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>
25 #include <trace/events/f2fs.h>
28 * Lock ordering for the change of data block address:
31 * update block addresses in the node page
33 static void __set_data_blkaddr(struct dnode_of_data
*dn
, block_t new_addr
)
37 struct page
*node_page
= dn
->node_page
;
38 unsigned int ofs_in_node
= dn
->ofs_in_node
;
40 f2fs_wait_on_page_writeback(node_page
, NODE
, false);
42 rn
= F2FS_NODE(node_page
);
44 /* Get physical address of data block */
45 addr_array
= blkaddr_in_node(rn
);
46 addr_array
[ofs_in_node
] = cpu_to_le32(new_addr
);
47 set_page_dirty(node_page
);
50 int reserve_new_block(struct dnode_of_data
*dn
)
52 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
54 if (is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
))
56 if (!inc_valid_block_count(sbi
, dn
->inode
, 1))
59 trace_f2fs_reserve_new_block(dn
->inode
, dn
->nid
, dn
->ofs_in_node
);
61 __set_data_blkaddr(dn
, NEW_ADDR
);
62 dn
->data_blkaddr
= NEW_ADDR
;
67 static int check_extent_cache(struct inode
*inode
, pgoff_t pgofs
,
68 struct buffer_head
*bh_result
)
70 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
71 pgoff_t start_fofs
, end_fofs
;
72 block_t start_blkaddr
;
74 read_lock(&fi
->ext
.ext_lock
);
75 if (fi
->ext
.len
== 0) {
76 read_unlock(&fi
->ext
.ext_lock
);
80 stat_inc_total_hit(inode
->i_sb
);
82 start_fofs
= fi
->ext
.fofs
;
83 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
84 start_blkaddr
= fi
->ext
.blk_addr
;
86 if (pgofs
>= start_fofs
&& pgofs
<= end_fofs
) {
87 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
90 clear_buffer_new(bh_result
);
91 map_bh(bh_result
, inode
->i_sb
,
92 start_blkaddr
+ pgofs
- start_fofs
);
93 count
= end_fofs
- pgofs
+ 1;
94 if (count
< (UINT_MAX
>> blkbits
))
95 bh_result
->b_size
= (count
<< blkbits
);
97 bh_result
->b_size
= UINT_MAX
;
99 stat_inc_read_hit(inode
->i_sb
);
100 read_unlock(&fi
->ext
.ext_lock
);
103 read_unlock(&fi
->ext
.ext_lock
);
107 void update_extent_cache(block_t blk_addr
, struct dnode_of_data
*dn
)
109 struct f2fs_inode_info
*fi
= F2FS_I(dn
->inode
);
110 pgoff_t fofs
, start_fofs
, end_fofs
;
111 block_t start_blkaddr
, end_blkaddr
;
113 f2fs_bug_on(blk_addr
== NEW_ADDR
);
114 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
), fi
) +
117 /* Update the page address in the parent node */
118 __set_data_blkaddr(dn
, blk_addr
);
120 write_lock(&fi
->ext
.ext_lock
);
122 start_fofs
= fi
->ext
.fofs
;
123 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
124 start_blkaddr
= fi
->ext
.blk_addr
;
125 end_blkaddr
= fi
->ext
.blk_addr
+ fi
->ext
.len
- 1;
127 /* Drop and initialize the matched extent */
128 if (fi
->ext
.len
== 1 && fofs
== start_fofs
)
132 if (fi
->ext
.len
== 0) {
133 if (blk_addr
!= NULL_ADDR
) {
135 fi
->ext
.blk_addr
= blk_addr
;
142 if (fofs
== start_fofs
- 1 && blk_addr
== start_blkaddr
- 1) {
150 if (fofs
== end_fofs
+ 1 && blk_addr
== end_blkaddr
+ 1) {
155 /* Split the existing extent */
156 if (fi
->ext
.len
> 1 &&
157 fofs
>= start_fofs
&& fofs
<= end_fofs
) {
158 if ((end_fofs
- fofs
) < (fi
->ext
.len
>> 1)) {
159 fi
->ext
.len
= fofs
- start_fofs
;
161 fi
->ext
.fofs
= fofs
+ 1;
162 fi
->ext
.blk_addr
= start_blkaddr
+
163 fofs
- start_fofs
+ 1;
164 fi
->ext
.len
-= fofs
- start_fofs
+ 1;
168 write_unlock(&fi
->ext
.ext_lock
);
172 write_unlock(&fi
->ext
.ext_lock
);
176 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
, bool sync
)
178 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
179 struct address_space
*mapping
= inode
->i_mapping
;
180 struct dnode_of_data dn
;
184 page
= find_get_page(mapping
, index
);
185 if (page
&& PageUptodate(page
))
187 f2fs_put_page(page
, 0);
189 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
190 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
195 if (dn
.data_blkaddr
== NULL_ADDR
)
196 return ERR_PTR(-ENOENT
);
198 /* By fallocate(), there is no cached page, but with NEW_ADDR */
199 if (dn
.data_blkaddr
== NEW_ADDR
)
200 return ERR_PTR(-EINVAL
);
202 page
= grab_cache_page_write_begin(mapping
, index
, AOP_FLAG_NOFS
);
204 return ERR_PTR(-ENOMEM
);
206 if (PageUptodate(page
)) {
211 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
,
212 sync
? READ_SYNC
: READA
);
214 wait_on_page_locked(page
);
215 if (!PageUptodate(page
)) {
216 f2fs_put_page(page
, 0);
217 return ERR_PTR(-EIO
);
224 * If it tries to access a hole, return an error.
225 * Because, the callers, functions in dir.c and GC, should be able to know
226 * whether this page exists or not.
228 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
)
230 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
231 struct address_space
*mapping
= inode
->i_mapping
;
232 struct dnode_of_data dn
;
237 page
= grab_cache_page_write_begin(mapping
, index
, AOP_FLAG_NOFS
);
239 return ERR_PTR(-ENOMEM
);
241 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
242 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
244 f2fs_put_page(page
, 1);
249 if (dn
.data_blkaddr
== NULL_ADDR
) {
250 f2fs_put_page(page
, 1);
251 return ERR_PTR(-ENOENT
);
254 if (PageUptodate(page
))
258 * A new dentry page is allocated but not able to be written, since its
259 * new inode page couldn't be allocated due to -ENOSPC.
260 * In such the case, its blkaddr can be remained as NEW_ADDR.
261 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
263 if (dn
.data_blkaddr
== NEW_ADDR
) {
264 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
265 SetPageUptodate(page
);
269 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
274 if (!PageUptodate(page
)) {
275 f2fs_put_page(page
, 1);
276 return ERR_PTR(-EIO
);
278 if (page
->mapping
!= mapping
) {
279 f2fs_put_page(page
, 1);
286 * Caller ensures that this data page is never allocated.
287 * A new zero-filled data page is allocated in the page cache.
289 * Also, caller should grab and release a mutex by calling mutex_lock_op() and
291 * Note that, npage is set only by make_empty_dir.
293 struct page
*get_new_data_page(struct inode
*inode
,
294 struct page
*npage
, pgoff_t index
, bool new_i_size
)
296 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
297 struct address_space
*mapping
= inode
->i_mapping
;
299 struct dnode_of_data dn
;
302 set_new_dnode(&dn
, inode
, npage
, npage
, 0);
303 err
= get_dnode_of_data(&dn
, index
, ALLOC_NODE
);
307 if (dn
.data_blkaddr
== NULL_ADDR
) {
308 if (reserve_new_block(&dn
)) {
311 return ERR_PTR(-ENOSPC
);
317 page
= grab_cache_page(mapping
, index
);
319 return ERR_PTR(-ENOMEM
);
321 if (PageUptodate(page
))
324 if (dn
.data_blkaddr
== NEW_ADDR
) {
325 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
326 SetPageUptodate(page
);
328 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
332 if (!PageUptodate(page
)) {
333 f2fs_put_page(page
, 1);
334 return ERR_PTR(-EIO
);
336 if (page
->mapping
!= mapping
) {
337 f2fs_put_page(page
, 1);
343 i_size_read(inode
) < ((index
+ 1) << PAGE_CACHE_SHIFT
)) {
344 i_size_write(inode
, ((index
+ 1) << PAGE_CACHE_SHIFT
));
345 /* Only the directory inode sets new_i_size */
346 set_inode_flag(F2FS_I(inode
), FI_UPDATE_DIR
);
347 mark_inode_dirty_sync(inode
);
352 static void read_end_io(struct bio
*bio
, int err
)
354 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
355 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
358 struct page
*page
= bvec
->bv_page
;
360 if (--bvec
>= bio
->bi_io_vec
)
361 prefetchw(&bvec
->bv_page
->flags
);
364 SetPageUptodate(page
);
366 ClearPageUptodate(page
);
370 } while (bvec
>= bio
->bi_io_vec
);
375 * Fill the locked page with data located in the block address.
376 * Return unlocked page.
378 int f2fs_readpage(struct f2fs_sb_info
*sbi
, struct page
*page
,
379 block_t blk_addr
, int type
)
381 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
384 trace_f2fs_readpage(page
, blk_addr
, type
);
386 down_read(&sbi
->bio_sem
);
388 /* Allocate a new bio */
389 bio
= f2fs_bio_alloc(bdev
, 1);
391 /* Initialize the bio */
392 bio
->bi_sector
= SECTOR_FROM_BLOCK(sbi
, blk_addr
);
393 bio
->bi_end_io
= read_end_io
;
395 if (bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) < PAGE_CACHE_SIZE
) {
397 up_read(&sbi
->bio_sem
);
398 f2fs_put_page(page
, 1);
402 submit_bio(type
, bio
);
403 up_read(&sbi
->bio_sem
);
408 * This function should be used by the data read flow only where it
409 * does not check the "create" flag that indicates block allocation.
410 * The reason for this special functionality is to exploit VFS readahead
413 static int get_data_block_ro(struct inode
*inode
, sector_t iblock
,
414 struct buffer_head
*bh_result
, int create
)
416 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
417 unsigned maxblocks
= bh_result
->b_size
>> blkbits
;
418 struct dnode_of_data dn
;
422 /* Get the page offset from the block offset(iblock) */
423 pgofs
= (pgoff_t
)(iblock
>> (PAGE_CACHE_SHIFT
- blkbits
));
425 if (check_extent_cache(inode
, pgofs
, bh_result
)) {
426 trace_f2fs_get_data_block(inode
, iblock
, bh_result
, 0);
430 /* When reading holes, we need its node page */
431 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
432 err
= get_dnode_of_data(&dn
, pgofs
, LOOKUP_NODE_RA
);
434 trace_f2fs_get_data_block(inode
, iblock
, bh_result
, err
);
435 return (err
== -ENOENT
) ? 0 : err
;
438 /* It does not support data allocation */
441 if (dn
.data_blkaddr
!= NEW_ADDR
&& dn
.data_blkaddr
!= NULL_ADDR
) {
443 unsigned int end_offset
;
445 end_offset
= IS_INODE(dn
.node_page
) ?
446 ADDRS_PER_INODE(F2FS_I(inode
)) :
449 clear_buffer_new(bh_result
);
451 /* Give more consecutive addresses for the read ahead */
452 for (i
= 0; i
< end_offset
- dn
.ofs_in_node
; i
++)
453 if (((datablock_addr(dn
.node_page
,
455 != (dn
.data_blkaddr
+ i
)) || maxblocks
== i
)
457 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
458 bh_result
->b_size
= (i
<< blkbits
);
461 trace_f2fs_get_data_block(inode
, iblock
, bh_result
, 0);
465 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
467 return mpage_readpage(page
, get_data_block_ro
);
470 static int f2fs_read_data_pages(struct file
*file
,
471 struct address_space
*mapping
,
472 struct list_head
*pages
, unsigned nr_pages
)
474 return mpage_readpages(mapping
, pages
, nr_pages
, get_data_block_ro
);
477 int do_write_data_page(struct page
*page
)
479 struct inode
*inode
= page
->mapping
->host
;
480 block_t old_blk_addr
, new_blk_addr
;
481 struct dnode_of_data dn
;
484 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
485 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
489 old_blk_addr
= dn
.data_blkaddr
;
491 /* This page is already truncated */
492 if (old_blk_addr
== NULL_ADDR
)
495 set_page_writeback(page
);
498 * If current allocation needs SSR,
499 * it had better in-place writes for updated data.
501 if (unlikely(old_blk_addr
!= NEW_ADDR
&&
502 !is_cold_data(page
) &&
503 need_inplace_update(inode
))) {
504 rewrite_data_page(F2FS_SB(inode
->i_sb
), page
,
507 write_data_page(inode
, page
, &dn
,
508 old_blk_addr
, &new_blk_addr
);
509 update_extent_cache(new_blk_addr
, &dn
);
516 static int f2fs_write_data_page(struct page
*page
,
517 struct writeback_control
*wbc
)
519 struct inode
*inode
= page
->mapping
->host
;
520 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
521 loff_t i_size
= i_size_read(inode
);
522 const pgoff_t end_index
= ((unsigned long long) i_size
)
525 bool need_balance_fs
= false;
528 if (page
->index
< end_index
)
532 * If the offset is out-of-range of file size,
533 * this page does not have to be written to disk.
535 offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
536 if ((page
->index
>= end_index
+ 1) || !offset
) {
537 if (S_ISDIR(inode
->i_mode
)) {
538 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
539 inode_dec_dirty_dents(inode
);
544 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
546 if (sbi
->por_doing
) {
547 err
= AOP_WRITEPAGE_ACTIVATE
;
551 /* Dentry blocks are controlled by checkpoint */
552 if (S_ISDIR(inode
->i_mode
)) {
553 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
554 inode_dec_dirty_dents(inode
);
555 err
= do_write_data_page(page
);
558 err
= do_write_data_page(page
);
560 need_balance_fs
= true;
567 if (wbc
->for_reclaim
)
568 f2fs_submit_bio(sbi
, DATA
, true);
570 clear_cold_data(page
);
574 f2fs_balance_fs(sbi
);
578 wbc
->pages_skipped
++;
579 set_page_dirty(page
);
583 #define MAX_DESIRED_PAGES_WP 4096
585 static int __f2fs_writepage(struct page
*page
, struct writeback_control
*wbc
,
588 struct address_space
*mapping
= data
;
589 int ret
= mapping
->a_ops
->writepage(page
, wbc
);
590 mapping_set_error(mapping
, ret
);
594 static int f2fs_write_data_pages(struct address_space
*mapping
,
595 struct writeback_control
*wbc
)
597 struct inode
*inode
= mapping
->host
;
598 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
601 long excess_nrtw
= 0, desired_nrtw
;
603 /* deal with chardevs and other special file */
604 if (!mapping
->a_ops
->writepage
)
607 if (wbc
->nr_to_write
< MAX_DESIRED_PAGES_WP
) {
608 desired_nrtw
= MAX_DESIRED_PAGES_WP
;
609 excess_nrtw
= desired_nrtw
- wbc
->nr_to_write
;
610 wbc
->nr_to_write
= desired_nrtw
;
613 if (!S_ISDIR(inode
->i_mode
)) {
614 mutex_lock(&sbi
->writepages
);
617 ret
= write_cache_pages(mapping
, wbc
, __f2fs_writepage
, mapping
);
619 mutex_unlock(&sbi
->writepages
);
620 f2fs_submit_bio(sbi
, DATA
, (wbc
->sync_mode
== WB_SYNC_ALL
));
622 remove_dirty_dir_inode(inode
);
624 wbc
->nr_to_write
-= excess_nrtw
;
628 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
629 loff_t pos
, unsigned len
, unsigned flags
,
630 struct page
**pagep
, void **fsdata
)
632 struct inode
*inode
= mapping
->host
;
633 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
635 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_CACHE_SHIFT
;
636 struct dnode_of_data dn
;
639 f2fs_balance_fs(sbi
);
641 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
648 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
649 err
= get_dnode_of_data(&dn
, index
, ALLOC_NODE
);
653 if (dn
.data_blkaddr
== NULL_ADDR
)
654 err
= reserve_new_block(&dn
);
662 if ((len
== PAGE_CACHE_SIZE
) || PageUptodate(page
))
665 if ((pos
& PAGE_CACHE_MASK
) >= i_size_read(inode
)) {
666 unsigned start
= pos
& (PAGE_CACHE_SIZE
- 1);
667 unsigned end
= start
+ len
;
669 /* Reading beyond i_size is simple: memset to zero */
670 zero_user_segments(page
, 0, start
, end
, PAGE_CACHE_SIZE
);
674 if (dn
.data_blkaddr
== NEW_ADDR
) {
675 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
677 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
681 if (!PageUptodate(page
)) {
682 f2fs_put_page(page
, 1);
685 if (page
->mapping
!= mapping
) {
686 f2fs_put_page(page
, 1);
691 SetPageUptodate(page
);
692 clear_cold_data(page
);
697 f2fs_put_page(page
, 1);
701 static int f2fs_write_end(struct file
*file
,
702 struct address_space
*mapping
,
703 loff_t pos
, unsigned len
, unsigned copied
,
704 struct page
*page
, void *fsdata
)
706 struct inode
*inode
= page
->mapping
->host
;
708 SetPageUptodate(page
);
709 set_page_dirty(page
);
711 if (pos
+ copied
> i_size_read(inode
)) {
712 i_size_write(inode
, pos
+ copied
);
713 mark_inode_dirty(inode
);
714 update_inode_page(inode
);
718 page_cache_release(page
);
722 static ssize_t
f2fs_direct_IO(int rw
, struct kiocb
*iocb
,
723 const struct iovec
*iov
, loff_t offset
, unsigned long nr_segs
)
725 struct file
*file
= iocb
->ki_filp
;
726 struct inode
*inode
= file
->f_mapping
->host
;
731 /* Needs synchronization with the cleaner */
732 return blockdev_direct_IO(rw
, iocb
, inode
, iov
, offset
, nr_segs
,
736 static void f2fs_invalidate_data_page(struct page
*page
, unsigned int offset
,
739 struct inode
*inode
= page
->mapping
->host
;
740 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
741 if (S_ISDIR(inode
->i_mode
) && PageDirty(page
)) {
742 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
743 inode_dec_dirty_dents(inode
);
745 ClearPagePrivate(page
);
748 static int f2fs_release_data_page(struct page
*page
, gfp_t wait
)
750 ClearPagePrivate(page
);
754 static int f2fs_set_data_page_dirty(struct page
*page
)
756 struct address_space
*mapping
= page
->mapping
;
757 struct inode
*inode
= mapping
->host
;
759 trace_f2fs_set_page_dirty(page
, DATA
);
761 SetPageUptodate(page
);
762 if (!PageDirty(page
)) {
763 __set_page_dirty_nobuffers(page
);
764 set_dirty_dir_page(inode
, page
);
770 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
772 return generic_block_bmap(mapping
, block
, get_data_block_ro
);
775 const struct address_space_operations f2fs_dblock_aops
= {
776 .readpage
= f2fs_read_data_page
,
777 .readpages
= f2fs_read_data_pages
,
778 .writepage
= f2fs_write_data_page
,
779 .writepages
= f2fs_write_data_pages
,
780 .write_begin
= f2fs_write_begin
,
781 .write_end
= f2fs_write_end
,
782 .set_page_dirty
= f2fs_set_data_page_dirty
,
783 .invalidatepage
= f2fs_invalidate_data_page
,
784 .releasepage
= f2fs_release_data_page
,
785 .direct_IO
= f2fs_direct_IO
,