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>
22 #include <linux/cleancache.h>
28 #include <trace/events/f2fs.h>
30 static void f2fs_read_end_io(struct bio
*bio
)
35 if (f2fs_bio_encrypted(bio
)) {
37 fscrypt_release_ctx(bio
->bi_private
);
39 fscrypt_decrypt_bio_pages(bio
->bi_private
, bio
);
44 bio_for_each_segment_all(bvec
, bio
, i
) {
45 struct page
*page
= bvec
->bv_page
;
48 SetPageUptodate(page
);
50 ClearPageUptodate(page
);
58 static void f2fs_write_end_io(struct bio
*bio
)
60 struct f2fs_sb_info
*sbi
= bio
->bi_private
;
64 bio_for_each_segment_all(bvec
, bio
, i
) {
65 struct page
*page
= bvec
->bv_page
;
67 fscrypt_pullback_bio_page(&page
, true);
69 if (unlikely(bio
->bi_error
)) {
70 set_bit(AS_EIO
, &page
->mapping
->flags
);
71 f2fs_stop_checkpoint(sbi
, true);
73 end_page_writeback(page
);
75 if (atomic_dec_and_test(&sbi
->nr_wb_bios
) &&
76 wq_has_sleeper(&sbi
->cp_wait
))
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 bio
= f2fs_bio_alloc(npages
);
92 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
93 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blk_addr
);
94 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
95 bio
->bi_private
= is_read
? NULL
: sbi
;
100 static inline void __submit_bio(struct f2fs_sb_info
*sbi
, struct bio
*bio
)
102 if (!is_read_io(bio_op(bio
)))
103 atomic_inc(&sbi
->nr_wb_bios
);
107 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
109 struct f2fs_io_info
*fio
= &io
->fio
;
114 if (is_read_io(fio
->op
))
115 trace_f2fs_submit_read_bio(io
->sbi
->sb
, fio
, io
->bio
);
117 trace_f2fs_submit_write_bio(io
->sbi
->sb
, fio
, io
->bio
);
119 bio_set_op_attrs(io
->bio
, fio
->op
, fio
->op_flags
);
121 __submit_bio(io
->sbi
, io
->bio
);
125 static bool __has_merged_page(struct f2fs_bio_info
*io
, struct inode
*inode
,
126 struct page
*page
, nid_t ino
)
128 struct bio_vec
*bvec
;
135 if (!inode
&& !page
&& !ino
)
138 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
140 if (bvec
->bv_page
->mapping
)
141 target
= bvec
->bv_page
;
143 target
= fscrypt_control_page(bvec
->bv_page
);
145 if (inode
&& inode
== target
->mapping
->host
)
147 if (page
&& page
== target
)
149 if (ino
&& ino
== ino_of_node(target
))
156 static bool has_merged_page(struct f2fs_sb_info
*sbi
, struct inode
*inode
,
157 struct page
*page
, nid_t ino
,
160 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
161 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
164 down_read(&io
->io_rwsem
);
165 ret
= __has_merged_page(io
, inode
, page
, ino
);
166 up_read(&io
->io_rwsem
);
170 static void __f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
171 struct inode
*inode
, struct page
*page
,
172 nid_t ino
, enum page_type type
, int rw
)
174 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
175 struct f2fs_bio_info
*io
;
177 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
179 down_write(&io
->io_rwsem
);
181 if (!__has_merged_page(io
, inode
, page
, ino
))
184 /* change META to META_FLUSH in the checkpoint procedure */
185 if (type
>= META_FLUSH
) {
186 io
->fio
.type
= META_FLUSH
;
187 io
->fio
.op
= REQ_OP_WRITE
;
188 if (test_opt(sbi
, NOBARRIER
))
189 io
->fio
.op_flags
= WRITE_FLUSH
| REQ_META
| REQ_PRIO
;
191 io
->fio
.op_flags
= WRITE_FLUSH_FUA
| REQ_META
|
194 __submit_merged_bio(io
);
196 up_write(&io
->io_rwsem
);
199 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
, enum page_type type
,
202 __f2fs_submit_merged_bio(sbi
, NULL
, NULL
, 0, type
, rw
);
205 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info
*sbi
,
206 struct inode
*inode
, struct page
*page
,
207 nid_t ino
, enum page_type type
, int rw
)
209 if (has_merged_page(sbi
, inode
, page
, ino
, type
))
210 __f2fs_submit_merged_bio(sbi
, inode
, page
, ino
, type
, rw
);
213 void f2fs_flush_merged_bios(struct f2fs_sb_info
*sbi
)
215 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
216 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
217 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
221 * Fill the locked page with data located in the block address.
222 * Return unlocked page.
224 int f2fs_submit_page_bio(struct f2fs_io_info
*fio
)
227 struct page
*page
= fio
->encrypted_page
?
228 fio
->encrypted_page
: fio
->page
;
230 trace_f2fs_submit_page_bio(page
, fio
);
231 f2fs_trace_ios(fio
, 0);
233 /* Allocate a new bio */
234 bio
= __bio_alloc(fio
->sbi
, fio
->new_blkaddr
, 1, is_read_io(fio
->op
));
236 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
240 bio
->bi_rw
= fio
->op_flags
;
241 bio_set_op_attrs(bio
, fio
->op
, fio
->op_flags
);
243 __submit_bio(fio
->sbi
, bio
);
247 void f2fs_submit_page_mbio(struct f2fs_io_info
*fio
)
249 struct f2fs_sb_info
*sbi
= fio
->sbi
;
250 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
251 struct f2fs_bio_info
*io
;
252 bool is_read
= is_read_io(fio
->op
);
253 struct page
*bio_page
;
255 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
257 if (fio
->old_blkaddr
!= NEW_ADDR
)
258 verify_block_addr(sbi
, fio
->old_blkaddr
);
259 verify_block_addr(sbi
, fio
->new_blkaddr
);
261 down_write(&io
->io_rwsem
);
263 if (io
->bio
&& (io
->last_block_in_bio
!= fio
->new_blkaddr
- 1 ||
264 (io
->fio
.op
!= fio
->op
|| io
->fio
.op_flags
!= fio
->op_flags
)))
265 __submit_merged_bio(io
);
267 if (io
->bio
== NULL
) {
268 int bio_blocks
= MAX_BIO_BLOCKS(sbi
);
270 io
->bio
= __bio_alloc(sbi
, fio
->new_blkaddr
,
271 bio_blocks
, is_read
);
275 bio_page
= fio
->encrypted_page
? fio
->encrypted_page
: fio
->page
;
277 if (bio_add_page(io
->bio
, bio_page
, PAGE_SIZE
, 0) <
279 __submit_merged_bio(io
);
283 io
->last_block_in_bio
= fio
->new_blkaddr
;
284 f2fs_trace_ios(fio
, 0);
286 up_write(&io
->io_rwsem
);
287 trace_f2fs_submit_page_mbio(fio
->page
, fio
);
290 static void __set_data_blkaddr(struct dnode_of_data
*dn
)
292 struct f2fs_node
*rn
= F2FS_NODE(dn
->node_page
);
295 /* Get physical address of data block */
296 addr_array
= blkaddr_in_node(rn
);
297 addr_array
[dn
->ofs_in_node
] = cpu_to_le32(dn
->data_blkaddr
);
301 * Lock ordering for the change of data block address:
304 * update block addresses in the node page
306 void set_data_blkaddr(struct dnode_of_data
*dn
)
308 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
309 __set_data_blkaddr(dn
);
310 if (set_page_dirty(dn
->node_page
))
311 dn
->node_changed
= true;
314 void f2fs_update_data_blkaddr(struct dnode_of_data
*dn
, block_t blkaddr
)
316 dn
->data_blkaddr
= blkaddr
;
317 set_data_blkaddr(dn
);
318 f2fs_update_extent_cache(dn
);
321 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
322 int reserve_new_blocks(struct dnode_of_data
*dn
, blkcnt_t count
)
324 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
329 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
331 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
334 trace_f2fs_reserve_new_blocks(dn
->inode
, dn
->nid
,
335 dn
->ofs_in_node
, count
);
337 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
339 for (; count
> 0; dn
->ofs_in_node
++) {
341 datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
342 if (blkaddr
== NULL_ADDR
) {
343 dn
->data_blkaddr
= NEW_ADDR
;
344 __set_data_blkaddr(dn
);
349 if (set_page_dirty(dn
->node_page
))
350 dn
->node_changed
= true;
352 mark_inode_dirty(dn
->inode
);
357 /* Should keep dn->ofs_in_node unchanged */
358 int reserve_new_block(struct dnode_of_data
*dn
)
360 unsigned int ofs_in_node
= dn
->ofs_in_node
;
363 ret
= reserve_new_blocks(dn
, 1);
364 dn
->ofs_in_node
= ofs_in_node
;
368 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
370 bool need_put
= dn
->inode_page
? false : true;
373 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
377 if (dn
->data_blkaddr
== NULL_ADDR
)
378 err
= reserve_new_block(dn
);
384 int f2fs_get_block(struct dnode_of_data
*dn
, pgoff_t index
)
386 struct extent_info ei
;
387 struct inode
*inode
= dn
->inode
;
389 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
390 dn
->data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
394 return f2fs_reserve_block(dn
, index
);
397 struct page
*get_read_data_page(struct inode
*inode
, pgoff_t index
,
398 int op_flags
, bool for_write
)
400 struct address_space
*mapping
= inode
->i_mapping
;
401 struct dnode_of_data dn
;
403 struct extent_info ei
;
405 struct f2fs_io_info fio
= {
406 .sbi
= F2FS_I_SB(inode
),
409 .op_flags
= op_flags
,
410 .encrypted_page
= NULL
,
413 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
414 return read_mapping_page(mapping
, index
, NULL
);
416 page
= f2fs_grab_cache_page(mapping
, index
, for_write
);
418 return ERR_PTR(-ENOMEM
);
420 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
421 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
425 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
426 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
431 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
436 if (PageUptodate(page
)) {
442 * A new dentry page is allocated but not able to be written, since its
443 * new inode page couldn't be allocated due to -ENOSPC.
444 * In such the case, its blkaddr can be remained as NEW_ADDR.
445 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
447 if (dn
.data_blkaddr
== NEW_ADDR
) {
448 zero_user_segment(page
, 0, PAGE_SIZE
);
449 SetPageUptodate(page
);
454 fio
.new_blkaddr
= fio
.old_blkaddr
= dn
.data_blkaddr
;
456 err
= f2fs_submit_page_bio(&fio
);
462 f2fs_put_page(page
, 1);
466 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
)
468 struct address_space
*mapping
= inode
->i_mapping
;
471 page
= find_get_page(mapping
, index
);
472 if (page
&& PageUptodate(page
))
474 f2fs_put_page(page
, 0);
476 page
= get_read_data_page(inode
, index
, READ_SYNC
, false);
480 if (PageUptodate(page
))
483 wait_on_page_locked(page
);
484 if (unlikely(!PageUptodate(page
))) {
485 f2fs_put_page(page
, 0);
486 return ERR_PTR(-EIO
);
492 * If it tries to access a hole, return an error.
493 * Because, the callers, functions in dir.c and GC, should be able to know
494 * whether this page exists or not.
496 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
,
499 struct address_space
*mapping
= inode
->i_mapping
;
502 page
= get_read_data_page(inode
, index
, READ_SYNC
, for_write
);
506 /* wait for read completion */
508 if (unlikely(!PageUptodate(page
))) {
509 f2fs_put_page(page
, 1);
510 return ERR_PTR(-EIO
);
512 if (unlikely(page
->mapping
!= mapping
)) {
513 f2fs_put_page(page
, 1);
520 * Caller ensures that this data page is never allocated.
521 * A new zero-filled data page is allocated in the page cache.
523 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
525 * Note that, ipage is set only by make_empty_dir, and if any error occur,
526 * ipage should be released by this function.
528 struct page
*get_new_data_page(struct inode
*inode
,
529 struct page
*ipage
, pgoff_t index
, bool new_i_size
)
531 struct address_space
*mapping
= inode
->i_mapping
;
533 struct dnode_of_data dn
;
536 page
= f2fs_grab_cache_page(mapping
, index
, true);
539 * before exiting, we should make sure ipage will be released
540 * if any error occur.
542 f2fs_put_page(ipage
, 1);
543 return ERR_PTR(-ENOMEM
);
546 set_new_dnode(&dn
, inode
, ipage
, NULL
, 0);
547 err
= f2fs_reserve_block(&dn
, index
);
549 f2fs_put_page(page
, 1);
555 if (PageUptodate(page
))
558 if (dn
.data_blkaddr
== NEW_ADDR
) {
559 zero_user_segment(page
, 0, PAGE_SIZE
);
560 SetPageUptodate(page
);
562 f2fs_put_page(page
, 1);
564 /* if ipage exists, blkaddr should be NEW_ADDR */
565 f2fs_bug_on(F2FS_I_SB(inode
), ipage
);
566 page
= get_lock_data_page(inode
, index
, true);
571 if (new_i_size
&& i_size_read(inode
) <
572 ((loff_t
)(index
+ 1) << PAGE_SHIFT
)) {
573 i_size_write(inode
, ((loff_t
)(index
+ 1) << PAGE_SHIFT
));
574 /* Only the directory inode sets new_i_size */
575 set_inode_flag(F2FS_I(inode
), FI_UPDATE_DIR
);
580 static int __allocate_data_block(struct dnode_of_data
*dn
)
582 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
583 struct f2fs_summary sum
;
585 int seg
= CURSEG_WARM_DATA
;
589 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
592 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
593 if (dn
->data_blkaddr
== NEW_ADDR
)
596 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
600 get_node_info(sbi
, dn
->nid
, &ni
);
601 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
603 if (dn
->ofs_in_node
== 0 && dn
->inode_page
== dn
->node_page
)
604 seg
= CURSEG_DIRECT_IO
;
606 allocate_data_block(sbi
, NULL
, dn
->data_blkaddr
, &dn
->data_blkaddr
,
608 set_data_blkaddr(dn
);
611 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), dn
->inode
) +
613 if (i_size_read(dn
->inode
) < ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
))
614 i_size_write(dn
->inode
,
615 ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
));
619 ssize_t
f2fs_preallocate_blocks(struct kiocb
*iocb
, struct iov_iter
*from
)
621 struct inode
*inode
= file_inode(iocb
->ki_filp
);
622 struct f2fs_map_blocks map
;
625 map
.m_lblk
= F2FS_BLK_ALIGN(iocb
->ki_pos
);
626 map
.m_len
= F2FS_BYTES_TO_BLK(iov_iter_count(from
));
627 map
.m_next_pgofs
= NULL
;
629 if (f2fs_encrypted_inode(inode
))
632 if (iocb
->ki_flags
& IOCB_DIRECT
) {
633 ret
= f2fs_convert_inline_inode(inode
);
636 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_DIO
);
638 if (iocb
->ki_pos
+ iov_iter_count(from
) > MAX_INLINE_DATA
) {
639 ret
= f2fs_convert_inline_inode(inode
);
643 if (!f2fs_has_inline_data(inode
))
644 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_AIO
);
649 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
650 * f2fs_map_blocks structure.
651 * If original data blocks are allocated, then give them to blockdev.
653 * a. preallocate requested block addresses
654 * b. do not use extent cache for better performance
655 * c. give the block addresses to blockdev
657 int f2fs_map_blocks(struct inode
*inode
, struct f2fs_map_blocks
*map
,
658 int create
, int flag
)
660 unsigned int maxblocks
= map
->m_len
;
661 struct dnode_of_data dn
;
662 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
663 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE_RA
;
664 pgoff_t pgofs
, end_offset
, end
;
665 int err
= 0, ofs
= 1;
666 unsigned int ofs_in_node
, last_ofs_in_node
;
668 struct extent_info ei
;
669 bool allocated
= false;
675 /* it only supports block size == page size */
676 pgofs
= (pgoff_t
)map
->m_lblk
;
677 end
= pgofs
+ maxblocks
;
679 if (!create
&& f2fs_lookup_extent_cache(inode
, pgofs
, &ei
)) {
680 map
->m_pblk
= ei
.blk
+ pgofs
- ei
.fofs
;
681 map
->m_len
= min((pgoff_t
)maxblocks
, ei
.fofs
+ ei
.len
- pgofs
);
682 map
->m_flags
= F2FS_MAP_MAPPED
;
690 /* When reading holes, we need its node page */
691 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
692 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
694 if (flag
== F2FS_GET_BLOCK_BMAP
)
696 if (err
== -ENOENT
) {
698 if (map
->m_next_pgofs
)
700 get_next_page_offset(&dn
, pgofs
);
706 ofs_in_node
= dn
.ofs_in_node
;
707 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, inode
);
710 blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
712 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
) {
714 if (unlikely(f2fs_cp_error(sbi
))) {
718 if (flag
== F2FS_GET_BLOCK_PRE_AIO
) {
719 if (blkaddr
== NULL_ADDR
) {
721 last_ofs_in_node
= dn
.ofs_in_node
;
724 err
= __allocate_data_block(&dn
);
726 set_inode_flag(F2FS_I(inode
),
733 map
->m_flags
= F2FS_MAP_NEW
;
734 blkaddr
= dn
.data_blkaddr
;
736 if (flag
== F2FS_GET_BLOCK_BMAP
) {
740 if (flag
== F2FS_GET_BLOCK_FIEMAP
&&
741 blkaddr
== NULL_ADDR
) {
742 if (map
->m_next_pgofs
)
743 *map
->m_next_pgofs
= pgofs
+ 1;
745 if (flag
!= F2FS_GET_BLOCK_FIEMAP
||
751 if (flag
== F2FS_GET_BLOCK_PRE_AIO
)
754 if (map
->m_len
== 0) {
755 /* preallocated unwritten block should be mapped for fiemap. */
756 if (blkaddr
== NEW_ADDR
)
757 map
->m_flags
|= F2FS_MAP_UNWRITTEN
;
758 map
->m_flags
|= F2FS_MAP_MAPPED
;
760 map
->m_pblk
= blkaddr
;
762 } else if ((map
->m_pblk
!= NEW_ADDR
&&
763 blkaddr
== (map
->m_pblk
+ ofs
)) ||
764 (map
->m_pblk
== NEW_ADDR
&& blkaddr
== NEW_ADDR
) ||
765 flag
== F2FS_GET_BLOCK_PRE_DIO
) {
776 /* preallocate blocks in batch for one dnode page */
777 if (flag
== F2FS_GET_BLOCK_PRE_AIO
&&
778 (pgofs
== end
|| dn
.ofs_in_node
== end_offset
)) {
780 dn
.ofs_in_node
= ofs_in_node
;
781 err
= reserve_new_blocks(&dn
, prealloc
);
785 map
->m_len
+= dn
.ofs_in_node
- ofs_in_node
;
786 if (prealloc
&& dn
.ofs_in_node
!= last_ofs_in_node
+ 1) {
790 dn
.ofs_in_node
= end_offset
;
795 else if (dn
.ofs_in_node
< end_offset
)
799 sync_inode_page(&dn
);
804 f2fs_balance_fs(sbi
, allocated
);
811 sync_inode_page(&dn
);
816 f2fs_balance_fs(sbi
, allocated
);
819 trace_f2fs_map_blocks(inode
, map
, err
);
823 static int __get_data_block(struct inode
*inode
, sector_t iblock
,
824 struct buffer_head
*bh
, int create
, int flag
,
827 struct f2fs_map_blocks map
;
831 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
832 map
.m_next_pgofs
= next_pgofs
;
834 ret
= f2fs_map_blocks(inode
, &map
, create
, flag
);
836 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
837 bh
->b_state
= (bh
->b_state
& ~F2FS_MAP_FLAGS
) | map
.m_flags
;
838 bh
->b_size
= map
.m_len
<< inode
->i_blkbits
;
843 static int get_data_block(struct inode
*inode
, sector_t iblock
,
844 struct buffer_head
*bh_result
, int create
, int flag
,
847 return __get_data_block(inode
, iblock
, bh_result
, create
,
851 static int get_data_block_dio(struct inode
*inode
, sector_t iblock
,
852 struct buffer_head
*bh_result
, int create
)
854 return __get_data_block(inode
, iblock
, bh_result
, create
,
855 F2FS_GET_BLOCK_DIO
, NULL
);
858 static int get_data_block_bmap(struct inode
*inode
, sector_t iblock
,
859 struct buffer_head
*bh_result
, int create
)
861 /* Block number less than F2FS MAX BLOCKS */
862 if (unlikely(iblock
>= F2FS_I_SB(inode
)->max_file_blocks
))
865 return __get_data_block(inode
, iblock
, bh_result
, create
,
866 F2FS_GET_BLOCK_BMAP
, NULL
);
869 static inline sector_t
logical_to_blk(struct inode
*inode
, loff_t offset
)
871 return (offset
>> inode
->i_blkbits
);
874 static inline loff_t
blk_to_logical(struct inode
*inode
, sector_t blk
)
876 return (blk
<< inode
->i_blkbits
);
879 int f2fs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
882 struct buffer_head map_bh
;
883 sector_t start_blk
, last_blk
;
886 u64 logical
= 0, phys
= 0, size
= 0;
890 ret
= fiemap_check_flags(fieinfo
, FIEMAP_FLAG_SYNC
);
894 if (f2fs_has_inline_data(inode
)) {
895 ret
= f2fs_inline_data_fiemap(inode
, fieinfo
, start
, len
);
902 isize
= i_size_read(inode
);
906 if (start
+ len
> isize
)
909 if (logical_to_blk(inode
, len
) == 0)
910 len
= blk_to_logical(inode
, 1);
912 start_blk
= logical_to_blk(inode
, start
);
913 last_blk
= logical_to_blk(inode
, start
+ len
- 1);
916 memset(&map_bh
, 0, sizeof(struct buffer_head
));
919 ret
= get_data_block(inode
, start_blk
, &map_bh
, 0,
920 F2FS_GET_BLOCK_FIEMAP
, &next_pgofs
);
925 if (!buffer_mapped(&map_bh
)) {
926 start_blk
= next_pgofs
;
927 /* Go through holes util pass the EOF */
928 if (blk_to_logical(inode
, start_blk
) < isize
)
930 /* Found a hole beyond isize means no more extents.
931 * Note that the premise is that filesystems don't
932 * punch holes beyond isize and keep size unchanged.
934 flags
|= FIEMAP_EXTENT_LAST
;
938 if (f2fs_encrypted_inode(inode
))
939 flags
|= FIEMAP_EXTENT_DATA_ENCRYPTED
;
941 ret
= fiemap_fill_next_extent(fieinfo
, logical
,
945 if (start_blk
> last_blk
|| ret
)
948 logical
= blk_to_logical(inode
, start_blk
);
949 phys
= blk_to_logical(inode
, map_bh
.b_blocknr
);
950 size
= map_bh
.b_size
;
952 if (buffer_unwritten(&map_bh
))
953 flags
= FIEMAP_EXTENT_UNWRITTEN
;
955 start_blk
+= logical_to_blk(inode
, size
);
959 if (fatal_signal_pending(current
))
972 * This function was originally taken from fs/mpage.c, and customized for f2fs.
973 * Major change was from block_size == page_size in f2fs by default.
975 static int f2fs_mpage_readpages(struct address_space
*mapping
,
976 struct list_head
*pages
, struct page
*page
,
979 struct bio
*bio
= NULL
;
981 sector_t last_block_in_bio
= 0;
982 struct inode
*inode
= mapping
->host
;
983 const unsigned blkbits
= inode
->i_blkbits
;
984 const unsigned blocksize
= 1 << blkbits
;
985 sector_t block_in_file
;
987 sector_t last_block_in_file
;
989 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
990 struct f2fs_map_blocks map
;
996 map
.m_next_pgofs
= NULL
;
998 for (page_idx
= 0; nr_pages
; page_idx
++, nr_pages
--) {
1000 prefetchw(&page
->flags
);
1002 page
= list_entry(pages
->prev
, struct page
, lru
);
1003 list_del(&page
->lru
);
1004 if (add_to_page_cache_lru(page
, mapping
,
1006 readahead_gfp_mask(mapping
)))
1010 block_in_file
= (sector_t
)page
->index
;
1011 last_block
= block_in_file
+ nr_pages
;
1012 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >>
1014 if (last_block
> last_block_in_file
)
1015 last_block
= last_block_in_file
;
1018 * Map blocks using the previous result first.
1020 if ((map
.m_flags
& F2FS_MAP_MAPPED
) &&
1021 block_in_file
> map
.m_lblk
&&
1022 block_in_file
< (map
.m_lblk
+ map
.m_len
))
1026 * Then do more f2fs_map_blocks() calls until we are
1027 * done with this page.
1031 if (block_in_file
< last_block
) {
1032 map
.m_lblk
= block_in_file
;
1033 map
.m_len
= last_block
- block_in_file
;
1035 if (f2fs_map_blocks(inode
, &map
, 0,
1036 F2FS_GET_BLOCK_READ
))
1037 goto set_error_page
;
1040 if ((map
.m_flags
& F2FS_MAP_MAPPED
)) {
1041 block_nr
= map
.m_pblk
+ block_in_file
- map
.m_lblk
;
1042 SetPageMappedToDisk(page
);
1044 if (!PageUptodate(page
) && !cleancache_get_page(page
)) {
1045 SetPageUptodate(page
);
1049 zero_user_segment(page
, 0, PAGE_SIZE
);
1050 SetPageUptodate(page
);
1056 * This page will go to BIO. Do we need to send this
1059 if (bio
&& (last_block_in_bio
!= block_nr
- 1)) {
1061 __submit_bio(F2FS_I_SB(inode
), bio
);
1065 struct fscrypt_ctx
*ctx
= NULL
;
1067 if (f2fs_encrypted_inode(inode
) &&
1068 S_ISREG(inode
->i_mode
)) {
1070 ctx
= fscrypt_get_ctx(inode
, GFP_NOFS
);
1072 goto set_error_page
;
1074 /* wait the page to be moved by cleaning */
1075 f2fs_wait_on_encrypted_page_writeback(
1076 F2FS_I_SB(inode
), block_nr
);
1079 bio
= bio_alloc(GFP_KERNEL
,
1080 min_t(int, nr_pages
, BIO_MAX_PAGES
));
1083 fscrypt_release_ctx(ctx
);
1084 goto set_error_page
;
1086 bio
->bi_bdev
= bdev
;
1087 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(block_nr
);
1088 bio
->bi_end_io
= f2fs_read_end_io
;
1089 bio
->bi_private
= ctx
;
1090 bio_set_op_attrs(bio
, REQ_OP_READ
, 0);
1093 if (bio_add_page(bio
, page
, blocksize
, 0) < blocksize
)
1094 goto submit_and_realloc
;
1096 last_block_in_bio
= block_nr
;
1100 zero_user_segment(page
, 0, PAGE_SIZE
);
1105 __submit_bio(F2FS_I_SB(inode
), bio
);
1113 BUG_ON(pages
&& !list_empty(pages
));
1115 __submit_bio(F2FS_I_SB(inode
), bio
);
1119 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
1121 struct inode
*inode
= page
->mapping
->host
;
1124 trace_f2fs_readpage(page
, DATA
);
1126 /* If the file has inline data, try to read it directly */
1127 if (f2fs_has_inline_data(inode
))
1128 ret
= f2fs_read_inline_data(inode
, page
);
1130 ret
= f2fs_mpage_readpages(page
->mapping
, NULL
, page
, 1);
1134 static int f2fs_read_data_pages(struct file
*file
,
1135 struct address_space
*mapping
,
1136 struct list_head
*pages
, unsigned nr_pages
)
1138 struct inode
*inode
= file
->f_mapping
->host
;
1139 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
1141 trace_f2fs_readpages(inode
, page
, nr_pages
);
1143 /* If the file has inline data, skip readpages */
1144 if (f2fs_has_inline_data(inode
))
1147 return f2fs_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
1150 int do_write_data_page(struct f2fs_io_info
*fio
)
1152 struct page
*page
= fio
->page
;
1153 struct inode
*inode
= page
->mapping
->host
;
1154 struct dnode_of_data dn
;
1157 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
1158 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
1162 fio
->old_blkaddr
= dn
.data_blkaddr
;
1164 /* This page is already truncated */
1165 if (fio
->old_blkaddr
== NULL_ADDR
) {
1166 ClearPageUptodate(page
);
1170 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1171 gfp_t gfp_flags
= GFP_NOFS
;
1173 /* wait for GCed encrypted page writeback */
1174 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode
),
1177 fio
->encrypted_page
= fscrypt_encrypt_page(inode
, fio
->page
,
1179 if (IS_ERR(fio
->encrypted_page
)) {
1180 err
= PTR_ERR(fio
->encrypted_page
);
1181 if (err
== -ENOMEM
) {
1182 /* flush pending ios and wait for a while */
1183 f2fs_flush_merged_bios(F2FS_I_SB(inode
));
1184 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1185 gfp_flags
|= __GFP_NOFAIL
;
1193 set_page_writeback(page
);
1196 * If current allocation needs SSR,
1197 * it had better in-place writes for updated data.
1199 if (unlikely(fio
->old_blkaddr
!= NEW_ADDR
&&
1200 !is_cold_data(page
) &&
1201 !IS_ATOMIC_WRITTEN_PAGE(page
) &&
1202 need_inplace_update(inode
))) {
1203 rewrite_data_page(fio
);
1204 set_inode_flag(F2FS_I(inode
), FI_UPDATE_WRITE
);
1205 trace_f2fs_do_write_data_page(page
, IPU
);
1207 write_data_page(&dn
, fio
);
1208 trace_f2fs_do_write_data_page(page
, OPU
);
1209 set_inode_flag(F2FS_I(inode
), FI_APPEND_WRITE
);
1210 if (page
->index
== 0)
1211 set_inode_flag(F2FS_I(inode
), FI_FIRST_BLOCK_WRITTEN
);
1214 f2fs_put_dnode(&dn
);
1218 static int f2fs_write_data_page(struct page
*page
,
1219 struct writeback_control
*wbc
)
1221 struct inode
*inode
= page
->mapping
->host
;
1222 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1223 loff_t i_size
= i_size_read(inode
);
1224 const pgoff_t end_index
= ((unsigned long long) i_size
)
1226 unsigned offset
= 0;
1227 bool need_balance_fs
= false;
1229 struct f2fs_io_info fio
= {
1233 .op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: 0,
1235 .encrypted_page
= NULL
,
1238 trace_f2fs_writepage(page
, DATA
);
1240 if (page
->index
< end_index
)
1244 * If the offset is out-of-range of file size,
1245 * this page does not have to be written to disk.
1247 offset
= i_size
& (PAGE_SIZE
- 1);
1248 if ((page
->index
>= end_index
+ 1) || !offset
)
1251 zero_user_segment(page
, offset
, PAGE_SIZE
);
1253 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1255 if (f2fs_is_drop_cache(inode
))
1257 /* we should not write 0'th page having journal header */
1258 if (f2fs_is_volatile_file(inode
) && (!page
->index
||
1259 (!wbc
->for_reclaim
&&
1260 available_free_memory(sbi
, BASE_CHECK
))))
1263 /* Dentry blocks are controlled by checkpoint */
1264 if (S_ISDIR(inode
->i_mode
)) {
1265 if (unlikely(f2fs_cp_error(sbi
)))
1267 err
= do_write_data_page(&fio
);
1271 /* we should bypass data pages to proceed the kworkder jobs */
1272 if (unlikely(f2fs_cp_error(sbi
))) {
1277 if (!wbc
->for_reclaim
)
1278 need_balance_fs
= true;
1279 else if (has_not_enough_free_secs(sbi
, 0))
1284 if (f2fs_has_inline_data(inode
))
1285 err
= f2fs_write_inline_data(inode
, page
);
1287 err
= do_write_data_page(&fio
);
1288 f2fs_unlock_op(sbi
);
1290 if (err
&& err
!= -ENOENT
)
1293 clear_cold_data(page
);
1295 inode_dec_dirty_pages(inode
);
1297 ClearPageUptodate(page
);
1299 if (wbc
->for_reclaim
) {
1300 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, DATA
, WRITE
);
1301 remove_dirty_inode(inode
);
1305 f2fs_balance_fs(sbi
, need_balance_fs
);
1307 if (unlikely(f2fs_cp_error(sbi
)))
1308 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1313 redirty_page_for_writepage(wbc
, page
);
1314 return AOP_WRITEPAGE_ACTIVATE
;
1317 static int __f2fs_writepage(struct page
*page
, struct writeback_control
*wbc
,
1320 struct address_space
*mapping
= data
;
1321 int ret
= mapping
->a_ops
->writepage(page
, wbc
);
1322 mapping_set_error(mapping
, ret
);
1327 * This function was copied from write_cche_pages from mm/page-writeback.c.
1328 * The major change is making write step of cold data page separately from
1329 * warm/hot data page.
1331 static int f2fs_write_cache_pages(struct address_space
*mapping
,
1332 struct writeback_control
*wbc
, writepage_t writepage
,
1337 struct pagevec pvec
;
1339 pgoff_t
uninitialized_var(writeback_index
);
1341 pgoff_t end
; /* Inclusive */
1344 int range_whole
= 0;
1348 pagevec_init(&pvec
, 0);
1350 if (wbc
->range_cyclic
) {
1351 writeback_index
= mapping
->writeback_index
; /* prev offset */
1352 index
= writeback_index
;
1359 index
= wbc
->range_start
>> PAGE_SHIFT
;
1360 end
= wbc
->range_end
>> PAGE_SHIFT
;
1361 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1363 cycled
= 1; /* ignore range_cyclic tests */
1365 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1366 tag
= PAGECACHE_TAG_TOWRITE
;
1368 tag
= PAGECACHE_TAG_DIRTY
;
1370 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1371 tag_pages_for_writeback(mapping
, index
, end
);
1373 while (!done
&& (index
<= end
)) {
1376 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
1377 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1);
1381 for (i
= 0; i
< nr_pages
; i
++) {
1382 struct page
*page
= pvec
.pages
[i
];
1384 if (page
->index
> end
) {
1389 done_index
= page
->index
;
1393 if (unlikely(page
->mapping
!= mapping
)) {
1399 if (!PageDirty(page
)) {
1400 /* someone wrote it for us */
1401 goto continue_unlock
;
1404 if (step
== is_cold_data(page
))
1405 goto continue_unlock
;
1407 if (PageWriteback(page
)) {
1408 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1409 f2fs_wait_on_page_writeback(page
,
1412 goto continue_unlock
;
1415 BUG_ON(PageWriteback(page
));
1416 if (!clear_page_dirty_for_io(page
))
1417 goto continue_unlock
;
1419 ret
= (*writepage
)(page
, wbc
, data
);
1420 if (unlikely(ret
)) {
1421 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
1425 done_index
= page
->index
+ 1;
1431 if (--wbc
->nr_to_write
<= 0 &&
1432 wbc
->sync_mode
== WB_SYNC_NONE
) {
1437 pagevec_release(&pvec
);
1446 if (!cycled
&& !done
) {
1449 end
= writeback_index
- 1;
1452 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1453 mapping
->writeback_index
= done_index
;
1458 static int f2fs_write_data_pages(struct address_space
*mapping
,
1459 struct writeback_control
*wbc
)
1461 struct inode
*inode
= mapping
->host
;
1462 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1463 bool locked
= false;
1467 /* deal with chardevs and other special file */
1468 if (!mapping
->a_ops
->writepage
)
1471 /* skip writing if there is no dirty page in this inode */
1472 if (!get_dirty_pages(inode
) && wbc
->sync_mode
== WB_SYNC_NONE
)
1475 if (S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_NONE
&&
1476 get_dirty_pages(inode
) < nr_pages_to_skip(sbi
, DATA
) &&
1477 available_free_memory(sbi
, DIRTY_DENTS
))
1480 /* skip writing during file defragment */
1481 if (is_inode_flag_set(F2FS_I(inode
), FI_DO_DEFRAG
))
1484 /* during POR, we don't need to trigger writepage at all. */
1485 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1488 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1490 diff
= nr_pages_to_write(sbi
, DATA
, wbc
);
1492 if (!S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_ALL
) {
1493 mutex_lock(&sbi
->writepages
);
1496 ret
= f2fs_write_cache_pages(mapping
, wbc
, __f2fs_writepage
, mapping
);
1497 f2fs_submit_merged_bio_cond(sbi
, inode
, NULL
, 0, DATA
, WRITE
);
1499 mutex_unlock(&sbi
->writepages
);
1501 remove_dirty_inode(inode
);
1503 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- diff
);
1507 wbc
->pages_skipped
+= get_dirty_pages(inode
);
1508 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1512 static void f2fs_write_failed(struct address_space
*mapping
, loff_t to
)
1514 struct inode
*inode
= mapping
->host
;
1515 loff_t i_size
= i_size_read(inode
);
1518 truncate_pagecache(inode
, i_size
);
1519 truncate_blocks(inode
, i_size
, true);
1523 static int prepare_write_begin(struct f2fs_sb_info
*sbi
,
1524 struct page
*page
, loff_t pos
, unsigned len
,
1525 block_t
*blk_addr
, bool *node_changed
)
1527 struct inode
*inode
= page
->mapping
->host
;
1528 pgoff_t index
= page
->index
;
1529 struct dnode_of_data dn
;
1531 bool locked
= false;
1532 struct extent_info ei
;
1536 * we already allocated all the blocks, so we don't need to get
1537 * the block addresses when there is no need to fill the page.
1539 if (!f2fs_has_inline_data(inode
) && !f2fs_encrypted_inode(inode
) &&
1543 if (f2fs_has_inline_data(inode
) ||
1544 (pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1549 /* check inline_data */
1550 ipage
= get_node_page(sbi
, inode
->i_ino
);
1551 if (IS_ERR(ipage
)) {
1552 err
= PTR_ERR(ipage
);
1556 set_new_dnode(&dn
, inode
, ipage
, ipage
, 0);
1558 if (f2fs_has_inline_data(inode
)) {
1559 if (pos
+ len
<= MAX_INLINE_DATA
) {
1560 read_inline_data(page
, ipage
);
1561 set_inode_flag(F2FS_I(inode
), FI_DATA_EXIST
);
1563 set_inline_node(ipage
);
1565 err
= f2fs_convert_inline_page(&dn
, page
);
1568 if (dn
.data_blkaddr
== NULL_ADDR
)
1569 err
= f2fs_get_block(&dn
, index
);
1571 } else if (locked
) {
1572 err
= f2fs_get_block(&dn
, index
);
1574 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
1575 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
1578 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
1579 if (err
|| dn
.data_blkaddr
== NULL_ADDR
) {
1580 f2fs_put_dnode(&dn
);
1588 /* convert_inline_page can make node_changed */
1589 *blk_addr
= dn
.data_blkaddr
;
1590 *node_changed
= dn
.node_changed
;
1592 f2fs_put_dnode(&dn
);
1595 f2fs_unlock_op(sbi
);
1599 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
1600 loff_t pos
, unsigned len
, unsigned flags
,
1601 struct page
**pagep
, void **fsdata
)
1603 struct inode
*inode
= mapping
->host
;
1604 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1605 struct page
*page
= NULL
;
1606 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_SHIFT
;
1607 bool need_balance
= false;
1608 block_t blkaddr
= NULL_ADDR
;
1611 trace_f2fs_write_begin(inode
, pos
, len
, flags
);
1614 * We should check this at this moment to avoid deadlock on inode page
1615 * and #0 page. The locking rule for inline_data conversion should be:
1616 * lock_page(page #0) -> lock_page(inode_page)
1619 err
= f2fs_convert_inline_inode(inode
);
1624 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1632 err
= prepare_write_begin(sbi
, page
, pos
, len
,
1633 &blkaddr
, &need_balance
);
1637 if (need_balance
&& has_not_enough_free_secs(sbi
, 0)) {
1639 f2fs_balance_fs(sbi
, true);
1641 if (page
->mapping
!= mapping
) {
1642 /* The page got truncated from under us */
1643 f2fs_put_page(page
, 1);
1648 f2fs_wait_on_page_writeback(page
, DATA
, false);
1650 /* wait for GCed encrypted page writeback */
1651 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1652 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
1654 if (len
== PAGE_SIZE
)
1656 if (PageUptodate(page
))
1659 if ((pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1660 unsigned start
= pos
& (PAGE_SIZE
- 1);
1661 unsigned end
= start
+ len
;
1663 /* Reading beyond i_size is simple: memset to zero */
1664 zero_user_segments(page
, 0, start
, end
, PAGE_SIZE
);
1668 if (blkaddr
== NEW_ADDR
) {
1669 zero_user_segment(page
, 0, PAGE_SIZE
);
1671 struct f2fs_io_info fio
= {
1675 .op_flags
= READ_SYNC
,
1676 .old_blkaddr
= blkaddr
,
1677 .new_blkaddr
= blkaddr
,
1679 .encrypted_page
= NULL
,
1681 err
= f2fs_submit_page_bio(&fio
);
1686 if (unlikely(!PageUptodate(page
))) {
1690 if (unlikely(page
->mapping
!= mapping
)) {
1691 f2fs_put_page(page
, 1);
1695 /* avoid symlink page */
1696 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1697 err
= fscrypt_decrypt_page(page
);
1703 SetPageUptodate(page
);
1705 clear_cold_data(page
);
1709 f2fs_put_page(page
, 1);
1710 f2fs_write_failed(mapping
, pos
+ len
);
1714 static int f2fs_write_end(struct file
*file
,
1715 struct address_space
*mapping
,
1716 loff_t pos
, unsigned len
, unsigned copied
,
1717 struct page
*page
, void *fsdata
)
1719 struct inode
*inode
= page
->mapping
->host
;
1721 trace_f2fs_write_end(inode
, pos
, len
, copied
);
1723 set_page_dirty(page
);
1725 if (pos
+ copied
> i_size_read(inode
)) {
1726 i_size_write(inode
, pos
+ copied
);
1727 mark_inode_dirty(inode
);
1730 f2fs_put_page(page
, 1);
1731 f2fs_update_time(F2FS_I_SB(inode
), REQ_TIME
);
1735 static int check_direct_IO(struct inode
*inode
, struct iov_iter
*iter
,
1738 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
1740 if (offset
& blocksize_mask
)
1743 if (iov_iter_alignment(iter
) & blocksize_mask
)
1749 static ssize_t
f2fs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
1751 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
1752 struct inode
*inode
= mapping
->host
;
1753 size_t count
= iov_iter_count(iter
);
1754 loff_t offset
= iocb
->ki_pos
;
1757 err
= check_direct_IO(inode
, iter
, offset
);
1761 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1764 trace_f2fs_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
1766 err
= blockdev_direct_IO(iocb
, inode
, iter
, get_data_block_dio
);
1767 if (iov_iter_rw(iter
) == WRITE
) {
1769 set_inode_flag(F2FS_I(inode
), FI_UPDATE_WRITE
);
1771 f2fs_write_failed(mapping
, offset
+ count
);
1774 trace_f2fs_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), err
);
1779 void f2fs_invalidate_page(struct page
*page
, unsigned int offset
,
1780 unsigned int length
)
1782 struct inode
*inode
= page
->mapping
->host
;
1783 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1785 if (inode
->i_ino
>= F2FS_ROOT_INO(sbi
) &&
1786 (offset
% PAGE_SIZE
|| length
!= PAGE_SIZE
))
1789 if (PageDirty(page
)) {
1790 if (inode
->i_ino
== F2FS_META_INO(sbi
))
1791 dec_page_count(sbi
, F2FS_DIRTY_META
);
1792 else if (inode
->i_ino
== F2FS_NODE_INO(sbi
))
1793 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1795 inode_dec_dirty_pages(inode
);
1798 /* This is atomic written page, keep Private */
1799 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1802 set_page_private(page
, 0);
1803 ClearPagePrivate(page
);
1806 int f2fs_release_page(struct page
*page
, gfp_t wait
)
1808 /* If this is dirty page, keep PagePrivate */
1809 if (PageDirty(page
))
1812 /* This is atomic written page, keep Private */
1813 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1816 set_page_private(page
, 0);
1817 ClearPagePrivate(page
);
1821 static int f2fs_set_data_page_dirty(struct page
*page
)
1823 struct address_space
*mapping
= page
->mapping
;
1824 struct inode
*inode
= mapping
->host
;
1826 trace_f2fs_set_page_dirty(page
, DATA
);
1828 SetPageUptodate(page
);
1830 if (f2fs_is_atomic_file(inode
)) {
1831 if (!IS_ATOMIC_WRITTEN_PAGE(page
)) {
1832 register_inmem_page(inode
, page
);
1836 * Previously, this page has been registered, we just
1842 if (!PageDirty(page
)) {
1843 __set_page_dirty_nobuffers(page
);
1844 update_dirty_page(inode
, page
);
1850 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
1852 struct inode
*inode
= mapping
->host
;
1854 if (f2fs_has_inline_data(inode
))
1857 /* make sure allocating whole blocks */
1858 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
1859 filemap_write_and_wait(mapping
);
1861 return generic_block_bmap(mapping
, block
, get_data_block_bmap
);
1864 const struct address_space_operations f2fs_dblock_aops
= {
1865 .readpage
= f2fs_read_data_page
,
1866 .readpages
= f2fs_read_data_pages
,
1867 .writepage
= f2fs_write_data_page
,
1868 .writepages
= f2fs_write_data_pages
,
1869 .write_begin
= f2fs_write_begin
,
1870 .write_end
= f2fs_write_end
,
1871 .set_page_dirty
= f2fs_set_data_page_dirty
,
1872 .invalidatepage
= f2fs_invalidate_page
,
1873 .releasepage
= f2fs_release_page
,
1874 .direct_IO
= f2fs_direct_IO
,