]>
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 | */ | |
1da177e4 | 9 | #include <linux/mm.h> |
5a0e3ad6 | 10 | #include <linux/gfp.h> |
1da177e4 LT |
11 | #include <linux/kernel_stat.h> |
12 | #include <linux/swap.h> | |
46017e95 | 13 | #include <linux/swapops.h> |
1da177e4 LT |
14 | #include <linux/init.h> |
15 | #include <linux/pagemap.h> | |
1da177e4 | 16 | #include <linux/backing-dev.h> |
3fb5c298 | 17 | #include <linux/blkdev.h> |
c484d410 | 18 | #include <linux/pagevec.h> |
b20a3503 | 19 | #include <linux/migrate.h> |
1da177e4 LT |
20 | |
21 | #include <asm/pgtable.h> | |
22 | ||
23 | /* | |
24 | * swapper_space is a fiction, retained to simplify the path through | |
7eaceacc | 25 | * vmscan's shrink_page_list. |
1da177e4 | 26 | */ |
f5e54d6e | 27 | static const struct address_space_operations swap_aops = { |
1da177e4 | 28 | .writepage = swap_writepage, |
62c230bc | 29 | .set_page_dirty = swap_set_page_dirty, |
1c93923c | 30 | #ifdef CONFIG_MIGRATION |
e965f963 | 31 | .migratepage = migrate_page, |
1c93923c | 32 | #endif |
1da177e4 LT |
33 | }; |
34 | ||
33806f06 SL |
35 | struct address_space swapper_spaces[MAX_SWAPFILES] = { |
36 | [0 ... MAX_SWAPFILES - 1] = { | |
37 | .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN), | |
4bb5f5d9 | 38 | .i_mmap_writable = ATOMIC_INIT(0), |
33806f06 | 39 | .a_ops = &swap_aops, |
371a096e HY |
40 | /* swap cache doesn't use writeback related tags */ |
41 | .flags = 1 << AS_NO_WRITEBACK_TAGS, | |
33806f06 | 42 | } |
1da177e4 | 43 | }; |
1da177e4 LT |
44 | |
45 | #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0) | |
46 | ||
47 | static struct { | |
48 | unsigned long add_total; | |
49 | unsigned long del_total; | |
50 | unsigned long find_success; | |
51 | unsigned long find_total; | |
1da177e4 LT |
52 | } swap_cache_info; |
53 | ||
33806f06 SL |
54 | unsigned long total_swapcache_pages(void) |
55 | { | |
56 | int i; | |
57 | unsigned long ret = 0; | |
58 | ||
59 | for (i = 0; i < MAX_SWAPFILES; i++) | |
60 | ret += swapper_spaces[i].nrpages; | |
61 | return ret; | |
62 | } | |
63 | ||
579f8290 SL |
64 | static atomic_t swapin_readahead_hits = ATOMIC_INIT(4); |
65 | ||
1da177e4 LT |
66 | void show_swap_cache_info(void) |
67 | { | |
33806f06 | 68 | printk("%lu pages in swap cache\n", total_swapcache_pages()); |
2c97b7fc | 69 | printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n", |
1da177e4 | 70 | swap_cache_info.add_total, swap_cache_info.del_total, |
bb63be0a | 71 | swap_cache_info.find_success, swap_cache_info.find_total); |
ec8acf20 SL |
72 | printk("Free swap = %ldkB\n", |
73 | get_nr_swap_pages() << (PAGE_SHIFT - 10)); | |
1da177e4 LT |
74 | printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10)); |
75 | } | |
76 | ||
77 | /* | |
31a56396 | 78 | * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space, |
1da177e4 LT |
79 | * but sets SwapCache flag and private instead of mapping and index. |
80 | */ | |
2f772e6c | 81 | int __add_to_swap_cache(struct page *page, swp_entry_t entry) |
1da177e4 LT |
82 | { |
83 | int error; | |
33806f06 | 84 | struct address_space *address_space; |
1da177e4 | 85 | |
309381fe SL |
86 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
87 | VM_BUG_ON_PAGE(PageSwapCache(page), page); | |
88 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
51726b12 | 89 | |
09cbfeaf | 90 | get_page(page); |
31a56396 DN |
91 | SetPageSwapCache(page); |
92 | set_page_private(page, entry.val); | |
93 | ||
33806f06 SL |
94 | address_space = swap_address_space(entry); |
95 | spin_lock_irq(&address_space->tree_lock); | |
96 | error = radix_tree_insert(&address_space->page_tree, | |
97 | entry.val, page); | |
31a56396 | 98 | if (likely(!error)) { |
33806f06 | 99 | address_space->nrpages++; |
11fb9989 | 100 | __inc_node_page_state(page, NR_FILE_PAGES); |
31a56396 DN |
101 | INC_CACHE_INFO(add_total); |
102 | } | |
33806f06 | 103 | spin_unlock_irq(&address_space->tree_lock); |
31a56396 DN |
104 | |
105 | if (unlikely(error)) { | |
2ca4532a DN |
106 | /* |
107 | * Only the context which have set SWAP_HAS_CACHE flag | |
108 | * would call add_to_swap_cache(). | |
109 | * So add_to_swap_cache() doesn't returns -EEXIST. | |
110 | */ | |
111 | VM_BUG_ON(error == -EEXIST); | |
31a56396 DN |
112 | set_page_private(page, 0UL); |
113 | ClearPageSwapCache(page); | |
09cbfeaf | 114 | put_page(page); |
31a56396 DN |
115 | } |
116 | ||
117 | return error; | |
118 | } | |
119 | ||
120 | ||
121 | int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask) | |
122 | { | |
123 | int error; | |
124 | ||
5e4c0d97 | 125 | error = radix_tree_maybe_preload(gfp_mask); |
35c754d7 | 126 | if (!error) { |
31a56396 | 127 | error = __add_to_swap_cache(page, entry); |
1da177e4 | 128 | radix_tree_preload_end(); |
fa1de900 | 129 | } |
1da177e4 LT |
130 | return error; |
131 | } | |
132 | ||
1da177e4 LT |
133 | /* |
134 | * This must be called only on pages that have | |
135 | * been verified to be in the swap cache. | |
136 | */ | |
137 | void __delete_from_swap_cache(struct page *page) | |
138 | { | |
33806f06 SL |
139 | swp_entry_t entry; |
140 | struct address_space *address_space; | |
141 | ||
309381fe SL |
142 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
143 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); | |
144 | VM_BUG_ON_PAGE(PageWriteback(page), page); | |
1da177e4 | 145 | |
33806f06 SL |
146 | entry.val = page_private(page); |
147 | address_space = swap_address_space(entry); | |
148 | radix_tree_delete(&address_space->page_tree, page_private(page)); | |
4c21e2f2 | 149 | set_page_private(page, 0); |
1da177e4 | 150 | ClearPageSwapCache(page); |
33806f06 | 151 | address_space->nrpages--; |
11fb9989 | 152 | __dec_node_page_state(page, NR_FILE_PAGES); |
1da177e4 LT |
153 | INC_CACHE_INFO(del_total); |
154 | } | |
155 | ||
156 | /** | |
157 | * add_to_swap - allocate swap space for a page | |
158 | * @page: page we want to move to swap | |
159 | * | |
160 | * Allocate swap space for the page and add the page to the | |
161 | * swap cache. Caller needs to hold the page lock. | |
162 | */ | |
5bc7b8ac | 163 | int add_to_swap(struct page *page, struct list_head *list) |
1da177e4 LT |
164 | { |
165 | swp_entry_t entry; | |
1da177e4 LT |
166 | int err; |
167 | ||
309381fe SL |
168 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
169 | VM_BUG_ON_PAGE(!PageUptodate(page), page); | |
1da177e4 | 170 | |
2ca4532a DN |
171 | entry = get_swap_page(); |
172 | if (!entry.val) | |
173 | return 0; | |
174 | ||
37e84351 VD |
175 | if (mem_cgroup_try_charge_swap(page, entry)) { |
176 | swapcache_free(entry); | |
177 | return 0; | |
178 | } | |
179 | ||
3f04f62f | 180 | if (unlikely(PageTransHuge(page))) |
5bc7b8ac | 181 | if (unlikely(split_huge_page_to_list(page, list))) { |
0a31bc97 | 182 | swapcache_free(entry); |
3f04f62f AA |
183 | return 0; |
184 | } | |
185 | ||
2ca4532a DN |
186 | /* |
187 | * Radix-tree node allocations from PF_MEMALLOC contexts could | |
188 | * completely exhaust the page allocator. __GFP_NOMEMALLOC | |
189 | * stops emergency reserves from being allocated. | |
190 | * | |
191 | * TODO: this could cause a theoretical memory reclaim | |
192 | * deadlock in the swap out path. | |
193 | */ | |
194 | /* | |
854e9ed0 | 195 | * Add it to the swap cache. |
2ca4532a DN |
196 | */ |
197 | err = add_to_swap_cache(page, entry, | |
198 | __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN); | |
199 | ||
854e9ed0 | 200 | if (!err) { |
2ca4532a DN |
201 | return 1; |
202 | } else { /* -ENOMEM radix-tree allocation failure */ | |
bd53b714 | 203 | /* |
2ca4532a DN |
204 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely |
205 | * clear SWAP_HAS_CACHE flag. | |
1da177e4 | 206 | */ |
0a31bc97 | 207 | swapcache_free(entry); |
2ca4532a | 208 | return 0; |
1da177e4 LT |
209 | } |
210 | } | |
211 | ||
212 | /* | |
213 | * This must be called only on pages that have | |
214 | * been verified to be in the swap cache and locked. | |
215 | * It will never put the page into the free list, | |
216 | * the caller has a reference on the page. | |
217 | */ | |
218 | void delete_from_swap_cache(struct page *page) | |
219 | { | |
220 | swp_entry_t entry; | |
33806f06 | 221 | struct address_space *address_space; |
1da177e4 | 222 | |
4c21e2f2 | 223 | entry.val = page_private(page); |
1da177e4 | 224 | |
33806f06 SL |
225 | address_space = swap_address_space(entry); |
226 | spin_lock_irq(&address_space->tree_lock); | |
1da177e4 | 227 | __delete_from_swap_cache(page); |
33806f06 | 228 | spin_unlock_irq(&address_space->tree_lock); |
1da177e4 | 229 | |
0a31bc97 | 230 | swapcache_free(entry); |
09cbfeaf | 231 | put_page(page); |
1da177e4 LT |
232 | } |
233 | ||
1da177e4 LT |
234 | /* |
235 | * If we are the only user, then try to free up the swap cache. | |
236 | * | |
237 | * Its ok to check for PageSwapCache without the page lock | |
a2c43eed HD |
238 | * here because we are going to recheck again inside |
239 | * try_to_free_swap() _with_ the lock. | |
1da177e4 LT |
240 | * - Marcelo |
241 | */ | |
242 | static inline void free_swap_cache(struct page *page) | |
243 | { | |
a2c43eed HD |
244 | if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) { |
245 | try_to_free_swap(page); | |
1da177e4 LT |
246 | unlock_page(page); |
247 | } | |
248 | } | |
249 | ||
250 | /* | |
251 | * Perform a free_page(), also freeing any swap cache associated with | |
b8072f09 | 252 | * this page if it is the last user of the page. |
1da177e4 LT |
253 | */ |
254 | void free_page_and_swap_cache(struct page *page) | |
255 | { | |
256 | free_swap_cache(page); | |
770a5370 GS |
257 | if (is_huge_zero_page(page)) |
258 | put_huge_zero_page(); | |
259 | else | |
260 | put_page(page); | |
1da177e4 LT |
261 | } |
262 | ||
263 | /* | |
264 | * Passed an array of pages, drop them all from swapcache and then release | |
265 | * them. They are removed from the LRU and freed if this is their last use. | |
266 | */ | |
267 | void free_pages_and_swap_cache(struct page **pages, int nr) | |
268 | { | |
1da177e4 | 269 | struct page **pagep = pages; |
aabfb572 | 270 | int i; |
1da177e4 LT |
271 | |
272 | lru_add_drain(); | |
aabfb572 MH |
273 | for (i = 0; i < nr; i++) |
274 | free_swap_cache(pagep[i]); | |
275 | release_pages(pagep, nr, false); | |
1da177e4 LT |
276 | } |
277 | ||
278 | /* | |
279 | * Lookup a swap entry in the swap cache. A found page will be returned | |
280 | * unlocked and with its refcount incremented - we rely on the kernel | |
281 | * lock getting page table operations atomic even if we drop the page | |
282 | * lock before returning. | |
283 | */ | |
284 | struct page * lookup_swap_cache(swp_entry_t entry) | |
285 | { | |
286 | struct page *page; | |
287 | ||
33806f06 | 288 | page = find_get_page(swap_address_space(entry), entry.val); |
1da177e4 | 289 | |
579f8290 | 290 | if (page) { |
1da177e4 | 291 | INC_CACHE_INFO(find_success); |
579f8290 SL |
292 | if (TestClearPageReadahead(page)) |
293 | atomic_inc(&swapin_readahead_hits); | |
294 | } | |
1da177e4 LT |
295 | |
296 | INC_CACHE_INFO(find_total); | |
297 | return page; | |
298 | } | |
299 | ||
5b999aad DS |
300 | struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
301 | struct vm_area_struct *vma, unsigned long addr, | |
302 | bool *new_page_allocated) | |
1da177e4 LT |
303 | { |
304 | struct page *found_page, *new_page = NULL; | |
5b999aad | 305 | struct address_space *swapper_space = swap_address_space(entry); |
1da177e4 | 306 | int err; |
5b999aad | 307 | *new_page_allocated = false; |
1da177e4 LT |
308 | |
309 | do { | |
310 | /* | |
311 | * First check the swap cache. Since this is normally | |
312 | * called after lookup_swap_cache() failed, re-calling | |
313 | * that would confuse statistics. | |
314 | */ | |
5b999aad | 315 | found_page = find_get_page(swapper_space, entry.val); |
1da177e4 LT |
316 | if (found_page) |
317 | break; | |
318 | ||
319 | /* | |
320 | * Get a new page to read into from swap. | |
321 | */ | |
322 | if (!new_page) { | |
02098fea | 323 | new_page = alloc_page_vma(gfp_mask, vma, addr); |
1da177e4 LT |
324 | if (!new_page) |
325 | break; /* Out of memory */ | |
326 | } | |
327 | ||
31a56396 DN |
328 | /* |
329 | * call radix_tree_preload() while we can wait. | |
330 | */ | |
5e4c0d97 | 331 | err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL); |
31a56396 DN |
332 | if (err) |
333 | break; | |
334 | ||
f000944d HD |
335 | /* |
336 | * Swap entry may have been freed since our caller observed it. | |
337 | */ | |
355cfa73 | 338 | err = swapcache_prepare(entry); |
cbab0e4e | 339 | if (err == -EEXIST) { |
31a56396 | 340 | radix_tree_preload_end(); |
cbab0e4e RA |
341 | /* |
342 | * We might race against get_swap_page() and stumble | |
343 | * across a SWAP_HAS_CACHE swap_map entry whose page | |
344 | * has not been brought into the swapcache yet, while | |
345 | * the other end is scheduled away waiting on discard | |
346 | * I/O completion at scan_swap_map(). | |
347 | * | |
348 | * In order to avoid turning this transitory state | |
349 | * into a permanent loop around this -EEXIST case | |
350 | * if !CONFIG_PREEMPT and the I/O completion happens | |
351 | * to be waiting on the CPU waitqueue where we are now | |
352 | * busy looping, we just conditionally invoke the | |
353 | * scheduler here, if there are some more important | |
354 | * tasks to run. | |
355 | */ | |
356 | cond_resched(); | |
355cfa73 | 357 | continue; |
31a56396 DN |
358 | } |
359 | if (err) { /* swp entry is obsolete ? */ | |
360 | radix_tree_preload_end(); | |
f000944d | 361 | break; |
31a56396 | 362 | } |
f000944d | 363 | |
2ca4532a | 364 | /* May fail (-ENOMEM) if radix-tree node allocation failed. */ |
48c935ad | 365 | __SetPageLocked(new_page); |
fa9949da | 366 | __SetPageSwapBacked(new_page); |
31a56396 | 367 | err = __add_to_swap_cache(new_page, entry); |
529ae9aa | 368 | if (likely(!err)) { |
31a56396 | 369 | radix_tree_preload_end(); |
1da177e4 LT |
370 | /* |
371 | * Initiate read into locked page and return. | |
372 | */ | |
c5fdae46 | 373 | lru_cache_add_anon(new_page); |
5b999aad | 374 | *new_page_allocated = true; |
1da177e4 LT |
375 | return new_page; |
376 | } | |
31a56396 | 377 | radix_tree_preload_end(); |
48c935ad | 378 | __ClearPageLocked(new_page); |
2ca4532a DN |
379 | /* |
380 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely | |
381 | * clear SWAP_HAS_CACHE flag. | |
382 | */ | |
0a31bc97 | 383 | swapcache_free(entry); |
f000944d | 384 | } while (err != -ENOMEM); |
1da177e4 LT |
385 | |
386 | if (new_page) | |
09cbfeaf | 387 | put_page(new_page); |
1da177e4 LT |
388 | return found_page; |
389 | } | |
46017e95 | 390 | |
5b999aad DS |
391 | /* |
392 | * Locate a page of swap in physical memory, reserving swap cache space | |
393 | * and reading the disk if it is not already cached. | |
394 | * A failure return means that either the page allocation failed or that | |
395 | * the swap entry is no longer in use. | |
396 | */ | |
397 | struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, | |
398 | struct vm_area_struct *vma, unsigned long addr) | |
399 | { | |
400 | bool page_was_allocated; | |
401 | struct page *retpage = __read_swap_cache_async(entry, gfp_mask, | |
402 | vma, addr, &page_was_allocated); | |
403 | ||
404 | if (page_was_allocated) | |
405 | swap_readpage(retpage); | |
406 | ||
407 | return retpage; | |
408 | } | |
409 | ||
579f8290 SL |
410 | static unsigned long swapin_nr_pages(unsigned long offset) |
411 | { | |
412 | static unsigned long prev_offset; | |
413 | unsigned int pages, max_pages, last_ra; | |
414 | static atomic_t last_readahead_pages; | |
415 | ||
4db0c3c2 | 416 | max_pages = 1 << READ_ONCE(page_cluster); |
579f8290 SL |
417 | if (max_pages <= 1) |
418 | return 1; | |
419 | ||
420 | /* | |
421 | * This heuristic has been found to work well on both sequential and | |
422 | * random loads, swapping to hard disk or to SSD: please don't ask | |
423 | * what the "+ 2" means, it just happens to work well, that's all. | |
424 | */ | |
425 | pages = atomic_xchg(&swapin_readahead_hits, 0) + 2; | |
426 | if (pages == 2) { | |
427 | /* | |
428 | * We can have no readahead hits to judge by: but must not get | |
429 | * stuck here forever, so check for an adjacent offset instead | |
430 | * (and don't even bother to check whether swap type is same). | |
431 | */ | |
432 | if (offset != prev_offset + 1 && offset != prev_offset - 1) | |
433 | pages = 1; | |
434 | prev_offset = offset; | |
435 | } else { | |
436 | unsigned int roundup = 4; | |
437 | while (roundup < pages) | |
438 | roundup <<= 1; | |
439 | pages = roundup; | |
440 | } | |
441 | ||
442 | if (pages > max_pages) | |
443 | pages = max_pages; | |
444 | ||
445 | /* Don't shrink readahead too fast */ | |
446 | last_ra = atomic_read(&last_readahead_pages) / 2; | |
447 | if (pages < last_ra) | |
448 | pages = last_ra; | |
449 | atomic_set(&last_readahead_pages, pages); | |
450 | ||
451 | return pages; | |
452 | } | |
453 | ||
46017e95 HD |
454 | /** |
455 | * swapin_readahead - swap in pages in hope we need them soon | |
456 | * @entry: swap entry of this memory | |
7682486b | 457 | * @gfp_mask: memory allocation flags |
46017e95 HD |
458 | * @vma: user vma this address belongs to |
459 | * @addr: target address for mempolicy | |
460 | * | |
461 | * Returns the struct page for entry and addr, after queueing swapin. | |
462 | * | |
463 | * Primitive swap readahead code. We simply read an aligned block of | |
464 | * (1 << page_cluster) entries in the swap area. This method is chosen | |
465 | * because it doesn't cost us any seek time. We also make sure to queue | |
466 | * the 'original' request together with the readahead ones... | |
467 | * | |
468 | * This has been extended to use the NUMA policies from the mm triggering | |
469 | * the readahead. | |
470 | * | |
471 | * Caller must hold down_read on the vma->vm_mm if vma is not NULL. | |
472 | */ | |
02098fea | 473 | struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask, |
46017e95 HD |
474 | struct vm_area_struct *vma, unsigned long addr) |
475 | { | |
46017e95 | 476 | struct page *page; |
579f8290 SL |
477 | unsigned long entry_offset = swp_offset(entry); |
478 | unsigned long offset = entry_offset; | |
67f96aa2 | 479 | unsigned long start_offset, end_offset; |
579f8290 | 480 | unsigned long mask; |
3fb5c298 | 481 | struct blk_plug plug; |
46017e95 | 482 | |
579f8290 SL |
483 | mask = swapin_nr_pages(offset) - 1; |
484 | if (!mask) | |
485 | goto skip; | |
486 | ||
67f96aa2 RR |
487 | /* Read a page_cluster sized and aligned cluster around offset. */ |
488 | start_offset = offset & ~mask; | |
489 | end_offset = offset | mask; | |
490 | if (!start_offset) /* First page is swap header. */ | |
491 | start_offset++; | |
492 | ||
3fb5c298 | 493 | blk_start_plug(&plug); |
67f96aa2 | 494 | for (offset = start_offset; offset <= end_offset ; offset++) { |
46017e95 HD |
495 | /* Ok, do the async read-ahead now */ |
496 | page = read_swap_cache_async(swp_entry(swp_type(entry), offset), | |
02098fea | 497 | gfp_mask, vma, addr); |
46017e95 | 498 | if (!page) |
67f96aa2 | 499 | continue; |
579f8290 SL |
500 | if (offset != entry_offset) |
501 | SetPageReadahead(page); | |
09cbfeaf | 502 | put_page(page); |
46017e95 | 503 | } |
3fb5c298 CE |
504 | blk_finish_plug(&plug); |
505 | ||
46017e95 | 506 | lru_add_drain(); /* Push any new pages onto the LRU now */ |
579f8290 | 507 | skip: |
02098fea | 508 | return read_swap_cache_async(entry, gfp_mask, vma, addr); |
46017e95 | 509 | } |