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_size = PAGE_SIZE;
  40                 bio->bi_end_io = end_io;
  41         }
  42         return bio;
  43 }
  44
  45 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         int ret = 0;
 189
 190         if (try_to_free_swap(page)) {
 191                 unlock_page(page);
 192                 goto out;
 193         }
 194         if (frontswap_store(page) == 0) {
 195                 set_page_writeback(page);
 196                 unlock_page(page);
 197                 end_page_writeback(page);
 198                 goto out;
 199         }
 200         ret = __swap_writepage(page, wbc, end_swap_bio_write);
 201 out:
 202         return ret;
 203 }
 204
 205 int __swap_writepage(struct page *page, struct writeback_control *wbc,
 206         void (*end_write_func)(struct bio *, int))
 207 {
 208         struct bio *bio;
 209         int ret = 0, rw = WRITE;
 210         struct swap_info_struct *sis = page_swap_info(page);
 211
 212         if (sis->flags & SWP_FILE) {
 213                 struct kiocb kiocb;
 214                 struct file *swap_file = sis->swap_file;
 215                 struct address_space *mapping = swap_file->f_mapping;
 216                 struct iovec iov = {
 217                         .iov_base = kmap(page),
 218                         .iov_len  = PAGE_SIZE,
 219                 };
 220
 221                 init_sync_kiocb(&kiocb, swap_file);
 222                 kiocb.ki_pos = page_file_offset(page);
 223                 kiocb.ki_left = PAGE_SIZE;
 224                 kiocb.ki_nbytes = PAGE_SIZE;
 225
 226                 set_page_writeback(page);
 227                 unlock_page(page);
 228                 ret = mapping->a_ops->direct_IO(KERNEL_WRITE,
 229                                                 &kiocb, &iov,
 230                                                 kiocb.ki_pos, 1);
 231                 kunmap(page);
 232                 if (ret == PAGE_SIZE) {
 233                         count_vm_event(PSWPOUT);
 234                         ret = 0;
 235                 } else {
 236                         /*
 237                          * In the case of swap-over-nfs, this can be a
 238                          * temporary failure if the system has limited
 239                          * memory for allocating transmit buffers.
 240                          * Mark the page dirty and avoid
 241                          * rotate_reclaimable_page but rate-limit the
 242                          * messages but do not flag PageError like
 243                          * the normal direct-to-bio case as it could
 244                          * be temporary.
 245                          */
 246                         set_page_dirty(page);
 247                         ClearPageReclaim(page);
 248                         pr_err_ratelimited("Write error on dio swapfile (%Lu)\n",
 249                                 page_file_offset(page));
 250                 }
 251                 end_page_writeback(page);
 252                 return ret;
 253         }
 254
 255         bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 256         if (bio == NULL) {
 257                 set_page_dirty(page);
 258                 unlock_page(page);
 259                 ret = -ENOMEM;
 260                 goto out;
 261         }
 262         if (wbc->sync_mode == WB_SYNC_ALL)
 263                 rw |= REQ_SYNC;
 264         count_vm_event(PSWPOUT);
 265         set_page_writeback(page);
 266         unlock_page(page);
 267         submit_bio(rw, bio);
 268 out:
 269         return ret;
 270 }
 271
 272 int swap_readpage(struct page *page)
 273 {
 274         struct bio *bio;
 275         int ret = 0;
 276         struct swap_info_struct *sis = page_swap_info(page);
 277
 278         VM_BUG_ON(!PageLocked(page));
 279         VM_BUG_ON(PageUptodate(page));
 280         if (frontswap_load(page) == 0) {
 281                 SetPageUptodate(page);
 282                 unlock_page(page);
 283                 goto out;
 284         }
 285
 286         if (sis->flags & SWP_FILE) {
 287                 struct file *swap_file = sis->swap_file;
 288                 struct address_space *mapping = swap_file->f_mapping;
 289
 290                 ret = mapping->a_ops->readpage(swap_file, page);
 291                 if (!ret)
 292                         count_vm_event(PSWPIN);
 293                 return ret;
 294         }
 295
 296         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 297         if (bio == NULL) {
 298                 unlock_page(page);
 299                 ret = -ENOMEM;
 300                 goto out;
 301         }
 302         count_vm_event(PSWPIN);
 303         submit_bio(READ, bio);
 304 out:
 305         return ret;
 306 }
 307
 308 int swap_set_page_dirty(struct page *page)
 309 {
 310         struct swap_info_struct *sis = page_swap_info(page);
 311
 312         if (sis->flags & SWP_FILE) {
 313                 struct address_space *mapping = sis->swap_file->f_mapping;
 314                 return mapping->a_ops->set_page_dirty(page);
 315         } else {
 316                 return __set_page_dirty_no_writeback(page);
 317         }
 318 }