X-Git-Url: http://git.cascardo.eti.br/?a=blobdiff_plain;f=lib%2Fnetlink-socket.c;h=1a1b5e42a3082f5cae4f119bb8d87f044681d4b7;hb=d71c423edf5d7e9377957dddc0c72d1d0c1375b1;hp=df889d3eb449974c4535ddfbe69c05f6ddd554f8;hpb=bdd0bd2873415838643e42a1e4c08319fe8d3b8b;p=cascardo%2Fovs.git

diff --git a/lib/netlink-socket.c b/lib/netlink-socket.c
index df889d3eb..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.
@@ -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. */
@@ -195,6 +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);
@@ -273,63 +260,27 @@ 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;
-    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);
+    uint32_t pid = 0;
+    int retval = 0;
+    DWORD bytes = 0;
 
-    if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
-                         txn.request->data, txn.request->size,
-                         txn.reply->data, txn.reply->size,
+    if (!DeviceIoControl(sock->handle, OVS_IOCTL_GET_PID,
+                         NULL, 0, &pid, sizeof(pid),
                          &bytes, NULL)) {
         retval = EINVAL;
-        goto done;
     } else {
-        if (bytes < ovs_msg_size) {
+        if (bytes < sizeof(pid)) {
             retval = EINVAL;
-            goto done;
-        }
-
-        nlmsg = nl_msg_nlmsghdr(txn.reply);
-        if (nlmsg->nlmsg_seq != seq) {
-            retval = EINVAL;
-            goto done;
+        } else {
+            sock->pid = pid;
         }
-        sock->pid = nlmsg->nlmsg_pid;
     }
-    retval = 0;
 
-done:
-    ofpbuf_uninit(&request);
-    ofpbuf_uninit(&reply);
     return retval;
 }
 #endif  /* _WIN32 */
@@ -408,8 +359,8 @@ nl_sock_subscribe_packets(struct nl_sock *sock)
 
     error = nl_sock_subscribe_packet__(sock, true);
     if (error) {
-        VLOG_WARN("could not unsubscribe packets (%s)",
-                  ovs_strerror(errno));
+        VLOG_WARN("could not subscribe packets (%s)",
+                  ovs_strerror(error));
         return error;
     }
     sock->read_ioctl = OVS_IOCTL_READ_PACKET;
@@ -424,8 +375,8 @@ nl_sock_unsubscribe_packets(struct nl_sock *sock)
 
     int error = nl_sock_subscribe_packet__(sock, false);
     if (error) {
-        VLOG_WARN("could not subscribe to packets (%s)",
-                  ovs_strerror(errno));
+        VLOG_WARN("could not unsubscribe to packets (%s)",
+                  ovs_strerror(error));
         return error;
     }
 
@@ -511,6 +462,8 @@ nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
             retval = -1;
             /* 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;
         }
@@ -600,7 +553,10 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
         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;
+            /* XXX: Map to a more appropriate error. */
             errno = EINVAL;
         } else {
             retval = bytes;
@@ -825,61 +781,78 @@ nl_sock_transact_multiple__(struct nl_sock *sock,
     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;
 
-        if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
-                             txn->request->data,
-                             txn->request->size,
-                             reply_buf, sizeof reply_buf,
-                             &reply_len, NULL)) {
+        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 < 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;
-        }
+        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;
+            /* 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));
+            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;
             }
-        } 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);
+
+            /* 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;
                 }
-                memcpy(txn->reply->data, reply_buf, reply_len);
-                txn->reply->size = reply_len;
             }
+            ofpbuf_uninit(&tmp_reply);
         }
-        ofpbuf_uninit(&tmp_reply);
 
         /* Count the number of successful transactions. */
         (*done)++;
@@ -945,6 +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;
+            }
         }
     }
 }
@@ -1187,10 +1165,17 @@ pend_io_request(struct nl_sock *sock)
     struct ovs_header *ovs_header;
     struct nlmsghdr *nlmsg;
     uint32_t seq;
-    int retval;
+    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);
@@ -1199,7 +1184,7 @@ pend_io_request(struct nl_sock *sock)
 
     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);
+                          cmd, OVS_WIN_CONTROL_VERSION);
     nlmsg = nl_msg_nlmsghdr(&request);
     nlmsg->nlmsg_seq = seq;
     nlmsg->nlmsg_pid = sock->pid;
@@ -1215,13 +1200,10 @@ pend_io_request(struct nl_sock *sock)
         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();
+        retval = EAGAIN;
     }
-    retval = 0;
 
 done:
     ofpbuf_uninit(&request);
@@ -1237,15 +1219,21 @@ nl_sock_wait(const struct nl_sock *sock, short int events)
 {
 #ifdef _WIN32
     if (sock->overlapped.Internal != STATUS_PENDING) {
-        pend_io_request(CONST_CAST(struct nl_sock *, sock));
-       /* XXX: poll_wevent_wait(sock->overlapped.hEvent); */
+        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);
     }
-    poll_immediate_wake(); /* XXX: temporary. */
 #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().
  *
@@ -1256,13 +1244,9 @@ nl_sock_wait(const struct nl_sock *sock, short int events)
 int
 nl_sock_fd(const struct nl_sock *sock)
 {
-#ifdef _WIN32
-    BUILD_ASSERT_DECL(sizeof sock->handle == sizeof(int));
-    return (int)sock->handle;
-#else
     return sock->fd;
-#endif
 }
+#endif
 
 /* Returns the PID associated with this socket. */
 uint32_t
@@ -1803,15 +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);
 }