1 /**************************************************************************
3 Copyright (c) 2008-2010, BitGravity Inc.
6 Redistribution and use in source and binary forms, with or without
7 modification, are permitted provided that the following conditions are met:
9 1. Redistributions of source code must retain the above copyright notice,
10 this list of conditions and the following disclaimer.
12 2. Neither the name of the BitGravity Corporation nor the names of its
13 contributors may be used to endorse or promote products derived from
14 this software without specific prior written permission.
16 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
17 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
20 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26 POSSIBILITY OF SUCH DAMAGE.
28 ***************************************************************************/
30 #include "opt_route.h"
31 #include "opt_mpath.h"
34 #include "opt_inet6.h"
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/bitstring.h>
42 #include <sys/condvar.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/kthread.h>
46 #include <sys/limits.h>
47 #include <sys/malloc.h>
51 #include <sys/sched.h>
53 #include <sys/socket.h>
54 #include <sys/syslog.h>
55 #include <sys/sysctl.h>
58 #include <net/if_llatbl.h>
59 #include <net/if_var.h>
60 #include <net/route.h>
61 #include <net/flowtable.h>
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/if_ether.h>
68 #include <netinet/ip.h>
70 #include <netinet/ip6.h>
72 #include <netinet/tcp.h>
73 #include <netinet/udp.h>
74 #include <netinet/sctp.h>
76 #include <libkern/jenkins.h>
80 uint16_t ip_sport; /* source port */
81 uint16_t ip_dport; /* destination port */
82 in_addr_t ip_saddr; /* source address */
83 in_addr_t ip_daddr; /* destination address */
87 struct ipv4_tuple ipf_ipt;
92 uint16_t ip_sport; /* source port */
93 uint16_t ip_dport; /* destination port */
94 struct in6_addr ip_saddr; /* source address */
95 struct in6_addr ip_daddr; /* destination address */
99 struct ipv6_tuple ipf_ipt;
104 volatile uint32_t f_fhash; /* hash flowing forward */
105 uint16_t f_flags; /* flow flags */
107 uint8_t f_proto; /* protocol */
108 uint32_t f_fibnum; /* fib index */
109 uint32_t f_uptime; /* uptime at last access */
110 struct flentry *f_next; /* pointer to collision entry */
111 volatile struct rtentry *f_rt; /* rtentry for flow */
112 volatile struct llentry *f_lle; /* llentry for flow */
116 struct flentry fl_entry;
117 union ipv4_flow fl_flow;
121 struct flentry fl_entry;
122 union ipv6_flow fl_flow;
125 #define fl_fhash fl_entry.fl_fhash
126 #define fl_flags fl_entry.fl_flags
127 #define fl_proto fl_entry.fl_proto
128 #define fl_uptime fl_entry.fl_uptime
129 #define fl_rt fl_entry.fl_rt
130 #define fl_lle fl_entry.fl_lle
132 #define SECS_PER_HOUR 3600
133 #define SECS_PER_DAY (24*SECS_PER_HOUR)
137 #define FIN_WAIT_IDLE 600
138 #define TCP_IDLE SECS_PER_DAY
141 typedef void fl_lock_t(struct flowtable *, uint32_t);
142 typedef void fl_rtalloc_t(struct route *, uint32_t, u_int);
145 struct flentry **global;
146 struct flentry **pcpu[MAXCPU];
149 struct flowtable_stats {
150 uint64_t ft_collisions;
151 uint64_t ft_allocated;
153 uint64_t ft_max_depth;
154 uint64_t ft_free_checks;
158 } __aligned(CACHE_LINE_SIZE);
161 struct flowtable_stats ft_stats[MAXCPU];
167 fl_lock_t *ft_unlock;
168 fl_rtalloc_t *ft_rtalloc;
170 * XXX need to pad out
172 struct mtx *ft_locks;
173 union flentryp ft_table;
174 bitstr_t *ft_masks[MAXCPU];
175 bitstr_t *ft_tmpmask;
176 struct flowtable *ft_next;
178 uint32_t ft_count __aligned(CACHE_LINE_SIZE);
179 uint32_t ft_udp_idle __aligned(CACHE_LINE_SIZE);
180 uint32_t ft_fin_wait_idle;
181 uint32_t ft_syn_idle;
182 uint32_t ft_tcp_idle;
184 } __aligned(CACHE_LINE_SIZE);
186 static struct proc *flowcleanerproc;
187 static VNET_DEFINE(struct flowtable *, flow_list_head);
188 static VNET_DEFINE(uint32_t, flow_hashjitter);
189 static VNET_DEFINE(uma_zone_t, flow_ipv4_zone);
190 static VNET_DEFINE(uma_zone_t, flow_ipv6_zone);
192 #define V_flow_list_head VNET(flow_list_head)
193 #define V_flow_hashjitter VNET(flow_hashjitter)
194 #define V_flow_ipv4_zone VNET(flow_ipv4_zone)
195 #define V_flow_ipv6_zone VNET(flow_ipv6_zone)
198 static struct cv flowclean_f_cv;
199 static struct cv flowclean_c_cv;
200 static struct mtx flowclean_lock;
201 static uint32_t flowclean_cycles;
202 static uint32_t flowclean_freq;
204 #ifdef FLOWTABLE_DEBUG
205 #define FLDPRINTF(ft, flags, fmt, ...) \
207 if ((ft)->ft_flags & (flags)) \
208 printf((fmt), __VA_ARGS__); \
212 #define FLDPRINTF(ft, flags, fmt, ...)
219 * - Make flowtable stats per-cpu, aggregated at sysctl call time,
220 * to avoid extra cache evictions caused by incrementing a shared
222 * - add sysctls to resize && flush flow tables
223 * - Add per flowtable sysctls for statistics and configuring timeouts
224 * - add saturation counter to rtentry to support per-packet load-balancing
225 * add flag to indicate round-robin flow, add list lookup from head
227 * - add sysctl / device node / syscall to support exporting and importing
228 * of flows with flag to indicate that a flow was imported so should
229 * not be considered for auto-cleaning
230 * - support explicit connection state (currently only ad-hoc for DSR)
231 * - idetach() cleanup for options VIMAGE builds.
233 VNET_DEFINE(int, flowtable_enable) = 1;
234 static VNET_DEFINE(int, flowtable_debug);
235 static VNET_DEFINE(int, flowtable_syn_expire) = SYN_IDLE;
236 static VNET_DEFINE(int, flowtable_udp_expire) = UDP_IDLE;
237 static VNET_DEFINE(int, flowtable_fin_wait_expire) = FIN_WAIT_IDLE;
238 static VNET_DEFINE(int, flowtable_tcp_expire) = TCP_IDLE;
239 static VNET_DEFINE(int, flowtable_nmbflows);
240 static VNET_DEFINE(int, flowtable_ready) = 0;
242 #define V_flowtable_enable VNET(flowtable_enable)
243 #define V_flowtable_debug VNET(flowtable_debug)
244 #define V_flowtable_syn_expire VNET(flowtable_syn_expire)
245 #define V_flowtable_udp_expire VNET(flowtable_udp_expire)
246 #define V_flowtable_fin_wait_expire VNET(flowtable_fin_wait_expire)
247 #define V_flowtable_tcp_expire VNET(flowtable_tcp_expire)
248 #define V_flowtable_nmbflows VNET(flowtable_nmbflows)
249 #define V_flowtable_ready VNET(flowtable_ready)
251 static SYSCTL_NODE(_net_inet, OID_AUTO, flowtable, CTLFLAG_RD, NULL,
253 SYSCTL_VNET_INT(_net_inet_flowtable, OID_AUTO, debug, CTLFLAG_RW,
254 &VNET_NAME(flowtable_debug), 0, "print debug info.");
255 SYSCTL_VNET_INT(_net_inet_flowtable, OID_AUTO, enable, CTLFLAG_RW,
256 &VNET_NAME(flowtable_enable), 0, "enable flowtable caching.");
259 * XXX This does not end up updating timeouts at runtime
260 * and only reflects the value for the last table added :-/
262 SYSCTL_VNET_INT(_net_inet_flowtable, OID_AUTO, syn_expire, CTLFLAG_RW,
263 &VNET_NAME(flowtable_syn_expire), 0,
264 "seconds after which to remove syn allocated flow.");
265 SYSCTL_VNET_INT(_net_inet_flowtable, OID_AUTO, udp_expire, CTLFLAG_RW,
266 &VNET_NAME(flowtable_udp_expire), 0,
267 "seconds after which to remove flow allocated to UDP.");
268 SYSCTL_VNET_INT(_net_inet_flowtable, OID_AUTO, fin_wait_expire, CTLFLAG_RW,
269 &VNET_NAME(flowtable_fin_wait_expire), 0,
270 "seconds after which to remove a flow in FIN_WAIT.");
271 SYSCTL_VNET_INT(_net_inet_flowtable, OID_AUTO, tcp_expire, CTLFLAG_RW,
272 &VNET_NAME(flowtable_tcp_expire), 0,
273 "seconds after which to remove flow allocated to a TCP connection.");
277 * Maximum number of flows that can be allocated of a given type.
279 * The table is allocated at boot time (for the pure caching case
280 * there is no reason why this could not be changed at runtime)
281 * and thus (currently) needs to be set with a tunable.
284 sysctl_nmbflows(SYSCTL_HANDLER_ARGS)
286 int error, newnmbflows;
288 newnmbflows = V_flowtable_nmbflows;
289 error = sysctl_handle_int(oidp, &newnmbflows, 0, req);
290 if (error == 0 && req->newptr) {
291 if (newnmbflows > V_flowtable_nmbflows) {
292 V_flowtable_nmbflows = newnmbflows;
293 uma_zone_set_max(V_flow_ipv4_zone,
294 V_flowtable_nmbflows);
295 uma_zone_set_max(V_flow_ipv6_zone,
296 V_flowtable_nmbflows);
302 SYSCTL_VNET_PROC(_net_inet_flowtable, OID_AUTO, nmbflows,
303 CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_nmbflows, "IU",
304 "Maximum number of flows allowed");
308 #define FS_PRINT(sb, field) sbuf_printf((sb), "\t%s: %jd\n", #field, fs->ft_##field)
311 fs_print(struct sbuf *sb, struct flowtable_stats *fs)
314 FS_PRINT(sb, collisions);
315 FS_PRINT(sb, allocated);
316 FS_PRINT(sb, misses);
317 FS_PRINT(sb, max_depth);
318 FS_PRINT(sb, free_checks);
321 FS_PRINT(sb, lookups);
325 flowtable_show_stats(struct sbuf *sb, struct flowtable *ft)
328 struct flowtable_stats fs, *pfs;
330 if (ft->ft_flags & FL_PCPU) {
331 bzero(&fs, sizeof(fs));
334 pfs->ft_collisions += ft->ft_stats[i].ft_collisions;
335 pfs->ft_allocated += ft->ft_stats[i].ft_allocated;
336 pfs->ft_misses += ft->ft_stats[i].ft_misses;
337 pfs->ft_free_checks += ft->ft_stats[i].ft_free_checks;
338 pfs->ft_frees += ft->ft_stats[i].ft_frees;
339 pfs->ft_hits += ft->ft_stats[i].ft_hits;
340 pfs->ft_lookups += ft->ft_stats[i].ft_lookups;
341 if (ft->ft_stats[i].ft_max_depth > pfs->ft_max_depth)
342 pfs->ft_max_depth = ft->ft_stats[i].ft_max_depth;
345 pfs = &ft->ft_stats[0];
351 sysctl_flowtable_stats(SYSCTL_HANDLER_ARGS)
353 struct flowtable *ft;
357 sb = sbuf_new(NULL, NULL, 64*1024, SBUF_FIXEDLEN);
359 ft = V_flow_list_head;
361 sbuf_printf(sb, "\ntable name: %s\n", ft->ft_name);
362 flowtable_show_stats(sb, ft);
366 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
371 SYSCTL_VNET_PROC(_net_inet_flowtable, OID_AUTO, stats, CTLTYPE_STRING|CTLFLAG_RD,
372 NULL, 0, sysctl_flowtable_stats, "A", "flowtable statistics");
377 rtalloc_ign_wrapper(struct route *ro, uint32_t hash, u_int fibnum)
380 rtalloc_ign_fib(ro, 0, fibnum);
385 flowtable_global_lock(struct flowtable *table, uint32_t hash)
387 int lock_index = (hash)&(table->ft_lock_count - 1);
389 mtx_lock(&table->ft_locks[lock_index]);
393 flowtable_global_unlock(struct flowtable *table, uint32_t hash)
395 int lock_index = (hash)&(table->ft_lock_count - 1);
397 mtx_unlock(&table->ft_locks[lock_index]);
401 flowtable_pcpu_lock(struct flowtable *table, uint32_t hash)
408 flowtable_pcpu_unlock(struct flowtable *table, uint32_t hash)
414 #define FL_ENTRY_INDEX(table, hash)((hash) % (table)->ft_size)
415 #define FL_ENTRY(table, hash) *flowtable_entry((table), (hash))
416 #define FL_ENTRY_LOCK(table, hash) (table)->ft_lock((table), (hash))
417 #define FL_ENTRY_UNLOCK(table, hash) (table)->ft_unlock((table), (hash))
419 #define FL_STALE (1<<8)
420 #define FL_OVERWRITE (1<<10)
423 flow_invalidate(struct flentry *fle)
426 fle->f_flags |= FL_STALE;
430 proto_to_flags(uint8_t proto)
453 flags_to_proto(int flags)
455 int proto, protoflags;
457 protoflags = flags & (FL_TCP|FL_SCTP|FL_UDP);
458 switch (protoflags) {
463 proto = IPPROTO_SCTP;
476 #ifdef FLOWTABLE_DEBUG
478 ipv4_flow_print_tuple(int flags, int proto, struct sockaddr_in *ssin,
479 struct sockaddr_in *dsin)
481 char saddr[4*sizeof "123"], daddr[4*sizeof "123"];
483 if (flags & FL_HASH_ALL) {
484 inet_ntoa_r(ssin->sin_addr, saddr);
485 inet_ntoa_r(dsin->sin_addr, daddr);
486 printf("proto=%d %s:%d->%s:%d\n",
487 proto, saddr, ntohs(ssin->sin_port), daddr,
488 ntohs(dsin->sin_port));
490 inet_ntoa_r(*(struct in_addr *) &dsin->sin_addr, daddr);
491 printf("proto=%d %s\n", proto, daddr);
498 ipv4_mbuf_demarshal(struct flowtable *ft, struct mbuf *m,
499 struct sockaddr_in *ssin, struct sockaddr_in *dsin, uint16_t *flags)
507 uint16_t sport, dport;
509 proto = sport = dport = 0;
510 ip = mtod(m, struct ip *);
511 dsin->sin_family = AF_INET;
512 dsin->sin_len = sizeof(*dsin);
513 dsin->sin_addr = ip->ip_dst;
514 ssin->sin_family = AF_INET;
515 ssin->sin_len = sizeof(*ssin);
516 ssin->sin_addr = ip->ip_src;
519 if ((*flags & FL_HASH_ALL) == 0) {
520 FLDPRINTF(ft, FL_DEBUG_ALL, "skip port check flags=0x%x ",
525 iphlen = ip->ip_hl << 2; /* XXX options? */
529 th = (struct tcphdr *)((caddr_t)ip + iphlen);
530 sport = th->th_sport;
531 dport = th->th_dport;
532 if ((*flags & FL_HASH_ALL) &&
533 (th->th_flags & (TH_RST|TH_FIN)))
537 uh = (struct udphdr *)((caddr_t)ip + iphlen);
538 sport = uh->uh_sport;
539 dport = uh->uh_dport;
542 sh = (struct sctphdr *)((caddr_t)ip + iphlen);
543 sport = sh->src_port;
544 dport = sh->dest_port;
547 FLDPRINTF(ft, FL_DEBUG_ALL, "proto=0x%x not supported\n", proto);
549 /* no port - hence not a protocol we care about */
555 *flags |= proto_to_flags(proto);
556 ssin->sin_port = sport;
557 dsin->sin_port = dport;
562 ipv4_flow_lookup_hash_internal(
563 struct sockaddr_in *ssin, struct sockaddr_in *dsin,
564 uint32_t *key, uint16_t flags)
566 uint16_t sport, dport;
570 if ((V_flowtable_enable == 0) || (V_flowtable_ready == 0))
572 proto = flags_to_proto(flags);
573 sport = dport = key[2] = key[1] = key[0] = 0;
574 if ((ssin != NULL) && (flags & FL_HASH_ALL)) {
575 key[1] = ssin->sin_addr.s_addr;
576 sport = ssin->sin_port;
579 key[2] = dsin->sin_addr.s_addr;
580 dport = dsin->sin_port;
582 if (flags & FL_HASH_ALL) {
583 ((uint16_t *)key)[0] = sport;
584 ((uint16_t *)key)[1] = dport;
586 offset = V_flow_hashjitter + proto;
588 return (jenkins_hashword(key, 3, offset));
591 static struct flentry *
592 flowtable_lookup_mbuf4(struct flowtable *ft, struct mbuf *m)
594 struct sockaddr_storage ssa, dsa;
596 struct sockaddr_in *dsin, *ssin;
598 dsin = (struct sockaddr_in *)&dsa;
599 ssin = (struct sockaddr_in *)&ssa;
600 bzero(dsin, sizeof(*dsin));
601 bzero(ssin, sizeof(*ssin));
602 flags = ft->ft_flags;
603 if (ipv4_mbuf_demarshal(ft, m, ssin, dsin, &flags) != 0)
606 return (flowtable_lookup(ft, &ssa, &dsa, M_GETFIB(m), flags));
610 flow_to_route(struct flentry *fle, struct route *ro)
612 uint32_t *hashkey = NULL;
613 struct sockaddr_in *sin;
615 sin = (struct sockaddr_in *)&ro->ro_dst;
616 sin->sin_family = AF_INET;
617 sin->sin_len = sizeof(*sin);
618 hashkey = ((struct flentry_v4 *)fle)->fl_flow.ipf_key;
619 sin->sin_addr.s_addr = hashkey[2];
620 ro->ro_rt = __DEVOLATILE(struct rtentry *, fle->f_rt);
621 ro->ro_lle = __DEVOLATILE(struct llentry *, fle->f_lle);
622 ro->ro_flags |= RT_NORTREF;
628 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
629 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
630 * pointer might become stale after other pullups (but we never use it
633 #define PULLUP_TO(_len, p, T) \
635 int x = (_len) + sizeof(T); \
636 if ((m)->m_len < x) { \
637 goto receive_failed; \
639 p = (mtod(m, char *) + (_len)); \
642 #define TCP(p) ((struct tcphdr *)(p))
643 #define SCTP(p) ((struct sctphdr *)(p))
644 #define UDP(p) ((struct udphdr *)(p))
647 ipv6_mbuf_demarshal(struct flowtable *ft, struct mbuf *m,
648 struct sockaddr_in6 *ssin6, struct sockaddr_in6 *dsin6, uint16_t *flags)
653 uint16_t src_port, dst_port;
657 offset = hlen = src_port = dst_port = 0;
659 ip6 = mtod(m, struct ip6_hdr *);
660 hlen = sizeof(struct ip6_hdr);
661 proto = ip6->ip6_nxt;
663 if ((*flags & FL_HASH_ALL) == 0)
666 while (ulp == NULL) {
669 case IPPROTO_OSPFIGP:
677 PULLUP_TO(hlen, ulp, struct tcphdr);
678 dst_port = TCP(ulp)->th_dport;
679 src_port = TCP(ulp)->th_sport;
680 if ((*flags & FL_HASH_ALL) &&
681 (TCP(ulp)->th_flags & (TH_RST|TH_FIN)))
685 PULLUP_TO(hlen, ulp, struct sctphdr);
686 src_port = SCTP(ulp)->src_port;
687 dst_port = SCTP(ulp)->dest_port;
690 PULLUP_TO(hlen, ulp, struct udphdr);
691 dst_port = UDP(ulp)->uh_dport;
692 src_port = UDP(ulp)->uh_sport;
694 case IPPROTO_HOPOPTS: /* RFC 2460 */
695 PULLUP_TO(hlen, ulp, struct ip6_hbh);
696 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
697 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
700 case IPPROTO_ROUTING: /* RFC 2460 */
701 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
702 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
703 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
706 case IPPROTO_FRAGMENT: /* RFC 2460 */
707 PULLUP_TO(hlen, ulp, struct ip6_frag);
708 hlen += sizeof (struct ip6_frag);
709 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
710 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
714 case IPPROTO_DSTOPTS: /* RFC 2460 */
715 PULLUP_TO(hlen, ulp, struct ip6_hbh);
716 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
717 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
720 case IPPROTO_AH: /* RFC 2402 */
721 PULLUP_TO(hlen, ulp, struct ip6_ext);
722 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
723 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
727 PULLUP_TO(hlen, ulp, struct ip6_ext);
738 dsin6->sin6_family = AF_INET6;
739 dsin6->sin6_len = sizeof(*dsin6);
740 dsin6->sin6_port = dst_port;
741 memcpy(&dsin6->sin6_addr, &ip6->ip6_dst, sizeof(struct in6_addr));
743 ssin6->sin6_family = AF_INET6;
744 ssin6->sin6_len = sizeof(*ssin6);
745 ssin6->sin6_port = src_port;
746 memcpy(&ssin6->sin6_addr, &ip6->ip6_src, sizeof(struct in6_addr));
747 *flags |= proto_to_flags(proto);
752 #define zero_key(key) \
766 ipv6_flow_lookup_hash_internal(
767 struct sockaddr_in6 *ssin6, struct sockaddr_in6 *dsin6,
768 uint32_t *key, uint16_t flags)
770 uint16_t sport, dport;
774 if ((V_flowtable_enable == 0) || (V_flowtable_ready == 0))
777 proto = flags_to_proto(flags);
781 memcpy(&key[1], &dsin6->sin6_addr, sizeof(struct in6_addr));
782 dport = dsin6->sin6_port;
784 if ((ssin6 != NULL) && (flags & FL_HASH_ALL)) {
785 memcpy(&key[5], &ssin6->sin6_addr, sizeof(struct in6_addr));
786 sport = ssin6->sin6_port;
788 if (flags & FL_HASH_ALL) {
789 ((uint16_t *)key)[0] = sport;
790 ((uint16_t *)key)[1] = dport;
792 offset = V_flow_hashjitter + proto;
794 return (jenkins_hashword(key, 9, offset));
797 static struct flentry *
798 flowtable_lookup_mbuf6(struct flowtable *ft, struct mbuf *m)
800 struct sockaddr_storage ssa, dsa;
801 struct sockaddr_in6 *dsin6, *ssin6;
804 dsin6 = (struct sockaddr_in6 *)&dsa;
805 ssin6 = (struct sockaddr_in6 *)&ssa;
806 bzero(dsin6, sizeof(*dsin6));
807 bzero(ssin6, sizeof(*ssin6));
808 flags = ft->ft_flags;
810 if (ipv6_mbuf_demarshal(ft, m, ssin6, dsin6, &flags) != 0)
813 return (flowtable_lookup(ft, &ssa, &dsa, M_GETFIB(m), flags));
817 flow_to_route_in6(struct flentry *fle, struct route_in6 *ro)
819 uint32_t *hashkey = NULL;
820 struct sockaddr_in6 *sin6;
822 sin6 = (struct sockaddr_in6 *)&ro->ro_dst;
824 sin6->sin6_family = AF_INET6;
825 sin6->sin6_len = sizeof(*sin6);
826 hashkey = ((struct flentry_v6 *)fle)->fl_flow.ipf_key;
827 memcpy(&sin6->sin6_addr, &hashkey[5], sizeof (struct in6_addr));
828 ro->ro_rt = __DEVOLATILE(struct rtentry *, fle->f_rt);
829 ro->ro_lle = __DEVOLATILE(struct llentry *, fle->f_lle);
830 ro->ro_flags |= RT_NORTREF;
835 flowtable_mask(struct flowtable *ft)
839 if (ft->ft_flags & FL_PCPU)
840 mask = ft->ft_masks[curcpu];
842 mask = ft->ft_masks[0];
847 static struct flentry **
848 flowtable_entry(struct flowtable *ft, uint32_t hash)
850 struct flentry **fle;
851 int index = (hash % ft->ft_size);
853 if (ft->ft_flags & FL_PCPU) {
854 KASSERT(&ft->ft_table.pcpu[curcpu][0] != NULL, ("pcpu not set"));
855 fle = &ft->ft_table.pcpu[curcpu][index];
857 KASSERT(&ft->ft_table.global[0] != NULL, ("global not set"));
858 fle = &ft->ft_table.global[index];
865 flow_stale(struct flowtable *ft, struct flentry *fle)
869 if ((fle->f_fhash == 0)
870 || ((fle->f_rt->rt_flags & RTF_HOST) &&
871 ((fle->f_rt->rt_flags & (RTF_UP))
873 || (fle->f_rt->rt_ifp == NULL)
874 || !RT_LINK_IS_UP(fle->f_rt->rt_ifp))
877 idle_time = time_uptime - fle->f_uptime;
879 if ((fle->f_flags & FL_STALE) ||
880 ((fle->f_flags & (TH_SYN|TH_ACK|TH_FIN)) == 0
881 && (idle_time > ft->ft_udp_idle)) ||
882 ((fle->f_flags & TH_FIN)
883 && (idle_time > ft->ft_fin_wait_idle)) ||
884 ((fle->f_flags & (TH_SYN|TH_ACK)) == TH_SYN
885 && (idle_time > ft->ft_syn_idle)) ||
886 ((fle->f_flags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)
887 && (idle_time > ft->ft_tcp_idle)) ||
888 ((fle->f_rt->rt_flags & RTF_UP) == 0 ||
889 (fle->f_rt->rt_ifp == NULL)))
896 flowtable_set_hashkey(struct flentry *fle, uint32_t *key)
901 if (fle->f_flags & FL_IPV6) {
903 hashkey = ((struct flentry_v4 *)fle)->fl_flow.ipf_key;
906 hashkey = ((struct flentry_v6 *)fle)->fl_flow.ipf_key;
909 for (i = 0; i < nwords; i++)
913 static struct flentry *
914 flow_alloc(struct flowtable *ft)
916 struct flentry *newfle;
920 zone = (ft->ft_flags & FL_IPV6) ? V_flow_ipv6_zone : V_flow_ipv4_zone;
922 newfle = uma_zalloc(zone, M_NOWAIT | M_ZERO);
924 atomic_add_int(&ft->ft_count, 1);
929 flow_free(struct flentry *fle, struct flowtable *ft)
933 zone = (ft->ft_flags & FL_IPV6) ? V_flow_ipv6_zone : V_flow_ipv4_zone;
934 atomic_add_int(&ft->ft_count, -1);
935 uma_zfree(zone, fle);
939 flow_full(struct flowtable *ft)
945 count = ft->ft_count;
947 if (full && (count < (V_flowtable_nmbflows - (V_flowtable_nmbflows >> 3))))
949 else if (!full && (count > (V_flowtable_nmbflows - (V_flowtable_nmbflows >> 5))))
952 if (full && !ft->ft_full) {
953 flowclean_freq = 4*hz;
954 if ((ft->ft_flags & FL_HASH_ALL) == 0)
955 ft->ft_udp_idle = ft->ft_fin_wait_idle =
956 ft->ft_syn_idle = ft->ft_tcp_idle = 5;
957 cv_broadcast(&flowclean_c_cv);
958 } else if (!full && ft->ft_full) {
959 flowclean_freq = 20*hz;
960 if ((ft->ft_flags & FL_HASH_ALL) == 0)
961 ft->ft_udp_idle = ft->ft_fin_wait_idle =
962 ft->ft_syn_idle = ft->ft_tcp_idle = 30;
965 return (ft->ft_full);
969 flowtable_insert(struct flowtable *ft, uint32_t hash, uint32_t *key,
970 uint32_t fibnum, struct route *ro, uint16_t flags)
972 struct flentry *fle, *fletail, *newfle, **flep;
973 struct flowtable_stats *fs = &ft->ft_stats[curcpu];
978 newfle = flow_alloc(ft);
982 newfle->f_flags |= (flags & FL_IPV6);
983 proto = flags_to_proto(flags);
985 FL_ENTRY_LOCK(ft, hash);
986 mask = flowtable_mask(ft);
987 flep = flowtable_entry(ft, hash);
988 fletail = fle = *flep;
991 bit_set(mask, FL_ENTRY_INDEX(ft, hash));
992 *flep = fle = newfle;
999 * find end of list and make sure that we were not
1000 * preempted by another thread handling this flow
1002 while (fle != NULL) {
1003 if (fle->f_fhash == hash && !flow_stale(ft, fle)) {
1005 * there was either a hash collision
1006 * or we lost a race to insert
1008 FL_ENTRY_UNLOCK(ft, hash);
1009 flow_free(newfle, ft);
1011 if (flags & FL_OVERWRITE)
1016 * re-visit this double condition XXX
1018 if (fletail->f_next != NULL)
1019 fletail = fle->f_next;
1025 if (depth > fs->ft_max_depth)
1026 fs->ft_max_depth = depth;
1027 fletail->f_next = newfle;
1030 flowtable_set_hashkey(fle, key);
1032 fle->f_proto = proto;
1033 fle->f_rt = ro->ro_rt;
1034 fle->f_lle = ro->ro_lle;
1035 fle->f_fhash = hash;
1036 fle->f_fibnum = fibnum;
1037 fle->f_uptime = time_uptime;
1038 FL_ENTRY_UNLOCK(ft, hash);
1043 kern_flowtable_insert(struct flowtable *ft,
1044 struct sockaddr_storage *ssa, struct sockaddr_storage *dsa,
1045 struct route *ro, uint32_t fibnum, int flags)
1047 uint32_t key[9], hash;
1049 flags = (ft->ft_flags | flags | FL_OVERWRITE);
1053 if (ssa->ss_family == AF_INET)
1054 hash = ipv4_flow_lookup_hash_internal((struct sockaddr_in *)ssa,
1055 (struct sockaddr_in *)dsa, key, flags);
1058 if (ssa->ss_family == AF_INET6)
1059 hash = ipv6_flow_lookup_hash_internal((struct sockaddr_in6 *)ssa,
1060 (struct sockaddr_in6 *)dsa, key, flags);
1062 if (ro->ro_rt == NULL || ro->ro_lle == NULL)
1065 FLDPRINTF(ft, FL_DEBUG,
1066 "kern_flowtable_insert: key=%x:%x:%x hash=%x fibnum=%d flags=%x\n",
1067 key[0], key[1], key[2], hash, fibnum, flags);
1068 return (flowtable_insert(ft, hash, key, fibnum, ro, flags));
1072 flowtable_key_equal(struct flentry *fle, uint32_t *key)
1077 if (fle->f_flags & FL_IPV6) {
1079 hashkey = ((struct flentry_v4 *)fle)->fl_flow.ipf_key;
1082 hashkey = ((struct flentry_v6 *)fle)->fl_flow.ipf_key;
1085 for (i = 0; i < nwords; i++)
1086 if (hashkey[i] != key[i])
1093 flowtable_lookup_mbuf(struct flowtable *ft, struct mbuf *m, int af)
1095 struct flentry *fle = NULL;
1099 fle = flowtable_lookup_mbuf4(ft, m);
1103 fle = flowtable_lookup_mbuf6(ft, m);
1105 if (fle != NULL && m != NULL && (m->m_flags & M_FLOWID) == 0) {
1106 m->m_flags |= M_FLOWID;
1107 m->m_pkthdr.flowid = fle->f_fhash;
1113 flowtable_lookup(struct flowtable *ft, struct sockaddr_storage *ssa,
1114 struct sockaddr_storage *dsa, uint32_t fibnum, int flags)
1116 uint32_t key[9], hash;
1117 struct flentry *fle;
1118 struct flowtable_stats *fs = &ft->ft_stats[curcpu];
1122 struct llentry *lle;
1123 struct route sro, *ro;
1124 struct route_in6 sro6;
1126 sro.ro_rt = sro6.ro_rt = NULL;
1127 sro.ro_lle = sro6.ro_lle = NULL;
1130 flags |= ft->ft_flags;
1131 proto = flags_to_proto(flags);
1133 if (ssa->ss_family == AF_INET) {
1134 struct sockaddr_in *ssin, *dsin;
1137 memcpy(&ro->ro_dst, dsa, sizeof(struct sockaddr_in));
1139 * The harvested source and destination addresses
1140 * may contain port information if the packet is
1141 * from a transport protocol (e.g. TCP/UDP). The
1142 * port field must be cleared before performing
1145 ((struct sockaddr_in *)&ro->ro_dst)->sin_port = 0;
1146 dsin = (struct sockaddr_in *)dsa;
1147 ssin = (struct sockaddr_in *)ssa;
1148 if ((dsin->sin_addr.s_addr == ssin->sin_addr.s_addr) ||
1149 (ntohl(dsin->sin_addr.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
1150 (ntohl(ssin->sin_addr.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
1153 hash = ipv4_flow_lookup_hash_internal(ssin, dsin, key, flags);
1157 if (ssa->ss_family == AF_INET6) {
1158 struct sockaddr_in6 *ssin6, *dsin6;
1160 ro = (struct route *)&sro6;
1161 memcpy(&sro6.ro_dst, dsa,
1162 sizeof(struct sockaddr_in6));
1163 ((struct sockaddr_in6 *)&ro->ro_dst)->sin6_port = 0;
1164 dsin6 = (struct sockaddr_in6 *)dsa;
1165 ssin6 = (struct sockaddr_in6 *)ssa;
1168 hash = ipv6_flow_lookup_hash_internal(ssin6, dsin6, key, flags);
1172 * Ports are zero and this isn't a transmit cache
1173 * - thus not a protocol for which we need to keep
1175 * FL_HASH_ALL => key[0] != 0 for TCP || UDP || SCTP
1177 if (hash == 0 || (key[0] == 0 && (ft->ft_flags & FL_HASH_ALL)))
1181 FL_ENTRY_LOCK(ft, hash);
1182 if ((fle = FL_ENTRY(ft, hash)) == NULL) {
1183 FL_ENTRY_UNLOCK(ft, hash);
1187 rt = __DEVOLATILE(struct rtentry *, fle->f_rt);
1188 lle = __DEVOLATILE(struct llentry *, fle->f_lle);
1191 && fle->f_fhash == hash
1192 && flowtable_key_equal(fle, key)
1193 && (proto == fle->f_proto)
1194 && (fibnum == fle->f_fibnum)
1195 && (rt->rt_flags & RTF_UP)
1196 && (rt->rt_ifp != NULL)
1197 && (lle->la_flags & LLE_VALID)) {
1199 fle->f_uptime = time_uptime;
1200 fle->f_flags |= flags;
1201 FL_ENTRY_UNLOCK(ft, hash);
1203 } else if (fle->f_next != NULL) {
1207 FL_ENTRY_UNLOCK(ft, hash);
1209 if (flags & FL_NOAUTO || flow_full(ft))
1214 * This bit of code ends up locking the
1215 * same route 3 times (just like ip_output + ether_output)
1217 * - in rt_check when called by arpresolve
1218 * - dropping the refcount for the rtentry
1220 * This could be consolidated to one if we wrote a variant
1221 * of arpresolve with an rt_check variant that expected to
1222 * receive the route locked
1226 if ((ro->ro_dst.sa_family != AF_INET) &&
1227 (ro->ro_dst.sa_family != AF_INET6))
1228 panic("sa_family == %d\n", ro->ro_dst.sa_family);
1231 ft->ft_rtalloc(ro, hash, fibnum);
1232 if (ro->ro_rt == NULL)
1233 error = ENETUNREACH;
1235 struct llentry *lle = NULL;
1236 struct sockaddr_storage *l3addr;
1237 struct rtentry *rt = ro->ro_rt;
1238 struct ifnet *ifp = rt->rt_ifp;
1240 if (ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) {
1246 if (ssa->ss_family == AF_INET6) {
1247 struct sockaddr_in6 *dsin6;
1249 dsin6 = (struct sockaddr_in6 *)dsa;
1250 if (in6_localaddr(&dsin6->sin6_addr)) {
1256 if (rt->rt_flags & RTF_GATEWAY)
1257 l3addr = (struct sockaddr_storage *)rt->rt_gateway;
1260 l3addr = (struct sockaddr_storage *)&ro->ro_dst;
1261 lle = llentry_alloc(ifp, LLTABLE6(ifp), l3addr);
1265 if (ssa->ss_family == AF_INET) {
1266 if (rt->rt_flags & RTF_GATEWAY)
1267 l3addr = (struct sockaddr_storage *)rt->rt_gateway;
1269 l3addr = (struct sockaddr_storage *)&ro->ro_dst;
1270 lle = llentry_alloc(ifp, LLTABLE(ifp), l3addr);
1281 error = flowtable_insert(ft, hash, key, fibnum, ro, flags);
1291 return ((error) ? NULL : fle);
1295 * used by the bit_alloc macro
1297 #define calloc(count, size) malloc((count)*(size), M_DEVBUF, M_WAITOK|M_ZERO)
1300 flowtable_alloc(char *name, int nentry, int flags)
1302 struct flowtable *ft, *fttail;
1305 if (V_flow_hashjitter == 0)
1306 V_flow_hashjitter = arc4random();
1308 KASSERT(nentry > 0, ("nentry must be > 0, is %d\n", nentry));
1310 ft = malloc(sizeof(struct flowtable),
1311 M_RTABLE, M_WAITOK | M_ZERO);
1314 ft->ft_flags = flags;
1315 ft->ft_size = nentry;
1317 ft->ft_rtalloc = rtalloc_mpath_fib;
1319 ft->ft_rtalloc = rtalloc_ign_wrapper;
1321 if (flags & FL_PCPU) {
1322 ft->ft_lock = flowtable_pcpu_lock;
1323 ft->ft_unlock = flowtable_pcpu_unlock;
1325 for (i = 0; i <= mp_maxid; i++) {
1326 ft->ft_table.pcpu[i] =
1327 malloc(nentry*sizeof(struct flentry *),
1328 M_RTABLE, M_WAITOK | M_ZERO);
1329 ft->ft_masks[i] = bit_alloc(nentry);
1332 ft->ft_lock_count = 2*(powerof2(mp_maxid + 1) ? (mp_maxid + 1):
1333 (fls(mp_maxid + 1) << 1));
1335 ft->ft_lock = flowtable_global_lock;
1336 ft->ft_unlock = flowtable_global_unlock;
1337 ft->ft_table.global =
1338 malloc(nentry*sizeof(struct flentry *),
1339 M_RTABLE, M_WAITOK | M_ZERO);
1340 ft->ft_locks = malloc(ft->ft_lock_count*sizeof(struct mtx),
1341 M_RTABLE, M_WAITOK | M_ZERO);
1342 for (i = 0; i < ft->ft_lock_count; i++)
1343 mtx_init(&ft->ft_locks[i], "flow", NULL, MTX_DEF|MTX_DUPOK);
1345 ft->ft_masks[0] = bit_alloc(nentry);
1347 ft->ft_tmpmask = bit_alloc(nentry);
1350 * In the local transmit case the table truly is
1351 * just a cache - so everything is eligible for
1352 * replacement after 5s of non-use
1354 if (flags & FL_HASH_ALL) {
1355 ft->ft_udp_idle = V_flowtable_udp_expire;
1356 ft->ft_syn_idle = V_flowtable_syn_expire;
1357 ft->ft_fin_wait_idle = V_flowtable_fin_wait_expire;
1358 ft->ft_tcp_idle = V_flowtable_fin_wait_expire;
1360 ft->ft_udp_idle = ft->ft_fin_wait_idle =
1361 ft->ft_syn_idle = ft->ft_tcp_idle = 30;
1366 * hook in to the cleaner list
1368 if (V_flow_list_head == NULL)
1369 V_flow_list_head = ft;
1371 fttail = V_flow_list_head;
1372 while (fttail->ft_next != NULL)
1373 fttail = fttail->ft_next;
1374 fttail->ft_next = ft;
1381 * The rest of the code is devoted to garbage collection of expired entries.
1382 * It is a new additon made necessary by the switch to dynamically allocating
1387 fle_free(struct flentry *fle, struct flowtable *ft)
1390 struct llentry *lle;
1392 rt = __DEVOLATILE(struct rtentry *, fle->f_rt);
1393 lle = __DEVOLATILE(struct llentry *, fle->f_lle);
1402 flowtable_free_stale(struct flowtable *ft, struct rtentry *rt)
1404 int curbit = 0, count;
1405 struct flentry *fle, **flehead, *fleprev;
1406 struct flentry *flefreehead, *flefreetail, *fletmp;
1407 bitstr_t *mask, *tmpmask;
1408 struct flowtable_stats *fs = &ft->ft_stats[curcpu];
1410 flefreehead = flefreetail = NULL;
1411 mask = flowtable_mask(ft);
1412 tmpmask = ft->ft_tmpmask;
1413 memcpy(tmpmask, mask, ft->ft_size/8);
1415 * XXX Note to self, bit_ffs operates at the byte level
1416 * and thus adds gratuitous overhead
1418 bit_ffs(tmpmask, ft->ft_size, &curbit);
1419 while (curbit != -1) {
1420 if (curbit >= ft->ft_size || curbit < -1) {
1422 "warning: bad curbit value %d \n",
1427 FL_ENTRY_LOCK(ft, curbit);
1428 flehead = flowtable_entry(ft, curbit);
1429 fle = fleprev = *flehead;
1431 fs->ft_free_checks++;
1433 if (fle == NULL && curbit > 0) {
1435 "warning bit=%d set, but no fle found\n",
1439 while (fle != NULL) {
1441 if (__DEVOLATILE(struct rtentry *, fle->f_rt) != rt) {
1446 } else if (!flow_stale(ft, fle)) {
1452 * delete head of the list
1454 if (fleprev == *flehead) {
1456 if (fle == fleprev) {
1457 fleprev = *flehead = fle->f_next;
1459 fleprev = *flehead = fle;
1463 * don't advance fleprev
1466 fleprev->f_next = fle->f_next;
1467 fle = fleprev->f_next;
1470 if (flefreehead == NULL)
1471 flefreehead = flefreetail = fletmp;
1473 flefreetail->f_next = fletmp;
1474 flefreetail = fletmp;
1476 fletmp->f_next = NULL;
1478 if (*flehead == NULL)
1479 bit_clear(mask, curbit);
1480 FL_ENTRY_UNLOCK(ft, curbit);
1481 bit_clear(tmpmask, curbit);
1482 bit_ffs(tmpmask, ft->ft_size, &curbit);
1485 while ((fle = flefreehead) != NULL) {
1486 flefreehead = fle->f_next;
1491 if (V_flowtable_debug && count)
1492 log(LOG_DEBUG, "freed %d flow entries\n", count);
1496 flowtable_route_flush(struct flowtable *ft, struct rtentry *rt)
1500 if (ft->ft_flags & FL_PCPU) {
1502 if (smp_started == 1) {
1503 thread_lock(curthread);
1504 sched_bind(curthread, i);
1505 thread_unlock(curthread);
1508 flowtable_free_stale(ft, rt);
1510 if (smp_started == 1) {
1511 thread_lock(curthread);
1512 sched_unbind(curthread);
1513 thread_unlock(curthread);
1517 flowtable_free_stale(ft, rt);
1522 flowtable_clean_vnet(void)
1524 struct flowtable *ft;
1527 ft = V_flow_list_head;
1528 while (ft != NULL) {
1529 if (ft->ft_flags & FL_PCPU) {
1531 if (smp_started == 1) {
1532 thread_lock(curthread);
1533 sched_bind(curthread, i);
1534 thread_unlock(curthread);
1537 flowtable_free_stale(ft, NULL);
1539 if (smp_started == 1) {
1540 thread_lock(curthread);
1541 sched_unbind(curthread);
1542 thread_unlock(curthread);
1546 flowtable_free_stale(ft, NULL);
1553 flowtable_cleaner(void)
1555 VNET_ITERATOR_DECL(vnet_iter);
1559 log(LOG_INFO, "flowtable cleaner started\n");
1563 VNET_FOREACH(vnet_iter) {
1564 CURVNET_SET(vnet_iter);
1565 flowtable_clean_vnet();
1568 VNET_LIST_RUNLOCK();
1571 * The 10 second interval between cleaning checks
1574 mtx_lock(&flowclean_lock);
1576 sched_prio(td, PPAUSE);
1579 cv_broadcast(&flowclean_f_cv);
1580 cv_timedwait(&flowclean_c_cv, &flowclean_lock, flowclean_freq);
1581 mtx_unlock(&flowclean_lock);
1586 flowtable_flush(void *unused __unused)
1590 mtx_lock(&flowclean_lock);
1591 start = flowclean_cycles;
1592 while (start == flowclean_cycles) {
1593 cv_broadcast(&flowclean_c_cv);
1594 cv_wait(&flowclean_f_cv, &flowclean_lock);
1596 mtx_unlock(&flowclean_lock);
1599 static struct kproc_desc flow_kp = {
1604 SYSINIT(flowcleaner, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, kproc_start, &flow_kp);
1607 flowtable_init_vnet(const void *unused __unused)
1610 V_flowtable_nmbflows = 1024 + maxusers * 64 * mp_ncpus;
1611 V_flow_ipv4_zone = uma_zcreate("ip4flow", sizeof(struct flentry_v4),
1612 NULL, NULL, NULL, NULL, 64, UMA_ZONE_MAXBUCKET);
1613 V_flow_ipv6_zone = uma_zcreate("ip6flow", sizeof(struct flentry_v6),
1614 NULL, NULL, NULL, NULL, 64, UMA_ZONE_MAXBUCKET);
1615 uma_zone_set_max(V_flow_ipv4_zone, V_flowtable_nmbflows);
1616 uma_zone_set_max(V_flow_ipv6_zone, V_flowtable_nmbflows);
1617 V_flowtable_ready = 1;
1619 VNET_SYSINIT(flowtable_init_vnet, SI_SUB_SMP, SI_ORDER_ANY,
1620 flowtable_init_vnet, NULL);
1623 flowtable_init(const void *unused __unused)
1626 cv_init(&flowclean_c_cv, "c_flowcleanwait");
1627 cv_init(&flowclean_f_cv, "f_flowcleanwait");
1628 mtx_init(&flowclean_lock, "flowclean lock", NULL, MTX_DEF);
1629 EVENTHANDLER_REGISTER(ifnet_departure_event, flowtable_flush, NULL,
1630 EVENTHANDLER_PRI_ANY);
1631 flowclean_freq = 20*hz;
1633 SYSINIT(flowtable_init, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST,
1634 flowtable_init, NULL);
1639 flowtable_uninit(const void *unused __unused)
1642 V_flowtable_ready = 0;
1643 uma_zdestroy(V_flow_ipv4_zone);
1644 uma_zdestroy(V_flow_ipv6_zone);
1647 VNET_SYSUNINIT(flowtable_uninit, SI_SUB_KTHREAD_INIT, SI_ORDER_ANY,
1648 flowtable_uninit, NULL);
1653 flowtable_get_hashkey(struct flentry *fle)
1657 if (fle->f_flags & FL_IPV6)
1658 hashkey = ((struct flentry_v4 *)fle)->fl_flow.ipf_key;
1660 hashkey = ((struct flentry_v6 *)fle)->fl_flow.ipf_key;
1666 flowtable_mask_pcpu(struct flowtable *ft, int cpuid)
1670 if (ft->ft_flags & FL_PCPU)
1671 mask = ft->ft_masks[cpuid];
1673 mask = ft->ft_masks[0];
1678 static struct flentry **
1679 flowtable_entry_pcpu(struct flowtable *ft, uint32_t hash, int cpuid)
1681 struct flentry **fle;
1682 int index = (hash % ft->ft_size);
1684 if (ft->ft_flags & FL_PCPU) {
1685 fle = &ft->ft_table.pcpu[cpuid][index];
1687 fle = &ft->ft_table.global[index];
1694 flow_show(struct flowtable *ft, struct flentry *fle)
1697 int rt_valid, ifp_valid;
1698 uint16_t sport, dport;
1700 char saddr[4*sizeof "123"], daddr[4*sizeof "123"];
1701 volatile struct rtentry *rt;
1702 struct ifnet *ifp = NULL;
1704 idle_time = (int)(time_uptime - fle->f_uptime);
1706 rt_valid = rt != NULL;
1709 ifp_valid = ifp != NULL;
1710 hashkey = flowtable_get_hashkey(fle);
1711 if (fle->f_flags & FL_IPV6)
1714 inet_ntoa_r(*(struct in_addr *) &hashkey[2], daddr);
1715 if (ft->ft_flags & FL_HASH_ALL) {
1716 inet_ntoa_r(*(struct in_addr *) &hashkey[1], saddr);
1717 sport = ntohs(((uint16_t *)hashkey)[0]);
1718 dport = ntohs(((uint16_t *)hashkey)[1]);
1719 db_printf("%s:%d->%s:%d",
1720 saddr, sport, daddr,
1723 db_printf("%s ", daddr);
1726 if (fle->f_flags & FL_STALE)
1727 db_printf(" FL_STALE ");
1728 if (fle->f_flags & FL_TCP)
1729 db_printf(" FL_TCP ");
1730 if (fle->f_flags & FL_UDP)
1731 db_printf(" FL_UDP ");
1733 if (rt->rt_flags & RTF_UP)
1734 db_printf(" RTF_UP ");
1737 if (ifp->if_flags & IFF_LOOPBACK)
1738 db_printf(" IFF_LOOPBACK ");
1739 if (ifp->if_flags & IFF_UP)
1740 db_printf(" IFF_UP ");
1741 if (ifp->if_flags & IFF_POINTOPOINT)
1742 db_printf(" IFF_POINTOPOINT ");
1744 if (fle->f_flags & FL_IPV6)
1745 db_printf("\n\tkey=%08x:%08x:%08x%08x:%08x:%08x%08x:%08x:%08x",
1746 hashkey[0], hashkey[1], hashkey[2],
1747 hashkey[3], hashkey[4], hashkey[5],
1748 hashkey[6], hashkey[7], hashkey[8]);
1750 db_printf("\n\tkey=%08x:%08x:%08x ",
1751 hashkey[0], hashkey[1], hashkey[2]);
1752 db_printf("hash=%08x idle_time=%03d"
1753 "\n\tfibnum=%02d rt=%p",
1754 fle->f_fhash, idle_time, fle->f_fibnum, fle->f_rt);
1759 flowtable_show(struct flowtable *ft, int cpuid)
1762 struct flentry *fle, **flehead;
1763 bitstr_t *mask, *tmpmask;
1766 db_printf("cpu: %d\n", cpuid);
1767 mask = flowtable_mask_pcpu(ft, cpuid);
1768 tmpmask = ft->ft_tmpmask;
1769 memcpy(tmpmask, mask, ft->ft_size/8);
1771 * XXX Note to self, bit_ffs operates at the byte level
1772 * and thus adds gratuitous overhead
1774 bit_ffs(tmpmask, ft->ft_size, &curbit);
1775 while (curbit != -1) {
1776 if (curbit >= ft->ft_size || curbit < -1) {
1777 db_printf("warning: bad curbit value %d \n",
1782 flehead = flowtable_entry_pcpu(ft, curbit, cpuid);
1785 while (fle != NULL) {
1790 bit_clear(tmpmask, curbit);
1791 bit_ffs(tmpmask, ft->ft_size, &curbit);
1796 flowtable_show_vnet(void)
1798 struct flowtable *ft;
1801 ft = V_flow_list_head;
1802 while (ft != NULL) {
1803 printf("name: %s\n", ft->ft_name);
1804 if (ft->ft_flags & FL_PCPU) {
1806 flowtable_show(ft, i);
1809 flowtable_show(ft, -1);
1815 DB_SHOW_COMMAND(flowtables, db_show_flowtables)
1817 VNET_ITERATOR_DECL(vnet_iter);
1819 VNET_FOREACH(vnet_iter) {
1820 CURVNET_SET(vnet_iter);
1822 db_printf("vnet %p\n", vnet_iter);
1824 flowtable_show_vnet();