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
)
32 bio_for_each_segment_all(bvec
, bio
, i
) {
33 struct page
*page
= bvec
->bv_page
;
36 SetPageUptodate(page
);
38 ClearPageUptodate(page
);
46 static void f2fs_write_end_io(struct bio
*bio
, int err
)
48 struct f2fs_sb_info
*sbi
= bio
->bi_private
;
52 bio_for_each_segment_all(bvec
, bio
, i
) {
53 struct page
*page
= bvec
->bv_page
;
57 set_bit(AS_EIO
, &page
->mapping
->flags
);
58 f2fs_stop_checkpoint(sbi
);
60 end_page_writeback(page
);
61 dec_page_count(sbi
, F2FS_WRITEBACK
);
64 if (!get_pages(sbi
, F2FS_WRITEBACK
) &&
65 !list_empty(&sbi
->cp_wait
.task_list
))
66 wake_up(&sbi
->cp_wait
);
72 * Low-level block read/write IO operations.
74 static struct bio
*__bio_alloc(struct f2fs_sb_info
*sbi
, block_t blk_addr
,
75 int npages
, bool is_read
)
79 /* No failure on bio allocation */
80 bio
= bio_alloc(GFP_NOIO
, npages
);
82 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
83 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blk_addr
);
84 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
85 bio
->bi_private
= sbi
;
90 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
92 struct f2fs_io_info
*fio
= &io
->fio
;
97 if (is_read_io(fio
->rw
))
98 trace_f2fs_submit_read_bio(io
->sbi
->sb
, fio
->rw
,
101 trace_f2fs_submit_write_bio(io
->sbi
->sb
, fio
->rw
,
104 submit_bio(fio
->rw
, io
->bio
);
108 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
109 enum page_type type
, int rw
)
111 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
112 struct f2fs_bio_info
*io
;
114 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
116 down_write(&io
->io_rwsem
);
118 /* change META to META_FLUSH in the checkpoint procedure */
119 if (type
>= META_FLUSH
) {
120 io
->fio
.type
= META_FLUSH
;
121 if (test_opt(sbi
, NOBARRIER
))
122 io
->fio
.rw
= WRITE_FLUSH
| REQ_META
| REQ_PRIO
;
124 io
->fio
.rw
= WRITE_FLUSH_FUA
| REQ_META
| REQ_PRIO
;
126 __submit_merged_bio(io
);
127 up_write(&io
->io_rwsem
);
131 * Fill the locked page with data located in the block address.
132 * Return unlocked page.
134 int f2fs_submit_page_bio(struct f2fs_sb_info
*sbi
, struct page
*page
,
135 block_t blk_addr
, int rw
)
139 trace_f2fs_submit_page_bio(page
, blk_addr
, rw
);
141 /* Allocate a new bio */
142 bio
= __bio_alloc(sbi
, blk_addr
, 1, is_read_io(rw
));
144 if (bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) < PAGE_CACHE_SIZE
) {
146 f2fs_put_page(page
, 1);
154 void f2fs_submit_page_mbio(struct f2fs_sb_info
*sbi
, struct page
*page
,
155 block_t blk_addr
, struct f2fs_io_info
*fio
)
157 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
158 struct f2fs_bio_info
*io
;
159 bool is_read
= is_read_io(fio
->rw
);
161 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
163 verify_block_addr(sbi
, blk_addr
);
165 down_write(&io
->io_rwsem
);
168 inc_page_count(sbi
, F2FS_WRITEBACK
);
170 if (io
->bio
&& (io
->last_block_in_bio
!= blk_addr
- 1 ||
171 io
->fio
.rw
!= fio
->rw
))
172 __submit_merged_bio(io
);
174 if (io
->bio
== NULL
) {
175 int bio_blocks
= MAX_BIO_BLOCKS(sbi
);
177 io
->bio
= __bio_alloc(sbi
, blk_addr
, bio_blocks
, is_read
);
181 if (bio_add_page(io
->bio
, page
, PAGE_CACHE_SIZE
, 0) <
183 __submit_merged_bio(io
);
187 io
->last_block_in_bio
= blk_addr
;
189 up_write(&io
->io_rwsem
);
190 trace_f2fs_submit_page_mbio(page
, fio
->rw
, fio
->type
, blk_addr
);
194 * Lock ordering for the change of data block address:
197 * update block addresses in the node page
199 static void __set_data_blkaddr(struct dnode_of_data
*dn
, block_t new_addr
)
201 struct f2fs_node
*rn
;
203 struct page
*node_page
= dn
->node_page
;
204 unsigned int ofs_in_node
= dn
->ofs_in_node
;
206 f2fs_wait_on_page_writeback(node_page
, NODE
);
208 rn
= F2FS_NODE(node_page
);
210 /* Get physical address of data block */
211 addr_array
= blkaddr_in_node(rn
);
212 addr_array
[ofs_in_node
] = cpu_to_le32(new_addr
);
213 set_page_dirty(node_page
);
216 int reserve_new_block(struct dnode_of_data
*dn
)
218 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
220 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
222 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, 1)))
225 trace_f2fs_reserve_new_block(dn
->inode
, dn
->nid
, dn
->ofs_in_node
);
227 __set_data_blkaddr(dn
, NEW_ADDR
);
228 dn
->data_blkaddr
= NEW_ADDR
;
229 mark_inode_dirty(dn
->inode
);
234 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
236 bool need_put
= dn
->inode_page
? false : true;
239 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
243 if (dn
->data_blkaddr
== NULL_ADDR
)
244 err
= reserve_new_block(dn
);
250 static int check_extent_cache(struct inode
*inode
, pgoff_t pgofs
,
251 struct buffer_head
*bh_result
)
253 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
254 pgoff_t start_fofs
, end_fofs
;
255 block_t start_blkaddr
;
257 if (is_inode_flag_set(fi
, FI_NO_EXTENT
))
260 read_lock(&fi
->ext
.ext_lock
);
261 if (fi
->ext
.len
== 0) {
262 read_unlock(&fi
->ext
.ext_lock
);
266 stat_inc_total_hit(inode
->i_sb
);
268 start_fofs
= fi
->ext
.fofs
;
269 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
270 start_blkaddr
= fi
->ext
.blk_addr
;
272 if (pgofs
>= start_fofs
&& pgofs
<= end_fofs
) {
273 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
276 clear_buffer_new(bh_result
);
277 map_bh(bh_result
, inode
->i_sb
,
278 start_blkaddr
+ pgofs
- start_fofs
);
279 count
= end_fofs
- pgofs
+ 1;
280 if (count
< (UINT_MAX
>> blkbits
))
281 bh_result
->b_size
= (count
<< blkbits
);
283 bh_result
->b_size
= UINT_MAX
;
285 stat_inc_read_hit(inode
->i_sb
);
286 read_unlock(&fi
->ext
.ext_lock
);
289 read_unlock(&fi
->ext
.ext_lock
);
293 void update_extent_cache(block_t blk_addr
, struct dnode_of_data
*dn
)
295 struct f2fs_inode_info
*fi
= F2FS_I(dn
->inode
);
296 pgoff_t fofs
, start_fofs
, end_fofs
;
297 block_t start_blkaddr
, end_blkaddr
;
298 int need_update
= true;
300 f2fs_bug_on(F2FS_I_SB(dn
->inode
), blk_addr
== NEW_ADDR
);
301 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), fi
) +
304 /* Update the page address in the parent node */
305 __set_data_blkaddr(dn
, blk_addr
);
307 if (is_inode_flag_set(fi
, FI_NO_EXTENT
))
310 write_lock(&fi
->ext
.ext_lock
);
312 start_fofs
= fi
->ext
.fofs
;
313 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
314 start_blkaddr
= fi
->ext
.blk_addr
;
315 end_blkaddr
= fi
->ext
.blk_addr
+ fi
->ext
.len
- 1;
317 /* Drop and initialize the matched extent */
318 if (fi
->ext
.len
== 1 && fofs
== start_fofs
)
322 if (fi
->ext
.len
== 0) {
323 if (blk_addr
!= NULL_ADDR
) {
325 fi
->ext
.blk_addr
= blk_addr
;
332 if (fofs
== start_fofs
- 1 && blk_addr
== start_blkaddr
- 1) {
340 if (fofs
== end_fofs
+ 1 && blk_addr
== end_blkaddr
+ 1) {
345 /* Split the existing extent */
346 if (fi
->ext
.len
> 1 &&
347 fofs
>= start_fofs
&& fofs
<= end_fofs
) {
348 if ((end_fofs
- fofs
) < (fi
->ext
.len
>> 1)) {
349 fi
->ext
.len
= fofs
- start_fofs
;
351 fi
->ext
.fofs
= fofs
+ 1;
352 fi
->ext
.blk_addr
= start_blkaddr
+
353 fofs
- start_fofs
+ 1;
354 fi
->ext
.len
-= fofs
- start_fofs
+ 1;
360 /* Finally, if the extent is very fragmented, let's drop the cache. */
361 if (fi
->ext
.len
< F2FS_MIN_EXTENT_LEN
) {
363 set_inode_flag(fi
, FI_NO_EXTENT
);
367 write_unlock(&fi
->ext
.ext_lock
);
373 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
, bool sync
)
375 struct address_space
*mapping
= inode
->i_mapping
;
376 struct dnode_of_data dn
;
380 page
= find_get_page(mapping
, index
);
381 if (page
&& PageUptodate(page
))
383 f2fs_put_page(page
, 0);
385 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
386 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
391 if (dn
.data_blkaddr
== NULL_ADDR
)
392 return ERR_PTR(-ENOENT
);
394 /* By fallocate(), there is no cached page, but with NEW_ADDR */
395 if (unlikely(dn
.data_blkaddr
== NEW_ADDR
))
396 return ERR_PTR(-EINVAL
);
398 page
= grab_cache_page(mapping
, index
);
400 return ERR_PTR(-ENOMEM
);
402 if (PageUptodate(page
)) {
407 err
= f2fs_submit_page_bio(F2FS_I_SB(inode
), page
, dn
.data_blkaddr
,
408 sync
? READ_SYNC
: READA
);
413 wait_on_page_locked(page
);
414 if (unlikely(!PageUptodate(page
))) {
415 f2fs_put_page(page
, 0);
416 return ERR_PTR(-EIO
);
423 * If it tries to access a hole, return an error.
424 * Because, the callers, functions in dir.c and GC, should be able to know
425 * whether this page exists or not.
427 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
)
429 struct address_space
*mapping
= inode
->i_mapping
;
430 struct dnode_of_data dn
;
435 page
= grab_cache_page(mapping
, index
);
437 return ERR_PTR(-ENOMEM
);
439 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
440 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
442 f2fs_put_page(page
, 1);
447 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
448 f2fs_put_page(page
, 1);
449 return ERR_PTR(-ENOENT
);
452 if (PageUptodate(page
))
456 * A new dentry page is allocated but not able to be written, since its
457 * new inode page couldn't be allocated due to -ENOSPC.
458 * In such the case, its blkaddr can be remained as NEW_ADDR.
459 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
461 if (dn
.data_blkaddr
== NEW_ADDR
) {
462 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
463 SetPageUptodate(page
);
467 err
= f2fs_submit_page_bio(F2FS_I_SB(inode
), page
,
468 dn
.data_blkaddr
, READ_SYNC
);
473 if (unlikely(!PageUptodate(page
))) {
474 f2fs_put_page(page
, 1);
475 return ERR_PTR(-EIO
);
477 if (unlikely(page
->mapping
!= mapping
)) {
478 f2fs_put_page(page
, 1);
485 * Caller ensures that this data page is never allocated.
486 * A new zero-filled data page is allocated in the page cache.
488 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
490 * Note that, ipage is set only by make_empty_dir.
492 struct page
*get_new_data_page(struct inode
*inode
,
493 struct page
*ipage
, pgoff_t index
, bool new_i_size
)
495 struct address_space
*mapping
= inode
->i_mapping
;
497 struct dnode_of_data dn
;
500 set_new_dnode(&dn
, inode
, ipage
, NULL
, 0);
501 err
= f2fs_reserve_block(&dn
, index
);
505 page
= grab_cache_page(mapping
, index
);
511 if (PageUptodate(page
))
514 if (dn
.data_blkaddr
== NEW_ADDR
) {
515 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
516 SetPageUptodate(page
);
518 err
= f2fs_submit_page_bio(F2FS_I_SB(inode
), page
,
519 dn
.data_blkaddr
, READ_SYNC
);
524 if (unlikely(!PageUptodate(page
))) {
525 f2fs_put_page(page
, 1);
529 if (unlikely(page
->mapping
!= mapping
)) {
530 f2fs_put_page(page
, 1);
536 i_size_read(inode
) < ((index
+ 1) << PAGE_CACHE_SHIFT
)) {
537 i_size_write(inode
, ((index
+ 1) << PAGE_CACHE_SHIFT
));
538 /* Only the directory inode sets new_i_size */
539 set_inode_flag(F2FS_I(inode
), FI_UPDATE_DIR
);
548 static int __allocate_data_block(struct dnode_of_data
*dn
)
550 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
551 struct f2fs_inode_info
*fi
= F2FS_I(dn
->inode
);
552 struct f2fs_summary sum
;
558 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
560 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, 1)))
563 __set_data_blkaddr(dn
, NEW_ADDR
);
564 dn
->data_blkaddr
= NEW_ADDR
;
566 get_node_info(sbi
, dn
->nid
, &ni
);
567 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
569 type
= CURSEG_WARM_DATA
;
571 allocate_data_block(sbi
, NULL
, NULL_ADDR
, &new_blkaddr
, &sum
, type
);
573 /* direct IO doesn't use extent cache to maximize the performance */
574 set_inode_flag(F2FS_I(dn
->inode
), FI_NO_EXTENT
);
575 update_extent_cache(new_blkaddr
, dn
);
576 clear_inode_flag(F2FS_I(dn
->inode
), FI_NO_EXTENT
);
579 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), fi
) +
581 if (i_size_read(dn
->inode
) < ((fofs
+ 1) << PAGE_CACHE_SHIFT
))
582 i_size_write(dn
->inode
, ((fofs
+ 1) << PAGE_CACHE_SHIFT
));
584 dn
->data_blkaddr
= new_blkaddr
;
589 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
590 * If original data blocks are allocated, then give them to blockdev.
592 * a. preallocate requested block addresses
593 * b. do not use extent cache for better performance
594 * c. give the block addresses to blockdev
596 static int __get_data_block(struct inode
*inode
, sector_t iblock
,
597 struct buffer_head
*bh_result
, int create
, bool fiemap
)
599 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
600 unsigned maxblocks
= bh_result
->b_size
>> blkbits
;
601 struct dnode_of_data dn
;
602 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE_RA
;
603 pgoff_t pgofs
, end_offset
;
604 int err
= 0, ofs
= 1;
605 bool allocated
= false;
607 /* Get the page offset from the block offset(iblock) */
608 pgofs
= (pgoff_t
)(iblock
>> (PAGE_CACHE_SHIFT
- blkbits
));
610 if (check_extent_cache(inode
, pgofs
, bh_result
))
614 f2fs_balance_fs(F2FS_I_SB(inode
));
615 f2fs_lock_op(F2FS_I_SB(inode
));
618 /* When reading holes, we need its node page */
619 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
620 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
626 if (dn
.data_blkaddr
== NEW_ADDR
&& !fiemap
)
629 if (dn
.data_blkaddr
!= NULL_ADDR
) {
630 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
632 err
= __allocate_data_block(&dn
);
636 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
641 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, F2FS_I(inode
));
642 bh_result
->b_size
= (((size_t)1) << blkbits
);
647 if (dn
.ofs_in_node
>= end_offset
) {
649 sync_inode_page(&dn
);
653 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
654 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
660 if (dn
.data_blkaddr
== NEW_ADDR
&& !fiemap
)
663 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, F2FS_I(inode
));
666 if (maxblocks
> (bh_result
->b_size
>> blkbits
)) {
667 block_t blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
668 if (blkaddr
== NULL_ADDR
&& create
) {
669 err
= __allocate_data_block(&dn
);
673 blkaddr
= dn
.data_blkaddr
;
675 /* Give more consecutive addresses for the readahead */
676 if (blkaddr
== (bh_result
->b_blocknr
+ ofs
)) {
680 bh_result
->b_size
+= (((size_t)1) << blkbits
);
686 sync_inode_page(&dn
);
691 f2fs_unlock_op(F2FS_I_SB(inode
));
693 trace_f2fs_get_data_block(inode
, iblock
, bh_result
, err
);
697 static int get_data_block(struct inode
*inode
, sector_t iblock
,
698 struct buffer_head
*bh_result
, int create
)
700 return __get_data_block(inode
, iblock
, bh_result
, create
, false);
703 static int get_data_block_fiemap(struct inode
*inode
, sector_t iblock
,
704 struct buffer_head
*bh_result
, int create
)
706 return __get_data_block(inode
, iblock
, bh_result
, create
, true);
709 int f2fs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
712 return generic_block_fiemap(inode
, fieinfo
,
713 start
, len
, get_data_block_fiemap
);
716 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
718 struct inode
*inode
= page
->mapping
->host
;
721 trace_f2fs_readpage(page
, DATA
);
723 /* If the file has inline data, try to read it directly */
724 if (f2fs_has_inline_data(inode
))
725 ret
= f2fs_read_inline_data(inode
, page
);
727 ret
= mpage_readpage(page
, get_data_block
);
732 static int f2fs_read_data_pages(struct file
*file
,
733 struct address_space
*mapping
,
734 struct list_head
*pages
, unsigned nr_pages
)
736 struct inode
*inode
= file
->f_mapping
->host
;
738 /* If the file has inline data, skip readpages */
739 if (f2fs_has_inline_data(inode
))
742 return mpage_readpages(mapping
, pages
, nr_pages
, get_data_block
);
745 int do_write_data_page(struct page
*page
, struct f2fs_io_info
*fio
)
747 struct inode
*inode
= page
->mapping
->host
;
748 block_t old_blkaddr
, new_blkaddr
;
749 struct dnode_of_data dn
;
752 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
753 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
757 old_blkaddr
= dn
.data_blkaddr
;
759 /* This page is already truncated */
760 if (old_blkaddr
== NULL_ADDR
)
763 set_page_writeback(page
);
766 * If current allocation needs SSR,
767 * it had better in-place writes for updated data.
769 if (unlikely(old_blkaddr
!= NEW_ADDR
&&
770 !is_cold_data(page
) &&
771 need_inplace_update(inode
))) {
772 rewrite_data_page(page
, old_blkaddr
, fio
);
773 set_inode_flag(F2FS_I(inode
), FI_UPDATE_WRITE
);
775 write_data_page(page
, &dn
, &new_blkaddr
, fio
);
776 update_extent_cache(new_blkaddr
, &dn
);
777 set_inode_flag(F2FS_I(inode
), FI_APPEND_WRITE
);
784 static int f2fs_write_data_page(struct page
*page
,
785 struct writeback_control
*wbc
)
787 struct inode
*inode
= page
->mapping
->host
;
788 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
789 loff_t i_size
= i_size_read(inode
);
790 const pgoff_t end_index
= ((unsigned long long) i_size
)
793 bool need_balance_fs
= false;
795 struct f2fs_io_info fio
= {
797 .rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
,
800 trace_f2fs_writepage(page
, DATA
);
802 if (page
->index
< end_index
)
806 * If the offset is out-of-range of file size,
807 * this page does not have to be written to disk.
809 offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
810 if ((page
->index
>= end_index
+ 1) || !offset
)
813 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
815 if (unlikely(sbi
->por_doing
))
818 /* Dentry blocks are controlled by checkpoint */
819 if (S_ISDIR(inode
->i_mode
)) {
820 if (unlikely(f2fs_cp_error(sbi
)))
822 err
= do_write_data_page(page
, &fio
);
826 /* we should bypass data pages to proceed the kworkder jobs */
827 if (unlikely(f2fs_cp_error(sbi
))) {
833 if (!wbc
->for_reclaim
)
834 need_balance_fs
= true;
835 else if (has_not_enough_free_secs(sbi
, 0))
840 if (f2fs_has_inline_data(inode
))
841 err
= f2fs_write_inline_data(inode
, page
);
843 err
= do_write_data_page(page
, &fio
);
846 if (err
&& err
!= -ENOENT
)
849 clear_cold_data(page
);
851 inode_dec_dirty_pages(inode
);
854 f2fs_balance_fs(sbi
);
855 if (wbc
->for_reclaim
)
856 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
860 redirty_page_for_writepage(wbc
, page
);
861 return AOP_WRITEPAGE_ACTIVATE
;
864 static int __f2fs_writepage(struct page
*page
, struct writeback_control
*wbc
,
867 struct address_space
*mapping
= data
;
868 int ret
= mapping
->a_ops
->writepage(page
, wbc
);
869 mapping_set_error(mapping
, ret
);
873 static int f2fs_write_data_pages(struct address_space
*mapping
,
874 struct writeback_control
*wbc
)
876 struct inode
*inode
= mapping
->host
;
877 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
882 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
884 /* deal with chardevs and other special file */
885 if (!mapping
->a_ops
->writepage
)
888 if (S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_NONE
&&
889 get_dirty_pages(inode
) < nr_pages_to_skip(sbi
, DATA
) &&
890 available_free_memory(sbi
, DIRTY_DENTS
))
893 diff
= nr_pages_to_write(sbi
, DATA
, wbc
);
895 if (!S_ISDIR(inode
->i_mode
)) {
896 mutex_lock(&sbi
->writepages
);
899 ret
= write_cache_pages(mapping
, wbc
, __f2fs_writepage
, mapping
);
901 mutex_unlock(&sbi
->writepages
);
903 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
905 remove_dirty_dir_inode(inode
);
907 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- diff
);
911 wbc
->pages_skipped
+= get_dirty_pages(inode
);
915 static void f2fs_write_failed(struct address_space
*mapping
, loff_t to
)
917 struct inode
*inode
= mapping
->host
;
919 if (to
> inode
->i_size
) {
920 truncate_pagecache(inode
, inode
->i_size
);
921 truncate_blocks(inode
, inode
->i_size
, true);
925 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
926 loff_t pos
, unsigned len
, unsigned flags
,
927 struct page
**pagep
, void **fsdata
)
929 struct inode
*inode
= mapping
->host
;
930 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
931 struct page
*page
, *ipage
;
932 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_CACHE_SHIFT
;
933 struct dnode_of_data dn
;
936 trace_f2fs_write_begin(inode
, pos
, len
, flags
);
938 f2fs_balance_fs(sbi
);
940 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
950 /* check inline_data */
951 ipage
= get_node_page(sbi
, inode
->i_ino
);
955 set_new_dnode(&dn
, inode
, ipage
, ipage
, 0);
957 if (f2fs_has_inline_data(inode
)) {
958 if (pos
+ len
<= MAX_INLINE_DATA
) {
959 read_inline_data(page
, ipage
);
960 set_inode_flag(F2FS_I(inode
), FI_DATA_EXIST
);
961 sync_inode_page(&dn
);
963 } else if (page
->index
== 0) {
964 err
= f2fs_convert_inline_page(&dn
, page
);
968 struct page
*p
= grab_cache_page(inode
->i_mapping
, 0);
973 err
= f2fs_convert_inline_page(&dn
, p
);
979 err
= f2fs_reserve_block(&dn
, index
);
986 if ((len
== PAGE_CACHE_SIZE
) || PageUptodate(page
))
989 f2fs_wait_on_page_writeback(page
, DATA
);
991 if ((pos
& PAGE_CACHE_MASK
) >= i_size_read(inode
)) {
992 unsigned start
= pos
& (PAGE_CACHE_SIZE
- 1);
993 unsigned end
= start
+ len
;
995 /* Reading beyond i_size is simple: memset to zero */
996 zero_user_segments(page
, 0, start
, end
, PAGE_CACHE_SIZE
);
1000 if (dn
.data_blkaddr
== NEW_ADDR
) {
1001 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
1003 err
= f2fs_submit_page_bio(sbi
, page
, dn
.data_blkaddr
,
1009 if (unlikely(!PageUptodate(page
))) {
1010 f2fs_put_page(page
, 1);
1014 if (unlikely(page
->mapping
!= mapping
)) {
1015 f2fs_put_page(page
, 1);
1020 SetPageUptodate(page
);
1021 clear_cold_data(page
);
1025 f2fs_unlock_op(sbi
);
1026 f2fs_put_page(page
, 1);
1028 f2fs_write_failed(mapping
, pos
+ len
);
1032 static int f2fs_write_end(struct file
*file
,
1033 struct address_space
*mapping
,
1034 loff_t pos
, unsigned len
, unsigned copied
,
1035 struct page
*page
, void *fsdata
)
1037 struct inode
*inode
= page
->mapping
->host
;
1039 trace_f2fs_write_end(inode
, pos
, len
, copied
);
1041 set_page_dirty(page
);
1043 if (pos
+ copied
> i_size_read(inode
)) {
1044 i_size_write(inode
, pos
+ copied
);
1045 mark_inode_dirty(inode
);
1046 update_inode_page(inode
);
1049 f2fs_put_page(page
, 1);
1053 static int check_direct_IO(struct inode
*inode
, int rw
,
1054 struct iov_iter
*iter
, loff_t offset
)
1056 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
1061 if (offset
& blocksize_mask
)
1064 if (iov_iter_alignment(iter
) & blocksize_mask
)
1070 static ssize_t
f2fs_direct_IO(int rw
, struct kiocb
*iocb
,
1071 struct iov_iter
*iter
, loff_t offset
)
1073 struct file
*file
= iocb
->ki_filp
;
1074 struct address_space
*mapping
= file
->f_mapping
;
1075 struct inode
*inode
= mapping
->host
;
1076 size_t count
= iov_iter_count(iter
);
1079 /* we don't need to use inline_data strictly */
1080 if (f2fs_has_inline_data(inode
)) {
1081 err
= f2fs_convert_inline_inode(inode
);
1086 if (check_direct_IO(inode
, rw
, iter
, offset
))
1089 trace_f2fs_direct_IO_enter(inode
, offset
, count
, rw
);
1091 err
= blockdev_direct_IO(rw
, iocb
, inode
, iter
, offset
, get_data_block
);
1092 if (err
< 0 && (rw
& WRITE
))
1093 f2fs_write_failed(mapping
, offset
+ count
);
1095 trace_f2fs_direct_IO_exit(inode
, offset
, count
, rw
, err
);
1100 static void f2fs_invalidate_data_page(struct page
*page
, unsigned int offset
,
1101 unsigned int length
)
1103 struct inode
*inode
= page
->mapping
->host
;
1105 if (offset
% PAGE_CACHE_SIZE
|| length
!= PAGE_CACHE_SIZE
)
1108 if (f2fs_is_atomic_file(inode
) || f2fs_is_volatile_file(inode
))
1109 invalidate_inmem_page(inode
, page
);
1111 if (PageDirty(page
))
1112 inode_dec_dirty_pages(inode
);
1113 ClearPagePrivate(page
);
1116 static int f2fs_release_data_page(struct page
*page
, gfp_t wait
)
1118 ClearPagePrivate(page
);
1122 static int f2fs_set_data_page_dirty(struct page
*page
)
1124 struct address_space
*mapping
= page
->mapping
;
1125 struct inode
*inode
= mapping
->host
;
1127 trace_f2fs_set_page_dirty(page
, DATA
);
1129 SetPageUptodate(page
);
1131 if (f2fs_is_atomic_file(inode
) || f2fs_is_volatile_file(inode
)) {
1132 register_inmem_page(inode
, page
);
1136 mark_inode_dirty(inode
);
1138 if (!PageDirty(page
)) {
1139 __set_page_dirty_nobuffers(page
);
1140 update_dirty_page(inode
, page
);
1146 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
1148 struct inode
*inode
= mapping
->host
;
1150 /* we don't need to use inline_data strictly */
1151 if (f2fs_has_inline_data(inode
)) {
1152 int err
= f2fs_convert_inline_inode(inode
);
1156 return generic_block_bmap(mapping
, block
, get_data_block
);
1159 const struct address_space_operations f2fs_dblock_aops
= {
1160 .readpage
= f2fs_read_data_page
,
1161 .readpages
= f2fs_read_data_pages
,
1162 .writepage
= f2fs_write_data_page
,
1163 .writepages
= f2fs_write_data_pages
,
1164 .write_begin
= f2fs_write_begin
,
1165 .write_end
= f2fs_write_end
,
1166 .set_page_dirty
= f2fs_set_data_page_dirty
,
1167 .invalidatepage
= f2fs_invalidate_data_page
,
1168 .releasepage
= f2fs_release_data_page
,
1169 .direct_IO
= f2fs_direct_IO
,