xref: /freebsd/sys/netinet/ip_dummynet.h (revision 2e3f49888ec8851bafb22011533217487764fdb0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 1998-2010 Luigi Rizzo, Universita` di Pisa
5  * Portions Copyright (c) 2000 Akamba Corp.
6  * All rights reserved
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 THE AUTHOR 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
27  * SUCH DAMAGE.
28  */
29 
30 #ifndef _IP_DUMMYNET_H
31 #define _IP_DUMMYNET_H
32 #define NEW_AQM
33 /*
34  * Definition of the kernel-userland API for dummynet.
35  *
36  * Setsockopt() and getsockopt() pass a batch of objects, each
37  * of them starting with a "struct dn_id" which should fully identify
38  * the object and its relation with others in the sequence.
39  * The first object in each request should have
40  *	 type= DN_CMD_*, id = DN_API_VERSION.
41  * For other objects, type and subtype specify the object, len indicates
42  * the total length including the header, and 'id' identifies the specific
43  * object.
44  *
45  * Most objects are numbered with an identifier in the range 1..65535.
46  * DN_MAX_ID indicates the first value outside the range.
47  */
48 
49 #define	DN_API_VERSION	12500000
50 #define	DN_MAX_ID	0x10000
51 
52 struct dn_id {
53 	uint16_t	len;	/* total obj len including this header */
54 	uint8_t		type;
55 	uint8_t		subtype;
56 	uint32_t	id;	/* generic id */
57 };
58 
59 /*
60  * These values are in the type field of struct dn_id.
61  * To preserve the ABI, never rearrange the list or delete
62  * entries with the exception of DN_LAST
63  */
64 enum {
65 	DN_NONE = 0,
66 	DN_LINK = 1,
67 	DN_FS,
68 	DN_SCH,
69 	DN_SCH_I,
70 	DN_QUEUE,
71 	DN_DELAY_LINE,
72 	DN_PROFILE,
73 	DN_FLOW,		/* struct dn_flow */
74 	DN_TEXT,		/* opaque text is the object */
75 
76 	DN_CMD_CONFIG = 0x80,	/* objects follow */
77 	DN_CMD_DELETE,		/* subtype + list of entries */
78 	DN_CMD_GET,		/* subtype + list of entries */
79 	DN_CMD_FLUSH,
80 	/* for compatibility with FreeBSD 7.2/8 */
81 	DN_COMPAT_PIPE,
82 	DN_COMPAT_QUEUE,
83 	DN_GET_COMPAT,
84 
85 	/* special commands for emulation of sysctl variables */
86 	DN_SYSCTL_GET,
87 	DN_SYSCTL_SET,
88 #ifdef NEW_AQM
89 	/* subtypes used for setting/getting extra parameters.
90 	 * these subtypes used with IP_DUMMYNET3 command (get)
91 	 * and DN_TEXT (set). */
92 	DN_AQM_PARAMS, /* AQM extra params */
93 	DN_SCH_PARAMS, /* scheduler extra params */
94 #endif
95 	DN_LAST,
96 };
97 
98 enum { /* subtype for schedulers, flowset and the like */
99 	DN_SCHED_UNKNOWN = 0,
100 	DN_SCHED_FIFO = 1,
101 	DN_SCHED_WF2QP = 2,
102 	/* others are in individual modules */
103 };
104 
105 enum {	/* user flags */
106 	DN_HAVE_MASK	= 0x0001,	/* fs or sched has a mask */
107 	DN_NOERROR	= 0x0002,	/* do not report errors */
108 	DN_QHT_HASH	= 0x0004,	/* qht is a hash table */
109 	DN_QSIZE_BYTES	= 0x0008,	/* queue size is in bytes */
110 	DN_HAS_PROFILE	= 0x0010,	/* a link has a profile */
111 	DN_IS_RED	= 0x0020,
112 	DN_IS_GENTLE_RED= 0x0040,
113 	DN_IS_ECN	= 0x0080,
114 	#ifdef NEW_AQM
115 	DN_IS_AQM = 0x0100,     /* AQMs: e.g Codel & PIE */
116 	#endif
117 	DN_PIPE_CMD	= 0x1000,	/* pipe config... */
118 };
119 
120 /*
121  * link template.
122  */
123 struct dn_link {
124 	struct dn_id oid;
125 
126 	/*
127 	 * Userland sets bw and delay in bits/s and milliseconds.
128 	 * The kernel converts this back and forth to bits/tick and ticks.
129 	 * XXX what about burst ?
130 	 */
131 	int32_t		link_nr;
132 	uint32_t	bandwidth;	/* bit/s or bits/tick.   */
133 	int		delay;		/* ms and ticks */
134 	uint64_t	burst;		/* scaled. bits*Hz  XXX */
135 };
136 
137 /*
138  * A flowset, which is a template for flows. Contains parameters
139  * from the command line: id, target scheduler, queue sizes, plr,
140  * flow masks, buckets for the flow hash, and possibly scheduler-
141  * specific parameters (weight, quantum and so on).
142  */
143 struct dn_fs {
144 	struct dn_id oid;
145 	uint32_t fs_nr;		/* the flowset number */
146 	uint32_t flags;		/* userland flags */
147 	int qsize;		/* queue size in slots or bytes */
148 	int32_t pl_state;	/* packet loss state */
149 	uint32_t buckets;	/* buckets used for the queue hash table */
150 
151 	struct ipfw_flow_id flow_mask;
152 	uint32_t sched_nr;	/* the scheduler we attach to */
153 	/* generic scheduler parameters. Leave them at -1 if unset.
154 	 * Now we use 0: weight, 1: lmax, 2: priority
155 	 */
156 	int par[4];
157 
158 	/* RED/GRED parameters.
159 	 * weight and probabilities are in the range 0..1 represented
160 	 * in fixed point arithmetic with SCALE_RED decimal bits.
161 	 */
162 #define SCALE_RED	16
163 #define SCALE(x)	( (x) << SCALE_RED )
164 #define SCALE_VAL(x)	( (x) >> SCALE_RED )
165 #define SCALE_MUL(x,y)	( ( (x) * (y) ) >> SCALE_RED )
166 	int w_q ;		/* queue weight (scaled) */
167 	int max_th ;		/* maximum threshold for queue (scaled) */
168 	int min_th ;		/* minimum threshold for queue (scaled) */
169 	int max_p ;		/* maximum value for p_b (scaled) */
170 
171 	int32_t plr[4];		/* PLR, pkt loss rate (2^31-1 means 100%) */
172 };
173 
174 /*
175  * dn_flow collects flow_id and stats for queues and scheduler
176  * instances, and is used to pass these info to userland.
177  * oid.type/oid.subtype describe the object, oid.id is number
178  * of the parent object.
179  */
180 struct dn_flow {
181 	struct dn_id	oid;
182 	struct ipfw_flow_id fid;
183 	uint64_t	tot_pkts; /* statistics counters  */
184 	uint64_t	tot_bytes;
185 	uint32_t	length; /* Queue length, in packets */
186 	uint32_t	len_bytes; /* Queue length, in bytes */
187 	uint32_t	drops;
188 };
189 
190 /*
191  * Scheduler template, mostly indicating the name, number,
192  * sched_mask and buckets.
193  */
194 struct dn_sch {
195 	struct dn_id	oid;
196 	uint32_t	sched_nr; /* N, scheduler number */
197 	uint32_t	buckets; /* number of buckets for the instances */
198 	uint32_t	flags;	/* have_mask, ... */
199 
200 	char name[16];	/* null terminated */
201 	/* mask to select the appropriate scheduler instance */
202 	struct ipfw_flow_id sched_mask; /* M */
203 };
204 
205 /* A delay profile is attached to a link.
206  * Note that a profile, as any other object, cannot be longer than 2^16
207  */
208 #define	ED_MAX_SAMPLES_NO	1024
209 struct dn_profile {
210 	struct dn_id	oid;
211 	/* fields to simulate a delay profile */
212 #define ED_MAX_NAME_LEN		32
213 	char	name[ED_MAX_NAME_LEN];
214 	int	link_nr;
215 	int	loss_level;
216 	uint32_t	bandwidth;			// XXX use link bandwidth?
217 	int	samples_no;			/* actual len of samples[] */
218 	int	samples[ED_MAX_SAMPLES_NO];	/* may be shorter */
219 };
220 
221 #ifdef NEW_AQM
222 /* Extra parameters for AQM and scheduler.
223  * This struct is used to pass and retrieve parameters (configurations)
224  * to/from AQM and Scheduler.
225  */
226 struct dn_extra_parms {
227 	struct dn_id oid;
228 	char name[16];
229 	uint32_t nr;
230 #define DN_MAX_EXTRA_PARM	10
231 	int64_t par[DN_MAX_EXTRA_PARM];
232 };
233 #endif
234 
235 /*
236  * Overall structure of dummynet
237 
238 In dummynet, packets are selected with the firewall rules, and passed
239 to two different objects: PIPE or QUEUE (bad name).
240 
241 A QUEUE defines a classifier, which groups packets into flows
242 according to a 'mask', puts them into independent queues (one
243 per flow) with configurable size and queue management policy,
244 and passes flows to a scheduler:
245 
246                  (flow_mask|sched_mask)  sched_mask
247 	 +---------+   weight Wx  +-------------+
248          |         |->-[flow]-->--|             |-+
249     -->--| QUEUE x |   ...        |             | |
250          |         |->-[flow]-->--| SCHEDuler N | |
251 	 +---------+              |             | |
252 	     ...                  |             +--[LINK N]-->--
253 	 +---------+   weight Wy  |             | +--[LINK N]-->--
254          |         |->-[flow]-->--|             | |
255     -->--| QUEUE y |   ...        |             | |
256          |         |->-[flow]-->--|             | |
257 	 +---------+              +-------------+ |
258 	                            +-------------+
259 
260 Many QUEUE objects can connect to the same scheduler, each
261 QUEUE object can have its own set of parameters.
262 
263 In turn, the SCHEDuler 'forks' multiple instances according
264 to a 'sched_mask', each instance manages its own set of queues
265 and transmits on a private instance of a configurable LINK.
266 
267 A PIPE is a simplified version of the above, where there
268 is no flow_mask, and each scheduler instance handles a single queue.
269 
270 The following data structures (visible from userland) describe
271 the objects used by dummynet:
272 
273  + dn_link, contains the main configuration parameters related
274    to delay and bandwidth;
275  + dn_profile describes a delay profile;
276  + dn_flow describes the flow status (flow id, statistics)
277 
278  + dn_sch describes a scheduler
279  + dn_fs describes a flowset (msk, weight, queue parameters)
280 
281  *
282  */
283 
284 #endif /* _IP_DUMMYNET_H */
285