clarify get_user_pages() prototype
[linux-2.6.git] / mm / swap_state.c
index 3e3381d..42cd38e 100644 (file)
@@ -17,6 +17,7 @@
 #include <linux/backing-dev.h>
 #include <linux/pagevec.h>
 #include <linux/migrate.h>
+#include <linux/page_cgroup.h>
 
 #include <asm/pgtable.h>
 
@@ -33,13 +34,13 @@ static const struct address_space_operations swap_aops = {
 };
 
 static struct backing_dev_info swap_backing_dev_info = {
-       .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
+       .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
        .unplug_io_fn   = swap_unplug_io_fn,
 };
 
 struct address_space swapper_space = {
        .page_tree      = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
-       .tree_lock      = __RW_LOCK_UNLOCKED(swapper_space.tree_lock),
+       .tree_lock      = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
        .a_ops          = &swap_aops,
        .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
        .backing_dev_info = &swap_backing_dev_info,
@@ -56,10 +57,11 @@ static struct {
 
 void show_swap_cache_info(void)
 {
-       printk("Swap cache: add %lu, delete %lu, find %lu/%lu\n",
+       printk("%lu pages in swap cache\n", total_swapcache_pages);
+       printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
                swap_cache_info.add_total, swap_cache_info.del_total,
                swap_cache_info.find_success, swap_cache_info.find_total);
-       printk("Free swap  = %lukB\n", nr_swap_pages << (PAGE_SHIFT - 10));
+       printk("Free swap  = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
        printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
 }
 
@@ -71,16 +73,17 @@ int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
 {
        int error;
 
-       BUG_ON(!PageLocked(page));
-       BUG_ON(PageSwapCache(page));
-       BUG_ON(PagePrivate(page));
+       VM_BUG_ON(!PageLocked(page));
+       VM_BUG_ON(PageSwapCache(page));
+       VM_BUG_ON(!PageSwapBacked(page));
+
        error = radix_tree_preload(gfp_mask);
        if (!error) {
                page_cache_get(page);
                SetPageSwapCache(page);
                set_page_private(page, entry.val);
 
-               write_lock_irq(&swapper_space.tree_lock);
+               spin_lock_irq(&swapper_space.tree_lock);
                error = radix_tree_insert(&swapper_space.page_tree,
                                                entry.val, page);
                if (likely(!error)) {
@@ -88,7 +91,7 @@ int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
                        __inc_zone_page_state(page, NR_FILE_PAGES);
                        INC_CACHE_INFO(add_total);
                }
-               write_unlock_irq(&swapper_space.tree_lock);
+               spin_unlock_irq(&swapper_space.tree_lock);
                radix_tree_preload_end();
 
                if (unlikely(error)) {
@@ -106,10 +109,9 @@ int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
  */
 void __delete_from_swap_cache(struct page *page)
 {
-       BUG_ON(!PageLocked(page));
-       BUG_ON(!PageSwapCache(page));
-       BUG_ON(PageWriteback(page));
-       BUG_ON(PagePrivate(page));
+       VM_BUG_ON(!PageLocked(page));
+       VM_BUG_ON(!PageSwapCache(page));
+       VM_BUG_ON(PageWriteback(page));
 
        radix_tree_delete(&swapper_space.page_tree, page_private(page));
        set_page_private(page, 0);
@@ -122,18 +124,17 @@ void __delete_from_swap_cache(struct page *page)
 /**
  * add_to_swap - allocate swap space for a page
  * @page: page we want to move to swap
- * @gfp_mask: memory allocation flags
  *
  * Allocate swap space for the page and add the page to the
  * swap cache.  Caller needs to hold the page lock. 
  */
-int add_to_swap(struct page * page, gfp_t gfp_mask)
+int add_to_swap(struct page *page)
 {
        swp_entry_t entry;
        int err;
 
-       BUG_ON(!PageLocked(page));
-       BUG_ON(!PageUptodate(page));
+       VM_BUG_ON(!PageLocked(page));
+       VM_BUG_ON(!PageUptodate(page));
 
        for (;;) {
                entry = get_swap_page();
@@ -152,7 +153,7 @@ int add_to_swap(struct page * page, gfp_t gfp_mask)
                 * Add it to the swap cache and mark it dirty
                 */
                err = add_to_swap_cache(page, entry,
-                               gfp_mask|__GFP_NOMEMALLOC|__GFP_NOWARN);
+                               __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
 
                switch (err) {
                case 0:                         /* Success */
@@ -160,11 +161,11 @@ int add_to_swap(struct page * page, gfp_t gfp_mask)
                        return 1;
                case -EEXIST:
                        /* Raced with "speculative" read_swap_cache_async */
-                       swap_free(entry);
+                       swapcache_free(entry, NULL);
                        continue;
                default:
                        /* -ENOMEM radix-tree allocation failure */
-                       swap_free(entry);
+                       swapcache_free(entry, NULL);
                        return 0;
                }
        }
@@ -182,11 +183,11 @@ void delete_from_swap_cache(struct page *page)
 
        entry.val = page_private(page);
 
-       write_lock_irq(&swapper_space.tree_lock);
+       spin_lock_irq(&swapper_space.tree_lock);
        __delete_from_swap_cache(page);
-       write_unlock_irq(&swapper_space.tree_lock);
+       spin_unlock_irq(&swapper_space.tree_lock);
 
-       swap_free(entry);
+       swapcache_free(entry, page);
        page_cache_release(page);
 }
 
@@ -194,14 +195,14 @@ void delete_from_swap_cache(struct page *page)
  * If we are the only user, then try to free up the swap cache. 
  * 
  * Its ok to check for PageSwapCache without the page lock
- * here because we are going to recheck again inside 
- * exclusive_swap_page() _with_ the lock. 
+ * here because we are going to recheck again inside
+ * try_to_free_swap() _with_ the lock.
  *                                     - Marcelo
  */
 static inline void free_swap_cache(struct page *page)
 {
-       if (PageSwapCache(page) && !TestSetPageLocked(page)) {
-               remove_exclusive_swap_page(page);
+       if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
+               try_to_free_swap(page);
                unlock_page(page);
        }
 }
@@ -290,7 +291,10 @@ struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
                /*
                 * Swap entry may have been freed since our caller observed it.
                 */
-               if (!swap_duplicate(entry))
+               err = swapcache_prepare(entry);
+               if (err == -EEXIST) /* seems racy */
+                       continue;
+               if (err)           /* swp entry is obsolete ? */
                        break;
 
                /*
@@ -301,18 +305,20 @@ struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
                 * re-using the just freed swap entry for an existing page.
                 * May fail (-ENOMEM) if radix-tree node allocation failed.
                 */
-               SetPageLocked(new_page);
+               __set_page_locked(new_page);
+               SetPageSwapBacked(new_page);
                err = add_to_swap_cache(new_page, entry, gfp_mask & GFP_KERNEL);
-               if (!err) {
+               if (likely(!err)) {
                        /*
                         * Initiate read into locked page and return.
                         */
-                       lru_cache_add_active(new_page);
-                       swap_readpage(NULL, new_page);
+                       lru_cache_add_anon(new_page);
+                       swap_readpage(new_page);
                        return new_page;
                }
-               ClearPageLocked(new_page);
-               swap_free(entry);
+               ClearPageSwapBacked(new_page);
+               __clear_page_locked(new_page);
+               swapcache_free(entry, NULL);
        } while (err != -ENOMEM);
 
        if (new_page)