
3511 * 3. split the extent into two extents, zeroout the second half.
3512 * 4. split the extent into two extents with out zeroout.
3514 split_map.m_lblk = map->m_lblk;
3515 split_map.m_len = map->m_len;
3517 if (max_zeroout && (allocated > map->m_len)) {
3518 if (allocated <= max_zeroout) {
3521 cpu_to_le32(map->m_lblk);
3522 zero_ex.ee_len = cpu_to_le16(allocated);
3523 ext4_ext_store_pblock(&zero_ex,
3524 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3525 err = ext4_ext_zeroout(inode, &zero_ex);
3528 split_map.m_lblk = map->m_lblk;
3529 split_map.m_len = allocated;
3530 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3532 if (map->m_lblk != ee_block) {
3533 zero_ex.ee_block = ex->ee_block;
3534 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3536 ext4_ext_store_pblock(&zero_ex,
3537 ext4_ext_pblock(ex));
3538 err = ext4_ext_zeroout(inode, &zero_ex);
3543 split_map.m_lblk = ee_block;
3544 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3545 allocated = map->m_len;
3549 allocated = ext4_split_extent(handle, inode, path,
3550 &split_map, split_flag, flags);
3555 /* If we have gotten a failure, don't zero out status tree */
3557 err = ext4_zeroout_es(inode, &zero_ex);
3558 return err ? err : allocated;
3562 * This function is called by ext4_ext_map_blocks() from
3563 * ext4_get_blocks_dio_write() when DIO to write
3564 * to an uninitialized extent.
3566 * Writing to an uninitialized extent may result in splitting the uninitialized
3567 * extent into multiple initialized/uninitialized extents (up to three)
3568 * There are three possibilities:
3569 * a> There is no split required: Entire extent should be uninitialized
3570 * b> Splits in two extents: Write is happening at either end of the extent
3571 * c> Splits in three extents: Somone is writing in middle of the extent
3573 * One of more index blocks maybe needed if the extent tree grow after
3574 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3575 * complete, we need to split the uninitialized extent before DIO submit
3576 * the IO. The uninitialized extent called at this time will be split
3577 * into three uninitialized extent(at most). After IO complete, the part
3578 * being filled will be convert to initialized by the end_io callback function
3579 * via ext4_convert_unwritten_extents().
3581 * Returns the size of uninitialized extent to be written on success.
3583 static int ext4_split_unwritten_extents(handle_t *handle,
3584 struct inode *inode,
3585 struct ext4_map_blocks *map,
3586 struct ext4_ext_path *path,
3589 ext4_lblk_t eof_block;
3590 ext4_lblk_t ee_block;
3591 struct ext4_extent *ex;
3592 unsigned int ee_len;
3593 int split_flag = 0, depth;
3595 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3596 "block %llu, max_blocks %u\n", inode->i_ino,
3597 (unsigned long long)map->m_lblk, map->m_len);
3599 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3600 inode->i_sb->s_blocksize_bits;
3601 if (eof_block < map->m_lblk + map->m_len)
3602 eof_block = map->m_lblk + map->m_len;
3604 * It is safe to convert extent to initialized via explicit
3605 * zeroout only if extent is fully insde i_size or new_size.
3607 depth = ext_depth(inode);
3608 ex = path[depth].p_ext;
3609 ee_block = le32_to_cpu(ex->ee_block);
3610 ee_len = ext4_ext_get_actual_len(ex);
3612 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3613 split_flag |= EXT4_EXT_MARK_UNINIT2;
3614 if (flags & EXT4_GET_BLOCKS_CONVERT)
3615 split_flag |= EXT4_EXT_DATA_VALID2;
3616 flags |= EXT4_GET_BLOCKS_PRE_IO;
3617 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3620 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3621 struct inode *inode,
3622 struct ext4_map_blocks *map,
3623 struct ext4_ext_path *path)
3625 struct ext4_extent *ex;
3626 ext4_lblk_t ee_block;
3627 unsigned int ee_len;
3631 depth = ext_depth(inode);
3632 ex = path[depth].p_ext;
3633 ee_block = le32_to_cpu(ex->ee_block);
3634 ee_len = ext4_ext_get_actual_len(ex);
3636 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3637 "block %llu, max_blocks %u\n", inode->i_ino,
3638 (unsigned long long)ee_block, ee_len);
3640 /* If extent is larger than requested it is a clear sign that we still
3641 * have some extent state machine issues left. So extent_split is still
3643 * TODO: Once all related issues will be fixed this situation should be
3646 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3648 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3649 " len %u; IO logical block %llu, len %u\n",
3650 inode->i_ino, (unsigned long long)ee_block, ee_len,
3651 (unsigned long long)map->m_lblk, map->m_len);
3653 err = ext4_split_unwritten_extents(handle, inode, map, path,
3654 EXT4_GET_BLOCKS_CONVERT);
3657 ext4_ext_drop_refs(path);
3658 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3660 err = PTR_ERR(path);
3663 depth = ext_depth(inode);
3664 ex = path[depth].p_ext;
3667 err = ext4_ext_get_access(handle, inode, path + depth);
3670 /* first mark the extent as initialized */
3671 ext4_ext_mark_initialized(ex);
3673 /* note: ext4_ext_correct_indexes() isn't needed here because
3674 * borders are not changed
3676 ext4_ext_try_to_merge(handle, inode, path, ex);
3678 /* Mark modified extent as dirty */
3679 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3681 ext4_ext_show_leaf(inode, path);
3685 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3686 sector_t block, int count)
3689 for (i = 0; i < count; i++)
3690 unmap_underlying_metadata(bdev, block + i);
3694 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3696 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3698 struct ext4_ext_path *path,
3702 struct ext4_extent_header *eh;
3703 struct ext4_extent *last_ex;
3705 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3708 depth = ext_depth(inode);
3709 eh = path[depth].p_hdr;
3712 * We're going to remove EOFBLOCKS_FL entirely in future so we
3713 * do not care for this case anymore. Simply remove the flag
3714 * if there are no extents.
3716 if (unlikely(!eh->eh_entries))
3718 last_ex = EXT_LAST_EXTENT(eh);
3720 * We should clear the EOFBLOCKS_FL flag if we are writing the
3721 * last block in the last extent in the file. We test this by
3722 * first checking to see if the caller to
3723 * ext4_ext_get_blocks() was interested in the last block (or
3724 * a block beyond the last block) in the current extent. If
3725 * this turns out to be false, we can bail out from this
3726 * function immediately.
3728 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3729 ext4_ext_get_actual_len(last_ex))
3732 * If the caller does appear to be planning to write at or
3733 * beyond the end of the current extent, we then test to see
3734 * if the current extent is the last extent in the file, by
3735 * checking to make sure it was reached via the rightmost node
3736 * at each level of the tree.
3738 for (i = depth-1; i >= 0; i--)
3739 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3742 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3743 return ext4_mark_inode_dirty(handle, inode);
3747 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3749 * Return 1 if there is a delalloc block in the range, otherwise 0.
3751 int ext4_find_delalloc_range(struct inode *inode,
3752 ext4_lblk_t lblk_start,
3753 ext4_lblk_t lblk_end)
3755 struct extent_status es;
3757 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3759 return 0; /* there is no delay extent in this tree */
3760 else if (es.es_lblk <= lblk_start &&
3761 lblk_start < es.es_lblk + es.es_len)
3763 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3769 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3771 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3772 ext4_lblk_t lblk_start, lblk_end;
3773 lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3774 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3776 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3780 * Determines how many complete clusters (out of those specified by the 'map')
3781 * are under delalloc and were reserved quota for.
3782 * This function is called when we are writing out the blocks that were
3783 * originally written with their allocation delayed, but then the space was
3784 * allocated using fallocate() before the delayed allocation could be resolved.
3785 * The cases to look for are:
3786 * ('=' indicated delayed allocated blocks
3787 * '-' indicates non-delayed allocated blocks)
3788 * (a) partial clusters towards beginning and/or end outside of allocated range
3789 * are not delalloc'ed.
3791 * |----c---=|====c====|====c====|===-c----|
3792 * |++++++ allocated ++++++|
3793 * ==> 4 complete clusters in above example
3795 * (b) partial cluster (outside of allocated range) towards either end is
3796 * marked for delayed allocation. In this case, we will exclude that
3799 * |----====c========|========c========|
3800 * |++++++ allocated ++++++|
3801 * ==> 1 complete clusters in above example
3804 * |================c================|
3805 * |++++++ allocated ++++++|
3806 * ==> 0 complete clusters in above example
3808 * The ext4_da_update_reserve_space will be called only if we
3809 * determine here that there were some "entire" clusters that span
3810 * this 'allocated' range.
3811 * In the non-bigalloc case, this function will just end up returning num_blks
3812 * without ever calling ext4_find_delalloc_range.
3815 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3816 unsigned int num_blks)
3818 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3819 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3820 ext4_lblk_t lblk_from, lblk_to, c_offset;
3821 unsigned int allocated_clusters = 0;
3823 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3824 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3826 /* max possible clusters for this allocation */
3827 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3829 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3831 /* Check towards left side */
3832 c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3834 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3835 lblk_to = lblk_from + c_offset - 1;
3837 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3838 allocated_clusters--;
3841 /* Now check towards right. */
3842 c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3843 if (allocated_clusters && c_offset) {
3844 lblk_from = lblk_start + num_blks;
3845 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3847 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3848 allocated_clusters--;
3851 return allocated_clusters;
3855 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3856 struct ext4_map_blocks *map,
3857 struct ext4_ext_path *path, int flags,
3858 unsigned int allocated, ext4_fsblk_t newblock)
3862 ext4_io_end_t *io = ext4_inode_aio(inode);
3864 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3865 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3866 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3868 ext4_ext_show_leaf(inode, path);
3871 * When writing into uninitialized space, we should not fail to
3872 * allocate metadata blocks for the new extent block if needed.
3874 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3876 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3877 allocated, newblock);
3879 /* get_block() before submit the IO, split the extent */
3880 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3881 ret = ext4_split_unwritten_extents(handle, inode, map,
3886 * Flag the inode(non aio case) or end_io struct (aio case)
3887 * that this IO needs to conversion to written when IO is
3891 ext4_set_io_unwritten_flag(inode, io);
3893 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3894 map->m_flags |= EXT4_MAP_UNWRITTEN;
3895 if (ext4_should_dioread_nolock(inode))
3896 map->m_flags |= EXT4_MAP_UNINIT;
3899 /* IO end_io complete, convert the filled extent to written */
3900 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3901 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3904 ext4_update_inode_fsync_trans(handle, inode, 1);
3905 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3909 map->m_flags |= EXT4_MAP_MAPPED;
3910 if (allocated > map->m_len)
3911 allocated = map->m_len;
3912 map->m_len = allocated;
3915 /* buffered IO case */
3917 * repeat fallocate creation request
3918 * we already have an unwritten extent
3920 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
3921 map->m_flags |= EXT4_MAP_UNWRITTEN;
3925 /* buffered READ or buffered write_begin() lookup */
3926 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3928 * We have blocks reserved already. We
3929 * return allocated blocks so that delalloc
3930 * won't do block reservation for us. But
3931 * the buffer head will be unmapped so that
3932 * a read from the block returns 0s.
3934 map->m_flags |= EXT4_MAP_UNWRITTEN;
3938 /* buffered write, writepage time, convert*/
3939 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
3941 ext4_update_inode_fsync_trans(handle, inode, 1);
3948 map->m_flags |= EXT4_MAP_NEW;
3950 * if we allocated more blocks than requested
3951 * we need to make sure we unmap the extra block
3952 * allocated. The actual needed block will get
3953 * unmapped later when we find the buffer_head marked
3956 if (allocated > map->m_len) {
3957 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3958 newblock + map->m_len,
3959 allocated - map->m_len);
3960 allocated = map->m_len;
3962 map->m_len = allocated;
3965 * If we have done fallocate with the offset that is already
3966 * delayed allocated, we would have block reservation
3967 * and quota reservation done in the delayed write path.
3968 * But fallocate would have already updated quota and block
3969 * count for this offset. So cancel these reservation
3971 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3972 unsigned int reserved_clusters;
3973 reserved_clusters = get_reserved_cluster_alloc(inode,
3974 map->m_lblk, map->m_len);
3975 if (reserved_clusters)
3976 ext4_da_update_reserve_space(inode,
3982 map->m_flags |= EXT4_MAP_MAPPED;
3983 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3984 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3990 if (allocated > map->m_len)
3991 allocated = map->m_len;
3992 ext4_ext_show_leaf(inode, path);
3993 map->m_pblk = newblock;
3994 map->m_len = allocated;
3997 ext4_ext_drop_refs(path);
4000 return err ? err : allocated;
4004 * get_implied_cluster_alloc - check to see if the requested
4005 * allocation (in the map structure) overlaps with a cluster already
4006 * allocated in an extent.
4007 * @sb The filesystem superblock structure
4008 * @map The requested lblk->pblk mapping
4009 * @ex The extent structure which might contain an implied
4010 * cluster allocation
4012 * This function is called by ext4_ext_map_blocks() after we failed to
4013 * find blocks that were already in the inode's extent tree. Hence,
4014 * we know that the beginning of the requested region cannot overlap
4015 * the extent from the inode's extent tree. There are three cases we
4016 * want to catch. The first is this case:
4018 * |--- cluster # N--|
4019 * |--- extent ---| |---- requested region ---|
4022 * The second case that we need to test for is this one:
4024 * |--------- cluster # N ----------------|
4025 * |--- requested region --| |------- extent ----|
4026 * |=======================|
4028 * The third case is when the requested region lies between two extents
4029 * within the same cluster:
4030 * |------------- cluster # N-------------|
4031 * |----- ex -----| |---- ex_right ----|
4032 * |------ requested region ------|
4033 * |================|
4035 * In each of the above cases, we need to set the map->m_pblk and
4036 * map->m_len so it corresponds to the return the extent labelled as
4037 * "|====|" from cluster #N, since it is already in use for data in
4038 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4039 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4040 * as a new "allocated" block region. Otherwise, we will return 0 and
4041 * ext4_ext_map_blocks() will then allocate one or more new clusters
4042 * by calling ext4_mb_new_blocks().
4044 static int get_implied_cluster_alloc(struct super_block *sb,
4045 struct ext4_map_blocks *map,
4046 struct ext4_extent *ex,
4047 struct ext4_ext_path *path)
4049 struct ext4_sb_info *sbi = EXT4_SB(sb);
4050 ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4051 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4052 ext4_lblk_t rr_cluster_start;
4053 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4054 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4055 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4057 /* The extent passed in that we are trying to match */
4058 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4059 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4061 /* The requested region passed into ext4_map_blocks() */
4062 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4064 if ((rr_cluster_start == ex_cluster_end) ||
4065 (rr_cluster_start == ex_cluster_start)) {
4066 if (rr_cluster_start == ex_cluster_end)
4067 ee_start += ee_len - 1;
4068 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
4070 map->m_len = min(map->m_len,
4071 (unsigned) sbi->s_cluster_ratio - c_offset);
4073 * Check for and handle this case:
4075 * |--------- cluster # N-------------|
4076 * |------- extent ----|
4077 * |--- requested region ---|
4081 if (map->m_lblk < ee_block)
4082 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4085 * Check for the case where there is already another allocated
4086 * block to the right of 'ex' but before the end of the cluster.
4088 * |------------- cluster # N-------------|
4089 * |----- ex -----| |---- ex_right ----|
4090 * |------ requested region ------|
4091 * |================|
4093 if (map->m_lblk > ee_block) {
4094 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4095 map->m_len = min(map->m_len, next - map->m_lblk);
4098 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4102 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4108 * Block allocation/map/preallocation routine for extents based files
4111 * Need to be called with
4112 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4113 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4115 * return > 0, number of of blocks already mapped/allocated
4116 * if create == 0 and these are pre-allocated blocks
4117 * buffer head is unmapped
4118 * otherwise blocks are mapped
4120 * return = 0, if plain look up failed (blocks have not been allocated)
4121 * buffer head is unmapped
4123 * return < 0, error case.
4125 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4126 struct ext4_map_blocks *map, int flags)
4128 struct ext4_ext_path *path = NULL;
4129 struct ext4_extent newex, *ex, *ex2;
4130 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4131 ext4_fsblk_t newblock = 0;
4132 int free_on_err = 0, err = 0, depth;
4133 unsigned int allocated = 0, offset = 0;
4134 unsigned int allocated_clusters = 0;
4135 struct ext4_allocation_request ar;
4136 ext4_io_end_t *io = ext4_inode_aio(inode);
4137 ext4_lblk_t cluster_offset;
4138 int set_unwritten = 0;
4140 ext_debug("blocks %u/%u requested for inode %lu\n",
4141 map->m_lblk, map->m_len, inode->i_ino);
4142 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4144 /* find extent for this block */
4145 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4147 err = PTR_ERR(path);
4152 depth = ext_depth(inode);
4155 * consistent leaf must not be empty;
4156 * this situation is possible, though, _during_ tree modification;
4157 * this is why assert can't be put in ext4_ext_find_extent()
4159 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4160 EXT4_ERROR_INODE(inode, "bad extent address "
4161 "lblock: %lu, depth: %d pblock %lld",
4162 (unsigned long) map->m_lblk, depth,
4163 path[depth].p_block);
4168 ex = path[depth].p_ext;
4170 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4171 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4172 unsigned short ee_len;
4175 * Uninitialized extents are treated as holes, except that
4176 * we split out initialized portions during a write.
4178 ee_len = ext4_ext_get_actual_len(ex);
4180 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4182 /* if found extent covers block, simply return it */
4183 if (in_range(map->m_lblk, ee_block, ee_len)) {
4184 newblock = map->m_lblk - ee_block + ee_start;
4185 /* number of remaining blocks in the extent */
4186 allocated = ee_len - (map->m_lblk - ee_block);
4187 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4188 ee_block, ee_len, newblock);
4190 if (!ext4_ext_is_uninitialized(ex))
4193 allocated = ext4_ext_handle_uninitialized_extents(
4194 handle, inode, map, path, flags,
4195 allocated, newblock);
4200 if ((sbi->s_cluster_ratio > 1) &&
4201 ext4_find_delalloc_cluster(inode, map->m_lblk))
4202 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4205 * requested block isn't allocated yet;
4206 * we couldn't try to create block if create flag is zero
4208 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4210 * put just found gap into cache to speed up
4211 * subsequent requests
4213 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4214 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4219 * Okay, we need to do block allocation.
4221 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4222 newex.ee_block = cpu_to_le32(map->m_lblk);
4223 cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4226 * If we are doing bigalloc, check to see if the extent returned
4227 * by ext4_ext_find_extent() implies a cluster we can use.
4229 if (cluster_offset && ex &&
4230 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4231 ar.len = allocated = map->m_len;
4232 newblock = map->m_pblk;
4233 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4234 goto got_allocated_blocks;
4237 /* find neighbour allocated blocks */
4238 ar.lleft = map->m_lblk;
4239 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4242 ar.lright = map->m_lblk;
4244 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4248 /* Check if the extent after searching to the right implies a
4249 * cluster we can use. */
4250 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4251 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4252 ar.len = allocated = map->m_len;
4253 newblock = map->m_pblk;
4254 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4255 goto got_allocated_blocks;
4259 * See if request is beyond maximum number of blocks we can have in
4260 * a single extent. For an initialized extent this limit is
4261 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4262 * EXT_UNINIT_MAX_LEN.
4264 if (map->m_len > EXT_INIT_MAX_LEN &&
4265 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4266 map->m_len = EXT_INIT_MAX_LEN;
4267 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4268 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4269 map->m_len = EXT_UNINIT_MAX_LEN;
4271 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4272 newex.ee_len = cpu_to_le16(map->m_len);
4273 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4275 allocated = ext4_ext_get_actual_len(&newex);
4277 allocated = map->m_len;
4279 /* allocate new block */
4281 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4282 ar.logical = map->m_lblk;
4284 * We calculate the offset from the beginning of the cluster
4285 * for the logical block number, since when we allocate a
4286 * physical cluster, the physical block should start at the
4287 * same offset from the beginning of the cluster. This is
4288 * needed so that future calls to get_implied_cluster_alloc()
4291 offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4292 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4294 ar.logical -= offset;
4295 if (S_ISREG(inode->i_mode))
4296 ar.flags = EXT4_MB_HINT_DATA;
4298 /* disable in-core preallocation for non-regular files */
4300 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4301 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4302 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4305 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4306 ar.goal, newblock, allocated);
4308 allocated_clusters = ar.len;
4309 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4310 if (ar.len > allocated)
4313 got_allocated_blocks:
4314 /* try to insert new extent into found leaf and return */
4315 ext4_ext_store_pblock(&newex, newblock + offset);
4316 newex.ee_len = cpu_to_le16(ar.len);
4317 /* Mark uninitialized */
4318 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4319 ext4_ext_mark_uninitialized(&newex);
4320 map->m_flags |= EXT4_MAP_UNWRITTEN;
4322 * io_end structure was created for every IO write to an
4323 * uninitialized extent. To avoid unnecessary conversion,
4324 * here we flag the IO that really needs the conversion.
4325 * For non asycn direct IO case, flag the inode state
4326 * that we need to perform conversion when IO is done.
4328 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4330 if (ext4_should_dioread_nolock(inode))
4331 map->m_flags |= EXT4_MAP_UNINIT;
4335 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4336 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4339 err = ext4_ext_insert_extent(handle, inode, path,
4342 if (!err && set_unwritten) {
4344 ext4_set_io_unwritten_flag(inode, io);
4346 ext4_set_inode_state(inode,
4347 EXT4_STATE_DIO_UNWRITTEN);
4350 if (err && free_on_err) {
4351 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4352 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4353 /* free data blocks we just allocated */
4354 /* not a good idea to call discard here directly,
4355 * but otherwise we'd need to call it every free() */
4356 ext4_discard_preallocations(inode);
4357 ext4_free_blocks(handle, inode, NULL, newblock,
4358 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4362 /* previous routine could use block we allocated */
4363 newblock = ext4_ext_pblock(&newex);
4364 allocated = ext4_ext_get_actual_len(&newex);
4365 if (allocated > map->m_len)
4366 allocated = map->m_len;
4367 map->m_flags |= EXT4_MAP_NEW;
4370 * Update reserved blocks/metadata blocks after successful
4371 * block allocation which had been deferred till now.
4373 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4374 unsigned int reserved_clusters;
4376 * Check how many clusters we had reserved this allocated range
4378 reserved_clusters = get_reserved_cluster_alloc(inode,
4379 map->m_lblk, allocated);
4380 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4381 if (reserved_clusters) {
4383 * We have clusters reserved for this range.
4384 * But since we are not doing actual allocation
4385 * and are simply using blocks from previously
4386 * allocated cluster, we should release the
4387 * reservation and not claim quota.
4389 ext4_da_update_reserve_space(inode,
4390 reserved_clusters, 0);
4393 BUG_ON(allocated_clusters < reserved_clusters);
4394 if (reserved_clusters < allocated_clusters) {
4395 struct ext4_inode_info *ei = EXT4_I(inode);
4396 int reservation = allocated_clusters -
4399 * It seems we claimed few clusters outside of
4400 * the range of this allocation. We should give
4401 * it back to the reservation pool. This can
4402 * happen in the following case:
4404 * * Suppose s_cluster_ratio is 4 (i.e., each
4405 * cluster has 4 blocks. Thus, the clusters
4406 * are [0-3],[4-7],[8-11]...
4407 * * First comes delayed allocation write for
4408 * logical blocks 10 & 11. Since there were no
4409 * previous delayed allocated blocks in the
4410 * range [8-11], we would reserve 1 cluster
4412 * * Next comes write for logical blocks 3 to 8.
4413 * In this case, we will reserve 2 clusters
4414 * (for [0-3] and [4-7]; and not for [8-11] as
4415 * that range has a delayed allocated blocks.
4416 * Thus total reserved clusters now becomes 3.
4417 * * Now, during the delayed allocation writeout
4418 * time, we will first write blocks [3-8] and
4419 * allocate 3 clusters for writing these
4420 * blocks. Also, we would claim all these
4421 * three clusters above.
4422 * * Now when we come here to writeout the
4423 * blocks [10-11], we would expect to claim
4424 * the reservation of 1 cluster we had made
4425 * (and we would claim it since there are no
4426 * more delayed allocated blocks in the range
4427 * [8-11]. But our reserved cluster count had
4428 * already gone to 0.
4430 * Thus, at the step 4 above when we determine
4431 * that there are still some unwritten delayed
4432 * allocated blocks outside of our current
4433 * block range, we should increment the
4434 * reserved clusters count so that when the
4435 * remaining blocks finally gets written, we
4438 dquot_reserve_block(inode,
4439 EXT4_C2B(sbi, reservation));
4440 spin_lock(&ei->i_block_reservation_lock);
4441 ei->i_reserved_data_blocks += reservation;
4442 spin_unlock(&ei->i_block_reservation_lock);
4445 * We will claim quota for all newly allocated blocks.
4446 * We're updating the reserved space *after* the
4447 * correction above so we do not accidentally free
4448 * all the metadata reservation because we might
4449 * actually need it later on.
4451 ext4_da_update_reserve_space(inode, allocated_clusters,
4457 * Cache the extent and update transaction to commit on fdatasync only
4458 * when it is _not_ an uninitialized extent.
4460 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
4461 ext4_update_inode_fsync_trans(handle, inode, 1);
4463 ext4_update_inode_fsync_trans(handle, inode, 0);
4465 if (allocated > map->m_len)
4466 allocated = map->m_len;
4467 ext4_ext_show_leaf(inode, path);
4468 map->m_flags |= EXT4_MAP_MAPPED;
4469 map->m_pblk = newblock;
4470 map->m_len = allocated;
4473 ext4_ext_drop_refs(path);
4478 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4479 err ? err : allocated);
4480 ext4_es_lru_add(inode);
4481 return err ? err : allocated;
4484 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4486 struct super_block *sb = inode->i_sb;
4487 ext4_lblk_t last_block;
4491 * TODO: optimization is possible here.
4492 * Probably we need not scan at all,
4493 * because page truncation is enough.
4496 /* we have to know where to truncate from in crash case */
4497 EXT4_I(inode)->i_disksize = inode->i_size;
4498 ext4_mark_inode_dirty(handle, inode);
4500 last_block = (inode->i_size + sb->s_blocksize - 1)
4501 >> EXT4_BLOCK_SIZE_BITS(sb);
4503 err = ext4_es_remove_extent(inode, last_block,
4504 EXT_MAX_BLOCKS - last_block);
4505 if (err == -ENOMEM) {
4507 congestion_wait(BLK_RW_ASYNC, HZ/50);
4511 ext4_std_error(inode->i_sb, err);
4514 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4515 ext4_std_error(inode->i_sb, err);
4518 static void ext4_falloc_update_inode(struct inode *inode,
4519 int mode, loff_t new_size, int update_ctime)
4521 struct timespec now;
4524 now = current_fs_time(inode->i_sb);
4525 if (!timespec_equal(&inode->i_ctime, &now))
4526 inode->i_ctime = now;
4529 * Update only when preallocation was requested beyond
4532 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4533 if (new_size > i_size_read(inode))
4534 i_size_write(inode, new_size);
4535 if (new_size > EXT4_I(inode)->i_disksize)
4536 ext4_update_i_disksize(inode, new_size);
4539 * Mark that we allocate beyond EOF so the subsequent truncate
4540 * can proceed even if the new size is the same as i_size.
4542 if (new_size > i_size_read(inode))
4543 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4549 * preallocate space for a file. This implements ext4's fallocate file
4550 * operation, which gets called from sys_fallocate system call.
4551 * For block-mapped files, posix_fallocate should fall back to the method
4552 * of writing zeroes to the required new blocks (the same behavior which is
4553 * expected for file systems which do not support fallocate() system call).
4555 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4557 struct inode *inode = file_inode(file);
4560 unsigned int max_blocks;
4565 struct ext4_map_blocks map;
4566 unsigned int credits, blkbits = inode->i_blkbits;
4568 /* Return error if mode is not supported */
4569 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4572 if (mode & FALLOC_FL_PUNCH_HOLE)
4573 return ext4_punch_hole(inode, offset, len);
4575 ret = ext4_convert_inline_data(inode);
4580 * currently supporting (pre)allocate mode for extent-based
4583 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4586 trace_ext4_fallocate_enter(inode, offset, len, mode);
4587 map.m_lblk = offset >> blkbits;
4589 * We can't just convert len to max_blocks because
4590 * If blocksize = 4096 offset = 3072 and len = 2048
4592 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4595 * credits to insert 1 extent into extent tree
4597 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4598 mutex_lock(&inode->i_mutex);
4599 ret = inode_newsize_ok(inode, (len + offset));
4601 mutex_unlock(&inode->i_mutex);
4602 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4605 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4606 if (mode & FALLOC_FL_KEEP_SIZE)
4607 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4609 * Don't normalize the request if it can fit in one extent so
4610 * that it doesn't get unnecessarily split into multiple
4613 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4614 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4617 while (ret >= 0 && ret < max_blocks) {
4618 map.m_lblk = map.m_lblk + ret;
4619 map.m_len = max_blocks = max_blocks - ret;
4620 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4622 if (IS_ERR(handle)) {
4623 ret = PTR_ERR(handle);
4626 ret = ext4_map_blocks(handle, inode, &map, flags);
4629 ext4_warning(inode->i_sb,
4630 "inode #%lu: block %u: len %u: "
4631 "ext4_ext_map_blocks returned %d",
4632 inode->i_ino, map.m_lblk,
4635 ext4_mark_inode_dirty(handle, inode);
4636 ret2 = ext4_journal_stop(handle);
4639 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4640 blkbits) >> blkbits))
4641 new_size = offset + len;
4643 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4645 ext4_falloc_update_inode(inode, mode, new_size,
4646 (map.m_flags & EXT4_MAP_NEW));
4647 ext4_mark_inode_dirty(handle, inode);
4648 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4649 ext4_handle_sync(handle);
4650 ret2 = ext4_journal_stop(handle);
4654 if (ret == -ENOSPC &&
4655 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4659 mutex_unlock(&inode->i_mutex);
4660 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4661 ret > 0 ? ret2 : ret);
4662 return ret > 0 ? ret2 : ret;
4666 * This function convert a range of blocks to written extents
4667 * The caller of this function will pass the start offset and the size.
4668 * all unwritten extents within this range will be converted to
4671 * This function is called from the direct IO end io call back
4672 * function, to convert the fallocated extents after IO is completed.
4673 * Returns 0 on success.
4675 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4676 loff_t offset, ssize_t len)
4678 unsigned int max_blocks;
4681 struct ext4_map_blocks map;
4682 unsigned int credits, blkbits = inode->i_blkbits;
4684 map.m_lblk = offset >> blkbits;
4686 * We can't just convert len to max_blocks because
4687 * If blocksize = 4096 offset = 3072 and len = 2048
4689 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4692 * This is somewhat ugly but the idea is clear: When transaction is
4693 * reserved, everything goes into it. Otherwise we rather start several
4694 * smaller transactions for conversion of each extent separately.
4697 handle = ext4_journal_start_reserved(handle,
4698 EXT4_HT_EXT_CONVERT);
4700 return PTR_ERR(handle);
4704 * credits to insert 1 extent into extent tree
4706 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4708 while (ret >= 0 && ret < max_blocks) {
4710 map.m_len = (max_blocks -= ret);
4712 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4714 if (IS_ERR(handle)) {
4715 ret = PTR_ERR(handle);
4719 ret = ext4_map_blocks(handle, inode, &map,
4720 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4722 ext4_warning(inode->i_sb,
4723 "inode #%lu: block %u: len %u: "
4724 "ext4_ext_map_blocks returned %d",
4725 inode->i_ino, map.m_lblk,
4727 ext4_mark_inode_dirty(handle, inode);
4729 ret2 = ext4_journal_stop(handle);
4730 if (ret <= 0 || ret2)
4734 ret2 = ext4_journal_stop(handle);
4735 return ret > 0 ? ret2 : ret;
4739 * If newes is not existing extent (newes->ec_pblk equals zero) find
4740 * delayed extent at start of newes and update newes accordingly and
4741 * return start of the next delayed extent.
4743 * If newes is existing extent (newes->ec_pblk is not equal zero)
4744 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4745 * extent found. Leave newes unmodified.
4747 static int ext4_find_delayed_extent(struct inode *inode,
4748 struct extent_status *newes)
4750 struct extent_status es;
4751 ext4_lblk_t block, next_del;
4753 if (newes->es_pblk == 0) {
4754 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
4755 newes->es_lblk + newes->es_len - 1, &es);
4758 * No extent in extent-tree contains block @newes->es_pblk,
4759 * then the block may stay in 1)a hole or 2)delayed-extent.
4765 if (es.es_lblk > newes->es_lblk) {
4767 newes->es_len = min(es.es_lblk - newes->es_lblk,
4772 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
4775 block = newes->es_lblk + newes->es_len;
4776 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
4778 next_del = EXT_MAX_BLOCKS;
4780 next_del = es.es_lblk;
4784 /* fiemap flags we can handle specified here */
4785 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4787 static int ext4_xattr_fiemap(struct inode *inode,
4788 struct fiemap_extent_info *fieinfo)
4792 __u32 flags = FIEMAP_EXTENT_LAST;
4793 int blockbits = inode->i_sb->s_blocksize_bits;
4797 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4798 struct ext4_iloc iloc;
4799 int offset; /* offset of xattr in inode */
4801 error = ext4_get_inode_loc(inode, &iloc);
4804 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4805 offset = EXT4_GOOD_OLD_INODE_SIZE +
4806 EXT4_I(inode)->i_extra_isize;
4808 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4809 flags |= FIEMAP_EXTENT_DATA_INLINE;
4811 } else { /* external block */
4812 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4813 length = inode->i_sb->s_blocksize;
4817 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4819 return (error < 0 ? error : 0);
4822 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4823 __u64 start, __u64 len)
4825 ext4_lblk_t start_blk;
4828 if (ext4_has_inline_data(inode)) {
4831 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4837 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4838 error = ext4_ext_precache(inode);
4843 /* fallback to generic here if not in extents fmt */
4844 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4845 return generic_block_fiemap(inode, fieinfo, start, len,
4848 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4851 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4852 error = ext4_xattr_fiemap(inode, fieinfo);
4854 ext4_lblk_t len_blks;
4857 start_blk = start >> inode->i_sb->s_blocksize_bits;
4858 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4859 if (last_blk >= EXT_MAX_BLOCKS)
4860 last_blk = EXT_MAX_BLOCKS-1;
4861 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4864 * Walk the extent tree gathering extent information
4865 * and pushing extents back to the user.
4867 error = ext4_fill_fiemap_extents(inode, start_blk,
4870 ext4_es_lru_add(inode);