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 if (!PageUptodate(page
))
51 SetPageUptodate(page
);
53 ClearPageUptodate(page
);
61 static void f2fs_write_end_io(struct bio
*bio
)
63 struct f2fs_sb_info
*sbi
= bio
->bi_private
;
67 bio_for_each_segment_all(bvec
, bio
, i
) {
68 struct page
*page
= bvec
->bv_page
;
70 fscrypt_pullback_bio_page(&page
, true);
72 if (unlikely(bio
->bi_error
)) {
73 set_bit(AS_EIO
, &page
->mapping
->flags
);
74 f2fs_stop_checkpoint(sbi
, true);
76 end_page_writeback(page
);
78 if (atomic_dec_and_test(&sbi
->nr_wb_bios
) &&
79 wq_has_sleeper(&sbi
->cp_wait
))
80 wake_up(&sbi
->cp_wait
);
86 * Low-level block read/write IO operations.
88 static struct bio
*__bio_alloc(struct f2fs_sb_info
*sbi
, block_t blk_addr
,
89 int npages
, bool is_read
)
93 bio
= f2fs_bio_alloc(npages
);
95 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
96 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blk_addr
);
97 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
98 bio
->bi_private
= is_read
? NULL
: sbi
;
103 static inline void __submit_bio(struct f2fs_sb_info
*sbi
, int rw
,
104 struct bio
*bio
, enum page_type type
)
106 if (!is_read_io(rw
)) {
107 atomic_inc(&sbi
->nr_wb_bios
);
108 if (f2fs_sb_mounted_hmsmr(sbi
->sb
) &&
109 current
->plug
&& (type
== DATA
|| type
== NODE
))
110 blk_finish_plug(current
->plug
);
115 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
117 struct f2fs_io_info
*fio
= &io
->fio
;
122 if (is_read_io(fio
->rw
))
123 trace_f2fs_submit_read_bio(io
->sbi
->sb
, fio
, io
->bio
);
125 trace_f2fs_submit_write_bio(io
->sbi
->sb
, fio
, io
->bio
);
127 __submit_bio(io
->sbi
, fio
->rw
, io
->bio
, fio
->type
);
131 static bool __has_merged_page(struct f2fs_bio_info
*io
, struct inode
*inode
,
132 struct page
*page
, nid_t ino
)
134 struct bio_vec
*bvec
;
141 if (!inode
&& !page
&& !ino
)
144 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
146 if (bvec
->bv_page
->mapping
)
147 target
= bvec
->bv_page
;
149 target
= fscrypt_control_page(bvec
->bv_page
);
151 if (inode
&& inode
== target
->mapping
->host
)
153 if (page
&& page
== target
)
155 if (ino
&& ino
== ino_of_node(target
))
162 static bool has_merged_page(struct f2fs_sb_info
*sbi
, struct inode
*inode
,
163 struct page
*page
, nid_t ino
,
166 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
167 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
170 down_read(&io
->io_rwsem
);
171 ret
= __has_merged_page(io
, inode
, page
, ino
);
172 up_read(&io
->io_rwsem
);
176 static void __f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
177 struct inode
*inode
, struct page
*page
,
178 nid_t ino
, enum page_type type
, int rw
)
180 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
181 struct f2fs_bio_info
*io
;
183 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
185 down_write(&io
->io_rwsem
);
187 if (!__has_merged_page(io
, inode
, page
, ino
))
190 /* change META to META_FLUSH in the checkpoint procedure */
191 if (type
>= META_FLUSH
) {
192 io
->fio
.type
= META_FLUSH
;
193 if (test_opt(sbi
, NOBARRIER
))
194 io
->fio
.rw
= WRITE_FLUSH
| REQ_META
| REQ_PRIO
;
196 io
->fio
.rw
= WRITE_FLUSH_FUA
| REQ_META
| REQ_PRIO
;
198 __submit_merged_bio(io
);
200 up_write(&io
->io_rwsem
);
203 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
, enum page_type type
,
206 __f2fs_submit_merged_bio(sbi
, NULL
, NULL
, 0, type
, rw
);
209 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info
*sbi
,
210 struct inode
*inode
, struct page
*page
,
211 nid_t ino
, enum page_type type
, int rw
)
213 if (has_merged_page(sbi
, inode
, page
, ino
, type
))
214 __f2fs_submit_merged_bio(sbi
, inode
, page
, ino
, type
, rw
);
217 void f2fs_flush_merged_bios(struct f2fs_sb_info
*sbi
)
219 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
220 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
221 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
225 * Fill the locked page with data located in the block address.
226 * Return unlocked page.
228 int f2fs_submit_page_bio(struct f2fs_io_info
*fio
)
231 struct page
*page
= fio
->encrypted_page
?
232 fio
->encrypted_page
: fio
->page
;
234 trace_f2fs_submit_page_bio(page
, fio
);
235 f2fs_trace_ios(fio
, 0);
237 /* Allocate a new bio */
238 bio
= __bio_alloc(fio
->sbi
, fio
->new_blkaddr
, 1, is_read_io(fio
->rw
));
240 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
245 __submit_bio(fio
->sbi
, fio
->rw
, bio
, fio
->type
);
249 void f2fs_submit_page_mbio(struct f2fs_io_info
*fio
)
251 struct f2fs_sb_info
*sbi
= fio
->sbi
;
252 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
253 struct f2fs_bio_info
*io
;
254 bool is_read
= is_read_io(fio
->rw
);
255 struct page
*bio_page
;
257 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
259 if (fio
->old_blkaddr
!= NEW_ADDR
)
260 verify_block_addr(sbi
, fio
->old_blkaddr
);
261 verify_block_addr(sbi
, fio
->new_blkaddr
);
263 down_write(&io
->io_rwsem
);
265 if (io
->bio
&& (io
->last_block_in_bio
!= fio
->new_blkaddr
- 1 ||
266 io
->fio
.rw
!= fio
->rw
))
267 __submit_merged_bio(io
);
269 if (io
->bio
== NULL
) {
270 int bio_blocks
= MAX_BIO_BLOCKS(sbi
);
272 io
->bio
= __bio_alloc(sbi
, fio
->new_blkaddr
,
273 bio_blocks
, is_read
);
277 bio_page
= fio
->encrypted_page
? fio
->encrypted_page
: fio
->page
;
279 if (bio_add_page(io
->bio
, bio_page
, PAGE_SIZE
, 0) <
281 __submit_merged_bio(io
);
285 io
->last_block_in_bio
= fio
->new_blkaddr
;
286 f2fs_trace_ios(fio
, 0);
288 up_write(&io
->io_rwsem
);
289 trace_f2fs_submit_page_mbio(fio
->page
, fio
);
292 static void __set_data_blkaddr(struct dnode_of_data
*dn
)
294 struct f2fs_node
*rn
= F2FS_NODE(dn
->node_page
);
297 /* Get physical address of data block */
298 addr_array
= blkaddr_in_node(rn
);
299 addr_array
[dn
->ofs_in_node
] = cpu_to_le32(dn
->data_blkaddr
);
303 * Lock ordering for the change of data block address:
306 * update block addresses in the node page
308 void set_data_blkaddr(struct dnode_of_data
*dn
)
310 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
311 __set_data_blkaddr(dn
);
312 if (set_page_dirty(dn
->node_page
))
313 dn
->node_changed
= true;
316 void f2fs_update_data_blkaddr(struct dnode_of_data
*dn
, block_t blkaddr
)
318 dn
->data_blkaddr
= blkaddr
;
319 set_data_blkaddr(dn
);
320 f2fs_update_extent_cache(dn
);
323 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
324 int reserve_new_blocks(struct dnode_of_data
*dn
, blkcnt_t count
)
326 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
331 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
333 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
336 trace_f2fs_reserve_new_blocks(dn
->inode
, dn
->nid
,
337 dn
->ofs_in_node
, count
);
339 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
341 for (; count
> 0; dn
->ofs_in_node
++) {
343 datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
344 if (blkaddr
== NULL_ADDR
) {
345 dn
->data_blkaddr
= NEW_ADDR
;
346 __set_data_blkaddr(dn
);
351 if (set_page_dirty(dn
->node_page
))
352 dn
->node_changed
= true;
356 /* Should keep dn->ofs_in_node unchanged */
357 int reserve_new_block(struct dnode_of_data
*dn
)
359 unsigned int ofs_in_node
= dn
->ofs_in_node
;
362 ret
= reserve_new_blocks(dn
, 1);
363 dn
->ofs_in_node
= ofs_in_node
;
367 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
369 bool need_put
= dn
->inode_page
? false : true;
372 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
376 if (dn
->data_blkaddr
== NULL_ADDR
)
377 err
= reserve_new_block(dn
);
383 int f2fs_get_block(struct dnode_of_data
*dn
, pgoff_t index
)
385 struct extent_info ei
;
386 struct inode
*inode
= dn
->inode
;
388 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
389 dn
->data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
393 return f2fs_reserve_block(dn
, index
);
396 struct page
*get_read_data_page(struct inode
*inode
, pgoff_t index
,
397 int rw
, bool for_write
)
399 struct address_space
*mapping
= inode
->i_mapping
;
400 struct dnode_of_data dn
;
402 struct extent_info ei
;
404 struct f2fs_io_info fio
= {
405 .sbi
= F2FS_I_SB(inode
),
408 .encrypted_page
= NULL
,
411 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
412 return read_mapping_page(mapping
, index
, NULL
);
414 page
= f2fs_grab_cache_page(mapping
, index
, for_write
);
416 return ERR_PTR(-ENOMEM
);
418 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
419 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
423 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
424 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
429 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
434 if (PageUptodate(page
)) {
440 * A new dentry page is allocated but not able to be written, since its
441 * new inode page couldn't be allocated due to -ENOSPC.
442 * In such the case, its blkaddr can be remained as NEW_ADDR.
443 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
445 if (dn
.data_blkaddr
== NEW_ADDR
) {
446 zero_user_segment(page
, 0, PAGE_SIZE
);
447 if (!PageUptodate(page
))
448 SetPageUptodate(page
);
453 fio
.new_blkaddr
= fio
.old_blkaddr
= dn
.data_blkaddr
;
455 err
= f2fs_submit_page_bio(&fio
);
461 f2fs_put_page(page
, 1);
465 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
)
467 struct address_space
*mapping
= inode
->i_mapping
;
470 page
= find_get_page(mapping
, index
);
471 if (page
&& PageUptodate(page
))
473 f2fs_put_page(page
, 0);
475 page
= get_read_data_page(inode
, index
, READ_SYNC
, false);
479 if (PageUptodate(page
))
482 wait_on_page_locked(page
);
483 if (unlikely(!PageUptodate(page
))) {
484 f2fs_put_page(page
, 0);
485 return ERR_PTR(-EIO
);
491 * If it tries to access a hole, return an error.
492 * Because, the callers, functions in dir.c and GC, should be able to know
493 * whether this page exists or not.
495 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
,
498 struct address_space
*mapping
= inode
->i_mapping
;
501 page
= get_read_data_page(inode
, index
, READ_SYNC
, for_write
);
505 /* wait for read completion */
507 if (unlikely(page
->mapping
!= mapping
)) {
508 f2fs_put_page(page
, 1);
511 if (unlikely(!PageUptodate(page
))) {
512 f2fs_put_page(page
, 1);
513 return ERR_PTR(-EIO
);
519 * Caller ensures that this data page is never allocated.
520 * A new zero-filled data page is allocated in the page cache.
522 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
524 * Note that, ipage is set only by make_empty_dir, and if any error occur,
525 * ipage should be released by this function.
527 struct page
*get_new_data_page(struct inode
*inode
,
528 struct page
*ipage
, pgoff_t index
, bool new_i_size
)
530 struct address_space
*mapping
= inode
->i_mapping
;
532 struct dnode_of_data dn
;
535 page
= f2fs_grab_cache_page(mapping
, index
, true);
538 * before exiting, we should make sure ipage will be released
539 * if any error occur.
541 f2fs_put_page(ipage
, 1);
542 return ERR_PTR(-ENOMEM
);
545 set_new_dnode(&dn
, inode
, ipage
, NULL
, 0);
546 err
= f2fs_reserve_block(&dn
, index
);
548 f2fs_put_page(page
, 1);
554 if (PageUptodate(page
))
557 if (dn
.data_blkaddr
== NEW_ADDR
) {
558 zero_user_segment(page
, 0, PAGE_SIZE
);
559 if (!PageUptodate(page
))
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 f2fs_i_size_write(inode
, ((loff_t
)(index
+ 1) << PAGE_SHIFT
));
577 static int __allocate_data_block(struct dnode_of_data
*dn
)
579 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
580 struct f2fs_summary sum
;
582 int seg
= CURSEG_WARM_DATA
;
586 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
589 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
590 if (dn
->data_blkaddr
== NEW_ADDR
)
593 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
597 get_node_info(sbi
, dn
->nid
, &ni
);
598 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
600 if (dn
->ofs_in_node
== 0 && dn
->inode_page
== dn
->node_page
)
601 seg
= CURSEG_DIRECT_IO
;
603 allocate_data_block(sbi
, NULL
, dn
->data_blkaddr
, &dn
->data_blkaddr
,
605 set_data_blkaddr(dn
);
608 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), dn
->inode
) +
610 if (i_size_read(dn
->inode
) < ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
))
611 f2fs_i_size_write(dn
->inode
,
612 ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
));
616 ssize_t
f2fs_preallocate_blocks(struct kiocb
*iocb
, struct iov_iter
*from
)
618 struct inode
*inode
= file_inode(iocb
->ki_filp
);
619 struct f2fs_map_blocks map
;
622 map
.m_lblk
= F2FS_BLK_ALIGN(iocb
->ki_pos
);
623 map
.m_len
= F2FS_BYTES_TO_BLK(iov_iter_count(from
));
624 map
.m_next_pgofs
= NULL
;
626 if (f2fs_encrypted_inode(inode
))
629 if (iocb
->ki_flags
& IOCB_DIRECT
) {
630 ret
= f2fs_convert_inline_inode(inode
);
633 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_DIO
);
635 if (iocb
->ki_pos
+ iov_iter_count(from
) > MAX_INLINE_DATA
) {
636 ret
= f2fs_convert_inline_inode(inode
);
640 if (!f2fs_has_inline_data(inode
))
641 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_AIO
);
646 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
647 * f2fs_map_blocks structure.
648 * If original data blocks are allocated, then give them to blockdev.
650 * a. preallocate requested block addresses
651 * b. do not use extent cache for better performance
652 * c. give the block addresses to blockdev
654 int f2fs_map_blocks(struct inode
*inode
, struct f2fs_map_blocks
*map
,
655 int create
, int flag
)
657 unsigned int maxblocks
= map
->m_len
;
658 struct dnode_of_data dn
;
659 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
660 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE
;
661 pgoff_t pgofs
, end_offset
, end
;
662 int err
= 0, ofs
= 1;
663 unsigned int ofs_in_node
, last_ofs_in_node
;
665 struct extent_info ei
;
666 bool allocated
= false;
672 /* it only supports block size == page size */
673 pgofs
= (pgoff_t
)map
->m_lblk
;
674 end
= pgofs
+ maxblocks
;
676 if (!create
&& f2fs_lookup_extent_cache(inode
, pgofs
, &ei
)) {
677 map
->m_pblk
= ei
.blk
+ pgofs
- ei
.fofs
;
678 map
->m_len
= min((pgoff_t
)maxblocks
, ei
.fofs
+ ei
.len
- pgofs
);
679 map
->m_flags
= F2FS_MAP_MAPPED
;
687 /* When reading holes, we need its node page */
688 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
689 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
691 if (flag
== F2FS_GET_BLOCK_BMAP
)
693 if (err
== -ENOENT
) {
695 if (map
->m_next_pgofs
)
697 get_next_page_offset(&dn
, pgofs
);
703 ofs_in_node
= dn
.ofs_in_node
;
704 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, inode
);
707 blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
709 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
) {
711 if (unlikely(f2fs_cp_error(sbi
))) {
715 if (flag
== F2FS_GET_BLOCK_PRE_AIO
) {
716 if (blkaddr
== NULL_ADDR
) {
718 last_ofs_in_node
= dn
.ofs_in_node
;
721 err
= __allocate_data_block(&dn
);
723 set_inode_flag(inode
, FI_APPEND_WRITE
);
729 map
->m_flags
= F2FS_MAP_NEW
;
730 blkaddr
= dn
.data_blkaddr
;
732 if (flag
== F2FS_GET_BLOCK_BMAP
) {
736 if (flag
== F2FS_GET_BLOCK_FIEMAP
&&
737 blkaddr
== NULL_ADDR
) {
738 if (map
->m_next_pgofs
)
739 *map
->m_next_pgofs
= pgofs
+ 1;
741 if (flag
!= F2FS_GET_BLOCK_FIEMAP
||
747 if (flag
== F2FS_GET_BLOCK_PRE_AIO
)
750 if (map
->m_len
== 0) {
751 /* preallocated unwritten block should be mapped for fiemap. */
752 if (blkaddr
== NEW_ADDR
)
753 map
->m_flags
|= F2FS_MAP_UNWRITTEN
;
754 map
->m_flags
|= F2FS_MAP_MAPPED
;
756 map
->m_pblk
= blkaddr
;
758 } else if ((map
->m_pblk
!= NEW_ADDR
&&
759 blkaddr
== (map
->m_pblk
+ ofs
)) ||
760 (map
->m_pblk
== NEW_ADDR
&& blkaddr
== NEW_ADDR
) ||
761 flag
== F2FS_GET_BLOCK_PRE_DIO
) {
772 /* preallocate blocks in batch for one dnode page */
773 if (flag
== F2FS_GET_BLOCK_PRE_AIO
&&
774 (pgofs
== end
|| dn
.ofs_in_node
== end_offset
)) {
776 dn
.ofs_in_node
= ofs_in_node
;
777 err
= reserve_new_blocks(&dn
, prealloc
);
781 map
->m_len
+= dn
.ofs_in_node
- ofs_in_node
;
782 if (prealloc
&& dn
.ofs_in_node
!= last_ofs_in_node
+ 1) {
786 dn
.ofs_in_node
= end_offset
;
791 else if (dn
.ofs_in_node
< end_offset
)
798 f2fs_balance_fs(sbi
, allocated
);
808 f2fs_balance_fs(sbi
, allocated
);
811 trace_f2fs_map_blocks(inode
, map
, err
);
815 static int __get_data_block(struct inode
*inode
, sector_t iblock
,
816 struct buffer_head
*bh
, int create
, int flag
,
819 struct f2fs_map_blocks map
;
823 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
824 map
.m_next_pgofs
= next_pgofs
;
826 ret
= f2fs_map_blocks(inode
, &map
, create
, flag
);
828 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
829 bh
->b_state
= (bh
->b_state
& ~F2FS_MAP_FLAGS
) | map
.m_flags
;
830 bh
->b_size
= map
.m_len
<< inode
->i_blkbits
;
835 static int get_data_block(struct inode
*inode
, sector_t iblock
,
836 struct buffer_head
*bh_result
, int create
, int flag
,
839 return __get_data_block(inode
, iblock
, bh_result
, create
,
843 static int get_data_block_dio(struct inode
*inode
, sector_t iblock
,
844 struct buffer_head
*bh_result
, int create
)
846 return __get_data_block(inode
, iblock
, bh_result
, create
,
847 F2FS_GET_BLOCK_DIO
, NULL
);
850 static int get_data_block_bmap(struct inode
*inode
, sector_t iblock
,
851 struct buffer_head
*bh_result
, int create
)
853 /* Block number less than F2FS MAX BLOCKS */
854 if (unlikely(iblock
>= F2FS_I_SB(inode
)->max_file_blocks
))
857 return __get_data_block(inode
, iblock
, bh_result
, create
,
858 F2FS_GET_BLOCK_BMAP
, NULL
);
861 static inline sector_t
logical_to_blk(struct inode
*inode
, loff_t offset
)
863 return (offset
>> inode
->i_blkbits
);
866 static inline loff_t
blk_to_logical(struct inode
*inode
, sector_t blk
)
868 return (blk
<< inode
->i_blkbits
);
871 int f2fs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
874 struct buffer_head map_bh
;
875 sector_t start_blk
, last_blk
;
878 u64 logical
= 0, phys
= 0, size
= 0;
882 ret
= fiemap_check_flags(fieinfo
, FIEMAP_FLAG_SYNC
);
886 if (f2fs_has_inline_data(inode
)) {
887 ret
= f2fs_inline_data_fiemap(inode
, fieinfo
, start
, len
);
894 isize
= i_size_read(inode
);
898 if (start
+ len
> isize
)
901 if (logical_to_blk(inode
, len
) == 0)
902 len
= blk_to_logical(inode
, 1);
904 start_blk
= logical_to_blk(inode
, start
);
905 last_blk
= logical_to_blk(inode
, start
+ len
- 1);
908 memset(&map_bh
, 0, sizeof(struct buffer_head
));
911 ret
= get_data_block(inode
, start_blk
, &map_bh
, 0,
912 F2FS_GET_BLOCK_FIEMAP
, &next_pgofs
);
917 if (!buffer_mapped(&map_bh
)) {
918 start_blk
= next_pgofs
;
919 /* Go through holes util pass the EOF */
920 if (blk_to_logical(inode
, start_blk
) < isize
)
922 /* Found a hole beyond isize means no more extents.
923 * Note that the premise is that filesystems don't
924 * punch holes beyond isize and keep size unchanged.
926 flags
|= FIEMAP_EXTENT_LAST
;
930 if (f2fs_encrypted_inode(inode
))
931 flags
|= FIEMAP_EXTENT_DATA_ENCRYPTED
;
933 ret
= fiemap_fill_next_extent(fieinfo
, logical
,
937 if (start_blk
> last_blk
|| ret
)
940 logical
= blk_to_logical(inode
, start_blk
);
941 phys
= blk_to_logical(inode
, map_bh
.b_blocknr
);
942 size
= map_bh
.b_size
;
944 if (buffer_unwritten(&map_bh
))
945 flags
= FIEMAP_EXTENT_UNWRITTEN
;
947 start_blk
+= logical_to_blk(inode
, size
);
951 if (fatal_signal_pending(current
))
963 struct bio
*f2fs_grab_bio(struct inode
*inode
, block_t blkaddr
,
966 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
967 struct fscrypt_ctx
*ctx
= NULL
;
968 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
971 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
972 ctx
= fscrypt_get_ctx(inode
, GFP_NOFS
);
974 return ERR_CAST(ctx
);
976 /* wait the page to be moved by cleaning */
977 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
980 bio
= bio_alloc(GFP_KERNEL
, min_t(int, nr_pages
, BIO_MAX_PAGES
));
983 fscrypt_release_ctx(ctx
);
984 return ERR_PTR(-ENOMEM
);
987 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blkaddr
);
988 bio
->bi_end_io
= f2fs_read_end_io
;
989 bio
->bi_private
= ctx
;
995 * This function was originally taken from fs/mpage.c, and customized for f2fs.
996 * Major change was from block_size == page_size in f2fs by default.
998 static int f2fs_mpage_readpages(struct address_space
*mapping
,
999 struct list_head
*pages
, struct page
*page
,
1002 struct bio
*bio
= NULL
;
1004 sector_t last_block_in_bio
= 0;
1005 struct inode
*inode
= mapping
->host
;
1006 const unsigned blkbits
= inode
->i_blkbits
;
1007 const unsigned blocksize
= 1 << blkbits
;
1008 sector_t block_in_file
;
1009 sector_t last_block
;
1010 sector_t last_block_in_file
;
1012 struct f2fs_map_blocks map
;
1018 map
.m_next_pgofs
= NULL
;
1020 for (page_idx
= 0; nr_pages
; page_idx
++, nr_pages
--) {
1022 prefetchw(&page
->flags
);
1024 page
= list_entry(pages
->prev
, struct page
, lru
);
1025 list_del(&page
->lru
);
1026 if (add_to_page_cache_lru(page
, mapping
,
1027 page
->index
, GFP_KERNEL
))
1031 block_in_file
= (sector_t
)page
->index
;
1032 last_block
= block_in_file
+ nr_pages
;
1033 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >>
1035 if (last_block
> last_block_in_file
)
1036 last_block
= last_block_in_file
;
1039 * Map blocks using the previous result first.
1041 if ((map
.m_flags
& F2FS_MAP_MAPPED
) &&
1042 block_in_file
> map
.m_lblk
&&
1043 block_in_file
< (map
.m_lblk
+ map
.m_len
))
1047 * Then do more f2fs_map_blocks() calls until we are
1048 * done with this page.
1052 if (block_in_file
< last_block
) {
1053 map
.m_lblk
= block_in_file
;
1054 map
.m_len
= last_block
- block_in_file
;
1056 if (f2fs_map_blocks(inode
, &map
, 0,
1057 F2FS_GET_BLOCK_READ
))
1058 goto set_error_page
;
1061 if ((map
.m_flags
& F2FS_MAP_MAPPED
)) {
1062 block_nr
= map
.m_pblk
+ block_in_file
- map
.m_lblk
;
1063 SetPageMappedToDisk(page
);
1065 if (!PageUptodate(page
) && !cleancache_get_page(page
)) {
1066 SetPageUptodate(page
);
1070 zero_user_segment(page
, 0, PAGE_SIZE
);
1071 if (!PageUptodate(page
))
1072 SetPageUptodate(page
);
1078 * This page will go to BIO. Do we need to send this
1081 if (bio
&& (last_block_in_bio
!= block_nr
- 1)) {
1083 __submit_bio(F2FS_I_SB(inode
), READ
, bio
, DATA
);
1087 bio
= f2fs_grab_bio(inode
, block_nr
, nr_pages
);
1090 goto set_error_page
;
1094 if (bio_add_page(bio
, page
, blocksize
, 0) < blocksize
)
1095 goto submit_and_realloc
;
1097 last_block_in_bio
= block_nr
;
1101 zero_user_segment(page
, 0, PAGE_SIZE
);
1106 __submit_bio(F2FS_I_SB(inode
), READ
, bio
, DATA
);
1114 BUG_ON(pages
&& !list_empty(pages
));
1116 __submit_bio(F2FS_I_SB(inode
), READ
, bio
, DATA
);
1120 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
1122 struct inode
*inode
= page
->mapping
->host
;
1125 trace_f2fs_readpage(page
, DATA
);
1127 /* If the file has inline data, try to read it directly */
1128 if (f2fs_has_inline_data(inode
))
1129 ret
= f2fs_read_inline_data(inode
, page
);
1131 ret
= f2fs_mpage_readpages(page
->mapping
, NULL
, page
, 1);
1135 static int f2fs_read_data_pages(struct file
*file
,
1136 struct address_space
*mapping
,
1137 struct list_head
*pages
, unsigned nr_pages
)
1139 struct inode
*inode
= file
->f_mapping
->host
;
1140 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
1142 trace_f2fs_readpages(inode
, page
, nr_pages
);
1144 /* If the file has inline data, skip readpages */
1145 if (f2fs_has_inline_data(inode
))
1148 return f2fs_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
1151 int do_write_data_page(struct f2fs_io_info
*fio
)
1153 struct page
*page
= fio
->page
;
1154 struct inode
*inode
= page
->mapping
->host
;
1155 struct dnode_of_data dn
;
1158 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
1159 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
1163 fio
->old_blkaddr
= dn
.data_blkaddr
;
1165 /* This page is already truncated */
1166 if (fio
->old_blkaddr
== NULL_ADDR
) {
1167 ClearPageUptodate(page
);
1171 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1172 gfp_t gfp_flags
= GFP_NOFS
;
1174 /* wait for GCed encrypted page writeback */
1175 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode
),
1178 fio
->encrypted_page
= fscrypt_encrypt_page(inode
, fio
->page
,
1180 if (IS_ERR(fio
->encrypted_page
)) {
1181 err
= PTR_ERR(fio
->encrypted_page
);
1182 if (err
== -ENOMEM
) {
1183 /* flush pending ios and wait for a while */
1184 f2fs_flush_merged_bios(F2FS_I_SB(inode
));
1185 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1186 gfp_flags
|= __GFP_NOFAIL
;
1194 set_page_writeback(page
);
1197 * If current allocation needs SSR,
1198 * it had better in-place writes for updated data.
1200 if (unlikely(fio
->old_blkaddr
!= NEW_ADDR
&&
1201 !is_cold_data(page
) &&
1202 !IS_ATOMIC_WRITTEN_PAGE(page
) &&
1203 need_inplace_update(inode
))) {
1204 rewrite_data_page(fio
);
1205 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1206 trace_f2fs_do_write_data_page(page
, IPU
);
1208 write_data_page(&dn
, fio
);
1209 trace_f2fs_do_write_data_page(page
, OPU
);
1210 set_inode_flag(inode
, FI_APPEND_WRITE
);
1211 if (page
->index
== 0)
1212 set_inode_flag(inode
, FI_FIRST_BLOCK_WRITTEN
);
1215 f2fs_put_dnode(&dn
);
1219 static int f2fs_write_data_page(struct page
*page
,
1220 struct writeback_control
*wbc
)
1222 struct inode
*inode
= page
->mapping
->host
;
1223 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1224 loff_t i_size
= i_size_read(inode
);
1225 const pgoff_t end_index
= ((unsigned long long) i_size
)
1227 loff_t psize
= (page
->index
+ 1) << PAGE_SHIFT
;
1228 unsigned offset
= 0;
1229 bool need_balance_fs
= false;
1231 struct f2fs_io_info fio
= {
1234 .rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
,
1236 .encrypted_page
= NULL
,
1239 trace_f2fs_writepage(page
, DATA
);
1241 if (page
->index
< end_index
)
1245 * If the offset is out-of-range of file size,
1246 * this page does not have to be written to disk.
1248 offset
= i_size
& (PAGE_SIZE
- 1);
1249 if ((page
->index
>= end_index
+ 1) || !offset
)
1252 zero_user_segment(page
, offset
, PAGE_SIZE
);
1254 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1256 if (f2fs_is_drop_cache(inode
))
1258 /* we should not write 0'th page having journal header */
1259 if (f2fs_is_volatile_file(inode
) && (!page
->index
||
1260 (!wbc
->for_reclaim
&&
1261 available_free_memory(sbi
, BASE_CHECK
))))
1264 /* we should bypass data pages to proceed the kworkder jobs */
1265 if (unlikely(f2fs_cp_error(sbi
))) {
1266 mapping_set_error(page
->mapping
, -EIO
);
1270 /* Dentry blocks are controlled by checkpoint */
1271 if (S_ISDIR(inode
->i_mode
)) {
1272 err
= do_write_data_page(&fio
);
1276 if (!wbc
->for_reclaim
)
1277 need_balance_fs
= true;
1278 else if (has_not_enough_free_secs(sbi
, 0))
1283 if (f2fs_has_inline_data(inode
))
1284 err
= f2fs_write_inline_data(inode
, page
);
1286 err
= do_write_data_page(&fio
);
1287 if (F2FS_I(inode
)->last_disk_size
< psize
)
1288 F2FS_I(inode
)->last_disk_size
= psize
;
1289 f2fs_unlock_op(sbi
);
1291 if (err
&& err
!= -ENOENT
)
1294 clear_cold_data(page
);
1296 inode_dec_dirty_pages(inode
);
1298 ClearPageUptodate(page
);
1300 if (wbc
->for_reclaim
) {
1301 f2fs_submit_merged_bio_cond(sbi
, NULL
, page
, 0, DATA
, WRITE
);
1302 remove_dirty_inode(inode
);
1306 f2fs_balance_fs(sbi
, need_balance_fs
);
1308 if (unlikely(f2fs_cp_error(sbi
)))
1309 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1314 redirty_page_for_writepage(wbc
, page
);
1320 * This function was copied from write_cche_pages from mm/page-writeback.c.
1321 * The major change is making write step of cold data page separately from
1322 * warm/hot data page.
1324 static int f2fs_write_cache_pages(struct address_space
*mapping
,
1325 struct writeback_control
*wbc
)
1329 struct pagevec pvec
;
1331 pgoff_t
uninitialized_var(writeback_index
);
1333 pgoff_t end
; /* Inclusive */
1336 int range_whole
= 0;
1339 pagevec_init(&pvec
, 0);
1341 if (wbc
->range_cyclic
) {
1342 writeback_index
= mapping
->writeback_index
; /* prev offset */
1343 index
= writeback_index
;
1350 index
= wbc
->range_start
>> PAGE_SHIFT
;
1351 end
= wbc
->range_end
>> PAGE_SHIFT
;
1352 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1354 cycled
= 1; /* ignore range_cyclic tests */
1356 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1357 tag
= PAGECACHE_TAG_TOWRITE
;
1359 tag
= PAGECACHE_TAG_DIRTY
;
1361 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1362 tag_pages_for_writeback(mapping
, index
, end
);
1364 while (!done
&& (index
<= end
)) {
1367 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
1368 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1);
1372 for (i
= 0; i
< nr_pages
; i
++) {
1373 struct page
*page
= pvec
.pages
[i
];
1375 if (page
->index
> end
) {
1380 done_index
= page
->index
;
1384 if (unlikely(page
->mapping
!= mapping
)) {
1390 if (!PageDirty(page
)) {
1391 /* someone wrote it for us */
1392 goto continue_unlock
;
1395 if (PageWriteback(page
)) {
1396 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1397 f2fs_wait_on_page_writeback(page
,
1400 goto continue_unlock
;
1403 BUG_ON(PageWriteback(page
));
1404 if (!clear_page_dirty_for_io(page
))
1405 goto continue_unlock
;
1407 ret
= mapping
->a_ops
->writepage(page
, wbc
);
1408 if (unlikely(ret
)) {
1409 done_index
= page
->index
+ 1;
1414 if (--wbc
->nr_to_write
<= 0 &&
1415 wbc
->sync_mode
== WB_SYNC_NONE
) {
1420 pagevec_release(&pvec
);
1424 if (!cycled
&& !done
) {
1427 end
= writeback_index
- 1;
1430 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1431 mapping
->writeback_index
= done_index
;
1436 static int f2fs_write_data_pages(struct address_space
*mapping
,
1437 struct writeback_control
*wbc
)
1439 struct inode
*inode
= mapping
->host
;
1440 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1441 struct blk_plug plug
;
1444 /* deal with chardevs and other special file */
1445 if (!mapping
->a_ops
->writepage
)
1448 /* skip writing if there is no dirty page in this inode */
1449 if (!get_dirty_pages(inode
) && wbc
->sync_mode
== WB_SYNC_NONE
)
1452 if (S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_NONE
&&
1453 get_dirty_pages(inode
) < nr_pages_to_skip(sbi
, DATA
) &&
1454 available_free_memory(sbi
, DIRTY_DENTS
))
1457 /* skip writing during file defragment */
1458 if (is_inode_flag_set(inode
, FI_DO_DEFRAG
))
1461 /* during POR, we don't need to trigger writepage at all. */
1462 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1465 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1467 blk_start_plug(&plug
);
1468 ret
= f2fs_write_cache_pages(mapping
, wbc
);
1469 blk_finish_plug(&plug
);
1471 * if some pages were truncated, we cannot guarantee its mapping->host
1472 * to detect pending bios.
1474 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1476 remove_dirty_inode(inode
);
1480 wbc
->pages_skipped
+= get_dirty_pages(inode
);
1481 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1485 static void f2fs_write_failed(struct address_space
*mapping
, loff_t to
)
1487 struct inode
*inode
= mapping
->host
;
1488 loff_t i_size
= i_size_read(inode
);
1491 truncate_pagecache(inode
, i_size
);
1492 truncate_blocks(inode
, i_size
, true);
1496 static int prepare_write_begin(struct f2fs_sb_info
*sbi
,
1497 struct page
*page
, loff_t pos
, unsigned len
,
1498 block_t
*blk_addr
, bool *node_changed
)
1500 struct inode
*inode
= page
->mapping
->host
;
1501 pgoff_t index
= page
->index
;
1502 struct dnode_of_data dn
;
1504 bool locked
= false;
1505 struct extent_info ei
;
1509 * we already allocated all the blocks, so we don't need to get
1510 * the block addresses when there is no need to fill the page.
1512 if (!f2fs_has_inline_data(inode
) && !f2fs_encrypted_inode(inode
) &&
1516 if (f2fs_has_inline_data(inode
) ||
1517 (pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1522 /* check inline_data */
1523 ipage
= get_node_page(sbi
, inode
->i_ino
);
1524 if (IS_ERR(ipage
)) {
1525 err
= PTR_ERR(ipage
);
1529 set_new_dnode(&dn
, inode
, ipage
, ipage
, 0);
1531 if (f2fs_has_inline_data(inode
)) {
1532 if (pos
+ len
<= MAX_INLINE_DATA
) {
1533 read_inline_data(page
, ipage
);
1534 set_inode_flag(inode
, FI_DATA_EXIST
);
1536 set_inline_node(ipage
);
1538 err
= f2fs_convert_inline_page(&dn
, page
);
1541 if (dn
.data_blkaddr
== NULL_ADDR
)
1542 err
= f2fs_get_block(&dn
, index
);
1544 } else if (locked
) {
1545 err
= f2fs_get_block(&dn
, index
);
1547 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
1548 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
1551 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
1552 if (err
|| dn
.data_blkaddr
== NULL_ADDR
) {
1553 f2fs_put_dnode(&dn
);
1561 /* convert_inline_page can make node_changed */
1562 *blk_addr
= dn
.data_blkaddr
;
1563 *node_changed
= dn
.node_changed
;
1565 f2fs_put_dnode(&dn
);
1568 f2fs_unlock_op(sbi
);
1572 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
1573 loff_t pos
, unsigned len
, unsigned flags
,
1574 struct page
**pagep
, void **fsdata
)
1576 struct inode
*inode
= mapping
->host
;
1577 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1578 struct page
*page
= NULL
;
1579 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_SHIFT
;
1580 bool need_balance
= false;
1581 block_t blkaddr
= NULL_ADDR
;
1584 trace_f2fs_write_begin(inode
, pos
, len
, flags
);
1587 * We should check this at this moment to avoid deadlock on inode page
1588 * and #0 page. The locking rule for inline_data conversion should be:
1589 * lock_page(page #0) -> lock_page(inode_page)
1592 err
= f2fs_convert_inline_inode(inode
);
1597 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1605 err
= prepare_write_begin(sbi
, page
, pos
, len
,
1606 &blkaddr
, &need_balance
);
1610 if (need_balance
&& has_not_enough_free_secs(sbi
, 0)) {
1612 f2fs_balance_fs(sbi
, true);
1614 if (page
->mapping
!= mapping
) {
1615 /* The page got truncated from under us */
1616 f2fs_put_page(page
, 1);
1621 f2fs_wait_on_page_writeback(page
, DATA
, false);
1623 /* wait for GCed encrypted page writeback */
1624 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1625 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
1627 if (len
== PAGE_SIZE
)
1629 if (PageUptodate(page
))
1632 if ((pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1633 unsigned start
= pos
& (PAGE_SIZE
- 1);
1634 unsigned end
= start
+ len
;
1636 /* Reading beyond i_size is simple: memset to zero */
1637 zero_user_segments(page
, 0, start
, end
, PAGE_SIZE
);
1641 if (blkaddr
== NEW_ADDR
) {
1642 zero_user_segment(page
, 0, PAGE_SIZE
);
1646 bio
= f2fs_grab_bio(inode
, blkaddr
, 1);
1652 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
1658 __submit_bio(sbi
, READ_SYNC
, bio
, DATA
);
1661 if (unlikely(page
->mapping
!= mapping
)) {
1662 f2fs_put_page(page
, 1);
1665 if (unlikely(!PageUptodate(page
))) {
1671 if (!PageUptodate(page
))
1672 SetPageUptodate(page
);
1674 clear_cold_data(page
);
1678 f2fs_put_page(page
, 1);
1679 f2fs_write_failed(mapping
, pos
+ len
);
1683 static int f2fs_write_end(struct file
*file
,
1684 struct address_space
*mapping
,
1685 loff_t pos
, unsigned len
, unsigned copied
,
1686 struct page
*page
, void *fsdata
)
1688 struct inode
*inode
= page
->mapping
->host
;
1690 trace_f2fs_write_end(inode
, pos
, len
, copied
);
1692 set_page_dirty(page
);
1693 f2fs_put_page(page
, 1);
1695 if (pos
+ copied
> i_size_read(inode
))
1696 f2fs_i_size_write(inode
, pos
+ copied
);
1698 f2fs_update_time(F2FS_I_SB(inode
), REQ_TIME
);
1702 static int check_direct_IO(struct inode
*inode
, struct iov_iter
*iter
,
1705 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
1707 if (offset
& blocksize_mask
)
1710 if (iov_iter_alignment(iter
) & blocksize_mask
)
1716 static ssize_t
f2fs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
1718 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
1719 struct inode
*inode
= mapping
->host
;
1720 size_t count
= iov_iter_count(iter
);
1721 loff_t offset
= iocb
->ki_pos
;
1722 int rw
= iov_iter_rw(iter
);
1725 err
= check_direct_IO(inode
, iter
, offset
);
1729 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1731 if (test_opt(F2FS_I_SB(inode
), LFS
))
1734 trace_f2fs_direct_IO_enter(inode
, offset
, count
, rw
);
1736 down_read(&F2FS_I(inode
)->dio_rwsem
[rw
]);
1737 err
= blockdev_direct_IO(iocb
, inode
, iter
, get_data_block_dio
);
1738 up_read(&F2FS_I(inode
)->dio_rwsem
[rw
]);
1742 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1744 f2fs_write_failed(mapping
, offset
+ count
);
1747 trace_f2fs_direct_IO_exit(inode
, offset
, count
, rw
, err
);
1752 void f2fs_invalidate_page(struct page
*page
, unsigned int offset
,
1753 unsigned int length
)
1755 struct inode
*inode
= page
->mapping
->host
;
1756 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1758 if (inode
->i_ino
>= F2FS_ROOT_INO(sbi
) &&
1759 (offset
% PAGE_SIZE
|| length
!= PAGE_SIZE
))
1762 if (PageDirty(page
)) {
1763 if (inode
->i_ino
== F2FS_META_INO(sbi
))
1764 dec_page_count(sbi
, F2FS_DIRTY_META
);
1765 else if (inode
->i_ino
== F2FS_NODE_INO(sbi
))
1766 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1768 inode_dec_dirty_pages(inode
);
1771 /* This is atomic written page, keep Private */
1772 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1775 set_page_private(page
, 0);
1776 ClearPagePrivate(page
);
1779 int f2fs_release_page(struct page
*page
, gfp_t wait
)
1781 /* If this is dirty page, keep PagePrivate */
1782 if (PageDirty(page
))
1785 /* This is atomic written page, keep Private */
1786 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1789 set_page_private(page
, 0);
1790 ClearPagePrivate(page
);
1795 * This was copied from __set_page_dirty_buffers which gives higher performance
1796 * in very high speed storages. (e.g., pmem)
1798 void f2fs_set_page_dirty_nobuffers(struct page
*page
)
1800 struct address_space
*mapping
= page
->mapping
;
1801 unsigned long flags
;
1803 if (unlikely(!mapping
))
1806 spin_lock(&mapping
->private_lock
);
1807 lock_page_memcg(page
);
1809 spin_unlock(&mapping
->private_lock
);
1811 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
1812 WARN_ON_ONCE(!PageUptodate(page
));
1813 account_page_dirtied(page
, mapping
);
1814 radix_tree_tag_set(&mapping
->page_tree
,
1815 page_index(page
), PAGECACHE_TAG_DIRTY
);
1816 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
1817 unlock_page_memcg(page
);
1819 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1823 static int f2fs_set_data_page_dirty(struct page
*page
)
1825 struct address_space
*mapping
= page
->mapping
;
1826 struct inode
*inode
= mapping
->host
;
1828 trace_f2fs_set_page_dirty(page
, DATA
);
1830 if (!PageUptodate(page
))
1831 SetPageUptodate(page
);
1833 if (f2fs_is_atomic_file(inode
)) {
1834 if (!IS_ATOMIC_WRITTEN_PAGE(page
)) {
1835 register_inmem_page(inode
, page
);
1839 * Previously, this page has been registered, we just
1845 if (!PageDirty(page
)) {
1846 f2fs_set_page_dirty_nobuffers(page
);
1847 update_dirty_page(inode
, page
);
1853 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
1855 struct inode
*inode
= mapping
->host
;
1857 if (f2fs_has_inline_data(inode
))
1860 /* make sure allocating whole blocks */
1861 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
1862 filemap_write_and_wait(mapping
);
1864 return generic_block_bmap(mapping
, block
, get_data_block_bmap
);
1867 const struct address_space_operations f2fs_dblock_aops
= {
1868 .readpage
= f2fs_read_data_page
,
1869 .readpages
= f2fs_read_data_pages
,
1870 .writepage
= f2fs_write_data_page
,
1871 .writepages
= f2fs_write_data_pages
,
1872 .write_begin
= f2fs_write_begin
,
1873 .write_end
= f2fs_write_end
,
1874 .set_page_dirty
= f2fs_set_data_page_dirty
,
1875 .invalidatepage
= f2fs_invalidate_page
,
1876 .releasepage
= f2fs_release_page
,
1877 .direct_IO
= f2fs_direct_IO
,