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
  13 #define PAGE_OFS(ofs) ((ofs) & (PAGE_SIZE-1))
  14
  15 static void request_complete(struct bio *bio, int err)
  16 {
  17         complete((struct completion *)bio->bi_private);
  18 }
  19
  20 static int sync_request(struct page *page, struct block_device *bdev, int rw)
  21 {
  22         struct bio bio;
  23         struct bio_vec bio_vec;
  24         struct completion complete;
  25
  26         bio_init(&bio);
  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;
  31         bio.bi_vcnt = 1;
  32         bio.bi_idx = 0;
  33         bio.bi_size = PAGE_SIZE;
  34         bio.bi_bdev = bdev;
  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;
  39
  40         submit_bio(rw, &bio);
  41         generic_unplug_device(bdev_get_queue(bdev));
  42         wait_for_completion(&complete);
  43         return test_bit(BIO_UPTODATE, &bio.bi_flags) ? 0 : -EIO;
  44 }
  45
  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
  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         } while (bvec >= bio->bi_io_vec);
  84         bio_put(bio);
  85         if (atomic_dec_and_test(&super->s_pending_writes))
  86                 wake_up(&wq);
  87 }
  88
  89 static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
  90                 size_t nr_pages)
  91 {
  92         struct logfs_super *super = logfs_super(sb);
  93         struct address_space *mapping = super->s_mapping_inode->i_mapping;
  94         struct bio *bio;
  95         struct page *page;
  96         struct request_queue *q = bdev_get_queue(sb->s_bdev);
  97         unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
  98         int i;
  99
 100         bio = bio_alloc(GFP_NOFS, max_pages);
 101         BUG_ON(!bio); /* FIXME: handle this */
 102
 103         for (i = 0; i < nr_pages; i++) {
 104                 if (i >= max_pages) {
 105                         /* Block layer cannot split bios :( */
 106                         bio->bi_vcnt = i;
 107                         bio->bi_idx = 0;
 108                         bio->bi_size = i * PAGE_SIZE;
 109                         bio->bi_bdev = super->s_bdev;
 110                         bio->bi_sector = ofs >> 9;
 111                         bio->bi_private = sb;
 112                         bio->bi_end_io = writeseg_end_io;
 113                         atomic_inc(&super->s_pending_writes);
 114                         submit_bio(WRITE, bio);
 115
 116                         ofs += i * PAGE_SIZE;
 117                         index += i;
 118                         nr_pages -= i;
 119                         i = 0;
 120
 121                         bio = bio_alloc(GFP_NOFS, max_pages);
 122                         BUG_ON(!bio);
 123                 }
 124                 page = find_lock_page(mapping, index + i);
 125                 BUG_ON(!page);
 126                 bio->bi_io_vec[i].bv_page = page;
 127                 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
 128                 bio->bi_io_vec[i].bv_offset = 0;
 129
 130                 BUG_ON(PageWriteback(page));
 131                 set_page_writeback(page);
 132                 unlock_page(page);
 133         }
 134         bio->bi_vcnt = nr_pages;
 135         bio->bi_idx = 0;
 136         bio->bi_size = nr_pages * PAGE_SIZE;
 137         bio->bi_bdev = super->s_bdev;
 138         bio->bi_sector = ofs >> 9;
 139         bio->bi_private = sb;
 140         bio->bi_end_io = writeseg_end_io;
 141         atomic_inc(&super->s_pending_writes);
 142         submit_bio(WRITE, bio);
 143         return 0;
 144 }
 145
 146 static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
 147 {
 148         struct logfs_super *super = logfs_super(sb);
 149         int head;
 150
 151         BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
 152
 153         if (len == 0) {
 154                 /* This can happen when the object fit perfectly into a
 155                  * segment, the segment gets written per sync and subsequently
 156                  * closed.
 157                  */
 158                 return;
 159         }
 160         head = ofs & (PAGE_SIZE - 1);
 161         if (head) {
 162                 ofs -= head;
 163                 len += head;
 164         }
 165         len = PAGE_ALIGN(len);
 166         __bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
 167         generic_unplug_device(bdev_get_queue(logfs_super(sb)->s_bdev));
 168 }
 169
 170
 171 static void erase_end_io(struct bio *bio, int err)
 172 {
 173         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 174         struct super_block *sb = bio->bi_private;
 175         struct logfs_super *super = logfs_super(sb);
 176
 177         BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
 178         BUG_ON(err);
 179         BUG_ON(bio->bi_vcnt == 0);
 180         bio_put(bio);
 181         if (atomic_dec_and_test(&super->s_pending_writes))
 182                 wake_up(&wq);
 183 }
 184
 185 static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
 186                 size_t nr_pages)
 187 {
 188         struct logfs_super *super = logfs_super(sb);
 189         struct bio *bio;
 190         struct request_queue *q = bdev_get_queue(sb->s_bdev);
 191         unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
 192         int i;
 193
 194         bio = bio_alloc(GFP_NOFS, max_pages);
 195         BUG_ON(!bio); /* FIXME: handle this */
 196
 197         for (i = 0; i < nr_pages; i++) {
 198                 if (i >= max_pages) {
 199                         /* Block layer cannot split bios :( */
 200                         bio->bi_vcnt = i;
 201                         bio->bi_idx = 0;
 202                         bio->bi_size = i * PAGE_SIZE;
 203                         bio->bi_bdev = super->s_bdev;
 204                         bio->bi_sector = ofs >> 9;
 205                         bio->bi_private = sb;
 206                         bio->bi_end_io = erase_end_io;
 207                         atomic_inc(&super->s_pending_writes);
 208                         submit_bio(WRITE, bio);
 209
 210                         ofs += i * PAGE_SIZE;
 211                         index += i;
 212                         nr_pages -= i;
 213                         i = 0;
 214
 215                         bio = bio_alloc(GFP_NOFS, max_pages);
 216                         BUG_ON(!bio);
 217                 }
 218                 bio->bi_io_vec[i].bv_page = super->s_erase_page;
 219                 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
 220                 bio->bi_io_vec[i].bv_offset = 0;
 221         }
 222         bio->bi_vcnt = nr_pages;
 223         bio->bi_idx = 0;
 224         bio->bi_size = nr_pages * PAGE_SIZE;
 225         bio->bi_bdev = super->s_bdev;
 226         bio->bi_sector = ofs >> 9;
 227         bio->bi_private = sb;
 228         bio->bi_end_io = erase_end_io;
 229         atomic_inc(&super->s_pending_writes);
 230         submit_bio(WRITE, bio);
 231         return 0;
 232 }
 233
 234 static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
 235                 int ensure_write)
 236 {
 237         struct logfs_super *super = logfs_super(sb);
 238
 239         BUG_ON(to & (PAGE_SIZE - 1));
 240         BUG_ON(len & (PAGE_SIZE - 1));
 241
 242         if (super->s_flags & LOGFS_SB_FLAG_RO)
 243                 return -EROFS;
 244
 245         if (ensure_write) {
 246                 /*
 247                  * Object store doesn't care whether erases happen or not.
 248                  * But for the journal they are required.  Otherwise a scan
 249                  * can find an old commit entry and assume it is the current
 250                  * one, travelling back in time.
 251                  */
 252                 do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
 253         }
 254
 255         return 0;
 256 }
 257
 258 static void bdev_sync(struct super_block *sb)
 259 {
 260         struct logfs_super *super = logfs_super(sb);
 261
 262         wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
 263 }
 264
 265 static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
 266 {
 267         struct logfs_super *super = logfs_super(sb);
 268         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 269         filler_t *filler = bdev_readpage;
 270
 271         *ofs = 0;
 272         return read_cache_page(mapping, 0, filler, sb);
 273 }
 274
 275 static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
 276 {
 277         struct logfs_super *super = logfs_super(sb);
 278         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 279         filler_t *filler = bdev_readpage;
 280         u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
 281         pgoff_t index = pos >> PAGE_SHIFT;
 282
 283         *ofs = pos;
 284         return read_cache_page(mapping, index, filler, sb);
 285 }
 286
 287 static int bdev_write_sb(struct super_block *sb, struct page *page)
 288 {
 289         struct block_device *bdev = logfs_super(sb)->s_bdev;
 290
 291         /* Nothing special to do for block devices. */
 292         return sync_request(page, bdev, WRITE);
 293 }
 294
 295 static void bdev_put_device(struct super_block *sb)
 296 {
 297         close_bdev_exclusive(logfs_super(sb)->s_bdev, FMODE_READ|FMODE_WRITE);
 298 }
 299
 300 static const struct logfs_device_ops bd_devops = {
 301         .find_first_sb  = bdev_find_first_sb,
 302         .find_last_sb   = bdev_find_last_sb,
 303         .write_sb       = bdev_write_sb,
 304         .readpage       = bdev_readpage,
 305         .writeseg       = bdev_writeseg,
 306         .erase          = bdev_erase,
 307         .sync           = bdev_sync,
 308         .put_device     = bdev_put_device,
 309 };
 310
 311 int logfs_get_sb_bdev(struct file_system_type *type, int flags,
 312                 const char *devname, struct vfsmount *mnt)
 313 {
 314         struct block_device *bdev;
 315
 316         bdev = open_bdev_exclusive(devname, FMODE_READ|FMODE_WRITE, type);
 317         if (IS_ERR(bdev))
 318                 return PTR_ERR(bdev);
 319
 320         if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
 321                 int mtdnr = MINOR(bdev->bd_dev);
 322                 close_bdev_exclusive(bdev, FMODE_READ|FMODE_WRITE);
 323                 return logfs_get_sb_mtd(type, flags, mtdnr, mnt);
 324         }
 325
 326         return logfs_get_sb_device(type, flags, NULL, bdev, &bd_devops, mnt);
 327 }