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
  80         /*
  81          * There is no guarantee that the page is in swap cache - the software
  82          * suspend code (at least) uses end_swap_bio_read() against a non-
  83          * swapcache page.  So we must check PG_swapcache before proceeding with
  84          * this optimization.
  85          */
  86         if (unlikely(!PageSwapCache(page)))
  87                 return;
  88
  89         sis = page_swap_info(page);
  90         if (!(sis->flags & SWP_BLKDEV))
  91                 return;
  92
  93         /*
  94          * The swap subsystem performs lazy swap slot freeing,
  95          * expecting that the page will be swapped out again.
  96          * So we can avoid an unnecessary write if the page
  97          * isn't redirtied.
  98          * This is good for real swap storage because we can
  99          * reduce unnecessary I/O and enhance wear-leveling
 100          * if an SSD is used as the as swap device.
 101          * But if in-memory swap device (eg zram) is used,
 102          * this causes a duplicated copy between uncompressed
 103          * data in VM-owned memory and compressed data in
 104          * zram-owned memory.  So let's free zram-owned memory
 105          * and make the VM-owned decompressed page *dirty*,
 106          * so the page should be swapped out somewhere again if
 107          * we again wish to reclaim it.
 108          */
 109         disk = sis->bdev->bd_disk;
 110         if (disk->fops->swap_slot_free_notify) {
 111                 swp_entry_t entry;
 112                 unsigned long offset;
 113
 114                 entry.val = page_private(page);
 115                 offset = swp_offset(entry);
 116
 117                 SetPageDirty(page);
 118                 disk->fops->swap_slot_free_notify(sis->bdev,
 119                                 offset);
 120         }
 121 }
 122
 123 static void end_swap_bio_read(struct bio *bio)
 124 {
 125         struct page *page = bio->bi_io_vec[0].bv_page;
 126         struct task_struct *waiter = bio->bi_private;
 127
 128         if (bio->bi_status) {
 129                 SetPageError(page);
 130                 ClearPageUptodate(page);
 131                 pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
 132                          MAJOR(bio_dev(bio)), MINOR(bio_dev(bio)),
 133                          (unsigned long long)bio->bi_iter.bi_sector);
 134                 goto out;
 135         }
 136
 137         SetPageUptodate(page);
 138         swap_slot_free_notify(page);
 139 out:
 140         unlock_page(page);
 141         WRITE_ONCE(bio->bi_private, NULL);
 142         bio_put(bio);
 143         wake_up_process(waiter);
 144         put_task_struct(waiter);
 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 list.  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_FILE) {
 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, ITER_BVEC | 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 | wbc_to_write_flags(wbc);
 342         count_swpout_vm_event(page);
 343         set_page_writeback(page);
 344         unlock_page(page);
 345         submit_bio(bio);
 346 out:
 347         return ret;
 348 }
 349
 350 int swap_readpage(struct page *page, bool do_poll)
 351 {
 352         struct bio *bio;
 353         int ret = 0;
 354         struct swap_info_struct *sis = page_swap_info(page);
 355         blk_qc_t qc;
 356         struct gendisk *disk;
 357
 358         VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 359         VM_BUG_ON_PAGE(!PageLocked(page), page);
 360         VM_BUG_ON_PAGE(PageUptodate(page), page);
 361         if (frontswap_load(page) == 0) {
 362                 SetPageUptodate(page);
 363                 unlock_page(page);
 364                 goto out;
 365         }
 366
 367         if (sis->flags & SWP_FILE) {
 368                 struct file *swap_file = sis->swap_file;
 369                 struct address_space *mapping = swap_file->f_mapping;
 370
 371                 ret = mapping->a_ops->readpage(swap_file, page);
 372                 if (!ret)
 373                         count_vm_event(PSWPIN);
 374                 return ret;
 375         }
 376
 377         ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
 378         if (!ret) {
 379                 if (trylock_page(page)) {
 380                         swap_slot_free_notify(page);
 381                         unlock_page(page);
 382                 }
 383
 384                 count_vm_event(PSWPIN);
 385                 return 0;
 386         }
 387
 388         ret = 0;
 389         bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 390         if (bio == NULL) {
 391                 unlock_page(page);
 392                 ret = -ENOMEM;
 393                 goto out;
 394         }
 395         disk = bio->bi_disk;
 396         /*
 397          * Keep this task valid during swap readpage because the oom killer may
 398          * attempt to access it in the page fault retry time check.
 399          */
 400         get_task_struct(current);
 401         bio->bi_private = current;
 402         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 403         count_vm_event(PSWPIN);
 404         bio_get(bio);
 405         qc = submit_bio(bio);
 406         while (do_poll) {
 407                 set_current_state(TASK_UNINTERRUPTIBLE);
 408                 if (!READ_ONCE(bio->bi_private))
 409                         break;
 410
 411                 if (!blk_mq_poll(disk->queue, qc))
 412                         break;
 413         }
 414         __set_current_state(TASK_RUNNING);
 415         bio_put(bio);
 416
 417 out:
 418         return ret;
 419 }
 420
 421 int swap_set_page_dirty(struct page *page)
 422 {
 423         struct swap_info_struct *sis = page_swap_info(page);
 424
 425         if (sis->flags & SWP_FILE) {
 426                 struct address_space *mapping = sis->swap_file->f_mapping;
 427
 428                 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
 429                 return mapping->a_ops->set_page_dirty(page);
 430         } else {
 431                 return __set_page_dirty_no_writeback(page);
 432         }
 433 }