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/module.h>
  10 #include <linux/mm.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/buffer_head.h>
  17 #include <linux/backing-dev.h>
  18 #include <linux/pagevec.h>
  19 #include <linux/migrate.h>
  20
  21 #include <asm/pgtable.h>
  22
  23 /*
  24  * swapper_space is a fiction, retained to simplify the path through
  25  * vmscan's shrink_page_list, to make sync_page look nicer, and to allow
  26  * future use of radix_tree tags in the swap cache.
  27  */
  28 static const struct address_space_operations swap_aops = {
  29         .writepage      = swap_writepage,
  30         .sync_page      = block_sync_page,
  31         .set_page_dirty = __set_page_dirty_nobuffers,
  32         .migratepage    = migrate_page,
  33 };
  34
  35 static struct backing_dev_info swap_backing_dev_info = {
  36         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  37         .unplug_io_fn   = swap_unplug_io_fn,
  38 };
  39
  40 struct address_space swapper_space = {
  41         .page_tree      = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
  42         .tree_lock      = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
  43         .a_ops          = &swap_aops,
  44         .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
  45         .backing_dev_info = &swap_backing_dev_info,
  46 };
  47
  48 #define INC_CACHE_INFO(x)       do { swap_cache_info.x++; } while (0)
  49
  50 static struct {
  51         unsigned long add_total;
  52         unsigned long del_total;
  53         unsigned long find_success;
  54         unsigned long find_total;
  55 } swap_cache_info;
  56
  57 void show_swap_cache_info(void)
  58 {
  59         printk("Swap cache: add %lu, delete %lu, find %lu/%lu\n",
  60                 swap_cache_info.add_total, swap_cache_info.del_total,
  61                 swap_cache_info.find_success, swap_cache_info.find_total);
  62         printk("Free swap  = %lukB\n", nr_swap_pages << (PAGE_SHIFT - 10));
  63         printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
  64 }
  65
  66 /*
  67  * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
  68  * but sets SwapCache flag and private instead of mapping and index.
  69  */
  70 int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
  71 {
  72         int error;
  73
  74         BUG_ON(!PageLocked(page));
  75         BUG_ON(PageSwapCache(page));
  76         BUG_ON(PagePrivate(page));
  77         error = radix_tree_preload(gfp_mask);
  78         if (!error) {
  79                 page_cache_get(page);
  80                 SetPageSwapCache(page);
  81                 set_page_private(page, entry.val);
  82
  83                 spin_lock_irq(&swapper_space.tree_lock);
  84                 error = radix_tree_insert(&swapper_space.page_tree,
  85                                                 entry.val, page);
  86                 if (likely(!error)) {
  87                         total_swapcache_pages++;
  88                         __inc_zone_page_state(page, NR_FILE_PAGES);
  89                         INC_CACHE_INFO(add_total);
  90                 }
  91                 spin_unlock_irq(&swapper_space.tree_lock);
  92                 radix_tree_preload_end();
  93
  94                 if (unlikely(error)) {
  95                         set_page_private(page, 0UL);
  96                         ClearPageSwapCache(page);
  97                         page_cache_release(page);
  98                 }
  99         }
 100         return error;
 101 }
 102
 103 /*
 104  * This must be called only on pages that have
 105  * been verified to be in the swap cache.
 106  */
 107 void __delete_from_swap_cache(struct page *page)
 108 {
 109         BUG_ON(!PageLocked(page));
 110         BUG_ON(!PageSwapCache(page));
 111         BUG_ON(PageWriteback(page));
 112         BUG_ON(PagePrivate(page));
 113
 114         radix_tree_delete(&swapper_space.page_tree, page_private(page));
 115         set_page_private(page, 0);
 116         ClearPageSwapCache(page);
 117         total_swapcache_pages--;
 118         __dec_zone_page_state(page, NR_FILE_PAGES);
 119         INC_CACHE_INFO(del_total);
 120 }
 121
 122 /**
 123  * add_to_swap - allocate swap space for a page
 124  * @page: page we want to move to swap
 125  * @gfp_mask: memory allocation flags
 126  *
 127  * Allocate swap space for the page and add the page to the
 128  * swap cache.  Caller needs to hold the page lock.
 129  */
 130 int add_to_swap(struct page * page, gfp_t gfp_mask)
 131 {
 132         swp_entry_t entry;
 133         int err;
 134
 135         BUG_ON(!PageLocked(page));
 136         BUG_ON(!PageUptodate(page));
 137
 138         for (;;) {
 139                 entry = get_swap_page();
 140                 if (!entry.val)
 141                         return 0;
 142
 143                 /*
 144                  * Radix-tree node allocations from PF_MEMALLOC contexts could
 145                  * completely exhaust the page allocator. __GFP_NOMEMALLOC
 146                  * stops emergency reserves from being allocated.
 147                  *
 148                  * TODO: this could cause a theoretical memory reclaim
 149                  * deadlock in the swap out path.
 150                  */
 151                 /*
 152                  * Add it to the swap cache and mark it dirty
 153                  */
 154                 err = add_to_swap_cache(page, entry,
 155                                 gfp_mask|__GFP_NOMEMALLOC|__GFP_NOWARN);
 156
 157                 switch (err) {
 158                 case 0:                         /* Success */
 159                         SetPageDirty(page);
 160                         return 1;
 161                 case -EEXIST:
 162                         /* Raced with "speculative" read_swap_cache_async */
 163                         swap_free(entry);
 164                         continue;
 165                 default:
 166                         /* -ENOMEM radix-tree allocation failure */
 167                         swap_free(entry);
 168                         return 0;
 169                 }
 170         }
 171 }
 172
 173 /*
 174  * This must be called only on pages that have
 175  * been verified to be in the swap cache and locked.
 176  * It will never put the page into the free list,
 177  * the caller has a reference on the page.
 178  */
 179 void delete_from_swap_cache(struct page *page)
 180 {
 181         swp_entry_t entry;
 182
 183         entry.val = page_private(page);
 184
 185         spin_lock_irq(&swapper_space.tree_lock);
 186         __delete_from_swap_cache(page);
 187         spin_unlock_irq(&swapper_space.tree_lock);
 188
 189         swap_free(entry);
 190         page_cache_release(page);
 191 }
 192
 193 /*
 194  * If we are the only user, then try to free up the swap cache.
 195  *
 196  * Its ok to check for PageSwapCache without the page lock
 197  * here because we are going to recheck again inside
 198  * exclusive_swap_page() _with_ the lock.
 199  *                                      - Marcelo
 200  */
 201 static inline void free_swap_cache(struct page *page)
 202 {
 203         if (PageSwapCache(page) && !TestSetPageLocked(page)) {
 204                 remove_exclusive_swap_page(page);
 205                 unlock_page(page);
 206         }
 207 }
 208
 209 /*
 210  * Perform a free_page(), also freeing any swap cache associated with
 211  * this page if it is the last user of the page.
 212  */
 213 void free_page_and_swap_cache(struct page *page)
 214 {
 215         free_swap_cache(page);
 216         page_cache_release(page);
 217 }
 218
 219 /*
 220  * Passed an array of pages, drop them all from swapcache and then release
 221  * them.  They are removed from the LRU and freed if this is their last use.
 222  */
 223 void free_pages_and_swap_cache(struct page **pages, int nr)
 224 {
 225         struct page **pagep = pages;
 226
 227         lru_add_drain();
 228         while (nr) {
 229                 int todo = min(nr, PAGEVEC_SIZE);
 230                 int i;
 231
 232                 for (i = 0; i < todo; i++)
 233                         free_swap_cache(pagep[i]);
 234                 release_pages(pagep, todo, 0);
 235                 pagep += todo;
 236                 nr -= todo;
 237         }
 238 }
 239
 240 /*
 241  * Lookup a swap entry in the swap cache. A found page will be returned
 242  * unlocked and with its refcount incremented - we rely on the kernel
 243  * lock getting page table operations atomic even if we drop the page
 244  * lock before returning.
 245  */
 246 struct page * lookup_swap_cache(swp_entry_t entry)
 247 {
 248         struct page *page;
 249
 250         page = find_get_page(&swapper_space, entry.val);
 251
 252         if (page)
 253                 INC_CACHE_INFO(find_success);
 254
 255         INC_CACHE_INFO(find_total);
 256         return page;
 257 }
 258
 259 /*
 260  * Locate a page of swap in physical memory, reserving swap cache space
 261  * and reading the disk if it is not already cached.
 262  * A failure return means that either the page allocation failed or that
 263  * the swap entry is no longer in use.
 264  */
 265 struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
 266                         struct vm_area_struct *vma, unsigned long addr)
 267 {
 268         struct page *found_page, *new_page = NULL;
 269         int err;
 270
 271         do {
 272                 /*
 273                  * First check the swap cache.  Since this is normally
 274                  * called after lookup_swap_cache() failed, re-calling
 275                  * that would confuse statistics.
 276                  */
 277                 found_page = find_get_page(&swapper_space, entry.val);
 278                 if (found_page)
 279                         break;
 280
 281                 /*
 282                  * Get a new page to read into from swap.
 283                  */
 284                 if (!new_page) {
 285                         new_page = alloc_page_vma(gfp_mask, vma, addr);
 286                         if (!new_page)
 287                                 break;          /* Out of memory */
 288                 }
 289
 290                 /*
 291                  * Swap entry may have been freed since our caller observed it.
 292                  */
 293                 if (!swap_duplicate(entry))
 294                         break;
 295
 296                 /*
 297                  * Associate the page with swap entry in the swap cache.
 298                  * May fail (-EEXIST) if there is already a page associated
 299                  * with this entry in the swap cache: added by a racing
 300                  * read_swap_cache_async, or add_to_swap or shmem_writepage
 301                  * re-using the just freed swap entry for an existing page.
 302                  * May fail (-ENOMEM) if radix-tree node allocation failed.
 303                  */
 304                 SetPageLocked(new_page);
 305                 err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
 306                 if (!err) {
 307                         /*
 308                          * Initiate read into locked page and return.
 309                          */
 310                         lru_cache_add_active(new_page);
 311                         swap_readpage(NULL, new_page);
 312                         return new_page;
 313                 }
 314                 ClearPageLocked(new_page);
 315                 swap_free(entry);
 316         } while (err != -ENOMEM);
 317
 318         if (new_page)
 319                 page_cache_release(new_page);
 320         return found_page;
 321 }
 322
 323 /**
 324  * swapin_readahead - swap in pages in hope we need them soon
 325  * @entry: swap entry of this memory
 326  * @gfp_mask: memory allocation flags
 327  * @vma: user vma this address belongs to
 328  * @addr: target address for mempolicy
 329  *
 330  * Returns the struct page for entry and addr, after queueing swapin.
 331  *
 332  * Primitive swap readahead code. We simply read an aligned block of
 333  * (1 << page_cluster) entries in the swap area. This method is chosen
 334  * because it doesn't cost us any seek time.  We also make sure to queue
 335  * the 'original' request together with the readahead ones...
 336  *
 337  * This has been extended to use the NUMA policies from the mm triggering
 338  * the readahead.
 339  *
 340  * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
 341  */
 342 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
 343                         struct vm_area_struct *vma, unsigned long addr)
 344 {
 345         int nr_pages;
 346         struct page *page;
 347         unsigned long offset;
 348         unsigned long end_offset;
 349
 350         /*
 351          * Get starting offset for readaround, and number of pages to read.
 352          * Adjust starting address by readbehind (for NUMA interleave case)?
 353          * No, it's very unlikely that swap layout would follow vma layout,
 354          * more likely that neighbouring swap pages came from the same node:
 355          * so use the same "addr" to choose the same node for each swap read.
 356          */
 357         nr_pages = valid_swaphandles(entry, &offset);
 358         for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
 359                 /* Ok, do the async read-ahead now */
 360                 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
 361                                                 gfp_mask, vma, addr);
 362                 if (!page)
 363                         break;
 364                 page_cache_release(page);
 365         }
 366         lru_add_drain();        /* Push any new pages onto the LRU now */
 367         return read_swap_cache_async(entry, gfp_mask, vma, addr);
 368 }