2 * fs/logfs/dev_bdev.c - Device access methods for block devices
4 * As should be obvious for Linux kernel code, license is GPLv2
6 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
10 #include <linux/blkdev.h>
11 #include <linux/buffer_head.h>
12 #include <linux/gfp.h>
14 #define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
16 static void request_complete(struct bio
*bio
, int err
)
18 complete((struct completion
*)bio
->bi_private
);
21 static int sync_request(struct page
*page
, struct block_device
*bdev
, int rw
)
24 struct bio_vec bio_vec
;
25 struct completion complete
;
28 bio
.bi_io_vec
= &bio_vec
;
29 bio_vec
.bv_page
= page
;
30 bio_vec
.bv_len
= PAGE_SIZE
;
31 bio_vec
.bv_offset
= 0;
34 bio
.bi_size
= PAGE_SIZE
;
36 bio
.bi_sector
= page
->index
* (PAGE_SIZE
>> 9);
37 init_completion(&complete
);
38 bio
.bi_private
= &complete
;
39 bio
.bi_end_io
= request_complete
;
42 generic_unplug_device(bdev_get_queue(bdev
));
43 wait_for_completion(&complete
);
44 return test_bit(BIO_UPTODATE
, &bio
.bi_flags
) ? 0 : -EIO
;
47 static int bdev_readpage(void *_sb
, struct page
*page
)
49 struct super_block
*sb
= _sb
;
50 struct block_device
*bdev
= logfs_super(sb
)->s_bdev
;
53 err
= sync_request(page
, bdev
, READ
);
55 ClearPageUptodate(page
);
58 SetPageUptodate(page
);
65 static DECLARE_WAIT_QUEUE_HEAD(wq
);
67 static void writeseg_end_io(struct bio
*bio
, int err
)
69 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
70 struct bio_vec
*bvec
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
71 struct super_block
*sb
= bio
->bi_private
;
72 struct logfs_super
*super
= logfs_super(sb
);
75 BUG_ON(!uptodate
); /* FIXME: Retry io or write elsewhere */
77 BUG_ON(bio
->bi_vcnt
== 0);
80 if (--bvec
>= bio
->bi_io_vec
)
81 prefetchw(&bvec
->bv_page
->flags
);
83 end_page_writeback(page
);
84 } while (bvec
>= bio
->bi_io_vec
);
86 if (atomic_dec_and_test(&super
->s_pending_writes
))
90 static int __bdev_writeseg(struct super_block
*sb
, u64 ofs
, pgoff_t index
,
93 struct logfs_super
*super
= logfs_super(sb
);
94 struct address_space
*mapping
= super
->s_mapping_inode
->i_mapping
;
97 struct request_queue
*q
= bdev_get_queue(sb
->s_bdev
);
98 unsigned int max_pages
= queue_max_hw_sectors(q
) >> (PAGE_SHIFT
- 9);
101 bio
= bio_alloc(GFP_NOFS
, max_pages
);
102 BUG_ON(!bio
); /* FIXME: handle this */
104 for (i
= 0; i
< nr_pages
; i
++) {
105 if (i
>= max_pages
) {
106 /* Block layer cannot split bios :( */
109 bio
->bi_size
= i
* PAGE_SIZE
;
110 bio
->bi_bdev
= super
->s_bdev
;
111 bio
->bi_sector
= ofs
>> 9;
112 bio
->bi_private
= sb
;
113 bio
->bi_end_io
= writeseg_end_io
;
114 atomic_inc(&super
->s_pending_writes
);
115 submit_bio(WRITE
, bio
);
117 ofs
+= i
* PAGE_SIZE
;
122 bio
= bio_alloc(GFP_NOFS
, max_pages
);
125 page
= find_lock_page(mapping
, index
+ i
);
127 bio
->bi_io_vec
[i
].bv_page
= page
;
128 bio
->bi_io_vec
[i
].bv_len
= PAGE_SIZE
;
129 bio
->bi_io_vec
[i
].bv_offset
= 0;
131 BUG_ON(PageWriteback(page
));
132 set_page_writeback(page
);
135 bio
->bi_vcnt
= nr_pages
;
137 bio
->bi_size
= nr_pages
* PAGE_SIZE
;
138 bio
->bi_bdev
= super
->s_bdev
;
139 bio
->bi_sector
= ofs
>> 9;
140 bio
->bi_private
= sb
;
141 bio
->bi_end_io
= writeseg_end_io
;
142 atomic_inc(&super
->s_pending_writes
);
143 submit_bio(WRITE
, bio
);
147 static void bdev_writeseg(struct super_block
*sb
, u64 ofs
, size_t len
)
149 struct logfs_super
*super
= logfs_super(sb
);
152 BUG_ON(super
->s_flags
& LOGFS_SB_FLAG_RO
);
155 /* This can happen when the object fit perfectly into a
156 * segment, the segment gets written per sync and subsequently
161 head
= ofs
& (PAGE_SIZE
- 1);
166 len
= PAGE_ALIGN(len
);
167 __bdev_writeseg(sb
, ofs
, ofs
>> PAGE_SHIFT
, len
>> PAGE_SHIFT
);
168 generic_unplug_device(bdev_get_queue(logfs_super(sb
)->s_bdev
));
172 static void erase_end_io(struct bio
*bio
, int err
)
174 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
175 struct super_block
*sb
= bio
->bi_private
;
176 struct logfs_super
*super
= logfs_super(sb
);
178 BUG_ON(!uptodate
); /* FIXME: Retry io or write elsewhere */
180 BUG_ON(bio
->bi_vcnt
== 0);
182 if (atomic_dec_and_test(&super
->s_pending_writes
))
186 static int do_erase(struct super_block
*sb
, u64 ofs
, pgoff_t index
,
189 struct logfs_super
*super
= logfs_super(sb
);
191 struct request_queue
*q
= bdev_get_queue(sb
->s_bdev
);
192 unsigned int max_pages
= queue_max_hw_sectors(q
) >> (PAGE_SHIFT
- 9);
195 bio
= bio_alloc(GFP_NOFS
, max_pages
);
196 BUG_ON(!bio
); /* FIXME: handle this */
198 for (i
= 0; i
< nr_pages
; i
++) {
199 if (i
>= max_pages
) {
200 /* Block layer cannot split bios :( */
203 bio
->bi_size
= i
* PAGE_SIZE
;
204 bio
->bi_bdev
= super
->s_bdev
;
205 bio
->bi_sector
= ofs
>> 9;
206 bio
->bi_private
= sb
;
207 bio
->bi_end_io
= erase_end_io
;
208 atomic_inc(&super
->s_pending_writes
);
209 submit_bio(WRITE
, bio
);
211 ofs
+= i
* PAGE_SIZE
;
216 bio
= bio_alloc(GFP_NOFS
, max_pages
);
219 bio
->bi_io_vec
[i
].bv_page
= super
->s_erase_page
;
220 bio
->bi_io_vec
[i
].bv_len
= PAGE_SIZE
;
221 bio
->bi_io_vec
[i
].bv_offset
= 0;
223 bio
->bi_vcnt
= nr_pages
;
225 bio
->bi_size
= nr_pages
* PAGE_SIZE
;
226 bio
->bi_bdev
= super
->s_bdev
;
227 bio
->bi_sector
= ofs
>> 9;
228 bio
->bi_private
= sb
;
229 bio
->bi_end_io
= erase_end_io
;
230 atomic_inc(&super
->s_pending_writes
);
231 submit_bio(WRITE
, bio
);
235 static int bdev_erase(struct super_block
*sb
, loff_t to
, size_t len
,
238 struct logfs_super
*super
= logfs_super(sb
);
240 BUG_ON(to
& (PAGE_SIZE
- 1));
241 BUG_ON(len
& (PAGE_SIZE
- 1));
243 if (super
->s_flags
& LOGFS_SB_FLAG_RO
)
248 * Object store doesn't care whether erases happen or not.
249 * But for the journal they are required. Otherwise a scan
250 * can find an old commit entry and assume it is the current
251 * one, travelling back in time.
253 do_erase(sb
, to
, to
>> PAGE_SHIFT
, len
>> PAGE_SHIFT
);
259 static void bdev_sync(struct super_block
*sb
)
261 struct logfs_super
*super
= logfs_super(sb
);
263 wait_event(wq
, atomic_read(&super
->s_pending_writes
) == 0);
266 static struct page
*bdev_find_first_sb(struct super_block
*sb
, u64
*ofs
)
268 struct logfs_super
*super
= logfs_super(sb
);
269 struct address_space
*mapping
= super
->s_mapping_inode
->i_mapping
;
270 filler_t
*filler
= bdev_readpage
;
273 return read_cache_page(mapping
, 0, filler
, sb
);
276 static struct page
*bdev_find_last_sb(struct super_block
*sb
, u64
*ofs
)
278 struct logfs_super
*super
= logfs_super(sb
);
279 struct address_space
*mapping
= super
->s_mapping_inode
->i_mapping
;
280 filler_t
*filler
= bdev_readpage
;
281 u64 pos
= (super
->s_bdev
->bd_inode
->i_size
& ~0xfffULL
) - 0x1000;
282 pgoff_t index
= pos
>> PAGE_SHIFT
;
285 return read_cache_page(mapping
, index
, filler
, sb
);
288 static int bdev_write_sb(struct super_block
*sb
, struct page
*page
)
290 struct block_device
*bdev
= logfs_super(sb
)->s_bdev
;
292 /* Nothing special to do for block devices. */
293 return sync_request(page
, bdev
, WRITE
);
296 static void bdev_put_device(struct super_block
*sb
)
298 close_bdev_exclusive(logfs_super(sb
)->s_bdev
, FMODE_READ
|FMODE_WRITE
);
301 static const struct logfs_device_ops bd_devops
= {
302 .find_first_sb
= bdev_find_first_sb
,
303 .find_last_sb
= bdev_find_last_sb
,
304 .write_sb
= bdev_write_sb
,
305 .readpage
= bdev_readpage
,
306 .writeseg
= bdev_writeseg
,
309 .put_device
= bdev_put_device
,
312 int logfs_get_sb_bdev(struct file_system_type
*type
, int flags
,
313 const char *devname
, struct vfsmount
*mnt
)
315 struct block_device
*bdev
;
317 bdev
= open_bdev_exclusive(devname
, FMODE_READ
|FMODE_WRITE
, type
);
319 return PTR_ERR(bdev
);
321 if (MAJOR(bdev
->bd_dev
) == MTD_BLOCK_MAJOR
) {
322 int mtdnr
= MINOR(bdev
->bd_dev
);
323 close_bdev_exclusive(bdev
, FMODE_READ
|FMODE_WRITE
);
324 return logfs_get_sb_mtd(type
, flags
, mtdnr
, mnt
);
327 return logfs_get_sb_device(type
, flags
, NULL
, bdev
, &bd_devops
, mnt
);