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 int sync_request(struct page *page, struct block_device *bdev, int op)
  18 {
  19         struct bio bio;
  20         struct bio_vec bio_vec;
  21
  22         bio_init(&bio, &bio_vec, 1);
  23         bio.bi_bdev = bdev;
  24         bio_add_page(&bio, page, PAGE_SIZE, 0);
  25         bio.bi_iter.bi_sector = page->index * (PAGE_SIZE >> 9);
  26         bio_set_op_attrs(&bio, op, 0);
  27
  28         return submit_bio_wait(&bio);
  29 }
  30
  31 static int bdev_readpage(void *_sb, struct page *page)
  32 {
  33         struct super_block *sb = _sb;
  34         struct block_device *bdev = logfs_super(sb)->s_bdev;
  35         int err;
  36
  37         err = sync_request(page, bdev, READ);
  38         if (err) {
  39                 ClearPageUptodate(page);
  40                 SetPageError(page);
  41         } else {
  42                 SetPageUptodate(page);
  43                 ClearPageError(page);
  44         }
  45         unlock_page(page);
  46         return err;
  47 }
  48
  49 static DECLARE_WAIT_QUEUE_HEAD(wq);
  50
  51 static void writeseg_end_io(struct bio *bio)
  52 {
  53         struct bio_vec *bvec;
  54         int i;
  55         struct super_block *sb = bio->bi_private;
  56         struct logfs_super *super = logfs_super(sb);
  57
  58         BUG_ON(bio->bi_error); /* FIXME: Retry io or write elsewhere */
  59
  60         bio_for_each_segment_all(bvec, bio, i) {
  61                 end_page_writeback(bvec->bv_page);
  62                 put_page(bvec->bv_page);
  63         }
  64         bio_put(bio);
  65         if (atomic_dec_and_test(&super->s_pending_writes))
  66                 wake_up(&wq);
  67 }
  68
  69 static int __bdev_writeseg(struct super_block *sb, u64 ofs, pgoff_t index,
  70                 size_t nr_pages)
  71 {
  72         struct logfs_super *super = logfs_super(sb);
  73         struct address_space *mapping = super->s_mapping_inode->i_mapping;
  74         struct bio *bio = NULL;
  75         struct page *page;
  76         unsigned int max_pages;
  77         int i, ret;
  78
  79         max_pages = min_t(size_t, nr_pages, BIO_MAX_PAGES);
  80
  81         for (i = 0; i < nr_pages; i++) {
  82                 if (!bio) {
  83                         bio = bio_alloc(GFP_NOFS, max_pages);
  84                         BUG_ON(!bio);
  85
  86                         bio->bi_bdev = super->s_bdev;
  87                         bio->bi_iter.bi_sector = ofs >> 9;
  88                         bio->bi_private = sb;
  89                         bio->bi_end_io = writeseg_end_io;
  90                         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
  91                 }
  92                 page = find_lock_page(mapping, index + i);
  93                 BUG_ON(!page);
  94                 ret = bio_add_page(bio, page, PAGE_SIZE, 0);
  95
  96                 BUG_ON(PageWriteback(page));
  97                 set_page_writeback(page);
  98                 unlock_page(page);
  99
 100                 if (!ret) {
 101                         /* Block layer cannot split bios :( */
 102                         ofs += bio->bi_iter.bi_size;
 103                         atomic_inc(&super->s_pending_writes);
 104                         submit_bio(bio);
 105                         bio = NULL;
 106                 }
 107         }
 108
 109         if (bio) {
 110                 atomic_inc(&super->s_pending_writes);
 111                 submit_bio(bio);
 112         }
 113         return 0;
 114 }
 115
 116 static void bdev_writeseg(struct super_block *sb, u64 ofs, size_t len)
 117 {
 118         struct logfs_super *super = logfs_super(sb);
 119         int head;
 120
 121         BUG_ON(super->s_flags & LOGFS_SB_FLAG_RO);
 122
 123         if (len == 0) {
 124                 /* This can happen when the object fit perfectly into a
 125                  * segment, the segment gets written per sync and subsequently
 126                  * closed.
 127                  */
 128                 return;
 129         }
 130         head = ofs & (PAGE_SIZE - 1);
 131         if (head) {
 132                 ofs -= head;
 133                 len += head;
 134         }
 135         len = PAGE_ALIGN(len);
 136         __bdev_writeseg(sb, ofs, ofs >> PAGE_SHIFT, len >> PAGE_SHIFT);
 137 }
 138
 139
 140 static void erase_end_io(struct bio *bio)
 141 {
 142         struct super_block *sb = bio->bi_private;
 143         struct logfs_super *super = logfs_super(sb);
 144
 145         BUG_ON(bio->bi_error); /* FIXME: Retry io or write elsewhere */
 146         BUG_ON(bio->bi_vcnt == 0);
 147         bio_put(bio);
 148         if (atomic_dec_and_test(&super->s_pending_writes))
 149                 wake_up(&wq);
 150 }
 151
 152 static int do_erase(struct super_block *sb, u64 ofs, pgoff_t index,
 153                 size_t nr_pages)
 154 {
 155         struct logfs_super *super = logfs_super(sb);
 156         struct bio *bio;
 157         unsigned int max_pages;
 158         int i;
 159
 160         max_pages = min_t(size_t, nr_pages, BIO_MAX_PAGES);
 161
 162         bio = bio_alloc(GFP_NOFS, max_pages);
 163         BUG_ON(!bio);
 164
 165         for (i = 0; i < nr_pages; i++) {
 166                 if (i >= max_pages) {
 167                         /* Block layer cannot split bios :( */
 168                         bio->bi_vcnt = i;
 169                         bio->bi_iter.bi_size = i * PAGE_SIZE;
 170                         bio->bi_bdev = super->s_bdev;
 171                         bio->bi_iter.bi_sector = ofs >> 9;
 172                         bio->bi_private = sb;
 173                         bio->bi_end_io = erase_end_io;
 174                         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 175                         atomic_inc(&super->s_pending_writes);
 176                         submit_bio(bio);
 177
 178                         ofs += i * PAGE_SIZE;
 179                         index += i;
 180                         nr_pages -= i;
 181                         i = 0;
 182
 183                         bio = bio_alloc(GFP_NOFS, max_pages);
 184                         BUG_ON(!bio);
 185                 }
 186                 bio->bi_io_vec[i].bv_page = super->s_erase_page;
 187                 bio->bi_io_vec[i].bv_len = PAGE_SIZE;
 188                 bio->bi_io_vec[i].bv_offset = 0;
 189         }
 190         bio->bi_vcnt = nr_pages;
 191         bio->bi_iter.bi_size = nr_pages * PAGE_SIZE;
 192         bio->bi_bdev = super->s_bdev;
 193         bio->bi_iter.bi_sector = ofs >> 9;
 194         bio->bi_private = sb;
 195         bio->bi_end_io = erase_end_io;
 196         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 197         atomic_inc(&super->s_pending_writes);
 198         submit_bio(bio);
 199         return 0;
 200 }
 201
 202 static int bdev_erase(struct super_block *sb, loff_t to, size_t len,
 203                 int ensure_write)
 204 {
 205         struct logfs_super *super = logfs_super(sb);
 206
 207         BUG_ON(to & (PAGE_SIZE - 1));
 208         BUG_ON(len & (PAGE_SIZE - 1));
 209
 210         if (super->s_flags & LOGFS_SB_FLAG_RO)
 211                 return -EROFS;
 212
 213         if (ensure_write) {
 214                 /*
 215                  * Object store doesn't care whether erases happen or not.
 216                  * But for the journal they are required.  Otherwise a scan
 217                  * can find an old commit entry and assume it is the current
 218                  * one, travelling back in time.
 219                  */
 220                 do_erase(sb, to, to >> PAGE_SHIFT, len >> PAGE_SHIFT);
 221         }
 222
 223         return 0;
 224 }
 225
 226 static void bdev_sync(struct super_block *sb)
 227 {
 228         struct logfs_super *super = logfs_super(sb);
 229
 230         wait_event(wq, atomic_read(&super->s_pending_writes) == 0);
 231 }
 232
 233 static struct page *bdev_find_first_sb(struct super_block *sb, u64 *ofs)
 234 {
 235         struct logfs_super *super = logfs_super(sb);
 236         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 237         filler_t *filler = bdev_readpage;
 238
 239         *ofs = 0;
 240         return read_cache_page(mapping, 0, filler, sb);
 241 }
 242
 243 static struct page *bdev_find_last_sb(struct super_block *sb, u64 *ofs)
 244 {
 245         struct logfs_super *super = logfs_super(sb);
 246         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 247         filler_t *filler = bdev_readpage;
 248         u64 pos = (super->s_bdev->bd_inode->i_size & ~0xfffULL) - 0x1000;
 249         pgoff_t index = pos >> PAGE_SHIFT;
 250
 251         *ofs = pos;
 252         return read_cache_page(mapping, index, filler, sb);
 253 }
 254
 255 static int bdev_write_sb(struct super_block *sb, struct page *page)
 256 {
 257         struct block_device *bdev = logfs_super(sb)->s_bdev;
 258
 259         /* Nothing special to do for block devices. */
 260         return sync_request(page, bdev, WRITE);
 261 }
 262
 263 static void bdev_put_device(struct logfs_super *s)
 264 {
 265         blkdev_put(s->s_bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 266 }
 267
 268 static int bdev_can_write_buf(struct super_block *sb, u64 ofs)
 269 {
 270         return 0;
 271 }
 272
 273 static const struct logfs_device_ops bd_devops = {
 274         .find_first_sb  = bdev_find_first_sb,
 275         .find_last_sb   = bdev_find_last_sb,
 276         .write_sb       = bdev_write_sb,
 277         .readpage       = bdev_readpage,
 278         .writeseg       = bdev_writeseg,
 279         .erase          = bdev_erase,
 280         .can_write_buf  = bdev_can_write_buf,
 281         .sync           = bdev_sync,
 282         .put_device     = bdev_put_device,
 283 };
 284
 285 int logfs_get_sb_bdev(struct logfs_super *p, struct file_system_type *type,
 286                 const char *devname)
 287 {
 288         struct block_device *bdev;
 289
 290         bdev = blkdev_get_by_path(devname, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 291                                   type);
 292         if (IS_ERR(bdev))
 293                 return PTR_ERR(bdev);
 294
 295         if (MAJOR(bdev->bd_dev) == MTD_BLOCK_MAJOR) {
 296                 int mtdnr = MINOR(bdev->bd_dev);
 297                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 298                 return logfs_get_sb_mtd(p, mtdnr);
 299         }
 300
 301         p->s_bdev = bdev;
 302         p->s_mtd = NULL;
 303         p->s_devops = &bd_devops;
 304         return 0;
 305 }