4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/aio.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
25 #include <trace/events/f2fs.h>
27 static void f2fs_read_end_io(struct bio
*bio
, int err
)
29 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
30 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
33 struct page
*page
= bvec
->bv_page
;
35 if (--bvec
>= bio
->bi_io_vec
)
36 prefetchw(&bvec
->bv_page
->flags
);
38 if (unlikely(!uptodate
)) {
39 ClearPageUptodate(page
);
42 SetPageUptodate(page
);
45 } while (bvec
>= bio
->bi_io_vec
);
50 static void f2fs_write_end_io(struct bio
*bio
, int err
)
52 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
53 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
54 struct f2fs_sb_info
*sbi
= F2FS_SB(bvec
->bv_page
->mapping
->host
->i_sb
);
57 struct page
*page
= bvec
->bv_page
;
59 if (--bvec
>= bio
->bi_io_vec
)
60 prefetchw(&bvec
->bv_page
->flags
);
62 if (unlikely(!uptodate
)) {
64 set_bit(AS_EIO
, &page
->mapping
->flags
);
65 set_ckpt_flags(sbi
->ckpt
, CP_ERROR_FLAG
);
66 sbi
->sb
->s_flags
|= MS_RDONLY
;
68 end_page_writeback(page
);
69 dec_page_count(sbi
, F2FS_WRITEBACK
);
70 } while (bvec
>= bio
->bi_io_vec
);
73 complete(bio
->bi_private
);
75 if (!get_pages(sbi
, F2FS_WRITEBACK
) &&
76 !list_empty(&sbi
->cp_wait
.task_list
))
77 wake_up(&sbi
->cp_wait
);
83 * Low-level block read/write IO operations.
85 static struct bio
*__bio_alloc(struct f2fs_sb_info
*sbi
, block_t blk_addr
,
86 int npages
, bool is_read
)
90 /* No failure on bio allocation */
91 bio
= bio_alloc(GFP_NOIO
, npages
);
93 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
94 bio
->bi_sector
= SECTOR_FROM_BLOCK(sbi
, blk_addr
);
95 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
100 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
102 struct f2fs_io_info
*fio
= &io
->fio
;
110 if (is_read_io(rw
)) {
111 trace_f2fs_submit_read_bio(io
->sbi
->sb
, rw
,
113 submit_bio(rw
, io
->bio
);
115 trace_f2fs_submit_write_bio(io
->sbi
->sb
, rw
,
118 * META_FLUSH is only from the checkpoint procedure, and we
119 * should wait this metadata bio for FS consistency.
121 if (fio
->type
== META_FLUSH
) {
122 DECLARE_COMPLETION_ONSTACK(wait
);
123 io
->bio
->bi_private
= &wait
;
124 submit_bio(rw
, io
->bio
);
125 wait_for_completion(&wait
);
127 submit_bio(rw
, io
->bio
);
134 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
135 enum page_type type
, int rw
)
137 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
138 struct f2fs_bio_info
*io
;
140 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
142 mutex_lock(&io
->io_mutex
);
144 /* change META to META_FLUSH in the checkpoint procedure */
145 if (type
>= META_FLUSH
) {
146 io
->fio
.type
= META_FLUSH
;
147 io
->fio
.rw
= WRITE_FLUSH_FUA
;
149 __submit_merged_bio(io
);
150 mutex_unlock(&io
->io_mutex
);
154 * Fill the locked page with data located in the block address.
155 * Return unlocked page.
157 int f2fs_submit_page_bio(struct f2fs_sb_info
*sbi
, struct page
*page
,
158 block_t blk_addr
, int rw
)
162 trace_f2fs_submit_page_bio(page
, blk_addr
, rw
);
164 /* Allocate a new bio */
165 bio
= __bio_alloc(sbi
, blk_addr
, 1, is_read_io(rw
));
167 if (bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) < PAGE_CACHE_SIZE
) {
169 f2fs_put_page(page
, 1);
177 void f2fs_submit_page_mbio(struct f2fs_sb_info
*sbi
, struct page
*page
,
178 block_t blk_addr
, struct f2fs_io_info
*fio
)
180 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
181 struct f2fs_bio_info
*io
;
182 bool is_read
= is_read_io(fio
->rw
);
184 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
186 verify_block_addr(sbi
, blk_addr
);
188 mutex_lock(&io
->io_mutex
);
191 inc_page_count(sbi
, F2FS_WRITEBACK
);
193 if (io
->bio
&& (io
->last_block_in_bio
!= blk_addr
- 1 ||
194 io
->fio
.rw
!= fio
->rw
))
195 __submit_merged_bio(io
);
197 if (io
->bio
== NULL
) {
198 int bio_blocks
= MAX_BIO_BLOCKS(max_hw_blocks(sbi
));
200 io
->bio
= __bio_alloc(sbi
, blk_addr
, bio_blocks
, is_read
);
204 if (bio_add_page(io
->bio
, page
, PAGE_CACHE_SIZE
, 0) <
206 __submit_merged_bio(io
);
210 io
->last_block_in_bio
= blk_addr
;
212 mutex_unlock(&io
->io_mutex
);
213 trace_f2fs_submit_page_mbio(page
, fio
->rw
, fio
->type
, blk_addr
);
217 * Lock ordering for the change of data block address:
220 * update block addresses in the node page
222 static void __set_data_blkaddr(struct dnode_of_data
*dn
, block_t new_addr
)
224 struct f2fs_node
*rn
;
226 struct page
*node_page
= dn
->node_page
;
227 unsigned int ofs_in_node
= dn
->ofs_in_node
;
229 f2fs_wait_on_page_writeback(node_page
, NODE
, false);
231 rn
= F2FS_NODE(node_page
);
233 /* Get physical address of data block */
234 addr_array
= blkaddr_in_node(rn
);
235 addr_array
[ofs_in_node
] = cpu_to_le32(new_addr
);
236 set_page_dirty(node_page
);
239 int reserve_new_block(struct dnode_of_data
*dn
)
241 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
243 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
245 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, 1)))
248 trace_f2fs_reserve_new_block(dn
->inode
, dn
->nid
, dn
->ofs_in_node
);
250 __set_data_blkaddr(dn
, NEW_ADDR
);
251 dn
->data_blkaddr
= NEW_ADDR
;
256 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
258 bool need_put
= dn
->inode_page
? false : true;
261 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
264 if (dn
->data_blkaddr
== NULL_ADDR
)
265 err
= reserve_new_block(dn
);
272 static int check_extent_cache(struct inode
*inode
, pgoff_t pgofs
,
273 struct buffer_head
*bh_result
)
275 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
276 pgoff_t start_fofs
, end_fofs
;
277 block_t start_blkaddr
;
279 if (is_inode_flag_set(fi
, FI_NO_EXTENT
))
282 read_lock(&fi
->ext
.ext_lock
);
283 if (fi
->ext
.len
== 0) {
284 read_unlock(&fi
->ext
.ext_lock
);
288 stat_inc_total_hit(inode
->i_sb
);
290 start_fofs
= fi
->ext
.fofs
;
291 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
292 start_blkaddr
= fi
->ext
.blk_addr
;
294 if (pgofs
>= start_fofs
&& pgofs
<= end_fofs
) {
295 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
298 clear_buffer_new(bh_result
);
299 map_bh(bh_result
, inode
->i_sb
,
300 start_blkaddr
+ pgofs
- start_fofs
);
301 count
= end_fofs
- pgofs
+ 1;
302 if (count
< (UINT_MAX
>> blkbits
))
303 bh_result
->b_size
= (count
<< blkbits
);
305 bh_result
->b_size
= UINT_MAX
;
307 stat_inc_read_hit(inode
->i_sb
);
308 read_unlock(&fi
->ext
.ext_lock
);
311 read_unlock(&fi
->ext
.ext_lock
);
315 void update_extent_cache(block_t blk_addr
, struct dnode_of_data
*dn
)
317 struct f2fs_inode_info
*fi
= F2FS_I(dn
->inode
);
318 pgoff_t fofs
, start_fofs
, end_fofs
;
319 block_t start_blkaddr
, end_blkaddr
;
320 int need_update
= true;
322 f2fs_bug_on(blk_addr
== NEW_ADDR
);
323 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), fi
) +
326 /* Update the page address in the parent node */
327 __set_data_blkaddr(dn
, blk_addr
);
329 if (is_inode_flag_set(fi
, FI_NO_EXTENT
))
332 write_lock(&fi
->ext
.ext_lock
);
334 start_fofs
= fi
->ext
.fofs
;
335 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
336 start_blkaddr
= fi
->ext
.blk_addr
;
337 end_blkaddr
= fi
->ext
.blk_addr
+ fi
->ext
.len
- 1;
339 /* Drop and initialize the matched extent */
340 if (fi
->ext
.len
== 1 && fofs
== start_fofs
)
344 if (fi
->ext
.len
== 0) {
345 if (blk_addr
!= NULL_ADDR
) {
347 fi
->ext
.blk_addr
= blk_addr
;
354 if (fofs
== start_fofs
- 1 && blk_addr
== start_blkaddr
- 1) {
362 if (fofs
== end_fofs
+ 1 && blk_addr
== end_blkaddr
+ 1) {
367 /* Split the existing extent */
368 if (fi
->ext
.len
> 1 &&
369 fofs
>= start_fofs
&& fofs
<= end_fofs
) {
370 if ((end_fofs
- fofs
) < (fi
->ext
.len
>> 1)) {
371 fi
->ext
.len
= fofs
- start_fofs
;
373 fi
->ext
.fofs
= fofs
+ 1;
374 fi
->ext
.blk_addr
= start_blkaddr
+
375 fofs
- start_fofs
+ 1;
376 fi
->ext
.len
-= fofs
- start_fofs
+ 1;
382 /* Finally, if the extent is very fragmented, let's drop the cache. */
383 if (fi
->ext
.len
< F2FS_MIN_EXTENT_LEN
) {
385 set_inode_flag(fi
, FI_NO_EXTENT
);
389 write_unlock(&fi
->ext
.ext_lock
);
395 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
, bool sync
)
397 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
398 struct address_space
*mapping
= inode
->i_mapping
;
399 struct dnode_of_data dn
;
403 page
= find_get_page(mapping
, index
);
404 if (page
&& PageUptodate(page
))
406 f2fs_put_page(page
, 0);
408 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
409 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
414 if (dn
.data_blkaddr
== NULL_ADDR
)
415 return ERR_PTR(-ENOENT
);
417 /* By fallocate(), there is no cached page, but with NEW_ADDR */
418 if (unlikely(dn
.data_blkaddr
== NEW_ADDR
))
419 return ERR_PTR(-EINVAL
);
421 page
= grab_cache_page_write_begin(mapping
, index
, AOP_FLAG_NOFS
);
423 return ERR_PTR(-ENOMEM
);
425 if (PageUptodate(page
)) {
430 err
= f2fs_submit_page_bio(sbi
, page
, dn
.data_blkaddr
,
431 sync
? READ_SYNC
: READA
);
436 wait_on_page_locked(page
);
437 if (unlikely(!PageUptodate(page
))) {
438 f2fs_put_page(page
, 0);
439 return ERR_PTR(-EIO
);
446 * If it tries to access a hole, return an error.
447 * Because, the callers, functions in dir.c and GC, should be able to know
448 * whether this page exists or not.
450 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
)
452 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
453 struct address_space
*mapping
= inode
->i_mapping
;
454 struct dnode_of_data dn
;
459 page
= grab_cache_page_write_begin(mapping
, index
, AOP_FLAG_NOFS
);
461 return ERR_PTR(-ENOMEM
);
463 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
464 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
466 f2fs_put_page(page
, 1);
471 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
472 f2fs_put_page(page
, 1);
473 return ERR_PTR(-ENOENT
);
476 if (PageUptodate(page
))
480 * A new dentry page is allocated but not able to be written, since its
481 * new inode page couldn't be allocated due to -ENOSPC.
482 * In such the case, its blkaddr can be remained as NEW_ADDR.
483 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
485 if (dn
.data_blkaddr
== NEW_ADDR
) {
486 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
487 SetPageUptodate(page
);
491 err
= f2fs_submit_page_bio(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
496 if (unlikely(!PageUptodate(page
))) {
497 f2fs_put_page(page
, 1);
498 return ERR_PTR(-EIO
);
500 if (unlikely(page
->mapping
!= mapping
)) {
501 f2fs_put_page(page
, 1);
508 * Caller ensures that this data page is never allocated.
509 * A new zero-filled data page is allocated in the page cache.
511 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
513 * Note that, npage is set only by make_empty_dir.
515 struct page
*get_new_data_page(struct inode
*inode
,
516 struct page
*npage
, pgoff_t index
, bool new_i_size
)
518 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
519 struct address_space
*mapping
= inode
->i_mapping
;
521 struct dnode_of_data dn
;
524 set_new_dnode(&dn
, inode
, npage
, npage
, 0);
525 err
= f2fs_reserve_block(&dn
, index
);
529 page
= grab_cache_page(mapping
, index
);
531 return ERR_PTR(-ENOMEM
);
533 if (PageUptodate(page
))
536 if (dn
.data_blkaddr
== NEW_ADDR
) {
537 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
538 SetPageUptodate(page
);
540 err
= f2fs_submit_page_bio(sbi
, page
, dn
.data_blkaddr
,
545 if (unlikely(!PageUptodate(page
))) {
546 f2fs_put_page(page
, 1);
547 return ERR_PTR(-EIO
);
549 if (unlikely(page
->mapping
!= mapping
)) {
550 f2fs_put_page(page
, 1);
556 i_size_read(inode
) < ((index
+ 1) << PAGE_CACHE_SHIFT
)) {
557 i_size_write(inode
, ((index
+ 1) << PAGE_CACHE_SHIFT
));
558 /* Only the directory inode sets new_i_size */
559 set_inode_flag(F2FS_I(inode
), FI_UPDATE_DIR
);
560 mark_inode_dirty_sync(inode
);
565 static int __allocate_data_block(struct dnode_of_data
*dn
)
567 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
568 struct f2fs_summary sum
;
573 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
575 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, 1)))
578 __set_data_blkaddr(dn
, NEW_ADDR
);
579 dn
->data_blkaddr
= NEW_ADDR
;
581 get_node_info(sbi
, dn
->nid
, &ni
);
582 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
584 type
= CURSEG_WARM_DATA
;
586 allocate_data_block(sbi
, NULL
, NULL_ADDR
, &new_blkaddr
, &sum
, type
);
588 /* direct IO doesn't use extent cache to maximize the performance */
589 set_inode_flag(F2FS_I(dn
->inode
), FI_NO_EXTENT
);
590 update_extent_cache(new_blkaddr
, dn
);
591 clear_inode_flag(F2FS_I(dn
->inode
), FI_NO_EXTENT
);
593 dn
->data_blkaddr
= new_blkaddr
;
598 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
599 * If original data blocks are allocated, then give them to blockdev.
601 * a. preallocate requested block addresses
602 * b. do not use extent cache for better performance
603 * c. give the block addresses to blockdev
605 static int get_data_block(struct inode
*inode
, sector_t iblock
,
606 struct buffer_head
*bh_result
, int create
)
608 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
609 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
610 unsigned maxblocks
= bh_result
->b_size
>> blkbits
;
611 struct dnode_of_data dn
;
612 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE_RA
;
613 pgoff_t pgofs
, end_offset
;
614 int err
= 0, ofs
= 1;
615 bool allocated
= false;
617 /* Get the page offset from the block offset(iblock) */
618 pgofs
= (pgoff_t
)(iblock
>> (PAGE_CACHE_SHIFT
- blkbits
));
620 if (check_extent_cache(inode
, pgofs
, bh_result
))
626 /* When reading holes, we need its node page */
627 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
628 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
634 if (dn
.data_blkaddr
== NEW_ADDR
)
637 if (dn
.data_blkaddr
!= NULL_ADDR
) {
638 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
640 err
= __allocate_data_block(&dn
);
644 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
649 end_offset
= IS_INODE(dn
.node_page
) ?
650 ADDRS_PER_INODE(F2FS_I(inode
)) : ADDRS_PER_BLOCK
;
651 bh_result
->b_size
= (((size_t)1) << blkbits
);
656 if (dn
.ofs_in_node
>= end_offset
) {
658 sync_inode_page(&dn
);
662 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
663 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
669 if (dn
.data_blkaddr
== NEW_ADDR
)
672 end_offset
= IS_INODE(dn
.node_page
) ?
673 ADDRS_PER_INODE(F2FS_I(inode
)) : ADDRS_PER_BLOCK
;
676 if (maxblocks
> (bh_result
->b_size
>> blkbits
)) {
677 block_t blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
678 if (blkaddr
== NULL_ADDR
&& create
) {
679 err
= __allocate_data_block(&dn
);
683 blkaddr
= dn
.data_blkaddr
;
685 /* Give more consecutive addresses for the read ahead */
686 if (blkaddr
== (bh_result
->b_blocknr
+ ofs
)) {
690 bh_result
->b_size
+= (((size_t)1) << blkbits
);
696 sync_inode_page(&dn
);
703 trace_f2fs_get_data_block(inode
, iblock
, bh_result
, err
);
707 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
709 struct inode
*inode
= page
->mapping
->host
;
712 /* If the file has inline data, try to read it directlly */
713 if (f2fs_has_inline_data(inode
))
714 ret
= f2fs_read_inline_data(inode
, page
);
716 ret
= mpage_readpage(page
, get_data_block
);
721 static int f2fs_read_data_pages(struct file
*file
,
722 struct address_space
*mapping
,
723 struct list_head
*pages
, unsigned nr_pages
)
725 struct inode
*inode
= file
->f_mapping
->host
;
727 /* If the file has inline data, skip readpages */
728 if (f2fs_has_inline_data(inode
))
731 return mpage_readpages(mapping
, pages
, nr_pages
, get_data_block
);
734 int do_write_data_page(struct page
*page
, struct f2fs_io_info
*fio
)
736 struct inode
*inode
= page
->mapping
->host
;
737 block_t old_blkaddr
, new_blkaddr
;
738 struct dnode_of_data dn
;
741 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
742 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
746 old_blkaddr
= dn
.data_blkaddr
;
748 /* This page is already truncated */
749 if (old_blkaddr
== NULL_ADDR
)
752 set_page_writeback(page
);
755 * If current allocation needs SSR,
756 * it had better in-place writes for updated data.
758 if (unlikely(old_blkaddr
!= NEW_ADDR
&&
759 !is_cold_data(page
) &&
760 need_inplace_update(inode
))) {
761 rewrite_data_page(page
, old_blkaddr
, fio
);
763 write_data_page(page
, &dn
, &new_blkaddr
, fio
);
764 update_extent_cache(new_blkaddr
, &dn
);
771 static int f2fs_write_data_page(struct page
*page
,
772 struct writeback_control
*wbc
)
774 struct inode
*inode
= page
->mapping
->host
;
775 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
776 loff_t i_size
= i_size_read(inode
);
777 const pgoff_t end_index
= ((unsigned long long) i_size
)
780 bool need_balance_fs
= false;
782 struct f2fs_io_info fio
= {
784 .rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
,
787 if (page
->index
< end_index
)
791 * If the offset is out-of-range of file size,
792 * this page does not have to be written to disk.
794 offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
795 if ((page
->index
>= end_index
+ 1) || !offset
) {
796 if (S_ISDIR(inode
->i_mode
)) {
797 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
798 inode_dec_dirty_dents(inode
);
803 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
805 if (unlikely(sbi
->por_doing
)) {
806 err
= AOP_WRITEPAGE_ACTIVATE
;
810 /* Dentry blocks are controlled by checkpoint */
811 if (S_ISDIR(inode
->i_mode
)) {
812 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
813 inode_dec_dirty_dents(inode
);
814 err
= do_write_data_page(page
, &fio
);
818 if (f2fs_has_inline_data(inode
) || f2fs_may_inline(inode
)) {
819 err
= f2fs_write_inline_data(inode
, page
, offset
);
823 err
= do_write_data_page(page
, &fio
);
827 need_balance_fs
= true;
834 if (wbc
->for_reclaim
)
835 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
837 clear_cold_data(page
);
841 f2fs_balance_fs(sbi
);
845 wbc
->pages_skipped
++;
846 set_page_dirty(page
);
850 #define MAX_DESIRED_PAGES_WP 4096
852 static int __f2fs_writepage(struct page
*page
, struct writeback_control
*wbc
,
855 struct address_space
*mapping
= data
;
856 int ret
= mapping
->a_ops
->writepage(page
, wbc
);
857 mapping_set_error(mapping
, ret
);
861 static int f2fs_write_data_pages(struct address_space
*mapping
,
862 struct writeback_control
*wbc
)
864 struct inode
*inode
= mapping
->host
;
865 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
868 long excess_nrtw
= 0, desired_nrtw
;
870 /* deal with chardevs and other special file */
871 if (!mapping
->a_ops
->writepage
)
874 if (wbc
->nr_to_write
< MAX_DESIRED_PAGES_WP
) {
875 desired_nrtw
= MAX_DESIRED_PAGES_WP
;
876 excess_nrtw
= desired_nrtw
- wbc
->nr_to_write
;
877 wbc
->nr_to_write
= desired_nrtw
;
880 if (!S_ISDIR(inode
->i_mode
)) {
881 mutex_lock(&sbi
->writepages
);
884 ret
= write_cache_pages(mapping
, wbc
, __f2fs_writepage
, mapping
);
886 mutex_unlock(&sbi
->writepages
);
888 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
890 remove_dirty_dir_inode(inode
);
892 wbc
->nr_to_write
-= excess_nrtw
;
896 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
897 loff_t pos
, unsigned len
, unsigned flags
,
898 struct page
**pagep
, void **fsdata
)
900 struct inode
*inode
= mapping
->host
;
901 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
903 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_CACHE_SHIFT
;
904 struct dnode_of_data dn
;
907 f2fs_balance_fs(sbi
);
909 err
= f2fs_convert_inline_data(inode
, pos
+ len
);
913 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
918 if (f2fs_has_inline_data(inode
) && (pos
+ len
) <= MAX_INLINE_DATA
)
922 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
923 err
= f2fs_reserve_block(&dn
, index
);
927 f2fs_put_page(page
, 1);
931 if ((len
== PAGE_CACHE_SIZE
) || PageUptodate(page
))
934 if ((pos
& PAGE_CACHE_MASK
) >= i_size_read(inode
)) {
935 unsigned start
= pos
& (PAGE_CACHE_SIZE
- 1);
936 unsigned end
= start
+ len
;
938 /* Reading beyond i_size is simple: memset to zero */
939 zero_user_segments(page
, 0, start
, end
, PAGE_CACHE_SIZE
);
943 if (dn
.data_blkaddr
== NEW_ADDR
) {
944 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
946 if (f2fs_has_inline_data(inode
))
947 err
= f2fs_read_inline_data(inode
, page
);
949 err
= f2fs_submit_page_bio(sbi
, page
, dn
.data_blkaddr
,
954 if (unlikely(!PageUptodate(page
))) {
955 f2fs_put_page(page
, 1);
958 if (unlikely(page
->mapping
!= mapping
)) {
959 f2fs_put_page(page
, 1);
964 SetPageUptodate(page
);
965 clear_cold_data(page
);
969 static int f2fs_write_end(struct file
*file
,
970 struct address_space
*mapping
,
971 loff_t pos
, unsigned len
, unsigned copied
,
972 struct page
*page
, void *fsdata
)
974 struct inode
*inode
= page
->mapping
->host
;
976 SetPageUptodate(page
);
977 set_page_dirty(page
);
979 if (pos
+ copied
> i_size_read(inode
)) {
980 i_size_write(inode
, pos
+ copied
);
981 mark_inode_dirty(inode
);
982 update_inode_page(inode
);
985 f2fs_put_page(page
, 1);
989 static int check_direct_IO(struct inode
*inode
, int rw
,
990 const struct iovec
*iov
, loff_t offset
, unsigned long nr_segs
)
992 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
998 if (offset
& blocksize_mask
)
1001 for (i
= 0; i
< nr_segs
; i
++)
1002 if (iov
[i
].iov_len
& blocksize_mask
)
1007 static ssize_t
f2fs_direct_IO(int rw
, struct kiocb
*iocb
,
1008 const struct iovec
*iov
, loff_t offset
, unsigned long nr_segs
)
1010 struct file
*file
= iocb
->ki_filp
;
1011 struct inode
*inode
= file
->f_mapping
->host
;
1013 /* Let buffer I/O handle the inline data case. */
1014 if (f2fs_has_inline_data(inode
))
1017 if (check_direct_IO(inode
, rw
, iov
, offset
, nr_segs
))
1020 return blockdev_direct_IO(rw
, iocb
, inode
, iov
, offset
, nr_segs
,
1024 static void f2fs_invalidate_data_page(struct page
*page
, unsigned int offset
,
1025 unsigned int length
)
1027 struct inode
*inode
= page
->mapping
->host
;
1028 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
1029 if (S_ISDIR(inode
->i_mode
) && PageDirty(page
)) {
1030 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
1031 inode_dec_dirty_dents(inode
);
1033 ClearPagePrivate(page
);
1036 static int f2fs_release_data_page(struct page
*page
, gfp_t wait
)
1038 ClearPagePrivate(page
);
1042 static int f2fs_set_data_page_dirty(struct page
*page
)
1044 struct address_space
*mapping
= page
->mapping
;
1045 struct inode
*inode
= mapping
->host
;
1047 trace_f2fs_set_page_dirty(page
, DATA
);
1049 SetPageUptodate(page
);
1050 if (!PageDirty(page
)) {
1051 __set_page_dirty_nobuffers(page
);
1052 set_dirty_dir_page(inode
, page
);
1058 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
1060 return generic_block_bmap(mapping
, block
, get_data_block
);
1063 const struct address_space_operations f2fs_dblock_aops
= {
1064 .readpage
= f2fs_read_data_page
,
1065 .readpages
= f2fs_read_data_pages
,
1066 .writepage
= f2fs_write_data_page
,
1067 .writepages
= f2fs_write_data_pages
,
1068 .write_begin
= f2fs_write_begin
,
1069 .write_end
= f2fs_write_end
,
1070 .set_page_dirty
= f2fs_set_data_page_dirty
,
1071 .invalidatepage
= f2fs_invalidate_data_page
,
1072 .releasepage
= f2fs_release_data_page
,
1073 .direct_IO
= f2fs_direct_IO
,