blob: 539b8885e3d1d4942dbb905b60b87d40726e3de5 [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>
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -070017#include <linux/blkdev.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>
Huang, Ying4b3ef9d2017-02-22 15:45:26 -080020#include <linux/vmalloc.h>
Tim Chen67afa382017-02-22 15:45:39 -080021#include <linux/swap_slots.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,
Mel Gorman62c230b2012-07-31 16:44:55 -070031 .set_page_dirty = swap_set_page_dirty,
Andrew Morton1c939232014-10-09 15:27:59 -070032#ifdef CONFIG_MIGRATION
Christoph Lametere965f962006-02-01 03:05:41 -080033 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -070034#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070035};
36
Huang, Ying4b3ef9d2017-02-22 15:45:26 -080037struct address_space *swapper_spaces[MAX_SWAPFILES];
38static unsigned int nr_swapper_spaces[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
41
42static struct {
43 unsigned long add_total;
44 unsigned long del_total;
45 unsigned long find_success;
46 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047} swap_cache_info;
48
Shaohua Li33806f02013-02-22 16:34:37 -080049unsigned long total_swapcache_pages(void)
50{
Huang, Ying4b3ef9d2017-02-22 15:45:26 -080051 unsigned int i, j, nr;
Shaohua Li33806f02013-02-22 16:34:37 -080052 unsigned long ret = 0;
Huang, Ying4b3ef9d2017-02-22 15:45:26 -080053 struct address_space *spaces;
Shaohua Li33806f02013-02-22 16:34:37 -080054
Huang, Ying4b3ef9d2017-02-22 15:45:26 -080055 rcu_read_lock();
56 for (i = 0; i < MAX_SWAPFILES; i++) {
57 /*
58 * The corresponding entries in nr_swapper_spaces and
59 * swapper_spaces will be reused only after at least
60 * one grace period. So it is impossible for them
61 * belongs to different usage.
62 */
63 nr = nr_swapper_spaces[i];
64 spaces = rcu_dereference(swapper_spaces[i]);
65 if (!nr || !spaces)
66 continue;
67 for (j = 0; j < nr; j++)
68 ret += spaces[j].nrpages;
69 }
70 rcu_read_unlock();
Shaohua Li33806f02013-02-22 16:34:37 -080071 return ret;
72}
73
Shaohua Li579f8292014-02-06 12:04:21 -080074static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076void show_swap_cache_info(void)
77{
Shaohua Li33806f02013-02-22 16:34:37 -080078 printk("%lu pages in swap cache\n", total_swapcache_pages());
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070079 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080081 swap_cache_info.find_success, swap_cache_info.find_total);
Shaohua Liec8acf22013-02-22 16:34:38 -080082 printk("Free swap = %ldkB\n",
83 get_nr_swap_pages() << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
85}
86
87/*
Daisuke Nishimura31a56392009-09-21 17:02:50 -070088 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * but sets SwapCache flag and private instead of mapping and index.
90 */
Seth Jennings2f772e62013-04-29 15:08:34 -070091int __add_to_swap_cache(struct page *page, swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 int error;
Shaohua Li33806f02013-02-22 16:34:37 -080094 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Sasha Levin309381fea2014-01-23 15:52:54 -080096 VM_BUG_ON_PAGE(!PageLocked(page), page);
97 VM_BUG_ON_PAGE(PageSwapCache(page), page);
98 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins51726b12009-01-06 14:39:25 -080099
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300100 get_page(page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700101 SetPageSwapCache(page);
102 set_page_private(page, entry.val);
103
Shaohua Li33806f02013-02-22 16:34:37 -0800104 address_space = swap_address_space(entry);
105 spin_lock_irq(&address_space->tree_lock);
106 error = radix_tree_insert(&address_space->page_tree,
Huang Yingf6ab1f72016-10-07 17:00:21 -0700107 swp_offset(entry), page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700108 if (likely(!error)) {
Shaohua Li33806f02013-02-22 16:34:37 -0800109 address_space->nrpages++;
Mel Gorman11fb9982016-07-28 15:46:20 -0700110 __inc_node_page_state(page, NR_FILE_PAGES);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700111 INC_CACHE_INFO(add_total);
112 }
Shaohua Li33806f02013-02-22 16:34:37 -0800113 spin_unlock_irq(&address_space->tree_lock);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700114
115 if (unlikely(error)) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700116 /*
117 * Only the context which have set SWAP_HAS_CACHE flag
118 * would call add_to_swap_cache().
119 * So add_to_swap_cache() doesn't returns -EEXIST.
120 */
121 VM_BUG_ON(error == -EEXIST);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700122 set_page_private(page, 0UL);
123 ClearPageSwapCache(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300124 put_page(page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700125 }
126
127 return error;
128}
129
130
131int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
132{
133 int error;
134
Jan Kara5e4c0d972013-09-11 14:26:05 -0700135 error = radix_tree_maybe_preload(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (!error) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700137 error = __add_to_swap_cache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 radix_tree_preload_end();
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 return error;
141}
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143/*
144 * This must be called only on pages that have
145 * been verified to be in the swap cache.
146 */
147void __delete_from_swap_cache(struct page *page)
148{
Shaohua Li33806f02013-02-22 16:34:37 -0800149 swp_entry_t entry;
150 struct address_space *address_space;
151
Sasha Levin309381fea2014-01-23 15:52:54 -0800152 VM_BUG_ON_PAGE(!PageLocked(page), page);
153 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
154 VM_BUG_ON_PAGE(PageWriteback(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Shaohua Li33806f02013-02-22 16:34:37 -0800156 entry.val = page_private(page);
157 address_space = swap_address_space(entry);
Huang Yingf6ab1f72016-10-07 17:00:21 -0700158 radix_tree_delete(&address_space->page_tree, swp_offset(entry));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700159 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 ClearPageSwapCache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800161 address_space->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700162 __dec_node_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 INC_CACHE_INFO(del_total);
164}
165
166/**
167 * add_to_swap - allocate swap space for a page
168 * @page: page we want to move to swap
169 *
170 * Allocate swap space for the page and add the page to the
171 * swap cache. Caller needs to hold the page lock.
172 */
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700173int add_to_swap(struct page *page, struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
175 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 int err;
177
Sasha Levin309381fea2014-01-23 15:52:54 -0800178 VM_BUG_ON_PAGE(!PageLocked(page), page);
179 VM_BUG_ON_PAGE(!PageUptodate(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700181 entry = get_swap_page();
182 if (!entry.val)
183 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Vladimir Davydov37e84352016-01-20 15:02:56 -0800185 if (mem_cgroup_try_charge_swap(page, entry)) {
186 swapcache_free(entry);
187 return 0;
188 }
189
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800190 if (unlikely(PageTransHuge(page)))
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700191 if (unlikely(split_huge_page_to_list(page, list))) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700192 swapcache_free(entry);
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800193 return 0;
194 }
195
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700196 /*
197 * Radix-tree node allocations from PF_MEMALLOC contexts could
198 * completely exhaust the page allocator. __GFP_NOMEMALLOC
199 * stops emergency reserves from being allocated.
200 *
201 * TODO: this could cause a theoretical memory reclaim
202 * deadlock in the swap out path.
203 */
204 /*
Minchan Kim854e9ed2016-01-15 16:54:53 -0800205 * Add it to the swap cache.
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700206 */
207 err = add_to_swap_cache(page, entry,
208 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Minchan Kim854e9ed2016-01-15 16:54:53 -0800210 if (!err) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700211 return 1;
212 } else { /* -ENOMEM radix-tree allocation failure */
213 /*
214 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
215 * clear SWAP_HAS_CACHE flag.
216 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700217 swapcache_free(entry);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700218 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 }
220}
221
222/*
223 * This must be called only on pages that have
224 * been verified to be in the swap cache and locked.
225 * It will never put the page into the free list,
226 * the caller has a reference on the page.
227 */
228void delete_from_swap_cache(struct page *page)
229{
230 swp_entry_t entry;
Shaohua Li33806f02013-02-22 16:34:37 -0800231 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700233 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Shaohua Li33806f02013-02-22 16:34:37 -0800235 address_space = swap_address_space(entry);
236 spin_lock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 __delete_from_swap_cache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800238 spin_unlock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700240 swapcache_free(entry);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300241 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244/*
245 * If we are the only user, then try to free up the swap cache.
246 *
247 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800248 * here because we are going to recheck again inside
249 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 * - Marcelo
251 */
252static inline void free_swap_cache(struct page *page)
253{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800254 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
255 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 unlock_page(page);
257 }
258}
259
260/*
261 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700262 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
264void free_page_and_swap_cache(struct page *page)
265{
266 free_swap_cache(page);
Aaron Lu6fcb52a2016-10-07 17:00:08 -0700267 if (!is_huge_zero_page(page))
Gerald Schaefer770a5372016-06-08 15:33:50 -0700268 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269}
270
271/*
272 * Passed an array of pages, drop them all from swapcache and then release
273 * them. They are removed from the LRU and freed if this is their last use.
274 */
275void free_pages_and_swap_cache(struct page **pages, int nr)
276{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 struct page **pagep = pages;
Michal Hockoaabfb572014-10-09 15:28:52 -0700278 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
280 lru_add_drain();
Michal Hockoaabfb572014-10-09 15:28:52 -0700281 for (i = 0; i < nr; i++)
282 free_swap_cache(pagep[i]);
283 release_pages(pagep, nr, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
286/*
287 * Lookup a swap entry in the swap cache. A found page will be returned
288 * unlocked and with its refcount incremented - we rely on the kernel
289 * lock getting page table operations atomic even if we drop the page
290 * lock before returning.
291 */
292struct page * lookup_swap_cache(swp_entry_t entry)
293{
294 struct page *page;
295
Huang Yingf6ab1f72016-10-07 17:00:21 -0700296 page = find_get_page(swap_address_space(entry), swp_offset(entry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
Shaohua Li579f8292014-02-06 12:04:21 -0800298 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 INC_CACHE_INFO(find_success);
Shaohua Li579f8292014-02-06 12:04:21 -0800300 if (TestClearPageReadahead(page))
301 atomic_inc(&swapin_readahead_hits);
302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
304 INC_CACHE_INFO(find_total);
305 return page;
306}
307
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700308struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
309 struct vm_area_struct *vma, unsigned long addr,
310 bool *new_page_allocated)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
312 struct page *found_page, *new_page = NULL;
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700313 struct address_space *swapper_space = swap_address_space(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 int err;
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700315 *new_page_allocated = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 do {
318 /*
319 * First check the swap cache. Since this is normally
320 * called after lookup_swap_cache() failed, re-calling
321 * that would confuse statistics.
322 */
Huang Yingf6ab1f72016-10-07 17:00:21 -0700323 found_page = find_get_page(swapper_space, swp_offset(entry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 if (found_page)
325 break;
326
Huang Yingba81f832017-02-22 15:45:46 -0800327 /*
328 * Just skip read ahead for unused swap slot.
329 * During swap_off when swap_slot_cache is disabled,
330 * we have to handle the race between putting
331 * swap entry in swap cache and marking swap slot
332 * as SWAP_HAS_CACHE. That's done in later part of code or
333 * else swap_off will be aborted if we return NULL.
334 */
335 if (!__swp_swapcount(entry) && swap_slot_cache_enabled)
336 break;
Tim Chene8c26ab2017-02-22 15:45:29 -0800337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 /*
339 * Get a new page to read into from swap.
340 */
341 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800342 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 if (!new_page)
344 break; /* Out of memory */
345 }
346
347 /*
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700348 * call radix_tree_preload() while we can wait.
349 */
Jan Kara5e4c0d972013-09-11 14:26:05 -0700350 err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700351 if (err)
352 break;
353
354 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800355 * Swap entry may have been freed since our caller observed it.
356 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700357 err = swapcache_prepare(entry);
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700358 if (err == -EEXIST) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700359 radix_tree_preload_end();
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700360 /*
361 * We might race against get_swap_page() and stumble
362 * across a SWAP_HAS_CACHE swap_map entry whose page
Huang Ying9c1cc2e2017-05-03 14:54:33 -0700363 * has not been brought into the swapcache yet.
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700364 */
365 cond_resched();
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700366 continue;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700367 }
368 if (err) { /* swp entry is obsolete ? */
369 radix_tree_preload_end();
Hugh Dickinsf0009442008-02-04 22:28:49 -0800370 break;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700371 }
Hugh Dickinsf0009442008-02-04 22:28:49 -0800372
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700373 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800374 __SetPageLocked(new_page);
Hugh Dickinsfa9949d2016-05-19 17:12:41 -0700375 __SetPageSwapBacked(new_page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700376 err = __add_to_swap_cache(new_page, entry);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200377 if (likely(!err)) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700378 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 /*
380 * Initiate read into locked page and return.
381 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700382 lru_cache_add_anon(new_page);
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700383 *new_page_allocated = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return new_page;
385 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700386 radix_tree_preload_end();
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800387 __ClearPageLocked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700388 /*
389 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
390 * clear SWAP_HAS_CACHE flag.
391 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700392 swapcache_free(entry);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800393 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 if (new_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300396 put_page(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 return found_page;
398}
Hugh Dickins46017e92008-02-04 22:28:41 -0800399
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700400/*
401 * Locate a page of swap in physical memory, reserving swap cache space
402 * and reading the disk if it is not already cached.
403 * A failure return means that either the page allocation failed or that
404 * the swap entry is no longer in use.
405 */
406struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
407 struct vm_area_struct *vma, unsigned long addr)
408{
409 bool page_was_allocated;
410 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
411 vma, addr, &page_was_allocated);
412
413 if (page_was_allocated)
414 swap_readpage(retpage);
415
416 return retpage;
417}
418
Shaohua Li579f8292014-02-06 12:04:21 -0800419static unsigned long swapin_nr_pages(unsigned long offset)
420{
421 static unsigned long prev_offset;
422 unsigned int pages, max_pages, last_ra;
423 static atomic_t last_readahead_pages;
424
Jason Low4db0c3c2015-04-15 16:14:08 -0700425 max_pages = 1 << READ_ONCE(page_cluster);
Shaohua Li579f8292014-02-06 12:04:21 -0800426 if (max_pages <= 1)
427 return 1;
428
429 /*
430 * This heuristic has been found to work well on both sequential and
431 * random loads, swapping to hard disk or to SSD: please don't ask
432 * what the "+ 2" means, it just happens to work well, that's all.
433 */
434 pages = atomic_xchg(&swapin_readahead_hits, 0) + 2;
435 if (pages == 2) {
436 /*
437 * We can have no readahead hits to judge by: but must not get
438 * stuck here forever, so check for an adjacent offset instead
439 * (and don't even bother to check whether swap type is same).
440 */
441 if (offset != prev_offset + 1 && offset != prev_offset - 1)
442 pages = 1;
443 prev_offset = offset;
444 } else {
445 unsigned int roundup = 4;
446 while (roundup < pages)
447 roundup <<= 1;
448 pages = roundup;
449 }
450
451 if (pages > max_pages)
452 pages = max_pages;
453
454 /* Don't shrink readahead too fast */
455 last_ra = atomic_read(&last_readahead_pages) / 2;
456 if (pages < last_ra)
457 pages = last_ra;
458 atomic_set(&last_readahead_pages, pages);
459
460 return pages;
461}
462
Hugh Dickins46017e92008-02-04 22:28:41 -0800463/**
464 * swapin_readahead - swap in pages in hope we need them soon
465 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700466 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800467 * @vma: user vma this address belongs to
468 * @addr: target address for mempolicy
469 *
470 * Returns the struct page for entry and addr, after queueing swapin.
471 *
472 * Primitive swap readahead code. We simply read an aligned block of
473 * (1 << page_cluster) entries in the swap area. This method is chosen
474 * because it doesn't cost us any seek time. We also make sure to queue
475 * the 'original' request together with the readahead ones...
476 *
477 * This has been extended to use the NUMA policies from the mm triggering
478 * the readahead.
479 *
480 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
481 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800482struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800483 struct vm_area_struct *vma, unsigned long addr)
484{
Hugh Dickins46017e92008-02-04 22:28:41 -0800485 struct page *page;
Shaohua Li579f8292014-02-06 12:04:21 -0800486 unsigned long entry_offset = swp_offset(entry);
487 unsigned long offset = entry_offset;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700488 unsigned long start_offset, end_offset;
Shaohua Li579f8292014-02-06 12:04:21 -0800489 unsigned long mask;
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700490 struct blk_plug plug;
Hugh Dickins46017e92008-02-04 22:28:41 -0800491
Shaohua Li579f8292014-02-06 12:04:21 -0800492 mask = swapin_nr_pages(offset) - 1;
493 if (!mask)
494 goto skip;
495
Rik van Riel67f96aa2012-03-21 16:33:50 -0700496 /* Read a page_cluster sized and aligned cluster around offset. */
497 start_offset = offset & ~mask;
498 end_offset = offset | mask;
499 if (!start_offset) /* First page is swap header. */
500 start_offset++;
501
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700502 blk_start_plug(&plug);
Rik van Riel67f96aa2012-03-21 16:33:50 -0700503 for (offset = start_offset; offset <= end_offset ; offset++) {
Hugh Dickins46017e92008-02-04 22:28:41 -0800504 /* Ok, do the async read-ahead now */
505 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800506 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800507 if (!page)
Rik van Riel67f96aa2012-03-21 16:33:50 -0700508 continue;
Shaohua Li579f8292014-02-06 12:04:21 -0800509 if (offset != entry_offset)
510 SetPageReadahead(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300511 put_page(page);
Hugh Dickins46017e92008-02-04 22:28:41 -0800512 }
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700513 blk_finish_plug(&plug);
514
Hugh Dickins46017e92008-02-04 22:28:41 -0800515 lru_add_drain(); /* Push any new pages onto the LRU now */
Shaohua Li579f8292014-02-06 12:04:21 -0800516skip:
Hugh Dickins02098fe2008-02-04 22:28:42 -0800517 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800518}
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800519
520int init_swap_address_space(unsigned int type, unsigned long nr_pages)
521{
522 struct address_space *spaces, *space;
523 unsigned int i, nr;
524
525 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
Huang Ying54f180d2017-05-08 15:57:40 -0700526 spaces = kvzalloc(sizeof(struct address_space) * nr, GFP_KERNEL);
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800527 if (!spaces)
528 return -ENOMEM;
529 for (i = 0; i < nr; i++) {
530 space = spaces + i;
531 INIT_RADIX_TREE(&space->page_tree, GFP_ATOMIC|__GFP_NOWARN);
532 atomic_set(&space->i_mmap_writable, 0);
533 space->a_ops = &swap_aops;
534 /* swap cache doesn't use writeback related tags */
535 mapping_set_no_writeback_tags(space);
536 spin_lock_init(&space->tree_lock);
537 }
538 nr_swapper_spaces[type] = nr;
539 rcu_assign_pointer(swapper_spaces[type], spaces);
540
541 return 0;
542}
543
544void exit_swap_address_space(unsigned int type)
545{
546 struct address_space *spaces;
547
548 spaces = swapper_spaces[type];
549 nr_swapper_spaces[type] = 0;
550 rcu_assign_pointer(swapper_spaces[type], NULL);
551 synchronize_rcu();
552 kvfree(spaces);
553}