[prev in list] [next in list] [prev in thread] [next in thread] 

List:       opensolaris-dtrace-discuss
Subject:    Re: [dtrace-discuss] 32-bit/64-bit weirdness
From:       Jon Haslam <Jonathan.Haslam () Sun ! COM>
Date:       2009-12-03 11:37:25
Message-ID: 4B17A2F5.4000400 () sun ! com
[Download RAW message or body]

Hi Drew,

> I'm wondering if this is expected behavior, or a bug.

Expected but not all that obvious really. See below.

> Given this D:
> 
> # cat smb_logon.d
> pid$target::smb_logon:entry {
>  add=&((struct netr_client *)arg0)->username;
>  printf("%p",add);
>  exit();
> }
> 
> and this defintion:
> 
>  typedef struct netr_client {
>          uint16_t logon_level;
>          char *username;        /* request username */
>          char *domain;        /* request domain */
>          char *e_username;    /* effective username */
>          char *e_domain;        /* effective domain */
>  [...]
>  } netr_client_t;
> 
> On a 64-bit x86 system (SunOS smar-x 5.11 snv_127 i86pc):
> 
> # dtrace -s smb_logon.d -p `pgrep smbd`
> dtrace: script 'smb_logon.d' matched 1 probe
> CPU     ID                    FUNCTION:NAME
>   2  66800                  smb_logon:entry 810aef8
> 
> This works, but since it's a 32-bit application the address is 'wrong'. 
> It should be 810aef4.
> 
> # dtrace -32 -s smb_logon.d -p `pgrep smbd`
> dtrace: failed to compile script smb_logon.d: line 2: operator -> cannot 
> be applied to a forward declaration: no struct netr_client definition is 
> available
> 
> So... where is dtrace looking for the ctf info? It can't be in the 
> binary, as mdb can find the definition:
> 
> # mdb -p `pgrep smbd`

We currently don't employ any ctf that is available in userland.
The definition you are finding is most likely from the 'smbsrv'
kernel module.

The reason that it "works" on 64-bit but not on 32-bit is
that we don't support mixed-mode. This means that if we
have a 32-bit data model in userland and a 64-bit data model
in kernel we won't read any CTF from the kernel. In that
case we no longer have the netr_client_t definition
available and we can't carry on.

What you need to do is to #include the the header containing
the definition of netr_client_t and invoke the pre-processor
with the '-C' flag.

Jon.
_______________________________________________
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org
[prev in list] [next in list] [prev in thread] [next in thread]