2 * linux/fs/ext4/indirect.c
6 * linux/fs/ext4/inode.c
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
15 * linux/fs/minix/inode.c
17 * Copyright (C) 1991, 1992 Linus Torvalds
19 * Goal-directed block allocation by Stephen Tweedie
20 * (sct@redhat.com), 1993, 1998
23 #include "ext4_jbd2.h"
25 #include <linux/uio.h>
27 #include <trace/events/ext4.h>
32 struct buffer_head *bh;
35 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
42 * ext4_block_to_path - parse the block number into array of offsets
43 * @inode: inode in question (we are only interested in its superblock)
44 * @i_block: block number to be parsed
45 * @offsets: array to store the offsets in
46 * @boundary: set this non-zero if the referred-to block is likely to be
47 * followed (on disk) by an indirect block.
49 * To store the locations of file's data ext4 uses a data structure common
50 * for UNIX filesystems - tree of pointers anchored in the inode, with
51 * data blocks at leaves and indirect blocks in intermediate nodes.
52 * This function translates the block number into path in that tree -
53 * return value is the path length and @offsets[n] is the offset of
54 * pointer to (n+1)th node in the nth one. If @block is out of range
55 * (negative or too large) warning is printed and zero returned.
57 * Note: function doesn't find node addresses, so no IO is needed. All
58 * we need to know is the capacity of indirect blocks (taken from the
63 * Portability note: the last comparison (check that we fit into triple
64 * indirect block) is spelled differently, because otherwise on an
65 * architecture with 32-bit longs and 8Kb pages we might get into trouble
66 * if our filesystem had 8Kb blocks. We might use long long, but that would
67 * kill us on x86. Oh, well, at least the sign propagation does not matter -
68 * i_block would have to be negative in the very beginning, so we would not
72 static int ext4_block_to_path(struct inode *inode,
74 ext4_lblk_t offsets[4], int *boundary)
76 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
77 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
78 const long direct_blocks = EXT4_NDIR_BLOCKS,
79 indirect_blocks = ptrs,
80 double_blocks = (1 << (ptrs_bits * 2));
84 if (i_block < direct_blocks) {
85 offsets[n++] = i_block;
86 final = direct_blocks;
87 } else if ((i_block -= direct_blocks) < indirect_blocks) {
88 offsets[n++] = EXT4_IND_BLOCK;
89 offsets[n++] = i_block;
91 } else if ((i_block -= indirect_blocks) < double_blocks) {
92 offsets[n++] = EXT4_DIND_BLOCK;
93 offsets[n++] = i_block >> ptrs_bits;
94 offsets[n++] = i_block & (ptrs - 1);
96 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
97 offsets[n++] = EXT4_TIND_BLOCK;
98 offsets[n++] = i_block >> (ptrs_bits * 2);
99 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
100 offsets[n++] = i_block & (ptrs - 1);
103 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
104 i_block + direct_blocks +
105 indirect_blocks + double_blocks, inode->i_ino);
108 *boundary = final - 1 - (i_block & (ptrs - 1));
113 * ext4_get_branch - read the chain of indirect blocks leading to data
114 * @inode: inode in question
115 * @depth: depth of the chain (1 - direct pointer, etc.)
116 * @offsets: offsets of pointers in inode/indirect blocks
117 * @chain: place to store the result
118 * @err: here we store the error value
120 * Function fills the array of triples <key, p, bh> and returns %NULL
121 * if everything went OK or the pointer to the last filled triple
122 * (incomplete one) otherwise. Upon the return chain[i].key contains
123 * the number of (i+1)-th block in the chain (as it is stored in memory,
124 * i.e. little-endian 32-bit), chain[i].p contains the address of that
125 * number (it points into struct inode for i==0 and into the bh->b_data
126 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
127 * block for i>0 and NULL for i==0. In other words, it holds the block
128 * numbers of the chain, addresses they were taken from (and where we can
129 * verify that chain did not change) and buffer_heads hosting these
132 * Function stops when it stumbles upon zero pointer (absent block)
133 * (pointer to last triple returned, *@err == 0)
134 * or when it gets an IO error reading an indirect block
135 * (ditto, *@err == -EIO)
136 * or when it reads all @depth-1 indirect blocks successfully and finds
137 * the whole chain, all way to the data (returns %NULL, *err == 0).
139 * Need to be called with
140 * down_read(&EXT4_I(inode)->i_data_sem)
142 static Indirect *ext4_get_branch(struct inode *inode, int depth,
143 ext4_lblk_t *offsets,
144 Indirect chain[4], int *err)
146 struct super_block *sb = inode->i_sb;
148 struct buffer_head *bh;
152 /* i_data is not going away, no lock needed */
153 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
157 bh = sb_getblk(sb, le32_to_cpu(p->key));
163 if (!bh_uptodate_or_lock(bh)) {
164 if (bh_submit_read(bh) < 0) {
168 /* validate block references */
169 if (ext4_check_indirect_blockref(inode, bh)) {
175 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
189 * ext4_find_near - find a place for allocation with sufficient locality
191 * @ind: descriptor of indirect block.
193 * This function returns the preferred place for block allocation.
194 * It is used when heuristic for sequential allocation fails.
196 * + if there is a block to the left of our position - allocate near it.
197 * + if pointer will live in indirect block - allocate near that block.
198 * + if pointer will live in inode - allocate in the same
201 * In the latter case we colour the starting block by the callers PID to
202 * prevent it from clashing with concurrent allocations for a different inode
203 * in the same block group. The PID is used here so that functionally related
204 * files will be close-by on-disk.
206 * Caller must make sure that @ind is valid and will stay that way.
208 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
210 struct ext4_inode_info *ei = EXT4_I(inode);
211 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
214 /* Try to find previous block */
215 for (p = ind->p - 1; p >= start; p--) {
217 return le32_to_cpu(*p);
220 /* No such thing, so let's try location of indirect block */
222 return ind->bh->b_blocknr;
225 * It is going to be referred to from the inode itself? OK, just put it
226 * into the same cylinder group then.
228 return ext4_inode_to_goal_block(inode);
232 * ext4_find_goal - find a preferred place for allocation.
234 * @block: block we want
235 * @partial: pointer to the last triple within a chain
237 * Normally this function find the preferred place for block allocation,
239 * Because this is only used for non-extent files, we limit the block nr
242 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
248 * XXX need to get goal block from mballoc's data structures
251 goal = ext4_find_near(inode, partial);
252 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
257 * ext4_blks_to_allocate - Look up the block map and count the number
258 * of direct blocks need to be allocated for the given branch.
260 * @branch: chain of indirect blocks
261 * @k: number of blocks need for indirect blocks
262 * @blks: number of data blocks to be mapped.
263 * @blocks_to_boundary: the offset in the indirect block
265 * return the total number of blocks to be allocate, including the
266 * direct and indirect blocks.
268 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
269 int blocks_to_boundary)
271 unsigned int count = 0;
274 * Simple case, [t,d]Indirect block(s) has not allocated yet
275 * then it's clear blocks on that path have not allocated
278 /* right now we don't handle cross boundary allocation */
279 if (blks < blocks_to_boundary + 1)
282 count += blocks_to_boundary + 1;
287 while (count < blks && count <= blocks_to_boundary &&
288 le32_to_cpu(*(branch[0].p + count)) == 0) {
295 * ext4_alloc_branch - allocate and set up a chain of blocks.
296 * @handle: handle for this transaction
298 * @indirect_blks: number of allocated indirect blocks
299 * @blks: number of allocated direct blocks
300 * @goal: preferred place for allocation
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,
343 branch[i].key = cpu_to_le32(new_blocks[i]);
347 bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
353 BUFFER_TRACE(bh, "call get_create_access");
354 err = ext4_journal_get_create_access(handle, bh);
360 memset(bh->b_data, 0, bh->b_size);
361 p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
364 if (i == indirect_blks)
366 for (j = 0; j < len; j++)
367 *p++ = cpu_to_le32(b++);
369 BUFFER_TRACE(bh, "marking uptodate");
370 set_buffer_uptodate(bh);
373 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
374 err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
380 for (; i >= 0; i--) {
382 * We want to ext4_forget() only freshly allocated indirect
383 * blocks. Buffer for new_blocks[i-1] is at branch[i].bh and
384 * buffer at branch[0].bh is indirect block / inode already
385 * existing before ext4_alloc_branch() was called.
387 if (i > 0 && i != indirect_blks && branch[i].bh)
388 ext4_forget(handle, 1, ar->inode, branch[i].bh,
389 branch[i].bh->b_blocknr);
390 ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
391 (i == indirect_blks) ? ar->len : 1, 0);
397 * ext4_splice_branch - splice the allocated branch onto inode.
398 * @handle: handle for this transaction
400 * @block: (logical) number of block we are adding
401 * @chain: chain of indirect blocks (with a missing link - see
403 * @where: location of missing link
404 * @num: number of indirect blocks we are adding
405 * @blks: number of direct blocks we are adding
407 * This function fills the missing link and does all housekeeping needed in
408 * inode (->i_blocks, etc.). In case of success we end up with the full
409 * chain to new block and return 0.
411 static int ext4_splice_branch(handle_t *handle,
412 struct ext4_allocation_request *ar,
413 Indirect *where, int num)
417 ext4_fsblk_t current_block;
420 * If we're splicing into a [td]indirect block (as opposed to the
421 * inode) then we need to get write access to the [td]indirect block
425 BUFFER_TRACE(where->bh, "get_write_access");
426 err = ext4_journal_get_write_access(handle, where->bh);
432 *where->p = where->key;
435 * Update the host buffer_head or inode to point to more just allocated
436 * direct blocks blocks
438 if (num == 0 && ar->len > 1) {
439 current_block = le32_to_cpu(where->key) + 1;
440 for (i = 1; i < ar->len; i++)
441 *(where->p + i) = cpu_to_le32(current_block++);
444 /* We are done with atomic stuff, now do the rest of housekeeping */
445 /* had we spliced it onto indirect block? */
448 * If we spliced it onto an indirect block, we haven't
449 * altered the inode. Note however that if it is being spliced
450 * onto an indirect block at the very end of the file (the
451 * file is growing) then we *will* alter the inode to reflect
452 * the new i_size. But that is not done here - it is done in
453 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
455 jbd_debug(5, "splicing indirect only\n");
456 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
457 err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
462 * OK, we spliced it into the inode itself on a direct block.
464 ext4_mark_inode_dirty(handle, ar->inode);
465 jbd_debug(5, "splicing direct\n");
470 for (i = 1; i <= num; i++) {
472 * branch[i].bh is newly allocated, so there is no
473 * need to revoke the block, which is why we don't
474 * need to set EXT4_FREE_BLOCKS_METADATA.
476 ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
477 EXT4_FREE_BLOCKS_FORGET);
479 ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
486 * The ext4_ind_map_blocks() function handles non-extents inodes
487 * (i.e., using the traditional indirect/double-indirect i_blocks
488 * scheme) for ext4_map_blocks().
490 * Allocation strategy is simple: if we have to allocate something, we will
491 * have to go the whole way to leaf. So let's do it before attaching anything
492 * to tree, set linkage between the newborn blocks, write them if sync is
493 * required, recheck the path, free and repeat if check fails, otherwise
494 * set the last missing link (that will protect us from any truncate-generated
495 * removals - all blocks on the path are immune now) and possibly force the
496 * write on the parent block.
497 * That has a nice additional property: no special recovery from the failed
498 * allocations is needed - we simply release blocks and do not touch anything
499 * reachable from inode.
501 * `handle' can be NULL if create == 0.
503 * return > 0, # of blocks mapped or allocated.
504 * return = 0, if plain lookup failed.
505 * return < 0, error case.
507 * The ext4_ind_get_blocks() function should be called with
508 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
509 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
510 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
513 int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
514 struct ext4_map_blocks *map,
517 struct ext4_allocation_request ar;
519 ext4_lblk_t offsets[4];
523 int blocks_to_boundary = 0;
526 ext4_fsblk_t first_block = 0;
528 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
529 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
530 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
531 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
532 &blocks_to_boundary);
537 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
539 /* Simplest case - block found, no allocation needed */
541 first_block = le32_to_cpu(chain[depth - 1].key);
544 while (count < map->m_len && count <= blocks_to_boundary) {
547 blk = le32_to_cpu(*(chain[depth-1].p + count));
549 if (blk == first_block + count)
557 /* Next simple case - plain lookup or failed read of indirect block */
558 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
562 * Okay, we need to do block allocation.
564 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
565 EXT4_FEATURE_RO_COMPAT_BIGALLOC)) {
566 EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
567 "non-extent mapped inodes with bigalloc");
571 /* Set up for the direct block allocation */
572 memset(&ar, 0, sizeof(ar));
574 ar.logical = map->m_lblk;
575 if (S_ISREG(inode->i_mode))
576 ar.flags = EXT4_MB_HINT_DATA;
577 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
578 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
579 if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
580 ar.flags |= EXT4_MB_USE_RESERVED;
582 ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
584 /* the number of blocks need to allocate for [d,t]indirect blocks */
585 indirect_blks = (chain + depth) - partial - 1;
588 * Next look up the indirect map to count the totoal number of
589 * direct blocks to allocate for this branch.
591 ar.len = ext4_blks_to_allocate(partial, indirect_blks,
592 map->m_len, blocks_to_boundary);
595 * Block out ext4_truncate while we alter the tree
597 err = ext4_alloc_branch(handle, &ar, indirect_blks,
598 offsets + (partial - chain), partial);
601 * The ext4_splice_branch call will free and forget any buffers
602 * on the new chain if there is a failure, but that risks using
603 * up transaction credits, especially for bitmaps where the
604 * credits cannot be returned. Can we handle this somehow? We
605 * may need to return -EAGAIN upwards in the worst case. --sct
608 err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
612 map->m_flags |= EXT4_MAP_NEW;
614 ext4_update_inode_fsync_trans(handle, inode, 1);
617 map->m_flags |= EXT4_MAP_MAPPED;
618 map->m_pblk = le32_to_cpu(chain[depth-1].key);
620 if (count > blocks_to_boundary)
621 map->m_flags |= EXT4_MAP_BOUNDARY;
623 /* Clean up and exit */
624 partial = chain + depth - 1; /* the whole chain */
626 while (partial > chain) {
627 BUFFER_TRACE(partial->bh, "call brelse");
632 trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
637 * O_DIRECT for ext3 (or indirect map) based files
639 * If the O_DIRECT write will extend the file then add this inode to the
640 * orphan list. So recovery will truncate it back to the original size
641 * if the machine crashes during the write.
643 * If the O_DIRECT write is intantiating holes inside i_size and the machine
644 * crashes then stale disk data _may_ be exposed inside the file. But current
645 * VFS code falls back into buffered path in that case so we are safe.
647 ssize_t ext4_ind_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
650 struct file *file = iocb->ki_filp;
651 struct inode *inode = file->f_mapping->host;
652 struct ext4_inode_info *ei = EXT4_I(inode);
656 size_t count = iov_iter_count(iter);
659 if (iov_iter_rw(iter) == WRITE) {
660 loff_t final_size = offset + count;
662 if (final_size > inode->i_size) {
663 /* Credits for sb + inode write */
664 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
665 if (IS_ERR(handle)) {
666 ret = PTR_ERR(handle);
669 ret = ext4_orphan_add(handle, inode);
671 ext4_journal_stop(handle);
675 ei->i_disksize = inode->i_size;
676 ext4_journal_stop(handle);
681 if (iov_iter_rw(iter) == READ && ext4_should_dioread_nolock(inode)) {
683 * Nolock dioread optimization may be dynamically disabled
684 * via ext4_inode_block_unlocked_dio(). Check inode's state
685 * while holding extra i_dio_count ref.
687 inode_dio_begin(inode);
689 if (unlikely(ext4_test_inode_state(inode,
690 EXT4_STATE_DIOREAD_LOCK))) {
691 inode_dio_end(inode);
695 ret = dax_do_io(iocb, inode, iter, offset,
696 ext4_get_block, NULL, 0);
698 ret = __blockdev_direct_IO(iocb, inode,
699 inode->i_sb->s_bdev, iter,
700 offset, ext4_get_block, NULL,
702 inode_dio_end(inode);
706 ret = dax_do_io(iocb, inode, iter, offset,
707 ext4_get_block, NULL, DIO_LOCKING);
709 ret = blockdev_direct_IO(iocb, inode, iter, offset,
712 if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
713 loff_t isize = i_size_read(inode);
714 loff_t end = offset + count;
717 ext4_truncate_failed_write(inode);
720 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
726 /* Credits for sb + inode write */
727 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
728 if (IS_ERR(handle)) {
729 /* This is really bad luck. We've written the data
730 * but cannot extend i_size. Bail out and pretend
731 * the write failed... */
732 ret = PTR_ERR(handle);
734 ext4_orphan_del(NULL, inode);
739 ext4_orphan_del(handle, inode);
741 loff_t end = offset + ret;
742 if (end > inode->i_size) {
743 ei->i_disksize = end;
744 i_size_write(inode, end);
746 * We're going to return a positive `ret'
747 * here due to non-zero-length I/O, so there's
748 * no way of reporting error returns from
749 * ext4_mark_inode_dirty() to userspace. So
752 ext4_mark_inode_dirty(handle, inode);
755 err = ext4_journal_stop(handle);
764 * Calculate the number of metadata blocks need to reserve
765 * to allocate a new block at @lblocks for non extent file based file
767 int ext4_ind_calc_metadata_amount(struct inode *inode, sector_t lblock)
769 struct ext4_inode_info *ei = EXT4_I(inode);
770 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
773 if (lblock < EXT4_NDIR_BLOCKS)
776 lblock -= EXT4_NDIR_BLOCKS;
778 if (ei->i_da_metadata_calc_len &&
779 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
780 ei->i_da_metadata_calc_len++;
783 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
784 ei->i_da_metadata_calc_len = 1;
785 blk_bits = order_base_2(lblock);
786 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
790 * Calculate number of indirect blocks touched by mapping @nrblocks logically
793 int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
796 * With N contiguous data blocks, we need at most
797 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
798 * 2 dindirect blocks, and 1 tindirect block
800 return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
804 * Truncate transactions can be complex and absolutely huge. So we need to
805 * be able to restart the transaction at a conventient checkpoint to make
806 * sure we don't overflow the journal.
808 * Try to extend this transaction for the purposes of truncation. If
809 * extend fails, we need to propagate the failure up and restart the
810 * transaction in the top-level truncate loop. --sct
812 * Returns 0 if we managed to create more room. If we can't create more
813 * room, and the transaction must be restarted we return 1.
815 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
817 if (!ext4_handle_valid(handle))
819 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
821 if (!ext4_journal_extend(handle, ext4_blocks_for_truncate(inode)))
827 * Probably it should be a library function... search for first non-zero word
828 * or memcmp with zero_page, whatever is better for particular architecture.
831 static inline int all_zeroes(__le32 *p, __le32 *q)
840 * ext4_find_shared - find the indirect blocks for partial truncation.
841 * @inode: inode in question
842 * @depth: depth of the affected branch
843 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
844 * @chain: place to store the pointers to partial indirect blocks
845 * @top: place to the (detached) top of branch
847 * This is a helper function used by ext4_truncate().
849 * When we do truncate() we may have to clean the ends of several
850 * indirect blocks but leave the blocks themselves alive. Block is
851 * partially truncated if some data below the new i_size is referred
852 * from it (and it is on the path to the first completely truncated
853 * data block, indeed). We have to free the top of that path along
854 * with everything to the right of the path. Since no allocation
855 * past the truncation point is possible until ext4_truncate()
856 * finishes, we may safely do the latter, but top of branch may
857 * require special attention - pageout below the truncation point
858 * might try to populate it.
860 * We atomically detach the top of branch from the tree, store the
861 * block number of its root in *@top, pointers to buffer_heads of
862 * partially truncated blocks - in @chain[].bh and pointers to
863 * their last elements that should not be removed - in
864 * @chain[].p. Return value is the pointer to last filled element
867 * The work left to caller to do the actual freeing of subtrees:
868 * a) free the subtree starting from *@top
869 * b) free the subtrees whose roots are stored in
870 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
871 * c) free the subtrees growing from the inode past the @chain[0].
872 * (no partially truncated stuff there). */
874 static Indirect *ext4_find_shared(struct inode *inode, int depth,
875 ext4_lblk_t offsets[4], Indirect chain[4],
878 Indirect *partial, *p;
882 /* Make k index the deepest non-null offset + 1 */
883 for (k = depth; k > 1 && !offsets[k-1]; k--)
885 partial = ext4_get_branch(inode, k, offsets, chain, &err);
886 /* Writer: pointers */
888 partial = chain + k-1;
890 * If the branch acquired continuation since we've looked at it -
891 * fine, it should all survive and (new) top doesn't belong to us.
893 if (!partial->key && *partial->p)
896 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
899 * OK, we've found the last block that must survive. The rest of our
900 * branch should be detached before unlocking. However, if that rest
901 * of branch is all ours and does not grow immediately from the inode
902 * it's easier to cheat and just decrement partial->p.
904 if (p == chain + k - 1 && p > chain) {
908 /* Nope, don't do this in ext4. Must leave the tree intact */
915 while (partial > p) {
924 * Zero a number of block pointers in either an inode or an indirect block.
925 * If we restart the transaction we must again get write access to the
926 * indirect block for further modification.
928 * We release `count' blocks on disk, but (last - first) may be greater
929 * than `count' because there can be holes in there.
931 * Return 0 on success, 1 on invalid block range
932 * and < 0 on fatal error.
934 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
935 struct buffer_head *bh,
936 ext4_fsblk_t block_to_free,
937 unsigned long count, __le32 *first,
941 int flags = EXT4_FREE_BLOCKS_VALIDATED;
944 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
945 flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
946 else if (ext4_should_journal_data(inode))
947 flags |= EXT4_FREE_BLOCKS_FORGET;
949 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
951 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
952 "blocks %llu len %lu",
953 (unsigned long long) block_to_free, count);
957 if (try_to_extend_transaction(handle, inode)) {
959 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
960 err = ext4_handle_dirty_metadata(handle, inode, bh);
964 err = ext4_mark_inode_dirty(handle, inode);
967 err = ext4_truncate_restart_trans(handle, inode,
968 ext4_blocks_for_truncate(inode));
972 BUFFER_TRACE(bh, "retaking write access");
973 err = ext4_journal_get_write_access(handle, bh);
979 for (p = first; p < last; p++)
982 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
985 ext4_std_error(inode->i_sb, err);
990 * ext4_free_data - free a list of data blocks
991 * @handle: handle for this transaction
992 * @inode: inode we are dealing with
993 * @this_bh: indirect buffer_head which contains *@first and *@last
994 * @first: array of block numbers
995 * @last: points immediately past the end of array
997 * We are freeing all blocks referred from that array (numbers are stored as
998 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1000 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1001 * blocks are contiguous then releasing them at one time will only affect one
1002 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1003 * actually use a lot of journal space.
1005 * @this_bh will be %NULL if @first and @last point into the inode's direct
1008 static void ext4_free_data(handle_t *handle, struct inode *inode,
1009 struct buffer_head *this_bh,
1010 __le32 *first, __le32 *last)
1012 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
1013 unsigned long count = 0; /* Number of blocks in the run */
1014 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1017 ext4_fsblk_t nr; /* Current block # */
1018 __le32 *p; /* Pointer into inode/ind
1019 for current block */
1022 if (this_bh) { /* For indirect block */
1023 BUFFER_TRACE(this_bh, "get_write_access");
1024 err = ext4_journal_get_write_access(handle, this_bh);
1025 /* Important: if we can't update the indirect pointers
1026 * to the blocks, we can't free them. */
1031 for (p = first; p < last; p++) {
1032 nr = le32_to_cpu(*p);
1034 /* accumulate blocks to free if they're contiguous */
1037 block_to_free_p = p;
1039 } else if (nr == block_to_free + count) {
1042 err = ext4_clear_blocks(handle, inode, this_bh,
1043 block_to_free, count,
1044 block_to_free_p, p);
1048 block_to_free_p = p;
1054 if (!err && count > 0)
1055 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
1056 count, block_to_free_p, p);
1062 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
1065 * The buffer head should have an attached journal head at this
1066 * point. However, if the data is corrupted and an indirect
1067 * block pointed to itself, it would have been detached when
1068 * the block was cleared. Check for this instead of OOPSing.
1070 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
1071 ext4_handle_dirty_metadata(handle, inode, this_bh);
1073 EXT4_ERROR_INODE(inode,
1074 "circular indirect block detected at "
1076 (unsigned long long) this_bh->b_blocknr);
1081 * ext4_free_branches - free an array of branches
1082 * @handle: JBD handle for this transaction
1083 * @inode: inode we are dealing with
1084 * @parent_bh: the buffer_head which contains *@first and *@last
1085 * @first: array of block numbers
1086 * @last: pointer immediately past the end of array
1087 * @depth: depth of the branches to free
1089 * We are freeing all blocks referred from these branches (numbers are
1090 * stored as little-endian 32-bit) and updating @inode->i_blocks
1093 static void ext4_free_branches(handle_t *handle, struct inode *inode,
1094 struct buffer_head *parent_bh,
1095 __le32 *first, __le32 *last, int depth)
1100 if (ext4_handle_is_aborted(handle))
1104 struct buffer_head *bh;
1105 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1107 while (--p >= first) {
1108 nr = le32_to_cpu(*p);
1110 continue; /* A hole */
1112 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
1114 EXT4_ERROR_INODE(inode,
1115 "invalid indirect mapped "
1116 "block %lu (level %d)",
1117 (unsigned long) nr, depth);
1121 /* Go read the buffer for the next level down */
1122 bh = sb_bread(inode->i_sb, nr);
1125 * A read failure? Report error and clear slot
1129 EXT4_ERROR_INODE_BLOCK(inode, nr,
1134 /* This zaps the entire block. Bottom up. */
1135 BUFFER_TRACE(bh, "free child branches");
1136 ext4_free_branches(handle, inode, bh,
1137 (__le32 *) bh->b_data,
1138 (__le32 *) bh->b_data + addr_per_block,
1143 * Everything below this this pointer has been
1144 * released. Now let this top-of-subtree go.
1146 * We want the freeing of this indirect block to be
1147 * atomic in the journal with the updating of the
1148 * bitmap block which owns it. So make some room in
1151 * We zero the parent pointer *after* freeing its
1152 * pointee in the bitmaps, so if extend_transaction()
1153 * for some reason fails to put the bitmap changes and
1154 * the release into the same transaction, recovery
1155 * will merely complain about releasing a free block,
1156 * rather than leaking blocks.
1158 if (ext4_handle_is_aborted(handle))
1160 if (try_to_extend_transaction(handle, inode)) {
1161 ext4_mark_inode_dirty(handle, inode);
1162 ext4_truncate_restart_trans(handle, inode,
1163 ext4_blocks_for_truncate(inode));
1167 * The forget flag here is critical because if
1168 * we are journaling (and not doing data
1169 * journaling), we have to make sure a revoke
1170 * record is written to prevent the journal
1171 * replay from overwriting the (former)
1172 * indirect block if it gets reallocated as a
1173 * data block. This must happen in the same
1174 * transaction where the data blocks are
1177 ext4_free_blocks(handle, inode, NULL, nr, 1,
1178 EXT4_FREE_BLOCKS_METADATA|
1179 EXT4_FREE_BLOCKS_FORGET);
1183 * The block which we have just freed is
1184 * pointed to by an indirect block: journal it
1186 BUFFER_TRACE(parent_bh, "get_write_access");
1187 if (!ext4_journal_get_write_access(handle,
1190 BUFFER_TRACE(parent_bh,
1191 "call ext4_handle_dirty_metadata");
1192 ext4_handle_dirty_metadata(handle,
1199 /* We have reached the bottom of the tree. */
1200 BUFFER_TRACE(parent_bh, "free data blocks");
1201 ext4_free_data(handle, inode, parent_bh, first, last);
1205 void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1207 struct ext4_inode_info *ei = EXT4_I(inode);
1208 __le32 *i_data = ei->i_data;
1209 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1210 ext4_lblk_t offsets[4];
1215 ext4_lblk_t last_block, max_block;
1216 unsigned blocksize = inode->i_sb->s_blocksize;
1218 last_block = (inode->i_size + blocksize-1)
1219 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1220 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1221 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1223 if (last_block != max_block) {
1224 n = ext4_block_to_path(inode, last_block, offsets, NULL);
1229 ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1232 * The orphan list entry will now protect us from any crash which
1233 * occurs before the truncate completes, so it is now safe to propagate
1234 * the new, shorter inode size (held for now in i_size) into the
1235 * on-disk inode. We do this via i_disksize, which is the value which
1236 * ext4 *really* writes onto the disk inode.
1238 ei->i_disksize = inode->i_size;
1240 if (last_block == max_block) {
1242 * It is unnecessary to free any data blocks if last_block is
1243 * equal to the indirect block limit.
1246 } else if (n == 1) { /* direct blocks */
1247 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1248 i_data + EXT4_NDIR_BLOCKS);
1252 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1253 /* Kill the top of shared branch (not detached) */
1255 if (partial == chain) {
1256 /* Shared branch grows from the inode */
1257 ext4_free_branches(handle, inode, NULL,
1258 &nr, &nr+1, (chain+n-1) - partial);
1261 * We mark the inode dirty prior to restart,
1262 * and prior to stop. No need for it here.
1265 /* Shared branch grows from an indirect block */
1266 BUFFER_TRACE(partial->bh, "get_write_access");
1267 ext4_free_branches(handle, inode, partial->bh,
1269 partial->p+1, (chain+n-1) - partial);
1272 /* Clear the ends of indirect blocks on the shared branch */
1273 while (partial > chain) {
1274 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1275 (__le32*)partial->bh->b_data+addr_per_block,
1276 (chain+n-1) - partial);
1277 BUFFER_TRACE(partial->bh, "call brelse");
1278 brelse(partial->bh);
1282 /* Kill the remaining (whole) subtrees */
1283 switch (offsets[0]) {
1285 nr = i_data[EXT4_IND_BLOCK];
1287 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1288 i_data[EXT4_IND_BLOCK] = 0;
1290 case EXT4_IND_BLOCK:
1291 nr = i_data[EXT4_DIND_BLOCK];
1293 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1294 i_data[EXT4_DIND_BLOCK] = 0;
1296 case EXT4_DIND_BLOCK:
1297 nr = i_data[EXT4_TIND_BLOCK];
1299 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1300 i_data[EXT4_TIND_BLOCK] = 0;
1302 case EXT4_TIND_BLOCK:
1308 * ext4_ind_remove_space - remove space from the range
1309 * @handle: JBD handle for this transaction
1310 * @inode: inode we are dealing with
1311 * @start: First block to remove
1312 * @end: One block after the last block to remove (exclusive)
1314 * Free the blocks in the defined range (end is exclusive endpoint of
1315 * range). This is used by ext4_punch_hole().
1317 int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1318 ext4_lblk_t start, ext4_lblk_t end)
1320 struct ext4_inode_info *ei = EXT4_I(inode);
1321 __le32 *i_data = ei->i_data;
1322 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1323 ext4_lblk_t offsets[4], offsets2[4];
1324 Indirect chain[4], chain2[4];
1325 Indirect *partial, *partial2;
1326 ext4_lblk_t max_block;
1327 __le32 nr = 0, nr2 = 0;
1329 unsigned blocksize = inode->i_sb->s_blocksize;
1331 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1332 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1333 if (end >= max_block)
1335 if ((start >= end) || (start > max_block))
1338 n = ext4_block_to_path(inode, start, offsets, NULL);
1339 n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1343 if ((n == 1) && (n == n2)) {
1344 /* We're punching only within direct block range */
1345 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1346 i_data + offsets2[0]);
1348 } else if (n2 > n) {
1350 * Start and end are on a different levels so we're going to
1351 * free partial block at start, and partial block at end of
1352 * the range. If there are some levels in between then
1353 * do_indirects label will take care of that.
1358 * Start is at the direct block level, free
1359 * everything to the end of the level.
1361 ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1362 i_data + EXT4_NDIR_BLOCKS);
1367 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1369 if (partial == chain) {
1370 /* Shared branch grows from the inode */
1371 ext4_free_branches(handle, inode, NULL,
1372 &nr, &nr+1, (chain+n-1) - partial);
1375 /* Shared branch grows from an indirect block */
1376 BUFFER_TRACE(partial->bh, "get_write_access");
1377 ext4_free_branches(handle, inode, partial->bh,
1379 partial->p+1, (chain+n-1) - partial);
1384 * Clear the ends of indirect blocks on the shared branch
1385 * at the start of the range
1387 while (partial > chain) {
1388 ext4_free_branches(handle, inode, partial->bh,
1390 (__le32 *)partial->bh->b_data+addr_per_block,
1391 (chain+n-1) - partial);
1392 BUFFER_TRACE(partial->bh, "call brelse");
1393 brelse(partial->bh);
1398 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1400 if (partial2 == chain2) {
1402 * Remember, end is exclusive so here we're at
1403 * the start of the next level we're not going
1404 * to free. Everything was covered by the start
1411 * ext4_find_shared returns Indirect structure which
1412 * points to the last element which should not be
1413 * removed by truncate. But this is end of the range
1414 * in punch_hole so we need to point to the next element
1420 * Clear the ends of indirect blocks on the shared branch
1421 * at the end of the range
1423 while (partial2 > chain2) {
1424 ext4_free_branches(handle, inode, partial2->bh,
1425 (__le32 *)partial2->bh->b_data,
1427 (chain2+n2-1) - partial2);
1428 BUFFER_TRACE(partial2->bh, "call brelse");
1429 brelse(partial2->bh);
1435 /* Punch happened within the same level (n == n2) */
1436 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1437 partial2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1439 /* Free top, but only if partial2 isn't its subtree. */
1441 int level = min(partial - chain, partial2 - chain2);
1445 for (i = 0; i <= level; i++) {
1446 if (offsets[i] != offsets2[i]) {
1453 if (partial == chain) {
1454 /* Shared branch grows from the inode */
1455 ext4_free_branches(handle, inode, NULL,
1457 (chain+n-1) - partial);
1460 /* Shared branch grows from an indirect block */
1461 BUFFER_TRACE(partial->bh, "get_write_access");
1462 ext4_free_branches(handle, inode, partial->bh,
1465 (chain+n-1) - partial);
1472 * ext4_find_shared returns Indirect structure which
1473 * points to the last element which should not be
1474 * removed by truncate. But this is end of the range
1475 * in punch_hole so we need to point to the next element
1480 while (partial > chain || partial2 > chain2) {
1481 int depth = (chain+n-1) - partial;
1482 int depth2 = (chain2+n2-1) - partial2;
1484 if (partial > chain && partial2 > chain2 &&
1485 partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1487 * We've converged on the same block. Clear the range,
1490 ext4_free_branches(handle, inode, partial->bh,
1493 (chain+n-1) - partial);
1494 BUFFER_TRACE(partial->bh, "call brelse");
1495 brelse(partial->bh);
1496 BUFFER_TRACE(partial2->bh, "call brelse");
1497 brelse(partial2->bh);
1502 * The start and end partial branches may not be at the same
1503 * level even though the punch happened within one level. So, we
1504 * give them a chance to arrive at the same level, then walk
1505 * them in step with each other until we converge on the same
1508 if (partial > chain && depth <= depth2) {
1509 ext4_free_branches(handle, inode, partial->bh,
1511 (__le32 *)partial->bh->b_data+addr_per_block,
1512 (chain+n-1) - partial);
1513 BUFFER_TRACE(partial->bh, "call brelse");
1514 brelse(partial->bh);
1517 if (partial2 > chain2 && depth2 <= depth) {
1518 ext4_free_branches(handle, inode, partial2->bh,
1519 (__le32 *)partial2->bh->b_data,
1521 (chain2+n2-1) - partial2);
1522 BUFFER_TRACE(partial2->bh, "call brelse");
1523 brelse(partial2->bh);
1530 /* Kill the remaining (whole) subtrees */
1531 switch (offsets[0]) {
1535 nr = i_data[EXT4_IND_BLOCK];
1537 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1538 i_data[EXT4_IND_BLOCK] = 0;
1540 case EXT4_IND_BLOCK:
1543 nr = i_data[EXT4_DIND_BLOCK];
1545 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1546 i_data[EXT4_DIND_BLOCK] = 0;
1548 case EXT4_DIND_BLOCK:
1551 nr = i_data[EXT4_TIND_BLOCK];
1553 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1554 i_data[EXT4_TIND_BLOCK] = 0;
1556 case EXT4_TIND_BLOCK: