Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
5a0e3ad6 | 11 | #include <linux/gfp.h> |
1da177e4 LT |
12 | #include <linux/kernel_stat.h> |
13 | #include <linux/swap.h> | |
46017e95 | 14 | #include <linux/swapops.h> |
1da177e4 LT |
15 | #include <linux/init.h> |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/buffer_head.h> | |
18 | #include <linux/backing-dev.h> | |
c484d410 | 19 | #include <linux/pagevec.h> |
b20a3503 | 20 | #include <linux/migrate.h> |
8c7c6e34 | 21 | #include <linux/page_cgroup.h> |
1da177e4 LT |
22 | |
23 | #include <asm/pgtable.h> | |
24 | ||
25 | /* | |
26 | * swapper_space is a fiction, retained to simplify the path through | |
7eaceacc | 27 | * vmscan's shrink_page_list. |
1da177e4 | 28 | */ |
f5e54d6e | 29 | static const struct address_space_operations swap_aops = { |
1da177e4 | 30 | .writepage = swap_writepage, |
1da177e4 | 31 | .set_page_dirty = __set_page_dirty_nobuffers, |
e965f963 | 32 | .migratepage = migrate_page, |
1da177e4 LT |
33 | }; |
34 | ||
35 | static struct backing_dev_info swap_backing_dev_info = { | |
d993831f | 36 | .name = "swap", |
4f98a2fe | 37 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, |
1da177e4 LT |
38 | }; |
39 | ||
40 | struct address_space swapper_space = { | |
41 | .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN), | |
19fd6231 | 42 | .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock), |
1da177e4 LT |
43 | .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 | }; | |
1da177e4 LT |
47 | |
48 | #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0) | |
49 | ||
50 | static struct { | |
51 | unsigned long add_total; | |
52 | unsigned long del_total; | |
53 | unsigned long find_success; | |
54 | unsigned long find_total; | |
1da177e4 LT |
55 | } swap_cache_info; |
56 | ||
57 | void show_swap_cache_info(void) | |
58 | { | |
2c97b7fc JW |
59 | printk("%lu pages in swap cache\n", total_swapcache_pages); |
60 | printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n", | |
1da177e4 | 61 | swap_cache_info.add_total, swap_cache_info.del_total, |
bb63be0a | 62 | swap_cache_info.find_success, swap_cache_info.find_total); |
07279cdf | 63 | printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10)); |
1da177e4 LT |
64 | printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10)); |
65 | } | |
66 | ||
67 | /* | |
31a56396 | 68 | * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space, |
1da177e4 LT |
69 | * but sets SwapCache flag and private instead of mapping and index. |
70 | */ | |
31a56396 | 71 | static int __add_to_swap_cache(struct page *page, swp_entry_t entry) |
1da177e4 LT |
72 | { |
73 | int error; | |
74 | ||
51726b12 HD |
75 | VM_BUG_ON(!PageLocked(page)); |
76 | VM_BUG_ON(PageSwapCache(page)); | |
77 | VM_BUG_ON(!PageSwapBacked(page)); | |
78 | ||
31a56396 DN |
79 | 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)) { | |
2ca4532a DN |
93 | /* |
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); | |
31a56396 DN |
99 | set_page_private(page, 0UL); |
100 | ClearPageSwapCache(page); | |
101 | page_cache_release(page); | |
102 | } | |
103 | ||
104 | return error; | |
105 | } | |
106 | ||
107 | ||
108 | int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask) | |
109 | { | |
110 | int error; | |
111 | ||
35c754d7 BS |
112 | error = radix_tree_preload(gfp_mask); |
113 | if (!error) { | |
31a56396 | 114 | error = __add_to_swap_cache(page, entry); |
1da177e4 | 115 | radix_tree_preload_end(); |
fa1de900 | 116 | } |
1da177e4 LT |
117 | return error; |
118 | } | |
119 | ||
1da177e4 LT |
120 | /* |
121 | * This must be called only on pages that have | |
122 | * been verified to be in the swap cache. | |
123 | */ | |
124 | void __delete_from_swap_cache(struct page *page) | |
125 | { | |
51726b12 HD |
126 | VM_BUG_ON(!PageLocked(page)); |
127 | VM_BUG_ON(!PageSwapCache(page)); | |
128 | VM_BUG_ON(PageWriteback(page)); | |
1da177e4 | 129 | |
4c21e2f2 HD |
130 | radix_tree_delete(&swapper_space.page_tree, page_private(page)); |
131 | set_page_private(page, 0); | |
1da177e4 LT |
132 | ClearPageSwapCache(page); |
133 | total_swapcache_pages--; | |
347ce434 | 134 | __dec_zone_page_state(page, NR_FILE_PAGES); |
1da177e4 LT |
135 | 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 | */ | |
ac47b003 | 145 | int add_to_swap(struct page *page) |
1da177e4 LT |
146 | { |
147 | swp_entry_t entry; | |
1da177e4 LT |
148 | int err; |
149 | ||
51726b12 HD |
150 | VM_BUG_ON(!PageLocked(page)); |
151 | VM_BUG_ON(!PageUptodate(page)); | |
1da177e4 | 152 | |
2ca4532a DN |
153 | entry = get_swap_page(); |
154 | if (!entry.val) | |
155 | return 0; | |
156 | ||
3f04f62f AA |
157 | if (unlikely(PageTransHuge(page))) |
158 | if (unlikely(split_huge_page(page))) { | |
159 | swapcache_free(entry, NULL); | |
160 | return 0; | |
161 | } | |
162 | ||
2ca4532a DN |
163 | /* |
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); | |
176 | ||
177 | if (!err) { /* Success */ | |
178 | SetPageDirty(page); | |
179 | return 1; | |
180 | } else { /* -ENOMEM radix-tree allocation failure */ | |
bd53b714 | 181 | /* |
2ca4532a DN |
182 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely |
183 | * clear SWAP_HAS_CACHE flag. | |
1da177e4 | 184 | */ |
2ca4532a DN |
185 | swapcache_free(entry, NULL); |
186 | return 0; | |
1da177e4 LT |
187 | } |
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 | */ | |
196 | void delete_from_swap_cache(struct page *page) | |
197 | { | |
198 | swp_entry_t entry; | |
199 | ||
4c21e2f2 | 200 | entry.val = page_private(page); |
1da177e4 | 201 | |
19fd6231 | 202 | spin_lock_irq(&swapper_space.tree_lock); |
1da177e4 | 203 | __delete_from_swap_cache(page); |
19fd6231 | 204 | spin_unlock_irq(&swapper_space.tree_lock); |
1da177e4 | 205 | |
cb4b86ba | 206 | swapcache_free(entry, page); |
1da177e4 LT |
207 | page_cache_release(page); |
208 | } | |
209 | ||
1da177e4 LT |
210 | /* |
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 | |
a2c43eed HD |
214 | * here because we are going to recheck again inside |
215 | * try_to_free_swap() _with_ the lock. | |
1da177e4 LT |
216 | * - Marcelo |
217 | */ | |
218 | static inline void free_swap_cache(struct page *page) | |
219 | { | |
a2c43eed HD |
220 | if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) { |
221 | try_to_free_swap(page); | |
1da177e4 LT |
222 | unlock_page(page); |
223 | } | |
224 | } | |
225 | ||
226 | /* | |
227 | * Perform a free_page(), also freeing any swap cache associated with | |
b8072f09 | 228 | * this page if it is the last user of the page. |
1da177e4 LT |
229 | */ |
230 | void 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 | */ | |
240 | void free_pages_and_swap_cache(struct page **pages, int nr) | |
241 | { | |
1da177e4 LT |
242 | struct page **pagep = pages; |
243 | ||
244 | lru_add_drain(); | |
245 | while (nr) { | |
c484d410 | 246 | int todo = min(nr, PAGEVEC_SIZE); |
1da177e4 LT |
247 | 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 | */ | |
263 | struct 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 | */ | |
02098fea | 282 | struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
1da177e4 LT |
283 | 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) { | |
02098fea | 302 | new_page = alloc_page_vma(gfp_mask, vma, addr); |
1da177e4 LT |
303 | if (!new_page) |
304 | break; /* Out of memory */ | |
305 | } | |
306 | ||
31a56396 DN |
307 | /* |
308 | * call radix_tree_preload() while we can wait. | |
309 | */ | |
310 | err = radix_tree_preload(gfp_mask & GFP_KERNEL); | |
311 | if (err) | |
312 | break; | |
313 | ||
f000944d HD |
314 | /* |
315 | * Swap entry may have been freed since our caller observed it. | |
316 | */ | |
355cfa73 | 317 | err = swapcache_prepare(entry); |
31a56396 DN |
318 | if (err == -EEXIST) { /* seems racy */ |
319 | radix_tree_preload_end(); | |
355cfa73 | 320 | continue; |
31a56396 DN |
321 | } |
322 | if (err) { /* swp entry is obsolete ? */ | |
323 | radix_tree_preload_end(); | |
f000944d | 324 | break; |
31a56396 | 325 | } |
f000944d | 326 | |
2ca4532a | 327 | /* May fail (-ENOMEM) if radix-tree node allocation failed. */ |
f45840b5 | 328 | __set_page_locked(new_page); |
b2e18538 | 329 | SetPageSwapBacked(new_page); |
31a56396 | 330 | err = __add_to_swap_cache(new_page, entry); |
529ae9aa | 331 | if (likely(!err)) { |
31a56396 | 332 | radix_tree_preload_end(); |
1da177e4 LT |
333 | /* |
334 | * Initiate read into locked page and return. | |
335 | */ | |
c5fdae46 | 336 | lru_cache_add_anon(new_page); |
aca8bf32 | 337 | swap_readpage(new_page); |
1da177e4 LT |
338 | return new_page; |
339 | } | |
31a56396 | 340 | radix_tree_preload_end(); |
b2e18538 | 341 | ClearPageSwapBacked(new_page); |
f45840b5 | 342 | __clear_page_locked(new_page); |
2ca4532a DN |
343 | /* |
344 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely | |
345 | * clear SWAP_HAS_CACHE flag. | |
346 | */ | |
cb4b86ba | 347 | swapcache_free(entry, NULL); |
f000944d | 348 | } while (err != -ENOMEM); |
1da177e4 LT |
349 | |
350 | if (new_page) | |
351 | page_cache_release(new_page); | |
352 | return found_page; | |
353 | } | |
46017e95 HD |
354 | |
355 | /** | |
356 | * swapin_readahead - swap in pages in hope we need them soon | |
357 | * @entry: swap entry of this memory | |
7682486b | 358 | * @gfp_mask: memory allocation flags |
46017e95 HD |
359 | * @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 | */ | |
02098fea | 374 | struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask, |
46017e95 HD |
375 | 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), | |
02098fea | 393 | gfp_mask, vma, addr); |
46017e95 HD |
394 | if (!page) |
395 | break; | |
396 | page_cache_release(page); | |
397 | } | |
398 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
02098fea | 399 | return read_swap_cache_async(entry, gfp_mask, vma, addr); |
46017e95 | 400 | } |