blob: 1416e7e9e02db3b5da60fe3a270e98e7b84024bb [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>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080020#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22#include <asm/pgtable.h>
23
24/*
25 * swapper_space is a fiction, retained to simplify the path through
Anderson Briglia2706a1b2007-07-15 23:38:09 -070026 * vmscan's shrink_page_list, to make sync_page look nicer, and to allow
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 * future use of radix_tree tags in the swap cache.
28 */
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070029static const struct address_space_operations swap_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 .writepage = swap_writepage,
31 .sync_page = block_sync_page,
32 .set_page_dirty = __set_page_dirty_nobuffers,
Christoph Lametere965f962006-02-01 03:05:41 -080033 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -070034};
35
36static struct backing_dev_info swap_backing_dev_info = {
Rik van Riel4f98a2f2008-10-18 20:26:32 -070037 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 .unplug_io_fn = swap_unplug_io_fn,
39};
40
41struct address_space swapper_space = {
42 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
Nick Piggin19fd6232008-07-25 19:45:32 -070043 .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 .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};
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
50
51static struct {
52 unsigned long add_total;
53 unsigned long del_total;
54 unsigned long find_success;
55 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056} swap_cache_info;
57
58void show_swap_cache_info(void)
59{
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070060 printk("%lu pages in swap cache\n", total_swapcache_pages);
61 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080063 swap_cache_info.find_success, swap_cache_info.find_total);
Hugh Dickins07279cd2008-08-20 14:09:05 -070064 printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
66}
67
68/*
Nick Piggine2867812008-07-25 19:45:30 -070069 * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 * but sets SwapCache flag and private instead of mapping and index.
71 */
Hugh Dickins73b12622008-02-04 22:28:50 -080072int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073{
74 int error;
75
Hugh Dickins51726b12009-01-06 14:39:25 -080076 VM_BUG_ON(!PageLocked(page));
77 VM_BUG_ON(PageSwapCache(page));
78 VM_BUG_ON(!PageSwapBacked(page));
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 error = radix_tree_preload(gfp_mask);
81 if (!error) {
Nick Piggine2867812008-07-25 19:45:30 -070082 page_cache_get(page);
83 SetPageSwapCache(page);
84 set_page_private(page, entry.val);
85
Nick Piggin19fd6232008-07-25 19:45:32 -070086 spin_lock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 error = radix_tree_insert(&swapper_space.page_tree,
88 entry.val, page);
Nick Piggine2867812008-07-25 19:45:30 -070089 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 total_swapcache_pages++;
Christoph Lameter347ce432006-06-30 01:55:35 -070091 __inc_zone_page_state(page, NR_FILE_PAGES);
Hugh Dickinsbb63be02008-02-04 22:28:49 -080092 INC_CACHE_INFO(add_total);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 }
Nick Piggin19fd6232008-07-25 19:45:32 -070094 spin_unlock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 radix_tree_preload_end();
Nick Piggine2867812008-07-25 19:45:30 -070096
97 if (unlikely(error)) {
98 set_page_private(page, 0UL);
99 ClearPageSwapCache(page);
100 page_cache_release(page);
101 }
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800102 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 return error;
104}
105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106/*
107 * This must be called only on pages that have
108 * been verified to be in the swap cache.
109 */
110void __delete_from_swap_cache(struct page *page)
111{
Hugh Dickins51726b12009-01-06 14:39:25 -0800112 VM_BUG_ON(!PageLocked(page));
113 VM_BUG_ON(!PageSwapCache(page));
114 VM_BUG_ON(PageWriteback(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700116 radix_tree_delete(&swapper_space.page_tree, page_private(page));
117 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 ClearPageSwapCache(page);
119 total_swapcache_pages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700120 __dec_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 INC_CACHE_INFO(del_total);
122}
123
124/**
125 * add_to_swap - allocate swap space for a page
126 * @page: page we want to move to swap
Randy Dunlap76824862008-03-19 17:00:40 -0700127 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 *
129 * Allocate swap space for the page and add the page to the
130 * swap cache. Caller needs to hold the page lock.
131 */
Hugh Dickinsac47b002009-01-06 14:39:39 -0800132int add_to_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133{
134 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 int err;
136
Hugh Dickins51726b12009-01-06 14:39:25 -0800137 VM_BUG_ON(!PageLocked(page));
138 VM_BUG_ON(!PageUptodate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
140 for (;;) {
141 entry = get_swap_page();
142 if (!entry.val)
143 return 0;
144
Nick Pigginbd53b712005-05-01 08:58:37 -0700145 /*
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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 *
Nick Pigginbd53b712005-05-01 08:58:37 -0700150 * TODO: this could cause a theoretical memory reclaim
151 * deadlock in the swap out path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 /*
154 * Add it to the swap cache and mark it dirty
155 */
Hugh Dickinsf0009442008-02-04 22:28:49 -0800156 err = add_to_swap_cache(page, entry,
Hugh Dickinsac47b002009-01-06 14:39:39 -0800157 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159 switch (err) {
160 case 0: /* Success */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 SetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 return 1;
163 case -EEXIST:
164 /* Raced with "speculative" read_swap_cache_async */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 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 */
181void delete_from_swap_cache(struct page *page)
182{
183 swp_entry_t entry;
184
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700185 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Nick Piggin19fd6232008-07-25 19:45:32 -0700187 spin_lock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700189 spin_unlock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700191 mem_cgroup_uncharge_swapcache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 swap_free(entry);
193 page_cache_release(page);
194}
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196/*
197 * If we are the only user, then try to free up the swap cache.
198 *
199 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800200 * here because we are going to recheck again inside
201 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * - Marcelo
203 */
204static inline void free_swap_cache(struct page *page)
205{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800206 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
207 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 unlock_page(page);
209 }
210}
211
212/*
213 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700214 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
216void free_page_and_swap_cache(struct page *page)
217{
218 free_swap_cache(page);
219 page_cache_release(page);
220}
221
222/*
223 * Passed an array of pages, drop them all from swapcache and then release
224 * them. They are removed from the LRU and freed if this is their last use.
225 */
226void free_pages_and_swap_cache(struct page **pages, int nr)
227{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 struct page **pagep = pages;
229
230 lru_add_drain();
231 while (nr) {
Hugh Dickinsc484d412006-01-06 00:10:55 -0800232 int todo = min(nr, PAGEVEC_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 int i;
234
235 for (i = 0; i < todo; i++)
236 free_swap_cache(pagep[i]);
237 release_pages(pagep, todo, 0);
238 pagep += todo;
239 nr -= todo;
240 }
241}
242
243/*
244 * Lookup a swap entry in the swap cache. A found page will be returned
245 * unlocked and with its refcount incremented - we rely on the kernel
246 * lock getting page table operations atomic even if we drop the page
247 * lock before returning.
248 */
249struct page * lookup_swap_cache(swp_entry_t entry)
250{
251 struct page *page;
252
253 page = find_get_page(&swapper_space, entry.val);
254
255 if (page)
256 INC_CACHE_INFO(find_success);
257
258 INC_CACHE_INFO(find_total);
259 return page;
260}
261
262/*
263 * Locate a page of swap in physical memory, reserving swap cache space
264 * and reading the disk if it is not already cached.
265 * A failure return means that either the page allocation failed or that
266 * the swap entry is no longer in use.
267 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800268struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 struct vm_area_struct *vma, unsigned long addr)
270{
271 struct page *found_page, *new_page = NULL;
272 int err;
273
274 do {
275 /*
276 * First check the swap cache. Since this is normally
277 * called after lookup_swap_cache() failed, re-calling
278 * that would confuse statistics.
279 */
280 found_page = find_get_page(&swapper_space, entry.val);
281 if (found_page)
282 break;
283
284 /*
285 * Get a new page to read into from swap.
286 */
287 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800288 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 if (!new_page)
290 break; /* Out of memory */
291 }
292
293 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800294 * Swap entry may have been freed since our caller observed it.
295 */
296 if (!swap_duplicate(entry))
297 break;
298
299 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 * Associate the page with swap entry in the swap cache.
Hugh Dickinsf0009442008-02-04 22:28:49 -0800301 * May fail (-EEXIST) if there is already a page associated
302 * with this entry in the swap cache: added by a racing
303 * read_swap_cache_async, or add_to_swap or shmem_writepage
304 * re-using the just freed swap entry for an existing page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 * May fail (-ENOMEM) if radix-tree node allocation failed.
306 */
Nick Pigginf45840b2008-10-18 20:26:57 -0700307 __set_page_locked(new_page);
Rik van Rielb2e18532008-10-18 20:26:30 -0700308 SetPageSwapBacked(new_page);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800309 err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200310 if (likely(!err)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 /*
312 * Initiate read into locked page and return.
313 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700314 lru_cache_add_anon(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 swap_readpage(NULL, new_page);
316 return new_page;
317 }
Rik van Rielb2e18532008-10-18 20:26:30 -0700318 ClearPageSwapBacked(new_page);
Nick Pigginf45840b2008-10-18 20:26:57 -0700319 __clear_page_locked(new_page);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800320 swap_free(entry);
321 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323 if (new_page)
324 page_cache_release(new_page);
325 return found_page;
326}
Hugh Dickins46017e92008-02-04 22:28:41 -0800327
328/**
329 * swapin_readahead - swap in pages in hope we need them soon
330 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700331 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800332 * @vma: user vma this address belongs to
333 * @addr: target address for mempolicy
334 *
335 * Returns the struct page for entry and addr, after queueing swapin.
336 *
337 * Primitive swap readahead code. We simply read an aligned block of
338 * (1 << page_cluster) entries in the swap area. This method is chosen
339 * because it doesn't cost us any seek time. We also make sure to queue
340 * the 'original' request together with the readahead ones...
341 *
342 * This has been extended to use the NUMA policies from the mm triggering
343 * the readahead.
344 *
345 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
346 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800347struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800348 struct vm_area_struct *vma, unsigned long addr)
349{
350 int nr_pages;
351 struct page *page;
352 unsigned long offset;
353 unsigned long end_offset;
354
355 /*
356 * Get starting offset for readaround, and number of pages to read.
357 * Adjust starting address by readbehind (for NUMA interleave case)?
358 * No, it's very unlikely that swap layout would follow vma layout,
359 * more likely that neighbouring swap pages came from the same node:
360 * so use the same "addr" to choose the same node for each swap read.
361 */
362 nr_pages = valid_swaphandles(entry, &offset);
363 for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
364 /* Ok, do the async read-ahead now */
365 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800366 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800367 if (!page)
368 break;
369 page_cache_release(page);
370 }
371 lru_add_drain(); /* Push any new pages onto the LRU now */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800372 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800373}