ofp-util: Enforce OpenFlow 1.1+ table_id requirements in flow_mod messages.
[sliver-openvswitch.git] / lib / ofp-util.c
index d435e99..f0cd751 100644 (file)
@@ -1504,8 +1504,8 @@ ofputil_decode_flow_mod(struct ofputil_flow_mod *fm,
             return error;
         }
 
-        error = ofpacts_pull_openflow11_instructions(&b, oh->version,
-                                                     b.size, ofpacts);
+        error = ofpacts_pull_openflow_instructions(&b, b.size, oh->version,
+                                                   ofpacts);
         if (error) {
             return error;
         }
@@ -1530,7 +1530,19 @@ ofputil_decode_flow_mod(struct ofputil_flow_mod *fm,
         }
         fm->modify_cookie = false;
         fm->command = ofm->command;
+
+        /* Get table ID.
+         *
+         * OF1.1 entirely forbids table_id == 255.
+         * OF1.2+ allows table_id == 255 only for deletes. */
         fm->table_id = ofm->table_id;
+        if (fm->table_id == 255
+            && (oh->version == OFP11_VERSION
+                || (ofm->command != OFPFC_DELETE &&
+                    ofm->command != OFPFC_DELETE_STRICT))) {
+            return OFPERR_OFPFMFC_BAD_TABLE_ID;
+        }
+
         fm->idle_timeout = ntohs(ofm->idle_timeout);
         fm->hard_timeout = ntohs(ofm->hard_timeout);
         fm->buffer_id = ntohl(ofm->buffer_id);
@@ -1561,7 +1573,8 @@ ofputil_decode_flow_mod(struct ofputil_flow_mod *fm,
             ofputil_normalize_match(&fm->match);
 
             /* Now get the actions. */
-            error = ofpacts_pull_openflow10(&b, b.size, ofpacts);
+            error = ofpacts_pull_openflow_actions(&b, b.size, oh->version,
+                                                  ofpacts);
             if (error) {
                 return error;
             }
@@ -1594,7 +1607,8 @@ ofputil_decode_flow_mod(struct ofputil_flow_mod *fm,
             if (error) {
                 return error;
             }
-            error = ofpacts_pull_openflow10(&b, b.size, ofpacts);
+            error = ofpacts_pull_openflow_actions(&b, b.size, oh->version,
+                                                  ofpacts);
             if (error) {
                 return error;
             }
@@ -2056,7 +2070,14 @@ ofputil_encode_flow_mod(const struct ofputil_flow_mod *fm,
             ofm->cookie = fm->cookie;
         }
         ofm->cookie_mask = fm->cookie_mask;
-        ofm->table_id = fm->table_id;
+        if (fm->table_id != 255
+            || (protocol != OFPUTIL_P_OF11_STD
+                && (fm->command == OFPFC_DELETE ||
+                    fm->command == OFPFC_DELETE_STRICT))) {
+            ofm->table_id = fm->table_id;
+        } else {
+            ofm->table_id = 0;
+        }
         ofm->command = fm->command;
         ofm->idle_timeout = htons(fm->idle_timeout);
         ofm->hard_timeout = htons(fm->hard_timeout);
@@ -2066,7 +2087,8 @@ ofputil_encode_flow_mod(const struct ofputil_flow_mod *fm,
         ofm->out_group = htonl(fm->out_group);
         ofm->flags = raw_flags;
         ofputil_put_ofp11_match(msg, &fm->match, protocol);
-        ofpacts_put_openflow11_instructions(fm->ofpacts, fm->ofpacts_len, msg);
+        ofpacts_put_openflow_instructions(fm->ofpacts, fm->ofpacts_len, msg,
+                                          version);
         break;
     }
 
@@ -2086,7 +2108,8 @@ ofputil_encode_flow_mod(const struct ofputil_flow_mod *fm,
         ofm->buffer_id = htonl(fm->buffer_id);
         ofm->out_port = htons(ofp_to_u16(fm->out_port));
         ofm->flags = raw_flags;
-        ofpacts_put_openflow10(fm->ofpacts, fm->ofpacts_len, msg);
+        ofpacts_put_openflow_actions(fm->ofpacts, fm->ofpacts_len, msg,
+                                     version);
         break;
     }
 
@@ -2109,7 +2132,8 @@ ofputil_encode_flow_mod(const struct ofputil_flow_mod *fm,
         nfm->out_port = htons(ofp_to_u16(fm->out_port));
         nfm->flags = raw_flags;
         nfm->match_len = htons(match_len);
-        ofpacts_put_openflow10(fm->ofpacts, fm->ofpacts_len, msg);
+        ofpacts_put_openflow_actions(fm->ofpacts, fm->ofpacts_len, msg,
+                                     version);
         break;
     }
 
@@ -2638,9 +2662,9 @@ ofputil_decode_flow_stats_reply(struct ofputil_flow_stats *fs,
             return EINVAL;
         }
 
-        if (ofpacts_pull_openflow11_instructions(msg, oh->version,
-                                                 length - sizeof *ofs -
-                                                 padded_match_len, ofpacts)) {
+        if (ofpacts_pull_openflow_instructions(msg, length - sizeof *ofs -
+                                               padded_match_len, oh->version,
+                                               ofpacts)) {
             VLOG_WARN_RL(&bad_ofmsg_rl, "OFPST_FLOW reply bad instructions");
             return EINVAL;
         }
@@ -2683,7 +2707,8 @@ ofputil_decode_flow_stats_reply(struct ofputil_flow_stats *fs,
             return EINVAL;
         }
 
-        if (ofpacts_pull_openflow10(msg, length - sizeof *ofs, ofpacts)) {
+        if (ofpacts_pull_openflow_actions(msg, length - sizeof *ofs,
+                                          oh->version, ofpacts)) {
             return EINVAL;
         }
 
@@ -2723,7 +2748,8 @@ ofputil_decode_flow_stats_reply(struct ofputil_flow_stats *fs,
         }
 
         actions_len = length - sizeof *nfs - ROUND_UP(match_len, 8);
-        if (ofpacts_pull_openflow10(msg, actions_len, ofpacts)) {
+        if (ofpacts_pull_openflow_actions(msg, actions_len, oh->version,
+                                          ofpacts)) {
             return EINVAL;
         }
 
@@ -2777,16 +2803,16 @@ ofputil_append_flow_stats_reply(const struct ofputil_flow_stats *fs,
     struct ofpbuf *reply = ofpbuf_from_list(list_back(replies));
     size_t start_ofs = reply->size;
     enum ofpraw raw;
+    enum ofp_version version = ((struct ofp_header *)reply->data)->version;
 
     ofpraw_decode_partial(&raw, reply->data, reply->size);
     if (raw == OFPRAW_OFPST11_FLOW_REPLY || raw == OFPRAW_OFPST13_FLOW_REPLY) {
-        const struct ofp_header *oh = reply->data;
         struct ofp11_flow_stats *ofs;
 
         ofpbuf_put_uninit(reply, sizeof *ofs);
         oxm_put_match(reply, &fs->match);
-        ofpacts_put_openflow11_instructions(fs->ofpacts, fs->ofpacts_len,
-                                            reply);
+        ofpacts_put_openflow_instructions(fs->ofpacts, fs->ofpacts_len, reply,
+                                          version);
 
         ofs = ofpbuf_at_assert(reply, start_ofs, sizeof *ofs);
         ofs->length = htons(reply->size - start_ofs);
@@ -2798,7 +2824,7 @@ ofputil_append_flow_stats_reply(const struct ofputil_flow_stats *fs,
         ofs->idle_timeout = htons(fs->idle_timeout);
         ofs->hard_timeout = htons(fs->hard_timeout);
         if (raw == OFPRAW_OFPST13_FLOW_REPLY) {
-            ofs->flags = ofputil_encode_flow_mod_flags(fs->flags, oh->version);
+            ofs->flags = ofputil_encode_flow_mod_flags(fs->flags, version);
         } else {
             ofs->flags = 0;
         }
@@ -2810,8 +2836,8 @@ ofputil_append_flow_stats_reply(const struct ofputil_flow_stats *fs,
         struct ofp10_flow_stats *ofs;
 
         ofpbuf_put_uninit(reply, sizeof *ofs);
-        ofpacts_put_openflow10(fs->ofpacts, fs->ofpacts_len, reply);
-
+        ofpacts_put_openflow_actions(fs->ofpacts, fs->ofpacts_len, reply,
+                                     version);
         ofs = ofpbuf_at_assert(reply, start_ofs, sizeof *ofs);
         ofs->length = htons(reply->size - start_ofs);
         ofs->table_id = fs->table_id;
@@ -2834,8 +2860,8 @@ ofputil_append_flow_stats_reply(const struct ofputil_flow_stats *fs,
 
         ofpbuf_put_uninit(reply, sizeof *nfs);
         match_len = nx_put_match(reply, &fs->match, 0, 0);
-        ofpacts_put_openflow10(fs->ofpacts, fs->ofpacts_len, reply);
-
+        ofpacts_put_openflow_actions(fs->ofpacts, fs->ofpacts_len, reply,
+                                     version);
         nfs = ofpbuf_at_assert(reply, start_ofs, sizeof *nfs);
         nfs->length = htons(reply->size - start_ofs);
         nfs->table_id = fs->table_id;
@@ -3371,9 +3397,8 @@ ofputil_decode_packet_out(struct ofputil_packet_out *po,
             return error;
         }
 
-        error = ofpacts_pull_openflow11_actions(&b, oh->version,
-                                                ntohs(opo->actions_len),
-                                                ofpacts);
+        error = ofpacts_pull_openflow_actions(&b, ntohs(opo->actions_len),
+                                              oh->version, ofpacts);
         if (error) {
             return error;
         }
@@ -3384,7 +3409,8 @@ ofputil_decode_packet_out(struct ofputil_packet_out *po,
         po->buffer_id = ntohl(opo->buffer_id);
         po->in_port = u16_to_ofp(ntohs(opo->in_port));
 
-        error = ofpacts_pull_openflow10(&b, ntohs(opo->actions_len), ofpacts);
+        error = ofpacts_pull_openflow_actions(&b, ntohs(opo->actions_len),
+                                              oh->version, ofpacts);
         if (error) {
             return error;
         }
@@ -4501,6 +4527,7 @@ ofputil_decode_flow_update(struct ofputil_flow_update *update,
 {
     struct nx_flow_update_header *nfuh;
     unsigned int length;
+    struct ofp_header *oh;
 
     if (!msg->l2) {
         msg->l2 = msg->data;
@@ -4515,6 +4542,8 @@ ofputil_decode_flow_update(struct ofputil_flow_update *update,
         goto bad_len;
     }
 
+    oh = msg->l2;
+
     nfuh = msg->data;
     update->event = ntohs(nfuh->event);
     length = ntohs(nfuh->length);
@@ -4563,7 +4592,8 @@ ofputil_decode_flow_update(struct ofputil_flow_update *update,
         }
 
         actions_len = length - sizeof *nfuf - ROUND_UP(match_len, 8);
-        error = ofpacts_pull_openflow10(msg, actions_len, ofpacts);
+        error = ofpacts_pull_openflow_actions(msg, actions_len, oh->version,
+                                              ofpacts);
         if (error) {
             return error;
         }
@@ -4623,9 +4653,11 @@ ofputil_append_flow_update(const struct ofputil_flow_update *update,
     struct nx_flow_update_header *nfuh;
     struct ofpbuf *msg;
     size_t start_ofs;
+    enum ofp_version version;
 
     msg = ofpbuf_from_list(list_back(replies));
     start_ofs = msg->size;
+    version = ((struct ofp_header *)msg->l2)->version;
 
     if (update->event == NXFME_ABBREV) {
         struct nx_flow_update_abbrev *nfua;
@@ -4638,8 +4670,8 @@ ofputil_append_flow_update(const struct ofputil_flow_update *update,
 
         ofpbuf_put_zeros(msg, sizeof *nfuf);
         match_len = nx_put_match(msg, update->match, htonll(0), htonll(0));
-        ofpacts_put_openflow10(update->ofpacts, update->ofpacts_len, msg);
-
+        ofpacts_put_openflow_actions(update->ofpacts, update->ofpacts_len, msg,
+                                     version);
         nfuf = ofpbuf_at_assert(msg, start_ofs, sizeof *nfuf);
         nfuf->reason = htons(update->reason);
         nfuf->priority = htons(update->priority);
@@ -4678,7 +4710,8 @@ ofputil_encode_packet_out(const struct ofputil_packet_out *po,
         msg = ofpraw_alloc(OFPRAW_OFPT10_PACKET_OUT, OFP10_VERSION, size);
         ofpbuf_put_zeros(msg, sizeof *opo);
         actions_ofs = msg->size;
-        ofpacts_put_openflow10(po->ofpacts, po->ofpacts_len, msg);
+        ofpacts_put_openflow_actions(po->ofpacts, po->ofpacts_len, msg,
+                                     ofp_version);
 
         opo = msg->l3;
         opo->buffer_id = htonl(po->buffer_id);
@@ -4695,8 +4728,8 @@ ofputil_encode_packet_out(const struct ofputil_packet_out *po,
 
         msg = ofpraw_alloc(OFPRAW_OFPT11_PACKET_OUT, ofp_version, size);
         ofpbuf_put_zeros(msg, sizeof *opo);
-        len = ofpacts_put_openflow11_actions(po->ofpacts, po->ofpacts_len, msg);
-
+        len = ofpacts_put_openflow_actions(po->ofpacts, po->ofpacts_len, msg,
+                                           ofp_version);
         opo = msg->l3;
         opo->buffer_id = htonl(po->buffer_id);
         opo->in_port = ofputil_port_to_ofp11(po->in_port);
@@ -4833,31 +4866,6 @@ ofputil_port_to_ofp11(ofp_port_t ofp10_port)
                  : ofp_to_u16(ofp10_port) + OFPP11_OFFSET);
 }
 
-/* Checks that 'port' is a valid output port for the OFPAT10_OUTPUT action, given
- * that the switch will never have more than 'max_ports' ports.  Returns 0 if
- * 'port' is valid, otherwise an OpenFlow return code. */
-enum ofperr
-ofputil_check_output_port(ofp_port_t port, ofp_port_t max_ports)
-{
-    switch (port) {
-    case OFPP_IN_PORT:
-    case OFPP_TABLE:
-    case OFPP_NORMAL:
-    case OFPP_FLOOD:
-    case OFPP_ALL:
-    case OFPP_CONTROLLER:
-    case OFPP_NONE:
-    case OFPP_LOCAL:
-        return 0;
-
-    default:
-        if (ofp_to_u16(port) < ofp_to_u16(max_ports)) {
-            return 0;
-        }
-        return OFPERR_OFPBAC_BAD_OUT_PORT;
-    }
-}
-
 #define OFPUTIL_NAMED_PORTS                     \
         OFPUTIL_NAMED_PORT(IN_PORT)             \
         OFPUTIL_NAMED_PORT(TABLE)               \
@@ -5064,22 +5072,21 @@ size_t ofputil_count_phy_ports(uint8_t ofp_version, struct ofpbuf *b)
     return b->size / ofputil_get_phy_port_size(ofp_version);
 }
 
-/* Returns the 'enum ofputil_action_code' corresponding to 'name' (e.g. if
- * 'name' is "output" then the return value is OFPUTIL_OFPAT10_OUTPUT), or -1 if
- * 'name' is not the name of any action.
- *
- * ofp-util.def lists the mapping from names to action. */
-int
-ofputil_action_code_from_name(const char *name)
-{
-    static const char *const names[OFPUTIL_N_ACTIONS] = {
-        NULL,
+/* ofp-util.def lists the mapping from names to action. */
+static const char *const names[OFPUTIL_N_ACTIONS] = {
+    NULL,
 #define OFPAT10_ACTION(ENUM, STRUCT, NAME)             NAME,
 #define OFPAT11_ACTION(ENUM, STRUCT, EXTENSIBLE, NAME) NAME,
 #define NXAST_ACTION(ENUM, STRUCT, EXTENSIBLE, NAME)   NAME,
 #include "ofp-util.def"
-    };
+};
 
+/* Returns the 'enum ofputil_action_code' corresponding to 'name' (e.g. if
+ * 'name' is "output" then the return value is OFPUTIL_OFPAT10_OUTPUT), or -1
+ * if 'name' is not the name of any action. */
+int
+ofputil_action_code_from_name(const char *name)
+{
     const char *const *p;
 
     for (p = names; p < &names[ARRAY_SIZE(names)]; p++) {
@@ -5090,6 +5097,15 @@ ofputil_action_code_from_name(const char *name)
     return -1;
 }
 
+/* Returns name corresponding to the 'enum ofputil_action_code',
+ * or "Unkonwn action", if the name is not available. */
+const char *
+ofputil_action_name_from_code(enum ofputil_action_code code)
+{
+    return code < (int)OFPUTIL_N_ACTIONS && names[code] ? names[code]
+        : "Unknown action";
+}
+
 /* Appends an action of the type specified by 'code' to 'buf' and returns the
  * action.  Initializes the parts of 'action' that identify it as having type
  * <ENUM> and length 'sizeof *action' and zeros the rest.  For actions that
@@ -5972,6 +5988,7 @@ ofputil_append_group_desc_reply(const struct ofputil_group_desc *gds,
     struct ofp11_group_desc_stats *ogds;
     struct ofputil_bucket *bucket;
     size_t start_ogds;
+    enum ofp_version version = ((struct ofp_header *)reply->data)->version;
 
     start_ogds = reply->size;
     ofpbuf_put_zeros(reply, sizeof *ogds);
@@ -5981,9 +5998,8 @@ ofputil_append_group_desc_reply(const struct ofputil_group_desc *gds,
 
         start_ob = reply->size;
         ofpbuf_put_zeros(reply, sizeof *ob);
-        ofpacts_put_openflow11_actions(bucket->ofpacts,
-                                       bucket->ofpacts_len, reply);
-
+        ofpacts_put_openflow_actions(bucket->ofpacts, bucket->ofpacts_len,
+                                     reply, version);
         ob = ofpbuf_at_assert(reply, start_ob, sizeof *ob);
         ob->len = htons(reply->size - start_ob);
         ob->weight = htons(bucket->weight);
@@ -5999,8 +6015,8 @@ ofputil_append_group_desc_reply(const struct ofputil_group_desc *gds,
 }
 
 static enum ofperr
-ofputil_pull_buckets(struct ofpbuf *msg, enum ofp_version version,
-                     size_t buckets_length, struct list *buckets)
+ofputil_pull_buckets(struct ofpbuf *msg, size_t buckets_length,
+                     enum ofp_version version, struct list *buckets)
 {
     struct ofp11_bucket *ob;
 
@@ -6033,8 +6049,8 @@ ofputil_pull_buckets(struct ofpbuf *msg, enum ofp_version version,
         buckets_length -= ob_len;
 
         ofpbuf_init(&ofpacts, 0);
-        error = ofpacts_pull_openflow11_actions(msg, version,
-                                                ob_len - sizeof *ob, &ofpacts);
+        error = ofpacts_pull_openflow_actions(msg, ob_len - sizeof *ob,
+                                              version, &ofpacts);
         if (error) {
             ofpbuf_uninit(&ofpacts);
             ofputil_bucket_list_destroy(buckets);
@@ -6099,7 +6115,7 @@ ofputil_decode_group_desc_reply(struct ofputil_group_desc *gd,
         return OFPERR_OFPBRC_BAD_LEN;
     }
 
-    return ofputil_pull_buckets(msg, version, length - sizeof *ogds,
+    return ofputil_pull_buckets(msg, length - sizeof *ogds, version,
                                 &gd->buckets);
 }
 
@@ -6141,8 +6157,9 @@ ofputil_encode_group_mod(enum ofp_version ofp_version,
             start_bucket = b->size;
             ofpbuf_put_uninit(b, sizeof *ob);
             if (bucket->ofpacts && bucket->ofpacts_len) {
-                ofpacts_put_openflow11_actions(bucket->ofpacts,
-                                bucket->ofpacts_len, b);
+                ofpacts_put_openflow_actions(bucket->ofpacts,
+                                             bucket->ofpacts_len, b,
+                                             ofp_version);
             }
             ob = ofpbuf_at_assert(b, start_bucket, sizeof *ob);
             ob->len = htons(b->size - start_bucket);;
@@ -6183,7 +6200,7 @@ ofputil_decode_group_mod(const struct ofp_header *oh,
     gm->type = ogm->type;
     gm->group_id = ntohl(ogm->group_id);
 
-    return ofputil_pull_buckets(&msg, oh->version, msg.size, &gm->buckets);
+    return ofputil_pull_buckets(&msg, msg.size, oh->version, &gm->buckets);
 }
 
 /* Parse a queue status request message into 'oqsr'.