X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fcfm.c;h=1b326252902c21b98f644fc247f317ec9b9d23db;hb=5767a79a40599e5e6e70662833b27ffdbdab7e37;hp=8acbd0939aa77692b0428d1d5142e133e396a93e;hpb=75a4ead16d571a4e30cb4b54c5456e30a20c3d78;p=sliver-openvswitch.git diff --git a/lib/cfm.c b/lib/cfm.c index 8acbd0939..1b3262529 100644 --- a/lib/cfm.c +++ b/lib/cfm.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2010, 2011 Nicira Networks. + * Copyright (c) 2010, 2011, 2012, 2013, 2014 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -17,19 +17,22 @@ #include #include "cfm.h" -#include #include #include #include #include "byte-order.h" +#include "connectivity.h" #include "dynamic-string.h" #include "flow.h" #include "hash.h" #include "hmap.h" +#include "netdev.h" #include "ofpbuf.h" #include "packets.h" #include "poll-loop.h" +#include "random.h" +#include "seq.h" #include "timer.h" #include "timeval.h" #include "unixctl.h" @@ -59,41 +62,54 @@ static const uint8_t eth_addr_ccm_x[6] = { #define CCM_MAID_LEN 48 #define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */ #define CCM_RDI_MASK 0x80 +#define CFM_HEALTH_INTERVAL 6 + +OVS_PACKED( struct ccm { - uint8_t mdlevel_version; /* MD Level and Version */ - uint8_t opcode; - uint8_t flags; - uint8_t tlv_offset; + uint8_t mdlevel_version; /* MD Level and Version */ + uint8_t opcode; + uint8_t flags; + uint8_t tlv_offset; ovs_be32 seq; ovs_be16 mpid; - uint8_t maid[CCM_MAID_LEN]; + uint8_t maid[CCM_MAID_LEN]; /* Defined by ITU-T Y.1731 should be zero */ ovs_be16 interval_ms_x; /* Transmission interval in ms. */ ovs_be64 mpid64; /* MPID in extended mode. */ uint8_t opdown; /* Operationally down. */ - uint8_t zero[5]; + uint8_t zero[5]; /* TLV space. */ uint8_t end_tlv; -} __attribute__((packed)); +}); BUILD_ASSERT_DECL(CCM_LEN == sizeof(struct ccm)); struct cfm { - char *name; /* Name of this CFM object. */ + const char *name; /* Name of this CFM object. */ struct hmap_node hmap_node; /* Node in all_cfms list. */ + struct netdev *netdev; + uint64_t rx_packets; /* Packets received by 'netdev'. */ + uint64_t mpid; - bool extended; /* Extended mode. */ - bool fault; /* Indicates connectivity fault. */ - bool unexpected_recv; /* Received an unexpected CCM. */ + bool demand; /* Demand mode. */ + bool booted; /* A full fault interval has occurred. */ + enum cfm_fault_reason fault; /* Connectivity fault status. */ + enum cfm_fault_reason recv_fault; /* Bit mask of faults occurring on + receive. */ bool opup; /* Operational State. */ bool remote_opup; /* Remote Operational State. */ + int fault_override; /* Manual override of 'fault' status. + Ignored if negative. */ + uint32_t seq; /* The sequence number of our last CCM. */ uint8_t ccm_interval; /* The CCM transmission interval. */ int ccm_interval_ms; /* 'ccm_interval' in milliseconds. */ - uint16_t ccm_vlan; /* Vlan tag of CCM PDUs. */ + uint16_t ccm_vlan; /* Vlan tag of CCM PDUs. CFM_RANDOM_VLAN if + random. */ + uint8_t ccm_pcp; /* Priority of CCM PDUs. */ uint8_t maid[CCM_MAID_LEN]; /* The MAID of this CFM. */ struct timer tx_timer; /* Send CCM when expired. */ @@ -105,6 +121,27 @@ struct cfm { * avoid flapping. */ uint64_t *rmps_array; /* Cache of remote_mps. */ size_t rmps_array_len; /* Number of rmps in 'rmps_array'. */ + + int health; /* Percentage of the number of CCM frames + received. */ + int health_interval; /* Number of fault_intervals since health was + recomputed. */ + long long int last_tx; /* Last CCM transmission time. */ + + atomic_bool check_tnl_key; /* Verify the tunnel key of inbound packets? */ + atomic_bool extended; /* Extended mode. */ + struct ovs_refcount ref_cnt; + + uint64_t flap_count; /* Count the flaps since boot. */ + + /* True when the variables returned by cfm_get_*() are changed + * since last check. */ + bool status_changed; + + /* When 'cfm->demand' is set, at least one ccm is required to be received + * every 100 * cfm_interval. If ccm is not received within this interval, + * even if data packets are received, the cfm fault will be set. */ + struct timer demand_rx_ccm_t; }; /* Remote MPs represent foreign network entities that are configured to have @@ -114,25 +151,73 @@ struct remote_mp { struct hmap_node node; /* Node in 'remote_mps' map. */ bool recv; /* CCM was received since last fault check. */ - bool rdi; /* Remote Defect Indicator. Indicates remote_mp isn't - receiving CCMs that it's expecting to. */ bool opup; /* Operational State. */ + uint32_t seq; /* Most recently received sequence number. */ + uint8_t num_health_ccm; /* Number of received ccm frames every + CFM_HEALTH_INTERVAL * 'fault_interval'. */ + long long int last_rx; /* Last CCM reception time. */ + }; -static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); -static struct hmap all_cfms = HMAP_INITIALIZER(&all_cfms); +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 30); + +static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER; +static struct hmap all_cfms__ = HMAP_INITIALIZER(&all_cfms__); +static struct hmap *const all_cfms OVS_GUARDED_BY(mutex) = &all_cfms__; -static void cfm_unixctl_show(struct unixctl_conn *, const char *args, - void *aux); +static unixctl_cb_func cfm_unixctl_show; +static unixctl_cb_func cfm_unixctl_set_fault; + +static uint64_t +cfm_rx_packets(const struct cfm *cfm) OVS_REQUIRES(mutex) +{ + struct netdev_stats stats; + + if (!netdev_get_stats(cfm->netdev, &stats)) { + return stats.rx_packets; + } else { + return 0; + } +} static const uint8_t * -cfm_ccm_addr(const struct cfm *cfm) +cfm_ccm_addr(struct cfm *cfm) +{ + bool extended; + atomic_read(&cfm->extended, &extended); + return extended ? eth_addr_ccm_x : eth_addr_ccm; +} + +/* Returns the string representation of the given cfm_fault_reason 'reason'. */ +const char * +cfm_fault_reason_to_str(int reason) +{ + switch (reason) { +#define CFM_FAULT_REASON(NAME, STR) case CFM_FAULT_##NAME: return #STR; + CFM_FAULT_REASONS +#undef CFM_FAULT_REASON + default: return ""; + } +} + +static void +ds_put_cfm_fault(struct ds *ds, int fault) { - return cfm->extended ? eth_addr_ccm_x : eth_addr_ccm; + int i; + + for (i = 0; i < CFM_FAULT_N_REASONS; i++) { + int reason = 1 << i; + + if (fault & reason) { + ds_put_format(ds, "%s ", cfm_fault_reason_to_str(reason)); + } + } + + ds_chomp(ds, ' '); } static void -cfm_generate_maid(struct cfm *cfm) +cfm_generate_maid(struct cfm *cfm) OVS_REQUIRES(mutex) { const char *ovs_md_name = "ovs"; const char *ovs_ma_name = "ovs"; @@ -144,7 +229,7 @@ cfm_generate_maid(struct cfm *cfm) md_len = strlen(ovs_md_name); ma_len = strlen(ovs_ma_name); - assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN); + ovs_assert(md_len && ma_len && md_len + ma_len + 4 <= CCM_MAID_LEN); cfm->maid[0] = 4; /* MD name string format. */ cfm->maid[1] = md_len; /* MD name size. */ @@ -160,7 +245,7 @@ static int ccm_interval_to_ms(uint8_t interval) { switch (interval) { - case 0: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */ + case 0: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */ case 1: return 3; /* Not recommended due to timer resolution. */ case 2: return 10; /* Not recommended due to timer resolution. */ case 3: return 100; @@ -168,14 +253,14 @@ ccm_interval_to_ms(uint8_t interval) case 5: return 10000; case 6: return 60000; case 7: return 600000; - default: NOT_REACHED(); /* Explicitly not supported by 802.1ag. */ + default: OVS_NOT_REACHED(); /* Explicitly not supported by 802.1ag. */ } - NOT_REACHED(); + OVS_NOT_REACHED(); } static long long int -cfm_fault_interval(struct cfm *cfm) +cfm_fault_interval(struct cfm *cfm) OVS_REQUIRES(mutex) { /* According to the 802.1ag specification we should assume every other MP * with the same MAID has the same transmission interval that we have. If @@ -183,9 +268,13 @@ cfm_fault_interval(struct cfm *cfm) * as a fault (likely due to a configuration error). Thus we can check all * MPs at once making this quite a bit simpler. * - * According to the specification we should check when (ccm_interval_ms * - * 3.5)ms have passed. */ - return (cfm->ccm_interval_ms * 7) / 2; + * When cfm is not in demand mode, we check when (ccm_interval_ms * 3.5) ms + * have passed. When cfm is in demand mode, we check when + * (MAX(ccm_interval_ms, 500) * 3.5) ms have passed. This ensures that + * ovs-vswitchd has enough time to pull statistics from the datapath. */ + + return (MAX(cfm->ccm_interval_ms, cfm->demand ? 500 : cfm->ccm_interval_ms) + * 7) / 2; } static uint8_t @@ -205,7 +294,7 @@ ms_to_ccm_interval(int interval_ms) static uint32_t hash_mpid(uint64_t mpid) { - return hash_bytes(&mpid, sizeof mpid, 0); + return hash_uint64(mpid); } static bool @@ -217,7 +306,7 @@ cfm_is_valid_mpid(bool extended, uint64_t mpid) } static struct remote_mp * -lookup_remote_mp(const struct cfm *cfm, uint64_t mpid) +lookup_remote_mp(const struct cfm *cfm, uint64_t mpid) OVS_REQUIRES(mutex) { struct remote_mp *rmp; @@ -233,28 +322,42 @@ lookup_remote_mp(const struct cfm *cfm, uint64_t mpid) void cfm_init(void) { - unixctl_command_register("cfm/show", "[interface]", cfm_unixctl_show, + unixctl_command_register("cfm/show", "[interface]", 0, 1, cfm_unixctl_show, NULL); + unixctl_command_register("cfm/set-fault", "[interface] normal|false|true", + 1, 2, cfm_unixctl_set_fault, NULL); } /* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by * cfm_configure() before use. */ struct cfm * -cfm_create(const char *name) +cfm_create(const struct netdev *netdev) OVS_EXCLUDED(mutex) { struct cfm *cfm; cfm = xzalloc(sizeof *cfm); - cfm->name = xstrdup(name); + cfm->netdev = netdev_ref(netdev); + cfm->name = netdev_get_name(cfm->netdev); hmap_init(&cfm->remote_mps); - cfm_generate_maid(cfm); - hmap_insert(&all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0)); cfm->remote_opup = true; + cfm->fault_override = -1; + cfm->health = -1; + cfm->last_tx = 0; + cfm->flap_count = 0; + atomic_init(&cfm->extended, false); + atomic_init(&cfm->check_tnl_key, false); + ovs_refcount_init(&cfm->ref_cnt); + + ovs_mutex_lock(&mutex); + cfm_generate_maid(cfm); + hmap_insert(all_cfms, &cfm->hmap_node, hash_string(cfm->name, 0)); + ovs_mutex_unlock(&mutex); + return cfm; } void -cfm_destroy(struct cfm *cfm) +cfm_unref(struct cfm *cfm) OVS_EXCLUDED(mutex) { struct remote_mp *rmp, *rmp_next; @@ -262,97 +365,218 @@ cfm_destroy(struct cfm *cfm) return; } + if (ovs_refcount_unref(&cfm->ref_cnt) != 1) { + return; + } + + ovs_mutex_lock(&mutex); + hmap_remove(all_cfms, &cfm->hmap_node); + ovs_mutex_unlock(&mutex); + HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) { hmap_remove(&cfm->remote_mps, &rmp->node); free(rmp); } hmap_destroy(&cfm->remote_mps); - hmap_remove(&all_cfms, &cfm->hmap_node); + netdev_close(cfm->netdev); free(cfm->rmps_array); - free(cfm->name); + free(cfm); } +struct cfm * +cfm_ref(const struct cfm *cfm_) +{ + struct cfm *cfm = CONST_CAST(struct cfm *, cfm_); + if (cfm) { + ovs_refcount_ref(&cfm->ref_cnt); + } + return cfm; +} + +/* Records the status change and changes the global connectivity seq. */ +static void +cfm_status_changed(struct cfm *cfm) OVS_REQUIRES(mutex) +{ + seq_change(connectivity_seq_get()); + cfm->status_changed = true; +} + /* Should be run periodically to update fault statistics messages. */ void -cfm_run(struct cfm *cfm) +cfm_run(struct cfm *cfm) OVS_EXCLUDED(mutex) { + ovs_mutex_lock(&mutex); if (timer_expired(&cfm->fault_timer)) { long long int interval = cfm_fault_interval(cfm); struct remote_mp *rmp, *rmp_next; - - cfm->fault = cfm->unexpected_recv; - cfm->unexpected_recv = false; + enum cfm_fault_reason old_cfm_fault = cfm->fault; + uint64_t old_flap_count = cfm->flap_count; + int old_health = cfm->health; + size_t old_rmps_array_len = cfm->rmps_array_len; + bool old_rmps_deleted = false; + bool old_rmp_opup = cfm->remote_opup; + bool demand_override; + bool rmp_set_opup = false; + bool rmp_set_opdown = false; + + cfm->fault = cfm->recv_fault; + cfm->recv_fault = 0; cfm->rmps_array_len = 0; free(cfm->rmps_array); cfm->rmps_array = xmalloc(hmap_count(&cfm->remote_mps) * sizeof *cfm->rmps_array); - cfm->remote_opup = true; - HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) { + if (cfm->health_interval == CFM_HEALTH_INTERVAL) { + /* Calculate the cfm health of the interface. If the number of + * remote_mpids of a cfm interface is > 1, the cfm health is + * undefined. If the number of remote_mpids is 1, the cfm health is + * the percentage of the ccm frames received in the + * (CFM_HEALTH_INTERVAL * 3.5)ms, else it is 0. */ + if (hmap_count(&cfm->remote_mps) > 1) { + cfm->health = -1; + } else if (hmap_is_empty(&cfm->remote_mps)) { + cfm->health = 0; + } else { + int exp_ccm_recvd; + + rmp = CONTAINER_OF(hmap_first(&cfm->remote_mps), + struct remote_mp, node); + exp_ccm_recvd = (CFM_HEALTH_INTERVAL * 7) / 2; + /* Calculate the percentage of healthy ccm frames received. + * Since the 'fault_interval' is (3.5 * cfm_interval), and + * 1 CCM packet must be received every cfm_interval, + * the 'remote_mpid' health reports the percentage of + * healthy CCM frames received every + * 'CFM_HEALTH_INTERVAL'th 'fault_interval'. */ + cfm->health = (rmp->num_health_ccm * 100) / exp_ccm_recvd; + cfm->health = MIN(cfm->health, 100); + rmp->num_health_ccm = 0; + ovs_assert(cfm->health >= 0 && cfm->health <= 100); + } + cfm->health_interval = 0; + } + cfm->health_interval++; + + demand_override = false; + if (cfm->demand) { + uint64_t rx_packets = cfm_rx_packets(cfm); + demand_override = hmap_count(&cfm->remote_mps) == 1 + && rx_packets > cfm->rx_packets + && !timer_expired(&cfm->demand_rx_ccm_t); + cfm->rx_packets = rx_packets; + } + HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) { if (!rmp->recv) { - VLOG_DBG("%s: no CCM from RMP %"PRIu64" in the last %lldms", - cfm->name, rmp->mpid, interval); - hmap_remove(&cfm->remote_mps, &rmp->node); - free(rmp); + VLOG_INFO("%s: Received no CCM from RMP %"PRIu64" in the last" + " %lldms", cfm->name, rmp->mpid, + time_msec() - rmp->last_rx); + if (!demand_override) { + old_rmps_deleted = true; + hmap_remove(&cfm->remote_mps, &rmp->node); + free(rmp); + } } else { rmp->recv = false; - if (rmp->mpid == cfm->mpid) { - VLOG_WARN_RL(&rl,"%s: received CCM with local MPID" - " %"PRIu64, cfm->name, rmp->mpid); - cfm->fault = true; - } - - if (rmp->rdi) { - VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name, - rmp->mpid); - cfm->fault = true; - } - - if (!rmp->opup) { - cfm->remote_opup = rmp->opup; + if (rmp->opup) { + rmp_set_opup = true; + } else { + rmp_set_opdown = true; } cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid; } } + if (rmp_set_opdown) { + cfm->remote_opup = false; + } + else if (rmp_set_opup) { + cfm->remote_opup = true; + } + if (hmap_is_empty(&cfm->remote_mps)) { - cfm->fault = true; + cfm->fault |= CFM_FAULT_RECV; + } + + if (old_cfm_fault != cfm->fault) { + if (!VLOG_DROP_INFO(&rl)) { + struct ds ds = DS_EMPTY_INITIALIZER; + + ds_put_cstr(&ds, "from ["); + ds_put_cfm_fault(&ds, old_cfm_fault); + ds_put_cstr(&ds, "] to ["); + ds_put_cfm_fault(&ds, cfm->fault); + ds_put_char(&ds, ']'); + VLOG_INFO("%s: CFM faults changed %s.", cfm->name, ds_cstr(&ds)); + ds_destroy(&ds); + } + + /* If there is a flap, increments the counter. */ + if (old_cfm_fault == 0 || cfm->fault == 0) { + cfm->flap_count++; + } + } + + /* These variables represent the cfm session status, it is desirable + * to update them to database immediately after change. */ + if (old_health != cfm->health + || old_rmp_opup != cfm->remote_opup + || (old_rmps_array_len != cfm->rmps_array_len || old_rmps_deleted) + || old_cfm_fault != cfm->fault + || old_flap_count != cfm->flap_count) { + cfm_status_changed(cfm); } + cfm->booted = true; timer_set_duration(&cfm->fault_timer, interval); + VLOG_DBG("%s: new fault interval", cfm->name); } + ovs_mutex_unlock(&mutex); } /* Should be run periodically to check if the CFM module has a CCM message it * wishes to send. */ bool -cfm_should_send_ccm(struct cfm *cfm) +cfm_should_send_ccm(struct cfm *cfm) OVS_EXCLUDED(mutex) { - return timer_expired(&cfm->tx_timer); + bool ret; + + ovs_mutex_lock(&mutex); + ret = timer_expired(&cfm->tx_timer); + ovs_mutex_unlock(&mutex); + return ret; } /* Composes a CCM message into 'packet'. Messages generated with this function * should be sent whenever cfm_should_send_ccm() indicates. */ void cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet, - uint8_t eth_src[ETH_ADDR_LEN]) + uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex) { + uint16_t ccm_vlan; struct ccm *ccm; + bool extended; + ovs_mutex_lock(&mutex); timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms); eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM, sizeof *ccm); - if (cfm->ccm_vlan) { - eth_push_vlan(packet, htons(cfm->ccm_vlan)); + ccm_vlan = (cfm->ccm_vlan != CFM_RANDOM_VLAN + ? cfm->ccm_vlan + : random_uint16()); + ccm_vlan = ccm_vlan & VLAN_VID_MASK; + + if (ccm_vlan || cfm->ccm_pcp) { + uint16_t tci = ccm_vlan | (cfm->ccm_pcp << VLAN_PCP_SHIFT); + eth_push_vlan(packet, htons(ETH_TYPE_VLAN), htons(tci)); } - ccm = packet->l3; + ccm = ofpbuf_l3(packet); ccm->mdlevel_version = 0; ccm->opcode = CCM_OPCODE; ccm->tlv_offset = 70; @@ -362,7 +586,8 @@ cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet, memset(ccm->zero, 0, sizeof ccm->zero); ccm->end_tlv = 0; - if (cfm->extended) { + atomic_read(&cfm->extended, &extended); + if (extended) { ccm->mpid = htons(hash_mpid(cfm->mpid)); ccm->mpid64 = htonll(cfm->mpid); ccm->opdown = !cfm->opup; @@ -373,25 +598,56 @@ cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet, } if (cfm->ccm_interval == 0) { - assert(cfm->extended); + ovs_assert(extended); ccm->interval_ms_x = htons(cfm->ccm_interval_ms); + } else { + ccm->interval_ms_x = htons(0); } - if (hmap_is_empty(&cfm->remote_mps)) { + if (cfm->booted && hmap_is_empty(&cfm->remote_mps)) { ccm->flags |= CCM_RDI_MASK; } + + if (cfm->last_tx) { + long long int delay = time_msec() - cfm->last_tx; + if (delay > (cfm->ccm_interval_ms * 3 / 2)) { + VLOG_WARN("%s: long delay of %lldms (expected %dms) sending CCM" + " seq %"PRIu32, cfm->name, delay, cfm->ccm_interval_ms, + cfm->seq); + } + } + cfm->last_tx = time_msec(); + ovs_mutex_unlock(&mutex); } void -cfm_wait(struct cfm *cfm) +cfm_wait(struct cfm *cfm) OVS_EXCLUDED(mutex) { - timer_wait(&cfm->tx_timer); - timer_wait(&cfm->fault_timer); + poll_timer_wait_until(cfm_wake_time(cfm)); } + +/* Returns the next cfm wakeup time. */ +long long int +cfm_wake_time(struct cfm *cfm) OVS_EXCLUDED(mutex) +{ + long long int retval; + + if (!cfm) { + return LLONG_MAX; + } + + ovs_mutex_lock(&mutex); + retval = MIN(cfm->tx_timer.t, cfm->fault_timer.t); + ovs_mutex_unlock(&mutex); + return retval; +} + + /* Configures 'cfm' with settings from 's'. */ bool cfm_configure(struct cfm *cfm, const struct cfm_settings *s) + OVS_EXCLUDED(mutex) { uint8_t interval; int interval_ms; @@ -400,18 +656,31 @@ cfm_configure(struct cfm *cfm, const struct cfm_settings *s) return false; } + ovs_mutex_lock(&mutex); cfm->mpid = s->mpid; - cfm->extended = s->extended; cfm->opup = s->opup; interval = ms_to_ccm_interval(s->interval); interval_ms = ccm_interval_to_ms(interval); - cfm->ccm_vlan = s->ccm_vlan & VLAN_VID_MASK; - if (cfm->extended && interval_ms != s->interval) { + atomic_store(&cfm->check_tnl_key, s->check_tnl_key); + atomic_store(&cfm->extended, s->extended); + + cfm->ccm_vlan = s->ccm_vlan; + cfm->ccm_pcp = s->ccm_pcp & (VLAN_PCP_MASK >> VLAN_PCP_SHIFT); + if (s->extended && interval_ms != s->interval) { interval = 0; interval_ms = MIN(s->interval, UINT16_MAX); } + if (s->extended && s->demand) { + if (!cfm->demand) { + cfm->demand = true; + cfm->rx_packets = cfm_rx_packets(cfm); + } + } else { + cfm->demand = false; + } + if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) { cfm->ccm_interval = interval; cfm->ccm_interval_ms = interval_ms; @@ -420,15 +689,40 @@ cfm_configure(struct cfm *cfm, const struct cfm_settings *s) timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm)); } + ovs_mutex_unlock(&mutex); return true; } -/* Returns true if 'cfm' should process packets from 'flow'. */ +/* Must be called when the netdev owned by 'cfm' should change. */ +void +cfm_set_netdev(struct cfm *cfm, const struct netdev *netdev) + OVS_EXCLUDED(mutex) +{ + ovs_mutex_lock(&mutex); + if (cfm->netdev != netdev) { + netdev_close(cfm->netdev); + cfm->netdev = netdev_ref(netdev); + } + ovs_mutex_unlock(&mutex); +} + +/* Returns true if 'cfm' should process packets from 'flow'. Sets + * fields in 'wc' that were used to make the determination. */ bool -cfm_should_process_flow(const struct cfm *cfm, const struct flow *flow) +cfm_should_process_flow(const struct cfm *cfm_, const struct flow *flow, + struct flow_wildcards *wc) { + struct cfm *cfm = CONST_CAST(struct cfm *, cfm_); + bool check_tnl_key; + + atomic_read(&cfm->check_tnl_key, &check_tnl_key); + memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst); + if (check_tnl_key) { + memset(&wc->masks.tunnel.tun_id, 0xff, sizeof wc->masks.tunnel.tun_id); + } return (ntohs(flow->dl_type) == ETH_TYPE_CFM - && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm))); + && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm)) + && (!check_tnl_key || flow->tunnel.tun_id == htonll(0))); } /* Updates internal statistics relevant to packet 'p'. Should be called on @@ -436,23 +730,27 @@ cfm_should_process_flow(const struct cfm *cfm, const struct flow *flow) * cfm_should_process_flow. */ void cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p) + OVS_EXCLUDED(mutex) { struct ccm *ccm; struct eth_header *eth; - eth = p->l2; - ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_ACCEPT_LEN); + ovs_mutex_lock(&mutex); + + eth = ofpbuf_l2(p); + ccm = ofpbuf_at(p, (uint8_t *)ofpbuf_l3(p) - (uint8_t *)ofpbuf_data(p), + CCM_ACCEPT_LEN); if (!ccm) { VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.", cfm->name); - return; + goto out; } if (ccm->opcode != CCM_OPCODE) { VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. " "(opcode %u)", cfm->name, ccm->opcode); - return; + goto out; } /* According to the 802.1ag specification, reception of a CCM with an @@ -463,9 +761,9 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p) * expensive changes to the network topology. It seems prudent to trigger * them judiciously, especially when CFM is used to check slave status of * bonds. Furthermore, faults can be maliciously triggered by crafting - * invalid CCMs. */ + * unexpected CCMs. */ if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) { - cfm->unexpected_recv = true; + cfm->recv_fault |= CFM_FAULT_MAID; VLOG_WARN_RL(&rl, "%s: Received unexpected remote MAID from MAC " ETH_ADDR_FMT, cfm->name, ETH_ADDR_ARGS(eth->eth_src)); } else { @@ -475,25 +773,30 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p) struct remote_mp *rmp; uint64_t ccm_mpid; + uint32_t ccm_seq; bool ccm_opdown; + bool extended; + enum cfm_fault_reason cfm_fault = 0; - if (cfm->extended) { + atomic_read(&cfm->extended, &extended); + if (extended) { ccm_mpid = ntohll(ccm->mpid64); ccm_opdown = ccm->opdown; } else { ccm_mpid = ntohs(ccm->mpid); ccm_opdown = false; } + ccm_seq = ntohl(ccm->seq); if (ccm_interval != cfm->ccm_interval) { - VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid interval" + VLOG_WARN_RL(&rl, "%s: received a CCM with an unexpected interval" " (%"PRIu8") from RMP %"PRIu64, cfm->name, ccm_interval, ccm_mpid); } - if (cfm->extended && ccm_interval == 0 + if (extended && ccm_interval == 0 && ccm_interval_ms_x != cfm->ccm_interval_ms) { - VLOG_WARN_RL(&rl, "%s: received a CCM with an invalid extended" + VLOG_WARN_RL(&rl, "%s: received a CCM with an unexpected extended" " interval (%"PRIu16"ms) from RMP %"PRIu64, cfm->name, ccm_interval_ms_x, ccm_mpid); } @@ -501,10 +804,10 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p) rmp = lookup_remote_mp(cfm, ccm_mpid); if (!rmp) { if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) { - rmp = xmalloc(sizeof *rmp); + rmp = xzalloc(sizeof *rmp); hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(ccm_mpid)); } else { - cfm->unexpected_recv = true; + cfm_fault |= CFM_FAULT_OVERFLOW; VLOG_WARN_RL(&rl, "%s: dropped CCM with MPID %"PRIu64" from MAC " ETH_ADDR_FMT, cfm->name, ccm_mpid, @@ -512,54 +815,153 @@ cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p) } } + if (ccm_rdi) { + cfm_fault |= CFM_FAULT_RDI; + VLOG_DBG("%s: RDI bit flagged from RMP %"PRIu64, cfm->name, + ccm_mpid); + } + + VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")" + " (interval %"PRIu8") (RDI %s)", cfm->name, ccm_seq, + ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false"); + if (rmp) { + if (rmp->mpid == cfm->mpid) { + cfm_fault |= CFM_FAULT_LOOPBACK; + VLOG_WARN_RL(&rl,"%s: received CCM with local MPID" + " %"PRIu64, cfm->name, rmp->mpid); + } + + if (rmp->seq && ccm_seq != (rmp->seq + 1)) { + VLOG_WARN_RL(&rl, "%s: (mpid %"PRIu64") detected sequence" + " numbers which indicate possible connectivity" + " problems (previous %"PRIu32") (current %"PRIu32 + ")", cfm->name, ccm_mpid, rmp->seq, ccm_seq); + } + rmp->mpid = ccm_mpid; + if (!cfm_fault) { + rmp->num_health_ccm++; + if (cfm->demand) { + timer_set_duration(&cfm->demand_rx_ccm_t, + 100 * cfm->ccm_interval_ms); + } + } rmp->recv = true; - rmp->rdi = ccm_rdi; + cfm->recv_fault |= cfm_fault; + rmp->seq = ccm_seq; rmp->opup = !ccm_opdown; + rmp->last_rx = time_msec(); } - - VLOG_DBG("%s: received CCM (seq %"PRIu32") (mpid %"PRIu64")" - " (interval %"PRIu8") (RDI %s)", cfm->name, ntohl(ccm->seq), - ccm_mpid, ccm_interval, ccm_rdi ? "true" : "false"); } + +out: + ovs_mutex_unlock(&mutex); } -/* Gets the fault status of 'cfm'. Returns true when 'cfm' has detected - * connectivity problems, false otherwise. */ +/* Returns and resets the 'cfm->status_changed'. */ bool -cfm_get_fault(const struct cfm *cfm) +cfm_check_status_change(struct cfm *cfm) OVS_EXCLUDED(mutex) +{ + bool ret; + + ovs_mutex_lock(&mutex); + ret = cfm->status_changed; + cfm->status_changed = false; + ovs_mutex_unlock(&mutex); + + return ret; +} + +static int +cfm_get_fault__(const struct cfm *cfm) OVS_REQUIRES(mutex) { + if (cfm->fault_override >= 0) { + return cfm->fault_override ? CFM_FAULT_OVERRIDE : 0; + } return cfm->fault; } +/* Gets the fault status of 'cfm'. Returns a bit mask of 'cfm_fault_reason's + * indicating the cause of the connectivity fault, or zero if there is no + * fault. */ +int +cfm_get_fault(const struct cfm *cfm) OVS_EXCLUDED(mutex) +{ + int fault; + + ovs_mutex_lock(&mutex); + fault = cfm_get_fault__(cfm); + ovs_mutex_unlock(&mutex); + return fault; +} + +/* Gets the number of cfm fault flapping since start. */ +uint64_t +cfm_get_flap_count(const struct cfm *cfm) OVS_EXCLUDED(mutex) +{ + uint64_t flap_count; + ovs_mutex_lock(&mutex); + flap_count = cfm->flap_count; + ovs_mutex_unlock(&mutex); + return flap_count; +} + +/* Gets the health of 'cfm'. Returns an integer between 0 and 100 indicating + * the health of the link as a percentage of ccm frames received in + * CFM_HEALTH_INTERVAL * 'fault_interval' if there is only 1 remote_mpid, + * returns 0 if there are no remote_mpids, and returns -1 if there are more + * than 1 remote_mpids. */ +int +cfm_get_health(const struct cfm *cfm) OVS_EXCLUDED(mutex) +{ + int health; + + ovs_mutex_lock(&mutex); + health = cfm->health; + ovs_mutex_unlock(&mutex); + return health; +} + /* Gets the operational state of 'cfm'. 'cfm' is considered operationally down * if it has received a CCM with the operationally down bit set from any of its - * remote maintenance points. Returns true if 'cfm' is operationally up. False - * otherwise. */ -bool -cfm_get_opup(const struct cfm *cfm) + * remote maintenance points. Returns 1 if 'cfm' is operationally up, 0 if + * 'cfm' is operationally down, or -1 if 'cfm' has no operational state + * (because it isn't in extended mode). */ +int +cfm_get_opup(const struct cfm *cfm_) OVS_EXCLUDED(mutex) { - return cfm->remote_opup; + struct cfm *cfm = CONST_CAST(struct cfm *, cfm_); + bool extended; + int opup; + + ovs_mutex_lock(&mutex); + atomic_read(&cfm->extended, &extended); + opup = extended ? cfm->remote_opup : -1; + ovs_mutex_unlock(&mutex); + + return opup; } /* Populates 'rmps' with an array of remote maintenance points reachable by * 'cfm'. The number of remote maintenance points is written to 'n_rmps'. * 'cfm' retains ownership of the array written to 'rmps' */ void -cfm_get_remote_mpids(const struct cfm *cfm, const uint64_t **rmps, - size_t *n_rmps) +cfm_get_remote_mpids(const struct cfm *cfm, uint64_t **rmps, size_t *n_rmps) + OVS_EXCLUDED(mutex) { - *rmps = cfm->rmps_array; + ovs_mutex_lock(&mutex); + *rmps = xmemdup(cfm->rmps_array, cfm->rmps_array_len * sizeof **rmps); *n_rmps = cfm->rmps_array_len; + ovs_mutex_unlock(&mutex); } static struct cfm * -cfm_find(const char *name) +cfm_find(const char *name) OVS_REQUIRES(mutex) { struct cfm *cfm; - HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), &all_cfms) { + HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), all_cfms) { if (!strcmp(cfm->name, name)) { return cfm; } @@ -568,16 +970,31 @@ cfm_find(const char *name) } static void -cfm_print_details(struct ds *ds, const struct cfm *cfm) +cfm_print_details(struct ds *ds, struct cfm *cfm) OVS_REQUIRES(mutex) { struct remote_mp *rmp; + bool extended; + int fault; + + atomic_read(&cfm->extended, &extended); ds_put_format(ds, "---- %s ----\n", cfm->name); - ds_put_format(ds, "MPID %"PRIu64":%s%s%s\n", cfm->mpid, - cfm->extended ? " extended" : "", - cfm->fault ? " fault" : "", - cfm->unexpected_recv ? " unexpected_recv" : ""); + ds_put_format(ds, "MPID %"PRIu64":%s%s\n", cfm->mpid, + extended ? " extended" : "", + cfm->fault_override >= 0 ? " fault_override" : ""); + + fault = cfm_get_fault__(cfm); + if (fault) { + ds_put_cstr(ds, "\tfault: "); + ds_put_cfm_fault(ds, fault); + ds_put_cstr(ds, "\n"); + } + if (cfm->health == -1) { + ds_put_format(ds, "\taverage health: undefined\n"); + } else { + ds_put_format(ds, "\taverage health: %d\n", cfm->health); + } ds_put_format(ds, "\topstate: %s\n", cfm->opup ? "up" : "down"); ds_put_format(ds, "\tremote_opstate: %s\n", cfm->remote_opup ? "up" : "down"); @@ -588,9 +1005,7 @@ cfm_print_details(struct ds *ds, const struct cfm *cfm) timer_msecs_until_expired(&cfm->fault_timer)); HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) { - ds_put_format(ds, "Remote MPID %"PRIu64":%s\n", - rmp->mpid, - rmp->rdi ? " rdi" : ""); + ds_put_format(ds, "Remote MPID %"PRIu64"\n", rmp->mpid); ds_put_format(ds, "\trecv since check: %s\n", rmp->recv ? "true" : "false"); ds_put_format(ds, "\topstate: %s\n", rmp->opup? "up" : "down"); @@ -598,25 +1013,69 @@ cfm_print_details(struct ds *ds, const struct cfm *cfm) } static void -cfm_unixctl_show(struct unixctl_conn *conn, - const char *args, void *aux OVS_UNUSED) +cfm_unixctl_show(struct unixctl_conn *conn, int argc, const char *argv[], + void *aux OVS_UNUSED) OVS_EXCLUDED(mutex) { struct ds ds = DS_EMPTY_INITIALIZER; - const struct cfm *cfm; + struct cfm *cfm; - if (strlen(args)) { - cfm = cfm_find(args); + ovs_mutex_lock(&mutex); + if (argc > 1) { + cfm = cfm_find(argv[1]); if (!cfm) { - unixctl_command_reply(conn, 501, "no such CFM object"); - return; + unixctl_command_reply_error(conn, "no such CFM object"); + goto out; } cfm_print_details(&ds, cfm); } else { - HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) { + HMAP_FOR_EACH (cfm, hmap_node, all_cfms) { cfm_print_details(&ds, cfm); } } - unixctl_command_reply(conn, 200, ds_cstr(&ds)); + unixctl_command_reply(conn, ds_cstr(&ds)); ds_destroy(&ds); +out: + ovs_mutex_unlock(&mutex); +} + +static void +cfm_unixctl_set_fault(struct unixctl_conn *conn, int argc, const char *argv[], + void *aux OVS_UNUSED) OVS_EXCLUDED(mutex) +{ + const char *fault_str = argv[argc - 1]; + int fault_override; + struct cfm *cfm; + + ovs_mutex_lock(&mutex); + if (!strcasecmp("true", fault_str)) { + fault_override = 1; + } else if (!strcasecmp("false", fault_str)) { + fault_override = 0; + } else if (!strcasecmp("normal", fault_str)) { + fault_override = -1; + } else { + unixctl_command_reply_error(conn, "unknown fault string"); + goto out; + } + + if (argc > 2) { + cfm = cfm_find(argv[1]); + if (!cfm) { + unixctl_command_reply_error(conn, "no such CFM object"); + goto out; + } + cfm->fault_override = fault_override; + cfm_status_changed(cfm); + } else { + HMAP_FOR_EACH (cfm, hmap_node, all_cfms) { + cfm->fault_override = fault_override; + cfm_status_changed(cfm); + } + } + + unixctl_command_reply(conn, "OK"); + +out: + ovs_mutex_unlock(&mutex); }