X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fcfm.c;h=1b326252902c21b98f644fc247f317ec9b9d23db;hb=5767a79a40599e5e6e70662833b27ffdbdab7e37;hp=fc7486e2f905d6021122ddd6e2ff4c09850ed119;hpb=6fcdfcd065750940efd97b90f9746601d5b3f965;p=sliver-openvswitch.git diff --git a/lib/cfm.c b/lib/cfm.c index fc7486e2f..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. @@ -21,38 +21,231 @@ #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" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(cfm); -#define CCM_OPCODE 1 /* CFM message opcode meaning CCM. */ +#define CFM_MAX_RMPS 256 -struct cfm_internal { - struct cfm cfm; - uint32_t seq; /* The sequence number of our last CCM. */ +/* Ethernet destination address of CCM packets. */ +static const uint8_t eth_addr_ccm[6] = { 0x01, 0x80, 0xC2, 0x00, 0x00, 0x30 }; +static const uint8_t eth_addr_ccm_x[6] = { + 0x01, 0x23, 0x20, 0x00, 0x00, 0x30 +}; + +#define ETH_TYPE_CFM 0x8902 + +/* A 'ccm' represents a Continuity Check Message from the 802.1ag + * specification. Continuity Check Messages are broadcast periodically so that + * hosts can determine whom they have connectivity to. + * + * The minimum length of a CCM as specified by IEEE 802.1ag is 75 bytes. + * Previous versions of Open vSwitch generated 74-byte CCM messages, so we + * accept such messages too. */ +#define CCM_LEN 75 +#define CCM_ACCEPT_LEN 74 +#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; + ovs_be32 seq; + ovs_be16 mpid; + 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]; + + /* TLV space. */ + uint8_t end_tlv; +}); +BUILD_ASSERT_DECL(CCM_LEN == sizeof(struct ccm)); + +struct cfm { + 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 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. 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. */ + struct timer fault_timer; /* Check for faults when expired. */ + + struct hmap remote_mps; /* Remote MPs. */ + + /* Result of cfm_get_remote_mpids(). Updated only during fault check to + * 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; +}; - long long ccm_sent; /* The time we last sent a CCM. */ - long long fault_check; /* The time we last checked for faults. */ +/* Remote MPs represent foreign network entities that are configured to have + * the same MAID as this CFM instance. */ +struct remote_mp { + uint64_t mpid; /* The Maintenance Point ID of this 'remote_mp'. */ + struct hmap_node node; /* Node in 'remote_mps' map. */ + + bool recv; /* CCM was received since last fault check. */ + 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. */ - long long x_recv_time; }; +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 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(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) +{ + 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) OVS_REQUIRES(mutex) +{ + const char *ovs_md_name = "ovs"; + const char *ovs_ma_name = "ovs"; + uint8_t *ma_p; + size_t md_len, ma_len; + + memset(cfm->maid, 0, CCM_MAID_LEN); + + md_len = strlen(ovs_md_name); + ma_len = strlen(ovs_ma_name); + + 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. */ + memcpy(&cfm->maid[2], ovs_md_name, md_len); /* MD name. */ + + ma_p = cfm->maid + 2 + md_len; + ma_p[0] = 2; /* MA name string format. */ + ma_p[1] = ma_len; /* MA name size. */ + memcpy(&ma_p[2], ovs_ma_name, ma_len); /* MA name. */ +} + 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; @@ -60,10 +253,28 @@ 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) 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 + * an MP has a different interval, cfm_process_heartbeat will register it + * as a fault (likely due to a configuration error). Thus we can check all + * MPs at once making this quite a bit simpler. + * + * 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 @@ -80,31 +291,26 @@ ms_to_ccm_interval(int interval_ms) return 1; } -static struct cfm_internal * -cfm_to_internal(const struct cfm *cfm) -{ - return CONTAINER_OF(cfm, struct cfm_internal, cfm); -} - static uint32_t -hash_mpid(uint8_t mpid) +hash_mpid(uint64_t mpid) { - return hash_int(mpid, 0); + return hash_uint64(mpid); } static bool -cfm_is_valid_mpid(uint32_t mpid) +cfm_is_valid_mpid(bool extended, uint64_t mpid) { - /* 802.1ag specification requires MPIDs to be within the range [1, 8191] */ - return mpid >= 1 && mpid <= 8191; + /* 802.1ag specification requires MPIDs to be within the range [1, 8191]. + * In extended mode we relax this requirement. */ + return mpid >= 1 && (extended || mpid <= 8191); } static struct remote_mp * -lookup_remote_mp(const struct hmap *hmap, uint16_t mpid) +lookup_remote_mp(const struct cfm *cfm, uint64_t mpid) OVS_REQUIRES(mutex) { struct remote_mp *rmp; - HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), hmap) { + HMAP_FOR_EACH_IN_BUCKET (rmp, node, hash_mpid(mpid), &cfm->remote_mps) { if (rmp->mpid == mpid) { return rmp; } @@ -113,225 +319,410 @@ lookup_remote_mp(const struct hmap *hmap, uint16_t mpid) return NULL; } -/* Allocates a 'cfm' object. This object should have its 'mpid', 'maid', - * 'eth_src', and 'interval' filled out. When changes are made to the 'cfm' - * object, cfm_configure should be called before using it. */ +void +cfm_init(void) +{ + 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(void) +cfm_create(const struct netdev *netdev) OVS_EXCLUDED(mutex) { struct cfm *cfm; - struct cfm_internal *cfmi; - - cfmi = xzalloc(sizeof *cfmi); - cfm = &cfmi->cfm; - cfmi->x_recv_time = LLONG_MIN; + cfm = xzalloc(sizeof *cfm); + cfm->netdev = netdev_ref(netdev); + cfm->name = netdev_get_name(cfm->netdev); hmap_init(&cfm->remote_mps); + 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 cfm_internal *cfmi = cfm_to_internal(cfm); struct remote_mp *rmp, *rmp_next; if (!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); - free(cfmi); + netdev_close(cfm->netdev); + free(cfm->rmps_array); + + 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) { - long long now = time_msec(); - struct cfm_internal *cfmi = cfm_to_internal(cfm); - long long fault_interval; + ovs_mutex_lock(&mutex); + if (timer_expired(&cfm->fault_timer)) { + long long int interval = cfm_fault_interval(cfm); + struct remote_mp *rmp, *rmp_next; + 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); + + 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; + } - /* 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 - * an MP has a different interval, cfm_process_heartbeat will register it - * 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. */ - fault_interval = (cfmi->ccm_interval_ms * 7) / 2; - if (now >= cfmi->fault_check + fault_interval) { - bool fault; - struct remote_mp *rmp; + HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) { + if (!rmp->recv) { + 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->opup) { + rmp_set_opup = true; + } else { + rmp_set_opdown = true; + } + + cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid; + } + } - fault = now < cfmi->x_recv_time + fault_interval; + 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 |= CFM_FAULT_RECV; + } - HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) { - rmp->fault = rmp->fault || cfmi->fault_check > rmp->recv_time; - fault = rmp->fault || fault; + 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++; + } } - cfm->fault = fault; - cfmi->fault_check = now; + /* 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) { - struct cfm_internal *cfmi = cfm_to_internal(cfm); + bool ret; - return time_msec() >= cfmi->ccm_sent + cfmi->ccm_interval_ms; + ovs_mutex_lock(&mutex); + ret = timer_expired(&cfm->tx_timer); + ovs_mutex_unlock(&mutex); + return ret; } -/* Composes a CCM message into 'ccm'. Messages generated with this function +/* 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 ccm *ccm) +cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet, + uint8_t eth_src[ETH_ADDR_LEN]) OVS_EXCLUDED(mutex) { - struct cfm_internal *cfmi = cfm_to_internal(cfm); + 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); - cfmi->ccm_sent = time_msec(); + 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 = ofpbuf_l3(packet); ccm->mdlevel_version = 0; ccm->opcode = CCM_OPCODE; ccm->tlv_offset = 70; - ccm->seq = htonl(++cfmi->seq); - ccm->mpid = htons(cfmi->cfm.mpid); - ccm->flags = cfmi->ccm_interval; - memcpy(ccm->maid, cfmi->cfm.maid, sizeof ccm->maid); -} - -void -cfm_wait(struct cfm *cfm) -{ - long long wait; - struct cfm_internal *cfmi = cfm_to_internal(cfm); - - wait = MIN(cfmi->ccm_sent + cfmi->ccm_interval_ms, - cfmi->fault_check + cfmi->ccm_interval_ms * 4); - poll_timer_wait_until(wait); -} - -/* Should be called whenever a client of the cfm library changes the internals - * of 'cfm'. Returns true if 'cfm' is valid. */ -bool -cfm_configure(struct cfm *cfm) -{ - struct cfm_internal *cfmi; + ccm->seq = htonl(++cfm->seq); + ccm->flags = cfm->ccm_interval; + memcpy(ccm->maid, cfm->maid, sizeof ccm->maid); + memset(ccm->zero, 0, sizeof ccm->zero); + ccm->end_tlv = 0; + + atomic_read(&cfm->extended, &extended); + if (extended) { + ccm->mpid = htons(hash_mpid(cfm->mpid)); + ccm->mpid64 = htonll(cfm->mpid); + ccm->opdown = !cfm->opup; + } else { + ccm->mpid = htons(cfm->mpid); + ccm->mpid64 = htonll(0); + ccm->opdown = 0; + } - if (!cfm_is_valid_mpid(cfm->mpid) || !cfm->interval) { - return false; + if (cfm->ccm_interval == 0) { + ovs_assert(extended); + ccm->interval_ms_x = htons(cfm->ccm_interval_ms); + } else { + ccm->interval_ms_x = htons(0); } - cfmi = cfm_to_internal(cfm); - cfmi->ccm_interval = ms_to_ccm_interval(cfm->interval); - cfmi->ccm_interval_ms = ccm_interval_to_ms(cfmi->ccm_interval); + if (cfm->booted && hmap_is_empty(&cfm->remote_mps)) { + ccm->flags |= CCM_RDI_MASK; + } - /* Force a resend and check in case anything changed. */ - cfmi->ccm_sent = 0; - cfmi->fault_check = 0; - return true; + 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); } -/* Given an array of MPIDs, updates the 'remote_mps' map of 'cfm' to reflect - * it. Invalid MPIDs are skipped. */ void -cfm_update_remote_mps(struct cfm *cfm, const uint16_t *mpids, size_t n_mpids) +cfm_wait(struct cfm *cfm) OVS_EXCLUDED(mutex) { - size_t i; - struct hmap new_rmps; - struct remote_mp *rmp, *rmp_next; - - hmap_init(&new_rmps); - - for (i = 0; i < n_mpids; i++) { - uint16_t mpid = mpids[i]; - - if (!cfm_is_valid_mpid(mpid) - || lookup_remote_mp(&new_rmps, mpid)) { - continue; - } - - if ((rmp = lookup_remote_mp(&cfm->remote_mps, mpid))) { - hmap_remove(&cfm->remote_mps, &rmp->node); - } else { - rmp = xzalloc(sizeof *rmp); - rmp->mpid = mpid; - } + poll_timer_wait_until(cfm_wake_time(cfm)); +} - hmap_insert(&new_rmps, &rmp->node, hash_mpid(mpid)); - } - hmap_swap(&new_rmps, &cfm->remote_mps); +/* Returns the next cfm wakeup time. */ +long long int +cfm_wake_time(struct cfm *cfm) OVS_EXCLUDED(mutex) +{ + long long int retval; - HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &new_rmps) { - hmap_remove(&new_rmps, &rmp->node); - free(rmp); + if (!cfm) { + return LLONG_MAX; } - hmap_destroy(&new_rmps); + ovs_mutex_lock(&mutex); + retval = MIN(cfm->tx_timer.t, cfm->fault_timer.t); + ovs_mutex_unlock(&mutex); + return retval; } -/* Finds a 'remote_mp' with 'mpid' in 'cfm'. If no such 'remote_mp' exists - * returns NULL. */ -const struct remote_mp * -cfm_get_remote_mp(const struct cfm *cfm, uint16_t mpid) -{ - return lookup_remote_mp(&cfm->remote_mps, mpid); -} -/* Generates 'maid' from 'md_name' and 'ma_name'. A NULL parameter indicates - * the default should be used. Returns false if unsuccessful. */ +/* Configures 'cfm' with settings from 's'. */ bool -cfm_generate_maid(const char *md_name, const char *ma_name, - uint8_t maid[CCM_MAID_LEN]) +cfm_configure(struct cfm *cfm, const struct cfm_settings *s) + OVS_EXCLUDED(mutex) { - uint8_t *ma_p; - size_t md_len, ma_len; + uint8_t interval; + int interval_ms; - if (!md_name) { - md_name = "ovs"; + if (!cfm_is_valid_mpid(s->extended, s->mpid) || s->interval <= 0) { + return false; } - if (!ma_name) { - ma_name = "ovs"; - } + ovs_mutex_lock(&mutex); + cfm->mpid = s->mpid; + cfm->opup = s->opup; + interval = ms_to_ccm_interval(s->interval); + interval_ms = ccm_interval_to_ms(interval); - memset(maid, 0, CCM_MAID_LEN); + atomic_store(&cfm->check_tnl_key, s->check_tnl_key); + atomic_store(&cfm->extended, s->extended); - md_len = strlen(md_name); - ma_len = strlen(ma_name); + 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 (!md_len || !ma_len || md_len + ma_len + 4 > CCM_MAID_LEN) { - return false; + if (s->extended && s->demand) { + if (!cfm->demand) { + cfm->demand = true; + cfm->rx_packets = cfm_rx_packets(cfm); + } + } else { + cfm->demand = false; } - maid[0] = 4; /* MD name string format. */ - maid[1] = md_len; /* MD name size. */ - memcpy(&maid[2], md_name, md_len); /* MD name. */ + if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) { + cfm->ccm_interval = interval; + cfm->ccm_interval_ms = interval_ms; + + timer_set_expired(&cfm->tx_timer); + timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm)); + } - ma_p = maid + 2 + md_len; - ma_p[0] = 2; /* MA name string format. */ - ma_p[1] = ma_len; /* MA name size. */ - memcpy(&ma_p[2], ma_name, ma_len); /* MA name. */ + ovs_mutex_unlock(&mutex); return true; } -/* Returns true if the CFM library 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 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, eth_addr_ccm)); + && 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 @@ -339,79 +730,352 @@ cfm_should_process_flow(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; - uint16_t ccm_mpid; - uint8_t ccm_interval; - struct remote_mp *rmp; struct eth_header *eth; - struct cfm_internal *cfmi = cfm_to_internal(cfm); - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + ovs_mutex_lock(&mutex); - eth = p->l2; - ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_LEN); + 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, "Received an un-parseable 802.1ag CCM heartbeat."); - return; + VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.", + cfm->name); + goto out; } if (ccm->opcode != CCM_OPCODE) { - VLOG_INFO_RL(&rl, "Received an unsupported 802.1ag message. " - "(opcode %u)", ccm->opcode); - return; + VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. " + "(opcode %u)", cfm->name, ccm->opcode); + goto out; } + /* According to the 802.1ag specification, reception of a CCM with an + * incorrect ccm_interval, unexpected MAID, or unexpected MPID should + * trigger a fault. We ignore this requirement for several reasons. + * + * Faults can cause a controller or Open vSwitch to make potentially + * 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 + * unexpected CCMs. */ if (memcmp(ccm->maid, cfm->maid, sizeof ccm->maid)) { - cfmi->x_recv_time = time_msec(); - cfm->fault = true; - VLOG_WARN_RL(&rl, "Received unexpected remote MAID from MAC " - ETH_ADDR_FMT, ETH_ADDR_ARGS(eth->eth_src)); + 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 { - ccm_mpid = ntohs(ccm->mpid); - ccm_interval = ccm->flags & 0x7; + uint8_t ccm_interval = ccm->flags & 0x7; + bool ccm_rdi = ccm->flags & CCM_RDI_MASK; + uint16_t ccm_interval_ms_x = ntohs(ccm->interval_ms_x); - rmp = lookup_remote_mp(&cfm->remote_mps, ccm_mpid); + struct remote_mp *rmp; + uint64_t ccm_mpid; + uint32_t ccm_seq; + bool ccm_opdown; + bool extended; + enum cfm_fault_reason cfm_fault = 0; + + 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 unexpected interval" + " (%"PRIu8") from RMP %"PRIu64, cfm->name, + ccm_interval, ccm_mpid); + } + + if (extended && ccm_interval == 0 + && ccm_interval_ms_x != cfm->ccm_interval_ms) { + 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); + } + + rmp = lookup_remote_mp(cfm, ccm_mpid); + if (!rmp) { + if (hmap_count(&cfm->remote_mps) < CFM_MAX_RMPS) { + rmp = xzalloc(sizeof *rmp); + hmap_insert(&cfm->remote_mps, &rmp->node, hash_mpid(ccm_mpid)); + } else { + cfm_fault |= CFM_FAULT_OVERFLOW; + VLOG_WARN_RL(&rl, + "%s: dropped CCM with MPID %"PRIu64" from MAC " + ETH_ADDR_FMT, cfm->name, ccm_mpid, + ETH_ADDR_ARGS(eth->eth_src)); + } + } + + 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) { - rmp->recv_time = time_msec(); - rmp->fault = ccm_interval != cfmi->ccm_interval; - cfm->fault = rmp->fault || cfm->fault; - } else { - cfmi->x_recv_time = time_msec(); - cfm->fault = true; - VLOG_WARN_RL(&rl, "Received unexpected remote MPID %d from MAC " - ETH_ADDR_FMT, ccm_mpid, ETH_ADDR_ARGS(eth->eth_src)); + 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; + cfm->recv_fault |= cfm_fault; + rmp->seq = ccm_seq; + rmp->opup = !ccm_opdown; + rmp->last_rx = time_msec(); } } + +out: + ovs_mutex_unlock(&mutex); } +/* Returns and resets the 'cfm->status_changed'. */ +bool +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 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) +{ + 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_dump_ds(const struct cfm *cfm, struct ds *ds) +cfm_get_remote_mpids(const struct cfm *cfm, uint64_t **rmps, size_t *n_rmps) + OVS_EXCLUDED(mutex) +{ + 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) OVS_REQUIRES(mutex) +{ + struct cfm *cfm; + + HMAP_FOR_EACH_WITH_HASH (cfm, hmap_node, hash_string(name, 0), all_cfms) { + if (!strcmp(cfm->name, name)) { + return cfm; + } + } + return NULL; +} + +static void +cfm_print_details(struct ds *ds, struct cfm *cfm) OVS_REQUIRES(mutex) { - const struct cfm_internal *cfmi = cfm_to_internal(cfm); - long long int now = time_msec(); struct remote_mp *rmp; + bool extended; + int fault; - ds_put_format(ds, "MPID %"PRIu16": %s\n", cfm->mpid, - cfm->fault ? "fault" : ""); + atomic_read(&cfm->extended, &extended); - ds_put_format(ds, "\tinterval: %dms\n", cfmi->ccm_interval_ms); - ds_put_format(ds, "\ttime since CCM tx: %lldms\n", now - cfmi->ccm_sent); - ds_put_format(ds, "\ttime since fault check: %lldms\n", - now - cfmi->fault_check); + ds_put_format(ds, "---- %s ----\n", cfm->name); + 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 (cfmi->x_recv_time != LLONG_MIN) { - ds_put_format(ds, "\ttime since bad CCM rx: %lldms\n", - now - cfmi->x_recv_time); + 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"); + ds_put_format(ds, "\tinterval: %dms\n", cfm->ccm_interval_ms); + ds_put_format(ds, "\tnext CCM tx: %lldms\n", + timer_msecs_until_expired(&cfm->tx_timer)); + ds_put_format(ds, "\tnext fault check: %lldms\n", + timer_msecs_until_expired(&cfm->fault_timer)); - ds_put_cstr(ds, "\n"); HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) { - ds_put_format(ds, "Remote MPID %"PRIu16": %s\n", rmp->mpid, - rmp->fault ? "fault" : ""); - ds_put_format(ds, "\ttime since CCM rx: %lldms\n", - time_msec() - rmp->recv_time); + 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"); } } + +static void +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; + struct cfm *cfm; + + ovs_mutex_lock(&mutex); + if (argc > 1) { + cfm = cfm_find(argv[1]); + if (!cfm) { + 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) { + cfm_print_details(&ds, cfm); + } + } + + 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); +}