/*
- * Copyright (c) 2010, 2011 Nicira Networks.
+ * Copyright (c) 2010, 2011, 2012 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <stdlib.h>
#include <string.h>
+#include "byte-order.h"
+#include "dynamic-string.h"
#include "flow.h"
#include "hash.h"
#include "hmap.h"
#include "ofpbuf.h"
#include "packets.h"
#include "poll-loop.h"
+#include "random.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 DEST_ADDR UINT64_C(0x0180C2000030) /* MD level 0 CCM destination. */
+#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
+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;
+} __attribute__((packed));
+BUILD_ASSERT_DECL(CCM_LEN == sizeof(struct ccm));
+
+struct cfm {
+ char *name; /* Name of this CFM object. */
+ struct hmap_node hmap_node; /* Node in all_cfms list. */
+
+ uint64_t mpid;
+ bool check_tnl_key; /* Verify the tunnel key of inbound packets? */
+ bool extended; /* Extended mode. */
+ bool booted; /* A full fault interval has occured. */
+ enum cfm_fault_reason fault; /* Connectivity fault status. */
+ enum cfm_fault_reason recv_fault; /* Bit mask of faults occuring 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. */
+};
+
+/* 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 ccm_sent; /* The time we last sent a CCM. */
- long long fault_check; /* The time we last checked for faults. */
};
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(20, 30);
+static struct hmap all_cfms = HMAP_INITIALIZER(&all_cfms);
+
+static unixctl_cb_func cfm_unixctl_show;
+static unixctl_cb_func cfm_unixctl_set_fault;
+
+static const uint8_t *
+cfm_ccm_addr(const struct cfm *cfm)
+{
+ return cfm->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 "<unknown>";
+ }
+}
+
+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)
+{
+ 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)
{
NOT_REACHED();
}
+static long long int
+cfm_fault_interval(struct cfm *cfm)
+{
+ /* 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. */
+ return (cfm->ccm_interval_ms * 7) / 2;
+}
+
static uint8_t
ms_to_ccm_interval(int interval_ms)
{
return 1;
}
-static struct cfm_internal *
-cfm_to_internal(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_bytes(&mpid, sizeof mpid, 0);
}
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)
{
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;
}
return NULL;
}
-static struct ofpbuf *
-compose_ccm(struct cfm_internal *cfmi)
+void
+cfm_init(void)
{
- struct ccm *ccm;
- struct ofpbuf *packet;
- struct eth_header *eth;
-
- packet = ofpbuf_new(ETH_HEADER_LEN + CCM_LEN + 2);
-
- ofpbuf_reserve(packet, 2);
-
- eth = ofpbuf_put_zeros(packet, ETH_HEADER_LEN);
- ccm = ofpbuf_put_zeros(packet, CCM_LEN);
-
- eth_addr_from_uint64(DEST_ADDR, eth->eth_dst);
- memcpy(eth->eth_src, cfmi->cfm.eth_src, sizeof eth->eth_src);
- eth->eth_type = htons(ETH_TYPE_CFM);
-
- 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);
- return packet;
+ 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. 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. */
+/* Allocates a 'cfm' object called 'name'. 'cfm' should be initialized by
+ * cfm_configure() before use. */
struct cfm *
-cfm_create(void)
+cfm_create(const char *name)
{
struct cfm *cfm;
- struct cfm_internal *cfmi;
-
- cfmi = xzalloc(sizeof *cfmi);
- cfm = &cfmi->cfm;
+ cfm = xzalloc(sizeof *cfm);
+ cfm->name = xstrdup(name);
hmap_init(&cfm->remote_mps);
- hmap_init(&cfm->x_remote_mps);
- hmap_init(&cfm->x_remote_maids);
+ 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;
return cfm;
}
cfm_destroy(struct cfm *cfm)
{
struct remote_mp *rmp, *rmp_next;
- struct remote_maid *rmaid, *rmaid_next;
if (!cfm) {
return;
free(rmp);
}
- HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->x_remote_mps) {
- hmap_remove(&cfm->x_remote_mps, &rmp->node);
- free(rmp);
- }
-
- HMAP_FOR_EACH_SAFE (rmaid, rmaid_next, node, &cfm->x_remote_maids) {
- hmap_remove(&cfm->x_remote_maids, &rmaid->node);
- free(rmaid);
- }
-
hmap_destroy(&cfm->remote_mps);
- hmap_destroy(&cfm->x_remote_mps);
- hmap_destroy(&cfm->x_remote_maids);
- free(cfm_to_internal(cfm));
+ hmap_remove(&all_cfms, &cfm->hmap_node);
+ free(cfm->rmps_array);
+ free(cfm->name);
+ free(cfm);
}
-/* Should be run periodically to update fault statistics and generate CCM
- * messages. If necessary, returns a packet which the caller is responsible
- * for sending, un-initing, and deallocating. Otherwise returns NULL. */
-struct ofpbuf *
+/* Should be run periodically to update fault statistics messages. */
+void
cfm_run(struct cfm *cfm)
{
- long long now = time_msec();
- struct cfm_internal *cfmi = cfm_to_internal(cfm);
-
- /* 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. We changed the multiplier to 4 to avoid messy
- * floating point arithmetic and add a bit of wiggle room. */
- if (now >= cfmi->fault_check + cfmi->ccm_interval_ms * 4) {
- bool fault;
+ if (timer_expired(&cfm->fault_timer)) {
+ long long int interval = cfm_fault_interval(cfm);
struct remote_mp *rmp, *rmp_next;
- struct remote_maid *rmaid, *rmaid_next;
-
- fault = false;
-
- HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
- rmp->fault = rmp->fault || cfmi->fault_check > rmp->recv_time;
- fault = rmp->fault || fault;
+ bool old_cfm_fault = cfm->fault;
+
+ 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;
+ 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++;
+
+ HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->remote_mps) {
- HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &cfm->x_remote_mps) {
- if (cfmi->fault_check > rmp->recv_time) {
- hmap_remove(&cfm->x_remote_mps, &rmp->node);
+ 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);
+ hmap_remove(&cfm->remote_mps, &rmp->node);
free(rmp);
+ } else {
+ rmp->recv = false;
+
+ if (!rmp->opup) {
+ cfm->remote_opup = rmp->opup;
+ }
+
+ cfm->rmps_array[cfm->rmps_array_len++] = rmp->mpid;
}
}
- HMAP_FOR_EACH_SAFE (rmaid, rmaid_next, node, &cfm->x_remote_maids) {
- if (cfmi->fault_check > rmaid->recv_time) {
- hmap_remove(&cfm->x_remote_maids, &rmaid->node);
- free(rmaid);
- }
+ if (hmap_is_empty(&cfm->remote_mps)) {
+ cfm->fault |= CFM_FAULT_RECV;
}
- fault = (fault || !hmap_is_empty(&cfm->x_remote_mps)
- || !hmap_is_empty(&cfm->x_remote_maids));
+ if (old_cfm_fault != cfm->fault && !VLOG_DROP_INFO(&rl)) {
+ struct ds ds = DS_EMPTY_INITIALIZER;
- cfm->fault = fault;
- cfmi->fault_check = now;
- }
+ 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 (now >= cfmi->ccm_sent + cfmi->ccm_interval_ms) {
- cfmi->ccm_sent = now;
- return compose_ccm(cfmi);
+ cfm->booted = true;
+ timer_set_duration(&cfm->fault_timer, interval);
+ VLOG_DBG("%s: new fault interval", cfm->name);
}
-
- return NULL;
-}
-
-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. */
+/* Should be run periodically to check if the CFM module has a CCM message it
+ * wishes to send. */
bool
-cfm_configure(struct cfm *cfm)
+cfm_should_send_ccm(struct cfm *cfm)
{
- struct cfm_internal *cfmi;
-
- if (!cfm_is_valid_mpid(cfm->mpid) || !cfm->interval) {
- return false;
- }
-
- 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);
-
- /* Force a resend and check in case anything changed. */
- cfmi->ccm_sent = 0;
- cfmi->fault_check = 0;
- return true;
+ return timer_expired(&cfm->tx_timer);
}
-/* Given an array of MPIDs, updates the 'remote_mps' map of 'cfm' to reflect
- * it. Invalid MPIDs are skipped. */
+/* Composes a CCM message into 'packet'. Messages generated with this function
+ * should be sent whenever cfm_should_send_ccm() indicates. */
void
-cfm_update_remote_mps(struct cfm *cfm, const uint16_t *mpids, size_t n_mpids)
+cfm_compose_ccm(struct cfm *cfm, struct ofpbuf *packet,
+ uint8_t eth_src[ETH_ADDR_LEN])
{
- size_t i;
- struct hmap new_rmps;
- struct remote_mp *rmp, *rmp_next;
+ uint16_t ccm_vlan;
+ struct ccm *ccm;
- hmap_init(&new_rmps);
+ timer_set_duration(&cfm->tx_timer, cfm->ccm_interval_ms);
+ eth_compose(packet, cfm_ccm_addr(cfm), eth_src, ETH_TYPE_CFM, sizeof *ccm);
- for (i = 0; i < n_mpids; i++) {
- uint16_t mpid = mpids[i];
+ ccm_vlan = (cfm->ccm_vlan != CFM_RANDOM_VLAN
+ ? cfm->ccm_vlan
+ : random_uint16());
+ ccm_vlan = ccm_vlan & VLAN_VID_MASK;
- 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 if ((rmp = lookup_remote_mp(&cfm->x_remote_mps, mpid))) {
- hmap_remove(&cfm->x_remote_mps, &rmp->node);
- } else {
- rmp = xzalloc(sizeof *rmp);
- rmp->mpid = mpid;
- }
+ if (ccm_vlan || cfm->ccm_pcp) {
+ uint16_t tci = ccm_vlan | (cfm->ccm_pcp << VLAN_PCP_SHIFT);
+ eth_push_vlan(packet, htons(tci));
+ }
- hmap_insert(&new_rmps, &rmp->node, hash_mpid(mpid));
+ ccm = packet->l3;
+ ccm->mdlevel_version = 0;
+ ccm->opcode = CCM_OPCODE;
+ ccm->tlv_offset = 70;
+ 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;
+
+ if (cfm->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;
}
- hmap_swap(&new_rmps, &cfm->remote_mps);
+ if (cfm->ccm_interval == 0) {
+ ovs_assert(cfm->extended);
+ ccm->interval_ms_x = htons(cfm->ccm_interval_ms);
+ } else {
+ ccm->interval_ms_x = htons(0);
+ }
- HMAP_FOR_EACH_SAFE (rmp, rmp_next, node, &new_rmps) {
- hmap_remove(&new_rmps, &rmp->node);
- free(rmp);
+ if (cfm->booted && hmap_is_empty(&cfm->remote_mps)) {
+ ccm->flags |= CCM_RDI_MASK;
}
- hmap_destroy(&new_rmps);
+ 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();
}
-/* 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)
+void
+cfm_wait(struct cfm *cfm)
{
- return lookup_remote_mp(&cfm->remote_mps, mpid);
+ timer_wait(&cfm->tx_timer);
+ timer_wait(&cfm->fault_timer);
}
-/* 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)
{
- 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";
+ cfm->mpid = s->mpid;
+ cfm->check_tnl_key = s->check_tnl_key;
+ 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;
+ cfm->ccm_pcp = s->ccm_pcp & (VLAN_PCP_MASK >> VLAN_PCP_SHIFT);
+ if (cfm->extended && interval_ms != s->interval) {
+ interval = 0;
+ interval_ms = MIN(s->interval, UINT16_MAX);
}
- memset(maid, 0, CCM_MAID_LEN);
-
- md_len = strlen(md_name);
- ma_len = strlen(ma_name);
+ if (interval != cfm->ccm_interval || interval_ms != cfm->ccm_interval_ms) {
+ cfm->ccm_interval = interval;
+ cfm->ccm_interval_ms = interval_ms;
- if (!md_len || !ma_len || md_len + ma_len + 4 > CCM_MAID_LEN) {
- return false;
+ timer_set_expired(&cfm->tx_timer);
+ timer_set_duration(&cfm->fault_timer, cfm_fault_interval(cfm));
}
- maid[0] = 4; /* MD name string format. */
- maid[1] = md_len; /* MD name size. */
- memcpy(&maid[2], md_name, md_len); /* MD name. */
-
- 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. */
return true;
}
-/* Returns true if the CFM library should process packets from 'flow'. */
+/* Returns true if 'cfm' should process packets from 'flow'. */
bool
-cfm_should_process_flow(const struct flow *flow)
+cfm_should_process_flow(const struct cfm *cfm, const struct flow *flow)
{
return (ntohs(flow->dl_type) == ETH_TYPE_CFM
- && eth_addr_to_uint64(flow->dl_dst) == DEST_ADDR);
+ && eth_addr_equals(flow->dl_dst, cfm_ccm_addr(cfm))
+ && (!cfm->check_tnl_key || flow->tunnel.tun_id == htonll(0)));
}
/* Updates internal statistics relevant to packet 'p'. Should be called on
cfm_process_heartbeat(struct cfm *cfm, const struct ofpbuf *p)
{
struct ccm *ccm;
- uint16_t ccm_mpid;
- uint32_t ccm_seq;
- uint8_t ccm_interval;
- struct remote_mp *rmp;
-
- struct cfm_internal *cfmi = cfm_to_internal(cfm);
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
+ struct eth_header *eth;
- ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_LEN);
+ eth = p->l2;
+ ccm = ofpbuf_at(p, (uint8_t *)p->l3 - (uint8_t *)p->data, CCM_ACCEPT_LEN);
if (!ccm) {
- VLOG_INFO_RL(&rl, "Received an un-parseable 802.1ag CCM heartbeat.");
+ VLOG_INFO_RL(&rl, "%s: Received an unparseable 802.1ag CCM heartbeat.",
+ cfm->name);
return;
}
if (ccm->opcode != CCM_OPCODE) {
- VLOG_INFO_RL(&rl, "Received an unsupported 802.1ag message. "
- "(opcode %u)", ccm->opcode);
+ VLOG_INFO_RL(&rl, "%s: Received an unsupported 802.1ag message. "
+ "(opcode %u)", cfm->name, ccm->opcode);
return;
}
+ /* 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)) {
- uint32_t hash;
- struct remote_maid *rmaid;
+ 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 {
+ 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);
+
+ struct remote_mp *rmp;
+ uint64_t ccm_mpid;
+ uint32_t ccm_seq;
+ bool ccm_opdown;
+ enum cfm_fault_reason cfm_fault = 0;
+
+ if (cfm->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) {
+ cfm_fault |= CFM_FAULT_INTERVAL;
+ VLOG_WARN_RL(&rl, "%s: received a CCM with an unexpected interval"
+ " (%"PRIu8") from RMP %"PRIu64, cfm->name,
+ ccm_interval, ccm_mpid);
+ }
- hash = hash_bytes(ccm->maid, sizeof ccm->maid, 0);
+ if (cfm->extended && ccm_interval == 0
+ && ccm_interval_ms_x != cfm->ccm_interval_ms) {
+ cfm_fault |= CFM_FAULT_INTERVAL;
+ 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);
+ }
- HMAP_FOR_EACH_IN_BUCKET (rmaid, node, hash, &cfm->x_remote_maids) {
- if (memcmp(rmaid->maid, ccm->maid, sizeof rmaid->maid) == 0) {
- rmaid->recv_time = time_msec();
- return;
+ 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));
}
}
- rmaid = xzalloc(sizeof *rmaid);
- rmaid->recv_time = time_msec();
- memcpy(rmaid->maid, ccm->maid, sizeof rmaid->maid);
- hmap_insert(&cfm->x_remote_maids, &rmaid->node, hash);
- return;
+ 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++;
+ }
+ rmp->recv = true;
+ cfm->recv_fault |= cfm_fault;
+ rmp->seq = ccm_seq;
+ rmp->opup = !ccm_opdown;
+ rmp->last_rx = time_msec();
+ }
+ }
+}
+
+/* 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)
+{
+ if (cfm->fault_override >= 0) {
+ return cfm->fault_override ? CFM_FAULT_OVERRIDE : 0;
+ }
+ return cfm->fault;
+}
+
+/* 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)
+{
+ return cfm->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)
+{
+ if (cfm->extended) {
+ return cfm->remote_opup;
+ } else {
+ return -1;
+ }
+}
+
+/* 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)
+{
+ *rmps = cfm->rmps_array;
+ *n_rmps = cfm->rmps_array_len;
+}
+
+static struct cfm *
+cfm_find(const char *name)
+{
+ 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, const struct cfm *cfm)
+{
+ struct remote_mp *rmp;
+ int fault;
+
+ ds_put_format(ds, "---- %s ----\n", cfm->name);
+ ds_put_format(ds, "MPID %"PRIu64":%s%s\n", cfm->mpid,
+ cfm->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");
+ 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));
+
+ HMAP_FOR_EACH (rmp, node, &cfm->remote_mps) {
+ 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)
+{
+ struct ds ds = DS_EMPTY_INITIALIZER;
+ const struct cfm *cfm;
+
+ if (argc > 1) {
+ cfm = cfm_find(argv[1]);
+ if (!cfm) {
+ unixctl_command_reply_error(conn, "no such CFM object");
+ return;
+ }
+ cfm_print_details(&ds, cfm);
+ } else {
+ HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) {
+ cfm_print_details(&ds, cfm);
+ }
}
- ccm_mpid = ntohs(ccm->mpid);
- ccm_seq = ntohl(ccm->seq);
- ccm_interval = ccm->flags & 0x7;
+ unixctl_command_reply(conn, ds_cstr(&ds));
+ ds_destroy(&ds);
+}
- rmp = lookup_remote_mp(&cfm->remote_mps, ccm_mpid);
+static void
+cfm_unixctl_set_fault(struct unixctl_conn *conn, int argc, const char *argv[],
+ void *aux OVS_UNUSED)
+{
+ const char *fault_str = argv[argc - 1];
+ int fault_override;
+ struct cfm *cfm;
- if (!rmp) {
- rmp = lookup_remote_mp(&cfm->x_remote_mps, ccm_mpid);
+ 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");
+ return;
}
- if (!rmp) {
- rmp = xzalloc(sizeof *rmp);
- rmp->mpid = ccm_mpid;
- hmap_insert(&cfm->x_remote_mps, &rmp->node, hash_mpid(ccm_mpid));
- cfm->fault = true;
+ if (argc > 2) {
+ cfm = cfm_find(argv[1]);
+ if (!cfm) {
+ unixctl_command_reply_error(conn, "no such CFM object");
+ return;
+ }
+ cfm->fault_override = fault_override;
+ } else {
+ HMAP_FOR_EACH (cfm, hmap_node, &all_cfms) {
+ cfm->fault_override = fault_override;
+ }
}
- rmp->recv_time = time_msec();
- rmp->fault = ccm_interval != cfmi->ccm_interval;
+ unixctl_command_reply(conn, "OK");
}