2 // Copyright (c) 2017 Intel Corporation
4 // Licensed under the Apache License, Version 2.0 (the "License");
5 // you may not use this file except in compliance with the License.
6 // You may obtain a copy of the License at
8 // http://www.apache.org/licenses/LICENSE-2.0
10 // Unless required by applicable law or agreed to in writing, software
11 // distributed under the License is distributed on an "AS IS" BASIS,
12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 // See the License for the specific language governing permissions and
14 // limitations under the License.
19 * Pipeline VFW BE Implementation.
21 * Implementation of Pipeline VFW Back End (BE).
22 * Responsible for packet processing.
36 #include <rte_common.h>
37 #include <rte_malloc.h>
38 #include <rte_ether.h>
39 #include <rte_ethdev.h>
43 #include <rte_byteorder.h>
45 #include <rte_table_lpm.h>
46 #include <rte_table_hash.h>
47 #include <rte_table_array.h>
48 #include <rte_table_acl.h>
49 #include <rte_table_stub.h>
50 #include <rte_timer.h>
51 #include <rte_cycles.h>
52 #include <rte_pipeline.h>
53 #include <rte_spinlock.h>
54 #include <rte_prefetch.h>
55 #include "pipeline_actions_common.h"
56 #include "hash_func.h"
57 #include "pipeline_vfw.h"
58 #include "pipeline_vfw_be.h"
59 #include "rte_cnxn_tracking.h"
60 #include "pipeline_arpicmp_be.h"
61 #include "vnf_common.h"
62 #include "vnf_define.h"
65 #include "lib_icmpv6.h"
66 #include "pipeline_common_fe.h"
70 uint8_t firewall_flag = 1;
72 uint8_t cnxn_tracking_is_active = 1;
74 * A structure defining the VFW pipeline input port per thread data.
76 struct vfw_ports_in_args {
77 struct pipeline *pipe;
78 struct rte_ct_cnxn_tracker *cnxn_tracker;
79 } __rte_cache_aligned;
81 * A structure defining the VFW pipeline per thread data.
85 pipeline_msg_req_handler custom_handlers[PIPELINE_VFW_MSG_REQS];
87 struct rte_ct_cnxn_tracker *cnxn_tracker;
88 struct rte_VFW_counter_block *counters;
89 struct rte_mbuf *pkt_buffer[PKT_BUFFER_SIZE];
90 struct lib_acl *plib_acl;
91 /* timestamp retrieved during in-port computations */
95 uint8_t links_map[PIPELINE_MAX_PORT_IN];
96 uint8_t outport_id[PIPELINE_MAX_PORT_IN];
97 /* Local ARP & ND Tables */
98 struct lib_arp_route_table_entry
99 local_lib_arp_route_table[MAX_ARP_RT_ENTRY];
100 uint8_t local_lib_arp_route_ent_cnt;
101 struct lib_nd_route_table_entry
102 local_lib_nd_route_table[MAX_ND_RT_ENTRY];
103 uint8_t local_lib_nd_route_ent_cnt;
105 } __rte_cache_aligned;
107 * A structure defining the mbuf meta data for VFW.
109 struct mbuf_tcp_meta_data {
110 /* output port stored for RTE_PIPELINE_ACTION_PORT_META */
111 uint32_t output_port;
112 struct rte_mbuf *next; /* next pointer for chained buffers */
113 } __rte_cache_aligned;
115 #define DONT_CARE_TCP_PACKET 0
116 #define IS_NOT_TCP_PACKET 0
117 #define IS_TCP_PACKET 1
119 #define META_DATA_OFFSET 128
121 #define RTE_PKTMBUF_HEADROOM 128 /* where is this defined ? */
122 #define ETHERNET_START (META_DATA_OFFSET + RTE_PKTMBUF_HEADROOM)
123 #define ETH_HDR_SIZE 14
124 #define PROTOCOL_START (IP_START + 9)
126 #define TCP_START (IP_START + 20)
127 #define RTE_LB_PORT_OFFSET 204 /* TODO: Need definition in LB header */
128 #define TCP_START_IPV6 (IP_START + 40)
129 #define PROTOCOL_START_IPV6 (IP_START + 6)
130 #define IP_HDR_DSCP_OFST 1
132 #define TCP_PROTOCOL 6
133 #define UDP_PROTOCOL 17
135 #define DELETE_BUFFERED_PACKETS 0
136 #define FORWARD_BUFFERED_PACKETS 1
140 #define IPv4_HEADER_SIZE 20
141 #define IPv6_HEADER_SIZE 40
143 #define IP_VERSION_4 4
144 #define IP_VERSION_6 6
147 #define IP_HDR_SIZE_IPV6 40
148 #define IP_HDR_DSCP_OFST_IPV6 0
149 #define IP_HDR_LENGTH_OFST_IPV6 4
150 #define IP_HDR_PROTOCOL_OFST_IPV6 6
151 #define IP_HDR_DST_ADR_OFST_IPV6 24
152 #define MAX_NUM_LOCAL_MAC_ADDRESS 16
153 /** The counter table for VFW pipeline per thread data.*/
154 struct rte_VFW_counter_block rte_vfw_counter_table[MAX_VFW_INSTANCES]
156 int rte_VFW_hi_counter_block_in_use = -1;
158 /* a spin lock used during vfw initialization only */
159 rte_spinlock_t rte_VFW_init_lock = RTE_SPINLOCK_INITIALIZER;
162 struct pipeline_action_key *action_array_a;
163 struct pipeline_action_key *action_array_b;
164 struct pipeline_action_key *action_array_active;
165 struct pipeline_action_key *action_array_standby;
166 uint32_t action_array_size;
167 struct action_counter_block
168 action_counter_table[MAX_VFW_INSTANCES][action_array_max]
171 * Pipeline table strategy for firewall. Unfortunately, there does not seem to
172 * be any use for the built-in table lookup of ip_pipeline for the firewall.
173 * The main table requirement of the firewall is the hash table to maintain
174 * connection info, but that is implemented seperately in the connection
175 * tracking library. So a "dummy" table lookup will be performed.
176 * TODO: look into "stub" table and see if that can be used
177 * to avoid useless table lookup
179 /***** ARP local cache *****/
180 uint8_t link_hw_laddr_valid[MAX_NUM_LOCAL_MAC_ADDRESS] = {
181 0, 0, 0, 0, 0, 0, 0, 0,
182 0, 0, 0, 0, 0, 0, 0, 0
185 static struct ether_addr link_hw_laddr[MAX_NUM_LOCAL_MAC_ADDRESS] = {
186 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
187 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
188 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
189 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
190 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
191 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
192 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
193 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
194 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
195 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
196 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
197 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
198 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
199 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
200 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} },
201 {.addr_bytes = {0x00, 0x00, 0x00, 0x00, 0x00, 0x00} }
204 /* Start TSC measurement */
205 /* Prefetch counters and pipe before this function */
206 static inline void start_tsc_measure(struct pipeline_vfw *vfw_pipe) {
207 vfw_pipe->counters->entry_timestamp = rte_get_tsc_cycles();
208 if (likely(vfw_pipe->counters->exit_timestamp))
209 vfw_pipe->counters->external_time_sum +=
210 vfw_pipe->counters->entry_timestamp -
211 vfw_pipe->counters->exit_timestamp;
214 /* End TSC measurement */
215 static inline void end_tsc_measure(
216 struct pipeline_vfw *vfw_pipe,
219 if (likely(n_pkts > 1)) {
220 vfw_pipe->counters->exit_timestamp = rte_get_tsc_cycles();
221 vfw_pipe->counters->internal_time_sum +=
222 vfw_pipe->counters->exit_timestamp -
223 vfw_pipe->counters->entry_timestamp;
224 vfw_pipe->counters->time_measurements++;
226 /* small counts skew results, ignore */
227 vfw_pipe->counters->exit_timestamp = 0;
231 static struct ether_addr *get_local_link_hw_addr(uint8_t out_port)
233 return &link_hw_laddr[out_port];
236 static uint8_t local_dest_mac_present(uint8_t out_port)
238 return link_hw_laddr_valid[out_port];
241 static uint32_t local_get_nh_ipv4(
245 struct pipeline_vfw *vfw_pipe)
249 for (i = 0; i < vfw_pipe->local_lib_arp_route_ent_cnt; i++) {
250 if (((vfw_pipe->local_lib_arp_route_table[i].ip &
251 vfw_pipe->local_lib_arp_route_table[i].mask) ==
252 (ip & vfw_pipe->local_lib_arp_route_table[i].mask))) {
253 *port = vfw_pipe->local_lib_arp_route_table[i].port;
255 *nhip = vfw_pipe->local_lib_arp_route_table[i].nh;
262 static void do_local_nh_ipv4_cache(uint32_t dest_if,
263 struct pipeline_vfw *vfw_pipe)
266 /* Search for the entry and do local copy */
269 for (i = 0; i < MAX_ARP_RT_ENTRY; i++) {
270 if (lib_arp_route_table[i].port == dest_if) {
272 struct lib_arp_route_table_entry *lentry =
274 local_lib_arp_route_table[vfw_pipe->
275 local_lib_arp_route_ent_cnt];
277 lentry->ip = lib_arp_route_table[i].ip;
278 lentry->mask = lib_arp_route_table[i].mask;
279 lentry->port = lib_arp_route_table[i].port;
280 lentry->nh = lib_arp_route_table[i].nh;
282 vfw_pipe->local_lib_arp_route_ent_cnt++;
287 static uint32_t local_get_nh_ipv6(
291 struct pipeline_vfw *vfw_pipe)
293 uint8_t netmask_ipv6[IPV6_ADD_SIZE], netip_nd[IPV6_ADD_SIZE],
294 netip_in[IPV6_ADD_SIZE];
295 uint8_t i = 0, j = 0, k = 0, l = 0, depthflags = 0, depthflags1 = 0;
296 memset(netmask_ipv6, 0, sizeof(netmask_ipv6));
297 memset(netip_nd, 0, sizeof(netip_nd));
298 memset(netip_in, 0, sizeof(netip_in));
300 for (i = 0; i < vfw_pipe->local_lib_nd_route_ent_cnt; i++) {
302 convert_prefixlen_to_netmask_ipv6(
303 vfw_pipe->local_lib_nd_route_table[i].depth,
306 for (k = 0; k < IPV6_ADD_SIZE; k++)
307 if (vfw_pipe->local_lib_nd_route_table[i].ipv6[k] &
310 netip_nd[k] = vfw_pipe->
311 local_lib_nd_route_table[i].ipv6[k];
314 for (l = 0; l < IPV6_ADD_SIZE; l++)
315 if (ip[l] & netmask_ipv6[l]) {
321 if ((depthflags == depthflags1) && (memcmp(netip_nd, netip_in,
322 sizeof(netip_nd)) == 0)) {
324 *port = vfw_pipe->local_lib_nd_route_table[i].port;
326 for (j = 0; j < IPV6_ADD_SIZE; j++)
328 local_lib_nd_route_table[i].nhipv6[j];
338 static void do_local_nh_ipv6_cache(uint32_t dest_if,
339 struct pipeline_vfw *vfw_pipe)
341 /* Search for the entry and do local copy */
344 for (i = 0; i < MAX_ND_RT_ENTRY; i++) {
346 if (lib_nd_route_table[i].port == dest_if) {
348 struct lib_nd_route_table_entry *lentry = &vfw_pipe->
349 local_lib_nd_route_table[vfw_pipe->
350 local_lib_nd_route_ent_cnt];
352 for (l = 0; l < IPV6_ADD_SIZE; l++) {
354 lib_nd_route_table[i].ipv6[l];
356 lib_nd_route_table[i].nhipv6[l];
358 lentry->depth = lib_nd_route_table[i].depth;
359 lentry->port = lib_nd_route_table[i].port;
361 vfw_pipe->local_lib_nd_route_ent_cnt++;
367 * Print packet for debugging.
370 * A pointer to the packet.
373 static __rte_unused void print_pkt(struct rte_mbuf *pkt)
376 int size = (int)sizeof(struct mbuf_tcp_meta_data);
377 uint8_t *rd = RTE_MBUF_METADATA_UINT8_PTR(pkt, META_DATA_OFFSET);
379 printf("Meta-data:\n");
380 for (i = 0; i < size; i++) {
381 printf("%02x ", rd[i]);
382 if ((i & TWO_BYTE_PRINT) == TWO_BYTE_PRINT)
386 printf("IP and TCP/UDP headers:\n");
387 rd = RTE_MBUF_METADATA_UINT8_PTR(pkt, IP_START);
388 for (i = 0; i < IP_HDR_SIZE_IPV6; i++) {
389 printf("%02x ", rd[i]);
390 if ((i & TWO_BYTE_PRINT) == TWO_BYTE_PRINT)
396 /* TODO: are the protocol numbers defined somewhere with meaningful names? */
397 #define IP_ICMP_PROTOCOL 1
398 #define IP_TCP_PROTOCOL 6
399 #define IP_UDP_PROTOCOL 17
400 #define IPv6_FRAGMENT_HEADER 44
403 * Return ethernet header structure form packet.
406 * A pointer to the packet.
409 static inline struct ether_hdr *rte_vfw_get_ether_addr(struct rte_mbuf *pkt)
411 return (struct ether_hdr *)RTE_MBUF_METADATA_UINT32_PTR(pkt,
416 * Return IPV4 header structure form packet.
419 * A pointer to the packet.
423 static inline struct ipv4_hdr *rte_vfw_get_IPv4_hdr_addr(
424 struct rte_mbuf *pkt)
426 return (struct ipv4_hdr *)RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
429 static inline int rte_vfw_is_IPv4(struct rte_mbuf *pkt)
431 /* NOTE: Only supporting IP headers with no options,
432 * so header is fixed size */
433 uint8_t ip_type = RTE_MBUF_METADATA_UINT8(pkt, IP_START)
436 return ip_type == IPv4_HDR_VERSION;
439 static inline int rte_vfw_is_IPv6(struct rte_mbuf *pkt)
441 /* NOTE: Only supporting IP headers with no options,
442 * so header is fixed size */
443 uint8_t ip_type = RTE_MBUF_METADATA_UINT8(pkt, IP_START)
446 return ip_type == IPv6_HDR_VERSION;
449 static inline void rte_vfw_incr_drop_ctr(uint64_t *counter)
451 if (likely(firewall_flag))
455 static uint8_t check_arp_icmp(
456 struct rte_mbuf *pkt,
457 struct pipeline_vfw *vfw_pipe)
459 struct ether_hdr *ehdr;
460 struct app_link_params *link;
461 uint8_t solicited_node_multicast_addr[IPV6_ADD_SIZE] = {
462 0xff, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
463 0x00, 0x00, 0x00, 0x01, 0xff, 0x00, 0x00, 0x00};
465 /* ARP outport number */
466 uint16_t out_port = vfw_pipe->pipe.n_ports_out - 1;
467 struct ipv4_hdr *ipv4_h;
468 struct ipv6_hdr *ipv6_h;
469 link = &myApp->link_params[pkt->port];
471 ehdr = rte_vfw_get_ether_addr(pkt);
472 switch (rte_be_to_cpu_16(ehdr->ether_type)) {
475 rte_pipeline_port_out_packet_insert(
480 vfw_pipe->counters->arpicmpPktCount++;
484 ipv4_h = (struct ipv4_hdr *)
485 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
486 if ((ipv4_h->next_proto_id == IP_PROTOCOL_ICMP) &&
488 rte_be_to_cpu_32(ipv4_h->dst_addr)) {
489 if (is_phy_port_privte(pkt->port)) {
490 rte_pipeline_port_out_packet_insert(
495 vfw_pipe->counters->arpicmpPktCount++;
502 ipv6_h = (struct ipv6_hdr *)
503 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
505 if (ipv6_h->proto == ICMPV6_PROTOCOL_ID) {
506 if (!memcmp(ipv6_h->dst_addr, link->ipv6, IPV6_ADD_SIZE)
507 || !memcmp(ipv6_h->dst_addr,
508 solicited_node_multicast_addr,
509 IPV6_ADD_CMP_MULTI)) {
511 rte_pipeline_port_out_packet_insert(
516 vfw_pipe->counters->arpicmpPktCount++;
520 pkts_drop_unsupported_type++;
533 * Performs basic VFW ipv4 packet filtering.
535 * A pointer to the packets.
539 * A pointer to VFW pipeline.
543 rte_vfw_ipv4_packet_filter_and_process(struct rte_mbuf **pkts,
545 struct pipeline_vfw *vfw_pipe)
549 * Make use of cache prefetch. At beginning of loop, want to prefetch
550 * mbuf data for next iteration (not current one).
551 * Note that ethernet header (14 bytes) is cache aligned. IPv4 header
552 * is 20 bytes (extensions not supported), while the IPv6 header is 40
553 * bytes. TCP header is 20 bytes, UDP is 8. One cache line prefetch
554 * will cover IPv4 and TCP or UDP, but to get IPv6 and TCP,
555 * need two pre-fetches.
558 uint8_t pos, next_pos = 0;
559 uint64_t pkt_mask; /* bitmask representing a single packet */
560 struct rte_mbuf *pkt;
561 struct rte_mbuf *next_pkt = NULL;
562 struct ipv4_hdr *ihdr4;
563 void *next_iphdr = NULL;
565 if (unlikely(pkts_mask == 0))
567 pos = (uint8_t) __builtin_ctzll(pkts_mask);
568 pkt_mask = 1LLU << pos; /* bitmask representing only this packet */
571 uint64_t bytes_processed = 0;
572 /* bitmap of packets left to process */
573 uint64_t pkts_to_process = pkts_mask;
574 /* bitmap of valid packets to return */
575 uint64_t valid_packets = pkts_mask;
578 /* prefetch counters, updated below. Most likely counters to update
580 rte_prefetch0(&vfw_pipe->counters);
582 do { /* always execute at least once */
584 /* remove this packet from remaining list */
585 uint64_t next_pkts_to_process = pkts_to_process &= ~pkt_mask;
587 if (likely(next_pkts_to_process)) {
588 /* another packet to process after this, prefetch it */
591 (uint8_t) __builtin_ctzll(next_pkts_to_process);
592 next_pkt = pkts[next_pos];
593 next_iphdr = RTE_MBUF_METADATA_UINT32_PTR(next_pkt,
595 rte_prefetch0(next_iphdr);
599 /* remove this packet from remaining list */
600 pkts_to_process &= ~pkt_mask;
603 if (!check_arp_icmp(pkt, vfw_pipe))
606 uint32_t packet_length = rte_pktmbuf_pkt_len(pkt);
608 bytes_processed += packet_length;
610 ihdr4 = (struct ipv4_hdr *)
611 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
613 /* verify that packet size according to mbuf is at least
614 * as large as the size according to the IP header.
617 uint32_t ip_length = rte_bswap16(ihdr4->total_length);
620 (ip_length > (packet_length - ETH_HDR_SIZE))) {
622 vfw_pipe->counters->pkts_drop_bad_size++;
626 * IPv4 fragmented if: MF (more fragments) or Fragment
627 * Offset are non-zero. Header in Intel order, so flip
628 * constant to compensate. Note that IPv6 uses a header
629 * extension for identifying fragments.
632 int fragmented = (ihdr4->fragment_offset & 0xff3f) != 0;
633 uint8_t ttl = ihdr4->time_to_live;
635 if (unlikely(fragmented)) {
637 vfw_pipe->counters->pkts_drop_fragmented++;
640 if (unlikely(ttl <= 1)) {
642 * about to decrement to zero (or is somehow
643 * already zero), so discard
646 vfw_pipe->counters->pkts_drop_ttl++;
650 * Dropping the packets other than TCP AND UDP.
653 uint8_t proto = ihdr4->next_proto_id;
655 if (unlikely(!(proto == IP_TCP_PROTOCOL ||
656 proto == IP_UDP_PROTOCOL ||
657 proto == IP_ICMP_PROTOCOL))) {
660 pkts_drop_unsupported_type++;
663 if (unlikely(discard)) {
664 valid_packets &= ~pkt_mask;
667 /* make next packet data the current */
668 pkts_to_process = next_pkts_to_process;
672 pkt_mask = 1LLU << pos;
674 } while (pkts_to_process);
676 /* finalize counters, etc. */
677 vfw_pipe->counters->bytes_processed += bytes_processed;
679 if (likely(firewall_flag))
680 return valid_packets;
685 * Performs basic VFW IPV6 packet filtering.
687 * A pointer to the packets.
691 * A pointer to VFW pipeline.
694 rte_vfw_ipv6_packet_filter_and_process(struct rte_mbuf **pkts,
696 struct pipeline_vfw *vfw_pipe)
700 * Make use of cache prefetch. At beginning of loop, want to prefetch
701 * mbuf data for next iteration (not current one).
702 * Note that ethernet header (14 bytes) is cache aligned. IPv4 header
703 * is 20 bytes (extensions not supported), while the IPv6 header is 40
704 * bytes. TCP header is 20 bytes, UDP is 8. One cache line prefetch
705 * will cover IPv4 and TCP or UDP, but to get IPv6 and TCP,
706 * need two pre-fetches.
709 uint8_t pos, next_pos = 0;
710 uint64_t pkt_mask; /* bitmask representing a single packet */
711 struct rte_mbuf *pkt;
712 struct rte_mbuf *next_pkt = NULL;
713 struct ipv6_hdr *ihdr6;
714 void *next_iphdr = NULL;
716 if (unlikely(pkts_mask == 0))
718 pos = (uint8_t) __builtin_ctzll(pkts_mask);
719 pkt_mask = 1LLU << pos; /* bitmask representing only this packet */
722 uint64_t bytes_processed = 0;
723 /* bitmap of packets left to process */
724 uint64_t pkts_to_process = pkts_mask;
725 /* bitmap of valid packets to return */
726 uint64_t valid_packets = pkts_mask;
728 /* prefetch counters, updated below. Most likely counters to update
730 rte_prefetch0(&vfw_pipe->counters);
732 do { /* always execute at least once */
734 /* remove this packet from remaining list */
735 uint64_t next_pkts_to_process = pkts_to_process &= ~pkt_mask;
737 if (likely(next_pkts_to_process)) {
738 /* another packet to process after this, prefetch it */
741 (uint8_t) __builtin_ctzll(next_pkts_to_process);
742 next_pkt = pkts[next_pos];
744 RTE_MBUF_METADATA_UINT32_PTR(next_pkt, IP_START);
745 rte_prefetch0(next_iphdr);
749 /* remove this packet from remaining list */
750 pkts_to_process &= ~pkt_mask;
753 if (!check_arp_icmp(pkt, vfw_pipe))
756 uint32_t packet_length = rte_pktmbuf_pkt_len(pkt);
758 bytes_processed += packet_length;
760 ihdr6 = (struct ipv6_hdr *)
761 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
764 * verify that packet size according to mbuf is at least
765 * as large as the size according to the IP header.
766 * For IPv6, note that size includes header extensions
767 * but not the base header size
771 rte_bswap16(ihdr6->payload_len) + IPv6_HEADER_SIZE;
774 (ip_length > (packet_length - ETH_HDR_SIZE))) {
776 vfw_pipe->counters->pkts_drop_bad_size++;
780 * Dropping the packets other than TCP AND UDP.
783 uint8_t proto = ihdr6->proto;
785 if (unlikely(!(proto == IP_TCP_PROTOCOL ||
786 proto == IP_UDP_PROTOCOL ||
787 proto == IP_ICMP_PROTOCOL))) {
789 if (proto == IPv6_FRAGMENT_HEADER)
791 pkts_drop_fragmented++;
794 pkts_drop_unsupported_type++;
798 * Behave like a router, and decrement the TTL of an
799 * IP packet. If this causes the TTL to become zero,
800 * the packet will be discarded. Unlike a router,
801 * no ICMP code 11 (Time * Exceeded) message will be
802 * sent back to the packet originator.
805 if (unlikely(ihdr6->hop_limits <= 1)) {
807 * about to decrement to zero (or is somehow
808 * already zero), so discard
811 vfw_pipe->counters->pkts_drop_ttl++;
814 if (unlikely(discard))
815 valid_packets &= ~pkt_mask;
819 /* make next packet data the current */
820 pkts_to_process = next_pkts_to_process;
824 pkt_mask = 1LLU << pos;
826 } while (pkts_to_process);
828 /* finalize counters, etc. */
829 vfw_pipe->counters->bytes_processed += bytes_processed;
831 if (likely(firewall_flag))
832 return valid_packets;
838 * exchange the mac address so source becomes destination and vice versa.
841 * A pointer to the ethernet header.
844 static inline void rte_sp_exchange_mac_addresses(struct ether_hdr *ehdr)
846 struct ether_addr saved_copy;
848 ether_addr_copy(&ehdr->d_addr, &saved_copy);
849 ether_addr_copy(&ehdr->s_addr, &ehdr->d_addr);
850 ether_addr_copy(&saved_copy, &ehdr->s_addr);
855 * walk every valid mbuf (denoted by pkts_mask) and apply arp to the packet.
856 * To support synproxy, some (altered) packets may need to be sent back where
857 * they came from. The ip header has already been adjusted, but the ethernet
858 * header has not, so this must be performed here.
859 * Return an updated pkts_mask, since arp may drop some packets
862 * A pointer to the packet array.
864 * Packet num to start processing
867 * @param synproxy_reply_mask
868 * Reply Packet mask for Synproxy
870 * A pointer to VFW pipeline.
873 pkt4_work_vfw_arp_ipv4_packets(struct rte_mbuf **pkts,
876 uint64_t synproxy_reply_mask,
877 struct pipeline_vfw *vfw_pipe)
884 struct ether_addr hw_addr;
885 struct mbuf_tcp_meta_data *meta_data_addr;
886 struct ether_hdr *ehdr;
887 struct rte_mbuf *pkt;
890 for (i = 0; i < 4; i++) {
891 uint32_t dest_if = INVALID_DESTIF;
892 /* bitmask representing only this packet */
893 uint64_t pkt_mask = 1LLU << (pkt_num + i);
897 if(!(*pkts_mask & pkt_mask))
900 int must_reverse = ((synproxy_reply_mask & pkt_mask) != 0);
902 phy_port = pkt->port;
903 meta_data_addr = (struct mbuf_tcp_meta_data *)
904 RTE_MBUF_METADATA_UINT32_PTR(pkt, META_DATA_OFFSET);
905 ehdr = rte_vfw_get_ether_addr(pkt);
908 struct ipv4_hdr *ihdr = (struct ipv4_hdr *)
909 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
912 uint32_t dest_address = rte_bswap32(ihdr->dst_addr);
914 ret = local_get_nh_ipv4(dest_address, &dest_if,
917 rte_sp_exchange_mac_addresses(ehdr);
918 if (is_phy_port_privte(phy_port)) {
920 dest_if = get_pub_to_prv_port(
923 if (dest_if == INVALID_DESTIF) {
924 *pkts_mask &= ~pkt_mask;
926 pkts_drop_without_arp_entry++;
928 do_local_nh_ipv4_cache(
934 dest_if = get_prv_to_pub_port(
937 if (dest_if == INVALID_DESTIF) {
938 *pkts_mask &= ~pkt_mask;
940 pkts_drop_without_arp_entry++;
942 do_local_nh_ipv4_cache(dest_if,
946 } else if (is_phy_port_privte(phy_port)) {
948 dest_if = get_prv_to_pub_port(&dest_address,
950 if (dest_if == INVALID_DESTIF) {
951 *pkts_mask &= ~pkt_mask;
953 pkts_drop_without_arp_entry++;
955 do_local_nh_ipv4_cache(dest_if, vfw_pipe);
960 dest_if = get_pub_to_prv_port(&dest_address,
962 if (dest_if == INVALID_DESTIF) {
963 *pkts_mask &= ~pkt_mask;
965 pkts_drop_without_arp_entry++;
967 do_local_nh_ipv4_cache(dest_if, vfw_pipe);
971 meta_data_addr->output_port = vfw_pipe->outport_id[dest_if];
972 if (local_dest_mac_present(dest_if)) {
973 ether_addr_copy(get_local_link_hw_addr(dest_if),
975 ether_addr_copy(get_link_hw_addr(dest_if),
978 ret_mac = get_dest_mac_addr_port(dest_address,
980 if (ret_mac == ARP_FOUND) {
982 link_hw_laddr_valid[dest_if] = 1;
983 memcpy(&link_hw_laddr[dest_if], &hw_addr,
984 sizeof(struct ether_addr));
986 ether_addr_copy(&hw_addr, &ehdr->d_addr);
987 ether_addr_copy(get_link_hw_addr(dest_if),
990 if (vfw_debug >= DEBUG_LEVEL_4) {
991 char buf[HW_ADDR_SIZE];
993 ether_format_addr(buf, sizeof(buf),
995 printf("MAC found for ip 0x%"
996 PRIx32", dest_if %d: %s, ",
999 ether_format_addr(buf, sizeof(buf),
1001 printf("new eth hdr src: %s, ", buf);
1002 ether_format_addr(buf, sizeof(buf),
1004 printf("new eth hdr dst: %s\n", buf);
1009 if (vfw_debug >= DEBUG_LEVEL_4) {
1010 char buf[HW_ADDR_SIZE];
1012 ether_format_addr(buf, sizeof(buf),
1014 printf("MAC NOT FOUND for ip 0x%"
1020 /* ICMP req sent, drop packet by
1021 * changing the mask */
1022 *pkts_mask &= ~pkt_mask;
1024 counters->pkts_drop_without_arp_entry++;
1032 * walk every valid mbuf (denoted by pkts_mask) and apply arp to the packet.
1033 * To support synproxy, some (altered) packets may need to be sent back where
1034 * they came from. The ip header has already been adjusted, but the ethernet
1035 * header has not, so this must be performed here.
1036 * Return an updated pkts_mask, since arp may drop some packets
1039 * A pointer to the packet.
1041 * Packet number to process
1043 * Packet mask pointer
1044 * @param synproxy_reply_mask
1045 * Reply Packet mask for Synproxy
1047 * A pointer to VFW pipeline.
1050 pkt_work_vfw_arp_ipv4_packets(struct rte_mbuf *pkts,
1052 uint64_t *pkts_mask,
1053 uint64_t synproxy_reply_mask,
1054 struct pipeline_vfw *vfw_pipe)
1058 uint32_t dest_if = INVALID_DESTIF;
1061 struct ether_addr hw_addr;
1062 struct mbuf_tcp_meta_data *meta_data_addr;
1063 struct ether_hdr *ehdr;
1064 struct rte_mbuf *pkt;
1066 uint64_t pkt_mask = 1LLU << pkt_num;
1070 if(*pkts_mask & pkt_mask) {
1072 int must_reverse = ((synproxy_reply_mask & pkt_mask) != 0);
1074 phy_port = pkt->port;
1075 meta_data_addr = (struct mbuf_tcp_meta_data *)
1076 RTE_MBUF_METADATA_UINT32_PTR(pkt, META_DATA_OFFSET);
1077 ehdr = rte_vfw_get_ether_addr(pkt);
1080 struct ipv4_hdr *ihdr = (struct ipv4_hdr *)
1081 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
1084 uint32_t dest_address = rte_bswap32(ihdr->dst_addr);
1086 ret = local_get_nh_ipv4(dest_address, &dest_if,
1089 rte_sp_exchange_mac_addresses(ehdr);
1090 if (is_phy_port_privte(phy_port)) {
1092 dest_if = get_pub_to_prv_port(
1095 if (dest_if == INVALID_DESTIF) {
1096 *pkts_mask &= ~pkt_mask;
1097 vfw_pipe->counters->
1098 pkts_drop_without_arp_entry++;
1100 do_local_nh_ipv4_cache(
1106 dest_if = get_prv_to_pub_port(
1109 if (dest_if == INVALID_DESTIF) {
1110 *pkts_mask &= ~pkt_mask;
1111 vfw_pipe->counters->
1112 pkts_drop_without_arp_entry++;
1114 do_local_nh_ipv4_cache(dest_if,
1118 } else if (is_phy_port_privte(phy_port)) {
1120 dest_if = get_prv_to_pub_port(&dest_address,
1122 if (dest_if == INVALID_DESTIF) {
1123 *pkts_mask &= ~pkt_mask;
1124 vfw_pipe->counters->
1125 pkts_drop_without_arp_entry++;
1127 do_local_nh_ipv4_cache(dest_if, vfw_pipe);
1132 dest_if = get_pub_to_prv_port(&dest_address,
1134 if (dest_if == INVALID_DESTIF) {
1135 *pkts_mask &= ~pkt_mask;
1136 vfw_pipe->counters->
1137 pkts_drop_without_arp_entry++;
1139 do_local_nh_ipv4_cache(dest_if, vfw_pipe);
1143 meta_data_addr->output_port = vfw_pipe->outport_id[dest_if];
1144 if (local_dest_mac_present(dest_if)) {
1145 ether_addr_copy(get_local_link_hw_addr(dest_if),
1147 ether_addr_copy(get_link_hw_addr(dest_if),
1150 ret_mac = get_dest_mac_addr_port(dest_address,
1151 &dest_if, &hw_addr);
1153 link_hw_laddr_valid[dest_if] = 1;
1154 memcpy(&link_hw_laddr[dest_if], &hw_addr,
1155 sizeof(struct ether_addr));
1157 ether_addr_copy(&hw_addr, &ehdr->d_addr);
1158 ether_addr_copy(get_link_hw_addr(dest_if),
1161 if (vfw_debug >= DEBUG_LEVEL_4) {
1162 char buf[HW_ADDR_SIZE];
1164 ether_format_addr(buf, sizeof(buf),
1166 printf("MAC found for ip 0x%"
1167 PRIx32", dest_if %d: %s, ",
1170 ether_format_addr(buf, sizeof(buf),
1172 printf("new eth hdr src: %s, ", buf);
1173 ether_format_addr(buf, sizeof(buf),
1175 printf("new eth hdr dst: %s\n", buf);
1179 if (vfw_debug >= DEBUG_LEVEL_4) {
1180 char buf[HW_ADDR_SIZE];
1182 ether_format_addr(buf, sizeof(buf),
1184 printf("MAC NOT FOUND for ip 0x%"
1190 /* ICMP req sent, drop packet by
1191 * changing the mask */
1192 *pkts_mask &= ~pkt_mask;
1194 counters->pkts_drop_without_arp_entry++;
1203 * walk every valid mbuf (denoted by pkts_mask) and apply arp to the packet.
1204 * To support synproxy, some (altered) packets may need to be sent back where
1205 * they came from. The ip header has already been adjusted, but the ethernet
1206 * header has not, so this must be performed here.
1207 * Return an updated pkts_mask, since arp may drop some packets
1210 * A pointer to the packets array.
1212 * Packet number to start processing.
1214 * Packet mask pointer
1215 * @param synproxy_reply_mask
1216 * Reply Packet mask for Synproxy
1218 * A pointer to VFW pipeline.
1222 pkt4_work_vfw_arp_ipv6_packets(struct rte_mbuf **pkts,
1224 uint64_t *pkts_mask,
1225 uint64_t synproxy_reply_mask,
1226 struct pipeline_vfw *vfw_pipe)
1228 uint8_t nh_ipv6[IPV6_ADD_SIZE];
1230 struct ether_addr hw_addr;
1231 struct mbuf_tcp_meta_data *meta_data_addr;
1232 struct ether_hdr *ehdr;
1233 struct rte_mbuf *pkt;
1237 for (i = 0; i < 4; i++) {
1238 uint32_t dest_if = INVALID_DESTIF;
1239 /* bitmask representing only this packet */
1240 uint64_t pkt_mask = 1LLU << (pkt_num + i);
1244 if(!(*pkts_mask & pkt_mask))
1246 int must_reverse = ((synproxy_reply_mask & pkt_mask) != 0);
1248 phy_port = pkt->port;
1249 meta_data_addr = (struct mbuf_tcp_meta_data *)
1250 RTE_MBUF_METADATA_UINT32_PTR(pkt, META_DATA_OFFSET);
1251 ehdr = rte_vfw_get_ether_addr(pkt);
1253 struct ipv6_hdr *ihdr = (struct ipv6_hdr *)
1254 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
1256 uint8_t nhip[IPV6_ADD_SIZE];
1257 uint8_t dest_address[IPV6_ADD_SIZE];
1259 memset(nhip, 0, IPV6_ADD_SIZE);
1261 rte_mov16(dest_address, ihdr->dst_addr);
1262 ret = local_get_nh_ipv6(&dest_address[0], &dest_if,
1263 &nhip[0], vfw_pipe);
1265 rte_sp_exchange_mac_addresses(ehdr);
1266 if (is_phy_port_privte(phy_port)) {
1268 dest_if = get_pub_to_prv_port(
1272 if (dest_if == INVALID_DESTIF) {
1273 *pkts_mask &= ~pkt_mask;
1274 vfw_pipe->counters->
1275 pkts_drop_without_arp_entry++;
1277 do_local_nh_ipv6_cache(dest_if,
1283 dest_if = get_prv_to_pub_port(
1287 if (dest_if == INVALID_DESTIF) {
1288 *pkts_mask &= ~pkt_mask;
1289 vfw_pipe->counters->
1290 pkts_drop_without_arp_entry++;
1292 do_local_nh_ipv6_cache(dest_if,
1297 } else if (is_phy_port_privte(phy_port)) {
1299 dest_if = get_prv_to_pub_port((uint32_t *)
1300 &dest_address[0], IP_VERSION_6);
1301 if (dest_if == INVALID_DESTIF) {
1302 *pkts_mask &= ~pkt_mask;
1303 vfw_pipe->counters->
1304 pkts_drop_without_arp_entry++;
1306 do_local_nh_ipv6_cache(dest_if, vfw_pipe);
1311 dest_if = get_pub_to_prv_port((uint32_t *)
1312 &dest_address[0], IP_VERSION_6);
1313 if (dest_if == INVALID_DESTIF) {
1314 *pkts_mask &= ~pkt_mask;
1315 vfw_pipe->counters->
1316 pkts_drop_without_arp_entry++;
1319 do_local_nh_ipv6_cache(dest_if, vfw_pipe);
1324 meta_data_addr->output_port = vfw_pipe->outport_id[dest_if];
1326 memset(nh_ipv6, 0, IPV6_ADD_SIZE);
1327 if (get_dest_mac_address_ipv6_port(
1332 ether_addr_copy(&hw_addr, &ehdr->d_addr);
1333 ether_addr_copy(get_link_hw_addr(dest_if),
1336 if (vfw_debug >= DEBUG_LEVEL_4) {
1337 char buf[HW_ADDR_SIZE];
1339 ether_format_addr(buf, sizeof(buf),
1341 printf("MAC found for dest_if %d: %s, ",
1343 ether_format_addr(buf, sizeof(buf),
1345 printf("new eth hdr src: %s, ", buf);
1346 ether_format_addr(buf, sizeof(buf),
1348 printf("new eth hdr dst: %s\n", buf);
1352 printf("deleting ipv6\n");
1353 *pkts_mask &= ~pkt_mask;
1354 /*Next Neighbor is not yet implemented
1356 vfw_pipe->counters->
1357 pkts_drop_without_arp_entry++;
1365 * walk every valid mbuf (denoted by pkts_mask) and apply arp to the packet.
1366 * To support synproxy, some (altered) packets may need to be sent back where
1367 * they came from. The ip header has already been adjusted, but the ethernet
1368 * header has not, so this must be performed here.
1369 * Return an updated pkts_mask, since arp may drop some packets
1372 * A pointer to the packets.
1374 * Packet number to process.
1376 * Packet mask pointer
1377 * @param synproxy_reply_mask
1378 * Reply Packet mask for Synproxy
1380 * A pointer to VFW pipeline.
1384 pkt_work_vfw_arp_ipv6_packets(struct rte_mbuf *pkts,
1386 uint64_t *pkts_mask,
1387 uint64_t synproxy_reply_mask,
1388 struct pipeline_vfw *vfw_pipe)
1390 uint8_t nh_ipv6[IPV6_ADD_SIZE];
1392 struct ether_addr hw_addr;
1393 struct mbuf_tcp_meta_data *meta_data_addr;
1394 struct ether_hdr *ehdr;
1395 struct rte_mbuf *pkt;
1398 uint32_t dest_if = INVALID_DESTIF;
1399 /* bitmask representing only this packet */
1400 uint64_t pkt_mask = 1LLU << pkt_num;
1404 if(*pkts_mask & pkt_mask) {
1406 int must_reverse = ((synproxy_reply_mask & pkt_mask) != 0);
1408 phy_port = pkt->port;
1409 meta_data_addr = (struct mbuf_tcp_meta_data *)
1410 RTE_MBUF_METADATA_UINT32_PTR(pkt, META_DATA_OFFSET);
1411 ehdr = rte_vfw_get_ether_addr(pkt);
1413 struct ipv6_hdr *ihdr = (struct ipv6_hdr *)
1414 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
1416 uint8_t nhip[IPV6_ADD_SIZE];
1417 uint8_t dest_address[IPV6_ADD_SIZE];
1419 memset(nhip, 0, IPV6_ADD_SIZE);
1421 rte_mov16(dest_address, ihdr->dst_addr);
1422 ret = local_get_nh_ipv6(&dest_address[0], &dest_if,
1423 &nhip[0], vfw_pipe);
1425 rte_sp_exchange_mac_addresses(ehdr);
1426 if (is_phy_port_privte(phy_port)) {
1428 dest_if = get_pub_to_prv_port(
1432 if (dest_if == INVALID_DESTIF) {
1433 *pkts_mask &= ~pkt_mask;
1434 vfw_pipe->counters->
1435 pkts_drop_without_arp_entry++;
1437 do_local_nh_ipv6_cache(dest_if,
1443 dest_if = get_prv_to_pub_port(
1447 if (dest_if == INVALID_DESTIF) {
1448 *pkts_mask &= ~pkt_mask;
1449 vfw_pipe->counters->
1450 pkts_drop_without_arp_entry++;
1452 do_local_nh_ipv6_cache(dest_if,
1457 } else if (is_phy_port_privte(phy_port)) {
1459 dest_if = get_prv_to_pub_port((uint32_t *)
1460 &dest_address[0], IP_VERSION_6);
1461 if (dest_if == INVALID_DESTIF) {
1462 *pkts_mask &= ~pkt_mask;
1463 vfw_pipe->counters->
1464 pkts_drop_without_arp_entry++;
1466 do_local_nh_ipv6_cache(dest_if, vfw_pipe);
1471 dest_if = get_pub_to_prv_port((uint32_t *)
1472 &dest_address[0], IP_VERSION_6);
1473 if (dest_if == INVALID_DESTIF) {
1474 *pkts_mask &= ~pkt_mask;
1475 vfw_pipe->counters->
1476 pkts_drop_without_arp_entry++;
1479 do_local_nh_ipv6_cache(dest_if, vfw_pipe);
1484 meta_data_addr->output_port = vfw_pipe->outport_id[dest_if];
1486 memset(nh_ipv6, 0, IPV6_ADD_SIZE);
1487 if (get_dest_mac_address_ipv6_port(
1492 ether_addr_copy(&hw_addr, &ehdr->d_addr);
1493 ether_addr_copy(get_link_hw_addr(dest_if),
1496 if (vfw_debug >= DEBUG_LEVEL_4) {
1497 char buf[HW_ADDR_SIZE];
1499 ether_format_addr(buf, sizeof(buf),
1501 printf("MAC found for dest_if %d: %s, ",
1503 ether_format_addr(buf, sizeof(buf),
1505 printf("new eth hdr src: %s, ", buf);
1506 ether_format_addr(buf, sizeof(buf),
1508 printf("new eth hdr dst: %s\n", buf);
1512 printf("deleting ipv6\n");
1513 *pkts_mask &= ~pkt_mask;
1514 /*Next Neighbor is not yet implemented
1516 vfw_pipe->counters->
1517 pkts_drop_without_arp_entry++;
1527 * walk every valid mbuf (denoted by pkts_mask) and apply arp to the packet.
1528 * To support synproxy, some (altered) packets may need to be sent back where
1529 * they came from. The ip header has already been adjusted, but the ethernet
1530 * header has not, so this must be performed here.
1531 * Return an updated pkts_mask, since arp may drop some packets
1534 * A pointer to the packet.
1537 * @param synproxy_reply_mask
1538 * Reply Packet mask for Synproxy
1540 * A pointer to VFW pipeline.
1543 rte_vfw_arp_ipv4_packets(struct rte_mbuf **pkts,
1545 uint64_t synproxy_reply_mask,
1546 struct pipeline_vfw *vfw_pipe)
1548 uint64_t pkts_to_arp = pkts_mask;
1551 uint32_t dest_if = INVALID_DESTIF;
1553 for (; pkts_to_arp;) {
1554 struct ether_addr hw_addr;
1555 struct mbuf_tcp_meta_data *meta_data_addr;
1556 struct ether_hdr *ehdr;
1557 struct rte_mbuf *pkt;
1560 uint8_t pos = (uint8_t) __builtin_ctzll(pkts_to_arp);
1561 /* bitmask representing only this packet */
1562 uint64_t pkt_mask = 1LLU << pos;
1563 /* remove this packet from remaining list */
1564 pkts_to_arp &= ~pkt_mask;
1566 int must_reverse = ((synproxy_reply_mask & pkt_mask) != 0);
1568 phy_port = pkt->port;
1569 meta_data_addr = (struct mbuf_tcp_meta_data *)
1570 RTE_MBUF_METADATA_UINT32_PTR(pkt, META_DATA_OFFSET);
1571 ehdr = rte_vfw_get_ether_addr(pkt);
1574 struct ipv4_hdr *ihdr = (struct ipv4_hdr *)
1575 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
1578 uint32_t dest_address = rte_bswap32(ihdr->dst_addr);
1580 ret = local_get_nh_ipv4(dest_address, &dest_if,
1583 rte_sp_exchange_mac_addresses(ehdr);
1584 if (is_phy_port_privte(phy_port)) {
1586 dest_if = get_pub_to_prv_port(
1589 if (dest_if == INVALID_DESTIF) {
1590 pkts_mask &= ~pkt_mask;
1591 vfw_pipe->counters->
1592 pkts_drop_without_arp_entry++;
1594 do_local_nh_ipv4_cache(
1600 dest_if = get_prv_to_pub_port(
1603 if (dest_if == INVALID_DESTIF) {
1604 pkts_mask &= ~pkt_mask;
1605 vfw_pipe->counters->
1606 pkts_drop_without_arp_entry++;
1608 do_local_nh_ipv4_cache(dest_if,
1612 } else if (is_phy_port_privte(phy_port)) {
1614 dest_if = get_prv_to_pub_port(&dest_address,
1616 if (dest_if == INVALID_DESTIF) {
1617 pkts_mask &= ~pkt_mask;
1618 vfw_pipe->counters->
1619 pkts_drop_without_arp_entry++;
1621 do_local_nh_ipv4_cache(dest_if, vfw_pipe);
1626 dest_if = get_pub_to_prv_port(&dest_address,
1628 if (dest_if == INVALID_DESTIF) {
1629 pkts_mask &= ~pkt_mask;
1630 vfw_pipe->counters->
1631 pkts_drop_without_arp_entry++;
1633 do_local_nh_ipv4_cache(dest_if, vfw_pipe);
1637 meta_data_addr->output_port = vfw_pipe->outport_id[dest_if];
1638 if (local_dest_mac_present(dest_if)) {
1639 ether_addr_copy(get_local_link_hw_addr(dest_if),
1641 ether_addr_copy(get_link_hw_addr(dest_if),
1644 ret_mac = get_dest_mac_addr_port(dest_address,
1645 &dest_if, &hw_addr);
1647 link_hw_laddr_valid[dest_if] = 1;
1648 memcpy(&link_hw_laddr[dest_if], &hw_addr,
1649 sizeof(struct ether_addr));
1651 ether_addr_copy(&hw_addr, &ehdr->d_addr);
1652 ether_addr_copy(get_link_hw_addr(dest_if),
1655 if (vfw_debug >= DEBUG_LEVEL_4) {
1656 char buf[HW_ADDR_SIZE];
1658 ether_format_addr(buf, sizeof(buf),
1660 printf("MAC found for ip 0x%"
1661 PRIx32", dest_if %d: %s, ",
1664 ether_format_addr(buf, sizeof(buf),
1666 printf("new eth hdr src: %s, ", buf);
1667 ether_format_addr(buf, sizeof(buf),
1669 printf("new eth hdr dst: %s\n", buf);
1673 if (unlikely(ret_mac == 0))
1674 request_arp(meta_data_addr->output_port,
1677 if (vfw_debug >= DEBUG_LEVEL_4) {
1678 char buf[HW_ADDR_SIZE];
1680 ether_format_addr(buf, sizeof(buf),
1682 printf("MAC NOT FOUND for ip 0x%"
1688 /* ICMP req sent, drop packet by
1689 * changing the mask */
1690 pkts_mask &= ~pkt_mask;
1692 counters->pkts_drop_without_arp_entry++;
1701 * walk every valid mbuf (denoted by pkts_mask) and apply arp to the packet.
1702 * To support synproxy, some (altered) packets may need to be sent back where
1703 * they came from. The ip header has already been adjusted, but the ethernet
1704 * header has not, so this must be performed here.
1705 * Return an updated pkts_mask, since arp may drop some packets
1708 * A pointer to the packet.
1711 * @param synproxy_reply_mask
1712 * Reply Packet mask for Synproxy
1714 * A pointer to VFW pipeline.
1718 rte_vfw_arp_ipv6_packets(struct rte_mbuf **pkts,
1720 uint64_t synproxy_reply_mask,
1721 struct pipeline_vfw *vfw_pipe)
1723 uint64_t pkts_to_arp = pkts_mask;
1724 uint8_t nh_ipv6[IPV6_ADD_SIZE];
1726 uint32_t dest_if = INVALID_DESTIF;
1728 for (; pkts_to_arp;) {
1729 struct ether_addr hw_addr;
1730 struct mbuf_tcp_meta_data *meta_data_addr;
1731 struct ether_hdr *ehdr;
1732 struct rte_mbuf *pkt;
1735 uint8_t pos = (uint8_t) __builtin_ctzll(pkts_to_arp);
1736 /* bitmask representing only this packet */
1737 uint64_t pkt_mask = 1LLU << pos;
1738 /* remove this packet from remaining list */
1739 pkts_to_arp &= ~pkt_mask;
1741 int must_reverse = ((synproxy_reply_mask & pkt_mask) != 0);
1743 phy_port = pkt->port;
1744 meta_data_addr = (struct mbuf_tcp_meta_data *)
1745 RTE_MBUF_METADATA_UINT32_PTR(pkt, META_DATA_OFFSET);
1746 ehdr = rte_vfw_get_ether_addr(pkt);
1748 struct ipv6_hdr *ihdr = (struct ipv6_hdr *)
1749 RTE_MBUF_METADATA_UINT32_PTR(pkt, IP_START);
1751 uint8_t nhip[IPV6_ADD_SIZE];
1752 uint8_t dest_address[IPV6_ADD_SIZE];
1754 memset(nhip, 0, IPV6_ADD_SIZE);
1756 rte_mov16(dest_address, ihdr->dst_addr);
1757 ret = local_get_nh_ipv6(&dest_address[0], &dest_if,
1758 &nhip[0], vfw_pipe);
1760 rte_sp_exchange_mac_addresses(ehdr);
1761 if (is_phy_port_privte(phy_port)) {
1763 dest_if = get_pub_to_prv_port(
1767 if (dest_if == INVALID_DESTIF) {
1768 pkts_mask &= ~pkt_mask;
1769 vfw_pipe->counters->
1770 pkts_drop_without_arp_entry++;
1772 do_local_nh_ipv6_cache(dest_if,
1778 dest_if = get_prv_to_pub_port(
1782 if (dest_if == INVALID_DESTIF) {
1783 pkts_mask &= ~pkt_mask;
1784 vfw_pipe->counters->
1785 pkts_drop_without_arp_entry++;
1787 do_local_nh_ipv6_cache(dest_if,
1792 } else if (is_phy_port_privte(phy_port)) {
1794 dest_if = get_prv_to_pub_port((uint32_t *)
1795 &dest_address[0], IP_VERSION_6);
1796 if (dest_if == INVALID_DESTIF) {
1797 pkts_mask &= ~pkt_mask;
1798 vfw_pipe->counters->
1799 pkts_drop_without_arp_entry++;
1801 do_local_nh_ipv6_cache(dest_if, vfw_pipe);
1806 dest_if = get_pub_to_prv_port((uint32_t *)
1807 &dest_address[0], IP_VERSION_6);
1808 if (dest_if == INVALID_DESTIF) {
1809 pkts_mask &= ~pkt_mask;
1810 vfw_pipe->counters->
1811 pkts_drop_without_arp_entry++;
1814 do_local_nh_ipv6_cache(dest_if, vfw_pipe);
1819 meta_data_addr->output_port = vfw_pipe->outport_id[dest_if];
1821 memset(nh_ipv6, 0, IPV6_ADD_SIZE);
1822 if (get_dest_mac_address_ipv6_port(
1827 ether_addr_copy(&hw_addr, &ehdr->d_addr);
1828 ether_addr_copy(get_link_hw_addr(dest_if),
1831 if (vfw_debug >= DEBUG_LEVEL_4) {
1832 char buf[HW_ADDR_SIZE];
1834 ether_format_addr(buf, sizeof(buf),
1836 printf("MAC found for dest_if %d: %s, ",
1838 ether_format_addr(buf, sizeof(buf),
1840 printf("new eth hdr src: %s, ", buf);
1841 ether_format_addr(buf, sizeof(buf),
1843 printf("new eth hdr dst: %s\n", buf);
1847 printf("deleting ipv6\n");
1848 pkts_mask &= ~pkt_mask;
1849 /*Next Neighbor is not yet implemented
1851 vfw_pipe->counters->
1852 pkts_drop_without_arp_entry++;
1863 * Packets processing for connection tracking.
1866 * A pointer to the pipeline.
1868 * A pointer to the connetion tracker .
1870 * A pointer to a burst of packets.
1871 * @param packet_mask_in
1872 * Input packets Mask.
1876 vfw_process_buffered_pkts(__rte_unused struct pipeline_vfw *vfw_pipe,
1877 struct rte_ct_cnxn_tracker *ct,
1878 struct rte_mbuf **pkts, uint64_t packet_mask_in)
1880 uint64_t keep_mask = packet_mask_in;
1881 struct rte_synproxy_helper sp_helper; /* for synproxy */
1884 rte_ct_cnxn_tracker_batch_lookup_with_synproxy(ct, pkts, keep_mask,
1887 if (unlikely(sp_helper.hijack_mask))
1888 printf("buffered hijack pkts severe error\n");
1890 if (unlikely(sp_helper.reply_pkt_mask))
1891 printf("buffered reply pkts severe error\n");
1897 * Free Packets from mbuf.
1900 * A pointer to the connection tracker to increment drop counter.
1903 * Packet to be free.
1906 vfw_pktmbuf_free(struct rte_ct_cnxn_tracker *ct, struct rte_mbuf *pkt)
1908 ct->counters->pkts_drop++;
1909 rte_pktmbuf_free(pkt);
1913 vfw_output_or_delete_buffered_packets(struct rte_ct_cnxn_tracker *ct,
1914 struct rte_pipeline *p,
1915 struct rte_mbuf **pkts,
1916 int num_pkts, uint64_t pkts_mask)
1919 struct mbuf_tcp_meta_data *meta_data_addr;
1920 uint64_t pkt_mask = 1;
1922 /* any clear bits in low-order num_pkts bit of
1923 * pkt_mask must be discarded */
1925 for (i = 0; i < num_pkts; i++) {
1926 struct rte_mbuf *pkt = pkts[i];
1928 if (pkts_mask & pkt_mask) {
1929 printf("vfw_output_or_delete_buffered_packets\n");
1930 meta_data_addr = (struct mbuf_tcp_meta_data *)
1931 RTE_MBUF_METADATA_UINT32_PTR(pkt, META_DATA_OFFSET);
1932 rte_pipeline_port_out_packet_insert(
1933 p, meta_data_addr->output_port, pkt);
1936 vfw_pktmbuf_free(ct, pkt);
1939 pkt_mask = pkt_mask << 1;
1944 *Packet buffered for synproxy.
1947 * A pointer to the pipeline.
1949 * A pointer to the vfw pipeline.
1951 * A pointer to the connection tracker.
1952 * @param forward_pkts
1953 * Packet forwarded by synproxy.
1957 vfw_handle_buffered_packets(struct rte_pipeline *p,
1958 struct pipeline_vfw *vfw_pipe,
1959 struct rte_ct_cnxn_tracker *ct, int forward_pkts)
1961 struct rte_mbuf *pkt_list = rte_ct_get_buffered_synproxy_packets(ct);
1963 if (likely(pkt_list == NULL)) /* only during proxy setup is != NULL */
1967 uint64_t keep_mask = 0;
1968 struct rte_mbuf **pkts = vfw_pipe->pkt_buffer;
1969 struct rte_mbuf *pkt;
1971 while (pkt_list != NULL) {
1972 struct mbuf_tcp_meta_data *meta_data =
1973 (struct mbuf_tcp_meta_data *)
1974 RTE_MBUF_METADATA_UINT32_PTR(pkt_list, META_DATA_OFFSET);
1976 /* detach head of list and advance list */
1978 pkt_list = meta_data->next;
1982 pkts[pkt_count++] = pkt;
1984 if (pkt_count == PKT_BUFFER_SIZE) {
1985 /* need to send out packets */
1986 /* currently 0, set all bits */
1987 keep_mask = ~keep_mask;
1990 vfw_process_buffered_pkts(vfw_pipe,
1993 vfw_output_or_delete_buffered_packets(
2003 vfw_pktmbuf_free(ct, pkt);
2007 if (pkt_count != 0) {
2008 /* need to send out packets */
2009 keep_mask = RTE_LEN2MASK(pkt_count, uint64_t);
2012 vfw_process_buffered_pkts(vfw_pipe, ct, pkts,
2015 vfw_output_or_delete_buffered_packets(ct, p, pkts, pkt_count,
2023 * The pipeline port-in action is used to do all the firewall and
2024 * connection tracking work for IPV4 packets.
2027 * A pointer to the pipeline.
2029 * A pointer to a burst of packets.
2031 * Number of packets to process.
2033 * A pointer to pipeline specific data.
2036 * 0 on success, negative on error.
2040 vfw_port_in_action_ipv4(struct rte_pipeline *p,
2041 struct rte_mbuf **pkts,
2042 __rte_unused uint32_t n_pkts, __rte_unused void *arg)
2044 struct vfw_ports_in_args *port_in_args =
2045 (struct vfw_ports_in_args *)arg;
2046 struct pipeline_vfw *vfw_pipe =
2047 (struct pipeline_vfw *)port_in_args->pipe;
2048 struct rte_ct_cnxn_tracker *ct = port_in_args->cnxn_tracker;
2050 start_tsc_measure(vfw_pipe);
2052 uint64_t packet_mask_in = RTE_LEN2MASK(n_pkts, uint64_t);
2053 uint64_t pkts_drop_mask;
2054 uint64_t hijack_mask = 0;
2055 uint64_t synproxy_reply_mask = 0; /* for synproxy */
2056 uint64_t keep_mask = packet_mask_in;
2058 uint64_t conntrack_mask = 0, connexist_mask = 0;
2059 struct rte_CT_helper ct_helper;
2063 * This routine uses a bit mask to represent which packets in the
2064 * "pkts" table are considered valid. Any table entry which exists
2065 * and is considered valid has the corresponding bit in the mask set.
2066 * Otherwise, it is cleared. Note that the mask is 64 bits,
2067 * but the number of packets in the table may be considerably less.
2068 * Any mask bits which do correspond to actual packets are cleared.
2069 * Various routines are called which may determine that an existing
2070 * packet is somehow invalid. The routine will return an altered bit
2071 * mask, with the bit cleared. At the end of all the checks,
2072 * packets are dropped if their mask bit is a zero
2075 rte_prefetch0(& vfw_pipe->counters);
2078 /* Pre-fetch all rte_mbuf header */
2079 for(j = 0; j < n_pkts; j++)
2080 rte_prefetch0(pkts[j]);
2082 memset(&ct_helper, 0, sizeof(struct rte_CT_helper));
2084 rte_prefetch0(& vfw_pipe->counters->pkts_drop_ttl);
2085 rte_prefetch0(& vfw_pipe->counters->sum_latencies);
2088 if (unlikely(vfw_debug > 1))
2089 printf("Enter in-port action IPV4 with %p packet mask\n",
2090 (void *)packet_mask_in);
2091 vfw_pipe->counters->pkts_received =
2092 vfw_pipe->counters->pkts_received + n_pkts;
2094 if (unlikely(VFW_DEBUG))
2095 printf("vfw_port_in_action_ipv4 pkts_received: %" PRIu64
2097 vfw_pipe->counters->pkts_received, n_pkts);
2099 /* first handle handle any previously buffered packets now released */
2100 vfw_handle_buffered_packets(p, vfw_pipe, ct,
2101 FORWARD_BUFFERED_PACKETS);
2103 /* now handle any new packets on input ports */
2104 if (likely(firewall_flag)) {
2105 keep_mask = rte_vfw_ipv4_packet_filter_and_process(pkts,
2106 keep_mask, vfw_pipe);
2107 vfw_pipe->counters->pkts_fw_forwarded +=
2108 __builtin_popcountll(keep_mask);
2112 rte_prefetch0((void*)vfw_pipe->plib_acl);
2113 rte_prefetch0((void*)vfw_rule_table_ipv4_active);
2114 #endif /* EN_SWP_ACL */
2115 keep_mask = lib_acl_ipv4_pkt_work_key(
2116 vfw_pipe->plib_acl, pkts, keep_mask,
2117 &vfw_pipe->counters->pkts_drop_without_rule,
2118 vfw_rule_table_ipv4_active,
2119 action_array_active,
2120 action_counter_table,
2121 &conntrack_mask, &connexist_mask);
2122 vfw_pipe->counters->pkts_acl_forwarded +=
2123 __builtin_popcountll(keep_mask);
2124 if (conntrack_mask > 0) {
2125 keep_mask = conntrack_mask;
2126 ct_helper.no_new_cnxn_mask = connexist_mask;
2127 cnxn_tracking_is_active = 1;
2129 cnxn_tracking_is_active = 0;
2130 #endif /* ACL_ENABLE */
2132 if (likely(cnxn_tracking_is_active)) {
2133 rte_ct_cnxn_tracker_batch_lookup_type(ct, pkts,
2134 &keep_mask, &ct_helper, IPv4_HEADER_SIZE);
2135 synproxy_reply_mask = ct_helper.reply_pkt_mask;
2136 hijack_mask = ct_helper.hijack_mask;
2141 for(j = 0; j < (n_pkts & 0x3LLU); j++) {
2142 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2144 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2147 rte_prefetch0((void*)in_port_dir_a);
2148 rte_prefetch0((void*)prv_to_pub_map);
2151 for (i = 0; i < (n_pkts & (~0x3LLU)); i += 4) {
2152 for (j = i+4; ((j < n_pkts) && (j < i+8)); j++) {
2153 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2155 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2158 pkt4_work_vfw_arp_ipv4_packets(&pkts[i], i, &keep_mask,
2159 synproxy_reply_mask, vfw_pipe);
2161 for (j = i; j < n_pkts; j++) {
2162 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2164 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2167 for (; i < n_pkts; i++) {
2168 pkt_work_vfw_arp_ipv4_packets(pkts[i], i, &keep_mask,
2169 synproxy_reply_mask, vfw_pipe);
2172 rte_prefetch0((void*)in_port_dir_a);
2173 rte_prefetch0((void*)prv_to_pub_map);
2174 rte_prefetch0((void*) & vfw_pipe->local_lib_arp_route_table);
2175 keep_mask = rte_vfw_arp_ipv4_packets(pkts, keep_mask,
2176 synproxy_reply_mask, vfw_pipe);
2179 if (vfw_debug > 1) {
2180 printf(" Exit in-port action with %p packet mask\n",
2182 if (keep_mask != packet_mask_in)
2183 printf("dropped packets, %p in, %p out\n",
2184 (void *)packet_mask_in,
2188 /* Update mask before returning, so that bad packets are dropped */
2190 pkts_drop_mask = packet_mask_in & ~keep_mask;
2192 if (unlikely(pkts_drop_mask != 0)) {
2193 /* printf("drop %p\n", (void *) pkts_drop_mask); */
2194 rte_pipeline_ah_packet_drop(p, pkts_drop_mask);
2197 if (unlikely(hijack_mask != 0))
2198 rte_pipeline_ah_packet_hijack(p, hijack_mask);
2200 vfw_pipe->counters->num_batch_pkts_sum += n_pkts;
2201 vfw_pipe->counters->num_pkts_measurements++;
2203 end_tsc_measure(vfw_pipe, n_pkts);
2208 * The pipeline port-in action is used to do all the firewall and
2209 * connection tracking work for IPV6 packet.
2212 * A pointer to the pipeline.
2214 * A pointer to a burst of packets.
2216 * Number of packets to process.
2218 * A pointer to pipeline specific data.
2221 * 0 on success, negative on error.
2225 vfw_port_in_action_ipv6(struct rte_pipeline *p,
2226 struct rte_mbuf **pkts,
2227 __rte_unused uint32_t n_pkts, __rte_unused void *arg)
2229 struct vfw_ports_in_args *port_in_args =
2230 (struct vfw_ports_in_args *)arg;
2231 struct pipeline_vfw *vfw_pipe =
2232 (struct pipeline_vfw *)port_in_args->pipe;
2233 struct rte_ct_cnxn_tracker *ct = port_in_args->cnxn_tracker;
2235 start_tsc_measure(vfw_pipe);
2237 uint64_t packet_mask_in = RTE_LEN2MASK(n_pkts, uint64_t);
2238 uint64_t pkts_drop_mask;
2239 uint64_t hijack_mask = 0;
2240 uint64_t synproxy_reply_mask = 0; /* for synproxy */
2241 uint64_t keep_mask = packet_mask_in;
2243 uint64_t conntrack_mask = 0, connexist_mask = 0;
2244 struct rte_CT_helper ct_helper;
2248 * This routine uses a bit mask to represent which packets in the
2249 * "pkts" table are considered valid. Any table entry which exists
2250 * and is considered valid has the corresponding bit in the mask set.
2251 * Otherwise, it is cleared. Note that the mask is 64 bits,
2252 * but the number of packets in the table may be considerably less.
2253 * Any mask bits which do correspond to actual packets are cleared.
2254 * Various routines are called which may determine that an existing
2255 * packet is somehow invalid. The routine will return an altered bit
2256 * mask, with the bit cleared. At the end of all the checks,
2257 * packets are dropped if their mask bit is a zero
2260 rte_prefetch0(& vfw_pipe->counters);
2262 /* Pre-fetch all rte_mbuf header */
2263 for(j = 0; j < n_pkts; j++)
2264 rte_prefetch0(pkts[j]);
2266 memset(&ct_helper, 0, sizeof(struct rte_CT_helper));
2267 rte_prefetch0(& vfw_pipe->counters->pkts_drop_ttl);
2268 rte_prefetch0(& vfw_pipe->counters->sum_latencies);
2271 printf("Enter in-port action with %p packet mask\n",
2272 (void *)packet_mask_in);
2273 vfw_pipe->counters->pkts_received =
2274 vfw_pipe->counters->pkts_received + n_pkts;
2276 printf("vfw_port_in_action pkts_received: %" PRIu64
2278 vfw_pipe->counters->pkts_received, n_pkts);
2280 /* first handle handle any previously buffered packets now released */
2281 vfw_handle_buffered_packets(p, vfw_pipe, ct,
2282 FORWARD_BUFFERED_PACKETS);
2284 /* now handle any new packets on input ports */
2285 if (likely(firewall_flag)) {
2286 keep_mask = rte_vfw_ipv6_packet_filter_and_process(pkts,
2287 keep_mask, vfw_pipe);
2288 vfw_pipe->counters->pkts_fw_forwarded +=
2289 __builtin_popcountll(keep_mask);
2294 rte_prefetch0((void*)vfw_pipe->plib_acl);
2295 rte_prefetch0((void*)vfw_rule_table_ipv6_active);
2296 #endif /* EN_SWP_ACL */
2297 keep_mask = lib_acl_ipv6_pkt_work_key(
2298 vfw_pipe->plib_acl, pkts, keep_mask,
2299 &vfw_pipe->counters->pkts_drop_without_rule,
2300 vfw_rule_table_ipv6_active,
2301 action_array_active,
2302 action_counter_table,
2303 &conntrack_mask, &connexist_mask);
2304 vfw_pipe->counters->pkts_acl_forwarded +=
2305 __builtin_popcountll(keep_mask);
2306 if (conntrack_mask > 0) {
2307 keep_mask = conntrack_mask;
2308 ct_helper.no_new_cnxn_mask = connexist_mask;
2309 cnxn_tracking_is_active = 1;
2311 cnxn_tracking_is_active = 0;
2312 #endif /* ACL_ENABLE */
2313 if (likely(cnxn_tracking_is_active)) {
2314 rte_ct_cnxn_tracker_batch_lookup_type(ct, pkts,
2315 &keep_mask, &ct_helper, IPv6_HEADER_SIZE);
2316 synproxy_reply_mask = ct_helper.reply_pkt_mask;
2317 hijack_mask = ct_helper.hijack_mask;
2322 for(j = 0; j < (n_pkts & 0x3LLU); j++) {
2323 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2325 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2328 rte_prefetch0((void*)in_port_dir_a);
2329 rte_prefetch0(vfw_pipe->local_lib_nd_route_table);
2332 for (i = 0; i < (n_pkts & (~0x3LLU)); i += 4) {
2333 for (j = i+4; ((j < n_pkts) && (j < i+8)); j++) {
2334 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2336 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2339 pkt4_work_vfw_arp_ipv6_packets(&pkts[i], i, &keep_mask,
2340 synproxy_reply_mask, vfw_pipe);
2342 for (j = i; j < n_pkts; j++) {
2343 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2345 rte_prefetch0(RTE_MBUF_METADATA_UINT32_PTR(pkts[j],
2348 for (; i < n_pkts; i++) {
2349 pkt_work_vfw_arp_ipv6_packets(pkts[i], i, &keep_mask,
2350 synproxy_reply_mask, vfw_pipe);
2353 rte_prefetch0((void*)in_port_dir_a);
2354 rte_prefetch0((void*) & vfw_pipe->local_lib_arp_route_table);
2355 keep_mask = rte_vfw_arp_ipv6_packets(pkts, keep_mask,
2356 synproxy_reply_mask, vfw_pipe);
2359 if (vfw_debug > 1) {
2360 printf(" Exit in-port action with %p packet mask\n",
2362 if (keep_mask != packet_mask_in)
2363 printf("dropped packets, %p in, %p out\n",
2364 (void *)packet_mask_in,
2368 /* Update mask before returning, so that bad packets are dropped */
2370 pkts_drop_mask = packet_mask_in & ~keep_mask;
2372 if (unlikely(pkts_drop_mask != 0)) {
2373 /* printf("drop %p\n", (void *) pkts_drop_mask); */
2374 rte_pipeline_ah_packet_drop(p, pkts_drop_mask);
2377 if (unlikely(hijack_mask != 0))
2378 rte_pipeline_ah_packet_hijack(p, hijack_mask);
2380 vfw_pipe->counters->num_batch_pkts_sum += n_pkts;
2381 vfw_pipe->counters->num_pkts_measurements++;
2383 end_tsc_measure(vfw_pipe, n_pkts);
2390 * Parse arguments in config file.
2393 * A pointer to the pipeline.
2395 * A pointer to pipeline specific parameters.
2398 * 0 on success, negative on error.
2401 pipeline_vfw_parse_args(struct pipeline_vfw *vfw_pipe,
2402 struct pipeline_params *params)
2408 printf("VFW pipeline_vfw_parse_args params->n_args: %d\n",
2411 for (i = 0; i < params->n_args; i++) {
2412 char *arg_name = params->args_name[i];
2413 char *arg_value = params->args_value[i];
2415 printf("VFW args[%d]: %s %d, %s\n", i, arg_name,
2416 atoi(arg_value), arg_value);
2418 status = lib_acl_parse_config(vfw_pipe->plib_acl,
2419 arg_name, arg_value, &vfw_n_rules);
2421 printf("rte_ct_set_configuration_options =%s,%s",
2422 arg_name, arg_value);
2424 } else if (status == 0)
2427 #endif /* traffic_type */
2428 if (strcmp(arg_name, "traffic_type") == 0) {
2429 int traffic_type = atoi(arg_value);
2431 if (traffic_type == 0 ||
2432 !(traffic_type == IP_VERSION_4 ||
2433 traffic_type == IP_VERSION_6)) {
2434 printf("not IPV4/IPV6");
2438 vfw_pipe->traffic_type = traffic_type;
2444 if (strcmp(arg_name, "n_flows") == 0) {
2445 int n_flows = atoi(arg_value);
2450 /* must be power of 2, round up if not */
2451 if (!rte_is_power_of_2(n_flows))
2452 n_flows = rte_align32pow2(n_flows);
2454 vfw_pipe->n_flows = n_flows;
2458 /* not firewall option, process as cnxn tracking option */
2459 status = rte_ct_set_configuration_options(
2460 vfw_pipe->cnxn_tracker,
2461 arg_name, arg_value);
2463 printf("rte_ct_set_configuration_options =%s,%s",
2464 arg_name, arg_value);
2466 } else if (status == 0)
2474 static void *pipeline_vfw_msg_req_custom_handler(struct pipeline *p,
2477 static pipeline_msg_req_handler handlers[] = {
2478 [PIPELINE_MSG_REQ_PING] = pipeline_msg_req_ping_handler,
2479 [PIPELINE_MSG_REQ_STATS_PORT_IN] =
2480 pipeline_msg_req_stats_port_in_handler,
2481 [PIPELINE_MSG_REQ_STATS_PORT_OUT] =
2482 pipeline_msg_req_stats_port_out_handler,
2483 [PIPELINE_MSG_REQ_STATS_TABLE] = pipeline_msg_req_stats_table_handler,
2484 [PIPELINE_MSG_REQ_PORT_IN_ENABLE] =
2485 pipeline_msg_req_port_in_enable_handler,
2486 [PIPELINE_MSG_REQ_PORT_IN_DISABLE] =
2487 pipeline_msg_req_port_in_disable_handler,
2488 [PIPELINE_MSG_REQ_CUSTOM] = pipeline_vfw_msg_req_custom_handler,
2491 static void *pipeline_vfw_msg_req_synproxy_flag_handler(struct pipeline *p,
2493 static pipeline_msg_req_handler custom_handlers[] = {
2495 [PIPELINE_VFW_MSG_REQ_SYNPROXY_FLAGS] =
2496 pipeline_vfw_msg_req_synproxy_flag_handler
2500 * Create and initialize Pipeline Back End (BE).
2503 * A pointer to the pipeline specific parameters..
2505 * A pointer to pipeline specific data.
2508 * A pointer to the pipeline create, NULL on error.
2511 *pipeline_vfw_init(struct pipeline_params *params, __rte_unused void *arg)
2515 /* Check input arguments */
2516 if ((params == NULL) ||
2517 (params->n_ports_in == 0) || (params->n_ports_out == 0))
2521 printf("num ports in %d / num ports out %d\n",
2522 params->n_ports_in, params->n_ports_out);
2524 /* Create a single pipeline instance and initialize. */
2525 struct pipeline_vfw *pipe_vfw;
2527 size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct pipeline_vfw));
2528 pipe_vfw = rte_zmalloc(NULL, size, RTE_CACHE_LINE_SIZE);
2530 if (pipe_vfw == NULL)
2533 struct pipeline *pipe;
2535 pipe = &pipe_vfw->pipe;
2537 strncpy(pipe->name, params->name, sizeof(pipe->name));
2538 pipe->log_level = params->log_level;
2539 pipe_vfw->n_flows = 4096; /* small default value */
2540 pipe_vfw->traffic_type = MIX;
2541 pipe_vfw->pipeline_num = 0xff;
2542 for (i = 0; i < PIPELINE_MAX_PORT_IN; i++) {
2543 pipe_vfw->links_map[i] = 0xff;
2544 pipe_vfw->outport_id[i] = 0xff;
2546 PLOG(pipe, HIGH, "VFW");
2548 /* Create a firewall instance and initialize. */
2549 pipe_vfw->cnxn_tracker =
2550 rte_zmalloc(NULL, rte_ct_get_cnxn_tracker_size(),
2551 RTE_CACHE_LINE_SIZE);
2553 if (pipe_vfw->cnxn_tracker == NULL)
2556 /* Create a acl instance and initialize. */
2557 pipe_vfw->plib_acl =
2558 rte_zmalloc(NULL, sizeof(struct lib_acl),
2559 RTE_CACHE_LINE_SIZE);
2561 if (pipe_vfw->plib_acl == NULL)
2564 timer_lcore = rte_lcore_id();
2566 * Now allocate a counter block entry. It appears that the
2567 * initialization of all instances is serialized on core 0,
2568 * so no lock is necessary.
2570 struct rte_VFW_counter_block *counter_ptr;
2572 if (rte_VFW_hi_counter_block_in_use == MAX_VFW_INSTANCES)
2573 /* error, exceeded table bounds */
2576 rte_VFW_hi_counter_block_in_use++;
2578 &rte_vfw_counter_table[rte_VFW_hi_counter_block_in_use];
2579 strncpy(counter_ptr->name, params->name, sizeof(counter_ptr->name));
2581 pipe_vfw->counters = counter_ptr;
2583 rte_ct_initialize_default_timeouts(pipe_vfw->cnxn_tracker);
2584 /* Parse arguments */
2585 if (pipeline_vfw_parse_args(pipe_vfw, params))
2588 uint16_t pointers_offset =
2589 META_DATA_OFFSET + offsetof(struct mbuf_tcp_meta_data, next);
2591 if (pipe_vfw->n_flows > 0)
2592 rte_ct_initialize_cnxn_tracker_with_synproxy(
2593 pipe_vfw->cnxn_tracker,
2598 pipe_vfw->counters->ct_counters =
2599 rte_ct_get_counter_address(pipe_vfw->cnxn_tracker);
2603 struct rte_pipeline_params pipeline_params = {
2604 .name = params->name,
2605 .socket_id = params->socket_id,
2606 .offset_port_id = META_DATA_OFFSET +
2607 offsetof(struct mbuf_tcp_meta_data, output_port)
2610 pipe->p = rte_pipeline_create(&pipeline_params);
2611 if (pipe->p == NULL) {
2620 * create a different "arg_ah" for each input port.
2621 * They differ only in the recorded port number. Unfortunately,
2622 * IP_PIPELINE does not pass port number in to input port handler
2625 uint32_t in_ports_arg_size =
2626 RTE_CACHE_LINE_ROUNDUP((sizeof(struct vfw_ports_in_args)) *
2627 (params->n_ports_in));
2628 struct vfw_ports_in_args *port_in_args =
2629 (struct vfw_ports_in_args *)
2630 rte_zmalloc(NULL, in_ports_arg_size, RTE_CACHE_LINE_SIZE);
2632 if (port_in_args == NULL)
2635 pipe->n_ports_in = params->n_ports_in;
2636 for (i = 0; i < pipe->n_ports_in; i++) {
2638 /* initialize this instance of port_in_args as necessary */
2639 port_in_args[i].pipe = pipe;
2640 port_in_args[i].cnxn_tracker = pipe_vfw->cnxn_tracker;
2642 struct rte_pipeline_port_in_params port_params = {
2644 pipeline_port_in_params_get_ops(¶ms->port_in
2647 pipeline_port_in_params_convert(¶ms->port_in
2649 .f_action = vfw_port_in_action_ipv4,
2650 .arg_ah = &(port_in_args[i]),
2651 .burst_size = params->port_in[i].burst_size,
2653 if (pipe_vfw->traffic_type == IP_VERSION_6)
2654 port_params.f_action = vfw_port_in_action_ipv6;
2655 int status = rte_pipeline_port_in_create(pipe->p, &port_params,
2656 &pipe->port_in_id[i]);
2659 rte_pipeline_free(pipe->p);
2666 pipe->n_ports_out = params->n_ports_out;
2667 for (i = 0; i < pipe->n_ports_out; i++) {
2668 struct rte_pipeline_port_out_params port_params = {
2669 .ops = pipeline_port_out_params_get_ops(
2670 ¶ms->port_out[i]),
2671 .arg_create = pipeline_port_out_params_convert(
2672 ¶ms->port_out[i]),
2677 int status = rte_pipeline_port_out_create(pipe->p, &port_params,
2678 &pipe->port_out_id[i]);
2681 rte_pipeline_free(pipe->p);
2687 int pipeline_num = 0;
2688 int dont_care = sscanf(params->name, "PIPELINE%d", &pipeline_num);
2691 printf("sscanf unble to read pipeline id\n");
2692 pipe_vfw->pipeline_num = (uint8_t) pipeline_num;
2693 register_pipeline_Qs(pipe_vfw->pipeline_num, pipe);
2694 set_link_map(pipe_vfw->pipeline_num, pipe, pipe_vfw->links_map);
2695 set_outport_id(pipe_vfw->pipeline_num, pipe,
2696 pipe_vfw->outport_id);
2697 printf("pipeline_num=%d\n", pipeline_num);
2699 /*If this is the first VFW thread, create common VFW Rule tables*/
2700 if (rte_VFW_hi_counter_block_in_use == 0) {
2701 vfw_rule_table_ipv4_active =
2702 lib_acl_create_active_standby_table_ipv4(1,
2704 if (vfw_rule_table_ipv4_active == NULL) {
2705 printf("Failed to create active table for IPV4\n");
2706 rte_pipeline_free(pipe->p);
2707 rte_free(pipe_vfw->cnxn_tracker);
2708 rte_free(pipe_vfw->plib_acl);
2712 vfw_rule_table_ipv4_standby =
2713 lib_acl_create_active_standby_table_ipv4(2,
2715 if (vfw_rule_table_ipv4_standby == NULL) {
2716 printf("Failed to create standby table for IPV4\n");
2717 rte_pipeline_free(pipe->p);
2718 rte_free(pipe_vfw->cnxn_tracker);
2719 rte_free(pipe_vfw->plib_acl);
2724 vfw_rule_table_ipv6_active =
2725 lib_acl_create_active_standby_table_ipv6(1,
2728 if (vfw_rule_table_ipv6_active == NULL) {
2729 printf("Failed to create active table for IPV6\n");
2730 rte_pipeline_free(pipe->p);
2731 rte_free(pipe_vfw->cnxn_tracker);
2732 rte_free(pipe_vfw->plib_acl);
2736 vfw_rule_table_ipv6_standby =
2737 lib_acl_create_active_standby_table_ipv6(2,
2739 if (vfw_rule_table_ipv6_standby == NULL) {
2740 printf("Failed to create standby table for IPV6\n");
2741 rte_pipeline_free(pipe->p);
2742 rte_free(pipe_vfw->cnxn_tracker);
2743 rte_free(pipe_vfw->plib_acl);
2755 struct rte_pipeline_table_params table_params = {
2756 .ops = &rte_table_stub_ops,
2758 .f_action_hit = NULL,
2759 .f_action_miss = NULL,
2761 .action_data_size = 0,
2764 int status = rte_pipeline_table_create(pipe->p,
2766 &pipe->table_id[0]);
2769 rte_pipeline_free(pipe->p);
2774 struct rte_pipeline_table_entry default_entry = {
2775 .action = RTE_PIPELINE_ACTION_PORT_META
2778 struct rte_pipeline_table_entry *default_entry_ptr;
2780 status = rte_pipeline_table_default_entry_add(pipe->p,
2783 &default_entry_ptr);
2786 rte_pipeline_free(pipe->p);
2790 for (i = 0; i < pipe->n_ports_in; i++) {
2791 int status = rte_pipeline_port_in_connect_to_table(
2793 pipe->port_in_id[i],
2797 rte_pipeline_free(pipe->p);
2803 /* Enable input ports */
2804 for (i = 0; i < pipe->n_ports_in; i++) {
2806 rte_pipeline_port_in_enable(pipe->p, pipe->port_in_id[i]);
2809 rte_pipeline_free(pipe->p);
2815 /* Check pipeline consistency */
2816 if (rte_pipeline_check(pipe->p) < 0) {
2817 rte_pipeline_free(pipe->p);
2822 /* Message queues */
2823 pipe->n_msgq = params->n_msgq;
2824 for (i = 0; i < pipe->n_msgq; i++)
2825 pipe->msgq_in[i] = params->msgq_in[i];
2827 for (i = 0; i < pipe->n_msgq; i++)
2828 pipe->msgq_out[i] = params->msgq_out[i];
2830 /* Message handlers */
2831 memcpy(pipe->handlers, handlers, sizeof(pipe->handlers));
2832 memcpy(pipe_vfw->custom_handlers, custom_handlers,
2833 sizeof(pipe_vfw->custom_handlers));
2839 * Free resources and delete pipeline.
2842 * A pointer to the pipeline.
2845 * 0 on success, negative on error.
2847 static int pipeline_vfw_free(void *pipeline)
2849 struct pipeline *p = (struct pipeline *)pipeline;
2851 /* Check input arguments */
2855 /* Free resources */
2856 rte_pipeline_free(p->p);
2862 * Callback function to map input/output ports.
2865 * A pointer to the pipeline.
2869 * A pointer to the Output port.
2872 * 0 on success, negative on error.
2875 pipeline_vfw_track(void *pipeline, __rte_unused uint32_t port_in,
2878 struct pipeline *p = (struct pipeline *)pipeline;
2880 /* Check input arguments */
2881 if ((p == NULL) || (port_in >= p->n_ports_in) || (port_out == NULL))
2884 if (p->n_ports_in == 1) {
2893 * Callback function to process timers.
2896 * A pointer to the pipeline.
2899 * 0 on success, negative on error.
2901 static int pipeline_vfw_timer(void *pipeline)
2903 struct pipeline_vfw *p = (struct pipeline_vfw *)pipeline;
2906 * handle any good buffered packets released by synproxy before checking
2907 * for packets relased by synproxy due to timeout.
2908 * Don't want packets missed
2911 vfw_handle_buffered_packets(p->pipe.p, p, p->cnxn_tracker,
2912 FORWARD_BUFFERED_PACKETS);
2914 pipeline_msg_req_handle(&p->pipe);
2915 rte_pipeline_flush(p->pipe.p);
2917 rte_ct_handle_expired_timers(p->cnxn_tracker);
2919 /* now handle packets released by synproxy due to timeout. */
2920 vfw_handle_buffered_packets(p->pipe.p, p, p->cnxn_tracker,
2921 DELETE_BUFFERED_PACKETS);
2927 * Callback function to process CLI commands from FE.
2930 * A pointer to the pipeline.
2932 * A pointer to command specific data.
2935 * A pointer to message handler on success,
2936 * pipeline_msg_req_invalid_hander on error.
2938 void *pipeline_vfw_msg_req_custom_handler(struct pipeline *p, void *msg)
2940 struct pipeline_vfw *pipe_vfw = (struct pipeline_vfw *)p;
2941 struct pipeline_custom_msg_req *req = msg;
2942 pipeline_msg_req_handler f_handle;
2944 f_handle = (req->subtype < PIPELINE_VFW_MSG_REQS) ?
2945 pipe_vfw->custom_handlers[req->subtype] :
2946 pipeline_msg_req_invalid_handler;
2948 if (f_handle == NULL)
2949 f_handle = pipeline_msg_req_invalid_handler;
2951 return f_handle(p, req);
2955 * Handler for synproxy ON/OFF CLI command.
2958 * A pointer to the pipeline.
2960 * A pointer to command specific data.
2963 * Response message contains status.
2966 void *pipeline_vfw_msg_req_synproxy_flag_handler(struct pipeline *p,
2969 struct pipeline_vfw *pipe_vfw = (struct pipeline_vfw *)p;
2970 struct pipeline_vfw_synproxy_flag_msg_req *req = msg;
2971 struct pipeline_vfw_synproxy_flag_msg_rsp *rsp = msg;
2973 if (req->synproxy_flag == 0) {
2974 rte_ct_disable_synproxy(pipe_vfw->cnxn_tracker);
2976 printf("synproxy turned OFF for %s\n", p->name);
2977 } else if (req->synproxy_flag == 1) {
2978 rte_ct_enable_synproxy(pipe_vfw->cnxn_tracker);
2980 printf("synproxy turned ON for %s\n", p->name);
2982 printf("Invalid synproxy setting\n");
2989 struct pipeline_be_ops pipeline_vfw_be_ops = {
2990 .f_init = pipeline_vfw_init,
2991 .f_free = pipeline_vfw_free,
2993 .f_timer = pipeline_vfw_timer,
2994 .f_track = pipeline_vfw_track,