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 index = page->index;
2165 BUG_ON(!PageLocked(page));
2166 BUG_ON(PageWriteback(page));
2167 block_invalidatepage(page, 0);
2168 ClearPageUptodate(page);
2175 static void ext4_print_free_blocks(struct inode *inode)
2177 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2178 printk(KERN_CRIT "Total free blocks count %lld\n",
2179 ext4_count_free_blocks(inode->i_sb));
2180 printk(KERN_CRIT "Free/Dirty block details\n");
2181 printk(KERN_CRIT "free_blocks=%lld\n",
2182 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2183 printk(KERN_CRIT "dirty_blocks=%lld\n",
2184 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2185 printk(KERN_CRIT "Block reservation details\n");
2186 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2187 EXT4_I(inode)->i_reserved_data_blocks);
2188 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2189 EXT4_I(inode)->i_reserved_meta_blocks);
2194 * mpage_da_map_blocks - go through given space
2196 * @mpd - bh describing space
2198 * The function skips space we know is already mapped to disk blocks.
2201 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2203 int err, blks, get_blocks_flags;
2204 struct buffer_head new;
2205 sector_t next = mpd->b_blocknr;
2206 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2207 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2208 handle_t *handle = NULL;
2211 * We consider only non-mapped and non-allocated blocks
2213 if ((mpd->b_state & (1 << BH_Mapped)) &&
2214 !(mpd->b_state & (1 << BH_Delay)) &&
2215 !(mpd->b_state & (1 << BH_Unwritten)))
2219 * If we didn't accumulate anything to write simply return
2224 handle = ext4_journal_current_handle();
2228 * Call ext4_get_blocks() to allocate any delayed allocation
2229 * blocks, or to convert an uninitialized extent to be
2230 * initialized (in the case where we have written into
2231 * one or more preallocated blocks).
2233 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2234 * indicate that we are on the delayed allocation path. This
2235 * affects functions in many different parts of the allocation
2236 * call path. This flag exists primarily because we don't
2237 * want to change *many* call functions, so ext4_get_blocks()
2238 * will set the magic i_delalloc_reserved_flag once the
2239 * inode's allocation semaphore is taken.
2241 * If the blocks in questions were delalloc blocks, set
2242 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2243 * variables are updated after the blocks have been allocated.
2246 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2247 if (ext4_should_dioread_nolock(mpd->inode))
2248 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2249 if (mpd->b_state & (1 << BH_Delay))
2250 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2252 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2253 &new, get_blocks_flags);
2257 * If get block returns with error we simply
2258 * return. Later writepage will redirty the page and
2259 * writepages will find the dirty page again
2264 if (err == -ENOSPC &&
2265 ext4_count_free_blocks(mpd->inode->i_sb)) {
2271 * get block failure will cause us to loop in
2272 * writepages, because a_ops->writepage won't be able
2273 * to make progress. The page will be redirtied by
2274 * writepage and writepages will again try to write
2277 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2278 "delayed block allocation failed for inode %lu at "
2279 "logical offset %llu with max blocks %zd with "
2280 "error %d\n", mpd->inode->i_ino,
2281 (unsigned long long) next,
2282 mpd->b_size >> mpd->inode->i_blkbits, err);
2283 printk(KERN_CRIT "This should not happen!! "
2284 "Data will be lost\n");
2285 if (err == -ENOSPC) {
2286 ext4_print_free_blocks(mpd->inode);
2288 /* invalidate all the pages */
2289 ext4_da_block_invalidatepages(mpd, next,
2290 mpd->b_size >> mpd->inode->i_blkbits);
2295 new.b_size = (blks << mpd->inode->i_blkbits);
2297 if (buffer_new(&new))
2298 __unmap_underlying_blocks(mpd->inode, &new);
2301 * If blocks are delayed marked, we need to
2302 * put actual blocknr and drop delayed bit
2304 if ((mpd->b_state & (1 << BH_Delay)) ||
2305 (mpd->b_state & (1 << BH_Unwritten)))
2306 mpage_put_bnr_to_bhs(mpd, next, &new);
2308 if (ext4_should_order_data(mpd->inode)) {
2309 err = ext4_jbd2_file_inode(handle, mpd->inode);
2315 * Update on-disk size along with block allocation.
2317 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2318 if (disksize > i_size_read(mpd->inode))
2319 disksize = i_size_read(mpd->inode);
2320 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2321 ext4_update_i_disksize(mpd->inode, disksize);
2322 return ext4_mark_inode_dirty(handle, mpd->inode);
2328 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2329 (1 << BH_Delay) | (1 << BH_Unwritten))
2332 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2334 * @mpd->lbh - extent of blocks
2335 * @logical - logical number of the block in the file
2336 * @bh - bh of the block (used to access block's state)
2338 * the function is used to collect contig. blocks in same state
2340 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2341 sector_t logical, size_t b_size,
2342 unsigned long b_state)
2345 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2347 /* check if thereserved journal credits might overflow */
2348 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2349 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2351 * With non-extent format we are limited by the journal
2352 * credit available. Total credit needed to insert
2353 * nrblocks contiguous blocks is dependent on the
2354 * nrblocks. So limit nrblocks.
2357 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2358 EXT4_MAX_TRANS_DATA) {
2360 * Adding the new buffer_head would make it cross the
2361 * allowed limit for which we have journal credit
2362 * reserved. So limit the new bh->b_size
2364 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2365 mpd->inode->i_blkbits;
2366 /* we will do mpage_da_submit_io in the next loop */
2370 * First block in the extent
2372 if (mpd->b_size == 0) {
2373 mpd->b_blocknr = logical;
2374 mpd->b_size = b_size;
2375 mpd->b_state = b_state & BH_FLAGS;
2379 next = mpd->b_blocknr + nrblocks;
2381 * Can we merge the block to our big extent?
2383 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2384 mpd->b_size += b_size;
2390 * We couldn't merge the block to our extent, so we
2391 * need to flush current extent and start new one
2393 if (mpage_da_map_blocks(mpd) == 0)
2394 mpage_da_submit_io(mpd);
2399 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2401 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2405 * __mpage_da_writepage - finds extent of pages and blocks
2407 * @page: page to consider
2408 * @wbc: not used, we just follow rules
2411 * The function finds extents of pages and scan them for all blocks.
2413 static int __mpage_da_writepage(struct page *page,
2414 struct writeback_control *wbc, void *data)
2416 struct mpage_da_data *mpd = data;
2417 struct inode *inode = mpd->inode;
2418 struct buffer_head *bh, *head;
2423 * Rest of the page in the page_vec
2424 * redirty then and skip then. We will
2425 * try to write them again after
2426 * starting a new transaction
2428 redirty_page_for_writepage(wbc, page);
2430 return MPAGE_DA_EXTENT_TAIL;
2433 * Can we merge this page to current extent?
2435 if (mpd->next_page != page->index) {
2437 * Nope, we can't. So, we map non-allocated blocks
2438 * and start IO on them using writepage()
2440 if (mpd->next_page != mpd->first_page) {
2441 if (mpage_da_map_blocks(mpd) == 0)
2442 mpage_da_submit_io(mpd);
2444 * skip rest of the page in the page_vec
2447 redirty_page_for_writepage(wbc, page);
2449 return MPAGE_DA_EXTENT_TAIL;
2453 * Start next extent of pages ...
2455 mpd->first_page = page->index;
2465 mpd->next_page = page->index + 1;
2466 logical = (sector_t) page->index <<
2467 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2469 if (!page_has_buffers(page)) {
2470 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2471 (1 << BH_Dirty) | (1 << BH_Uptodate));
2473 return MPAGE_DA_EXTENT_TAIL;
2476 * Page with regular buffer heads, just add all dirty ones
2478 head = page_buffers(page);
2481 BUG_ON(buffer_locked(bh));
2483 * We need to try to allocate
2484 * unmapped blocks in the same page.
2485 * Otherwise we won't make progress
2486 * with the page in ext4_writepage
2488 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2489 mpage_add_bh_to_extent(mpd, logical,
2493 return MPAGE_DA_EXTENT_TAIL;
2494 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2496 * mapped dirty buffer. We need to update
2497 * the b_state because we look at
2498 * b_state in mpage_da_map_blocks. We don't
2499 * update b_size because if we find an
2500 * unmapped buffer_head later we need to
2501 * use the b_state flag of that buffer_head.
2503 if (mpd->b_size == 0)
2504 mpd->b_state = bh->b_state & BH_FLAGS;
2507 } while ((bh = bh->b_this_page) != head);
2514 * This is a special get_blocks_t callback which is used by
2515 * ext4_da_write_begin(). It will either return mapped block or
2516 * reserve space for a single block.
2518 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2519 * We also have b_blocknr = -1 and b_bdev initialized properly
2521 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2522 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2523 * initialized properly.
2525 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2526 struct buffer_head *bh_result, int create)
2529 sector_t invalid_block = ~((sector_t) 0xffff);
2531 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2534 BUG_ON(create == 0);
2535 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2538 * first, we need to know whether the block is allocated already
2539 * preallocated blocks are unmapped but should treated
2540 * the same as allocated blocks.
2542 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2543 if ((ret == 0) && !buffer_delay(bh_result)) {
2544 /* the block isn't (pre)allocated yet, let's reserve space */
2546 * XXX: __block_prepare_write() unmaps passed block,
2549 ret = ext4_da_reserve_space(inode, iblock);
2551 /* not enough space to reserve */
2554 map_bh(bh_result, inode->i_sb, invalid_block);
2555 set_buffer_new(bh_result);
2556 set_buffer_delay(bh_result);
2557 } else if (ret > 0) {
2558 bh_result->b_size = (ret << inode->i_blkbits);
2559 if (buffer_unwritten(bh_result)) {
2560 /* A delayed write to unwritten bh should
2561 * be marked new and mapped. Mapped ensures
2562 * that we don't do get_block multiple times
2563 * when we write to the same offset and new
2564 * ensures that we do proper zero out for
2567 set_buffer_new(bh_result);
2568 set_buffer_mapped(bh_result);
2577 * This function is used as a standard get_block_t calback function
2578 * when there is no desire to allocate any blocks. It is used as a
2579 * callback function for block_prepare_write(), nobh_writepage(), and
2580 * block_write_full_page(). These functions should only try to map a
2581 * single block at a time.
2583 * Since this function doesn't do block allocations even if the caller
2584 * requests it by passing in create=1, it is critically important that
2585 * any caller checks to make sure that any buffer heads are returned
2586 * by this function are either all already mapped or marked for
2587 * delayed allocation before calling nobh_writepage() or
2588 * block_write_full_page(). Otherwise, b_blocknr could be left
2589 * unitialized, and the page write functions will be taken by
2592 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2593 struct buffer_head *bh_result, int create)
2596 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2598 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2601 * we don't want to do block allocation in writepage
2602 * so call get_block_wrap with create = 0
2604 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2606 bh_result->b_size = (ret << inode->i_blkbits);
2612 static int bget_one(handle_t *handle, struct buffer_head *bh)
2618 static int bput_one(handle_t *handle, struct buffer_head *bh)
2624 static int __ext4_journalled_writepage(struct page *page,
2627 struct address_space *mapping = page->mapping;
2628 struct inode *inode = mapping->host;
2629 struct buffer_head *page_bufs;
2630 handle_t *handle = NULL;
2634 page_bufs = page_buffers(page);
2636 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2637 /* As soon as we unlock the page, it can go away, but we have
2638 * references to buffers so we are safe */
2641 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2642 if (IS_ERR(handle)) {
2643 ret = PTR_ERR(handle);
2647 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2648 do_journal_get_write_access);
2650 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2654 err = ext4_journal_stop(handle);
2658 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2659 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2664 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2665 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2668 * Note that we don't need to start a transaction unless we're journaling data
2669 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2670 * need to file the inode to the transaction's list in ordered mode because if
2671 * we are writing back data added by write(), the inode is already there and if
2672 * we are writing back data modified via mmap(), noone guarantees in which
2673 * transaction the data will hit the disk. In case we are journaling data, we
2674 * cannot start transaction directly because transaction start ranks above page
2675 * lock so we have to do some magic.
2677 * This function can get called via...
2678 * - ext4_da_writepages after taking page lock (have journal handle)
2679 * - journal_submit_inode_data_buffers (no journal handle)
2680 * - shrink_page_list via pdflush (no journal handle)
2681 * - grab_page_cache when doing write_begin (have journal handle)
2683 * We don't do any block allocation in this function. If we have page with
2684 * multiple blocks we need to write those buffer_heads that are mapped. This
2685 * is important for mmaped based write. So if we do with blocksize 1K
2686 * truncate(f, 1024);
2687 * a = mmap(f, 0, 4096);
2689 * truncate(f, 4096);
2690 * we have in the page first buffer_head mapped via page_mkwrite call back
2691 * but other bufer_heads would be unmapped but dirty(dirty done via the
2692 * do_wp_page). So writepage should write the first block. If we modify
2693 * the mmap area beyond 1024 we will again get a page_fault and the
2694 * page_mkwrite callback will do the block allocation and mark the
2695 * buffer_heads mapped.
2697 * We redirty the page if we have any buffer_heads that is either delay or
2698 * unwritten in the page.
2700 * We can get recursively called as show below.
2702 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2705 * But since we don't do any block allocation we should not deadlock.
2706 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2708 static int ext4_writepage(struct page *page,
2709 struct writeback_control *wbc)
2714 struct buffer_head *page_bufs = NULL;
2715 struct inode *inode = page->mapping->host;
2717 trace_ext4_writepage(inode, page);
2718 size = i_size_read(inode);
2719 if (page->index == size >> PAGE_CACHE_SHIFT)
2720 len = size & ~PAGE_CACHE_MASK;
2722 len = PAGE_CACHE_SIZE;
2724 if (page_has_buffers(page)) {
2725 page_bufs = page_buffers(page);
2726 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2727 ext4_bh_delay_or_unwritten)) {
2729 * We don't want to do block allocation
2730 * So redirty the page and return
2731 * We may reach here when we do a journal commit
2732 * via journal_submit_inode_data_buffers.
2733 * If we don't have mapping block we just ignore
2734 * them. We can also reach here via shrink_page_list
2736 redirty_page_for_writepage(wbc, page);
2742 * The test for page_has_buffers() is subtle:
2743 * We know the page is dirty but it lost buffers. That means
2744 * that at some moment in time after write_begin()/write_end()
2745 * has been called all buffers have been clean and thus they
2746 * must have been written at least once. So they are all
2747 * mapped and we can happily proceed with mapping them
2748 * and writing the page.
2750 * Try to initialize the buffer_heads and check whether
2751 * all are mapped and non delay. We don't want to
2752 * do block allocation here.
2754 ret = block_prepare_write(page, 0, len,
2755 noalloc_get_block_write);
2757 page_bufs = page_buffers(page);
2758 /* check whether all are mapped and non delay */
2759 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2760 ext4_bh_delay_or_unwritten)) {
2761 redirty_page_for_writepage(wbc, page);
2767 * We can't do block allocation here
2768 * so just redity the page and unlock
2771 redirty_page_for_writepage(wbc, page);
2775 /* now mark the buffer_heads as dirty and uptodate */
2776 block_commit_write(page, 0, len);
2779 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2781 * It's mmapped pagecache. Add buffers and journal it. There
2782 * doesn't seem much point in redirtying the page here.
2784 ClearPageChecked(page);
2785 return __ext4_journalled_writepage(page, len);
2788 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2789 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2790 else if (page_bufs && buffer_uninit(page_bufs)) {
2791 ext4_set_bh_endio(page_bufs, inode);
2792 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2793 wbc, ext4_end_io_buffer_write);
2795 ret = block_write_full_page(page, noalloc_get_block_write,
2802 * This is called via ext4_da_writepages() to
2803 * calulate the total number of credits to reserve to fit
2804 * a single extent allocation into a single transaction,
2805 * ext4_da_writpeages() will loop calling this before
2806 * the block allocation.
2809 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2811 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2814 * With non-extent format the journal credit needed to
2815 * insert nrblocks contiguous block is dependent on
2816 * number of contiguous block. So we will limit
2817 * number of contiguous block to a sane value
2819 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) &&
2820 (max_blocks > EXT4_MAX_TRANS_DATA))
2821 max_blocks = EXT4_MAX_TRANS_DATA;
2823 return ext4_chunk_trans_blocks(inode, max_blocks);
2826 static int ext4_da_writepages(struct address_space *mapping,
2827 struct writeback_control *wbc)
2830 int range_whole = 0;
2831 handle_t *handle = NULL;
2832 struct mpage_da_data mpd;
2833 struct inode *inode = mapping->host;
2834 int no_nrwrite_index_update;
2835 int pages_written = 0;
2837 unsigned int max_pages;
2838 int range_cyclic, cycled = 1, io_done = 0;
2839 int needed_blocks, ret = 0;
2840 long desired_nr_to_write, nr_to_writebump = 0;
2841 loff_t range_start = wbc->range_start;
2842 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2844 trace_ext4_da_writepages(inode, wbc);
2847 * No pages to write? This is mainly a kludge to avoid starting
2848 * a transaction for special inodes like journal inode on last iput()
2849 * because that could violate lock ordering on umount
2851 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2855 * If the filesystem has aborted, it is read-only, so return
2856 * right away instead of dumping stack traces later on that
2857 * will obscure the real source of the problem. We test
2858 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2859 * the latter could be true if the filesystem is mounted
2860 * read-only, and in that case, ext4_da_writepages should
2861 * *never* be called, so if that ever happens, we would want
2864 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2867 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2870 range_cyclic = wbc->range_cyclic;
2871 if (wbc->range_cyclic) {
2872 index = mapping->writeback_index;
2875 wbc->range_start = index << PAGE_CACHE_SHIFT;
2876 wbc->range_end = LLONG_MAX;
2877 wbc->range_cyclic = 0;
2879 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2882 * This works around two forms of stupidity. The first is in
2883 * the writeback code, which caps the maximum number of pages
2884 * written to be 1024 pages. This is wrong on multiple
2885 * levels; different architectues have a different page size,
2886 * which changes the maximum amount of data which gets
2887 * written. Secondly, 4 megabytes is way too small. XFS
2888 * forces this value to be 16 megabytes by multiplying
2889 * nr_to_write parameter by four, and then relies on its
2890 * allocator to allocate larger extents to make them
2891 * contiguous. Unfortunately this brings us to the second
2892 * stupidity, which is that ext4's mballoc code only allocates
2893 * at most 2048 blocks. So we force contiguous writes up to
2894 * the number of dirty blocks in the inode, or
2895 * sbi->max_writeback_mb_bump whichever is smaller.
2897 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2898 if (!range_cyclic && range_whole)
2899 desired_nr_to_write = wbc->nr_to_write * 8;
2901 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2903 if (desired_nr_to_write > max_pages)
2904 desired_nr_to_write = max_pages;
2906 if (wbc->nr_to_write < desired_nr_to_write) {
2907 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2908 wbc->nr_to_write = desired_nr_to_write;
2912 mpd.inode = mapping->host;
2915 * we don't want write_cache_pages to update
2916 * nr_to_write and writeback_index
2918 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2919 wbc->no_nrwrite_index_update = 1;
2920 pages_skipped = wbc->pages_skipped;
2923 while (!ret && wbc->nr_to_write > 0) {
2926 * we insert one extent at a time. So we need
2927 * credit needed for single extent allocation.
2928 * journalled mode is currently not supported
2931 BUG_ON(ext4_should_journal_data(inode));
2932 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2934 /* start a new transaction*/
2935 handle = ext4_journal_start(inode, needed_blocks);
2936 if (IS_ERR(handle)) {
2937 ret = PTR_ERR(handle);
2938 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2939 "%ld pages, ino %lu; err %d\n", __func__,
2940 wbc->nr_to_write, inode->i_ino, ret);
2941 goto out_writepages;
2945 * Now call __mpage_da_writepage to find the next
2946 * contiguous region of logical blocks that need
2947 * blocks to be allocated by ext4. We don't actually
2948 * submit the blocks for I/O here, even though
2949 * write_cache_pages thinks it will, and will set the
2950 * pages as clean for write before calling
2951 * __mpage_da_writepage().
2959 mpd.pages_written = 0;
2961 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2964 * If we have a contiguous extent of pages and we
2965 * haven't done the I/O yet, map the blocks and submit
2968 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2969 if (mpage_da_map_blocks(&mpd) == 0)
2970 mpage_da_submit_io(&mpd);
2972 ret = MPAGE_DA_EXTENT_TAIL;
2974 trace_ext4_da_write_pages(inode, &mpd);
2975 wbc->nr_to_write -= mpd.pages_written;
2977 ext4_journal_stop(handle);
2979 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2980 /* commit the transaction which would
2981 * free blocks released in the transaction
2984 jbd2_journal_force_commit_nested(sbi->s_journal);
2985 wbc->pages_skipped = pages_skipped;
2987 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2989 * got one extent now try with
2992 pages_written += mpd.pages_written;
2993 wbc->pages_skipped = pages_skipped;
2996 } else if (wbc->nr_to_write)
2998 * There is no more writeout needed
2999 * or we requested for a noblocking writeout
3000 * and we found the device congested
3004 if (!io_done && !cycled) {
3007 wbc->range_start = index << PAGE_CACHE_SHIFT;
3008 wbc->range_end = mapping->writeback_index - 1;
3011 if (pages_skipped != wbc->pages_skipped)
3012 ext4_msg(inode->i_sb, KERN_CRIT,
3013 "This should not happen leaving %s "
3014 "with nr_to_write = %ld ret = %d\n",
3015 __func__, wbc->nr_to_write, ret);
3018 index += pages_written;
3019 wbc->range_cyclic = range_cyclic;
3020 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3022 * set the writeback_index so that range_cyclic
3023 * mode will write it back later
3025 mapping->writeback_index = index;
3028 if (!no_nrwrite_index_update)
3029 wbc->no_nrwrite_index_update = 0;
3030 wbc->nr_to_write -= nr_to_writebump;
3031 wbc->range_start = range_start;
3032 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3036 #define FALL_BACK_TO_NONDELALLOC 1
3037 static int ext4_nonda_switch(struct super_block *sb)
3039 s64 free_blocks, dirty_blocks;
3040 struct ext4_sb_info *sbi = EXT4_SB(sb);
3043 * switch to non delalloc mode if we are running low
3044 * on free block. The free block accounting via percpu
3045 * counters can get slightly wrong with percpu_counter_batch getting
3046 * accumulated on each CPU without updating global counters
3047 * Delalloc need an accurate free block accounting. So switch
3048 * to non delalloc when we are near to error range.
3050 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3051 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3052 if (2 * free_blocks < 3 * dirty_blocks ||
3053 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3055 * free block count is less than 150% of dirty blocks
3056 * or free blocks is less than watermark
3061 * Even if we don't switch but are nearing capacity,
3062 * start pushing delalloc when 1/2 of free blocks are dirty.
3064 if (free_blocks < 2 * dirty_blocks)
3065 writeback_inodes_sb_if_idle(sb);
3070 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3071 loff_t pos, unsigned len, unsigned flags,
3072 struct page **pagep, void **fsdata)
3074 int ret, retries = 0, quota_retries = 0;
3078 struct inode *inode = mapping->host;
3081 index = pos >> PAGE_CACHE_SHIFT;
3082 from = pos & (PAGE_CACHE_SIZE - 1);
3085 if (ext4_nonda_switch(inode->i_sb)) {
3086 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3087 return ext4_write_begin(file, mapping, pos,
3088 len, flags, pagep, fsdata);
3090 *fsdata = (void *)0;
3091 trace_ext4_da_write_begin(inode, pos, len, flags);
3094 * With delayed allocation, we don't log the i_disksize update
3095 * if there is delayed block allocation. But we still need
3096 * to journalling the i_disksize update if writes to the end
3097 * of file which has an already mapped buffer.
3099 handle = ext4_journal_start(inode, 1);
3100 if (IS_ERR(handle)) {
3101 ret = PTR_ERR(handle);
3104 /* We cannot recurse into the filesystem as the transaction is already
3106 flags |= AOP_FLAG_NOFS;
3108 page = grab_cache_page_write_begin(mapping, index, flags);
3110 ext4_journal_stop(handle);
3116 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3117 ext4_da_get_block_prep);
3120 ext4_journal_stop(handle);
3121 page_cache_release(page);
3123 * block_write_begin may have instantiated a few blocks
3124 * outside i_size. Trim these off again. Don't need
3125 * i_size_read because we hold i_mutex.
3127 if (pos + len > inode->i_size)
3128 ext4_truncate_failed_write(inode);
3131 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3134 if ((ret == -EDQUOT) &&
3135 EXT4_I(inode)->i_reserved_meta_blocks &&
3136 (quota_retries++ < 3)) {
3138 * Since we often over-estimate the number of meta
3139 * data blocks required, we may sometimes get a
3140 * spurios out of quota error even though there would
3141 * be enough space once we write the data blocks and
3142 * find out how many meta data blocks were _really_
3143 * required. So try forcing the inode write to see if
3146 write_inode_now(inode, (quota_retries == 3));
3154 * Check if we should update i_disksize
3155 * when write to the end of file but not require block allocation
3157 static int ext4_da_should_update_i_disksize(struct page *page,
3158 unsigned long offset)
3160 struct buffer_head *bh;
3161 struct inode *inode = page->mapping->host;
3165 bh = page_buffers(page);
3166 idx = offset >> inode->i_blkbits;
3168 for (i = 0; i < idx; i++)
3169 bh = bh->b_this_page;
3171 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3176 static int ext4_da_write_end(struct file *file,
3177 struct address_space *mapping,
3178 loff_t pos, unsigned len, unsigned copied,
3179 struct page *page, void *fsdata)
3181 struct inode *inode = mapping->host;
3183 handle_t *handle = ext4_journal_current_handle();
3185 unsigned long start, end;
3186 int write_mode = (int)(unsigned long)fsdata;
3188 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3189 if (ext4_should_order_data(inode)) {
3190 return ext4_ordered_write_end(file, mapping, pos,
3191 len, copied, page, fsdata);
3192 } else if (ext4_should_writeback_data(inode)) {
3193 return ext4_writeback_write_end(file, mapping, pos,
3194 len, copied, page, fsdata);
3200 trace_ext4_da_write_end(inode, pos, len, copied);
3201 start = pos & (PAGE_CACHE_SIZE - 1);
3202 end = start + copied - 1;
3205 * generic_write_end() will run mark_inode_dirty() if i_size
3206 * changes. So let's piggyback the i_disksize mark_inode_dirty
3210 new_i_size = pos + copied;
3211 if (new_i_size > EXT4_I(inode)->i_disksize) {
3212 if (ext4_da_should_update_i_disksize(page, end)) {
3213 down_write(&EXT4_I(inode)->i_data_sem);
3214 if (new_i_size > EXT4_I(inode)->i_disksize) {
3216 * Updating i_disksize when extending file
3217 * without needing block allocation
3219 if (ext4_should_order_data(inode))
3220 ret = ext4_jbd2_file_inode(handle,
3223 EXT4_I(inode)->i_disksize = new_i_size;
3225 up_write(&EXT4_I(inode)->i_data_sem);
3226 /* We need to mark inode dirty even if
3227 * new_i_size is less that inode->i_size
3228 * bu greater than i_disksize.(hint delalloc)
3230 ext4_mark_inode_dirty(handle, inode);
3233 ret2 = generic_write_end(file, mapping, pos, len, copied,
3238 ret2 = ext4_journal_stop(handle);
3242 return ret ? ret : copied;
3245 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3248 * Drop reserved blocks
3250 BUG_ON(!PageLocked(page));
3251 if (!page_has_buffers(page))
3254 ext4_da_page_release_reservation(page, offset);
3257 ext4_invalidatepage(page, offset);
3263 * Force all delayed allocation blocks to be allocated for a given inode.
3265 int ext4_alloc_da_blocks(struct inode *inode)
3267 trace_ext4_alloc_da_blocks(inode);
3269 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3270 !EXT4_I(inode)->i_reserved_meta_blocks)
3274 * We do something simple for now. The filemap_flush() will
3275 * also start triggering a write of the data blocks, which is
3276 * not strictly speaking necessary (and for users of
3277 * laptop_mode, not even desirable). However, to do otherwise
3278 * would require replicating code paths in:
3280 * ext4_da_writepages() ->
3281 * write_cache_pages() ---> (via passed in callback function)
3282 * __mpage_da_writepage() -->
3283 * mpage_add_bh_to_extent()
3284 * mpage_da_map_blocks()
3286 * The problem is that write_cache_pages(), located in
3287 * mm/page-writeback.c, marks pages clean in preparation for
3288 * doing I/O, which is not desirable if we're not planning on
3291 * We could call write_cache_pages(), and then redirty all of
3292 * the pages by calling redirty_page_for_writeback() but that
3293 * would be ugly in the extreme. So instead we would need to
3294 * replicate parts of the code in the above functions,
3295 * simplifying them becuase we wouldn't actually intend to
3296 * write out the pages, but rather only collect contiguous
3297 * logical block extents, call the multi-block allocator, and
3298 * then update the buffer heads with the block allocations.
3300 * For now, though, we'll cheat by calling filemap_flush(),
3301 * which will map the blocks, and start the I/O, but not
3302 * actually wait for the I/O to complete.
3304 return filemap_flush(inode->i_mapping);
3308 * bmap() is special. It gets used by applications such as lilo and by
3309 * the swapper to find the on-disk block of a specific piece of data.
3311 * Naturally, this is dangerous if the block concerned is still in the
3312 * journal. If somebody makes a swapfile on an ext4 data-journaling
3313 * filesystem and enables swap, then they may get a nasty shock when the
3314 * data getting swapped to that swapfile suddenly gets overwritten by
3315 * the original zero's written out previously to the journal and
3316 * awaiting writeback in the kernel's buffer cache.
3318 * So, if we see any bmap calls here on a modified, data-journaled file,
3319 * take extra steps to flush any blocks which might be in the cache.
3321 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3323 struct inode *inode = mapping->host;
3327 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3328 test_opt(inode->i_sb, DELALLOC)) {
3330 * With delalloc we want to sync the file
3331 * so that we can make sure we allocate
3334 filemap_write_and_wait(mapping);
3337 if (EXT4_JOURNAL(inode) &&
3338 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3340 * This is a REALLY heavyweight approach, but the use of
3341 * bmap on dirty files is expected to be extremely rare:
3342 * only if we run lilo or swapon on a freshly made file
3343 * do we expect this to happen.
3345 * (bmap requires CAP_SYS_RAWIO so this does not
3346 * represent an unprivileged user DOS attack --- we'd be
3347 * in trouble if mortal users could trigger this path at
3350 * NB. EXT4_STATE_JDATA is not set on files other than
3351 * regular files. If somebody wants to bmap a directory
3352 * or symlink and gets confused because the buffer
3353 * hasn't yet been flushed to disk, they deserve
3354 * everything they get.
3357 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3358 journal = EXT4_JOURNAL(inode);
3359 jbd2_journal_lock_updates(journal);
3360 err = jbd2_journal_flush(journal);
3361 jbd2_journal_unlock_updates(journal);
3367 return generic_block_bmap(mapping, block, ext4_get_block);
3370 static int ext4_readpage(struct file *file, struct page *page)
3372 return mpage_readpage(page, ext4_get_block);
3376 ext4_readpages(struct file *file, struct address_space *mapping,
3377 struct list_head *pages, unsigned nr_pages)
3379 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3382 static void ext4_free_io_end(ext4_io_end_t *io)
3391 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3393 struct buffer_head *head, *bh;
3394 unsigned int curr_off = 0;
3396 if (!page_has_buffers(page))
3398 head = bh = page_buffers(page);
3400 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3402 ext4_free_io_end(bh->b_private);
3403 bh->b_private = NULL;
3404 bh->b_end_io = NULL;
3406 curr_off = curr_off + bh->b_size;
3407 bh = bh->b_this_page;
3408 } while (bh != head);
3411 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3413 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3416 * free any io_end structure allocated for buffers to be discarded
3418 if (ext4_should_dioread_nolock(page->mapping->host))
3419 ext4_invalidatepage_free_endio(page, offset);
3421 * If it's a full truncate we just forget about the pending dirtying
3424 ClearPageChecked(page);
3427 jbd2_journal_invalidatepage(journal, page, offset);
3429 block_invalidatepage(page, offset);
3432 static int ext4_releasepage(struct page *page, gfp_t wait)
3434 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3436 WARN_ON(PageChecked(page));
3437 if (!page_has_buffers(page))
3440 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3442 return try_to_free_buffers(page);
3446 * O_DIRECT for ext3 (or indirect map) based files
3448 * If the O_DIRECT write will extend the file then add this inode to the
3449 * orphan list. So recovery will truncate it back to the original size
3450 * if the machine crashes during the write.
3452 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3453 * crashes then stale disk data _may_ be exposed inside the file. But current
3454 * VFS code falls back into buffered path in that case so we are safe.
3456 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3457 const struct iovec *iov, loff_t offset,
3458 unsigned long nr_segs)
3460 struct file *file = iocb->ki_filp;
3461 struct inode *inode = file->f_mapping->host;
3462 struct ext4_inode_info *ei = EXT4_I(inode);
3466 size_t count = iov_length(iov, nr_segs);
3470 loff_t final_size = offset + count;
3472 if (final_size > inode->i_size) {
3473 /* Credits for sb + inode write */
3474 handle = ext4_journal_start(inode, 2);
3475 if (IS_ERR(handle)) {
3476 ret = PTR_ERR(handle);
3479 ret = ext4_orphan_add(handle, inode);
3481 ext4_journal_stop(handle);
3485 ei->i_disksize = inode->i_size;
3486 ext4_journal_stop(handle);
3491 if (rw == READ && ext4_should_dioread_nolock(inode))
3492 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
3493 inode->i_sb->s_bdev, iov,
3495 ext4_get_block, NULL);
3497 ret = blockdev_direct_IO(rw, iocb, inode,
3498 inode->i_sb->s_bdev, iov,
3500 ext4_get_block, NULL);
3501 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3507 /* Credits for sb + inode write */
3508 handle = ext4_journal_start(inode, 2);
3509 if (IS_ERR(handle)) {
3510 /* This is really bad luck. We've written the data
3511 * but cannot extend i_size. Bail out and pretend
3512 * the write failed... */
3513 ret = PTR_ERR(handle);
3515 ext4_orphan_del(NULL, inode);
3520 ext4_orphan_del(handle, inode);
3522 loff_t end = offset + ret;
3523 if (end > inode->i_size) {
3524 ei->i_disksize = end;
3525 i_size_write(inode, end);
3527 * We're going to return a positive `ret'
3528 * here due to non-zero-length I/O, so there's
3529 * no way of reporting error returns from
3530 * ext4_mark_inode_dirty() to userspace. So
3533 ext4_mark_inode_dirty(handle, inode);
3536 err = ext4_journal_stop(handle);
3544 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3545 struct buffer_head *bh_result, int create)
3547 handle_t *handle = ext4_journal_current_handle();
3549 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3553 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3554 inode->i_ino, create);
3556 * ext4_get_block in prepare for a DIO write or buffer write.
3557 * We allocate an uinitialized extent if blocks haven't been allocated.
3558 * The extent will be converted to initialized after IO complete.
3560 create = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3563 if (max_blocks > DIO_MAX_BLOCKS)
3564 max_blocks = DIO_MAX_BLOCKS;
3565 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3566 handle = ext4_journal_start(inode, dio_credits);
3567 if (IS_ERR(handle)) {
3568 ret = PTR_ERR(handle);
3574 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3577 bh_result->b_size = (ret << inode->i_blkbits);
3581 ext4_journal_stop(handle);
3586 static void dump_completed_IO(struct inode * inode)
3589 struct list_head *cur, *before, *after;
3590 ext4_io_end_t *io, *io0, *io1;
3591 unsigned long flags;
3593 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3594 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3598 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3599 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3600 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3603 io0 = container_of(before, ext4_io_end_t, list);
3605 io1 = container_of(after, ext4_io_end_t, list);
3607 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3608 io, inode->i_ino, io0, io1);
3610 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3615 * check a range of space and convert unwritten extents to written.
3617 static int ext4_end_io_nolock(ext4_io_end_t *io)
3619 struct inode *inode = io->inode;
3620 loff_t offset = io->offset;
3621 ssize_t size = io->size;
3624 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3625 "list->prev 0x%p\n",
3626 io, inode->i_ino, io->list.next, io->list.prev);
3628 if (list_empty(&io->list))
3631 if (io->flag != EXT4_IO_UNWRITTEN)
3634 ret = ext4_convert_unwritten_extents(inode, offset, size);
3636 printk(KERN_EMERG "%s: failed to convert unwritten"
3637 "extents to written extents, error is %d"
3638 " io is still on inode %lu aio dio list\n",
3639 __func__, ret, inode->i_ino);
3643 /* clear the DIO AIO unwritten flag */
3649 * work on completed aio dio IO, to convert unwritten extents to extents
3651 static void ext4_end_io_work(struct work_struct *work)
3653 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3654 struct inode *inode = io->inode;
3655 struct ext4_inode_info *ei = EXT4_I(inode);
3656 unsigned long flags;
3659 mutex_lock(&inode->i_mutex);
3660 ret = ext4_end_io_nolock(io);
3662 mutex_unlock(&inode->i_mutex);
3666 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3667 if (!list_empty(&io->list))
3668 list_del_init(&io->list);
3669 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3670 mutex_unlock(&inode->i_mutex);
3671 ext4_free_io_end(io);
3675 * This function is called from ext4_sync_file().
3677 * When IO is completed, the work to convert unwritten extents to
3678 * written is queued on workqueue but may not get immediately
3679 * scheduled. When fsync is called, we need to ensure the
3680 * conversion is complete before fsync returns.
3681 * The inode keeps track of a list of pending/completed IO that
3682 * might needs to do the conversion. This function walks through
3683 * the list and convert the related unwritten extents for completed IO
3685 * The function return the number of pending IOs on success.
3687 int flush_completed_IO(struct inode *inode)
3690 struct ext4_inode_info *ei = EXT4_I(inode);
3691 unsigned long flags;
3695 if (list_empty(&ei->i_completed_io_list))
3698 dump_completed_IO(inode);
3699 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3700 while (!list_empty(&ei->i_completed_io_list)){
3701 io = list_entry(ei->i_completed_io_list.next,
3702 ext4_io_end_t, list);
3704 * Calling ext4_end_io_nolock() to convert completed
3707 * When ext4_sync_file() is called, run_queue() may already
3708 * about to flush the work corresponding to this io structure.
3709 * It will be upset if it founds the io structure related
3710 * to the work-to-be schedule is freed.
3712 * Thus we need to keep the io structure still valid here after
3713 * convertion finished. The io structure has a flag to
3714 * avoid double converting from both fsync and background work
3717 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3718 ret = ext4_end_io_nolock(io);
3719 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3723 list_del_init(&io->list);
3725 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3726 return (ret2 < 0) ? ret2 : 0;
3729 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3731 ext4_io_end_t *io = NULL;
3733 io = kmalloc(sizeof(*io), flags);
3742 INIT_WORK(&io->work, ext4_end_io_work);
3743 INIT_LIST_HEAD(&io->list);
3749 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3750 ssize_t size, void *private)
3752 ext4_io_end_t *io_end = iocb->private;
3753 struct workqueue_struct *wq;
3754 unsigned long flags;
3755 struct ext4_inode_info *ei;
3757 /* if not async direct IO or dio with 0 bytes write, just return */
3758 if (!io_end || !size)
3761 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3762 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3763 iocb->private, io_end->inode->i_ino, iocb, offset,
3766 /* if not aio dio with unwritten extents, just free io and return */
3767 if (io_end->flag != EXT4_IO_UNWRITTEN){
3768 ext4_free_io_end(io_end);
3769 iocb->private = NULL;
3773 io_end->offset = offset;
3774 io_end->size = size;
3775 io_end->flag = EXT4_IO_UNWRITTEN;
3776 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3778 /* queue the work to convert unwritten extents to written */
3779 queue_work(wq, &io_end->work);
3781 /* Add the io_end to per-inode completed aio dio list*/
3782 ei = EXT4_I(io_end->inode);
3783 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3784 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3785 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3786 iocb->private = NULL;
3789 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3791 ext4_io_end_t *io_end = bh->b_private;
3792 struct workqueue_struct *wq;
3793 struct inode *inode;
3794 unsigned long flags;
3796 if (!test_clear_buffer_uninit(bh) || !io_end)
3799 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3800 printk("sb umounted, discard end_io request for inode %lu\n",
3801 io_end->inode->i_ino);
3802 ext4_free_io_end(io_end);
3806 io_end->flag = EXT4_IO_UNWRITTEN;
3807 inode = io_end->inode;
3809 /* Add the io_end to per-inode completed io list*/
3810 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3811 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3812 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3814 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3815 /* queue the work to convert unwritten extents to written */
3816 queue_work(wq, &io_end->work);
3818 bh->b_private = NULL;
3819 bh->b_end_io = NULL;
3820 clear_buffer_uninit(bh);
3821 end_buffer_async_write(bh, uptodate);
3824 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3826 ext4_io_end_t *io_end;
3827 struct page *page = bh->b_page;
3828 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3829 size_t size = bh->b_size;
3832 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3834 if (printk_ratelimit())
3835 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3839 io_end->offset = offset;
3840 io_end->size = size;
3842 * We need to hold a reference to the page to make sure it
3843 * doesn't get evicted before ext4_end_io_work() has a chance
3844 * to convert the extent from written to unwritten.
3846 io_end->page = page;
3847 get_page(io_end->page);
3849 bh->b_private = io_end;
3850 bh->b_end_io = ext4_end_io_buffer_write;
3855 * For ext4 extent files, ext4 will do direct-io write to holes,
3856 * preallocated extents, and those write extend the file, no need to
3857 * fall back to buffered IO.
3859 * For holes, we fallocate those blocks, mark them as unintialized
3860 * If those blocks were preallocated, we mark sure they are splited, but
3861 * still keep the range to write as unintialized.
3863 * The unwrritten extents will be converted to written when DIO is completed.
3864 * For async direct IO, since the IO may still pending when return, we
3865 * set up an end_io call back function, which will do the convertion
3866 * when async direct IO completed.
3868 * If the O_DIRECT write will extend the file then add this inode to the
3869 * orphan list. So recovery will truncate it back to the original size
3870 * if the machine crashes during the write.
3873 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3874 const struct iovec *iov, loff_t offset,
3875 unsigned long nr_segs)
3877 struct file *file = iocb->ki_filp;
3878 struct inode *inode = file->f_mapping->host;
3880 size_t count = iov_length(iov, nr_segs);
3882 loff_t final_size = offset + count;
3883 if (rw == WRITE && final_size <= inode->i_size) {
3885 * We could direct write to holes and fallocate.
3887 * Allocated blocks to fill the hole are marked as uninitialized
3888 * to prevent paralel buffered read to expose the stale data
3889 * before DIO complete the data IO.
3891 * As to previously fallocated extents, ext4 get_block
3892 * will just simply mark the buffer mapped but still
3893 * keep the extents uninitialized.
3895 * for non AIO case, we will convert those unwritten extents
3896 * to written after return back from blockdev_direct_IO.
3898 * for async DIO, the conversion needs to be defered when
3899 * the IO is completed. The ext4 end_io callback function
3900 * will be called to take care of the conversion work.
3901 * Here for async case, we allocate an io_end structure to
3904 iocb->private = NULL;
3905 EXT4_I(inode)->cur_aio_dio = NULL;
3906 if (!is_sync_kiocb(iocb)) {
3907 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3911 * we save the io structure for current async
3912 * direct IO, so that later ext4_get_blocks()
3913 * could flag the io structure whether there
3914 * is a unwritten extents needs to be converted
3915 * when IO is completed.
3917 EXT4_I(inode)->cur_aio_dio = iocb->private;
3920 ret = blockdev_direct_IO(rw, iocb, inode,
3921 inode->i_sb->s_bdev, iov,
3923 ext4_get_block_write,
3926 EXT4_I(inode)->cur_aio_dio = NULL;
3928 * The io_end structure takes a reference to the inode,
3929 * that structure needs to be destroyed and the
3930 * reference to the inode need to be dropped, when IO is
3931 * complete, even with 0 byte write, or failed.
3933 * In the successful AIO DIO case, the io_end structure will be
3934 * desctroyed and the reference to the inode will be dropped
3935 * after the end_io call back function is called.
3937 * In the case there is 0 byte write, or error case, since
3938 * VFS direct IO won't invoke the end_io call back function,
3939 * we need to free the end_io structure here.
3941 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3942 ext4_free_io_end(iocb->private);
3943 iocb->private = NULL;
3944 } else if (ret > 0 && ext4_test_inode_state(inode,
3945 EXT4_STATE_DIO_UNWRITTEN)) {
3948 * for non AIO case, since the IO is already
3949 * completed, we could do the convertion right here
3951 err = ext4_convert_unwritten_extents(inode,
3955 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3960 /* for write the the end of file case, we fall back to old way */
3961 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3964 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3965 const struct iovec *iov, loff_t offset,
3966 unsigned long nr_segs)
3968 struct file *file = iocb->ki_filp;
3969 struct inode *inode = file->f_mapping->host;
3971 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3972 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3974 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3978 * Pages can be marked dirty completely asynchronously from ext4's journalling
3979 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3980 * much here because ->set_page_dirty is called under VFS locks. The page is
3981 * not necessarily locked.
3983 * We cannot just dirty the page and leave attached buffers clean, because the
3984 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3985 * or jbddirty because all the journalling code will explode.
3987 * So what we do is to mark the page "pending dirty" and next time writepage
3988 * is called, propagate that into the buffers appropriately.
3990 static int ext4_journalled_set_page_dirty(struct page *page)
3992 SetPageChecked(page);
3993 return __set_page_dirty_nobuffers(page);
3996 static const struct address_space_operations ext4_ordered_aops = {
3997 .readpage = ext4_readpage,
3998 .readpages = ext4_readpages,
3999 .writepage = ext4_writepage,
4000 .sync_page = block_sync_page,
4001 .write_begin = ext4_write_begin,
4002 .write_end = ext4_ordered_write_end,
4004 .invalidatepage = ext4_invalidatepage,
4005 .releasepage = ext4_releasepage,
4006 .direct_IO = ext4_direct_IO,
4007 .migratepage = buffer_migrate_page,
4008 .is_partially_uptodate = block_is_partially_uptodate,
4009 .error_remove_page = generic_error_remove_page,
4012 static const struct address_space_operations ext4_writeback_aops = {
4013 .readpage = ext4_readpage,
4014 .readpages = ext4_readpages,
4015 .writepage = ext4_writepage,
4016 .sync_page = block_sync_page,
4017 .write_begin = ext4_write_begin,
4018 .write_end = ext4_writeback_write_end,
4020 .invalidatepage = ext4_invalidatepage,
4021 .releasepage = ext4_releasepage,
4022 .direct_IO = ext4_direct_IO,
4023 .migratepage = buffer_migrate_page,
4024 .is_partially_uptodate = block_is_partially_uptodate,
4025 .error_remove_page = generic_error_remove_page,
4028 static const struct address_space_operations ext4_journalled_aops = {
4029 .readpage = ext4_readpage,
4030 .readpages = ext4_readpages,
4031 .writepage = ext4_writepage,
4032 .sync_page = block_sync_page,
4033 .write_begin = ext4_write_begin,
4034 .write_end = ext4_journalled_write_end,
4035 .set_page_dirty = ext4_journalled_set_page_dirty,
4037 .invalidatepage = ext4_invalidatepage,
4038 .releasepage = ext4_releasepage,
4039 .is_partially_uptodate = block_is_partially_uptodate,
4040 .error_remove_page = generic_error_remove_page,
4043 static const struct address_space_operations ext4_da_aops = {
4044 .readpage = ext4_readpage,
4045 .readpages = ext4_readpages,
4046 .writepage = ext4_writepage,
4047 .writepages = ext4_da_writepages,
4048 .sync_page = block_sync_page,
4049 .write_begin = ext4_da_write_begin,
4050 .write_end = ext4_da_write_end,
4052 .invalidatepage = ext4_da_invalidatepage,
4053 .releasepage = ext4_releasepage,
4054 .direct_IO = ext4_direct_IO,
4055 .migratepage = buffer_migrate_page,
4056 .is_partially_uptodate = block_is_partially_uptodate,
4057 .error_remove_page = generic_error_remove_page,
4060 void ext4_set_aops(struct inode *inode)
4062 if (ext4_should_order_data(inode) &&
4063 test_opt(inode->i_sb, DELALLOC))
4064 inode->i_mapping->a_ops = &ext4_da_aops;
4065 else if (ext4_should_order_data(inode))
4066 inode->i_mapping->a_ops = &ext4_ordered_aops;
4067 else if (ext4_should_writeback_data(inode) &&
4068 test_opt(inode->i_sb, DELALLOC))
4069 inode->i_mapping->a_ops = &ext4_da_aops;
4070 else if (ext4_should_writeback_data(inode))
4071 inode->i_mapping->a_ops = &ext4_writeback_aops;
4073 inode->i_mapping->a_ops = &ext4_journalled_aops;
4077 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4078 * up to the end of the block which corresponds to `from'.
4079 * This required during truncate. We need to physically zero the tail end
4080 * of that block so it doesn't yield old data if the file is later grown.
4082 int ext4_block_truncate_page(handle_t *handle,
4083 struct address_space *mapping, loff_t from)
4085 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4086 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4087 unsigned blocksize, length, pos;
4089 struct inode *inode = mapping->host;
4090 struct buffer_head *bh;
4094 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4095 mapping_gfp_mask(mapping) & ~__GFP_FS);
4099 blocksize = inode->i_sb->s_blocksize;
4100 length = blocksize - (offset & (blocksize - 1));
4101 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4104 * For "nobh" option, we can only work if we don't need to
4105 * read-in the page - otherwise we create buffers to do the IO.
4107 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
4108 ext4_should_writeback_data(inode) && PageUptodate(page)) {
4109 zero_user(page, offset, length);
4110 set_page_dirty(page);
4114 if (!page_has_buffers(page))
4115 create_empty_buffers(page, blocksize, 0);
4117 /* Find the buffer that contains "offset" */
4118 bh = page_buffers(page);
4120 while (offset >= pos) {
4121 bh = bh->b_this_page;
4127 if (buffer_freed(bh)) {
4128 BUFFER_TRACE(bh, "freed: skip");
4132 if (!buffer_mapped(bh)) {
4133 BUFFER_TRACE(bh, "unmapped");
4134 ext4_get_block(inode, iblock, bh, 0);
4135 /* unmapped? It's a hole - nothing to do */
4136 if (!buffer_mapped(bh)) {
4137 BUFFER_TRACE(bh, "still unmapped");
4142 /* Ok, it's mapped. Make sure it's up-to-date */
4143 if (PageUptodate(page))
4144 set_buffer_uptodate(bh);
4146 if (!buffer_uptodate(bh)) {
4148 ll_rw_block(READ, 1, &bh);
4150 /* Uhhuh. Read error. Complain and punt. */
4151 if (!buffer_uptodate(bh))
4155 if (ext4_should_journal_data(inode)) {
4156 BUFFER_TRACE(bh, "get write access");
4157 err = ext4_journal_get_write_access(handle, bh);
4162 zero_user(page, offset, length);
4164 BUFFER_TRACE(bh, "zeroed end of block");
4167 if (ext4_should_journal_data(inode)) {
4168 err = ext4_handle_dirty_metadata(handle, inode, bh);
4170 if (ext4_should_order_data(inode))
4171 err = ext4_jbd2_file_inode(handle, inode);
4172 mark_buffer_dirty(bh);
4177 page_cache_release(page);
4182 * Probably it should be a library function... search for first non-zero word
4183 * or memcmp with zero_page, whatever is better for particular architecture.
4186 static inline int all_zeroes(__le32 *p, __le32 *q)
4195 * ext4_find_shared - find the indirect blocks for partial truncation.
4196 * @inode: inode in question
4197 * @depth: depth of the affected branch
4198 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4199 * @chain: place to store the pointers to partial indirect blocks
4200 * @top: place to the (detached) top of branch
4202 * This is a helper function used by ext4_truncate().
4204 * When we do truncate() we may have to clean the ends of several
4205 * indirect blocks but leave the blocks themselves alive. Block is
4206 * partially truncated if some data below the new i_size is refered
4207 * from it (and it is on the path to the first completely truncated
4208 * data block, indeed). We have to free the top of that path along
4209 * with everything to the right of the path. Since no allocation
4210 * past the truncation point is possible until ext4_truncate()
4211 * finishes, we may safely do the latter, but top of branch may
4212 * require special attention - pageout below the truncation point
4213 * might try to populate it.
4215 * We atomically detach the top of branch from the tree, store the
4216 * block number of its root in *@top, pointers to buffer_heads of
4217 * partially truncated blocks - in @chain[].bh and pointers to
4218 * their last elements that should not be removed - in
4219 * @chain[].p. Return value is the pointer to last filled element
4222 * The work left to caller to do the actual freeing of subtrees:
4223 * a) free the subtree starting from *@top
4224 * b) free the subtrees whose roots are stored in
4225 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4226 * c) free the subtrees growing from the inode past the @chain[0].
4227 * (no partially truncated stuff there). */
4229 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4230 ext4_lblk_t offsets[4], Indirect chain[4],
4233 Indirect *partial, *p;
4237 /* Make k index the deepest non-null offset + 1 */
4238 for (k = depth; k > 1 && !offsets[k-1]; k--)
4240 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4241 /* Writer: pointers */
4243 partial = chain + k-1;
4245 * If the branch acquired continuation since we've looked at it -
4246 * fine, it should all survive and (new) top doesn't belong to us.
4248 if (!partial->key && *partial->p)
4251 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4254 * OK, we've found the last block that must survive. The rest of our
4255 * branch should be detached before unlocking. However, if that rest
4256 * of branch is all ours and does not grow immediately from the inode
4257 * it's easier to cheat and just decrement partial->p.
4259 if (p == chain + k - 1 && p > chain) {
4263 /* Nope, don't do this in ext4. Must leave the tree intact */
4270 while (partial > p) {
4271 brelse(partial->bh);
4279 * Zero a number of block pointers in either an inode or an indirect block.
4280 * If we restart the transaction we must again get write access to the
4281 * indirect block for further modification.
4283 * We release `count' blocks on disk, but (last - first) may be greater
4284 * than `count' because there can be holes in there.
4286 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4287 struct buffer_head *bh,
4288 ext4_fsblk_t block_to_free,
4289 unsigned long count, __le32 *first,
4293 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4295 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4296 flags |= EXT4_FREE_BLOCKS_METADATA;
4298 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4300 ext4_error(inode->i_sb, "inode #%lu: "
4301 "attempt to clear blocks %llu len %lu, invalid",
4302 inode->i_ino, (unsigned long long) block_to_free,
4307 if (try_to_extend_transaction(handle, inode)) {
4309 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4310 ext4_handle_dirty_metadata(handle, inode, bh);
4312 ext4_mark_inode_dirty(handle, inode);
4313 ext4_truncate_restart_trans(handle, inode,
4314 blocks_for_truncate(inode));
4316 BUFFER_TRACE(bh, "retaking write access");
4317 ext4_journal_get_write_access(handle, bh);
4321 for (p = first; p < last; p++)
4324 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4329 * ext4_free_data - free a list of data blocks
4330 * @handle: handle for this transaction
4331 * @inode: inode we are dealing with
4332 * @this_bh: indirect buffer_head which contains *@first and *@last
4333 * @first: array of block numbers
4334 * @last: points immediately past the end of array
4336 * We are freeing all blocks refered from that array (numbers are stored as
4337 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4339 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4340 * blocks are contiguous then releasing them at one time will only affect one
4341 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4342 * actually use a lot of journal space.
4344 * @this_bh will be %NULL if @first and @last point into the inode's direct
4347 static void ext4_free_data(handle_t *handle, struct inode *inode,
4348 struct buffer_head *this_bh,
4349 __le32 *first, __le32 *last)
4351 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4352 unsigned long count = 0; /* Number of blocks in the run */
4353 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4356 ext4_fsblk_t nr; /* Current block # */
4357 __le32 *p; /* Pointer into inode/ind
4358 for current block */
4361 if (this_bh) { /* For indirect block */
4362 BUFFER_TRACE(this_bh, "get_write_access");
4363 err = ext4_journal_get_write_access(handle, this_bh);
4364 /* Important: if we can't update the indirect pointers
4365 * to the blocks, we can't free them. */
4370 for (p = first; p < last; p++) {
4371 nr = le32_to_cpu(*p);
4373 /* accumulate blocks to free if they're contiguous */
4376 block_to_free_p = p;
4378 } else if (nr == block_to_free + count) {
4381 if (ext4_clear_blocks(handle, inode, this_bh,
4382 block_to_free, count,
4383 block_to_free_p, p))
4386 block_to_free_p = p;
4393 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4394 count, block_to_free_p, p);
4397 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4400 * The buffer head should have an attached journal head at this
4401 * point. However, if the data is corrupted and an indirect
4402 * block pointed to itself, it would have been detached when
4403 * the block was cleared. Check for this instead of OOPSing.
4405 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4406 ext4_handle_dirty_metadata(handle, inode, this_bh);
4408 ext4_error(inode->i_sb,
4409 "circular indirect block detected, "
4410 "inode=%lu, block=%llu",
4412 (unsigned long long) this_bh->b_blocknr);
4417 * ext4_free_branches - free an array of branches
4418 * @handle: JBD handle for this transaction
4419 * @inode: inode we are dealing with
4420 * @parent_bh: the buffer_head which contains *@first and *@last
4421 * @first: array of block numbers
4422 * @last: pointer immediately past the end of array
4423 * @depth: depth of the branches to free
4425 * We are freeing all blocks refered from these branches (numbers are
4426 * stored as little-endian 32-bit) and updating @inode->i_blocks
4429 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4430 struct buffer_head *parent_bh,
4431 __le32 *first, __le32 *last, int depth)
4436 if (ext4_handle_is_aborted(handle))
4440 struct buffer_head *bh;
4441 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4443 while (--p >= first) {
4444 nr = le32_to_cpu(*p);
4446 continue; /* A hole */
4448 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4450 ext4_error(inode->i_sb,
4451 "indirect mapped block in inode "
4452 "#%lu invalid (level %d, blk #%lu)",
4453 inode->i_ino, depth,
4454 (unsigned long) nr);
4458 /* Go read the buffer for the next level down */
4459 bh = sb_bread(inode->i_sb, nr);
4462 * A read failure? Report error and clear slot
4466 ext4_error(inode->i_sb,
4467 "Read failure, inode=%lu, block=%llu",
4472 /* This zaps the entire block. Bottom up. */
4473 BUFFER_TRACE(bh, "free child branches");
4474 ext4_free_branches(handle, inode, bh,
4475 (__le32 *) bh->b_data,
4476 (__le32 *) bh->b_data + addr_per_block,
4480 * We've probably journalled the indirect block several
4481 * times during the truncate. But it's no longer
4482 * needed and we now drop it from the transaction via
4483 * jbd2_journal_revoke().
4485 * That's easy if it's exclusively part of this
4486 * transaction. But if it's part of the committing
4487 * transaction then jbd2_journal_forget() will simply
4488 * brelse() it. That means that if the underlying
4489 * block is reallocated in ext4_get_block(),
4490 * unmap_underlying_metadata() will find this block
4491 * and will try to get rid of it. damn, damn.
4493 * If this block has already been committed to the
4494 * journal, a revoke record will be written. And
4495 * revoke records must be emitted *before* clearing
4496 * this block's bit in the bitmaps.
4498 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
4501 * Everything below this this pointer has been
4502 * released. Now let this top-of-subtree go.
4504 * We want the freeing of this indirect block to be
4505 * atomic in the journal with the updating of the
4506 * bitmap block which owns it. So make some room in
4509 * We zero the parent pointer *after* freeing its
4510 * pointee in the bitmaps, so if extend_transaction()
4511 * for some reason fails to put the bitmap changes and
4512 * the release into the same transaction, recovery
4513 * will merely complain about releasing a free block,
4514 * rather than leaking blocks.
4516 if (ext4_handle_is_aborted(handle))
4518 if (try_to_extend_transaction(handle, inode)) {
4519 ext4_mark_inode_dirty(handle, inode);
4520 ext4_truncate_restart_trans(handle, inode,
4521 blocks_for_truncate(inode));
4524 ext4_free_blocks(handle, inode, 0, nr, 1,
4525 EXT4_FREE_BLOCKS_METADATA);
4529 * The block which we have just freed is
4530 * pointed to by an indirect block: journal it
4532 BUFFER_TRACE(parent_bh, "get_write_access");
4533 if (!ext4_journal_get_write_access(handle,
4536 BUFFER_TRACE(parent_bh,
4537 "call ext4_handle_dirty_metadata");
4538 ext4_handle_dirty_metadata(handle,
4545 /* We have reached the bottom of the tree. */
4546 BUFFER_TRACE(parent_bh, "free data blocks");
4547 ext4_free_data(handle, inode, parent_bh, first, last);
4551 int ext4_can_truncate(struct inode *inode)
4553 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4555 if (S_ISREG(inode->i_mode))
4557 if (S_ISDIR(inode->i_mode))
4559 if (S_ISLNK(inode->i_mode))
4560 return !ext4_inode_is_fast_symlink(inode);
4567 * We block out ext4_get_block() block instantiations across the entire
4568 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4569 * simultaneously on behalf of the same inode.
4571 * As we work through the truncate and commmit bits of it to the journal there
4572 * is one core, guiding principle: the file's tree must always be consistent on
4573 * disk. We must be able to restart the truncate after a crash.
4575 * The file's tree may be transiently inconsistent in memory (although it
4576 * probably isn't), but whenever we close off and commit a journal transaction,
4577 * the contents of (the filesystem + the journal) must be consistent and
4578 * restartable. It's pretty simple, really: bottom up, right to left (although
4579 * left-to-right works OK too).
4581 * Note that at recovery time, journal replay occurs *before* the restart of
4582 * truncate against the orphan inode list.
4584 * The committed inode has the new, desired i_size (which is the same as
4585 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4586 * that this inode's truncate did not complete and it will again call
4587 * ext4_truncate() to have another go. So there will be instantiated blocks
4588 * to the right of the truncation point in a crashed ext4 filesystem. But
4589 * that's fine - as long as they are linked from the inode, the post-crash
4590 * ext4_truncate() run will find them and release them.
4592 void ext4_truncate(struct inode *inode)
4595 struct ext4_inode_info *ei = EXT4_I(inode);
4596 __le32 *i_data = ei->i_data;
4597 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4598 struct address_space *mapping = inode->i_mapping;
4599 ext4_lblk_t offsets[4];
4604 ext4_lblk_t last_block;
4605 unsigned blocksize = inode->i_sb->s_blocksize;
4607 if (!ext4_can_truncate(inode))
4610 EXT4_I(inode)->i_flags &= ~EXT4_EOFBLOCKS_FL;
4612 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4613 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4615 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4616 ext4_ext_truncate(inode);
4620 handle = start_transaction(inode);
4622 return; /* AKPM: return what? */
4624 last_block = (inode->i_size + blocksize-1)
4625 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4627 if (inode->i_size & (blocksize - 1))
4628 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4631 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4633 goto out_stop; /* error */
4636 * OK. This truncate is going to happen. We add the inode to the
4637 * orphan list, so that if this truncate spans multiple transactions,
4638 * and we crash, we will resume the truncate when the filesystem
4639 * recovers. It also marks the inode dirty, to catch the new size.
4641 * Implication: the file must always be in a sane, consistent
4642 * truncatable state while each transaction commits.
4644 if (ext4_orphan_add(handle, inode))
4648 * From here we block out all ext4_get_block() callers who want to
4649 * modify the block allocation tree.
4651 down_write(&ei->i_data_sem);
4653 ext4_discard_preallocations(inode);
4656 * The orphan list entry will now protect us from any crash which
4657 * occurs before the truncate completes, so it is now safe to propagate
4658 * the new, shorter inode size (held for now in i_size) into the
4659 * on-disk inode. We do this via i_disksize, which is the value which
4660 * ext4 *really* writes onto the disk inode.
4662 ei->i_disksize = inode->i_size;
4664 if (n == 1) { /* direct blocks */
4665 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4666 i_data + EXT4_NDIR_BLOCKS);
4670 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4671 /* Kill the top of shared branch (not detached) */
4673 if (partial == chain) {
4674 /* Shared branch grows from the inode */
4675 ext4_free_branches(handle, inode, NULL,
4676 &nr, &nr+1, (chain+n-1) - partial);
4679 * We mark the inode dirty prior to restart,
4680 * and prior to stop. No need for it here.
4683 /* Shared branch grows from an indirect block */
4684 BUFFER_TRACE(partial->bh, "get_write_access");
4685 ext4_free_branches(handle, inode, partial->bh,
4687 partial->p+1, (chain+n-1) - partial);
4690 /* Clear the ends of indirect blocks on the shared branch */
4691 while (partial > chain) {
4692 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4693 (__le32*)partial->bh->b_data+addr_per_block,
4694 (chain+n-1) - partial);
4695 BUFFER_TRACE(partial->bh, "call brelse");
4696 brelse(partial->bh);
4700 /* Kill the remaining (whole) subtrees */
4701 switch (offsets[0]) {
4703 nr = i_data[EXT4_IND_BLOCK];
4705 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4706 i_data[EXT4_IND_BLOCK] = 0;
4708 case EXT4_IND_BLOCK:
4709 nr = i_data[EXT4_DIND_BLOCK];
4711 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4712 i_data[EXT4_DIND_BLOCK] = 0;
4714 case EXT4_DIND_BLOCK:
4715 nr = i_data[EXT4_TIND_BLOCK];
4717 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4718 i_data[EXT4_TIND_BLOCK] = 0;
4720 case EXT4_TIND_BLOCK:
4724 up_write(&ei->i_data_sem);
4725 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4726 ext4_mark_inode_dirty(handle, inode);
4729 * In a multi-transaction truncate, we only make the final transaction
4733 ext4_handle_sync(handle);
4736 * If this was a simple ftruncate(), and the file will remain alive
4737 * then we need to clear up the orphan record which we created above.
4738 * However, if this was a real unlink then we were called by
4739 * ext4_delete_inode(), and we allow that function to clean up the
4740 * orphan info for us.
4743 ext4_orphan_del(handle, inode);
4745 ext4_journal_stop(handle);
4749 * ext4_get_inode_loc returns with an extra refcount against the inode's
4750 * underlying buffer_head on success. If 'in_mem' is true, we have all
4751 * data in memory that is needed to recreate the on-disk version of this
4754 static int __ext4_get_inode_loc(struct inode *inode,
4755 struct ext4_iloc *iloc, int in_mem)
4757 struct ext4_group_desc *gdp;
4758 struct buffer_head *bh;
4759 struct super_block *sb = inode->i_sb;
4761 int inodes_per_block, inode_offset;
4764 if (!ext4_valid_inum(sb, inode->i_ino))
4767 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4768 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4773 * Figure out the offset within the block group inode table
4775 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4776 inode_offset = ((inode->i_ino - 1) %
4777 EXT4_INODES_PER_GROUP(sb));
4778 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4779 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4781 bh = sb_getblk(sb, block);
4783 ext4_error(sb, "unable to read inode block - "
4784 "inode=%lu, block=%llu", inode->i_ino, block);
4787 if (!buffer_uptodate(bh)) {
4791 * If the buffer has the write error flag, we have failed
4792 * to write out another inode in the same block. In this
4793 * case, we don't have to read the block because we may
4794 * read the old inode data successfully.
4796 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4797 set_buffer_uptodate(bh);
4799 if (buffer_uptodate(bh)) {
4800 /* someone brought it uptodate while we waited */
4806 * If we have all information of the inode in memory and this
4807 * is the only valid inode in the block, we need not read the
4811 struct buffer_head *bitmap_bh;
4814 start = inode_offset & ~(inodes_per_block - 1);
4816 /* Is the inode bitmap in cache? */
4817 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4822 * If the inode bitmap isn't in cache then the
4823 * optimisation may end up performing two reads instead
4824 * of one, so skip it.
4826 if (!buffer_uptodate(bitmap_bh)) {
4830 for (i = start; i < start + inodes_per_block; i++) {
4831 if (i == inode_offset)
4833 if (ext4_test_bit(i, bitmap_bh->b_data))
4837 if (i == start + inodes_per_block) {
4838 /* all other inodes are free, so skip I/O */
4839 memset(bh->b_data, 0, bh->b_size);
4840 set_buffer_uptodate(bh);
4848 * If we need to do any I/O, try to pre-readahead extra
4849 * blocks from the inode table.
4851 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4852 ext4_fsblk_t b, end, table;
4855 table = ext4_inode_table(sb, gdp);
4856 /* s_inode_readahead_blks is always a power of 2 */
4857 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4860 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4861 num = EXT4_INODES_PER_GROUP(sb);
4862 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4863 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4864 num -= ext4_itable_unused_count(sb, gdp);
4865 table += num / inodes_per_block;
4869 sb_breadahead(sb, b++);
4873 * There are other valid inodes in the buffer, this inode
4874 * has in-inode xattrs, or we don't have this inode in memory.
4875 * Read the block from disk.
4878 bh->b_end_io = end_buffer_read_sync;
4879 submit_bh(READ_META, bh);
4881 if (!buffer_uptodate(bh)) {
4882 ext4_error(sb, "unable to read inode block - inode=%lu,"
4883 " block=%llu", inode->i_ino, block);
4893 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4895 /* We have all inode data except xattrs in memory here. */
4896 return __ext4_get_inode_loc(inode, iloc,
4897 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4900 void ext4_set_inode_flags(struct inode *inode)
4902 unsigned int flags = EXT4_I(inode)->i_flags;
4904 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4905 if (flags & EXT4_SYNC_FL)
4906 inode->i_flags |= S_SYNC;
4907 if (flags & EXT4_APPEND_FL)
4908 inode->i_flags |= S_APPEND;
4909 if (flags & EXT4_IMMUTABLE_FL)
4910 inode->i_flags |= S_IMMUTABLE;
4911 if (flags & EXT4_NOATIME_FL)
4912 inode->i_flags |= S_NOATIME;
4913 if (flags & EXT4_DIRSYNC_FL)
4914 inode->i_flags |= S_DIRSYNC;
4917 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4918 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4920 unsigned int flags = ei->vfs_inode.i_flags;
4922 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4923 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4925 ei->i_flags |= EXT4_SYNC_FL;
4926 if (flags & S_APPEND)
4927 ei->i_flags |= EXT4_APPEND_FL;
4928 if (flags & S_IMMUTABLE)
4929 ei->i_flags |= EXT4_IMMUTABLE_FL;
4930 if (flags & S_NOATIME)
4931 ei->i_flags |= EXT4_NOATIME_FL;
4932 if (flags & S_DIRSYNC)
4933 ei->i_flags |= EXT4_DIRSYNC_FL;
4936 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4937 struct ext4_inode_info *ei)
4940 struct inode *inode = &(ei->vfs_inode);
4941 struct super_block *sb = inode->i_sb;
4943 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4944 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4945 /* we are using combined 48 bit field */
4946 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4947 le32_to_cpu(raw_inode->i_blocks_lo);
4948 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4949 /* i_blocks represent file system block size */
4950 return i_blocks << (inode->i_blkbits - 9);
4955 return le32_to_cpu(raw_inode->i_blocks_lo);
4959 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4961 struct ext4_iloc iloc;
4962 struct ext4_inode *raw_inode;
4963 struct ext4_inode_info *ei;
4964 struct inode *inode;
4965 journal_t *journal = EXT4_SB(sb)->s_journal;
4969 inode = iget_locked(sb, ino);
4971 return ERR_PTR(-ENOMEM);
4972 if (!(inode->i_state & I_NEW))
4978 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4981 raw_inode = ext4_raw_inode(&iloc);
4982 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4983 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4984 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4985 if (!(test_opt(inode->i_sb, NO_UID32))) {
4986 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4987 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4989 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4991 ei->i_state_flags = 0;
4992 ei->i_dir_start_lookup = 0;
4993 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4994 /* We now have enough fields to check if the inode was active or not.
4995 * This is needed because nfsd might try to access dead inodes
4996 * the test is that same one that e2fsck uses
4997 * NeilBrown 1999oct15
4999 if (inode->i_nlink == 0) {
5000 if (inode->i_mode == 0 ||
5001 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5002 /* this inode is deleted */
5006 /* The only unlinked inodes we let through here have
5007 * valid i_mode and are being read by the orphan
5008 * recovery code: that's fine, we're about to complete
5009 * the process of deleting those. */
5011 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5012 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5013 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5014 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5016 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5017 inode->i_size = ext4_isize(raw_inode);
5018 ei->i_disksize = inode->i_size;
5020 ei->i_reserved_quota = 0;
5022 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5023 ei->i_block_group = iloc.block_group;
5024 ei->i_last_alloc_group = ~0;
5026 * NOTE! The in-memory inode i_data array is in little-endian order
5027 * even on big-endian machines: we do NOT byteswap the block numbers!
5029 for (block = 0; block < EXT4_N_BLOCKS; block++)
5030 ei->i_data[block] = raw_inode->i_block[block];
5031 INIT_LIST_HEAD(&ei->i_orphan);
5034 * Set transaction id's of transactions that have to be committed
5035 * to finish f[data]sync. We set them to currently running transaction
5036 * as we cannot be sure that the inode or some of its metadata isn't
5037 * part of the transaction - the inode could have been reclaimed and
5038 * now it is reread from disk.
5041 transaction_t *transaction;
5044 spin_lock(&journal->j_state_lock);
5045 if (journal->j_running_transaction)
5046 transaction = journal->j_running_transaction;
5048 transaction = journal->j_committing_transaction;
5050 tid = transaction->t_tid;
5052 tid = journal->j_commit_sequence;
5053 spin_unlock(&journal->j_state_lock);
5054 ei->i_sync_tid = tid;
5055 ei->i_datasync_tid = tid;
5058 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5059 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5060 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5061 EXT4_INODE_SIZE(inode->i_sb)) {
5065 if (ei->i_extra_isize == 0) {
5066 /* The extra space is currently unused. Use it. */
5067 ei->i_extra_isize = sizeof(struct ext4_inode) -
5068 EXT4_GOOD_OLD_INODE_SIZE;
5070 __le32 *magic = (void *)raw_inode +
5071 EXT4_GOOD_OLD_INODE_SIZE +
5073 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5074 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5077 ei->i_extra_isize = 0;
5079 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5080 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5081 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5082 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5084 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5085 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5086 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5088 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5092 if (ei->i_file_acl &&
5093 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5094 ext4_error(sb, "bad extended attribute block %llu inode #%lu",
5095 ei->i_file_acl, inode->i_ino);
5098 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
5099 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5100 (S_ISLNK(inode->i_mode) &&
5101 !ext4_inode_is_fast_symlink(inode)))
5102 /* Validate extent which is part of inode */
5103 ret = ext4_ext_check_inode(inode);
5104 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5105 (S_ISLNK(inode->i_mode) &&
5106 !ext4_inode_is_fast_symlink(inode))) {
5107 /* Validate block references which are part of inode */
5108 ret = ext4_check_inode_blockref(inode);
5113 if (S_ISREG(inode->i_mode)) {
5114 inode->i_op = &ext4_file_inode_operations;
5115 inode->i_fop = &ext4_file_operations;
5116 ext4_set_aops(inode);
5117 } else if (S_ISDIR(inode->i_mode)) {
5118 inode->i_op = &ext4_dir_inode_operations;
5119 inode->i_fop = &ext4_dir_operations;
5120 } else if (S_ISLNK(inode->i_mode)) {
5121 if (ext4_inode_is_fast_symlink(inode)) {
5122 inode->i_op = &ext4_fast_symlink_inode_operations;
5123 nd_terminate_link(ei->i_data, inode->i_size,
5124 sizeof(ei->i_data) - 1);
5126 inode->i_op = &ext4_symlink_inode_operations;
5127 ext4_set_aops(inode);
5129 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5130 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5131 inode->i_op = &ext4_special_inode_operations;
5132 if (raw_inode->i_block[0])
5133 init_special_inode(inode, inode->i_mode,
5134 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5136 init_special_inode(inode, inode->i_mode,
5137 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5140 ext4_error(inode->i_sb, "bogus i_mode (%o) for inode=%lu",
5141 inode->i_mode, inode->i_ino);
5145 ext4_set_inode_flags(inode);
5146 unlock_new_inode(inode);
5152 return ERR_PTR(ret);
5155 static int ext4_inode_blocks_set(handle_t *handle,
5156 struct ext4_inode *raw_inode,
5157 struct ext4_inode_info *ei)
5159 struct inode *inode = &(ei->vfs_inode);
5160 u64 i_blocks = inode->i_blocks;
5161 struct super_block *sb = inode->i_sb;
5163 if (i_blocks <= ~0U) {
5165 * i_blocks can be represnted in a 32 bit variable
5166 * as multiple of 512 bytes
5168 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5169 raw_inode->i_blocks_high = 0;
5170 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
5173 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5176 if (i_blocks <= 0xffffffffffffULL) {
5178 * i_blocks can be represented in a 48 bit variable
5179 * as multiple of 512 bytes
5181 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5182 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5183 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
5185 ei->i_flags |= EXT4_HUGE_FILE_FL;
5186 /* i_block is stored in file system block size */
5187 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5188 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5189 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5195 * Post the struct inode info into an on-disk inode location in the
5196 * buffer-cache. This gobbles the caller's reference to the
5197 * buffer_head in the inode location struct.
5199 * The caller must have write access to iloc->bh.
5201 static int ext4_do_update_inode(handle_t *handle,
5202 struct inode *inode,
5203 struct ext4_iloc *iloc)
5205 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5206 struct ext4_inode_info *ei = EXT4_I(inode);
5207 struct buffer_head *bh = iloc->bh;
5208 int err = 0, rc, block;
5210 /* For fields not not tracking in the in-memory inode,
5211 * initialise them to zero for new inodes. */
5212 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5213 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5215 ext4_get_inode_flags(ei);
5216 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5217 if (!(test_opt(inode->i_sb, NO_UID32))) {
5218 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5219 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5221 * Fix up interoperability with old kernels. Otherwise, old inodes get
5222 * re-used with the upper 16 bits of the uid/gid intact
5225 raw_inode->i_uid_high =
5226 cpu_to_le16(high_16_bits(inode->i_uid));
5227 raw_inode->i_gid_high =
5228 cpu_to_le16(high_16_bits(inode->i_gid));
5230 raw_inode->i_uid_high = 0;
5231 raw_inode->i_gid_high = 0;
5234 raw_inode->i_uid_low =
5235 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5236 raw_inode->i_gid_low =
5237 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5238 raw_inode->i_uid_high = 0;
5239 raw_inode->i_gid_high = 0;
5241 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5243 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5244 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5245 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5246 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5248 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5250 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5251 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5252 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5253 cpu_to_le32(EXT4_OS_HURD))
5254 raw_inode->i_file_acl_high =
5255 cpu_to_le16(ei->i_file_acl >> 32);
5256 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5257 ext4_isize_set(raw_inode, ei->i_disksize);
5258 if (ei->i_disksize > 0x7fffffffULL) {
5259 struct super_block *sb = inode->i_sb;
5260 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5261 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5262 EXT4_SB(sb)->s_es->s_rev_level ==
5263 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5264 /* If this is the first large file
5265 * created, add a flag to the superblock.
5267 err = ext4_journal_get_write_access(handle,
5268 EXT4_SB(sb)->s_sbh);
5271 ext4_update_dynamic_rev(sb);
5272 EXT4_SET_RO_COMPAT_FEATURE(sb,
5273 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5275 ext4_handle_sync(handle);
5276 err = ext4_handle_dirty_metadata(handle, NULL,
5277 EXT4_SB(sb)->s_sbh);
5280 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5281 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5282 if (old_valid_dev(inode->i_rdev)) {
5283 raw_inode->i_block[0] =
5284 cpu_to_le32(old_encode_dev(inode->i_rdev));
5285 raw_inode->i_block[1] = 0;
5287 raw_inode->i_block[0] = 0;
5288 raw_inode->i_block[1] =
5289 cpu_to_le32(new_encode_dev(inode->i_rdev));
5290 raw_inode->i_block[2] = 0;
5293 for (block = 0; block < EXT4_N_BLOCKS; block++)
5294 raw_inode->i_block[block] = ei->i_data[block];
5296 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5297 if (ei->i_extra_isize) {
5298 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5299 raw_inode->i_version_hi =
5300 cpu_to_le32(inode->i_version >> 32);
5301 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5304 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5305 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5308 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5310 ext4_update_inode_fsync_trans(handle, inode, 0);
5313 ext4_std_error(inode->i_sb, err);
5318 * ext4_write_inode()
5320 * We are called from a few places:
5322 * - Within generic_file_write() for O_SYNC files.
5323 * Here, there will be no transaction running. We wait for any running
5324 * trasnaction to commit.
5326 * - Within sys_sync(), kupdate and such.
5327 * We wait on commit, if tol to.
5329 * - Within prune_icache() (PF_MEMALLOC == true)
5330 * Here we simply return. We can't afford to block kswapd on the
5333 * In all cases it is actually safe for us to return without doing anything,
5334 * because the inode has been copied into a raw inode buffer in
5335 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5338 * Note that we are absolutely dependent upon all inode dirtiers doing the
5339 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5340 * which we are interested.
5342 * It would be a bug for them to not do this. The code:
5344 * mark_inode_dirty(inode)
5346 * inode->i_size = expr;
5348 * is in error because a kswapd-driven write_inode() could occur while
5349 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5350 * will no longer be on the superblock's dirty inode list.
5352 int ext4_write_inode(struct inode *inode, int wait)
5356 if (current->flags & PF_MEMALLOC)
5359 if (EXT4_SB(inode->i_sb)->s_journal) {
5360 if (ext4_journal_current_handle()) {
5361 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5369 err = ext4_force_commit(inode->i_sb);
5371 struct ext4_iloc iloc;
5373 err = ext4_get_inode_loc(inode, &iloc);
5377 sync_dirty_buffer(iloc.bh);
5378 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5379 ext4_error(inode->i_sb, "IO error syncing inode, "
5380 "inode=%lu, block=%llu", inode->i_ino,
5381 (unsigned long long)iloc.bh->b_blocknr);
5391 * Called from notify_change.
5393 * We want to trap VFS attempts to truncate the file as soon as
5394 * possible. In particular, we want to make sure that when the VFS
5395 * shrinks i_size, we put the inode on the orphan list and modify
5396 * i_disksize immediately, so that during the subsequent flushing of
5397 * dirty pages and freeing of disk blocks, we can guarantee that any
5398 * commit will leave the blocks being flushed in an unused state on
5399 * disk. (On recovery, the inode will get truncated and the blocks will
5400 * be freed, so we have a strong guarantee that no future commit will
5401 * leave these blocks visible to the user.)
5403 * Another thing we have to assure is that if we are in ordered mode
5404 * and inode is still attached to the committing transaction, we must
5405 * we start writeout of all the dirty pages which are being truncated.
5406 * This way we are sure that all the data written in the previous
5407 * transaction are already on disk (truncate waits for pages under
5410 * Called with inode->i_mutex down.
5412 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5414 struct inode *inode = dentry->d_inode;
5416 const unsigned int ia_valid = attr->ia_valid;
5418 error = inode_change_ok(inode, attr);
5422 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5423 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5426 /* (user+group)*(old+new) structure, inode write (sb,
5427 * inode block, ? - but truncate inode update has it) */
5428 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5429 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5430 if (IS_ERR(handle)) {
5431 error = PTR_ERR(handle);
5434 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
5436 ext4_journal_stop(handle);
5439 /* Update corresponding info in inode so that everything is in
5440 * one transaction */
5441 if (attr->ia_valid & ATTR_UID)
5442 inode->i_uid = attr->ia_uid;
5443 if (attr->ia_valid & ATTR_GID)
5444 inode->i_gid = attr->ia_gid;
5445 error = ext4_mark_inode_dirty(handle, inode);
5446 ext4_journal_stop(handle);
5449 if (attr->ia_valid & ATTR_SIZE) {
5450 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5451 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5453 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5460 if (S_ISREG(inode->i_mode) &&
5461 attr->ia_valid & ATTR_SIZE &&
5462 (attr->ia_size < inode->i_size ||
5463 (EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL))) {
5466 handle = ext4_journal_start(inode, 3);
5467 if (IS_ERR(handle)) {
5468 error = PTR_ERR(handle);
5472 error = ext4_orphan_add(handle, inode);
5473 EXT4_I(inode)->i_disksize = attr->ia_size;
5474 rc = ext4_mark_inode_dirty(handle, inode);
5477 ext4_journal_stop(handle);
5479 if (ext4_should_order_data(inode)) {
5480 error = ext4_begin_ordered_truncate(inode,
5483 /* Do as much error cleanup as possible */
5484 handle = ext4_journal_start(inode, 3);
5485 if (IS_ERR(handle)) {
5486 ext4_orphan_del(NULL, inode);
5489 ext4_orphan_del(handle, inode);
5490 ext4_journal_stop(handle);
5494 /* ext4_truncate will clear the flag */
5495 if ((EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL))
5496 ext4_truncate(inode);
5499 rc = inode_setattr(inode, attr);
5501 /* If inode_setattr's call to ext4_truncate failed to get a
5502 * transaction handle at all, we need to clean up the in-core
5503 * orphan list manually. */
5505 ext4_orphan_del(NULL, inode);
5507 if (!rc && (ia_valid & ATTR_MODE))
5508 rc = ext4_acl_chmod(inode);
5511 ext4_std_error(inode->i_sb, error);
5517 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5520 struct inode *inode;
5521 unsigned long delalloc_blocks;
5523 inode = dentry->d_inode;
5524 generic_fillattr(inode, stat);
5527 * We can't update i_blocks if the block allocation is delayed
5528 * otherwise in the case of system crash before the real block
5529 * allocation is done, we will have i_blocks inconsistent with
5530 * on-disk file blocks.
5531 * We always keep i_blocks updated together with real
5532 * allocation. But to not confuse with user, stat
5533 * will return the blocks that include the delayed allocation
5534 * blocks for this file.
5536 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5537 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5538 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5540 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5544 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5549 /* if nrblocks are contiguous */
5552 * With N contiguous data blocks, it need at most
5553 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5554 * 2 dindirect blocks
5557 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5558 return indirects + 3;
5561 * if nrblocks are not contiguous, worse case, each block touch
5562 * a indirect block, and each indirect block touch a double indirect
5563 * block, plus a triple indirect block
5565 indirects = nrblocks * 2 + 1;
5569 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5571 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
5572 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5573 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5577 * Account for index blocks, block groups bitmaps and block group
5578 * descriptor blocks if modify datablocks and index blocks
5579 * worse case, the indexs blocks spread over different block groups
5581 * If datablocks are discontiguous, they are possible to spread over
5582 * different block groups too. If they are contiuguous, with flexbg,
5583 * they could still across block group boundary.
5585 * Also account for superblock, inode, quota and xattr blocks
5587 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5589 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5595 * How many index blocks need to touch to modify nrblocks?
5596 * The "Chunk" flag indicating whether the nrblocks is
5597 * physically contiguous on disk
5599 * For Direct IO and fallocate, they calls get_block to allocate
5600 * one single extent at a time, so they could set the "Chunk" flag
5602 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5607 * Now let's see how many group bitmaps and group descriptors need
5617 if (groups > ngroups)
5619 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5620 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5622 /* bitmaps and block group descriptor blocks */
5623 ret += groups + gdpblocks;
5625 /* Blocks for super block, inode, quota and xattr blocks */
5626 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5632 * Calulate the total number of credits to reserve to fit
5633 * the modification of a single pages into a single transaction,
5634 * which may include multiple chunks of block allocations.
5636 * This could be called via ext4_write_begin()
5638 * We need to consider the worse case, when
5639 * one new block per extent.
5641 int ext4_writepage_trans_blocks(struct inode *inode)
5643 int bpp = ext4_journal_blocks_per_page(inode);
5646 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5648 /* Account for data blocks for journalled mode */
5649 if (ext4_should_journal_data(inode))
5655 * Calculate the journal credits for a chunk of data modification.
5657 * This is called from DIO, fallocate or whoever calling
5658 * ext4_get_blocks() to map/allocate a chunk of contiguous disk blocks.
5660 * journal buffers for data blocks are not included here, as DIO
5661 * and fallocate do no need to journal data buffers.
5663 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5665 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5669 * The caller must have previously called ext4_reserve_inode_write().
5670 * Give this, we know that the caller already has write access to iloc->bh.
5672 int ext4_mark_iloc_dirty(handle_t *handle,
5673 struct inode *inode, struct ext4_iloc *iloc)
5677 if (test_opt(inode->i_sb, I_VERSION))
5678 inode_inc_iversion(inode);
5680 /* the do_update_inode consumes one bh->b_count */
5683 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5684 err = ext4_do_update_inode(handle, inode, iloc);
5690 * On success, We end up with an outstanding reference count against
5691 * iloc->bh. This _must_ be cleaned up later.
5695 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5696 struct ext4_iloc *iloc)
5700 err = ext4_get_inode_loc(inode, iloc);
5702 BUFFER_TRACE(iloc->bh, "get_write_access");
5703 err = ext4_journal_get_write_access(handle, iloc->bh);
5709 ext4_std_error(inode->i_sb, err);
5714 * Expand an inode by new_extra_isize bytes.
5715 * Returns 0 on success or negative error number on failure.
5717 static int ext4_expand_extra_isize(struct inode *inode,
5718 unsigned int new_extra_isize,
5719 struct ext4_iloc iloc,
5722 struct ext4_inode *raw_inode;
5723 struct ext4_xattr_ibody_header *header;
5724 struct ext4_xattr_entry *entry;
5726 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5729 raw_inode = ext4_raw_inode(&iloc);
5731 header = IHDR(inode, raw_inode);
5732 entry = IFIRST(header);
5734 /* No extended attributes present */
5735 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5736 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5737 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5739 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5743 /* try to expand with EAs present */
5744 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5749 * What we do here is to mark the in-core inode as clean with respect to inode
5750 * dirtiness (it may still be data-dirty).
5751 * This means that the in-core inode may be reaped by prune_icache
5752 * without having to perform any I/O. This is a very good thing,
5753 * because *any* task may call prune_icache - even ones which
5754 * have a transaction open against a different journal.
5756 * Is this cheating? Not really. Sure, we haven't written the
5757 * inode out, but prune_icache isn't a user-visible syncing function.
5758 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5759 * we start and wait on commits.
5761 * Is this efficient/effective? Well, we're being nice to the system
5762 * by cleaning up our inodes proactively so they can be reaped
5763 * without I/O. But we are potentially leaving up to five seconds'
5764 * worth of inodes floating about which prune_icache wants us to
5765 * write out. One way to fix that would be to get prune_icache()
5766 * to do a write_super() to free up some memory. It has the desired
5769 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5771 struct ext4_iloc iloc;
5772 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5773 static unsigned int mnt_count;
5777 err = ext4_reserve_inode_write(handle, inode, &iloc);
5778 if (ext4_handle_valid(handle) &&
5779 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5780 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5782 * We need extra buffer credits since we may write into EA block
5783 * with this same handle. If journal_extend fails, then it will
5784 * only result in a minor loss of functionality for that inode.
5785 * If this is felt to be critical, then e2fsck should be run to
5786 * force a large enough s_min_extra_isize.
5788 if ((jbd2_journal_extend(handle,
5789 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5790 ret = ext4_expand_extra_isize(inode,
5791 sbi->s_want_extra_isize,
5794 ext4_set_inode_state(inode,
5795 EXT4_STATE_NO_EXPAND);
5797 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5798 ext4_warning(inode->i_sb,
5799 "Unable to expand inode %lu. Delete"
5800 " some EAs or run e2fsck.",
5803 le16_to_cpu(sbi->s_es->s_mnt_count);
5809 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5814 * ext4_dirty_inode() is called from __mark_inode_dirty()
5816 * We're really interested in the case where a file is being extended.
5817 * i_size has been changed by generic_commit_write() and we thus need
5818 * to include the updated inode in the current transaction.
5820 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5821 * are allocated to the file.
5823 * If the inode is marked synchronous, we don't honour that here - doing
5824 * so would cause a commit on atime updates, which we don't bother doing.
5825 * We handle synchronous inodes at the highest possible level.
5827 void ext4_dirty_inode(struct inode *inode)
5831 handle = ext4_journal_start(inode, 2);
5835 ext4_mark_inode_dirty(handle, inode);
5837 ext4_journal_stop(handle);
5844 * Bind an inode's backing buffer_head into this transaction, to prevent
5845 * it from being flushed to disk early. Unlike
5846 * ext4_reserve_inode_write, this leaves behind no bh reference and
5847 * returns no iloc structure, so the caller needs to repeat the iloc
5848 * lookup to mark the inode dirty later.
5850 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5852 struct ext4_iloc iloc;
5856 err = ext4_get_inode_loc(inode, &iloc);
5858 BUFFER_TRACE(iloc.bh, "get_write_access");
5859 err = jbd2_journal_get_write_access(handle, iloc.bh);
5861 err = ext4_handle_dirty_metadata(handle,
5867 ext4_std_error(inode->i_sb, err);
5872 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5879 * We have to be very careful here: changing a data block's
5880 * journaling status dynamically is dangerous. If we write a
5881 * data block to the journal, change the status and then delete
5882 * that block, we risk forgetting to revoke the old log record
5883 * from the journal and so a subsequent replay can corrupt data.
5884 * So, first we make sure that the journal is empty and that
5885 * nobody is changing anything.
5888 journal = EXT4_JOURNAL(inode);
5891 if (is_journal_aborted(journal))
5894 jbd2_journal_lock_updates(journal);
5895 jbd2_journal_flush(journal);
5898 * OK, there are no updates running now, and all cached data is
5899 * synced to disk. We are now in a completely consistent state
5900 * which doesn't have anything in the journal, and we know that
5901 * no filesystem updates are running, so it is safe to modify
5902 * the inode's in-core data-journaling state flag now.
5906 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5908 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5909 ext4_set_aops(inode);
5911 jbd2_journal_unlock_updates(journal);
5913 /* Finally we can mark the inode as dirty. */
5915 handle = ext4_journal_start(inode, 1);
5917 return PTR_ERR(handle);
5919 err = ext4_mark_inode_dirty(handle, inode);
5920 ext4_handle_sync(handle);
5921 ext4_journal_stop(handle);
5922 ext4_std_error(inode->i_sb, err);
5927 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5929 return !buffer_mapped(bh);
5932 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5934 struct page *page = vmf->page;
5939 struct file *file = vma->vm_file;
5940 struct inode *inode = file->f_path.dentry->d_inode;
5941 struct address_space *mapping = inode->i_mapping;
5944 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5945 * get i_mutex because we are already holding mmap_sem.
5947 down_read(&inode->i_alloc_sem);
5948 size = i_size_read(inode);
5949 if (page->mapping != mapping || size <= page_offset(page)
5950 || !PageUptodate(page)) {
5951 /* page got truncated from under us? */
5955 if (PageMappedToDisk(page))
5958 if (page->index == size >> PAGE_CACHE_SHIFT)
5959 len = size & ~PAGE_CACHE_MASK;
5961 len = PAGE_CACHE_SIZE;
5965 * return if we have all the buffers mapped. This avoid
5966 * the need to call write_begin/write_end which does a
5967 * journal_start/journal_stop which can block and take
5970 if (page_has_buffers(page)) {
5971 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5972 ext4_bh_unmapped)) {
5979 * OK, we need to fill the hole... Do write_begin write_end
5980 * to do block allocation/reservation.We are not holding
5981 * inode.i__mutex here. That allow * parallel write_begin,
5982 * write_end call. lock_page prevent this from happening
5983 * on the same page though
5985 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5986 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5989 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5990 len, len, page, fsdata);
5996 ret = VM_FAULT_SIGBUS;
5997 up_read(&inode->i_alloc_sem);