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
17static int sync_request(struct page *page, struct block_device *bdev, int rw)
18{
19	struct bio bio;
20	struct bio_vec bio_vec;
21
22	bio_init(&bio);
23	bio.bi_max_vecs = 1;
24	bio.bi_io_vec = &bio_vec;
25	bio_vec.bv_page = page;
26	bio_vec.bv_len = PAGE_SIZE;
27	bio_vec.bv_offset = 0;
28	bio.bi_vcnt = 1;
29	bio.bi_bdev = bdev;
30	bio.bi_iter.bi_sector = page->index * (PAGE_SIZE >> 9);
31	bio.bi_iter.bi_size = PAGE_SIZE;
32
33	return submit_bio_wait(rw, &bio);
34}
35
36static int bdev_readpage(void *_sb, struct page *page)
37{
38	struct super_block *sb = _sb;
39	struct block_device *bdev = logfs_super(sb)->s_bdev;
40	int err;
41
42	err = sync_request(page, bdev, READ);
43	if (err) {
44		ClearPageUptodate(page);
45		SetPageError(page);
46	} else {
47		SetPageUptodate(page);
48		ClearPageError(page);
49	}
50	unlock_page(page);
51	return err;
52}
53
54static DECLARE_WAIT_QUEUE_HEAD(wq);
55
56static void writeseg_end_io(struct bio *bio)
57{
58	struct bio_vec *bvec;
59	int i;
60	struct super_block *sb = bio->bi_private;
61	struct logfs_super *super = logfs_super(sb);
62
63	BUG_ON(bio->bi_error); /* FIXME: Retry io or write elsewhere */
64
65	bio_for_each_segment_all(bvec, bio, i) {
66		end_page_writeback(bvec->bv_page);
67		page_cache_release(bvec->bv_page);
68	}
69	bio_put(bio);
70	if (atomic_dec_and_test(&super->s_pending_writes))
71		wake_up(&wq);
72}
73
74static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
75		size_t nr_pages)
76{
77	struct logfs_super *super = logfs_super(sb);
78	struct address_space *mapping = super->s_mapping_inode->i_mapping;
79	struct bio *bio;
80	struct page *page;
81	unsigned int max_pages;
82	int i;
83
84	max_pages = min_t(size_t, nr_pages, BIO_MAX_PAGES);
85
86	bio = bio_alloc(GFP_NOFS, max_pages);
87	BUG_ON(!bio);
88
89	for (i = 0; i < nr_pages; i++) {
90		if (i >= max_pages) {
91			/* Block layer cannot split bios :( */
92			bio->bi_vcnt = i;
93			bio->bi_iter.bi_size = i * PAGE_SIZE;
94			bio->bi_bdev = super->s_bdev;
95			bio->bi_iter.bi_sector = ofs >> 9;
96			bio->bi_private = sb;
97			bio->bi_end_io = writeseg_end_io;
98			atomic_inc(&super->s_pending_writes);
99			submit_bio(WRITE, bio);
100
101			ofs += i * PAGE_SIZE;
102			index += i;
103			nr_pages -= i;
104			i = 0;
105
106			bio = bio_alloc(GFP_NOFS, max_pages);
107			BUG_ON(!bio);
108		}
109		page = find_lock_page(mapping, index + i);
110		BUG_ON(!page);
111		bio->bi_io_vec[i].bv_page = page;
112		bio->bi_io_vec[i].bv_len = PAGE_SIZE;
113		bio->bi_io_vec[i].bv_offset = 0;
114
115		BUG_ON(PageWriteback(page));
116		set_page_writeback(page);
117		unlock_page(page);
118	}
119	bio->bi_vcnt = nr_pages;
120	bio->bi_iter.bi_size = nr_pages * PAGE_SIZE;
121	bio->bi_bdev = super->s_bdev;
122	bio->bi_iter.bi_sector = ofs >> 9;
123	bio->bi_private = sb;
124	bio->bi_end_io = writeseg_end_io;
125	atomic_inc(&super->s_pending_writes);
126	submit_bio(WRITE, bio);
127	return 0;
128}
129
130static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
131{
132	struct logfs_super *super = logfs_super(sb);
133	int head;
134
135	BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
136
137	if (len == 0) {
138		/* This can happen when the object fit perfectly into a
139		 * segment, the segment gets written per sync and subsequently
140		 * closed.
141		 */
142		return;
143	}
144	head = ofs & (PAGE_SIZE - 1);
145	if (head) {
146		ofs -= head;
147		len += head;
148	}
149	len = PAGE_ALIGN(len);
150	__bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
151}
152
153
154static void erase_end_io(struct bio *bio)
155{
156	struct super_block *sb = bio->bi_private;
157	struct logfs_super *super = logfs_super(sb);
158
159	BUG_ON(bio->bi_error); /* FIXME: Retry io or write elsewhere */
160	BUG_ON(bio->bi_vcnt == 0);
161	bio_put(bio);
162	if (atomic_dec_and_test(&super->s_pending_writes))
163		wake_up(&wq);
164}
165
166static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
167		size_t nr_pages)
168{
169	struct logfs_super *super = logfs_super(sb);
170	struct bio *bio;
171	unsigned int max_pages;
172	int i;
173
174	max_pages = min_t(size_t, nr_pages, BIO_MAX_PAGES);
175
176	bio = bio_alloc(GFP_NOFS, max_pages);
177	BUG_ON(!bio);
178
179	for (i = 0; i < nr_pages; i++) {
180		if (i >= max_pages) {
181			/* Block layer cannot split bios :( */
182			bio->bi_vcnt = i;
183			bio->bi_iter.bi_size = i * PAGE_SIZE;
184			bio->bi_bdev = super->s_bdev;
185			bio->bi_iter.bi_sector = ofs >> 9;
186			bio->bi_private = sb;
187			bio->bi_end_io = erase_end_io;
188			atomic_inc(&super->s_pending_writes);
189			submit_bio(WRITE, bio);
190
191			ofs += i * PAGE_SIZE;
192			index += i;
193			nr_pages -= i;
194			i = 0;
195
196			bio = bio_alloc(GFP_NOFS, max_pages);
197			BUG_ON(!bio);
198		}
199		bio->bi_io_vec[i].bv_page = super->s_erase_page;
200		bio->bi_io_vec[i].bv_len = PAGE_SIZE;
201		bio->bi_io_vec[i].bv_offset = 0;
202	}
203	bio->bi_vcnt = nr_pages;
204	bio->bi_iter.bi_size = nr_pages * PAGE_SIZE;
205	bio->bi_bdev = super->s_bdev;
206	bio->bi_iter.bi_sector = ofs >> 9;
207	bio->bi_private = sb;
208	bio->bi_end_io = erase_end_io;
209	atomic_inc(&super->s_pending_writes);
210	submit_bio(WRITE, bio);
211	return 0;
212}
213
214static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
215		int ensure_write)
216{
217	struct logfs_super *super = logfs_super(sb);
218
219	BUG_ON(to & (PAGE_SIZE - 1));
220	BUG_ON(len & (PAGE_SIZE - 1));
221
222	if (super->s_flags & LOGFS_SB_FLAG_RO)
223		return -EROFS;
224
225	if (ensure_write) {
226		/*
227		 * Object store doesn't care whether erases happen or not.
228		 * But for the journal they are required.  Otherwise a scan
229		 * can find an old commit entry and assume it is the current
230		 * one, travelling back in time.
231		 */
232		do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
233	}
234
235	return 0;
236}
237
238static void bdev_sync(struct super_block *sb)
239{
240	struct logfs_super *super = logfs_super(sb);
241
242	wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
243}
244
245static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
246{
247	struct logfs_super *super = logfs_super(sb);
248	struct address_space *mapping = super->s_mapping_inode->i_mapping;
249	filler_t *filler = bdev_readpage;
250
251	*ofs = 0;
252	return read_cache_page(mapping, 0, filler, sb);
253}
254
255static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
256{
257	struct logfs_super *super = logfs_super(sb);
258	struct address_space *mapping = super->s_mapping_inode->i_mapping;
259	filler_t *filler = bdev_readpage;
260	u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
261	pgoff_t index = pos >> PAGE_SHIFT;
262
263	*ofs = pos;
264	return read_cache_page(mapping, index, filler, sb);
265}
266
267static int bdev_write_sb(struct super_block *sb, struct page *page)
268{
269	struct block_device *bdev = logfs_super(sb)->s_bdev;
270
271	/* Nothing special to do for block devices. */
272	return sync_request(page, bdev, WRITE);
273}
274
275static void bdev_put_device(struct logfs_super *s)
276{
277	blkdev_put(s->s_bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
278}
279
280static int bdev_can_write_buf(struct super_block *sb, u64 ofs)
281{
282	return 0;
283}
284
285static const struct logfs_device_ops bd_devops = {
286	.find_first_sb	= bdev_find_first_sb,
287	.find_last_sb	= bdev_find_last_sb,
288	.write_sb	= bdev_write_sb,
289	.readpage	= bdev_readpage,
290	.writeseg	= bdev_writeseg,
291	.erase		= bdev_erase,
292	.can_write_buf	= bdev_can_write_buf,
293	.sync		= bdev_sync,
294	.put_device	= bdev_put_device,
295};
296
297int logfs_get_sb_bdev(struct logfs_super *p, struct file_system_type *type,
298		const char *devname)
299{
300	struct block_device *bdev;
301
302	bdev = blkdev_get_by_path(devname, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
303				  type);
304	if (IS_ERR(bdev))
305		return PTR_ERR(bdev);
306
307	if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
308		int mtdnr = MINOR(bdev->bd_dev);
309		blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
310		return logfs_get_sb_mtd(p, mtdnr);
311	}
312
313	p->s_bdev = bdev;
314	p->s_mtd = NULL;
315	p->s_devops = &bd_devops;
316	return 0;
317}
318