X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=9306100670d38015e374153ba2ad76a08d46b776;hb=259e0b1ad1bfea762a76f0098deb8f8d8db1dfa3;hp=1c5ddc53bf311baa321c03f08d05e659b9568e41;hpb=bd3932826b3d9153f6d026adaad2439ac23f2fcd;p=sliver-openvswitch.git diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 1c5ddc53b..930610067 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -18,17 +18,18 @@ #include "netdev-linux.h" -#include #include #include #include #include +#include #include +#include #include -#include #include #include #include +#include #include #include #include @@ -37,9 +38,7 @@ #include #include #include -#include #include -#include #include #include #include @@ -57,28 +56,31 @@ #include "hmap.h" #include "netdev-provider.h" #include "netdev-vport.h" -#include "netlink.h" +#include "netlink-notifier.h" #include "netlink-socket.h" +#include "netlink.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 "socket-util.h" #include "sset.h" +#include "timer.h" +#include "unaligned.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(netdev_linux); -COVERAGE_DEFINE(netdev_get_vlan_vid); COVERAGE_DEFINE(netdev_set_policing); COVERAGE_DEFINE(netdev_arp_lookup); COVERAGE_DEFINE(netdev_get_ifindex); COVERAGE_DEFINE(netdev_get_hwaddr); COVERAGE_DEFINE(netdev_set_hwaddr); -COVERAGE_DEFINE(netdev_ethtool); +COVERAGE_DEFINE(netdev_get_ethtool); +COVERAGE_DEFINE(netdev_set_ethtool); + /* These were introduced in Linux 2.6.14, so they might be missing if we have * old headers. */ @@ -89,13 +91,22 @@ COVERAGE_DEFINE(netdev_ethtool); #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 { @@ -104,15 +115,10 @@ 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 -}; - -struct tap_state { - int fd; - bool opened; + VALID_POLICING = 1 << 5, + VALID_VPORT_STAT_ERROR = 1 << 6, + VALID_DRVINFO = 1 << 7, + VALID_FEATURES = 1 << 8, }; /* Traffic control. */ @@ -129,6 +135,8 @@ struct tc { * Written only by TC implementation. */ }; +#define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) } + /* One traffic control queue. * * Each TC implementation subclasses this with whatever additional data it @@ -136,6 +144,7 @@ struct tc { struct tc_queue { struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ unsigned int queue_id; /* OpenFlow queue ID. */ + long long int created; /* Time queue was created, in msecs. */ }; /* A particular kind of traffic control. Each implementation generally maps to @@ -172,7 +181,7 @@ struct tc_ops { * * (This function is null for tc_ops_other, which cannot be installed. For * other TC classes it should always be nonnull.) */ - int (*tc_install)(struct netdev *netdev, const struct shash *details); + int (*tc_install)(struct netdev *netdev, const struct smap *details); /* Called when the netdev code determines (through a Netlink query) that * this TC class's qdisc is installed on 'netdev', but we didn't install @@ -212,7 +221,7 @@ struct tc_ops { * * This function may be null if 'tc' is not configurable. */ - int (*qdisc_get)(const struct netdev *netdev, struct shash *details); + int (*qdisc_get)(const struct netdev *netdev, struct smap *details); /* Reconfigures 'netdev->tc' according to 'details', performing any * required Netlink calls to complete the reconfiguration. @@ -223,7 +232,7 @@ struct tc_ops { * * This function may be null if 'tc' is not configurable. */ - int (*qdisc_set)(struct netdev *, const struct shash *details); + int (*qdisc_set)(struct netdev *, const struct smap *details); /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is * one of the 'struct tc_queue's within 'netdev->tc->queues'. @@ -239,7 +248,7 @@ struct tc_ops { * This function may be null if 'tc' does not have queues ('n_queues' is * 0). */ int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, - struct shash *details); + struct smap *details); /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to * 'details', perfoming any required Netlink calls to complete the @@ -253,7 +262,7 @@ struct tc_ops { * This function may be null if 'tc' does not have queues or its queues are * not configurable. */ int (*class_set)(struct netdev *, unsigned int queue_id, - const struct shash *details); + const struct smap *details); /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct * tc_queue's within 'netdev->tc->queues'. @@ -301,7 +310,7 @@ 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[] = { +static const struct tc_ops *const tcs[] = { &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ &tc_ops_hfsc, /* Hierarchical fair service curve. */ &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ @@ -320,6 +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); @@ -340,11 +352,15 @@ static void tc_put_rtab(struct ofpbuf *, uint16_t type, const struct tc_ratespec *rate); static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); -struct netdev_dev_linux { - struct netdev_dev netdev_dev; +struct netdev_linux { + struct netdev up; - 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. */ @@ -353,193 +369,241 @@ 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. */ + + struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ struct tc *tc; - union { - struct tap_state tap; - } state; + /* For devices of class netdev_tap_class only. */ + int tap_fd; }; -struct netdev_linux { - struct netdev netdev; +struct netdev_rx_linux { + struct netdev_rx up; + bool is_tap; int fd; }; -/* Sockets used for ioctl operations. */ -static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */ -static int af_packet_sock = -1; /* AF_PACKET, SOCK_RAW. */ - -/* 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; +static const struct netdev_rx_class netdev_rx_linux_class; /* 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); -static int netdev_linux_init(void); +static void netdev_linux_run(void); static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, int cmd, const char *cmd_name); -static int netdev_linux_do_ioctl(const 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 *, unsigned int *flags); +static int set_flags(const char *, 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) { - return netdev_class->init == netdev_linux_init; + return netdev_class->run == netdev_linux_run; } -static struct netdev_dev_linux * -netdev_dev_linux_cast(const struct netdev_dev *netdev_dev) +static bool +is_tap_netdev(const struct netdev *netdev) { - const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); - assert(is_netdev_linux_class(netdev_class)); - - return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev); + return netdev_get_class(netdev) == &netdev_tap_class; } static struct netdev_linux * netdev_linux_cast(const struct netdev *netdev) { - struct netdev_dev *netdev_dev = netdev_get_dev(netdev); - const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); - assert(is_netdev_linux_class(netdev_class)); + ovs_assert(is_netdev_linux_class(netdev_get_class(netdev))); - return CONTAINER_OF(netdev, struct netdev_linux, netdev); + return CONTAINER_OF(netdev, struct netdev_linux, up); } - -static int -netdev_linux_init(void) -{ - static int status = -1; - if (status < 0) { - /* Create AF_INET socket. */ - af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); - status = af_inet_sock >= 0 ? 0 : errno; - if (status) { - VLOG_ERR("failed to create inet socket: %s", strerror(status)); - } else { - /* Create AF_PACKET socket. */ - af_packet_sock = socket(AF_PACKET, SOCK_RAW, 0); - status = af_packet_sock >= 0 ? 0 : errno; - if (!status) { - set_nonblocking(af_packet_sock); - } else { - VLOG_ERR("failed to create packet socket: %s", - strerror(status)); - } - } - /* Create rtnetlink socket. */ - if (!status) { - status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); - if (status) { - VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", - strerror(status)); - } - } - } - return status; +static struct netdev_rx_linux * +netdev_rx_linux_cast(const struct netdev_rx *rx) +{ + netdev_rx_assert_class(rx, &netdev_rx_linux_class); + return CONTAINER_OF(rx, struct netdev_rx_linux, up); } - + static void netdev_linux_run(void) { - rtnetlink_link_notifier_run(); + rtnetlink_link_run(); + netdev_linux_miimon_run(); } static void netdev_linux_wait(void) { - rtnetlink_link_notifier_wait(); + rtnetlink_link_wait(); + netdev_linux_miimon_wait(); +} + +static void +netdev_linux_changed(struct netdev_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_linux_update(struct netdev_linux *dev, + const struct rtnetlink_link_change *change) +{ + if (change->nlmsg_type == RTM_NEWLINK) { + /* Keep drv-info */ + netdev_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_linux_changed(dev, change->ifi_flags, 0); + } } static void netdev_linux_cache_cb(const struct rtnetlink_link_change *change, void *aux OVS_UNUSED) { - struct netdev_dev_linux *dev; if (change) { - struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname); - if (base_dev) { - const struct netdev_class *netdev_class = - netdev_dev_get_class(base_dev); - - if (is_netdev_linux_class(netdev_class)) { - dev = netdev_dev_linux_cast(base_dev); - dev->cache_valid = 0; - } + struct netdev *base_dev = netdev_from_name(change->ifname); + if (base_dev && is_netdev_linux_class(netdev_get_class(base_dev))) { + netdev_linux_update(netdev_linux_cast(base_dev), change); + netdev_close(base_dev); } } else { struct shash device_shash; struct shash_node *node; shash_init(&device_shash); - netdev_dev_get_devices(&netdev_linux_class, &device_shash); + netdev_get_devices(&netdev_linux_class, &device_shash); SHASH_FOR_EACH (node, &device_shash) { - dev = node->data; - dev->cache_valid = 0; + struct netdev *netdev = node->data; + struct netdev_linux *dev = netdev_linux_cast(netdev); + unsigned int flags; + + get_flags(&dev->up, &flags); + netdev_linux_changed(dev, flags, 0); + netdev_close(netdev); } shash_destroy(&device_shash); } } +static int +cache_notifier_ref(void) +{ + if (!cache_notifier_refcount) { + ovs_assert(!netdev_linux_cache_notifier); + + netdev_linux_cache_notifier = + rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL); + + if (!netdev_linux_cache_notifier) { + return EINVAL; + } + } + cache_notifier_refcount++; + + return 0; +} + +static void +cache_notifier_unref(void) +{ + ovs_assert(cache_notifier_refcount > 0); + if (!--cache_notifier_refcount) { + ovs_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, const struct shash *args, - struct netdev_dev **netdev_devp) +netdev_linux_create(const struct netdev_class *class, const char *name, + struct netdev **netdevp) { - struct netdev_dev_linux *netdev_dev; + struct netdev_linux *netdev; int error; - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for %s devices should be empty", - name, class->type); + error = cache_notifier_ref(); + if (error) { + return error; } - if (!cache_notifier_refcount) { - error = rtnetlink_link_notifier_register(&netdev_linux_cache_notifier, - netdev_linux_cache_cb, NULL); - if (error) { - return error; + netdev = xzalloc(sizeof *netdev); + netdev->change_seq = 1; + netdev_init(&netdev->up, name, class); + error = get_flags(&netdev->up, &netdev->ifi_flags); + if (error == ENODEV) { + if (class != &netdev_internal_class) { + /* The device does not exist, so don't allow it to be opened. */ + netdev_uninit(&netdev->up, false); + cache_notifier_unref(); + free(netdev); + return ENODEV; + } else { + /* "Internal" netdevs 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(). + * Therefore, ignore the error. */ } } - cache_notifier_refcount++; - netdev_dev = xzalloc(sizeof *netdev_dev); - netdev_dev_init(&netdev_dev->netdev_dev, name, args, class); - - *netdev_devp = &netdev_dev->netdev_dev; + *netdevp = &netdev->up; return 0; } @@ -551,152 +615,118 @@ netdev_linux_create(const struct netdev_class *class, * 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 **netdevp) { - struct netdev_dev_linux *netdev_dev; - struct tap_state *state; + struct netdev_linux *netdev; static const char tap_dev[] = "/dev/net/tun"; struct ifreq ifr; int error; - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for TAP devices should be empty", name); - } + netdev = xzalloc(sizeof *netdev); + netdev->change_seq = 1; - 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) { + netdev->tap_fd = open(tap_dev, O_RDWR); + if (netdev->tap_fd < 0) { error = errno; - VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); - goto error; + VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); + goto error_unref_notifier; } /* Create tap device. */ ifr.ifr_flags = IFF_TAP | IFF_NO_PI; ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { + if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) { VLOG_WARN("%s: creating tap device failed: %s", name, - strerror(errno)); + ovs_strerror(errno)); error = errno; - goto error; + goto error_close; } /* Make non-blocking. */ - error = set_nonblocking(state->fd); + error = set_nonblocking(netdev->tap_fd); if (error) { - goto error; + goto error_close; } - netdev_dev_init(&netdev_dev->netdev_dev, name, args, &netdev_tap_class); - *netdev_devp = &netdev_dev->netdev_dev; + netdev_init(&netdev->up, name, &netdev_tap_class); + *netdevp = &netdev->up; return 0; +error_close: + close(netdev->tap_fd); +error_unref_notifier: + cache_notifier_unref(); error: - free(netdev_dev); + free(netdev); return error; } static void -destroy_tap(struct netdev_dev_linux *netdev_dev) +netdev_linux_destroy(struct netdev *netdev_) { - struct tap_state *state = &netdev_dev->state.tap; - - if (state->fd >= 0) { - close(state->fd); - } -} - -/* Destroys the netdev device 'netdev_dev_'. */ -static void -netdev_linux_destroy(struct netdev_dev *netdev_dev_) -{ - struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); - const struct netdev_class *class = netdev_dev_get_class(netdev_dev_); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) { - netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); + if (netdev->tc && netdev->tc->ops->tc_destroy) { + netdev->tc->ops->tc_destroy(netdev->tc); } - if (class == &netdev_linux_class || class == &netdev_internal_class) { - cache_notifier_refcount--; - - if (!cache_notifier_refcount) { - rtnetlink_link_notifier_unregister(&netdev_linux_cache_notifier); - } - } else if (class == &netdev_tap_class) { - destroy_tap(netdev_dev); - } else { - NOT_REACHED(); + if (netdev_get_class(netdev_) == &netdev_tap_class + && netdev->tap_fd >= 0) + { + close(netdev->tap_fd); } + free(netdev); - free(netdev_dev); + cache_notifier_unref(); } static int -netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, - struct netdev **netdevp) +netdev_linux_rx_open(struct netdev *netdev_, struct netdev_rx **rxp) { - struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); - struct netdev_linux *netdev; - enum netdev_flags flags; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + bool is_tap = is_tap_netdev(netdev_); + struct netdev_rx_linux *rx; int error; + int fd; - /* Allocate network device. */ - netdev = xzalloc(sizeof *netdev); - netdev->fd = -1; - netdev_init(&netdev->netdev, netdev_dev_); - - /* 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") && - !netdev_dev->state.tap.opened) { - - /* We assume that the first user of the tap device is the primary user - * and give them the tap FD. Subsequent users probably just expect - * this to be a system device so open it normally to avoid send/receive - * 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) { + if (is_tap) { + fd = netdev->tap_fd; + } else { struct sockaddr_ll sll; - int protocol; int ifindex; + /* Result of tcpdump -dd inbound */ + static const struct sock_filter filt[] = { + { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ + { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ + { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ + { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ + }; + static const struct sock_fprog fprog = { + ARRAY_SIZE(filt), (struct sock_filter *) filt + }; /* 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) { + fd = socket(PF_PACKET, SOCK_RAW, 0); + if (fd < 0) { error = errno; + VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); goto error; } /* Set non-blocking mode. */ - error = set_nonblocking(netdev->fd); + error = set_nonblocking(fd); if (error) { goto error; } /* Get ethernet device index. */ - error = get_ifindex(&netdev->netdev, &ifindex); + error = get_ifindex(&netdev->up, &ifindex); if (error) { goto error; } @@ -705,119 +735,96 @@ netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, memset(&sll, 0, sizeof sll); sll.sll_family = AF_PACKET; sll.sll_ifindex = ifindex; - if (bind(netdev->fd, - (struct sockaddr *) &sll, sizeof sll) < 0) { + 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("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_), - strerror(error)); + VLOG_ERR("%s: failed to bind raw socket (%s)", + netdev_get_name(netdev_), ovs_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); + /* Filter for only inbound packets. */ + error = setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, + sizeof fprog); if (error) { + error = errno; + VLOG_ERR("%s: failed to attach filter (%s)", + netdev_get_name(netdev_), ovs_strerror(error)); goto error; } } - *netdevp = &netdev->netdev; + rx = xmalloc(sizeof *rx); + netdev_rx_init(&rx->up, netdev_, &netdev_rx_linux_class); + rx->is_tap = is_tap; + rx->fd = fd; + + *rxp = &rx->up; return 0; error: - netdev_uninit(&netdev->netdev, true); + if (fd >= 0) { + close(fd); + } return error; } -/* Closes and destroys 'netdev'. */ static void -netdev_linux_close(struct netdev *netdev_) +netdev_rx_linux_destroy(struct netdev_rx *rx_) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); - if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) { - close(netdev->fd); + if (!rx->is_tap) { + close(rx->fd); } - free(netdev); + free(rx); } -/* Initializes 'sset' with a list of the names of all known network devices. */ static int -netdev_linux_enumerate(struct sset *sset) +netdev_rx_linux_recv(struct netdev_rx *rx_, void *data, size_t size) { - struct if_nameindex *names; + struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); + ssize_t retval; - names = if_nameindex(); - if (names) { - size_t i; + do { + retval = (rx->is_tap + ? read(rx->fd, data, size) + : recv(rx->fd, data, size, MSG_TRUNC)); + } while (retval < 0 && errno == EINTR); - for (i = 0; names[i].if_name != NULL; i++) { - sset_add(sset, names[i].if_name); - } - if_freenameindex(names); - return 0; + if (retval >= 0) { + return retval > size ? -EMSGSIZE : retval; } else { - VLOG_WARN("could not obtain list of network device names: %s", - strerror(errno)); - return errno; - } -} - -static int -netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) -{ - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - - if (netdev->fd < 0) { - /* Device was opened with NETDEV_ETH_TYPE_NONE. */ - return -EAGAIN; - } - - for (;;) { - ssize_t retval = read(netdev->fd, data, size); - if (retval >= 0) { - return retval; - } else if (errno != EINTR) { - if (errno != EAGAIN) { - VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", - strerror(errno), netdev_get_name(netdev_)); - } - return -errno; + if (errno != EAGAIN) { + VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", + ovs_strerror(errno), netdev_rx_get_name(rx_)); } + return -errno; } } -/* Registers with the poll loop to wake up from the next call to poll_block() - * when a packet is ready to be received with netdev_recv() on 'netdev'. */ static void -netdev_linux_recv_wait(struct netdev *netdev_) +netdev_rx_linux_wait(struct netdev_rx *rx_) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->fd >= 0) { - poll_fd_wait(netdev->fd, POLLIN); - } + struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); + poll_fd_wait(rx->fd, POLLIN); } -/* Discards all packets waiting to be received from 'netdev'. */ static int -netdev_linux_drain(struct netdev *netdev_) +netdev_rx_linux_drain(struct netdev_rx *rx_) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->fd < 0) { - return 0; - } else if (!strcmp(netdev_get_type(netdev_), "tap")) { + struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); + if (rx->is_tap) { struct ifreq ifr; - int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, - SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); + int error = af_inet_ifreq_ioctl(netdev_rx_get_name(rx_), &ifr, + SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); if (error) { return error; } - drain_fd(netdev->fd, ifr.ifr_qlen); + drain_fd(rx->fd, ifr.ifr_qlen); return 0; } else { - return drain_rcvbuf(netdev->fd); + return drain_rcvbuf(rx->fd); } } @@ -833,17 +840,22 @@ netdev_linux_drain(struct netdev *netdev_) 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; - if (netdev->fd < 0) { + if (!is_tap_netdev(netdev_)) { /* 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) { @@ -856,7 +868,7 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) sll.sll_family = AF_PACKET; sll.sll_ifindex = ifindex; - iov.iov_base = (void *) data; + iov.iov_base = CONST_CAST(void *, data); iov.iov_len = size; msg.msg_name = &sll; @@ -867,13 +879,16 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) msg.msg_controllen = 0; msg.msg_flags = 0; - retval = sendmsg(af_packet_sock, &msg, 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); + /* Use the tap 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. This doesn't occur on other interface types + * because we attach a socket filter to the rx socket. */ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + retval = write(netdev->tap_fd, data, size); } if (retval < 0) { @@ -886,7 +901,7 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) continue; } else if (errno != EAGAIN) { VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", - netdev_get_name(netdev_), strerror(errno)); + netdev_get_name(netdev_), ovs_strerror(errno)); } return errno; } else if (retval != size) { @@ -907,14 +922,9 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) * expected to do additional queuing of packets. Thus, this function is * unlikely to ever be used. It is included for completeness. */ static void -netdev_linux_send_wait(struct netdev *netdev_) +netdev_linux_send_wait(struct netdev *netdev) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (netdev->fd < 0) { - /* Nothing to do. */ - } else if (strcmp(netdev_get_type(netdev_), "tap")) { - poll_fd_wait(netdev->fd, POLLOUT); - } else { + if (is_tap_netdev(netdev)) { /* TAP device always accepts packets.*/ poll_immediate_wake(); } @@ -926,41 +936,58 @@ static int netdev_linux_set_etheraddr(struct netdev *netdev_, const uint8_t mac[ETH_ADDR_LEN]) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_saved_flags *sf = NULL; 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->cache_valid & VALID_ETHERADDR) { + if (netdev->ether_addr_error) { + return netdev->ether_addr_error; + } + if (eth_addr_equals(netdev->etheraddr, mac)) { + return 0; + } + netdev->cache_valid &= ~VALID_ETHERADDR; + } + + /* Tap devices must be brought down before setting the address. */ + if (is_tap_netdev(netdev_)) { + netdev_turn_flags_off(netdev_, NETDEV_UP, &sf); + } + error = set_etheraddr(netdev_get_name(netdev_), mac); + if (!error || error == ENODEV) { + netdev->ether_addr_error = error; + netdev->cache_valid |= VALID_ETHERADDR; if (!error) { - netdev_dev->cache_valid |= VALID_ETHERADDR; - memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN); + memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN); } - } else { - error = 0; } + + netdev_restore_flags(sf); + 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)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (!(netdev->cache_valid & VALID_ETHERADDR)) { int error = get_etheraddr(netdev_get_name(netdev_), - netdev_dev->etheraddr); - if (error) { - return error; - } - netdev_dev->cache_valid |= VALID_ETHERADDR; + netdev->etheraddr); + + netdev->ether_addr_error = error; + netdev->cache_valid |= VALID_ETHERADDR; } - memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); - return 0; + + if (!netdev->ether_addr_error) { + memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN); + } + + return netdev->ether_addr_error; } /* Returns the maximum size of transmitted (and received) packets on 'netdev', @@ -969,22 +996,53 @@ netdev_linux_get_etheraddr(const struct netdev *netdev_, static int netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - if (!(netdev_dev->cache_valid & VALID_MTU)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (!(netdev->cache_valid & VALID_MTU)) { struct ifreq ifr; int error; - error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, - SIOCGIFMTU, "SIOCGIFMTU"); - if (error) { - return error; + error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, + SIOCGIFMTU, "SIOCGIFMTU"); + + netdev->netdev_mtu_error = error; + netdev->mtu = ifr.ifr_mtu; + netdev->cache_valid |= VALID_MTU; + } + + if (!netdev->netdev_mtu_error) { + *mtup = netdev->mtu; + } + return netdev->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_linux *netdev = netdev_linux_cast(netdev_); + struct ifreq ifr; + int error; + + if (netdev->cache_valid & VALID_MTU) { + if (netdev->netdev_mtu_error) { + return netdev->netdev_mtu_error; } - netdev_dev->mtu = ifr.ifr_mtu; - netdev_dev->cache_valid |= VALID_MTU; + if (netdev->mtu == mtu) { + return 0; + } + netdev->cache_valid &= ~VALID_MTU; } - *mtup = netdev_dev->mtu; - return 0; + ifr.ifr_mtu = mtu; + error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, + SIOCSIFMTU, "SIOCSIFMTU"); + if (!error || error == ENODEV) { + netdev->netdev_mtu_error = error; + netdev->mtu = ifr.ifr_mtu; + netdev->cache_valid |= VALID_MTU; + } + return error; } /* Returns the ifindex of 'netdev', if successful, as a positive number. @@ -1001,92 +1059,52 @@ netdev_linux_get_ifindex(const struct netdev *netdev) static int 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; - } + struct netdev_linux *netdev = netdev_linux_cast(netdev_); - 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->miimon_interval > 0) { + *carrier = netdev->miimon; + } else { + *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; } - *carrier = netdev_dev->carrier; - error = 0; -exit: - if (fd >= 0) { - close(fd); - } - free(fn); - return error; + return 0; +} + +static long long int +netdev_linux_get_carrier_resets(const struct netdev *netdev) +{ + return netdev_linux_cast(netdev)->carrier_resets; } static int -netdev_linux_do_miimon(const struct netdev *netdev, int cmd, - const char *cmd_name, struct mii_ioctl_data *data) +netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, + struct mii_ioctl_data *data) { struct ifreq ifr; int error; memset(&ifr, 0, sizeof ifr); memcpy(&ifr.ifr_data, data, sizeof *data); - error = netdev_linux_do_ioctl(netdev_get_name(netdev), - &ifr, cmd, cmd_name); + error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name); memcpy(data, &ifr.ifr_data, sizeof *data); return error; } static int -netdev_linux_get_miimon(const struct netdev *netdev, bool *miimon) +netdev_linux_get_miimon(const char *name, bool *miimon) { - const char *name = netdev_get_name(netdev); struct mii_ioctl_data data; int error; *miimon = false; memset(&data, 0, sizeof data); - error = netdev_linux_do_miimon(netdev, SIOCGMIIPHY, "SIOCGMIIPHY", &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(netdev, SIOCGMIIREG, "SIOCGMIIREG", + error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", &data); if (!error) { @@ -1100,6 +1118,7 @@ netdev_linux_get_miimon(const struct netdev *netdev, bool *miimon) VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", name); + COVERAGE_INC(netdev_get_ethtool); memset(&ecmd, 0, sizeof ecmd); error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, "ETHTOOL_GLINK"); @@ -1116,6 +1135,72 @@ netdev_linux_get_miimon(const struct netdev *netdev, bool *miimon) return error; } +static int +netdev_linux_set_miimon_interval(struct netdev *netdev_, + long long int interval) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + interval = interval > 0 ? MAX(interval, 100) : 0; + if (netdev->miimon_interval != interval) { + netdev->miimon_interval = interval; + timer_set_expired(&netdev->miimon_timer); + } + + return 0; +} + +static void +netdev_linux_miimon_run(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev *netdev = node->data; + struct netdev_linux *dev = netdev_linux_cast(netdev); + bool miimon; + + if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) { + netdev_close(netdev); + continue; + } + + netdev_linux_get_miimon(dev->up.name, &miimon); + if (miimon != dev->miimon) { + dev->miimon = miimon; + netdev_linux_changed(dev, dev->ifi_flags, 0); + } + + timer_set_duration(&dev->miimon_timer, dev->miimon_interval); + netdev_close(netdev); + } + + shash_destroy(&device_shash); +} + +static void +netdev_linux_miimon_wait(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev *netdev = node->data; + struct netdev_linux *dev = netdev_linux_cast(netdev); + + if (dev->miimon_interval > 0) { + timer_wait(&dev->miimon_timer); + } + netdev_close(netdev); + } + 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. */ @@ -1138,66 +1223,188 @@ check_for_working_netlink_stats(void) } else { VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " "via proc (you are probably running a pre-2.6.19 " - "kernel)", strerror(error)); + "kernel)", ovs_strerror(error)); return false; } } } -/* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */ static void -netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev) +swap_uint64(uint64_t *a, uint64_t *b) { - if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) { - const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev); - const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev); + uint64_t tmp = *a; + *a = *b; + *b = tmp; +} + +/* Copies 'src' into 'dst', performing format conversion in the process. + * + * 'src' is allowed to be misaligned. */ +static void +netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, + const struct ovs_vport_stats *src) +{ + dst->rx_packets = get_unaligned_u64(&src->rx_packets); + dst->tx_packets = get_unaligned_u64(&src->tx_packets); + dst->rx_bytes = get_unaligned_u64(&src->rx_bytes); + dst->tx_bytes = get_unaligned_u64(&src->tx_bytes); + dst->rx_errors = get_unaligned_u64(&src->rx_errors); + dst->tx_errors = get_unaligned_u64(&src->tx_errors); + dst->rx_dropped = get_unaligned_u64(&src->rx_dropped); + dst->tx_dropped = get_unaligned_u64(&src->tx_dropped); + dst->multicast = 0; + dst->collisions = 0; + dst->rx_length_errors = 0; + dst->rx_over_errors = 0; + dst->rx_crc_errors = 0; + dst->rx_frame_errors = 0; + dst->rx_fifo_errors = 0; + dst->rx_missed_errors = 0; + dst->tx_aborted_errors = 0; + dst->tx_carrier_errors = 0; + dst->tx_fifo_errors = 0; + dst->tx_heartbeat_errors = 0; + dst->tx_window_errors = 0; +} - netdev_dev->is_tap = !strcmp(type, "tap"); - netdev_dev->is_internal = (!netdev_dev->is_tap - && dpif_linux_is_internal_device(name)); - netdev_dev->cache_valid |= VALID_IS_PSEUDO; +static int +get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats) +{ + struct dpif_linux_vport reply; + struct ofpbuf *buf; + int error; + + error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf); + if (error) { + return error; + } else if (!reply.stats) { + ofpbuf_delete(buf); + return EOPNOTSUPP; } + + netdev_stats_from_ovs_vport_stats(stats, reply.stats); + + ofpbuf_delete(buf); + + return 0; } static void -swap_uint64(uint64_t *a, uint64_t *b) +get_stats_via_vport(const struct netdev *netdev_, + struct netdev_stats *stats) { - uint64_t tmp = *a; - *a = *b; - *b = tmp; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (!netdev->vport_stats_error || + !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) { + int error; + + error = get_stats_via_vport__(netdev_, stats); + if (error && error != ENOENT) { + VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " + "(%s)", + netdev_get_name(netdev_), ovs_strerror(error)); + } + netdev->vport_stats_error = error; + netdev->cache_valid |= VALID_VPORT_STAT_ERROR; + } +} + +static int +netdev_linux_sys_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) +{ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static int use_netlink_stats; + int error; + + if (ovsthread_once_start(&once)) { + use_netlink_stats = check_for_working_netlink_stats(); + ovsthread_once_done(&once); + } + + if (use_netlink_stats) { + int ifindex; + + error = get_ifindex(netdev_, &ifindex); + if (!error) { + error = get_stats_via_netlink(ifindex, stats); + } + } else { + error = get_stats_via_proc(netdev_get_name(netdev_), stats); + } + + if (error) { + VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d", + netdev_get_name(netdev_), error); + } + return error; + } -/* Retrieves current device stats for 'netdev'. */ +/* Retrieves current device stats for 'netdev-linux'. */ static int netdev_linux_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - static int use_netlink_stats = -1; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_stats dev_stats; int error; - if (netdev_dev->have_vport_stats || - !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) { + get_stats_via_vport(netdev_, stats); - error = netdev_vport_get_stats(netdev_, stats); - netdev_dev->have_vport_stats = !error; - netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS; - } + error = netdev_linux_sys_get_stats(netdev_, &dev_stats); - if (!netdev_dev->have_vport_stats) { - if (use_netlink_stats < 0) { - use_netlink_stats = check_for_working_netlink_stats(); + if (error) { + if (netdev->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->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_linux *netdev = netdev_linux_cast(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->vport_stats_error) { + return error; } else { - error = get_stats_via_proc(netdev_get_name(netdev_), stats); + return 0; } } @@ -1207,9 +1414,8 @@ 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->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); @@ -1225,144 +1431,228 @@ 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; + + 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; +} - return error; +static int +netdev_internal_get_stats(const struct netdev *netdev_, + struct netdev_stats *stats) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + get_stats_via_vport(netdev_, stats); + return netdev->vport_stats_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 "enum ofp_port_features" bits, in host byte order. Returns 0 if - * successful, otherwise a positive errno value. */ static int -netdev_linux_get_features(const struct netdev *netdev, - uint32_t *current, uint32_t *advertised, - uint32_t *supported, uint32_t *peer) +netdev_internal_set_stats(struct netdev *netdev, + const struct netdev_stats *stats) +{ + struct ovs_vport_stats vport_stats; + struct dpif_linux_vport vport; + int err; + + vport_stats.rx_packets = stats->rx_packets; + vport_stats.tx_packets = stats->tx_packets; + vport_stats.rx_bytes = stats->rx_bytes; + vport_stats.tx_bytes = stats->tx_bytes; + vport_stats.rx_errors = stats->rx_errors; + vport_stats.tx_errors = stats->tx_errors; + vport_stats.rx_dropped = stats->rx_dropped; + vport_stats.tx_dropped = stats->tx_dropped; + + dpif_linux_vport_init(&vport); + vport.cmd = OVS_VPORT_CMD_SET; + vport.name = netdev_get_name(netdev); + vport.stats = &vport_stats; + + err = dpif_linux_vport_transact(&vport, NULL, NULL); + + /* If the vport layer doesn't know about the device, that doesn't mean it + * doesn't exist (after all were able to open it when netdev_open() was + * called), it just means that it isn't attached and we'll be getting + * stats a different way. */ + if (err == ENODEV) { + err = EOPNOTSUPP; + } + + return err; +} + +static void +netdev_linux_read_features(struct netdev_linux *netdev) { struct ethtool_cmd ecmd; + uint32_t speed; int error; + if (netdev->cache_valid & VALID_FEATURES) { + return; + } + + COVERAGE_INC(netdev_get_ethtool); memset(&ecmd, 0, sizeof ecmd); - error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, + error = netdev_linux_do_ethtool(netdev->up.name, &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); if (error) { - return error; + goto out; } /* Supported features. */ - *supported = 0; + netdev->supported = 0; if (ecmd.supported & SUPPORTED_10baseT_Half) { - *supported |= OFPPF_10MB_HD; + netdev->supported |= NETDEV_F_10MB_HD; } if (ecmd.supported & SUPPORTED_10baseT_Full) { - *supported |= OFPPF_10MB_FD; + netdev->supported |= NETDEV_F_10MB_FD; } if (ecmd.supported & SUPPORTED_100baseT_Half) { - *supported |= OFPPF_100MB_HD; + netdev->supported |= NETDEV_F_100MB_HD; } if (ecmd.supported & SUPPORTED_100baseT_Full) { - *supported |= OFPPF_100MB_FD; + netdev->supported |= NETDEV_F_100MB_FD; } if (ecmd.supported & SUPPORTED_1000baseT_Half) { - *supported |= OFPPF_1GB_HD; + netdev->supported |= NETDEV_F_1GB_HD; } if (ecmd.supported & SUPPORTED_1000baseT_Full) { - *supported |= OFPPF_1GB_FD; + netdev->supported |= NETDEV_F_1GB_FD; } if (ecmd.supported & SUPPORTED_10000baseT_Full) { - *supported |= OFPPF_10GB_FD; + netdev->supported |= NETDEV_F_10GB_FD; } if (ecmd.supported & SUPPORTED_TP) { - *supported |= OFPPF_COPPER; + netdev->supported |= NETDEV_F_COPPER; } if (ecmd.supported & SUPPORTED_FIBRE) { - *supported |= OFPPF_FIBER; + netdev->supported |= NETDEV_F_FIBER; } if (ecmd.supported & SUPPORTED_Autoneg) { - *supported |= OFPPF_AUTONEG; + netdev->supported |= NETDEV_F_AUTONEG; } if (ecmd.supported & SUPPORTED_Pause) { - *supported |= OFPPF_PAUSE; + netdev->supported |= NETDEV_F_PAUSE; } if (ecmd.supported & SUPPORTED_Asym_Pause) { - *supported |= OFPPF_PAUSE_ASYM; + netdev->supported |= NETDEV_F_PAUSE_ASYM; } /* Advertised features. */ - *advertised = 0; + netdev->advertised = 0; if (ecmd.advertising & ADVERTISED_10baseT_Half) { - *advertised |= OFPPF_10MB_HD; + netdev->advertised |= NETDEV_F_10MB_HD; } if (ecmd.advertising & ADVERTISED_10baseT_Full) { - *advertised |= OFPPF_10MB_FD; + netdev->advertised |= NETDEV_F_10MB_FD; } if (ecmd.advertising & ADVERTISED_100baseT_Half) { - *advertised |= OFPPF_100MB_HD; + netdev->advertised |= NETDEV_F_100MB_HD; } if (ecmd.advertising & ADVERTISED_100baseT_Full) { - *advertised |= OFPPF_100MB_FD; + netdev->advertised |= NETDEV_F_100MB_FD; } if (ecmd.advertising & ADVERTISED_1000baseT_Half) { - *advertised |= OFPPF_1GB_HD; + netdev->advertised |= NETDEV_F_1GB_HD; } if (ecmd.advertising & ADVERTISED_1000baseT_Full) { - *advertised |= OFPPF_1GB_FD; + netdev->advertised |= NETDEV_F_1GB_FD; } if (ecmd.advertising & ADVERTISED_10000baseT_Full) { - *advertised |= OFPPF_10GB_FD; + netdev->advertised |= NETDEV_F_10GB_FD; } if (ecmd.advertising & ADVERTISED_TP) { - *advertised |= OFPPF_COPPER; + netdev->advertised |= NETDEV_F_COPPER; } if (ecmd.advertising & ADVERTISED_FIBRE) { - *advertised |= OFPPF_FIBER; + netdev->advertised |= NETDEV_F_FIBER; } if (ecmd.advertising & ADVERTISED_Autoneg) { - *advertised |= OFPPF_AUTONEG; + netdev->advertised |= NETDEV_F_AUTONEG; } if (ecmd.advertising & ADVERTISED_Pause) { - *advertised |= OFPPF_PAUSE; + netdev->advertised |= NETDEV_F_PAUSE; } if (ecmd.advertising & ADVERTISED_Asym_Pause) { - *advertised |= OFPPF_PAUSE_ASYM; + netdev->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->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; + } else if (speed == SPEED_100) { + netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; + } else if (speed == SPEED_1000) { + netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; + } else if (speed == SPEED_10000) { + netdev->current = NETDEV_F_10GB_FD; + } else if (speed == 40000) { + netdev->current = NETDEV_F_40GB_FD; + } else if (speed == 100000) { + netdev->current = NETDEV_F_100GB_FD; + } else if (speed == 1000000) { + netdev->current = NETDEV_F_1TB_FD; } else { - *current = 0; + netdev->current = 0; } if (ecmd.port == PORT_TP) { - *current |= OFPPF_COPPER; + netdev->current |= NETDEV_F_COPPER; } else if (ecmd.port == PORT_FIBRE) { - *current |= OFPPF_FIBER; + netdev->current |= NETDEV_F_FIBER; } if (ecmd.autoneg) { - *current |= OFPPF_AUTONEG; + netdev->current |= NETDEV_F_AUTONEG; } - /* Peer advertisements. */ - *peer = 0; /* XXX */ +out: + netdev->cache_valid |= VALID_FEATURES; + netdev->get_features_error = error; +} - return 0; +/* Stores the features supported by 'netdev' into of '*current', '*advertised', + * '*supported', and '*peer'. 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_linux *netdev = netdev_linux_cast(netdev_); + + netdev_linux_read_features(netdev); + + if (!netdev->get_features_error) { + *current = netdev->current; + *advertised = netdev->advertised; + *supported = netdev->supported; + *peer = 0; /* XXX */ + } + return netdev->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; + COVERAGE_INC(netdev_get_ethtool); memset(&ecmd, 0, sizeof ecmd); error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, ETHTOOL_GSET, "ETHTOOL_GSET"); @@ -1371,195 +1661,116 @@ 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; } + COVERAGE_INC(netdev_set_ethtool); 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, +netdev_linux_set_policing(struct netdev *netdev_, uint32_t kbits_rate, uint32_t kbits_burst) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - const char *netdev_name = netdev_get_name(netdev); - char command[1024]; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + const char *netdev_name = netdev_get_name(netdev_); + 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->cache_valid & VALID_POLICING) { + if (netdev->netdev_policing_error) { + return netdev->netdev_policing_error; + } + + if (netdev->kbits_rate == kbits_rate && + netdev->kbits_burst == kbits_burst) { + /* Assume that settings haven't changed since we last set them. */ + return 0; + } + netdev->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, ovs_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, ovs_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, ovs_strerror(error)); + goto out; } - - netdev_dev->kbits_rate = kbits_rate; - netdev_dev->kbits_burst = kbits_burst; - netdev_dev->cache_valid |= VALID_POLICING; } - return 0; + netdev->kbits_rate = kbits_rate; + netdev->kbits_burst = kbits_burst; + +out: + if (!error || error == ENODEV) { + netdev->netdev_policing_error = error; + netdev->cache_valid |= VALID_POLICING; + } + return error; } static int netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, struct sset *types) { - const struct tc_ops **opsp; + const struct tc_ops *const *opsp; for (opsp = tcs; *opsp != NULL; opsp++) { const struct tc_ops *ops = *opsp; @@ -1573,7 +1784,7 @@ netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, static const struct tc_ops * tc_lookup_ovs_name(const char *name) { - const struct tc_ops **opsp; + const struct tc_ops *const *opsp; for (opsp = tcs; *opsp != NULL; opsp++) { const struct tc_ops *ops = *opsp; @@ -1587,7 +1798,7 @@ tc_lookup_ovs_name(const char *name) static const struct tc_ops * tc_lookup_linux_name(const char *name) { - const struct tc_ops **opsp; + const struct tc_ops *const *opsp; for (opsp = tcs; *opsp != NULL; opsp++) { const struct tc_ops *ops = *opsp; @@ -1599,14 +1810,13 @@ tc_lookup_linux_name(const char *name) } static struct tc_queue * -tc_find_queue__(const struct netdev *netdev, unsigned int queue_id, +tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id, size_t hash) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct tc_queue *queue; - HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) { + HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) { if (queue->queue_id == queue_id) { return queue; } @@ -1634,30 +1844,28 @@ netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, } static int -netdev_linux_get_qos(const struct netdev *netdev, - const char **typep, struct shash *details) +netdev_linux_get_qos(const struct netdev *netdev_, + const char **typep, struct smap *details) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; } - *typep = netdev_dev->tc->ops->ovs_name; - return (netdev_dev->tc->ops->qdisc_get - ? netdev_dev->tc->ops->qdisc_get(netdev, details) + *typep = netdev->tc->ops->ovs_name; + return (netdev->tc->ops->qdisc_get + ? netdev->tc->ops->qdisc_get(netdev_, details) : 0); } static int -netdev_linux_set_qos(struct netdev *netdev, - const char *type, const struct shash *details) +netdev_linux_set_qos(struct netdev *netdev_, + const char *type, const struct smap *details) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); const struct tc_ops *new_ops; int error; @@ -1666,106 +1874,104 @@ netdev_linux_set_qos(struct netdev *netdev, return EOPNOTSUPP; } - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; } - if (new_ops == netdev_dev->tc->ops) { - return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0; + if (new_ops == netdev->tc->ops) { + return new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; } else { /* Delete existing qdisc. */ - error = tc_del_qdisc(netdev); + error = tc_del_qdisc(netdev_); if (error) { return error; } - assert(netdev_dev->tc == NULL); + ovs_assert(netdev->tc == NULL); /* Install new qdisc. */ - error = new_ops->tc_install(netdev, details); - assert((error == 0) == (netdev_dev->tc != NULL)); + error = new_ops->tc_install(netdev_, details); + ovs_assert((error == 0) == (netdev->tc != NULL)); return error; } } static int -netdev_linux_get_queue(const struct netdev *netdev, - unsigned int queue_id, struct shash *details) +netdev_linux_get_queue(const struct netdev *netdev_, + unsigned int queue_id, struct smap *details) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; } else { - struct tc_queue *queue = tc_find_queue(netdev, queue_id); + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); return (queue - ? netdev_dev->tc->ops->class_get(netdev, queue, details) + ? netdev->tc->ops->class_get(netdev_, queue, details) : ENOENT); } } static int -netdev_linux_set_queue(struct netdev *netdev, - unsigned int queue_id, const struct shash *details) +netdev_linux_set_queue(struct netdev *netdev_, + unsigned int queue_id, const struct smap *details) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; - } else if (queue_id >= netdev_dev->tc->ops->n_queues - || !netdev_dev->tc->ops->class_set) { + } else if (queue_id >= netdev->tc->ops->n_queues + || !netdev->tc->ops->class_set) { return EINVAL; } - return netdev_dev->tc->ops->class_set(netdev, queue_id, details); + return netdev->tc->ops->class_set(netdev_, queue_id, details); } static int -netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id) +netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; - } else if (!netdev_dev->tc->ops->class_delete) { + } else if (!netdev->tc->ops->class_delete) { return EINVAL; } else { - struct tc_queue *queue = tc_find_queue(netdev, queue_id); + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); return (queue - ? netdev_dev->tc->ops->class_delete(netdev, queue) + ? netdev->tc->ops->class_delete(netdev_, queue) : ENOENT); } } static int -netdev_linux_get_queue_stats(const struct netdev *netdev, +netdev_linux_get_queue_stats(const struct netdev *netdev_, unsigned int queue_id, struct netdev_queue_stats *stats) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; - } else if (!netdev_dev->tc->ops->class_get_stats) { + } else if (!netdev->tc->ops->class_get_stats) { return EOPNOTSUPP; } else { - const struct tc_queue *queue = tc_find_queue(netdev, queue_id); - return (queue - ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats) - : ENOENT); + const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + if (!queue) { + return ENOENT; + } + stats->created = queue->created; + return netdev->tc->ops->class_get_stats(netdev_, queue, stats); } } @@ -1780,70 +1986,69 @@ start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) return false; } tcmsg->tcm_parent = 0; - nl_dump_start(dump, rtnl_sock, &request); + nl_dump_start(dump, NETLINK_ROUTE, &request); ofpbuf_uninit(&request); return true; } static int -netdev_linux_dump_queues(const struct netdev *netdev, +netdev_linux_dump_queues(const struct netdev *netdev_, netdev_dump_queues_cb *cb, void *aux) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - struct tc_queue *queue; - struct shash details; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct tc_queue *queue, *next_queue; + struct smap details; int last_error; int error; - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; - } else if (!netdev_dev->tc->ops->class_get) { + } else if (!netdev->tc->ops->class_get) { return EOPNOTSUPP; } last_error = 0; - shash_init(&details); - HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) { - shash_clear(&details); + smap_init(&details); + HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node, + &netdev->tc->queues) { + smap_clear(&details); - error = netdev_dev->tc->ops->class_get(netdev, queue, &details); + error = netdev->tc->ops->class_get(netdev_, queue, &details); if (!error) { (*cb)(queue->queue_id, &details, aux); } else { last_error = error; } } - shash_destroy(&details); + smap_destroy(&details); return last_error; } static int -netdev_linux_dump_queue_stats(const struct netdev *netdev, +netdev_linux_dump_queue_stats(const struct netdev *netdev_, netdev_dump_queue_stats_cb *cb, void *aux) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct nl_dump dump; struct ofpbuf msg; int last_error; int error; - error = tc_query_qdisc(netdev); + error = tc_query_qdisc(netdev_); if (error) { return error; - } else if (!netdev_dev->tc->ops->class_dump_stats) { + } else if (!netdev->tc->ops->class_dump_stats) { return EOPNOTSUPP; } last_error = 0; - if (!start_queue_dump(netdev, &dump)) { + if (!start_queue_dump(netdev_, &dump)) { return ENODEV; } while (nl_dump_next(&dump, &msg)) { - error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux); + error = netdev->tc->ops->class_dump_stats(netdev_, &msg, cb, aux); if (error) { last_error = error; } @@ -1857,28 +2062,27 @@ static int netdev_linux_get_in4(const struct netdev *netdev_, struct in_addr *address, struct in_addr *netmask) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); - if (!(netdev_dev->cache_valid & VALID_IN4)) { + if (!(netdev->cache_valid & VALID_IN4)) { int error; - error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address, + error = netdev_linux_get_ipv4(netdev_, &netdev->address, SIOCGIFADDR, "SIOCGIFADDR"); if (error) { return error; } - error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask, + error = netdev_linux_get_ipv4(netdev_, &netdev->netmask, SIOCGIFNETMASK, "SIOCGIFNETMASK"); if (error) { return error; } - netdev_dev->cache_valid |= VALID_IN4; + netdev->cache_valid |= VALID_IN4; } - *address = netdev_dev->address; - *netmask = netdev_dev->netmask; + *address = netdev->address; + *netmask = netdev->netmask; return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; } @@ -1886,15 +2090,14 @@ static int netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, struct in_addr netmask) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); if (!error) { - netdev_dev->cache_valid |= VALID_IN4; - netdev_dev->address = address; - netdev_dev->netmask = netmask; + netdev->cache_valid |= VALID_IN4; + netdev->address = address; + netdev->netmask = netmask; if (address.s_addr != INADDR_ANY) { error = do_set_addr(netdev_, SIOCSIFNETMASK, "SIOCSIFNETMASK", netmask); @@ -1924,13 +2127,12 @@ parse_if_inet6_line(const char *line, static int netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev_)); - if (!(netdev_dev->cache_valid & VALID_IN6)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + if (!(netdev->cache_valid & VALID_IN6)) { FILE *file; char line[128]; - netdev_dev->in6 = in6addr_any; + netdev->in6 = in6addr_any; file = fopen("/proc/net/if_inet6", "r"); if (file != NULL) { @@ -1941,15 +2143,15 @@ netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) if (parse_if_inet6_line(line, &in6_tmp, ifname) && !strcmp(name, ifname)) { - netdev_dev->in6 = in6_tmp; + netdev->in6 = in6_tmp; break; } } fclose(file); } - netdev_dev->cache_valid |= VALID_IN6; + netdev->cache_valid |= VALID_IN6; } - *in6 = netdev_dev->in6; + *in6 = netdev->in6; return 0; } @@ -1971,11 +2173,10 @@ do_set_addr(struct netdev *netdev, int ioctl_nr, const char *ioctl_name, struct in_addr addr) { struct ifreq ifr; - 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, - ioctl_name); + make_in4_sockaddr(&ifr.ifr_addr, addr); + return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, + ioctl_name); } /* Adds 'router' as a default IP gateway. */ @@ -1991,9 +2192,9 @@ 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; - error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; + error = af_inet_ioctl(SIOCADDRT, &rt); if (error) { - VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); + VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); } return error; } @@ -2010,7 +2211,7 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, *netdev_name = NULL; stream = fopen(fn, "r"); if (stream == NULL) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); + VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); return errno; } @@ -2018,7 +2219,7 @@ 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; @@ -2060,32 +2261,48 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, } static int -netdev_linux_get_status(const struct netdev *netdev, struct shash *sh) +netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) { - struct ethtool_drvinfo drvinfo; - int error; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error = 0; - memset(&drvinfo, 0, sizeof drvinfo); - error = netdev_linux_do_ethtool(netdev_get_name(netdev), - (struct ethtool_cmd *)&drvinfo, - ETHTOOL_GDRVINFO, - "ETHTOOL_GDRVINFO"); - if (!error) { - shash_add(sh, "driver_name", xstrdup(drvinfo.driver)); - shash_add(sh, "driver_version", xstrdup(drvinfo.version)); - shash_add(sh, "firmware_version", xstrdup(drvinfo.fw_version)); + if (!(netdev->cache_valid & VALID_DRVINFO)) { + struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; + + COVERAGE_INC(netdev_get_ethtool); + memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo); + error = netdev_linux_do_ethtool(netdev->up.name, + cmd, + ETHTOOL_GDRVINFO, + "ETHTOOL_GDRVINFO"); + if (!error) { + netdev->cache_valid |= VALID_DRVINFO; + } } + if (!error) { + smap_add(smap, "driver_name", netdev->drvinfo.driver); + smap_add(smap, "driver_version", netdev->drvinfo.version); + smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); + } return error; } +static int +netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, + struct smap *smap) +{ + smap_add(smap, "driver_name", "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, - uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) + ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) { struct arpreq r; struct sockaddr_in sin; @@ -2101,12 +2318,13 @@ netdev_linux_arp_lookup(const struct netdev *netdev, r.arp_flags = 0; 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; + retval = af_inet_ioctl(SIOCGARP, &r); if (!retval) { memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); } else if (retval != ENXIO) { VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", - netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval)); + netdev_get_name(netdev), IP_ARGS(ip), + ovs_strerror(retval)); } return retval; } @@ -2138,127 +2356,45 @@ iff_to_nd_flags(int iff) } static int -netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, +netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, enum netdev_flags on, enum netdev_flags *old_flagsp) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); 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); - } + old_flags = netdev->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_get_name(netdev_), new_flags); + get_flags(netdev_, &netdev->ifi_flags); } return error; } -static void -poll_notify(struct list *list) -{ - struct netdev_linux_notifier *notifier; - LIST_FOR_EACH (notifier, node, list) { - struct netdev_notifier *n = ¬ifier->notifier; - n->cb(n); - } -} - -static void -netdev_linux_poll_cb(const struct rtnetlink_link_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); - } - } -} - -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; - error = rtnetlink_link_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_) +static unsigned int +netdev_linux_change_seq(const struct netdev *netdev) { - 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_link_notifier_unregister(&netdev_linux_poll_notifier); - } + return netdev_linux_cast(netdev)->change_seq; } -#define NETDEV_LINUX_CLASS(NAME, CREATE, ENUMERATE, SET_STATS) \ +#define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \ + GET_FEATURES, GET_STATUS) \ { \ NAME, \ \ - netdev_linux_init, \ + NULL, \ netdev_linux_run, \ netdev_linux_wait, \ \ CREATE, \ netdev_linux_destroy, \ + NULL, /* get_config */ \ NULL, /* set_config */ \ + NULL, /* get_tunnel_config */ \ \ - netdev_linux_open, \ - netdev_linux_close, \ - \ - ENUMERATE, \ - \ - netdev_linux_recv, \ - netdev_linux_recv_wait, \ - netdev_linux_drain, \ + netdev_linux_rx_open, \ \ netdev_linux_send, \ netdev_linux_send_wait, \ @@ -2266,15 +2402,16 @@ netdev_linux_poll_remove(struct netdev_notifier *notifier_) 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_miimon, \ - netdev_linux_get_stats, \ + netdev_linux_get_carrier_resets, \ + netdev_linux_set_miimon_interval, \ + GET_STATS, \ SET_STATS, \ \ - netdev_linux_get_features, \ + GET_FEATURES, \ netdev_linux_set_advertisements, \ - netdev_linux_get_vlan_vid, \ \ netdev_linux_set_policing, \ netdev_linux_get_qos_types, \ @@ -2293,35 +2430,47 @@ netdev_linux_poll_remove(struct netdev_notifier *notifier_) netdev_linux_get_in6, \ netdev_linux_add_router, \ netdev_linux_get_next_hop, \ - netdev_linux_get_status, \ + GET_STATUS, \ netdev_linux_arp_lookup, \ \ netdev_linux_update_flags, \ \ - netdev_linux_poll_add, \ - netdev_linux_poll_remove \ + netdev_linux_change_seq \ } const struct netdev_class netdev_linux_class = NETDEV_LINUX_CLASS( "system", netdev_linux_create, - netdev_linux_enumerate, - NULL); /* set_stats */ + netdev_linux_get_stats, + NULL, /* set_stats */ + netdev_linux_get_features, + netdev_linux_get_status); const struct netdev_class netdev_tap_class = NETDEV_LINUX_CLASS( "tap", netdev_linux_create_tap, - NULL, /* enumerate */ - NULL); /* set_stats */ + netdev_tap_get_stats, + NULL, /* set_stats */ + netdev_linux_get_features, + netdev_linux_get_status); const struct netdev_class netdev_internal_class = NETDEV_LINUX_CLASS( "internal", netdev_linux_create, - NULL, /* enumerate */ - netdev_vport_set_stats); + netdev_internal_get_stats, + netdev_internal_set_stats, + NULL, /* get_features */ + netdev_internal_get_status); + +static const struct netdev_rx_class netdev_rx_linux_class = { + netdev_rx_linux_destroy, + netdev_rx_linux_recv, + netdev_rx_linux_wait, + netdev_rx_linux_drain, +}; /* HTB traffic control class. */ @@ -2341,25 +2490,23 @@ struct htb_class { }; static struct htb * -htb_get__(const struct netdev *netdev) +htb_get__(const struct netdev *netdev_) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - return CONTAINER_OF(netdev_dev->tc, struct htb, tc); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + return CONTAINER_OF(netdev->tc, struct htb, tc); } static void -htb_install__(struct netdev *netdev, uint64_t max_rate) +htb_install__(struct netdev *netdev_, uint64_t max_rate) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct htb *htb; htb = xmalloc(sizeof *htb); tc_init(&htb->tc, &tc_ops_htb); htb->max_rate = max_rate; - netdev_dev->tc = &htb->tc; + netdev->tc = &htb->tc; } /* Create an HTB qdisc. @@ -2410,11 +2557,11 @@ htb_setup_class__(struct netdev *netdev, unsigned int handle, int error; int mtu; - netdev_get_mtu(netdev, &mtu); - if (mtu == INT_MAX) { + 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 EINVAL; + return error; } memset(&opt, 0, sizeof opt); @@ -2446,7 +2593,7 @@ htb_setup_class__(struct netdev *netdev, unsigned int handle, tc_get_major(handle), tc_get_minor(handle), tc_get_major(parent), tc_get_minor(parent), class->min_rate, class->max_rate, - class->burst, class->priority, strerror(error)); + class->burst, class->priority, ovs_strerror(error)); } return error; } @@ -2507,17 +2654,17 @@ htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, static void htb_parse_qdisc_details__(struct netdev *netdev, - const struct shash *details, struct htb_class *hc) + const struct smap *details, struct htb_class *hc) { const char *max_rate_s; - max_rate_s = shash_find_data(details, "max-rate"); + max_rate_s = smap_get(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; + hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; } hc->min_rate = hc->max_rate; hc->burst = 0; @@ -2526,20 +2673,20 @@ htb_parse_qdisc_details__(struct netdev *netdev, static int htb_parse_class_details__(struct netdev *netdev, - const struct shash *details, struct htb_class *hc) + const struct smap *details, struct htb_class *hc) { const struct htb *htb = htb_get__(netdev); - const char *min_rate_s = shash_find_data(details, "min-rate"); - 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; + const char *min_rate_s = smap_get(details, "min-rate"); + const char *max_rate_s = smap_get(details, "max-rate"); + const char *burst_s = smap_get(details, "burst"); + const char *priority_s = smap_get(details, "priority"); + int mtu, error; - netdev_get_mtu(netdev, &mtu); - if (mtu == INT_MAX) { + 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 EINVAL; + return error; } /* HTB requires at least an mtu sized min-rate to send any traffic even @@ -2590,7 +2737,7 @@ htb_query_class__(const struct netdev *netdev, unsigned int handle, } static int -htb_tc_install(struct netdev *netdev, const struct shash *details) +htb_tc_install(struct netdev *netdev, const struct smap *details) { int error; @@ -2630,6 +2777,7 @@ htb_update_queue__(struct netdev *netdev, unsigned int queue_id, hcp = xmalloc(sizeof *hcp); queue = &hcp->tc_queue; queue->queue_id = queue_id; + queue->created = time_msec(); hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); } @@ -2682,15 +2830,15 @@ htb_tc_destroy(struct tc *tc) } static int -htb_qdisc_get(const struct netdev *netdev, struct shash *details) +htb_qdisc_get(const struct netdev *netdev, struct smap *details) { const struct htb *htb = htb_get__(netdev); - shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate)); + smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); return 0; } static int -htb_qdisc_set(struct netdev *netdev, const struct shash *details) +htb_qdisc_set(struct netdev *netdev, const struct smap *details) { struct htb_class hc; int error; @@ -2706,24 +2854,24 @@ htb_qdisc_set(struct netdev *netdev, const struct shash *details) static int htb_class_get(const struct netdev *netdev OVS_UNUSED, - const struct tc_queue *queue, struct shash *details) + const struct tc_queue *queue, struct smap *details) { const struct htb_class *hc = htb_class_cast__(queue); - shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate)); + smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); if (hc->min_rate != hc->max_rate) { - shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate)); + smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); } - shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst)); + smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); if (hc->priority) { - shash_add(details, "priority", xasprintf("%u", hc->priority)); + smap_add_format(details, "priority", "%u", hc->priority); } return 0; } static int htb_class_set(struct netdev *netdev, unsigned int queue_id, - const struct shash *details) + const struct smap *details) { struct htb_class hc; int error; @@ -2820,11 +2968,10 @@ struct hfsc_class { }; static struct hfsc * -hfsc_get__(const struct netdev *netdev) +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); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + return CONTAINER_OF(netdev->tc, struct hfsc, tc); } static struct hfsc_class * @@ -2834,16 +2981,15 @@ hfsc_class_cast__(const struct tc_queue *queue) } static void -hfsc_install__(struct netdev *netdev, uint32_t max_rate) +hfsc_install__(struct netdev *netdev_, uint32_t max_rate) { - struct netdev_dev_linux * netdev_dev; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); 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; + netdev->tc = &hfsc->tc; } static void @@ -2865,6 +3011,7 @@ hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, hcp = xmalloc(sizeof *hcp); queue = &hcp->tc_queue; queue->queue_id = queue_id; + queue->created = time_msec(); hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); } @@ -2983,20 +3130,20 @@ hfsc_query_class__(const struct netdev *netdev, unsigned int handle, } static void -hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details, +hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *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_s = smap_get(details, "max-rate"); max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; if (!max_rate) { - uint32_t current; + enum netdev_features current; netdev_get_features(netdev, ¤t, NULL, NULL, NULL); - max_rate = netdev_features_to_bps(current) / 8; + max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; } class->min_rate = max_rate; @@ -3005,7 +3152,7 @@ hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details, static int hfsc_parse_class_details__(struct netdev *netdev, - const struct shash *details, + const struct smap *details, struct hfsc_class * class) { const struct hfsc *hfsc; @@ -3013,8 +3160,8 @@ hfsc_parse_class_details__(struct netdev *netdev, 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_s = smap_get(details, "min-rate"); + max_rate_s = smap_get(details, "max-rate"); min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; min_rate = MAX(min_rate, 1); @@ -3108,14 +3255,14 @@ hfsc_setup_class__(struct netdev *netdev, unsigned int handle, 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)); + class->min_rate, class->max_rate, ovs_strerror(error)); } return error; } static int -hfsc_tc_install(struct netdev *netdev, const struct shash *details) +hfsc_tc_install(struct netdev *netdev, const struct smap *details) { int error; struct hfsc_class class; @@ -3183,16 +3330,16 @@ hfsc_tc_destroy(struct tc *tc) } static int -hfsc_qdisc_get(const struct netdev *netdev, struct shash *details) +hfsc_qdisc_get(const struct netdev *netdev, struct smap *details) { const struct hfsc *hfsc; hfsc = hfsc_get__(netdev); - shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate)); + smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); return 0; } static int -hfsc_qdisc_set(struct netdev *netdev, const struct shash *details) +hfsc_qdisc_set(struct netdev *netdev, const struct smap *details) { int error; struct hfsc_class class; @@ -3210,21 +3357,21 @@ hfsc_qdisc_set(struct netdev *netdev, const struct shash *details) static int hfsc_class_get(const struct netdev *netdev OVS_UNUSED, - const struct tc_queue *queue, struct shash *details) + const struct tc_queue *queue, struct smap *details) { const struct hfsc_class *hc; hc = hfsc_class_cast__(queue); - shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate)); + smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); if (hc->min_rate != hc->max_rate) { - shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate)); + smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); } return 0; } static int hfsc_class_set(struct netdev *netdev, unsigned int queue_id, - const struct shash *details) + const struct smap *details) { int error; struct hfsc_class class; @@ -3314,22 +3461,19 @@ static const struct tc_ops tc_ops_hfsc = { * the "" (empty string) QoS type in the OVS database. */ static void -default_install__(struct netdev *netdev) +default_install__(struct netdev *netdev_) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - static struct tc *tc; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); - if (!tc) { - tc = xmalloc(sizeof *tc); - tc_init(tc, &tc_ops_default); - } - netdev_dev->tc = tc; + /* Nothing but a tc class implementation is allowed to write to a tc. This + * class never does that, so we can legitimately use a const tc object. */ + netdev->tc = CONST_CAST(struct tc *, &tc); } static int default_tc_install(struct netdev *netdev, - const struct shash *details OVS_UNUSED) + const struct smap *details OVS_UNUSED) { default_install__(netdev); return 0; @@ -3363,17 +3507,14 @@ static const struct tc_ops tc_ops_default = { * */ static int -other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) +other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); - static struct tc *tc; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other); - if (!tc) { - tc = xmalloc(sizeof *tc); - tc_init(tc, &tc_ops_other); - } - netdev_dev->tc = tc; + /* Nothing but a tc class implementation is allowed to write to a tc. This + * class never does that, so we can legitimately use a const tc object. */ + netdev->tc = CONST_CAST(struct tc *, &tc); return 0; } @@ -3465,11 +3606,112 @@ tc_make_request(const struct netdev *netdev, int type, unsigned int flags, static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) { - int error = nl_sock_transact(rtnl_sock, request, replyp); + int error = nl_transact(NETLINK_ROUTE, request, replyp); ofpbuf_uninit(request); 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 * 1000)/8, 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) { @@ -3509,30 +3751,35 @@ read_psched(void) * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion * [6] 2.6.34 from kernel.org on KVM */ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; static const char fn[] = "/proc/net/psched"; unsigned int a, b, c, d; FILE *stream; + if (!ovsthread_once_start(&once)) { + return; + } + ticks_per_s = 1.0; buffer_hz = 100; stream = fopen(fn, "r"); if (!stream) { - VLOG_WARN("%s: open failed: %s", fn, strerror(errno)); - return; + VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); + goto exit; } if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { VLOG_WARN("%s: read failed", fn); fclose(stream); - return; + goto exit; } VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); fclose(stream); if (!a || !c) { VLOG_WARN("%s: invalid scheduler parameters", fn); - return; + goto exit; } ticks_per_s = (double) a * c / b; @@ -3543,6 +3790,9 @@ read_psched(void) fn, a, b, c, d); } VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); + +exit: + ovsthread_once_done(&once); } /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a @@ -3550,9 +3800,7 @@ read_psched(void) static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) { - if (!buffer_hz) { - read_psched(); - } + read_psched(); return (rate * ticks) / ticks_per_s; } @@ -3561,9 +3809,7 @@ tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size) { - if (!buffer_hz) { - read_psched(); - } + read_psched(); return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; } @@ -3572,9 +3818,7 @@ tc_bytes_to_ticks(unsigned int rate, unsigned int size) static unsigned int tc_buffer_per_jiffy(unsigned int rate) { - if (!buffer_hz) { - read_psched(); - } + read_psched(); return rate / buffer_hz; } @@ -3722,7 +3966,7 @@ tc_query_class(const struct netdev *netdev, netdev_get_name(netdev), tc_get_major(handle), tc_get_minor(handle), tc_get_major(parent), tc_get_minor(parent), - strerror(error)); + ovs_strerror(error)); } return error; } @@ -3747,22 +3991,21 @@ tc_delete_class(const struct netdev *netdev, unsigned int handle) VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", netdev_get_name(netdev), tc_get_major(handle), tc_get_minor(handle), - strerror(error)); + ovs_strerror(error)); } return error; } /* Equivalent to "tc qdisc del dev root". */ static int -tc_del_qdisc(struct netdev *netdev) +tc_del_qdisc(struct netdev *netdev_) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct ofpbuf request; struct tcmsg *tcmsg; int error; - tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); + tcmsg = tc_make_request(netdev_, RTM_DELQDISC, 0, &request); if (!tcmsg) { return ENODEV; } @@ -3775,11 +4018,11 @@ tc_del_qdisc(struct netdev *netdev) * case we've accomplished our purpose. */ error = 0; } - if (!error && netdev_dev->tc) { - if (netdev_dev->tc->ops->tc_destroy) { - netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); + if (!error && netdev->tc) { + if (netdev->tc->ops->tc_destroy) { + netdev->tc->ops->tc_destroy(netdev->tc); } - netdev_dev->tc = NULL; + netdev->tc = NULL; } return error; } @@ -3788,17 +4031,16 @@ tc_del_qdisc(struct netdev *netdev) * kernel to determine what they are. Returns 0 if successful, otherwise a * positive errno value. */ static int -tc_query_qdisc(const struct netdev *netdev) +tc_query_qdisc(const struct netdev *netdev_) { - struct netdev_dev_linux *netdev_dev = - netdev_dev_linux_cast(netdev_get_dev(netdev)); + struct netdev_linux *netdev = netdev_linux_cast(netdev_); struct ofpbuf request, *qdisc; const struct tc_ops *ops; struct tcmsg *tcmsg; int load_error; int error; - if (netdev_dev->tc) { + if (netdev->tc) { return 0; } @@ -3816,7 +4058,7 @@ tc_query_qdisc(const struct netdev *netdev) * * We could check for Linux 2.6.35+ and use a more straightforward method * there. */ - tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request); + tcmsg = tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, &request); if (!tcmsg) { return ENODEV; } @@ -3849,13 +4091,13 @@ tc_query_qdisc(const struct netdev *netdev) } else { /* Who knows? Maybe the device got deleted. */ VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", - netdev_get_name(netdev), strerror(error)); + netdev_get_name(netdev_), ovs_strerror(error)); ops = &tc_ops_other; } /* Instantiate it. */ - load_error = ops->tc_load((struct netdev *) netdev, qdisc); - assert((load_error == 0) == (netdev_dev->tc != NULL)); + load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc); + ovs_assert((load_error == 0) == (netdev->tc != NULL)); ofpbuf_delete(qdisc); return error ? error : load_error; @@ -3930,56 +4172,85 @@ tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); } -/* Public utility functions. */ - -#define COPY_NETDEV_STATS \ - dst->rx_packets = src->rx_packets; \ - dst->tx_packets = src->tx_packets; \ - dst->rx_bytes = src->rx_bytes; \ - dst->tx_bytes = src->tx_bytes; \ - dst->rx_errors = src->rx_errors; \ - dst->tx_errors = src->tx_errors; \ - dst->rx_dropped = src->rx_dropped; \ - dst->tx_dropped = src->tx_dropped; \ - dst->multicast = src->multicast; \ - dst->collisions = src->collisions; \ - dst->rx_length_errors = src->rx_length_errors; \ - dst->rx_over_errors = src->rx_over_errors; \ - dst->rx_crc_errors = src->rx_crc_errors; \ - dst->rx_frame_errors = src->rx_frame_errors; \ - dst->rx_fifo_errors = src->rx_fifo_errors; \ - dst->rx_missed_errors = src->rx_missed_errors; \ - dst->tx_aborted_errors = src->tx_aborted_errors; \ - dst->tx_carrier_errors = src->tx_carrier_errors; \ - dst->tx_fifo_errors = src->tx_fifo_errors; \ - dst->tx_heartbeat_errors = src->tx_heartbeat_errors; \ - dst->tx_window_errors = src->tx_window_errors +/* Linux-only functions declared in netdev-linux.h */ -/* Copies 'src' into 'dst', performing format conversion in the process. */ -void -netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, - const struct rtnl_link_stats *src) +/* 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) { - COPY_NETDEV_STATS; -} + const char *netdev_name = netdev_get_name(netdev); + struct ethtool_value evalue; + uint32_t new_flags; + int error; -/* Copies 'src' into 'dst', performing format conversion in the process. */ -void -netdev_stats_from_rtnl_link_stats64(struct netdev_stats *dst, - const struct rtnl_link_stats64 *src) -{ - COPY_NETDEV_STATS; + COVERAGE_INC(netdev_get_ethtool); + memset(&evalue, 0, sizeof evalue); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); + if (error) { + return error; + } + + COVERAGE_INC(netdev_set_ethtool); + 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; + } + + COVERAGE_INC(netdev_get_ethtool); + 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. */ -void -netdev_stats_to_rtnl_link_stats64(struct rtnl_link_stats64 *dst, - const struct netdev_stats *src) +static void +netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, + const struct rtnl_link_stats *src) { - COPY_NETDEV_STATS; + dst->rx_packets = src->rx_packets; + dst->tx_packets = src->tx_packets; + dst->rx_bytes = src->rx_bytes; + dst->tx_bytes = src->tx_bytes; + dst->rx_errors = src->rx_errors; + dst->tx_errors = src->tx_errors; + dst->rx_dropped = src->rx_dropped; + dst->tx_dropped = src->tx_dropped; + dst->multicast = src->multicast; + dst->collisions = src->collisions; + dst->rx_length_errors = src->rx_length_errors; + dst->rx_over_errors = src->rx_over_errors; + dst->rx_crc_errors = src->rx_crc_errors; + dst->rx_frame_errors = src->rx_frame_errors; + dst->rx_fifo_errors = src->rx_fifo_errors; + dst->rx_missed_errors = src->rx_missed_errors; + dst->tx_aborted_errors = src->tx_aborted_errors; + dst->tx_carrier_errors = src->tx_carrier_errors; + dst->tx_fifo_errors = src->tx_fifo_errors; + dst->tx_heartbeat_errors = src->tx_heartbeat_errors; + dst->tx_window_errors = src->tx_window_errors; } - -/* Utility functions. */ static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats) @@ -4005,7 +4276,7 @@ get_stats_via_netlink(int ifindex, struct netdev_stats *stats) ifi = ofpbuf_put_zeros(&request, sizeof *ifi); ifi->ifi_family = PF_UNSPEC; ifi->ifi_index = ifindex; - error = nl_sock_transact(rtnl_sock, &request, &reply); + error = nl_transact(NETLINK_ROUTE, &request, &reply); ofpbuf_uninit(&request); if (error) { return error; @@ -4041,7 +4312,7 @@ get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) stream = fopen(fn, "r"); if (!stream) { - VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); + VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); return errno; } @@ -4089,38 +4360,42 @@ 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, unsigned int *flags) { struct ifreq ifr; int error; - error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS, - "SIOCGIFFLAGS"); - *flags = ifr.ifr_flags; + *flags = 0; + error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); + if (!error) { + *flags = ifr.ifr_flags; + } return error; } static int -set_flags(struct netdev *netdev, int flags) +set_flags(const char *name, unsigned int flags) { struct ifreq ifr; ifr.ifr_flags = flags; - return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS, - "SIOCSIFFLAGS"); + return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); } static int do_get_ifindex(const char *netdev_name) { struct ifreq ifr; + int error; ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); COVERAGE_INC(netdev_get_ifindex); - if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { + + error = af_inet_ioctl(SIOCGIFINDEX, &ifr); + if (error) { VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", - netdev_name, strerror(errno)); - return -errno; + netdev_name, ovs_strerror(error)); + return -error; } return ifr.ifr_ifindex; } @@ -4128,19 +4403,23 @@ do_get_ifindex(const char *netdev_name) static int 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)) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + + if (!(netdev->cache_valid & VALID_IFINDEX)) { int ifindex = do_get_ifindex(netdev_get_name(netdev_)); + if (ifindex < 0) { - return -ifindex; + netdev->get_ifindex_error = -ifindex; + netdev->ifindex = 0; + } else { + netdev->get_ifindex_error = 0; + netdev->ifindex = ifindex; } - netdev_dev->cache_valid |= VALID_IFINDEX; - netdev_dev->ifindex = ifindex; + netdev->cache_valid |= VALID_IFINDEX; } - *ifindexp = netdev_dev->ifindex; - return 0; + + *ifindexp = netdev->ifindex; + return netdev->get_ifindex_error; } static int @@ -4148,14 +4427,20 @@ get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) { struct ifreq ifr; int hwaddr_family; + int error; memset(&ifr, 0, sizeof ifr); 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)); - return errno; + error = af_inet_ioctl(SIOCGIFHWADDR, &ifr); + if (error) { + /* 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(error == ENODEV ? VLL_INFO : VLL_ERR, + "ioctl(SIOCGIFHWADDR) on %s device failed: %s", + netdev_name, ovs_strerror(error)); + return error; } hwaddr_family = ifr.ifr_hwaddr.sa_family; if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { @@ -4167,22 +4452,23 @@ 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; + int error; memset(&ifr, 0, sizeof ifr); ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); - ifr.ifr_hwaddr.sa_family = hwaddr_family; + 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) { + error = af_inet_ioctl(SIOCSIFHWADDR, &ifr); + if (error) { VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", - netdev_name, strerror(errno)); - return errno; + netdev_name, ovs_strerror(error)); } - return 0; + return error; } static int @@ -4190,38 +4476,24 @@ netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, int cmd, const char *cmd_name) { struct ifreq ifr; + int error; memset(&ifr, 0, sizeof ifr); ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); ifr.ifr_data = (caddr_t) ecmd; ecmd->cmd = cmd; - COVERAGE_INC(netdev_ethtool); - if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { - return 0; - } else { - if (errno != EOPNOTSUPP) { + error = af_inet_ioctl(SIOCETHTOOL, &ifr); + if (error) { + if (error != EOPNOTSUPP) { VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " - "failed: %s", cmd_name, name, strerror(errno)); + "failed: %s", cmd_name, name, ovs_strerror(error)); } else { /* The device doesn't support this operation. That's pretty * common, so there's no point in logging anything. */ } - return errno; - } -} - -static int -netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, - const char *cmd_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)); - return errno; } - return 0; + return error; } static int @@ -4232,10 +4504,37 @@ netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, int error; ifr.ifr_addr.sa_family = AF_INET; - error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); + error = af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); if (!error) { - const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; + const struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, + &ifr.ifr_addr); *ip = sin->sin_addr; } return error; } + +/* Returns an AF_PACKET raw socket or a negative errno value. */ +static int +af_packet_sock(void) +{ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + static int sock; + + if (ovsthread_once_start(&once)) { + sock = socket(AF_PACKET, SOCK_RAW, 0); + if (sock >= 0) { + int error = set_nonblocking(sock); + if (error) { + close(sock); + sock = -error; + } + } else { + sock = -errno; + VLOG_ERR("failed to create packet socket: %s", + ovs_strerror(errno)); + } + ovsthread_once_done(&once); + } + + return sock; +}