--
cgit 1.2-0.3.lf.el7
+From 48744b6339cf649a69b55997e138c17df1ecc897 Mon Sep 17 00:00:00 2001\r
+From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>\r
+Date: Sat, 26 Jan 2019 20:00:51 +0900\r
+Subject: [PATCH] oom, oom_reaper: do not enqueue same task twice\r
+\r
+Arkadiusz reported that enabling memcg's group oom killing causes\r
+strange memcg statistics where there is no task in a memcg despite\r
+the number of tasks in that memcg is not 0. It turned out that there\r
+is a bug in wake_oom_reaper() which allows enqueuing same task twice\r
+which makes impossible to decrease the number of tasks in that memcg\r
+due to a refcount leak.\r
+\r
+This bug existed since the OOM reaper became invokable from\r
+task_will_free_mem(current) path in out_of_memory() in Linux 4.7,\r
+but memcg's group oom killing made it easier to trigger this bug by\r
+calling wake_oom_reaper() on the same task from one out_of_memory()\r
+request.\r
+\r
+Fix this bug using an approach used by commit 855b018325737f76\r
+("oom, oom_reaper: disable oom_reaper for oom_kill_allocating_task").\r
+Since task_will_free_mem(p) == false if p->mm == NULL, we can assume that\r
+p->mm != NULL when wake_oom_reaper() is called from task_will_free_mem()\r
+paths. As a side effect of this patch, this patch also avoids enqueuing\r
+multiple threads sharing memory via task_will_free_mem(current) path.\r
+\r
+Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>\r
+Reported-by: Arkadiusz Miśkiewicz <a.miskiewicz@gmail.com>\r
+Fixes: af8e15cc85a25315 ("oom, oom_reaper: do not enqueue task if it is on the oom_reaper_list head")\r
+---\r
+ mm/oom_kill.c | 28 +++++++++++++---------------\r
+ 1 file changed, 13 insertions(+), 15 deletions(-)\r
+\r
+diff --git a/mm/oom_kill.c b/mm/oom_kill.c\r
+index f0e8cd9..457f240 100644\r
+--- a/mm/oom_kill.c\r
++++ b/mm/oom_kill.c\r
+@@ -505,14 +505,6 @@ bool __oom_reap_task_mm(struct mm_struct *mm)\r
+ struct vm_area_struct *vma;\r
+ bool ret = true;\r
+ \r
+- /*\r
+- * Tell all users of get_user/copy_from_user etc... that the content\r
+- * is no longer stable. No barriers really needed because unmapping\r
+- * should imply barriers already and the reader would hit a page fault\r
+- * if it stumbled over a reaped memory.\r
+- */\r
+- set_bit(MMF_UNSTABLE, &mm->flags);\r
+-\r
+ for (vma = mm->mmap ; vma; vma = vma->vm_next) {\r
+ if (!can_madv_dontneed_vma(vma))\r
+ continue;\r
+@@ -645,10 +637,15 @@ static int oom_reaper(void *unused)\r
+ return 0;\r
+ }\r
+ \r
+-static void wake_oom_reaper(struct task_struct *tsk)\r
++static void wake_oom_reaper(struct task_struct *tsk, struct mm_struct *mm)\r
+ {\r
+- /* tsk is already queued? */\r
+- if (tsk == oom_reaper_list || tsk->oom_reaper_list)\r
++ /*\r
++ * Tell all users of get_user/copy_from_user etc... that the content\r
++ * is no longer stable. No barriers really needed because unmapping\r
++ * should imply barriers already and the reader would hit a page fault\r
++ * if it stumbled over a reaped memory.\r
++ */\r
++ if (test_and_set_bit(MMF_UNSTABLE, &mm->flags))\r
+ return;\r
+ \r
+ get_task_struct(tsk);\r
+@@ -668,7 +665,8 @@ static int __init oom_init(void)\r
+ }\r
+ subsys_initcall(oom_init)\r
+ #else\r
+-static inline void wake_oom_reaper(struct task_struct *tsk)\r
++static inline void wake_oom_reaper(struct task_struct *tsk,\r
++ struct mm_struct *mm)\r
+ {\r
+ }\r
+ #endif /* CONFIG_MMU */\r
+@@ -915,7 +913,7 @@ static void __oom_kill_process(struct task_struct *victim)\r
+ rcu_read_unlock();\r
+ \r
+ if (can_oom_reap)\r
+- wake_oom_reaper(victim);\r
++ wake_oom_reaper(victim, mm);\r
+ \r
+ mmdrop(mm);\r
+ put_task_struct(victim);\r
+@@ -955,7 +953,7 @@ static void oom_kill_process(struct oom_control *oc, const char *message)\r
+ task_lock(p);\r
+ if (task_will_free_mem(p)) {\r
+ mark_oom_victim(p);\r
+- wake_oom_reaper(p);\r
++ wake_oom_reaper(p, p->mm);\r
+ task_unlock(p);\r
+ put_task_struct(p);\r
+ return;\r
+@@ -1085,7 +1083,7 @@ bool out_of_memory(struct oom_control *oc)\r
+ */\r
+ if (task_will_free_mem(current)) {\r
+ mark_oom_victim(current);\r
+- wake_oom_reaper(current);\r
++ wake_oom_reaper(current, current->mm);\r
+ return true;\r
+ }\r
+ \r
+-- \r
+1.8.3.1\r
+\r