1 /*
2 * fs/logfs/dev_bdev.c - Device access methods for block devices
3 *
4 * As should be obvious for Linux kernel code, license is GPLv2
5 *
6 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
7 */
8 #include "logfs.h"
9 #include <linux/bio.h>
10 #include <linux/blkdev.h>
11 #include <linux/buffer_head.h>
12 #include <linux/gfp.h>
13
14 #define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
15
request_complete(struct bio * bio,int err)16 static void request_complete(struct bio *bio, int err)
17 {
18 complete((struct completion *)bio->bi_private);
19 }
20
sync_request(struct page * page,struct block_device * bdev,int rw)21 static int sync_request(struct page *page, struct block_device *bdev, int rw)
22 {
23 struct bio bio;
24 struct bio_vec bio_vec;
25 struct completion complete;
26
27 bio_init(&bio);
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;
32 bio.bi_vcnt = 1;
33 bio.bi_idx = 0;
34 bio.bi_size = PAGE_SIZE;
35 bio.bi_bdev = bdev;
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;
40
41 submit_bio(rw, &bio);
42 wait_for_completion(&complete);
43 return test_bit(BIO_UPTODATE, &bio.bi_flags) ? 0 : -EIO;
44 }
45
bdev_readpage(void * _sb,struct page * page)46 static int bdev_readpage(void *_sb, struct page *page)
47 {
48 struct super_block *sb = _sb;
49 struct block_device *bdev = logfs_super(sb)->s_bdev;
50 int err;
51
52 err = sync_request(page, bdev, READ);
53 if (err) {
54 ClearPageUptodate(page);
55 SetPageError(page);
56 } else {
57 SetPageUptodate(page);
58 ClearPageError(page);
59 }
60 unlock_page(page);
61 return err;
62 }
63
64 static DECLARE_WAIT_QUEUE_HEAD(wq);
65
writeseg_end_io(struct bio * bio,int err)66 static void writeseg_end_io(struct bio *bio, int err)
67 {
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);
72 struct page *page;
73
74 BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
75 BUG_ON(err);
76 BUG_ON(bio->bi_vcnt == 0);
77 do {
78 page = bvec->bv_page;
79 if (--bvec >= bio->bi_io_vec)
80 prefetchw(&bvec->bv_page->flags);
81
82 end_page_writeback(page);
83 page_cache_release(page);
84 } while (bvec >= bio->bi_io_vec);
85 bio_put(bio);
86 if (atomic_dec_and_test(&super->s_pending_writes))
87 wake_up(&wq);
88 }
89
__bdev_writeseg(struct super_block * sb,u64 ofs,pgoff_t index,size_t nr_pages)90 static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
91 size_t nr_pages)
92 {
93 struct logfs_super *super = logfs_super(sb);
94 struct address_space *mapping = super->s_mapping_inode->i_mapping;
95 struct bio *bio;
96 struct page *page;
97 struct request_queue *q = bdev_get_queue(sb->s_bdev);
98 unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
99 int i;
100
101 if (max_pages > BIO_MAX_PAGES)
102 max_pages = BIO_MAX_PAGES;
103 bio = bio_alloc(GFP_NOFS, max_pages);
104 BUG_ON(!bio);
105
106 for (i = 0; i < nr_pages; i++) {
107 if (i >= max_pages) {
108 /* Block layer cannot split bios :( */
109 bio->bi_vcnt = i;
110 bio->bi_idx = 0;
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);
118
119 ofs += i * PAGE_SIZE;
120 index += i;
121 nr_pages -= i;
122 i = 0;
123
124 bio = bio_alloc(GFP_NOFS, max_pages);
125 BUG_ON(!bio);
126 }
127 page = find_lock_page(mapping, index + i);
128 BUG_ON(!page);
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;
132
133 BUG_ON(PageWriteback(page));
134 set_page_writeback(page);
135 unlock_page(page);
136 }
137 bio->bi_vcnt = nr_pages;
138 bio->bi_idx = 0;
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);
146 return 0;
147 }
148
bdev_writeseg(struct super_block * sb,u64 ofs,size_t len)149 static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
150 {
151 struct logfs_super *super = logfs_super(sb);
152 int head;
153
154 BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
155
156 if (len == 0) {
157 /* This can happen when the object fit perfectly into a
158 * segment, the segment gets written per sync and subsequently
159 * closed.
160 */
161 return;
162 }
163 head = ofs & (PAGE_SIZE - 1);
164 if (head) {
165 ofs -= head;
166 len += head;
167 }
168 len = PAGE_ALIGN(len);
169 __bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
170 }
171
172
erase_end_io(struct bio * bio,int err)173 static void erase_end_io(struct bio *bio, int err)
174 {
175 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
176 struct super_block *sb = bio->bi_private;
177 struct logfs_super *super = logfs_super(sb);
178
179 BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
180 BUG_ON(err);
181 BUG_ON(bio->bi_vcnt == 0);
182 bio_put(bio);
183 if (atomic_dec_and_test(&super->s_pending_writes))
184 wake_up(&wq);
185 }
186
do_erase(struct super_block * sb,u64 ofs,pgoff_t index,size_t nr_pages)187 static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
188 size_t nr_pages)
189 {
190 struct logfs_super *super = logfs_super(sb);
191 struct bio *bio;
192 struct request_queue *q = bdev_get_queue(sb->s_bdev);
193 unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
194 int i;
195
196 if (max_pages > BIO_MAX_PAGES)
197 max_pages = BIO_MAX_PAGES;
198 bio = bio_alloc(GFP_NOFS, max_pages);
199 BUG_ON(!bio);
200
201 for (i = 0; i < nr_pages; i++) {
202 if (i >= max_pages) {
203 /* Block layer cannot split bios :( */
204 bio->bi_vcnt = i;
205 bio->bi_idx = 0;
206 bio->bi_size = i * PAGE_SIZE;
207 bio->bi_bdev = super->s_bdev;
208 bio->bi_sector = ofs >> 9;
209 bio->bi_private = sb;
210 bio->bi_end_io = erase_end_io;
211 atomic_inc(&super->s_pending_writes);
212 submit_bio(WRITE, bio);
213
214 ofs += i * PAGE_SIZE;
215 index += i;
216 nr_pages -= i;
217 i = 0;
218
219 bio = bio_alloc(GFP_NOFS, max_pages);
220 BUG_ON(!bio);
221 }
222 bio->bi_io_vec[i].bv_page = super->s_erase_page;
223 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
224 bio->bi_io_vec[i].bv_offset = 0;
225 }
226 bio->bi_vcnt = nr_pages;
227 bio->bi_idx = 0;
228 bio->bi_size = nr_pages * PAGE_SIZE;
229 bio->bi_bdev = super->s_bdev;
230 bio->bi_sector = ofs >> 9;
231 bio->bi_private = sb;
232 bio->bi_end_io = erase_end_io;
233 atomic_inc(&super->s_pending_writes);
234 submit_bio(WRITE, bio);
235 return 0;
236 }
237
bdev_erase(struct super_block * sb,loff_t to,size_t len,int ensure_write)238 static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
239 int ensure_write)
240 {
241 struct logfs_super *super = logfs_super(sb);
242
243 BUG_ON(to & (PAGE_SIZE - 1));
244 BUG_ON(len & (PAGE_SIZE - 1));
245
246 if (super->s_flags & LOGFS_SB_FLAG_RO)
247 return -EROFS;
248
249 if (ensure_write) {
250 /*
251 * Object store doesn't care whether erases happen or not.
252 * But for the journal they are required. Otherwise a scan
253 * can find an old commit entry and assume it is the current
254 * one, travelling back in time.
255 */
256 do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
257 }
258
259 return 0;
260 }
261
bdev_sync(struct super_block * sb)262 static void bdev_sync(struct super_block *sb)
263 {
264 struct logfs_super *super = logfs_super(sb);
265
266 wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
267 }
268
bdev_find_first_sb(struct super_block * sb,u64 * ofs)269 static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
270 {
271 struct logfs_super *super = logfs_super(sb);
272 struct address_space *mapping = super->s_mapping_inode->i_mapping;
273 filler_t *filler = bdev_readpage;
274
275 *ofs = 0;
276 return read_cache_page(mapping, 0, filler, sb);
277 }
278
bdev_find_last_sb(struct super_block * sb,u64 * ofs)279 static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
280 {
281 struct logfs_super *super = logfs_super(sb);
282 struct address_space *mapping = super->s_mapping_inode->i_mapping;
283 filler_t *filler = bdev_readpage;
284 u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
285 pgoff_t index = pos >> PAGE_SHIFT;
286
287 *ofs = pos;
288 return read_cache_page(mapping, index, filler, sb);
289 }
290
bdev_write_sb(struct super_block * sb,struct page * page)291 static int bdev_write_sb(struct super_block *sb, struct page *page)
292 {
293 struct block_device *bdev = logfs_super(sb)->s_bdev;
294
295 /* Nothing special to do for block devices. */
296 return sync_request(page, bdev, WRITE);
297 }
298
bdev_put_device(struct logfs_super * s)299 static void bdev_put_device(struct logfs_super *s)
300 {
301 blkdev_put(s->s_bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
302 }
303
bdev_can_write_buf(struct super_block * sb,u64 ofs)304 static int bdev_can_write_buf(struct super_block *sb, u64 ofs)
305 {
306 return 0;
307 }
308
309 static const struct logfs_device_ops bd_devops = {
310 .find_first_sb = bdev_find_first_sb,
311 .find_last_sb = bdev_find_last_sb,
312 .write_sb = bdev_write_sb,
313 .readpage = bdev_readpage,
314 .writeseg = bdev_writeseg,
315 .erase = bdev_erase,
316 .can_write_buf = bdev_can_write_buf,
317 .sync = bdev_sync,
318 .put_device = bdev_put_device,
319 };
320
logfs_get_sb_bdev(struct logfs_super * p,struct file_system_type * type,const char * devname)321 int logfs_get_sb_bdev(struct logfs_super *p, struct file_system_type *type,
322 const char *devname)
323 {
324 struct block_device *bdev;
325
326 bdev = blkdev_get_by_path(devname, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
327 type);
328 if (IS_ERR(bdev))
329 return PTR_ERR(bdev);
330
331 if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
332 int mtdnr = MINOR(bdev->bd_dev);
333 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
334 return logfs_get_sb_mtd(p, mtdnr);
335 }
336
337 p->s_bdev = bdev;
338 p->s_mtd = NULL;
339 p->s_devops = &bd_devops;
340 return 0;
341 }
342