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 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
21 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/jbd2.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
54 trace_ext4_begin_ordered_truncate(inode
, new_size
);
56 * If jinode is zero, then we never opened the file for
57 * writing, so there's no need to call
58 * jbd2_journal_begin_ordered_truncate() since there's no
59 * outstanding writes we need to flush.
61 if (!EXT4_I(inode
)->jinode
)
63 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
64 EXT4_I(inode
)->jinode
,
68 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
69 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
70 struct buffer_head
*bh_result
, int create
);
71 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
72 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
73 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
74 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
77 * Test whether an inode is a fast symlink.
79 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
81 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
82 (inode
->i_sb
->s_blocksize
>> 9) : 0;
84 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
88 * Restart the transaction associated with *handle. This does a commit,
89 * so before we call here everything must be consistently dirtied against
92 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
98 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
99 * moment, get_block can be called only for blocks inside i_size since
100 * page cache has been already dropped and writes are blocked by
101 * i_mutex. So we can safely drop the i_data_sem here.
103 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
104 jbd_debug(2, "restarting handle %p\n", handle
);
105 up_write(&EXT4_I(inode
)->i_data_sem
);
106 ret
= ext4_journal_restart(handle
, nblocks
);
107 down_write(&EXT4_I(inode
)->i_data_sem
);
108 ext4_discard_preallocations(inode
);
114 * Called at the last iput() if i_nlink is zero.
116 void ext4_evict_inode(struct inode
*inode
)
121 trace_ext4_evict_inode(inode
);
123 ext4_ioend_wait(inode
);
125 if (inode
->i_nlink
) {
127 * When journalling data dirty buffers are tracked only in the
128 * journal. So although mm thinks everything is clean and
129 * ready for reaping the inode might still have some pages to
130 * write in the running transaction or waiting to be
131 * checkpointed. Thus calling jbd2_journal_invalidatepage()
132 * (via truncate_inode_pages()) to discard these buffers can
133 * cause data loss. Also even if we did not discard these
134 * buffers, we would have no way to find them after the inode
135 * is reaped and thus user could see stale data if he tries to
136 * read them before the transaction is checkpointed. So be
137 * careful and force everything to disk here... We use
138 * ei->i_datasync_tid to store the newest transaction
139 * containing inode's data.
141 * Note that directories do not have this problem because they
142 * don't use page cache.
144 if (ext4_should_journal_data(inode
) &&
145 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
146 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
147 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
149 jbd2_log_start_commit(journal
, commit_tid
);
150 jbd2_log_wait_commit(journal
, commit_tid
);
151 filemap_write_and_wait(&inode
->i_data
);
153 truncate_inode_pages(&inode
->i_data
, 0);
157 if (!is_bad_inode(inode
))
158 dquot_initialize(inode
);
160 if (ext4_should_order_data(inode
))
161 ext4_begin_ordered_truncate(inode
, 0);
162 truncate_inode_pages(&inode
->i_data
, 0);
164 if (is_bad_inode(inode
))
167 handle
= ext4_journal_start(inode
, ext4_blocks_for_truncate(inode
)+3);
168 if (IS_ERR(handle
)) {
169 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
171 * If we're going to skip the normal cleanup, we still need to
172 * make sure that the in-core orphan linked list is properly
175 ext4_orphan_del(NULL
, inode
);
180 ext4_handle_sync(handle
);
182 err
= ext4_mark_inode_dirty(handle
, inode
);
184 ext4_warning(inode
->i_sb
,
185 "couldn't mark inode dirty (err %d)", err
);
189 ext4_truncate(inode
);
192 * ext4_ext_truncate() doesn't reserve any slop when it
193 * restarts journal transactions; therefore there may not be
194 * enough credits left in the handle to remove the inode from
195 * the orphan list and set the dtime field.
197 if (!ext4_handle_has_enough_credits(handle
, 3)) {
198 err
= ext4_journal_extend(handle
, 3);
200 err
= ext4_journal_restart(handle
, 3);
202 ext4_warning(inode
->i_sb
,
203 "couldn't extend journal (err %d)", err
);
205 ext4_journal_stop(handle
);
206 ext4_orphan_del(NULL
, inode
);
212 * Kill off the orphan record which ext4_truncate created.
213 * AKPM: I think this can be inside the above `if'.
214 * Note that ext4_orphan_del() has to be able to cope with the
215 * deletion of a non-existent orphan - this is because we don't
216 * know if ext4_truncate() actually created an orphan record.
217 * (Well, we could do this if we need to, but heck - it works)
219 ext4_orphan_del(handle
, inode
);
220 EXT4_I(inode
)->i_dtime
= get_seconds();
223 * One subtle ordering requirement: if anything has gone wrong
224 * (transaction abort, IO errors, whatever), then we can still
225 * do these next steps (the fs will already have been marked as
226 * having errors), but we can't free the inode if the mark_dirty
229 if (ext4_mark_inode_dirty(handle
, inode
))
230 /* If that failed, just do the required in-core inode clear. */
231 ext4_clear_inode(inode
);
233 ext4_free_inode(handle
, inode
);
234 ext4_journal_stop(handle
);
237 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
241 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
243 return &EXT4_I(inode
)->i_reserved_quota
;
248 * Calculate the number of metadata blocks need to reserve
249 * to allocate a block located at @lblock
251 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
253 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
254 return ext4_ext_calc_metadata_amount(inode
, lblock
);
256 return ext4_ind_calc_metadata_amount(inode
, lblock
);
260 * Called with i_data_sem down, which is important since we can call
261 * ext4_discard_preallocations() from here.
263 void ext4_da_update_reserve_space(struct inode
*inode
,
264 int used
, int quota_claim
)
266 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
267 struct ext4_inode_info
*ei
= EXT4_I(inode
);
269 spin_lock(&ei
->i_block_reservation_lock
);
270 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
271 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
272 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
273 "with only %d reserved data blocks\n",
274 __func__
, inode
->i_ino
, used
,
275 ei
->i_reserved_data_blocks
);
277 used
= ei
->i_reserved_data_blocks
;
280 /* Update per-inode reservations */
281 ei
->i_reserved_data_blocks
-= used
;
282 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
283 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
284 used
+ ei
->i_allocated_meta_blocks
);
285 ei
->i_allocated_meta_blocks
= 0;
287 if (ei
->i_reserved_data_blocks
== 0) {
289 * We can release all of the reserved metadata blocks
290 * only when we have written all of the delayed
293 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
294 ei
->i_reserved_meta_blocks
);
295 ei
->i_reserved_meta_blocks
= 0;
296 ei
->i_da_metadata_calc_len
= 0;
298 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
300 /* Update quota subsystem for data blocks */
302 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
305 * We did fallocate with an offset that is already delayed
306 * allocated. So on delayed allocated writeback we should
307 * not re-claim the quota for fallocated blocks.
309 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
313 * If we have done all the pending block allocations and if
314 * there aren't any writers on the inode, we can discard the
315 * inode's preallocations.
317 if ((ei
->i_reserved_data_blocks
== 0) &&
318 (atomic_read(&inode
->i_writecount
) == 0))
319 ext4_discard_preallocations(inode
);
322 static int __check_block_validity(struct inode
*inode
, const char *func
,
324 struct ext4_map_blocks
*map
)
326 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
328 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
329 "lblock %lu mapped to illegal pblock "
330 "(length %d)", (unsigned long) map
->m_lblk
,
337 #define check_block_validity(inode, map) \
338 __check_block_validity((inode), __func__, __LINE__, (map))
341 * Return the number of contiguous dirty pages in a given inode
342 * starting at page frame idx.
344 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
345 unsigned int max_pages
)
347 struct address_space
*mapping
= inode
->i_mapping
;
351 int i
, nr_pages
, done
= 0;
355 pagevec_init(&pvec
, 0);
358 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
360 (pgoff_t
)PAGEVEC_SIZE
);
363 for (i
= 0; i
< nr_pages
; i
++) {
364 struct page
*page
= pvec
.pages
[i
];
365 struct buffer_head
*bh
, *head
;
368 if (unlikely(page
->mapping
!= mapping
) ||
370 PageWriteback(page
) ||
371 page
->index
!= idx
) {
376 if (page_has_buffers(page
)) {
377 bh
= head
= page_buffers(page
);
379 if (!buffer_delay(bh
) &&
380 !buffer_unwritten(bh
))
382 bh
= bh
->b_this_page
;
383 } while (!done
&& (bh
!= head
));
390 if (num
>= max_pages
) {
395 pagevec_release(&pvec
);
401 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
403 static void set_buffers_da_mapped(struct inode
*inode
,
404 struct ext4_map_blocks
*map
)
406 struct address_space
*mapping
= inode
->i_mapping
;
411 index
= map
->m_lblk
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
412 end
= (map
->m_lblk
+ map
->m_len
- 1) >>
413 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
415 pagevec_init(&pvec
, 0);
416 while (index
<= end
) {
417 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
,
419 (pgoff_t
)PAGEVEC_SIZE
));
422 for (i
= 0; i
< nr_pages
; i
++) {
423 struct page
*page
= pvec
.pages
[i
];
424 struct buffer_head
*bh
, *head
;
426 if (unlikely(page
->mapping
!= mapping
) ||
430 if (page_has_buffers(page
)) {
431 bh
= head
= page_buffers(page
);
433 set_buffer_da_mapped(bh
);
434 bh
= bh
->b_this_page
;
435 } while (bh
!= head
);
439 pagevec_release(&pvec
);
444 * The ext4_map_blocks() function tries to look up the requested blocks,
445 * and returns if the blocks are already mapped.
447 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
448 * and store the allocated blocks in the result buffer head and mark it
451 * If file type is extents based, it will call ext4_ext_map_blocks(),
452 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
455 * On success, it returns the number of blocks being mapped or allocate.
456 * if create==0 and the blocks are pre-allocated and uninitialized block,
457 * the result buffer head is unmapped. If the create ==1, it will make sure
458 * the buffer head is mapped.
460 * It returns 0 if plain look up failed (blocks have not been allocated), in
461 * that case, buffer head is unmapped
463 * It returns the error in case of allocation failure.
465 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
466 struct ext4_map_blocks
*map
, int flags
)
471 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
472 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
473 (unsigned long) map
->m_lblk
);
475 * Try to see if we can get the block without requesting a new
478 down_read((&EXT4_I(inode
)->i_data_sem
));
479 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
480 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
482 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
484 up_read((&EXT4_I(inode
)->i_data_sem
));
486 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
487 int ret
= check_block_validity(inode
, map
);
492 /* If it is only a block(s) look up */
493 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
497 * Returns if the blocks have already allocated
499 * Note that if blocks have been preallocated
500 * ext4_ext_get_block() returns the create = 0
501 * with buffer head unmapped.
503 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
507 * When we call get_blocks without the create flag, the
508 * BH_Unwritten flag could have gotten set if the blocks
509 * requested were part of a uninitialized extent. We need to
510 * clear this flag now that we are committed to convert all or
511 * part of the uninitialized extent to be an initialized
512 * extent. This is because we need to avoid the combination
513 * of BH_Unwritten and BH_Mapped flags being simultaneously
514 * set on the buffer_head.
516 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
519 * New blocks allocate and/or writing to uninitialized extent
520 * will possibly result in updating i_data, so we take
521 * the write lock of i_data_sem, and call get_blocks()
522 * with create == 1 flag.
524 down_write((&EXT4_I(inode
)->i_data_sem
));
527 * if the caller is from delayed allocation writeout path
528 * we have already reserved fs blocks for allocation
529 * let the underlying get_block() function know to
530 * avoid double accounting
532 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
533 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
535 * We need to check for EXT4 here because migrate
536 * could have changed the inode type in between
538 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
539 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
541 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
543 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
545 * We allocated new blocks which will result in
546 * i_data's format changing. Force the migrate
547 * to fail by clearing migrate flags
549 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
553 * Update reserved blocks/metadata blocks after successful
554 * block allocation which had been deferred till now. We don't
555 * support fallocate for non extent files. So we can update
556 * reserve space here.
559 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
560 ext4_da_update_reserve_space(inode
, retval
, 1);
562 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
563 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
565 /* If we have successfully mapped the delayed allocated blocks,
566 * set the BH_Da_Mapped bit on them. Its important to do this
567 * under the protection of i_data_sem.
569 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
570 set_buffers_da_mapped(inode
, map
);
573 up_write((&EXT4_I(inode
)->i_data_sem
));
574 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
575 int ret
= check_block_validity(inode
, map
);
582 /* Maximum number of blocks we map for direct IO at once. */
583 #define DIO_MAX_BLOCKS 4096
585 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
586 struct buffer_head
*bh
, int flags
)
588 handle_t
*handle
= ext4_journal_current_handle();
589 struct ext4_map_blocks map
;
590 int ret
= 0, started
= 0;
594 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
596 if (flags
&& !handle
) {
597 /* Direct IO write... */
598 if (map
.m_len
> DIO_MAX_BLOCKS
)
599 map
.m_len
= DIO_MAX_BLOCKS
;
600 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
601 handle
= ext4_journal_start(inode
, dio_credits
);
602 if (IS_ERR(handle
)) {
603 ret
= PTR_ERR(handle
);
609 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
611 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
612 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
613 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
617 ext4_journal_stop(handle
);
621 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
622 struct buffer_head
*bh
, int create
)
624 return _ext4_get_block(inode
, iblock
, bh
,
625 create
? EXT4_GET_BLOCKS_CREATE
: 0);
629 * `handle' can be NULL if create is zero
631 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
632 ext4_lblk_t block
, int create
, int *errp
)
634 struct ext4_map_blocks map
;
635 struct buffer_head
*bh
;
638 J_ASSERT(handle
!= NULL
|| create
== 0);
642 err
= ext4_map_blocks(handle
, inode
, &map
,
643 create
? EXT4_GET_BLOCKS_CREATE
: 0);
651 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
656 if (map
.m_flags
& EXT4_MAP_NEW
) {
657 J_ASSERT(create
!= 0);
658 J_ASSERT(handle
!= NULL
);
661 * Now that we do not always journal data, we should
662 * keep in mind whether this should always journal the
663 * new buffer as metadata. For now, regular file
664 * writes use ext4_get_block instead, so it's not a
668 BUFFER_TRACE(bh
, "call get_create_access");
669 fatal
= ext4_journal_get_create_access(handle
, bh
);
670 if (!fatal
&& !buffer_uptodate(bh
)) {
671 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
672 set_buffer_uptodate(bh
);
675 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
676 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
680 BUFFER_TRACE(bh
, "not a new buffer");
690 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
691 ext4_lblk_t block
, int create
, int *err
)
693 struct buffer_head
*bh
;
695 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
698 if (buffer_uptodate(bh
))
700 ll_rw_block(READ_META
, 1, &bh
);
702 if (buffer_uptodate(bh
))
709 static int walk_page_buffers(handle_t
*handle
,
710 struct buffer_head
*head
,
714 int (*fn
)(handle_t
*handle
,
715 struct buffer_head
*bh
))
717 struct buffer_head
*bh
;
718 unsigned block_start
, block_end
;
719 unsigned blocksize
= head
->b_size
;
721 struct buffer_head
*next
;
723 for (bh
= head
, block_start
= 0;
724 ret
== 0 && (bh
!= head
|| !block_start
);
725 block_start
= block_end
, bh
= next
) {
726 next
= bh
->b_this_page
;
727 block_end
= block_start
+ blocksize
;
728 if (block_end
<= from
|| block_start
>= to
) {
729 if (partial
&& !buffer_uptodate(bh
))
733 err
= (*fn
)(handle
, bh
);
741 * To preserve ordering, it is essential that the hole instantiation and
742 * the data write be encapsulated in a single transaction. We cannot
743 * close off a transaction and start a new one between the ext4_get_block()
744 * and the commit_write(). So doing the jbd2_journal_start at the start of
745 * prepare_write() is the right place.
747 * Also, this function can nest inside ext4_writepage() ->
748 * block_write_full_page(). In that case, we *know* that ext4_writepage()
749 * has generated enough buffer credits to do the whole page. So we won't
750 * block on the journal in that case, which is good, because the caller may
753 * By accident, ext4 can be reentered when a transaction is open via
754 * quota file writes. If we were to commit the transaction while thus
755 * reentered, there can be a deadlock - we would be holding a quota
756 * lock, and the commit would never complete if another thread had a
757 * transaction open and was blocking on the quota lock - a ranking
760 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
761 * will _not_ run commit under these circumstances because handle->h_ref
762 * is elevated. We'll still have enough credits for the tiny quotafile
765 static int do_journal_get_write_access(handle_t
*handle
,
766 struct buffer_head
*bh
)
768 int dirty
= buffer_dirty(bh
);
771 if (!buffer_mapped(bh
) || buffer_freed(bh
))
774 * __block_write_begin() could have dirtied some buffers. Clean
775 * the dirty bit as jbd2_journal_get_write_access() could complain
776 * otherwise about fs integrity issues. Setting of the dirty bit
777 * by __block_write_begin() isn't a real problem here as we clear
778 * the bit before releasing a page lock and thus writeback cannot
779 * ever write the buffer.
782 clear_buffer_dirty(bh
);
783 ret
= ext4_journal_get_write_access(handle
, bh
);
785 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
789 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
790 struct buffer_head
*bh_result
, int create
);
791 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
792 loff_t pos
, unsigned len
, unsigned flags
,
793 struct page
**pagep
, void **fsdata
)
795 struct inode
*inode
= mapping
->host
;
796 int ret
, needed_blocks
;
803 trace_ext4_write_begin(inode
, pos
, len
, flags
);
805 * Reserve one block more for addition to orphan list in case
806 * we allocate blocks but write fails for some reason
808 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
809 index
= pos
>> PAGE_CACHE_SHIFT
;
810 from
= pos
& (PAGE_CACHE_SIZE
- 1);
814 handle
= ext4_journal_start(inode
, needed_blocks
);
815 if (IS_ERR(handle
)) {
816 ret
= PTR_ERR(handle
);
820 /* We cannot recurse into the filesystem as the transaction is already
822 flags
|= AOP_FLAG_NOFS
;
824 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
826 ext4_journal_stop(handle
);
832 if (ext4_should_dioread_nolock(inode
))
833 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
835 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
837 if (!ret
&& ext4_should_journal_data(inode
)) {
838 ret
= walk_page_buffers(handle
, page_buffers(page
),
839 from
, to
, NULL
, do_journal_get_write_access
);
844 page_cache_release(page
);
846 * __block_write_begin may have instantiated a few blocks
847 * outside i_size. Trim these off again. Don't need
848 * i_size_read because we hold i_mutex.
850 * Add inode to orphan list in case we crash before
853 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
854 ext4_orphan_add(handle
, inode
);
856 ext4_journal_stop(handle
);
857 if (pos
+ len
> inode
->i_size
) {
858 ext4_truncate_failed_write(inode
);
860 * If truncate failed early the inode might
861 * still be on the orphan list; we need to
862 * make sure the inode is removed from the
863 * orphan list in that case.
866 ext4_orphan_del(NULL
, inode
);
870 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
876 /* For write_end() in data=journal mode */
877 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
879 if (!buffer_mapped(bh
) || buffer_freed(bh
))
881 set_buffer_uptodate(bh
);
882 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
885 static int ext4_generic_write_end(struct file
*file
,
886 struct address_space
*mapping
,
887 loff_t pos
, unsigned len
, unsigned copied
,
888 struct page
*page
, void *fsdata
)
890 int i_size_changed
= 0;
891 struct inode
*inode
= mapping
->host
;
892 handle_t
*handle
= ext4_journal_current_handle();
894 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
897 * No need to use i_size_read() here, the i_size
898 * cannot change under us because we hold i_mutex.
900 * But it's important to update i_size while still holding page lock:
901 * page writeout could otherwise come in and zero beyond i_size.
903 if (pos
+ copied
> inode
->i_size
) {
904 i_size_write(inode
, pos
+ copied
);
908 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
909 /* We need to mark inode dirty even if
910 * new_i_size is less that inode->i_size
911 * bu greater than i_disksize.(hint delalloc)
913 ext4_update_i_disksize(inode
, (pos
+ copied
));
917 page_cache_release(page
);
920 * Don't mark the inode dirty under page lock. First, it unnecessarily
921 * makes the holding time of page lock longer. Second, it forces lock
922 * ordering of page lock and transaction start for journaling
926 ext4_mark_inode_dirty(handle
, inode
);
932 * We need to pick up the new inode size which generic_commit_write gave us
933 * `file' can be NULL - eg, when called from page_symlink().
935 * ext4 never places buffers on inode->i_mapping->private_list. metadata
936 * buffers are managed internally.
938 static int ext4_ordered_write_end(struct file
*file
,
939 struct address_space
*mapping
,
940 loff_t pos
, unsigned len
, unsigned copied
,
941 struct page
*page
, void *fsdata
)
943 handle_t
*handle
= ext4_journal_current_handle();
944 struct inode
*inode
= mapping
->host
;
947 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
948 ret
= ext4_jbd2_file_inode(handle
, inode
);
951 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
954 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
955 /* if we have allocated more blocks and copied
956 * less. We will have blocks allocated outside
957 * inode->i_size. So truncate them
959 ext4_orphan_add(handle
, inode
);
964 page_cache_release(page
);
967 ret2
= ext4_journal_stop(handle
);
971 if (pos
+ len
> inode
->i_size
) {
972 ext4_truncate_failed_write(inode
);
974 * If truncate failed early the inode might still be
975 * on the orphan list; we need to make sure the inode
976 * is removed from the orphan list in that case.
979 ext4_orphan_del(NULL
, inode
);
983 return ret
? ret
: copied
;
986 static int ext4_writeback_write_end(struct file
*file
,
987 struct address_space
*mapping
,
988 loff_t pos
, unsigned len
, unsigned copied
,
989 struct page
*page
, void *fsdata
)
991 handle_t
*handle
= ext4_journal_current_handle();
992 struct inode
*inode
= mapping
->host
;
995 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
996 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
999 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1000 /* if we have allocated more blocks and copied
1001 * less. We will have blocks allocated outside
1002 * inode->i_size. So truncate them
1004 ext4_orphan_add(handle
, inode
);
1009 ret2
= ext4_journal_stop(handle
);
1013 if (pos
+ len
> inode
->i_size
) {
1014 ext4_truncate_failed_write(inode
);
1016 * If truncate failed early the inode might still be
1017 * on the orphan list; we need to make sure the inode
1018 * is removed from the orphan list in that case.
1021 ext4_orphan_del(NULL
, inode
);
1024 return ret
? ret
: copied
;
1027 static int ext4_journalled_write_end(struct file
*file
,
1028 struct address_space
*mapping
,
1029 loff_t pos
, unsigned len
, unsigned copied
,
1030 struct page
*page
, void *fsdata
)
1032 handle_t
*handle
= ext4_journal_current_handle();
1033 struct inode
*inode
= mapping
->host
;
1039 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1040 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1043 BUG_ON(!ext4_handle_valid(handle
));
1046 if (!PageUptodate(page
))
1048 page_zero_new_buffers(page
, from
+copied
, to
);
1051 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1052 to
, &partial
, write_end_fn
);
1054 SetPageUptodate(page
);
1055 new_i_size
= pos
+ copied
;
1056 if (new_i_size
> inode
->i_size
)
1057 i_size_write(inode
, pos
+copied
);
1058 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1059 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1060 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1061 ext4_update_i_disksize(inode
, new_i_size
);
1062 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1068 page_cache_release(page
);
1069 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1070 /* if we have allocated more blocks and copied
1071 * less. We will have blocks allocated outside
1072 * inode->i_size. So truncate them
1074 ext4_orphan_add(handle
, inode
);
1076 ret2
= ext4_journal_stop(handle
);
1079 if (pos
+ len
> inode
->i_size
) {
1080 ext4_truncate_failed_write(inode
);
1082 * If truncate failed early the inode might still be
1083 * on the orphan list; we need to make sure the inode
1084 * is removed from the orphan list in that case.
1087 ext4_orphan_del(NULL
, inode
);
1090 return ret
? ret
: copied
;
1094 * Reserve a single cluster located at lblock
1096 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1099 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1100 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1101 unsigned int md_needed
;
1105 * recalculate the amount of metadata blocks to reserve
1106 * in order to allocate nrblocks
1107 * worse case is one extent per block
1110 spin_lock(&ei
->i_block_reservation_lock
);
1111 md_needed
= EXT4_NUM_B2C(sbi
,
1112 ext4_calc_metadata_amount(inode
, lblock
));
1113 trace_ext4_da_reserve_space(inode
, md_needed
);
1114 spin_unlock(&ei
->i_block_reservation_lock
);
1117 * We will charge metadata quota at writeout time; this saves
1118 * us from metadata over-estimation, though we may go over by
1119 * a small amount in the end. Here we just reserve for data.
1121 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1125 * We do still charge estimated metadata to the sb though;
1126 * we cannot afford to run out of free blocks.
1128 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1129 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1130 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1136 spin_lock(&ei
->i_block_reservation_lock
);
1137 ei
->i_reserved_data_blocks
++;
1138 ei
->i_reserved_meta_blocks
+= md_needed
;
1139 spin_unlock(&ei
->i_block_reservation_lock
);
1141 return 0; /* success */
1144 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1146 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1147 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1150 return; /* Nothing to release, exit */
1152 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1154 trace_ext4_da_release_space(inode
, to_free
);
1155 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1157 * if there aren't enough reserved blocks, then the
1158 * counter is messed up somewhere. Since this
1159 * function is called from invalidate page, it's
1160 * harmless to return without any action.
1162 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1163 "ino %lu, to_free %d with only %d reserved "
1164 "data blocks\n", inode
->i_ino
, to_free
,
1165 ei
->i_reserved_data_blocks
);
1167 to_free
= ei
->i_reserved_data_blocks
;
1169 ei
->i_reserved_data_blocks
-= to_free
;
1171 if (ei
->i_reserved_data_blocks
== 0) {
1173 * We can release all of the reserved metadata blocks
1174 * only when we have written all of the delayed
1175 * allocation blocks.
1176 * Note that in case of bigalloc, i_reserved_meta_blocks,
1177 * i_reserved_data_blocks, etc. refer to number of clusters.
1179 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1180 ei
->i_reserved_meta_blocks
);
1181 ei
->i_reserved_meta_blocks
= 0;
1182 ei
->i_da_metadata_calc_len
= 0;
1185 /* update fs dirty data blocks counter */
1186 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1188 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1190 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1193 static void ext4_da_page_release_reservation(struct page
*page
,
1194 unsigned long offset
)
1197 struct buffer_head
*head
, *bh
;
1198 unsigned int curr_off
= 0;
1199 struct inode
*inode
= page
->mapping
->host
;
1200 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1203 head
= page_buffers(page
);
1206 unsigned int next_off
= curr_off
+ bh
->b_size
;
1208 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1210 clear_buffer_delay(bh
);
1211 clear_buffer_da_mapped(bh
);
1213 curr_off
= next_off
;
1214 } while ((bh
= bh
->b_this_page
) != head
);
1216 /* If we have released all the blocks belonging to a cluster, then we
1217 * need to release the reserved space for that cluster. */
1218 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1219 while (num_clusters
> 0) {
1221 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1222 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1223 if (sbi
->s_cluster_ratio
== 1 ||
1224 !ext4_find_delalloc_cluster(inode
, lblk
, 1))
1225 ext4_da_release_space(inode
, 1);
1232 * Delayed allocation stuff
1236 * mpage_da_submit_io - walks through extent of pages and try to write
1237 * them with writepage() call back
1239 * @mpd->inode: inode
1240 * @mpd->first_page: first page of the extent
1241 * @mpd->next_page: page after the last page of the extent
1243 * By the time mpage_da_submit_io() is called we expect all blocks
1244 * to be allocated. this may be wrong if allocation failed.
1246 * As pages are already locked by write_cache_pages(), we can't use it
1248 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1249 struct ext4_map_blocks
*map
)
1251 struct pagevec pvec
;
1252 unsigned long index
, end
;
1253 int ret
= 0, err
, nr_pages
, i
;
1254 struct inode
*inode
= mpd
->inode
;
1255 struct address_space
*mapping
= inode
->i_mapping
;
1256 loff_t size
= i_size_read(inode
);
1257 unsigned int len
, block_start
;
1258 struct buffer_head
*bh
, *page_bufs
= NULL
;
1259 int journal_data
= ext4_should_journal_data(inode
);
1260 sector_t pblock
= 0, cur_logical
= 0;
1261 struct ext4_io_submit io_submit
;
1263 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1264 memset(&io_submit
, 0, sizeof(io_submit
));
1266 * We need to start from the first_page to the next_page - 1
1267 * to make sure we also write the mapped dirty buffer_heads.
1268 * If we look at mpd->b_blocknr we would only be looking
1269 * at the currently mapped buffer_heads.
1271 index
= mpd
->first_page
;
1272 end
= mpd
->next_page
- 1;
1274 pagevec_init(&pvec
, 0);
1275 while (index
<= end
) {
1276 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1279 for (i
= 0; i
< nr_pages
; i
++) {
1280 int commit_write
= 0, skip_page
= 0;
1281 struct page
*page
= pvec
.pages
[i
];
1283 index
= page
->index
;
1287 if (index
== size
>> PAGE_CACHE_SHIFT
)
1288 len
= size
& ~PAGE_CACHE_MASK
;
1290 len
= PAGE_CACHE_SIZE
;
1292 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1294 pblock
= map
->m_pblk
+ (cur_logical
-
1299 BUG_ON(!PageLocked(page
));
1300 BUG_ON(PageWriteback(page
));
1303 * If the page does not have buffers (for
1304 * whatever reason), try to create them using
1305 * __block_write_begin. If this fails,
1306 * skip the page and move on.
1308 if (!page_has_buffers(page
)) {
1309 if (__block_write_begin(page
, 0, len
,
1310 noalloc_get_block_write
)) {
1318 bh
= page_bufs
= page_buffers(page
);
1323 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1324 (cur_logical
<= (map
->m_lblk
+
1325 (map
->m_len
- 1)))) {
1326 if (buffer_delay(bh
)) {
1327 clear_buffer_delay(bh
);
1328 bh
->b_blocknr
= pblock
;
1330 if (buffer_da_mapped(bh
))
1331 clear_buffer_da_mapped(bh
);
1332 if (buffer_unwritten(bh
) ||
1334 BUG_ON(bh
->b_blocknr
!= pblock
);
1335 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1336 set_buffer_uninit(bh
);
1337 clear_buffer_unwritten(bh
);
1340 /* skip page if block allocation undone */
1341 if (buffer_delay(bh
) || buffer_unwritten(bh
))
1343 bh
= bh
->b_this_page
;
1344 block_start
+= bh
->b_size
;
1347 } while (bh
!= page_bufs
);
1353 /* mark the buffer_heads as dirty & uptodate */
1354 block_commit_write(page
, 0, len
);
1356 clear_page_dirty_for_io(page
);
1358 * Delalloc doesn't support data journalling,
1359 * but eventually maybe we'll lift this
1362 if (unlikely(journal_data
&& PageChecked(page
)))
1363 err
= __ext4_journalled_writepage(page
, len
);
1364 else if (test_opt(inode
->i_sb
, MBLK_IO_SUBMIT
))
1365 err
= ext4_bio_write_page(&io_submit
, page
,
1367 else if (buffer_uninit(page_bufs
)) {
1368 ext4_set_bh_endio(page_bufs
, inode
);
1369 err
= block_write_full_page_endio(page
,
1370 noalloc_get_block_write
,
1371 mpd
->wbc
, ext4_end_io_buffer_write
);
1373 err
= block_write_full_page(page
,
1374 noalloc_get_block_write
, mpd
->wbc
);
1377 mpd
->pages_written
++;
1379 * In error case, we have to continue because
1380 * remaining pages are still locked
1385 pagevec_release(&pvec
);
1387 ext4_io_submit(&io_submit
);
1391 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1395 struct pagevec pvec
;
1396 struct inode
*inode
= mpd
->inode
;
1397 struct address_space
*mapping
= inode
->i_mapping
;
1399 index
= mpd
->first_page
;
1400 end
= mpd
->next_page
- 1;
1401 while (index
<= end
) {
1402 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1405 for (i
= 0; i
< nr_pages
; i
++) {
1406 struct page
*page
= pvec
.pages
[i
];
1407 if (page
->index
> end
)
1409 BUG_ON(!PageLocked(page
));
1410 BUG_ON(PageWriteback(page
));
1411 block_invalidatepage(page
, 0);
1412 ClearPageUptodate(page
);
1415 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1416 pagevec_release(&pvec
);
1421 static void ext4_print_free_blocks(struct inode
*inode
)
1423 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1424 printk(KERN_CRIT
"Total free blocks count %lld\n",
1425 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1426 ext4_count_free_clusters(inode
->i_sb
)));
1427 printk(KERN_CRIT
"Free/Dirty block details\n");
1428 printk(KERN_CRIT
"free_blocks=%lld\n",
1429 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1430 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1431 printk(KERN_CRIT
"dirty_blocks=%lld\n",
1432 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1433 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1434 printk(KERN_CRIT
"Block reservation details\n");
1435 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
1436 EXT4_I(inode
)->i_reserved_data_blocks
);
1437 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
1438 EXT4_I(inode
)->i_reserved_meta_blocks
);
1443 * mpage_da_map_and_submit - go through given space, map them
1444 * if necessary, and then submit them for I/O
1446 * @mpd - bh describing space
1448 * The function skips space we know is already mapped to disk blocks.
1451 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1453 int err
, blks
, get_blocks_flags
;
1454 struct ext4_map_blocks map
, *mapp
= NULL
;
1455 sector_t next
= mpd
->b_blocknr
;
1456 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1457 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1458 handle_t
*handle
= NULL
;
1461 * If the blocks are mapped already, or we couldn't accumulate
1462 * any blocks, then proceed immediately to the submission stage.
1464 if ((mpd
->b_size
== 0) ||
1465 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1466 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1467 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1470 handle
= ext4_journal_current_handle();
1474 * Call ext4_map_blocks() to allocate any delayed allocation
1475 * blocks, or to convert an uninitialized extent to be
1476 * initialized (in the case where we have written into
1477 * one or more preallocated blocks).
1479 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1480 * indicate that we are on the delayed allocation path. This
1481 * affects functions in many different parts of the allocation
1482 * call path. This flag exists primarily because we don't
1483 * want to change *many* call functions, so ext4_map_blocks()
1484 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1485 * inode's allocation semaphore is taken.
1487 * If the blocks in questions were delalloc blocks, set
1488 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1489 * variables are updated after the blocks have been allocated.
1492 map
.m_len
= max_blocks
;
1493 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1494 if (ext4_should_dioread_nolock(mpd
->inode
))
1495 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1496 if (mpd
->b_state
& (1 << BH_Delay
))
1497 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1499 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1501 struct super_block
*sb
= mpd
->inode
->i_sb
;
1505 * If get block returns EAGAIN or ENOSPC and there
1506 * appears to be free blocks we will just let
1507 * mpage_da_submit_io() unlock all of the pages.
1512 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1518 * get block failure will cause us to loop in
1519 * writepages, because a_ops->writepage won't be able
1520 * to make progress. The page will be redirtied by
1521 * writepage and writepages will again try to write
1524 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1525 ext4_msg(sb
, KERN_CRIT
,
1526 "delayed block allocation failed for inode %lu "
1527 "at logical offset %llu with max blocks %zd "
1528 "with error %d", mpd
->inode
->i_ino
,
1529 (unsigned long long) next
,
1530 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1531 ext4_msg(sb
, KERN_CRIT
,
1532 "This should not happen!! Data will be lost\n");
1534 ext4_print_free_blocks(mpd
->inode
);
1536 /* invalidate all the pages */
1537 ext4_da_block_invalidatepages(mpd
);
1539 /* Mark this page range as having been completed */
1546 if (map
.m_flags
& EXT4_MAP_NEW
) {
1547 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1550 for (i
= 0; i
< map
.m_len
; i
++)
1551 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1553 if (ext4_should_order_data(mpd
->inode
)) {
1554 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
1556 /* Only if the journal is aborted */
1562 * Update on-disk size along with block allocation.
1564 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1565 if (disksize
> i_size_read(mpd
->inode
))
1566 disksize
= i_size_read(mpd
->inode
);
1567 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1568 ext4_update_i_disksize(mpd
->inode
, disksize
);
1569 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1571 ext4_error(mpd
->inode
->i_sb
,
1572 "Failed to mark inode %lu dirty",
1577 mpage_da_submit_io(mpd
, mapp
);
1581 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1582 (1 << BH_Delay) | (1 << BH_Unwritten))
1585 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1587 * @mpd->lbh - extent of blocks
1588 * @logical - logical number of the block in the file
1589 * @bh - bh of the block (used to access block's state)
1591 * the function is used to collect contig. blocks in same state
1593 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1594 sector_t logical
, size_t b_size
,
1595 unsigned long b_state
)
1598 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1601 * XXX Don't go larger than mballoc is willing to allocate
1602 * This is a stopgap solution. We eventually need to fold
1603 * mpage_da_submit_io() into this function and then call
1604 * ext4_map_blocks() multiple times in a loop
1606 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
1609 /* check if thereserved journal credits might overflow */
1610 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
1611 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1613 * With non-extent format we are limited by the journal
1614 * credit available. Total credit needed to insert
1615 * nrblocks contiguous blocks is dependent on the
1616 * nrblocks. So limit nrblocks.
1619 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1620 EXT4_MAX_TRANS_DATA
) {
1622 * Adding the new buffer_head would make it cross the
1623 * allowed limit for which we have journal credit
1624 * reserved. So limit the new bh->b_size
1626 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1627 mpd
->inode
->i_blkbits
;
1628 /* we will do mpage_da_submit_io in the next loop */
1632 * First block in the extent
1634 if (mpd
->b_size
== 0) {
1635 mpd
->b_blocknr
= logical
;
1636 mpd
->b_size
= b_size
;
1637 mpd
->b_state
= b_state
& BH_FLAGS
;
1641 next
= mpd
->b_blocknr
+ nrblocks
;
1643 * Can we merge the block to our big extent?
1645 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1646 mpd
->b_size
+= b_size
;
1652 * We couldn't merge the block to our extent, so we
1653 * need to flush current extent and start new one
1655 mpage_da_map_and_submit(mpd
);
1659 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1661 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1665 * This function is grabs code from the very beginning of
1666 * ext4_map_blocks, but assumes that the caller is from delayed write
1667 * time. This function looks up the requested blocks and sets the
1668 * buffer delay bit under the protection of i_data_sem.
1670 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1671 struct ext4_map_blocks
*map
,
1672 struct buffer_head
*bh
)
1675 sector_t invalid_block
= ~((sector_t
) 0xffff);
1677 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1681 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1682 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1683 (unsigned long) map
->m_lblk
);
1685 * Try to see if we can get the block without requesting a new
1686 * file system block.
1688 down_read((&EXT4_I(inode
)->i_data_sem
));
1689 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1690 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1692 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1696 * XXX: __block_prepare_write() unmaps passed block,
1699 /* If the block was allocated from previously allocated cluster,
1700 * then we dont need to reserve it again. */
1701 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1702 retval
= ext4_da_reserve_space(inode
, iblock
);
1704 /* not enough space to reserve */
1708 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1709 * and it should not appear on the bh->b_state.
1711 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1713 map_bh(bh
, inode
->i_sb
, invalid_block
);
1715 set_buffer_delay(bh
);
1719 up_read((&EXT4_I(inode
)->i_data_sem
));
1725 * This is a special get_blocks_t callback which is used by
1726 * ext4_da_write_begin(). It will either return mapped block or
1727 * reserve space for a single block.
1729 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1730 * We also have b_blocknr = -1 and b_bdev initialized properly
1732 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1733 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1734 * initialized properly.
1736 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1737 struct buffer_head
*bh
, int create
)
1739 struct ext4_map_blocks map
;
1742 BUG_ON(create
== 0);
1743 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1745 map
.m_lblk
= iblock
;
1749 * first, we need to know whether the block is allocated already
1750 * preallocated blocks are unmapped but should treated
1751 * the same as allocated blocks.
1753 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1757 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1758 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1760 if (buffer_unwritten(bh
)) {
1761 /* A delayed write to unwritten bh should be marked
1762 * new and mapped. Mapped ensures that we don't do
1763 * get_block multiple times when we write to the same
1764 * offset and new ensures that we do proper zero out
1765 * for partial write.
1768 set_buffer_mapped(bh
);
1774 * This function is used as a standard get_block_t calback function
1775 * when there is no desire to allocate any blocks. It is used as a
1776 * callback function for block_write_begin() and block_write_full_page().
1777 * These functions should only try to map a single block at a time.
1779 * Since this function doesn't do block allocations even if the caller
1780 * requests it by passing in create=1, it is critically important that
1781 * any caller checks to make sure that any buffer heads are returned
1782 * by this function are either all already mapped or marked for
1783 * delayed allocation before calling block_write_full_page(). Otherwise,
1784 * b_blocknr could be left unitialized, and the page write functions will
1785 * be taken by surprise.
1787 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
1788 struct buffer_head
*bh_result
, int create
)
1790 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
1791 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
1794 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1800 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1806 static int __ext4_journalled_writepage(struct page
*page
,
1809 struct address_space
*mapping
= page
->mapping
;
1810 struct inode
*inode
= mapping
->host
;
1811 struct buffer_head
*page_bufs
;
1812 handle_t
*handle
= NULL
;
1816 ClearPageChecked(page
);
1817 page_bufs
= page_buffers(page
);
1819 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
1820 /* As soon as we unlock the page, it can go away, but we have
1821 * references to buffers so we are safe */
1824 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1825 if (IS_ERR(handle
)) {
1826 ret
= PTR_ERR(handle
);
1830 BUG_ON(!ext4_handle_valid(handle
));
1832 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1833 do_journal_get_write_access
);
1835 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1839 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1840 err
= ext4_journal_stop(handle
);
1844 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
1845 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1850 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
1851 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
1854 * Note that we don't need to start a transaction unless we're journaling data
1855 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1856 * need to file the inode to the transaction's list in ordered mode because if
1857 * we are writing back data added by write(), the inode is already there and if
1858 * we are writing back data modified via mmap(), no one guarantees in which
1859 * transaction the data will hit the disk. In case we are journaling data, we
1860 * cannot start transaction directly because transaction start ranks above page
1861 * lock so we have to do some magic.
1863 * This function can get called via...
1864 * - ext4_da_writepages after taking page lock (have journal handle)
1865 * - journal_submit_inode_data_buffers (no journal handle)
1866 * - shrink_page_list via pdflush (no journal handle)
1867 * - grab_page_cache when doing write_begin (have journal handle)
1869 * We don't do any block allocation in this function. If we have page with
1870 * multiple blocks we need to write those buffer_heads that are mapped. This
1871 * is important for mmaped based write. So if we do with blocksize 1K
1872 * truncate(f, 1024);
1873 * a = mmap(f, 0, 4096);
1875 * truncate(f, 4096);
1876 * we have in the page first buffer_head mapped via page_mkwrite call back
1877 * but other bufer_heads would be unmapped but dirty(dirty done via the
1878 * do_wp_page). So writepage should write the first block. If we modify
1879 * the mmap area beyond 1024 we will again get a page_fault and the
1880 * page_mkwrite callback will do the block allocation and mark the
1881 * buffer_heads mapped.
1883 * We redirty the page if we have any buffer_heads that is either delay or
1884 * unwritten in the page.
1886 * We can get recursively called as show below.
1888 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1891 * But since we don't do any block allocation we should not deadlock.
1892 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1894 static int ext4_writepage(struct page
*page
,
1895 struct writeback_control
*wbc
)
1897 int ret
= 0, commit_write
= 0;
1900 struct buffer_head
*page_bufs
= NULL
;
1901 struct inode
*inode
= page
->mapping
->host
;
1903 trace_ext4_writepage(page
);
1904 size
= i_size_read(inode
);
1905 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1906 len
= size
& ~PAGE_CACHE_MASK
;
1908 len
= PAGE_CACHE_SIZE
;
1911 * If the page does not have buffers (for whatever reason),
1912 * try to create them using __block_write_begin. If this
1913 * fails, redirty the page and move on.
1915 if (!page_has_buffers(page
)) {
1916 if (__block_write_begin(page
, 0, len
,
1917 noalloc_get_block_write
)) {
1919 redirty_page_for_writepage(wbc
, page
);
1925 page_bufs
= page_buffers(page
);
1926 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1927 ext4_bh_delay_or_unwritten
)) {
1929 * We don't want to do block allocation, so redirty
1930 * the page and return. We may reach here when we do
1931 * a journal commit via journal_submit_inode_data_buffers.
1932 * We can also reach here via shrink_page_list
1937 /* now mark the buffer_heads as dirty and uptodate */
1938 block_commit_write(page
, 0, len
);
1940 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1942 * It's mmapped pagecache. Add buffers and journal it. There
1943 * doesn't seem much point in redirtying the page here.
1945 return __ext4_journalled_writepage(page
, len
);
1947 if (buffer_uninit(page_bufs
)) {
1948 ext4_set_bh_endio(page_bufs
, inode
);
1949 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
1950 wbc
, ext4_end_io_buffer_write
);
1952 ret
= block_write_full_page(page
, noalloc_get_block_write
,
1959 * This is called via ext4_da_writepages() to
1960 * calculate the total number of credits to reserve to fit
1961 * a single extent allocation into a single transaction,
1962 * ext4_da_writpeages() will loop calling this before
1963 * the block allocation.
1966 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
1968 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
1971 * With non-extent format the journal credit needed to
1972 * insert nrblocks contiguous block is dependent on
1973 * number of contiguous block. So we will limit
1974 * number of contiguous block to a sane value
1976 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
1977 (max_blocks
> EXT4_MAX_TRANS_DATA
))
1978 max_blocks
= EXT4_MAX_TRANS_DATA
;
1980 return ext4_chunk_trans_blocks(inode
, max_blocks
);
1984 * write_cache_pages_da - walk the list of dirty pages of the given
1985 * address space and accumulate pages that need writing, and call
1986 * mpage_da_map_and_submit to map a single contiguous memory region
1987 * and then write them.
1989 static int write_cache_pages_da(struct address_space
*mapping
,
1990 struct writeback_control
*wbc
,
1991 struct mpage_da_data
*mpd
,
1992 pgoff_t
*done_index
)
1994 struct buffer_head
*bh
, *head
;
1995 struct inode
*inode
= mapping
->host
;
1996 struct pagevec pvec
;
1997 unsigned int nr_pages
;
2000 long nr_to_write
= wbc
->nr_to_write
;
2001 int i
, tag
, ret
= 0;
2003 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2006 pagevec_init(&pvec
, 0);
2007 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2008 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2010 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2011 tag
= PAGECACHE_TAG_TOWRITE
;
2013 tag
= PAGECACHE_TAG_DIRTY
;
2015 *done_index
= index
;
2016 while (index
<= end
) {
2017 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2018 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2022 for (i
= 0; i
< nr_pages
; i
++) {
2023 struct page
*page
= pvec
.pages
[i
];
2026 * At this point, the page may be truncated or
2027 * invalidated (changing page->mapping to NULL), or
2028 * even swizzled back from swapper_space to tmpfs file
2029 * mapping. However, page->index will not change
2030 * because we have a reference on the page.
2032 if (page
->index
> end
)
2035 *done_index
= page
->index
+ 1;
2038 * If we can't merge this page, and we have
2039 * accumulated an contiguous region, write it
2041 if ((mpd
->next_page
!= page
->index
) &&
2042 (mpd
->next_page
!= mpd
->first_page
)) {
2043 mpage_da_map_and_submit(mpd
);
2044 goto ret_extent_tail
;
2050 * If the page is no longer dirty, or its
2051 * mapping no longer corresponds to inode we
2052 * are writing (which means it has been
2053 * truncated or invalidated), or the page is
2054 * already under writeback and we are not
2055 * doing a data integrity writeback, skip the page
2057 if (!PageDirty(page
) ||
2058 (PageWriteback(page
) &&
2059 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2060 unlikely(page
->mapping
!= mapping
)) {
2065 wait_on_page_writeback(page
);
2066 BUG_ON(PageWriteback(page
));
2068 if (mpd
->next_page
!= page
->index
)
2069 mpd
->first_page
= page
->index
;
2070 mpd
->next_page
= page
->index
+ 1;
2071 logical
= (sector_t
) page
->index
<<
2072 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2074 if (!page_has_buffers(page
)) {
2075 mpage_add_bh_to_extent(mpd
, logical
,
2077 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2079 goto ret_extent_tail
;
2082 * Page with regular buffer heads,
2083 * just add all dirty ones
2085 head
= page_buffers(page
);
2088 BUG_ON(buffer_locked(bh
));
2090 * We need to try to allocate
2091 * unmapped blocks in the same page.
2092 * Otherwise we won't make progress
2093 * with the page in ext4_writepage
2095 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2096 mpage_add_bh_to_extent(mpd
, logical
,
2100 goto ret_extent_tail
;
2101 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2103 * mapped dirty buffer. We need
2104 * to update the b_state
2105 * because we look at b_state
2106 * in mpage_da_map_blocks. We
2107 * don't update b_size because
2108 * if we find an unmapped
2109 * buffer_head later we need to
2110 * use the b_state flag of that
2113 if (mpd
->b_size
== 0)
2114 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2117 } while ((bh
= bh
->b_this_page
) != head
);
2120 if (nr_to_write
> 0) {
2122 if (nr_to_write
== 0 &&
2123 wbc
->sync_mode
== WB_SYNC_NONE
)
2125 * We stop writing back only if we are
2126 * not doing integrity sync. In case of
2127 * integrity sync we have to keep going
2128 * because someone may be concurrently
2129 * dirtying pages, and we might have
2130 * synced a lot of newly appeared dirty
2131 * pages, but have not synced all of the
2137 pagevec_release(&pvec
);
2142 ret
= MPAGE_DA_EXTENT_TAIL
;
2144 pagevec_release(&pvec
);
2150 static int ext4_da_writepages(struct address_space
*mapping
,
2151 struct writeback_control
*wbc
)
2154 int range_whole
= 0;
2155 handle_t
*handle
= NULL
;
2156 struct mpage_da_data mpd
;
2157 struct inode
*inode
= mapping
->host
;
2158 int pages_written
= 0;
2159 unsigned int max_pages
;
2160 int range_cyclic
, cycled
= 1, io_done
= 0;
2161 int needed_blocks
, ret
= 0;
2162 long desired_nr_to_write
, nr_to_writebump
= 0;
2163 loff_t range_start
= wbc
->range_start
;
2164 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2165 pgoff_t done_index
= 0;
2167 struct blk_plug plug
;
2169 trace_ext4_da_writepages(inode
, wbc
);
2172 * No pages to write? This is mainly a kludge to avoid starting
2173 * a transaction for special inodes like journal inode on last iput()
2174 * because that could violate lock ordering on umount
2176 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2180 * If the filesystem has aborted, it is read-only, so return
2181 * right away instead of dumping stack traces later on that
2182 * will obscure the real source of the problem. We test
2183 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2184 * the latter could be true if the filesystem is mounted
2185 * read-only, and in that case, ext4_da_writepages should
2186 * *never* be called, so if that ever happens, we would want
2189 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2192 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2195 range_cyclic
= wbc
->range_cyclic
;
2196 if (wbc
->range_cyclic
) {
2197 index
= mapping
->writeback_index
;
2200 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2201 wbc
->range_end
= LLONG_MAX
;
2202 wbc
->range_cyclic
= 0;
2205 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2206 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2210 * This works around two forms of stupidity. The first is in
2211 * the writeback code, which caps the maximum number of pages
2212 * written to be 1024 pages. This is wrong on multiple
2213 * levels; different architectues have a different page size,
2214 * which changes the maximum amount of data which gets
2215 * written. Secondly, 4 megabytes is way too small. XFS
2216 * forces this value to be 16 megabytes by multiplying
2217 * nr_to_write parameter by four, and then relies on its
2218 * allocator to allocate larger extents to make them
2219 * contiguous. Unfortunately this brings us to the second
2220 * stupidity, which is that ext4's mballoc code only allocates
2221 * at most 2048 blocks. So we force contiguous writes up to
2222 * the number of dirty blocks in the inode, or
2223 * sbi->max_writeback_mb_bump whichever is smaller.
2225 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2226 if (!range_cyclic
&& range_whole
) {
2227 if (wbc
->nr_to_write
== LONG_MAX
)
2228 desired_nr_to_write
= wbc
->nr_to_write
;
2230 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2232 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2234 if (desired_nr_to_write
> max_pages
)
2235 desired_nr_to_write
= max_pages
;
2237 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2238 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2239 wbc
->nr_to_write
= desired_nr_to_write
;
2243 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2244 tag_pages_for_writeback(mapping
, index
, end
);
2246 blk_start_plug(&plug
);
2247 while (!ret
&& wbc
->nr_to_write
> 0) {
2250 * we insert one extent at a time. So we need
2251 * credit needed for single extent allocation.
2252 * journalled mode is currently not supported
2255 BUG_ON(ext4_should_journal_data(inode
));
2256 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2258 /* start a new transaction*/
2259 handle
= ext4_journal_start(inode
, needed_blocks
);
2260 if (IS_ERR(handle
)) {
2261 ret
= PTR_ERR(handle
);
2262 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2263 "%ld pages, ino %lu; err %d", __func__
,
2264 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2265 goto out_writepages
;
2269 * Now call write_cache_pages_da() to find the next
2270 * contiguous region of logical blocks that need
2271 * blocks to be allocated by ext4 and submit them.
2273 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
, &done_index
);
2275 * If we have a contiguous extent of pages and we
2276 * haven't done the I/O yet, map the blocks and submit
2279 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2280 mpage_da_map_and_submit(&mpd
);
2281 ret
= MPAGE_DA_EXTENT_TAIL
;
2283 trace_ext4_da_write_pages(inode
, &mpd
);
2284 wbc
->nr_to_write
-= mpd
.pages_written
;
2286 ext4_journal_stop(handle
);
2288 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2289 /* commit the transaction which would
2290 * free blocks released in the transaction
2293 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2295 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2297 * got one extent now try with
2300 pages_written
+= mpd
.pages_written
;
2303 } else if (wbc
->nr_to_write
)
2305 * There is no more writeout needed
2306 * or we requested for a noblocking writeout
2307 * and we found the device congested
2311 blk_finish_plug(&plug
);
2312 if (!io_done
&& !cycled
) {
2315 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2316 wbc
->range_end
= mapping
->writeback_index
- 1;
2321 wbc
->range_cyclic
= range_cyclic
;
2322 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2324 * set the writeback_index so that range_cyclic
2325 * mode will write it back later
2327 mapping
->writeback_index
= done_index
;
2330 wbc
->nr_to_write
-= nr_to_writebump
;
2331 wbc
->range_start
= range_start
;
2332 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2336 #define FALL_BACK_TO_NONDELALLOC 1
2337 static int ext4_nonda_switch(struct super_block
*sb
)
2339 s64 free_blocks
, dirty_blocks
;
2340 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2343 * switch to non delalloc mode if we are running low
2344 * on free block. The free block accounting via percpu
2345 * counters can get slightly wrong with percpu_counter_batch getting
2346 * accumulated on each CPU without updating global counters
2347 * Delalloc need an accurate free block accounting. So switch
2348 * to non delalloc when we are near to error range.
2350 free_blocks
= EXT4_C2B(sbi
,
2351 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
2352 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2353 if (2 * free_blocks
< 3 * dirty_blocks
||
2354 free_blocks
< (dirty_blocks
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2356 * free block count is less than 150% of dirty blocks
2357 * or free blocks is less than watermark
2362 * Even if we don't switch but are nearing capacity,
2363 * start pushing delalloc when 1/2 of free blocks are dirty.
2365 if (free_blocks
< 2 * dirty_blocks
)
2366 writeback_inodes_sb_if_idle(sb
);
2371 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2372 loff_t pos
, unsigned len
, unsigned flags
,
2373 struct page
**pagep
, void **fsdata
)
2375 int ret
, retries
= 0;
2378 struct inode
*inode
= mapping
->host
;
2382 index
= pos
>> PAGE_CACHE_SHIFT
;
2384 if (ext4_nonda_switch(inode
->i_sb
)) {
2385 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2386 return ext4_write_begin(file
, mapping
, pos
,
2387 len
, flags
, pagep
, fsdata
);
2389 *fsdata
= (void *)0;
2390 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2393 * With delayed allocation, we don't log the i_disksize update
2394 * if there is delayed block allocation. But we still need
2395 * to journalling the i_disksize update if writes to the end
2396 * of file which has an already mapped buffer.
2398 handle
= ext4_journal_start(inode
, 1);
2399 if (IS_ERR(handle
)) {
2400 ret
= PTR_ERR(handle
);
2403 /* We cannot recurse into the filesystem as the transaction is already
2405 flags
|= AOP_FLAG_NOFS
;
2407 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2409 ext4_journal_stop(handle
);
2415 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2418 ext4_journal_stop(handle
);
2419 page_cache_release(page
);
2421 * block_write_begin may have instantiated a few blocks
2422 * outside i_size. Trim these off again. Don't need
2423 * i_size_read because we hold i_mutex.
2425 if (pos
+ len
> inode
->i_size
)
2426 ext4_truncate_failed_write(inode
);
2428 page_len
= pos
& (PAGE_CACHE_SIZE
- 1);
2430 ret
= ext4_discard_partial_page_buffers_no_lock(handle
,
2431 inode
, page
, pos
- page_len
, page_len
,
2432 EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
);
2436 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2443 * Check if we should update i_disksize
2444 * when write to the end of file but not require block allocation
2446 static int ext4_da_should_update_i_disksize(struct page
*page
,
2447 unsigned long offset
)
2449 struct buffer_head
*bh
;
2450 struct inode
*inode
= page
->mapping
->host
;
2454 bh
= page_buffers(page
);
2455 idx
= offset
>> inode
->i_blkbits
;
2457 for (i
= 0; i
< idx
; i
++)
2458 bh
= bh
->b_this_page
;
2460 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2465 static int ext4_da_write_end(struct file
*file
,
2466 struct address_space
*mapping
,
2467 loff_t pos
, unsigned len
, unsigned copied
,
2468 struct page
*page
, void *fsdata
)
2470 struct inode
*inode
= mapping
->host
;
2472 handle_t
*handle
= ext4_journal_current_handle();
2474 unsigned long start
, end
;
2475 int write_mode
= (int)(unsigned long)fsdata
;
2478 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2479 if (ext4_should_order_data(inode
)) {
2480 return ext4_ordered_write_end(file
, mapping
, pos
,
2481 len
, copied
, page
, fsdata
);
2482 } else if (ext4_should_writeback_data(inode
)) {
2483 return ext4_writeback_write_end(file
, mapping
, pos
,
2484 len
, copied
, page
, fsdata
);
2490 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2491 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2492 end
= start
+ copied
- 1;
2495 * generic_write_end() will run mark_inode_dirty() if i_size
2496 * changes. So let's piggyback the i_disksize mark_inode_dirty
2500 new_i_size
= pos
+ copied
;
2501 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2502 if (ext4_da_should_update_i_disksize(page
, end
)) {
2503 down_write(&EXT4_I(inode
)->i_data_sem
);
2504 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2506 * Updating i_disksize when extending file
2507 * without needing block allocation
2509 if (ext4_should_order_data(inode
))
2510 ret
= ext4_jbd2_file_inode(handle
,
2513 EXT4_I(inode
)->i_disksize
= new_i_size
;
2515 up_write(&EXT4_I(inode
)->i_data_sem
);
2516 /* We need to mark inode dirty even if
2517 * new_i_size is less that inode->i_size
2518 * bu greater than i_disksize.(hint delalloc)
2520 ext4_mark_inode_dirty(handle
, inode
);
2523 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2526 page_len
= PAGE_CACHE_SIZE
-
2527 ((pos
+ copied
- 1) & (PAGE_CACHE_SIZE
- 1));
2530 ret
= ext4_discard_partial_page_buffers_no_lock(handle
,
2531 inode
, page
, pos
+ copied
- 1, page_len
,
2532 EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
);
2538 ret2
= ext4_journal_stop(handle
);
2542 return ret
? ret
: copied
;
2545 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2548 * Drop reserved blocks
2550 BUG_ON(!PageLocked(page
));
2551 if (!page_has_buffers(page
))
2554 ext4_da_page_release_reservation(page
, offset
);
2557 ext4_invalidatepage(page
, offset
);
2563 * Force all delayed allocation blocks to be allocated for a given inode.
2565 int ext4_alloc_da_blocks(struct inode
*inode
)
2567 trace_ext4_alloc_da_blocks(inode
);
2569 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2570 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2574 * We do something simple for now. The filemap_flush() will
2575 * also start triggering a write of the data blocks, which is
2576 * not strictly speaking necessary (and for users of
2577 * laptop_mode, not even desirable). However, to do otherwise
2578 * would require replicating code paths in:
2580 * ext4_da_writepages() ->
2581 * write_cache_pages() ---> (via passed in callback function)
2582 * __mpage_da_writepage() -->
2583 * mpage_add_bh_to_extent()
2584 * mpage_da_map_blocks()
2586 * The problem is that write_cache_pages(), located in
2587 * mm/page-writeback.c, marks pages clean in preparation for
2588 * doing I/O, which is not desirable if we're not planning on
2591 * We could call write_cache_pages(), and then redirty all of
2592 * the pages by calling redirty_page_for_writepage() but that
2593 * would be ugly in the extreme. So instead we would need to
2594 * replicate parts of the code in the above functions,
2595 * simplifying them because we wouldn't actually intend to
2596 * write out the pages, but rather only collect contiguous
2597 * logical block extents, call the multi-block allocator, and
2598 * then update the buffer heads with the block allocations.
2600 * For now, though, we'll cheat by calling filemap_flush(),
2601 * which will map the blocks, and start the I/O, but not
2602 * actually wait for the I/O to complete.
2604 return filemap_flush(inode
->i_mapping
);
2608 * bmap() is special. It gets used by applications such as lilo and by
2609 * the swapper to find the on-disk block of a specific piece of data.
2611 * Naturally, this is dangerous if the block concerned is still in the
2612 * journal. If somebody makes a swapfile on an ext4 data-journaling
2613 * filesystem and enables swap, then they may get a nasty shock when the
2614 * data getting swapped to that swapfile suddenly gets overwritten by
2615 * the original zero's written out previously to the journal and
2616 * awaiting writeback in the kernel's buffer cache.
2618 * So, if we see any bmap calls here on a modified, data-journaled file,
2619 * take extra steps to flush any blocks which might be in the cache.
2621 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2623 struct inode
*inode
= mapping
->host
;
2627 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2628 test_opt(inode
->i_sb
, DELALLOC
)) {
2630 * With delalloc we want to sync the file
2631 * so that we can make sure we allocate
2634 filemap_write_and_wait(mapping
);
2637 if (EXT4_JOURNAL(inode
) &&
2638 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2640 * This is a REALLY heavyweight approach, but the use of
2641 * bmap on dirty files is expected to be extremely rare:
2642 * only if we run lilo or swapon on a freshly made file
2643 * do we expect this to happen.
2645 * (bmap requires CAP_SYS_RAWIO so this does not
2646 * represent an unprivileged user DOS attack --- we'd be
2647 * in trouble if mortal users could trigger this path at
2650 * NB. EXT4_STATE_JDATA is not set on files other than
2651 * regular files. If somebody wants to bmap a directory
2652 * or symlink and gets confused because the buffer
2653 * hasn't yet been flushed to disk, they deserve
2654 * everything they get.
2657 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2658 journal
= EXT4_JOURNAL(inode
);
2659 jbd2_journal_lock_updates(journal
);
2660 err
= jbd2_journal_flush(journal
);
2661 jbd2_journal_unlock_updates(journal
);
2667 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2670 static int ext4_readpage(struct file
*file
, struct page
*page
)
2672 trace_ext4_readpage(page
);
2673 return mpage_readpage(page
, ext4_get_block
);
2677 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2678 struct list_head
*pages
, unsigned nr_pages
)
2680 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2683 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
2685 struct buffer_head
*head
, *bh
;
2686 unsigned int curr_off
= 0;
2688 if (!page_has_buffers(page
))
2690 head
= bh
= page_buffers(page
);
2692 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
2694 ext4_free_io_end(bh
->b_private
);
2695 bh
->b_private
= NULL
;
2696 bh
->b_end_io
= NULL
;
2698 curr_off
= curr_off
+ bh
->b_size
;
2699 bh
= bh
->b_this_page
;
2700 } while (bh
!= head
);
2703 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2705 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2707 trace_ext4_invalidatepage(page
, offset
);
2710 * free any io_end structure allocated for buffers to be discarded
2712 if (ext4_should_dioread_nolock(page
->mapping
->host
))
2713 ext4_invalidatepage_free_endio(page
, offset
);
2715 * If it's a full truncate we just forget about the pending dirtying
2718 ClearPageChecked(page
);
2721 jbd2_journal_invalidatepage(journal
, page
, offset
);
2723 block_invalidatepage(page
, offset
);
2726 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2728 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2730 trace_ext4_releasepage(page
);
2732 WARN_ON(PageChecked(page
));
2733 if (!page_has_buffers(page
))
2736 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2738 return try_to_free_buffers(page
);
2742 * ext4_get_block used when preparing for a DIO write or buffer write.
2743 * We allocate an uinitialized extent if blocks haven't been allocated.
2744 * The extent will be converted to initialized after the IO is complete.
2746 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2747 struct buffer_head
*bh_result
, int create
)
2749 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2750 inode
->i_ino
, create
);
2751 return _ext4_get_block(inode
, iblock
, bh_result
,
2752 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2755 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2756 ssize_t size
, void *private, int ret
,
2759 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2760 ext4_io_end_t
*io_end
= iocb
->private;
2761 struct workqueue_struct
*wq
;
2762 unsigned long flags
;
2763 struct ext4_inode_info
*ei
;
2765 /* if not async direct IO or dio with 0 bytes write, just return */
2766 if (!io_end
|| !size
)
2769 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2770 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2771 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2774 /* if not aio dio with unwritten extents, just free io and return */
2775 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2776 ext4_free_io_end(io_end
);
2777 iocb
->private = NULL
;
2780 aio_complete(iocb
, ret
, 0);
2781 inode_dio_done(inode
);
2785 io_end
->offset
= offset
;
2786 io_end
->size
= size
;
2788 io_end
->iocb
= iocb
;
2789 io_end
->result
= ret
;
2791 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
2793 /* Add the io_end to per-inode completed aio dio list*/
2794 ei
= EXT4_I(io_end
->inode
);
2795 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
2796 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
2797 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
2799 /* queue the work to convert unwritten extents to written */
2800 queue_work(wq
, &io_end
->work
);
2801 iocb
->private = NULL
;
2803 /* XXX: probably should move into the real I/O completion handler */
2804 inode_dio_done(inode
);
2807 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
2809 ext4_io_end_t
*io_end
= bh
->b_private
;
2810 struct workqueue_struct
*wq
;
2811 struct inode
*inode
;
2812 unsigned long flags
;
2814 if (!test_clear_buffer_uninit(bh
) || !io_end
)
2817 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
2818 printk("sb umounted, discard end_io request for inode %lu\n",
2819 io_end
->inode
->i_ino
);
2820 ext4_free_io_end(io_end
);
2825 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2826 * but being more careful is always safe for the future change.
2828 inode
= io_end
->inode
;
2829 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2830 io_end
->flag
|= EXT4_IO_END_UNWRITTEN
;
2831 atomic_inc(&EXT4_I(inode
)->i_aiodio_unwritten
);
2834 /* Add the io_end to per-inode completed io list*/
2835 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2836 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
2837 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2839 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
2840 /* queue the work to convert unwritten extents to written */
2841 queue_work(wq
, &io_end
->work
);
2843 bh
->b_private
= NULL
;
2844 bh
->b_end_io
= NULL
;
2845 clear_buffer_uninit(bh
);
2846 end_buffer_async_write(bh
, uptodate
);
2849 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
2851 ext4_io_end_t
*io_end
;
2852 struct page
*page
= bh
->b_page
;
2853 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
2854 size_t size
= bh
->b_size
;
2857 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
2859 pr_warn_ratelimited("%s: allocation fail\n", __func__
);
2863 io_end
->offset
= offset
;
2864 io_end
->size
= size
;
2866 * We need to hold a reference to the page to make sure it
2867 * doesn't get evicted before ext4_end_io_work() has a chance
2868 * to convert the extent from written to unwritten.
2870 io_end
->page
= page
;
2871 get_page(io_end
->page
);
2873 bh
->b_private
= io_end
;
2874 bh
->b_end_io
= ext4_end_io_buffer_write
;
2879 * For ext4 extent files, ext4 will do direct-io write to holes,
2880 * preallocated extents, and those write extend the file, no need to
2881 * fall back to buffered IO.
2883 * For holes, we fallocate those blocks, mark them as uninitialized
2884 * If those blocks were preallocated, we mark sure they are splited, but
2885 * still keep the range to write as uninitialized.
2887 * The unwrritten extents will be converted to written when DIO is completed.
2888 * For async direct IO, since the IO may still pending when return, we
2889 * set up an end_io call back function, which will do the conversion
2890 * when async direct IO completed.
2892 * If the O_DIRECT write will extend the file then add this inode to the
2893 * orphan list. So recovery will truncate it back to the original size
2894 * if the machine crashes during the write.
2897 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
2898 const struct iovec
*iov
, loff_t offset
,
2899 unsigned long nr_segs
)
2901 struct file
*file
= iocb
->ki_filp
;
2902 struct inode
*inode
= file
->f_mapping
->host
;
2904 size_t count
= iov_length(iov
, nr_segs
);
2906 loff_t final_size
= offset
+ count
;
2907 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
2909 * We could direct write to holes and fallocate.
2911 * Allocated blocks to fill the hole are marked as uninitialized
2912 * to prevent parallel buffered read to expose the stale data
2913 * before DIO complete the data IO.
2915 * As to previously fallocated extents, ext4 get_block
2916 * will just simply mark the buffer mapped but still
2917 * keep the extents uninitialized.
2919 * for non AIO case, we will convert those unwritten extents
2920 * to written after return back from blockdev_direct_IO.
2922 * for async DIO, the conversion needs to be defered when
2923 * the IO is completed. The ext4 end_io callback function
2924 * will be called to take care of the conversion work.
2925 * Here for async case, we allocate an io_end structure to
2928 iocb
->private = NULL
;
2929 EXT4_I(inode
)->cur_aio_dio
= NULL
;
2930 if (!is_sync_kiocb(iocb
)) {
2931 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
2935 * we save the io structure for current async
2936 * direct IO, so that later ext4_map_blocks()
2937 * could flag the io structure whether there
2938 * is a unwritten extents needs to be converted
2939 * when IO is completed.
2941 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
2944 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
2945 inode
->i_sb
->s_bdev
, iov
,
2947 ext4_get_block_write
,
2950 DIO_LOCKING
| DIO_SKIP_HOLES
);
2952 EXT4_I(inode
)->cur_aio_dio
= NULL
;
2954 * The io_end structure takes a reference to the inode,
2955 * that structure needs to be destroyed and the
2956 * reference to the inode need to be dropped, when IO is
2957 * complete, even with 0 byte write, or failed.
2959 * In the successful AIO DIO case, the io_end structure will be
2960 * desctroyed and the reference to the inode will be dropped
2961 * after the end_io call back function is called.
2963 * In the case there is 0 byte write, or error case, since
2964 * VFS direct IO won't invoke the end_io call back function,
2965 * we need to free the end_io structure here.
2967 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
2968 ext4_free_io_end(iocb
->private);
2969 iocb
->private = NULL
;
2970 } else if (ret
> 0 && ext4_test_inode_state(inode
,
2971 EXT4_STATE_DIO_UNWRITTEN
)) {
2974 * for non AIO case, since the IO is already
2975 * completed, we could do the conversion right here
2977 err
= ext4_convert_unwritten_extents(inode
,
2981 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
2986 /* for write the the end of file case, we fall back to old way */
2987 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2990 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2991 const struct iovec
*iov
, loff_t offset
,
2992 unsigned long nr_segs
)
2994 struct file
*file
= iocb
->ki_filp
;
2995 struct inode
*inode
= file
->f_mapping
->host
;
2999 * If we are doing data journalling we don't support O_DIRECT
3001 if (ext4_should_journal_data(inode
))
3004 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3005 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3006 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3008 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3009 trace_ext4_direct_IO_exit(inode
, offset
,
3010 iov_length(iov
, nr_segs
), rw
, ret
);
3015 * Pages can be marked dirty completely asynchronously from ext4's journalling
3016 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3017 * much here because ->set_page_dirty is called under VFS locks. The page is
3018 * not necessarily locked.
3020 * We cannot just dirty the page and leave attached buffers clean, because the
3021 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3022 * or jbddirty because all the journalling code will explode.
3024 * So what we do is to mark the page "pending dirty" and next time writepage
3025 * is called, propagate that into the buffers appropriately.
3027 static int ext4_journalled_set_page_dirty(struct page
*page
)
3029 SetPageChecked(page
);
3030 return __set_page_dirty_nobuffers(page
);
3033 static const struct address_space_operations ext4_ordered_aops
= {
3034 .readpage
= ext4_readpage
,
3035 .readpages
= ext4_readpages
,
3036 .writepage
= ext4_writepage
,
3037 .write_begin
= ext4_write_begin
,
3038 .write_end
= ext4_ordered_write_end
,
3040 .invalidatepage
= ext4_invalidatepage
,
3041 .releasepage
= ext4_releasepage
,
3042 .direct_IO
= ext4_direct_IO
,
3043 .migratepage
= buffer_migrate_page
,
3044 .is_partially_uptodate
= block_is_partially_uptodate
,
3045 .error_remove_page
= generic_error_remove_page
,
3048 static const struct address_space_operations ext4_writeback_aops
= {
3049 .readpage
= ext4_readpage
,
3050 .readpages
= ext4_readpages
,
3051 .writepage
= ext4_writepage
,
3052 .write_begin
= ext4_write_begin
,
3053 .write_end
= ext4_writeback_write_end
,
3055 .invalidatepage
= ext4_invalidatepage
,
3056 .releasepage
= ext4_releasepage
,
3057 .direct_IO
= ext4_direct_IO
,
3058 .migratepage
= buffer_migrate_page
,
3059 .is_partially_uptodate
= block_is_partially_uptodate
,
3060 .error_remove_page
= generic_error_remove_page
,
3063 static const struct address_space_operations ext4_journalled_aops
= {
3064 .readpage
= ext4_readpage
,
3065 .readpages
= ext4_readpages
,
3066 .writepage
= ext4_writepage
,
3067 .write_begin
= ext4_write_begin
,
3068 .write_end
= ext4_journalled_write_end
,
3069 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3071 .invalidatepage
= ext4_invalidatepage
,
3072 .releasepage
= ext4_releasepage
,
3073 .direct_IO
= ext4_direct_IO
,
3074 .is_partially_uptodate
= block_is_partially_uptodate
,
3075 .error_remove_page
= generic_error_remove_page
,
3078 static const struct address_space_operations ext4_da_aops
= {
3079 .readpage
= ext4_readpage
,
3080 .readpages
= ext4_readpages
,
3081 .writepage
= ext4_writepage
,
3082 .writepages
= ext4_da_writepages
,
3083 .write_begin
= ext4_da_write_begin
,
3084 .write_end
= ext4_da_write_end
,
3086 .invalidatepage
= ext4_da_invalidatepage
,
3087 .releasepage
= ext4_releasepage
,
3088 .direct_IO
= ext4_direct_IO
,
3089 .migratepage
= buffer_migrate_page
,
3090 .is_partially_uptodate
= block_is_partially_uptodate
,
3091 .error_remove_page
= generic_error_remove_page
,
3094 void ext4_set_aops(struct inode
*inode
)
3096 if (ext4_should_order_data(inode
) &&
3097 test_opt(inode
->i_sb
, DELALLOC
))
3098 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3099 else if (ext4_should_order_data(inode
))
3100 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3101 else if (ext4_should_writeback_data(inode
) &&
3102 test_opt(inode
->i_sb
, DELALLOC
))
3103 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3104 else if (ext4_should_writeback_data(inode
))
3105 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3107 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3112 * ext4_discard_partial_page_buffers()
3113 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3114 * This function finds and locks the page containing the offset
3115 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3116 * Calling functions that already have the page locked should call
3117 * ext4_discard_partial_page_buffers_no_lock directly.
3119 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3120 struct address_space
*mapping
, loff_t from
,
3121 loff_t length
, int flags
)
3123 struct inode
*inode
= mapping
->host
;
3127 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3128 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3132 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3133 from
, length
, flags
);
3136 page_cache_release(page
);
3141 * ext4_discard_partial_page_buffers_no_lock()
3142 * Zeros a page range of length 'length' starting from offset 'from'.
3143 * Buffer heads that correspond to the block aligned regions of the
3144 * zeroed range will be unmapped. Unblock aligned regions
3145 * will have the corresponding buffer head mapped if needed so that
3146 * that region of the page can be updated with the partial zero out.
3148 * This function assumes that the page has already been locked. The
3149 * The range to be discarded must be contained with in the given page.
3150 * If the specified range exceeds the end of the page it will be shortened
3151 * to the end of the page that corresponds to 'from'. This function is
3152 * appropriate for updating a page and it buffer heads to be unmapped and
3153 * zeroed for blocks that have been either released, or are going to be
3156 * handle: The journal handle
3157 * inode: The files inode
3158 * page: A locked page that contains the offset "from"
3159 * from: The starting byte offset (from the begining of the file)
3160 * to begin discarding
3161 * len: The length of bytes to discard
3162 * flags: Optional flags that may be used:
3164 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3165 * Only zero the regions of the page whose buffer heads
3166 * have already been unmapped. This flag is appropriate
3167 * for updateing the contents of a page whose blocks may
3168 * have already been released, and we only want to zero
3169 * out the regions that correspond to those released blocks.
3171 * Returns zero on sucess or negative on failure.
3173 int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3174 struct inode
*inode
, struct page
*page
, loff_t from
,
3175 loff_t length
, int flags
)
3177 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3178 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3179 unsigned int blocksize
, max
, pos
;
3180 unsigned int end_of_block
, range_to_discard
;
3182 struct buffer_head
*bh
;
3185 blocksize
= inode
->i_sb
->s_blocksize
;
3186 max
= PAGE_CACHE_SIZE
- offset
;
3188 if (index
!= page
->index
)
3192 * correct length if it does not fall between
3193 * 'from' and the end of the page
3195 if (length
> max
|| length
< 0)
3198 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3200 if (!page_has_buffers(page
)) {
3202 * If the range to be discarded covers a partial block
3203 * we need to get the page buffers. This is because
3204 * partial blocks cannot be released and the page needs
3205 * to be updated with the contents of the block before
3206 * we write the zeros on top of it.
3208 if (!(from
& (blocksize
- 1)) ||
3209 !((from
+ length
) & (blocksize
- 1))) {
3210 create_empty_buffers(page
, blocksize
, 0);
3213 * If there are no partial blocks,
3214 * there is nothing to update,
3215 * so we can return now
3221 /* Find the buffer that contains "offset" */
3222 bh
= page_buffers(page
);
3224 while (offset
>= pos
) {
3225 bh
= bh
->b_this_page
;
3231 while (pos
< offset
+ length
) {
3234 /* The length of space left to zero and unmap */
3235 range_to_discard
= offset
+ length
- pos
;
3237 /* The length of space until the end of the block */
3238 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3241 * Do not unmap or zero past end of block
3242 * for this buffer head
3244 if (range_to_discard
> end_of_block
)
3245 range_to_discard
= end_of_block
;
3249 * Skip this buffer head if we are only zeroing unampped
3250 * regions of the page
3252 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3256 /* If the range is block aligned, unmap */
3257 if (range_to_discard
== blocksize
) {
3258 clear_buffer_dirty(bh
);
3260 clear_buffer_mapped(bh
);
3261 clear_buffer_req(bh
);
3262 clear_buffer_new(bh
);
3263 clear_buffer_delay(bh
);
3264 clear_buffer_unwritten(bh
);
3265 clear_buffer_uptodate(bh
);
3266 zero_user(page
, pos
, range_to_discard
);
3267 BUFFER_TRACE(bh
, "Buffer discarded");
3272 * If this block is not completely contained in the range
3273 * to be discarded, then it is not going to be released. Because
3274 * we need to keep this block, we need to make sure this part
3275 * of the page is uptodate before we modify it by writeing
3276 * partial zeros on it.
3278 if (!buffer_mapped(bh
)) {
3280 * Buffer head must be mapped before we can read
3283 BUFFER_TRACE(bh
, "unmapped");
3284 ext4_get_block(inode
, iblock
, bh
, 0);
3285 /* unmapped? It's a hole - nothing to do */
3286 if (!buffer_mapped(bh
)) {
3287 BUFFER_TRACE(bh
, "still unmapped");
3292 /* Ok, it's mapped. Make sure it's up-to-date */
3293 if (PageUptodate(page
))
3294 set_buffer_uptodate(bh
);
3296 if (!buffer_uptodate(bh
)) {
3298 ll_rw_block(READ
, 1, &bh
);
3300 /* Uhhuh. Read error. Complain and punt.*/
3301 if (!buffer_uptodate(bh
))
3305 if (ext4_should_journal_data(inode
)) {
3306 BUFFER_TRACE(bh
, "get write access");
3307 err
= ext4_journal_get_write_access(handle
, bh
);
3312 zero_user(page
, pos
, range_to_discard
);
3315 if (ext4_should_journal_data(inode
)) {
3316 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3318 mark_buffer_dirty(bh
);
3320 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3322 bh
= bh
->b_this_page
;
3324 pos
+= range_to_discard
;
3331 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3332 * up to the end of the block which corresponds to `from'.
3333 * This required during truncate. We need to physically zero the tail end
3334 * of that block so it doesn't yield old data if the file is later grown.
3336 int ext4_block_truncate_page(handle_t
*handle
,
3337 struct address_space
*mapping
, loff_t from
)
3339 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3342 struct inode
*inode
= mapping
->host
;
3344 blocksize
= inode
->i_sb
->s_blocksize
;
3345 length
= blocksize
- (offset
& (blocksize
- 1));
3347 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3351 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3352 * starting from file offset 'from'. The range to be zero'd must
3353 * be contained with in one block. If the specified range exceeds
3354 * the end of the block it will be shortened to end of the block
3355 * that cooresponds to 'from'
3357 int ext4_block_zero_page_range(handle_t
*handle
,
3358 struct address_space
*mapping
, loff_t from
, loff_t length
)
3360 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3361 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3362 unsigned blocksize
, max
, pos
;
3364 struct inode
*inode
= mapping
->host
;
3365 struct buffer_head
*bh
;
3369 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3370 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3374 blocksize
= inode
->i_sb
->s_blocksize
;
3375 max
= blocksize
- (offset
& (blocksize
- 1));
3378 * correct length if it does not fall between
3379 * 'from' and the end of the block
3381 if (length
> max
|| length
< 0)
3384 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3386 if (!page_has_buffers(page
))
3387 create_empty_buffers(page
, blocksize
, 0);
3389 /* Find the buffer that contains "offset" */
3390 bh
= page_buffers(page
);
3392 while (offset
>= pos
) {
3393 bh
= bh
->b_this_page
;
3399 if (buffer_freed(bh
)) {
3400 BUFFER_TRACE(bh
, "freed: skip");
3404 if (!buffer_mapped(bh
)) {
3405 BUFFER_TRACE(bh
, "unmapped");
3406 ext4_get_block(inode
, iblock
, bh
, 0);
3407 /* unmapped? It's a hole - nothing to do */
3408 if (!buffer_mapped(bh
)) {
3409 BUFFER_TRACE(bh
, "still unmapped");
3414 /* Ok, it's mapped. Make sure it's up-to-date */
3415 if (PageUptodate(page
))
3416 set_buffer_uptodate(bh
);
3418 if (!buffer_uptodate(bh
)) {
3420 ll_rw_block(READ
, 1, &bh
);
3422 /* Uhhuh. Read error. Complain and punt. */
3423 if (!buffer_uptodate(bh
))
3427 if (ext4_should_journal_data(inode
)) {
3428 BUFFER_TRACE(bh
, "get write access");
3429 err
= ext4_journal_get_write_access(handle
, bh
);
3434 zero_user(page
, offset
, length
);
3436 BUFFER_TRACE(bh
, "zeroed end of block");
3439 if (ext4_should_journal_data(inode
)) {
3440 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3442 mark_buffer_dirty(bh
);
3446 page_cache_release(page
);
3450 int ext4_can_truncate(struct inode
*inode
)
3452 if (S_ISREG(inode
->i_mode
))
3454 if (S_ISDIR(inode
->i_mode
))
3456 if (S_ISLNK(inode
->i_mode
))
3457 return !ext4_inode_is_fast_symlink(inode
);
3462 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3463 * associated with the given offset and length
3465 * @inode: File inode
3466 * @offset: The offset where the hole will begin
3467 * @len: The length of the hole
3469 * Returns: 0 on sucess or negative on failure
3472 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3474 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3475 if (!S_ISREG(inode
->i_mode
))
3478 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3479 /* TODO: Add support for non extent hole punching */
3483 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) {
3484 /* TODO: Add support for bigalloc file systems */
3488 return ext4_ext_punch_hole(file
, offset
, length
);
3494 * We block out ext4_get_block() block instantiations across the entire
3495 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3496 * simultaneously on behalf of the same inode.
3498 * As we work through the truncate and commmit bits of it to the journal there
3499 * is one core, guiding principle: the file's tree must always be consistent on
3500 * disk. We must be able to restart the truncate after a crash.
3502 * The file's tree may be transiently inconsistent in memory (although it
3503 * probably isn't), but whenever we close off and commit a journal transaction,
3504 * the contents of (the filesystem + the journal) must be consistent and
3505 * restartable. It's pretty simple, really: bottom up, right to left (although
3506 * left-to-right works OK too).
3508 * Note that at recovery time, journal replay occurs *before* the restart of
3509 * truncate against the orphan inode list.
3511 * The committed inode has the new, desired i_size (which is the same as
3512 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3513 * that this inode's truncate did not complete and it will again call
3514 * ext4_truncate() to have another go. So there will be instantiated blocks
3515 * to the right of the truncation point in a crashed ext4 filesystem. But
3516 * that's fine - as long as they are linked from the inode, the post-crash
3517 * ext4_truncate() run will find them and release them.
3519 void ext4_truncate(struct inode
*inode
)
3521 trace_ext4_truncate_enter(inode
);
3523 if (!ext4_can_truncate(inode
))
3526 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3528 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3529 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3531 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3532 ext4_ext_truncate(inode
);
3534 ext4_ind_truncate(inode
);
3536 trace_ext4_truncate_exit(inode
);
3540 * ext4_get_inode_loc returns with an extra refcount against the inode's
3541 * underlying buffer_head on success. If 'in_mem' is true, we have all
3542 * data in memory that is needed to recreate the on-disk version of this
3545 static int __ext4_get_inode_loc(struct inode
*inode
,
3546 struct ext4_iloc
*iloc
, int in_mem
)
3548 struct ext4_group_desc
*gdp
;
3549 struct buffer_head
*bh
;
3550 struct super_block
*sb
= inode
->i_sb
;
3552 int inodes_per_block
, inode_offset
;
3555 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3558 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3559 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3564 * Figure out the offset within the block group inode table
3566 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3567 inode_offset
= ((inode
->i_ino
- 1) %
3568 EXT4_INODES_PER_GROUP(sb
));
3569 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3570 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3572 bh
= sb_getblk(sb
, block
);
3574 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3575 "unable to read itable block");
3578 if (!buffer_uptodate(bh
)) {
3582 * If the buffer has the write error flag, we have failed
3583 * to write out another inode in the same block. In this
3584 * case, we don't have to read the block because we may
3585 * read the old inode data successfully.
3587 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3588 set_buffer_uptodate(bh
);
3590 if (buffer_uptodate(bh
)) {
3591 /* someone brought it uptodate while we waited */
3597 * If we have all information of the inode in memory and this
3598 * is the only valid inode in the block, we need not read the
3602 struct buffer_head
*bitmap_bh
;
3605 start
= inode_offset
& ~(inodes_per_block
- 1);
3607 /* Is the inode bitmap in cache? */
3608 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3613 * If the inode bitmap isn't in cache then the
3614 * optimisation may end up performing two reads instead
3615 * of one, so skip it.
3617 if (!buffer_uptodate(bitmap_bh
)) {
3621 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3622 if (i
== inode_offset
)
3624 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3628 if (i
== start
+ inodes_per_block
) {
3629 /* all other inodes are free, so skip I/O */
3630 memset(bh
->b_data
, 0, bh
->b_size
);
3631 set_buffer_uptodate(bh
);
3639 * If we need to do any I/O, try to pre-readahead extra
3640 * blocks from the inode table.
3642 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3643 ext4_fsblk_t b
, end
, table
;
3646 table
= ext4_inode_table(sb
, gdp
);
3647 /* s_inode_readahead_blks is always a power of 2 */
3648 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3651 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3652 num
= EXT4_INODES_PER_GROUP(sb
);
3653 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3654 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
3655 num
-= ext4_itable_unused_count(sb
, gdp
);
3656 table
+= num
/ inodes_per_block
;
3660 sb_breadahead(sb
, b
++);
3664 * There are other valid inodes in the buffer, this inode
3665 * has in-inode xattrs, or we don't have this inode in memory.
3666 * Read the block from disk.
3668 trace_ext4_load_inode(inode
);
3670 bh
->b_end_io
= end_buffer_read_sync
;
3671 submit_bh(READ_META
, bh
);
3673 if (!buffer_uptodate(bh
)) {
3674 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3675 "unable to read itable block");
3685 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3687 /* We have all inode data except xattrs in memory here. */
3688 return __ext4_get_inode_loc(inode
, iloc
,
3689 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3692 void ext4_set_inode_flags(struct inode
*inode
)
3694 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3696 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3697 if (flags
& EXT4_SYNC_FL
)
3698 inode
->i_flags
|= S_SYNC
;
3699 if (flags
& EXT4_APPEND_FL
)
3700 inode
->i_flags
|= S_APPEND
;
3701 if (flags
& EXT4_IMMUTABLE_FL
)
3702 inode
->i_flags
|= S_IMMUTABLE
;
3703 if (flags
& EXT4_NOATIME_FL
)
3704 inode
->i_flags
|= S_NOATIME
;
3705 if (flags
& EXT4_DIRSYNC_FL
)
3706 inode
->i_flags
|= S_DIRSYNC
;
3709 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3710 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3712 unsigned int vfs_fl
;
3713 unsigned long old_fl
, new_fl
;
3716 vfs_fl
= ei
->vfs_inode
.i_flags
;
3717 old_fl
= ei
->i_flags
;
3718 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3719 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3721 if (vfs_fl
& S_SYNC
)
3722 new_fl
|= EXT4_SYNC_FL
;
3723 if (vfs_fl
& S_APPEND
)
3724 new_fl
|= EXT4_APPEND_FL
;
3725 if (vfs_fl
& S_IMMUTABLE
)
3726 new_fl
|= EXT4_IMMUTABLE_FL
;
3727 if (vfs_fl
& S_NOATIME
)
3728 new_fl
|= EXT4_NOATIME_FL
;
3729 if (vfs_fl
& S_DIRSYNC
)
3730 new_fl
|= EXT4_DIRSYNC_FL
;
3731 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3734 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3735 struct ext4_inode_info
*ei
)
3738 struct inode
*inode
= &(ei
->vfs_inode
);
3739 struct super_block
*sb
= inode
->i_sb
;
3741 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3742 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3743 /* we are using combined 48 bit field */
3744 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3745 le32_to_cpu(raw_inode
->i_blocks_lo
);
3746 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3747 /* i_blocks represent file system block size */
3748 return i_blocks
<< (inode
->i_blkbits
- 9);
3753 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3757 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3759 struct ext4_iloc iloc
;
3760 struct ext4_inode
*raw_inode
;
3761 struct ext4_inode_info
*ei
;
3762 struct inode
*inode
;
3763 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3767 inode
= iget_locked(sb
, ino
);
3769 return ERR_PTR(-ENOMEM
);
3770 if (!(inode
->i_state
& I_NEW
))
3776 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3779 raw_inode
= ext4_raw_inode(&iloc
);
3780 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3781 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3782 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3783 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3784 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3785 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3787 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
3789 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3790 ei
->i_dir_start_lookup
= 0;
3791 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3792 /* We now have enough fields to check if the inode was active or not.
3793 * This is needed because nfsd might try to access dead inodes
3794 * the test is that same one that e2fsck uses
3795 * NeilBrown 1999oct15
3797 if (inode
->i_nlink
== 0) {
3798 if (inode
->i_mode
== 0 ||
3799 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3800 /* this inode is deleted */
3804 /* The only unlinked inodes we let through here have
3805 * valid i_mode and are being read by the orphan
3806 * recovery code: that's fine, we're about to complete
3807 * the process of deleting those. */
3809 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3810 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3811 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3812 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3814 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3815 inode
->i_size
= ext4_isize(raw_inode
);
3816 ei
->i_disksize
= inode
->i_size
;
3818 ei
->i_reserved_quota
= 0;
3820 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3821 ei
->i_block_group
= iloc
.block_group
;
3822 ei
->i_last_alloc_group
= ~0;
3824 * NOTE! The in-memory inode i_data array is in little-endian order
3825 * even on big-endian machines: we do NOT byteswap the block numbers!
3827 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3828 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3829 INIT_LIST_HEAD(&ei
->i_orphan
);
3832 * Set transaction id's of transactions that have to be committed
3833 * to finish f[data]sync. We set them to currently running transaction
3834 * as we cannot be sure that the inode or some of its metadata isn't
3835 * part of the transaction - the inode could have been reclaimed and
3836 * now it is reread from disk.
3839 transaction_t
*transaction
;
3842 read_lock(&journal
->j_state_lock
);
3843 if (journal
->j_running_transaction
)
3844 transaction
= journal
->j_running_transaction
;
3846 transaction
= journal
->j_committing_transaction
;
3848 tid
= transaction
->t_tid
;
3850 tid
= journal
->j_commit_sequence
;
3851 read_unlock(&journal
->j_state_lock
);
3852 ei
->i_sync_tid
= tid
;
3853 ei
->i_datasync_tid
= tid
;
3856 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3857 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3858 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3859 EXT4_INODE_SIZE(inode
->i_sb
)) {
3863 if (ei
->i_extra_isize
== 0) {
3864 /* The extra space is currently unused. Use it. */
3865 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3866 EXT4_GOOD_OLD_INODE_SIZE
;
3868 __le32
*magic
= (void *)raw_inode
+
3869 EXT4_GOOD_OLD_INODE_SIZE
+
3871 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
3872 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3875 ei
->i_extra_isize
= 0;
3877 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3878 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3879 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3880 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3882 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3883 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3884 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3886 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3890 if (ei
->i_file_acl
&&
3891 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
3892 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
3896 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3897 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3898 (S_ISLNK(inode
->i_mode
) &&
3899 !ext4_inode_is_fast_symlink(inode
)))
3900 /* Validate extent which is part of inode */
3901 ret
= ext4_ext_check_inode(inode
);
3902 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3903 (S_ISLNK(inode
->i_mode
) &&
3904 !ext4_inode_is_fast_symlink(inode
))) {
3905 /* Validate block references which are part of inode */
3906 ret
= ext4_ind_check_inode(inode
);
3911 if (S_ISREG(inode
->i_mode
)) {
3912 inode
->i_op
= &ext4_file_inode_operations
;
3913 inode
->i_fop
= &ext4_file_operations
;
3914 ext4_set_aops(inode
);
3915 } else if (S_ISDIR(inode
->i_mode
)) {
3916 inode
->i_op
= &ext4_dir_inode_operations
;
3917 inode
->i_fop
= &ext4_dir_operations
;
3918 } else if (S_ISLNK(inode
->i_mode
)) {
3919 if (ext4_inode_is_fast_symlink(inode
)) {
3920 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3921 nd_terminate_link(ei
->i_data
, inode
->i_size
,
3922 sizeof(ei
->i_data
) - 1);
3924 inode
->i_op
= &ext4_symlink_inode_operations
;
3925 ext4_set_aops(inode
);
3927 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
3928 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
3929 inode
->i_op
= &ext4_special_inode_operations
;
3930 if (raw_inode
->i_block
[0])
3931 init_special_inode(inode
, inode
->i_mode
,
3932 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3934 init_special_inode(inode
, inode
->i_mode
,
3935 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3938 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
3942 ext4_set_inode_flags(inode
);
3943 unlock_new_inode(inode
);
3949 return ERR_PTR(ret
);
3952 static int ext4_inode_blocks_set(handle_t
*handle
,
3953 struct ext4_inode
*raw_inode
,
3954 struct ext4_inode_info
*ei
)
3956 struct inode
*inode
= &(ei
->vfs_inode
);
3957 u64 i_blocks
= inode
->i_blocks
;
3958 struct super_block
*sb
= inode
->i_sb
;
3960 if (i_blocks
<= ~0U) {
3962 * i_blocks can be represnted in a 32 bit variable
3963 * as multiple of 512 bytes
3965 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3966 raw_inode
->i_blocks_high
= 0;
3967 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3970 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
3973 if (i_blocks
<= 0xffffffffffffULL
) {
3975 * i_blocks can be represented in a 48 bit variable
3976 * as multiple of 512 bytes
3978 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3979 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3980 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3982 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3983 /* i_block is stored in file system block size */
3984 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
3985 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3986 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3992 * Post the struct inode info into an on-disk inode location in the
3993 * buffer-cache. This gobbles the caller's reference to the
3994 * buffer_head in the inode location struct.
3996 * The caller must have write access to iloc->bh.
3998 static int ext4_do_update_inode(handle_t
*handle
,
3999 struct inode
*inode
,
4000 struct ext4_iloc
*iloc
)
4002 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4003 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4004 struct buffer_head
*bh
= iloc
->bh
;
4005 int err
= 0, rc
, block
;
4007 /* For fields not not tracking in the in-memory inode,
4008 * initialise them to zero for new inodes. */
4009 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4010 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4012 ext4_get_inode_flags(ei
);
4013 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4014 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4015 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4016 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4018 * Fix up interoperability with old kernels. Otherwise, old inodes get
4019 * re-used with the upper 16 bits of the uid/gid intact
4022 raw_inode
->i_uid_high
=
4023 cpu_to_le16(high_16_bits(inode
->i_uid
));
4024 raw_inode
->i_gid_high
=
4025 cpu_to_le16(high_16_bits(inode
->i_gid
));
4027 raw_inode
->i_uid_high
= 0;
4028 raw_inode
->i_gid_high
= 0;
4031 raw_inode
->i_uid_low
=
4032 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4033 raw_inode
->i_gid_low
=
4034 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4035 raw_inode
->i_uid_high
= 0;
4036 raw_inode
->i_gid_high
= 0;
4038 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4040 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4041 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4042 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4043 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4045 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4047 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4048 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4049 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4050 cpu_to_le32(EXT4_OS_HURD
))
4051 raw_inode
->i_file_acl_high
=
4052 cpu_to_le16(ei
->i_file_acl
>> 32);
4053 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4054 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4055 if (ei
->i_disksize
> 0x7fffffffULL
) {
4056 struct super_block
*sb
= inode
->i_sb
;
4057 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4058 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4059 EXT4_SB(sb
)->s_es
->s_rev_level
==
4060 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4061 /* If this is the first large file
4062 * created, add a flag to the superblock.
4064 err
= ext4_journal_get_write_access(handle
,
4065 EXT4_SB(sb
)->s_sbh
);
4068 ext4_update_dynamic_rev(sb
);
4069 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4070 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4072 ext4_handle_sync(handle
);
4073 err
= ext4_handle_dirty_metadata(handle
, NULL
,
4074 EXT4_SB(sb
)->s_sbh
);
4077 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4078 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4079 if (old_valid_dev(inode
->i_rdev
)) {
4080 raw_inode
->i_block
[0] =
4081 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4082 raw_inode
->i_block
[1] = 0;
4084 raw_inode
->i_block
[0] = 0;
4085 raw_inode
->i_block
[1] =
4086 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4087 raw_inode
->i_block
[2] = 0;
4090 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4091 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4093 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4094 if (ei
->i_extra_isize
) {
4095 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4096 raw_inode
->i_version_hi
=
4097 cpu_to_le32(inode
->i_version
>> 32);
4098 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4101 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4102 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4105 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4107 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4110 ext4_std_error(inode
->i_sb
, err
);
4115 * ext4_write_inode()
4117 * We are called from a few places:
4119 * - Within generic_file_write() for O_SYNC files.
4120 * Here, there will be no transaction running. We wait for any running
4121 * trasnaction to commit.
4123 * - Within sys_sync(), kupdate and such.
4124 * We wait on commit, if tol to.
4126 * - Within prune_icache() (PF_MEMALLOC == true)
4127 * Here we simply return. We can't afford to block kswapd on the
4130 * In all cases it is actually safe for us to return without doing anything,
4131 * because the inode has been copied into a raw inode buffer in
4132 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4135 * Note that we are absolutely dependent upon all inode dirtiers doing the
4136 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4137 * which we are interested.
4139 * It would be a bug for them to not do this. The code:
4141 * mark_inode_dirty(inode)
4143 * inode->i_size = expr;
4145 * is in error because a kswapd-driven write_inode() could occur while
4146 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4147 * will no longer be on the superblock's dirty inode list.
4149 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4153 if (current
->flags
& PF_MEMALLOC
)
4156 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4157 if (ext4_journal_current_handle()) {
4158 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4163 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4166 err
= ext4_force_commit(inode
->i_sb
);
4168 struct ext4_iloc iloc
;
4170 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4173 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4174 sync_dirty_buffer(iloc
.bh
);
4175 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4176 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4177 "IO error syncing inode");
4188 * Called from notify_change.
4190 * We want to trap VFS attempts to truncate the file as soon as
4191 * possible. In particular, we want to make sure that when the VFS
4192 * shrinks i_size, we put the inode on the orphan list and modify
4193 * i_disksize immediately, so that during the subsequent flushing of
4194 * dirty pages and freeing of disk blocks, we can guarantee that any
4195 * commit will leave the blocks being flushed in an unused state on
4196 * disk. (On recovery, the inode will get truncated and the blocks will
4197 * be freed, so we have a strong guarantee that no future commit will
4198 * leave these blocks visible to the user.)
4200 * Another thing we have to assure is that if we are in ordered mode
4201 * and inode is still attached to the committing transaction, we must
4202 * we start writeout of all the dirty pages which are being truncated.
4203 * This way we are sure that all the data written in the previous
4204 * transaction are already on disk (truncate waits for pages under
4207 * Called with inode->i_mutex down.
4209 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4211 struct inode
*inode
= dentry
->d_inode
;
4214 const unsigned int ia_valid
= attr
->ia_valid
;
4216 error
= inode_change_ok(inode
, attr
);
4220 if (is_quota_modification(inode
, attr
))
4221 dquot_initialize(inode
);
4222 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4223 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4226 /* (user+group)*(old+new) structure, inode write (sb,
4227 * inode block, ? - but truncate inode update has it) */
4228 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
4229 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
4230 if (IS_ERR(handle
)) {
4231 error
= PTR_ERR(handle
);
4234 error
= dquot_transfer(inode
, attr
);
4236 ext4_journal_stop(handle
);
4239 /* Update corresponding info in inode so that everything is in
4240 * one transaction */
4241 if (attr
->ia_valid
& ATTR_UID
)
4242 inode
->i_uid
= attr
->ia_uid
;
4243 if (attr
->ia_valid
& ATTR_GID
)
4244 inode
->i_gid
= attr
->ia_gid
;
4245 error
= ext4_mark_inode_dirty(handle
, inode
);
4246 ext4_journal_stop(handle
);
4249 if (attr
->ia_valid
& ATTR_SIZE
) {
4250 inode_dio_wait(inode
);
4252 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4253 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4255 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4260 if (S_ISREG(inode
->i_mode
) &&
4261 attr
->ia_valid
& ATTR_SIZE
&&
4262 (attr
->ia_size
< inode
->i_size
)) {
4265 handle
= ext4_journal_start(inode
, 3);
4266 if (IS_ERR(handle
)) {
4267 error
= PTR_ERR(handle
);
4270 if (ext4_handle_valid(handle
)) {
4271 error
= ext4_orphan_add(handle
, inode
);
4274 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4275 rc
= ext4_mark_inode_dirty(handle
, inode
);
4278 ext4_journal_stop(handle
);
4280 if (ext4_should_order_data(inode
)) {
4281 error
= ext4_begin_ordered_truncate(inode
,
4284 /* Do as much error cleanup as possible */
4285 handle
= ext4_journal_start(inode
, 3);
4286 if (IS_ERR(handle
)) {
4287 ext4_orphan_del(NULL
, inode
);
4290 ext4_orphan_del(handle
, inode
);
4292 ext4_journal_stop(handle
);
4298 if (attr
->ia_valid
& ATTR_SIZE
) {
4299 if (attr
->ia_size
!= i_size_read(inode
)) {
4300 truncate_setsize(inode
, attr
->ia_size
);
4301 ext4_truncate(inode
);
4302 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
4303 ext4_truncate(inode
);
4307 setattr_copy(inode
, attr
);
4308 mark_inode_dirty(inode
);
4312 * If the call to ext4_truncate failed to get a transaction handle at
4313 * all, we need to clean up the in-core orphan list manually.
4315 if (orphan
&& inode
->i_nlink
)
4316 ext4_orphan_del(NULL
, inode
);
4318 if (!rc
&& (ia_valid
& ATTR_MODE
))
4319 rc
= ext4_acl_chmod(inode
);
4322 ext4_std_error(inode
->i_sb
, error
);
4328 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4331 struct inode
*inode
;
4332 unsigned long delalloc_blocks
;
4334 inode
= dentry
->d_inode
;
4335 generic_fillattr(inode
, stat
);
4338 * We can't update i_blocks if the block allocation is delayed
4339 * otherwise in the case of system crash before the real block
4340 * allocation is done, we will have i_blocks inconsistent with
4341 * on-disk file blocks.
4342 * We always keep i_blocks updated together with real
4343 * allocation. But to not confuse with user, stat
4344 * will return the blocks that include the delayed allocation
4345 * blocks for this file.
4347 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4349 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4353 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4355 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4356 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4357 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4361 * Account for index blocks, block groups bitmaps and block group
4362 * descriptor blocks if modify datablocks and index blocks
4363 * worse case, the indexs blocks spread over different block groups
4365 * If datablocks are discontiguous, they are possible to spread over
4366 * different block groups too. If they are contiuguous, with flexbg,
4367 * they could still across block group boundary.
4369 * Also account for superblock, inode, quota and xattr blocks
4371 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4373 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4379 * How many index blocks need to touch to modify nrblocks?
4380 * The "Chunk" flag indicating whether the nrblocks is
4381 * physically contiguous on disk
4383 * For Direct IO and fallocate, they calls get_block to allocate
4384 * one single extent at a time, so they could set the "Chunk" flag
4386 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4391 * Now let's see how many group bitmaps and group descriptors need
4401 if (groups
> ngroups
)
4403 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4404 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4406 /* bitmaps and block group descriptor blocks */
4407 ret
+= groups
+ gdpblocks
;
4409 /* Blocks for super block, inode, quota and xattr blocks */
4410 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4416 * Calculate the total number of credits to reserve to fit
4417 * the modification of a single pages into a single transaction,
4418 * which may include multiple chunks of block allocations.
4420 * This could be called via ext4_write_begin()
4422 * We need to consider the worse case, when
4423 * one new block per extent.
4425 int ext4_writepage_trans_blocks(struct inode
*inode
)
4427 int bpp
= ext4_journal_blocks_per_page(inode
);
4430 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4432 /* Account for data blocks for journalled mode */
4433 if (ext4_should_journal_data(inode
))
4439 * Calculate the journal credits for a chunk of data modification.
4441 * This is called from DIO, fallocate or whoever calling
4442 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4444 * journal buffers for data blocks are not included here, as DIO
4445 * and fallocate do no need to journal data buffers.
4447 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4449 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4453 * The caller must have previously called ext4_reserve_inode_write().
4454 * Give this, we know that the caller already has write access to iloc->bh.
4456 int ext4_mark_iloc_dirty(handle_t
*handle
,
4457 struct inode
*inode
, struct ext4_iloc
*iloc
)
4461 if (test_opt(inode
->i_sb
, I_VERSION
))
4462 inode_inc_iversion(inode
);
4464 /* the do_update_inode consumes one bh->b_count */
4467 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4468 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4474 * On success, We end up with an outstanding reference count against
4475 * iloc->bh. This _must_ be cleaned up later.
4479 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4480 struct ext4_iloc
*iloc
)
4484 err
= ext4_get_inode_loc(inode
, iloc
);
4486 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4487 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4493 ext4_std_error(inode
->i_sb
, err
);
4498 * Expand an inode by new_extra_isize bytes.
4499 * Returns 0 on success or negative error number on failure.
4501 static int ext4_expand_extra_isize(struct inode
*inode
,
4502 unsigned int new_extra_isize
,
4503 struct ext4_iloc iloc
,
4506 struct ext4_inode
*raw_inode
;
4507 struct ext4_xattr_ibody_header
*header
;
4509 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4512 raw_inode
= ext4_raw_inode(&iloc
);
4514 header
= IHDR(inode
, raw_inode
);
4516 /* No extended attributes present */
4517 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4518 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4519 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4521 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4525 /* try to expand with EAs present */
4526 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4531 * What we do here is to mark the in-core inode as clean with respect to inode
4532 * dirtiness (it may still be data-dirty).
4533 * This means that the in-core inode may be reaped by prune_icache
4534 * without having to perform any I/O. This is a very good thing,
4535 * because *any* task may call prune_icache - even ones which
4536 * have a transaction open against a different journal.
4538 * Is this cheating? Not really. Sure, we haven't written the
4539 * inode out, but prune_icache isn't a user-visible syncing function.
4540 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4541 * we start and wait on commits.
4543 * Is this efficient/effective? Well, we're being nice to the system
4544 * by cleaning up our inodes proactively so they can be reaped
4545 * without I/O. But we are potentially leaving up to five seconds'
4546 * worth of inodes floating about which prune_icache wants us to
4547 * write out. One way to fix that would be to get prune_icache()
4548 * to do a write_super() to free up some memory. It has the desired
4551 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4553 struct ext4_iloc iloc
;
4554 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4555 static unsigned int mnt_count
;
4559 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4560 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4561 if (ext4_handle_valid(handle
) &&
4562 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4563 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4565 * We need extra buffer credits since we may write into EA block
4566 * with this same handle. If journal_extend fails, then it will
4567 * only result in a minor loss of functionality for that inode.
4568 * If this is felt to be critical, then e2fsck should be run to
4569 * force a large enough s_min_extra_isize.
4571 if ((jbd2_journal_extend(handle
,
4572 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4573 ret
= ext4_expand_extra_isize(inode
,
4574 sbi
->s_want_extra_isize
,
4577 ext4_set_inode_state(inode
,
4578 EXT4_STATE_NO_EXPAND
);
4580 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4581 ext4_warning(inode
->i_sb
,
4582 "Unable to expand inode %lu. Delete"
4583 " some EAs or run e2fsck.",
4586 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4592 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4597 * ext4_dirty_inode() is called from __mark_inode_dirty()
4599 * We're really interested in the case where a file is being extended.
4600 * i_size has been changed by generic_commit_write() and we thus need
4601 * to include the updated inode in the current transaction.
4603 * Also, dquot_alloc_block() will always dirty the inode when blocks
4604 * are allocated to the file.
4606 * If the inode is marked synchronous, we don't honour that here - doing
4607 * so would cause a commit on atime updates, which we don't bother doing.
4608 * We handle synchronous inodes at the highest possible level.
4610 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4614 handle
= ext4_journal_start(inode
, 2);
4618 ext4_mark_inode_dirty(handle
, inode
);
4620 ext4_journal_stop(handle
);
4627 * Bind an inode's backing buffer_head into this transaction, to prevent
4628 * it from being flushed to disk early. Unlike
4629 * ext4_reserve_inode_write, this leaves behind no bh reference and
4630 * returns no iloc structure, so the caller needs to repeat the iloc
4631 * lookup to mark the inode dirty later.
4633 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4635 struct ext4_iloc iloc
;
4639 err
= ext4_get_inode_loc(inode
, &iloc
);
4641 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4642 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4644 err
= ext4_handle_dirty_metadata(handle
,
4650 ext4_std_error(inode
->i_sb
, err
);
4655 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4662 * We have to be very careful here: changing a data block's
4663 * journaling status dynamically is dangerous. If we write a
4664 * data block to the journal, change the status and then delete
4665 * that block, we risk forgetting to revoke the old log record
4666 * from the journal and so a subsequent replay can corrupt data.
4667 * So, first we make sure that the journal is empty and that
4668 * nobody is changing anything.
4671 journal
= EXT4_JOURNAL(inode
);
4674 if (is_journal_aborted(journal
))
4677 jbd2_journal_lock_updates(journal
);
4678 jbd2_journal_flush(journal
);
4681 * OK, there are no updates running now, and all cached data is
4682 * synced to disk. We are now in a completely consistent state
4683 * which doesn't have anything in the journal, and we know that
4684 * no filesystem updates are running, so it is safe to modify
4685 * the inode's in-core data-journaling state flag now.
4689 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4691 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4692 ext4_set_aops(inode
);
4694 jbd2_journal_unlock_updates(journal
);
4696 /* Finally we can mark the inode as dirty. */
4698 handle
= ext4_journal_start(inode
, 1);
4700 return PTR_ERR(handle
);
4702 err
= ext4_mark_inode_dirty(handle
, inode
);
4703 ext4_handle_sync(handle
);
4704 ext4_journal_stop(handle
);
4705 ext4_std_error(inode
->i_sb
, err
);
4710 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4712 return !buffer_mapped(bh
);
4715 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4717 struct page
*page
= vmf
->page
;
4721 struct file
*file
= vma
->vm_file
;
4722 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4723 struct address_space
*mapping
= inode
->i_mapping
;
4725 get_block_t
*get_block
;
4729 * This check is racy but catches the common case. We rely on
4730 * __block_page_mkwrite() to do a reliable check.
4732 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
4733 /* Delalloc case is easy... */
4734 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4735 !ext4_should_journal_data(inode
) &&
4736 !ext4_nonda_switch(inode
->i_sb
)) {
4738 ret
= __block_page_mkwrite(vma
, vmf
,
4739 ext4_da_get_block_prep
);
4740 } while (ret
== -ENOSPC
&&
4741 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4746 size
= i_size_read(inode
);
4747 /* Page got truncated from under us? */
4748 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4750 ret
= VM_FAULT_NOPAGE
;
4754 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4755 len
= size
& ~PAGE_CACHE_MASK
;
4757 len
= PAGE_CACHE_SIZE
;
4759 * Return if we have all the buffers mapped. This avoids the need to do
4760 * journal_start/journal_stop which can block and take a long time
4762 if (page_has_buffers(page
)) {
4763 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4764 ext4_bh_unmapped
)) {
4765 /* Wait so that we don't change page under IO */
4766 wait_on_page_writeback(page
);
4767 ret
= VM_FAULT_LOCKED
;
4772 /* OK, we need to fill the hole... */
4773 if (ext4_should_dioread_nolock(inode
))
4774 get_block
= ext4_get_block_write
;
4776 get_block
= ext4_get_block
;
4778 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
4779 if (IS_ERR(handle
)) {
4780 ret
= VM_FAULT_SIGBUS
;
4783 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
4784 if (!ret
&& ext4_should_journal_data(inode
)) {
4785 if (walk_page_buffers(handle
, page_buffers(page
), 0,
4786 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
4788 ret
= VM_FAULT_SIGBUS
;
4791 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
4793 ext4_journal_stop(handle
);
4794 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
4797 ret
= block_page_mkwrite_return(ret
);