/*
- * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
-#include <linux/version.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
+#include <sys/utsname.h>
#include <netpacket/packet.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <unistd.h>
#include "coverage.h"
-#include "dpif-linux.h"
+#include "dp-packet.h"
+#include "dpif-netlink.h"
+#include "dpif-netdev.h"
#include "dynamic-string.h"
#include "fatal-signal.h"
#include "hash.h"
#include "netlink.h"
#include "ofpbuf.h"
#include "openflow/openflow.h"
+#include "ovs-atomic.h"
#include "packets.h"
#include "poll-loop.h"
-#include "rtnetlink-link.h"
+#include "rtnetlink.h"
#include "shash.h"
#include "socket-util.h"
#include "sset.h"
#include "timer.h"
#include "unaligned.h"
-#include "vlog.h"
+#include "openvswitch/vlog.h"
VLOG_DEFINE_THIS_MODULE(netdev_linux);
#define TC_RTAB_SIZE 1024
#endif
+/* Linux 2.6.21 introduced struct tpacket_auxdata.
+ * Linux 2.6.27 added the tp_vlan_tci member.
+ * Linux 3.0 defined TP_STATUS_VLAN_VALID.
+ * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined
+ * TP_STATUS_VLAN_TPID_VALID.
+ *
+ * With all this churn it's easiest to unconditionally define a replacement
+ * structure that has everything we want.
+ */
+#ifndef PACKET_AUXDATA
+#define PACKET_AUXDATA 8
+#endif
+#ifndef TP_STATUS_VLAN_VALID
+#define TP_STATUS_VLAN_VALID (1 << 4)
+#endif
+#ifndef TP_STATUS_VLAN_TPID_VALID
+#define TP_STATUS_VLAN_TPID_VALID (1 << 6)
+#endif
+#undef tpacket_auxdata
+#define tpacket_auxdata rpl_tpacket_auxdata
+struct tpacket_auxdata {
+ uint32_t tp_status;
+ uint32_t tp_len;
+ uint32_t tp_snaplen;
+ uint16_t tp_mac;
+ uint16_t tp_net;
+ uint16_t tp_vlan_tci;
+ uint16_t tp_vlan_tpid;
+};
+
+/* Linux 2.6.27 introduced ethtool_cmd_speed
+ *
+ * To avoid revisiting problems reported with using configure to detect
+ * compatibility (see report at
+ * http://openvswitch.org/pipermail/dev/2014-October/047978.html)
+ * unconditionally replace ethtool_cmd_speed. */
+#define ethtool_cmd_speed rpl_ethtool_cmd_speed
+static inline uint32_t rpl_ethtool_cmd_speed(const struct ethtool_cmd *ep)
+{
+ return ep->speed | (ep->speed_hi << 16);
+}
+
+/* Linux 2.6.30 introduced supported and advertised flags for
+ * 1G base KX, and 10G base KX4, KR and R. */
+#ifndef SUPPORTED_1000baseKX_Full
+#define SUPPORTED_1000baseKX_Full (1 << 17)
+#define SUPPORTED_10000baseKX4_Full (1 << 18)
+#define SUPPORTED_10000baseKR_Full (1 << 19)
+#define SUPPORTED_10000baseR_FEC (1 << 20)
+#define ADVERTISED_1000baseKX_Full (1 << 17)
+#define ADVERTISED_10000baseKX4_Full (1 << 18)
+#define ADVERTISED_10000baseKR_Full (1 << 19)
+#define ADVERTISED_10000baseR_FEC (1 << 20)
+#endif
+
+/* Linux 3.5 introduced supported and advertised flags for
+ * 40G base KR4, CR4, SR4 and LR4. */
+#ifndef SUPPORTED_40000baseKR4_Full
+#define SUPPORTED_40000baseKR4_Full (1 << 23)
+#define SUPPORTED_40000baseCR4_Full (1 << 24)
+#define SUPPORTED_40000baseSR4_Full (1 << 25)
+#define SUPPORTED_40000baseLR4_Full (1 << 26)
+#define ADVERTISED_40000baseKR4_Full (1 << 23)
+#define ADVERTISED_40000baseCR4_Full (1 << 24)
+#define ADVERTISED_40000baseSR4_Full (1 << 25)
+#define ADVERTISED_40000baseLR4_Full (1 << 26)
+#endif
+
+/* Linux 2.6.35 introduced IFLA_STATS64 and rtnl_link_stats64.
+ *
+ * Tests for rtnl_link_stats64 don't seem to consistently work, e.g. on
+ * 2.6.32-431.29.2.el6.x86_64 (see report at
+ * http://openvswitch.org/pipermail/dev/2014-October/047978.html). Maybe
+ * if_link.h is not self-contained on those kernels. It is easiest to
+ * unconditionally define a replacement. */
+#ifndef IFLA_STATS64
+#define IFLA_STATS64 23
+#endif
+#define rtnl_link_stats64 rpl_rtnl_link_stats64
+struct rtnl_link_stats64 {
+ uint64_t rx_packets;
+ uint64_t tx_packets;
+ uint64_t rx_bytes;
+ uint64_t tx_bytes;
+ uint64_t rx_errors;
+ uint64_t tx_errors;
+ uint64_t rx_dropped;
+ uint64_t tx_dropped;
+ uint64_t multicast;
+ uint64_t collisions;
+
+ uint64_t rx_length_errors;
+ uint64_t rx_over_errors;
+ uint64_t rx_crc_errors;
+ uint64_t rx_frame_errors;
+ uint64_t rx_fifo_errors;
+ uint64_t rx_missed_errors;
+
+ uint64_t tx_aborted_errors;
+ uint64_t tx_carrier_errors;
+ uint64_t tx_fifo_errors;
+ uint64_t tx_heartbeat_errors;
+ uint64_t tx_window_errors;
+
+ uint64_t rx_compressed;
+ uint64_t tx_compressed;
+};
+
enum {
VALID_IFINDEX = 1 << 0,
VALID_ETHERADDR = 1 << 1,
static const struct tc_ops tc_ops_htb;
static const struct tc_ops tc_ops_hfsc;
+static const struct tc_ops tc_ops_codel;
+static const struct tc_ops tc_ops_fqcodel;
+static const struct tc_ops tc_ops_sfq;
static const struct tc_ops tc_ops_default;
static const struct tc_ops tc_ops_other;
static const struct tc_ops *const tcs[] = {
&tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */
&tc_ops_hfsc, /* Hierarchical fair service curve. */
+ &tc_ops_codel, /* Controlled delay */
+ &tc_ops_fqcodel, /* Fair queue controlled delay */
+ &tc_ops_sfq, /* Stochastic fair queueing */
&tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */
&tc_ops_other, /* Some other qdisc. */
NULL
unsigned int flags, struct ofpbuf *);
static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp);
static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add);
-static int tc_add_policer(struct netdev *netdev, int kbits_rate,
- int kbits_burst);
+static int tc_add_policer(struct netdev *,
+ uint32_t kbits_rate, uint32_t kbits_burst);
static int tc_parse_qdisc(const struct ofpbuf *, const char **kind,
struct nlattr **options);
struct ovs_mutex mutex;
unsigned int cache_valid;
- unsigned int change_seq;
bool miimon; /* Link status of last poll. */
long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */
/* 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 eth_addr etheraddr;
struct in_addr address, netmask;
struct in6_addr in6;
int mtu;
int netdev_policing_error; /* Cached error code from set policing. */
int get_features_error; /* Cached error code from ETHTOOL_GSET. */
int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */
+ int in4_error; /* Cached error code from reading in4 addr. */
+ int in6_error; /* Cached error code from reading in6 addr. */
enum netdev_features current; /* Cached from ETHTOOL_GSET. */
enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */
int tap_fd;
};
-struct netdev_rx_linux {
- struct netdev_rx up;
+struct netdev_rxq_linux {
+ struct netdev_rxq up;
bool is_tap;
int fd;
};
* additional log messages. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
+/* Polling miimon status for all ports causes performance degradation when
+ * handling a large number of ports. If there are no devices using miimon, then
+ * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait().
+ *
+ * Readers do not depend on this variable synchronizing with the related
+ * changes in the device miimon status, so we can use atomic_count. */
+static atomic_count miimon_cnt = ATOMIC_COUNT_INIT(0);
+
static void netdev_linux_run(void);
static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
static int do_set_addr(struct netdev *netdev,
int ioctl_nr, const char *ioctl_name,
struct in_addr addr);
-static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
-static int set_etheraddr(const char *netdev_name, 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_etheraddr(const char *netdev_name, struct eth_addr *ea);
+static int set_etheraddr(const char *netdev_name, const struct eth_addr);
+static int get_stats_via_netlink(const struct netdev *, struct netdev_stats *);
static int af_packet_sock(void);
+static bool netdev_linux_miimon_enabled(void);
static void netdev_linux_miimon_run(void);
static void netdev_linux_miimon_wait(void);
+static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup);
static bool
is_netdev_linux_class(const struct netdev_class *netdev_class)
return CONTAINER_OF(netdev, struct netdev_linux, up);
}
-static struct netdev_rx_linux *
-netdev_rx_linux_cast(const struct netdev_rx *rx)
+static struct netdev_rxq_linux *
+netdev_rxq_linux_cast(const struct netdev_rxq *rx)
{
ovs_assert(is_netdev_linux_class(netdev_get_class(rx->netdev)));
- return CONTAINER_OF(rx, struct netdev_rx_linux, up);
+ return CONTAINER_OF(rx, struct netdev_rxq_linux, up);
}
\f
static void netdev_linux_update(struct netdev_linux *netdev,
- const struct rtnetlink_link_change *)
+ const struct rtnetlink_change *)
OVS_REQUIRES(netdev->mutex);
static void netdev_linux_changed(struct netdev_linux *netdev,
unsigned int ifi_flags, unsigned int mask)
OVS_REQUIRES(netdev->mutex);
-/* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK changes, or NULL
+/* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK,
+ * RTNLGRP_IPV4_IFADDR and RTNLGRP_IPV6_IFADDR changes, or NULL
* if no such socket could be created. */
static struct nl_sock *
netdev_linux_notify_sock(void)
{
static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
static struct nl_sock *sock;
+ unsigned int mcgroups[3] = {RTNLGRP_LINK, RTNLGRP_IPV4_IFADDR,
+ RTNLGRP_IPV6_IFADDR};
if (ovsthread_once_start(&once)) {
int error;
error = nl_sock_create(NETLINK_ROUTE, &sock);
if (!error) {
- error = nl_sock_join_mcgroup(sock, RTNLGRP_LINK);
- if (error) {
- nl_sock_destroy(sock);
- sock = NULL;
+ size_t i;
+
+ for (i = 0; i < ARRAY_SIZE(mcgroups); i++) {
+ error = nl_sock_join_mcgroup(sock, mcgroups[i]);
+ if (error) {
+ nl_sock_destroy(sock);
+ sock = NULL;
+ break;
+ }
}
}
ovsthread_once_done(&once);
return sock;
}
+static bool
+netdev_linux_miimon_enabled(void)
+{
+ return atomic_count_get(&miimon_cnt) > 0;
+}
+
static void
netdev_linux_run(void)
{
struct nl_sock *sock;
int error;
- netdev_linux_miimon_run();
+ if (netdev_linux_miimon_enabled()) {
+ netdev_linux_miimon_run();
+ }
sock = netdev_linux_notify_sock();
if (!sock) {
ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
error = nl_sock_recv(sock, &buf, false);
if (!error) {
- struct rtnetlink_link_change change;
+ struct rtnetlink_change change;
- if (rtnetlink_link_parse(&buf, &change)) {
+ if (rtnetlink_parse(&buf, &change)) {
struct netdev *netdev_ = netdev_from_name(change.ifname);
if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) {
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
{
struct nl_sock *sock;
- netdev_linux_miimon_wait();
+ if (netdev_linux_miimon_enabled()) {
+ netdev_linux_miimon_wait();
+ }
sock = netdev_linux_notify_sock();
if (sock) {
nl_sock_wait(sock, POLLIN);
unsigned int ifi_flags, unsigned int mask)
OVS_REQUIRES(dev->mutex)
{
- dev->change_seq++;
- if (!dev->change_seq) {
- dev->change_seq++;
- }
+ netdev_change_seq_changed(&dev->up);
if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) {
dev->carrier_resets++;
static void
netdev_linux_update(struct netdev_linux *dev,
- const struct rtnetlink_link_change *change)
+ const struct rtnetlink_change *change)
OVS_REQUIRES(dev->mutex)
{
- if (change->nlmsg_type == RTM_NEWLINK) {
- /* Keep drv-info */
- netdev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO);
+ if (rtnetlink_type_is_rtnlgrp_link(change->nlmsg_type)){
+ if (change->nlmsg_type == RTM_NEWLINK) {
+ /* Keep drv-info, in4, in6. */
+ netdev_linux_changed(dev, change->ifi_flags,
+ VALID_DRVINFO | VALID_IN4 | VALID_IN6);
+
+ /* Update netdev from rtnl-change msg. */
+ if (change->mtu) {
+ dev->mtu = change->mtu;
+ dev->cache_valid |= VALID_MTU;
+ dev->netdev_mtu_error = 0;
+ }
- /* Update netdev from rtnl-change msg. */
- if (change->mtu) {
- dev->mtu = change->mtu;
- dev->cache_valid |= VALID_MTU;
- dev->netdev_mtu_error = 0;
- }
+ if (!eth_addr_is_zero(change->mac)) {
+ dev->etheraddr = change->mac;
+ dev->cache_valid |= VALID_ETHERADDR;
+ dev->ether_addr_error = 0;
+ }
- if (!eth_addr_is_zero(change->addr)) {
- memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN);
- dev->cache_valid |= VALID_ETHERADDR;
- dev->ether_addr_error = 0;
+ dev->ifindex = change->if_index;
+ dev->cache_valid |= VALID_IFINDEX;
+ dev->get_ifindex_error = 0;
+ } else {
+ netdev_linux_changed(dev, change->ifi_flags, 0);
}
-
- dev->ifindex = change->ifi_index;
- dev->cache_valid |= VALID_IFINDEX;
- dev->get_ifindex_error = 0;
-
+ } else if (rtnetlink_type_is_rtnlgrp_addr(change->nlmsg_type)) {
+ /* Invalidates in4, in6. */
+ netdev_linux_changed(dev, dev->ifi_flags,
+ ~(VALID_IN4 | VALID_IN6));
} else {
- netdev_linux_changed(dev, change->ifi_flags, 0);
+ OVS_NOT_REACHED();
}
}
netdev_linux_common_construct(struct netdev_linux *netdev)
{
ovs_mutex_init(&netdev->mutex);
- netdev->change_seq = 1;
}
/* Creates system and internal devices. */
close(netdev->tap_fd);
}
+ if (netdev->miimon_interval > 0) {
+ atomic_count_dec(&miimon_cnt);
+ }
+
ovs_mutex_destroy(&netdev->mutex);
}
free(netdev);
}
-static struct netdev_rx *
-netdev_linux_rx_alloc(void)
+static struct netdev_rxq *
+netdev_linux_rxq_alloc(void)
{
- struct netdev_rx_linux *rx = xzalloc(sizeof *rx);
+ struct netdev_rxq_linux *rx = xzalloc(sizeof *rx);
return &rx->up;
}
static int
-netdev_linux_rx_construct(struct netdev_rx *rx_)
+netdev_linux_rxq_construct(struct netdev_rxq *rxq_)
{
- struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
+ struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
struct netdev *netdev_ = rx->up.netdev;
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
int error;
rx->fd = netdev->tap_fd;
} else {
struct sockaddr_ll sll;
- int ifindex;
+ int ifindex, val;
/* Result of tcpdump -dd inbound */
static const struct sock_filter filt[] = {
{ 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */
goto error;
}
+ val = 1;
+ if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) {
+ error = errno;
+ VLOG_ERR("%s: failed to mark socket for auxdata (%s)",
+ netdev_get_name(netdev_), ovs_strerror(error));
+ goto error;
+ }
+
/* Set non-blocking mode. */
error = set_nonblocking(rx->fd);
if (error) {
memset(&sll, 0, sizeof sll);
sll.sll_family = AF_PACKET;
sll.sll_ifindex = ifindex;
- sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL);
+ sll.sll_protocol = htons(ETH_P_ALL);
if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) {
error = errno;
VLOG_ERR("%s: failed to bind raw socket (%s)",
}
static void
-netdev_linux_rx_destruct(struct netdev_rx *rx_)
+netdev_linux_rxq_destruct(struct netdev_rxq *rxq_)
{
- struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
+ struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
if (!rx->is_tap) {
close(rx->fd);
}
static void
-netdev_linux_rx_dealloc(struct netdev_rx *rx_)
+netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_)
{
- struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
+ struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
free(rx);
}
+static ovs_be16
+auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux)
+{
+ if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) {
+ return htons(aux->tp_vlan_tpid);
+ } else {
+ return htons(ETH_TYPE_VLAN);
+ }
+}
+
+static bool
+auxdata_has_vlan_tci(const struct tpacket_auxdata *aux)
+{
+ return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID;
+}
+
static int
-netdev_linux_rx_recv(struct netdev_rx *rx_, void *data, size_t size)
+netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer)
{
- struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
+ size_t size;
ssize_t retval;
+ struct iovec iov;
+ struct cmsghdr *cmsg;
+ union {
+ struct cmsghdr cmsg;
+ char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
+ } cmsg_buffer;
+ struct msghdr msgh;
+
+ /* Reserve headroom for a single VLAN tag */
+ dp_packet_reserve(buffer, VLAN_HEADER_LEN);
+ size = dp_packet_tailroom(buffer);
+
+ iov.iov_base = dp_packet_data(buffer);
+ iov.iov_len = size;
+ msgh.msg_name = NULL;
+ msgh.msg_namelen = 0;
+ msgh.msg_iov = &iov;
+ msgh.msg_iovlen = 1;
+ msgh.msg_control = &cmsg_buffer;
+ msgh.msg_controllen = sizeof cmsg_buffer;
+ msgh.msg_flags = 0;
do {
- retval = (rx->is_tap
- ? read(rx->fd, data, size)
- : recv(rx->fd, data, size, MSG_TRUNC));
+ retval = recvmsg(fd, &msgh, MSG_TRUNC);
} while (retval < 0 && errno == EINTR);
- if (retval >= 0) {
- return retval > size ? -EMSGSIZE : retval;
- } else {
- if (errno != EAGAIN) {
+ if (retval < 0) {
+ return errno;
+ } else if (retval > size) {
+ return EMSGSIZE;
+ }
+
+ dp_packet_set_size(buffer, dp_packet_size(buffer) + retval);
+
+ for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
+ const struct tpacket_auxdata *aux;
+
+ if (cmsg->cmsg_level != SOL_PACKET
+ || cmsg->cmsg_type != PACKET_AUXDATA
+ || cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata))) {
+ continue;
+ }
+
+ aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg));
+ if (auxdata_has_vlan_tci(aux)) {
+ if (retval < ETH_HEADER_LEN) {
+ return EINVAL;
+ }
+
+ eth_push_vlan(buffer, auxdata_to_vlan_tpid(aux),
+ htons(aux->tp_vlan_tci));
+ break;
+ }
+ }
+
+ return 0;
+}
+
+static int
+netdev_linux_rxq_recv_tap(int fd, struct dp_packet *buffer)
+{
+ ssize_t retval;
+ size_t size = dp_packet_tailroom(buffer);
+
+ do {
+ retval = read(fd, dp_packet_data(buffer), size);
+ } while (retval < 0 && errno == EINTR);
+
+ if (retval < 0) {
+ return errno;
+ }
+
+ dp_packet_set_size(buffer, dp_packet_size(buffer) + retval);
+ return 0;
+}
+
+static int
+netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets,
+ int *c)
+{
+ struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
+ struct netdev *netdev = rx->up.netdev;
+ struct dp_packet *buffer;
+ ssize_t retval;
+ int mtu;
+
+ if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) {
+ mtu = ETH_PAYLOAD_MAX;
+ }
+
+ buffer = dp_packet_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu,
+ DP_NETDEV_HEADROOM);
+ retval = (rx->is_tap
+ ? netdev_linux_rxq_recv_tap(rx->fd, buffer)
+ : netdev_linux_rxq_recv_sock(rx->fd, buffer));
+
+ if (retval) {
+ if (retval != EAGAIN && retval != EMSGSIZE) {
VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
- ovs_strerror(errno), netdev_rx_get_name(rx_));
+ netdev_rxq_get_name(rxq_), ovs_strerror(errno));
}
- return -errno;
+ dp_packet_delete(buffer);
+ } else {
+ dp_packet_pad(buffer);
+ dp_packet_rss_invalidate(buffer);
+ packets[0] = buffer;
+ *c = 1;
}
+
+ return retval;
}
static void
-netdev_linux_rx_wait(struct netdev_rx *rx_)
+netdev_linux_rxq_wait(struct netdev_rxq *rxq_)
{
- struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
+ struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
poll_fd_wait(rx->fd, POLLIN);
}
static int
-netdev_linux_rx_drain(struct netdev_rx *rx_)
+netdev_linux_rxq_drain(struct netdev_rxq *rxq_)
{
- struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_);
+ struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_);
if (rx->is_tap) {
struct ifreq ifr;
- int error = af_inet_ifreq_ioctl(netdev_rx_get_name(rx_), &ifr,
+ int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr,
SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
if (error) {
return error;
* The kernel maintains a packet transmission queue, so the caller is not
* expected to do additional queuing of packets. */
static int
-netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
+netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED,
+ struct dp_packet **pkts, int cnt, bool may_steal)
{
- for (;;) {
+ int i;
+ int error = 0;
+
+ /* 'i' is incremented only if there's no error */
+ for (i = 0; i < cnt;) {
+ const void *data = dp_packet_data(pkts[i]);
+ size_t size = dp_packet_size(pkts[i]);
ssize_t retval;
if (!is_tap_netdev(netdev_)) {
/* The Linux AF_PACKET implementation never blocks waiting for room
* for packets, instead returning ENOBUFS. Translate this into
* EAGAIN for the caller. */
- if (errno == ENOBUFS) {
- return EAGAIN;
- } else if (errno == EINTR) {
+ error = errno == ENOBUFS ? EAGAIN : errno;
+ if (error == EINTR) {
+ /* continue without incrementing 'i', i.e. retry this packet */
continue;
- } else if (errno != EAGAIN) {
- VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
- netdev_get_name(netdev_), ovs_strerror(errno));
}
- return errno;
+ break;
} else if (retval != size) {
- VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
- "%zu) on %s", retval, size, netdev_get_name(netdev_));
- return EMSGSIZE;
- } else {
- return 0;
+ VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" bytes"
+ " of %"PRIuSIZE") on %s", retval, size,
+ netdev_get_name(netdev_));
+ error = EMSGSIZE;
+ break;
+ }
+
+ /* Process the next packet in the batch */
+ i++;
+ }
+
+ if (may_steal) {
+ for (i = 0; i < cnt; i++) {
+ dp_packet_delete(pkts[i]);
}
}
+
+ if (error && error != EAGAIN) {
+ VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
+ netdev_get_name(netdev_), ovs_strerror(error));
+ }
+
+ return error;
+
}
/* Registers with the poll loop to wake up from the next call to poll_block()
* expected to do additional queuing of packets. Thus, this function is
* unlikely to ever be used. It is included for completeness. */
static void
-netdev_linux_send_wait(struct netdev *netdev)
+netdev_linux_send_wait(struct netdev *netdev, int qid OVS_UNUSED)
{
if (is_tap_netdev(netdev)) {
/* TAP device always accepts packets.*/
/* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful,
* otherwise a positive errno value. */
static int
-netdev_linux_set_etheraddr(struct netdev *netdev_,
- const uint8_t mac[ETH_ADDR_LEN])
+netdev_linux_set_etheraddr(struct netdev *netdev_, const struct eth_addr mac)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
enum netdev_flags old_flags = 0;
netdev->ether_addr_error = error;
netdev->cache_valid |= VALID_ETHERADDR;
if (!error) {
- memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN);
+ netdev->etheraddr = mac;
}
}
/* Copies 'netdev''s MAC address to 'mac' which is passed as param. */
static int
-netdev_linux_get_etheraddr(const struct netdev *netdev_,
- uint8_t mac[ETH_ADDR_LEN])
+netdev_linux_get_etheraddr(const struct netdev *netdev_, struct eth_addr *mac)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
int error;
ovs_mutex_lock(&netdev->mutex);
if (!(netdev->cache_valid & VALID_ETHERADDR)) {
netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_),
- netdev->etheraddr);
+ &netdev->etheraddr);
netdev->cache_valid |= VALID_ETHERADDR;
}
error = netdev->ether_addr_error;
if (!error) {
- memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN);
+ *mac = netdev->etheraddr;
}
ovs_mutex_unlock(&netdev->mutex);
ovs_mutex_lock(&netdev->mutex);
interval = interval > 0 ? MAX(interval, 100) : 0;
if (netdev->miimon_interval != interval) {
+ if (interval && !netdev->miimon_interval) {
+ atomic_count_inc(&miimon_cnt);
+ } else if (!interval && netdev->miimon_interval) {
+ atomic_count_dec(&miimon_cnt);
+ }
+
netdev->miimon_interval = interval;
timer_set_expired(&netdev->miimon_timer);
}
shash_destroy(&device_shash);
}
-/* Check whether we can we use RTM_GETLINK to get network device statistics.
- * In pre-2.6.19 kernels, this was only available if wireless extensions were
- * enabled. */
-static bool
-check_for_working_netlink_stats(void)
-{
- /* Decide on the netdev_get_stats() implementation to use. Netlink is
- * preferable, so if that works, we'll use it. */
- int ifindex = do_get_ifindex("lo");
- if (ifindex < 0) {
- VLOG_WARN("failed to get ifindex for lo, "
- "obtaining netdev stats from proc");
- return false;
- } else {
- struct netdev_stats stats;
- int error = get_stats_via_netlink(ifindex, &stats);
- if (!error) {
- VLOG_DBG("obtaining netdev stats via rtnetlink");
- return true;
- } else {
- VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
- "via proc (you are probably running a pre-2.6.19 "
- "kernel)", ovs_strerror(error));
- return false;
- }
- }
-}
-
static void
swap_uint64(uint64_t *a, uint64_t *b)
{
netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst,
const struct ovs_vport_stats *src)
{
- dst->rx_packets = get_unaligned_u64(&src->rx_packets);
- dst->tx_packets = get_unaligned_u64(&src->tx_packets);
- dst->rx_bytes = get_unaligned_u64(&src->rx_bytes);
- dst->tx_bytes = get_unaligned_u64(&src->tx_bytes);
- dst->rx_errors = get_unaligned_u64(&src->rx_errors);
- dst->tx_errors = get_unaligned_u64(&src->tx_errors);
- dst->rx_dropped = get_unaligned_u64(&src->rx_dropped);
- dst->tx_dropped = get_unaligned_u64(&src->tx_dropped);
+ dst->rx_packets = get_32aligned_u64(&src->rx_packets);
+ dst->tx_packets = get_32aligned_u64(&src->tx_packets);
+ dst->rx_bytes = get_32aligned_u64(&src->rx_bytes);
+ dst->tx_bytes = get_32aligned_u64(&src->tx_bytes);
+ dst->rx_errors = get_32aligned_u64(&src->rx_errors);
+ dst->tx_errors = get_32aligned_u64(&src->tx_errors);
+ dst->rx_dropped = get_32aligned_u64(&src->rx_dropped);
+ dst->tx_dropped = get_32aligned_u64(&src->tx_dropped);
dst->multicast = 0;
dst->collisions = 0;
dst->rx_length_errors = 0;
static int
get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats)
{
- struct dpif_linux_vport reply;
+ struct dpif_netlink_vport reply;
struct ofpbuf *buf;
int error;
- error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf);
+ error = dpif_netlink_vport_get(netdev_get_name(netdev), &reply, &buf);
if (error) {
return error;
} else if (!reply.stats) {
int error;
error = get_stats_via_vport__(netdev_, stats);
- if (error && error != ENOENT) {
+ if (error && error != ENOENT && error != ENODEV) {
VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed "
"(%s)",
netdev_get_name(netdev_), ovs_strerror(error));
}
}
-static int
-netdev_linux_sys_get_stats(const struct netdev *netdev_,
- struct netdev_stats *stats)
-{
- static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
- static int use_netlink_stats;
- int error;
-
- if (ovsthread_once_start(&once)) {
- use_netlink_stats = check_for_working_netlink_stats();
- ovsthread_once_done(&once);
- }
-
- if (use_netlink_stats) {
- int ifindex;
-
- error = get_ifindex(netdev_, &ifindex);
- if (!error) {
- error = get_stats_via_netlink(ifindex, stats);
- }
- } else {
- error = get_stats_via_proc(netdev_get_name(netdev_), stats);
- }
-
- if (error) {
- VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d",
- netdev_get_name(netdev_), error);
- }
- return error;
-
-}
-
/* Retrieves current device stats for 'netdev-linux'. */
static int
netdev_linux_get_stats(const struct netdev *netdev_,
ovs_mutex_lock(&netdev->mutex);
get_stats_via_vport(netdev_, stats);
- error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
+ error = get_stats_via_netlink(netdev_, &dev_stats);
if (error) {
if (!netdev->vport_stats_error) {
error = 0;
}
} else if (netdev->vport_stats_error) {
- /* stats not available from OVS then use ioctl stats. */
+ /* stats not available from OVS then use netdev stats. */
*stats = dev_stats;
} else {
+ /* Use kernel netdev's packet and byte counts since vport's counters
+ * do not reflect packet counts on the wire when GSO, TSO or GRO are
+ * enabled. */
+ stats->rx_packets = dev_stats.rx_packets;
+ stats->rx_bytes = dev_stats.rx_bytes;
+ stats->tx_packets = dev_stats.tx_packets;
+ stats->tx_bytes = dev_stats.tx_bytes;
+
stats->rx_errors += dev_stats.rx_errors;
stats->tx_errors += dev_stats.tx_errors;
stats->rx_dropped += dev_stats.rx_dropped;
ovs_mutex_lock(&netdev->mutex);
get_stats_via_vport(netdev_, stats);
- error = netdev_linux_sys_get_stats(netdev_, &dev_stats);
+ error = get_stats_via_netlink(netdev_, &dev_stats);
if (error) {
if (!netdev->vport_stats_error) {
error = 0;
stats->tx_heartbeat_errors = 0;
stats->tx_window_errors = 0;
} else {
+ /* Use kernel netdev's packet and byte counts since vport counters
+ * do not reflect packet counts on the wire when GSO, TSO or GRO
+ * are enabled. */
+ stats->rx_packets = dev_stats.tx_packets;
+ stats->rx_bytes = dev_stats.tx_bytes;
+ stats->tx_packets = dev_stats.rx_packets;
+ stats->tx_bytes = dev_stats.rx_bytes;
+
stats->rx_dropped += dev_stats.tx_dropped;
stats->tx_dropped += dev_stats.rx_dropped;
return error;
}
-static int
-netdev_internal_set_stats(struct netdev *netdev,
- const struct netdev_stats *stats)
-{
- struct ovs_vport_stats vport_stats;
- struct dpif_linux_vport vport;
- int err;
-
- vport_stats.rx_packets = stats->rx_packets;
- vport_stats.tx_packets = stats->tx_packets;
- vport_stats.rx_bytes = stats->rx_bytes;
- vport_stats.tx_bytes = stats->tx_bytes;
- vport_stats.rx_errors = stats->rx_errors;
- vport_stats.tx_errors = stats->tx_errors;
- vport_stats.rx_dropped = stats->rx_dropped;
- vport_stats.tx_dropped = stats->tx_dropped;
-
- dpif_linux_vport_init(&vport);
- vport.cmd = OVS_VPORT_CMD_SET;
- vport.name = netdev_get_name(netdev);
- vport.stats = &vport_stats;
-
- err = dpif_linux_vport_transact(&vport, NULL, NULL);
-
- /* If the vport layer doesn't know about the device, that doesn't mean it
- * doesn't exist (after all were able to open it when netdev_open() was
- * called), it just means that it isn't attached and we'll be getting
- * stats a different way. */
- if (err == ENODEV) {
- err = EOPNOTSUPP;
- }
-
- return err;
-}
-
static void
netdev_linux_read_features(struct netdev_linux *netdev)
{
if (ecmd.supported & SUPPORTED_1000baseT_Half) {
netdev->supported |= NETDEV_F_1GB_HD;
}
- if (ecmd.supported & SUPPORTED_1000baseT_Full) {
+ if ((ecmd.supported & SUPPORTED_1000baseT_Full) ||
+ (ecmd.supported & SUPPORTED_1000baseKX_Full)) {
netdev->supported |= NETDEV_F_1GB_FD;
}
- if (ecmd.supported & SUPPORTED_10000baseT_Full) {
+ if ((ecmd.supported & SUPPORTED_10000baseT_Full) ||
+ (ecmd.supported & SUPPORTED_10000baseKX4_Full) ||
+ (ecmd.supported & SUPPORTED_10000baseKR_Full) ||
+ (ecmd.supported & SUPPORTED_10000baseR_FEC)) {
netdev->supported |= NETDEV_F_10GB_FD;
}
+ if ((ecmd.supported & SUPPORTED_40000baseKR4_Full) ||
+ (ecmd.supported & SUPPORTED_40000baseCR4_Full) ||
+ (ecmd.supported & SUPPORTED_40000baseSR4_Full) ||
+ (ecmd.supported & SUPPORTED_40000baseLR4_Full)) {
+ netdev->supported |= NETDEV_F_40GB_FD;
+ }
if (ecmd.supported & SUPPORTED_TP) {
netdev->supported |= NETDEV_F_COPPER;
}
if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
netdev->advertised |= NETDEV_F_1GB_HD;
}
- if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
+ if ((ecmd.advertising & ADVERTISED_1000baseT_Full) ||
+ (ecmd.advertising & ADVERTISED_1000baseKX_Full)) {
netdev->advertised |= NETDEV_F_1GB_FD;
}
- if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
+ if ((ecmd.advertising & ADVERTISED_10000baseT_Full) ||
+ (ecmd.advertising & ADVERTISED_10000baseKX4_Full) ||
+ (ecmd.advertising & ADVERTISED_10000baseKR_Full) ||
+ (ecmd.advertising & ADVERTISED_10000baseR_FEC)) {
netdev->advertised |= NETDEV_F_10GB_FD;
}
+ if ((ecmd.advertising & ADVERTISED_40000baseKR4_Full) ||
+ (ecmd.advertising & ADVERTISED_40000baseCR4_Full) ||
+ (ecmd.advertising & ADVERTISED_40000baseSR4_Full) ||
+ (ecmd.advertising & ADVERTISED_40000baseLR4_Full)) {
+ netdev->advertised |= NETDEV_F_40GB_FD;
+ }
if (ecmd.advertising & ADVERTISED_TP) {
netdev->advertised |= NETDEV_F_COPPER;
}
}
/* Current settings. */
- speed = ecmd.speed;
+ speed = ethtool_cmd_speed(&ecmd);
if (speed == SPEED_10) {
netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD;
} else if (speed == SPEED_100) {
return error;
}
+struct queue_dump_state {
+ struct nl_dump dump;
+ struct ofpbuf buf;
+};
+
static bool
-start_queue_dump(const struct netdev *netdev, struct nl_dump *dump)
+start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state)
{
struct ofpbuf request;
struct tcmsg *tcmsg;
return false;
}
tcmsg->tcm_parent = 0;
- nl_dump_start(dump, NETLINK_ROUTE, &request);
+ nl_dump_start(&state->dump, NETLINK_ROUTE, &request);
ofpbuf_uninit(&request);
+
+ ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE);
return true;
}
+static int
+finish_queue_dump(struct queue_dump_state *state)
+{
+ ofpbuf_uninit(&state->buf);
+ return nl_dump_done(&state->dump);
+}
+
struct netdev_linux_queue_state {
unsigned int *queues;
size_t cur_queue;
ovs_mutex_lock(&netdev->mutex);
error = tc_query_qdisc(netdev_);
if (!error) {
- struct nl_dump dump;
+ struct queue_dump_state state;
if (!netdev->tc->ops->class_dump_stats) {
error = EOPNOTSUPP;
- } else if (!start_queue_dump(netdev_, &dump)) {
+ } else if (!start_queue_dump(netdev_, &state)) {
error = ENODEV;
} else {
struct ofpbuf msg;
int retval;
- while (nl_dump_next(&dump, &msg)) {
+ while (nl_dump_next(&state.dump, &msg, &state.buf)) {
retval = netdev->tc->ops->class_dump_stats(netdev_, &msg,
cb, aux);
if (retval) {
}
}
- retval = nl_dump_done(&dump);
+ retval = finish_queue_dump(&state);
if (retval) {
error = retval;
}
if (!error) {
error = netdev_linux_get_ipv4(netdev_, &netdev->netmask,
SIOCGIFNETMASK, "SIOCGIFNETMASK");
- if (!error) {
- netdev->cache_valid |= VALID_IN4;
- }
}
+ netdev->in4_error = error;
+ netdev->cache_valid |= VALID_IN4;
} else {
- error = 0;
+ error = netdev->in4_error;
}
if (!error) {
ovs_mutex_lock(&netdev->mutex);
error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
if (!error) {
- netdev->cache_valid |= VALID_IN4;
netdev->address = address;
netdev->netmask = netmask;
if (address.s_addr != INADDR_ANY) {
"SIOCSIFNETMASK", netmask);
}
}
+
+ if (!error) {
+ netdev->cache_valid |= VALID_IN4;
+ netdev->in4_error = 0;
+ } else {
+ netdev->cache_valid &= ~VALID_IN4;
+ }
ovs_mutex_unlock(&netdev->mutex);
return error;
{
uint8_t *s6 = in6->s6_addr;
#define X8 "%2"SCNx8
- return sscanf(line,
- " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
- "%*x %*x %*x %*x %16s\n",
- &s6[0], &s6[1], &s6[2], &s6[3],
- &s6[4], &s6[5], &s6[6], &s6[7],
- &s6[8], &s6[9], &s6[10], &s6[11],
- &s6[12], &s6[13], &s6[14], &s6[15],
- ifname) == 17;
-}
-
-/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
- * 'in6' is non-null) and returns true. Otherwise, returns false. */
+ return ovs_scan(line,
+ " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
+ "%*x %*x %*x %*x %16s\n",
+ &s6[0], &s6[1], &s6[2], &s6[3],
+ &s6[4], &s6[5], &s6[6], &s6[7],
+ &s6[8], &s6[9], &s6[10], &s6[11],
+ &s6[12], &s6[13], &s6[14], &s6[15],
+ ifname);
+}
+
+/* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address.
+ * Otherwise, sets '*in6' to 'in6addr_any' and returns the corresponding
+ * error. */
static int
netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ int error;
ovs_mutex_lock(&netdev->mutex);
if (!(netdev->cache_valid & VALID_IN6)) {
char line[128];
netdev->in6 = in6addr_any;
+ netdev->in6_error = EADDRNOTAVAIL;
file = fopen("/proc/net/if_inet6", "r");
if (file != NULL) {
&& !strcmp(name, ifname))
{
netdev->in6 = in6_tmp;
+ netdev->in6_error = 0;
break;
}
}
fclose(file);
+ } else {
+ netdev->in6_error = EOPNOTSUPP;
}
netdev->cache_valid |= VALID_IN6;
}
*in6 = netdev->in6;
+ error = netdev->in6_error;
ovs_mutex_unlock(&netdev->mutex);
- return 0;
+ return error;
}
static void
int refcnt, metric, mtu;
unsigned int flags, use, window, irtt;
- if (sscanf(line,
- "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
- " %d %u %u\n",
- iface, &dest, &gateway, &flags, &refcnt,
- &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
-
+ if (!ovs_scan(line,
+ "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
+ " %d %u %u\n",
+ iface, &dest, &gateway, &flags, &refcnt,
+ &use, &metric, &mask, &mtu, &window, &irtt)) {
VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
fn, ln, line);
continue;
* ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
static int
netdev_linux_arp_lookup(const struct netdev *netdev,
- ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN])
+ ovs_be32 ip, struct eth_addr *mac)
{
struct arpreq r;
struct sockaddr_in sin;
if (nd & NETDEV_PROMISC) {
iff |= IFF_PROMISC;
}
+ if (nd & NETDEV_LOOPBACK) {
+ iff |= IFF_LOOPBACK;
+ }
return iff;
}
if (iff & IFF_PROMISC) {
nd |= NETDEV_PROMISC;
}
+ if (iff & IFF_LOOPBACK) {
+ nd |= NETDEV_LOOPBACK;
+ }
return nd;
}
return error;
}
-static unsigned int
-netdev_linux_change_seq(const struct netdev *netdev_)
-{
- struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- unsigned int change_seq;
-
- ovs_mutex_lock(&netdev->mutex);
- change_seq = netdev->change_seq;
- ovs_mutex_unlock(&netdev->mutex);
-
- return change_seq;
-}
-
-#define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, SET_STATS, \
+#define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, \
GET_FEATURES, GET_STATUS) \
{ \
NAME, \
NULL, /* get_config */ \
NULL, /* set_config */ \
NULL, /* get_tunnel_config */ \
+ NULL, /* build header */ \
+ NULL, /* push header */ \
+ NULL, /* pop header */ \
+ NULL, /* get_numa_id */ \
+ NULL, /* set_multiq */ \
\
netdev_linux_send, \
netdev_linux_send_wait, \
netdev_linux_get_carrier_resets, \
netdev_linux_set_miimon_interval, \
GET_STATS, \
- SET_STATS, \
\
GET_FEATURES, \
netdev_linux_set_advertisements, \
\
netdev_linux_update_flags, \
\
- netdev_linux_change_seq, \
- \
- netdev_linux_rx_alloc, \
- netdev_linux_rx_construct, \
- netdev_linux_rx_destruct, \
- netdev_linux_rx_dealloc, \
- netdev_linux_rx_recv, \
- netdev_linux_rx_wait, \
- netdev_linux_rx_drain, \
+ netdev_linux_rxq_alloc, \
+ netdev_linux_rxq_construct, \
+ netdev_linux_rxq_destruct, \
+ netdev_linux_rxq_dealloc, \
+ netdev_linux_rxq_recv, \
+ netdev_linux_rxq_wait, \
+ netdev_linux_rxq_drain, \
}
const struct netdev_class netdev_linux_class =
"system",
netdev_linux_construct,
netdev_linux_get_stats,
- NULL, /* set_stats */
netdev_linux_get_features,
netdev_linux_get_status);
"tap",
netdev_linux_construct_tap,
netdev_tap_get_stats,
- NULL, /* set_stats */
netdev_linux_get_features,
netdev_linux_get_status);
"internal",
netdev_linux_construct,
netdev_internal_get_stats,
- netdev_internal_set_stats,
NULL, /* get_features */
netdev_internal_get_status);
\f
-/* HTB traffic control class. */
-#define HTB_N_QUEUES 0xf000
+#define CODEL_N_QUEUES 0x0000
-struct htb {
- struct tc tc;
- unsigned int max_rate; /* In bytes/s. */
-};
+/* In sufficiently new kernel headers these are defined as enums in
+ * <linux/pkt_sched.h>. Define them here as macros to help out with older
+ * kernels. (This overrides any enum definition in the header file but that's
+ * harmless.) */
+#define TCA_CODEL_TARGET 1
+#define TCA_CODEL_LIMIT 2
+#define TCA_CODEL_INTERVAL 3
-struct htb_class {
- struct tc_queue tc_queue;
- unsigned int min_rate; /* In bytes/s. */
- unsigned int max_rate; /* In bytes/s. */
- unsigned int burst; /* In bytes. */
- unsigned int priority; /* Lower values are higher priorities. */
+struct codel {
+ struct tc tc;
+ uint32_t target;
+ uint32_t limit;
+ uint32_t interval;
};
-static struct htb *
-htb_get__(const struct netdev *netdev_)
+static struct codel *
+codel_get__(const struct netdev *netdev_)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- return CONTAINER_OF(netdev->tc, struct htb, tc);
+ return CONTAINER_OF(netdev->tc, struct codel, tc);
}
static void
-htb_install__(struct netdev *netdev_, uint64_t max_rate)
+codel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit,
+ uint32_t interval)
{
struct netdev_linux *netdev = netdev_linux_cast(netdev_);
- struct htb *htb;
+ struct codel *codel;
- htb = xmalloc(sizeof *htb);
- tc_init(&htb->tc, &tc_ops_htb);
- htb->max_rate = max_rate;
+ codel = xmalloc(sizeof *codel);
+ tc_init(&codel->tc, &tc_ops_codel);
+ codel->target = target;
+ codel->limit = limit;
+ codel->interval = interval;
- netdev->tc = &htb->tc;
+ netdev->tc = &codel->tc;
}
-/* Create an HTB qdisc.
- *
- * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
static int
-htb_setup_qdisc__(struct netdev *netdev)
+codel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit,
+ uint32_t interval)
{
size_t opt_offset;
- struct tc_htb_glob opt;
struct ofpbuf request;
struct tcmsg *tcmsg;
+ uint32_t otarget, olimit, ointerval;
+ int error;
tc_del_qdisc(netdev);
tcmsg->tcm_handle = tc_make_handle(1, 0);
tcmsg->tcm_parent = TC_H_ROOT;
- nl_msg_put_string(&request, TCA_KIND, "htb");
-
- memset(&opt, 0, sizeof opt);
- opt.rate2quantum = 10;
- opt.version = 3;
- opt.defcls = 1;
+ otarget = target ? target : 5000;
+ olimit = limit ? limit : 10240;
+ ointerval = interval ? interval : 100000;
+ nl_msg_put_string(&request, TCA_KIND, "codel");
opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
- nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
+ nl_msg_put_u32(&request, TCA_CODEL_TARGET, otarget);
+ nl_msg_put_u32(&request, TCA_CODEL_LIMIT, olimit);
+ nl_msg_put_u32(&request, TCA_CODEL_INTERVAL, ointerval);
nl_msg_end_nested(&request, opt_offset);
- return tc_transact(&request, NULL);
+ error = tc_transact(&request, NULL);
+ if (error) {
+ VLOG_WARN_RL(&rl, "failed to replace %s qdisc, "
+ "target %u, limit %u, interval %u error %d(%s)",
+ netdev_get_name(netdev),
+ otarget, olimit, ointerval,
+ error, ovs_strerror(error));
+ }
+ return error;
+}
+
+static void
+codel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED,
+ const struct smap *details, struct codel *codel)
+{
+ const char *target_s;
+ const char *limit_s;
+ const char *interval_s;
+
+ target_s = smap_get(details, "target");
+ limit_s = smap_get(details, "limit");
+ interval_s = smap_get(details, "interval");
+
+ codel->target = target_s ? strtoull(target_s, NULL, 10) : 0;
+ codel->limit = limit_s ? strtoull(limit_s, NULL, 10) : 0;
+ codel->interval = interval_s ? strtoull(interval_s, NULL, 10) : 0;
+
+ if (!codel->target) {
+ codel->target = 5000;
+ }
+ if (!codel->limit) {
+ codel->limit = 10240;
+ }
+ if (!codel->interval) {
+ codel->interval = 100000;
+ }
}
-/* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
- * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
static int
-htb_setup_class__(struct netdev *netdev, unsigned int handle,
- unsigned int parent, struct htb_class *class)
+codel_tc_install(struct netdev *netdev, const struct smap *details)
{
- size_t opt_offset;
- struct tc_htb_opt opt;
- struct ofpbuf request;
- struct tcmsg *tcmsg;
int error;
- int mtu;
+ struct codel codel;
- error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
- if (error) {
- VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
- netdev_get_name(netdev));
- return error;
+ codel_parse_qdisc_details__(netdev, details, &codel);
+ error = codel_setup_qdisc__(netdev, codel.target, codel.limit,
+ codel.interval);
+ if (!error) {
+ codel_install__(netdev, codel.target, codel.limit, codel.interval);
}
-
+ return error;
+}
+
+static int
+codel_parse_tca_options__(struct nlattr *nl_options, struct codel *codel)
+{
+ static const struct nl_policy tca_codel_policy[] = {
+ [TCA_CODEL_TARGET] = { .type = NL_A_U32 },
+ [TCA_CODEL_LIMIT] = { .type = NL_A_U32 },
+ [TCA_CODEL_INTERVAL] = { .type = NL_A_U32 }
+ };
+
+ struct nlattr *attrs[ARRAY_SIZE(tca_codel_policy)];
+
+ if (!nl_parse_nested(nl_options, tca_codel_policy,
+ attrs, ARRAY_SIZE(tca_codel_policy))) {
+ VLOG_WARN_RL(&rl, "failed to parse CoDel class options");
+ return EPROTO;
+ }
+
+ codel->target = nl_attr_get_u32(attrs[TCA_CODEL_TARGET]);
+ codel->limit = nl_attr_get_u32(attrs[TCA_CODEL_LIMIT]);
+ codel->interval = nl_attr_get_u32(attrs[TCA_CODEL_INTERVAL]);
+ return 0;
+}
+
+static int
+codel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg)
+{
+ struct nlattr *nlattr;
+ const char * kind;
+ int error;
+ struct codel codel;
+
+ error = tc_parse_qdisc(nlmsg, &kind, &nlattr);
+ if (error != 0) {
+ return error;
+ }
+
+ error = codel_parse_tca_options__(nlattr, &codel);
+ if (error != 0) {
+ return error;
+ }
+
+ codel_install__(netdev, codel.target, codel.limit, codel.interval);
+ return 0;
+}
+
+
+static void
+codel_tc_destroy(struct tc *tc)
+{
+ struct codel *codel = CONTAINER_OF(tc, struct codel, tc);
+ tc_destroy(tc);
+ free(codel);
+}
+
+static int
+codel_qdisc_get(const struct netdev *netdev, struct smap *details)
+{
+ const struct codel *codel = codel_get__(netdev);
+ smap_add_format(details, "target", "%u", codel->target);
+ smap_add_format(details, "limit", "%u", codel->limit);
+ smap_add_format(details, "interval", "%u", codel->interval);
+ return 0;
+}
+
+static int
+codel_qdisc_set(struct netdev *netdev, const struct smap *details)
+{
+ struct codel codel;
+
+ codel_parse_qdisc_details__(netdev, details, &codel);
+ codel_install__(netdev, codel.target, codel.limit, codel.interval);
+ codel_get__(netdev)->target = codel.target;
+ codel_get__(netdev)->limit = codel.limit;
+ codel_get__(netdev)->interval = codel.interval;
+ return 0;
+}
+
+static const struct tc_ops tc_ops_codel = {
+ "codel", /* linux_name */
+ "linux-codel", /* ovs_name */
+ CODEL_N_QUEUES, /* n_queues */
+ codel_tc_install,
+ codel_tc_load,
+ codel_tc_destroy,
+ codel_qdisc_get,
+ codel_qdisc_set,
+ NULL,
+ NULL,
+ NULL,
+ NULL,
+ NULL
+};
+\f
+/* FQ-CoDel traffic control class. */
+
+#define FQCODEL_N_QUEUES 0x0000
+
+/* In sufficiently new kernel headers these are defined as enums in
+ * <linux/pkt_sched.h>. Define them here as macros to help out with older
+ * kernels. (This overrides any enum definition in the header file but that's
+ * harmless.) */
+#define TCA_FQ_CODEL_TARGET 1
+#define TCA_FQ_CODEL_LIMIT 2
+#define TCA_FQ_CODEL_INTERVAL 3
+#define TCA_FQ_CODEL_ECN 4
+#define TCA_FQ_CODEL_FLOWS 5
+#define TCA_FQ_CODEL_QUANTUM 6
+
+struct fqcodel {
+ struct tc tc;
+ uint32_t target;
+ uint32_t limit;
+ uint32_t interval;
+ uint32_t flows;
+ uint32_t quantum;
+};
+
+static struct fqcodel *
+fqcodel_get__(const struct netdev *netdev_)
+{
+ struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ return CONTAINER_OF(netdev->tc, struct fqcodel, tc);
+}
+
+static void
+fqcodel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit,
+ uint32_t interval, uint32_t flows, uint32_t quantum)
+{
+ struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ struct fqcodel *fqcodel;
+
+ fqcodel = xmalloc(sizeof *fqcodel);
+ tc_init(&fqcodel->tc, &tc_ops_fqcodel);
+ fqcodel->target = target;
+ fqcodel->limit = limit;
+ fqcodel->interval = interval;
+ fqcodel->flows = flows;
+ fqcodel->quantum = quantum;
+
+ netdev->tc = &fqcodel->tc;
+}
+
+static int
+fqcodel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit,
+ uint32_t interval, uint32_t flows, uint32_t quantum)
+{
+ size_t opt_offset;
+ struct ofpbuf request;
+ struct tcmsg *tcmsg;
+ uint32_t otarget, olimit, ointerval, oflows, oquantum;
+ int error;
+
+ tc_del_qdisc(netdev);
+
+ tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
+ NLM_F_EXCL | NLM_F_CREATE, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
+ tcmsg->tcm_handle = tc_make_handle(1, 0);
+ tcmsg->tcm_parent = TC_H_ROOT;
+
+ otarget = target ? target : 5000;
+ olimit = limit ? limit : 10240;
+ ointerval = interval ? interval : 100000;
+ oflows = flows ? flows : 1024;
+ oquantum = quantum ? quantum : 1514; /* fq_codel default quantum is 1514
+ not mtu */
+
+ nl_msg_put_string(&request, TCA_KIND, "fq_codel");
+ opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
+ nl_msg_put_u32(&request, TCA_FQ_CODEL_TARGET, otarget);
+ nl_msg_put_u32(&request, TCA_FQ_CODEL_LIMIT, olimit);
+ nl_msg_put_u32(&request, TCA_FQ_CODEL_INTERVAL, ointerval);
+ nl_msg_put_u32(&request, TCA_FQ_CODEL_FLOWS, oflows);
+ nl_msg_put_u32(&request, TCA_FQ_CODEL_QUANTUM, oquantum);
+ nl_msg_end_nested(&request, opt_offset);
+
+ error = tc_transact(&request, NULL);
+ if (error) {
+ VLOG_WARN_RL(&rl, "failed to replace %s qdisc, "
+ "target %u, limit %u, interval %u, flows %u, quantum %u error %d(%s)",
+ netdev_get_name(netdev),
+ otarget, olimit, ointerval, oflows, oquantum,
+ error, ovs_strerror(error));
+ }
+ return error;
+}
+
+static void
+fqcodel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED,
+ const struct smap *details, struct fqcodel *fqcodel)
+{
+ const char *target_s;
+ const char *limit_s;
+ const char *interval_s;
+ const char *flows_s;
+ const char *quantum_s;
+
+ target_s = smap_get(details, "target");
+ limit_s = smap_get(details, "limit");
+ interval_s = smap_get(details, "interval");
+ flows_s = smap_get(details, "flows");
+ quantum_s = smap_get(details, "quantum");
+ fqcodel->target = target_s ? strtoull(target_s, NULL, 10) : 0;
+ fqcodel->limit = limit_s ? strtoull(limit_s, NULL, 10) : 0;
+ fqcodel->interval = interval_s ? strtoull(interval_s, NULL, 10) : 0;
+ fqcodel->flows = flows_s ? strtoull(flows_s, NULL, 10) : 0;
+ fqcodel->quantum = quantum_s ? strtoull(quantum_s, NULL, 10) : 0;
+ if (!fqcodel->target) {
+ fqcodel->target = 5000;
+ }
+ if (!fqcodel->limit) {
+ fqcodel->limit = 10240;
+ }
+ if (!fqcodel->interval) {
+ fqcodel->interval = 1000000;
+ }
+ if (!fqcodel->flows) {
+ fqcodel->flows = 1024;
+ }
+ if (!fqcodel->quantum) {
+ fqcodel->quantum = 1514;
+ }
+}
+
+static int
+fqcodel_tc_install(struct netdev *netdev, const struct smap *details)
+{
+ int error;
+ struct fqcodel fqcodel;
+
+ fqcodel_parse_qdisc_details__(netdev, details, &fqcodel);
+ error = fqcodel_setup_qdisc__(netdev, fqcodel.target, fqcodel.limit,
+ fqcodel.interval, fqcodel.flows,
+ fqcodel.quantum);
+ if (!error) {
+ fqcodel_install__(netdev, fqcodel.target, fqcodel.limit,
+ fqcodel.interval, fqcodel.flows, fqcodel.quantum);
+ }
+ return error;
+}
+
+static int
+fqcodel_parse_tca_options__(struct nlattr *nl_options, struct fqcodel *fqcodel)
+{
+ static const struct nl_policy tca_fqcodel_policy[] = {
+ [TCA_FQ_CODEL_TARGET] = { .type = NL_A_U32 },
+ [TCA_FQ_CODEL_LIMIT] = { .type = NL_A_U32 },
+ [TCA_FQ_CODEL_INTERVAL] = { .type = NL_A_U32 },
+ [TCA_FQ_CODEL_FLOWS] = { .type = NL_A_U32 },
+ [TCA_FQ_CODEL_QUANTUM] = { .type = NL_A_U32 }
+ };
+
+ struct nlattr *attrs[ARRAY_SIZE(tca_fqcodel_policy)];
+
+ if (!nl_parse_nested(nl_options, tca_fqcodel_policy,
+ attrs, ARRAY_SIZE(tca_fqcodel_policy))) {
+ VLOG_WARN_RL(&rl, "failed to parse FQ_CoDel class options");
+ return EPROTO;
+ }
+
+ fqcodel->target = nl_attr_get_u32(attrs[TCA_FQ_CODEL_TARGET]);
+ fqcodel->limit = nl_attr_get_u32(attrs[TCA_FQ_CODEL_LIMIT]);
+ fqcodel->interval =nl_attr_get_u32(attrs[TCA_FQ_CODEL_INTERVAL]);
+ fqcodel->flows = nl_attr_get_u32(attrs[TCA_FQ_CODEL_FLOWS]);
+ fqcodel->quantum = nl_attr_get_u32(attrs[TCA_FQ_CODEL_QUANTUM]);
+ return 0;
+}
+
+static int
+fqcodel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg)
+{
+ struct nlattr *nlattr;
+ const char * kind;
+ int error;
+ struct fqcodel fqcodel;
+
+ error = tc_parse_qdisc(nlmsg, &kind, &nlattr);
+ if (error != 0) {
+ return error;
+ }
+
+ error = fqcodel_parse_tca_options__(nlattr, &fqcodel);
+ if (error != 0) {
+ return error;
+ }
+
+ fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval,
+ fqcodel.flows, fqcodel.quantum);
+ return 0;
+}
+
+static void
+fqcodel_tc_destroy(struct tc *tc)
+{
+ struct fqcodel *fqcodel = CONTAINER_OF(tc, struct fqcodel, tc);
+ tc_destroy(tc);
+ free(fqcodel);
+}
+
+static int
+fqcodel_qdisc_get(const struct netdev *netdev, struct smap *details)
+{
+ const struct fqcodel *fqcodel = fqcodel_get__(netdev);
+ smap_add_format(details, "target", "%u", fqcodel->target);
+ smap_add_format(details, "limit", "%u", fqcodel->limit);
+ smap_add_format(details, "interval", "%u", fqcodel->interval);
+ smap_add_format(details, "flows", "%u", fqcodel->flows);
+ smap_add_format(details, "quantum", "%u", fqcodel->quantum);
+ return 0;
+}
+
+static int
+fqcodel_qdisc_set(struct netdev *netdev, const struct smap *details)
+{
+ struct fqcodel fqcodel;
+
+ fqcodel_parse_qdisc_details__(netdev, details, &fqcodel);
+ fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval,
+ fqcodel.flows, fqcodel.quantum);
+ fqcodel_get__(netdev)->target = fqcodel.target;
+ fqcodel_get__(netdev)->limit = fqcodel.limit;
+ fqcodel_get__(netdev)->interval = fqcodel.interval;
+ fqcodel_get__(netdev)->flows = fqcodel.flows;
+ fqcodel_get__(netdev)->quantum = fqcodel.quantum;
+ return 0;
+}
+
+static const struct tc_ops tc_ops_fqcodel = {
+ "fq_codel", /* linux_name */
+ "linux-fq_codel", /* ovs_name */
+ FQCODEL_N_QUEUES, /* n_queues */
+ fqcodel_tc_install,
+ fqcodel_tc_load,
+ fqcodel_tc_destroy,
+ fqcodel_qdisc_get,
+ fqcodel_qdisc_set,
+ NULL,
+ NULL,
+ NULL,
+ NULL,
+ NULL
+};
+\f
+/* SFQ traffic control class. */
+
+#define SFQ_N_QUEUES 0x0000
+
+struct sfq {
+ struct tc tc;
+ uint32_t quantum;
+ uint32_t perturb;
+};
+
+static struct sfq *
+sfq_get__(const struct netdev *netdev_)
+{
+ struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ return CONTAINER_OF(netdev->tc, struct sfq, tc);
+}
+
+static void
+sfq_install__(struct netdev *netdev_, uint32_t quantum, uint32_t perturb)
+{
+ struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ struct sfq *sfq;
+
+ sfq = xmalloc(sizeof *sfq);
+ tc_init(&sfq->tc, &tc_ops_sfq);
+ sfq->perturb = perturb;
+ sfq->quantum = quantum;
+
+ netdev->tc = &sfq->tc;
+}
+
+static int
+sfq_setup_qdisc__(struct netdev *netdev, uint32_t quantum, uint32_t perturb)
+{
+ struct tc_sfq_qopt opt;
+ struct ofpbuf request;
+ struct tcmsg *tcmsg;
+ int mtu;
+ int mtu_error, error;
+ mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
+
+ tc_del_qdisc(netdev);
+
+ tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
+ NLM_F_EXCL | NLM_F_CREATE, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
+ tcmsg->tcm_handle = tc_make_handle(1, 0);
+ tcmsg->tcm_parent = TC_H_ROOT;
+
+ memset(&opt, 0, sizeof opt);
+ if (!quantum) {
+ if (!mtu_error) {
+ opt.quantum = mtu; /* if we cannot find mtu, use default */
+ }
+ } else {
+ opt.quantum = quantum;
+ }
+
+ if (!perturb) {
+ opt.perturb_period = 10;
+ } else {
+ opt.perturb_period = perturb;
+ }
+
+ nl_msg_put_string(&request, TCA_KIND, "sfq");
+ nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt);
+
+ error = tc_transact(&request, NULL);
+ if (error) {
+ VLOG_WARN_RL(&rl, "failed to replace %s qdisc, "
+ "quantum %u, perturb %u error %d(%s)",
+ netdev_get_name(netdev),
+ opt.quantum, opt.perturb_period,
+ error, ovs_strerror(error));
+ }
+ return error;
+}
+
+static void
+sfq_parse_qdisc_details__(struct netdev *netdev,
+ const struct smap *details, struct sfq *sfq)
+{
+ const char *perturb_s;
+ const char *quantum_s;
+ int mtu;
+ int mtu_error;
+
+ perturb_s = smap_get(details, "perturb");
+ quantum_s = smap_get(details, "quantum");
+ sfq->perturb = perturb_s ? strtoull(perturb_s, NULL, 10) : 0;
+ sfq->quantum = quantum_s ? strtoull(quantum_s, NULL, 10) : 0;
+ if (!sfq->perturb) {
+ sfq->perturb = 10;
+ }
+
+ if (!sfq->quantum) {
+ mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
+ if (!mtu_error) {
+ sfq->quantum = mtu;
+ } else {
+ VLOG_WARN_RL(&rl, "when using SFQ, you must specify quantum on a "
+ "device without mtu");
+ return;
+ }
+ }
+}
+
+static int
+sfq_tc_install(struct netdev *netdev, const struct smap *details)
+{
+ int error;
+ struct sfq sfq;
+
+ sfq_parse_qdisc_details__(netdev, details, &sfq);
+ error = sfq_setup_qdisc__(netdev, sfq.quantum, sfq.perturb);
+ if (!error) {
+ sfq_install__(netdev, sfq.quantum, sfq.perturb);
+ }
+ return error;
+}
+
+static int
+sfq_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg)
+{
+ const struct tc_sfq_qopt *sfq;
+ struct nlattr *nlattr;
+ const char * kind;
+ int error;
+
+ error = tc_parse_qdisc(nlmsg, &kind, &nlattr);
+ if (error == 0) {
+ sfq = nl_attr_get(nlattr);
+ sfq_install__(netdev, sfq->perturb_period, sfq->quantum);
+ return 0;
+ }
+
+ return error;
+}
+
+static void
+sfq_tc_destroy(struct tc *tc)
+{
+ struct sfq *sfq = CONTAINER_OF(tc, struct sfq, tc);
+ tc_destroy(tc);
+ free(sfq);
+}
+
+static int
+sfq_qdisc_get(const struct netdev *netdev, struct smap *details)
+{
+ const struct sfq *sfq = sfq_get__(netdev);
+ smap_add_format(details, "quantum", "%u", sfq->quantum);
+ smap_add_format(details, "perturb", "%u", sfq->perturb);
+ return 0;
+}
+
+static int
+sfq_qdisc_set(struct netdev *netdev, const struct smap *details)
+{
+ struct sfq sfq;
+
+ sfq_parse_qdisc_details__(netdev, details, &sfq);
+ sfq_install__(netdev, sfq.quantum, sfq.perturb);
+ sfq_get__(netdev)->quantum = sfq.quantum;
+ sfq_get__(netdev)->perturb = sfq.perturb;
+ return 0;
+}
+
+static const struct tc_ops tc_ops_sfq = {
+ "sfq", /* linux_name */
+ "linux-sfq", /* ovs_name */
+ SFQ_N_QUEUES, /* n_queues */
+ sfq_tc_install,
+ sfq_tc_load,
+ sfq_tc_destroy,
+ sfq_qdisc_get,
+ sfq_qdisc_set,
+ NULL,
+ NULL,
+ NULL,
+ NULL,
+ NULL
+};
+\f
+/* HTB traffic control class. */
+
+#define HTB_N_QUEUES 0xf000
+#define HTB_RATE2QUANTUM 10
+
+struct htb {
+ struct tc tc;
+ unsigned int max_rate; /* In bytes/s. */
+};
+
+struct htb_class {
+ struct tc_queue tc_queue;
+ unsigned int min_rate; /* In bytes/s. */
+ unsigned int max_rate; /* In bytes/s. */
+ unsigned int burst; /* In bytes. */
+ unsigned int priority; /* Lower values are higher priorities. */
+};
+
+static struct htb *
+htb_get__(const struct netdev *netdev_)
+{
+ struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ return CONTAINER_OF(netdev->tc, struct htb, tc);
+}
+
+static void
+htb_install__(struct netdev *netdev_, uint64_t max_rate)
+{
+ struct netdev_linux *netdev = netdev_linux_cast(netdev_);
+ struct htb *htb;
+
+ htb = xmalloc(sizeof *htb);
+ tc_init(&htb->tc, &tc_ops_htb);
+ htb->max_rate = max_rate;
+
+ netdev->tc = &htb->tc;
+}
+
+/* Create an HTB qdisc.
+ *
+ * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */
+static int
+htb_setup_qdisc__(struct netdev *netdev)
+{
+ size_t opt_offset;
+ struct tc_htb_glob opt;
+ struct ofpbuf request;
+ struct tcmsg *tcmsg;
+
+ tc_del_qdisc(netdev);
+
+ tcmsg = tc_make_request(netdev, RTM_NEWQDISC,
+ NLM_F_EXCL | NLM_F_CREATE, &request);
+ if (!tcmsg) {
+ return ENODEV;
+ }
+ tcmsg->tcm_handle = tc_make_handle(1, 0);
+ tcmsg->tcm_parent = TC_H_ROOT;
+
+ nl_msg_put_string(&request, TCA_KIND, "htb");
+
+ memset(&opt, 0, sizeof opt);
+ opt.rate2quantum = HTB_RATE2QUANTUM;
+ opt.version = 3;
+ opt.defcls = 1;
+
+ opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS);
+ nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt);
+ nl_msg_end_nested(&request, opt_offset);
+
+ return tc_transact(&request, NULL);
+}
+
+/* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb
+ * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */
+static int
+htb_setup_class__(struct netdev *netdev, unsigned int handle,
+ unsigned int parent, struct htb_class *class)
+{
+ size_t opt_offset;
+ struct tc_htb_opt opt;
+ struct ofpbuf request;
+ struct tcmsg *tcmsg;
+ int error;
+ int mtu;
+
+ error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu);
+ if (error) {
+ VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU",
+ netdev_get_name(netdev));
+ return error;
+ }
+
memset(&opt, 0, sizeof opt);
tc_fill_rate(&opt.rate, class->min_rate, mtu);
tc_fill_rate(&opt.ceil, class->max_rate, mtu);
+ /* Makes sure the quantum is at least MTU. Setting quantum will
+ * make htb ignore the r2q for this class. */
+ if ((class->min_rate / HTB_RATE2QUANTUM) < mtu) {
+ opt.quantum = mtu;
+ }
opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst);
opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst);
opt.prio = class->priority;
htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
{
struct ofpbuf msg;
- struct nl_dump dump;
+ struct queue_dump_state state;
struct htb_class hc;
/* Get qdisc options. */
htb_install__(netdev, hc.max_rate);
/* Get queues. */
- if (!start_queue_dump(netdev, &dump)) {
+ if (!start_queue_dump(netdev, &state)) {
return ENODEV;
}
- while (nl_dump_next(&dump, &msg)) {
+ while (nl_dump_next(&state.dump, &msg, &state.buf)) {
unsigned int queue_id;
if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
htb_update_queue__(netdev, queue_id, &hc);
}
}
- nl_dump_done(&dump);
+ finish_queue_dump(&state);
return 0;
}
hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED)
{
struct ofpbuf msg;
- struct nl_dump dump;
+ struct queue_dump_state state;
struct hfsc_class hc;
hc.max_rate = 0;
hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL);
hfsc_install__(netdev, hc.max_rate);
- if (!start_queue_dump(netdev, &dump)) {
+ if (!start_queue_dump(netdev, &state)) {
return ENODEV;
}
- while (nl_dump_next(&dump, &msg)) {
+ while (nl_dump_next(&state.dump, &msg, &state.buf)) {
unsigned int queue_id;
if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) {
}
}
- nl_dump_done(&dump);
+ finish_queue_dump(&state);
return 0;
}
* mtu 65535 drop
*
* The configuration and stats may be seen with the following command:
- * /sbin/tc -s filter show <devname> eth0 parent ffff:
+ * /sbin/tc -s filter show dev <devname> parent ffff:
*
* Returns 0 if successful, otherwise a positive errno value.
*/
static int
-tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst)
+tc_add_policer(struct netdev *netdev,
+ uint32_t kbits_rate, uint32_t kbits_burst)
{
struct tc_police tc_police;
struct ofpbuf request;
memset(&tc_police, 0, sizeof tc_police);
tc_police.action = TC_POLICE_SHOT;
tc_police.mtu = mtu;
- tc_fill_rate(&tc_police.rate, (kbits_rate * 1000)/8, mtu);
- tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate,
- kbits_burst * 1024);
+ tc_fill_rate(&tc_police.rate, ((uint64_t) kbits_rate * 1000)/8, mtu);
+
+ /* The following appears wrong in two ways:
+ *
+ * - tc_bytes_to_ticks() should take "bytes" as quantity for both of its
+ * arguments (or at least consistently "bytes" as both or "bits" as
+ * both), but this supplies bytes for the first argument and bits for the
+ * second.
+ *
+ * - In networking a kilobit is usually 1000 bits but this uses 1024 bits.
+ *
+ * However if you "fix" those problems then "tc filter show ..." shows
+ * "125000b", meaning 125,000 bits, when OVS configures it for 1000 kbit ==
+ * 1,000,000 bits, whereas this actually ends up doing the right thing from
+ * tc's point of view. Whatever. */
+ tc_police.burst = tc_bytes_to_ticks(
+ tc_police.rate.rate, MIN(UINT32_MAX / 1024, kbits_burst) * 1024);
tcmsg = tc_make_request(netdev, RTM_NEWTFILTER,
NLM_F_EXCL | NLM_F_CREATE, &request);
return error;
}
+static bool
+getqdisc_is_safe(void)
+{
+ static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
+ static bool safe = false;
+
+ if (ovsthread_once_start(&once)) {
+ struct utsname utsname;
+ int major, minor;
+
+ if (uname(&utsname) == -1) {
+ VLOG_WARN("uname failed (%s)", ovs_strerror(errno));
+ } else if (!ovs_scan(utsname.release, "%d.%d", &major, &minor)) {
+ VLOG_WARN("uname reported bad OS release (%s)", utsname.release);
+ } else if (major < 2 || (major == 2 && minor < 35)) {
+ VLOG_INFO("disabling unsafe RTM_GETQDISC in Linux kernel %s",
+ utsname.release);
+ } else {
+ safe = true;
+ }
+ ovsthread_once_done(&once);
+ }
+ return safe;
+}
+
/* If 'netdev''s qdisc type and parameters are not yet known, queries the
* kernel to determine what they are. Returns 0 if successful, otherwise a
* positive errno value. */
* create will have a class with handle 1:0. The built-in qdiscs only have
* a class with handle 0:0.
*
- * We could check for Linux 2.6.35+ and use a more straightforward method
- * there. */
+ * On Linux 2.6.35+ we use the straightforward method because it allows us
+ * to handle non-builtin qdiscs without handle 1:0 (e.g. codel). However,
+ * in such a case we get no response at all from the kernel (!) if a
+ * builtin qdisc is in use (which is later caught by "!error &&
+ * !qdisc->size"). */
tcmsg = tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, &request);
if (!tcmsg) {
return ENODEV;
}
- tcmsg->tcm_handle = tc_make_handle(1, 0);
- tcmsg->tcm_parent = 0;
+ tcmsg->tcm_handle = tc_make_handle(getqdisc_is_safe() ? 0 : 1, 0);
+ tcmsg->tcm_parent = getqdisc_is_safe() ? TC_H_ROOT : 0;
/* Figure out what tc class to instantiate. */
error = tc_transact(&request, &qdisc);
- if (!error) {
+ if (!error && qdisc->size) {
const char *kind;
error = tc_parse_qdisc(qdisc, &kind, NULL);
ops = tc_lookup_linux_name(kind);
if (!ops) {
static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1);
- VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind);
+ VLOG_DBG_RL(&rl2, "unknown qdisc \"%s\"", kind);
ops = &tc_ops_other;
}
}
- } else if (error == ENOENT) {
- /* Either it's a built-in qdisc, or it's a qdisc set up by some
- * other entity that doesn't have a handle 1:0. We will assume
- * that it's the system default qdisc. */
+ } else if ((!error && !qdisc->size) || error == ENOENT) {
+ /* Either it's a built-in qdisc, or (on Linux pre-2.6.35) it's a qdisc
+ * set up by some other entity that doesn't have a handle 1:0. We will
+ * assume that it's the system default qdisc. */
ops = &tc_ops_default;
error = 0;
} else {
}
COVERAGE_INC(netdev_set_ethtool);
- evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
+ new_flags = (evalue.data & ~flag) | (enable ? flag : 0);
+ if (new_flags == evalue.data) {
+ return 0;
+ }
+ evalue.data = new_flags;
error = netdev_linux_do_ethtool(netdev_name,
(struct ethtool_cmd *)&evalue,
ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS");
dst->tx_window_errors = src->tx_window_errors;
}
-static int
-get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
+/* Copies 'src' into 'dst', performing format conversion in the process. */
+static void
+netdev_stats_from_rtnl_link_stats64(struct netdev_stats *dst,
+ const struct rtnl_link_stats64 *src)
{
- /* Policy for RTNLGRP_LINK messages.
- *
- * There are *many* more fields in these messages, but currently we only
- * care about these fields. */
- static const struct nl_policy rtnlgrp_link_policy[] = {
- [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
- [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
- .min_len = sizeof(struct rtnl_link_stats) },
- };
+ dst->rx_packets = src->rx_packets;
+ dst->tx_packets = src->tx_packets;
+ dst->rx_bytes = src->rx_bytes;
+ dst->tx_bytes = src->tx_bytes;
+ dst->rx_errors = src->rx_errors;
+ dst->tx_errors = src->tx_errors;
+ dst->rx_dropped = src->rx_dropped;
+ dst->tx_dropped = src->tx_dropped;
+ dst->multicast = src->multicast;
+ dst->collisions = src->collisions;
+ dst->rx_length_errors = src->rx_length_errors;
+ dst->rx_over_errors = src->rx_over_errors;
+ dst->rx_crc_errors = src->rx_crc_errors;
+ dst->rx_frame_errors = src->rx_frame_errors;
+ dst->rx_fifo_errors = src->rx_fifo_errors;
+ dst->rx_missed_errors = src->rx_missed_errors;
+ dst->tx_aborted_errors = src->tx_aborted_errors;
+ dst->tx_carrier_errors = src->tx_carrier_errors;
+ dst->tx_fifo_errors = src->tx_fifo_errors;
+ dst->tx_heartbeat_errors = src->tx_heartbeat_errors;
+ dst->tx_window_errors = src->tx_window_errors;
+}
+static int
+get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats)
+{
struct ofpbuf request;
struct ofpbuf *reply;
- struct ifinfomsg *ifi;
- struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
int error;
ofpbuf_init(&request, 0);
- nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST);
- ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
- ifi->ifi_family = PF_UNSPEC;
- ifi->ifi_index = ifindex;
+ nl_msg_put_nlmsghdr(&request,
+ sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ),
+ RTM_GETLINK, NLM_F_REQUEST);
+ ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg));
+ nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_));
error = nl_transact(NETLINK_ROUTE, &request, &reply);
ofpbuf_uninit(&request);
if (error) {
return error;
}
- if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
- rtnlgrp_link_policy,
- attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
- ofpbuf_delete(reply);
- return EPROTO;
- }
-
- if (!attrs[IFLA_STATS]) {
- VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
- ofpbuf_delete(reply);
- return EPROTO;
+ if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) {
+ const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS64);
+ if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats64)) {
+ netdev_stats_from_rtnl_link_stats64(stats, nl_attr_get(a));
+ error = 0;
+ } else {
+ const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS);
+ if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) {
+ netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a));
+ error = 0;
+ } else {
+ VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
+ error = EPROTO;
+ }
+ }
+ } else {
+ VLOG_WARN_RL(&rl, "short RTM_GETLINK reply");
+ error = EPROTO;
}
- netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS]));
ofpbuf_delete(reply);
-
- return 0;
-}
-
-static int
-get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
-{
- static const char fn[] = "/proc/net/dev";
- char line[1024];
- FILE *stream;
- int ln;
-
- stream = fopen(fn, "r");
- if (!stream) {
- VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno));
- return errno;
- }
-
- ln = 0;
- while (fgets(line, sizeof line, stream)) {
- if (++ln >= 3) {
- char devname[16];
-#define X64 "%"SCNu64
- if (sscanf(line,
- " %15[^:]:"
- X64 X64 X64 X64 X64 X64 X64 "%*u"
- X64 X64 X64 X64 X64 X64 X64 "%*u",
- devname,
- &stats->rx_bytes,
- &stats->rx_packets,
- &stats->rx_errors,
- &stats->rx_dropped,
- &stats->rx_fifo_errors,
- &stats->rx_frame_errors,
- &stats->multicast,
- &stats->tx_bytes,
- &stats->tx_packets,
- &stats->tx_errors,
- &stats->tx_dropped,
- &stats->tx_fifo_errors,
- &stats->collisions,
- &stats->tx_carrier_errors) != 15) {
- VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
- } else if (!strcmp(devname, netdev_name)) {
- stats->rx_length_errors = UINT64_MAX;
- stats->rx_over_errors = UINT64_MAX;
- stats->rx_crc_errors = UINT64_MAX;
- stats->rx_missed_errors = UINT64_MAX;
- stats->tx_aborted_errors = UINT64_MAX;
- stats->tx_heartbeat_errors = UINT64_MAX;
- stats->tx_window_errors = UINT64_MAX;
- fclose(stream);
- return 0;
- }
- }
- }
- VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
- fclose(stream);
- return ENODEV;
+ return error;
}
static int
}
static int
-get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
+get_etheraddr(const char *netdev_name, struct eth_addr *ea)
{
struct ifreq ifr;
int hwaddr_family;
}
hwaddr_family = ifr.ifr_hwaddr.sa_family;
if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
- VLOG_WARN("%s device has unknown hardware address family %d",
+ VLOG_INFO("%s device has unknown hardware address family %d",
netdev_name, hwaddr_family);
+ return EINVAL;
}
memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
return 0;
}
static int
-set_etheraddr(const char *netdev_name,
- const uint8_t mac[ETH_ADDR_LEN])
+set_etheraddr(const char *netdev_name, const struct eth_addr mac)
{
struct ifreq ifr;
int error;
memset(&ifr, 0, sizeof ifr);
ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
- memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
+ memcpy(ifr.ifr_hwaddr.sa_data, &mac, ETH_ADDR_LEN);
COVERAGE_INC(netdev_set_hwaddr);
error = af_inet_ioctl(SIOCSIFHWADDR, &ifr);
if (error) {
projects / cascardo / ovs.git / blobdiff