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 <linux/backing-dev.h>
43 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
50 * used by extent splitting.
52 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
54 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
55 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
57 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
60 static __le32 ext4_extent_block_csum(struct inode *inode,
61 struct ext4_extent_header *eh)
63 struct ext4_inode_info *ei = EXT4_I(inode);
64 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
67 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
68 EXT4_EXTENT_TAIL_OFFSET(eh));
69 return cpu_to_le32(csum);
72 static int ext4_extent_block_csum_verify(struct inode *inode,
73 struct ext4_extent_header *eh)
75 struct ext4_extent_tail *et;
77 if (!ext4_has_metadata_csum(inode->i_sb))
80 et = find_ext4_extent_tail(eh);
81 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
86 static void ext4_extent_block_csum_set(struct inode *inode,
87 struct ext4_extent_header *eh)
89 struct ext4_extent_tail *et;
91 if (!ext4_has_metadata_csum(inode->i_sb))
94 et = find_ext4_extent_tail(eh);
95 et->et_checksum = ext4_extent_block_csum(inode, eh);
98 static int ext4_split_extent(handle_t *handle,
100 struct ext4_ext_path **ppath,
101 struct ext4_map_blocks *map,
105 static int ext4_split_extent_at(handle_t *handle,
107 struct ext4_ext_path **ppath,
112 static int ext4_find_delayed_extent(struct inode *inode,
113 struct extent_status *newes);
115 static int ext4_ext_truncate_extend_restart(handle_t *handle,
121 if (!ext4_handle_valid(handle))
123 if (handle->h_buffer_credits > needed)
125 err = ext4_journal_extend(handle, needed);
128 err = ext4_truncate_restart_trans(handle, inode, needed);
140 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
141 struct ext4_ext_path *path)
144 /* path points to block */
145 BUFFER_TRACE(path->p_bh, "get_write_access");
146 return ext4_journal_get_write_access(handle, path->p_bh);
148 /* path points to leaf/index in inode body */
149 /* we use in-core data, no need to protect them */
159 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
160 struct inode *inode, struct ext4_ext_path *path)
164 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
166 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
167 /* path points to block */
168 err = __ext4_handle_dirty_metadata(where, line, handle,
171 /* path points to leaf/index in inode body */
172 err = ext4_mark_inode_dirty(handle, inode);
177 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
178 struct ext4_ext_path *path,
182 int depth = path->p_depth;
183 struct ext4_extent *ex;
186 * Try to predict block placement assuming that we are
187 * filling in a file which will eventually be
188 * non-sparse --- i.e., in the case of libbfd writing
189 * an ELF object sections out-of-order but in a way
190 * the eventually results in a contiguous object or
191 * executable file, or some database extending a table
192 * space file. However, this is actually somewhat
193 * non-ideal if we are writing a sparse file such as
194 * qemu or KVM writing a raw image file that is going
195 * to stay fairly sparse, since it will end up
196 * fragmenting the file system's free space. Maybe we
197 * should have some hueristics or some way to allow
198 * userspace to pass a hint to file system,
199 * especially if the latter case turns out to be
202 ex = path[depth].p_ext;
204 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
205 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
207 if (block > ext_block)
208 return ext_pblk + (block - ext_block);
210 return ext_pblk - (ext_block - block);
213 /* it looks like index is empty;
214 * try to find starting block from index itself */
215 if (path[depth].p_bh)
216 return path[depth].p_bh->b_blocknr;
219 /* OK. use inode's group */
220 return ext4_inode_to_goal_block(inode);
224 * Allocation for a meta data block
227 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
228 struct ext4_ext_path *path,
229 struct ext4_extent *ex, int *err, unsigned int flags)
231 ext4_fsblk_t goal, newblock;
233 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
234 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
239 static inline int ext4_ext_space_block(struct inode *inode, int check)
243 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
244 / sizeof(struct ext4_extent);
245 #ifdef AGGRESSIVE_TEST
246 if (!check && size > 6)
252 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
256 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
257 / sizeof(struct ext4_extent_idx);
258 #ifdef AGGRESSIVE_TEST
259 if (!check && size > 5)
265 static inline int ext4_ext_space_root(struct inode *inode, int check)
269 size = sizeof(EXT4_I(inode)->i_data);
270 size -= sizeof(struct ext4_extent_header);
271 size /= sizeof(struct ext4_extent);
272 #ifdef AGGRESSIVE_TEST
273 if (!check && size > 3)
279 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
283 size = sizeof(EXT4_I(inode)->i_data);
284 size -= sizeof(struct ext4_extent_header);
285 size /= sizeof(struct ext4_extent_idx);
286 #ifdef AGGRESSIVE_TEST
287 if (!check && size > 4)
294 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
295 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
298 struct ext4_ext_path *path = *ppath;
299 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
301 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
302 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
303 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
304 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
308 * Calculate the number of metadata blocks needed
309 * to allocate @blocks
310 * Worse case is one block per extent
312 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
314 struct ext4_inode_info *ei = EXT4_I(inode);
317 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
318 / sizeof(struct ext4_extent_idx));
321 * If the new delayed allocation block is contiguous with the
322 * previous da block, it can share index blocks with the
323 * previous block, so we only need to allocate a new index
324 * block every idxs leaf blocks. At ldxs**2 blocks, we need
325 * an additional index block, and at ldxs**3 blocks, yet
326 * another index blocks.
328 if (ei->i_da_metadata_calc_len &&
329 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
332 if ((ei->i_da_metadata_calc_len % idxs) == 0)
334 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
336 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
338 ei->i_da_metadata_calc_len = 0;
340 ei->i_da_metadata_calc_len++;
341 ei->i_da_metadata_calc_last_lblock++;
346 * In the worst case we need a new set of index blocks at
347 * every level of the inode's extent tree.
349 ei->i_da_metadata_calc_len = 1;
350 ei->i_da_metadata_calc_last_lblock = lblock;
351 return ext_depth(inode) + 1;
355 ext4_ext_max_entries(struct inode *inode, int depth)
359 if (depth == ext_depth(inode)) {
361 max = ext4_ext_space_root(inode, 1);
363 max = ext4_ext_space_root_idx(inode, 1);
366 max = ext4_ext_space_block(inode, 1);
368 max = ext4_ext_space_block_idx(inode, 1);
374 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
376 ext4_fsblk_t block = ext4_ext_pblock(ext);
377 int len = ext4_ext_get_actual_len(ext);
378 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
379 ext4_lblk_t last = lblock + len - 1;
381 if (len == 0 || lblock > last)
383 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
386 static int ext4_valid_extent_idx(struct inode *inode,
387 struct ext4_extent_idx *ext_idx)
389 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
391 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
394 static int ext4_valid_extent_entries(struct inode *inode,
395 struct ext4_extent_header *eh,
398 unsigned short entries;
399 if (eh->eh_entries == 0)
402 entries = le16_to_cpu(eh->eh_entries);
406 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
407 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
408 ext4_fsblk_t pblock = 0;
409 ext4_lblk_t lblock = 0;
410 ext4_lblk_t prev = 0;
413 if (!ext4_valid_extent(inode, ext))
416 /* Check for overlapping extents */
417 lblock = le32_to_cpu(ext->ee_block);
418 len = ext4_ext_get_actual_len(ext);
419 if ((lblock <= prev) && prev) {
420 pblock = ext4_ext_pblock(ext);
421 es->s_last_error_block = cpu_to_le64(pblock);
426 prev = lblock + len - 1;
429 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
431 if (!ext4_valid_extent_idx(inode, ext_idx))
440 static int __ext4_ext_check(const char *function, unsigned int line,
441 struct inode *inode, struct ext4_extent_header *eh,
442 int depth, ext4_fsblk_t pblk)
444 const char *error_msg;
445 int max = 0, err = -EFSCORRUPTED;
447 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
448 error_msg = "invalid magic";
451 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
452 error_msg = "unexpected eh_depth";
455 if (unlikely(eh->eh_max == 0)) {
456 error_msg = "invalid eh_max";
459 max = ext4_ext_max_entries(inode, depth);
460 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
461 error_msg = "too large eh_max";
464 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
465 error_msg = "invalid eh_entries";
468 if (!ext4_valid_extent_entries(inode, eh, depth)) {
469 error_msg = "invalid extent entries";
472 if (unlikely(depth > 32)) {
473 error_msg = "too large eh_depth";
476 /* Verify checksum on non-root extent tree nodes */
477 if (ext_depth(inode) != depth &&
478 !ext4_extent_block_csum_verify(inode, eh)) {
479 error_msg = "extent tree corrupted";
486 ext4_error_inode(inode, function, line, 0,
487 "pblk %llu bad header/extent: %s - magic %x, "
488 "entries %u, max %u(%u), depth %u(%u)",
489 (unsigned long long) pblk, error_msg,
490 le16_to_cpu(eh->eh_magic),
491 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
492 max, le16_to_cpu(eh->eh_depth), depth);
496 #define ext4_ext_check(inode, eh, depth, pblk) \
497 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
499 int ext4_ext_check_inode(struct inode *inode)
501 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
504 static struct buffer_head *
505 __read_extent_tree_block(const char *function, unsigned int line,
506 struct inode *inode, ext4_fsblk_t pblk, int depth,
509 struct buffer_head *bh;
512 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
514 return ERR_PTR(-ENOMEM);
516 if (!bh_uptodate_or_lock(bh)) {
517 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
518 err = bh_submit_read(bh);
522 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
524 err = __ext4_ext_check(function, line, inode,
525 ext_block_hdr(bh), depth, pblk);
528 set_buffer_verified(bh);
530 * If this is a leaf block, cache all of its entries
532 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
533 struct ext4_extent_header *eh = ext_block_hdr(bh);
534 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
535 ext4_lblk_t prev = 0;
538 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
539 unsigned int status = EXTENT_STATUS_WRITTEN;
540 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
541 int len = ext4_ext_get_actual_len(ex);
543 if (prev && (prev != lblk))
544 ext4_es_cache_extent(inode, prev,
548 if (ext4_ext_is_unwritten(ex))
549 status = EXTENT_STATUS_UNWRITTEN;
550 ext4_es_cache_extent(inode, lblk, len,
551 ext4_ext_pblock(ex), status);
562 #define read_extent_tree_block(inode, pblk, depth, flags) \
563 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
567 * This function is called to cache a file's extent information in the
570 int ext4_ext_precache(struct inode *inode)
572 struct ext4_inode_info *ei = EXT4_I(inode);
573 struct ext4_ext_path *path = NULL;
574 struct buffer_head *bh;
575 int i = 0, depth, ret = 0;
577 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
578 return 0; /* not an extent-mapped inode */
580 down_read(&ei->i_data_sem);
581 depth = ext_depth(inode);
583 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
586 up_read(&ei->i_data_sem);
590 /* Don't cache anything if there are no external extent blocks */
593 path[0].p_hdr = ext_inode_hdr(inode);
594 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
597 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
600 * If this is a leaf block or we've reached the end of
601 * the index block, go up
604 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
605 brelse(path[i].p_bh);
610 bh = read_extent_tree_block(inode,
611 ext4_idx_pblock(path[i].p_idx++),
613 EXT4_EX_FORCE_CACHE);
620 path[i].p_hdr = ext_block_hdr(bh);
621 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
623 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
625 up_read(&ei->i_data_sem);
626 ext4_ext_drop_refs(path);
632 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
634 int k, l = path->p_depth;
637 for (k = 0; k <= l; k++, path++) {
639 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
640 ext4_idx_pblock(path->p_idx));
641 } else if (path->p_ext) {
642 ext_debug(" %d:[%d]%d:%llu ",
643 le32_to_cpu(path->p_ext->ee_block),
644 ext4_ext_is_unwritten(path->p_ext),
645 ext4_ext_get_actual_len(path->p_ext),
646 ext4_ext_pblock(path->p_ext));
653 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
655 int depth = ext_depth(inode);
656 struct ext4_extent_header *eh;
657 struct ext4_extent *ex;
663 eh = path[depth].p_hdr;
664 ex = EXT_FIRST_EXTENT(eh);
666 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
668 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
669 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
670 ext4_ext_is_unwritten(ex),
671 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
676 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
677 ext4_fsblk_t newblock, int level)
679 int depth = ext_depth(inode);
680 struct ext4_extent *ex;
682 if (depth != level) {
683 struct ext4_extent_idx *idx;
684 idx = path[level].p_idx;
685 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
686 ext_debug("%d: move %d:%llu in new index %llu\n", level,
687 le32_to_cpu(idx->ei_block),
688 ext4_idx_pblock(idx),
696 ex = path[depth].p_ext;
697 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
698 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
699 le32_to_cpu(ex->ee_block),
701 ext4_ext_is_unwritten(ex),
702 ext4_ext_get_actual_len(ex),
709 #define ext4_ext_show_path(inode, path)
710 #define ext4_ext_show_leaf(inode, path)
711 #define ext4_ext_show_move(inode, path, newblock, level)
714 void ext4_ext_drop_refs(struct ext4_ext_path *path)
720 depth = path->p_depth;
721 for (i = 0; i <= depth; i++, path++)
729 * ext4_ext_binsearch_idx:
730 * binary search for the closest index of the given block
731 * the header must be checked before calling this
734 ext4_ext_binsearch_idx(struct inode *inode,
735 struct ext4_ext_path *path, ext4_lblk_t block)
737 struct ext4_extent_header *eh = path->p_hdr;
738 struct ext4_extent_idx *r, *l, *m;
741 ext_debug("binsearch for %u(idx): ", block);
743 l = EXT_FIRST_INDEX(eh) + 1;
744 r = EXT_LAST_INDEX(eh);
747 if (block < le32_to_cpu(m->ei_block))
751 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
752 m, le32_to_cpu(m->ei_block),
753 r, le32_to_cpu(r->ei_block));
757 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
758 ext4_idx_pblock(path->p_idx));
760 #ifdef CHECK_BINSEARCH
762 struct ext4_extent_idx *chix, *ix;
765 chix = ix = EXT_FIRST_INDEX(eh);
766 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
768 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
769 printk(KERN_DEBUG "k=%d, ix=0x%p, "
771 ix, EXT_FIRST_INDEX(eh));
772 printk(KERN_DEBUG "%u <= %u\n",
773 le32_to_cpu(ix->ei_block),
774 le32_to_cpu(ix[-1].ei_block));
776 BUG_ON(k && le32_to_cpu(ix->ei_block)
777 <= le32_to_cpu(ix[-1].ei_block));
778 if (block < le32_to_cpu(ix->ei_block))
782 BUG_ON(chix != path->p_idx);
789 * ext4_ext_binsearch:
790 * binary search for closest extent of the given block
791 * the header must be checked before calling this
794 ext4_ext_binsearch(struct inode *inode,
795 struct ext4_ext_path *path, ext4_lblk_t block)
797 struct ext4_extent_header *eh = path->p_hdr;
798 struct ext4_extent *r, *l, *m;
800 if (eh->eh_entries == 0) {
802 * this leaf is empty:
803 * we get such a leaf in split/add case
808 ext_debug("binsearch for %u: ", block);
810 l = EXT_FIRST_EXTENT(eh) + 1;
811 r = EXT_LAST_EXTENT(eh);
815 if (block < le32_to_cpu(m->ee_block))
819 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
820 m, le32_to_cpu(m->ee_block),
821 r, le32_to_cpu(r->ee_block));
825 ext_debug(" -> %d:%llu:[%d]%d ",
826 le32_to_cpu(path->p_ext->ee_block),
827 ext4_ext_pblock(path->p_ext),
828 ext4_ext_is_unwritten(path->p_ext),
829 ext4_ext_get_actual_len(path->p_ext));
831 #ifdef CHECK_BINSEARCH
833 struct ext4_extent *chex, *ex;
836 chex = ex = EXT_FIRST_EXTENT(eh);
837 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
838 BUG_ON(k && le32_to_cpu(ex->ee_block)
839 <= le32_to_cpu(ex[-1].ee_block));
840 if (block < le32_to_cpu(ex->ee_block))
844 BUG_ON(chex != path->p_ext);
850 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
852 struct ext4_extent_header *eh;
854 eh = ext_inode_hdr(inode);
857 eh->eh_magic = EXT4_EXT_MAGIC;
858 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
859 ext4_mark_inode_dirty(handle, inode);
863 struct ext4_ext_path *
864 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
865 struct ext4_ext_path **orig_path, int flags)
867 struct ext4_extent_header *eh;
868 struct buffer_head *bh;
869 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
870 short int depth, i, ppos = 0;
873 eh = ext_inode_hdr(inode);
874 depth = ext_depth(inode);
877 ext4_ext_drop_refs(path);
878 if (depth > path[0].p_maxdepth) {
880 *orig_path = path = NULL;
884 /* account possible depth increase */
885 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
888 return ERR_PTR(-ENOMEM);
889 path[0].p_maxdepth = depth + 1;
895 /* walk through the tree */
897 ext_debug("depth %d: num %d, max %d\n",
898 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
900 ext4_ext_binsearch_idx(inode, path + ppos, block);
901 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
902 path[ppos].p_depth = i;
903 path[ppos].p_ext = NULL;
905 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
912 eh = ext_block_hdr(bh);
914 if (unlikely(ppos > depth)) {
916 EXT4_ERROR_INODE(inode,
917 "ppos %d > depth %d", ppos, depth);
921 path[ppos].p_bh = bh;
922 path[ppos].p_hdr = eh;
925 path[ppos].p_depth = i;
926 path[ppos].p_ext = NULL;
927 path[ppos].p_idx = NULL;
930 ext4_ext_binsearch(inode, path + ppos, block);
931 /* if not an empty leaf */
932 if (path[ppos].p_ext)
933 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
935 ext4_ext_show_path(inode, path);
940 ext4_ext_drop_refs(path);
948 * ext4_ext_insert_index:
949 * insert new index [@logical;@ptr] into the block at @curp;
950 * check where to insert: before @curp or after @curp
952 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
953 struct ext4_ext_path *curp,
954 int logical, ext4_fsblk_t ptr)
956 struct ext4_extent_idx *ix;
959 err = ext4_ext_get_access(handle, inode, curp);
963 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
964 EXT4_ERROR_INODE(inode,
965 "logical %d == ei_block %d!",
966 logical, le32_to_cpu(curp->p_idx->ei_block));
967 return -EFSCORRUPTED;
970 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
971 >= le16_to_cpu(curp->p_hdr->eh_max))) {
972 EXT4_ERROR_INODE(inode,
973 "eh_entries %d >= eh_max %d!",
974 le16_to_cpu(curp->p_hdr->eh_entries),
975 le16_to_cpu(curp->p_hdr->eh_max));
976 return -EFSCORRUPTED;
979 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
981 ext_debug("insert new index %d after: %llu\n", logical, ptr);
982 ix = curp->p_idx + 1;
985 ext_debug("insert new index %d before: %llu\n", logical, ptr);
989 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
992 ext_debug("insert new index %d: "
993 "move %d indices from 0x%p to 0x%p\n",
994 logical, len, ix, ix + 1);
995 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
998 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
999 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1000 return -EFSCORRUPTED;
1003 ix->ei_block = cpu_to_le32(logical);
1004 ext4_idx_store_pblock(ix, ptr);
1005 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1007 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1008 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1009 return -EFSCORRUPTED;
1012 err = ext4_ext_dirty(handle, inode, curp);
1013 ext4_std_error(inode->i_sb, err);
1020 * inserts new subtree into the path, using free index entry
1022 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1023 * - makes decision where to split
1024 * - moves remaining extents and index entries (right to the split point)
1025 * into the newly allocated blocks
1026 * - initializes subtree
1028 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1030 struct ext4_ext_path *path,
1031 struct ext4_extent *newext, int at)
1033 struct buffer_head *bh = NULL;
1034 int depth = ext_depth(inode);
1035 struct ext4_extent_header *neh;
1036 struct ext4_extent_idx *fidx;
1037 int i = at, k, m, a;
1038 ext4_fsblk_t newblock, oldblock;
1040 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1043 /* make decision: where to split? */
1044 /* FIXME: now decision is simplest: at current extent */
1046 /* if current leaf will be split, then we should use
1047 * border from split point */
1048 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1049 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1050 return -EFSCORRUPTED;
1052 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1053 border = path[depth].p_ext[1].ee_block;
1054 ext_debug("leaf will be split."
1055 " next leaf starts at %d\n",
1056 le32_to_cpu(border));
1058 border = newext->ee_block;
1059 ext_debug("leaf will be added."
1060 " next leaf starts at %d\n",
1061 le32_to_cpu(border));
1065 * If error occurs, then we break processing
1066 * and mark filesystem read-only. index won't
1067 * be inserted and tree will be in consistent
1068 * state. Next mount will repair buffers too.
1072 * Get array to track all allocated blocks.
1073 * We need this to handle errors and free blocks
1076 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
1080 /* allocate all needed blocks */
1081 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1082 for (a = 0; a < depth - at; a++) {
1083 newblock = ext4_ext_new_meta_block(handle, inode, path,
1084 newext, &err, flags);
1087 ablocks[a] = newblock;
1090 /* initialize new leaf */
1091 newblock = ablocks[--a];
1092 if (unlikely(newblock == 0)) {
1093 EXT4_ERROR_INODE(inode, "newblock == 0!");
1094 err = -EFSCORRUPTED;
1097 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1098 if (unlikely(!bh)) {
1104 err = ext4_journal_get_create_access(handle, bh);
1108 neh = ext_block_hdr(bh);
1109 neh->eh_entries = 0;
1110 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1111 neh->eh_magic = EXT4_EXT_MAGIC;
1114 /* move remainder of path[depth] to the new leaf */
1115 if (unlikely(path[depth].p_hdr->eh_entries !=
1116 path[depth].p_hdr->eh_max)) {
1117 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1118 path[depth].p_hdr->eh_entries,
1119 path[depth].p_hdr->eh_max);
1120 err = -EFSCORRUPTED;
1123 /* start copy from next extent */
1124 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1125 ext4_ext_show_move(inode, path, newblock, depth);
1127 struct ext4_extent *ex;
1128 ex = EXT_FIRST_EXTENT(neh);
1129 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1130 le16_add_cpu(&neh->eh_entries, m);
1133 ext4_extent_block_csum_set(inode, neh);
1134 set_buffer_uptodate(bh);
1137 err = ext4_handle_dirty_metadata(handle, inode, bh);
1143 /* correct old leaf */
1145 err = ext4_ext_get_access(handle, inode, path + depth);
1148 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1149 err = ext4_ext_dirty(handle, inode, path + depth);
1155 /* create intermediate indexes */
1157 if (unlikely(k < 0)) {
1158 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1159 err = -EFSCORRUPTED;
1163 ext_debug("create %d intermediate indices\n", k);
1164 /* insert new index into current index block */
1165 /* current depth stored in i var */
1168 oldblock = newblock;
1169 newblock = ablocks[--a];
1170 bh = sb_getblk(inode->i_sb, newblock);
1171 if (unlikely(!bh)) {
1177 err = ext4_journal_get_create_access(handle, bh);
1181 neh = ext_block_hdr(bh);
1182 neh->eh_entries = cpu_to_le16(1);
1183 neh->eh_magic = EXT4_EXT_MAGIC;
1184 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1185 neh->eh_depth = cpu_to_le16(depth - i);
1186 fidx = EXT_FIRST_INDEX(neh);
1187 fidx->ei_block = border;
1188 ext4_idx_store_pblock(fidx, oldblock);
1190 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1191 i, newblock, le32_to_cpu(border), oldblock);
1193 /* move remainder of path[i] to the new index block */
1194 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1195 EXT_LAST_INDEX(path[i].p_hdr))) {
1196 EXT4_ERROR_INODE(inode,
1197 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1198 le32_to_cpu(path[i].p_ext->ee_block));
1199 err = -EFSCORRUPTED;
1202 /* start copy indexes */
1203 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1204 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1205 EXT_MAX_INDEX(path[i].p_hdr));
1206 ext4_ext_show_move(inode, path, newblock, i);
1208 memmove(++fidx, path[i].p_idx,
1209 sizeof(struct ext4_extent_idx) * m);
1210 le16_add_cpu(&neh->eh_entries, m);
1212 ext4_extent_block_csum_set(inode, neh);
1213 set_buffer_uptodate(bh);
1216 err = ext4_handle_dirty_metadata(handle, inode, bh);
1222 /* correct old index */
1224 err = ext4_ext_get_access(handle, inode, path + i);
1227 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1228 err = ext4_ext_dirty(handle, inode, path + i);
1236 /* insert new index */
1237 err = ext4_ext_insert_index(handle, inode, path + at,
1238 le32_to_cpu(border), newblock);
1242 if (buffer_locked(bh))
1248 /* free all allocated blocks in error case */
1249 for (i = 0; i < depth; i++) {
1252 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1253 EXT4_FREE_BLOCKS_METADATA);
1262 * ext4_ext_grow_indepth:
1263 * implements tree growing procedure:
1264 * - allocates new block
1265 * - moves top-level data (index block or leaf) into the new block
1266 * - initializes new top-level, creating index that points to the
1267 * just created block
1269 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1272 struct ext4_extent_header *neh;
1273 struct buffer_head *bh;
1274 ext4_fsblk_t newblock, goal = 0;
1275 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1278 /* Try to prepend new index to old one */
1279 if (ext_depth(inode))
1280 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1281 if (goal > le32_to_cpu(es->s_first_data_block)) {
1282 flags |= EXT4_MB_HINT_TRY_GOAL;
1285 goal = ext4_inode_to_goal_block(inode);
1286 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1291 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1296 err = ext4_journal_get_create_access(handle, bh);
1302 /* move top-level index/leaf into new block */
1303 memmove(bh->b_data, EXT4_I(inode)->i_data,
1304 sizeof(EXT4_I(inode)->i_data));
1306 /* set size of new block */
1307 neh = ext_block_hdr(bh);
1308 /* old root could have indexes or leaves
1309 * so calculate e_max right way */
1310 if (ext_depth(inode))
1311 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1313 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1314 neh->eh_magic = EXT4_EXT_MAGIC;
1315 ext4_extent_block_csum_set(inode, neh);
1316 set_buffer_uptodate(bh);
1319 err = ext4_handle_dirty_metadata(handle, inode, bh);
1323 /* Update top-level index: num,max,pointer */
1324 neh = ext_inode_hdr(inode);
1325 neh->eh_entries = cpu_to_le16(1);
1326 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1327 if (neh->eh_depth == 0) {
1328 /* Root extent block becomes index block */
1329 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1330 EXT_FIRST_INDEX(neh)->ei_block =
1331 EXT_FIRST_EXTENT(neh)->ee_block;
1333 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1334 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1335 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1336 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1338 le16_add_cpu(&neh->eh_depth, 1);
1339 ext4_mark_inode_dirty(handle, inode);
1347 * ext4_ext_create_new_leaf:
1348 * finds empty index and adds new leaf.
1349 * if no free index is found, then it requests in-depth growing.
1351 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1352 unsigned int mb_flags,
1353 unsigned int gb_flags,
1354 struct ext4_ext_path **ppath,
1355 struct ext4_extent *newext)
1357 struct ext4_ext_path *path = *ppath;
1358 struct ext4_ext_path *curp;
1359 int depth, i, err = 0;
1362 i = depth = ext_depth(inode);
1364 /* walk up to the tree and look for free index entry */
1365 curp = path + depth;
1366 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1371 /* we use already allocated block for index block,
1372 * so subsequent data blocks should be contiguous */
1373 if (EXT_HAS_FREE_INDEX(curp)) {
1374 /* if we found index with free entry, then use that
1375 * entry: create all needed subtree and add new leaf */
1376 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1381 path = ext4_find_extent(inode,
1382 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1385 err = PTR_ERR(path);
1387 /* tree is full, time to grow in depth */
1388 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1393 path = ext4_find_extent(inode,
1394 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1397 err = PTR_ERR(path);
1402 * only first (depth 0 -> 1) produces free space;
1403 * in all other cases we have to split the grown tree
1405 depth = ext_depth(inode);
1406 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1407 /* now we need to split */
1417 * search the closest allocated block to the left for *logical
1418 * and returns it at @logical + it's physical address at @phys
1419 * if *logical is the smallest allocated block, the function
1420 * returns 0 at @phys
1421 * return value contains 0 (success) or error code
1423 static int ext4_ext_search_left(struct inode *inode,
1424 struct ext4_ext_path *path,
1425 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1427 struct ext4_extent_idx *ix;
1428 struct ext4_extent *ex;
1431 if (unlikely(path == NULL)) {
1432 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1433 return -EFSCORRUPTED;
1435 depth = path->p_depth;
1438 if (depth == 0 && path->p_ext == NULL)
1441 /* usually extent in the path covers blocks smaller
1442 * then *logical, but it can be that extent is the
1443 * first one in the file */
1445 ex = path[depth].p_ext;
1446 ee_len = ext4_ext_get_actual_len(ex);
1447 if (*logical < le32_to_cpu(ex->ee_block)) {
1448 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1449 EXT4_ERROR_INODE(inode,
1450 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1451 *logical, le32_to_cpu(ex->ee_block));
1452 return -EFSCORRUPTED;
1454 while (--depth >= 0) {
1455 ix = path[depth].p_idx;
1456 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1457 EXT4_ERROR_INODE(inode,
1458 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1459 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1460 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1461 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1463 return -EFSCORRUPTED;
1469 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1470 EXT4_ERROR_INODE(inode,
1471 "logical %d < ee_block %d + ee_len %d!",
1472 *logical, le32_to_cpu(ex->ee_block), ee_len);
1473 return -EFSCORRUPTED;
1476 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1477 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1482 * search the closest allocated block to the right for *logical
1483 * and returns it at @logical + it's physical address at @phys
1484 * if *logical is the largest allocated block, the function
1485 * returns 0 at @phys
1486 * return value contains 0 (success) or error code
1488 static int ext4_ext_search_right(struct inode *inode,
1489 struct ext4_ext_path *path,
1490 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1491 struct ext4_extent **ret_ex)
1493 struct buffer_head *bh = NULL;
1494 struct ext4_extent_header *eh;
1495 struct ext4_extent_idx *ix;
1496 struct ext4_extent *ex;
1498 int depth; /* Note, NOT eh_depth; depth from top of tree */
1501 if (unlikely(path == NULL)) {
1502 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1503 return -EFSCORRUPTED;
1505 depth = path->p_depth;
1508 if (depth == 0 && path->p_ext == NULL)
1511 /* usually extent in the path covers blocks smaller
1512 * then *logical, but it can be that extent is the
1513 * first one in the file */
1515 ex = path[depth].p_ext;
1516 ee_len = ext4_ext_get_actual_len(ex);
1517 if (*logical < le32_to_cpu(ex->ee_block)) {
1518 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1519 EXT4_ERROR_INODE(inode,
1520 "first_extent(path[%d].p_hdr) != ex",
1522 return -EFSCORRUPTED;
1524 while (--depth >= 0) {
1525 ix = path[depth].p_idx;
1526 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1527 EXT4_ERROR_INODE(inode,
1528 "ix != EXT_FIRST_INDEX *logical %d!",
1530 return -EFSCORRUPTED;
1536 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1537 EXT4_ERROR_INODE(inode,
1538 "logical %d < ee_block %d + ee_len %d!",
1539 *logical, le32_to_cpu(ex->ee_block), ee_len);
1540 return -EFSCORRUPTED;
1543 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1544 /* next allocated block in this leaf */
1549 /* go up and search for index to the right */
1550 while (--depth >= 0) {
1551 ix = path[depth].p_idx;
1552 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1556 /* we've gone up to the root and found no index to the right */
1560 /* we've found index to the right, let's
1561 * follow it and find the closest allocated
1562 * block to the right */
1564 block = ext4_idx_pblock(ix);
1565 while (++depth < path->p_depth) {
1566 /* subtract from p_depth to get proper eh_depth */
1567 bh = read_extent_tree_block(inode, block,
1568 path->p_depth - depth, 0);
1571 eh = ext_block_hdr(bh);
1572 ix = EXT_FIRST_INDEX(eh);
1573 block = ext4_idx_pblock(ix);
1577 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1580 eh = ext_block_hdr(bh);
1581 ex = EXT_FIRST_EXTENT(eh);
1583 *logical = le32_to_cpu(ex->ee_block);
1584 *phys = ext4_ext_pblock(ex);
1592 * ext4_ext_next_allocated_block:
1593 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1594 * NOTE: it considers block number from index entry as
1595 * allocated block. Thus, index entries have to be consistent
1599 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1603 BUG_ON(path == NULL);
1604 depth = path->p_depth;
1606 if (depth == 0 && path->p_ext == NULL)
1607 return EXT_MAX_BLOCKS;
1609 while (depth >= 0) {
1610 if (depth == path->p_depth) {
1612 if (path[depth].p_ext &&
1613 path[depth].p_ext !=
1614 EXT_LAST_EXTENT(path[depth].p_hdr))
1615 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1618 if (path[depth].p_idx !=
1619 EXT_LAST_INDEX(path[depth].p_hdr))
1620 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1625 return EXT_MAX_BLOCKS;
1629 * ext4_ext_next_leaf_block:
1630 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1632 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1636 BUG_ON(path == NULL);
1637 depth = path->p_depth;
1639 /* zero-tree has no leaf blocks at all */
1641 return EXT_MAX_BLOCKS;
1643 /* go to index block */
1646 while (depth >= 0) {
1647 if (path[depth].p_idx !=
1648 EXT_LAST_INDEX(path[depth].p_hdr))
1649 return (ext4_lblk_t)
1650 le32_to_cpu(path[depth].p_idx[1].ei_block);
1654 return EXT_MAX_BLOCKS;
1658 * ext4_ext_correct_indexes:
1659 * if leaf gets modified and modified extent is first in the leaf,
1660 * then we have to correct all indexes above.
1661 * TODO: do we need to correct tree in all cases?
1663 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1664 struct ext4_ext_path *path)
1666 struct ext4_extent_header *eh;
1667 int depth = ext_depth(inode);
1668 struct ext4_extent *ex;
1672 eh = path[depth].p_hdr;
1673 ex = path[depth].p_ext;
1675 if (unlikely(ex == NULL || eh == NULL)) {
1676 EXT4_ERROR_INODE(inode,
1677 "ex %p == NULL or eh %p == NULL", ex, eh);
1678 return -EFSCORRUPTED;
1682 /* there is no tree at all */
1686 if (ex != EXT_FIRST_EXTENT(eh)) {
1687 /* we correct tree if first leaf got modified only */
1692 * TODO: we need correction if border is smaller than current one
1695 border = path[depth].p_ext->ee_block;
1696 err = ext4_ext_get_access(handle, inode, path + k);
1699 path[k].p_idx->ei_block = border;
1700 err = ext4_ext_dirty(handle, inode, path + k);
1705 /* change all left-side indexes */
1706 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1708 err = ext4_ext_get_access(handle, inode, path + k);
1711 path[k].p_idx->ei_block = border;
1712 err = ext4_ext_dirty(handle, inode, path + k);
1721 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1722 struct ext4_extent *ex2)
1724 unsigned short ext1_ee_len, ext2_ee_len;
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");
1946 return -EFSCORRUPTED;
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);
1953 return -EFSCORRUPTED;
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);
2180 err = -EFSCORRUPTED;
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");
2249 err = -EFSCORRUPTED;
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",
2272 err = -EFSCORRUPTED;
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);
2311 struct ext4_extent *ex;
2312 struct extent_status es;
2314 ex = path[depth].p_ext;
2316 /* there is no extent yet, so gap is [0;-] */
2318 len = EXT_MAX_BLOCKS;
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 } else if (block >= le32_to_cpu(ex->ee_block)
2328 + ext4_ext_get_actual_len(ex)) {
2330 lblock = le32_to_cpu(ex->ee_block)
2331 + ext4_ext_get_actual_len(ex);
2333 next = ext4_ext_next_allocated_block(path);
2334 ext_debug("cache gap(after): [%u:%u] %u",
2335 le32_to_cpu(ex->ee_block),
2336 ext4_ext_get_actual_len(ex),
2338 BUG_ON(next == lblock);
2339 len = next - lblock;
2344 ext4_es_find_delayed_extent_range(inode, lblock, lblock + len - 1, &es);
2346 /* There's delayed extent containing lblock? */
2347 if (es.es_lblk <= lblock)
2349 len = min(es.es_lblk - lblock, len);
2351 ext_debug(" -> %u:%u\n", lblock, len);
2352 ext4_es_insert_extent(inode, lblock, len, ~0, EXTENT_STATUS_HOLE);
2357 * removes index from the index block.
2359 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2360 struct ext4_ext_path *path, int depth)
2365 /* free index block */
2367 path = path + depth;
2368 leaf = ext4_idx_pblock(path->p_idx);
2369 if (unlikely(path->p_hdr->eh_entries == 0)) {
2370 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2371 return -EFSCORRUPTED;
2373 err = ext4_ext_get_access(handle, inode, path);
2377 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2378 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2379 len *= sizeof(struct ext4_extent_idx);
2380 memmove(path->p_idx, path->p_idx + 1, len);
2383 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2384 err = ext4_ext_dirty(handle, inode, path);
2387 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2388 trace_ext4_ext_rm_idx(inode, leaf);
2390 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2391 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2393 while (--depth >= 0) {
2394 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2397 err = ext4_ext_get_access(handle, inode, path);
2400 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2401 err = ext4_ext_dirty(handle, inode, path);
2409 * ext4_ext_calc_credits_for_single_extent:
2410 * This routine returns max. credits that needed to insert an extent
2411 * to the extent tree.
2412 * When pass the actual path, the caller should calculate credits
2415 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2416 struct ext4_ext_path *path)
2419 int depth = ext_depth(inode);
2422 /* probably there is space in leaf? */
2423 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2424 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2427 * There are some space in the leaf tree, no
2428 * need to account for leaf block credit
2430 * bitmaps and block group descriptor blocks
2431 * and other metadata blocks still need to be
2434 /* 1 bitmap, 1 block group descriptor */
2435 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2440 return ext4_chunk_trans_blocks(inode, nrblocks);
2444 * How many index/leaf blocks need to change/allocate to add @extents extents?
2446 * If we add a single extent, then in the worse case, each tree level
2447 * index/leaf need to be changed in case of the tree split.
2449 * If more extents are inserted, they could cause the whole tree split more
2450 * than once, but this is really rare.
2452 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2457 /* If we are converting the inline data, only one is needed here. */
2458 if (ext4_has_inline_data(inode))
2461 depth = ext_depth(inode);
2471 static inline int get_default_free_blocks_flags(struct inode *inode)
2473 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2474 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2475 else if (ext4_should_journal_data(inode))
2476 return EXT4_FREE_BLOCKS_FORGET;
2480 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2481 struct ext4_extent *ex,
2482 long long *partial_cluster,
2483 ext4_lblk_t from, ext4_lblk_t to)
2485 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2486 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2488 int flags = get_default_free_blocks_flags(inode);
2491 * For bigalloc file systems, we never free a partial cluster
2492 * at the beginning of the extent. Instead, we make a note
2493 * that we tried freeing the cluster, and check to see if we
2494 * need to free it on a subsequent call to ext4_remove_blocks,
2495 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2497 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2499 trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2501 * If we have a partial cluster, and it's different from the
2502 * cluster of the last block, we need to explicitly free the
2503 * partial cluster here.
2505 pblk = ext4_ext_pblock(ex) + ee_len - 1;
2506 if (*partial_cluster > 0 &&
2507 *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2508 ext4_free_blocks(handle, inode, NULL,
2509 EXT4_C2B(sbi, *partial_cluster),
2510 sbi->s_cluster_ratio, flags);
2511 *partial_cluster = 0;
2514 #ifdef EXTENTS_STATS
2516 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2517 spin_lock(&sbi->s_ext_stats_lock);
2518 sbi->s_ext_blocks += ee_len;
2519 sbi->s_ext_extents++;
2520 if (ee_len < sbi->s_ext_min)
2521 sbi->s_ext_min = ee_len;
2522 if (ee_len > sbi->s_ext_max)
2523 sbi->s_ext_max = ee_len;
2524 if (ext_depth(inode) > sbi->s_depth_max)
2525 sbi->s_depth_max = ext_depth(inode);
2526 spin_unlock(&sbi->s_ext_stats_lock);
2529 if (from >= le32_to_cpu(ex->ee_block)
2530 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2533 long long first_cluster;
2535 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2536 pblk = ext4_ext_pblock(ex) + ee_len - num;
2538 * Usually we want to free partial cluster at the end of the
2539 * extent, except for the situation when the cluster is still
2540 * used by any other extent (partial_cluster is negative).
2542 if (*partial_cluster < 0 &&
2543 *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1))
2544 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2546 ext_debug("free last %u blocks starting %llu partial %lld\n",
2547 num, pblk, *partial_cluster);
2548 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2550 * If the block range to be freed didn't start at the
2551 * beginning of a cluster, and we removed the entire
2552 * extent and the cluster is not used by any other extent,
2553 * save the partial cluster here, since we might need to
2554 * delete if we determine that the truncate or punch hole
2555 * operation has removed all of the blocks in the cluster.
2556 * If that cluster is used by another extent, preserve its
2557 * negative value so it isn't freed later on.
2559 * If the whole extent wasn't freed, we've reached the
2560 * start of the truncated/punched region and have finished
2561 * removing blocks. If there's a partial cluster here it's
2562 * shared with the remainder of the extent and is no longer
2563 * a candidate for removal.
2565 if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) {
2566 first_cluster = (long long) EXT4_B2C(sbi, pblk);
2567 if (first_cluster != -*partial_cluster)
2568 *partial_cluster = first_cluster;
2570 *partial_cluster = 0;
2573 ext4_error(sbi->s_sb, "strange request: removal(2) "
2574 "%u-%u from %u:%u\n",
2575 from, to, le32_to_cpu(ex->ee_block), ee_len);
2581 * ext4_ext_rm_leaf() Removes the extents associated with the
2582 * blocks appearing between "start" and "end". Both "start"
2583 * and "end" must appear in the same extent or EIO is returned.
2585 * @handle: The journal handle
2586 * @inode: The files inode
2587 * @path: The path to the leaf
2588 * @partial_cluster: The cluster which we'll have to free if all extents
2589 * has been released from it. However, if this value is
2590 * negative, it's a cluster just to the right of the
2591 * punched region and it must not be freed.
2592 * @start: The first block to remove
2593 * @end: The last block to remove
2596 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2597 struct ext4_ext_path *path,
2598 long long *partial_cluster,
2599 ext4_lblk_t start, ext4_lblk_t end)
2601 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2602 int err = 0, correct_index = 0;
2603 int depth = ext_depth(inode), credits;
2604 struct ext4_extent_header *eh;
2607 ext4_lblk_t ex_ee_block;
2608 unsigned short ex_ee_len;
2609 unsigned unwritten = 0;
2610 struct ext4_extent *ex;
2613 /* the header must be checked already in ext4_ext_remove_space() */
2614 ext_debug("truncate since %u in leaf to %u\n", start, end);
2615 if (!path[depth].p_hdr)
2616 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2617 eh = path[depth].p_hdr;
2618 if (unlikely(path[depth].p_hdr == NULL)) {
2619 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2620 return -EFSCORRUPTED;
2622 /* find where to start removing */
2623 ex = path[depth].p_ext;
2625 ex = EXT_LAST_EXTENT(eh);
2627 ex_ee_block = le32_to_cpu(ex->ee_block);
2628 ex_ee_len = ext4_ext_get_actual_len(ex);
2630 trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2632 while (ex >= EXT_FIRST_EXTENT(eh) &&
2633 ex_ee_block + ex_ee_len > start) {
2635 if (ext4_ext_is_unwritten(ex))
2640 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2641 unwritten, ex_ee_len);
2642 path[depth].p_ext = ex;
2644 a = ex_ee_block > start ? ex_ee_block : start;
2645 b = ex_ee_block+ex_ee_len - 1 < end ?
2646 ex_ee_block+ex_ee_len - 1 : end;
2648 ext_debug(" border %u:%u\n", a, b);
2650 /* If this extent is beyond the end of the hole, skip it */
2651 if (end < ex_ee_block) {
2653 * We're going to skip this extent and move to another,
2654 * so note that its first cluster is in use to avoid
2655 * freeing it when removing blocks. Eventually, the
2656 * right edge of the truncated/punched region will
2657 * be just to the left.
2659 if (sbi->s_cluster_ratio > 1) {
2660 pblk = ext4_ext_pblock(ex);
2662 -(long long) EXT4_B2C(sbi, pblk);
2665 ex_ee_block = le32_to_cpu(ex->ee_block);
2666 ex_ee_len = ext4_ext_get_actual_len(ex);
2668 } else if (b != ex_ee_block + ex_ee_len - 1) {
2669 EXT4_ERROR_INODE(inode,
2670 "can not handle truncate %u:%u "
2672 start, end, ex_ee_block,
2673 ex_ee_block + ex_ee_len - 1);
2674 err = -EFSCORRUPTED;
2676 } else if (a != ex_ee_block) {
2677 /* remove tail of the extent */
2678 num = a - ex_ee_block;
2680 /* remove whole extent: excellent! */
2684 * 3 for leaf, sb, and inode plus 2 (bmap and group
2685 * descriptor) for each block group; assume two block
2686 * groups plus ex_ee_len/blocks_per_block_group for
2689 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2690 if (ex == EXT_FIRST_EXTENT(eh)) {
2692 credits += (ext_depth(inode)) + 1;
2694 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2696 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2700 err = ext4_ext_get_access(handle, inode, path + depth);
2704 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2710 /* this extent is removed; mark slot entirely unused */
2711 ext4_ext_store_pblock(ex, 0);
2713 ex->ee_len = cpu_to_le16(num);
2715 * Do not mark unwritten if all the blocks in the
2716 * extent have been removed.
2718 if (unwritten && num)
2719 ext4_ext_mark_unwritten(ex);
2721 * If the extent was completely released,
2722 * we need to remove it from the leaf
2725 if (end != EXT_MAX_BLOCKS - 1) {
2727 * For hole punching, we need to scoot all the
2728 * extents up when an extent is removed so that
2729 * we dont have blank extents in the middle
2731 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2732 sizeof(struct ext4_extent));
2734 /* Now get rid of the one at the end */
2735 memset(EXT_LAST_EXTENT(eh), 0,
2736 sizeof(struct ext4_extent));
2738 le16_add_cpu(&eh->eh_entries, -1);
2741 err = ext4_ext_dirty(handle, inode, path + depth);
2745 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2746 ext4_ext_pblock(ex));
2748 ex_ee_block = le32_to_cpu(ex->ee_block);
2749 ex_ee_len = ext4_ext_get_actual_len(ex);
2752 if (correct_index && eh->eh_entries)
2753 err = ext4_ext_correct_indexes(handle, inode, path);
2756 * If there's a partial cluster and at least one extent remains in
2757 * the leaf, free the partial cluster if it isn't shared with the
2758 * current extent. If it is shared with the current extent
2759 * we zero partial_cluster because we've reached the start of the
2760 * truncated/punched region and we're done removing blocks.
2762 if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) {
2763 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2764 if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) {
2765 ext4_free_blocks(handle, inode, NULL,
2766 EXT4_C2B(sbi, *partial_cluster),
2767 sbi->s_cluster_ratio,
2768 get_default_free_blocks_flags(inode));
2770 *partial_cluster = 0;
2773 /* if this leaf is free, then we should
2774 * remove it from index block above */
2775 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2776 err = ext4_ext_rm_idx(handle, inode, path, depth);
2783 * ext4_ext_more_to_rm:
2784 * returns 1 if current index has to be freed (even partial)
2787 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2789 BUG_ON(path->p_idx == NULL);
2791 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2795 * if truncate on deeper level happened, it wasn't partial,
2796 * so we have to consider current index for truncation
2798 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2803 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2806 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2807 int depth = ext_depth(inode);
2808 struct ext4_ext_path *path = NULL;
2809 long long partial_cluster = 0;
2813 ext_debug("truncate since %u to %u\n", start, end);
2815 /* probably first extent we're gonna free will be last in block */
2816 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2818 return PTR_ERR(handle);
2821 trace_ext4_ext_remove_space(inode, start, end, depth);
2824 * Check if we are removing extents inside the extent tree. If that
2825 * is the case, we are going to punch a hole inside the extent tree
2826 * so we have to check whether we need to split the extent covering
2827 * the last block to remove so we can easily remove the part of it
2828 * in ext4_ext_rm_leaf().
2830 if (end < EXT_MAX_BLOCKS - 1) {
2831 struct ext4_extent *ex;
2832 ext4_lblk_t ee_block, ex_end, lblk;
2835 /* find extent for or closest extent to this block */
2836 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2838 ext4_journal_stop(handle);
2839 return PTR_ERR(path);
2841 depth = ext_depth(inode);
2842 /* Leaf not may not exist only if inode has no blocks at all */
2843 ex = path[depth].p_ext;
2846 EXT4_ERROR_INODE(inode,
2847 "path[%d].p_hdr == NULL",
2849 err = -EFSCORRUPTED;
2854 ee_block = le32_to_cpu(ex->ee_block);
2855 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2858 * See if the last block is inside the extent, if so split
2859 * the extent at 'end' block so we can easily remove the
2860 * tail of the first part of the split extent in
2861 * ext4_ext_rm_leaf().
2863 if (end >= ee_block && end < ex_end) {
2866 * If we're going to split the extent, note that
2867 * the cluster containing the block after 'end' is
2868 * in use to avoid freeing it when removing blocks.
2870 if (sbi->s_cluster_ratio > 1) {
2871 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2873 -(long long) EXT4_B2C(sbi, pblk);
2877 * Split the extent in two so that 'end' is the last
2878 * block in the first new extent. Also we should not
2879 * fail removing space due to ENOSPC so try to use
2880 * reserved block if that happens.
2882 err = ext4_force_split_extent_at(handle, inode, &path,
2887 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) {
2889 * If there's an extent to the right its first cluster
2890 * contains the immediate right boundary of the
2891 * truncated/punched region. Set partial_cluster to
2892 * its negative value so it won't be freed if shared
2893 * with the current extent. The end < ee_block case
2894 * is handled in ext4_ext_rm_leaf().
2897 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2903 -(long long) EXT4_B2C(sbi, pblk);
2907 * We start scanning from right side, freeing all the blocks
2908 * after i_size and walking into the tree depth-wise.
2910 depth = ext_depth(inode);
2915 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2917 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2920 ext4_journal_stop(handle);
2923 path[0].p_maxdepth = path[0].p_depth = depth;
2924 path[0].p_hdr = ext_inode_hdr(inode);
2927 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2928 err = -EFSCORRUPTED;
2934 while (i >= 0 && err == 0) {
2936 /* this is leaf block */
2937 err = ext4_ext_rm_leaf(handle, inode, path,
2938 &partial_cluster, start,
2940 /* root level has p_bh == NULL, brelse() eats this */
2941 brelse(path[i].p_bh);
2942 path[i].p_bh = NULL;
2947 /* this is index block */
2948 if (!path[i].p_hdr) {
2949 ext_debug("initialize header\n");
2950 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2953 if (!path[i].p_idx) {
2954 /* this level hasn't been touched yet */
2955 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2956 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2957 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2959 le16_to_cpu(path[i].p_hdr->eh_entries));
2961 /* we were already here, see at next index */
2965 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2966 i, EXT_FIRST_INDEX(path[i].p_hdr),
2968 if (ext4_ext_more_to_rm(path + i)) {
2969 struct buffer_head *bh;
2970 /* go to the next level */
2971 ext_debug("move to level %d (block %llu)\n",
2972 i + 1, ext4_idx_pblock(path[i].p_idx));
2973 memset(path + i + 1, 0, sizeof(*path));
2974 bh = read_extent_tree_block(inode,
2975 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2978 /* should we reset i_size? */
2982 /* Yield here to deal with large extent trees.
2983 * Should be a no-op if we did IO above. */
2985 if (WARN_ON(i + 1 > depth)) {
2986 err = -EFSCORRUPTED;
2989 path[i + 1].p_bh = bh;
2991 /* save actual number of indexes since this
2992 * number is changed at the next iteration */
2993 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2996 /* we finished processing this index, go up */
2997 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2998 /* index is empty, remove it;
2999 * handle must be already prepared by the
3000 * truncatei_leaf() */
3001 err = ext4_ext_rm_idx(handle, inode, path, i);
3003 /* root level has p_bh == NULL, brelse() eats this */
3004 brelse(path[i].p_bh);
3005 path[i].p_bh = NULL;
3007 ext_debug("return to level %d\n", i);
3011 trace_ext4_ext_remove_space_done(inode, start, end, depth,
3012 partial_cluster, path->p_hdr->eh_entries);
3015 * If we still have something in the partial cluster and we have removed
3016 * even the first extent, then we should free the blocks in the partial
3017 * cluster as well. (This code will only run when there are no leaves
3018 * to the immediate left of the truncated/punched region.)
3020 if (partial_cluster > 0 && err == 0) {
3021 /* don't zero partial_cluster since it's not used afterwards */
3022 ext4_free_blocks(handle, inode, NULL,
3023 EXT4_C2B(sbi, partial_cluster),
3024 sbi->s_cluster_ratio,
3025 get_default_free_blocks_flags(inode));
3028 /* TODO: flexible tree reduction should be here */
3029 if (path->p_hdr->eh_entries == 0) {
3031 * truncate to zero freed all the tree,
3032 * so we need to correct eh_depth
3034 err = ext4_ext_get_access(handle, inode, path);
3036 ext_inode_hdr(inode)->eh_depth = 0;
3037 ext_inode_hdr(inode)->eh_max =
3038 cpu_to_le16(ext4_ext_space_root(inode, 0));
3039 err = ext4_ext_dirty(handle, inode, path);
3043 ext4_ext_drop_refs(path);
3048 ext4_journal_stop(handle);
3054 * called at mount time
3056 void ext4_ext_init(struct super_block *sb)
3059 * possible initialization would be here
3062 if (ext4_has_feature_extents(sb)) {
3063 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3064 printk(KERN_INFO "EXT4-fs: file extents enabled"
3065 #ifdef AGGRESSIVE_TEST
3066 ", aggressive tests"
3068 #ifdef CHECK_BINSEARCH
3071 #ifdef EXTENTS_STATS
3076 #ifdef EXTENTS_STATS
3077 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3078 EXT4_SB(sb)->s_ext_min = 1 << 30;
3079 EXT4_SB(sb)->s_ext_max = 0;
3085 * called at umount time
3087 void ext4_ext_release(struct super_block *sb)
3089 if (!ext4_has_feature_extents(sb))
3092 #ifdef EXTENTS_STATS
3093 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3094 struct ext4_sb_info *sbi = EXT4_SB(sb);
3095 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3096 sbi->s_ext_blocks, sbi->s_ext_extents,
3097 sbi->s_ext_blocks / sbi->s_ext_extents);
3098 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3099 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3104 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3106 ext4_lblk_t ee_block;
3107 ext4_fsblk_t ee_pblock;
3108 unsigned int ee_len;
3110 ee_block = le32_to_cpu(ex->ee_block);
3111 ee_len = ext4_ext_get_actual_len(ex);
3112 ee_pblock = ext4_ext_pblock(ex);
3117 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3118 EXTENT_STATUS_WRITTEN);
3121 /* FIXME!! we need to try to merge to left or right after zero-out */
3122 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3124 ext4_fsblk_t ee_pblock;
3125 unsigned int ee_len;
3128 ee_len = ext4_ext_get_actual_len(ex);
3129 ee_pblock = ext4_ext_pblock(ex);
3131 if (ext4_encrypted_inode(inode))
3132 return ext4_encrypted_zeroout(inode, ex);
3134 ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
3142 * ext4_split_extent_at() splits an extent at given block.
3144 * @handle: the journal handle
3145 * @inode: the file inode
3146 * @path: the path to the extent
3147 * @split: the logical block where the extent is splitted.
3148 * @split_flags: indicates if the extent could be zeroout if split fails, and
3149 * the states(init or unwritten) of new extents.
3150 * @flags: flags used to insert new extent to extent tree.
3153 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3154 * of which are deterimined by split_flag.
3156 * There are two cases:
3157 * a> the extent are splitted into two extent.
3158 * b> split is not needed, and just mark the extent.
3160 * return 0 on success.
3162 static int ext4_split_extent_at(handle_t *handle,
3163 struct inode *inode,
3164 struct ext4_ext_path **ppath,
3169 struct ext4_ext_path *path = *ppath;
3170 ext4_fsblk_t newblock;
3171 ext4_lblk_t ee_block;
3172 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3173 struct ext4_extent *ex2 = NULL;
3174 unsigned int ee_len, depth;
3177 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3178 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3180 ext_debug("ext4_split_extents_at: inode %lu, logical"
3181 "block %llu\n", inode->i_ino, (unsigned long long)split);
3183 ext4_ext_show_leaf(inode, path);
3185 depth = ext_depth(inode);
3186 ex = path[depth].p_ext;
3187 ee_block = le32_to_cpu(ex->ee_block);
3188 ee_len = ext4_ext_get_actual_len(ex);
3189 newblock = split - ee_block + ext4_ext_pblock(ex);
3191 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3192 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3193 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3194 EXT4_EXT_MARK_UNWRIT1 |
3195 EXT4_EXT_MARK_UNWRIT2));
3197 err = ext4_ext_get_access(handle, inode, path + depth);
3201 if (split == ee_block) {
3203 * case b: block @split is the block that the extent begins with
3204 * then we just change the state of the extent, and splitting
3207 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3208 ext4_ext_mark_unwritten(ex);
3210 ext4_ext_mark_initialized(ex);
3212 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3213 ext4_ext_try_to_merge(handle, inode, path, ex);
3215 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3220 memcpy(&orig_ex, ex, sizeof(orig_ex));
3221 ex->ee_len = cpu_to_le16(split - ee_block);
3222 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3223 ext4_ext_mark_unwritten(ex);
3226 * path may lead to new leaf, not to original leaf any more
3227 * after ext4_ext_insert_extent() returns,
3229 err = ext4_ext_dirty(handle, inode, path + depth);
3231 goto fix_extent_len;
3234 ex2->ee_block = cpu_to_le32(split);
3235 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3236 ext4_ext_store_pblock(ex2, newblock);
3237 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3238 ext4_ext_mark_unwritten(ex2);
3240 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3241 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3242 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3243 if (split_flag & EXT4_EXT_DATA_VALID1) {
3244 err = ext4_ext_zeroout(inode, ex2);
3245 zero_ex.ee_block = ex2->ee_block;
3246 zero_ex.ee_len = cpu_to_le16(
3247 ext4_ext_get_actual_len(ex2));
3248 ext4_ext_store_pblock(&zero_ex,
3249 ext4_ext_pblock(ex2));
3251 err = ext4_ext_zeroout(inode, ex);
3252 zero_ex.ee_block = ex->ee_block;
3253 zero_ex.ee_len = cpu_to_le16(
3254 ext4_ext_get_actual_len(ex));
3255 ext4_ext_store_pblock(&zero_ex,
3256 ext4_ext_pblock(ex));
3259 err = ext4_ext_zeroout(inode, &orig_ex);
3260 zero_ex.ee_block = orig_ex.ee_block;
3261 zero_ex.ee_len = cpu_to_le16(
3262 ext4_ext_get_actual_len(&orig_ex));
3263 ext4_ext_store_pblock(&zero_ex,
3264 ext4_ext_pblock(&orig_ex));
3268 goto fix_extent_len;
3269 /* update the extent length and mark as initialized */
3270 ex->ee_len = cpu_to_le16(ee_len);
3271 ext4_ext_try_to_merge(handle, inode, path, ex);
3272 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3274 goto fix_extent_len;
3276 /* update extent status tree */
3277 err = ext4_zeroout_es(inode, &zero_ex);
3281 goto fix_extent_len;
3284 ext4_ext_show_leaf(inode, path);
3288 ex->ee_len = orig_ex.ee_len;
3289 ext4_ext_dirty(handle, inode, path + path->p_depth);
3294 * ext4_split_extents() splits an extent and mark extent which is covered
3295 * by @map as split_flags indicates
3297 * It may result in splitting the extent into multiple extents (up to three)
3298 * There are three possibilities:
3299 * a> There is no split required
3300 * b> Splits in two extents: Split is happening at either end of the extent
3301 * c> Splits in three extents: Somone is splitting in middle of the extent
3304 static int ext4_split_extent(handle_t *handle,
3305 struct inode *inode,
3306 struct ext4_ext_path **ppath,
3307 struct ext4_map_blocks *map,
3311 struct ext4_ext_path *path = *ppath;
3312 ext4_lblk_t ee_block;
3313 struct ext4_extent *ex;
3314 unsigned int ee_len, depth;
3317 int split_flag1, flags1;
3318 int allocated = map->m_len;
3320 depth = ext_depth(inode);
3321 ex = path[depth].p_ext;
3322 ee_block = le32_to_cpu(ex->ee_block);
3323 ee_len = ext4_ext_get_actual_len(ex);
3324 unwritten = ext4_ext_is_unwritten(ex);
3326 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3327 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3328 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3330 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3331 EXT4_EXT_MARK_UNWRIT2;
3332 if (split_flag & EXT4_EXT_DATA_VALID2)
3333 split_flag1 |= EXT4_EXT_DATA_VALID1;
3334 err = ext4_split_extent_at(handle, inode, ppath,
3335 map->m_lblk + map->m_len, split_flag1, flags1);
3339 allocated = ee_len - (map->m_lblk - ee_block);
3342 * Update path is required because previous ext4_split_extent_at() may
3343 * result in split of original leaf or extent zeroout.
3345 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3347 return PTR_ERR(path);
3348 depth = ext_depth(inode);
3349 ex = path[depth].p_ext;
3351 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3352 (unsigned long) map->m_lblk);
3353 return -EFSCORRUPTED;
3355 unwritten = ext4_ext_is_unwritten(ex);
3358 if (map->m_lblk >= ee_block) {
3359 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3361 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3362 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3363 EXT4_EXT_MARK_UNWRIT2);
3365 err = ext4_split_extent_at(handle, inode, ppath,
3366 map->m_lblk, split_flag1, flags);
3371 ext4_ext_show_leaf(inode, path);
3373 return err ? err : allocated;
3377 * This function is called by ext4_ext_map_blocks() if someone tries to write
3378 * to an unwritten extent. It may result in splitting the unwritten
3379 * extent into multiple extents (up to three - one initialized and two
3381 * There are three possibilities:
3382 * a> There is no split required: Entire extent should be initialized
3383 * b> Splits in two extents: Write is happening at either end of the extent
3384 * c> Splits in three extents: Somone is writing in middle of the extent
3387 * - The extent pointed to by 'path' is unwritten.
3388 * - The extent pointed to by 'path' contains a superset
3389 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3391 * Post-conditions on success:
3392 * - the returned value is the number of blocks beyond map->l_lblk
3393 * that are allocated and initialized.
3394 * It is guaranteed to be >= map->m_len.
3396 static int ext4_ext_convert_to_initialized(handle_t *handle,
3397 struct inode *inode,
3398 struct ext4_map_blocks *map,
3399 struct ext4_ext_path **ppath,
3402 struct ext4_ext_path *path = *ppath;
3403 struct ext4_sb_info *sbi;
3404 struct ext4_extent_header *eh;
3405 struct ext4_map_blocks split_map;
3406 struct ext4_extent zero_ex;
3407 struct ext4_extent *ex, *abut_ex;
3408 ext4_lblk_t ee_block, eof_block;
3409 unsigned int ee_len, depth, map_len = map->m_len;
3410 int allocated = 0, max_zeroout = 0;
3414 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3415 "block %llu, max_blocks %u\n", inode->i_ino,
3416 (unsigned long long)map->m_lblk, map_len);
3418 sbi = EXT4_SB(inode->i_sb);
3419 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3420 inode->i_sb->s_blocksize_bits;
3421 if (eof_block < map->m_lblk + map_len)
3422 eof_block = map->m_lblk + map_len;
3424 depth = ext_depth(inode);
3425 eh = path[depth].p_hdr;
3426 ex = path[depth].p_ext;
3427 ee_block = le32_to_cpu(ex->ee_block);
3428 ee_len = ext4_ext_get_actual_len(ex);
3431 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3433 /* Pre-conditions */
3434 BUG_ON(!ext4_ext_is_unwritten(ex));
3435 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3438 * Attempt to transfer newly initialized blocks from the currently
3439 * unwritten extent to its neighbor. This is much cheaper
3440 * than an insertion followed by a merge as those involve costly
3441 * memmove() calls. Transferring to the left is the common case in
3442 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3443 * followed by append writes.
3445 * Limitations of the current logic:
3446 * - L1: we do not deal with writes covering the whole extent.
3447 * This would require removing the extent if the transfer
3449 * - L2: we only attempt to merge with an extent stored in the
3450 * same extent tree node.
3452 if ((map->m_lblk == ee_block) &&
3453 /* See if we can merge left */
3454 (map_len < ee_len) && /*L1*/
3455 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3456 ext4_lblk_t prev_lblk;
3457 ext4_fsblk_t prev_pblk, ee_pblk;
3458 unsigned int prev_len;
3461 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3462 prev_len = ext4_ext_get_actual_len(abut_ex);
3463 prev_pblk = ext4_ext_pblock(abut_ex);
3464 ee_pblk = ext4_ext_pblock(ex);
3467 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3468 * upon those conditions:
3469 * - C1: abut_ex is initialized,
3470 * - C2: abut_ex is logically abutting ex,
3471 * - C3: abut_ex is physically abutting ex,
3472 * - C4: abut_ex can receive the additional blocks without
3473 * overflowing the (initialized) length limit.
3475 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3476 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3477 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3478 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3479 err = ext4_ext_get_access(handle, inode, path + depth);
3483 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3486 /* Shift the start of ex by 'map_len' blocks */
3487 ex->ee_block = cpu_to_le32(ee_block + map_len);
3488 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3489 ex->ee_len = cpu_to_le16(ee_len - map_len);
3490 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3492 /* Extend abut_ex by 'map_len' blocks */
3493 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3495 /* Result: number of initialized blocks past m_lblk */
3496 allocated = map_len;
3498 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3499 (map_len < ee_len) && /*L1*/
3500 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3501 /* See if we can merge right */
3502 ext4_lblk_t next_lblk;
3503 ext4_fsblk_t next_pblk, ee_pblk;
3504 unsigned int next_len;
3507 next_lblk = le32_to_cpu(abut_ex->ee_block);
3508 next_len = ext4_ext_get_actual_len(abut_ex);
3509 next_pblk = ext4_ext_pblock(abut_ex);
3510 ee_pblk = ext4_ext_pblock(ex);
3513 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3514 * upon those conditions:
3515 * - C1: abut_ex is initialized,
3516 * - C2: abut_ex is logically abutting ex,
3517 * - C3: abut_ex is physically abutting ex,
3518 * - C4: abut_ex can receive the additional blocks without
3519 * overflowing the (initialized) length limit.
3521 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3522 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3523 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3524 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3525 err = ext4_ext_get_access(handle, inode, path + depth);
3529 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3532 /* Shift the start of abut_ex by 'map_len' blocks */
3533 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3534 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3535 ex->ee_len = cpu_to_le16(ee_len - map_len);
3536 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3538 /* Extend abut_ex by 'map_len' blocks */
3539 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3541 /* Result: number of initialized blocks past m_lblk */
3542 allocated = map_len;
3546 /* Mark the block containing both extents as dirty */
3547 ext4_ext_dirty(handle, inode, path + depth);
3549 /* Update path to point to the right extent */
3550 path[depth].p_ext = abut_ex;
3553 allocated = ee_len - (map->m_lblk - ee_block);
3555 WARN_ON(map->m_lblk < ee_block);
3557 * It is safe to convert extent to initialized via explicit
3558 * zeroout only if extent is fully inside i_size or new_size.
3560 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3562 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3563 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3564 (inode->i_sb->s_blocksize_bits - 10);
3566 if (ext4_encrypted_inode(inode))
3569 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3570 if (max_zeroout && (ee_len <= max_zeroout)) {
3571 err = ext4_ext_zeroout(inode, ex);
3574 zero_ex.ee_block = ex->ee_block;
3575 zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex));
3576 ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex));
3578 err = ext4_ext_get_access(handle, inode, path + depth);
3581 ext4_ext_mark_initialized(ex);
3582 ext4_ext_try_to_merge(handle, inode, path, ex);
3583 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3589 * 1. split the extent into three extents.
3590 * 2. split the extent into two extents, zeroout the first half.
3591 * 3. split the extent into two extents, zeroout the second half.
3592 * 4. split the extent into two extents with out zeroout.
3594 split_map.m_lblk = map->m_lblk;
3595 split_map.m_len = map->m_len;
3597 if (max_zeroout && (allocated > map->m_len)) {
3598 if (allocated <= max_zeroout) {
3601 cpu_to_le32(map->m_lblk);
3602 zero_ex.ee_len = cpu_to_le16(allocated);
3603 ext4_ext_store_pblock(&zero_ex,
3604 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3605 err = ext4_ext_zeroout(inode, &zero_ex);
3608 split_map.m_lblk = map->m_lblk;
3609 split_map.m_len = allocated;
3610 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3612 if (map->m_lblk != ee_block) {
3613 zero_ex.ee_block = ex->ee_block;
3614 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3616 ext4_ext_store_pblock(&zero_ex,
3617 ext4_ext_pblock(ex));
3618 err = ext4_ext_zeroout(inode, &zero_ex);
3623 split_map.m_lblk = ee_block;
3624 split_map.m_len = map->m_lblk - ee_block + map->m_len;
3625 allocated = map->m_len;
3629 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3634 /* If we have gotten a failure, don't zero out status tree */
3636 err = ext4_zeroout_es(inode, &zero_ex);
3637 return err ? err : allocated;
3641 * This function is called by ext4_ext_map_blocks() from
3642 * ext4_get_blocks_dio_write() when DIO to write
3643 * to an unwritten extent.
3645 * Writing to an unwritten extent may result in splitting the unwritten
3646 * extent into multiple initialized/unwritten extents (up to three)
3647 * There are three possibilities:
3648 * a> There is no split required: Entire extent should be unwritten
3649 * b> Splits in two extents: Write is happening at either end of the extent
3650 * c> Splits in three extents: Somone is writing in middle of the extent
3652 * This works the same way in the case of initialized -> unwritten conversion.
3654 * One of more index blocks maybe needed if the extent tree grow after
3655 * the unwritten extent split. To prevent ENOSPC occur at the IO
3656 * complete, we need to split the unwritten extent before DIO submit
3657 * the IO. The unwritten extent called at this time will be split
3658 * into three unwritten extent(at most). After IO complete, the part
3659 * being filled will be convert to initialized by the end_io callback function
3660 * via ext4_convert_unwritten_extents().
3662 * Returns the size of unwritten extent to be written on success.
3664 static int ext4_split_convert_extents(handle_t *handle,
3665 struct inode *inode,
3666 struct ext4_map_blocks *map,
3667 struct ext4_ext_path **ppath,
3670 struct ext4_ext_path *path = *ppath;
3671 ext4_lblk_t eof_block;
3672 ext4_lblk_t ee_block;
3673 struct ext4_extent *ex;
3674 unsigned int ee_len;
3675 int split_flag = 0, depth;
3677 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3678 __func__, inode->i_ino,
3679 (unsigned long long)map->m_lblk, map->m_len);
3681 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3682 inode->i_sb->s_blocksize_bits;
3683 if (eof_block < map->m_lblk + map->m_len)
3684 eof_block = map->m_lblk + map->m_len;
3686 * It is safe to convert extent to initialized via explicit
3687 * zeroout only if extent is fully insde i_size or new_size.
3689 depth = ext_depth(inode);
3690 ex = path[depth].p_ext;
3691 ee_block = le32_to_cpu(ex->ee_block);
3692 ee_len = ext4_ext_get_actual_len(ex);
3694 /* Convert to unwritten */
3695 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3696 split_flag |= EXT4_EXT_DATA_VALID1;
3697 /* Convert to initialized */
3698 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3699 split_flag |= ee_block + ee_len <= eof_block ?
3700 EXT4_EXT_MAY_ZEROOUT : 0;
3701 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3703 flags |= EXT4_GET_BLOCKS_PRE_IO;
3704 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3707 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3708 struct inode *inode,
3709 struct ext4_map_blocks *map,
3710 struct ext4_ext_path **ppath)
3712 struct ext4_ext_path *path = *ppath;
3713 struct ext4_extent *ex;
3714 ext4_lblk_t ee_block;
3715 unsigned int ee_len;
3719 depth = ext_depth(inode);
3720 ex = path[depth].p_ext;
3721 ee_block = le32_to_cpu(ex->ee_block);
3722 ee_len = ext4_ext_get_actual_len(ex);
3724 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3725 "block %llu, max_blocks %u\n", inode->i_ino,
3726 (unsigned long long)ee_block, ee_len);
3728 /* If extent is larger than requested it is a clear sign that we still
3729 * have some extent state machine issues left. So extent_split is still
3731 * TODO: Once all related issues will be fixed this situation should be
3734 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3736 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3737 " len %u; IO logical block %llu, len %u\n",
3738 inode->i_ino, (unsigned long long)ee_block, ee_len,
3739 (unsigned long long)map->m_lblk, map->m_len);
3741 err = ext4_split_convert_extents(handle, inode, map, ppath,
3742 EXT4_GET_BLOCKS_CONVERT);
3745 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3747 return PTR_ERR(path);
3748 depth = ext_depth(inode);
3749 ex = path[depth].p_ext;
3752 err = ext4_ext_get_access(handle, inode, path + depth);
3755 /* first mark the extent as initialized */
3756 ext4_ext_mark_initialized(ex);
3758 /* note: ext4_ext_correct_indexes() isn't needed here because
3759 * borders are not changed
3761 ext4_ext_try_to_merge(handle, inode, path, ex);
3763 /* Mark modified extent as dirty */
3764 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3766 ext4_ext_show_leaf(inode, path);
3770 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3771 sector_t block, int count)
3774 for (i = 0; i < count; i++)
3775 unmap_underlying_metadata(bdev, block + i);
3779 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3781 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3783 struct ext4_ext_path *path,
3787 struct ext4_extent_header *eh;
3788 struct ext4_extent *last_ex;
3790 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3793 depth = ext_depth(inode);
3794 eh = path[depth].p_hdr;
3797 * We're going to remove EOFBLOCKS_FL entirely in future so we
3798 * do not care for this case anymore. Simply remove the flag
3799 * if there are no extents.
3801 if (unlikely(!eh->eh_entries))
3803 last_ex = EXT_LAST_EXTENT(eh);
3805 * We should clear the EOFBLOCKS_FL flag if we are writing the
3806 * last block in the last extent in the file. We test this by
3807 * first checking to see if the caller to
3808 * ext4_ext_get_blocks() was interested in the last block (or
3809 * a block beyond the last block) in the current extent. If
3810 * this turns out to be false, we can bail out from this
3811 * function immediately.
3813 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3814 ext4_ext_get_actual_len(last_ex))
3817 * If the caller does appear to be planning to write at or
3818 * beyond the end of the current extent, we then test to see
3819 * if the current extent is the last extent in the file, by
3820 * checking to make sure it was reached via the rightmost node
3821 * at each level of the tree.
3823 for (i = depth-1; i >= 0; i--)
3824 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3827 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3828 return ext4_mark_inode_dirty(handle, inode);
3832 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3834 * Return 1 if there is a delalloc block in the range, otherwise 0.
3836 int ext4_find_delalloc_range(struct inode *inode,
3837 ext4_lblk_t lblk_start,
3838 ext4_lblk_t lblk_end)
3840 struct extent_status es;
3842 ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es);
3844 return 0; /* there is no delay extent in this tree */
3845 else if (es.es_lblk <= lblk_start &&
3846 lblk_start < es.es_lblk + es.es_len)
3848 else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end)
3854 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk)
3856 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3857 ext4_lblk_t lblk_start, lblk_end;
3858 lblk_start = EXT4_LBLK_CMASK(sbi, lblk);
3859 lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3861 return ext4_find_delalloc_range(inode, lblk_start, lblk_end);
3865 * Determines how many complete clusters (out of those specified by the 'map')
3866 * are under delalloc and were reserved quota for.
3867 * This function is called when we are writing out the blocks that were
3868 * originally written with their allocation delayed, but then the space was
3869 * allocated using fallocate() before the delayed allocation could be resolved.
3870 * The cases to look for are:
3871 * ('=' indicated delayed allocated blocks
3872 * '-' indicates non-delayed allocated blocks)
3873 * (a) partial clusters towards beginning and/or end outside of allocated range
3874 * are not delalloc'ed.
3876 * |----c---=|====c====|====c====|===-c----|
3877 * |++++++ allocated ++++++|
3878 * ==> 4 complete clusters in above example
3880 * (b) partial cluster (outside of allocated range) towards either end is
3881 * marked for delayed allocation. In this case, we will exclude that
3884 * |----====c========|========c========|
3885 * |++++++ allocated ++++++|
3886 * ==> 1 complete clusters in above example
3889 * |================c================|
3890 * |++++++ allocated ++++++|
3891 * ==> 0 complete clusters in above example
3893 * The ext4_da_update_reserve_space will be called only if we
3894 * determine here that there were some "entire" clusters that span
3895 * this 'allocated' range.
3896 * In the non-bigalloc case, this function will just end up returning num_blks
3897 * without ever calling ext4_find_delalloc_range.
3900 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3901 unsigned int num_blks)
3903 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3904 ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3905 ext4_lblk_t lblk_from, lblk_to, c_offset;
3906 unsigned int allocated_clusters = 0;
3908 alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3909 alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3911 /* max possible clusters for this allocation */
3912 allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3914 trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3916 /* Check towards left side */
3917 c_offset = EXT4_LBLK_COFF(sbi, lblk_start);
3919 lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start);
3920 lblk_to = lblk_from + c_offset - 1;
3922 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3923 allocated_clusters--;
3926 /* Now check towards right. */
3927 c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks);
3928 if (allocated_clusters && c_offset) {
3929 lblk_from = lblk_start + num_blks;
3930 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3932 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to))
3933 allocated_clusters--;
3936 return allocated_clusters;
3940 convert_initialized_extent(handle_t *handle, struct inode *inode,
3941 struct ext4_map_blocks *map,
3942 struct ext4_ext_path **ppath, int flags,
3943 unsigned int allocated, ext4_fsblk_t newblock)
3945 struct ext4_ext_path *path = *ppath;
3946 struct ext4_extent *ex;
3947 ext4_lblk_t ee_block;
3948 unsigned int ee_len;
3953 * Make sure that the extent is no bigger than we support with
3956 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3957 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3959 depth = ext_depth(inode);
3960 ex = path[depth].p_ext;
3961 ee_block = le32_to_cpu(ex->ee_block);
3962 ee_len = ext4_ext_get_actual_len(ex);
3964 ext_debug("%s: inode %lu, logical"
3965 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3966 (unsigned long long)ee_block, ee_len);
3968 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3969 err = ext4_split_convert_extents(handle, inode, map, ppath,
3970 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3973 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3975 return PTR_ERR(path);
3976 depth = ext_depth(inode);
3977 ex = path[depth].p_ext;
3979 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3980 (unsigned long) map->m_lblk);
3981 return -EFSCORRUPTED;
3985 err = ext4_ext_get_access(handle, inode, path + depth);
3988 /* first mark the extent as unwritten */
3989 ext4_ext_mark_unwritten(ex);
3991 /* note: ext4_ext_correct_indexes() isn't needed here because
3992 * borders are not changed
3994 ext4_ext_try_to_merge(handle, inode, path, ex);
3996 /* Mark modified extent as dirty */
3997 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4000 ext4_ext_show_leaf(inode, path);
4002 ext4_update_inode_fsync_trans(handle, inode, 1);
4003 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4006 map->m_flags |= EXT4_MAP_UNWRITTEN;
4007 if (allocated > map->m_len)
4008 allocated = map->m_len;
4009 map->m_len = allocated;
4014 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4015 struct ext4_map_blocks *map,
4016 struct ext4_ext_path **ppath, int flags,
4017 unsigned int allocated, ext4_fsblk_t newblock)
4019 struct ext4_ext_path *path = *ppath;
4022 ext4_io_end_t *io = ext4_inode_aio(inode);
4024 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4025 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4026 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4028 ext4_ext_show_leaf(inode, path);
4031 * When writing into unwritten space, we should not fail to
4032 * allocate metadata blocks for the new extent block if needed.
4034 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4036 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4037 allocated, newblock);
4039 /* get_block() before submit the IO, split the extent */
4040 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4041 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4042 flags | EXT4_GET_BLOCKS_CONVERT);
4046 * Flag the inode(non aio case) or end_io struct (aio case)
4047 * that this IO needs to conversion to written when IO is
4051 ext4_set_io_unwritten_flag(inode, io);
4053 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4054 map->m_flags |= EXT4_MAP_UNWRITTEN;
4057 /* IO end_io complete, convert the filled extent to written */
4058 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4059 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4062 ext4_update_inode_fsync_trans(handle, inode, 1);
4063 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4067 map->m_flags |= EXT4_MAP_MAPPED;
4068 map->m_pblk = newblock;
4069 if (allocated > map->m_len)
4070 allocated = map->m_len;
4071 map->m_len = allocated;
4074 /* buffered IO case */
4076 * repeat fallocate creation request
4077 * we already have an unwritten extent
4079 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4080 map->m_flags |= EXT4_MAP_UNWRITTEN;
4084 /* buffered READ or buffered write_begin() lookup */
4085 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4087 * We have blocks reserved already. We
4088 * return allocated blocks so that delalloc
4089 * won't do block reservation for us. But
4090 * the buffer head will be unmapped so that
4091 * a read from the block returns 0s.
4093 map->m_flags |= EXT4_MAP_UNWRITTEN;
4097 /* buffered write, writepage time, convert*/
4098 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4100 ext4_update_inode_fsync_trans(handle, inode, 1);
4107 map->m_flags |= EXT4_MAP_NEW;
4109 * if we allocated more blocks than requested
4110 * we need to make sure we unmap the extra block
4111 * allocated. The actual needed block will get
4112 * unmapped later when we find the buffer_head marked
4115 if (allocated > map->m_len) {
4116 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
4117 newblock + map->m_len,
4118 allocated - map->m_len);
4119 allocated = map->m_len;
4121 map->m_len = allocated;
4124 * If we have done fallocate with the offset that is already
4125 * delayed allocated, we would have block reservation
4126 * and quota reservation done in the delayed write path.
4127 * But fallocate would have already updated quota and block
4128 * count for this offset. So cancel these reservation
4130 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4131 unsigned int reserved_clusters;
4132 reserved_clusters = get_reserved_cluster_alloc(inode,
4133 map->m_lblk, map->m_len);
4134 if (reserved_clusters)
4135 ext4_da_update_reserve_space(inode,
4141 map->m_flags |= EXT4_MAP_MAPPED;
4142 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4143 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4149 if (allocated > map->m_len)
4150 allocated = map->m_len;
4151 ext4_ext_show_leaf(inode, path);
4152 map->m_pblk = newblock;
4153 map->m_len = allocated;
4155 return err ? err : allocated;
4159 * get_implied_cluster_alloc - check to see if the requested
4160 * allocation (in the map structure) overlaps with a cluster already
4161 * allocated in an extent.
4162 * @sb The filesystem superblock structure
4163 * @map The requested lblk->pblk mapping
4164 * @ex The extent structure which might contain an implied
4165 * cluster allocation
4167 * This function is called by ext4_ext_map_blocks() after we failed to
4168 * find blocks that were already in the inode's extent tree. Hence,
4169 * we know that the beginning of the requested region cannot overlap
4170 * the extent from the inode's extent tree. There are three cases we
4171 * want to catch. The first is this case:
4173 * |--- cluster # N--|
4174 * |--- extent ---| |---- requested region ---|
4177 * The second case that we need to test for is this one:
4179 * |--------- cluster # N ----------------|
4180 * |--- requested region --| |------- extent ----|
4181 * |=======================|
4183 * The third case is when the requested region lies between two extents
4184 * within the same cluster:
4185 * |------------- cluster # N-------------|
4186 * |----- ex -----| |---- ex_right ----|
4187 * |------ requested region ------|
4188 * |================|
4190 * In each of the above cases, we need to set the map->m_pblk and
4191 * map->m_len so it corresponds to the return the extent labelled as
4192 * "|====|" from cluster #N, since it is already in use for data in
4193 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4194 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4195 * as a new "allocated" block region. Otherwise, we will return 0 and
4196 * ext4_ext_map_blocks() will then allocate one or more new clusters
4197 * by calling ext4_mb_new_blocks().
4199 static int get_implied_cluster_alloc(struct super_block *sb,
4200 struct ext4_map_blocks *map,
4201 struct ext4_extent *ex,
4202 struct ext4_ext_path *path)
4204 struct ext4_sb_info *sbi = EXT4_SB(sb);
4205 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4206 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4207 ext4_lblk_t rr_cluster_start;
4208 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4209 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4210 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4212 /* The extent passed in that we are trying to match */
4213 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4214 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4216 /* The requested region passed into ext4_map_blocks() */
4217 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4219 if ((rr_cluster_start == ex_cluster_end) ||
4220 (rr_cluster_start == ex_cluster_start)) {
4221 if (rr_cluster_start == ex_cluster_end)
4222 ee_start += ee_len - 1;
4223 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4224 map->m_len = min(map->m_len,
4225 (unsigned) sbi->s_cluster_ratio - c_offset);
4227 * Check for and handle this case:
4229 * |--------- cluster # N-------------|
4230 * |------- extent ----|
4231 * |--- requested region ---|
4235 if (map->m_lblk < ee_block)
4236 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4239 * Check for the case where there is already another allocated
4240 * block to the right of 'ex' but before the end of the cluster.
4242 * |------------- cluster # N-------------|
4243 * |----- ex -----| |---- ex_right ----|
4244 * |------ requested region ------|
4245 * |================|
4247 if (map->m_lblk > ee_block) {
4248 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4249 map->m_len = min(map->m_len, next - map->m_lblk);
4252 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4256 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4262 * Block allocation/map/preallocation routine for extents based files
4265 * Need to be called with
4266 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4267 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4269 * return > 0, number of of blocks already mapped/allocated
4270 * if create == 0 and these are pre-allocated blocks
4271 * buffer head is unmapped
4272 * otherwise blocks are mapped
4274 * return = 0, if plain look up failed (blocks have not been allocated)
4275 * buffer head is unmapped
4277 * return < 0, error case.
4279 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4280 struct ext4_map_blocks *map, int flags)
4282 struct ext4_ext_path *path = NULL;
4283 struct ext4_extent newex, *ex, *ex2;
4284 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4285 ext4_fsblk_t newblock = 0;
4286 int free_on_err = 0, err = 0, depth, ret;
4287 unsigned int allocated = 0, offset = 0;
4288 unsigned int allocated_clusters = 0;
4289 struct ext4_allocation_request ar;
4290 ext4_io_end_t *io = ext4_inode_aio(inode);
4291 ext4_lblk_t cluster_offset;
4292 int set_unwritten = 0;
4293 bool map_from_cluster = false;
4295 ext_debug("blocks %u/%u requested for inode %lu\n",
4296 map->m_lblk, map->m_len, inode->i_ino);
4297 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4299 /* find extent for this block */
4300 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4302 err = PTR_ERR(path);
4307 depth = ext_depth(inode);
4310 * consistent leaf must not be empty;
4311 * this situation is possible, though, _during_ tree modification;
4312 * this is why assert can't be put in ext4_find_extent()
4314 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4315 EXT4_ERROR_INODE(inode, "bad extent address "
4316 "lblock: %lu, depth: %d pblock %lld",
4317 (unsigned long) map->m_lblk, depth,
4318 path[depth].p_block);
4319 err = -EFSCORRUPTED;
4323 ex = path[depth].p_ext;
4325 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4326 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4327 unsigned short ee_len;
4331 * unwritten extents are treated as holes, except that
4332 * we split out initialized portions during a write.
4334 ee_len = ext4_ext_get_actual_len(ex);
4336 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4338 /* if found extent covers block, simply return it */
4339 if (in_range(map->m_lblk, ee_block, ee_len)) {
4340 newblock = map->m_lblk - ee_block + ee_start;
4341 /* number of remaining blocks in the extent */
4342 allocated = ee_len - (map->m_lblk - ee_block);
4343 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4344 ee_block, ee_len, newblock);
4347 * If the extent is initialized check whether the
4348 * caller wants to convert it to unwritten.
4350 if ((!ext4_ext_is_unwritten(ex)) &&
4351 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4352 allocated = convert_initialized_extent(
4353 handle, inode, map, &path,
4354 flags, allocated, newblock);
4356 } else if (!ext4_ext_is_unwritten(ex))
4359 ret = ext4_ext_handle_unwritten_extents(
4360 handle, inode, map, &path, flags,
4361 allocated, newblock);
4371 * requested block isn't allocated yet;
4372 * we couldn't try to create block if create flag is zero
4374 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4376 * put just found gap into cache to speed up
4377 * subsequent requests
4379 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4384 * Okay, we need to do block allocation.
4386 newex.ee_block = cpu_to_le32(map->m_lblk);
4387 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4390 * If we are doing bigalloc, check to see if the extent returned
4391 * by ext4_find_extent() implies a cluster we can use.
4393 if (cluster_offset && ex &&
4394 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4395 ar.len = allocated = map->m_len;
4396 newblock = map->m_pblk;
4397 map_from_cluster = true;
4398 goto got_allocated_blocks;
4401 /* find neighbour allocated blocks */
4402 ar.lleft = map->m_lblk;
4403 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4406 ar.lright = map->m_lblk;
4408 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4412 /* Check if the extent after searching to the right implies a
4413 * cluster we can use. */
4414 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4415 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4416 ar.len = allocated = map->m_len;
4417 newblock = map->m_pblk;
4418 map_from_cluster = true;
4419 goto got_allocated_blocks;
4423 * See if request is beyond maximum number of blocks we can have in
4424 * a single extent. For an initialized extent this limit is
4425 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4426 * EXT_UNWRITTEN_MAX_LEN.
4428 if (map->m_len > EXT_INIT_MAX_LEN &&
4429 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4430 map->m_len = EXT_INIT_MAX_LEN;
4431 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4432 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4433 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4435 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4436 newex.ee_len = cpu_to_le16(map->m_len);
4437 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4439 allocated = ext4_ext_get_actual_len(&newex);
4441 allocated = map->m_len;
4443 /* allocate new block */
4445 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4446 ar.logical = map->m_lblk;
4448 * We calculate the offset from the beginning of the cluster
4449 * for the logical block number, since when we allocate a
4450 * physical cluster, the physical block should start at the
4451 * same offset from the beginning of the cluster. This is
4452 * needed so that future calls to get_implied_cluster_alloc()
4455 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4456 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4458 ar.logical -= offset;
4459 if (S_ISREG(inode->i_mode))
4460 ar.flags = EXT4_MB_HINT_DATA;
4462 /* disable in-core preallocation for non-regular files */
4464 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4465 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4466 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4467 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4468 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4469 ar.flags |= EXT4_MB_USE_RESERVED;
4470 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4473 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4474 ar.goal, newblock, allocated);
4476 allocated_clusters = ar.len;
4477 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4478 if (ar.len > allocated)
4481 got_allocated_blocks:
4482 /* try to insert new extent into found leaf and return */
4483 ext4_ext_store_pblock(&newex, newblock + offset);
4484 newex.ee_len = cpu_to_le16(ar.len);
4485 /* Mark unwritten */
4486 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4487 ext4_ext_mark_unwritten(&newex);
4488 map->m_flags |= EXT4_MAP_UNWRITTEN;
4490 * io_end structure was created for every IO write to an
4491 * unwritten extent. To avoid unnecessary conversion,
4492 * here we flag the IO that really needs the conversion.
4493 * For non asycn direct IO case, flag the inode state
4494 * that we need to perform conversion when IO is done.
4496 if (flags & EXT4_GET_BLOCKS_PRE_IO)
4501 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4502 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4505 err = ext4_ext_insert_extent(handle, inode, &path,
4508 if (!err && set_unwritten) {
4510 ext4_set_io_unwritten_flag(inode, io);
4512 ext4_set_inode_state(inode,
4513 EXT4_STATE_DIO_UNWRITTEN);
4516 if (err && free_on_err) {
4517 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4518 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4519 /* free data blocks we just allocated */
4520 /* not a good idea to call discard here directly,
4521 * but otherwise we'd need to call it every free() */
4522 ext4_discard_preallocations(inode);
4523 ext4_free_blocks(handle, inode, NULL, newblock,
4524 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4528 /* previous routine could use block we allocated */
4529 newblock = ext4_ext_pblock(&newex);
4530 allocated = ext4_ext_get_actual_len(&newex);
4531 if (allocated > map->m_len)
4532 allocated = map->m_len;
4533 map->m_flags |= EXT4_MAP_NEW;
4536 * Update reserved blocks/metadata blocks after successful
4537 * block allocation which had been deferred till now.
4539 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4540 unsigned int reserved_clusters;
4542 * Check how many clusters we had reserved this allocated range
4544 reserved_clusters = get_reserved_cluster_alloc(inode,
4545 map->m_lblk, allocated);
4546 if (!map_from_cluster) {
4547 BUG_ON(allocated_clusters < reserved_clusters);
4548 if (reserved_clusters < allocated_clusters) {
4549 struct ext4_inode_info *ei = EXT4_I(inode);
4550 int reservation = allocated_clusters -
4553 * It seems we claimed few clusters outside of
4554 * the range of this allocation. We should give
4555 * it back to the reservation pool. This can
4556 * happen in the following case:
4558 * * Suppose s_cluster_ratio is 4 (i.e., each
4559 * cluster has 4 blocks. Thus, the clusters
4560 * are [0-3],[4-7],[8-11]...
4561 * * First comes delayed allocation write for
4562 * logical blocks 10 & 11. Since there were no
4563 * previous delayed allocated blocks in the
4564 * range [8-11], we would reserve 1 cluster
4566 * * Next comes write for logical blocks 3 to 8.
4567 * In this case, we will reserve 2 clusters
4568 * (for [0-3] and [4-7]; and not for [8-11] as
4569 * that range has a delayed allocated blocks.
4570 * Thus total reserved clusters now becomes 3.
4571 * * Now, during the delayed allocation writeout
4572 * time, we will first write blocks [3-8] and
4573 * allocate 3 clusters for writing these
4574 * blocks. Also, we would claim all these
4575 * three clusters above.
4576 * * Now when we come here to writeout the
4577 * blocks [10-11], we would expect to claim
4578 * the reservation of 1 cluster we had made
4579 * (and we would claim it since there are no
4580 * more delayed allocated blocks in the range
4581 * [8-11]. But our reserved cluster count had
4582 * already gone to 0.
4584 * Thus, at the step 4 above when we determine
4585 * that there are still some unwritten delayed
4586 * allocated blocks outside of our current
4587 * block range, we should increment the
4588 * reserved clusters count so that when the
4589 * remaining blocks finally gets written, we
4592 dquot_reserve_block(inode,
4593 EXT4_C2B(sbi, reservation));
4594 spin_lock(&ei->i_block_reservation_lock);
4595 ei->i_reserved_data_blocks += reservation;
4596 spin_unlock(&ei->i_block_reservation_lock);
4599 * We will claim quota for all newly allocated blocks.
4600 * We're updating the reserved space *after* the
4601 * correction above so we do not accidentally free
4602 * all the metadata reservation because we might
4603 * actually need it later on.
4605 ext4_da_update_reserve_space(inode, allocated_clusters,
4611 * Cache the extent and update transaction to commit on fdatasync only
4612 * when it is _not_ an unwritten extent.
4614 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4615 ext4_update_inode_fsync_trans(handle, inode, 1);
4617 ext4_update_inode_fsync_trans(handle, inode, 0);
4619 if (allocated > map->m_len)
4620 allocated = map->m_len;
4621 ext4_ext_show_leaf(inode, path);
4622 map->m_flags |= EXT4_MAP_MAPPED;
4623 map->m_pblk = newblock;
4624 map->m_len = allocated;
4626 ext4_ext_drop_refs(path);
4629 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4630 err ? err : allocated);
4631 return err ? err : allocated;
4634 void ext4_ext_truncate(handle_t *handle, struct inode *inode)
4636 struct super_block *sb = inode->i_sb;
4637 ext4_lblk_t last_block;
4641 * TODO: optimization is possible here.
4642 * Probably we need not scan at all,
4643 * because page truncation is enough.
4646 /* we have to know where to truncate from in crash case */
4647 EXT4_I(inode)->i_disksize = inode->i_size;
4648 ext4_mark_inode_dirty(handle, inode);
4650 last_block = (inode->i_size + sb->s_blocksize - 1)
4651 >> EXT4_BLOCK_SIZE_BITS(sb);
4653 err = ext4_es_remove_extent(inode, last_block,
4654 EXT_MAX_BLOCKS - last_block);
4655 if (err == -ENOMEM) {
4657 congestion_wait(BLK_RW_ASYNC, HZ/50);
4661 ext4_std_error(inode->i_sb, err);
4664 err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4665 ext4_std_error(inode->i_sb, err);
4668 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4669 ext4_lblk_t len, loff_t new_size,
4670 int flags, int mode)
4672 struct inode *inode = file_inode(file);
4678 struct ext4_map_blocks map;
4679 unsigned int credits;
4682 map.m_lblk = offset;
4685 * Don't normalize the request if it can fit in one extent so
4686 * that it doesn't get unnecessarily split into multiple
4689 if (len <= EXT_UNWRITTEN_MAX_LEN)
4690 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4693 * credits to insert 1 extent into extent tree
4695 credits = ext4_chunk_trans_blocks(inode, len);
4697 * We can only call ext_depth() on extent based inodes
4699 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4700 depth = ext_depth(inode);
4705 while (ret >= 0 && len) {
4707 * Recalculate credits when extent tree depth changes.
4709 if (depth >= 0 && depth != ext_depth(inode)) {
4710 credits = ext4_chunk_trans_blocks(inode, len);
4711 depth = ext_depth(inode);
4714 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4716 if (IS_ERR(handle)) {
4717 ret = PTR_ERR(handle);
4720 ret = ext4_map_blocks(handle, inode, &map, flags);
4722 ext4_debug("inode #%lu: block %u: len %u: "
4723 "ext4_ext_map_blocks returned %d",
4724 inode->i_ino, map.m_lblk,
4726 ext4_mark_inode_dirty(handle, inode);
4727 ret2 = ext4_journal_stop(handle);
4731 map.m_len = len = len - ret;
4732 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4733 inode->i_ctime = ext4_current_time(inode);
4735 if (epos > new_size)
4737 if (ext4_update_inode_size(inode, epos) & 0x1)
4738 inode->i_mtime = inode->i_ctime;
4740 if (epos > inode->i_size)
4741 ext4_set_inode_flag(inode,
4742 EXT4_INODE_EOFBLOCKS);
4744 ext4_mark_inode_dirty(handle, inode);
4745 ret2 = ext4_journal_stop(handle);
4749 if (ret == -ENOSPC &&
4750 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4755 return ret > 0 ? ret2 : ret;
4758 static long ext4_zero_range(struct file *file, loff_t offset,
4759 loff_t len, int mode)
4761 struct inode *inode = file_inode(file);
4762 handle_t *handle = NULL;
4763 unsigned int max_blocks;
4764 loff_t new_size = 0;
4768 int partial_begin, partial_end;
4771 unsigned int blkbits = inode->i_blkbits;
4773 trace_ext4_zero_range(inode, offset, len, mode);
4775 if (!S_ISREG(inode->i_mode))
4778 /* Call ext4_force_commit to flush all data in case of data=journal. */
4779 if (ext4_should_journal_data(inode)) {
4780 ret = ext4_force_commit(inode->i_sb);
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 mutex_lock(&inode->i_mutex);
4809 * Indirect files do not support unwritten extnets
4811 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4816 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4817 offset + len > i_size_read(inode)) {
4818 new_size = offset + len;
4819 ret = inode_newsize_ok(inode, new_size);
4824 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4825 if (mode & FALLOC_FL_KEEP_SIZE)
4826 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4828 /* Wait all existing dio workers, newcomers will block on i_mutex */
4829 ext4_inode_block_unlocked_dio(inode);
4830 inode_dio_wait(inode);
4832 /* Preallocate the range including the unaligned edges */
4833 if (partial_begin || partial_end) {
4834 ret = ext4_alloc_file_blocks(file,
4835 round_down(offset, 1 << blkbits) >> blkbits,
4836 (round_up((offset + len), 1 << blkbits) -
4837 round_down(offset, 1 << blkbits)) >> blkbits,
4838 new_size, flags, mode);
4844 /* Zero range excluding the unaligned edges */
4845 if (max_blocks > 0) {
4846 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4850 * Prevent page faults from reinstantiating pages we have
4851 * released from page cache.
4853 down_write(&EXT4_I(inode)->i_mmap_sem);
4854 ret = ext4_update_disksize_before_punch(inode, offset, len);
4856 up_write(&EXT4_I(inode)->i_mmap_sem);
4859 /* Now release the pages and zero block aligned part of pages */
4860 truncate_pagecache_range(inode, start, end - 1);
4861 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4863 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4865 up_write(&EXT4_I(inode)->i_mmap_sem);
4869 if (!partial_begin && !partial_end)
4873 * In worst case we have to writeout two nonadjacent unwritten
4874 * blocks and update the inode
4876 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4877 if (ext4_should_journal_data(inode))
4879 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4880 if (IS_ERR(handle)) {
4881 ret = PTR_ERR(handle);
4882 ext4_std_error(inode->i_sb, ret);
4886 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4888 ext4_update_inode_size(inode, new_size);
4891 * Mark that we allocate beyond EOF so the subsequent truncate
4892 * can proceed even if the new size is the same as i_size.
4894 if ((offset + len) > i_size_read(inode))
4895 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4897 ext4_mark_inode_dirty(handle, inode);
4899 /* Zero out partial block at the edges of the range */
4900 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4902 if (file->f_flags & O_SYNC)
4903 ext4_handle_sync(handle);
4905 ext4_journal_stop(handle);
4907 ext4_inode_resume_unlocked_dio(inode);
4909 mutex_unlock(&inode->i_mutex);
4914 * preallocate space for a file. This implements ext4's fallocate file
4915 * operation, which gets called from sys_fallocate system call.
4916 * For block-mapped files, posix_fallocate should fall back to the method
4917 * of writing zeroes to the required new blocks (the same behavior which is
4918 * expected for file systems which do not support fallocate() system call).
4920 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4922 struct inode *inode = file_inode(file);
4923 loff_t new_size = 0;
4924 unsigned int max_blocks;
4928 unsigned int blkbits = inode->i_blkbits;
4931 * Encrypted inodes can't handle collapse range or insert
4932 * range since we would need to re-encrypt blocks with a
4933 * different IV or XTS tweak (which are based on the logical
4936 * XXX It's not clear why zero range isn't working, but we'll
4937 * leave it disabled for encrypted inodes for now. This is a
4938 * bug we should fix....
4940 if (ext4_encrypted_inode(inode) &&
4941 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4942 FALLOC_FL_ZERO_RANGE)))
4945 /* Return error if mode is not supported */
4946 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4947 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4948 FALLOC_FL_INSERT_RANGE))
4951 if (mode & FALLOC_FL_PUNCH_HOLE)
4952 return ext4_punch_hole(inode, offset, len);
4954 ret = ext4_convert_inline_data(inode);
4958 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4959 return ext4_collapse_range(inode, offset, len);
4961 if (mode & FALLOC_FL_INSERT_RANGE)
4962 return ext4_insert_range(inode, offset, len);
4964 if (mode & FALLOC_FL_ZERO_RANGE)
4965 return ext4_zero_range(file, offset, len, mode);
4967 trace_ext4_fallocate_enter(inode, offset, len, mode);
4968 lblk = offset >> blkbits;
4970 * We can't just convert len to max_blocks because
4971 * If blocksize = 4096 offset = 3072 and len = 2048
4973 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4976 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4977 if (mode & FALLOC_FL_KEEP_SIZE)
4978 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4980 mutex_lock(&inode->i_mutex);
4983 * We only support preallocation for extent-based files only
4985 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4990 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4991 offset + len > i_size_read(inode)) {
4992 new_size = offset + len;
4993 ret = inode_newsize_ok(inode, new_size);
4998 /* Wait all existing dio workers, newcomers will block on i_mutex */
4999 ext4_inode_block_unlocked_dio(inode);
5000 inode_dio_wait(inode);
5002 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
5004 ext4_inode_resume_unlocked_dio(inode);
5008 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
5009 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5010 EXT4_I(inode)->i_sync_tid);
5013 mutex_unlock(&inode->i_mutex);
5014 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
5019 * This function convert a range of blocks to written extents
5020 * The caller of this function will pass the start offset and the size.
5021 * all unwritten extents within this range will be converted to
5024 * This function is called from the direct IO end io call back
5025 * function, to convert the fallocated extents after IO is completed.
5026 * Returns 0 on success.
5028 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
5029 loff_t offset, ssize_t len)
5031 unsigned int max_blocks;
5034 struct ext4_map_blocks map;
5035 unsigned int credits, blkbits = inode->i_blkbits;
5037 map.m_lblk = offset >> blkbits;
5039 * We can't just convert len to max_blocks because
5040 * If blocksize = 4096 offset = 3072 and len = 2048
5042 max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
5045 * This is somewhat ugly but the idea is clear: When transaction is
5046 * reserved, everything goes into it. Otherwise we rather start several
5047 * smaller transactions for conversion of each extent separately.
5050 handle = ext4_journal_start_reserved(handle,
5051 EXT4_HT_EXT_CONVERT);
5053 return PTR_ERR(handle);
5057 * credits to insert 1 extent into extent tree
5059 credits = ext4_chunk_trans_blocks(inode, max_blocks);
5061 while (ret >= 0 && ret < max_blocks) {
5063 map.m_len = (max_blocks -= ret);
5065 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
5067 if (IS_ERR(handle)) {
5068 ret = PTR_ERR(handle);
5072 ret = ext4_map_blocks(handle, inode, &map,
5073 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5075 ext4_warning(inode->i_sb,
5076 "inode #%lu: block %u: len %u: "
5077 "ext4_ext_map_blocks returned %d",
5078 inode->i_ino, map.m_lblk,
5080 ext4_mark_inode_dirty(handle, inode);
5082 ret2 = ext4_journal_stop(handle);
5083 if (ret <= 0 || ret2)
5087 ret2 = ext4_journal_stop(handle);
5088 return ret > 0 ? ret2 : ret;
5092 * If newes is not existing extent (newes->ec_pblk equals zero) find
5093 * delayed extent at start of newes and update newes accordingly and
5094 * return start of the next delayed extent.
5096 * If newes is existing extent (newes->ec_pblk is not equal zero)
5097 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5098 * extent found. Leave newes unmodified.
5100 static int ext4_find_delayed_extent(struct inode *inode,
5101 struct extent_status *newes)
5103 struct extent_status es;
5104 ext4_lblk_t block, next_del;
5106 if (newes->es_pblk == 0) {
5107 ext4_es_find_delayed_extent_range(inode, newes->es_lblk,
5108 newes->es_lblk + newes->es_len - 1, &es);
5111 * No extent in extent-tree contains block @newes->es_pblk,
5112 * then the block may stay in 1)a hole or 2)delayed-extent.
5118 if (es.es_lblk > newes->es_lblk) {
5120 newes->es_len = min(es.es_lblk - newes->es_lblk,
5125 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5128 block = newes->es_lblk + newes->es_len;
5129 ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es);
5131 next_del = EXT_MAX_BLOCKS;
5133 next_del = es.es_lblk;
5137 /* fiemap flags we can handle specified here */
5138 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5140 static int ext4_xattr_fiemap(struct inode *inode,
5141 struct fiemap_extent_info *fieinfo)
5145 __u32 flags = FIEMAP_EXTENT_LAST;
5146 int blockbits = inode->i_sb->s_blocksize_bits;
5150 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5151 struct ext4_iloc iloc;
5152 int offset; /* offset of xattr in inode */
5154 error = ext4_get_inode_loc(inode, &iloc);
5157 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5158 offset = EXT4_GOOD_OLD_INODE_SIZE +
5159 EXT4_I(inode)->i_extra_isize;
5161 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5162 flags |= FIEMAP_EXTENT_DATA_INLINE;
5164 } else { /* external block */
5165 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5166 length = inode->i_sb->s_blocksize;
5170 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5172 return (error < 0 ? error : 0);
5175 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5176 __u64 start, __u64 len)
5178 ext4_lblk_t start_blk;
5181 if (ext4_has_inline_data(inode)) {
5184 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5191 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5192 error = ext4_ext_precache(inode);
5197 /* fallback to generic here if not in extents fmt */
5198 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5199 return generic_block_fiemap(inode, fieinfo, start, len,
5202 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5205 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5206 error = ext4_xattr_fiemap(inode, fieinfo);
5208 ext4_lblk_t len_blks;
5211 start_blk = start >> inode->i_sb->s_blocksize_bits;
5212 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5213 if (last_blk >= EXT_MAX_BLOCKS)
5214 last_blk = EXT_MAX_BLOCKS-1;
5215 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5218 * Walk the extent tree gathering extent information
5219 * and pushing extents back to the user.
5221 error = ext4_fill_fiemap_extents(inode, start_blk,
5229 * Function to access the path buffer for marking it dirty.
5230 * It also checks if there are sufficient credits left in the journal handle
5234 ext4_access_path(handle_t *handle, struct inode *inode,
5235 struct ext4_ext_path *path)
5239 if (!ext4_handle_valid(handle))
5243 * Check if need to extend journal credits
5244 * 3 for leaf, sb, and inode plus 2 (bmap and group
5245 * descriptor) for each block group; assume two block
5248 if (handle->h_buffer_credits < 7) {
5249 credits = ext4_writepage_trans_blocks(inode);
5250 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5251 /* EAGAIN is success */
5252 if (err && err != -EAGAIN)
5256 err = ext4_ext_get_access(handle, inode, path);
5261 * ext4_ext_shift_path_extents:
5262 * Shift the extents of a path structure lying between path[depth].p_ext
5263 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5264 * if it is right shift or left shift operation.
5267 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5268 struct inode *inode, handle_t *handle,
5269 enum SHIFT_DIRECTION SHIFT)
5272 struct ext4_extent *ex_start, *ex_last;
5274 depth = path->p_depth;
5276 while (depth >= 0) {
5277 if (depth == path->p_depth) {
5278 ex_start = path[depth].p_ext;
5280 return -EFSCORRUPTED;
5282 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5284 err = ext4_access_path(handle, inode, path + depth);
5288 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5291 while (ex_start <= ex_last) {
5292 if (SHIFT == SHIFT_LEFT) {
5293 le32_add_cpu(&ex_start->ee_block,
5295 /* Try to merge to the left. */
5297 EXT_FIRST_EXTENT(path[depth].p_hdr))
5299 ext4_ext_try_to_merge_right(inode,
5300 path, ex_start - 1))
5305 le32_add_cpu(&ex_last->ee_block, shift);
5306 ext4_ext_try_to_merge_right(inode, path,
5311 err = ext4_ext_dirty(handle, inode, path + depth);
5315 if (--depth < 0 || !update)
5319 /* Update index too */
5320 err = ext4_access_path(handle, inode, path + depth);
5324 if (SHIFT == SHIFT_LEFT)
5325 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5327 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5328 err = ext4_ext_dirty(handle, inode, path + depth);
5332 /* we are done if current index is not a starting index */
5333 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5344 * ext4_ext_shift_extents:
5345 * All the extents which lies in the range from @start to the last allocated
5346 * block for the @inode are shifted either towards left or right (depending
5347 * upon @SHIFT) by @shift blocks.
5348 * On success, 0 is returned, error otherwise.
5351 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5352 ext4_lblk_t start, ext4_lblk_t shift,
5353 enum SHIFT_DIRECTION SHIFT)
5355 struct ext4_ext_path *path;
5357 struct ext4_extent *extent;
5358 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5360 /* Let path point to the last extent */
5361 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
5363 return PTR_ERR(path);
5365 depth = path->p_depth;
5366 extent = path[depth].p_ext;
5370 stop = le32_to_cpu(extent->ee_block) +
5371 ext4_ext_get_actual_len(extent);
5374 * In case of left shift, Don't start shifting extents until we make
5375 * sure the hole is big enough to accommodate the shift.
5377 if (SHIFT == SHIFT_LEFT) {
5378 path = ext4_find_extent(inode, start - 1, &path, 0);
5380 return PTR_ERR(path);
5381 depth = path->p_depth;
5382 extent = path[depth].p_ext;
5384 ex_start = le32_to_cpu(extent->ee_block);
5385 ex_end = le32_to_cpu(extent->ee_block) +
5386 ext4_ext_get_actual_len(extent);
5392 if ((start == ex_start && shift > ex_start) ||
5393 (shift > start - ex_end)) {
5394 ext4_ext_drop_refs(path);
5401 * In case of left shift, iterator points to start and it is increased
5402 * till we reach stop. In case of right shift, iterator points to stop
5403 * and it is decreased till we reach start.
5405 if (SHIFT == SHIFT_LEFT)
5410 /* Its safe to start updating extents */
5411 while (start < stop) {
5412 path = ext4_find_extent(inode, *iterator, &path, 0);
5414 return PTR_ERR(path);
5415 depth = path->p_depth;
5416 extent = path[depth].p_ext;
5418 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5419 (unsigned long) *iterator);
5420 return -EFSCORRUPTED;
5422 if (SHIFT == SHIFT_LEFT && *iterator >
5423 le32_to_cpu(extent->ee_block)) {
5424 /* Hole, move to the next extent */
5425 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5426 path[depth].p_ext++;
5428 *iterator = ext4_ext_next_allocated_block(path);
5433 if (SHIFT == SHIFT_LEFT) {
5434 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5435 *iterator = le32_to_cpu(extent->ee_block) +
5436 ext4_ext_get_actual_len(extent);
5438 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5439 *iterator = le32_to_cpu(extent->ee_block) > 0 ?
5440 le32_to_cpu(extent->ee_block) - 1 : 0;
5441 /* Update path extent in case we need to stop */
5442 while (le32_to_cpu(extent->ee_block) < start)
5444 path[depth].p_ext = extent;
5446 ret = ext4_ext_shift_path_extents(path, shift, inode,
5452 ext4_ext_drop_refs(path);
5458 * ext4_collapse_range:
5459 * This implements the fallocate's collapse range functionality for ext4
5460 * Returns: 0 and non-zero on error.
5462 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5464 struct super_block *sb = inode->i_sb;
5465 ext4_lblk_t punch_start, punch_stop;
5467 unsigned int credits;
5468 loff_t new_size, ioffset;
5472 * We need to test this early because xfstests assumes that a
5473 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5474 * system does not support collapse range.
5476 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5479 /* Collapse range works only on fs block size aligned offsets. */
5480 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5481 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5484 if (!S_ISREG(inode->i_mode))
5487 trace_ext4_collapse_range(inode, offset, len);
5489 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5490 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5492 /* Call ext4_force_commit to flush all data in case of data=journal. */
5493 if (ext4_should_journal_data(inode)) {
5494 ret = ext4_force_commit(inode->i_sb);
5499 mutex_lock(&inode->i_mutex);
5501 * There is no need to overlap collapse range with EOF, in which case
5502 * it is effectively a truncate operation
5504 if (offset + len >= i_size_read(inode)) {
5509 /* Currently just for extent based files */
5510 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5515 /* Wait for existing dio to complete */
5516 ext4_inode_block_unlocked_dio(inode);
5517 inode_dio_wait(inode);
5520 * Prevent page faults from reinstantiating pages we have released from
5523 down_write(&EXT4_I(inode)->i_mmap_sem);
5525 * Need to round down offset to be aligned with page size boundary
5526 * for page size > block size.
5528 ioffset = round_down(offset, PAGE_SIZE);
5530 * Write tail of the last page before removed range since it will get
5531 * removed from the page cache below.
5533 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5537 * Write data that will be shifted to preserve them when discarding
5538 * page cache below. We are also protected from pages becoming dirty
5541 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5545 truncate_pagecache(inode, ioffset);
5547 credits = ext4_writepage_trans_blocks(inode);
5548 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5549 if (IS_ERR(handle)) {
5550 ret = PTR_ERR(handle);
5554 down_write(&EXT4_I(inode)->i_data_sem);
5555 ext4_discard_preallocations(inode);
5557 ret = ext4_es_remove_extent(inode, punch_start,
5558 EXT_MAX_BLOCKS - punch_start);
5560 up_write(&EXT4_I(inode)->i_data_sem);
5564 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5566 up_write(&EXT4_I(inode)->i_data_sem);
5569 ext4_discard_preallocations(inode);
5571 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5572 punch_stop - punch_start, SHIFT_LEFT);
5574 up_write(&EXT4_I(inode)->i_data_sem);
5578 new_size = i_size_read(inode) - len;
5579 i_size_write(inode, new_size);
5580 EXT4_I(inode)->i_disksize = new_size;
5582 up_write(&EXT4_I(inode)->i_data_sem);
5584 ext4_handle_sync(handle);
5585 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5586 ext4_mark_inode_dirty(handle, inode);
5589 ext4_journal_stop(handle);
5591 up_write(&EXT4_I(inode)->i_mmap_sem);
5592 ext4_inode_resume_unlocked_dio(inode);
5594 mutex_unlock(&inode->i_mutex);
5599 * ext4_insert_range:
5600 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5601 * The data blocks starting from @offset to the EOF are shifted by @len
5602 * towards right to create a hole in the @inode. Inode size is increased
5604 * Returns 0 on success, error otherwise.
5606 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5608 struct super_block *sb = inode->i_sb;
5610 struct ext4_ext_path *path;
5611 struct ext4_extent *extent;
5612 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5613 unsigned int credits, ee_len;
5614 int ret = 0, depth, split_flag = 0;
5618 * We need to test this early because xfstests assumes that an
5619 * insert range of (0, 1) will return EOPNOTSUPP if the file
5620 * system does not support insert range.
5622 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5625 /* Insert range works only on fs block size aligned offsets. */
5626 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5627 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5630 if (!S_ISREG(inode->i_mode))
5633 trace_ext4_insert_range(inode, offset, len);
5635 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5636 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5638 /* Call ext4_force_commit to flush all data in case of data=journal */
5639 if (ext4_should_journal_data(inode)) {
5640 ret = ext4_force_commit(inode->i_sb);
5645 mutex_lock(&inode->i_mutex);
5646 /* Currently just for extent based files */
5647 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5652 /* Check for wrap through zero */
5653 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5658 /* Offset should be less than i_size */
5659 if (offset >= i_size_read(inode)) {
5664 /* Wait for existing dio to complete */
5665 ext4_inode_block_unlocked_dio(inode);
5666 inode_dio_wait(inode);
5669 * Prevent page faults from reinstantiating pages we have released from
5672 down_write(&EXT4_I(inode)->i_mmap_sem);
5674 * Need to round down to align start offset to page size boundary
5675 * for page size > block size.
5677 ioffset = round_down(offset, PAGE_SIZE);
5678 /* Write out all dirty pages */
5679 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5683 truncate_pagecache(inode, ioffset);
5685 credits = ext4_writepage_trans_blocks(inode);
5686 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5687 if (IS_ERR(handle)) {
5688 ret = PTR_ERR(handle);
5692 /* Expand file to avoid data loss if there is error while shifting */
5693 inode->i_size += len;
5694 EXT4_I(inode)->i_disksize += len;
5695 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5696 ret = ext4_mark_inode_dirty(handle, inode);
5700 down_write(&EXT4_I(inode)->i_data_sem);
5701 ext4_discard_preallocations(inode);
5703 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5705 up_write(&EXT4_I(inode)->i_data_sem);
5709 depth = ext_depth(inode);
5710 extent = path[depth].p_ext;
5712 ee_start_lblk = le32_to_cpu(extent->ee_block);
5713 ee_len = ext4_ext_get_actual_len(extent);
5716 * If offset_lblk is not the starting block of extent, split
5717 * the extent @offset_lblk
5719 if ((offset_lblk > ee_start_lblk) &&
5720 (offset_lblk < (ee_start_lblk + ee_len))) {
5721 if (ext4_ext_is_unwritten(extent))
5722 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5723 EXT4_EXT_MARK_UNWRIT2;
5724 ret = ext4_split_extent_at(handle, inode, &path,
5725 offset_lblk, split_flag,
5727 EXT4_GET_BLOCKS_PRE_IO |
5728 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5731 ext4_ext_drop_refs(path);
5734 up_write(&EXT4_I(inode)->i_data_sem);
5739 ret = ext4_es_remove_extent(inode, offset_lblk,
5740 EXT_MAX_BLOCKS - offset_lblk);
5742 up_write(&EXT4_I(inode)->i_data_sem);
5747 * if offset_lblk lies in a hole which is at start of file, use
5748 * ee_start_lblk to shift extents
5750 ret = ext4_ext_shift_extents(inode, handle,
5751 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5752 len_lblk, SHIFT_RIGHT);
5754 up_write(&EXT4_I(inode)->i_data_sem);
5756 ext4_handle_sync(handle);
5759 ext4_journal_stop(handle);
5761 up_write(&EXT4_I(inode)->i_mmap_sem);
5762 ext4_inode_resume_unlocked_dio(inode);
5764 mutex_unlock(&inode->i_mutex);
5769 * ext4_swap_extents - Swap extents between two inodes
5771 * @inode1: First inode
5772 * @inode2: Second inode
5773 * @lblk1: Start block for first inode
5774 * @lblk2: Start block for second inode
5775 * @count: Number of blocks to swap
5776 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5777 * @erp: Pointer to save error value
5779 * This helper routine does exactly what is promise "swap extents". All other
5780 * stuff such as page-cache locking consistency, bh mapping consistency or
5781 * extent's data copying must be performed by caller.
5783 * i_mutex is held for both inodes
5784 * i_data_sem is locked for write for both inodes
5786 * All pages from requested range are locked for both inodes
5789 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5790 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5791 ext4_lblk_t count, int unwritten, int *erp)
5793 struct ext4_ext_path *path1 = NULL;
5794 struct ext4_ext_path *path2 = NULL;
5795 int replaced_count = 0;
5797 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5798 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5799 BUG_ON(!mutex_is_locked(&inode1->i_mutex));
5800 BUG_ON(!mutex_is_locked(&inode2->i_mutex));
5802 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5805 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5810 struct ext4_extent *ex1, *ex2, tmp_ex;
5811 ext4_lblk_t e1_blk, e2_blk;
5812 int e1_len, e2_len, len;
5815 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5816 if (IS_ERR(path1)) {
5817 *erp = PTR_ERR(path1);
5823 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5824 if (IS_ERR(path2)) {
5825 *erp = PTR_ERR(path2);
5829 ex1 = path1[path1->p_depth].p_ext;
5830 ex2 = path2[path2->p_depth].p_ext;
5831 /* Do we have somthing to swap ? */
5832 if (unlikely(!ex2 || !ex1))
5835 e1_blk = le32_to_cpu(ex1->ee_block);
5836 e2_blk = le32_to_cpu(ex2->ee_block);
5837 e1_len = ext4_ext_get_actual_len(ex1);
5838 e2_len = ext4_ext_get_actual_len(ex2);
5841 if (!in_range(lblk1, e1_blk, e1_len) ||
5842 !in_range(lblk2, e2_blk, e2_len)) {
5843 ext4_lblk_t next1, next2;
5845 /* if hole after extent, then go to next extent */
5846 next1 = ext4_ext_next_allocated_block(path1);
5847 next2 = ext4_ext_next_allocated_block(path2);
5848 /* If hole before extent, then shift to that extent */
5853 /* Do we have something to swap */
5854 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5856 /* Move to the rightest boundary */
5857 len = next1 - lblk1;
5858 if (len < next2 - lblk2)
5859 len = next2 - lblk2;
5868 /* Prepare left boundary */
5869 if (e1_blk < lblk1) {
5871 *erp = ext4_force_split_extent_at(handle, inode1,
5876 if (e2_blk < lblk2) {
5878 *erp = ext4_force_split_extent_at(handle, inode2,
5883 /* ext4_split_extent_at() may result in leaf extent split,
5884 * path must to be revalidated. */
5888 /* Prepare right boundary */
5890 if (len > e1_blk + e1_len - lblk1)
5891 len = e1_blk + e1_len - lblk1;
5892 if (len > e2_blk + e2_len - lblk2)
5893 len = e2_blk + e2_len - lblk2;
5895 if (len != e1_len) {
5897 *erp = ext4_force_split_extent_at(handle, inode1,
5898 &path1, lblk1 + len, 0);
5902 if (len != e2_len) {
5904 *erp = ext4_force_split_extent_at(handle, inode2,
5905 &path2, lblk2 + len, 0);
5909 /* ext4_split_extent_at() may result in leaf extent split,
5910 * path must to be revalidated. */
5914 BUG_ON(e2_len != e1_len);
5915 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5918 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5922 /* Both extents are fully inside boundaries. Swap it now */
5924 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5925 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5926 ex1->ee_len = cpu_to_le16(e2_len);
5927 ex2->ee_len = cpu_to_le16(e1_len);
5929 ext4_ext_mark_unwritten(ex2);
5930 if (ext4_ext_is_unwritten(&tmp_ex))
5931 ext4_ext_mark_unwritten(ex1);
5933 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5934 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5935 *erp = ext4_ext_dirty(handle, inode2, path2 +
5939 *erp = ext4_ext_dirty(handle, inode1, path1 +
5942 * Looks scarry ah..? second inode already points to new blocks,
5943 * and it was successfully dirtied. But luckily error may happen
5944 * only due to journal error, so full transaction will be
5951 replaced_count += len;
5955 ext4_ext_drop_refs(path1);
5957 ext4_ext_drop_refs(path2);
5959 path1 = path2 = NULL;
5961 return replaced_count;