blob: 470038a9187383790751817cda61a07cb50f9512 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/backing-dev.h>
Hugh Dickinsc484d412006-01-06 00:10:55 -080017#include <linux/pagevec.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080018#include <linux/migrate.h>
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -080019#include <linux/page_cgroup.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
Jens Axboe7eaceac2011-03-10 08:52:07 +010025 * vmscan's shrink_page_list.
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 */
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070027static const struct address_space_operations swap_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 .writepage = swap_writepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 .set_page_dirty = __set_page_dirty_nobuffers,
Christoph Lametere965f962006-02-01 03:05:41 -080030 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -070031};
32
33static struct backing_dev_info swap_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020034 .name = "swap",
Rik van Riel4f98a2f2008-10-18 20:26:32 -070035 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -070036};
37
38struct address_space swapper_space = {
39 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
Nick Piggin19fd6232008-07-25 19:45:32 -070040 .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 .a_ops = &swap_aops,
42 .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
43 .backing_dev_info = &swap_backing_dev_info,
44};
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
47
48static struct {
49 unsigned long add_total;
50 unsigned long del_total;
51 unsigned long find_success;
52 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053} swap_cache_info;
54
55void show_swap_cache_info(void)
56{
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070057 printk("%lu pages in swap cache\n", total_swapcache_pages);
58 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080060 swap_cache_info.find_success, swap_cache_info.find_total);
Hugh Dickins07279cd2008-08-20 14:09:05 -070061 printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
63}
64
65/*
Daisuke Nishimura31a56392009-09-21 17:02:50 -070066 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * but sets SwapCache flag and private instead of mapping and index.
68 */
Daisuke Nishimura31a56392009-09-21 17:02:50 -070069static int __add_to_swap_cache(struct page *page, swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070{
71 int error;
72
Hugh Dickins51726b12009-01-06 14:39:25 -080073 VM_BUG_ON(!PageLocked(page));
74 VM_BUG_ON(PageSwapCache(page));
75 VM_BUG_ON(!PageSwapBacked(page));
76
Daisuke Nishimura31a56392009-09-21 17:02:50 -070077 page_cache_get(page);
78 SetPageSwapCache(page);
79 set_page_private(page, entry.val);
80
81 spin_lock_irq(&swapper_space.tree_lock);
82 error = radix_tree_insert(&swapper_space.page_tree, entry.val, page);
83 if (likely(!error)) {
84 total_swapcache_pages++;
85 __inc_zone_page_state(page, NR_FILE_PAGES);
86 INC_CACHE_INFO(add_total);
87 }
88 spin_unlock_irq(&swapper_space.tree_lock);
89
90 if (unlikely(error)) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -070091 /*
92 * Only the context which have set SWAP_HAS_CACHE flag
93 * would call add_to_swap_cache().
94 * So add_to_swap_cache() doesn't returns -EEXIST.
95 */
96 VM_BUG_ON(error == -EEXIST);
Daisuke Nishimura31a56392009-09-21 17:02:50 -070097 set_page_private(page, 0UL);
98 ClearPageSwapCache(page);
99 page_cache_release(page);
100 }
101
102 return error;
103}
104
105
106int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
107{
108 int error;
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 error = radix_tree_preload(gfp_mask);
111 if (!error) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700112 error = __add_to_swap_cache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 radix_tree_preload_end();
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 return error;
116}
117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118/*
119 * This must be called only on pages that have
120 * been verified to be in the swap cache.
121 */
122void __delete_from_swap_cache(struct page *page)
123{
Hugh Dickins51726b12009-01-06 14:39:25 -0800124 VM_BUG_ON(!PageLocked(page));
125 VM_BUG_ON(!PageSwapCache(page));
126 VM_BUG_ON(PageWriteback(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700128 radix_tree_delete(&swapper_space.page_tree, page_private(page));
129 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 ClearPageSwapCache(page);
131 total_swapcache_pages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700132 __dec_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 INC_CACHE_INFO(del_total);
134}
135
136/**
137 * add_to_swap - allocate swap space for a page
138 * @page: page we want to move to swap
139 *
140 * Allocate swap space for the page and add the page to the
141 * swap cache. Caller needs to hold the page lock.
142 */
Hugh Dickinsac47b002009-01-06 14:39:39 -0800143int add_to_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
145 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 int err;
147
Hugh Dickins51726b12009-01-06 14:39:25 -0800148 VM_BUG_ON(!PageLocked(page));
149 VM_BUG_ON(!PageUptodate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700151 entry = get_swap_page();
152 if (!entry.val)
153 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800155 if (unlikely(PageTransHuge(page)))
156 if (unlikely(split_huge_page(page))) {
157 swapcache_free(entry, NULL);
158 return 0;
159 }
160
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700161 /*
162 * Radix-tree node allocations from PF_MEMALLOC contexts could
163 * completely exhaust the page allocator. __GFP_NOMEMALLOC
164 * stops emergency reserves from being allocated.
165 *
166 * TODO: this could cause a theoretical memory reclaim
167 * deadlock in the swap out path.
168 */
169 /*
170 * Add it to the swap cache and mark it dirty
171 */
172 err = add_to_swap_cache(page, entry,
173 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700175 if (!err) { /* Success */
176 SetPageDirty(page);
177 return 1;
178 } else { /* -ENOMEM radix-tree allocation failure */
179 /*
180 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
181 * clear SWAP_HAS_CACHE flag.
182 */
183 swapcache_free(entry, NULL);
184 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
186}
187
188/*
189 * This must be called only on pages that have
190 * been verified to be in the swap cache and locked.
191 * It will never put the page into the free list,
192 * the caller has a reference on the page.
193 */
194void delete_from_swap_cache(struct page *page)
195{
196 swp_entry_t entry;
197
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700198 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
Nick Piggin19fd6232008-07-25 19:45:32 -0700200 spin_lock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 __delete_from_swap_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700202 spin_unlock_irq(&swapper_space.tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700204 swapcache_free(entry, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 page_cache_release(page);
206}
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208/*
209 * If we are the only user, then try to free up the swap cache.
210 *
211 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800212 * here because we are going to recheck again inside
213 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 * - Marcelo
215 */
216static inline void free_swap_cache(struct page *page)
217{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800218 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
219 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 unlock_page(page);
221 }
222}
223
224/*
225 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700226 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 */
228void free_page_and_swap_cache(struct page *page)
229{
230 free_swap_cache(page);
231 page_cache_release(page);
232}
233
234/*
235 * Passed an array of pages, drop them all from swapcache and then release
236 * them. They are removed from the LRU and freed if this is their last use.
237 */
238void free_pages_and_swap_cache(struct page **pages, int nr)
239{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 struct page **pagep = pages;
241
242 lru_add_drain();
243 while (nr) {
Hugh Dickinsc484d412006-01-06 00:10:55 -0800244 int todo = min(nr, PAGEVEC_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 int i;
246
247 for (i = 0; i < todo; i++)
248 free_swap_cache(pagep[i]);
249 release_pages(pagep, todo, 0);
250 pagep += todo;
251 nr -= todo;
252 }
253}
254
255/*
256 * Lookup a swap entry in the swap cache. A found page will be returned
257 * unlocked and with its refcount incremented - we rely on the kernel
258 * lock getting page table operations atomic even if we drop the page
259 * lock before returning.
260 */
261struct page * lookup_swap_cache(swp_entry_t entry)
262{
263 struct page *page;
264
265 page = find_get_page(&swapper_space, entry.val);
266
267 if (page)
268 INC_CACHE_INFO(find_success);
269
270 INC_CACHE_INFO(find_total);
271 return page;
272}
273
274/*
275 * Locate a page of swap in physical memory, reserving swap cache space
276 * and reading the disk if it is not already cached.
277 * A failure return means that either the page allocation failed or that
278 * the swap entry is no longer in use.
279 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800280struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 struct vm_area_struct *vma, unsigned long addr)
282{
283 struct page *found_page, *new_page = NULL;
284 int err;
285
286 do {
287 /*
288 * First check the swap cache. Since this is normally
289 * called after lookup_swap_cache() failed, re-calling
290 * that would confuse statistics.
291 */
292 found_page = find_get_page(&swapper_space, entry.val);
293 if (found_page)
294 break;
295
296 /*
297 * Get a new page to read into from swap.
298 */
299 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800300 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 if (!new_page)
302 break; /* Out of memory */
KAMEZAWA Hiroyuki4e5f01c2012-01-12 17:18:58 -0800303 /*
304 * The memcg-specific accounting when moving
305 * pages around the LRU lists relies on the
306 * page's owner (memcg) to be valid. Usually,
307 * pages are assigned to a new owner before
308 * being put on the LRU list, but since this
309 * is not the case here, the stale owner from
310 * a previous allocation cycle must be reset.
311 */
312 mem_cgroup_reset_owner(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 }
314
315 /*
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700316 * call radix_tree_preload() while we can wait.
317 */
318 err = radix_tree_preload(gfp_mask & GFP_KERNEL);
319 if (err)
320 break;
321
322 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800323 * Swap entry may have been freed since our caller observed it.
324 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700325 err = swapcache_prepare(entry);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700326 if (err == -EEXIST) { /* seems racy */
327 radix_tree_preload_end();
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700328 continue;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700329 }
330 if (err) { /* swp entry is obsolete ? */
331 radix_tree_preload_end();
Hugh Dickinsf0009442008-02-04 22:28:49 -0800332 break;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700333 }
Hugh Dickinsf0009442008-02-04 22:28:49 -0800334
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700335 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
Nick Pigginf45840b2008-10-18 20:26:57 -0700336 __set_page_locked(new_page);
Rik van Rielb2e18532008-10-18 20:26:30 -0700337 SetPageSwapBacked(new_page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700338 err = __add_to_swap_cache(new_page, entry);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200339 if (likely(!err)) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700340 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 /*
342 * Initiate read into locked page and return.
343 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700344 lru_cache_add_anon(new_page);
Minchan Kimaca8bf32009-06-16 15:33:02 -0700345 swap_readpage(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 return new_page;
347 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700348 radix_tree_preload_end();
Rik van Rielb2e18532008-10-18 20:26:30 -0700349 ClearPageSwapBacked(new_page);
Nick Pigginf45840b2008-10-18 20:26:57 -0700350 __clear_page_locked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700351 /*
352 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
353 * clear SWAP_HAS_CACHE flag.
354 */
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700355 swapcache_free(entry, NULL);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800356 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 if (new_page)
359 page_cache_release(new_page);
360 return found_page;
361}
Hugh Dickins46017e92008-02-04 22:28:41 -0800362
363/**
364 * swapin_readahead - swap in pages in hope we need them soon
365 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700366 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800367 * @vma: user vma this address belongs to
368 * @addr: target address for mempolicy
369 *
370 * Returns the struct page for entry and addr, after queueing swapin.
371 *
372 * Primitive swap readahead code. We simply read an aligned block of
373 * (1 << page_cluster) entries in the swap area. This method is chosen
374 * because it doesn't cost us any seek time. We also make sure to queue
375 * the 'original' request together with the readahead ones...
376 *
377 * This has been extended to use the NUMA policies from the mm triggering
378 * the readahead.
379 *
380 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
381 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800382struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800383 struct vm_area_struct *vma, unsigned long addr)
384{
385 int nr_pages;
386 struct page *page;
387 unsigned long offset;
388 unsigned long end_offset;
389
390 /*
391 * Get starting offset for readaround, and number of pages to read.
392 * Adjust starting address by readbehind (for NUMA interleave case)?
393 * No, it's very unlikely that swap layout would follow vma layout,
394 * more likely that neighbouring swap pages came from the same node:
395 * so use the same "addr" to choose the same node for each swap read.
396 */
397 nr_pages = valid_swaphandles(entry, &offset);
398 for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
399 /* Ok, do the async read-ahead now */
400 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800401 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800402 if (!page)
403 break;
404 page_cache_release(page);
405 }
406 lru_add_drain(); /* Push any new pages onto the LRU now */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800407 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800408}