blob: 4f251775ef902f3b83c5d680a9caa536773b3a60 [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>
13#include <linux/init.h>
14#include <linux/pagemap.h>
15#include <linux/buffer_head.h>
16#include <linux/backing-dev.h>
17
18#include <asm/pgtable.h>
19
20/*
21 * swapper_space is a fiction, retained to simplify the path through
22 * vmscan's shrink_list, to make sync_page look nicer, and to allow
23 * future use of radix_tree tags in the swap cache.
24 */
25static struct address_space_operations swap_aops = {
26 .writepage = swap_writepage,
27 .sync_page = block_sync_page,
28 .set_page_dirty = __set_page_dirty_nobuffers,
29};
30
31static struct backing_dev_info swap_backing_dev_info = {
32 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
33 .unplug_io_fn = swap_unplug_io_fn,
34};
35
36struct address_space swapper_space = {
37 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
38 .tree_lock = RW_LOCK_UNLOCKED,
39 .a_ops = &swap_aops,
40 .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
41 .backing_dev_info = &swap_backing_dev_info,
42};
43EXPORT_SYMBOL(swapper_space);
44
45#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
46
47static struct {
48 unsigned long add_total;
49 unsigned long del_total;
50 unsigned long find_success;
51 unsigned long find_total;
52 unsigned long noent_race;
53 unsigned long exist_race;
54} swap_cache_info;
55
56void show_swap_cache_info(void)
57{
58 printk("Swap cache: add %lu, delete %lu, find %lu/%lu, race %lu+%lu\n",
59 swap_cache_info.add_total, swap_cache_info.del_total,
60 swap_cache_info.find_success, swap_cache_info.find_total,
61 swap_cache_info.noent_race, swap_cache_info.exist_race);
62 printk("Free swap = %lukB\n", nr_swap_pages << (PAGE_SHIFT - 10));
63 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
64}
65
66/*
67 * __add_to_swap_cache resembles add_to_page_cache on swapper_space,
68 * but sets SwapCache flag and private instead of mapping and index.
69 */
70static int __add_to_swap_cache(struct page *page,
71 swp_entry_t entry, int gfp_mask)
72{
73 int error;
74
75 BUG_ON(PageSwapCache(page));
76 BUG_ON(PagePrivate(page));
77 error = radix_tree_preload(gfp_mask);
78 if (!error) {
79 write_lock_irq(&swapper_space.tree_lock);
80 error = radix_tree_insert(&swapper_space.page_tree,
81 entry.val, page);
82 if (!error) {
83 page_cache_get(page);
84 SetPageLocked(page);
85 SetPageSwapCache(page);
86 page->private = entry.val;
87 total_swapcache_pages++;
88 pagecache_acct(1);
89 }
90 write_unlock_irq(&swapper_space.tree_lock);
91 radix_tree_preload_end();
92 }
93 return error;
94}
95
96static int add_to_swap_cache(struct page *page, swp_entry_t entry)
97{
98 int error;
99
100 if (!swap_duplicate(entry)) {
101 INC_CACHE_INFO(noent_race);
102 return -ENOENT;
103 }
104 error = __add_to_swap_cache(page, entry, GFP_KERNEL);
105 /*
106 * Anon pages are already on the LRU, we don't run lru_cache_add here.
107 */
108 if (error) {
109 swap_free(entry);
110 if (error == -EEXIST)
111 INC_CACHE_INFO(exist_race);
112 return error;
113 }
114 INC_CACHE_INFO(add_total);
115 return 0;
116}
117
118/*
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{
124 BUG_ON(!PageLocked(page));
125 BUG_ON(!PageSwapCache(page));
126 BUG_ON(PageWriteback(page));
127
128 radix_tree_delete(&swapper_space.page_tree, page->private);
129 page->private = 0;
130 ClearPageSwapCache(page);
131 total_swapcache_pages--;
132 pagecache_acct(-1);
133 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 */
143int add_to_swap(struct page * page)
144{
145 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 int err;
147
148 if (!PageLocked(page))
149 BUG();
150
151 for (;;) {
152 entry = get_swap_page();
153 if (!entry.val)
154 return 0;
155
Nick Pigginbd53b712005-05-01 08:58:37 -0700156 /*
157 * Radix-tree node allocations from PF_MEMALLOC contexts could
158 * completely exhaust the page allocator. __GFP_NOMEMALLOC
159 * stops emergency reserves from being allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 *
Nick Pigginbd53b712005-05-01 08:58:37 -0700161 * TODO: this could cause a theoretical memory reclaim
162 * deadlock in the swap out path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 /*
165 * Add it to the swap cache and mark it dirty
166 */
Nick Pigginbd53b712005-05-01 08:58:37 -0700167 err = __add_to_swap_cache(page, entry,
168 GFP_ATOMIC|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170 switch (err) {
171 case 0: /* Success */
172 SetPageUptodate(page);
173 SetPageDirty(page);
174 INC_CACHE_INFO(add_total);
175 return 1;
176 case -EEXIST:
177 /* Raced with "speculative" read_swap_cache_async */
178 INC_CACHE_INFO(exist_race);
179 swap_free(entry);
180 continue;
181 default:
182 /* -ENOMEM radix-tree allocation failure */
183 swap_free(entry);
184 return 0;
185 }
186 }
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
199 BUG_ON(!PageSwapCache(page));
200 BUG_ON(!PageLocked(page));
201 BUG_ON(PageWriteback(page));
202 BUG_ON(PagePrivate(page));
203
204 entry.val = page->private;
205
206 write_lock_irq(&swapper_space.tree_lock);
207 __delete_from_swap_cache(page);
208 write_unlock_irq(&swapper_space.tree_lock);
209
210 swap_free(entry);
211 page_cache_release(page);
212}
213
214/*
215 * Strange swizzling function only for use by shmem_writepage
216 */
217int move_to_swap_cache(struct page *page, swp_entry_t entry)
218{
219 int err = __add_to_swap_cache(page, entry, GFP_ATOMIC);
220 if (!err) {
221 remove_from_page_cache(page);
222 page_cache_release(page); /* pagecache ref */
223 if (!swap_duplicate(entry))
224 BUG();
225 SetPageDirty(page);
226 INC_CACHE_INFO(add_total);
227 } else if (err == -EEXIST)
228 INC_CACHE_INFO(exist_race);
229 return err;
230}
231
232/*
233 * Strange swizzling function for shmem_getpage (and shmem_unuse)
234 */
235int move_from_swap_cache(struct page *page, unsigned long index,
236 struct address_space *mapping)
237{
238 int err = add_to_page_cache(page, mapping, index, GFP_ATOMIC);
239 if (!err) {
240 delete_from_swap_cache(page);
241 /* shift page from clean_pages to dirty_pages list */
242 ClearPageDirty(page);
243 set_page_dirty(page);
244 }
245 return err;
246}
247
248/*
249 * If we are the only user, then try to free up the swap cache.
250 *
251 * Its ok to check for PageSwapCache without the page lock
252 * here because we are going to recheck again inside
253 * exclusive_swap_page() _with_ the lock.
254 * - Marcelo
255 */
256static inline void free_swap_cache(struct page *page)
257{
258 if (PageSwapCache(page) && !TestSetPageLocked(page)) {
259 remove_exclusive_swap_page(page);
260 unlock_page(page);
261 }
262}
263
264/*
265 * Perform a free_page(), also freeing any swap cache associated with
266 * this page if it is the last user of the page. Can not do a lock_page,
267 * as we are holding the page_table_lock spinlock.
268 */
269void free_page_and_swap_cache(struct page *page)
270{
271 free_swap_cache(page);
272 page_cache_release(page);
273}
274
275/*
276 * Passed an array of pages, drop them all from swapcache and then release
277 * them. They are removed from the LRU and freed if this is their last use.
278 */
279void free_pages_and_swap_cache(struct page **pages, int nr)
280{
281 int chunk = 16;
282 struct page **pagep = pages;
283
284 lru_add_drain();
285 while (nr) {
286 int todo = min(chunk, nr);
287 int i;
288
289 for (i = 0; i < todo; i++)
290 free_swap_cache(pagep[i]);
291 release_pages(pagep, todo, 0);
292 pagep += todo;
293 nr -= todo;
294 }
295}
296
297/*
298 * Lookup a swap entry in the swap cache. A found page will be returned
299 * unlocked and with its refcount incremented - we rely on the kernel
300 * lock getting page table operations atomic even if we drop the page
301 * lock before returning.
302 */
303struct page * lookup_swap_cache(swp_entry_t entry)
304{
305 struct page *page;
306
307 page = find_get_page(&swapper_space, entry.val);
308
309 if (page)
310 INC_CACHE_INFO(find_success);
311
312 INC_CACHE_INFO(find_total);
313 return page;
314}
315
316/*
317 * Locate a page of swap in physical memory, reserving swap cache space
318 * and reading the disk if it is not already cached.
319 * A failure return means that either the page allocation failed or that
320 * the swap entry is no longer in use.
321 */
322struct page *read_swap_cache_async(swp_entry_t entry,
323 struct vm_area_struct *vma, unsigned long addr)
324{
325 struct page *found_page, *new_page = NULL;
326 int err;
327
328 do {
329 /*
330 * First check the swap cache. Since this is normally
331 * called after lookup_swap_cache() failed, re-calling
332 * that would confuse statistics.
333 */
334 found_page = find_get_page(&swapper_space, entry.val);
335 if (found_page)
336 break;
337
338 /*
339 * Get a new page to read into from swap.
340 */
341 if (!new_page) {
342 new_page = alloc_page_vma(GFP_HIGHUSER, vma, addr);
343 if (!new_page)
344 break; /* Out of memory */
345 }
346
347 /*
348 * Associate the page with swap entry in the swap cache.
349 * May fail (-ENOENT) if swap entry has been freed since
350 * our caller observed it. May fail (-EEXIST) if there
351 * is already a page associated with this entry in the
352 * swap cache: added by a racing read_swap_cache_async,
353 * or by try_to_swap_out (or shmem_writepage) re-using
354 * the just freed swap entry for an existing page.
355 * May fail (-ENOMEM) if radix-tree node allocation failed.
356 */
357 err = add_to_swap_cache(new_page, entry);
358 if (!err) {
359 /*
360 * Initiate read into locked page and return.
361 */
362 lru_cache_add_active(new_page);
363 swap_readpage(NULL, new_page);
364 return new_page;
365 }
366 } while (err != -ENOENT && err != -ENOMEM);
367
368 if (new_page)
369 page_cache_release(new_page);
370 return found_page;
371}