media: change media_device_request_alloc to match media_ioctl_info
Subsystems (3)
cover media: change media_device_request_alloc to match media_ioctl_info
*** BLURB HERE ***
Hello,
This patch is to modify media_device_request_alloc() so that it can pass CFI(Control Flow Integrity) check. I would like some review comments.
media_device_request_alloc() is saved in fn of media_ioctl_info struct, which is defined as long (*fn)(struct media_device *dev, void *arg). The type of the second parameter of media_device_request_alloc() is int* now, but it is void* in fn of media_ioctl_info. We got some ABI violation here. Therefore, we would like to use void* instead of int* for the second parameter of media_device_request_alloc().
static long media_device_request_alloc(struct media_device *mdev,
int *alloc_fd);
struct media_ioctl_info {
unsigned int cmd;
unsigned short flags;
long (*fn)(struct media_device *dev, void *arg);
long (*arg_from_user)(void *karg, void __user *uarg, unsigned int cmd);
long (*arg_to_user)(void __user *uarg, void *karg, unsigned int cmd);
};Here is an example. With Android’s CFI checking, we got the following error without this change.
[ 23.502477] mtk-cam soc:camisp: sd:mtk-cam raw-0 pad:2 set format w/h/code 2328/1748/0x3007 [ 23.518690] Kernel panic - not syncing: CFI failure (target: media_device_request_alloc+0x0/0x4) [ 23.519804] CPU: 7 PID: 818 Comm: mtkv4l2_ut Tainted: G S O 5.4.39-g02c0a4858bed-dirty #45 [ 23.521553] Call trace: [ 23.521868] dump_backtrace.cfi_jt+0x0/0x4 [ 23.522389] dump_stack+0xb8/0x114 [ 23.522824] panic+0x170/0x3e0 [ 23.523215] __ubsan_handle_cfi_check_fail_abort+0x0/0x14 [ 23.523896] perf_proc_update_handler+0x0/0xcc [ 23.524460] __cfi_check+0x610cc/0x68ef0 [ 23.524959] media_device_ioctl+0x218/0x238 [ 23.525488] media_device_compat_ioctl+0x60/0x7c [ 23.526072] media_compat_ioctl+0x58/0x9c [ 23.526581] __arm64_compat_sys_ioctl+0x10c/0x434 [ 23.527176] el0_svc_common+0xb4/0x18c [ 23.527651] el0_svc_compat_handler+0x1c/0x28 [ 23.528202] el0_svc_compat+0x8/0x24 [ 23.528659] SMP: stopping secondary CPUs [ 23.529161] Kernel Offset: 0x24b8c00000 from 0xffffffc010000000 [ 23.529906] PHYS_OFFSET: 0xffffffdd00000000 [ 23.530434] CPU features: 0x00000006,2a80a238 [ 23.530983] Memory Limit: none
PPL_LOG_STORE: check once, sig value 0x27, addr 0x116000.
Frederic Chen (1):
media: mc-device.c: change media_device_request_alloc to match
media_ioctl_info
drivers/media/mc/mc-device.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
PATCH 1/1 media: mc-device.c: change media_device_request_alloc to match media_ioctl_info
From: Frederic Chen <[email protected]>
We modified the type of media_device_request_alloc()'s second parameter from int* to void* so that it can match the interface defined in struct media_ioctl_info.
--- drivers/media/mc/mc-device.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/media/mc/mc-device.c +3 -2
@@ -370,9 +370,10 @@ static long media_device_get_topology(struct media_device *mdev, void *arg)return ret; } -static long media_device_request_alloc (struct media_device *mdev, -int *alloc_fd) +static long media_device_request_alloc (struct media_device *mdev,void *arg) { +int *alloc_fd = arg; +#ifdef CONFIG_MEDIA_CONTROLLER_REQUEST_API if (!mdev->ops || !mdev->ops->req_validate || !mdev->ops->req_queue)return -ENOTTY ;
Signature · 1 line
2.18.0
Re: [PATCH V1] media: mc-device.c: change media_device_request_alloc to match media_ioctl_info
Quoted 19 lines · On 06/08/2020 09:29, [email protected] wrote:
From: Frederic Chen <[email protected]>
We modified the type of media_device_request_alloc()'s second parameter from int* to void* so that it can match the interface defined in struct
media_ioctl_info.Signed-off-by Frederic Chen <[email protected]>--- drivers/media/mc/mc-device.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/media/mc/mc-device.c +0
-0@@ -370,9 +370,10 @@ static long media_device_get_topology(struct media_device *mdev, void *arg)return ret; }
drivers/media/mc/mc-device.c static long media_device_get_topology(struct media_device *mdev, void *arg) · line 379
Quoted 10 lines
drivers/media/mc/mc-device.c +3
-2-static long media_device_request_alloc (struct media_device *mdev, -int *alloc_fd) +static long media_device_request_alloc (struct media_device *mdev,void *arg) { +int *alloc_fd = arg; +#ifdef CONFIG_MEDIA_CONTROLLER_REQUEST_API if (!mdev->ops || !mdev->ops->req_validate || !mdev->ops->req_queue)return -ENOTTY ;
This change is fine, but the reason this wasn't noticed before is the cast in the MEDIA_IOC_ARG define:
#define MEDIA_IOC_ARG(__cmd, func, fl, from_user, to_user) \
[_IOC_NR(MEDIA_IOC_##__cmd)] = { \
.cmd = MEDIA_IOC_##__cmd, \
.fn = (long (*)(struct media_device *, void *))func, \
.flags = fl, \
.arg_from_user = from_user, \
.arg_to_user = to_user, \
}When assigning to .fn the func is cast to a specific function prototype. Without that cast the compiler would have warned about the mismatch.
I see no reason for that cast, so drop that cast as well.
Regards,
Hans
Re: [PATCH V1] media: mc-device.c: change media_device_request_alloc to match media_ioctl_info
Dear Hans,
Quoted 43 lines · Hans Verkuil
Quoted 19 lines · On 06/08/2020 09:29, [email protected] wrote:
From: Frederic Chen <[email protected]>
We modified the type of media_device_request_alloc()'s second parameter from int* to void* so that it can match the interface defined in struct
media_ioctl_info.Signed-off-by Frederic Chen <[email protected]>--- drivers/media/mc/mc-device.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/media/mc/mc-device.c +0
-0@@ -370,9 +370,10 @@ static long media_device_get_topology(struct media_device *mdev, void *arg)return ret; }drivers/media/mc/mc-device.c static long media_device_get_topology(struct media_device *mdev, void *arg) · line 379
Quoted 10 lines
drivers/media/mc/mc-device.c +3
-2-static long media_device_request_alloc (struct media_device *mdev, -int *alloc_fd) +static long media_device_request_alloc (struct media_device *mdev,void *arg) { +int *alloc_fd = arg; +#ifdef CONFIG_MEDIA_CONTROLLER_REQUEST_API if (!mdev->ops || !mdev->ops->req_validate || !mdev->ops->req_queue)return -ENOTTY ;This change is fine, but the reason this wasn't noticed before is the cast in the
MEDIA_IOC_ARGdefine:#define MEDIA_IOC_ARG(__cmd, func, fl, from_user, to_user) \ [_IOC_NR(MEDIA_IOC_##__cmd)] = { \ .cmd = MEDIA_IOC_##__cmd, \ .fn = (long (*)(struct media_device *, void *))func, \ .flags = fl, \ .arg_from_user = from_user, \ .arg_to_user = to_user, \ }
Quoted 4 lines
When assigning to .fn the func is cast to a specific function prototype. Without that cast the compiler would have warned about the mismatch.
I see no reason for that cast, so drop that cast as well.
I got it. I will remove the cast in the next patch.
Quoted 4 lines
Regards,
Hans
Sincerely,
Frederic Chen
_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek