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_status) {
  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         struct task_struct *waiter = bio->bi_private;
 121
 122         if (bio->bi_status) {
 123                 SetPageError(page);
 124                 ClearPageUptodate(page);
 125                 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
 126                          imajor(bio->bi_bdev->bd_inode),
 127                          iminor(bio->bi_bdev->bd_inode),
 128                          (unsigned long long)bio->bi_iter.bi_sector);
 129                 goto out;
 130         }
 131
 132         SetPageUptodate(page);
 133         swap_slot_free_notify(page);
 134 out:
 135         unlock_page(page);
 136         WRITE_ONCE(bio->bi_private, NULL);
 137         bio_put(bio);
 138         wake_up_process(waiter);
 139 }
 140
 141 int generic_swapfile_activate(struct swap_info_struct *sis,
 142                                 struct file *swap_file,
 143                                 sector_t *span)
 144 {
 145         struct address_space *mapping = swap_file->f_mapping;
 146         struct inode *inode = mapping->host;
 147         unsigned blocks_per_page;
 148         unsigned long page_no;
 149         unsigned blkbits;
 150         sector_t probe_block;
 151         sector_t last_block;
 152         sector_t lowest_block = -1;
 153         sector_t highest_block = 0;
 154         int nr_extents = 0;
 155         int ret;
 156
 157         blkbits = inode->i_blkbits;
 158         blocks_per_page = PAGE_SIZE >> blkbits;
 159
 160         /*
 161          * Map all the blocks into the extent list.  This code doesn't try
 162          * to be very smart.
 163          */
 164         probe_block = 0;
 165         page_no = 0;
 166         last_block = i_size_read(inode) >> blkbits;
 167         while ((probe_block + blocks_per_page) <= last_block &&
 168                         page_no < sis->max) {
 169                 unsigned block_in_page;
 170                 sector_t first_block;
 171
 172                 cond_resched();
 173
 174                 first_block = bmap(inode, probe_block);
 175                 if (first_block == 0)
 176                         goto bad_bmap;
 177
 178                 /*
 179                  * It must be PAGE_SIZE aligned on-disk
 180                  */
 181                 if (first_block & (blocks_per_page - 1)) {
 182                         probe_block++;
 183                         goto reprobe;
 184                 }
 185
 186                 for (block_in_page = 1; block_in_page < blocks_per_page;
 187                                         block_in_page++) {
 188                         sector_t block;
 189
 190                         block = bmap(inode, probe_block + block_in_page);
 191                         if (block == 0)
 192                                 goto bad_bmap;
 193                         if (block != first_block + block_in_page) {
 194                                 /* Discontiguity */
 195                                 probe_block++;
 196                                 goto reprobe;
 197                         }
 198                 }
 199
 200                 first_block >>= (PAGE_SHIFT - blkbits);
 201                 if (page_no) {  /* exclude the header page */
 202                         if (first_block < lowest_block)
 203                                 lowest_block = first_block;
 204                         if (first_block > highest_block)
 205                                 highest_block = first_block;
 206                 }
 207
 208                 /*
 209                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 210                  */
 211                 ret = add_swap_extent(sis, page_no, 1, first_block);
 212                 if (ret < 0)
 213                         goto out;
 214                 nr_extents += ret;
 215                 page_no++;
 216                 probe_block += blocks_per_page;
 217 reprobe:
 218                 continue;
 219         }
 220         ret = nr_extents;
 221         *span = 1 + highest_block - lowest_block;
 222         if (page_no == 0)
 223                 page_no = 1;    /* force Empty message */
 224         sis->max = page_no;
 225         sis->pages = page_no - 1;
 226         sis->highest_bit = page_no - 1;
 227 out:
 228         return ret;
 229 bad_bmap:
 230         pr_err("swapon: swapfile has holes\n");
 231         ret = -EINVAL;
 232         goto out;
 233 }
 234
 235 /*
 236  * We may have stale swap cache pages in memory: notice
 237  * them here and get rid of the unnecessary final write.
 238  */
 239 int swap_writepage(struct page *page, struct writeback_control *wbc)
 240 {
 241         int ret = 0;
 242
 243         if (try_to_free_swap(page)) {
 244                 unlock_page(page);
 245                 goto out;
 246         }
 247         if (frontswap_store(page) == 0) {
 248                 set_page_writeback(page);
 249                 unlock_page(page);
 250                 end_page_writeback(page);
 251                 goto out;
 252         }
 253         ret = __swap_writepage(page, wbc, end_swap_bio_write);
 254 out:
 255         return ret;
 256 }
 257
 258 static sector_t swap_page_sector(struct page *page)
 259 {
 260         return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
 261 }
 262
 263 int __swap_writepage(struct page *page, struct writeback_control *wbc,
 264                 bio_end_io_t end_write_func)
 265 {
 266         struct bio *bio;
 267         int ret;
 268         struct swap_info_struct *sis = page_swap_info(page);
 269
 270         VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 271         if (sis->flags & SWP_FILE) {
 272                 struct kiocb kiocb;
 273                 struct file *swap_file = sis->swap_file;
 274                 struct address_space *mapping = swap_file->f_mapping;
 275                 struct bio_vec bv = {
 276                         .bv_page = page,
 277                         .bv_len  = PAGE_SIZE,
 278                         .bv_offset = 0
 279                 };
 280                 struct iov_iter from;
 281
 282                 iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
 283                 init_sync_kiocb(&kiocb, swap_file);
 284                 kiocb.ki_pos = page_file_offset(page);
 285
 286                 set_page_writeback(page);
 287                 unlock_page(page);
 288                 ret = mapping->a_ops->direct_IO(&kiocb, &from);
 289                 if (ret == PAGE_SIZE) {
 290                         count_vm_event(PSWPOUT);
 291                         ret = 0;
 292                 } else {
 293                         /*
 294                          * In the case of swap-over-nfs, this can be a
 295                          * temporary failure if the system has limited
 296                          * memory for allocating transmit buffers.
 297                          * Mark the page dirty and avoid
 298                          * rotate_reclaimable_page but rate-limit the
 299                          * messages but do not flag PageError like
 300                          * the normal direct-to-bio case as it could
 301                          * be temporary.
 302                          */
 303                         set_page_dirty(page);
 304                         ClearPageReclaim(page);
 305                         pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
 306                                            page_file_offset(page));
 307                 }
 308                 end_page_writeback(page);
 309                 return ret;
 310         }
 311
 312         ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
 313         if (!ret) {
 314                 count_vm_event(PSWPOUT);
 315                 return 0;
 316         }
 317
 318         ret = 0;
 319         bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 320         if (bio == NULL) {
 321                 set_page_dirty(page);
 322                 unlock_page(page);
 323                 ret = -ENOMEM;
 324                 goto out;
 325         }
 326         bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
 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, bool do_poll)
 336 {
 337         struct bio *bio;
 338         int ret = 0;
 339         struct swap_info_struct *sis = page_swap_info(page);
 340         blk_qc_t qc;
 341         struct block_device *bdev;
 342
 343         VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 344         VM_BUG_ON_PAGE(!PageLocked(page), page);
 345         VM_BUG_ON_PAGE(PageUptodate(page), page);
 346         if (frontswap_load(page) == 0) {
 347                 SetPageUptodate(page);
 348                 unlock_page(page);
 349                 goto out;
 350         }
 351
 352         if (sis->flags & SWP_FILE) {
 353                 struct file *swap_file = sis->swap_file;
 354                 struct address_space *mapping = swap_file->f_mapping;
 355
 356                 ret = mapping->a_ops->readpage(swap_file, page);
 357                 if (!ret)
 358                         count_vm_event(PSWPIN);
 359                 return ret;
 360         }
 361
 362         ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
 363         if (!ret) {
 364                 if (trylock_page(page)) {
 365                         swap_slot_free_notify(page);
 366                         unlock_page(page);
 367                 }
 368
 369                 count_vm_event(PSWPIN);
 370                 return 0;
 371         }
 372
 373         ret = 0;
 374         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 375         if (bio == NULL) {
 376                 unlock_page(page);
 377                 ret = -ENOMEM;
 378                 goto out;
 379         }
 380         bdev = bio->bi_bdev;
 381         bio->bi_private = current;
 382         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 383         count_vm_event(PSWPIN);
 384         bio_get(bio);
 385         qc = submit_bio(bio);
 386         while (do_poll) {
 387                 set_current_state(TASK_UNINTERRUPTIBLE);
 388                 if (!READ_ONCE(bio->bi_private))
 389                         break;
 390
 391                 if (!blk_mq_poll(bdev_get_queue(bdev), qc))
 392                         break;
 393         }
 394         __set_current_state(TASK_RUNNING);
 395         bio_put(bio);
 396
 397 out:
 398         return ret;
 399 }
 400
 401 int swap_set_page_dirty(struct page *page)
 402 {
 403         struct swap_info_struct *sis = page_swap_info(page);
 404
 405         if (sis->flags & SWP_FILE) {
 406                 struct address_space *mapping = sis->swap_file->f_mapping;
 407
 408                 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 409                 return mapping->a_ops->set_page_dirty(page);
 410         } else {
 411                 return __set_page_dirty_no_writeback(page);
 412         }
 413 }