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
  10 #include <linux/mm.h>
  11 #include <linux/kernel_stat.h>
  12 #include <linux/swap.h>
  13 #include <linux/init.h>
  14 #include <linux/pagemap.h>
  15 #include <linux/backing-dev.h>
  16
  17 #include <asm/pgtable.h>
  18
  19 static struct backing_dev_info swap_backing_dev_info = {
  20         .ra_pages       = 0,    /* No readahead */
  21         .memory_backed  = 1,    /* Does not contribute to dirty memory */
  22 };
  23
  24 extern struct address_space_operations swap_aops;
  25
  26 struct address_space swapper_space = {
  27         .page_tree      = RADIX_TREE_INIT(GFP_ATOMIC),
  28         .page_lock      = SPIN_LOCK_UNLOCKED,
  29         .clean_pages    = LIST_HEAD_INIT(swapper_space.clean_pages),
  30         .dirty_pages    = LIST_HEAD_INIT(swapper_space.dirty_pages),
  31         .io_pages       = LIST_HEAD_INIT(swapper_space.io_pages),
  32         .locked_pages   = LIST_HEAD_INIT(swapper_space.locked_pages),
  33         .a_ops          = &swap_aops,
  34         .backing_dev_info = &swap_backing_dev_info,
  35         .i_mmap         = LIST_HEAD_INIT(swapper_space.i_mmap),
  36         .i_mmap_shared  = LIST_HEAD_INIT(swapper_space.i_mmap_shared),
  37         .i_shared_sem   = __MUTEX_INITIALIZER(swapper_space.i_shared_sem),
  38         .truncate_count  = ATOMIC_INIT(0),
  39         .private_lock   = SPIN_LOCK_UNLOCKED,
  40         .private_list   = LIST_HEAD_INIT(swapper_space.private_list),
  41 };
  42
  43 #define INC_CACHE_INFO(x)       do { swap_cache_info.x++; } while (0)
  44
  45 static struct {
  46         unsigned long add_total;
  47         unsigned long del_total;
  48         unsigned long find_success;
  49         unsigned long find_total;
  50         unsigned long noent_race;
  51         unsigned long exist_race;
  52 } swap_cache_info;
  53
  54 void show_swap_cache_info(void)
  55 {
  56         printk("Swap cache: add %lu, delete %lu, find %lu/%lu, race %lu+%lu\n",
  57                 swap_cache_info.add_total, swap_cache_info.del_total,
  58                 swap_cache_info.find_success, swap_cache_info.find_total,
  59                 swap_cache_info.noent_race, swap_cache_info.exist_race);
  60 }
  61
  62 static int add_to_swap_cache(struct page *page, swp_entry_t entry)
  63 {
  64         int error;
  65
  66         if (page->mapping)
  67                 BUG();
  68         if (!swap_duplicate(entry)) {
  69                 INC_CACHE_INFO(noent_race);
  70                 return -ENOENT;
  71         }
  72         error = add_to_page_cache(page, &swapper_space, entry.val, GFP_KERNEL);
  73         /*
  74          * Anon pages are already on the LRU, we don't run lru_cache_add here.
  75          */
  76         if (error != 0) {
  77                 swap_free(entry);
  78                 if (error == -EEXIST)
  79                         INC_CACHE_INFO(exist_race);
  80                 return error;
  81         }
  82         if (!PageLocked(page))
  83                 BUG();
  84         if (!PageSwapCache(page))
  85                 BUG();
  86         INC_CACHE_INFO(add_total);
  87         return 0;
  88 }
  89
  90 /*
  91  * This must be called only on pages that have
  92  * been verified to be in the swap cache.
  93  */
  94 void __delete_from_swap_cache(struct page *page)
  95 {
  96         BUG_ON(!PageLocked(page));
  97         BUG_ON(!PageSwapCache(page));
  98         BUG_ON(PageWriteback(page));
  99         __remove_from_page_cache(page);
 100         INC_CACHE_INFO(del_total);
 101 }
 102
 103 /**
 104  * add_to_swap - allocate swap space for a page
 105  * @page: page we want to move to swap
 106  *
 107  * Allocate swap space for the page and add the page to the
 108  * swap cache.  Caller needs to hold the page lock.
 109  */
 110 int add_to_swap(struct page * page)
 111 {
 112         swp_entry_t entry;
 113         int pf_flags;
 114         int err;
 115
 116         if (!PageLocked(page))
 117                 BUG();
 118
 119         for (;;) {
 120                 entry = get_swap_page();
 121                 if (!entry.val)
 122                         return 0;
 123
 124                 /* Radix-tree node allocations are performing
 125                  * GFP_ATOMIC allocations under PF_MEMALLOC.
 126                  * They can completely exhaust the page allocator.
 127                  *
 128                  * So PF_MEMALLOC is dropped here.  This causes the slab
 129                  * allocations to fail earlier, so radix-tree nodes will
 130                  * then be allocated from the mempool reserves.
 131                  *
 132                  * We're still using __GFP_HIGH for radix-tree node
 133                  * allocations, so some of the emergency pools are available,
 134                  * just not all of them.
 135                  */
 136
 137                 pf_flags = current->flags;
 138                 current->flags &= ~PF_MEMALLOC;
 139
 140                 /*
 141                  * Add it to the swap cache and mark it dirty
 142                  */
 143                 err = add_to_page_cache(page, &swapper_space,
 144                                         entry.val, GFP_ATOMIC);
 145
 146                 if (pf_flags & PF_MEMALLOC)
 147                         current->flags |= PF_MEMALLOC;
 148
 149                 switch (err) {
 150                 case 0:                         /* Success */
 151                         SetPageUptodate(page);
 152                         ClearPageDirty(page);
 153                         set_page_dirty(page);
 154                         INC_CACHE_INFO(add_total);
 155                         return 1;
 156                 case -EEXIST:
 157                         /* Raced with "speculative" read_swap_cache_async */
 158                         INC_CACHE_INFO(exist_race);
 159                         swap_free(entry);
 160                         continue;
 161                 default:
 162                         /* -ENOMEM radix-tree allocation failure */
 163                         swap_free(entry);
 164                         return 0;
 165                 }
 166         }
 167 }
 168
 169 /*
 170  * This must be called only on pages that have
 171  * been verified to be in the swap cache and locked.
 172  * It will never put the page into the free list,
 173  * the caller has a reference on the page.
 174  */
 175 void delete_from_swap_cache(struct page *page)
 176 {
 177         swp_entry_t entry;
 178
 179         BUG_ON(!PageLocked(page));
 180         BUG_ON(PageWriteback(page));
 181         BUG_ON(PagePrivate(page));
 182
 183         entry.val = page->index;
 184
 185         spin_lock(&swapper_space.page_lock);
 186         __delete_from_swap_cache(page);
 187         spin_unlock(&swapper_space.page_lock);
 188
 189         swap_free(entry);
 190         page_cache_release(page);
 191 }
 192
 193 int move_to_swap_cache(struct page *page, swp_entry_t entry)
 194 {
 195         struct address_space *mapping = page->mapping;
 196         int err;
 197
 198         spin_lock(&swapper_space.page_lock);
 199         spin_lock(&mapping->page_lock);
 200
 201         err = radix_tree_insert(&swapper_space.page_tree, entry.val, page);
 202         if (!err) {
 203                 __remove_from_page_cache(page);
 204                 ___add_to_page_cache(page, &swapper_space, entry.val);
 205         }
 206
 207         spin_unlock(&mapping->page_lock);
 208         spin_unlock(&swapper_space.page_lock);
 209
 210         if (!err) {
 211                 if (!swap_duplicate(entry))
 212                         BUG();
 213                 /* shift page from clean_pages to dirty_pages list */
 214                 BUG_ON(PageDirty(page));
 215                 set_page_dirty(page);
 216                 INC_CACHE_INFO(add_total);
 217         } else if (err == -EEXIST)
 218                 INC_CACHE_INFO(exist_race);
 219         return err;
 220 }
 221
 222 int move_from_swap_cache(struct page *page, unsigned long index,
 223                 struct address_space *mapping)
 224 {
 225         swp_entry_t entry;
 226         int err;
 227
 228         BUG_ON(!PageLocked(page));
 229         BUG_ON(PageWriteback(page));
 230         BUG_ON(PagePrivate(page));
 231
 232         entry.val = page->index;
 233
 234         spin_lock(&swapper_space.page_lock);
 235         spin_lock(&mapping->page_lock);
 236
 237         err = radix_tree_insert(&mapping->page_tree, index, page);
 238         if (!err) {
 239                 __delete_from_swap_cache(page);
 240                 ___add_to_page_cache(page, mapping, index);
 241         }
 242
 243         spin_unlock(&mapping->page_lock);
 244         spin_unlock(&swapper_space.page_lock);
 245
 246         if (!err) {
 247                 swap_free(entry);
 248                 /* shift page from clean_pages to dirty_pages list */
 249                 ClearPageDirty(page);
 250                 set_page_dirty(page);
 251         }
 252         return err;
 253 }
 254
 255
 256 /*
 257  * If we are the only user, then try to free up the swap cache.
 258  *
 259  * Its ok to check for PageSwapCache without the page lock
 260  * here because we are going to recheck again inside
 261  * exclusive_swap_page() _with_ the lock.
 262  *                                      - Marcelo
 263  */
 264 static inline void free_swap_cache(struct page *page)
 265 {
 266         if (PageSwapCache(page) && !TestSetPageLocked(page)) {
 267                 remove_exclusive_swap_page(page);
 268                 unlock_page(page);
 269         }
 270 }
 271
 272 /*
 273  * Perform a free_page(), also freeing any swap cache associated with
 274  * this page if it is the last user of the page. Can not do a lock_page,
 275  * as we are holding the page_table_lock spinlock.
 276  */
 277 void free_page_and_swap_cache(struct page *page)
 278 {
 279         free_swap_cache(page);
 280         page_cache_release(page);
 281 }
 282
 283 /*
 284  * Passed an array of pages, drop them all from swapcache and then release
 285  * them.  They are removed from the LRU and freed if this is their last use.
 286  */
 287 void free_pages_and_swap_cache(struct page **pages, int nr)
 288 {
 289         int chunk = 16;
 290         struct page **pagep = pages;
 291
 292         lru_add_drain();
 293         while (nr) {
 294                 int todo = min(chunk, nr);
 295                 int i;
 296
 297                 for (i = 0; i < todo; i++)
 298                         free_swap_cache(pagep[i]);
 299                 release_pages(pagep, todo, 0);
 300                 pagep += todo;
 301                 nr -= todo;
 302         }
 303 }
 304
 305 /*
 306  * Lookup a swap entry in the swap cache. A found page will be returned
 307  * unlocked and with its refcount incremented - we rely on the kernel
 308  * lock getting page table operations atomic even if we drop the page
 309  * lock before returning.
 310  */
 311 struct page * lookup_swap_cache(swp_entry_t entry)
 312 {
 313         struct page *found;
 314
 315         found = find_get_page(&swapper_space, entry.val);
 316         /*
 317          * Unsafe to assert PageSwapCache and mapping on page found:
 318          * if SMP nothing prevents swapoff from deleting this page from
 319          * the swap cache at this moment.  find_lock_page would prevent
 320          * that, but no need to change: we _have_ got the right page.
 321          */
 322         INC_CACHE_INFO(find_total);
 323         if (found)
 324                 INC_CACHE_INFO(find_success);
 325         return found;
 326 }
 327
 328 /*
 329  * Locate a page of swap in physical memory, reserving swap cache space
 330  * and reading the disk if it is not already cached.
 331  * A failure return means that either the page allocation failed or that
 332  * the swap entry is no longer in use.
 333  */
 334 struct page * read_swap_cache_async(swp_entry_t entry)
 335 {
 336         struct page *found_page, *new_page = NULL;
 337         int err;
 338
 339         do {
 340                 /*
 341                  * First check the swap cache.  Since this is normally
 342                  * called after lookup_swap_cache() failed, re-calling
 343                  * that would confuse statistics: use find_get_page()
 344                  * directly.
 345                  */
 346                 found_page = find_get_page(&swapper_space, entry.val);
 347                 if (found_page)
 348                         break;
 349
 350                 /*
 351                  * Get a new page to read into from swap.
 352                  */
 353                 if (!new_page) {
 354                         new_page = alloc_page(GFP_HIGHUSER);
 355                         if (!new_page)
 356                                 break;          /* Out of memory */
 357                 }
 358
 359                 /*
 360                  * Associate the page with swap entry in the swap cache.
 361                  * May fail (-ENOENT) if swap entry has been freed since
 362                  * our caller observed it.  May fail (-EEXIST) if there
 363                  * is already a page associated with this entry in the
 364                  * swap cache: added by a racing read_swap_cache_async,
 365                  * or by try_to_swap_out (or shmem_writepage) re-using
 366                  * the just freed swap entry for an existing page.
 367                  * May fail (-ENOMEM) if radix-tree node allocation failed.
 368                  */
 369                 err = add_to_swap_cache(new_page, entry);
 370                 if (!err) {
 371                         /*
 372                          * Initiate read into locked page and return.
 373                          */
 374                         lru_cache_add_active(new_page);
 375                         swap_readpage(NULL, new_page);
 376                         return new_page;
 377                 }
 378         } while (err != -ENOENT && err != -ENOMEM);
 379
 380         if (new_page)
 381                 page_cache_release(new_page);
 382         return found_page;
 383 }