X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=4d2f3ac0977c8f3c7dbb71318cd6a7157758dd91;hb=e0edde6fee279cdbbf3c179f5f50adaf0c7c7f1e;hp=0c5ba8b85c81fa969601b8e047648a9483dcc64a;hpb=d98e60075528c3065ad453f7add4b30f22edcde3;p=sliver-openvswitch.git diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 0c5ba8b85..4d2f3ac09 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, 2012 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -15,16 +15,21 @@ */ #include + +#include "netdev-linux.h" + #include #include #include #include #include #include +#include #include -#include #include #include +#include +#include #include #include #include @@ -33,9 +38,7 @@ #include #include #include -#include #include -#include #include #include #include @@ -46,6 +49,7 @@ #include #include "coverage.h" +#include "dpif-linux.h" #include "dynamic-string.h" #include "fatal-signal.h" #include "hash.h" @@ -53,17 +57,28 @@ #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); + +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 +89,22 @@ 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 = NULL; static int cache_notifier_refcount; enum { @@ -89,10 +113,10 @@ enum { VALID_IN4 = 1 << 2, 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 << 5, + VALID_VPORT_STAT_ERROR = 1 << 6, + VALID_DRVINFO = 1 << 7, + VALID_FEATURES = 1 << 8, }; 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 @@ -303,6 +329,9 @@ static unsigned int tc_buffer_per_jiffy(unsigned int rate); static struct tcmsg *tc_make_request(const struct netdev *, int type, 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_parse_qdisc(const struct ofpbuf *, const char **kind, struct nlattr **options); @@ -328,6 +357,11 @@ 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. */ @@ -336,12 +370,24 @@ struct netdev_dev_linux { struct in_addr address, netmask; struct in6_addr in6; int mtu; - int carrier; - bool is_internal; /* Is this an openvswitch internal device? */ - bool is_tap; /* Is this a tuntap device? */ + unsigned int ifi_flags; + long long int carrier_resets; uint32_t kbits_rate; /* Policing data. */ uint32_t kbits_burst; - bool have_vport_stats; + int vport_stats_error; /* Cached error code from vport_get_stats(). + 0 or an errno value. */ + int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */ + int ether_addr_error; /* Cached error code from set/get etheraddr. */ + 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. */ + + enum netdev_features current; /* Cached from ETHTOOL_GSET. */ + enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */ + enum netdev_features supported; /* Cached from ETHTOOL_GSET. */ + enum netdev_features peer; /* Cached from ETHTOOL_GSET. */ + + struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ struct tc *tc; union { @@ -354,21 +400,12 @@ 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); @@ -381,18 +418,20 @@ static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd, const char *cmd_name); static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, int cmd, const char *cmd_name); -static int get_flags(const struct netdev *, int *flagsp); -static int set_flags(struct netdev *, int flags); +static int get_flags(const struct netdev_dev *, unsigned int *flags); +static int set_flags(struct netdev *, unsigned int flags); static int do_get_ifindex(const char *netdev_name); static int get_ifindex(const struct netdev *, int *ifindexp); 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, int hwaddr_family, - const uint8_t[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 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 +472,7 @@ 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 +485,88 @@ netdev_linux_init(void) static void netdev_linux_run(void) { - rtnetlink_notifier_run(); + rtnetlink_link_run(); + netdev_linux_miimon_run(); } static void netdev_linux_wait(void) { - rtnetlink_notifier_wait(); + rtnetlink_link_wait(); + netdev_linux_miimon_wait(); +} + +static int +netdev_linux_get_drvinfo(struct netdev_dev_linux *netdev_dev) +{ + + int error; + + if (netdev_dev->cache_valid & VALID_DRVINFO) { + return 0; + } + + memset(&netdev_dev->drvinfo, 0, sizeof netdev_dev->drvinfo); + error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name, + (struct ethtool_cmd *)&netdev_dev->drvinfo, + ETHTOOL_GDRVINFO, + "ETHTOOL_GDRVINFO"); + if (!error) { + netdev_dev->cache_valid |= VALID_DRVINFO; + } + return error; +} + +static void +netdev_dev_linux_changed(struct netdev_dev_linux *dev, + unsigned int ifi_flags, + unsigned int mask) +{ + dev->change_seq++; + if (!dev->change_seq) { + dev->change_seq++; + } + + if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { + dev->carrier_resets++; + } + dev->ifi_flags = ifi_flags; + + dev->cache_valid &= mask; +} + +static void +netdev_dev_linux_update(struct netdev_dev_linux *dev, + const struct rtnetlink_link_change *change) +{ + if (change->nlmsg_type == RTM_NEWLINK) { + /* Keep drv-info */ + netdev_dev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO); + + /* 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->addr)) { + memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN); + dev->cache_valid |= VALID_ETHERADDR; + dev->ether_addr_error = 0; + } + + dev->ifindex = change->ifi_index; + dev->cache_valid |= VALID_IFINDEX; + dev->get_ifindex_error = 0; + + } else { + netdev_dev_linux_changed(dev, change->ifi_flags, 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 +578,7 @@ 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_update(dev, change); } } } else { @@ -478,37 +588,63 @@ netdev_linux_cache_cb(const struct rtnetlink_change *change, shash_init(&device_shash); netdev_dev_get_devices(&netdev_linux_class, &device_shash); SHASH_FOR_EACH (node, &device_shash) { + unsigned int flags; + dev = node->data; - dev->cache_valid = 0; + + get_flags(&dev->netdev_dev, &flags); + netdev_dev_linux_changed(dev, flags, 0); } shash_destroy(&device_shash); } } -/* Creates the netdev device of 'type' with 'name'. */ 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) +cache_notifier_ref(void) { - struct netdev_dev_linux *netdev_dev; - int error; + if (!cache_notifier_refcount) { + assert(!netdev_linux_cache_notifier); - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for system devices should be empty", name); - } + netdev_linux_cache_notifier = + rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL); - if (!cache_notifier_refcount) { - error = rtnetlink_notifier_register(&netdev_linux_cache_notifier, - netdev_linux_cache_cb, NULL); - if (error) { - return error; + if (!netdev_linux_cache_notifier) { + return EINVAL; } } cache_notifier_refcount++; + return 0; +} + +static void +cache_notifier_unref(void) +{ + assert(cache_notifier_refcount > 0); + if (!--cache_notifier_refcount) { + assert(netdev_linux_cache_notifier); + rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier); + netdev_linux_cache_notifier = NULL; + } +} + +/* Creates system and internal devices. */ +static int +netdev_linux_create(const struct netdev_class *class, const char *name, + struct netdev_dev **netdev_devp) +{ + struct netdev_dev_linux *netdev_dev; + int error; + + error = cache_notifier_ref(); + if (error) { + return error; + } + 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); + get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags); *netdev_devp = &netdev_dev->netdev_dev; return 0; @@ -522,8 +658,7 @@ 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; @@ -531,41 +666,44 @@ 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; + error = cache_notifier_ref(); + if (error) { + goto error; + } + /* Open tap device. */ state->fd = open(tap_dev, O_RDWR); if (state->fd < 0) { error = errno; VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); - goto error; + goto error_unref_notifier; } /* 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)); error = errno; - goto error; + goto error_unref_notifier; } /* Make non-blocking. */ error = set_nonblocking(state->fd); if (error) { - goto error; + goto error_unref_notifier; } netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class); *netdev_devp = &netdev_dev->netdev_dev; return 0; +error_unref_notifier: + cache_notifier_unref(); error: free(netdev_dev); return error; @@ -586,28 +724,22 @@ 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")) { - cache_notifier_refcount--; - - if (!cache_notifier_refcount) { - rtnetlink_notifier_unregister(&netdev_linux_cache_notifier); - } - } else if (!strcmp(type, "tap")) { + if (class == &netdev_tap_class) { destroy_tap(netdev_dev); } - free(netdev_dev); + + cache_notifier_unref(); } 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 +751,19 @@ 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 +775,6 @@ 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; @@ -702,26 +797,59 @@ netdev_linux_close(struct netdev *netdev_) free(netdev); } -/* Initializes 'svec' with a list of the names of all known network devices. */ static int -netdev_linux_enumerate(struct svec *svec) +netdev_linux_listen(struct netdev *netdev_) { - struct if_nameindex *names; - - names = if_nameindex(); - if (names) { - size_t i; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct sockaddr_ll sll; + int ifindex; + int error; + int fd; - for (i = 0; names[i].if_name != NULL; i++) { - svec_add(svec, names[i].if_name); - } - if_freenameindex(names); + if (netdev->fd >= 0) { return 0; - } else { - VLOG_WARN("could not obtain list of network device names: %s", - strerror(errno)); - return errno; } + + /* 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 @@ -730,14 +858,18 @@ 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; } for (;;) { - ssize_t retval = read(netdev->fd, data, size); + ssize_t retval; + + retval = (netdev_->netdev_dev->netdev_class == &netdev_tap_class + ? read(netdev->fd, data, size) + : recv(netdev->fd, data, size, MSG_TRUNC)); if (retval >= 0) { - return retval; + return retval <= size ? retval : -EMSGSIZE; } else if (errno != EINTR) { if (errno != EAGAIN) { VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", @@ -793,15 +925,54 @@ 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 @@ -856,37 +1027,49 @@ netdev_linux_set_etheraddr(struct netdev *netdev_, netdev_dev_linux_cast(netdev_get_dev(netdev_)); int error; - if (!(netdev_dev->cache_valid & VALID_ETHERADDR) - || !eth_addr_equals(netdev_dev->etheraddr, mac)) { - error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac); + if (netdev_dev->cache_valid & VALID_ETHERADDR) { + if (netdev_dev->ether_addr_error) { + return netdev_dev->ether_addr_error; + } + if (eth_addr_equals(netdev_dev->etheraddr, mac)) { + return 0; + } + netdev_dev->cache_valid &= ~VALID_ETHERADDR; + } + + error = set_etheraddr(netdev_get_name(netdev_), mac); + if (!error || error == ENODEV) { + netdev_dev->ether_addr_error = error; + netdev_dev->cache_valid |= VALID_ETHERADDR; if (!error) { - netdev_dev->cache_valid |= VALID_ETHERADDR; memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN); } - } else { - error = 0; } + return error; } -/* Returns a pointer to 'netdev''s MAC address. The caller must not modify or - * free the returned buffer. */ +/* 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]) { struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_)); + if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) { int error = get_etheraddr(netdev_get_name(netdev_), netdev_dev->etheraddr); - if (error) { - return error; - } + + netdev_dev->ether_addr_error = error; netdev_dev->cache_valid |= VALID_ETHERADDR; } - memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); - return 0; + + if (!netdev_dev->ether_addr_error) { + memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); + } + + return netdev_dev->ether_addr_error; } /* Returns the maximum size of transmitted (and received) packets on 'netdev', @@ -903,14 +1086,47 @@ netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); - if (error) { - return error; + + netdev_dev->netdev_mtu_error = error; + netdev_dev->mtu = ifr.ifr_mtu; + netdev_dev->cache_valid |= VALID_MTU; + } + + if (!netdev_dev->netdev_mtu_error) { + *mtup = netdev_dev->mtu; + } + return netdev_dev->netdev_mtu_error; +} + +/* 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; + + if (netdev_dev->cache_valid & VALID_MTU) { + if (netdev_dev->netdev_mtu_error) { + return netdev_dev->netdev_mtu_error; + } + if (netdev_dev->mtu == mtu) { + return 0; } + netdev_dev->cache_valid &= ~VALID_MTU; + } + ifr.ifr_mtu = mtu; + error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, + SIOCSIFMTU, "SIOCSIFMTU"); + if (!error || error == ENODEV) { + netdev_dev->netdev_mtu_error = error; netdev_dev->mtu = ifr.ifr_mtu; netdev_dev->cache_valid |= VALID_MTU; } - *mtup = netdev_dev->mtu; - return 0; + return error; } /* Returns the ifindex of 'netdev', if successful, as a positive number. @@ -929,157 +1145,296 @@ netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) { struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_)); - int error = 0; - char *fn = NULL; - int fd = -1; - - if (!(netdev_dev->cache_valid & VALID_CARRIER)) { - char line[8]; - int retval; - - fn = xasprintf("/sys/class/net/%s/carrier", - netdev_get_name(netdev_)); - fd = open(fn, O_RDONLY); - if (fd < 0) { - error = errno; - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); - goto exit; - } - - retval = read(fd, line, sizeof line); - if (retval < 0) { - error = errno; - if (error == EINVAL) { - /* This is the normal return value when we try to check carrier - * if the network device is not up. */ - } else { - VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error)); - } - goto exit; - } else if (retval == 0) { - error = EPROTO; - VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn); - goto exit; - } - if (line[0] != '0' && line[0] != '1') { - error = EPROTO; - VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)", - fn, line[0]); - goto exit; - } - netdev_dev->carrier = line[0] != '0'; - netdev_dev->cache_valid |= VALID_CARRIER; + if (netdev_dev->miimon_interval > 0) { + *carrier = netdev_dev->miimon; + } else { + *carrier = (netdev_dev->ifi_flags & IFF_RUNNING) != 0; } - *carrier = netdev_dev->carrier; - error = 0; -exit: - if (fd >= 0) { - close(fd); - } - free(fn); - return error; + return 0; } -/* 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) +static long long int +netdev_linux_get_carrier_resets(const struct netdev *netdev) { - /* 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)", strerror(error)); - return false; - } - } + return netdev_dev_linux_cast(netdev_get_dev(netdev))->carrier_resets; } -/* 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) +static int +netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, + struct mii_ioctl_data *data) { - if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) { - const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev); - const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev); - - netdev_dev->is_tap = !strcmp(type, "tap"); - netdev_dev->is_internal = false; - if (!netdev_dev->is_tap) { - struct ethtool_drvinfo drvinfo; - int error; - - memset(&drvinfo, 0, sizeof drvinfo); - error = netdev_linux_do_ethtool(name, - (struct ethtool_cmd *)&drvinfo, - ETHTOOL_GDRVINFO, - "ETHTOOL_GDRVINFO"); + struct ifreq ifr; + int error; - if (!error && !strcmp(drvinfo.driver, "openvswitch")) { - netdev_dev->is_internal = true; - } - } + 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); - netdev_dev->cache_valid |= VALID_IS_PSEUDO; - } + return error; } -static void -swap_uint64(uint64_t *a, uint64_t *b) +static int +netdev_linux_get_miimon(const char *name, bool *miimon) { - *a ^= *b; - *b ^= *a; - *a ^= *b; + 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; } -/* Retrieves current device stats for 'netdev'. */ static int -netdev_linux_get_stats(const struct netdev *netdev_, - struct netdev_stats *stats) +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, dev->ifi_flags, 0); + } + + 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. */ +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)", strerror(error)); + return false; + } + } +} + +static void +swap_uint64(uint64_t *a, uint64_t *b) +{ + uint64_t tmp = *a; + *a = *b; + *b = tmp; +} + +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_)); + + if (!netdev_dev->vport_stats_error || + !(netdev_dev->cache_valid & VALID_VPORT_STAT_ERROR)) { + int error; + + error = netdev_vport_get_stats(netdev_, stats); + if (error) { + VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " + "(%s)", netdev_get_name(netdev_), strerror(error)); + } + netdev_dev->vport_stats_error = error; + netdev_dev->cache_valid |= VALID_VPORT_STAT_ERROR; + } +} + +static int +netdev_linux_sys_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) +{ static int use_netlink_stats = -1; int error; - COVERAGE_INC(netdev_get_stats); + if (use_netlink_stats < 0) { + use_netlink_stats = check_for_working_netlink_stats(); + } + + if (use_netlink_stats) { + int ifindex; - if (netdev_dev->have_vport_stats || - !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) { + 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); + } - error = netdev_vport_get_stats(netdev_, stats); - netdev_dev->have_vport_stats = !error; - netdev_dev->cache_valid |= VALID_HAVE_VPORT_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_, + 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); - if (!netdev_dev->have_vport_stats) { - if (use_netlink_stats < 0) { - use_netlink_stats = check_for_working_netlink_stats(); + error = netdev_linux_sys_get_stats(netdev_, &dev_stats); + + if (error) { + if (netdev_dev->vport_stats_error) { + return error; + } else { + return 0; } - if (use_netlink_stats) { - int ifindex; + } - error = get_ifindex(netdev_, &ifindex); - if (!error) { - error = get_stats_via_netlink(ifindex, stats); - } + if (netdev_dev->vport_stats_error) { + /* 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; +} + +/* Retrieves current device stats for 'netdev-tap' netdev or + * netdev-internal. */ +static int +netdev_tap_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->vport_stats_error) { + return error; } else { - error = get_stats_via_proc(netdev_get_name(netdev_), stats); + return 0; } } @@ -1089,9 +1444,8 @@ 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->vport_stats_error) { + *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); @@ -1107,140 +1461,193 @@ 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', - * '*advertised', '*supported', and '*peer' that are non-null. Each value is a - * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if - * successful, otherwise a positive errno value. */ static int -netdev_linux_get_features(struct netdev *netdev, - uint32_t *current, uint32_t *advertised, - uint32_t *supported, uint32_t *peer) +netdev_internal_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) +{ + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); + + get_stats_via_vport(netdev_, stats); + return netdev_dev->vport_stats_error; +} + +static void +netdev_linux_read_features(struct netdev_dev_linux *netdev_dev) { struct ethtool_cmd ecmd; + uint32_t speed; int error; + if (netdev_dev->cache_valid & VALID_FEATURES) { + return; + } + memset(&ecmd, 0, sizeof ecmd); - error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, + error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); if (error) { - return error; + goto out; } /* Supported features. */ - *supported = 0; + netdev_dev->supported = 0; if (ecmd.supported & SUPPORTED_10baseT_Half) { - *supported |= OFPPF_10MB_HD; + netdev_dev->supported |= NETDEV_F_10MB_HD; } if (ecmd.supported & SUPPORTED_10baseT_Full) { - *supported |= OFPPF_10MB_FD; + netdev_dev->supported |= NETDEV_F_10MB_FD; } if (ecmd.supported & SUPPORTED_100baseT_Half) { - *supported |= OFPPF_100MB_HD; + netdev_dev->supported |= NETDEV_F_100MB_HD; } if (ecmd.supported & SUPPORTED_100baseT_Full) { - *supported |= OFPPF_100MB_FD; + netdev_dev->supported |= NETDEV_F_100MB_FD; } if (ecmd.supported & SUPPORTED_1000baseT_Half) { - *supported |= OFPPF_1GB_HD; + netdev_dev->supported |= NETDEV_F_1GB_HD; } if (ecmd.supported & SUPPORTED_1000baseT_Full) { - *supported |= OFPPF_1GB_FD; + netdev_dev->supported |= NETDEV_F_1GB_FD; } if (ecmd.supported & SUPPORTED_10000baseT_Full) { - *supported |= OFPPF_10GB_FD; + netdev_dev->supported |= NETDEV_F_10GB_FD; } if (ecmd.supported & SUPPORTED_TP) { - *supported |= OFPPF_COPPER; + netdev_dev->supported |= NETDEV_F_COPPER; } if (ecmd.supported & SUPPORTED_FIBRE) { - *supported |= OFPPF_FIBER; + netdev_dev->supported |= NETDEV_F_FIBER; } if (ecmd.supported & SUPPORTED_Autoneg) { - *supported |= OFPPF_AUTONEG; + netdev_dev->supported |= NETDEV_F_AUTONEG; } if (ecmd.supported & SUPPORTED_Pause) { - *supported |= OFPPF_PAUSE; + netdev_dev->supported |= NETDEV_F_PAUSE; } if (ecmd.supported & SUPPORTED_Asym_Pause) { - *supported |= OFPPF_PAUSE_ASYM; + netdev_dev->supported |= NETDEV_F_PAUSE_ASYM; } /* Advertised features. */ - *advertised = 0; + netdev_dev->advertised = 0; if (ecmd.advertising & ADVERTISED_10baseT_Half) { - *advertised |= OFPPF_10MB_HD; + netdev_dev->advertised |= NETDEV_F_10MB_HD; } if (ecmd.advertising & ADVERTISED_10baseT_Full) { - *advertised |= OFPPF_10MB_FD; + netdev_dev->advertised |= NETDEV_F_10MB_FD; } if (ecmd.advertising & ADVERTISED_100baseT_Half) { - *advertised |= OFPPF_100MB_HD; + netdev_dev->advertised |= NETDEV_F_100MB_HD; } if (ecmd.advertising & ADVERTISED_100baseT_Full) { - *advertised |= OFPPF_100MB_FD; + netdev_dev->advertised |= NETDEV_F_100MB_FD; } if (ecmd.advertising & ADVERTISED_1000baseT_Half) { - *advertised |= OFPPF_1GB_HD; + netdev_dev->advertised |= NETDEV_F_1GB_HD; } if (ecmd.advertising & ADVERTISED_1000baseT_Full) { - *advertised |= OFPPF_1GB_FD; + netdev_dev->advertised |= NETDEV_F_1GB_FD; } if (ecmd.advertising & ADVERTISED_10000baseT_Full) { - *advertised |= OFPPF_10GB_FD; + netdev_dev->advertised |= NETDEV_F_10GB_FD; } if (ecmd.advertising & ADVERTISED_TP) { - *advertised |= OFPPF_COPPER; + netdev_dev->advertised |= NETDEV_F_COPPER; } if (ecmd.advertising & ADVERTISED_FIBRE) { - *advertised |= OFPPF_FIBER; + netdev_dev->advertised |= NETDEV_F_FIBER; } if (ecmd.advertising & ADVERTISED_Autoneg) { - *advertised |= OFPPF_AUTONEG; + netdev_dev->advertised |= NETDEV_F_AUTONEG; } if (ecmd.advertising & ADVERTISED_Pause) { - *advertised |= OFPPF_PAUSE; + netdev_dev->advertised |= NETDEV_F_PAUSE; } if (ecmd.advertising & ADVERTISED_Asym_Pause) { - *advertised |= OFPPF_PAUSE_ASYM; + netdev_dev->advertised |= NETDEV_F_PAUSE_ASYM; } /* Current settings. */ - if (ecmd.speed == SPEED_10) { - *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD; - } else if (ecmd.speed == SPEED_100) { - *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD; - } else if (ecmd.speed == SPEED_1000) { - *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD; - } else if (ecmd.speed == SPEED_10000) { - *current = OFPPF_10GB_FD; + speed = ecmd.speed; + if (speed == SPEED_10) { + netdev_dev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; + } else if (speed == SPEED_100) { + netdev_dev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; + } else if (speed == SPEED_1000) { + netdev_dev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; + } else if (speed == SPEED_10000) { + netdev_dev->current = NETDEV_F_10GB_FD; + } else if (speed == 40000) { + netdev_dev->current = NETDEV_F_40GB_FD; + } else if (speed == 100000) { + netdev_dev->current = NETDEV_F_100GB_FD; + } else if (speed == 1000000) { + netdev_dev->current = NETDEV_F_1TB_FD; } else { - *current = 0; + netdev_dev->current = 0; } if (ecmd.port == PORT_TP) { - *current |= OFPPF_COPPER; + netdev_dev->current |= NETDEV_F_COPPER; } else if (ecmd.port == PORT_FIBRE) { - *current |= OFPPF_FIBER; + netdev_dev->current |= NETDEV_F_FIBER; } if (ecmd.autoneg) { - *current |= OFPPF_AUTONEG; + netdev_dev->current |= NETDEV_F_AUTONEG; } /* Peer advertisements. */ - *peer = 0; /* XXX */ + netdev_dev->peer = 0; /* XXX */ - return 0; +out: + netdev_dev->cache_valid |= VALID_FEATURES; + netdev_dev->get_features_error = error; +} + +/* Stores the features supported by 'netdev' into each of '*current', + * '*advertised', '*supported', and '*peer' that are non-null. Each value is a + * bitmap of NETDEV_* bits. Returns 0 if successful, otherwise a positive + * errno value. */ +static int +netdev_linux_get_features(const struct netdev *netdev_, + enum netdev_features *current, + enum netdev_features *advertised, + enum netdev_features *supported, + enum netdev_features *peer) +{ + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev_)); + + netdev_linux_read_features(netdev_dev); + + if (!netdev_dev->get_features_error) { + *current = netdev_dev->current; + *advertised = netdev_dev->advertised; + *supported = netdev_dev->supported; + *peer = netdev_dev->peer; + } + return netdev_dev->get_features_error; } /* Set the features advertised by 'netdev' to 'advertise'. */ static int -netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) +netdev_linux_set_advertisements(struct netdev *netdev, + enum netdev_features advertise) { struct ethtool_cmd ecmd; int error; @@ -1253,144 +1660,48 @@ netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) } ecmd.advertising = 0; - if (advertise & OFPPF_10MB_HD) { + if (advertise & NETDEV_F_10MB_HD) { ecmd.advertising |= ADVERTISED_10baseT_Half; } - if (advertise & OFPPF_10MB_FD) { + if (advertise & NETDEV_F_10MB_FD) { ecmd.advertising |= ADVERTISED_10baseT_Full; } - if (advertise & OFPPF_100MB_HD) { + if (advertise & NETDEV_F_100MB_HD) { ecmd.advertising |= ADVERTISED_100baseT_Half; } - if (advertise & OFPPF_100MB_FD) { + if (advertise & NETDEV_F_100MB_FD) { ecmd.advertising |= ADVERTISED_100baseT_Full; } - if (advertise & OFPPF_1GB_HD) { + if (advertise & NETDEV_F_1GB_HD) { ecmd.advertising |= ADVERTISED_1000baseT_Half; } - if (advertise & OFPPF_1GB_FD) { + if (advertise & NETDEV_F_1GB_FD) { ecmd.advertising |= ADVERTISED_1000baseT_Full; } - if (advertise & OFPPF_10GB_FD) { + if (advertise & NETDEV_F_10GB_FD) { ecmd.advertising |= ADVERTISED_10000baseT_Full; } - if (advertise & OFPPF_COPPER) { + if (advertise & NETDEV_F_COPPER) { ecmd.advertising |= ADVERTISED_TP; } - if (advertise & OFPPF_FIBER) { + if (advertise & NETDEV_F_FIBER) { ecmd.advertising |= ADVERTISED_FIBRE; } - if (advertise & OFPPF_AUTONEG) { + if (advertise & NETDEV_F_AUTONEG) { ecmd.advertising |= ADVERTISED_Autoneg; } - if (advertise & OFPPF_PAUSE) { + if (advertise & NETDEV_F_PAUSE) { ecmd.advertising |= ADVERTISED_Pause; } - if (advertise & OFPPF_PAUSE_ASYM) { + if (advertise & NETDEV_F_PAUSE_ASYM) { ecmd.advertising |= ADVERTISED_Asym_Pause; } return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, ETHTOOL_SSET, "ETHTOOL_SSET"); } -/* If 'netdev_name' is the name of a VLAN network device (e.g. one created with - * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device - * and returns 0. Otherwise returns a errno value (specifically ENOENT if - * 'netdev_name' is the name of a network device that is not a VLAN device) and - * sets '*vlan_vid' to -1. */ -static int -netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) -{ - const char *netdev_name = netdev_get_name(netdev); - struct ds line = DS_EMPTY_INITIALIZER; - FILE *stream = NULL; - int error; - char *fn; - - COVERAGE_INC(netdev_get_vlan_vid); - fn = xasprintf("/proc/net/vlan/%s", netdev_name); - stream = fopen(fn, "r"); - if (!stream) { - error = errno; - goto done; - } - - if (ds_get_line(&line, stream)) { - if (ferror(stream)) { - error = errno; - VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno)); - } else { - error = EPROTO; - VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn); - } - goto done; - } - - if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) { - error = EPROTO; - VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"", - fn, ds_cstr(&line)); - goto done; - } - - error = 0; - -done: - free(fn); - if (stream) { - fclose(stream); - } - ds_destroy(&line); - if (error) { - *vlan_vid = -1; - } - return error; -} - -#define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress" -#define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1" - -/* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a - * positive errno value. - * - * This function is equivalent to running - * /sbin/tc qdisc del dev %s handle ffff: ingress - * but it is much, much faster. - */ -static int -netdev_linux_remove_policing(struct netdev *netdev) -{ - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - const char *netdev_name = netdev_get_name(netdev); - - struct ofpbuf request; - struct tcmsg *tcmsg; - 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"); - nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); - - error = tc_transact(&request, NULL); - if (error && error != ENOENT && error != EINVAL) { - VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", - netdev_name, strerror(error)); - return error; - } - - netdev_dev->kbits_rate = 0; - netdev_dev->kbits_burst = 0; - netdev_dev->cache_valid |= VALID_POLICING; - return 0; -} - -/* Attempts to set input rate limiting (policing) policy. */ +/* Attempts to set input rate limiting (policing) policy. Returns 0 if + * successful, otherwise a positive errno value. */ static int netdev_linux_set_policing(struct netdev *netdev, uint32_t kbits_rate, uint32_t kbits_burst) @@ -1398,55 +1709,72 @@ netdev_linux_set_policing(struct netdev *netdev, struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); const char *netdev_name = netdev_get_name(netdev); - char command[1024]; + int error; - COVERAGE_INC(netdev_set_policing); kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ : kbits_burst); /* Stick with user-specified value. */ - if (netdev_dev->cache_valid & VALID_POLICING - && netdev_dev->kbits_rate == kbits_rate - && netdev_dev->kbits_burst == kbits_burst) { - /* Assume that settings haven't changed since we last set them. */ - return 0; + if (netdev_dev->cache_valid & VALID_POLICING) { + if (netdev_dev->netdev_policing_error) { + return netdev_dev->netdev_policing_error; + } + + if (netdev_dev->kbits_rate == kbits_rate && + netdev_dev->kbits_burst == kbits_burst) { + /* Assume that settings haven't changed since we last set them. */ + return 0; + } + netdev_dev->cache_valid &= ~VALID_POLICING; + } + + COVERAGE_INC(netdev_set_policing); + /* Remove any existing ingress qdisc. */ + error = tc_add_del_ingress_qdisc(netdev, false); + if (error) { + VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", + netdev_name, strerror(error)); + goto out; } - netdev_linux_remove_policing(netdev); if (kbits_rate) { - snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name); - if (system(command) != 0) { - VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name); - return -1; + error = tc_add_del_ingress_qdisc(netdev, true); + if (error) { + VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", + netdev_name, strerror(error)); + goto out; } - snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name, - kbits_rate, kbits_burst); - if (system(command) != 0) { - VLOG_WARN_RL(&rl, "%s: problem configuring policing", - netdev_name); - return -1; + error = tc_add_policer(netdev, kbits_rate, kbits_burst); + if (error){ + VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", + netdev_name, strerror(error)); + goto out; } + } + + netdev_dev->kbits_rate = kbits_rate; + netdev_dev->kbits_burst = kbits_burst; - netdev_dev->kbits_rate = kbits_rate; - netdev_dev->kbits_burst = kbits_burst; +out: + if (!error || error == ENODEV) { + netdev_dev->netdev_policing_error = error; netdev_dev->cache_valid |= VALID_POLICING; } - - return 0; + return error; } 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; @@ -1673,7 +2001,7 @@ netdev_linux_dump_queues(const struct netdev *netdev, { struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); - struct tc_queue *queue; + struct tc_queue *queue, *next_queue; struct shash details; int last_error; int error; @@ -1687,7 +2015,8 @@ netdev_linux_dump_queues(const struct netdev *netdev, last_error = 0; shash_init(&details); - HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) { + HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node, + &netdev_dev->tc->queues) { shash_clear(&details); error = netdev_dev->tc->ops->class_get(netdev, queue, &details); @@ -1853,7 +2182,7 @@ 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, @@ -1873,7 +2202,6 @@ 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)); @@ -1901,7 +2229,7 @@ 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; @@ -1942,26 +2270,50 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, return ENXIO; } -/* 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]) +netdev_linux_get_drv_info(const struct netdev *netdev, struct shash *sh) { - struct arpreq r; + int error; + struct netdev_dev_linux *netdev_dev = + netdev_dev_linux_cast(netdev_get_dev(netdev)); + + error = netdev_linux_get_drvinfo(netdev_dev); + if (!error) { + shash_add(sh, "driver_name", xstrdup(netdev_dev->drvinfo.driver)); + shash_add(sh, "driver_version", xstrdup(netdev_dev->drvinfo.version)); + shash_add(sh, "firmware_version", xstrdup(netdev_dev->drvinfo.fw_version)); + } + return error; +} + +static int +netdev_internal_get_drv_info(const struct netdev *netdev OVS_UNUSED, struct shash *sh) +{ + shash_add(sh, "driver_name", xstrdup("openvswitch")); + return 0; +} + +/* 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, + 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) { @@ -2003,222 +2355,117 @@ static int netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, enum netdev_flags on, enum netdev_flags *old_flagsp) { + struct netdev_dev_linux *netdev_dev; int old_flags, new_flags; - int error; + int error = 0; - error = get_flags(netdev, &old_flags); - if (!error) { - *old_flagsp = iff_to_nd_flags(old_flags); - new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); - if (new_flags != old_flags) { - error = set_flags(netdev, new_flags); - } + netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); + old_flags = netdev_dev->ifi_flags; + *old_flagsp = iff_to_nd_flags(old_flags); + new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); + if (new_flags != old_flags) { + error = set_flags(netdev, new_flags); + get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags); } 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, GET_STATS, SET_STATS, \ + GET_FEATURES, GET_STATUS) \ +{ \ + 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, \ + \ + 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_get_carrier_resets, \ + netdev_linux_set_miimon_interval, \ + GET_STATS, \ + SET_STATS, \ + \ + GET_FEATURES, \ + netdev_linux_set_advertisements, \ + \ + 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, \ + 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_get_stats, + NULL, /* set_stats */ + netdev_linux_get_features, + netdev_linux_get_drv_info); + +const struct netdev_class netdev_tap_class = + NETDEV_LINUX_CLASS( + "tap", + netdev_linux_create_tap, + netdev_tap_get_stats, + NULL, /* set_stats */ + netdev_linux_get_features, + netdev_linux_get_drv_info); + +const struct netdev_class netdev_internal_class = + NETDEV_LINUX_CLASS( + "internal", + netdev_linux_create, + netdev_internal_get_stats, + netdev_vport_set_stats, + NULL, /* get_features */ + netdev_internal_get_drv_info); /* HTB traffic control class. */ @@ -2245,7 +2492,7 @@ 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 = @@ -2257,14 +2504,11 @@ 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) { @@ -2310,7 +2554,12 @@ 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); @@ -2409,7 +2658,7 @@ htb_parse_qdisc_details__(struct netdev *netdev, max_rate_s = shash_find_data(details, "max-rate"); hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; if (!hc->max_rate) { - uint32_t current; + enum netdev_features current; netdev_get_features(netdev, ¤t, NULL, NULL, NULL); hc->max_rate = netdev_features_to_bps(current) / 8; @@ -2428,15 +2677,19 @@ 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 */ @@ -2455,214 +2708,716 @@ 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); - /* priority */ - hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; + /* priority */ + hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; + + return 0; +} + +static int +htb_query_class__(const struct netdev *netdev, unsigned int handle, + unsigned int parent, struct htb_class *options, + struct netdev_queue_stats *stats) +{ + struct ofpbuf *reply; + int error; + + error = tc_query_class(netdev, handle, parent, &reply); + if (!error) { + error = htb_parse_tcmsg__(reply, NULL, options, stats); + ofpbuf_delete(reply); + } + return error; +} + +static int +htb_tc_install(struct netdev *netdev, const struct shash *details) +{ + int error; + + error = htb_setup_qdisc__(netdev); + if (!error) { + struct htb_class hc; + + htb_parse_qdisc_details__(netdev, details, &hc); + error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), + tc_make_handle(1, 0), &hc); + if (!error) { + htb_install__(netdev, hc.max_rate); + } + } + return error; +} + +static struct htb_class * +htb_class_cast__(const struct tc_queue *queue) +{ + return CONTAINER_OF(queue, struct htb_class, tc_queue); +} + +static void +htb_update_queue__(struct netdev *netdev, unsigned int queue_id, + const struct htb_class *hc) +{ + struct htb *htb = htb_get__(netdev); + size_t hash = hash_int(queue_id, 0); + struct tc_queue *queue; + struct htb_class *hcp; + + queue = tc_find_queue__(netdev, queue_id, hash); + if (queue) { + hcp = htb_class_cast__(queue); + } else { + hcp = xmalloc(sizeof *hcp); + queue = &hcp->tc_queue; + queue->queue_id = queue_id; + hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); + } + + hcp->min_rate = hc->min_rate; + hcp->max_rate = hc->max_rate; + hcp->burst = hc->burst; + hcp->priority = hc->priority; +} + +static int +htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) +{ + struct ofpbuf msg; + struct nl_dump dump; + struct htb_class hc; + + /* Get qdisc options. */ + hc.max_rate = 0; + htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); + htb_install__(netdev, hc.max_rate); + + /* Get queues. */ + if (!start_queue_dump(netdev, &dump)) { + return ENODEV; + } + while (nl_dump_next(&dump, &msg)) { + unsigned int queue_id; + + if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { + htb_update_queue__(netdev, queue_id, &hc); + } + } + nl_dump_done(&dump); + + return 0; +} + +static void +htb_tc_destroy(struct tc *tc) +{ + struct htb *htb = CONTAINER_OF(tc, struct htb, tc); + struct htb_class *hc, *next; + + HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) { + hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); + free(hc); + } + tc_destroy(tc); + free(htb); +} + +static int +htb_qdisc_get(const struct netdev *netdev, struct shash *details) +{ + const struct htb *htb = htb_get__(netdev); + shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate)); + return 0; +} + +static int +htb_qdisc_set(struct netdev *netdev, const struct shash *details) +{ + struct htb_class hc; + int error; + + htb_parse_qdisc_details__(netdev, details, &hc); + error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), + tc_make_handle(1, 0), &hc); + if (!error) { + htb_get__(netdev)->max_rate = hc.max_rate; + } + return error; +} + +static int +htb_class_get(const struct netdev *netdev OVS_UNUSED, + const struct tc_queue *queue, struct shash *details) +{ + const struct htb_class *hc = htb_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)); + } + shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst)); + if (hc->priority) { + shash_add(details, "priority", xasprintf("%u", hc->priority)); + } + return 0; +} + +static int +htb_class_set(struct netdev *netdev, unsigned int queue_id, + const struct shash *details) +{ + struct htb_class hc; + int error; + + error = htb_parse_class_details__(netdev, details, &hc); + if (error) { + return error; + } + + error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), + tc_make_handle(1, 0xfffe), &hc); + if (error) { + return error; + } + + htb_update_queue__(netdev, queue_id, &hc); + return 0; +} + +static int +htb_class_delete(struct netdev *netdev, struct tc_queue *queue) +{ + struct htb_class *hc = htb_class_cast__(queue); + struct htb *htb = htb_get__(netdev); + int error; + + error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); + if (!error) { + hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); + free(hc); + } + return error; +} + +static int +htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, + struct netdev_queue_stats *stats) +{ + return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), + tc_make_handle(1, 0xfffe), NULL, stats); +} + +static int +htb_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 <= HTB_N_QUEUES) { + (*cb)(minor - 1, &stats, aux); + } + return 0; +} + +static const struct tc_ops tc_ops_htb = { + "htb", /* linux_name */ + "linux-htb", /* ovs_name */ + HTB_N_QUEUES, /* n_queues */ + htb_tc_install, + htb_tc_load, + htb_tc_destroy, + htb_qdisc_get, + htb_qdisc_set, + htb_class_get, + htb_class_set, + htb_class_delete, + htb_class_get_stats, + 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) { + enum netdev_features 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; - return 0; + 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 -htb_query_class__(const struct netdev *netdev, unsigned int handle, - unsigned int parent, struct htb_class *options, - struct netdev_queue_stats *stats) +hfsc_setup_class__(struct netdev *netdev, unsigned int handle, + unsigned int parent, struct hfsc_class *class) { - struct ofpbuf *reply; int error; + size_t opt_offset; + struct tcmsg *tcmsg; + struct ofpbuf request; + struct tc_service_curve min, max; - error = tc_query_class(netdev, handle, parent, &reply); - if (!error) { - error = htb_parse_tcmsg__(reply, NULL, options, stats); - ofpbuf_delete(reply); + 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 -htb_tc_install(struct netdev *netdev, const struct shash *details) +hfsc_tc_install(struct netdev *netdev, const struct shash *details) { int error; + struct hfsc_class class; - error = htb_setup_qdisc__(netdev); - if (!error) { - struct htb_class hc; + error = hfsc_setup_qdisc__(netdev); - htb_parse_qdisc_details__(netdev, details, &hc); - error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), - tc_make_handle(1, 0), &hc); - if (!error) { - htb_install__(netdev, hc.max_rate); - } + if (error) { + return error; } - return error; -} - -static struct htb_class * -htb_class_cast__(const struct tc_queue *queue) -{ - return CONTAINER_OF(queue, struct htb_class, tc_queue); -} -static void -htb_update_queue__(struct netdev *netdev, unsigned int queue_id, - const struct htb_class *hc) -{ - struct htb *htb = htb_get__(netdev); - size_t hash = hash_int(queue_id, 0); - struct tc_queue *queue; - struct htb_class *hcp; + hfsc_parse_qdisc_details__(netdev, details, &class); + error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), + tc_make_handle(1, 0), &class); - queue = tc_find_queue__(netdev, queue_id, hash); - if (queue) { - hcp = htb_class_cast__(queue); - } else { - hcp = xmalloc(sizeof *hcp); - queue = &hcp->tc_queue; - queue->queue_id = queue_id; - hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); + if (error) { + return error; } - hcp->min_rate = hc->min_rate; - hcp->max_rate = hc->max_rate; - hcp->burst = hc->burst; - hcp->priority = hc->priority; + hfsc_install__(netdev, class.max_rate); + return 0; } static int -htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) +hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) { struct ofpbuf msg; struct nl_dump dump; - struct htb_class hc; - struct htb *htb; + struct hfsc_class hc; - /* 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); + hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); + hfsc_install__(netdev, hc.max_rate); - /* Get queues. */ if (!start_queue_dump(netdev, &dump)) { return ENODEV; } + while (nl_dump_next(&dump, &msg)) { unsigned int queue_id; - if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { - htb_update_queue__(netdev, queue_id, &hc); + if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { + hfsc_update_queue__(netdev, queue_id, &hc); } } - nl_dump_done(&dump); + nl_dump_done(&dump); return 0; } static void -htb_tc_destroy(struct tc *tc) +hfsc_tc_destroy(struct tc *tc) { - struct htb *htb = CONTAINER_OF(tc, struct htb, tc); - struct htb_class *hc, *next; + struct hfsc *hfsc; + struct hfsc_class *hc, *next; - HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) { - hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); + 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(htb); + free(hfsc); } static int -htb_qdisc_get(const struct netdev *netdev, struct shash *details) +hfsc_qdisc_get(const struct netdev *netdev, struct shash *details) { - const struct htb *htb = htb_get__(netdev); - shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate)); + const struct hfsc *hfsc; + hfsc = hfsc_get__(netdev); + shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate)); return 0; } static int -htb_qdisc_set(struct netdev *netdev, const struct shash *details) +hfsc_qdisc_set(struct netdev *netdev, const struct shash *details) { - struct htb_class hc; 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); - htb_parse_qdisc_details__(netdev, details, &hc); - error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), - tc_make_handle(1, 0), &hc); if (!error) { - htb_get__(netdev)->max_rate = hc.max_rate; + hfsc_get__(netdev)->max_rate = class.max_rate; } + return error; } static int -htb_class_get(const struct netdev *netdev OVS_UNUSED, +hfsc_class_get(const struct netdev *netdev OVS_UNUSED, const struct tc_queue *queue, struct shash *details) { - const struct htb_class *hc = htb_class_cast__(queue); + 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)); } - shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst)); - if (hc->priority) { - shash_add(details, "priority", xasprintf("%u", hc->priority)); - } return 0; } static int -htb_class_set(struct netdev *netdev, unsigned int queue_id, - const struct shash *details) +hfsc_class_set(struct netdev *netdev, unsigned int queue_id, + const struct shash *details) { - struct htb_class hc; int error; + struct hfsc_class class; - error = htb_parse_class_details__(netdev, details, &hc); + error = hfsc_parse_class_details__(netdev, details, &class); if (error) { return error; } - error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), - tc_make_handle(1, 0xfffe), &hc); + error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), + tc_make_handle(1, 0xfffe), &class); if (error) { return error; } - htb_update_queue__(netdev, queue_id, &hc); + hfsc_update_queue__(netdev, queue_id, &class); return 0; } static int -htb_class_delete(struct netdev *netdev, struct tc_queue *queue) +hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) { - struct htb_class *hc = htb_class_cast__(queue); - struct htb *htb = htb_get__(netdev); 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(&htb->tc.queues, &hc->tc_queue.hmap_node); + hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); free(hc); } return error; } static int -htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, - struct netdev_queue_stats *stats) +hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, + struct netdev_queue_stats *stats) { - return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), + return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), tc_make_handle(1, 0xfffe), NULL, stats); } static int -htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, - const struct ofpbuf *nlmsg, - netdev_dump_queue_stats_cb *cb, void *aux) +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; @@ -2675,26 +3430,26 @@ htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, major = tc_get_major(handle); minor = tc_get_minor(handle); - if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { + if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { (*cb)(minor - 1, &stats, aux); } return 0; } -static const struct tc_ops tc_ops_htb = { - "htb", /* linux_name */ - "linux-htb", /* ovs_name */ - HTB_N_QUEUES, /* n_queues */ - htb_tc_install, - htb_tc_load, - htb_tc_destroy, - htb_qdisc_get, - htb_qdisc_set, - htb_class_get, - htb_class_set, - htb_class_delete, - htb_class_get_stats, - htb_class_dump_stats +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. @@ -2859,6 +3614,107 @@ tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) return error; } +/* Adds or deletes a root ingress qdisc on 'netdev'. We use this for + * policing configuration. + * + * This function is equivalent to running the following when 'add' is true: + * /sbin/tc qdisc add dev handle ffff: ingress + * + * This function is equivalent to running the following when 'add' is false: + * /sbin/tc qdisc del dev handle ffff: ingress + * + * The configuration and stats may be seen with the following command: + * /sbin/tc -s qdisc show dev + * + * Returns 0 if successful, otherwise a positive errno value. + */ +static int +tc_add_del_ingress_qdisc(struct netdev *netdev, bool add) +{ + struct ofpbuf request; + struct tcmsg *tcmsg; + int error; + int type = add ? RTM_NEWQDISC : RTM_DELQDISC; + int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0; + + tcmsg = tc_make_request(netdev, type, flags, &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"); + nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); + + error = tc_transact(&request, NULL); + if (error) { + /* If we're deleting the qdisc, don't worry about some of the + * error conditions. */ + if (!add && (error == ENOENT || error == EINVAL)) { + return 0; + } + return error; + } + + return 0; +} + +/* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size + * of 'kbits_burst'. + * + * This function is equivalent to running: + * /sbin/tc filter add dev parent ffff: protocol all prio 49 + * basic police rate kbit burst k + * mtu 65535 drop + * + * The configuration and stats may be seen with the following command: + * /sbin/tc -s filter show eth0 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) +{ + struct tc_police tc_police; + struct ofpbuf request; + struct tcmsg *tcmsg; + size_t basic_offset; + size_t police_offset; + int error; + int mtu = 65535; + + 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/8 * 1000, mtu); + tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate, + kbits_burst * 1024); + + tcmsg = tc_make_request(netdev, RTM_NEWTFILTER, + NLM_F_EXCL | NLM_F_CREATE, &request); + if (!tcmsg) { + return ENODEV; + } + tcmsg->tcm_parent = tc_make_handle(0xffff, 0); + tcmsg->tcm_info = tc_make_handle(49, + (OVS_FORCE uint16_t) htons(ETH_P_ALL)); + + nl_msg_put_string(&request, TCA_KIND, "basic"); + basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); + police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); + nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); + tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); + nl_msg_end_nested(&request, police_offset); + nl_msg_end_nested(&request, basic_offset); + + error = tc_transact(&request, NULL); + if (error) { + return error; + } + + return 0; +} + static void read_psched(void) { @@ -3318,10 +4174,92 @@ 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)); } + +/* Linux-only functions declared in netdev-linux.h */ + +/* Returns a fd for an AF_INET socket or a negative errno value. */ +int +netdev_linux_get_af_inet_sock(void) +{ + int error = netdev_linux_init(); + return error ? -error : af_inet_sock; +} + +/* 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; +} /* Utility functions. */ +/* 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; +} + static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats) { @@ -3338,7 +4276,6 @@ 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; @@ -3366,28 +4303,7 @@ 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); @@ -3452,19 +4368,22 @@ get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) } static int -get_flags(const struct netdev *netdev, int *flags) +get_flags(const struct netdev_dev *dev, unsigned int *flags) { struct ifreq ifr; int error; - error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS, + *flags = 0; + error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); - *flags = ifr.ifr_flags; + if (!error) { + *flags = ifr.ifr_flags; + } return error; } static int -set_flags(struct netdev *netdev, int flags) +set_flags(struct netdev *netdev, unsigned int flags) { struct ifreq ifr; @@ -3478,7 +4397,7 @@ 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", @@ -3493,17 +4412,22 @@ get_ifindex(const struct netdev *netdev_, int *ifindexp) { struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_)); - *ifindexp = 0; + if (!(netdev_dev->cache_valid & VALID_IFINDEX)) { int ifindex = do_get_ifindex(netdev_get_name(netdev_)); + if (ifindex < 0) { - return -ifindex; + netdev_dev->get_ifindex_error = -ifindex; + netdev_dev->ifindex = 0; + } else { + netdev_dev->get_ifindex_error = 0; + netdev_dev->ifindex = ifindex; } netdev_dev->cache_valid |= VALID_IFINDEX; - netdev_dev->ifindex = ifindex; } + *ifindexp = netdev_dev->ifindex; - return 0; + return netdev_dev->get_ifindex_error; } static int @@ -3513,11 +4437,15 @@ 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; @@ -3530,14 +4458,14 @@ get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) } static int -set_etheraddr(const char *netdev_name, int hwaddr_family, +set_etheraddr(const char *netdev_name, const uint8_t mac[ETH_ADDR_LEN]) { struct ifreq ifr; memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - ifr.ifr_hwaddr.sa_family = hwaddr_family; + 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); COVERAGE_INC(netdev_set_hwaddr); if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { @@ -3555,7 +4483,7 @@ 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; @@ -3578,7 +4506,7 @@ 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)); @@ -3602,3 +4530,22 @@ 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; +}