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