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/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
23 #include <linux/memcontrol.h>
24 #include <linux/cleancache.h>
30 #include <trace/events/f2fs.h>
32 static void f2fs_read_end_io(struct bio
*bio
)
37 if (f2fs_bio_encrypted(bio
)) {
39 fscrypt_release_ctx(bio
->bi_private
);
41 fscrypt_decrypt_bio_pages(bio
->bi_private
, bio
);
46 bio_for_each_segment_all(bvec
, bio
, i
) {
47 struct page
*page
= bvec
->bv_page
;
50 SetPageUptodate(page
);
52 ClearPageUptodate(page
);
60 static void f2fs_write_end_io(struct bio
*bio
)
62 struct f2fs_sb_info
*sbi
= bio
->bi_private
;
66 bio_for_each_segment_all(bvec
, bio
, i
) {
67 struct page
*page
= bvec
->bv_page
;
69 fscrypt_pullback_bio_page(&page
, true);
71 if (unlikely(bio
->bi_error
)) {
72 set_bit(AS_EIO
, &page
->mapping
->flags
);
73 f2fs_stop_checkpoint(sbi
, true);
75 end_page_writeback(page
);
77 if (atomic_dec_and_test(&sbi
->nr_wb_bios
) &&
78 wq_has_sleeper(&sbi
->cp_wait
))
79 wake_up(&sbi
->cp_wait
);
85 * Low-level block read/write IO operations.
87 static struct bio
*__bio_alloc(struct f2fs_sb_info
*sbi
, block_t blk_addr
,
88 int npages
, bool is_read
)
92 bio
= f2fs_bio_alloc(npages
);
94 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
95 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blk_addr
);
96 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
97 bio
->bi_private
= is_read
? NULL
: sbi
;
102 static inline void __submit_bio(struct f2fs_sb_info
*sbi
, int rw
,
103 struct bio
*bio
, enum page_type type
)
105 if (!is_read_io(rw
)) {
106 atomic_inc(&sbi
->nr_wb_bios
);
107 if (f2fs_sb_mounted_hmsmr(sbi
->sb
) &&
108 current
->plug
&& (type
== DATA
|| type
== NODE
))
109 blk_finish_plug(current
->plug
);
114 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
116 struct f2fs_io_info
*fio
= &io
->fio
;
121 if (is_read_io(fio
->rw
))
122 trace_f2fs_submit_read_bio(io
->sbi
->sb
, fio
, io
->bio
);
124 trace_f2fs_submit_write_bio(io
->sbi
->sb
, fio
, io
->bio
);
126 __submit_bio(io
->sbi
, fio
->rw
, io
->bio
, fio
->type
);
130 static bool __has_merged_page(struct f2fs_bio_info
*io
, struct inode
*inode
,
131 struct page
*page
, nid_t ino
)
133 struct bio_vec
*bvec
;
140 if (!inode
&& !page
&& !ino
)
143 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
145 if (bvec
->bv_page
->mapping
)
146 target
= bvec
->bv_page
;
148 target
= fscrypt_control_page(bvec
->bv_page
);
150 if (inode
&& inode
== target
->mapping
->host
)
152 if (page
&& page
== target
)
154 if (ino
&& ino
== ino_of_node(target
))
161 static bool has_merged_page(struct f2fs_sb_info
*sbi
, struct inode
*inode
,
162 struct page
*page
, nid_t ino
,
165 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
166 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
169 down_read(&io
->io_rwsem
);
170 ret
= __has_merged_page(io
, inode
, page
, ino
);
171 up_read(&io
->io_rwsem
);
175 static void __f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
176 struct inode
*inode
, struct page
*page
,
177 nid_t ino
, enum page_type type
, int rw
)
179 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
180 struct f2fs_bio_info
*io
;
182 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
184 down_write(&io
->io_rwsem
);
186 if (!__has_merged_page(io
, inode
, page
, ino
))
189 /* change META to META_FLUSH in the checkpoint procedure */
190 if (type
>= META_FLUSH
) {
191 io
->fio
.type
= META_FLUSH
;
192 if (test_opt(sbi
, NOBARRIER
))
193 io
->fio
.rw
= WRITE_FLUSH
| REQ_META
| REQ_PRIO
;
195 io
->fio
.rw
= WRITE_FLUSH_FUA
| REQ_META
| REQ_PRIO
;
197 __submit_merged_bio(io
);
199 up_write(&io
->io_rwsem
);
202 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
, enum page_type type
,
205 __f2fs_submit_merged_bio(sbi
, NULL
, NULL
, 0, type
, rw
);
208 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info
*sbi
,
209 struct inode
*inode
, struct page
*page
,
210 nid_t ino
, enum page_type type
, int rw
)
212 if (has_merged_page(sbi
, inode
, page
, ino
, type
))
213 __f2fs_submit_merged_bio(sbi
, inode
, page
, ino
, type
, rw
);
216 void f2fs_flush_merged_bios(struct f2fs_sb_info
*sbi
)
218 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
219 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
220 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
224 * Fill the locked page with data located in the block address.
225 * Return unlocked page.
227 int f2fs_submit_page_bio(struct f2fs_io_info
*fio
)
230 struct page
*page
= fio
->encrypted_page
?
231 fio
->encrypted_page
: fio
->page
;
233 trace_f2fs_submit_page_bio(page
, fio
);
234 f2fs_trace_ios(fio
, 0);
236 /* Allocate a new bio */
237 bio
= __bio_alloc(fio
->sbi
, fio
->new_blkaddr
, 1, is_read_io(fio
->rw
));
239 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
244 __submit_bio(fio
->sbi
, fio
->rw
, bio
, fio
->type
);
248 void f2fs_submit_page_mbio(struct f2fs_io_info
*fio
)
250 struct f2fs_sb_info
*sbi
= fio
->sbi
;
251 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
252 struct f2fs_bio_info
*io
;
253 bool is_read
= is_read_io(fio
->rw
);
254 struct page
*bio_page
;
256 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
258 if (fio
->old_blkaddr
!= NEW_ADDR
)
259 verify_block_addr(sbi
, fio
->old_blkaddr
);
260 verify_block_addr(sbi
, fio
->new_blkaddr
);
262 down_write(&io
->io_rwsem
);
264 if (io
->bio
&& (io
->last_block_in_bio
!= fio
->new_blkaddr
- 1 ||
265 io
->fio
.rw
!= fio
->rw
))
266 __submit_merged_bio(io
);
268 if (io
->bio
== NULL
) {
269 int bio_blocks
= MAX_BIO_BLOCKS(sbi
);
271 io
->bio
= __bio_alloc(sbi
, fio
->new_blkaddr
,
272 bio_blocks
, is_read
);
276 bio_page
= fio
->encrypted_page
? fio
->encrypted_page
: fio
->page
;
278 if (bio_add_page(io
->bio
, bio_page
, PAGE_SIZE
, 0) <
280 __submit_merged_bio(io
);
284 io
->last_block_in_bio
= fio
->new_blkaddr
;
285 f2fs_trace_ios(fio
, 0);
287 up_write(&io
->io_rwsem
);
288 trace_f2fs_submit_page_mbio(fio
->page
, fio
);
291 static void __set_data_blkaddr(struct dnode_of_data
*dn
)
293 struct f2fs_node
*rn
= F2FS_NODE(dn
->node_page
);
296 /* Get physical address of data block */
297 addr_array
= blkaddr_in_node(rn
);
298 addr_array
[dn
->ofs_in_node
] = cpu_to_le32(dn
->data_blkaddr
);
302 * Lock ordering for the change of data block address:
305 * update block addresses in the node page
307 void set_data_blkaddr(struct dnode_of_data
*dn
)
309 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
310 __set_data_blkaddr(dn
);
311 if (set_page_dirty(dn
->node_page
))
312 dn
->node_changed
= true;
315 void f2fs_update_data_blkaddr(struct dnode_of_data
*dn
, block_t blkaddr
)
317 dn
->data_blkaddr
= blkaddr
;
318 set_data_blkaddr(dn
);
319 f2fs_update_extent_cache(dn
);
322 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
323 int reserve_new_blocks(struct dnode_of_data
*dn
, blkcnt_t count
)
325 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
330 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
332 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
335 trace_f2fs_reserve_new_blocks(dn
->inode
, dn
->nid
,
336 dn
->ofs_in_node
, count
);
338 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
340 for (; count
> 0; dn
->ofs_in_node
++) {
342 datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
343 if (blkaddr
== NULL_ADDR
) {
344 dn
->data_blkaddr
= NEW_ADDR
;
345 __set_data_blkaddr(dn
);
350 if (set_page_dirty(dn
->node_page
))
351 dn
->node_changed
= true;
355 /* Should keep dn->ofs_in_node unchanged */
356 int reserve_new_block(struct dnode_of_data
*dn
)
358 unsigned int ofs_in_node
= dn
->ofs_in_node
;
361 ret
= reserve_new_blocks(dn
, 1);
362 dn
->ofs_in_node
= ofs_in_node
;
366 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
368 bool need_put
= dn
->inode_page
? false : true;
371 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
375 if (dn
->data_blkaddr
== NULL_ADDR
)
376 err
= reserve_new_block(dn
);
382 int f2fs_get_block(struct dnode_of_data
*dn
, pgoff_t index
)
384 struct extent_info ei
;
385 struct inode
*inode
= dn
->inode
;
387 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
388 dn
->data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
392 return f2fs_reserve_block(dn
, index
);
395 struct page
*get_read_data_page(struct inode
*inode
, pgoff_t index
,
396 int rw
, bool for_write
)
398 struct address_space
*mapping
= inode
->i_mapping
;
399 struct dnode_of_data dn
;
401 struct extent_info ei
;
403 struct f2fs_io_info fio
= {
404 .sbi
= F2FS_I_SB(inode
),
407 .encrypted_page
= NULL
,
410 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
411 return read_mapping_page(mapping
, index
, NULL
);
413 page
= f2fs_grab_cache_page(mapping
, index
, for_write
);
415 return ERR_PTR(-ENOMEM
);
417 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
418 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
422 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
423 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
428 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
433 if (PageUptodate(page
)) {
439 * A new dentry page is allocated but not able to be written, since its
440 * new inode page couldn't be allocated due to -ENOSPC.
441 * In such the case, its blkaddr can be remained as NEW_ADDR.
442 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
444 if (dn
.data_blkaddr
== NEW_ADDR
) {
445 zero_user_segment(page
, 0, PAGE_SIZE
);
446 SetPageUptodate(page
);
451 fio
.new_blkaddr
= fio
.old_blkaddr
= dn
.data_blkaddr
;
453 err
= f2fs_submit_page_bio(&fio
);
459 f2fs_put_page(page
, 1);
463 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
)
465 struct address_space
*mapping
= inode
->i_mapping
;
468 page
= find_get_page(mapping
, index
);
469 if (page
&& PageUptodate(page
))
471 f2fs_put_page(page
, 0);
473 page
= get_read_data_page(inode
, index
, READ_SYNC
, false);
477 if (PageUptodate(page
))
480 wait_on_page_locked(page
);
481 if (unlikely(!PageUptodate(page
))) {
482 f2fs_put_page(page
, 0);
483 return ERR_PTR(-EIO
);
489 * If it tries to access a hole, return an error.
490 * Because, the callers, functions in dir.c and GC, should be able to know
491 * whether this page exists or not.
493 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
,
496 struct address_space
*mapping
= inode
->i_mapping
;
499 page
= get_read_data_page(inode
, index
, READ_SYNC
, for_write
);
503 /* wait for read completion */
505 if (unlikely(page
->mapping
!= mapping
)) {
506 f2fs_put_page(page
, 1);
509 if (unlikely(!PageUptodate(page
))) {
510 f2fs_put_page(page
, 1);
511 return ERR_PTR(-EIO
);
517 * Caller ensures that this data page is never allocated.
518 * A new zero-filled data page is allocated in the page cache.
520 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
522 * Note that, ipage is set only by make_empty_dir, and if any error occur,
523 * ipage should be released by this function.
525 struct page
*get_new_data_page(struct inode
*inode
,
526 struct page
*ipage
, pgoff_t index
, bool new_i_size
)
528 struct address_space
*mapping
= inode
->i_mapping
;
530 struct dnode_of_data dn
;
533 page
= f2fs_grab_cache_page(mapping
, index
, true);
536 * before exiting, we should make sure ipage will be released
537 * if any error occur.
539 f2fs_put_page(ipage
, 1);
540 return ERR_PTR(-ENOMEM
);
543 set_new_dnode(&dn
, inode
, ipage
, NULL
, 0);
544 err
= f2fs_reserve_block(&dn
, index
);
546 f2fs_put_page(page
, 1);
552 if (PageUptodate(page
))
555 if (dn
.data_blkaddr
== NEW_ADDR
) {
556 zero_user_segment(page
, 0, PAGE_SIZE
);
557 SetPageUptodate(page
);
559 f2fs_put_page(page
, 1);
561 /* if ipage exists, blkaddr should be NEW_ADDR */
562 f2fs_bug_on(F2FS_I_SB(inode
), ipage
);
563 page
= get_lock_data_page(inode
, index
, true);
568 if (new_i_size
&& i_size_read(inode
) <
569 ((loff_t
)(index
+ 1) << PAGE_SHIFT
))
570 f2fs_i_size_write(inode
, ((loff_t
)(index
+ 1) << PAGE_SHIFT
));
574 static int __allocate_data_block(struct dnode_of_data
*dn
)
576 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
577 struct f2fs_summary sum
;
579 int seg
= CURSEG_WARM_DATA
;
583 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
586 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
587 if (dn
->data_blkaddr
== NEW_ADDR
)
590 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
594 get_node_info(sbi
, dn
->nid
, &ni
);
595 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
597 if (dn
->ofs_in_node
== 0 && dn
->inode_page
== dn
->node_page
)
598 seg
= CURSEG_DIRECT_IO
;
600 allocate_data_block(sbi
, NULL
, dn
->data_blkaddr
, &dn
->data_blkaddr
,
602 set_data_blkaddr(dn
);
605 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), dn
->inode
) +
607 if (i_size_read(dn
->inode
) < ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
))
608 f2fs_i_size_write(dn
->inode
,
609 ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
));
613 ssize_t
f2fs_preallocate_blocks(struct kiocb
*iocb
, struct iov_iter
*from
)
615 struct inode
*inode
= file_inode(iocb
->ki_filp
);
616 struct f2fs_map_blocks map
;
619 map
.m_lblk
= F2FS_BLK_ALIGN(iocb
->ki_pos
);
620 map
.m_len
= F2FS_BYTES_TO_BLK(iov_iter_count(from
));
621 map
.m_next_pgofs
= NULL
;
623 if (f2fs_encrypted_inode(inode
))
626 if (iocb
->ki_flags
& IOCB_DIRECT
) {
627 ret
= f2fs_convert_inline_inode(inode
);
630 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_DIO
);
632 if (iocb
->ki_pos
+ iov_iter_count(from
) > MAX_INLINE_DATA
) {
633 ret
= f2fs_convert_inline_inode(inode
);
637 if (!f2fs_has_inline_data(inode
))
638 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_AIO
);
643 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
644 * f2fs_map_blocks structure.
645 * If original data blocks are allocated, then give them to blockdev.
647 * a. preallocate requested block addresses
648 * b. do not use extent cache for better performance
649 * c. give the block addresses to blockdev
651 int f2fs_map_blocks(struct inode
*inode
, struct f2fs_map_blocks
*map
,
652 int create
, int flag
)
654 unsigned int maxblocks
= map
->m_len
;
655 struct dnode_of_data dn
;
656 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
657 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE
;
658 pgoff_t pgofs
, end_offset
, end
;
659 int err
= 0, ofs
= 1;
660 unsigned int ofs_in_node
, last_ofs_in_node
;
662 struct extent_info ei
;
663 bool allocated
= false;
669 /* it only supports block size == page size */
670 pgofs
= (pgoff_t
)map
->m_lblk
;
671 end
= pgofs
+ maxblocks
;
673 if (!create
&& f2fs_lookup_extent_cache(inode
, pgofs
, &ei
)) {
674 map
->m_pblk
= ei
.blk
+ pgofs
- ei
.fofs
;
675 map
->m_len
= min((pgoff_t
)maxblocks
, ei
.fofs
+ ei
.len
- pgofs
);
676 map
->m_flags
= F2FS_MAP_MAPPED
;
684 /* When reading holes, we need its node page */
685 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
686 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
688 if (flag
== F2FS_GET_BLOCK_BMAP
)
690 if (err
== -ENOENT
) {
692 if (map
->m_next_pgofs
)
694 get_next_page_offset(&dn
, pgofs
);
700 ofs_in_node
= dn
.ofs_in_node
;
701 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, inode
);
704 blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
706 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
) {
708 if (unlikely(f2fs_cp_error(sbi
))) {
712 if (flag
== F2FS_GET_BLOCK_PRE_AIO
) {
713 if (blkaddr
== NULL_ADDR
) {
715 last_ofs_in_node
= dn
.ofs_in_node
;
718 err
= __allocate_data_block(&dn
);
720 set_inode_flag(inode
, FI_APPEND_WRITE
);
726 map
->m_flags
= F2FS_MAP_NEW
;
727 blkaddr
= dn
.data_blkaddr
;
729 if (flag
== F2FS_GET_BLOCK_BMAP
) {
733 if (flag
== F2FS_GET_BLOCK_FIEMAP
&&
734 blkaddr
== NULL_ADDR
) {
735 if (map
->m_next_pgofs
)
736 *map
->m_next_pgofs
= pgofs
+ 1;
738 if (flag
!= F2FS_GET_BLOCK_FIEMAP
||
744 if (flag
== F2FS_GET_BLOCK_PRE_AIO
)
747 if (map
->m_len
== 0) {
748 /* preallocated unwritten block should be mapped for fiemap. */
749 if (blkaddr
== NEW_ADDR
)
750 map
->m_flags
|= F2FS_MAP_UNWRITTEN
;
751 map
->m_flags
|= F2FS_MAP_MAPPED
;
753 map
->m_pblk
= blkaddr
;
755 } else if ((map
->m_pblk
!= NEW_ADDR
&&
756 blkaddr
== (map
->m_pblk
+ ofs
)) ||
757 (map
->m_pblk
== NEW_ADDR
&& blkaddr
== NEW_ADDR
) ||
758 flag
== F2FS_GET_BLOCK_PRE_DIO
) {
769 /* preallocate blocks in batch for one dnode page */
770 if (flag
== F2FS_GET_BLOCK_PRE_AIO
&&
771 (pgofs
== end
|| dn
.ofs_in_node
== end_offset
)) {
773 dn
.ofs_in_node
= ofs_in_node
;
774 err
= reserve_new_blocks(&dn
, prealloc
);
778 map
->m_len
+= dn
.ofs_in_node
- ofs_in_node
;
779 if (prealloc
&& dn
.ofs_in_node
!= last_ofs_in_node
+ 1) {
783 dn
.ofs_in_node
= end_offset
;
788 else if (dn
.ofs_in_node
< end_offset
)
795 f2fs_balance_fs(sbi
, allocated
);
805 f2fs_balance_fs(sbi
, allocated
);
808 trace_f2fs_map_blocks(inode
, map
, err
);
812 static int __get_data_block(struct inode
*inode
, sector_t iblock
,
813 struct buffer_head
*bh
, int create
, int flag
,
816 struct f2fs_map_blocks map
;
820 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
821 map
.m_next_pgofs
= next_pgofs
;
823 ret
= f2fs_map_blocks(inode
, &map
, create
, flag
);
825 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
826 bh
->b_state
= (bh
->b_state
& ~F2FS_MAP_FLAGS
) | map
.m_flags
;
827 bh
->b_size
= map
.m_len
<< inode
->i_blkbits
;
832 static int get_data_block(struct inode
*inode
, sector_t iblock
,
833 struct buffer_head
*bh_result
, int create
, int flag
,
836 return __get_data_block(inode
, iblock
, bh_result
, create
,
840 static int get_data_block_dio(struct inode
*inode
, sector_t iblock
,
841 struct buffer_head
*bh_result
, int create
)
843 return __get_data_block(inode
, iblock
, bh_result
, create
,
844 F2FS_GET_BLOCK_DIO
, NULL
);
847 static int get_data_block_bmap(struct inode
*inode
, sector_t iblock
,
848 struct buffer_head
*bh_result
, int create
)
850 /* Block number less than F2FS MAX BLOCKS */
851 if (unlikely(iblock
>= F2FS_I_SB(inode
)->max_file_blocks
))
854 return __get_data_block(inode
, iblock
, bh_result
, create
,
855 F2FS_GET_BLOCK_BMAP
, NULL
);
858 static inline sector_t
logical_to_blk(struct inode
*inode
, loff_t offset
)
860 return (offset
>> inode
->i_blkbits
);
863 static inline loff_t
blk_to_logical(struct inode
*inode
, sector_t blk
)
865 return (blk
<< inode
->i_blkbits
);
868 int f2fs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
871 struct buffer_head map_bh
;
872 sector_t start_blk
, last_blk
;
875 u64 logical
= 0, phys
= 0, size
= 0;
879 ret
= fiemap_check_flags(fieinfo
, FIEMAP_FLAG_SYNC
);
883 if (f2fs_has_inline_data(inode
)) {
884 ret
= f2fs_inline_data_fiemap(inode
, fieinfo
, start
, len
);
891 isize
= i_size_read(inode
);
895 if (start
+ len
> isize
)
898 if (logical_to_blk(inode
, len
) == 0)
899 len
= blk_to_logical(inode
, 1);
901 start_blk
= logical_to_blk(inode
, start
);
902 last_blk
= logical_to_blk(inode
, start
+ len
- 1);
905 memset(&map_bh
, 0, sizeof(struct buffer_head
));
908 ret
= get_data_block(inode
, start_blk
, &map_bh
, 0,
909 F2FS_GET_BLOCK_FIEMAP
, &next_pgofs
);
914 if (!buffer_mapped(&map_bh
)) {
915 start_blk
= next_pgofs
;
916 /* Go through holes util pass the EOF */
917 if (blk_to_logical(inode
, start_blk
) < isize
)
919 /* Found a hole beyond isize means no more extents.
920 * Note that the premise is that filesystems don't
921 * punch holes beyond isize and keep size unchanged.
923 flags
|= FIEMAP_EXTENT_LAST
;
927 if (f2fs_encrypted_inode(inode
))
928 flags
|= FIEMAP_EXTENT_DATA_ENCRYPTED
;
930 ret
= fiemap_fill_next_extent(fieinfo
, logical
,
934 if (start_blk
> last_blk
|| ret
)
937 logical
= blk_to_logical(inode
, start_blk
);
938 phys
= blk_to_logical(inode
, map_bh
.b_blocknr
);
939 size
= map_bh
.b_size
;
941 if (buffer_unwritten(&map_bh
))
942 flags
= FIEMAP_EXTENT_UNWRITTEN
;
944 start_blk
+= logical_to_blk(inode
, size
);
948 if (fatal_signal_pending(current
))
960 struct bio
*f2fs_grab_bio(struct inode
*inode
, block_t blkaddr
,
963 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
964 struct fscrypt_ctx
*ctx
= NULL
;
965 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
968 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
969 ctx
= fscrypt_get_ctx(inode
, GFP_NOFS
);
971 return ERR_CAST(ctx
);
973 /* wait the page to be moved by cleaning */
974 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
977 bio
= bio_alloc(GFP_KERNEL
, min_t(int, nr_pages
, BIO_MAX_PAGES
));
980 fscrypt_release_ctx(ctx
);
981 return ERR_PTR(-ENOMEM
);
984 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blkaddr
);
985 bio
->bi_end_io
= f2fs_read_end_io
;
986 bio
->bi_private
= ctx
;
992 * This function was originally taken from fs/mpage.c, and customized for f2fs.
993 * Major change was from block_size == page_size in f2fs by default.
995 static int f2fs_mpage_readpages(struct address_space
*mapping
,
996 struct list_head
*pages
, struct page
*page
,
999 struct bio
*bio
= NULL
;
1001 sector_t last_block_in_bio
= 0;
1002 struct inode
*inode
= mapping
->host
;
1003 const unsigned blkbits
= inode
->i_blkbits
;
1004 const unsigned blocksize
= 1 << blkbits
;
1005 sector_t block_in_file
;
1006 sector_t last_block
;
1007 sector_t last_block_in_file
;
1009 struct f2fs_map_blocks map
;
1015 map
.m_next_pgofs
= NULL
;
1017 for (page_idx
= 0; nr_pages
; page_idx
++, nr_pages
--) {
1019 prefetchw(&page
->flags
);
1021 page
= list_entry(pages
->prev
, struct page
, lru
);
1022 list_del(&page
->lru
);
1023 if (add_to_page_cache_lru(page
, mapping
,
1024 page
->index
, GFP_KERNEL
))
1028 block_in_file
= (sector_t
)page
->index
;
1029 last_block
= block_in_file
+ nr_pages
;
1030 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >>
1032 if (last_block
> last_block_in_file
)
1033 last_block
= last_block_in_file
;
1036 * Map blocks using the previous result first.
1038 if ((map
.m_flags
& F2FS_MAP_MAPPED
) &&
1039 block_in_file
> map
.m_lblk
&&
1040 block_in_file
< (map
.m_lblk
+ map
.m_len
))
1044 * Then do more f2fs_map_blocks() calls until we are
1045 * done with this page.
1049 if (block_in_file
< last_block
) {
1050 map
.m_lblk
= block_in_file
;
1051 map
.m_len
= last_block
- block_in_file
;
1053 if (f2fs_map_blocks(inode
, &map
, 0,
1054 F2FS_GET_BLOCK_READ
))
1055 goto set_error_page
;
1058 if ((map
.m_flags
& F2FS_MAP_MAPPED
)) {
1059 block_nr
= map
.m_pblk
+ block_in_file
- map
.m_lblk
;
1060 SetPageMappedToDisk(page
);
1062 if (!PageUptodate(page
) && !cleancache_get_page(page
)) {
1063 SetPageUptodate(page
);
1067 zero_user_segment(page
, 0, PAGE_SIZE
);
1068 SetPageUptodate(page
);
1074 * This page will go to BIO. Do we need to send this
1077 if (bio
&& (last_block_in_bio
!= block_nr
- 1)) {
1079 __submit_bio(F2FS_I_SB(inode
), READ
, bio
, DATA
);
1083 bio
= f2fs_grab_bio(inode
, block_nr
, nr_pages
);
1085 goto set_error_page
;
1088 if (bio_add_page(bio
, page
, blocksize
, 0) < blocksize
)
1089 goto submit_and_realloc
;
1091 last_block_in_bio
= block_nr
;
1095 zero_user_segment(page
, 0, PAGE_SIZE
);
1100 __submit_bio(F2FS_I_SB(inode
), READ
, bio
, DATA
);
1108 BUG_ON(pages
&& !list_empty(pages
));
1110 __submit_bio(F2FS_I_SB(inode
), READ
, bio
, DATA
);
1114 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
1116 struct inode
*inode
= page
->mapping
->host
;
1119 trace_f2fs_readpage(page
, DATA
);
1121 /* If the file has inline data, try to read it directly */
1122 if (f2fs_has_inline_data(inode
))
1123 ret
= f2fs_read_inline_data(inode
, page
);
1125 ret
= f2fs_mpage_readpages(page
->mapping
, NULL
, page
, 1);
1129 static int f2fs_read_data_pages(struct file
*file
,
1130 struct address_space
*mapping
,
1131 struct list_head
*pages
, unsigned nr_pages
)
1133 struct inode
*inode
= file
->f_mapping
->host
;
1134 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
1136 trace_f2fs_readpages(inode
, page
, nr_pages
);
1138 /* If the file has inline data, skip readpages */
1139 if (f2fs_has_inline_data(inode
))
1142 return f2fs_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
1145 int do_write_data_page(struct f2fs_io_info
*fio
)
1147 struct page
*page
= fio
->page
;
1148 struct inode
*inode
= page
->mapping
->host
;
1149 struct dnode_of_data dn
;
1152 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
1153 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
1157 fio
->old_blkaddr
= dn
.data_blkaddr
;
1159 /* This page is already truncated */
1160 if (fio
->old_blkaddr
== NULL_ADDR
) {
1161 ClearPageUptodate(page
);
1165 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1166 gfp_t gfp_flags
= GFP_NOFS
;
1168 /* wait for GCed encrypted page writeback */
1169 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode
),
1172 fio
->encrypted_page
= fscrypt_encrypt_page(inode
, fio
->page
,
1174 if (IS_ERR(fio
->encrypted_page
)) {
1175 err
= PTR_ERR(fio
->encrypted_page
);
1176 if (err
== -ENOMEM
) {
1177 /* flush pending ios and wait for a while */
1178 f2fs_flush_merged_bios(F2FS_I_SB(inode
));
1179 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1180 gfp_flags
|= __GFP_NOFAIL
;
1188 set_page_writeback(page
);
1191 * If current allocation needs SSR,
1192 * it had better in-place writes for updated data.
1194 if (unlikely(fio
->old_blkaddr
!= NEW_ADDR
&&
1195 !is_cold_data(page
) &&
1196 !IS_ATOMIC_WRITTEN_PAGE(page
) &&
1197 need_inplace_update(inode
))) {
1198 rewrite_data_page(fio
);
1199 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1200 trace_f2fs_do_write_data_page(page
, IPU
);
1202 write_data_page(&dn
, fio
);
1203 trace_f2fs_do_write_data_page(page
, OPU
);
1204 set_inode_flag(inode
, FI_APPEND_WRITE
);
1205 if (page
->index
== 0)
1206 set_inode_flag(inode
, FI_FIRST_BLOCK_WRITTEN
);
1209 f2fs_put_dnode(&dn
);
1213 static int f2fs_write_data_page(struct page
*page
,
1214 struct writeback_control
*wbc
)
1216 struct inode
*inode
= page
->mapping
->host
;
1217 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1218 loff_t i_size
= i_size_read(inode
);
1219 const pgoff_t end_index
= ((unsigned long long) i_size
)
1221 loff_t psize
= (page
->index
+ 1) << PAGE_SHIFT
;
1222 unsigned offset
= 0;
1223 bool need_balance_fs
= false;
1225 struct f2fs_io_info fio
= {
1228 .rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
,
1230 .encrypted_page
= NULL
,
1233 trace_f2fs_writepage(page
, DATA
);
1235 if (page
->index
< end_index
)
1239 * If the offset is out-of-range of file size,
1240 * this page does not have to be written to disk.
1242 offset
= i_size
& (PAGE_SIZE
- 1);
1243 if ((page
->index
>= end_index
+ 1) || !offset
)
1246 zero_user_segment(page
, offset
, PAGE_SIZE
);
1248 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1250 if (f2fs_is_drop_cache(inode
))
1252 /* we should not write 0'th page having journal header */
1253 if (f2fs_is_volatile_file(inode
) && (!page
->index
||
1254 (!wbc
->for_reclaim
&&
1255 available_free_memory(sbi
, BASE_CHECK
))))
1258 /* we should bypass data pages to proceed the kworkder jobs */
1259 if (unlikely(f2fs_cp_error(sbi
))) {
1260 mapping_set_error(page
->mapping
, -EIO
);
1264 /* Dentry blocks are controlled by checkpoint */
1265 if (S_ISDIR(inode
->i_mode
)) {
1266 err
= do_write_data_page(&fio
);
1270 if (!wbc
->for_reclaim
)
1271 need_balance_fs
= true;
1272 else if (has_not_enough_free_secs(sbi
, 0))
1277 if (f2fs_has_inline_data(inode
))
1278 err
= f2fs_write_inline_data(inode
, page
);
1280 err
= do_write_data_page(&fio
);
1281 if (F2FS_I(inode
)->last_disk_size
< psize
)
1282 F2FS_I(inode
)->last_disk_size
= psize
;
1283 f2fs_unlock_op(sbi
);
1285 if (err
&& err
!= -ENOENT
)
1288 clear_cold_data(page
);
1290 inode_dec_dirty_pages(inode
);
1292 ClearPageUptodate(page
);
1294 if (wbc
->for_reclaim
) {
1295 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, DATA
, WRITE
);
1296 remove_dirty_inode(inode
);
1300 f2fs_balance_fs(sbi
, need_balance_fs
);
1302 if (unlikely(f2fs_cp_error(sbi
)))
1303 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1308 redirty_page_for_writepage(wbc
, page
);
1314 * This function was copied from write_cche_pages from mm/page-writeback.c.
1315 * The major change is making write step of cold data page separately from
1316 * warm/hot data page.
1318 static int f2fs_write_cache_pages(struct address_space
*mapping
,
1319 struct writeback_control
*wbc
)
1323 struct pagevec pvec
;
1325 pgoff_t
uninitialized_var(writeback_index
);
1327 pgoff_t end
; /* Inclusive */
1330 int range_whole
= 0;
1333 pagevec_init(&pvec
, 0);
1335 if (wbc
->range_cyclic
) {
1336 writeback_index
= mapping
->writeback_index
; /* prev offset */
1337 index
= writeback_index
;
1344 index
= wbc
->range_start
>> PAGE_SHIFT
;
1345 end
= wbc
->range_end
>> PAGE_SHIFT
;
1346 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1348 cycled
= 1; /* ignore range_cyclic tests */
1350 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1351 tag
= PAGECACHE_TAG_TOWRITE
;
1353 tag
= PAGECACHE_TAG_DIRTY
;
1355 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1356 tag_pages_for_writeback(mapping
, index
, end
);
1358 while (!done
&& (index
<= end
)) {
1361 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
1362 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1);
1366 for (i
= 0; i
< nr_pages
; i
++) {
1367 struct page
*page
= pvec
.pages
[i
];
1369 if (page
->index
> end
) {
1374 done_index
= page
->index
;
1378 if (unlikely(page
->mapping
!= mapping
)) {
1384 if (!PageDirty(page
)) {
1385 /* someone wrote it for us */
1386 goto continue_unlock
;
1389 if (PageWriteback(page
)) {
1390 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1391 f2fs_wait_on_page_writeback(page
,
1394 goto continue_unlock
;
1397 BUG_ON(PageWriteback(page
));
1398 if (!clear_page_dirty_for_io(page
))
1399 goto continue_unlock
;
1401 ret
= mapping
->a_ops
->writepage(page
, wbc
);
1402 if (unlikely(ret
)) {
1403 done_index
= page
->index
+ 1;
1408 if (--wbc
->nr_to_write
<= 0 &&
1409 wbc
->sync_mode
== WB_SYNC_NONE
) {
1414 pagevec_release(&pvec
);
1418 if (!cycled
&& !done
) {
1421 end
= writeback_index
- 1;
1424 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1425 mapping
->writeback_index
= done_index
;
1430 static int f2fs_write_data_pages(struct address_space
*mapping
,
1431 struct writeback_control
*wbc
)
1433 struct inode
*inode
= mapping
->host
;
1434 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1437 /* deal with chardevs and other special file */
1438 if (!mapping
->a_ops
->writepage
)
1441 /* skip writing if there is no dirty page in this inode */
1442 if (!get_dirty_pages(inode
) && wbc
->sync_mode
== WB_SYNC_NONE
)
1445 if (S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_NONE
&&
1446 get_dirty_pages(inode
) < nr_pages_to_skip(sbi
, DATA
) &&
1447 available_free_memory(sbi
, DIRTY_DENTS
))
1450 /* skip writing during file defragment */
1451 if (is_inode_flag_set(inode
, FI_DO_DEFRAG
))
1454 /* during POR, we don't need to trigger writepage at all. */
1455 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1458 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1460 ret
= f2fs_write_cache_pages(mapping
, wbc
);
1462 * if some pages were truncated, we cannot guarantee its mapping->host
1463 * to detect pending bios.
1465 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1467 remove_dirty_inode(inode
);
1471 wbc
->pages_skipped
+= get_dirty_pages(inode
);
1472 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1476 static void f2fs_write_failed(struct address_space
*mapping
, loff_t to
)
1478 struct inode
*inode
= mapping
->host
;
1479 loff_t i_size
= i_size_read(inode
);
1482 truncate_pagecache(inode
, i_size
);
1483 truncate_blocks(inode
, i_size
, true);
1487 static int prepare_write_begin(struct f2fs_sb_info
*sbi
,
1488 struct page
*page
, loff_t pos
, unsigned len
,
1489 block_t
*blk_addr
, bool *node_changed
)
1491 struct inode
*inode
= page
->mapping
->host
;
1492 pgoff_t index
= page
->index
;
1493 struct dnode_of_data dn
;
1495 bool locked
= false;
1496 struct extent_info ei
;
1500 * we already allocated all the blocks, so we don't need to get
1501 * the block addresses when there is no need to fill the page.
1503 if (!f2fs_has_inline_data(inode
) && !f2fs_encrypted_inode(inode
) &&
1507 if (f2fs_has_inline_data(inode
) ||
1508 (pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1513 /* check inline_data */
1514 ipage
= get_node_page(sbi
, inode
->i_ino
);
1515 if (IS_ERR(ipage
)) {
1516 err
= PTR_ERR(ipage
);
1520 set_new_dnode(&dn
, inode
, ipage
, ipage
, 0);
1522 if (f2fs_has_inline_data(inode
)) {
1523 if (pos
+ len
<= MAX_INLINE_DATA
) {
1524 read_inline_data(page
, ipage
);
1525 set_inode_flag(inode
, FI_DATA_EXIST
);
1527 set_inline_node(ipage
);
1529 err
= f2fs_convert_inline_page(&dn
, page
);
1532 if (dn
.data_blkaddr
== NULL_ADDR
)
1533 err
= f2fs_get_block(&dn
, index
);
1535 } else if (locked
) {
1536 err
= f2fs_get_block(&dn
, index
);
1538 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
1539 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
1542 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
1543 if (err
|| dn
.data_blkaddr
== NULL_ADDR
) {
1544 f2fs_put_dnode(&dn
);
1552 /* convert_inline_page can make node_changed */
1553 *blk_addr
= dn
.data_blkaddr
;
1554 *node_changed
= dn
.node_changed
;
1556 f2fs_put_dnode(&dn
);
1559 f2fs_unlock_op(sbi
);
1563 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
1564 loff_t pos
, unsigned len
, unsigned flags
,
1565 struct page
**pagep
, void **fsdata
)
1567 struct inode
*inode
= mapping
->host
;
1568 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1569 struct page
*page
= NULL
;
1570 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_SHIFT
;
1571 bool need_balance
= false;
1572 block_t blkaddr
= NULL_ADDR
;
1575 trace_f2fs_write_begin(inode
, pos
, len
, flags
);
1578 * We should check this at this moment to avoid deadlock on inode page
1579 * and #0 page. The locking rule for inline_data conversion should be:
1580 * lock_page(page #0) -> lock_page(inode_page)
1583 err
= f2fs_convert_inline_inode(inode
);
1588 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1596 err
= prepare_write_begin(sbi
, page
, pos
, len
,
1597 &blkaddr
, &need_balance
);
1601 if (need_balance
&& has_not_enough_free_secs(sbi
, 0)) {
1603 f2fs_balance_fs(sbi
, true);
1605 if (page
->mapping
!= mapping
) {
1606 /* The page got truncated from under us */
1607 f2fs_put_page(page
, 1);
1612 f2fs_wait_on_page_writeback(page
, DATA
, false);
1614 /* wait for GCed encrypted page writeback */
1615 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1616 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
1618 if (len
== PAGE_SIZE
)
1620 if (PageUptodate(page
))
1623 if ((pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1624 unsigned start
= pos
& (PAGE_SIZE
- 1);
1625 unsigned end
= start
+ len
;
1627 /* Reading beyond i_size is simple: memset to zero */
1628 zero_user_segments(page
, 0, start
, end
, PAGE_SIZE
);
1632 if (blkaddr
== NEW_ADDR
) {
1633 zero_user_segment(page
, 0, PAGE_SIZE
);
1637 bio
= f2fs_grab_bio(inode
, blkaddr
, 1);
1643 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
1649 __submit_bio(sbi
, READ_SYNC
, bio
, DATA
);
1652 if (unlikely(page
->mapping
!= mapping
)) {
1653 f2fs_put_page(page
, 1);
1656 if (unlikely(!PageUptodate(page
))) {
1662 SetPageUptodate(page
);
1664 clear_cold_data(page
);
1668 f2fs_put_page(page
, 1);
1669 f2fs_write_failed(mapping
, pos
+ len
);
1673 static int f2fs_write_end(struct file
*file
,
1674 struct address_space
*mapping
,
1675 loff_t pos
, unsigned len
, unsigned copied
,
1676 struct page
*page
, void *fsdata
)
1678 struct inode
*inode
= page
->mapping
->host
;
1680 trace_f2fs_write_end(inode
, pos
, len
, copied
);
1682 set_page_dirty(page
);
1684 if (pos
+ copied
> i_size_read(inode
))
1685 f2fs_i_size_write(inode
, pos
+ copied
);
1687 f2fs_put_page(page
, 1);
1688 f2fs_update_time(F2FS_I_SB(inode
), REQ_TIME
);
1692 static int check_direct_IO(struct inode
*inode
, struct iov_iter
*iter
,
1695 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
1697 if (offset
& blocksize_mask
)
1700 if (iov_iter_alignment(iter
) & blocksize_mask
)
1706 static ssize_t
f2fs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
1708 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
1709 struct inode
*inode
= mapping
->host
;
1710 size_t count
= iov_iter_count(iter
);
1711 loff_t offset
= iocb
->ki_pos
;
1714 err
= check_direct_IO(inode
, iter
, offset
);
1718 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1720 if (test_opt(F2FS_I_SB(inode
), LFS
))
1723 trace_f2fs_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
1725 err
= blockdev_direct_IO(iocb
, inode
, iter
, get_data_block_dio
);
1726 if (iov_iter_rw(iter
) == WRITE
) {
1728 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1730 f2fs_write_failed(mapping
, offset
+ count
);
1733 trace_f2fs_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), err
);
1738 void f2fs_invalidate_page(struct page
*page
, unsigned int offset
,
1739 unsigned int length
)
1741 struct inode
*inode
= page
->mapping
->host
;
1742 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1744 if (inode
->i_ino
>= F2FS_ROOT_INO(sbi
) &&
1745 (offset
% PAGE_SIZE
|| length
!= PAGE_SIZE
))
1748 if (PageDirty(page
)) {
1749 if (inode
->i_ino
== F2FS_META_INO(sbi
))
1750 dec_page_count(sbi
, F2FS_DIRTY_META
);
1751 else if (inode
->i_ino
== F2FS_NODE_INO(sbi
))
1752 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1754 inode_dec_dirty_pages(inode
);
1757 /* This is atomic written page, keep Private */
1758 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1761 set_page_private(page
, 0);
1762 ClearPagePrivate(page
);
1765 int f2fs_release_page(struct page
*page
, gfp_t wait
)
1767 /* If this is dirty page, keep PagePrivate */
1768 if (PageDirty(page
))
1771 /* This is atomic written page, keep Private */
1772 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1775 set_page_private(page
, 0);
1776 ClearPagePrivate(page
);
1781 * This was copied from __set_page_dirty_buffers which gives higher performance
1782 * in very high speed storages. (e.g., pmem)
1784 void f2fs_set_page_dirty_nobuffers(struct page
*page
)
1786 struct address_space
*mapping
= page
->mapping
;
1787 unsigned long flags
;
1789 if (unlikely(!mapping
))
1792 spin_lock(&mapping
->private_lock
);
1793 lock_page_memcg(page
);
1795 spin_unlock(&mapping
->private_lock
);
1797 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
1798 WARN_ON_ONCE(!PageUptodate(page
));
1799 account_page_dirtied(page
, mapping
);
1800 radix_tree_tag_set(&mapping
->page_tree
,
1801 page_index(page
), PAGECACHE_TAG_DIRTY
);
1802 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
1803 unlock_page_memcg(page
);
1805 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1809 static int f2fs_set_data_page_dirty(struct page
*page
)
1811 struct address_space
*mapping
= page
->mapping
;
1812 struct inode
*inode
= mapping
->host
;
1814 trace_f2fs_set_page_dirty(page
, DATA
);
1816 SetPageUptodate(page
);
1818 if (f2fs_is_atomic_file(inode
)) {
1819 if (!IS_ATOMIC_WRITTEN_PAGE(page
)) {
1820 register_inmem_page(inode
, page
);
1824 * Previously, this page has been registered, we just
1830 if (!PageDirty(page
)) {
1831 f2fs_set_page_dirty_nobuffers(page
);
1832 update_dirty_page(inode
, page
);
1838 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
1840 struct inode
*inode
= mapping
->host
;
1842 if (f2fs_has_inline_data(inode
))
1845 /* make sure allocating whole blocks */
1846 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
1847 filemap_write_and_wait(mapping
);
1849 return generic_block_bmap(mapping
, block
, get_data_block_bmap
);
1852 const struct address_space_operations f2fs_dblock_aops
= {
1853 .readpage
= f2fs_read_data_page
,
1854 .readpages
= f2fs_read_data_pages
,
1855 .writepage
= f2fs_write_data_page
,
1856 .writepages
= f2fs_write_data_pages
,
1857 .write_begin
= f2fs_write_begin
,
1858 .write_end
= f2fs_write_end
,
1859 .set_page_dirty
= f2fs_set_data_page_dirty
,
1860 .invalidatepage
= f2fs_invalidate_page
,
1861 .releasepage
= f2fs_release_page
,
1862 .direct_IO
= f2fs_direct_IO
,