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>
13 #define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
15 static void request_complete(struct bio *bio, int err)
17 complete((struct completion *)bio->bi_private);
20 static int sync_request(struct page *page, struct block_device *bdev, int rw)
23 struct bio_vec bio_vec;
24 struct completion complete;
27 bio.bi_io_vec = &bio_vec;
28 bio_vec.bv_page = page;
29 bio_vec.bv_len = PAGE_SIZE;
30 bio_vec.bv_offset = 0;
33 bio.bi_size = PAGE_SIZE;
35 bio.bi_sector = page->index * (PAGE_SIZE >> 9);
36 init_completion(&complete);
37 bio.bi_private = &complete;
38 bio.bi_end_io = request_complete;
41 generic_unplug_device(bdev_get_queue(bdev));
42 wait_for_completion(&complete);
43 return test_bit(BIO_UPTODATE, &bio.bi_flags) ? 0 : -EIO;
46 static int bdev_readpage(void *_sb, struct page *page)
48 struct super_block *sb = _sb;
49 struct block_device *bdev = logfs_super(sb)->s_bdev;
52 err = sync_request(page, bdev, READ);
54 ClearPageUptodate(page);
57 SetPageUptodate(page);
64 static DECLARE_WAIT_QUEUE_HEAD(wq);
66 static void writeseg_end_io(struct bio *bio, int err)
68 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
69 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
70 struct super_block *sb = bio->bi_private;
71 struct logfs_super *super = logfs_super(sb);
74 BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
76 BUG_ON(bio->bi_vcnt == 0);
79 if (--bvec >= bio->bi_io_vec)
80 prefetchw(&bvec->bv_page->flags);
82 end_page_writeback(page);
83 page_cache_release(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 if (max_pages > BIO_MAX_PAGES)
102 max_pages = BIO_MAX_PAGES;
103 bio = bio_alloc(GFP_NOFS, max_pages);
106 for (i = 0; i < nr_pages; i++) {
107 if (i >= max_pages) {
108 /* Block layer cannot split bios :( */
111 bio->bi_size = i * PAGE_SIZE;
112 bio->bi_bdev = super->s_bdev;
113 bio->bi_sector = ofs >> 9;
114 bio->bi_private = sb;
115 bio->bi_end_io = writeseg_end_io;
116 atomic_inc(&super->s_pending_writes);
117 submit_bio(WRITE, bio);
119 ofs += i * PAGE_SIZE;
124 bio = bio_alloc(GFP_NOFS, max_pages);
127 page = find_lock_page(mapping, index + i);
129 bio->bi_io_vec[i].bv_page = page;
130 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
131 bio->bi_io_vec[i].bv_offset = 0;
133 BUG_ON(PageWriteback(page));
134 set_page_writeback(page);
137 bio->bi_vcnt = nr_pages;
139 bio->bi_size = nr_pages * PAGE_SIZE;
140 bio->bi_bdev = super->s_bdev;
141 bio->bi_sector = ofs >> 9;
142 bio->bi_private = sb;
143 bio->bi_end_io = writeseg_end_io;
144 atomic_inc(&super->s_pending_writes);
145 submit_bio(WRITE, bio);
149 static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
151 struct logfs_super *super = logfs_super(sb);
154 BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
157 /* This can happen when the object fit perfectly into a
158 * segment, the segment gets written per sync and subsequently
163 head = ofs & (PAGE_SIZE - 1);
168 len = PAGE_ALIGN(len);
169 __bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
170 generic_unplug_device(bdev_get_queue(logfs_super(sb)->s_bdev));
174 static void erase_end_io(struct bio *bio, int err)
176 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
177 struct super_block *sb = bio->bi_private;
178 struct logfs_super *super = logfs_super(sb);
180 BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
182 BUG_ON(bio->bi_vcnt == 0);
184 if (atomic_dec_and_test(&super->s_pending_writes))
188 static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
191 struct logfs_super *super = logfs_super(sb);
193 struct request_queue *q = bdev_get_queue(sb->s_bdev);
194 unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
197 if (max_pages > BIO_MAX_PAGES)
198 max_pages = BIO_MAX_PAGES;
199 bio = bio_alloc(GFP_NOFS, max_pages);
202 for (i = 0; i < nr_pages; i++) {
203 if (i >= max_pages) {
204 /* Block layer cannot split bios :( */
207 bio->bi_size = i * PAGE_SIZE;
208 bio->bi_bdev = super->s_bdev;
209 bio->bi_sector = ofs >> 9;
210 bio->bi_private = sb;
211 bio->bi_end_io = erase_end_io;
212 atomic_inc(&super->s_pending_writes);
213 submit_bio(WRITE, bio);
215 ofs += i * PAGE_SIZE;
220 bio = bio_alloc(GFP_NOFS, max_pages);
223 bio->bi_io_vec[i].bv_page = super->s_erase_page;
224 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
225 bio->bi_io_vec[i].bv_offset = 0;
227 bio->bi_vcnt = nr_pages;
229 bio->bi_size = nr_pages * PAGE_SIZE;
230 bio->bi_bdev = super->s_bdev;
231 bio->bi_sector = ofs >> 9;
232 bio->bi_private = sb;
233 bio->bi_end_io = erase_end_io;
234 atomic_inc(&super->s_pending_writes);
235 submit_bio(WRITE, bio);
239 static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
242 struct logfs_super *super = logfs_super(sb);
244 BUG_ON(to & (PAGE_SIZE - 1));
245 BUG_ON(len & (PAGE_SIZE - 1));
247 if (super->s_flags & LOGFS_SB_FLAG_RO)
252 * Object store doesn't care whether erases happen or not.
253 * But for the journal they are required. Otherwise a scan
254 * can find an old commit entry and assume it is the current
255 * one, travelling back in time.
257 do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
263 static void bdev_sync(struct super_block *sb)
265 struct logfs_super *super = logfs_super(sb);
267 wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
270 static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
272 struct logfs_super *super = logfs_super(sb);
273 struct address_space *mapping = super->s_mapping_inode->i_mapping;
274 filler_t *filler = bdev_readpage;
277 return read_cache_page(mapping, 0, filler, sb);
280 static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
282 struct logfs_super *super = logfs_super(sb);
283 struct address_space *mapping = super->s_mapping_inode->i_mapping;
284 filler_t *filler = bdev_readpage;
285 u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
286 pgoff_t index = pos >> PAGE_SHIFT;
289 return read_cache_page(mapping, index, filler, sb);
292 static int bdev_write_sb(struct super_block *sb, struct page *page)
294 struct block_device *bdev = logfs_super(sb)->s_bdev;
296 /* Nothing special to do for block devices. */
297 return sync_request(page, bdev, WRITE);
300 static void bdev_put_device(struct super_block *sb)
302 close_bdev_exclusive(logfs_super(sb)->s_bdev, FMODE_READ|FMODE_WRITE);
305 static const struct logfs_device_ops bd_devops = {
306 .find_first_sb = bdev_find_first_sb,
307 .find_last_sb = bdev_find_last_sb,
308 .write_sb = bdev_write_sb,
309 .readpage = bdev_readpage,
310 .writeseg = bdev_writeseg,
313 .put_device = bdev_put_device,
316 int logfs_get_sb_bdev(struct file_system_type *type, int flags,
317 const char *devname, struct vfsmount *mnt)
319 struct block_device *bdev;
321 bdev = open_bdev_exclusive(devname, FMODE_READ|FMODE_WRITE, type);
323 return PTR_ERR(bdev);
325 if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
326 int mtdnr = MINOR(bdev->bd_dev);
327 close_bdev_exclusive(bdev, FMODE_READ|FMODE_WRITE);
328 return logfs_get_sb_mtd(type, flags, mtdnr, mnt);
331 return logfs_get_sb_device(type, flags, NULL, bdev, &bd_devops, mnt);