blob: 46680461785bef647c3618bc7d493d1909aea80f [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090011#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/kernel_stat.h>
13#include <linux/swap.h>
Hugh Dickins46017e92008-02-04 22:28:41 -080014#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/init.h>
16#include <linux/pagemap.h>
17#include <linux/buffer_head.h>
18#include <linux/backing-dev.h>
Hugh Dickinsc484d412006-01-06 00:10:55 -080019#include <linux/pagevec.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080020#include <linux/migrate.h>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080021#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
23#include <asm/pgtable.h>
24
25/*
26 * swapper_space is a fiction, retained to simplify the path through
Jens Axboe7eaceac2011-03-10 08:52:07 +010027 * vmscan's shrink_page_list.
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 */
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,
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 .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 = {
Jens Axboed9938312009-06-12 14:45:52 +020036 .name = "swap",
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};
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/*
Daisuke Nishimura31a56392009-09-21 17:02:50 -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 */
Daisuke Nishimura31a56392009-09-21 17:02:50 -070071static int __add_to_swap_cache(struct page *page, swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 int error;
74
Hugh Dickins51726b12009-01-06 14:39:25 -080075 VM_BUG_ON(!PageLocked(page));
76 VM_BUG_ON(PageSwapCache(page));
77 VM_BUG_ON(!PageSwapBacked(page));
78
Daisuke Nishimura31a56392009-09-21 17:02:50 -070079 page_cache_get(page);
80 SetPageSwapCache(page);
81 set_page_private(page, entry.val);
82
83 spin_lock_irq(&swapper_space.tree_lock);
84 error = radix_tree_insert(&swapper_space.page_tree, entry.val, page);
85 if (likely(!error)) {
86 total_swapcache_pages++;
87 __inc_zone_page_state(page, NR_FILE_PAGES);
88 INC_CACHE_INFO(add_total);
89 }
90 spin_unlock_irq(&swapper_space.tree_lock);
91
92 if (unlikely(error)) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -070093 /*
94 * Only the context which have set SWAP_HAS_CACHE flag
95 * would call add_to_swap_cache().
96 * So add_to_swap_cache() doesn't returns -EEXIST.
97 */
98 VM_BUG_ON(error == -EEXIST);
Daisuke Nishimura31a56392009-09-21 17:02:50 -070099 set_page_private(page, 0UL);
100 ClearPageSwapCache(page);
101 page_cache_release(page);
102 }
103
104 return error;
105}
106
107
108int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
109{
110 int error;
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 error = radix_tree_preload(gfp_mask);
113 if (!error) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700114 error = __add_to_swap_cache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 radix_tree_preload_end();
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 return error;
118}
119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120/*
121 * This must be called only on pages that have
122 * been verified to be in the swap cache.
123 */
124void __delete_from_swap_cache(struct page *page)
125{
Hugh Dickins51726b12009-01-06 14:39:25 -0800126 VM_BUG_ON(!PageLocked(page));
127 VM_BUG_ON(!PageSwapCache(page));
128 VM_BUG_ON(PageWriteback(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700130 radix_tree_delete(&swapper_space.page_tree, page_private(page));
131 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 ClearPageSwapCache(page);
133 total_swapcache_pages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700134 __dec_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 INC_CACHE_INFO(del_total);
136}
137
138/**
139 * add_to_swap - allocate swap space for a page
140 * @page: page we want to move to swap
141 *
142 * Allocate swap space for the page and add the page to the
143 * swap cache. Caller needs to hold the page lock.
144 */
Hugh Dickinsac47b002009-01-06 14:39:39 -0800145int add_to_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 int err;
149
Hugh Dickins51726b12009-01-06 14:39:25 -0800150 VM_BUG_ON(!PageLocked(page));
151 VM_BUG_ON(!PageUptodate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700153 entry = get_swap_page();
154 if (!entry.val)
155 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800157 if (unlikely(PageTransHuge(page)))
158 if (unlikely(split_huge_page(page))) {
159 swapcache_free(entry, NULL);
160 return 0;
161 }
162
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700163 /*
164 * Radix-tree node allocations from PF_MEMALLOC contexts could
165 * completely exhaust the page allocator. __GFP_NOMEMALLOC
166 * stops emergency reserves from being allocated.
167 *
168 * TODO: this could cause a theoretical memory reclaim
169 * deadlock in the swap out path.
170 */
171 /*
172 * Add it to the swap cache and mark it dirty
173 */
174 err = add_to_swap_cache(page, entry,
175 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700177 if (!err) { /* Success */
178 SetPageDirty(page);
179 return 1;
180 } else { /* -ENOMEM radix-tree allocation failure */
181 /*
182 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
183 * clear SWAP_HAS_CACHE flag.
184 */
185 swapcache_free(entry, NULL);
186 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 }
188}
189
190/*
191 * This must be called only on pages that have
192 * been verified to be in the swap cache and locked.
193 * It will never put the page into the free list,
194 * the caller has a reference on the page.
195 */
196void delete_from_swap_cache(struct page *page)
197{
198 swp_entry_t entry;
199
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700200 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Nick Piggin19fd6232008-07-25 19:45:32 -0700202 spin_lock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700204 spin_unlock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700206 swapcache_free(entry, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 page_cache_release(page);
208}
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210/*
211 * If we are the only user, then try to free up the swap cache.
212 *
213 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800214 * here because we are going to recheck again inside
215 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 * - Marcelo
217 */
218static inline void free_swap_cache(struct page *page)
219{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800220 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
221 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 unlock_page(page);
223 }
224}
225
226/*
227 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700228 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
230void free_page_and_swap_cache(struct page *page)
231{
232 free_swap_cache(page);
233 page_cache_release(page);
234}
235
236/*
237 * Passed an array of pages, drop them all from swapcache and then release
238 * them. They are removed from the LRU and freed if this is their last use.
239 */
240void free_pages_and_swap_cache(struct page **pages, int nr)
241{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 struct page **pagep = pages;
243
244 lru_add_drain();
245 while (nr) {
Hugh Dickinsc484d412006-01-06 00:10:55 -0800246 int todo = min(nr, PAGEVEC_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 int i;
248
249 for (i = 0; i < todo; i++)
250 free_swap_cache(pagep[i]);
251 release_pages(pagep, todo, 0);
252 pagep += todo;
253 nr -= todo;
254 }
255}
256
257/*
258 * Lookup a swap entry in the swap cache. A found page will be returned
259 * unlocked and with its refcount incremented - we rely on the kernel
260 * lock getting page table operations atomic even if we drop the page
261 * lock before returning.
262 */
263struct page * lookup_swap_cache(swp_entry_t entry)
264{
265 struct page *page;
266
267 page = find_get_page(&swapper_space, entry.val);
268
269 if (page)
270 INC_CACHE_INFO(find_success);
271
272 INC_CACHE_INFO(find_total);
273 return page;
274}
275
276/*
277 * Locate a page of swap in physical memory, reserving swap cache space
278 * and reading the disk if it is not already cached.
279 * A failure return means that either the page allocation failed or that
280 * the swap entry is no longer in use.
281 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800282struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 struct vm_area_struct *vma, unsigned long addr)
284{
285 struct page *found_page, *new_page = NULL;
286 int err;
287
288 do {
289 /*
290 * First check the swap cache. Since this is normally
291 * called after lookup_swap_cache() failed, re-calling
292 * that would confuse statistics.
293 */
294 found_page = find_get_page(&swapper_space, entry.val);
295 if (found_page)
296 break;
297
298 /*
299 * Get a new page to read into from swap.
300 */
301 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800302 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 if (!new_page)
304 break; /* Out of memory */
305 }
306
307 /*
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700308 * call radix_tree_preload() while we can wait.
309 */
310 err = radix_tree_preload(gfp_mask & GFP_KERNEL);
311 if (err)
312 break;
313
314 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800315 * Swap entry may have been freed since our caller observed it.
316 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700317 err = swapcache_prepare(entry);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700318 if (err == -EEXIST) { /* seems racy */
319 radix_tree_preload_end();
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700320 continue;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700321 }
322 if (err) { /* swp entry is obsolete ? */
323 radix_tree_preload_end();
Hugh Dickinsf0009442008-02-04 22:28:49 -0800324 break;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700325 }
Hugh Dickinsf0009442008-02-04 22:28:49 -0800326
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700327 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
Nick Pigginf45840b2008-10-18 20:26:57 -0700328 __set_page_locked(new_page);
Rik van Rielb2e18532008-10-18 20:26:30 -0700329 SetPageSwapBacked(new_page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700330 err = __add_to_swap_cache(new_page, entry);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200331 if (likely(!err)) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700332 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 /*
334 * Initiate read into locked page and return.
335 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700336 lru_cache_add_anon(new_page);
Minchan Kimaca8bf32009-06-16 15:33:02 -0700337 swap_readpage(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 return new_page;
339 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700340 radix_tree_preload_end();
Rik van Rielb2e18532008-10-18 20:26:30 -0700341 ClearPageSwapBacked(new_page);
Nick Pigginf45840b2008-10-18 20:26:57 -0700342 __clear_page_locked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700343 /*
344 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
345 * clear SWAP_HAS_CACHE flag.
346 */
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700347 swapcache_free(entry, NULL);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800348 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350 if (new_page)
351 page_cache_release(new_page);
352 return found_page;
353}
Hugh Dickins46017e92008-02-04 22:28:41 -0800354
355/**
356 * swapin_readahead - swap in pages in hope we need them soon
357 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700358 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800359 * @vma: user vma this address belongs to
360 * @addr: target address for mempolicy
361 *
362 * Returns the struct page for entry and addr, after queueing swapin.
363 *
364 * Primitive swap readahead code. We simply read an aligned block of
365 * (1 << page_cluster) entries in the swap area. This method is chosen
366 * because it doesn't cost us any seek time. We also make sure to queue
367 * the 'original' request together with the readahead ones...
368 *
369 * This has been extended to use the NUMA policies from the mm triggering
370 * the readahead.
371 *
372 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
373 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800374struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800375 struct vm_area_struct *vma, unsigned long addr)
376{
377 int nr_pages;
378 struct page *page;
379 unsigned long offset;
380 unsigned long end_offset;
381
382 /*
383 * Get starting offset for readaround, and number of pages to read.
384 * Adjust starting address by readbehind (for NUMA interleave case)?
385 * No, it's very unlikely that swap layout would follow vma layout,
386 * more likely that neighbouring swap pages came from the same node:
387 * so use the same "addr" to choose the same node for each swap read.
388 */
389 nr_pages = valid_swaphandles(entry, &offset);
390 for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
391 /* Ok, do the async read-ahead now */
392 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800393 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800394 if (!page)
395 break;
396 page_cache_release(page);
397 }
398 lru_add_drain(); /* Push any new pages onto the LRU now */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800399 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800400}