memcg: fix shmem's swap accounting
KAMEZAWA Hiroyuki [Thu, 8 Jan 2009 02:08:35 +0000 (18:08 -0800)]
Now, you can see following even when swap accounting is enabled.

 1. Create Group 01, and 02.
 2. allocate a "file" on tmpfs by a task under 01.
 3. swap out the "file" (by memory pressure)
 4. Read "file" from a task in group 02.
 5. the charge of "file" is moved to group 02.

This is not ideal behavior. This is because SwapCache which was loaded
by read-ahead is not taken into account..

This is a patch to fix shmem's swapcache behavior.
  - remove mem_cgroup_cache_charge_swapin().
  - Add SwapCache handler routine to mem_cgroup_cache_charge().
    By this, shmem's file cache is charged at add_to_page_cache()
    with GFP_NOWAIT.
  - pass the page of swapcache to shrink_mem_cgroup.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/memcontrol.h
include/linux/swap.h
mm/memcontrol.c
mm/shmem.c

index 8ae6ece..326f45c 100644 (file)
@@ -56,7 +56,8 @@ extern void mem_cgroup_move_lists(struct page *page,
                                  enum lru_list from, enum lru_list to);
 extern void mem_cgroup_uncharge_page(struct page *page);
 extern void mem_cgroup_uncharge_cache_page(struct page *page);
-extern int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask);
+extern int mem_cgroup_shrink_usage(struct page *page,
+                       struct mm_struct *mm, gfp_t gfp_mask);
 
 extern unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
                                        struct list_head *dst,
@@ -155,7 +156,8 @@ static inline void mem_cgroup_uncharge_cache_page(struct page *page)
 {
 }
 
-static inline int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
+static inline int mem_cgroup_shrink_usage(struct page *page,
+                       struct mm_struct *mm, gfp_t gfp_mask)
 {
        return 0;
 }
index 4ccca25..d302155 100644 (file)
@@ -335,16 +335,8 @@ static inline void disable_swap_token(void)
 }
 
 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
-extern int mem_cgroup_cache_charge_swapin(struct page *page,
-                               struct mm_struct *mm, gfp_t mask, bool locked);
 extern void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent);
 #else
-static inline
-int mem_cgroup_cache_charge_swapin(struct page *page,
-                               struct mm_struct *mm, gfp_t mask, bool locked)
-{
-       return 0;
-}
 static inline void
 mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent)
 {
index f50cb7b..93a7928 100644 (file)
@@ -893,6 +893,23 @@ nomem:
        return -ENOMEM;
 }
 
+static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
+{
+       struct mem_cgroup *mem;
+       swp_entry_t ent;
+
+       if (!PageSwapCache(page))
+               return NULL;
+
+       ent.val = page_private(page);
+       mem = lookup_swap_cgroup(ent);
+       if (!mem)
+               return NULL;
+       if (!css_tryget(&mem->css))
+               return NULL;
+       return mem;
+}
+
 /*
  * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
  * USED state. If already USED, uncharge and return.
@@ -1084,6 +1101,9 @@ int mem_cgroup_newpage_charge(struct page *page,
 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
                                gfp_t gfp_mask)
 {
+       struct mem_cgroup *mem = NULL;
+       int ret;
+
        if (mem_cgroup_disabled())
                return 0;
        if (PageCompound(page))
@@ -1096,6 +1116,8 @@ int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
         * For GFP_NOWAIT case, the page may be pre-charged before calling
         * add_to_page_cache(). (See shmem.c) check it here and avoid to call
         * charge twice. (It works but has to pay a bit larger cost.)
+        * And when the page is SwapCache, it should take swap information
+        * into account. This is under lock_page() now.
         */
        if (!(gfp_mask & __GFP_WAIT)) {
                struct page_cgroup *pc;
@@ -1112,15 +1134,40 @@ int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
                unlock_page_cgroup(pc);
        }
 
-       if (unlikely(!mm))
+       if (do_swap_account && PageSwapCache(page)) {
+               mem = try_get_mem_cgroup_from_swapcache(page);
+               if (mem)
+                       mm = NULL;
+                 else
+                       mem = NULL;
+               /* SwapCache may be still linked to LRU now. */
+               mem_cgroup_lru_del_before_commit_swapcache(page);
+       }
+
+       if (unlikely(!mm && !mem))
                mm = &init_mm;
 
        if (page_is_file_cache(page))
                return mem_cgroup_charge_common(page, mm, gfp_mask,
                                MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
-       else
-               return mem_cgroup_charge_common(page, mm, gfp_mask,
-                               MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
+
+       ret = mem_cgroup_charge_common(page, mm, gfp_mask,
+                               MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
+       if (mem)
+               css_put(&mem->css);
+       if (PageSwapCache(page))
+               mem_cgroup_lru_add_after_commit_swapcache(page);
+
+       if (do_swap_account && !ret && PageSwapCache(page)) {
+               swp_entry_t ent = {.val = page_private(page)};
+               /* avoid double counting */
+               mem = swap_cgroup_record(ent, NULL);
+               if (mem) {
+                       res_counter_uncharge(&mem->memsw, PAGE_SIZE);
+                       mem_cgroup_put(mem);
+               }
+       }
+       return ret;
 }
 
 /*
@@ -1134,7 +1181,6 @@ int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
                                 gfp_t mask, struct mem_cgroup **ptr)
 {
        struct mem_cgroup *mem;
-       swp_entry_t     ent;
        int ret;
 
        if (mem_cgroup_disabled())
@@ -1142,7 +1188,6 @@ int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
 
        if (!do_swap_account)
                goto charge_cur_mm;
-
        /*
         * A racing thread's fault, or swapoff, may have already updated
         * the pte, and even removed page from swap cache: return success
@@ -1150,14 +1195,9 @@ int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
         */
        if (!PageSwapCache(page))
                return 0;
-
-       ent.val = page_private(page);
-
-       mem = lookup_swap_cgroup(ent);
+       mem = try_get_mem_cgroup_from_swapcache(page);
        if (!mem)
                goto charge_cur_mm;
-       if (!css_tryget(&mem->css))
-               goto charge_cur_mm;
        *ptr = mem;
        ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
        /* drop extra refcnt from tryget */
@@ -1169,62 +1209,6 @@ charge_cur_mm:
        return __mem_cgroup_try_charge(mm, mask, ptr, true);
 }
 
-#ifdef CONFIG_SWAP
-
-int mem_cgroup_cache_charge_swapin(struct page *page,
-                       struct mm_struct *mm, gfp_t mask, bool locked)
-{
-       int ret = 0;
-
-       if (mem_cgroup_disabled())
-               return 0;
-       if (unlikely(!mm))
-               mm = &init_mm;
-       if (!locked)
-               lock_page(page);
-       /*
-        * If not locked, the page can be dropped from SwapCache until
-        * we reach here.
-        */
-       if (PageSwapCache(page)) {
-               struct mem_cgroup *mem = NULL;
-               swp_entry_t ent;
-
-               ent.val = page_private(page);
-               if (do_swap_account) {
-                       mem = lookup_swap_cgroup(ent);
-                       if (mem) {
-                               if (css_tryget(&mem->css))
-                                       mm = NULL; /* charge to recorded */
-                               else
-                                       mem = NULL; /* charge to current */
-                       }
-               }
-               /* SwapCache may be still linked to LRU now. */
-               mem_cgroup_lru_del_before_commit_swapcache(page);
-               ret = mem_cgroup_charge_common(page, mm, mask,
-                               MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
-               mem_cgroup_lru_add_after_commit_swapcache(page);
-               /* drop extra refcnt from tryget */
-               if (mem)
-                       css_put(&mem->css);
-
-               if (!ret && do_swap_account) {
-                       /* avoid double counting */
-                       mem = swap_cgroup_record(ent, NULL);
-                       if (mem) {
-                               res_counter_uncharge(&mem->memsw, PAGE_SIZE);
-                               mem_cgroup_put(mem);
-                       }
-               }
-       }
-       if (!locked)
-               unlock_page(page);
-
-       return ret;
-}
-#endif
-
 void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
 {
        struct page_cgroup *pc;
@@ -1486,18 +1470,20 @@ void mem_cgroup_end_migration(struct mem_cgroup *mem,
  * This is typically used for page reclaiming for shmem for reducing side
  * effect of page allocation from shmem, which is used by some mem_cgroup.
  */
-int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
+int mem_cgroup_shrink_usage(struct page *page,
+                           struct mm_struct *mm,
+                           gfp_t gfp_mask)
 {
-       struct mem_cgroup *mem;
+       struct mem_cgroup *mem = NULL;
        int progress = 0;
        int retry = MEM_CGROUP_RECLAIM_RETRIES;
 
        if (mem_cgroup_disabled())
                return 0;
-       if (!mm)
-               return 0;
-
-       mem = try_get_mem_cgroup_from_mm(mm);
+       if (page)
+               mem = try_get_mem_cgroup_from_swapcache(page);
+       if (!mem && mm)
+               mem = try_get_mem_cgroup_from_mm(mm);
        if (unlikely(!mem))
                return 0;
 
index bbb7b04..5d0de96 100644 (file)
@@ -929,11 +929,11 @@ found:
        if (!inode)
                goto out;
        /*
-        * Charge page using GFP_HIGHUSER_MOVABLE while we can wait.
-        * charged back to the user(not to caller) when swap account is used.
+        * Charge page using GFP_KERNEL while we can wait.
+        * Charged back to the user(not to caller) when swap account is used.
+        * add_to_page_cache() will be called with GFP_NOWAIT.
         */
-       error = mem_cgroup_cache_charge_swapin(page, current->mm, GFP_KERNEL,
-                                       true);
+       error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
        if (error)
                goto out;
        error = radix_tree_preload(GFP_KERNEL);
@@ -1270,16 +1270,6 @@ repeat:
                                goto repeat;
                        }
                        wait_on_page_locked(swappage);
-                       /*
-                        * We want to avoid charge at add_to_page_cache().
-                        * charge against this swap cache here.
-                        */
-                       if (mem_cgroup_cache_charge_swapin(swappage,
-                               current->mm, gfp & GFP_RECLAIM_MASK, false)) {
-                               page_cache_release(swappage);
-                               error = -ENOMEM;
-                               goto failed;
-                       }
                        page_cache_release(swappage);
                        goto repeat;
                }
@@ -1334,15 +1324,19 @@ repeat:
                } else {
                        shmem_swp_unmap(entry);
                        spin_unlock(&info->lock);
-                       unlock_page(swappage);
-                       page_cache_release(swappage);
                        if (error == -ENOMEM) {
                                /* allow reclaim from this memory cgroup */
-                               error = mem_cgroup_shrink_usage(current->mm,
+                               error = mem_cgroup_shrink_usage(swappage,
+                                                               current->mm,
                                                                gfp);
-                               if (error)
+                               if (error) {
+                                       unlock_page(swappage);
+                                       page_cache_release(swappage);
                                        goto failed;
+                               }
                        }
+                       unlock_page(swappage);
+                       page_cache_release(swappage);
                        goto repeat;
                }
        } else if (sgp == SGP_READ && !filepage) {