linux/mm/page_io.c
<<
>>
Prefs
   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
  27static 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
  44void end_swap_bio_write(struct bio *bio)
  45{
  46        struct page *page = bio->bi_io_vec[0].bv_page;
  47
  48        if (bio->bi_error) {
  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
  69static 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
 117static void end_swap_bio_read(struct bio *bio)
 118{
 119        struct page *page = bio->bi_io_vec[0].bv_page;
 120
 121        if (bio->bi_error) {
 122                SetPageError(page);
 123                ClearPageUptodate(page);
 124                pr_alert("Read-error on swap-device (%u:%u:%llu)\n",
 125                         imajor(bio->bi_bdev->bd_inode),
 126                         iminor(bio->bi_bdev->bd_inode),
 127                         (unsigned long long)bio->bi_iter.bi_sector);
 128                goto out;
 129        }
 130
 131        SetPageUptodate(page);
 132        swap_slot_free_notify(page);
 133out:
 134        unlock_page(page);
 135        bio_put(bio);
 136}
 137
 138int generic_swapfile_activate(struct swap_info_struct *sis,
 139                                struct file *swap_file,
 140                                sector_t *span)
 141{
 142        struct address_space *mapping = swap_file->f_mapping;
 143        struct inode *inode = mapping->host;
 144        unsigned blocks_per_page;
 145        unsigned long page_no;
 146        unsigned blkbits;
 147        sector_t probe_block;
 148        sector_t last_block;
 149        sector_t lowest_block = -1;
 150        sector_t highest_block = 0;
 151        int nr_extents = 0;
 152        int ret;
 153
 154        blkbits = inode->i_blkbits;
 155        blocks_per_page = PAGE_SIZE >> blkbits;
 156
 157        /*
 158         * Map all the blocks into the extent list.  This code doesn't try
 159         * to be very smart.
 160         */
 161        probe_block = 0;
 162        page_no = 0;
 163        last_block = i_size_read(inode) >> blkbits;
 164        while ((probe_block + blocks_per_page) <= last_block &&
 165                        page_no < sis->max) {
 166                unsigned block_in_page;
 167                sector_t first_block;
 168
 169                first_block = bmap(inode, probe_block);
 170                if (first_block == 0)
 171                        goto bad_bmap;
 172
 173                /*
 174                 * It must be PAGE_SIZE aligned on-disk
 175                 */
 176                if (first_block & (blocks_per_page - 1)) {
 177                        probe_block++;
 178                        goto reprobe;
 179                }
 180
 181                for (block_in_page = 1; block_in_page < blocks_per_page;
 182                                        block_in_page++) {
 183                        sector_t block;
 184
 185                        block = bmap(inode, probe_block + block_in_page);
 186                        if (block == 0)
 187                                goto bad_bmap;
 188                        if (block != first_block + block_in_page) {
 189                                /* Discontiguity */
 190                                probe_block++;
 191                                goto reprobe;
 192                        }
 193                }
 194
 195                first_block >>= (PAGE_SHIFT - blkbits);
 196                if (page_no) {  /* exclude the header page */
 197                        if (first_block < lowest_block)
 198                                lowest_block = first_block;
 199                        if (first_block > highest_block)
 200                                highest_block = first_block;
 201                }
 202
 203                /*
 204                 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
 205                 */
 206                ret = add_swap_extent(sis, page_no, 1, first_block);
 207                if (ret < 0)
 208                        goto out;
 209                nr_extents += ret;
 210                page_no++;
 211                probe_block += blocks_per_page;
 212reprobe:
 213                continue;
 214        }
 215        ret = nr_extents;
 216        *span = 1 + highest_block - lowest_block;
 217        if (page_no == 0)
 218                page_no = 1;    /* force Empty message */
 219        sis->max = page_no;
 220        sis->pages = page_no - 1;
 221        sis->highest_bit = page_no - 1;
 222out:
 223        return ret;
 224bad_bmap:
 225        pr_err("swapon: swapfile has holes\n");
 226        ret = -EINVAL;
 227        goto out;
 228}
 229
 230/*
 231 * We may have stale swap cache pages in memory: notice
 232 * them here and get rid of the unnecessary final write.
 233 */
 234int swap_writepage(struct page *page, struct writeback_control *wbc)
 235{
 236        int ret = 0;
 237
 238        if (try_to_free_swap(page)) {
 239                unlock_page(page);
 240                goto out;
 241        }
 242        if (frontswap_store(page) == 0) {
 243                set_page_writeback(page);
 244                unlock_page(page);
 245                end_page_writeback(page);
 246                goto out;
 247        }
 248        ret = __swap_writepage(page, wbc, end_swap_bio_write);
 249out:
 250        return ret;
 251}
 252
 253static sector_t swap_page_sector(struct page *page)
 254{
 255        return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
 256}
 257
 258int __swap_writepage(struct page *page, struct writeback_control *wbc,
 259                bio_end_io_t end_write_func)
 260{
 261        struct bio *bio;
 262        int ret, rw = WRITE;
 263        struct swap_info_struct *sis = page_swap_info(page);
 264
 265        if (sis->flags & SWP_FILE) {
 266                struct kiocb kiocb;
 267                struct file *swap_file = sis->swap_file;
 268                struct address_space *mapping = swap_file->f_mapping;
 269                struct bio_vec bv = {
 270                        .bv_page = page,
 271                        .bv_len  = PAGE_SIZE,
 272                        .bv_offset = 0
 273                };
 274                struct iov_iter from;
 275
 276                iov_iter_bvec(&from, ITER_BVEC | WRITE, &bv, 1, PAGE_SIZE);
 277                init_sync_kiocb(&kiocb, swap_file);
 278                kiocb.ki_pos = page_file_offset(page);
 279
 280                set_page_writeback(page);
 281                unlock_page(page);
 282                ret = mapping->a_ops->direct_IO(&kiocb, &from);
 283                if (ret == PAGE_SIZE) {
 284                        count_vm_event(PSWPOUT);
 285                        ret = 0;
 286                } else {
 287                        /*
 288                         * In the case of swap-over-nfs, this can be a
 289                         * temporary failure if the system has limited
 290                         * memory for allocating transmit buffers.
 291                         * Mark the page dirty and avoid
 292                         * rotate_reclaimable_page but rate-limit the
 293                         * messages but do not flag PageError like
 294                         * the normal direct-to-bio case as it could
 295                         * be temporary.
 296                         */
 297                        set_page_dirty(page);
 298                        ClearPageReclaim(page);
 299                        pr_err_ratelimited("Write error on dio swapfile (%llu)\n",
 300                                           page_file_offset(page));
 301                }
 302                end_page_writeback(page);
 303                return ret;
 304        }
 305
 306        ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
 307        if (!ret) {
 308                count_vm_event(PSWPOUT);
 309                return 0;
 310        }
 311
 312        ret = 0;
 313        bio = get_swap_bio(GFP_NOIO, page, end_write_func);
 314        if (bio == NULL) {
 315                set_page_dirty(page);
 316                unlock_page(page);
 317                ret = -ENOMEM;
 318                goto out;
 319        }
 320        if (wbc->sync_mode == WB_SYNC_ALL)
 321                rw |= REQ_SYNC;
 322        count_vm_event(PSWPOUT);
 323        set_page_writeback(page);
 324        unlock_page(page);
 325        submit_bio(rw, bio);
 326out:
 327        return ret;
 328}
 329
 330int swap_readpage(struct page *page)
 331{
 332        struct bio *bio;
 333        int ret = 0;
 334        struct swap_info_struct *sis = page_swap_info(page);
 335
 336        VM_BUG_ON_PAGE(!PageLocked(page), page);
 337        VM_BUG_ON_PAGE(PageUptodate(page), page);
 338        if (frontswap_load(page) == 0) {
 339                SetPageUptodate(page);
 340                unlock_page(page);
 341                goto out;
 342        }
 343
 344        if (sis->flags & SWP_FILE) {
 345                struct file *swap_file = sis->swap_file;
 346                struct address_space *mapping = swap_file->f_mapping;
 347
 348                ret = mapping->a_ops->readpage(swap_file, page);
 349                if (!ret)
 350                        count_vm_event(PSWPIN);
 351                return ret;
 352        }
 353
 354        ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
 355        if (!ret) {
 356                if (trylock_page(page)) {
 357                        swap_slot_free_notify(page);
 358                        unlock_page(page);
 359                }
 360
 361                count_vm_event(PSWPIN);
 362                return 0;
 363        }
 364
 365        ret = 0;
 366        bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
 367        if (bio == NULL) {
 368                unlock_page(page);
 369                ret = -ENOMEM;
 370                goto out;
 371        }
 372        count_vm_event(PSWPIN);
 373        submit_bio(READ, bio);
 374out:
 375        return ret;
 376}
 377
 378int swap_set_page_dirty(struct page *page)
 379{
 380        struct swap_info_struct *sis = page_swap_info(page);
 381
 382        if (sis->flags & SWP_FILE) {
 383                struct address_space *mapping = sis->swap_file->f_mapping;
 384                return mapping->a_ops->set_page_dirty(page);
 385        } else {
 386                return __set_page_dirty_no_writeback(page);
 387        }
 388}
 389