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_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         wait_for_completion(&complete);
  44         return test_bit(BIO_UPTODATE, &bio.bi_flags) ? 0 : -EIO;
  45 }
  46
  47 static int bdev_readpage(void *_sb, struct page *page)
  48 {
  49         struct super_block *sb = _sb;
  50         struct block_device *bdev = logfs_super(sb)->s_bdev;
  51         int err;
  52
  53         err = sync_request(page, bdev, READ);
  54         if (err) {
  55                 ClearPageUptodate(page);
  56                 SetPageError(page);
  57         } else {
  58                 SetPageUptodate(page);
  59                 ClearPageError(page);
  60         }
  61         unlock_page(page);
  62         return err;
  63 }
  64
  65 static DECLARE_WAIT_QUEUE_HEAD(wq);
  66
  67 static void writeseg_end_io(struct bio *bio, int err)
  68 {
  69         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  70         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
  71         struct super_block *sb = bio->bi_private;
  72         struct logfs_super *super = logfs_super(sb);
  73         struct page *page;
  74
  75         BUG_ON(!uptodate); /* FIXME: Retry io or write elsewhere */
  76         BUG_ON(err);
  77         BUG_ON(bio->bi_vcnt == 0);
  78         do {
  79                 page = bvec->bv_page;
  80                 if (--bvec >= bio->bi_io_vec)
  81                         prefetchw(&bvec->bv_page->flags);
  82
  83                 end_page_writeback(page);
  84                 page_cache_release(page);
  85         } while (bvec >= bio->bi_io_vec);
  86         bio_put(bio);
  87         if (atomic_dec_and_test(&super->s_pending_writes))
  88                 wake_up(&wq);
  89 }
  90
  91 static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
  92                 size_t nr_pages)
  93 {
  94         struct logfs_super *super = logfs_super(sb);
  95         struct address_space *mapping = super->s_mapping_inode->i_mapping;
  96         struct bio *bio;
  97         struct page *page;
  98         unsigned int max_pages;
  99         int i;
 100
 101         max_pages = min(nr_pages, (size_t) bio_get_nr_vecs(super->s_bdev));
 102
 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_size = i * PAGE_SIZE;
 111                         bio->bi_bdev = super->s_bdev;
 112                         bio->bi_sector = ofs >> 9;
 113                         bio->bi_private = sb;
 114                         bio->bi_end_io = writeseg_end_io;
 115                         atomic_inc(&super->s_pending_writes);
 116                         submit_bio(WRITE, bio);
 117
 118                         ofs += i * PAGE_SIZE;
 119                         index += i;
 120                         nr_pages -= i;
 121                         i = 0;
 122
 123                         bio = bio_alloc(GFP_NOFS, max_pages);
 124                         BUG_ON(!bio);
 125                 }
 126                 page = find_lock_page(mapping, index + i);
 127                 BUG_ON(!page);
 128                 bio->bi_io_vec[i].bv_page = page;
 129                 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
 130                 bio->bi_io_vec[i].bv_offset = 0;
 131
 132                 BUG_ON(PageWriteback(page));
 133                 set_page_writeback(page);
 134                 unlock_page(page);
 135         }
 136         bio->bi_vcnt = nr_pages;
 137         bio->bi_size = nr_pages * PAGE_SIZE;
 138         bio->bi_bdev = super->s_bdev;
 139         bio->bi_sector = ofs >> 9;
 140         bio->bi_private = sb;
 141         bio->bi_end_io = writeseg_end_io;
 142         atomic_inc(&super->s_pending_writes);
 143         submit_bio(WRITE, bio);
 144         return 0;
 145 }
 146
 147 static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
 148 {
 149         struct logfs_super *super = logfs_super(sb);
 150         int head;
 151
 152         BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
 153
 154         if (len == 0) {
 155                 /* This can happen when the object fit perfectly into a
 156                  * segment, the segment gets written per sync and subsequently
 157                  * closed.
 158                  */
 159                 return;
 160         }
 161         head = ofs & (PAGE_SIZE - 1);
 162         if (head) {
 163                 ofs -= head;
 164                 len += head;
 165         }
 166         len = PAGE_ALIGN(len);
 167         __bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
 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         unsigned int max_pages;
 191         int i;
 192
 193         max_pages = min(nr_pages, (size_t) bio_get_nr_vecs(super->s_bdev));
 194
 195         bio = bio_alloc(GFP_NOFS, max_pages);
 196         BUG_ON(!bio);
 197
 198         for (i = 0; i < nr_pages; i++) {
 199                 if (i >= max_pages) {
 200                         /* Block layer cannot split bios :( */
 201                         bio->bi_vcnt = i;
 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_size = nr_pages * PAGE_SIZE;
 224         bio->bi_bdev = super->s_bdev;
 225         bio->bi_sector = ofs >> 9;
 226         bio->bi_private = sb;
 227         bio->bi_end_io = erase_end_io;
 228         atomic_inc(&super->s_pending_writes);
 229         submit_bio(WRITE, bio);
 230         return 0;
 231 }
 232
 233 static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
 234                 int ensure_write)
 235 {
 236         struct logfs_super *super = logfs_super(sb);
 237
 238         BUG_ON(to & (PAGE_SIZE - 1));
 239         BUG_ON(len & (PAGE_SIZE - 1));
 240
 241         if (super->s_flags & LOGFS_SB_FLAG_RO)
 242                 return -EROFS;
 243
 244         if (ensure_write) {
 245                 /*
 246                  * Object store doesn't care whether erases happen or not.
 247                  * But for the journal they are required.  Otherwise a scan
 248                  * can find an old commit entry and assume it is the current
 249                  * one, travelling back in time.
 250                  */
 251                 do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
 252         }
 253
 254         return 0;
 255 }
 256
 257 static void bdev_sync(struct super_block *sb)
 258 {
 259         struct logfs_super *super = logfs_super(sb);
 260
 261         wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
 262 }
 263
 264 static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
 265 {
 266         struct logfs_super *super = logfs_super(sb);
 267         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 268         filler_t *filler = bdev_readpage;
 269
 270         *ofs = 0;
 271         return read_cache_page(mapping, 0, filler, sb);
 272 }
 273
 274 static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
 275 {
 276         struct logfs_super *super = logfs_super(sb);
 277         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 278         filler_t *filler = bdev_readpage;
 279         u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
 280         pgoff_t index = pos >> PAGE_SHIFT;
 281
 282         *ofs = pos;
 283         return read_cache_page(mapping, index, filler, sb);
 284 }
 285
 286 static int bdev_write_sb(struct super_block *sb, struct page *page)
 287 {
 288         struct block_device *bdev = logfs_super(sb)->s_bdev;
 289
 290         /* Nothing special to do for block devices. */
 291         return sync_request(page, bdev, WRITE);
 292 }
 293
 294 static void bdev_put_device(struct logfs_super *s)
 295 {
 296         blkdev_put(s->s_bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 297 }
 298
 299 static int bdev_can_write_buf(struct super_block *sb, u64 ofs)
 300 {
 301         return 0;
 302 }
 303
 304 static const struct logfs_device_ops bd_devops = {
 305         .find_first_sb  = bdev_find_first_sb,
 306         .find_last_sb   = bdev_find_last_sb,
 307         .write_sb       = bdev_write_sb,
 308         .readpage       = bdev_readpage,
 309         .writeseg       = bdev_writeseg,
 310         .erase          = bdev_erase,
 311         .can_write_buf  = bdev_can_write_buf,
 312         .sync           = bdev_sync,
 313         .put_device     = bdev_put_device,
 314 };
 315
 316 int logfs_get_sb_bdev(struct logfs_super *p, struct file_system_type *type,
 317                 const char *devname)
 318 {
 319         struct block_device *bdev;
 320
 321         bdev = blkdev_get_by_path(devname, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 322                                   type);
 323         if (IS_ERR(bdev))
 324                 return PTR_ERR(bdev);
 325
 326         if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
 327                 int mtdnr = MINOR(bdev->bd_dev);
 328                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 329                 return logfs_get_sb_mtd(p, mtdnr);
 330         }
 331
 332         p->s_bdev = bdev;
 333         p->s_mtd = NULL;
 334         p->s_devops = &bd_devops;
 335         return 0;
 336 }