dtrace discuss - Nov 2005 - Capturing return values && inet_* routines?

Howdy,

Is there by any chance a way to capture the value returned from a function? I
am tracing a specific function that returns a char *, and would like to record
this value each time the function returns (I am currently using truss, but
DTrace
would be an optimal fit for what I am trying to do). Also, are the inet_*
routines by any chance available to manipulate IPv4 and IPv6 addresses? I
checked
the Variables and Actions and Subroutines sections in the DTrace users guide,
but wasn''t able to find an answer to my questions.

Thanks,
- Ryan
--
UNIX Administrator
http://daemons.net/~matty

Check out the chapters on the fbt and pid providers. The return value for
non-tail call returns will be in arg1.

Adam

On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
> 
> Howdy,
> 
> Is there by any chance a way to capture the value returned from a function?
> I
> am tracing a specific function that returns a char *, and would like to 
> record
> this value each time the function returns (I am currently using truss, but 
> DTrace
> would be an optimal fit for what I am trying to do). Also, are the inet_*
> routines by any chance available to manipulate IPv4 and IPv6 addresses? I 
> checked
> the Variables and Actions and Subroutines sections in the DTrace users 
> guide,
> but wasn''t able to find an answer to my questions.
> 
> Thanks,
> - Ryan
> --
> UNIX Administrator
> http://daemons.net/~matty
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
-- 
Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl

On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
> 
> Howdy,
> 
> Is there by any chance a way to capture the value returned from a function?
> I
> am tracing a specific function that returns a char *, and would like
> to record this value each time the function returns (I am currently
> using truss, but DTrace would be an optimal fit for what I am trying
> to do).
Assuming this is a userland process (since you are using truss),

pid$target::funcname:return /arg1 != NULL/ { printf("%s",
copyinstr(arg1)); }
> Also, are the inet_* routines by any chance available to
> manipulate IPv4 and IPv6 addresses? I checked the Variables and
> Actions and Subroutines sections in the DTrace users guide, but
wasn''t
> able to find an answer to my questions.
I don''t believe so;  which ones in particular would you like to have?

Cheers,
- jonathan
> Thanks,
> - Ryan
> --
> UNIX Administrator
> http://daemons.net/~matty
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
-- 
Jonathan Adams, Solaris Kernel Development

On Wed, 9 Nov 2005, Adam Leventhal wrote:
> Check out the chapters on the fbt and pid providers. The return value for
> non-tail call returns will be in arg1.
Thanks Adam and Jonathon -- arg1 worked like a charm!!!! As a follow up, 
is there a way to get the value of a global variable with the pid 
provider? I am DTracing a function that accesses several global variables,
and would like to be able to print the global varaibles value each time 
they''re accessed.

Thanks again for the awesome feedback!
- Ryan

--
UNIX Administrator
http://daemons.net/~matty
> Adam
>
> On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
>>
>> Howdy,
>>
>> Is there by any chance a way to capture the value returned from a
function?
>> I
>> am tracing a specific function that returns a char *, and would like to
>> record
>> this value each time the function returns (I am currently using truss,
but
>> DTrace
>> would be an optimal fit for what I am trying to do). Also, are the
inet_*
>> routines by any chance available to manipulate IPv4 and IPv6 addresses?
I
>> checked
>> the Variables and Actions and Subroutines sections in the DTrace users
>> guide,
>> but wasn''t able to find an answer to my questions.
>>
>> Thanks,
>> - Ryan
>> --
>> UNIX Administrator
>> http://daemons.net/~matty
>> _______________________________________________
>> dtrace-discuss mailing list
>> dtrace-discuss at opensolaris.org
>
> -- 
> Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl
>

Hey Ryan,

That''s not possible from DTrace yet. What you can do is find the
address
of the global variable using mdb and then stuff that address into your
script. It''s a bit coarse, but it should work.

Adam

On Wed, Nov 09, 2005 at 05:45:29PM -0500, Matty wrote:
> On Wed, 9 Nov 2005, Adam Leventhal wrote:
> 
> >Check out the chapters on the fbt and pid providers. The return value
for
> >non-tail call returns will be in arg1.
> 
> Thanks Adam and Jonathon -- arg1 worked like a charm!!!! As a follow up, 
> is there a way to get the value of a global variable with the pid 
> provider? I am DTracing a function that accesses several global variables,
> and would like to be able to print the global varaibles value each time 
> they''re accessed.
> 
> Thanks again for the awesome feedback!
> - Ryan
> 
> --
> UNIX Administrator
> http://daemons.net/~matty
> 
> >Adam
> >
> >On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
> >>
> >>Howdy,
> >>
> >>Is there by any chance a way to capture the value returned from a 
> >>function?
> >>I
> >>am tracing a specific function that returns a char *, and would
like to
> >>record
> >>this value each time the function returns (I am currently using
truss, but
> >>DTrace
> >>would be an optimal fit for what I am trying to do). Also, are the
inet_*
> >>routines by any chance available to manipulate IPv4 and IPv6
addresses? I
> >>checked
> >>the Variables and Actions and Subroutines sections in the DTrace
users
> >>guide,
> >>but wasn''t able to find an answer to my questions.
> >>
> >>Thanks,
> >>- Ryan
> >>--
> >>UNIX Administrator
> >>http://daemons.net/~matty
> >>_______________________________________________
> >>dtrace-discuss mailing list
> >>dtrace-discuss at opensolaris.org
> >
> >-- 
> >Adam Leventhal, Solaris Kernel Development      
http://blogs.sun.com/ahl
> >
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
-- 
Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl

Howdy,

Is it possible to use offsets from arg1 in a copyin() instruction? Given 
the following function prototype:

request_rec *ap_read_request(conn_rec *conn)

and the following format for request_rec:

struct request_rec {
     apr_pool_t *pool;
     conn_rec *connection;
     server_rec *server;
     request_rec *next;
     request_rec *prev;
     request_rec *main;
     char *the_request;
     [ ... ]
}

I would like to be able to do something like this:

pid$target::ap_read_request:return
{
    printf("Request line: %s\n",copyinstr(arg1 + 48));
}

If my math is right for a 64-bit platform, I would expect
"the_request" to
be offset 48-bytes from the address in arg1 (this is for testing purposes 
-- the correct solution will use defined types). I verified this by 
printing the addresses to a logfile:

[Thu Nov 10 16:05:30 2005] [error] [client 192.168.1.8] Address 
request_rec: 149c68 Address request_rec->hostname: 149c98

Given this information, is there a reason that this doesn''t work?:

$ cat viewreq.d

pid$target::ap_read_request:return
{
    printf("Address: %x Hostname: %s\n",arg1 ,copyinstr(arg1 + 48));
}

$ dtrace -p 3796 -s viewreqs.d

dtrace: script ''viewreqs.d'' matched 1 probe
CPU     ID                    FUNCTION:NAME
   0      4           ap_read_request:return Address: 149c68 Hostname:


dtrace: error on enabled probe ID 1 (ID 4: 
pid7006:httpd:ap_read_request:return): invalid address (0x0) in action #2 
at DIF offset 36

Thanks for any insight,
- Ryan
--
UNIX Administrator
http://daemons.net/~matty


On Wed, 9 Nov 2005, Adam Leventhal wrote:
> Check out the chapters on the fbt and pid providers. The return value for
> non-tail call returns will be in arg1.
>
> Adam
>
> On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
>>
>> Howdy,
>>
>> Is there by any chance a way to capture the value returned from a
function?
>> I
>> am tracing a specific function that returns a char *, and would like to
>> record
>> this value each time the function returns (I am currently using truss,
but
>> DTrace
>> would be an optimal fit for what I am trying to do). Also, are the
inet_*
>> routines by any chance available to manipulate IPv4 and IPv6 addresses?
I
>> checked
>> the Variables and Actions and Subroutines sections in the DTrace users
>> guide,
>> but wasn''t able to find an answer to my questions.
>>
>> Thanks,
>> - Ryan
>> --
>> UNIX Administrator
>> http://daemons.net/~matty
>> _______________________________________________
>> dtrace-discuss mailing list
>> dtrace-discuss at opensolaris.org
>
> -- 
> Adam Leventhal, Solaris Kernel Development       http://blogs.sun.com/ahl
>

On Thu, Nov 10, 2005 at 04:24:57PM -0500, Matty wrote:
> 
> Howdy,
> 
> Is it possible to use offsets from arg1 in a copyin() instruction? Given 
> the following function prototype:
> 
> request_rec *ap_read_request(conn_rec *conn)
> 
> and the following format for request_rec:
> 
> struct request_rec {
>     apr_pool_t *pool;
>     conn_rec *connection;
>     server_rec *server;
>     request_rec *next;
>     request_rec *prev;
>     request_rec *main;
>     char *the_request;
>     [ ... ]
> }
> 
> I would like to be able to do something like this:
> 
> pid$target::ap_read_request:return
> {
>    printf("Request line: %s\n",copyinstr(arg1 + 48));
> }
Nope;  you need to copy it in twice; once for the pointer, then for the string:

pid$target::ap_read_request:return
{
	this->ptr = (uintptr_t *)copyin(arg1 + 48, sizeof (uintptr_t));
	printf("Request line: %s\n", copyinstr(*this->ptr));
}

(uintptr_t is a pointer-sized integer)

Cheers,
- jonathan
> 
> If my math is right for a 64-bit platform, I would expect
"the_request" to
> be offset 48-bytes from the address in arg1 (this is for testing purposes 
> -- the correct solution will use defined types). I verified this by 
> printing the addresses to a logfile:
> 
> [Thu Nov 10 16:05:30 2005] [error] [client 192.168.1.8] Address 
> request_rec: 149c68 Address request_rec->hostname: 149c98
> 
> Given this information, is there a reason that this doesn''t work?:
> 
> $ cat viewreq.d
> 
> pid$target::ap_read_request:return
> {
>    printf("Address: %x Hostname: %s\n",arg1 ,copyinstr(arg1 +
48));
> }
> 
> $ dtrace -p 3796 -s viewreqs.d
> 
> dtrace: script ''viewreqs.d'' matched 1 probe
> CPU     ID                    FUNCTION:NAME
>   0      4           ap_read_request:return Address: 149c68 Hostname:
> 
> 
> dtrace: error on enabled probe ID 1 (ID 4: 
> pid7006:httpd:ap_read_request:return): invalid address (0x0) in action #2 
> at DIF offset 36
> 
> Thanks for any insight,
> - Ryan
> --
> UNIX Administrator
> http://daemons.net/~matty
> 
> 
> On Wed, 9 Nov 2005, Adam Leventhal wrote:
> 
> >Check out the chapters on the fbt and pid providers. The return value
for
> >non-tail call returns will be in arg1.
> >
> >Adam
> >
> >On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
> >>
> >>Howdy,
> >>
> >>Is there by any chance a way to capture the value returned from a 
> >>function?
> >>I
> >>am tracing a specific function that returns a char *, and would
like to
> >>record
> >>this value each time the function returns (I am currently using
truss, but
> >>DTrace
> >>would be an optimal fit for what I am trying to do). Also, are the
inet_*
> >>routines by any chance available to manipulate IPv4 and IPv6
addresses? I
> >>checked
> >>the Variables and Actions and Subroutines sections in the DTrace
users
> >>guide,
> >>but wasn''t able to find an answer to my questions.
> >>
> >>Thanks,
> >>- Ryan
> >>--
> >>UNIX Administrator
> >>http://daemons.net/~matty
> >>_______________________________________________
> >>dtrace-discuss mailing list
> >>dtrace-discuss at opensolaris.org
> >
> >-- 
> >Adam Leventhal, Solaris Kernel Development      
http://blogs.sun.com/ahl
> >
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
-- 
Jonathan Adams, Solaris Kernel Development

On Thu, 10 Nov 2005, Jonatha
> Nope;  you need to copy it in twice; once for the pointer, then for the
string:
>
> pid$target::ap_read_request:return
> {
> 	this->ptr = (uintptr_t *)copyin(arg1 + 48, sizeof (uintptr_t));
> 	printf("Request line: %s\n", copyinstr(*this->ptr));
> }
>
> (uintptr_t is a pointer-sized integer)
>
Hi Jonathon,

Thanks for the pointer, but it still seems to be angry:

$ dtrace -p 7133 -s viewreqs.d
dtrace: script ''viewreqs.d'' matched 1 probe
dtrace: error on enabled probe ID 1 (ID 36860:
pid7133:httpd:ap_read_request:return):
invalid address (0x0) in action #2 at DIF offset 20

dtrace: error on enabled probe ID 1 (ID 36860:
pid7133:httpd:ap_read_request:return):
invalid address (0x0) in action #2 at DIF offset 20

Does the error refer to the address of the pointer, or the address of the 
first byte of the item the pointer points to?

Thanks for your help!
- Ryan
--
UNIX Administrator
http://daemons.net/~matty

> Cheers,
> - jonathan
>
>>
>> If my math is right for a 64-bit platform, I would expect
"the_request" to
>> be offset 48-bytes from the address in arg1 (this is for testing
purposes
>> -- the correct solution will use defined types). I verified this by
>> printing the addresses to a logfile:
>>
>> [Thu Nov 10 16:05:30 2005] [error] [client 192.168.1.8] Address
>> request_rec: 149c68 Address request_rec->hostname: 149c98
>>
>> Given this information, is there a reason that this doesn''t
work?:
>>
>> $ cat viewreq.d
>>
>> pid$target::ap_read_request:return
>> {
>>    printf("Address: %x Hostname: %s\n",arg1 ,copyinstr(arg1 +
48));
>> }
>>
>> $ dtrace -p 3796 -s viewreqs.d
>>
>> dtrace: script ''viewreqs.d'' matched 1 probe
>> CPU     ID                    FUNCTION:NAME
>>   0      4           ap_read_request:return Address: 149c68 Hostname:
>>
>>
>> dtrace: error on enabled probe ID 1 (ID 4:
>> pid7006:httpd:ap_read_request:return): invalid address (0x0) in action
#2
>> at DIF offset 36
>>
>> Thanks for any insight,
>> - Ryan
>> --
>> UNIX Administrator
>> http://daemons.net/~matty
>>
>>
>> On Wed, 9 Nov 2005, Adam Leventhal wrote:
>>
>>> Check out the chapters on the fbt and pid providers. The return
value for
>>> non-tail call returns will be in arg1.
>>>
>>> Adam
>>>
>>> On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
>>>>
>>>> Howdy,
>>>>
>>>> Is there by any chance a way to capture the value returned from
a
>>>> function?
>>>> I
>>>> am tracing a specific function that returns a char *, and would
like to
>>>> record
>>>> this value each time the function returns (I am currently using
truss, but
>>>> DTrace
>>>> would be an optimal fit for what I am trying to do). Also, are
the inet_*
>>>> routines by any chance available to manipulate IPv4 and IPv6
addresses? I
>>>> checked
>>>> the Variables and Actions and Subroutines sections in the
DTrace users
>>>> guide,
>>>> but wasn''t able to find an answer to my questions.
>>>>
>>>> Thanks,
>>>> - Ryan
>>>> --
>>>> UNIX Administrator
>>>> http://daemons.net/~matty
>>>> _______________________________________________
>>>> dtrace-discuss mailing list
>>>> dtrace-discuss at opensolaris.org
>>>
>>> --
>>> Adam Leventhal, Solaris Kernel Development      
http://blogs.sun.com/ahl
>>>
>> _______________________________________________
>> dtrace-discuss mailing list
>> dtrace-discuss at opensolaris.org
>
> -- 
> Jonathan Adams, Solaris Kernel Development
>

On Thu, Nov 10, 2005 at 04:47:46PM -0500, Matty wrote:
> 
> On Thu, 10 Nov 2005, Jonatha
> >Nope;  you need to copy it in twice; once for the pointer, then for the
> >string:
> >
> >pid$target::ap_read_request:return
> >{
> >	this->ptr = (uintptr_t *)copyin(arg1 + 48, sizeof (uintptr_t));
> >	printf("Request line: %s\n", copyinstr(*this->ptr));
> >}
> >
> >(uintptr_t is a pointer-sized integer)
> >
> 
> Hi Jonathon,
> 
> Thanks for the pointer, but it still seems to be angry:
> 
> $ dtrace -p 7133 -s viewreqs.d
> dtrace: script ''viewreqs.d'' matched 1 probe
> dtrace: error on enabled probe ID 1 (ID 36860: 
> pid7133:httpd:ap_read_request:return):
> invalid address (0x0) in action #2 at DIF offset 20
> 
> dtrace: error on enabled probe ID 1 (ID 36860: 
> pid7133:httpd:ap_read_request:return):
> invalid address (0x0) in action #2 at DIF offset 20
> 
> Does the error refer to the address of the pointer, or the address of the 
> first byte of the item the pointer points to?
If you''re running this on sparc, it''s hard to tell;  the
"invalid address"
is rounded down to the nearest page.  On Intel, you get the exact address.

To find out what pointer is NULL, you could do:


pid$target::ap_read_request:return
{
	printf("arg1 == %p\n", arg1);
}

pid$target::ap_read_request:return
/(this->ptr = *(uintptr_t *)copyin(arg1 + 48, sizeof (uintptr_t))) != NULL/
{
	printf("Request line: %s\n", copyinstr(this->ptr));
}

pid$target::ap_read_request:return
/this->ptr == NULL/
{
	printf("request line is NULL\n");
}

Cheers,
- jonathan
> >
> >>
> >>If my math is right for a 64-bit platform, I would expect
"the_request" to
> >>be offset 48-bytes from the address in arg1 (this is for testing
purposes
> >>-- the correct solution will use defined types). I verified this by
> >>printing the addresses to a logfile:
> >>
> >>[Thu Nov 10 16:05:30 2005] [error] [client 192.168.1.8] Address
> >>request_rec: 149c68 Address request_rec->hostname: 149c98
> >>
> >>Given this information, is there a reason that this
doesn''t work?:
> >>
> >>$ cat viewreq.d
> >>
> >>pid$target::ap_read_request:return
> >>{
> >>   printf("Address: %x Hostname: %s\n",arg1
,copyinstr(arg1 + 48));
> >>}
> >>
> >>$ dtrace -p 3796 -s viewreqs.d
> >>
> >>dtrace: script ''viewreqs.d'' matched 1 probe
> >>CPU     ID                    FUNCTION:NAME
> >>  0      4           ap_read_request:return Address: 149c68
Hostname:
> >>
> >>
> >>dtrace: error on enabled probe ID 1 (ID 4:
> >>pid7006:httpd:ap_read_request:return): invalid address (0x0) in
action #2
> >>at DIF offset 36
> >>
> >>Thanks for any insight,
> >>- Ryan
> >>--
> >>UNIX Administrator
> >>http://daemons.net/~matty
> >>
> >>
> >>On Wed, 9 Nov 2005, Adam Leventhal wrote:
> >>
> >>>Check out the chapters on the fbt and pid providers. The return
value for
> >>>non-tail call returns will be in arg1.
> >>>
> >>>Adam
> >>>
> >>>On Wed, Nov 09, 2005 at 04:03:37PM -0500, Matty wrote:
> >>>>
> >>>>Howdy,
> >>>>
> >>>>Is there by any chance a way to capture the value returned
from a
> >>>>function?
> >>>>I
> >>>>am tracing a specific function that returns a char *, and
would like to
> >>>>record
> >>>>this value each time the function returns (I am currently
using truss,
> >>>>but
> >>>>DTrace
> >>>>would be an optimal fit for what I am trying to do). Also,
are the
> >>>>inet_*
> >>>>routines by any chance available to manipulate IPv4 and
IPv6 addresses?
> >>>>I
> >>>>checked
> >>>>the Variables and Actions and Subroutines sections in the
DTrace users
> >>>>guide,
> >>>>but wasn''t able to find an answer to my questions.
> >>>>
> >>>>Thanks,
> >>>>- Ryan
> >>>>--
> >>>>UNIX Administrator
> >>>>http://daemons.net/~matty
> >>>>_______________________________________________
> >>>>dtrace-discuss mailing list
> >>>>dtrace-discuss at opensolaris.org
> >>>
> >>>--
> >>>Adam Leventhal, Solaris Kernel Development      
http://blogs.sun.com/ahl
> >>>
> >>_______________________________________________
> >>dtrace-discuss mailing list
> >>dtrace-discuss at opensolaris.org
> >
> >-- 
> >Jonathan Adams, Solaris Kernel Development
> >
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
-- 
Jonathan Adams, Solaris Kernel Development