Re: [PATCH] cgroup: protect modifications to cgroup_idr with cgroup_mutex

[Michal Hocko <mhocko-AlSwsSmVLrQ@public.gmane.org>]

Hi Li,
good work in reproducing the issue so quickly!
I have tried to backport this patch to 3.12 kernel but the code has
changed since then.
The only two instances of idr_remove which are called outside of
cgroup_mutex seem to be:
 - cgroup_create calling it from err_free_id: path
 - css_free_work_fn
     mem_cgroup_css_free
       __mem_cgroup_free
         free_css_id

The second one takes ss->id_lock spinlock which should be sufficient
to exclude get_new_cssid but cgroup_mount and cgroup_create don't use
id_lock. They do hold cgroup_mutex though. So I guess I need something
like the following (I will have it tested):
---
Date: Tue, 11 Feb 2014 16:05:46 +0800
From: Li Zefan <lizefan-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
Subject: [PATCH - for 3.12] cgroup: protect modifications to cgroup_idr with cgroup_mutex

Setup cgroupfs like this:
  # mount -t cgroup -o cpuacct xxx /cgroup
  # mkdir /cgroup/sub1
  # mkdir /cgroup/sub2

Then run these two commands:
  # for ((; ;)) { mkdir /cgroup/sub1/tmp && rmdir /mnt/sub1/tmp; } &
  # for ((; ;)) { mkdir /cgroup/sub2/tmp && rmdir /mnt/sub2/tmp; } &

After seconds you may see this warning:

------------[ cut here ]------------
WARNING: CPU: 1 PID: 25243 at lib/idr.c:527 sub_remove+0x87/0x1b0()
idr_remove called for id=6 which is not allocated.
...
Call Trace:
 [<ffffffff8156063c>] dump_stack+0x7a/0x96
 [<ffffffff810591ac>] warn_slowpath_common+0x8c/0xc0
 [<ffffffff81059296>] warn_slowpath_fmt+0x46/0x50
 [<ffffffff81300aa7>] sub_remove+0x87/0x1b0
 [<ffffffff810f3f02>] ? css_killed_work_fn+0x32/0x1b0
 [<ffffffff81300bf5>] idr_remove+0x25/0xd0
 [<ffffffff810f2bab>] cgroup_destroy_css_killed+0x5b/0xc0
 [<ffffffff810f4000>] css_killed_work_fn+0x130/0x1b0
 [<ffffffff8107cdbc>] process_one_work+0x26c/0x550
 [<ffffffff8107eefe>] worker_thread+0x12e/0x3b0
 [<ffffffff81085f96>] kthread+0xe6/0xf0
 [<ffffffff81570bac>] ret_from_fork+0x7c/0xb0
---[ end trace 2d1577ec10cf80d0 ]---

It's because allocating/removing cgroup ID is not properly synchronized.

The bug was introduced when we converted cgroup_ida to cgroup_idr.
While synchronization is already done inside ida_simple_{get,remove}(),
users are responsible for concurrent calls to idr_{alloc,remove}().

[mhocko-AlSwsSmVLrQ@public.gmane.org: ported to 3.12]
Fixes: 4e96ee8e981b ("cgroup: convert cgroup_ida to cgroup_idr")
Cc: <stable-u79uwXL29TY76Z2rM5mHXA@public.gmane.org> #3.12+
Reported-by: Michal Hocko <mhocko-AlSwsSmVLrQ@public.gmane.org>
Signed-off-by: Li Zefan <lizefan-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>
---
 include/linux/cgroup.h |    2 ++
 kernel/cgroup.c        |   25 ++++++++++++++-----------
 2 files changed, 16 insertions(+), 11 deletions(-)

--- a/include/linux/cgroup.h
+++ b/include/linux/cgroup.h
@@ -169,6 +169,8 @@ struct cgroup {
 	 *
 	 * The ID of the root cgroup is always 0, and a new cgroup
 	 * will be assigned with a smallest available ID.
+	 *
+	 * Allocating/Removing ID must be protected by cgroup_mutex.
 	 */
 	int id;
 
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -4410,14 +4410,6 @@ static long cgroup_create(struct cgroup
 	rcu_assign_pointer(cgrp->name, name);
 
 	/*
-	 * Temporarily set the pointer to NULL, so idr_find() won't return
-	 * a half-baked cgroup.
-	 */
-	cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL);
-	if (cgrp->id < 0)
-		goto err_free_name;
-
-	/*
 	 * Only live parents can have children.  Note that the liveliness
 	 * check isn't strictly necessary because cgroup_mkdir() and
 	 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
@@ -4426,9 +4418,17 @@ static long cgroup_create(struct cgroup
 	 */
 	if (!cgroup_lock_live_group(parent)) {
 		err = -ENODEV;
-		goto err_free_id;
+		goto err_free_name;
 	}
 
+	/*
+	 * Temporarily set the pointer to NULL, so idr_find() won't return
+	 * a half-baked cgroup.
+	 */
+	cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL);
+	if (cgrp->id < 0)
+		goto err_unlock;
+
 	/* Grab a reference on the superblock so the hierarchy doesn't
 	 * get deleted on unmount if there are child cgroups.  This
 	 * can be done outside cgroup_mutex, since the sb can't
@@ -4542,11 +4542,11 @@ err_free_all:
 			ss->css_free(css);
 		}
 	}
+	idr_remove(&root->cgroup_idr, cgrp->id);
+err_unlock:
 	mutex_unlock(&cgroup_mutex);
 	/* Release the reference count that we took on the superblock */
 	deactivate_super(sb);
-err_free_id:
-	idr_remove(&root->cgroup_idr, cgrp->id);
 err_free_name:
 	kfree(rcu_dereference_raw(cgrp->name));
 err_free_cgrp:
@@ -5622,9 +5622,12 @@ void free_css_id(struct cgroup_subsys *s
 
 	rcu_assign_pointer(id->css, NULL);
 	rcu_assign_pointer(css->id, NULL);
+
+	mutex_lock(&cgroup_mutex);
 	spin_lock(&ss->id_lock);
 	idr_remove(&ss->idr, id->id);
 	spin_unlock(&ss->id_lock);
+	mutex_unlock(&cgroup_mutex);
 	kfree_rcu(id, rcu_head);
 }
 EXPORT_SYMBOL_GPL(free_css_id);
-- 
Michal Hocko
SUSE Labs