X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Frconn.c;h=c8bddab52efb104b752ef6e83adb90444cda096d;hb=refs%2Fheads%2Ffor-nox%2F0.4;hp=e5f377b9670fbe9ed39703db7c9c06c8cbc65af8;hpb=cd13e4cbc5f056a83e1e1542e0b34c0eb0ec08d1;p=sliver-openvswitch.git diff --git a/lib/rconn.c b/lib/rconn.c index e5f377b96..c8bddab52 100644 --- a/lib/rconn.c +++ b/lib/rconn.c @@ -39,7 +39,7 @@ #include #include #include "ofpbuf.h" -#include "openflow.h" +#include "openflow/openflow.h" #include "poll-loop.h" #include "sat-math.h" #include "timeval.h" @@ -92,6 +92,17 @@ struct rconn { time_t last_connected; unsigned int packets_sent; + /* In S_ACTIVE and S_IDLE, probably_admitted reports whether we believe + * that the peer has made a (positive) admission control decision on our + * connection. If we have not yet been (probably) admitted, then the + * connection does not reset the timer used for deciding whether the switch + * should go into fail-open mode. + * + * last_admitted reports the last time we believe such a positive admission + * control decision was made. */ + bool probably_admitted; + time_t last_admitted; + /* These values are simply for statistics reporting, not used directly by * anything internal to the rconn (or the secchan for that matter). */ unsigned int packets_received; @@ -131,6 +142,8 @@ static void disconnect(struct rconn *, int error); static void flush_queue(struct rconn *); static void question_connectivity(struct rconn *); static void copy_to_monitor(struct rconn *, const struct ofpbuf *); +static bool is_connected_state(enum state); +static bool is_admitted_msg(const struct ofpbuf *); /* Creates a new rconn, connects it (reliably) to 'name', and returns it. */ struct rconn * @@ -184,6 +197,9 @@ rconn_create(int probe_interval, int max_backoff) rc->packets_sent = 0; + rc->probably_admitted = false; + rc->last_admitted = time_now(); + rc->packets_received = 0; rc->n_attempted_connections = 0; rc->n_successful_connections = 0; @@ -248,10 +264,15 @@ void rconn_destroy(struct rconn *rc) { if (rc) { + size_t i; + free(rc->name); vconn_close(rc->vconn); flush_queue(rc); queue_destroy(&rc->txq); + for (i = 0; i < rc->n_monitors; i++) { + vconn_close(rc->monitors[i]); + } free(rc); } } @@ -359,10 +380,14 @@ run_ACTIVE(struct rconn *rc) { if (timed_out(rc)) { unsigned int base = MAX(rc->last_received, rc->state_entered); - rconn_send(rc, make_echo_request(), NULL); VLOG_DBG("%s: idle %u seconds, sending inactivity probe", rc->name, (unsigned int) (time_now() - base)); + + /* Ordering is important here: rconn_send() can transition to BACKOFF, + * and we don't want to transition back to IDLE if so, because then we + * can end up queuing a packet with vconn == NULL and then *boom*. */ state_transition(rc, S_IDLE); + rconn_send(rc, make_echo_request(), NULL); return; } @@ -436,6 +461,11 @@ rconn_recv(struct rconn *rc) int error = vconn_recv(rc->vconn, &buffer); if (!error) { copy_to_monitor(rc, buffer); + if (is_admitted_msg(buffer) + || time_now() - rc->last_connected >= 30) { + rc->probably_admitted = true; + rc->last_admitted = time_now(); + } rc->last_received = time_now(); rc->packets_received++; if (rc->state == S_IDLE) { @@ -558,15 +588,18 @@ rconn_is_alive(const struct rconn *rconn) bool rconn_is_connected(const struct rconn *rconn) { - return rconn->state & (S_ACTIVE | S_IDLE); + return is_connected_state(rconn->state); } -/* Returns 0 if 'rconn' is connected, otherwise the number of seconds that it - * has been disconnected. */ +/* Returns 0 if 'rconn' is connected. Otherwise, if 'rconn' is in a "failure + * mode" (that is, it is not connected), returns the number of seconds that it + * has been in failure mode, ignoring any times that it connected but the + * controller's admission control policy caused it to be quickly + * disconnected. */ int -rconn_disconnected_duration(const struct rconn *rconn) +rconn_failure_duration(const struct rconn *rconn) { - return rconn_is_connected(rconn) ? 0 : time_now() - rconn->last_received; + return rconn_is_connected(rconn) ? 0 : time_now() - rconn->last_admitted; } /* Returns the IP address of the peer, or 0 if the peer is not connected over @@ -700,7 +733,7 @@ disconnect(struct rconn *rc, int error) rc->name, strerror(error)); } else if (error == EOF) { if (rc->reliable) { - VLOG_WARN("%s: connection closed", rc->name); + VLOG_WARN("%s: connection closed by peer", rc->name); } } else { VLOG_WARN("%s: connection dropped", rc->name); @@ -773,6 +806,9 @@ timed_out(const struct rconn *rc) static void state_transition(struct rconn *rc, enum state state) { + if (is_connected_state(state) && !is_connected_state(rc->state)) { + rc->probably_admitted = false; + } if (rconn_is_connected(rc)) { rc->total_time_connected += elapsed_in_this_state(rc); } @@ -818,3 +854,27 @@ copy_to_monitor(struct rconn *rc, const struct ofpbuf *b) } ofpbuf_delete(clone); } + +static bool +is_connected_state(enum state state) +{ + return (state & (S_ACTIVE | S_IDLE)) != 0; +} + +static bool +is_admitted_msg(const struct ofpbuf *b) +{ + struct ofp_header *oh = b->data; + uint8_t type = oh->type; + return !(type < 32 + && (1u << type) & ((1u << OFPT_HELLO) | + (1u << OFPT_ERROR) | + (1u << OFPT_ECHO_REQUEST) | + (1u << OFPT_ECHO_REPLY) | + (1u << OFPT_VENDOR) | + (1u << OFPT_FEATURES_REQUEST) | + (1u << OFPT_FEATURES_REPLY) | + (1u << OFPT_GET_CONFIG_REQUEST) | + (1u << OFPT_GET_CONFIG_REPLY) | + (1u << OFPT_SET_CONFIG))); +}