* Cannot access data passed in via gettimeofday
@ 2006-12-14 23:20 Perry Cheng
2006-12-15 3:16 ` Eugene Teo
2006-12-15 21:08 ` Frank Ch. Eigler
0 siblings, 2 replies; 4+ messages in thread
From: Perry Cheng @ 2006-12-14 23:20 UTC (permalink / raw)
To: systemtap
Hi all,
I am having trouble getting parameters out of what seems to be a simple
but buggy system-tap script. The script is included below and the test
case (a short C program) follows. Basically, I am trying to hi-jack
gettimeofday and piggyback some information back by treating the struct it
passes in as a larger buffer than struct timeval. However, I need to
separate regular calls to gettimeofday from the special ones where the
special path triggers. To do this, I though I could treat the struct as
an incoming parameter by looking for unusual bit patterns in the struct.
Unfortunately, I can't seem to see the data at all despite using the
copy_from_user function to copy data from user to kernel space. Any idea
what's going on here? I additionally hijack stime so i can distinguish
in the output my special call to gettimeofday. The sample out below
shows that the special values 0xaaaaaaaa and 0xbbbbbbbb are not
transmitted. If I use settimeofday instead of gettimeofday, then this
program seems to work. It feels like there is some other mechanism at
work here that I don't know about.
---------------------------
gettimeofday 0: sec = 4ec7ed3c usec = 4eb6db96
---------------------------
**********************system tap script**********************
probe kernel.function("sys_stime")
{
doTime()
}
function doTime()
%{
_stp_printf("---------------------------\n");
%}
probe kernel.function("sys_gettimeofday")
{
doGetTimeOfDay($tv)
}
function doGetTimeOfDay(tv:long)
%{
struct timeval *tv = (struct timeval *) ((long)THIS->tv);
struct timeval ltv;
int notCopied = 0;
notCopied = copy_from_user(<v, tv, sizeof(ltv));
_stp_printf("gettimeofday %d: sec = %x usec = %x\n", notCopied,
ltv.tv_sec, ltv.tv_usec);
%}
**********************test program**********************
#include <sys/time.h>
#include <stdio.h>
#include <stdlib.h>
main() {
struct timeval tv;
time_t t;
stime(&t);
tv.tv_sec = 0xaaaaaaaa;
tv.tv_usec = 0xbbbbbbbb;
gettimeofday(&tv, NULL);
stime(&t);
}
Perry
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Cannot access data passed in via gettimeofday
2006-12-14 23:20 Cannot access data passed in via gettimeofday Perry Cheng
@ 2006-12-15 3:16 ` Eugene Teo
2006-12-15 21:08 ` Frank Ch. Eigler
1 sibling, 0 replies; 4+ messages in thread
From: Eugene Teo @ 2006-12-15 3:16 UTC (permalink / raw)
To: Perry Cheng; +Cc: systemtap
Perry Cheng wrote:
> I am having trouble getting parameters out of what seems to be a simple
> but buggy system-tap script. The script is included below and the test
> case (a short C program) follows. Basically, I am trying to hi-jack
> gettimeofday and piggyback some information back by treating the struct it
> passes in as a larger buffer than struct timeval. However, I need to
> separate regular calls to gettimeofday from the special ones where the
> special path triggers. To do this, I though I could treat the struct as
> an incoming parameter by looking for unusual bit patterns in the struct.
> Unfortunately, I can't seem to see the data at all despite using the
> copy_from_user function to copy data from user to kernel space. Any idea
> what's going on here? I additionally hijack stime so i can distinguish
> in the output my special call to gettimeofday. The sample out below
> shows that the special values 0xaaaaaaaa and 0xbbbbbbbb are not
> transmitted. If I use settimeofday instead of gettimeofday, then this
> program seems to work. It feels like there is some other mechanism at
> work here that I don't know about.
[eteo@kerndev tmp]$ stap -g test.stp -c ./test
Password:
---------------------------
gettimeofday 0: sec = aaaaaaaa usec = bbbbbbbb
---------------------------
gettimeofday 0: sec = 23a usec = 1
...
Eugene
--
1024D/58DF8823 print 47B9 90F6 AE4A 9C51 37E0 D6E1 EA84 C6A2 58DF 8823
main(i) { putchar(182623909 >> (i-1) * 5&31|!!(i<7)<<6) && main(++i); }
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Cannot access data passed in via gettimeofday
2006-12-14 23:20 Cannot access data passed in via gettimeofday Perry Cheng
2006-12-15 3:16 ` Eugene Teo
@ 2006-12-15 21:08 ` Frank Ch. Eigler
1 sibling, 0 replies; 4+ messages in thread
From: Frank Ch. Eigler @ 2006-12-15 21:08 UTC (permalink / raw)
To: Perry Cheng; +Cc: systemtap
Perry Cheng <perryche@us.ibm.com> writes:
> [...] I can't seem to see the data at all despite using the
> copy_from_user function to copy data from user to kernel space.
> [...]
Please be aware that might_sleep functions such as copy_from_user
should not generally be called from systemtap probe handlers.
Instead, one needs to use the atomic variants provided in the standard
tapset ... except we seem to lack a variant for copying over a struct
or an integer! We have had one old bug to improve target expression
syntax so this would be automated (#2049, to allow e.g. $tv=>tv_sec).
All that doesn't explain though why you're getting odd results. Does
the $tv pointer value itself seem plausible? Kernel/gcc version?
- FChE
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Cannot access data passed in via gettimeofday
[not found] <OF326DF462.A6D86C8A-ON85257245.00099860-85257245.00099F31@mck.us.ray.com>
@ 2006-12-15 3:45 ` Dave Sperry
0 siblings, 0 replies; 4+ messages in thread
From: Dave Sperry @ 2006-12-15 3:45 UTC (permalink / raw)
To: Perry Cheng, systemtap
Perry,
The test worked the first time I ran it. The second time I ran it I got
wacky results. Then I noticed I had a number of runaway processes after
the second test.
You may want to try a reboot and run the test again
Dave
> Perry Cheng wrote:
> > I am having trouble getting parameters out of what seems to be a simple
> > but buggy system-tap script. The script is included below and the test
> > case (a short C program) follows. Basically, I am trying to hi-jack
> > gettimeofday and piggyback some information back by treating the struct
> it
> > passes in as a larger buffer than struct timeval. However, I need to
> > separate regular calls to gettimeofday from the special ones where the
> > special path triggers. To do this, I though I could treat the struct as
> > an incoming parameter by looking for unusual bit patterns in the struct.
> > Unfortunately, I can't seem to see the data at all despite using the
> > copy_from_user function to copy data from user to kernel space. Any idea
>
> > what's going on here? I additionally hijack stime so i can distinguish
> > in the output my special call to gettimeofday. The sample out below
> > shows that the special values 0xaaaaaaaa and 0xbbbbbbbb are not
> > transmitted. If I use settimeofday instead of gettimeofday, then this
> > program seems to work. It feels like there is some other mechanism at
> > work here that I don't know about.
>
> [eteo@kerndev tmp]$ stap -g test.stp -c ./test
> Password:
> ---------------------------
> gettimeofday 0: sec = aaaaaaaa usec = bbbbbbbb
> ---------------------------
> gettimeofday 0: sec = 23a usec = 1
> ...
>
> Eugene
> --
> 1024D/58DF8823 print 47B9 90F6 AE4A 9C51 37E0 D6E1 EA84 C6A2 58DF 8823
> main(i) { putchar(182623909 >> (i-1) * 5&31|!!(i<7)<<6) && main(++i); }
>
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2006-12-15 13:17 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2006-12-14 23:20 Cannot access data passed in via gettimeofday Perry Cheng
2006-12-15 3:16 ` Eugene Teo
2006-12-15 21:08 ` Frank Ch. Eigler
[not found] <OF326DF462.A6D86C8A-ON85257245.00099860-85257245.00099F31@mck.us.ray.com>
2006-12-15 3:45 ` Dave Sperry
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).