/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <config.h>
#include "netlink-socket.h"
-#include <assert.h>
#include <errno.h>
#include <inttypes.h>
#include <stdlib.h>
#include <sys/types.h>
+#include <sys/uio.h>
#include <unistd.h>
#include "coverage.h"
#include "dynamic-string.h"
+#include "hash.h"
+#include "hmap.h"
#include "netlink.h"
#include "netlink-protocol.h"
+#include "odp-netlink.h"
#include "ofpbuf.h"
+#include "ovs-thread.h"
#include "poll-loop.h"
+#include "seq.h"
#include "socket-util.h"
-#include "stress.h"
+#include "util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(netlink_socket);
COVERAGE_DEFINE(netlink_overflow);
COVERAGE_DEFINE(netlink_received);
-COVERAGE_DEFINE(netlink_recv_retry);
-COVERAGE_DEFINE(netlink_send);
+COVERAGE_DEFINE(netlink_recv_jumbo);
COVERAGE_DEFINE(netlink_sent);
/* Linux header file confusion causes this to be undefined. */
#define SOL_NETLINK 270
#endif
+#ifdef _WIN32
+static struct ovs_mutex portid_mutex = OVS_MUTEX_INITIALIZER;
+static uint32_t g_last_portid = 0;
+
+/* Port IDs must be unique! */
+static uint32_t
+portid_next(void)
+ OVS_GUARDED_BY(portid_mutex)
+{
+ g_last_portid++;
+ return g_last_portid;
+}
+#endif /* _WIN32 */
+
/* A single (bad) Netlink message can in theory dump out many, many log
* messages, so the burst size is set quite high here to avoid missing useful
* information. Also, at high logging levels we log *all* Netlink messages. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
+static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
static void log_nlmsg(const char *function, int error,
const void *message, size_t size, int protocol);
+#ifdef _WIN32
+static int get_sock_pid_from_kernel(struct nl_sock *sock);
+#endif
\f
/* Netlink sockets. */
-struct nl_sock
-{
+struct nl_sock {
+#ifdef _WIN32
+ HANDLE handle;
+ OVERLAPPED overlapped;
+ DWORD read_ioctl;
+#else
int fd;
+#endif
+ uint32_t next_seq;
uint32_t pid;
int protocol;
- bool any_groups;
- struct nl_dump *dump;
+ unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
};
-static int alloc_pid(uint32_t *);
-static void free_pid(uint32_t);
-static int nl_sock_cow__(struct nl_sock *);
+/* Compile-time limit on iovecs, so that we can allocate a maximum-size array
+ * of iovecs on the stack. */
+#define MAX_IOVS 128
+
+/* Maximum number of iovecs that may be passed to sendmsg, capped at a
+ * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
+ *
+ * Initialized by nl_sock_create(). */
+static int max_iovs;
+
+static int nl_pool_alloc(int protocol, struct nl_sock **sockp);
+static void nl_pool_release(struct nl_sock *);
/* Creates a new netlink socket for the given netlink 'protocol'
* (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
- * new socket if successful, otherwise returns a positive errno value. */
+ * new socket if successful, otherwise returns a positive errno value. */
int
nl_sock_create(int protocol, struct nl_sock **sockp)
{
+ static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
struct nl_sock *sock;
+#ifndef _WIN32
struct sockaddr_nl local, remote;
+#endif
+ socklen_t local_size;
+ int rcvbuf;
int retval = 0;
+ if (ovsthread_once_start(&once)) {
+ int save_errno = errno;
+ errno = 0;
+
+ max_iovs = sysconf(_SC_UIO_MAXIOV);
+ if (max_iovs < _XOPEN_IOV_MAX) {
+ if (max_iovs == -1 && errno) {
+ VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno));
+ }
+ max_iovs = _XOPEN_IOV_MAX;
+ } else if (max_iovs > MAX_IOVS) {
+ max_iovs = MAX_IOVS;
+ }
+
+ errno = save_errno;
+ ovsthread_once_done(&once);
+ }
+
*sockp = NULL;
- sock = malloc(sizeof *sock);
- if (sock == NULL) {
- return ENOMEM;
+ sock = xmalloc(sizeof *sock);
+
+#ifdef _WIN32
+ sock->handle = CreateFile(OVS_DEVICE_NAME_USER,
+ GENERIC_READ | GENERIC_WRITE,
+ FILE_SHARE_READ | FILE_SHARE_WRITE,
+ NULL, OPEN_EXISTING,
+ FILE_FLAG_OVERLAPPED, NULL);
+
+ if (sock->handle == INVALID_HANDLE_VALUE) {
+ VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
+ goto error;
}
+ memset(&sock->overlapped, 0, sizeof sock->overlapped);
+ sock->overlapped.hEvent = CreateEvent(NULL, FALSE, FALSE, NULL);
+ if (sock->overlapped.hEvent == NULL) {
+ VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
+ goto error;
+ }
+ /* Initialize the type/ioctl to Generic */
+ sock->read_ioctl = OVS_IOCTL_READ;
+#else
sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
if (sock->fd < 0) {
- VLOG_ERR("fcntl: %s", strerror(errno));
+ VLOG_ERR("fcntl: %s", ovs_strerror(errno));
goto error;
}
+#endif
+
sock->protocol = protocol;
- sock->any_groups = false;
+ sock->next_seq = 1;
- retval = alloc_pid(&sock->pid);
- if (retval) {
+ rcvbuf = 1024 * 1024;
+#ifdef _WIN32
+ sock->rcvbuf = rcvbuf;
+ retval = get_sock_pid_from_kernel(sock);
+ if (retval != 0) {
goto error;
}
+#else
+ if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
+ &rcvbuf, sizeof rcvbuf)) {
+ /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
+ * Warn only if the failure is therefore unexpected. */
+ if (errno != EPERM) {
+ VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
+ "(%s)", rcvbuf, ovs_strerror(errno));
+ }
+ }
- /* Bind local address as our selected pid. */
- memset(&local, 0, sizeof local);
- local.nl_family = AF_NETLINK;
- local.nl_pid = sock->pid;
- if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) {
- VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno));
- goto error_free_pid;
+ retval = get_socket_rcvbuf(sock->fd);
+ if (retval < 0) {
+ retval = -retval;
+ goto error;
}
+ sock->rcvbuf = retval;
- /* Bind remote address as the kernel (pid 0). */
+ /* Connect to kernel (pid 0) as remote address. */
memset(&remote, 0, sizeof remote);
remote.nl_family = AF_NETLINK;
remote.nl_pid = 0;
if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
- VLOG_ERR("connect(0): %s", strerror(errno));
- goto error_free_pid;
+ VLOG_ERR("connect(0): %s", ovs_strerror(errno));
+ goto error;
+ }
+
+ /* Obtain pid assigned by kernel. */
+ local_size = sizeof local;
+ if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
+ VLOG_ERR("getsockname: %s", ovs_strerror(errno));
+ goto error;
+ }
+ if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
+ VLOG_ERR("getsockname returned bad Netlink name");
+ retval = EINVAL;
+ goto error;
}
+ sock->pid = local.nl_pid;
+#endif
*sockp = sock;
return 0;
-error_free_pid:
- free_pid(sock->pid);
error:
if (retval == 0) {
retval = errno;
retval = EINVAL;
}
}
+#ifdef _WIN32
+ if (sock->overlapped.hEvent) {
+ CloseHandle(sock->overlapped.hEvent);
+ }
+ if (sock->handle != INVALID_HANDLE_VALUE) {
+ CloseHandle(sock->handle);
+ }
+#else
if (sock->fd >= 0) {
close(sock->fd);
}
+#endif
free(sock);
return retval;
}
nl_sock_destroy(struct nl_sock *sock)
{
if (sock) {
- if (sock->dump) {
- sock->dump = NULL;
- } else {
- close(sock->fd);
- free_pid(sock->pid);
- free(sock);
+#ifdef _WIN32
+ if (sock->overlapped.hEvent) {
+ CloseHandle(sock->overlapped.hEvent);
}
+ CloseHandle(sock->handle);
+#else
+ close(sock->fd);
+#endif
+ free(sock);
}
}
+#ifdef _WIN32
+/* Reads the pid for 'sock' generated in the kernel datapath. The function
+ * follows a transaction semantic. Eventually this function should call into
+ * nl_transact. */
+static int
+get_sock_pid_from_kernel(struct nl_sock *sock)
+{
+ struct nl_transaction txn;
+ struct ofpbuf request;
+ uint64_t request_stub[128];
+ struct ofpbuf reply;
+ uint64_t reply_stub[128];
+ struct ovs_header *ovs_header;
+ struct nlmsghdr *nlmsg;
+ uint32_t seq;
+ int retval;
+ DWORD bytes;
+ int ovs_msg_size = sizeof (struct nlmsghdr) + sizeof (struct genlmsghdr) +
+ sizeof (struct ovs_header);
+
+ ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
+ txn.request = &request;
+ ofpbuf_use_stub(&reply, reply_stub, sizeof reply_stub);
+ txn.reply = &reply;
+
+ seq = nl_sock_allocate_seq(sock, 1);
+ nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
+ OVS_CTRL_CMD_WIN_GET_PID, OVS_WIN_CONTROL_VERSION);
+ nlmsg = nl_msg_nlmsghdr(txn.request);
+ nlmsg->nlmsg_seq = seq;
+
+ ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
+ ovs_header->dp_ifindex = 0;
+ ovs_header = ofpbuf_put_uninit(&reply, ovs_msg_size);
+
+ if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
+ ofpbuf_data(txn.request), ofpbuf_size(txn.request),
+ ofpbuf_data(txn.reply), ofpbuf_size(txn.reply),
+ &bytes, NULL)) {
+ retval = EINVAL;
+ goto done;
+ } else {
+ if (bytes < ovs_msg_size) {
+ retval = EINVAL;
+ goto done;
+ }
+
+ nlmsg = nl_msg_nlmsghdr(txn.reply);
+ if (nlmsg->nlmsg_seq != seq) {
+ retval = EINVAL;
+ goto done;
+ }
+ sock->pid = nlmsg->nlmsg_pid;
+ }
+ retval = 0;
+
+done:
+ ofpbuf_uninit(&request);
+ ofpbuf_uninit(&reply);
+ return retval;
+}
+#endif /* _WIN32 */
+
+#ifdef _WIN32
+static int __inline
+nl_sock_mcgroup(struct nl_sock *sock, unsigned int multicast_group, bool join)
+{
+ struct ofpbuf request;
+ uint64_t request_stub[128];
+ struct ovs_header *ovs_header;
+ struct nlmsghdr *nlmsg;
+ int error;
+
+ ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
+
+ nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
+ OVS_CTRL_CMD_MC_SUBSCRIBE_REQ,
+ OVS_WIN_CONTROL_VERSION);
+
+ ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
+ ovs_header->dp_ifindex = 0;
+
+ nl_msg_put_u32(&request, OVS_NL_ATTR_MCAST_GRP, multicast_group);
+ nl_msg_put_u8(&request, OVS_NL_ATTR_MCAST_JOIN, join ? 1 : 0);
+
+ error = nl_sock_send(sock, &request, true);
+ ofpbuf_uninit(&request);
+ return error;
+}
+#endif
/* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
* successful, otherwise a positive errno value.
*
+ * A socket that is subscribed to a multicast group that receives asynchronous
+ * notifications must not be used for Netlink transactions or dumps, because
+ * transactions and dumps can cause notifications to be lost.
+ *
* Multicast group numbers are always positive.
*
* It is not an error to attempt to join a multicast group to which a socket
int
nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
{
- int error = nl_sock_cow__(sock);
+#ifdef _WIN32
+ /* Set the socket type as a "multicast" socket */
+ sock->read_ioctl = OVS_IOCTL_READ_EVENT;
+ int error = nl_sock_mcgroup(sock, multicast_group, true);
if (error) {
- return error;
+ sock->read_ioctl = OVS_IOCTL_READ;
+ VLOG_WARN("could not join multicast group %u (%s)",
+ multicast_group, ovs_strerror(errno));
+ return errno;
}
+#else
if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
&multicast_group, sizeof multicast_group) < 0) {
VLOG_WARN("could not join multicast group %u (%s)",
- multicast_group, strerror(errno));
+ multicast_group, ovs_strerror(errno));
return errno;
}
- sock->any_groups = true;
+#endif
return 0;
}
int
nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
{
- assert(!sock->dump);
+#ifdef _WIN32
+ int error = nl_sock_mcgroup(sock, multicast_group, false);
+ if (error) {
+ VLOG_WARN("could not leave multicast group %u (%s)",
+ multicast_group, ovs_strerror(errno));
+ return errno;
+ }
+ sock->read_ioctl = OVS_IOCTL_READ;
+#else
if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
&multicast_group, sizeof multicast_group) < 0) {
VLOG_WARN("could not leave multicast group %u (%s)",
- multicast_group, strerror(errno));
+ multicast_group, ovs_strerror(errno));
return errno;
}
+#endif
return 0;
}
static int
-nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
+nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
+ uint32_t nlmsg_seq, bool wait)
{
struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
int error;
- nlmsg->nlmsg_len = msg->size;
+ nlmsg->nlmsg_len = ofpbuf_size(msg);
+ nlmsg->nlmsg_seq = nlmsg_seq;
nlmsg->nlmsg_pid = sock->pid;
do {
int retval;
- retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
+#ifdef _WIN32
+ DWORD bytes;
+
+ if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
+ ofpbuf_data(msg), ofpbuf_size(msg), NULL, 0,
+ &bytes, NULL)) {
+ retval = -1;
+ /* XXX: Map to a more appropriate error based on GetLastError(). */
+ errno = EINVAL;
+ } else {
+ retval = ofpbuf_size(msg);
+ }
+#else
+ retval = send(sock->fd, ofpbuf_data(msg), ofpbuf_size(msg),
+ wait ? 0 : MSG_DONTWAIT);
+#endif
error = retval < 0 ? errno : 0;
} while (error == EINTR);
- log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
+ log_nlmsg(__func__, error, ofpbuf_data(msg), ofpbuf_size(msg), sock->protocol);
if (!error) {
COVERAGE_INC(netlink_sent);
}
}
/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
- * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, and
- * nlmsg_pid will be set to 'sock''s pid, before the message is sent.
+ * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid
+ * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
+ * sequence number, before the message is sent.
*
* Returns 0 if successful, otherwise a positive errno value. If
* 'wait' is true, then the send will wait until buffer space is ready;
int
nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
{
- int error = nl_sock_cow__(sock);
- if (error) {
- return error;
- }
- return nl_sock_send__(sock, msg, wait);
+ return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
}
-/* Tries to send the 'n_iov' chunks of data in 'iov' to the kernel on 'sock' as
- * a single Netlink message. (The message must be fully formed and not require
- * finalization of its nlmsg_len or nlmsg_pid fields.)
+/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
+ * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid
+ * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
+ * 'nlmsg_seq', before the message is sent.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. If
+ * 'wait' is true, then the send will wait until buffer space is ready;
+ * otherwise, returns EAGAIN if the 'sock' send buffer is full.
*
- * Returns 0 if successful, otherwise a positive errno value. If 'wait' is
- * true, then the send will wait until buffer space is ready; otherwise,
- * returns EAGAIN if the 'sock' send buffer is full. */
+ * This function is suitable for sending a reply to a request that was received
+ * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
int
-nl_sock_sendv(struct nl_sock *sock, const struct iovec iov[], size_t n_iov,
- bool wait)
+nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
+ uint32_t nlmsg_seq, bool wait)
+{
+ return nl_sock_send__(sock, msg, nlmsg_seq, wait);
+}
+
+static int
+nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
{
+ /* We can't accurately predict the size of the data to be received. The
+ * caller is supposed to have allocated enough space in 'buf' to handle the
+ * "typical" case. To handle exceptions, we make available enough space in
+ * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
+ * figure since that's the maximum length of a Netlink attribute). */
+ struct nlmsghdr *nlmsghdr;
+ uint8_t tail[65536];
+ struct iovec iov[2];
struct msghdr msg;
+ ssize_t retval;
int error;
- COVERAGE_INC(netlink_send);
+ ovs_assert(buf->allocated >= sizeof *nlmsghdr);
+ ofpbuf_clear(buf);
+
+ iov[0].iov_base = ofpbuf_base(buf);
+ iov[0].iov_len = buf->allocated;
+ iov[1].iov_base = tail;
+ iov[1].iov_len = sizeof tail;
+
memset(&msg, 0, sizeof msg);
- msg.msg_iov = (struct iovec *) iov;
- msg.msg_iovlen = n_iov;
+ msg.msg_iov = iov;
+ msg.msg_iovlen = 2;
+
+ /* Receive a Netlink message from the kernel.
+ *
+ * This works around a kernel bug in which the kernel returns an error code
+ * as if it were the number of bytes read. It doesn't actually modify
+ * anything in the receive buffer in that case, so we can initialize the
+ * Netlink header with an impossible message length and then, upon success,
+ * check whether it changed. */
+ nlmsghdr = ofpbuf_base(buf);
do {
- int retval;
- retval = sendmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
- error = retval < 0 ? errno : 0;
+ nlmsghdr->nlmsg_len = UINT32_MAX;
+#ifdef _WIN32
+ DWORD bytes;
+ if (!DeviceIoControl(sock->handle, sock->read_ioctl,
+ NULL, 0, tail, sizeof tail, &bytes, NULL)) {
+ retval = -1;
+ errno = EINVAL;
+ } else {
+ retval = bytes;
+ if (retval == 0) {
+ retval = -1;
+ errno = EAGAIN;
+ } else {
+ if (retval >= buf->allocated) {
+ ofpbuf_reinit(buf, retval);
+ }
+ memcpy(ofpbuf_data(buf), tail, retval);
+ ofpbuf_set_size(buf, retval);
+ }
+ }
+#else
+ retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
+#endif
+ error = (retval < 0 ? errno
+ : retval == 0 ? ECONNRESET /* not possible? */
+ : nlmsghdr->nlmsg_len != UINT32_MAX ? 0
+ : retval);
} while (error == EINTR);
- if (error != EAGAIN) {
- log_nlmsg(__func__, error, iov[0].iov_base, iov[0].iov_len,
- sock->protocol);
- if (!error) {
- COVERAGE_INC(netlink_sent);
+ if (error) {
+ if (error == ENOBUFS) {
+ /* Socket receive buffer overflow dropped one or more messages that
+ * the kernel tried to send to us. */
+ COVERAGE_INC(netlink_overflow);
}
+ return error;
}
- return error;
-}
-
-/* This stress option is useful for testing that OVS properly tolerates
- * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
- * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
- * reply to a request. They can also occur if messages arrive on a multicast
- * channel faster than OVS can process them. */
-STRESS_OPTION(
- netlink_overflow, "simulate netlink socket receive buffer overflow",
- 5, 1, -1, 100);
-
-static int
-nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
-{
- uint8_t tmp;
- ssize_t bufsize = 2048;
- ssize_t nbytes, nbytes2;
- struct ofpbuf *buf;
- struct nlmsghdr *nlmsghdr;
- struct iovec iov;
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &iov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = 0
- };
- buf = ofpbuf_new(bufsize);
- *bufp = NULL;
-
-try_again:
- /* Attempt to read the message. We don't know the size of the data
- * yet, so we take a guess at 2048. If we're wrong, we keep trying
- * and doubling the buffer size each time.
- */
- nlmsghdr = ofpbuf_put_uninit(buf, bufsize);
- iov.iov_base = nlmsghdr;
- iov.iov_len = bufsize;
- do {
- nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK);
- } while (nbytes < 0 && errno == EINTR);
- if (nbytes < 0) {
- ofpbuf_delete(buf);
- return errno;
- }
if (msg.msg_flags & MSG_TRUNC) {
- COVERAGE_INC(netlink_recv_retry);
- bufsize *= 2;
- ofpbuf_reinit(buf, bufsize);
- goto try_again;
+ VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)",
+ sizeof tail);
+ return E2BIG;
}
- buf->size = nbytes;
- /* We successfully read the message, so recv again to clear the queue */
- iov.iov_base = &tmp;
- iov.iov_len = 1;
- do {
- nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT);
- } while (nbytes2 < 0 && errno == EINTR);
- if (nbytes2 < 0) {
- if (errno == ENOBUFS) {
- /* The kernel is notifying us that a message it tried to send to us
- * was dropped. We have to pass this along to the caller in case
- * it wants to retry a request. So kill the buffer, which we can
- * re-read next time. */
- COVERAGE_INC(netlink_overflow);
- ofpbuf_delete(buf);
- return ENOBUFS;
- } else {
- VLOG_ERR_RL(&rl, "failed to remove nlmsg from socket: %s\n",
- strerror(errno));
- }
- }
- if (nbytes < sizeof *nlmsghdr
+ if (retval < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len < sizeof *nlmsghdr
- || nlmsghdr->nlmsg_len > nbytes) {
- VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)",
- bufsize, NLMSG_HDRLEN);
- ofpbuf_delete(buf);
+ || nlmsghdr->nlmsg_len > retval) {
+ VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE" bytes < %"PRIuSIZE")",
+ retval, sizeof *nlmsghdr);
return EPROTO;
}
-
- if (STRESS(netlink_overflow)) {
- ofpbuf_delete(buf);
- return ENOBUFS;
+#ifndef _WIN32
+ ofpbuf_set_size(buf, MIN(retval, buf->allocated));
+ if (retval > buf->allocated) {
+ COVERAGE_INC(netlink_recv_jumbo);
+ ofpbuf_put(buf, tail, retval - buf->allocated);
}
+#endif
- *bufp = buf;
- log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
+ log_nlmsg(__func__, 0, ofpbuf_data(buf), ofpbuf_size(buf), sock->protocol);
COVERAGE_INC(netlink_received);
return 0;
}
-/* Tries to receive a netlink message from the kernel on 'sock'. If
- * successful, stores the received message into '*bufp' and returns 0. The
- * caller is responsible for destroying the message with ofpbuf_delete(). On
- * failure, returns a positive errno value and stores a null pointer into
- * '*bufp'.
+/* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
+ * 'wait' is true, waits for a message to be ready. Otherwise, fails with
+ * EAGAIN if the 'sock' receive buffer is empty.
+ *
+ * The caller must have initialized 'buf' with an allocation of at least
+ * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
+ * space for a "typical" message.
*
- * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
- * returns EAGAIN if the 'sock' receive buffer is empty. */
+ * On success, returns 0 and replaces 'buf''s previous content by the received
+ * message. This function expands 'buf''s allocated memory, as necessary, to
+ * hold the actual size of the received message.
+ *
+ * On failure, returns a positive errno value and clears 'buf' to zero length.
+ * 'buf' retains its previous memory allocation.
+ *
+ * Regardless of success or failure, this function resets 'buf''s headroom to
+ * 0. */
int
-nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
+nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
{
- int error = nl_sock_cow__(sock);
- if (error) {
- return error;
+ return nl_sock_recv__(sock, buf, wait);
+}
+
+static void
+nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
+ int error)
+{
+ size_t i;
+
+ for (i = 0; i < n; i++) {
+ struct nl_transaction *txn = transactions[i];
+
+ txn->error = error;
+ if (txn->reply) {
+ ofpbuf_clear(txn->reply);
+ }
}
- return nl_sock_recv__(sock, bufp, wait);
}
-/* Sends 'request' to the kernel via 'sock' and waits for a response. If
- * successful, returns 0. On failure, returns a positive errno value.
- *
- * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
- * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
- * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
- * reply, if any, is discarded.
- *
- * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
- * be set to 'sock''s pid, before the message is sent. NLM_F_ACK will be set
- * in nlmsg_flags.
- *
- * The caller is responsible for destroying 'request'.
- *
- * Bare Netlink is an unreliable transport protocol. This function layers
- * reliable delivery and reply semantics on top of bare Netlink.
- *
- * In Netlink, sending a request to the kernel is reliable enough, because the
- * kernel will tell us if the message cannot be queued (and we will in that
- * case put it on the transmit queue and wait until it can be delivered).
- *
- * Receiving the reply is the real problem: if the socket buffer is full when
- * the kernel tries to send the reply, the reply will be dropped. However, the
- * kernel sets a flag that a reply has been dropped. The next call to recv
- * then returns ENOBUFS. We can then re-send the request.
- *
- * Caveats:
- *
- * 1. Netlink depends on sequence numbers to match up requests and
- * replies. The sender of a request supplies a sequence number, and
- * the reply echos back that sequence number.
- *
- * This is fine, but (1) some kernel netlink implementations are
- * broken, in that they fail to echo sequence numbers and (2) this
- * function will drop packets with non-matching sequence numbers, so
- * that only a single request can be usefully transacted at a time.
- *
- * 2. Resending the request causes it to be re-executed, so the request
- * needs to be idempotent.
- */
-int
-nl_sock_transact(struct nl_sock *sock,
- const struct ofpbuf *request, struct ofpbuf **replyp)
+static int
+nl_sock_transact_multiple__(struct nl_sock *sock,
+ struct nl_transaction **transactions, size_t n,
+ size_t *done)
{
- uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
- struct nlmsghdr *nlmsghdr;
- struct ofpbuf *reply;
- int retval;
+ uint64_t tmp_reply_stub[1024 / 8];
+ struct nl_transaction tmp_txn;
+ struct ofpbuf tmp_reply;
- if (replyp) {
- *replyp = NULL;
+ uint32_t base_seq;
+ struct iovec iovs[MAX_IOVS];
+ struct msghdr msg;
+ int error;
+ int i;
+
+ base_seq = nl_sock_allocate_seq(sock, n);
+ *done = 0;
+ for (i = 0; i < n; i++) {
+ struct nl_transaction *txn = transactions[i];
+ struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
+
+ nlmsg->nlmsg_len = ofpbuf_size(txn->request);
+ nlmsg->nlmsg_seq = base_seq + i;
+ nlmsg->nlmsg_pid = sock->pid;
+
+ iovs[i].iov_base = ofpbuf_data(txn->request);
+ iovs[i].iov_len = ofpbuf_size(txn->request);
}
- /* Ensure that we get a reply even if this message doesn't ordinarily call
- * for one. */
- nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK;
+#ifndef _WIN32
+ memset(&msg, 0, sizeof msg);
+ msg.msg_iov = iovs;
+ msg.msg_iovlen = n;
+ do {
+ error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
+ } while (error == EINTR);
+
+ for (i = 0; i < n; i++) {
+ struct nl_transaction *txn = transactions[i];
-send:
- retval = nl_sock_send(sock, request, true);
- if (retval) {
- return retval;
+ log_nlmsg(__func__, error, ofpbuf_data(txn->request),
+ ofpbuf_size(txn->request), sock->protocol);
+ }
+ if (!error) {
+ COVERAGE_ADD(netlink_sent, n);
}
-recv:
- retval = nl_sock_recv(sock, &reply, true);
- if (retval) {
- if (retval == ENOBUFS) {
- COVERAGE_INC(netlink_overflow);
- VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
- goto send;
+ if (error) {
+ return error;
+ }
+
+ ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
+ tmp_txn.request = NULL;
+ tmp_txn.reply = &tmp_reply;
+ tmp_txn.error = 0;
+ while (n > 0) {
+ struct nl_transaction *buf_txn, *txn;
+ uint32_t seq;
+
+ /* Find a transaction whose buffer we can use for receiving a reply.
+ * If no such transaction is left, use tmp_txn. */
+ buf_txn = &tmp_txn;
+ for (i = 0; i < n; i++) {
+ if (transactions[i]->reply) {
+ buf_txn = transactions[i];
+ break;
+ }
+ }
+
+ /* Receive a reply. */
+ error = nl_sock_recv__(sock, buf_txn->reply, false);
+ if (error) {
+ if (error == EAGAIN) {
+ nl_sock_record_errors__(transactions, n, 0);
+ *done += n;
+ error = 0;
+ }
+ break;
+ }
+
+ /* Match the reply up with a transaction. */
+ seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
+ if (seq < base_seq || seq >= base_seq + n) {
+ VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
+ continue;
+ }
+ i = seq - base_seq;
+ txn = transactions[i];
+
+ /* Fill in the results for 'txn'. */
+ if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
+ if (txn->reply) {
+ ofpbuf_clear(txn->reply);
+ }
+ if (txn->error) {
+ VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
+ error, ovs_strerror(txn->error));
+ }
} else {
- return retval;
+ txn->error = 0;
+ if (txn->reply && txn != buf_txn) {
+ /* Swap buffers. */
+ struct ofpbuf *reply = buf_txn->reply;
+ buf_txn->reply = txn->reply;
+ txn->reply = reply;
+ }
}
+
+ /* Fill in the results for transactions before 'txn'. (We have to do
+ * this after the results for 'txn' itself because of the buffer swap
+ * above.) */
+ nl_sock_record_errors__(transactions, i, 0);
+
+ /* Advance. */
+ *done += i + 1;
+ transactions += i + 1;
+ n -= i + 1;
+ base_seq += i + 1;
}
- nlmsghdr = nl_msg_nlmsghdr(reply);
- if (seq != nlmsghdr->nlmsg_seq) {
- VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
- nl_msg_nlmsghdr(reply)->nlmsg_seq, seq);
- ofpbuf_delete(reply);
- goto recv;
- }
+ ofpbuf_uninit(&tmp_reply);
+#else
+ error = 0;
+ for (i = 0; i < n; i++) {
+ DWORD reply_len;
+ uint8_t tail[65536];
+ struct nl_transaction *txn = transactions[i];
+ struct nlmsghdr *request_nlmsg, *reply_nlmsg;
+
+ if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
+ ofpbuf_data(txn->request),
+ ofpbuf_size(txn->request),
+ txn->reply ? tail : 0,
+ txn->reply ? sizeof tail : 0,
+ &reply_len, NULL)) {
+ /* XXX: Map to a more appropriate error. */
+ error = EINVAL;
+ break;
+ }
- /* If the reply is an error, discard the reply and return the error code.
- *
- * Except: if the reply is just an acknowledgement (error code of 0), and
- * the caller is interested in the reply (replyp != NULL), pass the reply
- * up to the caller. Otherwise the caller will get a return value of 0
- * and null '*replyp', which makes unwary callers likely to segfault. */
- if (nl_msg_nlmsgerr(reply, &retval) && (retval || !replyp)) {
- ofpbuf_delete(reply);
- if (retval) {
- VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
- retval, strerror(retval));
+ if (txn->reply) {
+ if (reply_len < sizeof *reply_nlmsg) {
+ VLOG_DBG_RL(&rl, "insufficient length of reply %#"PRIu32,
+ reply_len);
+ break;
+ }
+
+ /* Validate the sequence number in the reply. */
+ request_nlmsg = nl_msg_nlmsghdr(txn->request);
+ reply_nlmsg = (struct nlmsghdr *)tail;
+
+ if (request_nlmsg->nlmsg_seq != reply_nlmsg->nlmsg_seq) {
+ ovs_assert(request_nlmsg->nlmsg_seq == reply_nlmsg->nlmsg_seq);
+ VLOG_DBG_RL(&rl, "mismatched seq request %#"PRIx32
+ ", reply %#"PRIx32, request_nlmsg->nlmsg_seq,
+ reply_nlmsg->nlmsg_seq);
+ break;
+ }
+
+ /* If reply was expected, verify if there was indeed a reply
+ * received. */
+ if (reply_len == 0) {
+ nl_sock_record_errors__(transactions, n, 0);
+ VLOG_DBG_RL(&rl, "reply not seen when expected seq %#"PRIx32,
+ request_nlmsg->nlmsg_seq);
+ break;
+ }
+
+ /* Copy the reply to the buffer specified by the caller. */
+ if (reply_len > txn->reply->allocated) {
+ ofpbuf_reinit(txn->reply, reply_len);
+ }
+ memcpy(ofpbuf_data(txn->reply), tail, reply_len);
+ ofpbuf_set_size(txn->reply, reply_len);
+
+ /* Handle errors embedded within the netlink message. */
+ if (nl_msg_nlmsgerr(txn->reply, &txn->error)) {
+ if (txn->reply) {
+ ofpbuf_clear(txn->reply);
+ }
+ if (txn->error) {
+ VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
+ error, ovs_strerror(txn->error));
+ }
+ } else {
+ txn->error = 0;
+ }
}
- return retval != EAGAIN ? retval : EPROTO;
+
+ /* Count the number of successful transactions. */
+ (*done)++;
}
- if (replyp) {
- *replyp = reply;
- } else {
- ofpbuf_delete(reply);
+ if (!error) {
+ COVERAGE_ADD(netlink_sent, n);
}
- return 0;
+#endif
+
+ return error;
}
-/* Drain all the messages currently in 'sock''s receive queue. */
-int
-nl_sock_drain(struct nl_sock *sock)
+static void
+nl_sock_transact_multiple(struct nl_sock *sock,
+ struct nl_transaction **transactions, size_t n)
{
- int error = nl_sock_cow__(sock);
- if (error) {
- return error;
+ int max_batch_count;
+ int error;
+
+ if (!n) {
+ return;
+ }
+
+ /* In theory, every request could have a 64 kB reply. But the default and
+ * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
+ * be a bit below 128 kB, so that would only allow a single message in a
+ * "batch". So we assume that replies average (at most) 4 kB, which allows
+ * a good deal of batching.
+ *
+ * In practice, most of the requests that we batch either have no reply at
+ * all or a brief reply. */
+ max_batch_count = MAX(sock->rcvbuf / 4096, 1);
+ max_batch_count = MIN(max_batch_count, max_iovs);
+
+ while (n > 0) {
+ size_t count, bytes;
+ size_t done;
+
+ /* Batch up to 'max_batch_count' transactions. But cap it at about a
+ * page of requests total because big skbuffs are expensive to
+ * allocate in the kernel. */
+#if defined(PAGESIZE)
+ enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
+#else
+ enum { MAX_BATCH_BYTES = 4096 - 512 };
+#endif
+ bytes = ofpbuf_size(transactions[0]->request);
+ for (count = 1; count < n && count < max_batch_count; count++) {
+ if (bytes + ofpbuf_size(transactions[count]->request) > MAX_BATCH_BYTES) {
+ break;
+ }
+ bytes += ofpbuf_size(transactions[count]->request);
+ }
+
+ error = nl_sock_transact_multiple__(sock, transactions, count, &done);
+ transactions += done;
+ n -= done;
+
+ if (error == ENOBUFS) {
+ VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
+ } else if (error) {
+ VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error));
+ nl_sock_record_errors__(transactions, n, error);
+ }
}
- return drain_rcvbuf(sock->fd);
}
-/* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
- * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
- * old fd over to the dump. */
static int
-nl_sock_cow__(struct nl_sock *sock)
+nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
+ struct ofpbuf **replyp)
{
- struct nl_sock *copy;
- uint32_t tmp_pid;
- int tmp_fd;
- int error;
-
- if (!sock->dump) {
- return 0;
- }
+ struct nl_transaction *transactionp;
+ struct nl_transaction transaction;
- error = nl_sock_clone(sock, ©);
- if (error) {
- return error;
- }
+ transaction.request = CONST_CAST(struct ofpbuf *, request);
+ transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
+ transactionp = &transaction;
- tmp_fd = sock->fd;
- sock->fd = copy->fd;
- copy->fd = tmp_fd;
+ nl_sock_transact_multiple(sock, &transactionp, 1);
- tmp_pid = sock->pid;
- sock->pid = copy->pid;
- copy->pid = tmp_pid;
+ if (replyp) {
+ if (transaction.error) {
+ ofpbuf_delete(transaction.reply);
+ *replyp = NULL;
+ } else {
+ *replyp = transaction.reply;
+ }
+ }
- sock->dump->sock = copy;
- sock->dump = NULL;
+ return transaction.error;
+}
+/* Drain all the messages currently in 'sock''s receive queue. */
+int
+nl_sock_drain(struct nl_sock *sock)
+{
+#ifdef _WIN32
return 0;
+#else
+ return drain_rcvbuf(sock->fd);
+#endif
}
-/* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
- * 'sock', and initializes 'dump' to reflect the state of the operation.
- *
- * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
- * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
- * NLM_F_ACK will be set in nlmsg_flags.
- *
- * This Netlink socket library is designed to ensure that the dump is reliable
- * and that it will not interfere with other operations on 'sock', including
- * destroying or sending and receiving messages on 'sock'. One corner case is
- * not handled:
+/* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a
+ * Netlink socket created with the given 'protocol', and initializes 'dump' to
+ * reflect the state of the operation.
*
- * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
- * whose response has not yet been received (e.g. with nl_sock_recv()).
- * This is unusual: usually nl_sock_transact() is used to send a message
- * and receive its reply all in one go.
+ * 'request' must contain a Netlink message. Before sending the message,
+ * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be
+ * set to the Netlink socket's pid. NLM_F_DUMP and NLM_F_ACK will be set in
+ * nlmsg_flags.
*
- * This function provides no status indication. An error status for the entire
- * dump operation is provided when it is completed by calling nl_dump_done().
+ * The design of this Netlink socket library ensures that the dump is reliable.
*
- * The caller is responsible for destroying 'request'.
+ * This function provides no status indication. nl_dump_done() provides an
+ * error status for the entire dump operation.
*
- * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
- * in either order.
+ * The caller must eventually destroy 'request'.
*/
void
-nl_dump_start(struct nl_dump *dump,
- struct nl_sock *sock, const struct ofpbuf *request)
-{
- struct nlmsghdr *nlmsghdr = nl_msg_nlmsghdr(request);
- nlmsghdr->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
- dump->seq = nlmsghdr->nlmsg_seq;
- dump->buffer = NULL;
- if (sock->any_groups || sock->dump) {
- /* 'sock' might belong to some multicast group, or it already has an
- * onoging dump. Clone the socket to avoid possibly intermixing
- * multicast messages or previous dump results with our results. */
- dump->status = nl_sock_clone(sock, &dump->sock);
- if (dump->status) {
- return;
- }
- } else {
- sock->dump = dump;
- dump->sock = sock;
- dump->status = 0;
+nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request)
+{
+ nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
+
+ ovs_mutex_init(&dump->mutex);
+ ovs_mutex_lock(&dump->mutex);
+ dump->status = nl_pool_alloc(protocol, &dump->sock);
+ if (!dump->status) {
+ dump->status = nl_sock_send__(dump->sock, request,
+ nl_sock_allocate_seq(dump->sock, 1),
+ true);
}
- dump->status = nl_sock_send__(sock, request, true);
+ dump->nl_seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
+ ovs_mutex_unlock(&dump->mutex);
}
-/* Helper function for nl_dump_next(). */
static int
-nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp)
+nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer)
+ OVS_REQUIRES(dump->mutex)
{
struct nlmsghdr *nlmsghdr;
- struct ofpbuf *buffer;
- int retval;
+ int error;
- retval = nl_sock_recv__(dump->sock, bufferp, true);
- if (retval) {
- return retval == EINTR ? EAGAIN : retval;
- }
- buffer = *bufferp;
+ while (!ofpbuf_size(buffer)) {
+ error = nl_sock_recv__(dump->sock, buffer, false);
+ if (error) {
+ /* The kernel never blocks providing the results of a dump, so
+ * error == EAGAIN means that we've read the whole thing, and
+ * therefore transform it into EOF. (The kernel always provides
+ * NLMSG_DONE as a sentinel. Some other thread must have received
+ * that already but not yet signaled it in 'status'.)
+ *
+ * Any other error is just an error. */
+ return error == EAGAIN ? EOF : error;
+ }
- nlmsghdr = nl_msg_nlmsghdr(buffer);
- if (dump->seq != nlmsghdr->nlmsg_seq) {
- VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
- nlmsghdr->nlmsg_seq, dump->seq);
- return EAGAIN;
+ nlmsghdr = nl_msg_nlmsghdr(buffer);
+ if (dump->nl_seq != nlmsghdr->nlmsg_seq) {
+ VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
+ nlmsghdr->nlmsg_seq, dump->nl_seq);
+ ofpbuf_clear(buffer);
+ }
}
- if (nl_msg_nlmsgerr(buffer, &retval)) {
+ if (nl_msg_nlmsgerr(buffer, &error) && error) {
VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
- strerror(retval));
- return retval && retval != EAGAIN ? retval : EPROTO;
+ ovs_strerror(error));
+ ofpbuf_clear(buffer);
+ return error;
}
return 0;
}
-/* Attempts to retrieve another reply from 'dump', which must have been
- * initialized with nl_dump_start().
+static int
+nl_dump_next__(struct ofpbuf *reply, struct ofpbuf *buffer)
+{
+ struct nlmsghdr *nlmsghdr = nl_msg_next(buffer, reply);
+ if (!nlmsghdr) {
+ VLOG_WARN_RL(&rl, "netlink dump contains message fragment");
+ return EPROTO;
+ } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
+ return EOF;
+ } else {
+ return 0;
+ }
+}
+
+/* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must
+ * have been initialized with nl_dump_start(), and 'buffer' must have been
+ * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long.
*
- * If successful, returns true and points 'reply->data' and 'reply->size' to
- * the message that was retrieved. The caller must not modify 'reply' (because
- * it points into the middle of a larger buffer).
+ * If successful, returns true and points 'reply->data' and
+ * 'ofpbuf_size(reply)' to the message that was retrieved. The caller must not
+ * modify 'reply' (because it points within 'buffer', which will be used by
+ * future calls to this function).
*
- * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
- * to 0. Failure might indicate an actual error or merely the end of replies.
- * An error status for the entire dump operation is provided when it is
- * completed by calling nl_dump_done().
+ * On failure, returns false and sets 'reply->data' to NULL and
+ * 'ofpbuf_size(reply)' to 0. Failure might indicate an actual error or merely
+ * the end of replies. An error status for the entire dump operation is
+ * provided when it is completed by calling nl_dump_done().
+ *
+ * Multiple threads may call this function, passing the same nl_dump, however
+ * each must provide independent buffers. This function may cache multiple
+ * replies in the buffer, and these will be processed before more replies are
+ * fetched. When this function returns false, other threads may continue to
+ * process replies in their buffers, but they will not fetch more replies.
*/
bool
-nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
+nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer)
{
- struct nlmsghdr *nlmsghdr;
+ int retval = 0;
- reply->data = NULL;
- reply->size = 0;
- if (dump->status) {
- return false;
+ /* If the buffer is empty, refill it.
+ *
+ * If the buffer is not empty, we don't check the dump's status.
+ * Otherwise, we could end up skipping some of the dump results if thread A
+ * hits EOF while thread B is in the midst of processing a batch. */
+ if (!ofpbuf_size(buffer)) {
+ ovs_mutex_lock(&dump->mutex);
+ if (!dump->status) {
+ /* Take the mutex here to avoid an in-kernel race. If two threads
+ * try to read from a Netlink dump socket at once, then the socket
+ * error can be set to EINVAL, which will be encountered on the
+ * next recv on that socket, which could be anywhere due to the way
+ * that we pool Netlink sockets. Serializing the recv calls avoids
+ * the issue. */
+ dump->status = nl_dump_refill(dump, buffer);
+ }
+ retval = dump->status;
+ ovs_mutex_unlock(&dump->mutex);
}
- if (dump->buffer && !dump->buffer->size) {
- ofpbuf_delete(dump->buffer);
- dump->buffer = NULL;
- }
- while (!dump->buffer) {
- int retval = nl_dump_recv(dump, &dump->buffer);
+ /* Fetch the next message from the buffer. */
+ if (!retval) {
+ retval = nl_dump_next__(reply, buffer);
if (retval) {
- ofpbuf_delete(dump->buffer);
- dump->buffer = NULL;
- if (retval != EAGAIN) {
+ /* Record 'retval' as the dump status, but don't overwrite an error
+ * with EOF. */
+ ovs_mutex_lock(&dump->mutex);
+ if (dump->status <= 0) {
dump->status = retval;
- return false;
}
+ ovs_mutex_unlock(&dump->mutex);
}
}
- nlmsghdr = nl_msg_next(dump->buffer, reply);
- if (!nlmsghdr) {
- VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
- dump->status = EPROTO;
- return false;
- } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
- dump->status = EOF;
- return false;
+ if (retval) {
+ ofpbuf_set_data(reply, NULL);
+ ofpbuf_set_size(reply, 0);
}
-
- return true;
+ return !retval;
}
/* Completes Netlink dump operation 'dump', which must have been initialized
int
nl_dump_done(struct nl_dump *dump)
{
+ int status;
+
+ ovs_mutex_lock(&dump->mutex);
+ status = dump->status;
+ ovs_mutex_unlock(&dump->mutex);
+
/* Drain any remaining messages that the client didn't read. Otherwise the
- * kernel will continue to queue them up and waste buffer space. */
- while (!dump->status) {
- struct ofpbuf reply;
- if (!nl_dump_next(dump, &reply)) {
- assert(dump->status);
+ * kernel will continue to queue them up and waste buffer space.
+ *
+ * XXX We could just destroy and discard the socket in this case. */
+ if (!status) {
+ uint64_t tmp_reply_stub[NL_DUMP_BUFSIZE / 8];
+ struct ofpbuf reply, buf;
+
+ ofpbuf_use_stub(&buf, tmp_reply_stub, sizeof tmp_reply_stub);
+ while (nl_dump_next(dump, &reply, &buf)) {
+ /* Nothing to do. */
}
+ ofpbuf_uninit(&buf);
+
+ ovs_mutex_lock(&dump->mutex);
+ status = dump->status;
+ ovs_mutex_unlock(&dump->mutex);
+ ovs_assert(status);
}
- if (dump->sock) {
- if (dump->sock->dump) {
- dump->sock->dump = NULL;
- } else {
- nl_sock_destroy(dump->sock);
+ nl_pool_release(dump->sock);
+ ovs_mutex_destroy(&dump->mutex);
+
+ return status == EOF ? 0 : status;
+}
+
+#ifdef _WIN32
+/* Pend an I/O request in the driver. The driver completes the I/O whenever
+ * an event or a packet is ready to be read. Once the I/O is completed
+ * the overlapped structure event associated with the pending I/O will be set
+ */
+static int
+pend_io_request(const struct nl_sock *sock)
+{
+ struct ofpbuf request;
+ uint64_t request_stub[128];
+ struct ovs_header *ovs_header;
+ struct nlmsghdr *nlmsg;
+ uint32_t seq;
+ int retval;
+ int error;
+ DWORD bytes;
+ OVERLAPPED *overlapped = CONST_CAST(OVERLAPPED *, &sock->overlapped);
+
+ int ovs_msg_size = sizeof (struct nlmsghdr) + sizeof (struct genlmsghdr) +
+ sizeof (struct ovs_header);
+
+ ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
+
+ seq = nl_sock_allocate_seq(sock, 1);
+ nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
+ OVS_CTRL_CMD_WIN_PEND_REQ, OVS_WIN_CONTROL_VERSION);
+ nlmsg = nl_msg_nlmsghdr(&request);
+ nlmsg->nlmsg_seq = seq;
+
+ ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
+ ovs_header->dp_ifindex = 0;
+
+ if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
+ ofpbuf_data(&request), ofpbuf_size(&request),
+ NULL, 0, &bytes, overlapped)) {
+ error = GetLastError();
+ /* Check if the I/O got pended */
+ if (error != ERROR_IO_INCOMPLETE && error != ERROR_IO_PENDING) {
+ VLOG_ERR("nl_sock_wait failed - %s\n", ovs_format_message(error));
+ retval = EINVAL;
+ goto done;
}
+ } else {
+ /* The I/O was completed synchronously */
+ poll_immediate_wake();
}
- ofpbuf_delete(dump->buffer);
- return dump->status == EOF ? 0 : dump->status;
+ retval = 0;
+
+done:
+ ofpbuf_uninit(&request);
+ return retval;
}
+#endif /* _WIN32 */
/* Causes poll_block() to wake up when any of the specified 'events' (which is
* a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
void
nl_sock_wait(const struct nl_sock *sock, short int events)
{
+#ifdef _WIN32
+ if (sock->overlapped.Internal != STATUS_PENDING) {
+ pend_io_request(sock);
+ }
+ poll_fd_wait(sock->handle, events);
+#else
poll_fd_wait(sock->fd, events);
+#endif
+}
+
+/* Returns the underlying fd for 'sock', for use in "poll()"-like operations
+ * that can't use nl_sock_wait().
+ *
+ * It's a little tricky to use the returned fd correctly, because nl_sock does
+ * "copy on write" to allow a single nl_sock to be used for notifications,
+ * transactions, and dumps. If 'sock' is used only for notifications and
+ * transactions (and never for dump) then the usage is safe. */
+int
+nl_sock_fd(const struct nl_sock *sock)
+{
+#ifdef _WIN32
+ return sock->handle;
+#else
+ return sock->fd;
+#endif
+}
+
+/* Returns the PID associated with this socket. */
+uint32_t
+nl_sock_pid(const struct nl_sock *sock)
+{
+ return sock->pid;
}
\f
/* Miscellaneous. */
+struct genl_family {
+ struct hmap_node hmap_node;
+ uint16_t id;
+ char *name;
+};
+
+static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
+
static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
[CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
+ [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
};
-static int do_lookup_genl_family(const char *name)
+static struct genl_family *
+find_genl_family_by_id(uint16_t id)
+{
+ struct genl_family *family;
+
+ HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
+ &genl_families) {
+ if (family->id == id) {
+ return family;
+ }
+ }
+ return NULL;
+}
+
+static void
+define_genl_family(uint16_t id, const char *name)
+{
+ struct genl_family *family = find_genl_family_by_id(id);
+
+ if (family) {
+ if (!strcmp(family->name, name)) {
+ return;
+ }
+ free(family->name);
+ } else {
+ family = xmalloc(sizeof *family);
+ family->id = id;
+ hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
+ }
+ family->name = xstrdup(name);
+}
+
+static const char *
+genl_family_to_name(uint16_t id)
+{
+ if (id == GENL_ID_CTRL) {
+ return "control";
+ } else {
+ struct genl_family *family = find_genl_family_by_id(id);
+ return family ? family->name : "unknown";
+ }
+}
+
+#ifndef _WIN32
+static int
+do_lookup_genl_family(const char *name, struct nlattr **attrs,
+ struct ofpbuf **replyp)
{
struct nl_sock *sock;
struct ofpbuf request, *reply;
- struct nlattr *attrs[ARRAY_SIZE(family_policy)];
- int retval;
+ int error;
- retval = nl_sock_create(NETLINK_GENERIC, &sock);
- if (retval) {
- return -retval;
+ *replyp = NULL;
+ error = nl_sock_create(NETLINK_GENERIC, &sock);
+ if (error) {
+ return error;
}
ofpbuf_init(&request, 0);
nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
CTRL_CMD_GETFAMILY, 1);
nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
- retval = nl_sock_transact(sock, &request, &reply);
+ error = nl_sock_transact(sock, &request, &reply);
ofpbuf_uninit(&request);
- if (retval) {
+ if (error) {
nl_sock_destroy(sock);
- return -retval;
+ return error;
}
if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
- family_policy, attrs, ARRAY_SIZE(family_policy))) {
+ family_policy, attrs, ARRAY_SIZE(family_policy))
+ || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
nl_sock_destroy(sock);
ofpbuf_delete(reply);
- return -EPROTO;
+ return EPROTO;
}
- retval = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
- if (retval == 0) {
- retval = -EPROTO;
- }
nl_sock_destroy(sock);
+ *replyp = reply;
+ return 0;
+}
+#else
+static int
+do_lookup_genl_family(const char *name, struct nlattr **attrs,
+ struct ofpbuf **replyp)
+{
+ struct nlmsghdr *nlmsg;
+ struct ofpbuf *reply;
+ int error;
+ uint16_t family_id;
+ const char *family_name;
+ uint32_t family_version;
+ uint32_t family_attrmax;
+ uint32_t mcgrp_id = OVS_WIN_NL_INVALID_MCGRP_ID;
+ const char *mcgrp_name = NULL;
+
+ *replyp = NULL;
+ reply = ofpbuf_new(1024);
+
+ /* CTRL_ATTR_MCAST_GROUPS is supported only for VPORT family. */
+ if (!strcmp(name, OVS_WIN_CONTROL_FAMILY)) {
+ family_id = OVS_WIN_NL_CTRL_FAMILY_ID;
+ family_name = OVS_WIN_CONTROL_FAMILY;
+ family_version = OVS_WIN_CONTROL_VERSION;
+ family_attrmax = OVS_WIN_CONTROL_ATTR_MAX;
+ } else if (!strcmp(name, OVS_DATAPATH_FAMILY)) {
+ family_id = OVS_WIN_NL_DATAPATH_FAMILY_ID;
+ family_name = OVS_DATAPATH_FAMILY;
+ family_version = OVS_DATAPATH_VERSION;
+ family_attrmax = OVS_DP_ATTR_MAX;
+ } else if (!strcmp(name, OVS_PACKET_FAMILY)) {
+ family_id = OVS_WIN_NL_PACKET_FAMILY_ID;
+ family_name = OVS_PACKET_FAMILY;
+ family_version = OVS_PACKET_VERSION;
+ family_attrmax = OVS_PACKET_ATTR_MAX;
+ } else if (!strcmp(name, OVS_VPORT_FAMILY)) {
+ family_id = OVS_WIN_NL_VPORT_FAMILY_ID;
+ family_name = OVS_VPORT_FAMILY;
+ family_version = OVS_VPORT_VERSION;
+ family_attrmax = OVS_VPORT_ATTR_MAX;
+ mcgrp_id = OVS_WIN_NL_VPORT_MCGRP_ID;
+ mcgrp_name = OVS_VPORT_MCGROUP;
+ } else if (!strcmp(name, OVS_FLOW_FAMILY)) {
+ family_id = OVS_WIN_NL_FLOW_FAMILY_ID;
+ family_name = OVS_FLOW_FAMILY;
+ family_version = OVS_FLOW_VERSION;
+ family_attrmax = OVS_FLOW_ATTR_MAX;
+ } else if (!strcmp(name, OVS_WIN_NETDEV_FAMILY)) {
+ family_id = OVS_WIN_NL_NETDEV_FAMILY_ID;
+ family_name = OVS_WIN_NETDEV_FAMILY;
+ family_version = OVS_WIN_NETDEV_VERSION;
+ family_attrmax = OVS_WIN_NETDEV_ATTR_MAX;
+ } else {
+ ofpbuf_delete(reply);
+ return EINVAL;
+ }
+
+ nl_msg_put_genlmsghdr(reply, 0, GENL_ID_CTRL, 0,
+ CTRL_CMD_NEWFAMILY, family_version);
+ /* CTRL_ATTR_HDRSIZE and CTRL_ATTR_OPS are not populated, but the
+ * callers do not seem to need them. */
+ nl_msg_put_u16(reply, CTRL_ATTR_FAMILY_ID, family_id);
+ nl_msg_put_string(reply, CTRL_ATTR_FAMILY_NAME, family_name);
+ nl_msg_put_u32(reply, CTRL_ATTR_VERSION, family_version);
+ nl_msg_put_u32(reply, CTRL_ATTR_MAXATTR, family_attrmax);
+
+ if (mcgrp_id != OVS_WIN_NL_INVALID_MCGRP_ID) {
+ size_t mcgrp_ofs1 = nl_msg_start_nested(reply, CTRL_ATTR_MCAST_GROUPS);
+ size_t mcgrp_ofs2= nl_msg_start_nested(reply,
+ OVS_WIN_NL_VPORT_MCGRP_ID - OVS_WIN_NL_MCGRP_START_ID);
+ nl_msg_put_u32(reply, CTRL_ATTR_MCAST_GRP_ID, mcgrp_id);
+ ovs_assert(mcgrp_name != NULL);
+ nl_msg_put_string(reply, CTRL_ATTR_MCAST_GRP_NAME, mcgrp_name);
+ nl_msg_end_nested(reply, mcgrp_ofs2);
+ nl_msg_end_nested(reply, mcgrp_ofs1);
+ }
+
+ /* Set the total length of the netlink message. */
+ nlmsg = nl_msg_nlmsghdr(reply);
+ nlmsg->nlmsg_len = ofpbuf_size(reply);
+
+ if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
+ family_policy, attrs, ARRAY_SIZE(family_policy))
+ || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
+ ofpbuf_delete(reply);
+ return EPROTO;
+ }
+
+ *replyp = reply;
+ return 0;
+}
+#endif
+
+/* Finds the multicast group called 'group_name' in genl family 'family_name'.
+ * When successful, writes its result to 'multicast_group' and returns 0.
+ * Otherwise, clears 'multicast_group' and returns a positive error code.
+ */
+int
+nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
+ unsigned int *multicast_group)
+{
+ struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
+ const struct nlattr *mc;
+ struct ofpbuf *reply;
+ unsigned int left;
+ int error;
+
+ *multicast_group = 0;
+ error = do_lookup_genl_family(family_name, family_attrs, &reply);
+ if (error) {
+ return error;
+ }
+
+ if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
+ error = EPROTO;
+ goto exit;
+ }
+
+ NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
+ static const struct nl_policy mc_policy[] = {
+ [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
+ [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
+ };
+
+ struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
+ const char *mc_name;
+
+ if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
+ error = EPROTO;
+ goto exit;
+ }
+
+ mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
+ if (!strcmp(group_name, mc_name)) {
+ *multicast_group =
+ nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
+ error = 0;
+ goto exit;
+ }
+ }
+ error = EPROTO;
+
+exit:
ofpbuf_delete(reply);
- return retval;
+ return error;
}
/* If '*number' is 0, translates the given Generic Netlink family 'name' to a
nl_lookup_genl_family(const char *name, int *number)
{
if (*number == 0) {
- *number = do_lookup_genl_family(name);
- assert(*number != 0);
+ struct nlattr *attrs[ARRAY_SIZE(family_policy)];
+ struct ofpbuf *reply;
+ int error;
+
+ error = do_lookup_genl_family(name, attrs, &reply);
+ if (!error) {
+ *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
+ define_genl_family(*number, name);
+ } else {
+ *number = -error;
+ }
+ ofpbuf_delete(reply);
+
+ ovs_assert(*number != 0);
}
return *number > 0 ? 0 : -*number;
}
\f
-/* Netlink PID.
+struct nl_pool {
+ struct nl_sock *socks[16];
+ int n;
+};
+
+static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER;
+static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex);
+
+static int
+nl_pool_alloc(int protocol, struct nl_sock **sockp)
+{
+ struct nl_sock *sock = NULL;
+ struct nl_pool *pool;
+
+ ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools));
+
+ ovs_mutex_lock(&pool_mutex);
+ pool = &pools[protocol];
+ if (pool->n > 0) {
+ sock = pool->socks[--pool->n];
+ }
+ ovs_mutex_unlock(&pool_mutex);
+
+ if (sock) {
+ *sockp = sock;
+ return 0;
+ } else {
+ return nl_sock_create(protocol, sockp);
+ }
+}
+
+static void
+nl_pool_release(struct nl_sock *sock)
+{
+ if (sock) {
+ struct nl_pool *pool = &pools[sock->protocol];
+
+ ovs_mutex_lock(&pool_mutex);
+ if (pool->n < ARRAY_SIZE(pool->socks)) {
+ pool->socks[pool->n++] = sock;
+ sock = NULL;
+ }
+ ovs_mutex_unlock(&pool_mutex);
+
+ nl_sock_destroy(sock);
+ }
+}
+
+/* Sends 'request' to the kernel on a Netlink socket for the given 'protocol'
+ * (e.g. NETLINK_ROUTE or NETLINK_GENERIC) and waits for a response. If
+ * successful, returns 0. On failure, returns a positive errno value.
+ *
+ * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
+ * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
+ * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
+ * reply, if any, is discarded.
+ *
+ * Before the message is sent, nlmsg_len in 'request' will be finalized to
+ * match ofpbuf_size(msg), nlmsg_pid will be set to the pid of the socket used
+ * for sending the request, and nlmsg_seq will be initialized.
+ *
+ * The caller is responsible for destroying 'request'.
+ *
+ * Bare Netlink is an unreliable transport protocol. This function layers
+ * reliable delivery and reply semantics on top of bare Netlink.
+ *
+ * In Netlink, sending a request to the kernel is reliable enough, because the
+ * kernel will tell us if the message cannot be queued (and we will in that
+ * case put it on the transmit queue and wait until it can be delivered).
+ *
+ * Receiving the reply is the real problem: if the socket buffer is full when
+ * the kernel tries to send the reply, the reply will be dropped. However, the
+ * kernel sets a flag that a reply has been dropped. The next call to recv
+ * then returns ENOBUFS. We can then re-send the request.
+ *
+ * Caveats:
+ *
+ * 1. Netlink depends on sequence numbers to match up requests and
+ * replies. The sender of a request supplies a sequence number, and
+ * the reply echos back that sequence number.
*
- * Every Netlink socket must be bound to a unique 32-bit PID. By convention,
- * programs that have a single Netlink socket use their Unix process ID as PID,
- * and programs with multiple Netlink sockets add a unique per-socket
- * identifier in the bits above the Unix process ID.
+ * This is fine, but (1) some kernel netlink implementations are
+ * broken, in that they fail to echo sequence numbers and (2) this
+ * function will drop packets with non-matching sequence numbers, so
+ * that only a single request can be usefully transacted at a time.
*
- * The kernel has Netlink PID 0.
+ * 2. Resending the request causes it to be re-executed, so the request
+ * needs to be idempotent.
*/
+int
+nl_transact(int protocol, const struct ofpbuf *request,
+ struct ofpbuf **replyp)
+{
+ struct nl_sock *sock;
+ int error;
-/* Parameters for how many bits in the PID should come from the Unix process ID
- * and how many unique per-socket. */
-#define SOCKET_BITS 10
-#define MAX_SOCKETS (1u << SOCKET_BITS)
+ error = nl_pool_alloc(protocol, &sock);
+ if (error) {
+ *replyp = NULL;
+ return error;
+ }
-#define PROCESS_BITS (32 - SOCKET_BITS)
-#define MAX_PROCESSES (1u << PROCESS_BITS)
-#define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1))
+ error = nl_sock_transact(sock, request, replyp);
-/* Bit vector of unused socket identifiers. */
-static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)];
+ nl_pool_release(sock);
+ return error;
+}
-/* Allocates and returns a new Netlink PID. */
-static int
-alloc_pid(uint32_t *pid)
+/* Sends the 'request' member of the 'n' transactions in 'transactions' on a
+ * Netlink socket for the given 'protocol' (e.g. NETLINK_ROUTE or
+ * NETLINK_GENERIC), in order, and receives responses to all of them. Fills in
+ * the 'error' member of each transaction with 0 if it was successful,
+ * otherwise with a positive errno value. If 'reply' is nonnull, then it will
+ * be filled with the reply if the message receives a detailed reply. In other
+ * cases, i.e. where the request failed or had no reply beyond an indication of
+ * success, 'reply' will be cleared if it is nonnull.
+ *
+ * The caller is responsible for destroying each request and reply, and the
+ * transactions array itself.
+ *
+ * Before sending each message, this function will finalize nlmsg_len in each
+ * 'request' to match the ofpbuf's size, set nlmsg_pid to the pid of the socket
+ * used for the transaction, and initialize nlmsg_seq.
+ *
+ * Bare Netlink is an unreliable transport protocol. This function layers
+ * reliable delivery and reply semantics on top of bare Netlink. See
+ * nl_transact() for some caveats.
+ */
+void
+nl_transact_multiple(int protocol,
+ struct nl_transaction **transactions, size_t n)
{
- int i;
+ struct nl_sock *sock;
+ int error;
- for (i = 0; i < MAX_SOCKETS; i++) {
- if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) {
- avail_sockets[i / 32] |= 1u << (i % 32);
- *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS);
- return 0;
- }
+ error = nl_pool_alloc(protocol, &sock);
+ if (!error) {
+ nl_sock_transact_multiple(sock, transactions, n);
+ nl_pool_release(sock);
+ } else {
+ nl_sock_record_errors__(transactions, n, error);
}
- VLOG_ERR("netlink pid space exhausted");
- return ENOBUFS;
}
-/* Makes the specified 'pid' available for reuse. */
-static void
-free_pid(uint32_t pid)
+\f
+static uint32_t
+nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
{
- int sock = pid >> PROCESS_BITS;
- assert(avail_sockets[sock / 32] & (1u << (sock % 32)));
- avail_sockets[sock / 32] &= ~(1u << (sock % 32));
+ uint32_t seq = sock->next_seq;
+
+ sock->next_seq += n;
+
+ /* Make it impossible for the next request for sequence numbers to wrap
+ * around to 0. Start over with 1 to avoid ever using a sequence number of
+ * 0, because the kernel uses sequence number 0 for notifications. */
+ if (sock->next_seq >= UINT32_MAX / 2) {
+ sock->next_seq = 1;
+ }
+
+ return seq;
}
-\f
+
static void
-nlmsghdr_to_string(const struct nlmsghdr *h, struct ds *ds)
+nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
{
struct nlmsg_flag {
unsigned int bits;
ds_put_cstr(ds, "(overrun)");
} else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
ds_put_cstr(ds, "(reserved)");
+ } else if (protocol == NETLINK_GENERIC) {
+ ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
} else {
ds_put_cstr(ds, "(family-defined)");
}
if (flags_left) {
ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
}
- ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32"(%d:%d))",
- h->nlmsg_seq, h->nlmsg_pid,
- (int) (h->nlmsg_pid & PROCESS_MASK),
- (int) (h->nlmsg_pid >> PROCESS_BITS));
+ ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
+ h->nlmsg_seq, h->nlmsg_pid);
}
static char *
struct ds ds = DS_EMPTY_INITIALIZER;
const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
if (h) {
- nlmsghdr_to_string(h, &ds);
+ nlmsghdr_to_string(h, protocol, &ds);
if (h->nlmsg_type == NLMSG_ERROR) {
const struct nlmsgerr *e;
e = ofpbuf_at(buffer, NLMSG_HDRLEN,
if (e) {
ds_put_format(&ds, " error(%d", e->error);
if (e->error < 0) {
- ds_put_format(&ds, "(%s)", strerror(-e->error));
+ ds_put_format(&ds, "(%s)", ovs_strerror(-e->error));
}
ds_put_cstr(&ds, ", in-reply-to(");
- nlmsghdr_to_string(&e->msg, &ds);
+ nlmsghdr_to_string(&e->msg, protocol, &ds);
ds_put_cstr(&ds, "))");
} else {
ds_put_cstr(&ds, " error(truncated)");
if (error) {
ds_put_format(&ds, " done(%d", *error);
if (*error < 0) {
- ds_put_format(&ds, "(%s)", strerror(-*error));
+ ds_put_format(&ds, "(%s)", ovs_strerror(-*error));
}
ds_put_cstr(&ds, ")");
} else {
ofpbuf_use_const(&buffer, message, size);
nlmsg = nlmsg_to_string(&buffer, protocol);
- VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);
+ VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg);
free(nlmsg);
}
-
-