2 * Copyright (c) 2012 Giuseppe Lettieri
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include <sys/types.h>
22 #include <arpa/inet.h>
23 #include <sys/ioctl.h>
24 #include <sys/socket.h>
26 #include <net/if_arp.h>
27 #include <linux/if_tun.h>
28 #include <netinet/in.h>
33 #include "netdev-provider.h"
35 #include "ofp-print.h"
38 #include "poll-loop.h"
42 #include "socket-util.h"
46 VLOG_DEFINE_THIS_MODULE(netdev_pltap);
51 struct netdev_stats stats;
52 enum netdev_flags new_flags;
53 enum netdev_flags flags;
55 struct sockaddr_in local_addr;
58 bool valid_local_netmask;
59 bool sync_flags_needed;
60 struct list sync_list;
61 unsigned int change_seq;
64 static struct list sync_list;
66 struct netdev_rx_pltap {
71 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
73 static struct shash pltap_netdevs = SHASH_INITIALIZER(&pltap_netdevs);
75 static int netdev_pltap_construct(struct netdev *netdev_);
77 static void netdev_pltap_update_seq(struct netdev_pltap *);
78 static int get_flags(struct netdev_pltap *dev, enum netdev_flags *flags);
81 netdev_pltap_finalized(struct netdev_pltap *dev)
83 return dev->valid_local_ip && dev->valid_local_netmask;
87 is_netdev_pltap_class(const struct netdev_class *class)
89 return class->construct == netdev_pltap_construct;
92 static struct netdev_pltap *
93 netdev_pltap_cast(const struct netdev *netdev)
95 ovs_assert(is_netdev_pltap_class(netdev_get_class(netdev)));
96 return CONTAINER_OF(netdev, struct netdev_pltap, up);
99 static struct netdev_rx_pltap*
100 netdev_rx_pltap_cast(const struct netdev_rx *rx)
102 ovs_assert(is_netdev_pltap_class(netdev_get_class(rx->netdev)));
103 return CONTAINER_OF(rx, struct netdev_rx_pltap, up);
106 static void sync_needed(struct netdev_pltap *dev)
108 if (dev->sync_flags_needed)
111 dev->sync_flags_needed = true;
112 list_insert(&sync_list, &dev->sync_list);
116 static void sync_done(struct netdev_pltap *dev)
118 if (!dev->sync_flags_needed)
121 (void) list_remove(&dev->sync_list);
122 dev->sync_flags_needed = false;
125 static struct netdev *
126 netdev_pltap_alloc(void)
128 struct netdev_pltap *netdev = xzalloc(sizeof *netdev);
133 netdev_pltap_construct(struct netdev *netdev_)
135 struct netdev_pltap *netdev = netdev_pltap_cast(netdev_);
138 netdev->real_name = xzalloc(IFNAMSIZ + 1);
139 memset(&netdev->local_addr, 0, sizeof(netdev->local_addr));
140 netdev->valid_local_ip = false;
141 netdev->valid_local_netmask = false;
143 netdev->sync_flags_needed = false;
144 netdev->change_seq = 1;
145 list_init(&netdev->sync_list);
148 /* Open tap device. */
149 netdev->fd = tun_alloc(IFF_TAP, netdev->real_name);
150 if (netdev->fd < 0) {
152 VLOG_WARN("tun_alloc(IFF_TAP, %s) failed: %s",
153 netdev_get_name(netdev_), ovs_strerror(error));
156 VLOG_DBG("real_name = %s", netdev->real_name);
158 /* Make non-blocking. */
159 error = set_nonblocking(netdev->fd);
164 shash_add(&pltap_netdevs, netdev_get_name(netdev_), netdev);
169 netdev_pltap_destruct(struct netdev *netdev_)
171 struct netdev_pltap *netdev = netdev_pltap_cast(netdev_);
173 if (netdev->fd != -1)
178 shash_find_and_delete(&pltap_netdevs,
179 netdev_get_name(netdev_));
183 netdev_pltap_dealloc(struct netdev *netdev_)
185 struct netdev_pltap *netdev = netdev_pltap_cast(netdev_);
189 static int netdev_pltap_up(struct netdev_pltap *dev);
191 static struct netdev_rx *
192 netdev_pltap_rx_alloc(void)
194 struct netdev_rx_pltap *rx = xzalloc(sizeof *rx);
199 netdev_pltap_rx_construct(struct netdev_rx *rx_)
201 struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
202 struct netdev *netdev_ = rx->up.netdev;
203 struct netdev_pltap *netdev =
204 netdev_pltap_cast(netdev_);
208 if (!netdev_pltap_finalized(netdev))
210 error = netdev_pltap_up(netdev);
218 netdev_pltap_rx_destruct(struct netdev_rx *rx_ OVS_UNUSED)
223 netdev_pltap_rx_dealloc(struct netdev_rx *rx_)
225 struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
230 static int vsys_transaction(const char *script,
231 const char **preply, char *format, ...)
233 char *msg = NULL, *reply = NULL;
234 const size_t reply_size = 1024;
235 int ifd = -1, ofd = -1, maxfd;
236 size_t bytes_to_write, bytes_to_read,
237 bytes_written = 0, bytes_read = 0;
239 char *ofname = NULL, *ifname = NULL;
242 va_start(args, format);
243 msg = xvasprintf(format, args);
245 reply = (char*)xmalloc(reply_size);
246 if (!msg || !reply) {
247 VLOG_ERR("Out of memory");
252 ofname = xasprintf("/vsys/%s.out", script);
253 ifname = xasprintf("/vsys/%s.in", script);
254 if (!ofname || !ifname) {
255 VLOG_ERR("Out of memory");
260 ofd = open(ofname, O_RDONLY | O_NONBLOCK);
262 VLOG_ERR("Cannot open %s: %s", ofname, ovs_strerror(errno));
266 ifd = open(ifname, O_WRONLY | O_NONBLOCK);
268 VLOG_ERR("Cannot open %s: %s", ifname, ovs_strerror(errno));
272 maxfd = (ifd < ofd) ? ofd : ifd;
274 bytes_to_write = strlen(msg);
275 bytes_to_read = reply_size;
276 while (bytes_to_write || bytes_to_read) {
277 fd_set readset, writeset, errorset;
282 if (bytes_to_write) {
283 FD_SET(ifd, &writeset);
284 FD_SET(ifd, &errorset);
286 FD_SET(ofd, &readset);
287 FD_SET(ofd, &errorset);
288 if (select(maxfd + 1, &readset, &writeset, &errorset, NULL) < 0) {
291 VLOG_ERR("selec error: %s", ovs_strerror(errno));
295 if (FD_ISSET(ifd, &errorset) || FD_ISSET(ofd, &errorset)) {
296 VLOG_ERR("error condition on ifd or ofd");
299 if (FD_ISSET(ifd, &writeset)) {
300 ssize_t n = write(ifd, msg + bytes_written, bytes_to_write);
302 if (errno != EAGAIN && errno != EINTR) {
303 VLOG_ERR("write on %s: %s", ifname, ovs_strerror(errno));
310 if (bytes_to_write == 0)
314 if (FD_ISSET(ofd, &readset)) {
315 ssize_t n = read(ofd, reply + bytes_read, bytes_to_read);
317 if (errno != EAGAIN && errno != EINTR) {
318 VLOG_ERR("read on %s: %s", ofname, ovs_strerror(errno));
331 reply[bytes_read] = '\0';
334 reply = NULL; /* prevent freeing the reply msg */
336 VLOG_ERR("%s returned: %s", script, reply);
353 netdev_pltap_up(struct netdev_pltap *dev)
355 if (!netdev_pltap_finalized(dev)) {
359 return vsys_transaction("vif_up", NULL, "%s\n"IP_FMT"\n%d\n",
361 IP_ARGS(dev->local_addr.sin_addr.s_addr),
366 netdev_pltap_down(struct netdev_pltap *dev)
368 if (!netdev_pltap_finalized(dev)) {
372 return vsys_transaction("vif_down", NULL, "%s\n", dev->real_name);
376 netdev_pltap_promisc(struct netdev_pltap *dev, bool promisc)
378 if (!netdev_pltap_finalized(dev)) {
382 return vsys_transaction("promisc", NULL, "%s\n%s",
384 (promisc ? "" : "-\n"));
388 netdev_pltap_sync_flags(struct netdev_pltap *dev)
391 if (dev->fd < 0 || !netdev_pltap_finalized(dev)) {
396 VLOG_DBG("sync_flags(%s): current: %s %s target: %s %s",
398 (dev->flags & NETDEV_UP ? "UP" : "-"),
399 (dev->flags & NETDEV_PROMISC ? "PROMISC" : "-"),
400 (dev->new_flags & NETDEV_UP ? "UP" : "-"),
401 (dev->new_flags & NETDEV_PROMISC ? "PROMISC" : "-"));
403 if ((dev->new_flags & NETDEV_UP) && !(dev->flags & NETDEV_UP)) {
404 (void) netdev_pltap_up(dev);
405 } else if (!(dev->new_flags & NETDEV_UP) && (dev->flags & NETDEV_UP)) {
406 (void) netdev_pltap_down(dev);
409 if ((dev->new_flags & NETDEV_PROMISC) ^ (dev->flags & NETDEV_PROMISC)) {
410 (void) netdev_pltap_promisc(dev, dev->new_flags & NETDEV_PROMISC);
413 netdev_pltap_update_seq(dev);
419 netdev_pltap_get_config(const struct netdev *dev_, struct smap *args)
421 struct netdev_pltap *netdev = netdev_pltap_cast(dev_);
423 if (netdev->valid_local_ip)
424 smap_add_format(args, "local_ip", IP_FMT,
425 IP_ARGS(netdev->local_addr.sin_addr.s_addr));
426 if (netdev->valid_local_netmask)
427 smap_add_format(args, "local_netmask", "%"PRIu32,
428 ntohs(netdev->local_netmask));
433 netdev_pltap_set_config(struct netdev *dev_, const struct smap *args)
435 struct netdev_pltap *netdev = netdev_pltap_cast(dev_);
436 struct shash_node *node;
438 VLOG_DBG("pltap_set_config(%s)", netdev_get_name(dev_));
439 SMAP_FOR_EACH(node, args) {
440 VLOG_DBG("arg: %s->%s", node->name, (char*)node->data);
441 if (!strcmp(node->name, "local_ip")) {
443 if (lookup_ip(node->data, &addr)) {
444 VLOG_WARN("%s: bad 'local_ip'", node->name);
446 netdev->local_addr.sin_addr = addr;
447 netdev->valid_local_ip = true;
449 } else if (!strcmp(node->name, "local_netmask")) {
450 netdev->local_netmask = atoi(node->data);
451 // XXX check valididy
452 netdev->valid_local_netmask = true;
454 VLOG_WARN("%s: unknown argument '%s'",
455 netdev_get_name(dev_), node->name);
458 if (netdev_pltap_finalized(netdev)) {
459 netdev->new_flags |= NETDEV_UP;
466 netdev_pltap_rx_recv(struct netdev_rx *rx_, void *buffer, size_t size)
468 struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
470 struct iovec iov[2] = {
471 { .iov_base = &pi, .iov_len = sizeof(pi) },
472 { .iov_base = buffer, .iov_len = size }
476 retval = readv(rx->fd, iov, 2);
478 if (retval <= size) {
483 } else if (errno != EINTR) {
484 if (errno != EAGAIN) {
485 VLOG_WARN_RL(&rl, "error receiveing Ethernet packet on %s: %s",
486 netdev_rx_get_name(rx_), ovs_strerror(errno));
494 netdev_pltap_rx_wait(struct netdev_rx *rx_)
496 struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
497 struct netdev_pltap *netdev =
498 netdev_pltap_cast(rx->up.netdev);
499 if (rx->fd >= 0 && netdev_pltap_finalized(netdev)) {
500 poll_fd_wait(rx->fd, POLLIN);
505 netdev_pltap_send(struct netdev *netdev_, const void *buffer, size_t size)
507 struct netdev_pltap *dev =
508 netdev_pltap_cast(netdev_);
509 struct tun_pi pi = { 0, 0x86 };
510 struct iovec iov[2] = {
511 { .iov_base = &pi, .iov_len = sizeof(pi) },
512 { .iov_base = (char*) buffer, .iov_len = size }
518 retval = writev(dev->fd, iov, 2);
520 if (retval != size + 4) {
521 VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of %zu) on %s",
522 retval, size + 4, netdev_get_name(netdev_));
525 } else if (errno != EINTR) {
526 if (errno != EAGAIN) {
527 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
528 netdev_get_name(netdev_), ovs_strerror(errno));
536 netdev_pltap_send_wait(struct netdev *netdev_)
538 struct netdev_pltap *dev =
539 netdev_pltap_cast(netdev_);
540 if (dev->fd >= 0 && netdev_pltap_finalized(dev)) {
541 poll_fd_wait(dev->fd, POLLOUT);
546 netdev_pltap_rx_drain(struct netdev_rx *rx_)
548 struct netdev_rx_pltap *rx = netdev_rx_pltap_cast(rx_);
555 error = recv(rx->fd, buffer, 128, MSG_TRUNC);
557 if (error == -EAGAIN)
559 else if (error != -EMSGSIZE)
567 netdev_pltap_set_etheraddr(struct netdev *netdevi OVS_UNUSED,
568 const uint8_t mac[ETH_ADDR_LEN] OVS_UNUSED)
574 // XXX from netdev-linux.c
576 get_etheraddr(struct netdev_pltap *dev, uint8_t ea[ETH_ADDR_LEN])
582 memset(&ifr, 0, sizeof ifr);
583 ovs_strzcpy(ifr.ifr_name, dev->real_name, sizeof ifr.ifr_name);
584 error = af_inet_ifreq_ioctl(dev->real_name, &ifr,
585 SIOCGIFHWADDR, "SIOCGIFHWADDR");
589 hwaddr_family = ifr.ifr_hwaddr.sa_family;
590 if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
591 VLOG_WARN("%s device has unknown hardware address family %d",
592 dev->real_name, hwaddr_family);
594 memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
599 get_flags(struct netdev_pltap *dev, enum netdev_flags *flags)
604 error = af_inet_ifreq_ioctl(dev->real_name, &ifr,
605 SIOCGIFFLAGS, "SIOCGIFFLAGS");
610 if (ifr.ifr_flags & IFF_UP)
612 if (ifr.ifr_flags & IFF_PROMISC)
613 *flags |= NETDEV_PROMISC;
618 netdev_pltap_get_etheraddr(const struct netdev *netdev,
619 uint8_t mac[ETH_ADDR_LEN])
621 struct netdev_pltap *dev =
622 netdev_pltap_cast(netdev);
625 return get_etheraddr(dev, mac);
629 // XXX can we read stats in planetlab?
631 netdev_pltap_get_stats(const struct netdev *netdev OVS_UNUSED, struct netdev_stats *stats OVS_UNUSED)
637 netdev_pltap_set_stats(struct netdev *netdev OVS_UNUSED, const struct netdev_stats *stats OVS_UNUSED)
644 netdev_pltap_update_flags(struct netdev *dev_,
645 enum netdev_flags off, enum netdev_flags on,
646 enum netdev_flags *old_flagsp)
648 struct netdev_pltap *netdev =
649 netdev_pltap_cast(dev_);
652 if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) {
656 if (netdev_pltap_finalized(netdev)) {
657 error = get_flags(netdev, &netdev->flags);
659 *old_flagsp = netdev->flags;
660 netdev->new_flags |= on;
661 netdev->new_flags &= ~off;
662 if (netdev->flags != netdev->new_flags) {
663 /* we cannot sync here, since we may be in a signal handler */
671 netdev_pltap_change_seq(const struct netdev *netdev)
673 return netdev_pltap_cast(netdev)->change_seq;
676 /* Helper functions. */
679 netdev_pltap_update_seq(struct netdev_pltap *dev)
682 if (!dev->change_seq) {
688 netdev_pltap_get_real_name(struct unixctl_conn *conn,
689 int argc OVS_UNUSED, const char *argv[], void *aux OVS_UNUSED)
691 struct netdev_pltap *pltap_dev;
693 pltap_dev = shash_find_data(&pltap_netdevs, argv[1]);
695 unixctl_command_reply_error(conn, "no such pltap netdev");
698 if (pltap_dev->fd < 0) {
699 unixctl_command_reply_error(conn, "no real device attached");
703 unixctl_command_reply(conn, pltap_dev->real_name);
707 netdev_pltap_init(void)
709 list_init(&sync_list);
710 unixctl_command_register("netdev-pltap/get-tapname", "port",
711 1, 1, netdev_pltap_get_real_name, NULL);
716 netdev_pltap_run(void)
718 struct netdev_pltap *iter, *next;
719 LIST_FOR_EACH_SAFE(iter, next, sync_list, &sync_list) {
720 netdev_pltap_sync_flags(iter);
725 netdev_pltap_wait(void)
727 if (!list_is_empty(&sync_list)) {
728 VLOG_DBG("netdev_pltap: scheduling sync");
729 poll_immediate_wake();
733 const struct netdev_class netdev_pltap_class = {
740 netdev_pltap_construct,
741 netdev_pltap_destruct,
742 netdev_pltap_dealloc,
743 netdev_pltap_get_config,
744 netdev_pltap_set_config,
745 NULL, /* get_tunnel_config */
748 netdev_pltap_send_wait,
750 netdev_pltap_set_etheraddr,
751 netdev_pltap_get_etheraddr,
754 NULL, /* get_ifindex */
755 NULL, /* get_carrier */
756 NULL, /* get_carrier_resets */
757 NULL, /* get_miimon */
758 netdev_pltap_get_stats,
759 netdev_pltap_set_stats,
761 NULL, /* get_features */
762 NULL, /* set_advertisements */
764 NULL, /* set_policing */
765 NULL, /* get_qos_types */
766 NULL, /* get_qos_capabilities */
769 NULL, /* get_queue */
770 NULL, /* set_queue */
771 NULL, /* delete_queue */
772 NULL, /* get_queue_stats */
773 NULL, /* dump_queues */
774 NULL, /* dump_queue_stats */
779 NULL, /* add_router */
780 NULL, /* get_next_hop */
781 NULL, /* get_drv_info */
782 NULL, /* arp_lookup */
784 netdev_pltap_update_flags,
786 netdev_pltap_change_seq,
788 netdev_pltap_rx_alloc,
789 netdev_pltap_rx_construct,
790 netdev_pltap_rx_destruct,
791 netdev_pltap_rx_dealloc,
792 netdev_pltap_rx_recv,
793 netdev_pltap_rx_wait,
794 netdev_pltap_rx_drain,