[PATCH resend] x86, tls: Don't validate lm in set_thread_area after all

[PATCH resend] x86, tls: Don't validate lm in set_thread_area after all

[Andy Lutomirski]

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
Cc: stable@vger.kernel.org # Only if 0e58af4e1d21 is backported
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---

I think that this got eaten by gmail's SMTP server.  It showed up in
my inbox, but it never made it to lkml.

 arch/x86/include/uapi/asm/ldt.h | 7 +++++++
 arch/x86/kernel/tls.c           | 6 ------
 2 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/arch/x86/include/uapi/asm/ldt.h b/arch/x86/include/uapi/asm/ldt.h
index 46727eb37bfe..6e1aaf73852a 100644
--- a/arch/x86/include/uapi/asm/ldt.h
+++ b/arch/x86/include/uapi/asm/ldt.h
@@ -28,6 +28,13 @@ struct user_desc {
 	unsigned int  seg_not_present:1;
 	unsigned int  useable:1;
 #ifdef __x86_64__
+	/*
+	 * Because this bit is not present in 32-bit user code, user
+	 * programs can pass uninitialized values here.  Therefore, in
+	 * any context in which a user_desc comes from a 32-bit program,
+	 * the kernel must act as though lm == 0, regardless of the
+	 * actual value.
+	 */
 	unsigned int  lm:1;
 #endif
 };
diff --git a/arch/x86/kernel/tls.c b/arch/x86/kernel/tls.c
index 3e551eee87b9..4e942f31b1a7 100644
--- a/arch/x86/kernel/tls.c
+++ b/arch/x86/kernel/tls.c
@@ -55,12 +55,6 @@ static bool tls_desc_okay(const struct user_desc *info)
 	if (info->seg_not_present)
 		return false;
 
-#ifdef CONFIG_X86_64
-	/* The L bit makes no sense for data. */
-	if (info->lm)
-		return false;
-#endif
-
 	return true;
 }
 
-- 
2.1.0

[tip:x86/urgent] x86/tls: Don't validate lm in set_thread_area() after all

[tip-bot for Andy Lutomirski]

Commit-ID:  3fb2f4237bb452eb4e98f6a5dbd5a445b4fed9d0
Gitweb:     http://git.kernel.org/tip/3fb2f4237bb452eb4e98f6a5dbd5a445b4fed9d0
Author:     Andy Lutomirski <luto@amacapital.net>
AuthorDate: Wed, 17 Dec 2014 14:48:30 -0800
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Thu, 18 Dec 2014 12:12:26 +0100

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 | 7 +++++++
 arch/x86/kernel/tls.c           | 6 ------
 2 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/arch/x86/include/uapi/asm/ldt.h b/arch/x86/include/uapi/asm/ldt.h
index 46727eb..6e1aaf7 100644
--- a/arch/x86/include/uapi/asm/ldt.h
+++ b/arch/x86/include/uapi/asm/ldt.h
@@ -28,6 +28,13 @@ struct user_desc {
 	unsigned int  seg_not_present:1;
 	unsigned int  useable:1;
 #ifdef __x86_64__
+	/*
+	 * Because this bit is not present in 32-bit user code, user
+	 * programs can pass uninitialized values here.  Therefore, in
+	 * any context in which a user_desc comes from a 32-bit program,
+	 * the kernel must act as though lm == 0, regardless of the
+	 * actual value.
+	 */
 	unsigned int  lm:1;
 #endif
 };
diff --git a/arch/x86/kernel/tls.c b/arch/x86/kernel/tls.c
index 3e551ee..4e942f3 100644
--- a/arch/x86/kernel/tls.c
+++ b/arch/x86/kernel/tls.c
@@ -55,12 +55,6 @@ static bool tls_desc_okay(const struct user_desc *info)
 	if (info->seg_not_present)
 		return false;
 
-#ifdef CONFIG_X86_64
-	/* The L bit makes no sense for data. */
-	if (info->lm)
-		return false;
-#endif
-
 	return true;
 }
 

Re: [tip:x86/urgent] x86/tls: Don't validate lm in set_thread_area() after all

[H. Peter Anvin]

On 12/18/2014 03:16 AM, tip-bot for Andy Lutomirski wrote:
> Commit-ID:  3fb2f4237bb452eb4e98f6a5dbd5a445b4fed9d0
> Gitweb:     http://git.kernel.org/tip/3fb2f4237bb452eb4e98f6a5dbd5a445b4fed9d0
> Author:     Andy Lutomirski <luto@amacapital.net>
> AuthorDate: Wed, 17 Dec 2014 14:48:30 -0800
> Committer:  Ingo Molnar <mingo@kernel.org>
> CommitDate: Thu, 18 Dec 2014 12:12:26 +0100
> 
> 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.
> 

No, it won't.  However, if you initialize this dynamically field by
field rather than as an initializer, then you are correct.

	-hpa

Re: [tip:x86/urgent] x86/tls: Don't validate lm in set_thread_area() after all

[Andy Lutomirski]

On Thu, Dec 18, 2014 at 8:59 AM, H. Peter Anvin <hpa@zytor.com> wrote:
> On 12/18/2014 03:16 AM, tip-bot for Andy Lutomirski wrote:
>> Commit-ID:  3fb2f4237bb452eb4e98f6a5dbd5a445b4fed9d0
>> Gitweb:     http://git.kernel.org/tip/3fb2f4237bb452eb4e98f6a5dbd5a445b4fed9d0
>> Author:     Andy Lutomirski <luto@amacapital.net>
>> AuthorDate: Wed, 17 Dec 2014 14:48:30 -0800
>> Committer:  Ingo Molnar <mingo@kernel.org>
>> CommitDate: Thu, 18 Dec 2014 12:12:26 +0100
>>
>> 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.
>>
>
> No, it won't.  However, if you initialize this dynamically field by
> field rather than as an initializer, then you are correct.
>

I tried the code above in function scope.

--Andy

>         -hpa
>



-- 
Andy Lutomirski
AMA Capital Management, LLC

Re: [tip:x86/urgent] x86/tls: Don't validate lm in set_thread_area() after all

[Linus Torvalds]

On Thu, Dec 18, 2014 at 8:59 AM, H. Peter Anvin <hpa@zytor.com> wrote:
>>
>> will leave .lm uninitialized.  This means that anything in the
>> kernel that reads user_desc.lm for 32-bit tasks is unreliable.
>
> No, it won't.  However, if you initialize this dynamically field by
> field rather than as an initializer, then you are correct.

Actually, even with a full initializer, unnamed parts of a structure
(so padding bytes between things, but for bitfields also unnamed
alignment fields etc) are basically "all bets are off". They are *not*
guaranteed to be initialized to zero.

So if you have a structure like

   struct {
       unsigned int a:5;
       unsigned int b;
   } x = { .a = 0, .b = 0 };

afaik the compiler is not guaranteed to initialize the left-over bits
in the first word. Because they simply don't "exist" as far as the C
language is concerned.

On the other hand, if you do

   struct {
        unsigned int a:5, unused:27;
        unsigned int b;
   } x = { .a = 0, .b = 0 };

then the 'unused' bits are guaranteed to be initialized to zero.

(Static allocations in the BSS are obviously zeroed for other reasons,
so there are no "left-over" bits there to worry about,. So in practice
the above is only about dynamic initializers).

                         Linus