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/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
47 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode
)->jinode
,
51 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
54 * Test whether an inode is a fast symlink.
56 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
58 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
59 (inode
->i_sb
->s_blocksize
>> 9) : 0;
61 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
65 * The ext4 forget function must perform a revoke if we are freeing data
66 * which has been journaled. Metadata (eg. indirect blocks) must be
67 * revoked in all cases.
69 * "bh" may be NULL: a metadata block may have been freed from memory
70 * but there may still be a record of it in the journal, and that record
71 * still needs to be revoked.
73 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
74 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
80 BUFFER_TRACE(bh
, "enter");
82 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
84 bh
, is_metadata
, inode
->i_mode
,
85 test_opt(inode
->i_sb
, DATA_FLAGS
));
87 /* Never use the revoke function if we are doing full data
88 * journaling: there is no need to, and a V1 superblock won't
89 * support it. Otherwise, only skip the revoke on un-journaled
92 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
93 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
95 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
96 return ext4_journal_forget(handle
, bh
);
102 * data!=journal && (is_metadata || should_journal_data(inode))
104 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
105 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
107 ext4_abort(inode
->i_sb
, __func__
,
108 "error %d when attempting revoke", err
);
109 BUFFER_TRACE(bh
, "exit");
114 * Work out how many blocks we need to proceed with the next chunk of a
115 * truncate transaction.
117 static unsigned long blocks_for_truncate(struct inode
*inode
)
121 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
123 /* Give ourselves just enough room to cope with inodes in which
124 * i_blocks is corrupt: we've seen disk corruptions in the past
125 * which resulted in random data in an inode which looked enough
126 * like a regular file for ext4 to try to delete it. Things
127 * will go a bit crazy if that happens, but at least we should
128 * try not to panic the whole kernel. */
132 /* But we need to bound the transaction so we don't overflow the
134 if (needed
> EXT4_MAX_TRANS_DATA
)
135 needed
= EXT4_MAX_TRANS_DATA
;
137 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
141 * Truncate transactions can be complex and absolutely huge. So we need to
142 * be able to restart the transaction at a conventient checkpoint to make
143 * sure we don't overflow the journal.
145 * start_transaction gets us a new handle for a truncate transaction,
146 * and extend_transaction tries to extend the existing one a bit. If
147 * extend fails, we need to propagate the failure up and restart the
148 * transaction in the top-level truncate loop. --sct
150 static handle_t
*start_transaction(struct inode
*inode
)
154 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
158 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
163 * Try to extend this transaction for the purposes of truncation.
165 * Returns 0 if we managed to create more room. If we can't create more
166 * room, and the transaction must be restarted we return 1.
168 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
170 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
172 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
178 * Restart the transaction associated with *handle. This does a commit,
179 * so before we call here everything must be consistently dirtied against
182 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
184 jbd_debug(2, "restarting handle %p\n", handle
);
185 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
189 * Called at the last iput() if i_nlink is zero.
191 void ext4_delete_inode (struct inode
* inode
)
195 if (ext4_should_order_data(inode
))
196 ext4_begin_ordered_truncate(inode
, 0);
197 truncate_inode_pages(&inode
->i_data
, 0);
199 if (is_bad_inode(inode
))
202 handle
= start_transaction(inode
);
203 if (IS_ERR(handle
)) {
205 * If we're going to skip the normal cleanup, we still need to
206 * make sure that the in-core orphan linked list is properly
209 ext4_orphan_del(NULL
, inode
);
217 ext4_truncate(inode
);
219 * Kill off the orphan record which ext4_truncate created.
220 * AKPM: I think this can be inside the above `if'.
221 * Note that ext4_orphan_del() has to be able to cope with the
222 * deletion of a non-existent orphan - this is because we don't
223 * know if ext4_truncate() actually created an orphan record.
224 * (Well, we could do this if we need to, but heck - it works)
226 ext4_orphan_del(handle
, inode
);
227 EXT4_I(inode
)->i_dtime
= get_seconds();
230 * One subtle ordering requirement: if anything has gone wrong
231 * (transaction abort, IO errors, whatever), then we can still
232 * do these next steps (the fs will already have been marked as
233 * having errors), but we can't free the inode if the mark_dirty
236 if (ext4_mark_inode_dirty(handle
, inode
))
237 /* If that failed, just do the required in-core inode clear. */
240 ext4_free_inode(handle
, inode
);
241 ext4_journal_stop(handle
);
244 clear_inode(inode
); /* We must guarantee clearing of inode... */
250 struct buffer_head
*bh
;
253 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
255 p
->key
= *(p
->p
= v
);
260 * ext4_block_to_path - parse the block number into array of offsets
261 * @inode: inode in question (we are only interested in its superblock)
262 * @i_block: block number to be parsed
263 * @offsets: array to store the offsets in
264 * @boundary: set this non-zero if the referred-to block is likely to be
265 * followed (on disk) by an indirect block.
267 * To store the locations of file's data ext4 uses a data structure common
268 * for UNIX filesystems - tree of pointers anchored in the inode, with
269 * data blocks at leaves and indirect blocks in intermediate nodes.
270 * This function translates the block number into path in that tree -
271 * return value is the path length and @offsets[n] is the offset of
272 * pointer to (n+1)th node in the nth one. If @block is out of range
273 * (negative or too large) warning is printed and zero returned.
275 * Note: function doesn't find node addresses, so no IO is needed. All
276 * we need to know is the capacity of indirect blocks (taken from the
281 * Portability note: the last comparison (check that we fit into triple
282 * indirect block) is spelled differently, because otherwise on an
283 * architecture with 32-bit longs and 8Kb pages we might get into trouble
284 * if our filesystem had 8Kb blocks. We might use long long, but that would
285 * kill us on x86. Oh, well, at least the sign propagation does not matter -
286 * i_block would have to be negative in the very beginning, so we would not
290 static int ext4_block_to_path(struct inode
*inode
,
292 ext4_lblk_t offsets
[4], int *boundary
)
294 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
295 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
296 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
297 indirect_blocks
= ptrs
,
298 double_blocks
= (1 << (ptrs_bits
* 2));
303 ext4_warning (inode
->i_sb
, "ext4_block_to_path", "block < 0");
304 } else if (i_block
< direct_blocks
) {
305 offsets
[n
++] = i_block
;
306 final
= direct_blocks
;
307 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
308 offsets
[n
++] = EXT4_IND_BLOCK
;
309 offsets
[n
++] = i_block
;
311 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
312 offsets
[n
++] = EXT4_DIND_BLOCK
;
313 offsets
[n
++] = i_block
>> ptrs_bits
;
314 offsets
[n
++] = i_block
& (ptrs
- 1);
316 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
317 offsets
[n
++] = EXT4_TIND_BLOCK
;
318 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
319 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
320 offsets
[n
++] = i_block
& (ptrs
- 1);
323 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
325 i_block
+ direct_blocks
+
326 indirect_blocks
+ double_blocks
);
329 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
334 * ext4_get_branch - read the chain of indirect blocks leading to data
335 * @inode: inode in question
336 * @depth: depth of the chain (1 - direct pointer, etc.)
337 * @offsets: offsets of pointers in inode/indirect blocks
338 * @chain: place to store the result
339 * @err: here we store the error value
341 * Function fills the array of triples <key, p, bh> and returns %NULL
342 * if everything went OK or the pointer to the last filled triple
343 * (incomplete one) otherwise. Upon the return chain[i].key contains
344 * the number of (i+1)-th block in the chain (as it is stored in memory,
345 * i.e. little-endian 32-bit), chain[i].p contains the address of that
346 * number (it points into struct inode for i==0 and into the bh->b_data
347 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
348 * block for i>0 and NULL for i==0. In other words, it holds the block
349 * numbers of the chain, addresses they were taken from (and where we can
350 * verify that chain did not change) and buffer_heads hosting these
353 * Function stops when it stumbles upon zero pointer (absent block)
354 * (pointer to last triple returned, *@err == 0)
355 * or when it gets an IO error reading an indirect block
356 * (ditto, *@err == -EIO)
357 * or when it reads all @depth-1 indirect blocks successfully and finds
358 * the whole chain, all way to the data (returns %NULL, *err == 0).
360 * Need to be called with
361 * down_read(&EXT4_I(inode)->i_data_sem)
363 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
364 ext4_lblk_t
*offsets
,
365 Indirect chain
[4], int *err
)
367 struct super_block
*sb
= inode
->i_sb
;
369 struct buffer_head
*bh
;
372 /* i_data is not going away, no lock needed */
373 add_chain (chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
377 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
380 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
394 * ext4_find_near - find a place for allocation with sufficient locality
396 * @ind: descriptor of indirect block.
398 * This function returns the preferred place for block allocation.
399 * It is used when heuristic for sequential allocation fails.
401 * + if there is a block to the left of our position - allocate near it.
402 * + if pointer will live in indirect block - allocate near that block.
403 * + if pointer will live in inode - allocate in the same
406 * In the latter case we colour the starting block by the callers PID to
407 * prevent it from clashing with concurrent allocations for a different inode
408 * in the same block group. The PID is used here so that functionally related
409 * files will be close-by on-disk.
411 * Caller must make sure that @ind is valid and will stay that way.
413 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
415 struct ext4_inode_info
*ei
= EXT4_I(inode
);
416 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
418 ext4_fsblk_t bg_start
;
419 ext4_fsblk_t last_block
;
420 ext4_grpblk_t colour
;
422 /* Try to find previous block */
423 for (p
= ind
->p
- 1; p
>= start
; p
--) {
425 return le32_to_cpu(*p
);
428 /* No such thing, so let's try location of indirect block */
430 return ind
->bh
->b_blocknr
;
433 * It is going to be referred to from the inode itself? OK, just put it
434 * into the same cylinder group then.
436 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
437 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
439 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
440 colour
= (current
->pid
% 16) *
441 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
443 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
444 return bg_start
+ colour
;
448 * ext4_find_goal - find a preferred place for allocation.
450 * @block: block we want
451 * @partial: pointer to the last triple within a chain
453 * Normally this function find the preferred place for block allocation,
456 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
459 struct ext4_block_alloc_info
*block_i
;
461 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
464 * try the heuristic for sequential allocation,
465 * failing that at least try to get decent locality.
467 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
468 && (block_i
->last_alloc_physical_block
!= 0)) {
469 return block_i
->last_alloc_physical_block
+ 1;
472 return ext4_find_near(inode
, partial
);
476 * ext4_blks_to_allocate: Look up the block map and count the number
477 * of direct blocks need to be allocated for the given branch.
479 * @branch: chain of indirect blocks
480 * @k: number of blocks need for indirect blocks
481 * @blks: number of data blocks to be mapped.
482 * @blocks_to_boundary: the offset in the indirect block
484 * return the total number of blocks to be allocate, including the
485 * direct and indirect blocks.
487 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
488 int blocks_to_boundary
)
490 unsigned long count
= 0;
493 * Simple case, [t,d]Indirect block(s) has not allocated yet
494 * then it's clear blocks on that path have not allocated
497 /* right now we don't handle cross boundary allocation */
498 if (blks
< blocks_to_boundary
+ 1)
501 count
+= blocks_to_boundary
+ 1;
506 while (count
< blks
&& count
<= blocks_to_boundary
&&
507 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
514 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
515 * @indirect_blks: the number of blocks need to allocate for indirect
518 * @new_blocks: on return it will store the new block numbers for
519 * the indirect blocks(if needed) and the first direct block,
520 * @blks: on return it will store the total number of allocated
523 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
524 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
525 int indirect_blks
, int blks
,
526 ext4_fsblk_t new_blocks
[4], int *err
)
529 unsigned long count
= 0, blk_allocated
= 0;
531 ext4_fsblk_t current_block
= 0;
535 * Here we try to allocate the requested multiple blocks at once,
536 * on a best-effort basis.
537 * To build a branch, we should allocate blocks for
538 * the indirect blocks(if not allocated yet), and at least
539 * the first direct block of this branch. That's the
540 * minimum number of blocks need to allocate(required)
542 /* first we try to allocate the indirect blocks */
543 target
= indirect_blks
;
546 /* allocating blocks for indirect blocks and direct blocks */
547 current_block
= ext4_new_meta_blocks(handle
, inode
,
553 /* allocate blocks for indirect blocks */
554 while (index
< indirect_blks
&& count
) {
555 new_blocks
[index
++] = current_block
++;
560 * save the new block number
561 * for the first direct block
563 new_blocks
[index
] = current_block
;
564 printk(KERN_INFO
"%s returned more blocks than "
565 "requested\n", __func__
);
571 target
= blks
- count
;
572 blk_allocated
= count
;
575 /* Now allocate data blocks */
577 /* allocating blocks for data blocks */
578 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
580 if (*err
&& (target
== blks
)) {
582 * if the allocation failed and we didn't allocate
588 if (target
== blks
) {
590 * save the new block number
591 * for the first direct block
593 new_blocks
[index
] = current_block
;
595 blk_allocated
+= count
;
598 /* total number of blocks allocated for direct blocks */
603 for (i
= 0; i
<index
; i
++)
604 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
609 * ext4_alloc_branch - allocate and set up a chain of blocks.
611 * @indirect_blks: number of allocated indirect blocks
612 * @blks: number of allocated direct blocks
613 * @offsets: offsets (in the blocks) to store the pointers to next.
614 * @branch: place to store the chain in.
616 * This function allocates blocks, zeroes out all but the last one,
617 * links them into chain and (if we are synchronous) writes them to disk.
618 * In other words, it prepares a branch that can be spliced onto the
619 * inode. It stores the information about that chain in the branch[], in
620 * the same format as ext4_get_branch() would do. We are calling it after
621 * we had read the existing part of chain and partial points to the last
622 * triple of that (one with zero ->key). Upon the exit we have the same
623 * picture as after the successful ext4_get_block(), except that in one
624 * place chain is disconnected - *branch->p is still zero (we did not
625 * set the last link), but branch->key contains the number that should
626 * be placed into *branch->p to fill that gap.
628 * If allocation fails we free all blocks we've allocated (and forget
629 * their buffer_heads) and return the error value the from failed
630 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
631 * as described above and return 0.
633 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
634 ext4_lblk_t iblock
, int indirect_blks
,
635 int *blks
, ext4_fsblk_t goal
,
636 ext4_lblk_t
*offsets
, Indirect
*branch
)
638 int blocksize
= inode
->i_sb
->s_blocksize
;
641 struct buffer_head
*bh
;
643 ext4_fsblk_t new_blocks
[4];
644 ext4_fsblk_t current_block
;
646 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
647 *blks
, new_blocks
, &err
);
651 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
653 * metadata blocks and data blocks are allocated.
655 for (n
= 1; n
<= indirect_blks
; n
++) {
657 * Get buffer_head for parent block, zero it out
658 * and set the pointer to new one, then send
661 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
664 BUFFER_TRACE(bh
, "call get_create_access");
665 err
= ext4_journal_get_create_access(handle
, bh
);
672 memset(bh
->b_data
, 0, blocksize
);
673 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
674 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
675 *branch
[n
].p
= branch
[n
].key
;
676 if ( n
== indirect_blks
) {
677 current_block
= new_blocks
[n
];
679 * End of chain, update the last new metablock of
680 * the chain to point to the new allocated
681 * data blocks numbers
683 for (i
=1; i
< num
; i
++)
684 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
686 BUFFER_TRACE(bh
, "marking uptodate");
687 set_buffer_uptodate(bh
);
690 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
691 err
= ext4_journal_dirty_metadata(handle
, bh
);
698 /* Allocation failed, free what we already allocated */
699 for (i
= 1; i
<= n
; i
++) {
700 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
701 ext4_journal_forget(handle
, branch
[i
].bh
);
703 for (i
= 0; i
<indirect_blks
; i
++)
704 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
706 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
712 * ext4_splice_branch - splice the allocated branch onto inode.
714 * @block: (logical) number of block we are adding
715 * @chain: chain of indirect blocks (with a missing link - see
717 * @where: location of missing link
718 * @num: number of indirect blocks we are adding
719 * @blks: number of direct blocks we are adding
721 * This function fills the missing link and does all housekeeping needed in
722 * inode (->i_blocks, etc.). In case of success we end up with the full
723 * chain to new block and return 0.
725 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
726 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
730 struct ext4_block_alloc_info
*block_i
;
731 ext4_fsblk_t current_block
;
733 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
735 * If we're splicing into a [td]indirect block (as opposed to the
736 * inode) then we need to get write access to the [td]indirect block
740 BUFFER_TRACE(where
->bh
, "get_write_access");
741 err
= ext4_journal_get_write_access(handle
, where
->bh
);
747 *where
->p
= where
->key
;
750 * Update the host buffer_head or inode to point to more just allocated
751 * direct blocks blocks
753 if (num
== 0 && blks
> 1) {
754 current_block
= le32_to_cpu(where
->key
) + 1;
755 for (i
= 1; i
< blks
; i
++)
756 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
760 * update the most recently allocated logical & physical block
761 * in i_block_alloc_info, to assist find the proper goal block for next
765 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
766 block_i
->last_alloc_physical_block
=
767 le32_to_cpu(where
[num
].key
) + blks
- 1;
770 /* We are done with atomic stuff, now do the rest of housekeeping */
772 inode
->i_ctime
= ext4_current_time(inode
);
773 ext4_mark_inode_dirty(handle
, inode
);
775 /* had we spliced it onto indirect block? */
778 * If we spliced it onto an indirect block, we haven't
779 * altered the inode. Note however that if it is being spliced
780 * onto an indirect block at the very end of the file (the
781 * file is growing) then we *will* alter the inode to reflect
782 * the new i_size. But that is not done here - it is done in
783 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
785 jbd_debug(5, "splicing indirect only\n");
786 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
787 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
792 * OK, we spliced it into the inode itself on a direct block.
793 * Inode was dirtied above.
795 jbd_debug(5, "splicing direct\n");
800 for (i
= 1; i
<= num
; i
++) {
801 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
802 ext4_journal_forget(handle
, where
[i
].bh
);
803 ext4_free_blocks(handle
, inode
,
804 le32_to_cpu(where
[i
-1].key
), 1, 0);
806 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
812 * Allocation strategy is simple: if we have to allocate something, we will
813 * have to go the whole way to leaf. So let's do it before attaching anything
814 * to tree, set linkage between the newborn blocks, write them if sync is
815 * required, recheck the path, free and repeat if check fails, otherwise
816 * set the last missing link (that will protect us from any truncate-generated
817 * removals - all blocks on the path are immune now) and possibly force the
818 * write on the parent block.
819 * That has a nice additional property: no special recovery from the failed
820 * allocations is needed - we simply release blocks and do not touch anything
821 * reachable from inode.
823 * `handle' can be NULL if create == 0.
825 * return > 0, # of blocks mapped or allocated.
826 * return = 0, if plain lookup failed.
827 * return < 0, error case.
830 * Need to be called with
831 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
832 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
834 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
835 ext4_lblk_t iblock
, unsigned long maxblocks
,
836 struct buffer_head
*bh_result
,
837 int create
, int extend_disksize
)
840 ext4_lblk_t offsets
[4];
845 int blocks_to_boundary
= 0;
847 struct ext4_inode_info
*ei
= EXT4_I(inode
);
849 ext4_fsblk_t first_block
= 0;
853 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
854 J_ASSERT(handle
!= NULL
|| create
== 0);
855 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
856 &blocks_to_boundary
);
861 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
863 /* Simplest case - block found, no allocation needed */
865 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
866 clear_buffer_new(bh_result
);
869 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
872 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
874 if (blk
== first_block
+ count
)
882 /* Next simple case - plain lookup or failed read of indirect block */
883 if (!create
|| err
== -EIO
)
887 * Okay, we need to do block allocation. Lazily initialize the block
888 * allocation info here if necessary
890 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
891 ext4_init_block_alloc_info(inode
);
893 goal
= ext4_find_goal(inode
, iblock
, partial
);
895 /* the number of blocks need to allocate for [d,t]indirect blocks */
896 indirect_blks
= (chain
+ depth
) - partial
- 1;
899 * Next look up the indirect map to count the totoal number of
900 * direct blocks to allocate for this branch.
902 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
903 maxblocks
, blocks_to_boundary
);
905 * Block out ext4_truncate while we alter the tree
907 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
909 offsets
+ (partial
- chain
), partial
);
912 * The ext4_splice_branch call will free and forget any buffers
913 * on the new chain if there is a failure, but that risks using
914 * up transaction credits, especially for bitmaps where the
915 * credits cannot be returned. Can we handle this somehow? We
916 * may need to return -EAGAIN upwards in the worst case. --sct
919 err
= ext4_splice_branch(handle
, inode
, iblock
,
920 partial
, indirect_blks
, count
);
922 * i_disksize growing is protected by i_data_sem. Don't forget to
923 * protect it if you're about to implement concurrent
924 * ext4_get_block() -bzzz
926 if (!err
&& extend_disksize
) {
927 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
928 if (disksize
> i_size_read(inode
))
929 disksize
= i_size_read(inode
);
930 if (disksize
> ei
->i_disksize
)
931 ei
->i_disksize
= disksize
;
936 set_buffer_new(bh_result
);
938 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
939 if (count
> blocks_to_boundary
)
940 set_buffer_boundary(bh_result
);
942 /* Clean up and exit */
943 partial
= chain
+ depth
- 1; /* the whole chain */
945 while (partial
> chain
) {
946 BUFFER_TRACE(partial
->bh
, "call brelse");
950 BUFFER_TRACE(bh_result
, "returned");
955 /* Maximum number of blocks we map for direct IO at once. */
956 #define DIO_MAX_BLOCKS 4096
958 * Number of credits we need for writing DIO_MAX_BLOCKS:
959 * We need sb + group descriptor + bitmap + inode -> 4
960 * For B blocks with A block pointers per block we need:
961 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
962 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
964 #define DIO_CREDITS 25
970 * ext4_ext4 get_block() wrapper function
971 * It will do a look up first, and returns if the blocks already mapped.
972 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
973 * and store the allocated blocks in the result buffer head and mark it
976 * If file type is extents based, it will call ext4_ext_get_blocks(),
977 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
980 * On success, it returns the number of blocks being mapped or allocate.
981 * if create==0 and the blocks are pre-allocated and uninitialized block,
982 * the result buffer head is unmapped. If the create ==1, it will make sure
983 * the buffer head is mapped.
985 * It returns 0 if plain look up failed (blocks have not been allocated), in
986 * that casem, buffer head is unmapped
988 * It returns the error in case of allocation failure.
990 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
991 unsigned long max_blocks
, struct buffer_head
*bh
,
992 int create
, int extend_disksize
, int flag
)
996 clear_buffer_mapped(bh
);
999 * Try to see if we can get the block without requesting
1000 * for new file system block.
1002 down_read((&EXT4_I(inode
)->i_data_sem
));
1003 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1004 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1007 retval
= ext4_get_blocks_handle(handle
,
1008 inode
, block
, max_blocks
, bh
, 0, 0);
1010 up_read((&EXT4_I(inode
)->i_data_sem
));
1012 /* If it is only a block(s) look up */
1017 * Returns if the blocks have already allocated
1019 * Note that if blocks have been preallocated
1020 * ext4_ext_get_block() returns th create = 0
1021 * with buffer head unmapped.
1023 if (retval
> 0 && buffer_mapped(bh
))
1027 * New blocks allocate and/or writing to uninitialized extent
1028 * will possibly result in updating i_data, so we take
1029 * the write lock of i_data_sem, and call get_blocks()
1030 * with create == 1 flag.
1032 down_write((&EXT4_I(inode
)->i_data_sem
));
1035 * if the caller is from delayed allocation writeout path
1036 * we have already reserved fs blocks for allocation
1037 * let the underlying get_block() function know to
1038 * avoid double accounting
1041 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1043 * We need to check for EXT4 here because migrate
1044 * could have changed the inode type in between
1046 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1047 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1048 bh
, create
, extend_disksize
);
1050 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1051 max_blocks
, bh
, create
, extend_disksize
);
1053 if (retval
> 0 && buffer_new(bh
)) {
1055 * We allocated new blocks which will result in
1056 * i_data's format changing. Force the migrate
1057 * to fail by clearing migrate flags
1059 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1065 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1067 * Update reserved blocks/metadata blocks
1068 * after successful block allocation
1069 * which were deferred till now
1071 if ((retval
> 0) && buffer_delay(bh
))
1072 ext4_da_release_space(inode
, retval
, 0);
1075 up_write((&EXT4_I(inode
)->i_data_sem
));
1079 static int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1080 struct buffer_head
*bh_result
, int create
)
1082 handle_t
*handle
= ext4_journal_current_handle();
1083 int ret
= 0, started
= 0;
1084 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1086 if (create
&& !handle
) {
1087 /* Direct IO write... */
1088 if (max_blocks
> DIO_MAX_BLOCKS
)
1089 max_blocks
= DIO_MAX_BLOCKS
;
1090 handle
= ext4_journal_start(inode
, DIO_CREDITS
+
1091 2 * EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
));
1092 if (IS_ERR(handle
)) {
1093 ret
= PTR_ERR(handle
);
1099 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1100 max_blocks
, bh_result
, create
, 0, 0);
1102 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1106 ext4_journal_stop(handle
);
1112 * `handle' can be NULL if create is zero
1114 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1115 ext4_lblk_t block
, int create
, int *errp
)
1117 struct buffer_head dummy
;
1120 J_ASSERT(handle
!= NULL
|| create
== 0);
1123 dummy
.b_blocknr
= -1000;
1124 buffer_trace_init(&dummy
.b_history
);
1125 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1126 &dummy
, create
, 1, 0);
1128 * ext4_get_blocks_handle() returns number of blocks
1129 * mapped. 0 in case of a HOLE.
1137 if (!err
&& buffer_mapped(&dummy
)) {
1138 struct buffer_head
*bh
;
1139 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1144 if (buffer_new(&dummy
)) {
1145 J_ASSERT(create
!= 0);
1146 J_ASSERT(handle
!= NULL
);
1149 * Now that we do not always journal data, we should
1150 * keep in mind whether this should always journal the
1151 * new buffer as metadata. For now, regular file
1152 * writes use ext4_get_block instead, so it's not a
1156 BUFFER_TRACE(bh
, "call get_create_access");
1157 fatal
= ext4_journal_get_create_access(handle
, bh
);
1158 if (!fatal
&& !buffer_uptodate(bh
)) {
1159 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1160 set_buffer_uptodate(bh
);
1163 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1164 err
= ext4_journal_dirty_metadata(handle
, bh
);
1168 BUFFER_TRACE(bh
, "not a new buffer");
1181 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1182 ext4_lblk_t block
, int create
, int *err
)
1184 struct buffer_head
* bh
;
1186 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1189 if (buffer_uptodate(bh
))
1191 ll_rw_block(READ_META
, 1, &bh
);
1193 if (buffer_uptodate(bh
))
1200 static int walk_page_buffers( handle_t
*handle
,
1201 struct buffer_head
*head
,
1205 int (*fn
)( handle_t
*handle
,
1206 struct buffer_head
*bh
))
1208 struct buffer_head
*bh
;
1209 unsigned block_start
, block_end
;
1210 unsigned blocksize
= head
->b_size
;
1212 struct buffer_head
*next
;
1214 for ( bh
= head
, block_start
= 0;
1215 ret
== 0 && (bh
!= head
|| !block_start
);
1216 block_start
= block_end
, bh
= next
)
1218 next
= bh
->b_this_page
;
1219 block_end
= block_start
+ blocksize
;
1220 if (block_end
<= from
|| block_start
>= to
) {
1221 if (partial
&& !buffer_uptodate(bh
))
1225 err
= (*fn
)(handle
, bh
);
1233 * To preserve ordering, it is essential that the hole instantiation and
1234 * the data write be encapsulated in a single transaction. We cannot
1235 * close off a transaction and start a new one between the ext4_get_block()
1236 * and the commit_write(). So doing the jbd2_journal_start at the start of
1237 * prepare_write() is the right place.
1239 * Also, this function can nest inside ext4_writepage() ->
1240 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1241 * has generated enough buffer credits to do the whole page. So we won't
1242 * block on the journal in that case, which is good, because the caller may
1245 * By accident, ext4 can be reentered when a transaction is open via
1246 * quota file writes. If we were to commit the transaction while thus
1247 * reentered, there can be a deadlock - we would be holding a quota
1248 * lock, and the commit would never complete if another thread had a
1249 * transaction open and was blocking on the quota lock - a ranking
1252 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1253 * will _not_ run commit under these circumstances because handle->h_ref
1254 * is elevated. We'll still have enough credits for the tiny quotafile
1257 static int do_journal_get_write_access(handle_t
*handle
,
1258 struct buffer_head
*bh
)
1260 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1262 return ext4_journal_get_write_access(handle
, bh
);
1265 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1266 loff_t pos
, unsigned len
, unsigned flags
,
1267 struct page
**pagep
, void **fsdata
)
1269 struct inode
*inode
= mapping
->host
;
1270 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1277 index
= pos
>> PAGE_CACHE_SHIFT
;
1278 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1282 handle
= ext4_journal_start(inode
, needed_blocks
);
1283 if (IS_ERR(handle
)) {
1284 ret
= PTR_ERR(handle
);
1288 page
= __grab_cache_page(mapping
, index
);
1290 ext4_journal_stop(handle
);
1296 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1299 if (!ret
&& ext4_should_journal_data(inode
)) {
1300 ret
= walk_page_buffers(handle
, page_buffers(page
),
1301 from
, to
, NULL
, do_journal_get_write_access
);
1306 ext4_journal_stop(handle
);
1307 page_cache_release(page
);
1310 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1316 /* For write_end() in data=journal mode */
1317 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1319 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1321 set_buffer_uptodate(bh
);
1322 return ext4_journal_dirty_metadata(handle
, bh
);
1326 * We need to pick up the new inode size which generic_commit_write gave us
1327 * `file' can be NULL - eg, when called from page_symlink().
1329 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1330 * buffers are managed internally.
1332 static int ext4_ordered_write_end(struct file
*file
,
1333 struct address_space
*mapping
,
1334 loff_t pos
, unsigned len
, unsigned copied
,
1335 struct page
*page
, void *fsdata
)
1337 handle_t
*handle
= ext4_journal_current_handle();
1338 struct inode
*inode
= mapping
->host
;
1342 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1345 ret
= ext4_jbd2_file_inode(handle
, inode
);
1349 * generic_write_end() will run mark_inode_dirty() if i_size
1350 * changes. So let's piggyback the i_disksize mark_inode_dirty
1355 new_i_size
= pos
+ copied
;
1356 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1357 EXT4_I(inode
)->i_disksize
= new_i_size
;
1358 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1364 ret2
= ext4_journal_stop(handle
);
1368 return ret
? ret
: copied
;
1371 static int ext4_writeback_write_end(struct file
*file
,
1372 struct address_space
*mapping
,
1373 loff_t pos
, unsigned len
, unsigned copied
,
1374 struct page
*page
, void *fsdata
)
1376 handle_t
*handle
= ext4_journal_current_handle();
1377 struct inode
*inode
= mapping
->host
;
1381 new_i_size
= pos
+ copied
;
1382 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1383 EXT4_I(inode
)->i_disksize
= new_i_size
;
1385 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1391 ret2
= ext4_journal_stop(handle
);
1395 return ret
? ret
: copied
;
1398 static int ext4_journalled_write_end(struct file
*file
,
1399 struct address_space
*mapping
,
1400 loff_t pos
, unsigned len
, unsigned copied
,
1401 struct page
*page
, void *fsdata
)
1403 handle_t
*handle
= ext4_journal_current_handle();
1404 struct inode
*inode
= mapping
->host
;
1409 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1413 if (!PageUptodate(page
))
1415 page_zero_new_buffers(page
, from
+copied
, to
);
1418 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1419 to
, &partial
, write_end_fn
);
1421 SetPageUptodate(page
);
1422 if (pos
+copied
> inode
->i_size
)
1423 i_size_write(inode
, pos
+copied
);
1424 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1425 if (inode
->i_size
> EXT4_I(inode
)->i_disksize
) {
1426 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
1427 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1433 ret2
= ext4_journal_stop(handle
);
1436 page_cache_release(page
);
1438 return ret
? ret
: copied
;
1441 * Calculate the number of metadata blocks need to reserve
1442 * to allocate @blocks for non extent file based file
1444 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
1446 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1447 int ind_blks
, dind_blks
, tind_blks
;
1449 /* number of new indirect blocks needed */
1450 ind_blks
= (blocks
+ icap
- 1) / icap
;
1452 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1456 return ind_blks
+ dind_blks
+ tind_blks
;
1460 * Calculate the number of metadata blocks need to reserve
1461 * to allocate given number of blocks
1463 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1465 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1466 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1468 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1471 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1473 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1474 unsigned long md_needed
, mdblocks
, total
= 0;
1477 * recalculate the amount of metadata blocks to reserve
1478 * in order to allocate nrblocks
1479 * worse case is one extent per block
1481 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1482 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1483 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1484 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1486 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1487 total
= md_needed
+ nrblocks
;
1489 if (ext4_has_free_blocks(sbi
, total
) < total
) {
1490 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1494 /* reduce fs free blocks counter */
1495 percpu_counter_sub(&sbi
->s_freeblocks_counter
, total
);
1497 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1498 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1500 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1501 return 0; /* success */
1504 void ext4_da_release_space(struct inode
*inode
, int used
, int to_free
)
1506 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1507 int total
, mdb
, mdb_free
, release
;
1509 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1510 /* recalculate the number of metablocks still need to be reserved */
1511 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
- to_free
;
1512 mdb
= ext4_calc_metadata_amount(inode
, total
);
1514 /* figure out how many metablocks to release */
1515 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1516 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1518 /* Account for allocated meta_blocks */
1519 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1521 release
= to_free
+ mdb_free
;
1523 /* update fs free blocks counter for truncate case */
1524 percpu_counter_add(&sbi
->s_freeblocks_counter
, release
);
1526 /* update per-inode reservations */
1527 BUG_ON(used
+ to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1528 EXT4_I(inode
)->i_reserved_data_blocks
-= (used
+ to_free
);
1530 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1531 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1532 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1533 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1536 static void ext4_da_page_release_reservation(struct page
*page
,
1537 unsigned long offset
)
1540 struct buffer_head
*head
, *bh
;
1541 unsigned int curr_off
= 0;
1543 head
= page_buffers(page
);
1546 unsigned int next_off
= curr_off
+ bh
->b_size
;
1548 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1550 clear_buffer_delay(bh
);
1552 curr_off
= next_off
;
1553 } while ((bh
= bh
->b_this_page
) != head
);
1554 ext4_da_release_space(page
->mapping
->host
, 0, to_release
);
1558 * Delayed allocation stuff
1561 struct mpage_da_data
{
1562 struct inode
*inode
;
1563 struct buffer_head lbh
; /* extent of blocks */
1564 unsigned long first_page
, next_page
; /* extent of pages */
1565 get_block_t
*get_block
;
1566 struct writeback_control
*wbc
;
1570 * mpage_da_submit_io - walks through extent of pages and try to write
1571 * them with __mpage_writepage()
1573 * @mpd->inode: inode
1574 * @mpd->first_page: first page of the extent
1575 * @mpd->next_page: page after the last page of the extent
1576 * @mpd->get_block: the filesystem's block mapper function
1578 * By the time mpage_da_submit_io() is called we expect all blocks
1579 * to be allocated. this may be wrong if allocation failed.
1581 * As pages are already locked by write_cache_pages(), we can't use it
1583 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1585 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
1586 struct mpage_data mpd_pp
= {
1588 .last_block_in_bio
= 0,
1589 .get_block
= mpd
->get_block
,
1592 int ret
= 0, err
, nr_pages
, i
;
1593 unsigned long index
, end
;
1594 struct pagevec pvec
;
1596 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1598 pagevec_init(&pvec
, 0);
1599 index
= mpd
->first_page
;
1600 end
= mpd
->next_page
- 1;
1602 while (index
<= end
) {
1603 /* XXX: optimize tail */
1604 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1607 for (i
= 0; i
< nr_pages
; i
++) {
1608 struct page
*page
= pvec
.pages
[i
];
1610 index
= page
->index
;
1615 err
= __mpage_writepage(page
, mpd
->wbc
, &mpd_pp
);
1618 * In error case, we have to continue because
1619 * remaining pages are still locked
1620 * XXX: unlock and re-dirty them?
1625 pagevec_release(&pvec
);
1628 mpage_bio_submit(WRITE
, mpd_pp
.bio
);
1634 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1636 * @mpd->inode - inode to walk through
1637 * @exbh->b_blocknr - first block on a disk
1638 * @exbh->b_size - amount of space in bytes
1639 * @logical - first logical block to start assignment with
1641 * the function goes through all passed space and put actual disk
1642 * block numbers into buffer heads, dropping BH_Delay
1644 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1645 struct buffer_head
*exbh
)
1647 struct inode
*inode
= mpd
->inode
;
1648 struct address_space
*mapping
= inode
->i_mapping
;
1649 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1650 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1651 struct buffer_head
*head
, *bh
;
1652 unsigned long index
, end
;
1653 struct pagevec pvec
;
1656 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1657 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1658 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1660 pagevec_init(&pvec
, 0);
1662 while (index
<= end
) {
1663 /* XXX: optimize tail */
1664 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1667 for (i
= 0; i
< nr_pages
; i
++) {
1668 struct page
*page
= pvec
.pages
[i
];
1670 index
= page
->index
;
1675 BUG_ON(!PageLocked(page
));
1676 BUG_ON(PageWriteback(page
));
1677 BUG_ON(!page_has_buffers(page
));
1679 bh
= page_buffers(page
);
1682 /* skip blocks out of the range */
1684 if (cur_logical
>= logical
)
1687 } while ((bh
= bh
->b_this_page
) != head
);
1690 if (cur_logical
>= logical
+ blocks
)
1692 if (buffer_delay(bh
)) {
1693 bh
->b_blocknr
= pblock
;
1694 clear_buffer_delay(bh
);
1695 } else if (buffer_mapped(bh
))
1696 BUG_ON(bh
->b_blocknr
!= pblock
);
1700 } while ((bh
= bh
->b_this_page
) != head
);
1702 pagevec_release(&pvec
);
1708 * __unmap_underlying_blocks - just a helper function to unmap
1709 * set of blocks described by @bh
1711 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1712 struct buffer_head
*bh
)
1714 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1717 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1718 for (i
= 0; i
< blocks
; i
++)
1719 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1723 * mpage_da_map_blocks - go through given space
1725 * @mpd->lbh - bh describing space
1726 * @mpd->get_block - the filesystem's block mapper function
1728 * The function skips space we know is already mapped to disk blocks.
1730 * The function ignores errors ->get_block() returns, thus real
1731 * error handling is postponed to __mpage_writepage()
1733 static void mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1735 struct buffer_head
*lbh
= &mpd
->lbh
;
1736 int err
= 0, remain
= lbh
->b_size
;
1737 sector_t next
= lbh
->b_blocknr
;
1738 struct buffer_head
new;
1741 * We consider only non-mapped and non-allocated blocks
1743 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1747 new.b_state
= lbh
->b_state
;
1749 new.b_size
= remain
;
1750 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1753 * Rather than implement own error handling
1754 * here, we just leave remaining blocks
1755 * unallocated and try again with ->writepage()
1759 BUG_ON(new.b_size
== 0);
1761 if (buffer_new(&new))
1762 __unmap_underlying_blocks(mpd
->inode
, &new);
1765 * If blocks are delayed marked, we need to
1766 * put actual blocknr and drop delayed bit
1768 if (buffer_delay(lbh
))
1769 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1771 /* go for the remaining blocks */
1772 next
+= new.b_size
>> mpd
->inode
->i_blkbits
;
1773 remain
-= new.b_size
;
1777 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | (1 << BH_Delay))
1780 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1782 * @mpd->lbh - extent of blocks
1783 * @logical - logical number of the block in the file
1784 * @bh - bh of the block (used to access block's state)
1786 * the function is used to collect contig. blocks in same state
1788 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1789 sector_t logical
, struct buffer_head
*bh
)
1791 struct buffer_head
*lbh
= &mpd
->lbh
;
1794 next
= lbh
->b_blocknr
+ (lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1797 * First block in the extent
1799 if (lbh
->b_size
== 0) {
1800 lbh
->b_blocknr
= logical
;
1801 lbh
->b_size
= bh
->b_size
;
1802 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1807 * Can we merge the block to our big extent?
1809 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1810 lbh
->b_size
+= bh
->b_size
;
1815 * We couldn't merge the block to our extent, so we
1816 * need to flush current extent and start new one
1818 mpage_da_map_blocks(mpd
);
1821 * Now start a new extent
1823 lbh
->b_size
= bh
->b_size
;
1824 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1825 lbh
->b_blocknr
= logical
;
1829 * __mpage_da_writepage - finds extent of pages and blocks
1831 * @page: page to consider
1832 * @wbc: not used, we just follow rules
1835 * The function finds extents of pages and scan them for all blocks.
1837 static int __mpage_da_writepage(struct page
*page
,
1838 struct writeback_control
*wbc
, void *data
)
1840 struct mpage_da_data
*mpd
= data
;
1841 struct inode
*inode
= mpd
->inode
;
1842 struct buffer_head
*bh
, *head
, fake
;
1846 * Can we merge this page to current extent?
1848 if (mpd
->next_page
!= page
->index
) {
1850 * Nope, we can't. So, we map non-allocated blocks
1851 * and start IO on them using __mpage_writepage()
1853 if (mpd
->next_page
!= mpd
->first_page
) {
1854 mpage_da_map_blocks(mpd
);
1855 mpage_da_submit_io(mpd
);
1859 * Start next extent of pages ...
1861 mpd
->first_page
= page
->index
;
1866 mpd
->lbh
.b_size
= 0;
1867 mpd
->lbh
.b_state
= 0;
1868 mpd
->lbh
.b_blocknr
= 0;
1871 mpd
->next_page
= page
->index
+ 1;
1872 logical
= (sector_t
) page
->index
<<
1873 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1875 if (!page_has_buffers(page
)) {
1877 * There is no attached buffer heads yet (mmap?)
1878 * we treat the page asfull of dirty blocks
1881 bh
->b_size
= PAGE_CACHE_SIZE
;
1883 set_buffer_dirty(bh
);
1884 set_buffer_uptodate(bh
);
1885 mpage_add_bh_to_extent(mpd
, logical
, bh
);
1888 * Page with regular buffer heads, just add all dirty ones
1890 head
= page_buffers(page
);
1893 BUG_ON(buffer_locked(bh
));
1894 if (buffer_dirty(bh
))
1895 mpage_add_bh_to_extent(mpd
, logical
, bh
);
1897 } while ((bh
= bh
->b_this_page
) != head
);
1904 * mpage_da_writepages - walk the list of dirty pages of the given
1905 * address space, allocates non-allocated blocks, maps newly-allocated
1906 * blocks to existing bhs and issue IO them
1908 * @mapping: address space structure to write
1909 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1910 * @get_block: the filesystem's block mapper function.
1912 * This is a library function, which implements the writepages()
1913 * address_space_operation.
1915 * In order to avoid duplication of logic that deals with partial pages,
1916 * multiple bio per page, etc, we find non-allocated blocks, allocate
1917 * them with minimal calls to ->get_block() and re-use __mpage_writepage()
1919 * It's important that we call __mpage_writepage() only once for each
1920 * involved page, otherwise we'd have to implement more complicated logic
1921 * to deal with pages w/o PG_lock or w/ PG_writeback and so on.
1923 * See comments to mpage_writepages()
1925 static int mpage_da_writepages(struct address_space
*mapping
,
1926 struct writeback_control
*wbc
,
1927 get_block_t get_block
)
1929 struct mpage_da_data mpd
;
1933 return generic_writepages(mapping
, wbc
);
1936 mpd
.inode
= mapping
->host
;
1938 mpd
.lbh
.b_state
= 0;
1939 mpd
.lbh
.b_blocknr
= 0;
1942 mpd
.get_block
= get_block
;
1944 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, &mpd
);
1947 * Handle last extent of pages
1949 if (mpd
.next_page
!= mpd
.first_page
) {
1950 mpage_da_map_blocks(&mpd
);
1951 mpage_da_submit_io(&mpd
);
1958 * this is a special callback for ->write_begin() only
1959 * it's intention is to return mapped block or reserve space
1961 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1962 struct buffer_head
*bh_result
, int create
)
1966 BUG_ON(create
== 0);
1967 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
1970 * first, we need to know whether the block is allocated already
1971 * preallocated blocks are unmapped but should treated
1972 * the same as allocated blocks.
1974 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
1975 if ((ret
== 0) && !buffer_delay(bh_result
)) {
1976 /* the block isn't (pre)allocated yet, let's reserve space */
1978 * XXX: __block_prepare_write() unmaps passed block,
1981 ret
= ext4_da_reserve_space(inode
, 1);
1983 /* not enough space to reserve */
1986 map_bh(bh_result
, inode
->i_sb
, 0);
1987 set_buffer_new(bh_result
);
1988 set_buffer_delay(bh_result
);
1989 } else if (ret
> 0) {
1990 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1996 #define EXT4_DELALLOC_RSVED 1
1997 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
1998 struct buffer_head
*bh_result
, int create
)
2001 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2002 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2003 handle_t
*handle
= NULL
;
2005 handle
= ext4_journal_current_handle();
2006 BUG_ON(handle
== NULL
);
2007 BUG_ON(create
== 0);
2009 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2010 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2012 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2015 * Update on-disk size along with block allocation
2016 * we don't use 'extend_disksize' as size may change
2017 * within already allocated block -bzzz
2019 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2020 if (disksize
> i_size_read(inode
))
2021 disksize
= i_size_read(inode
);
2022 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2024 * XXX: replace with spinlock if seen contended -bzzz
2026 down_write(&EXT4_I(inode
)->i_data_sem
);
2027 if (disksize
> EXT4_I(inode
)->i_disksize
)
2028 EXT4_I(inode
)->i_disksize
= disksize
;
2029 up_write(&EXT4_I(inode
)->i_data_sem
);
2031 if (EXT4_I(inode
)->i_disksize
== disksize
) {
2032 ret
= ext4_mark_inode_dirty(handle
, inode
);
2041 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2043 return !buffer_mapped(bh
) || buffer_delay(bh
);
2047 * get called vi ext4_da_writepages after taking page lock
2048 * We may end up doing block allocation here in case
2049 * mpage_da_map_blocks failed to allocate blocks.
2051 * We also get called via journal_submit_inode_data_buffers
2053 static int ext4_da_writepage(struct page
*page
,
2054 struct writeback_control
*wbc
)
2059 handle_t
*handle
= NULL
;
2060 struct buffer_head
*page_bufs
;
2061 struct inode
*inode
= page
->mapping
->host
;
2063 handle
= ext4_journal_current_handle();
2066 * This can happen when we aren't called via
2067 * ext4_da_writepages() but directly (shrink_page_list).
2068 * We cannot easily start a transaction here so we just skip
2069 * writing the page in case we would have to do so.
2070 * We reach here also via journal_submit_inode_data_buffers
2072 size
= i_size_read(inode
);
2074 page_bufs
= page_buffers(page
);
2075 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2076 len
= size
& ~PAGE_CACHE_MASK
;
2078 len
= PAGE_CACHE_SIZE
;
2080 if (walk_page_buffers(NULL
, page_bufs
, 0,
2081 len
, NULL
, ext4_bh_unmapped_or_delay
)) {
2083 * We can't do block allocation under
2084 * page lock without a handle . So redirty
2085 * the page and return
2086 * We may reach here when we do a journal commit
2087 * via journal_submit_inode_data_buffers.
2088 * If we don't have mapping block we just ignore
2091 redirty_page_for_writepage(wbc
, page
);
2097 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2098 ret
= nobh_writepage(page
, ext4_da_get_block_write
, wbc
);
2100 ret
= block_write_full_page(page
, ext4_da_get_block_write
, wbc
);
2106 * For now just follow the DIO way to estimate the max credits
2107 * needed to write out EXT4_MAX_WRITEBACK_PAGES.
2108 * todo: need to calculate the max credits need for
2109 * extent based files, currently the DIO credits is based on
2110 * indirect-blocks mapping way.
2112 * Probably should have a generic way to calculate credits
2113 * for DIO, writepages, and truncate
2115 #define EXT4_MAX_WRITEBACK_PAGES DIO_MAX_BLOCKS
2116 #define EXT4_MAX_WRITEBACK_CREDITS DIO_CREDITS
2118 static int ext4_da_writepages(struct address_space
*mapping
,
2119 struct writeback_control
*wbc
)
2121 struct inode
*inode
= mapping
->host
;
2122 handle_t
*handle
= NULL
;
2126 loff_t range_start
= 0;
2129 * No pages to write? This is mainly a kludge to avoid starting
2130 * a transaction for special inodes like journal inode on last iput()
2131 * because that could violate lock ordering on umount
2133 if (!mapping
->nrpages
)
2137 * Estimate the worse case needed credits to write out
2138 * EXT4_MAX_BUF_BLOCKS pages
2140 needed_blocks
= EXT4_MAX_WRITEBACK_CREDITS
;
2142 to_write
= wbc
->nr_to_write
;
2143 if (!wbc
->range_cyclic
) {
2145 * If range_cyclic is not set force range_cont
2146 * and save the old writeback_index
2148 wbc
->range_cont
= 1;
2149 range_start
= wbc
->range_start
;
2152 while (!ret
&& to_write
) {
2153 /* start a new transaction*/
2154 handle
= ext4_journal_start(inode
, needed_blocks
);
2155 if (IS_ERR(handle
)) {
2156 ret
= PTR_ERR(handle
);
2157 goto out_writepages
;
2159 if (ext4_should_order_data(inode
)) {
2161 * With ordered mode we need to add
2162 * the inode to the journal handle
2163 * when we do block allocation.
2165 ret
= ext4_jbd2_file_inode(handle
, inode
);
2167 ext4_journal_stop(handle
);
2168 goto out_writepages
;
2173 * set the max dirty pages could be write at a time
2174 * to fit into the reserved transaction credits
2176 if (wbc
->nr_to_write
> EXT4_MAX_WRITEBACK_PAGES
)
2177 wbc
->nr_to_write
= EXT4_MAX_WRITEBACK_PAGES
;
2179 to_write
-= wbc
->nr_to_write
;
2180 ret
= mpage_da_writepages(mapping
, wbc
,
2181 ext4_da_get_block_write
);
2182 ext4_journal_stop(handle
);
2183 if (wbc
->nr_to_write
) {
2185 * There is no more writeout needed
2186 * or we requested for a noblocking writeout
2187 * and we found the device congested
2189 to_write
+= wbc
->nr_to_write
;
2192 wbc
->nr_to_write
= to_write
;
2196 wbc
->nr_to_write
= to_write
;
2198 wbc
->range_start
= range_start
;
2202 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2203 loff_t pos
, unsigned len
, unsigned flags
,
2204 struct page
**pagep
, void **fsdata
)
2206 int ret
, retries
= 0;
2210 struct inode
*inode
= mapping
->host
;
2213 index
= pos
>> PAGE_CACHE_SHIFT
;
2214 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2219 * With delayed allocation, we don't log the i_disksize update
2220 * if there is delayed block allocation. But we still need
2221 * to journalling the i_disksize update if writes to the end
2222 * of file which has an already mapped buffer.
2224 handle
= ext4_journal_start(inode
, 1);
2225 if (IS_ERR(handle
)) {
2226 ret
= PTR_ERR(handle
);
2230 page
= __grab_cache_page(mapping
, index
);
2235 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2236 ext4_da_get_block_prep
);
2239 ext4_journal_stop(handle
);
2240 page_cache_release(page
);
2243 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2249 static int ext4_da_write_end(struct file
*file
,
2250 struct address_space
*mapping
,
2251 loff_t pos
, unsigned len
, unsigned copied
,
2252 struct page
*page
, void *fsdata
)
2254 struct inode
*inode
= mapping
->host
;
2256 handle_t
*handle
= ext4_journal_current_handle();
2260 * generic_write_end() will run mark_inode_dirty() if i_size
2261 * changes. So let's piggyback the i_disksize mark_inode_dirty
2265 new_i_size
= pos
+ copied
;
2266 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2267 if (!walk_page_buffers(NULL
, page_buffers(page
),
2268 0, len
, NULL
, ext4_bh_unmapped_or_delay
)){
2270 * Updating i_disksize when extending file without
2271 * needing block allocation
2273 if (ext4_should_order_data(inode
))
2274 ret
= ext4_jbd2_file_inode(handle
, inode
);
2276 EXT4_I(inode
)->i_disksize
= new_i_size
;
2278 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2283 ret2
= ext4_journal_stop(handle
);
2287 return ret
? ret
: copied
;
2290 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2293 * Drop reserved blocks
2295 BUG_ON(!PageLocked(page
));
2296 if (!page_has_buffers(page
))
2299 ext4_da_page_release_reservation(page
, offset
);
2302 ext4_invalidatepage(page
, offset
);
2309 * bmap() is special. It gets used by applications such as lilo and by
2310 * the swapper to find the on-disk block of a specific piece of data.
2312 * Naturally, this is dangerous if the block concerned is still in the
2313 * journal. If somebody makes a swapfile on an ext4 data-journaling
2314 * filesystem and enables swap, then they may get a nasty shock when the
2315 * data getting swapped to that swapfile suddenly gets overwritten by
2316 * the original zero's written out previously to the journal and
2317 * awaiting writeback in the kernel's buffer cache.
2319 * So, if we see any bmap calls here on a modified, data-journaled file,
2320 * take extra steps to flush any blocks which might be in the cache.
2322 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2324 struct inode
*inode
= mapping
->host
;
2328 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2329 test_opt(inode
->i_sb
, DELALLOC
)) {
2331 * With delalloc we want to sync the file
2332 * so that we can make sure we allocate
2335 filemap_write_and_wait(mapping
);
2338 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2340 * This is a REALLY heavyweight approach, but the use of
2341 * bmap on dirty files is expected to be extremely rare:
2342 * only if we run lilo or swapon on a freshly made file
2343 * do we expect this to happen.
2345 * (bmap requires CAP_SYS_RAWIO so this does not
2346 * represent an unprivileged user DOS attack --- we'd be
2347 * in trouble if mortal users could trigger this path at
2350 * NB. EXT4_STATE_JDATA is not set on files other than
2351 * regular files. If somebody wants to bmap a directory
2352 * or symlink and gets confused because the buffer
2353 * hasn't yet been flushed to disk, they deserve
2354 * everything they get.
2357 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2358 journal
= EXT4_JOURNAL(inode
);
2359 jbd2_journal_lock_updates(journal
);
2360 err
= jbd2_journal_flush(journal
);
2361 jbd2_journal_unlock_updates(journal
);
2367 return generic_block_bmap(mapping
,block
,ext4_get_block
);
2370 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2376 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2383 * Note that we don't need to start a transaction unless we're journaling data
2384 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2385 * need to file the inode to the transaction's list in ordered mode because if
2386 * we are writing back data added by write(), the inode is already there and if
2387 * we are writing back data modified via mmap(), noone guarantees in which
2388 * transaction the data will hit the disk. In case we are journaling data, we
2389 * cannot start transaction directly because transaction start ranks above page
2390 * lock so we have to do some magic.
2392 * In all journaling modes block_write_full_page() will start the I/O.
2396 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2401 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2403 * Same applies to ext4_get_block(). We will deadlock on various things like
2404 * lock_journal and i_data_sem
2406 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2409 * 16May01: If we're reentered then journal_current_handle() will be
2410 * non-zero. We simply *return*.
2412 * 1 July 2001: @@@ FIXME:
2413 * In journalled data mode, a data buffer may be metadata against the
2414 * current transaction. But the same file is part of a shared mapping
2415 * and someone does a writepage() on it.
2417 * We will move the buffer onto the async_data list, but *after* it has
2418 * been dirtied. So there's a small window where we have dirty data on
2421 * Note that this only applies to the last partial page in the file. The
2422 * bit which block_write_full_page() uses prepare/commit for. (That's
2423 * broken code anyway: it's wrong for msync()).
2425 * It's a rare case: affects the final partial page, for journalled data
2426 * where the file is subject to bith write() and writepage() in the same
2427 * transction. To fix it we'll need a custom block_write_full_page().
2428 * We'll probably need that anyway for journalling writepage() output.
2430 * We don't honour synchronous mounts for writepage(). That would be
2431 * disastrous. Any write() or metadata operation will sync the fs for
2435 static int __ext4_normal_writepage(struct page
*page
,
2436 struct writeback_control
*wbc
)
2438 struct inode
*inode
= page
->mapping
->host
;
2440 if (test_opt(inode
->i_sb
, NOBH
))
2441 return nobh_writepage(page
, ext4_get_block
, wbc
);
2443 return block_write_full_page(page
, ext4_get_block
, wbc
);
2447 static int ext4_normal_writepage(struct page
*page
,
2448 struct writeback_control
*wbc
)
2450 struct inode
*inode
= page
->mapping
->host
;
2451 loff_t size
= i_size_read(inode
);
2454 J_ASSERT(PageLocked(page
));
2455 J_ASSERT(page_has_buffers(page
));
2456 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2457 len
= size
& ~PAGE_CACHE_MASK
;
2459 len
= PAGE_CACHE_SIZE
;
2460 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2461 ext4_bh_unmapped_or_delay
));
2463 if (!ext4_journal_current_handle())
2464 return __ext4_normal_writepage(page
, wbc
);
2466 redirty_page_for_writepage(wbc
, page
);
2471 static int __ext4_journalled_writepage(struct page
*page
,
2472 struct writeback_control
*wbc
)
2474 struct address_space
*mapping
= page
->mapping
;
2475 struct inode
*inode
= mapping
->host
;
2476 struct buffer_head
*page_bufs
;
2477 handle_t
*handle
= NULL
;
2481 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
, ext4_get_block
);
2485 page_bufs
= page_buffers(page
);
2486 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2488 /* As soon as we unlock the page, it can go away, but we have
2489 * references to buffers so we are safe */
2492 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2493 if (IS_ERR(handle
)) {
2494 ret
= PTR_ERR(handle
);
2498 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2499 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2501 err
= walk_page_buffers(handle
, page_bufs
, 0,
2502 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2505 err
= ext4_journal_stop(handle
);
2509 walk_page_buffers(handle
, page_bufs
, 0,
2510 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2511 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2520 static int ext4_journalled_writepage(struct page
*page
,
2521 struct writeback_control
*wbc
)
2523 struct inode
*inode
= page
->mapping
->host
;
2524 loff_t size
= i_size_read(inode
);
2527 J_ASSERT(PageLocked(page
));
2528 J_ASSERT(page_has_buffers(page
));
2529 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2530 len
= size
& ~PAGE_CACHE_MASK
;
2532 len
= PAGE_CACHE_SIZE
;
2533 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2534 ext4_bh_unmapped_or_delay
));
2536 if (ext4_journal_current_handle())
2539 if (PageChecked(page
)) {
2541 * It's mmapped pagecache. Add buffers and journal it. There
2542 * doesn't seem much point in redirtying the page here.
2544 ClearPageChecked(page
);
2545 return __ext4_journalled_writepage(page
, wbc
);
2548 * It may be a page full of checkpoint-mode buffers. We don't
2549 * really know unless we go poke around in the buffer_heads.
2550 * But block_write_full_page will do the right thing.
2552 return block_write_full_page(page
, ext4_get_block
, wbc
);
2555 redirty_page_for_writepage(wbc
, page
);
2560 static int ext4_readpage(struct file
*file
, struct page
*page
)
2562 return mpage_readpage(page
, ext4_get_block
);
2566 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2567 struct list_head
*pages
, unsigned nr_pages
)
2569 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2572 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2574 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2577 * If it's a full truncate we just forget about the pending dirtying
2580 ClearPageChecked(page
);
2582 jbd2_journal_invalidatepage(journal
, page
, offset
);
2585 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2587 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2589 WARN_ON(PageChecked(page
));
2590 if (!page_has_buffers(page
))
2592 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2596 * If the O_DIRECT write will extend the file then add this inode to the
2597 * orphan list. So recovery will truncate it back to the original size
2598 * if the machine crashes during the write.
2600 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2601 * crashes then stale disk data _may_ be exposed inside the file. But current
2602 * VFS code falls back into buffered path in that case so we are safe.
2604 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2605 const struct iovec
*iov
, loff_t offset
,
2606 unsigned long nr_segs
)
2608 struct file
*file
= iocb
->ki_filp
;
2609 struct inode
*inode
= file
->f_mapping
->host
;
2610 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2614 size_t count
= iov_length(iov
, nr_segs
);
2617 loff_t final_size
= offset
+ count
;
2619 if (final_size
> inode
->i_size
) {
2620 /* Credits for sb + inode write */
2621 handle
= ext4_journal_start(inode
, 2);
2622 if (IS_ERR(handle
)) {
2623 ret
= PTR_ERR(handle
);
2626 ret
= ext4_orphan_add(handle
, inode
);
2628 ext4_journal_stop(handle
);
2632 ei
->i_disksize
= inode
->i_size
;
2633 ext4_journal_stop(handle
);
2637 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
2639 ext4_get_block
, NULL
);
2644 /* Credits for sb + inode write */
2645 handle
= ext4_journal_start(inode
, 2);
2646 if (IS_ERR(handle
)) {
2647 /* This is really bad luck. We've written the data
2648 * but cannot extend i_size. Bail out and pretend
2649 * the write failed... */
2650 ret
= PTR_ERR(handle
);
2654 ext4_orphan_del(handle
, inode
);
2656 loff_t end
= offset
+ ret
;
2657 if (end
> inode
->i_size
) {
2658 ei
->i_disksize
= end
;
2659 i_size_write(inode
, end
);
2661 * We're going to return a positive `ret'
2662 * here due to non-zero-length I/O, so there's
2663 * no way of reporting error returns from
2664 * ext4_mark_inode_dirty() to userspace. So
2667 ext4_mark_inode_dirty(handle
, inode
);
2670 err
= ext4_journal_stop(handle
);
2679 * Pages can be marked dirty completely asynchronously from ext4's journalling
2680 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2681 * much here because ->set_page_dirty is called under VFS locks. The page is
2682 * not necessarily locked.
2684 * We cannot just dirty the page and leave attached buffers clean, because the
2685 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2686 * or jbddirty because all the journalling code will explode.
2688 * So what we do is to mark the page "pending dirty" and next time writepage
2689 * is called, propagate that into the buffers appropriately.
2691 static int ext4_journalled_set_page_dirty(struct page
*page
)
2693 SetPageChecked(page
);
2694 return __set_page_dirty_nobuffers(page
);
2697 static const struct address_space_operations ext4_ordered_aops
= {
2698 .readpage
= ext4_readpage
,
2699 .readpages
= ext4_readpages
,
2700 .writepage
= ext4_normal_writepage
,
2701 .sync_page
= block_sync_page
,
2702 .write_begin
= ext4_write_begin
,
2703 .write_end
= ext4_ordered_write_end
,
2705 .invalidatepage
= ext4_invalidatepage
,
2706 .releasepage
= ext4_releasepage
,
2707 .direct_IO
= ext4_direct_IO
,
2708 .migratepage
= buffer_migrate_page
,
2711 static const struct address_space_operations ext4_writeback_aops
= {
2712 .readpage
= ext4_readpage
,
2713 .readpages
= ext4_readpages
,
2714 .writepage
= ext4_normal_writepage
,
2715 .sync_page
= block_sync_page
,
2716 .write_begin
= ext4_write_begin
,
2717 .write_end
= ext4_writeback_write_end
,
2719 .invalidatepage
= ext4_invalidatepage
,
2720 .releasepage
= ext4_releasepage
,
2721 .direct_IO
= ext4_direct_IO
,
2722 .migratepage
= buffer_migrate_page
,
2725 static const struct address_space_operations ext4_journalled_aops
= {
2726 .readpage
= ext4_readpage
,
2727 .readpages
= ext4_readpages
,
2728 .writepage
= ext4_journalled_writepage
,
2729 .sync_page
= block_sync_page
,
2730 .write_begin
= ext4_write_begin
,
2731 .write_end
= ext4_journalled_write_end
,
2732 .set_page_dirty
= ext4_journalled_set_page_dirty
,
2734 .invalidatepage
= ext4_invalidatepage
,
2735 .releasepage
= ext4_releasepage
,
2738 static const struct address_space_operations ext4_da_aops
= {
2739 .readpage
= ext4_readpage
,
2740 .readpages
= ext4_readpages
,
2741 .writepage
= ext4_da_writepage
,
2742 .writepages
= ext4_da_writepages
,
2743 .sync_page
= block_sync_page
,
2744 .write_begin
= ext4_da_write_begin
,
2745 .write_end
= ext4_da_write_end
,
2747 .invalidatepage
= ext4_da_invalidatepage
,
2748 .releasepage
= ext4_releasepage
,
2749 .direct_IO
= ext4_direct_IO
,
2750 .migratepage
= buffer_migrate_page
,
2753 void ext4_set_aops(struct inode
*inode
)
2755 if (ext4_should_order_data(inode
) &&
2756 test_opt(inode
->i_sb
, DELALLOC
))
2757 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2758 else if (ext4_should_order_data(inode
))
2759 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
2760 else if (ext4_should_writeback_data(inode
) &&
2761 test_opt(inode
->i_sb
, DELALLOC
))
2762 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2763 else if (ext4_should_writeback_data(inode
))
2764 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
2766 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
2770 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2771 * up to the end of the block which corresponds to `from'.
2772 * This required during truncate. We need to physically zero the tail end
2773 * of that block so it doesn't yield old data if the file is later grown.
2775 int ext4_block_truncate_page(handle_t
*handle
,
2776 struct address_space
*mapping
, loff_t from
)
2778 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
2779 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2780 unsigned blocksize
, length
, pos
;
2782 struct inode
*inode
= mapping
->host
;
2783 struct buffer_head
*bh
;
2787 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
2791 blocksize
= inode
->i_sb
->s_blocksize
;
2792 length
= blocksize
- (offset
& (blocksize
- 1));
2793 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2796 * For "nobh" option, we can only work if we don't need to
2797 * read-in the page - otherwise we create buffers to do the IO.
2799 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
2800 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
2801 zero_user(page
, offset
, length
);
2802 set_page_dirty(page
);
2806 if (!page_has_buffers(page
))
2807 create_empty_buffers(page
, blocksize
, 0);
2809 /* Find the buffer that contains "offset" */
2810 bh
= page_buffers(page
);
2812 while (offset
>= pos
) {
2813 bh
= bh
->b_this_page
;
2819 if (buffer_freed(bh
)) {
2820 BUFFER_TRACE(bh
, "freed: skip");
2824 if (!buffer_mapped(bh
)) {
2825 BUFFER_TRACE(bh
, "unmapped");
2826 ext4_get_block(inode
, iblock
, bh
, 0);
2827 /* unmapped? It's a hole - nothing to do */
2828 if (!buffer_mapped(bh
)) {
2829 BUFFER_TRACE(bh
, "still unmapped");
2834 /* Ok, it's mapped. Make sure it's up-to-date */
2835 if (PageUptodate(page
))
2836 set_buffer_uptodate(bh
);
2838 if (!buffer_uptodate(bh
)) {
2840 ll_rw_block(READ
, 1, &bh
);
2842 /* Uhhuh. Read error. Complain and punt. */
2843 if (!buffer_uptodate(bh
))
2847 if (ext4_should_journal_data(inode
)) {
2848 BUFFER_TRACE(bh
, "get write access");
2849 err
= ext4_journal_get_write_access(handle
, bh
);
2854 zero_user(page
, offset
, length
);
2856 BUFFER_TRACE(bh
, "zeroed end of block");
2859 if (ext4_should_journal_data(inode
)) {
2860 err
= ext4_journal_dirty_metadata(handle
, bh
);
2862 if (ext4_should_order_data(inode
))
2863 err
= ext4_jbd2_file_inode(handle
, inode
);
2864 mark_buffer_dirty(bh
);
2869 page_cache_release(page
);
2874 * Probably it should be a library function... search for first non-zero word
2875 * or memcmp with zero_page, whatever is better for particular architecture.
2878 static inline int all_zeroes(__le32
*p
, __le32
*q
)
2887 * ext4_find_shared - find the indirect blocks for partial truncation.
2888 * @inode: inode in question
2889 * @depth: depth of the affected branch
2890 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
2891 * @chain: place to store the pointers to partial indirect blocks
2892 * @top: place to the (detached) top of branch
2894 * This is a helper function used by ext4_truncate().
2896 * When we do truncate() we may have to clean the ends of several
2897 * indirect blocks but leave the blocks themselves alive. Block is
2898 * partially truncated if some data below the new i_size is refered
2899 * from it (and it is on the path to the first completely truncated
2900 * data block, indeed). We have to free the top of that path along
2901 * with everything to the right of the path. Since no allocation
2902 * past the truncation point is possible until ext4_truncate()
2903 * finishes, we may safely do the latter, but top of branch may
2904 * require special attention - pageout below the truncation point
2905 * might try to populate it.
2907 * We atomically detach the top of branch from the tree, store the
2908 * block number of its root in *@top, pointers to buffer_heads of
2909 * partially truncated blocks - in @chain[].bh and pointers to
2910 * their last elements that should not be removed - in
2911 * @chain[].p. Return value is the pointer to last filled element
2914 * The work left to caller to do the actual freeing of subtrees:
2915 * a) free the subtree starting from *@top
2916 * b) free the subtrees whose roots are stored in
2917 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2918 * c) free the subtrees growing from the inode past the @chain[0].
2919 * (no partially truncated stuff there). */
2921 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
2922 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
2924 Indirect
*partial
, *p
;
2928 /* Make k index the deepest non-null offest + 1 */
2929 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
2931 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
2932 /* Writer: pointers */
2934 partial
= chain
+ k
-1;
2936 * If the branch acquired continuation since we've looked at it -
2937 * fine, it should all survive and (new) top doesn't belong to us.
2939 if (!partial
->key
&& *partial
->p
)
2942 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
2945 * OK, we've found the last block that must survive. The rest of our
2946 * branch should be detached before unlocking. However, if that rest
2947 * of branch is all ours and does not grow immediately from the inode
2948 * it's easier to cheat and just decrement partial->p.
2950 if (p
== chain
+ k
- 1 && p
> chain
) {
2954 /* Nope, don't do this in ext4. Must leave the tree intact */
2961 while(partial
> p
) {
2962 brelse(partial
->bh
);
2970 * Zero a number of block pointers in either an inode or an indirect block.
2971 * If we restart the transaction we must again get write access to the
2972 * indirect block for further modification.
2974 * We release `count' blocks on disk, but (last - first) may be greater
2975 * than `count' because there can be holes in there.
2977 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
2978 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
2979 unsigned long count
, __le32
*first
, __le32
*last
)
2982 if (try_to_extend_transaction(handle
, inode
)) {
2984 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
2985 ext4_journal_dirty_metadata(handle
, bh
);
2987 ext4_mark_inode_dirty(handle
, inode
);
2988 ext4_journal_test_restart(handle
, inode
);
2990 BUFFER_TRACE(bh
, "retaking write access");
2991 ext4_journal_get_write_access(handle
, bh
);
2996 * Any buffers which are on the journal will be in memory. We find
2997 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
2998 * on them. We've already detached each block from the file, so
2999 * bforget() in jbd2_journal_forget() should be safe.
3001 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3003 for (p
= first
; p
< last
; p
++) {
3004 u32 nr
= le32_to_cpu(*p
);
3006 struct buffer_head
*tbh
;
3009 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3010 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3014 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3018 * ext4_free_data - free a list of data blocks
3019 * @handle: handle for this transaction
3020 * @inode: inode we are dealing with
3021 * @this_bh: indirect buffer_head which contains *@first and *@last
3022 * @first: array of block numbers
3023 * @last: points immediately past the end of array
3025 * We are freeing all blocks refered from that array (numbers are stored as
3026 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3028 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3029 * blocks are contiguous then releasing them at one time will only affect one
3030 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3031 * actually use a lot of journal space.
3033 * @this_bh will be %NULL if @first and @last point into the inode's direct
3036 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3037 struct buffer_head
*this_bh
,
3038 __le32
*first
, __le32
*last
)
3040 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3041 unsigned long count
= 0; /* Number of blocks in the run */
3042 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3045 ext4_fsblk_t nr
; /* Current block # */
3046 __le32
*p
; /* Pointer into inode/ind
3047 for current block */
3050 if (this_bh
) { /* For indirect block */
3051 BUFFER_TRACE(this_bh
, "get_write_access");
3052 err
= ext4_journal_get_write_access(handle
, this_bh
);
3053 /* Important: if we can't update the indirect pointers
3054 * to the blocks, we can't free them. */
3059 for (p
= first
; p
< last
; p
++) {
3060 nr
= le32_to_cpu(*p
);
3062 /* accumulate blocks to free if they're contiguous */
3065 block_to_free_p
= p
;
3067 } else if (nr
== block_to_free
+ count
) {
3070 ext4_clear_blocks(handle
, inode
, this_bh
,
3072 count
, block_to_free_p
, p
);
3074 block_to_free_p
= p
;
3081 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3082 count
, block_to_free_p
, p
);
3085 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3088 * The buffer head should have an attached journal head at this
3089 * point. However, if the data is corrupted and an indirect
3090 * block pointed to itself, it would have been detached when
3091 * the block was cleared. Check for this instead of OOPSing.
3094 ext4_journal_dirty_metadata(handle
, this_bh
);
3096 ext4_error(inode
->i_sb
, __func__
,
3097 "circular indirect block detected, "
3098 "inode=%lu, block=%llu",
3100 (unsigned long long) this_bh
->b_blocknr
);
3105 * ext4_free_branches - free an array of branches
3106 * @handle: JBD handle for this transaction
3107 * @inode: inode we are dealing with
3108 * @parent_bh: the buffer_head which contains *@first and *@last
3109 * @first: array of block numbers
3110 * @last: pointer immediately past the end of array
3111 * @depth: depth of the branches to free
3113 * We are freeing all blocks refered from these branches (numbers are
3114 * stored as little-endian 32-bit) and updating @inode->i_blocks
3117 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3118 struct buffer_head
*parent_bh
,
3119 __le32
*first
, __le32
*last
, int depth
)
3124 if (is_handle_aborted(handle
))
3128 struct buffer_head
*bh
;
3129 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3131 while (--p
>= first
) {
3132 nr
= le32_to_cpu(*p
);
3134 continue; /* A hole */
3136 /* Go read the buffer for the next level down */
3137 bh
= sb_bread(inode
->i_sb
, nr
);
3140 * A read failure? Report error and clear slot
3144 ext4_error(inode
->i_sb
, "ext4_free_branches",
3145 "Read failure, inode=%lu, block=%llu",
3150 /* This zaps the entire block. Bottom up. */
3151 BUFFER_TRACE(bh
, "free child branches");
3152 ext4_free_branches(handle
, inode
, bh
,
3153 (__le32
*)bh
->b_data
,
3154 (__le32
*)bh
->b_data
+ addr_per_block
,
3158 * We've probably journalled the indirect block several
3159 * times during the truncate. But it's no longer
3160 * needed and we now drop it from the transaction via
3161 * jbd2_journal_revoke().
3163 * That's easy if it's exclusively part of this
3164 * transaction. But if it's part of the committing
3165 * transaction then jbd2_journal_forget() will simply
3166 * brelse() it. That means that if the underlying
3167 * block is reallocated in ext4_get_block(),
3168 * unmap_underlying_metadata() will find this block
3169 * and will try to get rid of it. damn, damn.
3171 * If this block has already been committed to the
3172 * journal, a revoke record will be written. And
3173 * revoke records must be emitted *before* clearing
3174 * this block's bit in the bitmaps.
3176 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3179 * Everything below this this pointer has been
3180 * released. Now let this top-of-subtree go.
3182 * We want the freeing of this indirect block to be
3183 * atomic in the journal with the updating of the
3184 * bitmap block which owns it. So make some room in
3187 * We zero the parent pointer *after* freeing its
3188 * pointee in the bitmaps, so if extend_transaction()
3189 * for some reason fails to put the bitmap changes and
3190 * the release into the same transaction, recovery
3191 * will merely complain about releasing a free block,
3192 * rather than leaking blocks.
3194 if (is_handle_aborted(handle
))
3196 if (try_to_extend_transaction(handle
, inode
)) {
3197 ext4_mark_inode_dirty(handle
, inode
);
3198 ext4_journal_test_restart(handle
, inode
);
3201 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3205 * The block which we have just freed is
3206 * pointed to by an indirect block: journal it
3208 BUFFER_TRACE(parent_bh
, "get_write_access");
3209 if (!ext4_journal_get_write_access(handle
,
3212 BUFFER_TRACE(parent_bh
,
3213 "call ext4_journal_dirty_metadata");
3214 ext4_journal_dirty_metadata(handle
,
3220 /* We have reached the bottom of the tree. */
3221 BUFFER_TRACE(parent_bh
, "free data blocks");
3222 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3226 int ext4_can_truncate(struct inode
*inode
)
3228 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3230 if (S_ISREG(inode
->i_mode
))
3232 if (S_ISDIR(inode
->i_mode
))
3234 if (S_ISLNK(inode
->i_mode
))
3235 return !ext4_inode_is_fast_symlink(inode
);
3242 * We block out ext4_get_block() block instantiations across the entire
3243 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3244 * simultaneously on behalf of the same inode.
3246 * As we work through the truncate and commmit bits of it to the journal there
3247 * is one core, guiding principle: the file's tree must always be consistent on
3248 * disk. We must be able to restart the truncate after a crash.
3250 * The file's tree may be transiently inconsistent in memory (although it
3251 * probably isn't), but whenever we close off and commit a journal transaction,
3252 * the contents of (the filesystem + the journal) must be consistent and
3253 * restartable. It's pretty simple, really: bottom up, right to left (although
3254 * left-to-right works OK too).
3256 * Note that at recovery time, journal replay occurs *before* the restart of
3257 * truncate against the orphan inode list.
3259 * The committed inode has the new, desired i_size (which is the same as
3260 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3261 * that this inode's truncate did not complete and it will again call
3262 * ext4_truncate() to have another go. So there will be instantiated blocks
3263 * to the right of the truncation point in a crashed ext4 filesystem. But
3264 * that's fine - as long as they are linked from the inode, the post-crash
3265 * ext4_truncate() run will find them and release them.
3267 void ext4_truncate(struct inode
*inode
)
3270 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3271 __le32
*i_data
= ei
->i_data
;
3272 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3273 struct address_space
*mapping
= inode
->i_mapping
;
3274 ext4_lblk_t offsets
[4];
3279 ext4_lblk_t last_block
;
3280 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3282 if (!ext4_can_truncate(inode
))
3285 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3286 ext4_ext_truncate(inode
);
3290 handle
= start_transaction(inode
);
3292 return; /* AKPM: return what? */
3294 last_block
= (inode
->i_size
+ blocksize
-1)
3295 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3297 if (inode
->i_size
& (blocksize
- 1))
3298 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3301 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3303 goto out_stop
; /* error */
3306 * OK. This truncate is going to happen. We add the inode to the
3307 * orphan list, so that if this truncate spans multiple transactions,
3308 * and we crash, we will resume the truncate when the filesystem
3309 * recovers. It also marks the inode dirty, to catch the new size.
3311 * Implication: the file must always be in a sane, consistent
3312 * truncatable state while each transaction commits.
3314 if (ext4_orphan_add(handle
, inode
))
3318 * The orphan list entry will now protect us from any crash which
3319 * occurs before the truncate completes, so it is now safe to propagate
3320 * the new, shorter inode size (held for now in i_size) into the
3321 * on-disk inode. We do this via i_disksize, which is the value which
3322 * ext4 *really* writes onto the disk inode.
3324 ei
->i_disksize
= inode
->i_size
;
3327 * From here we block out all ext4_get_block() callers who want to
3328 * modify the block allocation tree.
3330 down_write(&ei
->i_data_sem
);
3332 if (n
== 1) { /* direct blocks */
3333 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3334 i_data
+ EXT4_NDIR_BLOCKS
);
3338 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3339 /* Kill the top of shared branch (not detached) */
3341 if (partial
== chain
) {
3342 /* Shared branch grows from the inode */
3343 ext4_free_branches(handle
, inode
, NULL
,
3344 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3347 * We mark the inode dirty prior to restart,
3348 * and prior to stop. No need for it here.
3351 /* Shared branch grows from an indirect block */
3352 BUFFER_TRACE(partial
->bh
, "get_write_access");
3353 ext4_free_branches(handle
, inode
, partial
->bh
,
3355 partial
->p
+1, (chain
+n
-1) - partial
);
3358 /* Clear the ends of indirect blocks on the shared branch */
3359 while (partial
> chain
) {
3360 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3361 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3362 (chain
+n
-1) - partial
);
3363 BUFFER_TRACE(partial
->bh
, "call brelse");
3364 brelse (partial
->bh
);
3368 /* Kill the remaining (whole) subtrees */
3369 switch (offsets
[0]) {
3371 nr
= i_data
[EXT4_IND_BLOCK
];
3373 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3374 i_data
[EXT4_IND_BLOCK
] = 0;
3376 case EXT4_IND_BLOCK
:
3377 nr
= i_data
[EXT4_DIND_BLOCK
];
3379 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3380 i_data
[EXT4_DIND_BLOCK
] = 0;
3382 case EXT4_DIND_BLOCK
:
3383 nr
= i_data
[EXT4_TIND_BLOCK
];
3385 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3386 i_data
[EXT4_TIND_BLOCK
] = 0;
3388 case EXT4_TIND_BLOCK
:
3392 ext4_discard_reservation(inode
);
3394 up_write(&ei
->i_data_sem
);
3395 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3396 ext4_mark_inode_dirty(handle
, inode
);
3399 * In a multi-transaction truncate, we only make the final transaction
3406 * If this was a simple ftruncate(), and the file will remain alive
3407 * then we need to clear up the orphan record which we created above.
3408 * However, if this was a real unlink then we were called by
3409 * ext4_delete_inode(), and we allow that function to clean up the
3410 * orphan info for us.
3413 ext4_orphan_del(handle
, inode
);
3415 ext4_journal_stop(handle
);
3418 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
3419 unsigned long ino
, struct ext4_iloc
*iloc
)
3421 ext4_group_t block_group
;
3422 unsigned long offset
;
3424 struct ext4_group_desc
*gdp
;
3426 if (!ext4_valid_inum(sb
, ino
)) {
3428 * This error is already checked for in namei.c unless we are
3429 * looking at an NFS filehandle, in which case no error
3435 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3436 gdp
= ext4_get_group_desc(sb
, block_group
, NULL
);
3441 * Figure out the offset within the block group inode table
3443 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
3444 EXT4_INODE_SIZE(sb
);
3445 block
= ext4_inode_table(sb
, gdp
) +
3446 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
3448 iloc
->block_group
= block_group
;
3449 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
3454 * ext4_get_inode_loc returns with an extra refcount against the inode's
3455 * underlying buffer_head on success. If 'in_mem' is true, we have all
3456 * data in memory that is needed to recreate the on-disk version of this
3459 static int __ext4_get_inode_loc(struct inode
*inode
,
3460 struct ext4_iloc
*iloc
, int in_mem
)
3463 struct buffer_head
*bh
;
3465 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
3469 bh
= sb_getblk(inode
->i_sb
, block
);
3471 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
3472 "unable to read inode block - "
3473 "inode=%lu, block=%llu",
3474 inode
->i_ino
, block
);
3477 if (!buffer_uptodate(bh
)) {
3479 if (buffer_uptodate(bh
)) {
3480 /* someone brought it uptodate while we waited */
3486 * If we have all information of the inode in memory and this
3487 * is the only valid inode in the block, we need not read the
3491 struct buffer_head
*bitmap_bh
;
3492 struct ext4_group_desc
*desc
;
3493 int inodes_per_buffer
;
3494 int inode_offset
, i
;
3495 ext4_group_t block_group
;
3498 block_group
= (inode
->i_ino
- 1) /
3499 EXT4_INODES_PER_GROUP(inode
->i_sb
);
3500 inodes_per_buffer
= bh
->b_size
/
3501 EXT4_INODE_SIZE(inode
->i_sb
);
3502 inode_offset
= ((inode
->i_ino
- 1) %
3503 EXT4_INODES_PER_GROUP(inode
->i_sb
));
3504 start
= inode_offset
& ~(inodes_per_buffer
- 1);
3506 /* Is the inode bitmap in cache? */
3507 desc
= ext4_get_group_desc(inode
->i_sb
,
3512 bitmap_bh
= sb_getblk(inode
->i_sb
,
3513 ext4_inode_bitmap(inode
->i_sb
, desc
));
3518 * If the inode bitmap isn't in cache then the
3519 * optimisation may end up performing two reads instead
3520 * of one, so skip it.
3522 if (!buffer_uptodate(bitmap_bh
)) {
3526 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
3527 if (i
== inode_offset
)
3529 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3533 if (i
== start
+ inodes_per_buffer
) {
3534 /* all other inodes are free, so skip I/O */
3535 memset(bh
->b_data
, 0, bh
->b_size
);
3536 set_buffer_uptodate(bh
);
3544 * There are other valid inodes in the buffer, this inode
3545 * has in-inode xattrs, or we don't have this inode in memory.
3546 * Read the block from disk.
3549 bh
->b_end_io
= end_buffer_read_sync
;
3550 submit_bh(READ_META
, bh
);
3552 if (!buffer_uptodate(bh
)) {
3553 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
3554 "unable to read inode block - "
3555 "inode=%lu, block=%llu",
3556 inode
->i_ino
, block
);
3566 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3568 /* We have all inode data except xattrs in memory here. */
3569 return __ext4_get_inode_loc(inode
, iloc
,
3570 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3573 void ext4_set_inode_flags(struct inode
*inode
)
3575 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3577 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3578 if (flags
& EXT4_SYNC_FL
)
3579 inode
->i_flags
|= S_SYNC
;
3580 if (flags
& EXT4_APPEND_FL
)
3581 inode
->i_flags
|= S_APPEND
;
3582 if (flags
& EXT4_IMMUTABLE_FL
)
3583 inode
->i_flags
|= S_IMMUTABLE
;
3584 if (flags
& EXT4_NOATIME_FL
)
3585 inode
->i_flags
|= S_NOATIME
;
3586 if (flags
& EXT4_DIRSYNC_FL
)
3587 inode
->i_flags
|= S_DIRSYNC
;
3590 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3591 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3593 unsigned int flags
= ei
->vfs_inode
.i_flags
;
3595 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3596 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
3598 ei
->i_flags
|= EXT4_SYNC_FL
;
3599 if (flags
& S_APPEND
)
3600 ei
->i_flags
|= EXT4_APPEND_FL
;
3601 if (flags
& S_IMMUTABLE
)
3602 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
3603 if (flags
& S_NOATIME
)
3604 ei
->i_flags
|= EXT4_NOATIME_FL
;
3605 if (flags
& S_DIRSYNC
)
3606 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
3608 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3609 struct ext4_inode_info
*ei
)
3612 struct inode
*inode
= &(ei
->vfs_inode
);
3613 struct super_block
*sb
= inode
->i_sb
;
3615 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3616 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3617 /* we are using combined 48 bit field */
3618 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3619 le32_to_cpu(raw_inode
->i_blocks_lo
);
3620 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
3621 /* i_blocks represent file system block size */
3622 return i_blocks
<< (inode
->i_blkbits
- 9);
3627 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3631 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3633 struct ext4_iloc iloc
;
3634 struct ext4_inode
*raw_inode
;
3635 struct ext4_inode_info
*ei
;
3636 struct buffer_head
*bh
;
3637 struct inode
*inode
;
3641 inode
= iget_locked(sb
, ino
);
3643 return ERR_PTR(-ENOMEM
);
3644 if (!(inode
->i_state
& I_NEW
))
3648 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
3649 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
3650 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
3652 ei
->i_block_alloc_info
= NULL
;
3654 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3658 raw_inode
= ext4_raw_inode(&iloc
);
3659 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3660 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3661 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3662 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
3663 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3664 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3666 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
3669 ei
->i_dir_start_lookup
= 0;
3670 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3671 /* We now have enough fields to check if the inode was active or not.
3672 * This is needed because nfsd might try to access dead inodes
3673 * the test is that same one that e2fsck uses
3674 * NeilBrown 1999oct15
3676 if (inode
->i_nlink
== 0) {
3677 if (inode
->i_mode
== 0 ||
3678 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3679 /* this inode is deleted */
3684 /* The only unlinked inodes we let through here have
3685 * valid i_mode and are being read by the orphan
3686 * recovery code: that's fine, we're about to complete
3687 * the process of deleting those. */
3689 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3690 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3691 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3692 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
3693 cpu_to_le32(EXT4_OS_HURD
)) {
3695 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3697 inode
->i_size
= ext4_isize(raw_inode
);
3698 ei
->i_disksize
= inode
->i_size
;
3699 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3700 ei
->i_block_group
= iloc
.block_group
;
3702 * NOTE! The in-memory inode i_data array is in little-endian order
3703 * even on big-endian machines: we do NOT byteswap the block numbers!
3705 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3706 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3707 INIT_LIST_HEAD(&ei
->i_orphan
);
3709 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3710 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3711 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3712 EXT4_INODE_SIZE(inode
->i_sb
)) {
3717 if (ei
->i_extra_isize
== 0) {
3718 /* The extra space is currently unused. Use it. */
3719 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3720 EXT4_GOOD_OLD_INODE_SIZE
;
3722 __le32
*magic
= (void *)raw_inode
+
3723 EXT4_GOOD_OLD_INODE_SIZE
+
3725 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
3726 ei
->i_state
|= EXT4_STATE_XATTR
;
3729 ei
->i_extra_isize
= 0;
3731 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3732 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3733 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3734 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3736 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3737 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3738 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3740 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3743 if (S_ISREG(inode
->i_mode
)) {
3744 inode
->i_op
= &ext4_file_inode_operations
;
3745 inode
->i_fop
= &ext4_file_operations
;
3746 ext4_set_aops(inode
);
3747 } else if (S_ISDIR(inode
->i_mode
)) {
3748 inode
->i_op
= &ext4_dir_inode_operations
;
3749 inode
->i_fop
= &ext4_dir_operations
;
3750 } else if (S_ISLNK(inode
->i_mode
)) {
3751 if (ext4_inode_is_fast_symlink(inode
))
3752 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3754 inode
->i_op
= &ext4_symlink_inode_operations
;
3755 ext4_set_aops(inode
);
3758 inode
->i_op
= &ext4_special_inode_operations
;
3759 if (raw_inode
->i_block
[0])
3760 init_special_inode(inode
, inode
->i_mode
,
3761 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3763 init_special_inode(inode
, inode
->i_mode
,
3764 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3767 ext4_set_inode_flags(inode
);
3768 unlock_new_inode(inode
);
3773 return ERR_PTR(ret
);
3776 static int ext4_inode_blocks_set(handle_t
*handle
,
3777 struct ext4_inode
*raw_inode
,
3778 struct ext4_inode_info
*ei
)
3780 struct inode
*inode
= &(ei
->vfs_inode
);
3781 u64 i_blocks
= inode
->i_blocks
;
3782 struct super_block
*sb
= inode
->i_sb
;
3785 if (i_blocks
<= ~0U) {
3787 * i_blocks can be represnted in a 32 bit variable
3788 * as multiple of 512 bytes
3790 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3791 raw_inode
->i_blocks_high
= 0;
3792 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
3793 } else if (i_blocks
<= 0xffffffffffffULL
) {
3795 * i_blocks can be represented in a 48 bit variable
3796 * as multiple of 512 bytes
3798 err
= ext4_update_rocompat_feature(handle
, sb
,
3799 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
3802 /* i_block is stored in the split 48 bit fields */
3803 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3804 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3805 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
3808 * i_blocks should be represented in a 48 bit variable
3809 * as multiple of file system block size
3811 err
= ext4_update_rocompat_feature(handle
, sb
,
3812 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
3815 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
3816 /* i_block is stored in file system block size */
3817 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
3818 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3819 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3826 * Post the struct inode info into an on-disk inode location in the
3827 * buffer-cache. This gobbles the caller's reference to the
3828 * buffer_head in the inode location struct.
3830 * The caller must have write access to iloc->bh.
3832 static int ext4_do_update_inode(handle_t
*handle
,
3833 struct inode
*inode
,
3834 struct ext4_iloc
*iloc
)
3836 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
3837 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3838 struct buffer_head
*bh
= iloc
->bh
;
3839 int err
= 0, rc
, block
;
3841 /* For fields not not tracking in the in-memory inode,
3842 * initialise them to zero for new inodes. */
3843 if (ei
->i_state
& EXT4_STATE_NEW
)
3844 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
3846 ext4_get_inode_flags(ei
);
3847 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
3848 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
3849 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
3850 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
3852 * Fix up interoperability with old kernels. Otherwise, old inodes get
3853 * re-used with the upper 16 bits of the uid/gid intact
3856 raw_inode
->i_uid_high
=
3857 cpu_to_le16(high_16_bits(inode
->i_uid
));
3858 raw_inode
->i_gid_high
=
3859 cpu_to_le16(high_16_bits(inode
->i_gid
));
3861 raw_inode
->i_uid_high
= 0;
3862 raw_inode
->i_gid_high
= 0;
3865 raw_inode
->i_uid_low
=
3866 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
3867 raw_inode
->i_gid_low
=
3868 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
3869 raw_inode
->i_uid_high
= 0;
3870 raw_inode
->i_gid_high
= 0;
3872 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
3874 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
3875 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
3876 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
3877 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
3879 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
3881 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
3882 /* clear the migrate flag in the raw_inode */
3883 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
3884 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
3885 cpu_to_le32(EXT4_OS_HURD
))
3886 raw_inode
->i_file_acl_high
=
3887 cpu_to_le16(ei
->i_file_acl
>> 32);
3888 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
3889 ext4_isize_set(raw_inode
, ei
->i_disksize
);
3890 if (ei
->i_disksize
> 0x7fffffffULL
) {
3891 struct super_block
*sb
= inode
->i_sb
;
3892 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3893 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
3894 EXT4_SB(sb
)->s_es
->s_rev_level
==
3895 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
3896 /* If this is the first large file
3897 * created, add a flag to the superblock.
3899 err
= ext4_journal_get_write_access(handle
,
3900 EXT4_SB(sb
)->s_sbh
);
3903 ext4_update_dynamic_rev(sb
);
3904 EXT4_SET_RO_COMPAT_FEATURE(sb
,
3905 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
3908 err
= ext4_journal_dirty_metadata(handle
,
3909 EXT4_SB(sb
)->s_sbh
);
3912 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
3913 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
3914 if (old_valid_dev(inode
->i_rdev
)) {
3915 raw_inode
->i_block
[0] =
3916 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
3917 raw_inode
->i_block
[1] = 0;
3919 raw_inode
->i_block
[0] = 0;
3920 raw_inode
->i_block
[1] =
3921 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
3922 raw_inode
->i_block
[2] = 0;
3924 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3925 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
3927 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
3928 if (ei
->i_extra_isize
) {
3929 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3930 raw_inode
->i_version_hi
=
3931 cpu_to_le32(inode
->i_version
>> 32);
3932 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
3936 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3937 rc
= ext4_journal_dirty_metadata(handle
, bh
);
3940 ei
->i_state
&= ~EXT4_STATE_NEW
;
3944 ext4_std_error(inode
->i_sb
, err
);
3949 * ext4_write_inode()
3951 * We are called from a few places:
3953 * - Within generic_file_write() for O_SYNC files.
3954 * Here, there will be no transaction running. We wait for any running
3955 * trasnaction to commit.
3957 * - Within sys_sync(), kupdate and such.
3958 * We wait on commit, if tol to.
3960 * - Within prune_icache() (PF_MEMALLOC == true)
3961 * Here we simply return. We can't afford to block kswapd on the
3964 * In all cases it is actually safe for us to return without doing anything,
3965 * because the inode has been copied into a raw inode buffer in
3966 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3969 * Note that we are absolutely dependent upon all inode dirtiers doing the
3970 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3971 * which we are interested.
3973 * It would be a bug for them to not do this. The code:
3975 * mark_inode_dirty(inode)
3977 * inode->i_size = expr;
3979 * is in error because a kswapd-driven write_inode() could occur while
3980 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3981 * will no longer be on the superblock's dirty inode list.
3983 int ext4_write_inode(struct inode
*inode
, int wait
)
3985 if (current
->flags
& PF_MEMALLOC
)
3988 if (ext4_journal_current_handle()) {
3989 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3997 return ext4_force_commit(inode
->i_sb
);
4003 * Called from notify_change.
4005 * We want to trap VFS attempts to truncate the file as soon as
4006 * possible. In particular, we want to make sure that when the VFS
4007 * shrinks i_size, we put the inode on the orphan list and modify
4008 * i_disksize immediately, so that during the subsequent flushing of
4009 * dirty pages and freeing of disk blocks, we can guarantee that any
4010 * commit will leave the blocks being flushed in an unused state on
4011 * disk. (On recovery, the inode will get truncated and the blocks will
4012 * be freed, so we have a strong guarantee that no future commit will
4013 * leave these blocks visible to the user.)
4015 * Another thing we have to assure is that if we are in ordered mode
4016 * and inode is still attached to the committing transaction, we must
4017 * we start writeout of all the dirty pages which are being truncated.
4018 * This way we are sure that all the data written in the previous
4019 * transaction are already on disk (truncate waits for pages under
4022 * Called with inode->i_mutex down.
4024 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4026 struct inode
*inode
= dentry
->d_inode
;
4028 const unsigned int ia_valid
= attr
->ia_valid
;
4030 error
= inode_change_ok(inode
, attr
);
4034 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4035 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4038 /* (user+group)*(old+new) structure, inode write (sb,
4039 * inode block, ? - but truncate inode update has it) */
4040 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4041 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4042 if (IS_ERR(handle
)) {
4043 error
= PTR_ERR(handle
);
4046 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4048 ext4_journal_stop(handle
);
4051 /* Update corresponding info in inode so that everything is in
4052 * one transaction */
4053 if (attr
->ia_valid
& ATTR_UID
)
4054 inode
->i_uid
= attr
->ia_uid
;
4055 if (attr
->ia_valid
& ATTR_GID
)
4056 inode
->i_gid
= attr
->ia_gid
;
4057 error
= ext4_mark_inode_dirty(handle
, inode
);
4058 ext4_journal_stop(handle
);
4061 if (attr
->ia_valid
& ATTR_SIZE
) {
4062 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4063 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4065 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4072 if (S_ISREG(inode
->i_mode
) &&
4073 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4076 handle
= ext4_journal_start(inode
, 3);
4077 if (IS_ERR(handle
)) {
4078 error
= PTR_ERR(handle
);
4082 error
= ext4_orphan_add(handle
, inode
);
4083 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4084 rc
= ext4_mark_inode_dirty(handle
, inode
);
4087 ext4_journal_stop(handle
);
4089 if (ext4_should_order_data(inode
)) {
4090 error
= ext4_begin_ordered_truncate(inode
,
4093 /* Do as much error cleanup as possible */
4094 handle
= ext4_journal_start(inode
, 3);
4095 if (IS_ERR(handle
)) {
4096 ext4_orphan_del(NULL
, inode
);
4099 ext4_orphan_del(handle
, inode
);
4100 ext4_journal_stop(handle
);
4106 rc
= inode_setattr(inode
, attr
);
4108 /* If inode_setattr's call to ext4_truncate failed to get a
4109 * transaction handle at all, we need to clean up the in-core
4110 * orphan list manually. */
4112 ext4_orphan_del(NULL
, inode
);
4114 if (!rc
&& (ia_valid
& ATTR_MODE
))
4115 rc
= ext4_acl_chmod(inode
);
4118 ext4_std_error(inode
->i_sb
, error
);
4126 * How many blocks doth make a writepage()?
4128 * With N blocks per page, it may be:
4133 * N+5 bitmap blocks (from the above)
4134 * N+5 group descriptor summary blocks
4137 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
4139 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
4141 * With ordered or writeback data it's the same, less the N data blocks.
4143 * If the inode's direct blocks can hold an integral number of pages then a
4144 * page cannot straddle two indirect blocks, and we can only touch one indirect
4145 * and dindirect block, and the "5" above becomes "3".
4147 * This still overestimates under most circumstances. If we were to pass the
4148 * start and end offsets in here as well we could do block_to_path() on each
4149 * block and work out the exact number of indirects which are touched. Pah.
4152 int ext4_writepage_trans_blocks(struct inode
*inode
)
4154 int bpp
= ext4_journal_blocks_per_page(inode
);
4155 int indirects
= (EXT4_NDIR_BLOCKS
% bpp
) ? 5 : 3;
4158 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
4159 return ext4_ext_writepage_trans_blocks(inode
, bpp
);
4161 if (ext4_should_journal_data(inode
))
4162 ret
= 3 * (bpp
+ indirects
) + 2;
4164 ret
= 2 * (bpp
+ indirects
) + 2;
4167 /* We know that structure was already allocated during DQUOT_INIT so
4168 * we will be updating only the data blocks + inodes */
4169 ret
+= 2*EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
4176 * The caller must have previously called ext4_reserve_inode_write().
4177 * Give this, we know that the caller already has write access to iloc->bh.
4179 int ext4_mark_iloc_dirty(handle_t
*handle
,
4180 struct inode
*inode
, struct ext4_iloc
*iloc
)
4184 if (test_opt(inode
->i_sb
, I_VERSION
))
4185 inode_inc_iversion(inode
);
4187 /* the do_update_inode consumes one bh->b_count */
4190 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4191 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4197 * On success, We end up with an outstanding reference count against
4198 * iloc->bh. This _must_ be cleaned up later.
4202 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4203 struct ext4_iloc
*iloc
)
4207 err
= ext4_get_inode_loc(inode
, iloc
);
4209 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4210 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4217 ext4_std_error(inode
->i_sb
, err
);
4222 * Expand an inode by new_extra_isize bytes.
4223 * Returns 0 on success or negative error number on failure.
4225 static int ext4_expand_extra_isize(struct inode
*inode
,
4226 unsigned int new_extra_isize
,
4227 struct ext4_iloc iloc
,
4230 struct ext4_inode
*raw_inode
;
4231 struct ext4_xattr_ibody_header
*header
;
4232 struct ext4_xattr_entry
*entry
;
4234 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4237 raw_inode
= ext4_raw_inode(&iloc
);
4239 header
= IHDR(inode
, raw_inode
);
4240 entry
= IFIRST(header
);
4242 /* No extended attributes present */
4243 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4244 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4245 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4247 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4251 /* try to expand with EAs present */
4252 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4257 * What we do here is to mark the in-core inode as clean with respect to inode
4258 * dirtiness (it may still be data-dirty).
4259 * This means that the in-core inode may be reaped by prune_icache
4260 * without having to perform any I/O. This is a very good thing,
4261 * because *any* task may call prune_icache - even ones which
4262 * have a transaction open against a different journal.
4264 * Is this cheating? Not really. Sure, we haven't written the
4265 * inode out, but prune_icache isn't a user-visible syncing function.
4266 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4267 * we start and wait on commits.
4269 * Is this efficient/effective? Well, we're being nice to the system
4270 * by cleaning up our inodes proactively so they can be reaped
4271 * without I/O. But we are potentially leaving up to five seconds'
4272 * worth of inodes floating about which prune_icache wants us to
4273 * write out. One way to fix that would be to get prune_icache()
4274 * to do a write_super() to free up some memory. It has the desired
4277 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4279 struct ext4_iloc iloc
;
4280 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4281 static unsigned int mnt_count
;
4285 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4286 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4287 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4289 * We need extra buffer credits since we may write into EA block
4290 * with this same handle. If journal_extend fails, then it will
4291 * only result in a minor loss of functionality for that inode.
4292 * If this is felt to be critical, then e2fsck should be run to
4293 * force a large enough s_min_extra_isize.
4295 if ((jbd2_journal_extend(handle
,
4296 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4297 ret
= ext4_expand_extra_isize(inode
,
4298 sbi
->s_want_extra_isize
,
4301 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4303 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4304 ext4_warning(inode
->i_sb
, __func__
,
4305 "Unable to expand inode %lu. Delete"
4306 " some EAs or run e2fsck.",
4309 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4315 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4320 * ext4_dirty_inode() is called from __mark_inode_dirty()
4322 * We're really interested in the case where a file is being extended.
4323 * i_size has been changed by generic_commit_write() and we thus need
4324 * to include the updated inode in the current transaction.
4326 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4327 * are allocated to the file.
4329 * If the inode is marked synchronous, we don't honour that here - doing
4330 * so would cause a commit on atime updates, which we don't bother doing.
4331 * We handle synchronous inodes at the highest possible level.
4333 void ext4_dirty_inode(struct inode
*inode
)
4335 handle_t
*current_handle
= ext4_journal_current_handle();
4338 handle
= ext4_journal_start(inode
, 2);
4341 if (current_handle
&&
4342 current_handle
->h_transaction
!= handle
->h_transaction
) {
4343 /* This task has a transaction open against a different fs */
4344 printk(KERN_EMERG
"%s: transactions do not match!\n",
4347 jbd_debug(5, "marking dirty. outer handle=%p\n",
4349 ext4_mark_inode_dirty(handle
, inode
);
4351 ext4_journal_stop(handle
);
4358 * Bind an inode's backing buffer_head into this transaction, to prevent
4359 * it from being flushed to disk early. Unlike
4360 * ext4_reserve_inode_write, this leaves behind no bh reference and
4361 * returns no iloc structure, so the caller needs to repeat the iloc
4362 * lookup to mark the inode dirty later.
4364 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4366 struct ext4_iloc iloc
;
4370 err
= ext4_get_inode_loc(inode
, &iloc
);
4372 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4373 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4375 err
= ext4_journal_dirty_metadata(handle
,
4380 ext4_std_error(inode
->i_sb
, err
);
4385 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4392 * We have to be very careful here: changing a data block's
4393 * journaling status dynamically is dangerous. If we write a
4394 * data block to the journal, change the status and then delete
4395 * that block, we risk forgetting to revoke the old log record
4396 * from the journal and so a subsequent replay can corrupt data.
4397 * So, first we make sure that the journal is empty and that
4398 * nobody is changing anything.
4401 journal
= EXT4_JOURNAL(inode
);
4402 if (is_journal_aborted(journal
))
4405 jbd2_journal_lock_updates(journal
);
4406 jbd2_journal_flush(journal
);
4409 * OK, there are no updates running now, and all cached data is
4410 * synced to disk. We are now in a completely consistent state
4411 * which doesn't have anything in the journal, and we know that
4412 * no filesystem updates are running, so it is safe to modify
4413 * the inode's in-core data-journaling state flag now.
4417 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4419 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4420 ext4_set_aops(inode
);
4422 jbd2_journal_unlock_updates(journal
);
4424 /* Finally we can mark the inode as dirty. */
4426 handle
= ext4_journal_start(inode
, 1);
4428 return PTR_ERR(handle
);
4430 err
= ext4_mark_inode_dirty(handle
, inode
);
4432 ext4_journal_stop(handle
);
4433 ext4_std_error(inode
->i_sb
, err
);
4438 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4440 return !buffer_mapped(bh
);
4443 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4448 struct file
*file
= vma
->vm_file
;
4449 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4450 struct address_space
*mapping
= inode
->i_mapping
;
4453 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4454 * get i_mutex because we are already holding mmap_sem.
4456 down_read(&inode
->i_alloc_sem
);
4457 size
= i_size_read(inode
);
4458 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4459 || !PageUptodate(page
)) {
4460 /* page got truncated from under us? */
4464 if (PageMappedToDisk(page
))
4467 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4468 len
= size
& ~PAGE_CACHE_MASK
;
4470 len
= PAGE_CACHE_SIZE
;
4472 if (page_has_buffers(page
)) {
4473 /* return if we have all the buffers mapped */
4474 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4479 * OK, we need to fill the hole... Do write_begin write_end
4480 * to do block allocation/reservation.We are not holding
4481 * inode.i__mutex here. That allow * parallel write_begin,
4482 * write_end call. lock_page prevent this from happening
4483 * on the same page though
4485 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4486 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, NULL
);
4489 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4490 len
, len
, page
, NULL
);
4495 up_read(&inode
->i_alloc_sem
);