Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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/init.h> |
| 14 | #include <linux/pagemap.h> |
| 15 | #include <linux/buffer_head.h> |
| 16 | #include <linux/backing-dev.h> |
Hugh Dickins | c484d41 | 2006-01-06 00:10:55 -0800 | [diff] [blame] | 17 | #include <linux/pagevec.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 18 | #include <linux/migrate.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | |
| 20 | #include <asm/pgtable.h> |
| 21 | |
| 22 | /* |
| 23 | * swapper_space is a fiction, retained to simplify the path through |
| 24 | * vmscan's shrink_list, to make sync_page look nicer, and to allow |
| 25 | * future use of radix_tree tags in the swap cache. |
| 26 | */ |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 27 | static const struct address_space_operations swap_aops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | .writepage = swap_writepage, |
| 29 | .sync_page = block_sync_page, |
| 30 | .set_page_dirty = __set_page_dirty_nobuffers, |
Christoph Lameter | e965f96 | 2006-02-01 03:05:41 -0800 | [diff] [blame] | 31 | .migratepage = migrate_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | }; |
| 33 | |
| 34 | static struct backing_dev_info swap_backing_dev_info = { |
| 35 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, |
| 36 | .unplug_io_fn = swap_unplug_io_fn, |
| 37 | }; |
| 38 | |
| 39 | struct address_space swapper_space = { |
| 40 | .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN), |
| 41 | .tree_lock = RW_LOCK_UNLOCKED, |
| 42 | .a_ops = &swap_aops, |
| 43 | .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear), |
| 44 | .backing_dev_info = &swap_backing_dev_info, |
| 45 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | unsigned long noent_race; |
| 55 | unsigned long exist_race; |
| 56 | } swap_cache_info; |
| 57 | |
| 58 | void show_swap_cache_info(void) |
| 59 | { |
| 60 | printk("Swap cache: add %lu, delete %lu, find %lu/%lu, race %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 | swap_cache_info.noent_race, swap_cache_info.exist_race); |
| 64 | printk("Free swap = %lukB\n", nr_swap_pages << (PAGE_SHIFT - 10)); |
| 65 | printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10)); |
| 66 | } |
| 67 | |
| 68 | /* |
| 69 | * __add_to_swap_cache resembles add_to_page_cache on swapper_space, |
| 70 | * but sets SwapCache flag and private instead of mapping and index. |
| 71 | */ |
Victor Fusco | 9de75d1 | 2005-09-10 00:26:46 -0700 | [diff] [blame] | 72 | static int __add_to_swap_cache(struct page *page, swp_entry_t entry, |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 73 | gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | { |
| 75 | int error; |
| 76 | |
| 77 | BUG_ON(PageSwapCache(page)); |
| 78 | BUG_ON(PagePrivate(page)); |
| 79 | error = radix_tree_preload(gfp_mask); |
| 80 | if (!error) { |
| 81 | write_lock_irq(&swapper_space.tree_lock); |
| 82 | error = radix_tree_insert(&swapper_space.page_tree, |
| 83 | entry.val, page); |
| 84 | if (!error) { |
| 85 | page_cache_get(page); |
| 86 | SetPageLocked(page); |
| 87 | SetPageSwapCache(page); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 88 | set_page_private(page, entry.val); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | total_swapcache_pages++; |
| 90 | pagecache_acct(1); |
| 91 | } |
| 92 | write_unlock_irq(&swapper_space.tree_lock); |
| 93 | radix_tree_preload_end(); |
| 94 | } |
| 95 | return error; |
| 96 | } |
| 97 | |
| 98 | static int add_to_swap_cache(struct page *page, swp_entry_t entry) |
| 99 | { |
| 100 | int error; |
| 101 | |
| 102 | if (!swap_duplicate(entry)) { |
| 103 | INC_CACHE_INFO(noent_race); |
| 104 | return -ENOENT; |
| 105 | } |
| 106 | error = __add_to_swap_cache(page, entry, GFP_KERNEL); |
| 107 | /* |
| 108 | * Anon pages are already on the LRU, we don't run lru_cache_add here. |
| 109 | */ |
| 110 | if (error) { |
| 111 | swap_free(entry); |
| 112 | if (error == -EEXIST) |
| 113 | INC_CACHE_INFO(exist_race); |
| 114 | return error; |
| 115 | } |
| 116 | INC_CACHE_INFO(add_total); |
| 117 | return 0; |
| 118 | } |
| 119 | |
| 120 | /* |
| 121 | * This must be called only on pages that have |
| 122 | * been verified to be in the swap cache. |
| 123 | */ |
| 124 | void __delete_from_swap_cache(struct page *page) |
| 125 | { |
| 126 | BUG_ON(!PageLocked(page)); |
| 127 | BUG_ON(!PageSwapCache(page)); |
| 128 | BUG_ON(PageWriteback(page)); |
Hugh Dickins | 3279ffd | 2005-09-03 15:54:43 -0700 | [diff] [blame] | 129 | BUG_ON(PagePrivate(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 131 | radix_tree_delete(&swapper_space.page_tree, page_private(page)); |
| 132 | set_page_private(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | ClearPageSwapCache(page); |
| 134 | total_swapcache_pages--; |
| 135 | pagecache_acct(-1); |
| 136 | INC_CACHE_INFO(del_total); |
| 137 | } |
| 138 | |
| 139 | /** |
| 140 | * add_to_swap - allocate swap space for a page |
| 141 | * @page: page we want to move to swap |
| 142 | * |
| 143 | * Allocate swap space for the page and add the page to the |
| 144 | * swap cache. Caller needs to hold the page lock. |
| 145 | */ |
Christoph Lameter | 1480a54 | 2006-01-08 01:00:53 -0800 | [diff] [blame] | 146 | int add_to_swap(struct page * page, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | { |
| 148 | swp_entry_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | int err; |
| 150 | |
Eric Sesterhenn | e74ca2b | 2006-04-01 01:25:12 +0200 | [diff] [blame] | 151 | BUG_ON(!PageLocked(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | |
| 153 | for (;;) { |
| 154 | entry = get_swap_page(); |
| 155 | if (!entry.val) |
| 156 | return 0; |
| 157 | |
Nick Piggin | bd53b71 | 2005-05-01 08:58:37 -0700 | [diff] [blame] | 158 | /* |
| 159 | * Radix-tree node allocations from PF_MEMALLOC contexts could |
| 160 | * completely exhaust the page allocator. __GFP_NOMEMALLOC |
| 161 | * stops emergency reserves from being allocated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | * |
Nick Piggin | bd53b71 | 2005-05-01 08:58:37 -0700 | [diff] [blame] | 163 | * TODO: this could cause a theoretical memory reclaim |
| 164 | * deadlock in the swap out path. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | /* |
| 167 | * Add it to the swap cache and mark it dirty |
| 168 | */ |
Nick Piggin | bd53b71 | 2005-05-01 08:58:37 -0700 | [diff] [blame] | 169 | err = __add_to_swap_cache(page, entry, |
Christoph Lameter | 1480a54 | 2006-01-08 01:00:53 -0800 | [diff] [blame] | 170 | gfp_mask|__GFP_NOMEMALLOC|__GFP_NOWARN); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | |
| 172 | switch (err) { |
| 173 | case 0: /* Success */ |
| 174 | SetPageUptodate(page); |
| 175 | SetPageDirty(page); |
| 176 | INC_CACHE_INFO(add_total); |
| 177 | return 1; |
| 178 | case -EEXIST: |
| 179 | /* Raced with "speculative" read_swap_cache_async */ |
| 180 | INC_CACHE_INFO(exist_race); |
| 181 | swap_free(entry); |
| 182 | continue; |
| 183 | default: |
| 184 | /* -ENOMEM radix-tree allocation failure */ |
| 185 | swap_free(entry); |
| 186 | return 0; |
| 187 | } |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | /* |
| 192 | * This must be called only on pages that have |
| 193 | * been verified to be in the swap cache and locked. |
| 194 | * It will never put the page into the free list, |
| 195 | * the caller has a reference on the page. |
| 196 | */ |
| 197 | void delete_from_swap_cache(struct page *page) |
| 198 | { |
| 199 | swp_entry_t entry; |
| 200 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 201 | entry.val = page_private(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | |
| 203 | write_lock_irq(&swapper_space.tree_lock); |
| 204 | __delete_from_swap_cache(page); |
| 205 | write_unlock_irq(&swapper_space.tree_lock); |
| 206 | |
| 207 | swap_free(entry); |
| 208 | page_cache_release(page); |
| 209 | } |
| 210 | |
| 211 | /* |
| 212 | * Strange swizzling function only for use by shmem_writepage |
| 213 | */ |
| 214 | int move_to_swap_cache(struct page *page, swp_entry_t entry) |
| 215 | { |
| 216 | int err = __add_to_swap_cache(page, entry, GFP_ATOMIC); |
| 217 | if (!err) { |
| 218 | remove_from_page_cache(page); |
| 219 | page_cache_release(page); /* pagecache ref */ |
| 220 | if (!swap_duplicate(entry)) |
| 221 | BUG(); |
| 222 | SetPageDirty(page); |
| 223 | INC_CACHE_INFO(add_total); |
| 224 | } else if (err == -EEXIST) |
| 225 | INC_CACHE_INFO(exist_race); |
| 226 | return err; |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * Strange swizzling function for shmem_getpage (and shmem_unuse) |
| 231 | */ |
| 232 | int move_from_swap_cache(struct page *page, unsigned long index, |
| 233 | struct address_space *mapping) |
| 234 | { |
| 235 | int err = add_to_page_cache(page, mapping, index, GFP_ATOMIC); |
| 236 | if (!err) { |
| 237 | delete_from_swap_cache(page); |
| 238 | /* shift page from clean_pages to dirty_pages list */ |
| 239 | ClearPageDirty(page); |
| 240 | set_page_dirty(page); |
| 241 | } |
| 242 | return err; |
| 243 | } |
| 244 | |
| 245 | /* |
| 246 | * If we are the only user, then try to free up the swap cache. |
| 247 | * |
| 248 | * Its ok to check for PageSwapCache without the page lock |
| 249 | * here because we are going to recheck again inside |
| 250 | * exclusive_swap_page() _with_ the lock. |
| 251 | * - Marcelo |
| 252 | */ |
| 253 | static inline void free_swap_cache(struct page *page) |
| 254 | { |
| 255 | if (PageSwapCache(page) && !TestSetPageLocked(page)) { |
| 256 | remove_exclusive_swap_page(page); |
| 257 | unlock_page(page); |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | /* |
| 262 | * Perform a free_page(), also freeing any swap cache associated with |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 263 | * this page if it is the last user of the page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | */ |
| 265 | void free_page_and_swap_cache(struct page *page) |
| 266 | { |
| 267 | free_swap_cache(page); |
| 268 | page_cache_release(page); |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * Passed an array of pages, drop them all from swapcache and then release |
| 273 | * them. They are removed from the LRU and freed if this is their last use. |
| 274 | */ |
| 275 | void free_pages_and_swap_cache(struct page **pages, int nr) |
| 276 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | struct page **pagep = pages; |
| 278 | |
| 279 | lru_add_drain(); |
| 280 | while (nr) { |
Hugh Dickins | c484d41 | 2006-01-06 00:10:55 -0800 | [diff] [blame] | 281 | int todo = min(nr, PAGEVEC_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | int i; |
| 283 | |
| 284 | for (i = 0; i < todo; i++) |
| 285 | free_swap_cache(pagep[i]); |
| 286 | release_pages(pagep, todo, 0); |
| 287 | pagep += todo; |
| 288 | nr -= todo; |
| 289 | } |
| 290 | } |
| 291 | |
| 292 | /* |
| 293 | * Lookup a swap entry in the swap cache. A found page will be returned |
| 294 | * unlocked and with its refcount incremented - we rely on the kernel |
| 295 | * lock getting page table operations atomic even if we drop the page |
| 296 | * lock before returning. |
| 297 | */ |
| 298 | struct page * lookup_swap_cache(swp_entry_t entry) |
| 299 | { |
| 300 | struct page *page; |
| 301 | |
| 302 | page = find_get_page(&swapper_space, entry.val); |
| 303 | |
| 304 | if (page) |
| 305 | INC_CACHE_INFO(find_success); |
| 306 | |
| 307 | INC_CACHE_INFO(find_total); |
| 308 | return page; |
| 309 | } |
| 310 | |
| 311 | /* |
| 312 | * Locate a page of swap in physical memory, reserving swap cache space |
| 313 | * and reading the disk if it is not already cached. |
| 314 | * A failure return means that either the page allocation failed or that |
| 315 | * the swap entry is no longer in use. |
| 316 | */ |
| 317 | struct page *read_swap_cache_async(swp_entry_t entry, |
| 318 | struct vm_area_struct *vma, unsigned long addr) |
| 319 | { |
| 320 | struct page *found_page, *new_page = NULL; |
| 321 | int err; |
| 322 | |
| 323 | do { |
| 324 | /* |
| 325 | * First check the swap cache. Since this is normally |
| 326 | * called after lookup_swap_cache() failed, re-calling |
| 327 | * that would confuse statistics. |
| 328 | */ |
| 329 | found_page = find_get_page(&swapper_space, entry.val); |
| 330 | if (found_page) |
| 331 | break; |
| 332 | |
| 333 | /* |
| 334 | * Get a new page to read into from swap. |
| 335 | */ |
| 336 | if (!new_page) { |
| 337 | new_page = alloc_page_vma(GFP_HIGHUSER, vma, addr); |
| 338 | if (!new_page) |
| 339 | break; /* Out of memory */ |
| 340 | } |
| 341 | |
| 342 | /* |
| 343 | * Associate the page with swap entry in the swap cache. |
| 344 | * May fail (-ENOENT) if swap entry has been freed since |
| 345 | * our caller observed it. May fail (-EEXIST) if there |
| 346 | * is already a page associated with this entry in the |
| 347 | * swap cache: added by a racing read_swap_cache_async, |
| 348 | * or by try_to_swap_out (or shmem_writepage) re-using |
| 349 | * the just freed swap entry for an existing page. |
| 350 | * May fail (-ENOMEM) if radix-tree node allocation failed. |
| 351 | */ |
| 352 | err = add_to_swap_cache(new_page, entry); |
| 353 | if (!err) { |
| 354 | /* |
| 355 | * Initiate read into locked page and return. |
| 356 | */ |
| 357 | lru_cache_add_active(new_page); |
| 358 | swap_readpage(NULL, new_page); |
| 359 | return new_page; |
| 360 | } |
| 361 | } while (err != -ENOENT && err != -ENOMEM); |
| 362 | |
| 363 | if (new_page) |
| 364 | page_cache_release(new_page); |
| 365 | return found_page; |
| 366 | } |