From: Ben Pfaff Date: Thu, 12 May 2011 19:05:42 +0000 (-0700) Subject: Merge 'master' into 'next'. X-Git-Tag: v1.2.0~329^2~14 X-Git-Url: http://git.onelab.eu/?p=sliver-openvswitch.git;a=commitdiff_plain;h=007948177581f3b3dad188221593d0e4bdca6ba0 Merge 'master' into 'next'. --- 007948177581f3b3dad188221593d0e4bdca6ba0 diff --cc include/openflow/nicira-ext.h index 7df978b0c,241cfe99f..738fd90d8 --- a/include/openflow/nicira-ext.h +++ b/include/openflow/nicira-ext.h @@@ -177,63 -167,6 +174,53 @@@ enum nicira_stats_type NXST_AGGREGATE /* Analogous to OFPST_AGGREGATE. */ }; - /* NXT_TUN_ID_FROM_COOKIE request. */ - struct nxt_tun_id_cookie { - struct ofp_header header; - ovs_be32 vendor; /* NX_VENDOR_ID. */ - ovs_be32 subtype; /* NXT_TUN_ID_FROM_COOKIE */ - uint8_t set; /* Nonzero to enable, zero to disable. */ - uint8_t pad[7]; - }; - OFP_ASSERT(sizeof(struct nxt_tun_id_cookie) == 24); - +/* This command enables or disables an Open vSwitch extension that allows a + * controller to specify the OpenFlow table to which a flow should be added, + * instead of having the switch decide which table is most appropriate as + * required by OpenFlow 1.0. By default, the extension is disabled. + * + * When this feature is enabled, Open vSwitch treats struct ofp_flow_mod's + * 16-bit 'command' member as two separate fields. The upper 8 bits are used + * as the table ID, the lower 8 bits specify the command as usual. A table ID + * of 0xff is treated like a wildcarded table ID. + * + * The specific treatment of the table ID depends on the type of flow mod: + * + * - OFPFC_ADD: Given a specific table ID, the flow is always placed in that + * table. If an identical flow already exists in that table only, then it + * is replaced. If the flow cannot be placed in the specified table, + * either because the table is full or because the table cannot support + * flows of the given type, the switch replies with an + * OFPFMFC_ALL_TABLES_FULL error. (A controller can distinguish these + * cases by comparing the current and maximum number of entries reported + * in ofp_table_stats.) + * + * If the table ID is wildcarded, the switch picks an appropriate table + * itself. If an identical flow already exist in the selected flow table, + * then it is replaced. The choice of table might depend on the flows + * that are already in the switch; for example, if one table fills up then + * the switch might fall back to another one. + * + * - OFPFC_MODIFY, OFPFC_DELETE: Given a specific table ID, only flows + * within that table are matched and modified or deleted. If the table ID + * is wildcarded, flows within any table may be matched and modified or + * deleted. + * + * - OFPFC_MODIFY_STRICT, OFPFC_DELETE_STRICT: Given a specific table ID, + * only a flow within that table may be matched and modified or deleted. + * If the table ID is wildcarded and exactly one flow within any table + * matches, then it is modified or deleted; if flows in more than one + * table match, then none is modified or deleted. + */ +struct nxt_flow_mod_table_id { + struct ofp_header header; + uint32_t vendor; /* NX_VENDOR_ID. */ + uint32_t subtype; /* NXT_FLOW_MOD_TABLE_ID. */ + uint8_t set; /* Nonzero to enable, zero to disable. */ + uint8_t pad[7]; +}; +OFP_ASSERT(sizeof(struct nxt_flow_mod_table_id) == 24); + /* Configures the "role" of the sending controller. The default role is: * * - Other (NX_ROLE_OTHER), which allows the controller access to all diff --cc lib/bond.h index 8736f4cd4,c2efb8b02..fe587928d --- a/lib/bond.h +++ b/lib/bond.h @@@ -74,8 -75,7 +75,8 @@@ void bond_destroy(struct bond *) bool bond_reconfigure(struct bond *, const struct bond_settings *); void bond_slave_register(struct bond *, void *slave_, - uint16_t stable_id, struct netdev *); + uint32_t stable_id, struct netdev *); +void bond_slave_set_netdev(struct bond *, void *slave_, struct netdev *); void bond_slave_unregister(struct bond *, const void *slave); void bond_run(struct bond *, struct tag_set *, bool lacp_negotiated); diff --cc lib/flow.h index f5f965c5b,60229f58d..6bd2c3d13 --- a/lib/flow.h +++ b/lib/flow.h @@@ -118,7 -118,7 +118,7 @@@ typedef unsigned int OVS_BITWISE flow_w #define FWW_TP_DST ((OVS_FORCE flow_wildcards_t) (1 << 7)) /* Same meanings as corresponding OFPFW_* bits, but differ in value. */ #define FWW_NW_TOS ((OVS_FORCE flow_wildcards_t) (1 << 1)) --/* No corresponding OFPFW_* or OVSFW_* bits. */ ++/* No corresponding OFPFW_* bits. */ #define FWW_ETH_MCAST ((OVS_FORCE flow_wildcards_t) (1 << 8)) /* multicast bit only */ #define FWW_ARP_SHA ((OVS_FORCE flow_wildcards_t) (1 << 9)) diff --cc lib/learning-switch.c index 75e04df62,e003b8f99..6bd228623 --- a/lib/learning-switch.c +++ b/lib/learning-switch.c @@@ -238,10 -238,8 +238,9 @@@ lswitch_process_packet(struct lswitch * case OFPUTIL_OFPST_PORT_REPLY: case OFPUTIL_OFPST_TABLE_REPLY: case OFPUTIL_OFPST_AGGREGATE_REPLY: - case OFPUTIL_NXT_TUN_ID_FROM_COOKIE: case OFPUTIL_NXT_ROLE_REQUEST: case OFPUTIL_NXT_ROLE_REPLY: + case OFPUTIL_NXT_FLOW_MOD_TABLE_ID: case OFPUTIL_NXT_SET_FLOW_FORMAT: case OFPUTIL_NXT_FLOW_MOD: case OFPUTIL_NXT_FLOW_REMOVED: diff --cc lib/ofp-parse.c index c7ff4c323,f45c450c6..29137ac9d --- a/lib/ofp-parse.c +++ b/lib/ofp-parse.c @@@ -850,10 -856,10 +850,10 @@@ parse_ofp_flow_mod_str(struct list *pac struct flow_mod fm; ofpbuf_init(&actions, 64); - parse_ofp_str(&fm, NULL, is_del ? NULL : &actions, string); + parse_ofp_str(&fm, is_del ? NULL : &actions, string); fm.command = command; - min_format = ofputil_min_flow_format(&fm.cr, true, fm.cookie); + min_format = ofputil_min_flow_format(&fm.cr); next_format = MAX(*cur_format, min_format); if (next_format != *cur_format) { struct ofpbuf *sff = ofputil_make_set_flow_format(next_format); diff --cc lib/ofp-print.c index 5b04d1923,65dea728b..cbac1e44a --- a/lib/ofp-print.c +++ b/lib/ofp-print.c @@@ -837,7 -834,7 +834,7 @@@ ofp_print_flow_mod(struct ds *s, const bool need_priority; int error; - error = ofputil_decode_flow_mod(&fm, oh, NXFF_OPENFLOW10, true); - error = ofputil_decode_flow_mod(&fm, oh); ++ error = ofputil_decode_flow_mod(&fm, oh, true); if (error) { ofp_print_error(s, error); return; diff --cc lib/ofp-util.c index 962727cc9,ddac77270..3487cbac1 --- a/lib/ofp-util.c +++ b/lib/ofp-util.c @@@ -408,14 -361,7 +370,10 @@@ static in ofputil_decode_vendor(const struct ofp_header *oh, const struct ofputil_msg_type **typep) { + BUILD_ASSERT_DECL(sizeof(struct nxt_set_flow_format) + != sizeof(struct nxt_flow_mod_table_id)); + static const struct ofputil_msg_type nxt_messages[] = { - { OFPUTIL_NXT_TUN_ID_FROM_COOKIE, - NXT_TUN_ID_FROM_COOKIE, "NXT_TUN_ID_FROM_COOKIE", - sizeof(struct nxt_tun_id_cookie), 0 }, - { OFPUTIL_NXT_ROLE_REQUEST, NXT_ROLE_REQUEST, "NXT_ROLE_REQUEST", sizeof(struct nx_role_request), 0 }, @@@ -997,21 -856,11 +885,16 @@@ ofputil_make_flow_mod_table_id(bool flo * flow_mod in 'fm'. Returns 0 if successful, otherwise an OpenFlow error * code. * - * For OFPT_FLOW_MOD messages, 'flow_format' should be the current flow format - * at the time when the message was received. Otherwise 'flow_format' is - * ignored. - * + * 'flow_mod_table_id' should be true if the NXT_FLOW_MOD_TABLE_ID extension is + * enabled, false otherwise. + * * Does not validate the flow_mod actions. */ int -ofputil_decode_flow_mod(struct flow_mod *fm, const struct ofp_header *oh) +ofputil_decode_flow_mod(struct flow_mod *fm, const struct ofp_header *oh, - enum nx_flow_format flow_format, + bool flow_mod_table_id) { const struct ofputil_msg_type *type; + uint16_t command; struct ofpbuf b; ofpbuf_use_const(&b, oh, ntohs(oh->length)); @@@ -1049,10 -898,9 +932,9 @@@ } /* Translate the message. */ - ofputil_cls_rule_from_match(&match, ntohs(ofm->priority), flow_format, - ofm->cookie, &fm->cr); + ofputil_cls_rule_from_match(&match, ntohs(ofm->priority), &fm->cr); fm->cookie = ofm->cookie; - fm->command = ntohs(ofm->command); + command = ntohs(ofm->command); fm->idle_timeout = ntohs(ofm->idle_timeout); fm->hard_timeout = ntohs(ofm->hard_timeout); fm->buffer_id = ntohl(ofm->buffer_id); @@@ -1110,21 -946,15 +992,19 @@@ ofputil_encode_flow_mod(const struct fl { size_t actions_len = fm->n_actions * sizeof *fm->actions; struct ofpbuf *msg; + uint16_t command; + + command = (flow_mod_table_id + ? (fm->command & 0xff) | (fm->table_id << 8) + : fm->command); - if (flow_format == NXFF_OPENFLOW10 - || flow_format == NXFF_TUN_ID_FROM_COOKIE) { + if (flow_format == NXFF_OPENFLOW10) { struct ofp_flow_mod *ofm; msg = ofpbuf_new(sizeof *ofm + actions_len); ofm = put_openflow(sizeof *ofm, OFPT_FLOW_MOD, msg); - ofputil_cls_rule_to_match(&fm->cr, flow_format, &ofm->match, - fm->cookie, &ofm->cookie); - ofm->command = htons(command); + ofputil_cls_rule_to_match(&fm->cr, &ofm->match); - ofm->cookie = fm->cookie; + ofm->command = htons(fm->command); ofm->idle_timeout = htons(fm->idle_timeout); ofm->hard_timeout = htons(fm->hard_timeout); ofm->priority = htons(fm->cr.priority); diff --cc lib/ofp-util.h index 55c2c5889,6e69bae05..c87b04b74 --- a/lib/ofp-util.h +++ b/lib/ofp-util.h @@@ -141,11 -130,9 +135,11 @@@ struct flow_mod size_t n_actions; }; -int ofputil_decode_flow_mod(struct flow_mod *, const struct ofp_header *); +int ofputil_decode_flow_mod(struct flow_mod *, const struct ofp_header *, - enum nx_flow_format, bool flow_mod_table_id); ++ bool flow_mod_table_id); struct ofpbuf *ofputil_encode_flow_mod(const struct flow_mod *, - enum nx_flow_format); + enum nx_flow_format, + bool flow_mod_table_id); /* Flow stats or aggregate stats request, independent of flow format. */ struct flow_stats_request { diff --cc ofproto/ofproto-dpif.c index ea0e9e491,000000000..93acd429b mode 100644,000000..100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@@ -1,3932 -1,0 +1,3934 @@@ +/* + * Copyright (c) 2009, 2010, 2011 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include + +#include "ofproto/private.h" + +#include + +#include "autopath.h" +#include "bond.h" +#include "byte-order.h" +#include "connmgr.h" +#include "coverage.h" +#include "cfm.h" +#include "dpif.h" +#include "dynamic-string.h" +#include "fail-open.h" +#include "hmapx.h" +#include "lacp.h" +#include "mac-learning.h" +#include "multipath.h" +#include "netdev.h" +#include "netlink.h" +#include "nx-match.h" +#include "odp-util.h" +#include "ofp-util.h" +#include "ofpbuf.h" +#include "ofp-print.h" +#include "ofproto-sflow.h" +#include "poll-loop.h" +#include "timer.h" +#include "unaligned.h" +#include "unixctl.h" +#include "vlan-bitmap.h" +#include "vlog.h" + +VLOG_DEFINE_THIS_MODULE(ofproto_dpif); + +COVERAGE_DEFINE(ofproto_dpif_ctlr_action); +COVERAGE_DEFINE(ofproto_dpif_expired); +COVERAGE_DEFINE(ofproto_dpif_no_packet_in); +COVERAGE_DEFINE(ofproto_dpif_xlate); +COVERAGE_DEFINE(facet_changed_rule); +COVERAGE_DEFINE(facet_invalidated); +COVERAGE_DEFINE(facet_revalidate); +COVERAGE_DEFINE(facet_unexpected); + +/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a + * flow translation. */ +#define MAX_RESUBMIT_RECURSION 16 + +struct ofport_dpif; +struct ofproto_dpif; + +struct rule_dpif { + struct rule up; + + long long int used; /* Time last used; time created if not used. */ + + /* These statistics: + * + * - Do include packets and bytes from facets that have been deleted or + * whose own statistics have been folded into the rule. + * + * - Do include packets and bytes sent "by hand" that were accounted to + * the rule without any facet being involved (this is a rare corner + * case in rule_execute()). + * + * - Do not include packet or bytes that can be obtained from any facet's + * packet_count or byte_count member or that can be obtained from the + * datapath by, e.g., dpif_flow_get() for any facet. + */ + uint64_t packet_count; /* Number of packets received. */ + uint64_t byte_count; /* Number of bytes received. */ + + struct list facets; /* List of "struct facet"s. */ +}; + +static struct rule_dpif *rule_dpif_cast(const struct rule *rule) +{ + return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; +} + +static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *ofproto, + const struct flow *flow); + +#define MAX_MIRRORS 32 +typedef uint32_t mirror_mask_t; +#define MIRROR_MASK_C(X) UINT32_C(X) +BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); +struct ofmirror { + struct ofproto_dpif *ofproto; /* Owning ofproto. */ + size_t idx; /* In ofproto's "mirrors" array. */ + void *aux; /* Key supplied by ofproto's client. */ + char *name; /* Identifier for log messages. */ + + /* Selection criteria. */ + struct hmapx srcs; /* Contains "struct ofbundle *"s. */ + struct hmapx dsts; /* Contains "struct ofbundle *"s. */ + unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ + + /* Output (mutually exclusive). */ + struct ofbundle *out; /* Output port or NULL. */ + int out_vlan; /* Output VLAN or -1. */ +}; + +static void mirror_destroy(struct ofmirror *); + +/* A group of one or more OpenFlow ports. */ +#define OFBUNDLE_FLOOD ((struct ofbundle *) 1) +struct ofbundle { + struct ofproto_dpif *ofproto; /* Owning ofproto. */ + struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ + void *aux; /* Key supplied by ofproto's client. */ + char *name; /* Identifier for log messages. */ + + /* Configuration. */ + struct list ports; /* Contains "struct ofport"s. */ + int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ + unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. + * NULL if all VLANs are trunked. */ + struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ + struct bond *bond; /* Nonnull iff more than one port. */ + + /* Status. */ + bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */ + + /* Port mirroring info. */ + mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ + mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ + mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ +}; + +static void bundle_remove(struct ofport *); +static void bundle_destroy(struct ofbundle *); +static void bundle_del_port(struct ofport_dpif *); +static void bundle_run(struct ofbundle *); +static void bundle_wait(struct ofbundle *); + +struct action_xlate_ctx { +/* action_xlate_ctx_init() initializes these members. */ + + /* The ofproto. */ + struct ofproto_dpif *ofproto; + + /* Flow to which the OpenFlow actions apply. xlate_actions() will modify + * this flow when actions change header fields. */ + struct flow flow; + + /* The packet corresponding to 'flow', or a null pointer if we are + * revalidating without a packet to refer to. */ + const struct ofpbuf *packet; + + /* If nonnull, called just before executing a resubmit action. + * + * This is normally null so the client has to set it manually after + * calling action_xlate_ctx_init(). */ + void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); + +/* xlate_actions() initializes and uses these members. The client might want + * to look at them after it returns. */ + + struct ofpbuf *odp_actions; /* Datapath actions. */ + tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ + bool may_set_up_flow; /* True ordinarily; false if the actions must + * be reassessed for every packet. */ + uint16_t nf_output_iface; /* Output interface index for NetFlow. */ + +/* xlate_actions() initializes and uses these members, but the client has no + * reason to look at them. */ + + int recurse; /* Recursion level, via xlate_table_action. */ + int last_pop_priority; /* Offset in 'odp_actions' just past most + * recent ODP_ACTION_ATTR_SET_PRIORITY. */ +}; + +static void action_xlate_ctx_init(struct action_xlate_ctx *, + struct ofproto_dpif *, const struct flow *, + const struct ofpbuf *); +static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, + const union ofp_action *in, size_t n_in); + +/* An exact-match instantiation of an OpenFlow flow. */ +struct facet { + long long int used; /* Time last used; time created if not used. */ + + /* These statistics: + * + * - Do include packets and bytes sent "by hand", e.g. with + * dpif_execute(). + * + * - Do include packets and bytes that were obtained from the datapath + * when a flow was deleted (e.g. dpif_flow_del()) or when its + * statistics were reset (e.g. dpif_flow_put() with + * DPIF_FP_ZERO_STATS). + * + * - Do not include any packets or bytes that can currently be obtained + * from the datapath by, e.g., dpif_flow_get(). + */ + uint64_t packet_count; /* Number of packets received. */ + uint64_t byte_count; /* Number of bytes received. */ + + uint64_t dp_packet_count; /* Last known packet count in the datapath. */ + uint64_t dp_byte_count; /* Last known byte count in the datapath. */ + + uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ + uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ + long long int rs_used; /* Used time pushed to resubmit children. */ + + /* Number of bytes passed to account_cb. This may include bytes that can + * currently obtained from the datapath (thus, it can be greater than + * byte_count). */ + uint64_t accounted_bytes; + + struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ + struct list list_node; /* In owning rule's 'facets' list. */ + struct rule_dpif *rule; /* Owning rule. */ + struct flow flow; /* Exact-match flow. */ + bool installed; /* Installed in datapath? */ + bool may_install; /* True ordinarily; false if actions must + * be reassessed for every packet. */ + size_t actions_len; /* Number of bytes in actions[]. */ + struct nlattr *actions; /* Datapath actions. */ + tag_type tags; /* Tags. */ + struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ +}; + +static struct facet *facet_create(struct rule_dpif *, const struct flow *, + const struct ofpbuf *packet); +static void facet_remove(struct ofproto_dpif *, struct facet *); +static void facet_free(struct facet *); + +static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); +static struct facet *facet_lookup_valid(struct ofproto_dpif *, + const struct flow *); +static bool facet_revalidate(struct ofproto_dpif *, struct facet *); + +static void facet_execute(struct ofproto_dpif *, struct facet *, + struct ofpbuf *packet); + +static int facet_put__(struct ofproto_dpif *, struct facet *, + const struct nlattr *actions, size_t actions_len, + struct dpif_flow_stats *); +static void facet_install(struct ofproto_dpif *, struct facet *, + bool zero_stats); +static void facet_uninstall(struct ofproto_dpif *, struct facet *); +static void facet_flush_stats(struct ofproto_dpif *, struct facet *); + +static void facet_make_actions(struct ofproto_dpif *, struct facet *, + const struct ofpbuf *packet); +static void facet_update_time(struct ofproto_dpif *, struct facet *, + long long int used); +static void facet_update_stats(struct ofproto_dpif *, struct facet *, + const struct dpif_flow_stats *); +static void facet_push_stats(struct facet *); +static void facet_account(struct ofproto_dpif *, struct facet *, + uint64_t extra_bytes); + +static bool facet_is_controller_flow(struct facet *); + +static void flow_push_stats(const struct rule_dpif *, + struct flow *, uint64_t packets, uint64_t bytes, + long long int used); + +struct ofport_dpif { + struct ofport up; + + uint32_t odp_port; + struct ofbundle *bundle; /* Bundle that contains this port, if any. */ + struct list bundle_node; /* In struct ofbundle's "ports" list. */ + struct cfm *cfm; /* Connectivity Fault Management, if any. */ + tag_type tag; /* Tag associated with this port. */ ++ uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ +}; + +static struct ofport_dpif * +ofport_dpif_cast(const struct ofport *ofport) +{ + assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); + return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; +} + +static void port_run(struct ofport_dpif *); +static void port_wait(struct ofport_dpif *); +static int set_cfm(struct ofport *, const struct cfm *, + const uint16_t *remote_mps, size_t n_remote_mps); + +struct ofproto_dpif { + struct ofproto up; + struct dpif *dpif; + int max_ports; + + /* Statistics. */ + uint64_t n_matches; + + /* Bridging. */ + struct netflow *netflow; + struct ofproto_sflow *sflow; + struct hmap bundles; /* Contains "struct ofbundle"s. */ + struct mac_learning *ml; + struct ofmirror *mirrors[MAX_MIRRORS]; + bool has_bonded_bundles; + + /* Expiration. */ + struct timer next_expiration; + + /* Facets. */ + struct hmap facets; + bool need_revalidate; + struct tag_set revalidate_set; +}; + +static void ofproto_dpif_unixctl_init(void); + +static struct ofproto_dpif * +ofproto_dpif_cast(const struct ofproto *ofproto) +{ + assert(ofproto->ofproto_class == &ofproto_dpif_class); + return CONTAINER_OF(ofproto, struct ofproto_dpif, up); +} + +static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, + uint16_t ofp_port); +static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, + uint32_t odp_port); + +/* Packet processing. */ +static void update_learning_table(struct ofproto_dpif *, + const struct flow *, int vlan, + struct ofbundle *); +static bool is_admissible(struct ofproto_dpif *, const struct flow *, + bool have_packet, tag_type *, int *vlanp, + struct ofbundle **in_bundlep); +static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *); + +/* Flow expiration. */ +static int expire(struct ofproto_dpif *); + +/* Utilities. */ +static int send_packet(struct ofproto_dpif *, + uint32_t odp_port, uint16_t vlan_tci, + const struct ofpbuf *packet); + +/* Global variables. */ +static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + +/* Factory functions. */ + +static void +enumerate_types(struct sset *types) +{ + dp_enumerate_types(types); +} + +static int +enumerate_names(const char *type, struct sset *names) +{ + return dp_enumerate_names(type, names); +} + +static int +del(const char *type, const char *name) +{ + struct dpif *dpif; + int error; + + error = dpif_open(name, type, &dpif); + if (!error) { + error = dpif_delete(dpif); + dpif_close(dpif); + } + return error; +} + +/* Basic life-cycle. */ + +static struct ofproto * +alloc(void) +{ + struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); + return &ofproto->up; +} + +static void +dealloc(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + free(ofproto); +} + +static int +construct(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + const char *name = ofproto->up.name; + int error; + int i; + + error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); + if (error) { + VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); + return error; + } + + ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); + ofproto->n_matches = 0; + + error = dpif_recv_set_mask(ofproto->dpif, + ((1u << DPIF_UC_MISS) | + (1u << DPIF_UC_ACTION) | + (1u << DPIF_UC_SAMPLE))); + if (error) { + VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); + dpif_close(ofproto->dpif); + return error; + } + dpif_flow_flush(ofproto->dpif); + dpif_recv_purge(ofproto->dpif); + + ofproto->netflow = NULL; + ofproto->sflow = NULL; + hmap_init(&ofproto->bundles); + ofproto->ml = mac_learning_create(); + for (i = 0; i < MAX_MIRRORS; i++) { + ofproto->mirrors[i] = NULL; + } + ofproto->has_bonded_bundles = false; + + timer_set_duration(&ofproto->next_expiration, 1000); + + hmap_init(&ofproto->facets); + ofproto->need_revalidate = false; + tag_set_init(&ofproto->revalidate_set); + + ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables); + classifier_init(&ofproto->up.tables[0]); + ofproto->up.n_tables = 1; + + ofproto_dpif_unixctl_init(); + + return 0; +} + +static void +destruct(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + int i; + + for (i = 0; i < MAX_MIRRORS; i++) { + mirror_destroy(ofproto->mirrors[i]); + } + + netflow_destroy(ofproto->netflow); + ofproto_sflow_destroy(ofproto->sflow); + hmap_destroy(&ofproto->bundles); + mac_learning_destroy(ofproto->ml); + + hmap_destroy(&ofproto->facets); + + dpif_close(ofproto->dpif); +} + +static int +run(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct ofport_dpif *ofport; + struct ofbundle *bundle; + int i; + + dpif_run(ofproto->dpif); + + for (i = 0; i < 50; i++) { + struct dpif_upcall packet; + int error; + + error = dpif_recv(ofproto->dpif, &packet); + if (error) { + if (error == ENODEV) { + /* Datapath destroyed. */ + return error; + } + break; + } + + handle_upcall(ofproto, &packet); + } + + if (timer_expired(&ofproto->next_expiration)) { + int delay = expire(ofproto); + timer_set_duration(&ofproto->next_expiration, delay); + } + + if (ofproto->netflow) { + netflow_run(ofproto->netflow); + } + if (ofproto->sflow) { + ofproto_sflow_run(ofproto->sflow); + } + + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + port_run(ofport); + } + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + bundle_run(bundle); + } + + /* Now revalidate if there's anything to do. */ + if (ofproto->need_revalidate + || !tag_set_is_empty(&ofproto->revalidate_set)) { + struct tag_set revalidate_set = ofproto->revalidate_set; + bool revalidate_all = ofproto->need_revalidate; + struct facet *facet, *next; + + /* Clear the revalidation flags. */ + tag_set_init(&ofproto->revalidate_set); + ofproto->need_revalidate = false; + + HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { + if (revalidate_all + || tag_set_intersects(&revalidate_set, facet->tags)) { + facet_revalidate(ofproto, facet); + } + } + } + + return 0; +} + +static void +wait(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct ofport_dpif *ofport; + struct ofbundle *bundle; + + dpif_wait(ofproto->dpif); + dpif_recv_wait(ofproto->dpif); + if (ofproto->sflow) { + ofproto_sflow_wait(ofproto->sflow); + } + if (!tag_set_is_empty(&ofproto->revalidate_set)) { + poll_immediate_wake(); + } + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + port_wait(ofport); + } + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + bundle_wait(bundle); + } + if (ofproto->need_revalidate) { + /* Shouldn't happen, but if it does just go around again. */ + VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); + poll_immediate_wake(); + } else { + timer_wait(&ofproto->next_expiration); + } +} + +static void +flush(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct facet *facet, *next_facet; + + HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { + /* Mark the facet as not installed so that facet_remove() doesn't + * bother trying to uninstall it. There is no point in uninstalling it + * individually since we are about to blow away all the facets with + * dpif_flow_flush(). */ + facet->installed = false; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; + facet_remove(ofproto, facet); + } + dpif_flow_flush(ofproto->dpif); +} + +static void +get_features(struct ofproto *ofproto_ OVS_UNUSED, + bool *arp_match_ip, uint32_t *actions) +{ + *arp_match_ip = true; + *actions = ((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)); +} + +static void +get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct odp_stats s; + + strcpy(ots->name, "classifier"); + + dpif_get_dp_stats(ofproto->dpif, &s); + put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); + put_32aligned_be64(&ots->matched_count, + htonll(s.n_hit + ofproto->n_matches)); +} + +static int +set_netflow(struct ofproto *ofproto_, + const struct netflow_options *netflow_options) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + + if (netflow_options) { + if (!ofproto->netflow) { + ofproto->netflow = netflow_create(); + } + return netflow_set_options(ofproto->netflow, netflow_options); + } else { + netflow_destroy(ofproto->netflow); + ofproto->netflow = NULL; + return 0; + } +} + +static struct ofport * +port_alloc(void) +{ + struct ofport_dpif *port = xmalloc(sizeof *port); + return &port->up; +} + +static void +port_dealloc(struct ofport *port_) +{ + struct ofport_dpif *port = ofport_dpif_cast(port_); + free(port); +} + +static int +port_construct(struct ofport *port_) +{ + struct ofport_dpif *port = ofport_dpif_cast(port_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); + + port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); + port->bundle = NULL; + port->cfm = NULL; + port->tag = tag_create_random(); + + if (ofproto->sflow) { + ofproto_sflow_add_port(ofproto->sflow, port->odp_port, + netdev_get_name(port->up.netdev)); + } + + return 0; +} + +static void +port_destruct(struct ofport *port_) +{ + struct ofport_dpif *port = ofport_dpif_cast(port_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); + + bundle_remove(port_); + set_cfm(port_, NULL, NULL, 0); + if (ofproto->sflow) { + ofproto_sflow_del_port(ofproto->sflow, port->odp_port); + } +} + +static void +port_modified(struct ofport *port_) +{ + struct ofport_dpif *port = ofport_dpif_cast(port_); + + if (port->bundle && port->bundle->bond) { + bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); + } +} + +static void +port_reconfigured(struct ofport *port_, ovs_be32 old_config) +{ + struct ofport_dpif *port = ofport_dpif_cast(port_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); + ovs_be32 changed = old_config ^ port->up.opp.config; + + if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP | + OFPPC_NO_FWD | OFPPC_NO_FLOOD)) { + ofproto->need_revalidate = true; + } +} + +static int +set_sflow(struct ofproto *ofproto_, + const struct ofproto_sflow_options *sflow_options) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct ofproto_sflow *os = ofproto->sflow; + if (sflow_options) { + if (!os) { + struct ofport_dpif *ofport; + + os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + ofproto_sflow_add_port(os, ofport->odp_port, + netdev_get_name(ofport->up.netdev)); + } + } + ofproto_sflow_set_options(os, sflow_options); + } else { + ofproto_sflow_destroy(os); + ofproto->sflow = NULL; + } + return 0; +} + +static int +set_cfm(struct ofport *ofport_, const struct cfm *cfm, + const uint16_t *remote_mps, size_t n_remote_mps) +{ + struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + int error; + + if (!cfm) { + error = 0; + } else { + if (!ofport->cfm) { + ofport->cfm = cfm_create(); + } + + ofport->cfm->mpid = cfm->mpid; + ofport->cfm->interval = cfm->interval; + memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN); + + cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps); + + if (cfm_configure(ofport->cfm)) { + return 0; + } + + error = EINVAL; + } + cfm_destroy(ofport->cfm); + ofport->cfm = NULL; + return error; +} + +static int +get_cfm(struct ofport *ofport_, const struct cfm **cfmp) +{ + struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + *cfmp = ofport->cfm; + return 0; +} + +/* Bundles. */ + +/* Expires all MAC learning entries associated with 'port' and forces ofproto + * to revalidate every flow. */ +static void +bundle_flush_macs(struct ofbundle *bundle) +{ + struct ofproto_dpif *ofproto = bundle->ofproto; + struct mac_learning *ml = ofproto->ml; + struct mac_entry *mac, *next_mac; + + ofproto->need_revalidate = true; + LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { + if (mac->port.p == bundle) { + mac_learning_expire(ml, mac); + } + } +} + +static struct ofbundle * +bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) +{ + struct ofbundle *bundle; + + HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), + &ofproto->bundles) { + if (bundle->aux == aux) { + return bundle; + } + } + return NULL; +} + +/* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the + * ones that are found to 'bundles'. */ +static void +bundle_lookup_multiple(struct ofproto_dpif *ofproto, + void **auxes, size_t n_auxes, + struct hmapx *bundles) +{ + size_t i; + + hmapx_init(bundles); + for (i = 0; i < n_auxes; i++) { + struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); + if (bundle) { + hmapx_add(bundles, bundle); + } + } +} + +static void +bundle_del_port(struct ofport_dpif *port) +{ + struct ofbundle *bundle = port->bundle; + + list_remove(&port->bundle_node); + port->bundle = NULL; + + if (bundle->lacp) { + lacp_slave_unregister(bundle->lacp, port); + } + if (bundle->bond) { + bond_slave_unregister(bundle->bond, port); + } + + bundle->floodable = true; + LIST_FOR_EACH (port, bundle_node, &bundle->ports) { + if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { + bundle->floodable = false; + } + } +} + +static bool +bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, - struct lacp_slave_settings *lacp) ++ struct lacp_slave_settings *lacp, ++ uint32_t bond_stable_id) +{ + struct ofport_dpif *port; + + port = get_ofp_port(bundle->ofproto, ofp_port); + if (!port) { + return false; + } + + if (port->bundle != bundle) { + if (port->bundle) { + bundle_del_port(port); + } + + port->bundle = bundle; + list_push_back(&bundle->ports, &port->bundle_node); + if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { + bundle->floodable = false; + } + } + if (lacp) { + lacp_slave_register(bundle->lacp, port, lacp); + } + ++ port->bond_stable_id = bond_stable_id; ++ + return true; +} + +static void +bundle_destroy(struct ofbundle *bundle) +{ + struct ofproto_dpif *ofproto; + struct ofport_dpif *port, *next_port; + int i; + + if (!bundle) { + return; + } + + ofproto = bundle->ofproto; + for (i = 0; i < MAX_MIRRORS; i++) { + struct ofmirror *m = ofproto->mirrors[i]; + if (m) { + if (m->out == bundle) { + mirror_destroy(m); + } else if (hmapx_find_and_delete(&m->srcs, bundle) + || hmapx_find_and_delete(&m->dsts, bundle)) { + ofproto->need_revalidate = true; + } + } + } + + LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { + bundle_del_port(port); + } + + bundle_flush_macs(bundle); + hmap_remove(&ofproto->bundles, &bundle->hmap_node); + free(bundle->name); + free(bundle->trunks); + lacp_destroy(bundle->lacp); + bond_destroy(bundle->bond); + free(bundle); +} + +static int +bundle_set(struct ofproto *ofproto_, void *aux, + const struct ofproto_bundle_settings *s) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + bool need_flush = false; + const unsigned long *trunks; + struct ofport_dpif *port; + struct ofbundle *bundle; + size_t i; + bool ok; + + if (!s) { + bundle_destroy(bundle_lookup(ofproto, aux)); + return 0; + } + + assert(s->n_slaves == 1 || s->bond != NULL); + assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); + + bundle = bundle_lookup(ofproto, aux); + if (!bundle) { + bundle = xmalloc(sizeof *bundle); + + bundle->ofproto = ofproto; + hmap_insert(&ofproto->bundles, &bundle->hmap_node, + hash_pointer(aux, 0)); + bundle->aux = aux; + bundle->name = NULL; + + list_init(&bundle->ports); + bundle->vlan = -1; + bundle->trunks = NULL; + bundle->lacp = NULL; + bundle->bond = NULL; + + bundle->floodable = true; + + bundle->src_mirrors = 0; + bundle->dst_mirrors = 0; + bundle->mirror_out = 0; + } + + if (!bundle->name || strcmp(s->name, bundle->name)) { + free(bundle->name); + bundle->name = xstrdup(s->name); + } + + /* LACP. */ + if (s->lacp) { + if (!bundle->lacp) { + bundle->lacp = lacp_create(); + } + lacp_configure(bundle->lacp, s->lacp); + } else { + lacp_destroy(bundle->lacp); + bundle->lacp = NULL; + } + + /* Update set of ports. */ + ok = true; + for (i = 0; i < s->n_slaves; i++) { + if (!bundle_add_port(bundle, s->slaves[i], - s->lacp ? &s->lacp_slaves[i] : NULL)) { ++ s->lacp ? &s->lacp_slaves[i] : NULL, ++ s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { + ok = false; + } + } + if (!ok || list_size(&bundle->ports) != s->n_slaves) { + struct ofport_dpif *next_port; + + LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { + for (i = 0; i < s->n_slaves; i++) { + if (s->slaves[i] == odp_port_to_ofp_port(port->odp_port)) { + goto found; + } + } + + bundle_del_port(port); + found: ; + } + } + assert(list_size(&bundle->ports) <= s->n_slaves); + + if (list_is_empty(&bundle->ports)) { + bundle_destroy(bundle); + return EINVAL; + } + + /* Set VLAN tag. */ + if (s->vlan != bundle->vlan) { + bundle->vlan = s->vlan; + need_flush = true; + } + + /* Get trunked VLANs. */ + trunks = s->vlan == -1 ? NULL : s->trunks; + if (!vlan_bitmap_equal(trunks, bundle->trunks)) { + free(bundle->trunks); + bundle->trunks = vlan_bitmap_clone(trunks); + need_flush = true; + } + + /* Bonding. */ + if (!list_is_short(&bundle->ports)) { + bundle->ofproto->has_bonded_bundles = true; + if (bundle->bond) { + if (bond_reconfigure(bundle->bond, s->bond)) { + ofproto->need_revalidate = true; + } + } else { + bundle->bond = bond_create(s->bond); + } + + LIST_FOR_EACH (port, bundle_node, &bundle->ports) { - uint16_t stable_id = (bundle->lacp - ? lacp_slave_get_port_id(bundle->lacp, port) - : port->odp_port); - bond_slave_register(bundle->bond, port, stable_id, ++ bond_slave_register(bundle->bond, port, port->bond_stable_id, + port->up.netdev); + } + } else { + bond_destroy(bundle->bond); + bundle->bond = NULL; + } + + /* If we changed something that would affect MAC learning, un-learn + * everything on this port and force flow revalidation. */ + if (need_flush) { + bundle_flush_macs(bundle); + } + + return 0; +} + +static void +bundle_remove(struct ofport *port_) +{ + struct ofport_dpif *port = ofport_dpif_cast(port_); + struct ofbundle *bundle = port->bundle; + + if (bundle) { + bundle_del_port(port); + if (list_is_empty(&bundle->ports)) { + bundle_destroy(bundle); + } else if (list_is_short(&bundle->ports)) { + bond_destroy(bundle->bond); + bundle->bond = NULL; + } + } +} + +static void +send_pdu_cb(void *port_, const struct lacp_pdu *pdu) +{ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); + struct ofport_dpif *port = port_; + uint8_t ea[ETH_ADDR_LEN]; + int error; + + error = netdev_get_etheraddr(port->up.netdev, ea); + if (!error) { + struct lacp_pdu *packet_pdu; + struct ofpbuf packet; + + ofpbuf_init(&packet, 0); + packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, + sizeof *packet_pdu); + *packet_pdu = *pdu; + error = netdev_send(port->up.netdev, &packet); + if (error) { + VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " + "(%s)", port->bundle->name, + netdev_get_name(port->up.netdev), strerror(error)); + } + ofpbuf_uninit(&packet); + } else { + VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " + "%s (%s)", port->bundle->name, + netdev_get_name(port->up.netdev), strerror(error)); + } +} + +static void +bundle_send_learning_packets(struct ofbundle *bundle) +{ + struct ofproto_dpif *ofproto = bundle->ofproto; + int error, n_packets, n_errors; + struct mac_entry *e; + + error = n_packets = n_errors = 0; + LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { + if (e->port.p != bundle) { + int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan); + if (ret) { + error = ret; + n_errors++; + } + n_packets++; + } + } + + if (n_errors) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " + "packets, last error was: %s", + bundle->name, n_errors, n_packets, strerror(error)); + } else { + VLOG_DBG("bond %s: sent %d gratuitous learning packets", + bundle->name, n_packets); + } +} + +static void +bundle_run(struct ofbundle *bundle) +{ + if (bundle->lacp) { + lacp_run(bundle->lacp, send_pdu_cb); + } + if (bundle->bond) { + struct ofport_dpif *port; + + LIST_FOR_EACH (port, bundle_node, &bundle->ports) { + bool may_enable = lacp_slave_may_enable(bundle->lacp, port); + bond_slave_set_lacp_may_enable(bundle->bond, port, may_enable); + } + + bond_run(bundle->bond, &bundle->ofproto->revalidate_set, + lacp_negotiated(bundle->lacp)); + if (bond_should_send_learning_packets(bundle->bond)) { + bundle_send_learning_packets(bundle); + } + } +} + +static void +bundle_wait(struct ofbundle *bundle) +{ + if (bundle->lacp) { + lacp_wait(bundle->lacp); + } + if (bundle->bond) { + bond_wait(bundle->bond); + } +} + +/* Mirrors. */ + +static int +mirror_scan(struct ofproto_dpif *ofproto) +{ + int idx; + + for (idx = 0; idx < MAX_MIRRORS; idx++) { + if (!ofproto->mirrors[idx]) { + return idx; + } + } + return -1; +} + +static struct ofmirror * +mirror_lookup(struct ofproto_dpif *ofproto, void *aux) +{ + int i; + + for (i = 0; i < MAX_MIRRORS; i++) { + struct ofmirror *mirror = ofproto->mirrors[i]; + if (mirror && mirror->aux == aux) { + return mirror; + } + } + + return NULL; +} + +static int +mirror_set(struct ofproto *ofproto_, void *aux, + const struct ofproto_mirror_settings *s) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + mirror_mask_t mirror_bit; + struct ofbundle *bundle; + struct ofmirror *mirror; + struct ofbundle *out; + struct hmapx srcs; /* Contains "struct ofbundle *"s. */ + struct hmapx dsts; /* Contains "struct ofbundle *"s. */ + int out_vlan; + + mirror = mirror_lookup(ofproto, aux); + if (!s) { + mirror_destroy(mirror); + return 0; + } + if (!mirror) { + int idx; + + idx = mirror_scan(ofproto); + if (idx < 0) { + VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " + "cannot create %s", + ofproto->up.name, MAX_MIRRORS, s->name); + return EFBIG; + } + + mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); + mirror->ofproto = ofproto; + mirror->idx = idx; + mirror->out_vlan = -1; + mirror->name = NULL; + } + + if (!mirror->name || strcmp(s->name, mirror->name)) { + free(mirror->name); + mirror->name = xstrdup(s->name); + } + + /* Get the new configuration. */ + if (s->out_bundle) { + out = bundle_lookup(ofproto, s->out_bundle); + if (!out) { + mirror_destroy(mirror); + return EINVAL; + } + out_vlan = -1; + } else { + out = NULL; + out_vlan = s->out_vlan; + } + bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); + bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); + + /* If the configuration has not changed, do nothing. */ + if (hmapx_equals(&srcs, &mirror->srcs) + && hmapx_equals(&dsts, &mirror->dsts) + && vlan_bitmap_equal(mirror->vlans, s->src_vlans) + && mirror->out == out + && mirror->out_vlan == out_vlan) + { + hmapx_destroy(&srcs); + hmapx_destroy(&dsts); + return 0; + } + + hmapx_swap(&srcs, &mirror->srcs); + hmapx_destroy(&srcs); + + hmapx_swap(&dsts, &mirror->dsts); + hmapx_destroy(&dsts); + + free(mirror->vlans); + mirror->vlans = vlan_bitmap_clone(s->src_vlans); + + mirror->out = out; + mirror->out_vlan = out_vlan; + + /* Update bundles. */ + mirror_bit = MIRROR_MASK_C(1) << mirror->idx; + HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { + if (hmapx_contains(&mirror->srcs, bundle)) { + bundle->src_mirrors |= mirror_bit; + } else { + bundle->src_mirrors &= ~mirror_bit; + } + + if (hmapx_contains(&mirror->dsts, bundle)) { + bundle->dst_mirrors |= mirror_bit; + } else { + bundle->dst_mirrors &= ~mirror_bit; + } + + if (mirror->out == bundle) { + bundle->mirror_out |= mirror_bit; + } else { + bundle->mirror_out &= ~mirror_bit; + } + } + + ofproto->need_revalidate = true; + mac_learning_flush(ofproto->ml); + + return 0; +} + +static void +mirror_destroy(struct ofmirror *mirror) +{ + struct ofproto_dpif *ofproto; + mirror_mask_t mirror_bit; + struct ofbundle *bundle; + + if (!mirror) { + return; + } + + ofproto = mirror->ofproto; + ofproto->need_revalidate = true; + mac_learning_flush(ofproto->ml); + + mirror_bit = MIRROR_MASK_C(1) << mirror->idx; + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + bundle->src_mirrors &= ~mirror_bit; + bundle->dst_mirrors &= ~mirror_bit; + bundle->mirror_out &= ~mirror_bit; + } + + hmapx_destroy(&mirror->srcs); + hmapx_destroy(&mirror->dsts); + free(mirror->vlans); + + ofproto->mirrors[mirror->idx] = NULL; + free(mirror->name); + free(mirror); +} + +static int +set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { + ofproto->need_revalidate = true; + mac_learning_flush(ofproto->ml); + } + return 0; +} + +static bool +is_mirror_output_bundle(struct ofproto *ofproto_, void *aux) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct ofbundle *bundle = bundle_lookup(ofproto, aux); + return bundle && bundle->mirror_out != 0; +} + +/* Ports. */ + +static struct ofport_dpif * +get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) +{ + return ofport_dpif_cast(ofproto_get_port(&ofproto->up, ofp_port)); +} + +static struct ofport_dpif * +get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) +{ + return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); +} + +static void +ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, + struct dpif_port *dpif_port) +{ + ofproto_port->name = dpif_port->name; + ofproto_port->type = dpif_port->type; + ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); +} + +static void +port_run(struct ofport_dpif *ofport) +{ + if (ofport->cfm) { + cfm_run(ofport->cfm); + + if (cfm_should_send_ccm(ofport->cfm)) { + struct ofpbuf packet; + struct ccm *ccm; + + ofpbuf_init(&packet, 0); + ccm = eth_compose(&packet, eth_addr_ccm, ofport->up.opp.hw_addr, + ETH_TYPE_CFM, sizeof *ccm); + cfm_compose_ccm(ofport->cfm, ccm); + send_packet(ofproto_dpif_cast(ofport->up.ofproto), + ofport->odp_port, 0, &packet); + ofpbuf_uninit(&packet); + } + } +} + +static void +port_wait(struct ofport_dpif *ofport) +{ + if (ofport->cfm) { + cfm_wait(ofport->cfm); + } +} + +static int +port_query_by_name(const struct ofproto *ofproto_, const char *devname, + struct ofproto_port *ofproto_port) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct dpif_port dpif_port; + int error; + + error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); + if (!error) { + ofproto_port_from_dpif_port(ofproto_port, &dpif_port); + } + return error; +} + +static int +port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + uint16_t odp_port; + int error; + + error = dpif_port_add(ofproto->dpif, netdev, &odp_port); + if (!error) { + *ofp_portp = odp_port_to_ofp_port(odp_port); + } + return error; +} + +static int +port_del(struct ofproto *ofproto_, uint16_t ofp_port) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + int error; + + error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); + if (!error) { + struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); + if (ofport) { + /* The caller is going to close ofport->up.netdev. If this is a + * bonded port, then the bond is using that netdev, so remove it + * from the bond. The client will need to reconfigure everything + * after deleting ports, so then the slave will get re-added. */ + bundle_remove(&ofport->up); + } + } + return error; +} + +struct port_dump_state { + struct dpif_port_dump dump; + bool done; +}; + +static int +port_dump_start(const struct ofproto *ofproto_, void **statep) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct port_dump_state *state; + + *statep = state = xmalloc(sizeof *state); + dpif_port_dump_start(&state->dump, ofproto->dpif); + state->done = false; + return 0; +} + +static int +port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, + struct ofproto_port *port) +{ + struct port_dump_state *state = state_; + struct dpif_port dpif_port; + + if (dpif_port_dump_next(&state->dump, &dpif_port)) { + ofproto_port_from_dpif_port(port, &dpif_port); + return 0; + } else { + int error = dpif_port_dump_done(&state->dump); + state->done = true; + return error ? error : EOF; + } +} + +static int +port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) +{ + struct port_dump_state *state = state_; + + if (!state->done) { + dpif_port_dump_done(&state->dump); + } + free(state); + return 0; +} + +static int +port_poll(const struct ofproto *ofproto_, char **devnamep) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + return dpif_port_poll(ofproto->dpif, devnamep); +} + +static void +port_poll_wait(const struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + dpif_port_poll_wait(ofproto->dpif); +} + +static int +port_is_lacp_current(const struct ofport *ofport_) +{ + const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + return (ofport->bundle && ofport->bundle->lacp + ? lacp_slave_is_current(ofport->bundle->lacp, ofport) + : -1); +} + +/* Upcall handling. */ + +/* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an + * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to + * their individual configurations. + * + * If 'clone' is true, the caller retains ownership of 'upcall->packet'. + * Otherwise, ownership is transferred to this function. */ +static void +send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall, + const struct flow *flow, bool clone) +{ + struct ofputil_packet_in pin; + + pin.packet = upcall->packet; + pin.in_port = flow->in_port; + pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION; + pin.buffer_id = 0; /* not yet known */ + pin.send_len = upcall->userdata; + connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, + clone ? NULL : upcall->packet); +} + +static bool +process_special(struct ofproto_dpif *ofproto, const struct flow *flow, + const struct ofpbuf *packet) +{ + if (cfm_should_process_flow(flow)) { + struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); + if (ofport && ofport->cfm) { + cfm_process_heartbeat(ofport->cfm, packet); + } + return true; + } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { + struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); + if (port && port->bundle && port->bundle->lacp) { + const struct lacp_pdu *pdu = parse_lacp_packet(packet); + if (pdu) { + lacp_process_pdu(port->bundle->lacp, port, pdu); + } + return true; + } + } + return false; +} + +static void +handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) +{ + struct facet *facet; + struct flow flow; + + /* Obtain in_port and tun_id, at least. */ + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + + /* Set header pointers in 'flow'. */ + flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); + + /* Handle 802.1ag and LACP. */ + if (process_special(ofproto, &flow, upcall->packet)) { + ofpbuf_delete(upcall->packet); + ofproto->n_matches++; + return; + } + + /* Check with in-band control to see if this packet should be sent + * to the local port regardless of the flow table. */ + if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) { + send_packet(ofproto, OFPP_LOCAL, 0, upcall->packet); + } + + facet = facet_lookup_valid(ofproto, &flow); + if (!facet) { + struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow); + if (!rule) { + /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ + struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port); + if (port) { + if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { + COVERAGE_INC(ofproto_dpif_no_packet_in); + /* XXX install 'drop' flow entry */ + ofpbuf_delete(upcall->packet); + return; + } + } else { + VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, + flow.in_port); + } + + send_packet_in(ofproto, upcall, &flow, false); + return; + } + + facet = facet_create(rule, &flow, upcall->packet); + } else if (!facet->may_install) { + /* The facet is not installable, that is, we need to process every + * packet, so process the current packet's actions into 'facet'. */ + facet_make_actions(ofproto, facet, upcall->packet); + } + + if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { + /* + * Extra-special case for fail-open mode. + * + * We are in fail-open mode and the packet matched the fail-open rule, + * but we are connected to a controller too. We should send the packet + * up to the controller in the hope that it will try to set up a flow + * and thereby allow us to exit fail-open. + * + * See the top-level comment in fail-open.c for more information. + */ + send_packet_in(ofproto, upcall, &flow, true); + } + + facet_execute(ofproto, facet, upcall->packet); + facet_install(ofproto, facet, false); + ofproto->n_matches++; +} + +static void +handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) +{ + struct flow flow; + + switch (upcall->type) { + case DPIF_UC_ACTION: + COVERAGE_INC(ofproto_dpif_ctlr_action); + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + send_packet_in(ofproto, upcall, &flow, false); + break; + + case DPIF_UC_SAMPLE: + if (ofproto->sflow) { + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + ofproto_sflow_received(ofproto->sflow, upcall, &flow); + } + ofpbuf_delete(upcall->packet); + break; + + case DPIF_UC_MISS: + handle_miss_upcall(ofproto, upcall); + break; + + case DPIF_N_UC_TYPES: + default: + VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); + break; + } +} + +/* Flow expiration. */ + +static int facet_max_idle(const struct ofproto_dpif *); +static void update_stats(struct ofproto_dpif *); +static void rule_expire(struct rule_dpif *); +static void expire_facets(struct ofproto_dpif *, int dp_max_idle); + +/* This function is called periodically by run(). Its job is to collect + * updates for the flows that have been installed into the datapath, most + * importantly when they last were used, and then use that information to + * expire flows that have not been used recently. + * + * Returns the number of milliseconds after which it should be called again. */ +static int +expire(struct ofproto_dpif *ofproto) +{ + struct rule_dpif *rule, *next_rule; + struct cls_cursor cursor; + int dp_max_idle; + + /* Update stats for each flow in the datapath. */ + update_stats(ofproto); + + /* Expire facets that have been idle too long. */ + dp_max_idle = facet_max_idle(ofproto); + expire_facets(ofproto, dp_max_idle); + + /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ + cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { + rule_expire(rule); + } + + /* All outstanding data in existing flows has been accounted, so it's a + * good time to do bond rebalancing. */ + if (ofproto->has_bonded_bundles) { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + if (bundle->bond) { + bond_rebalance(bundle->bond, &ofproto->revalidate_set); + } + } + } + + return MIN(dp_max_idle, 1000); +} + +/* Update 'packet_count', 'byte_count', and 'used' members of installed facets. + * + * This function also pushes statistics updates to rules which each facet + * resubmits into. Generally these statistics will be accurate. However, if a + * facet changes the rule it resubmits into at some time in between + * update_stats() runs, it is possible that statistics accrued to the + * old rule will be incorrectly attributed to the new rule. This could be + * avoided by calling update_stats() whenever rules are created or + * deleted. However, the performance impact of making so many calls to the + * datapath do not justify the benefit of having perfectly accurate statistics. + */ +static void +update_stats(struct ofproto_dpif *p) +{ + const struct dpif_flow_stats *stats; + struct dpif_flow_dump dump; + const struct nlattr *key; + size_t key_len; + + dpif_flow_dump_start(&dump, p->dpif); + while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { + struct facet *facet; + struct flow flow; + + if (odp_flow_key_to_flow(key, key_len, &flow)) { + struct ds s; + + ds_init(&s); + odp_flow_key_format(key, key_len, &s); + VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", + ds_cstr(&s)); + ds_destroy(&s); + + continue; + } + facet = facet_find(p, &flow); + + if (facet && facet->installed) { + + if (stats->n_packets >= facet->dp_packet_count) { + uint64_t extra = stats->n_packets - facet->dp_packet_count; + facet->packet_count += extra; + } else { + VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); + } + + if (stats->n_bytes >= facet->dp_byte_count) { + facet->byte_count += stats->n_bytes - facet->dp_byte_count; + } else { + VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); + } + + facet->dp_packet_count = stats->n_packets; + facet->dp_byte_count = stats->n_bytes; + + facet_update_time(p, facet, stats->used); + facet_account(p, facet, stats->n_bytes); + facet_push_stats(facet); + } else { + /* There's a flow in the datapath that we know nothing about. + * Delete it. */ + COVERAGE_INC(facet_unexpected); + dpif_flow_del(p->dpif, key, key_len, NULL); + } + } + dpif_flow_dump_done(&dump); +} + +/* Calculates and returns the number of milliseconds of idle time after which + * facets should expire from the datapath and we should fold their statistics + * into their parent rules in userspace. */ +static int +facet_max_idle(const struct ofproto_dpif *ofproto) +{ + /* + * Idle time histogram. + * + * Most of the time a switch has a relatively small number of facets. When + * this is the case we might as well keep statistics for all of them in + * userspace and to cache them in the kernel datapath for performance as + * well. + * + * As the number of facets increases, the memory required to maintain + * statistics about them in userspace and in the kernel becomes + * significant. However, with a large number of facets it is likely that + * only a few of them are "heavy hitters" that consume a large amount of + * bandwidth. At this point, only heavy hitters are worth caching in the + * kernel and maintaining in userspaces; other facets we can discard. + * + * The technique used to compute the idle time is to build a histogram with + * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet + * that is installed in the kernel gets dropped in the appropriate bucket. + * After the histogram has been built, we compute the cutoff so that only + * the most-recently-used 1% of facets (but at least 1000 flows) are kept + * cached. At least the most-recently-used bucket of facets is kept, so + * actually an arbitrary number of facets can be kept in any given + * expiration run (though the next run will delete most of those unless + * they receive additional data). + * + * This requires a second pass through the facets, in addition to the pass + * made by update_stats(), because the former function never looks + * at uninstallable facets. + */ + enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; + enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; + int buckets[N_BUCKETS] = { 0 }; + struct facet *facet; + int total, bucket; + long long int now; + int i; + + total = hmap_count(&ofproto->facets); + if (total <= 1000) { + return N_BUCKETS * BUCKET_WIDTH; + } + + /* Build histogram. */ + now = time_msec(); + HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { + long long int idle = now - facet->used; + int bucket = (idle <= 0 ? 0 + : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 + : (unsigned int) idle / BUCKET_WIDTH); + buckets[bucket]++; + } + + /* Find the first bucket whose flows should be expired. */ + for (bucket = 0; bucket < N_BUCKETS; bucket++) { + if (buckets[bucket]) { + int subtotal = 0; + do { + subtotal += buckets[bucket++]; + } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); + break; + } + } + + if (VLOG_IS_DBG_ENABLED()) { + struct ds s; + + ds_init(&s); + ds_put_cstr(&s, "keep"); + for (i = 0; i < N_BUCKETS; i++) { + if (i == bucket) { + ds_put_cstr(&s, ", drop"); + } + if (buckets[i]) { + ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); + } + } + VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); + ds_destroy(&s); + } + + return bucket * BUCKET_WIDTH; +} + +static void +facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) +{ + if (ofproto->netflow && !facet_is_controller_flow(facet) && + netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { + struct ofexpired expired; + + if (facet->installed) { + struct dpif_flow_stats stats; + + facet_put__(ofproto, facet, facet->actions, facet->actions_len, + &stats); + facet_update_stats(ofproto, facet, &stats); + } + + expired.flow = facet->flow; + expired.packet_count = facet->packet_count; + expired.byte_count = facet->byte_count; + expired.used = facet->used; + netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); + } +} + +static void +expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle) +{ + long long int cutoff = time_msec() - dp_max_idle; + struct facet *facet, *next_facet; + + HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { + facet_active_timeout(ofproto, facet); + if (facet->used < cutoff) { + facet_remove(ofproto, facet); + } + } +} + +/* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, + * then delete it entirely. */ +static void +rule_expire(struct rule_dpif *rule) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct facet *facet, *next_facet; + long long int now; + uint8_t reason; + + /* Has 'rule' expired? */ + now = time_msec(); + if (rule->up.hard_timeout + && now > rule->up.created + rule->up.hard_timeout * 1000) { + reason = OFPRR_HARD_TIMEOUT; + } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) + && now > rule->used + rule->up.idle_timeout * 1000) { + reason = OFPRR_IDLE_TIMEOUT; + } else { + return; + } + + COVERAGE_INC(ofproto_dpif_expired); + + /* Update stats. (This is a no-op if the rule expired due to an idle + * timeout, because that only happens when the rule has no facets left.) */ + LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { + facet_remove(ofproto, facet); + } + + /* Get rid of the rule. */ + ofproto_rule_expire(&rule->up, reason); +} + +/* Facets. */ + +/* Creates and returns a new facet owned by 'rule', given a 'flow' and an + * example 'packet' within that flow. + * + * The caller must already have determined that no facet with an identical + * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in + * the ofproto's classifier table. */ +static struct facet * +facet_create(struct rule_dpif *rule, const struct flow *flow, + const struct ofpbuf *packet) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct facet *facet; + + facet = xzalloc(sizeof *facet); + facet->used = time_msec(); + hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); + list_push_back(&rule->facets, &facet->list_node); + facet->rule = rule; + facet->flow = *flow; + netflow_flow_init(&facet->nf_flow); + netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); + + facet_make_actions(ofproto, facet, packet); + + return facet; +} + +static void +facet_free(struct facet *facet) +{ + free(facet->actions); + free(facet); +} + +/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on + * 'packet', which arrived on 'in_port'. + * + * Takes ownership of 'packet'. */ +static bool +execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, + const struct nlattr *odp_actions, size_t actions_len, + struct ofpbuf *packet) +{ + if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) + && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) { + /* As an optimization, avoid a round-trip from userspace to kernel to + * userspace. This also avoids possibly filling up kernel packet + * buffers along the way. */ + struct dpif_upcall upcall; + + upcall.type = DPIF_UC_ACTION; + upcall.packet = packet; + upcall.key = NULL; + upcall.key_len = 0; + upcall.userdata = nl_attr_get_u64(odp_actions); + upcall.sample_pool = 0; + upcall.actions = NULL; + upcall.actions_len = 0; + + send_packet_in(ofproto, &upcall, flow, false); + + return true; + } else { + int error; + + error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet); + ofpbuf_delete(packet); + return !error; + } +} + +/* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s + * statistics appropriately. 'packet' must have at least sizeof(struct + * ofp_packet_in) bytes of headroom. + * + * For correct results, 'packet' must actually be in 'facet''s flow; that is, + * applying flow_extract() to 'packet' would yield the same flow as + * 'facet->flow'. + * + * 'facet' must have accurately composed ODP actions; that is, it must not be + * in need of revalidation. + * + * Takes ownership of 'packet'. */ +static void +facet_execute(struct ofproto_dpif *ofproto, struct facet *facet, + struct ofpbuf *packet) +{ + struct dpif_flow_stats stats; + + assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); + + flow_extract_stats(&facet->flow, packet, &stats); + stats.used = time_msec(); + if (execute_odp_actions(ofproto, &facet->flow, + facet->actions, facet->actions_len, packet)) { + facet_update_stats(ofproto, facet, &stats); + } +} + +/* Remove 'facet' from 'ofproto' and free up the associated memory: + * + * - If 'facet' was installed in the datapath, uninstalls it and updates its + * rule's statistics, via facet_uninstall(). + * + * - Removes 'facet' from its rule and from ofproto->facets. + */ +static void +facet_remove(struct ofproto_dpif *ofproto, struct facet *facet) +{ + facet_uninstall(ofproto, facet); + facet_flush_stats(ofproto, facet); + hmap_remove(&ofproto->facets, &facet->hmap_node); + list_remove(&facet->list_node); + facet_free(facet); +} + +/* Composes the ODP actions for 'facet' based on its rule's actions. */ +static void +facet_make_actions(struct ofproto_dpif *p, struct facet *facet, + const struct ofpbuf *packet) +{ + const struct rule_dpif *rule = facet->rule; + struct ofpbuf *odp_actions; + struct action_xlate_ctx ctx; + + action_xlate_ctx_init(&ctx, p, &facet->flow, packet); + odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); + facet->tags = ctx.tags; + facet->may_install = ctx.may_set_up_flow; + facet->nf_flow.output_iface = ctx.nf_output_iface; + + if (facet->actions_len != odp_actions->size + || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { + free(facet->actions); + facet->actions_len = odp_actions->size; + facet->actions = xmemdup(odp_actions->data, odp_actions->size); + } + + ofpbuf_delete(odp_actions); +} + +static int +facet_put__(struct ofproto_dpif *ofproto, struct facet *facet, + const struct nlattr *actions, size_t actions_len, + struct dpif_flow_stats *stats) +{ + struct odputil_keybuf keybuf; + enum dpif_flow_put_flags flags; + struct ofpbuf key; + + flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; + if (stats) { + flags |= DPIF_FP_ZERO_STATS; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; + } + + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, &facet->flow); + + return dpif_flow_put(ofproto->dpif, flags, key.data, key.size, + actions, actions_len, stats); +} + +/* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If + * 'zero_stats' is true, clears any existing statistics from the datapath for + * 'facet'. */ +static void +facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats) +{ + struct dpif_flow_stats stats; + + if (facet->may_install + && !facet_put__(p, facet, facet->actions, facet->actions_len, + zero_stats ? &stats : NULL)) { + facet->installed = true; + } +} + +static void +facet_account(struct ofproto_dpif *ofproto, + struct facet *facet, uint64_t extra_bytes) +{ + uint64_t total_bytes, n_bytes; + struct ofbundle *in_bundle; + const struct nlattr *a; + tag_type dummy = 0; + unsigned int left; + int vlan; + + total_bytes = facet->byte_count + extra_bytes; + if (total_bytes <= facet->accounted_bytes) { + return; + } + n_bytes = total_bytes - facet->accounted_bytes; + facet->accounted_bytes = total_bytes; + + /* Test that 'tags' is nonzero to ensure that only flows that include an + * OFPP_NORMAL action are used for learning and bond slave rebalancing. + * This works because OFPP_NORMAL always sets a nonzero tag value. + * + * Feed information from the active flows back into the learning table to + * ensure that table is always in sync with what is actually flowing + * through the datapath. */ + if (!facet->tags + || !is_admissible(ofproto, &facet->flow, false, &dummy, + &vlan, &in_bundle)) { + return; + } + + update_learning_table(ofproto, &facet->flow, vlan, in_bundle); + + if (!ofproto->has_bonded_bundles) { + return; + } + NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) { + if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) { + struct ofport_dpif *port; + + port = get_odp_port(ofproto, nl_attr_get_u32(a)); + if (port && port->bundle && port->bundle->bond) { + bond_account(port->bundle->bond, &facet->flow, vlan, n_bytes); + } + } + } +} + +/* If 'rule' is installed in the datapath, uninstalls it. */ +static void +facet_uninstall(struct ofproto_dpif *p, struct facet *facet) +{ + if (facet->installed) { + struct odputil_keybuf keybuf; + struct dpif_flow_stats stats; + struct ofpbuf key; + + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, &facet->flow); + + if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) { + facet_update_stats(p, facet, &stats); + } + facet->installed = false; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; + } else { + assert(facet->dp_packet_count == 0); + assert(facet->dp_byte_count == 0); + } +} + +/* Returns true if the only action for 'facet' is to send to the controller. + * (We don't report NetFlow expiration messages for such facets because they + * are just part of the control logic for the network, not real traffic). */ +static bool +facet_is_controller_flow(struct facet *facet) +{ + return (facet + && facet->rule->up.n_actions == 1 + && action_outputs_to_port(&facet->rule->up.actions[0], + htons(OFPP_CONTROLLER))); +} + +/* Folds all of 'facet''s statistics into its rule. Also updates the + * accounting ofhook and emits a NetFlow expiration if appropriate. All of + * 'facet''s statistics in the datapath should have been zeroed and folded into + * its packet and byte counts before this function is called. */ +static void +facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) +{ + assert(!facet->dp_byte_count); + assert(!facet->dp_packet_count); + + facet_push_stats(facet); + facet_account(ofproto, facet, 0); + + if (ofproto->netflow && !facet_is_controller_flow(facet)) { + struct ofexpired expired; + expired.flow = facet->flow; + expired.packet_count = facet->packet_count; + expired.byte_count = facet->byte_count; + expired.used = facet->used; + netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); + } + + facet->rule->packet_count += facet->packet_count; + facet->rule->byte_count += facet->byte_count; + + /* Reset counters to prevent double counting if 'facet' ever gets + * reinstalled. */ + facet->packet_count = 0; + facet->byte_count = 0; + facet->rs_packet_count = 0; + facet->rs_byte_count = 0; + facet->accounted_bytes = 0; + + netflow_flow_clear(&facet->nf_flow); +} + +/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. + * Returns it if found, otherwise a null pointer. + * + * The returned facet might need revalidation; use facet_lookup_valid() + * instead if that is important. */ +static struct facet * +facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) +{ + struct facet *facet; + + HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), + &ofproto->facets) { + if (flow_equal(flow, &facet->flow)) { + return facet; + } + } + + return NULL; +} + +/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. + * Returns it if found, otherwise a null pointer. + * + * The returned facet is guaranteed to be valid. */ +static struct facet * +facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) +{ + struct facet *facet = facet_find(ofproto, flow); + + /* The facet we found might not be valid, since we could be in need of + * revalidation. If it is not valid, don't return it. */ + if (facet + && ofproto->need_revalidate + && !facet_revalidate(ofproto, facet)) { + COVERAGE_INC(facet_invalidated); + return NULL; + } + + return facet; +} + +/* Re-searches 'ofproto''s classifier for a rule matching 'facet': + * + * - If the rule found is different from 'facet''s current rule, moves + * 'facet' to the new rule and recompiles its actions. + * + * - If the rule found is the same as 'facet''s current rule, leaves 'facet' + * where it is and recompiles its actions anyway. + * + * - If there is none, destroys 'facet'. + * + * Returns true if 'facet' still exists, false if it has been destroyed. */ +static bool +facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) +{ + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + struct rule_dpif *new_rule; + bool actions_changed; + + COVERAGE_INC(facet_revalidate); + + /* Determine the new rule. */ + new_rule = rule_dpif_lookup(ofproto, &facet->flow); + if (!new_rule) { + /* No new rule, so delete the facet. */ + facet_remove(ofproto, facet); + return false; + } + + /* Calculate new ODP actions. + * + * We do not modify any 'facet' state yet, because we might need to, e.g., + * emit a NetFlow expiration and, if so, we need to have the old state + * around to properly compose it. */ + action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); + odp_actions = xlate_actions(&ctx, + new_rule->up.actions, new_rule->up.n_actions); + actions_changed = (facet->actions_len != odp_actions->size + || memcmp(facet->actions, odp_actions->data, + facet->actions_len)); + + /* If the ODP actions changed or the installability changed, then we need + * to talk to the datapath. */ + if (actions_changed || ctx.may_set_up_flow != facet->installed) { + if (ctx.may_set_up_flow) { + struct dpif_flow_stats stats; + + facet_put__(ofproto, facet, + odp_actions->data, odp_actions->size, &stats); + facet_update_stats(ofproto, facet, &stats); + } else { + facet_uninstall(ofproto, facet); + } + + /* The datapath flow is gone or has zeroed stats, so push stats out of + * 'facet' into 'rule'. */ + facet_flush_stats(ofproto, facet); + } + + /* Update 'facet' now that we've taken care of all the old state. */ + facet->tags = ctx.tags; + facet->nf_flow.output_iface = ctx.nf_output_iface; + facet->may_install = ctx.may_set_up_flow; + if (actions_changed) { + free(facet->actions); + facet->actions_len = odp_actions->size; + facet->actions = xmemdup(odp_actions->data, odp_actions->size); + } + if (facet->rule != new_rule) { + COVERAGE_INC(facet_changed_rule); + list_remove(&facet->list_node); + list_push_back(&new_rule->facets, &facet->list_node); + facet->rule = new_rule; + facet->used = new_rule->up.created; + facet->rs_used = facet->used; + } + + ofpbuf_delete(odp_actions); + + return true; +} + +/* Updates 'facet''s used time. Caller is responsible for calling + * facet_push_stats() to update the flows which 'facet' resubmits into. */ +static void +facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet, + long long int used) +{ + if (used > facet->used) { + facet->used = used; + if (used > facet->rule->used) { + facet->rule->used = used; + } + netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); + } +} + +/* Folds the statistics from 'stats' into the counters in 'facet'. + * + * Because of the meaning of a facet's counters, it only makes sense to do this + * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a + * packet that was sent by hand or if it represents statistics that have been + * cleared out of the datapath. */ +static void +facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet, + const struct dpif_flow_stats *stats) +{ + if (stats->n_packets || stats->used > facet->used) { + facet_update_time(ofproto, facet, stats->used); + facet->packet_count += stats->n_packets; + facet->byte_count += stats->n_bytes; + facet_push_stats(facet); + netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); + } +} + +static void +facet_push_stats(struct facet *facet) +{ + uint64_t rs_packets, rs_bytes; + + assert(facet->packet_count >= facet->rs_packet_count); + assert(facet->byte_count >= facet->rs_byte_count); + assert(facet->used >= facet->rs_used); + + rs_packets = facet->packet_count - facet->rs_packet_count; + rs_bytes = facet->byte_count - facet->rs_byte_count; + + if (rs_packets || rs_bytes || facet->used > facet->rs_used) { + facet->rs_packet_count = facet->packet_count; + facet->rs_byte_count = facet->byte_count; + facet->rs_used = facet->used; + + flow_push_stats(facet->rule, &facet->flow, + rs_packets, rs_bytes, facet->used); + } +} + +struct ofproto_push { + struct action_xlate_ctx ctx; + uint64_t packets; + uint64_t bytes; + long long int used; +}; + +static void +push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) +{ + struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); + + if (rule) { + rule->packet_count += push->packets; + rule->byte_count += push->bytes; + rule->used = MAX(push->used, rule->used); + } +} + +/* Pushes flow statistics to the rules which 'flow' resubmits into given + * 'rule''s actions. */ +static void +flow_push_stats(const struct rule_dpif *rule, + struct flow *flow, uint64_t packets, uint64_t bytes, + long long int used) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct ofproto_push push; + + push.packets = packets; + push.bytes = bytes; + push.used = used; + + action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); + push.ctx.resubmit_hook = push_resubmit; + ofpbuf_delete(xlate_actions(&push.ctx, + rule->up.actions, rule->up.n_actions)); +} + +/* Rules. */ + +static struct rule_dpif * +rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) +{ + return rule_dpif_cast(rule_from_cls_rule( + classifier_lookup(&ofproto->up.tables[0], + flow))); +} + +static struct rule * +rule_alloc(void) +{ + struct rule_dpif *rule = xmalloc(sizeof *rule); + return &rule->up; +} + +static void +rule_dealloc(struct rule *rule_) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + free(rule); +} + +static int +rule_construct(struct rule *rule_) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct rule_dpif *old_rule; + int error; + + error = validate_actions(rule->up.actions, rule->up.n_actions, + &rule->up.cr.flow, ofproto->max_ports); + if (error) { + return error; + } + + old_rule = rule_dpif_cast(rule_from_cls_rule(classifier_find_rule_exactly( + &ofproto->up.tables[0], + &rule->up.cr))); + if (old_rule) { + ofproto_rule_destroy(&old_rule->up); + } + + rule->used = rule->up.created; + rule->packet_count = 0; + rule->byte_count = 0; + list_init(&rule->facets); + classifier_insert(&ofproto->up.tables[0], &rule->up.cr); + + ofproto->need_revalidate = true; + + return 0; +} + +static void +rule_destruct(struct rule *rule_) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct facet *facet, *next_facet; + + classifier_remove(&ofproto->up.tables[0], &rule->up.cr); + LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { + facet_revalidate(ofproto, facet); + } + ofproto->need_revalidate = true; +} + +static void +rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + struct facet *facet; + + /* Start from historical data for 'rule' itself that are no longer tracked + * in facets. This counts, for example, facets that have expired. */ + *packets = rule->packet_count; + *bytes = rule->byte_count; + + /* Add any statistics that are tracked by facets. This includes + * statistical data recently updated by ofproto_update_stats() as well as + * stats for packets that were executed "by hand" via dpif_execute(). */ + LIST_FOR_EACH (facet, list_node, &rule->facets) { + *packets += facet->packet_count; + *bytes += facet->byte_count; + } +} + +static int +rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + struct facet *facet; + size_t size; + + /* First look for a related facet. If we find one, account it to that. */ + facet = facet_lookup_valid(ofproto, flow); + if (facet && facet->rule == rule) { + facet_execute(ofproto, facet, packet); + return 0; + } + + /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then + * create a new facet for it and use that. */ + if (rule_dpif_lookup(ofproto, flow) == rule) { + facet = facet_create(rule, flow, packet); + facet_execute(ofproto, facet, packet); + facet_install(ofproto, facet, true); + return 0; + } + + /* We can't account anything to a facet. If we were to try, then that + * facet would have a non-matching rule, busting our invariants. */ + action_xlate_ctx_init(&ctx, ofproto, flow, packet); + odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); + size = packet->size; + if (execute_odp_actions(ofproto, flow, odp_actions->data, + odp_actions->size, packet)) { + rule->used = time_msec(); + rule->packet_count++; + rule->byte_count += size; + flow_push_stats(rule, flow, 1, size, rule->used); + } + ofpbuf_delete(odp_actions); + + return 0; +} + +static int +rule_modify_actions(struct rule *rule_, + const union ofp_action *actions, size_t n_actions) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + int error; + + error = validate_actions(actions, n_actions, &rule->up.cr.flow, + ofproto->max_ports); + if (!error) { + ofproto->need_revalidate = true; + } + return error; +} + +/* Sends 'packet' out of port 'odp_port' within 'ofproto'. If 'vlan_tci' is + * zero the packet will not have any 802.1Q hader; if it is nonzero, then the + * packet will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'. + * + * Returns 0 if successful, otherwise a positive errno value. */ +static int +send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, uint16_t vlan_tci, + const struct ofpbuf *packet) +{ + struct ofpbuf odp_actions; + int error; + + ofpbuf_init(&odp_actions, 32); + if (vlan_tci != 0) { + nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, + ntohs(vlan_tci & ~VLAN_CFI)); + } + nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); + error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size, + packet); + ofpbuf_uninit(&odp_actions); + + if (error) { + VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", + ofproto->up.name, odp_port, strerror(error)); + } + return error; +} + +/* OpenFlow to ODP action translation. */ + +static void do_xlate_actions(const union ofp_action *in, size_t n_in, + struct action_xlate_ctx *ctx); +static bool xlate_normal(struct action_xlate_ctx *); + +static void +add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) +{ + const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); + uint16_t odp_port = ofp_port_to_odp_port(ofp_port); + + if (ofport) { + if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) { + /* Forwarding disabled on port. */ + return; + } + } else { + /* + * We don't have an ofport record for this port, but it doesn't hurt to + * allow forwarding to it anyhow. Maybe such a port will appear later + * and we're pre-populating the flow table. + */ + } + + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); + ctx->nf_output_iface = ofp_port; +} + +static void +xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) +{ + if (ctx->recurse < MAX_RESUBMIT_RECURSION) { + struct rule_dpif *rule; + uint16_t old_in_port; + + /* Look up a flow with 'in_port' as the input port. Then restore the + * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will + * have surprising behavior). */ + old_in_port = ctx->flow.in_port; + ctx->flow.in_port = in_port; + rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow); + ctx->flow.in_port = old_in_port; + + if (ctx->resubmit_hook) { + ctx->resubmit_hook(ctx, rule); + } + + if (rule) { + ctx->recurse++; + do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); + ctx->recurse--; + } + } else { + static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); + + VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", + MAX_RESUBMIT_RECURSION); + } +} + +static void +flood_packets(struct ofproto_dpif *ofproto, + uint16_t ofp_in_port, ovs_be32 mask, + uint16_t *nf_output_iface, struct ofpbuf *odp_actions) +{ + struct ofport_dpif *ofport; + + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + uint16_t ofp_port = ofport->up.ofp_port; + if (ofp_port != ofp_in_port && !(ofport->up.opp.config & mask)) { + nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, + ofport->odp_port); + } + } + *nf_output_iface = NF_OUT_FLOOD; +} + +static void +xlate_output_action__(struct action_xlate_ctx *ctx, + uint16_t port, uint16_t max_len) +{ + uint16_t prev_nf_output_iface = ctx->nf_output_iface; + + ctx->nf_output_iface = NF_OUT_DROP; + + switch (port) { + case OFPP_IN_PORT: + add_output_action(ctx, ctx->flow.in_port); + break; + case OFPP_TABLE: + xlate_table_action(ctx, ctx->flow.in_port); + break; + case OFPP_NORMAL: + xlate_normal(ctx); + break; + case OFPP_FLOOD: + flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD), + &ctx->nf_output_iface, ctx->odp_actions); + break; + case OFPP_ALL: + flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0), + &ctx->nf_output_iface, ctx->odp_actions); + break; + case OFPP_CONTROLLER: + nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len); + break; + case OFPP_LOCAL: + add_output_action(ctx, OFPP_LOCAL); + break; + default: + if (port != ctx->flow.in_port) { + add_output_action(ctx, port); + } + break; + } + + if (prev_nf_output_iface == NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_FLOOD; + } else if (ctx->nf_output_iface == NF_OUT_DROP) { + ctx->nf_output_iface = prev_nf_output_iface; + } else if (prev_nf_output_iface != NF_OUT_DROP && + ctx->nf_output_iface != NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_MULTI; + } +} + +static void +xlate_output_action(struct action_xlate_ctx *ctx, + const struct ofp_action_output *oao) +{ + xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); +} + +/* If the final ODP action in 'ctx' is "pop priority", drop it, as an + * optimization, because we're going to add another action that sets the + * priority immediately after, or because there are no actions following the + * pop. */ +static void +remove_pop_action(struct action_xlate_ctx *ctx) +{ + if (ctx->odp_actions->size == ctx->last_pop_priority) { + ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN); + ctx->last_pop_priority = -1; + } +} + +static void +add_pop_action(struct action_xlate_ctx *ctx) +{ + if (ctx->odp_actions->size != ctx->last_pop_priority) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY); + ctx->last_pop_priority = ctx->odp_actions->size; + } +} + +static void +xlate_enqueue_action(struct action_xlate_ctx *ctx, + const struct ofp_action_enqueue *oae) +{ + uint16_t ofp_port, odp_port; + uint32_t priority; + int error; + + error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), + &priority); + if (error) { + /* Fall back to ordinary output action. */ + xlate_output_action__(ctx, ntohs(oae->port), 0); + return; + } + + /* Figure out ODP output port. */ + ofp_port = ntohs(oae->port); + if (ofp_port == OFPP_IN_PORT) { + ofp_port = ctx->flow.in_port; + } + odp_port = ofp_port_to_odp_port(ofp_port); + + /* Add ODP actions. */ + remove_pop_action(ctx); + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); + add_output_action(ctx, odp_port); + add_pop_action(ctx); + + /* Update NetFlow output port. */ + if (ctx->nf_output_iface == NF_OUT_DROP) { + ctx->nf_output_iface = odp_port; + } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_MULTI; + } +} + +static void +xlate_set_queue_action(struct action_xlate_ctx *ctx, + const struct nx_action_set_queue *nasq) +{ + uint32_t priority; + int error; + + error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), + &priority); + if (error) { + /* Couldn't translate queue to a priority, so ignore. A warning + * has already been logged. */ + return; + } + + remove_pop_action(ctx); + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); +} + +static void +xlate_set_dl_tci(struct action_xlate_ctx *ctx) +{ + ovs_be16 tci = ctx->flow.vlan_tci; + if (!(tci & htons(VLAN_CFI))) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); + } else { + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, + tci & ~htons(VLAN_CFI)); + } +} + +struct xlate_reg_state { + ovs_be16 vlan_tci; + ovs_be64 tun_id; +}; + +static void +save_reg_state(const struct action_xlate_ctx *ctx, + struct xlate_reg_state *state) +{ + state->vlan_tci = ctx->flow.vlan_tci; + state->tun_id = ctx->flow.tun_id; +} + +static void +update_reg_state(struct action_xlate_ctx *ctx, + const struct xlate_reg_state *state) +{ + if (ctx->flow.vlan_tci != state->vlan_tci) { + xlate_set_dl_tci(ctx); + } + if (ctx->flow.tun_id != state->tun_id) { + nl_msg_put_be64(ctx->odp_actions, + ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id); + } +} + +static void +xlate_autopath(struct action_xlate_ctx *ctx, + const struct nx_action_autopath *naa) +{ + uint16_t ofp_port = ntohl(naa->id); + struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); + + if (!port || !port->bundle) { + ofp_port = OFPP_NONE; + } else if (port->bundle->bond) { + /* Autopath does not support VLAN hashing. */ + struct ofport_dpif *slave = bond_choose_output_slave( + port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags); + if (slave) { + ofp_port = slave->up.ofp_port; + } + } + autopath_execute(naa, &ctx->flow, ofp_port); +} + +static void +xlate_nicira_action(struct action_xlate_ctx *ctx, + const struct nx_action_header *nah) +{ + const struct nx_action_resubmit *nar; + const struct nx_action_set_tunnel *nast; + const struct nx_action_set_queue *nasq; + const struct nx_action_multipath *nam; + const struct nx_action_autopath *naa; + enum nx_action_subtype subtype = ntohs(nah->subtype); + struct xlate_reg_state state; + ovs_be64 tun_id; + + assert(nah->vendor == htonl(NX_VENDOR_ID)); + switch (subtype) { + case NXAST_RESUBMIT: + nar = (const struct nx_action_resubmit *) nah; + xlate_table_action(ctx, ntohs(nar->in_port)); + break; + + case NXAST_SET_TUNNEL: + nast = (const struct nx_action_set_tunnel *) nah; + tun_id = htonll(ntohl(nast->tun_id)); + nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); + ctx->flow.tun_id = tun_id; + break; + + case NXAST_DROP_SPOOFED_ARP: + if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { + nl_msg_put_flag(ctx->odp_actions, + ODP_ACTION_ATTR_DROP_SPOOFED_ARP); + } + break; + + case NXAST_SET_QUEUE: + nasq = (const struct nx_action_set_queue *) nah; + xlate_set_queue_action(ctx, nasq); + break; + + case NXAST_POP_QUEUE: + add_pop_action(ctx); + break; + + case NXAST_REG_MOVE: + save_reg_state(ctx, &state); + nxm_execute_reg_move((const struct nx_action_reg_move *) nah, + &ctx->flow); + update_reg_state(ctx, &state); + break; + + case NXAST_REG_LOAD: + save_reg_state(ctx, &state); + nxm_execute_reg_load((const struct nx_action_reg_load *) nah, + &ctx->flow); + update_reg_state(ctx, &state); + break; + + case NXAST_NOTE: + /* Nothing to do. */ + break; + + case NXAST_SET_TUNNEL64: + tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id; + nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); + ctx->flow.tun_id = tun_id; + break; + + case NXAST_MULTIPATH: + nam = (const struct nx_action_multipath *) nah; + multipath_execute(nam, &ctx->flow); + break; + + case NXAST_AUTOPATH: + naa = (const struct nx_action_autopath *) nah; + xlate_autopath(ctx, naa); + break; + + /* If you add a new action here that modifies flow data, don't forget to + * update the flow key in ctx->flow at the same time. */ + + case NXAST_SNAT__OBSOLETE: + default: + VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype); + break; + } +} + +static void +do_xlate_actions(const union ofp_action *in, size_t n_in, + struct action_xlate_ctx *ctx) +{ + const struct ofport_dpif *port; + struct actions_iterator iter; + const union ofp_action *ia; + + port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); + if (port + && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && + port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) + ? htonl(OFPPC_NO_RECV_STP) + : htonl(OFPPC_NO_RECV))) { + /* Drop this flow. */ + return; + } + + for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { + enum ofp_action_type type = ntohs(ia->type); + const struct ofp_action_dl_addr *oada; + + switch (type) { + case OFPAT_OUTPUT: + xlate_output_action(ctx, &ia->output); + break; + + case OFPAT_SET_VLAN_VID: + ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); + ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); + xlate_set_dl_tci(ctx); + break; + + case OFPAT_SET_VLAN_PCP: + ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); + ctx->flow.vlan_tci |= htons( + (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); + xlate_set_dl_tci(ctx); + break; + + case OFPAT_STRIP_VLAN: + ctx->flow.vlan_tci = htons(0); + xlate_set_dl_tci(ctx); + break; + + case OFPAT_SET_DL_SRC: + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC, + oada->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); + break; + + case OFPAT_SET_DL_DST: + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST, + oada->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); + break; + + case OFPAT_SET_NW_SRC: + nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, + ia->nw_addr.nw_addr); + ctx->flow.nw_src = ia->nw_addr.nw_addr; + break; + + case OFPAT_SET_NW_DST: + nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST, + ia->nw_addr.nw_addr); + ctx->flow.nw_dst = ia->nw_addr.nw_addr; + break; + + case OFPAT_SET_NW_TOS: + nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, + ia->nw_tos.nw_tos); + ctx->flow.nw_tos = ia->nw_tos.nw_tos; + break; + + case OFPAT_SET_TP_SRC: + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, + ia->tp_port.tp_port); + ctx->flow.tp_src = ia->tp_port.tp_port; + break; + + case OFPAT_SET_TP_DST: + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST, + ia->tp_port.tp_port); + ctx->flow.tp_dst = ia->tp_port.tp_port; + break; + + case OFPAT_VENDOR: + xlate_nicira_action(ctx, (const struct nx_action_header *) ia); + break; + + case OFPAT_ENQUEUE: + xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); + break; + + default: + VLOG_DBG_RL(&rl, "unknown action type %d", (int) type); + break; + } + } +} + +static void +action_xlate_ctx_init(struct action_xlate_ctx *ctx, + struct ofproto_dpif *ofproto, const struct flow *flow, + const struct ofpbuf *packet) +{ + ctx->ofproto = ofproto; + ctx->flow = *flow; + ctx->packet = packet; + ctx->resubmit_hook = NULL; +} + +static struct ofpbuf * +xlate_actions(struct action_xlate_ctx *ctx, + const union ofp_action *in, size_t n_in) +{ + COVERAGE_INC(ofproto_dpif_xlate); + + ctx->odp_actions = ofpbuf_new(512); + ctx->tags = 0; + ctx->may_set_up_flow = true; + ctx->nf_output_iface = NF_OUT_DROP; + ctx->recurse = 0; + ctx->last_pop_priority = -1; + + if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { + ctx->may_set_up_flow = false; + } else { + do_xlate_actions(in, n_in, ctx); + } + + remove_pop_action(ctx); + + /* Check with in-band control to see if we're allowed to set up this + * flow. */ + if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, + ctx->odp_actions->data, + ctx->odp_actions->size)) { + ctx->may_set_up_flow = false; + } + + return ctx->odp_actions; +} + +/* OFPP_NORMAL implementation. */ + +struct dst { + struct ofport_dpif *port; + uint16_t vlan; +}; + +struct dst_set { + struct dst builtin[32]; + struct dst *dsts; + size_t n, allocated; +}; + +static void dst_set_init(struct dst_set *); +static void dst_set_add(struct dst_set *, const struct dst *); +static void dst_set_free(struct dst_set *); + +static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); + +static bool +set_dst(struct action_xlate_ctx *ctx, struct dst *dst, + const struct ofbundle *in_bundle, const struct ofbundle *out_bundle) +{ + dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE + : in_bundle->vlan >= 0 ? in_bundle->vlan + : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE + : vlan_tci_to_vid(ctx->flow.vlan_tci)); + + dst->port = (!out_bundle->bond + ? ofbundle_get_a_port(out_bundle) + : bond_choose_output_slave(out_bundle->bond, &ctx->flow, + dst->vlan, &ctx->tags)); + + return dst->port != NULL; +} + +static int +mirror_mask_ffs(mirror_mask_t mask) +{ + BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); + return ffs(mask); +} + +static void +dst_set_init(struct dst_set *set) +{ + set->dsts = set->builtin; + set->n = 0; + set->allocated = ARRAY_SIZE(set->builtin); +} + +static void +dst_set_add(struct dst_set *set, const struct dst *dst) +{ + if (set->n >= set->allocated) { + size_t new_allocated; + struct dst *new_dsts; + + new_allocated = set->allocated * 2; + new_dsts = xmalloc(new_allocated * sizeof *new_dsts); + memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); + + dst_set_free(set); + + set->dsts = new_dsts; + set->allocated = new_allocated; + } + set->dsts[set->n++] = *dst; +} + +static void +dst_set_free(struct dst_set *set) +{ + if (set->dsts != set->builtin) { + free(set->dsts); + } +} + +static bool +dst_is_duplicate(const struct dst_set *set, const struct dst *test) +{ + size_t i; + for (i = 0; i < set->n; i++) { + if (set->dsts[i].vlan == test->vlan + && set->dsts[i].port == test->port) { + return true; + } + } + return false; +} + +static bool +ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) +{ + return bundle->vlan < 0 && vlan_bitmap_contains(bundle->trunks, vlan); +} + +static bool +ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) +{ + return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); +} + +/* Returns an arbitrary interface within 'bundle'. */ +static struct ofport_dpif * +ofbundle_get_a_port(const struct ofbundle *bundle) +{ + return CONTAINER_OF(list_front(&bundle->ports), + struct ofport_dpif, bundle_node); +} + +static void +compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan, + const struct ofbundle *in_bundle, + const struct ofbundle *out_bundle, struct dst_set *set) +{ + struct dst dst; + + if (out_bundle == OFBUNDLE_FLOOD) { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { + if (bundle != in_bundle + && ofbundle_includes_vlan(bundle, vlan) + && bundle->floodable + && !bundle->mirror_out + && set_dst(ctx, &dst, in_bundle, bundle)) { + dst_set_add(set, &dst); + } + } + ctx->nf_output_iface = NF_OUT_FLOOD; + } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) { + dst_set_add(set, &dst); + ctx->nf_output_iface = dst.port->odp_port; + } +} + +static bool +vlan_is_mirrored(const struct ofmirror *m, int vlan) +{ + return vlan_bitmap_contains(m->vlans, vlan); +} + +static void +compose_mirror_dsts(struct action_xlate_ctx *ctx, + uint16_t vlan, const struct ofbundle *in_bundle, + struct dst_set *set) +{ + struct ofproto_dpif *ofproto = ctx->ofproto; + mirror_mask_t mirrors; + int flow_vlan; + size_t i; + + mirrors = in_bundle->src_mirrors; + for (i = 0; i < set->n; i++) { + mirrors |= set->dsts[i].port->bundle->dst_mirrors; + } + + if (!mirrors) { + return; + } + + flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); + if (flow_vlan == 0) { + flow_vlan = OFP_VLAN_NONE; + } + + while (mirrors) { + struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; + if (vlan_is_mirrored(m, vlan)) { + struct dst dst; + + if (m->out) { + if (set_dst(ctx, &dst, in_bundle, m->out) + && !dst_is_duplicate(set, &dst)) { + dst_set_add(set, &dst); + } + } else { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + if (ofbundle_includes_vlan(bundle, m->out_vlan) + && set_dst(ctx, &dst, in_bundle, bundle)) + { + if (bundle->vlan < 0) { + dst.vlan = m->out_vlan; + } + if (dst_is_duplicate(set, &dst)) { + continue; + } + + /* Use the vlan tag on the original flow instead of + * the one passed in the vlan parameter. This ensures + * that we compare the vlan from before any implicit + * tagging tags place. This is necessary because + * dst->vlan is the final vlan, after removing implicit + * tags. */ + if (bundle == in_bundle && dst.vlan == flow_vlan) { + /* Don't send out input port on same VLAN. */ + continue; + } + dst_set_add(set, &dst); + } + } + } + } + mirrors &= mirrors - 1; + } +} + +static void +compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, + const struct ofbundle *in_bundle, + const struct ofbundle *out_bundle) +{ + uint16_t initial_vlan, cur_vlan; + const struct dst *dst; + struct dst_set set; + + dst_set_init(&set); + compose_dsts(ctx, vlan, in_bundle, out_bundle, &set); + compose_mirror_dsts(ctx, vlan, in_bundle, &set); + + /* Output all the packets we can without having to change the VLAN. */ + initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); + if (initial_vlan == 0) { + initial_vlan = OFP_VLAN_NONE; + } + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan != initial_vlan) { + continue; + } + nl_msg_put_u32(ctx->odp_actions, + ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); + } + + /* Then output the rest. */ + cur_vlan = initial_vlan; + for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { + if (dst->vlan == initial_vlan) { + continue; + } + if (dst->vlan != cur_vlan) { + if (dst->vlan == OFP_VLAN_NONE) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); + } else { + ovs_be16 tci; + tci = htons(dst->vlan & VLAN_VID_MASK); + tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); + nl_msg_put_be16(ctx->odp_actions, + ODP_ACTION_ATTR_SET_DL_TCI, tci); + } + cur_vlan = dst->vlan; + } + nl_msg_put_u32(ctx->odp_actions, + ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); + } + + dst_set_free(&set); +} + +/* Returns the effective vlan of a packet, taking into account both the + * 802.1Q header and implicitly tagged ports. A value of 0 indicates that + * the packet is untagged and -1 indicates it has an invalid header and + * should be dropped. */ +static int +flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow, + struct ofbundle *in_bundle, bool have_packet) +{ + int vlan = vlan_tci_to_vid(flow->vlan_tci); + if (in_bundle->vlan >= 0) { + if (vlan) { + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " + "packet received on port %s configured with " + "implicit VLAN %"PRIu16, + ofproto->up.name, vlan, + in_bundle->name, in_bundle->vlan); + } + return -1; + } + vlan = in_bundle->vlan; + } else { + if (!ofbundle_includes_vlan(in_bundle, vlan)) { + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " + "packet received on port %s not configured for " + "trunking VLAN %d", + ofproto->up.name, vlan, in_bundle->name, vlan); + } + return -1; + } + } + + return vlan; +} + +/* A VM broadcasts a gratuitous ARP to indicate that it has resumed after + * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to + * indicate this; newer upstream kernels use gratuitous ARP requests. */ +static bool +is_gratuitous_arp(const struct flow *flow) +{ + return (flow->dl_type == htons(ETH_TYPE_ARP) + && eth_addr_is_broadcast(flow->dl_dst) + && (flow->nw_proto == ARP_OP_REPLY + || (flow->nw_proto == ARP_OP_REQUEST + && flow->nw_src == flow->nw_dst))); +} + +static void +update_learning_table(struct ofproto_dpif *ofproto, + const struct flow *flow, int vlan, + struct ofbundle *in_bundle) +{ + struct mac_entry *mac; + + if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { + return; + } + + mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); + if (is_gratuitous_arp(flow)) { + /* We don't want to learn from gratuitous ARP packets that are + * reflected back over bond slaves so we lock the learning table. */ + if (!in_bundle->bond) { + mac_entry_set_grat_arp_lock(mac); + } else if (mac_entry_is_grat_arp_locked(mac)) { + return; + } + } + + if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { + /* The log messages here could actually be useful in debugging, + * so keep the rate limit relatively high. */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); + VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " + "on port %s in VLAN %d", + ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), + in_bundle->name, vlan); + + mac->port.p = in_bundle; + tag_set_add(&ofproto->revalidate_set, + mac_learning_changed(ofproto->ml, mac)); + } +} + +/* Determines whether packets in 'flow' within 'br' should be forwarded or + * dropped. Returns true if they may be forwarded, false if they should be + * dropped. + * + * If 'have_packet' is true, it indicates that the caller is processing a + * received packet. If 'have_packet' is false, then the caller is just + * revalidating an existing flow because configuration has changed. Either + * way, 'have_packet' only affects logging (there is no point in logging errors + * during revalidation). + * + * Sets '*in_portp' to the input port. This will be a null pointer if + * flow->in_port does not designate a known input port (in which case + * is_admissible() returns false). + * + * When returning true, sets '*vlanp' to the effective VLAN of the input + * packet, as returned by flow_get_vlan(). + * + * May also add tags to '*tags', although the current implementation only does + * so in one special case. + */ +static bool +is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, + bool have_packet, + tag_type *tags, int *vlanp, struct ofbundle **in_bundlep) +{ + struct ofport_dpif *in_port; + struct ofbundle *in_bundle; + int vlan; + + /* Find the port and bundle for the received packet. */ + in_port = get_ofp_port(ofproto, flow->in_port); + *in_bundlep = in_bundle = in_port->bundle; + if (!in_port || !in_bundle) { + /* No interface? Something fishy... */ + if (have_packet) { + /* Odd. A few possible reasons here: + * + * - We deleted a port but there are still a few packets queued up + * from it. + * + * - Someone externally added a port (e.g. "ovs-dpctl add-if") that + * we don't know about. + * + * - Packet arrived on the local port but the local port is not + * part of a bundle. + */ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + + VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " + "port %"PRIu16, + ofproto->up.name, flow->in_port); + } + return false; + } + *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet); + if (vlan < 0) { + return false; + } + + /* Drop frames for reserved multicast addresses. */ + if (eth_addr_is_reserved(flow->dl_dst)) { + return false; + } + + /* Drop frames on bundles reserved for mirroring. */ + if (in_bundle->mirror_out) { + if (have_packet) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " + "%s, which is reserved exclusively for mirroring", + ofproto->up.name, in_bundle->name); + } + return false; + } + + if (in_bundle->bond) { + struct mac_entry *mac; + + switch (bond_check_admissibility(in_bundle->bond, in_port, + flow->dl_dst, tags)) { + case BV_ACCEPT: + break; + + case BV_DROP: + return false; + + case BV_DROP_IF_MOVED: + mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); + if (mac && mac->port.p != in_bundle && + (!is_gratuitous_arp(flow) + || mac_entry_is_grat_arp_locked(mac))) { + return false; + } + break; + } + } + + return true; +} + +/* If the composed actions may be applied to any packet in the given 'flow', + * returns true. Otherwise, the actions should only be applied to 'packet', or + * not at all, if 'packet' was NULL. */ +static bool +xlate_normal(struct action_xlate_ctx *ctx) +{ + struct ofbundle *in_bundle; + struct ofbundle *out_bundle; + struct mac_entry *mac; + int vlan; + + /* Check whether we should drop packets in this flow. */ + if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL, + &ctx->tags, &vlan, &in_bundle)) { + out_bundle = NULL; + goto done; + } + + /* Learn source MAC (but don't try to learn from revalidation). */ + if (ctx->packet) { + update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); + } + + /* Determine output bundle. */ + mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, + &ctx->tags); + if (mac) { + out_bundle = mac->port.p; + } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) { + /* If we are revalidating but don't have a learning entry then eject + * the flow. Installing a flow that floods packets opens up a window + * of time where we could learn from a packet reflected on a bond and + * blackhole packets before the learning table is updated to reflect + * the correct port. */ + return false; + } else { + out_bundle = OFBUNDLE_FLOOD; + } + + /* Don't send packets out their input bundles. */ + if (in_bundle == out_bundle) { + out_bundle = NULL; + } + +done: + if (in_bundle) { + compose_actions(ctx, vlan, in_bundle, out_bundle); + } + + return true; +} + +static bool +get_drop_frags(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + bool drop_frags; + + dpif_get_drop_frags(ofproto->dpif, &drop_frags); + return drop_frags; +} + +static void +set_drop_frags(struct ofproto *ofproto_, bool drop_frags) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + + dpif_set_drop_frags(ofproto->dpif, drop_frags); +} + +static int +packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, + const struct flow *flow, + const union ofp_action *ofp_actions, size_t n_ofp_actions) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + int error; + + error = validate_actions(ofp_actions, n_ofp_actions, flow, + ofproto->max_ports); + if (!error) { + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + + action_xlate_ctx_init(&ctx, ofproto, flow, packet); + odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); + dpif_execute(ofproto->dpif, odp_actions->data, odp_actions->size, + packet); + ofpbuf_delete(odp_actions); + } + return error; +} + +static void +get_netflow_ids(const struct ofproto *ofproto_, + uint8_t *engine_type, uint8_t *engine_id) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + + dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); +} + +static struct ofproto_dpif * +ofproto_dpif_lookup(const char *name) +{ + struct ofproto *ofproto = ofproto_lookup(name); + return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class + ? ofproto_dpif_cast(ofproto) + : NULL); +} + +static void +ofproto_unixctl_fdb_show(struct unixctl_conn *conn, + const char *args, void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct ofproto_dpif *ofproto; + const struct mac_entry *e; + + ofproto = ofproto_dpif_lookup(args); + if (!ofproto) { + unixctl_command_reply(conn, 501, "no such bridge"); + return; + } + + ds_put_cstr(&ds, " port VLAN MAC Age\n"); + LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { + struct ofbundle *bundle = e->port.p; + ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", + ofbundle_get_a_port(bundle)->odp_port, + e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); + } + unixctl_command_reply(conn, 200, ds_cstr(&ds)); + ds_destroy(&ds); +} + +struct ofproto_trace { + struct action_xlate_ctx ctx; + struct flow flow; + struct ds *result; +}; + +static void +trace_format_rule(struct ds *result, int level, const struct rule *rule) +{ + ds_put_char_multiple(result, '\t', level); + if (!rule) { + ds_put_cstr(result, "No match\n"); + return; + } + + ds_put_format(result, "Rule: cookie=%#"PRIx64" ", + ntohll(rule->flow_cookie)); + cls_rule_format(&rule->cr, result); + ds_put_char(result, '\n'); + + ds_put_char_multiple(result, '\t', level); + ds_put_cstr(result, "OpenFlow "); + ofp_print_actions(result, (const struct ofp_action_header *) rule->actions, + rule->n_actions * sizeof *rule->actions); + ds_put_char(result, '\n'); +} + +static void +trace_format_flow(struct ds *result, int level, const char *title, + struct ofproto_trace *trace) +{ + ds_put_char_multiple(result, '\t', level); + ds_put_format(result, "%s: ", title); + if (flow_equal(&trace->ctx.flow, &trace->flow)) { + ds_put_cstr(result, "unchanged"); + } else { + flow_format(result, &trace->ctx.flow); + trace->flow = trace->ctx.flow; + } + ds_put_char(result, '\n'); +} + +static void +trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) +{ + struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); + struct ds *result = trace->result; + + ds_put_char(result, '\n'); + trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); + trace_format_rule(result, ctx->recurse + 1, &rule->up); +} + +static void +ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) +{ + char *dpname, *in_port_s, *tun_id_s, *packet_s; + char *args = xstrdup(args_); + char *save_ptr = NULL; + struct ofproto_dpif *ofproto; + struct ofpbuf packet; + struct rule_dpif *rule; + struct ds result; + struct flow flow; + uint16_t in_port; + ovs_be64 tun_id; + char *s; + + ofpbuf_init(&packet, strlen(args) / 2); + ds_init(&result); + + dpname = strtok_r(args, " ", &save_ptr); + tun_id_s = strtok_r(NULL, " ", &save_ptr); + in_port_s = strtok_r(NULL, " ", &save_ptr); + packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ + if (!dpname || !in_port_s || !packet_s) { + unixctl_command_reply(conn, 501, "Bad command syntax"); + goto exit; + } + + ofproto = ofproto_dpif_lookup(dpname); + if (!ofproto) { + unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " + "for help)"); + goto exit; + } + + tun_id = htonll(strtoull(tun_id_s, NULL, 0)); + in_port = ofp_port_to_odp_port(atoi(in_port_s)); + + packet_s = ofpbuf_put_hex(&packet, packet_s, NULL); + packet_s += strspn(packet_s, " "); + if (*packet_s != '\0') { + unixctl_command_reply(conn, 501, "Trailing garbage in command"); + goto exit; + } + if (packet.size < ETH_HEADER_LEN) { + unixctl_command_reply(conn, 501, "Packet data too short for Ethernet"); + goto exit; + } + + ds_put_cstr(&result, "Packet: "); + s = ofp_packet_to_string(packet.data, packet.size, packet.size); + ds_put_cstr(&result, s); + free(s); + + flow_extract(&packet, tun_id, in_port, &flow); + ds_put_cstr(&result, "Flow: "); + flow_format(&result, &flow); + ds_put_char(&result, '\n'); + + rule = rule_dpif_lookup(ofproto, &flow); + trace_format_rule(&result, 0, &rule->up); + if (rule) { + struct ofproto_trace trace; + struct ofpbuf *odp_actions; + + trace.result = &result; + trace.flow = flow; + action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet); + trace.ctx.resubmit_hook = trace_resubmit; + odp_actions = xlate_actions(&trace.ctx, + rule->up.actions, rule->up.n_actions); + + ds_put_char(&result, '\n'); + trace_format_flow(&result, 0, "Final flow", &trace); + ds_put_cstr(&result, "Datapath actions: "); + format_odp_actions(&result, odp_actions->data, odp_actions->size); + ofpbuf_delete(odp_actions); + } + + unixctl_command_reply(conn, 200, ds_cstr(&result)); + +exit: + ds_destroy(&result); + ofpbuf_uninit(&packet); + free(args); +} + +static void +ofproto_dpif_unixctl_init(void) +{ + static bool registered; + if (registered) { + return; + } + registered = true; + + unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); + unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL); +} + +const struct ofproto_class ofproto_dpif_class = { + enumerate_types, + enumerate_names, + del, + alloc, + construct, + destruct, + dealloc, + run, + wait, + flush, + get_features, + get_tables, + port_alloc, + port_construct, + port_destruct, + port_dealloc, + port_modified, + port_reconfigured, + port_query_by_name, + port_add, + port_del, + port_dump_start, + port_dump_next, + port_dump_done, + port_poll, + port_poll_wait, + port_is_lacp_current, + rule_alloc, + rule_construct, + rule_destruct, + rule_dealloc, + rule_get_stats, + rule_execute, + rule_modify_actions, + get_drop_frags, + set_drop_frags, + packet_out, + set_netflow, + get_netflow_ids, + set_sflow, + set_cfm, + get_cfm, + bundle_set, + bundle_remove, + mirror_set, + set_flood_vlans, + is_mirror_output_bundle, +}; diff --cc ofproto/ofproto.c index f0fc4ec73,4ed291437..bfcae57a1 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@@ -1668,28 -2771,19 +1668,21 @@@ handle_table_stats_request(struct ofcon 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); + msg = start_ofp_stats_reply(request, sizeof *ots * p->n_tables); - /* Classifier table. */ - ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); - memset(ots, 0, sizeof *ots); - strcpy(ots->name, "classifier"); - ots->wildcards = htonl(OFPFW_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%d", i); - ots[i].wildcards = htonl(OVSFW_ALL); ++ 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); - if (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10) { - /* OpenFlow 1.0 only supports the OFPFW_* bits. */ - for (i = 0; i < p->n_tables; i++) { - ots[i].wildcards &= htonl(OFPFW_ALL); - } - } - ofconn_send_reply(ofconn, msg); return 0; } @@@ -1770,10 -2864,8 +1763,9 @@@ static voi 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)) { @@@ -1787,11 -2879,10 +1779,10 @@@ ofs = append_ofp_stats_reply(len, ofconn, replyp); ofs->length = htons(len); - ofs->table_id = 0; + ofs->table_id = rule->table_id; 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); + ofputil_cls_rule_to_match(&rule->cr, &ofs->match); + put_32aligned_be64(&ofs->cookie, rule->flow_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); @@@ -1860,12 -2919,13 +1851,12 @@@ handle_flow_stats_request(struct ofcon COVERAGE_INC(ofproto_flows_req); reply = start_ofp_stats_reply(oh, 1024); - ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, &target); - if (is_valid_table(fsr->table_id)) { ++ ofputil_cls_rule_from_match(&fsr->match, 0, &target); + FOR_EACH_MATCHING_TABLE (cls, fsr->table_id, ofproto) { struct cls_cursor cursor; - struct cls_rule target; struct rule *rule; - ofputil_cls_rule_from_match(&fsr->match, 0, &target); - cls_cursor_init(&cursor, &ofproto->cls, &target); + cls_cursor_init(&cursor, cls, &target); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); } @@@ -2503,8 -3543,7 +2493,8 @@@ handle_flow_mod(struct ofconn *ofconn, return error; } - error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn), - error = ofputil_decode_flow_mod(&fm, oh); ++ error = ofputil_decode_flow_mod(&fm, oh, + ofconn_get_flow_mod_table_id(ofconn)); if (error) { return error; } diff --cc ofproto/ofproto.h index ed8d45b3f,9a56beec2..ea9f98b3d --- a/ofproto/ofproto.h +++ b/ofproto/ofproto.h @@@ -163,68 -121,22 +163,69 @@@ void ofproto_set_desc(struct ofproto * int ofproto_set_snoops(struct ofproto *, const struct sset *snoops); int ofproto_set_netflow(struct ofproto *, const struct netflow_options *nf_options); -void ofproto_set_sflow(struct ofproto *, const struct ofproto_sflow_options *); +int ofproto_set_sflow(struct ofproto *, const struct ofproto_sflow_options *); -/* Configuration of individual interfaces. */ -struct cfm; +/* Configuration of ports. */ + +void ofproto_port_unregister(struct ofproto *, uint16_t ofp_port); + +void ofproto_port_clear_cfm(struct ofproto *, uint16_t ofp_port); +void ofproto_port_set_cfm(struct ofproto *, uint16_t ofp_port, + const struct cfm *, + const uint16_t *remote_mps, size_t n_remote_mps); +const struct cfm *ofproto_port_get_cfm(struct ofproto *, uint16_t ofp_port); +int ofproto_port_is_lacp_current(struct ofproto *, uint16_t ofp_port); -void ofproto_iface_clear_cfm(struct ofproto *, uint32_t port_no); -void ofproto_iface_set_cfm(struct ofproto *, uint32_t port_no, - const struct cfm *, - const uint16_t *remote_mps, size_t n_remote_mps); -const struct cfm *ofproto_iface_get_cfm(struct ofproto *, uint32_t port_no); +/* Configuration of bundles. */ +struct ofproto_bundle_settings { + char *name; /* For use in log messages. */ + + uint16_t *slaves; /* OpenFlow port numbers for slaves. */ + size_t n_slaves; + + int vlan; /* VLAN if access port, -1 if trunk port. */ + unsigned long *trunks; /* vlan_bitmap, NULL to trunk all VLANs. */ + + struct bond_settings *bond; /* Must be nonnull iff if n_slaves > 1. */ ++ uint32_t *bond_stable_ids; /* Array of n_slaves elements. */ + + struct lacp_settings *lacp; /* Nonnull to enable LACP. */ + struct lacp_slave_settings *lacp_slaves; /* Array of n_slaves elements. */ +}; + +int ofproto_bundle_register(struct ofproto *, void *aux, + const struct ofproto_bundle_settings *); +int ofproto_bundle_unregister(struct ofproto *, void *aux); + +/* Configuration of mirrors. */ +struct ofproto_mirror_settings { + /* Name for log messages. */ + char *name; + + /* Bundles that select packets for mirroring upon ingress. */ + void **srcs; /* A set of registered ofbundle handles. */ + size_t n_srcs; + + /* Bundles that select packets for mirroring upon egress. */ + void **dsts; /* A set of registered ofbundle handles. */ + size_t n_dsts; + + /* VLANs of packets to select for mirroring. */ + unsigned long *src_vlans; /* vlan_bitmap, NULL selects all VLANs. */ + + /* Output (mutually exclusive). */ + void *out_bundle; /* A registered ofbundle handle or NULL. */ + uint16_t out_vlan; /* Output VLAN, only if out_bundle is NULL. */ +}; + +int ofproto_mirror_register(struct ofproto *, void *aux, + const struct ofproto_mirror_settings *); +int ofproto_mirror_unregister(struct ofproto *, void *aux); + +int ofproto_set_flood_vlans(struct ofproto *, unsigned long *flood_vlans); +bool ofproto_is_mirror_output_bundle(struct ofproto *, void *aux); /* Configuration querying. */ -uint64_t ofproto_get_datapath_id(const struct ofproto *); -bool ofproto_has_primary_controller(const struct ofproto *); -enum ofproto_fail_mode ofproto_get_fail_mode(const struct ofproto *); -void ofproto_get_listeners(const struct ofproto *, struct sset *); bool ofproto_has_snoops(const struct ofproto *); void ofproto_get_snoops(const struct ofproto *, struct sset *); void ofproto_get_all_flows(struct ofproto *p, struct ds *); diff --cc ofproto/private.h index f2e320459,000000000..35a6d1c59 mode 100644,000000..100644 --- a/ofproto/private.h +++ b/ofproto/private.h @@@ -1,743 -1,0 +1,743 @@@ +/* + * Copyright (c) 2009, 2010, 2011 Nicira Networks. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#ifndef OFPROTO_PRIVATE_H +#define OFPROTO_PRIVATE_H 1 + +/* Definitions for use within ofproto. */ + +#include "ofproto/ofproto.h" +#include "classifier.h" +#include "list.h" +#include "shash.h" +#include "timeval.h" + +/* An OpenFlow switch. + * + * With few exceptions, ofproto implementations may look at these fields but + * should not modify them. */ +struct ofproto { + const struct ofproto_class *ofproto_class; + char *type; /* Datapath type. */ + char *name; /* Datapath name. */ + struct hmap_node hmap_node; /* In global 'all_ofprotos' hmap. */ + + /* Settings. */ + uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ + uint64_t datapath_id; /* Datapath ID. */ + char *mfr_desc; /* Manufacturer. */ + char *hw_desc; /* Hardware. */ + char *sw_desc; /* Software version. */ + char *serial_desc; /* Serial number. */ + char *dp_desc; /* Datapath description. */ + + /* Datapath. */ + struct netdev_monitor *netdev_monitor; + struct hmap ports; /* Contains "struct ofport"s. */ + struct shash port_by_name; + + /* Flow tables. */ + struct classifier *tables; /* Each classifier contains "struct rule"s. */ + int n_tables; + + /* OpenFlow connections. */ + struct connmgr *connmgr; +}; + +struct ofproto *ofproto_lookup(const char *name); +struct ofport *ofproto_get_port(const struct ofproto *, uint16_t ofp_port); + +/* An OpenFlow port within a "struct ofproto". + * + * With few exceptions, ofproto implementations may look at these fields but + * should not modify them. */ +struct ofport { + struct ofproto *ofproto; /* The ofproto that contains this port. */ + struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ + struct netdev *netdev; + struct ofp_phy_port opp; + uint16_t ofp_port; /* OpenFlow port number. */ +}; + +/* An OpenFlow flow within a "struct ofproto". + * + * With few exceptions, ofproto implementations may look at these fields but + * should not modify them. */ +struct rule { + struct ofproto *ofproto; /* The ofproto that contains this rule. */ + struct cls_rule cr; /* In owning ofproto's classifier. */ + + ovs_be64 flow_cookie; /* Controller-issued identifier. */ + + long long int created; /* Creation time. */ + uint16_t idle_timeout; /* In seconds from time of last use. */ + uint16_t hard_timeout; /* In seconds from time of creation. */ + uint8_t table_id; /* Index in ofproto's 'tables' array. */ + bool send_flow_removed; /* Send a flow removed message? */ + + union ofp_action *actions; /* OpenFlow actions. */ + int n_actions; /* Number of elements in actions[]. */ +}; + +static inline struct rule * +rule_from_cls_rule(const struct cls_rule *cls_rule) +{ + return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; +} + +void ofproto_rule_expire(struct rule *, uint8_t reason); +void ofproto_rule_destroy(struct rule *); + +/* ofproto class structure, to be defined by each ofproto implementation. + * + * + * Data Structures + * =============== + * + * These functions work primarily with three different kinds of data + * structures: + * + * - "struct ofproto", which represents an OpenFlow switch. + * + * - "struct ofport", which represents a port within an ofproto. + * + * - "struct rule", which represents an OpenFlow flow within an ofproto. + * + * Each of these data structures contains all of the implementation-independent + * generic state for the respective concept, called the "base" state. None of + * them contains any extra space for ofproto implementations to use. Instead, + * each implementation is expected to declare its own data structure that + * contains an instance of the generic data structure plus additional + * implementation-specific members, called the "derived" state. The + * implementation can use casts or (preferably) the CONTAINER_OF macro to + * obtain access to derived state given only a pointer to the embedded generic + * data structure. + * + * + * Life Cycle + * ========== + * + * Four stylized functions accompany each of these data structures: + * + * "alloc" "construct" "destruct" "dealloc" + * ------------ ---------------- --------------- -------------- + * ofproto ->alloc ->construct ->destruct ->dealloc + * ofport ->port_alloc ->port_construct ->port_destruct ->port_dealloc + * rule ->rule_alloc ->rule_construct ->rule_destruct ->rule_dealloc + * + * Any instance of a given data structure goes through the following life + * cycle: + * + * 1. The client calls the "alloc" function to obtain raw memory. If "alloc" + * fails, skip all the other steps. + * + * 2. The client initializes all of the data structure's base state. If this + * fails, skip to step 7. + * + * 3. The client calls the "construct" function. The implementation + * initializes derived state. It may refer to the already-initialized + * base state. If "construct" fails, skip to step 6. + * + * 4. The data structure is now initialized and in use. + * + * 5. When the data structure is no longer needed, the client calls the + * "destruct" function. The implementation uninitializes derived state. + * The base state has not been uninitialized yet, so the implementation + * may still refer to it. + * + * 6. The client uninitializes all of the data structure's base state. + * + * 7. The client calls the "dealloc" to free the raw memory. The + * implementation must not refer to base or derived state in the data + * structure, because it has already been uninitialized. + * + * Each "alloc" function allocates and returns a new instance of the respective + * data structure. The "alloc" function is not given any information about the + * use of the new data structure, so it cannot perform much initialization. + * Its purpose is just to ensure that the new data structure has enough room + * for base and derived state. It may return a null pointer if memory is not + * available, in which case none of the other functions is called. + * + * Each "construct" function initializes derived state in its respective data + * structure. When "construct" is called, all of the base state has already + * been initialized, so the "construct" function may refer to it. The + * "construct" function is allowed to fail, in which case the client calls the + * "dealloc" function (but not the "destruct" function). + * + * Each "destruct" function uninitializes and frees derived state in its + * respective data structure. When "destruct" is called, the base state has + * not yet been uninitialized, so the "destruct" function may refer to it. The + * "destruct" function is not allowed to fail. + * + * Each "dealloc" function frees raw memory that was allocated by the the + * "alloc" function. The memory's base and derived members might not have ever + * been initialized (but if "construct" returned successfully, then it has been + * "destruct"ed already). The "dealloc" function is not allowed to fail. + * + * + * Conventions + * =========== + * + * Most of these functions return 0 if they are successful or a positive error + * code on failure. Depending on the function, valid error codes are either + * errno values or OpenFlow error codes constructed with ofp_mkerr(). + * + * Most of these functions are expected to execute synchronously, that is, to + * block as necessary to obtain a result. Thus, these functions may return + * EAGAIN (or EWOULDBLOCK or EINPROGRESS) only where the function descriptions + * explicitly say those errors are a possibility. We may relax this + * requirement in the future if and when we encounter performance problems. */ +struct ofproto_class { +/* ## ----------------- ## */ +/* ## Factory Functions ## */ +/* ## ----------------- ## */ + + /* Enumerates the types of all support ofproto types into 'types'. The + * caller has already initialized 'types' and other ofproto classes might + * already have added names to it. */ + void (*enumerate_types)(struct sset *types); + + /* Enumerates the names of all existing datapath of the specified 'type' + * into 'names' 'all_dps'. The caller has already initialized 'names' as + * an empty sset. + * + * 'type' is one of the types enumerated by ->enumerate_types(). + * + * Returns 0 if successful, otherwise a positive errno value. + */ + int (*enumerate_names)(const char *type, struct sset *names); + + /* Deletes the datapath with the specified 'type' and 'name'. The caller + * should have closed any open ofproto with this 'type' and 'name'; this + * function is allowed to fail if that is not the case. + * + * 'type' is one of the types enumerated by ->enumerate_types(). + * 'name' is one of the names enumerated by ->enumerate_names() for 'type'. + * + * Returns 0 if successful, otherwise a positive errno value. + */ + int (*del)(const char *type, const char *name); + +/* ## --------------------------- ## */ +/* ## Top-Level ofproto Functions ## */ +/* ## --------------------------- ## */ + + /* Life-cycle functions for an "ofproto" (see "Life Cycle" above). + * + * ->construct() should not modify most base members of the ofproto. In + * particular, the client will initialize the ofproto's 'ports' member + * after construction is complete. + * + * ->construct() should initialize the base 'n_tables' member to the number + * of flow tables supported by the datapath (between 1 and 254, inclusive), + * initialize the base 'tables' member with space for one classifier per + * table, and initialize each classifier with classifier_init. Each flow + * table should be initially empty, so ->construct() should delete flows + * from the underlying datapath, if necessary, rather than populating the + * tables. + * + * Only one ofproto instance needs to be supported for any given datapath. + * If a datapath is already open as part of one "ofproto", then another + * attempt to "construct" the same datapath as part of another ofproto is + * allowed to fail with an error. + * + * ->construct() returns 0 if successful, otherwise a positive errno + * value. */ + struct ofproto *(*alloc)(void); + int (*construct)(struct ofproto *ofproto); + void (*destruct)(struct ofproto *ofproto); + void (*dealloc)(struct ofproto *ofproto); + + /* Performs any periodic activity required by 'ofproto'. It should: + * + * - Call connmgr_send_packet_in() for each received packet that missed + * in the OpenFlow flow table or that had a OFPP_CONTROLLER output + * action. + * + * - Call ofproto_rule_expire() for each OpenFlow flow that has reached + * its hard_timeout or idle_timeout, to expire the flow. + * + * Returns 0 if successful, otherwise a positive errno value. The ENODEV + * return value specifically means that the datapath underlying 'ofproto' + * has been destroyed (externally, e.g. by an admin running ovs-dpctl). + */ + int (*run)(struct ofproto *ofproto); + + /* Causes the poll loop to wake up when 'ofproto''s 'run' function needs to + * be called, e.g. by calling the timer or fd waiting functions in + * poll-loop.h. */ + void (*wait)(struct ofproto *ofproto); + + /* Every "struct rule" in 'ofproto' is about to be deleted, one by one. + * This function may prepare for that, for example by clearing state in + * advance. It should *not* actually delete any "struct rule"s from + * 'ofproto', only prepare for it. + * + * This function is optional; it's really just for optimization in case + * it's cheaper to delete all the flows from your hardware in a single pass + * than to do it one by one. */ + void (*flush)(struct ofproto *ofproto); + + /* Helper for the OpenFlow OFPT_FEATURES_REQUEST request. + * + * The implementation should store true in '*arp_match_ip' if the switch + * supports matching IP addresses inside ARP requests and replies, false + * otherwise. + * + * The implementation should store in '*actions' a bitmap of the supported + * OpenFlow actions: the bit with value (1 << n) should be set to 1 if the + * implementation supports the action with value 'n', and to 0 otherwise. + * For example, if the implementation supports the OFPAT_OUTPUT and + * OFPAT_ENQUEUE actions, but no others, it would set '*actions' to (1 << + * OFPAT_OUTPUT) | (1 << OFPAT_ENQUEUE). Vendor actions are not included + * in '*actions'. */ + void (*get_features)(struct ofproto *ofproto, + bool *arp_match_ip, uint32_t *actions); + + /* Helper for the OpenFlow OFPST_TABLE statistics request. + * + * The 'ots' array contains 'ofproto->n_tables' elements. Each element is + * initialized as: + * + * - 'table_id' to the array index. + * + * - 'name' to "table#" where # is the table ID. + * - * - 'wildcards' to OVSFW_ALL. ++ * - 'wildcards' to OFPFW_ALL. + * + * - 'max_entries' to 1,000,000. + * + * - 'active_count' to the classifier_count() for the table. + * + * - 'lookup_count' and 'matched_count' to 0. + * + * The implementation should update any members in each element for which + * it has better values: + * + * - 'name' to a more meaningful name. + * + * - 'wildcards' to the set of wildcards actually supported by the table + * (if it doesn't support all OpenFlow wildcards). + * + * - 'max_entries' to the maximum number of flows actually supported by + * the hardware. + * + * - 'lookup_count' to the number of packets looked up in this flow table + * so far. + * + * - 'matched_count' to the number of packets looked up in this flow + * table so far that matched one of the flow entries. + * + * Keep in mind that all of the members of struct ofp_table_stats are in + * network byte order. + */ + void (*get_tables)(struct ofproto *ofproto, struct ofp_table_stats *ots); + +/* ## ---------------- ## */ +/* ## ofport Functions ## */ +/* ## ---------------- ## */ + + /* Life-cycle functions for a "struct ofport" (see "Life Cycle" above). + * + * ->port_construct() should not modify any base members of the ofport. + * + * ofports are managed by the base ofproto code. The ofproto + * implementation should only create and destroy them in response to calls + * to these functions. The base ofproto code will create and destroy + * ofports in the following situations: + * + * - Just after the ->construct() function is called, the base ofproto + * iterates over all of the implementation's ports, using + * ->port_dump_start() and related functions, and constructs an ofport + * for each dumped port. + * + * - If ->port_poll() reports that a specific port has changed, then the + * base ofproto will query that port with ->port_query_by_name() and + * construct or destruct ofports as necessary to reflect the updated + * set of ports. + * + * - If ->port_poll() returns ENOBUFS to report an unspecified port set + * change, then the base ofproto will iterate over all of the + * implementation's ports, in the same way as at ofproto + * initialization, and construct and destruct ofports to reflect all of + * the changes. + * + * ->port_construct() returns 0 if successful, otherwise a positive errno + * value. + */ + struct ofport *(*port_alloc)(void); + int (*port_construct)(struct ofport *ofport); + void (*port_destruct)(struct ofport *ofport); + void (*port_dealloc)(struct ofport *ofport); + + /* Called after 'ofport->netdev' is replaced by a new netdev object. If + * the ofproto implementation uses the ofport's netdev internally, then it + * should switch to using the new one. The old one has been closed. + * + * An ofproto implementation that doesn't need to do anything in this + * function may use a null pointer. */ + void (*port_modified)(struct ofport *ofport); + + /* Called after an OpenFlow OFPT_PORT_MOD request changes a port's + * configuration. 'ofport->opp.config' contains the new configuration. + * 'old_config' contains the previous configuration. + * + * The caller implements OFPPC_PORT_DOWN using netdev functions to turn + * NETDEV_UP on and off, so this function doesn't have to do anything for + * that bit (and it won't be called if that is the only bit that + * changes). */ + void (*port_reconfigured)(struct ofport *ofport, ovs_be32 old_config); + + /* Looks up a port named 'devname' in 'ofproto'. On success, initializes + * '*port' appropriately. + * + * The caller owns the data in 'port' and must free it with + * ofproto_port_destroy() when it is no longer needed. */ + int (*port_query_by_name)(const struct ofproto *ofproto, + const char *devname, struct ofproto_port *port); + + /* Attempts to add 'netdev' as a port on 'ofproto'. Returns 0 if + * successful, otherwise a positive errno value. If successful, sets + * '*ofp_portp' to the new port's port number. + * + * It doesn't matter whether the new port will be returned by a later call + * to ->port_poll(); the implementation may do whatever is more + * convenient. */ + int (*port_add)(struct ofproto *ofproto, struct netdev *netdev, + uint16_t *ofp_portp); + + /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. Returns + * 0 if successful, otherwise a positive errno value. + * + * It doesn't matter whether the new port will be returned by a later call + * to ->port_poll(); the implementation may do whatever is more + * convenient. */ + int (*port_del)(struct ofproto *ofproto, uint16_t ofp_port); + + /* Attempts to begin dumping the ports in 'ofproto'. On success, returns 0 + * and initializes '*statep' with any data needed for iteration. On + * failure, returns a positive errno value. */ + int (*port_dump_start)(const struct ofproto *ofproto, void **statep); + + /* Attempts to retrieve another port from 'ofproto' for 'state', which was + * initialized by a successful call to the 'port_dump_start' function for + * 'ofproto'. On success, stores a new ofproto_port into 'port' and + * returns 0. Returns EOF if the end of the port table has been reached, + * or a positive errno value on error. This function will not be called + * again once it returns nonzero once for a given iteration (but the + * 'port_dump_done' function will be called afterward). + * + * The ofproto provider retains ownership of the data stored in 'port'. It + * must remain valid until at least the next call to 'port_dump_next' or + * 'port_dump_done' for 'state'. */ + int (*port_dump_next)(const struct ofproto *ofproto, void *state, + struct ofproto_port *port); + + /* Releases resources from 'ofproto' for 'state', which was initialized by + * a successful call to the 'port_dump_start' function for 'ofproto'. */ + int (*port_dump_done)(const struct ofproto *ofproto, void *state); + + /* Polls for changes in the set of ports in 'ofproto'. If the set of ports + * in 'ofproto' has changed, then this function should do one of the + * following: + * + * - Preferably: store the name of the device that was added to or deleted + * from 'ofproto' in '*devnamep' and return 0. The caller is responsible + * for freeing '*devnamep' (with free()) when it no longer needs it. + * + * - Alternatively: return ENOBUFS, without indicating the device that was + * added or deleted. + * + * Occasional 'false positives', in which the function returns 0 while + * indicating a device that was not actually added or deleted or returns + * ENOBUFS without any change, are acceptable. + * + * The purpose of 'port_poll' is to let 'ofproto' know about changes made + * externally to the 'ofproto' object, e.g. by a system administrator via + * ovs-dpctl. Therefore, it's OK, and even preferable, for port_poll() to + * not report changes made through calls to 'port_add' or 'port_del' on the + * same 'ofproto' object. (But it's OK for it to report them too, just + * slightly less efficient.) + * + * If the set of ports in 'ofproto' has not changed, returns EAGAIN. May + * also return other positive errno values to indicate that something has + * gone wrong. + * + * If the set of ports in a datapath is fixed, or if the only way that the + * set of ports in a datapath can change is through ->port_add() and + * ->port_del(), then this function may be a null pointer. + */ + int (*port_poll)(const struct ofproto *ofproto, char **devnamep); + + /* Arranges for the poll loop to wake up when ->port_poll() will return a + * value other than EAGAIN. + * + * If the set of ports in a datapath is fixed, or if the only way that the + * set of ports in a datapath can change is through ->port_add() and + * ->port_del(), or if the poll loop will always wake up anyway when + * ->port_poll() will return a value other than EAGAIN, then this function + * may be a null pointer. + */ + void (*port_poll_wait)(const struct ofproto *ofproto); + + /* Checks the status of LACP negotiation for 'port'. Returns 1 if LACP + * partner information for 'port' is up-to-date, 0 if LACP partner + * information is not current (generally indicating a connectivity + * problem), or -1 if LACP is not enabled on 'port'. + * + * This function may be a null pointer if the ofproto implementation does + * not support LACP. */ + int (*port_is_lacp_current)(const struct ofport *port); + +/* ## ----------------------- ## */ +/* ## OpenFlow Rule Functions ## */ +/* ## ----------------------- ## */ + + /* Life-cycle functions for a "struct rule" (see "Life Cycle" above). + * + * ->rule_construct() should first check whether the rule is acceptable: + * + * - Validate that the matching rule in 'rule->cr' is supported by the + * datapath. If not, then return an OpenFlow error code (as returned + * by ofp_mkerr()). + * + * For example, if the datapath does not support registers, then it + * should return an error if 'rule->cr' does not wildcard all + * registers. + * + * - Validate that 'rule->actions' and 'rule->n_actions' are well-formed + * OpenFlow actions that can be correctly implemented by the datapath. + * If not, then return an OpenFlow error code (as returned by + * ofp_mkerr()). + * + * The validate_actions() function (in ofp-util.c) can be useful as a + * model for action validation, but it accepts all of the OpenFlow + * actions that OVS understands. If your ofproto implementation only + * implements a subset of those, then you should implement your own + * action validation. + * + * If the rule is acceptable, then ->rule_construct() should modify the + * flow table: + * + * - If there was already a rule with exactly the same matching criteria + * and priority in the classifier, then it should destroy it (with + * ofproto_rule_destroy()). + * + * To the greatest extent possible, the old rule should be destroyed + * only if inserting the new rule succeeds; that is, ->rule_construct() + * should be transactional. + * + * The function classifier_find_rule_exactly() can locate such a rule. + * + * - Insert the new rule into the ofproto's 'cls' classifier, and into + * the datapath flow table. + * + * The function classifier_insert() inserts a rule into the classifier. + * + * Other than inserting 'rule->cr' into the classifier, ->rule_construct() + * should not modify any base members of struct rule. + * + * ->rule_destruct() should remove 'rule' from the ofproto's 'cls' + * classifier (e.g. with classifier_remove()) and from the datapath flow + * table. */ + struct rule *(*rule_alloc)(void); + int (*rule_construct)(struct rule *rule); + void (*rule_destruct)(struct rule *rule); + void (*rule_dealloc)(struct rule *rule); + + /* Obtains statistics for 'rule', storing the number of packets that have + * matched it in '*packet_count' and the number of bytes in those packets + * in '*byte_count'. */ + void (*rule_get_stats)(struct rule *rule, uint64_t *packet_count, + uint64_t *byte_count); + + /* Applies the actions in 'rule' to 'packet'. (This implements sending + * buffered packets for OpenFlow OFPT_FLOW_MOD commands.) + * + * Takes ownership of 'packet' (so it should eventually free it, with + * ofpbuf_delete()). + * + * 'flow' reflects the flow information for 'packet'. All of the + * information in 'flow' is extracted from 'packet', except for + * flow->tun_id and flow->in_port, which are assigned the correct values + * for the incoming packet. The register values are zeroed. + * + * The statistics for 'packet' should be included in 'rule'. + * + * Returns 0 if successful, otherwise an OpenFlow error code (as returned + * by ofp_mkerr()). */ + int (*rule_execute)(struct rule *rule, struct flow *flow, + struct ofpbuf *packet); + + /* Validates that the 'n' elements in 'actions' are well-formed OpenFlow + * actions that can be correctly implemented by the datapath. If not, then + * return an OpenFlow error code (as returned by ofp_mkerr()). If so, + * then update the datapath to implement the new actions and return 0. + * + * When this function runs, 'rule' still has its original actions. If this + * function returns 0, then the caller will update 'rule' with the new + * actions and free the old ones. */ + int (*rule_modify_actions)(struct rule *rule, + const union ofp_action *actions, size_t n); + + /* These functions implement the OpenFlow IP fragment handling policy. By + * default ('drop_frags' == false), an OpenFlow switch should treat IP + * fragments the same way as other packets (although TCP and UDP port + * numbers cannot be determined). With 'drop_frags' == true, the switch + * should drop all IP fragments without passing them through the flow + * table. */ + bool (*get_drop_frags)(struct ofproto *ofproto); + void (*set_drop_frags)(struct ofproto *ofproto, bool drop_frags); + + /* Implements the OpenFlow OFPT_PACKET_OUT command. The datapath should + * execute the 'n_actions' in the 'actions' array on 'packet'. + * + * The caller retains ownership of 'packet', so ->packet_out() should not + * modify or free it. + * + * This function must validate that the 'n_actions' elements in 'actions' + * are well-formed OpenFlow actions that can be correctly implemented by + * the datapath. If not, then it should return an OpenFlow error code (as + * returned by ofp_mkerr()). + * + * 'flow' reflects the flow information for 'packet'. All of the + * information in 'flow' is extracted from 'packet', except for + * flow->in_port, which is taken from the OFPT_PACKET_OUT message. + * flow->tun_id and its register values are zeroed. + * + * 'packet' is not matched against the OpenFlow flow table, so its + * statistics should not be included in OpenFlow flow statistics. + * + * Returns 0 if successful, otherwise an OpenFlow error code (as returned + * by ofp_mkerr()). */ + int (*packet_out)(struct ofproto *ofproto, struct ofpbuf *packet, + const struct flow *flow, + const union ofp_action *actions, + size_t n_actions); + +/* ## ------------------------- ## */ +/* ## OFPP_NORMAL configuration ## */ +/* ## ------------------------- ## */ + + /* Configures NetFlow on 'ofproto' according to the options in + * 'netflow_options', or turns off NetFlow if 'netflow_options' is NULL. + * + * EOPNOTSUPP as a return value indicates that 'ofproto' does not support + * sFlow, as does a null pointer. */ + int (*set_netflow)(struct ofproto *ofproto, + const struct netflow_options *netflow_options); + + void (*get_netflow_ids)(const struct ofproto *ofproto, + uint8_t *engine_type, uint8_t *engine_id); + + /* Configures sFlow on 'ofproto' according to the options in + * 'sflow_options', or turns off sFlow if 'sflow_options' is NULL. + * + * EOPNOTSUPP as a return value indicates that 'ofproto' does not support + * sFlow, as does a null pointer. */ + int (*set_sflow)(struct ofproto *ofproto, + const struct ofproto_sflow_options *sflow_options); + + /* Configures connectivity fault management on 'ofport'. + * + * If 'cfm' is nonnull, 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'. + * + * If 'cfm' is null, removes any connectivity fault management + * configuration from 'ofport'. + * + * EOPNOTSUPP as a return value indicates that this ofproto_class does not + * support CFM, as does a null pointer. */ + int (*set_cfm)(struct ofport *ofport, const struct cfm *cfm, + const uint16_t *remote_mps, size_t n_remote_mps); + + /* Stores the connectivity fault management object associated with 'ofport' + * in '*cfmp'. Stores a null pointer in '*cfmp' if CFM is not configured + * on 'ofport'. The caller must not modify or destroy the returned object. + * + * This function may be NULL if this ofproto_class does not support CFM. */ + int (*get_cfm)(struct ofport *ofport, const struct cfm **cfmp); + + /* If 's' is nonnull, this function registers a "bundle" associated with + * client data pointer 'aux' in 'ofproto'. A bundle is the same concept as + * a Port in OVSDB, that is, it consists of one or more "slave" devices + * (Interfaces, in OVSDB) along with VLAN and LACP configuration and, if + * there is more than one slave, a bonding configuration. If 'aux' is + * already registered then this function updates its configuration to 's'. + * Otherwise, this function registers a new bundle. + * + * If 's' is NULL, this function unregisters the bundle registered on + * 'ofproto' associated with client data pointer 'aux'. If no such bundle + * has been registered, this has no effect. + * + * This function affects only the behavior of the NXAST_AUTOPATH action and + * output to the OFPP_NORMAL port. An implementation that does not support + * it at all may set it to NULL or return EOPNOTSUPP. An implementation + * that supports only a subset of the functionality should implement what + * it can and return 0. */ + int (*bundle_set)(struct ofproto *ofproto, void *aux, + const struct ofproto_bundle_settings *s); + + /* If 'port' is part of any bundle, removes it from that bundle. If the + * bundle now has no ports, deletes the bundle. If the bundle now has only + * one port, deconfigures the bundle's bonding configuration. */ + void (*bundle_remove)(struct ofport *ofport); + + /* If 's' is nonnull, this function registers a mirror associated with + * client data pointer 'aux' in 'ofproto'. A mirror is the same concept as + * a Mirror in OVSDB. If 'aux' is already registered then this function + * updates its configuration to 's'. Otherwise, this function registers a + * new mirror. + * + * If 's' is NULL, this function unregisters the mirror registered on + * 'ofproto' associated with client data pointer 'aux'. If no such mirror + * has been registered, this has no effect. + * + * This function affects only the behavior of the OFPP_NORMAL action. An + * implementation that does not support it at all may set it to NULL or + * return EOPNOTSUPP. An implementation that supports only a subset of the + * functionality should implement what it can and return 0. */ + int (*mirror_set)(struct ofproto *ofproto, void *aux, + const struct ofproto_mirror_settings *s); + + /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs + * on which all packets are flooded, instead of using MAC learning. If + * 'flood_vlans' is NULL, then MAC learning applies to all VLANs. + * + * This function affects only the behavior of the OFPP_NORMAL action. An + * implementation that does not support it may set it to NULL or return + * EOPNOTSUPP. */ + int (*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 (*is_mirror_output_bundle)(struct ofproto *ofproto, void *aux); +}; + +extern const struct ofproto_class ofproto_dpif_class; + +int ofproto_class_register(const struct ofproto_class *); +int ofproto_class_unregister(const struct ofproto_class *); + +void ofproto_add_flow(struct ofproto *, const struct cls_rule *, + const union ofp_action *, size_t n_actions); +void ofproto_delete_flow(struct ofproto *, const struct cls_rule *); +void ofproto_flush_flows(struct ofproto *); + +#endif /* ofproto/private.h */ diff --cc tests/ofproto.at index 6f785e97d,10080e671..713cc7695 --- a/tests/ofproto.at +++ b/tests/ofproto.at @@@ -48,13 -48,9 +48,10 @@@ AT_CHECK([ovs-ofctl dump-flows br0 | ST ]) AT_CHECK([echo 'in_port=1,actions=0' | ovs-ofctl add-flows br0 -]) AT_CHECK([ovs-ofctl add-flow br0 in_port=0,actions=1]) - dnl Tests for a bug in which ofproto ignored tun_id in tun_id_from_cookie - dnl flow_mod commands. - AT_CHECK([ovs-ofctl add-flow -F tun_id_from_cookie br0 tun_id=1,actions=mod_vlan_vid:4]) AT_CHECK([ovs-ofctl dump-flows br0 | STRIP_XIDS | STRIP_DURATION | sort], [0], [dnl + cookie=0x0, duration=?s, table_id=0, n_packets=0, n_bytes=0, in_port=0 actions=output:1 cookie=0x0, duration=?s, table_id=0, n_packets=0, n_bytes=0, in_port=1 actions=output:0 - cookie=0x100000000, duration=?s, table_id=0, n_packets=0, n_bytes=0, tun_id=0x1 actions=mod_vlan_vid:4 + cookie=0x0, duration=?s, table_id=0, n_packets=0, n_bytes=0, in_port=65534 actions=output:1 NXST_FLOW reply: ]) AT_CHECK([ovs-ofctl del-flows br0]) diff --cc tests/ovs-ofctl.at index a9c3c2053,101471082..a8cdbe9a9 --- a/tests/ovs-ofctl.at +++ b/tests/ovs-ofctl.at @@@ -42,17 -40,11 +40,11 @@@ normalization changed ofp_match, detail pre: wildcards= 0x3820f8 in_port=65534 dl_src=00:0a:e4:25:6b:b0 dl_dst=00:00:00:00:00:00 dl_vlan= 9 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 post: wildcards= 0x3ffff8 in_port=65534 dl_src=00:0a:e4:25:6b:b0 dl_dst=00:00:00:00:00:00 dl_vlan= 9 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 normalization changed ofp_match, details: - pre: wildcards= 0x3820ff in_port=65534 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 -post: wildcards= 0x3fffff in_port=65534 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 + pre: wildcards= 0x3820ff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 +post: wildcards= 0x3fffff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 normalization changed ofp_match, details: - pre: wildcards= 0x3820ff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 - post: wildcards= 0x3fffff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 - normalization changed ofp_match, details: - pre: wildcards= 0x3820ff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 - post: wildcards= 0x3fffff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 - normalization changed ofp_match, details: - pre: wildcards= 0x23820ff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 - post: wildcards= 0x23fffff in_port= 0 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 + pre: wildcards= 0x3820ff in_port=65534 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 + post: wildcards= 0x3fffff in_port=65534 dl_src=00:00:00:00:00:00 dl_dst=00:00:00:00:00:00 dl_vlan= 0 dl_vlan_pcp= 0 dl_type= 0 nw_tos= 0 nw_proto= 0 nw_src= 0 nw_dst= 0 tp_src= 0 tp_dst= 0 ]) AT_CLEANUP diff --cc vswitchd/bridge.c index 761d02ba3,55c9f40af..127e0b7cf --- a/vswitchd/bridge.c +++ b/vswitchd/bridge.c @@@ -2019,9 -2908,9 +2019,12 @@@ port_del_ifaces(struct port *port /* Collect list of new interfaces. */ sset_init(&new_ifaces); - for (i = 0; i < cfg->n_interfaces; i++) { - const char *name = cfg->interfaces[i]->name; - sset_add(&new_ifaces, name); + for (i = 0; i < port->cfg->n_interfaces; i++) { + const char *name = port->cfg->interfaces[i]->name; - sset_add(&new_ifaces, name); ++ const char *type = port->cfg->interfaces[i]->name; ++ if (strcmp(type, "null")) { ++ sset_add(&new_ifaces, name); ++ } } /* Get rid of deleted interfaces. */ @@@ -2034,38 -2923,46 +2037,39 @@@ sset_destroy(&new_ifaces); } -/* Expires all MAC learning entries associated with 'port' and forces ofproto - * to revalidate every flow. */ +/* Adds new interfaces to 'port' and updates 'type' and 'cfg' members of + * existing ones. */ static void -port_flush_macs(struct port *port) +port_add_ifaces(struct port *port) { - struct bridge *br = port->bridge; - struct mac_learning *ml = br->ml; - struct mac_entry *mac, *next_mac; + struct shash new_ifaces; + struct shash_node *node; + size_t i; - bridge_flush(br); - LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { - if (mac->port.p == port) { - mac_learning_expire(ml, mac); + /* Collect new ifaces. */ + shash_init(&new_ifaces); + for (i = 0; i < port->cfg->n_interfaces; i++) { + const struct ovsrec_interface *cfg = port->cfg->interfaces[i]; - if (!shash_add_once(&new_ifaces, cfg->name, cfg)) { ++ if (strcmp(cfg->type, "null") ++ && !shash_add_once(&new_ifaces, cfg->name, cfg)) { + VLOG_WARN("port %s: %s specified twice as port interface", + port->name, cfg->name); + iface_set_ofport(cfg, -1); } } -} - -static void -port_reconfigure(struct port *port, const struct ovsrec_port *cfg) -{ - struct sset new_ifaces; - bool need_flush = false; - unsigned long *trunks; - int vlan; - size_t i; - - port->cfg = cfg; - /* Add new interfaces and update 'cfg' member of existing ones. */ - sset_init(&new_ifaces); - for (i = 0; i < cfg->n_interfaces; i++) { - const struct ovsrec_interface *if_cfg = cfg->interfaces[i]; + /* Create new interfaces. + * Update interface types and 'cfg' members. */ + SHASH_FOR_EACH (node, &new_ifaces) { + const struct ovsrec_interface *cfg = node->data; + const char *iface_name = node->name; struct iface *iface; - const char *type; - if (!sset_add(&new_ifaces, if_cfg->name)) { - VLOG_WARN("port %s: %s specified twice as port interface", - port->name, if_cfg->name); - iface_set_ofport(if_cfg, -1); - continue; + iface = iface_lookup(port->bridge, iface_name); + if (!iface) { + iface = iface_create(port, cfg); + } else { + iface->cfg = cfg; } /* Determine interface type. The local port always has type @@@ -2155,89 -3192,140 +2159,98 @@@ port_configure_lacp(struct port *port, /* Prefer bondable links if unspecified. */ priority = atoi(get_port_other_config(port->cfg, "lacp-system-priority", "0")); - s.priority = (priority > 0 && priority <= UINT16_MAX - ? priority - : UINT16_MAX - !list_is_short(&port->ifaces)); + s->priority = (priority > 0 && priority <= UINT16_MAX + ? priority + : UINT16_MAX - !list_is_short(&port->ifaces)); + - s->strict = !strcmp(get_port_other_config(port->cfg, "lacp-strict", - "false"), - "true"); ++ s->heartbeat = !strcmp(get_port_other_config(port->cfg, ++ "lacp-heartbeat", ++ "false"), "true"); + - s.heartbeat = !strcmp(get_port_other_config(port->cfg, - "lacp-heartbeat", - "false"), "true"); lacp_time = get_port_other_config(port->cfg, "lacp-time", "slow"); custom_time = atoi(lacp_time); if (!strcmp(lacp_time, "fast")) { - s.lacp_time = LACP_TIME_FAST; + s->lacp_time = LACP_TIME_FAST; } else if (!strcmp(lacp_time, "slow")) { - s.lacp_time = LACP_TIME_SLOW; + s->lacp_time = LACP_TIME_SLOW; } else if (custom_time > 0) { - s.lacp_time = LACP_TIME_CUSTOM; - s.custom_time = custom_time; + s->lacp_time = LACP_TIME_CUSTOM; + s->custom_time = custom_time; } else { - s.lacp_time = LACP_TIME_SLOW; + s->lacp_time = LACP_TIME_SLOW; } - if (!port->lacp) { - port->lacp = lacp_create(); + return s; +} + +static void +iface_configure_lacp(struct iface *iface, struct lacp_slave_settings *s) +{ - int priority, portid; ++ int priority, portid, key; + + portid = atoi(get_interface_other_config(iface->cfg, "lacp-port-id", "0")); + priority = atoi(get_interface_other_config(iface->cfg, + "lacp-port-priority", "0")); ++ key = atoi(get_interface_other_config(iface->cfg, "lacp-aggregation-key", ++ "0")); + + if (portid <= 0 || portid > UINT16_MAX) { + portid = iface->ofp_port; } - lacp_configure(port->lacp, &s); + if (priority <= 0 || priority > UINT16_MAX) { + priority = UINT16_MAX; + } - LIST_FOR_EACH (iface, port_elem, &port->ifaces) { - iface_reconfigure_lacp(iface); ++ if (key < 0 || key > UINT16_MAX) { ++ key = 0; + } ++ + s->name = iface->name; + s->id = portid; + s->priority = priority; ++ s->key = key; } static void -port_reconfigure_bond(struct port *port) +port_configure_bond(struct port *port, struct bond_settings *s) { - struct bond_settings s; const char *detect_s; - struct iface *iface; - if (list_is_short(&port->ifaces)) { - /* Not a bonded port. */ - bond_destroy(port->bond); - port->bond = NULL; - return; - } - - port->bridge->has_bonded_ports = true; - - s.name = port->name; - s.balance = BM_SLB; + s->name = port->name; + s->balance = BM_SLB; if (port->cfg->bond_mode - && !bond_mode_from_string(&s.balance, port->cfg->bond_mode)) { + && !bond_mode_from_string(&s->balance, port->cfg->bond_mode)) { VLOG_WARN("port %s: unknown bond_mode %s, defaulting to %s", port->name, port->cfg->bond_mode, - bond_mode_to_string(s.balance)); + bond_mode_to_string(s->balance)); } - s.detect = BLSM_CARRIER; + s->detect = BLSM_CARRIER; detect_s = get_port_other_config(port->cfg, "bond-detect-mode", NULL); - if (detect_s && !bond_detect_mode_from_string(&s.detect, detect_s)) { + if (detect_s && !bond_detect_mode_from_string(&s->detect, detect_s)) { VLOG_WARN("port %s: unsupported bond-detect-mode %s, " "defaulting to %s", - port->name, detect_s, bond_detect_mode_to_string(s.detect)); + port->name, detect_s, bond_detect_mode_to_string(s->detect)); } - s.miimon_interval = atoi( + s->miimon_interval = atoi( get_port_other_config(port->cfg, "bond-miimon-interval", "200")); - if (s.miimon_interval < 100) { - s.miimon_interval = 100; + if (s->miimon_interval < 100) { + s->miimon_interval = 100; } - s.up_delay = MAX(0, port->cfg->bond_updelay); - s.down_delay = MAX(0, port->cfg->bond_downdelay); - s.basis = atoi(get_port_other_config(port->cfg, "bond-hash-basis", "0")); - - s.rebalance_interval = atoi( + s->up_delay = MAX(0, port->cfg->bond_updelay); + s->down_delay = MAX(0, port->cfg->bond_downdelay); ++ s->basis = atoi(get_port_other_config(port->cfg, "bond-hash-basis", "0")); + s->rebalance_interval = atoi( get_port_other_config(port->cfg, "bond-rebalance-interval", "10000")); - if (s.rebalance_interval < 1000) { - s.rebalance_interval = 1000; - } - - s.fake_iface = port->cfg->bond_fake_iface; - - if (!port->bond) { - port->bond = bond_create(&s); - } else { - if (bond_reconfigure(port->bond, &s)) { - bridge_flush(port->bridge); - } - } - - LIST_FOR_EACH (iface, port_elem, &port->ifaces) { - long long stable_id; - - stable_id = atoll(get_interface_other_config(iface->cfg, - "bond-stable-id", "0")); - - if (stable_id <= 0 || stable_id >= UINT32_MAX) { - stable_id = odp_port_to_ofp_port(iface->dp_ifidx); - } - - bond_slave_register(iface->port->bond, iface, stable_id, - iface->netdev); - } -} - -static void -port_send_learning_packets(struct port *port) -{ - struct bridge *br = port->bridge; - int error, n_packets, n_errors; - struct mac_entry *e; - - error = n_packets = n_errors = 0; - LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { - if (e->port.p != port) { - int ret = bond_send_learning_packet(port->bond, e->mac, e->vlan); - if (ret) { - error = ret; - n_errors++; - } - n_packets++; - } + if (s->rebalance_interval < 1000) { + s->rebalance_interval = 1000; } - if (n_errors) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " - "packets, last error was: %s", - port->name, n_errors, n_packets, strerror(error)); - } else { - VLOG_DBG("bond %s: sent %d gratuitous learning packets", - port->name, n_packets); - } + s->fake_iface = port->cfg->bond_fake_iface; } /* Interface functions. */ @@@ -2444,9 -3532,9 +2457,9 @@@ iface_delete_queues(unsigned int queue_ } static void -iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) +iface_configure_qos(struct iface *iface, const struct ovsrec_qos *qos) { - if (!qos || qos->type[0] == '\0') { + if (!qos || qos->type[0] == '\0' || qos->n_queues < 1) { netdev_set_qos(iface->netdev, NULL, NULL); } else { struct iface_delete_queues_cbdata cbdata;