This commit was generated by cvs2svn to compensate for changes in r786,
[libnl.git] / lib / nl.c
diff --git a/lib/nl.c b/lib/nl.c
new file mode 100644 (file)
index 0000000..600f93e
--- /dev/null
+++ b/lib/nl.c
@@ -0,0 +1,1080 @@
+/*
+ * lib/nl.c            Core Netlink Interface
+ *
+ *     This library is free software; you can redistribute it and/or
+ *     modify it under the terms of the GNU Lesser General Public
+ *     License as published by the Free Software Foundation version 2.1
+ *     of the License.
+ *
+ * Copyright (c) 2003-2006 Thomas Graf <tgraf@suug.ch>
+ */
+
+/**
+ * @defgroup nl Core Netlink API
+ * @brief
+ *
+ * @par 1) Creating the netlink handle
+ * @code
+ * struct nl_handle *handle;
+ *
+ * // Allocate and initialize a new netlink handle
+ * handle = nl_handle_new();
+ *
+ * // Are multiple handles being allocated? You have to provide a unique
+ * // netlink process id and overwrite the default local process id.
+ * nl_handle_set_pid(handle, MY_UNIQUE_PID);
+ *
+ * // Is this socket used for event processing? You need to disable sequence
+ * // number checking in order to be able to receive messages not explicitely
+ * // requested.
+ * nl_disable_sequence_check(handle);
+ *
+ * // Use nl_handle_get_fd() to fetch the file description, for example to
+ * // put a socket into non-blocking i/o mode.
+ * fcntl(nl_handle_get_fd(handle), F_SETFL, O_NONBLOCK);
+ * @endcode
+ *
+ * @par 2) Joining Groups
+ * @code
+ * // You may join/subscribe to as many groups as you want, don't forget
+ * // to eventually disable sequence number checking. Note: Joining must
+ * // be done before connecting/binding the socket.
+ * nl_join_groups(handle, GROUP_ID1 | GROUP_ID2);
+ * @endcode
+ *
+ * @par 3) Connecting the socket
+ * @code
+ * // Bind and connect the socket to a protocol, NETLINK_ROUTE in this example.
+ * nl_connect(handle, NETLINK_ROUTE);
+ * @endcode
+ *
+ * @par 4) Sending data
+ * @code
+ * // The most rudimentary method is to use nl_sendto() simply pushing
+ * // a piece of data to the other netlink peer. This method is not
+ * // recommended.
+ * const char buf[] = { 0x01, 0x02, 0x03, 0x04 };
+ * nl_sendto(handle, buf, sizeof(buf));
+ *
+ * // A more comfortable interface is nl_send() taking a pointer to
+ * // a netlink message.
+ * struct nl_msg *msg = my_msg_builder();
+ * nl_send(handle, nlmsg_hdr(msg));
+ *
+ * // nl_sendmsg() provides additional control over the sendmsg() message
+ * // header in order to allow more specific addressing of multiple peers etc.
+ * struct msghdr hdr = { ... };
+ * nl_sendmsg(handle, nlmsg_hdr(msg), &hdr);
+ *
+ * // You're probably too lazy to fill out the netlink pid, sequence number
+ * // and message flags all the time. nl_send_auto_complete() automatically
+ * // extends your message header as needed with an appropriate sequence
+ * // number, the netlink pid stored in the netlink handle and the message
+ * // flags NLM_F_REQUEST and NLM_F_ACK
+ * nl_send_auto_complete(handle, nlmsg_hdr(msg));
+ *
+ * // Simple protocols don't require the complex message construction interface
+ * // and may favour nl_send_simple() to easly send a bunch of payload
+ * // encapsulated in a netlink message header.
+ * nl_send_simple(handle, MY_MSG_TYPE, 0, buf, sizeof(buf));
+ * @endcode
+ *
+ * @par 5) Receiving data
+ * @code
+ * // nl_recv() receives a single message allocating a buffer for the message
+ * // content and gives back the pointer to you.
+ * struct sockaddr_nl peer;
+ * unsigned char *msg;
+ * nl_recv(handle, &peer, &msg);
+ *
+ * // nl_recvmsgs() receives a bunch of messages until the callback system
+ * // orders it to state, usually after receving a compolete multi part
+ * // message series.
+ * nl_recvmsgs(handle, my_callback_configuration);
+ *
+ * // nl_recvmsgs_def() acts just like nl_recvmsg() but uses the callback
+ * // configuration stored in the handle.
+ * nl_recvmsgs_def(handle);
+ *
+ * // In case you want to wait for the ACK to be recieved that you requested
+ * // with your latest message, you can call nl_wait_for_ack()
+ * nl_wait_for_ack(handle);
+ * @endcode
+ *
+ * @par 6) Cleaning up
+ * @code
+ * // Close the socket first to release kernel memory
+ * nl_close(handle);
+ *
+ * // Finally destroy the netlink handle
+ * nl_handle_destroy(handle);
+ * @endcode
+ * 
+ * @{
+ */
+
+#include <netlink-local.h>
+#include <netlink/netlink.h>
+#include <netlink/utils.h>
+#include <netlink/handlers.h>
+#include <netlink/msg.h>
+#include <netlink/attr.h>
+
+/**
+ * @name Handle Management
+ * @{
+ */
+
+/**
+ * Allocate and initialize new non-default netlink handle.
+ * @arg kind           Kind of callback handler to use per default.
+ *
+ * Allocates and initializes a new netlink handle, the netlink process id
+ * is set to the local process id which may conflict if multiple handles
+ * are created, therefore you may have to overwrite it using
+ * nl_handle_set_pid(). The initial sequence number is initialized to the
+ * current UNIX time.
+ *
+ * @return Newly allocated netlink handle or NULL.
+ */
+struct nl_handle *nl_handle_alloc_nondefault(enum nl_cb_kind kind)
+{
+       struct nl_handle *handle;
+       
+       handle = calloc(1, sizeof(*handle));
+       if (!handle)
+               goto errout;
+
+       handle->h_cb = nl_cb_new(kind);
+       if (!handle->h_cb)
+               goto errout;
+       
+       handle->h_local.nl_family = AF_NETLINK;
+       handle->h_peer.nl_family = AF_NETLINK;
+       handle->h_local.nl_pid = getpid();
+       handle->h_seq_expect = handle->h_seq_next = time(0);
+
+       return handle;
+errout:
+       nl_handle_destroy(handle);
+       nl_errno(ENOMEM);
+       return NULL;
+}
+
+/**
+ * Allocate and initialize new netlink handle.
+ *
+ * Allocates and initializes a new netlink handle, the netlink process id
+ * is set to the local process id which may conflict if multiple handles
+ * are created, therefore you may have to overwrite it using
+ * nl_handle_set_pid(). The initial sequence number is initialized to the
+ * current UNIX time. The default callback (NL_CB_DEFAULT) handlers are
+ * being used.
+ *
+ * @return Newly allocated netlink handle or NULL.
+ */
+struct nl_handle *nl_handle_alloc(void)
+{
+       return nl_handle_alloc_nondefault(NL_CB_DEFAULT);
+}
+
+/**
+ * Destroy netlink handle.
+ * @arg handle         Netlink handle.
+ */
+void nl_handle_destroy(struct nl_handle *handle)
+{
+       if (!handle)
+               return;
+
+       nl_cb_destroy(handle->h_cb);
+       free(handle);
+}
+
+/** @} */
+
+/**
+ * @name Utilities
+ * @{
+ */
+
+/**
+ * Set socket buffer size of netlink handle.
+ * @arg handle         Netlink handle.
+ * @arg rxbuf          New receive socket buffer size in bytes.
+ * @arg txbuf          New transmit socket buffer size in bytes.
+ *
+ * Sets the socket buffer size of a netlink handle to the specified
+ * values \c rxbuf and \c txbuf. Providing a value of \c 0 assumes a
+ * good default value.
+ *
+ * @note It is not required to call this function prior to nl_connect().
+ * @return 0 on sucess or a negative error code.
+ */
+int nl_set_buffer_size(struct nl_handle *handle, int rxbuf, int txbuf)
+{
+       int err;
+
+       if (rxbuf <= 0)
+               rxbuf = 32768;
+
+       if (txbuf <= 0)
+               txbuf = 32768;
+       
+       err = setsockopt(handle->h_fd, SOL_SOCKET, SO_SNDBUF,
+                        &txbuf, sizeof(txbuf));
+       if (err < 0)
+               return nl_error(errno, "setsockopt(SO_SNDBUF) failed");
+
+       err = setsockopt(handle->h_fd, SOL_SOCKET, SO_RCVBUF,
+                        &rxbuf, sizeof(rxbuf));
+       if (err < 0)
+               return nl_error(errno, "setsockopt(SO_RCVBUF) failed");
+
+       handle->h_flags |= NL_SOCK_BUFSIZE_SET;
+
+       return 0;
+}
+
+/**
+ * Enable/disable credential passing on netlink handle.
+ * @arg handle         Netlink handle
+ * @arg state          New state (0 - disabled, 1 - enabled)
+ */
+int nl_set_passcred(struct nl_handle *handle, int state)
+{
+       int err;
+
+       err = setsockopt(handle->h_fd, SOL_SOCKET, SO_PASSCRED,
+                        &state, sizeof(state));
+       if (err < 0)
+               return nl_error(errno, "setsockopt(SO_PASSCRED) failed");
+
+       if (state)
+               handle->h_flags |= NL_SOCK_PASSCRED;
+       else
+               handle->h_flags &= ~NL_SOCK_PASSCRED;
+
+       return 0;
+}
+
+/**
+ * Join multicast groups.
+ * @arg handle         Netlink handle.
+ * @arg groups         Bitmask of groups to join.
+ *
+ * @note Joining of groups must be done prior to connecting/binding
+ *       the socket (nl_connect()).
+ */
+void nl_join_groups(struct nl_handle *handle, int groups)
+{
+       handle->h_local.nl_groups |= groups;
+}
+
+#ifndef SOL_NETLINK
+#define SOL_NETLINK 270
+#endif
+
+int nl_join_group(struct nl_handle *handle, int group)
+{
+       int err;
+
+       err = setsockopt(handle->h_fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
+                        &group, sizeof(group));
+       if (err < 0)
+               return nl_error(errno, "setsockopt(NETLINK_ADD_MEMBERSHIP) "
+                                      "failed");
+
+       return 0;
+}
+
+static int noop_seq_check(struct nl_msg *msg, void *arg)
+{
+       return NL_PROCEED;
+}
+
+/**
+ * Disable sequence number checking.
+ * @arg handle         Netlink handle.
+ *
+ * Disables checking of sequence numbers on the netlink handle. This is
+ * required to allow messages to be processed which were not requested by
+ * a preceding request message, e.g. netlink events.
+ */
+void nl_disable_sequence_check(struct nl_handle *handle)
+{
+       nl_cb_set(nl_handle_get_cb(handle), NL_CB_SEQ_CHECK,
+                 NL_CB_CUSTOM, noop_seq_check, NULL);
+}
+
+/** @} */
+
+/**
+ * @name Acccess Functions
+ * @{
+ */
+
+/**
+ * Get netlink process identifier of netlink handle.
+ * @arg handle         Netlink handle.
+ * @return Netlink process identifier.
+ */
+pid_t nl_handle_get_pid(struct nl_handle *handle)
+{
+       return handle->h_local.nl_pid;
+}
+
+/**
+ * Set netlink process identifier of netlink handle.
+ * @arg handle         Netlink handle.
+ * @arg pid            New netlink process identifier.
+ */
+void nl_handle_set_pid(struct nl_handle *handle, pid_t pid)
+{
+       handle->h_local.nl_pid = pid;
+}
+
+/**
+ * Get netlink process identifier of peer from netlink handle.
+ * @arg handle         Netlink handle.
+ * @return Netlink process identifier.
+ */
+pid_t nl_handle_get_peer_pid(struct nl_handle *handle)
+{
+       return handle->h_peer.nl_pid;
+}
+
+/**
+ * Set netlink process identifier of peer in netlink handle.
+ * @arg handle         Netlink handle.
+ * @arg pid            New netlink process identifier.
+ */
+void nl_handle_set_peer_pid(struct nl_handle *handle, pid_t pid)
+{
+       handle->h_peer.nl_pid = pid;
+}
+
+/**
+ * Get file descriptor of netlink handle.
+ * @arg handle         Netlink handle.
+ * @return File descriptor of netlink socket or -1 if not connected.
+ */
+int nl_handle_get_fd(struct nl_handle *handle)
+{
+       return handle->h_fd;
+}
+
+/**
+ * Get local netlink address of netlink handle.
+ * @arg handle         Netlink handle.
+ * @return Local netlink address.
+ */
+struct sockaddr_nl *nl_handle_get_local_addr(struct nl_handle *handle)
+{
+       return &handle->h_local;
+}
+
+/**
+ * Get peer netlink address of netlink handle.
+ * @arg handle         Netlink handle.
+ * @note The peer address is undefined while the socket is unconnected.
+ * @return Netlink address of the peer.
+ */
+struct sockaddr_nl *nl_handle_get_peer_addr(struct nl_handle *handle)
+{
+       return &handle->h_peer;
+}
+
+/**
+ * Get callback configuration of netlink handle.
+ * @arg handle         Netlink handle.
+ * @return Currently active callback configuration or NULL if not available.
+ */
+struct nl_cb *nl_handle_get_cb(struct nl_handle *handle)
+{
+       return handle->h_cb;
+}
+
+/** @} */
+
+/**
+ * @name Connection Management
+ * @{
+ */
+
+/**
+ * Create and connect netlink socket.
+ * @arg handle         Netlink handle.
+ * @arg protocol       Netlink protocol to use.
+ *
+ * Creates a netlink socket using the specified protocol, binds the socket
+ * and issues a connection attempt.
+ *
+ * @return 0 on success or a negative error code.
+ */
+int nl_connect(struct nl_handle *handle, int protocol)
+{
+       int err;
+       socklen_t addrlen;
+
+       handle->h_fd = socket(AF_NETLINK, SOCK_RAW, protocol);
+       if (handle->h_fd < 0)
+               return nl_error(1, "socket(AF_NETLINK, ...) failed");
+
+       if (!(handle->h_flags & NL_SOCK_BUFSIZE_SET)) {
+               err = nl_set_buffer_size(handle, 0, 0);
+               if (err < 0)
+                       return err;
+       }
+
+       err = bind(handle->h_fd, (struct sockaddr*) &handle->h_local,
+                  sizeof(handle->h_local));
+       if (err < 0)
+               return nl_error(1, "bind() failed");
+
+       addrlen = sizeof(handle->h_local);
+       err = getsockname(handle->h_fd, (struct sockaddr *) &handle->h_local,
+                         &addrlen);
+       if (err < 0)
+               return nl_error(1, "getsockname failed");
+
+       if (addrlen != sizeof(handle->h_local))
+               return nl_error(EADDRNOTAVAIL, "Invalid address length");
+
+       if (handle->h_local.nl_family != AF_NETLINK)
+               return nl_error(EPFNOSUPPORT, "Address format not supported");
+
+       handle->h_proto = protocol;
+
+       return 0;
+}
+
+/**
+ * Close/Disconnect netlink socket.
+ * @arg handle         Netlink handle
+ */
+void nl_close(struct nl_handle *handle)
+{
+       if (handle->h_fd >= 0) {
+               close(handle->h_fd);
+               handle->h_fd = -1;
+       }
+
+       handle->h_proto = 0;
+}
+
+/** @} */
+
+/**
+ * @name Send
+ * @{
+ */
+
+/**
+ * Send raw data over netlink socket.
+ * @arg handle         Netlink handle.
+ * @arg buf            Data buffer.
+ * @arg size           Size of data buffer.
+ * @return Number of characters written on success or a negative error code.
+ */
+int nl_sendto(struct nl_handle *handle, void *buf, size_t size)
+{
+       int ret;
+
+       ret = sendto(handle->h_fd, buf, size, 0, (struct sockaddr *)
+                    &handle->h_peer, sizeof(handle->h_peer));
+       if (ret < 0)
+               return nl_errno(errno);
+
+       return ret;
+}
+
+/**
+ * Send netlink message with control over sendmsg() message header.
+ * @arg handle         Netlink handle.
+ * @arg msg            Netlink message to be sent.
+ * @arg hdr            Sendmsg() message header.
+ * @return Number of characters sent on sucess or a negative error code.
+ */
+int nl_sendmsg(struct nl_handle *handle, struct nl_msg *msg, struct msghdr *hdr)
+{
+       struct nl_cb *cb;
+       int ret;
+
+       struct iovec iov = {
+               .iov_base = (void *) nlmsg_hdr(msg),
+               .iov_len = nlmsg_hdr(msg)->nlmsg_len,
+       };
+
+       hdr->msg_iov = &iov;
+       hdr->msg_iovlen = 1;
+
+       nlmsg_set_src(msg, &handle->h_local);
+
+       cb = nl_handle_get_cb(handle);
+       if (cb->cb_set[NL_CB_MSG_OUT])
+               if (nl_cb_call(cb, NL_CB_MSG_OUT, msg) != NL_PROCEED)
+                       return 0;
+
+       ret = sendmsg(handle->h_fd, hdr, 0);
+       if (ret < 0)
+               return nl_errno(errno);
+
+       return ret;
+}
+
+
+/**
+ * Send netlink message.
+ * @arg handle         Netlink handle
+ * @arg msg            Netlink message to be sent.
+ * @see nl_sendmsg()
+ * @return Number of characters sent on success or a negative error code.
+ */
+int nl_send(struct nl_handle *handle, struct nl_msg *msg)
+{
+       struct sockaddr_nl *dst;
+       struct ucred *creds;
+       
+       struct msghdr hdr = {
+               .msg_name = (void *) &handle->h_peer,
+               .msg_namelen = sizeof(struct sockaddr_nl),
+       };
+
+       /* Overwrite destination if specified in the message itself, defaults
+        * to the peer address of the handle.
+        */
+       dst = nlmsg_get_dst(msg);
+       if (dst->nl_family == AF_NETLINK)
+               hdr.msg_name = dst;
+
+       /* Add credentials if present. */
+       creds = nlmsg_get_creds(msg);
+       if (creds != NULL) {
+               char buf[CMSG_SPACE(sizeof(struct ucred))];
+               struct cmsghdr *cmsg;
+
+               hdr.msg_control = buf;
+               hdr.msg_controllen = sizeof(buf);
+
+               cmsg = CMSG_FIRSTHDR(&hdr);
+               cmsg->cmsg_level = SOL_SOCKET;
+               cmsg->cmsg_type = SCM_CREDENTIALS;
+               cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
+               memcpy(CMSG_DATA(cmsg), creds, sizeof(struct ucred));
+       }
+
+       return nl_sendmsg(handle, msg, &hdr);
+}
+
+/**
+ * Send netlink message and check & extend header values as needed.
+ * @arg handle         Netlink handle.
+ * @arg msg            Netlink message to be sent.
+ *
+ * Checks the netlink message \c nlh for completness and extends it
+ * as required before sending it out. Checked fields include pid,
+ * sequence nr, and flags.
+ *
+ * @see nl_send()
+ * @return Number of characters sent or a negative error code.
+ */
+int nl_send_auto_complete(struct nl_handle *handle, struct nl_msg *msg)
+{
+       struct nlmsghdr *nlh;
+
+       nlh = nlmsg_hdr(msg);
+       if (nlh->nlmsg_pid == 0)
+               nlh->nlmsg_pid = handle->h_local.nl_pid;
+
+       if (nlh->nlmsg_seq == 0)
+               nlh->nlmsg_seq = handle->h_seq_next++;
+       
+       nlh->nlmsg_flags |= (NLM_F_REQUEST | NLM_F_ACK);
+
+       if (handle->h_cb->cb_send_ow)
+               return handle->h_cb->cb_send_ow(handle, msg);
+       else
+               return nl_send(handle, msg);
+}
+
+/**
+ * Send simple netlink message using nl_send_auto_complete()
+ * @arg handle         Netlink handle.
+ * @arg type           Netlink message type.
+ * @arg flags          Netlink message flags.
+ * @arg buf            Data buffer.
+ * @arg size           Size of data buffer.
+ *
+ * Builds a netlink message with the specified type and flags and
+ * appends the specified data as payload to the message.
+ *
+ * @see nl_send_auto_complete()
+ * @return Number of characters sent on success or a negative error code.
+ */
+int nl_send_simple(struct nl_handle *handle, int type, int flags, void *buf,
+                  size_t size)
+{
+       int err;
+       struct nl_msg *msg;
+       struct nlmsghdr nlh = {
+               .nlmsg_len = nlmsg_msg_size(0),
+               .nlmsg_type = type,
+               .nlmsg_flags = flags,
+       };
+
+       msg = nlmsg_build(&nlh);
+       if (!msg)
+               return nl_errno(ENOMEM);
+
+       if (buf && size)
+               nlmsg_append(msg, buf, size, 1);
+
+       err = nl_send_auto_complete(handle, msg);
+       nlmsg_free(msg);
+
+       return err;
+}
+
+/** @} */
+
+/**
+ * @name Receive
+ * @{
+ */
+
+/**
+ * Receive netlink message from netlink socket.
+ * @arg handle         Netlink handle.
+ * @arg nla            Destination pointer for peer's netlink address.
+ * @arg buf            Destination pointer for message content.
+ * @arg creds          Destination pointer for credentials.
+ *
+ * Receives a netlink message, allocates a buffer in \c *buf and
+ * stores the message content. The peer's netlink address is stored
+ * in \c *nla. The caller is responsible for freeing the buffer allocated
+ * in \c *buf if a positive value is returned.  Interruped system calls
+ * are handled by repeating the read. The input buffer size is determined
+ * by peeking before the actual read is done.
+ *
+ * A non-blocking sockets causes the function to return immediately if
+ * no data is available.
+ *
+ * @return Number of octets read, 0 on EOF or a negative error code.
+ */
+int nl_recv(struct nl_handle *handle, struct sockaddr_nl *nla,
+           unsigned char **buf, struct ucred **creds)
+{
+       int n;
+       int flags = MSG_PEEK;
+
+       struct iovec iov = {
+               .iov_len = 4096,
+       };
+
+       struct msghdr msg = {
+               .msg_name = (void *) nla,
+               .msg_namelen = sizeof(sizeof(struct sockaddr_nl)),
+               .msg_iov = &iov,
+               .msg_iovlen = 1,
+               .msg_control = NULL,
+               .msg_controllen = 0,
+               .msg_flags = 0,
+       };
+       struct cmsghdr *cmsg;
+
+       iov.iov_base = *buf = calloc(1, iov.iov_len);
+
+       if (handle->h_flags & NL_SOCK_PASSCRED) {
+               msg.msg_controllen = CMSG_SPACE(sizeof(struct ucred));
+               msg.msg_control = calloc(1, msg.msg_controllen);
+       }
+retry:
+
+       if ((n = recvmsg(handle->h_fd, &msg, flags)) <= 0) {
+               if (!n)
+                       goto abort;
+               else if (n < 0) {
+                       if (errno == EINTR)
+                               goto retry;
+                       else if (errno == EAGAIN)
+                               goto abort;
+                       else {
+                               free(msg.msg_control);
+                               free(*buf);
+                               return nl_error(errno, "recvmsg failed");
+                       }
+               }
+       }
+       
+       if (iov.iov_len < n) {
+               /* Provided buffer is not long enough, enlarge it
+                * and try again. */
+               iov.iov_len *= 2;
+               iov.iov_base = *buf = realloc(*buf, iov.iov_len);
+               goto retry;
+       } else if (msg.msg_flags & MSG_CTRUNC) {
+               msg.msg_controllen *= 2;
+               msg.msg_control = realloc(msg.msg_control, msg.msg_controllen);
+               goto retry;
+       } else if (flags != 0) {
+               /* Buffer is big enough, do the actual reading */
+               flags = 0;
+               goto retry;
+       }
+
+       if (msg.msg_namelen != sizeof(struct sockaddr_nl)) {
+               free(msg.msg_control);
+               free(*buf);
+               return nl_error(EADDRNOTAVAIL, "socket address size mismatch");
+       }
+
+       for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
+               if (cmsg->cmsg_level == SOL_SOCKET &&
+                   cmsg->cmsg_type == SCM_CREDENTIALS) {
+                       *creds = calloc(1, sizeof(struct ucred));
+                       memcpy(*creds, CMSG_DATA(cmsg), sizeof(struct ucred));
+                       break;
+               }
+       }
+
+       free(msg.msg_control);
+       return n;
+
+abort:
+       free(msg.msg_control);
+       free(*buf);
+       return 0;
+}
+
+
+/**
+ * Receive a set of messages from a netlink socket.
+ * @arg handle         netlink handle
+ * @arg cb             set of callbacks to control the behaviour.
+ *
+ * Repeatedly calls nl_recv() and parses the messages as netlink
+ * messages. Stops reading if one of the callbacks returns
+ * NL_EXIT or nl_recv returns either 0 or a negative error code.
+ *
+ * A non-blocking sockets causes the function to return immediately if
+ * no data is available.
+ *
+ * @return 0 on success or a negative error code from nl_recv().
+ */
+int nl_recvmsgs(struct nl_handle *handle, struct nl_cb *cb)
+{
+       int n, err = 0;
+       unsigned char *buf = NULL;
+       struct nlmsghdr *hdr;
+       struct sockaddr_nl nla = {0};
+       struct nl_msg *msg = NULL;
+       struct ucred *creds = NULL;
+
+continue_reading:
+       if (cb->cb_recv_ow)
+               n = cb->cb_recv_ow(handle, &nla, &buf, &creds);
+       else
+               n = nl_recv(handle, &nla, &buf, &creds);
+
+       if (n <= 0)
+               return n;
+
+       NL_DBG(3, "recvmsgs(%p): Read %d bytes\n", handle, n);
+
+       hdr = (struct nlmsghdr *) buf;
+       while (nlmsg_ok(hdr, n)) {
+               NL_DBG(3, "recgmsgs(%p): Processing valid message...\n",
+                      handle);
+
+               nlmsg_free(msg);
+               msg = nlmsg_convert(hdr);
+               if (!msg) {
+                       err = nl_errno(ENOMEM);
+                       goto out;
+               }
+
+               nlmsg_set_proto(msg, handle->h_proto);
+               nlmsg_set_src(msg, &nla);
+               if (creds)
+                       nlmsg_set_creds(msg, creds);
+
+               /* Raw callback is the first, it gives the most control
+                * to the user and he can do his very own parsing. */
+               if (cb->cb_set[NL_CB_MSG_IN]) {
+                       err = nl_cb_call(cb, NL_CB_MSG_IN, msg);
+                       if (err == NL_SKIP)
+                               goto skip;
+                       else if (err == NL_EXIT || err < 0)
+                               goto out;
+               }
+
+               /* Sequence number checking. The check may be done by
+                * the user, otherwise a very simple check is applied
+                * enforcing strict ordering */
+               if (cb->cb_set[NL_CB_SEQ_CHECK]) {
+                       err = nl_cb_call(cb, NL_CB_SEQ_CHECK, msg);
+                       if (err == NL_SKIP)
+                               goto skip;
+                       else if (err == NL_EXIT || err < 0)
+                               goto out;
+               } else if (hdr->nlmsg_seq != handle->h_seq_expect) {
+                       if (cb->cb_set[NL_CB_INVALID]) {
+                               err = nl_cb_call(cb, NL_CB_INVALID, msg);
+                               if (err == NL_SKIP)
+                                       goto skip;
+                               else if (err == NL_EXIT || err < 0)
+                                       goto out;
+                       } else
+                               goto out;
+               }
+
+               if (hdr->nlmsg_type == NLMSG_DONE ||
+                   hdr->nlmsg_type == NLMSG_ERROR ||
+                   hdr->nlmsg_type == NLMSG_NOOP ||
+                   hdr->nlmsg_type == NLMSG_OVERRUN) {
+                       /* We can't check for !NLM_F_MULTI since some netlink
+                        * users in the kernel are broken. */
+                       handle->h_seq_expect++;
+                       NL_DBG(3, "recvmsgs(%p): Increased expected " \
+                              "sequence number to %d\n",
+                              handle, handle->h_seq_expect);
+               }
+       
+               /* Other side wishes to see an ack for this message */
+               if (hdr->nlmsg_flags & NLM_F_ACK) {
+                       if (cb->cb_set[NL_CB_SEND_ACK]) {
+                               err = nl_cb_call(cb, NL_CB_SEND_ACK, msg);
+                               if (err == NL_SKIP)
+                                       goto skip;
+                               else if (err == NL_EXIT || err < 0)
+                                       goto out;
+                       } else {
+                               /* FIXME: implement */
+                       }
+               }
+
+               /* messages terminates a multpart message, this is
+                * usually the end of a message and therefore we slip
+                * out of the loop by default. the user may overrule
+                * this action by skipping this packet. */
+               if (hdr->nlmsg_type == NLMSG_DONE) {
+                       if (cb->cb_set[NL_CB_FINISH]) {
+                               err = nl_cb_call(cb, NL_CB_FINISH, msg);
+                               if (err == NL_SKIP)
+                                       goto skip;
+                               else if (err == NL_EXIT || err < 0)
+                                       goto out;
+                       }
+                       err = 0;
+                       goto out;
+               }
+
+               /* Message to be ignored, the default action is to
+                * skip this message if no callback is specified. The
+                * user may overrule this action by returning
+                * NL_PROCEED. */
+               else if (hdr->nlmsg_type == NLMSG_NOOP) {
+                       if (cb->cb_set[NL_CB_SKIPPED]) {
+                               err = nl_cb_call(cb, NL_CB_SKIPPED, msg);
+                               if (err == NL_SKIP)
+                                       goto skip;
+                               else if (err == NL_EXIT || err < 0)
+                                       goto out;
+                       } else
+                               goto skip;
+               }
+
+               /* Data got lost, report back to user. The default action is to
+                * quit parsing. The user may overrule this action by retuning
+                * NL_SKIP or NL_PROCEED (dangerous) */
+               else if (hdr->nlmsg_type == NLMSG_OVERRUN) {
+                       if (cb->cb_set[NL_CB_OVERRUN]) {
+                               err = nl_cb_call(cb, NL_CB_OVERRUN, msg);
+                               if (err == NL_SKIP)
+                                       goto skip;
+                               else if (err == NL_EXIT || err < 0)
+                                       goto out;
+                       } else
+                               goto out;
+               }
+
+               /* Message carries a nlmsgerr */
+               else if (hdr->nlmsg_type == NLMSG_ERROR) {
+                       struct nlmsgerr *e = nlmsg_data(hdr);
+
+                       if (hdr->nlmsg_len < nlmsg_msg_size(sizeof(*e))) {
+                               /* Truncated error message, the default action
+                                * is to stop parsing. The user may overrule
+                                * this action by returning NL_SKIP or
+                                * NL_PROCEED (dangerous) */
+                               if (cb->cb_set[NL_CB_INVALID]) {
+                                       err = nl_cb_call(cb, NL_CB_INVALID,
+                                                        msg);
+                                       if (err == NL_SKIP)
+                                               goto skip;
+                                       else if (err == NL_EXIT || err < 0)
+                                               goto out;
+                               } else
+                                       goto out;
+                       } else if (e->error) {
+                               /* Error message reported back from kernel. */
+                               if (cb->cb_err) {
+                                       err = cb->cb_err(&nla, e,
+                                                          cb->cb_err_arg);
+                                       if (err == NL_SKIP)
+                                               goto skip;
+                                       else if (err == NL_EXIT || err < 0) {
+                                               nl_error(-e->error,
+                                                        "Netlink Error");
+                                               err = e->error;
+                                               goto out;
+                                       }
+                               } else {
+                                       nl_error(-e->error, "Netlink Error");
+                                       err = e->error;
+                                       goto out;
+                               }
+                       } else if (cb->cb_set[NL_CB_ACK]) {
+                               /* ACK */
+                               err = nl_cb_call(cb, NL_CB_ACK, msg);
+                               if (err == NL_SKIP)
+                                       goto skip;
+                               else if (err == NL_EXIT || err < 0)
+                                       goto out;
+                       }
+               } else {
+                       /* Valid message (not checking for MULTIPART bit to
+                        * get along with broken kernels. NL_SKIP has no
+                        * effect on this.  */
+                       if (cb->cb_set[NL_CB_VALID]) {
+                               err = nl_cb_call(cb, NL_CB_VALID, msg);
+                               if (err == NL_SKIP)
+                                       goto skip;
+                               else if (err == NL_EXIT || err < 0)
+                                       goto out;
+                       }
+               }
+skip:
+               hdr = nlmsg_next(hdr, &n);
+       }
+       
+       nlmsg_free(msg);
+       free(buf);
+       free(creds);
+       buf = NULL;
+       msg = NULL;
+       creds = NULL;
+
+       /* Multipart message not yet complete, continue reading */
+       goto continue_reading;
+
+out:
+       nlmsg_free(msg);
+       free(buf);
+       free(creds);
+
+       return err;
+}
+
+/**
+ * Receive a set of message from a netlink socket using handlers in nl_handle.
+ * @arg handle         netlink handle
+ *
+ * Calls nl_recvmsgs() with the handlers configured in the netlink handle.
+ */
+int nl_recvmsgs_def(struct nl_handle *handle)
+{
+       if (handle->h_cb->cb_recvmsgs_ow)
+               return handle->h_cb->cb_recvmsgs_ow(handle, handle->h_cb);
+       else
+               return nl_recvmsgs(handle, handle->h_cb);
+}
+
+static int ack_wait_handler(struct nl_msg *msg, void *arg)
+{
+       return NL_EXIT;
+}
+
+/**
+ * Wait for ACK.
+ * @arg handle         netlink handle
+ * @pre The netlink socket must be in blocking state.
+ *
+ * Waits until an ACK is received for the latest not yet acknowledged
+ * netlink message.
+ */
+int nl_wait_for_ack(struct nl_handle *handle)
+{
+       int err;
+       struct nl_cb *cb = nl_cb_clone(nl_handle_get_cb(handle));
+
+       nl_cb_set(cb, NL_CB_ACK, NL_CB_CUSTOM, ack_wait_handler, NULL);
+
+       err = nl_recvmsgs(handle, cb);
+       nl_cb_destroy(cb);
+
+       return err;
+}
+
+/** @} */
+
+/**
+ * @name Netlink Family Translations
+ * @{
+ */
+
+static struct trans_tbl nlfamilies[] = {
+       __ADD(NETLINK_ROUTE,route)
+       __ADD(NETLINK_W1,w1)
+       __ADD(NETLINK_USERSOCK,usersock)
+       __ADD(NETLINK_FIREWALL,firewall)
+       __ADD(NETLINK_INET_DIAG,inetdiag)
+       __ADD(NETLINK_NFLOG,nflog)
+       __ADD(NETLINK_XFRM,xfrm)
+       __ADD(NETLINK_SELINUX,selinux)
+       __ADD(NETLINK_ISCSI,iscsi)
+       __ADD(NETLINK_AUDIT,audit)
+       __ADD(NETLINK_FIB_LOOKUP,fib_lookup)
+       __ADD(NETLINK_CONNECTOR,connector)
+       __ADD(NETLINK_NETFILTER,netfilter)
+       __ADD(NETLINK_IP6_FW,ip6_fw)
+       __ADD(NETLINK_DNRTMSG,dnrtmsg)
+       __ADD(NETLINK_KOBJECT_UEVENT,kobject_uevent)
+       __ADD(NETLINK_GENERIC,generic)
+};
+
+/**
+ * Convert netlink family to character string.
+ * @arg family         Netlink family.
+ * @arg buf            Destination buffer.
+ * @arg size           Size of destination buffer.
+ *
+ * Converts a netlink family to a character string and stores it in
+ * the specified destination buffer.
+ *
+ * @return The destination buffer or the family encoded in hexidecimal
+ *         form if no match was found.
+ */
+char * nl_nlfamily2str(int family, char *buf, size_t size)
+{
+       return __type2str(family, buf, size, nlfamilies,
+                         ARRAY_SIZE(nlfamilies));
+}
+
+/**
+ * Convert character string to netlink family.
+ * @arg name           Name of netlink family.
+ *
+ * Converts the provided character string specifying a netlink
+ * family to the corresponding numeric value.
+ *
+ * @return Numeric netlink family or a negative value if no match was found.
+ */
+int nl_str2nlfamily(const char *name)
+{
+       return __str2type(name, nlfamilies, ARRAY_SIZE(nlfamilies));
+}
+
+/** @} */
+/** @} */