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 #include <linux/memcontrol.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, to make sync_page look nicer, and to allow
  27  * future use of radix_tree tags in the swap cache.
  28  */
  29 static const struct address_space_operations swap_aops = {
  30         .writepage      = swap_writepage,
  31         .sync_page      = block_sync_page,
  32         .set_page_dirty = __set_page_dirty_nobuffers,
  33         .migratepage    = migrate_page,
  34 };
  35
  36 static struct backing_dev_info swap_backing_dev_info = {
  37         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
  38         .unplug_io_fn   = swap_unplug_io_fn,
  39 };
  40
  41 struct address_space swapper_space = {
  42         .page_tree      = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
  43         .tree_lock      = __RW_LOCK_UNLOCKED(swapper_space.tree_lock),
  44         .a_ops          = &swap_aops,
  45         .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
  46         .backing_dev_info = &swap_backing_dev_info,
  47 };
  48
  49 #define INC_CACHE_INFO(x)       do { swap_cache_info.x++; } while (0)
  50
  51 static struct {
  52         unsigned long add_total;
  53         unsigned long del_total;
  54         unsigned long find_success;
  55         unsigned long find_total;
  56 } swap_cache_info;
  57
  58 void show_swap_cache_info(void)
  59 {
  60         printk("Swap cache: add %lu, delete %lu, find %lu/%lu\n",
  61                 swap_cache_info.add_total, swap_cache_info.del_total,
  62                 swap_cache_info.find_success, swap_cache_info.find_total);
  63         printk("Free swap  = %lukB\n", nr_swap_pages << (PAGE_SHIFT - 10));
  64         printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
  65 }
  66
  67 /*
  68  * add_to_swap_cache resembles add_to_page_cache on swapper_space,
  69  * but sets SwapCache flag and private instead of mapping and index.
  70  */
  71 int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
  72 {
  73         int error;
  74
  75         BUG_ON(!PageLocked(page));
  76         BUG_ON(PageSwapCache(page));
  77         BUG_ON(PagePrivate(page));
  78         error = radix_tree_preload(gfp_mask);
  79         if (!error) {
  80
  81                 error = mem_cgroup_charge(page, current->mm);
  82                 if (error)
  83                         goto out;
  84
  85                 write_lock_irq(&swapper_space.tree_lock);
  86                 error = radix_tree_insert(&swapper_space.page_tree,
  87                                                 entry.val, page);
  88                 if (!error) {
  89                         page_cache_get(page);
  90                         SetPageSwapCache(page);
  91                         set_page_private(page, entry.val);
  92                         total_swapcache_pages++;
  93                         __inc_zone_page_state(page, NR_FILE_PAGES);
  94                         INC_CACHE_INFO(add_total);
  95                 } else {
  96                         mem_cgroup_uncharge_page(page);
  97                 }
  98                 write_unlock_irq(&swapper_space.tree_lock);
  99                 radix_tree_preload_end();
 100         }
 101 out:
 102         return error;
 103 }
 104
 105 /*
 106  * This must be called only on pages that have
 107  * been verified to be in the swap cache.
 108  */
 109 void __delete_from_swap_cache(struct page *page)
 110 {
 111         BUG_ON(!PageLocked(page));
 112         BUG_ON(!PageSwapCache(page));
 113         BUG_ON(PageWriteback(page));
 114         BUG_ON(PagePrivate(page));
 115
 116         mem_cgroup_uncharge_page(page);
 117         radix_tree_delete(&swapper_space.page_tree, page_private(page));
 118         set_page_private(page, 0);
 119         ClearPageSwapCache(page);
 120         total_swapcache_pages--;
 121         __dec_zone_page_state(page, NR_FILE_PAGES);
 122         INC_CACHE_INFO(del_total);
 123 }
 124
 125 /**
 126  * add_to_swap - allocate swap space for a page
 127  * @page: page we want to move to swap
 128  *
 129  * Allocate swap space for the page and add the page to the
 130  * swap cache.  Caller needs to hold the page lock.
 131  */
 132 int add_to_swap(struct page * page, gfp_t gfp_mask)
 133 {
 134         swp_entry_t entry;
 135         int err;
 136
 137         BUG_ON(!PageLocked(page));
 138         BUG_ON(!PageUptodate(page));
 139
 140         for (;;) {
 141                 entry = get_swap_page();
 142                 if (!entry.val)
 143                         return 0;
 144
 145                 /*
 146                  * Radix-tree node allocations from PF_MEMALLOC contexts could
 147                  * completely exhaust the page allocator. __GFP_NOMEMALLOC
 148                  * stops emergency reserves from being allocated.
 149                  *
 150                  * TODO: this could cause a theoretical memory reclaim
 151                  * deadlock in the swap out path.
 152                  */
 153                 /*
 154                  * Add it to the swap cache and mark it dirty
 155                  */
 156                 err = add_to_swap_cache(page, entry,
 157                                 gfp_mask|__GFP_NOMEMALLOC|__GFP_NOWARN);
 158
 159                 switch (err) {
 160                 case 0:                         /* Success */
 161                         SetPageDirty(page);
 162                         return 1;
 163                 case -EEXIST:
 164                         /* Raced with "speculative" read_swap_cache_async */
 165                         swap_free(entry);
 166                         continue;
 167                 default:
 168                         /* -ENOMEM radix-tree allocation failure */
 169                         swap_free(entry);
 170                         return 0;
 171                 }
 172         }
 173 }
 174
 175 /*
 176  * This must be called only on pages that have
 177  * been verified to be in the swap cache and locked.
 178  * It will never put the page into the free list,
 179  * the caller has a reference on the page.
 180  */
 181 void delete_from_swap_cache(struct page *page)
 182 {
 183         swp_entry_t entry;
 184
 185         entry.val = page_private(page);
 186
 187         write_lock_irq(&swapper_space.tree_lock);
 188         __delete_from_swap_cache(page);
 189         write_unlock_irq(&swapper_space.tree_lock);
 190
 191         swap_free(entry);
 192         page_cache_release(page);
 193 }
 194
 195 /*
 196  * If we are the only user, then try to free up the swap cache.
 197  *
 198  * Its ok to check for PageSwapCache without the page lock
 199  * here because we are going to recheck again inside
 200  * exclusive_swap_page() _with_ the lock.
 201  *                                      - Marcelo
 202  */
 203 static inline void free_swap_cache(struct page *page)
 204 {
 205         if (PageSwapCache(page) && !TestSetPageLocked(page)) {
 206                 remove_exclusive_swap_page(page);
 207                 unlock_page(page);
 208         }
 209 }
 210
 211 /*
 212  * Perform a free_page(), also freeing any swap cache associated with
 213  * this page if it is the last user of the page.
 214  */
 215 void free_page_and_swap_cache(struct page *page)
 216 {
 217         free_swap_cache(page);
 218         page_cache_release(page);
 219 }
 220
 221 /*
 222  * Passed an array of pages, drop them all from swapcache and then release
 223  * them.  They are removed from the LRU and freed if this is their last use.
 224  */
 225 void free_pages_and_swap_cache(struct page **pages, int nr)
 226 {
 227         struct page **pagep = pages;
 228
 229         lru_add_drain();
 230         while (nr) {
 231                 int todo = min(nr, PAGEVEC_SIZE);
 232                 int i;
 233
 234                 for (i = 0; i < todo; i++)
 235                         free_swap_cache(pagep[i]);
 236                 release_pages(pagep, todo, 0);
 237                 pagep += todo;
 238                 nr -= todo;
 239         }
 240 }
 241
 242 /*
 243  * Lookup a swap entry in the swap cache. A found page will be returned
 244  * unlocked and with its refcount incremented - we rely on the kernel
 245  * lock getting page table operations atomic even if we drop the page
 246  * lock before returning.
 247  */
 248 struct page * lookup_swap_cache(swp_entry_t entry)
 249 {
 250         struct page *page;
 251
 252         page = find_get_page(&swapper_space, entry.val);
 253
 254         if (page)
 255                 INC_CACHE_INFO(find_success);
 256
 257         INC_CACHE_INFO(find_total);
 258         return page;
 259 }
 260
 261 /*
 262  * Locate a page of swap in physical memory, reserving swap cache space
 263  * and reading the disk if it is not already cached.
 264  * A failure return means that either the page allocation failed or that
 265  * the swap entry is no longer in use.
 266  */
 267 struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
 268                         struct vm_area_struct *vma, unsigned long addr)
 269 {
 270         struct page *found_page, *new_page = NULL;
 271         int err;
 272
 273         do {
 274                 /*
 275                  * First check the swap cache.  Since this is normally
 276                  * called after lookup_swap_cache() failed, re-calling
 277                  * that would confuse statistics.
 278                  */
 279                 found_page = find_get_page(&swapper_space, entry.val);
 280                 if (found_page)
 281                         break;
 282
 283                 /*
 284                  * Get a new page to read into from swap.
 285                  */
 286                 if (!new_page) {
 287                         new_page = alloc_page_vma(gfp_mask, vma, addr);
 288                         if (!new_page)
 289                                 break;          /* Out of memory */
 290                 }
 291
 292                 /*
 293                  * Swap entry may have been freed since our caller observed it.
 294                  */
 295                 if (!swap_duplicate(entry))
 296                         break;
 297
 298                 /*
 299                  * Associate the page with swap entry in the swap cache.
 300                  * May fail (-EEXIST) if there is already a page associated
 301                  * with this entry in the swap cache: added by a racing
 302                  * read_swap_cache_async, or add_to_swap or shmem_writepage
 303                  * re-using the just freed swap entry for an existing page.
 304                  * May fail (-ENOMEM) if radix-tree node allocation failed.
 305                  */
 306                 SetPageLocked(new_page);
 307                 err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
 308                 if (!err) {
 309                         /*
 310                          * Initiate read into locked page and return.
 311                          */
 312                         lru_cache_add_active(new_page);
 313                         swap_readpage(NULL, new_page);
 314                         return new_page;
 315                 }
 316                 ClearPageLocked(new_page);
 317                 swap_free(entry);
 318         } while (err != -ENOMEM);
 319
 320         if (new_page)
 321                 page_cache_release(new_page);
 322         return found_page;
 323 }
 324
 325 /**
 326  * swapin_readahead - swap in pages in hope we need them soon
 327  * @entry: swap entry of this memory
 328  * @vma: user vma this address belongs to
 329  * @addr: target address for mempolicy
 330  *
 331  * Returns the struct page for entry and addr, after queueing swapin.
 332  *
 333  * Primitive swap readahead code. We simply read an aligned block of
 334  * (1 << page_cluster) entries in the swap area. This method is chosen
 335  * because it doesn't cost us any seek time.  We also make sure to queue
 336  * the 'original' request together with the readahead ones...
 337  *
 338  * This has been extended to use the NUMA policies from the mm triggering
 339  * the readahead.
 340  *
 341  * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
 342  */
 343 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
 344                         struct vm_area_struct *vma, unsigned long addr)
 345 {
 346         int nr_pages;
 347         struct page *page;
 348         unsigned long offset;
 349         unsigned long end_offset;
 350
 351         /*
 352          * Get starting offset for readaround, and number of pages to read.
 353          * Adjust starting address by readbehind (for NUMA interleave case)?
 354          * No, it's very unlikely that swap layout would follow vma layout,
 355          * more likely that neighbouring swap pages came from the same node:
 356          * so use the same "addr" to choose the same node for each swap read.
 357          */
 358         nr_pages = valid_swaphandles(entry, &offset);
 359         for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
 360                 /* Ok, do the async read-ahead now */
 361                 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
 362                                                 gfp_mask, vma, addr);
 363                 if (!page)
 364                         break;
 365                 page_cache_release(page);
 366         }
 367         lru_add_drain();        /* Push any new pages onto the LRU now */
 368         return read_swap_cache_async(entry, gfp_mask, vma, addr);
 369 }