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 <linux/blkdev.h>
  25 #include <asm/pgtable.h>
  26
  27 static struct bio *get_swap_bio(gfp_t gfp_flags,
  28                                 struct page *page, bio_end_io_t end_io)
  29 {
  30         struct bio *bio;
  31
  32         bio = bio_alloc(gfp_flags, 1);
  33         if (bio) {
  34                 bio->bi_sector = map_swap_page(page, &bio->bi_bdev);
  35                 bio->bi_sector <<= PAGE_SHIFT - 9;
  36                 bio->bi_io_vec[0].bv_page = page;
  37                 bio->bi_io_vec[0].bv_len = PAGE_SIZE;
  38                 bio->bi_io_vec[0].bv_offset = 0;
  39                 bio->bi_vcnt = 1;
  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 #ifndef CONFIG_VNSWAP
  63                 printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n",
  64                                 imajor(bio->bi_bdev->bd_inode),
  65                                 iminor(bio->bi_bdev->bd_inode),
  66                                 (unsigned long long)bio->bi_sector);
  67 #endif
  68                 ClearPageReclaim(page);
  69         }
  70         end_page_writeback(page);
  71         bio_put(bio);
  72 }
  73
  74 void end_swap_bio_read(struct bio *bio, int err)
  75 {
  76         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
  77         struct page *page = bio->bi_io_vec[0].bv_page;
  78
  79         if (!uptodate) {
  80                 SetPageError(page);
  81                 ClearPageUptodate(page);
  82                 printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n",
  83                                 imajor(bio->bi_bdev->bd_inode),
  84                                 iminor(bio->bi_bdev->bd_inode),
  85                                 (unsigned long long)bio->bi_sector);
  86                 goto out;
  87         }
  88
  89         SetPageUptodate(page);
  90
  91         /*
  92          * There is no guarantee that the page is in swap cache - the software
  93          * suspend code (at least) uses end_swap_bio_read() against a non-
  94          * swapcache page.  So we must check PG_swapcache before proceeding with
  95          * this optimization.
  96          */
  97         if (likely(PageSwapCache(page))) {
  98                 struct swap_info_struct *sis;
  99
 100                 sis = page_swap_info(page);
 101                 if (sis->flags & SWP_BLKDEV) {
 102                         /*
 103                          * The swap subsystem performs lazy swap slot freeing,
 104                          * expecting that the page will be swapped out again.
 105                          * So we can avoid an unnecessary write if the page
 106                          * isn't redirtied.
 107                          * This is good for real swap storage because we can
 108                          * reduce unnecessary I/O and enhance wear-leveling
 109                          * if an SSD is used as the as swap device.
 110                          * But if in-memory swap device (eg zram) is used,
 111                          * this causes a duplicated copy between uncompressed
 112                          * data in VM-owned memory and compressed data in
 113                          * zram-owned memory.  So let's free zram-owned memory
 114                          * and make the VM-owned decompressed page *dirty*,
 115                          * so the page should be swapped out somewhere again if
 116                          * we again wish to reclaim it.
 117                          */
 118                         struct gendisk *disk = sis->bdev->bd_disk;
 119                         if (disk->fops->swap_slot_free_notify) {
 120                                 swp_entry_t entry;
 121                                 unsigned long offset;
 122
 123                                 entry.val = page_private(page);
 124                                 offset = swp_offset(entry);
 125
 126                                 SetPageDirty(page);
 127                                 disk->fops->swap_slot_free_notify(sis->bdev,
 128                                                 offset);
 129                         }
 130                 }
 131         }
 132
 133 out:
 134         unlock_page(page);
 135         bio_put(bio);
 136 }
 137
 138 int generic_swapfile_activate(struct swap_info_struct *sis,
 139                                 struct file *swap_file,
 140                                 sector_t *span)
 141 {
 142         struct address_space *mapping = swap_file->f_mapping;
 143         struct inode *inode = mapping->host;
 144         unsigned blocks_per_page;
 145         unsigned long page_no;
 146         unsigned blkbits;
 147         sector_t probe_block;
 148         sector_t last_block;
 149         sector_t lowest_block = -1;
 150         sector_t highest_block = 0;
 151         int nr_extents = 0;
 152         int ret;
 153
 154         blkbits = inode->i_blkbits;
 155         blocks_per_page = PAGE_SIZE >> blkbits;
 156
 157         /*
 158          * Map all the blocks into the extent list.  This code doesn't try
 159          * to be very smart.
 160          */
 161         probe_block = 0;
 162         page_no = 0;
 163         last_block = i_size_read(inode) >> blkbits;
 164         while ((probe_block + blocks_per_page) <= last_block &&
 165                         page_no < sis->max) {
 166                 unsigned block_in_page;
 167                 sector_t first_block;
 168
 169                 first_block = bmap(inode, probe_block);
 170                 if (first_block == 0)
 171                         goto bad_bmap;
 172
 173                 /*
 174                  * It must be PAGE_SIZE aligned on-disk
 175                  */
 176                 if (first_block & (blocks_per_page - 1)) {
 177                         probe_block++;
 178                         goto reprobe;
 179                 }
 180
 181                 for (block_in_page = 1; block_in_page < blocks_per_page;
 182                                         block_in_page++) {
 183                         sector_t block;
 184
 185                         block = bmap(inode, probe_block + block_in_page);
 186                         if (block == 0)
 187                                 goto bad_bmap;
 188                         if (block != first_block + block_in_page) {
 189                                 /* Discontiguity */
 190                                 probe_block++;
 191                                 goto reprobe;
 192                         }
 193                 }
 194
 195                 first_block >>= (PAGE_SHIFT - blkbits);
 196                 if (page_no) {  /* exclude the header page */
 197                         if (first_block < lowest_block)
 198                                 lowest_block = first_block;
 199                         if (first_block > highest_block)
 200                                 highest_block = first_block;
 201                 }
 202
 203                 /*
 204                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 205                  */
 206                 ret = add_swap_extent(sis, page_no, 1, first_block);
 207                 if (ret < 0)
 208                         goto out;
 209                 nr_extents += ret;
 210                 page_no++;
 211                 probe_block += blocks_per_page;
 212 reprobe:
 213                 continue;
 214         }
 215         ret = nr_extents;
 216         *span = 1 + highest_block - lowest_block;
 217         if (page_no == 0)
 218                 page_no = 1;    /* force Empty message */
 219         sis->max = page_no;
 220         sis->pages = page_no - 1;
 221         sis->highest_bit = page_no - 1;
 222 out:
 223         return ret;
 224 bad_bmap:
 225         printk(KERN_ERR "swapon: swapfile has holes\n");
 226         ret = -EINVAL;
 227         goto out;
 228 }
 229
 230 /*
 231  * We may have stale swap cache pages in memory: notice
 232  * them here and get rid of the unnecessary final write.
 233  */
 234 int swap_writepage(struct page *page, struct writeback_control *wbc)
 235 {
 236         int ret = 0;
 237
 238         if (try_to_free_swap(page)) {
 239                 unlock_page(page);
 240                 goto out;
 241         }
 242         if (frontswap_store(page) == 0) {
 243                 set_page_writeback(page);
 244                 unlock_page(page);
 245                 end_page_writeback(page);
 246                 goto out;
 247         }
 248         ret = __swap_writepage(page, wbc, end_swap_bio_write);
 249 out:
 250         return ret;
 251 }
 252
 253 int __swap_writepage(struct page *page, struct writeback_control *wbc,
 254         void (*end_write_func)(struct bio *, int))
 255 {
 256         struct bio *bio;
 257         int ret = 0, rw = WRITE;
 258         struct swap_info_struct *sis = page_swap_info(page);
 259
 260         if (sis->flags & SWP_FILE) {
 261                 struct kiocb kiocb;
 262                 struct file *swap_file = sis->swap_file;
 263                 struct address_space *mapping = swap_file->f_mapping;
 264                 struct iovec iov = {
 265                         .iov_base = kmap(page),
 266                         .iov_len  = PAGE_SIZE,
 267                 };
 268
 269                 init_sync_kiocb(&kiocb, swap_file);
 270                 kiocb.ki_pos = page_file_offset(page);
 271                 kiocb.ki_left = PAGE_SIZE;
 272                 kiocb.ki_nbytes = PAGE_SIZE;
 273
 274                 set_page_writeback(page);
 275                 unlock_page(page);
 276                 ret = mapping->a_ops->direct_IO(KERNEL_WRITE,
 277                                                 &kiocb, &iov,
 278                                                 kiocb.ki_pos, 1);
 279                 kunmap(page);
 280                 if (ret == PAGE_SIZE) {
 281                         count_vm_event(PSWPOUT);
 282                         ret = 0;
 283                 } else {
 284                         /*
 285                          * In the case of swap-over-nfs, this can be a
 286                          * temporary failure if the system has limited
 287                          * memory for allocating transmit buffers.
 288                          * Mark the page dirty and avoid
 289                          * rotate_reclaimable_page but rate-limit the
 290                          * messages but do not flag PageError like
 291                          * the normal direct-to-bio case as it could
 292                          * be temporary.
 293                          */
 294                         set_page_dirty(page);
 295                         ClearPageReclaim(page);
 296                         pr_err_ratelimited("Write error on dio swapfile (%Lu)\n",
 297                                 page_file_offset(page));
 298                 }
 299                 end_page_writeback(page);
 300                 return ret;
 301         }
 302
 303         bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 304         if (bio == NULL) {
 305                 set_page_dirty(page);
 306                 unlock_page(page);
 307                 ret = -ENOMEM;
 308                 goto out;
 309         }
 310         if (wbc->sync_mode == WB_SYNC_ALL)
 311                 rw |= REQ_SYNC;
 312         count_vm_event(PSWPOUT);
 313         set_page_writeback(page);
 314         unlock_page(page);
 315         submit_bio(rw, bio);
 316 out:
 317         return ret;
 318 }
 319
 320 int swap_readpage(struct page *page)
 321 {
 322         struct bio *bio;
 323         int ret = 0;
 324         struct swap_info_struct *sis = page_swap_info(page);
 325
 326         VM_BUG_ON(!PageLocked(page));
 327         VM_BUG_ON(PageUptodate(page));
 328         if (frontswap_load(page) == 0) {
 329                 SetPageUptodate(page);
 330                 unlock_page(page);
 331                 goto out;
 332         }
 333
 334         if (sis->flags & SWP_FILE) {
 335                 struct file *swap_file = sis->swap_file;
 336                 struct address_space *mapping = swap_file->f_mapping;
 337
 338                 ret = mapping->a_ops->readpage(swap_file, page);
 339                 if (!ret)
 340                         count_vm_event(PSWPIN);
 341                 return ret;
 342         }
 343
 344         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 345         if (bio == NULL) {
 346                 unlock_page(page);
 347                 ret = -ENOMEM;
 348                 goto out;
 349         }
 350         count_vm_event(PSWPIN);
 351         submit_bio(READ, bio);
 352 out:
 353         return ret;
 354 }
 355
 356 int swap_set_page_dirty(struct page *page)
 357 {
 358         struct swap_info_struct *sis = page_swap_info(page);
 359
 360         if (sis->flags & SWP_FILE) {
 361                 struct address_space *mapping = sis->swap_file->f_mapping;
 362                 return mapping->a_ops->set_page_dirty(page);
 363         } else {
 364                 return __set_page_dirty_no_writeback(page);
 365         }
 366 }