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                 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
  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
 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         generic_unplug_device(bdev_get_queue(logfs_super(sb)->s_bdev));
 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         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
 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
 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
 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
 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
 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
 300 static void bdev_put_device(struct super_block *sb)
 301 {
 302         close_bdev_exclusive(logfs_super(sb)->s_bdev, FMODE_READ|FMODE_WRITE);
 303 }
 304
 305 static const struct logfs_device_ops bd_devops = {
 306         .find_first_sb  = bdev_find_first_sb,
 307         .find_last_sb   = bdev_find_last_sb,
 308         .write_sb       = bdev_write_sb,
 309         .readpage       = bdev_readpage,
 310         .writeseg       = bdev_writeseg,
 311         .erase          = bdev_erase,
 312         .sync           = bdev_sync,
 313         .put_device     = bdev_put_device,
 314 };
 315
 316 int logfs_get_sb_bdev(struct file_system_type *type, int flags,
 317                 const char *devname, struct vfsmount *mnt)
 318 {
 319         struct block_device *bdev;
 320
 321         bdev = open_bdev_exclusive(devname, FMODE_READ|FMODE_WRITE, type);
 322         if (IS_ERR(bdev))
 323                 return PTR_ERR(bdev);
 324
 325         if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
 326                 int mtdnr = MINOR(bdev->bd_dev);
 327                 close_bdev_exclusive(bdev, FMODE_READ|FMODE_WRITE);
 328                 return logfs_get_sb_mtd(type, flags, mtdnr, mnt);
 329         }
 330
 331         return logfs_get_sb_device(type, flags, NULL, bdev, &bd_devops, mnt);
 332 }