page_count is copied from userspace. agp_allocate_memory() tries to
check whether this number is too big, but doesn't take into account the
wrap case. Also agp_create_user_memory() doesn't check whether
alloc_size is calculated from num_agp_pages variable without overflow.
This may lead to allocation of too small buffer with following buffer
overflow.
Another problem in agp code is not addressed in the patch - kernel memory
exhaustion (AGPIOC_RESERVE and AGPIOC_ALLOCATE ioctls). It is not checked
whether requested pid is a pid of the caller (no check in agpioc_reserve_wrap()).
Each allocation is limited to 16KB, though, there is no per-process limit.
This might lead to OOM situation, which is not even solved in case of the
caller death by OOM killer - the memory is allocated for another (faked) process.
Signed-off-by: Vasiliy Kulikov <[email protected]>
---
drivers/char/agp/generic.c | 8 +++++++-
1 files changed, 7 insertions(+), 1 deletions(-)
diff --git a/drivers/char/agp/generic.c b/drivers/char/agp/generic.c
index 745e7ba..342df51 100644
--- a/drivers/char/agp/generic.c
+++ b/drivers/char/agp/generic.c
@@ -115,6 +115,9 @@ static struct agp_memory *agp_create_user_memory(unsigned long num_agp_pages)
struct agp_memory *new;
unsigned long alloc_size = num_agp_pages*sizeof(struct page *);
+ if (INT_MAX/sizeof(struct page *) < num_agp_pages)
+ return NULL;
+
new = kzalloc(sizeof(struct agp_memory), GFP_KERNEL);
if (new == NULL)
return NULL;
@@ -234,11 +237,14 @@ struct agp_memory *agp_allocate_memory(struct agp_bridge_data *bridge,
int scratch_pages;
struct agp_memory *new;
size_t i;
+ int cur_memory;
if (!bridge)
return NULL;
- if ((atomic_read(&bridge->current_memory_agp) + page_count) > bridge->max_memory_agp)
+ cur_memory = atomic_read(&bridge->current_memory_agp);
+ if ((cur_memory + page_count > bridge->max_memory_agp) ||
+ (cur_memory + page_count < page_count))
return NULL;
if (type >= AGP_USER_TYPES) {
--
1.7.0.4
> - ? ? ? if ((atomic_read(&bridge->current_memory_agp) + page_count) > bridge->max_memory_agp)
> + ? ? ? cur_memory = atomic_read(&bridge->current_memory_agp);
> + ? ? ? if ((cur_memory + page_count > bridge->max_memory_agp) ||
> + ? ? ? ? ? (cur_memory + page_count < page_count))
I should have said that current_memory_agp is already protected by
agp_fe.agp_mutex, so the patch doesn't break atomic operation
(otherwise there is also a race between the check and use of
current_memory_agp).
On Thu, Apr 14, 2011 at 08:55:19PM +0400, Vasiliy Kulikov wrote:
> page_count is copied from userspace. agp_allocate_memory() tries to
> check whether this number is too big, but doesn't take into account the
> wrap case. Also agp_create_user_memory() doesn't check whether
> alloc_size is calculated from num_agp_pages variable without overflow.
> This may lead to allocation of too small buffer with following buffer
> overflow.
>
> Another problem in agp code is not addressed in the patch - kernel memory
> exhaustion (AGPIOC_RESERVE and AGPIOC_ALLOCATE ioctls). It is not checked
> whether requested pid is a pid of the caller (no check in agpioc_reserve_wrap()).
> Each allocation is limited to 16KB, though, there is no per-process limit.
> This might lead to OOM situation, which is not even solved in case of the
> caller death by OOM killer - the memory is allocated for another (faked) process.
>
> Signed-off-by: Vasiliy Kulikov <[email protected]>
David, what ever happened to this, and the other agp bugfix from
Vasiliy?
thanks,
greg k-h
> ---
> drivers/char/agp/generic.c | 8 +++++++-
> 1 files changed, 7 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/char/agp/generic.c b/drivers/char/agp/generic.c
> index 745e7ba..342df51 100644
> --- a/drivers/char/agp/generic.c
> +++ b/drivers/char/agp/generic.c
> @@ -115,6 +115,9 @@ static struct agp_memory *agp_create_user_memory(unsigned long num_agp_pages)
> struct agp_memory *new;
> unsigned long alloc_size = num_agp_pages*sizeof(struct page *);
>
> + if (INT_MAX/sizeof(struct page *) < num_agp_pages)
> + return NULL;
> +
> new = kzalloc(sizeof(struct agp_memory), GFP_KERNEL);
> if (new == NULL)
> return NULL;
> @@ -234,11 +237,14 @@ struct agp_memory *agp_allocate_memory(struct agp_bridge_data *bridge,
> int scratch_pages;
> struct agp_memory *new;
> size_t i;
> + int cur_memory;
>
> if (!bridge)
> return NULL;
>
> - if ((atomic_read(&bridge->current_memory_agp) + page_count) > bridge->max_memory_agp)
> + cur_memory = atomic_read(&bridge->current_memory_agp);
> + if ((cur_memory + page_count > bridge->max_memory_agp) ||
> + (cur_memory + page_count < page_count))
> return NULL;
>
> if (type >= AGP_USER_TYPES) {
> --
> 1.7.0.4
>
> _______________________________________________
> Security mailing list
> [email protected]
> http://linux.kernel.org/mailman/listinfo/security
On Tue, Apr 19, 2011 at 09:21 -0700, Greg KH wrote:
> David, what ever happened to this, and the other agp bugfix from
> Vasiliy?
I got emails that both fixes went into -mm.
Thanks,
--
Vasiliy Kulikov
http://www.openwall.com - bringing security into open computing environments
On Tue, Apr 19, 2011 at 09:48:13PM +0400, Vasiliy Kulikov wrote:
> On Tue, Apr 19, 2011 at 09:21 -0700, Greg KH wrote:
> > David, what ever happened to this, and the other agp bugfix from
> > Vasiliy?
>
> I got emails that both fixes went into -mm.
That's good, but David needs to take them and push them to Linus...
> On Tue, Apr 19, 2011 at 09:48:13PM +0400, Vasiliy Kulikov wrote:
> > On Tue, Apr 19, 2011 at 09:21 -0700, Greg KH wrote:
> > > David, what ever happened to this, and the other agp bugfix from
> > > Vasiliy?
> >
> > I got emails that both fixes went into -mm.
>
> That's good, but David needs to take them and push them to Linus...
>
>
I'm ruminating, while they are security fixes, I'm hesitant to push them
ASAP, without further testing, the /dev/agpgart hasn't really been used
by much in around 5 years, so getting a userspace and hardware that can
test it is quite a lot of fun, though I suppse on the other side, nobody
will notice if it breaks anything for 2 years.
Dave.