Eugene:
> Can Chris or someone please tell me what is the reason behind leaving
> __kernel_uid_t == short on x86 architecture, and adding
> __kernel_uid32_t? To me, it sounds more reasonable to rather make
> __kernel_uid_t == int and add __kernel_uid16_t == short for binary
> compatibility things.
That's what I did originally. Unfortunately, it breaks both libc5 and
glibc. In the case of libc5, every time you compile a program, you are
including header files from the kernel source tree (normally installed in
/usr/include/linux and /usr/include/asm). If we were to change the types
of uid_t and gid_t directly in the kernel, then anyone trying to compile a
program would be using the wrong sizes and would end up with a miscompiled
program.
This is why we needed the following ugly construction in
include/linux/types.h:
#ifdef __KERNEL__
typedef __kernel_uid32_t uid_t;
typedef __kernel_gid32_t gid_t;
#else
typedef __kernel_uid_t uid_t;
typedef __kernel_gid_t gid_t;
#endif /* __KERNEL__ */
> As a result of current status of things, 32bit uid support does *not*
> become visible to the userspace when you rebuild glibc with the new
> kernel source. To make it take effect, you also need to modify glibc
> source, and in a rather illogical way.
This is quite deliberate. If we were to change the type of __kernel_uid_t,
and you recompiled glibc, you would *not* get a working glibc, you would
get a glibc that worked only partially and caused various programs to
crash. Even worse, if you tried to use this glibc on a system with an
older kernel, every program would crash.
Old versions of glibc simply do not expect the type of __kernel_uid_t to
change.
Instead, we added a new type, __kernel_uid32_t, and new system calls
(sys_setuid32, sys_chown32, etc.). Old versions of glibc do not know about
these, and will continue to work properly if compiled against newer kernel
headers.
Future versions of glibc will check to see if __kernel_uid32_t is defined
when they are being compiled, and if so, will include everything that is
necessary to make use of 32-bit UIDs on new kernels, and still work
properly on older ones.
I'll post a message to the kernel list when I have a patch for glibc 2.1
for testing. The older patches on my web site do not yet work for the
32-bit UID support now in Linux 2.3.
Thanks,
Chris Wing
wingc@engin.umich.edu
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Jan 15 2000 - 21:00:24 EST