1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/indirect.c
7 * linux/fs/ext4/inode.c
9 * Copyright (C) 1992, 1993, 1994, 1995
10 * Remy Card (card@masi.ibp.fr)
11 * Laboratoire MASI - Institut Blaise Pascal
12 * Universite Pierre et Marie Curie (Paris VI)
16 * linux/fs/minix/inode.c
18 * Copyright (C) 1991, 1992 Linus Torvalds
20 * Goal-directed block allocation by Stephen Tweedie
21 * (sct@redhat.com), 1993, 1998
24 #include "ext4_jbd2.h"
26 #include <linux/dax.h>
27 #include <linux/uio.h>
29 #include <trace/events/ext4.h>
34 struct buffer_head
*bh
;
37 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
44 * ext4_block_to_path - parse the block number into array of offsets
45 * @inode: inode in question (we are only interested in its superblock)
46 * @i_block: block number to be parsed
47 * @offsets: array to store the offsets in
48 * @boundary: set this non-zero if the referred-to block is likely to be
49 * followed (on disk) by an indirect block.
51 * To store the locations of file's data ext4 uses a data structure common
52 * for UNIX filesystems - tree of pointers anchored in the inode, with
53 * data blocks at leaves and indirect blocks in intermediate nodes.
54 * This function translates the block number into path in that tree -
55 * return value is the path length and @offsets[n] is the offset of
56 * pointer to (n+1)th node in the nth one. If @block is out of range
57 * (negative or too large) warning is printed and zero returned.
59 * Note: function doesn't find node addresses, so no IO is needed. All
60 * we need to know is the capacity of indirect blocks (taken from the
65 * Portability note: the last comparison (check that we fit into triple
66 * indirect block) is spelled differently, because otherwise on an
67 * architecture with 32-bit longs and 8Kb pages we might get into trouble
68 * if our filesystem had 8Kb blocks. We might use long long, but that would
69 * kill us on x86. Oh, well, at least the sign propagation does not matter -
70 * i_block would have to be negative in the very beginning, so we would not
74 static int ext4_block_to_path(struct inode
*inode
,
76 ext4_lblk_t offsets
[4], int *boundary
)
78 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
79 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
80 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
81 indirect_blocks
= ptrs
,
82 double_blocks
= (1 << (ptrs_bits
* 2));
86 if (i_block
< direct_blocks
) {
87 offsets
[n
++] = i_block
;
88 final
= direct_blocks
;
89 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
90 offsets
[n
++] = EXT4_IND_BLOCK
;
91 offsets
[n
++] = i_block
;
93 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
94 offsets
[n
++] = EXT4_DIND_BLOCK
;
95 offsets
[n
++] = i_block
>> ptrs_bits
;
96 offsets
[n
++] = i_block
& (ptrs
- 1);
98 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
99 offsets
[n
++] = EXT4_TIND_BLOCK
;
100 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
101 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
102 offsets
[n
++] = i_block
& (ptrs
- 1);
105 ext4_warning(inode
->i_sb
, "block %lu > max in inode %lu",
106 i_block
+ direct_blocks
+
107 indirect_blocks
+ double_blocks
, inode
->i_ino
);
110 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
115 * ext4_get_branch - read the chain of indirect blocks leading to data
116 * @inode: inode in question
117 * @depth: depth of the chain (1 - direct pointer, etc.)
118 * @offsets: offsets of pointers in inode/indirect blocks
119 * @chain: place to store the result
120 * @err: here we store the error value
122 * Function fills the array of triples <key, p, bh> and returns %NULL
123 * if everything went OK or the pointer to the last filled triple
124 * (incomplete one) otherwise. Upon the return chain[i].key contains
125 * the number of (i+1)-th block in the chain (as it is stored in memory,
126 * i.e. little-endian 32-bit), chain[i].p contains the address of that
127 * number (it points into struct inode for i==0 and into the bh->b_data
128 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
129 * block for i>0 and NULL for i==0. In other words, it holds the block
130 * numbers of the chain, addresses they were taken from (and where we can
131 * verify that chain did not change) and buffer_heads hosting these
134 * Function stops when it stumbles upon zero pointer (absent block)
135 * (pointer to last triple returned, *@err == 0)
136 * or when it gets an IO error reading an indirect block
137 * (ditto, *@err == -EIO)
138 * or when it reads all @depth-1 indirect blocks successfully and finds
139 * the whole chain, all way to the data (returns %NULL, *err == 0).
141 * Need to be called with
142 * down_read(&EXT4_I(inode)->i_data_sem)
144 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
145 ext4_lblk_t
*offsets
,
146 Indirect chain
[4], int *err
)
148 struct super_block
*sb
= inode
->i_sb
;
150 struct buffer_head
*bh
;
155 /* i_data is not going away, no lock needed */
156 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
160 key
= le32_to_cpu(p
->key
);
161 if (key
> ext4_blocks_count(EXT4_SB(sb
)->s_es
)) {
162 /* the block was out of range */
166 bh
= sb_getblk(sb
, key
);
172 if (!bh_uptodate_or_lock(bh
)) {
173 if (ext4_read_bh(bh
, 0, NULL
, false) < 0) {
177 /* validate block references */
178 if (ext4_check_indirect_blockref(inode
, bh
)) {
184 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
198 * ext4_find_near - find a place for allocation with sufficient locality
200 * @ind: descriptor of indirect block.
202 * This function returns the preferred place for block allocation.
203 * It is used when heuristic for sequential allocation fails.
205 * + if there is a block to the left of our position - allocate near it.
206 * + if pointer will live in indirect block - allocate near that block.
207 * + if pointer will live in inode - allocate in the same
210 * In the latter case we colour the starting block by the callers PID to
211 * prevent it from clashing with concurrent allocations for a different inode
212 * in the same block group. The PID is used here so that functionally related
213 * files will be close-by on-disk.
215 * Caller must make sure that @ind is valid and will stay that way.
217 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
219 struct ext4_inode_info
*ei
= EXT4_I(inode
);
220 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
223 /* Try to find previous block */
224 for (p
= ind
->p
- 1; p
>= start
; p
--) {
226 return le32_to_cpu(*p
);
229 /* No such thing, so let's try location of indirect block */
231 return ind
->bh
->b_blocknr
;
234 * It is going to be referred to from the inode itself? OK, just put it
235 * into the same cylinder group then.
237 return ext4_inode_to_goal_block(inode
);
241 * ext4_find_goal - find a preferred place for allocation.
243 * @block: block we want
244 * @partial: pointer to the last triple within a chain
246 * Normally this function find the preferred place for block allocation,
248 * Because this is only used for non-extent files, we limit the block nr
251 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
257 * XXX need to get goal block from mballoc's data structures
260 goal
= ext4_find_near(inode
, partial
);
261 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
266 * ext4_blks_to_allocate - Look up the block map and count the number
267 * of direct blocks need to be allocated for the given branch.
269 * @branch: chain of indirect blocks
270 * @k: number of blocks need for indirect blocks
271 * @blks: number of data blocks to be mapped.
272 * @blocks_to_boundary: the offset in the indirect block
274 * return the total number of blocks to be allocate, including the
275 * direct and indirect blocks.
277 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
278 int blocks_to_boundary
)
280 unsigned int count
= 0;
283 * Simple case, [t,d]Indirect block(s) has not allocated yet
284 * then it's clear blocks on that path have not allocated
287 /* right now we don't handle cross boundary allocation */
288 if (blks
< blocks_to_boundary
+ 1)
291 count
+= blocks_to_boundary
+ 1;
296 while (count
< blks
&& count
<= blocks_to_boundary
&&
297 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
304 * ext4_alloc_branch() - allocate and set up a chain of blocks
305 * @handle: handle for this transaction
306 * @ar: structure describing the allocation request
307 * @indirect_blks: number of allocated indirect blocks
308 * @offsets: offsets (in the blocks) to store the pointers to next.
309 * @branch: place to store the chain in.
311 * This function allocates blocks, zeroes out all but the last one,
312 * links them into chain and (if we are synchronous) writes them to disk.
313 * In other words, it prepares a branch that can be spliced onto the
314 * inode. It stores the information about that chain in the branch[], in
315 * the same format as ext4_get_branch() would do. We are calling it after
316 * we had read the existing part of chain and partial points to the last
317 * triple of that (one with zero ->key). Upon the exit we have the same
318 * picture as after the successful ext4_get_block(), except that in one
319 * place chain is disconnected - *branch->p is still zero (we did not
320 * set the last link), but branch->key contains the number that should
321 * be placed into *branch->p to fill that gap.
323 * If allocation fails we free all blocks we've allocated (and forget
324 * their buffer_heads) and return the error value the from failed
325 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
326 * as described above and return 0.
328 static int ext4_alloc_branch(handle_t
*handle
,
329 struct ext4_allocation_request
*ar
,
330 int indirect_blks
, ext4_lblk_t
*offsets
,
333 struct buffer_head
* bh
;
334 ext4_fsblk_t b
, new_blocks
[4];
336 int i
, j
, err
, len
= 1;
338 for (i
= 0; i
<= indirect_blks
; i
++) {
339 if (i
== indirect_blks
) {
340 new_blocks
[i
] = ext4_mb_new_blocks(handle
, ar
, &err
);
342 ar
->goal
= new_blocks
[i
] = ext4_new_meta_blocks(handle
,
344 ar
->flags
& EXT4_MB_DELALLOC_RESERVED
,
346 /* Simplify error cleanup... */
347 branch
[i
+1].bh
= NULL
;
353 branch
[i
].key
= cpu_to_le32(new_blocks
[i
]);
357 bh
= branch
[i
].bh
= sb_getblk(ar
->inode
->i_sb
, new_blocks
[i
-1]);
363 BUFFER_TRACE(bh
, "call get_create_access");
364 err
= ext4_journal_get_create_access(handle
, ar
->inode
->i_sb
,
371 memset(bh
->b_data
, 0, bh
->b_size
);
372 p
= branch
[i
].p
= (__le32
*) bh
->b_data
+ offsets
[i
];
375 if (i
== indirect_blks
)
377 for (j
= 0; j
< len
; j
++)
378 *p
++ = cpu_to_le32(b
++);
380 BUFFER_TRACE(bh
, "marking uptodate");
381 set_buffer_uptodate(bh
);
384 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
385 err
= ext4_handle_dirty_metadata(handle
, ar
->inode
, bh
);
391 if (i
== indirect_blks
) {
392 /* Free data blocks */
393 ext4_free_blocks(handle
, ar
->inode
, NULL
, new_blocks
[i
],
397 for (; i
>= 0; i
--) {
399 * We want to ext4_forget() only freshly allocated indirect
400 * blocks. Buffer for new_blocks[i] is at branch[i+1].bh
401 * (buffer at branch[0].bh is indirect block / inode already
402 * existing before ext4_alloc_branch() was called). Also
403 * because blocks are freshly allocated, we don't need to
404 * revoke them which is why we don't set
405 * EXT4_FREE_BLOCKS_METADATA.
407 ext4_free_blocks(handle
, ar
->inode
, branch
[i
+1].bh
,
409 branch
[i
+1].bh
? EXT4_FREE_BLOCKS_FORGET
: 0);
415 * ext4_splice_branch() - splice the allocated branch onto inode.
416 * @handle: handle for this transaction
417 * @ar: structure describing the allocation request
418 * @where: location of missing link
419 * @num: number of indirect blocks we are adding
421 * This function fills the missing link and does all housekeeping needed in
422 * inode (->i_blocks, etc.). In case of success we end up with the full
423 * chain to new block and return 0.
425 static int ext4_splice_branch(handle_t
*handle
,
426 struct ext4_allocation_request
*ar
,
427 Indirect
*where
, int num
)
431 ext4_fsblk_t current_block
;
434 * If we're splicing into a [td]indirect block (as opposed to the
435 * inode) then we need to get write access to the [td]indirect block
439 BUFFER_TRACE(where
->bh
, "get_write_access");
440 err
= ext4_journal_get_write_access(handle
, ar
->inode
->i_sb
,
441 where
->bh
, EXT4_JTR_NONE
);
447 *where
->p
= where
->key
;
450 * Update the host buffer_head or inode to point to more just allocated
451 * direct blocks blocks
453 if (num
== 0 && ar
->len
> 1) {
454 current_block
= le32_to_cpu(where
->key
) + 1;
455 for (i
= 1; i
< ar
->len
; i
++)
456 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
459 /* We are done with atomic stuff, now do the rest of housekeeping */
460 /* had we spliced it onto indirect block? */
463 * If we spliced it onto an indirect block, we haven't
464 * altered the inode. Note however that if it is being spliced
465 * onto an indirect block at the very end of the file (the
466 * file is growing) then we *will* alter the inode to reflect
467 * the new i_size. But that is not done here - it is done in
468 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
470 ext4_debug("splicing indirect only\n");
471 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
472 err
= ext4_handle_dirty_metadata(handle
, ar
->inode
, where
->bh
);
477 * OK, we spliced it into the inode itself on a direct block.
479 err
= ext4_mark_inode_dirty(handle
, ar
->inode
);
482 ext4_debug("splicing direct\n");
487 for (i
= 1; i
<= num
; i
++) {
489 * branch[i].bh is newly allocated, so there is no
490 * need to revoke the block, which is why we don't
491 * need to set EXT4_FREE_BLOCKS_METADATA.
493 ext4_free_blocks(handle
, ar
->inode
, where
[i
].bh
, 0, 1,
494 EXT4_FREE_BLOCKS_FORGET
);
496 ext4_free_blocks(handle
, ar
->inode
, NULL
, le32_to_cpu(where
[num
].key
),
503 * The ext4_ind_map_blocks() function handles non-extents inodes
504 * (i.e., using the traditional indirect/double-indirect i_blocks
505 * scheme) for ext4_map_blocks().
507 * Allocation strategy is simple: if we have to allocate something, we will
508 * have to go the whole way to leaf. So let's do it before attaching anything
509 * to tree, set linkage between the newborn blocks, write them if sync is
510 * required, recheck the path, free and repeat if check fails, otherwise
511 * set the last missing link (that will protect us from any truncate-generated
512 * removals - all blocks on the path are immune now) and possibly force the
513 * write on the parent block.
514 * That has a nice additional property: no special recovery from the failed
515 * allocations is needed - we simply release blocks and do not touch anything
516 * reachable from inode.
518 * `handle' can be NULL if create == 0.
520 * return > 0, # of blocks mapped or allocated.
521 * return = 0, if plain lookup failed.
522 * return < 0, error case.
524 * The ext4_ind_get_blocks() function should be called with
525 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
526 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
527 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
530 int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
531 struct ext4_map_blocks
*map
,
534 struct ext4_allocation_request ar
;
536 ext4_lblk_t offsets
[4];
540 int blocks_to_boundary
= 0;
543 ext4_fsblk_t first_block
= 0;
545 trace_ext4_ind_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
546 ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
547 ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
548 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
549 &blocks_to_boundary
);
554 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
556 /* Simplest case - block found, no allocation needed */
558 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
561 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
564 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
566 if (blk
== first_block
+ count
)
574 /* Next simple case - plain lookup failed */
575 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
576 unsigned epb
= inode
->i_sb
->s_blocksize
/ sizeof(u32
);
580 * Count number blocks in a subtree under 'partial'. At each
581 * level we count number of complete empty subtrees beyond
582 * current offset and then descend into the subtree only
583 * partially beyond current offset.
586 for (i
= partial
- chain
+ 1; i
< depth
; i
++)
587 count
= count
* epb
+ (epb
- offsets
[i
] - 1);
589 /* Fill in size of a hole we found */
591 map
->m_len
= min_t(unsigned int, map
->m_len
, count
);
595 /* Failed read of indirect block */
600 * Okay, we need to do block allocation.
602 if (ext4_has_feature_bigalloc(inode
->i_sb
)) {
603 EXT4_ERROR_INODE(inode
, "Can't allocate blocks for "
604 "non-extent mapped inodes with bigalloc");
609 /* Set up for the direct block allocation */
610 memset(&ar
, 0, sizeof(ar
));
612 ar
.logical
= map
->m_lblk
;
613 if (S_ISREG(inode
->i_mode
))
614 ar
.flags
= EXT4_MB_HINT_DATA
;
615 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
616 ar
.flags
|= EXT4_MB_DELALLOC_RESERVED
;
617 if (flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
618 ar
.flags
|= EXT4_MB_USE_RESERVED
;
620 ar
.goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
622 /* the number of blocks need to allocate for [d,t]indirect blocks */
623 indirect_blks
= (chain
+ depth
) - partial
- 1;
626 * Next look up the indirect map to count the totoal number of
627 * direct blocks to allocate for this branch.
629 ar
.len
= ext4_blks_to_allocate(partial
, indirect_blks
,
630 map
->m_len
, blocks_to_boundary
);
633 * Block out ext4_truncate while we alter the tree
635 err
= ext4_alloc_branch(handle
, &ar
, indirect_blks
,
636 offsets
+ (partial
- chain
), partial
);
639 * The ext4_splice_branch call will free and forget any buffers
640 * on the new chain if there is a failure, but that risks using
641 * up transaction credits, especially for bitmaps where the
642 * credits cannot be returned. Can we handle this somehow? We
643 * may need to return -EAGAIN upwards in the worst case. --sct
646 err
= ext4_splice_branch(handle
, &ar
, partial
, indirect_blks
);
650 map
->m_flags
|= EXT4_MAP_NEW
;
652 ext4_update_inode_fsync_trans(handle
, inode
, 1);
656 map
->m_flags
|= EXT4_MAP_MAPPED
;
657 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
659 if (count
> blocks_to_boundary
)
660 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
662 /* Clean up and exit */
663 partial
= chain
+ depth
- 1; /* the whole chain */
665 while (partial
> chain
) {
666 BUFFER_TRACE(partial
->bh
, "call brelse");
671 trace_ext4_ind_map_blocks_exit(inode
, flags
, map
, err
);
676 * Calculate number of indirect blocks touched by mapping @nrblocks logically
679 int ext4_ind_trans_blocks(struct inode
*inode
, int nrblocks
)
682 * With N contiguous data blocks, we need at most
683 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
684 * 2 dindirect blocks, and 1 tindirect block
686 return DIV_ROUND_UP(nrblocks
, EXT4_ADDR_PER_BLOCK(inode
->i_sb
)) + 4;
689 static int ext4_ind_trunc_restart_fn(handle_t
*handle
, struct inode
*inode
,
690 struct buffer_head
*bh
, int *dropped
)
695 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
696 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
700 err
= ext4_mark_inode_dirty(handle
, inode
);
704 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
705 * moment, get_block can be called only for blocks inside i_size since
706 * page cache has been already dropped and writes are blocked by
707 * i_rwsem. So we can safely drop the i_data_sem here.
709 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
710 ext4_discard_preallocations(inode
);
711 up_write(&EXT4_I(inode
)->i_data_sem
);
717 * Truncate transactions can be complex and absolutely huge. So we need to
718 * be able to restart the transaction at a convenient checkpoint to make
719 * sure we don't overflow the journal.
721 * Try to extend this transaction for the purposes of truncation. If
722 * extend fails, we restart transaction.
724 static int ext4_ind_truncate_ensure_credits(handle_t
*handle
,
726 struct buffer_head
*bh
,
732 ret
= ext4_journal_ensure_credits_fn(handle
, EXT4_RESERVE_TRANS_BLOCKS
,
733 ext4_blocks_for_truncate(inode
), revoke_creds
,
734 ext4_ind_trunc_restart_fn(handle
, inode
, bh
, &dropped
));
736 down_write(&EXT4_I(inode
)->i_data_sem
);
740 BUFFER_TRACE(bh
, "retaking write access");
741 ret
= ext4_journal_get_write_access(handle
, inode
->i_sb
, bh
,
750 * Probably it should be a library function... search for first non-zero word
751 * or memcmp with zero_page, whatever is better for particular architecture.
754 static inline int all_zeroes(__le32
*p
, __le32
*q
)
763 * ext4_find_shared - find the indirect blocks for partial truncation.
764 * @inode: inode in question
765 * @depth: depth of the affected branch
766 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
767 * @chain: place to store the pointers to partial indirect blocks
768 * @top: place to the (detached) top of branch
770 * This is a helper function used by ext4_truncate().
772 * When we do truncate() we may have to clean the ends of several
773 * indirect blocks but leave the blocks themselves alive. Block is
774 * partially truncated if some data below the new i_size is referred
775 * from it (and it is on the path to the first completely truncated
776 * data block, indeed). We have to free the top of that path along
777 * with everything to the right of the path. Since no allocation
778 * past the truncation point is possible until ext4_truncate()
779 * finishes, we may safely do the latter, but top of branch may
780 * require special attention - pageout below the truncation point
781 * might try to populate it.
783 * We atomically detach the top of branch from the tree, store the
784 * block number of its root in *@top, pointers to buffer_heads of
785 * partially truncated blocks - in @chain[].bh and pointers to
786 * their last elements that should not be removed - in
787 * @chain[].p. Return value is the pointer to last filled element
790 * The work left to caller to do the actual freeing of subtrees:
791 * a) free the subtree starting from *@top
792 * b) free the subtrees whose roots are stored in
793 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
794 * c) free the subtrees growing from the inode past the @chain[0].
795 * (no partially truncated stuff there). */
797 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
798 ext4_lblk_t offsets
[4], Indirect chain
[4],
801 Indirect
*partial
, *p
;
805 /* Make k index the deepest non-null offset + 1 */
806 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
808 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
809 /* Writer: pointers */
811 partial
= chain
+ k
-1;
813 * If the branch acquired continuation since we've looked at it -
814 * fine, it should all survive and (new) top doesn't belong to us.
816 if (!partial
->key
&& *partial
->p
)
819 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
822 * OK, we've found the last block that must survive. The rest of our
823 * branch should be detached before unlocking. However, if that rest
824 * of branch is all ours and does not grow immediately from the inode
825 * it's easier to cheat and just decrement partial->p.
827 if (p
== chain
+ k
- 1 && p
> chain
) {
831 /* Nope, don't do this in ext4. Must leave the tree intact */
838 while (partial
> p
) {
847 * Zero a number of block pointers in either an inode or an indirect block.
848 * If we restart the transaction we must again get write access to the
849 * indirect block for further modification.
851 * We release `count' blocks on disk, but (last - first) may be greater
852 * than `count' because there can be holes in there.
854 * Return 0 on success, 1 on invalid block range
855 * and < 0 on fatal error.
857 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
858 struct buffer_head
*bh
,
859 ext4_fsblk_t block_to_free
,
860 unsigned long count
, __le32
*first
,
864 int flags
= EXT4_FREE_BLOCKS_VALIDATED
;
867 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
) ||
868 ext4_test_inode_flag(inode
, EXT4_INODE_EA_INODE
))
869 flags
|= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_METADATA
;
870 else if (ext4_should_journal_data(inode
))
871 flags
|= EXT4_FREE_BLOCKS_FORGET
;
873 if (!ext4_inode_block_valid(inode
, block_to_free
, count
)) {
874 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
875 "blocks %llu len %lu",
876 (unsigned long long) block_to_free
, count
);
880 err
= ext4_ind_truncate_ensure_credits(handle
, inode
, bh
,
881 ext4_free_data_revoke_credits(inode
, count
));
885 for (p
= first
; p
< last
; p
++)
888 ext4_free_blocks(handle
, inode
, NULL
, block_to_free
, count
, flags
);
891 ext4_std_error(inode
->i_sb
, err
);
896 * ext4_free_data - free a list of data blocks
897 * @handle: handle for this transaction
898 * @inode: inode we are dealing with
899 * @this_bh: indirect buffer_head which contains *@first and *@last
900 * @first: array of block numbers
901 * @last: points immediately past the end of array
903 * We are freeing all blocks referred from that array (numbers are stored as
904 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
906 * We accumulate contiguous runs of blocks to free. Conveniently, if these
907 * blocks are contiguous then releasing them at one time will only affect one
908 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
909 * actually use a lot of journal space.
911 * @this_bh will be %NULL if @first and @last point into the inode's direct
914 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
915 struct buffer_head
*this_bh
,
916 __le32
*first
, __le32
*last
)
918 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
919 unsigned long count
= 0; /* Number of blocks in the run */
920 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
923 ext4_fsblk_t nr
; /* Current block # */
924 __le32
*p
; /* Pointer into inode/ind
928 if (this_bh
) { /* For indirect block */
929 BUFFER_TRACE(this_bh
, "get_write_access");
930 err
= ext4_journal_get_write_access(handle
, inode
->i_sb
,
931 this_bh
, EXT4_JTR_NONE
);
932 /* Important: if we can't update the indirect pointers
933 * to the blocks, we can't free them. */
938 for (p
= first
; p
< last
; p
++) {
939 nr
= le32_to_cpu(*p
);
941 /* accumulate blocks to free if they're contiguous */
946 } else if (nr
== block_to_free
+ count
) {
949 err
= ext4_clear_blocks(handle
, inode
, this_bh
,
950 block_to_free
, count
,
961 if (!err
&& count
> 0)
962 err
= ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
963 count
, block_to_free_p
, p
);
969 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
972 * The buffer head should have an attached journal head at this
973 * point. However, if the data is corrupted and an indirect
974 * block pointed to itself, it would have been detached when
975 * the block was cleared. Check for this instead of OOPSing.
977 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
978 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
980 EXT4_ERROR_INODE(inode
,
981 "circular indirect block detected at "
983 (unsigned long long) this_bh
->b_blocknr
);
988 * ext4_free_branches - free an array of branches
989 * @handle: JBD handle for this transaction
990 * @inode: inode we are dealing with
991 * @parent_bh: the buffer_head which contains *@first and *@last
992 * @first: array of block numbers
993 * @last: pointer immediately past the end of array
994 * @depth: depth of the branches to free
996 * We are freeing all blocks referred from these branches (numbers are
997 * stored as little-endian 32-bit) and updating @inode->i_blocks
1000 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
1001 struct buffer_head
*parent_bh
,
1002 __le32
*first
, __le32
*last
, int depth
)
1007 if (ext4_handle_is_aborted(handle
))
1011 struct buffer_head
*bh
;
1012 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1014 while (--p
>= first
) {
1015 nr
= le32_to_cpu(*p
);
1017 continue; /* A hole */
1019 if (!ext4_inode_block_valid(inode
, nr
, 1)) {
1020 EXT4_ERROR_INODE(inode
,
1021 "invalid indirect mapped "
1022 "block %lu (level %d)",
1023 (unsigned long) nr
, depth
);
1027 /* Go read the buffer for the next level down */
1028 bh
= ext4_sb_bread(inode
->i_sb
, nr
, 0);
1031 * A read failure? Report error and clear slot
1035 ext4_error_inode_block(inode
, nr
, -PTR_ERR(bh
),
1040 /* This zaps the entire block. Bottom up. */
1041 BUFFER_TRACE(bh
, "free child branches");
1042 ext4_free_branches(handle
, inode
, bh
,
1043 (__le32
*) bh
->b_data
,
1044 (__le32
*) bh
->b_data
+ addr_per_block
,
1049 * Everything below this pointer has been
1050 * released. Now let this top-of-subtree go.
1052 * We want the freeing of this indirect block to be
1053 * atomic in the journal with the updating of the
1054 * bitmap block which owns it. So make some room in
1057 * We zero the parent pointer *after* freeing its
1058 * pointee in the bitmaps, so if extend_transaction()
1059 * for some reason fails to put the bitmap changes and
1060 * the release into the same transaction, recovery
1061 * will merely complain about releasing a free block,
1062 * rather than leaking blocks.
1064 if (ext4_handle_is_aborted(handle
))
1066 if (ext4_ind_truncate_ensure_credits(handle
, inode
,
1068 ext4_free_metadata_revoke_credits(
1069 inode
->i_sb
, 1)) < 0)
1073 * The forget flag here is critical because if
1074 * we are journaling (and not doing data
1075 * journaling), we have to make sure a revoke
1076 * record is written to prevent the journal
1077 * replay from overwriting the (former)
1078 * indirect block if it gets reallocated as a
1079 * data block. This must happen in the same
1080 * transaction where the data blocks are
1083 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1084 EXT4_FREE_BLOCKS_METADATA
|
1085 EXT4_FREE_BLOCKS_FORGET
);
1089 * The block which we have just freed is
1090 * pointed to by an indirect block: journal it
1092 BUFFER_TRACE(parent_bh
, "get_write_access");
1093 if (!ext4_journal_get_write_access(handle
,
1094 inode
->i_sb
, parent_bh
,
1097 BUFFER_TRACE(parent_bh
,
1098 "call ext4_handle_dirty_metadata");
1099 ext4_handle_dirty_metadata(handle
,
1106 /* We have reached the bottom of the tree. */
1107 BUFFER_TRACE(parent_bh
, "free data blocks");
1108 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
1112 void ext4_ind_truncate(handle_t
*handle
, struct inode
*inode
)
1114 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1115 __le32
*i_data
= ei
->i_data
;
1116 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1117 ext4_lblk_t offsets
[4];
1122 ext4_lblk_t last_block
, max_block
;
1123 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1125 last_block
= (inode
->i_size
+ blocksize
-1)
1126 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1127 max_block
= (EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
+ blocksize
-1)
1128 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1130 if (last_block
!= max_block
) {
1131 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
1136 ext4_es_remove_extent(inode
, last_block
, EXT_MAX_BLOCKS
- last_block
);
1139 * The orphan list entry will now protect us from any crash which
1140 * occurs before the truncate completes, so it is now safe to propagate
1141 * the new, shorter inode size (held for now in i_size) into the
1142 * on-disk inode. We do this via i_disksize, which is the value which
1143 * ext4 *really* writes onto the disk inode.
1145 ei
->i_disksize
= inode
->i_size
;
1147 if (last_block
== max_block
) {
1149 * It is unnecessary to free any data blocks if last_block is
1150 * equal to the indirect block limit.
1153 } else if (n
== 1) { /* direct blocks */
1154 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
1155 i_data
+ EXT4_NDIR_BLOCKS
);
1159 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1160 /* Kill the top of shared branch (not detached) */
1162 if (partial
== chain
) {
1163 /* Shared branch grows from the inode */
1164 ext4_free_branches(handle
, inode
, NULL
,
1165 &nr
, &nr
+1, (chain
+n
-1) - partial
);
1168 * We mark the inode dirty prior to restart,
1169 * and prior to stop. No need for it here.
1172 /* Shared branch grows from an indirect block */
1173 BUFFER_TRACE(partial
->bh
, "get_write_access");
1174 ext4_free_branches(handle
, inode
, partial
->bh
,
1176 partial
->p
+1, (chain
+n
-1) - partial
);
1179 /* Clear the ends of indirect blocks on the shared branch */
1180 while (partial
> chain
) {
1181 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
1182 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1183 (chain
+n
-1) - partial
);
1184 BUFFER_TRACE(partial
->bh
, "call brelse");
1185 brelse(partial
->bh
);
1189 /* Kill the remaining (whole) subtrees */
1190 switch (offsets
[0]) {
1192 nr
= i_data
[EXT4_IND_BLOCK
];
1194 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
1195 i_data
[EXT4_IND_BLOCK
] = 0;
1198 case EXT4_IND_BLOCK
:
1199 nr
= i_data
[EXT4_DIND_BLOCK
];
1201 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
1202 i_data
[EXT4_DIND_BLOCK
] = 0;
1205 case EXT4_DIND_BLOCK
:
1206 nr
= i_data
[EXT4_TIND_BLOCK
];
1208 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
1209 i_data
[EXT4_TIND_BLOCK
] = 0;
1212 case EXT4_TIND_BLOCK
:
1218 * ext4_ind_remove_space - remove space from the range
1219 * @handle: JBD handle for this transaction
1220 * @inode: inode we are dealing with
1221 * @start: First block to remove
1222 * @end: One block after the last block to remove (exclusive)
1224 * Free the blocks in the defined range (end is exclusive endpoint of
1225 * range). This is used by ext4_punch_hole().
1227 int ext4_ind_remove_space(handle_t
*handle
, struct inode
*inode
,
1228 ext4_lblk_t start
, ext4_lblk_t end
)
1230 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1231 __le32
*i_data
= ei
->i_data
;
1232 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1233 ext4_lblk_t offsets
[4], offsets2
[4];
1234 Indirect chain
[4], chain2
[4];
1235 Indirect
*partial
, *partial2
;
1236 Indirect
*p
= NULL
, *p2
= NULL
;
1237 ext4_lblk_t max_block
;
1238 __le32 nr
= 0, nr2
= 0;
1240 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1242 max_block
= (EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
+ blocksize
-1)
1243 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1244 if (end
>= max_block
)
1246 if ((start
>= end
) || (start
> max_block
))
1249 n
= ext4_block_to_path(inode
, start
, offsets
, NULL
);
1250 n2
= ext4_block_to_path(inode
, end
, offsets2
, NULL
);
1254 if ((n
== 1) && (n
== n2
)) {
1255 /* We're punching only within direct block range */
1256 ext4_free_data(handle
, inode
, NULL
, i_data
+ offsets
[0],
1257 i_data
+ offsets2
[0]);
1259 } else if (n2
> n
) {
1261 * Start and end are on a different levels so we're going to
1262 * free partial block at start, and partial block at end of
1263 * the range. If there are some levels in between then
1264 * do_indirects label will take care of that.
1269 * Start is at the direct block level, free
1270 * everything to the end of the level.
1272 ext4_free_data(handle
, inode
, NULL
, i_data
+ offsets
[0],
1273 i_data
+ EXT4_NDIR_BLOCKS
);
1278 partial
= p
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1280 if (partial
== chain
) {
1281 /* Shared branch grows from the inode */
1282 ext4_free_branches(handle
, inode
, NULL
,
1283 &nr
, &nr
+1, (chain
+n
-1) - partial
);
1286 /* Shared branch grows from an indirect block */
1287 BUFFER_TRACE(partial
->bh
, "get_write_access");
1288 ext4_free_branches(handle
, inode
, partial
->bh
,
1290 partial
->p
+1, (chain
+n
-1) - partial
);
1295 * Clear the ends of indirect blocks on the shared branch
1296 * at the start of the range
1298 while (partial
> chain
) {
1299 ext4_free_branches(handle
, inode
, partial
->bh
,
1301 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1302 (chain
+n
-1) - partial
);
1307 partial2
= p2
= ext4_find_shared(inode
, n2
, offsets2
, chain2
, &nr2
);
1309 if (partial2
== chain2
) {
1311 * Remember, end is exclusive so here we're at
1312 * the start of the next level we're not going
1313 * to free. Everything was covered by the start
1320 * ext4_find_shared returns Indirect structure which
1321 * points to the last element which should not be
1322 * removed by truncate. But this is end of the range
1323 * in punch_hole so we need to point to the next element
1329 * Clear the ends of indirect blocks on the shared branch
1330 * at the end of the range
1332 while (partial2
> chain2
) {
1333 ext4_free_branches(handle
, inode
, partial2
->bh
,
1334 (__le32
*)partial2
->bh
->b_data
,
1336 (chain2
+n2
-1) - partial2
);
1342 /* Punch happened within the same level (n == n2) */
1343 partial
= p
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1344 partial2
= p2
= ext4_find_shared(inode
, n2
, offsets2
, chain2
, &nr2
);
1346 /* Free top, but only if partial2 isn't its subtree. */
1348 int level
= min(partial
- chain
, partial2
- chain2
);
1352 for (i
= 0; i
<= level
; i
++) {
1353 if (offsets
[i
] != offsets2
[i
]) {
1360 if (partial
== chain
) {
1361 /* Shared branch grows from the inode */
1362 ext4_free_branches(handle
, inode
, NULL
,
1364 (chain
+n
-1) - partial
);
1367 /* Shared branch grows from an indirect block */
1368 BUFFER_TRACE(partial
->bh
, "get_write_access");
1369 ext4_free_branches(handle
, inode
, partial
->bh
,
1372 (chain
+n
-1) - partial
);
1379 * ext4_find_shared returns Indirect structure which
1380 * points to the last element which should not be
1381 * removed by truncate. But this is end of the range
1382 * in punch_hole so we need to point to the next element
1387 while (partial
> chain
|| partial2
> chain2
) {
1388 int depth
= (chain
+n
-1) - partial
;
1389 int depth2
= (chain2
+n2
-1) - partial2
;
1391 if (partial
> chain
&& partial2
> chain2
&&
1392 partial
->bh
->b_blocknr
== partial2
->bh
->b_blocknr
) {
1394 * We've converged on the same block. Clear the range,
1397 ext4_free_branches(handle
, inode
, partial
->bh
,
1400 (chain
+n
-1) - partial
);
1405 * The start and end partial branches may not be at the same
1406 * level even though the punch happened within one level. So, we
1407 * give them a chance to arrive at the same level, then walk
1408 * them in step with each other until we converge on the same
1411 if (partial
> chain
&& depth
<= depth2
) {
1412 ext4_free_branches(handle
, inode
, partial
->bh
,
1414 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1415 (chain
+n
-1) - partial
);
1418 if (partial2
> chain2
&& depth2
<= depth
) {
1419 ext4_free_branches(handle
, inode
, partial2
->bh
,
1420 (__le32
*)partial2
->bh
->b_data
,
1422 (chain2
+n2
-1) - partial2
);
1428 while (p
&& p
> chain
) {
1429 BUFFER_TRACE(p
->bh
, "call brelse");
1433 while (p2
&& p2
> chain2
) {
1434 BUFFER_TRACE(p2
->bh
, "call brelse");
1441 /* Kill the remaining (whole) subtrees */
1442 switch (offsets
[0]) {
1446 nr
= i_data
[EXT4_IND_BLOCK
];
1448 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
1449 i_data
[EXT4_IND_BLOCK
] = 0;
1452 case EXT4_IND_BLOCK
:
1455 nr
= i_data
[EXT4_DIND_BLOCK
];
1457 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
1458 i_data
[EXT4_DIND_BLOCK
] = 0;
1461 case EXT4_DIND_BLOCK
:
1464 nr
= i_data
[EXT4_TIND_BLOCK
];
1466 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
1467 i_data
[EXT4_TIND_BLOCK
] = 0;
1470 case EXT4_TIND_BLOCK
: