Fault Injection in strace [GSOC 2016]. A sample implementation.

[haris iqbal]

Hey,

My apologies. Can I get some comments on my POC.

Also I would want to know how exactly should faults be injected into
the syscalls. My knowledge says that this injection of faults would be
specific for every system calls.

Another things is, in my example, one cannot use the -e option and the
-g option together. To make that possible, one would need another
array of flags to determine the chosen system calls for the -g option.
I am thinking the right way?


On Tue, Mar 8, 2016 at 10:36 PM, haris iqbal <haris.phnx at gmail.com> wrote:
> On Tue, Mar 8, 2016 at 10:34 PM, haris iqbal <haris.phnx at gmail.com> wrote:
>> On Tue, Mar 8, 2016 at 10:00 PM, Gabriel Laskar <gabriel at lse.epita.fr> wrote:
>>> On Tue, 8 Mar 2016 18:30:37 +0530
>>> haris iqbal <haris.phnx at gmail.com> wrote:
>>>
>>>> Hello,
>>>>
>>>> I have implemented a very rudimentary option in strace to inject
>>>> faults for certain system calls. It uses the code written for the -e
>>>> option. One has to provide the system calls to be failed using the
>>>> newly implemented -g option. It seems to be working.
>>>>
>>>> Since it is not a patch, how should I share the code for everyone to
>>>> see.
>>>>
>>>
>>> Can't you give us a diff?
>>
>> Yes. Here it is. I did this today evening, so the code might seem like
>> very temporary. I just wanted to check whether the way I am thinking
>> would work or not.
>>
>> diff --git a/strace.c b/strace.c
>> index 49d6f3d..42cd9a6 100644
>> --- a/strace.c
>> +++ b/strace.c
>> @@ -1487,6 +1487,11 @@ get_os_release(void)
>>   * Don't want main() to inline us and defeat the reason
>>   * we have a separate function.
>>   */
>> +
>> +void fail_syscall(int);
>> +void set_fail_flag(void);
>> +void set_failing_parameters(char* str);
>> +
>>  static void ATTRIBUTE_NOINLINE
>>  init(int argc, char *argv[])
>>  {
>> @@ -1523,7 +1528,7 @@ init(int argc, char *argv[])
>>          "k"
>>  #endif
>>          "D"
>> -        "a:e:o:O:p:s:S:u:E:P:I:")) != EOF) {
>> +        "a:e:g:j:o:O:p:s:S:u:E:P:I:")) != EOF) {
>>          switch (c) {
>>          case 'b':
>>              if (strcmp(optarg, "execve") != 0)
>> @@ -1600,6 +1605,13 @@ init(int argc, char *argv[])
>>          case 'e':
>>              qualify(optarg);
>>              break;
>> +        case 'g':
>> +            qualify(optarg);
>> +            set_fail_flag();
>> +            break;
>> +        case 'j':
>> +            set_failing_parameters(optarg);
>> +            break;
>>          case 'o':
>>              outfname = xstrdup(optarg);
>>              break;
>> @@ -2322,7 +2334,8 @@ show_stopsig:
>>       * This should be syscall entry or exit.
>>       * Handle it.
>>       */
>> -    if (trace_syscall(tcp) < 0) {
>> +    int temp = trace_syscall(tcp);
>> +    if (temp < 0) {
>>          /*
>>           * ptrace() failed in trace_syscall().
>>           * Likely a result of process disappearing mid-flight.
>> @@ -2337,6 +2350,11 @@ show_stopsig:
>>          return true;
>>      }
>>
>> +    if(temp == 3)
>> +    {
>> +        fail_syscall(pid);
>> +    }
>> +
>>  restart_tracee_with_sig_0:
>>      sig = 0;
>>
>> diff --git a/syscall.c b/syscall.c
>> index 6efcde5..44fe5f3 100644
>> --- a/syscall.c
>> +++ b/syscall.c
>> @@ -785,12 +785,43 @@ static void get_error(struct tcb *, const bool);
>>  static int getregs_old(pid_t);
>>  #endif
>>
>> +/* custom for POC of fault injection*/
>> +
>> +void fail_syscall(int);
>> +void set_fail_flag(void);
>> +void set_failing_parameters(const char*);
>> +
>> +
>> +
>> +unsigned int fail_flag = 0;
>> +int probability = 100;
>> +void set_fail_flag()
>> +{
>> +    fail_flag = 1;
>> +    return;
>> +}
>> +
>> +void set_failing_parameters(const char* str)
>> +{
>> +    probability = atoi(str);
>> +    return;
>> +}
>> +
>> +/* END */
>> +
>> +
>>  static int
>>  trace_syscall_entering(struct tcb *tcp)
>>  {
>>      int res, scno_good;
>>
>>      scno_good = res = get_scno(tcp);
>> +    if ((tcp->qual_flg & QUAL_TRACE) && fail_flag == 1)
>> +    {
>> +        // probablity needs to be added using randon number genaratiion
>> +        tprintf("%s system called failed with probability %d\n",
>> syscall_name(tcp->scno), probability);
>> +        return 3;
>> +    }
>>      if (res == 0)
>>          return res;
>>      if (res == 1)
>> @@ -1241,6 +1272,16 @@ get_regset(pid_t pid)
>>  }
>>  #endif /* ARCH_REGS_FOR_GETREGSET */
>>
>> +void fail_syscall(pid)
>> +{
>> +    void * fail_struct = calloc(1, sizeof(ARCH_IOVEC_FOR_GETREGSET));
>> +    tprintf("done");
>> +
>> +    ptrace(PTRACE_SETREGS, pid, NT_PRSTATUS, fail_struct);
>> +    free(fail_struct);
>> +}
>
> I used this technique just to fail the system call. I know this is not
> a practical way. I just did this to test whether the idea would work
> or not.
>
> What should be the proper way to fail a given system call?
>
>> +
>> +
>>  void
>>  get_regs(pid_t pid)
>>  {
>>
>>
>>>
>>> --
>>> Gabriel Laskar
>>
>>
>>
>> --
>>
>> With regards,
>>
>> Md Haris Iqbal,
>> Placement Coordinator, MTech IT
>> NITK Surathkal,
>> Contact: +91 8861996962
>
>
>
> --
>
> With regards,
>
> Md Haris Iqbal,
> Placement Coordinator, MTech IT
> NITK Surathkal,
> Contact: +91 8861996962



-- 

With regards,

Md Haris Iqbal,
Placement Coordinator, MTech IT
NITK Surathkal,
Contact: +91 8861996962