X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=ee3c5f52a5cc931e59caf8444586728768606b4c;hb=f613a0d72c521ca3a4eeb2c29ac523f6fdf72667;hp=06f739cdc0f9ca39a0229776e7ad93ddb309aa58;hpb=da3827b5518ed6cd2256ba87eddd1c40bbca75f0;p=sliver-openvswitch.git diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 06f739cdc..ee3c5f52a 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010 Nicira Networks. + * Copyright (c) 2009, 2010, 2011 Nicira Networks. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -15,6 +15,9 @@ */ #include + +#include "netdev-linux.h" + #include #include #include @@ -25,6 +28,7 @@ #include #include #include +#include #include #include #include @@ -46,6 +50,7 @@ #include #include "coverage.h" +#include "dpif-linux.h" #include "dynamic-string.h" #include "fatal-signal.h" #include "hash.h" @@ -53,17 +58,28 @@ #include "netdev-provider.h" #include "netdev-vport.h" #include "netlink.h" +#include "netlink-notifier.h" +#include "netlink-socket.h" #include "ofpbuf.h" #include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" -#include "rtnetlink.h" +#include "rtnetlink-link.h" #include "socket-util.h" #include "shash.h" -#include "svec.h" +#include "sset.h" +#include "timer.h" #include "vlog.h" -VLOG_DEFINE_THIS_MODULE(netdev_linux) +VLOG_DEFINE_THIS_MODULE(netdev_linux); + +COVERAGE_DEFINE(netdev_get_vlan_vid); +COVERAGE_DEFINE(netdev_set_policing); +COVERAGE_DEFINE(netdev_arp_lookup); +COVERAGE_DEFINE(netdev_get_ifindex); +COVERAGE_DEFINE(netdev_get_hwaddr); +COVERAGE_DEFINE(netdev_set_hwaddr); +COVERAGE_DEFINE(netdev_ethtool); /* These were introduced in Linux 2.6.14, so they might be missing if we have * old headers. */ @@ -74,13 +90,22 @@ VLOG_DEFINE_THIS_MODULE(netdev_linux) #define ADVERTISED_Asym_Pause (1 << 14) #endif +/* These were introduced in Linux 2.6.24, so they might be missing if we + * have old headers. */ +#ifndef ETHTOOL_GFLAGS +#define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */ +#endif +#ifndef ETHTOOL_SFLAGS +#define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */ +#endif + /* This was introduced in Linux 2.6.25, so it might be missing if we have old * headers. */ #ifndef TC_RTAB_SIZE #define TC_RTAB_SIZE 1024 #endif -static struct rtnetlink_notifier netdev_linux_cache_notifier; +static struct nln_notifier netdev_linux_cache_notifier; static int cache_notifier_refcount; enum { @@ -90,9 +115,8 @@ enum { VALID_IN6 = 1 << 3, VALID_MTU = 1 << 4, VALID_CARRIER = 1 << 5, - VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */ - VALID_POLICING = 1 << 7, - VALID_HAVE_VPORT_STATS = 1 << 8 + VALID_POLICING = 1 << 6, + VALID_HAVE_VPORT_STATS = 1 << 7 }; struct tap_state { @@ -282,11 +306,13 @@ tc_destroy(struct tc *tc) } static const struct tc_ops tc_ops_htb; +static const struct tc_ops tc_ops_hfsc; static const struct tc_ops tc_ops_default; static const struct tc_ops tc_ops_other; static const struct tc_ops *tcs[] = { &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ + &tc_ops_hfsc, /* Hierarchical fair service curve. */ &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ &tc_ops_other, /* Some other qdisc. */ NULL @@ -328,6 +354,11 @@ struct netdev_dev_linux { struct shash_node *shash_node; 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. */ + struct timer miimon_timer; /* The following are figured out "on demand" only. They are only valid * when the corresponding VALID_* bit in 'cache_valid' is set. */ @@ -337,8 +368,6 @@ struct netdev_dev_linux { 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; bool have_vport_stats; @@ -354,21 +383,12 @@ struct netdev_linux { int fd; }; -/* An AF_INET socket (used for ioctl operations). */ -static int af_inet_sock = -1; +/* Sockets used for ioctl operations. */ +static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */ /* A Netlink routing socket that is not subscribed to any multicast groups. */ static struct nl_sock *rtnl_sock; -struct netdev_linux_notifier { - struct netdev_notifier notifier; - struct list node; -}; - -static struct shash netdev_linux_notifiers = - SHASH_INITIALIZER(&netdev_linux_notifiers); -static struct rtnetlink_notifier netdev_linux_poll_notifier; - /* This is set pretty low because we probably won't learn anything from the * additional log messages. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); @@ -393,6 +413,9 @@ static int set_etheraddr(const char *netdev_name, int hwaddr_family, 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 af_packet_sock(void); +static void netdev_linux_miimon_run(void); +static void netdev_linux_miimon_wait(void); static bool is_netdev_linux_class(const struct netdev_class *netdev_class) @@ -433,7 +456,7 @@ netdev_linux_init(void) /* Create rtnetlink socket. */ if (!status) { - status = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock); + status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); if (status) { VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", strerror(status)); @@ -446,17 +469,29 @@ netdev_linux_init(void) static void netdev_linux_run(void) { - rtnetlink_notifier_run(); + rtnetlink_link_notifier_run(); + netdev_linux_miimon_run(); } static void netdev_linux_wait(void) { - rtnetlink_notifier_wait(); + rtnetlink_link_notifier_wait(); + netdev_linux_miimon_wait(); +} + +static void +netdev_dev_linux_changed(struct netdev_dev_linux *dev) +{ + dev->change_seq++; + if (!dev->change_seq) { + dev->change_seq++; + } + dev->cache_valid = 0; } static void -netdev_linux_cache_cb(const struct rtnetlink_change *change, +netdev_linux_cache_cb(const struct rtnetlink_link_change *change, void *aux OVS_UNUSED) { struct netdev_dev_linux *dev; @@ -468,7 +503,7 @@ netdev_linux_cache_cb(const struct rtnetlink_change *change, if (is_netdev_linux_class(netdev_class)) { dev = netdev_dev_linux_cast(base_dev); - dev->cache_valid = 0; + netdev_dev_linux_changed(dev); } } } else { @@ -479,28 +514,23 @@ netdev_linux_cache_cb(const struct rtnetlink_change *change, netdev_dev_get_devices(&netdev_linux_class, &device_shash); SHASH_FOR_EACH (node, &device_shash) { dev = node->data; - dev->cache_valid = 0; + netdev_dev_linux_changed(dev); } shash_destroy(&device_shash); } } -/* Creates the netdev device of 'type' with 'name'. */ +/* Creates system and internal devices. */ static int -netdev_linux_create_system(const struct netdev_class *class OVS_UNUSED, - const char *name, const struct shash *args, - struct netdev_dev **netdev_devp) +netdev_linux_create(const struct netdev_class *class, const char *name, + struct netdev_dev **netdev_devp) { struct netdev_dev_linux *netdev_dev; int 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); + error = rtnetlink_link_notifier_register(&netdev_linux_cache_notifier, + netdev_linux_cache_cb, NULL); if (error) { return error; } @@ -508,7 +538,8 @@ netdev_linux_create_system(const struct netdev_class *class OVS_UNUSED, cache_notifier_refcount++; netdev_dev = xzalloc(sizeof *netdev_dev); - netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class); + netdev_dev->change_seq = 1; + netdev_dev_init(&netdev_dev->netdev_dev, name, class); *netdev_devp = &netdev_dev->netdev_dev; return 0; @@ -522,8 +553,7 @@ netdev_linux_create_system(const struct netdev_class *class OVS_UNUSED, * be unavailable to other reads for tap devices. */ static int netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, - const char *name, const struct shash *args, - struct netdev_dev **netdev_devp) + const char *name, struct netdev_dev **netdev_devp) { struct netdev_dev_linux *netdev_dev; struct tap_state *state; @@ -531,10 +561,6 @@ netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, struct ifreq ifr; int 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; @@ -548,7 +574,7 @@ netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, /* Create tap device. */ ifr.ifr_flags = IFF_TAP | IFF_NO_PI; - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); + ovs_strzcpy(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)); @@ -586,28 +612,29 @@ 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_); + const struct netdev_class *class = netdev_dev_get_class(netdev_dev_); if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) { netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); } - if (!strcmp(type, "system")) { + if (class == &netdev_linux_class || class == &netdev_internal_class) { cache_notifier_refcount--; if (!cache_notifier_refcount) { - rtnetlink_notifier_unregister(&netdev_linux_cache_notifier); + rtnetlink_link_notifier_unregister(&netdev_linux_cache_notifier); } - } else if (!strcmp(type, "tap")) { + } else if (class == &netdev_tap_class) { destroy_tap(netdev_dev); + } else { + NOT_REACHED(); } free(netdev_dev); } static int -netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, - struct netdev **netdevp) +netdev_linux_open(struct netdev_dev *netdev_dev_, struct netdev **netdevp) { struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); struct netdev_linux *netdev; @@ -619,9 +646,19 @@ netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, netdev->fd = -1; netdev_init(&netdev->netdev, netdev_dev_); - error = netdev_get_flags(&netdev->netdev, &flags); - if (error == ENODEV) { - goto error; + /* Verify that the device really exists, by attempting to read its flags. + * (The flags might be cached, in which case this won't actually do an + * ioctl.) + * + * Don't do this for "internal" netdevs, though, because those have to be + * created as netdev objects before they exist in the kernel, because + * creating them in the kernel happens by passing a netdev object to + * dpif_port_add(). */ + if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) { + error = netdev_get_flags(&netdev->netdev, &flags); + if (error == ENODEV) { + goto error; + } } if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") && @@ -633,53 +670,6 @@ netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, * directions appearing to be reversed. */ netdev->fd = netdev_dev->state.tap.fd; netdev_dev->state.tap.opened = true; - } else if (ethertype != NETDEV_ETH_TYPE_NONE) { - struct sockaddr_ll sll; - int protocol; - int ifindex; - - /* Create file descriptor. */ - protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL - : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 - : ethertype); - netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); - if (netdev->fd < 0) { - error = errno; - goto error; - } - - /* Set non-blocking mode. */ - error = set_nonblocking(netdev->fd); - if (error) { - goto error; - } - - /* Get ethernet device index. */ - error = get_ifindex(&netdev->netdev, &ifindex); - if (error) { - goto error; - } - - /* Bind to specific ethernet device. */ - memset(&sll, 0, sizeof sll); - sll.sll_family = AF_PACKET; - sll.sll_ifindex = ifindex; - if (bind(netdev->fd, - (struct sockaddr *) &sll, sizeof sll) < 0) { - error = errno; - VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_), - strerror(error)); - goto error; - } - - /* Between the socket() and bind() calls above, the socket receives all - * 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->fd); - if (error) { - goto error; - } } *netdevp = &netdev->netdev; @@ -702,9 +692,9 @@ netdev_linux_close(struct netdev *netdev_) free(netdev); } -/* Initializes 'svec' with a list of the names of all known network devices. */ +/* Initializes 'sset' with a list of the names of all known network devices. */ static int -netdev_linux_enumerate(struct svec *svec) +netdev_linux_enumerate(struct sset *sset) { struct if_nameindex *names; @@ -713,7 +703,7 @@ netdev_linux_enumerate(struct svec *svec) size_t i; for (i = 0; names[i].if_name != NULL; i++) { - svec_add(svec, names[i].if_name); + sset_add(sset, names[i].if_name); } if_freenameindex(names); return 0; @@ -724,13 +714,68 @@ netdev_linux_enumerate(struct svec *svec) } } +static int +netdev_linux_listen(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct sockaddr_ll sll; + int ifindex; + int error; + int fd; + + if (netdev->fd >= 0) { + return 0; + } + + /* Create file descriptor. */ + fd = socket(PF_PACKET, SOCK_RAW, 0); + if (fd < 0) { + error = errno; + VLOG_ERR("failed to create raw socket (%s)", strerror(error)); + goto error; + } + + /* Set non-blocking mode. */ + error = set_nonblocking(fd); + if (error) { + goto error; + } + + /* Get ethernet device index. */ + error = get_ifindex(&netdev->netdev, &ifindex); + if (error) { + goto error; + } + + /* Bind to specific ethernet device. */ + 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); + if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) { + error = errno; + VLOG_ERR("%s: failed to bind raw socket (%s)", + netdev_get_name(netdev_), strerror(error)); + goto error; + } + + netdev->fd = fd; + return 0; + +error: + if (fd >= 0) { + close(fd); + } + return error; +} + static int netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); if (netdev->fd < 0) { - /* Device was opened with NETDEV_ETH_TYPE_NONE. */ + /* Device is not listening. */ return -EAGAIN; } @@ -793,15 +838,54 @@ static int netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + for (;;) { + ssize_t retval; - /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. - */ - if (netdev->fd < 0) { - return EPIPE; - } + if (netdev->fd < 0) { + /* Use our AF_PACKET socket to send to this device. */ + struct sockaddr_ll sll; + struct msghdr msg; + struct iovec iov; + int ifindex; + int error; + int sock; + + sock = af_packet_sock(); + if (sock < 0) { + return sock; + } + + error = get_ifindex(netdev_, &ifindex); + if (error) { + return error; + } + + /* We don't bother setting most fields in sockaddr_ll because the + * kernel ignores them for SOCK_RAW. */ + memset(&sll, 0, sizeof sll); + sll.sll_family = AF_PACKET; + sll.sll_ifindex = ifindex; + + iov.iov_base = (void *) data; + iov.iov_len = size; + + msg.msg_name = &sll; + msg.msg_namelen = sizeof sll; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + msg.msg_control = NULL; + msg.msg_controllen = 0; + msg.msg_flags = 0; + + retval = sendmsg(sock, &msg, 0); + } else { + /* Use the netdev's own fd to send to this device. This is + * essential for tap devices, because packets sent to a tap device + * with an AF_PACKET socket will loop back to be *received* again + * on the tap device. */ + retval = write(netdev->fd, data, size); + } - for (;;) { - 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 @@ -913,6 +997,29 @@ netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) return 0; } +/* Sets the maximum size of transmitted (MTU) for given device using linux + * networking ioctl interface. + */ +static int +netdev_linux_set_mtu(const struct netdev *netdev_, int mtu) +{ + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); + struct ifreq ifr; + int error; + + ifr.ifr_mtu = mtu; + error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, + SIOCSIFMTU, "SIOCSIFMTU"); + if (error) { + return error; + } + + netdev_dev->mtu = ifr.ifr_mtu; + netdev_dev->cache_valid |= VALID_MTU; + return 0; +} + /* Returns the ifindex of 'netdev', if successful, as a positive number. * On failure, returns a negative errno value. */ static int @@ -933,6 +1040,11 @@ netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) char *fn = NULL; int fd = -1; + if (netdev_dev->miimon_interval > 0) { + *carrier = netdev_dev->miimon; + return 0; + } + if (!(netdev_dev->cache_valid & VALID_CARRIER)) { char line[8]; int retval; @@ -982,6 +1094,127 @@ exit: return error; } +static int +netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, + struct mii_ioctl_data *data) +{ + struct ifreq ifr; + int error; + + memset(&ifr, 0, sizeof ifr); + memcpy(&ifr.ifr_data, data, sizeof *data); + error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name); + memcpy(data, &ifr.ifr_data, sizeof *data); + + return error; +} + +static int +netdev_linux_get_miimon(const char *name, bool *miimon) +{ + struct mii_ioctl_data data; + int error; + + *miimon = false; + + memset(&data, 0, sizeof data); + error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data); + if (!error) { + /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ + data.reg_num = MII_BMSR; + error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", + &data); + + if (!error) { + *miimon = !!(data.val_out & BMSR_LSTATUS); + } else { + VLOG_WARN_RL(&rl, "%s: failed to query MII", name); + } + } else { + struct ethtool_cmd ecmd; + + VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", + name); + + memset(&ecmd, 0, sizeof ecmd); + error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, + "ETHTOOL_GLINK"); + if (!error) { + struct ethtool_value eval; + + memcpy(&eval, &ecmd, sizeof eval); + *miimon = !!eval.data; + } else { + VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); + } + } + + return error; +} + +static int +netdev_linux_set_miimon_interval(struct netdev *netdev_, + long long int interval) +{ + struct netdev_dev_linux *netdev_dev; + + netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_)); + + interval = interval > 0 ? MAX(interval, 100) : 0; + if (netdev_dev->miimon_interval != interval) { + netdev_dev->miimon_interval = interval; + timer_set_expired(&netdev_dev->miimon_timer); + } + + return 0; +} + +static void +netdev_linux_miimon_run(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_dev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev_dev_linux *dev = node->data; + bool miimon; + + if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) { + continue; + } + + netdev_linux_get_miimon(dev->netdev_dev.name, &miimon); + if (miimon != dev->miimon) { + dev->miimon = miimon; + netdev_dev_linux_changed(dev); + } + + timer_set_duration(&dev->miimon_timer, dev->miimon_interval); + } + + shash_destroy(&device_shash); +} + +static void +netdev_linux_miimon_wait(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_dev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev_dev_linux *dev = node->data; + + if (dev->miimon_interval > 0) { + timer_wait(&dev->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. */ @@ -1010,76 +1243,131 @@ check_for_working_netlink_stats(void) } } -/* 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) +swap_uint64(uint64_t *a, uint64_t *b) { - 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); + uint64_t tmp = *a; + *a = *b; + *b = tmp; +} - netdev_dev->is_tap = !strcmp(type, "tap"); - netdev_dev->is_internal = false; - if (!netdev_dev->is_tap) { - struct ethtool_drvinfo drvinfo; - int error; +static void +get_stats_via_vport(const struct netdev *netdev_, + struct netdev_stats *stats) +{ + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); - memset(&drvinfo, 0, sizeof drvinfo); - error = netdev_linux_do_ethtool(name, - (struct ethtool_cmd *)&drvinfo, - ETHTOOL_GDRVINFO, - "ETHTOOL_GDRVINFO"); + if (netdev_dev->have_vport_stats || + !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) { + int error; - if (!error && !strcmp(drvinfo.driver, "openvswitch")) { - netdev_dev->is_internal = true; - } + error = netdev_vport_get_stats(netdev_, stats); + if (error) { + VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed %d", + netdev_get_name(netdev_), error); } - - netdev_dev->cache_valid |= VALID_IS_PSEUDO; + netdev_dev->have_vport_stats = !error; + netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS; } } -static void -swap_uint64(uint64_t *a, uint64_t *b) +static int +netdev_linux_sys_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) { - *a ^= *b; - *b ^= *a; - *a ^= *b; + static int use_netlink_stats = -1; + int error; + + if (use_netlink_stats < 0) { + use_netlink_stats = check_for_working_netlink_stats(); + } + + 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'. */ +/* Retrieves current device stats for 'netdev-linux'. */ static int 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; + struct netdev_stats dev_stats; int error; - COVERAGE_INC(netdev_get_stats); + get_stats_via_vport(netdev_, stats); - if (netdev_dev->have_vport_stats || - !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) { + error = netdev_linux_sys_get_stats(netdev_, &dev_stats); - error = netdev_vport_get_stats(netdev_, stats); - netdev_dev->have_vport_stats = !error; - netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS; + if (error) { + if (!netdev_dev->have_vport_stats) { + return error; + } else { + return 0; + } } if (!netdev_dev->have_vport_stats) { - if (use_netlink_stats < 0) { - use_netlink_stats = check_for_working_netlink_stats(); - } - if (use_netlink_stats) { - int ifindex; + /* stats not available from OVS then use ioctl stats. */ + *stats = dev_stats; + } else { + stats->rx_errors += dev_stats.rx_errors; + stats->tx_errors += dev_stats.tx_errors; + stats->rx_dropped += dev_stats.rx_dropped; + stats->tx_dropped += dev_stats.tx_dropped; + stats->multicast += dev_stats.multicast; + stats->collisions += dev_stats.collisions; + stats->rx_length_errors += dev_stats.rx_length_errors; + stats->rx_over_errors += dev_stats.rx_over_errors; + stats->rx_crc_errors += dev_stats.rx_crc_errors; + stats->rx_frame_errors += dev_stats.rx_frame_errors; + stats->rx_fifo_errors += dev_stats.rx_fifo_errors; + stats->rx_missed_errors += dev_stats.rx_missed_errors; + stats->tx_aborted_errors += dev_stats.tx_aborted_errors; + stats->tx_carrier_errors += dev_stats.tx_carrier_errors; + stats->tx_fifo_errors += dev_stats.tx_fifo_errors; + stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; + stats->tx_window_errors += dev_stats.tx_window_errors; + } + return 0; +} - error = get_ifindex(netdev_, &ifindex); - if (!error) { - error = get_stats_via_netlink(ifindex, stats); - } +/* Retrieves current device stats for 'netdev-tap' netdev or + * netdev-internal. */ +static int +netdev_pseudo_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) +{ + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); + struct netdev_stats dev_stats; + int error; + + get_stats_via_vport(netdev_, stats); + + error = netdev_linux_sys_get_stats(netdev_, &dev_stats); + if (error) { + if (!netdev_dev->have_vport_stats) { + return error; } else { - error = get_stats_via_proc(netdev_get_name(netdev_), stats); + return 0; } } @@ -1089,9 +1377,8 @@ netdev_linux_get_stats(const struct netdev *netdev_, * them back here. This does not apply if we are getting stats from the * vport layer because it always tracks stats from the perspective of the * switch. */ - netdev_linux_update_is_pseudo(netdev_dev); - if (!error && !netdev_dev->have_vport_stats && - (netdev_dev->is_internal || netdev_dev->is_tap)) { + if (!netdev_dev->have_vport_stats) { + *stats = dev_stats; swap_uint64(&stats->rx_packets, &stats->tx_packets); swap_uint64(&stats->rx_bytes, &stats->tx_bytes); swap_uint64(&stats->rx_errors, &stats->tx_errors); @@ -1107,9 +1394,17 @@ netdev_linux_get_stats(const struct netdev *netdev_, stats->tx_fifo_errors = 0; stats->tx_heartbeat_errors = 0; stats->tx_window_errors = 0; - } + } else { + stats->rx_dropped += dev_stats.tx_dropped; + stats->tx_dropped += dev_stats.rx_dropped; - return error; + stats->rx_errors += dev_stats.tx_errors; + stats->tx_errors += dev_stats.rx_errors; + + stats->multicast += dev_stats.multicast; + stats->collisions += dev_stats.collisions; + } + return 0; } /* Stores the features supported by 'netdev' into each of '*current', @@ -1117,7 +1412,7 @@ netdev_linux_get_stats(const struct netdev *netdev_, * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if * successful, otherwise a positive errno value. */ static int -netdev_linux_get_features(struct netdev *netdev, +netdev_linux_get_features(const struct netdev *netdev, uint32_t *current, uint32_t *advertised, uint32_t *supported, uint32_t *peer) { @@ -1369,6 +1664,9 @@ netdev_linux_remove_policing(struct netdev *netdev) int error; tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); + if (!tcmsg) { + return ENODEV; + } tcmsg->tcm_handle = tc_make_handle(0xffff, 0); tcmsg->tcm_parent = TC_H_INGRESS; nl_msg_put_string(&request, TCA_KIND, "ingress"); @@ -1436,14 +1734,14 @@ netdev_linux_set_policing(struct netdev *netdev, static int netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, - struct svec *types) + struct sset *types) { const struct tc_ops **opsp; for (opsp = tcs; *opsp != NULL; opsp++) { const struct tc_ops *ops = *opsp; if (ops->tc_install && ops->ovs_name[0] != '\0') { - svec_add(types, ops->ovs_name); + sset_add(types, ops->ovs_name); } } return 0; @@ -1648,16 +1946,20 @@ netdev_linux_get_queue_stats(const struct netdev *netdev, } } -static void +static bool start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) { struct ofpbuf request; struct tcmsg *tcmsg; tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request); + if (!tcmsg) { + return false; + } tcmsg->tcm_parent = 0; nl_dump_start(dump, rtnl_sock, &request); ofpbuf_uninit(&request); + return true; } static int @@ -1714,7 +2016,9 @@ netdev_linux_dump_queue_stats(const struct netdev *netdev, } last_error = 0; - start_queue_dump(netdev, &dump); + if (!start_queue_dump(netdev, &dump)) { + return ENODEV; + } while (nl_dump_next(&dump, &msg)) { error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux); if (error) { @@ -1844,7 +2148,7 @@ do_set_addr(struct netdev *netdev, int ioctl_nr, const char *ioctl_name, struct in_addr addr) { struct ifreq ifr; - strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name); + ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name); make_in4_sockaddr(&ifr.ifr_addr, addr); return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, @@ -1864,7 +2168,6 @@ netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) make_in4_sockaddr(&rt.rt_gateway, router); make_in4_sockaddr(&rt.rt_genmask, any); rt.rt_flags = RTF_UP | RTF_GATEWAY; - COVERAGE_INC(netdev_add_router); error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; if (error) { VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); @@ -1892,7 +2195,7 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, while (fgets(line, sizeof line, stream)) { if (++ln >= 2) { char iface[17]; - uint32_t dest, gateway, mask; + ovs_be32 dest, gateway, mask; int refcnt, metric, mtu; unsigned int flags, use, window, irtt; @@ -1933,26 +2236,47 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, return ENXIO; } +static int +netdev_linux_get_status(const struct netdev *netdev, struct shash *sh) +{ + struct ethtool_drvinfo drvinfo; + int error; + + memset(&drvinfo, 0, sizeof drvinfo); + error = netdev_linux_do_ethtool(netdev_get_name(netdev), + (struct ethtool_cmd *)&drvinfo, + ETHTOOL_GDRVINFO, + "ETHTOOL_GDRVINFO"); + if (!error) { + shash_add(sh, "driver_name", xstrdup(drvinfo.driver)); + shash_add(sh, "driver_version", xstrdup(drvinfo.version)); + shash_add(sh, "firmware_version", xstrdup(drvinfo.fw_version)); + } + + return error; +} + /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be * successfully retrieved, it stores the corresponding MAC address in 'mac' and * returns 0. Otherwise, it returns a positive errno value; in particular, * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ static int netdev_linux_arp_lookup(const struct netdev *netdev, - uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) + ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) { struct arpreq r; struct sockaddr_in sin; int retval; memset(&r, 0, sizeof r); + memset(&sin, 0, sizeof sin); 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_get_name(netdev), sizeof r.arp_dev); + ovs_strzcpy(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) { @@ -2008,208 +2332,100 @@ netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, return error; } -static void -poll_notify(struct list *list) +static unsigned int +netdev_linux_change_seq(const struct netdev *netdev) { - struct netdev_linux_notifier *notifier; - LIST_FOR_EACH (notifier, node, list) { - struct netdev_notifier *n = ¬ifier->notifier; - n->cb(n); - } + return netdev_dev_linux_cast(netdev_get_dev(netdev))->change_seq; } -static void -netdev_linux_poll_cb(const struct rtnetlink_change *change, - void *aux OVS_UNUSED) -{ - if (change) { - struct list *list = shash_find_data(&netdev_linux_notifiers, - change->ifname); - if (list) { - poll_notify(list); - } - } else { - struct shash_node *node; - SHASH_FOR_EACH (node, &netdev_linux_notifiers) { - poll_notify(node->data); - } - } +#define NETDEV_LINUX_CLASS(NAME, CREATE, ENUMERATE, GET_STATS, SET_STATS) \ +{ \ + NAME, \ + \ + netdev_linux_init, \ + netdev_linux_run, \ + netdev_linux_wait, \ + \ + CREATE, \ + netdev_linux_destroy, \ + NULL, /* get_config */ \ + NULL, /* set_config */ \ + \ + netdev_linux_open, \ + netdev_linux_close, \ + \ + ENUMERATE, \ + \ + netdev_linux_listen, \ + netdev_linux_recv, \ + netdev_linux_recv_wait, \ + netdev_linux_drain, \ + \ + netdev_linux_send, \ + netdev_linux_send_wait, \ + \ + netdev_linux_set_etheraddr, \ + netdev_linux_get_etheraddr, \ + netdev_linux_get_mtu, \ + netdev_linux_set_mtu, \ + netdev_linux_get_ifindex, \ + netdev_linux_get_carrier, \ + netdev_linux_set_miimon_interval, \ + GET_STATS, \ + SET_STATS, \ + \ + netdev_linux_get_features, \ + netdev_linux_set_advertisements, \ + netdev_linux_get_vlan_vid, \ + \ + netdev_linux_set_policing, \ + netdev_linux_get_qos_types, \ + netdev_linux_get_qos_capabilities, \ + netdev_linux_get_qos, \ + netdev_linux_set_qos, \ + netdev_linux_get_queue, \ + netdev_linux_set_queue, \ + netdev_linux_delete_queue, \ + netdev_linux_get_queue_stats, \ + netdev_linux_dump_queues, \ + netdev_linux_dump_queue_stats, \ + \ + netdev_linux_get_in4, \ + netdev_linux_set_in4, \ + netdev_linux_get_in6, \ + netdev_linux_add_router, \ + netdev_linux_get_next_hop, \ + netdev_linux_get_status, \ + netdev_linux_arp_lookup, \ + \ + netdev_linux_update_flags, \ + \ + netdev_linux_change_seq \ } -static int -netdev_linux_poll_add(struct netdev *netdev, - void (*cb)(struct netdev_notifier *), void *aux, - struct netdev_notifier **notifierp) -{ - const char *netdev_name = netdev_get_name(netdev); - struct netdev_linux_notifier *notifier; - struct list *list; - - if (shash_is_empty(&netdev_linux_notifiers)) { - int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier, - netdev_linux_poll_cb, NULL); - if (error) { - return error; - } - } - - list = shash_find_data(&netdev_linux_notifiers, netdev_name); - if (!list) { - list = xmalloc(sizeof *list); - list_init(list); - shash_add(&netdev_linux_notifiers, netdev_name, list); - } - - notifier = xmalloc(sizeof *notifier); - netdev_notifier_init(¬ifier->notifier, netdev, cb, aux); - list_push_back(list, ¬ifier->node); - *notifierp = ¬ifier->notifier; - return 0; -} - -static void -netdev_linux_poll_remove(struct netdev_notifier *notifier_) -{ - struct netdev_linux_notifier *notifier = - CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier); - struct list *list; - - /* Remove 'notifier' from its list. */ - list = list_remove(¬ifier->node); - if (list_is_empty(list)) { - /* The list is now empty. Remove it from the hash and free it. */ - const char *netdev_name = netdev_get_name(notifier->notifier.netdev); - shash_delete(&netdev_linux_notifiers, - shash_find(&netdev_linux_notifiers, netdev_name)); - free(list); - } - free(notifier); - - /* If that was the last notifier, unregister. */ - if (shash_is_empty(&netdev_linux_notifiers)) { - rtnetlink_notifier_unregister(&netdev_linux_poll_notifier); - } -} - -const struct netdev_class netdev_linux_class = { - "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_enumerate, - - netdev_linux_recv, - netdev_linux_recv_wait, - netdev_linux_drain, - - netdev_linux_send, - netdev_linux_send_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, - netdev_vport_set_stats, - - netdev_linux_get_features, - netdev_linux_set_advertisements, - netdev_linux_get_vlan_vid, - - netdev_linux_set_policing, - netdev_linux_get_qos_types, - netdev_linux_get_qos_capabilities, - netdev_linux_get_qos, - netdev_linux_set_qos, - netdev_linux_get_queue, - netdev_linux_set_queue, - netdev_linux_delete_queue, - netdev_linux_get_queue_stats, - netdev_linux_dump_queues, - netdev_linux_dump_queue_stats, - - netdev_linux_get_in4, - netdev_linux_set_in4, - netdev_linux_get_in6, - netdev_linux_add_router, - netdev_linux_get_next_hop, - netdev_linux_arp_lookup, - - netdev_linux_update_flags, - - netdev_linux_poll_add, - netdev_linux_poll_remove, -}; - -const struct netdev_class netdev_tap_class = { - "tap", - - netdev_linux_init, - netdev_linux_run, - netdev_linux_wait, - - netdev_linux_create_tap, - netdev_linux_destroy, - NULL, /* reconfigure */ - - netdev_linux_open, - netdev_linux_close, - - NULL, /* enumerate */ - - netdev_linux_recv, - netdev_linux_recv_wait, - netdev_linux_drain, - - netdev_linux_send, - netdev_linux_send_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_get_vlan_vid, - - netdev_linux_set_policing, - netdev_linux_get_qos_types, - netdev_linux_get_qos_capabilities, - netdev_linux_get_qos, - netdev_linux_set_qos, - netdev_linux_get_queue, - netdev_linux_set_queue, - netdev_linux_delete_queue, - netdev_linux_get_queue_stats, - netdev_linux_dump_queues, - netdev_linux_dump_queue_stats, - - netdev_linux_get_in4, - netdev_linux_set_in4, - netdev_linux_get_in6, - netdev_linux_add_router, - netdev_linux_get_next_hop, - netdev_linux_arp_lookup, - - netdev_linux_update_flags, - - netdev_linux_poll_add, - netdev_linux_poll_remove, -}; +const struct netdev_class netdev_linux_class = + NETDEV_LINUX_CLASS( + "system", + netdev_linux_create, + netdev_linux_enumerate, + netdev_linux_get_stats, + NULL); /* set_stats */ + +const struct netdev_class netdev_tap_class = + NETDEV_LINUX_CLASS( + "tap", + netdev_linux_create_tap, + NULL, /* enumerate */ + netdev_pseudo_get_stats, + NULL); /* set_stats */ + +const struct netdev_class netdev_internal_class = + NETDEV_LINUX_CLASS( + "internal", + netdev_linux_create, + NULL, /* enumerate */ + netdev_pseudo_get_stats, + netdev_vport_set_stats); /* HTB traffic control class. */ @@ -2236,7 +2452,7 @@ htb_get__(const struct netdev *netdev) return CONTAINER_OF(netdev_dev->tc, struct htb, tc); } -static struct htb * +static void htb_install__(struct netdev *netdev, uint64_t max_rate) { struct netdev_dev_linux *netdev_dev = @@ -2248,14 +2464,11 @@ htb_install__(struct netdev *netdev, uint64_t max_rate) htb->max_rate = max_rate; netdev_dev->tc = &htb->tc; - - return htb; } /* Create an HTB qdisc. * - * Equivalent to "tc qdisc add dev root handle 1: htb default - * 0". */ + * Equivalent to "tc qdisc add dev root handle 1: htb default 1". */ static int htb_setup_qdisc__(struct netdev *netdev) { @@ -2268,6 +2481,9 @@ htb_setup_qdisc__(struct netdev *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; @@ -2276,7 +2492,7 @@ htb_setup_qdisc__(struct netdev *netdev) memset(&opt, 0, sizeof opt); opt.rate2quantum = 10; opt.version = 3; - opt.defcls = 0; + opt.defcls = 1; opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); @@ -2298,7 +2514,12 @@ htb_setup_class__(struct netdev *netdev, unsigned int handle, int error; int mtu; - netdev_get_mtu(netdev, &mtu); + error = netdev_get_mtu(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); @@ -2308,6 +2529,9 @@ htb_setup_class__(struct netdev *netdev, unsigned int handle, opt.prio = class->priority; tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); + if (!tcmsg) { + return ENODEV; + } tcmsg->tcm_handle = handle; tcmsg->tcm_parent = parent; @@ -2413,15 +2637,19 @@ htb_parse_class_details__(struct netdev *netdev, const char *max_rate_s = shash_find_data(details, "max-rate"); const char *burst_s = shash_find_data(details, "burst"); const char *priority_s = shash_find_data(details, "priority"); - int mtu; + int mtu, error; - /* min-rate. Don't allow a min-rate below 1500 bytes/s. */ - if (!min_rate_s) { - /* min-rate is required. */ - return EINVAL; + error = netdev_get_mtu(netdev, &mtu); + if (error) { + VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", + netdev_get_name(netdev)); + return error; } - hc->min_rate = strtoull(min_rate_s, NULL, 10) / 8; - hc->min_rate = MAX(hc->min_rate, 1500); + + /* HTB requires at least an mtu sized min-rate to send any traffic even + * on uncongested links. */ + hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; + hc->min_rate = MAX(hc->min_rate, mtu); hc->min_rate = MIN(hc->min_rate, htb->max_rate); /* max-rate */ @@ -2440,7 +2668,6 @@ htb_parse_class_details__(struct netdev *netdev, * doesn't include the Ethernet header, we need to add at least 14 (18?) to * the MTU. We actually add 64, instead of 14, as a guard against * additional headers get tacked on somewhere that we're not aware of. */ - netdev_get_mtu(netdev, &mtu); hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0; hc->burst = MAX(hc->burst, mtu + 64); @@ -2519,20 +2746,19 @@ htb_update_queue__(struct netdev *netdev, unsigned int queue_id, static int htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) { - struct shash details = SHASH_INITIALIZER(&details); struct ofpbuf msg; struct nl_dump dump; struct htb_class hc; - struct htb *htb; /* Get qdisc options. */ hc.max_rate = 0; htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); - htb = htb_install__(netdev, hc.max_rate); + htb_install__(netdev, hc.max_rate); /* Get queues. */ - start_queue_dump(netdev, &dump); - shash_init(&details); + if (!start_queue_dump(netdev, &dump)) { + return ENODEV; + } while (nl_dump_next(&dump, &msg)) { unsigned int queue_id; @@ -2682,6 +2908,510 @@ static const struct tc_ops tc_ops_htb = { htb_class_dump_stats }; +/* "linux-hfsc" traffic control class. */ + +#define HFSC_N_QUEUES 0xf000 + +struct hfsc { + struct tc tc; + uint32_t max_rate; +}; + +struct hfsc_class { + struct tc_queue tc_queue; + uint32_t min_rate; + uint32_t max_rate; +}; + +static struct hfsc * +hfsc_get__(const struct netdev *netdev) +{ + struct netdev_dev_linux *netdev_dev; + netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); + return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc); +} + +static struct hfsc_class * +hfsc_class_cast__(const struct tc_queue *queue) +{ + return CONTAINER_OF(queue, struct hfsc_class, tc_queue); +} + +static void +hfsc_install__(struct netdev *netdev, uint32_t max_rate) +{ + struct netdev_dev_linux * netdev_dev; + struct hfsc *hfsc; + + netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); + hfsc = xmalloc(sizeof *hfsc); + tc_init(&hfsc->tc, &tc_ops_hfsc); + hfsc->max_rate = max_rate; + netdev_dev->tc = &hfsc->tc; +} + +static void +hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, + const struct hfsc_class *hc) +{ + size_t hash; + struct hfsc *hfsc; + struct hfsc_class *hcp; + struct tc_queue *queue; + + hfsc = hfsc_get__(netdev); + hash = hash_int(queue_id, 0); + + queue = tc_find_queue__(netdev, queue_id, hash); + if (queue) { + hcp = hfsc_class_cast__(queue); + } else { + hcp = xmalloc(sizeof *hcp); + queue = &hcp->tc_queue; + queue->queue_id = queue_id; + hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); + } + + hcp->min_rate = hc->min_rate; + hcp->max_rate = hc->max_rate; +} + +static int +hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) +{ + const struct tc_service_curve *rsc, *fsc, *usc; + static const struct nl_policy tca_hfsc_policy[] = { + [TCA_HFSC_RSC] = { + .type = NL_A_UNSPEC, + .optional = false, + .min_len = sizeof(struct tc_service_curve), + }, + [TCA_HFSC_FSC] = { + .type = NL_A_UNSPEC, + .optional = false, + .min_len = sizeof(struct tc_service_curve), + }, + [TCA_HFSC_USC] = { + .type = NL_A_UNSPEC, + .optional = false, + .min_len = sizeof(struct tc_service_curve), + }, + }; + struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; + + if (!nl_parse_nested(nl_options, tca_hfsc_policy, + attrs, ARRAY_SIZE(tca_hfsc_policy))) { + VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); + return EPROTO; + } + + rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); + fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); + usc = nl_attr_get(attrs[TCA_HFSC_USC]); + + if (rsc->m1 != 0 || rsc->d != 0 || + fsc->m1 != 0 || fsc->d != 0 || + usc->m1 != 0 || usc->d != 0) { + VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " + "Non-linear service curves are not supported."); + return EPROTO; + } + + if (rsc->m2 != fsc->m2) { + VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " + "Real-time service curves are not supported "); + return EPROTO; + } + + if (rsc->m2 > usc->m2) { + VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " + "Min-rate service curve is greater than " + "the max-rate service curve."); + return EPROTO; + } + + class->min_rate = fsc->m2; + class->max_rate = usc->m2; + return 0; +} + +static int +hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, + struct hfsc_class *options, + struct netdev_queue_stats *stats) +{ + int error; + unsigned int handle; + struct nlattr *nl_options; + + error = tc_parse_class(tcmsg, &handle, &nl_options, stats); + if (error) { + return error; + } + + if (queue_id) { + unsigned int major, minor; + + major = tc_get_major(handle); + minor = tc_get_minor(handle); + if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { + *queue_id = minor - 1; + } else { + return EPROTO; + } + } + + if (options) { + error = hfsc_parse_tca_options__(nl_options, options); + } + + return error; +} + +static int +hfsc_query_class__(const struct netdev *netdev, unsigned int handle, + unsigned int parent, struct hfsc_class *options, + struct netdev_queue_stats *stats) +{ + int error; + struct ofpbuf *reply; + + error = tc_query_class(netdev, handle, parent, &reply); + if (error) { + return error; + } + + error = hfsc_parse_tcmsg__(reply, NULL, options, stats); + ofpbuf_delete(reply); + return error; +} + +static void +hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details, + struct hfsc_class *class) +{ + uint32_t max_rate; + const char *max_rate_s; + + max_rate_s = shash_find_data(details, "max-rate"); + max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; + + if (!max_rate) { + uint32_t current; + + netdev_get_features(netdev, ¤t, NULL, NULL, NULL); + max_rate = netdev_features_to_bps(current) / 8; + } + + class->min_rate = max_rate; + class->max_rate = max_rate; +} + +static int +hfsc_parse_class_details__(struct netdev *netdev, + const struct shash *details, + struct hfsc_class * class) +{ + const struct hfsc *hfsc; + uint32_t min_rate, max_rate; + const char *min_rate_s, *max_rate_s; + + hfsc = hfsc_get__(netdev); + min_rate_s = shash_find_data(details, "min-rate"); + max_rate_s = shash_find_data(details, "max-rate"); + + min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; + min_rate = MAX(min_rate, 1); + min_rate = MIN(min_rate, hfsc->max_rate); + + max_rate = (max_rate_s + ? strtoull(max_rate_s, NULL, 10) / 8 + : hfsc->max_rate); + max_rate = MAX(max_rate, min_rate); + max_rate = MIN(max_rate, hfsc->max_rate); + + class->min_rate = min_rate; + class->max_rate = max_rate; + + return 0; +} + +/* Create an HFSC qdisc. + * + * Equivalent to "tc qdisc add dev root handle 1: hfsc default 1". */ +static int +hfsc_setup_qdisc__(struct netdev * netdev) +{ + struct tcmsg *tcmsg; + struct ofpbuf request; + struct tc_hfsc_qopt opt; + + 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); + opt.defcls = 1; + + nl_msg_put_string(&request, TCA_KIND, "hfsc"); + nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); + + return tc_transact(&request, NULL); +} + +/* Create an HFSC class. + * + * Equivalent to "tc class add parent classid hfsc + * sc rate ul rate " */ +static int +hfsc_setup_class__(struct netdev *netdev, unsigned int handle, + unsigned int parent, struct hfsc_class *class) +{ + int error; + size_t opt_offset; + struct tcmsg *tcmsg; + struct ofpbuf request; + struct tc_service_curve min, max; + + tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); + + if (!tcmsg) { + return ENODEV; + } + + tcmsg->tcm_handle = handle; + tcmsg->tcm_parent = parent; + + min.m1 = 0; + min.d = 0; + min.m2 = class->min_rate; + + max.m1 = 0; + max.d = 0; + max.m2 = class->max_rate; + + nl_msg_put_string(&request, TCA_KIND, "hfsc"); + opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); + nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); + nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); + nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); + nl_msg_end_nested(&request, opt_offset); + + error = tc_transact(&request, NULL); + if (error) { + VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " + "min-rate %ubps, max-rate %ubps (%s)", + netdev_get_name(netdev), + tc_get_major(handle), tc_get_minor(handle), + tc_get_major(parent), tc_get_minor(parent), + class->min_rate, class->max_rate, strerror(error)); + } + + return error; +} + +static int +hfsc_tc_install(struct netdev *netdev, const struct shash *details) +{ + int error; + struct hfsc_class class; + + error = hfsc_setup_qdisc__(netdev); + + if (error) { + return error; + } + + hfsc_parse_qdisc_details__(netdev, details, &class); + error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), + tc_make_handle(1, 0), &class); + + if (error) { + return error; + } + + hfsc_install__(netdev, class.max_rate); + return 0; +} + +static int +hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) +{ + struct ofpbuf msg; + struct nl_dump dump; + 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)) { + return ENODEV; + } + + while (nl_dump_next(&dump, &msg)) { + unsigned int queue_id; + + if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { + hfsc_update_queue__(netdev, queue_id, &hc); + } + } + + nl_dump_done(&dump); + return 0; +} + +static void +hfsc_tc_destroy(struct tc *tc) +{ + struct hfsc *hfsc; + struct hfsc_class *hc, *next; + + hfsc = CONTAINER_OF(tc, struct hfsc, tc); + + HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { + hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); + free(hc); + } + + tc_destroy(tc); + free(hfsc); +} + +static int +hfsc_qdisc_get(const struct netdev *netdev, struct shash *details) +{ + const struct hfsc *hfsc; + hfsc = hfsc_get__(netdev); + shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate)); + return 0; +} + +static int +hfsc_qdisc_set(struct netdev *netdev, const struct shash *details) +{ + int error; + struct hfsc_class class; + + hfsc_parse_qdisc_details__(netdev, details, &class); + error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), + tc_make_handle(1, 0), &class); + + if (!error) { + hfsc_get__(netdev)->max_rate = class.max_rate; + } + + return error; +} + +static int +hfsc_class_get(const struct netdev *netdev OVS_UNUSED, + const struct tc_queue *queue, struct shash *details) +{ + const struct hfsc_class *hc; + + hc = hfsc_class_cast__(queue); + shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate)); + if (hc->min_rate != hc->max_rate) { + shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate)); + } + return 0; +} + +static int +hfsc_class_set(struct netdev *netdev, unsigned int queue_id, + const struct shash *details) +{ + int error; + struct hfsc_class class; + + error = hfsc_parse_class_details__(netdev, details, &class); + if (error) { + return error; + } + + error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), + tc_make_handle(1, 0xfffe), &class); + if (error) { + return error; + } + + hfsc_update_queue__(netdev, queue_id, &class); + return 0; +} + +static int +hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) +{ + int error; + struct hfsc *hfsc; + struct hfsc_class *hc; + + hc = hfsc_class_cast__(queue); + hfsc = hfsc_get__(netdev); + + error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); + if (!error) { + hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); + free(hc); + } + return error; +} + +static int +hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, + struct netdev_queue_stats *stats) +{ + return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), + tc_make_handle(1, 0xfffe), NULL, stats); +} + +static int +hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, + const struct ofpbuf *nlmsg, + netdev_dump_queue_stats_cb *cb, void *aux) +{ + struct netdev_queue_stats stats; + unsigned int handle, major, minor; + int error; + + error = tc_parse_class(nlmsg, &handle, NULL, &stats); + if (error) { + return error; + } + + major = tc_get_major(handle); + minor = tc_get_minor(handle); + if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { + (*cb)(minor - 1, &stats, aux); + } + return 0; +} + +static const struct tc_ops tc_ops_hfsc = { + "hfsc", /* linux_name */ + "linux-hfsc", /* ovs_name */ + HFSC_N_QUEUES, /* n_queues */ + hfsc_tc_install, /* tc_install */ + hfsc_tc_load, /* tc_load */ + hfsc_tc_destroy, /* tc_destroy */ + hfsc_qdisc_get, /* qdisc_get */ + hfsc_qdisc_set, /* qdisc_set */ + hfsc_class_get, /* class_get */ + hfsc_class_set, /* class_set */ + hfsc_class_delete, /* class_delete */ + hfsc_class_get_stats, /* class_get_stats */ + hfsc_class_dump_stats /* class_dump_stats */ +}; + /* "linux-default" traffic control class. * * This class represents the default, unnamed Linux qdisc. It corresponds to @@ -3084,6 +3814,9 @@ tc_query_class(const struct netdev *netdev, int error; tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request); + if (!tcmsg) { + return ENODEV; + } tcmsg->tcm_handle = handle; tcmsg->tcm_parent = parent; @@ -3107,6 +3840,9 @@ tc_delete_class(const struct netdev *netdev, unsigned int handle) int error; tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request); + if (!tcmsg) { + return ENODEV; + } tcmsg->tcm_handle = handle; tcmsg->tcm_parent = 0; @@ -3131,6 +3867,9 @@ tc_del_qdisc(struct netdev *netdev) int error; tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); + if (!tcmsg) { + return ENODEV; + } tcmsg->tcm_handle = tc_make_handle(1, 0); tcmsg->tcm_parent = TC_H_ROOT; @@ -3182,6 +3921,9 @@ tc_query_qdisc(const struct netdev *netdev) * We could check for Linux 2.6.35+ and use a more straightforward method * there. */ 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; @@ -3291,6 +4033,34 @@ tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); } + +/* Copies 'src' into 'dst', performing format conversion in the process. */ +static void +netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, + const struct rtnl_link_stats *src) +{ + 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; +} /* Utility functions. */ @@ -3311,7 +4081,6 @@ get_stats_via_netlink(int ifindex, struct netdev_stats *stats) struct ofpbuf request; struct ofpbuf *reply; struct ifinfomsg *ifi; - const struct rtnl_link_stats *rtnl_stats; struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; int error; @@ -3339,28 +4108,7 @@ get_stats_via_netlink(int ifindex, struct netdev_stats *stats) return EPROTO; } - rtnl_stats = nl_attr_get(attrs[IFLA_STATS]); - stats->rx_packets = rtnl_stats->rx_packets; - stats->tx_packets = rtnl_stats->tx_packets; - stats->rx_bytes = rtnl_stats->rx_bytes; - stats->tx_bytes = rtnl_stats->tx_bytes; - stats->rx_errors = rtnl_stats->rx_errors; - stats->tx_errors = rtnl_stats->tx_errors; - stats->rx_dropped = rtnl_stats->rx_dropped; - stats->tx_dropped = rtnl_stats->tx_dropped; - stats->multicast = rtnl_stats->multicast; - stats->collisions = rtnl_stats->collisions; - stats->rx_length_errors = rtnl_stats->rx_length_errors; - stats->rx_over_errors = rtnl_stats->rx_over_errors; - stats->rx_crc_errors = rtnl_stats->rx_crc_errors; - stats->rx_frame_errors = rtnl_stats->rx_frame_errors; - stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors; - stats->rx_missed_errors = rtnl_stats->rx_missed_errors; - stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors; - stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors; - stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors; - stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors; - stats->tx_window_errors = rtnl_stats->tx_window_errors; + netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS])); ofpbuf_delete(reply); @@ -3451,7 +4199,7 @@ do_get_ifindex(const char *netdev_name) { struct ifreq ifr; - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); + ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); COVERAGE_INC(netdev_get_ifindex); if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", @@ -3486,11 +4234,15 @@ get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) int hwaddr_family; memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); + ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); COVERAGE_INC(netdev_get_hwaddr); if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s", - netdev_name, strerror(errno)); + /* ENODEV probably means that a vif disappeared asynchronously and + * hasn't been removed from the database yet, so reduce the log level + * to INFO for that case. */ + VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR, + "ioctl(SIOCGIFHWADDR) on %s device failed: %s", + netdev_name, strerror(errno)); return errno; } hwaddr_family = ifr.ifr_hwaddr.sa_family; @@ -3509,7 +4261,7 @@ set_etheraddr(const char *netdev_name, int hwaddr_family, struct ifreq ifr; memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); + ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); ifr.ifr_hwaddr.sa_family = hwaddr_family; memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); COVERAGE_INC(netdev_set_hwaddr); @@ -3528,7 +4280,7 @@ netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, struct ifreq ifr; memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); + ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); ifr.ifr_data = (caddr_t) ecmd; ecmd->cmd = cmd; @@ -3547,11 +4299,56 @@ netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, } } +/* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If + * 'enable' is true, the bit is set. Otherwise, it is cleared. */ +int +netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, + const char *flag_name, bool enable) +{ + const char *netdev_name = netdev_get_name(netdev); + struct ethtool_value evalue; + uint32_t new_flags; + int error; + + memset(&evalue, 0, sizeof evalue); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); + if (error) { + return error; + } + + evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); + if (error) { + return error; + } + + memset(&evalue, 0, sizeof evalue); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); + if (error) { + return error; + } + + if (new_flags != evalue.data) { + VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " + "device %s failed", enable ? "enable" : "disable", + flag_name, netdev_name); + return EOPNOTSUPP; + } + + return 0; +} + static int netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, const char *cmd_name) { - strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name); + ovs_strzcpy(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", name, cmd_name, strerror(errno)); @@ -3575,3 +4372,22 @@ netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, } return error; } + +/* Returns an AF_PACKET raw socket or a negative errno value. */ +static int +af_packet_sock(void) +{ + static int sock = INT_MIN; + + if (sock == INT_MIN) { + sock = socket(AF_PACKET, SOCK_RAW, 0); + if (sock >= 0) { + set_nonblocking(sock); + } else { + sock = -errno; + VLOG_ERR("failed to create packet socket: %s", strerror(errno)); + } + } + + return sock; +}