1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
6 * Architecture independence:
7 * Copyright (c) 2005, Bull S.A.
8 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
12 * Extents support for EXT4
15 * - ext4*_error() should be used in some situations
16 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17 * - smart tree reduction
21 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <linux/fiemap.h>
30 #include <linux/backing-dev.h>
31 #include "ext4_jbd2.h"
32 #include "ext4_extents.h"
35 #include <trace/events/ext4.h>
38 * used by extent splitting.
40 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
42 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
43 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
45 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
46 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
48 static __le32 ext4_extent_block_csum(struct inode *inode,
49 struct ext4_extent_header *eh)
51 struct ext4_inode_info *ei = EXT4_I(inode);
52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
55 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
56 EXT4_EXTENT_TAIL_OFFSET(eh));
57 return cpu_to_le32(csum);
60 static int ext4_extent_block_csum_verify(struct inode *inode,
61 struct ext4_extent_header *eh)
63 struct ext4_extent_tail *et;
65 if (!ext4_has_metadata_csum(inode->i_sb))
68 et = find_ext4_extent_tail(eh);
69 if (et->et_checksum != ext4_extent_block_csum(inode, eh))
74 static void ext4_extent_block_csum_set(struct inode *inode,
75 struct ext4_extent_header *eh)
77 struct ext4_extent_tail *et;
79 if (!ext4_has_metadata_csum(inode->i_sb))
82 et = find_ext4_extent_tail(eh);
83 et->et_checksum = ext4_extent_block_csum(inode, eh);
86 static int ext4_split_extent(handle_t *handle,
88 struct ext4_ext_path **ppath,
89 struct ext4_map_blocks *map,
93 static int ext4_split_extent_at(handle_t *handle,
95 struct ext4_ext_path **ppath,
100 static int ext4_find_delayed_extent(struct inode *inode,
101 struct extent_status *newes);
103 static int ext4_ext_truncate_extend_restart(handle_t *handle,
109 if (!ext4_handle_valid(handle))
111 if (handle->h_buffer_credits >= needed)
114 * If we need to extend the journal get a few extra blocks
115 * while we're at it for efficiency's sake.
118 err = ext4_journal_extend(handle, needed - handle->h_buffer_credits);
121 err = ext4_truncate_restart_trans(handle, inode, needed);
133 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
134 struct ext4_ext_path *path)
137 /* path points to block */
138 BUFFER_TRACE(path->p_bh, "get_write_access");
139 return ext4_journal_get_write_access(handle, path->p_bh);
141 /* path points to leaf/index in inode body */
142 /* we use in-core data, no need to protect them */
152 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
153 struct inode *inode, struct ext4_ext_path *path)
157 WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
159 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
160 /* path points to block */
161 err = __ext4_handle_dirty_metadata(where, line, handle,
164 /* path points to leaf/index in inode body */
165 err = ext4_mark_inode_dirty(handle, inode);
170 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
171 struct ext4_ext_path *path,
175 int depth = path->p_depth;
176 struct ext4_extent *ex;
179 * Try to predict block placement assuming that we are
180 * filling in a file which will eventually be
181 * non-sparse --- i.e., in the case of libbfd writing
182 * an ELF object sections out-of-order but in a way
183 * the eventually results in a contiguous object or
184 * executable file, or some database extending a table
185 * space file. However, this is actually somewhat
186 * non-ideal if we are writing a sparse file such as
187 * qemu or KVM writing a raw image file that is going
188 * to stay fairly sparse, since it will end up
189 * fragmenting the file system's free space. Maybe we
190 * should have some hueristics or some way to allow
191 * userspace to pass a hint to file system,
192 * especially if the latter case turns out to be
195 ex = path[depth].p_ext;
197 ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
198 ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
200 if (block > ext_block)
201 return ext_pblk + (block - ext_block);
203 return ext_pblk - (ext_block - block);
206 /* it looks like index is empty;
207 * try to find starting block from index itself */
208 if (path[depth].p_bh)
209 return path[depth].p_bh->b_blocknr;
212 /* OK. use inode's group */
213 return ext4_inode_to_goal_block(inode);
217 * Allocation for a meta data block
220 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
221 struct ext4_ext_path *path,
222 struct ext4_extent *ex, int *err, unsigned int flags)
224 ext4_fsblk_t goal, newblock;
226 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
227 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
232 static inline int ext4_ext_space_block(struct inode *inode, int check)
236 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
237 / sizeof(struct ext4_extent);
238 #ifdef AGGRESSIVE_TEST
239 if (!check && size > 6)
245 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
249 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
250 / sizeof(struct ext4_extent_idx);
251 #ifdef AGGRESSIVE_TEST
252 if (!check && size > 5)
258 static inline int ext4_ext_space_root(struct inode *inode, int check)
262 size = sizeof(EXT4_I(inode)->i_data);
263 size -= sizeof(struct ext4_extent_header);
264 size /= sizeof(struct ext4_extent);
265 #ifdef AGGRESSIVE_TEST
266 if (!check && size > 3)
272 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
276 size = sizeof(EXT4_I(inode)->i_data);
277 size -= sizeof(struct ext4_extent_header);
278 size /= sizeof(struct ext4_extent_idx);
279 #ifdef AGGRESSIVE_TEST
280 if (!check && size > 4)
287 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
288 struct ext4_ext_path **ppath, ext4_lblk_t lblk,
291 struct ext4_ext_path *path = *ppath;
292 int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
294 return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
295 EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
296 EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
297 (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
301 * Calculate the number of metadata blocks needed
302 * to allocate @blocks
303 * Worse case is one block per extent
305 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
307 struct ext4_inode_info *ei = EXT4_I(inode);
310 idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
311 / sizeof(struct ext4_extent_idx));
314 * If the new delayed allocation block is contiguous with the
315 * previous da block, it can share index blocks with the
316 * previous block, so we only need to allocate a new index
317 * block every idxs leaf blocks. At ldxs**2 blocks, we need
318 * an additional index block, and at ldxs**3 blocks, yet
319 * another index blocks.
321 if (ei->i_da_metadata_calc_len &&
322 ei->i_da_metadata_calc_last_lblock+1 == lblock) {
325 if ((ei->i_da_metadata_calc_len % idxs) == 0)
327 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
329 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
331 ei->i_da_metadata_calc_len = 0;
333 ei->i_da_metadata_calc_len++;
334 ei->i_da_metadata_calc_last_lblock++;
339 * In the worst case we need a new set of index blocks at
340 * every level of the inode's extent tree.
342 ei->i_da_metadata_calc_len = 1;
343 ei->i_da_metadata_calc_last_lblock = lblock;
344 return ext_depth(inode) + 1;
348 ext4_ext_max_entries(struct inode *inode, int depth)
352 if (depth == ext_depth(inode)) {
354 max = ext4_ext_space_root(inode, 1);
356 max = ext4_ext_space_root_idx(inode, 1);
359 max = ext4_ext_space_block(inode, 1);
361 max = ext4_ext_space_block_idx(inode, 1);
367 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
369 ext4_fsblk_t block = ext4_ext_pblock(ext);
370 int len = ext4_ext_get_actual_len(ext);
371 ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
376 * - overflow/wrap-around
378 if (lblock + len <= lblock)
380 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
383 static int ext4_valid_extent_idx(struct inode *inode,
384 struct ext4_extent_idx *ext_idx)
386 ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
388 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
391 static int ext4_valid_extent_entries(struct inode *inode,
392 struct ext4_extent_header *eh,
395 unsigned short entries;
396 if (eh->eh_entries == 0)
399 entries = le16_to_cpu(eh->eh_entries);
403 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
404 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
405 ext4_fsblk_t pblock = 0;
406 ext4_lblk_t lblock = 0;
407 ext4_lblk_t prev = 0;
410 if (!ext4_valid_extent(inode, ext))
413 /* Check for overlapping extents */
414 lblock = le32_to_cpu(ext->ee_block);
415 len = ext4_ext_get_actual_len(ext);
416 if ((lblock <= prev) && prev) {
417 pblock = ext4_ext_pblock(ext);
418 es->s_last_error_block = cpu_to_le64(pblock);
423 prev = lblock + len - 1;
426 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
428 if (!ext4_valid_extent_idx(inode, ext_idx))
437 static int __ext4_ext_check(const char *function, unsigned int line,
438 struct inode *inode, struct ext4_extent_header *eh,
439 int depth, ext4_fsblk_t pblk)
441 const char *error_msg;
442 int max = 0, err = -EFSCORRUPTED;
444 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
445 error_msg = "invalid magic";
448 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
449 error_msg = "unexpected eh_depth";
452 if (unlikely(eh->eh_max == 0)) {
453 error_msg = "invalid eh_max";
456 max = ext4_ext_max_entries(inode, depth);
457 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
458 error_msg = "too large eh_max";
461 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
462 error_msg = "invalid eh_entries";
465 if (!ext4_valid_extent_entries(inode, eh, depth)) {
466 error_msg = "invalid extent entries";
469 if (unlikely(depth > 32)) {
470 error_msg = "too large eh_depth";
473 /* Verify checksum on non-root extent tree nodes */
474 if (ext_depth(inode) != depth &&
475 !ext4_extent_block_csum_verify(inode, eh)) {
476 error_msg = "extent tree corrupted";
483 ext4_error_inode(inode, function, line, 0,
484 "pblk %llu bad header/extent: %s - magic %x, "
485 "entries %u, max %u(%u), depth %u(%u)",
486 (unsigned long long) pblk, error_msg,
487 le16_to_cpu(eh->eh_magic),
488 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
489 max, le16_to_cpu(eh->eh_depth), depth);
493 #define ext4_ext_check(inode, eh, depth, pblk) \
494 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
496 int ext4_ext_check_inode(struct inode *inode)
498 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
501 static struct buffer_head *
502 __read_extent_tree_block(const char *function, unsigned int line,
503 struct inode *inode, ext4_fsblk_t pblk, int depth,
506 struct buffer_head *bh;
509 bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
511 return ERR_PTR(-ENOMEM);
513 if (!bh_uptodate_or_lock(bh)) {
514 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
515 err = bh_submit_read(bh);
519 if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
521 err = __ext4_ext_check(function, line, inode,
522 ext_block_hdr(bh), depth, pblk);
525 set_buffer_verified(bh);
527 * If this is a leaf block, cache all of its entries
529 if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
530 struct ext4_extent_header *eh = ext_block_hdr(bh);
531 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
532 ext4_lblk_t prev = 0;
535 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
536 unsigned int status = EXTENT_STATUS_WRITTEN;
537 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
538 int len = ext4_ext_get_actual_len(ex);
540 if (prev && (prev != lblk))
541 ext4_es_cache_extent(inode, prev,
545 if (ext4_ext_is_unwritten(ex))
546 status = EXTENT_STATUS_UNWRITTEN;
547 ext4_es_cache_extent(inode, lblk, len,
548 ext4_ext_pblock(ex), status);
559 #define read_extent_tree_block(inode, pblk, depth, flags) \
560 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
564 * This function is called to cache a file's extent information in the
567 int ext4_ext_precache(struct inode *inode)
569 struct ext4_inode_info *ei = EXT4_I(inode);
570 struct ext4_ext_path *path = NULL;
571 struct buffer_head *bh;
572 int i = 0, depth, ret = 0;
574 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
575 return 0; /* not an extent-mapped inode */
577 down_read(&ei->i_data_sem);
578 depth = ext_depth(inode);
580 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
583 up_read(&ei->i_data_sem);
587 /* Don't cache anything if there are no external extent blocks */
590 path[0].p_hdr = ext_inode_hdr(inode);
591 ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
594 path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
597 * If this is a leaf block or we've reached the end of
598 * the index block, go up
601 path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
602 brelse(path[i].p_bh);
607 bh = read_extent_tree_block(inode,
608 ext4_idx_pblock(path[i].p_idx++),
610 EXT4_EX_FORCE_CACHE);
617 path[i].p_hdr = ext_block_hdr(bh);
618 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
620 ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
622 up_read(&ei->i_data_sem);
623 ext4_ext_drop_refs(path);
629 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
631 int k, l = path->p_depth;
634 for (k = 0; k <= l; k++, path++) {
636 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
637 ext4_idx_pblock(path->p_idx));
638 } else if (path->p_ext) {
639 ext_debug(" %d:[%d]%d:%llu ",
640 le32_to_cpu(path->p_ext->ee_block),
641 ext4_ext_is_unwritten(path->p_ext),
642 ext4_ext_get_actual_len(path->p_ext),
643 ext4_ext_pblock(path->p_ext));
650 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
652 int depth = ext_depth(inode);
653 struct ext4_extent_header *eh;
654 struct ext4_extent *ex;
660 eh = path[depth].p_hdr;
661 ex = EXT_FIRST_EXTENT(eh);
663 ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
665 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
666 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
667 ext4_ext_is_unwritten(ex),
668 ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
673 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
674 ext4_fsblk_t newblock, int level)
676 int depth = ext_depth(inode);
677 struct ext4_extent *ex;
679 if (depth != level) {
680 struct ext4_extent_idx *idx;
681 idx = path[level].p_idx;
682 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
683 ext_debug("%d: move %d:%llu in new index %llu\n", level,
684 le32_to_cpu(idx->ei_block),
685 ext4_idx_pblock(idx),
693 ex = path[depth].p_ext;
694 while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
695 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
696 le32_to_cpu(ex->ee_block),
698 ext4_ext_is_unwritten(ex),
699 ext4_ext_get_actual_len(ex),
706 #define ext4_ext_show_path(inode, path)
707 #define ext4_ext_show_leaf(inode, path)
708 #define ext4_ext_show_move(inode, path, newblock, level)
711 void ext4_ext_drop_refs(struct ext4_ext_path *path)
717 depth = path->p_depth;
718 for (i = 0; i <= depth; i++, path++)
726 * ext4_ext_binsearch_idx:
727 * binary search for the closest index of the given block
728 * the header must be checked before calling this
731 ext4_ext_binsearch_idx(struct inode *inode,
732 struct ext4_ext_path *path, ext4_lblk_t block)
734 struct ext4_extent_header *eh = path->p_hdr;
735 struct ext4_extent_idx *r, *l, *m;
738 ext_debug("binsearch for %u(idx): ", block);
740 l = EXT_FIRST_INDEX(eh) + 1;
741 r = EXT_LAST_INDEX(eh);
744 if (block < le32_to_cpu(m->ei_block))
748 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
749 m, le32_to_cpu(m->ei_block),
750 r, le32_to_cpu(r->ei_block));
754 ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
755 ext4_idx_pblock(path->p_idx));
757 #ifdef CHECK_BINSEARCH
759 struct ext4_extent_idx *chix, *ix;
762 chix = ix = EXT_FIRST_INDEX(eh);
763 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
765 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
766 printk(KERN_DEBUG "k=%d, ix=0x%p, "
768 ix, EXT_FIRST_INDEX(eh));
769 printk(KERN_DEBUG "%u <= %u\n",
770 le32_to_cpu(ix->ei_block),
771 le32_to_cpu(ix[-1].ei_block));
773 BUG_ON(k && le32_to_cpu(ix->ei_block)
774 <= le32_to_cpu(ix[-1].ei_block));
775 if (block < le32_to_cpu(ix->ei_block))
779 BUG_ON(chix != path->p_idx);
786 * ext4_ext_binsearch:
787 * binary search for closest extent of the given block
788 * the header must be checked before calling this
791 ext4_ext_binsearch(struct inode *inode,
792 struct ext4_ext_path *path, ext4_lblk_t block)
794 struct ext4_extent_header *eh = path->p_hdr;
795 struct ext4_extent *r, *l, *m;
797 if (eh->eh_entries == 0) {
799 * this leaf is empty:
800 * we get such a leaf in split/add case
805 ext_debug("binsearch for %u: ", block);
807 l = EXT_FIRST_EXTENT(eh) + 1;
808 r = EXT_LAST_EXTENT(eh);
812 if (block < le32_to_cpu(m->ee_block))
816 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
817 m, le32_to_cpu(m->ee_block),
818 r, le32_to_cpu(r->ee_block));
822 ext_debug(" -> %d:%llu:[%d]%d ",
823 le32_to_cpu(path->p_ext->ee_block),
824 ext4_ext_pblock(path->p_ext),
825 ext4_ext_is_unwritten(path->p_ext),
826 ext4_ext_get_actual_len(path->p_ext));
828 #ifdef CHECK_BINSEARCH
830 struct ext4_extent *chex, *ex;
833 chex = ex = EXT_FIRST_EXTENT(eh);
834 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
835 BUG_ON(k && le32_to_cpu(ex->ee_block)
836 <= le32_to_cpu(ex[-1].ee_block));
837 if (block < le32_to_cpu(ex->ee_block))
841 BUG_ON(chex != path->p_ext);
847 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
849 struct ext4_extent_header *eh;
851 eh = ext_inode_hdr(inode);
854 eh->eh_magic = EXT4_EXT_MAGIC;
855 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
856 ext4_mark_inode_dirty(handle, inode);
860 struct ext4_ext_path *
861 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
862 struct ext4_ext_path **orig_path, int flags)
864 struct ext4_extent_header *eh;
865 struct buffer_head *bh;
866 struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
867 short int depth, i, ppos = 0;
870 eh = ext_inode_hdr(inode);
871 depth = ext_depth(inode);
872 if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
873 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
880 ext4_ext_drop_refs(path);
881 if (depth > path[0].p_maxdepth) {
883 *orig_path = path = NULL;
887 /* account possible depth increase */
888 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
891 return ERR_PTR(-ENOMEM);
892 path[0].p_maxdepth = depth + 1;
898 /* walk through the tree */
900 ext_debug("depth %d: num %d, max %d\n",
901 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
903 ext4_ext_binsearch_idx(inode, path + ppos, block);
904 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
905 path[ppos].p_depth = i;
906 path[ppos].p_ext = NULL;
908 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
915 eh = ext_block_hdr(bh);
917 path[ppos].p_bh = bh;
918 path[ppos].p_hdr = eh;
921 path[ppos].p_depth = i;
922 path[ppos].p_ext = NULL;
923 path[ppos].p_idx = NULL;
926 ext4_ext_binsearch(inode, path + ppos, block);
927 /* if not an empty leaf */
928 if (path[ppos].p_ext)
929 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
931 ext4_ext_show_path(inode, path);
936 ext4_ext_drop_refs(path);
944 * ext4_ext_insert_index:
945 * insert new index [@logical;@ptr] into the block at @curp;
946 * check where to insert: before @curp or after @curp
948 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
949 struct ext4_ext_path *curp,
950 int logical, ext4_fsblk_t ptr)
952 struct ext4_extent_idx *ix;
955 err = ext4_ext_get_access(handle, inode, curp);
959 if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
960 EXT4_ERROR_INODE(inode,
961 "logical %d == ei_block %d!",
962 logical, le32_to_cpu(curp->p_idx->ei_block));
963 return -EFSCORRUPTED;
966 if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
967 >= le16_to_cpu(curp->p_hdr->eh_max))) {
968 EXT4_ERROR_INODE(inode,
969 "eh_entries %d >= eh_max %d!",
970 le16_to_cpu(curp->p_hdr->eh_entries),
971 le16_to_cpu(curp->p_hdr->eh_max));
972 return -EFSCORRUPTED;
975 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
977 ext_debug("insert new index %d after: %llu\n", logical, ptr);
978 ix = curp->p_idx + 1;
981 ext_debug("insert new index %d before: %llu\n", logical, ptr);
985 len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
988 ext_debug("insert new index %d: "
989 "move %d indices from 0x%p to 0x%p\n",
990 logical, len, ix, ix + 1);
991 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
994 if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
995 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
996 return -EFSCORRUPTED;
999 ix->ei_block = cpu_to_le32(logical);
1000 ext4_idx_store_pblock(ix, ptr);
1001 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1003 if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1004 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1005 return -EFSCORRUPTED;
1008 err = ext4_ext_dirty(handle, inode, curp);
1009 ext4_std_error(inode->i_sb, err);
1016 * inserts new subtree into the path, using free index entry
1018 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1019 * - makes decision where to split
1020 * - moves remaining extents and index entries (right to the split point)
1021 * into the newly allocated blocks
1022 * - initializes subtree
1024 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1026 struct ext4_ext_path *path,
1027 struct ext4_extent *newext, int at)
1029 struct buffer_head *bh = NULL;
1030 int depth = ext_depth(inode);
1031 struct ext4_extent_header *neh;
1032 struct ext4_extent_idx *fidx;
1033 int i = at, k, m, a;
1034 ext4_fsblk_t newblock, oldblock;
1036 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1038 size_t ext_size = 0;
1040 /* make decision: where to split? */
1041 /* FIXME: now decision is simplest: at current extent */
1043 /* if current leaf will be split, then we should use
1044 * border from split point */
1045 if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1046 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1047 return -EFSCORRUPTED;
1049 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1050 border = path[depth].p_ext[1].ee_block;
1051 ext_debug("leaf will be split."
1052 " next leaf starts at %d\n",
1053 le32_to_cpu(border));
1055 border = newext->ee_block;
1056 ext_debug("leaf will be added."
1057 " next leaf starts at %d\n",
1058 le32_to_cpu(border));
1062 * If error occurs, then we break processing
1063 * and mark filesystem read-only. index won't
1064 * be inserted and tree will be in consistent
1065 * state. Next mount will repair buffers too.
1069 * Get array to track all allocated blocks.
1070 * We need this to handle errors and free blocks
1073 ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1077 /* allocate all needed blocks */
1078 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1079 for (a = 0; a < depth - at; a++) {
1080 newblock = ext4_ext_new_meta_block(handle, inode, path,
1081 newext, &err, flags);
1084 ablocks[a] = newblock;
1087 /* initialize new leaf */
1088 newblock = ablocks[--a];
1089 if (unlikely(newblock == 0)) {
1090 EXT4_ERROR_INODE(inode, "newblock == 0!");
1091 err = -EFSCORRUPTED;
1094 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1095 if (unlikely(!bh)) {
1101 err = ext4_journal_get_create_access(handle, bh);
1105 neh = ext_block_hdr(bh);
1106 neh->eh_entries = 0;
1107 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1108 neh->eh_magic = EXT4_EXT_MAGIC;
1111 /* move remainder of path[depth] to the new leaf */
1112 if (unlikely(path[depth].p_hdr->eh_entries !=
1113 path[depth].p_hdr->eh_max)) {
1114 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1115 path[depth].p_hdr->eh_entries,
1116 path[depth].p_hdr->eh_max);
1117 err = -EFSCORRUPTED;
1120 /* start copy from next extent */
1121 m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1122 ext4_ext_show_move(inode, path, newblock, depth);
1124 struct ext4_extent *ex;
1125 ex = EXT_FIRST_EXTENT(neh);
1126 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1127 le16_add_cpu(&neh->eh_entries, m);
1130 /* zero out unused area in the extent block */
1131 ext_size = sizeof(struct ext4_extent_header) +
1132 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1133 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1134 ext4_extent_block_csum_set(inode, neh);
1135 set_buffer_uptodate(bh);
1138 err = ext4_handle_dirty_metadata(handle, inode, bh);
1144 /* correct old leaf */
1146 err = ext4_ext_get_access(handle, inode, path + depth);
1149 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1150 err = ext4_ext_dirty(handle, inode, path + depth);
1156 /* create intermediate indexes */
1158 if (unlikely(k < 0)) {
1159 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1160 err = -EFSCORRUPTED;
1164 ext_debug("create %d intermediate indices\n", k);
1165 /* insert new index into current index block */
1166 /* current depth stored in i var */
1169 oldblock = newblock;
1170 newblock = ablocks[--a];
1171 bh = sb_getblk(inode->i_sb, newblock);
1172 if (unlikely(!bh)) {
1178 err = ext4_journal_get_create_access(handle, bh);
1182 neh = ext_block_hdr(bh);
1183 neh->eh_entries = cpu_to_le16(1);
1184 neh->eh_magic = EXT4_EXT_MAGIC;
1185 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1186 neh->eh_depth = cpu_to_le16(depth - i);
1187 fidx = EXT_FIRST_INDEX(neh);
1188 fidx->ei_block = border;
1189 ext4_idx_store_pblock(fidx, oldblock);
1191 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1192 i, newblock, le32_to_cpu(border), oldblock);
1194 /* move remainder of path[i] to the new index block */
1195 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1196 EXT_LAST_INDEX(path[i].p_hdr))) {
1197 EXT4_ERROR_INODE(inode,
1198 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1199 le32_to_cpu(path[i].p_ext->ee_block));
1200 err = -EFSCORRUPTED;
1203 /* start copy indexes */
1204 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1205 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1206 EXT_MAX_INDEX(path[i].p_hdr));
1207 ext4_ext_show_move(inode, path, newblock, i);
1209 memmove(++fidx, path[i].p_idx,
1210 sizeof(struct ext4_extent_idx) * m);
1211 le16_add_cpu(&neh->eh_entries, m);
1213 /* zero out unused area in the extent block */
1214 ext_size = sizeof(struct ext4_extent_header) +
1215 (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1216 memset(bh->b_data + ext_size, 0,
1217 inode->i_sb->s_blocksize - ext_size);
1218 ext4_extent_block_csum_set(inode, neh);
1219 set_buffer_uptodate(bh);
1222 err = ext4_handle_dirty_metadata(handle, inode, bh);
1228 /* correct old index */
1230 err = ext4_ext_get_access(handle, inode, path + i);
1233 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1234 err = ext4_ext_dirty(handle, inode, path + i);
1242 /* insert new index */
1243 err = ext4_ext_insert_index(handle, inode, path + at,
1244 le32_to_cpu(border), newblock);
1248 if (buffer_locked(bh))
1254 /* free all allocated blocks in error case */
1255 for (i = 0; i < depth; i++) {
1258 ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1259 EXT4_FREE_BLOCKS_METADATA);
1268 * ext4_ext_grow_indepth:
1269 * implements tree growing procedure:
1270 * - allocates new block
1271 * - moves top-level data (index block or leaf) into the new block
1272 * - initializes new top-level, creating index that points to the
1273 * just created block
1275 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1278 struct ext4_extent_header *neh;
1279 struct buffer_head *bh;
1280 ext4_fsblk_t newblock, goal = 0;
1281 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1283 size_t ext_size = 0;
1285 /* Try to prepend new index to old one */
1286 if (ext_depth(inode))
1287 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1288 if (goal > le32_to_cpu(es->s_first_data_block)) {
1289 flags |= EXT4_MB_HINT_TRY_GOAL;
1292 goal = ext4_inode_to_goal_block(inode);
1293 newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1298 bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1303 err = ext4_journal_get_create_access(handle, bh);
1309 ext_size = sizeof(EXT4_I(inode)->i_data);
1310 /* move top-level index/leaf into new block */
1311 memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1312 /* zero out unused area in the extent block */
1313 memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1315 /* set size of new block */
1316 neh = ext_block_hdr(bh);
1317 /* old root could have indexes or leaves
1318 * so calculate e_max right way */
1319 if (ext_depth(inode))
1320 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1322 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1323 neh->eh_magic = EXT4_EXT_MAGIC;
1324 ext4_extent_block_csum_set(inode, neh);
1325 set_buffer_uptodate(bh);
1328 err = ext4_handle_dirty_metadata(handle, inode, bh);
1332 /* Update top-level index: num,max,pointer */
1333 neh = ext_inode_hdr(inode);
1334 neh->eh_entries = cpu_to_le16(1);
1335 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1336 if (neh->eh_depth == 0) {
1337 /* Root extent block becomes index block */
1338 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1339 EXT_FIRST_INDEX(neh)->ei_block =
1340 EXT_FIRST_EXTENT(neh)->ee_block;
1342 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1343 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1344 le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1345 ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1347 le16_add_cpu(&neh->eh_depth, 1);
1348 ext4_mark_inode_dirty(handle, inode);
1356 * ext4_ext_create_new_leaf:
1357 * finds empty index and adds new leaf.
1358 * if no free index is found, then it requests in-depth growing.
1360 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1361 unsigned int mb_flags,
1362 unsigned int gb_flags,
1363 struct ext4_ext_path **ppath,
1364 struct ext4_extent *newext)
1366 struct ext4_ext_path *path = *ppath;
1367 struct ext4_ext_path *curp;
1368 int depth, i, err = 0;
1371 i = depth = ext_depth(inode);
1373 /* walk up to the tree and look for free index entry */
1374 curp = path + depth;
1375 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1380 /* we use already allocated block for index block,
1381 * so subsequent data blocks should be contiguous */
1382 if (EXT_HAS_FREE_INDEX(curp)) {
1383 /* if we found index with free entry, then use that
1384 * entry: create all needed subtree and add new leaf */
1385 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1390 path = ext4_find_extent(inode,
1391 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1394 err = PTR_ERR(path);
1396 /* tree is full, time to grow in depth */
1397 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1402 path = ext4_find_extent(inode,
1403 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1406 err = PTR_ERR(path);
1411 * only first (depth 0 -> 1) produces free space;
1412 * in all other cases we have to split the grown tree
1414 depth = ext_depth(inode);
1415 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1416 /* now we need to split */
1426 * search the closest allocated block to the left for *logical
1427 * and returns it at @logical + it's physical address at @phys
1428 * if *logical is the smallest allocated block, the function
1429 * returns 0 at @phys
1430 * return value contains 0 (success) or error code
1432 static int ext4_ext_search_left(struct inode *inode,
1433 struct ext4_ext_path *path,
1434 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1436 struct ext4_extent_idx *ix;
1437 struct ext4_extent *ex;
1440 if (unlikely(path == NULL)) {
1441 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1442 return -EFSCORRUPTED;
1444 depth = path->p_depth;
1447 if (depth == 0 && path->p_ext == NULL)
1450 /* usually extent in the path covers blocks smaller
1451 * then *logical, but it can be that extent is the
1452 * first one in the file */
1454 ex = path[depth].p_ext;
1455 ee_len = ext4_ext_get_actual_len(ex);
1456 if (*logical < le32_to_cpu(ex->ee_block)) {
1457 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1458 EXT4_ERROR_INODE(inode,
1459 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1460 *logical, le32_to_cpu(ex->ee_block));
1461 return -EFSCORRUPTED;
1463 while (--depth >= 0) {
1464 ix = path[depth].p_idx;
1465 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1466 EXT4_ERROR_INODE(inode,
1467 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1468 ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1469 EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1470 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1472 return -EFSCORRUPTED;
1478 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1479 EXT4_ERROR_INODE(inode,
1480 "logical %d < ee_block %d + ee_len %d!",
1481 *logical, le32_to_cpu(ex->ee_block), ee_len);
1482 return -EFSCORRUPTED;
1485 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1486 *phys = ext4_ext_pblock(ex) + ee_len - 1;
1491 * search the closest allocated block to the right for *logical
1492 * and returns it at @logical + it's physical address at @phys
1493 * if *logical is the largest allocated block, the function
1494 * returns 0 at @phys
1495 * return value contains 0 (success) or error code
1497 static int ext4_ext_search_right(struct inode *inode,
1498 struct ext4_ext_path *path,
1499 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1500 struct ext4_extent **ret_ex)
1502 struct buffer_head *bh = NULL;
1503 struct ext4_extent_header *eh;
1504 struct ext4_extent_idx *ix;
1505 struct ext4_extent *ex;
1507 int depth; /* Note, NOT eh_depth; depth from top of tree */
1510 if (unlikely(path == NULL)) {
1511 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1512 return -EFSCORRUPTED;
1514 depth = path->p_depth;
1517 if (depth == 0 && path->p_ext == NULL)
1520 /* usually extent in the path covers blocks smaller
1521 * then *logical, but it can be that extent is the
1522 * first one in the file */
1524 ex = path[depth].p_ext;
1525 ee_len = ext4_ext_get_actual_len(ex);
1526 if (*logical < le32_to_cpu(ex->ee_block)) {
1527 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1528 EXT4_ERROR_INODE(inode,
1529 "first_extent(path[%d].p_hdr) != ex",
1531 return -EFSCORRUPTED;
1533 while (--depth >= 0) {
1534 ix = path[depth].p_idx;
1535 if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1536 EXT4_ERROR_INODE(inode,
1537 "ix != EXT_FIRST_INDEX *logical %d!",
1539 return -EFSCORRUPTED;
1545 if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1546 EXT4_ERROR_INODE(inode,
1547 "logical %d < ee_block %d + ee_len %d!",
1548 *logical, le32_to_cpu(ex->ee_block), ee_len);
1549 return -EFSCORRUPTED;
1552 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1553 /* next allocated block in this leaf */
1558 /* go up and search for index to the right */
1559 while (--depth >= 0) {
1560 ix = path[depth].p_idx;
1561 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1565 /* we've gone up to the root and found no index to the right */
1569 /* we've found index to the right, let's
1570 * follow it and find the closest allocated
1571 * block to the right */
1573 block = ext4_idx_pblock(ix);
1574 while (++depth < path->p_depth) {
1575 /* subtract from p_depth to get proper eh_depth */
1576 bh = read_extent_tree_block(inode, block,
1577 path->p_depth - depth, 0);
1580 eh = ext_block_hdr(bh);
1581 ix = EXT_FIRST_INDEX(eh);
1582 block = ext4_idx_pblock(ix);
1586 bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1589 eh = ext_block_hdr(bh);
1590 ex = EXT_FIRST_EXTENT(eh);
1592 *logical = le32_to_cpu(ex->ee_block);
1593 *phys = ext4_ext_pblock(ex);
1601 * ext4_ext_next_allocated_block:
1602 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1603 * NOTE: it considers block number from index entry as
1604 * allocated block. Thus, index entries have to be consistent
1608 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1612 BUG_ON(path == NULL);
1613 depth = path->p_depth;
1615 if (depth == 0 && path->p_ext == NULL)
1616 return EXT_MAX_BLOCKS;
1618 while (depth >= 0) {
1619 if (depth == path->p_depth) {
1621 if (path[depth].p_ext &&
1622 path[depth].p_ext !=
1623 EXT_LAST_EXTENT(path[depth].p_hdr))
1624 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1627 if (path[depth].p_idx !=
1628 EXT_LAST_INDEX(path[depth].p_hdr))
1629 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1634 return EXT_MAX_BLOCKS;
1638 * ext4_ext_next_leaf_block:
1639 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1641 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1645 BUG_ON(path == NULL);
1646 depth = path->p_depth;
1648 /* zero-tree has no leaf blocks at all */
1650 return EXT_MAX_BLOCKS;
1652 /* go to index block */
1655 while (depth >= 0) {
1656 if (path[depth].p_idx !=
1657 EXT_LAST_INDEX(path[depth].p_hdr))
1658 return (ext4_lblk_t)
1659 le32_to_cpu(path[depth].p_idx[1].ei_block);
1663 return EXT_MAX_BLOCKS;
1667 * ext4_ext_correct_indexes:
1668 * if leaf gets modified and modified extent is first in the leaf,
1669 * then we have to correct all indexes above.
1670 * TODO: do we need to correct tree in all cases?
1672 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1673 struct ext4_ext_path *path)
1675 struct ext4_extent_header *eh;
1676 int depth = ext_depth(inode);
1677 struct ext4_extent *ex;
1681 eh = path[depth].p_hdr;
1682 ex = path[depth].p_ext;
1684 if (unlikely(ex == NULL || eh == NULL)) {
1685 EXT4_ERROR_INODE(inode,
1686 "ex %p == NULL or eh %p == NULL", ex, eh);
1687 return -EFSCORRUPTED;
1691 /* there is no tree at all */
1695 if (ex != EXT_FIRST_EXTENT(eh)) {
1696 /* we correct tree if first leaf got modified only */
1701 * TODO: we need correction if border is smaller than current one
1704 border = path[depth].p_ext->ee_block;
1705 err = ext4_ext_get_access(handle, inode, path + k);
1708 path[k].p_idx->ei_block = border;
1709 err = ext4_ext_dirty(handle, inode, path + k);
1714 /* change all left-side indexes */
1715 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1717 err = ext4_ext_get_access(handle, inode, path + k);
1720 path[k].p_idx->ei_block = border;
1721 err = ext4_ext_dirty(handle, inode, path + k);
1730 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1731 struct ext4_extent *ex2)
1733 unsigned short ext1_ee_len, ext2_ee_len;
1735 if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1738 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1739 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1741 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1742 le32_to_cpu(ex2->ee_block))
1746 * To allow future support for preallocated extents to be added
1747 * as an RO_COMPAT feature, refuse to merge to extents if
1748 * this can result in the top bit of ee_len being set.
1750 if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1753 * The check for IO to unwritten extent is somewhat racy as we
1754 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1755 * dropping i_data_sem. But reserved blocks should save us in that
1758 if (ext4_ext_is_unwritten(ex1) &&
1759 (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1760 atomic_read(&EXT4_I(inode)->i_unwritten) ||
1761 (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1763 #ifdef AGGRESSIVE_TEST
1764 if (ext1_ee_len >= 4)
1768 if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1774 * This function tries to merge the "ex" extent to the next extent in the tree.
1775 * It always tries to merge towards right. If you want to merge towards
1776 * left, pass "ex - 1" as argument instead of "ex".
1777 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1778 * 1 if they got merged.
1780 static int ext4_ext_try_to_merge_right(struct inode *inode,
1781 struct ext4_ext_path *path,
1782 struct ext4_extent *ex)
1784 struct ext4_extent_header *eh;
1785 unsigned int depth, len;
1786 int merge_done = 0, unwritten;
1788 depth = ext_depth(inode);
1789 BUG_ON(path[depth].p_hdr == NULL);
1790 eh = path[depth].p_hdr;
1792 while (ex < EXT_LAST_EXTENT(eh)) {
1793 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1795 /* merge with next extent! */
1796 unwritten = ext4_ext_is_unwritten(ex);
1797 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1798 + ext4_ext_get_actual_len(ex + 1));
1800 ext4_ext_mark_unwritten(ex);
1802 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1803 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1804 * sizeof(struct ext4_extent);
1805 memmove(ex + 1, ex + 2, len);
1807 le16_add_cpu(&eh->eh_entries, -1);
1809 WARN_ON(eh->eh_entries == 0);
1810 if (!eh->eh_entries)
1811 EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1818 * This function does a very simple check to see if we can collapse
1819 * an extent tree with a single extent tree leaf block into the inode.
1821 static void ext4_ext_try_to_merge_up(handle_t *handle,
1822 struct inode *inode,
1823 struct ext4_ext_path *path)
1826 unsigned max_root = ext4_ext_space_root(inode, 0);
1829 if ((path[0].p_depth != 1) ||
1830 (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1831 (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1835 * We need to modify the block allocation bitmap and the block
1836 * group descriptor to release the extent tree block. If we
1837 * can't get the journal credits, give up.
1839 if (ext4_journal_extend(handle, 2))
1843 * Copy the extent data up to the inode
1845 blk = ext4_idx_pblock(path[0].p_idx);
1846 s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1847 sizeof(struct ext4_extent_idx);
1848 s += sizeof(struct ext4_extent_header);
1850 path[1].p_maxdepth = path[0].p_maxdepth;
1851 memcpy(path[0].p_hdr, path[1].p_hdr, s);
1852 path[0].p_depth = 0;
1853 path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1854 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1855 path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1857 brelse(path[1].p_bh);
1858 ext4_free_blocks(handle, inode, NULL, blk, 1,
1859 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1863 * This function tries to merge the @ex extent to neighbours in the tree.
1864 * return 1 if merge left else 0.
1866 static void ext4_ext_try_to_merge(handle_t *handle,
1867 struct inode *inode,
1868 struct ext4_ext_path *path,
1869 struct ext4_extent *ex) {
1870 struct ext4_extent_header *eh;
1874 depth = ext_depth(inode);
1875 BUG_ON(path[depth].p_hdr == NULL);
1876 eh = path[depth].p_hdr;
1878 if (ex > EXT_FIRST_EXTENT(eh))
1879 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1882 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1884 ext4_ext_try_to_merge_up(handle, inode, path);
1888 * check if a portion of the "newext" extent overlaps with an
1891 * If there is an overlap discovered, it updates the length of the newext
1892 * such that there will be no overlap, and then returns 1.
1893 * If there is no overlap found, it returns 0.
1895 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1896 struct inode *inode,
1897 struct ext4_extent *newext,
1898 struct ext4_ext_path *path)
1901 unsigned int depth, len1;
1902 unsigned int ret = 0;
1904 b1 = le32_to_cpu(newext->ee_block);
1905 len1 = ext4_ext_get_actual_len(newext);
1906 depth = ext_depth(inode);
1907 if (!path[depth].p_ext)
1909 b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1912 * get the next allocated block if the extent in the path
1913 * is before the requested block(s)
1916 b2 = ext4_ext_next_allocated_block(path);
1917 if (b2 == EXT_MAX_BLOCKS)
1919 b2 = EXT4_LBLK_CMASK(sbi, b2);
1922 /* check for wrap through zero on extent logical start block*/
1923 if (b1 + len1 < b1) {
1924 len1 = EXT_MAX_BLOCKS - b1;
1925 newext->ee_len = cpu_to_le16(len1);
1929 /* check for overlap */
1930 if (b1 + len1 > b2) {
1931 newext->ee_len = cpu_to_le16(b2 - b1);
1939 * ext4_ext_insert_extent:
1940 * tries to merge requsted extent into the existing extent or
1941 * inserts requested extent as new one into the tree,
1942 * creating new leaf in the no-space case.
1944 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1945 struct ext4_ext_path **ppath,
1946 struct ext4_extent *newext, int gb_flags)
1948 struct ext4_ext_path *path = *ppath;
1949 struct ext4_extent_header *eh;
1950 struct ext4_extent *ex, *fex;
1951 struct ext4_extent *nearex; /* nearest extent */
1952 struct ext4_ext_path *npath = NULL;
1953 int depth, len, err;
1955 int mb_flags = 0, unwritten;
1957 if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1958 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1959 if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1960 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1961 return -EFSCORRUPTED;
1963 depth = ext_depth(inode);
1964 ex = path[depth].p_ext;
1965 eh = path[depth].p_hdr;
1966 if (unlikely(path[depth].p_hdr == NULL)) {
1967 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1968 return -EFSCORRUPTED;
1971 /* try to insert block into found extent and return */
1972 if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1975 * Try to see whether we should rather test the extent on
1976 * right from ex, or from the left of ex. This is because
1977 * ext4_find_extent() can return either extent on the
1978 * left, or on the right from the searched position. This
1979 * will make merging more effective.
1981 if (ex < EXT_LAST_EXTENT(eh) &&
1982 (le32_to_cpu(ex->ee_block) +
1983 ext4_ext_get_actual_len(ex) <
1984 le32_to_cpu(newext->ee_block))) {
1987 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1988 (le32_to_cpu(newext->ee_block) +
1989 ext4_ext_get_actual_len(newext) <
1990 le32_to_cpu(ex->ee_block)))
1993 /* Try to append newex to the ex */
1994 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1995 ext_debug("append [%d]%d block to %u:[%d]%d"
1997 ext4_ext_is_unwritten(newext),
1998 ext4_ext_get_actual_len(newext),
1999 le32_to_cpu(ex->ee_block),
2000 ext4_ext_is_unwritten(ex),
2001 ext4_ext_get_actual_len(ex),
2002 ext4_ext_pblock(ex));
2003 err = ext4_ext_get_access(handle, inode,
2007 unwritten = ext4_ext_is_unwritten(ex);
2008 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2009 + ext4_ext_get_actual_len(newext));
2011 ext4_ext_mark_unwritten(ex);
2012 eh = path[depth].p_hdr;
2018 /* Try to prepend newex to the ex */
2019 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2020 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2022 le32_to_cpu(newext->ee_block),
2023 ext4_ext_is_unwritten(newext),
2024 ext4_ext_get_actual_len(newext),
2025 le32_to_cpu(ex->ee_block),
2026 ext4_ext_is_unwritten(ex),
2027 ext4_ext_get_actual_len(ex),
2028 ext4_ext_pblock(ex));
2029 err = ext4_ext_get_access(handle, inode,
2034 unwritten = ext4_ext_is_unwritten(ex);
2035 ex->ee_block = newext->ee_block;
2036 ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2037 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2038 + ext4_ext_get_actual_len(newext));
2040 ext4_ext_mark_unwritten(ex);
2041 eh = path[depth].p_hdr;
2047 depth = ext_depth(inode);
2048 eh = path[depth].p_hdr;
2049 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2052 /* probably next leaf has space for us? */
2053 fex = EXT_LAST_EXTENT(eh);
2054 next = EXT_MAX_BLOCKS;
2055 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2056 next = ext4_ext_next_leaf_block(path);
2057 if (next != EXT_MAX_BLOCKS) {
2058 ext_debug("next leaf block - %u\n", next);
2059 BUG_ON(npath != NULL);
2060 npath = ext4_find_extent(inode, next, NULL, 0);
2062 return PTR_ERR(npath);
2063 BUG_ON(npath->p_depth != path->p_depth);
2064 eh = npath[depth].p_hdr;
2065 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2066 ext_debug("next leaf isn't full(%d)\n",
2067 le16_to_cpu(eh->eh_entries));
2071 ext_debug("next leaf has no free space(%d,%d)\n",
2072 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2076 * There is no free space in the found leaf.
2077 * We're gonna add a new leaf in the tree.
2079 if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2080 mb_flags |= EXT4_MB_USE_RESERVED;
2081 err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2085 depth = ext_depth(inode);
2086 eh = path[depth].p_hdr;
2089 nearex = path[depth].p_ext;
2091 err = ext4_ext_get_access(handle, inode, path + depth);
2096 /* there is no extent in this leaf, create first one */
2097 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2098 le32_to_cpu(newext->ee_block),
2099 ext4_ext_pblock(newext),
2100 ext4_ext_is_unwritten(newext),
2101 ext4_ext_get_actual_len(newext));
2102 nearex = EXT_FIRST_EXTENT(eh);
2104 if (le32_to_cpu(newext->ee_block)
2105 > le32_to_cpu(nearex->ee_block)) {
2107 ext_debug("insert %u:%llu:[%d]%d before: "
2109 le32_to_cpu(newext->ee_block),
2110 ext4_ext_pblock(newext),
2111 ext4_ext_is_unwritten(newext),
2112 ext4_ext_get_actual_len(newext),
2117 BUG_ON(newext->ee_block == nearex->ee_block);
2118 ext_debug("insert %u:%llu:[%d]%d after: "
2120 le32_to_cpu(newext->ee_block),
2121 ext4_ext_pblock(newext),
2122 ext4_ext_is_unwritten(newext),
2123 ext4_ext_get_actual_len(newext),
2126 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2128 ext_debug("insert %u:%llu:[%d]%d: "
2129 "move %d extents from 0x%p to 0x%p\n",
2130 le32_to_cpu(newext->ee_block),
2131 ext4_ext_pblock(newext),
2132 ext4_ext_is_unwritten(newext),
2133 ext4_ext_get_actual_len(newext),
2134 len, nearex, nearex + 1);
2135 memmove(nearex + 1, nearex,
2136 len * sizeof(struct ext4_extent));
2140 le16_add_cpu(&eh->eh_entries, 1);
2141 path[depth].p_ext = nearex;
2142 nearex->ee_block = newext->ee_block;
2143 ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2144 nearex->ee_len = newext->ee_len;
2147 /* try to merge extents */
2148 if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2149 ext4_ext_try_to_merge(handle, inode, path, nearex);
2152 /* time to correct all indexes above */
2153 err = ext4_ext_correct_indexes(handle, inode, path);
2157 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2160 ext4_ext_drop_refs(npath);
2165 static int ext4_fill_fiemap_extents(struct inode *inode,
2166 ext4_lblk_t block, ext4_lblk_t num,
2167 struct fiemap_extent_info *fieinfo)
2169 struct ext4_ext_path *path = NULL;
2170 struct ext4_extent *ex;
2171 struct extent_status es;
2172 ext4_lblk_t next, next_del, start = 0, end = 0;
2173 ext4_lblk_t last = block + num;
2174 int exists, depth = 0, err = 0;
2175 unsigned int flags = 0;
2176 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2178 while (block < last && block != EXT_MAX_BLOCKS) {
2180 /* find extent for this block */
2181 down_read(&EXT4_I(inode)->i_data_sem);
2183 path = ext4_find_extent(inode, block, &path, 0);
2185 up_read(&EXT4_I(inode)->i_data_sem);
2186 err = PTR_ERR(path);
2191 depth = ext_depth(inode);
2192 if (unlikely(path[depth].p_hdr == NULL)) {
2193 up_read(&EXT4_I(inode)->i_data_sem);
2194 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2195 err = -EFSCORRUPTED;
2198 ex = path[depth].p_ext;
2199 next = ext4_ext_next_allocated_block(path);
2204 /* there is no extent yet, so try to allocate
2205 * all requested space */
2208 } else if (le32_to_cpu(ex->ee_block) > block) {
2209 /* need to allocate space before found extent */
2211 end = le32_to_cpu(ex->ee_block);
2212 if (block + num < end)
2214 } else if (block >= le32_to_cpu(ex->ee_block)
2215 + ext4_ext_get_actual_len(ex)) {
2216 /* need to allocate space after found extent */
2221 } else if (block >= le32_to_cpu(ex->ee_block)) {
2223 * some part of requested space is covered
2227 end = le32_to_cpu(ex->ee_block)
2228 + ext4_ext_get_actual_len(ex);
2229 if (block + num < end)
2235 BUG_ON(end <= start);
2239 es.es_len = end - start;
2242 es.es_lblk = le32_to_cpu(ex->ee_block);
2243 es.es_len = ext4_ext_get_actual_len(ex);
2244 es.es_pblk = ext4_ext_pblock(ex);
2245 if (ext4_ext_is_unwritten(ex))
2246 flags |= FIEMAP_EXTENT_UNWRITTEN;
2250 * Find delayed extent and update es accordingly. We call
2251 * it even in !exists case to find out whether es is the
2252 * last existing extent or not.
2254 next_del = ext4_find_delayed_extent(inode, &es);
2255 if (!exists && next_del) {
2257 flags |= (FIEMAP_EXTENT_DELALLOC |
2258 FIEMAP_EXTENT_UNKNOWN);
2260 up_read(&EXT4_I(inode)->i_data_sem);
2262 if (unlikely(es.es_len == 0)) {
2263 EXT4_ERROR_INODE(inode, "es.es_len == 0");
2264 err = -EFSCORRUPTED;
2269 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2270 * we need to check next == EXT_MAX_BLOCKS because it is
2271 * possible that an extent is with unwritten and delayed
2272 * status due to when an extent is delayed allocated and
2273 * is allocated by fallocate status tree will track both of
2276 * So we could return a unwritten and delayed extent, and
2277 * its block is equal to 'next'.
2279 if (next == next_del && next == EXT_MAX_BLOCKS) {
2280 flags |= FIEMAP_EXTENT_LAST;
2281 if (unlikely(next_del != EXT_MAX_BLOCKS ||
2282 next != EXT_MAX_BLOCKS)) {
2283 EXT4_ERROR_INODE(inode,
2284 "next extent == %u, next "
2285 "delalloc extent = %u",
2287 err = -EFSCORRUPTED;
2293 err = fiemap_fill_next_extent(fieinfo,
2294 (__u64)es.es_lblk << blksize_bits,
2295 (__u64)es.es_pblk << blksize_bits,
2296 (__u64)es.es_len << blksize_bits,
2306 block = es.es_lblk + es.es_len;
2309 ext4_ext_drop_refs(path);
2315 * ext4_ext_determine_hole - determine hole around given block
2316 * @inode: inode we lookup in
2317 * @path: path in extent tree to @lblk
2318 * @lblk: pointer to logical block around which we want to determine hole
2320 * Determine hole length (and start if easily possible) around given logical
2321 * block. We don't try too hard to find the beginning of the hole but @path
2322 * actually points to extent before @lblk, we provide it.
2324 * The function returns the length of a hole starting at @lblk. We update @lblk
2325 * to the beginning of the hole if we managed to find it.
2327 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2328 struct ext4_ext_path *path,
2331 int depth = ext_depth(inode);
2332 struct ext4_extent *ex;
2335 ex = path[depth].p_ext;
2337 /* there is no extent yet, so gap is [0;-] */
2339 len = EXT_MAX_BLOCKS;
2340 } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2341 len = le32_to_cpu(ex->ee_block) - *lblk;
2342 } else if (*lblk >= le32_to_cpu(ex->ee_block)
2343 + ext4_ext_get_actual_len(ex)) {
2346 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2347 next = ext4_ext_next_allocated_block(path);
2348 BUG_ON(next == *lblk);
2357 * ext4_ext_put_gap_in_cache:
2358 * calculate boundaries of the gap that the requested block fits into
2359 * and cache this gap
2362 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2363 ext4_lblk_t hole_len)
2365 struct extent_status es;
2367 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2368 hole_start + hole_len - 1, &es);
2370 /* There's delayed extent containing lblock? */
2371 if (es.es_lblk <= hole_start)
2373 hole_len = min(es.es_lblk - hole_start, hole_len);
2375 ext_debug(" -> %u:%u\n", hole_start, hole_len);
2376 ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2377 EXTENT_STATUS_HOLE);
2382 * removes index from the index block.
2384 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2385 struct ext4_ext_path *path, int depth)
2390 /* free index block */
2392 path = path + depth;
2393 leaf = ext4_idx_pblock(path->p_idx);
2394 if (unlikely(path->p_hdr->eh_entries == 0)) {
2395 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2396 return -EFSCORRUPTED;
2398 err = ext4_ext_get_access(handle, inode, path);
2402 if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2403 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2404 len *= sizeof(struct ext4_extent_idx);
2405 memmove(path->p_idx, path->p_idx + 1, len);
2408 le16_add_cpu(&path->p_hdr->eh_entries, -1);
2409 err = ext4_ext_dirty(handle, inode, path);
2412 ext_debug("index is empty, remove it, free block %llu\n", leaf);
2413 trace_ext4_ext_rm_idx(inode, leaf);
2415 ext4_free_blocks(handle, inode, NULL, leaf, 1,
2416 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2418 while (--depth >= 0) {
2419 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2422 err = ext4_ext_get_access(handle, inode, path);
2425 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2426 err = ext4_ext_dirty(handle, inode, path);
2434 * ext4_ext_calc_credits_for_single_extent:
2435 * This routine returns max. credits that needed to insert an extent
2436 * to the extent tree.
2437 * When pass the actual path, the caller should calculate credits
2440 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2441 struct ext4_ext_path *path)
2444 int depth = ext_depth(inode);
2447 /* probably there is space in leaf? */
2448 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2449 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2452 * There are some space in the leaf tree, no
2453 * need to account for leaf block credit
2455 * bitmaps and block group descriptor blocks
2456 * and other metadata blocks still need to be
2459 /* 1 bitmap, 1 block group descriptor */
2460 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2465 return ext4_chunk_trans_blocks(inode, nrblocks);
2469 * How many index/leaf blocks need to change/allocate to add @extents extents?
2471 * If we add a single extent, then in the worse case, each tree level
2472 * index/leaf need to be changed in case of the tree split.
2474 * If more extents are inserted, they could cause the whole tree split more
2475 * than once, but this is really rare.
2477 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2482 /* If we are converting the inline data, only one is needed here. */
2483 if (ext4_has_inline_data(inode))
2486 depth = ext_depth(inode);
2496 static inline int get_default_free_blocks_flags(struct inode *inode)
2498 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2499 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2500 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2501 else if (ext4_should_journal_data(inode))
2502 return EXT4_FREE_BLOCKS_FORGET;
2507 * ext4_rereserve_cluster - increment the reserved cluster count when
2508 * freeing a cluster with a pending reservation
2510 * @inode - file containing the cluster
2511 * @lblk - logical block in cluster to be reserved
2513 * Increments the reserved cluster count and adjusts quota in a bigalloc
2514 * file system when freeing a partial cluster containing at least one
2515 * delayed and unwritten block. A partial cluster meeting that
2516 * requirement will have a pending reservation. If so, the
2517 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2518 * defer reserved and allocated space accounting to a subsequent call
2521 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2523 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2524 struct ext4_inode_info *ei = EXT4_I(inode);
2526 dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2528 spin_lock(&ei->i_block_reservation_lock);
2529 ei->i_reserved_data_blocks++;
2530 percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2531 spin_unlock(&ei->i_block_reservation_lock);
2533 percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2534 ext4_remove_pending(inode, lblk);
2537 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2538 struct ext4_extent *ex,
2539 struct partial_cluster *partial,
2540 ext4_lblk_t from, ext4_lblk_t to)
2542 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2543 unsigned short ee_len = ext4_ext_get_actual_len(ex);
2544 ext4_fsblk_t last_pblk, pblk;
2548 /* only extent tail removal is allowed */
2549 if (from < le32_to_cpu(ex->ee_block) ||
2550 to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2551 ext4_error(sbi->s_sb,
2552 "strange request: removal(2) %u-%u from %u:%u",
2553 from, to, le32_to_cpu(ex->ee_block), ee_len);
2557 #ifdef EXTENTS_STATS
2558 spin_lock(&sbi->s_ext_stats_lock);
2559 sbi->s_ext_blocks += ee_len;
2560 sbi->s_ext_extents++;
2561 if (ee_len < sbi->s_ext_min)
2562 sbi->s_ext_min = ee_len;
2563 if (ee_len > sbi->s_ext_max)
2564 sbi->s_ext_max = ee_len;
2565 if (ext_depth(inode) > sbi->s_depth_max)
2566 sbi->s_depth_max = ext_depth(inode);
2567 spin_unlock(&sbi->s_ext_stats_lock);
2570 trace_ext4_remove_blocks(inode, ex, from, to, partial);
2573 * if we have a partial cluster, and it's different from the
2574 * cluster of the last block in the extent, we free it
2576 last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2578 if (partial->state != initial &&
2579 partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2580 if (partial->state == tofree) {
2581 flags = get_default_free_blocks_flags(inode);
2582 if (ext4_is_pending(inode, partial->lblk))
2583 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2584 ext4_free_blocks(handle, inode, NULL,
2585 EXT4_C2B(sbi, partial->pclu),
2586 sbi->s_cluster_ratio, flags);
2587 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2588 ext4_rereserve_cluster(inode, partial->lblk);
2590 partial->state = initial;
2593 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2594 pblk = ext4_ext_pblock(ex) + ee_len - num;
2597 * We free the partial cluster at the end of the extent (if any),
2598 * unless the cluster is used by another extent (partial_cluster
2599 * state is nofree). If a partial cluster exists here, it must be
2600 * shared with the last block in the extent.
2602 flags = get_default_free_blocks_flags(inode);
2604 /* partial, left end cluster aligned, right end unaligned */
2605 if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2606 (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2607 (partial->state != nofree)) {
2608 if (ext4_is_pending(inode, to))
2609 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2610 ext4_free_blocks(handle, inode, NULL,
2611 EXT4_PBLK_CMASK(sbi, last_pblk),
2612 sbi->s_cluster_ratio, flags);
2613 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2614 ext4_rereserve_cluster(inode, to);
2615 partial->state = initial;
2616 flags = get_default_free_blocks_flags(inode);
2619 flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2622 * For bigalloc file systems, we never free a partial cluster
2623 * at the beginning of the extent. Instead, we check to see if we
2624 * need to free it on a subsequent call to ext4_remove_blocks,
2625 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2627 flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2628 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2630 /* reset the partial cluster if we've freed past it */
2631 if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2632 partial->state = initial;
2635 * If we've freed the entire extent but the beginning is not left
2636 * cluster aligned and is not marked as ineligible for freeing we
2637 * record the partial cluster at the beginning of the extent. It
2638 * wasn't freed by the preceding ext4_free_blocks() call, and we
2639 * need to look farther to the left to determine if it's to be freed
2640 * (not shared with another extent). Else, reset the partial
2641 * cluster - we're either done freeing or the beginning of the
2642 * extent is left cluster aligned.
2644 if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2645 if (partial->state == initial) {
2646 partial->pclu = EXT4_B2C(sbi, pblk);
2647 partial->lblk = from;
2648 partial->state = tofree;
2651 partial->state = initial;
2658 * ext4_ext_rm_leaf() Removes the extents associated with the
2659 * blocks appearing between "start" and "end". Both "start"
2660 * and "end" must appear in the same extent or EIO is returned.
2662 * @handle: The journal handle
2663 * @inode: The files inode
2664 * @path: The path to the leaf
2665 * @partial_cluster: The cluster which we'll have to free if all extents
2666 * has been released from it. However, if this value is
2667 * negative, it's a cluster just to the right of the
2668 * punched region and it must not be freed.
2669 * @start: The first block to remove
2670 * @end: The last block to remove
2673 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2674 struct ext4_ext_path *path,
2675 struct partial_cluster *partial,
2676 ext4_lblk_t start, ext4_lblk_t end)
2678 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2679 int err = 0, correct_index = 0;
2680 int depth = ext_depth(inode), credits;
2681 struct ext4_extent_header *eh;
2684 ext4_lblk_t ex_ee_block;
2685 unsigned short ex_ee_len;
2686 unsigned unwritten = 0;
2687 struct ext4_extent *ex;
2690 /* the header must be checked already in ext4_ext_remove_space() */
2691 ext_debug("truncate since %u in leaf to %u\n", start, end);
2692 if (!path[depth].p_hdr)
2693 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2694 eh = path[depth].p_hdr;
2695 if (unlikely(path[depth].p_hdr == NULL)) {
2696 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2697 return -EFSCORRUPTED;
2699 /* find where to start removing */
2700 ex = path[depth].p_ext;
2702 ex = EXT_LAST_EXTENT(eh);
2704 ex_ee_block = le32_to_cpu(ex->ee_block);
2705 ex_ee_len = ext4_ext_get_actual_len(ex);
2707 trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2709 while (ex >= EXT_FIRST_EXTENT(eh) &&
2710 ex_ee_block + ex_ee_len > start) {
2712 if (ext4_ext_is_unwritten(ex))
2717 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2718 unwritten, ex_ee_len);
2719 path[depth].p_ext = ex;
2721 a = ex_ee_block > start ? ex_ee_block : start;
2722 b = ex_ee_block+ex_ee_len - 1 < end ?
2723 ex_ee_block+ex_ee_len - 1 : end;
2725 ext_debug(" border %u:%u\n", a, b);
2727 /* If this extent is beyond the end of the hole, skip it */
2728 if (end < ex_ee_block) {
2730 * We're going to skip this extent and move to another,
2731 * so note that its first cluster is in use to avoid
2732 * freeing it when removing blocks. Eventually, the
2733 * right edge of the truncated/punched region will
2734 * be just to the left.
2736 if (sbi->s_cluster_ratio > 1) {
2737 pblk = ext4_ext_pblock(ex);
2738 partial->pclu = EXT4_B2C(sbi, pblk);
2739 partial->state = nofree;
2742 ex_ee_block = le32_to_cpu(ex->ee_block);
2743 ex_ee_len = ext4_ext_get_actual_len(ex);
2745 } else if (b != ex_ee_block + ex_ee_len - 1) {
2746 EXT4_ERROR_INODE(inode,
2747 "can not handle truncate %u:%u "
2749 start, end, ex_ee_block,
2750 ex_ee_block + ex_ee_len - 1);
2751 err = -EFSCORRUPTED;
2753 } else if (a != ex_ee_block) {
2754 /* remove tail of the extent */
2755 num = a - ex_ee_block;
2757 /* remove whole extent: excellent! */
2761 * 3 for leaf, sb, and inode plus 2 (bmap and group
2762 * descriptor) for each block group; assume two block
2763 * groups plus ex_ee_len/blocks_per_block_group for
2766 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2767 if (ex == EXT_FIRST_EXTENT(eh)) {
2769 credits += (ext_depth(inode)) + 1;
2771 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2773 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2777 err = ext4_ext_get_access(handle, inode, path + depth);
2781 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2786 /* this extent is removed; mark slot entirely unused */
2787 ext4_ext_store_pblock(ex, 0);
2789 ex->ee_len = cpu_to_le16(num);
2791 * Do not mark unwritten if all the blocks in the
2792 * extent have been removed.
2794 if (unwritten && num)
2795 ext4_ext_mark_unwritten(ex);
2797 * If the extent was completely released,
2798 * we need to remove it from the leaf
2801 if (end != EXT_MAX_BLOCKS - 1) {
2803 * For hole punching, we need to scoot all the
2804 * extents up when an extent is removed so that
2805 * we dont have blank extents in the middle
2807 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2808 sizeof(struct ext4_extent));
2810 /* Now get rid of the one at the end */
2811 memset(EXT_LAST_EXTENT(eh), 0,
2812 sizeof(struct ext4_extent));
2814 le16_add_cpu(&eh->eh_entries, -1);
2817 err = ext4_ext_dirty(handle, inode, path + depth);
2821 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2822 ext4_ext_pblock(ex));
2824 ex_ee_block = le32_to_cpu(ex->ee_block);
2825 ex_ee_len = ext4_ext_get_actual_len(ex);
2828 if (correct_index && eh->eh_entries)
2829 err = ext4_ext_correct_indexes(handle, inode, path);
2832 * If there's a partial cluster and at least one extent remains in
2833 * the leaf, free the partial cluster if it isn't shared with the
2834 * current extent. If it is shared with the current extent
2835 * we reset the partial cluster because we've reached the start of the
2836 * truncated/punched region and we're done removing blocks.
2838 if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2839 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2840 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2841 int flags = get_default_free_blocks_flags(inode);
2843 if (ext4_is_pending(inode, partial->lblk))
2844 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2845 ext4_free_blocks(handle, inode, NULL,
2846 EXT4_C2B(sbi, partial->pclu),
2847 sbi->s_cluster_ratio, flags);
2848 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2849 ext4_rereserve_cluster(inode, partial->lblk);
2851 partial->state = initial;
2854 /* if this leaf is free, then we should
2855 * remove it from index block above */
2856 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2857 err = ext4_ext_rm_idx(handle, inode, path, depth);
2864 * ext4_ext_more_to_rm:
2865 * returns 1 if current index has to be freed (even partial)
2868 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2870 BUG_ON(path->p_idx == NULL);
2872 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2876 * if truncate on deeper level happened, it wasn't partial,
2877 * so we have to consider current index for truncation
2879 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2884 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2887 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2888 int depth = ext_depth(inode);
2889 struct ext4_ext_path *path = NULL;
2890 struct partial_cluster partial;
2896 partial.state = initial;
2898 ext_debug("truncate since %u to %u\n", start, end);
2900 /* probably first extent we're gonna free will be last in block */
2901 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2903 return PTR_ERR(handle);
2906 trace_ext4_ext_remove_space(inode, start, end, depth);
2909 * Check if we are removing extents inside the extent tree. If that
2910 * is the case, we are going to punch a hole inside the extent tree
2911 * so we have to check whether we need to split the extent covering
2912 * the last block to remove so we can easily remove the part of it
2913 * in ext4_ext_rm_leaf().
2915 if (end < EXT_MAX_BLOCKS - 1) {
2916 struct ext4_extent *ex;
2917 ext4_lblk_t ee_block, ex_end, lblk;
2920 /* find extent for or closest extent to this block */
2921 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2923 ext4_journal_stop(handle);
2924 return PTR_ERR(path);
2926 depth = ext_depth(inode);
2927 /* Leaf not may not exist only if inode has no blocks at all */
2928 ex = path[depth].p_ext;
2931 EXT4_ERROR_INODE(inode,
2932 "path[%d].p_hdr == NULL",
2934 err = -EFSCORRUPTED;
2939 ee_block = le32_to_cpu(ex->ee_block);
2940 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2943 * See if the last block is inside the extent, if so split
2944 * the extent at 'end' block so we can easily remove the
2945 * tail of the first part of the split extent in
2946 * ext4_ext_rm_leaf().
2948 if (end >= ee_block && end < ex_end) {
2951 * If we're going to split the extent, note that
2952 * the cluster containing the block after 'end' is
2953 * in use to avoid freeing it when removing blocks.
2955 if (sbi->s_cluster_ratio > 1) {
2956 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2957 partial.pclu = EXT4_B2C(sbi, pblk);
2958 partial.state = nofree;
2962 * Split the extent in two so that 'end' is the last
2963 * block in the first new extent. Also we should not
2964 * fail removing space due to ENOSPC so try to use
2965 * reserved block if that happens.
2967 err = ext4_force_split_extent_at(handle, inode, &path,
2972 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2973 partial.state == initial) {
2975 * If we're punching, there's an extent to the right.
2976 * If the partial cluster hasn't been set, set it to
2977 * that extent's first cluster and its state to nofree
2978 * so it won't be freed should it contain blocks to be
2979 * removed. If it's already set (tofree/nofree), we're
2980 * retrying and keep the original partial cluster info
2981 * so a cluster marked tofree as a result of earlier
2982 * extent removal is not lost.
2985 err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2990 partial.pclu = EXT4_B2C(sbi, pblk);
2991 partial.state = nofree;
2996 * We start scanning from right side, freeing all the blocks
2997 * after i_size and walking into the tree depth-wise.
2999 depth = ext_depth(inode);
3004 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3006 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3009 ext4_journal_stop(handle);
3012 path[0].p_maxdepth = path[0].p_depth = depth;
3013 path[0].p_hdr = ext_inode_hdr(inode);
3016 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3017 err = -EFSCORRUPTED;
3023 while (i >= 0 && err == 0) {
3025 /* this is leaf block */
3026 err = ext4_ext_rm_leaf(handle, inode, path,
3027 &partial, start, end);
3028 /* root level has p_bh == NULL, brelse() eats this */
3029 brelse(path[i].p_bh);
3030 path[i].p_bh = NULL;
3035 /* this is index block */
3036 if (!path[i].p_hdr) {
3037 ext_debug("initialize header\n");
3038 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3041 if (!path[i].p_idx) {
3042 /* this level hasn't been touched yet */
3043 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3044 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3045 ext_debug("init index ptr: hdr 0x%p, num %d\n",
3047 le16_to_cpu(path[i].p_hdr->eh_entries));
3049 /* we were already here, see at next index */
3053 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3054 i, EXT_FIRST_INDEX(path[i].p_hdr),
3056 if (ext4_ext_more_to_rm(path + i)) {
3057 struct buffer_head *bh;
3058 /* go to the next level */
3059 ext_debug("move to level %d (block %llu)\n",
3060 i + 1, ext4_idx_pblock(path[i].p_idx));
3061 memset(path + i + 1, 0, sizeof(*path));
3062 bh = read_extent_tree_block(inode,
3063 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3066 /* should we reset i_size? */
3070 /* Yield here to deal with large extent trees.
3071 * Should be a no-op if we did IO above. */
3073 if (WARN_ON(i + 1 > depth)) {
3074 err = -EFSCORRUPTED;
3077 path[i + 1].p_bh = bh;
3079 /* save actual number of indexes since this
3080 * number is changed at the next iteration */
3081 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3084 /* we finished processing this index, go up */
3085 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3086 /* index is empty, remove it;
3087 * handle must be already prepared by the
3088 * truncatei_leaf() */
3089 err = ext4_ext_rm_idx(handle, inode, path, i);
3091 /* root level has p_bh == NULL, brelse() eats this */
3092 brelse(path[i].p_bh);
3093 path[i].p_bh = NULL;
3095 ext_debug("return to level %d\n", i);
3099 trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3100 path->p_hdr->eh_entries);
3103 * if there's a partial cluster and we have removed the first extent
3104 * in the file, then we also free the partial cluster, if any
3106 if (partial.state == tofree && err == 0) {
3107 int flags = get_default_free_blocks_flags(inode);
3109 if (ext4_is_pending(inode, partial.lblk))
3110 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3111 ext4_free_blocks(handle, inode, NULL,
3112 EXT4_C2B(sbi, partial.pclu),
3113 sbi->s_cluster_ratio, flags);
3114 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3115 ext4_rereserve_cluster(inode, partial.lblk);
3116 partial.state = initial;
3119 /* TODO: flexible tree reduction should be here */
3120 if (path->p_hdr->eh_entries == 0) {
3122 * truncate to zero freed all the tree,
3123 * so we need to correct eh_depth
3125 err = ext4_ext_get_access(handle, inode, path);
3127 ext_inode_hdr(inode)->eh_depth = 0;
3128 ext_inode_hdr(inode)->eh_max =
3129 cpu_to_le16(ext4_ext_space_root(inode, 0));
3130 err = ext4_ext_dirty(handle, inode, path);
3134 ext4_ext_drop_refs(path);
3139 ext4_journal_stop(handle);
3145 * called at mount time
3147 void ext4_ext_init(struct super_block *sb)
3150 * possible initialization would be here
3153 if (ext4_has_feature_extents(sb)) {
3154 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3155 printk(KERN_INFO "EXT4-fs: file extents enabled"
3156 #ifdef AGGRESSIVE_TEST
3157 ", aggressive tests"
3159 #ifdef CHECK_BINSEARCH
3162 #ifdef EXTENTS_STATS
3167 #ifdef EXTENTS_STATS
3168 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3169 EXT4_SB(sb)->s_ext_min = 1 << 30;
3170 EXT4_SB(sb)->s_ext_max = 0;
3176 * called at umount time
3178 void ext4_ext_release(struct super_block *sb)
3180 if (!ext4_has_feature_extents(sb))
3183 #ifdef EXTENTS_STATS
3184 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3185 struct ext4_sb_info *sbi = EXT4_SB(sb);
3186 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3187 sbi->s_ext_blocks, sbi->s_ext_extents,
3188 sbi->s_ext_blocks / sbi->s_ext_extents);
3189 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3190 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3195 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3197 ext4_lblk_t ee_block;
3198 ext4_fsblk_t ee_pblock;
3199 unsigned int ee_len;
3201 ee_block = le32_to_cpu(ex->ee_block);
3202 ee_len = ext4_ext_get_actual_len(ex);
3203 ee_pblock = ext4_ext_pblock(ex);
3208 return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3209 EXTENT_STATUS_WRITTEN);
3212 /* FIXME!! we need to try to merge to left or right after zero-out */
3213 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3215 ext4_fsblk_t ee_pblock;
3216 unsigned int ee_len;
3218 ee_len = ext4_ext_get_actual_len(ex);
3219 ee_pblock = ext4_ext_pblock(ex);
3220 return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3225 * ext4_split_extent_at() splits an extent at given block.
3227 * @handle: the journal handle
3228 * @inode: the file inode
3229 * @path: the path to the extent
3230 * @split: the logical block where the extent is splitted.
3231 * @split_flags: indicates if the extent could be zeroout if split fails, and
3232 * the states(init or unwritten) of new extents.
3233 * @flags: flags used to insert new extent to extent tree.
3236 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3237 * of which are deterimined by split_flag.
3239 * There are two cases:
3240 * a> the extent are splitted into two extent.
3241 * b> split is not needed, and just mark the extent.
3243 * return 0 on success.
3245 static int ext4_split_extent_at(handle_t *handle,
3246 struct inode *inode,
3247 struct ext4_ext_path **ppath,
3252 struct ext4_ext_path *path = *ppath;
3253 ext4_fsblk_t newblock;
3254 ext4_lblk_t ee_block;
3255 struct ext4_extent *ex, newex, orig_ex, zero_ex;
3256 struct ext4_extent *ex2 = NULL;
3257 unsigned int ee_len, depth;
3260 BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3261 (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3263 ext_debug("ext4_split_extents_at: inode %lu, logical"
3264 "block %llu\n", inode->i_ino, (unsigned long long)split);
3266 ext4_ext_show_leaf(inode, path);
3268 depth = ext_depth(inode);
3269 ex = path[depth].p_ext;
3270 ee_block = le32_to_cpu(ex->ee_block);
3271 ee_len = ext4_ext_get_actual_len(ex);
3272 newblock = split - ee_block + ext4_ext_pblock(ex);
3274 BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3275 BUG_ON(!ext4_ext_is_unwritten(ex) &&
3276 split_flag & (EXT4_EXT_MAY_ZEROOUT |
3277 EXT4_EXT_MARK_UNWRIT1 |
3278 EXT4_EXT_MARK_UNWRIT2));
3280 err = ext4_ext_get_access(handle, inode, path + depth);
3284 if (split == ee_block) {
3286 * case b: block @split is the block that the extent begins with
3287 * then we just change the state of the extent, and splitting
3290 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3291 ext4_ext_mark_unwritten(ex);
3293 ext4_ext_mark_initialized(ex);
3295 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3296 ext4_ext_try_to_merge(handle, inode, path, ex);
3298 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3303 memcpy(&orig_ex, ex, sizeof(orig_ex));
3304 ex->ee_len = cpu_to_le16(split - ee_block);
3305 if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3306 ext4_ext_mark_unwritten(ex);
3309 * path may lead to new leaf, not to original leaf any more
3310 * after ext4_ext_insert_extent() returns,
3312 err = ext4_ext_dirty(handle, inode, path + depth);
3314 goto fix_extent_len;
3317 ex2->ee_block = cpu_to_le32(split);
3318 ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block));
3319 ext4_ext_store_pblock(ex2, newblock);
3320 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3321 ext4_ext_mark_unwritten(ex2);
3323 err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3324 if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3325 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3326 if (split_flag & EXT4_EXT_DATA_VALID1) {
3327 err = ext4_ext_zeroout(inode, ex2);
3328 zero_ex.ee_block = ex2->ee_block;
3329 zero_ex.ee_len = cpu_to_le16(
3330 ext4_ext_get_actual_len(ex2));
3331 ext4_ext_store_pblock(&zero_ex,
3332 ext4_ext_pblock(ex2));
3334 err = ext4_ext_zeroout(inode, ex);
3335 zero_ex.ee_block = ex->ee_block;
3336 zero_ex.ee_len = cpu_to_le16(
3337 ext4_ext_get_actual_len(ex));
3338 ext4_ext_store_pblock(&zero_ex,
3339 ext4_ext_pblock(ex));
3342 err = ext4_ext_zeroout(inode, &orig_ex);
3343 zero_ex.ee_block = orig_ex.ee_block;
3344 zero_ex.ee_len = cpu_to_le16(
3345 ext4_ext_get_actual_len(&orig_ex));
3346 ext4_ext_store_pblock(&zero_ex,
3347 ext4_ext_pblock(&orig_ex));
3351 goto fix_extent_len;
3352 /* update the extent length and mark as initialized */
3353 ex->ee_len = cpu_to_le16(ee_len);
3354 ext4_ext_try_to_merge(handle, inode, path, ex);
3355 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3357 goto fix_extent_len;
3359 /* update extent status tree */
3360 err = ext4_zeroout_es(inode, &zero_ex);
3364 goto fix_extent_len;
3367 ext4_ext_show_leaf(inode, path);
3371 ex->ee_len = orig_ex.ee_len;
3372 ext4_ext_dirty(handle, inode, path + path->p_depth);
3377 * ext4_split_extents() splits an extent and mark extent which is covered
3378 * by @map as split_flags indicates
3380 * It may result in splitting the extent into multiple extents (up to three)
3381 * There are three possibilities:
3382 * a> There is no split required
3383 * b> Splits in two extents: Split is happening at either end of the extent
3384 * c> Splits in three extents: Somone is splitting in middle of the extent
3387 static int ext4_split_extent(handle_t *handle,
3388 struct inode *inode,
3389 struct ext4_ext_path **ppath,
3390 struct ext4_map_blocks *map,
3394 struct ext4_ext_path *path = *ppath;
3395 ext4_lblk_t ee_block;
3396 struct ext4_extent *ex;
3397 unsigned int ee_len, depth;
3400 int split_flag1, flags1;
3401 int allocated = map->m_len;
3403 depth = ext_depth(inode);
3404 ex = path[depth].p_ext;
3405 ee_block = le32_to_cpu(ex->ee_block);
3406 ee_len = ext4_ext_get_actual_len(ex);
3407 unwritten = ext4_ext_is_unwritten(ex);
3409 if (map->m_lblk + map->m_len < ee_block + ee_len) {
3410 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3411 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3413 split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3414 EXT4_EXT_MARK_UNWRIT2;
3415 if (split_flag & EXT4_EXT_DATA_VALID2)
3416 split_flag1 |= EXT4_EXT_DATA_VALID1;
3417 err = ext4_split_extent_at(handle, inode, ppath,
3418 map->m_lblk + map->m_len, split_flag1, flags1);
3422 allocated = ee_len - (map->m_lblk - ee_block);
3425 * Update path is required because previous ext4_split_extent_at() may
3426 * result in split of original leaf or extent zeroout.
3428 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3430 return PTR_ERR(path);
3431 depth = ext_depth(inode);
3432 ex = path[depth].p_ext;
3434 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3435 (unsigned long) map->m_lblk);
3436 return -EFSCORRUPTED;
3438 unwritten = ext4_ext_is_unwritten(ex);
3441 if (map->m_lblk >= ee_block) {
3442 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3444 split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3445 split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3446 EXT4_EXT_MARK_UNWRIT2);
3448 err = ext4_split_extent_at(handle, inode, ppath,
3449 map->m_lblk, split_flag1, flags);
3454 ext4_ext_show_leaf(inode, path);
3456 return err ? err : allocated;
3460 * This function is called by ext4_ext_map_blocks() if someone tries to write
3461 * to an unwritten extent. It may result in splitting the unwritten
3462 * extent into multiple extents (up to three - one initialized and two
3464 * There are three possibilities:
3465 * a> There is no split required: Entire extent should be initialized
3466 * b> Splits in two extents: Write is happening at either end of the extent
3467 * c> Splits in three extents: Somone is writing in middle of the extent
3470 * - The extent pointed to by 'path' is unwritten.
3471 * - The extent pointed to by 'path' contains a superset
3472 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3474 * Post-conditions on success:
3475 * - the returned value is the number of blocks beyond map->l_lblk
3476 * that are allocated and initialized.
3477 * It is guaranteed to be >= map->m_len.
3479 static int ext4_ext_convert_to_initialized(handle_t *handle,
3480 struct inode *inode,
3481 struct ext4_map_blocks *map,
3482 struct ext4_ext_path **ppath,
3485 struct ext4_ext_path *path = *ppath;
3486 struct ext4_sb_info *sbi;
3487 struct ext4_extent_header *eh;
3488 struct ext4_map_blocks split_map;
3489 struct ext4_extent zero_ex1, zero_ex2;
3490 struct ext4_extent *ex, *abut_ex;
3491 ext4_lblk_t ee_block, eof_block;
3492 unsigned int ee_len, depth, map_len = map->m_len;
3493 int allocated = 0, max_zeroout = 0;
3495 int split_flag = EXT4_EXT_DATA_VALID2;
3497 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3498 "block %llu, max_blocks %u\n", inode->i_ino,
3499 (unsigned long long)map->m_lblk, map_len);
3501 sbi = EXT4_SB(inode->i_sb);
3502 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3503 inode->i_sb->s_blocksize_bits;
3504 if (eof_block < map->m_lblk + map_len)
3505 eof_block = map->m_lblk + map_len;
3507 depth = ext_depth(inode);
3508 eh = path[depth].p_hdr;
3509 ex = path[depth].p_ext;
3510 ee_block = le32_to_cpu(ex->ee_block);
3511 ee_len = ext4_ext_get_actual_len(ex);
3512 zero_ex1.ee_len = 0;
3513 zero_ex2.ee_len = 0;
3515 trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3517 /* Pre-conditions */
3518 BUG_ON(!ext4_ext_is_unwritten(ex));
3519 BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3522 * Attempt to transfer newly initialized blocks from the currently
3523 * unwritten extent to its neighbor. This is much cheaper
3524 * than an insertion followed by a merge as those involve costly
3525 * memmove() calls. Transferring to the left is the common case in
3526 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3527 * followed by append writes.
3529 * Limitations of the current logic:
3530 * - L1: we do not deal with writes covering the whole extent.
3531 * This would require removing the extent if the transfer
3533 * - L2: we only attempt to merge with an extent stored in the
3534 * same extent tree node.
3536 if ((map->m_lblk == ee_block) &&
3537 /* See if we can merge left */
3538 (map_len < ee_len) && /*L1*/
3539 (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/
3540 ext4_lblk_t prev_lblk;
3541 ext4_fsblk_t prev_pblk, ee_pblk;
3542 unsigned int prev_len;
3545 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3546 prev_len = ext4_ext_get_actual_len(abut_ex);
3547 prev_pblk = ext4_ext_pblock(abut_ex);
3548 ee_pblk = ext4_ext_pblock(ex);
3551 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3552 * upon those conditions:
3553 * - C1: abut_ex is initialized,
3554 * - C2: abut_ex is logically abutting ex,
3555 * - C3: abut_ex is physically abutting ex,
3556 * - C4: abut_ex can receive the additional blocks without
3557 * overflowing the (initialized) length limit.
3559 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3560 ((prev_lblk + prev_len) == ee_block) && /*C2*/
3561 ((prev_pblk + prev_len) == ee_pblk) && /*C3*/
3562 (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3563 err = ext4_ext_get_access(handle, inode, path + depth);
3567 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3570 /* Shift the start of ex by 'map_len' blocks */
3571 ex->ee_block = cpu_to_le32(ee_block + map_len);
3572 ext4_ext_store_pblock(ex, ee_pblk + map_len);
3573 ex->ee_len = cpu_to_le16(ee_len - map_len);
3574 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3576 /* Extend abut_ex by 'map_len' blocks */
3577 abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3579 /* Result: number of initialized blocks past m_lblk */
3580 allocated = map_len;
3582 } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3583 (map_len < ee_len) && /*L1*/
3584 ex < EXT_LAST_EXTENT(eh)) { /*L2*/
3585 /* See if we can merge right */
3586 ext4_lblk_t next_lblk;
3587 ext4_fsblk_t next_pblk, ee_pblk;
3588 unsigned int next_len;
3591 next_lblk = le32_to_cpu(abut_ex->ee_block);
3592 next_len = ext4_ext_get_actual_len(abut_ex);
3593 next_pblk = ext4_ext_pblock(abut_ex);
3594 ee_pblk = ext4_ext_pblock(ex);
3597 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3598 * upon those conditions:
3599 * - C1: abut_ex is initialized,
3600 * - C2: abut_ex is logically abutting ex,
3601 * - C3: abut_ex is physically abutting ex,
3602 * - C4: abut_ex can receive the additional blocks without
3603 * overflowing the (initialized) length limit.
3605 if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/
3606 ((map->m_lblk + map_len) == next_lblk) && /*C2*/
3607 ((ee_pblk + ee_len) == next_pblk) && /*C3*/
3608 (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/
3609 err = ext4_ext_get_access(handle, inode, path + depth);
3613 trace_ext4_ext_convert_to_initialized_fastpath(inode,
3616 /* Shift the start of abut_ex by 'map_len' blocks */
3617 abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3618 ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3619 ex->ee_len = cpu_to_le16(ee_len - map_len);
3620 ext4_ext_mark_unwritten(ex); /* Restore the flag */
3622 /* Extend abut_ex by 'map_len' blocks */
3623 abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3625 /* Result: number of initialized blocks past m_lblk */
3626 allocated = map_len;
3630 /* Mark the block containing both extents as dirty */
3631 ext4_ext_dirty(handle, inode, path + depth);
3633 /* Update path to point to the right extent */
3634 path[depth].p_ext = abut_ex;
3637 allocated = ee_len - (map->m_lblk - ee_block);
3639 WARN_ON(map->m_lblk < ee_block);
3641 * It is safe to convert extent to initialized via explicit
3642 * zeroout only if extent is fully inside i_size or new_size.
3644 split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3646 if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3647 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3648 (inode->i_sb->s_blocksize_bits - 10);
3650 if (IS_ENCRYPTED(inode))
3655 * 1. split the extent into three extents.
3656 * 2. split the extent into two extents, zeroout the head of the first
3658 * 3. split the extent into two extents, zeroout the tail of the second
3660 * 4. split the extent into two extents with out zeroout.
3661 * 5. no splitting needed, just possibly zeroout the head and / or the
3662 * tail of the extent.
3664 split_map.m_lblk = map->m_lblk;
3665 split_map.m_len = map->m_len;
3667 if (max_zeroout && (allocated > split_map.m_len)) {
3668 if (allocated <= max_zeroout) {
3671 cpu_to_le32(split_map.m_lblk +
3674 cpu_to_le16(allocated - split_map.m_len);
3675 ext4_ext_store_pblock(&zero_ex1,
3676 ext4_ext_pblock(ex) + split_map.m_lblk +
3677 split_map.m_len - ee_block);
3678 err = ext4_ext_zeroout(inode, &zero_ex1);
3681 split_map.m_len = allocated;
3683 if (split_map.m_lblk - ee_block + split_map.m_len <
3686 if (split_map.m_lblk != ee_block) {
3687 zero_ex2.ee_block = ex->ee_block;
3688 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3690 ext4_ext_store_pblock(&zero_ex2,
3691 ext4_ext_pblock(ex));
3692 err = ext4_ext_zeroout(inode, &zero_ex2);
3697 split_map.m_len += split_map.m_lblk - ee_block;
3698 split_map.m_lblk = ee_block;
3699 allocated = map->m_len;
3703 err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3708 /* If we have gotten a failure, don't zero out status tree */
3710 err = ext4_zeroout_es(inode, &zero_ex1);
3712 err = ext4_zeroout_es(inode, &zero_ex2);
3714 return err ? err : allocated;
3718 * This function is called by ext4_ext_map_blocks() from
3719 * ext4_get_blocks_dio_write() when DIO to write
3720 * to an unwritten extent.
3722 * Writing to an unwritten extent may result in splitting the unwritten
3723 * extent into multiple initialized/unwritten extents (up to three)
3724 * There are three possibilities:
3725 * a> There is no split required: Entire extent should be unwritten
3726 * b> Splits in two extents: Write is happening at either end of the extent
3727 * c> Splits in three extents: Somone is writing in middle of the extent
3729 * This works the same way in the case of initialized -> unwritten conversion.
3731 * One of more index blocks maybe needed if the extent tree grow after
3732 * the unwritten extent split. To prevent ENOSPC occur at the IO
3733 * complete, we need to split the unwritten extent before DIO submit
3734 * the IO. The unwritten extent called at this time will be split
3735 * into three unwritten extent(at most). After IO complete, the part
3736 * being filled will be convert to initialized by the end_io callback function
3737 * via ext4_convert_unwritten_extents().
3739 * Returns the size of unwritten extent to be written on success.
3741 static int ext4_split_convert_extents(handle_t *handle,
3742 struct inode *inode,
3743 struct ext4_map_blocks *map,
3744 struct ext4_ext_path **ppath,
3747 struct ext4_ext_path *path = *ppath;
3748 ext4_lblk_t eof_block;
3749 ext4_lblk_t ee_block;
3750 struct ext4_extent *ex;
3751 unsigned int ee_len;
3752 int split_flag = 0, depth;
3754 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3755 __func__, inode->i_ino,
3756 (unsigned long long)map->m_lblk, map->m_len);
3758 eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3759 inode->i_sb->s_blocksize_bits;
3760 if (eof_block < map->m_lblk + map->m_len)
3761 eof_block = map->m_lblk + map->m_len;
3763 * It is safe to convert extent to initialized via explicit
3764 * zeroout only if extent is fully insde i_size or new_size.
3766 depth = ext_depth(inode);
3767 ex = path[depth].p_ext;
3768 ee_block = le32_to_cpu(ex->ee_block);
3769 ee_len = ext4_ext_get_actual_len(ex);
3771 /* Convert to unwritten */
3772 if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3773 split_flag |= EXT4_EXT_DATA_VALID1;
3774 /* Convert to initialized */
3775 } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3776 split_flag |= ee_block + ee_len <= eof_block ?
3777 EXT4_EXT_MAY_ZEROOUT : 0;
3778 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3780 flags |= EXT4_GET_BLOCKS_PRE_IO;
3781 return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3784 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3785 struct inode *inode,
3786 struct ext4_map_blocks *map,
3787 struct ext4_ext_path **ppath)
3789 struct ext4_ext_path *path = *ppath;
3790 struct ext4_extent *ex;
3791 ext4_lblk_t ee_block;
3792 unsigned int ee_len;
3796 depth = ext_depth(inode);
3797 ex = path[depth].p_ext;
3798 ee_block = le32_to_cpu(ex->ee_block);
3799 ee_len = ext4_ext_get_actual_len(ex);
3801 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3802 "block %llu, max_blocks %u\n", inode->i_ino,
3803 (unsigned long long)ee_block, ee_len);
3805 /* If extent is larger than requested it is a clear sign that we still
3806 * have some extent state machine issues left. So extent_split is still
3808 * TODO: Once all related issues will be fixed this situation should be
3811 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3813 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3814 " len %u; IO logical block %llu, len %u",
3815 inode->i_ino, (unsigned long long)ee_block, ee_len,
3816 (unsigned long long)map->m_lblk, map->m_len);
3818 err = ext4_split_convert_extents(handle, inode, map, ppath,
3819 EXT4_GET_BLOCKS_CONVERT);
3822 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3824 return PTR_ERR(path);
3825 depth = ext_depth(inode);
3826 ex = path[depth].p_ext;
3829 err = ext4_ext_get_access(handle, inode, path + depth);
3832 /* first mark the extent as initialized */
3833 ext4_ext_mark_initialized(ex);
3835 /* note: ext4_ext_correct_indexes() isn't needed here because
3836 * borders are not changed
3838 ext4_ext_try_to_merge(handle, inode, path, ex);
3840 /* Mark modified extent as dirty */
3841 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3843 ext4_ext_show_leaf(inode, path);
3848 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3850 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3852 struct ext4_ext_path *path,
3856 struct ext4_extent_header *eh;
3857 struct ext4_extent *last_ex;
3859 if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3862 depth = ext_depth(inode);
3863 eh = path[depth].p_hdr;
3866 * We're going to remove EOFBLOCKS_FL entirely in future so we
3867 * do not care for this case anymore. Simply remove the flag
3868 * if there are no extents.
3870 if (unlikely(!eh->eh_entries))
3872 last_ex = EXT_LAST_EXTENT(eh);
3874 * We should clear the EOFBLOCKS_FL flag if we are writing the
3875 * last block in the last extent in the file. We test this by
3876 * first checking to see if the caller to
3877 * ext4_ext_get_blocks() was interested in the last block (or
3878 * a block beyond the last block) in the current extent. If
3879 * this turns out to be false, we can bail out from this
3880 * function immediately.
3882 if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3883 ext4_ext_get_actual_len(last_ex))
3886 * If the caller does appear to be planning to write at or
3887 * beyond the end of the current extent, we then test to see
3888 * if the current extent is the last extent in the file, by
3889 * checking to make sure it was reached via the rightmost node
3890 * at each level of the tree.
3892 for (i = depth-1; i >= 0; i--)
3893 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3896 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3897 return ext4_mark_inode_dirty(handle, inode);
3901 convert_initialized_extent(handle_t *handle, struct inode *inode,
3902 struct ext4_map_blocks *map,
3903 struct ext4_ext_path **ppath,
3904 unsigned int allocated)
3906 struct ext4_ext_path *path = *ppath;
3907 struct ext4_extent *ex;
3908 ext4_lblk_t ee_block;
3909 unsigned int ee_len;
3914 * Make sure that the extent is no bigger than we support with
3917 if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3918 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3920 depth = ext_depth(inode);
3921 ex = path[depth].p_ext;
3922 ee_block = le32_to_cpu(ex->ee_block);
3923 ee_len = ext4_ext_get_actual_len(ex);
3925 ext_debug("%s: inode %lu, logical"
3926 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3927 (unsigned long long)ee_block, ee_len);
3929 if (ee_block != map->m_lblk || ee_len > map->m_len) {
3930 err = ext4_split_convert_extents(handle, inode, map, ppath,
3931 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3934 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3936 return PTR_ERR(path);
3937 depth = ext_depth(inode);
3938 ex = path[depth].p_ext;
3940 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3941 (unsigned long) map->m_lblk);
3942 return -EFSCORRUPTED;
3946 err = ext4_ext_get_access(handle, inode, path + depth);
3949 /* first mark the extent as unwritten */
3950 ext4_ext_mark_unwritten(ex);
3952 /* note: ext4_ext_correct_indexes() isn't needed here because
3953 * borders are not changed
3955 ext4_ext_try_to_merge(handle, inode, path, ex);
3957 /* Mark modified extent as dirty */
3958 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3961 ext4_ext_show_leaf(inode, path);
3963 ext4_update_inode_fsync_trans(handle, inode, 1);
3964 err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3967 map->m_flags |= EXT4_MAP_UNWRITTEN;
3968 if (allocated > map->m_len)
3969 allocated = map->m_len;
3970 map->m_len = allocated;
3975 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3976 struct ext4_map_blocks *map,
3977 struct ext4_ext_path **ppath, int flags,
3978 unsigned int allocated, ext4_fsblk_t newblock)
3980 struct ext4_ext_path *path = *ppath;
3984 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
3985 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3986 inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3988 ext4_ext_show_leaf(inode, path);
3991 * When writing into unwritten space, we should not fail to
3992 * allocate metadata blocks for the new extent block if needed.
3994 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3996 trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3997 allocated, newblock);
3999 /* get_block() before submit the IO, split the extent */
4000 if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4001 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4002 flags | EXT4_GET_BLOCKS_CONVERT);
4005 map->m_flags |= EXT4_MAP_UNWRITTEN;
4008 /* IO end_io complete, convert the filled extent to written */
4009 if (flags & EXT4_GET_BLOCKS_CONVERT) {
4010 if (flags & EXT4_GET_BLOCKS_ZERO) {
4011 if (allocated > map->m_len)
4012 allocated = map->m_len;
4013 err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4018 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4021 ext4_update_inode_fsync_trans(handle, inode, 1);
4022 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4026 map->m_flags |= EXT4_MAP_MAPPED;
4027 map->m_pblk = newblock;
4028 if (allocated > map->m_len)
4029 allocated = map->m_len;
4030 map->m_len = allocated;
4033 /* buffered IO case */
4035 * repeat fallocate creation request
4036 * we already have an unwritten extent
4038 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4039 map->m_flags |= EXT4_MAP_UNWRITTEN;
4043 /* buffered READ or buffered write_begin() lookup */
4044 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4046 * We have blocks reserved already. We
4047 * return allocated blocks so that delalloc
4048 * won't do block reservation for us. But
4049 * the buffer head will be unmapped so that
4050 * a read from the block returns 0s.
4052 map->m_flags |= EXT4_MAP_UNWRITTEN;
4056 /* buffered write, writepage time, convert*/
4057 ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4059 ext4_update_inode_fsync_trans(handle, inode, 1);
4066 map->m_flags |= EXT4_MAP_NEW;
4067 if (allocated > map->m_len)
4068 allocated = map->m_len;
4069 map->m_len = allocated;
4072 map->m_flags |= EXT4_MAP_MAPPED;
4073 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4074 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4080 if (allocated > map->m_len)
4081 allocated = map->m_len;
4082 ext4_ext_show_leaf(inode, path);
4083 map->m_pblk = newblock;
4084 map->m_len = allocated;
4086 return err ? err : allocated;
4090 * get_implied_cluster_alloc - check to see if the requested
4091 * allocation (in the map structure) overlaps with a cluster already
4092 * allocated in an extent.
4093 * @sb The filesystem superblock structure
4094 * @map The requested lblk->pblk mapping
4095 * @ex The extent structure which might contain an implied
4096 * cluster allocation
4098 * This function is called by ext4_ext_map_blocks() after we failed to
4099 * find blocks that were already in the inode's extent tree. Hence,
4100 * we know that the beginning of the requested region cannot overlap
4101 * the extent from the inode's extent tree. There are three cases we
4102 * want to catch. The first is this case:
4104 * |--- cluster # N--|
4105 * |--- extent ---| |---- requested region ---|
4108 * The second case that we need to test for is this one:
4110 * |--------- cluster # N ----------------|
4111 * |--- requested region --| |------- extent ----|
4112 * |=======================|
4114 * The third case is when the requested region lies between two extents
4115 * within the same cluster:
4116 * |------------- cluster # N-------------|
4117 * |----- ex -----| |---- ex_right ----|
4118 * |------ requested region ------|
4119 * |================|
4121 * In each of the above cases, we need to set the map->m_pblk and
4122 * map->m_len so it corresponds to the return the extent labelled as
4123 * "|====|" from cluster #N, since it is already in use for data in
4124 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4125 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4126 * as a new "allocated" block region. Otherwise, we will return 0 and
4127 * ext4_ext_map_blocks() will then allocate one or more new clusters
4128 * by calling ext4_mb_new_blocks().
4130 static int get_implied_cluster_alloc(struct super_block *sb,
4131 struct ext4_map_blocks *map,
4132 struct ext4_extent *ex,
4133 struct ext4_ext_path *path)
4135 struct ext4_sb_info *sbi = EXT4_SB(sb);
4136 ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4137 ext4_lblk_t ex_cluster_start, ex_cluster_end;
4138 ext4_lblk_t rr_cluster_start;
4139 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4140 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4141 unsigned short ee_len = ext4_ext_get_actual_len(ex);
4143 /* The extent passed in that we are trying to match */
4144 ex_cluster_start = EXT4_B2C(sbi, ee_block);
4145 ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4147 /* The requested region passed into ext4_map_blocks() */
4148 rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4150 if ((rr_cluster_start == ex_cluster_end) ||
4151 (rr_cluster_start == ex_cluster_start)) {
4152 if (rr_cluster_start == ex_cluster_end)
4153 ee_start += ee_len - 1;
4154 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4155 map->m_len = min(map->m_len,
4156 (unsigned) sbi->s_cluster_ratio - c_offset);
4158 * Check for and handle this case:
4160 * |--------- cluster # N-------------|
4161 * |------- extent ----|
4162 * |--- requested region ---|
4166 if (map->m_lblk < ee_block)
4167 map->m_len = min(map->m_len, ee_block - map->m_lblk);
4170 * Check for the case where there is already another allocated
4171 * block to the right of 'ex' but before the end of the cluster.
4173 * |------------- cluster # N-------------|
4174 * |----- ex -----| |---- ex_right ----|
4175 * |------ requested region ------|
4176 * |================|
4178 if (map->m_lblk > ee_block) {
4179 ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4180 map->m_len = min(map->m_len, next - map->m_lblk);
4183 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4187 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4193 * Block allocation/map/preallocation routine for extents based files
4196 * Need to be called with
4197 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4198 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4200 * return > 0, number of of blocks already mapped/allocated
4201 * if create == 0 and these are pre-allocated blocks
4202 * buffer head is unmapped
4203 * otherwise blocks are mapped
4205 * return = 0, if plain look up failed (blocks have not been allocated)
4206 * buffer head is unmapped
4208 * return < 0, error case.
4210 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4211 struct ext4_map_blocks *map, int flags)
4213 struct ext4_ext_path *path = NULL;
4214 struct ext4_extent newex, *ex, *ex2;
4215 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4216 ext4_fsblk_t newblock = 0;
4217 int free_on_err = 0, err = 0, depth, ret;
4218 unsigned int allocated = 0, offset = 0;
4219 unsigned int allocated_clusters = 0;
4220 struct ext4_allocation_request ar;
4221 ext4_lblk_t cluster_offset;
4222 bool map_from_cluster = false;
4224 ext_debug("blocks %u/%u requested for inode %lu\n",
4225 map->m_lblk, map->m_len, inode->i_ino);
4226 trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4228 /* find extent for this block */
4229 path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4231 err = PTR_ERR(path);
4236 depth = ext_depth(inode);
4239 * consistent leaf must not be empty;
4240 * this situation is possible, though, _during_ tree modification;
4241 * this is why assert can't be put in ext4_find_extent()
4243 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4244 EXT4_ERROR_INODE(inode, "bad extent address "
4245 "lblock: %lu, depth: %d pblock %lld",
4246 (unsigned long) map->m_lblk, depth,
4247 path[depth].p_block);
4248 err = -EFSCORRUPTED;
4252 ex = path[depth].p_ext;
4254 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4255 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4256 unsigned short ee_len;
4260 * unwritten extents are treated as holes, except that
4261 * we split out initialized portions during a write.
4263 ee_len = ext4_ext_get_actual_len(ex);
4265 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4267 /* if found extent covers block, simply return it */
4268 if (in_range(map->m_lblk, ee_block, ee_len)) {
4269 newblock = map->m_lblk - ee_block + ee_start;
4270 /* number of remaining blocks in the extent */
4271 allocated = ee_len - (map->m_lblk - ee_block);
4272 ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4273 ee_block, ee_len, newblock);
4276 * If the extent is initialized check whether the
4277 * caller wants to convert it to unwritten.
4279 if ((!ext4_ext_is_unwritten(ex)) &&
4280 (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4281 allocated = convert_initialized_extent(
4282 handle, inode, map, &path,
4285 } else if (!ext4_ext_is_unwritten(ex))
4288 ret = ext4_ext_handle_unwritten_extents(
4289 handle, inode, map, &path, flags,
4290 allocated, newblock);
4300 * requested block isn't allocated yet;
4301 * we couldn't try to create block if create flag is zero
4303 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4304 ext4_lblk_t hole_start, hole_len;
4306 hole_start = map->m_lblk;
4307 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4309 * put just found gap into cache to speed up
4310 * subsequent requests
4312 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4314 /* Update hole_len to reflect hole size after map->m_lblk */
4315 if (hole_start != map->m_lblk)
4316 hole_len -= map->m_lblk - hole_start;
4318 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4324 * Okay, we need to do block allocation.
4326 newex.ee_block = cpu_to_le32(map->m_lblk);
4327 cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4330 * If we are doing bigalloc, check to see if the extent returned
4331 * by ext4_find_extent() implies a cluster we can use.
4333 if (cluster_offset && ex &&
4334 get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4335 ar.len = allocated = map->m_len;
4336 newblock = map->m_pblk;
4337 map_from_cluster = true;
4338 goto got_allocated_blocks;
4341 /* find neighbour allocated blocks */
4342 ar.lleft = map->m_lblk;
4343 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4346 ar.lright = map->m_lblk;
4348 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4352 /* Check if the extent after searching to the right implies a
4353 * cluster we can use. */
4354 if ((sbi->s_cluster_ratio > 1) && ex2 &&
4355 get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4356 ar.len = allocated = map->m_len;
4357 newblock = map->m_pblk;
4358 map_from_cluster = true;
4359 goto got_allocated_blocks;
4363 * See if request is beyond maximum number of blocks we can have in
4364 * a single extent. For an initialized extent this limit is
4365 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4366 * EXT_UNWRITTEN_MAX_LEN.
4368 if (map->m_len > EXT_INIT_MAX_LEN &&
4369 !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4370 map->m_len = EXT_INIT_MAX_LEN;
4371 else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4372 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4373 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4375 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4376 newex.ee_len = cpu_to_le16(map->m_len);
4377 err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4379 allocated = ext4_ext_get_actual_len(&newex);
4381 allocated = map->m_len;
4383 /* allocate new block */
4385 ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4386 ar.logical = map->m_lblk;
4388 * We calculate the offset from the beginning of the cluster
4389 * for the logical block number, since when we allocate a
4390 * physical cluster, the physical block should start at the
4391 * same offset from the beginning of the cluster. This is
4392 * needed so that future calls to get_implied_cluster_alloc()
4395 offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4396 ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4398 ar.logical -= offset;
4399 if (S_ISREG(inode->i_mode))
4400 ar.flags = EXT4_MB_HINT_DATA;
4402 /* disable in-core preallocation for non-regular files */
4404 if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4405 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4406 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4407 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4408 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4409 ar.flags |= EXT4_MB_USE_RESERVED;
4410 newblock = ext4_mb_new_blocks(handle, &ar, &err);
4413 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4414 ar.goal, newblock, allocated);
4416 allocated_clusters = ar.len;
4417 ar.len = EXT4_C2B(sbi, ar.len) - offset;
4418 if (ar.len > allocated)
4421 got_allocated_blocks:
4422 /* try to insert new extent into found leaf and return */
4423 ext4_ext_store_pblock(&newex, newblock + offset);
4424 newex.ee_len = cpu_to_le16(ar.len);
4425 /* Mark unwritten */
4426 if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4427 ext4_ext_mark_unwritten(&newex);
4428 map->m_flags |= EXT4_MAP_UNWRITTEN;
4432 if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4433 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4436 err = ext4_ext_insert_extent(handle, inode, &path,
4439 if (err && free_on_err) {
4440 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4441 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4442 /* free data blocks we just allocated */
4443 /* not a good idea to call discard here directly,
4444 * but otherwise we'd need to call it every free() */
4445 ext4_discard_preallocations(inode);
4446 ext4_free_blocks(handle, inode, NULL, newblock,
4447 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4451 /* previous routine could use block we allocated */
4452 newblock = ext4_ext_pblock(&newex);
4453 allocated = ext4_ext_get_actual_len(&newex);
4454 if (allocated > map->m_len)
4455 allocated = map->m_len;
4456 map->m_flags |= EXT4_MAP_NEW;
4459 * Reduce the reserved cluster count to reflect successful deferred
4460 * allocation of delayed allocated clusters or direct allocation of
4461 * clusters discovered to be delayed allocated. Once allocated, a
4462 * cluster is not included in the reserved count.
4464 if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4465 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4467 * When allocating delayed allocated clusters, simply
4468 * reduce the reserved cluster count and claim quota
4470 ext4_da_update_reserve_space(inode, allocated_clusters,
4473 ext4_lblk_t lblk, len;
4477 * When allocating non-delayed allocated clusters
4478 * (from fallocate, filemap, DIO, or clusters
4479 * allocated when delalloc has been disabled by
4480 * ext4_nonda_switch), reduce the reserved cluster
4481 * count by the number of allocated clusters that
4482 * have previously been delayed allocated. Quota
4483 * has been claimed by ext4_mb_new_blocks() above,
4484 * so release the quota reservations made for any
4485 * previously delayed allocated clusters.
4487 lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4488 len = allocated_clusters << sbi->s_cluster_bits;
4489 n = ext4_es_delayed_clu(inode, lblk, len);
4491 ext4_da_update_reserve_space(inode, (int) n, 0);
4496 * Cache the extent and update transaction to commit on fdatasync only
4497 * when it is _not_ an unwritten extent.
4499 if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4500 ext4_update_inode_fsync_trans(handle, inode, 1);
4502 ext4_update_inode_fsync_trans(handle, inode, 0);
4504 if (allocated > map->m_len)
4505 allocated = map->m_len;
4506 ext4_ext_show_leaf(inode, path);
4507 map->m_flags |= EXT4_MAP_MAPPED;
4508 map->m_pblk = newblock;
4509 map->m_len = allocated;
4511 ext4_ext_drop_refs(path);
4514 trace_ext4_ext_map_blocks_exit(inode, flags, map,
4515 err ? err : allocated);
4516 return err ? err : allocated;
4519 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4521 struct super_block *sb = inode->i_sb;
4522 ext4_lblk_t last_block;
4526 * TODO: optimization is possible here.
4527 * Probably we need not scan at all,
4528 * because page truncation is enough.
4531 /* we have to know where to truncate from in crash case */
4532 EXT4_I(inode)->i_disksize = inode->i_size;
4533 err = ext4_mark_inode_dirty(handle, inode);
4537 last_block = (inode->i_size + sb->s_blocksize - 1)
4538 >> EXT4_BLOCK_SIZE_BITS(sb);
4540 err = ext4_es_remove_extent(inode, last_block,
4541 EXT_MAX_BLOCKS - last_block);
4542 if (err == -ENOMEM) {
4544 congestion_wait(BLK_RW_ASYNC, HZ/50);
4549 return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4552 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4553 ext4_lblk_t len, loff_t new_size,
4556 struct inode *inode = file_inode(file);
4562 struct ext4_map_blocks map;
4563 unsigned int credits;
4566 BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4567 map.m_lblk = offset;
4570 * Don't normalize the request if it can fit in one extent so
4571 * that it doesn't get unnecessarily split into multiple
4574 if (len <= EXT_UNWRITTEN_MAX_LEN)
4575 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4578 * credits to insert 1 extent into extent tree
4580 credits = ext4_chunk_trans_blocks(inode, len);
4581 depth = ext_depth(inode);
4584 while (ret >= 0 && len) {
4586 * Recalculate credits when extent tree depth changes.
4588 if (depth != ext_depth(inode)) {
4589 credits = ext4_chunk_trans_blocks(inode, len);
4590 depth = ext_depth(inode);
4593 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4595 if (IS_ERR(handle)) {
4596 ret = PTR_ERR(handle);
4599 ret = ext4_map_blocks(handle, inode, &map, flags);
4601 ext4_debug("inode #%lu: block %u: len %u: "
4602 "ext4_ext_map_blocks returned %d",
4603 inode->i_ino, map.m_lblk,
4605 ext4_mark_inode_dirty(handle, inode);
4606 ret2 = ext4_journal_stop(handle);
4610 map.m_len = len = len - ret;
4611 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4612 inode->i_ctime = current_time(inode);
4614 if (epos > new_size)
4616 if (ext4_update_inode_size(inode, epos) & 0x1)
4617 inode->i_mtime = inode->i_ctime;
4619 if (epos > inode->i_size)
4620 ext4_set_inode_flag(inode,
4621 EXT4_INODE_EOFBLOCKS);
4623 ext4_mark_inode_dirty(handle, inode);
4624 ext4_update_inode_fsync_trans(handle, inode, 1);
4625 ret2 = ext4_journal_stop(handle);
4629 if (ret == -ENOSPC &&
4630 ext4_should_retry_alloc(inode->i_sb, &retries)) {
4635 return ret > 0 ? ret2 : ret;
4638 static long ext4_zero_range(struct file *file, loff_t offset,
4639 loff_t len, int mode)
4641 struct inode *inode = file_inode(file);
4642 handle_t *handle = NULL;
4643 unsigned int max_blocks;
4644 loff_t new_size = 0;
4648 int partial_begin, partial_end;
4651 unsigned int blkbits = inode->i_blkbits;
4653 trace_ext4_zero_range(inode, offset, len, mode);
4655 if (!S_ISREG(inode->i_mode))
4658 /* Call ext4_force_commit to flush all data in case of data=journal. */
4659 if (ext4_should_journal_data(inode)) {
4660 ret = ext4_force_commit(inode->i_sb);
4666 * Round up offset. This is not fallocate, we neet to zero out
4667 * blocks, so convert interior block aligned part of the range to
4668 * unwritten and possibly manually zero out unaligned parts of the
4671 start = round_up(offset, 1 << blkbits);
4672 end = round_down((offset + len), 1 << blkbits);
4674 if (start < offset || end > offset + len)
4676 partial_begin = offset & ((1 << blkbits) - 1);
4677 partial_end = (offset + len) & ((1 << blkbits) - 1);
4679 lblk = start >> blkbits;
4680 max_blocks = (end >> blkbits);
4681 if (max_blocks < lblk)
4689 * Indirect files do not support unwritten extnets
4691 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4696 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4697 (offset + len > i_size_read(inode) ||
4698 offset + len > EXT4_I(inode)->i_disksize)) {
4699 new_size = offset + len;
4700 ret = inode_newsize_ok(inode, new_size);
4705 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4706 if (mode & FALLOC_FL_KEEP_SIZE)
4707 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4709 /* Wait all existing dio workers, newcomers will block on i_mutex */
4710 inode_dio_wait(inode);
4712 /* Preallocate the range including the unaligned edges */
4713 if (partial_begin || partial_end) {
4714 ret = ext4_alloc_file_blocks(file,
4715 round_down(offset, 1 << blkbits) >> blkbits,
4716 (round_up((offset + len), 1 << blkbits) -
4717 round_down(offset, 1 << blkbits)) >> blkbits,
4724 /* Zero range excluding the unaligned edges */
4725 if (max_blocks > 0) {
4726 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4730 * Prevent page faults from reinstantiating pages we have
4731 * released from page cache.
4733 down_write(&EXT4_I(inode)->i_mmap_sem);
4735 ret = ext4_break_layouts(inode);
4737 up_write(&EXT4_I(inode)->i_mmap_sem);
4741 ret = ext4_update_disksize_before_punch(inode, offset, len);
4743 up_write(&EXT4_I(inode)->i_mmap_sem);
4746 /* Now release the pages and zero block aligned part of pages */
4747 truncate_pagecache_range(inode, start, end - 1);
4748 inode->i_mtime = inode->i_ctime = current_time(inode);
4750 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4752 up_write(&EXT4_I(inode)->i_mmap_sem);
4756 if (!partial_begin && !partial_end)
4760 * In worst case we have to writeout two nonadjacent unwritten
4761 * blocks and update the inode
4763 credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4764 if (ext4_should_journal_data(inode))
4766 handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4767 if (IS_ERR(handle)) {
4768 ret = PTR_ERR(handle);
4769 ext4_std_error(inode->i_sb, ret);
4773 inode->i_mtime = inode->i_ctime = current_time(inode);
4775 ext4_update_inode_size(inode, new_size);
4778 * Mark that we allocate beyond EOF so the subsequent truncate
4779 * can proceed even if the new size is the same as i_size.
4781 if ((offset + len) > i_size_read(inode))
4782 ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4784 ext4_mark_inode_dirty(handle, inode);
4786 /* Zero out partial block at the edges of the range */
4787 ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4789 ext4_update_inode_fsync_trans(handle, inode, 1);
4791 if (file->f_flags & O_SYNC)
4792 ext4_handle_sync(handle);
4794 ext4_journal_stop(handle);
4796 inode_unlock(inode);
4801 * preallocate space for a file. This implements ext4's fallocate file
4802 * operation, which gets called from sys_fallocate system call.
4803 * For block-mapped files, posix_fallocate should fall back to the method
4804 * of writing zeroes to the required new blocks (the same behavior which is
4805 * expected for file systems which do not support fallocate() system call).
4807 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4809 struct inode *inode = file_inode(file);
4810 loff_t new_size = 0;
4811 unsigned int max_blocks;
4815 unsigned int blkbits = inode->i_blkbits;
4818 * Encrypted inodes can't handle collapse range or insert
4819 * range since we would need to re-encrypt blocks with a
4820 * different IV or XTS tweak (which are based on the logical
4823 * XXX It's not clear why zero range isn't working, but we'll
4824 * leave it disabled for encrypted inodes for now. This is a
4825 * bug we should fix....
4827 if (IS_ENCRYPTED(inode) &&
4828 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4829 FALLOC_FL_ZERO_RANGE)))
4832 /* Return error if mode is not supported */
4833 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4834 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4835 FALLOC_FL_INSERT_RANGE))
4838 if (mode & FALLOC_FL_PUNCH_HOLE)
4839 return ext4_punch_hole(inode, offset, len);
4841 ret = ext4_convert_inline_data(inode);
4845 if (mode & FALLOC_FL_COLLAPSE_RANGE)
4846 return ext4_collapse_range(inode, offset, len);
4848 if (mode & FALLOC_FL_INSERT_RANGE)
4849 return ext4_insert_range(inode, offset, len);
4851 if (mode & FALLOC_FL_ZERO_RANGE)
4852 return ext4_zero_range(file, offset, len, mode);
4854 trace_ext4_fallocate_enter(inode, offset, len, mode);
4855 lblk = offset >> blkbits;
4857 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4858 flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4859 if (mode & FALLOC_FL_KEEP_SIZE)
4860 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4865 * We only support preallocation for extent-based files only
4867 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4872 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4873 (offset + len > i_size_read(inode) ||
4874 offset + len > EXT4_I(inode)->i_disksize)) {
4875 new_size = offset + len;
4876 ret = inode_newsize_ok(inode, new_size);
4881 /* Wait all existing dio workers, newcomers will block on i_mutex */
4882 inode_dio_wait(inode);
4884 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4888 if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4889 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4890 EXT4_I(inode)->i_sync_tid);
4893 inode_unlock(inode);
4894 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4899 * This function convert a range of blocks to written extents
4900 * The caller of this function will pass the start offset and the size.
4901 * all unwritten extents within this range will be converted to
4904 * This function is called from the direct IO end io call back
4905 * function, to convert the fallocated extents after IO is completed.
4906 * Returns 0 on success.
4908 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4909 loff_t offset, ssize_t len)
4911 unsigned int max_blocks;
4914 struct ext4_map_blocks map;
4915 unsigned int credits, blkbits = inode->i_blkbits;
4917 map.m_lblk = offset >> blkbits;
4918 max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4921 * This is somewhat ugly but the idea is clear: When transaction is
4922 * reserved, everything goes into it. Otherwise we rather start several
4923 * smaller transactions for conversion of each extent separately.
4926 handle = ext4_journal_start_reserved(handle,
4927 EXT4_HT_EXT_CONVERT);
4929 return PTR_ERR(handle);
4933 * credits to insert 1 extent into extent tree
4935 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4937 while (ret >= 0 && ret < max_blocks) {
4939 map.m_len = (max_blocks -= ret);
4941 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4943 if (IS_ERR(handle)) {
4944 ret = PTR_ERR(handle);
4948 ret = ext4_map_blocks(handle, inode, &map,
4949 EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4951 ext4_warning(inode->i_sb,
4952 "inode #%lu: block %u: len %u: "
4953 "ext4_ext_map_blocks returned %d",
4954 inode->i_ino, map.m_lblk,
4956 ext4_mark_inode_dirty(handle, inode);
4958 ret2 = ext4_journal_stop(handle);
4959 if (ret <= 0 || ret2)
4963 ret2 = ext4_journal_stop(handle);
4964 return ret > 0 ? ret2 : ret;
4968 * If newes is not existing extent (newes->ec_pblk equals zero) find
4969 * delayed extent at start of newes and update newes accordingly and
4970 * return start of the next delayed extent.
4972 * If newes is existing extent (newes->ec_pblk is not equal zero)
4973 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4974 * extent found. Leave newes unmodified.
4976 static int ext4_find_delayed_extent(struct inode *inode,
4977 struct extent_status *newes)
4979 struct extent_status es;
4980 ext4_lblk_t block, next_del;
4982 if (newes->es_pblk == 0) {
4983 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
4985 newes->es_lblk + newes->es_len - 1,
4989 * No extent in extent-tree contains block @newes->es_pblk,
4990 * then the block may stay in 1)a hole or 2)delayed-extent.
4996 if (es.es_lblk > newes->es_lblk) {
4998 newes->es_len = min(es.es_lblk - newes->es_lblk,
5003 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5006 block = newes->es_lblk + newes->es_len;
5007 ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5008 EXT_MAX_BLOCKS, &es);
5010 next_del = EXT_MAX_BLOCKS;
5012 next_del = es.es_lblk;
5016 /* fiemap flags we can handle specified here */
5017 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5019 static int ext4_xattr_fiemap(struct inode *inode,
5020 struct fiemap_extent_info *fieinfo)
5024 __u32 flags = FIEMAP_EXTENT_LAST;
5025 int blockbits = inode->i_sb->s_blocksize_bits;
5029 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5030 struct ext4_iloc iloc;
5031 int offset; /* offset of xattr in inode */
5033 error = ext4_get_inode_loc(inode, &iloc);
5036 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5037 offset = EXT4_GOOD_OLD_INODE_SIZE +
5038 EXT4_I(inode)->i_extra_isize;
5040 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5041 flags |= FIEMAP_EXTENT_DATA_INLINE;
5043 } else { /* external block */
5044 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5045 length = inode->i_sb->s_blocksize;
5049 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5051 return (error < 0 ? error : 0);
5054 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5055 __u64 start, __u64 len)
5057 ext4_lblk_t start_blk;
5060 if (ext4_has_inline_data(inode)) {
5063 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5070 if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5071 error = ext4_ext_precache(inode);
5076 /* fallback to generic here if not in extents fmt */
5077 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5078 return generic_block_fiemap(inode, fieinfo, start, len,
5081 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5084 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5085 error = ext4_xattr_fiemap(inode, fieinfo);
5087 ext4_lblk_t len_blks;
5090 start_blk = start >> inode->i_sb->s_blocksize_bits;
5091 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5092 if (last_blk >= EXT_MAX_BLOCKS)
5093 last_blk = EXT_MAX_BLOCKS-1;
5094 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5097 * Walk the extent tree gathering extent information
5098 * and pushing extents back to the user.
5100 error = ext4_fill_fiemap_extents(inode, start_blk,
5108 * Function to access the path buffer for marking it dirty.
5109 * It also checks if there are sufficient credits left in the journal handle
5113 ext4_access_path(handle_t *handle, struct inode *inode,
5114 struct ext4_ext_path *path)
5118 if (!ext4_handle_valid(handle))
5122 * Check if need to extend journal credits
5123 * 3 for leaf, sb, and inode plus 2 (bmap and group
5124 * descriptor) for each block group; assume two block
5127 if (handle->h_buffer_credits < 7) {
5128 credits = ext4_writepage_trans_blocks(inode);
5129 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5130 /* EAGAIN is success */
5131 if (err && err != -EAGAIN)
5135 err = ext4_ext_get_access(handle, inode, path);
5140 * ext4_ext_shift_path_extents:
5141 * Shift the extents of a path structure lying between path[depth].p_ext
5142 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5143 * if it is right shift or left shift operation.
5146 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5147 struct inode *inode, handle_t *handle,
5148 enum SHIFT_DIRECTION SHIFT)
5151 struct ext4_extent *ex_start, *ex_last;
5153 depth = path->p_depth;
5155 while (depth >= 0) {
5156 if (depth == path->p_depth) {
5157 ex_start = path[depth].p_ext;
5159 return -EFSCORRUPTED;
5161 ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5163 err = ext4_access_path(handle, inode, path + depth);
5167 if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5170 while (ex_start <= ex_last) {
5171 if (SHIFT == SHIFT_LEFT) {
5172 le32_add_cpu(&ex_start->ee_block,
5174 /* Try to merge to the left. */
5176 EXT_FIRST_EXTENT(path[depth].p_hdr))
5178 ext4_ext_try_to_merge_right(inode,
5179 path, ex_start - 1))
5184 le32_add_cpu(&ex_last->ee_block, shift);
5185 ext4_ext_try_to_merge_right(inode, path,
5190 err = ext4_ext_dirty(handle, inode, path + depth);
5194 if (--depth < 0 || !update)
5198 /* Update index too */
5199 err = ext4_access_path(handle, inode, path + depth);
5203 if (SHIFT == SHIFT_LEFT)
5204 le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5206 le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5207 err = ext4_ext_dirty(handle, inode, path + depth);
5211 /* we are done if current index is not a starting index */
5212 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5223 * ext4_ext_shift_extents:
5224 * All the extents which lies in the range from @start to the last allocated
5225 * block for the @inode are shifted either towards left or right (depending
5226 * upon @SHIFT) by @shift blocks.
5227 * On success, 0 is returned, error otherwise.
5230 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5231 ext4_lblk_t start, ext4_lblk_t shift,
5232 enum SHIFT_DIRECTION SHIFT)
5234 struct ext4_ext_path *path;
5236 struct ext4_extent *extent;
5237 ext4_lblk_t stop, *iterator, ex_start, ex_end;
5239 /* Let path point to the last extent */
5240 path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5243 return PTR_ERR(path);
5245 depth = path->p_depth;
5246 extent = path[depth].p_ext;
5250 stop = le32_to_cpu(extent->ee_block);
5253 * For left shifts, make sure the hole on the left is big enough to
5254 * accommodate the shift. For right shifts, make sure the last extent
5255 * won't be shifted beyond EXT_MAX_BLOCKS.
5257 if (SHIFT == SHIFT_LEFT) {
5258 path = ext4_find_extent(inode, start - 1, &path,
5261 return PTR_ERR(path);
5262 depth = path->p_depth;
5263 extent = path[depth].p_ext;
5265 ex_start = le32_to_cpu(extent->ee_block);
5266 ex_end = le32_to_cpu(extent->ee_block) +
5267 ext4_ext_get_actual_len(extent);
5273 if ((start == ex_start && shift > ex_start) ||
5274 (shift > start - ex_end)) {
5279 if (shift > EXT_MAX_BLOCKS -
5280 (stop + ext4_ext_get_actual_len(extent))) {
5287 * In case of left shift, iterator points to start and it is increased
5288 * till we reach stop. In case of right shift, iterator points to stop
5289 * and it is decreased till we reach start.
5291 if (SHIFT == SHIFT_LEFT)
5297 * Its safe to start updating extents. Start and stop are unsigned, so
5298 * in case of right shift if extent with 0 block is reached, iterator
5299 * becomes NULL to indicate the end of the loop.
5301 while (iterator && start <= stop) {
5302 path = ext4_find_extent(inode, *iterator, &path,
5305 return PTR_ERR(path);
5306 depth = path->p_depth;
5307 extent = path[depth].p_ext;
5309 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5310 (unsigned long) *iterator);
5311 return -EFSCORRUPTED;
5313 if (SHIFT == SHIFT_LEFT && *iterator >
5314 le32_to_cpu(extent->ee_block)) {
5315 /* Hole, move to the next extent */
5316 if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5317 path[depth].p_ext++;
5319 *iterator = ext4_ext_next_allocated_block(path);
5324 if (SHIFT == SHIFT_LEFT) {
5325 extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5326 *iterator = le32_to_cpu(extent->ee_block) +
5327 ext4_ext_get_actual_len(extent);
5329 extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5330 if (le32_to_cpu(extent->ee_block) > 0)
5331 *iterator = le32_to_cpu(extent->ee_block) - 1;
5333 /* Beginning is reached, end of the loop */
5335 /* Update path extent in case we need to stop */
5336 while (le32_to_cpu(extent->ee_block) < start)
5338 path[depth].p_ext = extent;
5340 ret = ext4_ext_shift_path_extents(path, shift, inode,
5346 ext4_ext_drop_refs(path);
5352 * ext4_collapse_range:
5353 * This implements the fallocate's collapse range functionality for ext4
5354 * Returns: 0 and non-zero on error.
5356 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5358 struct super_block *sb = inode->i_sb;
5359 ext4_lblk_t punch_start, punch_stop;
5361 unsigned int credits;
5362 loff_t new_size, ioffset;
5366 * We need to test this early because xfstests assumes that a
5367 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5368 * system does not support collapse range.
5370 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5373 /* Collapse range works only on fs block size aligned offsets. */
5374 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5375 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5378 if (!S_ISREG(inode->i_mode))
5381 trace_ext4_collapse_range(inode, offset, len);
5383 punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5384 punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5386 /* Call ext4_force_commit to flush all data in case of data=journal. */
5387 if (ext4_should_journal_data(inode)) {
5388 ret = ext4_force_commit(inode->i_sb);
5395 * There is no need to overlap collapse range with EOF, in which case
5396 * it is effectively a truncate operation
5398 if (offset + len >= i_size_read(inode)) {
5403 /* Currently just for extent based files */
5404 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5409 /* Wait for existing dio to complete */
5410 inode_dio_wait(inode);
5413 * Prevent page faults from reinstantiating pages we have released from
5416 down_write(&EXT4_I(inode)->i_mmap_sem);
5418 ret = ext4_break_layouts(inode);
5423 * Need to round down offset to be aligned with page size boundary
5424 * for page size > block size.
5426 ioffset = round_down(offset, PAGE_SIZE);
5428 * Write tail of the last page before removed range since it will get
5429 * removed from the page cache below.
5431 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5435 * Write data that will be shifted to preserve them when discarding
5436 * page cache below. We are also protected from pages becoming dirty
5439 ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5443 truncate_pagecache(inode, ioffset);
5445 credits = ext4_writepage_trans_blocks(inode);
5446 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5447 if (IS_ERR(handle)) {
5448 ret = PTR_ERR(handle);
5452 down_write(&EXT4_I(inode)->i_data_sem);
5453 ext4_discard_preallocations(inode);
5455 ret = ext4_es_remove_extent(inode, punch_start,
5456 EXT_MAX_BLOCKS - punch_start);
5458 up_write(&EXT4_I(inode)->i_data_sem);
5462 ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5464 up_write(&EXT4_I(inode)->i_data_sem);
5467 ext4_discard_preallocations(inode);
5469 ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5470 punch_stop - punch_start, SHIFT_LEFT);
5472 up_write(&EXT4_I(inode)->i_data_sem);
5476 new_size = i_size_read(inode) - len;
5477 i_size_write(inode, new_size);
5478 EXT4_I(inode)->i_disksize = new_size;
5480 up_write(&EXT4_I(inode)->i_data_sem);
5482 ext4_handle_sync(handle);
5483 inode->i_mtime = inode->i_ctime = current_time(inode);
5484 ext4_mark_inode_dirty(handle, inode);
5485 ext4_update_inode_fsync_trans(handle, inode, 1);
5488 ext4_journal_stop(handle);
5490 up_write(&EXT4_I(inode)->i_mmap_sem);
5492 inode_unlock(inode);
5497 * ext4_insert_range:
5498 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5499 * The data blocks starting from @offset to the EOF are shifted by @len
5500 * towards right to create a hole in the @inode. Inode size is increased
5502 * Returns 0 on success, error otherwise.
5504 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5506 struct super_block *sb = inode->i_sb;
5508 struct ext4_ext_path *path;
5509 struct ext4_extent *extent;
5510 ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5511 unsigned int credits, ee_len;
5512 int ret = 0, depth, split_flag = 0;
5516 * We need to test this early because xfstests assumes that an
5517 * insert range of (0, 1) will return EOPNOTSUPP if the file
5518 * system does not support insert range.
5520 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5523 /* Insert range works only on fs block size aligned offsets. */
5524 if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5525 len & (EXT4_CLUSTER_SIZE(sb) - 1))
5528 if (!S_ISREG(inode->i_mode))
5531 trace_ext4_insert_range(inode, offset, len);
5533 offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5534 len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5536 /* Call ext4_force_commit to flush all data in case of data=journal */
5537 if (ext4_should_journal_data(inode)) {
5538 ret = ext4_force_commit(inode->i_sb);
5544 /* Currently just for extent based files */
5545 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5550 /* Check for wrap through zero */
5551 if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5556 /* Offset should be less than i_size */
5557 if (offset >= i_size_read(inode)) {
5562 /* Wait for existing dio to complete */
5563 inode_dio_wait(inode);
5566 * Prevent page faults from reinstantiating pages we have released from
5569 down_write(&EXT4_I(inode)->i_mmap_sem);
5571 ret = ext4_break_layouts(inode);
5576 * Need to round down to align start offset to page size boundary
5577 * for page size > block size.
5579 ioffset = round_down(offset, PAGE_SIZE);
5580 /* Write out all dirty pages */
5581 ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5585 truncate_pagecache(inode, ioffset);
5587 credits = ext4_writepage_trans_blocks(inode);
5588 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5589 if (IS_ERR(handle)) {
5590 ret = PTR_ERR(handle);
5594 /* Expand file to avoid data loss if there is error while shifting */
5595 inode->i_size += len;
5596 EXT4_I(inode)->i_disksize += len;
5597 inode->i_mtime = inode->i_ctime = current_time(inode);
5598 ret = ext4_mark_inode_dirty(handle, inode);
5602 down_write(&EXT4_I(inode)->i_data_sem);
5603 ext4_discard_preallocations(inode);
5605 path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5607 up_write(&EXT4_I(inode)->i_data_sem);
5611 depth = ext_depth(inode);
5612 extent = path[depth].p_ext;
5614 ee_start_lblk = le32_to_cpu(extent->ee_block);
5615 ee_len = ext4_ext_get_actual_len(extent);
5618 * If offset_lblk is not the starting block of extent, split
5619 * the extent @offset_lblk
5621 if ((offset_lblk > ee_start_lblk) &&
5622 (offset_lblk < (ee_start_lblk + ee_len))) {
5623 if (ext4_ext_is_unwritten(extent))
5624 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5625 EXT4_EXT_MARK_UNWRIT2;
5626 ret = ext4_split_extent_at(handle, inode, &path,
5627 offset_lblk, split_flag,
5629 EXT4_GET_BLOCKS_PRE_IO |
5630 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5633 ext4_ext_drop_refs(path);
5636 up_write(&EXT4_I(inode)->i_data_sem);
5640 ext4_ext_drop_refs(path);
5644 ret = ext4_es_remove_extent(inode, offset_lblk,
5645 EXT_MAX_BLOCKS - offset_lblk);
5647 up_write(&EXT4_I(inode)->i_data_sem);
5652 * if offset_lblk lies in a hole which is at start of file, use
5653 * ee_start_lblk to shift extents
5655 ret = ext4_ext_shift_extents(inode, handle,
5656 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5657 len_lblk, SHIFT_RIGHT);
5659 up_write(&EXT4_I(inode)->i_data_sem);
5661 ext4_handle_sync(handle);
5663 ext4_update_inode_fsync_trans(handle, inode, 1);
5666 ext4_journal_stop(handle);
5668 up_write(&EXT4_I(inode)->i_mmap_sem);
5670 inode_unlock(inode);
5675 * ext4_swap_extents - Swap extents between two inodes
5677 * @inode1: First inode
5678 * @inode2: Second inode
5679 * @lblk1: Start block for first inode
5680 * @lblk2: Start block for second inode
5681 * @count: Number of blocks to swap
5682 * @unwritten: Mark second inode's extents as unwritten after swap
5683 * @erp: Pointer to save error value
5685 * This helper routine does exactly what is promise "swap extents". All other
5686 * stuff such as page-cache locking consistency, bh mapping consistency or
5687 * extent's data copying must be performed by caller.
5689 * i_mutex is held for both inodes
5690 * i_data_sem is locked for write for both inodes
5692 * All pages from requested range are locked for both inodes
5695 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5696 struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5697 ext4_lblk_t count, int unwritten, int *erp)
5699 struct ext4_ext_path *path1 = NULL;
5700 struct ext4_ext_path *path2 = NULL;
5701 int replaced_count = 0;
5703 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5704 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5705 BUG_ON(!inode_is_locked(inode1));
5706 BUG_ON(!inode_is_locked(inode2));
5708 *erp = ext4_es_remove_extent(inode1, lblk1, count);
5711 *erp = ext4_es_remove_extent(inode2, lblk2, count);
5716 struct ext4_extent *ex1, *ex2, tmp_ex;
5717 ext4_lblk_t e1_blk, e2_blk;
5718 int e1_len, e2_len, len;
5721 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5722 if (IS_ERR(path1)) {
5723 *erp = PTR_ERR(path1);
5729 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5730 if (IS_ERR(path2)) {
5731 *erp = PTR_ERR(path2);
5735 ex1 = path1[path1->p_depth].p_ext;
5736 ex2 = path2[path2->p_depth].p_ext;
5737 /* Do we have somthing to swap ? */
5738 if (unlikely(!ex2 || !ex1))
5741 e1_blk = le32_to_cpu(ex1->ee_block);
5742 e2_blk = le32_to_cpu(ex2->ee_block);
5743 e1_len = ext4_ext_get_actual_len(ex1);
5744 e2_len = ext4_ext_get_actual_len(ex2);
5747 if (!in_range(lblk1, e1_blk, e1_len) ||
5748 !in_range(lblk2, e2_blk, e2_len)) {
5749 ext4_lblk_t next1, next2;
5751 /* if hole after extent, then go to next extent */
5752 next1 = ext4_ext_next_allocated_block(path1);
5753 next2 = ext4_ext_next_allocated_block(path2);
5754 /* If hole before extent, then shift to that extent */
5759 /* Do we have something to swap */
5760 if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5762 /* Move to the rightest boundary */
5763 len = next1 - lblk1;
5764 if (len < next2 - lblk2)
5765 len = next2 - lblk2;
5774 /* Prepare left boundary */
5775 if (e1_blk < lblk1) {
5777 *erp = ext4_force_split_extent_at(handle, inode1,
5782 if (e2_blk < lblk2) {
5784 *erp = ext4_force_split_extent_at(handle, inode2,
5789 /* ext4_split_extent_at() may result in leaf extent split,
5790 * path must to be revalidated. */
5794 /* Prepare right boundary */
5796 if (len > e1_blk + e1_len - lblk1)
5797 len = e1_blk + e1_len - lblk1;
5798 if (len > e2_blk + e2_len - lblk2)
5799 len = e2_blk + e2_len - lblk2;
5801 if (len != e1_len) {
5803 *erp = ext4_force_split_extent_at(handle, inode1,
5804 &path1, lblk1 + len, 0);
5808 if (len != e2_len) {
5810 *erp = ext4_force_split_extent_at(handle, inode2,
5811 &path2, lblk2 + len, 0);
5815 /* ext4_split_extent_at() may result in leaf extent split,
5816 * path must to be revalidated. */
5820 BUG_ON(e2_len != e1_len);
5821 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5824 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5828 /* Both extents are fully inside boundaries. Swap it now */
5830 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5831 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5832 ex1->ee_len = cpu_to_le16(e2_len);
5833 ex2->ee_len = cpu_to_le16(e1_len);
5835 ext4_ext_mark_unwritten(ex2);
5836 if (ext4_ext_is_unwritten(&tmp_ex))
5837 ext4_ext_mark_unwritten(ex1);
5839 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5840 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5841 *erp = ext4_ext_dirty(handle, inode2, path2 +
5845 *erp = ext4_ext_dirty(handle, inode1, path1 +
5848 * Looks scarry ah..? second inode already points to new blocks,
5849 * and it was successfully dirtied. But luckily error may happen
5850 * only due to journal error, so full transaction will be
5857 replaced_count += len;
5861 ext4_ext_drop_refs(path1);
5863 ext4_ext_drop_refs(path2);
5865 path1 = path2 = NULL;
5867 return replaced_count;
5871 * ext4_clu_mapped - determine whether any block in a logical cluster has
5872 * been mapped to a physical cluster
5874 * @inode - file containing the logical cluster
5875 * @lclu - logical cluster of interest
5877 * Returns 1 if any block in the logical cluster is mapped, signifying
5878 * that a physical cluster has been allocated for it. Otherwise,
5879 * returns 0. Can also return negative error codes. Derived from
5880 * ext4_ext_map_blocks().
5882 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5884 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5885 struct ext4_ext_path *path;
5886 int depth, mapped = 0, err = 0;
5887 struct ext4_extent *extent;
5888 ext4_lblk_t first_lblk, first_lclu, last_lclu;
5890 /* search for the extent closest to the first block in the cluster */
5891 path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5893 err = PTR_ERR(path);
5898 depth = ext_depth(inode);
5901 * A consistent leaf must not be empty. This situation is possible,
5902 * though, _during_ tree modification, and it's why an assert can't
5903 * be put in ext4_find_extent().
5905 if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5906 EXT4_ERROR_INODE(inode,
5907 "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5908 (unsigned long) EXT4_C2B(sbi, lclu),
5909 depth, path[depth].p_block);
5910 err = -EFSCORRUPTED;
5914 extent = path[depth].p_ext;
5916 /* can't be mapped if the extent tree is empty */
5920 first_lblk = le32_to_cpu(extent->ee_block);
5921 first_lclu = EXT4_B2C(sbi, first_lblk);
5924 * Three possible outcomes at this point - found extent spanning
5925 * the target cluster, to the left of the target cluster, or to the
5926 * right of the target cluster. The first two cases are handled here.
5927 * The last case indicates the target cluster is not mapped.
5929 if (lclu >= first_lclu) {
5930 last_lclu = EXT4_B2C(sbi, first_lblk +
5931 ext4_ext_get_actual_len(extent) - 1);
5932 if (lclu <= last_lclu) {
5935 first_lblk = ext4_ext_next_allocated_block(path);
5936 first_lclu = EXT4_B2C(sbi, first_lblk);
5937 if (lclu == first_lclu)
5943 ext4_ext_drop_refs(path);
5946 return err ? err : mapped;