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) < 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);
655 map->m_flags |= EXT4_MAP_MAPPED;
656 map->m_pblk = le32_to_cpu(chain[depth-1].key);
658 if (count > blocks_to_boundary)
659 map->m_flags |= EXT4_MAP_BOUNDARY;
661 /* Clean up and exit */
662 partial = chain + depth - 1; /* the whole chain */
664 while (partial > chain) {
665 BUFFER_TRACE(partial->bh, "call brelse");
670 trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
675 * Calculate number of indirect blocks touched by mapping @nrblocks logically
678 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
681 * With N contiguous data blocks, we need at most
682 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
683 * 2 dindirect blocks, and 1 tindirect block
685 return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
688 static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
689 struct buffer_head *bh, int *dropped)
694 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
695 err = ext4_handle_dirty_metadata(handle, inode, bh);
699 err = ext4_mark_inode_dirty(handle, inode);
703 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
704 * moment, get_block can be called only for blocks inside i_size since
705 * page cache has been already dropped and writes are blocked by
706 * i_rwsem. So we can safely drop the i_data_sem here.
708 BUG_ON(EXT4_JOURNAL(inode) == NULL);
709 ext4_discard_preallocations(inode, 0);
710 up_write(&EXT4_I(inode)->i_data_sem);
716 * Truncate transactions can be complex and absolutely huge. So we need to
717 * be able to restart the transaction at a convenient checkpoint to make
718 * sure we don't overflow the journal.
720 * Try to extend this transaction for the purposes of truncation. If
721 * extend fails, we restart transaction.
723 static int ext4_ind_truncate_ensure_credits(handle_t *handle,
725 struct buffer_head *bh,
731 ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS,
732 ext4_blocks_for_truncate(inode), revoke_creds,
733 ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped));
735 down_write(&EXT4_I(inode)->i_data_sem);
739 BUFFER_TRACE(bh, "retaking write access");
740 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
749 * Probably it should be a library function... search for first non-zero word
750 * or memcmp with zero_page, whatever is better for particular architecture.
753 static inline int all_zeroes(__le32 *p, __le32 *q)
762 * ext4_find_shared - find the indirect blocks for partial truncation.
763 * @inode: inode in question
764 * @depth: depth of the affected branch
765 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
766 * @chain: place to store the pointers to partial indirect blocks
767 * @top: place to the (detached) top of branch
769 * This is a helper function used by ext4_truncate().
771 * When we do truncate() we may have to clean the ends of several
772 * indirect blocks but leave the blocks themselves alive. Block is
773 * partially truncated if some data below the new i_size is referred
774 * from it (and it is on the path to the first completely truncated
775 * data block, indeed). We have to free the top of that path along
776 * with everything to the right of the path. Since no allocation
777 * past the truncation point is possible until ext4_truncate()
778 * finishes, we may safely do the latter, but top of branch may
779 * require special attention - pageout below the truncation point
780 * might try to populate it.
782 * We atomically detach the top of branch from the tree, store the
783 * block number of its root in *@top, pointers to buffer_heads of
784 * partially truncated blocks - in @chain[].bh and pointers to
785 * their last elements that should not be removed - in
786 * @chain[].p. Return value is the pointer to last filled element
789 * The work left to caller to do the actual freeing of subtrees:
790 * a) free the subtree starting from *@top
791 * b) free the subtrees whose roots are stored in
792 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
793 * c) free the subtrees growing from the inode past the @chain[0].
794 * (no partially truncated stuff there). */
796 static Indirect *ext4_find_shared(struct inode *inode, int depth,
797 ext4_lblk_t offsets[4], Indirect chain[4],
800 Indirect *partial, *p;
804 /* Make k index the deepest non-null offset + 1 */
805 for (k = depth; k > 1 && !offsets[k-1]; k--)
807 partial = ext4_get_branch(inode, k, offsets, chain, &err);
808 /* Writer: pointers */
810 partial = chain + k-1;
812 * If the branch acquired continuation since we've looked at it -
813 * fine, it should all survive and (new) top doesn't belong to us.
815 if (!partial->key && *partial->p)
818 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
821 * OK, we've found the last block that must survive. The rest of our
822 * branch should be detached before unlocking. However, if that rest
823 * of branch is all ours and does not grow immediately from the inode
824 * it's easier to cheat and just decrement partial->p.
826 if (p == chain + k - 1 && p > chain) {
830 /* Nope, don't do this in ext4. Must leave the tree intact */
837 while (partial > p) {
846 * Zero a number of block pointers in either an inode or an indirect block.
847 * If we restart the transaction we must again get write access to the
848 * indirect block for further modification.
850 * We release `count' blocks on disk, but (last - first) may be greater
851 * than `count' because there can be holes in there.
853 * Return 0 on success, 1 on invalid block range
854 * and < 0 on fatal error.
856 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
857 struct buffer_head *bh,
858 ext4_fsblk_t block_to_free,
859 unsigned long count, __le32 *first,
863 int flags = EXT4_FREE_BLOCKS_VALIDATED;
866 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
867 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
868 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
869 else if (ext4_should_journal_data(inode))
870 flags |= EXT4_FREE_BLOCKS_FORGET;
872 if (!ext4_inode_block_valid(inode, block_to_free, count)) {
873 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
874 "blocks %llu len %lu",
875 (unsigned long long) block_to_free, count);
879 err = ext4_ind_truncate_ensure_credits(handle, inode, bh,
880 ext4_free_data_revoke_credits(inode, count));
884 for (p = first; p < last; p++)
887 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
890 ext4_std_error(inode->i_sb, err);
895 * ext4_free_data - free a list of data blocks
896 * @handle: handle for this transaction
897 * @inode: inode we are dealing with
898 * @this_bh: indirect buffer_head which contains *@first and *@last
899 * @first: array of block numbers
900 * @last: points immediately past the end of array
902 * We are freeing all blocks referred from that array (numbers are stored as
903 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
905 * We accumulate contiguous runs of blocks to free. Conveniently, if these
906 * blocks are contiguous then releasing them at one time will only affect one
907 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
908 * actually use a lot of journal space.
910 * @this_bh will be %NULL if @first and @last point into the inode's direct
913 static void ext4_free_data(handle_t *handle, struct inode *inode,
914 struct buffer_head *this_bh,
915 __le32 *first, __le32 *last)
917 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
918 unsigned long count = 0; /* Number of blocks in the run */
919 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
922 ext4_fsblk_t nr; /* Current block # */
923 __le32 *p; /* Pointer into inode/ind
927 if (this_bh) { /* For indirect block */
928 BUFFER_TRACE(this_bh, "get_write_access");
929 err = ext4_journal_get_write_access(handle, inode->i_sb,
930 this_bh, EXT4_JTR_NONE);
931 /* Important: if we can't update the indirect pointers
932 * to the blocks, we can't free them. */
937 for (p = first; p < last; p++) {
938 nr = le32_to_cpu(*p);
940 /* accumulate blocks to free if they're contiguous */
945 } else if (nr == block_to_free + count) {
948 err = ext4_clear_blocks(handle, inode, this_bh,
949 block_to_free, count,
960 if (!err && count > 0)
961 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
962 count, block_to_free_p, p);
968 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
971 * The buffer head should have an attached journal head at this
972 * point. However, if the data is corrupted and an indirect
973 * block pointed to itself, it would have been detached when
974 * the block was cleared. Check for this instead of OOPSing.
976 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
977 ext4_handle_dirty_metadata(handle, inode, this_bh);
979 EXT4_ERROR_INODE(inode,
980 "circular indirect block detected at "
982 (unsigned long long) this_bh->b_blocknr);
987 * ext4_free_branches - free an array of branches
988 * @handle: JBD handle for this transaction
989 * @inode: inode we are dealing with
990 * @parent_bh: the buffer_head which contains *@first and *@last
991 * @first: array of block numbers
992 * @last: pointer immediately past the end of array
993 * @depth: depth of the branches to free
995 * We are freeing all blocks referred from these branches (numbers are
996 * stored as little-endian 32-bit) and updating @inode->i_blocks
999 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1000 struct buffer_head *parent_bh,
1001 __le32 *first, __le32 *last, int depth)
1006 if (ext4_handle_is_aborted(handle))
1010 struct buffer_head *bh;
1011 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1013 while (--p >= first) {
1014 nr = le32_to_cpu(*p);
1016 continue; /* A hole */
1018 if (!ext4_inode_block_valid(inode, nr, 1)) {
1019 EXT4_ERROR_INODE(inode,
1020 "invalid indirect mapped "
1021 "block %lu (level %d)",
1022 (unsigned long) nr, depth);
1026 /* Go read the buffer for the next level down */
1027 bh = ext4_sb_bread(inode->i_sb, nr, 0);
1030 * A read failure? Report error and clear slot
1034 ext4_error_inode_block(inode, nr, -PTR_ERR(bh),
1039 /* This zaps the entire block. Bottom up. */
1040 BUFFER_TRACE(bh, "free child branches");
1041 ext4_free_branches(handle, inode, bh,
1042 (__le32 *) bh->b_data,
1043 (__le32 *) bh->b_data + addr_per_block,
1048 * Everything below this pointer has been
1049 * released. Now let this top-of-subtree go.
1051 * We want the freeing of this indirect block to be
1052 * atomic in the journal with the updating of the
1053 * bitmap block which owns it. So make some room in
1056 * We zero the parent pointer *after* freeing its
1057 * pointee in the bitmaps, so if extend_transaction()
1058 * for some reason fails to put the bitmap changes and
1059 * the release into the same transaction, recovery
1060 * will merely complain about releasing a free block,
1061 * rather than leaking blocks.
1063 if (ext4_handle_is_aborted(handle))
1065 if (ext4_ind_truncate_ensure_credits(handle, inode,
1067 ext4_free_metadata_revoke_credits(
1068 inode->i_sb, 1)) < 0)
1072 * The forget flag here is critical because if
1073 * we are journaling (and not doing data
1074 * journaling), we have to make sure a revoke
1075 * record is written to prevent the journal
1076 * replay from overwriting the (former)
1077 * indirect block if it gets reallocated as a
1078 * data block. This must happen in the same
1079 * transaction where the data blocks are
1082 ext4_free_blocks(handle, inode, NULL, nr, 1,
1083 EXT4_FREE_BLOCKS_METADATA|
1084 EXT4_FREE_BLOCKS_FORGET);
1088 * The block which we have just freed is
1089 * pointed to by an indirect block: journal it
1091 BUFFER_TRACE(parent_bh, "get_write_access");
1092 if (!ext4_journal_get_write_access(handle,
1093 inode->i_sb, parent_bh,
1096 BUFFER_TRACE(parent_bh,
1097 "call ext4_handle_dirty_metadata");
1098 ext4_handle_dirty_metadata(handle,
1105 /* We have reached the bottom of the tree. */
1106 BUFFER_TRACE(parent_bh, "free data blocks");
1107 ext4_free_data(handle, inode, parent_bh, first, last);
1111 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1113 struct ext4_inode_info *ei = EXT4_I(inode);
1114 __le32 *i_data = ei->i_data;
1115 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1116 ext4_lblk_t offsets[4];
1121 ext4_lblk_t last_block, max_block;
1122 unsigned blocksize = inode->i_sb->s_blocksize;
1124 last_block = (inode->i_size + blocksize-1)
1125 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1126 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1127 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1129 if (last_block != max_block) {
1130 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1135 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1138 * The orphan list entry will now protect us from any crash which
1139 * occurs before the truncate completes, so it is now safe to propagate
1140 * the new, shorter inode size (held for now in i_size) into the
1141 * on-disk inode. We do this via i_disksize, which is the value which
1142 * ext4 *really* writes onto the disk inode.
1144 ei->i_disksize = inode->i_size;
1146 if (last_block == max_block) {
1148 * It is unnecessary to free any data blocks if last_block is
1149 * equal to the indirect block limit.
1152 } else if (n == 1) { /* direct blocks */
1153 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1154 i_data + EXT4_NDIR_BLOCKS);
1158 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1159 /* Kill the top of shared branch (not detached) */
1161 if (partial == chain) {
1162 /* Shared branch grows from the inode */
1163 ext4_free_branches(handle, inode, NULL,
1164 &nr, &nr+1, (chain+n-1) - partial);
1167 * We mark the inode dirty prior to restart,
1168 * and prior to stop. No need for it here.
1171 /* Shared branch grows from an indirect block */
1172 BUFFER_TRACE(partial->bh, "get_write_access");
1173 ext4_free_branches(handle, inode, partial->bh,
1175 partial->p+1, (chain+n-1) - partial);
1178 /* Clear the ends of indirect blocks on the shared branch */
1179 while (partial > chain) {
1180 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1181 (__le32*)partial->bh->b_data+addr_per_block,
1182 (chain+n-1) - partial);
1183 BUFFER_TRACE(partial->bh, "call brelse");
1184 brelse(partial->bh);
1188 /* Kill the remaining (whole) subtrees */
1189 switch (offsets[0]) {
1191 nr = i_data[EXT4_IND_BLOCK];
1193 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1194 i_data[EXT4_IND_BLOCK] = 0;
1197 case EXT4_IND_BLOCK:
1198 nr = i_data[EXT4_DIND_BLOCK];
1200 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1201 i_data[EXT4_DIND_BLOCK] = 0;
1204 case EXT4_DIND_BLOCK:
1205 nr = i_data[EXT4_TIND_BLOCK];
1207 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1208 i_data[EXT4_TIND_BLOCK] = 0;
1211 case EXT4_TIND_BLOCK:
1217 * ext4_ind_remove_space - remove space from the range
1218 * @handle: JBD handle for this transaction
1219 * @inode: inode we are dealing with
1220 * @start: First block to remove
1221 * @end: One block after the last block to remove (exclusive)
1223 * Free the blocks in the defined range (end is exclusive endpoint of
1224 * range). This is used by ext4_punch_hole().
1226 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1227 ext4_lblk_t start, ext4_lblk_t end)
1229 struct ext4_inode_info *ei = EXT4_I(inode);
1230 __le32 *i_data = ei->i_data;
1231 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1232 ext4_lblk_t offsets[4], offsets2[4];
1233 Indirect chain[4], chain2[4];
1234 Indirect *partial, *partial2;
1235 Indirect *p = NULL, *p2 = NULL;
1236 ext4_lblk_t max_block;
1237 __le32 nr = 0, nr2 = 0;
1239 unsigned blocksize = inode->i_sb->s_blocksize;
1241 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1242 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1243 if (end >= max_block)
1245 if ((start >= end) || (start > max_block))
1248 n = ext4_block_to_path(inode, start, offsets, NULL);
1249 n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1253 if ((n == 1) && (n == n2)) {
1254 /* We're punching only within direct block range */
1255 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1256 i_data + offsets2[0]);
1258 } else if (n2 > n) {
1260 * Start and end are on a different levels so we're going to
1261 * free partial block at start, and partial block at end of
1262 * the range. If there are some levels in between then
1263 * do_indirects label will take care of that.
1268 * Start is at the direct block level, free
1269 * everything to the end of the level.
1271 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1272 i_data + EXT4_NDIR_BLOCKS);
1277 partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1279 if (partial == chain) {
1280 /* Shared branch grows from the inode */
1281 ext4_free_branches(handle, inode, NULL,
1282 &nr, &nr+1, (chain+n-1) - partial);
1285 /* Shared branch grows from an indirect block */
1286 BUFFER_TRACE(partial->bh, "get_write_access");
1287 ext4_free_branches(handle, inode, partial->bh,
1289 partial->p+1, (chain+n-1) - partial);
1294 * Clear the ends of indirect blocks on the shared branch
1295 * at the start of the range
1297 while (partial > chain) {
1298 ext4_free_branches(handle, inode, partial->bh,
1300 (__le32 *)partial->bh->b_data+addr_per_block,
1301 (chain+n-1) - partial);
1306 partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1308 if (partial2 == chain2) {
1310 * Remember, end is exclusive so here we're at
1311 * the start of the next level we're not going
1312 * to free. Everything was covered by the start
1319 * ext4_find_shared returns Indirect structure which
1320 * points to the last element which should not be
1321 * removed by truncate. But this is end of the range
1322 * in punch_hole so we need to point to the next element
1328 * Clear the ends of indirect blocks on the shared branch
1329 * at the end of the range
1331 while (partial2 > chain2) {
1332 ext4_free_branches(handle, inode, partial2->bh,
1333 (__le32 *)partial2->bh->b_data,
1335 (chain2+n2-1) - partial2);
1341 /* Punch happened within the same level (n == n2) */
1342 partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1343 partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1345 /* Free top, but only if partial2 isn't its subtree. */
1347 int level = min(partial - chain, partial2 - chain2);
1351 for (i = 0; i <= level; i++) {
1352 if (offsets[i] != offsets2[i]) {
1359 if (partial == chain) {
1360 /* Shared branch grows from the inode */
1361 ext4_free_branches(handle, inode, NULL,
1363 (chain+n-1) - partial);
1366 /* Shared branch grows from an indirect block */
1367 BUFFER_TRACE(partial->bh, "get_write_access");
1368 ext4_free_branches(handle, inode, partial->bh,
1371 (chain+n-1) - partial);
1378 * ext4_find_shared returns Indirect structure which
1379 * points to the last element which should not be
1380 * removed by truncate. But this is end of the range
1381 * in punch_hole so we need to point to the next element
1386 while (partial > chain || partial2 > chain2) {
1387 int depth = (chain+n-1) - partial;
1388 int depth2 = (chain2+n2-1) - partial2;
1390 if (partial > chain && partial2 > chain2 &&
1391 partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1393 * We've converged on the same block. Clear the range,
1396 ext4_free_branches(handle, inode, partial->bh,
1399 (chain+n-1) - partial);
1404 * The start and end partial branches may not be at the same
1405 * level even though the punch happened within one level. So, we
1406 * give them a chance to arrive at the same level, then walk
1407 * them in step with each other until we converge on the same
1410 if (partial > chain && depth <= depth2) {
1411 ext4_free_branches(handle, inode, partial->bh,
1413 (__le32 *)partial->bh->b_data+addr_per_block,
1414 (chain+n-1) - partial);
1417 if (partial2 > chain2 && depth2 <= depth) {
1418 ext4_free_branches(handle, inode, partial2->bh,
1419 (__le32 *)partial2->bh->b_data,
1421 (chain2+n2-1) - partial2);
1427 while (p && p > chain) {
1428 BUFFER_TRACE(p->bh, "call brelse");
1432 while (p2 && p2 > chain2) {
1433 BUFFER_TRACE(p2->bh, "call brelse");
1440 /* Kill the remaining (whole) subtrees */
1441 switch (offsets[0]) {
1445 nr = i_data[EXT4_IND_BLOCK];
1447 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1448 i_data[EXT4_IND_BLOCK] = 0;
1451 case EXT4_IND_BLOCK:
1454 nr = i_data[EXT4_DIND_BLOCK];
1456 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1457 i_data[EXT4_DIND_BLOCK] = 0;
1460 case EXT4_DIND_BLOCK:
1463 nr = i_data[EXT4_TIND_BLOCK];
1465 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1466 i_data[EXT4_TIND_BLOCK] = 0;
1469 case EXT4_TIND_BLOCK: