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 bool __is_cp_guaranteed(struct page
*page
)
34 struct address_space
*mapping
= page
->mapping
;
36 struct f2fs_sb_info
*sbi
;
41 inode
= mapping
->host
;
42 sbi
= F2FS_I_SB(inode
);
44 if (inode
->i_ino
== F2FS_META_INO(sbi
) ||
45 inode
->i_ino
== F2FS_NODE_INO(sbi
) ||
46 S_ISDIR(inode
->i_mode
) ||
52 static void f2fs_read_end_io(struct bio
*bio
)
57 #ifdef CONFIG_F2FS_FAULT_INJECTION
58 if (time_to_inject(F2FS_P_SB(bio
->bi_io_vec
->bv_page
), FAULT_IO
))
62 if (f2fs_bio_encrypted(bio
)) {
64 fscrypt_release_ctx(bio
->bi_private
);
66 fscrypt_decrypt_bio_pages(bio
->bi_private
, bio
);
71 bio_for_each_segment_all(bvec
, bio
, i
) {
72 struct page
*page
= bvec
->bv_page
;
75 if (!PageUptodate(page
))
76 SetPageUptodate(page
);
78 ClearPageUptodate(page
);
86 static void f2fs_write_end_io(struct bio
*bio
)
88 struct f2fs_sb_info
*sbi
= bio
->bi_private
;
92 bio_for_each_segment_all(bvec
, bio
, i
) {
93 struct page
*page
= bvec
->bv_page
;
94 enum count_type type
= WB_DATA_TYPE(page
);
96 fscrypt_pullback_bio_page(&page
, true);
98 if (unlikely(bio
->bi_error
)) {
99 mapping_set_error(page
->mapping
, -EIO
);
100 f2fs_stop_checkpoint(sbi
, true);
102 dec_page_count(sbi
, type
);
103 clear_cold_data(page
);
104 end_page_writeback(page
);
106 if (!get_pages(sbi
, F2FS_WB_CP_DATA
) &&
107 wq_has_sleeper(&sbi
->cp_wait
))
108 wake_up(&sbi
->cp_wait
);
114 * Return true, if pre_bio's bdev is same as its target device.
116 struct block_device
*f2fs_target_device(struct f2fs_sb_info
*sbi
,
117 block_t blk_addr
, struct bio
*bio
)
119 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
122 for (i
= 0; i
< sbi
->s_ndevs
; i
++) {
123 if (FDEV(i
).start_blk
<= blk_addr
&&
124 FDEV(i
).end_blk
>= blk_addr
) {
125 blk_addr
-= FDEV(i
).start_blk
;
132 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blk_addr
);
137 int f2fs_target_device_index(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
141 for (i
= 0; i
< sbi
->s_ndevs
; i
++)
142 if (FDEV(i
).start_blk
<= blkaddr
&& FDEV(i
).end_blk
>= blkaddr
)
147 static bool __same_bdev(struct f2fs_sb_info
*sbi
,
148 block_t blk_addr
, struct bio
*bio
)
150 return f2fs_target_device(sbi
, blk_addr
, NULL
) == bio
->bi_bdev
;
154 * Low-level block read/write IO operations.
156 static struct bio
*__bio_alloc(struct f2fs_sb_info
*sbi
, block_t blk_addr
,
157 int npages
, bool is_read
)
161 bio
= f2fs_bio_alloc(npages
);
163 f2fs_target_device(sbi
, blk_addr
, bio
);
164 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
165 bio
->bi_private
= is_read
? NULL
: sbi
;
170 static inline void __submit_bio(struct f2fs_sb_info
*sbi
,
171 struct bio
*bio
, enum page_type type
)
173 if (!is_read_io(bio_op(bio
))) {
174 if (f2fs_sb_mounted_blkzoned(sbi
->sb
) &&
175 current
->plug
&& (type
== DATA
|| type
== NODE
))
176 blk_finish_plug(current
->plug
);
181 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
183 struct f2fs_io_info
*fio
= &io
->fio
;
188 if (is_read_io(fio
->op
))
189 trace_f2fs_submit_read_bio(io
->sbi
->sb
, fio
, io
->bio
);
191 trace_f2fs_submit_write_bio(io
->sbi
->sb
, fio
, io
->bio
);
193 bio_set_op_attrs(io
->bio
, fio
->op
, fio
->op_flags
);
195 __submit_bio(io
->sbi
, io
->bio
, fio
->type
);
199 static bool __has_merged_page(struct f2fs_bio_info
*io
, struct inode
*inode
,
200 struct page
*page
, nid_t ino
)
202 struct bio_vec
*bvec
;
209 if (!inode
&& !page
&& !ino
)
212 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
214 if (bvec
->bv_page
->mapping
)
215 target
= bvec
->bv_page
;
217 target
= fscrypt_control_page(bvec
->bv_page
);
219 if (inode
&& inode
== target
->mapping
->host
)
221 if (page
&& page
== target
)
223 if (ino
&& ino
== ino_of_node(target
))
230 static bool has_merged_page(struct f2fs_sb_info
*sbi
, struct inode
*inode
,
231 struct page
*page
, nid_t ino
,
234 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
235 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
238 down_read(&io
->io_rwsem
);
239 ret
= __has_merged_page(io
, inode
, page
, ino
);
240 up_read(&io
->io_rwsem
);
244 static void __f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
245 struct inode
*inode
, struct page
*page
,
246 nid_t ino
, enum page_type type
, int rw
)
248 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
249 struct f2fs_bio_info
*io
;
251 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
253 down_write(&io
->io_rwsem
);
255 if (!__has_merged_page(io
, inode
, page
, ino
))
258 /* change META to META_FLUSH in the checkpoint procedure */
259 if (type
>= META_FLUSH
) {
260 io
->fio
.type
= META_FLUSH
;
261 io
->fio
.op
= REQ_OP_WRITE
;
262 io
->fio
.op_flags
= REQ_PREFLUSH
| REQ_META
| REQ_PRIO
;
263 if (!test_opt(sbi
, NOBARRIER
))
264 io
->fio
.op_flags
|= REQ_FUA
;
266 __submit_merged_bio(io
);
268 up_write(&io
->io_rwsem
);
271 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
, enum page_type type
,
274 __f2fs_submit_merged_bio(sbi
, NULL
, NULL
, 0, type
, rw
);
277 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info
*sbi
,
278 struct inode
*inode
, struct page
*page
,
279 nid_t ino
, enum page_type type
, int rw
)
281 if (has_merged_page(sbi
, inode
, page
, ino
, type
))
282 __f2fs_submit_merged_bio(sbi
, inode
, page
, ino
, type
, rw
);
285 void f2fs_flush_merged_bios(struct f2fs_sb_info
*sbi
)
287 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
288 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
289 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
293 * Fill the locked page with data located in the block address.
294 * Return unlocked page.
296 int f2fs_submit_page_bio(struct f2fs_io_info
*fio
)
299 struct page
*page
= fio
->encrypted_page
?
300 fio
->encrypted_page
: fio
->page
;
302 trace_f2fs_submit_page_bio(page
, fio
);
303 f2fs_trace_ios(fio
, 0);
305 /* Allocate a new bio */
306 bio
= __bio_alloc(fio
->sbi
, fio
->new_blkaddr
, 1, is_read_io(fio
->op
));
308 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
312 bio_set_op_attrs(bio
, fio
->op
, fio
->op_flags
);
314 __submit_bio(fio
->sbi
, bio
, fio
->type
);
318 void f2fs_submit_page_mbio(struct f2fs_io_info
*fio
)
320 struct f2fs_sb_info
*sbi
= fio
->sbi
;
321 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
322 struct f2fs_bio_info
*io
;
323 bool is_read
= is_read_io(fio
->op
);
324 struct page
*bio_page
;
326 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
328 if (fio
->old_blkaddr
!= NEW_ADDR
)
329 verify_block_addr(sbi
, fio
->old_blkaddr
);
330 verify_block_addr(sbi
, fio
->new_blkaddr
);
332 bio_page
= fio
->encrypted_page
? fio
->encrypted_page
: fio
->page
;
335 inc_page_count(sbi
, WB_DATA_TYPE(bio_page
));
337 down_write(&io
->io_rwsem
);
339 if (io
->bio
&& (io
->last_block_in_bio
!= fio
->new_blkaddr
- 1 ||
340 (io
->fio
.op
!= fio
->op
|| io
->fio
.op_flags
!= fio
->op_flags
) ||
341 !__same_bdev(sbi
, fio
->new_blkaddr
, io
->bio
)))
342 __submit_merged_bio(io
);
344 if (io
->bio
== NULL
) {
345 io
->bio
= __bio_alloc(sbi
, fio
->new_blkaddr
,
346 BIO_MAX_PAGES
, is_read
);
350 if (bio_add_page(io
->bio
, bio_page
, PAGE_SIZE
, 0) <
352 __submit_merged_bio(io
);
356 io
->last_block_in_bio
= fio
->new_blkaddr
;
357 f2fs_trace_ios(fio
, 0);
359 up_write(&io
->io_rwsem
);
360 trace_f2fs_submit_page_mbio(fio
->page
, fio
);
363 static void __set_data_blkaddr(struct dnode_of_data
*dn
)
365 struct f2fs_node
*rn
= F2FS_NODE(dn
->node_page
);
368 /* Get physical address of data block */
369 addr_array
= blkaddr_in_node(rn
);
370 addr_array
[dn
->ofs_in_node
] = cpu_to_le32(dn
->data_blkaddr
);
374 * Lock ordering for the change of data block address:
377 * update block addresses in the node page
379 void set_data_blkaddr(struct dnode_of_data
*dn
)
381 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
382 __set_data_blkaddr(dn
);
383 if (set_page_dirty(dn
->node_page
))
384 dn
->node_changed
= true;
387 void f2fs_update_data_blkaddr(struct dnode_of_data
*dn
, block_t blkaddr
)
389 dn
->data_blkaddr
= blkaddr
;
390 set_data_blkaddr(dn
);
391 f2fs_update_extent_cache(dn
);
394 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
395 int reserve_new_blocks(struct dnode_of_data
*dn
, blkcnt_t count
)
397 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
402 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
404 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
407 trace_f2fs_reserve_new_blocks(dn
->inode
, dn
->nid
,
408 dn
->ofs_in_node
, count
);
410 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
412 for (; count
> 0; dn
->ofs_in_node
++) {
414 datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
415 if (blkaddr
== NULL_ADDR
) {
416 dn
->data_blkaddr
= NEW_ADDR
;
417 __set_data_blkaddr(dn
);
422 if (set_page_dirty(dn
->node_page
))
423 dn
->node_changed
= true;
427 /* Should keep dn->ofs_in_node unchanged */
428 int reserve_new_block(struct dnode_of_data
*dn
)
430 unsigned int ofs_in_node
= dn
->ofs_in_node
;
433 ret
= reserve_new_blocks(dn
, 1);
434 dn
->ofs_in_node
= ofs_in_node
;
438 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
440 bool need_put
= dn
->inode_page
? false : true;
443 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
447 if (dn
->data_blkaddr
== NULL_ADDR
)
448 err
= reserve_new_block(dn
);
454 int f2fs_get_block(struct dnode_of_data
*dn
, pgoff_t index
)
456 struct extent_info ei
;
457 struct inode
*inode
= dn
->inode
;
459 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
460 dn
->data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
464 return f2fs_reserve_block(dn
, index
);
467 struct page
*get_read_data_page(struct inode
*inode
, pgoff_t index
,
468 int op_flags
, bool for_write
)
470 struct address_space
*mapping
= inode
->i_mapping
;
471 struct dnode_of_data dn
;
473 struct extent_info ei
;
475 struct f2fs_io_info fio
= {
476 .sbi
= F2FS_I_SB(inode
),
479 .op_flags
= op_flags
,
480 .encrypted_page
= NULL
,
483 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
484 return read_mapping_page(mapping
, index
, NULL
);
486 page
= f2fs_grab_cache_page(mapping
, index
, for_write
);
488 return ERR_PTR(-ENOMEM
);
490 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
491 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
495 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
496 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
501 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
506 if (PageUptodate(page
)) {
512 * A new dentry page is allocated but not able to be written, since its
513 * new inode page couldn't be allocated due to -ENOSPC.
514 * In such the case, its blkaddr can be remained as NEW_ADDR.
515 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
517 if (dn
.data_blkaddr
== NEW_ADDR
) {
518 zero_user_segment(page
, 0, PAGE_SIZE
);
519 if (!PageUptodate(page
))
520 SetPageUptodate(page
);
525 fio
.new_blkaddr
= fio
.old_blkaddr
= dn
.data_blkaddr
;
527 err
= f2fs_submit_page_bio(&fio
);
533 f2fs_put_page(page
, 1);
537 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
)
539 struct address_space
*mapping
= inode
->i_mapping
;
542 page
= find_get_page(mapping
, index
);
543 if (page
&& PageUptodate(page
))
545 f2fs_put_page(page
, 0);
547 page
= get_read_data_page(inode
, index
, 0, false);
551 if (PageUptodate(page
))
554 wait_on_page_locked(page
);
555 if (unlikely(!PageUptodate(page
))) {
556 f2fs_put_page(page
, 0);
557 return ERR_PTR(-EIO
);
563 * If it tries to access a hole, return an error.
564 * Because, the callers, functions in dir.c and GC, should be able to know
565 * whether this page exists or not.
567 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
,
570 struct address_space
*mapping
= inode
->i_mapping
;
573 page
= get_read_data_page(inode
, index
, 0, for_write
);
577 /* wait for read completion */
579 if (unlikely(page
->mapping
!= mapping
)) {
580 f2fs_put_page(page
, 1);
583 if (unlikely(!PageUptodate(page
))) {
584 f2fs_put_page(page
, 1);
585 return ERR_PTR(-EIO
);
591 * Caller ensures that this data page is never allocated.
592 * A new zero-filled data page is allocated in the page cache.
594 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
596 * Note that, ipage is set only by make_empty_dir, and if any error occur,
597 * ipage should be released by this function.
599 struct page
*get_new_data_page(struct inode
*inode
,
600 struct page
*ipage
, pgoff_t index
, bool new_i_size
)
602 struct address_space
*mapping
= inode
->i_mapping
;
604 struct dnode_of_data dn
;
607 page
= f2fs_grab_cache_page(mapping
, index
, true);
610 * before exiting, we should make sure ipage will be released
611 * if any error occur.
613 f2fs_put_page(ipage
, 1);
614 return ERR_PTR(-ENOMEM
);
617 set_new_dnode(&dn
, inode
, ipage
, NULL
, 0);
618 err
= f2fs_reserve_block(&dn
, index
);
620 f2fs_put_page(page
, 1);
626 if (PageUptodate(page
))
629 if (dn
.data_blkaddr
== NEW_ADDR
) {
630 zero_user_segment(page
, 0, PAGE_SIZE
);
631 if (!PageUptodate(page
))
632 SetPageUptodate(page
);
634 f2fs_put_page(page
, 1);
636 /* if ipage exists, blkaddr should be NEW_ADDR */
637 f2fs_bug_on(F2FS_I_SB(inode
), ipage
);
638 page
= get_lock_data_page(inode
, index
, true);
643 if (new_i_size
&& i_size_read(inode
) <
644 ((loff_t
)(index
+ 1) << PAGE_SHIFT
))
645 f2fs_i_size_write(inode
, ((loff_t
)(index
+ 1) << PAGE_SHIFT
));
649 static int __allocate_data_block(struct dnode_of_data
*dn
)
651 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
652 struct f2fs_summary sum
;
657 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
660 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
661 if (dn
->data_blkaddr
== NEW_ADDR
)
664 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
668 get_node_info(sbi
, dn
->nid
, &ni
);
669 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
671 allocate_data_block(sbi
, NULL
, dn
->data_blkaddr
, &dn
->data_blkaddr
,
672 &sum
, CURSEG_WARM_DATA
);
673 set_data_blkaddr(dn
);
676 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), dn
->inode
) +
678 if (i_size_read(dn
->inode
) < ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
))
679 f2fs_i_size_write(dn
->inode
,
680 ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
));
684 static inline bool __force_buffered_io(struct inode
*inode
, int rw
)
686 return ((f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) ||
687 (rw
== WRITE
&& test_opt(F2FS_I_SB(inode
), LFS
)) ||
688 F2FS_I_SB(inode
)->s_ndevs
);
691 int f2fs_preallocate_blocks(struct kiocb
*iocb
, struct iov_iter
*from
)
693 struct inode
*inode
= file_inode(iocb
->ki_filp
);
694 struct f2fs_map_blocks map
;
697 map
.m_lblk
= F2FS_BLK_ALIGN(iocb
->ki_pos
);
698 map
.m_len
= F2FS_BYTES_TO_BLK(iocb
->ki_pos
+ iov_iter_count(from
));
699 if (map
.m_len
> map
.m_lblk
)
700 map
.m_len
-= map
.m_lblk
;
704 map
.m_next_pgofs
= NULL
;
706 if (iocb
->ki_flags
& IOCB_DIRECT
) {
707 err
= f2fs_convert_inline_inode(inode
);
710 return f2fs_map_blocks(inode
, &map
, 1,
711 __force_buffered_io(inode
, WRITE
) ?
712 F2FS_GET_BLOCK_PRE_AIO
:
713 F2FS_GET_BLOCK_PRE_DIO
);
715 if (iocb
->ki_pos
+ iov_iter_count(from
) > MAX_INLINE_DATA
) {
716 err
= f2fs_convert_inline_inode(inode
);
720 if (!f2fs_has_inline_data(inode
))
721 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_AIO
);
726 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
727 * f2fs_map_blocks structure.
728 * If original data blocks are allocated, then give them to blockdev.
730 * a. preallocate requested block addresses
731 * b. do not use extent cache for better performance
732 * c. give the block addresses to blockdev
734 int f2fs_map_blocks(struct inode
*inode
, struct f2fs_map_blocks
*map
,
735 int create
, int flag
)
737 unsigned int maxblocks
= map
->m_len
;
738 struct dnode_of_data dn
;
739 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
740 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE
;
741 pgoff_t pgofs
, end_offset
, end
;
742 int err
= 0, ofs
= 1;
743 unsigned int ofs_in_node
, last_ofs_in_node
;
745 struct extent_info ei
;
754 /* it only supports block size == page size */
755 pgofs
= (pgoff_t
)map
->m_lblk
;
756 end
= pgofs
+ maxblocks
;
758 if (!create
&& f2fs_lookup_extent_cache(inode
, pgofs
, &ei
)) {
759 map
->m_pblk
= ei
.blk
+ pgofs
- ei
.fofs
;
760 map
->m_len
= min((pgoff_t
)maxblocks
, ei
.fofs
+ ei
.len
- pgofs
);
761 map
->m_flags
= F2FS_MAP_MAPPED
;
769 /* When reading holes, we need its node page */
770 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
771 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
773 if (flag
== F2FS_GET_BLOCK_BMAP
)
775 if (err
== -ENOENT
) {
777 if (map
->m_next_pgofs
)
779 get_next_page_offset(&dn
, pgofs
);
785 last_ofs_in_node
= ofs_in_node
= dn
.ofs_in_node
;
786 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, inode
);
789 blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
791 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
) {
793 if (unlikely(f2fs_cp_error(sbi
))) {
797 if (flag
== F2FS_GET_BLOCK_PRE_AIO
) {
798 if (blkaddr
== NULL_ADDR
) {
800 last_ofs_in_node
= dn
.ofs_in_node
;
803 err
= __allocate_data_block(&dn
);
805 set_inode_flag(inode
, FI_APPEND_WRITE
);
809 map
->m_flags
= F2FS_MAP_NEW
;
810 blkaddr
= dn
.data_blkaddr
;
812 if (flag
== F2FS_GET_BLOCK_BMAP
) {
816 if (flag
== F2FS_GET_BLOCK_FIEMAP
&&
817 blkaddr
== NULL_ADDR
) {
818 if (map
->m_next_pgofs
)
819 *map
->m_next_pgofs
= pgofs
+ 1;
821 if (flag
!= F2FS_GET_BLOCK_FIEMAP
||
827 if (flag
== F2FS_GET_BLOCK_PRE_AIO
)
830 if (map
->m_len
== 0) {
831 /* preallocated unwritten block should be mapped for fiemap. */
832 if (blkaddr
== NEW_ADDR
)
833 map
->m_flags
|= F2FS_MAP_UNWRITTEN
;
834 map
->m_flags
|= F2FS_MAP_MAPPED
;
836 map
->m_pblk
= blkaddr
;
838 } else if ((map
->m_pblk
!= NEW_ADDR
&&
839 blkaddr
== (map
->m_pblk
+ ofs
)) ||
840 (map
->m_pblk
== NEW_ADDR
&& blkaddr
== NEW_ADDR
) ||
841 flag
== F2FS_GET_BLOCK_PRE_DIO
) {
852 /* preallocate blocks in batch for one dnode page */
853 if (flag
== F2FS_GET_BLOCK_PRE_AIO
&&
854 (pgofs
== end
|| dn
.ofs_in_node
== end_offset
)) {
856 dn
.ofs_in_node
= ofs_in_node
;
857 err
= reserve_new_blocks(&dn
, prealloc
);
861 map
->m_len
+= dn
.ofs_in_node
- ofs_in_node
;
862 if (prealloc
&& dn
.ofs_in_node
!= last_ofs_in_node
+ 1) {
866 dn
.ofs_in_node
= end_offset
;
871 else if (dn
.ofs_in_node
< end_offset
)
878 f2fs_balance_fs(sbi
, dn
.node_changed
);
887 f2fs_balance_fs(sbi
, dn
.node_changed
);
890 trace_f2fs_map_blocks(inode
, map
, err
);
894 static int __get_data_block(struct inode
*inode
, sector_t iblock
,
895 struct buffer_head
*bh
, int create
, int flag
,
898 struct f2fs_map_blocks map
;
902 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
903 map
.m_next_pgofs
= next_pgofs
;
905 err
= f2fs_map_blocks(inode
, &map
, create
, flag
);
907 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
908 bh
->b_state
= (bh
->b_state
& ~F2FS_MAP_FLAGS
) | map
.m_flags
;
909 bh
->b_size
= map
.m_len
<< inode
->i_blkbits
;
914 static int get_data_block(struct inode
*inode
, sector_t iblock
,
915 struct buffer_head
*bh_result
, int create
, int flag
,
918 return __get_data_block(inode
, iblock
, bh_result
, create
,
922 static int get_data_block_dio(struct inode
*inode
, sector_t iblock
,
923 struct buffer_head
*bh_result
, int create
)
925 return __get_data_block(inode
, iblock
, bh_result
, create
,
926 F2FS_GET_BLOCK_DIO
, NULL
);
929 static int get_data_block_bmap(struct inode
*inode
, sector_t iblock
,
930 struct buffer_head
*bh_result
, int create
)
932 /* Block number less than F2FS MAX BLOCKS */
933 if (unlikely(iblock
>= F2FS_I_SB(inode
)->max_file_blocks
))
936 return __get_data_block(inode
, iblock
, bh_result
, create
,
937 F2FS_GET_BLOCK_BMAP
, NULL
);
940 static inline sector_t
logical_to_blk(struct inode
*inode
, loff_t offset
)
942 return (offset
>> inode
->i_blkbits
);
945 static inline loff_t
blk_to_logical(struct inode
*inode
, sector_t blk
)
947 return (blk
<< inode
->i_blkbits
);
950 int f2fs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
953 struct buffer_head map_bh
;
954 sector_t start_blk
, last_blk
;
956 u64 logical
= 0, phys
= 0, size
= 0;
960 ret
= fiemap_check_flags(fieinfo
, FIEMAP_FLAG_SYNC
);
964 if (f2fs_has_inline_data(inode
)) {
965 ret
= f2fs_inline_data_fiemap(inode
, fieinfo
, start
, len
);
972 if (logical_to_blk(inode
, len
) == 0)
973 len
= blk_to_logical(inode
, 1);
975 start_blk
= logical_to_blk(inode
, start
);
976 last_blk
= logical_to_blk(inode
, start
+ len
- 1);
979 memset(&map_bh
, 0, sizeof(struct buffer_head
));
982 ret
= get_data_block(inode
, start_blk
, &map_bh
, 0,
983 F2FS_GET_BLOCK_FIEMAP
, &next_pgofs
);
988 if (!buffer_mapped(&map_bh
)) {
989 start_blk
= next_pgofs
;
991 if (blk_to_logical(inode
, start_blk
) < blk_to_logical(inode
,
992 F2FS_I_SB(inode
)->max_file_blocks
))
995 flags
|= FIEMAP_EXTENT_LAST
;
999 if (f2fs_encrypted_inode(inode
))
1000 flags
|= FIEMAP_EXTENT_DATA_ENCRYPTED
;
1002 ret
= fiemap_fill_next_extent(fieinfo
, logical
,
1006 if (start_blk
> last_blk
|| ret
)
1009 logical
= blk_to_logical(inode
, start_blk
);
1010 phys
= blk_to_logical(inode
, map_bh
.b_blocknr
);
1011 size
= map_bh
.b_size
;
1013 if (buffer_unwritten(&map_bh
))
1014 flags
= FIEMAP_EXTENT_UNWRITTEN
;
1016 start_blk
+= logical_to_blk(inode
, size
);
1020 if (fatal_signal_pending(current
))
1028 inode_unlock(inode
);
1032 static struct bio
*f2fs_grab_bio(struct inode
*inode
, block_t blkaddr
,
1035 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1036 struct fscrypt_ctx
*ctx
= NULL
;
1039 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1040 ctx
= fscrypt_get_ctx(inode
, GFP_NOFS
);
1042 return ERR_CAST(ctx
);
1044 /* wait the page to be moved by cleaning */
1045 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
1048 bio
= bio_alloc(GFP_KERNEL
, min_t(int, nr_pages
, BIO_MAX_PAGES
));
1051 fscrypt_release_ctx(ctx
);
1052 return ERR_PTR(-ENOMEM
);
1054 f2fs_target_device(sbi
, blkaddr
, bio
);
1055 bio
->bi_end_io
= f2fs_read_end_io
;
1056 bio
->bi_private
= ctx
;
1062 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1063 * Major change was from block_size == page_size in f2fs by default.
1065 static int f2fs_mpage_readpages(struct address_space
*mapping
,
1066 struct list_head
*pages
, struct page
*page
,
1069 struct bio
*bio
= NULL
;
1071 sector_t last_block_in_bio
= 0;
1072 struct inode
*inode
= mapping
->host
;
1073 const unsigned blkbits
= inode
->i_blkbits
;
1074 const unsigned blocksize
= 1 << blkbits
;
1075 sector_t block_in_file
;
1076 sector_t last_block
;
1077 sector_t last_block_in_file
;
1079 struct f2fs_map_blocks map
;
1085 map
.m_next_pgofs
= NULL
;
1087 for (page_idx
= 0; nr_pages
; page_idx
++, nr_pages
--) {
1089 prefetchw(&page
->flags
);
1091 page
= list_entry(pages
->prev
, struct page
, lru
);
1092 list_del(&page
->lru
);
1093 if (add_to_page_cache_lru(page
, mapping
,
1095 readahead_gfp_mask(mapping
)))
1099 block_in_file
= (sector_t
)page
->index
;
1100 last_block
= block_in_file
+ nr_pages
;
1101 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >>
1103 if (last_block
> last_block_in_file
)
1104 last_block
= last_block_in_file
;
1107 * Map blocks using the previous result first.
1109 if ((map
.m_flags
& F2FS_MAP_MAPPED
) &&
1110 block_in_file
> map
.m_lblk
&&
1111 block_in_file
< (map
.m_lblk
+ map
.m_len
))
1115 * Then do more f2fs_map_blocks() calls until we are
1116 * done with this page.
1120 if (block_in_file
< last_block
) {
1121 map
.m_lblk
= block_in_file
;
1122 map
.m_len
= last_block
- block_in_file
;
1124 if (f2fs_map_blocks(inode
, &map
, 0,
1125 F2FS_GET_BLOCK_READ
))
1126 goto set_error_page
;
1129 if ((map
.m_flags
& F2FS_MAP_MAPPED
)) {
1130 block_nr
= map
.m_pblk
+ block_in_file
- map
.m_lblk
;
1131 SetPageMappedToDisk(page
);
1133 if (!PageUptodate(page
) && !cleancache_get_page(page
)) {
1134 SetPageUptodate(page
);
1138 zero_user_segment(page
, 0, PAGE_SIZE
);
1139 if (!PageUptodate(page
))
1140 SetPageUptodate(page
);
1146 * This page will go to BIO. Do we need to send this
1149 if (bio
&& (last_block_in_bio
!= block_nr
- 1 ||
1150 !__same_bdev(F2FS_I_SB(inode
), block_nr
, bio
))) {
1152 __submit_bio(F2FS_I_SB(inode
), bio
, DATA
);
1156 bio
= f2fs_grab_bio(inode
, block_nr
, nr_pages
);
1159 goto set_error_page
;
1161 bio_set_op_attrs(bio
, REQ_OP_READ
, 0);
1164 if (bio_add_page(bio
, page
, blocksize
, 0) < blocksize
)
1165 goto submit_and_realloc
;
1167 last_block_in_bio
= block_nr
;
1171 zero_user_segment(page
, 0, PAGE_SIZE
);
1176 __submit_bio(F2FS_I_SB(inode
), bio
, DATA
);
1184 BUG_ON(pages
&& !list_empty(pages
));
1186 __submit_bio(F2FS_I_SB(inode
), bio
, DATA
);
1190 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
1192 struct inode
*inode
= page
->mapping
->host
;
1195 trace_f2fs_readpage(page
, DATA
);
1197 /* If the file has inline data, try to read it directly */
1198 if (f2fs_has_inline_data(inode
))
1199 ret
= f2fs_read_inline_data(inode
, page
);
1201 ret
= f2fs_mpage_readpages(page
->mapping
, NULL
, page
, 1);
1205 static int f2fs_read_data_pages(struct file
*file
,
1206 struct address_space
*mapping
,
1207 struct list_head
*pages
, unsigned nr_pages
)
1209 struct inode
*inode
= file
->f_mapping
->host
;
1210 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
1212 trace_f2fs_readpages(inode
, page
, nr_pages
);
1214 /* If the file has inline data, skip readpages */
1215 if (f2fs_has_inline_data(inode
))
1218 return f2fs_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
1221 int do_write_data_page(struct f2fs_io_info
*fio
)
1223 struct page
*page
= fio
->page
;
1224 struct inode
*inode
= page
->mapping
->host
;
1225 struct dnode_of_data dn
;
1228 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
1229 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
1233 fio
->old_blkaddr
= dn
.data_blkaddr
;
1235 /* This page is already truncated */
1236 if (fio
->old_blkaddr
== NULL_ADDR
) {
1237 ClearPageUptodate(page
);
1241 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1242 gfp_t gfp_flags
= GFP_NOFS
;
1244 /* wait for GCed encrypted page writeback */
1245 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode
),
1248 fio
->encrypted_page
= fscrypt_encrypt_page(inode
, fio
->page
,
1252 if (IS_ERR(fio
->encrypted_page
)) {
1253 err
= PTR_ERR(fio
->encrypted_page
);
1254 if (err
== -ENOMEM
) {
1255 /* flush pending ios and wait for a while */
1256 f2fs_flush_merged_bios(F2FS_I_SB(inode
));
1257 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1258 gfp_flags
|= __GFP_NOFAIL
;
1266 set_page_writeback(page
);
1269 * If current allocation needs SSR,
1270 * it had better in-place writes for updated data.
1272 if (unlikely(fio
->old_blkaddr
!= NEW_ADDR
&&
1273 !is_cold_data(page
) &&
1274 !IS_ATOMIC_WRITTEN_PAGE(page
) &&
1275 need_inplace_update(inode
))) {
1276 rewrite_data_page(fio
);
1277 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1278 trace_f2fs_do_write_data_page(page
, IPU
);
1280 write_data_page(&dn
, fio
);
1281 trace_f2fs_do_write_data_page(page
, OPU
);
1282 set_inode_flag(inode
, FI_APPEND_WRITE
);
1283 if (page
->index
== 0)
1284 set_inode_flag(inode
, FI_FIRST_BLOCK_WRITTEN
);
1287 f2fs_put_dnode(&dn
);
1291 static int f2fs_write_data_page(struct page
*page
,
1292 struct writeback_control
*wbc
)
1294 struct inode
*inode
= page
->mapping
->host
;
1295 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1296 loff_t i_size
= i_size_read(inode
);
1297 const pgoff_t end_index
= ((unsigned long long) i_size
)
1299 loff_t psize
= (page
->index
+ 1) << PAGE_SHIFT
;
1300 unsigned offset
= 0;
1301 bool need_balance_fs
= false;
1303 struct f2fs_io_info fio
= {
1307 .op_flags
= wbc_to_write_flags(wbc
),
1309 .encrypted_page
= NULL
,
1312 trace_f2fs_writepage(page
, DATA
);
1314 if (page
->index
< end_index
)
1318 * If the offset is out-of-range of file size,
1319 * this page does not have to be written to disk.
1321 offset
= i_size
& (PAGE_SIZE
- 1);
1322 if ((page
->index
>= end_index
+ 1) || !offset
)
1325 zero_user_segment(page
, offset
, PAGE_SIZE
);
1327 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1329 if (f2fs_is_drop_cache(inode
))
1331 /* we should not write 0'th page having journal header */
1332 if (f2fs_is_volatile_file(inode
) && (!page
->index
||
1333 (!wbc
->for_reclaim
&&
1334 available_free_memory(sbi
, BASE_CHECK
))))
1337 /* we should bypass data pages to proceed the kworkder jobs */
1338 if (unlikely(f2fs_cp_error(sbi
))) {
1339 mapping_set_error(page
->mapping
, -EIO
);
1343 /* Dentry blocks are controlled by checkpoint */
1344 if (S_ISDIR(inode
->i_mode
)) {
1345 err
= do_write_data_page(&fio
);
1349 if (!wbc
->for_reclaim
)
1350 need_balance_fs
= true;
1351 else if (has_not_enough_free_secs(sbi
, 0, 0))
1356 if (f2fs_has_inline_data(inode
))
1357 err
= f2fs_write_inline_data(inode
, page
);
1359 err
= do_write_data_page(&fio
);
1360 if (F2FS_I(inode
)->last_disk_size
< psize
)
1361 F2FS_I(inode
)->last_disk_size
= psize
;
1362 f2fs_unlock_op(sbi
);
1364 if (err
&& err
!= -ENOENT
)
1368 inode_dec_dirty_pages(inode
);
1370 ClearPageUptodate(page
);
1372 if (wbc
->for_reclaim
) {
1373 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, DATA
, WRITE
);
1374 remove_dirty_inode(inode
);
1378 f2fs_balance_fs(sbi
, need_balance_fs
);
1380 if (unlikely(f2fs_cp_error(sbi
)))
1381 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1386 redirty_page_for_writepage(wbc
, page
);
1388 return AOP_WRITEPAGE_ACTIVATE
;
1394 * This function was copied from write_cche_pages from mm/page-writeback.c.
1395 * The major change is making write step of cold data page separately from
1396 * warm/hot data page.
1398 static int f2fs_write_cache_pages(struct address_space
*mapping
,
1399 struct writeback_control
*wbc
)
1403 struct pagevec pvec
;
1405 pgoff_t
uninitialized_var(writeback_index
);
1407 pgoff_t end
; /* Inclusive */
1410 int range_whole
= 0;
1414 pagevec_init(&pvec
, 0);
1416 if (wbc
->range_cyclic
) {
1417 writeback_index
= mapping
->writeback_index
; /* prev offset */
1418 index
= writeback_index
;
1425 index
= wbc
->range_start
>> PAGE_SHIFT
;
1426 end
= wbc
->range_end
>> PAGE_SHIFT
;
1427 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1429 cycled
= 1; /* ignore range_cyclic tests */
1431 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1432 tag
= PAGECACHE_TAG_TOWRITE
;
1434 tag
= PAGECACHE_TAG_DIRTY
;
1436 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1437 tag_pages_for_writeback(mapping
, index
, end
);
1439 while (!done
&& (index
<= end
)) {
1442 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
1443 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1);
1447 for (i
= 0; i
< nr_pages
; i
++) {
1448 struct page
*page
= pvec
.pages
[i
];
1450 if (page
->index
> end
) {
1455 done_index
= page
->index
;
1459 if (unlikely(page
->mapping
!= mapping
)) {
1465 if (!PageDirty(page
)) {
1466 /* someone wrote it for us */
1467 goto continue_unlock
;
1470 if (PageWriteback(page
)) {
1471 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1472 f2fs_wait_on_page_writeback(page
,
1475 goto continue_unlock
;
1478 BUG_ON(PageWriteback(page
));
1479 if (!clear_page_dirty_for_io(page
))
1480 goto continue_unlock
;
1482 ret
= mapping
->a_ops
->writepage(page
, wbc
);
1483 if (unlikely(ret
)) {
1485 * keep nr_to_write, since vfs uses this to
1486 * get # of written pages.
1488 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
1493 done_index
= page
->index
+ 1;
1500 if (--wbc
->nr_to_write
<= 0 &&
1501 wbc
->sync_mode
== WB_SYNC_NONE
) {
1506 pagevec_release(&pvec
);
1510 if (!cycled
&& !done
) {
1513 end
= writeback_index
- 1;
1516 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1517 mapping
->writeback_index
= done_index
;
1520 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping
), mapping
->host
,
1521 NULL
, 0, DATA
, WRITE
);
1526 static int f2fs_write_data_pages(struct address_space
*mapping
,
1527 struct writeback_control
*wbc
)
1529 struct inode
*inode
= mapping
->host
;
1530 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1531 struct blk_plug plug
;
1534 /* deal with chardevs and other special file */
1535 if (!mapping
->a_ops
->writepage
)
1538 /* skip writing if there is no dirty page in this inode */
1539 if (!get_dirty_pages(inode
) && wbc
->sync_mode
== WB_SYNC_NONE
)
1542 if (S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_NONE
&&
1543 get_dirty_pages(inode
) < nr_pages_to_skip(sbi
, DATA
) &&
1544 available_free_memory(sbi
, DIRTY_DENTS
))
1547 /* skip writing during file defragment */
1548 if (is_inode_flag_set(inode
, FI_DO_DEFRAG
))
1551 /* during POR, we don't need to trigger writepage at all. */
1552 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1555 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1557 blk_start_plug(&plug
);
1558 ret
= f2fs_write_cache_pages(mapping
, wbc
);
1559 blk_finish_plug(&plug
);
1561 * if some pages were truncated, we cannot guarantee its mapping->host
1562 * to detect pending bios.
1565 remove_dirty_inode(inode
);
1569 wbc
->pages_skipped
+= get_dirty_pages(inode
);
1570 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1574 static void f2fs_write_failed(struct address_space
*mapping
, loff_t to
)
1576 struct inode
*inode
= mapping
->host
;
1577 loff_t i_size
= i_size_read(inode
);
1580 truncate_pagecache(inode
, i_size
);
1581 truncate_blocks(inode
, i_size
, true);
1585 static int prepare_write_begin(struct f2fs_sb_info
*sbi
,
1586 struct page
*page
, loff_t pos
, unsigned len
,
1587 block_t
*blk_addr
, bool *node_changed
)
1589 struct inode
*inode
= page
->mapping
->host
;
1590 pgoff_t index
= page
->index
;
1591 struct dnode_of_data dn
;
1593 bool locked
= false;
1594 struct extent_info ei
;
1598 * we already allocated all the blocks, so we don't need to get
1599 * the block addresses when there is no need to fill the page.
1601 if (!f2fs_has_inline_data(inode
) && len
== PAGE_SIZE
)
1604 if (f2fs_has_inline_data(inode
) ||
1605 (pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1610 /* check inline_data */
1611 ipage
= get_node_page(sbi
, inode
->i_ino
);
1612 if (IS_ERR(ipage
)) {
1613 err
= PTR_ERR(ipage
);
1617 set_new_dnode(&dn
, inode
, ipage
, ipage
, 0);
1619 if (f2fs_has_inline_data(inode
)) {
1620 if (pos
+ len
<= MAX_INLINE_DATA
) {
1621 read_inline_data(page
, ipage
);
1622 set_inode_flag(inode
, FI_DATA_EXIST
);
1624 set_inline_node(ipage
);
1626 err
= f2fs_convert_inline_page(&dn
, page
);
1629 if (dn
.data_blkaddr
== NULL_ADDR
)
1630 err
= f2fs_get_block(&dn
, index
);
1632 } else if (locked
) {
1633 err
= f2fs_get_block(&dn
, index
);
1635 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
1636 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
1639 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
1640 if (err
|| dn
.data_blkaddr
== NULL_ADDR
) {
1641 f2fs_put_dnode(&dn
);
1649 /* convert_inline_page can make node_changed */
1650 *blk_addr
= dn
.data_blkaddr
;
1651 *node_changed
= dn
.node_changed
;
1653 f2fs_put_dnode(&dn
);
1656 f2fs_unlock_op(sbi
);
1660 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
1661 loff_t pos
, unsigned len
, unsigned flags
,
1662 struct page
**pagep
, void **fsdata
)
1664 struct inode
*inode
= mapping
->host
;
1665 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1666 struct page
*page
= NULL
;
1667 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_SHIFT
;
1668 bool need_balance
= false;
1669 block_t blkaddr
= NULL_ADDR
;
1672 trace_f2fs_write_begin(inode
, pos
, len
, flags
);
1675 * We should check this at this moment to avoid deadlock on inode page
1676 * and #0 page. The locking rule for inline_data conversion should be:
1677 * lock_page(page #0) -> lock_page(inode_page)
1680 err
= f2fs_convert_inline_inode(inode
);
1685 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1693 err
= prepare_write_begin(sbi
, page
, pos
, len
,
1694 &blkaddr
, &need_balance
);
1698 if (need_balance
&& has_not_enough_free_secs(sbi
, 0, 0)) {
1700 f2fs_balance_fs(sbi
, true);
1702 if (page
->mapping
!= mapping
) {
1703 /* The page got truncated from under us */
1704 f2fs_put_page(page
, 1);
1709 f2fs_wait_on_page_writeback(page
, DATA
, false);
1711 /* wait for GCed encrypted page writeback */
1712 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1713 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
1715 if (len
== PAGE_SIZE
|| PageUptodate(page
))
1718 if (blkaddr
== NEW_ADDR
) {
1719 zero_user_segment(page
, 0, PAGE_SIZE
);
1720 SetPageUptodate(page
);
1724 bio
= f2fs_grab_bio(inode
, blkaddr
, 1);
1729 bio
->bi_opf
= REQ_OP_READ
;
1730 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
1736 __submit_bio(sbi
, bio
, DATA
);
1739 if (unlikely(page
->mapping
!= mapping
)) {
1740 f2fs_put_page(page
, 1);
1743 if (unlikely(!PageUptodate(page
))) {
1751 f2fs_put_page(page
, 1);
1752 f2fs_write_failed(mapping
, pos
+ len
);
1756 static int f2fs_write_end(struct file
*file
,
1757 struct address_space
*mapping
,
1758 loff_t pos
, unsigned len
, unsigned copied
,
1759 struct page
*page
, void *fsdata
)
1761 struct inode
*inode
= page
->mapping
->host
;
1763 trace_f2fs_write_end(inode
, pos
, len
, copied
);
1766 * This should be come from len == PAGE_SIZE, and we expect copied
1767 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1768 * let generic_perform_write() try to copy data again through copied=0.
1770 if (!PageUptodate(page
)) {
1771 if (unlikely(copied
!= PAGE_SIZE
))
1774 SetPageUptodate(page
);
1779 set_page_dirty(page
);
1781 if (pos
+ copied
> i_size_read(inode
))
1782 f2fs_i_size_write(inode
, pos
+ copied
);
1784 f2fs_put_page(page
, 1);
1785 f2fs_update_time(F2FS_I_SB(inode
), REQ_TIME
);
1789 static int check_direct_IO(struct inode
*inode
, struct iov_iter
*iter
,
1792 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
1794 if (offset
& blocksize_mask
)
1797 if (iov_iter_alignment(iter
) & blocksize_mask
)
1803 static ssize_t
f2fs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
1805 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
1806 struct inode
*inode
= mapping
->host
;
1807 size_t count
= iov_iter_count(iter
);
1808 loff_t offset
= iocb
->ki_pos
;
1809 int rw
= iov_iter_rw(iter
);
1812 err
= check_direct_IO(inode
, iter
, offset
);
1816 if (__force_buffered_io(inode
, rw
))
1819 trace_f2fs_direct_IO_enter(inode
, offset
, count
, rw
);
1821 down_read(&F2FS_I(inode
)->dio_rwsem
[rw
]);
1822 err
= blockdev_direct_IO(iocb
, inode
, iter
, get_data_block_dio
);
1823 up_read(&F2FS_I(inode
)->dio_rwsem
[rw
]);
1827 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1829 f2fs_write_failed(mapping
, offset
+ count
);
1832 trace_f2fs_direct_IO_exit(inode
, offset
, count
, rw
, err
);
1837 void f2fs_invalidate_page(struct page
*page
, unsigned int offset
,
1838 unsigned int length
)
1840 struct inode
*inode
= page
->mapping
->host
;
1841 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1843 if (inode
->i_ino
>= F2FS_ROOT_INO(sbi
) &&
1844 (offset
% PAGE_SIZE
|| length
!= PAGE_SIZE
))
1847 if (PageDirty(page
)) {
1848 if (inode
->i_ino
== F2FS_META_INO(sbi
)) {
1849 dec_page_count(sbi
, F2FS_DIRTY_META
);
1850 } else if (inode
->i_ino
== F2FS_NODE_INO(sbi
)) {
1851 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1853 inode_dec_dirty_pages(inode
);
1854 remove_dirty_inode(inode
);
1858 /* This is atomic written page, keep Private */
1859 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1862 set_page_private(page
, 0);
1863 ClearPagePrivate(page
);
1866 int f2fs_release_page(struct page
*page
, gfp_t wait
)
1868 /* If this is dirty page, keep PagePrivate */
1869 if (PageDirty(page
))
1872 /* This is atomic written page, keep Private */
1873 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1876 set_page_private(page
, 0);
1877 ClearPagePrivate(page
);
1882 * This was copied from __set_page_dirty_buffers which gives higher performance
1883 * in very high speed storages. (e.g., pmem)
1885 void f2fs_set_page_dirty_nobuffers(struct page
*page
)
1887 struct address_space
*mapping
= page
->mapping
;
1888 unsigned long flags
;
1890 if (unlikely(!mapping
))
1893 spin_lock(&mapping
->private_lock
);
1894 lock_page_memcg(page
);
1896 spin_unlock(&mapping
->private_lock
);
1898 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
1899 WARN_ON_ONCE(!PageUptodate(page
));
1900 account_page_dirtied(page
, mapping
);
1901 radix_tree_tag_set(&mapping
->page_tree
,
1902 page_index(page
), PAGECACHE_TAG_DIRTY
);
1903 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
1904 unlock_page_memcg(page
);
1906 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1910 static int f2fs_set_data_page_dirty(struct page
*page
)
1912 struct address_space
*mapping
= page
->mapping
;
1913 struct inode
*inode
= mapping
->host
;
1915 trace_f2fs_set_page_dirty(page
, DATA
);
1917 if (!PageUptodate(page
))
1918 SetPageUptodate(page
);
1920 if (f2fs_is_atomic_file(inode
)) {
1921 if (!IS_ATOMIC_WRITTEN_PAGE(page
)) {
1922 register_inmem_page(inode
, page
);
1926 * Previously, this page has been registered, we just
1932 if (!PageDirty(page
)) {
1933 f2fs_set_page_dirty_nobuffers(page
);
1934 update_dirty_page(inode
, page
);
1940 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
1942 struct inode
*inode
= mapping
->host
;
1944 if (f2fs_has_inline_data(inode
))
1947 /* make sure allocating whole blocks */
1948 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
1949 filemap_write_and_wait(mapping
);
1951 return generic_block_bmap(mapping
, block
, get_data_block_bmap
);
1954 #ifdef CONFIG_MIGRATION
1955 #include <linux/migrate.h>
1957 int f2fs_migrate_page(struct address_space
*mapping
,
1958 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
1960 int rc
, extra_count
;
1961 struct f2fs_inode_info
*fi
= F2FS_I(mapping
->host
);
1962 bool atomic_written
= IS_ATOMIC_WRITTEN_PAGE(page
);
1964 BUG_ON(PageWriteback(page
));
1966 /* migrating an atomic written page is safe with the inmem_lock hold */
1967 if (atomic_written
&& !mutex_trylock(&fi
->inmem_lock
))
1971 * A reference is expected if PagePrivate set when move mapping,
1972 * however F2FS breaks this for maintaining dirty page counts when
1973 * truncating pages. So here adjusting the 'extra_count' make it work.
1975 extra_count
= (atomic_written
? 1 : 0) - page_has_private(page
);
1976 rc
= migrate_page_move_mapping(mapping
, newpage
,
1977 page
, NULL
, mode
, extra_count
);
1978 if (rc
!= MIGRATEPAGE_SUCCESS
) {
1980 mutex_unlock(&fi
->inmem_lock
);
1984 if (atomic_written
) {
1985 struct inmem_pages
*cur
;
1986 list_for_each_entry(cur
, &fi
->inmem_pages
, list
)
1987 if (cur
->page
== page
) {
1988 cur
->page
= newpage
;
1991 mutex_unlock(&fi
->inmem_lock
);
1996 if (PagePrivate(page
))
1997 SetPagePrivate(newpage
);
1998 set_page_private(newpage
, page_private(page
));
2000 migrate_page_copy(newpage
, page
);
2002 return MIGRATEPAGE_SUCCESS
;
2006 const struct address_space_operations f2fs_dblock_aops
= {
2007 .readpage
= f2fs_read_data_page
,
2008 .readpages
= f2fs_read_data_pages
,
2009 .writepage
= f2fs_write_data_page
,
2010 .writepages
= f2fs_write_data_pages
,
2011 .write_begin
= f2fs_write_begin
,
2012 .write_end
= f2fs_write_end
,
2013 .set_page_dirty
= f2fs_set_data_page_dirty
,
2014 .invalidatepage
= f2fs_invalidate_page
,
2015 .releasepage
= f2fs_release_page
,
2016 .direct_IO
= f2fs_direct_IO
,
2018 #ifdef CONFIG_MIGRATION
2019 .migratepage
= f2fs_migrate_page
,