Re: [PATCH] mm: memcontrol: fix potential oom_lock recursion deadlock

[Michal Hocko <mhocko@suse.com>]

On Thu 21-07-22 08:49:57, Tetsuo Handa wrote:
> syzbot is reporting fs_reclaim allocation with oom_lock held [1]. We
> must make sure that such allocation won't hit __alloc_pages_may_oom()
> path which will retry forever if oom_lock is already held.
> 
> I choose GFP_ATOMIC than GFP_NOWAIT, for since global OOM situation will
> likely be avoided by killing some process in memcg, and memory will be
> released after printk(), trying a little hard will be acceptable.

Nope, this is not a proper fix. You are making memory.stat more likely
to fail.

An uncoditional GFP_KERNEL allocation is certainly not good but is there
any reason to not use GFP_NOIO instead?

Or even better. In an ideal world we won't allocate from here at
all. Can we pre-allocate that single page and re-use it for the oom
report?
--- 
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index abec50f31fe6..13483cb278bb 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -1460,14 +1460,12 @@ static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg,
 	return memcg_page_state(memcg, item) * memcg_page_state_unit(item);
 }
 
-static char *memory_stat_format(struct mem_cgroup *memcg)
+void memory_stat_format(struct mem_cgroup *memcg, char *buf)
 {
 	struct seq_buf s;
 	int i;
 
-	seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE);
-	if (!s.buffer)
-		return NULL;
+	seq_buf_init(&s, buf, PAGE_SIZE);
 
 	/*
 	 * Provide statistics on the state of the memory subsystem as
@@ -1533,8 +1531,6 @@ static char *memory_stat_format(struct mem_cgroup *memcg)
 
 	/* The above should easily fit into one page */
 	WARN_ON_ONCE(seq_buf_has_overflowed(&s));
-
-	return s.buffer;
 }
 
 #define K(x) ((x) << (PAGE_SHIFT-10))
@@ -1563,6 +1559,12 @@ void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *
 	rcu_read_unlock();
 }
 
+/*
+ * preallocated buffer to collect memory stats for the oom situation.
+ * Usage protected by oom_lock
+ */
+char oombuf[PAGE_SIZE];
+
 /**
  * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to
  * memory controller.
@@ -1570,7 +1572,7 @@ void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *
  */
 void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
 {
-	char *buf;
+	lockdep_assert_held(&oom_lock);
 
 	pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
 		K((u64)page_counter_read(&memcg->memory)),
@@ -1591,11 +1593,8 @@ void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
 	pr_info("Memory cgroup stats for ");
 	pr_cont_cgroup_path(memcg->css.cgroup);
 	pr_cont(":");
-	buf = memory_stat_format(memcg);
-	if (!buf)
-		return;
-	pr_info("%s", buf);
-	kfree(buf);
+	memory_stat_format(memcg, oombuf);
+	pr_info("%s", oombuf);
 }
 
 /*
@@ -6335,11 +6334,11 @@ static int memory_events_local_show(struct seq_file *m, void *v)
 static int memory_stat_show(struct seq_file *m, void *v)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
-	char *buf;
+	char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 
-	buf = memory_stat_format(memcg);
 	if (!buf)
 		return -ENOMEM;
+	memory_stat_format(memcg, buf);
 	seq_puts(m, buf);
 	kfree(buf);
 	return 0;

-- 
Michal Hocko
SUSE Labs