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);
229 * Kill off the orphan record which ext4_truncate created.
230 * AKPM: I think this can be inside the above `if'.
231 * Note that ext4_orphan_del() has to be able to cope with the
232 * deletion of a non-existent orphan - this is because we don't
233 * know if ext4_truncate() actually created an orphan record.
234 * (Well, we could do this if we need to, but heck - it works)
236 ext4_orphan_del(handle, inode);
237 EXT4_I(inode)->i_dtime = get_seconds();
240 * One subtle ordering requirement: if anything has gone wrong
241 * (transaction abort, IO errors, whatever), then we can still
242 * do these next steps (the fs will already have been marked as
243 * having errors), but we can't free the inode if the mark_dirty
246 if (ext4_mark_inode_dirty(handle, inode))
247 /* If that failed, just do the required in-core inode clear. */
250 ext4_free_inode(handle, inode);
251 ext4_journal_stop(handle);
254 clear_inode(inode); /* We must guarantee clearing of inode... */
260 struct buffer_head *bh;
263 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
265 p->key = *(p->p = v);
270 * ext4_block_to_path - parse the block number into array of offsets
271 * @inode: inode in question (we are only interested in its superblock)
272 * @i_block: block number to be parsed
273 * @offsets: array to store the offsets in
274 * @boundary: set this non-zero if the referred-to block is likely to be
275 * followed (on disk) by an indirect block.
277 * To store the locations of file's data ext4 uses a data structure common
278 * for UNIX filesystems - tree of pointers anchored in the inode, with
279 * data blocks at leaves and indirect blocks in intermediate nodes.
280 * This function translates the block number into path in that tree -
281 * return value is the path length and @offsets[n] is the offset of
282 * pointer to (n+1)th node in the nth one. If @block is out of range
283 * (negative or too large) warning is printed and zero returned.
285 * Note: function doesn't find node addresses, so no IO is needed. All
286 * we need to know is the capacity of indirect blocks (taken from the
291 * Portability note: the last comparison (check that we fit into triple
292 * indirect block) is spelled differently, because otherwise on an
293 * architecture with 32-bit longs and 8Kb pages we might get into trouble
294 * if our filesystem had 8Kb blocks. We might use long long, but that would
295 * kill us on x86. Oh, well, at least the sign propagation does not matter -
296 * i_block would have to be negative in the very beginning, so we would not
300 static int ext4_block_to_path(struct inode *inode,
302 ext4_lblk_t offsets[4], int *boundary)
304 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
305 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
306 const long direct_blocks = EXT4_NDIR_BLOCKS,
307 indirect_blocks = ptrs,
308 double_blocks = (1 << (ptrs_bits * 2));
312 if (i_block < direct_blocks) {
313 offsets[n++] = i_block;
314 final = direct_blocks;
315 } else if ((i_block -= direct_blocks) < indirect_blocks) {
316 offsets[n++] = EXT4_IND_BLOCK;
317 offsets[n++] = i_block;
319 } else if ((i_block -= indirect_blocks) < double_blocks) {
320 offsets[n++] = EXT4_DIND_BLOCK;
321 offsets[n++] = i_block >> ptrs_bits;
322 offsets[n++] = i_block & (ptrs - 1);
324 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
325 offsets[n++] = EXT4_TIND_BLOCK;
326 offsets[n++] = i_block >> (ptrs_bits * 2);
327 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
328 offsets[n++] = i_block & (ptrs - 1);
331 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
332 i_block + direct_blocks +
333 indirect_blocks + double_blocks, inode->i_ino);
336 *boundary = final - 1 - (i_block & (ptrs - 1));
340 static int __ext4_check_blockref(const char *function, unsigned int line,
342 __le32 *p, unsigned int max)
347 while (bref < p+max) {
348 blk = le32_to_cpu(*bref++);
350 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
352 ext4_error_inode(inode, function, line, blk,
361 #define ext4_check_indirect_blockref(inode, bh) \
362 __ext4_check_blockref(__func__, __LINE__, inode, \
363 (__le32 *)(bh)->b_data, \
364 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
366 #define ext4_check_inode_blockref(inode) \
367 __ext4_check_blockref(__func__, __LINE__, inode, \
368 EXT4_I(inode)->i_data, \
372 * ext4_get_branch - read the chain of indirect blocks leading to data
373 * @inode: inode in question
374 * @depth: depth of the chain (1 - direct pointer, etc.)
375 * @offsets: offsets of pointers in inode/indirect blocks
376 * @chain: place to store the result
377 * @err: here we store the error value
379 * Function fills the array of triples <key, p, bh> and returns %NULL
380 * if everything went OK or the pointer to the last filled triple
381 * (incomplete one) otherwise. Upon the return chain[i].key contains
382 * the number of (i+1)-th block in the chain (as it is stored in memory,
383 * i.e. little-endian 32-bit), chain[i].p contains the address of that
384 * number (it points into struct inode for i==0 and into the bh->b_data
385 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
386 * block for i>0 and NULL for i==0. In other words, it holds the block
387 * numbers of the chain, addresses they were taken from (and where we can
388 * verify that chain did not change) and buffer_heads hosting these
391 * Function stops when it stumbles upon zero pointer (absent block)
392 * (pointer to last triple returned, *@err == 0)
393 * or when it gets an IO error reading an indirect block
394 * (ditto, *@err == -EIO)
395 * or when it reads all @depth-1 indirect blocks successfully and finds
396 * the whole chain, all way to the data (returns %NULL, *err == 0).
398 * Need to be called with
399 * down_read(&EXT4_I(inode)->i_data_sem)
401 static Indirect *ext4_get_branch(struct inode *inode, int depth,
402 ext4_lblk_t *offsets,
403 Indirect chain[4], int *err)
405 struct super_block *sb = inode->i_sb;
407 struct buffer_head *bh;
410 /* i_data is not going away, no lock needed */
411 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
415 bh = sb_getblk(sb, le32_to_cpu(p->key));
419 if (!bh_uptodate_or_lock(bh)) {
420 if (bh_submit_read(bh) < 0) {
424 /* validate block references */
425 if (ext4_check_indirect_blockref(inode, bh)) {
431 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
445 * ext4_find_near - find a place for allocation with sufficient locality
447 * @ind: descriptor of indirect block.
449 * This function returns the preferred place for block allocation.
450 * It is used when heuristic for sequential allocation fails.
452 * + if there is a block to the left of our position - allocate near it.
453 * + if pointer will live in indirect block - allocate near that block.
454 * + if pointer will live in inode - allocate in the same
457 * In the latter case we colour the starting block by the callers PID to
458 * prevent it from clashing with concurrent allocations for a different inode
459 * in the same block group. The PID is used here so that functionally related
460 * files will be close-by on-disk.
462 * Caller must make sure that @ind is valid and will stay that way.
464 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
466 struct ext4_inode_info *ei = EXT4_I(inode);
467 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
469 ext4_fsblk_t bg_start;
470 ext4_fsblk_t last_block;
471 ext4_grpblk_t colour;
472 ext4_group_t block_group;
473 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
475 /* Try to find previous block */
476 for (p = ind->p - 1; p >= start; p--) {
478 return le32_to_cpu(*p);
481 /* No such thing, so let's try location of indirect block */
483 return ind->bh->b_blocknr;
486 * It is going to be referred to from the inode itself? OK, just put it
487 * into the same cylinder group then.
489 block_group = ei->i_block_group;
490 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
491 block_group &= ~(flex_size-1);
492 if (S_ISREG(inode->i_mode))
495 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
496 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
499 * If we are doing delayed allocation, we don't need take
500 * colour into account.
502 if (test_opt(inode->i_sb, DELALLOC))
505 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
506 colour = (current->pid % 16) *
507 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
509 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
510 return bg_start + colour;
514 * ext4_find_goal - find a preferred place for allocation.
516 * @block: block we want
517 * @partial: pointer to the last triple within a chain
519 * Normally this function find the preferred place for block allocation,
521 * Because this is only used for non-extent files, we limit the block nr
524 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
530 * XXX need to get goal block from mballoc's data structures
533 goal = ext4_find_near(inode, partial);
534 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
539 * ext4_blks_to_allocate: Look up the block map and count the number
540 * of direct blocks need to be allocated for the given branch.
542 * @branch: chain of indirect blocks
543 * @k: number of blocks need for indirect blocks
544 * @blks: number of data blocks to be mapped.
545 * @blocks_to_boundary: the offset in the indirect block
547 * return the total number of blocks to be allocate, including the
548 * direct and indirect blocks.
550 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
551 int blocks_to_boundary)
553 unsigned int count = 0;
556 * Simple case, [t,d]Indirect block(s) has not allocated yet
557 * then it's clear blocks on that path have not allocated
560 /* right now we don't handle cross boundary allocation */
561 if (blks < blocks_to_boundary + 1)
564 count += blocks_to_boundary + 1;
569 while (count < blks && count <= blocks_to_boundary &&
570 le32_to_cpu(*(branch[0].p + count)) == 0) {
577 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
578 * @indirect_blks: the number of blocks need to allocate for indirect
581 * @new_blocks: on return it will store the new block numbers for
582 * the indirect blocks(if needed) and the first direct block,
583 * @blks: on return it will store the total number of allocated
586 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
587 ext4_lblk_t iblock, ext4_fsblk_t goal,
588 int indirect_blks, int blks,
589 ext4_fsblk_t new_blocks[4], int *err)
591 struct ext4_allocation_request ar;
593 unsigned long count = 0, blk_allocated = 0;
595 ext4_fsblk_t current_block = 0;
599 * Here we try to allocate the requested multiple blocks at once,
600 * on a best-effort basis.
601 * To build a branch, we should allocate blocks for
602 * the indirect blocks(if not allocated yet), and at least
603 * the first direct block of this branch. That's the
604 * minimum number of blocks need to allocate(required)
606 /* first we try to allocate the indirect blocks */
607 target = indirect_blks;
610 /* allocating blocks for indirect blocks and direct blocks */
611 current_block = ext4_new_meta_blocks(handle, inode,
616 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
617 EXT4_ERROR_INODE(inode,
618 "current_block %llu + count %lu > %d!",
619 current_block, count,
620 EXT4_MAX_BLOCK_FILE_PHYS);
626 /* allocate blocks for indirect blocks */
627 while (index < indirect_blks && count) {
628 new_blocks[index++] = current_block++;
633 * save the new block number
634 * for the first direct block
636 new_blocks[index] = current_block;
637 printk(KERN_INFO "%s returned more blocks than "
638 "requested\n", __func__);
644 target = blks - count ;
645 blk_allocated = count;
648 /* Now allocate data blocks */
649 memset(&ar, 0, sizeof(ar));
654 if (S_ISREG(inode->i_mode))
655 /* enable in-core preallocation only for regular files */
656 ar.flags = EXT4_MB_HINT_DATA;
658 current_block = ext4_mb_new_blocks(handle, &ar, err);
659 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
660 EXT4_ERROR_INODE(inode,
661 "current_block %llu + ar.len %d > %d!",
662 current_block, ar.len,
663 EXT4_MAX_BLOCK_FILE_PHYS);
668 if (*err && (target == blks)) {
670 * if the allocation failed and we didn't allocate
676 if (target == blks) {
678 * save the new block number
679 * for the first direct block
681 new_blocks[index] = current_block;
683 blk_allocated += ar.len;
686 /* total number of blocks allocated for direct blocks */
691 for (i = 0; i < index; i++)
692 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
697 * ext4_alloc_branch - allocate and set up a chain of blocks.
699 * @indirect_blks: number of allocated indirect blocks
700 * @blks: number of allocated direct blocks
701 * @offsets: offsets (in the blocks) to store the pointers to next.
702 * @branch: place to store the chain in.
704 * This function allocates blocks, zeroes out all but the last one,
705 * links them into chain and (if we are synchronous) writes them to disk.
706 * In other words, it prepares a branch that can be spliced onto the
707 * inode. It stores the information about that chain in the branch[], in
708 * the same format as ext4_get_branch() would do. We are calling it after
709 * we had read the existing part of chain and partial points to the last
710 * triple of that (one with zero ->key). Upon the exit we have the same
711 * picture as after the successful ext4_get_block(), except that in one
712 * place chain is disconnected - *branch->p is still zero (we did not
713 * set the last link), but branch->key contains the number that should
714 * be placed into *branch->p to fill that gap.
716 * If allocation fails we free all blocks we've allocated (and forget
717 * their buffer_heads) and return the error value the from failed
718 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
719 * as described above and return 0.
721 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
722 ext4_lblk_t iblock, int indirect_blks,
723 int *blks, ext4_fsblk_t goal,
724 ext4_lblk_t *offsets, Indirect *branch)
726 int blocksize = inode->i_sb->s_blocksize;
729 struct buffer_head *bh;
731 ext4_fsblk_t new_blocks[4];
732 ext4_fsblk_t current_block;
734 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
735 *blks, new_blocks, &err);
739 branch[0].key = cpu_to_le32(new_blocks[0]);
741 * metadata blocks and data blocks are allocated.
743 for (n = 1; n <= indirect_blks; n++) {
745 * Get buffer_head for parent block, zero it out
746 * and set the pointer to new one, then send
749 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
752 BUFFER_TRACE(bh, "call get_create_access");
753 err = ext4_journal_get_create_access(handle, bh);
755 /* Don't brelse(bh) here; it's done in
756 * ext4_journal_forget() below */
761 memset(bh->b_data, 0, blocksize);
762 branch[n].p = (__le32 *) bh->b_data + offsets[n];
763 branch[n].key = cpu_to_le32(new_blocks[n]);
764 *branch[n].p = branch[n].key;
765 if (n == indirect_blks) {
766 current_block = new_blocks[n];
768 * End of chain, update the last new metablock of
769 * the chain to point to the new allocated
770 * data blocks numbers
772 for (i = 1; i < num; i++)
773 *(branch[n].p + i) = cpu_to_le32(++current_block);
775 BUFFER_TRACE(bh, "marking uptodate");
776 set_buffer_uptodate(bh);
779 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
780 err = ext4_handle_dirty_metadata(handle, inode, bh);
787 /* Allocation failed, free what we already allocated */
788 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
789 for (i = 1; i <= n ; i++) {
791 * branch[i].bh is newly allocated, so there is no
792 * need to revoke the block, which is why we don't
793 * need to set EXT4_FREE_BLOCKS_METADATA.
795 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
796 EXT4_FREE_BLOCKS_FORGET);
798 for (i = n+1; i < indirect_blks; i++)
799 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
801 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
807 * ext4_splice_branch - splice the allocated branch onto inode.
809 * @block: (logical) number of block we are adding
810 * @chain: chain of indirect blocks (with a missing link - see
812 * @where: location of missing link
813 * @num: number of indirect blocks we are adding
814 * @blks: number of direct blocks we are adding
816 * This function fills the missing link and does all housekeeping needed in
817 * inode (->i_blocks, etc.). In case of success we end up with the full
818 * chain to new block and return 0.
820 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
821 ext4_lblk_t block, Indirect *where, int num,
826 ext4_fsblk_t current_block;
829 * If we're splicing into a [td]indirect block (as opposed to the
830 * inode) then we need to get write access to the [td]indirect block
834 BUFFER_TRACE(where->bh, "get_write_access");
835 err = ext4_journal_get_write_access(handle, where->bh);
841 *where->p = where->key;
844 * Update the host buffer_head or inode to point to more just allocated
845 * direct blocks blocks
847 if (num == 0 && blks > 1) {
848 current_block = le32_to_cpu(where->key) + 1;
849 for (i = 1; i < blks; i++)
850 *(where->p + i) = cpu_to_le32(current_block++);
853 /* We are done with atomic stuff, now do the rest of housekeeping */
854 /* had we spliced it onto indirect block? */
857 * If we spliced it onto an indirect block, we haven't
858 * altered the inode. Note however that if it is being spliced
859 * onto an indirect block at the very end of the file (the
860 * file is growing) then we *will* alter the inode to reflect
861 * the new i_size. But that is not done here - it is done in
862 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
864 jbd_debug(5, "splicing indirect only\n");
865 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
866 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
871 * OK, we spliced it into the inode itself on a direct block.
873 ext4_mark_inode_dirty(handle, inode);
874 jbd_debug(5, "splicing direct\n");
879 for (i = 1; i <= num; i++) {
881 * branch[i].bh is newly allocated, so there is no
882 * need to revoke the block, which is why we don't
883 * need to set EXT4_FREE_BLOCKS_METADATA.
885 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
886 EXT4_FREE_BLOCKS_FORGET);
888 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
895 * The ext4_ind_map_blocks() function handles non-extents inodes
896 * (i.e., using the traditional indirect/double-indirect i_blocks
897 * scheme) for ext4_map_blocks().
899 * Allocation strategy is simple: if we have to allocate something, we will
900 * have to go the whole way to leaf. So let's do it before attaching anything
901 * to tree, set linkage between the newborn blocks, write them if sync is
902 * required, recheck the path, free and repeat if check fails, otherwise
903 * set the last missing link (that will protect us from any truncate-generated
904 * removals - all blocks on the path are immune now) and possibly force the
905 * write on the parent block.
906 * That has a nice additional property: no special recovery from the failed
907 * allocations is needed - we simply release blocks and do not touch anything
908 * reachable from inode.
910 * `handle' can be NULL if create == 0.
912 * return > 0, # of blocks mapped or allocated.
913 * return = 0, if plain lookup failed.
914 * return < 0, error case.
916 * The ext4_ind_get_blocks() function should be called with
917 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
918 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
919 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
922 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
923 struct ext4_map_blocks *map,
927 ext4_lblk_t offsets[4];
932 int blocks_to_boundary = 0;
935 ext4_fsblk_t first_block = 0;
937 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
938 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
939 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
940 &blocks_to_boundary);
945 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
947 /* Simplest case - block found, no allocation needed */
949 first_block = le32_to_cpu(chain[depth - 1].key);
952 while (count < map->m_len && count <= blocks_to_boundary) {
955 blk = le32_to_cpu(*(chain[depth-1].p + count));
957 if (blk == first_block + count)
965 /* Next simple case - plain lookup or failed read of indirect block */
966 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
970 * Okay, we need to do block allocation.
972 goal = ext4_find_goal(inode, map->m_lblk, partial);
974 /* the number of blocks need to allocate for [d,t]indirect blocks */
975 indirect_blks = (chain + depth) - partial - 1;
978 * Next look up the indirect map to count the totoal number of
979 * direct blocks to allocate for this branch.
981 count = ext4_blks_to_allocate(partial, indirect_blks,
982 map->m_len, blocks_to_boundary);
984 * Block out ext4_truncate while we alter the tree
986 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
988 offsets + (partial - chain), partial);
991 * The ext4_splice_branch call will free and forget any buffers
992 * on the new chain if there is a failure, but that risks using
993 * up transaction credits, especially for bitmaps where the
994 * credits cannot be returned. Can we handle this somehow? We
995 * may need to return -EAGAIN upwards in the worst case. --sct
998 err = ext4_splice_branch(handle, inode, map->m_lblk,
999 partial, indirect_blks, count);
1003 map->m_flags |= EXT4_MAP_NEW;
1005 ext4_update_inode_fsync_trans(handle, inode, 1);
1007 map->m_flags |= EXT4_MAP_MAPPED;
1008 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1010 if (count > blocks_to_boundary)
1011 map->m_flags |= EXT4_MAP_BOUNDARY;
1013 /* Clean up and exit */
1014 partial = chain + depth - 1; /* the whole chain */
1016 while (partial > chain) {
1017 BUFFER_TRACE(partial->bh, "call brelse");
1018 brelse(partial->bh);
1026 qsize_t *ext4_get_reserved_space(struct inode *inode)
1028 return &EXT4_I(inode)->i_reserved_quota;
1033 * Calculate the number of metadata blocks need to reserve
1034 * to allocate a new block at @lblocks for non extent file based file
1036 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1039 struct ext4_inode_info *ei = EXT4_I(inode);
1040 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1043 if (lblock < EXT4_NDIR_BLOCKS)
1046 lblock -= EXT4_NDIR_BLOCKS;
1048 if (ei->i_da_metadata_calc_len &&
1049 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1050 ei->i_da_metadata_calc_len++;
1053 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1054 ei->i_da_metadata_calc_len = 1;
1055 blk_bits = order_base_2(lblock);
1056 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1060 * Calculate the number of metadata blocks need to reserve
1061 * to allocate a block located at @lblock
1063 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1065 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1066 return ext4_ext_calc_metadata_amount(inode, lblock);
1068 return ext4_indirect_calc_metadata_amount(inode, lblock);
1072 * Called with i_data_sem down, which is important since we can call
1073 * ext4_discard_preallocations() from here.
1075 void ext4_da_update_reserve_space(struct inode *inode,
1076 int used, int quota_claim)
1078 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1079 struct ext4_inode_info *ei = EXT4_I(inode);
1081 spin_lock(&ei->i_block_reservation_lock);
1082 trace_ext4_da_update_reserve_space(inode, used);
1083 if (unlikely(used > ei->i_reserved_data_blocks)) {
1084 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1085 "with only %d reserved data blocks\n",
1086 __func__, inode->i_ino, used,
1087 ei->i_reserved_data_blocks);
1089 used = ei->i_reserved_data_blocks;
1092 /* Update per-inode reservations */
1093 ei->i_reserved_data_blocks -= used;
1094 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1095 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1096 used + ei->i_allocated_meta_blocks);
1097 ei->i_allocated_meta_blocks = 0;
1099 if (ei->i_reserved_data_blocks == 0) {
1101 * We can release all of the reserved metadata blocks
1102 * only when we have written all of the delayed
1103 * allocation blocks.
1105 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1106 ei->i_reserved_meta_blocks);
1107 ei->i_reserved_meta_blocks = 0;
1108 ei->i_da_metadata_calc_len = 0;
1110 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1112 /* Update quota subsystem for data blocks */
1114 dquot_claim_block(inode, used);
1117 * We did fallocate with an offset that is already delayed
1118 * allocated. So on delayed allocated writeback we should
1119 * not re-claim the quota for fallocated blocks.
1121 dquot_release_reservation_block(inode, used);
1125 * If we have done all the pending block allocations and if
1126 * there aren't any writers on the inode, we can discard the
1127 * inode's preallocations.
1129 if ((ei->i_reserved_data_blocks == 0) &&
1130 (atomic_read(&inode->i_writecount) == 0))
1131 ext4_discard_preallocations(inode);
1134 static int __check_block_validity(struct inode *inode, const char *func,
1136 struct ext4_map_blocks *map)
1138 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1140 ext4_error_inode(inode, func, line, map->m_pblk,
1141 "lblock %lu mapped to illegal pblock "
1142 "(length %d)", (unsigned long) map->m_lblk,
1149 #define check_block_validity(inode, map) \
1150 __check_block_validity((inode), __func__, __LINE__, (map))
1153 * Return the number of contiguous dirty pages in a given inode
1154 * starting at page frame idx.
1156 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1157 unsigned int max_pages)
1159 struct address_space *mapping = inode->i_mapping;
1161 struct pagevec pvec;
1163 int i, nr_pages, done = 0;
1167 pagevec_init(&pvec, 0);
1170 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1171 PAGECACHE_TAG_DIRTY,
1172 (pgoff_t)PAGEVEC_SIZE);
1175 for (i = 0; i < nr_pages; i++) {
1176 struct page *page = pvec.pages[i];
1177 struct buffer_head *bh, *head;
1180 if (unlikely(page->mapping != mapping) ||
1182 PageWriteback(page) ||
1183 page->index != idx) {
1188 if (page_has_buffers(page)) {
1189 bh = head = page_buffers(page);
1191 if (!buffer_delay(bh) &&
1192 !buffer_unwritten(bh))
1194 bh = bh->b_this_page;
1195 } while (!done && (bh != head));
1202 if (num >= max_pages)
1205 pagevec_release(&pvec);
1211 * The ext4_map_blocks() function tries to look up the requested blocks,
1212 * and returns if the blocks are already mapped.
1214 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1215 * and store the allocated blocks in the result buffer head and mark it
1218 * If file type is extents based, it will call ext4_ext_map_blocks(),
1219 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1222 * On success, it returns the number of blocks being mapped or allocate.
1223 * if create==0 and the blocks are pre-allocated and uninitialized block,
1224 * the result buffer head is unmapped. If the create ==1, it will make sure
1225 * the buffer head is mapped.
1227 * It returns 0 if plain look up failed (blocks have not been allocated), in
1228 * that casem, buffer head is unmapped
1230 * It returns the error in case of allocation failure.
1232 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1233 struct ext4_map_blocks *map, int flags)
1238 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1239 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1240 (unsigned long) map->m_lblk);
1242 * Try to see if we can get the block without requesting a new
1243 * file system block.
1245 down_read((&EXT4_I(inode)->i_data_sem));
1246 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1247 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1249 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1251 up_read((&EXT4_I(inode)->i_data_sem));
1253 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1254 int ret = check_block_validity(inode, map);
1259 /* If it is only a block(s) look up */
1260 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1264 * Returns if the blocks have already allocated
1266 * Note that if blocks have been preallocated
1267 * ext4_ext_get_block() returns th create = 0
1268 * with buffer head unmapped.
1270 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1274 * When we call get_blocks without the create flag, the
1275 * BH_Unwritten flag could have gotten set if the blocks
1276 * requested were part of a uninitialized extent. We need to
1277 * clear this flag now that we are committed to convert all or
1278 * part of the uninitialized extent to be an initialized
1279 * extent. This is because we need to avoid the combination
1280 * of BH_Unwritten and BH_Mapped flags being simultaneously
1281 * set on the buffer_head.
1283 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1286 * New blocks allocate and/or writing to uninitialized extent
1287 * will possibly result in updating i_data, so we take
1288 * the write lock of i_data_sem, and call get_blocks()
1289 * with create == 1 flag.
1291 down_write((&EXT4_I(inode)->i_data_sem));
1294 * if the caller is from delayed allocation writeout path
1295 * we have already reserved fs blocks for allocation
1296 * let the underlying get_block() function know to
1297 * avoid double accounting
1299 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1300 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1302 * We need to check for EXT4 here because migrate
1303 * could have changed the inode type in between
1305 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1306 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1308 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1310 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1312 * We allocated new blocks which will result in
1313 * i_data's format changing. Force the migrate
1314 * to fail by clearing migrate flags
1316 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1320 * Update reserved blocks/metadata blocks after successful
1321 * block allocation which had been deferred till now. We don't
1322 * support fallocate for non extent files. So we can update
1323 * reserve space here.
1326 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1327 ext4_da_update_reserve_space(inode, retval, 1);
1329 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1330 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1332 up_write((&EXT4_I(inode)->i_data_sem));
1333 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1334 int ret = check_block_validity(inode, map);
1341 /* Maximum number of blocks we map for direct IO at once. */
1342 #define DIO_MAX_BLOCKS 4096
1344 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1345 struct buffer_head *bh, int flags)
1347 handle_t *handle = ext4_journal_current_handle();
1348 struct ext4_map_blocks map;
1349 int ret = 0, started = 0;
1352 map.m_lblk = iblock;
1353 map.m_len = bh->b_size >> inode->i_blkbits;
1355 if (flags && !handle) {
1356 /* Direct IO write... */
1357 if (map.m_len > DIO_MAX_BLOCKS)
1358 map.m_len = DIO_MAX_BLOCKS;
1359 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1360 handle = ext4_journal_start(inode, dio_credits);
1361 if (IS_ERR(handle)) {
1362 ret = PTR_ERR(handle);
1368 ret = ext4_map_blocks(handle, inode, &map, flags);
1370 map_bh(bh, inode->i_sb, map.m_pblk);
1371 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1372 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1376 ext4_journal_stop(handle);
1380 int ext4_get_block(struct inode *inode, sector_t iblock,
1381 struct buffer_head *bh, int create)
1383 return _ext4_get_block(inode, iblock, bh,
1384 create ? EXT4_GET_BLOCKS_CREATE : 0);
1388 * `handle' can be NULL if create is zero
1390 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1391 ext4_lblk_t block, int create, int *errp)
1393 struct ext4_map_blocks map;
1394 struct buffer_head *bh;
1397 J_ASSERT(handle != NULL || create == 0);
1401 err = ext4_map_blocks(handle, inode, &map,
1402 create ? EXT4_GET_BLOCKS_CREATE : 0);
1410 bh = sb_getblk(inode->i_sb, map.m_pblk);
1415 if (map.m_flags & EXT4_MAP_NEW) {
1416 J_ASSERT(create != 0);
1417 J_ASSERT(handle != NULL);
1420 * Now that we do not always journal data, we should
1421 * keep in mind whether this should always journal the
1422 * new buffer as metadata. For now, regular file
1423 * writes use ext4_get_block instead, so it's not a
1427 BUFFER_TRACE(bh, "call get_create_access");
1428 fatal = ext4_journal_get_create_access(handle, bh);
1429 if (!fatal && !buffer_uptodate(bh)) {
1430 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1431 set_buffer_uptodate(bh);
1434 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1435 err = ext4_handle_dirty_metadata(handle, inode, bh);
1439 BUFFER_TRACE(bh, "not a new buffer");
1449 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1450 ext4_lblk_t block, int create, int *err)
1452 struct buffer_head *bh;
1454 bh = ext4_getblk(handle, inode, block, create, err);
1457 if (buffer_uptodate(bh))
1459 ll_rw_block(READ_META, 1, &bh);
1461 if (buffer_uptodate(bh))
1468 static int walk_page_buffers(handle_t *handle,
1469 struct buffer_head *head,
1473 int (*fn)(handle_t *handle,
1474 struct buffer_head *bh))
1476 struct buffer_head *bh;
1477 unsigned block_start, block_end;
1478 unsigned blocksize = head->b_size;
1480 struct buffer_head *next;
1482 for (bh = head, block_start = 0;
1483 ret == 0 && (bh != head || !block_start);
1484 block_start = block_end, bh = next) {
1485 next = bh->b_this_page;
1486 block_end = block_start + blocksize;
1487 if (block_end <= from || block_start >= to) {
1488 if (partial && !buffer_uptodate(bh))
1492 err = (*fn)(handle, bh);
1500 * To preserve ordering, it is essential that the hole instantiation and
1501 * the data write be encapsulated in a single transaction. We cannot
1502 * close off a transaction and start a new one between the ext4_get_block()
1503 * and the commit_write(). So doing the jbd2_journal_start at the start of
1504 * prepare_write() is the right place.
1506 * Also, this function can nest inside ext4_writepage() ->
1507 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1508 * has generated enough buffer credits to do the whole page. So we won't
1509 * block on the journal in that case, which is good, because the caller may
1512 * By accident, ext4 can be reentered when a transaction is open via
1513 * quota file writes. If we were to commit the transaction while thus
1514 * reentered, there can be a deadlock - we would be holding a quota
1515 * lock, and the commit would never complete if another thread had a
1516 * transaction open and was blocking on the quota lock - a ranking
1519 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1520 * will _not_ run commit under these circumstances because handle->h_ref
1521 * is elevated. We'll still have enough credits for the tiny quotafile
1524 static int do_journal_get_write_access(handle_t *handle,
1525 struct buffer_head *bh)
1527 if (!buffer_mapped(bh) || buffer_freed(bh))
1529 return ext4_journal_get_write_access(handle, bh);
1533 * Truncate blocks that were not used by write. We have to truncate the
1534 * pagecache as well so that corresponding buffers get properly unmapped.
1536 static void ext4_truncate_failed_write(struct inode *inode)
1538 truncate_inode_pages(inode->i_mapping, inode->i_size);
1539 ext4_truncate(inode);
1542 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1543 struct buffer_head *bh_result, int create);
1544 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1545 loff_t pos, unsigned len, unsigned flags,
1546 struct page **pagep, void **fsdata)
1548 struct inode *inode = mapping->host;
1549 int ret, needed_blocks;
1556 trace_ext4_write_begin(inode, pos, len, flags);
1558 * Reserve one block more for addition to orphan list in case
1559 * we allocate blocks but write fails for some reason
1561 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1562 index = pos >> PAGE_CACHE_SHIFT;
1563 from = pos & (PAGE_CACHE_SIZE - 1);
1567 handle = ext4_journal_start(inode, needed_blocks);
1568 if (IS_ERR(handle)) {
1569 ret = PTR_ERR(handle);
1573 /* We cannot recurse into the filesystem as the transaction is already
1575 flags |= AOP_FLAG_NOFS;
1577 page = grab_cache_page_write_begin(mapping, index, flags);
1579 ext4_journal_stop(handle);
1585 if (ext4_should_dioread_nolock(inode))
1586 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1587 fsdata, ext4_get_block_write);
1589 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1590 fsdata, ext4_get_block);
1592 if (!ret && ext4_should_journal_data(inode)) {
1593 ret = walk_page_buffers(handle, page_buffers(page),
1594 from, to, NULL, do_journal_get_write_access);
1599 page_cache_release(page);
1601 * block_write_begin may have instantiated a few blocks
1602 * outside i_size. Trim these off again. Don't need
1603 * i_size_read because we hold i_mutex.
1605 * Add inode to orphan list in case we crash before
1608 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1609 ext4_orphan_add(handle, inode);
1611 ext4_journal_stop(handle);
1612 if (pos + len > inode->i_size) {
1613 ext4_truncate_failed_write(inode);
1615 * If truncate failed early the inode might
1616 * still be on the orphan list; we need to
1617 * make sure the inode is removed from the
1618 * orphan list in that case.
1621 ext4_orphan_del(NULL, inode);
1625 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1631 /* For write_end() in data=journal mode */
1632 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1634 if (!buffer_mapped(bh) || buffer_freed(bh))
1636 set_buffer_uptodate(bh);
1637 return ext4_handle_dirty_metadata(handle, NULL, bh);
1640 static int ext4_generic_write_end(struct file *file,
1641 struct address_space *mapping,
1642 loff_t pos, unsigned len, unsigned copied,
1643 struct page *page, void *fsdata)
1645 int i_size_changed = 0;
1646 struct inode *inode = mapping->host;
1647 handle_t *handle = ext4_journal_current_handle();
1649 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1652 * No need to use i_size_read() here, the i_size
1653 * cannot change under us because we hold i_mutex.
1655 * But it's important to update i_size while still holding page lock:
1656 * page writeout could otherwise come in and zero beyond i_size.
1658 if (pos + copied > inode->i_size) {
1659 i_size_write(inode, pos + copied);
1663 if (pos + copied > EXT4_I(inode)->i_disksize) {
1664 /* We need to mark inode dirty even if
1665 * new_i_size is less that inode->i_size
1666 * bu greater than i_disksize.(hint delalloc)
1668 ext4_update_i_disksize(inode, (pos + copied));
1672 page_cache_release(page);
1675 * Don't mark the inode dirty under page lock. First, it unnecessarily
1676 * makes the holding time of page lock longer. Second, it forces lock
1677 * ordering of page lock and transaction start for journaling
1681 ext4_mark_inode_dirty(handle, inode);
1687 * We need to pick up the new inode size which generic_commit_write gave us
1688 * `file' can be NULL - eg, when called from page_symlink().
1690 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1691 * buffers are managed internally.
1693 static int ext4_ordered_write_end(struct file *file,
1694 struct address_space *mapping,
1695 loff_t pos, unsigned len, unsigned copied,
1696 struct page *page, void *fsdata)
1698 handle_t *handle = ext4_journal_current_handle();
1699 struct inode *inode = mapping->host;
1702 trace_ext4_ordered_write_end(inode, pos, len, copied);
1703 ret = ext4_jbd2_file_inode(handle, inode);
1706 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1709 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1710 /* if we have allocated more blocks and copied
1711 * less. We will have blocks allocated outside
1712 * inode->i_size. So truncate them
1714 ext4_orphan_add(handle, inode);
1718 ret2 = ext4_journal_stop(handle);
1722 if (pos + len > inode->i_size) {
1723 ext4_truncate_failed_write(inode);
1725 * If truncate failed early the inode might still be
1726 * on the orphan list; we need to make sure the inode
1727 * is removed from the orphan list in that case.
1730 ext4_orphan_del(NULL, inode);
1734 return ret ? ret : copied;
1737 static int ext4_writeback_write_end(struct file *file,
1738 struct address_space *mapping,
1739 loff_t pos, unsigned len, unsigned copied,
1740 struct page *page, void *fsdata)
1742 handle_t *handle = ext4_journal_current_handle();
1743 struct inode *inode = mapping->host;
1746 trace_ext4_writeback_write_end(inode, pos, len, copied);
1747 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1750 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1751 /* if we have allocated more blocks and copied
1752 * less. We will have blocks allocated outside
1753 * inode->i_size. So truncate them
1755 ext4_orphan_add(handle, inode);
1760 ret2 = ext4_journal_stop(handle);
1764 if (pos + len > inode->i_size) {
1765 ext4_truncate_failed_write(inode);
1767 * If truncate failed early the inode might still be
1768 * on the orphan list; we need to make sure the inode
1769 * is removed from the orphan list in that case.
1772 ext4_orphan_del(NULL, inode);
1775 return ret ? ret : copied;
1778 static int ext4_journalled_write_end(struct file *file,
1779 struct address_space *mapping,
1780 loff_t pos, unsigned len, unsigned copied,
1781 struct page *page, void *fsdata)
1783 handle_t *handle = ext4_journal_current_handle();
1784 struct inode *inode = mapping->host;
1790 trace_ext4_journalled_write_end(inode, pos, len, copied);
1791 from = pos & (PAGE_CACHE_SIZE - 1);
1795 if (!PageUptodate(page))
1797 page_zero_new_buffers(page, from+copied, to);
1800 ret = walk_page_buffers(handle, page_buffers(page), from,
1801 to, &partial, write_end_fn);
1803 SetPageUptodate(page);
1804 new_i_size = pos + copied;
1805 if (new_i_size > inode->i_size)
1806 i_size_write(inode, pos+copied);
1807 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1808 if (new_i_size > EXT4_I(inode)->i_disksize) {
1809 ext4_update_i_disksize(inode, new_i_size);
1810 ret2 = ext4_mark_inode_dirty(handle, inode);
1816 page_cache_release(page);
1817 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1818 /* if we have allocated more blocks and copied
1819 * less. We will have blocks allocated outside
1820 * inode->i_size. So truncate them
1822 ext4_orphan_add(handle, inode);
1824 ret2 = ext4_journal_stop(handle);
1827 if (pos + len > inode->i_size) {
1828 ext4_truncate_failed_write(inode);
1830 * If truncate failed early the inode might still be
1831 * on the orphan list; we need to make sure the inode
1832 * is removed from the orphan list in that case.
1835 ext4_orphan_del(NULL, inode);
1838 return ret ? ret : copied;
1842 * Reserve a single block located at lblock
1844 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1847 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1848 struct ext4_inode_info *ei = EXT4_I(inode);
1849 unsigned long md_needed;
1853 * recalculate the amount of metadata blocks to reserve
1854 * in order to allocate nrblocks
1855 * worse case is one extent per block
1858 spin_lock(&ei->i_block_reservation_lock);
1859 md_needed = ext4_calc_metadata_amount(inode, lblock);
1860 trace_ext4_da_reserve_space(inode, md_needed);
1861 spin_unlock(&ei->i_block_reservation_lock);
1864 * We will charge metadata quota at writeout time; this saves
1865 * us from metadata over-estimation, though we may go over by
1866 * a small amount in the end. Here we just reserve for data.
1868 ret = dquot_reserve_block(inode, 1);
1872 * We do still charge estimated metadata to the sb though;
1873 * we cannot afford to run out of free blocks.
1875 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1876 dquot_release_reservation_block(inode, 1);
1877 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1883 spin_lock(&ei->i_block_reservation_lock);
1884 ei->i_reserved_data_blocks++;
1885 ei->i_reserved_meta_blocks += md_needed;
1886 spin_unlock(&ei->i_block_reservation_lock);
1888 return 0; /* success */
1891 static void ext4_da_release_space(struct inode *inode, int to_free)
1893 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1894 struct ext4_inode_info *ei = EXT4_I(inode);
1897 return; /* Nothing to release, exit */
1899 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1901 trace_ext4_da_release_space(inode, to_free);
1902 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1904 * if there aren't enough reserved blocks, then the
1905 * counter is messed up somewhere. Since this
1906 * function is called from invalidate page, it's
1907 * harmless to return without any action.
1909 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1910 "ino %lu, to_free %d with only %d reserved "
1911 "data blocks\n", inode->i_ino, to_free,
1912 ei->i_reserved_data_blocks);
1914 to_free = ei->i_reserved_data_blocks;
1916 ei->i_reserved_data_blocks -= to_free;
1918 if (ei->i_reserved_data_blocks == 0) {
1920 * We can release all of the reserved metadata blocks
1921 * only when we have written all of the delayed
1922 * allocation blocks.
1924 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1925 ei->i_reserved_meta_blocks);
1926 ei->i_reserved_meta_blocks = 0;
1927 ei->i_da_metadata_calc_len = 0;
1930 /* update fs dirty data blocks counter */
1931 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1933 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1935 dquot_release_reservation_block(inode, to_free);
1938 static void ext4_da_page_release_reservation(struct page *page,
1939 unsigned long offset)
1942 struct buffer_head *head, *bh;
1943 unsigned int curr_off = 0;
1945 head = page_buffers(page);
1948 unsigned int next_off = curr_off + bh->b_size;
1950 if ((offset <= curr_off) && (buffer_delay(bh))) {
1952 clear_buffer_delay(bh);
1954 curr_off = next_off;
1955 } while ((bh = bh->b_this_page) != head);
1956 ext4_da_release_space(page->mapping->host, to_release);
1960 * Delayed allocation stuff
1964 * mpage_da_submit_io - walks through extent of pages and try to write
1965 * them with writepage() call back
1967 * @mpd->inode: inode
1968 * @mpd->first_page: first page of the extent
1969 * @mpd->next_page: page after the last page of the extent
1971 * By the time mpage_da_submit_io() is called we expect all blocks
1972 * to be allocated. this may be wrong if allocation failed.
1974 * As pages are already locked by write_cache_pages(), we can't use it
1976 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1979 struct pagevec pvec;
1980 unsigned long index, end;
1981 int ret = 0, err, nr_pages, i;
1982 struct inode *inode = mpd->inode;
1983 struct address_space *mapping = inode->i_mapping;
1985 BUG_ON(mpd->next_page <= mpd->first_page);
1987 * We need to start from the first_page to the next_page - 1
1988 * to make sure we also write the mapped dirty buffer_heads.
1989 * If we look at mpd->b_blocknr we would only be looking
1990 * at the currently mapped buffer_heads.
1992 index = mpd->first_page;
1993 end = mpd->next_page - 1;
1995 pagevec_init(&pvec, 0);
1996 while (index <= end) {
1997 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2000 for (i = 0; i < nr_pages; i++) {
2001 struct page *page = pvec.pages[i];
2003 index = page->index;
2008 BUG_ON(!PageLocked(page));
2009 BUG_ON(PageWriteback(page));
2011 pages_skipped = mpd->wbc->pages_skipped;
2012 err = mapping->a_ops->writepage(page, mpd->wbc);
2013 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2015 * have successfully written the page
2016 * without skipping the same
2018 mpd->pages_written++;
2020 * In error case, we have to continue because
2021 * remaining pages are still locked
2022 * XXX: unlock and re-dirty them?
2027 pagevec_release(&pvec);
2033 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2035 * the function goes through all passed space and put actual disk
2036 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2038 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd,
2039 struct ext4_map_blocks *map)
2041 struct inode *inode = mpd->inode;
2042 struct address_space *mapping = inode->i_mapping;
2043 int blocks = map->m_len;
2044 sector_t pblock = map->m_pblk, cur_logical;
2045 struct buffer_head *head, *bh;
2047 struct pagevec pvec;
2050 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2051 end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2052 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2054 pagevec_init(&pvec, 0);
2056 while (index <= end) {
2057 /* XXX: optimize tail */
2058 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2061 for (i = 0; i < nr_pages; i++) {
2062 struct page *page = pvec.pages[i];
2064 index = page->index;
2069 BUG_ON(!PageLocked(page));
2070 BUG_ON(PageWriteback(page));
2071 BUG_ON(!page_has_buffers(page));
2073 bh = page_buffers(page);
2076 /* skip blocks out of the range */
2078 if (cur_logical >= map->m_lblk)
2081 } while ((bh = bh->b_this_page) != head);
2084 if (cur_logical >= map->m_lblk + blocks)
2087 if (buffer_delay(bh) || buffer_unwritten(bh)) {
2089 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2091 if (buffer_delay(bh)) {
2092 clear_buffer_delay(bh);
2093 bh->b_blocknr = pblock;
2096 * unwritten already should have
2097 * blocknr assigned. Verify that
2099 clear_buffer_unwritten(bh);
2100 BUG_ON(bh->b_blocknr != pblock);
2103 } else if (buffer_mapped(bh))
2104 BUG_ON(bh->b_blocknr != pblock);
2106 if (map->m_flags & EXT4_MAP_UNINIT)
2107 set_buffer_uninit(bh);
2110 } while ((bh = bh->b_this_page) != head);
2112 pagevec_release(&pvec);
2117 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2118 sector_t logical, long blk_cnt)
2122 struct pagevec pvec;
2123 struct inode *inode = mpd->inode;
2124 struct address_space *mapping = inode->i_mapping;
2126 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2127 end = (logical + blk_cnt - 1) >>
2128 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2129 while (index <= end) {
2130 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2133 for (i = 0; i < nr_pages; i++) {
2134 struct page *page = pvec.pages[i];
2135 if (page->index > end)
2137 BUG_ON(!PageLocked(page));
2138 BUG_ON(PageWriteback(page));
2139 block_invalidatepage(page, 0);
2140 ClearPageUptodate(page);
2143 index = pvec.pages[nr_pages - 1]->index + 1;
2144 pagevec_release(&pvec);
2149 static void ext4_print_free_blocks(struct inode *inode)
2151 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2152 printk(KERN_CRIT "Total free blocks count %lld\n",
2153 ext4_count_free_blocks(inode->i_sb));
2154 printk(KERN_CRIT "Free/Dirty block details\n");
2155 printk(KERN_CRIT "free_blocks=%lld\n",
2156 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2157 printk(KERN_CRIT "dirty_blocks=%lld\n",
2158 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2159 printk(KERN_CRIT "Block reservation details\n");
2160 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2161 EXT4_I(inode)->i_reserved_data_blocks);
2162 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2163 EXT4_I(inode)->i_reserved_meta_blocks);
2168 * mpage_da_map_blocks - go through given space
2170 * @mpd - bh describing space
2172 * The function skips space we know is already mapped to disk blocks.
2175 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2177 int err, blks, get_blocks_flags;
2178 struct ext4_map_blocks map;
2179 sector_t next = mpd->b_blocknr;
2180 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2181 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2182 handle_t *handle = NULL;
2185 * We consider only non-mapped and non-allocated blocks
2187 if ((mpd->b_state & (1 << BH_Mapped)) &&
2188 !(mpd->b_state & (1 << BH_Delay)) &&
2189 !(mpd->b_state & (1 << BH_Unwritten)))
2193 * If we didn't accumulate anything to write simply return
2198 handle = ext4_journal_current_handle();
2202 * Call ext4_get_blocks() to allocate any delayed allocation
2203 * blocks, or to convert an uninitialized extent to be
2204 * initialized (in the case where we have written into
2205 * one or more preallocated blocks).
2207 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2208 * indicate that we are on the delayed allocation path. This
2209 * affects functions in many different parts of the allocation
2210 * call path. This flag exists primarily because we don't
2211 * want to change *many* call functions, so ext4_get_blocks()
2212 * will set the magic i_delalloc_reserved_flag once the
2213 * inode's allocation semaphore is taken.
2215 * If the blocks in questions were delalloc blocks, set
2216 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2217 * variables are updated after the blocks have been allocated.
2220 map.m_len = max_blocks;
2221 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2222 if (ext4_should_dioread_nolock(mpd->inode))
2223 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2224 if (mpd->b_state & (1 << BH_Delay))
2225 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2227 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2231 * If get block returns with error we simply
2232 * return. Later writepage will redirty the page and
2233 * writepages will find the dirty page again
2238 if (err == -ENOSPC &&
2239 ext4_count_free_blocks(mpd->inode->i_sb)) {
2245 * get block failure will cause us to loop in
2246 * writepages, because a_ops->writepage won't be able
2247 * to make progress. The page will be redirtied by
2248 * writepage and writepages will again try to write
2251 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2252 "delayed block allocation failed for inode %lu at "
2253 "logical offset %llu with max blocks %zd with "
2254 "error %d", mpd->inode->i_ino,
2255 (unsigned long long) next,
2256 mpd->b_size >> mpd->inode->i_blkbits, err);
2257 printk(KERN_CRIT "This should not happen!! "
2258 "Data will be lost\n");
2259 if (err == -ENOSPC) {
2260 ext4_print_free_blocks(mpd->inode);
2262 /* invalidate all the pages */
2263 ext4_da_block_invalidatepages(mpd, next,
2264 mpd->b_size >> mpd->inode->i_blkbits);
2269 if (map.m_flags & EXT4_MAP_NEW) {
2270 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2273 for (i = 0; i < map.m_len; i++)
2274 unmap_underlying_metadata(bdev, map.m_pblk + i);
2278 * If blocks are delayed marked, we need to
2279 * put actual blocknr and drop delayed bit
2281 if ((mpd->b_state & (1 << BH_Delay)) ||
2282 (mpd->b_state & (1 << BH_Unwritten)))
2283 mpage_put_bnr_to_bhs(mpd, &map);
2285 if (ext4_should_order_data(mpd->inode)) {
2286 err = ext4_jbd2_file_inode(handle, mpd->inode);
2292 * Update on-disk size along with block allocation.
2294 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2295 if (disksize > i_size_read(mpd->inode))
2296 disksize = i_size_read(mpd->inode);
2297 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2298 ext4_update_i_disksize(mpd->inode, disksize);
2299 return ext4_mark_inode_dirty(handle, mpd->inode);
2305 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2306 (1 << BH_Delay) | (1 << BH_Unwritten))
2309 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2311 * @mpd->lbh - extent of blocks
2312 * @logical - logical number of the block in the file
2313 * @bh - bh of the block (used to access block's state)
2315 * the function is used to collect contig. blocks in same state
2317 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2318 sector_t logical, size_t b_size,
2319 unsigned long b_state)
2322 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2325 * XXX Don't go larger than mballoc is willing to allocate
2326 * This is a stopgap solution. We eventually need to fold
2327 * mpage_da_submit_io() into this function and then call
2328 * ext4_get_blocks() multiple times in a loop
2330 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2333 /* check if thereserved journal credits might overflow */
2334 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2335 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2337 * With non-extent format we are limited by the journal
2338 * credit available. Total credit needed to insert
2339 * nrblocks contiguous blocks is dependent on the
2340 * nrblocks. So limit nrblocks.
2343 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2344 EXT4_MAX_TRANS_DATA) {
2346 * Adding the new buffer_head would make it cross the
2347 * allowed limit for which we have journal credit
2348 * reserved. So limit the new bh->b_size
2350 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2351 mpd->inode->i_blkbits;
2352 /* we will do mpage_da_submit_io in the next loop */
2356 * First block in the extent
2358 if (mpd->b_size == 0) {
2359 mpd->b_blocknr = logical;
2360 mpd->b_size = b_size;
2361 mpd->b_state = b_state & BH_FLAGS;
2365 next = mpd->b_blocknr + nrblocks;
2367 * Can we merge the block to our big extent?
2369 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2370 mpd->b_size += b_size;
2376 * We couldn't merge the block to our extent, so we
2377 * need to flush current extent and start new one
2379 if (mpage_da_map_blocks(mpd) == 0)
2380 mpage_da_submit_io(mpd);
2385 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2387 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2391 * __mpage_da_writepage - finds extent of pages and blocks
2393 * @page: page to consider
2394 * @wbc: not used, we just follow rules
2397 * The function finds extents of pages and scan them for all blocks.
2399 static int __mpage_da_writepage(struct page *page,
2400 struct writeback_control *wbc, void *data)
2402 struct mpage_da_data *mpd = data;
2403 struct inode *inode = mpd->inode;
2404 struct buffer_head *bh, *head;
2408 * Can we merge this page to current extent?
2410 if (mpd->next_page != page->index) {
2412 * Nope, we can't. So, we map non-allocated blocks
2413 * and start IO on them using writepage()
2415 if (mpd->next_page != mpd->first_page) {
2416 if (mpage_da_map_blocks(mpd) == 0)
2417 mpage_da_submit_io(mpd);
2419 * skip rest of the page in the page_vec
2422 redirty_page_for_writepage(wbc, page);
2424 return MPAGE_DA_EXTENT_TAIL;
2428 * Start next extent of pages ...
2430 mpd->first_page = page->index;
2440 mpd->next_page = page->index + 1;
2441 logical = (sector_t) page->index <<
2442 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2444 if (!page_has_buffers(page)) {
2445 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2446 (1 << BH_Dirty) | (1 << BH_Uptodate));
2448 return MPAGE_DA_EXTENT_TAIL;
2451 * Page with regular buffer heads, just add all dirty ones
2453 head = page_buffers(page);
2456 BUG_ON(buffer_locked(bh));
2458 * We need to try to allocate
2459 * unmapped blocks in the same page.
2460 * Otherwise we won't make progress
2461 * with the page in ext4_writepage
2463 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2464 mpage_add_bh_to_extent(mpd, logical,
2468 return MPAGE_DA_EXTENT_TAIL;
2469 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2471 * mapped dirty buffer. We need to update
2472 * the b_state because we look at
2473 * b_state in mpage_da_map_blocks. We don't
2474 * update b_size because if we find an
2475 * unmapped buffer_head later we need to
2476 * use the b_state flag of that buffer_head.
2478 if (mpd->b_size == 0)
2479 mpd->b_state = bh->b_state & BH_FLAGS;
2482 } while ((bh = bh->b_this_page) != head);
2489 * This is a special get_blocks_t callback which is used by
2490 * ext4_da_write_begin(). It will either return mapped block or
2491 * reserve space for a single block.
2493 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2494 * We also have b_blocknr = -1 and b_bdev initialized properly
2496 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2497 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2498 * initialized properly.
2500 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2501 struct buffer_head *bh, int create)
2503 struct ext4_map_blocks map;
2505 sector_t invalid_block = ~((sector_t) 0xffff);
2507 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2510 BUG_ON(create == 0);
2511 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2513 map.m_lblk = iblock;
2517 * first, we need to know whether the block is allocated already
2518 * preallocated blocks are unmapped but should treated
2519 * the same as allocated blocks.
2521 ret = ext4_map_blocks(NULL, inode, &map, 0);
2525 if (buffer_delay(bh))
2526 return 0; /* Not sure this could or should happen */
2528 * XXX: __block_prepare_write() unmaps passed block,
2531 ret = ext4_da_reserve_space(inode, iblock);
2533 /* not enough space to reserve */
2536 map_bh(bh, inode->i_sb, invalid_block);
2538 set_buffer_delay(bh);
2542 map_bh(bh, inode->i_sb, map.m_pblk);
2543 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2545 if (buffer_unwritten(bh)) {
2546 /* A delayed write to unwritten bh should be marked
2547 * new and mapped. Mapped ensures that we don't do
2548 * get_block multiple times when we write to the same
2549 * offset and new ensures that we do proper zero out
2550 * for partial write.
2553 set_buffer_mapped(bh);
2559 * This function is used as a standard get_block_t calback function
2560 * when there is no desire to allocate any blocks. It is used as a
2561 * callback function for block_prepare_write() and block_write_full_page().
2562 * These functions should only try to map a single block at a time.
2564 * Since this function doesn't do block allocations even if the caller
2565 * requests it by passing in create=1, it is critically important that
2566 * any caller checks to make sure that any buffer heads are returned
2567 * by this function are either all already mapped or marked for
2568 * delayed allocation before calling block_write_full_page(). Otherwise,
2569 * b_blocknr could be left unitialized, and the page write functions will
2570 * be taken by surprise.
2572 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2573 struct buffer_head *bh_result, int create)
2575 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2576 return _ext4_get_block(inode, iblock, bh_result, 0);
2579 static int bget_one(handle_t *handle, struct buffer_head *bh)
2585 static int bput_one(handle_t *handle, struct buffer_head *bh)
2591 static int __ext4_journalled_writepage(struct page *page,
2594 struct address_space *mapping = page->mapping;
2595 struct inode *inode = mapping->host;
2596 struct buffer_head *page_bufs;
2597 handle_t *handle = NULL;
2601 page_bufs = page_buffers(page);
2603 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2604 /* As soon as we unlock the page, it can go away, but we have
2605 * references to buffers so we are safe */
2608 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2609 if (IS_ERR(handle)) {
2610 ret = PTR_ERR(handle);
2614 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2615 do_journal_get_write_access);
2617 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2621 err = ext4_journal_stop(handle);
2625 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2626 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2631 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2632 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2635 * Note that we don't need to start a transaction unless we're journaling data
2636 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2637 * need to file the inode to the transaction's list in ordered mode because if
2638 * we are writing back data added by write(), the inode is already there and if
2639 * we are writing back data modified via mmap(), noone guarantees in which
2640 * transaction the data will hit the disk. In case we are journaling data, we
2641 * cannot start transaction directly because transaction start ranks above page
2642 * lock so we have to do some magic.
2644 * This function can get called via...
2645 * - ext4_da_writepages after taking page lock (have journal handle)
2646 * - journal_submit_inode_data_buffers (no journal handle)
2647 * - shrink_page_list via pdflush (no journal handle)
2648 * - grab_page_cache when doing write_begin (have journal handle)
2650 * We don't do any block allocation in this function. If we have page with
2651 * multiple blocks we need to write those buffer_heads that are mapped. This
2652 * is important for mmaped based write. So if we do with blocksize 1K
2653 * truncate(f, 1024);
2654 * a = mmap(f, 0, 4096);
2656 * truncate(f, 4096);
2657 * we have in the page first buffer_head mapped via page_mkwrite call back
2658 * but other bufer_heads would be unmapped but dirty(dirty done via the
2659 * do_wp_page). So writepage should write the first block. If we modify
2660 * the mmap area beyond 1024 we will again get a page_fault and the
2661 * page_mkwrite callback will do the block allocation and mark the
2662 * buffer_heads mapped.
2664 * We redirty the page if we have any buffer_heads that is either delay or
2665 * unwritten in the page.
2667 * We can get recursively called as show below.
2669 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2672 * But since we don't do any block allocation we should not deadlock.
2673 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2675 static int ext4_writepage(struct page *page,
2676 struct writeback_control *wbc)
2681 struct buffer_head *page_bufs = NULL;
2682 struct inode *inode = page->mapping->host;
2684 trace_ext4_writepage(inode, page);
2685 size = i_size_read(inode);
2686 if (page->index == size >> PAGE_CACHE_SHIFT)
2687 len = size & ~PAGE_CACHE_MASK;
2689 len = PAGE_CACHE_SIZE;
2691 if (page_has_buffers(page)) {
2692 page_bufs = page_buffers(page);
2693 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2694 ext4_bh_delay_or_unwritten)) {
2696 * We don't want to do block allocation
2697 * So redirty the page and return
2698 * We may reach here when we do a journal commit
2699 * via journal_submit_inode_data_buffers.
2700 * If we don't have mapping block we just ignore
2701 * them. We can also reach here via shrink_page_list
2703 redirty_page_for_writepage(wbc, page);
2709 * The test for page_has_buffers() is subtle:
2710 * We know the page is dirty but it lost buffers. That means
2711 * that at some moment in time after write_begin()/write_end()
2712 * has been called all buffers have been clean and thus they
2713 * must have been written at least once. So they are all
2714 * mapped and we can happily proceed with mapping them
2715 * and writing the page.
2717 * Try to initialize the buffer_heads and check whether
2718 * all are mapped and non delay. We don't want to
2719 * do block allocation here.
2721 ret = block_prepare_write(page, 0, len,
2722 noalloc_get_block_write);
2724 page_bufs = page_buffers(page);
2725 /* check whether all are mapped and non delay */
2726 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2727 ext4_bh_delay_or_unwritten)) {
2728 redirty_page_for_writepage(wbc, page);
2734 * We can't do block allocation here
2735 * so just redity the page and unlock
2738 redirty_page_for_writepage(wbc, page);
2742 /* now mark the buffer_heads as dirty and uptodate */
2743 block_commit_write(page, 0, len);
2746 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2748 * It's mmapped pagecache. Add buffers and journal it. There
2749 * doesn't seem much point in redirtying the page here.
2751 ClearPageChecked(page);
2752 return __ext4_journalled_writepage(page, len);
2755 if (page_bufs && buffer_uninit(page_bufs)) {
2756 ext4_set_bh_endio(page_bufs, inode);
2757 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2758 wbc, ext4_end_io_buffer_write);
2760 ret = block_write_full_page(page, noalloc_get_block_write,
2767 * This is called via ext4_da_writepages() to
2768 * calulate the total number of credits to reserve to fit
2769 * a single extent allocation into a single transaction,
2770 * ext4_da_writpeages() will loop calling this before
2771 * the block allocation.
2774 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2776 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2779 * With non-extent format the journal credit needed to
2780 * insert nrblocks contiguous block is dependent on
2781 * number of contiguous block. So we will limit
2782 * number of contiguous block to a sane value
2784 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2785 (max_blocks > EXT4_MAX_TRANS_DATA))
2786 max_blocks = EXT4_MAX_TRANS_DATA;
2788 return ext4_chunk_trans_blocks(inode, max_blocks);
2792 * write_cache_pages_da - walk the list of dirty pages of the given
2793 * address space and call the callback function (which usually writes
2796 * This is a forked version of write_cache_pages(). Differences:
2797 * Range cyclic is ignored.
2798 * no_nrwrite_index_update is always presumed true
2800 static int write_cache_pages_da(struct address_space *mapping,
2801 struct writeback_control *wbc,
2802 struct mpage_da_data *mpd)
2806 struct pagevec pvec;
2809 pgoff_t end; /* Inclusive */
2810 long nr_to_write = wbc->nr_to_write;
2812 pagevec_init(&pvec, 0);
2813 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2814 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2816 while (!done && (index <= end)) {
2819 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2820 PAGECACHE_TAG_DIRTY,
2821 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2825 for (i = 0; i < nr_pages; i++) {
2826 struct page *page = pvec.pages[i];
2829 * At this point, the page may be truncated or
2830 * invalidated (changing page->mapping to NULL), or
2831 * even swizzled back from swapper_space to tmpfs file
2832 * mapping. However, page->index will not change
2833 * because we have a reference on the page.
2835 if (page->index > end) {
2843 * Page truncated or invalidated. We can freely skip it
2844 * then, even for data integrity operations: the page
2845 * has disappeared concurrently, so there could be no
2846 * real expectation of this data interity operation
2847 * even if there is now a new, dirty page at the same
2848 * pagecache address.
2850 if (unlikely(page->mapping != mapping)) {
2856 if (!PageDirty(page)) {
2857 /* someone wrote it for us */
2858 goto continue_unlock;
2861 if (PageWriteback(page)) {
2862 if (wbc->sync_mode != WB_SYNC_NONE)
2863 wait_on_page_writeback(page);
2865 goto continue_unlock;
2868 BUG_ON(PageWriteback(page));
2869 if (!clear_page_dirty_for_io(page))
2870 goto continue_unlock;
2872 ret = __mpage_da_writepage(page, wbc, mpd);
2873 if (unlikely(ret)) {
2874 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2883 if (nr_to_write > 0) {
2885 if (nr_to_write == 0 &&
2886 wbc->sync_mode == WB_SYNC_NONE) {
2888 * We stop writing back only if we are
2889 * not doing integrity sync. In case of
2890 * integrity sync we have to keep going
2891 * because someone may be concurrently
2892 * dirtying pages, and we might have
2893 * synced a lot of newly appeared dirty
2894 * pages, but have not synced all of the
2902 pagevec_release(&pvec);
2909 static int ext4_da_writepages(struct address_space *mapping,
2910 struct writeback_control *wbc)
2913 int range_whole = 0;
2914 handle_t *handle = NULL;
2915 struct mpage_da_data mpd;
2916 struct inode *inode = mapping->host;
2917 int pages_written = 0;
2919 unsigned int max_pages;
2920 int range_cyclic, cycled = 1, io_done = 0;
2921 int needed_blocks, ret = 0;
2922 long desired_nr_to_write, nr_to_writebump = 0;
2923 loff_t range_start = wbc->range_start;
2924 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2926 trace_ext4_da_writepages(inode, wbc);
2929 * No pages to write? This is mainly a kludge to avoid starting
2930 * a transaction for special inodes like journal inode on last iput()
2931 * because that could violate lock ordering on umount
2933 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2937 * If the filesystem has aborted, it is read-only, so return
2938 * right away instead of dumping stack traces later on that
2939 * will obscure the real source of the problem. We test
2940 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2941 * the latter could be true if the filesystem is mounted
2942 * read-only, and in that case, ext4_da_writepages should
2943 * *never* be called, so if that ever happens, we would want
2946 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2949 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2952 range_cyclic = wbc->range_cyclic;
2953 if (wbc->range_cyclic) {
2954 index = mapping->writeback_index;
2957 wbc->range_start = index << PAGE_CACHE_SHIFT;
2958 wbc->range_end = LLONG_MAX;
2959 wbc->range_cyclic = 0;
2961 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2964 * This works around two forms of stupidity. The first is in
2965 * the writeback code, which caps the maximum number of pages
2966 * written to be 1024 pages. This is wrong on multiple
2967 * levels; different architectues have a different page size,
2968 * which changes the maximum amount of data which gets
2969 * written. Secondly, 4 megabytes is way too small. XFS
2970 * forces this value to be 16 megabytes by multiplying
2971 * nr_to_write parameter by four, and then relies on its
2972 * allocator to allocate larger extents to make them
2973 * contiguous. Unfortunately this brings us to the second
2974 * stupidity, which is that ext4's mballoc code only allocates
2975 * at most 2048 blocks. So we force contiguous writes up to
2976 * the number of dirty blocks in the inode, or
2977 * sbi->max_writeback_mb_bump whichever is smaller.
2979 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2980 if (!range_cyclic && range_whole)
2981 desired_nr_to_write = wbc->nr_to_write * 8;
2983 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2985 if (desired_nr_to_write > max_pages)
2986 desired_nr_to_write = max_pages;
2988 if (wbc->nr_to_write < desired_nr_to_write) {
2989 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2990 wbc->nr_to_write = desired_nr_to_write;
2994 mpd.inode = mapping->host;
2996 pages_skipped = wbc->pages_skipped;
2999 while (!ret && wbc->nr_to_write > 0) {
3002 * we insert one extent at a time. So we need
3003 * credit needed for single extent allocation.
3004 * journalled mode is currently not supported
3007 BUG_ON(ext4_should_journal_data(inode));
3008 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3010 /* start a new transaction*/
3011 handle = ext4_journal_start(inode, needed_blocks);
3012 if (IS_ERR(handle)) {
3013 ret = PTR_ERR(handle);
3014 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3015 "%ld pages, ino %lu; err %d", __func__,
3016 wbc->nr_to_write, inode->i_ino, ret);
3017 goto out_writepages;
3021 * Now call __mpage_da_writepage to find the next
3022 * contiguous region of logical blocks that need
3023 * blocks to be allocated by ext4. We don't actually
3024 * submit the blocks for I/O here, even though
3025 * write_cache_pages thinks it will, and will set the
3026 * pages as clean for write before calling
3027 * __mpage_da_writepage().
3035 mpd.pages_written = 0;
3037 ret = write_cache_pages_da(mapping, wbc, &mpd);
3039 * If we have a contiguous extent of pages and we
3040 * haven't done the I/O yet, map the blocks and submit
3043 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3044 if (mpage_da_map_blocks(&mpd) == 0)
3045 mpage_da_submit_io(&mpd);
3047 ret = MPAGE_DA_EXTENT_TAIL;
3049 trace_ext4_da_write_pages(inode, &mpd);
3050 wbc->nr_to_write -= mpd.pages_written;
3052 ext4_journal_stop(handle);
3054 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3055 /* commit the transaction which would
3056 * free blocks released in the transaction
3059 jbd2_journal_force_commit_nested(sbi->s_journal);
3060 wbc->pages_skipped = pages_skipped;
3062 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3064 * got one extent now try with
3067 pages_written += mpd.pages_written;
3068 wbc->pages_skipped = pages_skipped;
3071 } else if (wbc->nr_to_write)
3073 * There is no more writeout needed
3074 * or we requested for a noblocking writeout
3075 * and we found the device congested
3079 if (!io_done && !cycled) {
3082 wbc->range_start = index << PAGE_CACHE_SHIFT;
3083 wbc->range_end = mapping->writeback_index - 1;
3086 if (pages_skipped != wbc->pages_skipped)
3087 ext4_msg(inode->i_sb, KERN_CRIT,
3088 "This should not happen leaving %s "
3089 "with nr_to_write = %ld ret = %d",
3090 __func__, wbc->nr_to_write, ret);
3093 index += pages_written;
3094 wbc->range_cyclic = range_cyclic;
3095 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3097 * set the writeback_index so that range_cyclic
3098 * mode will write it back later
3100 mapping->writeback_index = index;
3103 wbc->nr_to_write -= nr_to_writebump;
3104 wbc->range_start = range_start;
3105 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3109 #define FALL_BACK_TO_NONDELALLOC 1
3110 static int ext4_nonda_switch(struct super_block *sb)
3112 s64 free_blocks, dirty_blocks;
3113 struct ext4_sb_info *sbi = EXT4_SB(sb);
3116 * switch to non delalloc mode if we are running low
3117 * on free block. The free block accounting via percpu
3118 * counters can get slightly wrong with percpu_counter_batch getting
3119 * accumulated on each CPU without updating global counters
3120 * Delalloc need an accurate free block accounting. So switch
3121 * to non delalloc when we are near to error range.
3123 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3124 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3125 if (2 * free_blocks < 3 * dirty_blocks ||
3126 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3128 * free block count is less than 150% of dirty blocks
3129 * or free blocks is less than watermark
3134 * Even if we don't switch but are nearing capacity,
3135 * start pushing delalloc when 1/2 of free blocks are dirty.
3137 if (free_blocks < 2 * dirty_blocks)
3138 writeback_inodes_sb_if_idle(sb);
3143 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3144 loff_t pos, unsigned len, unsigned flags,
3145 struct page **pagep, void **fsdata)
3147 int ret, retries = 0;
3150 struct inode *inode = mapping->host;
3153 index = pos >> PAGE_CACHE_SHIFT;
3155 if (ext4_nonda_switch(inode->i_sb)) {
3156 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3157 return ext4_write_begin(file, mapping, pos,
3158 len, flags, pagep, fsdata);
3160 *fsdata = (void *)0;
3161 trace_ext4_da_write_begin(inode, pos, len, flags);
3164 * With delayed allocation, we don't log the i_disksize update
3165 * if there is delayed block allocation. But we still need
3166 * to journalling the i_disksize update if writes to the end
3167 * of file which has an already mapped buffer.
3169 handle = ext4_journal_start(inode, 1);
3170 if (IS_ERR(handle)) {
3171 ret = PTR_ERR(handle);
3174 /* We cannot recurse into the filesystem as the transaction is already
3176 flags |= AOP_FLAG_NOFS;
3178 page = grab_cache_page_write_begin(mapping, index, flags);
3180 ext4_journal_stop(handle);
3186 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3187 ext4_da_get_block_prep);
3190 ext4_journal_stop(handle);
3191 page_cache_release(page);
3193 * block_write_begin may have instantiated a few blocks
3194 * outside i_size. Trim these off again. Don't need
3195 * i_size_read because we hold i_mutex.
3197 if (pos + len > inode->i_size)
3198 ext4_truncate_failed_write(inode);
3201 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3208 * Check if we should update i_disksize
3209 * when write to the end of file but not require block allocation
3211 static int ext4_da_should_update_i_disksize(struct page *page,
3212 unsigned long offset)
3214 struct buffer_head *bh;
3215 struct inode *inode = page->mapping->host;
3219 bh = page_buffers(page);
3220 idx = offset >> inode->i_blkbits;
3222 for (i = 0; i < idx; i++)
3223 bh = bh->b_this_page;
3225 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3230 static int ext4_da_write_end(struct file *file,
3231 struct address_space *mapping,
3232 loff_t pos, unsigned len, unsigned copied,
3233 struct page *page, void *fsdata)
3235 struct inode *inode = mapping->host;
3237 handle_t *handle = ext4_journal_current_handle();
3239 unsigned long start, end;
3240 int write_mode = (int)(unsigned long)fsdata;
3242 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3243 if (ext4_should_order_data(inode)) {
3244 return ext4_ordered_write_end(file, mapping, pos,
3245 len, copied, page, fsdata);
3246 } else if (ext4_should_writeback_data(inode)) {
3247 return ext4_writeback_write_end(file, mapping, pos,
3248 len, copied, page, fsdata);
3254 trace_ext4_da_write_end(inode, pos, len, copied);
3255 start = pos & (PAGE_CACHE_SIZE - 1);
3256 end = start + copied - 1;
3259 * generic_write_end() will run mark_inode_dirty() if i_size
3260 * changes. So let's piggyback the i_disksize mark_inode_dirty
3264 new_i_size = pos + copied;
3265 if (new_i_size > EXT4_I(inode)->i_disksize) {
3266 if (ext4_da_should_update_i_disksize(page, end)) {
3267 down_write(&EXT4_I(inode)->i_data_sem);
3268 if (new_i_size > EXT4_I(inode)->i_disksize) {
3270 * Updating i_disksize when extending file
3271 * without needing block allocation
3273 if (ext4_should_order_data(inode))
3274 ret = ext4_jbd2_file_inode(handle,
3277 EXT4_I(inode)->i_disksize = new_i_size;
3279 up_write(&EXT4_I(inode)->i_data_sem);
3280 /* We need to mark inode dirty even if
3281 * new_i_size is less that inode->i_size
3282 * bu greater than i_disksize.(hint delalloc)
3284 ext4_mark_inode_dirty(handle, inode);
3287 ret2 = generic_write_end(file, mapping, pos, len, copied,
3292 ret2 = ext4_journal_stop(handle);
3296 return ret ? ret : copied;
3299 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3302 * Drop reserved blocks
3304 BUG_ON(!PageLocked(page));
3305 if (!page_has_buffers(page))
3308 ext4_da_page_release_reservation(page, offset);
3311 ext4_invalidatepage(page, offset);
3317 * Force all delayed allocation blocks to be allocated for a given inode.
3319 int ext4_alloc_da_blocks(struct inode *inode)
3321 trace_ext4_alloc_da_blocks(inode);
3323 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3324 !EXT4_I(inode)->i_reserved_meta_blocks)
3328 * We do something simple for now. The filemap_flush() will
3329 * also start triggering a write of the data blocks, which is
3330 * not strictly speaking necessary (and for users of
3331 * laptop_mode, not even desirable). However, to do otherwise
3332 * would require replicating code paths in:
3334 * ext4_da_writepages() ->
3335 * write_cache_pages() ---> (via passed in callback function)
3336 * __mpage_da_writepage() -->
3337 * mpage_add_bh_to_extent()
3338 * mpage_da_map_blocks()
3340 * The problem is that write_cache_pages(), located in
3341 * mm/page-writeback.c, marks pages clean in preparation for
3342 * doing I/O, which is not desirable if we're not planning on
3345 * We could call write_cache_pages(), and then redirty all of
3346 * the pages by calling redirty_page_for_writeback() but that
3347 * would be ugly in the extreme. So instead we would need to
3348 * replicate parts of the code in the above functions,
3349 * simplifying them becuase we wouldn't actually intend to
3350 * write out the pages, but rather only collect contiguous
3351 * logical block extents, call the multi-block allocator, and
3352 * then update the buffer heads with the block allocations.
3354 * For now, though, we'll cheat by calling filemap_flush(),
3355 * which will map the blocks, and start the I/O, but not
3356 * actually wait for the I/O to complete.
3358 return filemap_flush(inode->i_mapping);
3362 * bmap() is special. It gets used by applications such as lilo and by
3363 * the swapper to find the on-disk block of a specific piece of data.
3365 * Naturally, this is dangerous if the block concerned is still in the
3366 * journal. If somebody makes a swapfile on an ext4 data-journaling
3367 * filesystem and enables swap, then they may get a nasty shock when the
3368 * data getting swapped to that swapfile suddenly gets overwritten by
3369 * the original zero's written out previously to the journal and
3370 * awaiting writeback in the kernel's buffer cache.
3372 * So, if we see any bmap calls here on a modified, data-journaled file,
3373 * take extra steps to flush any blocks which might be in the cache.
3375 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3377 struct inode *inode = mapping->host;
3381 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3382 test_opt(inode->i_sb, DELALLOC)) {
3384 * With delalloc we want to sync the file
3385 * so that we can make sure we allocate
3388 filemap_write_and_wait(mapping);
3391 if (EXT4_JOURNAL(inode) &&
3392 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3394 * This is a REALLY heavyweight approach, but the use of
3395 * bmap on dirty files is expected to be extremely rare:
3396 * only if we run lilo or swapon on a freshly made file
3397 * do we expect this to happen.
3399 * (bmap requires CAP_SYS_RAWIO so this does not
3400 * represent an unprivileged user DOS attack --- we'd be
3401 * in trouble if mortal users could trigger this path at
3404 * NB. EXT4_STATE_JDATA is not set on files other than
3405 * regular files. If somebody wants to bmap a directory
3406 * or symlink and gets confused because the buffer
3407 * hasn't yet been flushed to disk, they deserve
3408 * everything they get.
3411 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3412 journal = EXT4_JOURNAL(inode);
3413 jbd2_journal_lock_updates(journal);
3414 err = jbd2_journal_flush(journal);
3415 jbd2_journal_unlock_updates(journal);
3421 return generic_block_bmap(mapping, block, ext4_get_block);
3424 static int ext4_readpage(struct file *file, struct page *page)
3426 return mpage_readpage(page, ext4_get_block);
3430 ext4_readpages(struct file *file, struct address_space *mapping,
3431 struct list_head *pages, unsigned nr_pages)
3433 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3436 static void ext4_free_io_end(ext4_io_end_t *io)
3445 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3447 struct buffer_head *head, *bh;
3448 unsigned int curr_off = 0;
3450 if (!page_has_buffers(page))
3452 head = bh = page_buffers(page);
3454 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3456 ext4_free_io_end(bh->b_private);
3457 bh->b_private = NULL;
3458 bh->b_end_io = NULL;
3460 curr_off = curr_off + bh->b_size;
3461 bh = bh->b_this_page;
3462 } while (bh != head);
3465 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3467 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3470 * free any io_end structure allocated for buffers to be discarded
3472 if (ext4_should_dioread_nolock(page->mapping->host))
3473 ext4_invalidatepage_free_endio(page, offset);
3475 * If it's a full truncate we just forget about the pending dirtying
3478 ClearPageChecked(page);
3481 jbd2_journal_invalidatepage(journal, page, offset);
3483 block_invalidatepage(page, offset);
3486 static int ext4_releasepage(struct page *page, gfp_t wait)
3488 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3490 WARN_ON(PageChecked(page));
3491 if (!page_has_buffers(page))
3494 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3496 return try_to_free_buffers(page);
3500 * O_DIRECT for ext3 (or indirect map) based files
3502 * If the O_DIRECT write will extend the file then add this inode to the
3503 * orphan list. So recovery will truncate it back to the original size
3504 * if the machine crashes during the write.
3506 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3507 * crashes then stale disk data _may_ be exposed inside the file. But current
3508 * VFS code falls back into buffered path in that case so we are safe.
3510 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3511 const struct iovec *iov, loff_t offset,
3512 unsigned long nr_segs)
3514 struct file *file = iocb->ki_filp;
3515 struct inode *inode = file->f_mapping->host;
3516 struct ext4_inode_info *ei = EXT4_I(inode);
3520 size_t count = iov_length(iov, nr_segs);
3524 loff_t final_size = offset + count;
3526 if (final_size > inode->i_size) {
3527 /* Credits for sb + inode write */
3528 handle = ext4_journal_start(inode, 2);
3529 if (IS_ERR(handle)) {
3530 ret = PTR_ERR(handle);
3533 ret = ext4_orphan_add(handle, inode);
3535 ext4_journal_stop(handle);
3539 ei->i_disksize = inode->i_size;
3540 ext4_journal_stop(handle);
3545 if (rw == READ && ext4_should_dioread_nolock(inode))
3546 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
3547 inode->i_sb->s_bdev, iov,
3549 ext4_get_block, NULL);
3551 ret = blockdev_direct_IO(rw, iocb, inode,
3552 inode->i_sb->s_bdev, iov,
3554 ext4_get_block, NULL);
3555 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3561 /* Credits for sb + inode write */
3562 handle = ext4_journal_start(inode, 2);
3563 if (IS_ERR(handle)) {
3564 /* This is really bad luck. We've written the data
3565 * but cannot extend i_size. Bail out and pretend
3566 * the write failed... */
3567 ret = PTR_ERR(handle);
3569 ext4_orphan_del(NULL, inode);
3574 ext4_orphan_del(handle, inode);
3576 loff_t end = offset + ret;
3577 if (end > inode->i_size) {
3578 ei->i_disksize = end;
3579 i_size_write(inode, end);
3581 * We're going to return a positive `ret'
3582 * here due to non-zero-length I/O, so there's
3583 * no way of reporting error returns from
3584 * ext4_mark_inode_dirty() to userspace. So
3587 ext4_mark_inode_dirty(handle, inode);
3590 err = ext4_journal_stop(handle);
3599 * ext4_get_block used when preparing for a DIO write or buffer write.
3600 * We allocate an uinitialized extent if blocks haven't been allocated.
3601 * The extent will be converted to initialized after the IO is complete.
3603 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3604 struct buffer_head *bh_result, int create)
3606 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3607 inode->i_ino, create);
3608 return _ext4_get_block(inode, iblock, bh_result,
3609 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3612 static void dump_completed_IO(struct inode * inode)
3615 struct list_head *cur, *before, *after;
3616 ext4_io_end_t *io, *io0, *io1;
3617 unsigned long flags;
3619 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3620 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3624 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3625 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3626 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3629 io0 = container_of(before, ext4_io_end_t, list);
3631 io1 = container_of(after, ext4_io_end_t, list);
3633 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3634 io, inode->i_ino, io0, io1);
3636 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3641 * check a range of space and convert unwritten extents to written.
3643 static int ext4_end_io_nolock(ext4_io_end_t *io)
3645 struct inode *inode = io->inode;
3646 loff_t offset = io->offset;
3647 ssize_t size = io->size;
3650 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3651 "list->prev 0x%p\n",
3652 io, inode->i_ino, io->list.next, io->list.prev);
3654 if (list_empty(&io->list))
3657 if (io->flag != EXT4_IO_UNWRITTEN)
3660 ret = ext4_convert_unwritten_extents(inode, offset, size);
3662 printk(KERN_EMERG "%s: failed to convert unwritten"
3663 "extents to written extents, error is %d"
3664 " io is still on inode %lu aio dio list\n",
3665 __func__, ret, inode->i_ino);
3669 /* clear the DIO AIO unwritten flag */
3675 * work on completed aio dio IO, to convert unwritten extents to extents
3677 static void ext4_end_io_work(struct work_struct *work)
3679 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3680 struct inode *inode = io->inode;
3681 struct ext4_inode_info *ei = EXT4_I(inode);
3682 unsigned long flags;
3685 mutex_lock(&inode->i_mutex);
3686 ret = ext4_end_io_nolock(io);
3688 mutex_unlock(&inode->i_mutex);
3692 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3693 if (!list_empty(&io->list))
3694 list_del_init(&io->list);
3695 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3696 mutex_unlock(&inode->i_mutex);
3697 ext4_free_io_end(io);
3701 * This function is called from ext4_sync_file().
3703 * When IO is completed, the work to convert unwritten extents to
3704 * written is queued on workqueue but may not get immediately
3705 * scheduled. When fsync is called, we need to ensure the
3706 * conversion is complete before fsync returns.
3707 * The inode keeps track of a list of pending/completed IO that
3708 * might needs to do the conversion. This function walks through
3709 * the list and convert the related unwritten extents for completed IO
3711 * The function return the number of pending IOs on success.
3713 int flush_completed_IO(struct inode *inode)
3716 struct ext4_inode_info *ei = EXT4_I(inode);
3717 unsigned long flags;
3721 if (list_empty(&ei->i_completed_io_list))
3724 dump_completed_IO(inode);
3725 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3726 while (!list_empty(&ei->i_completed_io_list)){
3727 io = list_entry(ei->i_completed_io_list.next,
3728 ext4_io_end_t, list);
3730 * Calling ext4_end_io_nolock() to convert completed
3733 * When ext4_sync_file() is called, run_queue() may already
3734 * about to flush the work corresponding to this io structure.
3735 * It will be upset if it founds the io structure related
3736 * to the work-to-be schedule is freed.
3738 * Thus we need to keep the io structure still valid here after
3739 * convertion finished. The io structure has a flag to
3740 * avoid double converting from both fsync and background work
3743 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3744 ret = ext4_end_io_nolock(io);
3745 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3749 list_del_init(&io->list);
3751 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3752 return (ret2 < 0) ? ret2 : 0;
3755 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3757 ext4_io_end_t *io = NULL;
3759 io = kmalloc(sizeof(*io), flags);
3768 INIT_WORK(&io->work, ext4_end_io_work);
3769 INIT_LIST_HEAD(&io->list);
3775 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3776 ssize_t size, void *private)
3778 ext4_io_end_t *io_end = iocb->private;
3779 struct workqueue_struct *wq;
3780 unsigned long flags;
3781 struct ext4_inode_info *ei;
3783 /* if not async direct IO or dio with 0 bytes write, just return */
3784 if (!io_end || !size)
3787 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3788 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3789 iocb->private, io_end->inode->i_ino, iocb, offset,
3792 /* if not aio dio with unwritten extents, just free io and return */
3793 if (io_end->flag != EXT4_IO_UNWRITTEN){
3794 ext4_free_io_end(io_end);
3795 iocb->private = NULL;
3799 io_end->offset = offset;
3800 io_end->size = size;
3801 io_end->flag = EXT4_IO_UNWRITTEN;
3802 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3804 /* queue the work to convert unwritten extents to written */
3805 queue_work(wq, &io_end->work);
3807 /* Add the io_end to per-inode completed aio dio list*/
3808 ei = EXT4_I(io_end->inode);
3809 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3810 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3811 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3812 iocb->private = NULL;
3815 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3817 ext4_io_end_t *io_end = bh->b_private;
3818 struct workqueue_struct *wq;
3819 struct inode *inode;
3820 unsigned long flags;
3822 if (!test_clear_buffer_uninit(bh) || !io_end)
3825 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3826 printk("sb umounted, discard end_io request for inode %lu\n",
3827 io_end->inode->i_ino);
3828 ext4_free_io_end(io_end);
3832 io_end->flag = EXT4_IO_UNWRITTEN;
3833 inode = io_end->inode;
3835 /* Add the io_end to per-inode completed io list*/
3836 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3837 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3838 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3840 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3841 /* queue the work to convert unwritten extents to written */
3842 queue_work(wq, &io_end->work);
3844 bh->b_private = NULL;
3845 bh->b_end_io = NULL;
3846 clear_buffer_uninit(bh);
3847 end_buffer_async_write(bh, uptodate);
3850 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3852 ext4_io_end_t *io_end;
3853 struct page *page = bh->b_page;
3854 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3855 size_t size = bh->b_size;
3858 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3860 if (printk_ratelimit())
3861 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3865 io_end->offset = offset;
3866 io_end->size = size;
3868 * We need to hold a reference to the page to make sure it
3869 * doesn't get evicted before ext4_end_io_work() has a chance
3870 * to convert the extent from written to unwritten.
3872 io_end->page = page;
3873 get_page(io_end->page);
3875 bh->b_private = io_end;
3876 bh->b_end_io = ext4_end_io_buffer_write;
3881 * For ext4 extent files, ext4 will do direct-io write to holes,
3882 * preallocated extents, and those write extend the file, no need to
3883 * fall back to buffered IO.
3885 * For holes, we fallocate those blocks, mark them as unintialized
3886 * If those blocks were preallocated, we mark sure they are splited, but
3887 * still keep the range to write as unintialized.
3889 * The unwrritten extents will be converted to written when DIO is completed.
3890 * For async direct IO, since the IO may still pending when return, we
3891 * set up an end_io call back function, which will do the convertion
3892 * when async direct IO completed.
3894 * If the O_DIRECT write will extend the file then add this inode to the
3895 * orphan list. So recovery will truncate it back to the original size
3896 * if the machine crashes during the write.
3899 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3900 const struct iovec *iov, loff_t offset,
3901 unsigned long nr_segs)
3903 struct file *file = iocb->ki_filp;
3904 struct inode *inode = file->f_mapping->host;
3906 size_t count = iov_length(iov, nr_segs);
3908 loff_t final_size = offset + count;
3909 if (rw == WRITE && final_size <= inode->i_size) {
3911 * We could direct write to holes and fallocate.
3913 * Allocated blocks to fill the hole are marked as uninitialized
3914 * to prevent paralel buffered read to expose the stale data
3915 * before DIO complete the data IO.
3917 * As to previously fallocated extents, ext4 get_block
3918 * will just simply mark the buffer mapped but still
3919 * keep the extents uninitialized.
3921 * for non AIO case, we will convert those unwritten extents
3922 * to written after return back from blockdev_direct_IO.
3924 * for async DIO, the conversion needs to be defered when
3925 * the IO is completed. The ext4 end_io callback function
3926 * will be called to take care of the conversion work.
3927 * Here for async case, we allocate an io_end structure to
3930 iocb->private = NULL;
3931 EXT4_I(inode)->cur_aio_dio = NULL;
3932 if (!is_sync_kiocb(iocb)) {
3933 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3937 * we save the io structure for current async
3938 * direct IO, so that later ext4_get_blocks()
3939 * could flag the io structure whether there
3940 * is a unwritten extents needs to be converted
3941 * when IO is completed.
3943 EXT4_I(inode)->cur_aio_dio = iocb->private;
3946 ret = blockdev_direct_IO(rw, iocb, inode,
3947 inode->i_sb->s_bdev, iov,
3949 ext4_get_block_write,
3952 EXT4_I(inode)->cur_aio_dio = NULL;
3954 * The io_end structure takes a reference to the inode,
3955 * that structure needs to be destroyed and the
3956 * reference to the inode need to be dropped, when IO is
3957 * complete, even with 0 byte write, or failed.
3959 * In the successful AIO DIO case, the io_end structure will be
3960 * desctroyed and the reference to the inode will be dropped
3961 * after the end_io call back function is called.
3963 * In the case there is 0 byte write, or error case, since
3964 * VFS direct IO won't invoke the end_io call back function,
3965 * we need to free the end_io structure here.
3967 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3968 ext4_free_io_end(iocb->private);
3969 iocb->private = NULL;
3970 } else if (ret > 0 && ext4_test_inode_state(inode,
3971 EXT4_STATE_DIO_UNWRITTEN)) {
3974 * for non AIO case, since the IO is already
3975 * completed, we could do the convertion right here
3977 err = ext4_convert_unwritten_extents(inode,
3981 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3986 /* for write the the end of file case, we fall back to old way */
3987 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3990 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3991 const struct iovec *iov, loff_t offset,
3992 unsigned long nr_segs)
3994 struct file *file = iocb->ki_filp;
3995 struct inode *inode = file->f_mapping->host;
3997 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3998 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
4000 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4004 * Pages can be marked dirty completely asynchronously from ext4's journalling
4005 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4006 * much here because ->set_page_dirty is called under VFS locks. The page is
4007 * not necessarily locked.
4009 * We cannot just dirty the page and leave attached buffers clean, because the
4010 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4011 * or jbddirty because all the journalling code will explode.
4013 * So what we do is to mark the page "pending dirty" and next time writepage
4014 * is called, propagate that into the buffers appropriately.
4016 static int ext4_journalled_set_page_dirty(struct page *page)
4018 SetPageChecked(page);
4019 return __set_page_dirty_nobuffers(page);
4022 static const struct address_space_operations ext4_ordered_aops = {
4023 .readpage = ext4_readpage,
4024 .readpages = ext4_readpages,
4025 .writepage = ext4_writepage,
4026 .sync_page = block_sync_page,
4027 .write_begin = ext4_write_begin,
4028 .write_end = ext4_ordered_write_end,
4030 .invalidatepage = ext4_invalidatepage,
4031 .releasepage = ext4_releasepage,
4032 .direct_IO = ext4_direct_IO,
4033 .migratepage = buffer_migrate_page,
4034 .is_partially_uptodate = block_is_partially_uptodate,
4035 .error_remove_page = generic_error_remove_page,
4038 static const struct address_space_operations ext4_writeback_aops = {
4039 .readpage = ext4_readpage,
4040 .readpages = ext4_readpages,
4041 .writepage = ext4_writepage,
4042 .sync_page = block_sync_page,
4043 .write_begin = ext4_write_begin,
4044 .write_end = ext4_writeback_write_end,
4046 .invalidatepage = ext4_invalidatepage,
4047 .releasepage = ext4_releasepage,
4048 .direct_IO = ext4_direct_IO,
4049 .migratepage = buffer_migrate_page,
4050 .is_partially_uptodate = block_is_partially_uptodate,
4051 .error_remove_page = generic_error_remove_page,
4054 static const struct address_space_operations ext4_journalled_aops = {
4055 .readpage = ext4_readpage,
4056 .readpages = ext4_readpages,
4057 .writepage = ext4_writepage,
4058 .sync_page = block_sync_page,
4059 .write_begin = ext4_write_begin,
4060 .write_end = ext4_journalled_write_end,
4061 .set_page_dirty = ext4_journalled_set_page_dirty,
4063 .invalidatepage = ext4_invalidatepage,
4064 .releasepage = ext4_releasepage,
4065 .is_partially_uptodate = block_is_partially_uptodate,
4066 .error_remove_page = generic_error_remove_page,
4069 static const struct address_space_operations ext4_da_aops = {
4070 .readpage = ext4_readpage,
4071 .readpages = ext4_readpages,
4072 .writepage = ext4_writepage,
4073 .writepages = ext4_da_writepages,
4074 .sync_page = block_sync_page,
4075 .write_begin = ext4_da_write_begin,
4076 .write_end = ext4_da_write_end,
4078 .invalidatepage = ext4_da_invalidatepage,
4079 .releasepage = ext4_releasepage,
4080 .direct_IO = ext4_direct_IO,
4081 .migratepage = buffer_migrate_page,
4082 .is_partially_uptodate = block_is_partially_uptodate,
4083 .error_remove_page = generic_error_remove_page,
4086 void ext4_set_aops(struct inode *inode)
4088 if (ext4_should_order_data(inode) &&
4089 test_opt(inode->i_sb, DELALLOC))
4090 inode->i_mapping->a_ops = &ext4_da_aops;
4091 else if (ext4_should_order_data(inode))
4092 inode->i_mapping->a_ops = &ext4_ordered_aops;
4093 else if (ext4_should_writeback_data(inode) &&
4094 test_opt(inode->i_sb, DELALLOC))
4095 inode->i_mapping->a_ops = &ext4_da_aops;
4096 else if (ext4_should_writeback_data(inode))
4097 inode->i_mapping->a_ops = &ext4_writeback_aops;
4099 inode->i_mapping->a_ops = &ext4_journalled_aops;
4103 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4104 * up to the end of the block which corresponds to `from'.
4105 * This required during truncate. We need to physically zero the tail end
4106 * of that block so it doesn't yield old data if the file is later grown.
4108 int ext4_block_truncate_page(handle_t *handle,
4109 struct address_space *mapping, loff_t from)
4111 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4112 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4113 unsigned blocksize, length, pos;
4115 struct inode *inode = mapping->host;
4116 struct buffer_head *bh;
4120 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4121 mapping_gfp_mask(mapping) & ~__GFP_FS);
4125 blocksize = inode->i_sb->s_blocksize;
4126 length = blocksize - (offset & (blocksize - 1));
4127 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4129 if (!page_has_buffers(page))
4130 create_empty_buffers(page, blocksize, 0);
4132 /* Find the buffer that contains "offset" */
4133 bh = page_buffers(page);
4135 while (offset >= pos) {
4136 bh = bh->b_this_page;
4142 if (buffer_freed(bh)) {
4143 BUFFER_TRACE(bh, "freed: skip");
4147 if (!buffer_mapped(bh)) {
4148 BUFFER_TRACE(bh, "unmapped");
4149 ext4_get_block(inode, iblock, bh, 0);
4150 /* unmapped? It's a hole - nothing to do */
4151 if (!buffer_mapped(bh)) {
4152 BUFFER_TRACE(bh, "still unmapped");
4157 /* Ok, it's mapped. Make sure it's up-to-date */
4158 if (PageUptodate(page))
4159 set_buffer_uptodate(bh);
4161 if (!buffer_uptodate(bh)) {
4163 ll_rw_block(READ, 1, &bh);
4165 /* Uhhuh. Read error. Complain and punt. */
4166 if (!buffer_uptodate(bh))
4170 if (ext4_should_journal_data(inode)) {
4171 BUFFER_TRACE(bh, "get write access");
4172 err = ext4_journal_get_write_access(handle, bh);
4177 zero_user(page, offset, length);
4179 BUFFER_TRACE(bh, "zeroed end of block");
4182 if (ext4_should_journal_data(inode)) {
4183 err = ext4_handle_dirty_metadata(handle, inode, bh);
4185 if (ext4_should_order_data(inode))
4186 err = ext4_jbd2_file_inode(handle, inode);
4187 mark_buffer_dirty(bh);
4192 page_cache_release(page);
4197 * Probably it should be a library function... search for first non-zero word
4198 * or memcmp with zero_page, whatever is better for particular architecture.
4201 static inline int all_zeroes(__le32 *p, __le32 *q)
4210 * ext4_find_shared - find the indirect blocks for partial truncation.
4211 * @inode: inode in question
4212 * @depth: depth of the affected branch
4213 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4214 * @chain: place to store the pointers to partial indirect blocks
4215 * @top: place to the (detached) top of branch
4217 * This is a helper function used by ext4_truncate().
4219 * When we do truncate() we may have to clean the ends of several
4220 * indirect blocks but leave the blocks themselves alive. Block is
4221 * partially truncated if some data below the new i_size is refered
4222 * from it (and it is on the path to the first completely truncated
4223 * data block, indeed). We have to free the top of that path along
4224 * with everything to the right of the path. Since no allocation
4225 * past the truncation point is possible until ext4_truncate()
4226 * finishes, we may safely do the latter, but top of branch may
4227 * require special attention - pageout below the truncation point
4228 * might try to populate it.
4230 * We atomically detach the top of branch from the tree, store the
4231 * block number of its root in *@top, pointers to buffer_heads of
4232 * partially truncated blocks - in @chain[].bh and pointers to
4233 * their last elements that should not be removed - in
4234 * @chain[].p. Return value is the pointer to last filled element
4237 * The work left to caller to do the actual freeing of subtrees:
4238 * a) free the subtree starting from *@top
4239 * b) free the subtrees whose roots are stored in
4240 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4241 * c) free the subtrees growing from the inode past the @chain[0].
4242 * (no partially truncated stuff there). */
4244 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4245 ext4_lblk_t offsets[4], Indirect chain[4],
4248 Indirect *partial, *p;
4252 /* Make k index the deepest non-null offset + 1 */
4253 for (k = depth; k > 1 && !offsets[k-1]; k--)
4255 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4256 /* Writer: pointers */
4258 partial = chain + k-1;
4260 * If the branch acquired continuation since we've looked at it -
4261 * fine, it should all survive and (new) top doesn't belong to us.
4263 if (!partial->key && *partial->p)
4266 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4269 * OK, we've found the last block that must survive. The rest of our
4270 * branch should be detached before unlocking. However, if that rest
4271 * of branch is all ours and does not grow immediately from the inode
4272 * it's easier to cheat and just decrement partial->p.
4274 if (p == chain + k - 1 && p > chain) {
4278 /* Nope, don't do this in ext4. Must leave the tree intact */
4285 while (partial > p) {
4286 brelse(partial->bh);
4294 * Zero a number of block pointers in either an inode or an indirect block.
4295 * If we restart the transaction we must again get write access to the
4296 * indirect block for further modification.
4298 * We release `count' blocks on disk, but (last - first) may be greater
4299 * than `count' because there can be holes in there.
4301 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4302 struct buffer_head *bh,
4303 ext4_fsblk_t block_to_free,
4304 unsigned long count, __le32 *first,
4308 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4310 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4311 flags |= EXT4_FREE_BLOCKS_METADATA;
4313 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4315 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4316 "blocks %llu len %lu",
4317 (unsigned long long) block_to_free, count);
4321 if (try_to_extend_transaction(handle, inode)) {
4323 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4324 ext4_handle_dirty_metadata(handle, inode, bh);
4326 ext4_mark_inode_dirty(handle, inode);
4327 ext4_truncate_restart_trans(handle, inode,
4328 blocks_for_truncate(inode));
4330 BUFFER_TRACE(bh, "retaking write access");
4331 ext4_journal_get_write_access(handle, bh);
4335 for (p = first; p < last; p++)
4338 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4343 * ext4_free_data - free a list of data blocks
4344 * @handle: handle for this transaction
4345 * @inode: inode we are dealing with
4346 * @this_bh: indirect buffer_head which contains *@first and *@last
4347 * @first: array of block numbers
4348 * @last: points immediately past the end of array
4350 * We are freeing all blocks refered from that array (numbers are stored as
4351 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4353 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4354 * blocks are contiguous then releasing them at one time will only affect one
4355 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4356 * actually use a lot of journal space.
4358 * @this_bh will be %NULL if @first and @last point into the inode's direct
4361 static void ext4_free_data(handle_t *handle, struct inode *inode,
4362 struct buffer_head *this_bh,
4363 __le32 *first, __le32 *last)
4365 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4366 unsigned long count = 0; /* Number of blocks in the run */
4367 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4370 ext4_fsblk_t nr; /* Current block # */
4371 __le32 *p; /* Pointer into inode/ind
4372 for current block */
4375 if (this_bh) { /* For indirect block */
4376 BUFFER_TRACE(this_bh, "get_write_access");
4377 err = ext4_journal_get_write_access(handle, this_bh);
4378 /* Important: if we can't update the indirect pointers
4379 * to the blocks, we can't free them. */
4384 for (p = first; p < last; p++) {
4385 nr = le32_to_cpu(*p);
4387 /* accumulate blocks to free if they're contiguous */
4390 block_to_free_p = p;
4392 } else if (nr == block_to_free + count) {
4395 if (ext4_clear_blocks(handle, inode, this_bh,
4396 block_to_free, count,
4397 block_to_free_p, p))
4400 block_to_free_p = p;
4407 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4408 count, block_to_free_p, p);
4411 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4414 * The buffer head should have an attached journal head at this
4415 * point. However, if the data is corrupted and an indirect
4416 * block pointed to itself, it would have been detached when
4417 * the block was cleared. Check for this instead of OOPSing.
4419 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4420 ext4_handle_dirty_metadata(handle, inode, this_bh);
4422 EXT4_ERROR_INODE(inode,
4423 "circular indirect block detected at "
4425 (unsigned long long) this_bh->b_blocknr);
4430 * ext4_free_branches - free an array of branches
4431 * @handle: JBD handle for this transaction
4432 * @inode: inode we are dealing with
4433 * @parent_bh: the buffer_head which contains *@first and *@last
4434 * @first: array of block numbers
4435 * @last: pointer immediately past the end of array
4436 * @depth: depth of the branches to free
4438 * We are freeing all blocks refered from these branches (numbers are
4439 * stored as little-endian 32-bit) and updating @inode->i_blocks
4442 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4443 struct buffer_head *parent_bh,
4444 __le32 *first, __le32 *last, int depth)
4449 if (ext4_handle_is_aborted(handle))
4453 struct buffer_head *bh;
4454 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4456 while (--p >= first) {
4457 nr = le32_to_cpu(*p);
4459 continue; /* A hole */
4461 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4463 EXT4_ERROR_INODE(inode,
4464 "invalid indirect mapped "
4465 "block %lu (level %d)",
4466 (unsigned long) nr, depth);
4470 /* Go read the buffer for the next level down */
4471 bh = sb_bread(inode->i_sb, nr);
4474 * A read failure? Report error and clear slot
4478 EXT4_ERROR_INODE_BLOCK(inode, nr,
4483 /* This zaps the entire block. Bottom up. */
4484 BUFFER_TRACE(bh, "free child branches");
4485 ext4_free_branches(handle, inode, bh,
4486 (__le32 *) bh->b_data,
4487 (__le32 *) bh->b_data + addr_per_block,
4491 * We've probably journalled the indirect block several
4492 * times during the truncate. But it's no longer
4493 * needed and we now drop it from the transaction via
4494 * jbd2_journal_revoke().
4496 * That's easy if it's exclusively part of this
4497 * transaction. But if it's part of the committing
4498 * transaction then jbd2_journal_forget() will simply
4499 * brelse() it. That means that if the underlying
4500 * block is reallocated in ext4_get_block(),
4501 * unmap_underlying_metadata() will find this block
4502 * and will try to get rid of it. damn, damn.
4504 * If this block has already been committed to the
4505 * journal, a revoke record will be written. And
4506 * revoke records must be emitted *before* clearing
4507 * this block's bit in the bitmaps.
4509 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
4512 * Everything below this this pointer has been
4513 * released. Now let this top-of-subtree go.
4515 * We want the freeing of this indirect block to be
4516 * atomic in the journal with the updating of the
4517 * bitmap block which owns it. So make some room in
4520 * We zero the parent pointer *after* freeing its
4521 * pointee in the bitmaps, so if extend_transaction()
4522 * for some reason fails to put the bitmap changes and
4523 * the release into the same transaction, recovery
4524 * will merely complain about releasing a free block,
4525 * rather than leaking blocks.
4527 if (ext4_handle_is_aborted(handle))
4529 if (try_to_extend_transaction(handle, inode)) {
4530 ext4_mark_inode_dirty(handle, inode);
4531 ext4_truncate_restart_trans(handle, inode,
4532 blocks_for_truncate(inode));
4535 ext4_free_blocks(handle, inode, 0, nr, 1,
4536 EXT4_FREE_BLOCKS_METADATA);
4540 * The block which we have just freed is
4541 * pointed to by an indirect block: journal it
4543 BUFFER_TRACE(parent_bh, "get_write_access");
4544 if (!ext4_journal_get_write_access(handle,
4547 BUFFER_TRACE(parent_bh,
4548 "call ext4_handle_dirty_metadata");
4549 ext4_handle_dirty_metadata(handle,
4556 /* We have reached the bottom of the tree. */
4557 BUFFER_TRACE(parent_bh, "free data blocks");
4558 ext4_free_data(handle, inode, parent_bh, first, last);
4562 int ext4_can_truncate(struct inode *inode)
4564 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4566 if (S_ISREG(inode->i_mode))
4568 if (S_ISDIR(inode->i_mode))
4570 if (S_ISLNK(inode->i_mode))
4571 return !ext4_inode_is_fast_symlink(inode);
4578 * We block out ext4_get_block() block instantiations across the entire
4579 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4580 * simultaneously on behalf of the same inode.
4582 * As we work through the truncate and commmit bits of it to the journal there
4583 * is one core, guiding principle: the file's tree must always be consistent on
4584 * disk. We must be able to restart the truncate after a crash.
4586 * The file's tree may be transiently inconsistent in memory (although it
4587 * probably isn't), but whenever we close off and commit a journal transaction,
4588 * the contents of (the filesystem + the journal) must be consistent and
4589 * restartable. It's pretty simple, really: bottom up, right to left (although
4590 * left-to-right works OK too).
4592 * Note that at recovery time, journal replay occurs *before* the restart of
4593 * truncate against the orphan inode list.
4595 * The committed inode has the new, desired i_size (which is the same as
4596 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4597 * that this inode's truncate did not complete and it will again call
4598 * ext4_truncate() to have another go. So there will be instantiated blocks
4599 * to the right of the truncation point in a crashed ext4 filesystem. But
4600 * that's fine - as long as they are linked from the inode, the post-crash
4601 * ext4_truncate() run will find them and release them.
4603 void ext4_truncate(struct inode *inode)
4606 struct ext4_inode_info *ei = EXT4_I(inode);
4607 __le32 *i_data = ei->i_data;
4608 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4609 struct address_space *mapping = inode->i_mapping;
4610 ext4_lblk_t offsets[4];
4615 ext4_lblk_t last_block;
4616 unsigned blocksize = inode->i_sb->s_blocksize;
4618 if (!ext4_can_truncate(inode))
4621 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4623 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4624 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4626 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4627 ext4_ext_truncate(inode);
4631 handle = start_transaction(inode);
4633 return; /* AKPM: return what? */
4635 last_block = (inode->i_size + blocksize-1)
4636 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4638 if (inode->i_size & (blocksize - 1))
4639 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4642 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4644 goto out_stop; /* error */
4647 * OK. This truncate is going to happen. We add the inode to the
4648 * orphan list, so that if this truncate spans multiple transactions,
4649 * and we crash, we will resume the truncate when the filesystem
4650 * recovers. It also marks the inode dirty, to catch the new size.
4652 * Implication: the file must always be in a sane, consistent
4653 * truncatable state while each transaction commits.
4655 if (ext4_orphan_add(handle, inode))
4659 * From here we block out all ext4_get_block() callers who want to
4660 * modify the block allocation tree.
4662 down_write(&ei->i_data_sem);
4664 ext4_discard_preallocations(inode);
4667 * The orphan list entry will now protect us from any crash which
4668 * occurs before the truncate completes, so it is now safe to propagate
4669 * the new, shorter inode size (held for now in i_size) into the
4670 * on-disk inode. We do this via i_disksize, which is the value which
4671 * ext4 *really* writes onto the disk inode.
4673 ei->i_disksize = inode->i_size;
4675 if (n == 1) { /* direct blocks */
4676 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4677 i_data + EXT4_NDIR_BLOCKS);
4681 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4682 /* Kill the top of shared branch (not detached) */
4684 if (partial == chain) {
4685 /* Shared branch grows from the inode */
4686 ext4_free_branches(handle, inode, NULL,
4687 &nr, &nr+1, (chain+n-1) - partial);
4690 * We mark the inode dirty prior to restart,
4691 * and prior to stop. No need for it here.
4694 /* Shared branch grows from an indirect block */
4695 BUFFER_TRACE(partial->bh, "get_write_access");
4696 ext4_free_branches(handle, inode, partial->bh,
4698 partial->p+1, (chain+n-1) - partial);
4701 /* Clear the ends of indirect blocks on the shared branch */
4702 while (partial > chain) {
4703 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4704 (__le32*)partial->bh->b_data+addr_per_block,
4705 (chain+n-1) - partial);
4706 BUFFER_TRACE(partial->bh, "call brelse");
4707 brelse(partial->bh);
4711 /* Kill the remaining (whole) subtrees */
4712 switch (offsets[0]) {
4714 nr = i_data[EXT4_IND_BLOCK];
4716 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4717 i_data[EXT4_IND_BLOCK] = 0;
4719 case EXT4_IND_BLOCK:
4720 nr = i_data[EXT4_DIND_BLOCK];
4722 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4723 i_data[EXT4_DIND_BLOCK] = 0;
4725 case EXT4_DIND_BLOCK:
4726 nr = i_data[EXT4_TIND_BLOCK];
4728 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4729 i_data[EXT4_TIND_BLOCK] = 0;
4731 case EXT4_TIND_BLOCK:
4735 up_write(&ei->i_data_sem);
4736 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4737 ext4_mark_inode_dirty(handle, inode);
4740 * In a multi-transaction truncate, we only make the final transaction
4744 ext4_handle_sync(handle);
4747 * If this was a simple ftruncate(), and the file will remain alive
4748 * then we need to clear up the orphan record which we created above.
4749 * However, if this was a real unlink then we were called by
4750 * ext4_delete_inode(), and we allow that function to clean up the
4751 * orphan info for us.
4754 ext4_orphan_del(handle, inode);
4756 ext4_journal_stop(handle);
4760 * ext4_get_inode_loc returns with an extra refcount against the inode's
4761 * underlying buffer_head on success. If 'in_mem' is true, we have all
4762 * data in memory that is needed to recreate the on-disk version of this
4765 static int __ext4_get_inode_loc(struct inode *inode,
4766 struct ext4_iloc *iloc, int in_mem)
4768 struct ext4_group_desc *gdp;
4769 struct buffer_head *bh;
4770 struct super_block *sb = inode->i_sb;
4772 int inodes_per_block, inode_offset;
4775 if (!ext4_valid_inum(sb, inode->i_ino))
4778 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4779 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4784 * Figure out the offset within the block group inode table
4786 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4787 inode_offset = ((inode->i_ino - 1) %
4788 EXT4_INODES_PER_GROUP(sb));
4789 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4790 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4792 bh = sb_getblk(sb, block);
4794 EXT4_ERROR_INODE_BLOCK(inode, block,
4795 "unable to read itable block");
4798 if (!buffer_uptodate(bh)) {
4802 * If the buffer has the write error flag, we have failed
4803 * to write out another inode in the same block. In this
4804 * case, we don't have to read the block because we may
4805 * read the old inode data successfully.
4807 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4808 set_buffer_uptodate(bh);
4810 if (buffer_uptodate(bh)) {
4811 /* someone brought it uptodate while we waited */
4817 * If we have all information of the inode in memory and this
4818 * is the only valid inode in the block, we need not read the
4822 struct buffer_head *bitmap_bh;
4825 start = inode_offset & ~(inodes_per_block - 1);
4827 /* Is the inode bitmap in cache? */
4828 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4833 * If the inode bitmap isn't in cache then the
4834 * optimisation may end up performing two reads instead
4835 * of one, so skip it.
4837 if (!buffer_uptodate(bitmap_bh)) {
4841 for (i = start; i < start + inodes_per_block; i++) {
4842 if (i == inode_offset)
4844 if (ext4_test_bit(i, bitmap_bh->b_data))
4848 if (i == start + inodes_per_block) {
4849 /* all other inodes are free, so skip I/O */
4850 memset(bh->b_data, 0, bh->b_size);
4851 set_buffer_uptodate(bh);
4859 * If we need to do any I/O, try to pre-readahead extra
4860 * blocks from the inode table.
4862 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4863 ext4_fsblk_t b, end, table;
4866 table = ext4_inode_table(sb, gdp);
4867 /* s_inode_readahead_blks is always a power of 2 */
4868 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4871 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4872 num = EXT4_INODES_PER_GROUP(sb);
4873 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4874 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4875 num -= ext4_itable_unused_count(sb, gdp);
4876 table += num / inodes_per_block;
4880 sb_breadahead(sb, b++);
4884 * There are other valid inodes in the buffer, this inode
4885 * has in-inode xattrs, or we don't have this inode in memory.
4886 * Read the block from disk.
4889 bh->b_end_io = end_buffer_read_sync;
4890 submit_bh(READ_META, bh);
4892 if (!buffer_uptodate(bh)) {
4893 EXT4_ERROR_INODE_BLOCK(inode, block,
4894 "unable to read itable block");
4904 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4906 /* We have all inode data except xattrs in memory here. */
4907 return __ext4_get_inode_loc(inode, iloc,
4908 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4911 void ext4_set_inode_flags(struct inode *inode)
4913 unsigned int flags = EXT4_I(inode)->i_flags;
4915 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4916 if (flags & EXT4_SYNC_FL)
4917 inode->i_flags |= S_SYNC;
4918 if (flags & EXT4_APPEND_FL)
4919 inode->i_flags |= S_APPEND;
4920 if (flags & EXT4_IMMUTABLE_FL)
4921 inode->i_flags |= S_IMMUTABLE;
4922 if (flags & EXT4_NOATIME_FL)
4923 inode->i_flags |= S_NOATIME;
4924 if (flags & EXT4_DIRSYNC_FL)
4925 inode->i_flags |= S_DIRSYNC;
4928 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4929 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4931 unsigned int vfs_fl;
4932 unsigned long old_fl, new_fl;
4935 vfs_fl = ei->vfs_inode.i_flags;
4936 old_fl = ei->i_flags;
4937 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4938 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4940 if (vfs_fl & S_SYNC)
4941 new_fl |= EXT4_SYNC_FL;
4942 if (vfs_fl & S_APPEND)
4943 new_fl |= EXT4_APPEND_FL;
4944 if (vfs_fl & S_IMMUTABLE)
4945 new_fl |= EXT4_IMMUTABLE_FL;
4946 if (vfs_fl & S_NOATIME)
4947 new_fl |= EXT4_NOATIME_FL;
4948 if (vfs_fl & S_DIRSYNC)
4949 new_fl |= EXT4_DIRSYNC_FL;
4950 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4953 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4954 struct ext4_inode_info *ei)
4957 struct inode *inode = &(ei->vfs_inode);
4958 struct super_block *sb = inode->i_sb;
4960 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4961 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4962 /* we are using combined 48 bit field */
4963 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4964 le32_to_cpu(raw_inode->i_blocks_lo);
4965 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4966 /* i_blocks represent file system block size */
4967 return i_blocks << (inode->i_blkbits - 9);
4972 return le32_to_cpu(raw_inode->i_blocks_lo);
4976 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4978 struct ext4_iloc iloc;
4979 struct ext4_inode *raw_inode;
4980 struct ext4_inode_info *ei;
4981 struct inode *inode;
4982 journal_t *journal = EXT4_SB(sb)->s_journal;
4986 inode = iget_locked(sb, ino);
4988 return ERR_PTR(-ENOMEM);
4989 if (!(inode->i_state & I_NEW))
4995 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4998 raw_inode = ext4_raw_inode(&iloc);
4999 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5000 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5001 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5002 if (!(test_opt(inode->i_sb, NO_UID32))) {
5003 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5004 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5006 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
5008 ei->i_state_flags = 0;
5009 ei->i_dir_start_lookup = 0;
5010 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5011 /* We now have enough fields to check if the inode was active or not.
5012 * This is needed because nfsd might try to access dead inodes
5013 * the test is that same one that e2fsck uses
5014 * NeilBrown 1999oct15
5016 if (inode->i_nlink == 0) {
5017 if (inode->i_mode == 0 ||
5018 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5019 /* this inode is deleted */
5023 /* The only unlinked inodes we let through here have
5024 * valid i_mode and are being read by the orphan
5025 * recovery code: that's fine, we're about to complete
5026 * the process of deleting those. */
5028 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5029 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5030 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5031 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5033 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5034 inode->i_size = ext4_isize(raw_inode);
5035 ei->i_disksize = inode->i_size;
5037 ei->i_reserved_quota = 0;
5039 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5040 ei->i_block_group = iloc.block_group;
5041 ei->i_last_alloc_group = ~0;
5043 * NOTE! The in-memory inode i_data array is in little-endian order
5044 * even on big-endian machines: we do NOT byteswap the block numbers!
5046 for (block = 0; block < EXT4_N_BLOCKS; block++)
5047 ei->i_data[block] = raw_inode->i_block[block];
5048 INIT_LIST_HEAD(&ei->i_orphan);
5051 * Set transaction id's of transactions that have to be committed
5052 * to finish f[data]sync. We set them to currently running transaction
5053 * as we cannot be sure that the inode or some of its metadata isn't
5054 * part of the transaction - the inode could have been reclaimed and
5055 * now it is reread from disk.
5058 transaction_t *transaction;
5061 spin_lock(&journal->j_state_lock);
5062 if (journal->j_running_transaction)
5063 transaction = journal->j_running_transaction;
5065 transaction = journal->j_committing_transaction;
5067 tid = transaction->t_tid;
5069 tid = journal->j_commit_sequence;
5070 spin_unlock(&journal->j_state_lock);
5071 ei->i_sync_tid = tid;
5072 ei->i_datasync_tid = tid;
5075 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5076 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5077 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5078 EXT4_INODE_SIZE(inode->i_sb)) {
5082 if (ei->i_extra_isize == 0) {
5083 /* The extra space is currently unused. Use it. */
5084 ei->i_extra_isize = sizeof(struct ext4_inode) -
5085 EXT4_GOOD_OLD_INODE_SIZE;
5087 __le32 *magic = (void *)raw_inode +
5088 EXT4_GOOD_OLD_INODE_SIZE +
5090 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5091 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5094 ei->i_extra_isize = 0;
5096 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5097 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5098 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5099 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5101 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5102 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5103 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5105 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5109 if (ei->i_file_acl &&
5110 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5111 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5115 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5116 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5117 (S_ISLNK(inode->i_mode) &&
5118 !ext4_inode_is_fast_symlink(inode)))
5119 /* Validate extent which is part of inode */
5120 ret = ext4_ext_check_inode(inode);
5121 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5122 (S_ISLNK(inode->i_mode) &&
5123 !ext4_inode_is_fast_symlink(inode))) {
5124 /* Validate block references which are part of inode */
5125 ret = ext4_check_inode_blockref(inode);
5130 if (S_ISREG(inode->i_mode)) {
5131 inode->i_op = &ext4_file_inode_operations;
5132 inode->i_fop = &ext4_file_operations;
5133 ext4_set_aops(inode);
5134 } else if (S_ISDIR(inode->i_mode)) {
5135 inode->i_op = &ext4_dir_inode_operations;
5136 inode->i_fop = &ext4_dir_operations;
5137 } else if (S_ISLNK(inode->i_mode)) {
5138 if (ext4_inode_is_fast_symlink(inode)) {
5139 inode->i_op = &ext4_fast_symlink_inode_operations;
5140 nd_terminate_link(ei->i_data, inode->i_size,
5141 sizeof(ei->i_data) - 1);
5143 inode->i_op = &ext4_symlink_inode_operations;
5144 ext4_set_aops(inode);
5146 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5147 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5148 inode->i_op = &ext4_special_inode_operations;
5149 if (raw_inode->i_block[0])
5150 init_special_inode(inode, inode->i_mode,
5151 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5153 init_special_inode(inode, inode->i_mode,
5154 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5157 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5161 ext4_set_inode_flags(inode);
5162 unlock_new_inode(inode);
5168 return ERR_PTR(ret);
5171 static int ext4_inode_blocks_set(handle_t *handle,
5172 struct ext4_inode *raw_inode,
5173 struct ext4_inode_info *ei)
5175 struct inode *inode = &(ei->vfs_inode);
5176 u64 i_blocks = inode->i_blocks;
5177 struct super_block *sb = inode->i_sb;
5179 if (i_blocks <= ~0U) {
5181 * i_blocks can be represnted in a 32 bit variable
5182 * as multiple of 512 bytes
5184 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5185 raw_inode->i_blocks_high = 0;
5186 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5189 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5192 if (i_blocks <= 0xffffffffffffULL) {
5194 * i_blocks can be represented in a 48 bit variable
5195 * as multiple of 512 bytes
5197 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5198 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5199 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5201 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5202 /* i_block is stored in file system block size */
5203 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5204 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5205 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5211 * Post the struct inode info into an on-disk inode location in the
5212 * buffer-cache. This gobbles the caller's reference to the
5213 * buffer_head in the inode location struct.
5215 * The caller must have write access to iloc->bh.
5217 static int ext4_do_update_inode(handle_t *handle,
5218 struct inode *inode,
5219 struct ext4_iloc *iloc)
5221 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5222 struct ext4_inode_info *ei = EXT4_I(inode);
5223 struct buffer_head *bh = iloc->bh;
5224 int err = 0, rc, block;
5226 /* For fields not not tracking in the in-memory inode,
5227 * initialise them to zero for new inodes. */
5228 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5229 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5231 ext4_get_inode_flags(ei);
5232 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5233 if (!(test_opt(inode->i_sb, NO_UID32))) {
5234 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5235 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5237 * Fix up interoperability with old kernels. Otherwise, old inodes get
5238 * re-used with the upper 16 bits of the uid/gid intact
5241 raw_inode->i_uid_high =
5242 cpu_to_le16(high_16_bits(inode->i_uid));
5243 raw_inode->i_gid_high =
5244 cpu_to_le16(high_16_bits(inode->i_gid));
5246 raw_inode->i_uid_high = 0;
5247 raw_inode->i_gid_high = 0;
5250 raw_inode->i_uid_low =
5251 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5252 raw_inode->i_gid_low =
5253 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5254 raw_inode->i_uid_high = 0;
5255 raw_inode->i_gid_high = 0;
5257 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5259 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5260 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5261 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5262 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5264 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5266 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5267 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5268 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5269 cpu_to_le32(EXT4_OS_HURD))
5270 raw_inode->i_file_acl_high =
5271 cpu_to_le16(ei->i_file_acl >> 32);
5272 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5273 ext4_isize_set(raw_inode, ei->i_disksize);
5274 if (ei->i_disksize > 0x7fffffffULL) {
5275 struct super_block *sb = inode->i_sb;
5276 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5277 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5278 EXT4_SB(sb)->s_es->s_rev_level ==
5279 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5280 /* If this is the first large file
5281 * created, add a flag to the superblock.
5283 err = ext4_journal_get_write_access(handle,
5284 EXT4_SB(sb)->s_sbh);
5287 ext4_update_dynamic_rev(sb);
5288 EXT4_SET_RO_COMPAT_FEATURE(sb,
5289 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5291 ext4_handle_sync(handle);
5292 err = ext4_handle_dirty_metadata(handle, NULL,
5293 EXT4_SB(sb)->s_sbh);
5296 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5297 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5298 if (old_valid_dev(inode->i_rdev)) {
5299 raw_inode->i_block[0] =
5300 cpu_to_le32(old_encode_dev(inode->i_rdev));
5301 raw_inode->i_block[1] = 0;
5303 raw_inode->i_block[0] = 0;
5304 raw_inode->i_block[1] =
5305 cpu_to_le32(new_encode_dev(inode->i_rdev));
5306 raw_inode->i_block[2] = 0;
5309 for (block = 0; block < EXT4_N_BLOCKS; block++)
5310 raw_inode->i_block[block] = ei->i_data[block];
5312 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5313 if (ei->i_extra_isize) {
5314 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5315 raw_inode->i_version_hi =
5316 cpu_to_le32(inode->i_version >> 32);
5317 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5320 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5321 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5324 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5326 ext4_update_inode_fsync_trans(handle, inode, 0);
5329 ext4_std_error(inode->i_sb, err);
5334 * ext4_write_inode()
5336 * We are called from a few places:
5338 * - Within generic_file_write() for O_SYNC files.
5339 * Here, there will be no transaction running. We wait for any running
5340 * trasnaction to commit.
5342 * - Within sys_sync(), kupdate and such.
5343 * We wait on commit, if tol to.
5345 * - Within prune_icache() (PF_MEMALLOC == true)
5346 * Here we simply return. We can't afford to block kswapd on the
5349 * In all cases it is actually safe for us to return without doing anything,
5350 * because the inode has been copied into a raw inode buffer in
5351 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5354 * Note that we are absolutely dependent upon all inode dirtiers doing the
5355 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5356 * which we are interested.
5358 * It would be a bug for them to not do this. The code:
5360 * mark_inode_dirty(inode)
5362 * inode->i_size = expr;
5364 * is in error because a kswapd-driven write_inode() could occur while
5365 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5366 * will no longer be on the superblock's dirty inode list.
5368 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5372 if (current->flags & PF_MEMALLOC)
5375 if (EXT4_SB(inode->i_sb)->s_journal) {
5376 if (ext4_journal_current_handle()) {
5377 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5382 if (wbc->sync_mode != WB_SYNC_ALL)
5385 err = ext4_force_commit(inode->i_sb);
5387 struct ext4_iloc iloc;
5389 err = __ext4_get_inode_loc(inode, &iloc, 0);
5392 if (wbc->sync_mode == WB_SYNC_ALL)
5393 sync_dirty_buffer(iloc.bh);
5394 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5395 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5396 "IO error syncing inode");
5407 * Called from notify_change.
5409 * We want to trap VFS attempts to truncate the file as soon as
5410 * possible. In particular, we want to make sure that when the VFS
5411 * shrinks i_size, we put the inode on the orphan list and modify
5412 * i_disksize immediately, so that during the subsequent flushing of
5413 * dirty pages and freeing of disk blocks, we can guarantee that any
5414 * commit will leave the blocks being flushed in an unused state on
5415 * disk. (On recovery, the inode will get truncated and the blocks will
5416 * be freed, so we have a strong guarantee that no future commit will
5417 * leave these blocks visible to the user.)
5419 * Another thing we have to assure is that if we are in ordered mode
5420 * and inode is still attached to the committing transaction, we must
5421 * we start writeout of all the dirty pages which are being truncated.
5422 * This way we are sure that all the data written in the previous
5423 * transaction are already on disk (truncate waits for pages under
5426 * Called with inode->i_mutex down.
5428 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5430 struct inode *inode = dentry->d_inode;
5432 const unsigned int ia_valid = attr->ia_valid;
5434 error = inode_change_ok(inode, attr);
5438 if (is_quota_modification(inode, attr))
5439 dquot_initialize(inode);
5440 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5441 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5444 /* (user+group)*(old+new) structure, inode write (sb,
5445 * inode block, ? - but truncate inode update has it) */
5446 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5447 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5448 if (IS_ERR(handle)) {
5449 error = PTR_ERR(handle);
5452 error = dquot_transfer(inode, attr);
5454 ext4_journal_stop(handle);
5457 /* Update corresponding info in inode so that everything is in
5458 * one transaction */
5459 if (attr->ia_valid & ATTR_UID)
5460 inode->i_uid = attr->ia_uid;
5461 if (attr->ia_valid & ATTR_GID)
5462 inode->i_gid = attr->ia_gid;
5463 error = ext4_mark_inode_dirty(handle, inode);
5464 ext4_journal_stop(handle);
5467 if (attr->ia_valid & ATTR_SIZE) {
5468 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5469 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5471 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5478 if (S_ISREG(inode->i_mode) &&
5479 attr->ia_valid & ATTR_SIZE &&
5480 (attr->ia_size < inode->i_size ||
5481 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5484 handle = ext4_journal_start(inode, 3);
5485 if (IS_ERR(handle)) {
5486 error = PTR_ERR(handle);
5490 error = ext4_orphan_add(handle, inode);
5491 EXT4_I(inode)->i_disksize = attr->ia_size;
5492 rc = ext4_mark_inode_dirty(handle, inode);
5495 ext4_journal_stop(handle);
5497 if (ext4_should_order_data(inode)) {
5498 error = ext4_begin_ordered_truncate(inode,
5501 /* Do as much error cleanup as possible */
5502 handle = ext4_journal_start(inode, 3);
5503 if (IS_ERR(handle)) {
5504 ext4_orphan_del(NULL, inode);
5507 ext4_orphan_del(handle, inode);
5508 ext4_journal_stop(handle);
5512 /* ext4_truncate will clear the flag */
5513 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5514 ext4_truncate(inode);
5517 rc = inode_setattr(inode, attr);
5519 /* If inode_setattr's call to ext4_truncate failed to get a
5520 * transaction handle at all, we need to clean up the in-core
5521 * orphan list manually. */
5523 ext4_orphan_del(NULL, inode);
5525 if (!rc && (ia_valid & ATTR_MODE))
5526 rc = ext4_acl_chmod(inode);
5529 ext4_std_error(inode->i_sb, error);
5535 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5538 struct inode *inode;
5539 unsigned long delalloc_blocks;
5541 inode = dentry->d_inode;
5542 generic_fillattr(inode, stat);
5545 * We can't update i_blocks if the block allocation is delayed
5546 * otherwise in the case of system crash before the real block
5547 * allocation is done, we will have i_blocks inconsistent with
5548 * on-disk file blocks.
5549 * We always keep i_blocks updated together with real
5550 * allocation. But to not confuse with user, stat
5551 * will return the blocks that include the delayed allocation
5552 * blocks for this file.
5554 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5555 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5556 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5558 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5562 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5567 /* if nrblocks are contiguous */
5570 * With N contiguous data blocks, it need at most
5571 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5572 * 2 dindirect blocks
5575 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5576 return indirects + 3;
5579 * if nrblocks are not contiguous, worse case, each block touch
5580 * a indirect block, and each indirect block touch a double indirect
5581 * block, plus a triple indirect block
5583 indirects = nrblocks * 2 + 1;
5587 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5589 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5590 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5591 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5595 * Account for index blocks, block groups bitmaps and block group
5596 * descriptor blocks if modify datablocks and index blocks
5597 * worse case, the indexs blocks spread over different block groups
5599 * If datablocks are discontiguous, they are possible to spread over
5600 * different block groups too. If they are contiuguous, with flexbg,
5601 * they could still across block group boundary.
5603 * Also account for superblock, inode, quota and xattr blocks
5605 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5607 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5613 * How many index blocks need to touch to modify nrblocks?
5614 * The "Chunk" flag indicating whether the nrblocks is
5615 * physically contiguous on disk
5617 * For Direct IO and fallocate, they calls get_block to allocate
5618 * one single extent at a time, so they could set the "Chunk" flag
5620 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5625 * Now let's see how many group bitmaps and group descriptors need
5635 if (groups > ngroups)
5637 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5638 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5640 /* bitmaps and block group descriptor blocks */
5641 ret += groups + gdpblocks;
5643 /* Blocks for super block, inode, quota and xattr blocks */
5644 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5650 * Calulate the total number of credits to reserve to fit
5651 * the modification of a single pages into a single transaction,
5652 * which may include multiple chunks of block allocations.
5654 * This could be called via ext4_write_begin()
5656 * We need to consider the worse case, when
5657 * one new block per extent.
5659 int ext4_writepage_trans_blocks(struct inode *inode)
5661 int bpp = ext4_journal_blocks_per_page(inode);
5664 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5666 /* Account for data blocks for journalled mode */
5667 if (ext4_should_journal_data(inode))
5673 * Calculate the journal credits for a chunk of data modification.
5675 * This is called from DIO, fallocate or whoever calling
5676 * ext4_get_blocks() to map/allocate a chunk of contiguous disk blocks.
5678 * journal buffers for data blocks are not included here, as DIO
5679 * and fallocate do no need to journal data buffers.
5681 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5683 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5687 * The caller must have previously called ext4_reserve_inode_write().
5688 * Give this, we know that the caller already has write access to iloc->bh.
5690 int ext4_mark_iloc_dirty(handle_t *handle,
5691 struct inode *inode, struct ext4_iloc *iloc)
5695 if (test_opt(inode->i_sb, I_VERSION))
5696 inode_inc_iversion(inode);
5698 /* the do_update_inode consumes one bh->b_count */
5701 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5702 err = ext4_do_update_inode(handle, inode, iloc);
5708 * On success, We end up with an outstanding reference count against
5709 * iloc->bh. This _must_ be cleaned up later.
5713 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5714 struct ext4_iloc *iloc)
5718 err = ext4_get_inode_loc(inode, iloc);
5720 BUFFER_TRACE(iloc->bh, "get_write_access");
5721 err = ext4_journal_get_write_access(handle, iloc->bh);
5727 ext4_std_error(inode->i_sb, err);
5732 * Expand an inode by new_extra_isize bytes.
5733 * Returns 0 on success or negative error number on failure.
5735 static int ext4_expand_extra_isize(struct inode *inode,
5736 unsigned int new_extra_isize,
5737 struct ext4_iloc iloc,
5740 struct ext4_inode *raw_inode;
5741 struct ext4_xattr_ibody_header *header;
5743 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5746 raw_inode = ext4_raw_inode(&iloc);
5748 header = IHDR(inode, raw_inode);
5750 /* No extended attributes present */
5751 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5752 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5753 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5755 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5759 /* try to expand with EAs present */
5760 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5765 * What we do here is to mark the in-core inode as clean with respect to inode
5766 * dirtiness (it may still be data-dirty).
5767 * This means that the in-core inode may be reaped by prune_icache
5768 * without having to perform any I/O. This is a very good thing,
5769 * because *any* task may call prune_icache - even ones which
5770 * have a transaction open against a different journal.
5772 * Is this cheating? Not really. Sure, we haven't written the
5773 * inode out, but prune_icache isn't a user-visible syncing function.
5774 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5775 * we start and wait on commits.
5777 * Is this efficient/effective? Well, we're being nice to the system
5778 * by cleaning up our inodes proactively so they can be reaped
5779 * without I/O. But we are potentially leaving up to five seconds'
5780 * worth of inodes floating about which prune_icache wants us to
5781 * write out. One way to fix that would be to get prune_icache()
5782 * to do a write_super() to free up some memory. It has the desired
5785 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5787 struct ext4_iloc iloc;
5788 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5789 static unsigned int mnt_count;
5793 err = ext4_reserve_inode_write(handle, inode, &iloc);
5794 if (ext4_handle_valid(handle) &&
5795 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5796 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5798 * We need extra buffer credits since we may write into EA block
5799 * with this same handle. If journal_extend fails, then it will
5800 * only result in a minor loss of functionality for that inode.
5801 * If this is felt to be critical, then e2fsck should be run to
5802 * force a large enough s_min_extra_isize.
5804 if ((jbd2_journal_extend(handle,
5805 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5806 ret = ext4_expand_extra_isize(inode,
5807 sbi->s_want_extra_isize,
5810 ext4_set_inode_state(inode,
5811 EXT4_STATE_NO_EXPAND);
5813 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5814 ext4_warning(inode->i_sb,
5815 "Unable to expand inode %lu. Delete"
5816 " some EAs or run e2fsck.",
5819 le16_to_cpu(sbi->s_es->s_mnt_count);
5825 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5830 * ext4_dirty_inode() is called from __mark_inode_dirty()
5832 * We're really interested in the case where a file is being extended.
5833 * i_size has been changed by generic_commit_write() and we thus need
5834 * to include the updated inode in the current transaction.
5836 * Also, dquot_alloc_block() will always dirty the inode when blocks
5837 * are allocated to the file.
5839 * If the inode is marked synchronous, we don't honour that here - doing
5840 * so would cause a commit on atime updates, which we don't bother doing.
5841 * We handle synchronous inodes at the highest possible level.
5843 void ext4_dirty_inode(struct inode *inode)
5847 handle = ext4_journal_start(inode, 2);
5851 ext4_mark_inode_dirty(handle, inode);
5853 ext4_journal_stop(handle);
5860 * Bind an inode's backing buffer_head into this transaction, to prevent
5861 * it from being flushed to disk early. Unlike
5862 * ext4_reserve_inode_write, this leaves behind no bh reference and
5863 * returns no iloc structure, so the caller needs to repeat the iloc
5864 * lookup to mark the inode dirty later.
5866 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5868 struct ext4_iloc iloc;
5872 err = ext4_get_inode_loc(inode, &iloc);
5874 BUFFER_TRACE(iloc.bh, "get_write_access");
5875 err = jbd2_journal_get_write_access(handle, iloc.bh);
5877 err = ext4_handle_dirty_metadata(handle,
5883 ext4_std_error(inode->i_sb, err);
5888 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5895 * We have to be very careful here: changing a data block's
5896 * journaling status dynamically is dangerous. If we write a
5897 * data block to the journal, change the status and then delete
5898 * that block, we risk forgetting to revoke the old log record
5899 * from the journal and so a subsequent replay can corrupt data.
5900 * So, first we make sure that the journal is empty and that
5901 * nobody is changing anything.
5904 journal = EXT4_JOURNAL(inode);
5907 if (is_journal_aborted(journal))
5910 jbd2_journal_lock_updates(journal);
5911 jbd2_journal_flush(journal);
5914 * OK, there are no updates running now, and all cached data is
5915 * synced to disk. We are now in a completely consistent state
5916 * which doesn't have anything in the journal, and we know that
5917 * no filesystem updates are running, so it is safe to modify
5918 * the inode's in-core data-journaling state flag now.
5922 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5924 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5925 ext4_set_aops(inode);
5927 jbd2_journal_unlock_updates(journal);
5929 /* Finally we can mark the inode as dirty. */
5931 handle = ext4_journal_start(inode, 1);
5933 return PTR_ERR(handle);
5935 err = ext4_mark_inode_dirty(handle, inode);
5936 ext4_handle_sync(handle);
5937 ext4_journal_stop(handle);
5938 ext4_std_error(inode->i_sb, err);
5943 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5945 return !buffer_mapped(bh);
5948 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5950 struct page *page = vmf->page;
5955 struct file *file = vma->vm_file;
5956 struct inode *inode = file->f_path.dentry->d_inode;
5957 struct address_space *mapping = inode->i_mapping;
5960 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5961 * get i_mutex because we are already holding mmap_sem.
5963 down_read(&inode->i_alloc_sem);
5964 size = i_size_read(inode);
5965 if (page->mapping != mapping || size <= page_offset(page)
5966 || !PageUptodate(page)) {
5967 /* page got truncated from under us? */
5971 if (PageMappedToDisk(page))
5974 if (page->index == size >> PAGE_CACHE_SHIFT)
5975 len = size & ~PAGE_CACHE_MASK;
5977 len = PAGE_CACHE_SIZE;
5981 * return if we have all the buffers mapped. This avoid
5982 * the need to call write_begin/write_end which does a
5983 * journal_start/journal_stop which can block and take
5986 if (page_has_buffers(page)) {
5987 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5988 ext4_bh_unmapped)) {
5995 * OK, we need to fill the hole... Do write_begin write_end
5996 * to do block allocation/reservation.We are not holding
5997 * inode.i__mutex here. That allow * parallel write_begin,
5998 * write_end call. lock_page prevent this from happening
5999 * on the same page though
6001 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
6002 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
6005 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
6006 len, len, page, fsdata);
6012 ret = VM_FAULT_SIGBUS;
6013 up_read(&inode->i_alloc_sem);