Fix control messages in sys_rtdm_recvmsg
cover Fix control messages in sys_rtdm_recvmsg
This patch-set shall fix the recvmsg system-call and its handling of control messages and the msghdr structure. It is necessary to be able to apply another patch to RTnet, which depends on a correct handling of msg_controllen.
Jan pointed out, that all driver code that relies on the current interface has to be adapted to the new one. I checked the driver code and I found can drivers to be the only ones affected. Finally, the changes to the system call have been documented in rtdm_driver.h header file. Please let me know if it's unclear...
I used a recent linux kernel as reference of how to implement the handling of msghdr. The patch changes the hadnling of msg_namelen as it won't get written back to user-space if it has been modified by driver code. That's what the Kernel does. However, there are also drivers in the Kernel tree that write to msg_namelen, so I'm not really sure what's the correct behaviour. So this might break user code that relies on this write-back although I can't really imagine a use-case where you actually need to check the size. Usually you have to know the address type (structure) to be able to interepret the address correctly.
Manuel Huber (3):
rtdm: Fix msghdr struct (cmsg) in sys_rtdm_recvmsg
rtcan: Adapt rtcan_raw to changes in rtdm
doc: Add documentation to rtdm_recvmsg_handler_t
include/rtdm/rtdm_driver.h | 9 ++++++++- ksrc/drivers/can/rtcan_raw.c | 33 ++++++++++++++++----------------- ksrc/skins/rtdm/syscall.c | 18 ++++++++++++++---- 3 files changed, 38 insertions(+), 22 deletions(-)
Signature · 1 line
1.8.3
PATCH 1/3 rtdm: Fix msghdr struct (cmsg) in sys_rtdm_recvmsg
From: Manuel Huber <[email protected]>
Whenever a new control message is put into msg_control buffer the actual address and the space left is saved to msg_control and msg_controllen. This allows adding messages as long as there is enough space left in the user-supplied buffer. Both fields have to be fixed again before passing them to the user by copying the original starting address of the buffer to msg_control and saving the actual amount of bytes written to the buffer to msg_controllen.
Note: Changes of msg_namelen (in driver code) will not be reflected to user-space anymore. This change could break user-space applications that depend on the previous implementation.
* Explicit use of __xn_put_user rather then __xn_copy_to_user
* Don't write back msg->msg_namelen
* Reset msg_flags before passing msg to driver code
--- ksrc/skins/rtdm/syscall.c | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-)
ksrc/skins/rtdm/syscall.c +14 -4
@@ -79,22 +79,32 @@ static int sys_rtdm_recvmsg(struct pt_regs *regs) {struct task_struct *p = current;struct msghdr krnl_msg; +uintptr_t cmsg_ptr; +struct msghdr __user *usr_msg;int ret; -if (unlikely (!access_wok (__xn_reg_arg2 (regs), + usr_msg = (void __user *)__xn_reg_arg2 (regs); + +if (unlikely (!access_wok ((void __user *)usr_msg,sizeof (krnl_msg)) ||__xn_copy_from_user (&krnl_msg, - (void __user *)__xn_reg_arg2 (regs), + (void __user *)usr_msg,sizeof (krnl_msg))))return -EFAULT ; + cmsg_ptr = (uintptr_t )krnl_msg.msg_control; + krnl_msg.msg_flags =0 ; + ret =__rt_dev_recvmsg (p,__xn_reg_arg1 (regs), &krnl_msg,__xn_reg_arg3 (regs));if (unlikely (ret <0 ))return ret; -if (unlikely (__xn_copy_to_user ((void __user *)__xn_reg_arg2 (regs), - &krnl_msg,sizeof (krnl_msg)))) +if (unlikely (__xn_put_user ((typeof (krnl_msg.msg_controllen)) + ((uintptr_t )krnl_msg.msg_control - cmsg_ptr), + (void __user *)&usr_msg->msg_controllen) || +__xn_put_user (krnl_msg.msg_flags, + (void __user *)&(usr_msg->msg_flags))))return -EFAULT ;return ret;
Signature · 1 line
1.8.3
PATCH 2/3 rtcan: Adapt rtcan_raw to changes in rtdm
In reply to Manuel HuberFrom: Manuel Huber <[email protected]>
--- ksrc/drivers/can/rtcan_raw.c | 33 ++++++++++++++++----------------- 1 file changed, 16 insertions(+), 17 deletions(-)
ksrc/drivers/can/rtcan_raw.c +16 -17
@@ -559,6 +559,7 @@ ssize_t rtcan_raw_recvmsg(struct rtdm_dev_context *context,size_t payload_size;rtdm_lockctx_t lock_ctx;int ret; +int copy_timestamp =0 ;/* Clear frame memory location */ memset (&frame,0 ,sizeof (can_frame_t));
@@ -610,8 +611,10 @@ ssize_t rtcan_raw_recvmsg(struct rtdm_dev_context *context, } if (msg->msg_control != NULL) { - if (msg->msg_controllen < sizeof(nanosecs_abs_t)) - return -EINVAL; + if (msg->msg_controllen >= sizeof(timestamp)) + copy_timestamp = 1; + else + msg->msg_flags |= MSG_CTRUNC; if (user_info) { if (!rtdm_rw_user_ok(user_info, msg->msg_control,
@@ -686,7 +689,7 @@ ssize_t rtcan_raw_recvmsg(struct rtdm_dev_context *context,/* Is a timestamp available and is the caller actually interested? */ -if (msg->msg_controllen && (can_dlc &RTCAN_HAS_TIMESTAMP )) { +if (copy_timestamp && (can_dlc &RTCAN_HAS_TIMESTAMP )) {/* Copy timestamp */ MEMCPY_FROM_RING_BUF (×tamp,RTCAN_TIMESTAMP_SIZE ); }
@@ -741,15 +744,13 @@ ssize_t rtcan_raw_recvmsg(struct rtdm_dev_context *context,return -EFAULT ;/* Copy timestamp if existent and wanted */ -if (msg->msg_controllen) { -if (can_dlc &RTCAN_HAS_TIMESTAMP ) { -if (rtdm_copy_to_user (user_info, msg->msg_control, - ×tamp,RTCAN_TIMESTAMP_SIZE )) -return -EFAULT ; +if (copy_timestamp && (can_dlc &RTCAN_HAS_TIMESTAMP )) { +if (rtdm_copy_to_user (user_info, msg->msg_control, + ×tamp,RTCAN_TIMESTAMP_SIZE )) +return -EFAULT ; - msg->msg_controllen =RTCAN_TIMESTAMP_SIZE ; - }else - msg->msg_controllen =0 ; + msg->msg_controllen -=RTCAN_TIMESTAMP_SIZE ; + msg->msg_control +=RTCAN_TIMESTAMP_SIZE ; } }else {
@@ -768,12 +769,10 @@ ssize_t rtcan_raw_recvmsg(struct rtdm_dev_context *context, iov->iov_len -=sizeof (can_frame_t);/* Copy timestamp if existent and wanted */ -if (msg->msg_controllen) { -if (can_dlc &RTCAN_HAS_TIMESTAMP ) { -memcpy (msg->msg_control, ×tamp,RTCAN_TIMESTAMP_SIZE ); - msg->msg_controllen =RTCAN_TIMESTAMP_SIZE ; - }else - msg->msg_controllen =0 ; +if (copy_timestamp && (can_dlc &RTCAN_HAS_TIMESTAMP )) { +memcpy (msg->msg_control, ×tamp,RTCAN_TIMESTAMP_SIZE ); + msg->msg_controllen -=RTCAN_TIMESTAMP_SIZE ; + msg->msg_control +=RTCAN_TIMESTAMP_SIZE ; } }
Signature · 1 line
1.8.3
PATCH 3/3 doc: Add documentation to rtdm_recvmsg_handler_t
In reply to Manuel HuberFrom: Manuel Huber <[email protected]>
Describes consequences of changes in sys_rtdm_recvmsg to the interface regarding the msghdr struct.
--- include/rtdm/rtdm_driver.h | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
include/rtdm/rtdm_driver.h +8 -1
@@ -294,7 +294,14 @@ typedef ssize_t (*rtdm_write_handler_t)(struct rtdm_dev_context *context,* @param[in] user_info Opaque pointer to information about user mode caller, * NULL if kernel mode call * @param[in,out] msg Message descriptor as passed by the user, automatically -* mirrored to safe kernel memory in case of user mode call +* mirrored to safe kernel memory in case of user mode call. If msg_control +* points to a valid buffer (not NULL) and msg_controllen is > 0, cmsghdr +* structures can be put into msg_control buffer by driver code. Driver code +* has to move msg_control to point to the next free byte (or the end of the +* buffer). The core layer will update msg_controllen to reflect the actual +* number of meaningful bytes written to msg_control buffer. If there is +* insufficient space to write all control messages, MSG_CTRUNC has to be set +* by the driver (do not abort). * @param[in] flags Message flags as passed by the user * * @return On success, the number of bytes received. On failure return
Signature · 1 line
1.8.3