Re: RFC: introduce struct ksymbol

[Frederic Weisbecker <fweisbec@gmail.com>]

On Wed, Apr 15, 2009 at 08:31:06AM +0200, Ingo Molnar wrote:
> 
> * Pekka Enberg <penberg@cs.helsinki.fi> wrote:
> 
> > Hi Ingo,
> > 
> > On Wed, Apr 15, 2009 at 8:58 AM, Ingo Molnar <mingo@elte.hu> wrote:
> > >
> > > (Sam and Rusty Cc:-ed)
> > >
> > > * Joe Perches <joe@perches.com> wrote:
> > >
> > >> On Wed, 2009-04-15 at 02:00 +0200, Frederic Weisbecker wrote:
> > >> > arch/blackfin/kernel/traps.c:   symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf);
> > >> > arch/powerpc/xmon/xmon.c:               name = kallsyms_lookup(pc, &size, &offset, NULL, tmpstr);
> > >> > arch/sh/kernel/cpu/sh5/unwind.c:        sym = kallsyms_lookup(pc, NULL, &offset, NULL, namebuf);
> > >> > arch/x86/kernel/ftrace.c:       kallsyms_lookup((unsigned long) syscall, NULL, NULL, NULL, str);
> > >> > kernel/kprobes.c:               sym = kallsyms_lookup((unsigned long)p->addr, NULL,
> > >> > kernel/lockdep.c:       return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
> > >> > kernel/trace/ftrace.c:  kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
> > >> > kernel/trace/ftrace.c:  kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
> > >> > kernel/trace/ftrace.c:  kallsyms_lookup((unsigned long)rec->ops->func, NULL, NULL, NULL, str);
> > >> > kernel/trace/ftrace.c:  kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
> > >> > kernel/trace/ftrace.c:  kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
> > >> > kernel/trace/ftrace.c:  kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
> > >> > kernel/trace/ftrace.c:  kallsyms_lookup(*ptr, NULL, NULL, NULL, str);
> > >> > kernel/trace/trace_functions.c: kallsyms_lookup(ip, NULL, NULL, NULL, str);
> > >> > kernel/trace/trace_output.c:    kallsyms_lookup(address, NULL, NULL, NULL, str);
> > >>
> > >> Perhaps a conversion from
> > >>
> > >> "char str[KSYM_SYMBOL_LEN]"
> > >> to
> > >> "struct ksymbol sym"?
> > >>
> > >> could be useful.
> > >>
> > >> There are a few places that use a hard coded length of 128
> > >> instead of KSYM_SYMBOL_LENGTH that are also converted.
> > >>
> > >> Compile tested only
> > >
> > > Why not 'struct ksym'? That name is unused right now, it is shorter
> > > and just as descriptive.
> > >
> > > Regarding the change... dunno. Sam, Rusty - what do you think?
> > >
> > > Downsides would be loss of awareness of stack footprint impact. A
> > > plain struct is easy to slap on, and it's not immediately visible
> > > that it carries 128 bytes of weight. It might also be confusing in
> > > terms of the nature of the interface - whether it's a pointery
> > > object or not.
> > >
> > > Prior use:
> > >
> > >        char str[KSYM_SYMBOL_LEN];
> > >
> > >        kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
> > >
> > > New use:
> > >
> > >        struct ksym sym;
> > >
> > >        kallsyms_lookup(rec->ip, NULL, NULL, NULL, &sym);
> > >
> > > Dunno.
> > 
> > The change makes sense to me. Passing a raw char pointer down the 
> > call-chain is a buffer overflow waiting to happen and as a matter 
> > of fact, we've had bugs in this area before. See commit 
> > 9c24624727f6d6c460e45762a408ca5f5b9b8ef2 ("KSYM_SYMBOL_LEN fixes") 
> > for an example.
> 
> OK.
> 
> Albeit i think that particular bug happened due to the ambigious 
> namingof KSYM_SYMBOL_LEN versus KSYM_NAME_LEN. Who would have 
> thought that there's a difference between the two and that there's a 
> 'ksym symbol' versus 'ksym name' distinction?
> 
> It would be more robust to have just one length (the longer one), 
> and maybe have the shorter one as __KSYM_NAME_LEN - for expert use.


IMHO, I would prefer that rather than the struct ksym because of
the obfuscation reasons you talked about previously.

Frederic.

 
> 	Ingo