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