Skip to contentSkip to conversation

mm, memcg: cgroup v1 and v2 tunable load/store tearing fixes

PATCH Superseded Yue Zhao 4 participants 12 messages Mar 6, 2023 – Mar 8, 2023

Subsystems (4)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Yue Zhao

author
#0 lore raw

v2 cover mm, memcg: cgroup v1 and v2 tunable load/store tearing fixes

This patch series helps to prevent load/store tearing in several cgroup knobs.

As kindly pointed out by Michal Hocko and Roman Gushchin , the changelog has been rephrased.

Besides, more knobs were checked, according to kind suggestions from Shakeel Butt and Muchun Song.

v1:
- Add [WRITE|READ]_ONCE for memcg->oom_group v2:
- Rephrase changelog
- Add [WRITE|READ]_ONCE for memcg->oom_kill_disable, memcg->swappiness, vm_swappiness and memcg->soft_limit

Yue Zhao (4):
mm, memcg: Prevent memory.oom.group load/store tearing mm, memcg: Prevent memory.swappiness load/store tearing mm, memcg: Prevent memory.oom_control load/store tearing mm, memcg: Prevent memory.soft_limit_in_bytes load/store tearing

 include/linux/swap.h |  8 ++++----
 mm/memcontrol.c      | 18 +++++++++---------
 2 files changed, 13 insertions(+), 13 deletions(-)
Signature · 1 line
2.17.1
author
#1 lore raw

PATCH v2 1/4 mm, memcg: Prevent memory.oom.group load/store tearing

The knob for cgroup v2 memory controller: memory.oom.group is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely (the knob is usually configured long before any workloads hits actual memcg oom) but it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->oom_group is lockless, so it can be concurrently set at the same time as we are trying to read it.

---
 mm/memcontrol.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
mm/memcontrol.c +3 -3
index 5abffe6f8389..06821e5f7604 100644
@@ -2067,7 +2067,7 @@ struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
 	 * highest-level memory cgroup with oom.group set.
 	 */
 	for (; memcg; memcg = parent_mem_cgroup(memcg)) {
-		if (memcg->oom_group)
+		if (READ_ONCE(memcg->oom_group))
 			oom_group = memcg;
 
 		if (memcg == oom_domain)
@@ -6623,7 +6623,7 @@ static int memory_oom_group_show(struct seq_file *m, void *v)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
 
-	seq_printf(m, "%d\n", memcg->oom_group);
+	seq_printf(m, "%d\n", READ_ONCE(memcg->oom_group));
 
 	return 0;
 }
@@ -6645,7 +6645,7 @@ static ssize_t memory_oom_group_write(struct kernfs_open_file *of,
 	if (oom_group != 0 && oom_group != 1)
 		return -EINVAL;
 
-	memcg->oom_group = oom_group;
+	WRITE_ONCE(memcg->oom_group, oom_group);
 
 	return nbytes;
 }
Signature · 1 line
2.17.1
Acked-by on v2 1/4
#2 lore raw

Re: [PATCH v2, 1/4] mm, memcg: Prevent memory.oom.group load/store tearing

Quoted 11 lines · Yue Zhao

The knob for cgroup v2 memory controller: memory.oom.group is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely (the knob is usually configured long before any workloads hits actual memcg oom) but it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->oom_group is lockless, so it can be concurrently set at the same time as we are trying to read it.

Thanks!

Quoted 37 lines
---
 mm/memcontrol.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
mm/memcontrol.c +3 -3
index 5abffe6f8389..06821e5f7604 100644
@@ -2067,7 +2067,7 @@ struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
 	 * highest-level memory cgroup with oom.group set.
 	 */
 	for (; memcg; memcg = parent_mem_cgroup(memcg)) {
-		if (memcg->oom_group)
+		if (READ_ONCE(memcg->oom_group))
 			oom_group = memcg;
 
 		if (memcg == oom_domain)
@@ -6623,7 +6623,7 @@ static int memory_oom_group_show(struct seq_file *m, void *v)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
 
-	seq_printf(m, "%d\n", memcg->oom_group);
+	seq_printf(m, "%d\n", READ_ONCE(memcg->oom_group));
 
 	return 0;
 }
@@ -6645,7 +6645,7 @@ static ssize_t memory_oom_group_write(struct kernfs_open_file *of,
 	if (oom_group != 0 && oom_group != 1)
 		return -EINVAL;
 
-	memcg->oom_group = oom_group;
+	WRITE_ONCE(memcg->oom_group, oom_group);
 
 	return nbytes;
 }
Signature · 1 line
2.17.1
Signature · 2 lines
Michal Hocko
SUSE Labs
author
#3 lore raw

PATCH v2 2/4 mm, memcg: Prevent memory.swappiness load/store tearing

In reply to Yue Zhao

The knob for cgroup v1 memory controller: memory.swappiness is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely. But it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->swappiness and vm_swappiness is lockless, so both of them can be concurrently set at the same time as we are trying to read them.

---
 include/linux/swap.h | 8 ++++----
 mm/memcontrol.c      | 4 ++--
 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/swap.h +4 -4
index 209a425739a9..3f3fe43d1766 100644
@@ -620,18 +620,18 @@ static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
 {
 	/* Cgroup2 doesn't have per-cgroup swappiness */
 	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
-		return vm_swappiness;
+		return READ_ONCE(vm_swappiness);
 
 	/* root ? */
 	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
-		return vm_swappiness;
+		return READ_ONCE(vm_swappiness);
 
-	return memcg->swappiness;
+	return READ_ONCE(memcg->swappiness);
 }
 #else
 static inline int mem_cgroup_swappiness(struct mem_cgroup *mem)
 {
-	return vm_swappiness;
+	return READ_ONCE(vm_swappiness);
 }
 #endif
 
mm/memcontrol.c +2 -2
index 06821e5f7604..dca895c66a9b 100644
@@ -4179,9 +4179,9 @@ static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
 		return -EINVAL;
 
 	if (!mem_cgroup_is_root(memcg))
-		memcg->swappiness = val;
+		WRITE_ONCE(memcg->swappiness, val);
 	else
-		vm_swappiness = val;
+		WRITE_ONCE(vm_swappiness, val);
 
 	return 0;
 }
Signature · 1 line
2.17.1
Acked-by on v2 2/4
#4 lore raw

Re: [PATCH v2, 2/4] mm, memcg: Prevent memory.swappiness load/store tearing

Quoted 11 lines · Yue Zhao

The knob for cgroup v1 memory controller: memory.swappiness is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely. But it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->swappiness and vm_swappiness is lockless, so both of them can be concurrently set at the same time as we are trying to read them.

Thanks!

Quoted 50 lines
---
 include/linux/swap.h | 8 ++++----
 mm/memcontrol.c      | 4 ++--
 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/swap.h +4 -4
index 209a425739a9..3f3fe43d1766 100644
@@ -620,18 +620,18 @@ static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
 {
 	/* Cgroup2 doesn't have per-cgroup swappiness */
 	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
-		return vm_swappiness;
+		return READ_ONCE(vm_swappiness);
 
 	/* root ? */
 	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
-		return vm_swappiness;
+		return READ_ONCE(vm_swappiness);
 
-	return memcg->swappiness;
+	return READ_ONCE(memcg->swappiness);
 }
 #else
 static inline int mem_cgroup_swappiness(struct mem_cgroup *mem)
 {
-	return vm_swappiness;
+	return READ_ONCE(vm_swappiness);
 }
 #endif
 
mm/memcontrol.c +2 -2
index 06821e5f7604..dca895c66a9b 100644
@@ -4179,9 +4179,9 @@ static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
 		return -EINVAL;
 
 	if (!mem_cgroup_is_root(memcg))
-		memcg->swappiness = val;
+		WRITE_ONCE(memcg->swappiness, val);
 	else
-		vm_swappiness = val;
+		WRITE_ONCE(vm_swappiness, val);
 
 	return 0;
 }
Signature · 1 line
2.17.1
Signature · 2 lines
Michal Hocko
SUSE Labs
author
#5 lore raw

PATCH v2 3/4 mm, memcg: Prevent memory.oom_control load/store tearing

In reply to Yue Zhao

The knob for cgroup v1 memory controller: memory.oom_control is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely. But it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->oom_kill_disable is lockless, so it can be concurrently set at the same time as we are trying to read it.

---
 mm/memcontrol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
mm/memcontrol.c +2 -2
index dca895c66a9b..26605b2f51b1 100644
@@ -4515,7 +4515,7 @@ static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_seq(sf);
 
-	seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
+	seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable));
 	seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
 	seq_printf(sf, "oom_kill %lu\n",
 		   atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL]));
@@ -4531,7 +4531,7 @@ static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
 	if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1)))
 		return -EINVAL;
 
-	memcg->oom_kill_disable = val;
+	WRITE_ONCE(memcg->oom_kill_disable, val);
 	if (!val)
 		memcg_oom_recover(memcg);
 
Signature · 1 line
2.17.1
#6 lore raw

Re: [PATCH v2, 3/4] mm, memcg: Prevent memory.oom_control load/store tearing

Quoted 29 lines · Yue Zhao

The knob for cgroup v1 memory controller: memory.oom_control is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely. But it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->oom_kill_disable is lockless, so it can be concurrently set at the same time as we are trying to read it.

---
 mm/memcontrol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
mm/memcontrol.c +1 -1
index dca895c66a9b..26605b2f51b1 100644
@@ -4515,7 +4515,7 @@ static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_seq(sf);
 
-	seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
+	seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable));
 	seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
 	seq_printf(sf, "oom_kill %lu\n",
 		   atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL]));

mm/memcontrol.c static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, · line 4536

Quoted 7 lines
mm/memcontrol.c +1 -1
@@ -4531,7 +4531,7 @@ static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
 	if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1)))
 		return -EINVAL;
 
-	memcg->oom_kill_disable = val;
+	WRITE_ONCE(memcg->oom_kill_disable, val);
 	if (!val)
 		memcg_oom_recover(memcg);

Any specific reasons you haven't covered other accesses (mem_cgroup_css_alloc, mem_cgroup_oom, mem_cgroup_oom_synchronize)?

Quoted 3 lines
Signature · 2 lines
-- 
2.17.1
Signature · 2 lines
Michal Hocko
SUSE Labs
#7 lore raw

Re: [PATCH v2, 3/4] mm, memcg: Prevent memory.oom_control load/store tearing

Quoted 34 lines · On Tue, Mar 7, 2023 at 1:53 AM Michal Hocko…
Quoted 32 lines · Yue Zhao

The knob for cgroup v1 memory controller: memory.oom_control is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely. But it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->oom_kill_disable is lockless, so it can be concurrently set at the same time as we are trying to read it.

---
 mm/memcontrol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
mm/memcontrol.c +1 -1
index dca895c66a9b..26605b2f51b1 100644
@@ -4515,7 +4515,7 @@ static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
 {
      struct mem_cgroup *memcg = mem_cgroup_from_seq(sf);
 
-     seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
+     seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable));
      seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
      seq_printf(sf, "oom_kill %lu\n",
                 atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL]));
@@ -4531,7 +4531,7 @@ static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
      if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1)))
              return -EINVAL;
 
Quoted 7 lines

mm/memcontrol.c static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, · line 4536

Quoted 4 lines
mm/memcontrol.c +1 -1
-     memcg->oom_kill_disable = val;
+     WRITE_ONCE(memcg->oom_kill_disable, val);
      if (!val)
              memcg_oom_recover(memcg);

Any specific reasons you haven't covered other accesses (mem_cgroup_css_alloc, mem_cgroup_oom, mem_cgroup_oom_synchronize)?

Thanks for point this out, you are right, we should add [READ|WRITE]_ONCE for all used places.
Let me create PATCH v3 later.
Also for the memcg->soft_limit, I will update as well.

Quoted 7 lines
Quoted 3 lines
mm/memcontrol.c +0 -1
--

2.17.1

--
Michal Hocko
SUSE Labs

author
#8 lore raw

PATCH v2 4/4 mm, memcg: Prevent memory.soft_limit_in_bytes load/store tearing

In reply to Yue Zhao

The knob for cgroup v1 memory controller: memory.soft_limit_in_bytes is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely. But it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->soft_limit is lockless,
so it can be concurrently set at the same time as we are trying to read it.

---
 mm/memcontrol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
mm/memcontrol.c +2 -2
index 26605b2f51b1..20566f59bbcb 100644
@@ -3728,7 +3728,7 @@ static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
 	case RES_FAILCNT:
 		return counter->failcnt;
 	case RES_SOFT_LIMIT:
-		return (u64)memcg->soft_limit * PAGE_SIZE;
+		return (u64)READ_ONCE(memcg->soft_limit) * PAGE_SIZE;
 	default:
 		BUG();
 	}
@@ -3870,7 +3870,7 @@ static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
 		if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
 			ret = -EOPNOTSUPP;
 		} else {
-			memcg->soft_limit = nr_pages;
+			WRITE_ONCE(memcg->soft_limit, nr_pages);
 			ret = 0;
 		}
 		break;
Signature · 1 line
2.17.1
#9 lore raw

Re: [PATCH v2, 4/4] mm, memcg: Prevent memory.soft_limit_in_bytes load/store tearing

Quoted 9 lines · Yue Zhao

The knob for cgroup v1 memory controller: memory.soft_limit_in_bytes is not protected by any locking so it can be modified while it is used. This is not an actual problem because races are unlikely. But it is better to use READ_ONCE/WRITE_ONCE to prevent compiler from doing anything funky.

The access of memcg->soft_limit is lockless,
so it can be concurrently set at the same time as we are trying to read it.

Similar here. mem_cgroup_css_reset and mem_cgroup_css_alloc are not covered.

Quoted 30 lines
---
 mm/memcontrol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
mm/memcontrol.c +2 -2
index 26605b2f51b1..20566f59bbcb 100644
@@ -3728,7 +3728,7 @@ static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
 	case RES_FAILCNT:
 		return counter->failcnt;
 	case RES_SOFT_LIMIT:
-		return (u64)memcg->soft_limit * PAGE_SIZE;
+		return (u64)READ_ONCE(memcg->soft_limit) * PAGE_SIZE;
 	default:
 		BUG();
 	}
@@ -3870,7 +3870,7 @@ static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
 		if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
 			ret = -EOPNOTSUPP;
 		} else {
-			memcg->soft_limit = nr_pages;
+			WRITE_ONCE(memcg->soft_limit, nr_pages);
 			ret = 0;
 		}
 		break;
Signature · 1 line
2.17.1
Signature · 2 lines
Michal Hocko
SUSE Labs
Acked-by on the v2 series
#10 lore raw
In reply to Yue Zhao
Quoted 10 lines · On Mon, Mar 6, 2023 at 7:42 AM Yue Zhao…

This patch series helps to prevent load/store tearing in several cgroup knobs.

As kindly pointed out by Michal Hocko and Roman Gushchin , the changelog has been rephrased.

Besides, more knobs were checked, according to kind suggestions from Shakeel Butt and Muchun Song.

Quoted 7 lines

v1:
- Add [WRITE|READ]_ONCE for memcg->oom_group v2:
- Rephrase changelog
- Add [WRITE|READ]_ONCE for memcg->oom_kill_disable, memcg->swappiness, vm_swappiness and memcg->soft_limit

Thanks Yue and for the whole series:

maintainer Acked-by on the v2 series
#11 lore raw
In reply to Yue Zhao
Quoted 16 lines · Yue Zhao

This patch series helps to prevent load/store tearing in several cgroup knobs.

As kindly pointed out by Michal Hocko and Roman Gushchin , the changelog has been rephrased.

Besides, more knobs were checked, according to kind suggestions from Shakeel Butt and Muchun Song.

v1:
- Add [WRITE|READ]_ONCE for memcg->oom_group v2:
- Rephrase changelog
- Add [WRITE|READ]_ONCE for memcg->oom_kill_disable, memcg->swappiness, vm_swappiness and memcg->soft_limit

Quoted 5 lines

Yue Zhao (4):
mm, memcg: Prevent memory.oom.group load/store tearing mm, memcg: Prevent memory.swappiness load/store tearing mm, memcg: Prevent memory.oom_control load/store tearing mm, memcg: Prevent memory.soft_limit_in_bytes load/store tearing

for the series.

Thank you!