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 <linux/aio.h>
  24 #include <asm/pgtable.h>
  25
  26 static struct bio *get_swap_bio(gfp_t gfp_flags,
  27                                 struct page *page, bio_end_io_t end_io)
  28 {
  29         struct bio *bio;
  30
  31         bio = bio_alloc(gfp_flags, 1);
  32         if (bio) {
  33                 bio->bi_sector = map_swap_page(page, &bio->bi_bdev);
  34                 bio->bi_sector <<= PAGE_SHIFT - 9;
  35                 bio->bi_io_vec[0].bv_page = page;
  36                 bio->bi_io_vec[0].bv_len = PAGE_SIZE;
  37                 bio->bi_io_vec[0].bv_offset = 0;
  38                 bio->bi_vcnt = 1;
  39                 bio->bi_idx = 0;
  40                 bio->bi_size = PAGE_SIZE;
  41                 bio->bi_end_io = end_io;
  42         }
  43         return bio;
  44 }
  45
  46 void end_swap_bio_write(struct bio *bio, int err)
  47 {
  48         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  49         struct page *page = bio->bi_io_vec[0].bv_page;
  50
  51         if (!uptodate) {
  52                 SetPageError(page);
  53                 /*
  54                  * We failed to write the page out to swap-space.
  55                  * Re-dirty the page in order to avoid it being reclaimed.
  56                  * Also print a dire warning that things will go BAD (tm)
  57                  * very quickly.
  58                  *
  59                  * Also clear PG_reclaim to avoid rotate_reclaimable_page()
  60                  */
  61                 set_page_dirty(page);
  62                 printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n",
  63                                 imajor(bio->bi_bdev->bd_inode),
  64                                 iminor(bio->bi_bdev->bd_inode),
  65                                 (unsigned long long)bio->bi_sector);
  66                 ClearPageReclaim(page);
  67         }
  68         end_page_writeback(page);
  69         bio_put(bio);
  70 }
  71
  72 void end_swap_bio_read(struct bio *bio, int err)
  73 {
  74         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  75         struct page *page = bio->bi_io_vec[0].bv_page;
  76
  77         if (!uptodate) {
  78                 SetPageError(page);
  79                 ClearPageUptodate(page);
  80                 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
  81                                 imajor(bio->bi_bdev->bd_inode),
  82                                 iminor(bio->bi_bdev->bd_inode),
  83                                 (unsigned long long)bio->bi_sector);
  84         } else {
  85                 SetPageUptodate(page);
  86         }
  87         unlock_page(page);
  88         bio_put(bio);
  89 }
  90
  91 int generic_swapfile_activate(struct swap_info_struct *sis,
  92                                 struct file *swap_file,
  93                                 sector_t *span)
  94 {
  95         struct address_space *mapping = swap_file->f_mapping;
  96         struct inode *inode = mapping->host;
  97         unsigned blocks_per_page;
  98         unsigned long page_no;
  99         unsigned blkbits;
 100         sector_t probe_block;
 101         sector_t last_block;
 102         sector_t lowest_block = -1;
 103         sector_t highest_block = 0;
 104         int nr_extents = 0;
 105         int ret;
 106
 107         blkbits = inode->i_blkbits;
 108         blocks_per_page = PAGE_SIZE >> blkbits;
 109
 110         /*
 111          * Map all the blocks into the extent list.  This code doesn't try
 112          * to be very smart.
 113          */
 114         probe_block = 0;
 115         page_no = 0;
 116         last_block = i_size_read(inode) >> blkbits;
 117         while ((probe_block + blocks_per_page) <= last_block &&
 118                         page_no < sis->max) {
 119                 unsigned block_in_page;
 120                 sector_t first_block;
 121
 122                 first_block = bmap(inode, probe_block);
 123                 if (first_block == 0)
 124                         goto bad_bmap;
 125
 126                 /*
 127                  * It must be PAGE_SIZE aligned on-disk
 128                  */
 129                 if (first_block & (blocks_per_page - 1)) {
 130                         probe_block++;
 131                         goto reprobe;
 132                 }
 133
 134                 for (block_in_page = 1; block_in_page < blocks_per_page;
 135                                         block_in_page++) {
 136                         sector_t block;
 137
 138                         block = bmap(inode, probe_block + block_in_page);
 139                         if (block == 0)
 140                                 goto bad_bmap;
 141                         if (block != first_block + block_in_page) {
 142                                 /* Discontiguity */
 143                                 probe_block++;
 144                                 goto reprobe;
 145                         }
 146                 }
 147
 148                 first_block >>= (PAGE_SHIFT - blkbits);
 149                 if (page_no) {  /* exclude the header page */
 150                         if (first_block < lowest_block)
 151                                 lowest_block = first_block;
 152                         if (first_block > highest_block)
 153                                 highest_block = first_block;
 154                 }
 155
 156                 /*
 157                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 158                  */
 159                 ret = add_swap_extent(sis, page_no, 1, first_block);
 160                 if (ret < 0)
 161                         goto out;
 162                 nr_extents += ret;
 163                 page_no++;
 164                 probe_block += blocks_per_page;
 165 reprobe:
 166                 continue;
 167         }
 168         ret = nr_extents;
 169         *span = 1 + highest_block - lowest_block;
 170         if (page_no == 0)
 171                 page_no = 1;    /* force Empty message */
 172         sis->max = page_no;
 173         sis->pages = page_no - 1;
 174         sis->highest_bit = page_no - 1;
 175 out:
 176         return ret;
 177 bad_bmap:
 178         printk(KERN_ERR "swapon: swapfile has holes\n");
 179         ret = -EINVAL;
 180         goto out;
 181 }
 182
 183 /*
 184  * We may have stale swap cache pages in memory: notice
 185  * them here and get rid of the unnecessary final write.
 186  */
 187 int swap_writepage(struct page *page, struct writeback_control *wbc)
 188 {
 189         int ret = 0;
 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         ret = __swap_writepage(page, wbc, end_swap_bio_write);
 202 out:
 203         return ret;
 204 }
 205
 206 int __swap_writepage(struct page *page, struct writeback_control *wbc,
 207         void (*end_write_func)(struct bio *, int))
 208 {
 209         struct bio *bio;
 210         int ret = 0, rw = WRITE;
 211         struct swap_info_struct *sis = page_swap_info(page);
 212
 213         if (sis->flags & SWP_FILE) {
 214                 struct kiocb kiocb;
 215                 struct file *swap_file = sis->swap_file;
 216                 struct address_space *mapping = swap_file->f_mapping;
 217                 struct iovec iov = {
 218                         .iov_base = kmap(page),
 219                         .iov_len  = PAGE_SIZE,
 220                 };
 221
 222                 init_sync_kiocb(&kiocb, swap_file);
 223                 kiocb.ki_pos = page_file_offset(page);
 224                 kiocb.ki_left = PAGE_SIZE;
 225                 kiocb.ki_nbytes = PAGE_SIZE;
 226
 227                 set_page_writeback(page);
 228                 unlock_page(page);
 229                 ret = mapping->a_ops->direct_IO(KERNEL_WRITE,
 230                                                 &kiocb, &iov,
 231                                                 kiocb.ki_pos, 1);
 232                 kunmap(page);
 233                 if (ret == PAGE_SIZE) {
 234                         count_vm_event(PSWPOUT);
 235                         ret = 0;
 236                 } else {
 237                         /*
 238                          * In the case of swap-over-nfs, this can be a
 239                          * temporary failure if the system has limited
 240                          * memory for allocating transmit buffers.
 241                          * Mark the page dirty and avoid
 242                          * rotate_reclaimable_page but rate-limit the
 243                          * messages but do not flag PageError like
 244                          * the normal direct-to-bio case as it could
 245                          * be temporary.
 246                          */
 247                         set_page_dirty(page);
 248                         ClearPageReclaim(page);
 249                         pr_err_ratelimited("Write error on dio swapfile (%Lu)\n",
 250                                 page_file_offset(page));
 251                 }
 252                 end_page_writeback(page);
 253                 return ret;
 254         }
 255
 256         bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 257         if (bio == NULL) {
 258                 set_page_dirty(page);
 259                 unlock_page(page);
 260                 ret = -ENOMEM;
 261                 goto out;
 262         }
 263         if (wbc->sync_mode == WB_SYNC_ALL)
 264                 rw |= REQ_SYNC;
 265         count_vm_event(PSWPOUT);
 266         set_page_writeback(page);
 267         unlock_page(page);
 268         submit_bio(rw, bio);
 269 out:
 270         return ret;
 271 }
 272
 273 int swap_readpage(struct page *page)
 274 {
 275         struct bio *bio;
 276         int ret = 0;
 277         struct swap_info_struct *sis = page_swap_info(page);
 278
 279         VM_BUG_ON(!PageLocked(page));
 280         VM_BUG_ON(PageUptodate(page));
 281         if (frontswap_load(page) == 0) {
 282                 SetPageUptodate(page);
 283                 unlock_page(page);
 284                 goto out;
 285         }
 286
 287         if (sis->flags & SWP_FILE) {
 288                 struct file *swap_file = sis->swap_file;
 289                 struct address_space *mapping = swap_file->f_mapping;
 290
 291                 ret = mapping->a_ops->readpage(swap_file, page);
 292                 if (!ret)
 293                         count_vm_event(PSWPIN);
 294                 return ret;
 295         }
 296
 297         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 298         if (bio == NULL) {
 299                 unlock_page(page);
 300                 ret = -ENOMEM;
 301                 goto out;
 302         }
 303         count_vm_event(PSWPIN);
 304         submit_bio(READ, bio);
 305 out:
 306         return ret;
 307 }
 308
 309 int swap_set_page_dirty(struct page *page)
 310 {
 311         struct swap_info_struct *sis = page_swap_info(page);
 312
 313         if (sis->flags & SWP_FILE) {
 314                 struct address_space *mapping = sis->swap_file->f_mapping;
 315                 return mapping->a_ops->set_page_dirty(page);
 316         } else {
 317                 return __set_page_dirty_no_writeback(page);
 318         }
 319 }