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 #include <linux/prefetch.h>
14
15 #define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
16
request_complete(struct bio * bio,int err)17 static void request_complete(struct bio *bio, int err)
18 {
19 complete((struct completion *)bio->bi_private);
20 }
21
sync_request(struct page * page,struct block_device * bdev,int rw)22 static int sync_request(struct page *page, struct block_device *bdev, int rw)
23 {
24 struct bio bio;
25 struct bio_vec bio_vec;
26 struct completion complete;
27
28 bio_init(&bio);
29 bio.bi_io_vec = &bio_vec;
30 bio_vec.bv_page = page;
31 bio_vec.bv_len = PAGE_SIZE;
32 bio_vec.bv_offset = 0;
33 bio.bi_vcnt = 1;
34 bio.bi_idx = 0;
35 bio.bi_size = PAGE_SIZE;
36 bio.bi_bdev = bdev;
37 bio.bi_sector = page->index * (PAGE_SIZE >> 9);
38 init_completion(&complete);
39 bio.bi_private = &complete;
40 bio.bi_end_io = request_complete;
41
42 submit_bio(rw, &bio);
43 wait_for_completion(&complete);
44 return test_bit(BIO_UPTODATE, &bio.bi_flags) ? 0 : -EIO;
45 }
46
bdev_readpage(void * _sb,struct page * page)47 static int bdev_readpage(void *_sb, struct page *page)
48 {
49 struct super_block *sb = _sb;
50 struct block_device *bdev = logfs_super(sb)->s_bdev;
51 int err;
52
53 err = sync_request(page, bdev, READ);
54 if (err) {
55 ClearPageUptodate(page);
56 SetPageError(page);
57 } else {
58 SetPageUptodate(page);
59 ClearPageError(page);
60 }
61 unlock_page(page);
62 return err;
63 }
64
65 static DECLARE_WAIT_QUEUE_HEAD(wq);
66
writeseg_end_io(struct bio * bio,int err)67 static void writeseg_end_io(struct bio *bio, int err)
68 {
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);
73 struct page *page;
74
75 BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
76 BUG_ON(err);
77 BUG_ON(bio->bi_vcnt == 0);
78 do {
79 page = bvec->bv_page;
80 if (--bvec >= bio->bi_io_vec)
81 prefetchw(&bvec->bv_page->flags);
82
83 end_page_writeback(page);
84 page_cache_release(page);
85 } while (bvec >= bio->bi_io_vec);
86 bio_put(bio);
87 if (atomic_dec_and_test(&super->s_pending_writes))
88 wake_up(&wq);
89 }
90
__bdev_writeseg(struct super_block * sb,u64 ofs,pgoff_t index,size_t nr_pages)91 static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
92 size_t nr_pages)
93 {
94 struct logfs_super *super = logfs_super(sb);
95 struct address_space *mapping = super->s_mapping_inode->i_mapping;
96 struct bio *bio;
97 struct page *page;
98 struct request_queue *q = bdev_get_queue(sb->s_bdev);
99 unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
100 int i;
101
102 if (max_pages > BIO_MAX_PAGES)
103 max_pages = BIO_MAX_PAGES;
104 bio = bio_alloc(GFP_NOFS, max_pages);
105 BUG_ON(!bio);
106
107 for (i = 0; i < nr_pages; i++) {
108 if (i >= max_pages) {
109 /* Block layer cannot split bios :( */
110 bio->bi_vcnt = i;
111 bio->bi_idx = 0;
112 bio->bi_size = i * PAGE_SIZE;
113 bio->bi_bdev = super->s_bdev;
114 bio->bi_sector = ofs >> 9;
115 bio->bi_private = sb;
116 bio->bi_end_io = writeseg_end_io;
117 atomic_inc(&super->s_pending_writes);
118 submit_bio(WRITE, bio);
119
120 ofs += i * PAGE_SIZE;
121 index += i;
122 nr_pages -= i;
123 i = 0;
124
125 bio = bio_alloc(GFP_NOFS, max_pages);
126 BUG_ON(!bio);
127 }
128 page = find_lock_page(mapping, index + i);
129 BUG_ON(!page);
130 bio->bi_io_vec[i].bv_page = page;
131 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
132 bio->bi_io_vec[i].bv_offset = 0;
133
134 BUG_ON(PageWriteback(page));
135 set_page_writeback(page);
136 unlock_page(page);
137 }
138 bio->bi_vcnt = nr_pages;
139 bio->bi_idx = 0;
140 bio->bi_size = nr_pages * PAGE_SIZE;
141 bio->bi_bdev = super->s_bdev;
142 bio->bi_sector = ofs >> 9;
143 bio->bi_private = sb;
144 bio->bi_end_io = writeseg_end_io;
145 atomic_inc(&super->s_pending_writes);
146 submit_bio(WRITE, bio);
147 return 0;
148 }
149
bdev_writeseg(struct super_block * sb,u64 ofs,size_t len)150 static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
151 {
152 struct logfs_super *super = logfs_super(sb);
153 int head;
154
155 BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
156
157 if (len == 0) {
158 /* This can happen when the object fit perfectly into a
159 * segment, the segment gets written per sync and subsequently
160 * closed.
161 */
162 return;
163 }
164 head = ofs & (PAGE_SIZE - 1);
165 if (head) {
166 ofs -= head;
167 len += head;
168 }
169 len = PAGE_ALIGN(len);
170 __bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
171 }
172
173
erase_end_io(struct bio * bio,int err)174 static void erase_end_io(struct bio *bio, int err)
175 {
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);
179
180 BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
181 BUG_ON(err);
182 BUG_ON(bio->bi_vcnt == 0);
183 bio_put(bio);
184 if (atomic_dec_and_test(&super->s_pending_writes))
185 wake_up(&wq);
186 }
187
do_erase(struct super_block * sb,u64 ofs,pgoff_t index,size_t nr_pages)188 static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
189 size_t nr_pages)
190 {
191 struct logfs_super *super = logfs_super(sb);
192 struct bio *bio;
193 struct request_queue *q = bdev_get_queue(sb->s_bdev);
194 unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
195 int i;
196
197 if (max_pages > BIO_MAX_PAGES)
198 max_pages = BIO_MAX_PAGES;
199 bio = bio_alloc(GFP_NOFS, max_pages);
200 BUG_ON(!bio);
201
202 for (i = 0; i < nr_pages; i++) {
203 if (i >= max_pages) {
204 /* Block layer cannot split bios :( */
205 bio->bi_vcnt = i;
206 bio->bi_idx = 0;
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);
214
215 ofs += i * PAGE_SIZE;
216 index += i;
217 nr_pages -= i;
218 i = 0;
219
220 bio = bio_alloc(GFP_NOFS, max_pages);
221 BUG_ON(!bio);
222 }
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;
226 }
227 bio->bi_vcnt = nr_pages;
228 bio->bi_idx = 0;
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);
236 return 0;
237 }
238
bdev_erase(struct super_block * sb,loff_t to,size_t len,int ensure_write)239 static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
240 int ensure_write)
241 {
242 struct logfs_super *super = logfs_super(sb);
243
244 BUG_ON(to & (PAGE_SIZE - 1));
245 BUG_ON(len & (PAGE_SIZE - 1));
246
247 if (super->s_flags & LOGFS_SB_FLAG_RO)
248 return -EROFS;
249
250 if (ensure_write) {
251 /*
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.
256 */
257 do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
258 }
259
260 return 0;
261 }
262
bdev_sync(struct super_block * sb)263 static void bdev_sync(struct super_block *sb)
264 {
265 struct logfs_super *super = logfs_super(sb);
266
267 wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
268 }
269
bdev_find_first_sb(struct super_block * sb,u64 * ofs)270 static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
271 {
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;
275
276 *ofs = 0;
277 return read_cache_page(mapping, 0, filler, sb);
278 }
279
bdev_find_last_sb(struct super_block * sb,u64 * ofs)280 static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
281 {
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;
287
288 *ofs = pos;
289 return read_cache_page(mapping, index, filler, sb);
290 }
291
bdev_write_sb(struct super_block * sb,struct page * page)292 static int bdev_write_sb(struct super_block *sb, struct page *page)
293 {
294 struct block_device *bdev = logfs_super(sb)->s_bdev;
295
296 /* Nothing special to do for block devices. */
297 return sync_request(page, bdev, WRITE);
298 }
299
bdev_put_device(struct logfs_super * s)300 static void bdev_put_device(struct logfs_super *s)
301 {
302 blkdev_put(s->s_bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
303 }
304
bdev_can_write_buf(struct super_block * sb,u64 ofs)305 static int bdev_can_write_buf(struct super_block *sb, u64 ofs)
306 {
307 return 0;
308 }
309
310 static const struct logfs_device_ops bd_devops = {
311 .find_first_sb = bdev_find_first_sb,
312 .find_last_sb = bdev_find_last_sb,
313 .write_sb = bdev_write_sb,
314 .readpage = bdev_readpage,
315 .writeseg = bdev_writeseg,
316 .erase = bdev_erase,
317 .can_write_buf = bdev_can_write_buf,
318 .sync = bdev_sync,
319 .put_device = bdev_put_device,
320 };
321
logfs_get_sb_bdev(struct logfs_super * p,struct file_system_type * type,const char * devname)322 int logfs_get_sb_bdev(struct logfs_super *p, struct file_system_type *type,
323 const char *devname)
324 {
325 struct block_device *bdev;
326
327 bdev = blkdev_get_by_path(devname, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
328 type);
329 if (IS_ERR(bdev))
330 return PTR_ERR(bdev);
331
332 if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
333 int mtdnr = MINOR(bdev->bd_dev);
334 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
335 return logfs_get_sb_mtd(p, mtdnr);
336 }
337
338 p->s_bdev = bdev;
339 p->s_mtd = NULL;
340 p->s_devops = &bd_devops;
341 return 0;
342 }
343