/*
- * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * 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 "ofpbuf.h"
#include "ovs-thread.h"
#include "poll-loop.h"
+#include "seq.h"
#include "socket-util.h"
#include "util.h"
#include "vlog.h"
#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;
+}
+
+static void
+set_sock_pid_in_kernel(HANDLE handle, uint32_t pid)
+ OVS_GUARDED_BY(portid_mutex)
+{
+ struct nlmsghdr msg = { 0 };
+
+ msg.nlmsg_len = sizeof(struct nlmsghdr);
+ msg.nlmsg_type = 80; /* target = set file pid */
+ msg.nlmsg_flags = 0;
+ msg.nlmsg_seq = 0;
+ msg.nlmsg_pid = pid;
+
+ WriteFile(handle, &msg, sizeof(struct nlmsghdr), NULL, NULL);
+}
+#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. */
/* Netlink sockets. */
struct nl_sock {
+#ifdef _WIN32
+ HANDLE handle;
+#else
int fd;
+#endif
uint32_t next_seq;
uint32_t pid;
int protocol;
{
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;
*sockp = NULL;
sock = xmalloc(sizeof *sock);
+#ifdef _WIN32
+ sock->handle = CreateFileA("\\\\.\\OpenVSwitchDevice",
+ GENERIC_READ | GENERIC_WRITE,
+ FILE_SHARE_READ | FILE_SHARE_WRITE,
+ NULL, OPEN_EXISTING,
+ FILE_ATTRIBUTE_NORMAL, NULL);
+
+ int last_error = GetLastError();
+
+ if (sock->handle == INVALID_HANDLE_VALUE) {
+ VLOG_ERR("fcntl: %s", ovs_strerror(last_error));
+ goto error;
+ }
+#else
sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
if (sock->fd < 0) {
VLOG_ERR("fcntl: %s", ovs_strerror(errno));
goto error;
}
+#endif
+
sock->protocol = protocol;
sock->next_seq = 1;
rcvbuf = 1024 * 1024;
+#ifdef _WIN32
+ sock->rcvbuf = rcvbuf;
+ ovs_mutex_lock(&portid_mutex);
+ sock->pid = portid_next();
+ set_sock_pid_in_kernel(sock->handle, sock->pid);
+ ovs_mutex_unlock(&portid_mutex);
+#else
if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
&rcvbuf, sizeof rcvbuf)) {
/* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
goto error;
}
sock->pid = local.nl_pid;
+#endif
*sockp = sock;
return 0;
retval = EINVAL;
}
}
+#ifdef _WIN32
+ 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) {
+#ifdef _WIN32
+ CloseHandle(sock->handle);
+#else
close(sock->fd);
+#endif
free(sock);
}
}
int
nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
{
+#ifdef _WIN32
+#define OVS_VPORT_MCGROUP_FALLBACK_ID 33
+ struct ofpbuf msg_buf;
+ struct message_multicast
+ {
+ struct nlmsghdr;
+ /* if true, join; if else, leave */
+ unsigned char join;
+ unsigned int groupId;
+ };
+
+ struct message_multicast msg = { 0 };
+
+ msg.nlmsg_len = sizeof(struct message_multicast);
+ msg.nlmsg_type = OVS_VPORT_MCGROUP_FALLBACK_ID;
+ msg.nlmsg_flags = 0;
+ msg.nlmsg_seq = 0;
+ msg.nlmsg_pid = sock->pid;
+
+ msg.join = 1;
+ msg.groupId = multicast_group;
+ msg_buf.base_ = &msg;
+ msg_buf.data_ = &msg;
+ msg_buf.size_ = msg.nlmsg_len;
+
+ nl_sock_send__(sock, &msg_buf, msg.nlmsg_seq, 0);
+#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, ovs_strerror(errno));
return errno;
}
+#endif
return 0;
}
int
nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
{
+#ifdef _WIN32
+ struct ofpbuf msg_buf;
+ struct message_multicast
+ {
+ struct nlmsghdr;
+ /* if true, join; if else, leave*/
+ unsigned char join;
+ };
+
+ struct message_multicast msg = { 0 };
+ nl_msg_put_nlmsghdr(&msg, sizeof(struct message_multicast),
+ multicast_group, 0);
+ msg.join = 0;
+
+ msg_buf.base_ = &msg;
+ msg_buf.data_ = &msg;
+ msg_buf.size_ = msg.nlmsg_len;
+
+ nl_sock_send__(sock, &msg_buf, msg.nlmsg_seq, 0);
+#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, ovs_strerror(errno));
return errno;
}
+#endif
return 0;
}
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
+ bool result;
+ DWORD last_error = 0;
+ result = WriteFile(sock->handle, ofpbuf_data(msg), ofpbuf_size(msg),
+ &retval, NULL);
+ last_error = GetLastError();
+ if (last_error != ERROR_SUCCESS && !result) {
+ retval = -1;
+ errno = EAGAIN;
+ }
+#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, nlmsg_pid
+ * '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.
*
}
/* 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, nlmsg_pid
+ * '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.
*
* '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;
+#ifdef _WIN32
+#define MAX_STACK_LENGTH 81920
+ uint8_t tail[MAX_STACK_LENGTH];
+#else
uint8_t tail[65536];
+#endif
struct iovec iov[2];
struct msghdr msg;
ssize_t retval;
+ int error;
ovs_assert(buf->allocated >= sizeof *nlmsghdr);
ofpbuf_clear(buf);
- iov[0].iov_base = buf->base;
+ iov[0].iov_base = ofpbuf_base(buf);
iov[0].iov_len = buf->allocated;
iov[1].iov_base = tail;
iov[1].iov_len = sizeof tail;
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 {
+ nlmsghdr->nlmsg_len = UINT32_MAX;
+#ifdef _WIN32
+ boolean result = false;
+ DWORD last_error = 0;
+ result = ReadFile(sock->handle, tail, MAX_STACK_LENGTH, &retval, NULL);
+ last_error = GetLastError();
+ if (last_error != ERROR_SUCCESS && !result) {
+ retval = -1;
+ errno = EAGAIN;
+ } else {
+ ofpbuf_put(buf, tail, retval);
+ }
+#else
retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
- } while (retval < 0 && errno == EINTR);
-
- if (retval < 0) {
- int error = errno;
+#endif
+ error = (retval < 0 ? errno
+ : retval == 0 ? ECONNRESET /* not possible? */
+ : nlmsghdr->nlmsg_len != UINT32_MAX ? 0
+ : retval);
+ } while (error == EINTR);
+ if (error) {
if (error == ENOBUFS) {
/* Socket receive buffer overflow dropped one or more messages that
* the kernel tried to send to us. */
return E2BIG;
}
- nlmsghdr = buf->data;
if (retval < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len > retval) {
- VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE"d bytes < %"PRIuSIZE")",
+ VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE" bytes < %"PRIuSIZE")",
retval, sizeof *nlmsghdr);
return EPROTO;
}
-
- buf->size = MIN(retval, buf->allocated);
+#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
- 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;
struct nl_transaction *txn = transactions[i];
struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
- nlmsg->nlmsg_len = txn->request->size;
+ nlmsg->nlmsg_len = ofpbuf_size(txn->request);
nlmsg->nlmsg_seq = base_seq + i;
nlmsg->nlmsg_pid = sock->pid;
- iovs[i].iov_base = txn->request->data;
- iovs[i].iov_len = txn->request->size;
+ iovs[i].iov_base = ofpbuf_data(txn->request);
+ iovs[i].iov_len = ofpbuf_size(txn->request);
}
memset(&msg, 0, sizeof msg);
msg.msg_iov = iovs;
msg.msg_iovlen = n;
do {
+#ifdef _WIN32
+ DWORD last_error = 0;
+ bool result = FALSE;
+ for (i = 0; i < n; i++) {
+ result = WriteFile((HANDLE)sock->handle, iovs[i].iov_base, iovs[i].iov_len,
+ &error, NULL);
+ last_error = GetLastError();
+ if (last_error != ERROR_SUCCESS && !result) {
+ error = EAGAIN;
+ errno = EAGAIN;
+ } else {
+ error = 0;
+ }
+ }
+#else
error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
+#endif
} while (error == EINTR);
for (i = 0; i < n; i++) {
struct nl_transaction *txn = transactions[i];
- log_nlmsg(__func__, error, txn->request->data, txn->request->size,
+ log_nlmsg(__func__, error, ofpbuf_data(txn->request), ofpbuf_size(txn->request),
sock->protocol);
}
if (!error) {
return error;
}
-/* Sends the 'request' member of the 'n' transactions in 'transactions' on
- * 'sock', 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 'sock''s pid, 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_sock_transact() for some caveats.
- */
-void
+static void
nl_sock_transact_multiple(struct nl_sock *sock,
struct nl_transaction **transactions, size_t n)
{
#else
enum { MAX_BATCH_BYTES = 4096 - 512 };
#endif
- bytes = transactions[0]->request->size;
+ bytes = ofpbuf_size(transactions[0]->request);
for (count = 1; count < n && count < max_batch_count; count++) {
- if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
+ if (bytes + ofpbuf_size(transactions[count]->request) > MAX_BATCH_BYTES) {
break;
}
- bytes += transactions[count]->request->size;
+ bytes += ofpbuf_size(transactions[count]->request);
}
error = nl_sock_transact_multiple__(sock, transactions, count, &done);
}
}
-/* 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.
- *
- * Before the message is sent, nlmsg_len in 'request' will be finalized to
- * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
- * be initialized, 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
+static int
nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
struct ofpbuf **replyp)
{
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 on a
* Netlink socket created with the given 'protocol', 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 the Netlink socket's pid, before the message is sent. NLM_F_DUMP
- * and NLM_F_ACK will be set in nlmsg_flags.
+ * '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.
*
* The design of this Netlink socket library ensures that the dump is reliable.
*
- * 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().
+ * This function provides no status indication. nl_dump_done() provides an
+ * error status for the entire dump operation.
*
- * The caller is responsible for destroying 'request'.
+ * The caller must eventually destroy 'request'.
*/
void
nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request)
{
- ofpbuf_init(&dump->buffer, 4096);
+ 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) {
- return;
+ if (!dump->status) {
+ dump->status = nl_sock_send__(dump->sock, request,
+ nl_sock_allocate_seq(dump->sock, 1),
+ true);
}
-
- nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
- dump->status = nl_sock_send__(dump->sock, request,
- nl_sock_allocate_seq(dump->sock, 1), true);
- dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
+ 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)
+nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer)
+ OVS_REQUIRES(dump->mutex)
{
struct nlmsghdr *nlmsghdr;
- int retval;
+ int error;
- retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
- if (retval) {
- return retval == EINTR ? EAGAIN : retval;
- }
+ 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(&dump->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(&dump->buffer, &retval)) {
+ if (nl_msg_nlmsgerr(buffer, &error) && error) {
VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
- ovs_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
+ * '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).
*
- * 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).
+ * 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().
*
- * 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().
+ * 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);
}
- while (!dump->buffer.size) {
- int retval = nl_dump_recv(dump);
+ /* Fetch the next message from the buffer. */
+ if (!retval) {
+ retval = nl_dump_next__(reply, buffer);
if (retval) {
- ofpbuf_clear(&dump->buffer);
- 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.
*
* XXX We could just destroy and discard the socket in this case. */
- while (!dump->status) {
- struct ofpbuf reply;
- if (!nl_dump_next(dump, &reply)) {
- ovs_assert(dump->status);
+ 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);
}
+
nl_pool_release(dump->sock);
- ofpbuf_uninit(&dump->buffer);
- return dump->status == EOF ? 0 : dump->status;
+ ovs_mutex_destroy(&dump->mutex);
+
+ return status == EOF ? 0 : status;
}
/* Causes poll_block() to wake up when any of the specified 'events' (which is
void
nl_sock_wait(const struct nl_sock *sock, short int events)
{
+#ifdef _WIN32
+ 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
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. */
}
}
+/* 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.
+ *
+ * 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_transact(int protocol, const struct ofpbuf *request,
struct ofpbuf **replyp)
return error;
}
+/* 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)