blob: 797c3831cbec7401ccfecd3ecdc98c7667c8d6bb [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Hugh Dickins46017e92008-02-04 22:28:41 -080013#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/init.h>
15#include <linux/pagemap.h>
16#include <linux/buffer_head.h>
17#include <linux/backing-dev.h>
Hugh Dickinsc484d412006-01-06 00:10:55 -080018#include <linux/pagevec.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080019#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
21#include <asm/pgtable.h>
22
23/*
24 * swapper_space is a fiction, retained to simplify the path through
Anderson Briglia2706a1b2007-07-15 23:38:09 -070025 * vmscan's shrink_page_list, to make sync_page look nicer, and to allow
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 * future use of radix_tree tags in the swap cache.
27 */
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070028static const struct address_space_operations swap_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 .writepage = swap_writepage,
30 .sync_page = block_sync_page,
31 .set_page_dirty = __set_page_dirty_nobuffers,
Christoph Lametere965f962006-02-01 03:05:41 -080032 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -070033};
34
35static struct backing_dev_info swap_backing_dev_info = {
Miklos Szeredie4ad08f2008-04-30 00:54:37 -070036 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 .unplug_io_fn = swap_unplug_io_fn,
38};
39
40struct address_space swapper_space = {
41 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
Nick Piggin19fd6232008-07-25 19:45:32 -070042 .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 .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};
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
49
50static struct {
51 unsigned long add_total;
52 unsigned long del_total;
53 unsigned long find_success;
54 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055} swap_cache_info;
56
57void show_swap_cache_info(void)
58{
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070059 printk("%lu pages in swap cache\n", total_swapcache_pages);
60 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080062 swap_cache_info.find_success, swap_cache_info.find_total);
Hugh Dickins07279cd2008-08-20 14:09:05 -070063 printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
65}
66
67/*
Nick Piggine2867812008-07-25 19:45:30 -070068 * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * but sets SwapCache flag and private instead of mapping and index.
70 */
Hugh Dickins73b12622008-02-04 22:28:50 -080071int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 int error;
74
Nick Pigginb55ed812007-10-16 01:24:42 -070075 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 BUG_ON(PageSwapCache(page));
77 BUG_ON(PagePrivate(page));
78 error = radix_tree_preload(gfp_mask);
79 if (!error) {
Nick Piggine2867812008-07-25 19:45:30 -070080 page_cache_get(page);
81 SetPageSwapCache(page);
82 set_page_private(page, entry.val);
83
Nick Piggin19fd6232008-07-25 19:45:32 -070084 spin_lock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 error = radix_tree_insert(&swapper_space.page_tree,
86 entry.val, page);
Nick Piggine2867812008-07-25 19:45:30 -070087 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 total_swapcache_pages++;
Christoph Lameter347ce432006-06-30 01:55:35 -070089 __inc_zone_page_state(page, NR_FILE_PAGES);
Hugh Dickinsbb63be02008-02-04 22:28:49 -080090 INC_CACHE_INFO(add_total);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 }
Nick Piggin19fd6232008-07-25 19:45:32 -070092 spin_unlock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 radix_tree_preload_end();
Nick Piggine2867812008-07-25 19:45:30 -070094
95 if (unlikely(error)) {
96 set_page_private(page, 0UL);
97 ClearPageSwapCache(page);
98 page_cache_release(page);
99 }
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 return error;
102}
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104/*
105 * This must be called only on pages that have
106 * been verified to be in the swap cache.
107 */
108void __delete_from_swap_cache(struct page *page)
109{
110 BUG_ON(!PageLocked(page));
111 BUG_ON(!PageSwapCache(page));
112 BUG_ON(PageWriteback(page));
Hugh Dickins3279ffd2005-09-03 15:54:43 -0700113 BUG_ON(PagePrivate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700115 radix_tree_delete(&swapper_space.page_tree, page_private(page));
116 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 ClearPageSwapCache(page);
118 total_swapcache_pages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700119 __dec_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 INC_CACHE_INFO(del_total);
121}
122
123/**
124 * add_to_swap - allocate swap space for a page
125 * @page: page we want to move to swap
Randy Dunlap76824862008-03-19 17:00:40 -0700126 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 *
128 * Allocate swap space for the page and add the page to the
129 * swap cache. Caller needs to hold the page lock.
130 */
Christoph Lameter1480a542006-01-08 01:00:53 -0800131int add_to_swap(struct page * page, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
133 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 int err;
135
Eric Sesterhenne74ca2b2006-04-01 01:25:12 +0200136 BUG_ON(!PageLocked(page));
Nick Piggin0ed361d2008-02-04 22:29:34 -0800137 BUG_ON(!PageUptodate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
139 for (;;) {
140 entry = get_swap_page();
141 if (!entry.val)
142 return 0;
143
Nick Pigginbd53b712005-05-01 08:58:37 -0700144 /*
145 * Radix-tree node allocations from PF_MEMALLOC contexts could
146 * completely exhaust the page allocator. __GFP_NOMEMALLOC
147 * stops emergency reserves from being allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 *
Nick Pigginbd53b712005-05-01 08:58:37 -0700149 * TODO: this could cause a theoretical memory reclaim
150 * deadlock in the swap out path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 /*
153 * Add it to the swap cache and mark it dirty
154 */
Hugh Dickinsf0009442008-02-04 22:28:49 -0800155 err = add_to_swap_cache(page, entry,
Christoph Lameter1480a542006-01-08 01:00:53 -0800156 gfp_mask|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 switch (err) {
159 case 0: /* Success */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 SetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 return 1;
162 case -EEXIST:
163 /* Raced with "speculative" read_swap_cache_async */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 swap_free(entry);
165 continue;
166 default:
167 /* -ENOMEM radix-tree allocation failure */
168 swap_free(entry);
169 return 0;
170 }
171 }
172}
173
174/*
175 * This must be called only on pages that have
176 * been verified to be in the swap cache and locked.
177 * It will never put the page into the free list,
178 * the caller has a reference on the page.
179 */
180void delete_from_swap_cache(struct page *page)
181{
182 swp_entry_t entry;
183
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700184 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Nick Piggin19fd6232008-07-25 19:45:32 -0700186 spin_lock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700188 spin_unlock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
190 swap_free(entry);
191 page_cache_release(page);
192}
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/*
195 * If we are the only user, then try to free up the swap cache.
196 *
197 * Its ok to check for PageSwapCache without the page lock
198 * here because we are going to recheck again inside
199 * exclusive_swap_page() _with_ the lock.
200 * - Marcelo
201 */
202static inline void free_swap_cache(struct page *page)
203{
Nick Piggin529ae9a2008-08-02 12:01:03 +0200204 if (PageSwapCache(page) && trylock_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 remove_exclusive_swap_page(page);
206 unlock_page(page);
207 }
208}
209
210/*
211 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700212 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 */
214void free_page_and_swap_cache(struct page *page)
215{
216 free_swap_cache(page);
217 page_cache_release(page);
218}
219
220/*
221 * Passed an array of pages, drop them all from swapcache and then release
222 * them. They are removed from the LRU and freed if this is their last use.
223 */
224void free_pages_and_swap_cache(struct page **pages, int nr)
225{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 struct page **pagep = pages;
227
228 lru_add_drain();
229 while (nr) {
Hugh Dickinsc484d412006-01-06 00:10:55 -0800230 int todo = min(nr, PAGEVEC_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 int i;
232
233 for (i = 0; i < todo; i++)
234 free_swap_cache(pagep[i]);
235 release_pages(pagep, todo, 0);
236 pagep += todo;
237 nr -= todo;
238 }
239}
240
241/*
242 * Lookup a swap entry in the swap cache. A found page will be returned
243 * unlocked and with its refcount incremented - we rely on the kernel
244 * lock getting page table operations atomic even if we drop the page
245 * lock before returning.
246 */
247struct page * lookup_swap_cache(swp_entry_t entry)
248{
249 struct page *page;
250
251 page = find_get_page(&swapper_space, entry.val);
252
253 if (page)
254 INC_CACHE_INFO(find_success);
255
256 INC_CACHE_INFO(find_total);
257 return page;
258}
259
260/*
261 * Locate a page of swap in physical memory, reserving swap cache space
262 * and reading the disk if it is not already cached.
263 * A failure return means that either the page allocation failed or that
264 * the swap entry is no longer in use.
265 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800266struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 struct vm_area_struct *vma, unsigned long addr)
268{
269 struct page *found_page, *new_page = NULL;
270 int err;
271
272 do {
273 /*
274 * First check the swap cache. Since this is normally
275 * called after lookup_swap_cache() failed, re-calling
276 * that would confuse statistics.
277 */
278 found_page = find_get_page(&swapper_space, entry.val);
279 if (found_page)
280 break;
281
282 /*
283 * Get a new page to read into from swap.
284 */
285 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800286 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 if (!new_page)
288 break; /* Out of memory */
289 }
290
291 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800292 * Swap entry may have been freed since our caller observed it.
293 */
294 if (!swap_duplicate(entry))
295 break;
296
297 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 * Associate the page with swap entry in the swap cache.
Hugh Dickinsf0009442008-02-04 22:28:49 -0800299 * May fail (-EEXIST) if there is already a page associated
300 * with this entry in the swap cache: added by a racing
301 * read_swap_cache_async, or add_to_swap or shmem_writepage
302 * re-using the just freed swap entry for an existing page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 * May fail (-ENOMEM) if radix-tree node allocation failed.
304 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200305 set_page_locked(new_page);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800306 err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200307 if (likely(!err)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 /*
309 * Initiate read into locked page and return.
310 */
311 lru_cache_add_active(new_page);
312 swap_readpage(NULL, new_page);
313 return new_page;
314 }
Nick Piggin529ae9a2008-08-02 12:01:03 +0200315 clear_page_locked(new_page);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800316 swap_free(entry);
317 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 if (new_page)
320 page_cache_release(new_page);
321 return found_page;
322}
Hugh Dickins46017e92008-02-04 22:28:41 -0800323
324/**
325 * swapin_readahead - swap in pages in hope we need them soon
326 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700327 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800328 * @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 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800343struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800344 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),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800362 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800363 if (!page)
364 break;
365 page_cache_release(page);
366 }
367 lru_add_drain(); /* Push any new pages onto the LRU now */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800368 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800369}