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>
42 #include <linux/slab.h>
44 #include "ext4_jbd2.h"
47 #include "ext4_extents.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static inline int ext4_begin_ordered_truncate(struct inode *inode,
56 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode->i_sb)->s_journal,
58 &EXT4_I(inode)->jinode,
62 static void ext4_invalidatepage(struct page *page, unsigned long offset);
65 * Test whether an inode is a fast symlink.
67 static int ext4_inode_is_fast_symlink(struct inode *inode)
69 int ea_blocks = EXT4_I(inode)->i_file_acl ?
70 (inode->i_sb->s_blocksize >> 9) : 0;
72 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
76 * Work out how many blocks we need to proceed with the next chunk of a
77 * truncate transaction.
79 static unsigned long blocks_for_truncate(struct inode *inode)
83 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
85 /* Give ourselves just enough room to cope with inodes in which
86 * i_blocks is corrupt: we've seen disk corruptions in the past
87 * which resulted in random data in an inode which looked enough
88 * like a regular file for ext4 to try to delete it. Things
89 * will go a bit crazy if that happens, but at least we should
90 * try not to panic the whole kernel. */
94 /* But we need to bound the transaction so we don't overflow the
96 if (needed > EXT4_MAX_TRANS_DATA)
97 needed = EXT4_MAX_TRANS_DATA;
99 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
103 * Truncate transactions can be complex and absolutely huge. So we need to
104 * be able to restart the transaction at a conventient checkpoint to make
105 * sure we don't overflow the journal.
107 * start_transaction gets us a new handle for a truncate transaction,
108 * and extend_transaction tries to extend the existing one a bit. If
109 * extend fails, we need to propagate the failure up and restart the
110 * transaction in the top-level truncate loop. --sct
112 static handle_t *start_transaction(struct inode *inode)
116 result = ext4_journal_start(inode, blocks_for_truncate(inode));
120 ext4_std_error(inode->i_sb, PTR_ERR(result));
125 * Try to extend this transaction for the purposes of truncation.
127 * Returns 0 if we managed to create more room. If we can't create more
128 * room, and the transaction must be restarted we return 1.
130 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
132 if (!ext4_handle_valid(handle))
134 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
136 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
142 * Restart the transaction associated with *handle. This does a commit,
143 * so before we call here everything must be consistently dirtied against
146 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
152 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
153 * moment, get_block can be called only for blocks inside i_size since
154 * page cache has been already dropped and writes are blocked by
155 * i_mutex. So we can safely drop the i_data_sem here.
157 BUG_ON(EXT4_JOURNAL(inode) == NULL);
158 jbd_debug(2, "restarting handle %p\n", handle);
159 up_write(&EXT4_I(inode)->i_data_sem);
160 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
161 down_write(&EXT4_I(inode)->i_data_sem);
162 ext4_discard_preallocations(inode);
168 * Called at the last iput() if i_nlink is zero.
170 void ext4_delete_inode(struct inode *inode)
175 if (!is_bad_inode(inode))
176 dquot_initialize(inode);
178 if (ext4_should_order_data(inode))
179 ext4_begin_ordered_truncate(inode, 0);
180 truncate_inode_pages(&inode->i_data, 0);
182 if (is_bad_inode(inode))
185 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
186 if (IS_ERR(handle)) {
187 ext4_std_error(inode->i_sb, PTR_ERR(handle));
189 * If we're going to skip the normal cleanup, we still need to
190 * make sure that the in-core orphan linked list is properly
193 ext4_orphan_del(NULL, inode);
198 ext4_handle_sync(handle);
200 err = ext4_mark_inode_dirty(handle, inode);
202 ext4_warning(inode->i_sb,
203 "couldn't mark inode dirty (err %d)", err);
207 ext4_truncate(inode);
210 * ext4_ext_truncate() doesn't reserve any slop when it
211 * restarts journal transactions; therefore there may not be
212 * enough credits left in the handle to remove the inode from
213 * the orphan list and set the dtime field.
215 if (!ext4_handle_has_enough_credits(handle, 3)) {
216 err = ext4_journal_extend(handle, 3);
218 err = ext4_journal_restart(handle, 3);
220 ext4_warning(inode->i_sb,
221 "couldn't extend journal (err %d)", err);
223 ext4_journal_stop(handle);
224 ext4_orphan_del(NULL, inode);
230 * Kill off the orphan record which ext4_truncate created.
231 * AKPM: I think this can be inside the above `if'.
232 * Note that ext4_orphan_del() has to be able to cope with the
233 * deletion of a non-existent orphan - this is because we don't
234 * know if ext4_truncate() actually created an orphan record.
235 * (Well, we could do this if we need to, but heck - it works)
237 ext4_orphan_del(handle, inode);
238 EXT4_I(inode)->i_dtime = get_seconds();
241 * One subtle ordering requirement: if anything has gone wrong
242 * (transaction abort, IO errors, whatever), then we can still
243 * do these next steps (the fs will already have been marked as
244 * having errors), but we can't free the inode if the mark_dirty
247 if (ext4_mark_inode_dirty(handle, inode))
248 /* If that failed, just do the required in-core inode clear. */
251 ext4_free_inode(handle, inode);
252 ext4_journal_stop(handle);
255 clear_inode(inode); /* We must guarantee clearing of inode... */
261 struct buffer_head *bh;
264 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
266 p->key = *(p->p = v);
271 * ext4_block_to_path - parse the block number into array of offsets
272 * @inode: inode in question (we are only interested in its superblock)
273 * @i_block: block number to be parsed
274 * @offsets: array to store the offsets in
275 * @boundary: set this non-zero if the referred-to block is likely to be
276 * followed (on disk) by an indirect block.
278 * To store the locations of file's data ext4 uses a data structure common
279 * for UNIX filesystems - tree of pointers anchored in the inode, with
280 * data blocks at leaves and indirect blocks in intermediate nodes.
281 * This function translates the block number into path in that tree -
282 * return value is the path length and @offsets[n] is the offset of
283 * pointer to (n+1)th node in the nth one. If @block is out of range
284 * (negative or too large) warning is printed and zero returned.
286 * Note: function doesn't find node addresses, so no IO is needed. All
287 * we need to know is the capacity of indirect blocks (taken from the
292 * Portability note: the last comparison (check that we fit into triple
293 * indirect block) is spelled differently, because otherwise on an
294 * architecture with 32-bit longs and 8Kb pages we might get into trouble
295 * if our filesystem had 8Kb blocks. We might use long long, but that would
296 * kill us on x86. Oh, well, at least the sign propagation does not matter -
297 * i_block would have to be negative in the very beginning, so we would not
301 static int ext4_block_to_path(struct inode *inode,
303 ext4_lblk_t offsets[4], int *boundary)
305 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
306 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
307 const long direct_blocks = EXT4_NDIR_BLOCKS,
308 indirect_blocks = ptrs,
309 double_blocks = (1 << (ptrs_bits * 2));
313 if (i_block < direct_blocks) {
314 offsets[n++] = i_block;
315 final = direct_blocks;
316 } else if ((i_block -= direct_blocks) < indirect_blocks) {
317 offsets[n++] = EXT4_IND_BLOCK;
318 offsets[n++] = i_block;
320 } else if ((i_block -= indirect_blocks) < double_blocks) {
321 offsets[n++] = EXT4_DIND_BLOCK;
322 offsets[n++] = i_block >> ptrs_bits;
323 offsets[n++] = i_block & (ptrs - 1);
325 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
326 offsets[n++] = EXT4_TIND_BLOCK;
327 offsets[n++] = i_block >> (ptrs_bits * 2);
328 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
329 offsets[n++] = i_block & (ptrs - 1);
332 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
333 i_block + direct_blocks +
334 indirect_blocks + double_blocks, inode->i_ino);
337 *boundary = final - 1 - (i_block & (ptrs - 1));
341 static int __ext4_check_blockref(const char *function, unsigned int line,
343 __le32 *p, unsigned int max)
345 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
349 while (bref < p+max) {
350 blk = le32_to_cpu(*bref++);
352 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
354 es->s_last_error_block = cpu_to_le64(blk);
355 ext4_error_inode(inode, function, line, blk,
364 #define ext4_check_indirect_blockref(inode, bh) \
365 __ext4_check_blockref(__func__, __LINE__, inode, \
366 (__le32 *)(bh)->b_data, \
367 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
369 #define ext4_check_inode_blockref(inode) \
370 __ext4_check_blockref(__func__, __LINE__, inode, \
371 EXT4_I(inode)->i_data, \
375 * ext4_get_branch - read the chain of indirect blocks leading to data
376 * @inode: inode in question
377 * @depth: depth of the chain (1 - direct pointer, etc.)
378 * @offsets: offsets of pointers in inode/indirect blocks
379 * @chain: place to store the result
380 * @err: here we store the error value
382 * Function fills the array of triples <key, p, bh> and returns %NULL
383 * if everything went OK or the pointer to the last filled triple
384 * (incomplete one) otherwise. Upon the return chain[i].key contains
385 * the number of (i+1)-th block in the chain (as it is stored in memory,
386 * i.e. little-endian 32-bit), chain[i].p contains the address of that
387 * number (it points into struct inode for i==0 and into the bh->b_data
388 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
389 * block for i>0 and NULL for i==0. In other words, it holds the block
390 * numbers of the chain, addresses they were taken from (and where we can
391 * verify that chain did not change) and buffer_heads hosting these
394 * Function stops when it stumbles upon zero pointer (absent block)
395 * (pointer to last triple returned, *@err == 0)
396 * or when it gets an IO error reading an indirect block
397 * (ditto, *@err == -EIO)
398 * or when it reads all @depth-1 indirect blocks successfully and finds
399 * the whole chain, all way to the data (returns %NULL, *err == 0).
401 * Need to be called with
402 * down_read(&EXT4_I(inode)->i_data_sem)
404 static Indirect *ext4_get_branch(struct inode *inode, int depth,
405 ext4_lblk_t *offsets,
406 Indirect chain[4], int *err)
408 struct super_block *sb = inode->i_sb;
410 struct buffer_head *bh;
413 /* i_data is not going away, no lock needed */
414 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
418 bh = sb_getblk(sb, le32_to_cpu(p->key));
422 if (!bh_uptodate_or_lock(bh)) {
423 if (bh_submit_read(bh) < 0) {
427 /* validate block references */
428 if (ext4_check_indirect_blockref(inode, bh)) {
434 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
448 * ext4_find_near - find a place for allocation with sufficient locality
450 * @ind: descriptor of indirect block.
452 * This function returns the preferred place for block allocation.
453 * It is used when heuristic for sequential allocation fails.
455 * + if there is a block to the left of our position - allocate near it.
456 * + if pointer will live in indirect block - allocate near that block.
457 * + if pointer will live in inode - allocate in the same
460 * In the latter case we colour the starting block by the callers PID to
461 * prevent it from clashing with concurrent allocations for a different inode
462 * in the same block group. The PID is used here so that functionally related
463 * files will be close-by on-disk.
465 * Caller must make sure that @ind is valid and will stay that way.
467 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
469 struct ext4_inode_info *ei = EXT4_I(inode);
470 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
472 ext4_fsblk_t bg_start;
473 ext4_fsblk_t last_block;
474 ext4_grpblk_t colour;
475 ext4_group_t block_group;
476 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
478 /* Try to find previous block */
479 for (p = ind->p - 1; p >= start; p--) {
481 return le32_to_cpu(*p);
484 /* No such thing, so let's try location of indirect block */
486 return ind->bh->b_blocknr;
489 * It is going to be referred to from the inode itself? OK, just put it
490 * into the same cylinder group then.
492 block_group = ei->i_block_group;
493 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
494 block_group &= ~(flex_size-1);
495 if (S_ISREG(inode->i_mode))
498 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
499 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
502 * If we are doing delayed allocation, we don't need take
503 * colour into account.
505 if (test_opt(inode->i_sb, DELALLOC))
508 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
509 colour = (current->pid % 16) *
510 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
512 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
513 return bg_start + colour;
517 * ext4_find_goal - find a preferred place for allocation.
519 * @block: block we want
520 * @partial: pointer to the last triple within a chain
522 * Normally this function find the preferred place for block allocation,
524 * Because this is only used for non-extent files, we limit the block nr
527 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
533 * XXX need to get goal block from mballoc's data structures
536 goal = ext4_find_near(inode, partial);
537 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
542 * ext4_blks_to_allocate: Look up the block map and count the number
543 * of direct blocks need to be allocated for the given branch.
545 * @branch: chain of indirect blocks
546 * @k: number of blocks need for indirect blocks
547 * @blks: number of data blocks to be mapped.
548 * @blocks_to_boundary: the offset in the indirect block
550 * return the total number of blocks to be allocate, including the
551 * direct and indirect blocks.
553 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
554 int blocks_to_boundary)
556 unsigned int count = 0;
559 * Simple case, [t,d]Indirect block(s) has not allocated yet
560 * then it's clear blocks on that path have not allocated
563 /* right now we don't handle cross boundary allocation */
564 if (blks < blocks_to_boundary + 1)
567 count += blocks_to_boundary + 1;
572 while (count < blks && count <= blocks_to_boundary &&
573 le32_to_cpu(*(branch[0].p + count)) == 0) {
580 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
581 * @indirect_blks: the number of blocks need to allocate for indirect
584 * @new_blocks: on return it will store the new block numbers for
585 * the indirect blocks(if needed) and the first direct block,
586 * @blks: on return it will store the total number of allocated
589 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
590 ext4_lblk_t iblock, ext4_fsblk_t goal,
591 int indirect_blks, int blks,
592 ext4_fsblk_t new_blocks[4], int *err)
594 struct ext4_allocation_request ar;
596 unsigned long count = 0, blk_allocated = 0;
598 ext4_fsblk_t current_block = 0;
602 * Here we try to allocate the requested multiple blocks at once,
603 * on a best-effort basis.
604 * To build a branch, we should allocate blocks for
605 * the indirect blocks(if not allocated yet), and at least
606 * the first direct block of this branch. That's the
607 * minimum number of blocks need to allocate(required)
609 /* first we try to allocate the indirect blocks */
610 target = indirect_blks;
613 /* allocating blocks for indirect blocks and direct blocks */
614 current_block = ext4_new_meta_blocks(handle, inode,
619 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
620 EXT4_ERROR_INODE(inode,
621 "current_block %llu + count %lu > %d!",
622 current_block, count,
623 EXT4_MAX_BLOCK_FILE_PHYS);
629 /* allocate blocks for indirect blocks */
630 while (index < indirect_blks && count) {
631 new_blocks[index++] = current_block++;
636 * save the new block number
637 * for the first direct block
639 new_blocks[index] = current_block;
640 printk(KERN_INFO "%s returned more blocks than "
641 "requested\n", __func__);
647 target = blks - count ;
648 blk_allocated = count;
651 /* Now allocate data blocks */
652 memset(&ar, 0, sizeof(ar));
657 if (S_ISREG(inode->i_mode))
658 /* enable in-core preallocation only for regular files */
659 ar.flags = EXT4_MB_HINT_DATA;
661 current_block = ext4_mb_new_blocks(handle, &ar, err);
662 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
663 EXT4_ERROR_INODE(inode,
664 "current_block %llu + ar.len %d > %d!",
665 current_block, ar.len,
666 EXT4_MAX_BLOCK_FILE_PHYS);
671 if (*err && (target == blks)) {
673 * if the allocation failed and we didn't allocate
679 if (target == blks) {
681 * save the new block number
682 * for the first direct block
684 new_blocks[index] = current_block;
686 blk_allocated += ar.len;
689 /* total number of blocks allocated for direct blocks */
694 for (i = 0; i < index; i++)
695 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
700 * ext4_alloc_branch - allocate and set up a chain of blocks.
702 * @indirect_blks: number of allocated indirect blocks
703 * @blks: number of allocated direct blocks
704 * @offsets: offsets (in the blocks) to store the pointers to next.
705 * @branch: place to store the chain in.
707 * This function allocates blocks, zeroes out all but the last one,
708 * links them into chain and (if we are synchronous) writes them to disk.
709 * In other words, it prepares a branch that can be spliced onto the
710 * inode. It stores the information about that chain in the branch[], in
711 * the same format as ext4_get_branch() would do. We are calling it after
712 * we had read the existing part of chain and partial points to the last
713 * triple of that (one with zero ->key). Upon the exit we have the same
714 * picture as after the successful ext4_get_block(), except that in one
715 * place chain is disconnected - *branch->p is still zero (we did not
716 * set the last link), but branch->key contains the number that should
717 * be placed into *branch->p to fill that gap.
719 * If allocation fails we free all blocks we've allocated (and forget
720 * their buffer_heads) and return the error value the from failed
721 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
722 * as described above and return 0.
724 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
725 ext4_lblk_t iblock, int indirect_blks,
726 int *blks, ext4_fsblk_t goal,
727 ext4_lblk_t *offsets, Indirect *branch)
729 int blocksize = inode->i_sb->s_blocksize;
732 struct buffer_head *bh;
734 ext4_fsblk_t new_blocks[4];
735 ext4_fsblk_t current_block;
737 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
738 *blks, new_blocks, &err);
742 branch[0].key = cpu_to_le32(new_blocks[0]);
744 * metadata blocks and data blocks are allocated.
746 for (n = 1; n <= indirect_blks; n++) {
748 * Get buffer_head for parent block, zero it out
749 * and set the pointer to new one, then send
752 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
755 BUFFER_TRACE(bh, "call get_create_access");
756 err = ext4_journal_get_create_access(handle, bh);
758 /* Don't brelse(bh) here; it's done in
759 * ext4_journal_forget() below */
764 memset(bh->b_data, 0, blocksize);
765 branch[n].p = (__le32 *) bh->b_data + offsets[n];
766 branch[n].key = cpu_to_le32(new_blocks[n]);
767 *branch[n].p = branch[n].key;
768 if (n == indirect_blks) {
769 current_block = new_blocks[n];
771 * End of chain, update the last new metablock of
772 * the chain to point to the new allocated
773 * data blocks numbers
775 for (i = 1; i < num; i++)
776 *(branch[n].p + i) = cpu_to_le32(++current_block);
778 BUFFER_TRACE(bh, "marking uptodate");
779 set_buffer_uptodate(bh);
782 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
783 err = ext4_handle_dirty_metadata(handle, inode, bh);
790 /* Allocation failed, free what we already allocated */
791 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
792 for (i = 1; i <= n ; i++) {
794 * branch[i].bh is newly allocated, so there is no
795 * need to revoke the block, which is why we don't
796 * need to set EXT4_FREE_BLOCKS_METADATA.
798 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
799 EXT4_FREE_BLOCKS_FORGET);
801 for (i = n+1; i < indirect_blks; i++)
802 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
804 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
810 * ext4_splice_branch - splice the allocated branch onto inode.
812 * @block: (logical) number of block we are adding
813 * @chain: chain of indirect blocks (with a missing link - see
815 * @where: location of missing link
816 * @num: number of indirect blocks we are adding
817 * @blks: number of direct blocks we are adding
819 * This function fills the missing link and does all housekeeping needed in
820 * inode (->i_blocks, etc.). In case of success we end up with the full
821 * chain to new block and return 0.
823 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
824 ext4_lblk_t block, Indirect *where, int num,
829 ext4_fsblk_t current_block;
832 * If we're splicing into a [td]indirect block (as opposed to the
833 * inode) then we need to get write access to the [td]indirect block
837 BUFFER_TRACE(where->bh, "get_write_access");
838 err = ext4_journal_get_write_access(handle, where->bh);
844 *where->p = where->key;
847 * Update the host buffer_head or inode to point to more just allocated
848 * direct blocks blocks
850 if (num == 0 && blks > 1) {
851 current_block = le32_to_cpu(where->key) + 1;
852 for (i = 1; i < blks; i++)
853 *(where->p + i) = cpu_to_le32(current_block++);
856 /* We are done with atomic stuff, now do the rest of housekeeping */
857 /* had we spliced it onto indirect block? */
860 * If we spliced it onto an indirect block, we haven't
861 * altered the inode. Note however that if it is being spliced
862 * onto an indirect block at the very end of the file (the
863 * file is growing) then we *will* alter the inode to reflect
864 * the new i_size. But that is not done here - it is done in
865 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
867 jbd_debug(5, "splicing indirect only\n");
868 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
869 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
874 * OK, we spliced it into the inode itself on a direct block.
876 ext4_mark_inode_dirty(handle, inode);
877 jbd_debug(5, "splicing direct\n");
882 for (i = 1; i <= num; i++) {
884 * branch[i].bh is newly allocated, so there is no
885 * need to revoke the block, which is why we don't
886 * need to set EXT4_FREE_BLOCKS_METADATA.
888 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
889 EXT4_FREE_BLOCKS_FORGET);
891 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
898 * The ext4_ind_map_blocks() function handles non-extents inodes
899 * (i.e., using the traditional indirect/double-indirect i_blocks
900 * scheme) for ext4_map_blocks().
902 * Allocation strategy is simple: if we have to allocate something, we will
903 * have to go the whole way to leaf. So let's do it before attaching anything
904 * to tree, set linkage between the newborn blocks, write them if sync is
905 * required, recheck the path, free and repeat if check fails, otherwise
906 * set the last missing link (that will protect us from any truncate-generated
907 * removals - all blocks on the path are immune now) and possibly force the
908 * write on the parent block.
909 * That has a nice additional property: no special recovery from the failed
910 * allocations is needed - we simply release blocks and do not touch anything
911 * reachable from inode.
913 * `handle' can be NULL if create == 0.
915 * return > 0, # of blocks mapped or allocated.
916 * return = 0, if plain lookup failed.
917 * return < 0, error case.
919 * The ext4_ind_get_blocks() function should be called with
920 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
921 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
922 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
925 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
926 struct ext4_map_blocks *map,
930 ext4_lblk_t offsets[4];
935 int blocks_to_boundary = 0;
938 ext4_fsblk_t first_block = 0;
940 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
941 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
942 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
943 &blocks_to_boundary);
948 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
950 /* Simplest case - block found, no allocation needed */
952 first_block = le32_to_cpu(chain[depth - 1].key);
955 while (count < map->m_len && count <= blocks_to_boundary) {
958 blk = le32_to_cpu(*(chain[depth-1].p + count));
960 if (blk == first_block + count)
968 /* Next simple case - plain lookup or failed read of indirect block */
969 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
973 * Okay, we need to do block allocation.
975 goal = ext4_find_goal(inode, map->m_lblk, partial);
977 /* the number of blocks need to allocate for [d,t]indirect blocks */
978 indirect_blks = (chain + depth) - partial - 1;
981 * Next look up the indirect map to count the totoal number of
982 * direct blocks to allocate for this branch.
984 count = ext4_blks_to_allocate(partial, indirect_blks,
985 map->m_len, blocks_to_boundary);
987 * Block out ext4_truncate while we alter the tree
989 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
991 offsets + (partial - chain), partial);
994 * The ext4_splice_branch call will free and forget any buffers
995 * on the new chain if there is a failure, but that risks using
996 * up transaction credits, especially for bitmaps where the
997 * credits cannot be returned. Can we handle this somehow? We
998 * may need to return -EAGAIN upwards in the worst case. --sct
1001 err = ext4_splice_branch(handle, inode, map->m_lblk,
1002 partial, indirect_blks, count);
1006 map->m_flags |= EXT4_MAP_NEW;
1008 ext4_update_inode_fsync_trans(handle, inode, 1);
1010 map->m_flags |= EXT4_MAP_MAPPED;
1011 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1013 if (count > blocks_to_boundary)
1014 map->m_flags |= EXT4_MAP_BOUNDARY;
1016 /* Clean up and exit */
1017 partial = chain + depth - 1; /* the whole chain */
1019 while (partial > chain) {
1020 BUFFER_TRACE(partial->bh, "call brelse");
1021 brelse(partial->bh);
1029 qsize_t *ext4_get_reserved_space(struct inode *inode)
1031 return &EXT4_I(inode)->i_reserved_quota;
1036 * Calculate the number of metadata blocks need to reserve
1037 * to allocate a new block at @lblocks for non extent file based file
1039 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1042 struct ext4_inode_info *ei = EXT4_I(inode);
1043 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1046 if (lblock < EXT4_NDIR_BLOCKS)
1049 lblock -= EXT4_NDIR_BLOCKS;
1051 if (ei->i_da_metadata_calc_len &&
1052 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1053 ei->i_da_metadata_calc_len++;
1056 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1057 ei->i_da_metadata_calc_len = 1;
1058 blk_bits = order_base_2(lblock);
1059 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1063 * Calculate the number of metadata blocks need to reserve
1064 * to allocate a block located at @lblock
1066 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1068 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1069 return ext4_ext_calc_metadata_amount(inode, lblock);
1071 return ext4_indirect_calc_metadata_amount(inode, lblock);
1075 * Called with i_data_sem down, which is important since we can call
1076 * ext4_discard_preallocations() from here.
1078 void ext4_da_update_reserve_space(struct inode *inode,
1079 int used, int quota_claim)
1081 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1082 struct ext4_inode_info *ei = EXT4_I(inode);
1084 spin_lock(&ei->i_block_reservation_lock);
1085 trace_ext4_da_update_reserve_space(inode, used);
1086 if (unlikely(used > ei->i_reserved_data_blocks)) {
1087 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1088 "with only %d reserved data blocks\n",
1089 __func__, inode->i_ino, used,
1090 ei->i_reserved_data_blocks);
1092 used = ei->i_reserved_data_blocks;
1095 /* Update per-inode reservations */
1096 ei->i_reserved_data_blocks -= used;
1097 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1098 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1099 used + ei->i_allocated_meta_blocks);
1100 ei->i_allocated_meta_blocks = 0;
1102 if (ei->i_reserved_data_blocks == 0) {
1104 * We can release all of the reserved metadata blocks
1105 * only when we have written all of the delayed
1106 * allocation blocks.
1108 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1109 ei->i_reserved_meta_blocks);
1110 ei->i_reserved_meta_blocks = 0;
1111 ei->i_da_metadata_calc_len = 0;
1113 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1115 /* Update quota subsystem for data blocks */
1117 dquot_claim_block(inode, used);
1120 * We did fallocate with an offset that is already delayed
1121 * allocated. So on delayed allocated writeback we should
1122 * not re-claim the quota for fallocated blocks.
1124 dquot_release_reservation_block(inode, used);
1128 * If we have done all the pending block allocations and if
1129 * there aren't any writers on the inode, we can discard the
1130 * inode's preallocations.
1132 if ((ei->i_reserved_data_blocks == 0) &&
1133 (atomic_read(&inode->i_writecount) == 0))
1134 ext4_discard_preallocations(inode);
1137 static int __check_block_validity(struct inode *inode, const char *func,
1139 struct ext4_map_blocks *map)
1141 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1143 ext4_error_inode(inode, func, line, map->m_pblk,
1144 "lblock %lu mapped to illegal pblock "
1145 "(length %d)", (unsigned long) map->m_lblk,
1152 #define check_block_validity(inode, map) \
1153 __check_block_validity((inode), __func__, __LINE__, (map))
1156 * Return the number of contiguous dirty pages in a given inode
1157 * starting at page frame idx.
1159 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1160 unsigned int max_pages)
1162 struct address_space *mapping = inode->i_mapping;
1164 struct pagevec pvec;
1166 int i, nr_pages, done = 0;
1170 pagevec_init(&pvec, 0);
1173 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1174 PAGECACHE_TAG_DIRTY,
1175 (pgoff_t)PAGEVEC_SIZE);
1178 for (i = 0; i < nr_pages; i++) {
1179 struct page *page = pvec.pages[i];
1180 struct buffer_head *bh, *head;
1183 if (unlikely(page->mapping != mapping) ||
1185 PageWriteback(page) ||
1186 page->index != idx) {
1191 if (page_has_buffers(page)) {
1192 bh = head = page_buffers(page);
1194 if (!buffer_delay(bh) &&
1195 !buffer_unwritten(bh))
1197 bh = bh->b_this_page;
1198 } while (!done && (bh != head));
1205 if (num >= max_pages)
1208 pagevec_release(&pvec);
1214 * The ext4_map_blocks() function tries to look up the requested blocks,
1215 * and returns if the blocks are already mapped.
1217 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1218 * and store the allocated blocks in the result buffer head and mark it
1221 * If file type is extents based, it will call ext4_ext_map_blocks(),
1222 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1225 * On success, it returns the number of blocks being mapped or allocate.
1226 * if create==0 and the blocks are pre-allocated and uninitialized block,
1227 * the result buffer head is unmapped. If the create ==1, it will make sure
1228 * the buffer head is mapped.
1230 * It returns 0 if plain look up failed (blocks have not been allocated), in
1231 * that casem, buffer head is unmapped
1233 * It returns the error in case of allocation failure.
1235 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1236 struct ext4_map_blocks *map, int flags)
1241 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1242 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1243 (unsigned long) map->m_lblk);
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_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1250 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1252 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1254 up_read((&EXT4_I(inode)->i_data_sem));
1256 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1257 int ret = check_block_validity(inode, map);
1262 /* If it is only a block(s) look up */
1263 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1267 * Returns if the blocks have already allocated
1269 * Note that if blocks have been preallocated
1270 * ext4_ext_get_block() returns th create = 0
1271 * with buffer head unmapped.
1273 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1277 * When we call get_blocks without the create flag, the
1278 * BH_Unwritten flag could have gotten set if the blocks
1279 * requested were part of a uninitialized extent. We need to
1280 * clear this flag now that we are committed to convert all or
1281 * part of the uninitialized extent to be an initialized
1282 * extent. This is because we need to avoid the combination
1283 * of BH_Unwritten and BH_Mapped flags being simultaneously
1284 * set on the buffer_head.
1286 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1289 * New blocks allocate and/or writing to uninitialized extent
1290 * will possibly result in updating i_data, so we take
1291 * the write lock of i_data_sem, and call get_blocks()
1292 * with create == 1 flag.
1294 down_write((&EXT4_I(inode)->i_data_sem));
1297 * if the caller is from delayed allocation writeout path
1298 * we have already reserved fs blocks for allocation
1299 * let the underlying get_block() function know to
1300 * avoid double accounting
1302 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1303 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1305 * We need to check for EXT4 here because migrate
1306 * could have changed the inode type in between
1308 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1309 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1311 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1313 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1315 * We allocated new blocks which will result in
1316 * i_data's format changing. Force the migrate
1317 * to fail by clearing migrate flags
1319 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1323 * Update reserved blocks/metadata blocks after successful
1324 * block allocation which had been deferred till now. We don't
1325 * support fallocate for non extent files. So we can update
1326 * reserve space here.
1329 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1330 ext4_da_update_reserve_space(inode, retval, 1);
1332 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1333 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1335 up_write((&EXT4_I(inode)->i_data_sem));
1336 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1337 int ret = check_block_validity(inode, map);
1344 /* Maximum number of blocks we map for direct IO at once. */
1345 #define DIO_MAX_BLOCKS 4096
1347 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1348 struct buffer_head *bh, int flags)
1350 handle_t *handle = ext4_journal_current_handle();
1351 struct ext4_map_blocks map;
1352 int ret = 0, started = 0;
1355 map.m_lblk = iblock;
1356 map.m_len = bh->b_size >> inode->i_blkbits;
1358 if (flags && !handle) {
1359 /* Direct IO write... */
1360 if (map.m_len > DIO_MAX_BLOCKS)
1361 map.m_len = DIO_MAX_BLOCKS;
1362 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1363 handle = ext4_journal_start(inode, dio_credits);
1364 if (IS_ERR(handle)) {
1365 ret = PTR_ERR(handle);
1371 ret = ext4_map_blocks(handle, inode, &map, flags);
1373 map_bh(bh, inode->i_sb, map.m_pblk);
1374 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1375 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1379 ext4_journal_stop(handle);
1383 int ext4_get_block(struct inode *inode, sector_t iblock,
1384 struct buffer_head *bh, int create)
1386 return _ext4_get_block(inode, iblock, bh,
1387 create ? EXT4_GET_BLOCKS_CREATE : 0);
1391 * `handle' can be NULL if create is zero
1393 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1394 ext4_lblk_t block, int create, int *errp)
1396 struct ext4_map_blocks map;
1397 struct buffer_head *bh;
1400 J_ASSERT(handle != NULL || create == 0);
1404 err = ext4_map_blocks(handle, inode, &map,
1405 create ? EXT4_GET_BLOCKS_CREATE : 0);
1413 bh = sb_getblk(inode->i_sb, map.m_pblk);
1418 if (map.m_flags & EXT4_MAP_NEW) {
1419 J_ASSERT(create != 0);
1420 J_ASSERT(handle != NULL);
1423 * Now that we do not always journal data, we should
1424 * keep in mind whether this should always journal the
1425 * new buffer as metadata. For now, regular file
1426 * writes use ext4_get_block instead, so it's not a
1430 BUFFER_TRACE(bh, "call get_create_access");
1431 fatal = ext4_journal_get_create_access(handle, bh);
1432 if (!fatal && !buffer_uptodate(bh)) {
1433 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1434 set_buffer_uptodate(bh);
1437 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1438 err = ext4_handle_dirty_metadata(handle, inode, bh);
1442 BUFFER_TRACE(bh, "not a new buffer");
1452 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1453 ext4_lblk_t block, int create, int *err)
1455 struct buffer_head *bh;
1457 bh = ext4_getblk(handle, inode, block, create, err);
1460 if (buffer_uptodate(bh))
1462 ll_rw_block(READ_META, 1, &bh);
1464 if (buffer_uptodate(bh))
1471 static int walk_page_buffers(handle_t *handle,
1472 struct buffer_head *head,
1476 int (*fn)(handle_t *handle,
1477 struct buffer_head *bh))
1479 struct buffer_head *bh;
1480 unsigned block_start, block_end;
1481 unsigned blocksize = head->b_size;
1483 struct buffer_head *next;
1485 for (bh = head, block_start = 0;
1486 ret == 0 && (bh != head || !block_start);
1487 block_start = block_end, bh = next) {
1488 next = bh->b_this_page;
1489 block_end = block_start + blocksize;
1490 if (block_end <= from || block_start >= to) {
1491 if (partial && !buffer_uptodate(bh))
1495 err = (*fn)(handle, bh);
1503 * To preserve ordering, it is essential that the hole instantiation and
1504 * the data write be encapsulated in a single transaction. We cannot
1505 * close off a transaction and start a new one between the ext4_get_block()
1506 * and the commit_write(). So doing the jbd2_journal_start at the start of
1507 * prepare_write() is the right place.
1509 * Also, this function can nest inside ext4_writepage() ->
1510 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1511 * has generated enough buffer credits to do the whole page. So we won't
1512 * block on the journal in that case, which is good, because the caller may
1515 * By accident, ext4 can be reentered when a transaction is open via
1516 * quota file writes. If we were to commit the transaction while thus
1517 * reentered, there can be a deadlock - we would be holding a quota
1518 * lock, and the commit would never complete if another thread had a
1519 * transaction open and was blocking on the quota lock - a ranking
1522 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1523 * will _not_ run commit under these circumstances because handle->h_ref
1524 * is elevated. We'll still have enough credits for the tiny quotafile
1527 static int do_journal_get_write_access(handle_t *handle,
1528 struct buffer_head *bh)
1530 int dirty = buffer_dirty(bh);
1533 if (!buffer_mapped(bh) || buffer_freed(bh))
1536 * __block_prepare_write() could have dirtied some buffers. Clean
1537 * the dirty bit as jbd2_journal_get_write_access() could complain
1538 * otherwise about fs integrity issues. Setting of the dirty bit
1539 * by __block_prepare_write() isn't a real problem here as we clear
1540 * the bit before releasing a page lock and thus writeback cannot
1541 * ever write the buffer.
1544 clear_buffer_dirty(bh);
1545 ret = ext4_journal_get_write_access(handle, bh);
1547 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1552 * Truncate blocks that were not used by write. We have to truncate the
1553 * pagecache as well so that corresponding buffers get properly unmapped.
1555 static void ext4_truncate_failed_write(struct inode *inode)
1557 truncate_inode_pages(inode->i_mapping, inode->i_size);
1558 ext4_truncate(inode);
1561 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1562 struct buffer_head *bh_result, int create);
1563 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1564 loff_t pos, unsigned len, unsigned flags,
1565 struct page **pagep, void **fsdata)
1567 struct inode *inode = mapping->host;
1568 int ret, needed_blocks;
1575 trace_ext4_write_begin(inode, pos, len, flags);
1577 * Reserve one block more for addition to orphan list in case
1578 * we allocate blocks but write fails for some reason
1580 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1581 index = pos >> PAGE_CACHE_SHIFT;
1582 from = pos & (PAGE_CACHE_SIZE - 1);
1586 handle = ext4_journal_start(inode, needed_blocks);
1587 if (IS_ERR(handle)) {
1588 ret = PTR_ERR(handle);
1592 /* We cannot recurse into the filesystem as the transaction is already
1594 flags |= AOP_FLAG_NOFS;
1596 page = grab_cache_page_write_begin(mapping, index, flags);
1598 ext4_journal_stop(handle);
1604 if (ext4_should_dioread_nolock(inode))
1605 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1606 fsdata, ext4_get_block_write);
1608 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1609 fsdata, ext4_get_block);
1611 if (!ret && ext4_should_journal_data(inode)) {
1612 ret = walk_page_buffers(handle, page_buffers(page),
1613 from, to, NULL, do_journal_get_write_access);
1618 page_cache_release(page);
1620 * block_write_begin may have instantiated a few blocks
1621 * outside i_size. Trim these off again. Don't need
1622 * i_size_read because we hold i_mutex.
1624 * Add inode to orphan list in case we crash before
1627 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1628 ext4_orphan_add(handle, inode);
1630 ext4_journal_stop(handle);
1631 if (pos + len > inode->i_size) {
1632 ext4_truncate_failed_write(inode);
1634 * If truncate failed early the inode might
1635 * still be on the orphan list; we need to
1636 * make sure the inode is removed from the
1637 * orphan list in that case.
1640 ext4_orphan_del(NULL, inode);
1644 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1650 /* For write_end() in data=journal mode */
1651 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1653 if (!buffer_mapped(bh) || buffer_freed(bh))
1655 set_buffer_uptodate(bh);
1656 return ext4_handle_dirty_metadata(handle, NULL, bh);
1659 static int ext4_generic_write_end(struct file *file,
1660 struct address_space *mapping,
1661 loff_t pos, unsigned len, unsigned copied,
1662 struct page *page, void *fsdata)
1664 int i_size_changed = 0;
1665 struct inode *inode = mapping->host;
1666 handle_t *handle = ext4_journal_current_handle();
1668 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1671 * No need to use i_size_read() here, the i_size
1672 * cannot change under us because we hold i_mutex.
1674 * But it's important to update i_size while still holding page lock:
1675 * page writeout could otherwise come in and zero beyond i_size.
1677 if (pos + copied > inode->i_size) {
1678 i_size_write(inode, pos + copied);
1682 if (pos + copied > EXT4_I(inode)->i_disksize) {
1683 /* We need to mark inode dirty even if
1684 * new_i_size is less that inode->i_size
1685 * bu greater than i_disksize.(hint delalloc)
1687 ext4_update_i_disksize(inode, (pos + copied));
1691 page_cache_release(page);
1694 * Don't mark the inode dirty under page lock. First, it unnecessarily
1695 * makes the holding time of page lock longer. Second, it forces lock
1696 * ordering of page lock and transaction start for journaling
1700 ext4_mark_inode_dirty(handle, inode);
1706 * We need to pick up the new inode size which generic_commit_write gave us
1707 * `file' can be NULL - eg, when called from page_symlink().
1709 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1710 * buffers are managed internally.
1712 static int ext4_ordered_write_end(struct file *file,
1713 struct address_space *mapping,
1714 loff_t pos, unsigned len, unsigned copied,
1715 struct page *page, void *fsdata)
1717 handle_t *handle = ext4_journal_current_handle();
1718 struct inode *inode = mapping->host;
1721 trace_ext4_ordered_write_end(inode, pos, len, copied);
1722 ret = ext4_jbd2_file_inode(handle, inode);
1725 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1728 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1729 /* if we have allocated more blocks and copied
1730 * less. We will have blocks allocated outside
1731 * inode->i_size. So truncate them
1733 ext4_orphan_add(handle, inode);
1737 ret2 = ext4_journal_stop(handle);
1741 if (pos + len > inode->i_size) {
1742 ext4_truncate_failed_write(inode);
1744 * If truncate failed early the inode might still be
1745 * on the orphan list; we need to make sure the inode
1746 * is removed from the orphan list in that case.
1749 ext4_orphan_del(NULL, inode);
1753 return ret ? ret : copied;
1756 static int ext4_writeback_write_end(struct file *file,
1757 struct address_space *mapping,
1758 loff_t pos, unsigned len, unsigned copied,
1759 struct page *page, void *fsdata)
1761 handle_t *handle = ext4_journal_current_handle();
1762 struct inode *inode = mapping->host;
1765 trace_ext4_writeback_write_end(inode, pos, len, copied);
1766 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1769 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1770 /* if we have allocated more blocks and copied
1771 * less. We will have blocks allocated outside
1772 * inode->i_size. So truncate them
1774 ext4_orphan_add(handle, inode);
1779 ret2 = ext4_journal_stop(handle);
1783 if (pos + len > inode->i_size) {
1784 ext4_truncate_failed_write(inode);
1786 * If truncate failed early the inode might still be
1787 * on the orphan list; we need to make sure the inode
1788 * is removed from the orphan list in that case.
1791 ext4_orphan_del(NULL, inode);
1794 return ret ? ret : copied;
1797 static int ext4_journalled_write_end(struct file *file,
1798 struct address_space *mapping,
1799 loff_t pos, unsigned len, unsigned copied,
1800 struct page *page, void *fsdata)
1802 handle_t *handle = ext4_journal_current_handle();
1803 struct inode *inode = mapping->host;
1809 trace_ext4_journalled_write_end(inode, pos, len, copied);
1810 from = pos & (PAGE_CACHE_SIZE - 1);
1814 if (!PageUptodate(page))
1816 page_zero_new_buffers(page, from+copied, to);
1819 ret = walk_page_buffers(handle, page_buffers(page), from,
1820 to, &partial, write_end_fn);
1822 SetPageUptodate(page);
1823 new_i_size = pos + copied;
1824 if (new_i_size > inode->i_size)
1825 i_size_write(inode, pos+copied);
1826 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1827 if (new_i_size > EXT4_I(inode)->i_disksize) {
1828 ext4_update_i_disksize(inode, new_i_size);
1829 ret2 = ext4_mark_inode_dirty(handle, inode);
1835 page_cache_release(page);
1836 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1837 /* if we have allocated more blocks and copied
1838 * less. We will have blocks allocated outside
1839 * inode->i_size. So truncate them
1841 ext4_orphan_add(handle, inode);
1843 ret2 = ext4_journal_stop(handle);
1846 if (pos + len > inode->i_size) {
1847 ext4_truncate_failed_write(inode);
1849 * If truncate failed early the inode might still be
1850 * on the orphan list; we need to make sure the inode
1851 * is removed from the orphan list in that case.
1854 ext4_orphan_del(NULL, inode);
1857 return ret ? ret : copied;
1861 * Reserve a single block located at lblock
1863 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1866 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1867 struct ext4_inode_info *ei = EXT4_I(inode);
1868 unsigned long md_needed;
1872 * recalculate the amount of metadata blocks to reserve
1873 * in order to allocate nrblocks
1874 * worse case is one extent per block
1877 spin_lock(&ei->i_block_reservation_lock);
1878 md_needed = ext4_calc_metadata_amount(inode, lblock);
1879 trace_ext4_da_reserve_space(inode, md_needed);
1880 spin_unlock(&ei->i_block_reservation_lock);
1883 * We will charge metadata quota at writeout time; this saves
1884 * us from metadata over-estimation, though we may go over by
1885 * a small amount in the end. Here we just reserve for data.
1887 ret = dquot_reserve_block(inode, 1);
1891 * We do still charge estimated metadata to the sb though;
1892 * we cannot afford to run out of free blocks.
1894 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1895 dquot_release_reservation_block(inode, 1);
1896 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1902 spin_lock(&ei->i_block_reservation_lock);
1903 ei->i_reserved_data_blocks++;
1904 ei->i_reserved_meta_blocks += md_needed;
1905 spin_unlock(&ei->i_block_reservation_lock);
1907 return 0; /* success */
1910 static void ext4_da_release_space(struct inode *inode, int to_free)
1912 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1913 struct ext4_inode_info *ei = EXT4_I(inode);
1916 return; /* Nothing to release, exit */
1918 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1920 trace_ext4_da_release_space(inode, to_free);
1921 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1923 * if there aren't enough reserved blocks, then the
1924 * counter is messed up somewhere. Since this
1925 * function is called from invalidate page, it's
1926 * harmless to return without any action.
1928 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1929 "ino %lu, to_free %d with only %d reserved "
1930 "data blocks\n", inode->i_ino, to_free,
1931 ei->i_reserved_data_blocks);
1933 to_free = ei->i_reserved_data_blocks;
1935 ei->i_reserved_data_blocks -= to_free;
1937 if (ei->i_reserved_data_blocks == 0) {
1939 * We can release all of the reserved metadata blocks
1940 * only when we have written all of the delayed
1941 * allocation blocks.
1943 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1944 ei->i_reserved_meta_blocks);
1945 ei->i_reserved_meta_blocks = 0;
1946 ei->i_da_metadata_calc_len = 0;
1949 /* update fs dirty data blocks counter */
1950 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1952 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1954 dquot_release_reservation_block(inode, to_free);
1957 static void ext4_da_page_release_reservation(struct page *page,
1958 unsigned long offset)
1961 struct buffer_head *head, *bh;
1962 unsigned int curr_off = 0;
1964 head = page_buffers(page);
1967 unsigned int next_off = curr_off + bh->b_size;
1969 if ((offset <= curr_off) && (buffer_delay(bh))) {
1971 clear_buffer_delay(bh);
1973 curr_off = next_off;
1974 } while ((bh = bh->b_this_page) != head);
1975 ext4_da_release_space(page->mapping->host, to_release);
1979 * Delayed allocation stuff
1983 * mpage_da_submit_io - walks through extent of pages and try to write
1984 * them with writepage() call back
1986 * @mpd->inode: inode
1987 * @mpd->first_page: first page of the extent
1988 * @mpd->next_page: page after the last page of the extent
1990 * By the time mpage_da_submit_io() is called we expect all blocks
1991 * to be allocated. this may be wrong if allocation failed.
1993 * As pages are already locked by write_cache_pages(), we can't use it
1995 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1998 struct pagevec pvec;
1999 unsigned long index, end;
2000 int ret = 0, err, nr_pages, i;
2001 struct inode *inode = mpd->inode;
2002 struct address_space *mapping = inode->i_mapping;
2004 BUG_ON(mpd->next_page <= mpd->first_page);
2006 * We need to start from the first_page to the next_page - 1
2007 * to make sure we also write the mapped dirty buffer_heads.
2008 * If we look at mpd->b_blocknr we would only be looking
2009 * at the currently mapped buffer_heads.
2011 index = mpd->first_page;
2012 end = mpd->next_page - 1;
2014 pagevec_init(&pvec, 0);
2015 while (index <= end) {
2016 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2019 for (i = 0; i < nr_pages; i++) {
2020 struct page *page = pvec.pages[i];
2022 index = page->index;
2027 BUG_ON(!PageLocked(page));
2028 BUG_ON(PageWriteback(page));
2030 pages_skipped = mpd->wbc->pages_skipped;
2031 err = mapping->a_ops->writepage(page, mpd->wbc);
2032 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2034 * have successfully written the page
2035 * without skipping the same
2037 mpd->pages_written++;
2039 * In error case, we have to continue because
2040 * remaining pages are still locked
2041 * XXX: unlock and re-dirty them?
2046 pagevec_release(&pvec);
2052 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2054 * the function goes through all passed space and put actual disk
2055 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2057 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd,
2058 struct ext4_map_blocks *map)
2060 struct inode *inode = mpd->inode;
2061 struct address_space *mapping = inode->i_mapping;
2062 int blocks = map->m_len;
2063 sector_t pblock = map->m_pblk, cur_logical;
2064 struct buffer_head *head, *bh;
2066 struct pagevec pvec;
2069 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2070 end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2071 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2073 pagevec_init(&pvec, 0);
2075 while (index <= end) {
2076 /* XXX: optimize tail */
2077 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2080 for (i = 0; i < nr_pages; i++) {
2081 struct page *page = pvec.pages[i];
2083 index = page->index;
2088 BUG_ON(!PageLocked(page));
2089 BUG_ON(PageWriteback(page));
2090 BUG_ON(!page_has_buffers(page));
2092 bh = page_buffers(page);
2095 /* skip blocks out of the range */
2097 if (cur_logical >= map->m_lblk)
2100 } while ((bh = bh->b_this_page) != head);
2103 if (cur_logical >= map->m_lblk + blocks)
2106 if (buffer_delay(bh) || buffer_unwritten(bh)) {
2108 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2110 if (buffer_delay(bh)) {
2111 clear_buffer_delay(bh);
2112 bh->b_blocknr = pblock;
2115 * unwritten already should have
2116 * blocknr assigned. Verify that
2118 clear_buffer_unwritten(bh);
2119 BUG_ON(bh->b_blocknr != pblock);
2122 } else if (buffer_mapped(bh))
2123 BUG_ON(bh->b_blocknr != pblock);
2125 if (map->m_flags & EXT4_MAP_UNINIT)
2126 set_buffer_uninit(bh);
2129 } while ((bh = bh->b_this_page) != head);
2131 pagevec_release(&pvec);
2136 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2137 sector_t logical, long blk_cnt)
2141 struct pagevec pvec;
2142 struct inode *inode = mpd->inode;
2143 struct address_space *mapping = inode->i_mapping;
2145 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2146 end = (logical + blk_cnt - 1) >>
2147 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2148 while (index <= end) {
2149 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2152 for (i = 0; i < nr_pages; i++) {
2153 struct page *page = pvec.pages[i];
2154 if (page->index > end)
2156 BUG_ON(!PageLocked(page));
2157 BUG_ON(PageWriteback(page));
2158 block_invalidatepage(page, 0);
2159 ClearPageUptodate(page);
2162 index = pvec.pages[nr_pages - 1]->index + 1;
2163 pagevec_release(&pvec);
2168 static void ext4_print_free_blocks(struct inode *inode)
2170 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2171 printk(KERN_CRIT "Total free blocks count %lld\n",
2172 ext4_count_free_blocks(inode->i_sb));
2173 printk(KERN_CRIT "Free/Dirty block details\n");
2174 printk(KERN_CRIT "free_blocks=%lld\n",
2175 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2176 printk(KERN_CRIT "dirty_blocks=%lld\n",
2177 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2178 printk(KERN_CRIT "Block reservation details\n");
2179 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2180 EXT4_I(inode)->i_reserved_data_blocks);
2181 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2182 EXT4_I(inode)->i_reserved_meta_blocks);
2187 * mpage_da_map_blocks - go through given space
2189 * @mpd - bh describing space
2191 * The function skips space we know is already mapped to disk blocks.
2194 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2196 int err, blks, get_blocks_flags;
2197 struct ext4_map_blocks map;
2198 sector_t next = mpd->b_blocknr;
2199 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2200 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2201 handle_t *handle = NULL;
2204 * We consider only non-mapped and non-allocated blocks
2206 if ((mpd->b_state & (1 << BH_Mapped)) &&
2207 !(mpd->b_state & (1 << BH_Delay)) &&
2208 !(mpd->b_state & (1 << BH_Unwritten)))
2212 * If we didn't accumulate anything to write simply return
2217 handle = ext4_journal_current_handle();
2221 * Call ext4_map_blocks() to allocate any delayed allocation
2222 * blocks, or to convert an uninitialized extent to be
2223 * initialized (in the case where we have written into
2224 * one or more preallocated blocks).
2226 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2227 * indicate that we are on the delayed allocation path. This
2228 * affects functions in many different parts of the allocation
2229 * call path. This flag exists primarily because we don't
2230 * want to change *many* call functions, so ext4_map_blocks()
2231 * will set the magic i_delalloc_reserved_flag once the
2232 * inode's allocation semaphore is taken.
2234 * If the blocks in questions were delalloc blocks, set
2235 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2236 * variables are updated after the blocks have been allocated.
2239 map.m_len = max_blocks;
2240 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2241 if (ext4_should_dioread_nolock(mpd->inode))
2242 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2243 if (mpd->b_state & (1 << BH_Delay))
2244 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2246 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2248 struct super_block *sb = mpd->inode->i_sb;
2252 * If get block returns with error we simply
2253 * return. Later writepage will redirty the page and
2254 * writepages will find the dirty page again
2259 if (err == -ENOSPC &&
2260 ext4_count_free_blocks(sb)) {
2266 * get block failure will cause us to loop in
2267 * writepages, because a_ops->writepage won't be able
2268 * to make progress. The page will be redirtied by
2269 * writepage and writepages will again try to write
2272 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2273 ext4_msg(sb, KERN_CRIT,
2274 "delayed block allocation failed for inode %lu "
2275 "at logical offset %llu with max blocks %zd "
2276 "with error %d", mpd->inode->i_ino,
2277 (unsigned long long) next,
2278 mpd->b_size >> mpd->inode->i_blkbits, err);
2279 ext4_msg(sb, KERN_CRIT,
2280 "This should not happen!! Data will be lost\n");
2282 ext4_print_free_blocks(mpd->inode);
2284 /* invalidate all the pages */
2285 ext4_da_block_invalidatepages(mpd, next,
2286 mpd->b_size >> mpd->inode->i_blkbits);
2291 if (map.m_flags & EXT4_MAP_NEW) {
2292 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2295 for (i = 0; i < map.m_len; i++)
2296 unmap_underlying_metadata(bdev, map.m_pblk + i);
2300 * If blocks are delayed marked, we need to
2301 * put actual blocknr and drop delayed bit
2303 if ((mpd->b_state & (1 << BH_Delay)) ||
2304 (mpd->b_state & (1 << BH_Unwritten)))
2305 mpage_put_bnr_to_bhs(mpd, &map);
2307 if (ext4_should_order_data(mpd->inode)) {
2308 err = ext4_jbd2_file_inode(handle, mpd->inode);
2314 * Update on-disk size along with block allocation.
2316 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2317 if (disksize > i_size_read(mpd->inode))
2318 disksize = i_size_read(mpd->inode);
2319 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2320 ext4_update_i_disksize(mpd->inode, disksize);
2321 return ext4_mark_inode_dirty(handle, mpd->inode);
2327 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2328 (1 << BH_Delay) | (1 << BH_Unwritten))
2331 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2333 * @mpd->lbh - extent of blocks
2334 * @logical - logical number of the block in the file
2335 * @bh - bh of the block (used to access block's state)
2337 * the function is used to collect contig. blocks in same state
2339 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2340 sector_t logical, size_t b_size,
2341 unsigned long b_state)
2344 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2347 * XXX Don't go larger than mballoc is willing to allocate
2348 * This is a stopgap solution. We eventually need to fold
2349 * mpage_da_submit_io() into this function and then call
2350 * ext4_map_blocks() multiple times in a loop
2352 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2355 /* check if thereserved journal credits might overflow */
2356 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2357 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2359 * With non-extent format we are limited by the journal
2360 * credit available. Total credit needed to insert
2361 * nrblocks contiguous blocks is dependent on the
2362 * nrblocks. So limit nrblocks.
2365 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2366 EXT4_MAX_TRANS_DATA) {
2368 * Adding the new buffer_head would make it cross the
2369 * allowed limit for which we have journal credit
2370 * reserved. So limit the new bh->b_size
2372 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2373 mpd->inode->i_blkbits;
2374 /* we will do mpage_da_submit_io in the next loop */
2378 * First block in the extent
2380 if (mpd->b_size == 0) {
2381 mpd->b_blocknr = logical;
2382 mpd->b_size = b_size;
2383 mpd->b_state = b_state & BH_FLAGS;
2387 next = mpd->b_blocknr + nrblocks;
2389 * Can we merge the block to our big extent?
2391 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2392 mpd->b_size += b_size;
2398 * We couldn't merge the block to our extent, so we
2399 * need to flush current extent and start new one
2401 if (mpage_da_map_blocks(mpd) == 0)
2402 mpage_da_submit_io(mpd);
2407 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2409 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2413 * __mpage_da_writepage - finds extent of pages and blocks
2415 * @page: page to consider
2416 * @wbc: not used, we just follow rules
2419 * The function finds extents of pages and scan them for all blocks.
2421 static int __mpage_da_writepage(struct page *page,
2422 struct writeback_control *wbc, void *data)
2424 struct mpage_da_data *mpd = data;
2425 struct inode *inode = mpd->inode;
2426 struct buffer_head *bh, *head;
2430 * Can we merge this page to current extent?
2432 if (mpd->next_page != page->index) {
2434 * Nope, we can't. So, we map non-allocated blocks
2435 * and start IO on them using writepage()
2437 if (mpd->next_page != mpd->first_page) {
2438 if (mpage_da_map_blocks(mpd) == 0)
2439 mpage_da_submit_io(mpd);
2441 * skip rest of the page in the page_vec
2444 redirty_page_for_writepage(wbc, page);
2446 return MPAGE_DA_EXTENT_TAIL;
2450 * Start next extent of pages ...
2452 mpd->first_page = page->index;
2462 mpd->next_page = page->index + 1;
2463 logical = (sector_t) page->index <<
2464 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2466 if (!page_has_buffers(page)) {
2467 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2468 (1 << BH_Dirty) | (1 << BH_Uptodate));
2470 return MPAGE_DA_EXTENT_TAIL;
2473 * Page with regular buffer heads, just add all dirty ones
2475 head = page_buffers(page);
2478 BUG_ON(buffer_locked(bh));
2480 * We need to try to allocate
2481 * unmapped blocks in the same page.
2482 * Otherwise we won't make progress
2483 * with the page in ext4_writepage
2485 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2486 mpage_add_bh_to_extent(mpd, logical,
2490 return MPAGE_DA_EXTENT_TAIL;
2491 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2493 * mapped dirty buffer. We need to update
2494 * the b_state because we look at
2495 * b_state in mpage_da_map_blocks. We don't
2496 * update b_size because if we find an
2497 * unmapped buffer_head later we need to
2498 * use the b_state flag of that buffer_head.
2500 if (mpd->b_size == 0)
2501 mpd->b_state = bh->b_state & BH_FLAGS;
2504 } while ((bh = bh->b_this_page) != head);
2511 * This is a special get_blocks_t callback which is used by
2512 * ext4_da_write_begin(). It will either return mapped block or
2513 * reserve space for a single block.
2515 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2516 * We also have b_blocknr = -1 and b_bdev initialized properly
2518 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2519 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2520 * initialized properly.
2522 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2523 struct buffer_head *bh, int create)
2525 struct ext4_map_blocks map;
2527 sector_t invalid_block = ~((sector_t) 0xffff);
2529 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2532 BUG_ON(create == 0);
2533 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2535 map.m_lblk = iblock;
2539 * first, we need to know whether the block is allocated already
2540 * preallocated blocks are unmapped but should treated
2541 * the same as allocated blocks.
2543 ret = ext4_map_blocks(NULL, inode, &map, 0);
2547 if (buffer_delay(bh))
2548 return 0; /* Not sure this could or should happen */
2550 * XXX: __block_prepare_write() unmaps passed block,
2553 ret = ext4_da_reserve_space(inode, iblock);
2555 /* not enough space to reserve */
2558 map_bh(bh, inode->i_sb, invalid_block);
2560 set_buffer_delay(bh);
2564 map_bh(bh, inode->i_sb, map.m_pblk);
2565 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2567 if (buffer_unwritten(bh)) {
2568 /* A delayed write to unwritten bh should be marked
2569 * new and mapped. Mapped ensures that we don't do
2570 * get_block multiple times when we write to the same
2571 * offset and new ensures that we do proper zero out
2572 * for partial write.
2575 set_buffer_mapped(bh);
2581 * This function is used as a standard get_block_t calback function
2582 * when there is no desire to allocate any blocks. It is used as a
2583 * callback function for block_prepare_write() and block_write_full_page().
2584 * These functions should only try to map a single block at a time.
2586 * Since this function doesn't do block allocations even if the caller
2587 * requests it by passing in create=1, it is critically important that
2588 * any caller checks to make sure that any buffer heads are returned
2589 * by this function are either all already mapped or marked for
2590 * delayed allocation before calling block_write_full_page(). Otherwise,
2591 * b_blocknr could be left unitialized, and the page write functions will
2592 * be taken by surprise.
2594 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2595 struct buffer_head *bh_result, int create)
2597 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2598 return _ext4_get_block(inode, iblock, bh_result, 0);
2601 static int bget_one(handle_t *handle, struct buffer_head *bh)
2607 static int bput_one(handle_t *handle, struct buffer_head *bh)
2613 static int __ext4_journalled_writepage(struct page *page,
2616 struct address_space *mapping = page->mapping;
2617 struct inode *inode = mapping->host;
2618 struct buffer_head *page_bufs;
2619 handle_t *handle = NULL;
2623 page_bufs = page_buffers(page);
2625 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2626 /* As soon as we unlock the page, it can go away, but we have
2627 * references to buffers so we are safe */
2630 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2631 if (IS_ERR(handle)) {
2632 ret = PTR_ERR(handle);
2636 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2637 do_journal_get_write_access);
2639 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2643 err = ext4_journal_stop(handle);
2647 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2648 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2653 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2654 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2657 * Note that we don't need to start a transaction unless we're journaling data
2658 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2659 * need to file the inode to the transaction's list in ordered mode because if
2660 * we are writing back data added by write(), the inode is already there and if
2661 * we are writing back data modified via mmap(), noone guarantees in which
2662 * transaction the data will hit the disk. In case we are journaling data, we
2663 * cannot start transaction directly because transaction start ranks above page
2664 * lock so we have to do some magic.
2666 * This function can get called via...
2667 * - ext4_da_writepages after taking page lock (have journal handle)
2668 * - journal_submit_inode_data_buffers (no journal handle)
2669 * - shrink_page_list via pdflush (no journal handle)
2670 * - grab_page_cache when doing write_begin (have journal handle)
2672 * We don't do any block allocation in this function. If we have page with
2673 * multiple blocks we need to write those buffer_heads that are mapped. This
2674 * is important for mmaped based write. So if we do with blocksize 1K
2675 * truncate(f, 1024);
2676 * a = mmap(f, 0, 4096);
2678 * truncate(f, 4096);
2679 * we have in the page first buffer_head mapped via page_mkwrite call back
2680 * but other bufer_heads would be unmapped but dirty(dirty done via the
2681 * do_wp_page). So writepage should write the first block. If we modify
2682 * the mmap area beyond 1024 we will again get a page_fault and the
2683 * page_mkwrite callback will do the block allocation and mark the
2684 * buffer_heads mapped.
2686 * We redirty the page if we have any buffer_heads that is either delay or
2687 * unwritten in the page.
2689 * We can get recursively called as show below.
2691 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2694 * But since we don't do any block allocation we should not deadlock.
2695 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2697 static int ext4_writepage(struct page *page,
2698 struct writeback_control *wbc)
2703 struct buffer_head *page_bufs = NULL;
2704 struct inode *inode = page->mapping->host;
2706 trace_ext4_writepage(inode, page);
2707 size = i_size_read(inode);
2708 if (page->index == size >> PAGE_CACHE_SHIFT)
2709 len = size & ~PAGE_CACHE_MASK;
2711 len = PAGE_CACHE_SIZE;
2713 if (page_has_buffers(page)) {
2714 page_bufs = page_buffers(page);
2715 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2716 ext4_bh_delay_or_unwritten)) {
2718 * We don't want to do block allocation
2719 * So redirty the page and return
2720 * We may reach here when we do a journal commit
2721 * via journal_submit_inode_data_buffers.
2722 * If we don't have mapping block we just ignore
2723 * them. We can also reach here via shrink_page_list
2725 redirty_page_for_writepage(wbc, page);
2731 * The test for page_has_buffers() is subtle:
2732 * We know the page is dirty but it lost buffers. That means
2733 * that at some moment in time after write_begin()/write_end()
2734 * has been called all buffers have been clean and thus they
2735 * must have been written at least once. So they are all
2736 * mapped and we can happily proceed with mapping them
2737 * and writing the page.
2739 * Try to initialize the buffer_heads and check whether
2740 * all are mapped and non delay. We don't want to
2741 * do block allocation here.
2743 ret = block_prepare_write(page, 0, len,
2744 noalloc_get_block_write);
2746 page_bufs = page_buffers(page);
2747 /* check whether all are mapped and non delay */
2748 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2749 ext4_bh_delay_or_unwritten)) {
2750 redirty_page_for_writepage(wbc, page);
2756 * We can't do block allocation here
2757 * so just redity the page and unlock
2760 redirty_page_for_writepage(wbc, page);
2764 /* now mark the buffer_heads as dirty and uptodate */
2765 block_commit_write(page, 0, len);
2768 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2770 * It's mmapped pagecache. Add buffers and journal it. There
2771 * doesn't seem much point in redirtying the page here.
2773 ClearPageChecked(page);
2774 return __ext4_journalled_writepage(page, len);
2777 if (page_bufs && buffer_uninit(page_bufs)) {
2778 ext4_set_bh_endio(page_bufs, inode);
2779 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2780 wbc, ext4_end_io_buffer_write);
2782 ret = block_write_full_page(page, noalloc_get_block_write,
2789 * This is called via ext4_da_writepages() to
2790 * calulate the total number of credits to reserve to fit
2791 * a single extent allocation into a single transaction,
2792 * ext4_da_writpeages() will loop calling this before
2793 * the block allocation.
2796 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2798 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2801 * With non-extent format the journal credit needed to
2802 * insert nrblocks contiguous block is dependent on
2803 * number of contiguous block. So we will limit
2804 * number of contiguous block to a sane value
2806 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2807 (max_blocks > EXT4_MAX_TRANS_DATA))
2808 max_blocks = EXT4_MAX_TRANS_DATA;
2810 return ext4_chunk_trans_blocks(inode, max_blocks);
2814 * write_cache_pages_da - walk the list of dirty pages of the given
2815 * address space and call the callback function (which usually writes
2818 * This is a forked version of write_cache_pages(). Differences:
2819 * Range cyclic is ignored.
2820 * no_nrwrite_index_update is always presumed true
2822 static int write_cache_pages_da(struct address_space *mapping,
2823 struct writeback_control *wbc,
2824 struct mpage_da_data *mpd)
2828 struct pagevec pvec;
2831 pgoff_t end; /* Inclusive */
2832 long nr_to_write = wbc->nr_to_write;
2834 pagevec_init(&pvec, 0);
2835 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2836 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2838 while (!done && (index <= end)) {
2841 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2842 PAGECACHE_TAG_DIRTY,
2843 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2847 for (i = 0; i < nr_pages; i++) {
2848 struct page *page = pvec.pages[i];
2851 * At this point, the page may be truncated or
2852 * invalidated (changing page->mapping to NULL), or
2853 * even swizzled back from swapper_space to tmpfs file
2854 * mapping. However, page->index will not change
2855 * because we have a reference on the page.
2857 if (page->index > end) {
2865 * Page truncated or invalidated. We can freely skip it
2866 * then, even for data integrity operations: the page
2867 * has disappeared concurrently, so there could be no
2868 * real expectation of this data interity operation
2869 * even if there is now a new, dirty page at the same
2870 * pagecache address.
2872 if (unlikely(page->mapping != mapping)) {
2878 if (!PageDirty(page)) {
2879 /* someone wrote it for us */
2880 goto continue_unlock;
2883 if (PageWriteback(page)) {
2884 if (wbc->sync_mode != WB_SYNC_NONE)
2885 wait_on_page_writeback(page);
2887 goto continue_unlock;
2890 BUG_ON(PageWriteback(page));
2891 if (!clear_page_dirty_for_io(page))
2892 goto continue_unlock;
2894 ret = __mpage_da_writepage(page, wbc, mpd);
2895 if (unlikely(ret)) {
2896 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2905 if (nr_to_write > 0) {
2907 if (nr_to_write == 0 &&
2908 wbc->sync_mode == WB_SYNC_NONE) {
2910 * We stop writing back only if we are
2911 * not doing integrity sync. In case of
2912 * integrity sync we have to keep going
2913 * because someone may be concurrently
2914 * dirtying pages, and we might have
2915 * synced a lot of newly appeared dirty
2916 * pages, but have not synced all of the
2924 pagevec_release(&pvec);
2931 static int ext4_da_writepages(struct address_space *mapping,
2932 struct writeback_control *wbc)
2935 int range_whole = 0;
2936 handle_t *handle = NULL;
2937 struct mpage_da_data mpd;
2938 struct inode *inode = mapping->host;
2939 int pages_written = 0;
2941 unsigned int max_pages;
2942 int range_cyclic, cycled = 1, io_done = 0;
2943 int needed_blocks, ret = 0;
2944 long desired_nr_to_write, nr_to_writebump = 0;
2945 loff_t range_start = wbc->range_start;
2946 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2948 trace_ext4_da_writepages(inode, wbc);
2951 * No pages to write? This is mainly a kludge to avoid starting
2952 * a transaction for special inodes like journal inode on last iput()
2953 * because that could violate lock ordering on umount
2955 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2959 * If the filesystem has aborted, it is read-only, so return
2960 * right away instead of dumping stack traces later on that
2961 * will obscure the real source of the problem. We test
2962 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2963 * the latter could be true if the filesystem is mounted
2964 * read-only, and in that case, ext4_da_writepages should
2965 * *never* be called, so if that ever happens, we would want
2968 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2971 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2974 range_cyclic = wbc->range_cyclic;
2975 if (wbc->range_cyclic) {
2976 index = mapping->writeback_index;
2979 wbc->range_start = index << PAGE_CACHE_SHIFT;
2980 wbc->range_end = LLONG_MAX;
2981 wbc->range_cyclic = 0;
2983 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2986 * This works around two forms of stupidity. The first is in
2987 * the writeback code, which caps the maximum number of pages
2988 * written to be 1024 pages. This is wrong on multiple
2989 * levels; different architectues have a different page size,
2990 * which changes the maximum amount of data which gets
2991 * written. Secondly, 4 megabytes is way too small. XFS
2992 * forces this value to be 16 megabytes by multiplying
2993 * nr_to_write parameter by four, and then relies on its
2994 * allocator to allocate larger extents to make them
2995 * contiguous. Unfortunately this brings us to the second
2996 * stupidity, which is that ext4's mballoc code only allocates
2997 * at most 2048 blocks. So we force contiguous writes up to
2998 * the number of dirty blocks in the inode, or
2999 * sbi->max_writeback_mb_bump whichever is smaller.
3001 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
3002 if (!range_cyclic && range_whole)
3003 desired_nr_to_write = wbc->nr_to_write * 8;
3005 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
3007 if (desired_nr_to_write > max_pages)
3008 desired_nr_to_write = max_pages;
3010 if (wbc->nr_to_write < desired_nr_to_write) {
3011 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
3012 wbc->nr_to_write = desired_nr_to_write;
3016 mpd.inode = mapping->host;
3018 pages_skipped = wbc->pages_skipped;
3021 while (!ret && wbc->nr_to_write > 0) {
3024 * we insert one extent at a time. So we need
3025 * credit needed for single extent allocation.
3026 * journalled mode is currently not supported
3029 BUG_ON(ext4_should_journal_data(inode));
3030 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3032 /* start a new transaction*/
3033 handle = ext4_journal_start(inode, needed_blocks);
3034 if (IS_ERR(handle)) {
3035 ret = PTR_ERR(handle);
3036 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3037 "%ld pages, ino %lu; err %d", __func__,
3038 wbc->nr_to_write, inode->i_ino, ret);
3039 goto out_writepages;
3043 * Now call __mpage_da_writepage to find the next
3044 * contiguous region of logical blocks that need
3045 * blocks to be allocated by ext4. We don't actually
3046 * submit the blocks for I/O here, even though
3047 * write_cache_pages thinks it will, and will set the
3048 * pages as clean for write before calling
3049 * __mpage_da_writepage().
3057 mpd.pages_written = 0;
3059 ret = write_cache_pages_da(mapping, wbc, &mpd);
3061 * If we have a contiguous extent of pages and we
3062 * haven't done the I/O yet, map the blocks and submit
3065 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3066 if (mpage_da_map_blocks(&mpd) == 0)
3067 mpage_da_submit_io(&mpd);
3069 ret = MPAGE_DA_EXTENT_TAIL;
3071 trace_ext4_da_write_pages(inode, &mpd);
3072 wbc->nr_to_write -= mpd.pages_written;
3074 ext4_journal_stop(handle);
3076 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3077 /* commit the transaction which would
3078 * free blocks released in the transaction
3081 jbd2_journal_force_commit_nested(sbi->s_journal);
3082 wbc->pages_skipped = pages_skipped;
3084 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3086 * got one extent now try with
3089 pages_written += mpd.pages_written;
3090 wbc->pages_skipped = pages_skipped;
3093 } else if (wbc->nr_to_write)
3095 * There is no more writeout needed
3096 * or we requested for a noblocking writeout
3097 * and we found the device congested
3101 if (!io_done && !cycled) {
3104 wbc->range_start = index << PAGE_CACHE_SHIFT;
3105 wbc->range_end = mapping->writeback_index - 1;
3108 if (pages_skipped != wbc->pages_skipped)
3109 ext4_msg(inode->i_sb, KERN_CRIT,
3110 "This should not happen leaving %s "
3111 "with nr_to_write = %ld ret = %d",
3112 __func__, wbc->nr_to_write, ret);
3115 index += pages_written;
3116 wbc->range_cyclic = range_cyclic;
3117 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3119 * set the writeback_index so that range_cyclic
3120 * mode will write it back later
3122 mapping->writeback_index = index;
3125 wbc->nr_to_write -= nr_to_writebump;
3126 wbc->range_start = range_start;
3127 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3131 #define FALL_BACK_TO_NONDELALLOC 1
3132 static int ext4_nonda_switch(struct super_block *sb)
3134 s64 free_blocks, dirty_blocks;
3135 struct ext4_sb_info *sbi = EXT4_SB(sb);
3138 * switch to non delalloc mode if we are running low
3139 * on free block. The free block accounting via percpu
3140 * counters can get slightly wrong with percpu_counter_batch getting
3141 * accumulated on each CPU without updating global counters
3142 * Delalloc need an accurate free block accounting. So switch
3143 * to non delalloc when we are near to error range.
3145 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3146 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3147 if (2 * free_blocks < 3 * dirty_blocks ||
3148 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3150 * free block count is less than 150% of dirty blocks
3151 * or free blocks is less than watermark
3156 * Even if we don't switch but are nearing capacity,
3157 * start pushing delalloc when 1/2 of free blocks are dirty.
3159 if (free_blocks < 2 * dirty_blocks)
3160 writeback_inodes_sb_if_idle(sb);
3165 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3166 loff_t pos, unsigned len, unsigned flags,
3167 struct page **pagep, void **fsdata)
3169 int ret, retries = 0;
3172 struct inode *inode = mapping->host;
3175 index = pos >> PAGE_CACHE_SHIFT;
3177 if (ext4_nonda_switch(inode->i_sb)) {
3178 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3179 return ext4_write_begin(file, mapping, pos,
3180 len, flags, pagep, fsdata);
3182 *fsdata = (void *)0;
3183 trace_ext4_da_write_begin(inode, pos, len, flags);
3186 * With delayed allocation, we don't log the i_disksize update
3187 * if there is delayed block allocation. But we still need
3188 * to journalling the i_disksize update if writes to the end
3189 * of file which has an already mapped buffer.
3191 handle = ext4_journal_start(inode, 1);
3192 if (IS_ERR(handle)) {
3193 ret = PTR_ERR(handle);
3196 /* We cannot recurse into the filesystem as the transaction is already
3198 flags |= AOP_FLAG_NOFS;
3200 page = grab_cache_page_write_begin(mapping, index, flags);
3202 ext4_journal_stop(handle);
3208 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3209 ext4_da_get_block_prep);
3212 ext4_journal_stop(handle);
3213 page_cache_release(page);
3215 * block_write_begin may have instantiated a few blocks
3216 * outside i_size. Trim these off again. Don't need
3217 * i_size_read because we hold i_mutex.
3219 if (pos + len > inode->i_size)
3220 ext4_truncate_failed_write(inode);
3223 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3230 * Check if we should update i_disksize
3231 * when write to the end of file but not require block allocation
3233 static int ext4_da_should_update_i_disksize(struct page *page,
3234 unsigned long offset)
3236 struct buffer_head *bh;
3237 struct inode *inode = page->mapping->host;
3241 bh = page_buffers(page);
3242 idx = offset >> inode->i_blkbits;
3244 for (i = 0; i < idx; i++)
3245 bh = bh->b_this_page;
3247 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3252 static int ext4_da_write_end(struct file *file,
3253 struct address_space *mapping,
3254 loff_t pos, unsigned len, unsigned copied,
3255 struct page *page, void *fsdata)
3257 struct inode *inode = mapping->host;
3259 handle_t *handle = ext4_journal_current_handle();
3261 unsigned long start, end;
3262 int write_mode = (int)(unsigned long)fsdata;
3264 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3265 if (ext4_should_order_data(inode)) {
3266 return ext4_ordered_write_end(file, mapping, pos,
3267 len, copied, page, fsdata);
3268 } else if (ext4_should_writeback_data(inode)) {
3269 return ext4_writeback_write_end(file, mapping, pos,
3270 len, copied, page, fsdata);
3276 trace_ext4_da_write_end(inode, pos, len, copied);
3277 start = pos & (PAGE_CACHE_SIZE - 1);
3278 end = start + copied - 1;
3281 * generic_write_end() will run mark_inode_dirty() if i_size
3282 * changes. So let's piggyback the i_disksize mark_inode_dirty
3286 new_i_size = pos + copied;
3287 if (new_i_size > EXT4_I(inode)->i_disksize) {
3288 if (ext4_da_should_update_i_disksize(page, end)) {
3289 down_write(&EXT4_I(inode)->i_data_sem);
3290 if (new_i_size > EXT4_I(inode)->i_disksize) {
3292 * Updating i_disksize when extending file
3293 * without needing block allocation
3295 if (ext4_should_order_data(inode))
3296 ret = ext4_jbd2_file_inode(handle,
3299 EXT4_I(inode)->i_disksize = new_i_size;
3301 up_write(&EXT4_I(inode)->i_data_sem);
3302 /* We need to mark inode dirty even if
3303 * new_i_size is less that inode->i_size
3304 * bu greater than i_disksize.(hint delalloc)
3306 ext4_mark_inode_dirty(handle, inode);
3309 ret2 = generic_write_end(file, mapping, pos, len, copied,
3314 ret2 = ext4_journal_stop(handle);
3318 return ret ? ret : copied;
3321 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3324 * Drop reserved blocks
3326 BUG_ON(!PageLocked(page));
3327 if (!page_has_buffers(page))
3330 ext4_da_page_release_reservation(page, offset);
3333 ext4_invalidatepage(page, offset);
3339 * Force all delayed allocation blocks to be allocated for a given inode.
3341 int ext4_alloc_da_blocks(struct inode *inode)
3343 trace_ext4_alloc_da_blocks(inode);
3345 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3346 !EXT4_I(inode)->i_reserved_meta_blocks)
3350 * We do something simple for now. The filemap_flush() will
3351 * also start triggering a write of the data blocks, which is
3352 * not strictly speaking necessary (and for users of
3353 * laptop_mode, not even desirable). However, to do otherwise
3354 * would require replicating code paths in:
3356 * ext4_da_writepages() ->
3357 * write_cache_pages() ---> (via passed in callback function)
3358 * __mpage_da_writepage() -->
3359 * mpage_add_bh_to_extent()
3360 * mpage_da_map_blocks()
3362 * The problem is that write_cache_pages(), located in
3363 * mm/page-writeback.c, marks pages clean in preparation for
3364 * doing I/O, which is not desirable if we're not planning on
3367 * We could call write_cache_pages(), and then redirty all of
3368 * the pages by calling redirty_page_for_writeback() but that
3369 * would be ugly in the extreme. So instead we would need to
3370 * replicate parts of the code in the above functions,
3371 * simplifying them becuase we wouldn't actually intend to
3372 * write out the pages, but rather only collect contiguous
3373 * logical block extents, call the multi-block allocator, and
3374 * then update the buffer heads with the block allocations.
3376 * For now, though, we'll cheat by calling filemap_flush(),
3377 * which will map the blocks, and start the I/O, but not
3378 * actually wait for the I/O to complete.
3380 return filemap_flush(inode->i_mapping);
3384 * bmap() is special. It gets used by applications such as lilo and by
3385 * the swapper to find the on-disk block of a specific piece of data.
3387 * Naturally, this is dangerous if the block concerned is still in the
3388 * journal. If somebody makes a swapfile on an ext4 data-journaling
3389 * filesystem and enables swap, then they may get a nasty shock when the
3390 * data getting swapped to that swapfile suddenly gets overwritten by
3391 * the original zero's written out previously to the journal and
3392 * awaiting writeback in the kernel's buffer cache.
3394 * So, if we see any bmap calls here on a modified, data-journaled file,
3395 * take extra steps to flush any blocks which might be in the cache.
3397 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3399 struct inode *inode = mapping->host;
3403 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3404 test_opt(inode->i_sb, DELALLOC)) {
3406 * With delalloc we want to sync the file
3407 * so that we can make sure we allocate
3410 filemap_write_and_wait(mapping);
3413 if (EXT4_JOURNAL(inode) &&
3414 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3416 * This is a REALLY heavyweight approach, but the use of
3417 * bmap on dirty files is expected to be extremely rare:
3418 * only if we run lilo or swapon on a freshly made file
3419 * do we expect this to happen.
3421 * (bmap requires CAP_SYS_RAWIO so this does not
3422 * represent an unprivileged user DOS attack --- we'd be
3423 * in trouble if mortal users could trigger this path at
3426 * NB. EXT4_STATE_JDATA is not set on files other than
3427 * regular files. If somebody wants to bmap a directory
3428 * or symlink and gets confused because the buffer
3429 * hasn't yet been flushed to disk, they deserve
3430 * everything they get.
3433 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3434 journal = EXT4_JOURNAL(inode);
3435 jbd2_journal_lock_updates(journal);
3436 err = jbd2_journal_flush(journal);
3437 jbd2_journal_unlock_updates(journal);
3443 return generic_block_bmap(mapping, block, ext4_get_block);
3446 static int ext4_readpage(struct file *file, struct page *page)
3448 return mpage_readpage(page, ext4_get_block);
3452 ext4_readpages(struct file *file, struct address_space *mapping,
3453 struct list_head *pages, unsigned nr_pages)
3455 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3458 static void ext4_free_io_end(ext4_io_end_t *io)
3467 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3469 struct buffer_head *head, *bh;
3470 unsigned int curr_off = 0;
3472 if (!page_has_buffers(page))
3474 head = bh = page_buffers(page);
3476 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3478 ext4_free_io_end(bh->b_private);
3479 bh->b_private = NULL;
3480 bh->b_end_io = NULL;
3482 curr_off = curr_off + bh->b_size;
3483 bh = bh->b_this_page;
3484 } while (bh != head);
3487 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3489 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3492 * free any io_end structure allocated for buffers to be discarded
3494 if (ext4_should_dioread_nolock(page->mapping->host))
3495 ext4_invalidatepage_free_endio(page, offset);
3497 * If it's a full truncate we just forget about the pending dirtying
3500 ClearPageChecked(page);
3503 jbd2_journal_invalidatepage(journal, page, offset);
3505 block_invalidatepage(page, offset);
3508 static int ext4_releasepage(struct page *page, gfp_t wait)
3510 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3512 WARN_ON(PageChecked(page));
3513 if (!page_has_buffers(page))
3516 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3518 return try_to_free_buffers(page);
3522 * O_DIRECT for ext3 (or indirect map) based files
3524 * If the O_DIRECT write will extend the file then add this inode to the
3525 * orphan list. So recovery will truncate it back to the original size
3526 * if the machine crashes during the write.
3528 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3529 * crashes then stale disk data _may_ be exposed inside the file. But current
3530 * VFS code falls back into buffered path in that case so we are safe.
3532 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3533 const struct iovec *iov, loff_t offset,
3534 unsigned long nr_segs)
3536 struct file *file = iocb->ki_filp;
3537 struct inode *inode = file->f_mapping->host;
3538 struct ext4_inode_info *ei = EXT4_I(inode);
3542 size_t count = iov_length(iov, nr_segs);
3546 loff_t final_size = offset + count;
3548 if (final_size > inode->i_size) {
3549 /* Credits for sb + inode write */
3550 handle = ext4_journal_start(inode, 2);
3551 if (IS_ERR(handle)) {
3552 ret = PTR_ERR(handle);
3555 ret = ext4_orphan_add(handle, inode);
3557 ext4_journal_stop(handle);
3561 ei->i_disksize = inode->i_size;
3562 ext4_journal_stop(handle);
3567 if (rw == READ && ext4_should_dioread_nolock(inode))
3568 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
3569 inode->i_sb->s_bdev, iov,
3571 ext4_get_block, NULL);
3573 ret = blockdev_direct_IO(rw, iocb, inode,
3574 inode->i_sb->s_bdev, iov,
3576 ext4_get_block, NULL);
3577 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3583 /* Credits for sb + inode write */
3584 handle = ext4_journal_start(inode, 2);
3585 if (IS_ERR(handle)) {
3586 /* This is really bad luck. We've written the data
3587 * but cannot extend i_size. Bail out and pretend
3588 * the write failed... */
3589 ret = PTR_ERR(handle);
3591 ext4_orphan_del(NULL, inode);
3596 ext4_orphan_del(handle, inode);
3598 loff_t end = offset + ret;
3599 if (end > inode->i_size) {
3600 ei->i_disksize = end;
3601 i_size_write(inode, end);
3603 * We're going to return a positive `ret'
3604 * here due to non-zero-length I/O, so there's
3605 * no way of reporting error returns from
3606 * ext4_mark_inode_dirty() to userspace. So
3609 ext4_mark_inode_dirty(handle, inode);
3612 err = ext4_journal_stop(handle);
3621 * ext4_get_block used when preparing for a DIO write or buffer write.
3622 * We allocate an uinitialized extent if blocks haven't been allocated.
3623 * The extent will be converted to initialized after the IO is complete.
3625 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3626 struct buffer_head *bh_result, int create)
3628 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3629 inode->i_ino, create);
3630 return _ext4_get_block(inode, iblock, bh_result,
3631 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3634 static void dump_completed_IO(struct inode * inode)
3637 struct list_head *cur, *before, *after;
3638 ext4_io_end_t *io, *io0, *io1;
3639 unsigned long flags;
3641 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3642 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3646 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3647 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3648 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3651 io0 = container_of(before, ext4_io_end_t, list);
3653 io1 = container_of(after, ext4_io_end_t, list);
3655 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3656 io, inode->i_ino, io0, io1);
3658 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3663 * check a range of space and convert unwritten extents to written.
3665 static int ext4_end_io_nolock(ext4_io_end_t *io)
3667 struct inode *inode = io->inode;
3668 loff_t offset = io->offset;
3669 ssize_t size = io->size;
3672 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3673 "list->prev 0x%p\n",
3674 io, inode->i_ino, io->list.next, io->list.prev);
3676 if (list_empty(&io->list))
3679 if (io->flag != EXT4_IO_UNWRITTEN)
3682 ret = ext4_convert_unwritten_extents(inode, offset, size);
3684 printk(KERN_EMERG "%s: failed to convert unwritten"
3685 "extents to written extents, error is %d"
3686 " io is still on inode %lu aio dio list\n",
3687 __func__, ret, inode->i_ino);
3692 aio_complete(io->iocb, io->result, 0);
3693 /* clear the DIO AIO unwritten flag */
3699 * work on completed aio dio IO, to convert unwritten extents to extents
3701 static void ext4_end_io_work(struct work_struct *work)
3703 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3704 struct inode *inode = io->inode;
3705 struct ext4_inode_info *ei = EXT4_I(inode);
3706 unsigned long flags;
3709 mutex_lock(&inode->i_mutex);
3710 ret = ext4_end_io_nolock(io);
3712 mutex_unlock(&inode->i_mutex);
3716 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3717 if (!list_empty(&io->list))
3718 list_del_init(&io->list);
3719 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3720 mutex_unlock(&inode->i_mutex);
3721 ext4_free_io_end(io);
3725 * This function is called from ext4_sync_file().
3727 * When IO is completed, the work to convert unwritten extents to
3728 * written is queued on workqueue but may not get immediately
3729 * scheduled. When fsync is called, we need to ensure the
3730 * conversion is complete before fsync returns.
3731 * The inode keeps track of a list of pending/completed IO that
3732 * might needs to do the conversion. This function walks through
3733 * the list and convert the related unwritten extents for completed IO
3735 * The function return the number of pending IOs on success.
3737 int flush_completed_IO(struct inode *inode)
3740 struct ext4_inode_info *ei = EXT4_I(inode);
3741 unsigned long flags;
3745 if (list_empty(&ei->i_completed_io_list))
3748 dump_completed_IO(inode);
3749 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3750 while (!list_empty(&ei->i_completed_io_list)){
3751 io = list_entry(ei->i_completed_io_list.next,
3752 ext4_io_end_t, list);
3754 * Calling ext4_end_io_nolock() to convert completed
3757 * When ext4_sync_file() is called, run_queue() may already
3758 * about to flush the work corresponding to this io structure.
3759 * It will be upset if it founds the io structure related
3760 * to the work-to-be schedule is freed.
3762 * Thus we need to keep the io structure still valid here after
3763 * convertion finished. The io structure has a flag to
3764 * avoid double converting from both fsync and background work
3767 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3768 ret = ext4_end_io_nolock(io);
3769 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3773 list_del_init(&io->list);
3775 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3776 return (ret2 < 0) ? ret2 : 0;
3779 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3781 ext4_io_end_t *io = NULL;
3783 io = kmalloc(sizeof(*io), flags);
3794 INIT_WORK(&io->work, ext4_end_io_work);
3795 INIT_LIST_HEAD(&io->list);
3801 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3802 ssize_t size, void *private, int ret,
3805 ext4_io_end_t *io_end = iocb->private;
3806 struct workqueue_struct *wq;
3807 unsigned long flags;
3808 struct ext4_inode_info *ei;
3810 /* if not async direct IO or dio with 0 bytes write, just return */
3811 if (!io_end || !size)
3814 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3815 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3816 iocb->private, io_end->inode->i_ino, iocb, offset,
3819 /* if not aio dio with unwritten extents, just free io and return */
3820 if (io_end->flag != EXT4_IO_UNWRITTEN){
3821 ext4_free_io_end(io_end);
3822 iocb->private = NULL;
3825 aio_complete(iocb, ret, 0);
3829 io_end->offset = offset;
3830 io_end->size = size;
3832 io_end->iocb = iocb;
3833 io_end->result = ret;
3835 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3837 /* queue the work to convert unwritten extents to written */
3838 queue_work(wq, &io_end->work);
3840 /* Add the io_end to per-inode completed aio dio list*/
3841 ei = EXT4_I(io_end->inode);
3842 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3843 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3844 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3845 iocb->private = NULL;
3848 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3850 ext4_io_end_t *io_end = bh->b_private;
3851 struct workqueue_struct *wq;
3852 struct inode *inode;
3853 unsigned long flags;
3855 if (!test_clear_buffer_uninit(bh) || !io_end)
3858 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3859 printk("sb umounted, discard end_io request for inode %lu\n",
3860 io_end->inode->i_ino);
3861 ext4_free_io_end(io_end);
3865 io_end->flag = EXT4_IO_UNWRITTEN;
3866 inode = io_end->inode;
3868 /* Add the io_end to per-inode completed io list*/
3869 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3870 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3871 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3873 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3874 /* queue the work to convert unwritten extents to written */
3875 queue_work(wq, &io_end->work);
3877 bh->b_private = NULL;
3878 bh->b_end_io = NULL;
3879 clear_buffer_uninit(bh);
3880 end_buffer_async_write(bh, uptodate);
3883 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3885 ext4_io_end_t *io_end;
3886 struct page *page = bh->b_page;
3887 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3888 size_t size = bh->b_size;
3891 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3893 if (printk_ratelimit())
3894 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3898 io_end->offset = offset;
3899 io_end->size = size;
3901 * We need to hold a reference to the page to make sure it
3902 * doesn't get evicted before ext4_end_io_work() has a chance
3903 * to convert the extent from written to unwritten.
3905 io_end->page = page;
3906 get_page(io_end->page);
3908 bh->b_private = io_end;
3909 bh->b_end_io = ext4_end_io_buffer_write;
3914 * For ext4 extent files, ext4 will do direct-io write to holes,
3915 * preallocated extents, and those write extend the file, no need to
3916 * fall back to buffered IO.
3918 * For holes, we fallocate those blocks, mark them as unintialized
3919 * If those blocks were preallocated, we mark sure they are splited, but
3920 * still keep the range to write as unintialized.
3922 * The unwrritten extents will be converted to written when DIO is completed.
3923 * For async direct IO, since the IO may still pending when return, we
3924 * set up an end_io call back function, which will do the convertion
3925 * when async direct IO completed.
3927 * If the O_DIRECT write will extend the file then add this inode to the
3928 * orphan list. So recovery will truncate it back to the original size
3929 * if the machine crashes during the write.
3932 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3933 const struct iovec *iov, loff_t offset,
3934 unsigned long nr_segs)
3936 struct file *file = iocb->ki_filp;
3937 struct inode *inode = file->f_mapping->host;
3939 size_t count = iov_length(iov, nr_segs);
3941 loff_t final_size = offset + count;
3942 if (rw == WRITE && final_size <= inode->i_size) {
3944 * We could direct write to holes and fallocate.
3946 * Allocated blocks to fill the hole are marked as uninitialized
3947 * to prevent paralel buffered read to expose the stale data
3948 * before DIO complete the data IO.
3950 * As to previously fallocated extents, ext4 get_block
3951 * will just simply mark the buffer mapped but still
3952 * keep the extents uninitialized.
3954 * for non AIO case, we will convert those unwritten extents
3955 * to written after return back from blockdev_direct_IO.
3957 * for async DIO, the conversion needs to be defered when
3958 * the IO is completed. The ext4 end_io callback function
3959 * will be called to take care of the conversion work.
3960 * Here for async case, we allocate an io_end structure to
3963 iocb->private = NULL;
3964 EXT4_I(inode)->cur_aio_dio = NULL;
3965 if (!is_sync_kiocb(iocb)) {
3966 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3970 * we save the io structure for current async
3971 * direct IO, so that later ext4_map_blocks()
3972 * could flag the io structure whether there
3973 * is a unwritten extents needs to be converted
3974 * when IO is completed.
3976 EXT4_I(inode)->cur_aio_dio = iocb->private;
3979 ret = blockdev_direct_IO(rw, iocb, inode,
3980 inode->i_sb->s_bdev, iov,
3982 ext4_get_block_write,
3985 EXT4_I(inode)->cur_aio_dio = NULL;
3987 * The io_end structure takes a reference to the inode,
3988 * that structure needs to be destroyed and the
3989 * reference to the inode need to be dropped, when IO is
3990 * complete, even with 0 byte write, or failed.
3992 * In the successful AIO DIO case, the io_end structure will be
3993 * desctroyed and the reference to the inode will be dropped
3994 * after the end_io call back function is called.
3996 * In the case there is 0 byte write, or error case, since
3997 * VFS direct IO won't invoke the end_io call back function,
3998 * we need to free the end_io structure here.
4000 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
4001 ext4_free_io_end(iocb->private);
4002 iocb->private = NULL;
4003 } else if (ret > 0 && ext4_test_inode_state(inode,
4004 EXT4_STATE_DIO_UNWRITTEN)) {
4007 * for non AIO case, since the IO is already
4008 * completed, we could do the convertion right here
4010 err = ext4_convert_unwritten_extents(inode,
4014 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4019 /* for write the the end of file case, we fall back to old way */
4020 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4023 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
4024 const struct iovec *iov, loff_t offset,
4025 unsigned long nr_segs)
4027 struct file *file = iocb->ki_filp;
4028 struct inode *inode = file->f_mapping->host;
4030 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4031 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
4033 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4037 * Pages can be marked dirty completely asynchronously from ext4's journalling
4038 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4039 * much here because ->set_page_dirty is called under VFS locks. The page is
4040 * not necessarily locked.
4042 * We cannot just dirty the page and leave attached buffers clean, because the
4043 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4044 * or jbddirty because all the journalling code will explode.
4046 * So what we do is to mark the page "pending dirty" and next time writepage
4047 * is called, propagate that into the buffers appropriately.
4049 static int ext4_journalled_set_page_dirty(struct page *page)
4051 SetPageChecked(page);
4052 return __set_page_dirty_nobuffers(page);
4055 static const struct address_space_operations ext4_ordered_aops = {
4056 .readpage = ext4_readpage,
4057 .readpages = ext4_readpages,
4058 .writepage = ext4_writepage,
4059 .sync_page = block_sync_page,
4060 .write_begin = ext4_write_begin,
4061 .write_end = ext4_ordered_write_end,
4063 .invalidatepage = ext4_invalidatepage,
4064 .releasepage = ext4_releasepage,
4065 .direct_IO = ext4_direct_IO,
4066 .migratepage = buffer_migrate_page,
4067 .is_partially_uptodate = block_is_partially_uptodate,
4068 .error_remove_page = generic_error_remove_page,
4071 static const struct address_space_operations ext4_writeback_aops = {
4072 .readpage = ext4_readpage,
4073 .readpages = ext4_readpages,
4074 .writepage = ext4_writepage,
4075 .sync_page = block_sync_page,
4076 .write_begin = ext4_write_begin,
4077 .write_end = ext4_writeback_write_end,
4079 .invalidatepage = ext4_invalidatepage,
4080 .releasepage = ext4_releasepage,
4081 .direct_IO = ext4_direct_IO,
4082 .migratepage = buffer_migrate_page,
4083 .is_partially_uptodate = block_is_partially_uptodate,
4084 .error_remove_page = generic_error_remove_page,
4087 static const struct address_space_operations ext4_journalled_aops = {
4088 .readpage = ext4_readpage,
4089 .readpages = ext4_readpages,
4090 .writepage = ext4_writepage,
4091 .sync_page = block_sync_page,
4092 .write_begin = ext4_write_begin,
4093 .write_end = ext4_journalled_write_end,
4094 .set_page_dirty = ext4_journalled_set_page_dirty,
4096 .invalidatepage = ext4_invalidatepage,
4097 .releasepage = ext4_releasepage,
4098 .is_partially_uptodate = block_is_partially_uptodate,
4099 .error_remove_page = generic_error_remove_page,
4102 static const struct address_space_operations ext4_da_aops = {
4103 .readpage = ext4_readpage,
4104 .readpages = ext4_readpages,
4105 .writepage = ext4_writepage,
4106 .writepages = ext4_da_writepages,
4107 .sync_page = block_sync_page,
4108 .write_begin = ext4_da_write_begin,
4109 .write_end = ext4_da_write_end,
4111 .invalidatepage = ext4_da_invalidatepage,
4112 .releasepage = ext4_releasepage,
4113 .direct_IO = ext4_direct_IO,
4114 .migratepage = buffer_migrate_page,
4115 .is_partially_uptodate = block_is_partially_uptodate,
4116 .error_remove_page = generic_error_remove_page,
4119 void ext4_set_aops(struct inode *inode)
4121 if (ext4_should_order_data(inode) &&
4122 test_opt(inode->i_sb, DELALLOC))
4123 inode->i_mapping->a_ops = &ext4_da_aops;
4124 else if (ext4_should_order_data(inode))
4125 inode->i_mapping->a_ops = &ext4_ordered_aops;
4126 else if (ext4_should_writeback_data(inode) &&
4127 test_opt(inode->i_sb, DELALLOC))
4128 inode->i_mapping->a_ops = &ext4_da_aops;
4129 else if (ext4_should_writeback_data(inode))
4130 inode->i_mapping->a_ops = &ext4_writeback_aops;
4132 inode->i_mapping->a_ops = &ext4_journalled_aops;
4136 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4137 * up to the end of the block which corresponds to `from'.
4138 * This required during truncate. We need to physically zero the tail end
4139 * of that block so it doesn't yield old data if the file is later grown.
4141 int ext4_block_truncate_page(handle_t *handle,
4142 struct address_space *mapping, loff_t from)
4144 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4145 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4146 unsigned blocksize, length, pos;
4148 struct inode *inode = mapping->host;
4149 struct buffer_head *bh;
4153 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4154 mapping_gfp_mask(mapping) & ~__GFP_FS);
4158 blocksize = inode->i_sb->s_blocksize;
4159 length = blocksize - (offset & (blocksize - 1));
4160 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4162 if (!page_has_buffers(page))
4163 create_empty_buffers(page, blocksize, 0);
4165 /* Find the buffer that contains "offset" */
4166 bh = page_buffers(page);
4168 while (offset >= pos) {
4169 bh = bh->b_this_page;
4175 if (buffer_freed(bh)) {
4176 BUFFER_TRACE(bh, "freed: skip");
4180 if (!buffer_mapped(bh)) {
4181 BUFFER_TRACE(bh, "unmapped");
4182 ext4_get_block(inode, iblock, bh, 0);
4183 /* unmapped? It's a hole - nothing to do */
4184 if (!buffer_mapped(bh)) {
4185 BUFFER_TRACE(bh, "still unmapped");
4190 /* Ok, it's mapped. Make sure it's up-to-date */
4191 if (PageUptodate(page))
4192 set_buffer_uptodate(bh);
4194 if (!buffer_uptodate(bh)) {
4196 ll_rw_block(READ, 1, &bh);
4198 /* Uhhuh. Read error. Complain and punt. */
4199 if (!buffer_uptodate(bh))
4203 if (ext4_should_journal_data(inode)) {
4204 BUFFER_TRACE(bh, "get write access");
4205 err = ext4_journal_get_write_access(handle, bh);
4210 zero_user(page, offset, length);
4212 BUFFER_TRACE(bh, "zeroed end of block");
4215 if (ext4_should_journal_data(inode)) {
4216 err = ext4_handle_dirty_metadata(handle, inode, bh);
4218 if (ext4_should_order_data(inode))
4219 err = ext4_jbd2_file_inode(handle, inode);
4220 mark_buffer_dirty(bh);
4225 page_cache_release(page);
4230 * Probably it should be a library function... search for first non-zero word
4231 * or memcmp with zero_page, whatever is better for particular architecture.
4234 static inline int all_zeroes(__le32 *p, __le32 *q)
4243 * ext4_find_shared - find the indirect blocks for partial truncation.
4244 * @inode: inode in question
4245 * @depth: depth of the affected branch
4246 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4247 * @chain: place to store the pointers to partial indirect blocks
4248 * @top: place to the (detached) top of branch
4250 * This is a helper function used by ext4_truncate().
4252 * When we do truncate() we may have to clean the ends of several
4253 * indirect blocks but leave the blocks themselves alive. Block is
4254 * partially truncated if some data below the new i_size is refered
4255 * from it (and it is on the path to the first completely truncated
4256 * data block, indeed). We have to free the top of that path along
4257 * with everything to the right of the path. Since no allocation
4258 * past the truncation point is possible until ext4_truncate()
4259 * finishes, we may safely do the latter, but top of branch may
4260 * require special attention - pageout below the truncation point
4261 * might try to populate it.
4263 * We atomically detach the top of branch from the tree, store the
4264 * block number of its root in *@top, pointers to buffer_heads of
4265 * partially truncated blocks - in @chain[].bh and pointers to
4266 * their last elements that should not be removed - in
4267 * @chain[].p. Return value is the pointer to last filled element
4270 * The work left to caller to do the actual freeing of subtrees:
4271 * a) free the subtree starting from *@top
4272 * b) free the subtrees whose roots are stored in
4273 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4274 * c) free the subtrees growing from the inode past the @chain[0].
4275 * (no partially truncated stuff there). */
4277 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4278 ext4_lblk_t offsets[4], Indirect chain[4],
4281 Indirect *partial, *p;
4285 /* Make k index the deepest non-null offset + 1 */
4286 for (k = depth; k > 1 && !offsets[k-1]; k--)
4288 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4289 /* Writer: pointers */
4291 partial = chain + k-1;
4293 * If the branch acquired continuation since we've looked at it -
4294 * fine, it should all survive and (new) top doesn't belong to us.
4296 if (!partial->key && *partial->p)
4299 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4302 * OK, we've found the last block that must survive. The rest of our
4303 * branch should be detached before unlocking. However, if that rest
4304 * of branch is all ours and does not grow immediately from the inode
4305 * it's easier to cheat and just decrement partial->p.
4307 if (p == chain + k - 1 && p > chain) {
4311 /* Nope, don't do this in ext4. Must leave the tree intact */
4318 while (partial > p) {
4319 brelse(partial->bh);
4327 * Zero a number of block pointers in either an inode or an indirect block.
4328 * If we restart the transaction we must again get write access to the
4329 * indirect block for further modification.
4331 * We release `count' blocks on disk, but (last - first) may be greater
4332 * than `count' because there can be holes in there.
4334 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4335 struct buffer_head *bh,
4336 ext4_fsblk_t block_to_free,
4337 unsigned long count, __le32 *first,
4341 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4343 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4344 flags |= EXT4_FREE_BLOCKS_METADATA;
4346 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4348 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4349 "blocks %llu len %lu",
4350 (unsigned long long) block_to_free, count);
4354 if (try_to_extend_transaction(handle, inode)) {
4356 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4357 ext4_handle_dirty_metadata(handle, inode, bh);
4359 ext4_mark_inode_dirty(handle, inode);
4360 ext4_truncate_restart_trans(handle, inode,
4361 blocks_for_truncate(inode));
4363 BUFFER_TRACE(bh, "retaking write access");
4364 ext4_journal_get_write_access(handle, bh);
4368 for (p = first; p < last; p++)
4371 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4376 * ext4_free_data - free a list of data blocks
4377 * @handle: handle for this transaction
4378 * @inode: inode we are dealing with
4379 * @this_bh: indirect buffer_head which contains *@first and *@last
4380 * @first: array of block numbers
4381 * @last: points immediately past the end of array
4383 * We are freeing all blocks refered from that array (numbers are stored as
4384 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4386 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4387 * blocks are contiguous then releasing them at one time will only affect one
4388 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4389 * actually use a lot of journal space.
4391 * @this_bh will be %NULL if @first and @last point into the inode's direct
4394 static void ext4_free_data(handle_t *handle, struct inode *inode,
4395 struct buffer_head *this_bh,
4396 __le32 *first, __le32 *last)
4398 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4399 unsigned long count = 0; /* Number of blocks in the run */
4400 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4403 ext4_fsblk_t nr; /* Current block # */
4404 __le32 *p; /* Pointer into inode/ind
4405 for current block */
4408 if (this_bh) { /* For indirect block */
4409 BUFFER_TRACE(this_bh, "get_write_access");
4410 err = ext4_journal_get_write_access(handle, this_bh);
4411 /* Important: if we can't update the indirect pointers
4412 * to the blocks, we can't free them. */
4417 for (p = first; p < last; p++) {
4418 nr = le32_to_cpu(*p);
4420 /* accumulate blocks to free if they're contiguous */
4423 block_to_free_p = p;
4425 } else if (nr == block_to_free + count) {
4428 if (ext4_clear_blocks(handle, inode, this_bh,
4429 block_to_free, count,
4430 block_to_free_p, p))
4433 block_to_free_p = p;
4440 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4441 count, block_to_free_p, p);
4444 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4447 * The buffer head should have an attached journal head at this
4448 * point. However, if the data is corrupted and an indirect
4449 * block pointed to itself, it would have been detached when
4450 * the block was cleared. Check for this instead of OOPSing.
4452 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4453 ext4_handle_dirty_metadata(handle, inode, this_bh);
4455 EXT4_ERROR_INODE(inode,
4456 "circular indirect block detected at "
4458 (unsigned long long) this_bh->b_blocknr);
4463 * ext4_free_branches - free an array of branches
4464 * @handle: JBD handle for this transaction
4465 * @inode: inode we are dealing with
4466 * @parent_bh: the buffer_head which contains *@first and *@last
4467 * @first: array of block numbers
4468 * @last: pointer immediately past the end of array
4469 * @depth: depth of the branches to free
4471 * We are freeing all blocks refered from these branches (numbers are
4472 * stored as little-endian 32-bit) and updating @inode->i_blocks
4475 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4476 struct buffer_head *parent_bh,
4477 __le32 *first, __le32 *last, int depth)
4482 if (ext4_handle_is_aborted(handle))
4486 struct buffer_head *bh;
4487 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4489 while (--p >= first) {
4490 nr = le32_to_cpu(*p);
4492 continue; /* A hole */
4494 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4496 EXT4_ERROR_INODE(inode,
4497 "invalid indirect mapped "
4498 "block %lu (level %d)",
4499 (unsigned long) nr, depth);
4503 /* Go read the buffer for the next level down */
4504 bh = sb_bread(inode->i_sb, nr);
4507 * A read failure? Report error and clear slot
4511 EXT4_ERROR_INODE_BLOCK(inode, nr,
4516 /* This zaps the entire block. Bottom up. */
4517 BUFFER_TRACE(bh, "free child branches");
4518 ext4_free_branches(handle, inode, bh,
4519 (__le32 *) bh->b_data,
4520 (__le32 *) bh->b_data + addr_per_block,
4524 * Everything below this this pointer has been
4525 * released. Now let this top-of-subtree go.
4527 * We want the freeing of this indirect block to be
4528 * atomic in the journal with the updating of the
4529 * bitmap block which owns it. So make some room in
4532 * We zero the parent pointer *after* freeing its
4533 * pointee in the bitmaps, so if extend_transaction()
4534 * for some reason fails to put the bitmap changes and
4535 * the release into the same transaction, recovery
4536 * will merely complain about releasing a free block,
4537 * rather than leaking blocks.
4539 if (ext4_handle_is_aborted(handle))
4541 if (try_to_extend_transaction(handle, inode)) {
4542 ext4_mark_inode_dirty(handle, inode);
4543 ext4_truncate_restart_trans(handle, inode,
4544 blocks_for_truncate(inode));
4548 * The forget flag here is critical because if
4549 * we are journaling (and not doing data
4550 * journaling), we have to make sure a revoke
4551 * record is written to prevent the journal
4552 * replay from overwriting the (former)
4553 * indirect block if it gets reallocated as a
4554 * data block. This must happen in the same
4555 * transaction where the data blocks are
4558 ext4_free_blocks(handle, inode, 0, nr, 1,
4559 EXT4_FREE_BLOCKS_METADATA|
4560 EXT4_FREE_BLOCKS_FORGET);
4564 * The block which we have just freed is
4565 * pointed to by an indirect block: journal it
4567 BUFFER_TRACE(parent_bh, "get_write_access");
4568 if (!ext4_journal_get_write_access(handle,
4571 BUFFER_TRACE(parent_bh,
4572 "call ext4_handle_dirty_metadata");
4573 ext4_handle_dirty_metadata(handle,
4580 /* We have reached the bottom of the tree. */
4581 BUFFER_TRACE(parent_bh, "free data blocks");
4582 ext4_free_data(handle, inode, parent_bh, first, last);
4586 int ext4_can_truncate(struct inode *inode)
4588 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4590 if (S_ISREG(inode->i_mode))
4592 if (S_ISDIR(inode->i_mode))
4594 if (S_ISLNK(inode->i_mode))
4595 return !ext4_inode_is_fast_symlink(inode);
4602 * We block out ext4_get_block() block instantiations across the entire
4603 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4604 * simultaneously on behalf of the same inode.
4606 * As we work through the truncate and commmit bits of it to the journal there
4607 * is one core, guiding principle: the file's tree must always be consistent on
4608 * disk. We must be able to restart the truncate after a crash.
4610 * The file's tree may be transiently inconsistent in memory (although it
4611 * probably isn't), but whenever we close off and commit a journal transaction,
4612 * the contents of (the filesystem + the journal) must be consistent and
4613 * restartable. It's pretty simple, really: bottom up, right to left (although
4614 * left-to-right works OK too).
4616 * Note that at recovery time, journal replay occurs *before* the restart of
4617 * truncate against the orphan inode list.
4619 * The committed inode has the new, desired i_size (which is the same as
4620 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4621 * that this inode's truncate did not complete and it will again call
4622 * ext4_truncate() to have another go. So there will be instantiated blocks
4623 * to the right of the truncation point in a crashed ext4 filesystem. But
4624 * that's fine - as long as they are linked from the inode, the post-crash
4625 * ext4_truncate() run will find them and release them.
4627 void ext4_truncate(struct inode *inode)
4630 struct ext4_inode_info *ei = EXT4_I(inode);
4631 __le32 *i_data = ei->i_data;
4632 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4633 struct address_space *mapping = inode->i_mapping;
4634 ext4_lblk_t offsets[4];
4639 ext4_lblk_t last_block;
4640 unsigned blocksize = inode->i_sb->s_blocksize;
4642 if (!ext4_can_truncate(inode))
4645 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4647 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4648 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4650 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4651 ext4_ext_truncate(inode);
4655 handle = start_transaction(inode);
4657 return; /* AKPM: return what? */
4659 last_block = (inode->i_size + blocksize-1)
4660 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4662 if (inode->i_size & (blocksize - 1))
4663 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4666 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4668 goto out_stop; /* error */
4671 * OK. This truncate is going to happen. We add the inode to the
4672 * orphan list, so that if this truncate spans multiple transactions,
4673 * and we crash, we will resume the truncate when the filesystem
4674 * recovers. It also marks the inode dirty, to catch the new size.
4676 * Implication: the file must always be in a sane, consistent
4677 * truncatable state while each transaction commits.
4679 if (ext4_orphan_add(handle, inode))
4683 * From here we block out all ext4_get_block() callers who want to
4684 * modify the block allocation tree.
4686 down_write(&ei->i_data_sem);
4688 ext4_discard_preallocations(inode);
4691 * The orphan list entry will now protect us from any crash which
4692 * occurs before the truncate completes, so it is now safe to propagate
4693 * the new, shorter inode size (held for now in i_size) into the
4694 * on-disk inode. We do this via i_disksize, which is the value which
4695 * ext4 *really* writes onto the disk inode.
4697 ei->i_disksize = inode->i_size;
4699 if (n == 1) { /* direct blocks */
4700 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4701 i_data + EXT4_NDIR_BLOCKS);
4705 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4706 /* Kill the top of shared branch (not detached) */
4708 if (partial == chain) {
4709 /* Shared branch grows from the inode */
4710 ext4_free_branches(handle, inode, NULL,
4711 &nr, &nr+1, (chain+n-1) - partial);
4714 * We mark the inode dirty prior to restart,
4715 * and prior to stop. No need for it here.
4718 /* Shared branch grows from an indirect block */
4719 BUFFER_TRACE(partial->bh, "get_write_access");
4720 ext4_free_branches(handle, inode, partial->bh,
4722 partial->p+1, (chain+n-1) - partial);
4725 /* Clear the ends of indirect blocks on the shared branch */
4726 while (partial > chain) {
4727 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4728 (__le32*)partial->bh->b_data+addr_per_block,
4729 (chain+n-1) - partial);
4730 BUFFER_TRACE(partial->bh, "call brelse");
4731 brelse(partial->bh);
4735 /* Kill the remaining (whole) subtrees */
4736 switch (offsets[0]) {
4738 nr = i_data[EXT4_IND_BLOCK];
4740 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4741 i_data[EXT4_IND_BLOCK] = 0;
4743 case EXT4_IND_BLOCK:
4744 nr = i_data[EXT4_DIND_BLOCK];
4746 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4747 i_data[EXT4_DIND_BLOCK] = 0;
4749 case EXT4_DIND_BLOCK:
4750 nr = i_data[EXT4_TIND_BLOCK];
4752 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4753 i_data[EXT4_TIND_BLOCK] = 0;
4755 case EXT4_TIND_BLOCK:
4759 up_write(&ei->i_data_sem);
4760 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4761 ext4_mark_inode_dirty(handle, inode);
4764 * In a multi-transaction truncate, we only make the final transaction
4768 ext4_handle_sync(handle);
4771 * If this was a simple ftruncate(), and the file will remain alive
4772 * then we need to clear up the orphan record which we created above.
4773 * However, if this was a real unlink then we were called by
4774 * ext4_delete_inode(), and we allow that function to clean up the
4775 * orphan info for us.
4778 ext4_orphan_del(handle, inode);
4780 ext4_journal_stop(handle);
4784 * ext4_get_inode_loc returns with an extra refcount against the inode's
4785 * underlying buffer_head on success. If 'in_mem' is true, we have all
4786 * data in memory that is needed to recreate the on-disk version of this
4789 static int __ext4_get_inode_loc(struct inode *inode,
4790 struct ext4_iloc *iloc, int in_mem)
4792 struct ext4_group_desc *gdp;
4793 struct buffer_head *bh;
4794 struct super_block *sb = inode->i_sb;
4796 int inodes_per_block, inode_offset;
4799 if (!ext4_valid_inum(sb, inode->i_ino))
4802 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4803 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4808 * Figure out the offset within the block group inode table
4810 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4811 inode_offset = ((inode->i_ino - 1) %
4812 EXT4_INODES_PER_GROUP(sb));
4813 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4814 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4816 bh = sb_getblk(sb, block);
4818 EXT4_ERROR_INODE_BLOCK(inode, block,
4819 "unable to read itable block");
4822 if (!buffer_uptodate(bh)) {
4826 * If the buffer has the write error flag, we have failed
4827 * to write out another inode in the same block. In this
4828 * case, we don't have to read the block because we may
4829 * read the old inode data successfully.
4831 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4832 set_buffer_uptodate(bh);
4834 if (buffer_uptodate(bh)) {
4835 /* someone brought it uptodate while we waited */
4841 * If we have all information of the inode in memory and this
4842 * is the only valid inode in the block, we need not read the
4846 struct buffer_head *bitmap_bh;
4849 start = inode_offset & ~(inodes_per_block - 1);
4851 /* Is the inode bitmap in cache? */
4852 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4857 * If the inode bitmap isn't in cache then the
4858 * optimisation may end up performing two reads instead
4859 * of one, so skip it.
4861 if (!buffer_uptodate(bitmap_bh)) {
4865 for (i = start; i < start + inodes_per_block; i++) {
4866 if (i == inode_offset)
4868 if (ext4_test_bit(i, bitmap_bh->b_data))
4872 if (i == start + inodes_per_block) {
4873 /* all other inodes are free, so skip I/O */
4874 memset(bh->b_data, 0, bh->b_size);
4875 set_buffer_uptodate(bh);
4883 * If we need to do any I/O, try to pre-readahead extra
4884 * blocks from the inode table.
4886 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4887 ext4_fsblk_t b, end, table;
4890 table = ext4_inode_table(sb, gdp);
4891 /* s_inode_readahead_blks is always a power of 2 */
4892 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4895 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4896 num = EXT4_INODES_PER_GROUP(sb);
4897 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4898 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4899 num -= ext4_itable_unused_count(sb, gdp);
4900 table += num / inodes_per_block;
4904 sb_breadahead(sb, b++);
4908 * There are other valid inodes in the buffer, this inode
4909 * has in-inode xattrs, or we don't have this inode in memory.
4910 * Read the block from disk.
4913 bh->b_end_io = end_buffer_read_sync;
4914 submit_bh(READ_META, bh);
4916 if (!buffer_uptodate(bh)) {
4917 EXT4_ERROR_INODE_BLOCK(inode, block,
4918 "unable to read itable block");
4928 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4930 /* We have all inode data except xattrs in memory here. */
4931 return __ext4_get_inode_loc(inode, iloc,
4932 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4935 void ext4_set_inode_flags(struct inode *inode)
4937 unsigned int flags = EXT4_I(inode)->i_flags;
4939 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4940 if (flags & EXT4_SYNC_FL)
4941 inode->i_flags |= S_SYNC;
4942 if (flags & EXT4_APPEND_FL)
4943 inode->i_flags |= S_APPEND;
4944 if (flags & EXT4_IMMUTABLE_FL)
4945 inode->i_flags |= S_IMMUTABLE;
4946 if (flags & EXT4_NOATIME_FL)
4947 inode->i_flags |= S_NOATIME;
4948 if (flags & EXT4_DIRSYNC_FL)
4949 inode->i_flags |= S_DIRSYNC;
4952 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4953 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4955 unsigned int vfs_fl;
4956 unsigned long old_fl, new_fl;
4959 vfs_fl = ei->vfs_inode.i_flags;
4960 old_fl = ei->i_flags;
4961 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4962 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4964 if (vfs_fl & S_SYNC)
4965 new_fl |= EXT4_SYNC_FL;
4966 if (vfs_fl & S_APPEND)
4967 new_fl |= EXT4_APPEND_FL;
4968 if (vfs_fl & S_IMMUTABLE)
4969 new_fl |= EXT4_IMMUTABLE_FL;
4970 if (vfs_fl & S_NOATIME)
4971 new_fl |= EXT4_NOATIME_FL;
4972 if (vfs_fl & S_DIRSYNC)
4973 new_fl |= EXT4_DIRSYNC_FL;
4974 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4977 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4978 struct ext4_inode_info *ei)
4981 struct inode *inode = &(ei->vfs_inode);
4982 struct super_block *sb = inode->i_sb;
4984 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4985 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4986 /* we are using combined 48 bit field */
4987 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4988 le32_to_cpu(raw_inode->i_blocks_lo);
4989 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4990 /* i_blocks represent file system block size */
4991 return i_blocks << (inode->i_blkbits - 9);
4996 return le32_to_cpu(raw_inode->i_blocks_lo);
5000 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
5002 struct ext4_iloc iloc;
5003 struct ext4_inode *raw_inode;
5004 struct ext4_inode_info *ei;
5005 struct inode *inode;
5006 journal_t *journal = EXT4_SB(sb)->s_journal;
5010 inode = iget_locked(sb, ino);
5012 return ERR_PTR(-ENOMEM);
5013 if (!(inode->i_state & I_NEW))
5019 ret = __ext4_get_inode_loc(inode, &iloc, 0);
5022 raw_inode = ext4_raw_inode(&iloc);
5023 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5024 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5025 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5026 if (!(test_opt(inode->i_sb, NO_UID32))) {
5027 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5028 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5030 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
5032 ei->i_state_flags = 0;
5033 ei->i_dir_start_lookup = 0;
5034 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5035 /* We now have enough fields to check if the inode was active or not.
5036 * This is needed because nfsd might try to access dead inodes
5037 * the test is that same one that e2fsck uses
5038 * NeilBrown 1999oct15
5040 if (inode->i_nlink == 0) {
5041 if (inode->i_mode == 0 ||
5042 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5043 /* this inode is deleted */
5047 /* The only unlinked inodes we let through here have
5048 * valid i_mode and are being read by the orphan
5049 * recovery code: that's fine, we're about to complete
5050 * the process of deleting those. */
5052 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5053 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5054 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5055 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5057 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5058 inode->i_size = ext4_isize(raw_inode);
5059 ei->i_disksize = inode->i_size;
5061 ei->i_reserved_quota = 0;
5063 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5064 ei->i_block_group = iloc.block_group;
5065 ei->i_last_alloc_group = ~0;
5067 * NOTE! The in-memory inode i_data array is in little-endian order
5068 * even on big-endian machines: we do NOT byteswap the block numbers!
5070 for (block = 0; block < EXT4_N_BLOCKS; block++)
5071 ei->i_data[block] = raw_inode->i_block[block];
5072 INIT_LIST_HEAD(&ei->i_orphan);
5075 * Set transaction id's of transactions that have to be committed
5076 * to finish f[data]sync. We set them to currently running transaction
5077 * as we cannot be sure that the inode or some of its metadata isn't
5078 * part of the transaction - the inode could have been reclaimed and
5079 * now it is reread from disk.
5082 transaction_t *transaction;
5085 read_lock(&journal->j_state_lock);
5086 if (journal->j_running_transaction)
5087 transaction = journal->j_running_transaction;
5089 transaction = journal->j_committing_transaction;
5091 tid = transaction->t_tid;
5093 tid = journal->j_commit_sequence;
5094 read_unlock(&journal->j_state_lock);
5095 ei->i_sync_tid = tid;
5096 ei->i_datasync_tid = tid;
5099 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5100 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5101 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5102 EXT4_INODE_SIZE(inode->i_sb)) {
5106 if (ei->i_extra_isize == 0) {
5107 /* The extra space is currently unused. Use it. */
5108 ei->i_extra_isize = sizeof(struct ext4_inode) -
5109 EXT4_GOOD_OLD_INODE_SIZE;
5111 __le32 *magic = (void *)raw_inode +
5112 EXT4_GOOD_OLD_INODE_SIZE +
5114 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5115 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5118 ei->i_extra_isize = 0;
5120 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5121 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5122 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5123 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5125 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5126 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5127 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5129 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5133 if (ei->i_file_acl &&
5134 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5135 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5139 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5140 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5141 (S_ISLNK(inode->i_mode) &&
5142 !ext4_inode_is_fast_symlink(inode)))
5143 /* Validate extent which is part of inode */
5144 ret = ext4_ext_check_inode(inode);
5145 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5146 (S_ISLNK(inode->i_mode) &&
5147 !ext4_inode_is_fast_symlink(inode))) {
5148 /* Validate block references which are part of inode */
5149 ret = ext4_check_inode_blockref(inode);
5154 if (S_ISREG(inode->i_mode)) {
5155 inode->i_op = &ext4_file_inode_operations;
5156 inode->i_fop = &ext4_file_operations;
5157 ext4_set_aops(inode);
5158 } else if (S_ISDIR(inode->i_mode)) {
5159 inode->i_op = &ext4_dir_inode_operations;
5160 inode->i_fop = &ext4_dir_operations;
5161 } else if (S_ISLNK(inode->i_mode)) {
5162 if (ext4_inode_is_fast_symlink(inode)) {
5163 inode->i_op = &ext4_fast_symlink_inode_operations;
5164 nd_terminate_link(ei->i_data, inode->i_size,
5165 sizeof(ei->i_data) - 1);
5167 inode->i_op = &ext4_symlink_inode_operations;
5168 ext4_set_aops(inode);
5170 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5171 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5172 inode->i_op = &ext4_special_inode_operations;
5173 if (raw_inode->i_block[0])
5174 init_special_inode(inode, inode->i_mode,
5175 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5177 init_special_inode(inode, inode->i_mode,
5178 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5181 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5185 ext4_set_inode_flags(inode);
5186 unlock_new_inode(inode);
5192 return ERR_PTR(ret);
5195 static int ext4_inode_blocks_set(handle_t *handle,
5196 struct ext4_inode *raw_inode,
5197 struct ext4_inode_info *ei)
5199 struct inode *inode = &(ei->vfs_inode);
5200 u64 i_blocks = inode->i_blocks;
5201 struct super_block *sb = inode->i_sb;
5203 if (i_blocks <= ~0U) {
5205 * i_blocks can be represnted in a 32 bit variable
5206 * as multiple of 512 bytes
5208 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5209 raw_inode->i_blocks_high = 0;
5210 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5213 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5216 if (i_blocks <= 0xffffffffffffULL) {
5218 * i_blocks can be represented in a 48 bit variable
5219 * as multiple of 512 bytes
5221 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5222 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5223 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5225 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5226 /* i_block is stored in file system block size */
5227 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5228 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5229 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5235 * Post the struct inode info into an on-disk inode location in the
5236 * buffer-cache. This gobbles the caller's reference to the
5237 * buffer_head in the inode location struct.
5239 * The caller must have write access to iloc->bh.
5241 static int ext4_do_update_inode(handle_t *handle,
5242 struct inode *inode,
5243 struct ext4_iloc *iloc)
5245 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5246 struct ext4_inode_info *ei = EXT4_I(inode);
5247 struct buffer_head *bh = iloc->bh;
5248 int err = 0, rc, block;
5250 /* For fields not not tracking in the in-memory inode,
5251 * initialise them to zero for new inodes. */
5252 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5253 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5255 ext4_get_inode_flags(ei);
5256 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5257 if (!(test_opt(inode->i_sb, NO_UID32))) {
5258 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5259 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5261 * Fix up interoperability with old kernels. Otherwise, old inodes get
5262 * re-used with the upper 16 bits of the uid/gid intact
5265 raw_inode->i_uid_high =
5266 cpu_to_le16(high_16_bits(inode->i_uid));
5267 raw_inode->i_gid_high =
5268 cpu_to_le16(high_16_bits(inode->i_gid));
5270 raw_inode->i_uid_high = 0;
5271 raw_inode->i_gid_high = 0;
5274 raw_inode->i_uid_low =
5275 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5276 raw_inode->i_gid_low =
5277 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5278 raw_inode->i_uid_high = 0;
5279 raw_inode->i_gid_high = 0;
5281 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5283 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5284 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5285 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5286 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5288 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5290 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5291 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5292 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5293 cpu_to_le32(EXT4_OS_HURD))
5294 raw_inode->i_file_acl_high =
5295 cpu_to_le16(ei->i_file_acl >> 32);
5296 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5297 ext4_isize_set(raw_inode, ei->i_disksize);
5298 if (ei->i_disksize > 0x7fffffffULL) {
5299 struct super_block *sb = inode->i_sb;
5300 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5301 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5302 EXT4_SB(sb)->s_es->s_rev_level ==
5303 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5304 /* If this is the first large file
5305 * created, add a flag to the superblock.
5307 err = ext4_journal_get_write_access(handle,
5308 EXT4_SB(sb)->s_sbh);
5311 ext4_update_dynamic_rev(sb);
5312 EXT4_SET_RO_COMPAT_FEATURE(sb,
5313 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5315 ext4_handle_sync(handle);
5316 err = ext4_handle_dirty_metadata(handle, NULL,
5317 EXT4_SB(sb)->s_sbh);
5320 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5321 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5322 if (old_valid_dev(inode->i_rdev)) {
5323 raw_inode->i_block[0] =
5324 cpu_to_le32(old_encode_dev(inode->i_rdev));
5325 raw_inode->i_block[1] = 0;
5327 raw_inode->i_block[0] = 0;
5328 raw_inode->i_block[1] =
5329 cpu_to_le32(new_encode_dev(inode->i_rdev));
5330 raw_inode->i_block[2] = 0;
5333 for (block = 0; block < EXT4_N_BLOCKS; block++)
5334 raw_inode->i_block[block] = ei->i_data[block];
5336 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5337 if (ei->i_extra_isize) {
5338 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5339 raw_inode->i_version_hi =
5340 cpu_to_le32(inode->i_version >> 32);
5341 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5344 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5345 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5348 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5350 ext4_update_inode_fsync_trans(handle, inode, 0);
5353 ext4_std_error(inode->i_sb, err);
5358 * ext4_write_inode()
5360 * We are called from a few places:
5362 * - Within generic_file_write() for O_SYNC files.
5363 * Here, there will be no transaction running. We wait for any running
5364 * trasnaction to commit.
5366 * - Within sys_sync(), kupdate and such.
5367 * We wait on commit, if tol to.
5369 * - Within prune_icache() (PF_MEMALLOC == true)
5370 * Here we simply return. We can't afford to block kswapd on the
5373 * In all cases it is actually safe for us to return without doing anything,
5374 * because the inode has been copied into a raw inode buffer in
5375 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5378 * Note that we are absolutely dependent upon all inode dirtiers doing the
5379 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5380 * which we are interested.
5382 * It would be a bug for them to not do this. The code:
5384 * mark_inode_dirty(inode)
5386 * inode->i_size = expr;
5388 * is in error because a kswapd-driven write_inode() could occur while
5389 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5390 * will no longer be on the superblock's dirty inode list.
5392 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5396 if (current->flags & PF_MEMALLOC)
5399 if (EXT4_SB(inode->i_sb)->s_journal) {
5400 if (ext4_journal_current_handle()) {
5401 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5406 if (wbc->sync_mode != WB_SYNC_ALL)
5409 err = ext4_force_commit(inode->i_sb);
5411 struct ext4_iloc iloc;
5413 err = __ext4_get_inode_loc(inode, &iloc, 0);
5416 if (wbc->sync_mode == WB_SYNC_ALL)
5417 sync_dirty_buffer(iloc.bh);
5418 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5419 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5420 "IO error syncing inode");
5431 * Called from notify_change.
5433 * We want to trap VFS attempts to truncate the file as soon as
5434 * possible. In particular, we want to make sure that when the VFS
5435 * shrinks i_size, we put the inode on the orphan list and modify
5436 * i_disksize immediately, so that during the subsequent flushing of
5437 * dirty pages and freeing of disk blocks, we can guarantee that any
5438 * commit will leave the blocks being flushed in an unused state on
5439 * disk. (On recovery, the inode will get truncated and the blocks will
5440 * be freed, so we have a strong guarantee that no future commit will
5441 * leave these blocks visible to the user.)
5443 * Another thing we have to assure is that if we are in ordered mode
5444 * and inode is still attached to the committing transaction, we must
5445 * we start writeout of all the dirty pages which are being truncated.
5446 * This way we are sure that all the data written in the previous
5447 * transaction are already on disk (truncate waits for pages under
5450 * Called with inode->i_mutex down.
5452 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5454 struct inode *inode = dentry->d_inode;
5456 const unsigned int ia_valid = attr->ia_valid;
5458 error = inode_change_ok(inode, attr);
5462 if (is_quota_modification(inode, attr))
5463 dquot_initialize(inode);
5464 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5465 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5468 /* (user+group)*(old+new) structure, inode write (sb,
5469 * inode block, ? - but truncate inode update has it) */
5470 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5471 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5472 if (IS_ERR(handle)) {
5473 error = PTR_ERR(handle);
5476 error = dquot_transfer(inode, attr);
5478 ext4_journal_stop(handle);
5481 /* Update corresponding info in inode so that everything is in
5482 * one transaction */
5483 if (attr->ia_valid & ATTR_UID)
5484 inode->i_uid = attr->ia_uid;
5485 if (attr->ia_valid & ATTR_GID)
5486 inode->i_gid = attr->ia_gid;
5487 error = ext4_mark_inode_dirty(handle, inode);
5488 ext4_journal_stop(handle);
5491 if (attr->ia_valid & ATTR_SIZE) {
5492 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5493 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5495 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5500 if (S_ISREG(inode->i_mode) &&
5501 attr->ia_valid & ATTR_SIZE &&
5502 (attr->ia_size < inode->i_size ||
5503 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5506 handle = ext4_journal_start(inode, 3);
5507 if (IS_ERR(handle)) {
5508 error = PTR_ERR(handle);
5512 error = ext4_orphan_add(handle, inode);
5513 EXT4_I(inode)->i_disksize = attr->ia_size;
5514 rc = ext4_mark_inode_dirty(handle, inode);
5517 ext4_journal_stop(handle);
5519 if (ext4_should_order_data(inode)) {
5520 error = ext4_begin_ordered_truncate(inode,
5523 /* Do as much error cleanup as possible */
5524 handle = ext4_journal_start(inode, 3);
5525 if (IS_ERR(handle)) {
5526 ext4_orphan_del(NULL, inode);
5529 ext4_orphan_del(handle, inode);
5530 ext4_journal_stop(handle);
5534 /* ext4_truncate will clear the flag */
5535 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5536 ext4_truncate(inode);
5539 rc = inode_setattr(inode, attr);
5541 /* If inode_setattr's call to ext4_truncate failed to get a
5542 * transaction handle at all, we need to clean up the in-core
5543 * orphan list manually. */
5545 ext4_orphan_del(NULL, inode);
5547 if (!rc && (ia_valid & ATTR_MODE))
5548 rc = ext4_acl_chmod(inode);
5551 ext4_std_error(inode->i_sb, error);
5557 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5560 struct inode *inode;
5561 unsigned long delalloc_blocks;
5563 inode = dentry->d_inode;
5564 generic_fillattr(inode, stat);
5567 * We can't update i_blocks if the block allocation is delayed
5568 * otherwise in the case of system crash before the real block
5569 * allocation is done, we will have i_blocks inconsistent with
5570 * on-disk file blocks.
5571 * We always keep i_blocks updated together with real
5572 * allocation. But to not confuse with user, stat
5573 * will return the blocks that include the delayed allocation
5574 * blocks for this file.
5576 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5577 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5578 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5580 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5584 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5589 /* if nrblocks are contiguous */
5592 * With N contiguous data blocks, it need at most
5593 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5594 * 2 dindirect blocks
5597 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5598 return indirects + 3;
5601 * if nrblocks are not contiguous, worse case, each block touch
5602 * a indirect block, and each indirect block touch a double indirect
5603 * block, plus a triple indirect block
5605 indirects = nrblocks * 2 + 1;
5609 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5611 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5612 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5613 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5617 * Account for index blocks, block groups bitmaps and block group
5618 * descriptor blocks if modify datablocks and index blocks
5619 * worse case, the indexs blocks spread over different block groups
5621 * If datablocks are discontiguous, they are possible to spread over
5622 * different block groups too. If they are contiuguous, with flexbg,
5623 * they could still across block group boundary.
5625 * Also account for superblock, inode, quota and xattr blocks
5627 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5629 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5635 * How many index blocks need to touch to modify nrblocks?
5636 * The "Chunk" flag indicating whether the nrblocks is
5637 * physically contiguous on disk
5639 * For Direct IO and fallocate, they calls get_block to allocate
5640 * one single extent at a time, so they could set the "Chunk" flag
5642 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5647 * Now let's see how many group bitmaps and group descriptors need
5657 if (groups > ngroups)
5659 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5660 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5662 /* bitmaps and block group descriptor blocks */
5663 ret += groups + gdpblocks;
5665 /* Blocks for super block, inode, quota and xattr blocks */
5666 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5672 * Calulate the total number of credits to reserve to fit
5673 * the modification of a single pages into a single transaction,
5674 * which may include multiple chunks of block allocations.
5676 * This could be called via ext4_write_begin()
5678 * We need to consider the worse case, when
5679 * one new block per extent.
5681 int ext4_writepage_trans_blocks(struct inode *inode)
5683 int bpp = ext4_journal_blocks_per_page(inode);
5686 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5688 /* Account for data blocks for journalled mode */
5689 if (ext4_should_journal_data(inode))
5695 * Calculate the journal credits for a chunk of data modification.
5697 * This is called from DIO, fallocate or whoever calling
5698 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5700 * journal buffers for data blocks are not included here, as DIO
5701 * and fallocate do no need to journal data buffers.
5703 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5705 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5709 * The caller must have previously called ext4_reserve_inode_write().
5710 * Give this, we know that the caller already has write access to iloc->bh.
5712 int ext4_mark_iloc_dirty(handle_t *handle,
5713 struct inode *inode, struct ext4_iloc *iloc)
5717 if (test_opt(inode->i_sb, I_VERSION))
5718 inode_inc_iversion(inode);
5720 /* the do_update_inode consumes one bh->b_count */
5723 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5724 err = ext4_do_update_inode(handle, inode, iloc);
5730 * On success, We end up with an outstanding reference count against
5731 * iloc->bh. This _must_ be cleaned up later.
5735 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5736 struct ext4_iloc *iloc)
5740 err = ext4_get_inode_loc(inode, iloc);
5742 BUFFER_TRACE(iloc->bh, "get_write_access");
5743 err = ext4_journal_get_write_access(handle, iloc->bh);
5749 ext4_std_error(inode->i_sb, err);
5754 * Expand an inode by new_extra_isize bytes.
5755 * Returns 0 on success or negative error number on failure.
5757 static int ext4_expand_extra_isize(struct inode *inode,
5758 unsigned int new_extra_isize,
5759 struct ext4_iloc iloc,
5762 struct ext4_inode *raw_inode;
5763 struct ext4_xattr_ibody_header *header;
5765 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5768 raw_inode = ext4_raw_inode(&iloc);
5770 header = IHDR(inode, raw_inode);
5772 /* No extended attributes present */
5773 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5774 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5775 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5777 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5781 /* try to expand with EAs present */
5782 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5787 * What we do here is to mark the in-core inode as clean with respect to inode
5788 * dirtiness (it may still be data-dirty).
5789 * This means that the in-core inode may be reaped by prune_icache
5790 * without having to perform any I/O. This is a very good thing,
5791 * because *any* task may call prune_icache - even ones which
5792 * have a transaction open against a different journal.
5794 * Is this cheating? Not really. Sure, we haven't written the
5795 * inode out, but prune_icache isn't a user-visible syncing function.
5796 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5797 * we start and wait on commits.
5799 * Is this efficient/effective? Well, we're being nice to the system
5800 * by cleaning up our inodes proactively so they can be reaped
5801 * without I/O. But we are potentially leaving up to five seconds'
5802 * worth of inodes floating about which prune_icache wants us to
5803 * write out. One way to fix that would be to get prune_icache()
5804 * to do a write_super() to free up some memory. It has the desired
5807 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5809 struct ext4_iloc iloc;
5810 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5811 static unsigned int mnt_count;
5815 err = ext4_reserve_inode_write(handle, inode, &iloc);
5816 if (ext4_handle_valid(handle) &&
5817 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5818 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5820 * We need extra buffer credits since we may write into EA block
5821 * with this same handle. If journal_extend fails, then it will
5822 * only result in a minor loss of functionality for that inode.
5823 * If this is felt to be critical, then e2fsck should be run to
5824 * force a large enough s_min_extra_isize.
5826 if ((jbd2_journal_extend(handle,
5827 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5828 ret = ext4_expand_extra_isize(inode,
5829 sbi->s_want_extra_isize,
5832 ext4_set_inode_state(inode,
5833 EXT4_STATE_NO_EXPAND);
5835 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5836 ext4_warning(inode->i_sb,
5837 "Unable to expand inode %lu. Delete"
5838 " some EAs or run e2fsck.",
5841 le16_to_cpu(sbi->s_es->s_mnt_count);
5847 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5852 * ext4_dirty_inode() is called from __mark_inode_dirty()
5854 * We're really interested in the case where a file is being extended.
5855 * i_size has been changed by generic_commit_write() and we thus need
5856 * to include the updated inode in the current transaction.
5858 * Also, dquot_alloc_block() will always dirty the inode when blocks
5859 * are allocated to the file.
5861 * If the inode is marked synchronous, we don't honour that here - doing
5862 * so would cause a commit on atime updates, which we don't bother doing.
5863 * We handle synchronous inodes at the highest possible level.
5865 void ext4_dirty_inode(struct inode *inode)
5869 handle = ext4_journal_start(inode, 2);
5873 ext4_mark_inode_dirty(handle, inode);
5875 ext4_journal_stop(handle);
5882 * Bind an inode's backing buffer_head into this transaction, to prevent
5883 * it from being flushed to disk early. Unlike
5884 * ext4_reserve_inode_write, this leaves behind no bh reference and
5885 * returns no iloc structure, so the caller needs to repeat the iloc
5886 * lookup to mark the inode dirty later.
5888 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5890 struct ext4_iloc iloc;
5894 err = ext4_get_inode_loc(inode, &iloc);
5896 BUFFER_TRACE(iloc.bh, "get_write_access");
5897 err = jbd2_journal_get_write_access(handle, iloc.bh);
5899 err = ext4_handle_dirty_metadata(handle,
5905 ext4_std_error(inode->i_sb, err);
5910 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5917 * We have to be very careful here: changing a data block's
5918 * journaling status dynamically is dangerous. If we write a
5919 * data block to the journal, change the status and then delete
5920 * that block, we risk forgetting to revoke the old log record
5921 * from the journal and so a subsequent replay can corrupt data.
5922 * So, first we make sure that the journal is empty and that
5923 * nobody is changing anything.
5926 journal = EXT4_JOURNAL(inode);
5929 if (is_journal_aborted(journal))
5932 jbd2_journal_lock_updates(journal);
5933 jbd2_journal_flush(journal);
5936 * OK, there are no updates running now, and all cached data is
5937 * synced to disk. We are now in a completely consistent state
5938 * which doesn't have anything in the journal, and we know that
5939 * no filesystem updates are running, so it is safe to modify
5940 * the inode's in-core data-journaling state flag now.
5944 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5946 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5947 ext4_set_aops(inode);
5949 jbd2_journal_unlock_updates(journal);
5951 /* Finally we can mark the inode as dirty. */
5953 handle = ext4_journal_start(inode, 1);
5955 return PTR_ERR(handle);
5957 err = ext4_mark_inode_dirty(handle, inode);
5958 ext4_handle_sync(handle);
5959 ext4_journal_stop(handle);
5960 ext4_std_error(inode->i_sb, err);
5965 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5967 return !buffer_mapped(bh);
5970 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5972 struct page *page = vmf->page;
5977 struct file *file = vma->vm_file;
5978 struct inode *inode = file->f_path.dentry->d_inode;
5979 struct address_space *mapping = inode->i_mapping;
5982 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5983 * get i_mutex because we are already holding mmap_sem.
5985 down_read(&inode->i_alloc_sem);
5986 size = i_size_read(inode);
5987 if (page->mapping != mapping || size <= page_offset(page)
5988 || !PageUptodate(page)) {
5989 /* page got truncated from under us? */
5993 if (PageMappedToDisk(page))
5996 if (page->index == size >> PAGE_CACHE_SHIFT)
5997 len = size & ~PAGE_CACHE_MASK;
5999 len = PAGE_CACHE_SIZE;
6003 * return if we have all the buffers mapped. This avoid
6004 * the need to call write_begin/write_end which does a
6005 * journal_start/journal_stop which can block and take
6008 if (page_has_buffers(page)) {
6009 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
6010 ext4_bh_unmapped)) {
6017 * OK, we need to fill the hole... Do write_begin write_end
6018 * to do block allocation/reservation.We are not holding
6019 * inode.i__mutex here. That allow * parallel write_begin,
6020 * write_end call. lock_page prevent this from happening
6021 * on the same page though
6023 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
6024 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
6027 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
6028 len, len, page, fsdata);
6034 ret = VM_FAULT_SIGBUS;
6035 up_read(&inode->i_alloc_sem);