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>
18 #include <asm/pgtable.h>
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.
25 static struct address_space_operations swap_aops = {
26 .writepage = swap_writepage,
27 .sync_page = block_sync_page,
28 .set_page_dirty = __set_page_dirty_nobuffers,
31 static struct backing_dev_info swap_backing_dev_info = {
32 .memory_backed = 1, /* Does not contribute to dirty memory */
33 .unplug_io_fn = swap_unplug_io_fn,
36 struct address_space swapper_space = {
37 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
38 .tree_lock = SPIN_LOCK_UNLOCKED,
40 .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
41 .backing_dev_info = &swap_backing_dev_info,
43 EXPORT_SYMBOL(swapper_space);
45 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
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;
56 void show_swap_cache_info(void)
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));
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.
70 static int __add_to_swap_cache(struct page *page,
71 swp_entry_t entry, int gfp_mask)
75 BUG_ON(PageSwapCache(page));
76 BUG_ON(PagePrivate(page));
77 error = radix_tree_preload(gfp_mask);
79 spin_lock_irq(&swapper_space.tree_lock);
80 error = radix_tree_insert(&swapper_space.page_tree,
85 SetPageSwapCache(page);
86 page->private = entry.val;
87 total_swapcache_pages++;
90 spin_unlock_irq(&swapper_space.tree_lock);
91 radix_tree_preload_end();
96 static int add_to_swap_cache(struct page *page, swp_entry_t entry)
100 if (!swap_duplicate(entry)) {
101 INC_CACHE_INFO(noent_race);
104 error = __add_to_swap_cache(page, entry, GFP_KERNEL);
106 * Anon pages are already on the LRU, we don't run lru_cache_add here.
110 if (error == -EEXIST)
111 INC_CACHE_INFO(exist_race);
114 INC_CACHE_INFO(add_total);
119 * This must be called only on pages that have
120 * been verified to be in the swap cache.
122 void __delete_from_swap_cache(struct page *page)
124 BUG_ON(!PageLocked(page));
125 BUG_ON(!PageSwapCache(page));
126 BUG_ON(PageWriteback(page));
128 radix_tree_delete(&swapper_space.page_tree, page->private);
130 ClearPageSwapCache(page);
131 total_swapcache_pages--;
133 INC_CACHE_INFO(del_total);
137 * add_to_swap - allocate swap space for a page
138 * @page: page we want to move to swap
140 * Allocate swap space for the page and add the page to the
141 * swap cache. Caller needs to hold the page lock.
143 int add_to_swap(struct page * page)
149 if (!PageLocked(page))
153 entry = get_swap_page();
157 /* Radix-tree node allocations are performing
158 * GFP_ATOMIC allocations under PF_MEMALLOC.
159 * They can completely exhaust the page allocator.
161 * So PF_MEMALLOC is dropped here. This causes the slab
162 * allocations to fail earlier, so radix-tree nodes will
163 * then be allocated from the mempool reserves.
165 * We're still using __GFP_HIGH for radix-tree node
166 * allocations, so some of the emergency pools are available,
167 * just not all of them.
170 pf_flags = current->flags;
171 current->flags &= ~PF_MEMALLOC;
174 * Add it to the swap cache and mark it dirty
176 err = __add_to_swap_cache(page, entry, GFP_ATOMIC|__GFP_NOWARN);
178 if (pf_flags & PF_MEMALLOC)
179 current->flags |= PF_MEMALLOC;
182 case 0: /* Success */
183 SetPageUptodate(page);
185 INC_CACHE_INFO(add_total);
188 /* Raced with "speculative" read_swap_cache_async */
189 INC_CACHE_INFO(exist_race);
193 /* -ENOMEM radix-tree allocation failure */
201 * This must be called only on pages that have
202 * been verified to be in the swap cache and locked.
203 * It will never put the page into the free list,
204 * the caller has a reference on the page.
206 void delete_from_swap_cache(struct page *page)
210 BUG_ON(!PageSwapCache(page));
211 BUG_ON(!PageLocked(page));
212 BUG_ON(PageWriteback(page));
213 BUG_ON(PagePrivate(page));
215 entry.val = page->private;
217 spin_lock_irq(&swapper_space.tree_lock);
218 __delete_from_swap_cache(page);
219 spin_unlock_irq(&swapper_space.tree_lock);
222 page_cache_release(page);
226 * Strange swizzling function only for use by shmem_writepage
228 int move_to_swap_cache(struct page *page, swp_entry_t entry)
230 int err = __add_to_swap_cache(page, entry, GFP_ATOMIC);
232 remove_from_page_cache(page);
233 page_cache_release(page); /* pagecache ref */
234 if (!swap_duplicate(entry))
237 INC_CACHE_INFO(add_total);
238 } else if (err == -EEXIST)
239 INC_CACHE_INFO(exist_race);
244 * Strange swizzling function for shmem_getpage (and shmem_unuse)
246 int move_from_swap_cache(struct page *page, unsigned long index,
247 struct address_space *mapping)
249 int err = add_to_page_cache(page, mapping, index, GFP_ATOMIC);
251 delete_from_swap_cache(page);
252 /* shift page from clean_pages to dirty_pages list */
253 ClearPageDirty(page);
254 set_page_dirty(page);
260 * If we are the only user, then try to free up the swap cache.
262 * Its ok to check for PageSwapCache without the page lock
263 * here because we are going to recheck again inside
264 * exclusive_swap_page() _with_ the lock.
267 static inline void free_swap_cache(struct page *page)
269 if (PageSwapCache(page) && !TestSetPageLocked(page)) {
270 remove_exclusive_swap_page(page);
276 * Perform a free_page(), also freeing any swap cache associated with
277 * this page if it is the last user of the page. Can not do a lock_page,
278 * as we are holding the page_table_lock spinlock.
280 void free_page_and_swap_cache(struct page *page)
282 free_swap_cache(page);
283 page_cache_release(page);
287 * Passed an array of pages, drop them all from swapcache and then release
288 * them. They are removed from the LRU and freed if this is their last use.
290 void free_pages_and_swap_cache(struct page **pages, int nr)
293 struct page **pagep = pages;
297 int todo = min(chunk, nr);
300 for (i = 0; i < todo; i++)
301 free_swap_cache(pagep[i]);
302 release_pages(pagep, todo, 0);
309 * Lookup a swap entry in the swap cache. A found page will be returned
310 * unlocked and with its refcount incremented - we rely on the kernel
311 * lock getting page table operations atomic even if we drop the page
312 * lock before returning.
314 struct page * lookup_swap_cache(swp_entry_t entry)
318 spin_lock_irq(&swapper_space.tree_lock);
319 page = radix_tree_lookup(&swapper_space.page_tree, entry.val);
321 page_cache_get(page);
322 INC_CACHE_INFO(find_success);
324 spin_unlock_irq(&swapper_space.tree_lock);
325 INC_CACHE_INFO(find_total);
330 * Locate a page of swap in physical memory, reserving swap cache space
331 * and reading the disk if it is not already cached.
332 * A failure return means that either the page allocation failed or that
333 * the swap entry is no longer in use.
335 struct page *read_swap_cache_async(swp_entry_t entry,
336 struct vm_area_struct *vma, unsigned long addr)
338 struct page *found_page, *new_page = NULL;
343 * First check the swap cache. Since this is normally
344 * called after lookup_swap_cache() failed, re-calling
345 * that would confuse statistics.
347 spin_lock_irq(&swapper_space.tree_lock);
348 found_page = radix_tree_lookup(&swapper_space.page_tree,
351 page_cache_get(found_page);
352 spin_unlock_irq(&swapper_space.tree_lock);
357 * Get a new page to read into from swap.
360 new_page = alloc_page_vma(GFP_HIGHUSER, vma, addr);
362 break; /* Out of memory */
366 * Associate the page with swap entry in the swap cache.
367 * May fail (-ENOENT) if swap entry has been freed since
368 * our caller observed it. May fail (-EEXIST) if there
369 * is already a page associated with this entry in the
370 * swap cache: added by a racing read_swap_cache_async,
371 * or by try_to_swap_out (or shmem_writepage) re-using
372 * the just freed swap entry for an existing page.
373 * May fail (-ENOMEM) if radix-tree node allocation failed.
375 err = add_to_swap_cache(new_page, entry);
378 * Initiate read into locked page and return.
380 lru_cache_add_active(new_page);
381 swap_readpage(NULL, new_page);
384 } while (err != -ENOENT && err != -ENOMEM);
387 page_cache_release(new_page);