2005-03-06 14:25:37

by Yoichi Yuasa

Subject: [PATCH 2.6.11-mm1] mips: more convert verify_area to access_ok

This patch converts verify_area to access_ok for include/asm-mips.

Yoichi

Signed-off-by: Yoichi Yuasa <[email protected]>

diff -urN -X dontdiff a-orig/include/asm-mips/uaccess.h a/include/asm-mips/uaccess.h
--- a-orig/include/asm-mips/uaccess.h Sat Mar 5 04:15:22 2005
+++ a/include/asm-mips/uaccess.h Sun Mar 6 15:51:02 2005
@@ -254,13 +254,11 @@
({ \
__typeof__(*(ptr)) __gu_val = 0; \
long __gu_addr; \
- long __gu_err; \
+ long __gu_err = -EFAULT; \
\
might_sleep(); \
__gu_addr = (long) (ptr); \
- __gu_err = verify_area(VERIFY_READ, (void *) __gu_addr, size); \
- \
- if (likely(!__gu_err)) { \
+ if (access_ok(VERIFY_READ, (void *) __gu_addr, size)) { \
switch (size) { \
case 1: __get_user_asm("lb", __gu_err); break; \
case 2: __get_user_asm("lh", __gu_err); break; \
@@ -348,14 +346,12 @@
({ \
__typeof__(*(ptr)) __pu_val; \
long __pu_addr; \
- long __pu_err; \
+ long __pu_err = -EFAULT; \
\
might_sleep(); \
__pu_val = (x); \
__pu_addr = (long) (ptr); \
- __pu_err = verify_area(VERIFY_WRITE, (void *) __pu_addr, size); \
- \
- if (likely(!__pu_err)) { \
+ if (access_ok(VERIFY_WRITE, (void *) __pu_addr, size)) { \
switch (size) { \
case 1: __put_user_asm("sb", __pu_val); break; \
case 2: __put_user_asm("sh", __pu_val); break; \

2005-03-07 00:17:16

by Jesper Juhl

Subject: Re: [PATCH 2.6.11-mm1] mips: more convert verify_area to access_ok

On Sun, 6 Mar 2005, Yoichi Yuasa wrote:

> This patch converts verify_area to access_ok for include/asm-mips.
>
Yeah, that's one of the few bits I had not done yet. Thank you for taking
a look at that.

I don't believe your patch is correct though. See below for what I think
is a better one.


> Yoichi
>
> Signed-off-by: Yoichi Yuasa <[email protected]>
>
> diff -urN -X dontdiff a-orig/include/asm-mips/uaccess.h a/include/asm-mips/uaccess.h
> --- a-orig/include/asm-mips/uaccess.h Sat Mar 5 04:15:22 2005
> +++ a/include/asm-mips/uaccess.h Sun Mar 6 15:51:02 2005
> @@ -254,13 +254,11 @@
> ({ \
> __typeof__(*(ptr)) __gu_val = 0; \
> long __gu_addr; \
> - long __gu_err; \
> + long __gu_err = -EFAULT; \
> \
> might_sleep(); \
> __gu_addr = (long) (ptr); \
> - __gu_err = verify_area(VERIFY_READ, (void *) __gu_addr, size); \
> - \
> - if (likely(!__gu_err)) { \
> + if (access_ok(VERIFY_READ, (void *) __gu_addr, size)) { \
> switch (size) { \
> case 1: __get_user_asm("lb", __gu_err); break; \
> case 2: __get_user_asm("lh", __gu_err); break; \

with this change, __gu_err will always be -EFAULT. With the original code
it was either -EFAULT or 0 depending on the return value from verify_area.
Same goes for the next hunk in your patch.

I believe a more correct patch would be this :

Signed-off-by: Jesper Juhl <[email protected]>

diff -up linux-2.6.11-mm1-orig/include/asm-mips/uaccess.h linux-2.6.11-mm1/include/asm-mips/uaccess.h
--- linux-2.6.11-mm1-orig/include/asm-mips/uaccess.h 2005-03-05 00:39:40.000000000 +0100
+++ linux-2.6.11-mm1/include/asm-mips/uaccess.h 2005-03-07 00:49:24.000000000 +0100
@@ -258,7 +258,8 @@ struct __large_struct { unsigned long bu
\
might_sleep(); \
__gu_addr = (long) (ptr); \
- __gu_err = verify_area(VERIFY_READ, (void *) __gu_addr, size); \
+ __gu_err = access_ok(VERIFY_READ, (void *) __gu_addr, size) \
+ ? 0 : -EFAULT; \
\
if (likely(!__gu_err)) { \
switch (size) { \
@@ -353,7 +354,8 @@ extern void __get_user_unknown(void);
might_sleep(); \
__pu_val = (x); \
__pu_addr = (long) (ptr); \
- __pu_err = verify_area(VERIFY_WRITE, (void *) __pu_addr, size); \
+ __pu_err = access_ok(VERIFY_WRITE, (void *) __pu_addr, size) \
+ ? 0 : -EFAULT; \
\
if (likely(!__pu_err)) { \
switch (size) { \




It preserves the exact behaviour of the original.


--
Jesper

2005-03-07 06:38:43

by Yoichi Yuasa

Subject: Re: [PATCH 2.6.11-mm1] mips: more convert verify_area to access_ok

On Mon, 7 Mar 2005 00:55:30 +0100 (CET)
Jesper Juhl <[email protected]> wrote:

> On Sun, 6 Mar 2005, Yoichi Yuasa wrote:
>
> > This patch converts verify_area to access_ok for include/asm-mips.
> >
> Yeah, that's one of the few bits I had not done yet. Thank you for taking
> a look at that.
>
> I don't believe your patch is correct though. See below for what I think
> is a better one.

That's right.
I forgot to assign 0 to __gu_err/__pu_err after access_ok().

Thanks,

Yoichi