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
;
154 /* i_data is not going away, no lock needed */
155 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
159 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
165 if (!bh_uptodate_or_lock(bh
)) {
166 if (bh_submit_read(bh
) < 0) {
170 /* validate block references */
171 if (ext4_check_indirect_blockref(inode
, bh
)) {
177 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
191 * ext4_find_near - find a place for allocation with sufficient locality
193 * @ind: descriptor of indirect block.
195 * This function returns the preferred place for block allocation.
196 * It is used when heuristic for sequential allocation fails.
198 * + if there is a block to the left of our position - allocate near it.
199 * + if pointer will live in indirect block - allocate near that block.
200 * + if pointer will live in inode - allocate in the same
203 * In the latter case we colour the starting block by the callers PID to
204 * prevent it from clashing with concurrent allocations for a different inode
205 * in the same block group. The PID is used here so that functionally related
206 * files will be close-by on-disk.
208 * Caller must make sure that @ind is valid and will stay that way.
210 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
212 struct ext4_inode_info
*ei
= EXT4_I(inode
);
213 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
216 /* Try to find previous block */
217 for (p
= ind
->p
- 1; p
>= start
; p
--) {
219 return le32_to_cpu(*p
);
222 /* No such thing, so let's try location of indirect block */
224 return ind
->bh
->b_blocknr
;
227 * It is going to be referred to from the inode itself? OK, just put it
228 * into the same cylinder group then.
230 return ext4_inode_to_goal_block(inode
);
234 * ext4_find_goal - find a preferred place for allocation.
236 * @block: block we want
237 * @partial: pointer to the last triple within a chain
239 * Normally this function find the preferred place for block allocation,
241 * Because this is only used for non-extent files, we limit the block nr
244 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
250 * XXX need to get goal block from mballoc's data structures
253 goal
= ext4_find_near(inode
, partial
);
254 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
259 * ext4_blks_to_allocate - Look up the block map and count the number
260 * of direct blocks need to be allocated for the given branch.
262 * @branch: chain of indirect blocks
263 * @k: number of blocks need for indirect blocks
264 * @blks: number of data blocks to be mapped.
265 * @blocks_to_boundary: the offset in the indirect block
267 * return the total number of blocks to be allocate, including the
268 * direct and indirect blocks.
270 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
271 int blocks_to_boundary
)
273 unsigned int count
= 0;
276 * Simple case, [t,d]Indirect block(s) has not allocated yet
277 * then it's clear blocks on that path have not allocated
280 /* right now we don't handle cross boundary allocation */
281 if (blks
< blocks_to_boundary
+ 1)
284 count
+= blocks_to_boundary
+ 1;
289 while (count
< blks
&& count
<= blocks_to_boundary
&&
290 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
297 * ext4_alloc_branch - allocate and set up a chain of blocks.
298 * @handle: handle for this transaction
300 * @indirect_blks: number of allocated indirect blocks
301 * @blks: number of allocated direct blocks
302 * @goal: preferred place for allocation
303 * @offsets: offsets (in the blocks) to store the pointers to next.
304 * @branch: place to store the chain in.
306 * This function allocates blocks, zeroes out all but the last one,
307 * links them into chain and (if we are synchronous) writes them to disk.
308 * In other words, it prepares a branch that can be spliced onto the
309 * inode. It stores the information about that chain in the branch[], in
310 * the same format as ext4_get_branch() would do. We are calling it after
311 * we had read the existing part of chain and partial points to the last
312 * triple of that (one with zero ->key). Upon the exit we have the same
313 * picture as after the successful ext4_get_block(), except that in one
314 * place chain is disconnected - *branch->p is still zero (we did not
315 * set the last link), but branch->key contains the number that should
316 * be placed into *branch->p to fill that gap.
318 * If allocation fails we free all blocks we've allocated (and forget
319 * their buffer_heads) and return the error value the from failed
320 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
321 * as described above and return 0.
323 static int ext4_alloc_branch(handle_t
*handle
,
324 struct ext4_allocation_request
*ar
,
325 int indirect_blks
, ext4_lblk_t
*offsets
,
328 struct buffer_head
* bh
;
329 ext4_fsblk_t b
, new_blocks
[4];
331 int i
, j
, err
, len
= 1;
333 for (i
= 0; i
<= indirect_blks
; i
++) {
334 if (i
== indirect_blks
) {
335 new_blocks
[i
] = ext4_mb_new_blocks(handle
, ar
, &err
);
337 ar
->goal
= new_blocks
[i
] = ext4_new_meta_blocks(handle
,
339 ar
->flags
& EXT4_MB_DELALLOC_RESERVED
,
345 branch
[i
].key
= cpu_to_le32(new_blocks
[i
]);
349 bh
= branch
[i
].bh
= sb_getblk(ar
->inode
->i_sb
, new_blocks
[i
-1]);
355 BUFFER_TRACE(bh
, "call get_create_access");
356 err
= ext4_journal_get_create_access(handle
, bh
);
362 memset(bh
->b_data
, 0, bh
->b_size
);
363 p
= branch
[i
].p
= (__le32
*) bh
->b_data
+ offsets
[i
];
366 if (i
== indirect_blks
)
368 for (j
= 0; j
< len
; j
++)
369 *p
++ = cpu_to_le32(b
++);
371 BUFFER_TRACE(bh
, "marking uptodate");
372 set_buffer_uptodate(bh
);
375 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
376 err
= ext4_handle_dirty_metadata(handle
, ar
->inode
, bh
);
382 for (; i
>= 0; i
--) {
384 * We want to ext4_forget() only freshly allocated indirect
385 * blocks. Buffer for new_blocks[i-1] is at branch[i].bh and
386 * buffer at branch[0].bh is indirect block / inode already
387 * existing before ext4_alloc_branch() was called.
389 if (i
> 0 && i
!= indirect_blks
&& branch
[i
].bh
)
390 ext4_forget(handle
, 1, ar
->inode
, branch
[i
].bh
,
391 branch
[i
].bh
->b_blocknr
);
392 ext4_free_blocks(handle
, ar
->inode
, NULL
, new_blocks
[i
],
393 (i
== indirect_blks
) ? ar
->len
: 1, 0);
399 * ext4_splice_branch - splice the allocated branch onto inode.
400 * @handle: handle for this transaction
402 * @block: (logical) number of block we are adding
403 * @chain: chain of indirect blocks (with a missing link - see
405 * @where: location of missing link
406 * @num: number of indirect blocks we are adding
407 * @blks: number of direct blocks we are adding
409 * This function fills the missing link and does all housekeeping needed in
410 * inode (->i_blocks, etc.). In case of success we end up with the full
411 * chain to new block and return 0.
413 static int ext4_splice_branch(handle_t
*handle
,
414 struct ext4_allocation_request
*ar
,
415 Indirect
*where
, int num
)
419 ext4_fsblk_t current_block
;
422 * If we're splicing into a [td]indirect block (as opposed to the
423 * inode) then we need to get write access to the [td]indirect block
427 BUFFER_TRACE(where
->bh
, "get_write_access");
428 err
= ext4_journal_get_write_access(handle
, where
->bh
);
434 *where
->p
= where
->key
;
437 * Update the host buffer_head or inode to point to more just allocated
438 * direct blocks blocks
440 if (num
== 0 && ar
->len
> 1) {
441 current_block
= le32_to_cpu(where
->key
) + 1;
442 for (i
= 1; i
< ar
->len
; i
++)
443 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
446 /* We are done with atomic stuff, now do the rest of housekeeping */
447 /* had we spliced it onto indirect block? */
450 * If we spliced it onto an indirect block, we haven't
451 * altered the inode. Note however that if it is being spliced
452 * onto an indirect block at the very end of the file (the
453 * file is growing) then we *will* alter the inode to reflect
454 * the new i_size. But that is not done here - it is done in
455 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
457 jbd_debug(5, "splicing indirect only\n");
458 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
459 err
= ext4_handle_dirty_metadata(handle
, ar
->inode
, where
->bh
);
464 * OK, we spliced it into the inode itself on a direct block.
466 ext4_mark_inode_dirty(handle
, ar
->inode
);
467 jbd_debug(5, "splicing direct\n");
472 for (i
= 1; i
<= num
; i
++) {
474 * branch[i].bh is newly allocated, so there is no
475 * need to revoke the block, which is why we don't
476 * need to set EXT4_FREE_BLOCKS_METADATA.
478 ext4_free_blocks(handle
, ar
->inode
, where
[i
].bh
, 0, 1,
479 EXT4_FREE_BLOCKS_FORGET
);
481 ext4_free_blocks(handle
, ar
->inode
, NULL
, le32_to_cpu(where
[num
].key
),
488 * The ext4_ind_map_blocks() function handles non-extents inodes
489 * (i.e., using the traditional indirect/double-indirect i_blocks
490 * scheme) for ext4_map_blocks().
492 * Allocation strategy is simple: if we have to allocate something, we will
493 * have to go the whole way to leaf. So let's do it before attaching anything
494 * to tree, set linkage between the newborn blocks, write them if sync is
495 * required, recheck the path, free and repeat if check fails, otherwise
496 * set the last missing link (that will protect us from any truncate-generated
497 * removals - all blocks on the path are immune now) and possibly force the
498 * write on the parent block.
499 * That has a nice additional property: no special recovery from the failed
500 * allocations is needed - we simply release blocks and do not touch anything
501 * reachable from inode.
503 * `handle' can be NULL if create == 0.
505 * return > 0, # of blocks mapped or allocated.
506 * return = 0, if plain lookup failed.
507 * return < 0, error case.
509 * The ext4_ind_get_blocks() function should be called with
510 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
511 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
512 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
515 int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
516 struct ext4_map_blocks
*map
,
519 struct ext4_allocation_request ar
;
521 ext4_lblk_t offsets
[4];
525 int blocks_to_boundary
= 0;
528 ext4_fsblk_t first_block
= 0;
530 trace_ext4_ind_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
531 J_ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
532 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
533 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
534 &blocks_to_boundary
);
539 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
541 /* Simplest case - block found, no allocation needed */
543 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
546 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
549 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
551 if (blk
== first_block
+ count
)
559 /* Next simple case - plain lookup failed */
560 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
561 unsigned epb
= inode
->i_sb
->s_blocksize
/ sizeof(u32
);
564 /* Count number blocks in a subtree under 'partial' */
566 for (i
= 0; partial
+ i
!= chain
+ depth
- 1; i
++)
568 /* Fill in size of a hole we found */
570 map
->m_len
= min_t(unsigned int, map
->m_len
, count
);
574 /* Failed read of indirect block */
579 * Okay, we need to do block allocation.
581 if (ext4_has_feature_bigalloc(inode
->i_sb
)) {
582 EXT4_ERROR_INODE(inode
, "Can't allocate blocks for "
583 "non-extent mapped inodes with bigalloc");
584 return -EFSCORRUPTED
;
587 /* Set up for the direct block allocation */
588 memset(&ar
, 0, sizeof(ar
));
590 ar
.logical
= map
->m_lblk
;
591 if (S_ISREG(inode
->i_mode
))
592 ar
.flags
= EXT4_MB_HINT_DATA
;
593 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
594 ar
.flags
|= EXT4_MB_DELALLOC_RESERVED
;
595 if (flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
596 ar
.flags
|= EXT4_MB_USE_RESERVED
;
598 ar
.goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
600 /* the number of blocks need to allocate for [d,t]indirect blocks */
601 indirect_blks
= (chain
+ depth
) - partial
- 1;
604 * Next look up the indirect map to count the totoal number of
605 * direct blocks to allocate for this branch.
607 ar
.len
= ext4_blks_to_allocate(partial
, indirect_blks
,
608 map
->m_len
, blocks_to_boundary
);
611 * Block out ext4_truncate while we alter the tree
613 err
= ext4_alloc_branch(handle
, &ar
, indirect_blks
,
614 offsets
+ (partial
- chain
), partial
);
617 * The ext4_splice_branch call will free and forget any buffers
618 * on the new chain if there is a failure, but that risks using
619 * up transaction credits, especially for bitmaps where the
620 * credits cannot be returned. Can we handle this somehow? We
621 * may need to return -EAGAIN upwards in the worst case. --sct
624 err
= ext4_splice_branch(handle
, &ar
, partial
, indirect_blks
);
628 map
->m_flags
|= EXT4_MAP_NEW
;
630 ext4_update_inode_fsync_trans(handle
, inode
, 1);
633 map
->m_flags
|= EXT4_MAP_MAPPED
;
634 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
636 if (count
> blocks_to_boundary
)
637 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
639 /* Clean up and exit */
640 partial
= chain
+ depth
- 1; /* the whole chain */
642 while (partial
> chain
) {
643 BUFFER_TRACE(partial
->bh
, "call brelse");
648 trace_ext4_ind_map_blocks_exit(inode
, flags
, map
, err
);
653 * Calculate the number of metadata blocks need to reserve
654 * to allocate a new block at @lblocks for non extent file based file
656 int ext4_ind_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
658 struct ext4_inode_info
*ei
= EXT4_I(inode
);
659 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
662 if (lblock
< EXT4_NDIR_BLOCKS
)
665 lblock
-= EXT4_NDIR_BLOCKS
;
667 if (ei
->i_da_metadata_calc_len
&&
668 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
669 ei
->i_da_metadata_calc_len
++;
672 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
673 ei
->i_da_metadata_calc_len
= 1;
674 blk_bits
= order_base_2(lblock
);
675 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
679 * Calculate number of indirect blocks touched by mapping @nrblocks logically
682 int ext4_ind_trans_blocks(struct inode
*inode
, int nrblocks
)
685 * With N contiguous data blocks, we need at most
686 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
687 * 2 dindirect blocks, and 1 tindirect block
689 return DIV_ROUND_UP(nrblocks
, EXT4_ADDR_PER_BLOCK(inode
->i_sb
)) + 4;
693 * Truncate transactions can be complex and absolutely huge. So we need to
694 * be able to restart the transaction at a conventient checkpoint to make
695 * sure we don't overflow the journal.
697 * Try to extend this transaction for the purposes of truncation. If
698 * extend fails, we need to propagate the failure up and restart the
699 * transaction in the top-level truncate loop. --sct
701 * Returns 0 if we managed to create more room. If we can't create more
702 * room, and the transaction must be restarted we return 1.
704 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
706 if (!ext4_handle_valid(handle
))
708 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
710 if (!ext4_journal_extend(handle
, ext4_blocks_for_truncate(inode
)))
716 * Probably it should be a library function... search for first non-zero word
717 * or memcmp with zero_page, whatever is better for particular architecture.
720 static inline int all_zeroes(__le32
*p
, __le32
*q
)
729 * ext4_find_shared - find the indirect blocks for partial truncation.
730 * @inode: inode in question
731 * @depth: depth of the affected branch
732 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
733 * @chain: place to store the pointers to partial indirect blocks
734 * @top: place to the (detached) top of branch
736 * This is a helper function used by ext4_truncate().
738 * When we do truncate() we may have to clean the ends of several
739 * indirect blocks but leave the blocks themselves alive. Block is
740 * partially truncated if some data below the new i_size is referred
741 * from it (and it is on the path to the first completely truncated
742 * data block, indeed). We have to free the top of that path along
743 * with everything to the right of the path. Since no allocation
744 * past the truncation point is possible until ext4_truncate()
745 * finishes, we may safely do the latter, but top of branch may
746 * require special attention - pageout below the truncation point
747 * might try to populate it.
749 * We atomically detach the top of branch from the tree, store the
750 * block number of its root in *@top, pointers to buffer_heads of
751 * partially truncated blocks - in @chain[].bh and pointers to
752 * their last elements that should not be removed - in
753 * @chain[].p. Return value is the pointer to last filled element
756 * The work left to caller to do the actual freeing of subtrees:
757 * a) free the subtree starting from *@top
758 * b) free the subtrees whose roots are stored in
759 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
760 * c) free the subtrees growing from the inode past the @chain[0].
761 * (no partially truncated stuff there). */
763 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
764 ext4_lblk_t offsets
[4], Indirect chain
[4],
767 Indirect
*partial
, *p
;
771 /* Make k index the deepest non-null offset + 1 */
772 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
774 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
775 /* Writer: pointers */
777 partial
= chain
+ k
-1;
779 * If the branch acquired continuation since we've looked at it -
780 * fine, it should all survive and (new) top doesn't belong to us.
782 if (!partial
->key
&& *partial
->p
)
785 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
788 * OK, we've found the last block that must survive. The rest of our
789 * branch should be detached before unlocking. However, if that rest
790 * of branch is all ours and does not grow immediately from the inode
791 * it's easier to cheat and just decrement partial->p.
793 if (p
== chain
+ k
- 1 && p
> chain
) {
797 /* Nope, don't do this in ext4. Must leave the tree intact */
804 while (partial
> p
) {
813 * Zero a number of block pointers in either an inode or an indirect block.
814 * If we restart the transaction we must again get write access to the
815 * indirect block for further modification.
817 * We release `count' blocks on disk, but (last - first) may be greater
818 * than `count' because there can be holes in there.
820 * Return 0 on success, 1 on invalid block range
821 * and < 0 on fatal error.
823 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
824 struct buffer_head
*bh
,
825 ext4_fsblk_t block_to_free
,
826 unsigned long count
, __le32
*first
,
830 int flags
= EXT4_FREE_BLOCKS_VALIDATED
;
833 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
) ||
834 ext4_test_inode_flag(inode
, EXT4_INODE_EA_INODE
))
835 flags
|= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_METADATA
;
836 else if (ext4_should_journal_data(inode
))
837 flags
|= EXT4_FREE_BLOCKS_FORGET
;
839 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
841 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
842 "blocks %llu len %lu",
843 (unsigned long long) block_to_free
, count
);
847 if (try_to_extend_transaction(handle
, inode
)) {
849 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
850 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
854 err
= ext4_mark_inode_dirty(handle
, inode
);
857 err
= ext4_truncate_restart_trans(handle
, inode
,
858 ext4_blocks_for_truncate(inode
));
862 BUFFER_TRACE(bh
, "retaking write access");
863 err
= ext4_journal_get_write_access(handle
, bh
);
869 for (p
= first
; p
< last
; p
++)
872 ext4_free_blocks(handle
, inode
, NULL
, block_to_free
, count
, flags
);
875 ext4_std_error(inode
->i_sb
, err
);
880 * ext4_free_data - free a list of data blocks
881 * @handle: handle for this transaction
882 * @inode: inode we are dealing with
883 * @this_bh: indirect buffer_head which contains *@first and *@last
884 * @first: array of block numbers
885 * @last: points immediately past the end of array
887 * We are freeing all blocks referred from that array (numbers are stored as
888 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
890 * We accumulate contiguous runs of blocks to free. Conveniently, if these
891 * blocks are contiguous then releasing them at one time will only affect one
892 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
893 * actually use a lot of journal space.
895 * @this_bh will be %NULL if @first and @last point into the inode's direct
898 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
899 struct buffer_head
*this_bh
,
900 __le32
*first
, __le32
*last
)
902 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
903 unsigned long count
= 0; /* Number of blocks in the run */
904 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
907 ext4_fsblk_t nr
; /* Current block # */
908 __le32
*p
; /* Pointer into inode/ind
912 if (this_bh
) { /* For indirect block */
913 BUFFER_TRACE(this_bh
, "get_write_access");
914 err
= ext4_journal_get_write_access(handle
, this_bh
);
915 /* Important: if we can't update the indirect pointers
916 * to the blocks, we can't free them. */
921 for (p
= first
; p
< last
; p
++) {
922 nr
= le32_to_cpu(*p
);
924 /* accumulate blocks to free if they're contiguous */
929 } else if (nr
== block_to_free
+ count
) {
932 err
= ext4_clear_blocks(handle
, inode
, this_bh
,
933 block_to_free
, count
,
944 if (!err
&& count
> 0)
945 err
= ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
946 count
, block_to_free_p
, p
);
952 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
955 * The buffer head should have an attached journal head at this
956 * point. However, if the data is corrupted and an indirect
957 * block pointed to itself, it would have been detached when
958 * the block was cleared. Check for this instead of OOPSing.
960 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
961 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
963 EXT4_ERROR_INODE(inode
,
964 "circular indirect block detected at "
966 (unsigned long long) this_bh
->b_blocknr
);
971 * ext4_free_branches - free an array of branches
972 * @handle: JBD handle for this transaction
973 * @inode: inode we are dealing with
974 * @parent_bh: the buffer_head which contains *@first and *@last
975 * @first: array of block numbers
976 * @last: pointer immediately past the end of array
977 * @depth: depth of the branches to free
979 * We are freeing all blocks referred from these branches (numbers are
980 * stored as little-endian 32-bit) and updating @inode->i_blocks
983 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
984 struct buffer_head
*parent_bh
,
985 __le32
*first
, __le32
*last
, int depth
)
990 if (ext4_handle_is_aborted(handle
))
994 struct buffer_head
*bh
;
995 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
997 while (--p
>= first
) {
998 nr
= le32_to_cpu(*p
);
1000 continue; /* A hole */
1002 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
1004 EXT4_ERROR_INODE(inode
,
1005 "invalid indirect mapped "
1006 "block %lu (level %d)",
1007 (unsigned long) nr
, depth
);
1011 /* Go read the buffer for the next level down */
1012 bh
= sb_bread(inode
->i_sb
, nr
);
1015 * A read failure? Report error and clear slot
1019 EXT4_ERROR_INODE_BLOCK(inode
, nr
,
1024 /* This zaps the entire block. Bottom up. */
1025 BUFFER_TRACE(bh
, "free child branches");
1026 ext4_free_branches(handle
, inode
, bh
,
1027 (__le32
*) bh
->b_data
,
1028 (__le32
*) bh
->b_data
+ addr_per_block
,
1033 * Everything below this this pointer has been
1034 * released. Now let this top-of-subtree go.
1036 * We want the freeing of this indirect block to be
1037 * atomic in the journal with the updating of the
1038 * bitmap block which owns it. So make some room in
1041 * We zero the parent pointer *after* freeing its
1042 * pointee in the bitmaps, so if extend_transaction()
1043 * for some reason fails to put the bitmap changes and
1044 * the release into the same transaction, recovery
1045 * will merely complain about releasing a free block,
1046 * rather than leaking blocks.
1048 if (ext4_handle_is_aborted(handle
))
1050 if (try_to_extend_transaction(handle
, inode
)) {
1051 ext4_mark_inode_dirty(handle
, inode
);
1052 ext4_truncate_restart_trans(handle
, inode
,
1053 ext4_blocks_for_truncate(inode
));
1057 * The forget flag here is critical because if
1058 * we are journaling (and not doing data
1059 * journaling), we have to make sure a revoke
1060 * record is written to prevent the journal
1061 * replay from overwriting the (former)
1062 * indirect block if it gets reallocated as a
1063 * data block. This must happen in the same
1064 * transaction where the data blocks are
1067 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1068 EXT4_FREE_BLOCKS_METADATA
|
1069 EXT4_FREE_BLOCKS_FORGET
);
1073 * The block which we have just freed is
1074 * pointed to by an indirect block: journal it
1076 BUFFER_TRACE(parent_bh
, "get_write_access");
1077 if (!ext4_journal_get_write_access(handle
,
1080 BUFFER_TRACE(parent_bh
,
1081 "call ext4_handle_dirty_metadata");
1082 ext4_handle_dirty_metadata(handle
,
1089 /* We have reached the bottom of the tree. */
1090 BUFFER_TRACE(parent_bh
, "free data blocks");
1091 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
1095 void ext4_ind_truncate(handle_t
*handle
, struct inode
*inode
)
1097 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1098 __le32
*i_data
= ei
->i_data
;
1099 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1100 ext4_lblk_t offsets
[4];
1105 ext4_lblk_t last_block
, max_block
;
1106 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1108 last_block
= (inode
->i_size
+ blocksize
-1)
1109 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1110 max_block
= (EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
+ blocksize
-1)
1111 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1113 if (last_block
!= max_block
) {
1114 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
1119 ext4_es_remove_extent(inode
, last_block
, EXT_MAX_BLOCKS
- last_block
);
1122 * The orphan list entry will now protect us from any crash which
1123 * occurs before the truncate completes, so it is now safe to propagate
1124 * the new, shorter inode size (held for now in i_size) into the
1125 * on-disk inode. We do this via i_disksize, which is the value which
1126 * ext4 *really* writes onto the disk inode.
1128 ei
->i_disksize
= inode
->i_size
;
1130 if (last_block
== max_block
) {
1132 * It is unnecessary to free any data blocks if last_block is
1133 * equal to the indirect block limit.
1136 } else if (n
== 1) { /* direct blocks */
1137 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
1138 i_data
+ EXT4_NDIR_BLOCKS
);
1142 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1143 /* Kill the top of shared branch (not detached) */
1145 if (partial
== chain
) {
1146 /* Shared branch grows from the inode */
1147 ext4_free_branches(handle
, inode
, NULL
,
1148 &nr
, &nr
+1, (chain
+n
-1) - partial
);
1151 * We mark the inode dirty prior to restart,
1152 * and prior to stop. No need for it here.
1155 /* Shared branch grows from an indirect block */
1156 BUFFER_TRACE(partial
->bh
, "get_write_access");
1157 ext4_free_branches(handle
, inode
, partial
->bh
,
1159 partial
->p
+1, (chain
+n
-1) - partial
);
1162 /* Clear the ends of indirect blocks on the shared branch */
1163 while (partial
> chain
) {
1164 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
1165 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1166 (chain
+n
-1) - partial
);
1167 BUFFER_TRACE(partial
->bh
, "call brelse");
1168 brelse(partial
->bh
);
1172 /* Kill the remaining (whole) subtrees */
1173 switch (offsets
[0]) {
1175 nr
= i_data
[EXT4_IND_BLOCK
];
1177 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
1178 i_data
[EXT4_IND_BLOCK
] = 0;
1180 case EXT4_IND_BLOCK
:
1181 nr
= i_data
[EXT4_DIND_BLOCK
];
1183 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
1184 i_data
[EXT4_DIND_BLOCK
] = 0;
1186 case EXT4_DIND_BLOCK
:
1187 nr
= i_data
[EXT4_TIND_BLOCK
];
1189 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
1190 i_data
[EXT4_TIND_BLOCK
] = 0;
1192 case EXT4_TIND_BLOCK
:
1198 * ext4_ind_remove_space - remove space from the range
1199 * @handle: JBD handle for this transaction
1200 * @inode: inode we are dealing with
1201 * @start: First block to remove
1202 * @end: One block after the last block to remove (exclusive)
1204 * Free the blocks in the defined range (end is exclusive endpoint of
1205 * range). This is used by ext4_punch_hole().
1207 int ext4_ind_remove_space(handle_t
*handle
, struct inode
*inode
,
1208 ext4_lblk_t start
, ext4_lblk_t end
)
1210 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1211 __le32
*i_data
= ei
->i_data
;
1212 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1213 ext4_lblk_t offsets
[4], offsets2
[4];
1214 Indirect chain
[4], chain2
[4];
1215 Indirect
*partial
, *partial2
;
1216 ext4_lblk_t max_block
;
1217 __le32 nr
= 0, nr2
= 0;
1219 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1221 max_block
= (EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
+ blocksize
-1)
1222 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1223 if (end
>= max_block
)
1225 if ((start
>= end
) || (start
> max_block
))
1228 n
= ext4_block_to_path(inode
, start
, offsets
, NULL
);
1229 n2
= ext4_block_to_path(inode
, end
, offsets2
, NULL
);
1233 if ((n
== 1) && (n
== n2
)) {
1234 /* We're punching only within direct block range */
1235 ext4_free_data(handle
, inode
, NULL
, i_data
+ offsets
[0],
1236 i_data
+ offsets2
[0]);
1238 } else if (n2
> n
) {
1240 * Start and end are on a different levels so we're going to
1241 * free partial block at start, and partial block at end of
1242 * the range. If there are some levels in between then
1243 * do_indirects label will take care of that.
1248 * Start is at the direct block level, free
1249 * everything to the end of the level.
1251 ext4_free_data(handle
, inode
, NULL
, i_data
+ offsets
[0],
1252 i_data
+ EXT4_NDIR_BLOCKS
);
1257 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1259 if (partial
== chain
) {
1260 /* Shared branch grows from the inode */
1261 ext4_free_branches(handle
, inode
, NULL
,
1262 &nr
, &nr
+1, (chain
+n
-1) - partial
);
1265 /* Shared branch grows from an indirect block */
1266 BUFFER_TRACE(partial
->bh
, "get_write_access");
1267 ext4_free_branches(handle
, inode
, partial
->bh
,
1269 partial
->p
+1, (chain
+n
-1) - partial
);
1274 * Clear the ends of indirect blocks on the shared branch
1275 * at the start of the range
1277 while (partial
> chain
) {
1278 ext4_free_branches(handle
, inode
, partial
->bh
,
1280 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1281 (chain
+n
-1) - partial
);
1282 BUFFER_TRACE(partial
->bh
, "call brelse");
1283 brelse(partial
->bh
);
1288 partial2
= ext4_find_shared(inode
, n2
, offsets2
, chain2
, &nr2
);
1290 if (partial2
== chain2
) {
1292 * Remember, end is exclusive so here we're at
1293 * the start of the next level we're not going
1294 * to free. Everything was covered by the start
1301 * ext4_find_shared returns Indirect structure which
1302 * points to the last element which should not be
1303 * removed by truncate. But this is end of the range
1304 * in punch_hole so we need to point to the next element
1310 * Clear the ends of indirect blocks on the shared branch
1311 * at the end of the range
1313 while (partial2
> chain2
) {
1314 ext4_free_branches(handle
, inode
, partial2
->bh
,
1315 (__le32
*)partial2
->bh
->b_data
,
1317 (chain2
+n2
-1) - partial2
);
1318 BUFFER_TRACE(partial2
->bh
, "call brelse");
1319 brelse(partial2
->bh
);
1325 /* Punch happened within the same level (n == n2) */
1326 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1327 partial2
= ext4_find_shared(inode
, n2
, offsets2
, chain2
, &nr2
);
1329 /* Free top, but only if partial2 isn't its subtree. */
1331 int level
= min(partial
- chain
, partial2
- chain2
);
1335 for (i
= 0; i
<= level
; i
++) {
1336 if (offsets
[i
] != offsets2
[i
]) {
1343 if (partial
== chain
) {
1344 /* Shared branch grows from the inode */
1345 ext4_free_branches(handle
, inode
, NULL
,
1347 (chain
+n
-1) - partial
);
1350 /* Shared branch grows from an indirect block */
1351 BUFFER_TRACE(partial
->bh
, "get_write_access");
1352 ext4_free_branches(handle
, inode
, partial
->bh
,
1355 (chain
+n
-1) - partial
);
1362 * ext4_find_shared returns Indirect structure which
1363 * points to the last element which should not be
1364 * removed by truncate. But this is end of the range
1365 * in punch_hole so we need to point to the next element
1370 while (partial
> chain
|| partial2
> chain2
) {
1371 int depth
= (chain
+n
-1) - partial
;
1372 int depth2
= (chain2
+n2
-1) - partial2
;
1374 if (partial
> chain
&& partial2
> chain2
&&
1375 partial
->bh
->b_blocknr
== partial2
->bh
->b_blocknr
) {
1377 * We've converged on the same block. Clear the range,
1380 ext4_free_branches(handle
, inode
, partial
->bh
,
1383 (chain
+n
-1) - partial
);
1384 BUFFER_TRACE(partial
->bh
, "call brelse");
1385 brelse(partial
->bh
);
1386 BUFFER_TRACE(partial2
->bh
, "call brelse");
1387 brelse(partial2
->bh
);
1392 * The start and end partial branches may not be at the same
1393 * level even though the punch happened within one level. So, we
1394 * give them a chance to arrive at the same level, then walk
1395 * them in step with each other until we converge on the same
1398 if (partial
> chain
&& depth
<= depth2
) {
1399 ext4_free_branches(handle
, inode
, partial
->bh
,
1401 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1402 (chain
+n
-1) - partial
);
1403 BUFFER_TRACE(partial
->bh
, "call brelse");
1404 brelse(partial
->bh
);
1407 if (partial2
> chain2
&& depth2
<= depth
) {
1408 ext4_free_branches(handle
, inode
, partial2
->bh
,
1409 (__le32
*)partial2
->bh
->b_data
,
1411 (chain2
+n2
-1) - partial2
);
1412 BUFFER_TRACE(partial2
->bh
, "call brelse");
1413 brelse(partial2
->bh
);
1420 /* Kill the remaining (whole) subtrees */
1421 switch (offsets
[0]) {
1425 nr
= i_data
[EXT4_IND_BLOCK
];
1427 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
1428 i_data
[EXT4_IND_BLOCK
] = 0;
1430 case EXT4_IND_BLOCK
:
1433 nr
= i_data
[EXT4_DIND_BLOCK
];
1435 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
1436 i_data
[EXT4_DIND_BLOCK
] = 0;
1438 case EXT4_DIND_BLOCK
:
1441 nr
= i_data
[EXT4_TIND_BLOCK
];
1443 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
1444 i_data
[EXT4_TIND_BLOCK
] = 0;
1446 case EXT4_TIND_BLOCK
: