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_metadata_csum(inode->i_sb))
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
86 static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
89 struct ext4_extent_tail *et;
91 if (!ext4_has_metadata_csum(inode->i_sb))
94 et = find_ext4_extent_tail(eh);
95 et->et_checksum = ext4_extent_block_csum(inode, eh);
98 static int ext4_split_extent(handle_t *handle,
100 struct ext4_ext_path *path,
101 struct ext4_map_blocks *map,
105 static int ext4_split_extent_at(handle_t *handle,
107 struct ext4_ext_path *path,
112 static int ext4_find_delayed_extent(struct inode *inode,
113 struct extent_status *newes);
115 static int ext4_ext_truncate_extend_restart(handle_t *handle,
121 if (!ext4_handle_valid(handle))
123 if (handle->h_buffer_credits > needed)
125 err = ext4_journal_extend(handle, needed);
128 err = ext4_truncate_restart_trans(handle, inode, needed);
140 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
141 struct ext4_ext_path *path)
144 /* path points to block */
145 return ext4_journal_get_write_access(handle, path->p_bh);
147 /* path points to leaf/index in inode body */
148 /* we use in-core data, no need to protect them */
158 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
159 struct inode *inode, struct ext4_ext_path *path)
163 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
164 /* path points to block */
165 err = __ext4_handle_dirty_metadata(where, line, handle,
168 /* path points to leaf/index in inode body */
169 err = ext4_mark_inode_dirty(handle, inode);
174 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
175 struct ext4_ext_path *path,
179 int depth = path->p_depth;
180 struct ext4_extent *ex;
183 * Try to predict block placement assuming that we are
184 * filling in a file which will eventually be
185 * non-sparse --- i.e., in the case of libbfd writing
186 * an ELF object sections out-of-order but in a way
187 * the eventually results in a contiguous object or
188 * executable file, or some database extending a table
189 * space file. However, this is actually somewhat
190 * non-ideal if we are writing a sparse file such as
191 * qemu or KVM writing a raw image file that is going
192 * to stay fairly sparse, since it will end up
193 * fragmenting the file system's free space. Maybe we
194 * should have some hueristics or some way to allow
195 * userspace to pass a hint to file system,
196 * especially if the latter case turns out to be
199 ex = path[depth].p_ext;
201 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
202 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
204 if (block > ext_block)
205 return ext_pblk + (block - ext_block);
207 return ext_pblk - (ext_block - block);
210 /* it looks like index is empty;
211 * try to find starting block from index itself */
212 if (path[depth].p_bh)
213 return path[depth].p_bh->b_blocknr;
216 /* OK. use inode's group */
217 return ext4_inode_to_goal_block(inode);
221 * Allocation for a meta data block
224 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
225 struct ext4_ext_path *path,
226 struct ext4_extent *ex, int *err, unsigned int flags)
228 ext4_fsblk_t goal, newblock;
230 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
231 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
236 static inline int ext4_ext_space_block(struct inode *inode, int check)
240 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
241 / sizeof(struct ext4_extent);
242 #ifdef AGGRESSIVE_TEST
243 if (!check && size > 6)
249 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
253 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
254 / sizeof(struct ext4_extent_idx);
255 #ifdef AGGRESSIVE_TEST
256 if (!check && size > 5)
262 static inline int ext4_ext_space_root(struct inode *inode, int check)
266 size = sizeof(EXT4_I(inode)->i_data);
267 size -= sizeof(struct ext4_extent_header);
268 size /= sizeof(struct ext4_extent);
269 #ifdef AGGRESSIVE_TEST
270 if (!check && size > 3)
276 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
280 size = sizeof(EXT4_I(inode)->i_data);
281 size -= sizeof(struct ext4_extent_header);
282 size /= sizeof(struct ext4_extent_idx);
283 #ifdef AGGRESSIVE_TEST
284 if (!check && size > 4)
291 * Calculate the number of metadata blocks needed
292 * to allocate @blocks
293 * Worse case is one block per extent
295 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
297 struct ext4_inode_info *ei = EXT4_I(inode);
300 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
301 / sizeof(struct ext4_extent_idx));
304 * If the new delayed allocation block is contiguous with the
305 * previous da block, it can share index blocks with the
306 * previous block, so we only need to allocate a new index
307 * block every idxs leaf blocks. At ldxs**2 blocks, we need
308 * an additional index block, and at ldxs**3 blocks, yet
309 * another index blocks.
311 if (ei->i_da_metadata_calc_len &&
312 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
315 if ((ei->i_da_metadata_calc_len % idxs) == 0)
317 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
319 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
321 ei->i_da_metadata_calc_len = 0;
323 ei->i_da_metadata_calc_len++;
324 ei->i_da_metadata_calc_last_lblock++;
329 * In the worst case we need a new set of index blocks at
330 * every level of the inode's extent tree.
332 ei->i_da_metadata_calc_len = 1;
333 ei->i_da_metadata_calc_last_lblock = lblock;
334 return ext_depth(inode) + 1;
338 ext4_ext_max_entries(struct inode *inode, int depth)
342 if (depth == ext_depth(inode)) {
344 max = ext4_ext_space_root(inode, 1);
346 max = ext4_ext_space_root_idx(inode, 1);
349 max = ext4_ext_space_block(inode, 1);
351 max = ext4_ext_space_block_idx(inode, 1);
357 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
359 ext4_fsblk_t block = ext4_ext_pblock(ext);
360 int len = ext4_ext_get_actual_len(ext);
361 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
362 ext4_lblk_t last = lblock + len - 1;
366 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
369 static int ext4_valid_extent_idx(struct inode *inode,
370 struct ext4_extent_idx *ext_idx)
372 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
374 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
377 static int ext4_valid_extent_entries(struct inode *inode,
378 struct ext4_extent_header *eh,
381 unsigned short entries;
382 if (eh->eh_entries == 0)
385 entries = le16_to_cpu(eh->eh_entries);
389 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
390 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
391 ext4_fsblk_t pblock = 0;
392 ext4_lblk_t lblock = 0;
393 ext4_lblk_t prev = 0;
396 if (!ext4_valid_extent(inode, ext))
399 /* Check for overlapping extents */
400 lblock = le32_to_cpu(ext->ee_block);
401 len = ext4_ext_get_actual_len(ext);
402 if ((lblock <= prev) && prev) {
403 pblock = ext4_ext_pblock(ext);
404 es->s_last_error_block = cpu_to_le64(pblock);
409 prev = lblock + len - 1;
412 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
414 if (!ext4_valid_extent_idx(inode, ext_idx))
423 static int __ext4_ext_check(const char *function, unsigned int line,
424 struct inode *inode, struct ext4_extent_header *eh,
425 int depth, ext4_fsblk_t pblk)
427 const char *error_msg;
430 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
431 error_msg = "invalid magic";
434 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
435 error_msg = "unexpected eh_depth";
438 if (unlikely(eh->eh_max == 0)) {
439 error_msg = "invalid eh_max";
442 max = ext4_ext_max_entries(inode, depth);
443 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
444 error_msg = "too large eh_max";
447 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
448 error_msg = "invalid eh_entries";
451 if (!ext4_valid_extent_entries(inode, eh, depth)) {
452 error_msg = "invalid extent entries";
455 /* Verify checksum on non-root extent tree nodes */
456 if (ext_depth(inode) != depth &&
457 !ext4_extent_block_csum_verify(inode, eh)) {
458 error_msg = "extent tree corrupted";
464 ext4_error_inode(inode, function, line, 0,
465 "pblk %llu bad header/extent: %s - magic %x, "
466 "entries %u, max %u(%u), depth %u(%u)",
467 (unsigned long long) pblk, error_msg,
468 le16_to_cpu(eh->eh_magic),
469 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
470 max, le16_to_cpu(eh->eh_depth), depth);
474 #define ext4_ext_check(inode, eh, depth, pblk) \
475 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
477 int ext4_ext_check_inode(struct inode *inode)
479 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
482 static struct buffer_head *
483 __read_extent_tree_block(const char *function, unsigned int line,
484 struct inode *inode, ext4_fsblk_t pblk, int depth,
487 struct buffer_head *bh;
490 bh = sb_getblk(inode->i_sb, pblk);
492 return ERR_PTR(-ENOMEM);
494 if (!bh_uptodate_or_lock(bh)) {
495 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
496 err = bh_submit_read(bh);
500 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
502 err = __ext4_ext_check(function, line, inode,
503 ext_block_hdr(bh), depth, pblk);
506 set_buffer_verified(bh);
508 * If this is a leaf block, cache all of its entries
510 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
511 struct ext4_extent_header *eh = ext_block_hdr(bh);
512 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
513 ext4_lblk_t prev = 0;
516 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
517 unsigned int status = EXTENT_STATUS_WRITTEN;
518 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
519 int len = ext4_ext_get_actual_len(ex);
521 if (prev && (prev != lblk))
522 ext4_es_cache_extent(inode, prev,
526 if (ext4_ext_is_uninitialized(ex))
527 status = EXTENT_STATUS_UNWRITTEN;
528 ext4_es_cache_extent(inode, lblk, len,
529 ext4_ext_pblock(ex), status);
540 #define read_extent_tree_block(inode, pblk, depth, flags) \
541 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
545 * This function is called to cache a file's extent information in the
548 int ext4_ext_precache(struct inode *inode)
550 struct ext4_inode_info *ei = EXT4_I(inode);
551 struct ext4_ext_path *path = NULL;
552 struct buffer_head *bh;
553 int i = 0, depth, ret = 0;
555 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
556 return 0; /* not an extent-mapped inode */
558 down_read(&ei->i_data_sem);
559 depth = ext_depth(inode);
561 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
564 up_read(&ei->i_data_sem);
568 /* Don't cache anything if there are no external extent blocks */
571 path[0].p_hdr = ext_inode_hdr(inode);
572 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
575 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
578 * If this is a leaf block or we've reached the end of
579 * the index block, go up
582 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
583 brelse(path[i].p_bh);
588 bh = read_extent_tree_block(inode,
589 ext4_idx_pblock(path[i].p_idx++),
591 EXT4_EX_FORCE_CACHE);
598 path[i].p_hdr = ext_block_hdr(bh);
599 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
601 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
603 up_read(&ei->i_data_sem);
604 ext4_ext_drop_refs(path);
610 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
612 int k, l = path->p_depth;
615 for (k = 0; k <= l; k++, path++) {
617 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
618 ext4_idx_pblock(path->p_idx));
619 } else if (path->p_ext) {
620 ext_debug(" %d:[%d]%d:%llu ",
621 le32_to_cpu(path->p_ext->ee_block),
622 ext4_ext_is_uninitialized(path->p_ext),
623 ext4_ext_get_actual_len(path->p_ext),
624 ext4_ext_pblock(path->p_ext));
631 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
633 int depth = ext_depth(inode);
634 struct ext4_extent_header *eh;
635 struct ext4_extent *ex;
641 eh = path[depth].p_hdr;
642 ex = EXT_FIRST_EXTENT(eh);
644 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
646 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
647 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
648 ext4_ext_is_uninitialized(ex),
649 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
654 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
655 ext4_fsblk_t newblock, int level)
657 int depth = ext_depth(inode);
658 struct ext4_extent *ex;
660 if (depth != level) {
661 struct ext4_extent_idx *idx;
662 idx = path[level].p_idx;
663 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
664 ext_debug("%d: move %d:%llu in new index %llu\n", level,
665 le32_to_cpu(idx->ei_block),
666 ext4_idx_pblock(idx),
674 ex = path[depth].p_ext;
675 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
676 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
677 le32_to_cpu(ex->ee_block),
679 ext4_ext_is_uninitialized(ex),
680 ext4_ext_get_actual_len(ex),
687 #define ext4_ext_show_path(inode, path)
688 #define ext4_ext_show_leaf(inode, path)
689 #define ext4_ext_show_move(inode, path, newblock, level)
692 void ext4_ext_drop_refs(struct ext4_ext_path *path)
694 int depth = path->p_depth;
697 for (i = 0; i <= depth; i++, path++)
705 * ext4_ext_binsearch_idx:
706 * binary search for the closest index of the given block
707 * the header must be checked before calling this
710 ext4_ext_binsearch_idx(struct inode *inode,
711 struct ext4_ext_path *path, ext4_lblk_t block)
713 struct ext4_extent_header *eh = path->p_hdr;
714 struct ext4_extent_idx *r, *l, *m;
717 ext_debug("binsearch for %u(idx): ", block);
719 l = EXT_FIRST_INDEX(eh) + 1;
720 r = EXT_LAST_INDEX(eh);
723 if (block < le32_to_cpu(m->ei_block))
727 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
728 m, le32_to_cpu(m->ei_block),
729 r, le32_to_cpu(r->ei_block));
733 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
734 ext4_idx_pblock(path->p_idx));
736 #ifdef CHECK_BINSEARCH
738 struct ext4_extent_idx *chix, *ix;
741 chix = ix = EXT_FIRST_INDEX(eh);
742 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
744 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
745 printk(KERN_DEBUG "k=%d, ix=0x%p, "
747 ix, EXT_FIRST_INDEX(eh));
748 printk(KERN_DEBUG "%u <= %u\n",
749 le32_to_cpu(ix->ei_block),
750 le32_to_cpu(ix[-1].ei_block));
752 BUG_ON(k && le32_to_cpu(ix->ei_block)
753 <= le32_to_cpu(ix[-1].ei_block));
754 if (block < le32_to_cpu(ix->ei_block))
758 BUG_ON(chix != path->p_idx);
765 * ext4_ext_binsearch:
766 * binary search for closest extent of the given block
767 * the header must be checked before calling this
770 ext4_ext_binsearch(struct inode *inode,
771 struct ext4_ext_path *path, ext4_lblk_t block)
773 struct ext4_extent_header *eh = path->p_hdr;
774 struct ext4_extent *r, *l, *m;
776 if (eh->eh_entries == 0) {
778 * this leaf is empty:
779 * we get such a leaf in split/add case
784 ext_debug("binsearch for %u: ", block);
786 l = EXT_FIRST_EXTENT(eh) + 1;
787 r = EXT_LAST_EXTENT(eh);
791 if (block < le32_to_cpu(m->ee_block))
795 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
796 m, le32_to_cpu(m->ee_block),
797 r, le32_to_cpu(r->ee_block));
801 ext_debug(" -> %d:%llu:[%d]%d ",
802 le32_to_cpu(path->p_ext->ee_block),
803 ext4_ext_pblock(path->p_ext),
804 ext4_ext_is_uninitialized(path->p_ext),
805 ext4_ext_get_actual_len(path->p_ext));
807 #ifdef CHECK_BINSEARCH
809 struct ext4_extent *chex, *ex;
812 chex = ex = EXT_FIRST_EXTENT(eh);
813 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
814 BUG_ON(k && le32_to_cpu(ex->ee_block)
815 <= le32_to_cpu(ex[-1].ee_block));
816 if (block < le32_to_cpu(ex->ee_block))
820 BUG_ON(chex != path->p_ext);
826 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
828 struct ext4_extent_header *eh;
830 eh = ext_inode_hdr(inode);
833 eh->eh_magic = EXT4_EXT_MAGIC;
834 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
835 ext4_mark_inode_dirty(handle, inode);
839 struct ext4_ext_path *
840 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
841 struct ext4_ext_path *path, int flags)
843 struct ext4_extent_header *eh;
844 struct buffer_head *bh;
845 short int depth, i, ppos = 0, alloc = 0;
848 eh = ext_inode_hdr(inode);
849 depth = ext_depth(inode);
851 /* account possible depth increase */
853 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
856 return ERR_PTR(-ENOMEM);
863 /* walk through the tree */
865 ext_debug("depth %d: num %d, max %d\n",
866 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
868 ext4_ext_binsearch_idx(inode, path + ppos, block);
869 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
870 path[ppos].p_depth = i;
871 path[ppos].p_ext = NULL;
873 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
880 eh = ext_block_hdr(bh);
882 if (unlikely(ppos > depth)) {
884 EXT4_ERROR_INODE(inode,
885 "ppos %d > depth %d", ppos, depth);
889 path[ppos].p_bh = bh;
890 path[ppos].p_hdr = eh;
893 path[ppos].p_depth = i;
894 path[ppos].p_ext = NULL;
895 path[ppos].p_idx = NULL;
898 ext4_ext_binsearch(inode, path + ppos, block);
899 /* if not an empty leaf */
900 if (path[ppos].p_ext)
901 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
903 ext4_ext_show_path(inode, path);
908 ext4_ext_drop_refs(path);
915 * ext4_ext_insert_index:
916 * insert new index [@logical;@ptr] into the block at @curp;
917 * check where to insert: before @curp or after @curp
919 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
920 struct ext4_ext_path *curp,
921 int logical, ext4_fsblk_t ptr)
923 struct ext4_extent_idx *ix;
926 err = ext4_ext_get_access(handle, inode, curp);
930 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
931 EXT4_ERROR_INODE(inode,
932 "logical %d == ei_block %d!",
933 logical, le32_to_cpu(curp->p_idx->ei_block));
937 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
938 >= le16_to_cpu(curp->p_hdr->eh_max))) {
939 EXT4_ERROR_INODE(inode,
940 "eh_entries %d >= eh_max %d!",
941 le16_to_cpu(curp->p_hdr->eh_entries),
942 le16_to_cpu(curp->p_hdr->eh_max));
946 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
948 ext_debug("insert new index %d after: %llu\n", logical, ptr);
949 ix = curp->p_idx + 1;
952 ext_debug("insert new index %d before: %llu\n", logical, ptr);
956 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
959 ext_debug("insert new index %d: "
960 "move %d indices from 0x%p to 0x%p\n",
961 logical, len, ix, ix + 1);
962 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
965 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
966 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
970 ix->ei_block = cpu_to_le32(logical);
971 ext4_idx_store_pblock(ix, ptr);
972 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
974 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
975 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
979 err = ext4_ext_dirty(handle, inode, curp);
980 ext4_std_error(inode->i_sb, err);
987 * inserts new subtree into the path, using free index entry
989 * - allocates all needed blocks (new leaf and all intermediate index blocks)
990 * - makes decision where to split
991 * - moves remaining extents and index entries (right to the split point)
992 * into the newly allocated blocks
993 * - initializes subtree
995 static int ext4_ext_split(handle_t *handle, struct inode *inode,
997 struct ext4_ext_path *path,
998 struct ext4_extent *newext, int at)
1000 struct buffer_head *bh = NULL;
1001 int depth = ext_depth(inode);
1002 struct ext4_extent_header *neh;
1003 struct ext4_extent_idx *fidx;
1004 int i = at, k, m, a;
1005 ext4_fsblk_t newblock, oldblock;
1007 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1010 /* make decision: where to split? */
1011 /* FIXME: now decision is simplest: at current extent */
1013 /* if current leaf will be split, then we should use
1014 * border from split point */
1015 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1016 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1019 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1020 border = path[depth].p_ext[1].ee_block;
1021 ext_debug("leaf will be split."
1022 " next leaf starts at %d\n",
1023 le32_to_cpu(border));
1025 border = newext->ee_block;
1026 ext_debug("leaf will be added."
1027 " next leaf starts at %d\n",
1028 le32_to_cpu(border));
1032 * If error occurs, then we break processing
1033 * and mark filesystem read-only. index won't
1034 * be inserted and tree will be in consistent
1035 * state. Next mount will repair buffers too.
1039 * Get array to track all allocated blocks.
1040 * We need this to handle errors and free blocks
1043 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1047 /* allocate all needed blocks */
1048 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1049 for (a = 0; a < depth - at; a++) {
1050 newblock = ext4_ext_new_meta_block(handle, inode, path,
1051 newext, &err, flags);
1054 ablocks[a] = newblock;
1057 /* initialize new leaf */
1058 newblock = ablocks[--a];
1059 if (unlikely(newblock == 0)) {
1060 EXT4_ERROR_INODE(inode, "newblock == 0!");
1064 bh = sb_getblk(inode->i_sb, newblock);
1065 if (unlikely(!bh)) {
1071 err = ext4_journal_get_create_access(handle, bh);
1075 neh = ext_block_hdr(bh);
1076 neh->eh_entries = 0;
1077 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1078 neh->eh_magic = EXT4_EXT_MAGIC;
1081 /* move remainder of path[depth] to the new leaf */
1082 if (unlikely(path[depth].p_hdr->eh_entries !=
1083 path[depth].p_hdr->eh_max)) {
1084 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1085 path[depth].p_hdr->eh_entries,
1086 path[depth].p_hdr->eh_max);
1090 /* start copy from next extent */
1091 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1092 ext4_ext_show_move(inode, path, newblock, depth);
1094 struct ext4_extent *ex;
1095 ex = EXT_FIRST_EXTENT(neh);
1096 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1097 le16_add_cpu(&neh->eh_entries, m);
1100 ext4_extent_block_csum_set(inode, neh);
1101 set_buffer_uptodate(bh);
1104 err = ext4_handle_dirty_metadata(handle, inode, bh);
1110 /* correct old leaf */
1112 err = ext4_ext_get_access(handle, inode, path + depth);
1115 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1116 err = ext4_ext_dirty(handle, inode, path + depth);
1122 /* create intermediate indexes */
1124 if (unlikely(k < 0)) {
1125 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1130 ext_debug("create %d intermediate indices\n", k);
1131 /* insert new index into current index block */
1132 /* current depth stored in i var */
1135 oldblock = newblock;
1136 newblock = ablocks[--a];
1137 bh = sb_getblk(inode->i_sb, newblock);
1138 if (unlikely(!bh)) {
1144 err = ext4_journal_get_create_access(handle, bh);
1148 neh = ext_block_hdr(bh);
1149 neh->eh_entries = cpu_to_le16(1);
1150 neh->eh_magic = EXT4_EXT_MAGIC;
1151 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1152 neh->eh_depth = cpu_to_le16(depth - i);
1153 fidx = EXT_FIRST_INDEX(neh);
1154 fidx->ei_block = border;
1155 ext4_idx_store_pblock(fidx, oldblock);
1157 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1158 i, newblock, le32_to_cpu(border), oldblock);
1160 /* move remainder of path[i] to the new index block */
1161 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1162 EXT_LAST_INDEX(path[i].p_hdr))) {
1163 EXT4_ERROR_INODE(inode,
1164 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1165 le32_to_cpu(path[i].p_ext->ee_block));
1169 /* start copy indexes */
1170 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1171 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1172 EXT_MAX_INDEX(path[i].p_hdr));
1173 ext4_ext_show_move(inode, path, newblock, i);
1175 memmove(++fidx, path[i].p_idx,
1176 sizeof(struct ext4_extent_idx) * m);
1177 le16_add_cpu(&neh->eh_entries, m);
1179 ext4_extent_block_csum_set(inode, neh);
1180 set_buffer_uptodate(bh);
1183 err = ext4_handle_dirty_metadata(handle, inode, bh);
1189 /* correct old index */
1191 err = ext4_ext_get_access(handle, inode, path + i);
1194 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1195 err = ext4_ext_dirty(handle, inode, path + i);
1203 /* insert new index */
1204 err = ext4_ext_insert_index(handle, inode, path + at,
1205 le32_to_cpu(border), newblock);
1209 if (buffer_locked(bh))
1215 /* free all allocated blocks in error case */
1216 for (i = 0; i < depth; i++) {
1219 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1220 EXT4_FREE_BLOCKS_METADATA);
1229 * ext4_ext_grow_indepth:
1230 * implements tree growing procedure:
1231 * - allocates new block
1232 * - moves top-level data (index block or leaf) into the new block
1233 * - initializes new top-level, creating index that points to the
1234 * just created block
1236 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1238 struct ext4_extent *newext)
1240 struct ext4_extent_header *neh;
1241 struct buffer_head *bh;
1242 ext4_fsblk_t newblock;
1245 newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1246 newext, &err, flags);
1250 bh = sb_getblk(inode->i_sb, newblock);
1255 err = ext4_journal_get_create_access(handle, bh);
1261 /* move top-level index/leaf into new block */
1262 memmove(bh->b_data, EXT4_I(inode)->i_data,
1263 sizeof(EXT4_I(inode)->i_data));
1265 /* set size of new block */
1266 neh = ext_block_hdr(bh);
1267 /* old root could have indexes or leaves
1268 * so calculate e_max right way */
1269 if (ext_depth(inode))
1270 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1272 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1273 neh->eh_magic = EXT4_EXT_MAGIC;
1274 ext4_extent_block_csum_set(inode, neh);
1275 set_buffer_uptodate(bh);
1278 err = ext4_handle_dirty_metadata(handle, inode, bh);
1282 /* Update top-level index: num,max,pointer */
1283 neh = ext_inode_hdr(inode);
1284 neh->eh_entries = cpu_to_le16(1);
1285 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1286 if (neh->eh_depth == 0) {
1287 /* Root extent block becomes index block */
1288 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1289 EXT_FIRST_INDEX(neh)->ei_block =
1290 EXT_FIRST_EXTENT(neh)->ee_block;
1292 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1293 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1294 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1295 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1297 le16_add_cpu(&neh->eh_depth, 1);
1298 ext4_mark_inode_dirty(handle, inode);
1306 * ext4_ext_create_new_leaf:
1307 * finds empty index and adds new leaf.
1308 * if no free index is found, then it requests in-depth growing.
1310 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1311 unsigned int mb_flags,
1312 unsigned int gb_flags,
1313 struct ext4_ext_path *path,
1314 struct ext4_extent *newext)
1316 struct ext4_ext_path *curp;
1317 int depth, i, err = 0;
1320 i = depth = ext_depth(inode);
1322 /* walk up to the tree and look for free index entry */
1323 curp = path + depth;
1324 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1329 /* we use already allocated block for index block,
1330 * so subsequent data blocks should be contiguous */
1331 if (EXT_HAS_FREE_INDEX(curp)) {
1332 /* if we found index with free entry, then use that
1333 * entry: create all needed subtree and add new leaf */
1334 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1339 ext4_ext_drop_refs(path);
1340 path = ext4_ext_find_extent(inode,
1341 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1344 err = PTR_ERR(path);
1346 /* tree is full, time to grow in depth */
1347 err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext);
1352 ext4_ext_drop_refs(path);
1353 path = ext4_ext_find_extent(inode,
1354 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1357 err = PTR_ERR(path);
1362 * only first (depth 0 -> 1) produces free space;
1363 * in all other cases we have to split the grown tree
1365 depth = ext_depth(inode);
1366 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1367 /* now we need to split */
1377 * search the closest allocated block to the left for *logical
1378 * and returns it at @logical + it's physical address at @phys
1379 * if *logical is the smallest allocated block, the function
1380 * returns 0 at @phys
1381 * return value contains 0 (success) or error code
1383 static int ext4_ext_search_left(struct inode *inode,
1384 struct ext4_ext_path *path,
1385 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1387 struct ext4_extent_idx *ix;
1388 struct ext4_extent *ex;
1391 if (unlikely(path == NULL)) {
1392 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1395 depth = path->p_depth;
1398 if (depth == 0 && path->p_ext == NULL)
1401 /* usually extent in the path covers blocks smaller
1402 * then *logical, but it can be that extent is the
1403 * first one in the file */
1405 ex = path[depth].p_ext;
1406 ee_len = ext4_ext_get_actual_len(ex);
1407 if (*logical < le32_to_cpu(ex->ee_block)) {
1408 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1409 EXT4_ERROR_INODE(inode,
1410 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1411 *logical, le32_to_cpu(ex->ee_block));
1414 while (--depth >= 0) {
1415 ix = path[depth].p_idx;
1416 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1417 EXT4_ERROR_INODE(inode,
1418 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1419 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1420 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1421 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1429 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1430 EXT4_ERROR_INODE(inode,
1431 "logical %d < ee_block %d + ee_len %d!",
1432 *logical, le32_to_cpu(ex->ee_block), ee_len);
1436 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1437 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1442 * search the closest allocated block to the right for *logical
1443 * and returns it at @logical + it's physical address at @phys
1444 * if *logical is the largest allocated block, the function
1445 * returns 0 at @phys
1446 * return value contains 0 (success) or error code
1448 static int ext4_ext_search_right(struct inode *inode,
1449 struct ext4_ext_path *path,
1450 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1451 struct ext4_extent **ret_ex)
1453 struct buffer_head *bh = NULL;
1454 struct ext4_extent_header *eh;
1455 struct ext4_extent_idx *ix;
1456 struct ext4_extent *ex;
1458 int depth; /* Note, NOT eh_depth; depth from top of tree */
1461 if (unlikely(path == NULL)) {
1462 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1465 depth = path->p_depth;
1468 if (depth == 0 && path->p_ext == NULL)
1471 /* usually extent in the path covers blocks smaller
1472 * then *logical, but it can be that extent is the
1473 * first one in the file */
1475 ex = path[depth].p_ext;
1476 ee_len = ext4_ext_get_actual_len(ex);
1477 if (*logical < le32_to_cpu(ex->ee_block)) {
1478 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1479 EXT4_ERROR_INODE(inode,
1480 "first_extent(path[%d].p_hdr) != ex",
1484 while (--depth >= 0) {
1485 ix = path[depth].p_idx;
1486 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1487 EXT4_ERROR_INODE(inode,
1488 "ix != EXT_FIRST_INDEX *logical %d!",
1496 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1497 EXT4_ERROR_INODE(inode,
1498 "logical %d < ee_block %d + ee_len %d!",
1499 *logical, le32_to_cpu(ex->ee_block), ee_len);
1503 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1504 /* next allocated block in this leaf */
1509 /* go up and search for index to the right */
1510 while (--depth >= 0) {
1511 ix = path[depth].p_idx;
1512 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1516 /* we've gone up to the root and found no index to the right */
1520 /* we've found index to the right, let's
1521 * follow it and find the closest allocated
1522 * block to the right */
1524 block = ext4_idx_pblock(ix);
1525 while (++depth < path->p_depth) {
1526 /* subtract from p_depth to get proper eh_depth */
1527 bh = read_extent_tree_block(inode, block,
1528 path->p_depth - depth, 0);
1531 eh = ext_block_hdr(bh);
1532 ix = EXT_FIRST_INDEX(eh);
1533 block = ext4_idx_pblock(ix);
1537 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1540 eh = ext_block_hdr(bh);
1541 ex = EXT_FIRST_EXTENT(eh);
1543 *logical = le32_to_cpu(ex->ee_block);
1544 *phys = ext4_ext_pblock(ex);
1552 * ext4_ext_next_allocated_block:
1553 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1554 * NOTE: it considers block number from index entry as
1555 * allocated block. Thus, index entries have to be consistent
1559 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1563 BUG_ON(path == NULL);
1564 depth = path->p_depth;
1566 if (depth == 0 && path->p_ext == NULL)
1567 return EXT_MAX_BLOCKS;
1569 while (depth >= 0) {
1570 if (depth == path->p_depth) {
1572 if (path[depth].p_ext &&
1573 path[depth].p_ext !=
1574 EXT_LAST_EXTENT(path[depth].p_hdr))
1575 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1578 if (path[depth].p_idx !=
1579 EXT_LAST_INDEX(path[depth].p_hdr))
1580 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1585 return EXT_MAX_BLOCKS;
1589 * ext4_ext_next_leaf_block:
1590 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1592 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1596 BUG_ON(path == NULL);
1597 depth = path->p_depth;
1599 /* zero-tree has no leaf blocks at all */
1601 return EXT_MAX_BLOCKS;
1603 /* go to index block */
1606 while (depth >= 0) {
1607 if (path[depth].p_idx !=
1608 EXT_LAST_INDEX(path[depth].p_hdr))
1609 return (ext4_lblk_t)
1610 le32_to_cpu(path[depth].p_idx[1].ei_block);
1614 return EXT_MAX_BLOCKS;
1618 * ext4_ext_correct_indexes:
1619 * if leaf gets modified and modified extent is first in the leaf,
1620 * then we have to correct all indexes above.
1621 * TODO: do we need to correct tree in all cases?
1623 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1624 struct ext4_ext_path *path)
1626 struct ext4_extent_header *eh;
1627 int depth = ext_depth(inode);
1628 struct ext4_extent *ex;
1632 eh = path[depth].p_hdr;
1633 ex = path[depth].p_ext;
1635 if (unlikely(ex == NULL || eh == NULL)) {
1636 EXT4_ERROR_INODE(inode,
1637 "ex %p == NULL or eh %p == NULL", ex, eh);
1642 /* there is no tree at all */
1646 if (ex != EXT_FIRST_EXTENT(eh)) {
1647 /* we correct tree if first leaf got modified only */
1652 * TODO: we need correction if border is smaller than current one
1655 border = path[depth].p_ext->ee_block;
1656 err = ext4_ext_get_access(handle, inode, path + k);
1659 path[k].p_idx->ei_block = border;
1660 err = ext4_ext_dirty(handle, inode, path + k);
1665 /* change all left-side indexes */
1666 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1668 err = ext4_ext_get_access(handle, inode, path + k);
1671 path[k].p_idx->ei_block = border;
1672 err = ext4_ext_dirty(handle, inode, path + k);
1681 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1682 struct ext4_extent *ex2)
1684 unsigned short ext1_ee_len, ext2_ee_len;
1687 * Make sure that both extents are initialized. We don't merge
1688 * uninitialized extents so that we can be sure that end_io code has
1689 * the extent that was written properly split out and conversion to
1690 * initialized is trivial.
1692 if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
1695 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1696 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1698 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1699 le32_to_cpu(ex2->ee_block))
1703 * To allow future support for preallocated extents to be added
1704 * as an RO_COMPAT feature, refuse to merge to extents if
1705 * this can result in the top bit of ee_len being set.
1707 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1709 #ifdef AGGRESSIVE_TEST
1710 if (ext1_ee_len >= 4)
1714 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1720 * This function tries to merge the "ex" extent to the next extent in the tree.
1721 * It always tries to merge towards right. If you want to merge towards
1722 * left, pass "ex - 1" as argument instead of "ex".
1723 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1724 * 1 if they got merged.
1726 static int ext4_ext_try_to_merge_right(struct inode *inode,
1727 struct ext4_ext_path *path,
1728 struct ext4_extent *ex)
1730 struct ext4_extent_header *eh;
1731 unsigned int depth, len;
1734 depth = ext_depth(inode);
1735 BUG_ON(path[depth].p_hdr == NULL);
1736 eh = path[depth].p_hdr;
1738 while (ex < EXT_LAST_EXTENT(eh)) {
1739 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1741 /* merge with next extent! */
1742 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1743 + ext4_ext_get_actual_len(ex + 1));
1745 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1746 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1747 * sizeof(struct ext4_extent);
1748 memmove(ex + 1, ex + 2, len);
1750 le16_add_cpu(&eh->eh_entries, -1);
1752 WARN_ON(eh->eh_entries == 0);
1753 if (!eh->eh_entries)
1754 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1761 * This function does a very simple check to see if we can collapse
1762 * an extent tree with a single extent tree leaf block into the inode.
1764 static void ext4_ext_try_to_merge_up(handle_t *handle,
1765 struct inode *inode,
1766 struct ext4_ext_path *path)
1769 unsigned max_root = ext4_ext_space_root(inode, 0);
1772 if ((path[0].p_depth != 1) ||
1773 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1774 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1778 * We need to modify the block allocation bitmap and the block
1779 * group descriptor to release the extent tree block. If we
1780 * can't get the journal credits, give up.
1782 if (ext4_journal_extend(handle, 2))
1786 * Copy the extent data up to the inode
1788 blk = ext4_idx_pblock(path[0].p_idx);
1789 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1790 sizeof(struct ext4_extent_idx);
1791 s += sizeof(struct ext4_extent_header);
1793 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1794 path[0].p_depth = 0;
1795 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1796 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1797 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1799 brelse(path[1].p_bh);
1800 ext4_free_blocks(handle, inode, NULL, blk, 1,
1801 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET |
1802 EXT4_FREE_BLOCKS_RESERVE);
1806 * This function tries to merge the @ex extent to neighbours in the tree.
1807 * return 1 if merge left else 0.
1809 static void ext4_ext_try_to_merge(handle_t *handle,
1810 struct inode *inode,
1811 struct ext4_ext_path *path,
1812 struct ext4_extent *ex) {
1813 struct ext4_extent_header *eh;
1817 depth = ext_depth(inode);
1818 BUG_ON(path[depth].p_hdr == NULL);
1819 eh = path[depth].p_hdr;
1821 if (ex > EXT_FIRST_EXTENT(eh))
1822 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1825 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1827 ext4_ext_try_to_merge_up(handle, inode, path);
1831 * check if a portion of the "newext" extent overlaps with an
1834 * If there is an overlap discovered, it updates the length of the newext
1835 * such that there will be no overlap, and then returns 1.
1836 * If there is no overlap found, it returns 0.
1838 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1839 struct inode *inode,
1840 struct ext4_extent *newext,
1841 struct ext4_ext_path *path)
1844 unsigned int depth, len1;
1845 unsigned int ret = 0;
1847 b1 = le32_to_cpu(newext->ee_block);
1848 len1 = ext4_ext_get_actual_len(newext);
1849 depth = ext_depth(inode);
1850 if (!path[depth].p_ext)
1852 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1855 * get the next allocated block if the extent in the path
1856 * is before the requested block(s)
1859 b2 = ext4_ext_next_allocated_block(path);
1860 if (b2 == EXT_MAX_BLOCKS)
1862 b2 = EXT4_LBLK_CMASK(sbi, b2);
1865 /* check for wrap through zero on extent logical start block*/
1866 if (b1 + len1 < b1) {
1867 len1 = EXT_MAX_BLOCKS - b1;
1868 newext->ee_len = cpu_to_le16(len1);
1872 /* check for overlap */
1873 if (b1 + len1 > b2) {
1874 newext->ee_len = cpu_to_le16(b2 - b1);
1882 * ext4_ext_insert_extent:
1883 * tries to merge requsted extent into the existing extent or
1884 * inserts requested extent as new one into the tree,
1885 * creating new leaf in the no-space case.
1887 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1888 struct ext4_ext_path *path,
1889 struct ext4_extent *newext, int gb_flags)
1891 struct ext4_extent_header *eh;
1892 struct ext4_extent *ex, *fex;
1893 struct ext4_extent *nearex; /* nearest extent */
1894 struct ext4_ext_path *npath = NULL;
1895 int depth, len, err;
1899 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1900 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1903 depth = ext_depth(inode);
1904 ex = path[depth].p_ext;
1905 eh = path[depth].p_hdr;
1906 if (unlikely(path[depth].p_hdr == NULL)) {
1907 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1911 /* try to insert block into found extent and return */
1912 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1915 * Try to see whether we should rather test the extent on
1916 * right from ex, or from the left of ex. This is because
1917 * ext4_ext_find_extent() can return either extent on the
1918 * left, or on the right from the searched position. This
1919 * will make merging more effective.
1921 if (ex < EXT_LAST_EXTENT(eh) &&
1922 (le32_to_cpu(ex->ee_block) +
1923 ext4_ext_get_actual_len(ex) <
1924 le32_to_cpu(newext->ee_block))) {
1927 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1928 (le32_to_cpu(newext->ee_block) +
1929 ext4_ext_get_actual_len(newext) <
1930 le32_to_cpu(ex->ee_block)))
1933 /* Try to append newex to the ex */
1934 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1935 ext_debug("append [%d]%d block to %u:[%d]%d"
1937 ext4_ext_is_uninitialized(newext),
1938 ext4_ext_get_actual_len(newext),
1939 le32_to_cpu(ex->ee_block),
1940 ext4_ext_is_uninitialized(ex),
1941 ext4_ext_get_actual_len(ex),
1942 ext4_ext_pblock(ex));
1943 err = ext4_ext_get_access(handle, inode,
1948 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1949 + ext4_ext_get_actual_len(newext));
1950 eh = path[depth].p_hdr;
1956 /* Try to prepend newex to the ex */
1957 if (ext4_can_extents_be_merged(inode, newext, ex)) {
1958 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1960 le32_to_cpu(newext->ee_block),
1961 ext4_ext_is_uninitialized(newext),
1962 ext4_ext_get_actual_len(newext),
1963 le32_to_cpu(ex->ee_block),
1964 ext4_ext_is_uninitialized(ex),
1965 ext4_ext_get_actual_len(ex),
1966 ext4_ext_pblock(ex));
1967 err = ext4_ext_get_access(handle, inode,
1972 ex->ee_block = newext->ee_block;
1973 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
1974 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1975 + ext4_ext_get_actual_len(newext));
1976 eh = path[depth].p_hdr;
1982 depth = ext_depth(inode);
1983 eh = path[depth].p_hdr;
1984 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1987 /* probably next leaf has space for us? */
1988 fex = EXT_LAST_EXTENT(eh);
1989 next = EXT_MAX_BLOCKS;
1990 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1991 next = ext4_ext_next_leaf_block(path);
1992 if (next != EXT_MAX_BLOCKS) {
1993 ext_debug("next leaf block - %u\n", next);
1994 BUG_ON(npath != NULL);
1995 npath = ext4_ext_find_extent(inode, next, NULL, 0);
1997 return PTR_ERR(npath);
1998 BUG_ON(npath->p_depth != path->p_depth);
1999 eh = npath[depth].p_hdr;
2000 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2001 ext_debug("next leaf isn't full(%d)\n",
2002 le16_to_cpu(eh->eh_entries));
2006 ext_debug("next leaf has no free space(%d,%d)\n",
2007 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2011 * There is no free space in the found leaf.
2012 * We're gonna add a new leaf in the tree.
2014 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2015 mb_flags = EXT4_MB_USE_RESERVED;
2016 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2020 depth = ext_depth(inode);
2021 eh = path[depth].p_hdr;
2024 nearex = path[depth].p_ext;
2026 err = ext4_ext_get_access(handle, inode, path + depth);
2031 /* there is no extent in this leaf, create first one */
2032 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2033 le32_to_cpu(newext->ee_block),
2034 ext4_ext_pblock(newext),
2035 ext4_ext_is_uninitialized(newext),
2036 ext4_ext_get_actual_len(newext));
2037 nearex = EXT_FIRST_EXTENT(eh);
2039 if (le32_to_cpu(newext->ee_block)
2040 > le32_to_cpu(nearex->ee_block)) {
2042 ext_debug("insert %u:%llu:[%d]%d before: "
2044 le32_to_cpu(newext->ee_block),
2045 ext4_ext_pblock(newext),
2046 ext4_ext_is_uninitialized(newext),
2047 ext4_ext_get_actual_len(newext),
2052 BUG_ON(newext->ee_block == nearex->ee_block);
2053 ext_debug("insert %u:%llu:[%d]%d after: "
2055 le32_to_cpu(newext->ee_block),
2056 ext4_ext_pblock(newext),
2057 ext4_ext_is_uninitialized(newext),
2058 ext4_ext_get_actual_len(newext),
2061 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2063 ext_debug("insert %u:%llu:[%d]%d: "
2064 "move %d extents from 0x%p to 0x%p\n",
2065 le32_to_cpu(newext->ee_block),
2066 ext4_ext_pblock(newext),
2067 ext4_ext_is_uninitialized(newext),
2068 ext4_ext_get_actual_len(newext),
2069 len, nearex, nearex + 1);
2070 memmove(nearex + 1, nearex,
2071 len * sizeof(struct ext4_extent));
2075 le16_add_cpu(&eh->eh_entries, 1);
2076 path[depth].p_ext = nearex;
2077 nearex->ee_block = newext->ee_block;
2078 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2079 nearex->ee_len = newext->ee_len;
2082 /* try to merge extents */
2083 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2084 ext4_ext_try_to_merge(handle, inode, path, nearex);
2087 /* time to correct all indexes above */
2088 err = ext4_ext_correct_indexes(handle, inode, path);
2092 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2096 ext4_ext_drop_refs(npath);
2102 static int ext4_fill_fiemap_extents(struct inode *inode,
2103 ext4_lblk_t block, ext4_lblk_t num,
2104 struct fiemap_extent_info *fieinfo)
2106 struct ext4_ext_path *path = NULL;
2107 struct ext4_extent *ex;
2108 struct extent_status es;
2109 ext4_lblk_t next, next_del, start = 0, end = 0;
2110 ext4_lblk_t last = block + num;
2111 int exists, depth = 0, err = 0;
2112 unsigned int flags = 0;
2113 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2115 while (block < last && block != EXT_MAX_BLOCKS) {
2117 /* find extent for this block */
2118 down_read(&EXT4_I(inode)->i_data_sem);
2120 if (path && ext_depth(inode) != depth) {
2121 /* depth was changed. we have to realloc path */
2126 path = ext4_ext_find_extent(inode, block, path, 0);
2128 up_read(&EXT4_I(inode)->i_data_sem);
2129 err = PTR_ERR(path);
2134 depth = ext_depth(inode);
2135 if (unlikely(path[depth].p_hdr == NULL)) {
2136 up_read(&EXT4_I(inode)->i_data_sem);
2137 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2141 ex = path[depth].p_ext;
2142 next = ext4_ext_next_allocated_block(path);
2143 ext4_ext_drop_refs(path);
2148 /* there is no extent yet, so try to allocate
2149 * all requested space */
2152 } else if (le32_to_cpu(ex->ee_block) > block) {
2153 /* need to allocate space before found extent */
2155 end = le32_to_cpu(ex->ee_block);
2156 if (block + num < end)
2158 } else if (block >= le32_to_cpu(ex->ee_block)
2159 + ext4_ext_get_actual_len(ex)) {
2160 /* need to allocate space after found extent */
2165 } else if (block >= le32_to_cpu(ex->ee_block)) {
2167 * some part of requested space is covered
2171 end = le32_to_cpu(ex->ee_block)
2172 + ext4_ext_get_actual_len(ex);
2173 if (block + num < end)
2179 BUG_ON(end <= start);
2183 es.es_len = end - start;
2186 es.es_lblk = le32_to_cpu(ex->ee_block);
2187 es.es_len = ext4_ext_get_actual_len(ex);
2188 es.es_pblk = ext4_ext_pblock(ex);
2189 if (ext4_ext_is_uninitialized(ex))
2190 flags |= FIEMAP_EXTENT_UNWRITTEN;
2194 * Find delayed extent and update es accordingly. We call
2195 * it even in !exists case to find out whether es is the
2196 * last existing extent or not.
2198 next_del = ext4_find_delayed_extent(inode, &es);
2199 if (!exists && next_del) {
2201 flags |= (FIEMAP_EXTENT_DELALLOC |
2202 FIEMAP_EXTENT_UNKNOWN);
2204 up_read(&EXT4_I(inode)->i_data_sem);
2206 if (unlikely(es.es_len == 0)) {
2207 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2213 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2214 * we need to check next == EXT_MAX_BLOCKS because it is
2215 * possible that an extent is with unwritten and delayed
2216 * status due to when an extent is delayed allocated and
2217 * is allocated by fallocate status tree will track both of
2220 * So we could return a unwritten and delayed extent, and
2221 * its block is equal to 'next'.
2223 if (next == next_del && next == EXT_MAX_BLOCKS) {
2224 flags |= FIEMAP_EXTENT_LAST;
2225 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2226 next != EXT_MAX_BLOCKS)) {
2227 EXT4_ERROR_INODE(inode,
2228 "next extent == %u, next "
2229 "delalloc extent = %u",
2237 err = fiemap_fill_next_extent(fieinfo,
2238 (__u64)es.es_lblk << blksize_bits,
2239 (__u64)es.es_pblk << blksize_bits,
2240 (__u64)es.es_len << blksize_bits,
2250 block = es.es_lblk + es.es_len;
2254 ext4_ext_drop_refs(path);
2262 * ext4_ext_put_gap_in_cache:
2263 * calculate boundaries of the gap that the requested block fits into
2264 * and cache this gap
2267 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2270 int depth = ext_depth(inode);
2271 unsigned long len = 0;
2272 ext4_lblk_t lblock = 0;
2273 struct ext4_extent *ex;
2275 ex = path[depth].p_ext;
2278 * there is no extent yet, so gap is [0;-] and we
2281 ext_debug("cache gap(whole file):");
2282 } else if (block < le32_to_cpu(ex->ee_block)) {
2284 len = le32_to_cpu(ex->ee_block) - block;
2285 ext_debug("cache gap(before): %u [%u:%u]",
2287 le32_to_cpu(ex->ee_block),
2288 ext4_ext_get_actual_len(ex));
2289 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2290 ext4_es_insert_extent(inode, lblock, len, ~0,
2291 EXTENT_STATUS_HOLE);
2292 } else if (block >= le32_to_cpu(ex->ee_block)
2293 + ext4_ext_get_actual_len(ex)) {
2295 lblock = le32_to_cpu(ex->ee_block)
2296 + ext4_ext_get_actual_len(ex);
2298 next = ext4_ext_next_allocated_block(path);
2299 ext_debug("cache gap(after): [%u:%u] %u",
2300 le32_to_cpu(ex->ee_block),
2301 ext4_ext_get_actual_len(ex),
2303 BUG_ON(next == lblock);
2304 len = next - lblock;
2305 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2306 ext4_es_insert_extent(inode, lblock, len, ~0,
2307 EXTENT_STATUS_HOLE);
2312 ext_debug(" -> %u:%lu\n", lblock, len);
2317 * removes index from the index block.
2319 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2320 struct ext4_ext_path *path, int depth)
2325 /* free index block */
2327 path = path + depth;
2328 leaf = ext4_idx_pblock(path->p_idx);
2329 if (unlikely(path->p_hdr->eh_entries == 0)) {
2330 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2333 err = ext4_ext_get_access(handle, inode, path);
2337 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2338 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2339 len *= sizeof(struct ext4_extent_idx);
2340 memmove(path->p_idx, path->p_idx + 1, len);
2343 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2344 err = ext4_ext_dirty(handle, inode, path);
2347 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2348 trace_ext4_ext_rm_idx(inode, leaf);
2350 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2351 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2353 while (--depth >= 0) {
2354 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2357 err = ext4_ext_get_access(handle, inode, path);
2360 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2361 err = ext4_ext_dirty(handle, inode, path);
2369 * ext4_ext_calc_credits_for_single_extent:
2370 * This routine returns max. credits that needed to insert an extent
2371 * to the extent tree.
2372 * When pass the actual path, the caller should calculate credits
2375 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2376 struct ext4_ext_path *path)
2379 int depth = ext_depth(inode);
2382 /* probably there is space in leaf? */
2383 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2384 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2387 * There are some space in the leaf tree, no
2388 * need to account for leaf block credit
2390 * bitmaps and block group descriptor blocks
2391 * and other metadata blocks still need to be
2394 /* 1 bitmap, 1 block group descriptor */
2395 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2400 return ext4_chunk_trans_blocks(inode, nrblocks);
2404 * How many index/leaf blocks need to change/allocate to add @extents extents?
2406 * If we add a single extent, then in the worse case, each tree level
2407 * index/leaf need to be changed in case of the tree split.
2409 * If more extents are inserted, they could cause the whole tree split more
2410 * than once, but this is really rare.
2412 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2417 /* If we are converting the inline data, only one is needed here. */
2418 if (ext4_has_inline_data(inode))
2421 depth = ext_depth(inode);
2431 static inline int get_default_free_blocks_flags(struct inode *inode)
2433 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2434 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2435 else if (ext4_should_journal_data(inode))
2436 return EXT4_FREE_BLOCKS_FORGET;
2440 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2441 struct ext4_extent *ex,
2442 long long *partial_cluster,
2443 ext4_lblk_t from, ext4_lblk_t to)
2445 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2446 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2448 int flags = get_default_free_blocks_flags(inode);
2451 * For bigalloc file systems, we never free a partial cluster
2452 * at the beginning of the extent. Instead, we make a note
2453 * that we tried freeing the cluster, and check to see if we
2454 * need to free it on a subsequent call to ext4_remove_blocks,
2455 * or at the end of the ext4_truncate() operation.
2457 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2459 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2461 * If we have a partial cluster, and it's different from the
2462 * cluster of the last block, we need to explicitly free the
2463 * partial cluster here.
2465 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2466 if ((*partial_cluster > 0) &&
2467 (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2468 ext4_free_blocks(handle, inode, NULL,
2469 EXT4_C2B(sbi, *partial_cluster),
2470 sbi->s_cluster_ratio, flags);
2471 *partial_cluster = 0;
2474 #ifdef EXTENTS_STATS
2476 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2477 spin_lock(&sbi->s_ext_stats_lock);
2478 sbi->s_ext_blocks += ee_len;
2479 sbi->s_ext_extents++;
2480 if (ee_len < sbi->s_ext_min)
2481 sbi->s_ext_min = ee_len;
2482 if (ee_len > sbi->s_ext_max)
2483 sbi->s_ext_max = ee_len;
2484 if (ext_depth(inode) > sbi->s_depth_max)
2485 sbi->s_depth_max = ext_depth(inode);
2486 spin_unlock(&sbi->s_ext_stats_lock);
2489 if (from >= le32_to_cpu(ex->ee_block)
2490 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2493 unsigned int unaligned;
2495 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2496 pblk = ext4_ext_pblock(ex) + ee_len - num;
2498 * Usually we want to free partial cluster at the end of the
2499 * extent, except for the situation when the cluster is still
2500 * used by any other extent (partial_cluster is negative).
2502 if (*partial_cluster < 0 &&
2503 -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1))
2504 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2506 ext_debug("free last %u blocks starting %llu partial %lld\n",
2507 num, pblk, *partial_cluster);
2508 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2510 * If the block range to be freed didn't start at the
2511 * beginning of a cluster, and we removed the entire
2512 * extent and the cluster is not used by any other extent,
2513 * save the partial cluster here, since we might need to
2514 * delete if we determine that the truncate operation has
2515 * removed all of the blocks in the cluster.
2517 * On the other hand, if we did not manage to free the whole
2518 * extent, we have to mark the cluster as used (store negative
2519 * cluster number in partial_cluster).
2521 unaligned = EXT4_PBLK_COFF(sbi, pblk);
2522 if (unaligned && (ee_len == num) &&
2523 (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk))))
2524 *partial_cluster = EXT4_B2C(sbi, pblk);
2526 *partial_cluster = -((long long)EXT4_B2C(sbi, pblk));
2527 else if (*partial_cluster > 0)
2528 *partial_cluster = 0;
2530 ext4_error(sbi->s_sb, "strange request: removal(2) "
2531 "%u-%u from %u:%u\n",
2532 from, to, le32_to_cpu(ex->ee_block), ee_len);
2538 * ext4_ext_rm_leaf() Removes the extents associated with the
2539 * blocks appearing between "start" and "end", and splits the extents
2540 * if "start" and "end" appear in the same extent
2542 * @handle: The journal handle
2543 * @inode: The files inode
2544 * @path: The path to the leaf
2545 * @partial_cluster: The cluster which we'll have to free if all extents
2546 * has been released from it. It gets negative in case
2547 * that the cluster is still used.
2548 * @start: The first block to remove
2549 * @end: The last block to remove
2552 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2553 struct ext4_ext_path *path,
2554 long long *partial_cluster,
2555 ext4_lblk_t start, ext4_lblk_t end)
2557 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2558 int err = 0, correct_index = 0;
2559 int depth = ext_depth(inode), credits;
2560 struct ext4_extent_header *eh;
2563 ext4_lblk_t ex_ee_block;
2564 unsigned short ex_ee_len;
2565 unsigned uninitialized = 0;
2566 struct ext4_extent *ex;
2569 /* the header must be checked already in ext4_ext_remove_space() */
2570 ext_debug("truncate since %u in leaf to %u\n", start, end);
2571 if (!path[depth].p_hdr)
2572 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2573 eh = path[depth].p_hdr;
2574 if (unlikely(path[depth].p_hdr == NULL)) {
2575 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2578 /* find where to start removing */
2579 ex = path[depth].p_ext;
2581 ex = EXT_LAST_EXTENT(eh);
2583 ex_ee_block = le32_to_cpu(ex->ee_block);
2584 ex_ee_len = ext4_ext_get_actual_len(ex);
2587 * If we're starting with an extent other than the last one in the
2588 * node, we need to see if it shares a cluster with the extent to
2589 * the right (towards the end of the file). If its leftmost cluster
2590 * is this extent's rightmost cluster and it is not cluster aligned,
2591 * we'll mark it as a partial that is not to be deallocated.
2594 if (ex != EXT_LAST_EXTENT(eh)) {
2595 ext4_fsblk_t current_pblk, right_pblk;
2596 long long current_cluster, right_cluster;
2598 current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2599 current_cluster = (long long)EXT4_B2C(sbi, current_pblk);
2600 right_pblk = ext4_ext_pblock(ex + 1);
2601 right_cluster = (long long)EXT4_B2C(sbi, right_pblk);
2602 if (current_cluster == right_cluster &&
2603 EXT4_PBLK_COFF(sbi, right_pblk))
2604 *partial_cluster = -right_cluster;
2607 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2609 while (ex >= EXT_FIRST_EXTENT(eh) &&
2610 ex_ee_block + ex_ee_len > start) {
2612 if (ext4_ext_is_uninitialized(ex))
2617 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2618 uninitialized, ex_ee_len);
2619 path[depth].p_ext = ex;
2621 a = ex_ee_block > start ? ex_ee_block : start;
2622 b = ex_ee_block+ex_ee_len - 1 < end ?
2623 ex_ee_block+ex_ee_len - 1 : end;
2625 ext_debug(" border %u:%u\n", a, b);
2627 /* If this extent is beyond the end of the hole, skip it */
2628 if (end < ex_ee_block) {
2630 * We're going to skip this extent and move to another,
2631 * so if this extent is not cluster aligned we have
2632 * to mark the current cluster as used to avoid
2633 * accidentally freeing it later on
2635 pblk = ext4_ext_pblock(ex);
2636 if (EXT4_PBLK_COFF(sbi, pblk))
2638 -((long long)EXT4_B2C(sbi, pblk));
2640 ex_ee_block = le32_to_cpu(ex->ee_block);
2641 ex_ee_len = ext4_ext_get_actual_len(ex);
2643 } else if (b != ex_ee_block + ex_ee_len - 1) {
2644 EXT4_ERROR_INODE(inode,
2645 "can not handle truncate %u:%u "
2647 start, end, ex_ee_block,
2648 ex_ee_block + ex_ee_len - 1);
2651 } else if (a != ex_ee_block) {
2652 /* remove tail of the extent */
2653 num = a - ex_ee_block;
2655 /* remove whole extent: excellent! */
2659 * 3 for leaf, sb, and inode plus 2 (bmap and group
2660 * descriptor) for each block group; assume two block
2661 * groups plus ex_ee_len/blocks_per_block_group for
2664 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2665 if (ex == EXT_FIRST_EXTENT(eh)) {
2667 credits += (ext_depth(inode)) + 1;
2669 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2671 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2675 err = ext4_ext_get_access(handle, inode, path + depth);
2679 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2685 /* this extent is removed; mark slot entirely unused */
2686 ext4_ext_store_pblock(ex, 0);
2688 ex->ee_len = cpu_to_le16(num);
2690 * Do not mark uninitialized if all the blocks in the
2691 * extent have been removed.
2693 if (uninitialized && num)
2694 ext4_ext_mark_uninitialized(ex);
2696 * If the extent was completely released,
2697 * we need to remove it from the leaf
2700 if (end != EXT_MAX_BLOCKS - 1) {
2702 * For hole punching, we need to scoot all the
2703 * extents up when an extent is removed so that
2704 * we dont have blank extents in the middle
2706 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2707 sizeof(struct ext4_extent));
2709 /* Now get rid of the one at the end */
2710 memset(EXT_LAST_EXTENT(eh), 0,
2711 sizeof(struct ext4_extent));
2713 le16_add_cpu(&eh->eh_entries, -1);
2714 } else if (*partial_cluster > 0)
2715 *partial_cluster = 0;
2717 err = ext4_ext_dirty(handle, inode, path + depth);
2721 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2722 ext4_ext_pblock(ex));
2724 ex_ee_block = le32_to_cpu(ex->ee_block);
2725 ex_ee_len = ext4_ext_get_actual_len(ex);
2728 if (correct_index && eh->eh_entries)
2729 err = ext4_ext_correct_indexes(handle, inode, path);
2732 * If there's a partial cluster and at least one extent remains in
2733 * the leaf, free the partial cluster if it isn't shared with the
2734 * current extent. If there's a partial cluster and no extents
2735 * remain in the leaf, it can't be freed here. It can only be
2736 * freed when it's possible to determine if it's not shared with
2737 * any other extent - when the next leaf is processed or when space
2738 * removal is complete.
2740 if (*partial_cluster > 0 && eh->eh_entries &&
2741 (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2742 *partial_cluster)) {
2743 int flags = get_default_free_blocks_flags(inode);
2745 ext4_free_blocks(handle, inode, NULL,
2746 EXT4_C2B(sbi, *partial_cluster),
2747 sbi->s_cluster_ratio, flags);
2748 *partial_cluster = 0;
2751 /* if this leaf is free, then we should
2752 * remove it from index block above */
2753 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2754 err = ext4_ext_rm_idx(handle, inode, path, depth);
2761 * ext4_ext_more_to_rm:
2762 * returns 1 if current index has to be freed (even partial)
2765 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2767 BUG_ON(path->p_idx == NULL);
2769 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2773 * if truncate on deeper level happened, it wasn't partial,
2774 * so we have to consider current index for truncation
2776 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2781 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2784 struct super_block *sb = inode->i_sb;
2785 int depth = ext_depth(inode);
2786 struct ext4_ext_path *path = NULL;
2787 long long partial_cluster = 0;
2791 ext_debug("truncate since %u to %u\n", start, end);
2793 /* probably first extent we're gonna free will be last in block */
2794 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2796 return PTR_ERR(handle);
2799 trace_ext4_ext_remove_space(inode, start, end, depth);
2802 * Check if we are removing extents inside the extent tree. If that
2803 * is the case, we are going to punch a hole inside the extent tree
2804 * so we have to check whether we need to split the extent covering
2805 * the last block to remove so we can easily remove the part of it
2806 * in ext4_ext_rm_leaf().
2808 if (end < EXT_MAX_BLOCKS - 1) {
2809 struct ext4_extent *ex;
2810 ext4_lblk_t ee_block;
2812 /* find extent for this block */
2813 path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2815 ext4_journal_stop(handle);
2816 return PTR_ERR(path);
2818 depth = ext_depth(inode);
2819 /* Leaf not may not exist only if inode has no blocks at all */
2820 ex = path[depth].p_ext;
2823 EXT4_ERROR_INODE(inode,
2824 "path[%d].p_hdr == NULL",
2831 ee_block = le32_to_cpu(ex->ee_block);
2834 * See if the last block is inside the extent, if so split
2835 * the extent at 'end' block so we can easily remove the
2836 * tail of the first part of the split extent in
2837 * ext4_ext_rm_leaf().
2839 if (end >= ee_block &&
2840 end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2843 if (ext4_ext_is_uninitialized(ex))
2844 split_flag = EXT4_EXT_MARK_UNINIT1 |
2845 EXT4_EXT_MARK_UNINIT2;
2848 * Split the extent in two so that 'end' is the last
2849 * block in the first new extent. Also we should not
2850 * fail removing space due to ENOSPC so try to use
2851 * reserved block if that happens.
2853 err = ext4_split_extent_at(handle, inode, path,
2854 end + 1, split_flag,
2856 EXT4_GET_BLOCKS_PRE_IO |
2857 EXT4_GET_BLOCKS_METADATA_NOFAIL);
2864 * We start scanning from right side, freeing all the blocks
2865 * after i_size and walking into the tree depth-wise.
2867 depth = ext_depth(inode);
2872 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2874 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2877 ext4_journal_stop(handle);
2880 path[0].p_depth = depth;
2881 path[0].p_hdr = ext_inode_hdr(inode);
2884 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2891 while (i >= 0 && err == 0) {
2893 /* this is leaf block */
2894 err = ext4_ext_rm_leaf(handle, inode, path,
2895 &partial_cluster, start,
2897 /* root level has p_bh == NULL, brelse() eats this */
2898 brelse(path[i].p_bh);
2899 path[i].p_bh = NULL;
2904 /* this is index block */
2905 if (!path[i].p_hdr) {
2906 ext_debug("initialize header\n");
2907 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2910 if (!path[i].p_idx) {
2911 /* this level hasn't been touched yet */
2912 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2913 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2914 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2916 le16_to_cpu(path[i].p_hdr->eh_entries));
2918 /* we were already here, see at next index */
2922 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2923 i, EXT_FIRST_INDEX(path[i].p_hdr),
2925 if (ext4_ext_more_to_rm(path + i)) {
2926 struct buffer_head *bh;
2927 /* go to the next level */
2928 ext_debug("move to level %d (block %llu)\n",
2929 i + 1, ext4_idx_pblock(path[i].p_idx));
2930 memset(path + i + 1, 0, sizeof(*path));
2931 bh = read_extent_tree_block(inode,
2932 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2935 /* should we reset i_size? */
2939 /* Yield here to deal with large extent trees.
2940 * Should be a no-op if we did IO above. */
2942 if (WARN_ON(i + 1 > depth)) {
2946 path[i + 1].p_bh = bh;
2948 /* save actual number of indexes since this
2949 * number is changed at the next iteration */
2950 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2953 /* we finished processing this index, go up */
2954 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2955 /* index is empty, remove it;
2956 * handle must be already prepared by the
2957 * truncatei_leaf() */
2958 err = ext4_ext_rm_idx(handle, inode, path, i);
2960 /* root level has p_bh == NULL, brelse() eats this */
2961 brelse(path[i].p_bh);
2962 path[i].p_bh = NULL;
2964 ext_debug("return to level %d\n", i);
2968 trace_ext4_ext_remove_space_done(inode, start, end, depth,
2969 partial_cluster, path->p_hdr->eh_entries);
2971 /* If we still have something in the partial cluster and we have removed
2972 * even the first extent, then we should free the blocks in the partial
2973 * cluster as well. */
2974 if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) {
2975 int flags = get_default_free_blocks_flags(inode);
2977 ext4_free_blocks(handle, inode, NULL,
2978 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2979 EXT4_SB(sb)->s_cluster_ratio, flags);
2980 partial_cluster = 0;
2983 /* TODO: flexible tree reduction should be here */
2984 if (path->p_hdr->eh_entries == 0) {
2986 * truncate to zero freed all the tree,
2987 * so we need to correct eh_depth
2989 err = ext4_ext_get_access(handle, inode, path);
2991 ext_inode_hdr(inode)->eh_depth = 0;
2992 ext_inode_hdr(inode)->eh_max =
2993 cpu_to_le16(ext4_ext_space_root(inode, 0));
2994 err = ext4_ext_dirty(handle, inode, path);
2998 ext4_ext_drop_refs(path);
3000 if (err == -EAGAIN) {
3004 ext4_journal_stop(handle);
3010 * called at mount time
3012 void ext4_ext_init(struct super_block *sb)
3015 * possible initialization would be here
3018 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3019 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3020 printk(KERN_INFO "EXT4-fs: file extents enabled"
3021 #ifdef AGGRESSIVE_TEST
3022 ", aggressive tests"
3024 #ifdef CHECK_BINSEARCH
3027 #ifdef EXTENTS_STATS
3032 #ifdef EXTENTS_STATS
3033 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3034 EXT4_SB(sb)->s_ext_min = 1 << 30;
3035 EXT4_SB(sb)->s_ext_max = 0;
3041 * called at umount time
3043 void ext4_ext_release(struct super_block *sb)
3045 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3048 #ifdef EXTENTS_STATS
3049 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3050 struct ext4_sb_info *sbi = EXT4_SB(sb);
3051 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3052 sbi->s_ext_blocks, sbi->s_ext_extents,
3053 sbi->s_ext_blocks / sbi->s_ext_extents);
3054 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3055 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3060 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3062 ext4_lblk_t ee_block;
3063 ext4_fsblk_t ee_pblock;
3064 unsigned int ee_len;
3066 ee_block = le32_to_cpu(ex->ee_block);
3067 ee_len = ext4_ext_get_actual_len(ex);
3068 ee_pblock = ext4_ext_pblock(ex);
3073 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3074 EXTENT_STATUS_WRITTEN);
3077 /* FIXME!! we need to try to merge to left or right after zero-out */
3078 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3080 ext4_fsblk_t ee_pblock;
3081 unsigned int ee_len;
3084 ee_len = ext4_ext_get_actual_len(ex);
3085 ee_pblock = ext4_ext_pblock(ex);
3087 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3095 * ext4_split_extent_at() splits an extent at given block.
3097 * @handle: the journal handle
3098 * @inode: the file inode
3099 * @path: the path to the extent
3100 * @split: the logical block where the extent is splitted.
3101 * @split_flags: indicates if the extent could be zeroout if split fails, and
3102 * the states(init or uninit) of new extents.
3103 * @flags: flags used to insert new extent to extent tree.
3106 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3107 * of which are deterimined by split_flag.
3109 * There are two cases:
3110 * a> the extent are splitted into two extent.
3111 * b> split is not needed, and just mark the extent.
3113 * return 0 on success.
3115 static int ext4_split_extent_at(handle_t *handle,
3116 struct inode *inode,
3117 struct ext4_ext_path *path,
3122 ext4_fsblk_t newblock;
3123 ext4_lblk_t ee_block;
3124 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3125 struct ext4_extent *ex2 = NULL;
3126 unsigned int ee_len, depth;
3129 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3130 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3132 ext_debug("ext4_split_extents_at: inode %lu, logical"
3133 "block %llu\n", inode->i_ino, (unsigned long long)split);
3135 ext4_ext_show_leaf(inode, path);
3137 depth = ext_depth(inode);
3138 ex = path[depth].p_ext;
3139 ee_block = le32_to_cpu(ex->ee_block);
3140 ee_len = ext4_ext_get_actual_len(ex);
3141 newblock = split - ee_block + ext4_ext_pblock(ex);
3143 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3144 BUG_ON(!ext4_ext_is_uninitialized(ex) &&
3145 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3146 EXT4_EXT_MARK_UNINIT1 |
3147 EXT4_EXT_MARK_UNINIT2));
3149 err = ext4_ext_get_access(handle, inode, path + depth);
3153 if (split == ee_block) {
3155 * case b: block @split is the block that the extent begins with
3156 * then we just change the state of the extent, and splitting
3159 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3160 ext4_ext_mark_uninitialized(ex);
3162 ext4_ext_mark_initialized(ex);
3164 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3165 ext4_ext_try_to_merge(handle, inode, path, ex);
3167 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3172 memcpy(&orig_ex, ex, sizeof(orig_ex));
3173 ex->ee_len = cpu_to_le16(split - ee_block);
3174 if (split_flag & EXT4_EXT_MARK_UNINIT1)
3175 ext4_ext_mark_uninitialized(ex);
3178 * path may lead to new leaf, not to original leaf any more
3179 * after ext4_ext_insert_extent() returns,
3181 err = ext4_ext_dirty(handle, inode, path + depth);
3183 goto fix_extent_len;
3186 ex2->ee_block = cpu_to_le32(split);
3187 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3188 ext4_ext_store_pblock(ex2, newblock);
3189 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3190 ext4_ext_mark_uninitialized(ex2);
3192 err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3193 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3194 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3195 if (split_flag & EXT4_EXT_DATA_VALID1) {
3196 err = ext4_ext_zeroout(inode, ex2);
3197 zero_ex.ee_block = ex2->ee_block;
3198 zero_ex.ee_len = cpu_to_le16(
3199 ext4_ext_get_actual_len(ex2));
3200 ext4_ext_store_pblock(&zero_ex,
3201 ext4_ext_pblock(ex2));
3203 err = ext4_ext_zeroout(inode, ex);
3204 zero_ex.ee_block = ex->ee_block;
3205 zero_ex.ee_len = cpu_to_le16(
3206 ext4_ext_get_actual_len(ex));
3207 ext4_ext_store_pblock(&zero_ex,
3208 ext4_ext_pblock(ex));
3211 err = ext4_ext_zeroout(inode, &orig_ex);
3212 zero_ex.ee_block = orig_ex.ee_block;
3213 zero_ex.ee_len = cpu_to_le16(
3214 ext4_ext_get_actual_len(&orig_ex));
3215 ext4_ext_store_pblock(&zero_ex,
3216 ext4_ext_pblock(&orig_ex));
3220 goto fix_extent_len;
3221 /* update the extent length and mark as initialized */
3222 ex->ee_len = cpu_to_le16(ee_len);
3223 ext4_ext_try_to_merge(handle, inode, path, ex);
3224 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3226 goto fix_extent_len;
3228 /* update extent status tree */
3229 err = ext4_zeroout_es(inode, &zero_ex);
3233 goto fix_extent_len;
3236 ext4_ext_show_leaf(inode, path);
3240 ex->ee_len = orig_ex.ee_len;
3241 ext4_ext_dirty(handle, inode, path + depth);
3246 * ext4_split_extents() splits an extent and mark extent which is covered
3247 * by @map as split_flags indicates
3249 * It may result in splitting the extent into multiple extents (up to three)
3250 * There are three possibilities:
3251 * a> There is no split required
3252 * b> Splits in two extents: Split is happening at either end of the extent
3253 * c> Splits in three extents: Somone is splitting in middle of the extent
3256 static int ext4_split_extent(handle_t *handle,
3257 struct inode *inode,
3258 struct ext4_ext_path *path,
3259 struct ext4_map_blocks *map,
3263 ext4_lblk_t ee_block;
3264 struct ext4_extent *ex;
3265 unsigned int ee_len, depth;
3268 int split_flag1, flags1;
3269 int allocated = map->m_len;
3271 depth = ext_depth(inode);
3272 ex = path[depth].p_ext;
3273 ee_block = le32_to_cpu(ex->ee_block);
3274 ee_len = ext4_ext_get_actual_len(ex);
3275 uninitialized = ext4_ext_is_uninitialized(ex);
3277 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3278 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3279 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3281 split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3282 EXT4_EXT_MARK_UNINIT2;
3283 if (split_flag & EXT4_EXT_DATA_VALID2)
3284 split_flag1 |= EXT4_EXT_DATA_VALID1;
3285 err = ext4_split_extent_at(handle, inode, path,
3286 map->m_lblk + map->m_len, split_flag1, flags1);
3290 allocated = ee_len - (map->m_lblk - ee_block);
3293 * Update path is required because previous ext4_split_extent_at() may
3294 * result in split of original leaf or extent zeroout.
3296 ext4_ext_drop_refs(path);
3297 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3299 return PTR_ERR(path);
3300 depth = ext_depth(inode);
3301 ex = path[depth].p_ext;
3302 uninitialized = ext4_ext_is_uninitialized(ex);
3305 if (map->m_lblk >= ee_block) {
3306 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3307 if (uninitialized) {
3308 split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3309 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3310 EXT4_EXT_MARK_UNINIT2);
3312 err = ext4_split_extent_at(handle, inode, path,
3313 map->m_lblk, split_flag1, flags);
3318 ext4_ext_show_leaf(inode, path);
3320 return err ? err : allocated;
3324 * This function is called by ext4_ext_map_blocks() if someone tries to write
3325 * to an uninitialized extent. It may result in splitting the uninitialized
3326 * extent into multiple extents (up to three - one initialized and two
3328 * There are three possibilities:
3329 * a> There is no split required: Entire extent should be initialized
3330 * b> Splits in two extents: Write is happening at either end of the extent
3331 * c> Splits in three extents: Somone is writing in middle of the extent
3334 * - The extent pointed to by 'path' is uninitialized.
3335 * - The extent pointed to by 'path' contains a superset
3336 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3338 * Post-conditions on success:
3339 * - the returned value is the number of blocks beyond map->l_lblk
3340 * that are allocated and initialized.
3341 * It is guaranteed to be >= map->m_len.
3343 static int ext4_ext_convert_to_initialized(handle_t *handle,
3344 struct inode *inode,
3345 struct ext4_map_blocks *map,
3346 struct ext4_ext_path *path,
3349 struct ext4_sb_info *sbi;
3350 struct ext4_extent_header *eh;
3351 struct ext4_map_blocks split_map;
3352 struct ext4_extent zero_ex;
3353 struct ext4_extent *ex, *abut_ex;
3354 ext4_lblk_t ee_block, eof_block;
3355 unsigned int ee_len, depth, map_len = map->m_len;
3356 int allocated = 0, max_zeroout = 0;
3360 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3361 "block %llu, max_blocks %u\n", inode->i_ino,
3362 (unsigned long long)map->m_lblk, map_len);
3364 sbi = EXT4_SB(inode->i_sb);
3365 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3366 inode->i_sb->s_blocksize_bits;
3367 if (eof_block < map->m_lblk + map_len)
3368 eof_block = map->m_lblk + map_len;
3370 depth = ext_depth(inode);
3371 eh = path[depth].p_hdr;
3372 ex = path[depth].p_ext;
3373 ee_block = le32_to_cpu(ex->ee_block);
3374 ee_len = ext4_ext_get_actual_len(ex);
3377 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3379 /* Pre-conditions */
3380 BUG_ON(!ext4_ext_is_uninitialized(ex));
3381 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3384 * Attempt to transfer newly initialized blocks from the currently
3385 * uninitialized extent to its neighbor. This is much cheaper
3386 * than an insertion followed by a merge as those involve costly
3387 * memmove() calls. Transferring to the left is the common case in
3388 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3389 * followed by append writes.
3391 * Limitations of the current logic:
3392 * - L1: we do not deal with writes covering the whole extent.
3393 * This would require removing the extent if the transfer
3395 * - L2: we only attempt to merge with an extent stored in the
3396 * same extent tree node.
3398 if ((map->m_lblk == ee_block) &&
3399 /* See if we can merge left */
3400 (map_len < ee_len) && /*L1*/
3401 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3402 ext4_lblk_t prev_lblk;
3403 ext4_fsblk_t prev_pblk, ee_pblk;
3404 unsigned int prev_len;
3407 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3408 prev_len = ext4_ext_get_actual_len(abut_ex);
3409 prev_pblk = ext4_ext_pblock(abut_ex);
3410 ee_pblk = ext4_ext_pblock(ex);
3413 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3414 * upon those conditions:
3415 * - C1: abut_ex is initialized,
3416 * - C2: abut_ex is logically abutting ex,
3417 * - C3: abut_ex is physically abutting ex,
3418 * - C4: abut_ex can receive the additional blocks without
3419 * overflowing the (initialized) length limit.
3421 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3422 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3423 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3424 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3425 err = ext4_ext_get_access(handle, inode, path + depth);
3429 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3432 /* Shift the start of ex by 'map_len' blocks */
3433 ex->ee_block = cpu_to_le32(ee_block + map_len);
3434 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3435 ex->ee_len = cpu_to_le16(ee_len - map_len);
3436 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3438 /* Extend abut_ex by 'map_len' blocks */
3439 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3441 /* Result: number of initialized blocks past m_lblk */
3442 allocated = map_len;
3444 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3445 (map_len < ee_len) && /*L1*/
3446 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3447 /* See if we can merge right */
3448 ext4_lblk_t next_lblk;
3449 ext4_fsblk_t next_pblk, ee_pblk;
3450 unsigned int next_len;
3453 next_lblk = le32_to_cpu(abut_ex->ee_block);
3454 next_len = ext4_ext_get_actual_len(abut_ex);
3455 next_pblk = ext4_ext_pblock(abut_ex);
3456 ee_pblk = ext4_ext_pblock(ex);
3459 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3460 * upon those conditions:
3461 * - C1: abut_ex is initialized,
3462 * - C2: abut_ex is logically abutting ex,
3463 * - C3: abut_ex is physically abutting ex,
3464 * - C4: abut_ex can receive the additional blocks without
3465 * overflowing the (initialized) length limit.
3467 if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/
3468 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3469 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3470 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3471 err = ext4_ext_get_access(handle, inode, path + depth);
3475 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3478 /* Shift the start of abut_ex by 'map_len' blocks */
3479 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3480 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3481 ex->ee_len = cpu_to_le16(ee_len - map_len);
3482 ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3484 /* Extend abut_ex by 'map_len' blocks */
3485 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3487 /* Result: number of initialized blocks past m_lblk */
3488 allocated = map_len;
3492 /* Mark the block containing both extents as dirty */
3493 ext4_ext_dirty(handle, inode, path + depth);
3495 /* Update path to point to the right extent */
3496 path[depth].p_ext = abut_ex;
3499 allocated = ee_len - (map->m_lblk - ee_block);
3501 WARN_ON(map->m_lblk < ee_block);
3503 * It is safe to convert extent to initialized via explicit
3504 * zeroout only if extent is fully inside i_size or new_size.
3506 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3508 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3509 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3510 (inode->i_sb->s_blocksize_bits - 10);
3512 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3513 if (max_zeroout && (ee_len <= max_zeroout)) {
3514 err = ext4_ext_zeroout(inode, ex);
3517 zero_ex.ee_block = ex->ee_block;
3518 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3519 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3521 err = ext4_ext_get_access(handle, inode, path + depth);
3524 ext4_ext_mark_initialized(ex);
3525 ext4_ext_try_to_merge(handle, inode, path, ex);
3526 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3532 * 1. split the extent into three extents.
3533 * 2. split the extent into two extents, zeroout the first half.
3534 * 3. split the extent into two extents, zeroout the second half.
3535 * 4. split the extent into two extents with out zeroout.
3537 split_map.m_lblk = map->m_lblk;
3538 split_map.m_len = map->m_len;
3540 if (max_zeroout && (allocated > map->m_len)) {
3541 if (allocated <= max_zeroout) {
3544 cpu_to_le32(map->m_lblk);
3545 zero_ex.ee_len = cpu_to_le16(allocated);
3546 ext4_ext_store_pblock(&zero_ex,
3547 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3548 err = ext4_ext_zeroout(inode, &zero_ex);
3551 split_map.m_lblk = map->m_lblk;
3552 split_map.m_len = allocated;
3553 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3555 if (map->m_lblk != ee_block) {
3556 zero_ex.ee_block = ex->ee_block;
3557 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3559 ext4_ext_store_pblock(&zero_ex,
3560 ext4_ext_pblock(ex));
3561 err = ext4_ext_zeroout(inode, &zero_ex);
3566 split_map.m_lblk = ee_block;
3567 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3568 allocated = map->m_len;
3572 allocated = ext4_split_extent(handle, inode, path,
3573 &split_map, split_flag, flags);
3578 /* If we have gotten a failure, don't zero out status tree */
3580 err = ext4_zeroout_es(inode, &zero_ex);
3581 return err ? err : allocated;
3585 * This function is called by ext4_ext_map_blocks() from
3586 * ext4_get_blocks_dio_write() when DIO to write
3587 * to an uninitialized extent.
3589 * Writing to an uninitialized extent may result in splitting the uninitialized
3590 * extent into multiple initialized/uninitialized extents (up to three)
3591 * There are three possibilities:
3592 * a> There is no split required: Entire extent should be uninitialized
3593 * b> Splits in two extents: Write is happening at either end of the extent
3594 * c> Splits in three extents: Somone is writing in middle of the extent
3596 * One of more index blocks maybe needed if the extent tree grow after
3597 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3598 * complete, we need to split the uninitialized extent before DIO submit
3599 * the IO. The uninitialized extent called at this time will be split
3600 * into three uninitialized extent(at most). After IO complete, the part
3601 * being filled will be convert to initialized by the end_io callback function
3602 * via ext4_convert_unwritten_extents().
3604 * Returns the size of uninitialized extent to be written on success.
3606 static int ext4_split_unwritten_extents(handle_t *handle,
3607 struct inode *inode,
3608 struct ext4_map_blocks *map,
3609 struct ext4_ext_path *path,
3612 ext4_lblk_t eof_block;
3613 ext4_lblk_t ee_block;
3614 struct ext4_extent *ex;
3615 unsigned int ee_len;
3616 int split_flag = 0, depth;
3618 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3619 "block %llu, max_blocks %u\n", inode->i_ino,
3620 (unsigned long long)map->m_lblk, map->m_len);
3622 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3623 inode->i_sb->s_blocksize_bits;
3624 if (eof_block < map->m_lblk + map->m_len)
3625 eof_block = map->m_lblk + map->m_len;
3627 * It is safe to convert extent to initialized via explicit
3628 * zeroout only if extent is fully insde i_size or new_size.
3630 depth = ext_depth(inode);
3631 ex = path[depth].p_ext;
3632 ee_block = le32_to_cpu(ex->ee_block);
3633 ee_len = ext4_ext_get_actual_len(ex);
3635 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3636 split_flag |= EXT4_EXT_MARK_UNINIT2;
3637 if (flags & EXT4_GET_BLOCKS_CONVERT)
3638 split_flag |= EXT4_EXT_DATA_VALID2;
3639 flags |= EXT4_GET_BLOCKS_PRE_IO;
3640 return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3643 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3644 struct inode *inode,
3645 struct ext4_map_blocks *map,
3646 struct ext4_ext_path *path)
3648 struct ext4_extent *ex;
3649 ext4_lblk_t ee_block;
3650 unsigned int ee_len;
3654 depth = ext_depth(inode);
3655 ex = path[depth].p_ext;
3656 ee_block = le32_to_cpu(ex->ee_block);
3657 ee_len = ext4_ext_get_actual_len(ex);
3659 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3660 "block %llu, max_blocks %u\n", inode->i_ino,
3661 (unsigned long long)ee_block, ee_len);
3663 /* If extent is larger than requested it is a clear sign that we still
3664 * have some extent state machine issues left. So extent_split is still
3666 * TODO: Once all related issues will be fixed this situation should be
3669 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3671 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3672 " len %u; IO logical block %llu, len %u\n",
3673 inode->i_ino, (unsigned long long)ee_block, ee_len,
3674 (unsigned long long)map->m_lblk, map->m_len);
3676 err = ext4_split_unwritten_extents(handle, inode, map, path,
3677 EXT4_GET_BLOCKS_CONVERT);
3680 ext4_ext_drop_refs(path);
3681 path = ext4_ext_find_extent(inode, map->m_lblk, path, 0);
3683 err = PTR_ERR(path);
3686 depth = ext_depth(inode);
3687 ex = path[depth].p_ext;
3690 err = ext4_ext_get_access(handle, inode, path + depth);
3693 /* first mark the extent as initialized */
3694 ext4_ext_mark_initialized(ex);
3696 /* note: ext4_ext_correct_indexes() isn't needed here because
3697 * borders are not changed
3699 ext4_ext_try_to_merge(handle, inode, path, ex);
3701 /* Mark modified extent as dirty */
3702 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3704 ext4_ext_show_leaf(inode, path);
3708 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3709 sector_t block, int count)
3712 for (i = 0; i < count; i++)
3713 unmap_underlying_metadata(bdev, block + i);
3717 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3719 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3721 struct ext4_ext_path *path,
3725 struct ext4_extent_header *eh;
3726 struct ext4_extent *last_ex;
3728 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3731 depth = ext_depth(inode);
3732 eh = path[depth].p_hdr;
3735 * We're going to remove EOFBLOCKS_FL entirely in future so we
3736 * do not care for this case anymore. Simply remove the flag
3737 * if there are no extents.
3739 if (unlikely(!eh->eh_entries))
3741 last_ex = EXT_LAST_EXTENT(eh);
3743 * We should clear the EOFBLOCKS_FL flag if we are writing the
3744 * last block in the last extent in the file. We test this by
3745 * first checking to see if the caller to
3746 * ext4_ext_get_blocks() was interested in the last block (or
3747 * a block beyond the last block) in the current extent. If
3748 * this turns out to be false, we can bail out from this
3749 * function immediately.
3751 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3752 ext4_ext_get_actual_len(last_ex))
3755 * If the caller does appear to be planning to write at or
3756 * beyond the end of the current extent, we then test to see
3757 * if the current extent is the last extent in the file, by
3758 * checking to make sure it was reached via the rightmost node
3759 * at each level of the tree.
3761 for (i = depth-1; i >= 0; i--)
3762 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3765 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3766 return ext4_mark_inode_dirty(handle, inode);
3770 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3772 * Return 1 if there is a delalloc block in the range, otherwise 0.
3774 int ext4_find_delalloc_range(struct inode *inode,
3775 ext4_lblk_t lblk_start,
3776 ext4_lblk_t lblk_end)
3778 struct extent_status es;
3780 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3782 return 0; /* there is no delay extent in this tree */
3783 else if (es.es_lblk <= lblk_start &&
3784 lblk_start < es.es_lblk + es.es_len)
3786 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3792 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3794 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3795 ext4_lblk_t lblk_start, lblk_end;
3796 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3797 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3799 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3803 * Determines how many complete clusters (out of those specified by the 'map')
3804 * are under delalloc and were reserved quota for.
3805 * This function is called when we are writing out the blocks that were
3806 * originally written with their allocation delayed, but then the space was
3807 * allocated using fallocate() before the delayed allocation could be resolved.
3808 * The cases to look for are:
3809 * ('=' indicated delayed allocated blocks
3810 * '-' indicates non-delayed allocated blocks)
3811 * (a) partial clusters towards beginning and/or end outside of allocated range
3812 * are not delalloc'ed.
3814 * |----c---=|====c====|====c====|===-c----|
3815 * |++++++ allocated ++++++|
3816 * ==> 4 complete clusters in above example
3818 * (b) partial cluster (outside of allocated range) towards either end is
3819 * marked for delayed allocation. In this case, we will exclude that
3822 * |----====c========|========c========|
3823 * |++++++ allocated ++++++|
3824 * ==> 1 complete clusters in above example
3827 * |================c================|
3828 * |++++++ allocated ++++++|
3829 * ==> 0 complete clusters in above example
3831 * The ext4_da_update_reserve_space will be called only if we
3832 * determine here that there were some "entire" clusters that span
3833 * this 'allocated' range.
3834 * In the non-bigalloc case, this function will just end up returning num_blks
3835 * without ever calling ext4_find_delalloc_range.
3838 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3839 unsigned int num_blks)
3841 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3842 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3843 ext4_lblk_t lblk_from, lblk_to, c_offset;
3844 unsigned int allocated_clusters = 0;
3846 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3847 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3849 /* max possible clusters for this allocation */
3850 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3852 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3854 /* Check towards left side */
3855 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3857 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3858 lblk_to = lblk_from + c_offset - 1;
3860 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3861 allocated_clusters--;
3864 /* Now check towards right. */
3865 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3866 if (allocated_clusters && c_offset) {
3867 lblk_from = lblk_start + num_blks;
3868 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3870 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3871 allocated_clusters--;
3874 return allocated_clusters;
3878 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3879 struct ext4_map_blocks *map,
3880 struct ext4_ext_path *path, int flags,
3881 unsigned int allocated, ext4_fsblk_t newblock)
3885 ext4_io_end_t *io = ext4_inode_aio(inode);
3887 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3888 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3889 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3891 ext4_ext_show_leaf(inode, path);
3894 * When writing into uninitialized space, we should not fail to
3895 * allocate metadata blocks for the new extent block if needed.
3897 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3899 trace_ext4_ext_handle_uninitialized_extents(inode, map, flags,
3900 allocated, newblock);
3902 /* get_block() before submit the IO, split the extent */
3903 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3904 ret = ext4_split_unwritten_extents(handle, inode, map,
3909 * Flag the inode(non aio case) or end_io struct (aio case)
3910 * that this IO needs to conversion to written when IO is
3914 ext4_set_io_unwritten_flag(inode, io);
3916 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3917 map->m_flags |= EXT4_MAP_UNWRITTEN;
3918 if (ext4_should_dioread_nolock(inode))
3919 map->m_flags |= EXT4_MAP_UNINIT;
3922 /* IO end_io complete, convert the filled extent to written */
3923 if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3924 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3927 ext4_update_inode_fsync_trans(handle, inode, 1);
3928 err = check_eofblocks_fl(handle, inode, map->m_lblk,
3932 map->m_flags |= EXT4_MAP_MAPPED;
3933 map->m_pblk = newblock;
3934 if (allocated > map->m_len)
3935 allocated = map->m_len;
3936 map->m_len = allocated;
3939 /* buffered IO case */
3941 * repeat fallocate creation request
3942 * we already have an unwritten extent
3944 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) {
3945 map->m_flags |= EXT4_MAP_UNWRITTEN;
3949 /* buffered READ or buffered write_begin() lookup */
3950 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3952 * We have blocks reserved already. We
3953 * return allocated blocks so that delalloc
3954 * won't do block reservation for us. But
3955 * the buffer head will be unmapped so that
3956 * a read from the block returns 0s.
3958 map->m_flags |= EXT4_MAP_UNWRITTEN;
3962 /* buffered write, writepage time, convert*/
3963 ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags);
3965 ext4_update_inode_fsync_trans(handle, inode, 1);
3972 map->m_flags |= EXT4_MAP_NEW;
3974 * if we allocated more blocks than requested
3975 * we need to make sure we unmap the extra block
3976 * allocated. The actual needed block will get
3977 * unmapped later when we find the buffer_head marked
3980 if (allocated > map->m_len) {
3981 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3982 newblock + map->m_len,
3983 allocated - map->m_len);
3984 allocated = map->m_len;
3986 map->m_len = allocated;
3989 * If we have done fallocate with the offset that is already
3990 * delayed allocated, we would have block reservation
3991 * and quota reservation done in the delayed write path.
3992 * But fallocate would have already updated quota and block
3993 * count for this offset. So cancel these reservation
3995 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3996 unsigned int reserved_clusters;
3997 reserved_clusters = get_reserved_cluster_alloc(inode,
3998 map->m_lblk, map->m_len);
3999 if (reserved_clusters)
4000 ext4_da_update_reserve_space(inode,
4006 map->m_flags |= EXT4_MAP_MAPPED;
4007 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4008 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4014 if (allocated > map->m_len)
4015 allocated = map->m_len;
4016 ext4_ext_show_leaf(inode, path);
4017 map->m_pblk = newblock;
4018 map->m_len = allocated;
4021 ext4_ext_drop_refs(path);
4024 return err ? err : allocated;
4028 * get_implied_cluster_alloc - check to see if the requested
4029 * allocation (in the map structure) overlaps with a cluster already
4030 * allocated in an extent.
4031 * @sb The filesystem superblock structure
4032 * @map The requested lblk->pblk mapping
4033 * @ex The extent structure which might contain an implied
4034 * cluster allocation
4036 * This function is called by ext4_ext_map_blocks() after we failed to
4037 * find blocks that were already in the inode's extent tree. Hence,
4038 * we know that the beginning of the requested region cannot overlap
4039 * the extent from the inode's extent tree. There are three cases we
4040 * want to catch. The first is this case:
4042 * |--- cluster # N--|
4043 * |--- extent ---| |---- requested region ---|
4046 * The second case that we need to test for is this one:
4048 * |--------- cluster # N ----------------|
4049 * |--- requested region --| |------- extent ----|
4050 * |=======================|
4052 * The third case is when the requested region lies between two extents
4053 * within the same cluster:
4054 * |------------- cluster # N-------------|
4055 * |----- ex -----| |---- ex_right ----|
4056 * |------ requested region ------|
4057 * |================|
4059 * In each of the above cases, we need to set the map->m_pblk and
4060 * map->m_len so it corresponds to the return the extent labelled as
4061 * "|====|" from cluster #N, since it is already in use for data in
4062 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4063 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4064 * as a new "allocated" block region. Otherwise, we will return 0 and
4065 * ext4_ext_map_blocks() will then allocate one or more new clusters
4066 * by calling ext4_mb_new_blocks().
4068 static int get_implied_cluster_alloc(struct super_block *sb,
4069 struct ext4_map_blocks *map,
4070 struct ext4_extent *ex,
4071 struct ext4_ext_path *path)
4073 struct ext4_sb_info *sbi = EXT4_SB(sb);
4074 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4075 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4076 ext4_lblk_t rr_cluster_start;
4077 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4078 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4079 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4081 /* The extent passed in that we are trying to match */
4082 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4083 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4085 /* The requested region passed into ext4_map_blocks() */
4086 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4088 if ((rr_cluster_start == ex_cluster_end) ||
4089 (rr_cluster_start == ex_cluster_start)) {
4090 if (rr_cluster_start == ex_cluster_end)
4091 ee_start += ee_len - 1;
4092 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4093 map->m_len = min(map->m_len,
4094 (unsigned) sbi->s_cluster_ratio - c_offset);
4096 * Check for and handle this case:
4098 * |--------- cluster # N-------------|
4099 * |------- extent ----|
4100 * |--- requested region ---|
4104 if (map->m_lblk < ee_block)
4105 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4108 * Check for the case where there is already another allocated
4109 * block to the right of 'ex' but before the end of the cluster.
4111 * |------------- cluster # N-------------|
4112 * |----- ex -----| |---- ex_right ----|
4113 * |------ requested region ------|
4114 * |================|
4116 if (map->m_lblk > ee_block) {
4117 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4118 map->m_len = min(map->m_len, next - map->m_lblk);
4121 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4125 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4131 * Block allocation/map/preallocation routine for extents based files
4134 * Need to be called with
4135 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4136 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4138 * return > 0, number of of blocks already mapped/allocated
4139 * if create == 0 and these are pre-allocated blocks
4140 * buffer head is unmapped
4141 * otherwise blocks are mapped
4143 * return = 0, if plain look up failed (blocks have not been allocated)
4144 * buffer head is unmapped
4146 * return < 0, error case.
4148 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4149 struct ext4_map_blocks *map, int flags)
4151 struct ext4_ext_path *path = NULL;
4152 struct ext4_extent newex, *ex, *ex2;
4153 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4154 ext4_fsblk_t newblock = 0;
4155 int free_on_err = 0, err = 0, depth, ret;
4156 unsigned int allocated = 0, offset = 0;
4157 unsigned int allocated_clusters = 0;
4158 struct ext4_allocation_request ar;
4159 ext4_io_end_t *io = ext4_inode_aio(inode);
4160 ext4_lblk_t cluster_offset;
4161 int set_unwritten = 0;
4163 ext_debug("blocks %u/%u requested for inode %lu\n",
4164 map->m_lblk, map->m_len, inode->i_ino);
4165 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4167 /* find extent for this block */
4168 path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0);
4170 err = PTR_ERR(path);
4175 depth = ext_depth(inode);
4178 * consistent leaf must not be empty;
4179 * this situation is possible, though, _during_ tree modification;
4180 * this is why assert can't be put in ext4_ext_find_extent()
4182 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4183 EXT4_ERROR_INODE(inode, "bad extent address "
4184 "lblock: %lu, depth: %d pblock %lld",
4185 (unsigned long) map->m_lblk, depth,
4186 path[depth].p_block);
4191 ex = path[depth].p_ext;
4193 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4194 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4195 unsigned short ee_len;
4198 * Uninitialized extents are treated as holes, except that
4199 * we split out initialized portions during a write.
4201 ee_len = ext4_ext_get_actual_len(ex);
4203 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4205 /* if found extent covers block, simply return it */
4206 if (in_range(map->m_lblk, ee_block, ee_len)) {
4207 newblock = map->m_lblk - ee_block + ee_start;
4208 /* number of remaining blocks in the extent */
4209 allocated = ee_len - (map->m_lblk - ee_block);
4210 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4211 ee_block, ee_len, newblock);
4213 if (!ext4_ext_is_uninitialized(ex))
4216 ret = ext4_ext_handle_uninitialized_extents(
4217 handle, inode, map, path, flags,
4218 allocated, newblock);
4227 if ((sbi->s_cluster_ratio > 1) &&
4228 ext4_find_delalloc_cluster(inode, map->m_lblk))
4229 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4232 * requested block isn't allocated yet;
4233 * we couldn't try to create block if create flag is zero
4235 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4237 * put just found gap into cache to speed up
4238 * subsequent requests
4240 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4241 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4246 * Okay, we need to do block allocation.
4248 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4249 newex.ee_block = cpu_to_le32(map->m_lblk);
4250 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4253 * If we are doing bigalloc, check to see if the extent returned
4254 * by ext4_ext_find_extent() implies a cluster we can use.
4256 if (cluster_offset && ex &&
4257 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4258 ar.len = allocated = map->m_len;
4259 newblock = map->m_pblk;
4260 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4261 goto got_allocated_blocks;
4264 /* find neighbour allocated blocks */
4265 ar.lleft = map->m_lblk;
4266 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4269 ar.lright = map->m_lblk;
4271 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4275 /* Check if the extent after searching to the right implies a
4276 * cluster we can use. */
4277 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4278 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4279 ar.len = allocated = map->m_len;
4280 newblock = map->m_pblk;
4281 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4282 goto got_allocated_blocks;
4286 * See if request is beyond maximum number of blocks we can have in
4287 * a single extent. For an initialized extent this limit is
4288 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4289 * EXT_UNINIT_MAX_LEN.
4291 if (map->m_len > EXT_INIT_MAX_LEN &&
4292 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4293 map->m_len = EXT_INIT_MAX_LEN;
4294 else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4295 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4296 map->m_len = EXT_UNINIT_MAX_LEN;
4298 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4299 newex.ee_len = cpu_to_le16(map->m_len);
4300 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4302 allocated = ext4_ext_get_actual_len(&newex);
4304 allocated = map->m_len;
4306 /* allocate new block */
4308 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4309 ar.logical = map->m_lblk;
4311 * We calculate the offset from the beginning of the cluster
4312 * for the logical block number, since when we allocate a
4313 * physical cluster, the physical block should start at the
4314 * same offset from the beginning of the cluster. This is
4315 * needed so that future calls to get_implied_cluster_alloc()
4318 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4319 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4321 ar.logical -= offset;
4322 if (S_ISREG(inode->i_mode))
4323 ar.flags = EXT4_MB_HINT_DATA;
4325 /* disable in-core preallocation for non-regular files */
4327 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4328 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4329 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4332 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4333 ar.goal, newblock, allocated);
4335 allocated_clusters = ar.len;
4336 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4337 if (ar.len > allocated)
4340 got_allocated_blocks:
4341 /* try to insert new extent into found leaf and return */
4342 ext4_ext_store_pblock(&newex, newblock + offset);
4343 newex.ee_len = cpu_to_le16(ar.len);
4344 /* Mark uninitialized */
4345 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4346 ext4_ext_mark_uninitialized(&newex);
4347 map->m_flags |= EXT4_MAP_UNWRITTEN;
4349 * io_end structure was created for every IO write to an
4350 * uninitialized extent. To avoid unnecessary conversion,
4351 * here we flag the IO that really needs the conversion.
4352 * For non asycn direct IO case, flag the inode state
4353 * that we need to perform conversion when IO is done.
4355 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4357 if (ext4_should_dioread_nolock(inode))
4358 map->m_flags |= EXT4_MAP_UNINIT;
4362 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4363 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4366 err = ext4_ext_insert_extent(handle, inode, path,
4369 if (!err && set_unwritten) {
4371 ext4_set_io_unwritten_flag(inode, io);
4373 ext4_set_inode_state(inode,
4374 EXT4_STATE_DIO_UNWRITTEN);
4377 if (err && free_on_err) {
4378 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4379 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4380 /* free data blocks we just allocated */
4381 /* not a good idea to call discard here directly,
4382 * but otherwise we'd need to call it every free() */
4383 ext4_discard_preallocations(inode);
4384 ext4_free_blocks(handle, inode, NULL, newblock,
4385 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4389 /* previous routine could use block we allocated */
4390 newblock = ext4_ext_pblock(&newex);
4391 allocated = ext4_ext_get_actual_len(&newex);
4392 if (allocated > map->m_len)
4393 allocated = map->m_len;
4394 map->m_flags |= EXT4_MAP_NEW;
4397 * Update reserved blocks/metadata blocks after successful
4398 * block allocation which had been deferred till now.
4400 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4401 unsigned int reserved_clusters;
4403 * Check how many clusters we had reserved this allocated range
4405 reserved_clusters = get_reserved_cluster_alloc(inode,
4406 map->m_lblk, allocated);
4407 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4408 if (reserved_clusters) {
4410 * We have clusters reserved for this range.
4411 * But since we are not doing actual allocation
4412 * and are simply using blocks from previously
4413 * allocated cluster, we should release the
4414 * reservation and not claim quota.
4416 ext4_da_update_reserve_space(inode,
4417 reserved_clusters, 0);
4420 BUG_ON(allocated_clusters < reserved_clusters);
4421 if (reserved_clusters < allocated_clusters) {
4422 struct ext4_inode_info *ei = EXT4_I(inode);
4423 int reservation = allocated_clusters -
4426 * It seems we claimed few clusters outside of
4427 * the range of this allocation. We should give
4428 * it back to the reservation pool. This can
4429 * happen in the following case:
4431 * * Suppose s_cluster_ratio is 4 (i.e., each
4432 * cluster has 4 blocks. Thus, the clusters
4433 * are [0-3],[4-7],[8-11]...
4434 * * First comes delayed allocation write for
4435 * logical blocks 10 & 11. Since there were no
4436 * previous delayed allocated blocks in the
4437 * range [8-11], we would reserve 1 cluster
4439 * * Next comes write for logical blocks 3 to 8.
4440 * In this case, we will reserve 2 clusters
4441 * (for [0-3] and [4-7]; and not for [8-11] as
4442 * that range has a delayed allocated blocks.
4443 * Thus total reserved clusters now becomes 3.
4444 * * Now, during the delayed allocation writeout
4445 * time, we will first write blocks [3-8] and
4446 * allocate 3 clusters for writing these
4447 * blocks. Also, we would claim all these
4448 * three clusters above.
4449 * * Now when we come here to writeout the
4450 * blocks [10-11], we would expect to claim
4451 * the reservation of 1 cluster we had made
4452 * (and we would claim it since there are no
4453 * more delayed allocated blocks in the range
4454 * [8-11]. But our reserved cluster count had
4455 * already gone to 0.
4457 * Thus, at the step 4 above when we determine
4458 * that there are still some unwritten delayed
4459 * allocated blocks outside of our current
4460 * block range, we should increment the
4461 * reserved clusters count so that when the
4462 * remaining blocks finally gets written, we
4465 dquot_reserve_block(inode,
4466 EXT4_C2B(sbi, reservation));
4467 spin_lock(&ei->i_block_reservation_lock);
4468 ei->i_reserved_data_blocks += reservation;
4469 spin_unlock(&ei->i_block_reservation_lock);
4472 * We will claim quota for all newly allocated blocks.
4473 * We're updating the reserved space *after* the
4474 * correction above so we do not accidentally free
4475 * all the metadata reservation because we might
4476 * actually need it later on.
4478 ext4_da_update_reserve_space(inode, allocated_clusters,
4484 * Cache the extent and update transaction to commit on fdatasync only
4485 * when it is _not_ an uninitialized extent.
4487 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
4488 ext4_update_inode_fsync_trans(handle, inode, 1);
4490 ext4_update_inode_fsync_trans(handle, inode, 0);
4492 if (allocated > map->m_len)
4493 allocated = map->m_len;
4494 ext4_ext_show_leaf(inode, path);
4495 map->m_flags |= EXT4_MAP_MAPPED;
4496 map->m_pblk = newblock;
4497 map->m_len = allocated;
4500 ext4_ext_drop_refs(path);
4505 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4506 err ? err : allocated);
4507 ext4_es_lru_add(inode);
4508 return err ? err : allocated;
4511 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4513 struct super_block *sb = inode->i_sb;
4514 ext4_lblk_t last_block;
4518 * TODO: optimization is possible here.
4519 * Probably we need not scan at all,
4520 * because page truncation is enough.
4523 /* we have to know where to truncate from in crash case */
4524 EXT4_I(inode)->i_disksize = inode->i_size;
4525 ext4_mark_inode_dirty(handle, inode);
4527 last_block = (inode->i_size + sb->s_blocksize - 1)
4528 >> EXT4_BLOCK_SIZE_BITS(sb);
4530 err = ext4_es_remove_extent(inode, last_block,
4531 EXT_MAX_BLOCKS - last_block);
4532 if (err == -ENOMEM) {
4534 congestion_wait(BLK_RW_ASYNC, HZ/50);
4538 ext4_std_error(inode->i_sb, err);
4541 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4542 ext4_std_error(inode->i_sb, err);
4545 static void ext4_falloc_update_inode(struct inode *inode,
4546 int mode, loff_t new_size, int update_ctime)
4548 struct timespec now;
4551 now = current_fs_time(inode->i_sb);
4552 if (!timespec_equal(&inode->i_ctime, &now))
4553 inode->i_ctime = now;
4556 * Update only when preallocation was requested beyond
4559 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4560 if (new_size > i_size_read(inode))
4561 i_size_write(inode, new_size);
4562 if (new_size > EXT4_I(inode)->i_disksize)
4563 ext4_update_i_disksize(inode, new_size);
4566 * Mark that we allocate beyond EOF so the subsequent truncate
4567 * can proceed even if the new size is the same as i_size.
4569 if (new_size > i_size_read(inode))
4570 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4576 * preallocate space for a file. This implements ext4's fallocate file
4577 * operation, which gets called from sys_fallocate system call.
4578 * For block-mapped files, posix_fallocate should fall back to the method
4579 * of writing zeroes to the required new blocks (the same behavior which is
4580 * expected for file systems which do not support fallocate() system call).
4582 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4584 struct inode *inode = file_inode(file);
4587 unsigned int max_blocks;
4592 struct ext4_map_blocks map;
4593 unsigned int credits, blkbits = inode->i_blkbits;
4595 /* Return error if mode is not supported */
4596 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4599 if (mode & FALLOC_FL_PUNCH_HOLE)
4600 return ext4_punch_hole(inode, offset, len);
4602 ret = ext4_convert_inline_data(inode);
4607 * currently supporting (pre)allocate mode for extent-based
4610 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4613 trace_ext4_fallocate_enter(inode, offset, len, mode);
4614 map.m_lblk = offset >> blkbits;
4616 * We can't just convert len to max_blocks because
4617 * If blocksize = 4096 offset = 3072 and len = 2048
4619 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4622 * credits to insert 1 extent into extent tree
4624 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4625 mutex_lock(&inode->i_mutex);
4626 ret = inode_newsize_ok(inode, (len + offset));
4628 mutex_unlock(&inode->i_mutex);
4629 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4632 flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4633 if (mode & FALLOC_FL_KEEP_SIZE)
4634 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4636 * Don't normalize the request if it can fit in one extent so
4637 * that it doesn't get unnecessarily split into multiple
4640 if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4641 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4644 while (ret >= 0 && ret < max_blocks) {
4645 map.m_lblk = map.m_lblk + ret;
4646 map.m_len = max_blocks = max_blocks - ret;
4647 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4649 if (IS_ERR(handle)) {
4650 ret = PTR_ERR(handle);
4653 ret = ext4_map_blocks(handle, inode, &map, flags);
4656 ext4_warning(inode->i_sb,
4657 "inode #%lu: block %u: len %u: "
4658 "ext4_ext_map_blocks returned %d",
4659 inode->i_ino, map.m_lblk,
4662 ext4_mark_inode_dirty(handle, inode);
4663 ret2 = ext4_journal_stop(handle);
4666 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4667 blkbits) >> blkbits))
4668 new_size = offset + len;
4670 new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4672 ext4_falloc_update_inode(inode, mode, new_size,
4673 (map.m_flags & EXT4_MAP_NEW));
4674 ext4_mark_inode_dirty(handle, inode);
4675 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4676 ext4_handle_sync(handle);
4677 ret2 = ext4_journal_stop(handle);
4681 if (ret == -ENOSPC &&
4682 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4686 mutex_unlock(&inode->i_mutex);
4687 trace_ext4_fallocate_exit(inode, offset, max_blocks,
4688 ret > 0 ? ret2 : ret);
4689 return ret > 0 ? ret2 : ret;
4693 * This function convert a range of blocks to written extents
4694 * The caller of this function will pass the start offset and the size.
4695 * all unwritten extents within this range will be converted to
4698 * This function is called from the direct IO end io call back
4699 * function, to convert the fallocated extents after IO is completed.
4700 * Returns 0 on success.
4702 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4703 loff_t offset, ssize_t len)
4705 unsigned int max_blocks;
4708 struct ext4_map_blocks map;
4709 unsigned int credits, blkbits = inode->i_blkbits;
4711 map.m_lblk = offset >> blkbits;
4713 * We can't just convert len to max_blocks because
4714 * If blocksize = 4096 offset = 3072 and len = 2048
4716 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4719 * This is somewhat ugly but the idea is clear: When transaction is
4720 * reserved, everything goes into it. Otherwise we rather start several
4721 * smaller transactions for conversion of each extent separately.
4724 handle = ext4_journal_start_reserved(handle,
4725 EXT4_HT_EXT_CONVERT);
4727 return PTR_ERR(handle);
4731 * credits to insert 1 extent into extent tree
4733 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4735 while (ret >= 0 && ret < max_blocks) {
4737 map.m_len = (max_blocks -= ret);
4739 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4741 if (IS_ERR(handle)) {
4742 ret = PTR_ERR(handle);
4746 ret = ext4_map_blocks(handle, inode, &map,
4747 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4749 ext4_warning(inode->i_sb,
4750 "inode #%lu: block %u: len %u: "
4751 "ext4_ext_map_blocks returned %d",
4752 inode->i_ino, map.m_lblk,
4754 ext4_mark_inode_dirty(handle, inode);
4756 ret2 = ext4_journal_stop(handle);
4757 if (ret <= 0 || ret2)
4761 ret2 = ext4_journal_stop(handle);
4762 return ret > 0 ? ret2 : ret;
4766 * If newes is not existing extent (newes->ec_pblk equals zero) find
4767 * delayed extent at start of newes and update newes accordingly and
4768 * return start of the next delayed extent.
4770 * If newes is existing extent (newes->ec_pblk is not equal zero)
4771 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4772 * extent found. Leave newes unmodified.
4774 static int ext4_find_delayed_extent(struct inode *inode,
4775 struct extent_status *newes)
4777 struct extent_status es;
4778 ext4_lblk_t block, next_del;
4780 if (newes->es_pblk == 0) {
4781 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
4782 newes->es_lblk + newes->es_len - 1, &es);
4785 * No extent in extent-tree contains block @newes->es_pblk,
4786 * then the block may stay in 1)a hole or 2)delayed-extent.
4792 if (es.es_lblk > newes->es_lblk) {
4794 newes->es_len = min(es.es_lblk - newes->es_lblk,
4799 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
4802 block = newes->es_lblk + newes->es_len;
4803 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
4805 next_del = EXT_MAX_BLOCKS;
4807 next_del = es.es_lblk;
4811 /* fiemap flags we can handle specified here */
4812 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4814 static int ext4_xattr_fiemap(struct inode *inode,
4815 struct fiemap_extent_info *fieinfo)
4819 __u32 flags = FIEMAP_EXTENT_LAST;
4820 int blockbits = inode->i_sb->s_blocksize_bits;
4824 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4825 struct ext4_iloc iloc;
4826 int offset; /* offset of xattr in inode */
4828 error = ext4_get_inode_loc(inode, &iloc);
4831 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4832 offset = EXT4_GOOD_OLD_INODE_SIZE +
4833 EXT4_I(inode)->i_extra_isize;
4835 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4836 flags |= FIEMAP_EXTENT_DATA_INLINE;
4838 } else { /* external block */
4839 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4840 length = inode->i_sb->s_blocksize;
4844 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4846 return (error < 0 ? error : 0);
4849 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4850 __u64 start, __u64 len)
4852 ext4_lblk_t start_blk;
4855 if (ext4_has_inline_data(inode)) {
4858 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
4864 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4865 error = ext4_ext_precache(inode);
4870 /* fallback to generic here if not in extents fmt */
4871 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4872 return generic_block_fiemap(inode, fieinfo, start, len,
4875 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4878 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4879 error = ext4_xattr_fiemap(inode, fieinfo);
4881 ext4_lblk_t len_blks;
4884 start_blk = start >> inode->i_sb->s_blocksize_bits;
4885 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4886 if (last_blk >= EXT_MAX_BLOCKS)
4887 last_blk = EXT_MAX_BLOCKS-1;
4888 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4891 * Walk the extent tree gathering extent information
4892 * and pushing extents back to the user.
4894 error = ext4_fill_fiemap_extents(inode, start_blk,
4897 ext4_es_lru_add(inode);