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>
25 #include <linux/sched/signal.h>
31 #include <trace/events/f2fs.h>
33 static bool __is_cp_guaranteed(struct page
*page
)
35 struct address_space
*mapping
= page
->mapping
;
37 struct f2fs_sb_info
*sbi
;
42 inode
= mapping
->host
;
43 sbi
= F2FS_I_SB(inode
);
45 if (inode
->i_ino
== F2FS_META_INO(sbi
) ||
46 inode
->i_ino
== F2FS_NODE_INO(sbi
) ||
47 S_ISDIR(inode
->i_mode
) ||
53 static void f2fs_read_end_io(struct bio
*bio
)
58 #ifdef CONFIG_F2FS_FAULT_INJECTION
59 if (time_to_inject(F2FS_P_SB(bio
->bi_io_vec
->bv_page
), FAULT_IO
)) {
60 f2fs_show_injection_info(FAULT_IO
);
65 if (f2fs_bio_encrypted(bio
)) {
67 fscrypt_release_ctx(bio
->bi_private
);
69 fscrypt_decrypt_bio_pages(bio
->bi_private
, bio
);
74 bio_for_each_segment_all(bvec
, bio
, i
) {
75 struct page
*page
= bvec
->bv_page
;
78 if (!PageUptodate(page
))
79 SetPageUptodate(page
);
81 ClearPageUptodate(page
);
89 static void f2fs_write_end_io(struct bio
*bio
)
91 struct f2fs_sb_info
*sbi
= bio
->bi_private
;
95 bio_for_each_segment_all(bvec
, bio
, i
) {
96 struct page
*page
= bvec
->bv_page
;
97 enum count_type type
= WB_DATA_TYPE(page
);
99 if (IS_DUMMY_WRITTEN_PAGE(page
)) {
100 set_page_private(page
, (unsigned long)NULL
);
101 ClearPagePrivate(page
);
103 mempool_free(page
, sbi
->write_io_dummy
);
105 if (unlikely(bio
->bi_error
))
106 f2fs_stop_checkpoint(sbi
, true);
110 fscrypt_pullback_bio_page(&page
, true);
112 if (unlikely(bio
->bi_error
)) {
113 mapping_set_error(page
->mapping
, -EIO
);
114 f2fs_stop_checkpoint(sbi
, true);
116 dec_page_count(sbi
, type
);
117 clear_cold_data(page
);
118 end_page_writeback(page
);
120 if (!get_pages(sbi
, F2FS_WB_CP_DATA
) &&
121 wq_has_sleeper(&sbi
->cp_wait
))
122 wake_up(&sbi
->cp_wait
);
128 * Return true, if pre_bio's bdev is same as its target device.
130 struct block_device
*f2fs_target_device(struct f2fs_sb_info
*sbi
,
131 block_t blk_addr
, struct bio
*bio
)
133 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
136 for (i
= 0; i
< sbi
->s_ndevs
; i
++) {
137 if (FDEV(i
).start_blk
<= blk_addr
&&
138 FDEV(i
).end_blk
>= blk_addr
) {
139 blk_addr
-= FDEV(i
).start_blk
;
146 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blk_addr
);
151 int f2fs_target_device_index(struct f2fs_sb_info
*sbi
, block_t blkaddr
)
155 for (i
= 0; i
< sbi
->s_ndevs
; i
++)
156 if (FDEV(i
).start_blk
<= blkaddr
&& FDEV(i
).end_blk
>= blkaddr
)
161 static bool __same_bdev(struct f2fs_sb_info
*sbi
,
162 block_t blk_addr
, struct bio
*bio
)
164 return f2fs_target_device(sbi
, blk_addr
, NULL
) == bio
->bi_bdev
;
168 * Low-level block read/write IO operations.
170 static struct bio
*__bio_alloc(struct f2fs_sb_info
*sbi
, block_t blk_addr
,
171 int npages
, bool is_read
)
175 bio
= f2fs_bio_alloc(npages
);
177 f2fs_target_device(sbi
, blk_addr
, bio
);
178 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
179 bio
->bi_private
= is_read
? NULL
: sbi
;
184 static inline void __submit_bio(struct f2fs_sb_info
*sbi
,
185 struct bio
*bio
, enum page_type type
)
187 if (!is_read_io(bio_op(bio
))) {
190 if (f2fs_sb_mounted_blkzoned(sbi
->sb
) &&
191 current
->plug
&& (type
== DATA
|| type
== NODE
))
192 blk_finish_plug(current
->plug
);
194 if (type
!= DATA
&& type
!= NODE
)
197 start
= bio
->bi_iter
.bi_size
>> F2FS_BLKSIZE_BITS
;
198 start
%= F2FS_IO_SIZE(sbi
);
203 /* fill dummy pages */
204 for (; start
< F2FS_IO_SIZE(sbi
); start
++) {
206 mempool_alloc(sbi
->write_io_dummy
,
207 GFP_NOIO
| __GFP_ZERO
| __GFP_NOFAIL
);
208 f2fs_bug_on(sbi
, !page
);
210 SetPagePrivate(page
);
211 set_page_private(page
, (unsigned long)DUMMY_WRITTEN_PAGE
);
213 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
)
217 * In the NODE case, we lose next block address chain. So, we
218 * need to do checkpoint in f2fs_sync_file.
221 set_sbi_flag(sbi
, SBI_NEED_CP
);
224 if (is_read_io(bio_op(bio
)))
225 trace_f2fs_submit_read_bio(sbi
->sb
, type
, bio
);
227 trace_f2fs_submit_write_bio(sbi
->sb
, type
, bio
);
231 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
233 struct f2fs_io_info
*fio
= &io
->fio
;
238 bio_set_op_attrs(io
->bio
, fio
->op
, fio
->op_flags
);
240 if (is_read_io(fio
->op
))
241 trace_f2fs_prepare_read_bio(io
->sbi
->sb
, fio
->type
, io
->bio
);
243 trace_f2fs_prepare_write_bio(io
->sbi
->sb
, fio
->type
, io
->bio
);
245 __submit_bio(io
->sbi
, io
->bio
, fio
->type
);
249 static bool __has_merged_page(struct f2fs_bio_info
*io
,
250 struct inode
*inode
, nid_t ino
, pgoff_t idx
)
252 struct bio_vec
*bvec
;
262 bio_for_each_segment_all(bvec
, io
->bio
, i
) {
264 if (bvec
->bv_page
->mapping
)
265 target
= bvec
->bv_page
;
267 target
= fscrypt_control_page(bvec
->bv_page
);
269 if (idx
!= target
->index
)
272 if (inode
&& inode
== target
->mapping
->host
)
274 if (ino
&& ino
== ino_of_node(target
))
281 static bool has_merged_page(struct f2fs_sb_info
*sbi
, struct inode
*inode
,
282 nid_t ino
, pgoff_t idx
, enum page_type type
)
284 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
285 struct f2fs_bio_info
*io
= &sbi
->write_io
[btype
];
288 down_read(&io
->io_rwsem
);
289 ret
= __has_merged_page(io
, inode
, ino
, idx
);
290 up_read(&io
->io_rwsem
);
294 static void __f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
295 struct inode
*inode
, nid_t ino
, pgoff_t idx
,
296 enum page_type type
, int rw
)
298 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
299 struct f2fs_bio_info
*io
;
301 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
303 down_write(&io
->io_rwsem
);
305 if (!__has_merged_page(io
, inode
, ino
, idx
))
308 /* change META to META_FLUSH in the checkpoint procedure */
309 if (type
>= META_FLUSH
) {
310 io
->fio
.type
= META_FLUSH
;
311 io
->fio
.op
= REQ_OP_WRITE
;
312 io
->fio
.op_flags
= REQ_META
| REQ_PRIO
;
313 if (!test_opt(sbi
, NOBARRIER
))
314 io
->fio
.op_flags
|= REQ_PREFLUSH
| REQ_FUA
;
316 __submit_merged_bio(io
);
318 up_write(&io
->io_rwsem
);
321 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
, enum page_type type
,
324 __f2fs_submit_merged_bio(sbi
, NULL
, 0, 0, type
, rw
);
327 void f2fs_submit_merged_bio_cond(struct f2fs_sb_info
*sbi
,
328 struct inode
*inode
, nid_t ino
, pgoff_t idx
,
329 enum page_type type
, int rw
)
331 if (has_merged_page(sbi
, inode
, ino
, idx
, type
))
332 __f2fs_submit_merged_bio(sbi
, inode
, ino
, idx
, type
, rw
);
335 void f2fs_flush_merged_bios(struct f2fs_sb_info
*sbi
)
337 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
338 f2fs_submit_merged_bio(sbi
, NODE
, WRITE
);
339 f2fs_submit_merged_bio(sbi
, META
, WRITE
);
343 * Fill the locked page with data located in the block address.
344 * Return unlocked page.
346 int f2fs_submit_page_bio(struct f2fs_io_info
*fio
)
349 struct page
*page
= fio
->encrypted_page
?
350 fio
->encrypted_page
: fio
->page
;
352 trace_f2fs_submit_page_bio(page
, fio
);
353 f2fs_trace_ios(fio
, 0);
355 /* Allocate a new bio */
356 bio
= __bio_alloc(fio
->sbi
, fio
->new_blkaddr
, 1, is_read_io(fio
->op
));
358 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
362 bio_set_op_attrs(bio
, fio
->op
, fio
->op_flags
);
364 __submit_bio(fio
->sbi
, bio
, fio
->type
);
368 int f2fs_submit_page_mbio(struct f2fs_io_info
*fio
)
370 struct f2fs_sb_info
*sbi
= fio
->sbi
;
371 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
372 struct f2fs_bio_info
*io
;
373 bool is_read
= is_read_io(fio
->op
);
374 struct page
*bio_page
;
377 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
379 if (fio
->old_blkaddr
!= NEW_ADDR
)
380 verify_block_addr(sbi
, fio
->old_blkaddr
);
381 verify_block_addr(sbi
, fio
->new_blkaddr
);
383 bio_page
= fio
->encrypted_page
? fio
->encrypted_page
: fio
->page
;
385 /* set submitted = 1 as a return value */
389 inc_page_count(sbi
, WB_DATA_TYPE(bio_page
));
391 down_write(&io
->io_rwsem
);
393 if (io
->bio
&& (io
->last_block_in_bio
!= fio
->new_blkaddr
- 1 ||
394 (io
->fio
.op
!= fio
->op
|| io
->fio
.op_flags
!= fio
->op_flags
) ||
395 !__same_bdev(sbi
, fio
->new_blkaddr
, io
->bio
)))
396 __submit_merged_bio(io
);
398 if (io
->bio
== NULL
) {
399 if ((fio
->type
== DATA
|| fio
->type
== NODE
) &&
400 fio
->new_blkaddr
& F2FS_IO_SIZE_MASK(sbi
)) {
403 dec_page_count(sbi
, WB_DATA_TYPE(bio_page
));
406 io
->bio
= __bio_alloc(sbi
, fio
->new_blkaddr
,
407 BIO_MAX_PAGES
, is_read
);
411 if (bio_add_page(io
->bio
, bio_page
, PAGE_SIZE
, 0) <
413 __submit_merged_bio(io
);
417 io
->last_block_in_bio
= fio
->new_blkaddr
;
418 f2fs_trace_ios(fio
, 0);
420 up_write(&io
->io_rwsem
);
421 trace_f2fs_submit_page_mbio(fio
->page
, fio
);
425 static void __set_data_blkaddr(struct dnode_of_data
*dn
)
427 struct f2fs_node
*rn
= F2FS_NODE(dn
->node_page
);
430 /* Get physical address of data block */
431 addr_array
= blkaddr_in_node(rn
);
432 addr_array
[dn
->ofs_in_node
] = cpu_to_le32(dn
->data_blkaddr
);
436 * Lock ordering for the change of data block address:
439 * update block addresses in the node page
441 void set_data_blkaddr(struct dnode_of_data
*dn
)
443 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
444 __set_data_blkaddr(dn
);
445 if (set_page_dirty(dn
->node_page
))
446 dn
->node_changed
= true;
449 void f2fs_update_data_blkaddr(struct dnode_of_data
*dn
, block_t blkaddr
)
451 dn
->data_blkaddr
= blkaddr
;
452 set_data_blkaddr(dn
);
453 f2fs_update_extent_cache(dn
);
456 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
457 int reserve_new_blocks(struct dnode_of_data
*dn
, blkcnt_t count
)
459 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
464 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
466 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
469 trace_f2fs_reserve_new_blocks(dn
->inode
, dn
->nid
,
470 dn
->ofs_in_node
, count
);
472 f2fs_wait_on_page_writeback(dn
->node_page
, NODE
, true);
474 for (; count
> 0; dn
->ofs_in_node
++) {
476 datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
477 if (blkaddr
== NULL_ADDR
) {
478 dn
->data_blkaddr
= NEW_ADDR
;
479 __set_data_blkaddr(dn
);
484 if (set_page_dirty(dn
->node_page
))
485 dn
->node_changed
= true;
489 /* Should keep dn->ofs_in_node unchanged */
490 int reserve_new_block(struct dnode_of_data
*dn
)
492 unsigned int ofs_in_node
= dn
->ofs_in_node
;
495 ret
= reserve_new_blocks(dn
, 1);
496 dn
->ofs_in_node
= ofs_in_node
;
500 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
502 bool need_put
= dn
->inode_page
? false : true;
505 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
509 if (dn
->data_blkaddr
== NULL_ADDR
)
510 err
= reserve_new_block(dn
);
516 int f2fs_get_block(struct dnode_of_data
*dn
, pgoff_t index
)
518 struct extent_info ei
= {0,0,0};
519 struct inode
*inode
= dn
->inode
;
521 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
522 dn
->data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
526 return f2fs_reserve_block(dn
, index
);
529 struct page
*get_read_data_page(struct inode
*inode
, pgoff_t index
,
530 int op_flags
, bool for_write
)
532 struct address_space
*mapping
= inode
->i_mapping
;
533 struct dnode_of_data dn
;
535 struct extent_info ei
= {0,0,0};
537 struct f2fs_io_info fio
= {
538 .sbi
= F2FS_I_SB(inode
),
541 .op_flags
= op_flags
,
542 .encrypted_page
= NULL
,
545 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
546 return read_mapping_page(mapping
, index
, NULL
);
548 page
= f2fs_grab_cache_page(mapping
, index
, for_write
);
550 return ERR_PTR(-ENOMEM
);
552 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
553 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
557 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
558 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
563 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
568 if (PageUptodate(page
)) {
574 * A new dentry page is allocated but not able to be written, since its
575 * new inode page couldn't be allocated due to -ENOSPC.
576 * In such the case, its blkaddr can be remained as NEW_ADDR.
577 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
579 if (dn
.data_blkaddr
== NEW_ADDR
) {
580 zero_user_segment(page
, 0, PAGE_SIZE
);
581 if (!PageUptodate(page
))
582 SetPageUptodate(page
);
587 fio
.new_blkaddr
= fio
.old_blkaddr
= dn
.data_blkaddr
;
589 err
= f2fs_submit_page_bio(&fio
);
595 f2fs_put_page(page
, 1);
599 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
)
601 struct address_space
*mapping
= inode
->i_mapping
;
604 page
= find_get_page(mapping
, index
);
605 if (page
&& PageUptodate(page
))
607 f2fs_put_page(page
, 0);
609 page
= get_read_data_page(inode
, index
, 0, false);
613 if (PageUptodate(page
))
616 wait_on_page_locked(page
);
617 if (unlikely(!PageUptodate(page
))) {
618 f2fs_put_page(page
, 0);
619 return ERR_PTR(-EIO
);
625 * If it tries to access a hole, return an error.
626 * Because, the callers, functions in dir.c and GC, should be able to know
627 * whether this page exists or not.
629 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
,
632 struct address_space
*mapping
= inode
->i_mapping
;
635 page
= get_read_data_page(inode
, index
, 0, for_write
);
639 /* wait for read completion */
641 if (unlikely(page
->mapping
!= mapping
)) {
642 f2fs_put_page(page
, 1);
645 if (unlikely(!PageUptodate(page
))) {
646 f2fs_put_page(page
, 1);
647 return ERR_PTR(-EIO
);
653 * Caller ensures that this data page is never allocated.
654 * A new zero-filled data page is allocated in the page cache.
656 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
658 * Note that, ipage is set only by make_empty_dir, and if any error occur,
659 * ipage should be released by this function.
661 struct page
*get_new_data_page(struct inode
*inode
,
662 struct page
*ipage
, pgoff_t index
, bool new_i_size
)
664 struct address_space
*mapping
= inode
->i_mapping
;
666 struct dnode_of_data dn
;
669 page
= f2fs_grab_cache_page(mapping
, index
, true);
672 * before exiting, we should make sure ipage will be released
673 * if any error occur.
675 f2fs_put_page(ipage
, 1);
676 return ERR_PTR(-ENOMEM
);
679 set_new_dnode(&dn
, inode
, ipage
, NULL
, 0);
680 err
= f2fs_reserve_block(&dn
, index
);
682 f2fs_put_page(page
, 1);
688 if (PageUptodate(page
))
691 if (dn
.data_blkaddr
== NEW_ADDR
) {
692 zero_user_segment(page
, 0, PAGE_SIZE
);
693 if (!PageUptodate(page
))
694 SetPageUptodate(page
);
696 f2fs_put_page(page
, 1);
698 /* if ipage exists, blkaddr should be NEW_ADDR */
699 f2fs_bug_on(F2FS_I_SB(inode
), ipage
);
700 page
= get_lock_data_page(inode
, index
, true);
705 if (new_i_size
&& i_size_read(inode
) <
706 ((loff_t
)(index
+ 1) << PAGE_SHIFT
))
707 f2fs_i_size_write(inode
, ((loff_t
)(index
+ 1) << PAGE_SHIFT
));
711 static int __allocate_data_block(struct dnode_of_data
*dn
)
713 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
714 struct f2fs_summary sum
;
719 if (unlikely(is_inode_flag_set(dn
->inode
, FI_NO_ALLOC
)))
722 dn
->data_blkaddr
= datablock_addr(dn
->node_page
, dn
->ofs_in_node
);
723 if (dn
->data_blkaddr
== NEW_ADDR
)
726 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, &count
)))
730 get_node_info(sbi
, dn
->nid
, &ni
);
731 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
733 allocate_data_block(sbi
, NULL
, dn
->data_blkaddr
, &dn
->data_blkaddr
,
734 &sum
, CURSEG_WARM_DATA
);
735 set_data_blkaddr(dn
);
738 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), dn
->inode
) +
740 if (i_size_read(dn
->inode
) < ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
))
741 f2fs_i_size_write(dn
->inode
,
742 ((loff_t
)(fofs
+ 1) << PAGE_SHIFT
));
746 static inline bool __force_buffered_io(struct inode
*inode
, int rw
)
748 return ((f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) ||
749 (rw
== WRITE
&& test_opt(F2FS_I_SB(inode
), LFS
)) ||
750 F2FS_I_SB(inode
)->s_ndevs
);
753 int f2fs_preallocate_blocks(struct kiocb
*iocb
, struct iov_iter
*from
)
755 struct inode
*inode
= file_inode(iocb
->ki_filp
);
756 struct f2fs_map_blocks map
;
759 if (is_inode_flag_set(inode
, FI_NO_PREALLOC
))
762 map
.m_lblk
= F2FS_BLK_ALIGN(iocb
->ki_pos
);
763 map
.m_len
= F2FS_BYTES_TO_BLK(iocb
->ki_pos
+ iov_iter_count(from
));
764 if (map
.m_len
> map
.m_lblk
)
765 map
.m_len
-= map
.m_lblk
;
769 map
.m_next_pgofs
= NULL
;
771 if (iocb
->ki_flags
& IOCB_DIRECT
) {
772 err
= f2fs_convert_inline_inode(inode
);
775 return f2fs_map_blocks(inode
, &map
, 1,
776 __force_buffered_io(inode
, WRITE
) ?
777 F2FS_GET_BLOCK_PRE_AIO
:
778 F2FS_GET_BLOCK_PRE_DIO
);
780 if (iocb
->ki_pos
+ iov_iter_count(from
) > MAX_INLINE_DATA
) {
781 err
= f2fs_convert_inline_inode(inode
);
785 if (!f2fs_has_inline_data(inode
))
786 return f2fs_map_blocks(inode
, &map
, 1, F2FS_GET_BLOCK_PRE_AIO
);
791 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
792 * f2fs_map_blocks structure.
793 * If original data blocks are allocated, then give them to blockdev.
795 * a. preallocate requested block addresses
796 * b. do not use extent cache for better performance
797 * c. give the block addresses to blockdev
799 int f2fs_map_blocks(struct inode
*inode
, struct f2fs_map_blocks
*map
,
800 int create
, int flag
)
802 unsigned int maxblocks
= map
->m_len
;
803 struct dnode_of_data dn
;
804 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
805 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE
;
806 pgoff_t pgofs
, end_offset
, end
;
807 int err
= 0, ofs
= 1;
808 unsigned int ofs_in_node
, last_ofs_in_node
;
810 struct extent_info ei
= {0,0,0};
819 /* it only supports block size == page size */
820 pgofs
= (pgoff_t
)map
->m_lblk
;
821 end
= pgofs
+ maxblocks
;
823 if (!create
&& f2fs_lookup_extent_cache(inode
, pgofs
, &ei
)) {
824 map
->m_pblk
= ei
.blk
+ pgofs
- ei
.fofs
;
825 map
->m_len
= min((pgoff_t
)maxblocks
, ei
.fofs
+ ei
.len
- pgofs
);
826 map
->m_flags
= F2FS_MAP_MAPPED
;
834 /* When reading holes, we need its node page */
835 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
836 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
838 if (flag
== F2FS_GET_BLOCK_BMAP
)
840 if (err
== -ENOENT
) {
842 if (map
->m_next_pgofs
)
844 get_next_page_offset(&dn
, pgofs
);
850 last_ofs_in_node
= ofs_in_node
= dn
.ofs_in_node
;
851 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, inode
);
854 blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
856 if (blkaddr
== NEW_ADDR
|| blkaddr
== NULL_ADDR
) {
858 if (unlikely(f2fs_cp_error(sbi
))) {
862 if (flag
== F2FS_GET_BLOCK_PRE_AIO
) {
863 if (blkaddr
== NULL_ADDR
) {
865 last_ofs_in_node
= dn
.ofs_in_node
;
868 err
= __allocate_data_block(&dn
);
870 set_inode_flag(inode
, FI_APPEND_WRITE
);
874 map
->m_flags
|= F2FS_MAP_NEW
;
875 blkaddr
= dn
.data_blkaddr
;
877 if (flag
== F2FS_GET_BLOCK_BMAP
) {
881 if (flag
== F2FS_GET_BLOCK_FIEMAP
&&
882 blkaddr
== NULL_ADDR
) {
883 if (map
->m_next_pgofs
)
884 *map
->m_next_pgofs
= pgofs
+ 1;
886 if (flag
!= F2FS_GET_BLOCK_FIEMAP
||
892 if (flag
== F2FS_GET_BLOCK_PRE_AIO
)
895 if (map
->m_len
== 0) {
896 /* preallocated unwritten block should be mapped for fiemap. */
897 if (blkaddr
== NEW_ADDR
)
898 map
->m_flags
|= F2FS_MAP_UNWRITTEN
;
899 map
->m_flags
|= F2FS_MAP_MAPPED
;
901 map
->m_pblk
= blkaddr
;
903 } else if ((map
->m_pblk
!= NEW_ADDR
&&
904 blkaddr
== (map
->m_pblk
+ ofs
)) ||
905 (map
->m_pblk
== NEW_ADDR
&& blkaddr
== NEW_ADDR
) ||
906 flag
== F2FS_GET_BLOCK_PRE_DIO
) {
917 /* preallocate blocks in batch for one dnode page */
918 if (flag
== F2FS_GET_BLOCK_PRE_AIO
&&
919 (pgofs
== end
|| dn
.ofs_in_node
== end_offset
)) {
921 dn
.ofs_in_node
= ofs_in_node
;
922 err
= reserve_new_blocks(&dn
, prealloc
);
926 map
->m_len
+= dn
.ofs_in_node
- ofs_in_node
;
927 if (prealloc
&& dn
.ofs_in_node
!= last_ofs_in_node
+ 1) {
931 dn
.ofs_in_node
= end_offset
;
936 else if (dn
.ofs_in_node
< end_offset
)
943 f2fs_balance_fs(sbi
, dn
.node_changed
);
952 f2fs_balance_fs(sbi
, dn
.node_changed
);
955 trace_f2fs_map_blocks(inode
, map
, err
);
959 static int __get_data_block(struct inode
*inode
, sector_t iblock
,
960 struct buffer_head
*bh
, int create
, int flag
,
963 struct f2fs_map_blocks map
;
967 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
968 map
.m_next_pgofs
= next_pgofs
;
970 err
= f2fs_map_blocks(inode
, &map
, create
, flag
);
972 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
973 bh
->b_state
= (bh
->b_state
& ~F2FS_MAP_FLAGS
) | map
.m_flags
;
974 bh
->b_size
= (u64
)map
.m_len
<< inode
->i_blkbits
;
979 static int get_data_block(struct inode
*inode
, sector_t iblock
,
980 struct buffer_head
*bh_result
, int create
, int flag
,
983 return __get_data_block(inode
, iblock
, bh_result
, create
,
987 static int get_data_block_dio(struct inode
*inode
, sector_t iblock
,
988 struct buffer_head
*bh_result
, int create
)
990 return __get_data_block(inode
, iblock
, bh_result
, create
,
991 F2FS_GET_BLOCK_DIO
, NULL
);
994 static int get_data_block_bmap(struct inode
*inode
, sector_t iblock
,
995 struct buffer_head
*bh_result
, int create
)
997 /* Block number less than F2FS MAX BLOCKS */
998 if (unlikely(iblock
>= F2FS_I_SB(inode
)->max_file_blocks
))
1001 return __get_data_block(inode
, iblock
, bh_result
, create
,
1002 F2FS_GET_BLOCK_BMAP
, NULL
);
1005 static inline sector_t
logical_to_blk(struct inode
*inode
, loff_t offset
)
1007 return (offset
>> inode
->i_blkbits
);
1010 static inline loff_t
blk_to_logical(struct inode
*inode
, sector_t blk
)
1012 return (blk
<< inode
->i_blkbits
);
1015 int f2fs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
1018 struct buffer_head map_bh
;
1019 sector_t start_blk
, last_blk
;
1021 u64 logical
= 0, phys
= 0, size
= 0;
1025 ret
= fiemap_check_flags(fieinfo
, FIEMAP_FLAG_SYNC
);
1029 if (f2fs_has_inline_data(inode
)) {
1030 ret
= f2fs_inline_data_fiemap(inode
, fieinfo
, start
, len
);
1037 if (logical_to_blk(inode
, len
) == 0)
1038 len
= blk_to_logical(inode
, 1);
1040 start_blk
= logical_to_blk(inode
, start
);
1041 last_blk
= logical_to_blk(inode
, start
+ len
- 1);
1044 memset(&map_bh
, 0, sizeof(struct buffer_head
));
1045 map_bh
.b_size
= len
;
1047 ret
= get_data_block(inode
, start_blk
, &map_bh
, 0,
1048 F2FS_GET_BLOCK_FIEMAP
, &next_pgofs
);
1053 if (!buffer_mapped(&map_bh
)) {
1054 start_blk
= next_pgofs
;
1056 if (blk_to_logical(inode
, start_blk
) < blk_to_logical(inode
,
1057 F2FS_I_SB(inode
)->max_file_blocks
))
1060 flags
|= FIEMAP_EXTENT_LAST
;
1064 if (f2fs_encrypted_inode(inode
))
1065 flags
|= FIEMAP_EXTENT_DATA_ENCRYPTED
;
1067 ret
= fiemap_fill_next_extent(fieinfo
, logical
,
1071 if (start_blk
> last_blk
|| ret
)
1074 logical
= blk_to_logical(inode
, start_blk
);
1075 phys
= blk_to_logical(inode
, map_bh
.b_blocknr
);
1076 size
= map_bh
.b_size
;
1078 if (buffer_unwritten(&map_bh
))
1079 flags
= FIEMAP_EXTENT_UNWRITTEN
;
1081 start_blk
+= logical_to_blk(inode
, size
);
1085 if (fatal_signal_pending(current
))
1093 inode_unlock(inode
);
1097 static struct bio
*f2fs_grab_bio(struct inode
*inode
, block_t blkaddr
,
1100 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1101 struct fscrypt_ctx
*ctx
= NULL
;
1104 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1105 ctx
= fscrypt_get_ctx(inode
, GFP_NOFS
);
1107 return ERR_CAST(ctx
);
1109 /* wait the page to be moved by cleaning */
1110 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
1113 bio
= bio_alloc(GFP_KERNEL
, min_t(int, nr_pages
, BIO_MAX_PAGES
));
1116 fscrypt_release_ctx(ctx
);
1117 return ERR_PTR(-ENOMEM
);
1119 f2fs_target_device(sbi
, blkaddr
, bio
);
1120 bio
->bi_end_io
= f2fs_read_end_io
;
1121 bio
->bi_private
= ctx
;
1127 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1128 * Major change was from block_size == page_size in f2fs by default.
1130 static int f2fs_mpage_readpages(struct address_space
*mapping
,
1131 struct list_head
*pages
, struct page
*page
,
1134 struct bio
*bio
= NULL
;
1136 sector_t last_block_in_bio
= 0;
1137 struct inode
*inode
= mapping
->host
;
1138 const unsigned blkbits
= inode
->i_blkbits
;
1139 const unsigned blocksize
= 1 << blkbits
;
1140 sector_t block_in_file
;
1141 sector_t last_block
;
1142 sector_t last_block_in_file
;
1144 struct f2fs_map_blocks map
;
1150 map
.m_next_pgofs
= NULL
;
1152 for (page_idx
= 0; nr_pages
; page_idx
++, nr_pages
--) {
1154 prefetchw(&page
->flags
);
1156 page
= list_last_entry(pages
, struct page
, lru
);
1157 list_del(&page
->lru
);
1158 if (add_to_page_cache_lru(page
, mapping
,
1160 readahead_gfp_mask(mapping
)))
1164 block_in_file
= (sector_t
)page
->index
;
1165 last_block
= block_in_file
+ nr_pages
;
1166 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >>
1168 if (last_block
> last_block_in_file
)
1169 last_block
= last_block_in_file
;
1172 * Map blocks using the previous result first.
1174 if ((map
.m_flags
& F2FS_MAP_MAPPED
) &&
1175 block_in_file
> map
.m_lblk
&&
1176 block_in_file
< (map
.m_lblk
+ map
.m_len
))
1180 * Then do more f2fs_map_blocks() calls until we are
1181 * done with this page.
1185 if (block_in_file
< last_block
) {
1186 map
.m_lblk
= block_in_file
;
1187 map
.m_len
= last_block
- block_in_file
;
1189 if (f2fs_map_blocks(inode
, &map
, 0,
1190 F2FS_GET_BLOCK_READ
))
1191 goto set_error_page
;
1194 if ((map
.m_flags
& F2FS_MAP_MAPPED
)) {
1195 block_nr
= map
.m_pblk
+ block_in_file
- map
.m_lblk
;
1196 SetPageMappedToDisk(page
);
1198 if (!PageUptodate(page
) && !cleancache_get_page(page
)) {
1199 SetPageUptodate(page
);
1203 zero_user_segment(page
, 0, PAGE_SIZE
);
1204 if (!PageUptodate(page
))
1205 SetPageUptodate(page
);
1211 * This page will go to BIO. Do we need to send this
1214 if (bio
&& (last_block_in_bio
!= block_nr
- 1 ||
1215 !__same_bdev(F2FS_I_SB(inode
), block_nr
, bio
))) {
1217 __submit_bio(F2FS_I_SB(inode
), bio
, DATA
);
1221 bio
= f2fs_grab_bio(inode
, block_nr
, nr_pages
);
1224 goto set_error_page
;
1226 bio_set_op_attrs(bio
, REQ_OP_READ
, 0);
1229 if (bio_add_page(bio
, page
, blocksize
, 0) < blocksize
)
1230 goto submit_and_realloc
;
1232 last_block_in_bio
= block_nr
;
1236 zero_user_segment(page
, 0, PAGE_SIZE
);
1241 __submit_bio(F2FS_I_SB(inode
), bio
, DATA
);
1249 BUG_ON(pages
&& !list_empty(pages
));
1251 __submit_bio(F2FS_I_SB(inode
), bio
, DATA
);
1255 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
1257 struct inode
*inode
= page
->mapping
->host
;
1260 trace_f2fs_readpage(page
, DATA
);
1262 /* If the file has inline data, try to read it directly */
1263 if (f2fs_has_inline_data(inode
))
1264 ret
= f2fs_read_inline_data(inode
, page
);
1266 ret
= f2fs_mpage_readpages(page
->mapping
, NULL
, page
, 1);
1270 static int f2fs_read_data_pages(struct file
*file
,
1271 struct address_space
*mapping
,
1272 struct list_head
*pages
, unsigned nr_pages
)
1274 struct inode
*inode
= file
->f_mapping
->host
;
1275 struct page
*page
= list_last_entry(pages
, struct page
, lru
);
1277 trace_f2fs_readpages(inode
, page
, nr_pages
);
1279 /* If the file has inline data, skip readpages */
1280 if (f2fs_has_inline_data(inode
))
1283 return f2fs_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
1286 int do_write_data_page(struct f2fs_io_info
*fio
)
1288 struct page
*page
= fio
->page
;
1289 struct inode
*inode
= page
->mapping
->host
;
1290 struct dnode_of_data dn
;
1293 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
1294 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
1298 fio
->old_blkaddr
= dn
.data_blkaddr
;
1300 /* This page is already truncated */
1301 if (fio
->old_blkaddr
== NULL_ADDR
) {
1302 ClearPageUptodate(page
);
1306 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
)) {
1307 gfp_t gfp_flags
= GFP_NOFS
;
1309 /* wait for GCed encrypted page writeback */
1310 f2fs_wait_on_encrypted_page_writeback(F2FS_I_SB(inode
),
1313 fio
->encrypted_page
= fscrypt_encrypt_page(inode
, fio
->page
,
1317 if (IS_ERR(fio
->encrypted_page
)) {
1318 err
= PTR_ERR(fio
->encrypted_page
);
1319 if (err
== -ENOMEM
) {
1320 /* flush pending ios and wait for a while */
1321 f2fs_flush_merged_bios(F2FS_I_SB(inode
));
1322 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1323 gfp_flags
|= __GFP_NOFAIL
;
1331 set_page_writeback(page
);
1334 * If current allocation needs SSR,
1335 * it had better in-place writes for updated data.
1337 if (unlikely(fio
->old_blkaddr
!= NEW_ADDR
&&
1338 !is_cold_data(page
) &&
1339 !IS_ATOMIC_WRITTEN_PAGE(page
) &&
1340 need_inplace_update(inode
))) {
1341 rewrite_data_page(fio
);
1342 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1343 trace_f2fs_do_write_data_page(page
, IPU
);
1345 write_data_page(&dn
, fio
);
1346 trace_f2fs_do_write_data_page(page
, OPU
);
1347 set_inode_flag(inode
, FI_APPEND_WRITE
);
1348 if (page
->index
== 0)
1349 set_inode_flag(inode
, FI_FIRST_BLOCK_WRITTEN
);
1352 f2fs_put_dnode(&dn
);
1356 static int __write_data_page(struct page
*page
, bool *submitted
,
1357 struct writeback_control
*wbc
)
1359 struct inode
*inode
= page
->mapping
->host
;
1360 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1361 loff_t i_size
= i_size_read(inode
);
1362 const pgoff_t end_index
= ((unsigned long long) i_size
)
1364 loff_t psize
= (page
->index
+ 1) << PAGE_SHIFT
;
1365 unsigned offset
= 0;
1366 bool need_balance_fs
= false;
1368 struct f2fs_io_info fio
= {
1372 .op_flags
= wbc_to_write_flags(wbc
),
1374 .encrypted_page
= NULL
,
1378 trace_f2fs_writepage(page
, DATA
);
1380 if (page
->index
< end_index
)
1384 * If the offset is out-of-range of file size,
1385 * this page does not have to be written to disk.
1387 offset
= i_size
& (PAGE_SIZE
- 1);
1388 if ((page
->index
>= end_index
+ 1) || !offset
)
1391 zero_user_segment(page
, offset
, PAGE_SIZE
);
1393 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1395 if (f2fs_is_drop_cache(inode
))
1397 /* we should not write 0'th page having journal header */
1398 if (f2fs_is_volatile_file(inode
) && (!page
->index
||
1399 (!wbc
->for_reclaim
&&
1400 available_free_memory(sbi
, BASE_CHECK
))))
1403 /* we should bypass data pages to proceed the kworkder jobs */
1404 if (unlikely(f2fs_cp_error(sbi
))) {
1405 mapping_set_error(page
->mapping
, -EIO
);
1409 /* Dentry blocks are controlled by checkpoint */
1410 if (S_ISDIR(inode
->i_mode
)) {
1411 err
= do_write_data_page(&fio
);
1415 if (!wbc
->for_reclaim
)
1416 need_balance_fs
= true;
1417 else if (has_not_enough_free_secs(sbi
, 0, 0))
1421 if (f2fs_has_inline_data(inode
)) {
1422 err
= f2fs_write_inline_data(inode
, page
);
1428 err
= do_write_data_page(&fio
);
1429 if (F2FS_I(inode
)->last_disk_size
< psize
)
1430 F2FS_I(inode
)->last_disk_size
= psize
;
1431 f2fs_unlock_op(sbi
);
1433 if (err
&& err
!= -ENOENT
)
1437 inode_dec_dirty_pages(inode
);
1439 ClearPageUptodate(page
);
1441 if (wbc
->for_reclaim
) {
1442 f2fs_submit_merged_bio_cond(sbi
, inode
, 0, page
->index
,
1444 remove_dirty_inode(inode
);
1449 f2fs_balance_fs(sbi
, need_balance_fs
);
1451 if (unlikely(f2fs_cp_error(sbi
))) {
1452 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
1457 *submitted
= fio
.submitted
;
1462 redirty_page_for_writepage(wbc
, page
);
1464 return AOP_WRITEPAGE_ACTIVATE
;
1469 static int f2fs_write_data_page(struct page
*page
,
1470 struct writeback_control
*wbc
)
1472 return __write_data_page(page
, NULL
, wbc
);
1476 * This function was copied from write_cche_pages from mm/page-writeback.c.
1477 * The major change is making write step of cold data page separately from
1478 * warm/hot data page.
1480 static int f2fs_write_cache_pages(struct address_space
*mapping
,
1481 struct writeback_control
*wbc
)
1485 struct pagevec pvec
;
1487 pgoff_t
uninitialized_var(writeback_index
);
1489 pgoff_t end
; /* Inclusive */
1491 pgoff_t last_idx
= ULONG_MAX
;
1493 int range_whole
= 0;
1496 pagevec_init(&pvec
, 0);
1498 if (wbc
->range_cyclic
) {
1499 writeback_index
= mapping
->writeback_index
; /* prev offset */
1500 index
= writeback_index
;
1507 index
= wbc
->range_start
>> PAGE_SHIFT
;
1508 end
= wbc
->range_end
>> PAGE_SHIFT
;
1509 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
1511 cycled
= 1; /* ignore range_cyclic tests */
1513 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1514 tag
= PAGECACHE_TAG_TOWRITE
;
1516 tag
= PAGECACHE_TAG_DIRTY
;
1518 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1519 tag_pages_for_writeback(mapping
, index
, end
);
1521 while (!done
&& (index
<= end
)) {
1524 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
1525 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1);
1529 for (i
= 0; i
< nr_pages
; i
++) {
1530 struct page
*page
= pvec
.pages
[i
];
1531 bool submitted
= false;
1533 if (page
->index
> end
) {
1538 done_index
= page
->index
;
1542 if (unlikely(page
->mapping
!= mapping
)) {
1548 if (!PageDirty(page
)) {
1549 /* someone wrote it for us */
1550 goto continue_unlock
;
1553 if (PageWriteback(page
)) {
1554 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
1555 f2fs_wait_on_page_writeback(page
,
1558 goto continue_unlock
;
1561 BUG_ON(PageWriteback(page
));
1562 if (!clear_page_dirty_for_io(page
))
1563 goto continue_unlock
;
1565 ret
= __write_data_page(page
, &submitted
, wbc
);
1566 if (unlikely(ret
)) {
1568 * keep nr_to_write, since vfs uses this to
1569 * get # of written pages.
1571 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
1576 done_index
= page
->index
+ 1;
1579 } else if (submitted
) {
1580 last_idx
= page
->index
;
1583 if (--wbc
->nr_to_write
<= 0 &&
1584 wbc
->sync_mode
== WB_SYNC_NONE
) {
1589 pagevec_release(&pvec
);
1593 if (!cycled
&& !done
) {
1596 end
= writeback_index
- 1;
1599 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
1600 mapping
->writeback_index
= done_index
;
1602 if (last_idx
!= ULONG_MAX
)
1603 f2fs_submit_merged_bio_cond(F2FS_M_SB(mapping
), mapping
->host
,
1604 0, last_idx
, DATA
, WRITE
);
1609 static int f2fs_write_data_pages(struct address_space
*mapping
,
1610 struct writeback_control
*wbc
)
1612 struct inode
*inode
= mapping
->host
;
1613 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1614 struct blk_plug plug
;
1617 /* deal with chardevs and other special file */
1618 if (!mapping
->a_ops
->writepage
)
1621 /* skip writing if there is no dirty page in this inode */
1622 if (!get_dirty_pages(inode
) && wbc
->sync_mode
== WB_SYNC_NONE
)
1625 if (S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_NONE
&&
1626 get_dirty_pages(inode
) < nr_pages_to_skip(sbi
, DATA
) &&
1627 available_free_memory(sbi
, DIRTY_DENTS
))
1630 /* skip writing during file defragment */
1631 if (is_inode_flag_set(inode
, FI_DO_DEFRAG
))
1634 /* during POR, we don't need to trigger writepage at all. */
1635 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
1638 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1640 blk_start_plug(&plug
);
1641 ret
= f2fs_write_cache_pages(mapping
, wbc
);
1642 blk_finish_plug(&plug
);
1644 * if some pages were truncated, we cannot guarantee its mapping->host
1645 * to detect pending bios.
1648 remove_dirty_inode(inode
);
1652 wbc
->pages_skipped
+= get_dirty_pages(inode
);
1653 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
1657 static void f2fs_write_failed(struct address_space
*mapping
, loff_t to
)
1659 struct inode
*inode
= mapping
->host
;
1660 loff_t i_size
= i_size_read(inode
);
1663 truncate_pagecache(inode
, i_size
);
1664 truncate_blocks(inode
, i_size
, true);
1668 static int prepare_write_begin(struct f2fs_sb_info
*sbi
,
1669 struct page
*page
, loff_t pos
, unsigned len
,
1670 block_t
*blk_addr
, bool *node_changed
)
1672 struct inode
*inode
= page
->mapping
->host
;
1673 pgoff_t index
= page
->index
;
1674 struct dnode_of_data dn
;
1676 bool locked
= false;
1677 struct extent_info ei
= {0,0,0};
1681 * we already allocated all the blocks, so we don't need to get
1682 * the block addresses when there is no need to fill the page.
1684 if (!f2fs_has_inline_data(inode
) && len
== PAGE_SIZE
&&
1685 !is_inode_flag_set(inode
, FI_NO_PREALLOC
))
1688 if (f2fs_has_inline_data(inode
) ||
1689 (pos
& PAGE_MASK
) >= i_size_read(inode
)) {
1694 /* check inline_data */
1695 ipage
= get_node_page(sbi
, inode
->i_ino
);
1696 if (IS_ERR(ipage
)) {
1697 err
= PTR_ERR(ipage
);
1701 set_new_dnode(&dn
, inode
, ipage
, ipage
, 0);
1703 if (f2fs_has_inline_data(inode
)) {
1704 if (pos
+ len
<= MAX_INLINE_DATA
) {
1705 read_inline_data(page
, ipage
);
1706 set_inode_flag(inode
, FI_DATA_EXIST
);
1708 set_inline_node(ipage
);
1710 err
= f2fs_convert_inline_page(&dn
, page
);
1713 if (dn
.data_blkaddr
== NULL_ADDR
)
1714 err
= f2fs_get_block(&dn
, index
);
1716 } else if (locked
) {
1717 err
= f2fs_get_block(&dn
, index
);
1719 if (f2fs_lookup_extent_cache(inode
, index
, &ei
)) {
1720 dn
.data_blkaddr
= ei
.blk
+ index
- ei
.fofs
;
1723 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
1724 if (err
|| dn
.data_blkaddr
== NULL_ADDR
) {
1725 f2fs_put_dnode(&dn
);
1733 /* convert_inline_page can make node_changed */
1734 *blk_addr
= dn
.data_blkaddr
;
1735 *node_changed
= dn
.node_changed
;
1737 f2fs_put_dnode(&dn
);
1740 f2fs_unlock_op(sbi
);
1744 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
1745 loff_t pos
, unsigned len
, unsigned flags
,
1746 struct page
**pagep
, void **fsdata
)
1748 struct inode
*inode
= mapping
->host
;
1749 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1750 struct page
*page
= NULL
;
1751 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_SHIFT
;
1752 bool need_balance
= false;
1753 block_t blkaddr
= NULL_ADDR
;
1756 trace_f2fs_write_begin(inode
, pos
, len
, flags
);
1759 * We should check this at this moment to avoid deadlock on inode page
1760 * and #0 page. The locking rule for inline_data conversion should be:
1761 * lock_page(page #0) -> lock_page(inode_page)
1764 err
= f2fs_convert_inline_inode(inode
);
1770 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1771 * wait_for_stable_page. Will wait that below with our IO control.
1773 page
= pagecache_get_page(mapping
, index
,
1774 FGP_LOCK
| FGP_WRITE
| FGP_CREAT
, GFP_NOFS
);
1782 err
= prepare_write_begin(sbi
, page
, pos
, len
,
1783 &blkaddr
, &need_balance
);
1787 if (need_balance
&& has_not_enough_free_secs(sbi
, 0, 0)) {
1789 f2fs_balance_fs(sbi
, true);
1791 if (page
->mapping
!= mapping
) {
1792 /* The page got truncated from under us */
1793 f2fs_put_page(page
, 1);
1798 f2fs_wait_on_page_writeback(page
, DATA
, false);
1800 /* wait for GCed encrypted page writeback */
1801 if (f2fs_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
1802 f2fs_wait_on_encrypted_page_writeback(sbi
, blkaddr
);
1804 if (len
== PAGE_SIZE
|| PageUptodate(page
))
1807 if (!(pos
& (PAGE_SIZE
- 1)) && (pos
+ len
) >= i_size_read(inode
)) {
1808 zero_user_segment(page
, len
, PAGE_SIZE
);
1812 if (blkaddr
== NEW_ADDR
) {
1813 zero_user_segment(page
, 0, PAGE_SIZE
);
1814 SetPageUptodate(page
);
1818 bio
= f2fs_grab_bio(inode
, blkaddr
, 1);
1823 bio
->bi_opf
= REQ_OP_READ
;
1824 if (bio_add_page(bio
, page
, PAGE_SIZE
, 0) < PAGE_SIZE
) {
1830 __submit_bio(sbi
, bio
, DATA
);
1833 if (unlikely(page
->mapping
!= mapping
)) {
1834 f2fs_put_page(page
, 1);
1837 if (unlikely(!PageUptodate(page
))) {
1845 f2fs_put_page(page
, 1);
1846 f2fs_write_failed(mapping
, pos
+ len
);
1850 static int f2fs_write_end(struct file
*file
,
1851 struct address_space
*mapping
,
1852 loff_t pos
, unsigned len
, unsigned copied
,
1853 struct page
*page
, void *fsdata
)
1855 struct inode
*inode
= page
->mapping
->host
;
1857 trace_f2fs_write_end(inode
, pos
, len
, copied
);
1860 * This should be come from len == PAGE_SIZE, and we expect copied
1861 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
1862 * let generic_perform_write() try to copy data again through copied=0.
1864 if (!PageUptodate(page
)) {
1865 if (unlikely(copied
!= len
))
1868 SetPageUptodate(page
);
1873 set_page_dirty(page
);
1875 if (pos
+ copied
> i_size_read(inode
))
1876 f2fs_i_size_write(inode
, pos
+ copied
);
1878 f2fs_put_page(page
, 1);
1879 f2fs_update_time(F2FS_I_SB(inode
), REQ_TIME
);
1883 static int check_direct_IO(struct inode
*inode
, struct iov_iter
*iter
,
1886 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
1888 if (offset
& blocksize_mask
)
1891 if (iov_iter_alignment(iter
) & blocksize_mask
)
1897 static ssize_t
f2fs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
1899 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
1900 struct inode
*inode
= mapping
->host
;
1901 size_t count
= iov_iter_count(iter
);
1902 loff_t offset
= iocb
->ki_pos
;
1903 int rw
= iov_iter_rw(iter
);
1906 err
= check_direct_IO(inode
, iter
, offset
);
1910 if (__force_buffered_io(inode
, rw
))
1913 trace_f2fs_direct_IO_enter(inode
, offset
, count
, rw
);
1915 down_read(&F2FS_I(inode
)->dio_rwsem
[rw
]);
1916 err
= blockdev_direct_IO(iocb
, inode
, iter
, get_data_block_dio
);
1917 up_read(&F2FS_I(inode
)->dio_rwsem
[rw
]);
1921 set_inode_flag(inode
, FI_UPDATE_WRITE
);
1923 f2fs_write_failed(mapping
, offset
+ count
);
1926 trace_f2fs_direct_IO_exit(inode
, offset
, count
, rw
, err
);
1931 void f2fs_invalidate_page(struct page
*page
, unsigned int offset
,
1932 unsigned int length
)
1934 struct inode
*inode
= page
->mapping
->host
;
1935 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1937 if (inode
->i_ino
>= F2FS_ROOT_INO(sbi
) &&
1938 (offset
% PAGE_SIZE
|| length
!= PAGE_SIZE
))
1941 if (PageDirty(page
)) {
1942 if (inode
->i_ino
== F2FS_META_INO(sbi
)) {
1943 dec_page_count(sbi
, F2FS_DIRTY_META
);
1944 } else if (inode
->i_ino
== F2FS_NODE_INO(sbi
)) {
1945 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1947 inode_dec_dirty_pages(inode
);
1948 remove_dirty_inode(inode
);
1952 /* This is atomic written page, keep Private */
1953 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1956 set_page_private(page
, 0);
1957 ClearPagePrivate(page
);
1960 int f2fs_release_page(struct page
*page
, gfp_t wait
)
1962 /* If this is dirty page, keep PagePrivate */
1963 if (PageDirty(page
))
1966 /* This is atomic written page, keep Private */
1967 if (IS_ATOMIC_WRITTEN_PAGE(page
))
1970 set_page_private(page
, 0);
1971 ClearPagePrivate(page
);
1976 * This was copied from __set_page_dirty_buffers which gives higher performance
1977 * in very high speed storages. (e.g., pmem)
1979 void f2fs_set_page_dirty_nobuffers(struct page
*page
)
1981 struct address_space
*mapping
= page
->mapping
;
1982 unsigned long flags
;
1984 if (unlikely(!mapping
))
1987 spin_lock(&mapping
->private_lock
);
1988 lock_page_memcg(page
);
1990 spin_unlock(&mapping
->private_lock
);
1992 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
1993 WARN_ON_ONCE(!PageUptodate(page
));
1994 account_page_dirtied(page
, mapping
);
1995 radix_tree_tag_set(&mapping
->page_tree
,
1996 page_index(page
), PAGECACHE_TAG_DIRTY
);
1997 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
1998 unlock_page_memcg(page
);
2000 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
2004 static int f2fs_set_data_page_dirty(struct page
*page
)
2006 struct address_space
*mapping
= page
->mapping
;
2007 struct inode
*inode
= mapping
->host
;
2009 trace_f2fs_set_page_dirty(page
, DATA
);
2011 if (!PageUptodate(page
))
2012 SetPageUptodate(page
);
2014 if (f2fs_is_atomic_file(inode
) && !f2fs_is_commit_atomic_write(inode
)) {
2015 if (!IS_ATOMIC_WRITTEN_PAGE(page
)) {
2016 register_inmem_page(inode
, page
);
2020 * Previously, this page has been registered, we just
2026 if (!PageDirty(page
)) {
2027 f2fs_set_page_dirty_nobuffers(page
);
2028 update_dirty_page(inode
, page
);
2034 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
2036 struct inode
*inode
= mapping
->host
;
2038 if (f2fs_has_inline_data(inode
))
2041 /* make sure allocating whole blocks */
2042 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2043 filemap_write_and_wait(mapping
);
2045 return generic_block_bmap(mapping
, block
, get_data_block_bmap
);
2048 #ifdef CONFIG_MIGRATION
2049 #include <linux/migrate.h>
2051 int f2fs_migrate_page(struct address_space
*mapping
,
2052 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
2054 int rc
, extra_count
;
2055 struct f2fs_inode_info
*fi
= F2FS_I(mapping
->host
);
2056 bool atomic_written
= IS_ATOMIC_WRITTEN_PAGE(page
);
2058 BUG_ON(PageWriteback(page
));
2060 /* migrating an atomic written page is safe with the inmem_lock hold */
2061 if (atomic_written
&& !mutex_trylock(&fi
->inmem_lock
))
2065 * A reference is expected if PagePrivate set when move mapping,
2066 * however F2FS breaks this for maintaining dirty page counts when
2067 * truncating pages. So here adjusting the 'extra_count' make it work.
2069 extra_count
= (atomic_written
? 1 : 0) - page_has_private(page
);
2070 rc
= migrate_page_move_mapping(mapping
, newpage
,
2071 page
, NULL
, mode
, extra_count
);
2072 if (rc
!= MIGRATEPAGE_SUCCESS
) {
2074 mutex_unlock(&fi
->inmem_lock
);
2078 if (atomic_written
) {
2079 struct inmem_pages
*cur
;
2080 list_for_each_entry(cur
, &fi
->inmem_pages
, list
)
2081 if (cur
->page
== page
) {
2082 cur
->page
= newpage
;
2085 mutex_unlock(&fi
->inmem_lock
);
2090 if (PagePrivate(page
))
2091 SetPagePrivate(newpage
);
2092 set_page_private(newpage
, page_private(page
));
2094 migrate_page_copy(newpage
, page
);
2096 return MIGRATEPAGE_SUCCESS
;
2100 const struct address_space_operations f2fs_dblock_aops
= {
2101 .readpage
= f2fs_read_data_page
,
2102 .readpages
= f2fs_read_data_pages
,
2103 .writepage
= f2fs_write_data_page
,
2104 .writepages
= f2fs_write_data_pages
,
2105 .write_begin
= f2fs_write_begin
,
2106 .write_end
= f2fs_write_end
,
2107 .set_page_dirty
= f2fs_set_data_page_dirty
,
2108 .invalidatepage
= f2fs_invalidate_page
,
2109 .releasepage
= f2fs_release_page
,
2110 .direct_IO
= f2fs_direct_IO
,
2112 #ifdef CONFIG_MIGRATION
2113 .migratepage
= f2fs_migrate_page
,