mm/page_io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/mm/page_io.c
   4  *
   5  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   6  *
   7  *  Swap reorganised 29.12.95,
   8  *  Asynchronous swapping added 30.12.95. Stephen Tweedie
   9  *  Removed race in async swapping. 14.4.1996. Bruno Haible
  10  *  Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie
  11  *  Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman
  12  */
  13
  14 #include <linux/mm.h>
  15 #include <linux/kernel_stat.h>
  16 #include <linux/gfp.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/swap.h>
  19 #include <linux/bio.h>
  20 #include <linux/swapops.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/writeback.h>
  23 #include <linux/frontswap.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/psi.h>
  26 #include <linux/uio.h>
  27 #include <linux/sched/task.h>
  28 #include <asm/pgtable.h>
  29
  30 static struct bio *get_swap_bio(gfp_t gfp_flags,
  31                                 struct page *page, bio_end_io_t end_io)
  32 {
  33         struct bio *bio;
  34
  35         bio = bio_alloc(gfp_flags, 1);
  36         if (bio) {
  37                 struct block_device *bdev;
  38
  39                 bio->bi_iter.bi_sector = map_swap_page(page, &bdev);
  40                 bio_set_dev(bio, bdev);
  41                 bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
  42                 bio->bi_end_io = end_io;
  43
  44                 bio_add_page(bio, page, PAGE_SIZE * hpage_nr_pages(page), 0);
  45         }
  46         return bio;
  47 }
  48
  49 void end_swap_bio_write(struct bio *bio)
  50 {
  51         struct page *page = bio_first_page_all(bio);
  52
  53         if (bio->bi_status) {
  54                 SetPageError(page);
  55                 /*
  56                  * We failed to write the page out to swap-space.
  57                  * Re-dirty the page in order to avoid it being reclaimed.
  58                  * Also print a dire warning that things will go BAD (tm)
  59                  * very quickly.
  60                  *
  61                  * Also clear PG_reclaim to avoid rotate_reclaimable_page()
  62                  */
  63                 set_page_dirty(page);
  64                 pr_alert("Write-error on swap-device (%u:%u:%llu)\n",
  65                          MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
  66                          (unsigned long long)bio->bi_iter.bi_sector);
  67                 ClearPageReclaim(page);
  68         }
  69         end_page_writeback(page);
  70         bio_put(bio);
  71 }
  72
  73 static void swap_slot_free_notify(struct page *page)
  74 {
  75         struct swap_info_struct *sis;
  76         struct gendisk *disk;
  77         swp_entry_t entry;
  78
  79         /*
  80          * There is no guarantee that the page is in swap cache - the software
  81          * suspend code (at least) uses end_swap_bio_read() against a non-
  82          * swapcache page.  So we must check PG_swapcache before proceeding with
  83          * this optimization.
  84          */
  85         if (unlikely(!PageSwapCache(page)))
  86                 return;
  87
  88         sis = page_swap_info(page);
  89         if (!(sis->flags & SWP_BLKDEV))
  90                 return;
  91
  92         /*
  93          * The swap subsystem performs lazy swap slot freeing,
  94          * expecting that the page will be swapped out again.
  95          * So we can avoid an unnecessary write if the page
  96          * isn't redirtied.
  97          * This is good for real swap storage because we can
  98          * reduce unnecessary I/O and enhance wear-leveling
  99          * if an SSD is used as the as swap device.
 100          * But if in-memory swap device (eg zram) is used,
 101          * this causes a duplicated copy between uncompressed
 102          * data in VM-owned memory and compressed data in
 103          * zram-owned memory.  So let's free zram-owned memory
 104          * and make the VM-owned decompressed page *dirty*,
 105          * so the page should be swapped out somewhere again if
 106          * we again wish to reclaim it.
 107          */
 108         disk = sis->bdev->bd_disk;
 109         entry.val = page_private(page);
 110         if (disk->fops->swap_slot_free_notify && __swap_count(entry) == 1) {
 111                 unsigned long offset;
 112
 113                 offset = swp_offset(entry);
 114
 115                 SetPageDirty(page);
 116                 disk->fops->swap_slot_free_notify(sis->bdev,
 117                                 offset);
 118         }
 119 }
 120
 121 static void end_swap_bio_read(struct bio *bio)
 122 {
 123         struct page *page = bio_first_page_all(bio);
 124         struct task_struct *waiter = bio->bi_private;
 125
 126         if (bio->bi_status) {
 127                 SetPageError(page);
 128                 ClearPageUptodate(page);
 129                 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
 130                          MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
 131                          (unsigned long long)bio->bi_iter.bi_sector);
 132                 goto out;
 133         }
 134
 135         SetPageUptodate(page);
 136         swap_slot_free_notify(page);
 137 out:
 138         unlock_page(page);
 139         WRITE_ONCE(bio->bi_private, NULL);
 140         bio_put(bio);
 141         if (waiter) {
 142                 blk_wake_io_task(waiter);
 143                 put_task_struct(waiter);
 144         }
 145 }
 146
 147 int generic_swapfile_activate(struct swap_info_struct *sis,
 148                                 struct file *swap_file,
 149                                 sector_t *span)
 150 {
 151         struct address_space *mapping = swap_file->f_mapping;
 152         struct inode *inode = mapping->host;
 153         unsigned blocks_per_page;
 154         unsigned long page_no;
 155         unsigned blkbits;
 156         sector_t probe_block;
 157         sector_t last_block;
 158         sector_t lowest_block = -1;
 159         sector_t highest_block = 0;
 160         int nr_extents = 0;
 161         int ret;
 162
 163         blkbits = inode->i_blkbits;
 164         blocks_per_page = PAGE_SIZE >> blkbits;
 165
 166         /*
 167          * Map all the blocks into the extent tree.  This code doesn't try
 168          * to be very smart.
 169          */
 170         probe_block = 0;
 171         page_no = 0;
 172         last_block = i_size_read(inode) >> blkbits;
 173         while ((probe_block + blocks_per_page) <= last_block &&
 174                         page_no < sis->max) {
 175                 unsigned block_in_page;
 176                 sector_t first_block;
 177
 178                 cond_resched();
 179
 180                 first_block = bmap(inode, probe_block);
 181                 if (first_block == 0)
 182                         goto bad_bmap;
 183
 184                 /*
 185                  * It must be PAGE_SIZE aligned on-disk
 186                  */
 187                 if (first_block & (blocks_per_page - 1)) {
 188                         probe_block++;
 189                         goto reprobe;
 190                 }
 191
 192                 for (block_in_page = 1; block_in_page < blocks_per_page;
 193                                         block_in_page++) {
 194                         sector_t block;
 195
 196                         block = bmap(inode, probe_block + block_in_page);
 197                         if (block == 0)
 198                                 goto bad_bmap;
 199                         if (block != first_block + block_in_page) {
 200                                 /* Discontiguity */
 201                                 probe_block++;
 202                                 goto reprobe;
 203                         }
 204                 }
 205
 206                 first_block >>= (PAGE_SHIFT - blkbits);
 207                 if (page_no) {  /* exclude the header page */
 208                         if (first_block < lowest_block)
 209                                 lowest_block = first_block;
 210                         if (first_block > highest_block)
 211                                 highest_block = first_block;
 212                 }
 213
 214                 /*
 215                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 216                  */
 217                 ret = add_swap_extent(sis, page_no, 1, first_block);
 218                 if (ret < 0)
 219                         goto out;
 220                 nr_extents += ret;
 221                 page_no++;
 222                 probe_block += blocks_per_page;
 223 reprobe:
 224                 continue;
 225         }
 226         ret = nr_extents;
 227         *span = 1 + highest_block - lowest_block;
 228         if (page_no == 0)
 229                 page_no = 1;    /* force Empty message */
 230         sis->max = page_no;
 231         sis->pages = page_no - 1;
 232         sis->highest_bit = page_no - 1;
 233 out:
 234         return ret;
 235 bad_bmap:
 236         pr_err("swapon: swapfile has holes\n");
 237         ret = -EINVAL;
 238         goto out;
 239 }
 240
 241 /*
 242  * We may have stale swap cache pages in memory: notice
 243  * them here and get rid of the unnecessary final write.
 244  */
 245 int swap_writepage(struct page *page, struct writeback_control *wbc)
 246 {
 247         int ret = 0;
 248
 249         if (try_to_free_swap(page)) {
 250                 unlock_page(page);
 251                 goto out;
 252         }
 253         if (frontswap_store(page) == 0) {
 254                 set_page_writeback(page);
 255                 unlock_page(page);
 256                 end_page_writeback(page);
 257                 goto out;
 258         }
 259         ret = __swap_writepage(page, wbc, end_swap_bio_write);
 260 out:
 261         return ret;
 262 }
 263
 264 static sector_t swap_page_sector(struct page *page)
 265 {
 266         return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
 267 }
 268
 269 static inline void count_swpout_vm_event(struct page *page)
 270 {
 271 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 272         if (unlikely(PageTransHuge(page)))
 273                 count_vm_event(THP_SWPOUT);
 274 #endif
 275         count_vm_events(PSWPOUT, hpage_nr_pages(page));
 276 }
 277
 278 int __swap_writepage(struct page *page, struct writeback_control *wbc,
 279                 bio_end_io_t end_write_func)
 280 {
 281         struct bio *bio;
 282         int ret;
 283         struct swap_info_struct *sis = page_swap_info(page);
 284
 285         VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 286         if (sis->flags & SWP_FS) {
 287                 struct kiocb kiocb;
 288                 struct file *swap_file = sis->swap_file;
 289                 struct address_space *mapping = swap_file->f_mapping;
 290                 struct bio_vec bv = {
 291                         .bv_page = page,
 292                         .bv_len  = PAGE_SIZE,
 293                         .bv_offset = 0
 294                 };
 295                 struct iov_iter from;
 296
 297                 iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
 298                 init_sync_kiocb(&kiocb, swap_file);
 299                 kiocb.ki_pos = page_file_offset(page);
 300
 301                 set_page_writeback(page);
 302                 unlock_page(page);
 303                 ret = mapping->a_ops->direct_IO(&kiocb, &from);
 304                 if (ret == PAGE_SIZE) {
 305                         count_vm_event(PSWPOUT);
 306                         ret = 0;
 307                 } else {
 308                         /*
 309                          * In the case of swap-over-nfs, this can be a
 310                          * temporary failure if the system has limited
 311                          * memory for allocating transmit buffers.
 312                          * Mark the page dirty and avoid
 313                          * rotate_reclaimable_page but rate-limit the
 314                          * messages but do not flag PageError like
 315                          * the normal direct-to-bio case as it could
 316                          * be temporary.
 317                          */
 318                         set_page_dirty(page);
 319                         ClearPageReclaim(page);
 320                         pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
 321                                            page_file_offset(page));
 322                 }
 323                 end_page_writeback(page);
 324                 return ret;
 325         }
 326
 327         ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
 328         if (!ret) {
 329                 count_swpout_vm_event(page);
 330                 return 0;
 331         }
 332
 333         ret = 0;
 334         bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 335         if (bio == NULL) {
 336                 set_page_dirty(page);
 337                 unlock_page(page);
 338                 ret = -ENOMEM;
 339                 goto out;
 340         }
 341         bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
 342         bio_associate_blkg_from_page(bio, page);
 343         count_swpout_vm_event(page);
 344         set_page_writeback(page);
 345         unlock_page(page);
 346         submit_bio(bio);
 347 out:
 348         return ret;
 349 }
 350
 351 int swap_readpage(struct page *page, bool synchronous)
 352 {
 353         struct bio *bio;
 354         int ret = 0;
 355         struct swap_info_struct *sis = page_swap_info(page);
 356         blk_qc_t qc;
 357         struct gendisk *disk;
 358         unsigned long pflags;
 359
 360         VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
 361         VM_BUG_ON_PAGE(!PageLocked(page), page);
 362         VM_BUG_ON_PAGE(PageUptodate(page), page);
 363
 364         /*
 365          * Count submission time as memory stall. When the device is congested,
 366          * or the submitting cgroup IO-throttled, submission can be a
 367          * significant part of overall IO time.
 368          */
 369         psi_memstall_enter(&pflags);
 370
 371         if (frontswap_load(page) == 0) {
 372                 SetPageUptodate(page);
 373                 unlock_page(page);
 374                 goto out;
 375         }
 376
 377         if (sis->flags & SWP_FS) {
 378                 struct file *swap_file = sis->swap_file;
 379                 struct address_space *mapping = swap_file->f_mapping;
 380
 381                 ret = mapping->a_ops->readpage(swap_file, page);
 382                 if (!ret)
 383                         count_vm_event(PSWPIN);
 384                 goto out;
 385         }
 386
 387         ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
 388         if (!ret) {
 389                 if (trylock_page(page)) {
 390                         swap_slot_free_notify(page);
 391                         unlock_page(page);
 392                 }
 393
 394                 count_vm_event(PSWPIN);
 395                 goto out;
 396         }
 397
 398         ret = 0;
 399         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 400         if (bio == NULL) {
 401                 unlock_page(page);
 402                 ret = -ENOMEM;
 403                 goto out;
 404         }
 405         disk = bio->bi_disk;
 406         /*
 407          * Keep this task valid during swap readpage because the oom killer may
 408          * attempt to access it in the page fault retry time check.
 409          */
 410         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 411         if (synchronous) {
 412                 bio->bi_opf |= REQ_HIPRI;
 413                 get_task_struct(current);
 414                 bio->bi_private = current;
 415         }
 416         count_vm_event(PSWPIN);
 417         bio_get(bio);
 418         qc = submit_bio(bio);
 419         while (synchronous) {
 420                 set_current_state(TASK_UNINTERRUPTIBLE);
 421                 if (!READ_ONCE(bio->bi_private))
 422                         break;
 423
 424                 if (!blk_poll(disk->queue, qc, true))
 425                         io_schedule();
 426         }
 427         __set_current_state(TASK_RUNNING);
 428         bio_put(bio);
 429
 430 out:
 431         psi_memstall_leave(&pflags);
 432         return ret;
 433 }
 434
 435 int swap_set_page_dirty(struct page *page)
 436 {
 437         struct swap_info_struct *sis = page_swap_info(page);
 438
 439         if (sis->flags & SWP_FS) {
 440                 struct address_space *mapping = sis->swap_file->f_mapping;
 441
 442                 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 443                 return mapping->a_ops->set_page_dirty(page);
 444         } else {
 445                 return __set_page_dirty_no_writeback(page);
 446         }
 447 }