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
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
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);
565 * Count number blocks in a subtree under 'partial'. At each
566 * level we count number of complete empty subtrees beyond
567 * current offset and then descend into the subtree only
568 * partially beyond current offset.
571 for (i = partial - chain + 1; i < depth; i++)
572 count = count * epb + (epb - offsets[i] - 1);
574 /* Fill in size of a hole we found */
576 map->m_len = min_t(unsigned int, map->m_len, count);
580 /* Failed read of indirect block */
585 * Okay, we need to do block allocation.
587 if (ext4_has_feature_bigalloc(inode->i_sb)) {
588 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
589 "non-extent mapped inodes with bigalloc");
590 return -EFSCORRUPTED;
593 /* Set up for the direct block allocation */
594 memset(&ar, 0, sizeof(ar));
596 ar.logical = map->m_lblk;
597 if (S_ISREG(inode->i_mode))
598 ar.flags = EXT4_MB_HINT_DATA;
599 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
600 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
601 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
602 ar.flags |= EXT4_MB_USE_RESERVED;
604 ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
606 /* the number of blocks need to allocate for [d,t]indirect blocks */
607 indirect_blks = (chain + depth) - partial - 1;
610 * Next look up the indirect map to count the totoal number of
611 * direct blocks to allocate for this branch.
613 ar.len = ext4_blks_to_allocate(partial, indirect_blks,
614 map->m_len, blocks_to_boundary);
617 * Block out ext4_truncate while we alter the tree
619 err = ext4_alloc_branch(handle, &ar, indirect_blks,
620 offsets + (partial - chain), partial);
623 * The ext4_splice_branch call will free and forget any buffers
624 * on the new chain if there is a failure, but that risks using
625 * up transaction credits, especially for bitmaps where the
626 * credits cannot be returned. Can we handle this somehow? We
627 * may need to return -EAGAIN upwards in the worst case. --sct
630 err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
634 map->m_flags |= EXT4_MAP_NEW;
636 ext4_update_inode_fsync_trans(handle, inode, 1);
639 map->m_flags |= EXT4_MAP_MAPPED;
640 map->m_pblk = le32_to_cpu(chain[depth-1].key);
642 if (count > blocks_to_boundary)
643 map->m_flags |= EXT4_MAP_BOUNDARY;
645 /* Clean up and exit */
646 partial = chain + depth - 1; /* the whole chain */
648 while (partial > chain) {
649 BUFFER_TRACE(partial->bh, "call brelse");
654 trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
659 * Calculate the number of metadata blocks need to reserve
660 * to allocate a new block at @lblocks for non extent file based file
662 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
664 struct ext4_inode_info *ei = EXT4_I(inode);
665 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
668 if (lblock < EXT4_NDIR_BLOCKS)
671 lblock -= EXT4_NDIR_BLOCKS;
673 if (ei->i_da_metadata_calc_len &&
674 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
675 ei->i_da_metadata_calc_len++;
678 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
679 ei->i_da_metadata_calc_len = 1;
680 blk_bits = order_base_2(lblock);
681 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
685 * Calculate number of indirect blocks touched by mapping @nrblocks logically
688 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
691 * With N contiguous data blocks, we need at most
692 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
693 * 2 dindirect blocks, and 1 tindirect block
695 return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
699 * Truncate transactions can be complex and absolutely huge. So we need to
700 * be able to restart the transaction at a conventient checkpoint to make
701 * sure we don't overflow the journal.
703 * Try to extend this transaction for the purposes of truncation. If
704 * extend fails, we need to propagate the failure up and restart the
705 * transaction in the top-level truncate loop. --sct
707 * Returns 0 if we managed to create more room. If we can't create more
708 * room, and the transaction must be restarted we return 1.
710 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
712 if (!ext4_handle_valid(handle))
714 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
716 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
722 * Probably it should be a library function... search for first non-zero word
723 * or memcmp with zero_page, whatever is better for particular architecture.
726 static inline int all_zeroes(__le32 *p, __le32 *q)
735 * ext4_find_shared - find the indirect blocks for partial truncation.
736 * @inode: inode in question
737 * @depth: depth of the affected branch
738 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
739 * @chain: place to store the pointers to partial indirect blocks
740 * @top: place to the (detached) top of branch
742 * This is a helper function used by ext4_truncate().
744 * When we do truncate() we may have to clean the ends of several
745 * indirect blocks but leave the blocks themselves alive. Block is
746 * partially truncated if some data below the new i_size is referred
747 * from it (and it is on the path to the first completely truncated
748 * data block, indeed). We have to free the top of that path along
749 * with everything to the right of the path. Since no allocation
750 * past the truncation point is possible until ext4_truncate()
751 * finishes, we may safely do the latter, but top of branch may
752 * require special attention - pageout below the truncation point
753 * might try to populate it.
755 * We atomically detach the top of branch from the tree, store the
756 * block number of its root in *@top, pointers to buffer_heads of
757 * partially truncated blocks - in @chain[].bh and pointers to
758 * their last elements that should not be removed - in
759 * @chain[].p. Return value is the pointer to last filled element
762 * The work left to caller to do the actual freeing of subtrees:
763 * a) free the subtree starting from *@top
764 * b) free the subtrees whose roots are stored in
765 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
766 * c) free the subtrees growing from the inode past the @chain[0].
767 * (no partially truncated stuff there). */
769 static Indirect *ext4_find_shared(struct inode *inode, int depth,
770 ext4_lblk_t offsets[4], Indirect chain[4],
773 Indirect *partial, *p;
777 /* Make k index the deepest non-null offset + 1 */
778 for (k = depth; k > 1 && !offsets[k-1]; k--)
780 partial = ext4_get_branch(inode, k, offsets, chain, &err);
781 /* Writer: pointers */
783 partial = chain + k-1;
785 * If the branch acquired continuation since we've looked at it -
786 * fine, it should all survive and (new) top doesn't belong to us.
788 if (!partial->key && *partial->p)
791 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
794 * OK, we've found the last block that must survive. The rest of our
795 * branch should be detached before unlocking. However, if that rest
796 * of branch is all ours and does not grow immediately from the inode
797 * it's easier to cheat and just decrement partial->p.
799 if (p == chain + k - 1 && p > chain) {
803 /* Nope, don't do this in ext4. Must leave the tree intact */
810 while (partial > p) {
819 * Zero a number of block pointers in either an inode or an indirect block.
820 * If we restart the transaction we must again get write access to the
821 * indirect block for further modification.
823 * We release `count' blocks on disk, but (last - first) may be greater
824 * than `count' because there can be holes in there.
826 * Return 0 on success, 1 on invalid block range
827 * and < 0 on fatal error.
829 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
830 struct buffer_head *bh,
831 ext4_fsblk_t block_to_free,
832 unsigned long count, __le32 *first,
836 int flags = EXT4_FREE_BLOCKS_VALIDATED;
839 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
840 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
841 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
842 else if (ext4_should_journal_data(inode))
843 flags |= EXT4_FREE_BLOCKS_FORGET;
845 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
847 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
848 "blocks %llu len %lu",
849 (unsigned long long) block_to_free, count);
853 if (try_to_extend_transaction(handle, inode)) {
855 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
856 err = ext4_handle_dirty_metadata(handle, inode, bh);
860 err = ext4_mark_inode_dirty(handle, inode);
863 err = ext4_truncate_restart_trans(handle, inode,
864 ext4_blocks_for_truncate(inode));
868 BUFFER_TRACE(bh, "retaking write access");
869 err = ext4_journal_get_write_access(handle, bh);
875 for (p = first; p < last; p++)
878 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
881 ext4_std_error(inode->i_sb, err);
886 * ext4_free_data - free a list of data blocks
887 * @handle: handle for this transaction
888 * @inode: inode we are dealing with
889 * @this_bh: indirect buffer_head which contains *@first and *@last
890 * @first: array of block numbers
891 * @last: points immediately past the end of array
893 * We are freeing all blocks referred from that array (numbers are stored as
894 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
896 * We accumulate contiguous runs of blocks to free. Conveniently, if these
897 * blocks are contiguous then releasing them at one time will only affect one
898 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
899 * actually use a lot of journal space.
901 * @this_bh will be %NULL if @first and @last point into the inode's direct
904 static void ext4_free_data(handle_t *handle, struct inode *inode,
905 struct buffer_head *this_bh,
906 __le32 *first, __le32 *last)
908 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
909 unsigned long count = 0; /* Number of blocks in the run */
910 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
913 ext4_fsblk_t nr; /* Current block # */
914 __le32 *p; /* Pointer into inode/ind
918 if (this_bh) { /* For indirect block */
919 BUFFER_TRACE(this_bh, "get_write_access");
920 err = ext4_journal_get_write_access(handle, this_bh);
921 /* Important: if we can't update the indirect pointers
922 * to the blocks, we can't free them. */
927 for (p = first; p < last; p++) {
928 nr = le32_to_cpu(*p);
930 /* accumulate blocks to free if they're contiguous */
935 } else if (nr == block_to_free + count) {
938 err = ext4_clear_blocks(handle, inode, this_bh,
939 block_to_free, count,
950 if (!err && count > 0)
951 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
952 count, block_to_free_p, p);
958 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
961 * The buffer head should have an attached journal head at this
962 * point. However, if the data is corrupted and an indirect
963 * block pointed to itself, it would have been detached when
964 * the block was cleared. Check for this instead of OOPSing.
966 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
967 ext4_handle_dirty_metadata(handle, inode, this_bh);
969 EXT4_ERROR_INODE(inode,
970 "circular indirect block detected at "
972 (unsigned long long) this_bh->b_blocknr);
977 * ext4_free_branches - free an array of branches
978 * @handle: JBD handle for this transaction
979 * @inode: inode we are dealing with
980 * @parent_bh: the buffer_head which contains *@first and *@last
981 * @first: array of block numbers
982 * @last: pointer immediately past the end of array
983 * @depth: depth of the branches to free
985 * We are freeing all blocks referred from these branches (numbers are
986 * stored as little-endian 32-bit) and updating @inode->i_blocks
989 static void ext4_free_branches(handle_t *handle, struct inode *inode,
990 struct buffer_head *parent_bh,
991 __le32 *first, __le32 *last, int depth)
996 if (ext4_handle_is_aborted(handle))
1000 struct buffer_head *bh;
1001 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1003 while (--p >= first) {
1004 nr = le32_to_cpu(*p);
1006 continue; /* A hole */
1008 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1010 EXT4_ERROR_INODE(inode,
1011 "invalid indirect mapped "
1012 "block %lu (level %d)",
1013 (unsigned long) nr, depth);
1017 /* Go read the buffer for the next level down */
1018 bh = sb_bread(inode->i_sb, nr);
1021 * A read failure? Report error and clear slot
1025 EXT4_ERROR_INODE_BLOCK(inode, nr,
1030 /* This zaps the entire block. Bottom up. */
1031 BUFFER_TRACE(bh, "free child branches");
1032 ext4_free_branches(handle, inode, bh,
1033 (__le32 *) bh->b_data,
1034 (__le32 *) bh->b_data + addr_per_block,
1039 * Everything below this this pointer has been
1040 * released. Now let this top-of-subtree go.
1042 * We want the freeing of this indirect block to be
1043 * atomic in the journal with the updating of the
1044 * bitmap block which owns it. So make some room in
1047 * We zero the parent pointer *after* freeing its
1048 * pointee in the bitmaps, so if extend_transaction()
1049 * for some reason fails to put the bitmap changes and
1050 * the release into the same transaction, recovery
1051 * will merely complain about releasing a free block,
1052 * rather than leaking blocks.
1054 if (ext4_handle_is_aborted(handle))
1056 if (try_to_extend_transaction(handle, inode)) {
1057 ext4_mark_inode_dirty(handle, inode);
1058 ext4_truncate_restart_trans(handle, inode,
1059 ext4_blocks_for_truncate(inode));
1063 * The forget flag here is critical because if
1064 * we are journaling (and not doing data
1065 * journaling), we have to make sure a revoke
1066 * record is written to prevent the journal
1067 * replay from overwriting the (former)
1068 * indirect block if it gets reallocated as a
1069 * data block. This must happen in the same
1070 * transaction where the data blocks are
1073 ext4_free_blocks(handle, inode, NULL, nr, 1,
1074 EXT4_FREE_BLOCKS_METADATA|
1075 EXT4_FREE_BLOCKS_FORGET);
1079 * The block which we have just freed is
1080 * pointed to by an indirect block: journal it
1082 BUFFER_TRACE(parent_bh, "get_write_access");
1083 if (!ext4_journal_get_write_access(handle,
1086 BUFFER_TRACE(parent_bh,
1087 "call ext4_handle_dirty_metadata");
1088 ext4_handle_dirty_metadata(handle,
1095 /* We have reached the bottom of the tree. */
1096 BUFFER_TRACE(parent_bh, "free data blocks");
1097 ext4_free_data(handle, inode, parent_bh, first, last);
1101 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1103 struct ext4_inode_info *ei = EXT4_I(inode);
1104 __le32 *i_data = ei->i_data;
1105 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1106 ext4_lblk_t offsets[4];
1111 ext4_lblk_t last_block, max_block;
1112 unsigned blocksize = inode->i_sb->s_blocksize;
1114 last_block = (inode->i_size + blocksize-1)
1115 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1116 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1117 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1119 if (last_block != max_block) {
1120 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1125 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1128 * The orphan list entry will now protect us from any crash which
1129 * occurs before the truncate completes, so it is now safe to propagate
1130 * the new, shorter inode size (held for now in i_size) into the
1131 * on-disk inode. We do this via i_disksize, which is the value which
1132 * ext4 *really* writes onto the disk inode.
1134 ei->i_disksize = inode->i_size;
1136 if (last_block == max_block) {
1138 * It is unnecessary to free any data blocks if last_block is
1139 * equal to the indirect block limit.
1142 } else if (n == 1) { /* direct blocks */
1143 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1144 i_data + EXT4_NDIR_BLOCKS);
1148 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1149 /* Kill the top of shared branch (not detached) */
1151 if (partial == chain) {
1152 /* Shared branch grows from the inode */
1153 ext4_free_branches(handle, inode, NULL,
1154 &nr, &nr+1, (chain+n-1) - partial);
1157 * We mark the inode dirty prior to restart,
1158 * and prior to stop. No need for it here.
1161 /* Shared branch grows from an indirect block */
1162 BUFFER_TRACE(partial->bh, "get_write_access");
1163 ext4_free_branches(handle, inode, partial->bh,
1165 partial->p+1, (chain+n-1) - partial);
1168 /* Clear the ends of indirect blocks on the shared branch */
1169 while (partial > chain) {
1170 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1171 (__le32*)partial->bh->b_data+addr_per_block,
1172 (chain+n-1) - partial);
1173 BUFFER_TRACE(partial->bh, "call brelse");
1174 brelse(partial->bh);
1178 /* Kill the remaining (whole) subtrees */
1179 switch (offsets[0]) {
1181 nr = i_data[EXT4_IND_BLOCK];
1183 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1184 i_data[EXT4_IND_BLOCK] = 0;
1187 case EXT4_IND_BLOCK:
1188 nr = i_data[EXT4_DIND_BLOCK];
1190 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1191 i_data[EXT4_DIND_BLOCK] = 0;
1194 case EXT4_DIND_BLOCK:
1195 nr = i_data[EXT4_TIND_BLOCK];
1197 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1198 i_data[EXT4_TIND_BLOCK] = 0;
1201 case EXT4_TIND_BLOCK:
1207 * ext4_ind_remove_space - remove space from the range
1208 * @handle: JBD handle for this transaction
1209 * @inode: inode we are dealing with
1210 * @start: First block to remove
1211 * @end: One block after the last block to remove (exclusive)
1213 * Free the blocks in the defined range (end is exclusive endpoint of
1214 * range). This is used by ext4_punch_hole().
1216 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1217 ext4_lblk_t start, ext4_lblk_t end)
1219 struct ext4_inode_info *ei = EXT4_I(inode);
1220 __le32 *i_data = ei->i_data;
1221 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1222 ext4_lblk_t offsets[4], offsets2[4];
1223 Indirect chain[4], chain2[4];
1224 Indirect *partial, *partial2;
1225 Indirect *p = NULL, *p2 = NULL;
1226 ext4_lblk_t max_block;
1227 __le32 nr = 0, nr2 = 0;
1229 unsigned blocksize = inode->i_sb->s_blocksize;
1231 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1232 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1233 if (end >= max_block)
1235 if ((start >= end) || (start > max_block))
1238 n = ext4_block_to_path(inode, start, offsets, NULL);
1239 n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1243 if ((n == 1) && (n == n2)) {
1244 /* We're punching only within direct block range */
1245 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1246 i_data + offsets2[0]);
1248 } else if (n2 > n) {
1250 * Start and end are on a different levels so we're going to
1251 * free partial block at start, and partial block at end of
1252 * the range. If there are some levels in between then
1253 * do_indirects label will take care of that.
1258 * Start is at the direct block level, free
1259 * everything to the end of the level.
1261 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1262 i_data + EXT4_NDIR_BLOCKS);
1267 partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1269 if (partial == chain) {
1270 /* Shared branch grows from the inode */
1271 ext4_free_branches(handle, inode, NULL,
1272 &nr, &nr+1, (chain+n-1) - partial);
1275 /* Shared branch grows from an indirect block */
1276 BUFFER_TRACE(partial->bh, "get_write_access");
1277 ext4_free_branches(handle, inode, partial->bh,
1279 partial->p+1, (chain+n-1) - partial);
1284 * Clear the ends of indirect blocks on the shared branch
1285 * at the start of the range
1287 while (partial > chain) {
1288 ext4_free_branches(handle, inode, partial->bh,
1290 (__le32 *)partial->bh->b_data+addr_per_block,
1291 (chain+n-1) - partial);
1296 partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1298 if (partial2 == chain2) {
1300 * Remember, end is exclusive so here we're at
1301 * the start of the next level we're not going
1302 * to free. Everything was covered by the start
1309 * ext4_find_shared returns Indirect structure which
1310 * points to the last element which should not be
1311 * removed by truncate. But this is end of the range
1312 * in punch_hole so we need to point to the next element
1318 * Clear the ends of indirect blocks on the shared branch
1319 * at the end of the range
1321 while (partial2 > chain2) {
1322 ext4_free_branches(handle, inode, partial2->bh,
1323 (__le32 *)partial2->bh->b_data,
1325 (chain2+n2-1) - partial2);
1331 /* Punch happened within the same level (n == n2) */
1332 partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1333 partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1335 /* Free top, but only if partial2 isn't its subtree. */
1337 int level = min(partial - chain, partial2 - chain2);
1341 for (i = 0; i <= level; i++) {
1342 if (offsets[i] != offsets2[i]) {
1349 if (partial == chain) {
1350 /* Shared branch grows from the inode */
1351 ext4_free_branches(handle, inode, NULL,
1353 (chain+n-1) - partial);
1356 /* Shared branch grows from an indirect block */
1357 BUFFER_TRACE(partial->bh, "get_write_access");
1358 ext4_free_branches(handle, inode, partial->bh,
1361 (chain+n-1) - partial);
1368 * ext4_find_shared returns Indirect structure which
1369 * points to the last element which should not be
1370 * removed by truncate. But this is end of the range
1371 * in punch_hole so we need to point to the next element
1376 while (partial > chain || partial2 > chain2) {
1377 int depth = (chain+n-1) - partial;
1378 int depth2 = (chain2+n2-1) - partial2;
1380 if (partial > chain && partial2 > chain2 &&
1381 partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1383 * We've converged on the same block. Clear the range,
1386 ext4_free_branches(handle, inode, partial->bh,
1389 (chain+n-1) - partial);
1394 * The start and end partial branches may not be at the same
1395 * level even though the punch happened within one level. So, we
1396 * give them a chance to arrive at the same level, then walk
1397 * them in step with each other until we converge on the same
1400 if (partial > chain && depth <= depth2) {
1401 ext4_free_branches(handle, inode, partial->bh,
1403 (__le32 *)partial->bh->b_data+addr_per_block,
1404 (chain+n-1) - partial);
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);
1417 while (p && p > chain) {
1418 BUFFER_TRACE(p->bh, "call brelse");
1422 while (p2 && p2 > chain2) {
1423 BUFFER_TRACE(p2->bh, "call brelse");
1430 /* Kill the remaining (whole) subtrees */
1431 switch (offsets[0]) {
1435 nr = i_data[EXT4_IND_BLOCK];
1437 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1438 i_data[EXT4_IND_BLOCK] = 0;
1441 case EXT4_IND_BLOCK:
1444 nr = i_data[EXT4_DIND_BLOCK];
1446 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1447 i_data[EXT4_DIND_BLOCK] = 0;
1450 case EXT4_DIND_BLOCK:
1453 nr = i_data[EXT4_TIND_BLOCK];
1455 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1456 i_data[EXT4_TIND_BLOCK] = 0;
1459 case EXT4_TIND_BLOCK: