dtrace discuss - Apr 2009 - trying to find argv[0] for writing processes

Hi,

I''m trying to modify the dtrace toolkit script iotop to print argv[0]
rather than execname. This is because I have a very busy E25k running multiple
oracle instances so execname always gives me ''oracle'' while
argv[0] returns the oracle SID. I could post-process the output using the pid,
but I''d rather just have it print argv0. This is a sample script that
demonstrates my problem.

#!/usr/sbin/dtrace -s
#pragma D option quiet
io:genunix::done
{
    a1 = copyinstr((uintptr_t)(curpsinfo->pr_dmodel == PR_MODEL_ILP32 ?
*(uint32_t *)copyin(arg1, 4) : *(uint64_t *)copyin(arg1, 8)));
    printf("%d:%s\n", pid, a1);
}

This just seems to produce the following style errors
dtrace: error on enabled probe ID 1 (ID 8628: io:genunix:biodone:done): invalid
address (0x0) in action #1 at DIF offset 256
dtrace: error on enabled probe ID 1 (ID 8628: io:genunix:biodone:done): invalid
address (0x2a100a46000) in action #1 at DIF offset 256
dtrace: error on enabled probe ID 1 (ID 8628: io:genunix:biodone:done): invalid
address (0x0) in action #1 at DIF offset 256
dtrace: error on enabled probe ID 1 (ID 8628: io:genunix:biodone:done): invalid
address (0x0) in action #1 at DIF offset 256

I''ve tried io:genunix::done and io:genunix::start. Both behave the same
way.
It works most of the time with syscall::exece:entry, but that''s not
what I''m interested in.

I could use this
#!/usr/sbin/dtrace -s
#pragma D option quiet
io:genunix::start
{
    printf("%d:%s\n", pid, curpsinfo->pr_psargs);
}
which works a treat, but gives all of the argument vector, I only want the first
entry!

Does anyone know how to extract this info? I don''t want to use stop()
and inspect /proc because I really don''t want to put that sort of
overhead onto the system.

Cheers,
Peter
-- 
This message posted from opensolaris.org

On Tue, Apr 28, 2009 at 08:05:32AM -0700, Peter Stubbs
wrote:
> Hi,
> 
> I''m trying to modify the dtrace toolkit script iotop to print
argv[0]
> rather than execname. This is because I have a very busy E25k running
> multiple oracle instances so execname always gives me
''oracle'' while
> argv[0] returns the oracle SID. I could post-process the output using
> the pid, but I''d rather just have it print argv0. This is a sample
> script that demonstrates my problem.
> 
> #!/usr/sbin/dtrace -s
> #pragma D option quiet
> io:genunix::done
> {
>     a1 = copyinstr((uintptr_t)(curpsinfo->pr_dmodel == PR_MODEL_ILP32 ?
*(uint32_t *)copyin(arg1, 4) : *(uint64_t *)copyin(arg1, 8)));
>     printf("%d:%s\n", pid, a1);
> }
There''s some confusion here: arg1 is the second *argument to the
probe*, not
argv[1] for the current process.
> I''ve tried io:genunix::done and io:genunix::start. Both behave the
same way.
> 
> It works most of the time with syscall::exece:entry, but that''s
not
> what I''m interested in.
> I could use this
> #!/usr/sbin/dtrace -s
> #pragma D option quiet
> io:genunix::start
> {
>     printf("%d:%s\n", pid, curpsinfo->pr_psargs);
> }
> which works a treat, but gives all of the argument vector, I only want
> the first entry!
>
> Does anyone know how to extract this info? I don''t want to use
stop()
> and inspect /proc because I really don''t want to put that sort of
> overhead onto the system.
If you want the first argument, then:

io:genunix::start
{
	this->execname = strtok(curpsinfo->pr_psargs, " ");

	this->firstarg = strtok(NULL, " ");
	this->firstarg = (this->firstarg != NULL) ? this->firstarg :
"";

	printf("%d:%s %s\n", pid, this->execname, this->firstarg);
}

Cheers,
- jonathan

Jonathan,

Thankyou for putting me out of my misery! It''s works at last.

Thanks again!
Peter
-- 
This message posted from opensolaris.org

Hi Again,

That''s solved my problem, but what happens if argv[0] has a space in
it? That would be very uncommon on solaris, but on a Mac it''s more
common. It seems strange that all the arguments are only available bunched
together with spaces between them. Is there a more reliable way to get this
info?

Cheers,
Peter
-- 
This message posted from opensolaris.org

I''m not 100% sure how one finds it in Solaris, but isn''t
argv[0] on the main
thread''s stack?

On Wed, Apr 29, 2009 at 8:48 AM, Peter Stubbs <ps258 at hotmail.com>
wrote:
> Hi Again,
>
> That''s solved my problem, but what happens if argv[0] has a space
in it?
> That would be very uncommon on solaris, but on a Mac it''s more
common. It
> seems strange that all the arguments are only available bunched together
> with spaces between them. Is there a more reliable way to get this info?
>
> Cheers,
> Peter
> --
> This message posted from opensolaris.org
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
>


-- 
colin.burgess at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/dtrace-discuss/attachments/20090430/357f7b33/attachment.html>

argv[0] is on the stack but not always in the same place. See
http://blogs.sun.com/thejel/category/Sun for the
26 Jul 2007 entry that discusses this.

Jim
---
Colin Burgess wrote:
> I''m not 100% sure how one finds it in Solaris, but isn''t
argv[0] on
> the main thread''s stack?
>
> On Wed, Apr 29, 2009 at 8:48 AM, Peter Stubbs <ps258 at hotmail.com 
> <mailto:ps258 at hotmail.com>> wrote:
>
>     Hi Again,
>
>     That''s solved my problem, but what happens if argv[0] has a
space
>     in it? That would be very uncommon on solaris, but on a Mac
it''s
>     more common. It seems strange that all the arguments are only
>     available bunched together with spaces between them. Is there a
>     more reliable way to get this info?
>
>     Cheers,
>     Peter
>     --
>     This message posted from opensolaris.org <http://opensolaris.org>
>     _______________________________________________
>     dtrace-discuss mailing list
>     dtrace-discuss at opensolaris.org <mailto:dtrace-discuss at
opensolaris.org>
>
>
>
>
> -- 
> colin.burgess at gmail.com <mailto:colin.burgess at gmail.com>
> ------------------------------------------------------------------------
>
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
>

Yes - funnily enough that''s almost exactly the same as QNX  32bit
Intel. :-)

I guess it would be a bit evil in Dtrace to snoop through the various arrays
to find the argv[0] there... :-)

On Thu, Apr 30, 2009 at 5:53 PM, James Litchfield
<James.Litchfield at sun.com>wrote:
> argv[0] is on the stack but not always in the same place. See
> http://blogs.sun.com/thejel/category/Sun for the
> 26 Jul 2007 entry that discusses this.
>
> Jim
> ---
> Colin Burgess wrote:
>
>> I''m not 100% sure how one finds it in Solaris, but
isn''t argv[0] on the
>> main thread''s stack?
>>
>> On Wed, Apr 29, 2009 at 8:48 AM, Peter Stubbs <ps258 at hotmail.com
<mailto:
>> ps258 at hotmail.com>> wrote:
>>
>>    Hi Again,
>>
>>    That''s solved my problem, but what happens if argv[0] has a
space
>>    in it? That would be very uncommon on solaris, but on a Mac
it''s
>>    more common. It seems strange that all the arguments are only
>>    available bunched together with spaces between them. Is there a
>>    more reliable way to get this info?
>>
>>    Cheers,
>>    Peter
>>    --
>>    This message posted from opensolaris.org
<http://opensolaris.org>
>>    _______________________________________________
>>    dtrace-discuss mailing list
>>    dtrace-discuss at opensolaris.org <mailto:dtrace-discuss at
opensolaris.org>
>>
>>
>>
>>
>> --
>> colin.burgess at gmail.com <mailto:colin.burgess at gmail.com>
>>
------------------------------------------------------------------------
>>
>> _______________________________________________
>> dtrace-discuss mailing list
>> dtrace-discuss at opensolaris.org
>>
>>
>
>

-- 
colin.burgess at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/dtrace-discuss/attachments/20090430/5dfabf5d/attachment.html>