X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=d924cf3654c19fb27ef2b01ce1561faf5176d383;hb=45c8d3a189843f0f45398caa420b952d5acd1f19;hp=5f13f18a404cd07a1f5610b15b2d33ab7cf007b0;hpb=007948177581f3b3dad188221593d0e4bdca6ba0;p=sliver-openvswitch.git diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 5f13f18a4..d924cf365 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -58,16 +58,17 @@ #include "netdev-provider.h" #include "netdev-vport.h" #include "netlink.h" +#include "netlink-notifier.h" #include "netlink-socket.h" #include "ofpbuf.h" #include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" -#include "rtnetlink.h" #include "rtnetlink-link.h" #include "socket-util.h" #include "shash.h" #include "sset.h" +#include "timer.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(netdev_linux); @@ -89,6 +90,15 @@ 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 @@ -345,6 +355,11 @@ struct netdev_dev_linux { struct shash_node *shash_node; unsigned int cache_valid; + unsigned int change_seq; + + bool miimon; /* Link status of last poll. */ + long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */ + struct timer miimon_timer; /* The following are figured out "on demand" only. They are only valid * when the corresponding VALID_* bit in 'cache_valid' is set. */ @@ -373,20 +388,10 @@ struct netdev_linux { /* 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; - /* 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); @@ -411,6 +416,9 @@ static int set_etheraddr(const char *netdev_name, int hwaddr_family, const uint8_t[ETH_ADDR_LEN]); static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); +static int af_packet_sock(void); +static void netdev_linux_miimon_run(void); +static void netdev_linux_miimon_wait(void); static bool is_netdev_linux_class(const struct netdev_class *netdev_class) @@ -447,15 +455,6 @@ netdev_linux_init(void) 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) { - VLOG_ERR("failed to create packet socket: %s", - strerror(status)); - } - set_nonblocking(af_packet_sock); } /* Create rtnetlink socket. */ @@ -474,12 +473,24 @@ static void netdev_linux_run(void) { rtnetlink_link_notifier_run(); + netdev_linux_miimon_run(); } static void netdev_linux_wait(void) { rtnetlink_link_notifier_wait(); + netdev_linux_miimon_wait(); +} + +static void +netdev_dev_linux_changed(struct netdev_dev_linux *dev) +{ + dev->change_seq++; + if (!dev->change_seq) { + dev->change_seq++; + } + dev->cache_valid = 0; } static void @@ -495,7 +506,7 @@ netdev_linux_cache_cb(const struct rtnetlink_link_change *change, if (is_netdev_linux_class(netdev_class)) { dev = netdev_dev_linux_cast(base_dev); - dev->cache_valid = 0; + netdev_dev_linux_changed(dev); } } } else { @@ -506,7 +517,7 @@ netdev_linux_cache_cb(const struct rtnetlink_link_change *change, netdev_dev_get_devices(&netdev_linux_class, &device_shash); SHASH_FOR_EACH (node, &device_shash) { dev = node->data; - dev->cache_valid = 0; + netdev_dev_linux_changed(dev); } shash_destroy(&device_shash); } @@ -514,18 +525,12 @@ netdev_linux_cache_cb(const struct rtnetlink_link_change *change, /* 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_dev **netdev_devp) { struct netdev_dev_linux *netdev_dev; int error; - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for %s devices should be empty", - name, class->type); - } - if (!cache_notifier_refcount) { error = rtnetlink_link_notifier_register(&netdev_linux_cache_notifier, netdev_linux_cache_cb, NULL); @@ -536,7 +541,8 @@ netdev_linux_create(const struct netdev_class *class, cache_notifier_refcount++; netdev_dev = xzalloc(sizeof *netdev_dev); - netdev_dev_init(&netdev_dev->netdev_dev, name, args, class); + netdev_dev->change_seq = 1; + netdev_dev_init(&netdev_dev->netdev_dev, name, class); *netdev_devp = &netdev_dev->netdev_dev; return 0; @@ -550,8 +556,7 @@ 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_dev **netdev_devp) { struct netdev_dev_linux *netdev_dev; struct tap_state *state; @@ -559,10 +564,6 @@ netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, struct ifreq ifr; int error; - if (!shash_is_empty(args)) { - VLOG_WARN("%s: arguments for TAP devices should be empty", name); - } - netdev_dev = xzalloc(sizeof *netdev_dev); state = &netdev_dev->state.tap; @@ -590,7 +591,7 @@ netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, goto error; } - netdev_dev_init(&netdev_dev->netdev_dev, name, args, &netdev_tap_class); + netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class); *netdev_devp = &netdev_dev->netdev_dev; return 0; @@ -636,8 +637,7 @@ netdev_linux_destroy(struct netdev_dev *netdev_dev_) } static int -netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, - struct netdev **netdevp) +netdev_linux_open(struct netdev_dev *netdev_dev_, struct netdev **netdevp) { struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); struct netdev_linux *netdev; @@ -673,53 +673,6 @@ netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, * directions appearing to be reversed. */ netdev->fd = netdev_dev->state.tap.fd; netdev_dev->state.tap.opened = true; - } else if (ethertype != NETDEV_ETH_TYPE_NONE) { - struct sockaddr_ll sll; - int protocol; - int ifindex; - - /* Create file descriptor. */ - protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL - : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 - : ethertype); - netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); - if (netdev->fd < 0) { - error = errno; - goto error; - } - - /* Set non-blocking mode. */ - error = set_nonblocking(netdev->fd); - if (error) { - goto error; - } - - /* Get ethernet device index. */ - error = get_ifindex(&netdev->netdev, &ifindex); - if (error) { - goto error; - } - - /* Bind to specific ethernet device. */ - memset(&sll, 0, sizeof sll); - sll.sll_family = AF_PACKET; - sll.sll_ifindex = ifindex; - if (bind(netdev->fd, - (struct sockaddr *) &sll, sizeof sll) < 0) { - error = errno; - VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_), - strerror(error)); - goto error; - } - - /* Between the socket() and bind() calls above, the socket receives all - * packets of the requested type on all system interfaces. We do not - * want to receive that data, but there is no way to avoid it. So we - * must now drain out the receive queue. */ - error = drain_rcvbuf(netdev->fd); - if (error) { - goto error; - } } *netdevp = &netdev->netdev; @@ -764,13 +717,68 @@ netdev_linux_enumerate(struct sset *sset) } } +static int +netdev_linux_listen(struct netdev *netdev_) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct sockaddr_ll sll; + int ifindex; + int error; + int fd; + + if (netdev->fd >= 0) { + return 0; + } + + /* Create file descriptor. */ + fd = socket(PF_PACKET, SOCK_RAW, 0); + if (fd < 0) { + error = errno; + VLOG_ERR("failed to create raw socket (%s)", strerror(error)); + goto error; + } + + /* Set non-blocking mode. */ + error = set_nonblocking(fd); + if (error) { + goto error; + } + + /* Get ethernet device index. */ + error = get_ifindex(&netdev->netdev, &ifindex); + if (error) { + goto error; + } + + /* Bind to specific ethernet device. */ + memset(&sll, 0, sizeof sll); + sll.sll_family = AF_PACKET; + sll.sll_ifindex = ifindex; + sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL); + if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) { + error = errno; + VLOG_ERR("%s: failed to bind raw socket (%s)", + netdev_get_name(netdev_), strerror(error)); + goto error; + } + + netdev->fd = fd; + return 0; + +error: + if (fd >= 0) { + close(fd); + } + return error; +} + static int netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); if (netdev->fd < 0) { - /* Device was opened with NETDEV_ETH_TYPE_NONE. */ + /* Device is not listening. */ return -EAGAIN; } @@ -843,6 +851,12 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) 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) { @@ -866,7 +880,7 @@ 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 @@ -1006,6 +1020,11 @@ netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) char *fn = NULL; int fd = -1; + if (netdev_dev->miimon_interval > 0) { + *carrier = netdev_dev->miimon; + return 0; + } + if (!(netdev_dev->cache_valid & VALID_CARRIER)) { char line[8]; int retval; @@ -1056,36 +1075,34 @@ exit: } 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 = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name); memcpy(data, &ifr.ifr_data, sizeof *data); return error; } static int -netdev_linux_get_miimon(const 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) { @@ -1115,6 +1132,69 @@ 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_dev_linux *netdev_dev; + + netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_)); + + interval = interval > 0 ? MAX(interval, 100) : 0; + if (netdev_dev->miimon_interval != interval) { + netdev_dev->miimon_interval = interval; + timer_set_expired(&netdev_dev->miimon_timer); + } + + return 0; +} + +static void +netdev_linux_miimon_run(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_dev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev_dev_linux *dev = node->data; + bool miimon; + + if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) { + continue; + } + + netdev_linux_get_miimon(dev->netdev_dev.name, &miimon); + if (miimon != dev->miimon) { + dev->miimon = miimon; + netdev_dev_linux_changed(dev); + } + + timer_set_duration(&dev->miimon_timer, dev->miimon_interval); + } + + shash_destroy(&device_shash); +} + +static void +netdev_linux_miimon_wait(void) +{ + struct shash device_shash; + struct shash_node *node; + + shash_init(&device_shash); + netdev_dev_get_devices(&netdev_linux_class, &device_shash); + SHASH_FOR_EACH (node, &device_shash) { + struct netdev_dev_linux *dev = node->data; + + if (dev->miimon_interval > 0) { + timer_wait(&dev->miimon_timer); + } + } + shash_destroy(&device_shash); +} + /* Check whether we can we use RTM_GETLINK to get network device statistics. * In pre-2.6.19 kernels, this was only available if wireless extensions were * enabled. */ @@ -2154,88 +2234,10 @@ netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, 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_dev_linux_cast(netdev_get_dev(netdev))->change_seq; } #define NETDEV_LINUX_CLASS(NAME, CREATE, ENUMERATE, SET_STATS) \ @@ -2248,6 +2250,7 @@ netdev_linux_poll_remove(struct netdev_notifier *notifier_) \ CREATE, \ netdev_linux_destroy, \ + NULL, /* get_config */ \ NULL, /* set_config */ \ \ netdev_linux_open, \ @@ -2255,6 +2258,7 @@ netdev_linux_poll_remove(struct netdev_notifier *notifier_) \ ENUMERATE, \ \ + netdev_linux_listen, \ netdev_linux_recv, \ netdev_linux_recv_wait, \ netdev_linux_drain, \ @@ -2267,7 +2271,7 @@ netdev_linux_poll_remove(struct netdev_notifier *notifier_) netdev_linux_get_mtu, \ netdev_linux_get_ifindex, \ netdev_linux_get_carrier, \ - netdev_linux_get_miimon, \ + netdev_linux_set_miimon_interval, \ netdev_linux_get_stats, \ SET_STATS, \ \ @@ -2297,8 +2301,7 @@ netdev_linux_poll_remove(struct netdev_notifier *notifier_) \ netdev_linux_update_flags, \ \ - netdev_linux_poll_add, \ - netdev_linux_poll_remove \ + netdev_linux_change_seq \ } const struct netdev_class netdev_linux_class = @@ -3976,6 +3979,8 @@ netdev_stats_to_rtnl_link_stats64(struct rtnl_link_stats64 *dst, const struct netdev_stats *src) { COPY_NETDEV_STATS; + dst->rx_compressed = 0; + dst->tx_compressed = 0; } /* Utility functions. */ @@ -4152,8 +4157,12 @@ get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); COVERAGE_INC(netdev_get_hwaddr); if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { - VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s", - netdev_name, strerror(errno)); + /* ENODEV probably means that a vif disappeared asynchronously and + * hasn't been removed from the database yet, so reduce the log level + * to INFO for that case. */ + VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR, + "ioctl(SIOCGIFHWADDR) on %s device failed: %s", + netdev_name, strerror(errno)); return errno; } hwaddr_family = ifr.ifr_hwaddr.sa_family; @@ -4210,6 +4219,51 @@ netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, } } +/* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If + * 'enable' is true, the bit is set. Otherwise, it is cleared. */ +int +netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, + const char *flag_name, bool enable) +{ + const char *netdev_name = netdev_get_name(netdev); + struct ethtool_value evalue; + uint32_t new_flags; + int error; + + memset(&evalue, 0, sizeof evalue); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); + if (error) { + return error; + } + + evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); + if (error) { + return error; + } + + memset(&evalue, 0, sizeof evalue); + error = netdev_linux_do_ethtool(netdev_name, + (struct ethtool_cmd *)&evalue, + ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); + if (error) { + return error; + } + + if (new_flags != evalue.data) { + VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " + "device %s failed", enable ? "enable" : "disable", + flag_name, netdev_name); + return EOPNOTSUPP; + } + + return 0; +} + static int netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, const char *cmd_name) @@ -4238,3 +4292,22 @@ netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, } return error; } + +/* Returns an AF_PACKET raw socket or a negative errno value. */ +static int +af_packet_sock(void) +{ + static int sock = INT_MIN; + + if (sock == INT_MIN) { + sock = socket(AF_PACKET, SOCK_RAW, 0); + if (sock >= 0) { + set_nonblocking(sock); + } else { + sock = -errno; + VLOG_ERR("failed to create packet socket: %s", strerror(errno)); + } + } + + return sock; +}