x86/tls: Don't validate lm in set_thread_area() after all
authorAndy Lutomirski <luto@amacapital.net>
Wed, 17 Dec 2014 22:48:30 +0000 (14:48 -0800)
committerIngo Molnar <mingo@kernel.org>
Thu, 18 Dec 2014 11:12:26 +0000 (12:12 +0100)
commit3fb2f4237bb452eb4e98f6a5dbd5a445b4fed9d0
tree497a40782e96617b59efa507aa736bd09c037452
parent0e58af4e1d2166e9e33375a0f121e4867010d4f8
x86/tls: Don't validate lm in set_thread_area() after all

It turns out that there's a lurking ABI issue.  GCC, when
compiling this in a 32-bit program:

struct user_desc desc = {
.entry_number    = idx,
.base_addr       = base,
.limit           = 0xfffff,
.seg_32bit       = 1,
.contents        = 0, /* Data, grow-up */
.read_exec_only  = 0,
.limit_in_pages  = 1,
.seg_not_present = 0,
.useable         = 0,
};

will leave .lm uninitialized.  This means that anything in the
kernel that reads user_desc.lm for 32-bit tasks is unreliable.

Revert the .lm check in set_thread_area().  The value never did
anything in the first place.

Fixes: 0e58af4e1d21 ("x86/tls: Disallow unusual TLS segments")
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org # Only if 0e58af4e1d21 is backported
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/d7875b60e28c512f6a6fc0baf5714d58e7eaadbb.1418856405.git.luto@amacapital.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/include/uapi/asm/ldt.h
arch/x86/kernel/tls.c