dtrace discuss - Dec 2006 - capturing arguments to strcmp?

I''m trying to figure out why devfsadm is crashing in strcmp (actually,
I suspect it''s related to lofi breaking and this condition surviving
several reboots, but can''t track it down any further).

omega# devfsadm -C
zsh: segmentation fault (core dumped)  devfsadm -C
omega# pstack core
core ''core'' of 616:     devfsadm -C
 ff130910 strcmp   (10, feea2210, 503f0, 0, 19290, 5e350) + 170
 00017048 ???????? (ffbff850, 5e350, 36000, 5e328, 0, 503f0)
 00016eb0 ???????? (5e2a0, 5e350, ffbff9b8, ff377e88, 5e380, ff398680)
 ff3782fc walk_one_minor_list (0, 0, 10, ffbff9b8, 16dd4, 5e350) + b0
 ff3783f4 di_walk_minor (5dba8, 0, 10, ffbff9b8, 16dd4, 36000) + d0
 000151b0 ???????? (ffbff9b8, 37fc8, 0, 23804, 5e050, 1ae74)
 0001549c ???????? (0, 20df2f, 37838, 37fc8, dc00, df2f)
 00013f00 main     (2, ffbffb24, 0, 22438, 0, 36000) + 340
 00013ba8 _start   (0, 0, 0, 0, 0, 0) + 108

So I tried the following:

omega# cat strcmp.d
#!/usr/sbin/dtrace -s
::strcmp:entry
/execname == "devfsadm"/
{
        self->first = arg0;
        self->second = arg1;
}

::strcmp:return
/execname == "devfsadm"/
{
        printf("strcmp(''%s'',
''%s'');\n", copyinstr(self->first),
copyinstr(self->second));
        self->first = 0;
        self->second = 0;
}

But when I try running this, I get:

dtrace: script ''./strcmp.d'' matched 2 probes
dtrace: error on enabled probe ID 2 (ID 28794: fbt:genunix:strcmp:return):
invalid address (0x600031d6000) in action #1 at DIF offset 28
dtrace: error on enabled probe ID 2 (ID 28794: fbt:genunix:strcmp:return):
invalid address (0x600031d6000) in action #1 at DIF offset 28
dtrace: error on enabled probe ID 2 (ID 28794: fbt:genunix:strcmp:return):
invalid address (0x2a100678000) in action #1 at DIF offset 28
dtrace: error on enabled probe ID 2 (ID 28794: fbt:genunix:strcmp:return):
invalid address (0x2a100678000) in action #1 at DIF offset 28
[...]

Why is this happening?  Of course I''m not going to catch the crashing
invocation with the above, but I''m just trying to get a start at
looking into the function''s arguments.

--Nicholas
 
 
This message posted from opensolaris.org

>I''m trying to figure out why devfsadm is crashing in strcmp
(actually,
>I suspect it''s related to lofi breaking and this condition
surviving
>several reboots, but can''t track it down any further).
>omega# devfsadm -C
>zsh: segmentation fault (core dumped)  devfsadm -C
This is outside the kernel.
>omega# pstack core
>core ''core'' of 616:     devfsadm -C
> ff130910 strcmp   (10, feea2210, 503f0, 0, 19290, 5e350) + 170
> 00017048 ???????? (ffbff850, 5e350, 36000, 5e328, 0, 503f0)
> 00016eb0 ???????? (5e2a0, 5e350, ffbff9b8, ff377e88, 5e380, ff398680)
> ff3782fc walk_one_minor_list (0, 0, 10, ffbff9b8, 16dd4, 5e350) + b0
> ff3783f4 di_walk_minor (5dba8, 0, 10, ffbff9b8, 16dd4, 36000) + d0
> 000151b0 ???????? (ffbff9b8, 37fc8, 0, 23804, 5e050, 1ae74)
> 0001549c ???????? (0, 20df2f, 37838, 37fc8, dc00, df2f)
> 00013f00 main     (2, ffbffb24, 0, 22438, 0, 36000) + 340
> 00013ba8 _start   (0, 0, 0, 0, 0, 0) + 108
>
>So I tried the following:
>
>omega# cat strcmp.d
>#!/usr/sbin/dtrace -s
>::strcmp:entry
>/execname == "devfsadm"/
>{
>        self->first = arg0;
>        self->second = arg1;
This is inside the kernel.
>Why is this happening?  Of course I''m not going to catch the
crashing
>invocation with the above, but I''m just trying to get a start at
>looking into the function''s arguments.
You''re catching the wrong strcmp; and the one you are catching is
inside
the kernel and the arguments to it don''t need to be copied with
copyinstr().

Starting with "truss" or a debugger on the core looks more helpful.
(Or use the pid provider)

Casper