X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev-linux.c;h=9bdbbdf9f98e6751746e2ba77252730258e6b459;hb=da4a619179d6d6e9e6a82977c41cfcc2d1533ad8;hp=d146ccfcfe090e8ccf097d9fd7310bb2ae50291d;hpb=38e0065b1f25615bf69cfdc000d17935f99c022b;p=sliver-openvswitch.git diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index d146ccfcf..9bdbbdf9f 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -48,6 +48,7 @@ #include #include +#include "connectivity.h" #include "coverage.h" #include "dpif-linux.h" #include "dynamic-string.h" @@ -61,9 +62,11 @@ #include "netlink.h" #include "ofpbuf.h" #include "openflow/openflow.h" +#include "ovs-atomic.h" #include "packets.h" #include "poll-loop.h" #include "rtnetlink-link.h" +#include "seq.h" #include "shash.h" #include "socket-util.h" #include "sset.h" @@ -356,7 +359,6 @@ struct netdev_linux { struct ovs_mutex mutex; 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. */ @@ -402,6 +404,11 @@ struct netdev_rx_linux { * additional log messages. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); +/* Polling miimon status for all ports causes performance degradation when + * handling a large number of ports. If there are no devices using miimon, then + * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). */ +static atomic_int miimon_cnt = ATOMIC_VAR_INIT(0); + static void netdev_linux_run(void); static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, @@ -410,6 +417,9 @@ 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 *, unsigned int *flags); static int set_flags(const char *, unsigned int flags); +static int update_flags(struct netdev_linux *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) + OVS_REQUIRES(netdev->mutex); 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, @@ -420,6 +430,7 @@ 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 bool netdev_linux_miimon_enabled(void); static void netdev_linux_miimon_run(void); static void netdev_linux_miimon_wait(void); @@ -482,13 +493,24 @@ netdev_linux_notify_sock(void) return sock; } +static bool +netdev_linux_miimon_enabled(void) +{ + int miimon; + + atomic_read(&miimon_cnt, &miimon); + return miimon > 0; +} + static void netdev_linux_run(void) { struct nl_sock *sock; int error; - netdev_linux_miimon_run(); + if (netdev_linux_miimon_enabled()) { + netdev_linux_miimon_run(); + } sock = netdev_linux_notify_sock(); if (!sock) { @@ -550,7 +572,9 @@ netdev_linux_wait(void) { struct nl_sock *sock; - netdev_linux_miimon_wait(); + if (netdev_linux_miimon_enabled()) { + netdev_linux_miimon_wait(); + } sock = netdev_linux_notify_sock(); if (sock) { nl_sock_wait(sock, POLLIN); @@ -562,10 +586,7 @@ netdev_linux_changed(struct netdev_linux *dev, unsigned int ifi_flags, unsigned int mask) OVS_REQUIRES(dev->mutex) { - dev->change_seq++; - if (!dev->change_seq) { - dev->change_seq++; - } + seq_change(connectivity_seq_get()); if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { dev->carrier_resets++; @@ -616,8 +637,7 @@ netdev_linux_alloc(void) static void netdev_linux_common_construct(struct netdev_linux *netdev) { - ovs_mutex_init(&netdev->mutex, PTHREAD_MUTEX_NORMAL); - netdev->change_seq = 1; + ovs_mutex_init(&netdev->mutex); } /* Creates system and internal devices. */ @@ -708,6 +728,11 @@ netdev_linux_destruct(struct netdev *netdev_) close(netdev->tap_fd); } + if (netdev->miimon_interval > 0) { + int junk; + atomic_sub(&miimon_cnt, 1, &junk); + } + ovs_mutex_destroy(&netdev->mutex); } @@ -947,8 +972,8 @@ netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) } return errno; } else if (retval != size) { - VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " - "%zu) on %s", retval, size, netdev_get_name(netdev_)); + VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE"d bytes of " + "%"PRIuSIZE") on %s", retval, size, netdev_get_name(netdev_)); return EMSGSIZE; } else { return 0; @@ -979,7 +1004,7 @@ netdev_linux_set_etheraddr(struct netdev *netdev_, const uint8_t mac[ETH_ADDR_LEN]) { struct netdev_linux *netdev = netdev_linux_cast(netdev_); - struct netdev_saved_flags *sf = NULL; + enum netdev_flags old_flags = 0; int error; ovs_mutex_lock(&netdev->mutex); @@ -994,7 +1019,7 @@ netdev_linux_set_etheraddr(struct netdev *netdev_, /* Tap devices must be brought down before setting the address. */ if (is_tap_netdev(netdev_)) { - netdev_turn_flags_off(netdev_, NETDEV_UP, &sf); + update_flags(netdev, NETDEV_UP, 0, &old_flags); } error = set_etheraddr(netdev_get_name(netdev_), mac); if (!error || error == ENODEV) { @@ -1005,7 +1030,9 @@ netdev_linux_set_etheraddr(struct netdev *netdev_, } } - netdev_restore_flags(sf); + if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) { + update_flags(netdev, 0, NETDEV_UP, &old_flags); + } exit: ovs_mutex_unlock(&netdev->mutex); @@ -1036,21 +1063,16 @@ netdev_linux_get_etheraddr(const struct netdev *netdev_, return error; } -/* Returns the maximum size of transmitted (and received) packets on 'netdev', - * in bytes, not including the hardware header; thus, this is typically 1500 - * bytes for Ethernet devices. */ static int -netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) +netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; - ovs_mutex_lock(&netdev->mutex); if (!(netdev->cache_valid & VALID_MTU)) { struct ifreq ifr; netdev->netdev_mtu_error = af_inet_ifreq_ioctl( - netdev_get_name(netdev_), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); + netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); netdev->mtu = ifr.ifr_mtu; netdev->cache_valid |= VALID_MTU; } @@ -1059,6 +1081,21 @@ netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) if (!error) { *mtup = netdev->mtu; } + + return error; +} + +/* Returns the maximum size of transmitted (and received) packets on 'netdev', + * in bytes, not including the hardware header; thus, this is typically 1500 + * bytes for Ethernet devices. */ +static int +netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) +{ + struct netdev_linux *netdev = netdev_linux_cast(netdev_); + int error; + + ovs_mutex_lock(&netdev->mutex); + error = netdev_linux_get_mtu__(netdev, mtup); ovs_mutex_unlock(&netdev->mutex); return error; @@ -1207,6 +1244,14 @@ netdev_linux_set_miimon_interval(struct netdev *netdev_, ovs_mutex_lock(&netdev->mutex); interval = interval > 0 ? MAX(interval, 100) : 0; if (netdev->miimon_interval != interval) { + int junk; + + if (interval && !netdev->miimon_interval) { + atomic_add(&miimon_cnt, 1, &junk); + } else if (!interval && netdev->miimon_interval) { + atomic_sub(&miimon_cnt, 1, &junk); + } + netdev->miimon_interval = interval; timer_set_expired(&netdev->miimon_timer); } @@ -1557,7 +1602,6 @@ netdev_internal_set_stats(struct netdev *netdev, static void netdev_linux_read_features(struct netdev_linux *netdev) - OVS_REQUIRES(netdev->mutex) { struct ethtool_cmd ecmd; uint32_t speed; @@ -2086,35 +2130,35 @@ start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) return true; } +struct netdev_linux_queue_state { + unsigned int *queues; + size_t cur_queue; + size_t n_queues; +}; + static int -netdev_linux_dump_queues(const struct netdev *netdev_, - netdev_dump_queues_cb *cb, void *aux) +netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); + const struct netdev_linux *netdev = netdev_linux_cast(netdev_); int error; ovs_mutex_lock(&netdev->mutex); error = tc_query_qdisc(netdev_); if (!error) { if (netdev->tc->ops->class_get) { - struct tc_queue *queue, *next_queue; - struct smap details; - - smap_init(&details); - HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node, - &netdev->tc->queues) { - int retval; - - smap_clear(&details); - - retval = netdev->tc->ops->class_get(netdev_, queue, &details); - if (!retval) { - (*cb)(queue->queue_id, &details, aux); - } else { - error = retval; - } + struct netdev_linux_queue_state *state; + struct tc_queue *queue; + size_t i; + + *statep = state = xmalloc(sizeof *state); + state->n_queues = hmap_count(&netdev->tc->queues); + state->cur_queue = 0; + state->queues = xmalloc(state->n_queues * sizeof *state->queues); + + i = 0; + HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) { + state->queues[i++] = queue->queue_id; } - smap_destroy(&details); } else { error = EOPNOTSUPP; } @@ -2124,6 +2168,41 @@ netdev_linux_dump_queues(const struct netdev *netdev_, return error; } +static int +netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_, + unsigned int *queue_idp, struct smap *details) +{ + const struct netdev_linux *netdev = netdev_linux_cast(netdev_); + struct netdev_linux_queue_state *state = state_; + int error = EOF; + + ovs_mutex_lock(&netdev->mutex); + while (state->cur_queue < state->n_queues) { + unsigned int queue_id = state->queues[state->cur_queue++]; + struct tc_queue *queue = tc_find_queue(netdev_, queue_id); + + if (queue) { + *queue_idp = queue_id; + error = netdev->tc->ops->class_get(netdev_, queue, details); + break; + } + } + ovs_mutex_unlock(&netdev->mutex); + + return error; +} + +static int +netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED, + void *state_) +{ + struct netdev_linux_queue_state *state = state_; + + free(state->queues); + free(state); + return 0; +} + static int netdev_linux_dump_queue_stats(const struct netdev *netdev_, netdev_dump_queue_stats_cb *cb, void *aux) @@ -2227,14 +2306,14 @@ parse_if_inet6_line(const char *line, { uint8_t *s6 = in6->s6_addr; #define X8 "%2"SCNx8 - return sscanf(line, - " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 - "%*x %*x %*x %*x %16s\n", - &s6[0], &s6[1], &s6[2], &s6[3], - &s6[4], &s6[5], &s6[6], &s6[7], - &s6[8], &s6[9], &s6[10], &s6[11], - &s6[12], &s6[13], &s6[14], &s6[15], - ifname) == 17; + return ovs_scan(line, + " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 + "%*x %*x %*x %*x %16s\n", + &s6[0], &s6[1], &s6[2], &s6[3], + &s6[4], &s6[5], &s6[6], &s6[7], + &s6[8], &s6[9], &s6[10], &s6[11], + &s6[12], &s6[13], &s6[14], &s6[15], + ifname); } /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if @@ -2342,12 +2421,11 @@ netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, int refcnt, metric, mtu; unsigned int flags, use, window, irtt; - if (sscanf(line, - "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 - " %d %u %u\n", - iface, &dest, &gateway, &flags, &refcnt, - &use, &metric, &mask, &mtu, &window, &irtt) != 11) { - + if (!ovs_scan(line, + "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 + " %d %u %u\n", + iface, &dest, &gateway, &flags, &refcnt, + &use, &metric, &mask, &mtu, &window, &irtt)) { VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", fn, ln, line); continue; @@ -2461,6 +2539,9 @@ nd_to_iff_flags(enum netdev_flags nd) if (nd & NETDEV_PROMISC) { iff |= IFF_PROMISC; } + if (nd & NETDEV_LOOPBACK) { + iff |= IFF_LOOPBACK; + } return iff; } @@ -2474,41 +2555,43 @@ iff_to_nd_flags(int iff) if (iff & IFF_PROMISC) { nd |= NETDEV_PROMISC; } + if (iff & IFF_LOOPBACK) { + nd |= NETDEV_LOOPBACK; + } return nd; } static int -netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, - enum netdev_flags on, enum netdev_flags *old_flagsp) +update_flags(struct netdev_linux *netdev, enum netdev_flags off, + enum netdev_flags on, enum netdev_flags *old_flagsp) + OVS_REQUIRES(netdev->mutex) { - struct netdev_linux *netdev = netdev_linux_cast(netdev_); int old_flags, new_flags; int error = 0; - ovs_mutex_lock(&netdev->mutex); 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); + error = set_flags(netdev_get_name(&netdev->up), new_flags); + get_flags(&netdev->up, &netdev->ifi_flags); } - ovs_mutex_unlock(&netdev->mutex); return error; } -static unsigned int -netdev_linux_change_seq(const struct netdev *netdev_) +static int +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_); - unsigned int change_seq; + int error; ovs_mutex_lock(&netdev->mutex); - change_seq = netdev->change_seq; + error = update_flags(netdev, off, on, old_flagsp); ovs_mutex_unlock(&netdev->mutex); - return change_seq; + return error; } #define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, SET_STATS, \ @@ -2554,7 +2637,9 @@ netdev_linux_change_seq(const struct netdev *netdev_) netdev_linux_set_queue, \ netdev_linux_delete_queue, \ netdev_linux_get_queue_stats, \ - netdev_linux_dump_queues, \ + netdev_linux_queue_dump_start, \ + netdev_linux_queue_dump_next, \ + netdev_linux_queue_dump_done, \ netdev_linux_dump_queue_stats, \ \ netdev_linux_get_in4, \ @@ -2567,8 +2652,6 @@ netdev_linux_change_seq(const struct netdev *netdev_) \ netdev_linux_update_flags, \ \ - netdev_linux_change_seq, \ - \ netdev_linux_rx_alloc, \ netdev_linux_rx_construct, \ netdev_linux_rx_destruct, \ @@ -2690,7 +2773,7 @@ htb_setup_class__(struct netdev *netdev, unsigned int handle, int error; int mtu; - error = netdev_get_mtu(netdev, &mtu); + error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); if (error) { VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", netdev_get_name(netdev)); @@ -2786,9 +2869,10 @@ htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, } static void -htb_parse_qdisc_details__(struct netdev *netdev, +htb_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, struct htb_class *hc) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); const char *max_rate_s; max_rate_s = smap_get(details, "max-rate"); @@ -2796,7 +2880,8 @@ htb_parse_qdisc_details__(struct netdev *netdev, if (!hc->max_rate) { enum netdev_features current; - netdev_get_features(netdev, ¤t, NULL, NULL, NULL); + netdev_linux_read_features(netdev); + current = !netdev->get_features_error ? netdev->current : 0; hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; } hc->min_rate = hc->max_rate; @@ -2815,7 +2900,7 @@ htb_parse_class_details__(struct netdev *netdev, const char *priority_s = smap_get(details, "priority"); int mtu, error; - error = netdev_get_mtu(netdev, &mtu); + error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); if (error) { VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", netdev_get_name(netdev)); @@ -3263,9 +3348,10 @@ hfsc_query_class__(const struct netdev *netdev, unsigned int handle, } static void -hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details, +hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, struct hfsc_class *class) { + struct netdev_linux *netdev = netdev_linux_cast(netdev_); uint32_t max_rate; const char *max_rate_s; @@ -3275,7 +3361,8 @@ hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details, if (!max_rate) { enum netdev_features current; - netdev_get_features(netdev, ¤t, NULL, NULL, NULL); + netdev_linux_read_features(netdev); + current = !netdev->get_features_error ? netdev->current : 0; max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; } @@ -4454,25 +4541,25 @@ get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) if (++ln >= 3) { char devname[16]; #define X64 "%"SCNu64 - if (sscanf(line, - " %15[^:]:" - X64 X64 X64 X64 X64 X64 X64 "%*u" - X64 X64 X64 X64 X64 X64 X64 "%*u", - devname, - &stats->rx_bytes, - &stats->rx_packets, - &stats->rx_errors, - &stats->rx_dropped, - &stats->rx_fifo_errors, - &stats->rx_frame_errors, - &stats->multicast, - &stats->tx_bytes, - &stats->tx_packets, - &stats->tx_errors, - &stats->tx_dropped, - &stats->tx_fifo_errors, - &stats->collisions, - &stats->tx_carrier_errors) != 15) { + if (!ovs_scan(line, + " %15[^:]:" + X64 X64 X64 X64 X64 X64 X64 "%*u" + X64 X64 X64 X64 X64 X64 X64 "%*u", + devname, + &stats->rx_bytes, + &stats->rx_packets, + &stats->rx_errors, + &stats->rx_dropped, + &stats->rx_fifo_errors, + &stats->rx_frame_errors, + &stats->multicast, + &stats->tx_bytes, + &stats->tx_packets, + &stats->tx_errors, + &stats->tx_dropped, + &stats->tx_fifo_errors, + &stats->collisions, + &stats->tx_carrier_errors)) { VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln); } else if (!strcmp(devname, netdev_name)) { stats->rx_length_errors = UINT64_MAX;