[PATCH} Network interface for IPMI
Subsystems (3)
IPMI is a manage standard that allows intelligent management controllers to monitor things about the system (temperature, fan speed, etc.). The management controllers sit on a bus, and have addresses, and such. After seeing the ugliness required for the 32-bit ioctl compatability layers for 64-bit kernels, I have decided that the network interface for IPMI is a good thing, as the IPMI device ioctls have pointers and require ugly hacks. None should be needed for the network interface.
This patch adds that layer.
Relative to 2.6.10-rc2-mm1
-Corey
include/linux/net.h +1 -1
@@ -26,7 +26,7 @@struct poll_table_struct ;struct inode ; -#define NPROTO 32/* should be enough for now.. */ +#define NPROTO 64/* should be enough for now.. */ #define SYS_SOCKET 1/* sys_socket(2) */ #define SYS_BIND 2/* sys_bind(2) */
include/linux/socket.h +3 -1
@@ -177,7 +177,8 @@#define AF_WANPIPE 25/* Wanpipe API Sockets */ #define AF_LLC 26/* Linux LLC */ #define AF_BLUETOOTH 31/* Bluetooth sockets */ -#define AF_MAX 32/* For now.. */ +#define AF_IPMI 32/* IPMI sockers */ +#define AF_MAX 33/* For now.. */ /* Protocol families, same as address families. */ #define PF_UNSPEC AF_UNSPEC
@@ -209,6 +210,7 @@#define PF_WANPIPE AF_WANPIPE#define PF_LLC AF_LLC#define PF_BLUETOOTH AF_BLUETOOTH +#define PF_IPMI AF_IPMI#define PF_MAX AF_MAX/* Maximum queue length specifiable by listen. */
include/net/af_ipmi.h +59 -0
59 lines not on this page · raw message
net/ipmi/af_ipmi.c +612 -0
612 lines not on this page · raw message
net/ipmi/Makefile +1 -0
@@ -0,0 +1 @@ +obj-$(CONFIG_IPMI_SOCKET) = af_ipmi.o
net/Kconfig +11 -0
@@ -71,6 +71,17 @@ Say Y unless you know what you are doing. +config IPMI_SOCKET + tristate "IPMI sockets" + depends on IPMI_HANDLER + ---help--- + If you say Y here, you will include support for IPMI sockets; + This way you don't have to use devices to access IPMI. You + must also enable the IPMI message handler and a low-level + driver in the Character Drivers if you enable this. + + If unsure, say N. + config NET_KEY tristate "PF_KEY sockets" select XFRM
net/Makefile +1 -0
@@ -42,6 +42,7 @@ obj-$(CONFIG_ECONET) += econet/ obj-$(CONFIG_VLAN_8021Q) += 8021q/ obj-$(CONFIG_IP_SCTP) += sctp/ +obj-$(CONFIG_IPMI_SOCKET) += ipmi/ ifeq ($(CONFIG_NET),y) obj-$(CONFIG_SYSCTL) += sysctl_net.o
include/linux/net.h line 29
Quoted 13 lines · Corey Minyard
IPMI is a manage standard that allows intelligent management controllers to monitor things about the system (temperature, fan speed, etc.). The management controllers sit on a bus, and have addresses, and such. After seeing the ugliness required for the 32-bit ioctl compatability layers for 64-bit kernels, I have decided that the network interface for IPMI is a good thing, as the IPMI device ioctls have pointers and require ugly hacks. None should be needed for the network interface.
This patch adds that layer.
include/linux/net.h +1
-1-#define NPROTO 32/* should be enough for now.. */ +#define NPROTO 64/* should be enough for now.. */
Boy, that was a big bump. Was this intentional?
net/ipmi/af_ipmi.c line 519
Quoted 6 lines
net/ipmi/af_ipmi.c +6
-0+static struct ipmi_sock *ipmi_socket_create1 (struct socket *sock) +{ +struct ipmi_sock *i; + +if (atomic_read (&ipmi_nr_socks) >=2 *files_stat.max_files) +return NULL ;
Why this test?
net/Kconfig line 83
Quoted 10 lines
net/Kconfig +10
-0+config IPMI_SOCKET + tristate "IPMI sockets" + depends on IPMI_HANDLER + ---help--- + If you say Y here, you will include support for IPMI sockets; + This way you don't have to use devices to access IPMI. You + must also enable the IPMI message handler and a low-level + driver in the Character Drivers if you enable this. + + If unsure, say N.
Is this new kernel interface documented somewhere?
Quoted 5 lines · Andrew Morton
Quoted 2 lines · Corey Minyard
IPMI is a manage standard that allows intelligent management controllers to monitor things about the system (temperature, fan speed, etc.). The management controllers sit on a bus, and have addresses, and such. After seeing the ugliness required for the 32-bit ioctl compatability layers for 64-bit kernels, I have decided that the network interface for IPMI is a good thing, as the IPMI device ioctls have pointers and require ugly hacks. None should be needed for the network interface.
Quoted 10 lines
include/linux/net.h line 29
Quoted 6 lines
This patch adds that layer.
include/linux/net.h +1
-1-#define NPROTO 32/* should be enough for now.. */ +#define NPROTO 64/* should be enough for now.. */ Boy, that was a big bump. Was this intentional?
It's the next power of 2 :). Any value larger than 32 should work, I'll just set it to 33.
Quoted 14 lines
net/ipmi/af_ipmi.c line 519
Quoted 10 lines
net/ipmi/af_ipmi.c +6
-0+static struct ipmi_sock *ipmi_socket_create1 (struct socket *sock) +{ +struct ipmi_sock *i; + +if (atomic_read (&ipmi_nr_socks) >=2 *files_stat.max_files) +return NULL ;Why this test?
It snuck in from the original patch writer and I missed it. Not necessary.
Quoted 18 lines
Quoted 4 lines
net/Kconfig line 83
Quoted 10 lines
net/Kconfig +10
-0+config IPMI_SOCKET + tristate "IPMI sockets" + depends on IPMI_HANDLER + ---help--- + If you say Y here, you will include support for IPMI sockets; + This way you don't have to use devices to access IPMI. You + must also enable the IPMI message handler and a low-level + driver in the Character Drivers if you enable this. + + If unsure, say N.Is this new kernel interface documented somewhere?
I thought I had done it, but it doesn't appear to be there. I'll add some and repost a patch.
Thanks,
-Corey
Here is a patch that fixed the problems you mentioned.
Thanks,
-Corey
Documentation/IPMI.txt +121 -1
@@ -104,7 +104,7 @@ af_ipmi - A network socket interface to IPMI. This doesn't take up a character device in your system. -Note that the KCS-only interface ahs been removed. +Note that the KCS-only interface has been removed. Much documentation for the interface is in the include files. The IPMI include files are:
@@ -218,6 +218,125 @@ appropriately. +Using the Socket Interface +-------------------------- + +The socket interface to IPMI removes a lot of the evils of the ioctl +interface, including taking a character device and passing pointers +between userland and the kernel in ioctl structures. Plus, IPMI is +really kind of a network, you have addressing, message options, and +things of that nature that fit naturally into the socket interface. I +should have done that from the beginning. + +This is the recommended interface to use for IPMI. It is a lot +cleaner than the ioctl interface. + +To create an IPMI socket, do something like the following: + + int fd; + int rv; + struct sockaddr_ipmi addr; + + fd = socket(PF_IPMI, SOCK_DGRAM, 0); + if (fd == -1) + <return error> + addr.sipmi_family = AF_IPMI; + addr.if_num = if_num; + rv = bind(fd, (struct sockaddr *) &addr, sizeof(addr)); + if (rv != -1) { + close(fd) + <return error> + } + +The "if_num" variable above is which IPMI interface you are binding +to, generally zero. To send a message, do something like: + + struct sockaddr_ipmi saddr; + char smsg_data[sizeof(struct ipmi_sock_msg) + + IPMI_MAX_MSG_LENGTH]; + struct ipmi_sock_msg *smsg = (void *) smsg_data; + + saddr.sipmi_family = AF_IPMI; + memcpy(&saddr.ipmi_addr, addr, addr_len); + + smsg->netfn = netfn; + smsg->cmd = cmd; + smsg->data_len = data_len; + smsg->msgid = msgid; + memcpy(smsg->data, data, smsg->data_len); + + rv = sendto(fd, smsg, sizeof(*smsg) + msg->data_len, 0, + (struct sockaddr *) &saddr, + addr_len + SOCKADDR_IPMI_OVERHEAD); + +The "addr" variable here is an address as described by the addressing +section above. The message data is the same as described in the +messaging section above. + + struct sockaddr_ipmi addr; + socklen_t addr_len; + struct ipmi_sock_msg *smsg; + unsigned char data[MAX_IPMI_DATA_SIZE + sizeof(*smsg)]; + + + addr_len = sizeof(addr); + rv = recvfrom(fd, data, sizeof(data), 0, + (struct sockaddr *) &addr, &addr_len); + if (rv < 0) { + <handle error> + } + + if (rv < sizeof(*smsg)) { + <handle error, msg too short> + } + + smsg = (struct ipmi_sock_msg *) data; + if (rv < (sizeof(*smsg) + smsg->data_len)) { + <handle error, msg too short> + } + + recv_type = smsg->recv_type; + ipmi_addr = (struct ipmi_addr *) &addr.ipmi_addr; + ipmi_addr_len = addr_len - SOCKADDR_IPMI_OVERHEAD; + msgid = smsg->msgid; + msg.netfn = smsg->netfn; + msg.cmd = smsg->cmd; + msg.data = smsg->data; + msg.data_len = smsg->data_len; + +You must handle the received message based upon the receive type. + +There are a few other things that are ioctls for the socket. This is +a few things that cannot be done in any easy way besides an ioctl, but +these don't contain any nasty things like pointers. + +The following sets the local address for the BMC. IPMI provides no +way for the driver to discover what the IPMB address for the BMC +should be. So this has to be set by the software using it. + + unsigned int slave_addr; + rv = ioctl(fd, SIOCIPMISETADDR, &slave_addr); + +The following enabled events to be received from the interface. If +you set this to true, incoming events will be delivered to the fd. + + int enable_events; + rv = ioctl(fd, SIOCIPMIGETEVENT, &enable_events); + +To receive external command addressed to LUN 2 of the BMC, register +with the following ioctl: + + struct ipmi_cmdspec reg; + reg.netfn = netfn; + reg.cmd = cmd; + rv = ioctl(fd, SIOCIPMIREGCMD, ®); + +The deregistration call is: + + rv = ioctl(fd, SIOCIPMIUNREGCMD, ®); + + + The Upper Layer Interface (Message Handler) -------------------------------------------
@@ -293,6 +412,7 @@ Events and Incoming Commands +---------------------------- The driver takes care of polling for IPMI events and receiving commands (commands are messages that are not responses, they are
include/linux/net.h +1 -1
@@ -26,7 +26,7 @@struct poll_table_struct ;struct inode ; -#define NPROTO 32/* should be enough for now.. */ +#define NPROTO 33/* should be enough for now.. */ #define SYS_SOCKET 1/* sys_socket(2) */ #define SYS_BIND 2/* sys_bind(2) */
include/linux/socket.h +3 -1
@@ -177,7 +177,8 @@#define AF_WANPIPE 25/* Wanpipe API Sockets */ #define AF_LLC 26/* Linux LLC */ #define AF_BLUETOOTH 31/* Bluetooth sockets */ -#define AF_MAX 32/* For now.. */ +#define AF_IPMI 32/* IPMI sockers */ +#define AF_MAX 33/* For now.. */ /* Protocol families, same as address families. */ #define PF_UNSPEC AF_UNSPEC
@@ -209,6 +210,7 @@#define PF_WANPIPE AF_WANPIPE#define PF_LLC AF_LLC#define PF_BLUETOOTH AF_BLUETOOTH +#define PF_IPMI AF_IPMI#define PF_MAX AF_MAX/* Maximum queue length specifiable by listen. */
include/net/af_ipmi.h +59 -0
59 lines not on this page · raw message
net/ipmi/af_ipmi.c +608 -0
608 lines not on this page · raw message
net/ipmi/Makefile +1 -0
@@ -0,0 +1 @@ +obj-$(CONFIG_IPMI_SOCKET) = af_ipmi.o
net/Kconfig +11 -0
@@ -71,6 +71,17 @@ Say Y unless you know what you are doing. +config IPMI_SOCKET + tristate "IPMI sockets" + depends on IPMI_HANDLER + ---help--- + If you say Y here, you will include support for IPMI sockets; + This way you don't have to use devices to access IPMI. You + must also enable the IPMI message handler and a low-level + driver in the Character Drivers if you enable this. + + If unsure, say N. + config NET_KEY tristate "PF_KEY sockets" select XFRM
net/Makefile +1 -0
@@ -42,6 +42,7 @@ obj-$(CONFIG_ECONET) += econet/ obj-$(CONFIG_VLAN_8021Q) += 8021q/ obj-$(CONFIG_IP_SCTP) += sctp/ +obj-$(CONFIG_IPMI_SOCKET) += ipmi/ ifeq ($(CONFIG_NET),y) obj-$(CONFIG_SYSCTL) += sysctl_net.o