X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=b5d30352379828d32a9f65de4009ac26a9bda503;hb=df2fa9b50e1c0e93c6be1d31a5d69eb4f077add3;hp=1aa909345f6c241e3865d5ec3e0e84f47580e773;hpb=1de0e8ae5c4a793edc4f033deefb185d933238e4;p=openvswitch diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 1aa90934..b5d30352 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -15,6 +15,9 @@ */ #include + +#include "netdev-linux.h" + #include #include #include @@ -65,6 +68,7 @@ #include "socket-util.h" #include "shash.h" #include "sset.h" +#include "timer.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(netdev_linux); @@ -343,6 +347,10 @@ struct netdev_dev_linux { struct shash_node *shash_node; unsigned int cache_valid; + 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. */ int ifindex; @@ -370,7 +378,6 @@ 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; @@ -408,6 +415,10 @@ 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 poll_notify(struct list *); +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) @@ -444,14 +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)); - } } /* Create rtnetlink socket. */ @@ -470,12 +473,14 @@ 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 @@ -678,7 +683,8 @@ netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, 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)); + netdev->fd = socket(PF_PACKET, SOCK_RAW, + (OVS_FORCE int) htons(protocol)); if (netdev->fd < 0) { error = errno; goto error; @@ -828,36 +834,55 @@ netdev_linux_drain(struct netdev *netdev_) static int netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) { - struct sockaddr_ll sll; - struct msghdr msg; - struct iovec iov; - int ifindex; - int error; + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + for (;;) { + ssize_t retval; - error = get_ifindex(netdev_, &ifindex); - if (error) { - return error; - } + if (netdev->fd < 0) { + /* Use our AF_PACKET socket to send to this device. */ + struct sockaddr_ll sll; + struct msghdr msg; + struct iovec iov; + int ifindex; + int error; + int sock; + + sock = af_packet_sock(); + if (sock < 0) { + return sock; + } + + error = get_ifindex(netdev_, &ifindex); + if (error) { + return error; + } - /* We don't bother setting most fields in sockaddr_ll because the kernel - * ignores them for SOCK_RAW. */ - memset(&sll, 0, sizeof sll); - sll.sll_family = AF_PACKET; - sll.sll_ifindex = ifindex; + /* We don't bother setting most fields in sockaddr_ll because the + * kernel ignores them for SOCK_RAW. */ + memset(&sll, 0, sizeof sll); + sll.sll_family = AF_PACKET; + sll.sll_ifindex = ifindex; - iov.iov_base = (void *) data; - iov.iov_len = size; + iov.iov_base = (void *) data; + iov.iov_len = size; - msg.msg_name = &sll; - msg.msg_namelen = sizeof sll; - msg.msg_iov = &iov; - msg.msg_iovlen = 1; - msg.msg_control = NULL; - msg.msg_controllen = 0; - msg.msg_flags = 0; + msg.msg_name = &sll; + msg.msg_namelen = sizeof sll; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + msg.msg_control = NULL; + msg.msg_controllen = 0; + msg.msg_flags = 0; + + retval = sendmsg(sock, &msg, 0); + } else { + /* Use the netdev's own fd to send to this device. This is + * essential for tap devices, because packets sent to a tap device + * with an AF_PACKET socket will loop back to be *received* again + * on the tap device. */ + retval = write(netdev->fd, data, size); + } - for (;;) { - ssize_t retval = sendmsg(af_packet_sock, &msg, 0); if (retval < 0) { /* The Linux AF_PACKET implementation never blocks waiting for room * for packets, instead returning ENOBUFS. Translate this into @@ -989,6 +1014,11 @@ 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; @@ -1039,36 +1069,34 @@ 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) { @@ -1098,6 +1126,75 @@ 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) { + struct list *list; + + dev->miimon = miimon; + list = shash_find_data(&netdev_linux_notifiers, + dev->netdev_dev.name); + if (list) { + poll_notify(list); + } + } + + 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. */ @@ -2000,7 +2097,7 @@ 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; @@ -2067,7 +2164,7 @@ netdev_linux_get_status(const struct netdev *netdev, struct shash *sh) * 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; @@ -2250,7 +2347,7 @@ 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, \ \ @@ -3911,7 +4008,57 @@ tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); } - + +/* 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 + +/* 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) +{ + COPY_NETDEV_STATS; +} + +/* 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; +} + +/* 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) +{ + COPY_NETDEV_STATS; + dst->rx_compressed = 0; + dst->tx_compressed = 0; +} /* Utility functions. */ @@ -3931,7 +4078,6 @@ get_stats_via_netlink(int ifindex, struct netdev_stats *stats) struct ofpbuf request; struct ofpbuf *reply; struct ifinfomsg *ifi; - const struct rtnl_link_stats *rtnl_stats; struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; int error; @@ -3959,28 +4105,7 @@ get_stats_via_netlink(int ifindex, struct netdev_stats *stats) return EPROTO; } - rtnl_stats = nl_attr_get(attrs[IFLA_STATS]); - stats->rx_packets = rtnl_stats->rx_packets; - stats->tx_packets = rtnl_stats->tx_packets; - stats->rx_bytes = rtnl_stats->rx_bytes; - stats->tx_bytes = rtnl_stats->tx_bytes; - stats->rx_errors = rtnl_stats->rx_errors; - stats->tx_errors = rtnl_stats->tx_errors; - stats->rx_dropped = rtnl_stats->rx_dropped; - stats->tx_dropped = rtnl_stats->tx_dropped; - stats->multicast = rtnl_stats->multicast; - stats->collisions = rtnl_stats->collisions; - stats->rx_length_errors = rtnl_stats->rx_length_errors; - stats->rx_over_errors = rtnl_stats->rx_over_errors; - stats->rx_crc_errors = rtnl_stats->rx_crc_errors; - stats->rx_frame_errors = rtnl_stats->rx_frame_errors; - stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors; - stats->rx_missed_errors = rtnl_stats->rx_missed_errors; - stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors; - stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors; - stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors; - stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors; - stats->tx_window_errors = rtnl_stats->tx_window_errors; + netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS])); ofpbuf_delete(reply); @@ -4195,3 +4320,22 @@ 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; +}