X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fofp-util.c;h=c1e98ba224a1125e1dc5b1c183b0261571557926;hb=df2fa9b50e1c0e93c6be1d31a5d69eb4f077add3;hp=4d89e0ae4af2334206b22642b4794978fd6eeeac;hpb=d31f1109f10e5ffb9bf266306b913ebf23781666;p=openvswitch diff --git a/lib/ofp-util.c b/lib/ofp-util.c index 4d89e0ae..c1e98ba2 100644 --- a/lib/ofp-util.c +++ b/lib/ofp-util.c @@ -18,7 +18,9 @@ #include "ofp-print.h" #include #include +#include #include +#include "autopath.h" #include "byte-order.h" #include "classifier.h" #include "dynamic-string.h" @@ -29,6 +31,7 @@ #include "ofpbuf.h" #include "packets.h" #include "random.h" +#include "unaligned.h" #include "type-props.h" #include "vlog.h" @@ -92,41 +95,32 @@ ofputil_netmask_to_wcbits(ovs_be32 netmask) /* WC_INVARIANTS is the invariant bits (as defined on WC_INVARIANT_LIST) all * OR'd together. */ -enum { - WC_INVARIANTS = 0 +static const flow_wildcards_t WC_INVARIANTS = 0 #define WC_INVARIANT_BIT(NAME) | FWW_##NAME WC_INVARIANT_LIST #undef WC_INVARIANT_BIT -}; +; /* Converts the ofp_match in 'match' into a cls_rule in 'rule', with the given - * 'priority'. - * - * 'flow_format' must either NXFF_OPENFLOW10 or NXFF_TUN_ID_FROM_COOKIE. In - * the latter case only, 'flow''s tun_id field will be taken from the high bits - * of 'cookie', if 'match''s wildcards do not indicate that tun_id is - * wildcarded. */ + * 'priority'. */ void ofputil_cls_rule_from_match(const struct ofp_match *match, - unsigned int priority, - enum nx_flow_format flow_format, - ovs_be64 cookie, struct cls_rule *rule) + unsigned int priority, struct cls_rule *rule) { struct flow_wildcards *wc = &rule->wc; - unsigned int ofpfw; + uint32_t ofpfw; ovs_be16 vid, pcp; /* Initialize rule->priority. */ - ofpfw = ntohl(match->wildcards); - ofpfw &= flow_format == NXFF_TUN_ID_FROM_COOKIE ? OVSFW_ALL : OFPFW_ALL; + ofpfw = ntohl(match->wildcards) & OFPFW_ALL; rule->priority = !ofpfw ? UINT16_MAX : priority; /* Initialize most of rule->wc. */ flow_wildcards_init_catchall(wc); - wc->wildcards = ofpfw & WC_INVARIANTS; + wc->wildcards = (OVS_FORCE flow_wildcards_t) ofpfw & WC_INVARIANTS; /* Wildcard fields that aren't defined by ofp_match or tun_id. */ - wc->wildcards |= (FWW_ARP_SHA | FWW_ARP_THA); + wc->wildcards |= (FWW_ARP_SHA | FWW_ARP_THA | FWW_ND_TARGET); if (ofpfw & OFPFW_NW_TOS) { wc->wildcards |= FWW_NW_TOS; @@ -134,10 +128,6 @@ ofputil_cls_rule_from_match(const struct ofp_match *match, wc->nw_src_mask = ofputil_wcbits_to_netmask(ofpfw >> OFPFW_NW_SRC_SHIFT); wc->nw_dst_mask = ofputil_wcbits_to_netmask(ofpfw >> OFPFW_NW_DST_SHIFT); - if (flow_format == NXFF_TUN_ID_FROM_COOKIE && !(ofpfw & NXFW_TUN_ID)) { - rule->flow.tun_id = htonll(ntohll(cookie) >> 32); - } - if (ofpfw & OFPFW_DL_DST) { /* OpenFlow 1.0 OFPFW_DL_DST covers the whole Ethernet destination, but * Open vSwitch breaks the Ethernet destination into bits as FWW_DL_DST @@ -148,8 +138,7 @@ ofputil_cls_rule_from_match(const struct ofp_match *match, /* Initialize most of rule->flow. */ rule->flow.nw_src = match->nw_src; rule->flow.nw_dst = match->nw_dst; - rule->flow.in_port = (match->in_port == htons(OFPP_LOCAL) ? ODPP_LOCAL - : ntohs(match->in_port)); + rule->flow.in_port = ntohs(match->in_port); rule->flow.dl_type = ofputil_dl_type_from_openflow(match->dl_type); rule->flow.tp_src = match->tp_src; rule->flow.tp_dst = match->tp_dst; @@ -205,47 +194,21 @@ ofputil_cls_rule_from_match(const struct ofp_match *match, cls_rule_zero_wildcarded_fields(rule); } -/* Convert 'rule' into the OpenFlow match structure 'match'. 'flow_format' - * must either NXFF_OPENFLOW10 or NXFF_TUN_ID_FROM_COOKIE. - * - * The NXFF_TUN_ID_FROM_COOKIE flow format requires modifying the flow cookie. - * This function can help with that, if 'cookie_out' is nonnull. For - * NXFF_OPENFLOW10, or if the tunnel ID is wildcarded, 'cookie_in' will be - * copied directly to '*cookie_out'. For NXFF_TUN_ID_FROM_COOKIE when tunnel - * ID is matched, 'cookie_in' will be modified appropriately before setting - * '*cookie_out'. - */ +/* Convert 'rule' into the OpenFlow match structure 'match'. */ void -ofputil_cls_rule_to_match(const struct cls_rule *rule, - enum nx_flow_format flow_format, - struct ofp_match *match, - ovs_be64 cookie_in, ovs_be64 *cookie_out) +ofputil_cls_rule_to_match(const struct cls_rule *rule, struct ofp_match *match) { const struct flow_wildcards *wc = &rule->wc; - unsigned int ofpfw; + uint32_t ofpfw; /* Figure out most OpenFlow wildcards. */ - ofpfw = wc->wildcards & WC_INVARIANTS; + ofpfw = (OVS_FORCE uint32_t) (wc->wildcards & WC_INVARIANTS); ofpfw |= ofputil_netmask_to_wcbits(wc->nw_src_mask) << OFPFW_NW_SRC_SHIFT; ofpfw |= ofputil_netmask_to_wcbits(wc->nw_dst_mask) << OFPFW_NW_DST_SHIFT; if (wc->wildcards & FWW_NW_TOS) { ofpfw |= OFPFW_NW_TOS; } - /* Tunnel ID. */ - if (flow_format == NXFF_TUN_ID_FROM_COOKIE) { - if (wc->tun_id_mask == htonll(0)) { - ofpfw |= NXFW_TUN_ID; - } else { - uint32_t cookie_lo = ntohll(cookie_in); - uint32_t cookie_hi = ntohll(rule->flow.tun_id); - cookie_in = htonll(cookie_lo | ((uint64_t) cookie_hi << 32)); - } - } - if (cookie_out) { - *cookie_out = cookie_in; - } - /* Translate VLANs. */ match->dl_vlan = htons(0); match->dl_vlan_pcp = 0; @@ -270,8 +233,7 @@ ofputil_cls_rule_to_match(const struct cls_rule *rule, /* Compose most of the match structure. */ match->wildcards = htonl(ofpfw); - match->in_port = htons(rule->flow.in_port == ODPP_LOCAL ? OFPP_LOCAL - : rule->flow.in_port); + match->in_port = htons(rule->flow.in_port); memcpy(match->dl_src, rule->flow.dl_src, ETH_ADDR_LEN); memcpy(match->dl_dst, rule->flow.dl_dst, ETH_ADDR_LEN); match->dl_type = ofputil_dl_type_to_openflow(rule->flow.dl_type); @@ -397,19 +359,10 @@ static int ofputil_decode_vendor(const struct ofp_header *oh, const struct ofputil_msg_type **typep) { - static const struct ofputil_msg_type nxt_messages[] = { - { OFPUTIL_NXT_STATUS_REQUEST, - NXT_STATUS_REQUEST, "NXT_STATUS_REQUEST", - sizeof(struct nicira_header), 1 }, - - { OFPUTIL_NXT_STATUS_REPLY, - NXT_STATUS_REPLY, "NXT_STATUS_REPLY", - sizeof(struct nicira_header), 1 }, - - { OFPUTIL_NXT_TUN_ID_FROM_COOKIE, - NXT_TUN_ID_FROM_COOKIE, "NXT_TUN_ID_FROM_COOKIE", - sizeof(struct nxt_tun_id_cookie), 0 }, + 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_ROLE_REQUEST, NXT_ROLE_REQUEST, "NXT_ROLE_REQUEST", sizeof(struct nx_role_request), 0 }, @@ -455,6 +408,21 @@ ofputil_decode_vendor(const struct ofp_header *oh, } nh = (const struct nicira_header *) oh; + + if (nh->subtype == htonl(NXT_FLOW_MOD_TABLE_ID) + && oh->length == htons(sizeof(struct nxt_flow_mod_table_id))) { + /* NXT_SET_FLOW_FORMAT and NXT_FLOW_MOD_TABLE_ID accidentally have the + * same value but different lengths. ofputil_lookup_openflow_message() + * doesn't support this case, so special case it here. */ + static const struct ofputil_msg_type nxt_flow_mod_table_id = + { OFPUTIL_NXT_FLOW_MOD_TABLE_ID, + NXT_FLOW_MOD_TABLE_ID, "NXT_FLOW_MOD_TABLE_ID", + sizeof(struct nxt_flow_mod_table_id), 0 }; + + *typep = &nxt_flow_mod_table_id; + return 0; + } + return ofputil_lookup_openflow_message(&nxt_category, ntohl(nh->subtype), ntohs(oh->length), typep); } @@ -806,7 +774,6 @@ ofputil_flow_format_is_valid(enum nx_flow_format flow_format) { switch (flow_format) { case NXFF_OPENFLOW10: - case NXFF_TUN_ID_FROM_COOKIE: case NXFF_NXM: return true; } @@ -820,8 +787,6 @@ ofputil_flow_format_to_string(enum nx_flow_format flow_format) switch (flow_format) { case NXFF_OPENFLOW10: return "openflow10"; - case NXFF_TUN_ID_FROM_COOKIE: - return "tun_id_from_cookie"; case NXFF_NXM: return "nxm"; default: @@ -833,7 +798,6 @@ int ofputil_flow_format_from_string(const char *s) { return (!strcmp(s, "openflow10") ? NXFF_OPENFLOW10 - : !strcmp(s, "tun_id_from_cookie") ? NXFF_TUN_ID_FROM_COOKIE : !strcmp(s, "nxm") ? NXFF_NXM : -1); } @@ -851,89 +815,43 @@ regs_fully_wildcarded(const struct flow_wildcards *wc) return true; } -static inline bool -is_nxm_required(const struct cls_rule *rule, bool cookie_support, - ovs_be64 cookie) +/* Returns the minimum nx_flow_format to use for sending 'rule' to a switch + * (e.g. to add or remove a flow). Only NXM can handle tunnel IDs, registers, + * or fixing the Ethernet multicast bit. Otherwise, it's better to use + * NXFF_OPENFLOW10 for backward compatibility. */ +enum nx_flow_format +ofputil_min_flow_format(const struct cls_rule *rule) { const struct flow_wildcards *wc = &rule->wc; - ovs_be32 cookie_hi; /* Only NXM supports separately wildcards the Ethernet multicast bit. */ if (!(wc->wildcards & FWW_DL_DST) != !(wc->wildcards & FWW_ETH_MCAST)) { - return true; + return NXFF_NXM; } /* Only NXM supports matching ARP hardware addresses. */ if (!(wc->wildcards & FWW_ARP_SHA) || !(wc->wildcards & FWW_ARP_THA)) { - return true; + return NXFF_NXM; } /* Only NXM supports matching IPv6 traffic. */ if (!(wc->wildcards & FWW_DL_TYPE) && (rule->flow.dl_type == htons(ETH_TYPE_IPV6))) { - return true; + return NXFF_NXM; } /* Only NXM supports matching registers. */ if (!regs_fully_wildcarded(wc)) { - return true; + return NXFF_NXM; } - switch (wc->tun_id_mask) { - case CONSTANT_HTONLL(0): - /* Other formats can fully wildcard tun_id. */ - break; - - case CONSTANT_HTONLL(UINT64_MAX): - /* Only NXM supports matching tunnel ID, unless there is a cookie and - * the top 32 bits of the cookie are the desired tunnel ID value. */ - cookie_hi = htonl(ntohll(cookie) >> 32); - if (!cookie_support - || (cookie_hi && cookie_hi != ntohll(rule->flow.tun_id))) { - return true; - } - break; - - default: - /* Only NXM supports partial matches on tunnel ID. */ - return true; + /* Only NXM supports matching tun_id. */ + if (wc->tun_id_mask != htonll(0)) { + return NXFF_NXM; } /* Other formats can express this rule. */ - return false; -} - -/* Returns the minimum nx_flow_format to use for sending 'rule' to a switch - * (e.g. to add or remove a flow). 'cookie_support' should be true if the - * command to be sent includes a flow cookie (as OFPT_FLOW_MOD does, for - * example) or false if the command does not (OFPST_FLOW and OFPST_AGGREGATE do - * not, for example). If 'cookie_support' is true, then 'cookie' should be the - * cookie to be sent; otherwise its value is ignored. - * - * The "best" flow format is chosen on this basis: - * - * - It must be capable of expressing the rule. NXFF_OPENFLOW10 flows can't - * handle tunnel IDs. NXFF_TUN_ID_FROM_COOKIE flows can't handle registers - * or fixing the Ethernet multicast bit, and can't handle tunnel IDs that - * conflict with the high 32 bits of the cookie or commands that don't - * support cookies. - * - * - Otherwise, the chosen format should be as backward compatible as - * possible. (NXFF_OPENFLOW10 is more backward compatible than - * NXFF_TUN_ID_FROM_COOKIE, which is more backward compatible than - * NXFF_NXM.) - */ -enum nx_flow_format -ofputil_min_flow_format(const struct cls_rule *rule, bool cookie_support, - ovs_be64 cookie) -{ - if (is_nxm_required(rule, cookie_support, cookie)) { - return NXFF_NXM; - } else if (rule->wc.tun_id_mask != htonll(0)) { - return NXFF_TUN_ID_FROM_COOKIE; - } else { - return NXFF_OPENFLOW10; - } + return NXFF_OPENFLOW10; } /* Returns an OpenFlow message that can be used to set the flow format to @@ -941,21 +859,25 @@ ofputil_min_flow_format(const struct cls_rule *rule, bool cookie_support, struct ofpbuf * ofputil_make_set_flow_format(enum nx_flow_format flow_format) { + struct nxt_set_flow_format *sff; struct ofpbuf *msg; - if (flow_format == NXFF_OPENFLOW10 - || flow_format == NXFF_TUN_ID_FROM_COOKIE) { - struct nxt_tun_id_cookie *tic; + sff = make_nxmsg(sizeof *sff, NXT_SET_FLOW_FORMAT, &msg); + sff->format = htonl(flow_format); - tic = make_nxmsg(sizeof *tic, NXT_TUN_ID_FROM_COOKIE, &msg); - tic->set = flow_format == NXFF_TUN_ID_FROM_COOKIE; - } else { - struct nxt_set_flow_format *sff; + return msg; +} - sff = make_nxmsg(sizeof *sff, NXT_SET_FLOW_FORMAT, &msg); - sff->format = htonl(flow_format); - } +/* Returns an OpenFlow message that can be used to turn the flow_mod_table_id + * extension on or off (according to 'flow_mod_table_id'). */ +struct ofpbuf * +ofputil_make_flow_mod_table_id(bool flow_mod_table_id) +{ + struct nxt_flow_mod_table_id *nfmti; + struct ofpbuf *msg; + nfmti = make_nxmsg(sizeof *nfmti, NXT_FLOW_MOD_TABLE_ID, &msg); + nfmti->set = flow_mod_table_id; return msg; } @@ -963,16 +885,16 @@ ofputil_make_set_flow_format(enum nx_flow_format flow_format) * 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, - 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)); @@ -980,8 +902,8 @@ ofputil_decode_flow_mod(struct flow_mod *fm, const struct ofp_header *oh, ofputil_decode_msg_type(oh, &type); if (ofputil_msg_type_code(type) == OFPUTIL_OFPT_FLOW_MOD) { /* Standard OpenFlow flow_mod. */ - struct ofp_match match, orig_match; const struct ofp_flow_mod *ofm; + uint16_t priority; int error; /* Dissect the message. */ @@ -991,29 +913,22 @@ ofputil_decode_flow_mod(struct flow_mod *fm, const struct ofp_header *oh, return error; } - /* Normalize ofm->match. If normalization actually changes anything, - * then log the differences. */ - match = ofm->match; - match.pad1[0] = match.pad2[0] = 0; - orig_match = match; - normalize_match(&match); - if (memcmp(&match, &orig_match, sizeof orig_match)) { - if (!VLOG_DROP_INFO(&bad_ofmsg_rl)) { - char *old = ofp_match_to_literal_string(&orig_match); - char *new = ofp_match_to_literal_string(&match); - VLOG_INFO("normalization changed ofp_match, details:"); - VLOG_INFO(" pre: %s", old); - VLOG_INFO("post: %s", new); - free(old); - free(new); - } + /* Set priority based on original wildcards. Normally we'd allow + * ofputil_cls_rule_from_match() to do this for us, but + * ofputil_normalize_rule() can put wildcards where the original flow + * didn't have them. */ + priority = ntohs(ofm->priority); + if (!(ofm->match.wildcards & htonl(OFPFW_ALL))) { + priority = UINT16_MAX; } + /* Translate the rule. */ + ofputil_cls_rule_from_match(&ofm->match, priority, &fm->cr); + ofputil_normalize_rule(&fm->cr, NXFF_OPENFLOW10); + /* Translate the message. */ - ofputil_cls_rule_from_match(&match, ntohs(ofm->priority), flow_format, - ofm->cookie, &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); @@ -1038,7 +953,7 @@ ofputil_decode_flow_mod(struct flow_mod *fm, const struct ofp_header *oh, /* Translate the message. */ fm->cookie = nfm->cookie; - fm->command = ntohs(nfm->command); + command = ntohs(nfm->command); fm->idle_timeout = ntohs(nfm->idle_timeout); fm->hard_timeout = ntohs(nfm->hard_timeout); fm->buffer_id = ntohl(nfm->buffer_id); @@ -1048,26 +963,42 @@ ofputil_decode_flow_mod(struct flow_mod *fm, const struct ofp_header *oh, NOT_REACHED(); } + if (flow_mod_table_id) { + fm->command = command & 0xff; + fm->table_id = command >> 8; + } else { + fm->command = command; + fm->table_id = 0xff; + } + return 0; } /* Converts 'fm' into an OFPT_FLOW_MOD or NXT_FLOW_MOD message according to - * 'flow_format' and returns the message. */ + * 'flow_format' and returns the message. + * + * 'flow_mod_table_id' should be true if the NXT_FLOW_MOD_TABLE_ID extension is + * enabled, false otherwise. */ struct ofpbuf * ofputil_encode_flow_mod(const struct flow_mod *fm, - enum nx_flow_format flow_format) + enum nx_flow_format flow_format, + bool flow_mod_table_id) { 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); + 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); @@ -1085,7 +1016,7 @@ ofputil_encode_flow_mod(const struct flow_mod *fm, nfm = msg->data; nfm->cookie = fm->cookie; - nfm->command = htons(fm->command); + nfm->command = htons(command); nfm->idle_timeout = htons(fm->idle_timeout); nfm->hard_timeout = htons(fm->hard_timeout); nfm->priority = htons(fm->cr.priority); @@ -1105,13 +1036,12 @@ ofputil_encode_flow_mod(const struct flow_mod *fm, static int ofputil_decode_ofpst_flow_request(struct flow_stats_request *fsr, const struct ofp_header *oh, - enum nx_flow_format flow_format, bool aggregate) { const struct ofp_flow_stats_request *ofsr = ofputil_stats_body(oh); fsr->aggregate = aggregate; - ofputil_cls_rule_from_match(&ofsr->match, 0, flow_format, 0, &fsr->match); + ofputil_cls_rule_from_match(&ofsr->match, 0, &fsr->match); fsr->out_port = ntohs(ofsr->out_port); fsr->table_id = ofsr->table_id; @@ -1146,17 +1076,11 @@ ofputil_decode_nxst_flow_request(struct flow_stats_request *fsr, } /* Converts an OFPST_FLOW, OFPST_AGGREGATE, NXST_FLOW, or NXST_AGGREGATE - * message 'oh', received when the current flow format was 'flow_format', into - * an abstract flow_stats_request in 'fsr'. Returns 0 if successful, otherwise - * an OpenFlow error code. - * - * For OFPST_FLOW and OFPST_AGGREGATE messages, 'flow_format' should be the - * current flow format at the time when the message was received. Otherwise - * 'flow_format' is ignored. */ + * request 'oh', into an abstract flow_stats_request in 'fsr'. Returns 0 if + * successful, otherwise an OpenFlow error code. */ int ofputil_decode_flow_stats_request(struct flow_stats_request *fsr, - const struct ofp_header *oh, - enum nx_flow_format flow_format) + const struct ofp_header *oh) { const struct ofputil_msg_type *type; struct ofpbuf b; @@ -1168,10 +1092,10 @@ ofputil_decode_flow_stats_request(struct flow_stats_request *fsr, code = ofputil_msg_type_code(type); switch (code) { case OFPUTIL_OFPST_FLOW_REQUEST: - return ofputil_decode_ofpst_flow_request(fsr, oh, flow_format, false); + return ofputil_decode_ofpst_flow_request(fsr, oh, false); case OFPUTIL_OFPST_AGGREGATE_REQUEST: - return ofputil_decode_ofpst_flow_request(fsr, oh, flow_format, true); + return ofputil_decode_ofpst_flow_request(fsr, oh, true); case OFPUTIL_NXST_FLOW_REQUEST: return ofputil_decode_nxst_flow_request(fsr, oh, false); @@ -1186,7 +1110,7 @@ ofputil_decode_flow_stats_request(struct flow_stats_request *fsr, } /* Converts abstract flow_stats_request 'fsr' into an OFPST_FLOW, - * OFPST_AGGREGATE, NXST_FLOW, or NXST_AGGREGATE message 'oh' according to + * OFPST_AGGREGATE, NXST_FLOW, or NXST_AGGREGATE request 'oh' according to * 'flow_format', and returns the message. */ struct ofpbuf * ofputil_encode_flow_stats_request(const struct flow_stats_request *fsr, @@ -1194,8 +1118,7 @@ ofputil_encode_flow_stats_request(const struct flow_stats_request *fsr, { struct ofpbuf *msg; - if (flow_format == NXFF_OPENFLOW10 - || flow_format == NXFF_TUN_ID_FROM_COOKIE) { + if (flow_format == NXFF_OPENFLOW10) { struct ofp_flow_stats_request *ofsr; int type; @@ -1204,8 +1127,7 @@ ofputil_encode_flow_stats_request(const struct flow_stats_request *fsr, type = fsr->aggregate ? OFPST_AGGREGATE : OFPST_FLOW; ofsr = ofputil_make_stats_request(sizeof *ofsr, type, &msg); - ofputil_cls_rule_to_match(&fsr->match, flow_format, &ofsr->match, - 0, NULL); + ofputil_cls_rule_to_match(&fsr->match, &ofsr->match); ofsr->table_id = fsr->table_id; ofsr->out_port = htons(fsr->out_port); } else if (flow_format == NXFF_NXM) { @@ -1228,18 +1150,120 @@ ofputil_encode_flow_stats_request(const struct flow_stats_request *fsr, return msg; } -/* Converts an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message 'oh', received - * when the current flow format was 'flow_format', into an abstract - * ofputil_flow_removed in 'fr'. Returns 0 if successful, otherwise an - * OpenFlow error code. +/* Converts an OFPST_FLOW or NXST_FLOW reply in 'msg' into an abstract + * ofputil_flow_stats in 'fs'. + * + * Multiple OFPST_FLOW or NXST_FLOW replies can be packed into a single + * OpenFlow message. Calling this function multiple times for a single 'msg' + * iterates through the replies. The caller must initially leave 'msg''s layer + * pointers null and not modify them between calls. * - * For OFPT_FLOW_REMOVED messages, 'flow_format' should be the current flow - * format at the time when the message was received. Otherwise 'flow_format' - * is ignored. */ + * Returns 0 if successful, EOF if no replies were left in this 'msg', + * otherwise a positive errno value. */ +int +ofputil_decode_flow_stats_reply(struct ofputil_flow_stats *fs, + struct ofpbuf *msg) +{ + const struct ofputil_msg_type *type; + int code; + + ofputil_decode_msg_type(msg->l2 ? msg->l2 : msg->data, &type); + code = ofputil_msg_type_code(type); + if (!msg->l2) { + msg->l2 = msg->data; + if (code == OFPUTIL_OFPST_FLOW_REPLY) { + ofpbuf_pull(msg, sizeof(struct ofp_stats_reply)); + } else if (code == OFPUTIL_NXST_FLOW_REPLY) { + ofpbuf_pull(msg, sizeof(struct nicira_stats_msg)); + } else { + NOT_REACHED(); + } + } + + if (!msg->size) { + return EOF; + } else if (code == OFPUTIL_OFPST_FLOW_REPLY) { + const struct ofp_flow_stats *ofs; + size_t length; + + ofs = ofpbuf_try_pull(msg, sizeof *ofs); + if (!ofs) { + VLOG_WARN_RL(&bad_ofmsg_rl, "OFPST_FLOW reply has %zu leftover " + "bytes at end", msg->size); + return EINVAL; + } + + length = ntohs(ofs->length); + if (length < sizeof *ofs) { + VLOG_WARN_RL(&bad_ofmsg_rl, "OFPST_FLOW reply claims invalid " + "length %zu", length); + return EINVAL; + } + + if (ofputil_pull_actions(msg, length - sizeof *ofs, + &fs->actions, &fs->n_actions)) { + return EINVAL; + } + + fs->cookie = get_32aligned_be64(&ofs->cookie); + ofputil_cls_rule_from_match(&ofs->match, ntohs(ofs->priority), + &fs->rule); + fs->table_id = ofs->table_id; + fs->duration_sec = ntohl(ofs->duration_sec); + fs->duration_nsec = ntohl(ofs->duration_nsec); + fs->idle_timeout = ntohs(ofs->idle_timeout); + fs->hard_timeout = ntohs(ofs->hard_timeout); + fs->packet_count = ntohll(get_32aligned_be64(&ofs->packet_count)); + fs->byte_count = ntohll(get_32aligned_be64(&ofs->byte_count)); + } else if (code == OFPUTIL_NXST_FLOW_REPLY) { + const struct nx_flow_stats *nfs; + size_t match_len, length; + + nfs = ofpbuf_try_pull(msg, sizeof *nfs); + if (!nfs) { + VLOG_WARN_RL(&bad_ofmsg_rl, "NXST_FLOW reply has %zu leftover " + "bytes at end", msg->size); + return EINVAL; + } + + length = ntohs(nfs->length); + match_len = ntohs(nfs->match_len); + if (length < sizeof *nfs + ROUND_UP(match_len, 8)) { + VLOG_WARN_RL(&bad_ofmsg_rl, "NXST_FLOW reply with match_len=%zu " + "claims invalid length %zu", match_len, length); + return EINVAL; + } + if (nx_pull_match(msg, match_len, ntohs(nfs->priority), &fs->rule)) { + return EINVAL; + } + + if (ofputil_pull_actions(msg, + length - sizeof *nfs - ROUND_UP(match_len, 8), + &fs->actions, &fs->n_actions)) { + return EINVAL; + } + + fs->cookie = nfs->cookie; + fs->table_id = nfs->table_id; + fs->duration_sec = ntohl(nfs->duration_sec); + fs->duration_nsec = ntohl(nfs->duration_nsec); + fs->idle_timeout = ntohs(nfs->idle_timeout); + fs->hard_timeout = ntohs(nfs->hard_timeout); + fs->packet_count = ntohll(nfs->packet_count); + fs->byte_count = ntohll(nfs->byte_count); + } else { + NOT_REACHED(); + } + + return 0; +} + +/* Converts an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message 'oh' into an + * abstract ofputil_flow_removed in 'fr'. Returns 0 if successful, otherwise + * an OpenFlow error code. */ int ofputil_decode_flow_removed(struct ofputil_flow_removed *fr, - const struct ofp_header *oh, - enum nx_flow_format flow_format) + const struct ofp_header *oh) { const struct ofputil_msg_type *type; enum ofputil_msg_code code; @@ -1251,7 +1275,7 @@ ofputil_decode_flow_removed(struct ofputil_flow_removed *fr, ofr = (const struct ofp_flow_removed *) oh; ofputil_cls_rule_from_match(&ofr->match, ntohs(ofr->priority), - flow_format, ofr->cookie, &fr->rule); + &fr->rule); fr->cookie = ofr->cookie; fr->reason = ofr->reason; fr->duration_sec = ntohl(ofr->duration_sec); @@ -1290,6 +1314,96 @@ ofputil_decode_flow_removed(struct ofputil_flow_removed *fr, return 0; } +/* Converts abstract ofputil_flow_removed 'fr' into an OFPT_FLOW_REMOVED or + * NXT_FLOW_REMOVED message 'oh' according to 'flow_format', and returns the + * message. */ +struct ofpbuf * +ofputil_encode_flow_removed(const struct ofputil_flow_removed *fr, + enum nx_flow_format flow_format) +{ + struct ofpbuf *msg; + + if (flow_format == NXFF_OPENFLOW10) { + struct ofp_flow_removed *ofr; + + ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), + &msg); + ofputil_cls_rule_to_match(&fr->rule, &ofr->match); + ofr->cookie = fr->cookie; + ofr->priority = htons(fr->rule.priority); + ofr->reason = fr->reason; + ofr->duration_sec = htonl(fr->duration_sec); + ofr->duration_nsec = htonl(fr->duration_nsec); + ofr->idle_timeout = htons(fr->idle_timeout); + ofr->packet_count = htonll(fr->packet_count); + ofr->byte_count = htonll(fr->byte_count); + } else if (flow_format == NXFF_NXM) { + struct nx_flow_removed *nfr; + int match_len; + + make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &msg); + match_len = nx_put_match(msg, &fr->rule); + + nfr = msg->data; + nfr->cookie = fr->cookie; + nfr->priority = htons(fr->rule.priority); + nfr->reason = fr->reason; + nfr->duration_sec = htonl(fr->duration_sec); + nfr->duration_nsec = htonl(fr->duration_nsec); + nfr->idle_timeout = htons(fr->idle_timeout); + nfr->match_len = htons(match_len); + nfr->packet_count = htonll(fr->packet_count); + nfr->byte_count = htonll(fr->byte_count); + } else { + NOT_REACHED(); + } + + return msg; +} + +/* Converts abstract ofputil_packet_in 'pin' into an OFPT_PACKET_IN message + * and returns the message. + * + * If 'rw_packet' is NULL, the caller takes ownership of the newly allocated + * returned ofpbuf. + * + * If 'rw_packet' is nonnull, then it must contain the same data as + * pin->packet. 'rw_packet' is allowed to be the same ofpbuf as pin->packet. + * It is modified in-place into an OFPT_PACKET_IN message according to 'pin', + * and then ofputil_encode_packet_in() returns 'rw_packet'. If 'rw_packet' has + * enough headroom to insert a "struct ofp_packet_in", this is more efficient + * than ofputil_encode_packet_in() because it does not copy the packet + * payload. */ +struct ofpbuf * +ofputil_encode_packet_in(const struct ofputil_packet_in *pin, + struct ofpbuf *rw_packet) +{ + int total_len = pin->packet->size; + struct ofp_packet_in *opi; + + if (rw_packet) { + if (pin->send_len < rw_packet->size) { + rw_packet->size = pin->send_len; + } + } else { + rw_packet = ofpbuf_clone_data_with_headroom( + pin->packet->data, MIN(pin->send_len, pin->packet->size), + offsetof(struct ofp_packet_in, data)); + } + + /* Add OFPT_PACKET_IN. */ + opi = ofpbuf_push_zeros(rw_packet, offsetof(struct ofp_packet_in, data)); + opi->header.version = OFP_VERSION; + opi->header.type = OFPT_PACKET_IN; + opi->total_len = htons(total_len); + opi->in_port = htons(pin->in_port); + opi->reason = pin->reason; + opi->buffer_id = htonl(pin->buffer_id); + update_openflow_length(rw_packet); + + return rw_packet; +} + /* Returns a string representing the message type of 'type'. The string is the * enumeration constant for the type, e.g. "OFPT_HELLO". For statistics * messages, the constant is followed by "request" or "reply", @@ -1510,7 +1624,7 @@ make_flow_mod(uint16_t command, const struct cls_rule *rule, ofm->header.length = htons(size); ofm->cookie = 0; ofm->priority = htons(MIN(rule->priority, UINT16_MAX)); - ofputil_cls_rule_to_match(rule, NXFF_OPENFLOW10, &ofm->match, 0, NULL); + ofputil_cls_rule_to_match(rule, &ofm->match); ofm->command = htons(command); return out; } @@ -1657,51 +1771,6 @@ make_echo_reply(const struct ofp_header *rq) return out; } -const struct ofp_flow_stats * -flow_stats_first(struct flow_stats_iterator *iter, - const struct ofp_stats_reply *osr) -{ - iter->pos = osr->body; - iter->end = osr->body + (ntohs(osr->header.length) - - offsetof(struct ofp_stats_reply, body)); - return flow_stats_next(iter); -} - -const struct ofp_flow_stats * -flow_stats_next(struct flow_stats_iterator *iter) -{ - ptrdiff_t bytes_left = iter->end - iter->pos; - const struct ofp_flow_stats *fs; - size_t length; - - if (bytes_left < sizeof *fs) { - if (bytes_left != 0) { - VLOG_WARN_RL(&bad_ofmsg_rl, - "%td leftover bytes in flow stats reply", bytes_left); - } - return NULL; - } - - fs = (const void *) iter->pos; - length = ntohs(fs->length); - if (length < sizeof *fs) { - VLOG_WARN_RL(&bad_ofmsg_rl, "flow stats length %zu is shorter than " - "min %zu", length, sizeof *fs); - return NULL; - } else if (length > bytes_left) { - VLOG_WARN_RL(&bad_ofmsg_rl, "flow stats length %zu but only %td " - "bytes left", length, bytes_left); - return NULL; - } else if ((length - sizeof *fs) % sizeof fs->actions[0]) { - VLOG_WARN_RL(&bad_ofmsg_rl, "flow stats length %zu has %zu bytes " - "left over in final action", length, - (length - sizeof *fs) % sizeof fs->actions[0]); - return NULL; - } - iter->pos += length; - return fs; -} - static int check_action_exact_len(const union ofp_action *a, unsigned int len, unsigned int required_len) @@ -1784,7 +1853,7 @@ check_nicira_action(const union ofp_action *a, unsigned int len, const struct flow *flow) { const struct nx_action_header *nah; - uint16_t subtype; + int subtype; int error; if (len < 16) { @@ -1796,7 +1865,7 @@ check_nicira_action(const union ofp_action *a, unsigned int len, subtype = ntohs(nah->subtype); if (subtype > TYPE_MAXIMUM(enum nx_action_subtype)) { - /* This is necessary because enum nx_action_subtype is probably an + /* This is necessary because enum nx_action_subtype may be an * 8-bit type, so the cast below throws away the top 8 bits. */ return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_VENDOR_TYPE); } @@ -1840,11 +1909,18 @@ check_nicira_action(const union ofp_action *a, unsigned int len, } return multipath_check((const struct nx_action_multipath *) a); + case NXAST_AUTOPATH: + error = check_nx_action_exact_len( + nah, len, sizeof(struct nx_action_autopath)); + if (error) { + return error; + } + return autopath_check((const struct nx_action_autopath *) a); + case NXAST_SNAT__OBSOLETE: default: VLOG_WARN_RL(&bad_ofmsg_rl, - "unknown Nicira vendor action subtype %"PRIu16, - ntohs(nah->subtype)); + "unknown Nicira vendor action subtype %d", subtype); return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_VENDOR_TYPE); } } @@ -1879,7 +1955,7 @@ check_action(const union ofp_action *a, unsigned int len, if (error) { return error; } - if (a->vlan_vid.vlan_vid & ~7) { + if (a->vlan_pcp.vlan_pcp & ~7) { return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_BAD_ARGUMENT); } return 0; @@ -1946,10 +2022,9 @@ validate_actions(const union ofp_action *actions, size_t n_actions, return 0; } -/* Returns true if 'action' outputs to 'port' (which must be in network byte - * order), false otherwise. */ +/* Returns true if 'action' outputs to 'port', false otherwise. */ bool -action_outputs_to_port(const union ofp_action *action, uint16_t port) +action_outputs_to_port(const union ofp_action *action, ovs_be16 port) { switch (ntohs(action->type)) { case OFPAT_OUTPUT: @@ -1985,123 +2060,115 @@ actions_next(struct actions_iterator *iter) } } +/* "Normalizes" the wildcards in 'rule'. That means: + * + * 1. If the type of level N is known, then only the valid fields for that + * level may be specified. For example, ARP does not have a TOS field, + * so nw_tos must be wildcarded if 'rule' specifies an ARP flow. + * Similarly, IPv4 does not have any IPv6 addresses, so ipv6_src and + * ipv6_dst (and other fields) must be wildcarded if 'rule' specifies an + * IPv4 flow. + * + * 2. If the type of level N is not known (or not understood by Open + * vSwitch), then no fields at all for that level may be specified. For + * example, Open vSwitch does not understand SCTP, an L4 protocol, so the + * L4 fields tp_src and tp_dst must be wildcarded if 'rule' specifies an + * SCTP flow. + * + * 'flow_format' specifies the format of the flow as received or as intended to + * be sent. This is important for IPv6 and ARP, for which NXM supports more + * detailed matching. */ void -normalize_match(struct ofp_match *m) -{ - enum { OFPFW_NW = (OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK | OFPFW_NW_PROTO - | OFPFW_NW_TOS) }; - enum { OFPFW_TP = OFPFW_TP_SRC | OFPFW_TP_DST }; - uint32_t wc; - - wc = ntohl(m->wildcards) & OVSFW_ALL; - if (wc & OFPFW_DL_TYPE) { - m->dl_type = 0; - - /* Can't sensibly match on network or transport headers if the - * data link type is unknown. */ - wc |= OFPFW_NW | OFPFW_TP; - m->nw_src = m->nw_dst = m->nw_proto = m->nw_tos = 0; - m->tp_src = m->tp_dst = 0; - } else if (m->dl_type == htons(ETH_TYPE_IP)) { - if (wc & OFPFW_NW_PROTO) { - m->nw_proto = 0; - - /* Can't sensibly match on transport headers if the network - * protocol is unknown. */ - wc |= OFPFW_TP; - m->tp_src = m->tp_dst = 0; - } else if (m->nw_proto == IPPROTO_TCP || - m->nw_proto == IPPROTO_UDP || - m->nw_proto == IPPROTO_ICMP) { - if (wc & OFPFW_TP_SRC) { - m->tp_src = 0; - } - if (wc & OFPFW_TP_DST) { - m->tp_dst = 0; - } - } else { - /* Transport layer fields will always be extracted as zeros, so we - * can do an exact-match on those values. */ - wc &= ~OFPFW_TP; - m->tp_src = m->tp_dst = 0; - } - if (wc & OFPFW_NW_SRC_MASK) { - m->nw_src &= ofputil_wcbits_to_netmask(wc >> OFPFW_NW_SRC_SHIFT); - } - if (wc & OFPFW_NW_DST_MASK) { - m->nw_dst &= ofputil_wcbits_to_netmask(wc >> OFPFW_NW_DST_SHIFT); - } - if (wc & OFPFW_NW_TOS) { - m->nw_tos = 0; - } else { - m->nw_tos &= IP_DSCP_MASK; +ofputil_normalize_rule(struct cls_rule *rule, enum nx_flow_format flow_format) +{ + enum { + MAY_NW_ADDR = 1 << 0, /* nw_src, nw_dst */ + MAY_TP_ADDR = 1 << 1, /* tp_src, tp_dst */ + MAY_NW_PROTO = 1 << 2, /* nw_proto */ + MAY_NW_TOS = 1 << 3, /* nw_tos */ + MAY_ARP_SHA = 1 << 4, /* arp_sha */ + MAY_ARP_THA = 1 << 5, /* arp_tha */ + MAY_IPV6_ADDR = 1 << 6, /* ipv6_src, ipv6_dst */ + MAY_ND_TARGET = 1 << 7 /* nd_target */ + } may_match; + + struct flow_wildcards wc; + + /* Figure out what fields may be matched. */ + if (rule->flow.dl_type == htons(ETH_TYPE_IP)) { + may_match = MAY_NW_PROTO | MAY_NW_TOS | MAY_NW_ADDR; + if (rule->flow.nw_proto == IPPROTO_TCP || + rule->flow.nw_proto == IPPROTO_UDP || + rule->flow.nw_proto == IPPROTO_ICMP) { + may_match |= MAY_TP_ADDR; } - } else if (m->dl_type == htons(ETH_TYPE_ARP)) { - if (wc & OFPFW_NW_PROTO) { - m->nw_proto = 0; - } - if (wc & OFPFW_NW_SRC_MASK) { - m->nw_src &= ofputil_wcbits_to_netmask(wc >> OFPFW_NW_SRC_SHIFT); + } else if (rule->flow.dl_type == htons(ETH_TYPE_IPV6) + && flow_format == NXFF_NXM) { + may_match = MAY_NW_PROTO | MAY_NW_TOS | MAY_IPV6_ADDR; + if (rule->flow.nw_proto == IPPROTO_TCP || + rule->flow.nw_proto == IPPROTO_UDP) { + may_match |= MAY_TP_ADDR; + } else if (rule->flow.nw_proto == IPPROTO_ICMPV6) { + may_match |= MAY_TP_ADDR; + if (rule->flow.tp_src == htons(ND_NEIGHBOR_SOLICIT)) { + may_match |= MAY_ND_TARGET | MAY_ARP_SHA; + } else if (rule->flow.tp_src == htons(ND_NEIGHBOR_ADVERT)) { + may_match |= MAY_ND_TARGET | MAY_ARP_THA; + } } - if (wc & OFPFW_NW_DST_MASK) { - m->nw_dst &= ofputil_wcbits_to_netmask(wc >> OFPFW_NW_DST_SHIFT); + } else if (rule->flow.dl_type == htons(ETH_TYPE_ARP)) { + may_match = MAY_NW_PROTO | MAY_NW_ADDR; + if (flow_format == NXFF_NXM) { + may_match |= MAY_ARP_SHA | MAY_ARP_THA; } - m->tp_src = m->tp_dst = m->nw_tos = 0; - } else if (m->dl_type == htons(ETH_TYPE_IPV6)) { - /* Don't normalize IPv6 traffic, since OpenFlow doesn't have a - * way to express it. */ } else { - /* Network and transport layer fields will always be extracted as - * zeros, so we can do an exact-match on those values. */ - wc &= ~(OFPFW_NW | OFPFW_TP); - m->nw_proto = m->nw_src = m->nw_dst = m->nw_tos = 0; - m->tp_src = m->tp_dst = 0; + may_match = 0; } - if (wc & OFPFW_DL_SRC) { - memset(m->dl_src, 0, sizeof m->dl_src); + + /* Clear the fields that may not be matched. */ + wc = rule->wc; + if (!(may_match & MAY_NW_ADDR)) { + wc.nw_src_mask = wc.nw_dst_mask = htonl(0); } - if (wc & OFPFW_DL_DST) { - memset(m->dl_dst, 0, sizeof m->dl_dst); + if (!(may_match & MAY_TP_ADDR)) { + wc.wildcards |= FWW_TP_SRC | FWW_TP_DST; + } + if (!(may_match & MAY_NW_PROTO)) { + wc.wildcards |= FWW_NW_PROTO; + } + if (!(may_match & MAY_NW_TOS)) { + wc.wildcards |= FWW_NW_TOS; + } + if (!(may_match & MAY_ARP_SHA)) { + wc.wildcards |= FWW_ARP_SHA; + } + if (!(may_match & MAY_ARP_THA)) { + wc.wildcards |= FWW_ARP_THA; + } + if (!(may_match & MAY_IPV6_ADDR)) { + wc.ipv6_src_mask = wc.ipv6_dst_mask = in6addr_any; + } + if (!(may_match & MAY_ND_TARGET)) { + wc.wildcards |= FWW_ND_TARGET; } - m->wildcards = htonl(wc); -} -/* Returns a string that describes 'match' in a very literal way, without - * interpreting its contents except in a very basic fashion. The returned - * string is intended to be fixed-length, so that it is easy to see differences - * between two such strings if one is put above another. This is useful for - * describing changes made by normalize_match(). - * - * The caller must free the returned string (with free()). */ -char * -ofp_match_to_literal_string(const struct ofp_match *match) -{ - return xasprintf("wildcards=%#10"PRIx32" " - " in_port=%5"PRId16" " - " dl_src="ETH_ADDR_FMT" " - " dl_dst="ETH_ADDR_FMT" " - " dl_vlan=%5"PRId16" " - " dl_vlan_pcp=%3"PRId8" " - " dl_type=%#6"PRIx16" " - " nw_tos=%#4"PRIx8" " - " nw_proto=%#4"PRIx16" " - " nw_src=%#10"PRIx32" " - " nw_dst=%#10"PRIx32" " - " tp_src=%5"PRId16" " - " tp_dst=%5"PRId16, - ntohl(match->wildcards), - ntohs(match->in_port), - ETH_ADDR_ARGS(match->dl_src), - ETH_ADDR_ARGS(match->dl_dst), - ntohs(match->dl_vlan), - match->dl_vlan_pcp, - ntohs(match->dl_type), - match->nw_tos, - match->nw_proto, - ntohl(match->nw_src), - ntohl(match->nw_dst), - ntohs(match->tp_src), - ntohs(match->tp_dst)); + /* Log any changes. */ + if (!flow_wildcards_equal(&wc, &rule->wc)) { + bool log = !VLOG_DROP_INFO(&bad_ofmsg_rl); + char *pre = log ? cls_rule_to_string(rule) : NULL; + + rule->wc = wc; + cls_rule_zero_wildcarded_fields(rule); + + if (log) { + char *post = cls_rule_to_string(rule); + VLOG_INFO("normalization changed ofp_match, details:"); + VLOG_INFO(" pre: %s", pre); + VLOG_INFO("post: %s", post); + free(pre); + free(post); + } + } } static uint32_t