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 = probe_block;
 181                 ret = bmap(inode, &first_block);
 182                 if (ret || !first_block)
 183                         goto bad_bmap;
 184
 185                 /*
 186                  * It must be PAGE_SIZE aligned on-disk
 187                  */
 188                 if (first_block & (blocks_per_page - 1)) {
 189                         probe_block++;
 190                         goto reprobe;
 191                 }
 192
 193                 for (block_in_page = 1; block_in_page < blocks_per_page;
 194                                         block_in_page++) {
 195                         sector_t block;
 196
 197                         block = probe_block + block_in_page;
 198                         ret = bmap(inode, &block);
 199                         if (ret || !block)
 200                                 goto bad_bmap;
 201
 202                         if (block != first_block + block_in_page) {
 203                                 /* Discontiguity */
 204                                 probe_block++;
 205                                 goto reprobe;
 206                         }
 207                 }
 208
 209                 first_block >>= (PAGE_SHIFT - blkbits);
 210                 if (page_no) {  /* exclude the header page */
 211                         if (first_block < lowest_block)
 212                                 lowest_block = first_block;
 213                         if (first_block > highest_block)
 214                                 highest_block = first_block;
 215                 }
 216
 217                 /*
 218                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 219                  */
 220                 ret = add_swap_extent(sis, page_no, 1, first_block);
 221                 if (ret < 0)
 222                         goto out;
 223                 nr_extents += ret;
 224                 page_no++;
 225                 probe_block += blocks_per_page;
 226 reprobe:
 227                 continue;
 228         }
 229         ret = nr_extents;
 230         *span = 1 + highest_block - lowest_block;
 231         if (page_no == 0)
 232                 page_no = 1;    /* force Empty message */
 233         sis->max = page_no;
 234         sis->pages = page_no - 1;
 235         sis->highest_bit = page_no - 1;
 236 out:
 237         return ret;
 238 bad_bmap:
 239         pr_err("swapon: swapfile has holes\n");
 240         ret = -EINVAL;
 241         goto out;
 242 }
 243
 244 /*
 245  * We may have stale swap cache pages in memory: notice
 246  * them here and get rid of the unnecessary final write.
 247  */
 248 int swap_writepage(struct page *page, struct writeback_control *wbc)
 249 {
 250         int ret = 0;
 251
 252         if (try_to_free_swap(page)) {
 253                 unlock_page(page);
 254                 goto out;
 255         }
 256         if (frontswap_store(page) == 0) {
 257                 set_page_writeback(page);
 258                 unlock_page(page);
 259                 end_page_writeback(page);
 260                 goto out;
 261         }
 262         ret = __swap_writepage(page, wbc, end_swap_bio_write);
 263 out:
 264         return ret;
 265 }
 266
 267 static sector_t swap_page_sector(struct page *page)
 268 {
 269         return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
 270 }
 271
 272 static inline void count_swpout_vm_event(struct page *page)
 273 {
 274 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 275         if (unlikely(PageTransHuge(page)))
 276                 count_vm_event(THP_SWPOUT);
 277 #endif
 278         count_vm_events(PSWPOUT, hpage_nr_pages(page));
 279 }
 280
 281 int __swap_writepage(struct page *page, struct writeback_control *wbc,
 282                 bio_end_io_t end_write_func)
 283 {
 284         struct bio *bio;
 285         int ret;
 286         struct swap_info_struct *sis = page_swap_info(page);
 287
 288         VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 289         if (sis->flags & SWP_FS) {
 290                 struct kiocb kiocb;
 291                 struct file *swap_file = sis->swap_file;
 292                 struct address_space *mapping = swap_file->f_mapping;
 293                 struct bio_vec bv = {
 294                         .bv_page = page,
 295                         .bv_len  = PAGE_SIZE,
 296                         .bv_offset = 0
 297                 };
 298                 struct iov_iter from;
 299
 300                 iov_iter_bvec(&from, WRITE, &bv, 1, PAGE_SIZE);
 301                 init_sync_kiocb(&kiocb, swap_file);
 302                 kiocb.ki_pos = page_file_offset(page);
 303
 304                 set_page_writeback(page);
 305                 unlock_page(page);
 306                 ret = mapping->a_ops->direct_IO(&kiocb, &from);
 307                 if (ret == PAGE_SIZE) {
 308                         count_vm_event(PSWPOUT);
 309                         ret = 0;
 310                 } else {
 311                         /*
 312                          * In the case of swap-over-nfs, this can be a
 313                          * temporary failure if the system has limited
 314                          * memory for allocating transmit buffers.
 315                          * Mark the page dirty and avoid
 316                          * rotate_reclaimable_page but rate-limit the
 317                          * messages but do not flag PageError like
 318                          * the normal direct-to-bio case as it could
 319                          * be temporary.
 320                          */
 321                         set_page_dirty(page);
 322                         ClearPageReclaim(page);
 323                         pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
 324                                            page_file_offset(page));
 325                 }
 326                 end_page_writeback(page);
 327                 return ret;
 328         }
 329
 330         ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
 331         if (!ret) {
 332                 count_swpout_vm_event(page);
 333                 return 0;
 334         }
 335
 336         ret = 0;
 337         bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 338         if (bio == NULL) {
 339                 set_page_dirty(page);
 340                 unlock_page(page);
 341                 ret = -ENOMEM;
 342                 goto out;
 343         }
 344         bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
 345         bio_associate_blkg_from_page(bio, page);
 346         count_swpout_vm_event(page);
 347         set_page_writeback(page);
 348         unlock_page(page);
 349         submit_bio(bio);
 350 out:
 351         return ret;
 352 }
 353
 354 int swap_readpage(struct page *page, bool synchronous)
 355 {
 356         struct bio *bio;
 357         int ret = 0;
 358         struct swap_info_struct *sis = page_swap_info(page);
 359         blk_qc_t qc;
 360         struct gendisk *disk;
 361         unsigned long pflags;
 362
 363         VM_BUG_ON_PAGE(!PageSwapCache(page) && !synchronous, page);
 364         VM_BUG_ON_PAGE(!PageLocked(page), page);
 365         VM_BUG_ON_PAGE(PageUptodate(page), page);
 366
 367         /*
 368          * Count submission time as memory stall. When the device is congested,
 369          * or the submitting cgroup IO-throttled, submission can be a
 370          * significant part of overall IO time.
 371          */
 372         psi_memstall_enter(&pflags);
 373
 374         if (frontswap_load(page) == 0) {
 375                 SetPageUptodate(page);
 376                 unlock_page(page);
 377                 goto out;
 378         }
 379
 380         if (sis->flags & SWP_FS) {
 381                 struct file *swap_file = sis->swap_file;
 382                 struct address_space *mapping = swap_file->f_mapping;
 383
 384                 ret = mapping->a_ops->readpage(swap_file, page);
 385                 if (!ret)
 386                         count_vm_event(PSWPIN);
 387                 goto out;
 388         }
 389
 390         ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
 391         if (!ret) {
 392                 if (trylock_page(page)) {
 393                         swap_slot_free_notify(page);
 394                         unlock_page(page);
 395                 }
 396
 397                 count_vm_event(PSWPIN);
 398                 goto out;
 399         }
 400
 401         ret = 0;
 402         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 403         if (bio == NULL) {
 404                 unlock_page(page);
 405                 ret = -ENOMEM;
 406                 goto out;
 407         }
 408         disk = bio->bi_disk;
 409         /*
 410          * Keep this task valid during swap readpage because the oom killer may
 411          * attempt to access it in the page fault retry time check.
 412          */
 413         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 414         if (synchronous) {
 415                 bio->bi_opf |= REQ_HIPRI;
 416                 get_task_struct(current);
 417                 bio->bi_private = current;
 418         }
 419         count_vm_event(PSWPIN);
 420         bio_get(bio);
 421         qc = submit_bio(bio);
 422         while (synchronous) {
 423                 set_current_state(TASK_UNINTERRUPTIBLE);
 424                 if (!READ_ONCE(bio->bi_private))
 425                         break;
 426
 427                 if (!blk_poll(disk->queue, qc, true))
 428                         io_schedule();
 429         }
 430         __set_current_state(TASK_RUNNING);
 431         bio_put(bio);
 432
 433 out:
 434         psi_memstall_leave(&pflags);
 435         return ret;
 436 }
 437
 438 int swap_set_page_dirty(struct page *page)
 439 {
 440         struct swap_info_struct *sis = page_swap_info(page);
 441
 442         if (sis->flags & SWP_FS) {
 443                 struct address_space *mapping = sis->swap_file->f_mapping;
 444
 445                 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 446                 return mapping->a_ops->set_page_dirty(page);
 447         } else {
 448                 return __set_page_dirty_no_writeback(page);
 449         }
 450 }