X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=7543e1140bbc1958b06b99407a62fcb14846cfcc;hb=e008bc13d03e2abb75be7d2e9271c3c50d0085a5;hp=9b7aec2ee266d4db11868b692fcecb753bcf10b3;hpb=9ed18e4635439cf6cf57a47e3d52b30fce235a70;p=openvswitch diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 9b7aec2e..7543e114 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" @@ -59,30 +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_unattached(struct ofproto *); +static struct ofopgroup *ofopgroup_create(struct ofproto *, 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 *, uint8_t table_id, - 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 *, + const struct ofputil_flow_mod *, + const struct ofp_header *); + +static bool handle_openflow(struct ofconn *, struct ofpbuf *); +static int handle_flow_mod__(struct ofproto *, struct ofconn *, + const struct ofputil_flow_mod *, + const struct ofp_header *); static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); +static void set_internal_devs_mtu(struct ofproto *); static void ofproto_unixctl_init(void); @@ -220,7 +285,9 @@ 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; @@ -251,26 +318,40 @@ 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->forward_bpdu = false; 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->frag_handling = OFPC_FRAG_NORMAL; hmap_init(&ofproto->ports); shash_init(&ofproto->port_by_name); ofproto->tables = NULL; ofproto->n_tables = 0; ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name); + ofproto->state = S_OPENFLOW; + list_init(&ofproto->pending); + ofproto->n_pending = 0; + 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)); ofproto_destroy__(ofproto); return error; } - assert(ofproto->n_tables > 0); + + 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); @@ -335,6 +416,33 @@ 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; + } +} + +/* If forward_bpdu is true, the NORMAL action will forward frames with + * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false, + * the NORMAL action will drop these frames. */ +void +ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu) +{ + bool old_val = ofproto->forward_bpdu; + ofproto->forward_bpdu = forward_bpdu; + if (old_val != ofproto->forward_bpdu) { + if (ofproto->ofproto_class->forward_bpdu_changed) { + ofproto->ofproto_class->forward_bpdu_changed(ofproto); + } + } +} + void ofproto_set_desc(struct ofproto *p, const char *mfr_desc, const char *hw_desc, @@ -422,6 +530,79 @@ ofproto_set_sflow(struct ofproto *ofproto, } } +/* Spanning Tree Protocol (STP) configuration. */ + +/* Configures STP on 'ofproto' using the settings defined in 's'. If + * 's' is NULL, disables STP. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +ofproto_set_stp(struct ofproto *ofproto, + const struct ofproto_stp_settings *s) +{ + return (ofproto->ofproto_class->set_stp + ? ofproto->ofproto_class->set_stp(ofproto, s) + : EOPNOTSUPP); +} + +/* Retrieves STP status of 'ofproto' and stores it in 's'. If the + * 'enabled' member of 's' is false, then the other members are not + * meaningful. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +ofproto_get_stp_status(struct ofproto *ofproto, + struct ofproto_stp_status *s) +{ + return (ofproto->ofproto_class->get_stp_status + ? ofproto->ofproto_class->get_stp_status(ofproto, s) + : EOPNOTSUPP); +} + +/* Configures STP on 'ofp_port' of 'ofproto' using the settings defined + * in 's'. The caller is responsible for assigning STP port numbers + * (using the 'port_num' member in the range of 1 through 255, inclusive) + * and ensuring there are no duplicates. If the 's' is NULL, then STP + * is disabled on the port. + * + * Returns 0 if successful, otherwise a positive errno value.*/ +int +ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port, + const struct ofproto_port_stp_settings *s) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + if (!ofport) { + VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16, + ofproto->name, ofp_port); + return ENODEV; + } + + return (ofproto->ofproto_class->set_stp_port + ? ofproto->ofproto_class->set_stp_port(ofport, s) + : EOPNOTSUPP); +} + +/* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in + * 's'. If the 'enabled' member in 's' is false, then the other members + * are not meaningful. + * + * Returns 0 if successful, otherwise a positive errno value.*/ +int +ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port, + struct ofproto_port_stp_status *s) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + if (!ofport) { + VLOG_WARN("%s: cannot get STP status on nonexistent port %"PRIu16, + ofproto->name, ofp_port); + return ENODEV; + } + + return (ofproto->ofproto_class->get_stp_port_status + ? ofproto->ofproto_class->get_stp_port_status(ofport, s) + : EOPNOTSUPP); +} + /* Connectivity Fault Management configuration. */ /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */ @@ -551,7 +732,7 @@ ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans) /* Returns true if 'aux' is a registered bundle that is currently in use as the * output for a mirror. */ bool -ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux) +ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux) { return (ofproto->ofproto_class->is_mirror_output_bundle ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux) @@ -570,10 +751,40 @@ 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_unattached(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) { - size_t i; + struct classifier *table; + + assert(list_is_empty(&ofproto->pending)); + assert(!ofproto->n_pending); connmgr_destroy(ofproto->connmgr); @@ -588,11 +799,14 @@ ofproto_destroy__(struct ofproto *ofproto) hmap_destroy(&ofproto->ports); shash_destroy(&ofproto->port_by_name); - for (i = 0; i < ofproto->n_tables; i++) { - classifier_destroy(&ofproto->tables[i]); + 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); } @@ -605,7 +819,7 @@ ofproto_destroy(struct ofproto *p) return; } - ofproto_flush_flows__(p); + ofproto_flush__(p); HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { ofport_destroy(ofport); } @@ -672,7 +886,24 @@ ofproto_run(struct ofproto *p) } } - 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; } @@ -692,7 +923,19 @@ ofproto_wait(struct ofproto *p) poll_immediate_wake(); } } - connmgr_wait(p->connmgr); + + 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 @@ -711,16 +954,7 @@ 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'. */ @@ -877,56 +1111,76 @@ ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port) * * 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, 0, actions, n_actions, 0, 0, 0, false, &rule); + const 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 ofputil_flow_mod fm; + + 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); + } +} + +/* Executes the flow modification specified in 'fm'. Returns 0 on success, an + * OpenFlow error code as encoded by ofp_mkerr() on failure, or + * OFPROTO_POSTPONE if the operation cannot be initiated now but may be retried + * later. + * + * This is a helper function for in-band control and fail-open. */ +int +ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm) +{ + return handle_flow_mod__(ofproto, NULL, fm, NULL); } /* 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. */ -void +bool 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], target)); - ofproto_rule_destroy(rule); -} - -static void -ofproto_flush_flows__(struct ofproto *ofproto) -{ - size_t i; - - COVERAGE_INC(ofproto_flush); - - if (ofproto->ofproto_class->flush) { - ofproto->ofproto_class->flush(ofproto); + 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_unattached(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; } - for (i = 0; i < ofproto->n_tables; i++) { - struct rule *rule, *next_rule; - struct cls_cursor cursor; - - cls_cursor_init(&cursor, &ofproto->tables[i], NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { - ofproto_rule_destroy(rule); - } - } } -/* Deletes all of the flows from all of ofproto's flow tables, then - * reintroduces rules required by in-band control and fail open. */ +/* 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 @@ -960,17 +1214,11 @@ 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)", @@ -1020,6 +1268,7 @@ ofport_install(struct ofproto *p, { const char *netdev_name = netdev_get_name(netdev); struct ofport *ofport; + int dev_mtu; int error; /* Create ofport. */ @@ -1038,6 +1287,13 @@ ofport_install(struct ofproto *p, hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0)); shash_add(&p->port_by_name, netdev_name, ofport); + if (!netdev_get_mtu(netdev, &dev_mtu)) { + set_internal_devs_mtu(p); + ofport->mtu = dev_mtu; + } else { + ofport->mtu = 0; + } + /* Let the ofproto_class initialize its private data. */ error = p->ofproto_class->port_construct(ofport); if (error) { @@ -1076,7 +1332,7 @@ ofport_remove_with_name(struct ofproto *ofproto, const char *name) } } -/* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'. +/* Updates 'port' with new 'opp' description. * * Does not handle a name or port number change. The caller must implement * such a change as a delete followed by an add. */ @@ -1095,6 +1351,17 @@ ofport_modified(struct ofport *port, struct ofp_phy_port *opp) connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY); } +/* Update OpenFlow 'state' in 'port' and notify controller. */ +void +ofproto_port_set_state(struct ofport *port, ovs_be32 state) +{ + if (port->opp.state != state) { + port->opp.state = state; + connmgr_send_port_status(port->ofproto->connmgr, &port->opp, + OFPPR_MODIFY); + } +} + void ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port) { @@ -1164,12 +1431,22 @@ update_port(struct ofproto *ofproto, const char *name) port = ofproto_get_port(ofproto, ofproto_port.ofp_port); if (port && !strcmp(netdev_get_name(port->netdev), name)) { struct netdev *old_netdev = port->netdev; + int dev_mtu; /* 'name' hasn't changed location. Any properties changed? */ if (!ofport_equal(&port->opp, &opp)) { ofport_modified(port, &opp); } + /* If this is a non-internal port and the MTU changed, check + * if the datapath's MTU needs to be updated. */ + if (strcmp(netdev_get_type(netdev), "internal") + && !netdev_get_mtu(netdev, &dev_mtu) + && port->mtu != dev_mtu) { + set_internal_devs_mtu(ofproto); + port->mtu = dev_mtu; + } + /* 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.*/ @@ -1225,72 +1502,53 @@ 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. */ + +/* Find the minimum MTU of all non-datapath devices attached to 'p'. + * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */ static int -rule_create(struct ofproto *ofproto, - const struct cls_rule *cls_rule, uint8_t table_id, - 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) +find_min_mtu(struct ofproto *p) { - struct rule *rule; - int error; + struct ofport *ofport; + int mtu = 0; - if (table_id == 0xff) { - if (ofproto->n_tables > 1) { - error = ofproto->ofproto_class->rule_choose_table(ofproto, - cls_rule, - &table_id); - if (error) { - return error; - } - assert(table_id < ofproto->n_tables); - } else { - table_id = 0; + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + struct netdev *netdev = ofport->netdev; + int dev_mtu; + + /* Skip any internal ports, since that's what we're trying to + * set. */ + if (!strcmp(netdev_get_type(netdev), "internal")) { + continue; } - } - rule = ofproto->ofproto_class->rule_alloc(); - if (!rule) { - error = ENOMEM; - goto error; + if (netdev_get_mtu(netdev, &dev_mtu)) { + continue; + } + if (!mtu || dev_mtu < mtu) { + mtu = dev_mtu; + } } - rule->ofproto = ofproto; - rule->cr = *cls_rule; - rule->table_id = table_id; - 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; + return mtu ? mtu: ETH_PAYLOAD_MAX; +} - error = ofproto->ofproto_class->rule_construct(rule); - if (error) { - ofproto_rule_destroy__(rule); - goto error; - } +/* Set the MTU of all datapath devices on 'p' to the minimum of the + * non-datapath ports. */ +static void +set_internal_devs_mtu(struct ofproto *p) +{ + struct ofport *ofport; + int mtu = find_min_mtu(p); - *rulep = rule; - return 0; + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + struct netdev *netdev = ofport->netdev; -error: - VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", - ofproto->name, strerror(error)); - *rulep = NULL; - return error; + if (!strcmp(netdev_get_type(netdev), "internal")) { + netdev_set_mtu(netdev, mtu); + } + } } - + static void ofproto_rule_destroy__(struct rule *rule) { @@ -1298,14 +1556,20 @@ 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 @@ -1315,13 +1579,12 @@ static bool 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 == OFPP_NONE) { return true; } - for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; - oa = actions_next(&i)) { + OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) { if (action_outputs_to_port(oa, htons(out_port))) { return true; } @@ -1402,18 +1665,12 @@ static int handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofpbuf *buf; struct ofp_switch_config *osc; - uint16_t flags; - bool drop_frags; - - /* Figure out flags. */ - drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto); - flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; + struct ofpbuf *buf; /* Send reply. */ osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); - osc->flags = htons(flags); + osc->flags = htons(ofproto->frag_handling); osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn)); ofconn_send_reply(ofconn, buf); @@ -1426,19 +1683,20 @@ handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) struct ofproto *ofproto = ofconn_get_ofproto(ofconn); uint16_t flags = ntohs(osc->flags); - if (ofconn_get_type(ofconn) == OFCONN_PRIMARY - && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) { - switch (flags & OFPC_FRAG_MASK) { - case OFPC_FRAG_NORMAL: - ofproto->ofproto_class->set_drop_frags(ofproto, false); - break; - case OFPC_FRAG_DROP: - ofproto->ofproto_class->set_drop_frags(ofproto, true); - break; - default: - VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", - osc->flags); - break; + if (ofconn_get_type(ofconn) != OFCONN_PRIMARY + || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) { + enum ofp_config_flags cur = ofproto->frag_handling; + enum ofp_config_flags next = flags & OFPC_FRAG_MASK; + + assert((cur & OFPC_FRAG_MASK) == cur); + if (cur != next) { + if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) { + ofproto->frag_handling = next; + } else { + VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s", + ofproto->name, + ofputil_frag_handling_to_string(next)); + } } } @@ -1477,7 +1735,6 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) struct ofpbuf request; struct flow flow; size_t n_ofp_actions; - uint16_t in_port; int error; COVERAGE_INC(ofproto_packet_out); @@ -1501,7 +1758,7 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) /* Get payload. */ if (opo->buffer_id != htonl(UINT32_MAX)) { error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id), - &buffer, &in_port); + &buffer, NULL); if (error || !buffer) { return error; } @@ -1744,6 +2001,9 @@ collect_rules_loose(struct ofproto *ofproto, uint8_t table_id, cls_cursor_init(&cursor, cls, match); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + 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); } @@ -1776,6 +2036,9 @@ collect_rules_strict(struct ofproto *ofproto, uint8_t table_id, 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); } @@ -1789,7 +2052,7 @@ handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_stats_msg *osm) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct flow_stats_request fsr; + struct ofputil_flow_stats_request fsr; struct list replies; struct list rules; struct rule *rule; @@ -1832,7 +2095,6 @@ 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); @@ -1847,8 +2109,8 @@ 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"); } @@ -1862,7 +2124,7 @@ ofproto_get_all_flows(struct ofproto *p, struct ds *results) { struct classifier *cls; - for (cls = &p->tables[0]; cls < &p->tables[p->n_tables]; cls++) { + OFPROTO_FOR_EACH_TABLE (cls, p) { struct cls_cursor cursor; struct rule *rule; @@ -1894,13 +2156,33 @@ ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port) : -1); } +/* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port' + * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by + * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a + * number less than 0 if CFM is not enabled on 'ofp_port'. */ +int +ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto, + uint16_t ofp_port, const uint64_t **rmps, + size_t *n_rmps) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + + *rmps = NULL; + *n_rmps = 0; + return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids + ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps, + n_rmps) + : -1); +} + static int handle_aggregate_stats_request(struct ofconn *ofconn, const struct ofp_stats_msg *osm) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct flow_stats_request request; + struct ofputil_flow_stats_request request; struct ofputil_aggregate_stats stats; + bool unknown_packets, unknown_bytes; struct ofpbuf *reply; struct list rules; struct rule *rule; @@ -1918,6 +2200,7 @@ handle_aggregate_stats_request(struct ofconn *ofconn, } 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; @@ -1925,10 +2208,26 @@ handle_aggregate_stats_request(struct ofconn *ofconn, ofproto->ofproto_class->rule_get_stats(rule, &packet_count, &byte_count); - stats.packet_count += packet_count; - stats.byte_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 = ofputil_encode_aggregate_stats_reply(&stats, osm); ofconn_send_reply(ofconn, reply); @@ -2016,70 +2315,123 @@ handle_queue_stats_request(struct ofconn *ofconn, 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. * * Adds the flow specified by 'ofm', which is followed by 'n_actions' - * ofp_actions, to the ofproto's flow table. Returns 0 on success or an - * OpenFlow error code as encoded by ofp_mkerr() on failure. + * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow + * error code as encoded by ofp_mkerr() on failure, or OFPROTO_POSTPONE if the + * operation cannot be initiated now but may be retried later. * * '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, + const struct ofputil_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) { - struct classifier *cls; - - FOR_EACH_MATCHING_TABLE (cls, fm->table_id, p) { - if (classifier_rule_overlaps(cls, &fm->cr)) { - return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); + /* 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); } - buf_err = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id, &packet, &in_port); - error = rule_create(p, &fm->cr, fm->table_id, 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; + /* Check for overlap, if requested. */ + if (fm->flags & OFPFF_CHECK_OVERLAP + && classifier_rule_overlaps(table, &fm->cr)) { + return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); } - if (packet) { - assert(!buf_err); - return rule_execute(rule, in_port, packet); + /* Serialize against pending deletion. */ + if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) { + return OFPROTO_POSTPONE; } - return buf_err; -} -static int -send_buffered_packet(struct ofconn *ofconn, - struct rule *rule, uint32_t buffer_id) -{ - struct ofpbuf *packet; - uint16_t in_port; - int error; + /* 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 = rule->modified = 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 = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); + ofoperation_create(group, rule, OFOPERATION_ADD); + rule->pending->victim = victim; - if (buffer_id == UINT32_MAX) { - return 0; + 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. */ @@ -2092,43 +2444,31 @@ send_buffered_packet(struct ofconn *ofconn, * * Returns 0 on success, otherwise an OpenFlow error code. */ static int -modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm, - struct list *rules) +modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request, struct list *rules) { - struct rule *match; + struct ofopgroup *group; struct rule *rule; - int error; - error = 0; - match = NULL; + group = ofopgroup_create(ofproto, 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)) { - int retval; - - retval = rule->ofproto->ofproto_class->rule_modify_actions( - rule, fm->actions, fm->n_actions); - if (!retval) { - match = rule; - free(rule->actions); - rule->actions = ofputil_actions_clone(fm->actions, - fm->n_actions); - rule->n_actions = fm->n_actions; - } else if (!error) { - error = retval; - } + 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); + } else { + rule->modified = time_msec(); } rule->flow_cookie = fm->cookie; } + ofopgroup_submit(group); - if (!error && 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 error; + return 0; } /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as @@ -2137,16 +2477,18 @@ modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm, * '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 ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) { - struct ofproto *p = ofconn_get_ofproto(ofconn); struct list rules; int error; - error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules); + error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, OFPP_NONE, + &rules); return (error ? error - : list_is_empty(&rules) ? add_flow(ofconn, fm) - : modify_flows__(ofconn, fm, &rules)); + : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request) + : modify_flows__(ofproto, ofconn, fm, request, &rules)); } /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error @@ -2155,16 +2497,19 @@ modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) * '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 ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) { - struct ofproto *p = ofconn_get_ofproto(ofconn); struct list rules; int error; - error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules); + error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, OFPP_NONE, + &rules); return (error ? error - : list_is_empty(&rules) ? add_flow(ofconn, fm) - : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, &rules) + : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request) + : list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn, + fm, request, &rules) : 0); } @@ -2174,43 +2519,56 @@ modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) * * Returns 0 on success, otherwise an OpenFlow error code. */ static int -delete_flows__(struct list *rules) +delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofp_header *request, struct list *rules) { struct rule *rule, *next; + struct ofopgroup *group; + group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX); LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) { ofproto_rule_send_removed(rule, OFPRR_DELETE); - ofproto_rule_destroy(rule); + + 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 0; } /* Implements OFPFC_DELETE. */ static int -delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) +delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) { struct list rules; int error; - error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port, + error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, fm->out_port, &rules); return (error ? error - : !list_is_empty(&rules) ? delete_flows__(&rules) + : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request, + &rules) : 0); } /* Implements OFPFC_DELETE_STRICT. */ static int -delete_flow_strict(struct ofproto *p, struct flow_mod *fm) +delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) { struct list rules; int error; - error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port, + error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, fm->out_port, &rules); return (error ? error - : list_is_singleton(&rules) ? delete_flows__(&rules) + : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn, + request, &rules) : 0); } @@ -2243,16 +2601,24 @@ 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_unattached(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 flow_mod fm; + struct ofputil_flow_mod fm; int error; error = reject_slave_controller(ofconn, "flow_mod"); @@ -2274,26 +2640,37 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); } - switch (fm.command) { + return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh); +} + +static int +handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *oh) +{ + if (ofproto->n_pending >= 50) { + assert(!list_is_empty(&ofproto->pending)); + return OFPROTO_POSTPONE; + } + + 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(ofproto, ofconn, fm, oh); case OFPFC_MODIFY_STRICT: - return modify_flow_strict(ofconn, &fm); + return modify_flow_strict(ofproto, ofconn, fm, oh); case OFPFC_DELETE: - delete_flows_loose(p, &fm); - return 0; + return delete_flows_loose(ofproto, ofconn, fm, oh); case OFPFC_DELETE_STRICT: - delete_flow_strict(p, &fm); - return 0; + return delete_flow_strict(ofproto, ofconn, fm, oh); default: - if (fm.command > 0xff) { + if (fm->command > 0xff) { VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but " "flow_mod_table_id extension is not enabled"); } @@ -2323,6 +2700,11 @@ 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); @@ -2351,13 +2733,18 @@ 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_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 @@ -2366,8 +2753,10 @@ 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; @@ -2449,7 +2838,7 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPUTIL_OFPST_QUEUE_REQUEST: return handle_queue_stats_request(ofconn, msg->data); - case OFPUTIL_INVALID: + case OFPUTIL_MSG_INVALID: case OFPUTIL_OFPT_HELLO: case OFPUTIL_OFPT_ERROR: case OFPUTIL_OFPT_FEATURES_REPLY: @@ -2484,14 +2873,253 @@ 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) { + 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_unattached(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 for 'ofproto'. + * + * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow + * connection. The 'request' and 'buffer_id' arguments are ignored. + * + * If 'ofconn' is nonnull, then the new ofopgroup 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(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofp_header *request, uint32_t buffer_id) +{ + struct ofopgroup *group = ofopgroup_create_unattached(ofproto); + if (ofconn) { + size_t request_len = ntohs(request->length); + + assert(ofconn_get_ofproto(ofconn) == ofproto); + + 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); + group->ofproto->n_pending++; + } +} + +static void +ofopgroup_destroy(struct ofopgroup *group) +{ + assert(list_is_empty(&group->ops)); + if (!list_is_empty(&group->ofproto_node)) { + assert(group->ofproto->n_pending > 0); + group->ofproto->n_pending--; + 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 'error' is 0, indicating success, the operation will be committed + * permanently to the flow table. There is one interesting subcase: + * + * - If 'op' is an "add flow" operation that is replacing an existing rule in + * the flow table (the "victim" rule) by a new one, then the caller must + * have uninitialized any derived state in the victim rule, as in step 5 in + * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete() + * performs steps 6 and 7 for the victim rule, most notably by calling its + * ->rule_dealloc() function. + * + * If 'error' is nonzero, then generally the operation will be rolled back: + * + * - If 'op' is an "add flow" operation, ofproto removes the new rule or + * restores the original rule. The caller must have uninitialized any + * derived state in the new rule, as in step 5 of in the "Life Cycle" in + * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and + * and 7 for the new rule, calling its ->rule_dealloc() function. + * + * - If 'op' is a "modify flow" operation, ofproto restores the original + * actions. + * + * - 'op' must not be a "delete flow" operation. Removing a rule is not + * allowed to fail. It must always succeed. + * + * 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) { + rule->modified = time_msec(); + } else { + 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 @@ -2562,5 +3190,5 @@ ofproto_unixctl_init(void) } registered = true; - unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL); + unixctl_command_register("ofproto/list", "", ofproto_unixctl_list, NULL); }