> > The bug itself has to do with the MSVC++ 6 version of strlen being
> > unable to handle empty string. I know this sounds unlikely, but if
> > I supply a NULL pointer to it, it generates an access violation in
> > debug mode and when I just run the same executable outside the
> > debugger it hangs.
>
> ...whereas on many UNIX systems (probably most of them, at this point)
> it just generates a segmentation violation, period, regardless of
> whether you're running it inside a debugger or not.
Sure does. If I read, on Solaris 2.5.1/SPARC with an Ethereal lacking
your change, the capture you sent me, it blows up in:
(gdb) where
#0 0xef2a3ec0 in strlen () from /usr/lib/libc.so.1
#1 0xa265c in strings (p=0xefffdb70, tmp=0x0) at snprintf.c:328
#2 0xa3804 in vsnprintf (string=0x1 <Address 0x1 out of bounds>,
length=4026522480, format=0x114590 "%s %s", args=0xefffdc84)
at snprintf.c:603
#3 0x9541c in col_add_fstr (fd=0x278478, el=30, format=0x114590 "%s %s")
at packet.c:694
#4 0x789f4 in dissect_transact_smb (pd=0x275e10 "\002?\202\213-g", offset=90,
fd=0x278478, parent=0x25a774, tree=0x25abac, si={tid = 2051, uid = 2051,
mid = 35904, pid = 49792, conversation = 0x27c578,
request_val = 0x27d588, unicode = 1}, max_data=1456, SMB_offset=58,
errcode=0, dirn=0) at packet-smb.c:9326
...
and your change fixes the problem. It'll probably blow up on a number
of other platforms as well.
