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
299 * @ar: structure describing the allocation request
300 * @indirect_blks: number of allocated indirect blocks
301 * @offsets: offsets (in the blocks) to store the pointers to next.
302 * @branch: place to store the chain in.
304 * This function allocates blocks, zeroes out all but the last one,
305 * links them into chain and (if we are synchronous) writes them to disk.
306 * In other words, it prepares a branch that can be spliced onto the
307 * inode. It stores the information about that chain in the branch[], in
308 * the same format as ext4_get_branch() would do. We are calling it after
309 * we had read the existing part of chain and partial points to the last
310 * triple of that (one with zero ->key). Upon the exit we have the same
311 * picture as after the successful ext4_get_block(), except that in one
312 * place chain is disconnected - *branch->p is still zero (we did not
313 * set the last link), but branch->key contains the number that should
314 * be placed into *branch->p to fill that gap.
316 * If allocation fails we free all blocks we've allocated (and forget
317 * their buffer_heads) and return the error value the from failed
318 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
319 * as described above and return 0.
321 static int ext4_alloc_branch(handle_t *handle,
322 struct ext4_allocation_request *ar,
323 int indirect_blks, ext4_lblk_t *offsets,
326 struct buffer_head * bh;
327 ext4_fsblk_t b, new_blocks[4];
329 int i, j, err, len = 1;
331 for (i = 0; i <= indirect_blks; i++) {
332 if (i == indirect_blks) {
333 new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
335 ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
337 ar->flags & EXT4_MB_DELALLOC_RESERVED,
339 /* Simplify error cleanup... */
340 branch[i+1].bh = NULL;
346 branch[i].key = cpu_to_le32(new_blocks[i]);
350 bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
356 BUFFER_TRACE(bh, "call get_create_access");
357 err = ext4_journal_get_create_access(handle, bh);
363 memset(bh->b_data, 0, bh->b_size);
364 p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
367 if (i == indirect_blks)
369 for (j = 0; j < len; j++)
370 *p++ = cpu_to_le32(b++);
372 BUFFER_TRACE(bh, "marking uptodate");
373 set_buffer_uptodate(bh);
376 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
377 err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
383 if (i == indirect_blks) {
384 /* Free data blocks */
385 ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
389 for (; i >= 0; i--) {
391 * We want to ext4_forget() only freshly allocated indirect
392 * blocks. Buffer for new_blocks[i] is at branch[i+1].bh
393 * (buffer at branch[0].bh is indirect block / inode already
394 * existing before ext4_alloc_branch() was called). Also
395 * because blocks are freshly allocated, we don't need to
396 * revoke them which is why we don't set
397 * EXT4_FREE_BLOCKS_METADATA.
399 ext4_free_blocks(handle, ar->inode, branch[i+1].bh,
401 branch[i+1].bh ? EXT4_FREE_BLOCKS_FORGET : 0);
407 * ext4_splice_branch() - splice the allocated branch onto inode.
408 * @handle: handle for this transaction
409 * @ar: structure describing the allocation request
410 * @where: location of missing link
411 * @num: number of indirect blocks we are adding
413 * This function fills the missing link and does all housekeeping needed in
414 * inode (->i_blocks, etc.). In case of success we end up with the full
415 * chain to new block and return 0.
417 static int ext4_splice_branch(handle_t *handle,
418 struct ext4_allocation_request *ar,
419 Indirect *where, int num)
423 ext4_fsblk_t current_block;
426 * If we're splicing into a [td]indirect block (as opposed to the
427 * inode) then we need to get write access to the [td]indirect block
431 BUFFER_TRACE(where->bh, "get_write_access");
432 err = ext4_journal_get_write_access(handle, where->bh);
438 *where->p = where->key;
441 * Update the host buffer_head or inode to point to more just allocated
442 * direct blocks blocks
444 if (num == 0 && ar->len > 1) {
445 current_block = le32_to_cpu(where->key) + 1;
446 for (i = 1; i < ar->len; i++)
447 *(where->p + i) = cpu_to_le32(current_block++);
450 /* We are done with atomic stuff, now do the rest of housekeeping */
451 /* had we spliced it onto indirect block? */
454 * If we spliced it onto an indirect block, we haven't
455 * altered the inode. Note however that if it is being spliced
456 * onto an indirect block at the very end of the file (the
457 * file is growing) then we *will* alter the inode to reflect
458 * the new i_size. But that is not done here - it is done in
459 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
461 jbd_debug(5, "splicing indirect only\n");
462 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
463 err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
468 * OK, we spliced it into the inode itself on a direct block.
470 ext4_mark_inode_dirty(handle, ar->inode);
471 jbd_debug(5, "splicing direct\n");
476 for (i = 1; i <= num; i++) {
478 * branch[i].bh is newly allocated, so there is no
479 * need to revoke the block, which is why we don't
480 * need to set EXT4_FREE_BLOCKS_METADATA.
482 ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
483 EXT4_FREE_BLOCKS_FORGET);
485 ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
492 * The ext4_ind_map_blocks() function handles non-extents inodes
493 * (i.e., using the traditional indirect/double-indirect i_blocks
494 * scheme) for ext4_map_blocks().
496 * Allocation strategy is simple: if we have to allocate something, we will
497 * have to go the whole way to leaf. So let's do it before attaching anything
498 * to tree, set linkage between the newborn blocks, write them if sync is
499 * required, recheck the path, free and repeat if check fails, otherwise
500 * set the last missing link (that will protect us from any truncate-generated
501 * removals - all blocks on the path are immune now) and possibly force the
502 * write on the parent block.
503 * That has a nice additional property: no special recovery from the failed
504 * allocations is needed - we simply release blocks and do not touch anything
505 * reachable from inode.
507 * `handle' can be NULL if create == 0.
509 * return > 0, # of blocks mapped or allocated.
510 * return = 0, if plain lookup failed.
511 * return < 0, error case.
513 * The ext4_ind_get_blocks() function should be called with
514 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
515 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
516 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
519 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
520 struct ext4_map_blocks *map,
523 struct ext4_allocation_request ar;
525 ext4_lblk_t offsets[4];
529 int blocks_to_boundary = 0;
532 ext4_fsblk_t first_block = 0;
534 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
535 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
536 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
537 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
538 &blocks_to_boundary);
543 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
545 /* Simplest case - block found, no allocation needed */
547 first_block = le32_to_cpu(chain[depth - 1].key);
550 while (count < map->m_len && count <= blocks_to_boundary) {
553 blk = le32_to_cpu(*(chain[depth-1].p + count));
555 if (blk == first_block + count)
563 /* Next simple case - plain lookup failed */
564 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
565 unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
569 * Count number blocks in a subtree under 'partial'. At each
570 * level we count number of complete empty subtrees beyond
571 * current offset and then descend into the subtree only
572 * partially beyond current offset.
575 for (i = partial - chain + 1; i < depth; i++)
576 count = count * epb + (epb - offsets[i] - 1);
578 /* Fill in size of a hole we found */
580 map->m_len = min_t(unsigned int, map->m_len, count);
584 /* Failed read of indirect block */
589 * Okay, we need to do block allocation.
591 if (ext4_has_feature_bigalloc(inode->i_sb)) {
592 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
593 "non-extent mapped inodes with bigalloc");
594 return -EFSCORRUPTED;
597 /* Set up for the direct block allocation */
598 memset(&ar, 0, sizeof(ar));
600 ar.logical = map->m_lblk;
601 if (S_ISREG(inode->i_mode))
602 ar.flags = EXT4_MB_HINT_DATA;
603 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
604 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
605 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
606 ar.flags |= EXT4_MB_USE_RESERVED;
608 ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
610 /* the number of blocks need to allocate for [d,t]indirect blocks */
611 indirect_blks = (chain + depth) - partial - 1;
614 * Next look up the indirect map to count the totoal number of
615 * direct blocks to allocate for this branch.
617 ar.len = ext4_blks_to_allocate(partial, indirect_blks,
618 map->m_len, blocks_to_boundary);
621 * Block out ext4_truncate while we alter the tree
623 err = ext4_alloc_branch(handle, &ar, indirect_blks,
624 offsets + (partial - chain), partial);
627 * The ext4_splice_branch call will free and forget any buffers
628 * on the new chain if there is a failure, but that risks using
629 * up transaction credits, especially for bitmaps where the
630 * credits cannot be returned. Can we handle this somehow? We
631 * may need to return -EAGAIN upwards in the worst case. --sct
634 err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
638 map->m_flags |= EXT4_MAP_NEW;
640 ext4_update_inode_fsync_trans(handle, inode, 1);
643 map->m_flags |= EXT4_MAP_MAPPED;
644 map->m_pblk = le32_to_cpu(chain[depth-1].key);
646 if (count > blocks_to_boundary)
647 map->m_flags |= EXT4_MAP_BOUNDARY;
649 /* Clean up and exit */
650 partial = chain + depth - 1; /* the whole chain */
652 while (partial > chain) {
653 BUFFER_TRACE(partial->bh, "call brelse");
658 trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
663 * Calculate the number of metadata blocks need to reserve
664 * to allocate a new block at @lblocks for non extent file based file
666 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
668 struct ext4_inode_info *ei = EXT4_I(inode);
669 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
672 if (lblock < EXT4_NDIR_BLOCKS)
675 lblock -= EXT4_NDIR_BLOCKS;
677 if (ei->i_da_metadata_calc_len &&
678 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
679 ei->i_da_metadata_calc_len++;
682 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
683 ei->i_da_metadata_calc_len = 1;
684 blk_bits = order_base_2(lblock);
685 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
689 * Calculate number of indirect blocks touched by mapping @nrblocks logically
692 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
695 * With N contiguous data blocks, we need at most
696 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
697 * 2 dindirect blocks, and 1 tindirect block
699 return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
702 static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
703 struct buffer_head *bh, int *dropped)
708 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
709 err = ext4_handle_dirty_metadata(handle, inode, bh);
713 err = ext4_mark_inode_dirty(handle, inode);
717 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
718 * moment, get_block can be called only for blocks inside i_size since
719 * page cache has been already dropped and writes are blocked by
720 * i_mutex. So we can safely drop the i_data_sem here.
722 BUG_ON(EXT4_JOURNAL(inode) == NULL);
723 ext4_discard_preallocations(inode);
724 up_write(&EXT4_I(inode)->i_data_sem);
730 * Truncate transactions can be complex and absolutely huge. So we need to
731 * be able to restart the transaction at a conventient checkpoint to make
732 * sure we don't overflow the journal.
734 * Try to extend this transaction for the purposes of truncation. If
735 * extend fails, we restart transaction.
737 static int ext4_ind_truncate_ensure_credits(handle_t *handle,
739 struct buffer_head *bh,
745 ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS,
746 ext4_blocks_for_truncate(inode), revoke_creds,
747 ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped));
749 down_write(&EXT4_I(inode)->i_data_sem);
753 BUFFER_TRACE(bh, "retaking write access");
754 ret = ext4_journal_get_write_access(handle, bh);
762 * Probably it should be a library function... search for first non-zero word
763 * or memcmp with zero_page, whatever is better for particular architecture.
766 static inline int all_zeroes(__le32 *p, __le32 *q)
775 * ext4_find_shared - find the indirect blocks for partial truncation.
776 * @inode: inode in question
777 * @depth: depth of the affected branch
778 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
779 * @chain: place to store the pointers to partial indirect blocks
780 * @top: place to the (detached) top of branch
782 * This is a helper function used by ext4_truncate().
784 * When we do truncate() we may have to clean the ends of several
785 * indirect blocks but leave the blocks themselves alive. Block is
786 * partially truncated if some data below the new i_size is referred
787 * from it (and it is on the path to the first completely truncated
788 * data block, indeed). We have to free the top of that path along
789 * with everything to the right of the path. Since no allocation
790 * past the truncation point is possible until ext4_truncate()
791 * finishes, we may safely do the latter, but top of branch may
792 * require special attention - pageout below the truncation point
793 * might try to populate it.
795 * We atomically detach the top of branch from the tree, store the
796 * block number of its root in *@top, pointers to buffer_heads of
797 * partially truncated blocks - in @chain[].bh and pointers to
798 * their last elements that should not be removed - in
799 * @chain[].p. Return value is the pointer to last filled element
802 * The work left to caller to do the actual freeing of subtrees:
803 * a) free the subtree starting from *@top
804 * b) free the subtrees whose roots are stored in
805 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
806 * c) free the subtrees growing from the inode past the @chain[0].
807 * (no partially truncated stuff there). */
809 static Indirect *ext4_find_shared(struct inode *inode, int depth,
810 ext4_lblk_t offsets[4], Indirect chain[4],
813 Indirect *partial, *p;
817 /* Make k index the deepest non-null offset + 1 */
818 for (k = depth; k > 1 && !offsets[k-1]; k--)
820 partial = ext4_get_branch(inode, k, offsets, chain, &err);
821 /* Writer: pointers */
823 partial = chain + k-1;
825 * If the branch acquired continuation since we've looked at it -
826 * fine, it should all survive and (new) top doesn't belong to us.
828 if (!partial->key && *partial->p)
831 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
834 * OK, we've found the last block that must survive. The rest of our
835 * branch should be detached before unlocking. However, if that rest
836 * of branch is all ours and does not grow immediately from the inode
837 * it's easier to cheat and just decrement partial->p.
839 if (p == chain + k - 1 && p > chain) {
843 /* Nope, don't do this in ext4. Must leave the tree intact */
850 while (partial > p) {
859 * Zero a number of block pointers in either an inode or an indirect block.
860 * If we restart the transaction we must again get write access to the
861 * indirect block for further modification.
863 * We release `count' blocks on disk, but (last - first) may be greater
864 * than `count' because there can be holes in there.
866 * Return 0 on success, 1 on invalid block range
867 * and < 0 on fatal error.
869 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
870 struct buffer_head *bh,
871 ext4_fsblk_t block_to_free,
872 unsigned long count, __le32 *first,
876 int flags = EXT4_FREE_BLOCKS_VALIDATED;
879 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
880 ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
881 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
882 else if (ext4_should_journal_data(inode))
883 flags |= EXT4_FREE_BLOCKS_FORGET;
885 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
887 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
888 "blocks %llu len %lu",
889 (unsigned long long) block_to_free, count);
893 err = ext4_ind_truncate_ensure_credits(handle, inode, bh,
894 ext4_free_data_revoke_credits(inode, count));
898 for (p = first; p < last; p++)
901 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
904 ext4_std_error(inode->i_sb, err);
909 * ext4_free_data - free a list of data blocks
910 * @handle: handle for this transaction
911 * @inode: inode we are dealing with
912 * @this_bh: indirect buffer_head which contains *@first and *@last
913 * @first: array of block numbers
914 * @last: points immediately past the end of array
916 * We are freeing all blocks referred from that array (numbers are stored as
917 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
919 * We accumulate contiguous runs of blocks to free. Conveniently, if these
920 * blocks are contiguous then releasing them at one time will only affect one
921 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
922 * actually use a lot of journal space.
924 * @this_bh will be %NULL if @first and @last point into the inode's direct
927 static void ext4_free_data(handle_t *handle, struct inode *inode,
928 struct buffer_head *this_bh,
929 __le32 *first, __le32 *last)
931 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
932 unsigned long count = 0; /* Number of blocks in the run */
933 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
936 ext4_fsblk_t nr; /* Current block # */
937 __le32 *p; /* Pointer into inode/ind
941 if (this_bh) { /* For indirect block */
942 BUFFER_TRACE(this_bh, "get_write_access");
943 err = ext4_journal_get_write_access(handle, this_bh);
944 /* Important: if we can't update the indirect pointers
945 * to the blocks, we can't free them. */
950 for (p = first; p < last; p++) {
951 nr = le32_to_cpu(*p);
953 /* accumulate blocks to free if they're contiguous */
958 } else if (nr == block_to_free + count) {
961 err = ext4_clear_blocks(handle, inode, this_bh,
962 block_to_free, count,
973 if (!err && count > 0)
974 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
975 count, block_to_free_p, p);
981 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
984 * The buffer head should have an attached journal head at this
985 * point. However, if the data is corrupted and an indirect
986 * block pointed to itself, it would have been detached when
987 * the block was cleared. Check for this instead of OOPSing.
989 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
990 ext4_handle_dirty_metadata(handle, inode, this_bh);
992 EXT4_ERROR_INODE(inode,
993 "circular indirect block detected at "
995 (unsigned long long) this_bh->b_blocknr);
1000 * ext4_free_branches - free an array of branches
1001 * @handle: JBD handle for this transaction
1002 * @inode: inode we are dealing with
1003 * @parent_bh: the buffer_head which contains *@first and *@last
1004 * @first: array of block numbers
1005 * @last: pointer immediately past the end of array
1006 * @depth: depth of the branches to free
1008 * We are freeing all blocks referred from these branches (numbers are
1009 * stored as little-endian 32-bit) and updating @inode->i_blocks
1012 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1013 struct buffer_head *parent_bh,
1014 __le32 *first, __le32 *last, int depth)
1019 if (ext4_handle_is_aborted(handle))
1023 struct buffer_head *bh;
1024 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1026 while (--p >= first) {
1027 nr = le32_to_cpu(*p);
1029 continue; /* A hole */
1031 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1033 EXT4_ERROR_INODE(inode,
1034 "invalid indirect mapped "
1035 "block %lu (level %d)",
1036 (unsigned long) nr, depth);
1040 /* Go read the buffer for the next level down */
1041 bh = sb_bread(inode->i_sb, nr);
1044 * A read failure? Report error and clear slot
1048 EXT4_ERROR_INODE_BLOCK(inode, nr,
1053 /* This zaps the entire block. Bottom up. */
1054 BUFFER_TRACE(bh, "free child branches");
1055 ext4_free_branches(handle, inode, bh,
1056 (__le32 *) bh->b_data,
1057 (__le32 *) bh->b_data + addr_per_block,
1062 * Everything below this this pointer has been
1063 * released. Now let this top-of-subtree go.
1065 * We want the freeing of this indirect block to be
1066 * atomic in the journal with the updating of the
1067 * bitmap block which owns it. So make some room in
1070 * We zero the parent pointer *after* freeing its
1071 * pointee in the bitmaps, so if extend_transaction()
1072 * for some reason fails to put the bitmap changes and
1073 * the release into the same transaction, recovery
1074 * will merely complain about releasing a free block,
1075 * rather than leaking blocks.
1077 if (ext4_handle_is_aborted(handle))
1079 if (ext4_ind_truncate_ensure_credits(handle, inode,
1081 ext4_free_metadata_revoke_credits(
1082 inode->i_sb, 1)) < 0)
1086 * The forget flag here is critical because if
1087 * we are journaling (and not doing data
1088 * journaling), we have to make sure a revoke
1089 * record is written to prevent the journal
1090 * replay from overwriting the (former)
1091 * indirect block if it gets reallocated as a
1092 * data block. This must happen in the same
1093 * transaction where the data blocks are
1096 ext4_free_blocks(handle, inode, NULL, nr, 1,
1097 EXT4_FREE_BLOCKS_METADATA|
1098 EXT4_FREE_BLOCKS_FORGET);
1102 * The block which we have just freed is
1103 * pointed to by an indirect block: journal it
1105 BUFFER_TRACE(parent_bh, "get_write_access");
1106 if (!ext4_journal_get_write_access(handle,
1109 BUFFER_TRACE(parent_bh,
1110 "call ext4_handle_dirty_metadata");
1111 ext4_handle_dirty_metadata(handle,
1118 /* We have reached the bottom of the tree. */
1119 BUFFER_TRACE(parent_bh, "free data blocks");
1120 ext4_free_data(handle, inode, parent_bh, first, last);
1124 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1126 struct ext4_inode_info *ei = EXT4_I(inode);
1127 __le32 *i_data = ei->i_data;
1128 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1129 ext4_lblk_t offsets[4];
1134 ext4_lblk_t last_block, max_block;
1135 unsigned blocksize = inode->i_sb->s_blocksize;
1137 last_block = (inode->i_size + blocksize-1)
1138 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1139 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1140 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1142 if (last_block != max_block) {
1143 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1148 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1151 * The orphan list entry will now protect us from any crash which
1152 * occurs before the truncate completes, so it is now safe to propagate
1153 * the new, shorter inode size (held for now in i_size) into the
1154 * on-disk inode. We do this via i_disksize, which is the value which
1155 * ext4 *really* writes onto the disk inode.
1157 ei->i_disksize = inode->i_size;
1159 if (last_block == max_block) {
1161 * It is unnecessary to free any data blocks if last_block is
1162 * equal to the indirect block limit.
1165 } else if (n == 1) { /* direct blocks */
1166 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1167 i_data + EXT4_NDIR_BLOCKS);
1171 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1172 /* Kill the top of shared branch (not detached) */
1174 if (partial == chain) {
1175 /* Shared branch grows from the inode */
1176 ext4_free_branches(handle, inode, NULL,
1177 &nr, &nr+1, (chain+n-1) - partial);
1180 * We mark the inode dirty prior to restart,
1181 * and prior to stop. No need for it here.
1184 /* Shared branch grows from an indirect block */
1185 BUFFER_TRACE(partial->bh, "get_write_access");
1186 ext4_free_branches(handle, inode, partial->bh,
1188 partial->p+1, (chain+n-1) - partial);
1191 /* Clear the ends of indirect blocks on the shared branch */
1192 while (partial > chain) {
1193 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1194 (__le32*)partial->bh->b_data+addr_per_block,
1195 (chain+n-1) - partial);
1196 BUFFER_TRACE(partial->bh, "call brelse");
1197 brelse(partial->bh);
1201 /* Kill the remaining (whole) subtrees */
1202 switch (offsets[0]) {
1204 nr = i_data[EXT4_IND_BLOCK];
1206 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1207 i_data[EXT4_IND_BLOCK] = 0;
1210 case EXT4_IND_BLOCK:
1211 nr = i_data[EXT4_DIND_BLOCK];
1213 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1214 i_data[EXT4_DIND_BLOCK] = 0;
1217 case EXT4_DIND_BLOCK:
1218 nr = i_data[EXT4_TIND_BLOCK];
1220 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1221 i_data[EXT4_TIND_BLOCK] = 0;
1224 case EXT4_TIND_BLOCK:
1230 * ext4_ind_remove_space - remove space from the range
1231 * @handle: JBD handle for this transaction
1232 * @inode: inode we are dealing with
1233 * @start: First block to remove
1234 * @end: One block after the last block to remove (exclusive)
1236 * Free the blocks in the defined range (end is exclusive endpoint of
1237 * range). This is used by ext4_punch_hole().
1239 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1240 ext4_lblk_t start, ext4_lblk_t end)
1242 struct ext4_inode_info *ei = EXT4_I(inode);
1243 __le32 *i_data = ei->i_data;
1244 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1245 ext4_lblk_t offsets[4], offsets2[4];
1246 Indirect chain[4], chain2[4];
1247 Indirect *partial, *partial2;
1248 Indirect *p = NULL, *p2 = NULL;
1249 ext4_lblk_t max_block;
1250 __le32 nr = 0, nr2 = 0;
1252 unsigned blocksize = inode->i_sb->s_blocksize;
1254 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1255 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1256 if (end >= max_block)
1258 if ((start >= end) || (start > max_block))
1261 n = ext4_block_to_path(inode, start, offsets, NULL);
1262 n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1266 if ((n == 1) && (n == n2)) {
1267 /* We're punching only within direct block range */
1268 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1269 i_data + offsets2[0]);
1271 } else if (n2 > n) {
1273 * Start and end are on a different levels so we're going to
1274 * free partial block at start, and partial block at end of
1275 * the range. If there are some levels in between then
1276 * do_indirects label will take care of that.
1281 * Start is at the direct block level, free
1282 * everything to the end of the level.
1284 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1285 i_data + EXT4_NDIR_BLOCKS);
1290 partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1292 if (partial == chain) {
1293 /* Shared branch grows from the inode */
1294 ext4_free_branches(handle, inode, NULL,
1295 &nr, &nr+1, (chain+n-1) - partial);
1298 /* Shared branch grows from an indirect block */
1299 BUFFER_TRACE(partial->bh, "get_write_access");
1300 ext4_free_branches(handle, inode, partial->bh,
1302 partial->p+1, (chain+n-1) - partial);
1307 * Clear the ends of indirect blocks on the shared branch
1308 * at the start of the range
1310 while (partial > chain) {
1311 ext4_free_branches(handle, inode, partial->bh,
1313 (__le32 *)partial->bh->b_data+addr_per_block,
1314 (chain+n-1) - partial);
1319 partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1321 if (partial2 == chain2) {
1323 * Remember, end is exclusive so here we're at
1324 * the start of the next level we're not going
1325 * to free. Everything was covered by the start
1332 * ext4_find_shared returns Indirect structure which
1333 * points to the last element which should not be
1334 * removed by truncate. But this is end of the range
1335 * in punch_hole so we need to point to the next element
1341 * Clear the ends of indirect blocks on the shared branch
1342 * at the end of the range
1344 while (partial2 > chain2) {
1345 ext4_free_branches(handle, inode, partial2->bh,
1346 (__le32 *)partial2->bh->b_data,
1348 (chain2+n2-1) - partial2);
1354 /* Punch happened within the same level (n == n2) */
1355 partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1356 partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1358 /* Free top, but only if partial2 isn't its subtree. */
1360 int level = min(partial - chain, partial2 - chain2);
1364 for (i = 0; i <= level; i++) {
1365 if (offsets[i] != offsets2[i]) {
1372 if (partial == chain) {
1373 /* Shared branch grows from the inode */
1374 ext4_free_branches(handle, inode, NULL,
1376 (chain+n-1) - partial);
1379 /* Shared branch grows from an indirect block */
1380 BUFFER_TRACE(partial->bh, "get_write_access");
1381 ext4_free_branches(handle, inode, partial->bh,
1384 (chain+n-1) - partial);
1391 * ext4_find_shared returns Indirect structure which
1392 * points to the last element which should not be
1393 * removed by truncate. But this is end of the range
1394 * in punch_hole so we need to point to the next element
1399 while (partial > chain || partial2 > chain2) {
1400 int depth = (chain+n-1) - partial;
1401 int depth2 = (chain2+n2-1) - partial2;
1403 if (partial > chain && partial2 > chain2 &&
1404 partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1406 * We've converged on the same block. Clear the range,
1409 ext4_free_branches(handle, inode, partial->bh,
1412 (chain+n-1) - partial);
1417 * The start and end partial branches may not be at the same
1418 * level even though the punch happened within one level. So, we
1419 * give them a chance to arrive at the same level, then walk
1420 * them in step with each other until we converge on the same
1423 if (partial > chain && depth <= depth2) {
1424 ext4_free_branches(handle, inode, partial->bh,
1426 (__le32 *)partial->bh->b_data+addr_per_block,
1427 (chain+n-1) - partial);
1430 if (partial2 > chain2 && depth2 <= depth) {
1431 ext4_free_branches(handle, inode, partial2->bh,
1432 (__le32 *)partial2->bh->b_data,
1434 (chain2+n2-1) - partial2);
1440 while (p && p > chain) {
1441 BUFFER_TRACE(p->bh, "call brelse");
1445 while (p2 && p2 > chain2) {
1446 BUFFER_TRACE(p2->bh, "call brelse");
1453 /* Kill the remaining (whole) subtrees */
1454 switch (offsets[0]) {
1458 nr = i_data[EXT4_IND_BLOCK];
1460 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1461 i_data[EXT4_IND_BLOCK] = 0;
1464 case EXT4_IND_BLOCK:
1467 nr = i_data[EXT4_DIND_BLOCK];
1469 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1470 i_data[EXT4_DIND_BLOCK] = 0;
1473 case EXT4_DIND_BLOCK:
1476 nr = i_data[EXT4_TIND_BLOCK];
1478 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1479 i_data[EXT4_TIND_BLOCK] = 0;
1482 case EXT4_TIND_BLOCK: