[PATCH 1/1] Smack:- Fix the issue of wrong info printed in ptrace error logs
Casey Schaufler
casey at schaufler-ca.com
Thu Feb 3 18:09:30 UTC 2022
On 2/3/2022 5:16 AM, Vishal Goel wrote:
> >>>>>>>> Currently tracer process info is printed in object field in
>>>>>>>>> smack error log for ptrace check which is wrong.
>>>>>>>>> Object process should print the tracee process info.
>>>>>>>>> Tracee info is not printed in the smack error logs.
>>>>>>>>> So it is not possible to debug the ptrace smack issues.
>>>>>>>>>
>>>>>>>>> Now changes has been done to print both tracer and tracee
>>>>>>>>> process info in smack error logs for ptrace scenarios
>>>>>>>>>
>>>>>>>>> Old logs:-
>>>>>>>>> [ 378.098330] audit: type=1400 audit(1637212273.300:2): lsm=SMACK fn=smack_ptrace_access_check action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=9397 comm="tst_pt" opid=9397 ocomm="tst_pt"
>>>>>>>>> [ 520.261605] audit: type=1400 audit(1637212415.464:3): lsm=SMACK fn=smack_ptrace_traceme action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=12685 comm="tst_pt_me" opid=12563 ocomm="bash"
>>>>>>>>> [ 1445.259319] audit: type=1400 audit(1637213340.460:5): lsm=SMACK fn=smack_bprm_set_creds action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=1778 comm="tst_bprm" opid=1776 ocomm="tst_bprm"
>>>>>>>>>
>>>>>>>>> New logs:-
>>>>>>>>> [ 378.098330] audit: type=1400 audit(1637212273.300:2): lsm=SMACK fn=smack_ptrace_access_check action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= tracer-pid=5189 tracer-comm="tst_pt" pid=5189 comm="tst_pt" tracee-pid=962 tracee-comm="test_tracee"
>>>>>>>>> [ 520.261605] audit: type=1400 audit(1637212415.464:3): lsm=SMACK fn=smack_ptrace_traceme action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= tracer-pid=6161 tracer-comm="bash" pid=6310 comm="tst_pt_me" tracee-pid=6310 tracee-comm="tst_pt_me"
>>>>>>>>> [ 1445.259319] audit: type=1400 audit(1637213340.460:5): lsm=SMACK fn=smack_bprm_set_creds action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= tracer-pid=6435 tracer-comm="tst_bprm" pid=6436 comm="tst_bprm" tracee-pid=6436 tracee-comm="tst_bprm"
>>>>>>>>>
>>>>>>>>> Signed-off-by: Vishal Goel <vishal.goel at samsung.com>
>>>>>>>> Does anyone from the audit side object to my taking this
>>>>>>>> in the Smack tree?
>>>>>>> The audit subsystem already has the "opid" and "ocomm" fields for
>>>>>>> reporting on the object task info and this is even available in
>>>>>>> dump_common_audit_data() via LSM_AUDIT_DATA_TASK; is there a reason
>>>>>>> that can't be used instead?
>>>>>> That info is not sufficient for debugging smack issues in ptrace calls.
>>>>>> Tracee information is not printed in the logs. For eg. in below log-
>>>>>> [ 378.098330] audit: type=1400 audit(1637212273.300:2): lsm=SMACK fn=smack_ptrace_access_check action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=9397 comm="tst_pt" opid=9397 ocomm="tst_pt"
>>>>>>
>>>>>> There is no information of the tracee process.
>>>>>> So to debug such ptrace issues, both tracer and tracee information is needed.
>>>>>> That's why added new type to print both info specifically for ptrace scenarios.
>>>>> From what I saw you are trying to record information about the tracer
>>>>> and the tracee, yes? The "pid", "comm", "opid", and "ocomm" fields
>>>>> should be used instead of adding new fields.
>>>> Actually in smack_ptrace_access_check() function, tracer process is current process.
>>>> While some other process is object process(tracee).
>>>> But in case of smack_ptrace_traceme() function, tracer process is parent process.
>>>> While current process is object process(tracee). So in this case, both pid/comm
>>>> and opid/ocomm will print current process info only i.e tracess process.
>>>> So tracer process info is not getting printed.
>>>> Similarly for smack_bprm_creds_for_exec(), tracer process is parent process.
>>>> And current process is tracee process.
>>>> So that's why we need to print separately tracer and tracee process info
>>>> without any confusion.
>>> The last time I checked, Smack's access controls operated as
>>> subject-verb-object triple, which should map nicely to the
>>> "pid"/"comm" and "opid"/"ocomm" fields; the former pair associated
>>> with the subject, the latter pair associated with the object. That
>>> combined with the "fn" field should give you all of the information
>>> relevant to the access control decision. If you feel that is not the
>>> case, perhaps that is an indicator that the information used in the
>>> access control decision is wrong, or there is a problem with the
>>> implementation.
>
>> Sorry that I've been absent from this discussion. Paul's right.
>> There's a whole lot of unnecessary complexity in this patch.
>> If you change the 2nd parameter of smk_ptrace_rule_check() to
>> be the task_struct of the tracee you should be able to do
>> exactly as Paul suggests.
> Yes,right. We can map the "opid/ocomm" pair to object/tracee process.
> But issue is to print subject info. In smack, subject can be current process
> or it can be parent of current process in some api.
> For eg. in smack_ptrace_access_check() function, subject/tracer process is current process.
> While in case of smack_ptrace_traceme() function, subject/tracer process is parent process.
> But as you know, that "pid/comm" pair always maps to current process according to below code:-
>
> static void dump_common_audit_data(struct audit_buffer *ab,
> struct common_audit_data *a)
> {
> audit_log_format(ab, " pid=%d comm=", task_tgid_nr(current));
> audit_log_untrustedstring(ab, memcpy(comm, current->comm, sizeof(comm)));
>
> So in some case, it prints correct info where current process is subject.
> But what about smack_ptrace_traceme() api, where subject is parent process not current process?
> So wrong subject info is getting printed currently.
> How can we map "pid/comm" to parent process which is actual subject in smack_ptrace_traceme() api case?
I'm not especially happy looking at the code as it is today.
Common code paths and helpers only work when you have commonality.
There have been changes made to the ptrace hooks that didn't take
these important differences into account. It looks like a rewrite
of smack_ptrace_traceme() that doesn't use the common path functions
is going to be necessary. Unfortunately, it's not going to be
elegant.
>
>
>
> --------- Original Message ---------
> Sender : Casey Schaufler <casey at schaufler-ca.com>
> Date : 2022-02-03 03:13 (GMT+9)
> Title : Re: [PATCH 1/1] Smack:- Fix the issue of wrong info printed in ptrace error logs
>
> On 2/2/2022 7:20 AM, Paul Moore wrote:
>> On Wed, Feb 2, 2022 at 5:38 AM Vishal Goel <vishal.goel at samsung.com> wrote:
>>>>>>>> Currently tracer process info is printed in object field in
>>>>>>>> smack error log for ptrace check which is wrong.
>>>>>>>> Object process should print the tracee process info.
>>>>>>>> Tracee info is not printed in the smack error logs.
>>>>>>>> So it is not possible to debug the ptrace smack issues.
>>>>>>>>
>>>>>>>> Now changes has been done to print both tracer and tracee
>>>>>>>> process info in smack error logs for ptrace scenarios
>>>>>>>>
>>>>>>>> Old logs:-
>>>>>>>> [ 378.098330] audit: type=1400 audit(1637212273.300:2): lsm=SMACK fn=smack_ptrace_access_check action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=9397 comm="tst_pt" opid=9397 ocomm="tst_pt"
>>>>>>>> [ 520.261605] audit: type=1400 audit(1637212415.464:3): lsm=SMACK fn=smack_ptrace_traceme action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=12685 comm="tst_pt_me" opid=12563 ocomm="bash"
>>>>>>>> [ 1445.259319] audit: type=1400 audit(1637213340.460:5): lsm=SMACK fn=smack_bprm_set_creds action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=1778 comm="tst_bprm" opid=1776 ocomm="tst_bprm"
>>>>>>>>
>>>>>>>> New logs:-
>>>>>>>> [ 378.098330] audit: type=1400 audit(1637212273.300:2): lsm=SMACK fn=smack_ptrace_access_check action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= tracer-pid=5189 tracer-comm="tst_pt" pid=5189 comm="tst_pt" tracee-pid=962 tracee-comm="test_tracee"
>>>>>>>> [ 520.261605] audit: type=1400 audit(1637212415.464:3): lsm=SMACK fn=smack_ptrace_traceme action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= tracer-pid=6161 tracer-comm="bash" pid=6310 comm="tst_pt_me" tracee-pid=6310 tracee-comm="tst_pt_me"
>>>>>>>> [ 1445.259319] audit: type=1400 audit(1637213340.460:5): lsm=SMACK fn=smack_bprm_set_creds action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= tracer-pid=6435 tracer-comm="tst_bprm" pid=6436 comm="tst_bprm" tracee-pid=6436 tracee-comm="tst_bprm"
>>>>>>>>
>>>>>>>> Signed-off-by: Vishal Goel <vishal.goel at samsung.com>
>>>>>>> Does anyone from the audit side object to my taking this
>>>>>>> in the Smack tree?
>>>>>> The audit subsystem already has the "opid" and "ocomm" fields for
>>>>>> reporting on the object task info and this is even available in
>>>>>> dump_common_audit_data() via LSM_AUDIT_DATA_TASK; is there a reason
>>>>>> that can't be used instead?
>>>>> That info is not sufficient for debugging smack issues in ptrace calls.
>>>>> Tracee information is not printed in the logs. For eg. in below log-
>>>>> [ 378.098330] audit: type=1400 audit(1637212273.300:2): lsm=SMACK fn=smack_ptrace_access_check action=denied subject="Tracer_lbl" object="Tracee_lbl" requested= pid=9397 comm="tst_pt" opid=9397 ocomm="tst_pt"
>>>>>
>>>>> There is no information of the tracee process.
>>>>> So to debug such ptrace issues, both tracer and tracee information is needed.
>>>>> That's why added new type to print both info specifically for ptrace scenarios.
>>>> From what I saw you are trying to record information about the tracer
>>>> and the tracee, yes? The "pid", "comm", "opid", and "ocomm" fields
>>>> should be used instead of adding new fields.
>>> Actually in smack_ptrace_access_check() function, tracer process is current process.
>>> While some other process is object process(tracee).
>>> But in case of smack_ptrace_traceme() function, tracer process is parent process.
>>> While current process is object process(tracee). So in this case, both pid/comm
>>> and opid/ocomm will print current process info only i.e tracess process.
>>> So tracer process info is not getting printed.
>>> Similarly for smack_bprm_creds_for_exec(), tracer process is parent process.
>>> And current process is tracee process.
>>> So that's why we need to print separately tracer and tracee process info
>>> without any confusion.
>> The last time I checked, Smack's access controls operated as
>> subject-verb-object triple, which should map nicely to the
>> "pid"/"comm" and "opid"/"ocomm" fields; the former pair associated
>> with the subject, the latter pair associated with the object. That
>> combined with the "fn" field should give you all of the information
>> relevant to the access control decision. If you feel that is not the
>> case, perhaps that is an indicator that the information used in the
>> access control decision is wrong, or there is a problem with the
>> implementation.
>
> Sorry that I've been absent from this discussion. Paul's right.
> There's a whole lot of unnecessary complexity in this patch.
> If you change the 2nd parameter of smk_ptrace_rule_check() to
> be the task_struct of the tracee you should be able to do
> exactly as Paul suggests.
>
>> --
>> paul-moore.com
>
More information about the Linux-audit
mailing list