blob: 78cc4d1f6ccefe617698761cf198a7eb8a6db2bc [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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/mm.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090010#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#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 Hiroyuki8c7c6e342009-01-07 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
Jens Axboe7eaceac2011-03-10 08:52:07 +010026 * vmscan's shrink_page_list.
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 */
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,
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 .set_page_dirty = __set_page_dirty_nobuffers,
Christoph Lametere965f962006-02-01 03:05:41 -080031 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -070032};
33
34static struct backing_dev_info swap_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020035 .name = "swap",
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -070037};
38
39struct address_space swapper_space = {
40 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
Nick Piggin19fd6232008-07-25 19:45:32 -070041 .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 .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 Torvalds1da177e2005-04-16 15:20:36 -070046
47#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
48
49static struct {
50 unsigned long add_total;
51 unsigned long del_total;
52 unsigned long find_success;
53 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054} swap_cache_info;
55
56void show_swap_cache_info(void)
57{
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070058 printk("%lu pages in swap cache\n", total_swapcache_pages);
59 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080061 swap_cache_info.find_success, swap_cache_info.find_total);
Hugh Dickins07279cd2008-08-20 14:09:05 -070062 printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
64}
65
66/*
Daisuke Nishimura31a56392009-09-21 17:02:50 -070067 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * but sets SwapCache flag and private instead of mapping and index.
69 */
Daisuke Nishimura31a56392009-09-21 17:02:50 -070070static int __add_to_swap_cache(struct page *page, swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
72 int error;
73
Hugh Dickins51726b12009-01-06 14:39:25 -080074 VM_BUG_ON(!PageLocked(page));
75 VM_BUG_ON(PageSwapCache(page));
76 VM_BUG_ON(!PageSwapBacked(page));
77
Daisuke Nishimura31a56392009-09-21 17:02:50 -070078 page_cache_get(page);
79 SetPageSwapCache(page);
80 set_page_private(page, entry.val);
81
82 spin_lock_irq(&swapper_space.tree_lock);
83 error = radix_tree_insert(&swapper_space.page_tree, entry.val, page);
84 if (likely(!error)) {
85 total_swapcache_pages++;
86 __inc_zone_page_state(page, NR_FILE_PAGES);
87 INC_CACHE_INFO(add_total);
88 }
89 spin_unlock_irq(&swapper_space.tree_lock);
90
91 if (unlikely(error)) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -070092 /*
93 * Only the context which have set SWAP_HAS_CACHE flag
94 * would call add_to_swap_cache().
95 * So add_to_swap_cache() doesn't returns -EEXIST.
96 */
97 VM_BUG_ON(error == -EEXIST);
Daisuke Nishimura31a56392009-09-21 17:02:50 -070098 set_page_private(page, 0UL);
99 ClearPageSwapCache(page);
100 page_cache_release(page);
101 }
102
103 return error;
104}
105
106
107int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
108{
109 int error;
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 error = radix_tree_preload(gfp_mask);
112 if (!error) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700113 error = __add_to_swap_cache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 radix_tree_preload_end();
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 return error;
117}
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119/*
120 * This must be called only on pages that have
121 * been verified to be in the swap cache.
122 */
123void __delete_from_swap_cache(struct page *page)
124{
Hugh Dickins51726b12009-01-06 14:39:25 -0800125 VM_BUG_ON(!PageLocked(page));
126 VM_BUG_ON(!PageSwapCache(page));
127 VM_BUG_ON(PageWriteback(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700129 radix_tree_delete(&swapper_space.page_tree, page_private(page));
130 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 ClearPageSwapCache(page);
132 total_swapcache_pages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700133 __dec_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 INC_CACHE_INFO(del_total);
135}
136
137/**
138 * add_to_swap - allocate swap space for a page
139 * @page: page we want to move to swap
140 *
141 * Allocate swap space for the page and add the page to the
142 * swap cache. Caller needs to hold the page lock.
143 */
Hugh Dickinsac47b002009-01-06 14:39:39 -0800144int add_to_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145{
146 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 int err;
148
Hugh Dickins51726b12009-01-06 14:39:25 -0800149 VM_BUG_ON(!PageLocked(page));
150 VM_BUG_ON(!PageUptodate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700152 entry = get_swap_page();
153 if (!entry.val)
154 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800156 if (unlikely(PageTransHuge(page)))
157 if (unlikely(split_huge_page(page))) {
158 swapcache_free(entry, NULL);
159 return 0;
160 }
161
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700162 /*
163 * Radix-tree node allocations from PF_MEMALLOC contexts could
164 * completely exhaust the page allocator. __GFP_NOMEMALLOC
165 * stops emergency reserves from being allocated.
166 *
167 * TODO: this could cause a theoretical memory reclaim
168 * deadlock in the swap out path.
169 */
170 /*
171 * Add it to the swap cache and mark it dirty
172 */
173 err = add_to_swap_cache(page, entry,
174 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700176 if (!err) { /* Success */
177 SetPageDirty(page);
178 return 1;
179 } else { /* -ENOMEM radix-tree allocation failure */
180 /*
181 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
182 * clear SWAP_HAS_CACHE flag.
183 */
184 swapcache_free(entry, NULL);
185 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
187}
188
189/*
190 * This must be called only on pages that have
191 * been verified to be in the swap cache and locked.
192 * It will never put the page into the free list,
193 * the caller has a reference on the page.
194 */
195void delete_from_swap_cache(struct page *page)
196{
197 swp_entry_t entry;
198
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700199 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Nick Piggin19fd6232008-07-25 19:45:32 -0700201 spin_lock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700203 spin_unlock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700205 swapcache_free(entry, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 page_cache_release(page);
207}
208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209/*
210 * If we are the only user, then try to free up the swap cache.
211 *
212 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800213 * here because we are going to recheck again inside
214 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 * - Marcelo
216 */
217static inline void free_swap_cache(struct page *page)
218{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800219 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
220 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 unlock_page(page);
222 }
223}
224
225/*
226 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700227 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 */
229void free_page_and_swap_cache(struct page *page)
230{
231 free_swap_cache(page);
232 page_cache_release(page);
233}
234
235/*
236 * Passed an array of pages, drop them all from swapcache and then release
237 * them. They are removed from the LRU and freed if this is their last use.
238 */
239void free_pages_and_swap_cache(struct page **pages, int nr)
240{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 struct page **pagep = pages;
242
243 lru_add_drain();
244 while (nr) {
Hugh Dickinsc484d412006-01-06 00:10:55 -0800245 int todo = min(nr, PAGEVEC_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 int i;
247
248 for (i = 0; i < todo; i++)
249 free_swap_cache(pagep[i]);
250 release_pages(pagep, todo, 0);
251 pagep += todo;
252 nr -= todo;
253 }
254}
255
256/*
257 * Lookup a swap entry in the swap cache. A found page will be returned
258 * unlocked and with its refcount incremented - we rely on the kernel
259 * lock getting page table operations atomic even if we drop the page
260 * lock before returning.
261 */
262struct page * lookup_swap_cache(swp_entry_t entry)
263{
264 struct page *page;
265
266 page = find_get_page(&swapper_space, entry.val);
267
268 if (page)
269 INC_CACHE_INFO(find_success);
270
271 INC_CACHE_INFO(find_total);
272 return page;
273}
274
275/*
276 * Locate a page of swap in physical memory, reserving swap cache space
277 * and reading the disk if it is not already cached.
278 * A failure return means that either the page allocation failed or that
279 * the swap entry is no longer in use.
280 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800281struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 struct vm_area_struct *vma, unsigned long addr)
283{
284 struct page *found_page, *new_page = NULL;
285 int err;
286
287 do {
288 /*
289 * First check the swap cache. Since this is normally
290 * called after lookup_swap_cache() failed, re-calling
291 * that would confuse statistics.
292 */
293 found_page = find_get_page(&swapper_space, entry.val);
294 if (found_page)
295 break;
296
297 /*
298 * Get a new page to read into from swap.
299 */
300 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800301 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 if (!new_page)
303 break; /* Out of memory */
304 }
305
306 /*
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700307 * call radix_tree_preload() while we can wait.
308 */
309 err = radix_tree_preload(gfp_mask & GFP_KERNEL);
310 if (err)
311 break;
312
313 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800314 * Swap entry may have been freed since our caller observed it.
315 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700316 err = swapcache_prepare(entry);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700317 if (err == -EEXIST) { /* seems racy */
318 radix_tree_preload_end();
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700319 continue;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700320 }
321 if (err) { /* swp entry is obsolete ? */
322 radix_tree_preload_end();
Hugh Dickinsf0009442008-02-04 22:28:49 -0800323 break;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700324 }
Hugh Dickinsf0009442008-02-04 22:28:49 -0800325
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700326 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
Nick Pigginf45840b2008-10-18 20:26:57 -0700327 __set_page_locked(new_page);
Rik van Rielb2e18532008-10-18 20:26:30 -0700328 SetPageSwapBacked(new_page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700329 err = __add_to_swap_cache(new_page, entry);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200330 if (likely(!err)) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700331 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 /*
333 * Initiate read into locked page and return.
334 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700335 lru_cache_add_anon(new_page);
Minchan Kimaca8bf32009-06-16 15:33:02 -0700336 swap_readpage(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return new_page;
338 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700339 radix_tree_preload_end();
Rik van Rielb2e18532008-10-18 20:26:30 -0700340 ClearPageSwapBacked(new_page);
Nick Pigginf45840b2008-10-18 20:26:57 -0700341 __clear_page_locked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700342 /*
343 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
344 * clear SWAP_HAS_CACHE flag.
345 */
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700346 swapcache_free(entry, NULL);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800347 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 if (new_page)
350 page_cache_release(new_page);
351 return found_page;
352}
Hugh Dickins46017e92008-02-04 22:28:41 -0800353
354/**
355 * swapin_readahead - swap in pages in hope we need them soon
356 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700357 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800358 * @vma: user vma this address belongs to
359 * @addr: target address for mempolicy
360 *
361 * Returns the struct page for entry and addr, after queueing swapin.
362 *
363 * Primitive swap readahead code. We simply read an aligned block of
364 * (1 << page_cluster) entries in the swap area. This method is chosen
365 * because it doesn't cost us any seek time. We also make sure to queue
366 * the 'original' request together with the readahead ones...
367 *
368 * This has been extended to use the NUMA policies from the mm triggering
369 * the readahead.
370 *
371 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
372 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800373struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800374 struct vm_area_struct *vma, unsigned long addr)
375{
376 int nr_pages;
377 struct page *page;
378 unsigned long offset;
379 unsigned long end_offset;
380
381 /*
382 * Get starting offset for readaround, and number of pages to read.
383 * Adjust starting address by readbehind (for NUMA interleave case)?
384 * No, it's very unlikely that swap layout would follow vma layout,
385 * more likely that neighbouring swap pages came from the same node:
386 * so use the same "addr" to choose the same node for each swap read.
387 */
388 nr_pages = valid_swaphandles(entry, &offset);
389 for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
390 /* Ok, do the async read-ahead now */
391 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800392 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800393 if (!page)
394 break;
395 page_cache_release(page);
396 }
397 lru_add_drain(); /* Push any new pages onto the LRU now */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800398 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800399}