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/blkdev.h>
18 #include <linux/bio.h>
19 #include <linux/prefetch.h>
26 * Lock ordering for the change of data block address:
29 * update block addresses in the node page
31 static void __set_data_blkaddr(struct dnode_of_data
*dn
, block_t new_addr
)
35 struct page
*node_page
= dn
->node_page
;
36 unsigned int ofs_in_node
= dn
->ofs_in_node
;
38 wait_on_page_writeback(node_page
);
40 rn
= (struct f2fs_node
*)page_address(node_page
);
42 /* Get physical address of data block */
43 addr_array
= blkaddr_in_node(rn
);
44 addr_array
[ofs_in_node
] = cpu_to_le32(new_addr
);
45 set_page_dirty(node_page
);
48 int reserve_new_block(struct dnode_of_data
*dn
)
50 struct f2fs_sb_info
*sbi
= F2FS_SB(dn
->inode
->i_sb
);
52 if (is_inode_flag_set(F2FS_I(dn
->inode
), FI_NO_ALLOC
))
54 if (!inc_valid_block_count(sbi
, dn
->inode
, 1))
57 __set_data_blkaddr(dn
, NEW_ADDR
);
58 dn
->data_blkaddr
= NEW_ADDR
;
63 static int check_extent_cache(struct inode
*inode
, pgoff_t pgofs
,
64 struct buffer_head
*bh_result
)
66 struct f2fs_inode_info
*fi
= F2FS_I(inode
);
67 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
68 pgoff_t start_fofs
, end_fofs
;
69 block_t start_blkaddr
;
71 read_lock(&fi
->ext
.ext_lock
);
72 if (fi
->ext
.len
== 0) {
73 read_unlock(&fi
->ext
.ext_lock
);
78 start_fofs
= fi
->ext
.fofs
;
79 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
80 start_blkaddr
= fi
->ext
.blk_addr
;
82 if (pgofs
>= start_fofs
&& pgofs
<= end_fofs
) {
83 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
86 clear_buffer_new(bh_result
);
87 map_bh(bh_result
, inode
->i_sb
,
88 start_blkaddr
+ pgofs
- start_fofs
);
89 count
= end_fofs
- pgofs
+ 1;
90 if (count
< (UINT_MAX
>> blkbits
))
91 bh_result
->b_size
= (count
<< blkbits
);
93 bh_result
->b_size
= UINT_MAX
;
96 read_unlock(&fi
->ext
.ext_lock
);
99 read_unlock(&fi
->ext
.ext_lock
);
103 void update_extent_cache(block_t blk_addr
, struct dnode_of_data
*dn
)
105 struct f2fs_inode_info
*fi
= F2FS_I(dn
->inode
);
106 pgoff_t fofs
, start_fofs
, end_fofs
;
107 block_t start_blkaddr
, end_blkaddr
;
109 BUG_ON(blk_addr
== NEW_ADDR
);
110 fofs
= start_bidx_of_node(ofs_of_node(dn
->node_page
)) + dn
->ofs_in_node
;
112 /* Update the page address in the parent node */
113 __set_data_blkaddr(dn
, blk_addr
);
115 write_lock(&fi
->ext
.ext_lock
);
117 start_fofs
= fi
->ext
.fofs
;
118 end_fofs
= fi
->ext
.fofs
+ fi
->ext
.len
- 1;
119 start_blkaddr
= fi
->ext
.blk_addr
;
120 end_blkaddr
= fi
->ext
.blk_addr
+ fi
->ext
.len
- 1;
122 /* Drop and initialize the matched extent */
123 if (fi
->ext
.len
== 1 && fofs
== start_fofs
)
127 if (fi
->ext
.len
== 0) {
128 if (blk_addr
!= NULL_ADDR
) {
130 fi
->ext
.blk_addr
= blk_addr
;
137 if (fofs
== start_fofs
- 1 && blk_addr
== start_blkaddr
- 1) {
145 if (fofs
== end_fofs
+ 1 && blk_addr
== end_blkaddr
+ 1) {
150 /* Split the existing extent */
151 if (fi
->ext
.len
> 1 &&
152 fofs
>= start_fofs
&& fofs
<= end_fofs
) {
153 if ((end_fofs
- fofs
) < (fi
->ext
.len
>> 1)) {
154 fi
->ext
.len
= fofs
- start_fofs
;
156 fi
->ext
.fofs
= fofs
+ 1;
157 fi
->ext
.blk_addr
= start_blkaddr
+
158 fofs
- start_fofs
+ 1;
159 fi
->ext
.len
-= fofs
- start_fofs
+ 1;
163 write_unlock(&fi
->ext
.ext_lock
);
167 write_unlock(&fi
->ext
.ext_lock
);
172 struct page
*find_data_page(struct inode
*inode
, pgoff_t index
)
174 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
175 struct address_space
*mapping
= inode
->i_mapping
;
176 struct dnode_of_data dn
;
180 page
= find_get_page(mapping
, index
);
181 if (page
&& PageUptodate(page
))
183 f2fs_put_page(page
, 0);
185 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
186 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
191 if (dn
.data_blkaddr
== NULL_ADDR
)
192 return ERR_PTR(-ENOENT
);
194 /* By fallocate(), there is no cached page, but with NEW_ADDR */
195 if (dn
.data_blkaddr
== NEW_ADDR
)
196 return ERR_PTR(-EINVAL
);
198 page
= grab_cache_page(mapping
, index
);
200 return ERR_PTR(-ENOMEM
);
202 if (PageUptodate(page
)) {
207 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
208 wait_on_page_locked(page
);
209 if (!PageUptodate(page
)) {
210 f2fs_put_page(page
, 0);
211 return ERR_PTR(-EIO
);
217 * If it tries to access a hole, return an error.
218 * Because, the callers, functions in dir.c and GC, should be able to know
219 * whether this page exists or not.
221 struct page
*get_lock_data_page(struct inode
*inode
, pgoff_t index
)
223 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
224 struct address_space
*mapping
= inode
->i_mapping
;
225 struct dnode_of_data dn
;
229 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
230 err
= get_dnode_of_data(&dn
, index
, LOOKUP_NODE
);
235 if (dn
.data_blkaddr
== NULL_ADDR
)
236 return ERR_PTR(-ENOENT
);
238 page
= grab_cache_page(mapping
, index
);
240 return ERR_PTR(-ENOMEM
);
242 if (PageUptodate(page
))
245 BUG_ON(dn
.data_blkaddr
== NEW_ADDR
);
246 BUG_ON(dn
.data_blkaddr
== NULL_ADDR
);
248 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
253 if (!PageUptodate(page
)) {
254 f2fs_put_page(page
, 1);
255 return ERR_PTR(-EIO
);
261 * Caller ensures that this data page is never allocated.
262 * A new zero-filled data page is allocated in the page cache.
264 struct page
*get_new_data_page(struct inode
*inode
, pgoff_t index
,
267 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
268 struct address_space
*mapping
= inode
->i_mapping
;
270 struct dnode_of_data dn
;
273 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
274 err
= get_dnode_of_data(&dn
, index
, ALLOC_NODE
);
278 if (dn
.data_blkaddr
== NULL_ADDR
) {
279 if (reserve_new_block(&dn
)) {
281 return ERR_PTR(-ENOSPC
);
286 page
= grab_cache_page(mapping
, index
);
288 return ERR_PTR(-ENOMEM
);
290 if (PageUptodate(page
))
293 if (dn
.data_blkaddr
== NEW_ADDR
) {
294 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
295 SetPageUptodate(page
);
297 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
301 if (!PageUptodate(page
)) {
302 f2fs_put_page(page
, 1);
303 return ERR_PTR(-EIO
);
308 i_size_read(inode
) < ((index
+ 1) << PAGE_CACHE_SHIFT
)) {
309 i_size_write(inode
, ((index
+ 1) << PAGE_CACHE_SHIFT
));
310 mark_inode_dirty_sync(inode
);
315 static void read_end_io(struct bio
*bio
, int err
)
317 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
318 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
321 struct page
*page
= bvec
->bv_page
;
323 if (--bvec
>= bio
->bi_io_vec
)
324 prefetchw(&bvec
->bv_page
->flags
);
327 SetPageUptodate(page
);
329 ClearPageUptodate(page
);
333 } while (bvec
>= bio
->bi_io_vec
);
334 kfree(bio
->bi_private
);
339 * Fill the locked page with data located in the block address.
340 * Return unlocked page.
342 int f2fs_readpage(struct f2fs_sb_info
*sbi
, struct page
*page
,
343 block_t blk_addr
, int type
)
345 struct block_device
*bdev
= sbi
->sb
->s_bdev
;
348 down_read(&sbi
->bio_sem
);
350 /* Allocate a new bio */
351 bio
= f2fs_bio_alloc(bdev
, 1);
353 /* Initialize the bio */
354 bio
->bi_sector
= SECTOR_FROM_BLOCK(sbi
, blk_addr
);
355 bio
->bi_end_io
= read_end_io
;
357 if (bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) < PAGE_CACHE_SIZE
) {
358 kfree(bio
->bi_private
);
360 up_read(&sbi
->bio_sem
);
361 f2fs_put_page(page
, 1);
365 submit_bio(type
, bio
);
366 up_read(&sbi
->bio_sem
);
371 * This function should be used by the data read flow only where it
372 * does not check the "create" flag that indicates block allocation.
373 * The reason for this special functionality is to exploit VFS readahead
376 static int get_data_block_ro(struct inode
*inode
, sector_t iblock
,
377 struct buffer_head
*bh_result
, int create
)
379 unsigned int blkbits
= inode
->i_sb
->s_blocksize_bits
;
380 unsigned maxblocks
= bh_result
->b_size
>> blkbits
;
381 struct dnode_of_data dn
;
385 /* Get the page offset from the block offset(iblock) */
386 pgofs
= (pgoff_t
)(iblock
>> (PAGE_CACHE_SHIFT
- blkbits
));
388 if (check_extent_cache(inode
, pgofs
, bh_result
))
391 /* When reading holes, we need its node page */
392 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
393 err
= get_dnode_of_data(&dn
, pgofs
, LOOKUP_NODE_RA
);
395 return (err
== -ENOENT
) ? 0 : err
;
397 /* It does not support data allocation */
400 if (dn
.data_blkaddr
!= NEW_ADDR
&& dn
.data_blkaddr
!= NULL_ADDR
) {
402 unsigned int end_offset
;
404 end_offset
= IS_INODE(dn
.node_page
) ?
408 clear_buffer_new(bh_result
);
410 /* Give more consecutive addresses for the read ahead */
411 for (i
= 0; i
< end_offset
- dn
.ofs_in_node
; i
++)
412 if (((datablock_addr(dn
.node_page
,
414 != (dn
.data_blkaddr
+ i
)) || maxblocks
== i
)
416 map_bh(bh_result
, inode
->i_sb
, dn
.data_blkaddr
);
417 bh_result
->b_size
= (i
<< blkbits
);
423 static int f2fs_read_data_page(struct file
*file
, struct page
*page
)
425 return mpage_readpage(page
, get_data_block_ro
);
428 static int f2fs_read_data_pages(struct file
*file
,
429 struct address_space
*mapping
,
430 struct list_head
*pages
, unsigned nr_pages
)
432 return mpage_readpages(mapping
, pages
, nr_pages
, get_data_block_ro
);
435 int do_write_data_page(struct page
*page
)
437 struct inode
*inode
= page
->mapping
->host
;
438 block_t old_blk_addr
, new_blk_addr
;
439 struct dnode_of_data dn
;
442 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
443 err
= get_dnode_of_data(&dn
, page
->index
, LOOKUP_NODE
);
447 old_blk_addr
= dn
.data_blkaddr
;
449 /* This page is already truncated */
450 if (old_blk_addr
== NULL_ADDR
)
453 set_page_writeback(page
);
456 * If current allocation needs SSR,
457 * it had better in-place writes for updated data.
459 if (old_blk_addr
!= NEW_ADDR
&& !is_cold_data(page
) &&
460 need_inplace_update(inode
)) {
461 rewrite_data_page(F2FS_SB(inode
->i_sb
), page
,
464 write_data_page(inode
, page
, &dn
,
465 old_blk_addr
, &new_blk_addr
);
466 update_extent_cache(new_blk_addr
, &dn
);
473 static int f2fs_write_data_page(struct page
*page
,
474 struct writeback_control
*wbc
)
476 struct inode
*inode
= page
->mapping
->host
;
477 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
478 loff_t i_size
= i_size_read(inode
);
479 const pgoff_t end_index
= ((unsigned long long) i_size
)
484 if (page
->index
< end_index
)
488 * If the offset is out-of-range of file size,
489 * this page does not have to be written to disk.
491 offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
492 if ((page
->index
>= end_index
+ 1) || !offset
) {
493 if (S_ISDIR(inode
->i_mode
)) {
494 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
495 inode_dec_dirty_dents(inode
);
500 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
505 if (wbc
->for_reclaim
&& !S_ISDIR(inode
->i_mode
) && !is_cold_data(page
))
508 mutex_lock_op(sbi
, DATA_WRITE
);
509 if (S_ISDIR(inode
->i_mode
)) {
510 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
511 inode_dec_dirty_dents(inode
);
513 err
= do_write_data_page(page
);
514 if (err
&& err
!= -ENOENT
) {
515 wbc
->pages_skipped
++;
516 set_page_dirty(page
);
518 mutex_unlock_op(sbi
, DATA_WRITE
);
520 if (wbc
->for_reclaim
)
521 f2fs_submit_bio(sbi
, DATA
, true);
526 clear_cold_data(page
);
529 if (!wbc
->for_reclaim
&& !S_ISDIR(inode
->i_mode
))
530 f2fs_balance_fs(sbi
);
535 return (err
== -ENOENT
) ? 0 : err
;
538 wbc
->pages_skipped
++;
539 set_page_dirty(page
);
540 return AOP_WRITEPAGE_ACTIVATE
;
543 #define MAX_DESIRED_PAGES_WP 4096
545 static int __f2fs_writepage(struct page
*page
, struct writeback_control
*wbc
,
548 struct address_space
*mapping
= data
;
549 int ret
= mapping
->a_ops
->writepage(page
, wbc
);
550 mapping_set_error(mapping
, ret
);
554 static int f2fs_write_data_pages(struct address_space
*mapping
,
555 struct writeback_control
*wbc
)
557 struct inode
*inode
= mapping
->host
;
558 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
560 long excess_nrtw
= 0, desired_nrtw
;
562 /* deal with chardevs and other special file */
563 if (!mapping
->a_ops
->writepage
)
566 if (wbc
->nr_to_write
< MAX_DESIRED_PAGES_WP
) {
567 desired_nrtw
= MAX_DESIRED_PAGES_WP
;
568 excess_nrtw
= desired_nrtw
- wbc
->nr_to_write
;
569 wbc
->nr_to_write
= desired_nrtw
;
572 if (!S_ISDIR(inode
->i_mode
))
573 mutex_lock(&sbi
->writepages
);
574 ret
= write_cache_pages(mapping
, wbc
, __f2fs_writepage
, mapping
);
575 if (!S_ISDIR(inode
->i_mode
))
576 mutex_unlock(&sbi
->writepages
);
577 f2fs_submit_bio(sbi
, DATA
, (wbc
->sync_mode
== WB_SYNC_ALL
));
579 remove_dirty_dir_inode(inode
);
581 wbc
->nr_to_write
-= excess_nrtw
;
585 static int f2fs_write_begin(struct file
*file
, struct address_space
*mapping
,
586 loff_t pos
, unsigned len
, unsigned flags
,
587 struct page
**pagep
, void **fsdata
)
589 struct inode
*inode
= mapping
->host
;
590 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
592 pgoff_t index
= ((unsigned long long) pos
) >> PAGE_CACHE_SHIFT
;
593 struct dnode_of_data dn
;
596 /* for nobh_write_end */
599 f2fs_balance_fs(sbi
);
601 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
606 mutex_lock_op(sbi
, DATA_NEW
);
608 set_new_dnode(&dn
, inode
, NULL
, NULL
, 0);
609 err
= get_dnode_of_data(&dn
, index
, ALLOC_NODE
);
611 mutex_unlock_op(sbi
, DATA_NEW
);
612 f2fs_put_page(page
, 1);
616 if (dn
.data_blkaddr
== NULL_ADDR
) {
617 err
= reserve_new_block(&dn
);
620 mutex_unlock_op(sbi
, DATA_NEW
);
621 f2fs_put_page(page
, 1);
627 mutex_unlock_op(sbi
, DATA_NEW
);
629 if ((len
== PAGE_CACHE_SIZE
) || PageUptodate(page
))
632 if ((pos
& PAGE_CACHE_MASK
) >= i_size_read(inode
)) {
633 unsigned start
= pos
& (PAGE_CACHE_SIZE
- 1);
634 unsigned end
= start
+ len
;
636 /* Reading beyond i_size is simple: memset to zero */
637 zero_user_segments(page
, 0, start
, end
, PAGE_CACHE_SIZE
);
641 if (dn
.data_blkaddr
== NEW_ADDR
) {
642 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
644 err
= f2fs_readpage(sbi
, page
, dn
.data_blkaddr
, READ_SYNC
);
648 if (!PageUptodate(page
)) {
649 f2fs_put_page(page
, 1);
654 SetPageUptodate(page
);
655 clear_cold_data(page
);
659 static ssize_t
f2fs_direct_IO(int rw
, struct kiocb
*iocb
,
660 const struct iovec
*iov
, loff_t offset
, unsigned long nr_segs
)
662 struct file
*file
= iocb
->ki_filp
;
663 struct inode
*inode
= file
->f_mapping
->host
;
668 /* Needs synchronization with the cleaner */
669 return blockdev_direct_IO(rw
, iocb
, inode
, iov
, offset
, nr_segs
,
673 static void f2fs_invalidate_data_page(struct page
*page
, unsigned long offset
)
675 struct inode
*inode
= page
->mapping
->host
;
676 struct f2fs_sb_info
*sbi
= F2FS_SB(inode
->i_sb
);
677 if (S_ISDIR(inode
->i_mode
) && PageDirty(page
)) {
678 dec_page_count(sbi
, F2FS_DIRTY_DENTS
);
679 inode_dec_dirty_dents(inode
);
681 ClearPagePrivate(page
);
684 static int f2fs_release_data_page(struct page
*page
, gfp_t wait
)
686 ClearPagePrivate(page
);
690 static int f2fs_set_data_page_dirty(struct page
*page
)
692 struct address_space
*mapping
= page
->mapping
;
693 struct inode
*inode
= mapping
->host
;
695 SetPageUptodate(page
);
696 if (!PageDirty(page
)) {
697 __set_page_dirty_nobuffers(page
);
698 set_dirty_dir_page(inode
, page
);
704 static sector_t
f2fs_bmap(struct address_space
*mapping
, sector_t block
)
706 return generic_block_bmap(mapping
, block
, get_data_block_ro
);
709 const struct address_space_operations f2fs_dblock_aops
= {
710 .readpage
= f2fs_read_data_page
,
711 .readpages
= f2fs_read_data_pages
,
712 .writepage
= f2fs_write_data_page
,
713 .writepages
= f2fs_write_data_pages
,
714 .write_begin
= f2fs_write_begin
,
715 .write_end
= nobh_write_end
,
716 .set_page_dirty
= f2fs_set_data_page_dirty
,
717 .invalidatepage
= f2fs_invalidate_data_page
,
718 .releasepage
= f2fs_release_data_page
,
719 .direct_IO
= f2fs_direct_IO
,