2 * linux/fs/ext4/indirect.c
6 * linux/fs/ext4/inode.c
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
15 * linux/fs/minix/inode.c
17 * Copyright (C) 1991, 1992 Linus Torvalds
19 * Goal-directed block allocation by Stephen Tweedie
20 * (sct@redhat.com), 1993, 1998
23 #include "ext4_jbd2.h"
25 #include <linux/dax.h>
26 #include <linux/uio.h>
28 #include <trace/events/ext4.h>
33 struct buffer_head
*bh
;
36 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
43 * ext4_block_to_path - parse the block number into array of offsets
44 * @inode: inode in question (we are only interested in its superblock)
45 * @i_block: block number to be parsed
46 * @offsets: array to store the offsets in
47 * @boundary: set this non-zero if the referred-to block is likely to be
48 * followed (on disk) by an indirect block.
50 * To store the locations of file's data ext4 uses a data structure common
51 * for UNIX filesystems - tree of pointers anchored in the inode, with
52 * data blocks at leaves and indirect blocks in intermediate nodes.
53 * This function translates the block number into path in that tree -
54 * return value is the path length and @offsets[n] is the offset of
55 * pointer to (n+1)th node in the nth one. If @block is out of range
56 * (negative or too large) warning is printed and zero returned.
58 * Note: function doesn't find node addresses, so no IO is needed. All
59 * we need to know is the capacity of indirect blocks (taken from the
64 * Portability note: the last comparison (check that we fit into triple
65 * indirect block) is spelled differently, because otherwise on an
66 * architecture with 32-bit longs and 8Kb pages we might get into trouble
67 * if our filesystem had 8Kb blocks. We might use long long, but that would
68 * kill us on x86. Oh, well, at least the sign propagation does not matter -
69 * i_block would have to be negative in the very beginning, so we would not
73 static int ext4_block_to_path(struct inode
*inode
,
75 ext4_lblk_t offsets
[4], int *boundary
)
77 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
78 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
79 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
80 indirect_blocks
= ptrs
,
81 double_blocks
= (1 << (ptrs_bits
* 2));
85 if (i_block
< direct_blocks
) {
86 offsets
[n
++] = i_block
;
87 final
= direct_blocks
;
88 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
89 offsets
[n
++] = EXT4_IND_BLOCK
;
90 offsets
[n
++] = i_block
;
92 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
93 offsets
[n
++] = EXT4_DIND_BLOCK
;
94 offsets
[n
++] = i_block
>> ptrs_bits
;
95 offsets
[n
++] = i_block
& (ptrs
- 1);
97 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
98 offsets
[n
++] = EXT4_TIND_BLOCK
;
99 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
100 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
101 offsets
[n
++] = i_block
& (ptrs
- 1);
104 ext4_warning(inode
->i_sb
, "block %lu > max in inode %lu",
105 i_block
+ direct_blocks
+
106 indirect_blocks
+ double_blocks
, inode
->i_ino
);
109 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
114 * ext4_get_branch - read the chain of indirect blocks leading to data
115 * @inode: inode in question
116 * @depth: depth of the chain (1 - direct pointer, etc.)
117 * @offsets: offsets of pointers in inode/indirect blocks
118 * @chain: place to store the result
119 * @err: here we store the error value
121 * Function fills the array of triples <key, p, bh> and returns %NULL
122 * if everything went OK or the pointer to the last filled triple
123 * (incomplete one) otherwise. Upon the return chain[i].key contains
124 * the number of (i+1)-th block in the chain (as it is stored in memory,
125 * i.e. little-endian 32-bit), chain[i].p contains the address of that
126 * number (it points into struct inode for i==0 and into the bh->b_data
127 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
128 * block for i>0 and NULL for i==0. In other words, it holds the block
129 * numbers of the chain, addresses they were taken from (and where we can
130 * verify that chain did not change) and buffer_heads hosting these
133 * Function stops when it stumbles upon zero pointer (absent block)
134 * (pointer to last triple returned, *@err == 0)
135 * or when it gets an IO error reading an indirect block
136 * (ditto, *@err == -EIO)
137 * or when it reads all @depth-1 indirect blocks successfully and finds
138 * the whole chain, all way to the data (returns %NULL, *err == 0).
140 * Need to be called with
141 * down_read(&EXT4_I(inode)->i_data_sem)
143 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
144 ext4_lblk_t
*offsets
,
145 Indirect chain
[4], int *err
)
147 struct super_block
*sb
= inode
->i_sb
;
149 struct buffer_head
*bh
;
153 /* i_data is not going away, no lock needed */
154 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
158 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
164 if (!bh_uptodate_or_lock(bh
)) {
165 if (bh_submit_read(bh
) < 0) {
169 /* validate block references */
170 if (ext4_check_indirect_blockref(inode
, bh
)) {
176 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
190 * ext4_find_near - find a place for allocation with sufficient locality
192 * @ind: descriptor of indirect block.
194 * This function returns the preferred place for block allocation.
195 * It is used when heuristic for sequential allocation fails.
197 * + if there is a block to the left of our position - allocate near it.
198 * + if pointer will live in indirect block - allocate near that block.
199 * + if pointer will live in inode - allocate in the same
202 * In the latter case we colour the starting block by the callers PID to
203 * prevent it from clashing with concurrent allocations for a different inode
204 * in the same block group. The PID is used here so that functionally related
205 * files will be close-by on-disk.
207 * Caller must make sure that @ind is valid and will stay that way.
209 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
211 struct ext4_inode_info
*ei
= EXT4_I(inode
);
212 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
215 /* Try to find previous block */
216 for (p
= ind
->p
- 1; p
>= start
; p
--) {
218 return le32_to_cpu(*p
);
221 /* No such thing, so let's try location of indirect block */
223 return ind
->bh
->b_blocknr
;
226 * It is going to be referred to from the inode itself? OK, just put it
227 * into the same cylinder group then.
229 return ext4_inode_to_goal_block(inode
);
233 * ext4_find_goal - find a preferred place for allocation.
235 * @block: block we want
236 * @partial: pointer to the last triple within a chain
238 * Normally this function find the preferred place for block allocation,
240 * Because this is only used for non-extent files, we limit the block nr
243 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
249 * XXX need to get goal block from mballoc's data structures
252 goal
= ext4_find_near(inode
, partial
);
253 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
258 * ext4_blks_to_allocate - Look up the block map and count the number
259 * of direct blocks need to be allocated for the given branch.
261 * @branch: chain of indirect blocks
262 * @k: number of blocks need for indirect blocks
263 * @blks: number of data blocks to be mapped.
264 * @blocks_to_boundary: the offset in the indirect block
266 * return the total number of blocks to be allocate, including the
267 * direct and indirect blocks.
269 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
270 int blocks_to_boundary
)
272 unsigned int count
= 0;
275 * Simple case, [t,d]Indirect block(s) has not allocated yet
276 * then it's clear blocks on that path have not allocated
279 /* right now we don't handle cross boundary allocation */
280 if (blks
< blocks_to_boundary
+ 1)
283 count
+= blocks_to_boundary
+ 1;
288 while (count
< blks
&& count
<= blocks_to_boundary
&&
289 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
296 * ext4_alloc_branch - allocate and set up a chain of blocks.
297 * @handle: handle for this transaction
299 * @indirect_blks: number of allocated indirect blocks
300 * @blks: number of allocated direct blocks
301 * @goal: preferred place for allocation
302 * @offsets: offsets (in the blocks) to store the pointers to next.
303 * @branch: place to store the chain in.
305 * This function allocates blocks, zeroes out all but the last one,
306 * links them into chain and (if we are synchronous) writes them to disk.
307 * In other words, it prepares a branch that can be spliced onto the
308 * inode. It stores the information about that chain in the branch[], in
309 * the same format as ext4_get_branch() would do. We are calling it after
310 * we had read the existing part of chain and partial points to the last
311 * triple of that (one with zero ->key). Upon the exit we have the same
312 * picture as after the successful ext4_get_block(), except that in one
313 * place chain is disconnected - *branch->p is still zero (we did not
314 * set the last link), but branch->key contains the number that should
315 * be placed into *branch->p to fill that gap.
317 * If allocation fails we free all blocks we've allocated (and forget
318 * their buffer_heads) and return the error value the from failed
319 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
320 * as described above and return 0.
322 static int ext4_alloc_branch(handle_t
*handle
,
323 struct ext4_allocation_request
*ar
,
324 int indirect_blks
, ext4_lblk_t
*offsets
,
327 struct buffer_head
* bh
;
328 ext4_fsblk_t b
, new_blocks
[4];
330 int i
, j
, err
, len
= 1;
332 for (i
= 0; i
<= indirect_blks
; i
++) {
333 if (i
== indirect_blks
) {
334 new_blocks
[i
] = ext4_mb_new_blocks(handle
, ar
, &err
);
336 ar
->goal
= new_blocks
[i
] = ext4_new_meta_blocks(handle
,
338 ar
->flags
& EXT4_MB_DELALLOC_RESERVED
,
344 branch
[i
].key
= cpu_to_le32(new_blocks
[i
]);
348 bh
= branch
[i
].bh
= sb_getblk(ar
->inode
->i_sb
, new_blocks
[i
-1]);
354 BUFFER_TRACE(bh
, "call get_create_access");
355 err
= ext4_journal_get_create_access(handle
, bh
);
361 memset(bh
->b_data
, 0, bh
->b_size
);
362 p
= branch
[i
].p
= (__le32
*) bh
->b_data
+ offsets
[i
];
365 if (i
== indirect_blks
)
367 for (j
= 0; j
< len
; j
++)
368 *p
++ = cpu_to_le32(b
++);
370 BUFFER_TRACE(bh
, "marking uptodate");
371 set_buffer_uptodate(bh
);
374 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
375 err
= ext4_handle_dirty_metadata(handle
, ar
->inode
, bh
);
381 for (; i
>= 0; i
--) {
383 * We want to ext4_forget() only freshly allocated indirect
384 * blocks. Buffer for new_blocks[i-1] is at branch[i].bh and
385 * buffer at branch[0].bh is indirect block / inode already
386 * existing before ext4_alloc_branch() was called.
388 if (i
> 0 && i
!= indirect_blks
&& branch
[i
].bh
)
389 ext4_forget(handle
, 1, ar
->inode
, branch
[i
].bh
,
390 branch
[i
].bh
->b_blocknr
);
391 ext4_free_blocks(handle
, ar
->inode
, NULL
, new_blocks
[i
],
392 (i
== indirect_blks
) ? ar
->len
: 1, 0);
398 * ext4_splice_branch - splice the allocated branch onto inode.
399 * @handle: handle for this transaction
401 * @block: (logical) number of block we are adding
402 * @chain: chain of indirect blocks (with a missing link - see
404 * @where: location of missing link
405 * @num: number of indirect blocks we are adding
406 * @blks: number of direct blocks we are adding
408 * This function fills the missing link and does all housekeeping needed in
409 * inode (->i_blocks, etc.). In case of success we end up with the full
410 * chain to new block and return 0.
412 static int ext4_splice_branch(handle_t
*handle
,
413 struct ext4_allocation_request
*ar
,
414 Indirect
*where
, int num
)
418 ext4_fsblk_t current_block
;
421 * If we're splicing into a [td]indirect block (as opposed to the
422 * inode) then we need to get write access to the [td]indirect block
426 BUFFER_TRACE(where
->bh
, "get_write_access");
427 err
= ext4_journal_get_write_access(handle
, where
->bh
);
433 *where
->p
= where
->key
;
436 * Update the host buffer_head or inode to point to more just allocated
437 * direct blocks blocks
439 if (num
== 0 && ar
->len
> 1) {
440 current_block
= le32_to_cpu(where
->key
) + 1;
441 for (i
= 1; i
< ar
->len
; i
++)
442 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
445 /* We are done with atomic stuff, now do the rest of housekeeping */
446 /* had we spliced it onto indirect block? */
449 * If we spliced it onto an indirect block, we haven't
450 * altered the inode. Note however that if it is being spliced
451 * onto an indirect block at the very end of the file (the
452 * file is growing) then we *will* alter the inode to reflect
453 * the new i_size. But that is not done here - it is done in
454 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
456 jbd_debug(5, "splicing indirect only\n");
457 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
458 err
= ext4_handle_dirty_metadata(handle
, ar
->inode
, where
->bh
);
463 * OK, we spliced it into the inode itself on a direct block.
465 ext4_mark_inode_dirty(handle
, ar
->inode
);
466 jbd_debug(5, "splicing direct\n");
471 for (i
= 1; i
<= num
; i
++) {
473 * branch[i].bh is newly allocated, so there is no
474 * need to revoke the block, which is why we don't
475 * need to set EXT4_FREE_BLOCKS_METADATA.
477 ext4_free_blocks(handle
, ar
->inode
, where
[i
].bh
, 0, 1,
478 EXT4_FREE_BLOCKS_FORGET
);
480 ext4_free_blocks(handle
, ar
->inode
, NULL
, le32_to_cpu(where
[num
].key
),
487 * The ext4_ind_map_blocks() function handles non-extents inodes
488 * (i.e., using the traditional indirect/double-indirect i_blocks
489 * scheme) for ext4_map_blocks().
491 * Allocation strategy is simple: if we have to allocate something, we will
492 * have to go the whole way to leaf. So let's do it before attaching anything
493 * to tree, set linkage between the newborn blocks, write them if sync is
494 * required, recheck the path, free and repeat if check fails, otherwise
495 * set the last missing link (that will protect us from any truncate-generated
496 * removals - all blocks on the path are immune now) and possibly force the
497 * write on the parent block.
498 * That has a nice additional property: no special recovery from the failed
499 * allocations is needed - we simply release blocks and do not touch anything
500 * reachable from inode.
502 * `handle' can be NULL if create == 0.
504 * return > 0, # of blocks mapped or allocated.
505 * return = 0, if plain lookup failed.
506 * return < 0, error case.
508 * The ext4_ind_get_blocks() function should be called with
509 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
510 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
511 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
514 int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
515 struct ext4_map_blocks
*map
,
518 struct ext4_allocation_request ar
;
520 ext4_lblk_t offsets
[4];
524 int blocks_to_boundary
= 0;
527 ext4_fsblk_t first_block
= 0;
529 trace_ext4_ind_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
530 J_ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
531 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
532 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
533 &blocks_to_boundary
);
538 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
540 /* Simplest case - block found, no allocation needed */
542 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
545 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
548 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
550 if (blk
== first_block
+ count
)
558 /* Next simple case - plain lookup failed */
559 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
560 unsigned epb
= inode
->i_sb
->s_blocksize
/ sizeof(u32
);
564 * Count number blocks in a subtree under 'partial'. At each
565 * level we count number of complete empty subtrees beyond
566 * current offset and then descend into the subtree only
567 * partially beyond current offset.
570 for (i
= partial
- chain
+ 1; i
< depth
; i
++)
571 count
= count
* epb
+ (epb
- offsets
[i
] - 1);
573 /* Fill in size of a hole we found */
575 map
->m_len
= min_t(unsigned int, map
->m_len
, count
);
579 /* Failed read of indirect block */
584 * Okay, we need to do block allocation.
586 if (ext4_has_feature_bigalloc(inode
->i_sb
)) {
587 EXT4_ERROR_INODE(inode
, "Can't allocate blocks for "
588 "non-extent mapped inodes with bigalloc");
589 return -EFSCORRUPTED
;
592 /* Set up for the direct block allocation */
593 memset(&ar
, 0, sizeof(ar
));
595 ar
.logical
= map
->m_lblk
;
596 if (S_ISREG(inode
->i_mode
))
597 ar
.flags
= EXT4_MB_HINT_DATA
;
598 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
599 ar
.flags
|= EXT4_MB_DELALLOC_RESERVED
;
600 if (flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
601 ar
.flags
|= EXT4_MB_USE_RESERVED
;
603 ar
.goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
605 /* the number of blocks need to allocate for [d,t]indirect blocks */
606 indirect_blks
= (chain
+ depth
) - partial
- 1;
609 * Next look up the indirect map to count the totoal number of
610 * direct blocks to allocate for this branch.
612 ar
.len
= ext4_blks_to_allocate(partial
, indirect_blks
,
613 map
->m_len
, blocks_to_boundary
);
616 * Block out ext4_truncate while we alter the tree
618 err
= ext4_alloc_branch(handle
, &ar
, indirect_blks
,
619 offsets
+ (partial
- chain
), partial
);
622 * The ext4_splice_branch call will free and forget any buffers
623 * on the new chain if there is a failure, but that risks using
624 * up transaction credits, especially for bitmaps where the
625 * credits cannot be returned. Can we handle this somehow? We
626 * may need to return -EAGAIN upwards in the worst case. --sct
629 err
= ext4_splice_branch(handle
, &ar
, partial
, indirect_blks
);
633 map
->m_flags
|= EXT4_MAP_NEW
;
635 ext4_update_inode_fsync_trans(handle
, inode
, 1);
638 map
->m_flags
|= EXT4_MAP_MAPPED
;
639 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
641 if (count
> blocks_to_boundary
)
642 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
644 /* Clean up and exit */
645 partial
= chain
+ depth
- 1; /* the whole chain */
647 while (partial
> chain
) {
648 BUFFER_TRACE(partial
->bh
, "call brelse");
653 trace_ext4_ind_map_blocks_exit(inode
, flags
, map
, err
);
658 * Calculate the number of metadata blocks need to reserve
659 * to allocate a new block at @lblocks for non extent file based file
661 int ext4_ind_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
663 struct ext4_inode_info
*ei
= EXT4_I(inode
);
664 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
667 if (lblock
< EXT4_NDIR_BLOCKS
)
670 lblock
-= EXT4_NDIR_BLOCKS
;
672 if (ei
->i_da_metadata_calc_len
&&
673 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
674 ei
->i_da_metadata_calc_len
++;
677 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
678 ei
->i_da_metadata_calc_len
= 1;
679 blk_bits
= order_base_2(lblock
);
680 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
684 * Calculate number of indirect blocks touched by mapping @nrblocks logically
687 int ext4_ind_trans_blocks(struct inode
*inode
, int nrblocks
)
690 * With N contiguous data blocks, we need at most
691 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
692 * 2 dindirect blocks, and 1 tindirect block
694 return DIV_ROUND_UP(nrblocks
, EXT4_ADDR_PER_BLOCK(inode
->i_sb
)) + 4;
698 * Truncate transactions can be complex and absolutely huge. So we need to
699 * be able to restart the transaction at a conventient checkpoint to make
700 * sure we don't overflow the journal.
702 * Try to extend this transaction for the purposes of truncation. If
703 * extend fails, we need to propagate the failure up and restart the
704 * transaction in the top-level truncate loop. --sct
706 * Returns 0 if we managed to create more room. If we can't create more
707 * room, and the transaction must be restarted we return 1.
709 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
711 if (!ext4_handle_valid(handle
))
713 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
715 if (!ext4_journal_extend(handle
, ext4_blocks_for_truncate(inode
)))
721 * Probably it should be a library function... search for first non-zero word
722 * or memcmp with zero_page, whatever is better for particular architecture.
725 static inline int all_zeroes(__le32
*p
, __le32
*q
)
734 * ext4_find_shared - find the indirect blocks for partial truncation.
735 * @inode: inode in question
736 * @depth: depth of the affected branch
737 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
738 * @chain: place to store the pointers to partial indirect blocks
739 * @top: place to the (detached) top of branch
741 * This is a helper function used by ext4_truncate().
743 * When we do truncate() we may have to clean the ends of several
744 * indirect blocks but leave the blocks themselves alive. Block is
745 * partially truncated if some data below the new i_size is referred
746 * from it (and it is on the path to the first completely truncated
747 * data block, indeed). We have to free the top of that path along
748 * with everything to the right of the path. Since no allocation
749 * past the truncation point is possible until ext4_truncate()
750 * finishes, we may safely do the latter, but top of branch may
751 * require special attention - pageout below the truncation point
752 * might try to populate it.
754 * We atomically detach the top of branch from the tree, store the
755 * block number of its root in *@top, pointers to buffer_heads of
756 * partially truncated blocks - in @chain[].bh and pointers to
757 * their last elements that should not be removed - in
758 * @chain[].p. Return value is the pointer to last filled element
761 * The work left to caller to do the actual freeing of subtrees:
762 * a) free the subtree starting from *@top
763 * b) free the subtrees whose roots are stored in
764 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
765 * c) free the subtrees growing from the inode past the @chain[0].
766 * (no partially truncated stuff there). */
768 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
769 ext4_lblk_t offsets
[4], Indirect chain
[4],
772 Indirect
*partial
, *p
;
776 /* Make k index the deepest non-null offset + 1 */
777 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
779 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
780 /* Writer: pointers */
782 partial
= chain
+ k
-1;
784 * If the branch acquired continuation since we've looked at it -
785 * fine, it should all survive and (new) top doesn't belong to us.
787 if (!partial
->key
&& *partial
->p
)
790 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
793 * OK, we've found the last block that must survive. The rest of our
794 * branch should be detached before unlocking. However, if that rest
795 * of branch is all ours and does not grow immediately from the inode
796 * it's easier to cheat and just decrement partial->p.
798 if (p
== chain
+ k
- 1 && p
> chain
) {
802 /* Nope, don't do this in ext4. Must leave the tree intact */
809 while (partial
> p
) {
818 * Zero a number of block pointers in either an inode or an indirect block.
819 * If we restart the transaction we must again get write access to the
820 * indirect block for further modification.
822 * We release `count' blocks on disk, but (last - first) may be greater
823 * than `count' because there can be holes in there.
825 * Return 0 on success, 1 on invalid block range
826 * and < 0 on fatal error.
828 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
829 struct buffer_head
*bh
,
830 ext4_fsblk_t block_to_free
,
831 unsigned long count
, __le32
*first
,
835 int flags
= EXT4_FREE_BLOCKS_VALIDATED
;
838 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
839 flags
|= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_METADATA
;
840 else if (ext4_should_journal_data(inode
))
841 flags
|= EXT4_FREE_BLOCKS_FORGET
;
843 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
845 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
846 "blocks %llu len %lu",
847 (unsigned long long) block_to_free
, count
);
851 if (try_to_extend_transaction(handle
, inode
)) {
853 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
854 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
858 err
= ext4_mark_inode_dirty(handle
, inode
);
861 err
= ext4_truncate_restart_trans(handle
, inode
,
862 ext4_blocks_for_truncate(inode
));
866 BUFFER_TRACE(bh
, "retaking write access");
867 err
= ext4_journal_get_write_access(handle
, bh
);
873 for (p
= first
; p
< last
; p
++)
876 ext4_free_blocks(handle
, inode
, NULL
, block_to_free
, count
, flags
);
879 ext4_std_error(inode
->i_sb
, err
);
884 * ext4_free_data - free a list of data blocks
885 * @handle: handle for this transaction
886 * @inode: inode we are dealing with
887 * @this_bh: indirect buffer_head which contains *@first and *@last
888 * @first: array of block numbers
889 * @last: points immediately past the end of array
891 * We are freeing all blocks referred from that array (numbers are stored as
892 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
894 * We accumulate contiguous runs of blocks to free. Conveniently, if these
895 * blocks are contiguous then releasing them at one time will only affect one
896 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
897 * actually use a lot of journal space.
899 * @this_bh will be %NULL if @first and @last point into the inode's direct
902 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
903 struct buffer_head
*this_bh
,
904 __le32
*first
, __le32
*last
)
906 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
907 unsigned long count
= 0; /* Number of blocks in the run */
908 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
911 ext4_fsblk_t nr
; /* Current block # */
912 __le32
*p
; /* Pointer into inode/ind
916 if (this_bh
) { /* For indirect block */
917 BUFFER_TRACE(this_bh
, "get_write_access");
918 err
= ext4_journal_get_write_access(handle
, this_bh
);
919 /* Important: if we can't update the indirect pointers
920 * to the blocks, we can't free them. */
925 for (p
= first
; p
< last
; p
++) {
926 nr
= le32_to_cpu(*p
);
928 /* accumulate blocks to free if they're contiguous */
933 } else if (nr
== block_to_free
+ count
) {
936 err
= ext4_clear_blocks(handle
, inode
, this_bh
,
937 block_to_free
, count
,
948 if (!err
&& count
> 0)
949 err
= ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
950 count
, block_to_free_p
, p
);
956 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
959 * The buffer head should have an attached journal head at this
960 * point. However, if the data is corrupted and an indirect
961 * block pointed to itself, it would have been detached when
962 * the block was cleared. Check for this instead of OOPSing.
964 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
965 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
967 EXT4_ERROR_INODE(inode
,
968 "circular indirect block detected at "
970 (unsigned long long) this_bh
->b_blocknr
);
975 * ext4_free_branches - free an array of branches
976 * @handle: JBD handle for this transaction
977 * @inode: inode we are dealing with
978 * @parent_bh: the buffer_head which contains *@first and *@last
979 * @first: array of block numbers
980 * @last: pointer immediately past the end of array
981 * @depth: depth of the branches to free
983 * We are freeing all blocks referred from these branches (numbers are
984 * stored as little-endian 32-bit) and updating @inode->i_blocks
987 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
988 struct buffer_head
*parent_bh
,
989 __le32
*first
, __le32
*last
, int depth
)
994 if (ext4_handle_is_aborted(handle
))
998 struct buffer_head
*bh
;
999 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1001 while (--p
>= first
) {
1002 nr
= le32_to_cpu(*p
);
1004 continue; /* A hole */
1006 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
1008 EXT4_ERROR_INODE(inode
,
1009 "invalid indirect mapped "
1010 "block %lu (level %d)",
1011 (unsigned long) nr
, depth
);
1015 /* Go read the buffer for the next level down */
1016 bh
= sb_bread(inode
->i_sb
, nr
);
1019 * A read failure? Report error and clear slot
1023 EXT4_ERROR_INODE_BLOCK(inode
, nr
,
1028 /* This zaps the entire block. Bottom up. */
1029 BUFFER_TRACE(bh
, "free child branches");
1030 ext4_free_branches(handle
, inode
, bh
,
1031 (__le32
*) bh
->b_data
,
1032 (__le32
*) bh
->b_data
+ addr_per_block
,
1037 * Everything below this this pointer has been
1038 * released. Now let this top-of-subtree go.
1040 * We want the freeing of this indirect block to be
1041 * atomic in the journal with the updating of the
1042 * bitmap block which owns it. So make some room in
1045 * We zero the parent pointer *after* freeing its
1046 * pointee in the bitmaps, so if extend_transaction()
1047 * for some reason fails to put the bitmap changes and
1048 * the release into the same transaction, recovery
1049 * will merely complain about releasing a free block,
1050 * rather than leaking blocks.
1052 if (ext4_handle_is_aborted(handle
))
1054 if (try_to_extend_transaction(handle
, inode
)) {
1055 ext4_mark_inode_dirty(handle
, inode
);
1056 ext4_truncate_restart_trans(handle
, inode
,
1057 ext4_blocks_for_truncate(inode
));
1061 * The forget flag here is critical because if
1062 * we are journaling (and not doing data
1063 * journaling), we have to make sure a revoke
1064 * record is written to prevent the journal
1065 * replay from overwriting the (former)
1066 * indirect block if it gets reallocated as a
1067 * data block. This must happen in the same
1068 * transaction where the data blocks are
1071 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1072 EXT4_FREE_BLOCKS_METADATA
|
1073 EXT4_FREE_BLOCKS_FORGET
);
1077 * The block which we have just freed is
1078 * pointed to by an indirect block: journal it
1080 BUFFER_TRACE(parent_bh
, "get_write_access");
1081 if (!ext4_journal_get_write_access(handle
,
1084 BUFFER_TRACE(parent_bh
,
1085 "call ext4_handle_dirty_metadata");
1086 ext4_handle_dirty_metadata(handle
,
1093 /* We have reached the bottom of the tree. */
1094 BUFFER_TRACE(parent_bh
, "free data blocks");
1095 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
1099 void ext4_ind_truncate(handle_t
*handle
, struct inode
*inode
)
1101 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1102 __le32
*i_data
= ei
->i_data
;
1103 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1104 ext4_lblk_t offsets
[4];
1109 ext4_lblk_t last_block
, max_block
;
1110 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1112 last_block
= (inode
->i_size
+ blocksize
-1)
1113 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1114 max_block
= (EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
+ blocksize
-1)
1115 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1117 if (last_block
!= max_block
) {
1118 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
1123 ext4_es_remove_extent(inode
, last_block
, EXT_MAX_BLOCKS
- last_block
);
1126 * The orphan list entry will now protect us from any crash which
1127 * occurs before the truncate completes, so it is now safe to propagate
1128 * the new, shorter inode size (held for now in i_size) into the
1129 * on-disk inode. We do this via i_disksize, which is the value which
1130 * ext4 *really* writes onto the disk inode.
1132 ei
->i_disksize
= inode
->i_size
;
1134 if (last_block
== max_block
) {
1136 * It is unnecessary to free any data blocks if last_block is
1137 * equal to the indirect block limit.
1140 } else if (n
== 1) { /* direct blocks */
1141 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
1142 i_data
+ EXT4_NDIR_BLOCKS
);
1146 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1147 /* Kill the top of shared branch (not detached) */
1149 if (partial
== chain
) {
1150 /* Shared branch grows from the inode */
1151 ext4_free_branches(handle
, inode
, NULL
,
1152 &nr
, &nr
+1, (chain
+n
-1) - partial
);
1155 * We mark the inode dirty prior to restart,
1156 * and prior to stop. No need for it here.
1159 /* Shared branch grows from an indirect block */
1160 BUFFER_TRACE(partial
->bh
, "get_write_access");
1161 ext4_free_branches(handle
, inode
, partial
->bh
,
1163 partial
->p
+1, (chain
+n
-1) - partial
);
1166 /* Clear the ends of indirect blocks on the shared branch */
1167 while (partial
> chain
) {
1168 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
1169 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1170 (chain
+n
-1) - partial
);
1171 BUFFER_TRACE(partial
->bh
, "call brelse");
1172 brelse(partial
->bh
);
1176 /* Kill the remaining (whole) subtrees */
1177 switch (offsets
[0]) {
1179 nr
= i_data
[EXT4_IND_BLOCK
];
1181 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
1182 i_data
[EXT4_IND_BLOCK
] = 0;
1184 case EXT4_IND_BLOCK
:
1185 nr
= i_data
[EXT4_DIND_BLOCK
];
1187 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
1188 i_data
[EXT4_DIND_BLOCK
] = 0;
1190 case EXT4_DIND_BLOCK
:
1191 nr
= i_data
[EXT4_TIND_BLOCK
];
1193 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
1194 i_data
[EXT4_TIND_BLOCK
] = 0;
1196 case EXT4_TIND_BLOCK
:
1202 * ext4_ind_remove_space - remove space from the range
1203 * @handle: JBD handle for this transaction
1204 * @inode: inode we are dealing with
1205 * @start: First block to remove
1206 * @end: One block after the last block to remove (exclusive)
1208 * Free the blocks in the defined range (end is exclusive endpoint of
1209 * range). This is used by ext4_punch_hole().
1211 int ext4_ind_remove_space(handle_t
*handle
, struct inode
*inode
,
1212 ext4_lblk_t start
, ext4_lblk_t end
)
1214 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1215 __le32
*i_data
= ei
->i_data
;
1216 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1217 ext4_lblk_t offsets
[4], offsets2
[4];
1218 Indirect chain
[4], chain2
[4];
1219 Indirect
*partial
, *partial2
;
1220 Indirect
*p
= NULL
, *p2
= NULL
;
1221 ext4_lblk_t max_block
;
1222 __le32 nr
= 0, nr2
= 0;
1224 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1226 max_block
= (EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
+ blocksize
-1)
1227 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
1228 if (end
>= max_block
)
1230 if ((start
>= end
) || (start
> max_block
))
1233 n
= ext4_block_to_path(inode
, start
, offsets
, NULL
);
1234 n2
= ext4_block_to_path(inode
, end
, offsets2
, NULL
);
1238 if ((n
== 1) && (n
== n2
)) {
1239 /* We're punching only within direct block range */
1240 ext4_free_data(handle
, inode
, NULL
, i_data
+ offsets
[0],
1241 i_data
+ offsets2
[0]);
1243 } else if (n2
> n
) {
1245 * Start and end are on a different levels so we're going to
1246 * free partial block at start, and partial block at end of
1247 * the range. If there are some levels in between then
1248 * do_indirects label will take care of that.
1253 * Start is at the direct block level, free
1254 * everything to the end of the level.
1256 ext4_free_data(handle
, inode
, NULL
, i_data
+ offsets
[0],
1257 i_data
+ EXT4_NDIR_BLOCKS
);
1262 partial
= p
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1264 if (partial
== chain
) {
1265 /* Shared branch grows from the inode */
1266 ext4_free_branches(handle
, inode
, NULL
,
1267 &nr
, &nr
+1, (chain
+n
-1) - partial
);
1270 /* Shared branch grows from an indirect block */
1271 BUFFER_TRACE(partial
->bh
, "get_write_access");
1272 ext4_free_branches(handle
, inode
, partial
->bh
,
1274 partial
->p
+1, (chain
+n
-1) - partial
);
1279 * Clear the ends of indirect blocks on the shared branch
1280 * at the start of the range
1282 while (partial
> chain
) {
1283 ext4_free_branches(handle
, inode
, partial
->bh
,
1285 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1286 (chain
+n
-1) - partial
);
1291 partial2
= p2
= ext4_find_shared(inode
, n2
, offsets2
, chain2
, &nr2
);
1293 if (partial2
== chain2
) {
1295 * Remember, end is exclusive so here we're at
1296 * the start of the next level we're not going
1297 * to free. Everything was covered by the start
1304 * ext4_find_shared returns Indirect structure which
1305 * points to the last element which should not be
1306 * removed by truncate. But this is end of the range
1307 * in punch_hole so we need to point to the next element
1313 * Clear the ends of indirect blocks on the shared branch
1314 * at the end of the range
1316 while (partial2
> chain2
) {
1317 ext4_free_branches(handle
, inode
, partial2
->bh
,
1318 (__le32
*)partial2
->bh
->b_data
,
1320 (chain2
+n2
-1) - partial2
);
1326 /* Punch happened within the same level (n == n2) */
1327 partial
= p
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
1328 partial2
= p2
= ext4_find_shared(inode
, n2
, offsets2
, chain2
, &nr2
);
1330 /* Free top, but only if partial2 isn't its subtree. */
1332 int level
= min(partial
- chain
, partial2
- chain2
);
1336 for (i
= 0; i
<= level
; i
++) {
1337 if (offsets
[i
] != offsets2
[i
]) {
1344 if (partial
== chain
) {
1345 /* Shared branch grows from the inode */
1346 ext4_free_branches(handle
, inode
, NULL
,
1348 (chain
+n
-1) - partial
);
1351 /* Shared branch grows from an indirect block */
1352 BUFFER_TRACE(partial
->bh
, "get_write_access");
1353 ext4_free_branches(handle
, inode
, partial
->bh
,
1356 (chain
+n
-1) - partial
);
1363 * ext4_find_shared returns Indirect structure which
1364 * points to the last element which should not be
1365 * removed by truncate. But this is end of the range
1366 * in punch_hole so we need to point to the next element
1371 while (partial
> chain
|| partial2
> chain2
) {
1372 int depth
= (chain
+n
-1) - partial
;
1373 int depth2
= (chain2
+n2
-1) - partial2
;
1375 if (partial
> chain
&& partial2
> chain2
&&
1376 partial
->bh
->b_blocknr
== partial2
->bh
->b_blocknr
) {
1378 * We've converged on the same block. Clear the range,
1381 ext4_free_branches(handle
, inode
, partial
->bh
,
1384 (chain
+n
-1) - partial
);
1389 * The start and end partial branches may not be at the same
1390 * level even though the punch happened within one level. So, we
1391 * give them a chance to arrive at the same level, then walk
1392 * them in step with each other until we converge on the same
1395 if (partial
> chain
&& depth
<= depth2
) {
1396 ext4_free_branches(handle
, inode
, partial
->bh
,
1398 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
1399 (chain
+n
-1) - partial
);
1402 if (partial2
> chain2
&& depth2
<= depth
) {
1403 ext4_free_branches(handle
, inode
, partial2
->bh
,
1404 (__le32
*)partial2
->bh
->b_data
,
1406 (chain2
+n2
-1) - partial2
);
1412 while (p
&& p
> chain
) {
1413 BUFFER_TRACE(p
->bh
, "call brelse");
1417 while (p2
&& p2
> chain2
) {
1418 BUFFER_TRACE(p2
->bh
, "call brelse");
1425 /* Kill the remaining (whole) subtrees */
1426 switch (offsets
[0]) {
1430 nr
= i_data
[EXT4_IND_BLOCK
];
1432 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
1433 i_data
[EXT4_IND_BLOCK
] = 0;
1435 case EXT4_IND_BLOCK
:
1438 nr
= i_data
[EXT4_DIND_BLOCK
];
1440 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
1441 i_data
[EXT4_DIND_BLOCK
] = 0;
1443 case EXT4_DIND_BLOCK
:
1446 nr
= i_data
[EXT4_TIND_BLOCK
];
1448 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
1449 i_data
[EXT4_TIND_BLOCK
] = 0;
1451 case EXT4_TIND_BLOCK
: