KVM: IRQ ACK notifier should be used with in-kernel irqchip
[linux-2.6.git] / mm / swapfile.c
index 8970c0b..54a9f87 100644 (file)
@@ -5,7 +5,6 @@
  *  Swap reorganised 29.12.95, Stephen Tweedie
  */
 
-#include <linux/config.h>
 #include <linux/mm.h>
 #include <linux/hugetlb.h>
 #include <linux/mman.h>
 #include <linux/rmap.h>
 #include <linux/security.h>
 #include <linux/backing-dev.h>
+#include <linux/mutex.h>
+#include <linux/capability.h>
 #include <linux/syscalls.h>
+#include <linux/memcontrol.h>
 
 #include <asm/pgtable.h>
 #include <asm/tlbflush.h>
 #include <linux/swapops.h>
 
-DEFINE_SPINLOCK(swap_lock);
-unsigned int nr_swapfiles;
+static DEFINE_SPINLOCK(swap_lock);
+static unsigned int nr_swapfiles;
 long total_swap_pages;
 static int swap_overflow;
-
-EXPORT_SYMBOL(total_swap_pages);
+static int least_priority;
 
 static const char Bad_file[] = "Bad swap file entry ";
 static const char Unused_file[] = "Unused swap file entry ";
 static const char Bad_offset[] = "Bad swap offset entry ";
 static const char Unused_offset[] = "Unused swap offset entry ";
 
-struct swap_list_t swap_list = {-1, -1};
+static struct swap_list_t swap_list = {-1, -1};
 
-struct swap_info_struct swap_info[MAX_SWAPFILES];
+static struct swap_info_struct swap_info[MAX_SWAPFILES];
 
-static DECLARE_MUTEX(swapon_sem);
+static DEFINE_MUTEX(swapon_mutex);
 
 /*
  * We need this because the bdev->unplug_fn can sleep and we cannot
  * hold swap_lock while calling the unplug_fn. And swap_lock
- * cannot be turned into a semaphore.
+ * cannot be turned into a mutex.
  */
 static DECLARE_RWSEM(swap_unplug_sem);
 
@@ -116,7 +117,7 @@ static inline unsigned long scan_swap_map(struct swap_info_struct *si)
                                last_in_cluster = offset + SWAPFILE_CLUSTER;
                        else if (offset == last_in_cluster) {
                                spin_lock(&swap_lock);
-                               si->cluster_next = offset-SWAPFILE_CLUSTER-1;
+                               si->cluster_next = offset-SWAPFILE_CLUSTER+1;
                                goto cluster;
                        }
                        if (unlikely(--latency_ration < 0)) {
@@ -213,6 +214,26 @@ noswap:
        return (swp_entry_t) {0};
 }
 
+swp_entry_t get_swap_page_of_type(int type)
+{
+       struct swap_info_struct *si;
+       pgoff_t offset;
+
+       spin_lock(&swap_lock);
+       si = swap_info + type;
+       if (si->flags & SWP_WRITEOK) {
+               nr_swap_pages--;
+               offset = scan_swap_map(si);
+               if (offset) {
+                       spin_unlock(&swap_lock);
+                       return swp_entry(type, offset);
+               }
+               nr_swap_pages++;
+       }
+       spin_unlock(&swap_lock);
+       return (swp_entry_t) {0};
+}
+
 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
 {
        struct swap_info_struct * p;
@@ -323,7 +344,7 @@ int can_share_swap_page(struct page *page)
  * Work out if there are any other processes sharing this
  * swap cache page. Free it if you can. Return success.
  */
-int remove_exclusive_swap_page(struct page *page)
+static int remove_exclusive_swap_page_count(struct page *page, int count)
 {
        int retval;
        struct swap_info_struct * p;
@@ -336,7 +357,7 @@ int remove_exclusive_swap_page(struct page *page)
                return 0;
        if (PageWriteback(page))
                return 0;
-       if (page_count(page) != 2) /* 2: us + cache */
+       if (page_count(page) != count) /* us + cache + ptes */
                return 0;
 
        entry.val = page_private(page);
@@ -348,13 +369,13 @@ int remove_exclusive_swap_page(struct page *page)
        retval = 0;
        if (p->swap_map[swp_offset(entry)] == 1) {
                /* Recheck the page count with the swapcache lock held.. */
-               write_lock_irq(&swapper_space.tree_lock);
-               if ((page_count(page) == 2) && !PageWriteback(page)) {
+               spin_lock_irq(&swapper_space.tree_lock);
+               if ((page_count(page) == count) && !PageWriteback(page)) {
                        __delete_from_swap_cache(page);
                        SetPageDirty(page);
                        retval = 1;
                }
-               write_unlock_irq(&swapper_space.tree_lock);
+               spin_unlock_irq(&swapper_space.tree_lock);
        }
        spin_unlock(&swap_lock);
 
@@ -367,6 +388,25 @@ int remove_exclusive_swap_page(struct page *page)
 }
 
 /*
+ * Most of the time the page should have two references: one for the
+ * process and one for the swap cache.
+ */
+int remove_exclusive_swap_page(struct page *page)
+{
+       return remove_exclusive_swap_page_count(page, 2);
+}
+
+/*
+ * The pageout code holds an extra reference to the page.  That raises
+ * the reference count to test for to 2 for a page that is only in the
+ * swap cache plus 1 for each process that maps the page.
+ */
+int remove_exclusive_swap_page_ref(struct page *page)
+{
+       return remove_exclusive_swap_page_count(page, 2 + page_mapcount(page));
+}
+
+/*
  * Free the swap entry like above, but also try to
  * free the page cache entry if it is the last user.
  */
@@ -375,20 +415,29 @@ void free_swap_and_cache(swp_entry_t entry)
        struct swap_info_struct * p;
        struct page *page = NULL;
 
+       if (is_migration_entry(entry))
+               return;
+
        p = swap_info_get(entry);
        if (p) {
-               if (swap_entry_free(p, swp_offset(entry)) == 1)
-                       page = find_trylock_page(&swapper_space, entry.val);
+               if (swap_entry_free(p, swp_offset(entry)) == 1) {
+                       page = find_get_page(&swapper_space, entry.val);
+                       if (page && !trylock_page(page)) {
+                               page_cache_release(page);
+                               page = NULL;
+                       }
+               }
                spin_unlock(&swap_lock);
        }
        if (page) {
                int one_user;
 
                BUG_ON(PagePrivate(page));
-               page_cache_get(page);
                one_user = (page_count(page) == 2);
                /* Only cache user (+us), or swap space full? Free it! */
-               if (!PageWriteback(page) && (one_user || vm_swap_full())) {
+               /* Also recheck PageSwapCache after page is locked (above) */
+               if (PageSwapCache(page) && !PageWriteback(page) &&
+                                       (one_user || vm_swap_full())) {
                        delete_from_swap_cache(page);
                        SetPageDirty(page);
                }
@@ -397,14 +446,105 @@ void free_swap_and_cache(swp_entry_t entry)
        }
 }
 
+#ifdef CONFIG_HIBERNATION
+/*
+ * Find the swap type that corresponds to given device (if any).
+ *
+ * @offset - number of the PAGE_SIZE-sized block of the device, starting
+ * from 0, in which the swap header is expected to be located.
+ *
+ * This is needed for the suspend to disk (aka swsusp).
+ */
+int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
+{
+       struct block_device *bdev = NULL;
+       int i;
+
+       if (device)
+               bdev = bdget(device);
+
+       spin_lock(&swap_lock);
+       for (i = 0; i < nr_swapfiles; i++) {
+               struct swap_info_struct *sis = swap_info + i;
+
+               if (!(sis->flags & SWP_WRITEOK))
+                       continue;
+
+               if (!bdev) {
+                       if (bdev_p)
+                               *bdev_p = sis->bdev;
+
+                       spin_unlock(&swap_lock);
+                       return i;
+               }
+               if (bdev == sis->bdev) {
+                       struct swap_extent *se;
+
+                       se = list_entry(sis->extent_list.next,
+                                       struct swap_extent, list);
+                       if (se->start_block == offset) {
+                               if (bdev_p)
+                                       *bdev_p = sis->bdev;
+
+                               spin_unlock(&swap_lock);
+                               bdput(bdev);
+                               return i;
+                       }
+               }
+       }
+       spin_unlock(&swap_lock);
+       if (bdev)
+               bdput(bdev);
+
+       return -ENODEV;
+}
+
+/*
+ * Return either the total number of swap pages of given type, or the number
+ * of free pages of that type (depending on @free)
+ *
+ * This is needed for software suspend
+ */
+unsigned int count_swap_pages(int type, int free)
+{
+       unsigned int n = 0;
+
+       if (type < nr_swapfiles) {
+               spin_lock(&swap_lock);
+               if (swap_info[type].flags & SWP_WRITEOK) {
+                       n = swap_info[type].pages;
+                       if (free)
+                               n -= swap_info[type].inuse_pages;
+               }
+               spin_unlock(&swap_lock);
+       }
+       return n;
+}
+#endif
+
 /*
  * No need to decide whether this PTE shares the swap entry with others,
  * just let do_wp_page work it out if a write is requested later - to
  * force COW, vm_page_prot omits write permission from any private vma.
  */
-static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
+static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
                unsigned long addr, swp_entry_t entry, struct page *page)
 {
+       spinlock_t *ptl;
+       pte_t *pte;
+       int ret = 1;
+
+       if (mem_cgroup_charge(page, vma->vm_mm, GFP_KERNEL))
+               ret = -ENOMEM;
+
+       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+       if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
+               if (ret > 0)
+                       mem_cgroup_uncharge_page(page);
+               ret = 0;
+               goto out;
+       }
+
        inc_mm_counter(vma->vm_mm, anon_rss);
        get_page(page);
        set_pte_at(vma->vm_mm, addr, pte,
@@ -416,6 +556,9 @@ static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
         * immediately swapped out again after swapon.
         */
        activate_page(page);
+out:
+       pte_unmap_unlock(pte, ptl);
+       return ret;
 }
 
 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
@@ -424,23 +567,34 @@ static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 {
        pte_t swp_pte = swp_entry_to_pte(entry);
        pte_t *pte;
-       spinlock_t *ptl;
-       int found = 0;
+       int ret = 0;
 
-       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+       /*
+        * We don't actually need pte lock while scanning for swp_pte: since
+        * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
+        * page table while we're scanning; though it could get zapped, and on
+        * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
+        * of unmatched parts which look like swp_pte, so unuse_pte must
+        * recheck under pte lock.  Scanning without pte lock lets it be
+        * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
+        */
+       pte = pte_offset_map(pmd, addr);
        do {
                /*
                 * swapoff spends a _lot_ of time in this loop!
                 * Test inline before going to call unuse_pte.
                 */
                if (unlikely(pte_same(*pte, swp_pte))) {
-                       unuse_pte(vma, pte++, addr, entry, page);
-                       found = 1;
-                       break;
+                       pte_unmap(pte);
+                       ret = unuse_pte(vma, pmd, addr, entry, page);
+                       if (ret)
+                               goto out;
+                       pte = pte_offset_map(pmd, addr);
                }
        } while (pte++, addr += PAGE_SIZE, addr != end);
-       pte_unmap_unlock(pte - 1, ptl);
-       return found;
+       pte_unmap(pte - 1);
+out:
+       return ret;
 }
 
 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
@@ -449,14 +603,16 @@ static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 {
        pmd_t *pmd;
        unsigned long next;
+       int ret;
 
        pmd = pmd_offset(pud, addr);
        do {
                next = pmd_addr_end(addr, end);
                if (pmd_none_or_clear_bad(pmd))
                        continue;
-               if (unuse_pte_range(vma, pmd, addr, next, entry, page))
-                       return 1;
+               ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
+               if (ret)
+                       return ret;
        } while (pmd++, addr = next, addr != end);
        return 0;
 }
@@ -467,14 +623,16 @@ static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 {
        pud_t *pud;
        unsigned long next;
+       int ret;
 
        pud = pud_offset(pgd, addr);
        do {
                next = pud_addr_end(addr, end);
                if (pud_none_or_clear_bad(pud))
                        continue;
-               if (unuse_pmd_range(vma, pud, addr, next, entry, page))
-                       return 1;
+               ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
+               if (ret)
+                       return ret;
        } while (pud++, addr = next, addr != end);
        return 0;
 }
@@ -484,6 +642,7 @@ static int unuse_vma(struct vm_area_struct *vma,
 {
        pgd_t *pgd;
        unsigned long addr, end, next;
+       int ret;
 
        if (page->mapping) {
                addr = page_address_in_vma(page, vma);
@@ -501,8 +660,9 @@ static int unuse_vma(struct vm_area_struct *vma,
                next = pgd_addr_end(addr, end);
                if (pgd_none_or_clear_bad(pgd))
                        continue;
-               if (unuse_pud_range(vma, pgd, addr, next, entry, page))
-                       return 1;
+               ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
+               if (ret)
+                       return ret;
        } while (pgd++, addr = next, addr != end);
        return 0;
 }
@@ -511,11 +671,12 @@ static int unuse_mm(struct mm_struct *mm,
                                swp_entry_t entry, struct page *page)
 {
        struct vm_area_struct *vma;
+       int ret = 0;
 
        if (!down_read_trylock(&mm->mmap_sem)) {
                /*
-                * Activate page so shrink_cache is unlikely to unmap its
-                * ptes while lock is dropped, so swapoff can make progress.
+                * Activate page so shrink_inactive_list is unlikely to unmap
+                * its ptes while lock is dropped, so swapoff can make progress.
                 */
                activate_page(page);
                unlock_page(page);
@@ -523,15 +684,11 @@ static int unuse_mm(struct mm_struct *mm,
                lock_page(page);
        }
        for (vma = mm->mmap; vma; vma = vma->vm_next) {
-               if (vma->anon_vma && unuse_vma(vma, entry, page))
+               if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
                        break;
        }
        up_read(&mm->mmap_sem);
-       /*
-        * Currently unuse_mm cannot fail, but leave error handling
-        * at call sites for now, since we change it from time to time.
-        */
-       return 0;
+       return (ret < 0)? ret: 0;
 }
 
 /*
@@ -626,7 +783,8 @@ static int try_to_unuse(unsigned int type)
                 */
                swap_map = &si->swap_map[i];
                entry = swp_entry(type, i);
-               page = read_swap_cache_async(entry, NULL, 0);
+               page = read_swap_cache_async(entry,
+                                       GFP_HIGHUSER_MOVABLE, NULL, 0);
                if (!page) {
                        /*
                         * Either swap_duplicate() failed because entry
@@ -685,13 +843,11 @@ static int try_to_unuse(unsigned int type)
                        atomic_inc(&new_start_mm->mm_users);
                        atomic_inc(&prev_mm->mm_users);
                        spin_lock(&mmlist_lock);
-                       while (*swap_map > 1 && !retval &&
+                       while (*swap_map > 1 && !retval && !shmem &&
                                        (p = p->next) != &start_mm->mmlist) {
                                mm = list_entry(p, struct mm_struct, mmlist);
-                               if (atomic_inc_return(&mm->mm_users) == 1) {
-                                       atomic_dec(&mm->mm_users);
+                               if (!atomic_inc_not_zero(&mm->mm_users))
                                        continue;
-                               }
                                spin_unlock(&mmlist_lock);
                                mmput(prev_mm);
                                prev_mm = mm;
@@ -719,6 +875,13 @@ static int try_to_unuse(unsigned int type)
                        mmput(start_mm);
                        start_mm = new_start_mm;
                }
+               if (shmem) {
+                       /* page has already been unlocked and released */
+                       if (shmem > 0)
+                               continue;
+                       retval = shmem;
+                       break;
+               }
                if (retval) {
                        unlock_page(page);
                        page_cache_release(page);
@@ -757,12 +920,6 @@ static int try_to_unuse(unsigned int type)
                 * read from disk into another page.  Splitting into two
                 * pages would be incorrect if swap supported "shared
                 * private" pages, but they are handled by tmpfs files.
-                *
-                * Note shmem_unuse already deleted a swappage from
-                * the swap cache, unless the move to filepage failed:
-                * in which case it left swappage in cache, lowered its
-                * swap count to pass quickly through the loops above,
-                * and now we must reincrement count to try again later.
                 */
                if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
                        struct writeback_control wbc = {
@@ -773,17 +930,13 @@ static int try_to_unuse(unsigned int type)
                        lock_page(page);
                        wait_on_page_writeback(page);
                }
-               if (PageSwapCache(page)) {
-                       if (shmem)
-                               swap_duplicate(entry);
-                       else
-                               delete_from_swap_cache(page);
-               }
+               if (PageSwapCache(page))
+                       delete_from_swap_cache(page);
 
                /*
                 * So we could skip searching mms once swap count went
                 * to 1, we did not mark any present ptes as dirty: must
-                * mark page dirty so shrink_list will preserve it.
+                * mark page dirty so shrink_page_list will preserve it.
                 */
                SetPageDirty(page);
                unlock_page(page);
@@ -849,6 +1002,23 @@ sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
        }
 }
 
+#ifdef CONFIG_HIBERNATION
+/*
+ * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
+ * corresponding to given index in swap_info (swap type).
+ */
+sector_t swapdev_block(int swap_type, pgoff_t offset)
+{
+       struct swap_info_struct *sis;
+
+       if (swap_type >= nr_swapfiles)
+               return 0;
+
+       sis = swap_info + swap_type;
+       return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
+}
+#endif /* CONFIG_HIBERNATION */
+
 /*
  * Free all of a swapdev's extent information
  */
@@ -1110,6 +1280,11 @@ asmlinkage long sys_swapoff(const char __user * specialfile)
                /* just pick something that's safe... */
                swap_list.next = swap_list.head;
        }
+       if (p->prio < 0) {
+               for (i = p->next; i >= 0; i = swap_info[i].next)
+                       swap_info[i].prio = p->prio--;
+               least_priority++;
+       }
        nr_swap_pages -= p->pages;
        total_swap_pages -= p->pages;
        p->flags &= ~SWP_WRITEOK;
@@ -1122,9 +1297,14 @@ asmlinkage long sys_swapoff(const char __user * specialfile)
        if (err) {
                /* re-insert swap space back into swap_list */
                spin_lock(&swap_lock);
-               for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
+               if (p->prio < 0)
+                       p->prio = --least_priority;
+               prev = -1;
+               for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
                        if (p->prio >= swap_info[i].prio)
                                break;
+                       prev = i;
+               }
                p->next = i;
                if (prev < 0)
                        swap_list.head = swap_list.next = p - swap_info;
@@ -1142,7 +1322,7 @@ asmlinkage long sys_swapoff(const char __user * specialfile)
        up_write(&swap_unplug_sem);
 
        destroy_swap_extents(p);
-       down(&swapon_sem);
+       mutex_lock(&swapon_mutex);
        spin_lock(&swap_lock);
        drain_mmlist();
 
@@ -1161,7 +1341,7 @@ asmlinkage long sys_swapoff(const char __user * specialfile)
        p->swap_map = NULL;
        p->flags = 0;
        spin_unlock(&swap_lock);
-       up(&swapon_sem);
+       mutex_unlock(&swapon_mutex);
        vfree(swap_map);
        inode = mapping->host;
        if (S_ISBLK(inode->i_mode)) {
@@ -1169,9 +1349,9 @@ asmlinkage long sys_swapoff(const char __user * specialfile)
                set_blocksize(bdev, p->old_block_size);
                bd_release(bdev);
        } else {
-               down(&inode->i_sem);
+               mutex_lock(&inode->i_mutex);
                inode->i_flags &= ~S_SWAPFILE;
-               up(&inode->i_sem);
+               mutex_unlock(&inode->i_mutex);
        }
        filp_close(swap_file, NULL);
        err = 0;
@@ -1190,12 +1370,15 @@ static void *swap_start(struct seq_file *swap, loff_t *pos)
        int i;
        loff_t l = *pos;
 
-       down(&swapon_sem);
+       mutex_lock(&swapon_mutex);
+
+       if (!l)
+               return SEQ_START_TOKEN;
 
        for (i = 0; i < nr_swapfiles; i++, ptr++) {
                if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
                        continue;
-               if (!l--)
+               if (!--l)
                        return ptr;
        }
 
@@ -1204,10 +1387,17 @@ static void *swap_start(struct seq_file *swap, loff_t *pos)
 
 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
 {
-       struct swap_info_struct *ptr = v;
+       struct swap_info_struct *ptr;
        struct swap_info_struct *endptr = swap_info + nr_swapfiles;
 
-       for (++ptr; ptr < endptr; ptr++) {
+       if (v == SEQ_START_TOKEN)
+               ptr = swap_info;
+       else {
+               ptr = v;
+               ptr++;
+       }
+
+       for (; ptr < endptr; ptr++) {
                if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
                        continue;
                ++*pos;
@@ -1219,7 +1409,7 @@ static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
 
 static void swap_stop(struct seq_file *swap, void *v)
 {
-       up(&swapon_sem);
+       mutex_unlock(&swapon_mutex);
 }
 
 static int swap_show(struct seq_file *swap, void *v)
@@ -1228,14 +1418,16 @@ static int swap_show(struct seq_file *swap, void *v)
        struct file *file;
        int len;
 
-       if (v == swap_info)
-               seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
+       if (ptr == SEQ_START_TOKEN) {
+               seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
+               return 0;
+       }
 
        file = ptr->swap_file;
-       len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
+       len = seq_path(swap, &file->f_path, " \t\n\\");
        seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
                       len < 40 ? 40 - len : 1, " ",
-                      S_ISBLK(file->f_dentry->d_inode->i_mode) ?
+                      S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
                                "partition" : "file\t",
                       ptr->pages << (PAGE_SHIFT - 10),
                       ptr->inuse_pages << (PAGE_SHIFT - 10),
@@ -1243,7 +1435,7 @@ static int swap_show(struct seq_file *swap, void *v)
        return 0;
 }
 
-static struct seq_operations swaps_op = {
+static const struct seq_operations swaps_op = {
        .start =        swap_start,
        .next =         swap_next,
        .stop =         swap_stop,
@@ -1255,7 +1447,7 @@ static int swaps_open(struct inode *inode, struct file *file)
        return seq_open(file, &swaps_op);
 }
 
-static struct file_operations proc_swaps_operations = {
+static const struct file_operations proc_swaps_operations = {
        .open           = swaps_open,
        .read           = seq_read,
        .llseek         = seq_lseek,
@@ -1264,16 +1456,21 @@ static struct file_operations proc_swaps_operations = {
 
 static int __init procswaps_init(void)
 {
-       struct proc_dir_entry *entry;
-
-       entry = create_proc_entry("swaps", 0, NULL);
-       if (entry)
-               entry->proc_fops = &proc_swaps_operations;
+       proc_create("swaps", 0, NULL, &proc_swaps_operations);
        return 0;
 }
 __initcall(procswaps_init);
 #endif /* CONFIG_PROC_FS */
 
+#ifdef MAX_SWAPFILES_CHECK
+static int __init max_swapfiles_check(void)
+{
+       MAX_SWAPFILES_CHECK();
+       return 0;
+}
+late_initcall(max_swapfiles_check);
+#endif
+
 /*
  * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
  *
@@ -1289,7 +1486,6 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
        unsigned int type;
        int i, prev;
        int error;
-       static int least_priority;
        union swap_header *swap_header = NULL;
        int swap_header_version;
        unsigned int nr_good_pages = 0;
@@ -1297,7 +1493,7 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
        sector_t span;
        unsigned long maxpages = 1;
        int swapfilesize;
-       unsigned short *swap_map;
+       unsigned short *swap_map = NULL;
        struct page *page = NULL;
        struct inode *inode = NULL;
        int did_down = 0;
@@ -1310,40 +1506,16 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
                if (!(p->flags & SWP_USED))
                        break;
        error = -EPERM;
-       /*
-        * Test if adding another swap device is possible. There are
-        * two limiting factors: 1) the number of bits for the swap
-        * type swp_entry_t definition and 2) the number of bits for
-        * the swap type in the swap ptes as defined by the different
-        * architectures. To honor both limitations a swap entry
-        * with swap offset 0 and swap type ~0UL is created, encoded
-        * to a swap pte, decoded to a swp_entry_t again and finally
-        * the swap type part is extracted. This will mask all bits
-        * from the initial ~0UL that can't be encoded in either the
-        * swp_entry_t or the architecture definition of a swap pte.
-        */
-       if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
+       if (type >= MAX_SWAPFILES) {
                spin_unlock(&swap_lock);
                goto out;
        }
        if (type >= nr_swapfiles)
                nr_swapfiles = type+1;
+       memset(p, 0, sizeof(*p));
        INIT_LIST_HEAD(&p->extent_list);
        p->flags = SWP_USED;
-       p->swap_file = NULL;
-       p->old_block_size = 0;
-       p->swap_map = NULL;
-       p->lowest_bit = 0;
-       p->highest_bit = 0;
-       p->cluster_nr = 0;
-       p->inuse_pages = 0;
        p->next = -1;
-       if (swap_flags & SWAP_FLAG_PREFER) {
-               p->prio =
-                 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
-       } else {
-               p->prio = --least_priority;
-       }
        spin_unlock(&swap_lock);
        name = getname(specialfile);
        error = PTR_ERR(name);
@@ -1388,7 +1560,7 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
                p->bdev = bdev;
        } else if (S_ISREG(inode->i_mode)) {
                p->bdev = inode->i_sb->s_bdev;
-               down(&inode->i_sem);
+               mutex_lock(&inode->i_mutex);
                did_down = 1;
                if (IS_SWAPFILE(inode)) {
                        error = -EBUSY;
@@ -1407,15 +1579,11 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
                error = -EINVAL;
                goto bad_swap;
        }
-       page = read_cache_page(mapping, 0,
-                       (filler_t *)mapping->a_ops->readpage, swap_file);
+       page = read_mapping_page(mapping, 0, swap_file);
        if (IS_ERR(page)) {
                error = PTR_ERR(page);
                goto bad_swap;
        }
-       wait_on_page_locked(page);
-       if (!PageUptodate(page))
-               goto bad_swap;
        kmap(page);
        swap_header = page_address(page);
 
@@ -1424,7 +1592,7 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
        else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
                swap_header_version = 2;
        else {
-               printk("Unable to find swap-space signature\n");
+               printk(KERN_ERR "Unable to find swap-space signature\n");
                error = -EINVAL;
                goto bad_swap;
        }
@@ -1436,6 +1604,14 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
                error = -EINVAL;
                goto bad_swap;
        case 2:
+               /* swap partition endianess hack... */
+               if (swab32(swap_header->info.version) == 1) {
+                       swab32s(&swap_header->info.version);
+                       swab32s(&swap_header->info.last_page);
+                       swab32s(&swap_header->info.nr_badpages);
+                       for (i = 0; i < swap_header->info.nr_badpages; i++)
+                               swab32s(&swap_header->info.badpages[i]);
+               }
                /* Check the swap header's sub-version and the size of
                    the swap file and bad block lists */
                if (swap_header->info.version != 1) {
@@ -1471,41 +1647,41 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
                error = -EINVAL;
                if (!maxpages)
                        goto bad_swap;
+               if (swapfilesize && maxpages > swapfilesize) {
+                       printk(KERN_WARNING
+                              "Swap area shorter than signature indicates\n");
+                       goto bad_swap;
+               }
                if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
                        goto bad_swap;
                if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
                        goto bad_swap;
-               
+
                /* OK, set up the swap map and apply the bad block list */
-               if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
+               swap_map = vmalloc(maxpages * sizeof(short));
+               if (!swap_map) {
                        error = -ENOMEM;
                        goto bad_swap;
                }
 
                error = 0;
-               memset(p->swap_map, 0, maxpages * sizeof(short));
-               for (i=0; i<swap_header->info.nr_badpages; i++) {
-                       int page = swap_header->info.badpages[i];
-                       if (page <= 0 || page >= swap_header->info.last_page)
+               memset(swap_map, 0, maxpages * sizeof(short));
+               for (i = 0; i < swap_header->info.nr_badpages; i++) {
+                       int page_nr = swap_header->info.badpages[i];
+                       if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
                                error = -EINVAL;
                        else
-                               p->swap_map[page] = SWAP_MAP_BAD;
+                               swap_map[page_nr] = SWAP_MAP_BAD;
                }
                nr_good_pages = swap_header->info.last_page -
                                swap_header->info.nr_badpages -
                                1 /* header page */;
-               if (error) 
+               if (error)
                        goto bad_swap;
        }
 
-       if (swapfilesize && maxpages > swapfilesize) {
-               printk(KERN_WARNING
-                      "Swap area shorter than signature indicates\n");
-               error = -EINVAL;
-               goto bad_swap;
-       }
        if (nr_good_pages) {
-               p->swap_map[0] = SWAP_MAP_BAD;
+               swap_map[0] = SWAP_MAP_BAD;
                p->max = maxpages;
                p->pages = nr_good_pages;
                nr_extents = setup_swap_extents(p, &span);
@@ -1521,8 +1697,14 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
                goto bad_swap;
        }
 
-       down(&swapon_sem);
+       mutex_lock(&swapon_mutex);
        spin_lock(&swap_lock);
+       if (swap_flags & SWAP_FLAG_PREFER)
+               p->prio =
+                 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
+       else
+               p->prio = --least_priority;
+       p->swap_map = swap_map;
        p->flags = SWP_ACTIVE;
        nr_swap_pages += nr_good_pages;
        total_swap_pages += nr_good_pages;
@@ -1547,7 +1729,7 @@ asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
                swap_info[prev].next = p - swap_info;
        }
        spin_unlock(&swap_lock);
-       up(&swapon_sem);
+       mutex_unlock(&swapon_mutex);
        error = 0;
        goto out;
 bad_swap:
@@ -1558,12 +1740,8 @@ bad_swap:
        destroy_swap_extents(p);
 bad_swap_2:
        spin_lock(&swap_lock);
-       swap_map = p->swap_map;
        p->swap_file = NULL;
-       p->swap_map = NULL;
        p->flags = 0;
-       if (!(swap_flags & SWAP_FLAG_PREFER))
-               ++least_priority;
        spin_unlock(&swap_lock);
        vfree(swap_map);
        if (swap_file)
@@ -1578,7 +1756,7 @@ out:
        if (did_down) {
                if (!error)
                        inode->i_flags |= S_SWAPFILE;
-               up(&inode->i_sem);
+               mutex_unlock(&inode->i_mutex);
        }
        return error;
 }
@@ -1612,6 +1790,9 @@ int swap_duplicate(swp_entry_t entry)
        unsigned long offset, type;
        int result = 0;
 
+       if (is_migration_entry(entry))
+               return 1;
+
        type = swp_type(entry);
        if (type >= nr_swapfiles)
                goto bad_file;
@@ -1651,30 +1832,48 @@ get_swap_info_struct(unsigned type)
  */
 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
 {
-       int ret = 0, i = 1 << page_cluster;
-       unsigned long toff;
-       struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
+       struct swap_info_struct *si;
+       int our_page_cluster = page_cluster;
+       pgoff_t target, toff;
+       pgoff_t base, end;
+       int nr_pages = 0;
 
-       if (!page_cluster)      /* no readahead */
+       if (!our_page_cluster)  /* no readahead */
                return 0;
-       toff = (swp_offset(entry) >> page_cluster) << page_cluster;
-       if (!toff)              /* first page is swap header */
-               toff++, i--;
-       *offset = toff;
+
+       si = &swap_info[swp_type(entry)];
+       target = swp_offset(entry);
+       base = (target >> our_page_cluster) << our_page_cluster;
+       end = base + (1 << our_page_cluster);
+       if (!base)              /* first page is swap header */
+               base++;
 
        spin_lock(&swap_lock);
-       do {
-               /* Don't read-ahead past the end of the swap area */
-               if (toff >= swapdev->max)
+       if (end > si->max)      /* don't go beyond end of map */
+               end = si->max;
+
+       /* Count contiguous allocated slots above our target */
+       for (toff = target; ++toff < end; nr_pages++) {
+               /* Don't read in free or bad pages */
+               if (!si->swap_map[toff])
+                       break;
+               if (si->swap_map[toff] == SWAP_MAP_BAD)
                        break;
+       }
+       /* Count contiguous allocated slots below our target */
+       for (toff = target; --toff >= base; nr_pages++) {
                /* Don't read in free or bad pages */
-               if (!swapdev->swap_map[toff])
+               if (!si->swap_map[toff])
                        break;
-               if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
+               if (si->swap_map[toff] == SWAP_MAP_BAD)
                        break;
-               toff++;
-               ret++;
-       } while (--i);
+       }
        spin_unlock(&swap_lock);
-       return ret;
+
+       /*
+        * Indicate starting offset, and return number of pages to get:
+        * if only 1, say 0, since there's then no readahead to be done.
+        */
+       *offset = ++toff;
+       return nr_pages? ++nr_pages: 0;
 }