Re: [PATCH] ima: fix deadlock within RCU list of ima_rules

From: liqiong
Date: Wed Aug 25 2021 - 03:06:08 EST


Hi Mimi,

Thanks for the advice,maybe i should trim the message,
here is a new copy:


subject: ima: fix deadlock when iterating over the init "ima_rules" list.

The init "ima_rules" list can't traverse back to head, if "ima_rules"
is being updated to "ima_policy_rules". It causes soft lockup and RCU stalls.
So we can introduce a duplicate of "ima_rules" for each "ima_rules" list loop.

Signed-off-by: liqiong <liqiong@xxxxxxxxxxxx>
---
This problem can happen in practice: updating the IMA policy
in the boot process while systemd-services are being checked.

security/integrity/ima/ima_policy.c | 17 ++++++++++++-----
1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/security/integrity/ima/ima_policy.c b/security/integrity/ima/ima_policy.c
index fd5d46e511f1..e92b197bfd3c 100644
--- a/security/integrity/ima/ima_policy.c
+++ b/security/integrity/ima/ima_policy.c


Regards,

liqiong

在 2021年08月24日 20:38, Mimi Zohar 写道:
> On Tue, 2021-08-24 at 20:09 +0800, liqiong wrote:
>> Hi Simon :
>>
>> ima: fix deadlock within RCU list of ima_rules.
>>
> Before the following paragraph, an introductory sentence is needed.
> Try adding a sentence to the affect that "ima_rules" initially points
> to the "ima_default_rules", but after loading a custom policy points to
> the "ima_policy_rules". Then describe the bug at a high level,
> something like - transitioning to the "ima_policy_rules" isn't being
> done safely.
>
> Followed by the details.
>
>> ima_match_policy() is looping on the policy ruleset while
>> ima_update_policy() updates the variable "ima_rules". This can
>> lead to a situation where ima_match_policy() can't exit the
>> 'list_for_each_entry_rcu' loop, causing RCU stalls
>> ("rcu_sched detected stall on CPU ...").
>>
>> This problem can happen in practice: updating the IMA policy
>> in the boot process while systemd-services are being checked.
>>
>> In addition to ima_match_policy(), other function with
>> "list_for_each_entry_rcu" should happen too. Fix locking by
>> introducing a duplicate of "ima_rules" for each
>> "list_for_each_entry_rcu".
>>
>>
>> How about this commit message ?
>>
>> I have tested this patch in lab, we can reproduced this error case,
>> have done reboot test many times. This patch should work.
> The above comment doesn't belong in the commit message, but is a
> message to the reviewers/maintainers and goes after the patch
> descriptions three dashes line.
>
> thanks,
>
> Mimi
>
>