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/slab.h>
  13 #include <linux/gfp.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_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         generic_unplug_device(bdev_get_queue(bdev));
  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         struct request_queue *q = bdev_get_queue(sb->s_bdev);
 100         unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
 101         int i;
 102
 103         if (max_pages > BIO_MAX_PAGES)
 104                 max_pages = BIO_MAX_PAGES;
 105         bio = bio_alloc(GFP_NOFS, max_pages);
 106         BUG_ON(!bio);
 107
 108         for (i = 0; i < nr_pages; i++) {
 109                 if (i >= max_pages) {
 110                         /* Block layer cannot split bios :( */
 111                         bio->bi_vcnt = i;
 112                         bio->bi_idx = 0;
 113                         bio->bi_size = i * PAGE_SIZE;
 114                         bio->bi_bdev = super->s_bdev;
 115                         bio->bi_sector = ofs >> 9;
 116                         bio->bi_private = sb;
 117                         bio->bi_end_io = writeseg_end_io;
 118                         atomic_inc(&super->s_pending_writes);
 119                         submit_bio(WRITE, bio);
 120
 121                         ofs += i * PAGE_SIZE;
 122                         index += i;
 123                         nr_pages -= i;
 124                         i = 0;
 125
 126                         bio = bio_alloc(GFP_NOFS, max_pages);
 127                         BUG_ON(!bio);
 128                 }
 129                 page = find_lock_page(mapping, index + i);
 130                 BUG_ON(!page);
 131                 bio->bi_io_vec[i].bv_page = page;
 132                 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
 133                 bio->bi_io_vec[i].bv_offset = 0;
 134
 135                 BUG_ON(PageWriteback(page));
 136                 set_page_writeback(page);
 137                 unlock_page(page);
 138         }
 139         bio->bi_vcnt = nr_pages;
 140         bio->bi_idx = 0;
 141         bio->bi_size = nr_pages * PAGE_SIZE;
 142         bio->bi_bdev = super->s_bdev;
 143         bio->bi_sector = ofs >> 9;
 144         bio->bi_private = sb;
 145         bio->bi_end_io = writeseg_end_io;
 146         atomic_inc(&super->s_pending_writes);
 147         submit_bio(WRITE, bio);
 148         return 0;
 149 }
 150
 151 static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
 152 {
 153         struct logfs_super *super = logfs_super(sb);
 154         int head;
 155
 156         BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
 157
 158         if (len == 0) {
 159                 /* This can happen when the object fit perfectly into a
 160                  * segment, the segment gets written per sync and subsequently
 161                  * closed.
 162                  */
 163                 return;
 164         }
 165         head = ofs & (PAGE_SIZE - 1);
 166         if (head) {
 167                 ofs -= head;
 168                 len += head;
 169         }
 170         len = PAGE_ALIGN(len);
 171         __bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
 172         generic_unplug_device(bdev_get_queue(logfs_super(sb)->s_bdev));
 173 }
 174
 175
 176 static void erase_end_io(struct bio *bio, int err)
 177 {
 178         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 179         struct super_block *sb = bio->bi_private;
 180         struct logfs_super *super = logfs_super(sb);
 181
 182         BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
 183         BUG_ON(err);
 184         BUG_ON(bio->bi_vcnt == 0);
 185         bio_put(bio);
 186         if (atomic_dec_and_test(&super->s_pending_writes))
 187                 wake_up(&wq);
 188 }
 189
 190 static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
 191                 size_t nr_pages)
 192 {
 193         struct logfs_super *super = logfs_super(sb);
 194         struct bio *bio;
 195         struct request_queue *q = bdev_get_queue(sb->s_bdev);
 196         unsigned int max_pages = queue_max_hw_sectors(q) >> (PAGE_SHIFT - 9);
 197         int i;
 198
 199         if (max_pages > BIO_MAX_PAGES)
 200                 max_pages = BIO_MAX_PAGES;
 201         bio = bio_alloc(GFP_NOFS, max_pages);
 202         BUG_ON(!bio);
 203
 204         for (i = 0; i < nr_pages; i++) {
 205                 if (i >= max_pages) {
 206                         /* Block layer cannot split bios :( */
 207                         bio->bi_vcnt = i;
 208                         bio->bi_idx = 0;
 209                         bio->bi_size = i * PAGE_SIZE;
 210                         bio->bi_bdev = super->s_bdev;
 211                         bio->bi_sector = ofs >> 9;
 212                         bio->bi_private = sb;
 213                         bio->bi_end_io = erase_end_io;
 214                         atomic_inc(&super->s_pending_writes);
 215                         submit_bio(WRITE, bio);
 216
 217                         ofs += i * PAGE_SIZE;
 218                         index += i;
 219                         nr_pages -= i;
 220                         i = 0;
 221
 222                         bio = bio_alloc(GFP_NOFS, max_pages);
 223                         BUG_ON(!bio);
 224                 }
 225                 bio->bi_io_vec[i].bv_page = super->s_erase_page;
 226                 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
 227                 bio->bi_io_vec[i].bv_offset = 0;
 228         }
 229         bio->bi_vcnt = nr_pages;
 230         bio->bi_idx = 0;
 231         bio->bi_size = nr_pages * PAGE_SIZE;
 232         bio->bi_bdev = super->s_bdev;
 233         bio->bi_sector = ofs >> 9;
 234         bio->bi_private = sb;
 235         bio->bi_end_io = erase_end_io;
 236         atomic_inc(&super->s_pending_writes);
 237         submit_bio(WRITE, bio);
 238         return 0;
 239 }
 240
 241 static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
 242                 int ensure_write)
 243 {
 244         struct logfs_super *super = logfs_super(sb);
 245
 246         BUG_ON(to & (PAGE_SIZE - 1));
 247         BUG_ON(len & (PAGE_SIZE - 1));
 248
 249         if (super->s_flags & LOGFS_SB_FLAG_RO)
 250                 return -EROFS;
 251
 252         if (ensure_write) {
 253                 /*
 254                  * Object store doesn't care whether erases happen or not.
 255                  * But for the journal they are required.  Otherwise a scan
 256                  * can find an old commit entry and assume it is the current
 257                  * one, travelling back in time.
 258                  */
 259                 do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
 260         }
 261
 262         return 0;
 263 }
 264
 265 static void bdev_sync(struct super_block *sb)
 266 {
 267         struct logfs_super *super = logfs_super(sb);
 268
 269         wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
 270 }
 271
 272 static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
 273 {
 274         struct logfs_super *super = logfs_super(sb);
 275         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 276         filler_t *filler = bdev_readpage;
 277
 278         *ofs = 0;
 279         return read_cache_page(mapping, 0, filler, sb);
 280 }
 281
 282 static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
 283 {
 284         struct logfs_super *super = logfs_super(sb);
 285         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 286         filler_t *filler = bdev_readpage;
 287         u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
 288         pgoff_t index = pos >> PAGE_SHIFT;
 289
 290         *ofs = pos;
 291         return read_cache_page(mapping, index, filler, sb);
 292 }
 293
 294 static int bdev_write_sb(struct super_block *sb, struct page *page)
 295 {
 296         struct block_device *bdev = logfs_super(sb)->s_bdev;
 297
 298         /* Nothing special to do for block devices. */
 299         return sync_request(page, bdev, WRITE);
 300 }
 301
 302 static void bdev_put_device(struct super_block *sb)
 303 {
 304         close_bdev_exclusive(logfs_super(sb)->s_bdev, FMODE_READ|FMODE_WRITE);
 305 }
 306
 307 static int bdev_can_write_buf(struct super_block *sb, u64 ofs)
 308 {
 309         return 0;
 310 }
 311
 312 static const struct logfs_device_ops bd_devops = {
 313         .find_first_sb  = bdev_find_first_sb,
 314         .find_last_sb   = bdev_find_last_sb,
 315         .write_sb       = bdev_write_sb,
 316         .readpage       = bdev_readpage,
 317         .writeseg       = bdev_writeseg,
 318         .erase          = bdev_erase,
 319         .can_write_buf  = bdev_can_write_buf,
 320         .sync           = bdev_sync,
 321         .put_device     = bdev_put_device,
 322 };
 323
 324 int logfs_get_sb_bdev(struct logfs_super *p,
 325                 struct file_system_type *type, int flags,
 326                 const char *devname, struct vfsmount *mnt)
 327 {
 328         struct block_device *bdev;
 329
 330         bdev = open_bdev_exclusive(devname, FMODE_READ|FMODE_WRITE, type);
 331         if (IS_ERR(bdev)) {
 332                 kfree(p);
 333                 return PTR_ERR(bdev);
 334         }
 335
 336         if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
 337                 int mtdnr = MINOR(bdev->bd_dev);
 338                 close_bdev_exclusive(bdev, FMODE_READ|FMODE_WRITE);
 339                 return logfs_get_sb_mtd(p, type, flags, mtdnr, mnt);
 340         }
 341
 342         p->s_bdev = bdev;
 343         p->s_mtd = NULL;
 344         p->s_devops = &bd_devops;
 345
 346         return logfs_get_sb_device(p, type, flags, mnt);
 347 }