/*
- * Copyright (c) 2009 Nicira Networks.
+ * Copyright (c) 2009, 2010 Nicira Networks.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <arpa/inet.h>
#include <inttypes.h>
#include <linux/if_tun.h>
+#include <linux/ip.h>
#include <linux/types.h>
#include <linux/ethtool.h>
+#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
#include <linux/version.h>
#include <netpacket/packet.h>
#include <net/ethernet.h>
#include <net/if.h>
+#include <linux/if_tunnel.h>
#include <net/if_arp.h>
#include <net/if_packet.h>
#include <net/route.h>
#include "netlink.h"
#include "ofpbuf.h"
#include "openflow/openflow.h"
+#include "openvswitch/internal_dev.h"
+#include "openvswitch/gre.h"
#include "packets.h"
#include "poll-loop.h"
#include "rtnetlink.h"
#define ADVERTISED_Asym_Pause (1 << 14)
#endif
-struct netdev_linux {
- struct netdev netdev;
-
- /* File descriptors. For ordinary network devices, the two fds below are
- * the same; for tap devices, they differ. */
- int netdev_fd; /* Network device. */
- int tap_fd; /* TAP character device, if any, otherwise the
- * network device. */
-
- struct netdev_linux_cache *cache;
-};
+static struct rtnetlink_notifier netdev_linux_cache_notifier;
+static int cache_notifier_refcount;
enum {
VALID_IFINDEX = 1 << 0,
VALID_IN4 = 1 << 2,
VALID_IN6 = 1 << 3,
VALID_MTU = 1 << 4,
- VALID_CARRIER = 1 << 5
+ VALID_CARRIER = 1 << 5,
+ VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */
+ VALID_POLICING = 1 << 7
+};
+
+struct tap_state {
+ int fd;
};
-/* Cached network device information. */
-struct netdev_linux_cache {
+struct netdev_dev_linux {
+ struct netdev_dev netdev_dev;
+
struct shash_node *shash_node;
- unsigned int valid;
- int ref_cnt;
+ unsigned int cache_valid;
+ /* The following are figured out "on demand" only. They are only valid
+ * when the corresponding VALID_* bit in 'cache_valid' is set. */
int ifindex;
uint8_t etheraddr[ETH_ADDR_LEN];
struct in_addr address, netmask;
struct in6_addr in6;
int mtu;
int carrier;
+ bool is_internal; /* Is this an openvswitch internal device? */
+ bool is_tap; /* Is this a tuntap device? */
+ uint32_t kbits_rate; /* Policing data. */
+ uint32_t kbits_burst;
+
+ union {
+ struct tap_state tap;
+ } state;
};
-static struct shash cache_map = SHASH_INITIALIZER(&cache_map);
-static struct rtnetlink_notifier netdev_linux_cache_notifier;
+struct netdev_linux {
+ struct netdev netdev;
+ int fd;
+};
/* An AF_INET socket (used for ioctl operations). */
static int af_inet_sock = -1;
* additional log messages. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
-static int netdev_linux_do_ethtool(struct netdev *, struct ethtool_cmd *,
+static int netdev_linux_init(void);
+
+static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
int cmd, const char *cmd_name);
-static int netdev_linux_do_ioctl(const struct netdev *, struct ifreq *,
- int cmd, const char *cmd_name);
+static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
+ const char *cmd_name);
static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
int cmd, const char *cmd_name);
static int get_flags(const struct netdev *, int *flagsp);
const uint8_t[ETH_ADDR_LEN]);
static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
+static int get_rtnl_sock(struct nl_sock **);
+
+static bool
+is_netdev_linux_class(const struct netdev_class *netdev_class)
+{
+ return netdev_class->init == netdev_linux_init;
+}
+
+static struct netdev_dev_linux *
+netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
+{
+ const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
+ assert(is_netdev_linux_class(netdev_class));
+
+ return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
+}
static struct netdev_linux *
netdev_linux_cast(const struct netdev *netdev)
{
- netdev_assert_class(netdev, &netdev_linux_class);
+ struct netdev_dev *netdev_dev = netdev_get_dev(netdev);
+ const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev);
+ assert(is_netdev_linux_class(netdev_class));
+
return CONTAINER_OF(netdev, struct netdev_linux, netdev);
}
static void
netdev_linux_cache_cb(const struct rtnetlink_change *change,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
- struct netdev_linux_cache *cache;
+ struct netdev_dev_linux *dev;
if (change) {
- cache = shash_find_data(&cache_map, change->ifname);
- if (cache) {
- cache->valid = 0;
+ struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
+ if (base_dev) {
+ const struct netdev_class *netdev_class =
+ netdev_dev_get_class(base_dev);
+
+ if (is_netdev_linux_class(netdev_class)) {
+ dev = netdev_dev_linux_cast(base_dev);
+ dev->cache_valid = 0;
+ }
}
} else {
+ struct shash device_shash;
struct shash_node *node;
- SHASH_FOR_EACH (node, &cache_map) {
- cache = node->data;
- cache->valid = 0;
+
+ shash_init(&device_shash);
+ netdev_dev_get_devices(&netdev_linux_class, &device_shash);
+ SHASH_FOR_EACH (node, &device_shash) {
+ dev = node->data;
+ dev->cache_valid = 0;
}
+ shash_destroy(&device_shash);
}
}
+/* Creates the netdev device of 'type' with 'name'. */
static int
-netdev_linux_open(const char *name, char *suffix, int ethertype,
- struct netdev **netdevp)
+netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
+ const struct shash *args, struct netdev_dev **netdev_devp)
{
- struct netdev_linux *netdev;
- enum netdev_flags flags;
+ struct netdev_dev_linux *netdev_dev;
int error;
- /* Allocate network device. */
- netdev = xcalloc(1, sizeof *netdev);
- netdev_init(&netdev->netdev, suffix, &netdev_linux_class);
- netdev->netdev_fd = -1;
- netdev->tap_fd = -1;
- netdev->cache = shash_find_data(&cache_map, suffix);
- if (!netdev->cache) {
- if (shash_is_empty(&cache_map)) {
- int error = rtnetlink_notifier_register(
- &netdev_linux_cache_notifier, netdev_linux_cache_cb, NULL);
- if (error) {
- netdev_close(&netdev->netdev);
- return error;
- }
+ if (!shash_is_empty(args)) {
+ VLOG_WARN("%s: arguments for system devices should be empty", name);
+ }
+
+ if (!cache_notifier_refcount) {
+ error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
+ netdev_linux_cache_cb, NULL);
+ if (error) {
+ return error;
}
- netdev->cache = xmalloc(sizeof *netdev->cache);
- netdev->cache->shash_node = shash_add(&cache_map, suffix,
- netdev->cache);
- netdev->cache->valid = 0;
- netdev->cache->ref_cnt = 0;
}
- netdev->cache->ref_cnt++;
+ cache_notifier_refcount++;
- if (!strncmp(name, "tap:", 4)) {
- static const char tap_dev[] = "/dev/net/tun";
- struct ifreq ifr;
+ netdev_dev = xzalloc(sizeof *netdev_dev);
+ netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
- /* Open tap device. */
- netdev->tap_fd = open(tap_dev, O_RDWR);
- if (netdev->tap_fd < 0) {
- error = errno;
- VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
- goto error;
- }
+ *netdev_devp = &netdev_dev->netdev_dev;
+ return 0;
+}
- /* Create tap device. */
- ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
- error = netdev_linux_do_ioctl(&netdev->netdev, &ifr,
- TUNSETIFF, "TUNSETIFF");
- if (error) {
- goto error;
- }
+/* For most types of netdevs we open the device for each call of
+ * netdev_open(). However, this is not the case with tap devices,
+ * since it is only possible to open the device once. In this
+ * situation we share a single file descriptor, and consequently
+ * buffers, across all readers. Therefore once data is read it will
+ * be unavailable to other reads for tap devices. */
+static int
+netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
+ const struct shash *args, struct netdev_dev **netdev_devp)
+{
+ struct netdev_dev_linux *netdev_dev;
+ struct tap_state *state;
+ static const char tap_dev[] = "/dev/net/tun";
+ struct ifreq ifr;
+ int error;
- /* Make non-blocking. */
- error = set_nonblocking(netdev->tap_fd);
- if (error) {
- goto error;
+ if (!shash_is_empty(args)) {
+ VLOG_WARN("%s: arguments for TAP devices should be empty", name);
+ }
+
+ netdev_dev = xzalloc(sizeof *netdev_dev);
+ state = &netdev_dev->state.tap;
+
+ /* Open tap device. */
+ state->fd = open(tap_dev, O_RDWR);
+ if (state->fd < 0) {
+ error = errno;
+ VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
+ goto error;
+ }
+
+ /* Create tap device. */
+ ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
+ strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
+ if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
+ VLOG_WARN("%s: creating tap device failed: %s", name,
+ strerror(errno));
+ error = errno;
+ goto error;
+ }
+
+ /* Make non-blocking. */
+ error = set_nonblocking(state->fd);
+ if (error) {
+ goto error;
+ }
+
+ netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
+ *netdev_devp = &netdev_dev->netdev_dev;
+ return 0;
+
+error:
+ free(netdev_dev);
+ return error;
+}
+
+static void
+destroy_tap(struct netdev_dev_linux *netdev_dev)
+{
+ struct tap_state *state = &netdev_dev->state.tap;
+
+ if (state->fd >= 0) {
+ close(state->fd);
+ }
+}
+
+/* Destroys the netdev device 'netdev_dev_'. */
+static void
+netdev_linux_destroy(struct netdev_dev *netdev_dev_)
+{
+ struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
+ const char *type = netdev_dev_get_type(netdev_dev_);
+
+ if (!strcmp(type, "system")) {
+ cache_notifier_refcount--;
+
+ if (!cache_notifier_refcount) {
+ rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
}
+ } else if (!strcmp(type, "tap")) {
+ destroy_tap(netdev_dev);
}
+ free(netdev_dev);
+}
+
+static int
+netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
+ struct netdev **netdevp)
+{
+ struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
+ struct netdev_linux *netdev;
+ enum netdev_flags flags;
+ int error;
+
+ /* Allocate network device. */
+ netdev = xzalloc(sizeof *netdev);
+ netdev->fd = -1;
+ netdev_init(&netdev->netdev, netdev_dev_);
+
error = netdev_get_flags(&netdev->netdev, &flags);
if (error == ENODEV) {
goto error;
}
- if (netdev->tap_fd >= 0 || ethertype != NETDEV_ETH_TYPE_NONE) {
+ if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
+ netdev->fd = netdev_dev->state.tap.fd;
+ } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
struct sockaddr_ll sll;
int protocol;
int ifindex;
protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
: ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
: ethertype);
- netdev->netdev_fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
- if (netdev->netdev_fd < 0) {
+ netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
+ if (netdev->fd < 0) {
error = errno;
goto error;
}
- if (netdev->tap_fd < 0) {
- netdev->tap_fd = netdev->netdev_fd;
- }
/* Set non-blocking mode. */
- error = set_nonblocking(netdev->netdev_fd);
+ error = set_nonblocking(netdev->fd);
if (error) {
goto error;
}
memset(&sll, 0, sizeof sll);
sll.sll_family = AF_PACKET;
sll.sll_ifindex = ifindex;
- if (bind(netdev->netdev_fd,
+ if (bind(netdev->fd,
(struct sockaddr *) &sll, sizeof sll) < 0) {
error = errno;
- VLOG_ERR("bind to %s failed: %s", suffix, strerror(error));
+ VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
+ strerror(error));
goto error;
}
* packets of the requested type on all system interfaces. We do not
* want to receive that data, but there is no way to avoid it. So we
* must now drain out the receive queue. */
- error = drain_rcvbuf(netdev->netdev_fd);
+ error = drain_rcvbuf(netdev->fd);
if (error) {
goto error;
}
return 0;
error:
- netdev_close(&netdev->netdev);
+ netdev_uninit(&netdev->netdev, true);
return error;
}
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->cache && !--netdev->cache->ref_cnt) {
- shash_delete(&cache_map, netdev->cache->shash_node);
- free(netdev->cache);
-
- if (shash_is_empty(&cache_map)) {
- rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
- }
- }
- if (netdev->netdev_fd >= 0) {
- close(netdev->netdev_fd);
- }
- if (netdev->tap_fd >= 0 && netdev->netdev_fd != netdev->tap_fd) {
- close(netdev->tap_fd);
+ if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
+ close(netdev->fd);
}
free(netdev);
}
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd < 0) {
+ if (netdev->fd < 0) {
/* Device was opened with NETDEV_ETH_TYPE_NONE. */
- return EAGAIN;
+ return -EAGAIN;
}
for (;;) {
- ssize_t retval = read(netdev->tap_fd, data, size);
+ ssize_t retval = read(netdev->fd, data, size);
if (retval >= 0) {
return retval;
} else if (errno != EINTR) {
VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
strerror(errno), netdev_get_name(netdev_));
}
- return errno;
+ return -errno;
}
}
}
netdev_linux_recv_wait(struct netdev *netdev_)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd >= 0) {
- poll_fd_wait(netdev->tap_fd, POLLIN);
+ if (netdev->fd >= 0) {
+ poll_fd_wait(netdev->fd, POLLIN);
}
}
netdev_linux_drain(struct netdev *netdev_)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
+ if (netdev->fd < 0) {
return 0;
- } else if (netdev->tap_fd != netdev->netdev_fd) {
+ } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
struct ifreq ifr;
- int error = netdev_linux_do_ioctl(netdev_, &ifr,
+ int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
if (error) {
return error;
}
- drain_fd(netdev->tap_fd, ifr.ifr_qlen);
+ drain_fd(netdev->fd, ifr.ifr_qlen);
return 0;
} else {
- return drain_rcvbuf(netdev->netdev_fd);
+ return drain_rcvbuf(netdev->fd);
}
}
/* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
*/
- if (netdev->tap_fd < 0) {
+ if (netdev->fd < 0) {
return EPIPE;
}
for (;;) {
- ssize_t retval = write(netdev->tap_fd, data, size);
+ ssize_t retval = write(netdev->fd, data, size);
if (retval < 0) {
/* The Linux AF_PACKET implementation never blocks waiting for room
* for packets, instead returning ENOBUFS. Translate this into
netdev_linux_send_wait(struct netdev *netdev_)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (netdev->tap_fd < 0 && netdev->netdev_fd < 0) {
+ if (netdev->fd < 0) {
/* Nothing to do. */
- } else if (netdev->tap_fd == netdev->netdev_fd) {
- poll_fd_wait(netdev->tap_fd, POLLOUT);
+ } else if (strcmp(netdev_get_type(netdev_), "tap")) {
+ poll_fd_wait(netdev->fd, POLLOUT);
} else {
/* TAP device always accepts packets.*/
poll_immediate_wake();
netdev_linux_set_etheraddr(struct netdev *netdev_,
const uint8_t mac[ETH_ADDR_LEN])
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- int error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
- if (!error) {
- memcpy(netdev->cache->etheraddr, mac, ETH_ADDR_LEN);
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
+ int error;
+
+ if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
+ || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
+ error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
+ if (!error) {
+ netdev_dev->cache_valid |= VALID_ETHERADDR;
+ memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
+ }
+ } else {
+ error = 0;
}
return error;
}
netdev_linux_get_etheraddr(const struct netdev *netdev_,
uint8_t mac[ETH_ADDR_LEN])
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (!(netdev->cache->valid & VALID_ETHERADDR)) {
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
+ if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
int error = get_etheraddr(netdev_get_name(netdev_),
- netdev->cache->etheraddr);
+ netdev_dev->etheraddr);
if (error) {
return error;
}
- netdev->cache->valid |= VALID_ETHERADDR;
+ netdev_dev->cache_valid |= VALID_ETHERADDR;
}
- memcpy(mac, netdev->cache->etheraddr, ETH_ADDR_LEN);
+ memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
return 0;
}
static int
netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (!(netdev->cache->valid & VALID_MTU)) {
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
+ if (!(netdev_dev->cache_valid & VALID_MTU)) {
struct ifreq ifr;
int error;
- error = netdev_linux_do_ioctl(netdev_, &ifr, SIOCGIFMTU, "SIOCGIFMTU");
+ error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
+ SIOCGIFMTU, "SIOCGIFMTU");
if (error) {
return error;
}
- netdev->cache->mtu = ifr.ifr_mtu;
- netdev->cache->valid |= VALID_MTU;
+ netdev_dev->mtu = ifr.ifr_mtu;
+ netdev_dev->cache_valid |= VALID_MTU;
}
- *mtup = netdev->cache->mtu;
+ *mtup = netdev_dev->mtu;
return 0;
}
+/* Returns the ifindex of 'netdev', if successful, as a positive number.
+ * On failure, returns a negative errno value. */
+static int
+netdev_linux_get_ifindex(const struct netdev *netdev)
+{
+ int ifindex, error;
+
+ error = get_ifindex(netdev, &ifindex);
+ return error ? -error : ifindex;
+}
+
static int
netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
int error = 0;
char *fn = NULL;
int fd = -1;
- if (!(netdev->cache->valid & VALID_CARRIER)) {
+ if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
char line[8];
int retval;
- fn = xasprintf("/sys/class/net/%s/carrier", netdev_get_name(netdev_));
+ fn = xasprintf("/sys/class/net/%s/carrier",
+ netdev_get_name(netdev_));
fd = open(fn, O_RDONLY);
if (fd < 0) {
error = errno;
fn, line[0]);
goto exit;
}
- netdev->cache->carrier = line[0] != '0';
- netdev->cache->valid |= VALID_CARRIER;
+ netdev_dev->carrier = line[0] != '0';
+ netdev_dev->cache_valid |= VALID_CARRIER;
}
- *carrier = netdev->cache->carrier;
+ *carrier = netdev_dev->carrier;
error = 0;
exit:
}
}
+/* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */
+static void
+netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev)
+{
+ if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) {
+ const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev);
+ const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev);
+
+ netdev_dev->is_tap = !strcmp(type, "tap");
+ netdev_dev->is_internal = false;
+ if (!netdev_dev->is_tap) {
+ struct ethtool_drvinfo drvinfo;
+ int error;
+
+ memset(&drvinfo, 0, sizeof drvinfo);
+ error = netdev_linux_do_ethtool(name,
+ (struct ethtool_cmd *)&drvinfo,
+ ETHTOOL_GDRVINFO,
+ "ETHTOOL_GDRVINFO");
+
+ if (!error && !strcmp(drvinfo.driver, "openvswitch")) {
+ netdev_dev->is_internal = true;
+ }
+ }
+
+ netdev_dev->cache_valid |= VALID_IS_PSEUDO;
+ }
+}
+
/* Retrieves current device stats for 'netdev'.
*
* XXX All of the members of struct netdev_stats are 64 bits wide, but on
* 32-bit architectures the Linux network stats are only 32 bits. */
static int
-netdev_linux_get_stats(const struct netdev *netdev, struct netdev_stats *stats)
+netdev_linux_get_stats(const struct netdev *netdev_,
+ struct netdev_stats *stats)
{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
static int use_netlink_stats = -1;
int error;
+ struct netdev_stats raw_stats;
+ struct netdev_stats *collect_stats = stats;
COVERAGE_INC(netdev_get_stats);
+
+ netdev_linux_update_is_pseudo(netdev_dev);
+ if (netdev_dev->is_internal) {
+ collect_stats = &raw_stats;
+ }
+
if (use_netlink_stats < 0) {
use_netlink_stats = check_for_working_netlink_stats();
}
if (use_netlink_stats) {
int ifindex;
- error = get_ifindex(netdev, &ifindex);
+ error = get_ifindex(netdev_, &ifindex);
if (!error) {
- error = get_stats_via_netlink(ifindex, stats);
+ error = get_stats_via_netlink(ifindex, collect_stats);
}
} else {
- error = get_stats_via_proc(netdev->name, stats);
+ error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
+ }
+
+ /* If this port is an internal port then the transmit and receive stats
+ * will appear to be swapped relative to the other ports since we are the
+ * one sending the data, not a remote computer. For consistency, we swap
+ * them back here. */
+ if (!error && (netdev_dev->is_internal || netdev_dev->is_tap)) {
+ stats->rx_packets = raw_stats.tx_packets;
+ stats->tx_packets = raw_stats.rx_packets;
+ stats->rx_bytes = raw_stats.tx_bytes;
+ stats->tx_bytes = raw_stats.rx_bytes;
+ stats->rx_errors = raw_stats.tx_errors;
+ stats->tx_errors = raw_stats.rx_errors;
+ stats->rx_dropped = raw_stats.tx_dropped;
+ stats->tx_dropped = raw_stats.rx_dropped;
+ stats->multicast = raw_stats.multicast;
+ stats->collisions = raw_stats.collisions;
+ stats->rx_length_errors = 0;
+ stats->rx_over_errors = 0;
+ stats->rx_crc_errors = 0;
+ stats->rx_frame_errors = 0;
+ stats->rx_fifo_errors = 0;
+ stats->rx_missed_errors = 0;
+ stats->tx_aborted_errors = 0;
+ stats->tx_carrier_errors = 0;
+ stats->tx_fifo_errors = 0;
+ stats->tx_heartbeat_errors = 0;
+ stats->tx_window_errors = 0;
}
+
return error;
}
+static int
+netdev_linux_set_stats(struct netdev *netdev,
+ const struct netdev_stats *stats)
+{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
+ struct internal_dev_stats dp_dev_stats;
+ struct ifreq ifr;
+
+ /* We must reject this call if 'netdev' is not an Open vSwitch internal
+ * port, because the ioctl that we are about to execute is in the "device
+ * private ioctls" range, which means that executing it on a device that
+ * is not the type we expect could do any random thing.
+ *
+ * (Amusingly, these ioctl numbers are commented "THESE IOCTLS ARE
+ * _DEPRECATED_ AND WILL DISAPPEAR IN 2.5.X" in linux/sockios.h. I guess
+ * DaveM is a little behind on that.) */
+ netdev_linux_update_is_pseudo(netdev_dev);
+ if (!netdev_dev->is_internal) {
+ return EOPNOTSUPP;
+ }
+
+ /* This actually only sets the *offset* that the dp_dev applies, but in our
+ * usage for fake bond devices the dp_dev never has any traffic of it own
+ * so it has the same effect. */
+ dp_dev_stats.rx_packets = stats->rx_packets;
+ dp_dev_stats.rx_bytes = stats->rx_bytes;
+ dp_dev_stats.tx_packets = stats->tx_packets;
+ dp_dev_stats.tx_bytes = stats->tx_bytes;
+ ifr.ifr_data = (void *) &dp_dev_stats;
+ return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr,
+ INTERNAL_DEV_SET_STATS,
+ "INTERNAL_DEV_SET_STATS");
+}
+
/* Stores the features supported by 'netdev' into each of '*current',
* '*advertised', '*supported', and '*peer' that are non-null. Each value is a
* bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if
- * successful, otherwise a positive errno value. On failure, all of the
- * passed-in values are set to 0. */
+ * successful, otherwise a positive errno value. */
static int
netdev_linux_get_features(struct netdev *netdev,
uint32_t *current, uint32_t *advertised,
int error;
memset(&ecmd, 0, sizeof ecmd);
- error = netdev_linux_do_ethtool(netdev, &ecmd,
+ error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
ETHTOOL_GSET, "ETHTOOL_GSET");
if (error) {
return error;
int error;
memset(&ecmd, 0, sizeof ecmd);
- error = netdev_linux_do_ethtool(netdev, &ecmd,
+ error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
ETHTOOL_GSET, "ETHTOOL_GSET");
if (error) {
return error;
if (advertise & OFPPF_PAUSE_ASYM) {
ecmd.advertising |= ADVERTISED_Asym_Pause;
}
- return netdev_linux_do_ethtool(netdev, &ecmd,
+ return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
ETHTOOL_SSET, "ETHTOOL_SSET");
}
#define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
#define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1"
-/* We redirect stderr to /dev/null because we often want to remove all
- * traffic control configuration on a port so its in a known state. If
- * this done when there is no such configuration, tc complains, so we just
- * always ignore it.
+
+/* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a
+ * positive errno value.
+ *
+ * This function is equivalent to running
+ * /sbin/tc qdisc del dev %s handle ffff: ingress
+ * but it is much, much faster.
*/
-#define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
+static int
+netdev_linux_remove_policing(struct netdev *netdev)
+{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
+ const char *netdev_name = netdev_get_name(netdev);
+
+ struct ofpbuf request;
+ struct ofpbuf *reply;
+ struct tcmsg *tcmsg;
+ struct nl_sock *rtnl_sock;
+ int ifindex;
+ int error;
+
+ error = get_ifindex(netdev, &ifindex);
+ if (error) {
+ return error;
+ }
+
+ error = get_rtnl_sock(&rtnl_sock);
+ if (error) {
+ return error;
+ }
+
+ ofpbuf_init(&request, 0);
+ nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *tcmsg,
+ RTM_DELQDISC, NLM_F_REQUEST);
+ tcmsg = ofpbuf_put_zeros(&request, sizeof *tcmsg);
+ tcmsg->tcm_family = AF_UNSPEC;
+ tcmsg->tcm_ifindex = ifindex;
+ tcmsg->tcm_handle = 0xffff0000;
+ tcmsg->tcm_parent = TC_H_INGRESS;
+ nl_msg_put_string(&request, TCA_KIND, "ingress");
+ nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0);
+ error = nl_sock_transact(rtnl_sock, &request, &reply);
+ ofpbuf_uninit(&request);
+ ofpbuf_delete(reply);
+ if (error && error != ENOENT) {
+ VLOG_WARN_RL(&rl, "%s: removing policing failed: %s",
+ netdev_name, strerror(error));
+ return error;
+ }
+
+ netdev_dev->kbits_rate = 0;
+ netdev_dev->kbits_burst = 0;
+ netdev_dev->cache_valid |= VALID_POLICING;
+ return 0;
+}
/* Attempts to set input rate limiting (policing) policy. */
static int
netdev_linux_set_policing(struct netdev *netdev,
uint32_t kbits_rate, uint32_t kbits_burst)
{
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev));
const char *netdev_name = netdev_get_name(netdev);
char command[1024];
COVERAGE_INC(netdev_set_policing);
- if (kbits_rate) {
- if (!kbits_burst) {
- /* Default to 10 kilobits if not specified. */
- kbits_burst = 10;
- }
- /* xxx This should be more careful about only adding if it
- * xxx actually exists, as opposed to always deleting it. */
- snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
- if (system(command) == -1) {
- VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
- }
+ kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */
+ : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */
+ : kbits_burst); /* Stick with user-specified value. */
+
+ if (netdev_dev->cache_valid & VALID_POLICING
+ && netdev_dev->kbits_rate == kbits_rate
+ && netdev_dev->kbits_burst == kbits_burst) {
+ /* Assume that settings haven't changed since we last set them. */
+ return 0;
+ }
+ netdev_linux_remove_policing(netdev);
+ if (kbits_rate) {
snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
if (system(command) != 0) {
VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
netdev_name);
return -1;
}
- } else {
- snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
- if (system(command) == -1) {
- VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
- }
+
+ netdev_dev->kbits_rate = kbits_rate;
+ netdev_dev->kbits_burst = kbits_burst;
+ netdev_dev->cache_valid |= VALID_POLICING;
}
return 0;
netdev_linux_get_in4(const struct netdev *netdev_,
struct in_addr *address, struct in_addr *netmask)
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (!(netdev->cache->valid & VALID_IN4)) {
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
+
+ if (!(netdev_dev->cache_valid & VALID_IN4)) {
int error;
- error = netdev_linux_get_ipv4(netdev_, &netdev->cache->address,
+ error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
SIOCGIFADDR, "SIOCGIFADDR");
if (error) {
return error;
}
- error = netdev_linux_get_ipv4(netdev_, &netdev->cache->netmask,
+ error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
SIOCGIFNETMASK, "SIOCGIFNETMASK");
if (error) {
return error;
}
- netdev->cache->valid |= VALID_IN4;
+ netdev_dev->cache_valid |= VALID_IN4;
}
- *address = netdev->cache->address;
- *netmask = netdev->cache->netmask;
+ *address = netdev_dev->address;
+ *netmask = netdev_dev->netmask;
return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
}
netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
struct in_addr netmask)
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
int error;
error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
if (!error) {
- netdev->cache->valid |= VALID_IN4;
- netdev->cache->address = address;
- netdev->cache->netmask = netmask;
+ netdev_dev->cache_valid |= VALID_IN4;
+ netdev_dev->address = address;
+ netdev_dev->netmask = netmask;
if (address.s_addr != INADDR_ANY) {
error = do_set_addr(netdev_, SIOCSIFNETMASK,
"SIOCSIFNETMASK", netmask);
static int
netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- if (!(netdev->cache->valid & VALID_IN6)) {
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
+ if (!(netdev_dev->cache_valid & VALID_IN6)) {
FILE *file;
char line[128];
- netdev->cache->in6 = in6addr_any;
+ netdev_dev->in6 = in6addr_any;
file = fopen("/proc/net/if_inet6", "r");
if (file != NULL) {
if (parse_if_inet6_line(line, &in6, ifname)
&& !strcmp(name, ifname))
{
- netdev->cache->in6 = in6;
+ netdev_dev->in6 = in6;
break;
}
}
fclose(file);
}
- netdev->cache->valid |= VALID_IN6;
+ netdev_dev->cache_valid |= VALID_IN6;
}
- *in6 = netdev->cache->in6;
+ *in6 = netdev_dev->in6;
return 0;
}
int ioctl_nr, const char *ioctl_name, struct in_addr addr)
{
struct ifreq ifr;
- strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
+ strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
make_in4_sockaddr(&ifr.ifr_addr, addr);
- return netdev_linux_do_ioctl(netdev, &ifr, ioctl_nr, ioctl_name);
+
+ return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
+ ioctl_name);
}
/* Adds 'router' as a default IP gateway. */
static int
-netdev_linux_add_router(struct netdev *netdev UNUSED, struct in_addr router)
+netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
{
struct in_addr any = { INADDR_ANY };
struct rtentry rt;
uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
{
struct arpreq r;
- struct sockaddr_in *pa;
+ struct sockaddr_in sin;
int retval;
memset(&r, 0, sizeof r);
- pa = (struct sockaddr_in *) &r.arp_pa;
- pa->sin_family = AF_INET;
- pa->sin_addr.s_addr = ip;
- pa->sin_port = 0;
+ sin.sin_family = AF_INET;
+ sin.sin_addr.s_addr = ip;
+ sin.sin_port = 0;
+ memcpy(&r.arp_pa, &sin, sizeof sin);
r.arp_ha.sa_family = ARPHRD_ETHER;
r.arp_flags = 0;
- strncpy(r.arp_dev, netdev->name, sizeof r.arp_dev);
+ strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
COVERAGE_INC(netdev_arp_lookup);
retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
if (!retval) {
memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
} else if (retval != ENXIO) {
VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
- netdev->name, IP_ARGS(&ip), strerror(retval));
+ netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
}
return retval;
}
static void
netdev_linux_poll_cb(const struct rtnetlink_change *change,
- void *aux UNUSED)
+ void *aux OVS_UNUSED)
{
if (change) {
struct list *list = shash_find_data(&netdev_linux_notifiers,
}
const struct netdev_class netdev_linux_class = {
- "", /* prefix */
- "linux", /* name */
+ "system",
netdev_linux_init,
netdev_linux_run,
netdev_linux_wait,
+ netdev_linux_create_system,
+ netdev_linux_destroy,
+ NULL, /* reconfigure */
+
netdev_linux_open,
netdev_linux_close,
netdev_linux_set_etheraddr,
netdev_linux_get_etheraddr,
netdev_linux_get_mtu,
+ netdev_linux_get_ifindex,
netdev_linux_get_carrier,
netdev_linux_get_stats,
+ netdev_linux_set_stats,
netdev_linux_get_features,
netdev_linux_set_advertisements,
};
const struct netdev_class netdev_tap_class = {
- "tap", /* prefix */
- "tap", /* name */
+ "tap",
netdev_linux_init,
- NULL, /* run */
- NULL, /* wait */
+ netdev_linux_run,
+ netdev_linux_wait,
+
+ netdev_linux_create_tap,
+ netdev_linux_destroy,
+ NULL, /* reconfigure */
netdev_linux_open,
netdev_linux_close,
- netdev_linux_enumerate,
+ NULL, /* enumerate */
netdev_linux_recv,
netdev_linux_recv_wait,
netdev_linux_set_etheraddr,
netdev_linux_get_etheraddr,
netdev_linux_get_mtu,
+ netdev_linux_get_ifindex,
netdev_linux_get_carrier,
netdev_linux_get_stats,
+ NULL, /* set_stats */
netdev_linux_get_features,
netdev_linux_set_advertisements,
netdev_linux_poll_add,
netdev_linux_poll_remove,
};
+
\f
static int
get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
.min_len = sizeof(struct rtnl_link_stats) },
};
-
- static struct nl_sock *rtnl_sock;
+ struct nl_sock *rtnl_sock;
struct ofpbuf request;
struct ofpbuf *reply;
struct ifinfomsg *ifi;
struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
int error;
- if (!rtnl_sock) {
- error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
- if (error) {
- VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
- strerror(error));
- return error;
- }
+ error = get_rtnl_sock(&rtnl_sock);
+ if (error) {
+ return error;
}
ofpbuf_init(&request, 0);
if (!attrs[IFLA_STATS]) {
VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
+ ofpbuf_delete(reply);
return EPROTO;
}
stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
stats->tx_window_errors = rtnl_stats->tx_window_errors;
+ ofpbuf_delete(reply);
+
return 0;
}
struct ifreq ifr;
int error;
- error = netdev_linux_do_ioctl(netdev, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS");
+ error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
+ "SIOCGIFFLAGS");
*flags = ifr.ifr_flags;
return error;
}
struct ifreq ifr;
ifr.ifr_flags = flags;
- return netdev_linux_do_ioctl(netdev, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS");
+ return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
+ "SIOCSIFFLAGS");
}
static int
static int
get_ifindex(const struct netdev *netdev_, int *ifindexp)
{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ struct netdev_dev_linux *netdev_dev =
+ netdev_dev_linux_cast(netdev_get_dev(netdev_));
*ifindexp = 0;
- if (!(netdev->cache->valid & VALID_IFINDEX)) {
+ if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
int ifindex = do_get_ifindex(netdev_get_name(netdev_));
if (ifindex < 0) {
return -ifindex;
}
- netdev->cache->valid |= VALID_IFINDEX;
- netdev->cache->ifindex = ifindex;
+ netdev_dev->cache_valid |= VALID_IFINDEX;
+ netdev_dev->ifindex = ifindex;
}
- *ifindexp = netdev->cache->ifindex;
+ *ifindexp = netdev_dev->ifindex;
return 0;
}
}
static int
-netdev_linux_do_ethtool(struct netdev *netdev, struct ethtool_cmd *ecmd,
+netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
int cmd, const char *cmd_name)
{
struct ifreq ifr;
memset(&ifr, 0, sizeof ifr);
- strncpy(ifr.ifr_name, netdev->name, sizeof ifr.ifr_name);
+ strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
ifr.ifr_data = (caddr_t) ecmd;
ecmd->cmd = cmd;
} else {
if (errno != EOPNOTSUPP) {
VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
- "failed: %s", cmd_name, netdev->name,
- strerror(errno));
+ "failed: %s", cmd_name, name, strerror(errno));
} else {
/* The device doesn't support this operation. That's pretty
* common, so there's no point in logging anything. */
}
static int
-netdev_linux_do_ioctl(const struct netdev *netdev, struct ifreq *ifr,
- int cmd, const char *cmd_name)
+netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
+ const char *cmd_name)
{
- strncpy(ifr->ifr_name, netdev_get_name(netdev), sizeof ifr->ifr_name);
+ strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
if (ioctl(af_inet_sock, cmd, ifr) == -1) {
- VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s",
- netdev_get_name(netdev), cmd_name, strerror(errno));
+ VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
+ strerror(errno));
return errno;
}
return 0;
int error;
ifr.ifr_addr.sa_family = AF_INET;
- error = netdev_linux_do_ioctl(netdev, &ifr, cmd, cmd_name);
+ error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
if (!error) {
const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
*ip = sin->sin_addr;
}
return error;
}
+
+/* Obtains a Netlink routing socket that is not subscribed to any multicast
+ * groups. Returns 0 if successful, otherwise a positive errno value. Stores
+ * the socket in '*rtnl_sockp' if successful, otherwise a null pointer. */
+static int
+get_rtnl_sock(struct nl_sock **rtnl_sockp)
+{
+ static struct nl_sock *sock;
+ int error;
+
+ if (!sock) {
+ error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &sock);
+ if (error) {
+ VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
+ strerror(error));
+ }
+ } else {
+ error = 0;
+ }
+
+ *rtnl_sockp = sock;
+ return error;
+}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff