fs/logfs/dev_bdev.c

   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
  17 static void request_complete(struct bio *bio, int err)
  18 {
  19         complete((struct completion *)bio->bi_private);
  20 }
  21
  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_max_vecs = 1;
  30         bio.bi_io_vec = &bio_vec;
  31         bio_vec.bv_page = page;
  32         bio_vec.bv_len = PAGE_SIZE;
  33         bio_vec.bv_offset = 0;
  34         bio.bi_vcnt = 1;
  35         bio.bi_idx = 0;
  36         bio.bi_size = PAGE_SIZE;
  37         bio.bi_bdev = bdev;
  38         bio.bi_sector = page->index * (PAGE_SIZE >> 9);
  39         init_completion(&complete);
  40         bio.bi_private = &complete;
  41         bio.bi_end_io = request_complete;
  42
  43         submit_bio(rw, &bio);
  44         wait_for_completion(&complete);
  45         return test_bit(BIO_UPTODATE, &bio.bi_flags) ? 0 : -EIO;
  46 }
  47
  48 static int bdev_readpage(void *_sb, struct page *page)
  49 {
  50         struct super_block *sb = _sb;
  51         struct block_device *bdev = logfs_super(sb)->s_bdev;
  52         int err;
  53
  54         err = sync_request(page, bdev, READ);
  55         if (err) {
  56                 ClearPageUptodate(page);
  57                 SetPageError(page);
  58         } else {
  59                 SetPageUptodate(page);
  60                 ClearPageError(page);
  61         }
  62         unlock_page(page);
  63         return err;
  64 }
  65
  66 static DECLARE_WAIT_QUEUE_HEAD(wq);
  67
  68 static void writeseg_end_io(struct bio *bio, int err)
  69 {
  70         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  71         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
  72         struct super_block *sb = bio->bi_private;
  73         struct logfs_super *super = logfs_super(sb);
  74         struct page *page;
  75
  76         BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
  77         BUG_ON(err);
  78         BUG_ON(bio->bi_vcnt == 0);
  79         do {
  80                 page = bvec->bv_page;
  81                 if (--bvec >= bio->bi_io_vec)
  82                         prefetchw(&bvec->bv_page->flags);
  83
  84                 end_page_writeback(page);
  85                 page_cache_release(page);
  86         } while (bvec >= bio->bi_io_vec);
  87         bio_put(bio);
  88         if (atomic_dec_and_test(&super->s_pending_writes))
  89                 wake_up(&wq);
  90 }
  91
  92 static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
  93                 size_t nr_pages)
  94 {
  95         struct logfs_super *super = logfs_super(sb);
  96         struct address_space *mapping = super->s_mapping_inode->i_mapping;
  97         struct bio *bio;
  98         struct page *page;
  99         unsigned int max_pages;
 100         int i;
 101
 102         max_pages = min(nr_pages, (size_t) bio_get_nr_vecs(super->s_bdev));
 103
 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
 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
 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
 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         unsigned int max_pages;
 194         int i;
 195
 196         max_pages = min(nr_pages, (size_t) bio_get_nr_vecs(super->s_bdev));
 197
 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
 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
 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
 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
 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
 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
 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
 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
 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 }