aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ext4/inode.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2011-08-01 19:56:03 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2011-08-01 19:56:03 -0400
commit60ad4466821a96913a9b567115e194ed1087c2d7 (patch)
treecd488ba72a60f856b85a467763fb633cbe7ef2d9 /fs/ext4/inode.c
parent1b8e94993c4752d98c33903aa836acc15f7e6d5c (diff)
parent79a77c5ac34cc27ccbfbdf7113b41cdd93534eab (diff)
Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (60 commits) ext4: prevent memory leaks from ext4_mb_init_backend() on error path ext4: use EXT4_BAD_INO for buddy cache to avoid colliding with valid inode # ext4: use ext4_msg() instead of printk in mballoc ext4: use ext4_kvzalloc()/ext4_kvmalloc() for s_group_desc and s_group_info ext4: introduce ext4_kvmalloc(), ext4_kzalloc(), and ext4_kvfree() ext4: use the correct error exit path in ext4_init_inode_table() ext4: add missing kfree() on error return path in add_new_gdb() ext4: change umode_t in tracepoint headers to be an explicit __u16 ext4: fix races in ext4_sync_parent() ext4: Fix overflow caused by missing cast in ext4_fallocate() ext4: add action of moving index in ext4_ext_rm_idx for Punch Hole ext4: simplify parameters of reserve_backup_gdb() ext4: simplify parameters of add_new_gdb() ext4: remove lock_buffer in bclean() and setup_new_group_blocks() ext4: simplify journal handling in setup_new_group_blocks() ext4: let setup_new_group_blocks() set multiple bits at a time ext4: fix a typo in ext4_group_extend() ext4: let ext4_group_add_blocks() handle 0 blocks quickly ext4: let ext4_group_add_blocks() return an error code ext4: rename ext4_add_groupblocks() to ext4_group_add_blocks() ... Fix up conflict in fs/ext4/inode.c: commit aacfc19c626e ("fs: simplify the blockdev_direct_IO prototype") had changed the ext4_ind_direct_IO() function for the new simplified calling convention, while commit dae1e52cb126 ("ext4: move ext4_ind_* functions from inode.c to indirect.c") moved the function to another file.
Diffstat (limited to 'fs/ext4/inode.c')
-rw-r--r--fs/ext4/inode.c1596
1 files changed, 37 insertions, 1559 deletions
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 3e5191f9f398..d47264cafee0 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -12,10 +12,6 @@
12 * 12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds 13 * Copyright (C) 1991, 1992 Linus Torvalds
14 * 14 *
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek 15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz) 16 * (jj@sunsite.ms.mff.cuni.cz)
21 * 17 *
@@ -47,6 +43,7 @@
47#include "xattr.h" 43#include "xattr.h"
48#include "acl.h" 44#include "acl.h"
49#include "ext4_extents.h" 45#include "ext4_extents.h"
46#include "truncate.h"
50 47
51#include <trace/events/ext4.h> 48#include <trace/events/ext4.h>
52 49
@@ -89,72 +86,6 @@ static int ext4_inode_is_fast_symlink(struct inode *inode)
89} 86}
90 87
91/* 88/*
92 * Work out how many blocks we need to proceed with the next chunk of a
93 * truncate transaction.
94 */
95static unsigned long blocks_for_truncate(struct inode *inode)
96{
97 ext4_lblk_t needed;
98
99 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
100
101 /* Give ourselves just enough room to cope with inodes in which
102 * i_blocks is corrupt: we've seen disk corruptions in the past
103 * which resulted in random data in an inode which looked enough
104 * like a regular file for ext4 to try to delete it. Things
105 * will go a bit crazy if that happens, but at least we should
106 * try not to panic the whole kernel. */
107 if (needed < 2)
108 needed = 2;
109
110 /* But we need to bound the transaction so we don't overflow the
111 * journal. */
112 if (needed > EXT4_MAX_TRANS_DATA)
113 needed = EXT4_MAX_TRANS_DATA;
114
115 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
116}
117
118/*
119 * Truncate transactions can be complex and absolutely huge. So we need to
120 * be able to restart the transaction at a conventient checkpoint to make
121 * sure we don't overflow the journal.
122 *
123 * start_transaction gets us a new handle for a truncate transaction,
124 * and extend_transaction tries to extend the existing one a bit. If
125 * extend fails, we need to propagate the failure up and restart the
126 * transaction in the top-level truncate loop. --sct
127 */
128static handle_t *start_transaction(struct inode *inode)
129{
130 handle_t *result;
131
132 result = ext4_journal_start(inode, blocks_for_truncate(inode));
133 if (!IS_ERR(result))
134 return result;
135
136 ext4_std_error(inode->i_sb, PTR_ERR(result));
137 return result;
138}
139
140/*
141 * Try to extend this transaction for the purposes of truncation.
142 *
143 * Returns 0 if we managed to create more room. If we can't create more
144 * room, and the transaction must be restarted we return 1.
145 */
146static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
147{
148 if (!ext4_handle_valid(handle))
149 return 0;
150 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
151 return 0;
152 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
153 return 0;
154 return 1;
155}
156
157/*
158 * Restart the transaction associated with *handle. This does a commit, 89 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against 90 * so before we call here everything must be consistently dirtied against
160 * this transaction. 91 * this transaction.
@@ -190,6 +121,33 @@ void ext4_evict_inode(struct inode *inode)
190 121
191 trace_ext4_evict_inode(inode); 122 trace_ext4_evict_inode(inode);
192 if (inode->i_nlink) { 123 if (inode->i_nlink) {
124 /*
125 * When journalling data dirty buffers are tracked only in the
126 * journal. So although mm thinks everything is clean and
127 * ready for reaping the inode might still have some pages to
128 * write in the running transaction or waiting to be
129 * checkpointed. Thus calling jbd2_journal_invalidatepage()
130 * (via truncate_inode_pages()) to discard these buffers can
131 * cause data loss. Also even if we did not discard these
132 * buffers, we would have no way to find them after the inode
133 * is reaped and thus user could see stale data if he tries to
134 * read them before the transaction is checkpointed. So be
135 * careful and force everything to disk here... We use
136 * ei->i_datasync_tid to store the newest transaction
137 * containing inode's data.
138 *
139 * Note that directories do not have this problem because they
140 * don't use page cache.
141 */
142 if (ext4_should_journal_data(inode) &&
143 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
144 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
145 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
146
147 jbd2_log_start_commit(journal, commit_tid);
148 jbd2_log_wait_commit(journal, commit_tid);
149 filemap_write_and_wait(&inode->i_data);
150 }
193 truncate_inode_pages(&inode->i_data, 0); 151 truncate_inode_pages(&inode->i_data, 0);
194 goto no_delete; 152 goto no_delete;
195 } 153 }
@@ -204,7 +162,7 @@ void ext4_evict_inode(struct inode *inode)
204 if (is_bad_inode(inode)) 162 if (is_bad_inode(inode))
205 goto no_delete; 163 goto no_delete;
206 164
207 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); 165 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
208 if (IS_ERR(handle)) { 166 if (IS_ERR(handle)) {
209 ext4_std_error(inode->i_sb, PTR_ERR(handle)); 167 ext4_std_error(inode->i_sb, PTR_ERR(handle));
210 /* 168 /*
@@ -277,793 +235,6 @@ no_delete:
277 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ 235 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
278} 236}
279 237
280typedef struct {
281 __le32 *p;
282 __le32 key;
283 struct buffer_head *bh;
284} Indirect;
285
286static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
287{
288 p->key = *(p->p = v);
289 p->bh = bh;
290}
291
292/**
293 * ext4_block_to_path - parse the block number into array of offsets
294 * @inode: inode in question (we are only interested in its superblock)
295 * @i_block: block number to be parsed
296 * @offsets: array to store the offsets in
297 * @boundary: set this non-zero if the referred-to block is likely to be
298 * followed (on disk) by an indirect block.
299 *
300 * To store the locations of file's data ext4 uses a data structure common
301 * for UNIX filesystems - tree of pointers anchored in the inode, with
302 * data blocks at leaves and indirect blocks in intermediate nodes.
303 * This function translates the block number into path in that tree -
304 * return value is the path length and @offsets[n] is the offset of
305 * pointer to (n+1)th node in the nth one. If @block is out of range
306 * (negative or too large) warning is printed and zero returned.
307 *
308 * Note: function doesn't find node addresses, so no IO is needed. All
309 * we need to know is the capacity of indirect blocks (taken from the
310 * inode->i_sb).
311 */
312
313/*
314 * Portability note: the last comparison (check that we fit into triple
315 * indirect block) is spelled differently, because otherwise on an
316 * architecture with 32-bit longs and 8Kb pages we might get into trouble
317 * if our filesystem had 8Kb blocks. We might use long long, but that would
318 * kill us on x86. Oh, well, at least the sign propagation does not matter -
319 * i_block would have to be negative in the very beginning, so we would not
320 * get there at all.
321 */
322
323static int ext4_block_to_path(struct inode *inode,
324 ext4_lblk_t i_block,
325 ext4_lblk_t offsets[4], int *boundary)
326{
327 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
328 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
329 const long direct_blocks = EXT4_NDIR_BLOCKS,
330 indirect_blocks = ptrs,
331 double_blocks = (1 << (ptrs_bits * 2));
332 int n = 0;
333 int final = 0;
334
335 if (i_block < direct_blocks) {
336 offsets[n++] = i_block;
337 final = direct_blocks;
338 } else if ((i_block -= direct_blocks) < indirect_blocks) {
339 offsets[n++] = EXT4_IND_BLOCK;
340 offsets[n++] = i_block;
341 final = ptrs;
342 } else if ((i_block -= indirect_blocks) < double_blocks) {
343 offsets[n++] = EXT4_DIND_BLOCK;
344 offsets[n++] = i_block >> ptrs_bits;
345 offsets[n++] = i_block & (ptrs - 1);
346 final = ptrs;
347 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
348 offsets[n++] = EXT4_TIND_BLOCK;
349 offsets[n++] = i_block >> (ptrs_bits * 2);
350 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
351 offsets[n++] = i_block & (ptrs - 1);
352 final = ptrs;
353 } else {
354 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
355 i_block + direct_blocks +
356 indirect_blocks + double_blocks, inode->i_ino);
357 }
358 if (boundary)
359 *boundary = final - 1 - (i_block & (ptrs - 1));
360 return n;
361}
362
363static int __ext4_check_blockref(const char *function, unsigned int line,
364 struct inode *inode,
365 __le32 *p, unsigned int max)
366{
367 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
368 __le32 *bref = p;
369 unsigned int blk;
370
371 while (bref < p+max) {
372 blk = le32_to_cpu(*bref++);
373 if (blk &&
374 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
375 blk, 1))) {
376 es->s_last_error_block = cpu_to_le64(blk);
377 ext4_error_inode(inode, function, line, blk,
378 "invalid block");
379 return -EIO;
380 }
381 }
382 return 0;
383}
384
385
386#define ext4_check_indirect_blockref(inode, bh) \
387 __ext4_check_blockref(__func__, __LINE__, inode, \
388 (__le32 *)(bh)->b_data, \
389 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
390
391#define ext4_check_inode_blockref(inode) \
392 __ext4_check_blockref(__func__, __LINE__, inode, \
393 EXT4_I(inode)->i_data, \
394 EXT4_NDIR_BLOCKS)
395
396/**
397 * ext4_get_branch - read the chain of indirect blocks leading to data
398 * @inode: inode in question
399 * @depth: depth of the chain (1 - direct pointer, etc.)
400 * @offsets: offsets of pointers in inode/indirect blocks
401 * @chain: place to store the result
402 * @err: here we store the error value
403 *
404 * Function fills the array of triples <key, p, bh> and returns %NULL
405 * if everything went OK or the pointer to the last filled triple
406 * (incomplete one) otherwise. Upon the return chain[i].key contains
407 * the number of (i+1)-th block in the chain (as it is stored in memory,
408 * i.e. little-endian 32-bit), chain[i].p contains the address of that
409 * number (it points into struct inode for i==0 and into the bh->b_data
410 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
411 * block for i>0 and NULL for i==0. In other words, it holds the block
412 * numbers of the chain, addresses they were taken from (and where we can
413 * verify that chain did not change) and buffer_heads hosting these
414 * numbers.
415 *
416 * Function stops when it stumbles upon zero pointer (absent block)
417 * (pointer to last triple returned, *@err == 0)
418 * or when it gets an IO error reading an indirect block
419 * (ditto, *@err == -EIO)
420 * or when it reads all @depth-1 indirect blocks successfully and finds
421 * the whole chain, all way to the data (returns %NULL, *err == 0).
422 *
423 * Need to be called with
424 * down_read(&EXT4_I(inode)->i_data_sem)
425 */
426static Indirect *ext4_get_branch(struct inode *inode, int depth,
427 ext4_lblk_t *offsets,
428 Indirect chain[4], int *err)
429{
430 struct super_block *sb = inode->i_sb;
431 Indirect *p = chain;
432 struct buffer_head *bh;
433
434 *err = 0;
435 /* i_data is not going away, no lock needed */
436 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
437 if (!p->key)
438 goto no_block;
439 while (--depth) {
440 bh = sb_getblk(sb, le32_to_cpu(p->key));
441 if (unlikely(!bh))
442 goto failure;
443
444 if (!bh_uptodate_or_lock(bh)) {
445 if (bh_submit_read(bh) < 0) {
446 put_bh(bh);
447 goto failure;
448 }
449 /* validate block references */
450 if (ext4_check_indirect_blockref(inode, bh)) {
451 put_bh(bh);
452 goto failure;
453 }
454 }
455
456 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
457 /* Reader: end */
458 if (!p->key)
459 goto no_block;
460 }
461 return NULL;
462
463failure:
464 *err = -EIO;
465no_block:
466 return p;
467}
468
469/**
470 * ext4_find_near - find a place for allocation with sufficient locality
471 * @inode: owner
472 * @ind: descriptor of indirect block.
473 *
474 * This function returns the preferred place for block allocation.
475 * It is used when heuristic for sequential allocation fails.
476 * Rules are:
477 * + if there is a block to the left of our position - allocate near it.
478 * + if pointer will live in indirect block - allocate near that block.
479 * + if pointer will live in inode - allocate in the same
480 * cylinder group.
481 *
482 * In the latter case we colour the starting block by the callers PID to
483 * prevent it from clashing with concurrent allocations for a different inode
484 * in the same block group. The PID is used here so that functionally related
485 * files will be close-by on-disk.
486 *
487 * Caller must make sure that @ind is valid and will stay that way.
488 */
489static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
490{
491 struct ext4_inode_info *ei = EXT4_I(inode);
492 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
493 __le32 *p;
494 ext4_fsblk_t bg_start;
495 ext4_fsblk_t last_block;
496 ext4_grpblk_t colour;
497 ext4_group_t block_group;
498 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
499
500 /* Try to find previous block */
501 for (p = ind->p - 1; p >= start; p--) {
502 if (*p)
503 return le32_to_cpu(*p);
504 }
505
506 /* No such thing, so let's try location of indirect block */
507 if (ind->bh)
508 return ind->bh->b_blocknr;
509
510 /*
511 * It is going to be referred to from the inode itself? OK, just put it
512 * into the same cylinder group then.
513 */
514 block_group = ei->i_block_group;
515 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
516 block_group &= ~(flex_size-1);
517 if (S_ISREG(inode->i_mode))
518 block_group++;
519 }
520 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
521 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
522
523 /*
524 * If we are doing delayed allocation, we don't need take
525 * colour into account.
526 */
527 if (test_opt(inode->i_sb, DELALLOC))
528 return bg_start;
529
530 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
531 colour = (current->pid % 16) *
532 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
533 else
534 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
535 return bg_start + colour;
536}
537
538/**
539 * ext4_find_goal - find a preferred place for allocation.
540 * @inode: owner
541 * @block: block we want
542 * @partial: pointer to the last triple within a chain
543 *
544 * Normally this function find the preferred place for block allocation,
545 * returns it.
546 * Because this is only used for non-extent files, we limit the block nr
547 * to 32 bits.
548 */
549static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
550 Indirect *partial)
551{
552 ext4_fsblk_t goal;
553
554 /*
555 * XXX need to get goal block from mballoc's data structures
556 */
557
558 goal = ext4_find_near(inode, partial);
559 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
560 return goal;
561}
562
563/**
564 * ext4_blks_to_allocate - Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
566 *
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
571 *
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
574 */
575static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
576 int blocks_to_boundary)
577{
578 unsigned int count = 0;
579
580 /*
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
583 */
584 if (k > 0) {
585 /* right now we don't handle cross boundary allocation */
586 if (blks < blocks_to_boundary + 1)
587 count += blks;
588 else
589 count += blocks_to_boundary + 1;
590 return count;
591 }
592
593 count++;
594 while (count < blks && count <= blocks_to_boundary &&
595 le32_to_cpu(*(branch[0].p + count)) == 0) {
596 count++;
597 }
598 return count;
599}
600
601/**
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @handle: handle for this transaction
604 * @inode: inode which needs allocated blocks
605 * @iblock: the logical block to start allocated at
606 * @goal: preferred physical block of allocation
607 * @indirect_blks: the number of blocks need to allocate for indirect
608 * blocks
609 * @blks: number of desired blocks
610 * @new_blocks: on return it will store the new block numbers for
611 * the indirect blocks(if needed) and the first direct block,
612 * @err: on return it will store the error code
613 *
614 * This function will return the number of blocks allocated as
615 * requested by the passed-in parameters.
616 */
617static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
618 ext4_lblk_t iblock, ext4_fsblk_t goal,
619 int indirect_blks, int blks,
620 ext4_fsblk_t new_blocks[4], int *err)
621{
622 struct ext4_allocation_request ar;
623 int target, i;
624 unsigned long count = 0, blk_allocated = 0;
625 int index = 0;
626 ext4_fsblk_t current_block = 0;
627 int ret = 0;
628
629 /*
630 * Here we try to allocate the requested multiple blocks at once,
631 * on a best-effort basis.
632 * To build a branch, we should allocate blocks for
633 * the indirect blocks(if not allocated yet), and at least
634 * the first direct block of this branch. That's the
635 * minimum number of blocks need to allocate(required)
636 */
637 /* first we try to allocate the indirect blocks */
638 target = indirect_blks;
639 while (target > 0) {
640 count = target;
641 /* allocating blocks for indirect blocks and direct blocks */
642 current_block = ext4_new_meta_blocks(handle, inode, goal,
643 0, &count, err);
644 if (*err)
645 goto failed_out;
646
647 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
648 EXT4_ERROR_INODE(inode,
649 "current_block %llu + count %lu > %d!",
650 current_block, count,
651 EXT4_MAX_BLOCK_FILE_PHYS);
652 *err = -EIO;
653 goto failed_out;
654 }
655
656 target -= count;
657 /* allocate blocks for indirect blocks */
658 while (index < indirect_blks && count) {
659 new_blocks[index++] = current_block++;
660 count--;
661 }
662 if (count > 0) {
663 /*
664 * save the new block number
665 * for the first direct block
666 */
667 new_blocks[index] = current_block;
668 printk(KERN_INFO "%s returned more blocks than "
669 "requested\n", __func__);
670 WARN_ON(1);
671 break;
672 }
673 }
674
675 target = blks - count ;
676 blk_allocated = count;
677 if (!target)
678 goto allocated;
679 /* Now allocate data blocks */
680 memset(&ar, 0, sizeof(ar));
681 ar.inode = inode;
682 ar.goal = goal;
683 ar.len = target;
684 ar.logical = iblock;
685 if (S_ISREG(inode->i_mode))
686 /* enable in-core preallocation only for regular files */
687 ar.flags = EXT4_MB_HINT_DATA;
688
689 current_block = ext4_mb_new_blocks(handle, &ar, err);
690 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
691 EXT4_ERROR_INODE(inode,
692 "current_block %llu + ar.len %d > %d!",
693 current_block, ar.len,
694 EXT4_MAX_BLOCK_FILE_PHYS);
695 *err = -EIO;
696 goto failed_out;
697 }
698
699 if (*err && (target == blks)) {
700 /*
701 * if the allocation failed and we didn't allocate
702 * any blocks before
703 */
704 goto failed_out;
705 }
706 if (!*err) {
707 if (target == blks) {
708 /*
709 * save the new block number
710 * for the first direct block
711 */
712 new_blocks[index] = current_block;
713 }
714 blk_allocated += ar.len;
715 }
716allocated:
717 /* total number of blocks allocated for direct blocks */
718 ret = blk_allocated;
719 *err = 0;
720 return ret;
721failed_out:
722 for (i = 0; i < index; i++)
723 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
724 return ret;
725}
726
727/**
728 * ext4_alloc_branch - allocate and set up a chain of blocks.
729 * @handle: handle for this transaction
730 * @inode: owner
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @goal: preferred place for allocation
734 * @offsets: offsets (in the blocks) to store the pointers to next.
735 * @branch: place to store the chain in.
736 *
737 * This function allocates blocks, zeroes out all but the last one,
738 * links them into chain and (if we are synchronous) writes them to disk.
739 * In other words, it prepares a branch that can be spliced onto the
740 * inode. It stores the information about that chain in the branch[], in
741 * the same format as ext4_get_branch() would do. We are calling it after
742 * we had read the existing part of chain and partial points to the last
743 * triple of that (one with zero ->key). Upon the exit we have the same
744 * picture as after the successful ext4_get_block(), except that in one
745 * place chain is disconnected - *branch->p is still zero (we did not
746 * set the last link), but branch->key contains the number that should
747 * be placed into *branch->p to fill that gap.
748 *
749 * If allocation fails we free all blocks we've allocated (and forget
750 * their buffer_heads) and return the error value the from failed
751 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
752 * as described above and return 0.
753 */
754static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
755 ext4_lblk_t iblock, int indirect_blks,
756 int *blks, ext4_fsblk_t goal,
757 ext4_lblk_t *offsets, Indirect *branch)
758{
759 int blocksize = inode->i_sb->s_blocksize;
760 int i, n = 0;
761 int err = 0;
762 struct buffer_head *bh;
763 int num;
764 ext4_fsblk_t new_blocks[4];
765 ext4_fsblk_t current_block;
766
767 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
768 *blks, new_blocks, &err);
769 if (err)
770 return err;
771
772 branch[0].key = cpu_to_le32(new_blocks[0]);
773 /*
774 * metadata blocks and data blocks are allocated.
775 */
776 for (n = 1; n <= indirect_blks; n++) {
777 /*
778 * Get buffer_head for parent block, zero it out
779 * and set the pointer to new one, then send
780 * parent to disk.
781 */
782 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
783 if (unlikely(!bh)) {
784 err = -EIO;
785 goto failed;
786 }
787
788 branch[n].bh = bh;
789 lock_buffer(bh);
790 BUFFER_TRACE(bh, "call get_create_access");
791 err = ext4_journal_get_create_access(handle, bh);
792 if (err) {
793 /* Don't brelse(bh) here; it's done in
794 * ext4_journal_forget() below */
795 unlock_buffer(bh);
796 goto failed;
797 }
798
799 memset(bh->b_data, 0, blocksize);
800 branch[n].p = (__le32 *) bh->b_data + offsets[n];
801 branch[n].key = cpu_to_le32(new_blocks[n]);
802 *branch[n].p = branch[n].key;
803 if (n == indirect_blks) {
804 current_block = new_blocks[n];
805 /*
806 * End of chain, update the last new metablock of
807 * the chain to point to the new allocated
808 * data blocks numbers
809 */
810 for (i = 1; i < num; i++)
811 *(branch[n].p + i) = cpu_to_le32(++current_block);
812 }
813 BUFFER_TRACE(bh, "marking uptodate");
814 set_buffer_uptodate(bh);
815 unlock_buffer(bh);
816
817 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
818 err = ext4_handle_dirty_metadata(handle, inode, bh);
819 if (err)
820 goto failed;
821 }
822 *blks = num;
823 return err;
824failed:
825 /* Allocation failed, free what we already allocated */
826 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
827 for (i = 1; i <= n ; i++) {
828 /*
829 * branch[i].bh is newly allocated, so there is no
830 * need to revoke the block, which is why we don't
831 * need to set EXT4_FREE_BLOCKS_METADATA.
832 */
833 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
834 EXT4_FREE_BLOCKS_FORGET);
835 }
836 for (i = n+1; i < indirect_blks; i++)
837 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
838
839 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
840
841 return err;
842}
843
844/**
845 * ext4_splice_branch - splice the allocated branch onto inode.
846 * @handle: handle for this transaction
847 * @inode: owner
848 * @block: (logical) number of block we are adding
849 * @chain: chain of indirect blocks (with a missing link - see
850 * ext4_alloc_branch)
851 * @where: location of missing link
852 * @num: number of indirect blocks we are adding
853 * @blks: number of direct blocks we are adding
854 *
855 * This function fills the missing link and does all housekeeping needed in
856 * inode (->i_blocks, etc.). In case of success we end up with the full
857 * chain to new block and return 0.
858 */
859static int ext4_splice_branch(handle_t *handle, struct inode *inode,
860 ext4_lblk_t block, Indirect *where, int num,
861 int blks)
862{
863 int i;
864 int err = 0;
865 ext4_fsblk_t current_block;
866
867 /*
868 * If we're splicing into a [td]indirect block (as opposed to the
869 * inode) then we need to get write access to the [td]indirect block
870 * before the splice.
871 */
872 if (where->bh) {
873 BUFFER_TRACE(where->bh, "get_write_access");
874 err = ext4_journal_get_write_access(handle, where->bh);
875 if (err)
876 goto err_out;
877 }
878 /* That's it */
879
880 *where->p = where->key;
881
882 /*
883 * Update the host buffer_head or inode to point to more just allocated
884 * direct blocks blocks
885 */
886 if (num == 0 && blks > 1) {
887 current_block = le32_to_cpu(where->key) + 1;
888 for (i = 1; i < blks; i++)
889 *(where->p + i) = cpu_to_le32(current_block++);
890 }
891
892 /* We are done with atomic stuff, now do the rest of housekeeping */
893 /* had we spliced it onto indirect block? */
894 if (where->bh) {
895 /*
896 * If we spliced it onto an indirect block, we haven't
897 * altered the inode. Note however that if it is being spliced
898 * onto an indirect block at the very end of the file (the
899 * file is growing) then we *will* alter the inode to reflect
900 * the new i_size. But that is not done here - it is done in
901 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
902 */
903 jbd_debug(5, "splicing indirect only\n");
904 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
905 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
906 if (err)
907 goto err_out;
908 } else {
909 /*
910 * OK, we spliced it into the inode itself on a direct block.
911 */
912 ext4_mark_inode_dirty(handle, inode);
913 jbd_debug(5, "splicing direct\n");
914 }
915 return err;
916
917err_out:
918 for (i = 1; i <= num; i++) {
919 /*
920 * branch[i].bh is newly allocated, so there is no
921 * need to revoke the block, which is why we don't
922 * need to set EXT4_FREE_BLOCKS_METADATA.
923 */
924 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
925 EXT4_FREE_BLOCKS_FORGET);
926 }
927 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
928 blks, 0);
929
930 return err;
931}
932
933/*
934 * The ext4_ind_map_blocks() function handles non-extents inodes
935 * (i.e., using the traditional indirect/double-indirect i_blocks
936 * scheme) for ext4_map_blocks().
937 *
938 * Allocation strategy is simple: if we have to allocate something, we will
939 * have to go the whole way to leaf. So let's do it before attaching anything
940 * to tree, set linkage between the newborn blocks, write them if sync is
941 * required, recheck the path, free and repeat if check fails, otherwise
942 * set the last missing link (that will protect us from any truncate-generated
943 * removals - all blocks on the path are immune now) and possibly force the
944 * write on the parent block.
945 * That has a nice additional property: no special recovery from the failed
946 * allocations is needed - we simply release blocks and do not touch anything
947 * reachable from inode.
948 *
949 * `handle' can be NULL if create == 0.
950 *
951 * return > 0, # of blocks mapped or allocated.
952 * return = 0, if plain lookup failed.
953 * return < 0, error case.
954 *
955 * The ext4_ind_get_blocks() function should be called with
956 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
957 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
958 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
959 * blocks.
960 */
961static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
962 struct ext4_map_blocks *map,
963 int flags)
964{
965 int err = -EIO;
966 ext4_lblk_t offsets[4];
967 Indirect chain[4];
968 Indirect *partial;
969 ext4_fsblk_t goal;
970 int indirect_blks;
971 int blocks_to_boundary = 0;
972 int depth;
973 int count = 0;
974 ext4_fsblk_t first_block = 0;
975
976 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
977 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
978 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
979 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
980 &blocks_to_boundary);
981
982 if (depth == 0)
983 goto out;
984
985 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
986
987 /* Simplest case - block found, no allocation needed */
988 if (!partial) {
989 first_block = le32_to_cpu(chain[depth - 1].key);
990 count++;
991 /*map more blocks*/
992 while (count < map->m_len && count <= blocks_to_boundary) {
993 ext4_fsblk_t blk;
994
995 blk = le32_to_cpu(*(chain[depth-1].p + count));
996
997 if (blk == first_block + count)
998 count++;
999 else
1000 break;
1001 }
1002 goto got_it;
1003 }
1004
1005 /* Next simple case - plain lookup or failed read of indirect block */
1006 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
1007 goto cleanup;
1008
1009 /*
1010 * Okay, we need to do block allocation.
1011 */
1012 goal = ext4_find_goal(inode, map->m_lblk, partial);
1013
1014 /* the number of blocks need to allocate for [d,t]indirect blocks */
1015 indirect_blks = (chain + depth) - partial - 1;
1016
1017 /*
1018 * Next look up the indirect map to count the totoal number of
1019 * direct blocks to allocate for this branch.
1020 */
1021 count = ext4_blks_to_allocate(partial, indirect_blks,
1022 map->m_len, blocks_to_boundary);
1023 /*
1024 * Block out ext4_truncate while we alter the tree
1025 */
1026 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1027 &count, goal,
1028 offsets + (partial - chain), partial);
1029
1030 /*
1031 * The ext4_splice_branch call will free and forget any buffers
1032 * on the new chain if there is a failure, but that risks using
1033 * up transaction credits, especially for bitmaps where the
1034 * credits cannot be returned. Can we handle this somehow? We
1035 * may need to return -EAGAIN upwards in the worst case. --sct
1036 */
1037 if (!err)
1038 err = ext4_splice_branch(handle, inode, map->m_lblk,
1039 partial, indirect_blks, count);
1040 if (err)
1041 goto cleanup;
1042
1043 map->m_flags |= EXT4_MAP_NEW;
1044
1045 ext4_update_inode_fsync_trans(handle, inode, 1);
1046got_it:
1047 map->m_flags |= EXT4_MAP_MAPPED;
1048 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1049 map->m_len = count;
1050 if (count > blocks_to_boundary)
1051 map->m_flags |= EXT4_MAP_BOUNDARY;
1052 err = count;
1053 /* Clean up and exit */
1054 partial = chain + depth - 1; /* the whole chain */
1055cleanup:
1056 while (partial > chain) {
1057 BUFFER_TRACE(partial->bh, "call brelse");
1058 brelse(partial->bh);
1059 partial--;
1060 }
1061out:
1062 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
1063 map->m_pblk, map->m_len, err);
1064 return err;
1065}
1066
1067#ifdef CONFIG_QUOTA 238#ifdef CONFIG_QUOTA
1068qsize_t *ext4_get_reserved_space(struct inode *inode) 239qsize_t *ext4_get_reserved_space(struct inode *inode)
1069{ 240{
@@ -1073,33 +244,6 @@ qsize_t *ext4_get_reserved_space(struct inode *inode)
1073 244
1074/* 245/*
1075 * Calculate the number of metadata blocks need to reserve 246 * Calculate the number of metadata blocks need to reserve
1076 * to allocate a new block at @lblocks for non extent file based file
1077 */
1078static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1079 sector_t lblock)
1080{
1081 struct ext4_inode_info *ei = EXT4_I(inode);
1082 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1083 int blk_bits;
1084
1085 if (lblock < EXT4_NDIR_BLOCKS)
1086 return 0;
1087
1088 lblock -= EXT4_NDIR_BLOCKS;
1089
1090 if (ei->i_da_metadata_calc_len &&
1091 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1092 ei->i_da_metadata_calc_len++;
1093 return 0;
1094 }
1095 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1096 ei->i_da_metadata_calc_len = 1;
1097 blk_bits = order_base_2(lblock);
1098 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1099}
1100
1101/*
1102 * Calculate the number of metadata blocks need to reserve
1103 * to allocate a block located at @lblock 247 * to allocate a block located at @lblock
1104 */ 248 */
1105static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) 249static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
@@ -1107,7 +251,7 @@ static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
1107 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) 251 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1108 return ext4_ext_calc_metadata_amount(inode, lblock); 252 return ext4_ext_calc_metadata_amount(inode, lblock);
1109 253
1110 return ext4_indirect_calc_metadata_amount(inode, lblock); 254 return ext4_ind_calc_metadata_amount(inode, lblock);
1111} 255}
1112 256
1113/* 257/*
@@ -1589,16 +733,6 @@ static int do_journal_get_write_access(handle_t *handle,
1589 return ret; 733 return ret;
1590} 734}
1591 735
1592/*
1593 * Truncate blocks that were not used by write. We have to truncate the
1594 * pagecache as well so that corresponding buffers get properly unmapped.
1595 */
1596static void ext4_truncate_failed_write(struct inode *inode)
1597{
1598 truncate_inode_pages(inode->i_mapping, inode->i_size);
1599 ext4_truncate(inode);
1600}
1601
1602static int ext4_get_block_write(struct inode *inode, sector_t iblock, 736static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1603 struct buffer_head *bh_result, int create); 737 struct buffer_head *bh_result, int create);
1604static int ext4_write_begin(struct file *file, struct address_space *mapping, 738static int ext4_write_begin(struct file *file, struct address_space *mapping,
@@ -1863,6 +997,7 @@ static int ext4_journalled_write_end(struct file *file,
1863 if (new_i_size > inode->i_size) 997 if (new_i_size > inode->i_size)
1864 i_size_write(inode, pos+copied); 998 i_size_write(inode, pos+copied);
1865 ext4_set_inode_state(inode, EXT4_STATE_JDATA); 999 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1000 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1866 if (new_i_size > EXT4_I(inode)->i_disksize) { 1001 if (new_i_size > EXT4_I(inode)->i_disksize) {
1867 ext4_update_i_disksize(inode, new_i_size); 1002 ext4_update_i_disksize(inode, new_i_size);
1868 ret2 = ext4_mark_inode_dirty(handle, inode); 1003 ret2 = ext4_mark_inode_dirty(handle, inode);
@@ -2571,6 +1706,7 @@ static int __ext4_journalled_writepage(struct page *page,
2571 write_end_fn); 1706 write_end_fn);
2572 if (ret == 0) 1707 if (ret == 0)
2573 ret = err; 1708 ret = err;
1709 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2574 err = ext4_journal_stop(handle); 1710 err = ext4_journal_stop(handle);
2575 if (!ret) 1711 if (!ret)
2576 ret = err; 1712 ret = err;
@@ -3450,112 +2586,6 @@ static int ext4_releasepage(struct page *page, gfp_t wait)
3450} 2586}
3451 2587
3452/* 2588/*
3453 * O_DIRECT for ext3 (or indirect map) based files
3454 *
3455 * If the O_DIRECT write will extend the file then add this inode to the
3456 * orphan list. So recovery will truncate it back to the original size
3457 * if the machine crashes during the write.
3458 *
3459 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3460 * crashes then stale disk data _may_ be exposed inside the file. But current
3461 * VFS code falls back into buffered path in that case so we are safe.
3462 */
3463static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3464 const struct iovec *iov, loff_t offset,
3465 unsigned long nr_segs)
3466{
3467 struct file *file = iocb->ki_filp;
3468 struct inode *inode = file->f_mapping->host;
3469 struct ext4_inode_info *ei = EXT4_I(inode);
3470 handle_t *handle;
3471 ssize_t ret;
3472 int orphan = 0;
3473 size_t count = iov_length(iov, nr_segs);
3474 int retries = 0;
3475
3476 if (rw == WRITE) {
3477 loff_t final_size = offset + count;
3478
3479 if (final_size > inode->i_size) {
3480 /* Credits for sb + inode write */
3481 handle = ext4_journal_start(inode, 2);
3482 if (IS_ERR(handle)) {
3483 ret = PTR_ERR(handle);
3484 goto out;
3485 }
3486 ret = ext4_orphan_add(handle, inode);
3487 if (ret) {
3488 ext4_journal_stop(handle);
3489 goto out;
3490 }
3491 orphan = 1;
3492 ei->i_disksize = inode->i_size;
3493 ext4_journal_stop(handle);
3494 }
3495 }
3496
3497retry:
3498 if (rw == READ && ext4_should_dioread_nolock(inode))
3499 ret = __blockdev_direct_IO(rw, iocb, inode,
3500 inode->i_sb->s_bdev, iov,
3501 offset, nr_segs,
3502 ext4_get_block, NULL, NULL, 0);
3503 else {
3504 ret = blockdev_direct_IO(rw, iocb, inode, iov,
3505 offset, nr_segs, ext4_get_block);
3506
3507 if (unlikely((rw & WRITE) && ret < 0)) {
3508 loff_t isize = i_size_read(inode);
3509 loff_t end = offset + iov_length(iov, nr_segs);
3510
3511 if (end > isize)
3512 ext4_truncate_failed_write(inode);
3513 }
3514 }
3515 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3516 goto retry;
3517
3518 if (orphan) {
3519 int err;
3520
3521 /* Credits for sb + inode write */
3522 handle = ext4_journal_start(inode, 2);
3523 if (IS_ERR(handle)) {
3524 /* This is really bad luck. We've written the data
3525 * but cannot extend i_size. Bail out and pretend
3526 * the write failed... */
3527 ret = PTR_ERR(handle);
3528 if (inode->i_nlink)
3529 ext4_orphan_del(NULL, inode);
3530
3531 goto out;
3532 }
3533 if (inode->i_nlink)
3534 ext4_orphan_del(handle, inode);
3535 if (ret > 0) {
3536 loff_t end = offset + ret;
3537 if (end > inode->i_size) {
3538 ei->i_disksize = end;
3539 i_size_write(inode, end);
3540 /*
3541 * We're going to return a positive `ret'
3542 * here due to non-zero-length I/O, so there's
3543 * no way of reporting error returns from
3544 * ext4_mark_inode_dirty() to userspace. So
3545 * ignore it.
3546 */
3547 ext4_mark_inode_dirty(handle, inode);
3548 }
3549 }
3550 err = ext4_journal_stop(handle);
3551 if (ret == 0)
3552 ret = err;
3553 }
3554out:
3555 return ret;
3556}
3557
3558/*
3559 * ext4_get_block used when preparing for a DIO write or buffer write. 2589 * ext4_get_block used when preparing for a DIO write or buffer write.
3560 * We allocate an uinitialized extent if blocks haven't been allocated. 2590 * We allocate an uinitialized extent if blocks haven't been allocated.
3561 * The extent will be converted to initialized after the IO is complete. 2591 * The extent will be converted to initialized after the IO is complete.
@@ -4033,383 +3063,6 @@ unlock:
4033 return err; 3063 return err;
4034} 3064}
4035 3065
4036/*
4037 * Probably it should be a library function... search for first non-zero word
4038 * or memcmp with zero_page, whatever is better for particular architecture.
4039 * Linus?
4040 */
4041static inline int all_zeroes(__le32 *p, __le32 *q)
4042{
4043 while (p < q)
4044 if (*p++)
4045 return 0;
4046 return 1;
4047}
4048
4049/**
4050 * ext4_find_shared - find the indirect blocks for partial truncation.
4051 * @inode: inode in question
4052 * @depth: depth of the affected branch
4053 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4054 * @chain: place to store the pointers to partial indirect blocks
4055 * @top: place to the (detached) top of branch
4056 *
4057 * This is a helper function used by ext4_truncate().
4058 *
4059 * When we do truncate() we may have to clean the ends of several
4060 * indirect blocks but leave the blocks themselves alive. Block is
4061 * partially truncated if some data below the new i_size is referred
4062 * from it (and it is on the path to the first completely truncated
4063 * data block, indeed). We have to free the top of that path along
4064 * with everything to the right of the path. Since no allocation
4065 * past the truncation point is possible until ext4_truncate()
4066 * finishes, we may safely do the latter, but top of branch may
4067 * require special attention - pageout below the truncation point
4068 * might try to populate it.
4069 *
4070 * We atomically detach the top of branch from the tree, store the
4071 * block number of its root in *@top, pointers to buffer_heads of
4072 * partially truncated blocks - in @chain[].bh and pointers to
4073 * their last elements that should not be removed - in
4074 * @chain[].p. Return value is the pointer to last filled element
4075 * of @chain.
4076 *
4077 * The work left to caller to do the actual freeing of subtrees:
4078 * a) free the subtree starting from *@top
4079 * b) free the subtrees whose roots are stored in
4080 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4081 * c) free the subtrees growing from the inode past the @chain[0].
4082 * (no partially truncated stuff there). */
4083
4084static Indirect *ext4_find_shared(struct inode *inode, int depth,
4085 ext4_lblk_t offsets[4], Indirect chain[4],
4086 __le32 *top)
4087{
4088 Indirect *partial, *p;
4089 int k, err;
4090
4091 *top = 0;
4092 /* Make k index the deepest non-null offset + 1 */
4093 for (k = depth; k > 1 && !offsets[k-1]; k--)
4094 ;
4095 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4096 /* Writer: pointers */
4097 if (!partial)
4098 partial = chain + k-1;
4099 /*
4100 * If the branch acquired continuation since we've looked at it -
4101 * fine, it should all survive and (new) top doesn't belong to us.
4102 */
4103 if (!partial->key && *partial->p)
4104 /* Writer: end */
4105 goto no_top;
4106 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4107 ;
4108 /*
4109 * OK, we've found the last block that must survive. The rest of our
4110 * branch should be detached before unlocking. However, if that rest
4111 * of branch is all ours and does not grow immediately from the inode
4112 * it's easier to cheat and just decrement partial->p.
4113 */
4114 if (p == chain + k - 1 && p > chain) {
4115 p->p--;
4116 } else {
4117 *top = *p->p;
4118 /* Nope, don't do this in ext4. Must leave the tree intact */
4119#if 0
4120 *p->p = 0;
4121#endif
4122 }
4123 /* Writer: end */
4124
4125 while (partial > p) {
4126 brelse(partial->bh);
4127 partial--;
4128 }
4129no_top:
4130 return partial;
4131}
4132
4133/*
4134 * Zero a number of block pointers in either an inode or an indirect block.
4135 * If we restart the transaction we must again get write access to the
4136 * indirect block for further modification.
4137 *
4138 * We release `count' blocks on disk, but (last - first) may be greater
4139 * than `count' because there can be holes in there.
4140 *
4141 * Return 0 on success, 1 on invalid block range
4142 * and < 0 on fatal error.
4143 */
4144static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4145 struct buffer_head *bh,
4146 ext4_fsblk_t block_to_free,
4147 unsigned long count, __le32 *first,
4148 __le32 *last)
4149{
4150 __le32 *p;
4151 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4152 int err;
4153
4154 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4155 flags |= EXT4_FREE_BLOCKS_METADATA;
4156
4157 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4158 count)) {
4159 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4160 "blocks %llu len %lu",
4161 (unsigned long long) block_to_free, count);
4162 return 1;
4163 }
4164
4165 if (try_to_extend_transaction(handle, inode)) {
4166 if (bh) {
4167 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4168 err = ext4_handle_dirty_metadata(handle, inode, bh);
4169 if (unlikely(err))
4170 goto out_err;
4171 }
4172 err = ext4_mark_inode_dirty(handle, inode);
4173 if (unlikely(err))
4174 goto out_err;
4175 err = ext4_truncate_restart_trans(handle, inode,
4176 blocks_for_truncate(inode));
4177 if (unlikely(err))
4178 goto out_err;
4179 if (bh) {
4180 BUFFER_TRACE(bh, "retaking write access");
4181 err = ext4_journal_get_write_access(handle, bh);
4182 if (unlikely(err))
4183 goto out_err;
4184 }
4185 }
4186
4187 for (p = first; p < last; p++)
4188 *p = 0;
4189
4190 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
4191 return 0;
4192out_err:
4193 ext4_std_error(inode->i_sb, err);
4194 return err;
4195}
4196
4197/**
4198 * ext4_free_data - free a list of data blocks
4199 * @handle: handle for this transaction
4200 * @inode: inode we are dealing with
4201 * @this_bh: indirect buffer_head which contains *@first and *@last
4202 * @first: array of block numbers
4203 * @last: points immediately past the end of array
4204 *
4205 * We are freeing all blocks referred from that array (numbers are stored as
4206 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4207 *
4208 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4209 * blocks are contiguous then releasing them at one time will only affect one
4210 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4211 * actually use a lot of journal space.
4212 *
4213 * @this_bh will be %NULL if @first and @last point into the inode's direct
4214 * block pointers.
4215 */
4216static void ext4_free_data(handle_t *handle, struct inode *inode,
4217 struct buffer_head *this_bh,
4218 __le32 *first, __le32 *last)
4219{
4220 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4221 unsigned long count = 0; /* Number of blocks in the run */
4222 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4223 corresponding to
4224 block_to_free */
4225 ext4_fsblk_t nr; /* Current block # */
4226 __le32 *p; /* Pointer into inode/ind
4227 for current block */
4228 int err = 0;
4229
4230 if (this_bh) { /* For indirect block */
4231 BUFFER_TRACE(this_bh, "get_write_access");
4232 err = ext4_journal_get_write_access(handle, this_bh);
4233 /* Important: if we can't update the indirect pointers
4234 * to the blocks, we can't free them. */
4235 if (err)
4236 return;
4237 }
4238
4239 for (p = first; p < last; p++) {
4240 nr = le32_to_cpu(*p);
4241 if (nr) {
4242 /* accumulate blocks to free if they're contiguous */
4243 if (count == 0) {
4244 block_to_free = nr;
4245 block_to_free_p = p;
4246 count = 1;
4247 } else if (nr == block_to_free + count) {
4248 count++;
4249 } else {
4250 err = ext4_clear_blocks(handle, inode, this_bh,
4251 block_to_free, count,
4252 block_to_free_p, p);
4253 if (err)
4254 break;
4255 block_to_free = nr;
4256 block_to_free_p = p;
4257 count = 1;
4258 }
4259 }
4260 }
4261
4262 if (!err && count > 0)
4263 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4264 count, block_to_free_p, p);
4265 if (err < 0)
4266 /* fatal error */
4267 return;
4268
4269 if (this_bh) {
4270 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4271
4272 /*
4273 * The buffer head should have an attached journal head at this
4274 * point. However, if the data is corrupted and an indirect
4275 * block pointed to itself, it would have been detached when
4276 * the block was cleared. Check for this instead of OOPSing.
4277 */
4278 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4279 ext4_handle_dirty_metadata(handle, inode, this_bh);
4280 else
4281 EXT4_ERROR_INODE(inode,
4282 "circular indirect block detected at "
4283 "block %llu",
4284 (unsigned long long) this_bh->b_blocknr);
4285 }
4286}
4287
4288/**
4289 * ext4_free_branches - free an array of branches
4290 * @handle: JBD handle for this transaction
4291 * @inode: inode we are dealing with
4292 * @parent_bh: the buffer_head which contains *@first and *@last
4293 * @first: array of block numbers
4294 * @last: pointer immediately past the end of array
4295 * @depth: depth of the branches to free
4296 *
4297 * We are freeing all blocks referred from these branches (numbers are
4298 * stored as little-endian 32-bit) and updating @inode->i_blocks
4299 * appropriately.
4300 */
4301static void ext4_free_branches(handle_t *handle, struct inode *inode,
4302 struct buffer_head *parent_bh,
4303 __le32 *first, __le32 *last, int depth)
4304{
4305 ext4_fsblk_t nr;
4306 __le32 *p;
4307
4308 if (ext4_handle_is_aborted(handle))
4309 return;
4310
4311 if (depth--) {
4312 struct buffer_head *bh;
4313 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4314 p = last;
4315 while (--p >= first) {
4316 nr = le32_to_cpu(*p);
4317 if (!nr)
4318 continue; /* A hole */
4319
4320 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4321 nr, 1)) {
4322 EXT4_ERROR_INODE(inode,
4323 "invalid indirect mapped "
4324 "block %lu (level %d)",
4325 (unsigned long) nr, depth);
4326 break;
4327 }
4328
4329 /* Go read the buffer for the next level down */
4330 bh = sb_bread(inode->i_sb, nr);
4331
4332 /*
4333 * A read failure? Report error and clear slot
4334 * (should be rare).
4335 */
4336 if (!bh) {
4337 EXT4_ERROR_INODE_BLOCK(inode, nr,
4338 "Read failure");
4339 continue;
4340 }
4341
4342 /* This zaps the entire block. Bottom up. */
4343 BUFFER_TRACE(bh, "free child branches");
4344 ext4_free_branches(handle, inode, bh,
4345 (__le32 *) bh->b_data,
4346 (__le32 *) bh->b_data + addr_per_block,
4347 depth);
4348 brelse(bh);
4349
4350 /*
4351 * Everything below this this pointer has been
4352 * released. Now let this top-of-subtree go.
4353 *
4354 * We want the freeing of this indirect block to be
4355 * atomic in the journal with the updating of the
4356 * bitmap block which owns it. So make some room in
4357 * the journal.
4358 *
4359 * We zero the parent pointer *after* freeing its
4360 * pointee in the bitmaps, so if extend_transaction()
4361 * for some reason fails to put the bitmap changes and
4362 * the release into the same transaction, recovery
4363 * will merely complain about releasing a free block,
4364 * rather than leaking blocks.
4365 */
4366 if (ext4_handle_is_aborted(handle))
4367 return;
4368 if (try_to_extend_transaction(handle, inode)) {
4369 ext4_mark_inode_dirty(handle, inode);
4370 ext4_truncate_restart_trans(handle, inode,
4371 blocks_for_truncate(inode));
4372 }
4373
4374 /*
4375 * The forget flag here is critical because if
4376 * we are journaling (and not doing data
4377 * journaling), we have to make sure a revoke
4378 * record is written to prevent the journal
4379 * replay from overwriting the (former)
4380 * indirect block if it gets reallocated as a
4381 * data block. This must happen in the same
4382 * transaction where the data blocks are
4383 * actually freed.
4384 */
4385 ext4_free_blocks(handle, inode, NULL, nr, 1,
4386 EXT4_FREE_BLOCKS_METADATA|
4387 EXT4_FREE_BLOCKS_FORGET);
4388
4389 if (parent_bh) {
4390 /*
4391 * The block which we have just freed is
4392 * pointed to by an indirect block: journal it
4393 */
4394 BUFFER_TRACE(parent_bh, "get_write_access");
4395 if (!ext4_journal_get_write_access(handle,
4396 parent_bh)){
4397 *p = 0;
4398 BUFFER_TRACE(parent_bh,
4399 "call ext4_handle_dirty_metadata");
4400 ext4_handle_dirty_metadata(handle,
4401 inode,
4402 parent_bh);
4403 }
4404 }
4405 }
4406 } else {
4407 /* We have reached the bottom of the tree. */
4408 BUFFER_TRACE(parent_bh, "free data blocks");
4409 ext4_free_data(handle, inode, parent_bh, first, last);
4410 }
4411}
4412
4413int ext4_can_truncate(struct inode *inode) 3066int ext4_can_truncate(struct inode *inode)
4414{ 3067{
4415 if (S_ISREG(inode->i_mode)) 3068 if (S_ISREG(inode->i_mode))
@@ -4476,19 +3129,6 @@ int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4476 */ 3129 */
4477void ext4_truncate(struct inode *inode) 3130void ext4_truncate(struct inode *inode)
4478{ 3131{
4479 handle_t *handle;
4480 struct ext4_inode_info *ei = EXT4_I(inode);
4481 __le32 *i_data = ei->i_data;
4482 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4483 struct address_space *mapping = inode->i_mapping;
4484 ext4_lblk_t offsets[4];
4485 Indirect chain[4];
4486 Indirect *partial;
4487 __le32 nr = 0;
4488 int n = 0;
4489 ext4_lblk_t last_block, max_block;
4490 unsigned blocksize = inode->i_sb->s_blocksize;
4491
4492 trace_ext4_truncate_enter(inode); 3132 trace_ext4_truncate_enter(inode);
4493 3133
4494 if (!ext4_can_truncate(inode)) 3134 if (!ext4_can_truncate(inode))
@@ -4499,149 +3139,11 @@ void ext4_truncate(struct inode *inode)
4499 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) 3139 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4500 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); 3140 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4501 3141
4502 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { 3142 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4503 ext4_ext_truncate(inode); 3143 ext4_ext_truncate(inode);
4504 trace_ext4_truncate_exit(inode); 3144 else
4505 return; 3145 ext4_ind_truncate(inode);
4506 }
4507
4508 handle = start_transaction(inode);
4509 if (IS_ERR(handle))
4510 return; /* AKPM: return what? */
4511
4512 last_block = (inode->i_size + blocksize-1)
4513 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4514 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
4515 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4516
4517 if (inode->i_size & (blocksize - 1))
4518 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4519 goto out_stop;
4520
4521 if (last_block != max_block) {
4522 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4523 if (n == 0)
4524 goto out_stop; /* error */
4525 }
4526
4527 /*
4528 * OK. This truncate is going to happen. We add the inode to the
4529 * orphan list, so that if this truncate spans multiple transactions,
4530 * and we crash, we will resume the truncate when the filesystem
4531 * recovers. It also marks the inode dirty, to catch the new size.
4532 *
4533 * Implication: the file must always be in a sane, consistent
4534 * truncatable state while each transaction commits.
4535 */
4536 if (ext4_orphan_add(handle, inode))
4537 goto out_stop;
4538
4539 /*
4540 * From here we block out all ext4_get_block() callers who want to
4541 * modify the block allocation tree.
4542 */
4543 down_write(&ei->i_data_sem);
4544
4545 ext4_discard_preallocations(inode);
4546
4547 /*
4548 * The orphan list entry will now protect us from any crash which
4549 * occurs before the truncate completes, so it is now safe to propagate
4550 * the new, shorter inode size (held for now in i_size) into the
4551 * on-disk inode. We do this via i_disksize, which is the value which
4552 * ext4 *really* writes onto the disk inode.
4553 */
4554 ei->i_disksize = inode->i_size;
4555
4556 if (last_block == max_block) {
4557 /*
4558 * It is unnecessary to free any data blocks if last_block is
4559 * equal to the indirect block limit.
4560 */
4561 goto out_unlock;
4562 } else if (n == 1) { /* direct blocks */
4563 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4564 i_data + EXT4_NDIR_BLOCKS);
4565 goto do_indirects;
4566 }
4567
4568 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4569 /* Kill the top of shared branch (not detached) */
4570 if (nr) {
4571 if (partial == chain) {
4572 /* Shared branch grows from the inode */
4573 ext4_free_branches(handle, inode, NULL,
4574 &nr, &nr+1, (chain+n-1) - partial);
4575 *partial->p = 0;
4576 /*
4577 * We mark the inode dirty prior to restart,
4578 * and prior to stop. No need for it here.
4579 */
4580 } else {
4581 /* Shared branch grows from an indirect block */
4582 BUFFER_TRACE(partial->bh, "get_write_access");
4583 ext4_free_branches(handle, inode, partial->bh,
4584 partial->p,
4585 partial->p+1, (chain+n-1) - partial);
4586 }
4587 }
4588 /* Clear the ends of indirect blocks on the shared branch */
4589 while (partial > chain) {
4590 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4591 (__le32*)partial->bh->b_data+addr_per_block,
4592 (chain+n-1) - partial);
4593 BUFFER_TRACE(partial->bh, "call brelse");
4594 brelse(partial->bh);
4595 partial--;
4596 }
4597do_indirects:
4598 /* Kill the remaining (whole) subtrees */
4599 switch (offsets[0]) {
4600 default:
4601 nr = i_data[EXT4_IND_BLOCK];
4602 if (nr) {
4603 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4604 i_data[EXT4_IND_BLOCK] = 0;
4605 }
4606 case EXT4_IND_BLOCK:
4607 nr = i_data[EXT4_DIND_BLOCK];
4608 if (nr) {
4609 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4610 i_data[EXT4_DIND_BLOCK] = 0;
4611 }
4612 case EXT4_DIND_BLOCK:
4613 nr = i_data[EXT4_TIND_BLOCK];
4614 if (nr) {
4615 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4616 i_data[EXT4_TIND_BLOCK] = 0;
4617 }
4618 case EXT4_TIND_BLOCK:
4619 ;
4620 }
4621
4622out_unlock:
4623 up_write(&ei->i_data_sem);
4624 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4625 ext4_mark_inode_dirty(handle, inode);
4626
4627 /*
4628 * In a multi-transaction truncate, we only make the final transaction
4629 * synchronous
4630 */
4631 if (IS_SYNC(inode))
4632 ext4_handle_sync(handle);
4633out_stop:
4634 /*
4635 * If this was a simple ftruncate(), and the file will remain alive
4636 * then we need to clear up the orphan record which we created above.
4637 * However, if this was a real unlink then we were called by
4638 * ext4_delete_inode(), and we allow that function to clean up the
4639 * orphan info for us.
4640 */
4641 if (inode->i_nlink)
4642 ext4_orphan_del(handle, inode);
4643 3146
4644 ext4_journal_stop(handle);
4645 trace_ext4_truncate_exit(inode); 3147 trace_ext4_truncate_exit(inode);
4646} 3148}
4647 3149
@@ -5012,7 +3514,7 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
5012 (S_ISLNK(inode->i_mode) && 3514 (S_ISLNK(inode->i_mode) &&
5013 !ext4_inode_is_fast_symlink(inode))) { 3515 !ext4_inode_is_fast_symlink(inode))) {
5014 /* Validate block references which are part of inode */ 3516 /* Validate block references which are part of inode */
5015 ret = ext4_check_inode_blockref(inode); 3517 ret = ext4_ind_check_inode(inode);
5016 } 3518 }
5017 if (ret) 3519 if (ret)
5018 goto bad_inode; 3520 goto bad_inode;
@@ -5459,34 +3961,10 @@ int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5459 return 0; 3961 return 0;
5460} 3962}
5461 3963
5462static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5463 int chunk)
5464{
5465 int indirects;
5466
5467 /* if nrblocks are contiguous */
5468 if (chunk) {
5469 /*
5470 * With N contiguous data blocks, we need at most
5471 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
5472 * 2 dindirect blocks, and 1 tindirect block
5473 */
5474 return DIV_ROUND_UP(nrblocks,
5475 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
5476 }
5477 /*
5478 * if nrblocks are not contiguous, worse case, each block touch
5479 * a indirect block, and each indirect block touch a double indirect
5480 * block, plus a triple indirect block
5481 */
5482 indirects = nrblocks * 2 + 1;
5483 return indirects;
5484}
5485
5486static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) 3964static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5487{ 3965{
5488 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 3966 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5489 return ext4_indirect_trans_blocks(inode, nrblocks, chunk); 3967 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
5490 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); 3968 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5491} 3969}
5492 3970
generated by cgit v1.2.2 (git 2.25.0) at 2025-09-07 19:19:05 -0400