ksm: let shared pages be swappable

Initial implementation for swapping out KSM's shared pages: add
page_referenced_ksm() and try_to_unmap_ksm(), which rmap.c calls when
faced with a PageKsm page.

Most of what's needed can be got from the rmap_items listed from the
stable_node of the ksm page, without discovering the actual vma: so in
this patch just fake up a struct vma for page_referenced_one() or
try_to_unmap_one(), then refine that in the next patch.

Add VM_NONLINEAR to ksm_madvise()'s list of exclusions: it has always been
implicit there (being only set with VM_SHARED, already excluded), but
let's make it explicit, to help justify the lack of nonlinear unmap.

Rely on the page lock to protect against concurrent modifications to that
page's node of the stable tree.

The awkward part is not swapout but swapin: do_swap_page() and
page_add_anon_rmap() now have to allow for new possibilities - perhaps a
ksm page still in swapcache, perhaps a swapcache page associated with one
location in one anon_vma now needed for another location or anon_vma.
(And the vma might even be no longer VM_MERGEABLE when that happens.)

ksm_might_need_to_copy() checks for that case, and supplies a duplicate
page when necessary, simply leaving it to a subsequent pass of ksmd to
rediscover the identity and merge them back into one ksm page.
Disappointingly primitive: but the alternative would have to accumulate
unswappable info about the swapped out ksm pages, limiting swappability.

Remove page_add_ksm_rmap(): page_add_anon_rmap() now has to allow for the
particular case it was handling, so just use it instead.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

4 files changed, 211 insertions, 43 deletions

diff --git a/mm/ksm.c b/mm/ksm.c
index af5f571185d5..2f58ceebfe8f 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -196,6 +196,13 @@ static DECLARE_WAIT_QUEUE_HEAD(ksm_thread_wait);
 static DEFINE_MUTEX(ksm_thread_mutex);
 static DEFINE_SPINLOCK(ksm_mmlist_lock);
 
+/*
+ * Temporary hack for page_referenced_ksm() and try_to_unmap_ksm(),
+ * later we rework things a little to get the right vma to them.
+ */
+static DEFINE_SPINLOCK(ksm_fallback_vma_lock);
+static struct vm_area_struct ksm_fallback_vma;
+
 #define KSM_KMEM_CACHE(__struct, __flags) kmem_cache_create("ksm_"#__struct,\
                 sizeof(struct __struct), __alignof__(struct __struct),\
                 (__flags), NULL)
@@ -445,14 +452,20 @@ static void remove_rmap_item_from_tree(struct rmap_item *rmap_item)
 {
         if (rmap_item->address & STABLE_FLAG) {
                 struct stable_node *stable_node;
+                struct page *page;
 
                 stable_node = rmap_item->head;
+                page = stable_node->page;
+                lock_page(page);
+
                 hlist_del(&rmap_item->hlist);
-                if (stable_node->hlist.first)
+                if (stable_node->hlist.first) {
+                        unlock_page(page);
                         ksm_pages_sharing--;
-                else {
-                        set_page_stable_node(stable_node->page, NULL);
-                        put_page(stable_node->page);
+                } else {
+                        set_page_stable_node(page, NULL);
+                        unlock_page(page);
+                        put_page(page);
 
                         rb_erase(&stable_node->node, &root_stable_tree);
                         free_stable_node(stable_node);
@@ -710,7 +723,7 @@ static int replace_page(struct vm_area_struct *vma, struct page *page,
         }
 
         get_page(kpage);
-        page_add_ksm_rmap(kpage);
+        page_add_anon_rmap(kpage, vma, addr);
 
         flush_cache_page(vma, addr, pte_pfn(*ptep));
         ptep_clear_flush(vma, addr, ptep);
@@ -763,8 +776,16 @@ static int try_to_merge_one_page(struct vm_area_struct *vma,
             pages_identical(page, kpage))
                 err = replace_page(vma, page, kpage, orig_pte);
 
-        if ((vma->vm_flags & VM_LOCKED) && !err)
+        if ((vma->vm_flags & VM_LOCKED) && !err) {
                 munlock_vma_page(page);
+                if (!PageMlocked(kpage)) {
+                        unlock_page(page);
+                        lru_add_drain();
+                        lock_page(kpage);
+                        mlock_vma_page(kpage);
+                        page = kpage;           /* for final unlock */
+                }
+        }
 
         unlock_page(page);
 out:
@@ -841,7 +862,11 @@ static struct page *try_to_merge_two_pages(struct rmap_item *rmap_item,
 
         copy_user_highpage(kpage, page, rmap_item->address, vma);
 
+        SetPageDirty(kpage);
+        __SetPageUptodate(kpage);
+        SetPageSwapBacked(kpage);
         set_page_stable_node(kpage, NULL);      /* mark it PageKsm */
+        lru_cache_add_lru(kpage, LRU_ACTIVE_ANON);
 
         err = try_to_merge_one_page(vma, page, kpage);
 up:
@@ -1071,7 +1096,9 @@ static void cmp_and_merge_page(struct page *page, struct rmap_item *rmap_item)
                          * The page was successfully merged:
                          * add its rmap_item to the stable tree.
                          */
+                        lock_page(kpage);
                         stable_tree_append(rmap_item, stable_node);
+                        unlock_page(kpage);
                 }
                 put_page(kpage);
                 return;
@@ -1112,11 +1139,13 @@ static void cmp_and_merge_page(struct page *page, struct rmap_item *rmap_item)
                 if (kpage) {
                         remove_rmap_item_from_tree(tree_rmap_item);
 
+                        lock_page(kpage);
                         stable_node = stable_tree_insert(kpage);
                         if (stable_node) {
                                 stable_tree_append(tree_rmap_item, stable_node);
                                 stable_tree_append(rmap_item, stable_node);
                         }
+                        unlock_page(kpage);
                         put_page(kpage);
 
                         /*
@@ -1285,14 +1314,6 @@ static void ksm_do_scan(unsigned int scan_npages)
                         return;
                 if (!PageKsm(page) || !in_stable_tree(rmap_item))
                         cmp_and_merge_page(page, rmap_item);
-                else if (page_mapcount(page) == 1) {
-                        /*
-                         * Replace now-unshared ksm page by ordinary page.
-                         */
-                        break_cow(rmap_item);
-                        remove_rmap_item_from_tree(rmap_item);
-                        rmap_item->oldchecksum = calc_checksum(page);
-                }
                 put_page(page);
         }
 }
@@ -1337,7 +1358,7 @@ int ksm_madvise(struct vm_area_struct *vma, unsigned long start,
                 if (*vm_flags & (VM_MERGEABLE | VM_SHARED  | VM_MAYSHARE   |
                                  VM_PFNMAP    | VM_IO      | VM_DONTEXPAND |
                                  VM_RESERVED  | VM_HUGETLB | VM_INSERTPAGE |
-                                 VM_MIXEDMAP  | VM_SAO))
+                                 VM_NONLINEAR | VM_MIXEDMAP | VM_SAO))
                         return 0;               /* just ignore the advice */
 
                 if (!test_bit(MMF_VM_MERGEABLE, &mm->flags)) {
@@ -1435,6 +1456,127 @@ void __ksm_exit(struct mm_struct *mm)
         }
 }
 
+struct page *ksm_does_need_to_copy(struct page *page,
+                        struct vm_area_struct *vma, unsigned long address)
+{
+        struct page *new_page;
+
+        unlock_page(page);      /* any racers will COW it, not modify it */
+
+        new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
+        if (new_page) {
+                copy_user_highpage(new_page, page, address, vma);
+
+                SetPageDirty(new_page);
+                __SetPageUptodate(new_page);
+                SetPageSwapBacked(new_page);
+                __set_page_locked(new_page);
+
+                if (page_evictable(new_page, vma))
+                        lru_cache_add_lru(new_page, LRU_ACTIVE_ANON);
+                else
+                        add_page_to_unevictable_list(new_page);
+        }
+
+        page_cache_release(page);
+        return new_page;
+}
+
+int page_referenced_ksm(struct page *page, struct mem_cgroup *memcg,
+                        unsigned long *vm_flags)
+{
+        struct stable_node *stable_node;
+        struct rmap_item *rmap_item;
+        struct hlist_node *hlist;
+        unsigned int mapcount = page_mapcount(page);
+        int referenced = 0;
+        struct vm_area_struct *vma;
+
+        VM_BUG_ON(!PageKsm(page));
+        VM_BUG_ON(!PageLocked(page));
+
+        stable_node = page_stable_node(page);
+        if (!stable_node)
+                return 0;
+
+        /*
+         * Temporary hack: really we need anon_vma in rmap_item, to
+         * provide the correct vma, and to find recently forked instances.
+         * Use zalloc to avoid weirdness if any other fields are involved.
+         */
+        vma = kmem_cache_zalloc(vm_area_cachep, GFP_ATOMIC);
+        if (!vma) {
+                spin_lock(&ksm_fallback_vma_lock);
+                vma = &ksm_fallback_vma;
+        }
+
+        hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
+                if (memcg && !mm_match_cgroup(rmap_item->mm, memcg))
+                        continue;
+
+                vma->vm_mm = rmap_item->mm;
+                vma->vm_start = rmap_item->address;
+                vma->vm_end = vma->vm_start + PAGE_SIZE;
+
+                referenced += page_referenced_one(page, vma,
+                                rmap_item->address, &mapcount, vm_flags);
+                if (!mapcount)
+                        goto out;
+        }
+out:
+        if (vma == &ksm_fallback_vma)
+                spin_unlock(&ksm_fallback_vma_lock);
+        else
+                kmem_cache_free(vm_area_cachep, vma);
+        return referenced;
+}
+
+int try_to_unmap_ksm(struct page *page, enum ttu_flags flags)
+{
+        struct stable_node *stable_node;
+        struct hlist_node *hlist;
+        struct rmap_item *rmap_item;
+        int ret = SWAP_AGAIN;
+        struct vm_area_struct *vma;
+
+        VM_BUG_ON(!PageKsm(page));
+        VM_BUG_ON(!PageLocked(page));
+
+        stable_node = page_stable_node(page);
+        if (!stable_node)
+                return SWAP_FAIL;
+
+        /*
+         * Temporary hack: really we need anon_vma in rmap_item, to
+         * provide the correct vma, and to find recently forked instances.
+         * Use zalloc to avoid weirdness if any other fields are involved.
+         */
+        if (TTU_ACTION(flags) != TTU_UNMAP)
+                return SWAP_FAIL;
+
+        vma = kmem_cache_zalloc(vm_area_cachep, GFP_ATOMIC);
+        if (!vma) {
+                spin_lock(&ksm_fallback_vma_lock);
+                vma = &ksm_fallback_vma;
+        }
+
+        hlist_for_each_entry(rmap_item, hlist, &stable_node->hlist, hlist) {
+                vma->vm_mm = rmap_item->mm;
+                vma->vm_start = rmap_item->address;
+                vma->vm_end = vma->vm_start + PAGE_SIZE;
+
+                ret = try_to_unmap_one(page, vma, rmap_item->address, flags);
+                if (ret != SWAP_AGAIN || !page_mapped(page))
+                        goto out;
+        }
+out:
+        if (vma == &ksm_fallback_vma)
+                spin_unlock(&ksm_fallback_vma_lock);
+        else
+                kmem_cache_free(vm_area_cachep, vma);
+        return ret;
+}
+
 #ifdef CONFIG_SYSFS
 /*
  * This all compiles without CONFIG_SYSFS, but is a waste of space.
diff --git a/mm/memory.c b/mm/memory.c
index 1c9dc46da3db..a54b2c498444 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2561,6 +2561,12 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
         lock_page(page);
         delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
 
+        page = ksm_might_need_to_copy(page, vma, address);
+        if (!page) {
+                ret = VM_FAULT_OOM;
+                goto out;
+        }
+
         if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
                 ret = VM_FAULT_OOM;
                 goto out_page;
diff --git a/mm/rmap.c b/mm/rmap.c
index ebee81688736..869aaa3206a2 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -49,6 +49,7 @@
 #include <linux/swapops.h>
 #include <linux/slab.h>
 #include <linux/init.h>
+#include <linux/ksm.h>
 #include <linux/rmap.h>
 #include <linux/rcupdate.h>
 #include <linux/module.h>
@@ -336,9 +337,9 @@ int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
  * Subfunctions of page_referenced: page_referenced_one called
  * repeatedly from either page_referenced_anon or page_referenced_file.
  */
-static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
-                               unsigned long address, unsigned int *mapcount,
-                               unsigned long *vm_flags)
+int page_referenced_one(struct page *page, struct vm_area_struct *vma,
+                        unsigned long address, unsigned int *mapcount,
+                        unsigned long *vm_flags)
 {
         struct mm_struct *mm = vma->vm_mm;
         pte_t *pte;
@@ -507,28 +508,33 @@ int page_referenced(struct page *page,
                     unsigned long *vm_flags)
 {
         int referenced = 0;
+        int we_locked = 0;
 
         if (TestClearPageReferenced(page))
                 referenced++;
 
         *vm_flags = 0;
         if (page_mapped(page) && page_rmapping(page)) {
-                if (PageAnon(page))
+                if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
+                        we_locked = trylock_page(page);
+                        if (!we_locked) {
+                                referenced++;
+                                goto out;
+                        }
+                }
+                if (unlikely(PageKsm(page)))
+                        referenced += page_referenced_ksm(page, mem_cont,
+                                                                vm_flags);
+                else if (PageAnon(page))
                         referenced += page_referenced_anon(page, mem_cont,
                                                                 vm_flags);
-                else if (is_locked)
+                else if (page->mapping)
                         referenced += page_referenced_file(page, mem_cont,
                                                                 vm_flags);
-                else if (!trylock_page(page))
-                        referenced++;
-                else {
-                        if (page->mapping)
-                                referenced += page_referenced_file(page,
-                                                        mem_cont, vm_flags);
+                if (we_locked)
                         unlock_page(page);
-                }
         }
-
+out:
         if (page_test_and_clear_young(page))
                 referenced++;
 
@@ -620,14 +626,7 @@ static void __page_set_anon_rmap(struct page *page,
         BUG_ON(!anon_vma);
         anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
         page->mapping = (struct address_space *) anon_vma;
-
         page->index = linear_page_index(vma, address);
-
-        /*
-         * nr_mapped state can be updated without turning off
-         * interrupts because it is not modified via interrupt.
-         */
-        __inc_zone_page_state(page, NR_ANON_PAGES);
 }
 
 /**
@@ -665,14 +664,21 @@ static void __page_check_anon_rmap(struct page *page,
  * @vma:        the vm area in which the mapping is added
  * @address:    the user virtual address mapped
  *
- * The caller needs to hold the pte lock and the page must be locked.
+ * The caller needs to hold the pte lock, and the page must be locked in
+ * the anon_vma case: to serialize mapping,index checking after setting.
  */
 void page_add_anon_rmap(struct page *page,
         struct vm_area_struct *vma, unsigned long address)
 {
+        int first = atomic_inc_and_test(&page->_mapcount);
+        if (first)
+                __inc_zone_page_state(page, NR_ANON_PAGES);
+        if (unlikely(PageKsm(page)))
+                return;
+
         VM_BUG_ON(!PageLocked(page));
         VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
-        if (atomic_inc_and_test(&page->_mapcount))
+        if (first)
                 __page_set_anon_rmap(page, vma, address);
         else
                 __page_check_anon_rmap(page, vma, address);
@@ -694,6 +700,7 @@ void page_add_new_anon_rmap(struct page *page,
         VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
         SetPageSwapBacked(page);
         atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
+        __inc_zone_page_state(page, NR_ANON_PAGES);
         __page_set_anon_rmap(page, vma, address);
         if (page_evictable(page, vma))
                 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
@@ -760,8 +767,8 @@ void page_remove_rmap(struct page *page)
  * Subfunctions of try_to_unmap: try_to_unmap_one called
  * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
  */
-static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
-                            unsigned long address, enum ttu_flags flags)
+int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
+                     unsigned long address, enum ttu_flags flags)
 {
         struct mm_struct *mm = vma->vm_mm;
         pte_t *pte;
@@ -1156,7 +1163,9 @@ int try_to_unmap(struct page *page, enum ttu_flags flags)
 
         BUG_ON(!PageLocked(page));
 
-        if (PageAnon(page))
+        if (unlikely(PageKsm(page)))
+                ret = try_to_unmap_ksm(page, flags);
+        else if (PageAnon(page))
                 ret = try_to_unmap_anon(page, flags);
         else
                 ret = try_to_unmap_file(page, flags);
@@ -1177,15 +1186,17 @@ int try_to_unmap(struct page *page, enum ttu_flags flags)
  *
  * SWAP_AGAIN   - no vma is holding page mlocked, or,
  * SWAP_AGAIN   - page mapped in mlocked vma -- couldn't acquire mmap sem
+ * SWAP_FAIL    - page cannot be located at present
  * SWAP_MLOCK   - page is now mlocked.
  */
 int try_to_munlock(struct page *page)
 {
         VM_BUG_ON(!PageLocked(page) || PageLRU(page));
 
-        if (PageAnon(page))
+        if (unlikely(PageKsm(page)))
+                return try_to_unmap_ksm(page, TTU_MUNLOCK);
+        else if (PageAnon(page))
                 return try_to_unmap_anon(page, TTU_MUNLOCK);
         else
                 return try_to_unmap_file(page, TTU_MUNLOCK);
 }
-
diff --git a/mm/swapfile.c b/mm/swapfile.c
index e74112e8e5f4..6c0585b16418 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -22,6 +22,7 @@
 #include <linux/seq_file.h>
 #include <linux/init.h>
 #include <linux/module.h>
+#include <linux/ksm.h>
 #include <linux/rmap.h>
 #include <linux/security.h>
 #include <linux/backing-dev.h>
@@ -650,6 +651,8 @@ int reuse_swap_page(struct page *page)
         int count;
 
         VM_BUG_ON(!PageLocked(page));
+        if (unlikely(PageKsm(page)))
+                return 0;
         count = page_mapcount(page);
         if (count <= 1 && PageSwapCache(page)) {
                 count += page_swapcount(page);
@@ -658,7 +661,7 @@ int reuse_swap_page(struct page *page)
                         SetPageDirty(page);
                 }
         }
-        return count == 1;
+        return count <= 1;
 }
 
 /*
@@ -1185,6 +1188,12 @@ 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.
+                 *
+                 * Given how unuse_vma() targets one particular offset
+                 * in an anon_vma, once the anon_vma has been determined,
+                 * this splitting happens to be just what is needed to
+                 * handle where KSM pages have been swapped out: re-reading
+                 * is unnecessarily slow, but we can fix that later on.
                  */
                 if (swap_count(*swap_map) &&
                      PageDirty(page) && PageSwapCache(page)) {