https://github.com/torvalds/linux
Tip revision: 59a3759d0fe8d969888c741bb33f4946e4d3750d authored by Linus Torvalds on 23 May 2009, 21:47:00 UTC
Linux 2.6.30-rc7
Linux 2.6.30-rc7
Tip revision: 59a3759
connector.txt
/*****************************************/
Kernel Connector.
/*****************************************/
Kernel connector - new netlink based userspace <-> kernel space easy
to use communication module.
Connector driver adds possibility to connect various agents using
netlink based network. One must register callback and
identifier. When driver receives special netlink message with
appropriate identifier, appropriate callback will be called.
From the userspace point of view it's quite straightforward:
socket();
bind();
send();
recv();
But if kernelspace want to use full power of such connections, driver
writer must create special sockets, must know about struct sk_buff
handling... Connector allows any kernelspace agents to use netlink
based networking for inter-process communication in a significantly
easier way:
int cn_add_callback(struct cb_id *id, char *name, void (*callback) (void *));
void cn_netlink_send(struct cn_msg *msg, u32 __group, int gfp_mask);
struct cb_id
{
__u32 idx;
__u32 val;
};
idx and val are unique identifiers which must be registered in
connector.h for in-kernel usage. void (*callback) (void *) - is a
callback function which will be called when message with above idx.val
will be received by connector core. Argument for that function must
be dereferenced to struct cn_msg *.
struct cn_msg
{
struct cb_id id;
__u32 seq;
__u32 ack;
__u32 len; /* Length of the following data */
__u8 data[0];
};
/*****************************************/
Connector interfaces.
/*****************************************/
int cn_add_callback(struct cb_id *id, char *name, void (*callback) (void *));
Registers new callback with connector core.
struct cb_id *id - unique connector's user identifier.
It must be registered in connector.h for legal in-kernel users.
char *name - connector's callback symbolic name.
void (*callback) (void *) - connector's callback.
Argument must be dereferenced to struct cn_msg *.
void cn_del_callback(struct cb_id *id);
Unregisters new callback with connector core.
struct cb_id *id - unique connector's user identifier.
int cn_netlink_send(struct cn_msg *msg, u32 __groups, int gfp_mask);
Sends message to the specified groups. It can be safely called from
softirq context, but may silently fail under strong memory pressure.
If there are no listeners for given group -ESRCH can be returned.
struct cn_msg * - message header(with attached data).
u32 __group - destination group.
If __group is zero, then appropriate group will
be searched through all registered connector users,
and message will be delivered to the group which was
created for user with the same ID as in msg.
If __group is not zero, then message will be delivered
to the specified group.
int gfp_mask - GFP mask.
Note: When registering new callback user, connector core assigns
netlink group to the user which is equal to it's id.idx.
/*****************************************/
Protocol description.
/*****************************************/
Current offers transport layer with fixed header. Recommended
protocol which uses such header is following:
msg->seq and msg->ack are used to determine message genealogy. When
someone sends message it puts there locally unique sequence and random
acknowledge numbers. Sequence number may be copied into
nlmsghdr->nlmsg_seq too.
Sequence number is incremented with each message to be sent.
If we expect reply to our message, then sequence number in received
message MUST be the same as in original message, and acknowledge
number MUST be the same + 1.
If we receive message and it's sequence number is not equal to one we
are expecting, then it is new message. If we receive message and it's
sequence number is the same as one we are expecting, but it's
acknowledge is not equal acknowledge number in original message + 1,
then it is new message.
Obviously, protocol header contains above id.
connector allows event notification in the following form: kernel
driver or userspace process can ask connector to notify it when
selected id's will be turned on or off(registered or unregistered it's
callback). It is done by sending special command to connector
driver(it also registers itself with id={-1, -1}).
As example of usage Documentation/connector now contains cn_test.c -
testing module which uses connector to request notification and to
send messages.
/*****************************************/
Reliability.
/*****************************************/
Netlink itself is not reliable protocol, that means that messages can
be lost due to memory pressure or process' receiving queue overflowed,
so caller is warned must be prepared. That is why struct cn_msg [main
connector's message header] contains u32 seq and u32 ack fields.
/*****************************************/
Userspace usage.
/*****************************************/
2.6.14 has a new netlink socket implementation, which by default does not
allow to send data to netlink groups other than 1.
So, if to use netlink socket (for example using connector)
with different group number userspace application must subscribe to
that group. It can be achieved by following pseudocode:
s = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_CONNECTOR);
l_local.nl_family = AF_NETLINK;
l_local.nl_groups = 12345;
l_local.nl_pid = 0;
if (bind(s, (struct sockaddr *)&l_local, sizeof(struct sockaddr_nl)) == -1) {
perror("bind");
close(s);
return -1;
}
{
int on = l_local.nl_groups;
setsockopt(s, 270, 1, &on, sizeof(on));
}
Where 270 above is SOL_NETLINK, and 1 is a NETLINK_ADD_MEMBERSHIP socket
option. To drop multicast subscription one should call above socket option
with NETLINK_DROP_MEMBERSHIP parameter which is defined as 0.
2.6.14 netlink code only allows to select a group which is less or equal to
the maximum group number, which is used at netlink_kernel_create() time.
In case of connector it is CN_NETLINK_USERS + 0xf, so if you want to use
group number 12345, you must increment CN_NETLINK_USERS to that number.
Additional 0xf numbers are allocated to be used by non-in-kernel users.
Due to this limitation, group 0xffffffff does not work now, so one can
not use add/remove connector's group notifications, but as far as I know,
only cn_test.c test module used it.
Some work in netlink area is still being done, so things can be changed in
2.6.15 timeframe, if it will happen, documentation will be updated for that
kernel.