1 /* $KAME: altq_subr.c,v 1.23 2004/04/20 16:10:06 itojun Exp $ */
2 /* $DragonFly: src/sys/net/altq/altq_subr.c,v 1.12 2008/05/14 11:59:23 sephe Exp $ */
5 * Copyright (C) 1997-2003
6 * Sony Computer Science Laboratories Inc. All rights reserved.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 #include "opt_inet6.h"
34 #include <sys/param.h>
35 #include <sys/malloc.h>
37 #include <sys/systm.h>
39 #include <sys/socket.h>
40 #include <sys/socketvar.h>
41 #include <sys/kernel.h>
42 #include <sys/callout.h>
43 #include <sys/errno.h>
44 #include <sys/syslog.h>
45 #include <sys/sysctl.h>
46 #include <sys/queue.h>
47 #include <sys/thread2.h>
50 #include <net/if_dl.h>
51 #include <net/if_types.h>
52 #include <net/ifq_var.h>
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/ip.h>
58 #include <netinet/ip6.h>
60 #include <netinet/tcp.h>
61 #include <netinet/udp.h>
63 #include <net/pf/pfvar.h>
64 #include <net/altq/altq.h>
66 /* machine dependent clock related includes */
67 #include <machine/clock.h> /* for tsc_frequency */
68 #include <machine/md_var.h> /* for cpu_feature */
69 #include <machine/specialreg.h> /* for CPUID_TSC */
72 * internal function prototypes
74 static void tbr_timeout(void *);
75 static int altq_enable_locked(struct ifaltq *);
76 static int altq_disable_locked(struct ifaltq *);
77 static int altq_detach_locked(struct ifaltq *);
78 static int tbr_set_locked(struct ifaltq *, struct tb_profile *);
80 int (*altq_input)(struct mbuf *, int) = NULL;
81 static int tbr_timer = 0; /* token bucket regulator timer */
82 static struct callout tbr_callout;
84 int pfaltq_running; /* keep track of running state */
86 MALLOC_DEFINE(M_ALTQ, "altq", "ALTQ structures");
89 * alternate queueing support routines
92 /* look up the queue state by the interface name and the queueing type. */
94 altq_lookup(const char *name, int type)
98 if ((ifp = ifunit(name)) != NULL) {
99 if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
100 return (ifp->if_snd.altq_disc);
107 altq_attach(struct ifaltq *ifq, int type, void *discipline,
108 ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request,
110 void *(*classify)(struct ifaltq *, struct mbuf *, struct altq_pktattr *))
112 if (!ifq_is_ready(ifq))
115 ifq->altq_type = type;
116 ifq->altq_disc = discipline;
117 ifq->altq_clfier = clfier;
118 ifq->altq_classify = classify;
119 ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
120 ifq_set_methods(ifq, enqueue, dequeue, request);
125 altq_detach_locked(struct ifaltq *ifq)
127 if (!ifq_is_ready(ifq))
129 if (ifq_is_enabled(ifq))
131 if (!ifq_is_attached(ifq))
134 ifq_set_classic(ifq);
135 ifq->altq_type = ALTQT_NONE;
136 ifq->altq_disc = NULL;
137 ifq->altq_clfier = NULL;
138 ifq->altq_classify = NULL;
139 ifq->altq_flags &= ALTQF_CANTCHANGE;
144 altq_detach(struct ifaltq *ifq)
149 error = altq_detach_locked(ifq);
155 altq_enable_locked(struct ifaltq *ifq)
157 if (!ifq_is_ready(ifq))
159 if (ifq_is_enabled(ifq))
162 ifq_purge_all_locked(ifq);
164 ifq->altq_flags |= ALTQF_ENABLED;
165 if (ifq->altq_clfier != NULL)
166 ifq->altq_flags |= ALTQF_CLASSIFY;
171 altq_enable(struct ifaltq *ifq)
176 error = altq_enable_locked(ifq);
182 altq_disable_locked(struct ifaltq *ifq)
184 if (!ifq_is_enabled(ifq))
187 ifq_purge_all_locked(ifq);
188 ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY);
193 altq_disable(struct ifaltq *ifq)
198 error = altq_disable_locked(ifq);
204 * internal representation of token bucket parameters
205 * rate: byte_per_unittime << 32
206 * (((bits_per_sec) / 8) << 32) / machclk_freq
211 #define TBR_SCALE(x) ((int64_t)(x) << TBR_SHIFT)
212 #define TBR_UNSCALE(x) ((x) >> TBR_SHIFT)
215 tbr_dequeue(struct ifaltq_subque *ifsq, struct mbuf *mpolled, int op)
217 struct ifaltq *ifq = ifsq->ifsq_altq;
218 struct tb_regulator *tbr;
223 if (ifsq_get_index(ifsq) != ALTQ_SUBQ_INDEX_DEFAULT) {
225 * Race happened, the unrelated subqueue was
226 * picked during the packet scheduler transition.
228 ifsq_classic_request(ifsq, ALTRQ_PURGE, NULL);
234 if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
235 /* if this is a remove after poll, bypass tbr check */
237 /* update token only when it is negative */
238 if (tbr->tbr_token <= 0) {
239 now = read_machclk();
240 interval = now - tbr->tbr_last;
241 if (interval >= tbr->tbr_filluptime)
242 tbr->tbr_token = tbr->tbr_depth;
244 tbr->tbr_token += interval * tbr->tbr_rate;
245 if (tbr->tbr_token > tbr->tbr_depth)
246 tbr->tbr_token = tbr->tbr_depth;
250 /* if token is still negative, don't allow dequeue */
251 if (tbr->tbr_token <= 0) {
257 if (ifq_is_enabled(ifq)) {
258 m = (*ifsq->ifsq_dequeue)(ifsq, mpolled, op);
259 } else if (op == ALTDQ_POLL) {
263 KKASSERT(mpolled == NULL || mpolled == m);
266 if (m != NULL && op == ALTDQ_REMOVE)
267 tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
268 tbr->tbr_lastop = op;
274 * set a token bucket regulator.
275 * if the specified rate is zero, the token bucket regulator is deleted.
278 tbr_set_locked(struct ifaltq *ifq, struct tb_profile *profile)
280 struct tb_regulator *tbr, *otbr;
282 if (machclk_freq == 0)
284 if (machclk_freq == 0) {
285 kprintf("%s: no cpu clock available!\n", __func__);
289 if (profile->rate == 0) {
290 /* delete this tbr */
291 if ((tbr = ifq->altq_tbr) == NULL)
293 ifq->altq_tbr = NULL;
298 tbr = kmalloc(sizeof(*tbr), M_ALTQ, M_WAITOK | M_ZERO);
299 tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
300 tbr->tbr_depth = TBR_SCALE(profile->depth);
301 if (tbr->tbr_rate > 0)
302 tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
304 tbr->tbr_filluptime = 0xffffffffffffffffLL;
305 tbr->tbr_token = tbr->tbr_depth;
306 tbr->tbr_last = read_machclk();
307 tbr->tbr_lastop = ALTDQ_REMOVE;
309 otbr = ifq->altq_tbr;
310 ifq->altq_tbr = tbr; /* set the new tbr */
314 else if (tbr_timer == 0) {
315 callout_reset(&tbr_callout, 1, tbr_timeout, NULL);
322 tbr_set(struct ifaltq *ifq, struct tb_profile *profile)
327 error = tbr_set_locked(ifq, profile);
333 * tbr_timeout goes through the interface list, and kicks the drivers
337 tbr_timeout(void *arg)
344 for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
345 struct ifaltq_subque *ifsq;
347 if (ifp->if_snd.altq_tbr == NULL)
350 ifsq = &ifp->if_snd.altq_subq[ALTQ_SUBQ_INDEX_DEFAULT];
352 if (!ifsq_is_empty(ifsq) && ifp->if_start != NULL) {
353 ifnet_serialize_tx(ifp, ifsq);
354 (*ifp->if_start)(ifp, ifsq);
355 ifnet_deserialize_tx(ifp, ifsq);
360 callout_reset(&tbr_callout, 1, tbr_timeout, NULL);
362 tbr_timer = 0; /* don't need tbr_timer anymore */
366 * get token bucket regulator profile
369 tbr_get(struct ifaltq *ifq, struct tb_profile *profile)
371 struct tb_regulator *tbr;
373 if ((tbr = ifq->altq_tbr) == NULL) {
378 (u_int)TBR_UNSCALE(tbr->tbr_rate * 8 * machclk_freq);
379 profile->depth = (u_int)TBR_UNSCALE(tbr->tbr_depth);
385 * attach a discipline to the interface. if one already exists, it is
389 altq_pfattach(struct pf_altq *a)
395 if (a->scheduler == ALTQT_NONE)
398 if (a->altq_disc == NULL)
401 ifp = ifunit(a->ifname);
408 switch (a->scheduler) {
411 error = cbq_pfattach(a, ifq);
416 error = priq_pfattach(a, ifq);
421 error = hfsc_pfattach(a, ifq);
426 error = fairq_pfattach(a, ifq);
434 /* if the state is running, enable altq */
435 if (error == 0 && pfaltq_running && ifq->altq_type != ALTQT_NONE &&
436 !ifq_is_enabled(ifq))
437 error = altq_enable_locked(ifq);
439 /* if altq is already enabled, reset set tokenbucket regulator */
440 if (error == 0 && ifq_is_enabled(ifq)) {
441 struct tb_profile tb;
443 tb.rate = a->ifbandwidth;
444 tb.depth = a->tbrsize;
445 error = tbr_set_locked(ifq, &tb);
453 * detach a discipline from the interface.
454 * it is possible that the discipline was already overridden by another
458 altq_pfdetach(struct pf_altq *a)
464 ifp = ifunit(a->ifname);
469 /* if this discipline is no longer referenced, just return */
470 if (a->altq_disc == NULL)
475 if (a->altq_disc != ifq->altq_disc)
478 if (ifq_is_enabled(ifq))
479 error = altq_disable_locked(ifq);
481 error = altq_detach_locked(ifq);
489 * add a discipline or a queue
492 altq_add(struct pf_altq *a)
496 if (a->qname[0] != 0)
497 return (altq_add_queue(a));
499 if (machclk_freq == 0)
501 if (machclk_freq == 0)
502 panic("altq_add: no cpu clock");
504 switch (a->scheduler) {
507 error = cbq_add_altq(a);
512 error = priq_add_altq(a);
517 error = hfsc_add_altq(a);
522 error = fairq_add_altq(a);
533 * remove a discipline or a queue
536 altq_remove(struct pf_altq *a)
540 if (a->qname[0] != 0)
541 return (altq_remove_queue(a));
543 switch (a->scheduler) {
546 error = cbq_remove_altq(a);
551 error = priq_remove_altq(a);
556 error = hfsc_remove_altq(a);
561 error = fairq_remove_altq(a);
572 * add a queue to the discipline
575 altq_add_queue(struct pf_altq *a)
579 switch (a->scheduler) {
582 error = cbq_add_queue(a);
587 error = priq_add_queue(a);
592 error = hfsc_add_queue(a);
597 error = fairq_add_queue(a);
608 * remove a queue from the discipline
611 altq_remove_queue(struct pf_altq *a)
615 switch (a->scheduler) {
618 error = cbq_remove_queue(a);
623 error = priq_remove_queue(a);
628 error = hfsc_remove_queue(a);
633 error = fairq_remove_queue(a);
644 * get queue statistics
647 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes)
651 switch (a->scheduler) {
654 error = cbq_getqstats(a, ubuf, nbytes);
659 error = priq_getqstats(a, ubuf, nbytes);
664 error = hfsc_getqstats(a, ubuf, nbytes);
669 error = fairq_getqstats(a, ubuf, nbytes);
680 * read and write diffserv field in IPv4 or IPv6 header
683 read_dsfield(struct mbuf *m, struct altq_pktattr *pktattr)
686 uint8_t ds_field = 0;
688 if (pktattr == NULL ||
689 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
692 /* verify that pattr_hdr is within the mbuf data */
693 for (m0 = m; m0 != NULL; m0 = m0->m_next) {
694 if ((pktattr->pattr_hdr >= m0->m_data) &&
695 (pktattr->pattr_hdr < m0->m_data + m0->m_len))
699 /* ick, pattr_hdr is stale */
700 pktattr->pattr_af = AF_UNSPEC;
702 kprintf("read_dsfield: can't locate header!\n");
707 if (pktattr->pattr_af == AF_INET) {
708 struct ip *ip = (struct ip *)pktattr->pattr_hdr;
711 return ((uint8_t)0); /* version mismatch! */
712 ds_field = ip->ip_tos;
715 else if (pktattr->pattr_af == AF_INET6) {
716 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
719 flowlabel = ntohl(ip6->ip6_flow);
720 if ((flowlabel >> 28) != 6)
721 return ((uint8_t)0); /* version mismatch! */
722 ds_field = (flowlabel >> 20) & 0xff;
729 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, uint8_t dsfield)
733 if (pktattr == NULL ||
734 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
737 /* verify that pattr_hdr is within the mbuf data */
738 for (m0 = m; m0 != NULL; m0 = m0->m_next) {
739 if ((pktattr->pattr_hdr >= m0->m_data) &&
740 (pktattr->pattr_hdr < m0->m_data + m0->m_len))
744 /* ick, pattr_hdr is stale */
745 pktattr->pattr_af = AF_UNSPEC;
747 kprintf("write_dsfield: can't locate header!\n");
752 if (pktattr->pattr_af == AF_INET) {
753 struct ip *ip = (struct ip *)pktattr->pattr_hdr;
758 return; /* version mismatch! */
760 dsfield |= old & 3; /* leave CU bits */
763 ip->ip_tos = dsfield;
765 * update checksum (from RFC1624)
766 * HC' = ~(~HC + ~m + m')
768 sum = ~ntohs(ip->ip_sum) & 0xffff;
769 sum += 0xff00 + (~old & 0xff) + dsfield;
770 sum = (sum >> 16) + (sum & 0xffff);
771 sum += (sum >> 16); /* add carry */
773 ip->ip_sum = htons(~sum & 0xffff);
776 else if (pktattr->pattr_af == AF_INET6) {
777 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
780 flowlabel = ntohl(ip6->ip6_flow);
781 if ((flowlabel >> 28) != 6)
782 return; /* version mismatch! */
783 flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
784 ip6->ip6_flow = htonl(flowlabel);
790 * high resolution clock support taking advantage of a machine dependent
791 * high resolution time counter (e.g., timestamp counter of intel pentium).
793 * - 64-bit-long monotonically-increasing counter
794 * - frequency range is 100M-4GHz (CPU speed)
796 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
797 #define MACHCLK_SHIFT 8
800 uint64_t machclk_freq = 0;
801 uint32_t machclk_per_tick = 0;
806 callout_init(&tbr_callout);
810 #if !defined(__i386__) || defined(ALTQ_NOPCC)
812 #elif defined(__DragonFly__)
814 #elif defined(__i386__)
815 /* check if TSC is available */
816 if (machclk_usepcc == 1 && (cpu_feature & CPUID_TSC) == 0)
820 if (machclk_usepcc == 0) {
821 /* emulate 256MHz using microtime() */
822 machclk_freq = 1000000LLU << MACHCLK_SHIFT;
823 machclk_per_tick = machclk_freq / hz;
825 kprintf("altq: emulate %juHz cpu clock\n", (uintmax_t)machclk_freq);
831 * if the clock frequency (of Pentium TSC or Alpha PCC) is
832 * accessible, just use it.
834 #ifdef _RDTSC_SUPPORTED_
835 if (cpu_feature & CPUID_TSC)
836 machclk_freq = (uint64_t)tsc_frequency;
840 * if we don't know the clock frequency, measure it.
842 if (machclk_freq == 0) {
844 struct timeval tv_start, tv_end;
845 uint64_t start, end, diff;
848 microtime(&tv_start);
849 start = read_machclk();
850 timo = hz; /* 1 sec */
851 tsleep(&wait, PCATCH, "init_machclk", timo);
853 end = read_machclk();
854 diff = (uint64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
855 + tv_end.tv_usec - tv_start.tv_usec;
857 machclk_freq = (end - start) * 1000000 / diff;
860 machclk_per_tick = machclk_freq / hz;
863 kprintf("altq: CPU clock: %juHz\n", (uintmax_t)machclk_freq);
872 if (machclk_usepcc) {
873 #ifdef _RDTSC_SUPPORTED_
876 panic("read_machclk");
882 val = (((uint64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
883 + tv.tv_usec) << MACHCLK_SHIFT);