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 = {
38 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
39 .unplug_io_fn = swap_unplug_io_fn,
42 struct address_space swapper_space = {
43 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
44 .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
46 .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
47 .backing_dev_info = &swap_backing_dev_info,
50 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
53 unsigned long add_total;
54 unsigned long del_total;
55 unsigned long find_success;
56 unsigned long find_total;
59 void show_swap_cache_info(void)
61 printk("%lu pages in swap cache\n", total_swapcache_pages);
62 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
63 swap_cache_info.add_total, swap_cache_info.del_total,
64 swap_cache_info.find_success, swap_cache_info.find_total);
65 printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
66 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
70 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
71 * but sets SwapCache flag and private instead of mapping and index.
73 static int __add_to_swap_cache(struct page *page, swp_entry_t entry)
77 VM_BUG_ON(!PageLocked(page));
78 VM_BUG_ON(PageSwapCache(page));
79 VM_BUG_ON(!PageSwapBacked(page));
82 SetPageSwapCache(page);
83 set_page_private(page, entry.val);
85 spin_lock_irq(&swapper_space.tree_lock);
86 error = radix_tree_insert(&swapper_space.page_tree, entry.val, page);
88 total_swapcache_pages++;
89 __inc_zone_page_state(page, NR_FILE_PAGES);
90 INC_CACHE_INFO(add_total);
92 spin_unlock_irq(&swapper_space.tree_lock);
94 if (unlikely(error)) {
96 * Only the context which have set SWAP_HAS_CACHE flag
97 * would call add_to_swap_cache().
98 * So add_to_swap_cache() doesn't returns -EEXIST.
100 VM_BUG_ON(error == -EEXIST);
101 set_page_private(page, 0UL);
102 ClearPageSwapCache(page);
103 page_cache_release(page);
110 int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
114 error = radix_tree_preload(gfp_mask);
116 error = __add_to_swap_cache(page, entry);
117 radix_tree_preload_end();
123 * This must be called only on pages that have
124 * been verified to be in the swap cache.
126 void __delete_from_swap_cache(struct page *page)
128 VM_BUG_ON(!PageLocked(page));
129 VM_BUG_ON(!PageSwapCache(page));
130 VM_BUG_ON(PageWriteback(page));
132 radix_tree_delete(&swapper_space.page_tree, page_private(page));
133 set_page_private(page, 0);
134 ClearPageSwapCache(page);
135 total_swapcache_pages--;
136 __dec_zone_page_state(page, NR_FILE_PAGES);
137 INC_CACHE_INFO(del_total);
141 * add_to_swap - allocate swap space for a page
142 * @page: page we want to move to swap
144 * Allocate swap space for the page and add the page to the
145 * swap cache. Caller needs to hold the page lock.
147 int add_to_swap(struct page *page)
152 VM_BUG_ON(!PageLocked(page));
153 VM_BUG_ON(!PageUptodate(page));
155 entry = get_swap_page();
160 * Radix-tree node allocations from PF_MEMALLOC contexts could
161 * completely exhaust the page allocator. __GFP_NOMEMALLOC
162 * stops emergency reserves from being allocated.
164 * TODO: this could cause a theoretical memory reclaim
165 * deadlock in the swap out path.
168 * Add it to the swap cache and mark it dirty
170 err = add_to_swap_cache(page, entry,
171 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
173 if (!err) { /* Success */
176 } else { /* -ENOMEM radix-tree allocation failure */
178 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
179 * clear SWAP_HAS_CACHE flag.
181 swapcache_free(entry, NULL);
187 * This must be called only on pages that have
188 * been verified to be in the swap cache and locked.
189 * It will never put the page into the free list,
190 * the caller has a reference on the page.
192 void delete_from_swap_cache(struct page *page)
196 entry.val = page_private(page);
198 spin_lock_irq(&swapper_space.tree_lock);
199 __delete_from_swap_cache(page);
200 spin_unlock_irq(&swapper_space.tree_lock);
202 swapcache_free(entry, page);
203 page_cache_release(page);
207 * If we are the only user, then try to free up the swap cache.
209 * Its ok to check for PageSwapCache without the page lock
210 * here because we are going to recheck again inside
211 * try_to_free_swap() _with_ the lock.
214 static inline void free_swap_cache(struct page *page)
216 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
217 try_to_free_swap(page);
223 * Perform a free_page(), also freeing any swap cache associated with
224 * this page if it is the last user of the page.
226 void free_page_and_swap_cache(struct page *page)
228 free_swap_cache(page);
229 page_cache_release(page);
233 * Passed an array of pages, drop them all from swapcache and then release
234 * them. They are removed from the LRU and freed if this is their last use.
236 void free_pages_and_swap_cache(struct page **pages, int nr)
238 struct page **pagep = pages;
242 int todo = min(nr, PAGEVEC_SIZE);
245 for (i = 0; i < todo; i++)
246 free_swap_cache(pagep[i]);
247 release_pages(pagep, todo, 0);
254 * Lookup a swap entry in the swap cache. A found page will be returned
255 * unlocked and with its refcount incremented - we rely on the kernel
256 * lock getting page table operations atomic even if we drop the page
257 * lock before returning.
259 struct page * lookup_swap_cache(swp_entry_t entry)
263 page = find_get_page(&swapper_space, entry.val);
266 INC_CACHE_INFO(find_success);
268 INC_CACHE_INFO(find_total);
273 * Locate a page of swap in physical memory, reserving swap cache space
274 * and reading the disk if it is not already cached.
275 * A failure return means that either the page allocation failed or that
276 * the swap entry is no longer in use.
278 struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
279 struct vm_area_struct *vma, unsigned long addr)
281 struct page *found_page, *new_page = NULL;
286 * First check the swap cache. Since this is normally
287 * called after lookup_swap_cache() failed, re-calling
288 * that would confuse statistics.
290 found_page = find_get_page(&swapper_space, entry.val);
295 * Get a new page to read into from swap.
298 new_page = alloc_page_vma(gfp_mask, vma, addr);
300 break; /* Out of memory */
304 * call radix_tree_preload() while we can wait.
306 err = radix_tree_preload(gfp_mask & GFP_KERNEL);
311 * Swap entry may have been freed since our caller observed it.
313 err = swapcache_prepare(entry);
314 if (err == -EEXIST) { /* seems racy */
315 radix_tree_preload_end();
318 if (err) { /* swp entry is obsolete ? */
319 radix_tree_preload_end();
323 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
324 __set_page_locked(new_page);
325 SetPageSwapBacked(new_page);
326 err = __add_to_swap_cache(new_page, entry);
328 radix_tree_preload_end();
330 * Initiate read into locked page and return.
332 lru_cache_add_anon(new_page);
333 swap_readpage(new_page);
336 radix_tree_preload_end();
337 ClearPageSwapBacked(new_page);
338 __clear_page_locked(new_page);
340 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
341 * clear SWAP_HAS_CACHE flag.
343 swapcache_free(entry, NULL);
344 } while (err != -ENOMEM);
347 page_cache_release(new_page);
352 * swapin_readahead - swap in pages in hope we need them soon
353 * @entry: swap entry of this memory
354 * @gfp_mask: memory allocation flags
355 * @vma: user vma this address belongs to
356 * @addr: target address for mempolicy
358 * Returns the struct page for entry and addr, after queueing swapin.
360 * Primitive swap readahead code. We simply read an aligned block of
361 * (1 << page_cluster) entries in the swap area. This method is chosen
362 * because it doesn't cost us any seek time. We also make sure to queue
363 * the 'original' request together with the readahead ones...
365 * This has been extended to use the NUMA policies from the mm triggering
368 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
370 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
371 struct vm_area_struct *vma, unsigned long addr)
375 unsigned long offset;
376 unsigned long end_offset;
379 * Get starting offset for readaround, and number of pages to read.
380 * Adjust starting address by readbehind (for NUMA interleave case)?
381 * No, it's very unlikely that swap layout would follow vma layout,
382 * more likely that neighbouring swap pages came from the same node:
383 * so use the same "addr" to choose the same node for each swap read.
385 nr_pages = valid_swaphandles(entry, &offset);
386 for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
387 /* Ok, do the async read-ahead now */
388 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
389 gfp_mask, vma, addr);
392 page_cache_release(page);
394 lru_add_drain(); /* Push any new pages onto the LRU now */
395 return read_swap_cache_async(entry, gfp_mask, vma, addr);