dtrace discuss - Jul 2008 - size and alignment bug with short in struct?

The following looks like a bug to me:

--- C program tst.c ----------------------------------

struct s1 {
     int x;
     short a;
};
struct s2 {
     struct s1 b;
     short c;
     int d;
} s = {
     { 1, 0x0102 },
     0x0304,
     0x05060708
};

void
fun(struct s2 *p)
{
         printf("   sizeof(struct s1) = %d\n", sizeof(struct s1));
         printf("   sizeof(struct s2) = %d\n", sizeof(struct s2));
         printf("   b.a %04x, c %04x, d %08x\n\n", p->b.a, p->c,
p->d);
}

int
main()
{
         fun(&s);
         return 0;
}

--- dtrace script ------------------------------

/usr/sbin/dtrace -q -s /dev/fd/0 -c tst <<''eof''

struct s1 {
     int x;
     short a;
};
struct s2 {
     struct s1 b;
     short c;
     int d;
};

pid$target::fun:entry
{
         printf("** sizeof(struct s1) = %d\n", sizeof(struct s1));
         printf("** sizeof(struct s2) = %d\n", sizeof(struct s2));
         p = (struct s2 *)copyin(arg0, sizeof(struct s2));
         printf("** b.a %04x, c %04x, d %08x\n", p->b.a, p->c,
p->d);
}

eof

--- output -------------------------------------

    sizeof(struct s1) = 8
    sizeof(struct s2) = 16
    b.a 0102, c 0304, d 05060708

** sizeof(struct s1) = 6
** sizeof(struct s2) = 12
** b.a 0102, c 0000, d 03040000

-------------------------------------------------

Dtrace shows wrong sizes for struct s1 and s2 and consequently accesses 
the wrong locations and outputs the wrong data. After adding a short to 
the end of s1 in the dtrace script (not in the C program)

struct s1 {
     int x;
     short a;
     short dummy;
};

everything works as expected. Is this a known bug?

Another question: Is there a simple way of making dtrace read from 
stdin? Having to use "dtrace -s /dev/fd/0" in the above script is a
bit
unusual for a unix tool. Implementing something like "dtrace -s -" 
should be easy I guess?

Michael

=== Michael Mueller =================Tel. + 49 8171 63600
Fax. + 49 8171 63615
Web: http://www.michael-mueller-it.de
======================================

On Mon, Jul 21, 2008 at 11:22:50AM +0200, Michael Mueller
wrote:
> The following looks like a bug to me:
[ snip ]
> everything works as expected. Is this a known bug?
That looks like a bug. Please file it if you haven''t already. It would
be
interesting to see how offsetof() results differed in C and D.
> Another question: Is there a simple way of making dtrace read from 
> stdin? Having to use "dtrace -s /dev/fd/0" in the above script is
a bit
> unusual for a unix tool. Implementing something like "dtrace -s
-"
> should be easy I guess?
It''s not much better, but we typically do ''dtrace -s
/dev/stdin''.

- ahl

-- 
Adam Leventhal, Fishworks                     http://blogs.sun.com/ahl