Re: [PATCH 1/4] um: Fix _FORTIFY_SOURCE=2 support for kernel modules

[Richard Weinberger <richard@nod.at>]

Am Mittwoch 29 Juni 2011, 18:42:43 schrieb Geert Uytterhoeven:
> On Wed, Jun 29, 2011 at 17:42, Richard Weinberger <richard@nod.at> wrote:
> > Am Mittwoch 29 Juni 2011, 17:37:54 schrieb Vitaliy Ivanov:
> >> On Tue, Jun 28, 2011 at 1:15 AM, Richard Weinberger <richard@nod.at> wrote:
> >> > When UML is compiled with _FORTIFY_SOURCE we have to
> >> > export all _chk() functions which are used in modules.
> >> > For now it's only the case for __sprintf_chk().
> >> > 
> >> > Tested-by: Florian Fainelli <florian@openwrt.org>
> >> > Reported-by: Florian Fainelli <florian@openwrt.org>
> >> > Signed-off-by: Richard Weinberger <richard@nod.at>
> >> > ---
> >> >  arch/um/os-Linux/user_syms.c |    5 +++++
> >> >  1 files changed, 5 insertions(+), 0 deletions(-)
> >> > 
> >> > diff --git a/arch/um/os-Linux/user_syms.c
> >> > b/arch/um/os-Linux/user_syms.c index 05f5ea8..45ffe46 100644
> >> > --- a/arch/um/os-Linux/user_syms.c
> >> > +++ b/arch/um/os-Linux/user_syms.c
> >> > @@ -113,3 +113,8 @@ EXPORT_SYMBOL(__stack_smash_handler);
> >> > 
> >> >  extern long __guard __attribute__((weak));
> >> >  EXPORT_SYMBOL(__guard);
> >> > +
> >> > +#ifdef _FORTIFY_SOURCE
> >> > +extern int __sprintf_chk(char *str, int flag, size_t strlen, const
> >> > char *format); +EXPORT_SYMBOL(__sprintf_chk);
> >> > +#endif
> >> > --
> >> 
> >> Can you please clarify why it's needed? I use FORTIFY_SOURCES on
> >> Linus' tree UML on Ubuntu w/ no problems and w/o this patch...
> > 
> > The problem appears only when a UML kernel module is using sprintf().
> > E.g: CONFIG_UML_WATCHDOG.
> > Then loading the module will fail because of the missing symbol
> > __sprintf_chk().
> 
> What about changing harddog_user.c to not use sprintf(), but snprintf()?

Then we still have to export __snprintf_chk().

Anyway, replacing all the ugly sprintf() by snprintf() is already on my TODO list.
harddog_user.c is not the only user.

> Besides, the code does look fishy to me:
> 
>     char pid_buf[sizeof("nnnnn\0")];
>     sprintf(pid_buf, "%d", os_getpid());
> 
> os_getpid() returns int:
> 
>     int os_getpid(void)
>     {
>             return syscall(__NR_getpid);
>     }
> 
> but sys_getpid() returns long.
> 
> So we truncate a (possibly 64-bit number) to 32-bit, and format it
> in a buffer that has space for 5 chars only...

True, the code assumes quite short pid numbers.
Thanks for pointing this out!

The sizeof("nnnnn\0") notation is very error-prone, UML is riddled with it.
It's time to get rid of it.

Thanks,
//richard