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/aio.h>
16 #include <linux/writeback.h>
17 #include <linux/backing-dev.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
26 #include <trace/events/f2fs.h>
28 static void f2fs_read_end_io(struct bio
*bio
, int err
)
33 bio_for_each_segment_all(bvec
, bio
, i
) {
34 struct page
*page
= bvec
->bv_page
;
37 SetPageUptodate(page
);
39 ClearPageUptodate(page
);
47 static void f2fs_write_end_io(struct bio
*bio
, int err
)
49 struct f2fs_sb_info
*sbi
= bio
->bi_private
;
53 bio_for_each_segment_all(bvec
, bio
, i
) {
54 struct page
*page
= bvec
->bv_page
;
58 set_bit(AS_EIO
, &page
->mapping
->flags
);
59 f2fs_stop_checkpoint(sbi
);
61 end_page_writeback(page
);
62 dec_page_count(sbi
, F2FS_WRITEBACK
);
65 if (!get_pages(sbi
, F2FS_WRITEBACK
) &&
66 !list_empty(&sbi
->cp_wait
.task_list
))
67 wake_up(&sbi
->cp_wait
);
73 * Low-level block read/write IO operations.
75 static struct bio
*__bio_alloc(struct f2fs_sb_info
*sbi
, block_t blk_addr
,
76 int npages
, bool is_read
)
80 /* No failure on bio allocation */
81 bio
= bio_alloc(GFP_NOIO
, npages
);
83 bio
->bi_bdev
= sbi
->sb
->s_bdev
;
84 bio
->bi_iter
.bi_sector
= SECTOR_FROM_BLOCK(blk_addr
);
85 bio
->bi_end_io
= is_read
? f2fs_read_end_io
: f2fs_write_end_io
;
86 bio
->bi_private
= sbi
;
91 static void __submit_merged_bio(struct f2fs_bio_info
*io
)
93 struct f2fs_io_info
*fio
= &io
->fio
;
98 if (is_read_io(fio
->rw
))
99 trace_f2fs_submit_read_bio(io
->sbi
->sb
, fio
, io
->bio
);
101 trace_f2fs_submit_write_bio(io
->sbi
->sb
, fio
, io
->bio
);
103 submit_bio(fio
->rw
, io
->bio
);
107 void f2fs_submit_merged_bio(struct f2fs_sb_info
*sbi
,
108 enum page_type type
, int rw
)
110 enum page_type btype
= PAGE_TYPE_OF_BIO(type
);
111 struct f2fs_bio_info
*io
;
113 io
= is_read_io(rw
) ? &sbi
->read_io
: &sbi
->write_io
[btype
];
115 down_write(&io
->io_rwsem
);
117 /* change META to META_FLUSH in the checkpoint procedure */
118 if (type
>= META_FLUSH
) {
119 io
->fio
.type
= META_FLUSH
;
120 if (test_opt(sbi
, NOBARRIER
))
121 io
->fio
.rw
= WRITE_FLUSH
| REQ_META
| REQ_PRIO
;
123 io
->fio
.rw
= WRITE_FLUSH_FUA
| REQ_META
| REQ_PRIO
;
125 __submit_merged_bio(io
);
126 up_write(&io
->io_rwsem
);
130 * Fill the locked page with data located in the block address.
131 * Return unlocked page.
133 int f2fs_submit_page_bio(struct f2fs_sb_info
*sbi
, struct page
*page
,
134 struct f2fs_io_info
*fio
)
138 trace_f2fs_submit_page_bio(page
, fio
);
139 f2fs_trace_ios(page
, fio
, 0);
141 /* Allocate a new bio */
142 bio
= __bio_alloc(sbi
, fio
->blk_addr
, 1, is_read_io(fio
->rw
));
144 if (bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) < PAGE_CACHE_SIZE
) {
146 f2fs_put_page(page
, 1);
150 submit_bio(fio
->rw
, bio
);
154 void f2fs_submit_page_mbio(struct f2fs_sb_info
*sbi
, struct page
*page
,
155 struct f2fs_io_info
*fio
)
157 enum page_type btype
= PAGE_TYPE_OF_BIO(fio
->type
);
158 struct f2fs_bio_info
*io
;
159 bool is_read
= is_read_io(fio
->rw
);
161 io
= is_read
? &sbi
->read_io
: &sbi
->write_io
[btype
];
163 verify_block_addr(sbi
, fio
->blk_addr
);
165 down_write(&io
->io_rwsem
);
168 inc_page_count(sbi
, F2FS_WRITEBACK
);
170 if (io
->bio
&& (io
->last_block_in_bio
!= fio
->blk_addr
- 1 ||
171 io
->fio
.rw
!= fio
->rw
))
172 __submit_merged_bio(io
);
174 if (io
->bio
== NULL
) {
175 int bio_blocks
= MAX_BIO_BLOCKS(sbi
);
177 io
->bio
= __bio_alloc(sbi
, fio
->blk_addr
, bio_blocks
, is_read
);
181 if (bio_add_page(io
->bio
, page
, PAGE_CACHE_SIZE
, 0) <
183 __submit_merged_bio(io
);
187 io
->last_block_in_bio
= fio
->blk_addr
;
188 f2fs_trace_ios(page
, fio
, 0);
190 up_write(&io
->io_rwsem
);
191 trace_f2fs_submit_page_mbio(page
, fio
);
195 * Lock ordering for the change of data block address:
198 * update block addresses in the node page
200 static void __set_data_blkaddr(struct dnode_of_data
*dn
)
202 struct f2fs_node
*rn
;
204 struct page
*node_page
= dn
->node_page
;
205 unsigned int ofs_in_node
= dn
->ofs_in_node
;
207 f2fs_wait_on_page_writeback(node_page
, NODE
);
209 rn
= F2FS_NODE(node_page
);
211 /* Get physical address of data block */
212 addr_array
= blkaddr_in_node(rn
);
213 addr_array
[ofs_in_node
] = cpu_to_le32(dn
->data_blkaddr
);
214 set_page_dirty(node_page
);
217 int reserve_new_block(struct dnode_of_data
*dn
)
219 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
221 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
223 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, 1)))
226 trace_f2fs_reserve_new_block(dn
->inode
, dn
->nid
, dn
->ofs_in_node
);
228 dn
->data_blkaddr
= NEW_ADDR
;
229 __set_data_blkaddr(dn
);
230 mark_inode_dirty(dn
->inode
);
235 int f2fs_reserve_block(struct dnode_of_data
*dn
, pgoff_t index
)
237 bool need_put
= dn
->inode_page
? false : true;
240 err
= get_dnode_of_data(dn
, index
, ALLOC_NODE
);
244 if (dn
->data_blkaddr
== NULL_ADDR
)
245 err
= reserve_new_block(dn
);
251 static int check_extent_cache(struct inode
*inode
, pgoff_t pgofs
,
252 struct buffer_head
*bh_result
)
254 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
255 pgoff_t start_fofs
, end_fofs
;
256 block_t start_blkaddr
;
258 if (is_inode_flag_set(fi
, FI_NO_EXTENT
))
261 read_lock(&fi
->ext
.ext_lock
);
262 if (fi
->ext
.len
== 0) {
263 read_unlock(&fi
->ext
.ext_lock
);
267 stat_inc_total_hit(inode
->i_sb
);
269 start_fofs
= fi
->ext
.fofs
;
270 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
271 start_blkaddr
= fi
->ext
.blk_addr
;
273 if (pgofs
>= start_fofs
&& pgofs
<= end_fofs
) {
274 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
277 clear_buffer_new(bh_result
);
278 map_bh(bh_result
, inode
->i_sb
,
279 start_blkaddr
+ pgofs
- start_fofs
);
280 count
= end_fofs
- pgofs
+ 1;
281 if (count
< (UINT_MAX
>> blkbits
))
282 bh_result
->b_size
= (count
<< blkbits
);
284 bh_result
->b_size
= UINT_MAX
;
286 stat_inc_read_hit(inode
->i_sb
);
287 read_unlock(&fi
->ext
.ext_lock
);
290 read_unlock(&fi
->ext
.ext_lock
);
294 void update_extent_cache(struct dnode_of_data
*dn
)
296 struct f2fs_inode_info
*fi
= F2FS_I(dn
->inode
);
297 pgoff_t fofs
, start_fofs
, end_fofs
;
298 block_t start_blkaddr
, end_blkaddr
;
299 int need_update
= true;
301 f2fs_bug_on(F2FS_I_SB(dn
->inode
), dn
->data_blkaddr
== NEW_ADDR
);
303 /* Update the page address in the parent node */
304 __set_data_blkaddr(dn
);
306 if (is_inode_flag_set(fi
, FI_NO_EXTENT
))
309 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), fi
) +
312 write_lock(&fi
->ext
.ext_lock
);
314 start_fofs
= fi
->ext
.fofs
;
315 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
316 start_blkaddr
= fi
->ext
.blk_addr
;
317 end_blkaddr
= fi
->ext
.blk_addr
+ fi
->ext
.len
- 1;
319 /* Drop and initialize the matched extent */
320 if (fi
->ext
.len
== 1 && fofs
== start_fofs
)
324 if (fi
->ext
.len
== 0) {
325 if (dn
->data_blkaddr
!= NULL_ADDR
) {
327 fi
->ext
.blk_addr
= dn
->data_blkaddr
;
334 if (fofs
== start_fofs
- 1 && dn
->data_blkaddr
== start_blkaddr
- 1) {
342 if (fofs
== end_fofs
+ 1 && dn
->data_blkaddr
== end_blkaddr
+ 1) {
347 /* Split the existing extent */
348 if (fi
->ext
.len
> 1 &&
349 fofs
>= start_fofs
&& fofs
<= end_fofs
) {
350 if ((end_fofs
- fofs
) < (fi
->ext
.len
>> 1)) {
351 fi
->ext
.len
= fofs
- start_fofs
;
353 fi
->ext
.fofs
= fofs
+ 1;
354 fi
->ext
.blk_addr
= start_blkaddr
+
355 fofs
- start_fofs
+ 1;
356 fi
->ext
.len
-= fofs
- start_fofs
+ 1;
362 /* Finally, if the extent is very fragmented, let's drop the cache. */
363 if (fi
->ext
.len
< F2FS_MIN_EXTENT_LEN
) {
365 set_inode_flag(fi
, FI_NO_EXTENT
);
369 write_unlock(&fi
->ext
.ext_lock
);
375 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
, bool sync
)
377 struct address_space
*mapping
= inode
->i_mapping
;
378 struct dnode_of_data dn
;
381 struct f2fs_io_info fio
= {
383 .rw
= sync
? READ_SYNC
: READA
,
386 page
= find_get_page(mapping
, index
);
387 if (page
&& PageUptodate(page
))
389 f2fs_put_page(page
, 0);
391 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
392 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
397 if (dn
.data_blkaddr
== NULL_ADDR
)
398 return ERR_PTR(-ENOENT
);
400 /* By fallocate(), there is no cached page, but with NEW_ADDR */
401 if (unlikely(dn
.data_blkaddr
== NEW_ADDR
))
402 return ERR_PTR(-EINVAL
);
404 page
= grab_cache_page(mapping
, index
);
406 return ERR_PTR(-ENOMEM
);
408 if (PageUptodate(page
)) {
413 fio
.blk_addr
= dn
.data_blkaddr
;
414 err
= f2fs_submit_page_bio(F2FS_I_SB(inode
), page
, &fio
);
419 wait_on_page_locked(page
);
420 if (unlikely(!PageUptodate(page
))) {
421 f2fs_put_page(page
, 0);
422 return ERR_PTR(-EIO
);
429 * If it tries to access a hole, return an error.
430 * Because, the callers, functions in dir.c and GC, should be able to know
431 * whether this page exists or not.
433 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
)
435 struct address_space
*mapping
= inode
->i_mapping
;
436 struct dnode_of_data dn
;
439 struct f2fs_io_info fio
= {
444 page
= grab_cache_page(mapping
, index
);
446 return ERR_PTR(-ENOMEM
);
448 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
449 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
451 f2fs_put_page(page
, 1);
456 if (unlikely(dn
.data_blkaddr
== NULL_ADDR
)) {
457 f2fs_put_page(page
, 1);
458 return ERR_PTR(-ENOENT
);
461 if (PageUptodate(page
))
465 * A new dentry page is allocated but not able to be written, since its
466 * new inode page couldn't be allocated due to -ENOSPC.
467 * In such the case, its blkaddr can be remained as NEW_ADDR.
468 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
470 if (dn
.data_blkaddr
== NEW_ADDR
) {
471 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
472 SetPageUptodate(page
);
476 fio
.blk_addr
= dn
.data_blkaddr
;
477 err
= f2fs_submit_page_bio(F2FS_I_SB(inode
), page
, &fio
);
482 if (unlikely(!PageUptodate(page
))) {
483 f2fs_put_page(page
, 1);
484 return ERR_PTR(-EIO
);
486 if (unlikely(page
->mapping
!= mapping
)) {
487 f2fs_put_page(page
, 1);
494 * Caller ensures that this data page is never allocated.
495 * A new zero-filled data page is allocated in the page cache.
497 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
499 * Note that, ipage is set only by make_empty_dir.
501 struct page
*get_new_data_page(struct inode
*inode
,
502 struct page
*ipage
, pgoff_t index
, bool new_i_size
)
504 struct address_space
*mapping
= inode
->i_mapping
;
506 struct dnode_of_data dn
;
509 set_new_dnode(&dn
, inode
, ipage
, NULL
, 0);
510 err
= f2fs_reserve_block(&dn
, index
);
514 page
= grab_cache_page(mapping
, index
);
520 if (PageUptodate(page
))
523 if (dn
.data_blkaddr
== NEW_ADDR
) {
524 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
525 SetPageUptodate(page
);
527 struct f2fs_io_info fio
= {
530 .blk_addr
= dn
.data_blkaddr
,
532 err
= f2fs_submit_page_bio(F2FS_I_SB(inode
), page
, &fio
);
537 if (unlikely(!PageUptodate(page
))) {
538 f2fs_put_page(page
, 1);
542 if (unlikely(page
->mapping
!= mapping
)) {
543 f2fs_put_page(page
, 1);
549 i_size_read(inode
) < ((index
+ 1) << PAGE_CACHE_SHIFT
)) {
550 i_size_write(inode
, ((index
+ 1) << PAGE_CACHE_SHIFT
));
551 /* Only the directory inode sets new_i_size */
552 set_inode_flag(F2FS_I(inode
), FI_UPDATE_DIR
);
561 static int __allocate_data_block(struct dnode_of_data
*dn
)
563 struct f2fs_sb_info
*sbi
= F2FS_I_SB(dn
->inode
);
564 struct f2fs_inode_info
*fi
= F2FS_I(dn
->inode
);
565 struct f2fs_summary sum
;
567 int seg
= CURSEG_WARM_DATA
;
570 if (unlikely(is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
)))
572 if (unlikely(!inc_valid_block_count(sbi
, dn
->inode
, 1)))
575 get_node_info(sbi
, dn
->nid
, &ni
);
576 set_summary(&sum
, dn
->nid
, dn
->ofs_in_node
, ni
.version
);
578 if (dn
->ofs_in_node
== 0 && dn
->inode_page
== dn
->node_page
)
579 seg
= CURSEG_DIRECT_IO
;
581 allocate_data_block(sbi
, NULL
, NULL_ADDR
, &dn
->data_blkaddr
, &sum
, seg
);
583 /* direct IO doesn't use extent cache to maximize the performance */
584 __set_data_blkaddr(dn
);
587 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), fi
) +
589 if (i_size_read(dn
->inode
) < ((fofs
+ 1) << PAGE_CACHE_SHIFT
))
590 i_size_write(dn
->inode
, ((fofs
+ 1) << PAGE_CACHE_SHIFT
));
596 * get_data_block() now supported readahead/bmap/rw direct_IO with mapped bh.
597 * If original data blocks are allocated, then give them to blockdev.
599 * a. preallocate requested block addresses
600 * b. do not use extent cache for better performance
601 * c. give the block addresses to blockdev
603 static int __get_data_block(struct inode
*inode
, sector_t iblock
,
604 struct buffer_head
*bh_result
, int create
, bool fiemap
)
606 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
607 unsigned maxblocks
= bh_result
->b_size
>> blkbits
;
608 struct dnode_of_data dn
;
609 int mode
= create
? ALLOC_NODE
: LOOKUP_NODE_RA
;
610 pgoff_t pgofs
, end_offset
;
611 int err
= 0, ofs
= 1;
612 bool allocated
= false;
614 /* Get the page offset from the block offset(iblock) */
615 pgofs
= (pgoff_t
)(iblock
>> (PAGE_CACHE_SHIFT
- blkbits
));
617 if (check_extent_cache(inode
, pgofs
, bh_result
))
621 f2fs_balance_fs(F2FS_I_SB(inode
));
622 f2fs_lock_op(F2FS_I_SB(inode
));
625 /* When reading holes, we need its node page */
626 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
627 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
633 if (dn
.data_blkaddr
== NEW_ADDR
&& !fiemap
)
636 if (dn
.data_blkaddr
!= NULL_ADDR
) {
637 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
639 err
= __allocate_data_block(&dn
);
643 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
648 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, F2FS_I(inode
));
649 bh_result
->b_size
= (((size_t)1) << blkbits
);
654 if (dn
.ofs_in_node
>= end_offset
) {
656 sync_inode_page(&dn
);
660 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
661 err
= get_dnode_of_data(&dn
, pgofs
, mode
);
667 if (dn
.data_blkaddr
== NEW_ADDR
&& !fiemap
)
670 end_offset
= ADDRS_PER_PAGE(dn
.node_page
, F2FS_I(inode
));
673 if (maxblocks
> (bh_result
->b_size
>> blkbits
)) {
674 block_t blkaddr
= datablock_addr(dn
.node_page
, dn
.ofs_in_node
);
675 if (blkaddr
== NULL_ADDR
&& create
) {
676 err
= __allocate_data_block(&dn
);
680 blkaddr
= dn
.data_blkaddr
;
682 /* Give more consecutive addresses for the readahead */
683 if (blkaddr
== (bh_result
->b_blocknr
+ ofs
)) {
687 bh_result
->b_size
+= (((size_t)1) << blkbits
);
693 sync_inode_page(&dn
);
698 f2fs_unlock_op(F2FS_I_SB(inode
));
700 trace_f2fs_get_data_block(inode
, iblock
, bh_result
, err
);
704 static int get_data_block(struct inode
*inode
, sector_t iblock
,
705 struct buffer_head
*bh_result
, int create
)
707 return __get_data_block(inode
, iblock
, bh_result
, create
, false);
710 static int get_data_block_fiemap(struct inode
*inode
, sector_t iblock
,
711 struct buffer_head
*bh_result
, int create
)
713 return __get_data_block(inode
, iblock
, bh_result
, create
, true);
716 int f2fs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
719 return generic_block_fiemap(inode
, fieinfo
,
720 start
, len
, get_data_block_fiemap
);
723 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
725 struct inode
*inode
= page
->mapping
->host
;
728 trace_f2fs_readpage(page
, DATA
);
730 /* If the file has inline data, try to read it directly */
731 if (f2fs_has_inline_data(inode
))
732 ret
= f2fs_read_inline_data(inode
, page
);
734 ret
= mpage_readpage(page
, get_data_block
);
739 static int f2fs_read_data_pages(struct file
*file
,
740 struct address_space
*mapping
,
741 struct list_head
*pages
, unsigned nr_pages
)
743 struct inode
*inode
= file
->f_mapping
->host
;
745 /* If the file has inline data, skip readpages */
746 if (f2fs_has_inline_data(inode
))
749 return mpage_readpages(mapping
, pages
, nr_pages
, get_data_block
);
752 int do_write_data_page(struct page
*page
, struct f2fs_io_info
*fio
)
754 struct inode
*inode
= page
->mapping
->host
;
755 struct dnode_of_data dn
;
758 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
759 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
763 fio
->blk_addr
= dn
.data_blkaddr
;
765 /* This page is already truncated */
766 if (fio
->blk_addr
== NULL_ADDR
)
769 set_page_writeback(page
);
772 * If current allocation needs SSR,
773 * it had better in-place writes for updated data.
775 if (unlikely(fio
->blk_addr
!= NEW_ADDR
&&
776 !is_cold_data(page
) &&
777 need_inplace_update(inode
))) {
778 rewrite_data_page(page
, fio
);
779 set_inode_flag(F2FS_I(inode
), FI_UPDATE_WRITE
);
781 write_data_page(page
, &dn
, fio
);
782 update_extent_cache(&dn
);
783 set_inode_flag(F2FS_I(inode
), FI_APPEND_WRITE
);
790 static int f2fs_write_data_page(struct page
*page
,
791 struct writeback_control
*wbc
)
793 struct inode
*inode
= page
->mapping
->host
;
794 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
795 loff_t i_size
= i_size_read(inode
);
796 const pgoff_t end_index
= ((unsigned long long) i_size
)
799 bool need_balance_fs
= false;
801 struct f2fs_io_info fio
= {
803 .rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
,
806 trace_f2fs_writepage(page
, DATA
);
808 if (page
->index
< end_index
)
812 * If the offset is out-of-range of file size,
813 * this page does not have to be written to disk.
815 offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
816 if ((page
->index
>= end_index
+ 1) || !offset
)
819 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
821 if (unlikely(is_sbi_flag_set(sbi
, SBI_POR_DOING
)))
823 if (f2fs_is_drop_cache(inode
))
825 if (f2fs_is_volatile_file(inode
) && !wbc
->for_reclaim
&&
826 available_free_memory(sbi
, BASE_CHECK
))
829 /* Dentry blocks are controlled by checkpoint */
830 if (S_ISDIR(inode
->i_mode
)) {
831 if (unlikely(f2fs_cp_error(sbi
)))
833 err
= do_write_data_page(page
, &fio
);
837 /* we should bypass data pages to proceed the kworkder jobs */
838 if (unlikely(f2fs_cp_error(sbi
))) {
843 if (!wbc
->for_reclaim
)
844 need_balance_fs
= true;
845 else if (has_not_enough_free_secs(sbi
, 0))
850 if (f2fs_has_inline_data(inode
))
851 err
= f2fs_write_inline_data(inode
, page
);
853 err
= do_write_data_page(page
, &fio
);
856 if (err
&& err
!= -ENOENT
)
859 clear_cold_data(page
);
861 inode_dec_dirty_pages(inode
);
864 f2fs_balance_fs(sbi
);
865 if (wbc
->for_reclaim
)
866 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
870 redirty_page_for_writepage(wbc
, page
);
871 return AOP_WRITEPAGE_ACTIVATE
;
874 static int __f2fs_writepage(struct page
*page
, struct writeback_control
*wbc
,
877 struct address_space
*mapping
= data
;
878 int ret
= mapping
->a_ops
->writepage(page
, wbc
);
879 mapping_set_error(mapping
, ret
);
883 static int f2fs_write_data_pages(struct address_space
*mapping
,
884 struct writeback_control
*wbc
)
886 struct inode
*inode
= mapping
->host
;
887 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
892 trace_f2fs_writepages(mapping
->host
, wbc
, DATA
);
894 /* deal with chardevs and other special file */
895 if (!mapping
->a_ops
->writepage
)
898 if (S_ISDIR(inode
->i_mode
) && wbc
->sync_mode
== WB_SYNC_NONE
&&
899 get_dirty_pages(inode
) < nr_pages_to_skip(sbi
, DATA
) &&
900 available_free_memory(sbi
, DIRTY_DENTS
))
903 diff
= nr_pages_to_write(sbi
, DATA
, wbc
);
905 if (!S_ISDIR(inode
->i_mode
)) {
906 mutex_lock(&sbi
->writepages
);
909 ret
= write_cache_pages(mapping
, wbc
, __f2fs_writepage
, mapping
);
911 mutex_unlock(&sbi
->writepages
);
913 f2fs_submit_merged_bio(sbi
, DATA
, WRITE
);
915 remove_dirty_dir_inode(inode
);
917 wbc
->nr_to_write
= max((long)0, wbc
->nr_to_write
- diff
);
921 wbc
->pages_skipped
+= get_dirty_pages(inode
);
925 static void f2fs_write_failed(struct address_space
*mapping
, loff_t to
)
927 struct inode
*inode
= mapping
->host
;
929 if (to
> inode
->i_size
) {
930 truncate_pagecache(inode
, inode
->i_size
);
931 truncate_blocks(inode
, inode
->i_size
, true);
935 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
936 loff_t pos
, unsigned len
, unsigned flags
,
937 struct page
**pagep
, void **fsdata
)
939 struct inode
*inode
= mapping
->host
;
940 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
941 struct page
*page
, *ipage
;
942 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_CACHE_SHIFT
;
943 struct dnode_of_data dn
;
946 trace_f2fs_write_begin(inode
, pos
, len
, flags
);
948 f2fs_balance_fs(sbi
);
951 * We should check this at this moment to avoid deadlock on inode page
952 * and #0 page. The locking rule for inline_data conversion should be:
953 * lock_page(page #0) -> lock_page(inode_page)
956 err
= f2fs_convert_inline_inode(inode
);
961 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
971 /* check inline_data */
972 ipage
= get_node_page(sbi
, inode
->i_ino
);
974 err
= PTR_ERR(ipage
);
978 set_new_dnode(&dn
, inode
, ipage
, ipage
, 0);
980 if (f2fs_has_inline_data(inode
)) {
981 if (pos
+ len
<= MAX_INLINE_DATA
) {
982 read_inline_data(page
, ipage
);
983 set_inode_flag(F2FS_I(inode
), FI_DATA_EXIST
);
984 sync_inode_page(&dn
);
987 err
= f2fs_convert_inline_page(&dn
, page
);
991 err
= f2fs_reserve_block(&dn
, index
);
998 if ((len
== PAGE_CACHE_SIZE
) || PageUptodate(page
))
1001 f2fs_wait_on_page_writeback(page
, DATA
);
1003 if ((pos
& PAGE_CACHE_MASK
) >= i_size_read(inode
)) {
1004 unsigned start
= pos
& (PAGE_CACHE_SIZE
- 1);
1005 unsigned end
= start
+ len
;
1007 /* Reading beyond i_size is simple: memset to zero */
1008 zero_user_segments(page
, 0, start
, end
, PAGE_CACHE_SIZE
);
1012 if (dn
.data_blkaddr
== NEW_ADDR
) {
1013 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
1015 struct f2fs_io_info fio
= {
1018 .blk_addr
= dn
.data_blkaddr
,
1020 err
= f2fs_submit_page_bio(sbi
, page
, &fio
);
1025 if (unlikely(!PageUptodate(page
))) {
1026 f2fs_put_page(page
, 1);
1030 if (unlikely(page
->mapping
!= mapping
)) {
1031 f2fs_put_page(page
, 1);
1036 SetPageUptodate(page
);
1037 clear_cold_data(page
);
1041 f2fs_put_dnode(&dn
);
1043 f2fs_unlock_op(sbi
);
1044 f2fs_put_page(page
, 1);
1046 f2fs_write_failed(mapping
, pos
+ len
);
1050 static int f2fs_write_end(struct file
*file
,
1051 struct address_space
*mapping
,
1052 loff_t pos
, unsigned len
, unsigned copied
,
1053 struct page
*page
, void *fsdata
)
1055 struct inode
*inode
= page
->mapping
->host
;
1057 trace_f2fs_write_end(inode
, pos
, len
, copied
);
1059 set_page_dirty(page
);
1061 if (pos
+ copied
> i_size_read(inode
)) {
1062 i_size_write(inode
, pos
+ copied
);
1063 mark_inode_dirty(inode
);
1064 update_inode_page(inode
);
1067 f2fs_put_page(page
, 1);
1071 static int check_direct_IO(struct inode
*inode
, int rw
,
1072 struct iov_iter
*iter
, loff_t offset
)
1074 unsigned blocksize_mask
= inode
->i_sb
->s_blocksize
- 1;
1079 if (offset
& blocksize_mask
)
1082 if (iov_iter_alignment(iter
) & blocksize_mask
)
1088 static ssize_t
f2fs_direct_IO(int rw
, struct kiocb
*iocb
,
1089 struct iov_iter
*iter
, loff_t offset
)
1091 struct file
*file
= iocb
->ki_filp
;
1092 struct address_space
*mapping
= file
->f_mapping
;
1093 struct inode
*inode
= mapping
->host
;
1094 size_t count
= iov_iter_count(iter
);
1097 /* we don't need to use inline_data strictly */
1098 if (f2fs_has_inline_data(inode
)) {
1099 err
= f2fs_convert_inline_inode(inode
);
1104 if (check_direct_IO(inode
, rw
, iter
, offset
))
1107 trace_f2fs_direct_IO_enter(inode
, offset
, count
, rw
);
1109 err
= blockdev_direct_IO(rw
, iocb
, inode
, iter
, offset
, get_data_block
);
1110 if (err
< 0 && (rw
& WRITE
))
1111 f2fs_write_failed(mapping
, offset
+ count
);
1113 trace_f2fs_direct_IO_exit(inode
, offset
, count
, rw
, err
);
1118 void f2fs_invalidate_page(struct page
*page
, unsigned int offset
,
1119 unsigned int length
)
1121 struct inode
*inode
= page
->mapping
->host
;
1122 struct f2fs_sb_info
*sbi
= F2FS_I_SB(inode
);
1124 if (inode
->i_ino
>= F2FS_ROOT_INO(sbi
) &&
1125 (offset
% PAGE_CACHE_SIZE
|| length
!= PAGE_CACHE_SIZE
))
1128 if (PageDirty(page
)) {
1129 if (inode
->i_ino
== F2FS_META_INO(sbi
))
1130 dec_page_count(sbi
, F2FS_DIRTY_META
);
1131 else if (inode
->i_ino
== F2FS_NODE_INO(sbi
))
1132 dec_page_count(sbi
, F2FS_DIRTY_NODES
);
1134 inode_dec_dirty_pages(inode
);
1136 ClearPagePrivate(page
);
1139 int f2fs_release_page(struct page
*page
, gfp_t wait
)
1141 /* If this is dirty page, keep PagePrivate */
1142 if (PageDirty(page
))
1145 ClearPagePrivate(page
);
1149 static int f2fs_set_data_page_dirty(struct page
*page
)
1151 struct address_space
*mapping
= page
->mapping
;
1152 struct inode
*inode
= mapping
->host
;
1154 trace_f2fs_set_page_dirty(page
, DATA
);
1156 SetPageUptodate(page
);
1158 if (f2fs_is_atomic_file(inode
)) {
1159 register_inmem_page(inode
, page
);
1163 mark_inode_dirty(inode
);
1165 if (!PageDirty(page
)) {
1166 __set_page_dirty_nobuffers(page
);
1167 update_dirty_page(inode
, page
);
1173 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
1175 struct inode
*inode
= mapping
->host
;
1177 /* we don't need to use inline_data strictly */
1178 if (f2fs_has_inline_data(inode
)) {
1179 int err
= f2fs_convert_inline_inode(inode
);
1183 return generic_block_bmap(mapping
, block
, get_data_block
);
1186 const struct address_space_operations f2fs_dblock_aops
= {
1187 .readpage
= f2fs_read_data_page
,
1188 .readpages
= f2fs_read_data_pages
,
1189 .writepage
= f2fs_write_data_page
,
1190 .writepages
= f2fs_write_data_pages
,
1191 .write_begin
= f2fs_write_begin
,
1192 .write_end
= f2fs_write_end
,
1193 .set_page_dirty
= f2fs_set_data_page_dirty
,
1194 .invalidatepage
= f2fs_invalidate_page
,
1195 .releasepage
= f2fs_release_page
,
1196 .direct_IO
= f2fs_direct_IO
,