2 * linux/mm/swap_state.c
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
7 * Rewritten to use page cache, (C) 1998 Stephen Tweedie
9 #include <linux/module.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 #include <linux/pagevec.h>
19 #include <asm/pgtable.h>
22 * swapper_space is a fiction, retained to simplify the path through
23 * vmscan's shrink_list, to make sync_page look nicer, and to allow
24 * future use of radix_tree tags in the swap cache.
26 static struct address_space_operations swap_aops = {
27 .writepage = swap_writepage,
28 .sync_page = block_sync_page,
29 .set_page_dirty = __set_page_dirty_nobuffers,
30 .migratepage = migrate_page,
33 static struct backing_dev_info swap_backing_dev_info = {
34 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
35 .unplug_io_fn = swap_unplug_io_fn,
38 struct address_space swapper_space = {
39 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
40 .tree_lock = RW_LOCK_UNLOCKED,
42 .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
43 .backing_dev_info = &swap_backing_dev_info,
46 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
49 unsigned long add_total;
50 unsigned long del_total;
51 unsigned long find_success;
52 unsigned long find_total;
53 unsigned long noent_race;
54 unsigned long exist_race;
57 void show_swap_cache_info(void)
59 printk("Swap cache: add %lu, delete %lu, find %lu/%lu, race %lu+%lu\n",
60 swap_cache_info.add_total, swap_cache_info.del_total,
61 swap_cache_info.find_success, swap_cache_info.find_total,
62 swap_cache_info.noent_race, swap_cache_info.exist_race);
63 printk("Free swap = %lukB\n", nr_swap_pages << (PAGE_SHIFT - 10));
64 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
68 * __add_to_swap_cache resembles add_to_page_cache on swapper_space,
69 * but sets SwapCache flag and private instead of mapping and index.
71 static int __add_to_swap_cache(struct page *page, swp_entry_t entry,
76 BUG_ON(PageSwapCache(page));
77 BUG_ON(PagePrivate(page));
78 error = radix_tree_preload(gfp_mask);
80 write_lock_irq(&swapper_space.tree_lock);
81 error = radix_tree_insert(&swapper_space.page_tree,
86 SetPageSwapCache(page);
87 set_page_private(page, entry.val);
88 total_swapcache_pages++;
91 write_unlock_irq(&swapper_space.tree_lock);
92 radix_tree_preload_end();
97 static int add_to_swap_cache(struct page *page, swp_entry_t entry)
101 if (!swap_duplicate(entry)) {
102 INC_CACHE_INFO(noent_race);
105 error = __add_to_swap_cache(page, entry, GFP_KERNEL);
107 * Anon pages are already on the LRU, we don't run lru_cache_add here.
111 if (error == -EEXIST)
112 INC_CACHE_INFO(exist_race);
115 INC_CACHE_INFO(add_total);
120 * This must be called only on pages that have
121 * been verified to be in the swap cache.
123 void __delete_from_swap_cache(struct page *page)
125 BUG_ON(!PageLocked(page));
126 BUG_ON(!PageSwapCache(page));
127 BUG_ON(PageWriteback(page));
128 BUG_ON(PagePrivate(page));
130 radix_tree_delete(&swapper_space.page_tree, page_private(page));
131 set_page_private(page, 0);
132 ClearPageSwapCache(page);
133 total_swapcache_pages--;
135 INC_CACHE_INFO(del_total);
139 * add_to_swap - allocate swap space for a page
140 * @page: page we want to move to swap
142 * Allocate swap space for the page and add the page to the
143 * swap cache. Caller needs to hold the page lock.
145 int add_to_swap(struct page * page, gfp_t gfp_mask)
150 if (!PageLocked(page))
154 entry = get_swap_page();
159 * Radix-tree node allocations from PF_MEMALLOC contexts could
160 * completely exhaust the page allocator. __GFP_NOMEMALLOC
161 * stops emergency reserves from being allocated.
163 * TODO: this could cause a theoretical memory reclaim
164 * deadlock in the swap out path.
167 * Add it to the swap cache and mark it dirty
169 err = __add_to_swap_cache(page, entry,
170 gfp_mask|__GFP_NOMEMALLOC|__GFP_NOWARN);
173 case 0: /* Success */
174 SetPageUptodate(page);
176 INC_CACHE_INFO(add_total);
179 /* Raced with "speculative" read_swap_cache_async */
180 INC_CACHE_INFO(exist_race);
184 /* -ENOMEM radix-tree allocation failure */
192 * This must be called only on pages that have
193 * been verified to be in the swap cache and locked.
194 * It will never put the page into the free list,
195 * the caller has a reference on the page.
197 void delete_from_swap_cache(struct page *page)
201 entry.val = page_private(page);
203 write_lock_irq(&swapper_space.tree_lock);
204 __delete_from_swap_cache(page);
205 write_unlock_irq(&swapper_space.tree_lock);
208 page_cache_release(page);
212 * Strange swizzling function only for use by shmem_writepage
214 int move_to_swap_cache(struct page *page, swp_entry_t entry)
216 int err = __add_to_swap_cache(page, entry, GFP_ATOMIC);
218 remove_from_page_cache(page);
219 page_cache_release(page); /* pagecache ref */
220 if (!swap_duplicate(entry))
223 INC_CACHE_INFO(add_total);
224 } else if (err == -EEXIST)
225 INC_CACHE_INFO(exist_race);
230 * Strange swizzling function for shmem_getpage (and shmem_unuse)
232 int move_from_swap_cache(struct page *page, unsigned long index,
233 struct address_space *mapping)
235 int err = add_to_page_cache(page, mapping, index, GFP_ATOMIC);
237 delete_from_swap_cache(page);
238 /* shift page from clean_pages to dirty_pages list */
239 ClearPageDirty(page);
240 set_page_dirty(page);
246 * If we are the only user, then try to free up the swap cache.
248 * Its ok to check for PageSwapCache without the page lock
249 * here because we are going to recheck again inside
250 * exclusive_swap_page() _with_ the lock.
253 static inline void free_swap_cache(struct page *page)
255 if (PageSwapCache(page) && !TestSetPageLocked(page)) {
256 remove_exclusive_swap_page(page);
262 * Perform a free_page(), also freeing any swap cache associated with
263 * this page if it is the last user of the page.
265 void free_page_and_swap_cache(struct page *page)
267 free_swap_cache(page);
268 page_cache_release(page);
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.
275 void free_pages_and_swap_cache(struct page **pages, int nr)
277 struct page **pagep = pages;
281 int todo = min(nr, PAGEVEC_SIZE);
284 for (i = 0; i < todo; i++)
285 free_swap_cache(pagep[i]);
286 release_pages(pagep, todo, 0);
293 * Lookup a swap entry in the swap cache. A found page will be returned
294 * unlocked and with its refcount incremented - we rely on the kernel
295 * lock getting page table operations atomic even if we drop the page
296 * lock before returning.
298 struct page * lookup_swap_cache(swp_entry_t entry)
302 page = find_get_page(&swapper_space, entry.val);
305 INC_CACHE_INFO(find_success);
307 INC_CACHE_INFO(find_total);
312 * Locate a page of swap in physical memory, reserving swap cache space
313 * and reading the disk if it is not already cached.
314 * A failure return means that either the page allocation failed or that
315 * the swap entry is no longer in use.
317 struct page *read_swap_cache_async(swp_entry_t entry,
318 struct vm_area_struct *vma, unsigned long addr)
320 struct page *found_page, *new_page = NULL;
325 * First check the swap cache. Since this is normally
326 * called after lookup_swap_cache() failed, re-calling
327 * that would confuse statistics.
329 found_page = find_get_page(&swapper_space, entry.val);
334 * Get a new page to read into from swap.
337 new_page = alloc_page_vma(GFP_HIGHUSER, vma, addr);
339 break; /* Out of memory */
343 * Associate the page with swap entry in the swap cache.
344 * May fail (-ENOENT) if swap entry has been freed since
345 * our caller observed it. May fail (-EEXIST) if there
346 * is already a page associated with this entry in the
347 * swap cache: added by a racing read_swap_cache_async,
348 * or by try_to_swap_out (or shmem_writepage) re-using
349 * the just freed swap entry for an existing page.
350 * May fail (-ENOMEM) if radix-tree node allocation failed.
352 err = add_to_swap_cache(new_page, entry);
355 * Initiate read into locked page and return.
357 lru_cache_add_active(new_page);
358 swap_readpage(NULL, new_page);
361 } while (err != -ENOENT && err != -ENOMEM);
364 page_cache_release(new_page);