mm/swap_state.c

   1 /*
   2  *  linux/mm/swap_state.c
   3  *
   4  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   5  *  Swap reorganised 29.12.95, Stephen Tweedie
   6  *
   7  *  Rewritten to use page cache, (C) 1998 Stephen Tweedie
   8  */
   9 #include <linux/mm.h>
  10 #include <linux/gfp.h>
  11 #include <linux/kernel_stat.h>
  12 #include <linux/swap.h>
  13 #include <linux/swapops.h>
  14 #include <linux/init.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/pagevec.h>
  19 #include <linux/migrate.h>
  20 #include <linux/page_cgroup.h>
  21
  22 #include <asm/pgtable.h>
  23
  24 /*
  25  * swapper_space is a fiction, retained to simplify the path through
  26  * vmscan's shrink_page_list.
  27  */
  28 static const struct address_space_operations swap_aops = {
  29         .writepage      = swap_writepage,
  30         .set_page_dirty = swap_set_page_dirty,
  31         .migratepage    = migrate_page,
  32 };
  33
  34 static struct backing_dev_info swap_backing_dev_info = {
  35         .name           = "swap",
  36         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
  37 };
  38
  39 struct address_space swapper_spaces[MAX_SWAPFILES] = {
  40         [0 ... MAX_SWAPFILES - 1] = {
  41                 .page_tree      = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
  42                 .a_ops          = &swap_aops,
  43                 .backing_dev_info = &swap_backing_dev_info,
  44         }
  45 };
  46
  47 #define INC_CACHE_INFO(x)       do { swap_cache_info.x++; } while (0)
  48
  49 static struct {
  50         unsigned long add_total;
  51         unsigned long del_total;
  52         unsigned long find_success;
  53         unsigned long find_total;
  54 } swap_cache_info;
  55
  56 unsigned long total_swapcache_pages(void)
  57 {
  58         int i;
  59         unsigned long ret = 0;
  60
  61         for (i = 0; i < MAX_SWAPFILES; i++)
  62                 ret += swapper_spaces[i].nrpages;
  63         return ret;
  64 }
  65
  66 void show_swap_cache_info(void)
  67 {
  68         printk("%lu pages in swap cache\n", total_swapcache_pages());
  69         printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
  70                 swap_cache_info.add_total, swap_cache_info.del_total,
  71                 swap_cache_info.find_success, swap_cache_info.find_total);
  72         printk("Free swap  = %ldkB\n",
  73                 get_nr_swap_pages() << (PAGE_SHIFT - 10));
  74         printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
  75 }
  76
  77 /*
  78  * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
  79  * but sets SwapCache flag and private instead of mapping and index.
  80  */
  81 int __add_to_swap_cache(struct page *page, swp_entry_t entry)
  82 {
  83         int error;
  84         struct address_space *address_space;
  85
  86         VM_BUG_ON(!PageLocked(page));
  87         VM_BUG_ON(PageSwapCache(page));
  88         VM_BUG_ON(!PageSwapBacked(page));
  89
  90         page_cache_get(page);
  91         SetPageSwapCache(page);
  92         set_page_private(page, entry.val);
  93
  94         address_space = swap_address_space(entry);
  95         spin_lock_irq(&address_space->tree_lock);
  96         error = radix_tree_insert(&address_space->page_tree,
  97                                         entry.val, page);
  98         if (likely(!error)) {
  99                 address_space->nrpages++;
 100                 __inc_zone_page_state(page, NR_FILE_PAGES);
 101                 INC_CACHE_INFO(add_total);
 102         }
 103         spin_unlock_irq(&address_space->tree_lock);
 104
 105         if (unlikely(error)) {
 106                 /*
 107                  * Only the context which have set SWAP_HAS_CACHE flag
 108                  * would call add_to_swap_cache().
 109                  * So add_to_swap_cache() doesn't returns -EEXIST.
 110                  */
 111                 VM_BUG_ON(error == -EEXIST);
 112                 set_page_private(page, 0UL);
 113                 ClearPageSwapCache(page);
 114                 page_cache_release(page);
 115         }
 116
 117         return error;
 118 }
 119
 120
 121 int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
 122 {
 123         int error;
 124
 125         error = radix_tree_preload(gfp_mask);
 126         if (!error) {
 127                 error = __add_to_swap_cache(page, entry);
 128                 radix_tree_preload_end();
 129         }
 130         return error;
 131 }
 132
 133 /*
 134  * This must be called only on pages that have
 135  * been verified to be in the swap cache.
 136  */
 137 void __delete_from_swap_cache(struct page *page)
 138 {
 139         swp_entry_t entry;
 140         struct address_space *address_space;
 141
 142         VM_BUG_ON(!PageLocked(page));
 143         VM_BUG_ON(!PageSwapCache(page));
 144         VM_BUG_ON(PageWriteback(page));
 145
 146         entry.val = page_private(page);
 147         address_space = swap_address_space(entry);
 148         radix_tree_delete(&address_space->page_tree, page_private(page));
 149         set_page_private(page, 0);
 150         ClearPageSwapCache(page);
 151         address_space->nrpages--;
 152         __dec_zone_page_state(page, NR_FILE_PAGES);
 153         INC_CACHE_INFO(del_total);
 154 }
 155
 156 /**
 157  * add_to_swap - allocate swap space for a page
 158  * @page: page we want to move to swap
 159  *
 160  * Allocate swap space for the page and add the page to the
 161  * swap cache.  Caller needs to hold the page lock.
 162  */
 163 int add_to_swap(struct page *page, struct list_head *list)
 164 {
 165         swp_entry_t entry;
 166         int err;
 167
 168         VM_BUG_ON(!PageLocked(page));
 169         VM_BUG_ON(!PageUptodate(page));
 170
 171 #ifndef CONFIG_MEMCG
 172         entry = get_swap_page();
 173 #else
 174         entry = get_swap_page_by_memcg(page);
 175 #endif
 176         if (!entry.val)
 177                 return 0;
 178
 179         if (unlikely(PageTransHuge(page)))
 180                 if (unlikely(split_huge_page_to_list(page, list))) {
 181                         swapcache_free(entry, NULL);
 182                         return 0;
 183                 }
 184
 185         /*
 186          * Radix-tree node allocations from PF_MEMALLOC contexts could
 187          * completely exhaust the page allocator. __GFP_NOMEMALLOC
 188          * stops emergency reserves from being allocated.
 189          *
 190          * TODO: this could cause a theoretical memory reclaim
 191          * deadlock in the swap out path.
 192          */
 193         /*
 194          * Add it to the swap cache and mark it dirty
 195          */
 196         err = add_to_swap_cache(page, entry,
 197                         __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
 198
 199         if (!err) {     /* Success */
 200                 SetPageDirty(page);
 201                 return 1;
 202         } else {        /* -ENOMEM radix-tree allocation failure */
 203                 /*
 204                  * add_to_swap_cache() doesn't return -EEXIST, so we can safely
 205                  * clear SWAP_HAS_CACHE flag.
 206                  */
 207                 swapcache_free(entry, NULL);
 208                 return 0;
 209         }
 210 }
 211
 212 /*
 213  * This must be called only on pages that have
 214  * been verified to be in the swap cache and locked.
 215  * It will never put the page into the free list,
 216  * the caller has a reference on the page.
 217  */
 218 void delete_from_swap_cache(struct page *page)
 219 {
 220         swp_entry_t entry;
 221         struct address_space *address_space;
 222
 223         entry.val = page_private(page);
 224
 225         address_space = swap_address_space(entry);
 226         spin_lock_irq(&address_space->tree_lock);
 227         __delete_from_swap_cache(page);
 228         spin_unlock_irq(&address_space->tree_lock);
 229
 230         swapcache_free(entry, page);
 231         page_cache_release(page);
 232 }
 233
 234 /*
 235  * If we are the only user, then try to free up the swap cache.
 236  *
 237  * Its ok to check for PageSwapCache without the page lock
 238  * here because we are going to recheck again inside
 239  * try_to_free_swap() _with_ the lock.
 240  *                                      - Marcelo
 241  */
 242 static inline void free_swap_cache(struct page *page)
 243 {
 244         if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
 245                 try_to_free_swap(page);
 246                 unlock_page(page);
 247         }
 248 }
 249
 250 /*
 251  * Perform a free_page(), also freeing any swap cache associated with
 252  * this page if it is the last user of the page.
 253  */
 254 void free_page_and_swap_cache(struct page *page)
 255 {
 256         free_swap_cache(page);
 257         page_cache_release(page);
 258 }
 259
 260 /*
 261  * Passed an array of pages, drop them all from swapcache and then release
 262  * them.  They are removed from the LRU and freed if this is their last use.
 263  */
 264 void free_pages_and_swap_cache(struct page **pages, int nr)
 265 {
 266         struct page **pagep = pages;
 267
 268         lru_add_drain();
 269         while (nr) {
 270                 int todo = min(nr, PAGEVEC_SIZE);
 271                 int i;
 272
 273                 for (i = 0; i < todo; i++)
 274                         free_swap_cache(pagep[i]);
 275                 release_pages(pagep, todo, 0);
 276                 pagep += todo;
 277                 nr -= todo;
 278         }
 279 }
 280
 281 /*
 282  * Lookup a swap entry in the swap cache. A found page will be returned
 283  * unlocked and with its refcount incremented - we rely on the kernel
 284  * lock getting page table operations atomic even if we drop the page
 285  * lock before returning.
 286  */
 287 struct page * lookup_swap_cache(swp_entry_t entry)
 288 {
 289         struct page *page;
 290
 291         page = find_get_page(swap_address_space(entry), entry.val);
 292
 293         if (page)
 294                 INC_CACHE_INFO(find_success);
 295
 296         INC_CACHE_INFO(find_total);
 297         return page;
 298 }
 299
 300 /*
 301  * Locate a page of swap in physical memory, reserving swap cache space
 302  * and reading the disk if it is not already cached.
 303  * A failure return means that either the page allocation failed or that
 304  * the swap entry is no longer in use.
 305  */
 306 struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
 307                         struct vm_area_struct *vma, unsigned long addr)
 308 {
 309         struct page *found_page, *new_page = NULL;
 310         int err;
 311
 312         do {
 313                 /*
 314                  * First check the swap cache.  Since this is normally
 315                  * called after lookup_swap_cache() failed, re-calling
 316                  * that would confuse statistics.
 317                  */
 318                 found_page = find_get_page(swap_address_space(entry),
 319                                         entry.val);
 320                 if (found_page)
 321                         break;
 322
 323                 /*
 324                  * Get a new page to read into from swap.
 325                  */
 326                 if (!new_page) {
 327                         new_page = alloc_page_vma(gfp_mask, vma, addr);
 328                         if (!new_page)
 329                                 break;          /* Out of memory */
 330                 }
 331
 332                 /*
 333                  * call radix_tree_preload() while we can wait.
 334                  */
 335                 err = radix_tree_preload(gfp_mask & GFP_KERNEL);
 336                 if (err)
 337                         break;
 338
 339                 /*
 340                  * Swap entry may have been freed since our caller observed it.
 341                  */
 342                 err = swapcache_prepare(entry);
 343                 if (err == -EEXIST) {
 344                         radix_tree_preload_end();
 345                         /*
 346                          * We might race against get_swap_page() and stumble
 347                          * across a SWAP_HAS_CACHE swap_map entry whose page
 348                          * has not been brought into the swapcache yet, while
 349                          * the other end is scheduled away waiting on discard
 350                          * I/O completion at scan_swap_map().
 351                          *
 352                          * In order to avoid turning this transitory state
 353                          * into a permanent loop around this -EEXIST case
 354                          * if !CONFIG_PREEMPT and the I/O completion happens
 355                          * to be waiting on the CPU waitqueue where we are now
 356                          * busy looping, we just conditionally invoke the
 357                          * scheduler here, if there are some more important
 358                          * tasks to run.
 359                          */
 360                         cond_resched();
 361                         continue;
 362                 }
 363                 if (err) {              /* swp entry is obsolete ? */
 364                         radix_tree_preload_end();
 365                         break;
 366                 }
 367
 368                 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
 369                 __set_page_locked(new_page);
 370                 SetPageSwapBacked(new_page);
 371                 err = __add_to_swap_cache(new_page, entry);
 372                 if (likely(!err)) {
 373                         radix_tree_preload_end();
 374                         /*
 375                          * Initiate read into locked page and return.
 376                          */
 377                         lru_cache_add_anon(new_page);
 378                         swap_readpage(new_page);
 379                         return new_page;
 380                 }
 381                 radix_tree_preload_end();
 382                 ClearPageSwapBacked(new_page);
 383                 __clear_page_locked(new_page);
 384                 /*
 385                  * add_to_swap_cache() doesn't return -EEXIST, so we can safely
 386                  * clear SWAP_HAS_CACHE flag.
 387                  */
 388                 swapcache_free(entry, NULL);
 389         } while (err != -ENOMEM);
 390
 391         if (new_page)
 392                 page_cache_release(new_page);
 393         return found_page;
 394 }
 395
 396 /**
 397  * swapin_readahead - swap in pages in hope we need them soon
 398  * @entry: swap entry of this memory
 399  * @gfp_mask: memory allocation flags
 400  * @vma: user vma this address belongs to
 401  * @addr: target address for mempolicy
 402  *
 403  * Returns the struct page for entry and addr, after queueing swapin.
 404  *
 405  * Primitive swap readahead code. We simply read an aligned block of
 406  * (1 << page_cluster) entries in the swap area. This method is chosen
 407  * because it doesn't cost us any seek time.  We also make sure to queue
 408  * the 'original' request together with the readahead ones...
 409  *
 410  * This has been extended to use the NUMA policies from the mm triggering
 411  * the readahead.
 412  *
 413  * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
 414  */
 415 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
 416                         struct vm_area_struct *vma, unsigned long addr)
 417 {
 418         struct page *page;
 419         unsigned long offset = swp_offset(entry);
 420         unsigned long start_offset, end_offset;
 421         unsigned long mask = (1UL << page_cluster) - 1;
 422         struct blk_plug plug;
 423
 424         /* Read a page_cluster sized and aligned cluster around offset. */
 425         start_offset = offset & ~mask;
 426         end_offset = offset | mask;
 427         if (!start_offset)      /* First page is swap header. */
 428                 start_offset++;
 429
 430         blk_start_plug(&plug);
 431         for (offset = start_offset; offset <= end_offset ; offset++) {
 432                 /* Ok, do the async read-ahead now */
 433                 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
 434                                                 gfp_mask, vma, addr);
 435                 if (!page)
 436                         continue;
 437                 page_cache_release(page);
 438         }
 439         blk_finish_plug(&plug);
 440
 441         lru_add_drain();        /* Push any new pages onto the LRU now */
 442         return read_swap_cache_async(entry, gfp_mask, vma, addr);
 443 }