mm/page_io.c

   1 /*
   2  *  linux/mm/page_io.c
   3  *
   4  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   5  *
   6  *  Swap reorganised 29.12.95,
   7  *  Asynchronous swapping added 30.12.95. Stephen Tweedie
   8  *  Removed race in async swapping. 14.4.1996. Bruno Haible
   9  *  Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie
  10  *  Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman
  11  */
  12
  13 #include <linux/mm.h>
  14 #include <linux/kernel_stat.h>
  15 #include <linux/gfp.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/swap.h>
  18 #include <linux/bio.h>
  19 #include <linux/swapops.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/writeback.h>
  22 #include <linux/frontswap.h>
  23 #include <asm/pgtable.h>
  24
  25 static struct bio *get_swap_bio(gfp_t gfp_flags,
  26                                 struct page *page, bio_end_io_t end_io)
  27 {
  28         struct bio *bio;
  29
  30         bio = bio_alloc(gfp_flags, 1);
  31         if (bio) {
  32                 bio->bi_sector = map_swap_page(page, &bio->bi_bdev);
  33                 bio->bi_sector <<= PAGE_SHIFT - 9;
  34                 bio->bi_io_vec[0].bv_page = page;
  35                 bio->bi_io_vec[0].bv_len = PAGE_SIZE;
  36                 bio->bi_io_vec[0].bv_offset = 0;
  37                 bio->bi_vcnt = 1;
  38                 bio->bi_idx = 0;
  39                 bio->bi_size = PAGE_SIZE;
  40                 bio->bi_end_io = end_io;
  41         }
  42         return bio;
  43 }
  44
  45 static void end_swap_bio_write(struct bio *bio, int err)
  46 {
  47         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  48         struct page *page = bio->bi_io_vec[0].bv_page;
  49
  50         if (!uptodate) {
  51                 SetPageError(page);
  52                 /*
  53                  * We failed to write the page out to swap-space.
  54                  * Re-dirty the page in order to avoid it being reclaimed.
  55                  * Also print a dire warning that things will go BAD (tm)
  56                  * very quickly.
  57                  *
  58                  * Also clear PG_reclaim to avoid rotate_reclaimable_page()
  59                  */
  60                 set_page_dirty(page);
  61                 printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n",
  62                                 imajor(bio->bi_bdev->bd_inode),
  63                                 iminor(bio->bi_bdev->bd_inode),
  64                                 (unsigned long long)bio->bi_sector);
  65                 ClearPageReclaim(page);
  66         }
  67         end_page_writeback(page);
  68         bio_put(bio);
  69 }
  70
  71 void end_swap_bio_read(struct bio *bio, int err)
  72 {
  73         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  74         struct page *page = bio->bi_io_vec[0].bv_page;
  75
  76         if (!uptodate) {
  77                 SetPageError(page);
  78                 ClearPageUptodate(page);
  79                 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
  80                                 imajor(bio->bi_bdev->bd_inode),
  81                                 iminor(bio->bi_bdev->bd_inode),
  82                                 (unsigned long long)bio->bi_sector);
  83         } else {
  84                 SetPageUptodate(page);
  85         }
  86         unlock_page(page);
  87         bio_put(bio);
  88 }
  89
  90 int generic_swapfile_activate(struct swap_info_struct *sis,
  91                                 struct file *swap_file,
  92                                 sector_t *span)
  93 {
  94         struct address_space *mapping = swap_file->f_mapping;
  95         struct inode *inode = mapping->host;
  96         unsigned blocks_per_page;
  97         unsigned long page_no;
  98         unsigned blkbits;
  99         sector_t probe_block;
 100         sector_t last_block;
 101         sector_t lowest_block = -1;
 102         sector_t highest_block = 0;
 103         int nr_extents = 0;
 104         int ret;
 105
 106         blkbits = inode->i_blkbits;
 107         blocks_per_page = PAGE_SIZE >> blkbits;
 108
 109         /*
 110          * Map all the blocks into the extent list.  This code doesn't try
 111          * to be very smart.
 112          */
 113         probe_block = 0;
 114         page_no = 0;
 115         last_block = i_size_read(inode) >> blkbits;
 116         while ((probe_block + blocks_per_page) <= last_block &&
 117                         page_no < sis->max) {
 118                 unsigned block_in_page;
 119                 sector_t first_block;
 120
 121                 first_block = bmap(inode, probe_block);
 122                 if (first_block == 0)
 123                         goto bad_bmap;
 124
 125                 /*
 126                  * It must be PAGE_SIZE aligned on-disk
 127                  */
 128                 if (first_block & (blocks_per_page - 1)) {
 129                         probe_block++;
 130                         goto reprobe;
 131                 }
 132
 133                 for (block_in_page = 1; block_in_page < blocks_per_page;
 134                                         block_in_page++) {
 135                         sector_t block;
 136
 137                         block = bmap(inode, probe_block + block_in_page);
 138                         if (block == 0)
 139                                 goto bad_bmap;
 140                         if (block != first_block + block_in_page) {
 141                                 /* Discontiguity */
 142                                 probe_block++;
 143                                 goto reprobe;
 144                         }
 145                 }
 146
 147                 first_block >>= (PAGE_SHIFT - blkbits);
 148                 if (page_no) {  /* exclude the header page */
 149                         if (first_block < lowest_block)
 150                                 lowest_block = first_block;
 151                         if (first_block > highest_block)
 152                                 highest_block = first_block;
 153                 }
 154
 155                 /*
 156                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 157                  */
 158                 ret = add_swap_extent(sis, page_no, 1, first_block);
 159                 if (ret < 0)
 160                         goto out;
 161                 nr_extents += ret;
 162                 page_no++;
 163                 probe_block += blocks_per_page;
 164 reprobe:
 165                 continue;
 166         }
 167         ret = nr_extents;
 168         *span = 1 + highest_block - lowest_block;
 169         if (page_no == 0)
 170                 page_no = 1;    /* force Empty message */
 171         sis->max = page_no;
 172         sis->pages = page_no - 1;
 173         sis->highest_bit = page_no - 1;
 174 out:
 175         return ret;
 176 bad_bmap:
 177         printk(KERN_ERR "swapon: swapfile has holes\n");
 178         ret = -EINVAL;
 179         goto out;
 180 }
 181
 182 /*
 183  * We may have stale swap cache pages in memory: notice
 184  * them here and get rid of the unnecessary final write.
 185  */
 186 int swap_writepage(struct page *page, struct writeback_control *wbc)
 187 {
 188         struct bio *bio;
 189         int ret = 0, rw = WRITE;
 190         struct swap_info_struct *sis = page_swap_info(page);
 191
 192         if (try_to_free_swap(page)) {
 193                 unlock_page(page);
 194                 goto out;
 195         }
 196         if (frontswap_store(page) == 0) {
 197                 set_page_writeback(page);
 198                 unlock_page(page);
 199                 end_page_writeback(page);
 200                 goto out;
 201         }
 202
 203         if (sis->flags & SWP_FILE) {
 204                 struct kiocb kiocb;
 205                 struct file *swap_file = sis->swap_file;
 206                 struct address_space *mapping = swap_file->f_mapping;
 207                 struct iovec iov = {
 208                         .iov_base = kmap(page),
 209                         .iov_len  = PAGE_SIZE,
 210                 };
 211
 212                 init_sync_kiocb(&kiocb, swap_file);
 213                 kiocb.ki_pos = page_file_offset(page);
 214                 kiocb.ki_left = PAGE_SIZE;
 215                 kiocb.ki_nbytes = PAGE_SIZE;
 216
 217                 unlock_page(page);
 218                 ret = mapping->a_ops->direct_IO(KERNEL_WRITE,
 219                                                 &kiocb, &iov,
 220                                                 kiocb.ki_pos, 1);
 221                 kunmap(page);
 222                 if (ret == PAGE_SIZE) {
 223                         count_vm_event(PSWPOUT);
 224                         ret = 0;
 225                 }
 226                 return ret;
 227         }
 228
 229         bio = get_swap_bio(GFP_NOIO, page, end_swap_bio_write);
 230         if (bio == NULL) {
 231                 set_page_dirty(page);
 232                 unlock_page(page);
 233                 ret = -ENOMEM;
 234                 goto out;
 235         }
 236         if (wbc->sync_mode == WB_SYNC_ALL)
 237                 rw |= REQ_SYNC;
 238         count_vm_event(PSWPOUT);
 239         set_page_writeback(page);
 240         unlock_page(page);
 241         submit_bio(rw, bio);
 242 out:
 243         return ret;
 244 }
 245
 246 int swap_readpage(struct page *page)
 247 {
 248         struct bio *bio;
 249         int ret = 0;
 250         struct swap_info_struct *sis = page_swap_info(page);
 251
 252         VM_BUG_ON(!PageLocked(page));
 253         VM_BUG_ON(PageUptodate(page));
 254         if (frontswap_load(page) == 0) {
 255                 SetPageUptodate(page);
 256                 unlock_page(page);
 257                 goto out;
 258         }
 259
 260         if (sis->flags & SWP_FILE) {
 261                 struct file *swap_file = sis->swap_file;
 262                 struct address_space *mapping = swap_file->f_mapping;
 263
 264                 ret = mapping->a_ops->readpage(swap_file, page);
 265                 if (!ret)
 266                         count_vm_event(PSWPIN);
 267                 return ret;
 268         }
 269
 270         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 271         if (bio == NULL) {
 272                 unlock_page(page);
 273                 ret = -ENOMEM;
 274                 goto out;
 275         }
 276         count_vm_event(PSWPIN);
 277         submit_bio(READ, bio);
 278 out:
 279         return ret;
 280 }
 281
 282 int swap_set_page_dirty(struct page *page)
 283 {
 284         struct swap_info_struct *sis = page_swap_info(page);
 285
 286         if (sis->flags & SWP_FILE) {
 287                 struct address_space *mapping = sis->swap_file->f_mapping;
 288                 return mapping->a_ops->set_page_dirty(page);
 289         } else {
 290                 return __set_page_dirty_no_writeback(page);
 291         }
 292 }