xref: /freebsd/share/man/man4/multicast.4 (revision 5203edcdc553fda6caa1da8826a89b1a02dad1bf)
1.\" Copyright (c) 2001-2003 International Computer Science Institute
2.\"
3.\" Permission is hereby granted, free of charge, to any person obtaining a
4.\" copy of this software and associated documentation files (the "Software"),
5.\" to deal in the Software without restriction, including without limitation
6.\" the rights to use, copy, modify, merge, publish, distribute, sublicense,
7.\" and/or sell copies of the Software, and to permit persons to whom the
8.\" Software is furnished to do so, subject to the following conditions:
9.\"
10.\" The above copyright notice and this permission notice shall be included in
11.\" all copies or substantial portions of the Software.
12.\"
13.\" The names and trademarks of copyright holders may not be used in
14.\" advertising or publicity pertaining to the software without specific
15.\" prior permission. Title to copyright in this software and any associated
16.\" documentation will at all times remain with the copyright holders.
17.\"
18.\" THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19.\" IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20.\" FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21.\" AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22.\" LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23.\" FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
24.\" DEALINGS IN THE SOFTWARE.
25.\"
26.\" $FreeBSD$
27.\"
28.Dd September 4, 2003
29.Dt MULTICAST 4
30.Os
31.\"
32.Sh NAME
33.Nm multicast
34.Nd Multicast Routing
35.\"
36.Sh SYNOPSIS
37.Cd "options MROUTING"
38.Pp
39.In sys/types.h
40.In sys/socket.h
41.In netinet/in.h
42.In netinet/ip_mroute.h
43.In netinet6/ip6_mroute.h
44.Ft int
45.Fn getsockopt "int s" IPPROTO_IP MRT_INIT "void *optval" "socklen_t *optlen"
46.Ft int
47.Fn setsockopt "int s" IPPROTO_IP MRT_INIT "const void *optval" "socklen_t optlen"
48.Ft int
49.Fn getsockopt "int s" IPPROTO_IPV6 MRT6_INIT "void *optval" "socklen_t *optlen"
50.Ft int
51.Fn setsockopt "int s" IPPROTO_IPV6 MRT6_INIT "const void *optval" "socklen_t optlen"
52.Sh DESCRIPTION
53.Tn "Multicast routing"
54is used to efficiently propagate data
55packets to a set of multicast listeners in multipoint networks.
56If unicast is used to replicate the data to all listeners,
57then some of the network links may carry multiple copies of the same
58data packets.
59With multicast routing, the overhead is reduced to one copy
60(at most) per network link.
61.Pp
62All multicast-capable routers must run a common multicast routing
63protocol.
64The Distance Vector Multicast Routing Protocol (DVMRP)
65was the first developed multicast routing protocol.
66Later, other protocols such as Multicast Extensions to OSPF (MOSPF),
67Core Based Trees (CBT),
68Protocol Independent Multicast - Sparse Mode (PIM-SM),
69and Protocol Independent Multicast - Dense Mode (PIM-DM)
70were developed as well.
71.Pp
72To start multicast routing,
73the user must enable multicast forwarding in the kernel
74(see
75.Sx SYNOPSIS
76about the kernel configuration options),
77and must run a multicast routing capable user-level process.
78From developer's point of view,
79the programming guide described in the
80.Sx "Programming Guide"
81section should be used to control the multicast forwarding in the kernel.
82.\"
83.Ss Programming Guide
84This section provides information about the basic multicast routing API.
85The so-called
86.Dq advanced multicast API
87is described in the
88.Sx "Advanced Multicast API Programming Guide"
89section.
90.Pp
91First, a multicast routing socket must be open.
92That socket would be used
93to control the multicast forwarding in the kernel.
94Note that most operations below require certain privilege
95(i.e., root privilege):
96.Pp
97.Bd -literal
98/* IPv4 */
99int mrouter_s4;
100mrouter_s4 = socket(AF_INET, SOCK_RAW, IPPROTO_IGMP);
101.Ed
102.Pp
103.Bd -literal
104int mrouter_s6;
105mrouter_s6 = socket(AF_INET6, SOCK_RAW, IPPROTO_ICMPV6);
106.Ed
107.Pp
108Note that if the router needs to open an IGMP or ICMPv6 socket
109(in case of IPv4 and IPv6 respectively)
110for sending or receiving of IGMP or MLD multicast group membership messages,
111then the same mrouter_s4 or mrouter_s6 sockets should be used
112for sending and receiving respectively IGMP or MLD messages.
113In case of BSD-derived kernel, it may be possible to open separate sockets
114for IGMP or MLD messages only.
115However, some other kernels (e.g., Linux) require that the multicast
116routing socket must be used for sending and receiving of IGMP or MLD
117messages.
118Therefore, for portability reason the multicast
119routing socket should be reused for IGMP and MLD messages as well.
120.Pp
121After the multicast routing socket is open, it can be used to enable
122or disable multicast forwarding in the kernel:
123.Bd -literal
124/* IPv4 */
125int v = 1;        /* 1 to enable, or 0 to disable */
126setsockopt(mrouter_s4, IPPROTO_IP, MRT_INIT, (void *)&v, sizeof(v));
127.Ed
128.Pp
129.Bd -literal
130/* IPv6 */
131int v = 1;        /* 1 to enable, or 0 to disable */
132setsockopt(mrouter_s6, IPPROTO_IPV6, MRT6_INIT, (void *)&v, sizeof(v));
133\&...
134/* If necessary, filter all ICMPv6 messages */
135struct icmp6_filter filter;
136ICMP6_FILTER_SETBLOCKALL(&filter);
137setsockopt(mrouter_s6, IPPROTO_ICMPV6, ICMP6_FILTER, (void *)&filter,
138           sizeof(filter));
139.Ed
140.Pp
141After multicast forwarding is enabled, the multicast routing socket
142can be used to enable PIM processing in the kernel if we are running PIM-SM or
143PIM-DM
144(see
145.Xr pim 4 ) .
146.Pp
147For each network interface (e.g., physical or a virtual tunnel)
148that would be used for multicast forwarding, a corresponding
149multicast interface must be added to the kernel:
150.Bd -literal
151/* IPv4 */
152struct vifctl vc;
153memset(&vc, 0, sizeof(vc));
154/* Assign all vifctl fields as appropriate */
155vc.vifc_vifi = vif_index;
156vc.vifc_flags = vif_flags;
157vc.vifc_threshold = min_ttl_threshold;
158vc.vifc_rate_limit = max_rate_limit;
159memcpy(&vc.vifc_lcl_addr, &vif_local_address, sizeof(vc.vifc_lcl_addr));
160if (vc.vifc_flags & VIFF_TUNNEL)
161    memcpy(&vc.vifc_rmt_addr, &vif_remote_address,
162           sizeof(vc.vifc_rmt_addr));
163setsockopt(mrouter_s4, IPPROTO_IP, MRT_ADD_VIF, (void *)&vc,
164           sizeof(vc));
165.Ed
166.Pp
167The
168.Dq vif_index
169must be unique per vif.
170The
171.Dq vif_flags
172contains the
173.Dq VIFF_*
174flags as defined in <netinet/ip_mroute.h>.
175The
176.Dq min_ttl_threshold
177contains the minimum TTL a multicast data packet must have to be
178forwarded on that vif.
179Typically, it would have value of 1.
180The
181.Dq max_rate_limit
182contains the maximum rate (in bits/s) of the multicast data packets forwarded
183on that vif.
184Value of 0 means no limit.
185The
186.Dq vif_local_address
187contains the local IP address of the corresponding local interface.
188The
189.Dq vif_remote_address
190contains the remote IP address in case of DVMRP multicast tunnels.
191.Pp
192.Bd -literal
193/* IPv6 */
194struct mif6ctl mc;
195memset(&mc, 0, sizeof(mc));
196/* Assign all mif6ctl fields as appropriate */
197mc.mif6c_mifi = mif_index;
198mc.mif6c_flags = mif_flags;
199mc.mif6c_pifi = pif_index;
200setsockopt(mrouter_s6, IPPROTO_IPV6, MRT6_ADD_MIF, (void *)&mc,
201           sizeof(mc));
202.Ed
203.Pp
204The
205.Dq mif_index
206must be unique per vif.
207The
208.Dq mif_flags
209contains the
210.Dq MIFF_*
211flags as defined in <netinet6/ip6_mroute.h>.
212The
213.Dq pif_index
214is the physical interface index of the corresponding local interface.
215.Pp
216A multicast interface is deleted by:
217.Bd -literal
218/* IPv4 */
219vifi_t vifi = vif_index;
220setsockopt(mrouter_s4, IPPROTO_IP, MRT_DEL_VIF, (void *)&vifi,
221           sizeof(vifi));
222.Ed
223.Pp
224.Bd -literal
225/* IPv6 */
226mifi_t mifi = mif_index;
227setsockopt(mrouter_s6, IPPROTO_IPV6, MRT6_DEL_MIF, (void *)&mifi,
228           sizeof(mifi));
229.Ed
230.Pp
231After the multicast forwarding is enabled, and the multicast virtual
232interfaces are
233added, the kernel may deliver upcall messages (also called signals
234later in this text) on the multicast routing socket that was open
235earlier with
236.Dq MRT_INIT
237or
238.Dq MRT6_INIT .
239The IPv4 upcalls have
240.Dq struct igmpmsg
241header (see <netinet/ip_mroute.h>) with field
242.Dq im_mbz
243set to zero.
244Note that this header follows the structure of
245.Dq struct ip
246with the protocol field
247.Dq ip_p
248set to zero.
249The IPv6 upcalls have
250.Dq struct mrt6msg
251header (see <netinet6/ip6_mroute.h>) with field
252.Dq im6_mbz
253set to zero.
254Note that this header follows the structure of
255.Dq struct ip6_hdr
256with the next header field
257.Dq ip6_nxt
258set to zero.
259.Pp
260The upcall header contains field
261.Dq im_msgtype
262and
263.Dq im6_msgtype
264with the type of the upcall
265.Dq IGMPMSG_*
266and
267.Dq MRT6MSG_*
268for IPv4 and IPv6 respectively.
269The values of the rest of the upcall header fields
270and the body of the upcall message depend on the particular upcall type.
271.Pp
272If the upcall message type is
273.Dq IGMPMSG_NOCACHE
274or
275.Dq MRT6MSG_NOCACHE ,
276this is an indication that a multicast packet has reached the multicast
277router, but the router has no forwarding state for that packet.
278Typically, the upcall would be a signal for the multicast routing
279user-level process to install the appropriate Multicast Forwarding
280Cache (MFC) entry in the kernel.
281.Pp
282A MFC entry is added by:
283.Bd -literal
284/* IPv4 */
285struct mfcctl mc;
286memset(&mc, 0, sizeof(mc));
287memcpy(&mc.mfcc_origin, &source_addr, sizeof(mc.mfcc_origin));
288memcpy(&mc.mfcc_mcastgrp, &group_addr, sizeof(mc.mfcc_mcastgrp));
289mc.mfcc_parent = iif_index;
290for (i = 0; i < maxvifs; i++)
291    mc.mfcc_ttls[i] = oifs_ttl[i];
292setsockopt(mrouter_s4, IPPROTO_IP, MRT_ADD_MFC,
293           (void *)&mc, sizeof(mc));
294.Ed
295.Pp
296.Bd -literal
297/* IPv6 */
298struct mf6cctl mc;
299memset(&mc, 0, sizeof(mc));
300memcpy(&mc.mf6cc_origin, &source_addr, sizeof(mc.mf6cc_origin));
301memcpy(&mc.mf6cc_mcastgrp, &group_addr, sizeof(mf6cc_mcastgrp));
302mc.mf6cc_parent = iif_index;
303for (i = 0; i < maxvifs; i++)
304    if (oifs_ttl[i] > 0)
305        IF_SET(i, &mc.mf6cc_ifset);
306setsockopt(mrouter_s4, IPPROTO_IPV6, MRT6_ADD_MFC,
307           (void *)&mc, sizeof(mc));
308.Ed
309.Pp
310The
311.Dq source_addr
312and
313.Dq group_addr
314are the source and group address of the multicast packet (as set
315in the upcall message).
316The
317.Dq iif_index
318is the virtual interface index of the multicast interface the multicast
319packets for this specific source and group address should be received on.
320The
321.Dq oifs_ttl[]
322array contains the minimum TTL (per interface) a multicast packet
323should have to be forwarded on an outgoing interface.
324If the TTL value is zero, the corresponding interface is not included
325in the set of outgoing interfaces.
326Note that in case of IPv6 only the set of outgoing interfaces can
327be specified.
328.Pp
329A MFC entry is deleted by:
330.Bd -literal
331/* IPv4 */
332struct mfcctl mc;
333memset(&mc, 0, sizeof(mc));
334memcpy(&mc.mfcc_origin, &source_addr, sizeof(mc.mfcc_origin));
335memcpy(&mc.mfcc_mcastgrp, &group_addr, sizeof(mc.mfcc_mcastgrp));
336setsockopt(mrouter_s4, IPPROTO_IP, MRT_DEL_MFC,
337           (void *)&mc, sizeof(mc));
338.Ed
339.Pp
340.Bd -literal
341/* IPv6 */
342struct mf6cctl mc;
343memset(&mc, 0, sizeof(mc));
344memcpy(&mc.mf6cc_origin, &source_addr, sizeof(mc.mf6cc_origin));
345memcpy(&mc.mf6cc_mcastgrp, &group_addr, sizeof(mf6cc_mcastgrp));
346setsockopt(mrouter_s4, IPPROTO_IPV6, MRT6_DEL_MFC,
347           (void *)&mc, sizeof(mc));
348.Ed
349.Pp
350The following method can be used to get various statistics per
351installed MFC entry in the kernel (e.g., the number of forwarded
352packets per source and group address):
353.Bd -literal
354/* IPv4 */
355struct sioc_sg_req sgreq;
356memset(&sgreq, 0, sizeof(sgreq));
357memcpy(&sgreq.src, &source_addr, sizeof(sgreq.src));
358memcpy(&sgreq.grp, &group_addr, sizeof(sgreq.grp));
359ioctl(mrouter_s4, SIOCGETSGCNT, &sgreq);
360.Ed
361.Pp
362.Bd -literal
363/* IPv6 */
364struct sioc_sg_req6 sgreq;
365memset(&sgreq, 0, sizeof(sgreq));
366memcpy(&sgreq.src, &source_addr, sizeof(sgreq.src));
367memcpy(&sgreq.grp, &group_addr, sizeof(sgreq.grp));
368ioctl(mrouter_s6, SIOCGETSGCNT_IN6, &sgreq);
369.Ed
370.Pp
371The following method can be used to get various statistics per
372multicast virtual interface in the kernel (e.g., the number of forwarded
373packets per interface):
374.Bd -literal
375/* IPv4 */
376struct sioc_vif_req vreq;
377memset(&vreq, 0, sizeof(vreq));
378vreq.vifi = vif_index;
379ioctl(mrouter_s4, SIOCGETVIFCNT, &vreq);
380.Ed
381.Pp
382.Bd -literal
383/* IPv6 */
384struct sioc_mif_req6 mreq;
385memset(&mreq, 0, sizeof(mreq));
386mreq.mifi = vif_index;
387ioctl(mrouter_s6, SIOCGETMIFCNT_IN6, &mreq);
388.Ed
389.Pp
390.Ss Advanced Multicast API Programming Guide
391If we want to add new features in the kernel, it becomes difficult
392to preserve backward compatibility (binary and API),
393and at the same time to allow user-level processes to take advantage of
394the new features (if the kernel supports them).
395.Pp
396One of the mechanisms that allows us to preserve the backward
397compatibility is a sort of negotiation
398between the user-level process and the kernel:
399.Bl -enum
400.It
401The user-level process tries to enable in the kernel the set of new
402features (and the corresponding API) it would like to use.
403.It
404The kernel returns the (sub)set of features it knows about
405and is willing to be enabled.
406.It
407The user-level process uses only that set of features
408the kernel has agreed on.
409.El
410.\"
411.Pp
412To support backward compatibility, if the user-level process doesn't
413ask for any new features, the kernel defaults to the basic
414multicast API (see the
415.Sx "Programming Guide"
416section).
417.\" XXX: edit as appropriate after the advanced multicast API is
418.\" supported under IPv6
419Currently, the advanced multicast API exists only for IPv4;
420in the future there will be IPv6 support as well.
421.Pp
422Below is a summary of the expandable API solution.
423Note that all new options and structures are defined
424in <netinet/ip_mroute.h> and <netinet6/ip6_mroute.h>,
425unless stated otherwise.
426.Pp
427The user-level process uses new get/setsockopt() options to
428perform the API features negotiation with the kernel.
429This negotiation must be performed right after the multicast routing
430socket is open.
431The set of desired/allowed features is stored in a bitset
432(currently, in uint32_t; i.e., maximum of 32 new features).
433The new get/setsockopt() options are
434.Dq MRT_API_SUPPORT
435and
436.Dq MRT_API_CONFIG .
437Example:
438.Bd -literal
439uint32_t v;
440getsockopt(sock, IPPROTO_IP, MRT_API_SUPPORT, (void *)&v, sizeof(v));
441.Ed
442.Pp
443would set in
444.Dq v
445the pre-defined bits that the kernel API supports.
446The eight least significant bits in uint32_t are same as the
447eight possible flags
448.Dq MRT_MFC_FLAGS_*
449that can be used in
450.Dq mfcc_flags
451as part of the new definition of
452.Dq struct mfcctl
453(see below about those flags), which leaves 24 flags for other new features.
454The value returned by getsockopt(MRT_API_SUPPORT) is read-only; in other
455words, setsockopt(MRT_API_SUPPORT) would fail.
456.Pp
457To modify the API, and to set some specific feature in the kernel, then:
458.Bd -literal
459uint32_t v = MRT_MFC_FLAGS_DISABLE_WRONGVIF;
460if (setsockopt(sock, IPPROTO_IP, MRT_API_CONFIG, (void *)&v, sizeof(v))
461    != 0) {
462    return (ERROR);
463}
464if (v & MRT_MFC_FLAGS_DISABLE_WRONGVIF)
465    return (OK);	/* Success */
466else
467    return (ERROR);
468.Ed
469.Pp
470In other words, when setsockopt(MRT_API_CONFIG) is called, the
471argument to it specifies the desired set of features to
472be enabled in the API and the kernel.
473The return value in
474.Dq v
475is the actual (sub)set of features that were enabled in the kernel.
476To obtain later the same set of features that were enabled, then:
477.Bd -literal
478getsockopt(sock, IPPROTO_IP, MRT_API_CONFIG, (void *)&v, sizeof(v));
479.Ed
480.Pp
481The set of enabled features is global.
482In other words, setsockopt(MRT_API_CONFIG)
483should be called right after setsockopt(MRT_INIT).
484.Pp
485Currently, the following set of new features is defined:
486.Bd -literal
487#define	MRT_MFC_FLAGS_DISABLE_WRONGVIF (1 << 0) /* disable WRONGVIF signals */
488#define	MRT_MFC_FLAGS_BORDER_VIF   (1 << 1)  /* border vif              */
489#define MRT_MFC_RP                 (1 << 8)  /* enable RP address	*/
490#define MRT_MFC_BW_UPCALL          (1 << 9)  /* enable bw upcalls	*/
491.Ed
492.\" .Pp
493.\" In the future there might be:
494.\" .Bd -literal
495.\" #define MRT_MFC_GROUP_SPECIFIC     (1 << 10) /* allow (*,G) MFC entries */
496.\" .Ed
497.\" .Pp
498.\" to allow (*,G) MFC entries (i.e., group-specific entries) in the kernel.
499.\" For now this is left-out until it is clear whether
500.\" (*,G) MFC support is the preferred solution instead of something more generic
501.\" solution for example.
502.\"
503.\" 2. The newly defined struct mfcctl2.
504.\"
505.Pp
506The advanced multicast API uses a newly defined
507.Dq struct mfcctl2
508instead of the traditional
509.Dq struct mfcctl .
510The original
511.Dq struct mfcctl
512is kept as is.
513The new
514.Dq struct mfcctl2
515is:
516.Bd -literal
517/*
518 * The new argument structure for MRT_ADD_MFC and MRT_DEL_MFC overlays
519 * and extends the old struct mfcctl.
520 */
521struct mfcctl2 {
522        /* the mfcctl fields */
523        struct in_addr  mfcc_origin;       /* ip origin of mcasts       */
524        struct in_addr  mfcc_mcastgrp;     /* multicast group associated*/
525        vifi_t          mfcc_parent;       /* incoming vif              */
526        u_char          mfcc_ttls[MAXVIFS];/* forwarding ttls on vifs   */
527
528        /* extension fields */
529        uint8_t         mfcc_flags[MAXVIFS];/* the MRT_MFC_FLAGS_* flags*/
530        struct in_addr  mfcc_rp;            /* the RP address           */
531};
532.Ed
533.Pp
534The new fields are
535.Dq mfcc_flags[MAXVIFS]
536and
537.Dq mfcc_rp .
538Note that for compatibility reasons they are added at the end.
539.Pp
540The
541.Dq mfcc_flags[MAXVIFS]
542field is used to set various flags per
543interface per (S,G) entry.
544Currently, the defined flags are:
545.Bd -literal
546#define	MRT_MFC_FLAGS_DISABLE_WRONGVIF (1 << 0) /* disable WRONGVIF signals */
547#define	MRT_MFC_FLAGS_BORDER_VIF       (1 << 1) /* border vif          */
548.Ed
549.Pp
550The
551.Dq MRT_MFC_FLAGS_DISABLE_WRONGVIF
552flag is used to explicitly disable the
553.Dq IGMPMSG_WRONGVIF
554kernel signal at the (S,G) granularity if a multicast data packet
555arrives on the wrong interface.
556Usually, this signal is used to
557complete the shortest-path switch in case of PIM-SM multicast routing,
558or to trigger a PIM assert message.
559However, it should not be delivered for interfaces that are not in
560the outgoing interface set, and that are not expecting to
561become an incoming interface.
562Hence, if the
563.Dq MRT_MFC_FLAGS_DISABLE_WRONGVIF
564flag is set for some of the
565interfaces, then a data packet that arrives on that interface for
566that MFC entry will NOT trigger a WRONGVIF signal.
567If that flag is not set, then a signal is triggered (the default action).
568.Pp
569The
570.Dq MRT_MFC_FLAGS_BORDER_VIF
571flag is used to specify whether the Border-bit in PIM
572Register messages should be set (in case when the Register encapsulation
573is performed inside the kernel).
574If it is set for the special PIM Register kernel virtual interface
575(see
576.Xr pim 4 ) ,
577the Border-bit in the Register messages sent to the RP will be set.
578.Pp
579The remaining six bits are reserved for future usage.
580.Pp
581The
582.Dq mfcc_rp
583field is used to specify the RP address (in case of PIM-SM multicast routing)
584for a multicast
585group G if we want to perform kernel-level PIM Register encapsulation.
586The
587.Dq mfcc_rp
588field is used only if the
589.Dq MRT_MFC_RP
590advanced API flag/capability has been successfully set by
591setsockopt(MRT_API_CONFIG).
592.Pp
593.\"
594.\" 3. Kernel-level PIM Register encapsulation
595.\"
596If the
597.Dq MRT_MFC_RP
598flag was successfully set by
599setsockopt(MRT_API_CONFIG), then the kernel will attempt to perform
600the PIM Register encapsulation itself instead of sending the
601multicast data packets to user level (inside IGMPMSG_WHOLEPKT
602upcalls) for user-level encapsulation.
603The RP address would be taken from the
604.Dq mfcc_rp
605field
606inside the new
607.Dq struct mfcctl2 .
608However, even if the
609.Dq MRT_MFC_RP
610flag was successfully set, if the
611.Dq mfcc_rp
612field was set to
613.Dq INADDR_ANY ,
614then the
615kernel will still deliver an IGMPMSG_WHOLEPKT upcall with the
616multicast data packet to the user-level process.
617.Pp
618In addition, if the multicast data packet is too large to fit within
619a single IP packet after the PIM Register encapsulation (e.g., if
620its size was on the order of 65500 bytes), the data packet will be
621fragmented, and then each of the fragments will be encapsulated
622separately.
623Note that typically a multicast data packet can be that
624large only if it was originated locally from the same hosts that
625performs the encapsulation; otherwise the transmission of the
626multicast data packet over Ethernet for example would have
627fragmented it into much smaller pieces.
628.\"
629.\" Note that if this code is ported to IPv6, we may need the kernel to
630.\" perform MTU discovery to the RP, and keep those discoveries inside
631.\" the kernel so the encapsulating router may send back ICMP
632.\" Fragmentation Required if the size of the multicast data packet is
633.\" too large (see "Encapsulating data packets in the Register Tunnel"
634.\" in Section 4.4.1 in the PIM-SM spec
635.\" draft-ietf-pim-sm-v2-new-05.{txt,ps}).
636.\" For IPv4 we may be able to get away without it, but for IPv6 we need
637.\" that.
638.\"
639.\" 4. Mechanism for "multicast bandwidth monitoring and upcalls".
640.\"
641.Pp
642Typically, a multicast routing user-level process would need to know the
643forwarding bandwidth for some data flow.
644For example, the multicast routing process may want to timeout idle MFC
645entries, or in case of PIM-SM it can initiate (S,G) shortest-path switch if
646the bandwidth rate is above a threshold for example.
647.Pp
648The original solution for measuring the bandwidth of a dataflow was
649that a user-level process would periodically
650query the kernel about the number of forwarded packets/bytes per
651(S,G), and then based on those numbers it would estimate whether a source
652has been idle, or whether the source's transmission bandwidth is above a
653threshold.
654That solution is far from being scalable, hence the need for a new
655mechanism for bandwidth monitoring.
656.Pp
657Below is a description of the bandwidth monitoring mechanism.
658.Bl -bullet
659.It
660If the bandwidth of a data flow satisfies some pre-defined filter,
661the kernel delivers an upcall on the multicast routing socket
662to the multicast routing process that has installed that filter.
663.It
664The bandwidth-upcall filters are installed per (S,G).
665There can be
666more than one filter per (S,G).
667.It
668Instead of supporting all possible comparison operations
669(i.e., < <= == != > >= ), there is support only for the
670<= and >= operations,
671because this makes the kernel-level implementation simpler,
672and because practically we need only those two.
673Further, the missing operations can be simulated by secondary
674user-level filtering of those <= and >= filters.
675For example, to simulate !=, then we need to install filter
676.Dq bw <= 0xffffffff ,
677and after an
678upcall is received, we need to check whether
679.Dq measured_bw != expected_bw .
680.It
681The bandwidth-upcall mechanism is enabled by
682setsockopt(MRT_API_CONFIG) for the MRT_MFC_BW_UPCALL flag.
683.It
684The bandwidth-upcall filters are added/deleted by the new
685setsockopt(MRT_ADD_BW_UPCALL) and setsockopt(MRT_DEL_BW_UPCALL)
686respectively (with the appropriate
687.Dq struct bw_upcall
688argument of course).
689.El
690.Pp
691From application point of view, a developer needs to know about
692the following:
693.Bd -literal
694/*
695 * Structure for installing or delivering an upcall if the
696 * measured bandwidth is above or below a threshold.
697 *
698 * User programs (e.g. daemons) may have a need to know when the
699 * bandwidth used by some data flow is above or below some threshold.
700 * This interface allows the userland to specify the threshold (in
701 * bytes and/or packets) and the measurement interval. Flows are
702 * all packet with the same source and destination IP address.
703 * At the moment the code is only used for multicast destinations
704 * but there is nothing that prevents its use for unicast.
705 *
706 * The measurement interval cannot be shorter than some Tmin (currently, 3s).
707 * The threshold is set in packets and/or bytes per_interval.
708 *
709 * Measurement works as follows:
710 *
711 * For >= measurements:
712 * The first packet marks the start of a measurement interval.
713 * During an interval we count packets and bytes, and when we
714 * pass the threshold we deliver an upcall and we are done.
715 * The first packet after the end of the interval resets the
716 * count and restarts the measurement.
717 *
718 * For <= measurement:
719 * We start a timer to fire at the end of the interval, and
720 * then for each incoming packet we count packets and bytes.
721 * When the timer fires, we compare the value with the threshold,
722 * schedule an upcall if we are below, and restart the measurement
723 * (reschedule timer and zero counters).
724 */
725
726struct bw_data {
727        struct timeval  b_time;
728        uint64_t        b_packets;
729        uint64_t        b_bytes;
730};
731
732struct bw_upcall {
733        struct in_addr  bu_src;         /* source address            */
734        struct in_addr  bu_dst;         /* destination address       */
735        uint32_t        bu_flags;       /* misc flags (see below)    */
736#define BW_UPCALL_UNIT_PACKETS (1 << 0) /* threshold (in packets)    */
737#define BW_UPCALL_UNIT_BYTES   (1 << 1) /* threshold (in bytes)      */
738#define BW_UPCALL_GEQ          (1 << 2) /* upcall if bw >= threshold */
739#define BW_UPCALL_LEQ          (1 << 3) /* upcall if bw <= threshold */
740#define BW_UPCALL_DELETE_ALL   (1 << 4) /* delete all upcalls for s,d*/
741        struct bw_data  bu_threshold;   /* the bw threshold          */
742        struct bw_data  bu_measured;    /* the measured bw           */
743};
744
745/* max. number of upcalls to deliver together */
746#define BW_UPCALLS_MAX				128
747/* min. threshold time interval for bandwidth measurement */
748#define BW_UPCALL_THRESHOLD_INTERVAL_MIN_SEC	3
749#define BW_UPCALL_THRESHOLD_INTERVAL_MIN_USEC	0
750.Ed
751.Pp
752The
753.Dq bw_upcall
754structure is used as an argument to
755setsockopt(MRT_ADD_BW_UPCALL) and setsockopt(MRT_DEL_BW_UPCALL).
756Each setsockopt(MRT_ADD_BW_UPCALL) installs a filter in the kernel
757for the source and destination address in the
758.Dq bw_upcall
759argument,
760and that filter will trigger an upcall according to the following
761pseudo-algorithm:
762.Bd -literal
763 if (bw_upcall_oper IS ">=") {
764    if (((bw_upcall_unit & PACKETS == PACKETS) &&
765         (measured_packets >= threshold_packets)) ||
766        ((bw_upcall_unit & BYTES == BYTES) &&
767         (measured_bytes >= threshold_bytes)))
768       SEND_UPCALL("measured bandwidth is >= threshold");
769  }
770  if (bw_upcall_oper IS "<=" && measured_interval >= threshold_interval) {
771    if (((bw_upcall_unit & PACKETS == PACKETS) &&
772         (measured_packets <= threshold_packets)) ||
773        ((bw_upcall_unit & BYTES == BYTES) &&
774         (measured_bytes <= threshold_bytes)))
775       SEND_UPCALL("measured bandwidth is <= threshold");
776  }
777.Ed
778.Pp
779In the same
780.Dq bw_upcall
781the unit can be specified in both BYTES and PACKETS.
782However, the GEQ and LEQ flags are mutually exclusive.
783.Pp
784Basically, an upcall is delivered if the measured bandwidth is >= or
785<= the threshold bandwidth (within the specified measurement
786interval).
787For practical reasons, the smallest value for the measurement
788interval is 3 seconds.
789If smaller values are allowed, then the bandwidth
790estimation may be less accurate, or the potentially very high frequency
791of the generated upcalls may introduce too much overhead.
792For the >= operation, the answer may be known before the end of
793.Dq threshold_interval ,
794therefore the upcall may be delivered earlier.
795For the <= operation however, we must wait
796until the threshold interval has expired to know the answer.
797.Pp
798Example of usage:
799.Bd -literal
800struct bw_upcall bw_upcall;
801/* Assign all bw_upcall fields as appropriate */
802memset(&bw_upcall, 0, sizeof(bw_upcall));
803memcpy(&bw_upcall.bu_src, &source, sizeof(bw_upcall.bu_src));
804memcpy(&bw_upcall.bu_dst, &group, sizeof(bw_upcall.bu_dst));
805bw_upcall.bu_threshold.b_data = threshold_interval;
806bw_upcall.bu_threshold.b_packets = threshold_packets;
807bw_upcall.bu_threshold.b_bytes = threshold_bytes;
808if (is_threshold_in_packets)
809    bw_upcall.bu_flags |= BW_UPCALL_UNIT_PACKETS;
810if (is_threshold_in_bytes)
811    bw_upcall.bu_flags |= BW_UPCALL_UNIT_BYTES;
812do {
813    if (is_geq_upcall) {
814        bw_upcall.bu_flags |= BW_UPCALL_GEQ;
815        break;
816    }
817    if (is_leq_upcall) {
818        bw_upcall.bu_flags |= BW_UPCALL_LEQ;
819        break;
820    }
821    return (ERROR);
822} while (0);
823setsockopt(mrouter_s4, IPPROTO_IP, MRT_ADD_BW_UPCALL,
824          (void *)&bw_upcall, sizeof(bw_upcall));
825.Ed
826.Pp
827To delete a single filter, then use MRT_DEL_BW_UPCALL,
828and the fields of bw_upcall must be set
829exactly same as when MRT_ADD_BW_UPCALL was called.
830.Pp
831To delete all bandwidth filters for a given (S,G), then
832only the
833.Dq bu_src
834and
835.Dq bu_dst
836fields in
837.Dq struct bw_upcall
838need to be set, and then just set only the
839.Dq BW_UPCALL_DELETE_ALL
840flag inside field
841.Dq bw_upcall.bu_flags .
842.Pp
843The bandwidth upcalls are received by aggregating them in the new upcall
844message:
845.Bd -literal
846#define IGMPMSG_BW_UPCALL  4  /* BW monitoring upcall */
847.Ed
848.Pp
849This message is an array of
850.Dq struct bw_upcall
851elements (up to BW_UPCALLS_MAX = 128).
852The upcalls are
853delivered when there are 128 pending upcalls, or when 1 second has
854expired since the previous upcall (whichever comes first).
855In an
856.Dq struct upcall
857element, the
858.Dq bu_measured
859field is filled-in to
860indicate the particular measured values.
861However, because of the way
862the particular intervals are measured, the user should be careful how
863bu_measured.b_time is used.
864For example, if the
865filter is installed to trigger an upcall if the number of packets
866is >= 1, then
867.Dq bu_measured
868may have a value of zero in the upcalls after the
869first one, because the measured interval for >= filters is
870.Dq clocked
871by the forwarded packets.
872Hence, this upcall mechanism should not be used for measuring
873the exact value of the bandwidth of the forwarded data.
874To measure the exact bandwidth, the user would need to
875get the forwarded packets statistics with the ioctl(SIOCGETSGCNT)
876mechanism
877(see the
878.Sx Programming Guide
879section) .
880.Pp
881Note that the upcalls for a filter are delivered until the specific
882filter is deleted, but no more frequently than once per
883.Dq bu_threshold.b_time .
884For example, if the filter is specified to
885deliver a signal if bw >= 1 packet, the first packet will trigger a
886signal, but the next upcall will be triggered no earlier than
887.Dq bu_threshold.b_time
888after the previous upcall.
889.Pp
890.\"
891.Sh SEE ALSO
892.Xr getsockopt 2 ,
893.Xr recvfrom 2 ,
894.Xr recvmsg 2 ,
895.Xr setsockopt 2 ,
896.Xr socket 2 ,
897.Xr icmp6 4 ,
898.Xr inet 4 ,
899.Xr inet6 4 ,
900.Xr intro 4 ,
901.Xr ip 4 ,
902.Xr ip6 4 ,
903.Xr pim 4
904.\"
905.Pp
906.Sh AUTHORS
907.An -nosplit
908The original multicast code was written by
909.An David Waitzman
910(BBN Labs),
911and later modified by the following individuals:
912.An Steve Deering
913(Stanford),
914.An Mark J. Steiglitz
915(Stanford),
916.An Van Jacobson
917(LBL),
918.An Ajit Thyagarajan
919(PARC),
920.An Bill Fenner
921(PARC).
922The IPv6 multicast support was implemented by the KAME project
923(http://www.kame.net), and was based on the IPv4 multicast code.
924The advanced multicast API and the multicast bandwidth
925monitoring were implemented by
926.An Pavlin Radoslavov
927(ICSI)
928in collaboration with
929.An Chris Brown
930(NextHop).
931.Pp
932This manual page was written by
933.An Pavlin Radoslavov
934(ICSI).
935