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 <asm/uaccess.h>
41 #include <linux/fiemap.h>
42 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
54 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static __le32 ext4_extent_block_csum(struct inode *inode,
60 struct ext4_extent_header *eh)
62 struct ext4_inode_info *ei = EXT4_I(inode);
63 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
67 EXT4_EXTENT_TAIL_OFFSET(eh));
68 return cpu_to_le32(csum);
71 static int ext4_extent_block_csum_verify(struct inode *inode,
72 struct ext4_extent_header *eh)
74 struct ext4_extent_tail *et;
76 if (!ext4_has_metadata_csum(inode->i_sb))
79 et = find_ext4_extent_tail(eh);
80 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
85 static void ext4_extent_block_csum_set(struct inode *inode,
86 struct ext4_extent_header *eh)
88 struct ext4_extent_tail *et;
90 if (!ext4_has_metadata_csum(inode->i_sb))
93 et = find_ext4_extent_tail(eh);
94 et->et_checksum = ext4_extent_block_csum(inode, eh);
97 static int ext4_split_extent(handle_t *handle,
99 struct ext4_ext_path **ppath,
100 struct ext4_map_blocks *map,
104 static int ext4_split_extent_at(handle_t *handle,
106 struct ext4_ext_path **ppath,
111 static int ext4_find_delayed_extent(struct inode *inode,
112 struct extent_status *newes);
114 static int ext4_ext_truncate_extend_restart(handle_t *handle,
120 if (!ext4_handle_valid(handle))
122 if (handle->h_buffer_credits > needed)
124 err = ext4_journal_extend(handle, needed);
127 err = ext4_truncate_restart_trans(handle, inode, needed);
139 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
140 struct ext4_ext_path *path)
143 /* path points to block */
144 BUFFER_TRACE(path->p_bh, "get_write_access");
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 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
165 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
166 /* path points to block */
167 err = __ext4_handle_dirty_metadata(where, line, handle,
170 /* path points to leaf/index in inode body */
171 err = ext4_mark_inode_dirty(handle, inode);
176 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
177 struct ext4_ext_path *path,
181 int depth = path->p_depth;
182 struct ext4_extent *ex;
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
198 * especially if the latter case turns out to be
201 ex = path[depth].p_ext;
203 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
204 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
206 if (block > ext_block)
207 return ext_pblk + (block - ext_block);
209 return ext_pblk - (ext_block - block);
212 /* it looks like index is empty;
213 * try to find starting block from index itself */
214 if (path[depth].p_bh)
215 return path[depth].p_bh->b_blocknr;
218 /* OK. use inode's group */
219 return ext4_inode_to_goal_block(inode);
223 * Allocation for a meta data block
226 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
227 struct ext4_ext_path *path,
228 struct ext4_extent *ex, int *err, unsigned int flags)
230 ext4_fsblk_t goal, newblock;
232 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
233 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
238 static inline int ext4_ext_space_block(struct inode *inode, int check)
242 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
243 / sizeof(struct ext4_extent);
244 #ifdef AGGRESSIVE_TEST
245 if (!check && size > 6)
251 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
255 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
256 / sizeof(struct ext4_extent_idx);
257 #ifdef AGGRESSIVE_TEST
258 if (!check && size > 5)
264 static inline int ext4_ext_space_root(struct inode *inode, int check)
268 size = sizeof(EXT4_I(inode)->i_data);
269 size -= sizeof(struct ext4_extent_header);
270 size /= sizeof(struct ext4_extent);
271 #ifdef AGGRESSIVE_TEST
272 if (!check && size > 3)
278 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
282 size = sizeof(EXT4_I(inode)->i_data);
283 size -= sizeof(struct ext4_extent_header);
284 size /= sizeof(struct ext4_extent_idx);
285 #ifdef AGGRESSIVE_TEST
286 if (!check && size > 4)
293 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
294 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
297 struct ext4_ext_path *path = *ppath;
298 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
300 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
301 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
302 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
303 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
307 * Calculate the number of metadata blocks needed
308 * to allocate @blocks
309 * Worse case is one block per extent
311 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
313 struct ext4_inode_info *ei = EXT4_I(inode);
316 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
317 / sizeof(struct ext4_extent_idx));
320 * If the new delayed allocation block is contiguous with the
321 * previous da block, it can share index blocks with the
322 * previous block, so we only need to allocate a new index
323 * block every idxs leaf blocks. At ldxs**2 blocks, we need
324 * an additional index block, and at ldxs**3 blocks, yet
325 * another index blocks.
327 if (ei->i_da_metadata_calc_len &&
328 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
331 if ((ei->i_da_metadata_calc_len % idxs) == 0)
333 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
335 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
337 ei->i_da_metadata_calc_len = 0;
339 ei->i_da_metadata_calc_len++;
340 ei->i_da_metadata_calc_last_lblock++;
345 * In the worst case we need a new set of index blocks at
346 * every level of the inode's extent tree.
348 ei->i_da_metadata_calc_len = 1;
349 ei->i_da_metadata_calc_last_lblock = lblock;
350 return ext_depth(inode) + 1;
354 ext4_ext_max_entries(struct inode *inode, int depth)
358 if (depth == ext_depth(inode)) {
360 max = ext4_ext_space_root(inode, 1);
362 max = ext4_ext_space_root_idx(inode, 1);
365 max = ext4_ext_space_block(inode, 1);
367 max = ext4_ext_space_block_idx(inode, 1);
373 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
375 ext4_fsblk_t block = ext4_ext_pblock(ext);
376 int len = ext4_ext_get_actual_len(ext);
377 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
378 ext4_lblk_t last = lblock + len - 1;
382 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
385 static int ext4_valid_extent_idx(struct inode *inode,
386 struct ext4_extent_idx *ext_idx)
388 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
390 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
393 static int ext4_valid_extent_entries(struct inode *inode,
394 struct ext4_extent_header *eh,
397 unsigned short entries;
398 if (eh->eh_entries == 0)
401 entries = le16_to_cpu(eh->eh_entries);
405 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
406 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
407 ext4_fsblk_t pblock = 0;
408 ext4_lblk_t lblock = 0;
409 ext4_lblk_t prev = 0;
412 if (!ext4_valid_extent(inode, ext))
415 /* Check for overlapping extents */
416 lblock = le32_to_cpu(ext->ee_block);
417 len = ext4_ext_get_actual_len(ext);
418 if ((lblock <= prev) && prev) {
419 pblock = ext4_ext_pblock(ext);
420 es->s_last_error_block = cpu_to_le64(pblock);
425 prev = lblock + len - 1;
428 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
430 if (!ext4_valid_extent_idx(inode, ext_idx))
439 static int __ext4_ext_check(const char *function, unsigned int line,
440 struct inode *inode, struct ext4_extent_header *eh,
441 int depth, ext4_fsblk_t pblk)
443 const char *error_msg;
446 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
447 error_msg = "invalid magic";
450 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
451 error_msg = "unexpected eh_depth";
454 if (unlikely(eh->eh_max == 0)) {
455 error_msg = "invalid eh_max";
458 max = ext4_ext_max_entries(inode, depth);
459 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
460 error_msg = "too large eh_max";
463 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
464 error_msg = "invalid eh_entries";
467 if (!ext4_valid_extent_entries(inode, eh, depth)) {
468 error_msg = "invalid extent entries";
471 /* Verify checksum on non-root extent tree nodes */
472 if (ext_depth(inode) != depth &&
473 !ext4_extent_block_csum_verify(inode, eh)) {
474 error_msg = "extent tree corrupted";
480 ext4_error_inode(inode, function, line, 0,
481 "pblk %llu bad header/extent: %s - magic %x, "
482 "entries %u, max %u(%u), depth %u(%u)",
483 (unsigned long long) pblk, error_msg,
484 le16_to_cpu(eh->eh_magic),
485 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
486 max, le16_to_cpu(eh->eh_depth), depth);
490 #define ext4_ext_check(inode, eh, depth, pblk) \
491 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
493 int ext4_ext_check_inode(struct inode *inode)
495 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
498 static struct buffer_head *
499 __read_extent_tree_block(const char *function, unsigned int line,
500 struct inode *inode, ext4_fsblk_t pblk, int depth,
503 struct buffer_head *bh;
506 bh = sb_getblk(inode->i_sb, pblk);
508 return ERR_PTR(-ENOMEM);
510 if (!bh_uptodate_or_lock(bh)) {
511 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
512 err = bh_submit_read(bh);
516 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
518 err = __ext4_ext_check(function, line, inode,
519 ext_block_hdr(bh), depth, pblk);
522 set_buffer_verified(bh);
524 * If this is a leaf block, cache all of its entries
526 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
527 struct ext4_extent_header *eh = ext_block_hdr(bh);
528 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
529 ext4_lblk_t prev = 0;
532 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
533 unsigned int status = EXTENT_STATUS_WRITTEN;
534 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
535 int len = ext4_ext_get_actual_len(ex);
537 if (prev && (prev != lblk))
538 ext4_es_cache_extent(inode, prev,
542 if (ext4_ext_is_unwritten(ex))
543 status = EXTENT_STATUS_UNWRITTEN;
544 ext4_es_cache_extent(inode, lblk, len,
545 ext4_ext_pblock(ex), status);
556 #define read_extent_tree_block(inode, pblk, depth, flags) \
557 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
561 * This function is called to cache a file's extent information in the
564 int ext4_ext_precache(struct inode *inode)
566 struct ext4_inode_info *ei = EXT4_I(inode);
567 struct ext4_ext_path *path = NULL;
568 struct buffer_head *bh;
569 int i = 0, depth, ret = 0;
571 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
572 return 0; /* not an extent-mapped inode */
574 down_read(&ei->i_data_sem);
575 depth = ext_depth(inode);
577 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
580 up_read(&ei->i_data_sem);
584 /* Don't cache anything if there are no external extent blocks */
587 path[0].p_hdr = ext_inode_hdr(inode);
588 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
591 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
594 * If this is a leaf block or we've reached the end of
595 * the index block, go up
598 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
599 brelse(path[i].p_bh);
604 bh = read_extent_tree_block(inode,
605 ext4_idx_pblock(path[i].p_idx++),
607 EXT4_EX_FORCE_CACHE);
614 path[i].p_hdr = ext_block_hdr(bh);
615 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
617 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
619 up_read(&ei->i_data_sem);
620 ext4_ext_drop_refs(path);
626 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
628 int k, l = path->p_depth;
631 for (k = 0; k <= l; k++, path++) {
633 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
634 ext4_idx_pblock(path->p_idx));
635 } else if (path->p_ext) {
636 ext_debug(" %d:[%d]%d:%llu ",
637 le32_to_cpu(path->p_ext->ee_block),
638 ext4_ext_is_unwritten(path->p_ext),
639 ext4_ext_get_actual_len(path->p_ext),
640 ext4_ext_pblock(path->p_ext));
647 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
649 int depth = ext_depth(inode);
650 struct ext4_extent_header *eh;
651 struct ext4_extent *ex;
657 eh = path[depth].p_hdr;
658 ex = EXT_FIRST_EXTENT(eh);
660 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
662 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
663 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
664 ext4_ext_is_unwritten(ex),
665 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
670 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
671 ext4_fsblk_t newblock, int level)
673 int depth = ext_depth(inode);
674 struct ext4_extent *ex;
676 if (depth != level) {
677 struct ext4_extent_idx *idx;
678 idx = path[level].p_idx;
679 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
680 ext_debug("%d: move %d:%llu in new index %llu\n", level,
681 le32_to_cpu(idx->ei_block),
682 ext4_idx_pblock(idx),
690 ex = path[depth].p_ext;
691 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
692 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
693 le32_to_cpu(ex->ee_block),
695 ext4_ext_is_unwritten(ex),
696 ext4_ext_get_actual_len(ex),
703 #define ext4_ext_show_path(inode, path)
704 #define ext4_ext_show_leaf(inode, path)
705 #define ext4_ext_show_move(inode, path, newblock, level)
708 void ext4_ext_drop_refs(struct ext4_ext_path *path)
714 depth = path->p_depth;
715 for (i = 0; i <= depth; i++, path++)
723 * ext4_ext_binsearch_idx:
724 * binary search for the closest index of the given block
725 * the header must be checked before calling this
728 ext4_ext_binsearch_idx(struct inode *inode,
729 struct ext4_ext_path *path, ext4_lblk_t block)
731 struct ext4_extent_header *eh = path->p_hdr;
732 struct ext4_extent_idx *r, *l, *m;
735 ext_debug("binsearch for %u(idx): ", block);
737 l = EXT_FIRST_INDEX(eh) + 1;
738 r = EXT_LAST_INDEX(eh);
741 if (block < le32_to_cpu(m->ei_block))
745 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
746 m, le32_to_cpu(m->ei_block),
747 r, le32_to_cpu(r->ei_block));
751 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
752 ext4_idx_pblock(path->p_idx));
754 #ifdef CHECK_BINSEARCH
756 struct ext4_extent_idx *chix, *ix;
759 chix = ix = EXT_FIRST_INDEX(eh);
760 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
762 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
763 printk(KERN_DEBUG "k=%d, ix=0x%p, "
765 ix, EXT_FIRST_INDEX(eh));
766 printk(KERN_DEBUG "%u <= %u\n",
767 le32_to_cpu(ix->ei_block),
768 le32_to_cpu(ix[-1].ei_block));
770 BUG_ON(k && le32_to_cpu(ix->ei_block)
771 <= le32_to_cpu(ix[-1].ei_block));
772 if (block < le32_to_cpu(ix->ei_block))
776 BUG_ON(chix != path->p_idx);
783 * ext4_ext_binsearch:
784 * binary search for closest extent of the given block
785 * the header must be checked before calling this
788 ext4_ext_binsearch(struct inode *inode,
789 struct ext4_ext_path *path, ext4_lblk_t block)
791 struct ext4_extent_header *eh = path->p_hdr;
792 struct ext4_extent *r, *l, *m;
794 if (eh->eh_entries == 0) {
796 * this leaf is empty:
797 * we get such a leaf in split/add case
802 ext_debug("binsearch for %u: ", block);
804 l = EXT_FIRST_EXTENT(eh) + 1;
805 r = EXT_LAST_EXTENT(eh);
809 if (block < le32_to_cpu(m->ee_block))
813 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
814 m, le32_to_cpu(m->ee_block),
815 r, le32_to_cpu(r->ee_block));
819 ext_debug(" -> %d:%llu:[%d]%d ",
820 le32_to_cpu(path->p_ext->ee_block),
821 ext4_ext_pblock(path->p_ext),
822 ext4_ext_is_unwritten(path->p_ext),
823 ext4_ext_get_actual_len(path->p_ext));
825 #ifdef CHECK_BINSEARCH
827 struct ext4_extent *chex, *ex;
830 chex = ex = EXT_FIRST_EXTENT(eh);
831 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
832 BUG_ON(k && le32_to_cpu(ex->ee_block)
833 <= le32_to_cpu(ex[-1].ee_block));
834 if (block < le32_to_cpu(ex->ee_block))
838 BUG_ON(chex != path->p_ext);
844 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
846 struct ext4_extent_header *eh;
848 eh = ext_inode_hdr(inode);
851 eh->eh_magic = EXT4_EXT_MAGIC;
852 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
853 ext4_mark_inode_dirty(handle, inode);
857 struct ext4_ext_path *
858 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
859 struct ext4_ext_path **orig_path, int flags)
861 struct ext4_extent_header *eh;
862 struct buffer_head *bh;
863 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
864 short int depth, i, ppos = 0;
867 eh = ext_inode_hdr(inode);
868 depth = ext_depth(inode);
871 ext4_ext_drop_refs(path);
872 if (depth > path[0].p_maxdepth) {
874 *orig_path = path = NULL;
878 /* account possible depth increase */
879 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
882 return ERR_PTR(-ENOMEM);
883 path[0].p_maxdepth = depth + 1;
889 /* walk through the tree */
891 ext_debug("depth %d: num %d, max %d\n",
892 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
894 ext4_ext_binsearch_idx(inode, path + ppos, block);
895 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
896 path[ppos].p_depth = i;
897 path[ppos].p_ext = NULL;
899 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
901 if (unlikely(IS_ERR(bh))) {
906 eh = ext_block_hdr(bh);
908 if (unlikely(ppos > depth)) {
910 EXT4_ERROR_INODE(inode,
911 "ppos %d > depth %d", ppos, depth);
915 path[ppos].p_bh = bh;
916 path[ppos].p_hdr = eh;
919 path[ppos].p_depth = i;
920 path[ppos].p_ext = NULL;
921 path[ppos].p_idx = NULL;
924 ext4_ext_binsearch(inode, path + ppos, block);
925 /* if not an empty leaf */
926 if (path[ppos].p_ext)
927 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
929 ext4_ext_show_path(inode, path);
934 ext4_ext_drop_refs(path);
942 * ext4_ext_insert_index:
943 * insert new index [@logical;@ptr] into the block at @curp;
944 * check where to insert: before @curp or after @curp
946 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
947 struct ext4_ext_path *curp,
948 int logical, ext4_fsblk_t ptr)
950 struct ext4_extent_idx *ix;
953 err = ext4_ext_get_access(handle, inode, curp);
957 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
958 EXT4_ERROR_INODE(inode,
959 "logical %d == ei_block %d!",
960 logical, le32_to_cpu(curp->p_idx->ei_block));
964 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
965 >= le16_to_cpu(curp->p_hdr->eh_max))) {
966 EXT4_ERROR_INODE(inode,
967 "eh_entries %d >= eh_max %d!",
968 le16_to_cpu(curp->p_hdr->eh_entries),
969 le16_to_cpu(curp->p_hdr->eh_max));
973 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
975 ext_debug("insert new index %d after: %llu\n", logical, ptr);
976 ix = curp->p_idx + 1;
979 ext_debug("insert new index %d before: %llu\n", logical, ptr);
983 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
986 ext_debug("insert new index %d: "
987 "move %d indices from 0x%p to 0x%p\n",
988 logical, len, ix, ix + 1);
989 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
992 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
993 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
997 ix->ei_block = cpu_to_le32(logical);
998 ext4_idx_store_pblock(ix, ptr);
999 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1001 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1002 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1006 err = ext4_ext_dirty(handle, inode, curp);
1007 ext4_std_error(inode->i_sb, err);
1014 * inserts new subtree into the path, using free index entry
1016 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1017 * - makes decision where to split
1018 * - moves remaining extents and index entries (right to the split point)
1019 * into the newly allocated blocks
1020 * - initializes subtree
1022 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1024 struct ext4_ext_path *path,
1025 struct ext4_extent *newext, int at)
1027 struct buffer_head *bh = NULL;
1028 int depth = ext_depth(inode);
1029 struct ext4_extent_header *neh;
1030 struct ext4_extent_idx *fidx;
1031 int i = at, k, m, a;
1032 ext4_fsblk_t newblock, oldblock;
1034 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1037 /* make decision: where to split? */
1038 /* FIXME: now decision is simplest: at current extent */
1040 /* if current leaf will be split, then we should use
1041 * border from split point */
1042 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1043 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1046 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1047 border = path[depth].p_ext[1].ee_block;
1048 ext_debug("leaf will be split."
1049 " next leaf starts at %d\n",
1050 le32_to_cpu(border));
1052 border = newext->ee_block;
1053 ext_debug("leaf will be added."
1054 " next leaf starts at %d\n",
1055 le32_to_cpu(border));
1059 * If error occurs, then we break processing
1060 * and mark filesystem read-only. index won't
1061 * be inserted and tree will be in consistent
1062 * state. Next mount will repair buffers too.
1066 * Get array to track all allocated blocks.
1067 * We need this to handle errors and free blocks
1070 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1074 /* allocate all needed blocks */
1075 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1076 for (a = 0; a < depth - at; a++) {
1077 newblock = ext4_ext_new_meta_block(handle, inode, path,
1078 newext, &err, flags);
1081 ablocks[a] = newblock;
1084 /* initialize new leaf */
1085 newblock = ablocks[--a];
1086 if (unlikely(newblock == 0)) {
1087 EXT4_ERROR_INODE(inode, "newblock == 0!");
1091 bh = sb_getblk(inode->i_sb, newblock);
1092 if (unlikely(!bh)) {
1098 err = ext4_journal_get_create_access(handle, bh);
1102 neh = ext_block_hdr(bh);
1103 neh->eh_entries = 0;
1104 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1105 neh->eh_magic = EXT4_EXT_MAGIC;
1108 /* move remainder of path[depth] to the new leaf */
1109 if (unlikely(path[depth].p_hdr->eh_entries !=
1110 path[depth].p_hdr->eh_max)) {
1111 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1112 path[depth].p_hdr->eh_entries,
1113 path[depth].p_hdr->eh_max);
1117 /* start copy from next extent */
1118 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1119 ext4_ext_show_move(inode, path, newblock, depth);
1121 struct ext4_extent *ex;
1122 ex = EXT_FIRST_EXTENT(neh);
1123 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1124 le16_add_cpu(&neh->eh_entries, m);
1127 ext4_extent_block_csum_set(inode, neh);
1128 set_buffer_uptodate(bh);
1131 err = ext4_handle_dirty_metadata(handle, inode, bh);
1137 /* correct old leaf */
1139 err = ext4_ext_get_access(handle, inode, path + depth);
1142 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1143 err = ext4_ext_dirty(handle, inode, path + depth);
1149 /* create intermediate indexes */
1151 if (unlikely(k < 0)) {
1152 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1157 ext_debug("create %d intermediate indices\n", k);
1158 /* insert new index into current index block */
1159 /* current depth stored in i var */
1162 oldblock = newblock;
1163 newblock = ablocks[--a];
1164 bh = sb_getblk(inode->i_sb, newblock);
1165 if (unlikely(!bh)) {
1171 err = ext4_journal_get_create_access(handle, bh);
1175 neh = ext_block_hdr(bh);
1176 neh->eh_entries = cpu_to_le16(1);
1177 neh->eh_magic = EXT4_EXT_MAGIC;
1178 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1179 neh->eh_depth = cpu_to_le16(depth - i);
1180 fidx = EXT_FIRST_INDEX(neh);
1181 fidx->ei_block = border;
1182 ext4_idx_store_pblock(fidx, oldblock);
1184 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1185 i, newblock, le32_to_cpu(border), oldblock);
1187 /* move remainder of path[i] to the new index block */
1188 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1189 EXT_LAST_INDEX(path[i].p_hdr))) {
1190 EXT4_ERROR_INODE(inode,
1191 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1192 le32_to_cpu(path[i].p_ext->ee_block));
1196 /* start copy indexes */
1197 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1198 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1199 EXT_MAX_INDEX(path[i].p_hdr));
1200 ext4_ext_show_move(inode, path, newblock, i);
1202 memmove(++fidx, path[i].p_idx,
1203 sizeof(struct ext4_extent_idx) * m);
1204 le16_add_cpu(&neh->eh_entries, m);
1206 ext4_extent_block_csum_set(inode, neh);
1207 set_buffer_uptodate(bh);
1210 err = ext4_handle_dirty_metadata(handle, inode, bh);
1216 /* correct old index */
1218 err = ext4_ext_get_access(handle, inode, path + i);
1221 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1222 err = ext4_ext_dirty(handle, inode, path + i);
1230 /* insert new index */
1231 err = ext4_ext_insert_index(handle, inode, path + at,
1232 le32_to_cpu(border), newblock);
1236 if (buffer_locked(bh))
1242 /* free all allocated blocks in error case */
1243 for (i = 0; i < depth; i++) {
1246 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1247 EXT4_FREE_BLOCKS_METADATA);
1256 * ext4_ext_grow_indepth:
1257 * implements tree growing procedure:
1258 * - allocates new block
1259 * - moves top-level data (index block or leaf) into the new block
1260 * - initializes new top-level, creating index that points to the
1261 * just created block
1263 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1266 struct ext4_extent_header *neh;
1267 struct buffer_head *bh;
1268 ext4_fsblk_t newblock, goal = 0;
1269 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1272 /* Try to prepend new index to old one */
1273 if (ext_depth(inode))
1274 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1275 if (goal > le32_to_cpu(es->s_first_data_block)) {
1276 flags |= EXT4_MB_HINT_TRY_GOAL;
1279 goal = ext4_inode_to_goal_block(inode);
1280 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1285 bh = sb_getblk(inode->i_sb, newblock);
1290 err = ext4_journal_get_create_access(handle, bh);
1296 /* move top-level index/leaf into new block */
1297 memmove(bh->b_data, EXT4_I(inode)->i_data,
1298 sizeof(EXT4_I(inode)->i_data));
1300 /* set size of new block */
1301 neh = ext_block_hdr(bh);
1302 /* old root could have indexes or leaves
1303 * so calculate e_max right way */
1304 if (ext_depth(inode))
1305 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1307 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1308 neh->eh_magic = EXT4_EXT_MAGIC;
1309 ext4_extent_block_csum_set(inode, neh);
1310 set_buffer_uptodate(bh);
1313 err = ext4_handle_dirty_metadata(handle, inode, bh);
1317 /* Update top-level index: num,max,pointer */
1318 neh = ext_inode_hdr(inode);
1319 neh->eh_entries = cpu_to_le16(1);
1320 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1321 if (neh->eh_depth == 0) {
1322 /* Root extent block becomes index block */
1323 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1324 EXT_FIRST_INDEX(neh)->ei_block =
1325 EXT_FIRST_EXTENT(neh)->ee_block;
1327 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1328 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1329 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1330 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1332 le16_add_cpu(&neh->eh_depth, 1);
1333 ext4_mark_inode_dirty(handle, inode);
1341 * ext4_ext_create_new_leaf:
1342 * finds empty index and adds new leaf.
1343 * if no free index is found, then it requests in-depth growing.
1345 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1346 unsigned int mb_flags,
1347 unsigned int gb_flags,
1348 struct ext4_ext_path **ppath,
1349 struct ext4_extent *newext)
1351 struct ext4_ext_path *path = *ppath;
1352 struct ext4_ext_path *curp;
1353 int depth, i, err = 0;
1356 i = depth = ext_depth(inode);
1358 /* walk up to the tree and look for free index entry */
1359 curp = path + depth;
1360 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1365 /* we use already allocated block for index block,
1366 * so subsequent data blocks should be contiguous */
1367 if (EXT_HAS_FREE_INDEX(curp)) {
1368 /* if we found index with free entry, then use that
1369 * entry: create all needed subtree and add new leaf */
1370 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1375 path = ext4_find_extent(inode,
1376 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1379 err = PTR_ERR(path);
1381 /* tree is full, time to grow in depth */
1382 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1387 path = ext4_find_extent(inode,
1388 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1391 err = PTR_ERR(path);
1396 * only first (depth 0 -> 1) produces free space;
1397 * in all other cases we have to split the grown tree
1399 depth = ext_depth(inode);
1400 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1401 /* now we need to split */
1411 * search the closest allocated block to the left for *logical
1412 * and returns it at @logical + it's physical address at @phys
1413 * if *logical is the smallest allocated block, the function
1414 * returns 0 at @phys
1415 * return value contains 0 (success) or error code
1417 static int ext4_ext_search_left(struct inode *inode,
1418 struct ext4_ext_path *path,
1419 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1421 struct ext4_extent_idx *ix;
1422 struct ext4_extent *ex;
1425 if (unlikely(path == NULL)) {
1426 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1429 depth = path->p_depth;
1432 if (depth == 0 && path->p_ext == NULL)
1435 /* usually extent in the path covers blocks smaller
1436 * then *logical, but it can be that extent is the
1437 * first one in the file */
1439 ex = path[depth].p_ext;
1440 ee_len = ext4_ext_get_actual_len(ex);
1441 if (*logical < le32_to_cpu(ex->ee_block)) {
1442 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1443 EXT4_ERROR_INODE(inode,
1444 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1445 *logical, le32_to_cpu(ex->ee_block));
1448 while (--depth >= 0) {
1449 ix = path[depth].p_idx;
1450 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1451 EXT4_ERROR_INODE(inode,
1452 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1453 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1454 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1455 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1463 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1464 EXT4_ERROR_INODE(inode,
1465 "logical %d < ee_block %d + ee_len %d!",
1466 *logical, le32_to_cpu(ex->ee_block), ee_len);
1470 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1471 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1476 * search the closest allocated block to the right for *logical
1477 * and returns it at @logical + it's physical address at @phys
1478 * if *logical is the largest allocated block, the function
1479 * returns 0 at @phys
1480 * return value contains 0 (success) or error code
1482 static int ext4_ext_search_right(struct inode *inode,
1483 struct ext4_ext_path *path,
1484 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1485 struct ext4_extent **ret_ex)
1487 struct buffer_head *bh = NULL;
1488 struct ext4_extent_header *eh;
1489 struct ext4_extent_idx *ix;
1490 struct ext4_extent *ex;
1492 int depth; /* Note, NOT eh_depth; depth from top of tree */
1495 if (unlikely(path == NULL)) {
1496 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1499 depth = path->p_depth;
1502 if (depth == 0 && path->p_ext == NULL)
1505 /* usually extent in the path covers blocks smaller
1506 * then *logical, but it can be that extent is the
1507 * first one in the file */
1509 ex = path[depth].p_ext;
1510 ee_len = ext4_ext_get_actual_len(ex);
1511 if (*logical < le32_to_cpu(ex->ee_block)) {
1512 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1513 EXT4_ERROR_INODE(inode,
1514 "first_extent(path[%d].p_hdr) != ex",
1518 while (--depth >= 0) {
1519 ix = path[depth].p_idx;
1520 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1521 EXT4_ERROR_INODE(inode,
1522 "ix != EXT_FIRST_INDEX *logical %d!",
1530 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1531 EXT4_ERROR_INODE(inode,
1532 "logical %d < ee_block %d + ee_len %d!",
1533 *logical, le32_to_cpu(ex->ee_block), ee_len);
1537 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1538 /* next allocated block in this leaf */
1543 /* go up and search for index to the right */
1544 while (--depth >= 0) {
1545 ix = path[depth].p_idx;
1546 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1550 /* we've gone up to the root and found no index to the right */
1554 /* we've found index to the right, let's
1555 * follow it and find the closest allocated
1556 * block to the right */
1558 block = ext4_idx_pblock(ix);
1559 while (++depth < path->p_depth) {
1560 /* subtract from p_depth to get proper eh_depth */
1561 bh = read_extent_tree_block(inode, block,
1562 path->p_depth - depth, 0);
1565 eh = ext_block_hdr(bh);
1566 ix = EXT_FIRST_INDEX(eh);
1567 block = ext4_idx_pblock(ix);
1571 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1574 eh = ext_block_hdr(bh);
1575 ex = EXT_FIRST_EXTENT(eh);
1577 *logical = le32_to_cpu(ex->ee_block);
1578 *phys = ext4_ext_pblock(ex);
1586 * ext4_ext_next_allocated_block:
1587 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1588 * NOTE: it considers block number from index entry as
1589 * allocated block. Thus, index entries have to be consistent
1593 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1597 BUG_ON(path == NULL);
1598 depth = path->p_depth;
1600 if (depth == 0 && path->p_ext == NULL)
1601 return EXT_MAX_BLOCKS;
1603 while (depth >= 0) {
1604 if (depth == path->p_depth) {
1606 if (path[depth].p_ext &&
1607 path[depth].p_ext !=
1608 EXT_LAST_EXTENT(path[depth].p_hdr))
1609 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1612 if (path[depth].p_idx !=
1613 EXT_LAST_INDEX(path[depth].p_hdr))
1614 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1619 return EXT_MAX_BLOCKS;
1623 * ext4_ext_next_leaf_block:
1624 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1626 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1630 BUG_ON(path == NULL);
1631 depth = path->p_depth;
1633 /* zero-tree has no leaf blocks at all */
1635 return EXT_MAX_BLOCKS;
1637 /* go to index block */
1640 while (depth >= 0) {
1641 if (path[depth].p_idx !=
1642 EXT_LAST_INDEX(path[depth].p_hdr))
1643 return (ext4_lblk_t)
1644 le32_to_cpu(path[depth].p_idx[1].ei_block);
1648 return EXT_MAX_BLOCKS;
1652 * ext4_ext_correct_indexes:
1653 * if leaf gets modified and modified extent is first in the leaf,
1654 * then we have to correct all indexes above.
1655 * TODO: do we need to correct tree in all cases?
1657 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1658 struct ext4_ext_path *path)
1660 struct ext4_extent_header *eh;
1661 int depth = ext_depth(inode);
1662 struct ext4_extent *ex;
1666 eh = path[depth].p_hdr;
1667 ex = path[depth].p_ext;
1669 if (unlikely(ex == NULL || eh == NULL)) {
1670 EXT4_ERROR_INODE(inode,
1671 "ex %p == NULL or eh %p == NULL", ex, eh);
1676 /* there is no tree at all */
1680 if (ex != EXT_FIRST_EXTENT(eh)) {
1681 /* we correct tree if first leaf got modified only */
1686 * TODO: we need correction if border is smaller than current one
1689 border = path[depth].p_ext->ee_block;
1690 err = ext4_ext_get_access(handle, inode, path + k);
1693 path[k].p_idx->ei_block = border;
1694 err = ext4_ext_dirty(handle, inode, path + k);
1699 /* change all left-side indexes */
1700 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1702 err = ext4_ext_get_access(handle, inode, path + k);
1705 path[k].p_idx->ei_block = border;
1706 err = ext4_ext_dirty(handle, inode, path + k);
1715 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1716 struct ext4_extent *ex2)
1718 unsigned short ext1_ee_len, ext2_ee_len;
1721 * Make sure that both extents are initialized. We don't merge
1722 * unwritten extents so that we can be sure that end_io code has
1723 * the extent that was written properly split out and conversion to
1724 * initialized is trivial.
1726 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1729 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1730 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1732 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1733 le32_to_cpu(ex2->ee_block))
1737 * To allow future support for preallocated extents to be added
1738 * as an RO_COMPAT feature, refuse to merge to extents if
1739 * this can result in the top bit of ee_len being set.
1741 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1743 if (ext4_ext_is_unwritten(ex1) &&
1744 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1745 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1746 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1748 #ifdef AGGRESSIVE_TEST
1749 if (ext1_ee_len >= 4)
1753 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1759 * This function tries to merge the "ex" extent to the next extent in the tree.
1760 * It always tries to merge towards right. If you want to merge towards
1761 * left, pass "ex - 1" as argument instead of "ex".
1762 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1763 * 1 if they got merged.
1765 static int ext4_ext_try_to_merge_right(struct inode *inode,
1766 struct ext4_ext_path *path,
1767 struct ext4_extent *ex)
1769 struct ext4_extent_header *eh;
1770 unsigned int depth, len;
1771 int merge_done = 0, unwritten;
1773 depth = ext_depth(inode);
1774 BUG_ON(path[depth].p_hdr == NULL);
1775 eh = path[depth].p_hdr;
1777 while (ex < EXT_LAST_EXTENT(eh)) {
1778 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1780 /* merge with next extent! */
1781 unwritten = ext4_ext_is_unwritten(ex);
1782 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1783 + ext4_ext_get_actual_len(ex + 1));
1785 ext4_ext_mark_unwritten(ex);
1787 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1788 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1789 * sizeof(struct ext4_extent);
1790 memmove(ex + 1, ex + 2, len);
1792 le16_add_cpu(&eh->eh_entries, -1);
1794 WARN_ON(eh->eh_entries == 0);
1795 if (!eh->eh_entries)
1796 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1803 * This function does a very simple check to see if we can collapse
1804 * an extent tree with a single extent tree leaf block into the inode.
1806 static void ext4_ext_try_to_merge_up(handle_t *handle,
1807 struct inode *inode,
1808 struct ext4_ext_path *path)
1811 unsigned max_root = ext4_ext_space_root(inode, 0);
1814 if ((path[0].p_depth != 1) ||
1815 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1816 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1820 * We need to modify the block allocation bitmap and the block
1821 * group descriptor to release the extent tree block. If we
1822 * can't get the journal credits, give up.
1824 if (ext4_journal_extend(handle, 2))
1828 * Copy the extent data up to the inode
1830 blk = ext4_idx_pblock(path[0].p_idx);
1831 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1832 sizeof(struct ext4_extent_idx);
1833 s += sizeof(struct ext4_extent_header);
1835 path[1].p_maxdepth = path[0].p_maxdepth;
1836 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1837 path[0].p_depth = 0;
1838 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1839 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1840 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1842 brelse(path[1].p_bh);
1843 ext4_free_blocks(handle, inode, NULL, blk, 1,
1844 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1848 * This function tries to merge the @ex extent to neighbours in the tree.
1849 * return 1 if merge left else 0.
1851 static void ext4_ext_try_to_merge(handle_t *handle,
1852 struct inode *inode,
1853 struct ext4_ext_path *path,
1854 struct ext4_extent *ex) {
1855 struct ext4_extent_header *eh;
1859 depth = ext_depth(inode);
1860 BUG_ON(path[depth].p_hdr == NULL);
1861 eh = path[depth].p_hdr;
1863 if (ex > EXT_FIRST_EXTENT(eh))
1864 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1867 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1869 ext4_ext_try_to_merge_up(handle, inode, path);
1873 * check if a portion of the "newext" extent overlaps with an
1876 * If there is an overlap discovered, it updates the length of the newext
1877 * such that there will be no overlap, and then returns 1.
1878 * If there is no overlap found, it returns 0.
1880 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1881 struct inode *inode,
1882 struct ext4_extent *newext,
1883 struct ext4_ext_path *path)
1886 unsigned int depth, len1;
1887 unsigned int ret = 0;
1889 b1 = le32_to_cpu(newext->ee_block);
1890 len1 = ext4_ext_get_actual_len(newext);
1891 depth = ext_depth(inode);
1892 if (!path[depth].p_ext)
1894 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1897 * get the next allocated block if the extent in the path
1898 * is before the requested block(s)
1901 b2 = ext4_ext_next_allocated_block(path);
1902 if (b2 == EXT_MAX_BLOCKS)
1904 b2 = EXT4_LBLK_CMASK(sbi, b2);
1907 /* check for wrap through zero on extent logical start block*/
1908 if (b1 + len1 < b1) {
1909 len1 = EXT_MAX_BLOCKS - b1;
1910 newext->ee_len = cpu_to_le16(len1);
1914 /* check for overlap */
1915 if (b1 + len1 > b2) {
1916 newext->ee_len = cpu_to_le16(b2 - b1);
1924 * ext4_ext_insert_extent:
1925 * tries to merge requsted extent into the existing extent or
1926 * inserts requested extent as new one into the tree,
1927 * creating new leaf in the no-space case.
1929 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1930 struct ext4_ext_path **ppath,
1931 struct ext4_extent *newext, int gb_flags)
1933 struct ext4_ext_path *path = *ppath;
1934 struct ext4_extent_header *eh;
1935 struct ext4_extent *ex, *fex;
1936 struct ext4_extent *nearex; /* nearest extent */
1937 struct ext4_ext_path *npath = NULL;
1938 int depth, len, err;
1940 int mb_flags = 0, unwritten;
1942 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1943 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1944 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1945 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1948 depth = ext_depth(inode);
1949 ex = path[depth].p_ext;
1950 eh = path[depth].p_hdr;
1951 if (unlikely(path[depth].p_hdr == NULL)) {
1952 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1956 /* try to insert block into found extent and return */
1957 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1960 * Try to see whether we should rather test the extent on
1961 * right from ex, or from the left of ex. This is because
1962 * ext4_find_extent() can return either extent on the
1963 * left, or on the right from the searched position. This
1964 * will make merging more effective.
1966 if (ex < EXT_LAST_EXTENT(eh) &&
1967 (le32_to_cpu(ex->ee_block) +
1968 ext4_ext_get_actual_len(ex) <
1969 le32_to_cpu(newext->ee_block))) {
1972 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1973 (le32_to_cpu(newext->ee_block) +
1974 ext4_ext_get_actual_len(newext) <
1975 le32_to_cpu(ex->ee_block)))
1978 /* Try to append newex to the ex */
1979 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1980 ext_debug("append [%d]%d block to %u:[%d]%d"
1982 ext4_ext_is_unwritten(newext),
1983 ext4_ext_get_actual_len(newext),
1984 le32_to_cpu(ex->ee_block),
1985 ext4_ext_is_unwritten(ex),
1986 ext4_ext_get_actual_len(ex),
1987 ext4_ext_pblock(ex));
1988 err = ext4_ext_get_access(handle, inode,
1992 unwritten = ext4_ext_is_unwritten(ex);
1993 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1994 + ext4_ext_get_actual_len(newext));
1996 ext4_ext_mark_unwritten(ex);
1997 eh = path[depth].p_hdr;
2003 /* Try to prepend newex to the ex */
2004 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2005 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2007 le32_to_cpu(newext->ee_block),
2008 ext4_ext_is_unwritten(newext),
2009 ext4_ext_get_actual_len(newext),
2010 le32_to_cpu(ex->ee_block),
2011 ext4_ext_is_unwritten(ex),
2012 ext4_ext_get_actual_len(ex),
2013 ext4_ext_pblock(ex));
2014 err = ext4_ext_get_access(handle, inode,
2019 unwritten = ext4_ext_is_unwritten(ex);
2020 ex->ee_block = newext->ee_block;
2021 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2022 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2023 + ext4_ext_get_actual_len(newext));
2025 ext4_ext_mark_unwritten(ex);
2026 eh = path[depth].p_hdr;
2032 depth = ext_depth(inode);
2033 eh = path[depth].p_hdr;
2034 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2037 /* probably next leaf has space for us? */
2038 fex = EXT_LAST_EXTENT(eh);
2039 next = EXT_MAX_BLOCKS;
2040 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2041 next = ext4_ext_next_leaf_block(path);
2042 if (next != EXT_MAX_BLOCKS) {
2043 ext_debug("next leaf block - %u\n", next);
2044 BUG_ON(npath != NULL);
2045 npath = ext4_find_extent(inode, next, NULL, 0);
2047 return PTR_ERR(npath);
2048 BUG_ON(npath->p_depth != path->p_depth);
2049 eh = npath[depth].p_hdr;
2050 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2051 ext_debug("next leaf isn't full(%d)\n",
2052 le16_to_cpu(eh->eh_entries));
2056 ext_debug("next leaf has no free space(%d,%d)\n",
2057 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2061 * There is no free space in the found leaf.
2062 * We're gonna add a new leaf in the tree.
2064 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2065 mb_flags |= EXT4_MB_USE_RESERVED;
2066 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2070 depth = ext_depth(inode);
2071 eh = path[depth].p_hdr;
2074 nearex = path[depth].p_ext;
2076 err = ext4_ext_get_access(handle, inode, path + depth);
2081 /* there is no extent in this leaf, create first one */
2082 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2083 le32_to_cpu(newext->ee_block),
2084 ext4_ext_pblock(newext),
2085 ext4_ext_is_unwritten(newext),
2086 ext4_ext_get_actual_len(newext));
2087 nearex = EXT_FIRST_EXTENT(eh);
2089 if (le32_to_cpu(newext->ee_block)
2090 > le32_to_cpu(nearex->ee_block)) {
2092 ext_debug("insert %u:%llu:[%d]%d before: "
2094 le32_to_cpu(newext->ee_block),
2095 ext4_ext_pblock(newext),
2096 ext4_ext_is_unwritten(newext),
2097 ext4_ext_get_actual_len(newext),
2102 BUG_ON(newext->ee_block == nearex->ee_block);
2103 ext_debug("insert %u:%llu:[%d]%d after: "
2105 le32_to_cpu(newext->ee_block),
2106 ext4_ext_pblock(newext),
2107 ext4_ext_is_unwritten(newext),
2108 ext4_ext_get_actual_len(newext),
2111 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2113 ext_debug("insert %u:%llu:[%d]%d: "
2114 "move %d extents from 0x%p to 0x%p\n",
2115 le32_to_cpu(newext->ee_block),
2116 ext4_ext_pblock(newext),
2117 ext4_ext_is_unwritten(newext),
2118 ext4_ext_get_actual_len(newext),
2119 len, nearex, nearex + 1);
2120 memmove(nearex + 1, nearex,
2121 len * sizeof(struct ext4_extent));
2125 le16_add_cpu(&eh->eh_entries, 1);
2126 path[depth].p_ext = nearex;
2127 nearex->ee_block = newext->ee_block;
2128 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2129 nearex->ee_len = newext->ee_len;
2132 /* try to merge extents */
2133 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2134 ext4_ext_try_to_merge(handle, inode, path, nearex);
2137 /* time to correct all indexes above */
2138 err = ext4_ext_correct_indexes(handle, inode, path);
2142 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2145 ext4_ext_drop_refs(npath);
2150 static int ext4_fill_fiemap_extents(struct inode *inode,
2151 ext4_lblk_t block, ext4_lblk_t num,
2152 struct fiemap_extent_info *fieinfo)
2154 struct ext4_ext_path *path = NULL;
2155 struct ext4_extent *ex;
2156 struct extent_status es;
2157 ext4_lblk_t next, next_del, start = 0, end = 0;
2158 ext4_lblk_t last = block + num;
2159 int exists, depth = 0, err = 0;
2160 unsigned int flags = 0;
2161 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2163 while (block < last && block != EXT_MAX_BLOCKS) {
2165 /* find extent for this block */
2166 down_read(&EXT4_I(inode)->i_data_sem);
2168 path = ext4_find_extent(inode, block, &path, 0);
2170 up_read(&EXT4_I(inode)->i_data_sem);
2171 err = PTR_ERR(path);
2176 depth = ext_depth(inode);
2177 if (unlikely(path[depth].p_hdr == NULL)) {
2178 up_read(&EXT4_I(inode)->i_data_sem);
2179 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2183 ex = path[depth].p_ext;
2184 next = ext4_ext_next_allocated_block(path);
2189 /* there is no extent yet, so try to allocate
2190 * all requested space */
2193 } else if (le32_to_cpu(ex->ee_block) > block) {
2194 /* need to allocate space before found extent */
2196 end = le32_to_cpu(ex->ee_block);
2197 if (block + num < end)
2199 } else if (block >= le32_to_cpu(ex->ee_block)
2200 + ext4_ext_get_actual_len(ex)) {
2201 /* need to allocate space after found extent */
2206 } else if (block >= le32_to_cpu(ex->ee_block)) {
2208 * some part of requested space is covered
2212 end = le32_to_cpu(ex->ee_block)
2213 + ext4_ext_get_actual_len(ex);
2214 if (block + num < end)
2220 BUG_ON(end <= start);
2224 es.es_len = end - start;
2227 es.es_lblk = le32_to_cpu(ex->ee_block);
2228 es.es_len = ext4_ext_get_actual_len(ex);
2229 es.es_pblk = ext4_ext_pblock(ex);
2230 if (ext4_ext_is_unwritten(ex))
2231 flags |= FIEMAP_EXTENT_UNWRITTEN;
2235 * Find delayed extent and update es accordingly. We call
2236 * it even in !exists case to find out whether es is the
2237 * last existing extent or not.
2239 next_del = ext4_find_delayed_extent(inode, &es);
2240 if (!exists && next_del) {
2242 flags |= (FIEMAP_EXTENT_DELALLOC |
2243 FIEMAP_EXTENT_UNKNOWN);
2245 up_read(&EXT4_I(inode)->i_data_sem);
2247 if (unlikely(es.es_len == 0)) {
2248 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2254 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2255 * we need to check next == EXT_MAX_BLOCKS because it is
2256 * possible that an extent is with unwritten and delayed
2257 * status due to when an extent is delayed allocated and
2258 * is allocated by fallocate status tree will track both of
2261 * So we could return a unwritten and delayed extent, and
2262 * its block is equal to 'next'.
2264 if (next == next_del && next == EXT_MAX_BLOCKS) {
2265 flags |= FIEMAP_EXTENT_LAST;
2266 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2267 next != EXT_MAX_BLOCKS)) {
2268 EXT4_ERROR_INODE(inode,
2269 "next extent == %u, next "
2270 "delalloc extent = %u",
2278 err = fiemap_fill_next_extent(fieinfo,
2279 (__u64)es.es_lblk << blksize_bits,
2280 (__u64)es.es_pblk << blksize_bits,
2281 (__u64)es.es_len << blksize_bits,
2291 block = es.es_lblk + es.es_len;
2294 ext4_ext_drop_refs(path);
2300 * ext4_ext_put_gap_in_cache:
2301 * calculate boundaries of the gap that the requested block fits into
2302 * and cache this gap
2305 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2308 int depth = ext_depth(inode);
2309 unsigned long len = 0;
2310 ext4_lblk_t lblock = 0;
2311 struct ext4_extent *ex;
2313 ex = path[depth].p_ext;
2316 * there is no extent yet, so gap is [0;-] and we
2319 ext_debug("cache gap(whole file):");
2320 } else if (block < le32_to_cpu(ex->ee_block)) {
2322 len = le32_to_cpu(ex->ee_block) - block;
2323 ext_debug("cache gap(before): %u [%u:%u]",
2325 le32_to_cpu(ex->ee_block),
2326 ext4_ext_get_actual_len(ex));
2327 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2328 ext4_es_insert_extent(inode, lblock, len, ~0,
2329 EXTENT_STATUS_HOLE);
2330 } else if (block >= le32_to_cpu(ex->ee_block)
2331 + ext4_ext_get_actual_len(ex)) {
2333 lblock = le32_to_cpu(ex->ee_block)
2334 + ext4_ext_get_actual_len(ex);
2336 next = ext4_ext_next_allocated_block(path);
2337 ext_debug("cache gap(after): [%u:%u] %u",
2338 le32_to_cpu(ex->ee_block),
2339 ext4_ext_get_actual_len(ex),
2341 BUG_ON(next == lblock);
2342 len = next - lblock;
2343 if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1))
2344 ext4_es_insert_extent(inode, lblock, len, ~0,
2345 EXTENT_STATUS_HOLE);
2350 ext_debug(" -> %u:%lu\n", lblock, len);
2355 * removes index from the index block.
2357 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2358 struct ext4_ext_path *path, int depth)
2363 /* free index block */
2365 path = path + depth;
2366 leaf = ext4_idx_pblock(path->p_idx);
2367 if (unlikely(path->p_hdr->eh_entries == 0)) {
2368 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2371 err = ext4_ext_get_access(handle, inode, path);
2375 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2376 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2377 len *= sizeof(struct ext4_extent_idx);
2378 memmove(path->p_idx, path->p_idx + 1, len);
2381 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2382 err = ext4_ext_dirty(handle, inode, path);
2385 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2386 trace_ext4_ext_rm_idx(inode, leaf);
2388 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2389 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2391 while (--depth >= 0) {
2392 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2395 err = ext4_ext_get_access(handle, inode, path);
2398 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2399 err = ext4_ext_dirty(handle, inode, path);
2407 * ext4_ext_calc_credits_for_single_extent:
2408 * This routine returns max. credits that needed to insert an extent
2409 * to the extent tree.
2410 * When pass the actual path, the caller should calculate credits
2413 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2414 struct ext4_ext_path *path)
2417 int depth = ext_depth(inode);
2420 /* probably there is space in leaf? */
2421 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2422 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2425 * There are some space in the leaf tree, no
2426 * need to account for leaf block credit
2428 * bitmaps and block group descriptor blocks
2429 * and other metadata blocks still need to be
2432 /* 1 bitmap, 1 block group descriptor */
2433 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2438 return ext4_chunk_trans_blocks(inode, nrblocks);
2442 * How many index/leaf blocks need to change/allocate to add @extents extents?
2444 * If we add a single extent, then in the worse case, each tree level
2445 * index/leaf need to be changed in case of the tree split.
2447 * If more extents are inserted, they could cause the whole tree split more
2448 * than once, but this is really rare.
2450 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2455 /* If we are converting the inline data, only one is needed here. */
2456 if (ext4_has_inline_data(inode))
2459 depth = ext_depth(inode);
2469 static inline int get_default_free_blocks_flags(struct inode *inode)
2471 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2472 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2473 else if (ext4_should_journal_data(inode))
2474 return EXT4_FREE_BLOCKS_FORGET;
2478 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2479 struct ext4_extent *ex,
2480 long long *partial_cluster,
2481 ext4_lblk_t from, ext4_lblk_t to)
2483 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2484 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2486 int flags = get_default_free_blocks_flags(inode);
2489 * For bigalloc file systems, we never free a partial cluster
2490 * at the beginning of the extent. Instead, we make a note
2491 * that we tried freeing the cluster, and check to see if we
2492 * need to free it on a subsequent call to ext4_remove_blocks,
2493 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2495 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2497 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2499 * If we have a partial cluster, and it's different from the
2500 * cluster of the last block, we need to explicitly free the
2501 * partial cluster here.
2503 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2504 if (*partial_cluster > 0 &&
2505 *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2506 ext4_free_blocks(handle, inode, NULL,
2507 EXT4_C2B(sbi, *partial_cluster),
2508 sbi->s_cluster_ratio, flags);
2509 *partial_cluster = 0;
2512 #ifdef EXTENTS_STATS
2514 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2515 spin_lock(&sbi->s_ext_stats_lock);
2516 sbi->s_ext_blocks += ee_len;
2517 sbi->s_ext_extents++;
2518 if (ee_len < sbi->s_ext_min)
2519 sbi->s_ext_min = ee_len;
2520 if (ee_len > sbi->s_ext_max)
2521 sbi->s_ext_max = ee_len;
2522 if (ext_depth(inode) > sbi->s_depth_max)
2523 sbi->s_depth_max = ext_depth(inode);
2524 spin_unlock(&sbi->s_ext_stats_lock);
2527 if (from >= le32_to_cpu(ex->ee_block)
2528 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2531 long long first_cluster;
2533 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2534 pblk = ext4_ext_pblock(ex) + ee_len - num;
2536 * Usually we want to free partial cluster at the end of the
2537 * extent, except for the situation when the cluster is still
2538 * used by any other extent (partial_cluster is negative).
2540 if (*partial_cluster < 0 &&
2541 *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2542 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2544 ext_debug("free last %u blocks starting %llu partial %lld\n",
2545 num, pblk, *partial_cluster);
2546 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2548 * If the block range to be freed didn't start at the
2549 * beginning of a cluster, and we removed the entire
2550 * extent and the cluster is not used by any other extent,
2551 * save the partial cluster here, since we might need to
2552 * delete if we determine that the truncate or punch hole
2553 * operation has removed all of the blocks in the cluster.
2554 * If that cluster is used by another extent, preserve its
2555 * negative value so it isn't freed later on.
2557 * If the whole extent wasn't freed, we've reached the
2558 * start of the truncated/punched region and have finished
2559 * removing blocks. If there's a partial cluster here it's
2560 * shared with the remainder of the extent and is no longer
2561 * a candidate for removal.
2563 if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2564 first_cluster = (long long) EXT4_B2C(sbi, pblk);
2565 if (first_cluster != -*partial_cluster)
2566 *partial_cluster = first_cluster;
2568 *partial_cluster = 0;
2571 ext4_error(sbi->s_sb, "strange request: removal(2) "
2572 "%u-%u from %u:%u\n",
2573 from, to, le32_to_cpu(ex->ee_block), ee_len);
2579 * ext4_ext_rm_leaf() Removes the extents associated with the
2580 * blocks appearing between "start" and "end". Both "start"
2581 * and "end" must appear in the same extent or EIO is returned.
2583 * @handle: The journal handle
2584 * @inode: The files inode
2585 * @path: The path to the leaf
2586 * @partial_cluster: The cluster which we'll have to free if all extents
2587 * has been released from it. However, if this value is
2588 * negative, it's a cluster just to the right of the
2589 * punched region and it must not be freed.
2590 * @start: The first block to remove
2591 * @end: The last block to remove
2594 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2595 struct ext4_ext_path *path,
2596 long long *partial_cluster,
2597 ext4_lblk_t start, ext4_lblk_t end)
2599 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2600 int err = 0, correct_index = 0;
2601 int depth = ext_depth(inode), credits;
2602 struct ext4_extent_header *eh;
2605 ext4_lblk_t ex_ee_block;
2606 unsigned short ex_ee_len;
2607 unsigned unwritten = 0;
2608 struct ext4_extent *ex;
2611 /* the header must be checked already in ext4_ext_remove_space() */
2612 ext_debug("truncate since %u in leaf to %u\n", start, end);
2613 if (!path[depth].p_hdr)
2614 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2615 eh = path[depth].p_hdr;
2616 if (unlikely(path[depth].p_hdr == NULL)) {
2617 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2620 /* find where to start removing */
2621 ex = path[depth].p_ext;
2623 ex = EXT_LAST_EXTENT(eh);
2625 ex_ee_block = le32_to_cpu(ex->ee_block);
2626 ex_ee_len = ext4_ext_get_actual_len(ex);
2628 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2630 while (ex >= EXT_FIRST_EXTENT(eh) &&
2631 ex_ee_block + ex_ee_len > start) {
2633 if (ext4_ext_is_unwritten(ex))
2638 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2639 unwritten, ex_ee_len);
2640 path[depth].p_ext = ex;
2642 a = ex_ee_block > start ? ex_ee_block : start;
2643 b = ex_ee_block+ex_ee_len - 1 < end ?
2644 ex_ee_block+ex_ee_len - 1 : end;
2646 ext_debug(" border %u:%u\n", a, b);
2648 /* If this extent is beyond the end of the hole, skip it */
2649 if (end < ex_ee_block) {
2651 * We're going to skip this extent and move to another,
2652 * so note that its first cluster is in use to avoid
2653 * freeing it when removing blocks. Eventually, the
2654 * right edge of the truncated/punched region will
2655 * be just to the left.
2657 if (sbi->s_cluster_ratio > 1) {
2658 pblk = ext4_ext_pblock(ex);
2660 -(long long) EXT4_B2C(sbi, pblk);
2663 ex_ee_block = le32_to_cpu(ex->ee_block);
2664 ex_ee_len = ext4_ext_get_actual_len(ex);
2666 } else if (b != ex_ee_block + ex_ee_len - 1) {
2667 EXT4_ERROR_INODE(inode,
2668 "can not handle truncate %u:%u "
2670 start, end, ex_ee_block,
2671 ex_ee_block + ex_ee_len - 1);
2674 } else if (a != ex_ee_block) {
2675 /* remove tail of the extent */
2676 num = a - ex_ee_block;
2678 /* remove whole extent: excellent! */
2682 * 3 for leaf, sb, and inode plus 2 (bmap and group
2683 * descriptor) for each block group; assume two block
2684 * groups plus ex_ee_len/blocks_per_block_group for
2687 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2688 if (ex == EXT_FIRST_EXTENT(eh)) {
2690 credits += (ext_depth(inode)) + 1;
2692 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2694 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2698 err = ext4_ext_get_access(handle, inode, path + depth);
2702 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2708 /* this extent is removed; mark slot entirely unused */
2709 ext4_ext_store_pblock(ex, 0);
2711 ex->ee_len = cpu_to_le16(num);
2713 * Do not mark unwritten if all the blocks in the
2714 * extent have been removed.
2716 if (unwritten && num)
2717 ext4_ext_mark_unwritten(ex);
2719 * If the extent was completely released,
2720 * we need to remove it from the leaf
2723 if (end != EXT_MAX_BLOCKS - 1) {
2725 * For hole punching, we need to scoot all the
2726 * extents up when an extent is removed so that
2727 * we dont have blank extents in the middle
2729 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2730 sizeof(struct ext4_extent));
2732 /* Now get rid of the one at the end */
2733 memset(EXT_LAST_EXTENT(eh), 0,
2734 sizeof(struct ext4_extent));
2736 le16_add_cpu(&eh->eh_entries, -1);
2739 err = ext4_ext_dirty(handle, inode, path + depth);
2743 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2744 ext4_ext_pblock(ex));
2746 ex_ee_block = le32_to_cpu(ex->ee_block);
2747 ex_ee_len = ext4_ext_get_actual_len(ex);
2750 if (correct_index && eh->eh_entries)
2751 err = ext4_ext_correct_indexes(handle, inode, path);
2754 * If there's a partial cluster and at least one extent remains in
2755 * the leaf, free the partial cluster if it isn't shared with the
2756 * current extent. If it is shared with the current extent
2757 * we zero partial_cluster because we've reached the start of the
2758 * truncated/punched region and we're done removing blocks.
2760 if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2761 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2762 if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2763 ext4_free_blocks(handle, inode, NULL,
2764 EXT4_C2B(sbi, *partial_cluster),
2765 sbi->s_cluster_ratio,
2766 get_default_free_blocks_flags(inode));
2768 *partial_cluster = 0;
2771 /* if this leaf is free, then we should
2772 * remove it from index block above */
2773 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2774 err = ext4_ext_rm_idx(handle, inode, path, depth);
2781 * ext4_ext_more_to_rm:
2782 * returns 1 if current index has to be freed (even partial)
2785 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2787 BUG_ON(path->p_idx == NULL);
2789 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2793 * if truncate on deeper level happened, it wasn't partial,
2794 * so we have to consider current index for truncation
2796 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2801 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2804 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2805 int depth = ext_depth(inode);
2806 struct ext4_ext_path *path = NULL;
2807 long long partial_cluster = 0;
2811 ext_debug("truncate since %u to %u\n", start, end);
2813 /* probably first extent we're gonna free will be last in block */
2814 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2816 return PTR_ERR(handle);
2819 trace_ext4_ext_remove_space(inode, start, end, depth);
2822 * Check if we are removing extents inside the extent tree. If that
2823 * is the case, we are going to punch a hole inside the extent tree
2824 * so we have to check whether we need to split the extent covering
2825 * the last block to remove so we can easily remove the part of it
2826 * in ext4_ext_rm_leaf().
2828 if (end < EXT_MAX_BLOCKS - 1) {
2829 struct ext4_extent *ex;
2830 ext4_lblk_t ee_block, ex_end, lblk;
2833 /* find extent for or closest extent to this block */
2834 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2836 ext4_journal_stop(handle);
2837 return PTR_ERR(path);
2839 depth = ext_depth(inode);
2840 /* Leaf not may not exist only if inode has no blocks at all */
2841 ex = path[depth].p_ext;
2844 EXT4_ERROR_INODE(inode,
2845 "path[%d].p_hdr == NULL",
2852 ee_block = le32_to_cpu(ex->ee_block);
2853 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2856 * See if the last block is inside the extent, if so split
2857 * the extent at 'end' block so we can easily remove the
2858 * tail of the first part of the split extent in
2859 * ext4_ext_rm_leaf().
2861 if (end >= ee_block && end < ex_end) {
2864 * If we're going to split the extent, note that
2865 * the cluster containing the block after 'end' is
2866 * in use to avoid freeing it when removing blocks.
2868 if (sbi->s_cluster_ratio > 1) {
2869 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2871 -(long long) EXT4_B2C(sbi, pblk);
2875 * Split the extent in two so that 'end' is the last
2876 * block in the first new extent. Also we should not
2877 * fail removing space due to ENOSPC so try to use
2878 * reserved block if that happens.
2880 err = ext4_force_split_extent_at(handle, inode, &path,
2885 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2887 * If there's an extent to the right its first cluster
2888 * contains the immediate right boundary of the
2889 * truncated/punched region. Set partial_cluster to
2890 * its negative value so it won't be freed if shared
2891 * with the current extent. The end < ee_block case
2892 * is handled in ext4_ext_rm_leaf().
2895 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2901 -(long long) EXT4_B2C(sbi, pblk);
2905 * We start scanning from right side, freeing all the blocks
2906 * after i_size and walking into the tree depth-wise.
2908 depth = ext_depth(inode);
2913 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2915 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2918 ext4_journal_stop(handle);
2921 path[0].p_maxdepth = path[0].p_depth = depth;
2922 path[0].p_hdr = ext_inode_hdr(inode);
2925 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2932 while (i >= 0 && err == 0) {
2934 /* this is leaf block */
2935 err = ext4_ext_rm_leaf(handle, inode, path,
2936 &partial_cluster, start,
2938 /* root level has p_bh == NULL, brelse() eats this */
2939 brelse(path[i].p_bh);
2940 path[i].p_bh = NULL;
2945 /* this is index block */
2946 if (!path[i].p_hdr) {
2947 ext_debug("initialize header\n");
2948 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2951 if (!path[i].p_idx) {
2952 /* this level hasn't been touched yet */
2953 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2954 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2955 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2957 le16_to_cpu(path[i].p_hdr->eh_entries));
2959 /* we were already here, see at next index */
2963 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2964 i, EXT_FIRST_INDEX(path[i].p_hdr),
2966 if (ext4_ext_more_to_rm(path + i)) {
2967 struct buffer_head *bh;
2968 /* go to the next level */
2969 ext_debug("move to level %d (block %llu)\n",
2970 i + 1, ext4_idx_pblock(path[i].p_idx));
2971 memset(path + i + 1, 0, sizeof(*path));
2972 bh = read_extent_tree_block(inode,
2973 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2976 /* should we reset i_size? */
2980 /* Yield here to deal with large extent trees.
2981 * Should be a no-op if we did IO above. */
2983 if (WARN_ON(i + 1 > depth)) {
2987 path[i + 1].p_bh = bh;
2989 /* save actual number of indexes since this
2990 * number is changed at the next iteration */
2991 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2994 /* we finished processing this index, go up */
2995 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2996 /* index is empty, remove it;
2997 * handle must be already prepared by the
2998 * truncatei_leaf() */
2999 err = ext4_ext_rm_idx(handle, inode, path, i);
3001 /* root level has p_bh == NULL, brelse() eats this */
3002 brelse(path[i].p_bh);
3003 path[i].p_bh = NULL;
3005 ext_debug("return to level %d\n", i);
3009 trace_ext4_ext_remove_space_done(inode, start, end, depth,
3010 partial_cluster, path->p_hdr->eh_entries);
3013 * If we still have something in the partial cluster and we have removed
3014 * even the first extent, then we should free the blocks in the partial
3015 * cluster as well. (This code will only run when there are no leaves
3016 * to the immediate left of the truncated/punched region.)
3018 if (partial_cluster > 0 && err == 0) {
3019 /* don't zero partial_cluster since it's not used afterwards */
3020 ext4_free_blocks(handle, inode, NULL,
3021 EXT4_C2B(sbi, partial_cluster),
3022 sbi->s_cluster_ratio,
3023 get_default_free_blocks_flags(inode));
3026 /* TODO: flexible tree reduction should be here */
3027 if (path->p_hdr->eh_entries == 0) {
3029 * truncate to zero freed all the tree,
3030 * so we need to correct eh_depth
3032 err = ext4_ext_get_access(handle, inode, path);
3034 ext_inode_hdr(inode)->eh_depth = 0;
3035 ext_inode_hdr(inode)->eh_max =
3036 cpu_to_le16(ext4_ext_space_root(inode, 0));
3037 err = ext4_ext_dirty(handle, inode, path);
3041 ext4_ext_drop_refs(path);
3046 ext4_journal_stop(handle);
3052 * called at mount time
3054 void ext4_ext_init(struct super_block *sb)
3057 * possible initialization would be here
3060 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
3061 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3062 printk(KERN_INFO "EXT4-fs: file extents enabled"
3063 #ifdef AGGRESSIVE_TEST
3064 ", aggressive tests"
3066 #ifdef CHECK_BINSEARCH
3069 #ifdef EXTENTS_STATS
3074 #ifdef EXTENTS_STATS
3075 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3076 EXT4_SB(sb)->s_ext_min = 1 << 30;
3077 EXT4_SB(sb)->s_ext_max = 0;
3083 * called at umount time
3085 void ext4_ext_release(struct super_block *sb)
3087 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3090 #ifdef EXTENTS_STATS
3091 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3092 struct ext4_sb_info *sbi = EXT4_SB(sb);
3093 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3094 sbi->s_ext_blocks, sbi->s_ext_extents,
3095 sbi->s_ext_blocks / sbi->s_ext_extents);
3096 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3097 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3102 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3104 ext4_lblk_t ee_block;
3105 ext4_fsblk_t ee_pblock;
3106 unsigned int ee_len;
3108 ee_block = le32_to_cpu(ex->ee_block);
3109 ee_len = ext4_ext_get_actual_len(ex);
3110 ee_pblock = ext4_ext_pblock(ex);
3115 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3116 EXTENT_STATUS_WRITTEN);
3119 /* FIXME!! we need to try to merge to left or right after zero-out */
3120 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3122 ext4_fsblk_t ee_pblock;
3123 unsigned int ee_len;
3126 ee_len = ext4_ext_get_actual_len(ex);
3127 ee_pblock = ext4_ext_pblock(ex);
3129 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3137 * ext4_split_extent_at() splits an extent at given block.
3139 * @handle: the journal handle
3140 * @inode: the file inode
3141 * @path: the path to the extent
3142 * @split: the logical block where the extent is splitted.
3143 * @split_flags: indicates if the extent could be zeroout if split fails, and
3144 * the states(init or unwritten) of new extents.
3145 * @flags: flags used to insert new extent to extent tree.
3148 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3149 * of which are deterimined by split_flag.
3151 * There are two cases:
3152 * a> the extent are splitted into two extent.
3153 * b> split is not needed, and just mark the extent.
3155 * return 0 on success.
3157 static int ext4_split_extent_at(handle_t *handle,
3158 struct inode *inode,
3159 struct ext4_ext_path **ppath,
3164 struct ext4_ext_path *path = *ppath;
3165 ext4_fsblk_t newblock;
3166 ext4_lblk_t ee_block;
3167 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3168 struct ext4_extent *ex2 = NULL;
3169 unsigned int ee_len, depth;
3172 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3173 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3175 ext_debug("ext4_split_extents_at: inode %lu, logical"
3176 "block %llu\n", inode->i_ino, (unsigned long long)split);
3178 ext4_ext_show_leaf(inode, path);
3180 depth = ext_depth(inode);
3181 ex = path[depth].p_ext;
3182 ee_block = le32_to_cpu(ex->ee_block);
3183 ee_len = ext4_ext_get_actual_len(ex);
3184 newblock = split - ee_block + ext4_ext_pblock(ex);
3186 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3187 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3188 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3189 EXT4_EXT_MARK_UNWRIT1 |
3190 EXT4_EXT_MARK_UNWRIT2));
3192 err = ext4_ext_get_access(handle, inode, path + depth);
3196 if (split == ee_block) {
3198 * case b: block @split is the block that the extent begins with
3199 * then we just change the state of the extent, and splitting
3202 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3203 ext4_ext_mark_unwritten(ex);
3205 ext4_ext_mark_initialized(ex);
3207 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3208 ext4_ext_try_to_merge(handle, inode, path, ex);
3210 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3215 memcpy(&orig_ex, ex, sizeof(orig_ex));
3216 ex->ee_len = cpu_to_le16(split - ee_block);
3217 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3218 ext4_ext_mark_unwritten(ex);
3221 * path may lead to new leaf, not to original leaf any more
3222 * after ext4_ext_insert_extent() returns,
3224 err = ext4_ext_dirty(handle, inode, path + depth);
3226 goto fix_extent_len;
3229 ex2->ee_block = cpu_to_le32(split);
3230 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3231 ext4_ext_store_pblock(ex2, newblock);
3232 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3233 ext4_ext_mark_unwritten(ex2);
3235 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3236 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3237 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3238 if (split_flag & EXT4_EXT_DATA_VALID1) {
3239 err = ext4_ext_zeroout(inode, ex2);
3240 zero_ex.ee_block = ex2->ee_block;
3241 zero_ex.ee_len = cpu_to_le16(
3242 ext4_ext_get_actual_len(ex2));
3243 ext4_ext_store_pblock(&zero_ex,
3244 ext4_ext_pblock(ex2));
3246 err = ext4_ext_zeroout(inode, ex);
3247 zero_ex.ee_block = ex->ee_block;
3248 zero_ex.ee_len = cpu_to_le16(
3249 ext4_ext_get_actual_len(ex));
3250 ext4_ext_store_pblock(&zero_ex,
3251 ext4_ext_pblock(ex));
3254 err = ext4_ext_zeroout(inode, &orig_ex);
3255 zero_ex.ee_block = orig_ex.ee_block;
3256 zero_ex.ee_len = cpu_to_le16(
3257 ext4_ext_get_actual_len(&orig_ex));
3258 ext4_ext_store_pblock(&zero_ex,
3259 ext4_ext_pblock(&orig_ex));
3263 goto fix_extent_len;
3264 /* update the extent length and mark as initialized */
3265 ex->ee_len = cpu_to_le16(ee_len);
3266 ext4_ext_try_to_merge(handle, inode, path, ex);
3267 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3269 goto fix_extent_len;
3271 /* update extent status tree */
3272 err = ext4_zeroout_es(inode, &zero_ex);
3276 goto fix_extent_len;
3279 ext4_ext_show_leaf(inode, path);
3283 ex->ee_len = orig_ex.ee_len;
3284 ext4_ext_dirty(handle, inode, path + path->p_depth);
3289 * ext4_split_extents() splits an extent and mark extent which is covered
3290 * by @map as split_flags indicates
3292 * It may result in splitting the extent into multiple extents (up to three)
3293 * There are three possibilities:
3294 * a> There is no split required
3295 * b> Splits in two extents: Split is happening at either end of the extent
3296 * c> Splits in three extents: Somone is splitting in middle of the extent
3299 static int ext4_split_extent(handle_t *handle,
3300 struct inode *inode,
3301 struct ext4_ext_path **ppath,
3302 struct ext4_map_blocks *map,
3306 struct ext4_ext_path *path = *ppath;
3307 ext4_lblk_t ee_block;
3308 struct ext4_extent *ex;
3309 unsigned int ee_len, depth;
3312 int split_flag1, flags1;
3313 int allocated = map->m_len;
3315 depth = ext_depth(inode);
3316 ex = path[depth].p_ext;
3317 ee_block = le32_to_cpu(ex->ee_block);
3318 ee_len = ext4_ext_get_actual_len(ex);
3319 unwritten = ext4_ext_is_unwritten(ex);
3321 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3322 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3323 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3325 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3326 EXT4_EXT_MARK_UNWRIT2;
3327 if (split_flag & EXT4_EXT_DATA_VALID2)
3328 split_flag1 |= EXT4_EXT_DATA_VALID1;
3329 err = ext4_split_extent_at(handle, inode, ppath,
3330 map->m_lblk + map->m_len, split_flag1, flags1);
3334 allocated = ee_len - (map->m_lblk - ee_block);
3337 * Update path is required because previous ext4_split_extent_at() may
3338 * result in split of original leaf or extent zeroout.
3340 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3342 return PTR_ERR(path);
3343 depth = ext_depth(inode);
3344 ex = path[depth].p_ext;
3346 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3347 (unsigned long) map->m_lblk);
3350 unwritten = ext4_ext_is_unwritten(ex);
3353 if (map->m_lblk >= ee_block) {
3354 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3356 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3357 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3358 EXT4_EXT_MARK_UNWRIT2);
3360 err = ext4_split_extent_at(handle, inode, ppath,
3361 map->m_lblk, split_flag1, flags);
3366 ext4_ext_show_leaf(inode, path);
3368 return err ? err : allocated;
3372 * This function is called by ext4_ext_map_blocks() if someone tries to write
3373 * to an unwritten extent. It may result in splitting the unwritten
3374 * extent into multiple extents (up to three - one initialized and two
3376 * There are three possibilities:
3377 * a> There is no split required: Entire extent should be initialized
3378 * b> Splits in two extents: Write is happening at either end of the extent
3379 * c> Splits in three extents: Somone is writing in middle of the extent
3382 * - The extent pointed to by 'path' is unwritten.
3383 * - The extent pointed to by 'path' contains a superset
3384 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3386 * Post-conditions on success:
3387 * - the returned value is the number of blocks beyond map->l_lblk
3388 * that are allocated and initialized.
3389 * It is guaranteed to be >= map->m_len.
3391 static int ext4_ext_convert_to_initialized(handle_t *handle,
3392 struct inode *inode,
3393 struct ext4_map_blocks *map,
3394 struct ext4_ext_path **ppath,
3397 struct ext4_ext_path *path = *ppath;
3398 struct ext4_sb_info *sbi;
3399 struct ext4_extent_header *eh;
3400 struct ext4_map_blocks split_map;
3401 struct ext4_extent zero_ex;
3402 struct ext4_extent *ex, *abut_ex;
3403 ext4_lblk_t ee_block, eof_block;
3404 unsigned int ee_len, depth, map_len = map->m_len;
3405 int allocated = 0, max_zeroout = 0;
3409 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3410 "block %llu, max_blocks %u\n", inode->i_ino,
3411 (unsigned long long)map->m_lblk, map_len);
3413 sbi = EXT4_SB(inode->i_sb);
3414 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3415 inode->i_sb->s_blocksize_bits;
3416 if (eof_block < map->m_lblk + map_len)
3417 eof_block = map->m_lblk + map_len;
3419 depth = ext_depth(inode);
3420 eh = path[depth].p_hdr;
3421 ex = path[depth].p_ext;
3422 ee_block = le32_to_cpu(ex->ee_block);
3423 ee_len = ext4_ext_get_actual_len(ex);
3426 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3428 /* Pre-conditions */
3429 BUG_ON(!ext4_ext_is_unwritten(ex));
3430 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3433 * Attempt to transfer newly initialized blocks from the currently
3434 * unwritten extent to its neighbor. This is much cheaper
3435 * than an insertion followed by a merge as those involve costly
3436 * memmove() calls. Transferring to the left is the common case in
3437 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3438 * followed by append writes.
3440 * Limitations of the current logic:
3441 * - L1: we do not deal with writes covering the whole extent.
3442 * This would require removing the extent if the transfer
3444 * - L2: we only attempt to merge with an extent stored in the
3445 * same extent tree node.
3447 if ((map->m_lblk == ee_block) &&
3448 /* See if we can merge left */
3449 (map_len < ee_len) && /*L1*/
3450 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3451 ext4_lblk_t prev_lblk;
3452 ext4_fsblk_t prev_pblk, ee_pblk;
3453 unsigned int prev_len;
3456 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3457 prev_len = ext4_ext_get_actual_len(abut_ex);
3458 prev_pblk = ext4_ext_pblock(abut_ex);
3459 ee_pblk = ext4_ext_pblock(ex);
3462 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3463 * upon those conditions:
3464 * - C1: abut_ex is initialized,
3465 * - C2: abut_ex is logically abutting ex,
3466 * - C3: abut_ex is physically abutting ex,
3467 * - C4: abut_ex can receive the additional blocks without
3468 * overflowing the (initialized) length limit.
3470 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3471 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3472 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3473 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3474 err = ext4_ext_get_access(handle, inode, path + depth);
3478 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3481 /* Shift the start of ex by 'map_len' blocks */
3482 ex->ee_block = cpu_to_le32(ee_block + map_len);
3483 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3484 ex->ee_len = cpu_to_le16(ee_len - map_len);
3485 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3487 /* Extend abut_ex by 'map_len' blocks */
3488 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3490 /* Result: number of initialized blocks past m_lblk */
3491 allocated = map_len;
3493 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3494 (map_len < ee_len) && /*L1*/
3495 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3496 /* See if we can merge right */
3497 ext4_lblk_t next_lblk;
3498 ext4_fsblk_t next_pblk, ee_pblk;
3499 unsigned int next_len;
3502 next_lblk = le32_to_cpu(abut_ex->ee_block);
3503 next_len = ext4_ext_get_actual_len(abut_ex);
3504 next_pblk = ext4_ext_pblock(abut_ex);
3505 ee_pblk = ext4_ext_pblock(ex);
3508 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3509 * upon those conditions:
3510 * - C1: abut_ex is initialized,
3511 * - C2: abut_ex is logically abutting ex,
3512 * - C3: abut_ex is physically abutting ex,
3513 * - C4: abut_ex can receive the additional blocks without
3514 * overflowing the (initialized) length limit.
3516 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3517 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3518 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3519 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3520 err = ext4_ext_get_access(handle, inode, path + depth);
3524 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3527 /* Shift the start of abut_ex by 'map_len' blocks */
3528 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3529 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3530 ex->ee_len = cpu_to_le16(ee_len - map_len);
3531 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3533 /* Extend abut_ex by 'map_len' blocks */
3534 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3536 /* Result: number of initialized blocks past m_lblk */
3537 allocated = map_len;
3541 /* Mark the block containing both extents as dirty */
3542 ext4_ext_dirty(handle, inode, path + depth);
3544 /* Update path to point to the right extent */
3545 path[depth].p_ext = abut_ex;
3548 allocated = ee_len - (map->m_lblk - ee_block);
3550 WARN_ON(map->m_lblk < ee_block);
3552 * It is safe to convert extent to initialized via explicit
3553 * zeroout only if extent is fully inside i_size or new_size.
3555 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3557 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3558 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3559 (inode->i_sb->s_blocksize_bits - 10);
3561 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3562 if (max_zeroout && (ee_len <= max_zeroout)) {
3563 err = ext4_ext_zeroout(inode, ex);
3566 zero_ex.ee_block = ex->ee_block;
3567 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3568 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3570 err = ext4_ext_get_access(handle, inode, path + depth);
3573 ext4_ext_mark_initialized(ex);
3574 ext4_ext_try_to_merge(handle, inode, path, ex);
3575 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3581 * 1. split the extent into three extents.
3582 * 2. split the extent into two extents, zeroout the first half.
3583 * 3. split the extent into two extents, zeroout the second half.
3584 * 4. split the extent into two extents with out zeroout.
3586 split_map.m_lblk = map->m_lblk;
3587 split_map.m_len = map->m_len;
3589 if (max_zeroout && (allocated > map->m_len)) {
3590 if (allocated <= max_zeroout) {
3593 cpu_to_le32(map->m_lblk);
3594 zero_ex.ee_len = cpu_to_le16(allocated);
3595 ext4_ext_store_pblock(&zero_ex,
3596 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3597 err = ext4_ext_zeroout(inode, &zero_ex);
3600 split_map.m_lblk = map->m_lblk;
3601 split_map.m_len = allocated;
3602 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3604 if (map->m_lblk != ee_block) {
3605 zero_ex.ee_block = ex->ee_block;
3606 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3608 ext4_ext_store_pblock(&zero_ex,
3609 ext4_ext_pblock(ex));
3610 err = ext4_ext_zeroout(inode, &zero_ex);
3615 split_map.m_lblk = ee_block;
3616 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3617 allocated = map->m_len;
3621 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3626 /* If we have gotten a failure, don't zero out status tree */
3628 err = ext4_zeroout_es(inode, &zero_ex);
3629 return err ? err : allocated;
3633 * This function is called by ext4_ext_map_blocks() from
3634 * ext4_get_blocks_dio_write() when DIO to write
3635 * to an unwritten extent.
3637 * Writing to an unwritten extent may result in splitting the unwritten
3638 * extent into multiple initialized/unwritten extents (up to three)
3639 * There are three possibilities:
3640 * a> There is no split required: Entire extent should be unwritten
3641 * b> Splits in two extents: Write is happening at either end of the extent
3642 * c> Splits in three extents: Somone is writing in middle of the extent
3644 * This works the same way in the case of initialized -> unwritten conversion.
3646 * One of more index blocks maybe needed if the extent tree grow after
3647 * the unwritten extent split. To prevent ENOSPC occur at the IO
3648 * complete, we need to split the unwritten extent before DIO submit
3649 * the IO. The unwritten extent called at this time will be split
3650 * into three unwritten extent(at most). After IO complete, the part
3651 * being filled will be convert to initialized by the end_io callback function
3652 * via ext4_convert_unwritten_extents().
3654 * Returns the size of unwritten extent to be written on success.
3656 static int ext4_split_convert_extents(handle_t *handle,
3657 struct inode *inode,
3658 struct ext4_map_blocks *map,
3659 struct ext4_ext_path **ppath,
3662 struct ext4_ext_path *path = *ppath;
3663 ext4_lblk_t eof_block;
3664 ext4_lblk_t ee_block;
3665 struct ext4_extent *ex;
3666 unsigned int ee_len;
3667 int split_flag = 0, depth;
3669 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3670 __func__, inode->i_ino,
3671 (unsigned long long)map->m_lblk, map->m_len);
3673 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3674 inode->i_sb->s_blocksize_bits;
3675 if (eof_block < map->m_lblk + map->m_len)
3676 eof_block = map->m_lblk + map->m_len;
3678 * It is safe to convert extent to initialized via explicit
3679 * zeroout only if extent is fully insde i_size or new_size.
3681 depth = ext_depth(inode);
3682 ex = path[depth].p_ext;
3683 ee_block = le32_to_cpu(ex->ee_block);
3684 ee_len = ext4_ext_get_actual_len(ex);
3686 /* Convert to unwritten */
3687 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3688 split_flag |= EXT4_EXT_DATA_VALID1;
3689 /* Convert to initialized */
3690 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3691 split_flag |= ee_block + ee_len <= eof_block ?
3692 EXT4_EXT_MAY_ZEROOUT : 0;
3693 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3695 flags |= EXT4_GET_BLOCKS_PRE_IO;
3696 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3699 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3700 struct inode *inode,
3701 struct ext4_map_blocks *map,
3702 struct ext4_ext_path **ppath)
3704 struct ext4_ext_path *path = *ppath;
3705 struct ext4_extent *ex;
3706 ext4_lblk_t ee_block;
3707 unsigned int ee_len;
3711 depth = ext_depth(inode);
3712 ex = path[depth].p_ext;
3713 ee_block = le32_to_cpu(ex->ee_block);
3714 ee_len = ext4_ext_get_actual_len(ex);
3716 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3717 "block %llu, max_blocks %u\n", inode->i_ino,
3718 (unsigned long long)ee_block, ee_len);
3720 /* If extent is larger than requested it is a clear sign that we still
3721 * have some extent state machine issues left. So extent_split is still
3723 * TODO: Once all related issues will be fixed this situation should be
3726 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3728 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3729 " len %u; IO logical block %llu, len %u\n",
3730 inode->i_ino, (unsigned long long)ee_block, ee_len,
3731 (unsigned long long)map->m_lblk, map->m_len);
3733 err = ext4_split_convert_extents(handle, inode, map, ppath,
3734 EXT4_GET_BLOCKS_CONVERT);
3737 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3739 return PTR_ERR(path);
3740 depth = ext_depth(inode);
3741 ex = path[depth].p_ext;
3744 err = ext4_ext_get_access(handle, inode, path + depth);
3747 /* first mark the extent as initialized */
3748 ext4_ext_mark_initialized(ex);
3750 /* note: ext4_ext_correct_indexes() isn't needed here because
3751 * borders are not changed
3753 ext4_ext_try_to_merge(handle, inode, path, ex);
3755 /* Mark modified extent as dirty */
3756 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3758 ext4_ext_show_leaf(inode, path);
3762 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3763 sector_t block, int count)
3766 for (i = 0; i < count; i++)
3767 unmap_underlying_metadata(bdev, block + i);
3771 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3773 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3775 struct ext4_ext_path *path,
3779 struct ext4_extent_header *eh;
3780 struct ext4_extent *last_ex;
3782 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3785 depth = ext_depth(inode);
3786 eh = path[depth].p_hdr;
3789 * We're going to remove EOFBLOCKS_FL entirely in future so we
3790 * do not care for this case anymore. Simply remove the flag
3791 * if there are no extents.
3793 if (unlikely(!eh->eh_entries))
3795 last_ex = EXT_LAST_EXTENT(eh);
3797 * We should clear the EOFBLOCKS_FL flag if we are writing the
3798 * last block in the last extent in the file. We test this by
3799 * first checking to see if the caller to
3800 * ext4_ext_get_blocks() was interested in the last block (or
3801 * a block beyond the last block) in the current extent. If
3802 * this turns out to be false, we can bail out from this
3803 * function immediately.
3805 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3806 ext4_ext_get_actual_len(last_ex))
3809 * If the caller does appear to be planning to write at or
3810 * beyond the end of the current extent, we then test to see
3811 * if the current extent is the last extent in the file, by
3812 * checking to make sure it was reached via the rightmost node
3813 * at each level of the tree.
3815 for (i = depth-1; i >= 0; i--)
3816 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3819 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3820 return ext4_mark_inode_dirty(handle, inode);
3824 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3826 * Return 1 if there is a delalloc block in the range, otherwise 0.
3828 int ext4_find_delalloc_range(struct inode *inode,
3829 ext4_lblk_t lblk_start,
3830 ext4_lblk_t lblk_end)
3832 struct extent_status es;
3834 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3836 return 0; /* there is no delay extent in this tree */
3837 else if (es.es_lblk <= lblk_start &&
3838 lblk_start < es.es_lblk + es.es_len)
3840 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3846 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3848 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3849 ext4_lblk_t lblk_start, lblk_end;
3850 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3851 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3853 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3857 * Determines how many complete clusters (out of those specified by the 'map')
3858 * are under delalloc and were reserved quota for.
3859 * This function is called when we are writing out the blocks that were
3860 * originally written with their allocation delayed, but then the space was
3861 * allocated using fallocate() before the delayed allocation could be resolved.
3862 * The cases to look for are:
3863 * ('=' indicated delayed allocated blocks
3864 * '-' indicates non-delayed allocated blocks)
3865 * (a) partial clusters towards beginning and/or end outside of allocated range
3866 * are not delalloc'ed.
3868 * |----c---=|====c====|====c====|===-c----|
3869 * |++++++ allocated ++++++|
3870 * ==> 4 complete clusters in above example
3872 * (b) partial cluster (outside of allocated range) towards either end is
3873 * marked for delayed allocation. In this case, we will exclude that
3876 * |----====c========|========c========|
3877 * |++++++ allocated ++++++|
3878 * ==> 1 complete clusters in above example
3881 * |================c================|
3882 * |++++++ allocated ++++++|
3883 * ==> 0 complete clusters in above example
3885 * The ext4_da_update_reserve_space will be called only if we
3886 * determine here that there were some "entire" clusters that span
3887 * this 'allocated' range.
3888 * In the non-bigalloc case, this function will just end up returning num_blks
3889 * without ever calling ext4_find_delalloc_range.
3892 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3893 unsigned int num_blks)
3895 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3896 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3897 ext4_lblk_t lblk_from, lblk_to, c_offset;
3898 unsigned int allocated_clusters = 0;
3900 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3901 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3903 /* max possible clusters for this allocation */
3904 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3906 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3908 /* Check towards left side */
3909 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3911 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3912 lblk_to = lblk_from + c_offset - 1;
3914 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3915 allocated_clusters--;
3918 /* Now check towards right. */
3919 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3920 if (allocated_clusters && c_offset) {
3921 lblk_from = lblk_start + num_blks;
3922 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3924 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3925 allocated_clusters--;
3928 return allocated_clusters;
3932 convert_initialized_extent(handle_t *handle, struct inode *inode,
3933 struct ext4_map_blocks *map,
3934 struct ext4_ext_path **ppath, int flags,
3935 unsigned int allocated, ext4_fsblk_t newblock)
3937 struct ext4_ext_path *path = *ppath;
3938 struct ext4_extent *ex;
3939 ext4_lblk_t ee_block;
3940 unsigned int ee_len;
3945 * Make sure that the extent is no bigger than we support with
3948 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3949 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3951 depth = ext_depth(inode);
3952 ex = path[depth].p_ext;
3953 ee_block = le32_to_cpu(ex->ee_block);
3954 ee_len = ext4_ext_get_actual_len(ex);
3956 ext_debug("%s: inode %lu, logical"
3957 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3958 (unsigned long long)ee_block, ee_len);
3960 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3961 err = ext4_split_convert_extents(handle, inode, map, ppath,
3962 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3965 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3967 return PTR_ERR(path);
3968 depth = ext_depth(inode);
3969 ex = path[depth].p_ext;
3971 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3972 (unsigned long) map->m_lblk);
3977 err = ext4_ext_get_access(handle, inode, path + depth);
3980 /* first mark the extent as unwritten */
3981 ext4_ext_mark_unwritten(ex);
3983 /* note: ext4_ext_correct_indexes() isn't needed here because
3984 * borders are not changed
3986 ext4_ext_try_to_merge(handle, inode, path, ex);
3988 /* Mark modified extent as dirty */
3989 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3992 ext4_ext_show_leaf(inode, path);
3994 ext4_update_inode_fsync_trans(handle, inode, 1);
3995 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3998 map->m_flags |= EXT4_MAP_UNWRITTEN;
3999 if (allocated > map->m_len)
4000 allocated = map->m_len;
4001 map->m_len = allocated;
4006 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4007 struct ext4_map_blocks *map,
4008 struct ext4_ext_path **ppath, int flags,
4009 unsigned int allocated, ext4_fsblk_t newblock)
4011 struct ext4_ext_path *path = *ppath;
4014 ext4_io_end_t *io = ext4_inode_aio(inode);
4016 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4017 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4018 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4020 ext4_ext_show_leaf(inode, path);
4023 * When writing into unwritten space, we should not fail to
4024 * allocate metadata blocks for the new extent block if needed.
4026 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4028 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4029 allocated, newblock);
4031 /* get_block() before submit the IO, split the extent */
4032 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4033 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4034 flags | EXT4_GET_BLOCKS_CONVERT);
4038 * Flag the inode(non aio case) or end_io struct (aio case)
4039 * that this IO needs to conversion to written when IO is
4043 ext4_set_io_unwritten_flag(inode, io);
4045 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4046 map->m_flags |= EXT4_MAP_UNWRITTEN;
4049 /* IO end_io complete, convert the filled extent to written */
4050 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4051 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4054 ext4_update_inode_fsync_trans(handle, inode, 1);
4055 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4059 map->m_flags |= EXT4_MAP_MAPPED;
4060 map->m_pblk = newblock;
4061 if (allocated > map->m_len)
4062 allocated = map->m_len;
4063 map->m_len = allocated;
4066 /* buffered IO case */
4068 * repeat fallocate creation request
4069 * we already have an unwritten extent
4071 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4072 map->m_flags |= EXT4_MAP_UNWRITTEN;
4076 /* buffered READ or buffered write_begin() lookup */
4077 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4079 * We have blocks reserved already. We
4080 * return allocated blocks so that delalloc
4081 * won't do block reservation for us. But
4082 * the buffer head will be unmapped so that
4083 * a read from the block returns 0s.
4085 map->m_flags |= EXT4_MAP_UNWRITTEN;
4089 /* buffered write, writepage time, convert*/
4090 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4092 ext4_update_inode_fsync_trans(handle, inode, 1);
4099 map->m_flags |= EXT4_MAP_NEW;
4101 * if we allocated more blocks than requested
4102 * we need to make sure we unmap the extra block
4103 * allocated. The actual needed block will get
4104 * unmapped later when we find the buffer_head marked
4107 if (allocated > map->m_len) {
4108 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4109 newblock + map->m_len,
4110 allocated - map->m_len);
4111 allocated = map->m_len;
4113 map->m_len = allocated;
4116 * If we have done fallocate with the offset that is already
4117 * delayed allocated, we would have block reservation
4118 * and quota reservation done in the delayed write path.
4119 * But fallocate would have already updated quota and block
4120 * count for this offset. So cancel these reservation
4122 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4123 unsigned int reserved_clusters;
4124 reserved_clusters = get_reserved_cluster_alloc(inode,
4125 map->m_lblk, map->m_len);
4126 if (reserved_clusters)
4127 ext4_da_update_reserve_space(inode,
4133 map->m_flags |= EXT4_MAP_MAPPED;
4134 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4135 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4141 if (allocated > map->m_len)
4142 allocated = map->m_len;
4143 ext4_ext_show_leaf(inode, path);
4144 map->m_pblk = newblock;
4145 map->m_len = allocated;
4147 return err ? err : allocated;
4151 * get_implied_cluster_alloc - check to see if the requested
4152 * allocation (in the map structure) overlaps with a cluster already
4153 * allocated in an extent.
4154 * @sb The filesystem superblock structure
4155 * @map The requested lblk->pblk mapping
4156 * @ex The extent structure which might contain an implied
4157 * cluster allocation
4159 * This function is called by ext4_ext_map_blocks() after we failed to
4160 * find blocks that were already in the inode's extent tree. Hence,
4161 * we know that the beginning of the requested region cannot overlap
4162 * the extent from the inode's extent tree. There are three cases we
4163 * want to catch. The first is this case:
4165 * |--- cluster # N--|
4166 * |--- extent ---| |---- requested region ---|
4169 * The second case that we need to test for is this one:
4171 * |--------- cluster # N ----------------|
4172 * |--- requested region --| |------- extent ----|
4173 * |=======================|
4175 * The third case is when the requested region lies between two extents
4176 * within the same cluster:
4177 * |------------- cluster # N-------------|
4178 * |----- ex -----| |---- ex_right ----|
4179 * |------ requested region ------|
4180 * |================|
4182 * In each of the above cases, we need to set the map->m_pblk and
4183 * map->m_len so it corresponds to the return the extent labelled as
4184 * "|====|" from cluster #N, since it is already in use for data in
4185 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4186 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4187 * as a new "allocated" block region. Otherwise, we will return 0 and
4188 * ext4_ext_map_blocks() will then allocate one or more new clusters
4189 * by calling ext4_mb_new_blocks().
4191 static int get_implied_cluster_alloc(struct super_block *sb,
4192 struct ext4_map_blocks *map,
4193 struct ext4_extent *ex,
4194 struct ext4_ext_path *path)
4196 struct ext4_sb_info *sbi = EXT4_SB(sb);
4197 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4198 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4199 ext4_lblk_t rr_cluster_start;
4200 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4201 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4202 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4204 /* The extent passed in that we are trying to match */
4205 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4206 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4208 /* The requested region passed into ext4_map_blocks() */
4209 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4211 if ((rr_cluster_start == ex_cluster_end) ||
4212 (rr_cluster_start == ex_cluster_start)) {
4213 if (rr_cluster_start == ex_cluster_end)
4214 ee_start += ee_len - 1;
4215 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4216 map->m_len = min(map->m_len,
4217 (unsigned) sbi->s_cluster_ratio - c_offset);
4219 * Check for and handle this case:
4221 * |--------- cluster # N-------------|
4222 * |------- extent ----|
4223 * |--- requested region ---|
4227 if (map->m_lblk < ee_block)
4228 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4231 * Check for the case where there is already another allocated
4232 * block to the right of 'ex' but before the end of the cluster.
4234 * |------------- cluster # N-------------|
4235 * |----- ex -----| |---- ex_right ----|
4236 * |------ requested region ------|
4237 * |================|
4239 if (map->m_lblk > ee_block) {
4240 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4241 map->m_len = min(map->m_len, next - map->m_lblk);
4244 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4248 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4254 * Block allocation/map/preallocation routine for extents based files
4257 * Need to be called with
4258 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4259 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4261 * return > 0, number of of blocks already mapped/allocated
4262 * if create == 0 and these are pre-allocated blocks
4263 * buffer head is unmapped
4264 * otherwise blocks are mapped
4266 * return = 0, if plain look up failed (blocks have not been allocated)
4267 * buffer head is unmapped
4269 * return < 0, error case.
4271 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4272 struct ext4_map_blocks *map, int flags)
4274 struct ext4_ext_path *path = NULL;
4275 struct ext4_extent newex, *ex, *ex2;
4276 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4277 ext4_fsblk_t newblock = 0;
4278 int free_on_err = 0, err = 0, depth, ret;
4279 unsigned int allocated = 0, offset = 0;
4280 unsigned int allocated_clusters = 0;
4281 struct ext4_allocation_request ar;
4282 ext4_io_end_t *io = ext4_inode_aio(inode);
4283 ext4_lblk_t cluster_offset;
4284 int set_unwritten = 0;
4285 bool map_from_cluster = false;
4287 ext_debug("blocks %u/%u requested for inode %lu\n",
4288 map->m_lblk, map->m_len, inode->i_ino);
4289 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4291 /* find extent for this block */
4292 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4294 err = PTR_ERR(path);
4299 depth = ext_depth(inode);
4302 * consistent leaf must not be empty;
4303 * this situation is possible, though, _during_ tree modification;
4304 * this is why assert can't be put in ext4_find_extent()
4306 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4307 EXT4_ERROR_INODE(inode, "bad extent address "
4308 "lblock: %lu, depth: %d pblock %lld",
4309 (unsigned long) map->m_lblk, depth,
4310 path[depth].p_block);
4315 ex = path[depth].p_ext;
4317 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4318 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4319 unsigned short ee_len;
4323 * unwritten extents are treated as holes, except that
4324 * we split out initialized portions during a write.
4326 ee_len = ext4_ext_get_actual_len(ex);
4328 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4330 /* if found extent covers block, simply return it */
4331 if (in_range(map->m_lblk, ee_block, ee_len)) {
4332 newblock = map->m_lblk - ee_block + ee_start;
4333 /* number of remaining blocks in the extent */
4334 allocated = ee_len - (map->m_lblk - ee_block);
4335 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4336 ee_block, ee_len, newblock);
4339 * If the extent is initialized check whether the
4340 * caller wants to convert it to unwritten.
4342 if ((!ext4_ext_is_unwritten(ex)) &&
4343 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4344 allocated = convert_initialized_extent(
4345 handle, inode, map, &path,
4346 flags, allocated, newblock);
4348 } else if (!ext4_ext_is_unwritten(ex))
4351 ret = ext4_ext_handle_unwritten_extents(
4352 handle, inode, map, &path, flags,
4353 allocated, newblock);
4363 * requested block isn't allocated yet;
4364 * we couldn't try to create block if create flag is zero
4366 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4368 * put just found gap into cache to speed up
4369 * subsequent requests
4371 if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0)
4372 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4377 * Okay, we need to do block allocation.
4379 newex.ee_block = cpu_to_le32(map->m_lblk);
4380 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4383 * If we are doing bigalloc, check to see if the extent returned
4384 * by ext4_find_extent() implies a cluster we can use.
4386 if (cluster_offset && ex &&
4387 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4388 ar.len = allocated = map->m_len;
4389 newblock = map->m_pblk;
4390 map_from_cluster = true;
4391 goto got_allocated_blocks;
4394 /* find neighbour allocated blocks */
4395 ar.lleft = map->m_lblk;
4396 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4399 ar.lright = map->m_lblk;
4401 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4405 /* Check if the extent after searching to the right implies a
4406 * cluster we can use. */
4407 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4408 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4409 ar.len = allocated = map->m_len;
4410 newblock = map->m_pblk;
4411 map_from_cluster = true;
4412 goto got_allocated_blocks;
4416 * See if request is beyond maximum number of blocks we can have in
4417 * a single extent. For an initialized extent this limit is
4418 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4419 * EXT_UNWRITTEN_MAX_LEN.
4421 if (map->m_len > EXT_INIT_MAX_LEN &&
4422 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4423 map->m_len = EXT_INIT_MAX_LEN;
4424 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4425 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4426 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4428 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4429 newex.ee_len = cpu_to_le16(map->m_len);
4430 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4432 allocated = ext4_ext_get_actual_len(&newex);
4434 allocated = map->m_len;
4436 /* allocate new block */
4438 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4439 ar.logical = map->m_lblk;
4441 * We calculate the offset from the beginning of the cluster
4442 * for the logical block number, since when we allocate a
4443 * physical cluster, the physical block should start at the
4444 * same offset from the beginning of the cluster. This is
4445 * needed so that future calls to get_implied_cluster_alloc()
4448 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4449 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4451 ar.logical -= offset;
4452 if (S_ISREG(inode->i_mode))
4453 ar.flags = EXT4_MB_HINT_DATA;
4455 /* disable in-core preallocation for non-regular files */
4457 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4458 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4459 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4460 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4461 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4464 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4465 ar.goal, newblock, allocated);
4467 allocated_clusters = ar.len;
4468 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4469 if (ar.len > allocated)
4472 got_allocated_blocks:
4473 /* try to insert new extent into found leaf and return */
4474 ext4_ext_store_pblock(&newex, newblock + offset);
4475 newex.ee_len = cpu_to_le16(ar.len);
4476 /* Mark unwritten */
4477 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4478 ext4_ext_mark_unwritten(&newex);
4479 map->m_flags |= EXT4_MAP_UNWRITTEN;
4481 * io_end structure was created for every IO write to an
4482 * unwritten extent. To avoid unnecessary conversion,
4483 * here we flag the IO that really needs the conversion.
4484 * For non asycn direct IO case, flag the inode state
4485 * that we need to perform conversion when IO is done.
4487 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4492 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4493 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4496 err = ext4_ext_insert_extent(handle, inode, &path,
4499 if (!err && set_unwritten) {
4501 ext4_set_io_unwritten_flag(inode, io);
4503 ext4_set_inode_state(inode,
4504 EXT4_STATE_DIO_UNWRITTEN);
4507 if (err && free_on_err) {
4508 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4509 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4510 /* free data blocks we just allocated */
4511 /* not a good idea to call discard here directly,
4512 * but otherwise we'd need to call it every free() */
4513 ext4_discard_preallocations(inode);
4514 ext4_free_blocks(handle, inode, NULL, newblock,
4515 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4519 /* previous routine could use block we allocated */
4520 newblock = ext4_ext_pblock(&newex);
4521 allocated = ext4_ext_get_actual_len(&newex);
4522 if (allocated > map->m_len)
4523 allocated = map->m_len;
4524 map->m_flags |= EXT4_MAP_NEW;
4527 * Update reserved blocks/metadata blocks after successful
4528 * block allocation which had been deferred till now.
4530 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4531 unsigned int reserved_clusters;
4533 * Check how many clusters we had reserved this allocated range
4535 reserved_clusters = get_reserved_cluster_alloc(inode,
4536 map->m_lblk, allocated);
4537 if (map_from_cluster) {
4538 if (reserved_clusters) {
4540 * We have clusters reserved for this range.
4541 * But since we are not doing actual allocation
4542 * and are simply using blocks from previously
4543 * allocated cluster, we should release the
4544 * reservation and not claim quota.
4546 ext4_da_update_reserve_space(inode,
4547 reserved_clusters, 0);
4550 BUG_ON(allocated_clusters < reserved_clusters);
4551 if (reserved_clusters < allocated_clusters) {
4552 struct ext4_inode_info *ei = EXT4_I(inode);
4553 int reservation = allocated_clusters -
4556 * It seems we claimed few clusters outside of
4557 * the range of this allocation. We should give
4558 * it back to the reservation pool. This can
4559 * happen in the following case:
4561 * * Suppose s_cluster_ratio is 4 (i.e., each
4562 * cluster has 4 blocks. Thus, the clusters
4563 * are [0-3],[4-7],[8-11]...
4564 * * First comes delayed allocation write for
4565 * logical blocks 10 & 11. Since there were no
4566 * previous delayed allocated blocks in the
4567 * range [8-11], we would reserve 1 cluster
4569 * * Next comes write for logical blocks 3 to 8.
4570 * In this case, we will reserve 2 clusters
4571 * (for [0-3] and [4-7]; and not for [8-11] as
4572 * that range has a delayed allocated blocks.
4573 * Thus total reserved clusters now becomes 3.
4574 * * Now, during the delayed allocation writeout
4575 * time, we will first write blocks [3-8] and
4576 * allocate 3 clusters for writing these
4577 * blocks. Also, we would claim all these
4578 * three clusters above.
4579 * * Now when we come here to writeout the
4580 * blocks [10-11], we would expect to claim
4581 * the reservation of 1 cluster we had made
4582 * (and we would claim it since there are no
4583 * more delayed allocated blocks in the range
4584 * [8-11]. But our reserved cluster count had
4585 * already gone to 0.
4587 * Thus, at the step 4 above when we determine
4588 * that there are still some unwritten delayed
4589 * allocated blocks outside of our current
4590 * block range, we should increment the
4591 * reserved clusters count so that when the
4592 * remaining blocks finally gets written, we
4595 dquot_reserve_block(inode,
4596 EXT4_C2B(sbi, reservation));
4597 spin_lock(&ei->i_block_reservation_lock);
4598 ei->i_reserved_data_blocks += reservation;
4599 spin_unlock(&ei->i_block_reservation_lock);
4602 * We will claim quota for all newly allocated blocks.
4603 * We're updating the reserved space *after* the
4604 * correction above so we do not accidentally free
4605 * all the metadata reservation because we might
4606 * actually need it later on.
4608 ext4_da_update_reserve_space(inode, allocated_clusters,
4614 * Cache the extent and update transaction to commit on fdatasync only
4615 * when it is _not_ an unwritten extent.
4617 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4618 ext4_update_inode_fsync_trans(handle, inode, 1);
4620 ext4_update_inode_fsync_trans(handle, inode, 0);
4622 if (allocated > map->m_len)
4623 allocated = map->m_len;
4624 ext4_ext_show_leaf(inode, path);
4625 map->m_flags |= EXT4_MAP_MAPPED;
4626 map->m_pblk = newblock;
4627 map->m_len = allocated;
4629 ext4_ext_drop_refs(path);
4632 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4633 err ? err : allocated);
4634 ext4_es_lru_add(inode);
4635 return err ? err : allocated;
4638 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4640 struct super_block *sb = inode->i_sb;
4641 ext4_lblk_t last_block;
4645 * TODO: optimization is possible here.
4646 * Probably we need not scan at all,
4647 * because page truncation is enough.
4650 /* we have to know where to truncate from in crash case */
4651 EXT4_I(inode)->i_disksize = inode->i_size;
4652 ext4_mark_inode_dirty(handle, inode);
4654 last_block = (inode->i_size + sb->s_blocksize - 1)
4655 >> EXT4_BLOCK_SIZE_BITS(sb);
4657 err = ext4_es_remove_extent(inode, last_block,
4658 EXT_MAX_BLOCKS - last_block);
4659 if (err == -ENOMEM) {
4661 congestion_wait(BLK_RW_ASYNC, HZ/50);
4665 ext4_std_error(inode->i_sb, err);
4668 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4669 ext4_std_error(inode->i_sb, err);
4672 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4673 ext4_lblk_t len, loff_t new_size,
4674 int flags, int mode)
4676 struct inode *inode = file_inode(file);
4681 struct ext4_map_blocks map;
4682 unsigned int credits;
4685 map.m_lblk = offset;
4688 * Don't normalize the request if it can fit in one extent so
4689 * that it doesn't get unnecessarily split into multiple
4692 if (len <= EXT_UNWRITTEN_MAX_LEN)
4693 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4696 * credits to insert 1 extent into extent tree
4698 credits = ext4_chunk_trans_blocks(inode, len);
4701 while (ret >= 0 && len) {
4702 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4704 if (IS_ERR(handle)) {
4705 ret = PTR_ERR(handle);
4708 ret = ext4_map_blocks(handle, inode, &map, flags);
4710 ext4_debug("inode #%lu: block %u: len %u: "
4711 "ext4_ext_map_blocks returned %d",
4712 inode->i_ino, map.m_lblk,
4714 ext4_mark_inode_dirty(handle, inode);
4715 ret2 = ext4_journal_stop(handle);
4719 map.m_len = len = len - ret;
4720 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4721 inode->i_ctime = ext4_current_time(inode);
4723 if (epos > new_size)
4725 if (ext4_update_inode_size(inode, epos) & 0x1)
4726 inode->i_mtime = inode->i_ctime;
4728 if (epos > inode->i_size)
4729 ext4_set_inode_flag(inode,
4730 EXT4_INODE_EOFBLOCKS);
4732 ext4_mark_inode_dirty(handle, inode);
4733 ret2 = ext4_journal_stop(handle);
4737 if (ret == -ENOSPC &&
4738 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4743 return ret > 0 ? ret2 : ret;
4746 static long ext4_zero_range(struct file *file, loff_t offset,
4747 loff_t len, int mode)
4749 struct inode *inode = file_inode(file);
4750 handle_t *handle = NULL;
4751 unsigned int max_blocks;
4752 loff_t new_size = 0;
4756 int partial_begin, partial_end;
4759 struct address_space *mapping = inode->i_mapping;
4760 unsigned int blkbits = inode->i_blkbits;
4762 trace_ext4_zero_range(inode, offset, len, mode);
4764 if (!S_ISREG(inode->i_mode))
4767 /* Call ext4_force_commit to flush all data in case of data=journal. */
4768 if (ext4_should_journal_data(inode)) {
4769 ret = ext4_force_commit(inode->i_sb);
4775 * Write out all dirty pages to avoid race conditions
4776 * Then release them.
4778 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4779 ret = filemap_write_and_wait_range(mapping, offset,
4786 * Round up offset. This is not fallocate, we neet to zero out
4787 * blocks, so convert interior block aligned part of the range to
4788 * unwritten and possibly manually zero out unaligned parts of the
4791 start = round_up(offset, 1 << blkbits);
4792 end = round_down((offset + len), 1 << blkbits);
4794 if (start < offset || end > offset + len)
4796 partial_begin = offset & ((1 << blkbits) - 1);
4797 partial_end = (offset + len) & ((1 << blkbits) - 1);
4799 lblk = start >> blkbits;
4800 max_blocks = (end >> blkbits);
4801 if (max_blocks < lblk)
4806 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
4807 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4809 if (mode & FALLOC_FL_KEEP_SIZE)
4810 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4812 mutex_lock(&inode->i_mutex);
4815 * Indirect files do not support unwritten extnets
4817 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4822 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4823 offset + len > i_size_read(inode)) {
4824 new_size = offset + len;
4825 ret = inode_newsize_ok(inode, new_size);
4829 * If we have a partial block after EOF we have to allocate
4836 if (max_blocks > 0) {
4838 /* Now release the pages and zero block aligned part of pages*/
4839 truncate_pagecache_range(inode, start, end - 1);
4840 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4842 /* Wait all existing dio workers, newcomers will block on i_mutex */
4843 ext4_inode_block_unlocked_dio(inode);
4844 inode_dio_wait(inode);
4846 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4851 * Remove entire range from the extent status tree.
4853 * ext4_es_remove_extent(inode, lblk, max_blocks) is
4854 * NOT sufficient. I'm not sure why this is the case,
4855 * but let's be conservative and remove the extent
4856 * status tree for the entire inode. There should be
4857 * no outstanding delalloc extents thanks to the
4858 * filemap_write_and_wait_range() call above.
4860 ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
4864 if (!partial_begin && !partial_end)
4868 * In worst case we have to writeout two nonadjacent unwritten
4869 * blocks and update the inode
4871 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4872 if (ext4_should_journal_data(inode))
4874 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4875 if (IS_ERR(handle)) {
4876 ret = PTR_ERR(handle);
4877 ext4_std_error(inode->i_sb, ret);
4881 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4883 ext4_update_inode_size(inode, new_size);
4886 * Mark that we allocate beyond EOF so the subsequent truncate
4887 * can proceed even if the new size is the same as i_size.
4889 if ((offset + len) > i_size_read(inode))
4890 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4892 ext4_mark_inode_dirty(handle, inode);
4894 /* Zero out partial block at the edges of the range */
4895 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4897 if (file->f_flags & O_SYNC)
4898 ext4_handle_sync(handle);
4900 ext4_journal_stop(handle);
4902 ext4_inode_resume_unlocked_dio(inode);
4904 mutex_unlock(&inode->i_mutex);
4909 * preallocate space for a file. This implements ext4's fallocate file
4910 * operation, which gets called from sys_fallocate system call.
4911 * For block-mapped files, posix_fallocate should fall back to the method
4912 * of writing zeroes to the required new blocks (the same behavior which is
4913 * expected for file systems which do not support fallocate() system call).
4915 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4917 struct inode *inode = file_inode(file);
4918 loff_t new_size = 0;
4919 unsigned int max_blocks;
4923 unsigned int blkbits = inode->i_blkbits;
4925 /* Return error if mode is not supported */
4926 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4927 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
4930 if (mode & FALLOC_FL_PUNCH_HOLE)
4931 return ext4_punch_hole(inode, offset, len);
4933 ret = ext4_convert_inline_data(inode);
4938 * currently supporting (pre)allocate mode for extent-based
4941 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4944 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4945 return ext4_collapse_range(inode, offset, len);
4947 if (mode & FALLOC_FL_ZERO_RANGE)
4948 return ext4_zero_range(file, offset, len, mode);
4950 trace_ext4_fallocate_enter(inode, offset, len, mode);
4951 lblk = offset >> blkbits;
4953 * We can't just convert len to max_blocks because
4954 * If blocksize = 4096 offset = 3072 and len = 2048
4956 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4959 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4960 if (mode & FALLOC_FL_KEEP_SIZE)
4961 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4963 mutex_lock(&inode->i_mutex);
4965 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4966 offset + len > i_size_read(inode)) {
4967 new_size = offset + len;
4968 ret = inode_newsize_ok(inode, new_size);
4973 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4978 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4979 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4980 EXT4_I(inode)->i_sync_tid);
4983 mutex_unlock(&inode->i_mutex);
4984 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4989 * This function convert a range of blocks to written extents
4990 * The caller of this function will pass the start offset and the size.
4991 * all unwritten extents within this range will be converted to
4994 * This function is called from the direct IO end io call back
4995 * function, to convert the fallocated extents after IO is completed.
4996 * Returns 0 on success.
4998 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4999 loff_t offset, ssize_t len)
5001 unsigned int max_blocks;
5004 struct ext4_map_blocks map;
5005 unsigned int credits, blkbits = inode->i_blkbits;
5007 map.m_lblk = offset >> blkbits;
5009 * We can't just convert len to max_blocks because
5010 * If blocksize = 4096 offset = 3072 and len = 2048
5012 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5015 * This is somewhat ugly but the idea is clear: When transaction is
5016 * reserved, everything goes into it. Otherwise we rather start several
5017 * smaller transactions for conversion of each extent separately.
5020 handle = ext4_journal_start_reserved(handle,
5021 EXT4_HT_EXT_CONVERT);
5023 return PTR_ERR(handle);
5027 * credits to insert 1 extent into extent tree
5029 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5031 while (ret >= 0 && ret < max_blocks) {
5033 map.m_len = (max_blocks -= ret);
5035 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5037 if (IS_ERR(handle)) {
5038 ret = PTR_ERR(handle);
5042 ret = ext4_map_blocks(handle, inode, &map,
5043 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5045 ext4_warning(inode->i_sb,
5046 "inode #%lu: block %u: len %u: "
5047 "ext4_ext_map_blocks returned %d",
5048 inode->i_ino, map.m_lblk,
5050 ext4_mark_inode_dirty(handle, inode);
5052 ret2 = ext4_journal_stop(handle);
5053 if (ret <= 0 || ret2)
5057 ret2 = ext4_journal_stop(handle);
5058 return ret > 0 ? ret2 : ret;
5062 * If newes is not existing extent (newes->ec_pblk equals zero) find
5063 * delayed extent at start of newes and update newes accordingly and
5064 * return start of the next delayed extent.
5066 * If newes is existing extent (newes->ec_pblk is not equal zero)
5067 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5068 * extent found. Leave newes unmodified.
5070 static int ext4_find_delayed_extent(struct inode *inode,
5071 struct extent_status *newes)
5073 struct extent_status es;
5074 ext4_lblk_t block, next_del;
5076 if (newes->es_pblk == 0) {
5077 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5078 newes->es_lblk + newes->es_len - 1, &es);
5081 * No extent in extent-tree contains block @newes->es_pblk,
5082 * then the block may stay in 1)a hole or 2)delayed-extent.
5088 if (es.es_lblk > newes->es_lblk) {
5090 newes->es_len = min(es.es_lblk - newes->es_lblk,
5095 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5098 block = newes->es_lblk + newes->es_len;
5099 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5101 next_del = EXT_MAX_BLOCKS;
5103 next_del = es.es_lblk;
5107 /* fiemap flags we can handle specified here */
5108 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5110 static int ext4_xattr_fiemap(struct inode *inode,
5111 struct fiemap_extent_info *fieinfo)
5115 __u32 flags = FIEMAP_EXTENT_LAST;
5116 int blockbits = inode->i_sb->s_blocksize_bits;
5120 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5121 struct ext4_iloc iloc;
5122 int offset; /* offset of xattr in inode */
5124 error = ext4_get_inode_loc(inode, &iloc);
5127 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5128 offset = EXT4_GOOD_OLD_INODE_SIZE +
5129 EXT4_I(inode)->i_extra_isize;
5131 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5132 flags |= FIEMAP_EXTENT_DATA_INLINE;
5134 } else { /* external block */
5135 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5136 length = inode->i_sb->s_blocksize;
5140 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5142 return (error < 0 ? error : 0);
5145 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5146 __u64 start, __u64 len)
5148 ext4_lblk_t start_blk;
5151 if (ext4_has_inline_data(inode)) {
5154 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline);
5160 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5161 error = ext4_ext_precache(inode);
5166 /* fallback to generic here if not in extents fmt */
5167 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5168 return generic_block_fiemap(inode, fieinfo, start, len,
5171 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5174 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5175 error = ext4_xattr_fiemap(inode, fieinfo);
5177 ext4_lblk_t len_blks;
5180 start_blk = start >> inode->i_sb->s_blocksize_bits;
5181 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5182 if (last_blk >= EXT_MAX_BLOCKS)
5183 last_blk = EXT_MAX_BLOCKS-1;
5184 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5187 * Walk the extent tree gathering extent information
5188 * and pushing extents back to the user.
5190 error = ext4_fill_fiemap_extents(inode, start_blk,
5193 ext4_es_lru_add(inode);
5199 * Function to access the path buffer for marking it dirty.
5200 * It also checks if there are sufficient credits left in the journal handle
5204 ext4_access_path(handle_t *handle, struct inode *inode,
5205 struct ext4_ext_path *path)
5209 if (!ext4_handle_valid(handle))
5213 * Check if need to extend journal credits
5214 * 3 for leaf, sb, and inode plus 2 (bmap and group
5215 * descriptor) for each block group; assume two block
5218 if (handle->h_buffer_credits < 7) {
5219 credits = ext4_writepage_trans_blocks(inode);
5220 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5221 /* EAGAIN is success */
5222 if (err && err != -EAGAIN)
5226 err = ext4_ext_get_access(handle, inode, path);
5231 * ext4_ext_shift_path_extents:
5232 * Shift the extents of a path structure lying between path[depth].p_ext
5233 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5234 * from starting block for each extent.
5237 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5238 struct inode *inode, handle_t *handle,
5242 struct ext4_extent *ex_start, *ex_last;
5244 depth = path->p_depth;
5246 while (depth >= 0) {
5247 if (depth == path->p_depth) {
5248 ex_start = path[depth].p_ext;
5252 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5256 err = ext4_access_path(handle, inode, path + depth);
5260 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5263 *start = le32_to_cpu(ex_last->ee_block) +
5264 ext4_ext_get_actual_len(ex_last);
5266 while (ex_start <= ex_last) {
5267 le32_add_cpu(&ex_start->ee_block, -shift);
5268 /* Try to merge to the left. */
5270 EXT_FIRST_EXTENT(path[depth].p_hdr)) &&
5271 ext4_ext_try_to_merge_right(inode,
5272 path, ex_start - 1))
5277 err = ext4_ext_dirty(handle, inode, path + depth);
5281 if (--depth < 0 || !update)
5285 /* Update index too */
5286 err = ext4_access_path(handle, inode, path + depth);
5290 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5291 err = ext4_ext_dirty(handle, inode, path + depth);
5295 /* we are done if current index is not a starting index */
5296 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5307 * ext4_ext_shift_extents:
5308 * All the extents which lies in the range from start to the last allocated
5309 * block for the file are shifted downwards by shift blocks.
5310 * On success, 0 is returned, error otherwise.
5313 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5314 ext4_lblk_t start, ext4_lblk_t shift)
5316 struct ext4_ext_path *path;
5318 struct ext4_extent *extent;
5319 ext4_lblk_t stop_block;
5320 ext4_lblk_t ex_start, ex_end;
5322 /* Let path point to the last extent */
5323 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5325 return PTR_ERR(path);
5327 depth = path->p_depth;
5328 extent = path[depth].p_ext;
5332 stop_block = le32_to_cpu(extent->ee_block) +
5333 ext4_ext_get_actual_len(extent);
5335 /* Nothing to shift, if hole is at the end of file */
5336 if (start >= stop_block)
5340 * Don't start shifting extents until we make sure the hole is big
5341 * enough to accomodate the shift.
5343 path = ext4_find_extent(inode, start - 1, &path, 0);
5345 return PTR_ERR(path);
5346 depth = path->p_depth;
5347 extent = path[depth].p_ext;
5349 ex_start = le32_to_cpu(extent->ee_block);
5350 ex_end = le32_to_cpu(extent->ee_block) +
5351 ext4_ext_get_actual_len(extent);
5357 if ((start == ex_start && shift > ex_start) ||
5358 (shift > start - ex_end))
5361 /* Its safe to start updating extents */
5362 while (start < stop_block) {
5363 path = ext4_find_extent(inode, start, &path, 0);
5365 return PTR_ERR(path);
5366 depth = path->p_depth;
5367 extent = path[depth].p_ext;
5369 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5370 (unsigned long) start);
5373 if (start > le32_to_cpu(extent->ee_block)) {
5374 /* Hole, move to the next extent */
5375 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5376 path[depth].p_ext++;
5378 start = ext4_ext_next_allocated_block(path);
5382 ret = ext4_ext_shift_path_extents(path, shift, inode,
5388 ext4_ext_drop_refs(path);
5394 * ext4_collapse_range:
5395 * This implements the fallocate's collapse range functionality for ext4
5396 * Returns: 0 and non-zero on error.
5398 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5400 struct super_block *sb = inode->i_sb;
5401 ext4_lblk_t punch_start, punch_stop;
5403 unsigned int credits;
5404 loff_t new_size, ioffset;
5407 /* Collapse range works only on fs block size aligned offsets. */
5408 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5409 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5412 if (!S_ISREG(inode->i_mode))
5415 trace_ext4_collapse_range(inode, offset, len);
5417 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5418 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5420 /* Call ext4_force_commit to flush all data in case of data=journal. */
5421 if (ext4_should_journal_data(inode)) {
5422 ret = ext4_force_commit(inode->i_sb);
5428 * Need to round down offset to be aligned with page size boundary
5429 * for page size > block size.
5431 ioffset = round_down(offset, PAGE_SIZE);
5433 /* Write out all dirty pages */
5434 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5439 /* Take mutex lock */
5440 mutex_lock(&inode->i_mutex);
5443 * There is no need to overlap collapse range with EOF, in which case
5444 * it is effectively a truncate operation
5446 if (offset + len >= i_size_read(inode)) {
5451 /* Currently just for extent based files */
5452 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5457 truncate_pagecache(inode, ioffset);
5459 /* Wait for existing dio to complete */
5460 ext4_inode_block_unlocked_dio(inode);
5461 inode_dio_wait(inode);
5463 credits = ext4_writepage_trans_blocks(inode);
5464 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5465 if (IS_ERR(handle)) {
5466 ret = PTR_ERR(handle);
5470 down_write(&EXT4_I(inode)->i_data_sem);
5471 ext4_discard_preallocations(inode);
5473 ret = ext4_es_remove_extent(inode, punch_start,
5474 EXT_MAX_BLOCKS - punch_start);
5476 up_write(&EXT4_I(inode)->i_data_sem);
5480 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5482 up_write(&EXT4_I(inode)->i_data_sem);
5485 ext4_discard_preallocations(inode);
5487 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5488 punch_stop - punch_start);
5490 up_write(&EXT4_I(inode)->i_data_sem);
5494 new_size = i_size_read(inode) - len;
5495 i_size_write(inode, new_size);
5496 EXT4_I(inode)->i_disksize = new_size;
5498 up_write(&EXT4_I(inode)->i_data_sem);
5500 ext4_handle_sync(handle);
5501 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5502 ext4_mark_inode_dirty(handle, inode);
5505 ext4_journal_stop(handle);
5507 ext4_inode_resume_unlocked_dio(inode);
5509 mutex_unlock(&inode->i_mutex);
5514 * ext4_swap_extents - Swap extents between two inodes
5516 * @inode1: First inode
5517 * @inode2: Second inode
5518 * @lblk1: Start block for first inode
5519 * @lblk2: Start block for second inode
5520 * @count: Number of blocks to swap
5521 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5522 * @erp: Pointer to save error value
5524 * This helper routine does exactly what is promise "swap extents". All other
5525 * stuff such as page-cache locking consistency, bh mapping consistency or
5526 * extent's data copying must be performed by caller.
5528 * i_mutex is held for both inodes
5529 * i_data_sem is locked for write for both inodes
5531 * All pages from requested range are locked for both inodes
5534 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5535 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5536 ext4_lblk_t count, int unwritten, int *erp)
5538 struct ext4_ext_path *path1 = NULL;
5539 struct ext4_ext_path *path2 = NULL;
5540 int replaced_count = 0;
5542 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5543 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5544 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5545 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5547 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5550 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5555 struct ext4_extent *ex1, *ex2, tmp_ex;
5556 ext4_lblk_t e1_blk, e2_blk;
5557 int e1_len, e2_len, len;
5560 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5561 if (unlikely(IS_ERR(path1))) {
5562 *erp = PTR_ERR(path1);
5568 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5569 if (unlikely(IS_ERR(path2))) {
5570 *erp = PTR_ERR(path2);
5574 ex1 = path1[path1->p_depth].p_ext;
5575 ex2 = path2[path2->p_depth].p_ext;
5576 /* Do we have somthing to swap ? */
5577 if (unlikely(!ex2 || !ex1))
5580 e1_blk = le32_to_cpu(ex1->ee_block);
5581 e2_blk = le32_to_cpu(ex2->ee_block);
5582 e1_len = ext4_ext_get_actual_len(ex1);
5583 e2_len = ext4_ext_get_actual_len(ex2);
5586 if (!in_range(lblk1, e1_blk, e1_len) ||
5587 !in_range(lblk2, e2_blk, e2_len)) {
5588 ext4_lblk_t next1, next2;
5590 /* if hole after extent, then go to next extent */
5591 next1 = ext4_ext_next_allocated_block(path1);
5592 next2 = ext4_ext_next_allocated_block(path2);
5593 /* If hole before extent, then shift to that extent */
5598 /* Do we have something to swap */
5599 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5601 /* Move to the rightest boundary */
5602 len = next1 - lblk1;
5603 if (len < next2 - lblk2)
5604 len = next2 - lblk2;
5613 /* Prepare left boundary */
5614 if (e1_blk < lblk1) {
5616 *erp = ext4_force_split_extent_at(handle, inode1,
5621 if (e2_blk < lblk2) {
5623 *erp = ext4_force_split_extent_at(handle, inode2,
5628 /* ext4_split_extent_at() may result in leaf extent split,
5629 * path must to be revalidated. */
5633 /* Prepare right boundary */
5635 if (len > e1_blk + e1_len - lblk1)
5636 len = e1_blk + e1_len - lblk1;
5637 if (len > e2_blk + e2_len - lblk2)
5638 len = e2_blk + e2_len - lblk2;
5640 if (len != e1_len) {
5642 *erp = ext4_force_split_extent_at(handle, inode1,
5643 &path1, lblk1 + len, 0);
5647 if (len != e2_len) {
5649 *erp = ext4_force_split_extent_at(handle, inode2,
5650 &path2, lblk2 + len, 0);
5654 /* ext4_split_extent_at() may result in leaf extent split,
5655 * path must to be revalidated. */
5659 BUG_ON(e2_len != e1_len);
5660 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5663 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5667 /* Both extents are fully inside boundaries. Swap it now */
5669 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5670 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5671 ex1->ee_len = cpu_to_le16(e2_len);
5672 ex2->ee_len = cpu_to_le16(e1_len);
5674 ext4_ext_mark_unwritten(ex2);
5675 if (ext4_ext_is_unwritten(&tmp_ex))
5676 ext4_ext_mark_unwritten(ex1);
5678 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5679 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5680 *erp = ext4_ext_dirty(handle, inode2, path2 +
5684 *erp = ext4_ext_dirty(handle, inode1, path1 +
5687 * Looks scarry ah..? second inode already points to new blocks,
5688 * and it was successfully dirtied. But luckily error may happen
5689 * only due to journal error, so full transaction will be
5696 replaced_count += len;
5700 ext4_ext_drop_refs(path1);
5702 ext4_ext_drop_refs(path2);
5704 path1 = path2 = NULL;
5706 return replaced_count;