X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=2dd3c9f71287d0380b41fd3dd48fe48c89acedc9;hb=b772ded808ccc1b3cce2c10377a35637b1beb28c;hp=c58b44837e9b9271959adb8e8433515638f5cc23;hpb=154896e3b99978317d74b5c9847fe07ec01b54b3;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index c58b44837..2dd3c9f71 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -33,13 +33,13 @@ #include "ofp-print.h" #include "ofp-util.h" #include "ofpbuf.h" +#include "ofproto-provider.h" #include "openflow/nicira-ext.h" #include "openflow/openflow.h" #include "packets.h" #include "pinsched.h" #include "pktbuf.h" #include "poll-loop.h" -#include "private.h" #include "shash.h" #include "sset.h" #include "timeval.h" @@ -49,9 +49,7 @@ VLOG_DEFINE_THIS_MODULE(ofproto); -COVERAGE_DEFINE(ofproto_agg_request); COVERAGE_DEFINE(ofproto_error); -COVERAGE_DEFINE(ofproto_flows_req); COVERAGE_DEFINE(ofproto_flush); COVERAGE_DEFINE(ofproto_no_packet_in); COVERAGE_DEFINE(ofproto_packet_out); @@ -61,25 +59,95 @@ COVERAGE_DEFINE(ofproto_reinit_ports); COVERAGE_DEFINE(ofproto_uninstallable); COVERAGE_DEFINE(ofproto_update_port); +enum ofproto_state { + S_OPENFLOW, /* Processing OpenFlow commands. */ + S_FLUSH, /* Deleting all flow table rules. */ +}; + +enum ofoperation_type { + OFOPERATION_ADD, + OFOPERATION_DELETE, + OFOPERATION_MODIFY +}; + +/* A single OpenFlow request can execute any number of operations. The + * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the + * ofconn to which an error reply should be sent if necessary. + * + * ofproto initiates some operations internally. These operations are still + * assigned to groups but will not have an associated ofconn. */ +struct ofopgroup { + struct ofproto *ofproto; /* Owning ofproto. */ + struct list ofproto_node; /* In ofproto's "pending" list. */ + struct list ops; /* List of "struct ofoperation"s. */ + + /* Data needed to send OpenFlow reply on failure or to send a buffered + * packet on success. + * + * If list_is_empty(ofconn_node) then this ofopgroup never had an + * associated ofconn or its ofconn's connection dropped after it initiated + * the operation. In the latter case 'ofconn' is a wild pointer that + * refers to freed memory, so the 'ofconn' member must be used only if + * !list_is_empty(ofconn_node). + */ + struct list ofconn_node; /* In ofconn's list of pending opgroups. */ + struct ofconn *ofconn; /* ofconn for reply (but see note above). */ + struct ofp_header *request; /* Original request (truncated at 64 bytes). */ + uint32_t buffer_id; /* Buffer id from original request. */ + int error; /* 0 if no error yet, otherwise error code. */ +}; + +static struct ofopgroup *ofopgroup_create(struct ofproto *); +static struct ofopgroup *ofopgroup_create_for_ofconn(struct ofconn *, + const struct ofp_header *, + uint32_t buffer_id); +static void ofopgroup_submit(struct ofopgroup *); +static void ofopgroup_destroy(struct ofopgroup *); + +/* A single flow table operation. */ +struct ofoperation { + struct ofopgroup *group; /* Owning group. */ + struct list group_node; /* In ofopgroup's "ops" list. */ + struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */ + struct rule *rule; /* Rule being operated upon. */ + enum ofoperation_type type; /* Type of operation. */ + int status; /* -1 if pending, otherwise 0 or error code. */ + struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */ + union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */ + int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */ + ovs_be64 flow_cookie; /* Rule's old flow cookie. */ +}; + +static void ofoperation_create(struct ofopgroup *, struct rule *, + enum ofoperation_type); +static void ofoperation_destroy(struct ofoperation *); + static void ofport_destroy__(struct ofport *); static void ofport_destroy(struct ofport *); -static int rule_create(struct ofproto *, const struct cls_rule *, - const union ofp_action *, size_t n_actions, - uint16_t idle_timeout, uint16_t hard_timeout, - ovs_be64 flow_cookie, bool send_flow_removed, - struct rule **rulep); - static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); static void ofproto_destroy__(struct ofproto *); -static void ofproto_flush_flows__(struct ofproto *); static void ofproto_rule_destroy__(struct rule *); static void ofproto_rule_send_removed(struct rule *, uint8_t reason); -static void handle_openflow(struct ofconn *, struct ofpbuf *); +static void ofopgroup_destroy(struct ofopgroup *); + +static int add_flow(struct ofproto *, struct ofconn *, struct flow_mod *, + const struct ofp_header *); + +/* This return value tells handle_openflow() that processing of the current + * OpenFlow message must be postponed until some ongoing operations have + * completed. + * + * This particular value is a good choice because it is negative (so it won't + * collide with any errno value or any value returned by ofp_mkerr()) and large + * (so it won't accidentally collide with EOF or a negative errno value). */ +enum { OFPROTO_POSTPONE = -100000 }; + +static bool handle_openflow(struct ofconn *, struct ofpbuf *); static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); @@ -221,7 +289,9 @@ ofproto_create(const char *datapath_name, const char *datapath_type, struct ofproto **ofprotop) { const struct ofproto_class *class; + struct classifier *table; struct ofproto *ofproto; + int n_tables; int error; *ofprotop = NULL; @@ -252,19 +322,24 @@ ofproto_create(const char *datapath_name, const char *datapath_type, hmap_insert(&all_ofprotos, &ofproto->hmap_node, hash_string(ofproto->name, 0)); ofproto->datapath_id = 0; + ofproto_set_flow_eviction_threshold(ofproto, + OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT); ofproto->fallback_dpid = pick_fallback_dpid(); ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC); ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC); ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC); ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC); - ofproto->netdev_monitor = netdev_monitor_create(); hmap_init(&ofproto->ports); shash_init(&ofproto->port_by_name); - classifier_init(&ofproto->cls); + ofproto->tables = NULL; + ofproto->n_tables = 0; ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name); + ofproto->state = S_OPENFLOW; + list_init(&ofproto->pending); + hmap_init(&ofproto->deletions); - error = ofproto->ofproto_class->construct(ofproto); + error = ofproto->ofproto_class->construct(ofproto, &n_tables); if (error) { VLOG_ERR("failed to open datapath %s: %s", datapath_name, strerror(error)); @@ -272,6 +347,13 @@ ofproto_create(const char *datapath_name, const char *datapath_type, return error; } + assert(n_tables >= 1 && n_tables <= 255); + ofproto->n_tables = n_tables; + ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables); + OFPROTO_FOR_EACH_TABLE (table, ofproto) { + classifier_init(table); + } + ofproto->datapath_id = pick_datapath_id(ofproto); VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id); init_ports(ofproto); @@ -335,6 +417,18 @@ ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id) connmgr_set_in_band_queue(ofproto->connmgr, queue_id); } +/* Sets the number of flows at which eviction from the kernel flow table + * will occur. */ +void +ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold) +{ + if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) { + ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN; + } else { + ofproto->flow_eviction_threshold = threshold; + } +} + void ofproto_set_desc(struct ofproto *p, const char *mfr_desc, const char *hw_desc, @@ -430,20 +524,19 @@ ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port) { struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); if (ofport && ofproto->ofproto_class->set_cfm) { - ofproto->ofproto_class->set_cfm(ofport, NULL, NULL, 0); + ofproto->ofproto_class->set_cfm(ofport, NULL); } } /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes - * basic configuration from the configuration members in 'cfm', and the set of - * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'. - * Ignores the statistics members of 'cfm'. + * basic configuration from the configuration members in 'cfm', and the remote + * maintenance point ID from remote_mpid. Ignores the statistics members of + * 'cfm'. * * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */ void ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port, - const struct cfm *cfm, - const uint16_t *remote_mps, size_t n_remote_mps) + const struct cfm_settings *s) { struct ofport *ofport; int error; @@ -455,9 +548,11 @@ ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port, return; } + /* XXX: For configuration simplicity, we only support one remote_mpid + * outside of the CFM module. It's not clear if this is the correct long + * term solution or not. */ error = (ofproto->ofproto_class->set_cfm - ? ofproto->ofproto_class->set_cfm(ofport, cfm, - remote_mps, n_remote_mps) + ? ofproto->ofproto_class->set_cfm(ofport, s) : EOPNOTSUPP); if (error) { VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)", @@ -466,22 +561,6 @@ ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port, } } -/* Returns the connectivity fault management object associated with 'ofp_port' - * within 'ofproto', or a null pointer if 'ofproto' does not have a port - * 'ofp_port' or if that port does not have CFM configured. The caller must - * not modify or destroy the returned object. */ -const struct cfm * -ofproto_port_get_cfm(struct ofproto *ofproto, uint16_t ofp_port) -{ - struct ofport *ofport; - const struct cfm *cfm; - - ofport = ofproto_get_port(ofproto, ofp_port); - return (ofport - && ofproto->ofproto_class->get_cfm - && !ofproto->ofproto_class->get_cfm(ofport, &cfm)) ? cfm : NULL; -} - /* Checks the status of LACP negotiation for 'ofp_port' within ofproto. * Returns 1 if LACP partner information for 'ofp_port' is up-to-date, * 0 if LACP partner information is not current (generally indicating a @@ -585,22 +664,60 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) connmgr_get_snoops(ofproto->connmgr, snoops); } +static void +ofproto_flush__(struct ofproto *ofproto) +{ + struct classifier *table; + struct ofopgroup *group; + + if (ofproto->ofproto_class->flush) { + ofproto->ofproto_class->flush(ofproto); + } + + group = ofopgroup_create(ofproto); + OFPROTO_FOR_EACH_TABLE (table, ofproto) { + struct rule *rule, *next_rule; + struct cls_cursor cursor; + + cls_cursor_init(&cursor, table, NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { + if (!rule->pending) { + ofoperation_create(group, rule, OFOPERATION_DELETE); + classifier_remove(table, &rule->cr); + ofproto->ofproto_class->rule_destruct(rule); + } + } + } + ofopgroup_submit(group); +} + static void ofproto_destroy__(struct ofproto *ofproto) { + struct classifier *table; + + assert(list_is_empty(&ofproto->pending)); + connmgr_destroy(ofproto->connmgr); hmap_remove(&all_ofprotos, &ofproto->hmap_node); free(ofproto->name); + free(ofproto->type); free(ofproto->mfr_desc); free(ofproto->hw_desc); free(ofproto->sw_desc); free(ofproto->serial_desc); free(ofproto->dp_desc); - netdev_monitor_destroy(ofproto->netdev_monitor); hmap_destroy(&ofproto->ports); shash_destroy(&ofproto->port_by_name); - classifier_destroy(&ofproto->cls); + + OFPROTO_FOR_EACH_TABLE (table, ofproto) { + assert(classifier_is_empty(table)); + classifier_destroy(table); + } + free(ofproto->tables); + + hmap_destroy(&ofproto->deletions); ofproto->ofproto_class->dealloc(ofproto); } @@ -614,9 +731,8 @@ ofproto_destroy(struct ofproto *p) return; } - ofproto_flush_flows__(p); + ofproto_flush__(p); HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { - hmap_remove(&p->ports, &ofport->hmap_node); ofport_destroy(ofport); } @@ -653,6 +769,7 @@ process_port_change(struct ofproto *ofproto, int error, char *devname) int ofproto_run(struct ofproto *p) { + struct ofport *ofport; char *devname; int error; @@ -672,12 +789,33 @@ ofproto_run(struct ofproto *p) process_port_change(p, error, devname); } } - while ((error = netdev_monitor_poll(p->netdev_monitor, - &devname)) != EAGAIN) { - process_port_change(p, error, devname); + + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + unsigned int change_seq = netdev_change_seq(ofport->netdev); + if (ofport->change_seq != change_seq) { + ofport->change_seq = change_seq; + update_port(p, netdev_get_name(ofport->netdev)); + } } - connmgr_run(p->connmgr, handle_openflow); + + switch (p->state) { + case S_OPENFLOW: + connmgr_run(p->connmgr, handle_openflow); + break; + + case S_FLUSH: + connmgr_run(p->connmgr, NULL); + ofproto_flush__(p); + if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + connmgr_flushed(p->connmgr); + p->state = S_OPENFLOW; + } + break; + + default: + NOT_REACHED(); + } return 0; } @@ -685,12 +823,31 @@ ofproto_run(struct ofproto *p) void ofproto_wait(struct ofproto *p) { + struct ofport *ofport; + p->ofproto_class->wait(p); if (p->ofproto_class->port_poll_wait) { p->ofproto_class->port_poll_wait(p); } - netdev_monitor_poll_wait(p->netdev_monitor); - connmgr_wait(p->connmgr); + + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + if (ofport->change_seq != netdev_change_seq(ofport->netdev)) { + poll_immediate_wake(); + } + } + + switch (p->state) { + case S_OPENFLOW: + connmgr_wait(p->connmgr, true); + break; + + case S_FLUSH: + connmgr_wait(p->connmgr, false); + if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + poll_immediate_wake(); + } + break; + } } bool @@ -709,16 +866,7 @@ ofproto_get_ofproto_controller_info(const struct ofproto *ofproto, void ofproto_free_ofproto_controller_info(struct shash *info) { - struct shash_node *node; - - SHASH_FOR_EACH (node, info) { - struct ofproto_controller_info *cinfo = node->data; - while (cinfo->pairs.n) { - free((char *) cinfo->pairs.values[--cinfo->pairs.n]); - } - free(cinfo); - } - shash_destroy(info); + connmgr_free_controller_info(info); } /* Makes a deep copy of 'old' into 'port'. */ @@ -863,7 +1011,7 @@ ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port) return error; } -/* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and +/* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and * performs the 'n_actions' actions in 'actions'. The new flow will not * timeout. * @@ -871,48 +1019,68 @@ ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port) * (0...65535, inclusive) then the flow will be visible to OpenFlow * controllers; otherwise, it will be hidden. * - * The caller retains ownership of 'cls_rule' and 'actions'. */ + * The caller retains ownership of 'cls_rule' and 'actions'. + * + * This is a helper function for in-band control and fail-open. */ void -ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, +ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule, const union ofp_action *actions, size_t n_actions) { - struct rule *rule; - rule_create(p, cls_rule, actions, n_actions, 0, 0, 0, false, &rule); -} + const struct rule *rule; -void -ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) -{ - struct rule *rule; + rule = rule_from_cls_rule(classifier_find_rule_exactly( + &ofproto->tables[0], cls_rule)); + if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions, + actions, n_actions)) { + struct flow_mod fm; - rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, - target)); - ofproto_rule_destroy(rule); + memset(&fm, 0, sizeof fm); + fm.cr = *cls_rule; + fm.buffer_id = UINT32_MAX; + fm.actions = (union ofp_action *) actions; + fm.n_actions = n_actions; + add_flow(ofproto, NULL, &fm, NULL); + } } -static void -ofproto_flush_flows__(struct ofproto *ofproto) +/* Searches for a rule with matching criteria exactly equal to 'target' in + * ofproto's table 0 and, if it finds one, deletes it. + * + * This is a helper function for in-band control and fail-open. */ +bool +ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) { - struct rule *rule, *next_rule; - struct cls_cursor cursor; - - COVERAGE_INC(ofproto_flush); + struct rule *rule; - if (ofproto->ofproto_class->flush) { - ofproto->ofproto_class->flush(ofproto); + rule = rule_from_cls_rule(classifier_find_rule_exactly( + &ofproto->tables[0], target)); + if (!rule) { + /* No such rule -> success. */ + return true; + } else if (rule->pending) { + /* An operation on the rule is already pending -> failure. + * Caller must retry later if it's important. */ + return false; + } else { + /* Initiate deletion -> success. */ + struct ofopgroup *group = ofopgroup_create(ofproto); + ofoperation_create(group, rule, OFOPERATION_DELETE); + classifier_remove(&ofproto->tables[rule->table_id], &rule->cr); + rule->ofproto->ofproto_class->rule_destruct(rule); + ofopgroup_submit(group); + return true; } - cls_cursor_init(&cursor, &ofproto->cls, NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { - ofproto_rule_destroy(rule); - } } +/* Starts the process of deleting all of the flows from all of ofproto's flow + * tables and then reintroducing the flows required by in-band control and + * fail-open. The process will complete in a later call to ofproto_run(). */ void ofproto_flush_flows(struct ofproto *ofproto) { - ofproto_flush_flows__(ofproto); - connmgr_flushed(ofproto->connmgr); + COVERAGE_INC(ofproto_flush); + ofproto->state = S_FLUSH; } static void @@ -946,17 +1114,11 @@ static struct netdev * ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp) { uint32_t curr, advertised, supported, peer; - struct netdev_options netdev_options; enum netdev_flags flags; struct netdev *netdev; int error; - memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = ofproto_port->name; - netdev_options.type = ofproto_port->type; - netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - - error = netdev_open(&netdev_options, &netdev); + error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev); if (error) { VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " "cannot be opened (%s)", @@ -1016,11 +1178,11 @@ ofport_install(struct ofproto *p, } ofport->ofproto = p; ofport->netdev = netdev; + ofport->change_seq = netdev_change_seq(netdev); ofport->opp = *opp; ofport->ofp_port = ntohs(opp->port_no); /* Add port to 'p'. */ - netdev_monitor_add(p->netdev_monitor, ofport->netdev); hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0)); shash_add(&p->port_by_name, netdev_name, ofport); @@ -1087,7 +1249,7 @@ ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port) struct ofport *port = ofproto_get_port(ofproto, ofp_port); if (port) { if (port->ofproto->ofproto_class->set_cfm) { - port->ofproto->ofproto_class->set_cfm(port, NULL, NULL, 0); + port->ofproto->ofproto_class->set_cfm(port, NULL); } if (port->ofproto->ofproto_class->bundle_remove) { port->ofproto->ofproto_class->bundle_remove(port); @@ -1101,7 +1263,6 @@ ofport_destroy__(struct ofport *port) struct ofproto *ofproto = port->ofproto; const char *name = netdev_get_name(port->netdev); - netdev_monitor_remove(ofproto->netdev_monitor, port->netdev); hmap_remove(&ofproto->ports, &port->hmap_node); shash_delete(&ofproto->port_by_name, shash_find(&ofproto->port_by_name, name)); @@ -1150,21 +1311,24 @@ update_port(struct ofproto *ofproto, const char *name) if (netdev) { port = ofproto_get_port(ofproto, ofproto_port.ofp_port); if (port && !strcmp(netdev_get_name(port->netdev), name)) { + struct netdev *old_netdev = port->netdev; + /* 'name' hasn't changed location. Any properties changed? */ if (!ofport_equal(&port->opp, &opp)) { ofport_modified(port, &opp); } - /* Install the newly opened netdev in case it has changed. */ - netdev_monitor_remove(ofproto->netdev_monitor, port->netdev); - netdev_monitor_add(ofproto->netdev_monitor, netdev); - - netdev_close(port->netdev); + /* Install the newly opened netdev in case it has changed. + * Don't close the old netdev yet in case port_modified has to + * remove a retained reference to it.*/ port->netdev = netdev; + port->change_seq = netdev_change_seq(netdev); if (port->ofproto->ofproto_class->port_modified) { port->ofproto->ofproto_class->port_modified(port); } + + netdev_close(old_netdev); } else { /* If 'port' is nonnull then its name differs from 'name' and thus * we should delete it. If we think there's a port named 'name' @@ -1210,55 +1374,6 @@ init_ports(struct ofproto *p) return 0; } -/* Creates a new rule initialized as specified, inserts it into 'ofproto''s - * flow table, and stores the new rule into '*rulep'. Returns 0 on success, - * otherwise a positive errno value or OpenFlow error code. */ -static int -rule_create(struct ofproto *ofproto, const struct cls_rule *cls_rule, - const union ofp_action *actions, size_t n_actions, - uint16_t idle_timeout, uint16_t hard_timeout, - ovs_be64 flow_cookie, bool send_flow_removed, - struct rule **rulep) -{ - struct rule *rule; - int error; - - rule = ofproto->ofproto_class->rule_alloc(); - if (!rule) { - error = ENOMEM; - goto error; - } - - rule->ofproto = ofproto; - rule->cr = *cls_rule; - rule->flow_cookie = flow_cookie; - rule->created = time_msec(); - rule->idle_timeout = idle_timeout; - rule->hard_timeout = hard_timeout; - rule->send_flow_removed = send_flow_removed; - if (n_actions > 0) { - rule->actions = xmemdup(actions, n_actions * sizeof *actions); - } else { - rule->actions = NULL; - } - rule->n_actions = n_actions; - - error = ofproto->ofproto_class->rule_construct(rule); - if (error) { - ofproto_rule_destroy__(rule); - goto error; - } - - *rulep = rule; - return 0; - -error: - VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", - ofproto->name, strerror(error)); - *rulep = NULL; - return error; -} - static void ofproto_rule_destroy__(struct rule *rule) { @@ -1266,31 +1381,36 @@ ofproto_rule_destroy__(struct rule *rule) rule->ofproto->ofproto_class->rule_dealloc(rule); } -/* Destroys 'rule' and removes it from the flow table and the datapath. */ +/* This function allows an ofproto implementation to destroy any rules that + * remain when its ->destruct() function is called. The caller must have + * already uninitialized any derived members of 'rule' (step 5 described in the + * large comment in ofproto/ofproto-provider.h titled "Life Cycle"). + * This function implements steps 6 and 7. + * + * This function should only be called from an ofproto implementation's + * ->destruct() function. It is not suitable elsewhere. */ void ofproto_rule_destroy(struct rule *rule) { - if (rule) { - rule->ofproto->ofproto_class->rule_destruct(rule); - ofproto_rule_destroy__(rule); - } + assert(!rule->pending); + classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr); + ofproto_rule_destroy__(rule); } /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't * count). */ static bool -rule_has_out_port(const struct rule *rule, ovs_be16 out_port) +rule_has_out_port(const struct rule *rule, uint16_t out_port) { const union ofp_action *oa; - struct actions_iterator i; + size_t left; - if (out_port == htons(OFPP_NONE)) { + if (out_port == OFPP_NONE) { return true; } - for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; - oa = actions_next(&i)) { - if (action_outputs_to_port(oa, out_port)) { + OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) { + if (action_outputs_to_port(oa, htons(out_port))) { return true; } } @@ -1327,17 +1447,6 @@ rule_is_hidden(const struct rule *rule) return rule->cr.priority > UINT16_MAX; } -static void -send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, - int error) -{ - struct ofpbuf *buf = ofputil_encode_error_msg(error, oh); - if (buf) { - COVERAGE_INC(ofproto_error); - ofconn_send_reply(ofconn, buf); - } -} - static int handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh) { @@ -1352,25 +1461,22 @@ handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) struct ofp_switch_features *osf; struct ofpbuf *buf; struct ofport *port; + bool arp_match_ip; + uint32_t actions; + + ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions); + assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */ osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); osf->datapath_id = htonll(ofproto->datapath_id); osf->n_buffers = htonl(pktbuf_capacity()); - osf->n_tables = 1; + osf->n_tables = ofproto->n_tables; osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | - OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); - osf->actions = htonl((1u << OFPAT_OUTPUT) | - (1u << OFPAT_SET_VLAN_VID) | - (1u << OFPAT_SET_VLAN_PCP) | - (1u << OFPAT_STRIP_VLAN) | - (1u << OFPAT_SET_DL_SRC) | - (1u << OFPAT_SET_DL_DST) | - (1u << OFPAT_SET_NW_SRC) | - (1u << OFPAT_SET_NW_DST) | - (1u << OFPAT_SET_NW_TOS) | - (1u << OFPAT_SET_TP_SRC) | - (1u << OFPAT_SET_TP_DST) | - (1u << OFPAT_ENQUEUE)); + OFPC_PORT_STATS); + if (arp_match_ip) { + osf->capabilities |= htonl(OFPC_ARP_MATCH_IP); + } + osf->actions = htonl(actions); HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { ofpbuf_put(buf, &port->opp, sizeof port->opp); @@ -1435,7 +1541,7 @@ handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) * * The log message mentions 'msg_type'. */ static int -reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) +reject_slave_controller(struct ofconn *ofconn, const char *msg_type) { if (ofconn_get_type(ofconn) == OFCONN_PRIMARY && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) { @@ -1551,89 +1657,15 @@ handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } -static struct ofpbuf * -make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) -{ - struct ofp_stats_reply *osr; - struct ofpbuf *msg; - - msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); - osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); - osr->type = type; - osr->flags = htons(0); - return msg; -} - -static struct ofpbuf * -start_ofp_stats_reply(const struct ofp_header *request, size_t body_len) -{ - const struct ofp_stats_request *osr - = (const struct ofp_stats_request *) request; - return make_ofp_stats_reply(osr->header.xid, osr->type, body_len); -} - -static void * -append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn, - struct ofpbuf **msgp) -{ - struct ofpbuf *msg = *msgp; - assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); - if (nbytes + msg->size > UINT16_MAX) { - struct ofp_stats_reply *reply = msg->data; - reply->flags = htons(OFPSF_REPLY_MORE); - *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes); - ofconn_send_reply(ofconn, msg); - } - return ofpbuf_put_uninit(*msgp, nbytes); -} - -static struct ofpbuf * -make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) -{ - struct nicira_stats_msg *nsm; - struct ofpbuf *msg; - - msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX)); - nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg); - nsm->type = htons(OFPST_VENDOR); - nsm->flags = htons(0); - nsm->vendor = htonl(NX_VENDOR_ID); - nsm->subtype = subtype; - return msg; -} - -static struct ofpbuf * -start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len) -{ - return make_nxstats_reply(request->header.xid, request->subtype, body_len); -} - -static void -append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, - struct ofpbuf **msgp) -{ - struct ofpbuf *msg = *msgp; - assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg)); - if (nbytes + msg->size > UINT16_MAX) { - struct nicira_stats_msg *reply = msg->data; - reply->flags = htons(OFPSF_REPLY_MORE); - *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes); - ofconn_send_reply(ofconn, msg); - } - ofpbuf_prealloc_tailroom(*msgp, nbytes); -} - static int handle_desc_stats_request(struct ofconn *ofconn, - const struct ofp_header *request) + const struct ofp_stats_msg *request) { struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofp_desc_stats *ods; struct ofpbuf *msg; - msg = start_ofp_stats_reply(request, sizeof *ods); - ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); - memset(ods, 0, sizeof *ods); + ods = ofputil_make_stats_reply(sizeof *ods, request, &msg); ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); @@ -1646,32 +1678,32 @@ handle_desc_stats_request(struct ofconn *ofconn, static int handle_table_stats_request(struct ofconn *ofconn, - const struct ofp_header *request) + const struct ofp_stats_msg *request) { struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofp_table_stats *ots; struct ofpbuf *msg; + size_t i; - msg = start_ofp_stats_reply(request, sizeof *ots * 2); + ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg); - /* Classifier table. */ - ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); - memset(ots, 0, sizeof *ots); - strcpy(ots->name, "classifier"); - ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10 - ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL)); - ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */ - ots->active_count = htonl(classifier_count(&p->cls)); - put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */ - put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */ + ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables); + for (i = 0; i < p->n_tables; i++) { + ots[i].table_id = i; + sprintf(ots[i].name, "table%zu", i); + ots[i].wildcards = htonl(OFPFW_ALL); + ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */ + ots[i].active_count = htonl(classifier_count(&p->tables[i])); + } + + p->ofproto_class->get_tables(p, ots); ofconn_send_reply(ofconn, msg); return 0; } static void -append_port_stat(struct ofport *port, struct ofconn *ofconn, - struct ofpbuf **msgp) +append_port_stat(struct ofport *port, struct list *replies) { struct netdev_stats stats; struct ofp_port_stats *ops; @@ -1681,7 +1713,7 @@ append_port_stat(struct ofport *port, struct ofconn *ofconn, * netdev_get_stats() will log errors. */ netdev_get_stats(port->netdev, &stats); - ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); + ops = ofputil_append_stats_reply(sizeof *ops, replies); ops->port_no = port->opp.port_no; memset(ops->pad, 0, sizeof ops->pad); put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets)); @@ -1699,27 +1731,26 @@ append_port_stat(struct ofport *port, struct ofconn *ofconn, } static int -handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +handle_port_stats_request(struct ofconn *ofconn, + const struct ofp_port_stats_request *psr) { struct ofproto *p = ofconn_get_ofproto(ofconn); - const struct ofp_port_stats_request *psr = ofputil_stats_body(oh); - struct ofp_port_stats *ops; - struct ofpbuf *msg; struct ofport *port; + struct list replies; - msg = start_ofp_stats_reply(oh, sizeof *ops * 16); + ofputil_start_stats_reply(&psr->osm, &replies); if (psr->port_no != htons(OFPP_NONE)) { port = ofproto_get_port(p, ntohs(psr->port_no)); if (port) { - append_port_stat(port, ofconn, &msg); + append_port_stat(port, &replies); } } else { HMAP_FOR_EACH (port, hmap_node, &p->ports) { - append_port_stat(port, ofconn, &msg); + append_port_stat(port, &replies); } } - ofconn_send_reply(ofconn, msg); + ofconn_send_replies(ofconn, &replies); return 0; } @@ -1731,169 +1762,162 @@ calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec) *nsec = (msecs % 1000) * (1000 * 1000); } -static void -calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be) -{ - uint32_t sec, nsec; - - calc_flow_duration__(start, &sec, &nsec); - *sec_be = htonl(sec); - *nsec_be = htonl(nsec); -} - -static void -put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, - ovs_be16 out_port, struct ofpbuf **replyp) -{ - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofp_flow_stats *ofs; - uint64_t packet_count, byte_count; - ovs_be64 cookie; - size_t act_len, len; - - if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { - return; - } - - act_len = sizeof *rule->actions * rule->n_actions; - len = offsetof(struct ofp_flow_stats, actions) + act_len; - - ofproto->ofproto_class->rule_get_stats(rule, &packet_count, &byte_count); - - ofs = append_ofp_stats_reply(len, ofconn, replyp); - ofs->length = htons(len); - ofs->table_id = 0; - ofs->pad = 0; - ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn), - &ofs->match, rule->flow_cookie, &cookie); - put_32aligned_be64(&ofs->cookie, cookie); - calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); - ofs->priority = htons(rule->cr.priority); - ofs->idle_timeout = htons(rule->idle_timeout); - ofs->hard_timeout = htons(rule->hard_timeout); - memset(ofs->pad2, 0, sizeof ofs->pad2); - put_32aligned_be64(&ofs->packet_count, htonll(packet_count)); - put_32aligned_be64(&ofs->byte_count, htonll(byte_count)); - if (rule->n_actions > 0) { - memcpy(ofs->actions, rule->actions, act_len); - } -} - -static bool -is_valid_table(uint8_t table_id) +static struct classifier * +first_matching_table(struct ofproto *ofproto, uint8_t table_id) { - if (table_id == 0 || table_id == 0xff) { - return true; + if (table_id == 0xff) { + return &ofproto->tables[0]; + } else if (table_id < ofproto->n_tables) { + return &ofproto->tables[table_id]; } else { /* It would probably be better to reply with an error but there doesn't * seem to be any appropriate value, so that might just be * confusing. */ VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8, table_id); - return false; + return NULL; } } +static struct classifier * +next_matching_table(struct ofproto *ofproto, + struct classifier *cls, uint8_t table_id) +{ + return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1] + ? cls + 1 + : NULL); +} + +/* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in + * OFPROTO: + * + * - If TABLE_ID is 0xff, this iterates over every classifier table in + * OFPROTO. + * + * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates + * only once, for that table. + * + * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning + * and does not enter the loop at all. + * + * All parameters are evaluated multiple times. + */ +#define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \ + for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \ + (CLS) != NULL; \ + (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID)) + +/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if + * 'table_id' is 0xff) that match 'match' in the "loose" way required for + * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list + * 'rules'. + * + * If 'out_port' is anything other than OFPP_NONE, then only rules that output + * to 'out_port' are included. + * + * Hidden rules are always omitted. + * + * Returns 0 on success, otherwise an OpenFlow error code. */ static int -handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +collect_rules_loose(struct ofproto *ofproto, uint8_t table_id, + const struct cls_rule *match, uint16_t out_port, + struct list *rules) { - const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh); - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofpbuf *reply; + struct classifier *cls; - COVERAGE_INC(ofproto_flows_req); - reply = start_ofp_stats_reply(oh, 1024); - if (is_valid_table(fsr->table_id)) { + list_init(rules); + FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) { struct cls_cursor cursor; - struct cls_rule target; struct rule *rule; - ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, - &target); - cls_cursor_init(&cursor, &ofproto->cls, &target); + cls_cursor_init(&cursor, cls, match); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); + if (rule->pending) { + return OFPROTO_POSTPONE; + } + if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { + list_push_back(rules, &rule->ofproto_node); + } } } - ofconn_send_reply(ofconn, reply); - return 0; } -static void -put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, - ovs_be16 out_port, struct ofpbuf **replyp) +/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if + * 'table_id' is 0xff) that match 'match' in the "strict" way required for + * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them + * on list 'rules'. + * + * If 'out_port' is anything other than OFPP_NONE, then only rules that output + * to 'out_port' are included. + * + * Hidden rules are always omitted. + * + * Returns 0 on success, otherwise an OpenFlow error code. */ +static int +collect_rules_strict(struct ofproto *ofproto, uint8_t table_id, + const struct cls_rule *match, uint16_t out_port, + struct list *rules) { - struct nx_flow_stats *nfs; - uint64_t packet_count, byte_count; - size_t act_len, start_len; - struct ofpbuf *reply; - - if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { - return; - } + struct classifier *cls; - rule->ofproto->ofproto_class->rule_get_stats(rule, - &packet_count, &byte_count); + list_init(rules); + FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) { + struct rule *rule; - act_len = sizeof *rule->actions * rule->n_actions; - - append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp); - start_len = (*replyp)->size; - reply = *replyp; - - nfs = ofpbuf_put_uninit(reply, sizeof *nfs); - nfs->table_id = 0; - nfs->pad = 0; - calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec); - nfs->cookie = rule->flow_cookie; - nfs->priority = htons(rule->cr.priority); - nfs->idle_timeout = htons(rule->idle_timeout); - nfs->hard_timeout = htons(rule->hard_timeout); - nfs->match_len = htons(nx_put_match(reply, &rule->cr)); - memset(nfs->pad2, 0, sizeof nfs->pad2); - nfs->packet_count = htonll(packet_count); - nfs->byte_count = htonll(byte_count); - if (rule->n_actions > 0) { - ofpbuf_put(reply, rule->actions, act_len); + rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match)); + if (rule) { + if (rule->pending) { + return OFPROTO_POSTPONE; + } + if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { + list_push_back(rules, &rule->ofproto_node); + } + } } - nfs->length = htons(reply->size - start_len); + return 0; } static int -handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh) +handle_flow_stats_request(struct ofconn *ofconn, + const struct ofp_stats_msg *osm) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct nx_flow_stats_request *nfsr; - struct cls_rule target; - struct ofpbuf *reply; - struct ofpbuf b; + struct flow_stats_request fsr; + struct list replies; + struct list rules; + struct rule *rule; int error; - ofpbuf_use_const(&b, oh, ntohs(oh->length)); - - /* Dissect the message. */ - nfsr = ofpbuf_pull(&b, sizeof *nfsr); - error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target); + error = ofputil_decode_flow_stats_request(&fsr, &osm->header); if (error) { return error; } - if (b.size) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + + error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match, + fsr.out_port, &rules); + if (error) { + return error; } - COVERAGE_INC(ofproto_flows_req); - reply = start_nxstats_reply(&nfsr->nsm, 1024); - if (is_valid_table(nfsr->table_id)) { - struct cls_cursor cursor; - struct rule *rule; + ofputil_start_stats_reply(osm, &replies); + LIST_FOR_EACH (rule, ofproto_node, &rules) { + struct ofputil_flow_stats fs; - cls_cursor_init(&cursor, &ofproto->cls, &target); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply); - } + fs.rule = rule->cr; + fs.cookie = rule->flow_cookie; + fs.table_id = rule->table_id; + calc_flow_duration__(rule->created, &fs.duration_sec, + &fs.duration_nsec); + fs.idle_timeout = rule->idle_timeout; + fs.hard_timeout = rule->hard_timeout; + ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count, + &fs.byte_count); + fs.actions = rule->actions; + fs.n_actions = rule->n_actions; + ofputil_append_flow_stats_reply(&fs, &replies); } - ofconn_send_reply(ofconn, reply); + ofconn_send_replies(ofconn, &replies); return 0; } @@ -1902,11 +1926,13 @@ static void flow_stats_ds(struct rule *rule, struct ds *results) { uint64_t packet_count, byte_count; - size_t act_len = sizeof *rule->actions * rule->n_actions; rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count, &byte_count); + if (rule->table_id != 0) { + ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id); + } ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); ds_put_format(results, "priority=%u, ", rule->cr.priority); @@ -1914,8 +1940,8 @@ flow_stats_ds(struct rule *rule, struct ds *results) ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); cls_rule_format(&rule->cr, results); ds_put_char(results, ','); - if (act_len > 0) { - ofp_print_actions(results, &rule->actions->header, act_len); + if (rule->n_actions > 0) { + ofp_print_actions(results, rule->actions, rule->n_actions); } else { ds_put_cstr(results, "drop"); } @@ -1927,12 +1953,16 @@ flow_stats_ds(struct rule *rule, struct ds *results) void ofproto_get_all_flows(struct ofproto *p, struct ds *results) { - struct cls_cursor cursor; - struct rule *rule; + struct classifier *cls; + + OFPROTO_FOR_EACH_TABLE (cls, p) { + struct cls_cursor cursor; + struct rule *rule; - cls_cursor_init(&cursor, &p->cls, NULL); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - flow_stats_ds(rule, results); + cls_cursor_init(&cursor, cls, NULL); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + flow_stats_ds(rule, results); + } } } @@ -1945,102 +1975,81 @@ ofproto_get_netflow_ids(const struct ofproto *ofproto, ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id); } -static void -query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, - ovs_be16 out_port, uint8_t table_id, - struct ofp_aggregate_stats_reply *oasr) +/* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1 + * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM + * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */ +int +ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port) { - uint64_t total_packets = 0; - uint64_t total_bytes = 0; - int n_flows = 0; - - COVERAGE_INC(ofproto_agg_request); - - if (is_valid_table(table_id)) { - struct cls_cursor cursor; - struct rule *rule; - - cls_cursor_init(&cursor, &ofproto->cls, target); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { - uint64_t packet_count; - uint64_t byte_count; - - ofproto->ofproto_class->rule_get_stats(rule, &packet_count, - &byte_count); - - total_packets += packet_count; - total_bytes += byte_count; - n_flows++; - } - } - } - - oasr->flow_count = htonl(n_flows); - put_32aligned_be64(&oasr->packet_count, htonll(total_packets)); - put_32aligned_be64(&oasr->byte_count, htonll(total_bytes)); - memset(oasr->pad, 0, sizeof oasr->pad); -} + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + return (ofport && ofproto->ofproto_class->get_cfm_fault + ? ofproto->ofproto_class->get_cfm_fault(ofport) + : -1); +} static int handle_aggregate_stats_request(struct ofconn *ofconn, - const struct ofp_header *oh) -{ - const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh); - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofp_aggregate_stats_reply *reply; - struct cls_rule target; - struct ofpbuf *msg; - - ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, - &target); - - msg = start_ofp_stats_reply(oh, sizeof *reply); - reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); - query_aggregate_stats(ofproto, &target, request->out_port, - request->table_id, reply); - ofconn_send_reply(ofconn, msg); - return 0; -} - -static int -handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh) + const struct ofp_stats_msg *osm) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct nx_aggregate_stats_request *request; - struct ofp_aggregate_stats_reply *reply; - struct cls_rule target; - struct ofpbuf b; - struct ofpbuf *buf; + struct flow_stats_request request; + struct ofputil_aggregate_stats stats; + bool unknown_packets, unknown_bytes; + struct ofpbuf *reply; + struct list rules; + struct rule *rule; int error; - ofpbuf_use_const(&b, oh, ntohs(oh->length)); + error = ofputil_decode_flow_stats_request(&request, &osm->header); + if (error) { + return error; + } - /* Dissect the message. */ - request = ofpbuf_pull(&b, sizeof *request); - error = nx_pull_match(&b, ntohs(request->match_len), 0, &target); + error = collect_rules_loose(ofproto, request.table_id, &request.match, + request.out_port, &rules); if (error) { return error; } - if (b.size) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + + memset(&stats, 0, sizeof stats); + unknown_packets = unknown_bytes = false; + LIST_FOR_EACH (rule, ofproto_node, &rules) { + uint64_t packet_count; + uint64_t byte_count; + + ofproto->ofproto_class->rule_get_stats(rule, &packet_count, + &byte_count); + + if (packet_count == UINT64_MAX) { + unknown_packets = true; + } else { + stats.packet_count += packet_count; + } + + if (byte_count == UINT64_MAX) { + unknown_bytes = true; + } else { + stats.byte_count += byte_count; + } + + stats.flow_count++; + } + if (unknown_packets) { + stats.packet_count = UINT64_MAX; + } + if (unknown_bytes) { + stats.byte_count = UINT64_MAX; } - /* Reply. */ - COVERAGE_INC(ofproto_flows_req); - buf = start_nxstats_reply(&request->nsm, sizeof *reply); - reply = ofpbuf_put_uninit(buf, sizeof *reply); - query_aggregate_stats(ofproto, &target, request->out_port, - request->table_id, reply); - ofconn_send_reply(ofconn, buf); + reply = ofputil_encode_aggregate_stats_reply(&stats, osm); + ofconn_send_reply(ofconn, reply); return 0; } struct queue_stats_cbdata { - struct ofconn *ofconn; struct ofport *ofport; - struct ofpbuf *msg; + struct list replies; }; static void @@ -2049,7 +2058,7 @@ put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, { struct ofp_queue_stats *reply; - reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); + reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies); reply->port_no = cbdata->ofport->opp.port_no; memset(reply->pad, 0, sizeof reply->pad); reply->queue_id = htonl(queue_id); @@ -2086,24 +2095,18 @@ handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, } static int -handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +handle_queue_stats_request(struct ofconn *ofconn, + const struct ofp_queue_stats_request *qsr) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - const struct ofp_queue_stats_request *qsr; struct queue_stats_cbdata cbdata; struct ofport *port; unsigned int port_no; uint32_t queue_id; - qsr = ofputil_stats_body(oh); - if (!qsr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - COVERAGE_INC(ofproto_queue_req); - cbdata.ofconn = ofconn; - cbdata.msg = start_ofp_stats_reply(oh, 128); + ofputil_start_stats_reply(&qsr->osm, &cbdata.replies); port_no = ntohs(qsr->port_no); queue_id = ntohl(qsr->queue_id); @@ -2117,14 +2120,34 @@ handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) handle_queue_stats_for_port(port, queue_id, &cbdata); } } else { - ofpbuf_delete(cbdata.msg); + ofpbuf_list_delete(&cbdata.replies); return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); } - ofconn_send_reply(ofconn, cbdata.msg); + ofconn_send_replies(ofconn, &cbdata.replies); return 0; } +static bool +is_flow_deletion_pending(const struct ofproto *ofproto, + const struct cls_rule *cls_rule, + uint8_t table_id) +{ + if (!hmap_is_empty(&ofproto->deletions)) { + struct ofoperation *op; + + HMAP_FOR_EACH_WITH_HASH (op, hmap_node, + cls_rule_hash(cls_rule, table_id), + &ofproto->deletions) { + if (cls_rule_equal(cls_rule, &op->rule->cr)) { + return true; + } + } + } + + return false; +} + /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT * in which no matching flow already exists in the flow table. * @@ -2135,214 +2158,231 @@ handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ static int -add_flow(struct ofconn *ofconn, struct flow_mod *fm) +add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct flow_mod *fm, + const struct ofp_header *request) { - struct ofproto *p = ofconn_get_ofproto(ofconn); - struct ofpbuf *packet; + struct classifier *table; + struct ofopgroup *group; + struct rule *victim; struct rule *rule; - uint16_t in_port; - int buf_err; int error; - if (fm->flags & OFPFF_CHECK_OVERLAP - && classifier_rule_overlaps(&p->cls, &fm->cr)) { - return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); - } + /* Check for overlap, if requested. */ + if (fm->flags & OFPFF_CHECK_OVERLAP) { + struct classifier *cls; - buf_err = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id, &packet, &in_port); - error = rule_create(p, &fm->cr, fm->actions, fm->n_actions, - fm->idle_timeout, fm->hard_timeout, fm->cookie, - fm->flags & OFPFF_SEND_FLOW_REM, &rule); - if (error) { - ofpbuf_delete(packet); - return error; + FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) { + if (classifier_rule_overlaps(cls, &fm->cr)) { + return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); + } + } } - if (packet) { - assert(!buf_err); - return rule_execute(rule, in_port, packet); + /* Pick table. */ + if (fm->table_id == 0xff) { + uint8_t table_id; + if (ofproto->ofproto_class->rule_choose_table) { + error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr, + &table_id); + if (error) { + return error; + } + assert(table_id < ofproto->n_tables); + table = &ofproto->tables[table_id]; + } else { + table = &ofproto->tables[0]; + } + } else if (fm->table_id < ofproto->n_tables) { + table = &ofproto->tables[fm->table_id]; + } else { + return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID); } - return buf_err; -} - -static struct rule * -find_flow_strict(struct ofproto *p, const struct flow_mod *fm) -{ - return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); -} -static int -send_buffered_packet(struct ofconn *ofconn, - struct rule *rule, uint32_t buffer_id) -{ - struct ofpbuf *packet; - uint16_t in_port; - int error; + /* Serialize against pending deletion. */ + if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) { + return OFPROTO_POSTPONE; + } - if (buffer_id == UINT32_MAX) { - return 0; + /* Allocate new rule. */ + rule = ofproto->ofproto_class->rule_alloc(); + if (!rule) { + VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", + ofproto->name, strerror(error)); + return ENOMEM; + } + rule->ofproto = ofproto; + rule->cr = fm->cr; + rule->pending = NULL; + rule->flow_cookie = fm->cookie; + rule->created = time_msec(); + rule->idle_timeout = fm->idle_timeout; + rule->hard_timeout = fm->hard_timeout; + rule->table_id = table - ofproto->tables; + rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0; + rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions); + rule->n_actions = fm->n_actions; + + /* Insert new rule. */ + victim = rule_from_cls_rule(classifier_replace(table, &rule->cr)); + if (victim && victim->pending) { + error = OFPROTO_POSTPONE; + } else { + group = (ofconn + ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id) + : ofopgroup_create(ofproto)); + ofoperation_create(group, rule, OFOPERATION_ADD); + rule->pending->victim = victim; + + error = ofproto->ofproto_class->rule_construct(rule); + if (error) { + ofoperation_destroy(rule->pending); + } + ofopgroup_submit(group); } - error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port); + /* Back out if an error occurred. */ if (error) { - return error; + if (victim) { + classifier_replace(table, &victim->cr); + } else { + classifier_remove(table, &rule->cr); + } + ofproto_rule_destroy__(rule); } - - return rule_execute(rule, in_port, packet); + return error; } /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ -struct modify_flows_cbdata { - struct ofproto *ofproto; - const struct flow_mod *fm; - struct rule *match; -}; +/* Modifies the rules listed in 'rules', changing their actions to match those + * in 'fm'. + * + * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, + * if any. + * + * Returns 0 on success, otherwise an OpenFlow error code. */ +static int +modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm, + const struct ofp_header *request, struct list *rules) +{ + struct ofopgroup *group; + struct rule *rule; + + group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id); + LIST_FOR_EACH (rule, ofproto_node, rules) { + if (!ofputil_actions_equal(fm->actions, fm->n_actions, + rule->actions, rule->n_actions)) { + ofoperation_create(group, rule, OFOPERATION_MODIFY); + rule->pending->actions = rule->actions; + rule->pending->n_actions = rule->n_actions; + rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions); + rule->n_actions = fm->n_actions; + rule->ofproto->ofproto_class->rule_modify_actions(rule); + } + rule->flow_cookie = fm->cookie; + } + ofopgroup_submit(group); -static int modify_flow(const struct flow_mod *, struct rule *); + return 0; +} /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as * encoded by ofp_mkerr() on failure. * - * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, * if any. */ static int -modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) +modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm, + const struct ofp_header *request) { struct ofproto *p = ofconn_get_ofproto(ofconn); - struct rule *match = NULL; - struct cls_cursor cursor; - struct rule *rule; + struct list rules; int error; - error = 0; - cls_cursor_init(&cursor, &p->cls, &fm->cr); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - if (!rule_is_hidden(rule)) { - int retval = modify_flow(fm, rule); - if (!retval) { - match = rule; - } else { - error = retval; - } - } - } - - if (error) { - return error; - } else if (match) { - /* This credits the packet to whichever flow happened to match last. - * That's weird. Maybe we should do a lookup for the flow that - * actually matches the packet? Who knows. */ - send_buffered_packet(ofconn, match, fm->buffer_id); - return 0; - } else { - return add_flow(ofconn, fm); - } + error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules); + return (error ? error + : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request) + : modify_flows__(ofconn, fm, request, &rules)); } /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error * code as encoded by ofp_mkerr() on failure. * - * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, * if any. */ static int -modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) +modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm, + const struct ofp_header *request) { struct ofproto *p = ofconn_get_ofproto(ofconn); - struct rule *rule = find_flow_strict(p, fm); - if (rule && !rule_is_hidden(rule)) { - int error = modify_flow(fm, rule); - if (!error) { - error = send_buffered_packet(ofconn, rule, fm->buffer_id); - } - return error; - } else { - return add_flow(ofconn, fm); - } + struct list rules; + int error; + + error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules); + return (error ? error + : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request) + : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request, + &rules) + : 0); } + +/* OFPFC_DELETE implementation. */ -/* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has - * been identified as a flow to be modified, by changing the rule's actions to - * match those in 'ofm' (which is followed by 'n_actions' ofp_action[] - * structures). */ +/* Deletes the rules listed in 'rules'. + * + * Returns 0 on success, otherwise an OpenFlow error code. */ static int -modify_flow(const struct flow_mod *fm, struct rule *rule) +delete_flows__(struct ofconn *ofconn, const struct ofp_header *request, + struct list *rules) { - size_t actions_len = fm->n_actions * sizeof *rule->actions; - int error; + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct rule *rule, *next; + struct ofopgroup *group; - if (fm->n_actions == rule->n_actions - && (!fm->n_actions - || !memcmp(fm->actions, rule->actions, actions_len))) { - error = 0; - } else { - error = rule->ofproto->ofproto_class->rule_modify_actions( - rule, fm->actions, fm->n_actions); - if (!error) { - free(rule->actions); - rule->actions = (fm->n_actions - ? xmemdup(fm->actions, actions_len) - : NULL); - rule->n_actions = fm->n_actions; - } - } + group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX); + LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) { + ofproto_rule_send_removed(rule, OFPRR_DELETE); - if (!error) { - rule->flow_cookie = fm->cookie; + ofoperation_create(group, rule, OFOPERATION_DELETE); + classifier_remove(&ofproto->tables[rule->table_id], &rule->cr); + rule->ofproto->ofproto_class->rule_destruct(rule); } + ofopgroup_submit(group); - return error; + return 0; } - -/* OFPFC_DELETE implementation. */ - -static void delete_flow(struct rule *, ovs_be16 out_port); /* Implements OFPFC_DELETE. */ -static void -delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) +static int +delete_flows_loose(struct ofconn *ofconn, const struct flow_mod *fm, + const struct ofp_header *request) { - struct rule *rule, *next_rule; - struct cls_cursor cursor; + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct list rules; + int error; - cls_cursor_init(&cursor, &p->cls, &fm->cr); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { - delete_flow(rule, htons(fm->out_port)); - } + error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port, + &rules); + return (error ? error + : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules) + : 0); } /* Implements OFPFC_DELETE_STRICT. */ -static void -delete_flow_strict(struct ofproto *p, struct flow_mod *fm) -{ - struct rule *rule = find_flow_strict(p, fm); - if (rule) { - delete_flow(rule, htons(fm->out_port)); - } -} - -/* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has - * been identified as a flow to delete from 'p''s flow table, by deleting the - * flow and sending out a OFPT_FLOW_REMOVED message to any interested - * controller. - * - * Will not delete 'rule' if it is hidden. Will delete 'rule' only if - * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the - * specified 'out_port'. */ -static void -delete_flow(struct rule *rule, ovs_be16 out_port) +static int +delete_flow_strict(struct ofconn *ofconn, struct flow_mod *fm, + const struct ofp_header *request) { - if (rule_is_hidden(rule)) { - return; - } - - if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { - return; - } + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct list rules; + int error; - ofproto_rule_send_removed(rule, OFPRR_DELETE); - ofproto_rule_destroy(rule); + error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port, + &rules); + return (error ? error + : list_is_singleton(&rules) ? delete_flows__(ofconn, request, + &rules) + : 0); } static void @@ -2374,15 +2414,24 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) void ofproto_rule_expire(struct rule *rule, uint8_t reason) { + struct ofproto *ofproto = rule->ofproto; + struct ofopgroup *group; + assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT); + ofproto_rule_send_removed(rule, reason); - ofproto_rule_destroy(rule); + + group = ofopgroup_create(ofproto); + ofoperation_create(group, rule, OFOPERATION_DELETE); + classifier_remove(&ofproto->tables[rule->table_id], &rule->cr); + rule->ofproto->ofproto_class->rule_destruct(rule); + ofopgroup_submit(group); } static int handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofproto *p = ofconn_get_ofproto(ofconn); + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct flow_mod fm; int error; @@ -2391,7 +2440,12 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) return error; } - error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn)); + if (list_size(&ofproto->pending) >= 50) { + return OFPROTO_POSTPONE; + } + + error = ofputil_decode_flow_mod(&fm, oh, + ofconn_get_flow_mod_table_id(ofconn)); if (error) { return error; } @@ -2406,40 +2460,29 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) switch (fm.command) { case OFPFC_ADD: - return add_flow(ofconn, &fm); + return add_flow(ofproto, ofconn, &fm, oh); case OFPFC_MODIFY: - return modify_flows_loose(ofconn, &fm); + return modify_flows_loose(ofconn, &fm, oh); case OFPFC_MODIFY_STRICT: - return modify_flow_strict(ofconn, &fm); + return modify_flow_strict(ofconn, &fm, oh); case OFPFC_DELETE: - delete_flows_loose(p, &fm); - return 0; + return delete_flows_loose(ofconn, &fm, oh); case OFPFC_DELETE_STRICT: - delete_flow_strict(p, &fm); - return 0; + return delete_flow_strict(ofconn, &fm, oh); default: + if (fm.command > 0xff) { + VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but " + "flow_mod_table_id extension is not enabled"); + } return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); } } -static int -handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh) -{ - const struct nxt_tun_id_cookie *msg - = (const struct nxt_tun_id_cookie *) oh; - enum nx_flow_format flow_format; - - flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; - ofconn_set_flow_format(ofconn, flow_format); - - return 0; -} - static int handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) { @@ -2462,6 +2505,11 @@ handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) return ofp_mkerr(OFPET_BAD_REQUEST, -1); } + if (ofconn_get_role(ofconn) != role + && ofconn_has_pending_opgroups(ofconn)) { + return OFPROTO_POSTPONE; + } + ofconn_set_role(ofconn, role); reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf); @@ -2471,6 +2519,17 @@ handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } +static int +handle_nxt_flow_mod_table_id(struct ofconn *ofconn, + const struct ofp_header *oh) +{ + const struct nxt_flow_mod_table_id *msg + = (const struct nxt_flow_mod_table_id *) oh; + + ofconn_set_flow_mod_table_id(ofconn, msg->set != 0); + return 0; +} + static int handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh) { @@ -2479,14 +2538,18 @@ handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh) uint32_t format; format = ntohl(msg->format); - if (format == NXFF_OPENFLOW10 - || format == NXFF_TUN_ID_FROM_COOKIE - || format == NXFF_NXM) { - ofconn_set_flow_format(ofconn, format); - return 0; - } else { + if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); } + + if (format != ofconn_get_flow_format(ofconn) + && ofconn_has_pending_opgroups(ofconn)) { + /* Avoid sending async messages in surprising flow format. */ + return OFPROTO_POSTPONE; + } + + ofconn_set_flow_format(ofconn, format); + return 0; } static int @@ -2495,8 +2558,10 @@ handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) struct ofp_header *ob; struct ofpbuf *buf; - /* Currently, everything executes synchronously, so we can just - * immediately send the barrier reply. */ + if (ofconn_has_pending_opgroups(ofconn)) { + return OFPROTO_POSTPONE; + } + ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); ofconn_send_reply(ofconn, buf); return 0; @@ -2545,45 +2610,40 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) return 0; /* Nicira extension requests. */ - case OFPUTIL_NXT_TUN_ID_FROM_COOKIE: - return handle_tun_id_from_cookie(ofconn, oh); - case OFPUTIL_NXT_ROLE_REQUEST: return handle_role_request(ofconn, oh); + case OFPUTIL_NXT_FLOW_MOD_TABLE_ID: + return handle_nxt_flow_mod_table_id(ofconn, oh); + case OFPUTIL_NXT_SET_FLOW_FORMAT: return handle_nxt_set_flow_format(ofconn, oh); case OFPUTIL_NXT_FLOW_MOD: return handle_flow_mod(ofconn, oh); - /* OpenFlow statistics requests. */ + /* Statistics requests. */ case OFPUTIL_OFPST_DESC_REQUEST: - return handle_desc_stats_request(ofconn, oh); + return handle_desc_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_FLOW_REQUEST: - return handle_flow_stats_request(ofconn, oh); + case OFPUTIL_NXST_FLOW_REQUEST: + return handle_flow_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_AGGREGATE_REQUEST: - return handle_aggregate_stats_request(ofconn, oh); + case OFPUTIL_NXST_AGGREGATE_REQUEST: + return handle_aggregate_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_TABLE_REQUEST: - return handle_table_stats_request(ofconn, oh); + return handle_table_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_PORT_REQUEST: - return handle_port_stats_request(ofconn, oh); + return handle_port_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_QUEUE_REQUEST: - return handle_queue_stats_request(ofconn, oh); - - /* Nicira extension statistics requests. */ - case OFPUTIL_NXST_FLOW_REQUEST: - return handle_nxst_flow(ofconn, oh); + return handle_queue_stats_request(ofconn, msg->data); - case OFPUTIL_NXST_AGGREGATE_REQUEST: - return handle_nxst_aggregate(ofconn, oh); - - case OFPUTIL_INVALID: + case OFPUTIL_MSG_INVALID: case OFPUTIL_OFPT_HELLO: case OFPUTIL_OFPT_ERROR: case OFPUTIL_OFPT_FEATURES_REPLY: @@ -2618,14 +2678,220 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) } } -static void +static bool handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) { int error = handle_openflow__(ofconn, ofp_msg); - if (error) { - send_error_oh(ofconn, ofp_msg->data, error); + if (error && error != OFPROTO_POSTPONE) { + ofconn_send_error(ofconn, ofp_msg->data, error); } COVERAGE_INC(ofproto_recv_openflow); + return error != OFPROTO_POSTPONE; +} + +/* Asynchronous operations. */ + +/* Creates and returns a new ofopgroup that is not associated with any + * OpenFlow connection. + * + * The caller should add operations to the returned group with + * ofoperation_create() and then submit it with ofopgroup_submit(). */ +static struct ofopgroup * +ofopgroup_create(struct ofproto *ofproto) +{ + struct ofopgroup *group = xzalloc(sizeof *group); + group->ofproto = ofproto; + list_init(&group->ofproto_node); + list_init(&group->ops); + list_init(&group->ofconn_node); + return group; +} + +/* Creates and returns a new ofopgroup that is associated with 'ofconn'. If + * the ofopgroup eventually fails, then the error reply will include 'request'. + * If the ofopgroup eventually succeeds, then the packet with buffer id + * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto. + * + * The caller should add operations to the returned group with + * ofoperation_create() and then submit it with ofopgroup_submit(). */ +static struct ofopgroup * +ofopgroup_create_for_ofconn(struct ofconn *ofconn, + const struct ofp_header *request, + uint32_t buffer_id) +{ + struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn)); + size_t request_len = ntohs(request->length); + + ofconn_add_opgroup(ofconn, &group->ofconn_node); + group->ofconn = ofconn; + group->request = xmemdup(request, MIN(request_len, 64)); + group->buffer_id = buffer_id; + + return group; +} + +/* Submits 'group' for processing. + * + * If 'group' contains no operations (e.g. none were ever added, or all of the + * ones that were added completed synchronously), then it is destroyed + * immediately. Otherwise it is added to the ofproto's list of pending + * groups. */ +static void +ofopgroup_submit(struct ofopgroup *group) +{ + if (list_is_empty(&group->ops)) { + ofopgroup_destroy(group); + } else { + list_push_back(&group->ofproto->pending, &group->ofproto_node); + } +} + +static void +ofopgroup_destroy(struct ofopgroup *group) +{ + assert(list_is_empty(&group->ops)); + if (!list_is_empty(&group->ofproto_node)) { + list_remove(&group->ofproto_node); + } + if (!list_is_empty(&group->ofconn_node)) { + list_remove(&group->ofconn_node); + if (group->error) { + ofconn_send_error(group->ofconn, group->request, group->error); + } + connmgr_retry(group->ofproto->connmgr); + } + free(group->request); + free(group); +} + +/* Initiates a new operation on 'rule', of the specified 'type', within + * 'group'. Prior to calling, 'rule' must not have any pending operation. */ +static void +ofoperation_create(struct ofopgroup *group, struct rule *rule, + enum ofoperation_type type) +{ + struct ofoperation *op; + + assert(!rule->pending); + + op = rule->pending = xzalloc(sizeof *op); + op->group = group; + list_push_back(&group->ops, &op->group_node); + op->rule = rule; + op->type = type; + op->status = -1; + op->flow_cookie = rule->flow_cookie; + + if (type == OFOPERATION_DELETE) { + hmap_insert(&op->group->ofproto->deletions, &op->hmap_node, + cls_rule_hash(&rule->cr, rule->table_id)); + } +} + +static void +ofoperation_destroy(struct ofoperation *op) +{ + struct ofopgroup *group = op->group; + + if (op->rule) { + op->rule->pending = NULL; + } + if (op->type == OFOPERATION_DELETE) { + hmap_remove(&group->ofproto->deletions, &op->hmap_node); + } + list_remove(&op->group_node); + free(op->actions); + free(op); + + if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) { + ofopgroup_destroy(group); + } +} + +/* Indicates that 'op' completed with status 'error', which is either 0 to + * indicate success or an OpenFlow error code (constructed with + * e.g. ofp_mkerr()). + * + * If 'op' is a "delete flow" operation, 'error' must be 0. That is, flow + * deletions are not allowed to fail. + * + * Please see the large comment in ofproto/ofproto-provider.h titled + * "Asynchronous Operation Support" for more information. */ +void +ofoperation_complete(struct ofoperation *op, int error) +{ + struct ofopgroup *group = op->group; + struct rule *rule = op->rule; + struct classifier *table = &rule->ofproto->tables[rule->table_id]; + + assert(rule->pending == op); + assert(op->status < 0); + assert(error >= 0); + + if (!error + && !group->error + && op->type != OFOPERATION_DELETE + && group->ofconn + && group->buffer_id != UINT32_MAX + && list_is_singleton(&op->group_node)) { + struct ofpbuf *packet; + uint16_t in_port; + + error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id, + &packet, &in_port); + if (packet) { + assert(!error); + error = rule_execute(rule, in_port, packet); + } + } + if (!group->error) { + group->error = error; + } + + switch (op->type) { + case OFOPERATION_ADD: + if (!error) { + if (op->victim) { + ofproto_rule_destroy__(op->victim); + } + } else { + if (op->victim) { + classifier_replace(table, &op->victim->cr); + op->victim = NULL; + } else { + classifier_remove(table, &rule->cr); + } + ofproto_rule_destroy__(rule); + } + op->victim = NULL; + break; + + case OFOPERATION_DELETE: + assert(!error); + ofproto_rule_destroy__(rule); + op->rule = NULL; + break; + + case OFOPERATION_MODIFY: + if (error) { + free(rule->actions); + rule->actions = op->actions; + rule->n_actions = op->n_actions; + op->actions = NULL; + } + break; + + default: + NOT_REACHED(); + } + ofoperation_destroy(op); +} + +struct rule * +ofoperation_get_victim(struct ofoperation *op) +{ + assert(op->type == OFOPERATION_ADD); + return op->victim; } static uint64_t