xref: /freebsd/sys/netinet/ip_dummynet.h (revision ddd5b8e9b4d8957fce018c520657cdfa4ecffad3)
1 /*-
2  * Copyright (c) 1998-2010 Luigi Rizzo, Universita` di Pisa
3  * Portions Copyright (c) 2000 Akamba Corp.
4  * All rights reserved
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD$
28  */
29 
30 #ifndef _IP_DUMMYNET_H
31 #define _IP_DUMMYNET_H
32 
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 
89 	DN_LAST,
90 };
91 
92 enum { /* subtype for schedulers, flowset and the like */
93 	DN_SCHED_UNKNOWN = 0,
94 	DN_SCHED_FIFO = 1,
95 	DN_SCHED_WF2QP = 2,
96 	/* others are in individual modules */
97 };
98 
99 enum {	/* user flags */
100 	DN_HAVE_MASK	= 0x0001,	/* fs or sched has a mask */
101 	DN_NOERROR	= 0x0002,	/* do not report errors */
102 	DN_QHT_HASH	= 0x0004,	/* qht is a hash table */
103 	DN_QSIZE_BYTES	= 0x0008,	/* queue size is in bytes */
104 	DN_HAS_PROFILE	= 0x0010,	/* a link has a profile */
105 	DN_IS_RED	= 0x0020,
106 	DN_IS_GENTLE_RED= 0x0040,
107 	DN_PIPE_CMD	= 0x1000,	/* pipe config... */
108 };
109 
110 /*
111  * link template.
112  */
113 struct dn_link {
114 	struct dn_id oid;
115 
116 	/*
117 	 * Userland sets bw and delay in bits/s and milliseconds.
118 	 * The kernel converts this back and forth to bits/tick and ticks.
119 	 * XXX what about burst ?
120 	 */
121 	int32_t		link_nr;
122 	int		bandwidth;	/* bit/s or bits/tick.   */
123 	int		delay;		/* ms and ticks */
124 	uint64_t	burst;		/* scaled. bits*Hz  XXX */
125 };
126 
127 /*
128  * A flowset, which is a template for flows. Contains parameters
129  * from the command line: id, target scheduler, queue sizes, plr,
130  * flow masks, buckets for the flow hash, and possibly scheduler-
131  * specific parameters (weight, quantum and so on).
132  */
133 struct dn_fs {
134 	struct dn_id oid;
135 	uint32_t fs_nr;		/* the flowset number */
136 	uint32_t flags;		/* userland flags */
137 	int qsize;		/* queue size in slots or bytes */
138 	int32_t plr;		/* PLR, pkt loss rate (2^31-1 means 100%) */
139 	uint32_t buckets;	/* buckets used for the queue hash table */
140 
141 	struct ipfw_flow_id flow_mask;
142 	uint32_t sched_nr;	/* the scheduler we attach to */
143 	/* generic scheduler parameters. Leave them at -1 if unset.
144 	 * Now we use 0: weight, 1: lmax, 2: priority
145 	 */
146 	int par[4];
147 
148 	/* RED/GRED parameters.
149 	 * weight and probabilities are in the range 0..1 represented
150 	 * in fixed point arithmetic with SCALE_RED decimal bits.
151 	 */
152 #define SCALE_RED	16
153 #define SCALE(x)	( (x) << SCALE_RED )
154 #define SCALE_VAL(x)	( (x) >> SCALE_RED )
155 #define SCALE_MUL(x,y)	( ( (x) * (y) ) >> SCALE_RED )
156 	int w_q ;		/* queue weight (scaled) */
157 	int max_th ;		/* maximum threshold for queue (scaled) */
158 	int min_th ;		/* minimum threshold for queue (scaled) */
159 	int max_p ;		/* maximum value for p_b (scaled) */
160 
161 };
162 
163 /*
164  * dn_flow collects flow_id and stats for queues and scheduler
165  * instances, and is used to pass these info to userland.
166  * oid.type/oid.subtype describe the object, oid.id is number
167  * of the parent object.
168  */
169 struct dn_flow {
170 	struct dn_id	oid;
171 	struct ipfw_flow_id fid;
172 	uint64_t	tot_pkts; /* statistics counters  */
173 	uint64_t	tot_bytes;
174 	uint32_t	length; /* Queue length, in packets */
175 	uint32_t	len_bytes; /* Queue length, in bytes */
176 	uint32_t	drops;
177 };
178 
179 
180 /*
181  * Scheduler template, mostly indicating the name, number,
182  * sched_mask and buckets.
183  */
184 struct dn_sch {
185 	struct dn_id	oid;
186 	uint32_t	sched_nr; /* N, scheduler number */
187 	uint32_t	buckets; /* number of buckets for the instances */
188 	uint32_t	flags;	/* have_mask, ... */
189 
190 	char name[16];	/* null terminated */
191 	/* mask to select the appropriate scheduler instance */
192 	struct ipfw_flow_id sched_mask; /* M */
193 };
194 
195 
196 /* A delay profile is attached to a link.
197  * Note that a profile, as any other object, cannot be longer than 2^16
198  */
199 #define	ED_MAX_SAMPLES_NO	1024
200 struct dn_profile {
201 	struct dn_id	oid;
202 	/* fields to simulate a delay profile */
203 #define ED_MAX_NAME_LEN		32
204 	char	name[ED_MAX_NAME_LEN];
205 	int	link_nr;
206 	int	loss_level;
207 	int	bandwidth;			// XXX use link bandwidth?
208 	int	samples_no;			/* actual len of samples[] */
209 	int	samples[ED_MAX_SAMPLES_NO];	/* may be shorter */
210 };
211 
212 
213 
214 /*
215  * Overall structure of dummynet
216 
217 In dummynet, packets are selected with the firewall rules, and passed
218 to two different objects: PIPE or QUEUE (bad name).
219 
220 A QUEUE defines a classifier, which groups packets into flows
221 according to a 'mask', puts them into independent queues (one
222 per flow) with configurable size and queue management policy,
223 and passes flows to a scheduler:
224 
225                  (flow_mask|sched_mask)  sched_mask
226 	 +---------+   weight Wx  +-------------+
227          |         |->-[flow]-->--|             |-+
228     -->--| QUEUE x |   ...        |             | |
229          |         |->-[flow]-->--| SCHEDuler N | |
230 	 +---------+              |             | |
231 	     ...                  |             +--[LINK N]-->--
232 	 +---------+   weight Wy  |             | +--[LINK N]-->--
233          |         |->-[flow]-->--|             | |
234     -->--| QUEUE y |   ...        |             | |
235          |         |->-[flow]-->--|             | |
236 	 +---------+              +-------------+ |
237 	                            +-------------+
238 
239 Many QUEUE objects can connect to the same scheduler, each
240 QUEUE object can have its own set of parameters.
241 
242 In turn, the SCHEDuler 'forks' multiple instances according
243 to a 'sched_mask', each instance manages its own set of queues
244 and transmits on a private instance of a configurable LINK.
245 
246 A PIPE is a simplified version of the above, where there
247 is no flow_mask, and each scheduler instance handles a single queue.
248 
249 The following data structures (visible from userland) describe
250 the objects used by dummynet:
251 
252  + dn_link, contains the main configuration parameters related
253    to delay and bandwidth;
254  + dn_profile describes a delay profile;
255  + dn_flow describes the flow status (flow id, statistics)
256 
257  + dn_sch describes a scheduler
258  + dn_fs describes a flowset (msk, weight, queue parameters)
259 
260  *
261  */
262 
263 #endif /* _IP_DUMMYNET_H */
264