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/swapops.h>
14 #include <linux/init.h>
15 #include <linux/pagemap.h>
16 #include <linux/buffer_head.h>
17 #include <linux/backing-dev.h>
18 #include <linux/pagevec.h>
19 #include <linux/migrate.h>
20 #include <linux/page_cgroup.h>
22 #include <asm/pgtable.h>
25 * swapper_space is a fiction, retained to simplify the path through
26 * vmscan's shrink_page_list, to make sync_page look nicer, and to allow
27 * future use of radix_tree tags in the swap cache.
29 static const struct address_space_operations swap_aops = {
30 .writepage = swap_writepage,
31 .sync_page = block_sync_page,
32 .set_page_dirty = __set_page_dirty_nobuffers,
33 .migratepage = migrate_page,
36 static struct backing_dev_info swap_backing_dev_info = {
37 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
38 .unplug_io_fn = swap_unplug_io_fn,
41 struct address_space swapper_space = {
42 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
43 .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
45 .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
46 .backing_dev_info = &swap_backing_dev_info,
49 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
52 unsigned long add_total;
53 unsigned long del_total;
54 unsigned long find_success;
55 unsigned long find_total;
58 void show_swap_cache_info(void)
60 printk("%lu pages in swap cache\n", total_swapcache_pages);
61 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
62 swap_cache_info.add_total, swap_cache_info.del_total,
63 swap_cache_info.find_success, swap_cache_info.find_total);
64 printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
65 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
69 * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
70 * but sets SwapCache flag and private instead of mapping and index.
72 int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
76 VM_BUG_ON(!PageLocked(page));
77 VM_BUG_ON(PageSwapCache(page));
78 VM_BUG_ON(!PageSwapBacked(page));
80 error = radix_tree_preload(gfp_mask);
83 SetPageSwapCache(page);
84 set_page_private(page, entry.val);
86 spin_lock_irq(&swapper_space.tree_lock);
87 error = radix_tree_insert(&swapper_space.page_tree,
90 total_swapcache_pages++;
91 __inc_zone_page_state(page, NR_FILE_PAGES);
92 INC_CACHE_INFO(add_total);
94 spin_unlock_irq(&swapper_space.tree_lock);
95 radix_tree_preload_end();
97 if (unlikely(error)) {
98 set_page_private(page, 0UL);
99 ClearPageSwapCache(page);
100 page_cache_release(page);
107 * This must be called only on pages that have
108 * been verified to be in the swap cache.
110 void __delete_from_swap_cache(struct page *page)
112 VM_BUG_ON(!PageLocked(page));
113 VM_BUG_ON(!PageSwapCache(page));
114 VM_BUG_ON(PageWriteback(page));
116 radix_tree_delete(&swapper_space.page_tree, page_private(page));
117 set_page_private(page, 0);
118 ClearPageSwapCache(page);
119 total_swapcache_pages--;
120 __dec_zone_page_state(page, NR_FILE_PAGES);
121 INC_CACHE_INFO(del_total);
125 * add_to_swap - allocate swap space for a page
126 * @page: page we want to move to swap
127 * @gfp_mask: memory allocation flags
129 * Allocate swap space for the page and add the page to the
130 * swap cache. Caller needs to hold the page lock.
132 int add_to_swap(struct page *page)
137 VM_BUG_ON(!PageLocked(page));
138 VM_BUG_ON(!PageUptodate(page));
141 entry = get_swap_page();
146 * Radix-tree node allocations from PF_MEMALLOC contexts could
147 * completely exhaust the page allocator. __GFP_NOMEMALLOC
148 * stops emergency reserves from being allocated.
150 * TODO: this could cause a theoretical memory reclaim
151 * deadlock in the swap out path.
154 * Add it to the swap cache and mark it dirty
156 err = add_to_swap_cache(page, entry,
157 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
160 case 0: /* Success */
164 /* Raced with "speculative" read_swap_cache_async */
168 /* -ENOMEM radix-tree allocation failure */
176 * This must be called only on pages that have
177 * been verified to be in the swap cache and locked.
178 * It will never put the page into the free list,
179 * the caller has a reference on the page.
181 void delete_from_swap_cache(struct page *page)
185 entry.val = page_private(page);
187 spin_lock_irq(&swapper_space.tree_lock);
188 __delete_from_swap_cache(page);
189 spin_unlock_irq(&swapper_space.tree_lock);
191 mem_cgroup_uncharge_swapcache(page, entry);
193 page_cache_release(page);
197 * If we are the only user, then try to free up the swap cache.
199 * Its ok to check for PageSwapCache without the page lock
200 * here because we are going to recheck again inside
201 * try_to_free_swap() _with_ the lock.
204 static inline void free_swap_cache(struct page *page)
206 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
207 try_to_free_swap(page);
213 * Perform a free_page(), also freeing any swap cache associated with
214 * this page if it is the last user of the page.
216 void free_page_and_swap_cache(struct page *page)
218 free_swap_cache(page);
219 page_cache_release(page);
223 * Passed an array of pages, drop them all from swapcache and then release
224 * them. They are removed from the LRU and freed if this is their last use.
226 void free_pages_and_swap_cache(struct page **pages, int nr)
228 struct page **pagep = pages;
232 int todo = min(nr, PAGEVEC_SIZE);
235 for (i = 0; i < todo; i++)
236 free_swap_cache(pagep[i]);
237 release_pages(pagep, todo, 0);
244 * Lookup a swap entry in the swap cache. A found page will be returned
245 * unlocked and with its refcount incremented - we rely on the kernel
246 * lock getting page table operations atomic even if we drop the page
247 * lock before returning.
249 struct page * lookup_swap_cache(swp_entry_t entry)
253 page = find_get_page(&swapper_space, entry.val);
256 INC_CACHE_INFO(find_success);
258 INC_CACHE_INFO(find_total);
263 * Locate a page of swap in physical memory, reserving swap cache space
264 * and reading the disk if it is not already cached.
265 * A failure return means that either the page allocation failed or that
266 * the swap entry is no longer in use.
268 struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
269 struct vm_area_struct *vma, unsigned long addr)
271 struct page *found_page, *new_page = NULL;
276 * First check the swap cache. Since this is normally
277 * called after lookup_swap_cache() failed, re-calling
278 * that would confuse statistics.
280 found_page = find_get_page(&swapper_space, entry.val);
285 * Get a new page to read into from swap.
288 new_page = alloc_page_vma(gfp_mask, vma, addr);
290 break; /* Out of memory */
294 * Swap entry may have been freed since our caller observed it.
296 if (!swap_duplicate(entry))
300 * Associate the page with swap entry in the swap cache.
301 * May fail (-EEXIST) if there is already a page associated
302 * with this entry in the swap cache: added by a racing
303 * read_swap_cache_async, or add_to_swap or shmem_writepage
304 * re-using the just freed swap entry for an existing page.
305 * May fail (-ENOMEM) if radix-tree node allocation failed.
307 __set_page_locked(new_page);
308 SetPageSwapBacked(new_page);
309 err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
312 * Initiate read into locked page and return.
314 lru_cache_add_anon(new_page);
315 swap_readpage(NULL, new_page);
318 ClearPageSwapBacked(new_page);
319 __clear_page_locked(new_page);
321 } while (err != -ENOMEM);
324 page_cache_release(new_page);
329 * swapin_readahead - swap in pages in hope we need them soon
330 * @entry: swap entry of this memory
331 * @gfp_mask: memory allocation flags
332 * @vma: user vma this address belongs to
333 * @addr: target address for mempolicy
335 * Returns the struct page for entry and addr, after queueing swapin.
337 * Primitive swap readahead code. We simply read an aligned block of
338 * (1 << page_cluster) entries in the swap area. This method is chosen
339 * because it doesn't cost us any seek time. We also make sure to queue
340 * the 'original' request together with the readahead ones...
342 * This has been extended to use the NUMA policies from the mm triggering
345 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
347 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
348 struct vm_area_struct *vma, unsigned long addr)
352 unsigned long offset;
353 unsigned long end_offset;
356 * Get starting offset for readaround, and number of pages to read.
357 * Adjust starting address by readbehind (for NUMA interleave case)?
358 * No, it's very unlikely that swap layout would follow vma layout,
359 * more likely that neighbouring swap pages came from the same node:
360 * so use the same "addr" to choose the same node for each swap read.
362 nr_pages = valid_swaphandles(entry, &offset);
363 for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
364 /* Ok, do the async read-ahead now */
365 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
366 gfp_mask, vma, addr);
369 page_cache_release(page);
371 lru_add_drain(); /* Push any new pages onto the LRU now */
372 return read_swap_cache_async(entry, gfp_mask, vma, addr);