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/blkdev.h>
  24 #include <linux/uio.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_iter.bi_sector = map_swap_page(page, &bio->bi_bdev);
  35                 bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
  36                 bio->bi_end_io = end_io;
  37
  38                 bio_add_page(bio, page, PAGE_SIZE, 0);
  39                 BUG_ON(bio->bi_iter.bi_size != PAGE_SIZE);
  40         }
  41         return bio;
  42 }
  43
  44 void end_swap_bio_write(struct bio *bio)
  45 {
  46         struct page *page = bio->bi_io_vec[0].bv_page;
  47
  48         if (bio->bi_error) {
  49                 SetPageError(page);
  50                 /*
  51                  * We failed to write the page out to swap-space.
  52                  * Re-dirty the page in order to avoid it being reclaimed.
  53                  * Also print a dire warning that things will go BAD (tm)
  54                  * very quickly.
  55                  *
  56                  * Also clear PG_reclaim to avoid rotate_reclaimable_page()
  57                  */
  58                 set_page_dirty(page);
  59                 pr_alert("Write-error on swap-device (%u:%u:%llu)\n",
  60                          imajor(bio->bi_bdev->bd_inode),
  61                          iminor(bio->bi_bdev->bd_inode),
  62                          (unsigned long long)bio->bi_iter.bi_sector);
  63                 ClearPageReclaim(page);
  64         }
  65         end_page_writeback(page);
  66         bio_put(bio);
  67 }
  68
  69 static void swap_slot_free_notify(struct page *page)
  70 {
  71         struct swap_info_struct *sis;
  72         struct gendisk *disk;
  73
  74         /*
  75          * There is no guarantee that the page is in swap cache - the software
  76          * suspend code (at least) uses end_swap_bio_read() against a non-
  77          * swapcache page.  So we must check PG_swapcache before proceeding with
  78          * this optimization.
  79          */
  80         if (unlikely(!PageSwapCache(page)))
  81                 return;
  82
  83         sis = page_swap_info(page);
  84         if (!(sis->flags & SWP_BLKDEV))
  85                 return;
  86
  87         /*
  88          * The swap subsystem performs lazy swap slot freeing,
  89          * expecting that the page will be swapped out again.
  90          * So we can avoid an unnecessary write if the page
  91          * isn't redirtied.
  92          * This is good for real swap storage because we can
  93          * reduce unnecessary I/O and enhance wear-leveling
  94          * if an SSD is used as the as swap device.
  95          * But if in-memory swap device (eg zram) is used,
  96          * this causes a duplicated copy between uncompressed
  97          * data in VM-owned memory and compressed data in
  98          * zram-owned memory.  So let's free zram-owned memory
  99          * and make the VM-owned decompressed page *dirty*,
 100          * so the page should be swapped out somewhere again if
 101          * we again wish to reclaim it.
 102          */
 103         disk = sis->bdev->bd_disk;
 104         if (disk->fops->swap_slot_free_notify) {
 105                 swp_entry_t entry;
 106                 unsigned long offset;
 107
 108                 entry.val = page_private(page);
 109                 offset = swp_offset(entry);
 110
 111                 SetPageDirty(page);
 112                 disk->fops->swap_slot_free_notify(sis->bdev,
 113                                 offset);
 114         }
 115 }
 116
 117 static void end_swap_bio_read(struct bio *bio)
 118 {
 119         struct page *page = bio->bi_io_vec[0].bv_page;
 120
 121         if (bio->bi_error) {
 122                 SetPageError(page);
 123                 ClearPageUptodate(page);
 124                 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
 125                          imajor(bio->bi_bdev->bd_inode),
 126                          iminor(bio->bi_bdev->bd_inode),
 127                          (unsigned long long)bio->bi_iter.bi_sector);
 128                 goto out;
 129         }
 130
 131         SetPageUptodate(page);
 132         swap_slot_free_notify(page);
 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                 cond_resched();
 170
 171                 first_block = bmap(inode, probe_block);
 172                 if (first_block == 0)
 173                         goto bad_bmap;
 174
 175                 /*
 176                  * It must be PAGE_SIZE aligned on-disk
 177                  */
 178                 if (first_block & (blocks_per_page - 1)) {
 179                         probe_block++;
 180                         goto reprobe;
 181                 }
 182
 183                 for (block_in_page = 1; block_in_page < blocks_per_page;
 184                                         block_in_page++) {
 185                         sector_t block;
 186
 187                         block = bmap(inode, probe_block + block_in_page);
 188                         if (block == 0)
 189                                 goto bad_bmap;
 190                         if (block != first_block + block_in_page) {
 191                                 /* Discontiguity */
 192                                 probe_block++;
 193                                 goto reprobe;
 194                         }
 195                 }
 196
 197                 first_block >>= (PAGE_SHIFT - blkbits);
 198                 if (page_no) {  /* exclude the header page */
 199                         if (first_block < lowest_block)
 200                                 lowest_block = first_block;
 201                         if (first_block > highest_block)
 202                                 highest_block = first_block;
 203                 }
 204
 205                 /*
 206                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 207                  */
 208                 ret = add_swap_extent(sis, page_no, 1, first_block);
 209                 if (ret < 0)
 210                         goto out;
 211                 nr_extents += ret;
 212                 page_no++;
 213                 probe_block += blocks_per_page;
 214 reprobe:
 215                 continue;
 216         }
 217         ret = nr_extents;
 218         *span = 1 + highest_block - lowest_block;
 219         if (page_no == 0)
 220                 page_no = 1;    /* force Empty message */
 221         sis->max = page_no;
 222         sis->pages = page_no - 1;
 223         sis->highest_bit = page_no - 1;
 224 out:
 225         return ret;
 226 bad_bmap:
 227         pr_err("swapon: swapfile has holes\n");
 228         ret = -EINVAL;
 229         goto out;
 230 }
 231
 232 /*
 233  * We may have stale swap cache pages in memory: notice
 234  * them here and get rid of the unnecessary final write.
 235  */
 236 int swap_writepage(struct page *page, struct writeback_control *wbc)
 237 {
 238         int ret = 0;
 239
 240         if (try_to_free_swap(page)) {
 241                 unlock_page(page);
 242                 goto out;
 243         }
 244         if (frontswap_store(page) == 0) {
 245                 set_page_writeback(page);
 246                 unlock_page(page);
 247                 end_page_writeback(page);
 248                 goto out;
 249         }
 250         ret = __swap_writepage(page, wbc, end_swap_bio_write);
 251 out:
 252         return ret;
 253 }
 254
 255 static sector_t swap_page_sector(struct page *page)
 256 {
 257         return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
 258 }
 259
 260 int __swap_writepage(struct page *page, struct writeback_control *wbc,
 261                 bio_end_io_t end_write_func)
 262 {
 263         struct bio *bio;
 264         int ret;
 265         struct swap_info_struct *sis = page_swap_info(page);
 266
 267         VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 268         if (sis->flags & SWP_FILE) {
 269                 struct kiocb kiocb;
 270                 struct file *swap_file = sis->swap_file;
 271                 struct address_space *mapping = swap_file->f_mapping;
 272                 struct bio_vec bv = {
 273                         .bv_page = page,
 274                         .bv_len  = PAGE_SIZE,
 275                         .bv_offset = 0
 276                 };
 277                 struct iov_iter from;
 278
 279                 iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
 280                 init_sync_kiocb(&kiocb, swap_file);
 281                 kiocb.ki_pos = page_file_offset(page);
 282
 283                 set_page_writeback(page);
 284                 unlock_page(page);
 285                 ret = mapping->a_ops->direct_IO(&kiocb, &from);
 286                 if (ret == PAGE_SIZE) {
 287                         count_vm_event(PSWPOUT);
 288                         ret = 0;
 289                 } else {
 290                         /*
 291                          * In the case of swap-over-nfs, this can be a
 292                          * temporary failure if the system has limited
 293                          * memory for allocating transmit buffers.
 294                          * Mark the page dirty and avoid
 295                          * rotate_reclaimable_page but rate-limit the
 296                          * messages but do not flag PageError like
 297                          * the normal direct-to-bio case as it could
 298                          * be temporary.
 299                          */
 300                         set_page_dirty(page);
 301                         ClearPageReclaim(page);
 302                         pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
 303                                            page_file_offset(page));
 304                 }
 305                 end_page_writeback(page);
 306                 return ret;
 307         }
 308
 309         ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
 310         if (!ret) {
 311                 count_vm_event(PSWPOUT);
 312                 return 0;
 313         }
 314
 315         ret = 0;
 316         bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 317         if (bio == NULL) {
 318                 set_page_dirty(page);
 319                 unlock_page(page);
 320                 ret = -ENOMEM;
 321                 goto out;
 322         }
 323         if (wbc->sync_mode == WB_SYNC_ALL)
 324                 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC);
 325         else
 326                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 327         count_vm_event(PSWPOUT);
 328         set_page_writeback(page);
 329         unlock_page(page);
 330         submit_bio(bio);
 331 out:
 332         return ret;
 333 }
 334
 335 int swap_readpage(struct page *page)
 336 {
 337         struct bio *bio;
 338         int ret = 0;
 339         struct swap_info_struct *sis = page_swap_info(page);
 340
 341         VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 342         VM_BUG_ON_PAGE(!PageLocked(page), page);
 343         VM_BUG_ON_PAGE(PageUptodate(page), page);
 344         if (frontswap_load(page) == 0) {
 345                 SetPageUptodate(page);
 346                 unlock_page(page);
 347                 goto out;
 348         }
 349
 350         if (sis->flags & SWP_FILE) {
 351                 struct file *swap_file = sis->swap_file;
 352                 struct address_space *mapping = swap_file->f_mapping;
 353
 354                 ret = mapping->a_ops->readpage(swap_file, page);
 355                 if (!ret)
 356                         count_vm_event(PSWPIN);
 357                 return ret;
 358         }
 359
 360         ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
 361         if (!ret) {
 362                 if (trylock_page(page)) {
 363                         swap_slot_free_notify(page);
 364                         unlock_page(page);
 365                 }
 366
 367                 count_vm_event(PSWPIN);
 368                 return 0;
 369         }
 370
 371         ret = 0;
 372         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 373         if (bio == NULL) {
 374                 unlock_page(page);
 375                 ret = -ENOMEM;
 376                 goto out;
 377         }
 378         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 379         count_vm_event(PSWPIN);
 380         submit_bio(bio);
 381 out:
 382         return ret;
 383 }
 384
 385 int swap_set_page_dirty(struct page *page)
 386 {
 387         struct swap_info_struct *sis = page_swap_info(page);
 388
 389         if (sis->flags & SWP_FILE) {
 390                 struct address_space *mapping = sis->swap_file->f_mapping;
 391
 392                 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 393                 return mapping->a_ops->set_page_dirty(page);
 394         } else {
 395                 return __set_page_dirty_no_writeback(page);
 396         }
 397 }