On 7/30/20 6:02 PM, Steven Rostedt wrote:
> On Thu, 30 Jul 2020 17:31:17 +0200
> peter enderborg <peter.enderborg@xxxxxxxx> wrote:
>
>> On 7/30/20 5:04 PM, Steven Rostedt wrote:
>>> On Thu, 30 Jul 2020 16:29:12 +0200
>>> peter enderborg <peter.enderborg@xxxxxxxx> wrote:
>>>
>>>> +#undef TRACE_SYSTEM
>>>> +#define TRACE_SYSTEM avc
>>>> +
>>>> +#if !defined(_TRACE_AVC_H) || defined(TRACE_HEADER_MULTI_READ)
>>>> +#define _TRACE_AVC_H
>>>> +
>>>> +#include <linux/tracepoint.h>
>>>> +TRACE_EVENT(avc_data,
>>>> + TP_PROTO(u32 requested,
>>>> + u32 denied,
>>>> + u32 audited,
>>>> + int result,
>>>> + const char *msg
>>>> + ),
>>>> +
>>>> + TP_ARGS(requested, denied, audited, result,msg),
>>>> +
>>>> + TP_STRUCT__entry(
>>>> + __field(u32, requested)
>>>> + __field(u32, denied)
>>>> + __field(u32, audited)
>>>> + __field(int, result)
>>>> + __array(char, msg, 255)
>>> You want to use __string() here, otherwise you are wasting a lot of
>>> buffer space.
>>>
>>> __string( msg, msg)
>> It should be a full structure with a lot of sub strings. But that make is even more relevant.
> So one event instance can have a list of strings recorded?
Yes, it is a list very similar to a normal trace. But it is more generic.
For example ino= is for filesystems that have inode, but for a
violation that send a signal that make no sense at all. Network
addresses is in many cases not applicable. laddr= is only exist for
for IP.
So if you just print them it will look like:
avc: denied { find } for interface=vendor.qti.hardware.perf::IPerf sid=u:r:permissioncontroller_app:s0:c230,c256,c512,c768 pid=9164 scontext=u:r:permissioncontroller_app:s0:c230,c256,c512,c768 tcontext=u:object_r:vendor_hal_perf_hwservice:s0 tclass=hwservice_manager permissive=0
avc: denied { execute } for pid=13914 comm="ScionFrontendAp" path="/data/user_de/0/com.google.android.gms/app_chimera/m/00000002/oat/arm64/DynamiteLoader.odex" dev="sda77" ino=204967 scontext=u:r:platform_app:s0:c512,c768 tcontext=u:object_r:privapp_data_file:s0:c512,c768 tclass=file permissive=0 ppid=788 pcomm="main" pgid=13914 pgcomm="on.updatecenter"
It omit the fields that are not used. Some parts are common some are not. So a correct format specification for trace will be problematic if there is no "optional" field indicator.
>
>>>
>>>> + ),
>>>> +
>>>> + TP_fast_assign(
>>>> + __entry->requested = requested;
>>>> + __entry->denied = denied;
>>>> + __entry->audited = audited;
>>>> + __entry->result = result;
>>>> + memcpy(__entry->msg, msg, 255);
>>> Not to mention, the above is a bug. As the msg being passed in, is
>>> highly unlikely to be 255 bytes. You just leaked all that memory after
>>> the sting to user space.
>>>
>>> Where you want here:
>>>
>>> __assign_str( msg, msg );
>> Directly in to the code. Was more in to get in to discussion on how complex we should have
>> the trace data. There is a lot of fields. Not all is always present. Is there any good way
>> to handle that? Like "something= somethingelse=42" or "something=nil somthingelse=42"
> Can you show what you want to record and what you want to display? I'm
> not totally understanding the request.
>
> -- Steve
>
>>>> + ),
>>>> +
>>>> + TP_printk("requested=0x%x denied=%d audited=%d result=%d
>>>> msg=%s",
>>>> + __entry->requested, __entry->denied, __entry->audited,
>>>> __entry->result, __entry->msg
>>>> + )
