X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=lib%2Fnetdev-linux.c;h=cd104b320c8650c255a58990ef3386c2427795f7;hb=d84d4b88d26e3f37ce24f1d3eebe0d70ef264f73;hp=a27a625f8b1d4af811f732a3e9c77041c21ff8ec;hpb=f2cc621bacf07a48e65e163e3420c7594fd65cfd;p=openvswitch diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index a27a625f..cd104b32 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -15,6 +15,9 @@ */ #include + +#include "netdev-linux.h" + #include #include #include @@ -64,7 +67,7 @@ #include "rtnetlink-link.h" #include "socket-util.h" #include "shash.h" -#include "svec.h" +#include "sset.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(netdev_linux); @@ -368,8 +371,8 @@ struct netdev_linux { int fd; }; -/* An AF_INET socket (used for ioctl operations). */ -static int af_inet_sock = -1; +/* Sockets used for ioctl operations. */ +static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */ /* A Netlink routing socket that is not subscribed to any multicast groups. */ static struct nl_sock *rtnl_sock; @@ -407,6 +410,7 @@ 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 bool is_netdev_linux_class(const struct netdev_class *netdev_class) @@ -729,9 +733,9 @@ netdev_linux_close(struct netdev *netdev_) free(netdev); } -/* Initializes 'svec' with a list of the names of all known network devices. */ +/* Initializes 'sset' with a list of the names of all known network devices. */ static int -netdev_linux_enumerate(struct svec *svec) +netdev_linux_enumerate(struct sset *sset) { struct if_nameindex *names; @@ -740,7 +744,7 @@ netdev_linux_enumerate(struct svec *svec) size_t i; for (i = 0; names[i].if_name != NULL; i++) { - svec_add(svec, names[i].if_name); + sset_add(sset, names[i].if_name); } if_freenameindex(names); return 0; @@ -820,15 +824,54 @@ static int netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); + for (;;) { + ssize_t retval; - /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. - */ - if (netdev->fd < 0) { - return EPIPE; - } + if (netdev->fd < 0) { + /* Use our AF_PACKET socket to send to this device. */ + struct sockaddr_ll sll; + struct msghdr msg; + struct iovec iov; + int ifindex; + int error; + int sock; + + sock = af_packet_sock(); + if (sock < 0) { + return sock; + } + + error = get_ifindex(netdev_, &ifindex); + if (error) { + return error; + } + + /* We don't bother setting most fields in sockaddr_ll because the + * kernel ignores them for SOCK_RAW. */ + memset(&sll, 0, sizeof sll); + sll.sll_family = AF_PACKET; + sll.sll_ifindex = ifindex; + + iov.iov_base = (void *) data; + iov.iov_len = size; + + msg.msg_name = &sll; + msg.msg_namelen = sizeof sll; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + msg.msg_control = NULL; + msg.msg_controllen = 0; + msg.msg_flags = 0; + + retval = sendmsg(sock, &msg, 0); + } else { + /* Use the netdev's own fd to send to this device. This is + * essential for tap devices, because packets sent to a tap device + * with an AF_PACKET socket will loop back to be *received* again + * on the tap device. */ + retval = write(netdev->fd, data, size); + } - for (;;) { - ssize_t retval = write(netdev->fd, data, size); if (retval < 0) { /* The Linux AF_PACKET implementation never blocks waiting for room * for packets, instead returning ENOBUFS. Translate this into @@ -1115,9 +1158,9 @@ netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev) static void swap_uint64(uint64_t *a, uint64_t *b) { - *a ^= *b; - *b ^= *a; - *a ^= *b; + uint64_t tmp = *a; + *a = *b; + *b = tmp; } /* Retrieves current device stats for 'netdev'. */ @@ -1510,14 +1553,14 @@ netdev_linux_set_policing(struct netdev *netdev, static int netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, - struct svec *types) + struct sset *types) { const struct tc_ops **opsp; for (opsp = tcs; *opsp != NULL; opsp++) { const struct tc_ops *ops = *opsp; if (ops->tc_install && ops->ovs_name[0] != '\0') { - svec_add(types, ops->ovs_name); + sset_add(types, ops->ovs_name); } } return 0; @@ -1971,7 +2014,7 @@ 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; @@ -2038,7 +2081,7 @@ 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; @@ -2301,7 +2344,7 @@ htb_get__(const struct netdev *netdev) return CONTAINER_OF(netdev_dev->tc, struct htb, tc); } -static struct htb * +static void htb_install__(struct netdev *netdev, uint64_t max_rate) { struct netdev_dev_linux *netdev_dev = @@ -2313,8 +2356,6 @@ htb_install__(struct netdev *netdev, uint64_t max_rate) htb->max_rate = max_rate; netdev_dev->tc = &htb->tc; - - return htb; } /* Create an HTB qdisc. @@ -2497,13 +2538,10 @@ htb_parse_class_details__(struct netdev *netdev, return EINVAL; } - /* min-rate. Don't allow a min-rate below 1500 bytes/s. */ - if (!min_rate_s) { - /* min-rate is required. */ - return EINVAL; - } - hc->min_rate = strtoull(min_rate_s, NULL, 10) / 8; - hc->min_rate = MAX(hc->min_rate, 1500); + /* HTB requires at least an mtu sized min-rate to send any traffic even + * on uncongested links. */ + hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; + hc->min_rate = MAX(hc->min_rate, mtu); hc->min_rate = MIN(hc->min_rate, htb->max_rate); /* max-rate */ @@ -2603,12 +2641,11 @@ htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) struct ofpbuf msg; struct nl_dump dump; struct htb_class hc; - struct htb *htb; /* Get qdisc options. */ hc.max_rate = 0; htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); - htb = htb_install__(netdev, hc.max_rate); + htb_install__(netdev, hc.max_rate); /* Get queues. */ if (!start_queue_dump(netdev, &dump)) { @@ -2792,7 +2829,7 @@ hfsc_class_cast__(const struct tc_queue *queue) return CONTAINER_OF(queue, struct hfsc_class, tc_queue); } -static struct hfsc * +static void hfsc_install__(struct netdev *netdev, uint32_t max_rate) { struct netdev_dev_linux * netdev_dev; @@ -2803,8 +2840,6 @@ hfsc_install__(struct netdev *netdev, uint32_t max_rate) tc_init(&hfsc->tc, &tc_ops_hfsc); hfsc->max_rate = max_rate; netdev_dev->tc = &hfsc->tc; - - return hfsc; } static void @@ -2977,12 +3012,8 @@ hfsc_parse_class_details__(struct netdev *netdev, min_rate_s = shash_find_data(details, "min-rate"); max_rate_s = shash_find_data(details, "max-rate"); - if (!min_rate_s) { - return EINVAL; - } - - min_rate = strtoull(min_rate_s, NULL, 10) / 8; - min_rate = MAX(min_rate, 1500); + min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; + min_rate = MAX(min_rate, 1); min_rate = MIN(min_rate, hfsc->max_rate); max_rate = (max_rate_s @@ -3107,13 +3138,12 @@ static int hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) { struct ofpbuf msg; - struct hfsc *hfsc; struct nl_dump dump; struct hfsc_class hc; hc.max_rate = 0; hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); - hfsc = hfsc_install__(netdev, hc.max_rate); + hfsc_install__(netdev, hc.max_rate); if (!start_queue_dump(netdev, &dump)) { return ENODEV; @@ -3895,7 +3925,55 @@ 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; +} /* Utility functions. */ @@ -3915,7 +3993,6 @@ 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; @@ -3943,28 +4020,7 @@ 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); @@ -4179,3 +4235,22 @@ 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; +}