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

[liqiong]

Hi Simon,

Thanks for you help, i may got it, here is the new commit message:


The current IMA ruleset is identified by the variable "ima_rules"
that default to "&ima_default_rules". When loading a custom policy
for the first time, the variable is updated to "&ima_policy_rules"
instead. That update isn't RCU-safe, and deadlocks are possible.
Because some functions like ima_match_policy() may loop indefinitely
over traversing the "ima_default_rules" as list_for_each_entry_rcu().

When iterating over the default ruleset back to head, value of
"&entry->list" is "&ima_default_rules", and "ima_rules" may have been
updated to "&ima_policy_rules", the loop condition (&entry->list != ima_rules)
stay alway true, traversing doesn't terminate, cause soft lockup and
RCU stalls.

Introduce a temporary value for "ima_rules" when iterating over
the ruleset to avoid the deadlocks.


Signed-off-by: liqiong <liqiong@xxxxxxxxxxxx>
---
 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


Thanks

liqiong

在 2021年08月26日 17:01, THOBY Simon 写道:
> Hi liqiong,
>
> On 8/26/21 10:15 AM, liqiong wrote:
>> Hi Simon,
>>
>> Thanks for your help, your advice is clear, can i just use it,
>> how about this:
>>
>>
>> The current IMA ruleset is identified by the variable "ima_rules",
>> and the pointer starts pointing at the list "ima_default_rules".
> After reading it again, maybe
> "The current IMA ruleset is identified by the variable "ima_rules",
> that defaults to "&ima_default_rules".'?
>
>> When loading a custom policy for the first time, the variable is
>> updated to point to the list "ima_policy_rules" instead. That update
>> isn't RCU-safe, and deadlocks are possible.
> I think what Mimi was trying to say is that you could add the high-level
> overview above, but keep the details. Sorry if I wasn't clearer in my
> earlier messages.
>
> Consider re-adding your previous paragraph
> """
> As a consequence, when traversing the ruleset, some functions like ima_match_policy()
> may loop indefinitely over "ima_default_rules" as list_for_each_entry_rcu() doesn't
> terminate (after the update, "ima_rules" no longer points to the list head, so the
> loop condition stays always true), causing RCU stalls.
> """
> (note: I tweaked it above, feel free to fix it)
>> Introduce a temporary value for "ima_rules" when iterating over
>> the ruleset to avoid the deadlocks.
> ... while keeping this a separate paragraph.
>
>>
>> Signed-off-by: liqiong <liqiong@xxxxxxxxxxxx>
>> ---
>>  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
>>
>>
>>
>> Thanks
>>
>> liqiong