[PATCH] cgroup: fix memory leak caused by missing cgroup_bpf_offline

From: Quanyang Wang <redacted>

When enabling CONFIG_CGROUP_BPF, kmemleak can be observed by running
the command as below:

    $mount -t cgroup -o none,name=foo cgroup cgroup/
    $umount cgroup/

unreferenced object 0xc3585c40 (size 64):
  comm "mount", pid 425, jiffies 4294959825 (age 31.990s)
  hex dump (first 32 bytes):
    01 00 00 80 84 8c 28 c0 00 00 00 00 00 00 00 00  ......(.........
    00 00 00 00 00 00 00 00 6c 43 a0 c3 00 00 00 00  ........lC......
  backtrace:
    [<e95a2f9e>] cgroup_bpf_inherit+0x44/0x24c
    [<1f03679c>] cgroup_setup_root+0x174/0x37c
    [<ed4b0ac5>] cgroup1_get_tree+0x2c0/0x4a0
    [<f85b12fd>] vfs_get_tree+0x24/0x108
    [<f55aec5c>] path_mount+0x384/0x988
    [<e2d5e9cd>] do_mount+0x64/0x9c
    [<208c9cfe>] sys_mount+0xfc/0x1f4
    [<06dd06e0>] ret_fast_syscall+0x0/0x48
    [<a8308cb3>] 0xbeb4daa8

This is because that root_cgrp->bpf.refcnt.data is allocated by the
function percpu_ref_init in cgroup_bpf_inherit which is called by
cgroup_setup_root when mounting, but not freed along with root_cgrp
when umounting. Adding cgroup_bpf_offline which calls percpu_ref_kill
to cgroup_kill_sb can free root_cgrp->bpf.refcnt.data in umount path.

Fixes: 2b0d3d3e4fcfb ("percpu_ref: reduce memory footprint of percpu_ref in fast path")
Signed-off-by: Quanyang Wang <redacted>
---
 kernel/cgroup/cgroup.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index c8b811e039cc2..ce636deec5e41 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c

@@ -2147,8 +2147,10 @@ static void cgroup_kill_sb(struct super_block *sb)
 	 * And don't kill the default root.
 	 */
 	if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
-	    !percpu_ref_is_dying(&root->cgrp.self.refcnt))
+			!percpu_ref_is_dying(&root->cgrp.self.refcnt)) {
+		cgroup_bpf_offline(&root->cgrp);
 		percpu_ref_kill(&root->cgrp.self.refcnt);
+	}
 	cgroup_put(&root->cgrp);
 	kernfs_kill_sb(sb);
 }

-- 
2.25.1

Hello.

On Thu, Oct 07, 2021 at 08:16:03PM +0800, quanyang.wang@windriver.com wrote:

This is because that root_cgrp->bpf.refcnt.data is allocated by the
function percpu_ref_init in cgroup_bpf_inherit which is called by
cgroup_setup_root when mounting, but not freed along with root_cgrp
when umounting.

Good catch!

Adding cgroup_bpf_offline which calls percpu_ref_kill to
cgroup_kill_sb can free root_cgrp->bpf.refcnt.data in umount path.

That is sensible.

Fixes: 2b0d3d3e4fcfb ("percpu_ref: reduce memory footprint of percpu_ref in fast path")

Why this Fixes:? Is the leak absent before the percpu_ref refactoring?
I guess the embedded data are free'd together with cgroup. Makes me
wonder why struct cgroup_bpf has a separate percpu_ref counter from
struct cgroup...

(You made some unnecessary whitespace here breaking indention :-)

 		percpu_ref_kill(&root->cgrp.self.refcnt);
+	}
 	cgroup_put(&root->cgrp);
 	kernfs_kill_sb(sb);
 }

On Mon, Oct 11, 2021 at 06:21:28PM +0200, Michal Koutny wrote:

Hello.

On Thu, Oct 07, 2021 at 08:16:03PM +0800, quanyang.wang@windriver.com wrote:

quoted

This is because that root_cgrp->bpf.refcnt.data is allocated by the
function percpu_ref_init in cgroup_bpf_inherit which is called by
cgroup_setup_root when mounting, but not freed along with root_cgrp
when umounting.

Good catch!

+1

quoted

Adding cgroup_bpf_offline which calls percpu_ref_kill to
cgroup_kill_sb can free root_cgrp->bpf.refcnt.data in umount path.

That is sensible.

quoted

Fixes: 2b0d3d3e4fcfb ("percpu_ref: reduce memory footprint of percpu_ref in fast path")

Why this Fixes:? Is the leak absent before the percpu_ref refactoring?

I agree, the "fixes" tag looks dubious to me.

I guess the embedded data are free'd together with cgroup. Makes me
wonder why struct cgroup_bpf has a separate percpu_ref counter from
struct cgroup...

This is because a cgroup can stay a long time (sometimes effectively forever)
in a dying state, so we want to release bpf structures earlier.

Thanks!

Hi Michal & Roman,

Thank you for your review.

On 10/12/21 12:21 AM, Michal Koutný wrote:

Hello.

On Thu, Oct 07, 2021 at 08:16:03PM +0800, quanyang.wang@windriver.com wrote:

quoted

This is because that root_cgrp->bpf.refcnt.data is allocated by the
function percpu_ref_init in cgroup_bpf_inherit which is called by
cgroup_setup_root when mounting, but not freed along with root_cgrp
when umounting.

Good catch!

quoted

Adding cgroup_bpf_offline which calls percpu_ref_kill to
cgroup_kill_sb can free root_cgrp->bpf.refcnt.data in umount path.

That is sensible.

quoted

Fixes: 2b0d3d3e4fcfb ("percpu_ref: reduce memory footprint of percpu_ref in fast path")

Why this Fixes:? Is the leak absent before the percpu_ref refactoring?

Before this commit, percpu_ref is embedded in cgroup, it can be freed 
along with cgroup, so there is no memory leak. Since this commit, it 
causes the memory leak.
Should I change it to "Fixes: 4bfc0bb2c60e ("bpf: decouple the lifetime 
of cgroup_bpf from cgroup itself")"?

I guess the embedded data are free'd together with cgroup. Makes me
wonder why struct cgroup_bpf has a separate percpu_ref counter from
struct cgroup...

quoted

+++ b/kernel/cgroup/cgroup.c

@@ -2147,8 +2147,10 @@ static void cgroup_kill_sb(struct super_block *sb)
  	 * And don't kill the default root.
  	 */
  	if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
-	    !percpu_ref_is_dying(&root->cgrp.self.refcnt))
+			!percpu_ref_is_dying(&root->cgrp.self.refcnt)) {
+		cgroup_bpf_offline(&root->cgrp);

(You made some unnecessary whitespace here breaking indention :-)

Thanks for pointing it out. I will send a V2 to fix this.

Thanks,
Quanyang

quoted

  		percpu_ref_kill(&root->cgrp.self.refcnt);
+	}
  	cgroup_put(&root->cgrp);
  	kernfs_kill_sb(sb);
  }

On Tue, Oct 12, 2021 at 02:22:13PM +0800, Quanyang Wang [off-list ref] wrote:

Before this commit, percpu_ref is embedded in cgroup, it can be freed along
with cgroup, so there is no memory leak. Since this commit, it causes the
memory leak.
Should I change it to "Fixes: 4bfc0bb2c60e ("bpf: decouple the lifetime of
cgroup_bpf from cgroup itself")"?

I see. The leak is a product so I'd tag both of them and explain it in
the commit message.

Thank you,
Michal

Roman Gushchin wrote:

On Mon, Oct 11, 2021 at 06:21:28PM +0200, Michal Koutny wrote:

quoted

Hello.

On Thu, Oct 07, 2021 at 08:16:03PM +0800, quanyang.wang@windriver.com wrote:

quoted

This is because that root_cgrp->bpf.refcnt.data is allocated by the
function percpu_ref_init in cgroup_bpf_inherit which is called by
cgroup_setup_root when mounting, but not freed along with root_cgrp
when umounting.

Good catch!

+1

quoted

quoted

Adding cgroup_bpf_offline which calls percpu_ref_kill to
cgroup_kill_sb can free root_cgrp->bpf.refcnt.data in umount path.

That is sensible.

quoted

Fixes: 2b0d3d3e4fcfb ("percpu_ref: reduce memory footprint of percpu_ref in fast path")

Why this Fixes:? Is the leak absent before the percpu_ref refactoring?

I agree, the "fixes" tag looks dubious to me.

quoted

I guess the embedded data are free'd together with cgroup. Makes me
wonder why struct cgroup_bpf has a separate percpu_ref counter from
struct cgroup...

This is because a cgroup can stay a long time (sometimes effectively forever)
in a dying state, so we want to release bpf structures earlier.

Thanks!

Other than whitespace LGTM.

Acked-by: John Fastabend <john.fastabend@gmail.com>