X-Git-Url: http://git.cascardo.eti.br/?a=blobdiff_plain;f=lib%2Fnetlink-socket.c;h=1a1b5e42a3082f5cae4f119bb8d87f044681d4b7;hb=HEAD;hp=4d9832f48d701ef566d14586827d01b9f2506659;hpb=ebac7fb759946f3e0660b882bfd4fefd39bc7b17;p=cascardo%2Fovs.git

diff --git a/lib/netlink-socket.c b/lib/netlink-socket.c
index 4d9832f48..1a1b5e42a 100644
--- a/lib/netlink-socket.c
+++ b/lib/netlink-socket.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2016 Nicira, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -35,7 +35,7 @@
 #include "seq.h"
 #include "socket-util.h"
 #include "util.h"
-#include "vlog.h"
+#include "openvswitch/vlog.h"
 
 VLOG_DEFINE_THIS_MODULE(netlink_socket);
 
@@ -49,20 +49,6 @@ COVERAGE_DEFINE(netlink_sent);
 #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. */
@@ -80,6 +66,8 @@ static int get_sock_pid_from_kernel(struct nl_sock *sock);
 struct nl_sock {
 #ifdef _WIN32
     HANDLE handle;
+    OVERLAPPED overlapped;
+    DWORD read_ioctl;
 #else
     int fd;
 #endif
@@ -139,18 +127,25 @@ nl_sock_create(int protocol, struct nl_sock **sockp)
     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();
+    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_strerror(last_error));
+        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) {
@@ -186,6 +181,7 @@ nl_sock_create(int protocol, struct nl_sock **sockp)
         goto error;
     }
     sock->rcvbuf = retval;
+    retval = 0;
 
     /* Connect to kernel (pid 0) as remote address. */
     memset(&remote, 0, sizeof remote);
@@ -221,6 +217,9 @@ error:
         }
     }
 #ifdef _WIN32
+    if (sock->overlapped.hEvent) {
+        CloseHandle(sock->overlapped.hEvent);
+    }
     if (sock->handle != INVALID_HANDLE_VALUE) {
         CloseHandle(sock->handle);
     }
@@ -248,6 +247,9 @@ nl_sock_destroy(struct nl_sock *sock)
 {
     if (sock) {
 #ifdef _WIN32
+        if (sock->overlapped.hEvent) {
+            CloseHandle(sock->overlapped.hEvent);
+        }
         CloseHandle(sock->handle);
 #else
         close(sock->fd);
@@ -258,67 +260,58 @@ nl_sock_destroy(struct nl_sock *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. */
+ * uses a separate IOCTL instead of a transaction semantic to avoid unnecessary
+ * message overhead. */
 static int
 get_sock_pid_from_kernel(struct nl_sock *sock)
 {
-    struct nl_transaction txn;
+    uint32_t pid = 0;
+    int retval = 0;
+    DWORD bytes = 0;
+
+    if (!DeviceIoControl(sock->handle, OVS_IOCTL_GET_PID,
+                         NULL, 0, &pid, sizeof(pid),
+                         &bytes, NULL)) {
+        retval = EINVAL;
+    } else {
+        if (bytes < sizeof(pid)) {
+            retval = EINVAL;
+        } else {
+            sock->pid = pid;
+        }
+    }
+
+    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 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);
+    int error;
 
     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_CTRL_CMD_MC_SUBSCRIBE_REQ,
+                          OVS_WIN_CONTROL_VERSION);
 
     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;
-        }
+    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);
 
-        nlmsg = nl_msg_nlmsghdr(txn.reply);
-        if (nlmsg->nlmsg_seq != seq) {
-            retval = EINVAL;
-            goto done;
-        }
-        sock->pid = nlmsg->nlmsg_pid;
-    }
-    retval = 0;
-
-done:
+    error = nl_sock_send(sock, &request, true);
     ofpbuf_uninit(&request);
-    ofpbuf_uninit(&reply);
-    return retval;
+    return error;
 }
-#endif  /* _WIN32 */
-
+#endif
 /* Tries to add 'sock' as a listener for 'multicast_group'.  Returns 0 if
  * successful, otherwise a positive errno value.
  *
@@ -334,31 +327,15 @@ 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);
+    /* 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) {
+        sock->read_ioctl = OVS_IOCTL_READ;
+        VLOG_WARN("could not join multicast group %u (%s)",
+                  multicast_group, ovs_strerror(error));
+        return error;
+    }
 #else
     if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
                    &multicast_group, sizeof multicast_group) < 0) {
@@ -370,6 +347,68 @@ nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
     return 0;
 }
 
+#ifdef _WIN32
+int
+nl_sock_subscribe_packets(struct nl_sock *sock)
+{
+    int error;
+
+    if (sock->read_ioctl != OVS_IOCTL_READ) {
+        return EINVAL;
+    }
+
+    error = nl_sock_subscribe_packet__(sock, true);
+    if (error) {
+        VLOG_WARN("could not subscribe packets (%s)",
+                  ovs_strerror(error));
+        return error;
+    }
+    sock->read_ioctl = OVS_IOCTL_READ_PACKET;
+
+    return 0;
+}
+
+int
+nl_sock_unsubscribe_packets(struct nl_sock *sock)
+{
+    ovs_assert(sock->read_ioctl == OVS_IOCTL_READ_PACKET);
+
+    int error = nl_sock_subscribe_packet__(sock, false);
+    if (error) {
+        VLOG_WARN("could not unsubscribe to packets (%s)",
+                  ovs_strerror(error));
+        return error;
+    }
+
+    sock->read_ioctl = OVS_IOCTL_READ;
+    return 0;
+}
+
+int
+nl_sock_subscribe_packet__(struct nl_sock *sock, bool subscribe)
+{
+    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_PACKET_SUBSCRIBE_REQ,
+                          OVS_WIN_CONTROL_VERSION);
+
+    ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
+    ovs_header->dp_ifindex = 0;
+    nl_msg_put_u8(&request, OVS_NL_ATTR_PACKET_SUBSCRIBE, subscribe ? 1 : 0);
+    nl_msg_put_u32(&request, OVS_NL_ATTR_PACKET_PID, sock->pid);
+
+    error = nl_sock_send(sock, &request, true);
+    ofpbuf_uninit(&request);
+    return error;
+}
+#endif
+
 /* Tries to make 'sock' stop listening to 'multicast_group'.  Returns 0 if
  * successful, otherwise a positive errno value.
  *
@@ -384,24 +423,13 @@ 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);
+    int error = nl_sock_mcgroup(sock, multicast_group, false);
+    if (error) {
+        VLOG_WARN("could not leave multicast group %u (%s)",
+                   multicast_group, ovs_strerror(error));
+        return error;
+    }
+    sock->read_ioctl = OVS_IOCTL_READ;
 #else
     if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
                    &multicast_group, sizeof multicast_group) < 0) {
@@ -420,27 +448,32 @@ nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
     struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
     int error;
 
-    nlmsg->nlmsg_len = ofpbuf_size(msg);
+    nlmsg->nlmsg_len = msg->size;
     nlmsg->nlmsg_seq = nlmsg_seq;
     nlmsg->nlmsg_pid = sock->pid;
     do {
         int retval;
 #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) {
+        DWORD bytes;
+
+        if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
+                             msg->data, msg->size, NULL, 0,
+                             &bytes, NULL)) {
             retval = -1;
-            errno = EAGAIN;
+            /* XXX: Map to a more appropriate error based on GetLastError(). */
+            errno = EINVAL;
+            VLOG_DBG_RL(&rl, "fatal driver failure in write: %s",
+                ovs_lasterror_to_string());
+        } else {
+            retval = msg->size;
         }
 #else
-        retval = send(sock->fd, ofpbuf_data(msg), ofpbuf_size(msg), wait ? 0 : MSG_DONTWAIT);
+        retval = send(sock->fd, msg->data, msg->size,
+                      wait ? 0 : MSG_DONTWAIT);
 #endif
         error = retval < 0 ? errno : 0;
     } while (error == EINTR);
-    log_nlmsg(__func__, error, ofpbuf_data(msg), ofpbuf_size(msg), sock->protocol);
+    log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
     if (!error) {
         COVERAGE_INC(netlink_sent);
     }
@@ -448,7 +481,7 @@ nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
 }
 
 /* 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
+ * 'sock'.  nlmsg_len in 'msg' will be finalized to match msg->size, 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.
  *
@@ -462,7 +495,7 @@ nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
 }
 
 /* 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
+ * 'sock'.  nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
  * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
  * 'nlmsg_seq', before the message is sent.
  *
@@ -488,12 +521,7 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
      * '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;
@@ -502,7 +530,7 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
     ovs_assert(buf->allocated >= sizeof *nlmsghdr);
     ofpbuf_clear(buf);
 
-    iov[0].iov_base = ofpbuf_base(buf);
+    iov[0].iov_base = buf->base;
     iov[0].iov_len = buf->allocated;
     iov[1].iov_base = tail;
     iov[1].iov_len = sizeof tail;
@@ -518,19 +546,32 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
      * 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);
+    nlmsghdr = buf->base;
     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) {
+        DWORD bytes;
+        if (!DeviceIoControl(sock->handle, sock->read_ioctl,
+                             NULL, 0, tail, sizeof tail, &bytes, NULL)) {
+            VLOG_DBG_RL(&rl, "fatal driver failure in transact: %s",
+                        ovs_lasterror_to_string());
             retval = -1;
-            errno = EAGAIN;
+            /* XXX: Map to a more appropriate error. */
+            errno = EINVAL;
         } else {
-            ofpbuf_put(buf, tail, retval);
+            retval = bytes;
+            if (retval == 0) {
+                retval = -1;
+                errno = EAGAIN;
+            } else {
+                if (retval >= buf->allocated) {
+                    ofpbuf_reinit(buf, retval);
+                    nlmsghdr = buf->base;
+                    nlmsghdr->nlmsg_len = UINT32_MAX;
+                }
+                memcpy(buf->data, tail, retval);
+                buf->size = retval;
+            }
         }
 #else
         retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
@@ -563,14 +604,14 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
         return EPROTO;
     }
 #ifndef _WIN32
-    ofpbuf_set_size(buf, MIN(retval, buf->allocated));
+    buf->size = 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, ofpbuf_data(buf), ofpbuf_size(buf), sock->protocol);
+    log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
     COVERAGE_INC(netlink_received);
 
     return 0;
@@ -636,42 +677,27 @@ nl_sock_transact_multiple__(struct nl_sock *sock,
         struct nl_transaction *txn = transactions[i];
         struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
 
-        nlmsg->nlmsg_len = ofpbuf_size(txn->request);
+        nlmsg->nlmsg_len = txn->request->size;
         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);
+        iovs[i].iov_base = txn->request->data;
+        iovs[i].iov_len = txn->request->size;
     }
 
+#ifndef _WIN32
     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, ofpbuf_data(txn->request), ofpbuf_size(txn->request),
-                  sock->protocol);
+        log_nlmsg(__func__, error, txn->request->data,
+                  txn->request->size, sock->protocol);
     }
     if (!error) {
         COVERAGE_ADD(netlink_sent, n);
@@ -750,6 +776,93 @@ nl_sock_transact_multiple__(struct nl_sock *sock,
         base_seq += i + 1;
     }
     ofpbuf_uninit(&tmp_reply);
+#else
+    error = 0;
+    uint8_t reply_buf[65536];
+    for (i = 0; i < n; i++) {
+        DWORD reply_len;
+        bool ret;
+        struct nl_transaction *txn = transactions[i];
+        struct nlmsghdr *request_nlmsg, *reply_nlmsg;
+
+        ret = DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
+                              txn->request->data,
+                              txn->request->size,
+                              reply_buf, sizeof reply_buf,
+                              &reply_len, NULL);
+
+        if (ret && reply_len == 0) {
+            /*
+             * The current transaction did not produce any data to read and that
+             * is not an error as such. Continue with the remainder of the
+             * transactions.
+             */
+            txn->error = 0;
+            if (txn->reply) {
+                ofpbuf_clear(txn->reply);
+            }
+        } else if (!ret) {
+            /* XXX: Map to a more appropriate error. */
+            error = EINVAL;
+            VLOG_DBG_RL(&rl, "fatal driver failure: %s",
+                ovs_lasterror_to_string());
+            break;
+        }
+
+        if (reply_len != 0) {
+            if (reply_len < sizeof *reply_nlmsg) {
+                nl_sock_record_errors__(transactions, n, 0);
+                VLOG_DBG_RL(&rl, "insufficient length of reply %#"PRIu32
+                    " for seq: %#"PRIx32, reply_len, request_nlmsg->nlmsg_seq);
+                break;
+            }
+
+            /* Validate the sequence number in the reply. */
+            request_nlmsg = nl_msg_nlmsghdr(txn->request);
+            reply_nlmsg = (struct nlmsghdr *)reply_buf;
+
+            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;
+            }
+
+            /* Handle errors embedded within the netlink message. */
+            ofpbuf_use_stub(&tmp_reply, reply_buf, sizeof reply_buf);
+            tmp_reply.size = sizeof reply_buf;
+            if (nl_msg_nlmsgerr(&tmp_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;
+                if (txn->reply) {
+                    /* Copy the reply to the buffer specified by the caller. */
+                    if (reply_len > txn->reply->allocated) {
+                        ofpbuf_reinit(txn->reply, reply_len);
+                    }
+                    memcpy(txn->reply->data, reply_buf, reply_len);
+                    txn->reply->size = reply_len;
+                }
+            }
+            ofpbuf_uninit(&tmp_reply);
+        }
+
+        /* Count the number of successful transactions. */
+        (*done)++;
+
+    }
+
+    if (!error) {
+        COVERAGE_ADD(netlink_sent, n);
+    }
+#endif
 
     return error;
 }
@@ -788,12 +901,12 @@ nl_sock_transact_multiple(struct nl_sock *sock,
 #else
         enum { MAX_BATCH_BYTES = 4096 - 512 };
 #endif
-        bytes = ofpbuf_size(transactions[0]->request);
+        bytes = transactions[0]->request->size;
         for (count = 1; count < n && count < max_batch_count; count++) {
-            if (bytes + ofpbuf_size(transactions[count]->request) > MAX_BATCH_BYTES) {
+            if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
                 break;
             }
-            bytes += ofpbuf_size(transactions[count]->request);
+            bytes += transactions[count]->request->size;
         }
 
         error = nl_sock_transact_multiple__(sock, transactions, count, &done);
@@ -805,6 +918,11 @@ nl_sock_transact_multiple(struct nl_sock *sock,
         } else if (error) {
             VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error));
             nl_sock_record_errors__(transactions, n, error);
+            if (error != EAGAIN) {
+                /* A fatal error has occurred.  Abort the rest of
+                 * transactions. */
+                break;
+            }
         }
     }
 }
@@ -885,7 +1003,7 @@ nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer)
     struct nlmsghdr *nlmsghdr;
     int error;
 
-    while (!ofpbuf_size(buffer)) {
+    while (!buffer->size) {
         error = nl_sock_recv__(dump->sock, buffer, false);
         if (error) {
             /* The kernel never blocks providing the results of a dump, so
@@ -935,12 +1053,12 @@ nl_dump_next__(struct ofpbuf *reply, struct ofpbuf *buffer)
  * 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
+ * 'reply->size' 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
- * 'ofpbuf_size(reply)' to 0.  Failure might indicate an actual error or merely
+ * '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().
  *
@@ -960,7 +1078,7 @@ nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer)
      * 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)) {
+    if (!buffer->size) {
         ovs_mutex_lock(&dump->mutex);
         if (!dump->status) {
             /* Take the mutex here to avoid an in-kernel race.  If two threads
@@ -990,8 +1108,8 @@ nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer)
     }
 
     if (retval) {
-        ofpbuf_set_data(reply, NULL);
-        ofpbuf_set_size(reply, 0);
+        reply->data = NULL;
+        reply->size = 0;
     }
     return !retval;
 }
@@ -1034,18 +1152,88 @@ nl_dump_done(struct nl_dump *dump)
     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(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 = 0;
+    int error;
+    DWORD bytes;
+    OVERLAPPED *overlapped = CONST_CAST(OVERLAPPED *, &sock->overlapped);
+    uint16_t cmd = OVS_CTRL_CMD_WIN_PEND_PACKET_REQ;
+
+    ovs_assert(sock->read_ioctl == OVS_IOCTL_READ_PACKET ||
+               sock->read_ioctl  == OVS_IOCTL_READ_EVENT);
+    if (sock->read_ioctl == OVS_IOCTL_READ_EVENT) {
+        cmd = OVS_CTRL_CMD_WIN_PEND_REQ;
+    }
+
+    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,
+                          cmd, OVS_WIN_CONTROL_VERSION);
+    nlmsg = nl_msg_nlmsghdr(&request);
+    nlmsg->nlmsg_seq = seq;
+    nlmsg->nlmsg_pid = sock->pid;
+
+    ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
+    ovs_header->dp_ifindex = 0;
+
+    if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
+                         request.data, request.size,
+                         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;
+        }
+    } else {
+        retval = EAGAIN;
+    }
+
+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'. */
+ * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'.
+ * On Windows, 'sock' is not treated as const, and may be modified. */
 void
 nl_sock_wait(const struct nl_sock *sock, short int events)
 {
 #ifdef _WIN32
-    poll_fd_wait(sock->handle, events);
+    if (sock->overlapped.Internal != STATUS_PENDING) {
+        int ret = pend_io_request(CONST_CAST(struct nl_sock *, sock));
+        if (ret == 0) {
+            poll_wevent_wait(sock->overlapped.hEvent);
+        } else {
+            poll_immediate_wake();
+        }
+    } else {
+        poll_wevent_wait(sock->overlapped.hEvent);
+    }
 #else
     poll_fd_wait(sock->fd, events);
 #endif
 }
 
+#ifndef _WIN32
 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
  * that can't use nl_sock_wait().
  *
@@ -1056,12 +1244,9 @@ nl_sock_wait(const struct nl_sock *sock, short int events)
 int
 nl_sock_fd(const struct nl_sock *sock)
 {
-#ifdef _WIN32
-    return sock->handle;
-#else
     return sock->fd;
-#endif
 }
+#endif
 
 /* Returns the PID associated with this socket. */
 uint32_t
@@ -1171,7 +1356,6 @@ static int
 do_lookup_genl_family(const char *name, struct nlattr **attrs,
                       struct ofpbuf **replyp)
 {
-    struct nl_sock *sock;
     struct nlmsghdr *nlmsg;
     struct ofpbuf *reply;
     int error;
@@ -1213,6 +1397,11 @@ do_lookup_genl_family(const char *name, struct nlattr **attrs,
         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;
@@ -1240,12 +1429,11 @@ do_lookup_genl_family(const char *name, struct nlattr **attrs,
 
     /* Set the total length of the netlink message. */
     nlmsg = nl_msg_nlmsghdr(reply);
-    nlmsg->nlmsg_len = ofpbuf_size(reply);
+    nlmsg->nlmsg_len = reply->size;
 
     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) {
-        nl_sock_destroy(sock);
         ofpbuf_delete(reply);
         return EPROTO;
     }
@@ -1393,7 +1581,7 @@ nl_pool_release(struct nl_sock *sock)
  * 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
+ * match msg->size, 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'.
@@ -1599,15 +1787,12 @@ static void
 log_nlmsg(const char *function, int error,
           const void *message, size_t size, int protocol)
 {
-    struct ofpbuf buffer;
-    char *nlmsg;
-
     if (!VLOG_IS_DBG_ENABLED()) {
         return;
     }
 
-    ofpbuf_use_const(&buffer, message, size);
-    nlmsg = nlmsg_to_string(&buffer, protocol);
+    struct ofpbuf buffer = ofpbuf_const_initializer(message, size);
+    char *nlmsg = nlmsg_to_string(&buffer, protocol);
     VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg);
     free(nlmsg);
 }