2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
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
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/workqueue.h>
41 #include <linux/kernel.h>
43 #include "ext4_jbd2.h"
46 #include "ext4_extents.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static inline int ext4_begin_ordered_truncate(struct inode *inode,
55 return jbd2_journal_begin_ordered_truncate(
56 EXT4_SB(inode->i_sb)->s_journal,
57 &EXT4_I(inode)->jinode,
61 static void ext4_invalidatepage(struct page *page, unsigned long offset);
64 * Test whether an inode is a fast symlink.
66 static int ext4_inode_is_fast_symlink(struct inode *inode)
68 int ea_blocks = EXT4_I(inode)->i_file_acl ?
69 (inode->i_sb->s_blocksize >> 9) : 0;
71 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
75 * Work out how many blocks we need to proceed with the next chunk of a
76 * truncate transaction.
78 static unsigned long blocks_for_truncate(struct inode *inode)
82 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
84 /* Give ourselves just enough room to cope with inodes in which
85 * i_blocks is corrupt: we've seen disk corruptions in the past
86 * which resulted in random data in an inode which looked enough
87 * like a regular file for ext4 to try to delete it. Things
88 * will go a bit crazy if that happens, but at least we should
89 * try not to panic the whole kernel. */
93 /* But we need to bound the transaction so we don't overflow the
95 if (needed > EXT4_MAX_TRANS_DATA)
96 needed = EXT4_MAX_TRANS_DATA;
98 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
102 * Truncate transactions can be complex and absolutely huge. So we need to
103 * be able to restart the transaction at a conventient checkpoint to make
104 * sure we don't overflow the journal.
106 * start_transaction gets us a new handle for a truncate transaction,
107 * and extend_transaction tries to extend the existing one a bit. If
108 * extend fails, we need to propagate the failure up and restart the
109 * transaction in the top-level truncate loop. --sct
111 static handle_t *start_transaction(struct inode *inode)
115 result = ext4_journal_start(inode, blocks_for_truncate(inode));
119 ext4_std_error(inode->i_sb, PTR_ERR(result));
124 * Try to extend this transaction for the purposes of truncation.
126 * Returns 0 if we managed to create more room. If we can't create more
127 * room, and the transaction must be restarted we return 1.
129 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
131 if (!ext4_handle_valid(handle))
133 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
135 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
141 * Restart the transaction associated with *handle. This does a commit,
142 * so before we call here everything must be consistently dirtied against
145 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
151 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
152 * moment, get_block can be called only for blocks inside i_size since
153 * page cache has been already dropped and writes are blocked by
154 * i_mutex. So we can safely drop the i_data_sem here.
156 BUG_ON(EXT4_JOURNAL(inode) == NULL);
157 jbd_debug(2, "restarting handle %p\n", handle);
158 up_write(&EXT4_I(inode)->i_data_sem);
159 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
160 down_write(&EXT4_I(inode)->i_data_sem);
161 ext4_discard_preallocations(inode);
167 * Called at the last iput() if i_nlink is zero.
169 void ext4_delete_inode(struct inode *inode)
174 if (ext4_should_order_data(inode))
175 ext4_begin_ordered_truncate(inode, 0);
176 truncate_inode_pages(&inode->i_data, 0);
178 if (is_bad_inode(inode))
181 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
182 if (IS_ERR(handle)) {
183 ext4_std_error(inode->i_sb, PTR_ERR(handle));
185 * If we're going to skip the normal cleanup, we still need to
186 * make sure that the in-core orphan linked list is properly
189 ext4_orphan_del(NULL, inode);
194 ext4_handle_sync(handle);
196 err = ext4_mark_inode_dirty(handle, inode);
198 ext4_warning(inode->i_sb,
199 "couldn't mark inode dirty (err %d)", err);
203 ext4_truncate(inode);
206 * ext4_ext_truncate() doesn't reserve any slop when it
207 * restarts journal transactions; therefore there may not be
208 * enough credits left in the handle to remove the inode from
209 * the orphan list and set the dtime field.
211 if (!ext4_handle_has_enough_credits(handle, 3)) {
212 err = ext4_journal_extend(handle, 3);
214 err = ext4_journal_restart(handle, 3);
216 ext4_warning(inode->i_sb,
217 "couldn't extend journal (err %d)", err);
219 ext4_journal_stop(handle);
225 * Kill off the orphan record which ext4_truncate created.
226 * AKPM: I think this can be inside the above `if'.
227 * Note that ext4_orphan_del() has to be able to cope with the
228 * deletion of a non-existent orphan - this is because we don't
229 * know if ext4_truncate() actually created an orphan record.
230 * (Well, we could do this if we need to, but heck - it works)
232 ext4_orphan_del(handle, inode);
233 EXT4_I(inode)->i_dtime = get_seconds();
236 * One subtle ordering requirement: if anything has gone wrong
237 * (transaction abort, IO errors, whatever), then we can still
238 * do these next steps (the fs will already have been marked as
239 * having errors), but we can't free the inode if the mark_dirty
242 if (ext4_mark_inode_dirty(handle, inode))
243 /* If that failed, just do the required in-core inode clear. */
246 ext4_free_inode(handle, inode);
247 ext4_journal_stop(handle);
250 clear_inode(inode); /* We must guarantee clearing of inode... */
256 struct buffer_head *bh;
259 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
261 p->key = *(p->p = v);
266 * ext4_block_to_path - parse the block number into array of offsets
267 * @inode: inode in question (we are only interested in its superblock)
268 * @i_block: block number to be parsed
269 * @offsets: array to store the offsets in
270 * @boundary: set this non-zero if the referred-to block is likely to be
271 * followed (on disk) by an indirect block.
273 * To store the locations of file's data ext4 uses a data structure common
274 * for UNIX filesystems - tree of pointers anchored in the inode, with
275 * data blocks at leaves and indirect blocks in intermediate nodes.
276 * This function translates the block number into path in that tree -
277 * return value is the path length and @offsets[n] is the offset of
278 * pointer to (n+1)th node in the nth one. If @block is out of range
279 * (negative or too large) warning is printed and zero returned.
281 * Note: function doesn't find node addresses, so no IO is needed. All
282 * we need to know is the capacity of indirect blocks (taken from the
287 * Portability note: the last comparison (check that we fit into triple
288 * indirect block) is spelled differently, because otherwise on an
289 * architecture with 32-bit longs and 8Kb pages we might get into trouble
290 * if our filesystem had 8Kb blocks. We might use long long, but that would
291 * kill us on x86. Oh, well, at least the sign propagation does not matter -
292 * i_block would have to be negative in the very beginning, so we would not
296 static int ext4_block_to_path(struct inode *inode,
298 ext4_lblk_t offsets[4], int *boundary)
300 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
301 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
302 const long direct_blocks = EXT4_NDIR_BLOCKS,
303 indirect_blocks = ptrs,
304 double_blocks = (1 << (ptrs_bits * 2));
308 if (i_block < direct_blocks) {
309 offsets[n++] = i_block;
310 final = direct_blocks;
311 } else if ((i_block -= direct_blocks) < indirect_blocks) {
312 offsets[n++] = EXT4_IND_BLOCK;
313 offsets[n++] = i_block;
315 } else if ((i_block -= indirect_blocks) < double_blocks) {
316 offsets[n++] = EXT4_DIND_BLOCK;
317 offsets[n++] = i_block >> ptrs_bits;
318 offsets[n++] = i_block & (ptrs - 1);
320 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
321 offsets[n++] = EXT4_TIND_BLOCK;
322 offsets[n++] = i_block >> (ptrs_bits * 2);
323 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
324 offsets[n++] = i_block & (ptrs - 1);
327 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
328 i_block + direct_blocks +
329 indirect_blocks + double_blocks, inode->i_ino);
332 *boundary = final - 1 - (i_block & (ptrs - 1));
336 static int __ext4_check_blockref(const char *function, struct inode *inode,
337 __le32 *p, unsigned int max)
342 while (bref < p+max) {
343 blk = le32_to_cpu(*bref++);
345 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
347 __ext4_error(inode->i_sb, function,
348 "invalid block reference %u "
349 "in inode #%lu", blk, inode->i_ino);
357 #define ext4_check_indirect_blockref(inode, bh) \
358 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
359 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
361 #define ext4_check_inode_blockref(inode) \
362 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
366 * ext4_get_branch - read the chain of indirect blocks leading to data
367 * @inode: inode in question
368 * @depth: depth of the chain (1 - direct pointer, etc.)
369 * @offsets: offsets of pointers in inode/indirect blocks
370 * @chain: place to store the result
371 * @err: here we store the error value
373 * Function fills the array of triples <key, p, bh> and returns %NULL
374 * if everything went OK or the pointer to the last filled triple
375 * (incomplete one) otherwise. Upon the return chain[i].key contains
376 * the number of (i+1)-th block in the chain (as it is stored in memory,
377 * i.e. little-endian 32-bit), chain[i].p contains the address of that
378 * number (it points into struct inode for i==0 and into the bh->b_data
379 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
380 * block for i>0 and NULL for i==0. In other words, it holds the block
381 * numbers of the chain, addresses they were taken from (and where we can
382 * verify that chain did not change) and buffer_heads hosting these
385 * Function stops when it stumbles upon zero pointer (absent block)
386 * (pointer to last triple returned, *@err == 0)
387 * or when it gets an IO error reading an indirect block
388 * (ditto, *@err == -EIO)
389 * or when it reads all @depth-1 indirect blocks successfully and finds
390 * the whole chain, all way to the data (returns %NULL, *err == 0).
392 * Need to be called with
393 * down_read(&EXT4_I(inode)->i_data_sem)
395 static Indirect *ext4_get_branch(struct inode *inode, int depth,
396 ext4_lblk_t *offsets,
397 Indirect chain[4], int *err)
399 struct super_block *sb = inode->i_sb;
401 struct buffer_head *bh;
404 /* i_data is not going away, no lock needed */
405 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
409 bh = sb_getblk(sb, le32_to_cpu(p->key));
413 if (!bh_uptodate_or_lock(bh)) {
414 if (bh_submit_read(bh) < 0) {
418 /* validate block references */
419 if (ext4_check_indirect_blockref(inode, bh)) {
425 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
439 * ext4_find_near - find a place for allocation with sufficient locality
441 * @ind: descriptor of indirect block.
443 * This function returns the preferred place for block allocation.
444 * It is used when heuristic for sequential allocation fails.
446 * + if there is a block to the left of our position - allocate near it.
447 * + if pointer will live in indirect block - allocate near that block.
448 * + if pointer will live in inode - allocate in the same
451 * In the latter case we colour the starting block by the callers PID to
452 * prevent it from clashing with concurrent allocations for a different inode
453 * in the same block group. The PID is used here so that functionally related
454 * files will be close-by on-disk.
456 * Caller must make sure that @ind is valid and will stay that way.
458 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
460 struct ext4_inode_info *ei = EXT4_I(inode);
461 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
463 ext4_fsblk_t bg_start;
464 ext4_fsblk_t last_block;
465 ext4_grpblk_t colour;
466 ext4_group_t block_group;
467 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
469 /* Try to find previous block */
470 for (p = ind->p - 1; p >= start; p--) {
472 return le32_to_cpu(*p);
475 /* No such thing, so let's try location of indirect block */
477 return ind->bh->b_blocknr;
480 * It is going to be referred to from the inode itself? OK, just put it
481 * into the same cylinder group then.
483 block_group = ei->i_block_group;
484 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
485 block_group &= ~(flex_size-1);
486 if (S_ISREG(inode->i_mode))
489 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
490 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
493 * If we are doing delayed allocation, we don't need take
494 * colour into account.
496 if (test_opt(inode->i_sb, DELALLOC))
499 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
500 colour = (current->pid % 16) *
501 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
503 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
504 return bg_start + colour;
508 * ext4_find_goal - find a preferred place for allocation.
510 * @block: block we want
511 * @partial: pointer to the last triple within a chain
513 * Normally this function find the preferred place for block allocation,
515 * Because this is only used for non-extent files, we limit the block nr
518 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
524 * XXX need to get goal block from mballoc's data structures
527 goal = ext4_find_near(inode, partial);
528 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
533 * ext4_blks_to_allocate: Look up the block map and count the number
534 * of direct blocks need to be allocated for the given branch.
536 * @branch: chain of indirect blocks
537 * @k: number of blocks need for indirect blocks
538 * @blks: number of data blocks to be mapped.
539 * @blocks_to_boundary: the offset in the indirect block
541 * return the total number of blocks to be allocate, including the
542 * direct and indirect blocks.
544 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
545 int blocks_to_boundary)
547 unsigned int count = 0;
550 * Simple case, [t,d]Indirect block(s) has not allocated yet
551 * then it's clear blocks on that path have not allocated
554 /* right now we don't handle cross boundary allocation */
555 if (blks < blocks_to_boundary + 1)
558 count += blocks_to_boundary + 1;
563 while (count < blks && count <= blocks_to_boundary &&
564 le32_to_cpu(*(branch[0].p + count)) == 0) {
571 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
572 * @indirect_blks: the number of blocks need to allocate for indirect
575 * @new_blocks: on return it will store the new block numbers for
576 * the indirect blocks(if needed) and the first direct block,
577 * @blks: on return it will store the total number of allocated
580 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
581 ext4_lblk_t iblock, ext4_fsblk_t goal,
582 int indirect_blks, int blks,
583 ext4_fsblk_t new_blocks[4], int *err)
585 struct ext4_allocation_request ar;
587 unsigned long count = 0, blk_allocated = 0;
589 ext4_fsblk_t current_block = 0;
593 * Here we try to allocate the requested multiple blocks at once,
594 * on a best-effort basis.
595 * To build a branch, we should allocate blocks for
596 * the indirect blocks(if not allocated yet), and at least
597 * the first direct block of this branch. That's the
598 * minimum number of blocks need to allocate(required)
600 /* first we try to allocate the indirect blocks */
601 target = indirect_blks;
604 /* allocating blocks for indirect blocks and direct blocks */
605 current_block = ext4_new_meta_blocks(handle, inode,
610 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
611 EXT4_ERROR_INODE(inode,
612 "current_block %llu + count %lu > %d!",
613 current_block, count,
614 EXT4_MAX_BLOCK_FILE_PHYS);
620 /* allocate blocks for indirect blocks */
621 while (index < indirect_blks && count) {
622 new_blocks[index++] = current_block++;
627 * save the new block number
628 * for the first direct block
630 new_blocks[index] = current_block;
631 printk(KERN_INFO "%s returned more blocks than "
632 "requested\n", __func__);
638 target = blks - count ;
639 blk_allocated = count;
642 /* Now allocate data blocks */
643 memset(&ar, 0, sizeof(ar));
648 if (S_ISREG(inode->i_mode))
649 /* enable in-core preallocation only for regular files */
650 ar.flags = EXT4_MB_HINT_DATA;
652 current_block = ext4_mb_new_blocks(handle, &ar, err);
653 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
654 EXT4_ERROR_INODE(inode,
655 "current_block %llu + ar.len %d > %d!",
656 current_block, ar.len,
657 EXT4_MAX_BLOCK_FILE_PHYS);
662 if (*err && (target == blks)) {
664 * if the allocation failed and we didn't allocate
670 if (target == blks) {
672 * save the new block number
673 * for the first direct block
675 new_blocks[index] = current_block;
677 blk_allocated += ar.len;
680 /* total number of blocks allocated for direct blocks */
685 for (i = 0; i < index; i++)
686 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
691 * ext4_alloc_branch - allocate and set up a chain of blocks.
693 * @indirect_blks: number of allocated indirect blocks
694 * @blks: number of allocated direct blocks
695 * @offsets: offsets (in the blocks) to store the pointers to next.
696 * @branch: place to store the chain in.
698 * This function allocates blocks, zeroes out all but the last one,
699 * links them into chain and (if we are synchronous) writes them to disk.
700 * In other words, it prepares a branch that can be spliced onto the
701 * inode. It stores the information about that chain in the branch[], in
702 * the same format as ext4_get_branch() would do. We are calling it after
703 * we had read the existing part of chain and partial points to the last
704 * triple of that (one with zero ->key). Upon the exit we have the same
705 * picture as after the successful ext4_get_block(), except that in one
706 * place chain is disconnected - *branch->p is still zero (we did not
707 * set the last link), but branch->key contains the number that should
708 * be placed into *branch->p to fill that gap.
710 * If allocation fails we free all blocks we've allocated (and forget
711 * their buffer_heads) and return the error value the from failed
712 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
713 * as described above and return 0.
715 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
716 ext4_lblk_t iblock, int indirect_blks,
717 int *blks, ext4_fsblk_t goal,
718 ext4_lblk_t *offsets, Indirect *branch)
720 int blocksize = inode->i_sb->s_blocksize;
723 struct buffer_head *bh;
725 ext4_fsblk_t new_blocks[4];
726 ext4_fsblk_t current_block;
728 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
729 *blks, new_blocks, &err);
733 branch[0].key = cpu_to_le32(new_blocks[0]);
735 * metadata blocks and data blocks are allocated.
737 for (n = 1; n <= indirect_blks; n++) {
739 * Get buffer_head for parent block, zero it out
740 * and set the pointer to new one, then send
743 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
746 BUFFER_TRACE(bh, "call get_create_access");
747 err = ext4_journal_get_create_access(handle, bh);
749 /* Don't brelse(bh) here; it's done in
750 * ext4_journal_forget() below */
755 memset(bh->b_data, 0, blocksize);
756 branch[n].p = (__le32 *) bh->b_data + offsets[n];
757 branch[n].key = cpu_to_le32(new_blocks[n]);
758 *branch[n].p = branch[n].key;
759 if (n == indirect_blks) {
760 current_block = new_blocks[n];
762 * End of chain, update the last new metablock of
763 * the chain to point to the new allocated
764 * data blocks numbers
766 for (i = 1; i < num; i++)
767 *(branch[n].p + i) = cpu_to_le32(++current_block);
769 BUFFER_TRACE(bh, "marking uptodate");
770 set_buffer_uptodate(bh);
773 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
774 err = ext4_handle_dirty_metadata(handle, inode, bh);
781 /* Allocation failed, free what we already allocated */
782 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
783 for (i = 1; i <= n ; i++) {
785 * branch[i].bh is newly allocated, so there is no
786 * need to revoke the block, which is why we don't
787 * need to set EXT4_FREE_BLOCKS_METADATA.
789 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
790 EXT4_FREE_BLOCKS_FORGET);
792 for (i = n+1; i < indirect_blks; i++)
793 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
795 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
801 * ext4_splice_branch - splice the allocated branch onto inode.
803 * @block: (logical) number of block we are adding
804 * @chain: chain of indirect blocks (with a missing link - see
806 * @where: location of missing link
807 * @num: number of indirect blocks we are adding
808 * @blks: number of direct blocks we are adding
810 * This function fills the missing link and does all housekeeping needed in
811 * inode (->i_blocks, etc.). In case of success we end up with the full
812 * chain to new block and return 0.
814 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
815 ext4_lblk_t block, Indirect *where, int num,
820 ext4_fsblk_t current_block;
823 * If we're splicing into a [td]indirect block (as opposed to the
824 * inode) then we need to get write access to the [td]indirect block
828 BUFFER_TRACE(where->bh, "get_write_access");
829 err = ext4_journal_get_write_access(handle, where->bh);
835 *where->p = where->key;
838 * Update the host buffer_head or inode to point to more just allocated
839 * direct blocks blocks
841 if (num == 0 && blks > 1) {
842 current_block = le32_to_cpu(where->key) + 1;
843 for (i = 1; i < blks; i++)
844 *(where->p + i) = cpu_to_le32(current_block++);
847 /* We are done with atomic stuff, now do the rest of housekeeping */
848 /* had we spliced it onto indirect block? */
851 * If we spliced it onto an indirect block, we haven't
852 * altered the inode. Note however that if it is being spliced
853 * onto an indirect block at the very end of the file (the
854 * file is growing) then we *will* alter the inode to reflect
855 * the new i_size. But that is not done here - it is done in
856 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
858 jbd_debug(5, "splicing indirect only\n");
859 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
860 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
865 * OK, we spliced it into the inode itself on a direct block.
867 ext4_mark_inode_dirty(handle, inode);
868 jbd_debug(5, "splicing direct\n");
873 for (i = 1; i <= num; i++) {
875 * branch[i].bh is newly allocated, so there is no
876 * need to revoke the block, which is why we don't
877 * need to set EXT4_FREE_BLOCKS_METADATA.
879 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
880 EXT4_FREE_BLOCKS_FORGET);
882 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
889 * The ext4_ind_get_blocks() function handles non-extents inodes
890 * (i.e., using the traditional indirect/double-indirect i_blocks
891 * scheme) for ext4_get_blocks().
893 * Allocation strategy is simple: if we have to allocate something, we will
894 * have to go the whole way to leaf. So let's do it before attaching anything
895 * to tree, set linkage between the newborn blocks, write them if sync is
896 * required, recheck the path, free and repeat if check fails, otherwise
897 * set the last missing link (that will protect us from any truncate-generated
898 * removals - all blocks on the path are immune now) and possibly force the
899 * write on the parent block.
900 * That has a nice additional property: no special recovery from the failed
901 * allocations is needed - we simply release blocks and do not touch anything
902 * reachable from inode.
904 * `handle' can be NULL if create == 0.
906 * return > 0, # of blocks mapped or allocated.
907 * return = 0, if plain lookup failed.
908 * return < 0, error case.
910 * The ext4_ind_get_blocks() function should be called with
911 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
912 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
913 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
916 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
917 ext4_lblk_t iblock, unsigned int maxblocks,
918 struct buffer_head *bh_result,
922 ext4_lblk_t offsets[4];
927 int blocks_to_boundary = 0;
930 ext4_fsblk_t first_block = 0;
932 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
933 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
934 depth = ext4_block_to_path(inode, iblock, offsets,
935 &blocks_to_boundary);
940 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
942 /* Simplest case - block found, no allocation needed */
944 first_block = le32_to_cpu(chain[depth - 1].key);
945 clear_buffer_new(bh_result);
948 while (count < maxblocks && count <= blocks_to_boundary) {
951 blk = le32_to_cpu(*(chain[depth-1].p + count));
953 if (blk == first_block + count)
961 /* Next simple case - plain lookup or failed read of indirect block */
962 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
966 * Okay, we need to do block allocation.
968 goal = ext4_find_goal(inode, iblock, partial);
970 /* the number of blocks need to allocate for [d,t]indirect blocks */
971 indirect_blks = (chain + depth) - partial - 1;
974 * Next look up the indirect map to count the totoal number of
975 * direct blocks to allocate for this branch.
977 count = ext4_blks_to_allocate(partial, indirect_blks,
978 maxblocks, blocks_to_boundary);
980 * Block out ext4_truncate while we alter the tree
982 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
984 offsets + (partial - chain), partial);
987 * The ext4_splice_branch call will free and forget any buffers
988 * on the new chain if there is a failure, but that risks using
989 * up transaction credits, especially for bitmaps where the
990 * credits cannot be returned. Can we handle this somehow? We
991 * may need to return -EAGAIN upwards in the worst case. --sct
994 err = ext4_splice_branch(handle, inode, iblock,
995 partial, indirect_blks, count);
999 set_buffer_new(bh_result);
1001 ext4_update_inode_fsync_trans(handle, inode, 1);
1003 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1004 if (count > blocks_to_boundary)
1005 set_buffer_boundary(bh_result);
1007 /* Clean up and exit */
1008 partial = chain + depth - 1; /* the whole chain */
1010 while (partial > chain) {
1011 BUFFER_TRACE(partial->bh, "call brelse");
1012 brelse(partial->bh);
1015 BUFFER_TRACE(bh_result, "returned");
1021 qsize_t *ext4_get_reserved_space(struct inode *inode)
1023 return &EXT4_I(inode)->i_reserved_quota;
1028 * Calculate the number of metadata blocks need to reserve
1029 * to allocate a new block at @lblocks for non extent file based file
1031 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1034 struct ext4_inode_info *ei = EXT4_I(inode);
1035 int dind_mask = EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1;
1038 if (lblock < EXT4_NDIR_BLOCKS)
1041 lblock -= EXT4_NDIR_BLOCKS;
1043 if (ei->i_da_metadata_calc_len &&
1044 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1045 ei->i_da_metadata_calc_len++;
1048 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1049 ei->i_da_metadata_calc_len = 1;
1050 blk_bits = roundup_pow_of_two(lblock + 1);
1051 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1055 * Calculate the number of metadata blocks need to reserve
1056 * to allocate a block located at @lblock
1058 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1060 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1061 return ext4_ext_calc_metadata_amount(inode, lblock);
1063 return ext4_indirect_calc_metadata_amount(inode, lblock);
1067 * Called with i_data_sem down, which is important since we can call
1068 * ext4_discard_preallocations() from here.
1070 void ext4_da_update_reserve_space(struct inode *inode,
1071 int used, int quota_claim)
1073 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1074 struct ext4_inode_info *ei = EXT4_I(inode);
1075 int mdb_free = 0, allocated_meta_blocks = 0;
1077 spin_lock(&ei->i_block_reservation_lock);
1078 trace_ext4_da_update_reserve_space(inode, used);
1079 if (unlikely(used > ei->i_reserved_data_blocks)) {
1080 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1081 "with only %d reserved data blocks\n",
1082 __func__, inode->i_ino, used,
1083 ei->i_reserved_data_blocks);
1085 used = ei->i_reserved_data_blocks;
1088 /* Update per-inode reservations */
1089 ei->i_reserved_data_blocks -= used;
1090 used += ei->i_allocated_meta_blocks;
1091 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1092 allocated_meta_blocks = ei->i_allocated_meta_blocks;
1093 ei->i_allocated_meta_blocks = 0;
1094 percpu_counter_sub(&sbi->s_dirtyblocks_counter, used);
1096 if (ei->i_reserved_data_blocks == 0) {
1098 * We can release all of the reserved metadata blocks
1099 * only when we have written all of the delayed
1100 * allocation blocks.
1102 mdb_free = ei->i_reserved_meta_blocks;
1103 ei->i_reserved_meta_blocks = 0;
1104 ei->i_da_metadata_calc_len = 0;
1105 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1107 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1109 /* Update quota subsystem */
1111 vfs_dq_claim_block(inode, used);
1113 vfs_dq_release_reservation_block(inode, mdb_free);
1116 * We did fallocate with an offset that is already delayed
1117 * allocated. So on delayed allocated writeback we should
1118 * not update the quota for allocated blocks. But then
1119 * converting an fallocate region to initialized region would
1120 * have caused a metadata allocation. So claim quota for
1123 if (allocated_meta_blocks)
1124 vfs_dq_claim_block(inode, allocated_meta_blocks);
1125 vfs_dq_release_reservation_block(inode, mdb_free + used);
1129 * If we have done all the pending block allocations and if
1130 * there aren't any writers on the inode, we can discard the
1131 * inode's preallocations.
1133 if ((ei->i_reserved_data_blocks == 0) &&
1134 (atomic_read(&inode->i_writecount) == 0))
1135 ext4_discard_preallocations(inode);
1138 static int check_block_validity(struct inode *inode, const char *msg,
1139 sector_t logical, sector_t phys, int len)
1141 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
1142 __ext4_error(inode->i_sb, msg,
1143 "inode #%lu logical block %llu mapped to %llu "
1144 "(size %d)", inode->i_ino,
1145 (unsigned long long) logical,
1146 (unsigned long long) phys, len);
1153 * Return the number of contiguous dirty pages in a given inode
1154 * starting at page frame idx.
1156 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1157 unsigned int max_pages)
1159 struct address_space *mapping = inode->i_mapping;
1161 struct pagevec pvec;
1163 int i, nr_pages, done = 0;
1167 pagevec_init(&pvec, 0);
1170 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1171 PAGECACHE_TAG_DIRTY,
1172 (pgoff_t)PAGEVEC_SIZE);
1175 for (i = 0; i < nr_pages; i++) {
1176 struct page *page = pvec.pages[i];
1177 struct buffer_head *bh, *head;
1180 if (unlikely(page->mapping != mapping) ||
1182 PageWriteback(page) ||
1183 page->index != idx) {
1188 if (page_has_buffers(page)) {
1189 bh = head = page_buffers(page);
1191 if (!buffer_delay(bh) &&
1192 !buffer_unwritten(bh))
1194 bh = bh->b_this_page;
1195 } while (!done && (bh != head));
1202 if (num >= max_pages)
1205 pagevec_release(&pvec);
1211 * The ext4_get_blocks() function tries to look up the requested blocks,
1212 * and returns if the blocks are already mapped.
1214 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1215 * and store the allocated blocks in the result buffer head and mark it
1218 * If file type is extents based, it will call ext4_ext_get_blocks(),
1219 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1222 * On success, it returns the number of blocks being mapped or allocate.
1223 * if create==0 and the blocks are pre-allocated and uninitialized block,
1224 * the result buffer head is unmapped. If the create ==1, it will make sure
1225 * the buffer head is mapped.
1227 * It returns 0 if plain look up failed (blocks have not been allocated), in
1228 * that casem, buffer head is unmapped
1230 * It returns the error in case of allocation failure.
1232 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1233 unsigned int max_blocks, struct buffer_head *bh,
1238 clear_buffer_mapped(bh);
1239 clear_buffer_unwritten(bh);
1241 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1242 "logical block %lu\n", inode->i_ino, flags, max_blocks,
1243 (unsigned long)block);
1245 * Try to see if we can get the block without requesting a new
1246 * file system block.
1248 down_read((&EXT4_I(inode)->i_data_sem));
1249 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1250 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1253 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1256 up_read((&EXT4_I(inode)->i_data_sem));
1258 if (retval > 0 && buffer_mapped(bh)) {
1259 int ret = check_block_validity(inode, "file system corruption",
1260 block, bh->b_blocknr, retval);
1265 /* If it is only a block(s) look up */
1266 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1270 * Returns if the blocks have already allocated
1272 * Note that if blocks have been preallocated
1273 * ext4_ext_get_block() returns th create = 0
1274 * with buffer head unmapped.
1276 if (retval > 0 && buffer_mapped(bh))
1280 * When we call get_blocks without the create flag, the
1281 * BH_Unwritten flag could have gotten set if the blocks
1282 * requested were part of a uninitialized extent. We need to
1283 * clear this flag now that we are committed to convert all or
1284 * part of the uninitialized extent to be an initialized
1285 * extent. This is because we need to avoid the combination
1286 * of BH_Unwritten and BH_Mapped flags being simultaneously
1287 * set on the buffer_head.
1289 clear_buffer_unwritten(bh);
1292 * New blocks allocate and/or writing to uninitialized extent
1293 * will possibly result in updating i_data, so we take
1294 * the write lock of i_data_sem, and call get_blocks()
1295 * with create == 1 flag.
1297 down_write((&EXT4_I(inode)->i_data_sem));
1300 * if the caller is from delayed allocation writeout path
1301 * we have already reserved fs blocks for allocation
1302 * let the underlying get_block() function know to
1303 * avoid double accounting
1305 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1306 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1308 * We need to check for EXT4 here because migrate
1309 * could have changed the inode type in between
1311 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1312 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1315 retval = ext4_ind_get_blocks(handle, inode, block,
1316 max_blocks, bh, flags);
1318 if (retval > 0 && buffer_new(bh)) {
1320 * We allocated new blocks which will result in
1321 * i_data's format changing. Force the migrate
1322 * to fail by clearing migrate flags
1324 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1328 * Update reserved blocks/metadata blocks after successful
1329 * block allocation which had been deferred till now. We don't
1330 * support fallocate for non extent files. So we can update
1331 * reserve space here.
1334 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1335 ext4_da_update_reserve_space(inode, retval, 1);
1337 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1338 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1340 up_write((&EXT4_I(inode)->i_data_sem));
1341 if (retval > 0 && buffer_mapped(bh)) {
1342 int ret = check_block_validity(inode, "file system "
1343 "corruption after allocation",
1344 block, bh->b_blocknr, retval);
1351 /* Maximum number of blocks we map for direct IO at once. */
1352 #define DIO_MAX_BLOCKS 4096
1354 int ext4_get_block(struct inode *inode, sector_t iblock,
1355 struct buffer_head *bh_result, int create)
1357 handle_t *handle = ext4_journal_current_handle();
1358 int ret = 0, started = 0;
1359 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1362 if (create && !handle) {
1363 /* Direct IO write... */
1364 if (max_blocks > DIO_MAX_BLOCKS)
1365 max_blocks = DIO_MAX_BLOCKS;
1366 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1367 handle = ext4_journal_start(inode, dio_credits);
1368 if (IS_ERR(handle)) {
1369 ret = PTR_ERR(handle);
1375 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1376 create ? EXT4_GET_BLOCKS_CREATE : 0);
1378 bh_result->b_size = (ret << inode->i_blkbits);
1382 ext4_journal_stop(handle);
1388 * `handle' can be NULL if create is zero
1390 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1391 ext4_lblk_t block, int create, int *errp)
1393 struct buffer_head dummy;
1397 J_ASSERT(handle != NULL || create == 0);
1400 dummy.b_blocknr = -1000;
1401 buffer_trace_init(&dummy.b_history);
1403 flags |= EXT4_GET_BLOCKS_CREATE;
1404 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1406 * ext4_get_blocks() returns number of blocks mapped. 0 in
1415 if (!err && buffer_mapped(&dummy)) {
1416 struct buffer_head *bh;
1417 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1422 if (buffer_new(&dummy)) {
1423 J_ASSERT(create != 0);
1424 J_ASSERT(handle != NULL);
1427 * Now that we do not always journal data, we should
1428 * keep in mind whether this should always journal the
1429 * new buffer as metadata. For now, regular file
1430 * writes use ext4_get_block instead, so it's not a
1434 BUFFER_TRACE(bh, "call get_create_access");
1435 fatal = ext4_journal_get_create_access(handle, bh);
1436 if (!fatal && !buffer_uptodate(bh)) {
1437 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1438 set_buffer_uptodate(bh);
1441 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1442 err = ext4_handle_dirty_metadata(handle, inode, bh);
1446 BUFFER_TRACE(bh, "not a new buffer");
1459 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1460 ext4_lblk_t block, int create, int *err)
1462 struct buffer_head *bh;
1464 bh = ext4_getblk(handle, inode, block, create, err);
1467 if (buffer_uptodate(bh))
1469 ll_rw_block(READ_META, 1, &bh);
1471 if (buffer_uptodate(bh))
1478 static int walk_page_buffers(handle_t *handle,
1479 struct buffer_head *head,
1483 int (*fn)(handle_t *handle,
1484 struct buffer_head *bh))
1486 struct buffer_head *bh;
1487 unsigned block_start, block_end;
1488 unsigned blocksize = head->b_size;
1490 struct buffer_head *next;
1492 for (bh = head, block_start = 0;
1493 ret == 0 && (bh != head || !block_start);
1494 block_start = block_end, bh = next) {
1495 next = bh->b_this_page;
1496 block_end = block_start + blocksize;
1497 if (block_end <= from || block_start >= to) {
1498 if (partial && !buffer_uptodate(bh))
1502 err = (*fn)(handle, bh);
1510 * To preserve ordering, it is essential that the hole instantiation and
1511 * the data write be encapsulated in a single transaction. We cannot
1512 * close off a transaction and start a new one between the ext4_get_block()
1513 * and the commit_write(). So doing the jbd2_journal_start at the start of
1514 * prepare_write() is the right place.
1516 * Also, this function can nest inside ext4_writepage() ->
1517 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1518 * has generated enough buffer credits to do the whole page. So we won't
1519 * block on the journal in that case, which is good, because the caller may
1522 * By accident, ext4 can be reentered when a transaction is open via
1523 * quota file writes. If we were to commit the transaction while thus
1524 * reentered, there can be a deadlock - we would be holding a quota
1525 * lock, and the commit would never complete if another thread had a
1526 * transaction open and was blocking on the quota lock - a ranking
1529 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1530 * will _not_ run commit under these circumstances because handle->h_ref
1531 * is elevated. We'll still have enough credits for the tiny quotafile
1534 static int do_journal_get_write_access(handle_t *handle,
1535 struct buffer_head *bh)
1537 if (!buffer_mapped(bh) || buffer_freed(bh))
1539 return ext4_journal_get_write_access(handle, bh);
1543 * Truncate blocks that were not used by write. We have to truncate the
1544 * pagecache as well so that corresponding buffers get properly unmapped.
1546 static void ext4_truncate_failed_write(struct inode *inode)
1548 truncate_inode_pages(inode->i_mapping, inode->i_size);
1549 ext4_truncate(inode);
1552 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1553 struct buffer_head *bh_result, int create);
1554 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1555 loff_t pos, unsigned len, unsigned flags,
1556 struct page **pagep, void **fsdata)
1558 struct inode *inode = mapping->host;
1559 int ret, needed_blocks;
1566 trace_ext4_write_begin(inode, pos, len, flags);
1568 * Reserve one block more for addition to orphan list in case
1569 * we allocate blocks but write fails for some reason
1571 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1572 index = pos >> PAGE_CACHE_SHIFT;
1573 from = pos & (PAGE_CACHE_SIZE - 1);
1577 handle = ext4_journal_start(inode, needed_blocks);
1578 if (IS_ERR(handle)) {
1579 ret = PTR_ERR(handle);
1583 /* We cannot recurse into the filesystem as the transaction is already
1585 flags |= AOP_FLAG_NOFS;
1587 page = grab_cache_page_write_begin(mapping, index, flags);
1589 ext4_journal_stop(handle);
1595 if (ext4_should_dioread_nolock(inode))
1596 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1597 fsdata, ext4_get_block_write);
1599 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1600 fsdata, ext4_get_block);
1602 if (!ret && ext4_should_journal_data(inode)) {
1603 ret = walk_page_buffers(handle, page_buffers(page),
1604 from, to, NULL, do_journal_get_write_access);
1609 page_cache_release(page);
1611 * block_write_begin may have instantiated a few blocks
1612 * outside i_size. Trim these off again. Don't need
1613 * i_size_read because we hold i_mutex.
1615 * Add inode to orphan list in case we crash before
1618 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1619 ext4_orphan_add(handle, inode);
1621 ext4_journal_stop(handle);
1622 if (pos + len > inode->i_size) {
1623 ext4_truncate_failed_write(inode);
1625 * If truncate failed early the inode might
1626 * still be on the orphan list; we need to
1627 * make sure the inode is removed from the
1628 * orphan list in that case.
1631 ext4_orphan_del(NULL, inode);
1635 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1641 /* For write_end() in data=journal mode */
1642 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1644 if (!buffer_mapped(bh) || buffer_freed(bh))
1646 set_buffer_uptodate(bh);
1647 return ext4_handle_dirty_metadata(handle, NULL, bh);
1650 static int ext4_generic_write_end(struct file *file,
1651 struct address_space *mapping,
1652 loff_t pos, unsigned len, unsigned copied,
1653 struct page *page, void *fsdata)
1655 int i_size_changed = 0;
1656 struct inode *inode = mapping->host;
1657 handle_t *handle = ext4_journal_current_handle();
1659 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1662 * No need to use i_size_read() here, the i_size
1663 * cannot change under us because we hold i_mutex.
1665 * But it's important to update i_size while still holding page lock:
1666 * page writeout could otherwise come in and zero beyond i_size.
1668 if (pos + copied > inode->i_size) {
1669 i_size_write(inode, pos + copied);
1673 if (pos + copied > EXT4_I(inode)->i_disksize) {
1674 /* We need to mark inode dirty even if
1675 * new_i_size is less that inode->i_size
1676 * bu greater than i_disksize.(hint delalloc)
1678 ext4_update_i_disksize(inode, (pos + copied));
1682 page_cache_release(page);
1685 * Don't mark the inode dirty under page lock. First, it unnecessarily
1686 * makes the holding time of page lock longer. Second, it forces lock
1687 * ordering of page lock and transaction start for journaling
1691 ext4_mark_inode_dirty(handle, inode);
1697 * We need to pick up the new inode size which generic_commit_write gave us
1698 * `file' can be NULL - eg, when called from page_symlink().
1700 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1701 * buffers are managed internally.
1703 static int ext4_ordered_write_end(struct file *file,
1704 struct address_space *mapping,
1705 loff_t pos, unsigned len, unsigned copied,
1706 struct page *page, void *fsdata)
1708 handle_t *handle = ext4_journal_current_handle();
1709 struct inode *inode = mapping->host;
1712 trace_ext4_ordered_write_end(inode, pos, len, copied);
1713 ret = ext4_jbd2_file_inode(handle, inode);
1716 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1719 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1720 /* if we have allocated more blocks and copied
1721 * less. We will have blocks allocated outside
1722 * inode->i_size. So truncate them
1724 ext4_orphan_add(handle, inode);
1728 ret2 = ext4_journal_stop(handle);
1732 if (pos + len > inode->i_size) {
1733 ext4_truncate_failed_write(inode);
1735 * If truncate failed early the inode might still be
1736 * on the orphan list; we need to make sure the inode
1737 * is removed from the orphan list in that case.
1740 ext4_orphan_del(NULL, inode);
1744 return ret ? ret : copied;
1747 static int ext4_writeback_write_end(struct file *file,
1748 struct address_space *mapping,
1749 loff_t pos, unsigned len, unsigned copied,
1750 struct page *page, void *fsdata)
1752 handle_t *handle = ext4_journal_current_handle();
1753 struct inode *inode = mapping->host;
1756 trace_ext4_writeback_write_end(inode, pos, len, copied);
1757 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1760 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1761 /* if we have allocated more blocks and copied
1762 * less. We will have blocks allocated outside
1763 * inode->i_size. So truncate them
1765 ext4_orphan_add(handle, inode);
1770 ret2 = ext4_journal_stop(handle);
1774 if (pos + len > inode->i_size) {
1775 ext4_truncate_failed_write(inode);
1777 * If truncate failed early the inode might still be
1778 * on the orphan list; we need to make sure the inode
1779 * is removed from the orphan list in that case.
1782 ext4_orphan_del(NULL, inode);
1785 return ret ? ret : copied;
1788 static int ext4_journalled_write_end(struct file *file,
1789 struct address_space *mapping,
1790 loff_t pos, unsigned len, unsigned copied,
1791 struct page *page, void *fsdata)
1793 handle_t *handle = ext4_journal_current_handle();
1794 struct inode *inode = mapping->host;
1800 trace_ext4_journalled_write_end(inode, pos, len, copied);
1801 from = pos & (PAGE_CACHE_SIZE - 1);
1805 if (!PageUptodate(page))
1807 page_zero_new_buffers(page, from+copied, to);
1810 ret = walk_page_buffers(handle, page_buffers(page), from,
1811 to, &partial, write_end_fn);
1813 SetPageUptodate(page);
1814 new_i_size = pos + copied;
1815 if (new_i_size > inode->i_size)
1816 i_size_write(inode, pos+copied);
1817 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1818 if (new_i_size > EXT4_I(inode)->i_disksize) {
1819 ext4_update_i_disksize(inode, new_i_size);
1820 ret2 = ext4_mark_inode_dirty(handle, inode);
1826 page_cache_release(page);
1827 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1828 /* if we have allocated more blocks and copied
1829 * less. We will have blocks allocated outside
1830 * inode->i_size. So truncate them
1832 ext4_orphan_add(handle, inode);
1834 ret2 = ext4_journal_stop(handle);
1837 if (pos + len > inode->i_size) {
1838 ext4_truncate_failed_write(inode);
1840 * If truncate failed early the inode might still be
1841 * on the orphan list; we need to make sure the inode
1842 * is removed from the orphan list in that case.
1845 ext4_orphan_del(NULL, inode);
1848 return ret ? ret : copied;
1852 * Reserve a single block located at lblock
1854 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1857 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1858 struct ext4_inode_info *ei = EXT4_I(inode);
1859 unsigned long md_needed, md_reserved;
1862 * recalculate the amount of metadata blocks to reserve
1863 * in order to allocate nrblocks
1864 * worse case is one extent per block
1867 spin_lock(&ei->i_block_reservation_lock);
1868 md_reserved = ei->i_reserved_meta_blocks;
1869 md_needed = ext4_calc_metadata_amount(inode, lblock);
1870 trace_ext4_da_reserve_space(inode, md_needed);
1871 spin_unlock(&ei->i_block_reservation_lock);
1874 * Make quota reservation here to prevent quota overflow
1875 * later. Real quota accounting is done at pages writeout
1878 if (vfs_dq_reserve_block(inode, md_needed + 1))
1881 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1882 vfs_dq_release_reservation_block(inode, md_needed + 1);
1883 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1889 spin_lock(&ei->i_block_reservation_lock);
1890 ei->i_reserved_data_blocks++;
1891 ei->i_reserved_meta_blocks += md_needed;
1892 spin_unlock(&ei->i_block_reservation_lock);
1894 return 0; /* success */
1897 static void ext4_da_release_space(struct inode *inode, int to_free)
1899 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1900 struct ext4_inode_info *ei = EXT4_I(inode);
1903 return; /* Nothing to release, exit */
1905 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1907 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1909 * if there aren't enough reserved blocks, then the
1910 * counter is messed up somewhere. Since this
1911 * function is called from invalidate page, it's
1912 * harmless to return without any action.
1914 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1915 "ino %lu, to_free %d with only %d reserved "
1916 "data blocks\n", inode->i_ino, to_free,
1917 ei->i_reserved_data_blocks);
1919 to_free = ei->i_reserved_data_blocks;
1921 ei->i_reserved_data_blocks -= to_free;
1923 if (ei->i_reserved_data_blocks == 0) {
1925 * We can release all of the reserved metadata blocks
1926 * only when we have written all of the delayed
1927 * allocation blocks.
1929 to_free += ei->i_reserved_meta_blocks;
1930 ei->i_reserved_meta_blocks = 0;
1931 ei->i_da_metadata_calc_len = 0;
1934 /* update fs dirty blocks counter */
1935 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1937 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1939 vfs_dq_release_reservation_block(inode, to_free);
1942 static void ext4_da_page_release_reservation(struct page *page,
1943 unsigned long offset)
1946 struct buffer_head *head, *bh;
1947 unsigned int curr_off = 0;
1949 head = page_buffers(page);
1952 unsigned int next_off = curr_off + bh->b_size;
1954 if ((offset <= curr_off) && (buffer_delay(bh))) {
1956 clear_buffer_delay(bh);
1958 curr_off = next_off;
1959 } while ((bh = bh->b_this_page) != head);
1960 ext4_da_release_space(page->mapping->host, to_release);
1964 * Delayed allocation stuff
1968 * mpage_da_submit_io - walks through extent of pages and try to write
1969 * them with writepage() call back
1971 * @mpd->inode: inode
1972 * @mpd->first_page: first page of the extent
1973 * @mpd->next_page: page after the last page of the extent
1975 * By the time mpage_da_submit_io() is called we expect all blocks
1976 * to be allocated. this may be wrong if allocation failed.
1978 * As pages are already locked by write_cache_pages(), we can't use it
1980 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1983 struct pagevec pvec;
1984 unsigned long index, end;
1985 int ret = 0, err, nr_pages, i;
1986 struct inode *inode = mpd->inode;
1987 struct address_space *mapping = inode->i_mapping;
1989 BUG_ON(mpd->next_page <= mpd->first_page);
1991 * We need to start from the first_page to the next_page - 1
1992 * to make sure we also write the mapped dirty buffer_heads.
1993 * If we look at mpd->b_blocknr we would only be looking
1994 * at the currently mapped buffer_heads.
1996 index = mpd->first_page;
1997 end = mpd->next_page - 1;
1999 pagevec_init(&pvec, 0);
2000 while (index <= end) {
2001 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2004 for (i = 0; i < nr_pages; i++) {
2005 struct page *page = pvec.pages[i];
2007 index = page->index;
2012 BUG_ON(!PageLocked(page));
2013 BUG_ON(PageWriteback(page));
2015 pages_skipped = mpd->wbc->pages_skipped;
2016 err = mapping->a_ops->writepage(page, mpd->wbc);
2017 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2019 * have successfully written the page
2020 * without skipping the same
2022 mpd->pages_written++;
2024 * In error case, we have to continue because
2025 * remaining pages are still locked
2026 * XXX: unlock and re-dirty them?
2031 pagevec_release(&pvec);
2037 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2039 * @mpd->inode - inode to walk through
2040 * @exbh->b_blocknr - first block on a disk
2041 * @exbh->b_size - amount of space in bytes
2042 * @logical - first logical block to start assignment with
2044 * the function goes through all passed space and put actual disk
2045 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2047 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2048 struct buffer_head *exbh)
2050 struct inode *inode = mpd->inode;
2051 struct address_space *mapping = inode->i_mapping;
2052 int blocks = exbh->b_size >> inode->i_blkbits;
2053 sector_t pblock = exbh->b_blocknr, cur_logical;
2054 struct buffer_head *head, *bh;
2056 struct pagevec pvec;
2059 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2060 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2061 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2063 pagevec_init(&pvec, 0);
2065 while (index <= end) {
2066 /* XXX: optimize tail */
2067 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2070 for (i = 0; i < nr_pages; i++) {
2071 struct page *page = pvec.pages[i];
2073 index = page->index;
2078 BUG_ON(!PageLocked(page));
2079 BUG_ON(PageWriteback(page));
2080 BUG_ON(!page_has_buffers(page));
2082 bh = page_buffers(page);
2085 /* skip blocks out of the range */
2087 if (cur_logical >= logical)
2090 } while ((bh = bh->b_this_page) != head);
2093 if (cur_logical >= logical + blocks)
2096 if (buffer_delay(bh) ||
2097 buffer_unwritten(bh)) {
2099 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2101 if (buffer_delay(bh)) {
2102 clear_buffer_delay(bh);
2103 bh->b_blocknr = pblock;
2106 * unwritten already should have
2107 * blocknr assigned. Verify that
2109 clear_buffer_unwritten(bh);
2110 BUG_ON(bh->b_blocknr != pblock);
2113 } else if (buffer_mapped(bh))
2114 BUG_ON(bh->b_blocknr != pblock);
2116 if (buffer_uninit(exbh))
2117 set_buffer_uninit(bh);
2120 } while ((bh = bh->b_this_page) != head);
2122 pagevec_release(&pvec);
2128 * __unmap_underlying_blocks - just a helper function to unmap
2129 * set of blocks described by @bh
2131 static inline void __unmap_underlying_blocks(struct inode *inode,
2132 struct buffer_head *bh)
2134 struct block_device *bdev = inode->i_sb->s_bdev;
2137 blocks = bh->b_size >> inode->i_blkbits;
2138 for (i = 0; i < blocks; i++)
2139 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2142 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2143 sector_t logical, long blk_cnt)
2147 struct pagevec pvec;
2148 struct inode *inode = mpd->inode;
2149 struct address_space *mapping = inode->i_mapping;
2151 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2152 end = (logical + blk_cnt - 1) >>
2153 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2154 while (index <= end) {
2155 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2158 for (i = 0; i < nr_pages; i++) {
2159 struct page *page = pvec.pages[i];
2160 if (page->index > end)
2162 BUG_ON(!PageLocked(page));
2163 BUG_ON(PageWriteback(page));
2164 block_invalidatepage(page, 0);
2165 ClearPageUptodate(page);
2168 index = pvec.pages[nr_pages - 1]->index + 1;
2169 pagevec_release(&pvec);
2174 static void ext4_print_free_blocks(struct inode *inode)
2176 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2177 printk(KERN_CRIT "Total free blocks count %lld\n",
2178 ext4_count_free_blocks(inode->i_sb));
2179 printk(KERN_CRIT "Free/Dirty block details\n");
2180 printk(KERN_CRIT "free_blocks=%lld\n",
2181 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2182 printk(KERN_CRIT "dirty_blocks=%lld\n",
2183 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2184 printk(KERN_CRIT "Block reservation details\n");
2185 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2186 EXT4_I(inode)->i_reserved_data_blocks);
2187 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2188 EXT4_I(inode)->i_reserved_meta_blocks);
2193 * mpage_da_map_blocks - go through given space
2195 * @mpd - bh describing space
2197 * The function skips space we know is already mapped to disk blocks.
2200 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2202 int err, blks, get_blocks_flags;
2203 struct buffer_head new;
2204 sector_t next = mpd->b_blocknr;
2205 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2206 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2207 handle_t *handle = NULL;
2210 * We consider only non-mapped and non-allocated blocks
2212 if ((mpd->b_state & (1 << BH_Mapped)) &&
2213 !(mpd->b_state & (1 << BH_Delay)) &&
2214 !(mpd->b_state & (1 << BH_Unwritten)))
2218 * If we didn't accumulate anything to write simply return
2223 handle = ext4_journal_current_handle();
2227 * Call ext4_get_blocks() to allocate any delayed allocation
2228 * blocks, or to convert an uninitialized extent to be
2229 * initialized (in the case where we have written into
2230 * one or more preallocated blocks).
2232 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2233 * indicate that we are on the delayed allocation path. This
2234 * affects functions in many different parts of the allocation
2235 * call path. This flag exists primarily because we don't
2236 * want to change *many* call functions, so ext4_get_blocks()
2237 * will set the magic i_delalloc_reserved_flag once the
2238 * inode's allocation semaphore is taken.
2240 * If the blocks in questions were delalloc blocks, set
2241 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2242 * variables are updated after the blocks have been allocated.
2245 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2246 if (ext4_should_dioread_nolock(mpd->inode))
2247 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2248 if (mpd->b_state & (1 << BH_Delay))
2249 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2251 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2252 &new, get_blocks_flags);
2256 * If get block returns with error we simply
2257 * return. Later writepage will redirty the page and
2258 * writepages will find the dirty page again
2263 if (err == -ENOSPC &&
2264 ext4_count_free_blocks(mpd->inode->i_sb)) {
2270 * get block failure will cause us to loop in
2271 * writepages, because a_ops->writepage won't be able
2272 * to make progress. The page will be redirtied by
2273 * writepage and writepages will again try to write
2276 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2277 "delayed block allocation failed for inode %lu at "
2278 "logical offset %llu with max blocks %zd with "
2279 "error %d\n", mpd->inode->i_ino,
2280 (unsigned long long) next,
2281 mpd->b_size >> mpd->inode->i_blkbits, err);
2282 printk(KERN_CRIT "This should not happen!! "
2283 "Data will be lost\n");
2284 if (err == -ENOSPC) {
2285 ext4_print_free_blocks(mpd->inode);
2287 /* invalidate all the pages */
2288 ext4_da_block_invalidatepages(mpd, next,
2289 mpd->b_size >> mpd->inode->i_blkbits);
2294 new.b_size = (blks << mpd->inode->i_blkbits);
2296 if (buffer_new(&new))
2297 __unmap_underlying_blocks(mpd->inode, &new);
2300 * If blocks are delayed marked, we need to
2301 * put actual blocknr and drop delayed bit
2303 if ((mpd->b_state & (1 << BH_Delay)) ||
2304 (mpd->b_state & (1 << BH_Unwritten)))
2305 mpage_put_bnr_to_bhs(mpd, next, &new);
2307 if (ext4_should_order_data(mpd->inode)) {
2308 err = ext4_jbd2_file_inode(handle, mpd->inode);
2314 * Update on-disk size along with block allocation.
2316 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2317 if (disksize > i_size_read(mpd->inode))
2318 disksize = i_size_read(mpd->inode);
2319 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2320 ext4_update_i_disksize(mpd->inode, disksize);
2321 return ext4_mark_inode_dirty(handle, mpd->inode);
2327 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2328 (1 << BH_Delay) | (1 << BH_Unwritten))
2331 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2333 * @mpd->lbh - extent of blocks
2334 * @logical - logical number of the block in the file
2335 * @bh - bh of the block (used to access block's state)
2337 * the function is used to collect contig. blocks in same state
2339 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2340 sector_t logical, size_t b_size,
2341 unsigned long b_state)
2344 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2346 /* check if thereserved journal credits might overflow */
2347 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2348 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2350 * With non-extent format we are limited by the journal
2351 * credit available. Total credit needed to insert
2352 * nrblocks contiguous blocks is dependent on the
2353 * nrblocks. So limit nrblocks.
2356 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2357 EXT4_MAX_TRANS_DATA) {
2359 * Adding the new buffer_head would make it cross the
2360 * allowed limit for which we have journal credit
2361 * reserved. So limit the new bh->b_size
2363 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2364 mpd->inode->i_blkbits;
2365 /* we will do mpage_da_submit_io in the next loop */
2369 * First block in the extent
2371 if (mpd->b_size == 0) {
2372 mpd->b_blocknr = logical;
2373 mpd->b_size = b_size;
2374 mpd->b_state = b_state & BH_FLAGS;
2378 next = mpd->b_blocknr + nrblocks;
2380 * Can we merge the block to our big extent?
2382 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2383 mpd->b_size += b_size;
2389 * We couldn't merge the block to our extent, so we
2390 * need to flush current extent and start new one
2392 if (mpage_da_map_blocks(mpd) == 0)
2393 mpage_da_submit_io(mpd);
2398 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2400 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2404 * __mpage_da_writepage - finds extent of pages and blocks
2406 * @page: page to consider
2407 * @wbc: not used, we just follow rules
2410 * The function finds extents of pages and scan them for all blocks.
2412 static int __mpage_da_writepage(struct page *page,
2413 struct writeback_control *wbc, void *data)
2415 struct mpage_da_data *mpd = data;
2416 struct inode *inode = mpd->inode;
2417 struct buffer_head *bh, *head;
2422 * Rest of the page in the page_vec
2423 * redirty then and skip then. We will
2424 * try to write them again after
2425 * starting a new transaction
2427 redirty_page_for_writepage(wbc, page);
2429 return MPAGE_DA_EXTENT_TAIL;
2432 * Can we merge this page to current extent?
2434 if (mpd->next_page != page->index) {
2436 * Nope, we can't. So, we map non-allocated blocks
2437 * and start IO on them using writepage()
2439 if (mpd->next_page != mpd->first_page) {
2440 if (mpage_da_map_blocks(mpd) == 0)
2441 mpage_da_submit_io(mpd);
2443 * skip rest of the page in the page_vec
2446 redirty_page_for_writepage(wbc, page);
2448 return MPAGE_DA_EXTENT_TAIL;
2452 * Start next extent of pages ...
2454 mpd->first_page = page->index;
2464 mpd->next_page = page->index + 1;
2465 logical = (sector_t) page->index <<
2466 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2468 if (!page_has_buffers(page)) {
2469 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2470 (1 << BH_Dirty) | (1 << BH_Uptodate));
2472 return MPAGE_DA_EXTENT_TAIL;
2475 * Page with regular buffer heads, just add all dirty ones
2477 head = page_buffers(page);
2480 BUG_ON(buffer_locked(bh));
2482 * We need to try to allocate
2483 * unmapped blocks in the same page.
2484 * Otherwise we won't make progress
2485 * with the page in ext4_writepage
2487 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2488 mpage_add_bh_to_extent(mpd, logical,
2492 return MPAGE_DA_EXTENT_TAIL;
2493 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2495 * mapped dirty buffer. We need to update
2496 * the b_state because we look at
2497 * b_state in mpage_da_map_blocks. We don't
2498 * update b_size because if we find an
2499 * unmapped buffer_head later we need to
2500 * use the b_state flag of that buffer_head.
2502 if (mpd->b_size == 0)
2503 mpd->b_state = bh->b_state & BH_FLAGS;
2506 } while ((bh = bh->b_this_page) != head);
2513 * This is a special get_blocks_t callback which is used by
2514 * ext4_da_write_begin(). It will either return mapped block or
2515 * reserve space for a single block.
2517 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2518 * We also have b_blocknr = -1 and b_bdev initialized properly
2520 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2521 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2522 * initialized properly.
2524 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2525 struct buffer_head *bh_result, int create)
2528 sector_t invalid_block = ~((sector_t) 0xffff);
2530 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2533 BUG_ON(create == 0);
2534 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2537 * first, we need to know whether the block is allocated already
2538 * preallocated blocks are unmapped but should treated
2539 * the same as allocated blocks.
2541 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2542 if ((ret == 0) && !buffer_delay(bh_result)) {
2543 /* the block isn't (pre)allocated yet, let's reserve space */
2545 * XXX: __block_prepare_write() unmaps passed block,
2548 ret = ext4_da_reserve_space(inode, iblock);
2550 /* not enough space to reserve */
2553 map_bh(bh_result, inode->i_sb, invalid_block);
2554 set_buffer_new(bh_result);
2555 set_buffer_delay(bh_result);
2556 } else if (ret > 0) {
2557 bh_result->b_size = (ret << inode->i_blkbits);
2558 if (buffer_unwritten(bh_result)) {
2559 /* A delayed write to unwritten bh should
2560 * be marked new and mapped. Mapped ensures
2561 * that we don't do get_block multiple times
2562 * when we write to the same offset and new
2563 * ensures that we do proper zero out for
2566 set_buffer_new(bh_result);
2567 set_buffer_mapped(bh_result);
2576 * This function is used as a standard get_block_t calback function
2577 * when there is no desire to allocate any blocks. It is used as a
2578 * callback function for block_prepare_write(), nobh_writepage(), and
2579 * block_write_full_page(). These functions should only try to map a
2580 * single block at a time.
2582 * Since this function doesn't do block allocations even if the caller
2583 * requests it by passing in create=1, it is critically important that
2584 * any caller checks to make sure that any buffer heads are returned
2585 * by this function are either all already mapped or marked for
2586 * delayed allocation before calling nobh_writepage() or
2587 * block_write_full_page(). Otherwise, b_blocknr could be left
2588 * unitialized, and the page write functions will be taken by
2591 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2592 struct buffer_head *bh_result, int create)
2595 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2597 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2600 * we don't want to do block allocation in writepage
2601 * so call get_block_wrap with create = 0
2603 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2605 bh_result->b_size = (ret << inode->i_blkbits);
2611 static int bget_one(handle_t *handle, struct buffer_head *bh)
2617 static int bput_one(handle_t *handle, struct buffer_head *bh)
2623 static int __ext4_journalled_writepage(struct page *page,
2626 struct address_space *mapping = page->mapping;
2627 struct inode *inode = mapping->host;
2628 struct buffer_head *page_bufs;
2629 handle_t *handle = NULL;
2633 page_bufs = page_buffers(page);
2635 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2636 /* As soon as we unlock the page, it can go away, but we have
2637 * references to buffers so we are safe */
2640 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2641 if (IS_ERR(handle)) {
2642 ret = PTR_ERR(handle);
2646 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2647 do_journal_get_write_access);
2649 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2653 err = ext4_journal_stop(handle);
2657 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2658 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2663 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2664 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2667 * Note that we don't need to start a transaction unless we're journaling data
2668 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2669 * need to file the inode to the transaction's list in ordered mode because if
2670 * we are writing back data added by write(), the inode is already there and if
2671 * we are writing back data modified via mmap(), noone guarantees in which
2672 * transaction the data will hit the disk. In case we are journaling data, we
2673 * cannot start transaction directly because transaction start ranks above page
2674 * lock so we have to do some magic.
2676 * This function can get called via...
2677 * - ext4_da_writepages after taking page lock (have journal handle)
2678 * - journal_submit_inode_data_buffers (no journal handle)
2679 * - shrink_page_list via pdflush (no journal handle)
2680 * - grab_page_cache when doing write_begin (have journal handle)
2682 * We don't do any block allocation in this function. If we have page with
2683 * multiple blocks we need to write those buffer_heads that are mapped. This
2684 * is important for mmaped based write. So if we do with blocksize 1K
2685 * truncate(f, 1024);
2686 * a = mmap(f, 0, 4096);
2688 * truncate(f, 4096);
2689 * we have in the page first buffer_head mapped via page_mkwrite call back
2690 * but other bufer_heads would be unmapped but dirty(dirty done via the
2691 * do_wp_page). So writepage should write the first block. If we modify
2692 * the mmap area beyond 1024 we will again get a page_fault and the
2693 * page_mkwrite callback will do the block allocation and mark the
2694 * buffer_heads mapped.
2696 * We redirty the page if we have any buffer_heads that is either delay or
2697 * unwritten in the page.
2699 * We can get recursively called as show below.
2701 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2704 * But since we don't do any block allocation we should not deadlock.
2705 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2707 static int ext4_writepage(struct page *page,
2708 struct writeback_control *wbc)
2713 struct buffer_head *page_bufs = NULL;
2714 struct inode *inode = page->mapping->host;
2716 trace_ext4_writepage(inode, page);
2717 size = i_size_read(inode);
2718 if (page->index == size >> PAGE_CACHE_SHIFT)
2719 len = size & ~PAGE_CACHE_MASK;
2721 len = PAGE_CACHE_SIZE;
2723 if (page_has_buffers(page)) {
2724 page_bufs = page_buffers(page);
2725 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2726 ext4_bh_delay_or_unwritten)) {
2728 * We don't want to do block allocation
2729 * So redirty the page and return
2730 * We may reach here when we do a journal commit
2731 * via journal_submit_inode_data_buffers.
2732 * If we don't have mapping block we just ignore
2733 * them. We can also reach here via shrink_page_list
2735 redirty_page_for_writepage(wbc, page);
2741 * The test for page_has_buffers() is subtle:
2742 * We know the page is dirty but it lost buffers. That means
2743 * that at some moment in time after write_begin()/write_end()
2744 * has been called all buffers have been clean and thus they
2745 * must have been written at least once. So they are all
2746 * mapped and we can happily proceed with mapping them
2747 * and writing the page.
2749 * Try to initialize the buffer_heads and check whether
2750 * all are mapped and non delay. We don't want to
2751 * do block allocation here.
2753 ret = block_prepare_write(page, 0, len,
2754 noalloc_get_block_write);
2756 page_bufs = page_buffers(page);
2757 /* check whether all are mapped and non delay */
2758 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2759 ext4_bh_delay_or_unwritten)) {
2760 redirty_page_for_writepage(wbc, page);
2766 * We can't do block allocation here
2767 * so just redity the page and unlock
2770 redirty_page_for_writepage(wbc, page);
2774 /* now mark the buffer_heads as dirty and uptodate */
2775 block_commit_write(page, 0, len);
2778 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2780 * It's mmapped pagecache. Add buffers and journal it. There
2781 * doesn't seem much point in redirtying the page here.
2783 ClearPageChecked(page);
2784 return __ext4_journalled_writepage(page, len);
2787 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2788 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2789 else if (page_bufs && buffer_uninit(page_bufs)) {
2790 ext4_set_bh_endio(page_bufs, inode);
2791 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2792 wbc, ext4_end_io_buffer_write);
2794 ret = block_write_full_page(page, noalloc_get_block_write,
2801 * This is called via ext4_da_writepages() to
2802 * calulate the total number of credits to reserve to fit
2803 * a single extent allocation into a single transaction,
2804 * ext4_da_writpeages() will loop calling this before
2805 * the block allocation.
2808 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2810 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2813 * With non-extent format the journal credit needed to
2814 * insert nrblocks contiguous block is dependent on
2815 * number of contiguous block. So we will limit
2816 * number of contiguous block to a sane value
2818 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) &&
2819 (max_blocks > EXT4_MAX_TRANS_DATA))
2820 max_blocks = EXT4_MAX_TRANS_DATA;
2822 return ext4_chunk_trans_blocks(inode, max_blocks);
2825 static int ext4_da_writepages(struct address_space *mapping,
2826 struct writeback_control *wbc)
2829 int range_whole = 0;
2830 handle_t *handle = NULL;
2831 struct mpage_da_data mpd;
2832 struct inode *inode = mapping->host;
2833 int no_nrwrite_index_update;
2834 int pages_written = 0;
2836 unsigned int max_pages;
2837 int range_cyclic, cycled = 1, io_done = 0;
2838 int needed_blocks, ret = 0;
2839 long desired_nr_to_write, nr_to_writebump = 0;
2840 loff_t range_start = wbc->range_start;
2841 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2843 trace_ext4_da_writepages(inode, wbc);
2846 * No pages to write? This is mainly a kludge to avoid starting
2847 * a transaction for special inodes like journal inode on last iput()
2848 * because that could violate lock ordering on umount
2850 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2854 * If the filesystem has aborted, it is read-only, so return
2855 * right away instead of dumping stack traces later on that
2856 * will obscure the real source of the problem. We test
2857 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2858 * the latter could be true if the filesystem is mounted
2859 * read-only, and in that case, ext4_da_writepages should
2860 * *never* be called, so if that ever happens, we would want
2863 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2866 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2869 range_cyclic = wbc->range_cyclic;
2870 if (wbc->range_cyclic) {
2871 index = mapping->writeback_index;
2874 wbc->range_start = index << PAGE_CACHE_SHIFT;
2875 wbc->range_end = LLONG_MAX;
2876 wbc->range_cyclic = 0;
2878 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2881 * This works around two forms of stupidity. The first is in
2882 * the writeback code, which caps the maximum number of pages
2883 * written to be 1024 pages. This is wrong on multiple
2884 * levels; different architectues have a different page size,
2885 * which changes the maximum amount of data which gets
2886 * written. Secondly, 4 megabytes is way too small. XFS
2887 * forces this value to be 16 megabytes by multiplying
2888 * nr_to_write parameter by four, and then relies on its
2889 * allocator to allocate larger extents to make them
2890 * contiguous. Unfortunately this brings us to the second
2891 * stupidity, which is that ext4's mballoc code only allocates
2892 * at most 2048 blocks. So we force contiguous writes up to
2893 * the number of dirty blocks in the inode, or
2894 * sbi->max_writeback_mb_bump whichever is smaller.
2896 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2897 if (!range_cyclic && range_whole)
2898 desired_nr_to_write = wbc->nr_to_write * 8;
2900 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2902 if (desired_nr_to_write > max_pages)
2903 desired_nr_to_write = max_pages;
2905 if (wbc->nr_to_write < desired_nr_to_write) {
2906 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2907 wbc->nr_to_write = desired_nr_to_write;
2911 mpd.inode = mapping->host;
2914 * we don't want write_cache_pages to update
2915 * nr_to_write and writeback_index
2917 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2918 wbc->no_nrwrite_index_update = 1;
2919 pages_skipped = wbc->pages_skipped;
2922 while (!ret && wbc->nr_to_write > 0) {
2925 * we insert one extent at a time. So we need
2926 * credit needed for single extent allocation.
2927 * journalled mode is currently not supported
2930 BUG_ON(ext4_should_journal_data(inode));
2931 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2933 /* start a new transaction*/
2934 handle = ext4_journal_start(inode, needed_blocks);
2935 if (IS_ERR(handle)) {
2936 ret = PTR_ERR(handle);
2937 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2938 "%ld pages, ino %lu; err %d\n", __func__,
2939 wbc->nr_to_write, inode->i_ino, ret);
2940 goto out_writepages;
2944 * Now call __mpage_da_writepage to find the next
2945 * contiguous region of logical blocks that need
2946 * blocks to be allocated by ext4. We don't actually
2947 * submit the blocks for I/O here, even though
2948 * write_cache_pages thinks it will, and will set the
2949 * pages as clean for write before calling
2950 * __mpage_da_writepage().
2958 mpd.pages_written = 0;
2960 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2963 * If we have a contiguous extent of pages and we
2964 * haven't done the I/O yet, map the blocks and submit
2967 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2968 if (mpage_da_map_blocks(&mpd) == 0)
2969 mpage_da_submit_io(&mpd);
2971 ret = MPAGE_DA_EXTENT_TAIL;
2973 trace_ext4_da_write_pages(inode, &mpd);
2974 wbc->nr_to_write -= mpd.pages_written;
2976 ext4_journal_stop(handle);
2978 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2979 /* commit the transaction which would
2980 * free blocks released in the transaction
2983 jbd2_journal_force_commit_nested(sbi->s_journal);
2984 wbc->pages_skipped = pages_skipped;
2986 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2988 * got one extent now try with
2991 pages_written += mpd.pages_written;
2992 wbc->pages_skipped = pages_skipped;
2995 } else if (wbc->nr_to_write)
2997 * There is no more writeout needed
2998 * or we requested for a noblocking writeout
2999 * and we found the device congested
3003 if (!io_done && !cycled) {
3006 wbc->range_start = index << PAGE_CACHE_SHIFT;
3007 wbc->range_end = mapping->writeback_index - 1;
3010 if (pages_skipped != wbc->pages_skipped)
3011 ext4_msg(inode->i_sb, KERN_CRIT,
3012 "This should not happen leaving %s "
3013 "with nr_to_write = %ld ret = %d\n",
3014 __func__, wbc->nr_to_write, ret);
3017 index += pages_written;
3018 wbc->range_cyclic = range_cyclic;
3019 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3021 * set the writeback_index so that range_cyclic
3022 * mode will write it back later
3024 mapping->writeback_index = index;
3027 if (!no_nrwrite_index_update)
3028 wbc->no_nrwrite_index_update = 0;
3029 wbc->nr_to_write -= nr_to_writebump;
3030 wbc->range_start = range_start;
3031 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3035 #define FALL_BACK_TO_NONDELALLOC 1
3036 static int ext4_nonda_switch(struct super_block *sb)
3038 s64 free_blocks, dirty_blocks;
3039 struct ext4_sb_info *sbi = EXT4_SB(sb);
3042 * switch to non delalloc mode if we are running low
3043 * on free block. The free block accounting via percpu
3044 * counters can get slightly wrong with percpu_counter_batch getting
3045 * accumulated on each CPU without updating global counters
3046 * Delalloc need an accurate free block accounting. So switch
3047 * to non delalloc when we are near to error range.
3049 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3050 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3051 if (2 * free_blocks < 3 * dirty_blocks ||
3052 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3054 * free block count is less than 150% of dirty blocks
3055 * or free blocks is less than watermark
3060 * Even if we don't switch but are nearing capacity,
3061 * start pushing delalloc when 1/2 of free blocks are dirty.
3063 if (free_blocks < 2 * dirty_blocks)
3064 writeback_inodes_sb_if_idle(sb);
3069 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3070 loff_t pos, unsigned len, unsigned flags,
3071 struct page **pagep, void **fsdata)
3073 int ret, retries = 0, quota_retries = 0;
3077 struct inode *inode = mapping->host;
3080 index = pos >> PAGE_CACHE_SHIFT;
3081 from = pos & (PAGE_CACHE_SIZE - 1);
3084 if (ext4_nonda_switch(inode->i_sb)) {
3085 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3086 return ext4_write_begin(file, mapping, pos,
3087 len, flags, pagep, fsdata);
3089 *fsdata = (void *)0;
3090 trace_ext4_da_write_begin(inode, pos, len, flags);
3093 * With delayed allocation, we don't log the i_disksize update
3094 * if there is delayed block allocation. But we still need
3095 * to journalling the i_disksize update if writes to the end
3096 * of file which has an already mapped buffer.
3098 handle = ext4_journal_start(inode, 1);
3099 if (IS_ERR(handle)) {
3100 ret = PTR_ERR(handle);
3103 /* We cannot recurse into the filesystem as the transaction is already
3105 flags |= AOP_FLAG_NOFS;
3107 page = grab_cache_page_write_begin(mapping, index, flags);
3109 ext4_journal_stop(handle);
3115 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3116 ext4_da_get_block_prep);
3119 ext4_journal_stop(handle);
3120 page_cache_release(page);
3122 * block_write_begin may have instantiated a few blocks
3123 * outside i_size. Trim these off again. Don't need
3124 * i_size_read because we hold i_mutex.
3126 if (pos + len > inode->i_size)
3127 ext4_truncate_failed_write(inode);
3130 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3133 if ((ret == -EDQUOT) &&
3134 EXT4_I(inode)->i_reserved_meta_blocks &&
3135 (quota_retries++ < 3)) {
3137 * Since we often over-estimate the number of meta
3138 * data blocks required, we may sometimes get a
3139 * spurios out of quota error even though there would
3140 * be enough space once we write the data blocks and
3141 * find out how many meta data blocks were _really_
3142 * required. So try forcing the inode write to see if
3145 write_inode_now(inode, (quota_retries == 3));
3153 * Check if we should update i_disksize
3154 * when write to the end of file but not require block allocation
3156 static int ext4_da_should_update_i_disksize(struct page *page,
3157 unsigned long offset)
3159 struct buffer_head *bh;
3160 struct inode *inode = page->mapping->host;
3164 bh = page_buffers(page);
3165 idx = offset >> inode->i_blkbits;
3167 for (i = 0; i < idx; i++)
3168 bh = bh->b_this_page;
3170 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3175 static int ext4_da_write_end(struct file *file,
3176 struct address_space *mapping,
3177 loff_t pos, unsigned len, unsigned copied,
3178 struct page *page, void *fsdata)
3180 struct inode *inode = mapping->host;
3182 handle_t *handle = ext4_journal_current_handle();
3184 unsigned long start, end;
3185 int write_mode = (int)(unsigned long)fsdata;
3187 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3188 if (ext4_should_order_data(inode)) {
3189 return ext4_ordered_write_end(file, mapping, pos,
3190 len, copied, page, fsdata);
3191 } else if (ext4_should_writeback_data(inode)) {
3192 return ext4_writeback_write_end(file, mapping, pos,
3193 len, copied, page, fsdata);
3199 trace_ext4_da_write_end(inode, pos, len, copied);
3200 start = pos & (PAGE_CACHE_SIZE - 1);
3201 end = start + copied - 1;
3204 * generic_write_end() will run mark_inode_dirty() if i_size
3205 * changes. So let's piggyback the i_disksize mark_inode_dirty
3209 new_i_size = pos + copied;
3210 if (new_i_size > EXT4_I(inode)->i_disksize) {
3211 if (ext4_da_should_update_i_disksize(page, end)) {
3212 down_write(&EXT4_I(inode)->i_data_sem);
3213 if (new_i_size > EXT4_I(inode)->i_disksize) {
3215 * Updating i_disksize when extending file
3216 * without needing block allocation
3218 if (ext4_should_order_data(inode))
3219 ret = ext4_jbd2_file_inode(handle,
3222 EXT4_I(inode)->i_disksize = new_i_size;
3224 up_write(&EXT4_I(inode)->i_data_sem);
3225 /* We need to mark inode dirty even if
3226 * new_i_size is less that inode->i_size
3227 * bu greater than i_disksize.(hint delalloc)
3229 ext4_mark_inode_dirty(handle, inode);
3232 ret2 = generic_write_end(file, mapping, pos, len, copied,
3237 ret2 = ext4_journal_stop(handle);
3241 return ret ? ret : copied;
3244 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3247 * Drop reserved blocks
3249 BUG_ON(!PageLocked(page));
3250 if (!page_has_buffers(page))
3253 ext4_da_page_release_reservation(page, offset);
3256 ext4_invalidatepage(page, offset);
3262 * Force all delayed allocation blocks to be allocated for a given inode.
3264 int ext4_alloc_da_blocks(struct inode *inode)
3266 trace_ext4_alloc_da_blocks(inode);
3268 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3269 !EXT4_I(inode)->i_reserved_meta_blocks)
3273 * We do something simple for now. The filemap_flush() will
3274 * also start triggering a write of the data blocks, which is
3275 * not strictly speaking necessary (and for users of
3276 * laptop_mode, not even desirable). However, to do otherwise
3277 * would require replicating code paths in:
3279 * ext4_da_writepages() ->
3280 * write_cache_pages() ---> (via passed in callback function)
3281 * __mpage_da_writepage() -->
3282 * mpage_add_bh_to_extent()
3283 * mpage_da_map_blocks()
3285 * The problem is that write_cache_pages(), located in
3286 * mm/page-writeback.c, marks pages clean in preparation for
3287 * doing I/O, which is not desirable if we're not planning on
3290 * We could call write_cache_pages(), and then redirty all of
3291 * the pages by calling redirty_page_for_writeback() but that
3292 * would be ugly in the extreme. So instead we would need to
3293 * replicate parts of the code in the above functions,
3294 * simplifying them becuase we wouldn't actually intend to
3295 * write out the pages, but rather only collect contiguous
3296 * logical block extents, call the multi-block allocator, and
3297 * then update the buffer heads with the block allocations.
3299 * For now, though, we'll cheat by calling filemap_flush(),
3300 * which will map the blocks, and start the I/O, but not
3301 * actually wait for the I/O to complete.
3303 return filemap_flush(inode->i_mapping);
3307 * bmap() is special. It gets used by applications such as lilo and by
3308 * the swapper to find the on-disk block of a specific piece of data.
3310 * Naturally, this is dangerous if the block concerned is still in the
3311 * journal. If somebody makes a swapfile on an ext4 data-journaling
3312 * filesystem and enables swap, then they may get a nasty shock when the
3313 * data getting swapped to that swapfile suddenly gets overwritten by
3314 * the original zero's written out previously to the journal and
3315 * awaiting writeback in the kernel's buffer cache.
3317 * So, if we see any bmap calls here on a modified, data-journaled file,
3318 * take extra steps to flush any blocks which might be in the cache.
3320 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3322 struct inode *inode = mapping->host;
3326 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3327 test_opt(inode->i_sb, DELALLOC)) {
3329 * With delalloc we want to sync the file
3330 * so that we can make sure we allocate
3333 filemap_write_and_wait(mapping);
3336 if (EXT4_JOURNAL(inode) &&
3337 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3339 * This is a REALLY heavyweight approach, but the use of
3340 * bmap on dirty files is expected to be extremely rare:
3341 * only if we run lilo or swapon on a freshly made file
3342 * do we expect this to happen.
3344 * (bmap requires CAP_SYS_RAWIO so this does not
3345 * represent an unprivileged user DOS attack --- we'd be
3346 * in trouble if mortal users could trigger this path at
3349 * NB. EXT4_STATE_JDATA is not set on files other than
3350 * regular files. If somebody wants to bmap a directory
3351 * or symlink and gets confused because the buffer
3352 * hasn't yet been flushed to disk, they deserve
3353 * everything they get.
3356 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3357 journal = EXT4_JOURNAL(inode);
3358 jbd2_journal_lock_updates(journal);
3359 err = jbd2_journal_flush(journal);
3360 jbd2_journal_unlock_updates(journal);
3366 return generic_block_bmap(mapping, block, ext4_get_block);
3369 static int ext4_readpage(struct file *file, struct page *page)
3371 return mpage_readpage(page, ext4_get_block);
3375 ext4_readpages(struct file *file, struct address_space *mapping,
3376 struct list_head *pages, unsigned nr_pages)
3378 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3381 static void ext4_free_io_end(ext4_io_end_t *io)
3390 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3392 struct buffer_head *head, *bh;
3393 unsigned int curr_off = 0;
3395 if (!page_has_buffers(page))
3397 head = bh = page_buffers(page);
3399 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3401 ext4_free_io_end(bh->b_private);
3402 bh->b_private = NULL;
3403 bh->b_end_io = NULL;
3405 curr_off = curr_off + bh->b_size;
3406 bh = bh->b_this_page;
3407 } while (bh != head);
3410 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3412 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3415 * free any io_end structure allocated for buffers to be discarded
3417 if (ext4_should_dioread_nolock(page->mapping->host))
3418 ext4_invalidatepage_free_endio(page, offset);
3420 * If it's a full truncate we just forget about the pending dirtying
3423 ClearPageChecked(page);
3426 jbd2_journal_invalidatepage(journal, page, offset);
3428 block_invalidatepage(page, offset);
3431 static int ext4_releasepage(struct page *page, gfp_t wait)
3433 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3435 WARN_ON(PageChecked(page));
3436 if (!page_has_buffers(page))
3439 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3441 return try_to_free_buffers(page);
3445 * O_DIRECT for ext3 (or indirect map) based files
3447 * If the O_DIRECT write will extend the file then add this inode to the
3448 * orphan list. So recovery will truncate it back to the original size
3449 * if the machine crashes during the write.
3451 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3452 * crashes then stale disk data _may_ be exposed inside the file. But current
3453 * VFS code falls back into buffered path in that case so we are safe.
3455 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3456 const struct iovec *iov, loff_t offset,
3457 unsigned long nr_segs)
3459 struct file *file = iocb->ki_filp;
3460 struct inode *inode = file->f_mapping->host;
3461 struct ext4_inode_info *ei = EXT4_I(inode);
3465 size_t count = iov_length(iov, nr_segs);
3469 loff_t final_size = offset + count;
3471 if (final_size > inode->i_size) {
3472 /* Credits for sb + inode write */
3473 handle = ext4_journal_start(inode, 2);
3474 if (IS_ERR(handle)) {
3475 ret = PTR_ERR(handle);
3478 ret = ext4_orphan_add(handle, inode);
3480 ext4_journal_stop(handle);
3484 ei->i_disksize = inode->i_size;
3485 ext4_journal_stop(handle);
3490 if (rw == READ && ext4_should_dioread_nolock(inode))
3491 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
3492 inode->i_sb->s_bdev, iov,
3494 ext4_get_block, NULL);
3496 ret = blockdev_direct_IO(rw, iocb, inode,
3497 inode->i_sb->s_bdev, iov,
3499 ext4_get_block, NULL);
3500 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3506 /* Credits for sb + inode write */
3507 handle = ext4_journal_start(inode, 2);
3508 if (IS_ERR(handle)) {
3509 /* This is really bad luck. We've written the data
3510 * but cannot extend i_size. Bail out and pretend
3511 * the write failed... */
3512 ret = PTR_ERR(handle);
3514 ext4_orphan_del(NULL, inode);
3519 ext4_orphan_del(handle, inode);
3521 loff_t end = offset + ret;
3522 if (end > inode->i_size) {
3523 ei->i_disksize = end;
3524 i_size_write(inode, end);
3526 * We're going to return a positive `ret'
3527 * here due to non-zero-length I/O, so there's
3528 * no way of reporting error returns from
3529 * ext4_mark_inode_dirty() to userspace. So
3532 ext4_mark_inode_dirty(handle, inode);
3535 err = ext4_journal_stop(handle);
3543 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3544 struct buffer_head *bh_result, int create)
3546 handle_t *handle = ext4_journal_current_handle();
3548 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3552 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3553 inode->i_ino, create);
3555 * ext4_get_block in prepare for a DIO write or buffer write.
3556 * We allocate an uinitialized extent if blocks haven't been allocated.
3557 * The extent will be converted to initialized after IO complete.
3559 create = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3562 if (max_blocks > DIO_MAX_BLOCKS)
3563 max_blocks = DIO_MAX_BLOCKS;
3564 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3565 handle = ext4_journal_start(inode, dio_credits);
3566 if (IS_ERR(handle)) {
3567 ret = PTR_ERR(handle);
3573 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3576 bh_result->b_size = (ret << inode->i_blkbits);
3580 ext4_journal_stop(handle);
3585 static void dump_completed_IO(struct inode * inode)
3588 struct list_head *cur, *before, *after;
3589 ext4_io_end_t *io, *io0, *io1;
3590 unsigned long flags;
3592 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3593 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3597 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3598 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3599 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3602 io0 = container_of(before, ext4_io_end_t, list);
3604 io1 = container_of(after, ext4_io_end_t, list);
3606 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3607 io, inode->i_ino, io0, io1);
3609 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3614 * check a range of space and convert unwritten extents to written.
3616 static int ext4_end_io_nolock(ext4_io_end_t *io)
3618 struct inode *inode = io->inode;
3619 loff_t offset = io->offset;
3620 ssize_t size = io->size;
3623 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3624 "list->prev 0x%p\n",
3625 io, inode->i_ino, io->list.next, io->list.prev);
3627 if (list_empty(&io->list))
3630 if (io->flag != EXT4_IO_UNWRITTEN)
3633 ret = ext4_convert_unwritten_extents(inode, offset, size);
3635 printk(KERN_EMERG "%s: failed to convert unwritten"
3636 "extents to written extents, error is %d"
3637 " io is still on inode %lu aio dio list\n",
3638 __func__, ret, inode->i_ino);
3642 /* clear the DIO AIO unwritten flag */
3648 * work on completed aio dio IO, to convert unwritten extents to extents
3650 static void ext4_end_io_work(struct work_struct *work)
3652 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3653 struct inode *inode = io->inode;
3654 struct ext4_inode_info *ei = EXT4_I(inode);
3655 unsigned long flags;
3658 mutex_lock(&inode->i_mutex);
3659 ret = ext4_end_io_nolock(io);
3661 mutex_unlock(&inode->i_mutex);
3665 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3666 if (!list_empty(&io->list))
3667 list_del_init(&io->list);
3668 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3669 mutex_unlock(&inode->i_mutex);
3670 ext4_free_io_end(io);
3674 * This function is called from ext4_sync_file().
3676 * When IO is completed, the work to convert unwritten extents to
3677 * written is queued on workqueue but may not get immediately
3678 * scheduled. When fsync is called, we need to ensure the
3679 * conversion is complete before fsync returns.
3680 * The inode keeps track of a list of pending/completed IO that
3681 * might needs to do the conversion. This function walks through
3682 * the list and convert the related unwritten extents for completed IO
3684 * The function return the number of pending IOs on success.
3686 int flush_completed_IO(struct inode *inode)
3689 struct ext4_inode_info *ei = EXT4_I(inode);
3690 unsigned long flags;
3694 if (list_empty(&ei->i_completed_io_list))
3697 dump_completed_IO(inode);
3698 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3699 while (!list_empty(&ei->i_completed_io_list)){
3700 io = list_entry(ei->i_completed_io_list.next,
3701 ext4_io_end_t, list);
3703 * Calling ext4_end_io_nolock() to convert completed
3706 * When ext4_sync_file() is called, run_queue() may already
3707 * about to flush the work corresponding to this io structure.
3708 * It will be upset if it founds the io structure related
3709 * to the work-to-be schedule is freed.
3711 * Thus we need to keep the io structure still valid here after
3712 * convertion finished. The io structure has a flag to
3713 * avoid double converting from both fsync and background work
3716 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3717 ret = ext4_end_io_nolock(io);
3718 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3722 list_del_init(&io->list);
3724 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3725 return (ret2 < 0) ? ret2 : 0;
3728 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3730 ext4_io_end_t *io = NULL;
3732 io = kmalloc(sizeof(*io), flags);
3741 INIT_WORK(&io->work, ext4_end_io_work);
3742 INIT_LIST_HEAD(&io->list);
3748 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3749 ssize_t size, void *private)
3751 ext4_io_end_t *io_end = iocb->private;
3752 struct workqueue_struct *wq;
3753 unsigned long flags;
3754 struct ext4_inode_info *ei;
3756 /* if not async direct IO or dio with 0 bytes write, just return */
3757 if (!io_end || !size)
3760 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3761 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3762 iocb->private, io_end->inode->i_ino, iocb, offset,
3765 /* if not aio dio with unwritten extents, just free io and return */
3766 if (io_end->flag != EXT4_IO_UNWRITTEN){
3767 ext4_free_io_end(io_end);
3768 iocb->private = NULL;
3772 io_end->offset = offset;
3773 io_end->size = size;
3774 io_end->flag = EXT4_IO_UNWRITTEN;
3775 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3777 /* queue the work to convert unwritten extents to written */
3778 queue_work(wq, &io_end->work);
3780 /* Add the io_end to per-inode completed aio dio list*/
3781 ei = EXT4_I(io_end->inode);
3782 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3783 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3784 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3785 iocb->private = NULL;
3788 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3790 ext4_io_end_t *io_end = bh->b_private;
3791 struct workqueue_struct *wq;
3792 struct inode *inode;
3793 unsigned long flags;
3795 if (!test_clear_buffer_uninit(bh) || !io_end)
3798 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3799 printk("sb umounted, discard end_io request for inode %lu\n",
3800 io_end->inode->i_ino);
3801 ext4_free_io_end(io_end);
3805 io_end->flag = EXT4_IO_UNWRITTEN;
3806 inode = io_end->inode;
3808 /* Add the io_end to per-inode completed io list*/
3809 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3810 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3811 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3813 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3814 /* queue the work to convert unwritten extents to written */
3815 queue_work(wq, &io_end->work);
3817 bh->b_private = NULL;
3818 bh->b_end_io = NULL;
3819 clear_buffer_uninit(bh);
3820 end_buffer_async_write(bh, uptodate);
3823 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3825 ext4_io_end_t *io_end;
3826 struct page *page = bh->b_page;
3827 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3828 size_t size = bh->b_size;
3831 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3833 if (printk_ratelimit())
3834 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3838 io_end->offset = offset;
3839 io_end->size = size;
3841 * We need to hold a reference to the page to make sure it
3842 * doesn't get evicted before ext4_end_io_work() has a chance
3843 * to convert the extent from written to unwritten.
3845 io_end->page = page;
3846 get_page(io_end->page);
3848 bh->b_private = io_end;
3849 bh->b_end_io = ext4_end_io_buffer_write;
3854 * For ext4 extent files, ext4 will do direct-io write to holes,
3855 * preallocated extents, and those write extend the file, no need to
3856 * fall back to buffered IO.
3858 * For holes, we fallocate those blocks, mark them as unintialized
3859 * If those blocks were preallocated, we mark sure they are splited, but
3860 * still keep the range to write as unintialized.
3862 * The unwrritten extents will be converted to written when DIO is completed.
3863 * For async direct IO, since the IO may still pending when return, we
3864 * set up an end_io call back function, which will do the convertion
3865 * when async direct IO completed.
3867 * If the O_DIRECT write will extend the file then add this inode to the
3868 * orphan list. So recovery will truncate it back to the original size
3869 * if the machine crashes during the write.
3872 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3873 const struct iovec *iov, loff_t offset,
3874 unsigned long nr_segs)
3876 struct file *file = iocb->ki_filp;
3877 struct inode *inode = file->f_mapping->host;
3879 size_t count = iov_length(iov, nr_segs);
3881 loff_t final_size = offset + count;
3882 if (rw == WRITE && final_size <= inode->i_size) {
3884 * We could direct write to holes and fallocate.
3886 * Allocated blocks to fill the hole are marked as uninitialized
3887 * to prevent paralel buffered read to expose the stale data
3888 * before DIO complete the data IO.
3890 * As to previously fallocated extents, ext4 get_block
3891 * will just simply mark the buffer mapped but still
3892 * keep the extents uninitialized.
3894 * for non AIO case, we will convert those unwritten extents
3895 * to written after return back from blockdev_direct_IO.
3897 * for async DIO, the conversion needs to be defered when
3898 * the IO is completed. The ext4 end_io callback function
3899 * will be called to take care of the conversion work.
3900 * Here for async case, we allocate an io_end structure to
3903 iocb->private = NULL;
3904 EXT4_I(inode)->cur_aio_dio = NULL;
3905 if (!is_sync_kiocb(iocb)) {
3906 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3910 * we save the io structure for current async
3911 * direct IO, so that later ext4_get_blocks()
3912 * could flag the io structure whether there
3913 * is a unwritten extents needs to be converted
3914 * when IO is completed.
3916 EXT4_I(inode)->cur_aio_dio = iocb->private;
3919 ret = blockdev_direct_IO(rw, iocb, inode,
3920 inode->i_sb->s_bdev, iov,
3922 ext4_get_block_write,
3925 EXT4_I(inode)->cur_aio_dio = NULL;
3927 * The io_end structure takes a reference to the inode,
3928 * that structure needs to be destroyed and the
3929 * reference to the inode need to be dropped, when IO is
3930 * complete, even with 0 byte write, or failed.
3932 * In the successful AIO DIO case, the io_end structure will be
3933 * desctroyed and the reference to the inode will be dropped
3934 * after the end_io call back function is called.
3936 * In the case there is 0 byte write, or error case, since
3937 * VFS direct IO won't invoke the end_io call back function,
3938 * we need to free the end_io structure here.
3940 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3941 ext4_free_io_end(iocb->private);
3942 iocb->private = NULL;
3943 } else if (ret > 0 && ext4_test_inode_state(inode,
3944 EXT4_STATE_DIO_UNWRITTEN)) {
3947 * for non AIO case, since the IO is already
3948 * completed, we could do the convertion right here
3950 err = ext4_convert_unwritten_extents(inode,
3954 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3959 /* for write the the end of file case, we fall back to old way */
3960 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3963 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3964 const struct iovec *iov, loff_t offset,
3965 unsigned long nr_segs)
3967 struct file *file = iocb->ki_filp;
3968 struct inode *inode = file->f_mapping->host;
3970 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3971 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3973 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3977 * Pages can be marked dirty completely asynchronously from ext4's journalling
3978 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3979 * much here because ->set_page_dirty is called under VFS locks. The page is
3980 * not necessarily locked.
3982 * We cannot just dirty the page and leave attached buffers clean, because the
3983 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3984 * or jbddirty because all the journalling code will explode.
3986 * So what we do is to mark the page "pending dirty" and next time writepage
3987 * is called, propagate that into the buffers appropriately.
3989 static int ext4_journalled_set_page_dirty(struct page *page)
3991 SetPageChecked(page);
3992 return __set_page_dirty_nobuffers(page);
3995 static const struct address_space_operations ext4_ordered_aops = {
3996 .readpage = ext4_readpage,
3997 .readpages = ext4_readpages,
3998 .writepage = ext4_writepage,
3999 .sync_page = block_sync_page,
4000 .write_begin = ext4_write_begin,
4001 .write_end = ext4_ordered_write_end,
4003 .invalidatepage = ext4_invalidatepage,
4004 .releasepage = ext4_releasepage,
4005 .direct_IO = ext4_direct_IO,
4006 .migratepage = buffer_migrate_page,
4007 .is_partially_uptodate = block_is_partially_uptodate,
4008 .error_remove_page = generic_error_remove_page,
4011 static const struct address_space_operations ext4_writeback_aops = {
4012 .readpage = ext4_readpage,
4013 .readpages = ext4_readpages,
4014 .writepage = ext4_writepage,
4015 .sync_page = block_sync_page,
4016 .write_begin = ext4_write_begin,
4017 .write_end = ext4_writeback_write_end,
4019 .invalidatepage = ext4_invalidatepage,
4020 .releasepage = ext4_releasepage,
4021 .direct_IO = ext4_direct_IO,
4022 .migratepage = buffer_migrate_page,
4023 .is_partially_uptodate = block_is_partially_uptodate,
4024 .error_remove_page = generic_error_remove_page,
4027 static const struct address_space_operations ext4_journalled_aops = {
4028 .readpage = ext4_readpage,
4029 .readpages = ext4_readpages,
4030 .writepage = ext4_writepage,
4031 .sync_page = block_sync_page,
4032 .write_begin = ext4_write_begin,
4033 .write_end = ext4_journalled_write_end,
4034 .set_page_dirty = ext4_journalled_set_page_dirty,
4036 .invalidatepage = ext4_invalidatepage,
4037 .releasepage = ext4_releasepage,
4038 .is_partially_uptodate = block_is_partially_uptodate,
4039 .error_remove_page = generic_error_remove_page,
4042 static const struct address_space_operations ext4_da_aops = {
4043 .readpage = ext4_readpage,
4044 .readpages = ext4_readpages,
4045 .writepage = ext4_writepage,
4046 .writepages = ext4_da_writepages,
4047 .sync_page = block_sync_page,
4048 .write_begin = ext4_da_write_begin,
4049 .write_end = ext4_da_write_end,
4051 .invalidatepage = ext4_da_invalidatepage,
4052 .releasepage = ext4_releasepage,
4053 .direct_IO = ext4_direct_IO,
4054 .migratepage = buffer_migrate_page,
4055 .is_partially_uptodate = block_is_partially_uptodate,
4056 .error_remove_page = generic_error_remove_page,
4059 void ext4_set_aops(struct inode *inode)
4061 if (ext4_should_order_data(inode) &&
4062 test_opt(inode->i_sb, DELALLOC))
4063 inode->i_mapping->a_ops = &ext4_da_aops;
4064 else if (ext4_should_order_data(inode))
4065 inode->i_mapping->a_ops = &ext4_ordered_aops;
4066 else if (ext4_should_writeback_data(inode) &&
4067 test_opt(inode->i_sb, DELALLOC))
4068 inode->i_mapping->a_ops = &ext4_da_aops;
4069 else if (ext4_should_writeback_data(inode))
4070 inode->i_mapping->a_ops = &ext4_writeback_aops;
4072 inode->i_mapping->a_ops = &ext4_journalled_aops;
4076 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4077 * up to the end of the block which corresponds to `from'.
4078 * This required during truncate. We need to physically zero the tail end
4079 * of that block so it doesn't yield old data if the file is later grown.
4081 int ext4_block_truncate_page(handle_t *handle,
4082 struct address_space *mapping, loff_t from)
4084 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4085 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4086 unsigned blocksize, length, pos;
4088 struct inode *inode = mapping->host;
4089 struct buffer_head *bh;
4093 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4094 mapping_gfp_mask(mapping) & ~__GFP_FS);
4098 blocksize = inode->i_sb->s_blocksize;
4099 length = blocksize - (offset & (blocksize - 1));
4100 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4103 * For "nobh" option, we can only work if we don't need to
4104 * read-in the page - otherwise we create buffers to do the IO.
4106 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
4107 ext4_should_writeback_data(inode) && PageUptodate(page)) {
4108 zero_user(page, offset, length);
4109 set_page_dirty(page);
4113 if (!page_has_buffers(page))
4114 create_empty_buffers(page, blocksize, 0);
4116 /* Find the buffer that contains "offset" */
4117 bh = page_buffers(page);
4119 while (offset >= pos) {
4120 bh = bh->b_this_page;
4126 if (buffer_freed(bh)) {
4127 BUFFER_TRACE(bh, "freed: skip");
4131 if (!buffer_mapped(bh)) {
4132 BUFFER_TRACE(bh, "unmapped");
4133 ext4_get_block(inode, iblock, bh, 0);
4134 /* unmapped? It's a hole - nothing to do */
4135 if (!buffer_mapped(bh)) {
4136 BUFFER_TRACE(bh, "still unmapped");
4141 /* Ok, it's mapped. Make sure it's up-to-date */
4142 if (PageUptodate(page))
4143 set_buffer_uptodate(bh);
4145 if (!buffer_uptodate(bh)) {
4147 ll_rw_block(READ, 1, &bh);
4149 /* Uhhuh. Read error. Complain and punt. */
4150 if (!buffer_uptodate(bh))
4154 if (ext4_should_journal_data(inode)) {
4155 BUFFER_TRACE(bh, "get write access");
4156 err = ext4_journal_get_write_access(handle, bh);
4161 zero_user(page, offset, length);
4163 BUFFER_TRACE(bh, "zeroed end of block");
4166 if (ext4_should_journal_data(inode)) {
4167 err = ext4_handle_dirty_metadata(handle, inode, bh);
4169 if (ext4_should_order_data(inode))
4170 err = ext4_jbd2_file_inode(handle, inode);
4171 mark_buffer_dirty(bh);
4176 page_cache_release(page);
4181 * Probably it should be a library function... search for first non-zero word
4182 * or memcmp with zero_page, whatever is better for particular architecture.
4185 static inline int all_zeroes(__le32 *p, __le32 *q)
4194 * ext4_find_shared - find the indirect blocks for partial truncation.
4195 * @inode: inode in question
4196 * @depth: depth of the affected branch
4197 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4198 * @chain: place to store the pointers to partial indirect blocks
4199 * @top: place to the (detached) top of branch
4201 * This is a helper function used by ext4_truncate().
4203 * When we do truncate() we may have to clean the ends of several
4204 * indirect blocks but leave the blocks themselves alive. Block is
4205 * partially truncated if some data below the new i_size is refered
4206 * from it (and it is on the path to the first completely truncated
4207 * data block, indeed). We have to free the top of that path along
4208 * with everything to the right of the path. Since no allocation
4209 * past the truncation point is possible until ext4_truncate()
4210 * finishes, we may safely do the latter, but top of branch may
4211 * require special attention - pageout below the truncation point
4212 * might try to populate it.
4214 * We atomically detach the top of branch from the tree, store the
4215 * block number of its root in *@top, pointers to buffer_heads of
4216 * partially truncated blocks - in @chain[].bh and pointers to
4217 * their last elements that should not be removed - in
4218 * @chain[].p. Return value is the pointer to last filled element
4221 * The work left to caller to do the actual freeing of subtrees:
4222 * a) free the subtree starting from *@top
4223 * b) free the subtrees whose roots are stored in
4224 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4225 * c) free the subtrees growing from the inode past the @chain[0].
4226 * (no partially truncated stuff there). */
4228 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4229 ext4_lblk_t offsets[4], Indirect chain[4],
4232 Indirect *partial, *p;
4236 /* Make k index the deepest non-null offset + 1 */
4237 for (k = depth; k > 1 && !offsets[k-1]; k--)
4239 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4240 /* Writer: pointers */
4242 partial = chain + k-1;
4244 * If the branch acquired continuation since we've looked at it -
4245 * fine, it should all survive and (new) top doesn't belong to us.
4247 if (!partial->key && *partial->p)
4250 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4253 * OK, we've found the last block that must survive. The rest of our
4254 * branch should be detached before unlocking. However, if that rest
4255 * of branch is all ours and does not grow immediately from the inode
4256 * it's easier to cheat and just decrement partial->p.
4258 if (p == chain + k - 1 && p > chain) {
4262 /* Nope, don't do this in ext4. Must leave the tree intact */
4269 while (partial > p) {
4270 brelse(partial->bh);
4278 * Zero a number of block pointers in either an inode or an indirect block.
4279 * If we restart the transaction we must again get write access to the
4280 * indirect block for further modification.
4282 * We release `count' blocks on disk, but (last - first) may be greater
4283 * than `count' because there can be holes in there.
4285 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4286 struct buffer_head *bh,
4287 ext4_fsblk_t block_to_free,
4288 unsigned long count, __le32 *first,
4292 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4294 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4295 flags |= EXT4_FREE_BLOCKS_METADATA;
4297 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4299 ext4_error(inode->i_sb, "inode #%lu: "
4300 "attempt to clear blocks %llu len %lu, invalid",
4301 inode->i_ino, (unsigned long long) block_to_free,
4306 if (try_to_extend_transaction(handle, inode)) {
4308 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4309 ext4_handle_dirty_metadata(handle, inode, bh);
4311 ext4_mark_inode_dirty(handle, inode);
4312 ext4_truncate_restart_trans(handle, inode,
4313 blocks_for_truncate(inode));
4315 BUFFER_TRACE(bh, "retaking write access");
4316 ext4_journal_get_write_access(handle, bh);
4320 for (p = first; p < last; p++)
4323 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4328 * ext4_free_data - free a list of data blocks
4329 * @handle: handle for this transaction
4330 * @inode: inode we are dealing with
4331 * @this_bh: indirect buffer_head which contains *@first and *@last
4332 * @first: array of block numbers
4333 * @last: points immediately past the end of array
4335 * We are freeing all blocks refered from that array (numbers are stored as
4336 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4338 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4339 * blocks are contiguous then releasing them at one time will only affect one
4340 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4341 * actually use a lot of journal space.
4343 * @this_bh will be %NULL if @first and @last point into the inode's direct
4346 static void ext4_free_data(handle_t *handle, struct inode *inode,
4347 struct buffer_head *this_bh,
4348 __le32 *first, __le32 *last)
4350 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4351 unsigned long count = 0; /* Number of blocks in the run */
4352 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4355 ext4_fsblk_t nr; /* Current block # */
4356 __le32 *p; /* Pointer into inode/ind
4357 for current block */
4360 if (this_bh) { /* For indirect block */
4361 BUFFER_TRACE(this_bh, "get_write_access");
4362 err = ext4_journal_get_write_access(handle, this_bh);
4363 /* Important: if we can't update the indirect pointers
4364 * to the blocks, we can't free them. */
4369 for (p = first; p < last; p++) {
4370 nr = le32_to_cpu(*p);
4372 /* accumulate blocks to free if they're contiguous */
4375 block_to_free_p = p;
4377 } else if (nr == block_to_free + count) {
4380 if (ext4_clear_blocks(handle, inode, this_bh,
4381 block_to_free, count,
4382 block_to_free_p, p))
4385 block_to_free_p = p;
4392 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4393 count, block_to_free_p, p);
4396 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4399 * The buffer head should have an attached journal head at this
4400 * point. However, if the data is corrupted and an indirect
4401 * block pointed to itself, it would have been detached when
4402 * the block was cleared. Check for this instead of OOPSing.
4404 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4405 ext4_handle_dirty_metadata(handle, inode, this_bh);
4407 ext4_error(inode->i_sb,
4408 "circular indirect block detected, "
4409 "inode=%lu, block=%llu",
4411 (unsigned long long) this_bh->b_blocknr);
4416 * ext4_free_branches - free an array of branches
4417 * @handle: JBD handle for this transaction
4418 * @inode: inode we are dealing with
4419 * @parent_bh: the buffer_head which contains *@first and *@last
4420 * @first: array of block numbers
4421 * @last: pointer immediately past the end of array
4422 * @depth: depth of the branches to free
4424 * We are freeing all blocks refered from these branches (numbers are
4425 * stored as little-endian 32-bit) and updating @inode->i_blocks
4428 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4429 struct buffer_head *parent_bh,
4430 __le32 *first, __le32 *last, int depth)
4435 if (ext4_handle_is_aborted(handle))
4439 struct buffer_head *bh;
4440 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4442 while (--p >= first) {
4443 nr = le32_to_cpu(*p);
4445 continue; /* A hole */
4447 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4449 ext4_error(inode->i_sb,
4450 "indirect mapped block in inode "
4451 "#%lu invalid (level %d, blk #%lu)",
4452 inode->i_ino, depth,
4453 (unsigned long) nr);
4457 /* Go read the buffer for the next level down */
4458 bh = sb_bread(inode->i_sb, nr);
4461 * A read failure? Report error and clear slot
4465 ext4_error(inode->i_sb,
4466 "Read failure, inode=%lu, block=%llu",
4471 /* This zaps the entire block. Bottom up. */
4472 BUFFER_TRACE(bh, "free child branches");
4473 ext4_free_branches(handle, inode, bh,
4474 (__le32 *) bh->b_data,
4475 (__le32 *) bh->b_data + addr_per_block,
4479 * We've probably journalled the indirect block several
4480 * times during the truncate. But it's no longer
4481 * needed and we now drop it from the transaction via
4482 * jbd2_journal_revoke().
4484 * That's easy if it's exclusively part of this
4485 * transaction. But if it's part of the committing
4486 * transaction then jbd2_journal_forget() will simply
4487 * brelse() it. That means that if the underlying
4488 * block is reallocated in ext4_get_block(),
4489 * unmap_underlying_metadata() will find this block
4490 * and will try to get rid of it. damn, damn.
4492 * If this block has already been committed to the
4493 * journal, a revoke record will be written. And
4494 * revoke records must be emitted *before* clearing
4495 * this block's bit in the bitmaps.
4497 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
4500 * Everything below this this pointer has been
4501 * released. Now let this top-of-subtree go.
4503 * We want the freeing of this indirect block to be
4504 * atomic in the journal with the updating of the
4505 * bitmap block which owns it. So make some room in
4508 * We zero the parent pointer *after* freeing its
4509 * pointee in the bitmaps, so if extend_transaction()
4510 * for some reason fails to put the bitmap changes and
4511 * the release into the same transaction, recovery
4512 * will merely complain about releasing a free block,
4513 * rather than leaking blocks.
4515 if (ext4_handle_is_aborted(handle))
4517 if (try_to_extend_transaction(handle, inode)) {
4518 ext4_mark_inode_dirty(handle, inode);
4519 ext4_truncate_restart_trans(handle, inode,
4520 blocks_for_truncate(inode));
4523 ext4_free_blocks(handle, inode, 0, nr, 1,
4524 EXT4_FREE_BLOCKS_METADATA);
4528 * The block which we have just freed is
4529 * pointed to by an indirect block: journal it
4531 BUFFER_TRACE(parent_bh, "get_write_access");
4532 if (!ext4_journal_get_write_access(handle,
4535 BUFFER_TRACE(parent_bh,
4536 "call ext4_handle_dirty_metadata");
4537 ext4_handle_dirty_metadata(handle,
4544 /* We have reached the bottom of the tree. */
4545 BUFFER_TRACE(parent_bh, "free data blocks");
4546 ext4_free_data(handle, inode, parent_bh, first, last);
4550 int ext4_can_truncate(struct inode *inode)
4552 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4554 if (S_ISREG(inode->i_mode))
4556 if (S_ISDIR(inode->i_mode))
4558 if (S_ISLNK(inode->i_mode))
4559 return !ext4_inode_is_fast_symlink(inode);
4566 * We block out ext4_get_block() block instantiations across the entire
4567 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4568 * simultaneously on behalf of the same inode.
4570 * As we work through the truncate and commmit bits of it to the journal there
4571 * is one core, guiding principle: the file's tree must always be consistent on
4572 * disk. We must be able to restart the truncate after a crash.
4574 * The file's tree may be transiently inconsistent in memory (although it
4575 * probably isn't), but whenever we close off and commit a journal transaction,
4576 * the contents of (the filesystem + the journal) must be consistent and
4577 * restartable. It's pretty simple, really: bottom up, right to left (although
4578 * left-to-right works OK too).
4580 * Note that at recovery time, journal replay occurs *before* the restart of
4581 * truncate against the orphan inode list.
4583 * The committed inode has the new, desired i_size (which is the same as
4584 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4585 * that this inode's truncate did not complete and it will again call
4586 * ext4_truncate() to have another go. So there will be instantiated blocks
4587 * to the right of the truncation point in a crashed ext4 filesystem. But
4588 * that's fine - as long as they are linked from the inode, the post-crash
4589 * ext4_truncate() run will find them and release them.
4591 void ext4_truncate(struct inode *inode)
4594 struct ext4_inode_info *ei = EXT4_I(inode);
4595 __le32 *i_data = ei->i_data;
4596 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4597 struct address_space *mapping = inode->i_mapping;
4598 ext4_lblk_t offsets[4];
4603 ext4_lblk_t last_block;
4604 unsigned blocksize = inode->i_sb->s_blocksize;
4606 if (!ext4_can_truncate(inode))
4609 EXT4_I(inode)->i_flags &= ~EXT4_EOFBLOCKS_FL;
4611 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4612 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4614 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4615 ext4_ext_truncate(inode);
4619 handle = start_transaction(inode);
4621 return; /* AKPM: return what? */
4623 last_block = (inode->i_size + blocksize-1)
4624 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4626 if (inode->i_size & (blocksize - 1))
4627 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4630 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4632 goto out_stop; /* error */
4635 * OK. This truncate is going to happen. We add the inode to the
4636 * orphan list, so that if this truncate spans multiple transactions,
4637 * and we crash, we will resume the truncate when the filesystem
4638 * recovers. It also marks the inode dirty, to catch the new size.
4640 * Implication: the file must always be in a sane, consistent
4641 * truncatable state while each transaction commits.
4643 if (ext4_orphan_add(handle, inode))
4647 * From here we block out all ext4_get_block() callers who want to
4648 * modify the block allocation tree.
4650 down_write(&ei->i_data_sem);
4652 ext4_discard_preallocations(inode);
4655 * The orphan list entry will now protect us from any crash which
4656 * occurs before the truncate completes, so it is now safe to propagate
4657 * the new, shorter inode size (held for now in i_size) into the
4658 * on-disk inode. We do this via i_disksize, which is the value which
4659 * ext4 *really* writes onto the disk inode.
4661 ei->i_disksize = inode->i_size;
4663 if (n == 1) { /* direct blocks */
4664 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4665 i_data + EXT4_NDIR_BLOCKS);
4669 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4670 /* Kill the top of shared branch (not detached) */
4672 if (partial == chain) {
4673 /* Shared branch grows from the inode */
4674 ext4_free_branches(handle, inode, NULL,
4675 &nr, &nr+1, (chain+n-1) - partial);
4678 * We mark the inode dirty prior to restart,
4679 * and prior to stop. No need for it here.
4682 /* Shared branch grows from an indirect block */
4683 BUFFER_TRACE(partial->bh, "get_write_access");
4684 ext4_free_branches(handle, inode, partial->bh,
4686 partial->p+1, (chain+n-1) - partial);
4689 /* Clear the ends of indirect blocks on the shared branch */
4690 while (partial > chain) {
4691 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4692 (__le32*)partial->bh->b_data+addr_per_block,
4693 (chain+n-1) - partial);
4694 BUFFER_TRACE(partial->bh, "call brelse");
4695 brelse(partial->bh);
4699 /* Kill the remaining (whole) subtrees */
4700 switch (offsets[0]) {
4702 nr = i_data[EXT4_IND_BLOCK];
4704 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4705 i_data[EXT4_IND_BLOCK] = 0;
4707 case EXT4_IND_BLOCK:
4708 nr = i_data[EXT4_DIND_BLOCK];
4710 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4711 i_data[EXT4_DIND_BLOCK] = 0;
4713 case EXT4_DIND_BLOCK:
4714 nr = i_data[EXT4_TIND_BLOCK];
4716 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4717 i_data[EXT4_TIND_BLOCK] = 0;
4719 case EXT4_TIND_BLOCK:
4723 up_write(&ei->i_data_sem);
4724 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4725 ext4_mark_inode_dirty(handle, inode);
4728 * In a multi-transaction truncate, we only make the final transaction
4732 ext4_handle_sync(handle);
4735 * If this was a simple ftruncate(), and the file will remain alive
4736 * then we need to clear up the orphan record which we created above.
4737 * However, if this was a real unlink then we were called by
4738 * ext4_delete_inode(), and we allow that function to clean up the
4739 * orphan info for us.
4742 ext4_orphan_del(handle, inode);
4744 ext4_journal_stop(handle);
4748 * ext4_get_inode_loc returns with an extra refcount against the inode's
4749 * underlying buffer_head on success. If 'in_mem' is true, we have all
4750 * data in memory that is needed to recreate the on-disk version of this
4753 static int __ext4_get_inode_loc(struct inode *inode,
4754 struct ext4_iloc *iloc, int in_mem)
4756 struct ext4_group_desc *gdp;
4757 struct buffer_head *bh;
4758 struct super_block *sb = inode->i_sb;
4760 int inodes_per_block, inode_offset;
4763 if (!ext4_valid_inum(sb, inode->i_ino))
4766 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4767 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4772 * Figure out the offset within the block group inode table
4774 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4775 inode_offset = ((inode->i_ino - 1) %
4776 EXT4_INODES_PER_GROUP(sb));
4777 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4778 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4780 bh = sb_getblk(sb, block);
4782 ext4_error(sb, "unable to read inode block - "
4783 "inode=%lu, block=%llu", inode->i_ino, block);
4786 if (!buffer_uptodate(bh)) {
4790 * If the buffer has the write error flag, we have failed
4791 * to write out another inode in the same block. In this
4792 * case, we don't have to read the block because we may
4793 * read the old inode data successfully.
4795 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4796 set_buffer_uptodate(bh);
4798 if (buffer_uptodate(bh)) {
4799 /* someone brought it uptodate while we waited */
4805 * If we have all information of the inode in memory and this
4806 * is the only valid inode in the block, we need not read the
4810 struct buffer_head *bitmap_bh;
4813 start = inode_offset & ~(inodes_per_block - 1);
4815 /* Is the inode bitmap in cache? */
4816 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4821 * If the inode bitmap isn't in cache then the
4822 * optimisation may end up performing two reads instead
4823 * of one, so skip it.
4825 if (!buffer_uptodate(bitmap_bh)) {
4829 for (i = start; i < start + inodes_per_block; i++) {
4830 if (i == inode_offset)
4832 if (ext4_test_bit(i, bitmap_bh->b_data))
4836 if (i == start + inodes_per_block) {
4837 /* all other inodes are free, so skip I/O */
4838 memset(bh->b_data, 0, bh->b_size);
4839 set_buffer_uptodate(bh);
4847 * If we need to do any I/O, try to pre-readahead extra
4848 * blocks from the inode table.
4850 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4851 ext4_fsblk_t b, end, table;
4854 table = ext4_inode_table(sb, gdp);
4855 /* s_inode_readahead_blks is always a power of 2 */
4856 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4859 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4860 num = EXT4_INODES_PER_GROUP(sb);
4861 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4862 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4863 num -= ext4_itable_unused_count(sb, gdp);
4864 table += num / inodes_per_block;
4868 sb_breadahead(sb, b++);
4872 * There are other valid inodes in the buffer, this inode
4873 * has in-inode xattrs, or we don't have this inode in memory.
4874 * Read the block from disk.
4877 bh->b_end_io = end_buffer_read_sync;
4878 submit_bh(READ_META, bh);
4880 if (!buffer_uptodate(bh)) {
4881 ext4_error(sb, "unable to read inode block - inode=%lu,"
4882 " block=%llu", inode->i_ino, block);
4892 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4894 /* We have all inode data except xattrs in memory here. */
4895 return __ext4_get_inode_loc(inode, iloc,
4896 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4899 void ext4_set_inode_flags(struct inode *inode)
4901 unsigned int flags = EXT4_I(inode)->i_flags;
4903 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4904 if (flags & EXT4_SYNC_FL)
4905 inode->i_flags |= S_SYNC;
4906 if (flags & EXT4_APPEND_FL)
4907 inode->i_flags |= S_APPEND;
4908 if (flags & EXT4_IMMUTABLE_FL)
4909 inode->i_flags |= S_IMMUTABLE;
4910 if (flags & EXT4_NOATIME_FL)
4911 inode->i_flags |= S_NOATIME;
4912 if (flags & EXT4_DIRSYNC_FL)
4913 inode->i_flags |= S_DIRSYNC;
4916 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4917 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4919 unsigned int flags = ei->vfs_inode.i_flags;
4921 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4922 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4924 ei->i_flags |= EXT4_SYNC_FL;
4925 if (flags & S_APPEND)
4926 ei->i_flags |= EXT4_APPEND_FL;
4927 if (flags & S_IMMUTABLE)
4928 ei->i_flags |= EXT4_IMMUTABLE_FL;
4929 if (flags & S_NOATIME)
4930 ei->i_flags |= EXT4_NOATIME_FL;
4931 if (flags & S_DIRSYNC)
4932 ei->i_flags |= EXT4_DIRSYNC_FL;
4935 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4936 struct ext4_inode_info *ei)
4939 struct inode *inode = &(ei->vfs_inode);
4940 struct super_block *sb = inode->i_sb;
4942 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4943 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4944 /* we are using combined 48 bit field */
4945 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4946 le32_to_cpu(raw_inode->i_blocks_lo);
4947 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4948 /* i_blocks represent file system block size */
4949 return i_blocks << (inode->i_blkbits - 9);
4954 return le32_to_cpu(raw_inode->i_blocks_lo);
4958 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4960 struct ext4_iloc iloc;
4961 struct ext4_inode *raw_inode;
4962 struct ext4_inode_info *ei;
4963 struct inode *inode;
4964 journal_t *journal = EXT4_SB(sb)->s_journal;
4968 inode = iget_locked(sb, ino);
4970 return ERR_PTR(-ENOMEM);
4971 if (!(inode->i_state & I_NEW))
4977 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4980 raw_inode = ext4_raw_inode(&iloc);
4981 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4982 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4983 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4984 if (!(test_opt(inode->i_sb, NO_UID32))) {
4985 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4986 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4988 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4990 ei->i_state_flags = 0;
4991 ei->i_dir_start_lookup = 0;
4992 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4993 /* We now have enough fields to check if the inode was active or not.
4994 * This is needed because nfsd might try to access dead inodes
4995 * the test is that same one that e2fsck uses
4996 * NeilBrown 1999oct15
4998 if (inode->i_nlink == 0) {
4999 if (inode->i_mode == 0 ||
5000 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5001 /* this inode is deleted */
5005 /* The only unlinked inodes we let through here have
5006 * valid i_mode and are being read by the orphan
5007 * recovery code: that's fine, we're about to complete
5008 * the process of deleting those. */
5010 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5011 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5012 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5013 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5015 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5016 inode->i_size = ext4_isize(raw_inode);
5017 ei->i_disksize = inode->i_size;
5019 ei->i_reserved_quota = 0;
5021 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5022 ei->i_block_group = iloc.block_group;
5023 ei->i_last_alloc_group = ~0;
5025 * NOTE! The in-memory inode i_data array is in little-endian order
5026 * even on big-endian machines: we do NOT byteswap the block numbers!
5028 for (block = 0; block < EXT4_N_BLOCKS; block++)
5029 ei->i_data[block] = raw_inode->i_block[block];
5030 INIT_LIST_HEAD(&ei->i_orphan);
5033 * Set transaction id's of transactions that have to be committed
5034 * to finish f[data]sync. We set them to currently running transaction
5035 * as we cannot be sure that the inode or some of its metadata isn't
5036 * part of the transaction - the inode could have been reclaimed and
5037 * now it is reread from disk.
5040 transaction_t *transaction;
5043 spin_lock(&journal->j_state_lock);
5044 if (journal->j_running_transaction)
5045 transaction = journal->j_running_transaction;
5047 transaction = journal->j_committing_transaction;
5049 tid = transaction->t_tid;
5051 tid = journal->j_commit_sequence;
5052 spin_unlock(&journal->j_state_lock);
5053 ei->i_sync_tid = tid;
5054 ei->i_datasync_tid = tid;
5057 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5058 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5059 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5060 EXT4_INODE_SIZE(inode->i_sb)) {
5064 if (ei->i_extra_isize == 0) {
5065 /* The extra space is currently unused. Use it. */
5066 ei->i_extra_isize = sizeof(struct ext4_inode) -
5067 EXT4_GOOD_OLD_INODE_SIZE;
5069 __le32 *magic = (void *)raw_inode +
5070 EXT4_GOOD_OLD_INODE_SIZE +
5072 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5073 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5076 ei->i_extra_isize = 0;
5078 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5079 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5080 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5081 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5083 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5084 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5085 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5087 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5091 if (ei->i_file_acl &&
5092 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5093 ext4_error(sb, "bad extended attribute block %llu inode #%lu",
5094 ei->i_file_acl, inode->i_ino);
5097 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
5098 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5099 (S_ISLNK(inode->i_mode) &&
5100 !ext4_inode_is_fast_symlink(inode)))
5101 /* Validate extent which is part of inode */
5102 ret = ext4_ext_check_inode(inode);
5103 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5104 (S_ISLNK(inode->i_mode) &&
5105 !ext4_inode_is_fast_symlink(inode))) {
5106 /* Validate block references which are part of inode */
5107 ret = ext4_check_inode_blockref(inode);
5112 if (S_ISREG(inode->i_mode)) {
5113 inode->i_op = &ext4_file_inode_operations;
5114 inode->i_fop = &ext4_file_operations;
5115 ext4_set_aops(inode);
5116 } else if (S_ISDIR(inode->i_mode)) {
5117 inode->i_op = &ext4_dir_inode_operations;
5118 inode->i_fop = &ext4_dir_operations;
5119 } else if (S_ISLNK(inode->i_mode)) {
5120 if (ext4_inode_is_fast_symlink(inode)) {
5121 inode->i_op = &ext4_fast_symlink_inode_operations;
5122 nd_terminate_link(ei->i_data, inode->i_size,
5123 sizeof(ei->i_data) - 1);
5125 inode->i_op = &ext4_symlink_inode_operations;
5126 ext4_set_aops(inode);
5128 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5129 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5130 inode->i_op = &ext4_special_inode_operations;
5131 if (raw_inode->i_block[0])
5132 init_special_inode(inode, inode->i_mode,
5133 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5135 init_special_inode(inode, inode->i_mode,
5136 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5139 ext4_error(inode->i_sb, "bogus i_mode (%o) for inode=%lu",
5140 inode->i_mode, inode->i_ino);
5144 ext4_set_inode_flags(inode);
5145 unlock_new_inode(inode);
5151 return ERR_PTR(ret);
5154 static int ext4_inode_blocks_set(handle_t *handle,
5155 struct ext4_inode *raw_inode,
5156 struct ext4_inode_info *ei)
5158 struct inode *inode = &(ei->vfs_inode);
5159 u64 i_blocks = inode->i_blocks;
5160 struct super_block *sb = inode->i_sb;
5162 if (i_blocks <= ~0U) {
5164 * i_blocks can be represnted in a 32 bit variable
5165 * as multiple of 512 bytes
5167 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5168 raw_inode->i_blocks_high = 0;
5169 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
5172 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5175 if (i_blocks <= 0xffffffffffffULL) {
5177 * i_blocks can be represented in a 48 bit variable
5178 * as multiple of 512 bytes
5180 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5181 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5182 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
5184 ei->i_flags |= EXT4_HUGE_FILE_FL;
5185 /* i_block is stored in file system block size */
5186 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5187 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5188 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5194 * Post the struct inode info into an on-disk inode location in the
5195 * buffer-cache. This gobbles the caller's reference to the
5196 * buffer_head in the inode location struct.
5198 * The caller must have write access to iloc->bh.
5200 static int ext4_do_update_inode(handle_t *handle,
5201 struct inode *inode,
5202 struct ext4_iloc *iloc)
5204 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5205 struct ext4_inode_info *ei = EXT4_I(inode);
5206 struct buffer_head *bh = iloc->bh;
5207 int err = 0, rc, block;
5209 /* For fields not not tracking in the in-memory inode,
5210 * initialise them to zero for new inodes. */
5211 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5212 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5214 ext4_get_inode_flags(ei);
5215 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5216 if (!(test_opt(inode->i_sb, NO_UID32))) {
5217 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5218 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5220 * Fix up interoperability with old kernels. Otherwise, old inodes get
5221 * re-used with the upper 16 bits of the uid/gid intact
5224 raw_inode->i_uid_high =
5225 cpu_to_le16(high_16_bits(inode->i_uid));
5226 raw_inode->i_gid_high =
5227 cpu_to_le16(high_16_bits(inode->i_gid));
5229 raw_inode->i_uid_high = 0;
5230 raw_inode->i_gid_high = 0;
5233 raw_inode->i_uid_low =
5234 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5235 raw_inode->i_gid_low =
5236 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5237 raw_inode->i_uid_high = 0;
5238 raw_inode->i_gid_high = 0;
5240 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5242 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5243 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5244 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5245 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5247 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5249 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5250 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5251 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5252 cpu_to_le32(EXT4_OS_HURD))
5253 raw_inode->i_file_acl_high =
5254 cpu_to_le16(ei->i_file_acl >> 32);
5255 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5256 ext4_isize_set(raw_inode, ei->i_disksize);
5257 if (ei->i_disksize > 0x7fffffffULL) {
5258 struct super_block *sb = inode->i_sb;
5259 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5260 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5261 EXT4_SB(sb)->s_es->s_rev_level ==
5262 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5263 /* If this is the first large file
5264 * created, add a flag to the superblock.
5266 err = ext4_journal_get_write_access(handle,
5267 EXT4_SB(sb)->s_sbh);
5270 ext4_update_dynamic_rev(sb);
5271 EXT4_SET_RO_COMPAT_FEATURE(sb,
5272 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5274 ext4_handle_sync(handle);
5275 err = ext4_handle_dirty_metadata(handle, NULL,
5276 EXT4_SB(sb)->s_sbh);
5279 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5280 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5281 if (old_valid_dev(inode->i_rdev)) {
5282 raw_inode->i_block[0] =
5283 cpu_to_le32(old_encode_dev(inode->i_rdev));
5284 raw_inode->i_block[1] = 0;
5286 raw_inode->i_block[0] = 0;
5287 raw_inode->i_block[1] =
5288 cpu_to_le32(new_encode_dev(inode->i_rdev));
5289 raw_inode->i_block[2] = 0;
5292 for (block = 0; block < EXT4_N_BLOCKS; block++)
5293 raw_inode->i_block[block] = ei->i_data[block];
5295 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5296 if (ei->i_extra_isize) {
5297 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5298 raw_inode->i_version_hi =
5299 cpu_to_le32(inode->i_version >> 32);
5300 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5303 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5304 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5307 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5309 ext4_update_inode_fsync_trans(handle, inode, 0);
5312 ext4_std_error(inode->i_sb, err);
5317 * ext4_write_inode()
5319 * We are called from a few places:
5321 * - Within generic_file_write() for O_SYNC files.
5322 * Here, there will be no transaction running. We wait for any running
5323 * trasnaction to commit.
5325 * - Within sys_sync(), kupdate and such.
5326 * We wait on commit, if tol to.
5328 * - Within prune_icache() (PF_MEMALLOC == true)
5329 * Here we simply return. We can't afford to block kswapd on the
5332 * In all cases it is actually safe for us to return without doing anything,
5333 * because the inode has been copied into a raw inode buffer in
5334 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5337 * Note that we are absolutely dependent upon all inode dirtiers doing the
5338 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5339 * which we are interested.
5341 * It would be a bug for them to not do this. The code:
5343 * mark_inode_dirty(inode)
5345 * inode->i_size = expr;
5347 * is in error because a kswapd-driven write_inode() could occur while
5348 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5349 * will no longer be on the superblock's dirty inode list.
5351 int ext4_write_inode(struct inode *inode, int wait)
5355 if (current->flags & PF_MEMALLOC)
5358 if (EXT4_SB(inode->i_sb)->s_journal) {
5359 if (ext4_journal_current_handle()) {
5360 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5368 err = ext4_force_commit(inode->i_sb);
5370 struct ext4_iloc iloc;
5372 err = ext4_get_inode_loc(inode, &iloc);
5376 sync_dirty_buffer(iloc.bh);
5377 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5378 ext4_error(inode->i_sb, "IO error syncing inode, "
5379 "inode=%lu, block=%llu", inode->i_ino,
5380 (unsigned long long)iloc.bh->b_blocknr);
5390 * Called from notify_change.
5392 * We want to trap VFS attempts to truncate the file as soon as
5393 * possible. In particular, we want to make sure that when the VFS
5394 * shrinks i_size, we put the inode on the orphan list and modify
5395 * i_disksize immediately, so that during the subsequent flushing of
5396 * dirty pages and freeing of disk blocks, we can guarantee that any
5397 * commit will leave the blocks being flushed in an unused state on
5398 * disk. (On recovery, the inode will get truncated and the blocks will
5399 * be freed, so we have a strong guarantee that no future commit will
5400 * leave these blocks visible to the user.)
5402 * Another thing we have to assure is that if we are in ordered mode
5403 * and inode is still attached to the committing transaction, we must
5404 * we start writeout of all the dirty pages which are being truncated.
5405 * This way we are sure that all the data written in the previous
5406 * transaction are already on disk (truncate waits for pages under
5409 * Called with inode->i_mutex down.
5411 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5413 struct inode *inode = dentry->d_inode;
5415 const unsigned int ia_valid = attr->ia_valid;
5417 error = inode_change_ok(inode, attr);
5421 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5422 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5425 /* (user+group)*(old+new) structure, inode write (sb,
5426 * inode block, ? - but truncate inode update has it) */
5427 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5428 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5429 if (IS_ERR(handle)) {
5430 error = PTR_ERR(handle);
5433 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
5435 ext4_journal_stop(handle);
5438 /* Update corresponding info in inode so that everything is in
5439 * one transaction */
5440 if (attr->ia_valid & ATTR_UID)
5441 inode->i_uid = attr->ia_uid;
5442 if (attr->ia_valid & ATTR_GID)
5443 inode->i_gid = attr->ia_gid;
5444 error = ext4_mark_inode_dirty(handle, inode);
5445 ext4_journal_stop(handle);
5448 if (attr->ia_valid & ATTR_SIZE) {
5449 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5450 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5452 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5459 if (S_ISREG(inode->i_mode) &&
5460 attr->ia_valid & ATTR_SIZE &&
5461 (attr->ia_size < inode->i_size ||
5462 (EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL))) {
5465 handle = ext4_journal_start(inode, 3);
5466 if (IS_ERR(handle)) {
5467 error = PTR_ERR(handle);
5471 error = ext4_orphan_add(handle, inode);
5472 EXT4_I(inode)->i_disksize = attr->ia_size;
5473 rc = ext4_mark_inode_dirty(handle, inode);
5476 ext4_journal_stop(handle);
5478 if (ext4_should_order_data(inode)) {
5479 error = ext4_begin_ordered_truncate(inode,
5482 /* Do as much error cleanup as possible */
5483 handle = ext4_journal_start(inode, 3);
5484 if (IS_ERR(handle)) {
5485 ext4_orphan_del(NULL, inode);
5488 ext4_orphan_del(handle, inode);
5489 ext4_journal_stop(handle);
5493 /* ext4_truncate will clear the flag */
5494 if ((EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL))
5495 ext4_truncate(inode);
5498 rc = inode_setattr(inode, attr);
5500 /* If inode_setattr's call to ext4_truncate failed to get a
5501 * transaction handle at all, we need to clean up the in-core
5502 * orphan list manually. */
5504 ext4_orphan_del(NULL, inode);
5506 if (!rc && (ia_valid & ATTR_MODE))
5507 rc = ext4_acl_chmod(inode);
5510 ext4_std_error(inode->i_sb, error);
5516 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5519 struct inode *inode;
5520 unsigned long delalloc_blocks;
5522 inode = dentry->d_inode;
5523 generic_fillattr(inode, stat);
5526 * We can't update i_blocks if the block allocation is delayed
5527 * otherwise in the case of system crash before the real block
5528 * allocation is done, we will have i_blocks inconsistent with
5529 * on-disk file blocks.
5530 * We always keep i_blocks updated together with real
5531 * allocation. But to not confuse with user, stat
5532 * will return the blocks that include the delayed allocation
5533 * blocks for this file.
5535 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5536 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5537 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5539 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5543 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5548 /* if nrblocks are contiguous */
5551 * With N contiguous data blocks, it need at most
5552 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5553 * 2 dindirect blocks
5556 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5557 return indirects + 3;
5560 * if nrblocks are not contiguous, worse case, each block touch
5561 * a indirect block, and each indirect block touch a double indirect
5562 * block, plus a triple indirect block
5564 indirects = nrblocks * 2 + 1;
5568 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5570 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
5571 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5572 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5576 * Account for index blocks, block groups bitmaps and block group
5577 * descriptor blocks if modify datablocks and index blocks
5578 * worse case, the indexs blocks spread over different block groups
5580 * If datablocks are discontiguous, they are possible to spread over
5581 * different block groups too. If they are contiuguous, with flexbg,
5582 * they could still across block group boundary.
5584 * Also account for superblock, inode, quota and xattr blocks
5586 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5588 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5594 * How many index blocks need to touch to modify nrblocks?
5595 * The "Chunk" flag indicating whether the nrblocks is
5596 * physically contiguous on disk
5598 * For Direct IO and fallocate, they calls get_block to allocate
5599 * one single extent at a time, so they could set the "Chunk" flag
5601 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5606 * Now let's see how many group bitmaps and group descriptors need
5616 if (groups > ngroups)
5618 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5619 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5621 /* bitmaps and block group descriptor blocks */
5622 ret += groups + gdpblocks;
5624 /* Blocks for super block, inode, quota and xattr blocks */
5625 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5631 * Calulate the total number of credits to reserve to fit
5632 * the modification of a single pages into a single transaction,
5633 * which may include multiple chunks of block allocations.
5635 * This could be called via ext4_write_begin()
5637 * We need to consider the worse case, when
5638 * one new block per extent.
5640 int ext4_writepage_trans_blocks(struct inode *inode)
5642 int bpp = ext4_journal_blocks_per_page(inode);
5645 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5647 /* Account for data blocks for journalled mode */
5648 if (ext4_should_journal_data(inode))
5654 * Calculate the journal credits for a chunk of data modification.
5656 * This is called from DIO, fallocate or whoever calling
5657 * ext4_get_blocks() to map/allocate a chunk of contiguous disk blocks.
5659 * journal buffers for data blocks are not included here, as DIO
5660 * and fallocate do no need to journal data buffers.
5662 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5664 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5668 * The caller must have previously called ext4_reserve_inode_write().
5669 * Give this, we know that the caller already has write access to iloc->bh.
5671 int ext4_mark_iloc_dirty(handle_t *handle,
5672 struct inode *inode, struct ext4_iloc *iloc)
5676 if (test_opt(inode->i_sb, I_VERSION))
5677 inode_inc_iversion(inode);
5679 /* the do_update_inode consumes one bh->b_count */
5682 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5683 err = ext4_do_update_inode(handle, inode, iloc);
5689 * On success, We end up with an outstanding reference count against
5690 * iloc->bh. This _must_ be cleaned up later.
5694 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5695 struct ext4_iloc *iloc)
5699 err = ext4_get_inode_loc(inode, iloc);
5701 BUFFER_TRACE(iloc->bh, "get_write_access");
5702 err = ext4_journal_get_write_access(handle, iloc->bh);
5708 ext4_std_error(inode->i_sb, err);
5713 * Expand an inode by new_extra_isize bytes.
5714 * Returns 0 on success or negative error number on failure.
5716 static int ext4_expand_extra_isize(struct inode *inode,
5717 unsigned int new_extra_isize,
5718 struct ext4_iloc iloc,
5721 struct ext4_inode *raw_inode;
5722 struct ext4_xattr_ibody_header *header;
5723 struct ext4_xattr_entry *entry;
5725 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5728 raw_inode = ext4_raw_inode(&iloc);
5730 header = IHDR(inode, raw_inode);
5731 entry = IFIRST(header);
5733 /* No extended attributes present */
5734 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5735 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5736 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5738 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5742 /* try to expand with EAs present */
5743 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5748 * What we do here is to mark the in-core inode as clean with respect to inode
5749 * dirtiness (it may still be data-dirty).
5750 * This means that the in-core inode may be reaped by prune_icache
5751 * without having to perform any I/O. This is a very good thing,
5752 * because *any* task may call prune_icache - even ones which
5753 * have a transaction open against a different journal.
5755 * Is this cheating? Not really. Sure, we haven't written the
5756 * inode out, but prune_icache isn't a user-visible syncing function.
5757 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5758 * we start and wait on commits.
5760 * Is this efficient/effective? Well, we're being nice to the system
5761 * by cleaning up our inodes proactively so they can be reaped
5762 * without I/O. But we are potentially leaving up to five seconds'
5763 * worth of inodes floating about which prune_icache wants us to
5764 * write out. One way to fix that would be to get prune_icache()
5765 * to do a write_super() to free up some memory. It has the desired
5768 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5770 struct ext4_iloc iloc;
5771 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5772 static unsigned int mnt_count;
5776 err = ext4_reserve_inode_write(handle, inode, &iloc);
5777 if (ext4_handle_valid(handle) &&
5778 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5779 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5781 * We need extra buffer credits since we may write into EA block
5782 * with this same handle. If journal_extend fails, then it will
5783 * only result in a minor loss of functionality for that inode.
5784 * If this is felt to be critical, then e2fsck should be run to
5785 * force a large enough s_min_extra_isize.
5787 if ((jbd2_journal_extend(handle,
5788 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5789 ret = ext4_expand_extra_isize(inode,
5790 sbi->s_want_extra_isize,
5793 ext4_set_inode_state(inode,
5794 EXT4_STATE_NO_EXPAND);
5796 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5797 ext4_warning(inode->i_sb,
5798 "Unable to expand inode %lu. Delete"
5799 " some EAs or run e2fsck.",
5802 le16_to_cpu(sbi->s_es->s_mnt_count);
5808 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5813 * ext4_dirty_inode() is called from __mark_inode_dirty()
5815 * We're really interested in the case where a file is being extended.
5816 * i_size has been changed by generic_commit_write() and we thus need
5817 * to include the updated inode in the current transaction.
5819 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5820 * are allocated to the file.
5822 * If the inode is marked synchronous, we don't honour that here - doing
5823 * so would cause a commit on atime updates, which we don't bother doing.
5824 * We handle synchronous inodes at the highest possible level.
5826 void ext4_dirty_inode(struct inode *inode)
5830 handle = ext4_journal_start(inode, 2);
5834 ext4_mark_inode_dirty(handle, inode);
5836 ext4_journal_stop(handle);
5843 * Bind an inode's backing buffer_head into this transaction, to prevent
5844 * it from being flushed to disk early. Unlike
5845 * ext4_reserve_inode_write, this leaves behind no bh reference and
5846 * returns no iloc structure, so the caller needs to repeat the iloc
5847 * lookup to mark the inode dirty later.
5849 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5851 struct ext4_iloc iloc;
5855 err = ext4_get_inode_loc(inode, &iloc);
5857 BUFFER_TRACE(iloc.bh, "get_write_access");
5858 err = jbd2_journal_get_write_access(handle, iloc.bh);
5860 err = ext4_handle_dirty_metadata(handle,
5866 ext4_std_error(inode->i_sb, err);
5871 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5878 * We have to be very careful here: changing a data block's
5879 * journaling status dynamically is dangerous. If we write a
5880 * data block to the journal, change the status and then delete
5881 * that block, we risk forgetting to revoke the old log record
5882 * from the journal and so a subsequent replay can corrupt data.
5883 * So, first we make sure that the journal is empty and that
5884 * nobody is changing anything.
5887 journal = EXT4_JOURNAL(inode);
5890 if (is_journal_aborted(journal))
5893 jbd2_journal_lock_updates(journal);
5894 jbd2_journal_flush(journal);
5897 * OK, there are no updates running now, and all cached data is
5898 * synced to disk. We are now in a completely consistent state
5899 * which doesn't have anything in the journal, and we know that
5900 * no filesystem updates are running, so it is safe to modify
5901 * the inode's in-core data-journaling state flag now.
5905 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5907 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5908 ext4_set_aops(inode);
5910 jbd2_journal_unlock_updates(journal);
5912 /* Finally we can mark the inode as dirty. */
5914 handle = ext4_journal_start(inode, 1);
5916 return PTR_ERR(handle);
5918 err = ext4_mark_inode_dirty(handle, inode);
5919 ext4_handle_sync(handle);
5920 ext4_journal_stop(handle);
5921 ext4_std_error(inode->i_sb, err);
5926 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5928 return !buffer_mapped(bh);
5931 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5933 struct page *page = vmf->page;
5938 struct file *file = vma->vm_file;
5939 struct inode *inode = file->f_path.dentry->d_inode;
5940 struct address_space *mapping = inode->i_mapping;
5943 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5944 * get i_mutex because we are already holding mmap_sem.
5946 down_read(&inode->i_alloc_sem);
5947 size = i_size_read(inode);
5948 if (page->mapping != mapping || size <= page_offset(page)
5949 || !PageUptodate(page)) {
5950 /* page got truncated from under us? */
5954 if (PageMappedToDisk(page))
5957 if (page->index == size >> PAGE_CACHE_SHIFT)
5958 len = size & ~PAGE_CACHE_MASK;
5960 len = PAGE_CACHE_SIZE;
5964 * return if we have all the buffers mapped. This avoid
5965 * the need to call write_begin/write_end which does a
5966 * journal_start/journal_stop which can block and take
5969 if (page_has_buffers(page)) {
5970 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5971 ext4_bh_unmapped)) {
5978 * OK, we need to fill the hole... Do write_begin write_end
5979 * to do block allocation/reservation.We are not holding
5980 * inode.i__mutex here. That allow * parallel write_begin,
5981 * write_end call. lock_page prevent this from happening
5982 * on the same page though
5984 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5985 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5988 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5989 len, len, page, fsdata);
5995 ret = VM_FAULT_SIGBUS;
5996 up_read(&inode->i_alloc_sem);