xref: /freebsd/sys/netinet/ip_dummynet.h (revision 77a0943ded95b9e6438f7db70c4a28e4d93946d4)
1 /*
2  * Copyright (c) 1998-2000 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 dummynet data structures.
35  * We first start with the heap which is used by the scheduler.
36  *
37  * Each list contains a set of parameters identifying the pipe, and
38  * a set of packets queued on the pipe itself.
39  *
40  * I could have used queue macros, but the management i have
41  * is pretty simple and this makes the code more portable.
42  */
43 
44 /*
45  * The key for the heap is used for two different values
46    1. timer ticks- max 10K/second, so 32 bits are enough
47    2. virtual times. These increase in steps of len/x, where len is the
48       packet length, and x is either the weight of the flow, or the
49       sum of all weights.
50       If we limit to max 1000 flows and a max weight of 100, then
51       x needs 17 bits. The packet size is 16 bits, so we can easily
52       overflow if we do not allow errors.
53 
54  */
55 typedef u_int64_t dn_key ;      /* sorting key */
56 #define DN_KEY_LT(a,b)     ((int64_t)((a)-(b)) < 0)
57 #define DN_KEY_LEQ(a,b)    ((int64_t)((a)-(b)) <= 0)
58 #define DN_KEY_GT(a,b)     ((int64_t)((a)-(b)) > 0)
59 #define DN_KEY_GEQ(a,b)    ((int64_t)((a)-(b)) >= 0)
60 /* XXX check names of next two macros */
61 #define MAX64(x,y)  (( (int64_t) ( (y)-(x) )) > 0 ) ? (y) : (x)
62 #define MY_M	16 /* number of left shift to obtain a larger precision */
63 /*
64  * XXX With this scaling, max 1000 flows, max weight 100, 1Gbit/s, the
65  * virtual time wraps every 15 days.
66  */
67 
68 #define OFFSET_OF(type, field) ((int)&( ((type *)0)->field) )
69 
70 struct dn_heap_entry {
71     dn_key key ;	/* sorting key. Topmost element is smallest one */
72     void *object ;	/* object pointer */
73 } ;
74 
75 struct dn_heap {
76     int size ;
77     int elements ;
78     int offset ; /* XXX if > 0 this is the offset of direct ptr to obj */
79     struct dn_heap_entry *p ;	/* really an array of "size" entries */
80 } ;
81 
82 /*
83  * MT_DUMMYNET is a new (fake) mbuf type that is prepended to the
84  * packet when it comes out of a pipe. The definition
85  * ought to go in /sys/sys/mbuf.h but here it is less intrusive.
86  */
87 
88 #define MT_DUMMYNET MT_CONTROL
89 
90 
91 /*
92  * struct dn_pkt identifies a packet in the dummynet queue. The
93  * first part is really an m_hdr for implementation purposes, and some
94  * fields are saved there. When passing the packet back to the ip_input/
95  * ip_output(), the struct is prepended to the mbuf chain with type
96  * MT_DUMMYNET, and contains the pointer to the matching rule.
97  */
98 struct dn_pkt {
99 	struct m_hdr hdr ;
100 #define dn_next	hdr.mh_nextpkt	/* next element in queue */
101 #define DN_NEXT(x)	(struct dn_pkt *)(x)->dn_next
102 #define dn_m	hdr.mh_next	/* packet to be forwarded */
103 #define dn_dir	hdr.mh_flags	/* action when pkt extracted from a queue */
104 #define DN_TO_IP_OUT	1
105 #define DN_TO_IP_IN	2
106 #define DN_TO_BDG_FWD	3
107 
108 	dn_key  output_time;    /* when the pkt is due for delivery */
109         struct ifnet *ifp;	/* interface, for ip_output		*/
110 	struct sockaddr_in *dn_dst ;
111         struct route ro;	/* route, for ip_output. MUST COPY	*/
112 	int flags ;		/* flags, for ip_output (IPv6 ?) */
113 };
114 
115 /*
116  * Overall structure (with WFQ):
117 
118 We have 3 data structures definining a pipe and associated queues:
119  + dn_pipe, which contains the main configuration parameters related
120    to delay and bandwidth
121  + dn_flow_set which contains WFQ configuration, flow
122    masks, plr and RED configuration
123  + dn_flow_queue which is the per-flow queue.
124  Multiple dn_flow_set can be linked to the same pipe, and multiple
125  dn_flow_queue can be linked to the same dn_flow_set.
126 
127  During configuration we set the dn_flow_set and dn_pipe parameters.
128  At runtime: packets are sent to the dn_flow_set (either WFQ ones, or
129  the one embedded in the dn_pipe for fixed-rate flows) which in turn
130  dispatches them to the appropriate dn_flow_queue (created dynamically
131  according to the masks).
132  The transmit clock for fixed rate flows (ready_event) selects the
133  dn_flow_queue to be used to transmit the next packet. For WF2Q,
134  wfq_ready_event() extract a pipe which in turn selects the right
135  flow using a number of heaps defined into the pipe.
136 
137  *
138  */
139 
140 /*
141  * We use per flow queues. Hashing is used to select the right slot,
142  * then we scan the list to match the flow-id.
143  */
144 struct dn_flow_queue {
145     struct dn_flow_queue *next ;
146     struct ipfw_flow_id id ;
147     struct dn_pkt *head, *tail ;	/* queue of packets */
148     u_int len ;
149     u_int len_bytes ;
150     long numbytes ;		/* credit for transmission (dynamic queues) */
151 
152     u_int64_t tot_pkts ;	/* statistics counters	*/
153     u_int64_t tot_bytes ;
154     u_int32_t drops ;
155     int hash_slot ;	/* debugging/diagnostic */
156 
157     /* RED parameters */
158     int avg ;                   /* average queue length est. (scaled) */
159     int count ;                 /* arrivals since last RED drop */
160     int random ;                /* random value (scaled) */
161     u_int32_t q_time ;          /* start of queue idle time */
162 
163     /* WF2Q+ support */
164     struct dn_flow_set *fs ; /* parent flow set */
165     int blh_pos ;	/* position in backlogged_heap */
166     dn_key sched_time ; /* current time when queue enters ready_heap */
167 
168     dn_key S,F ; /* start-time, finishing time */
169 } ;
170 
171 struct dn_flow_set {
172     struct dn_flow_set *next; /* next flow set in all_flow_sets list */
173 
174     u_short fs_nr ;             /* flow_set number       */
175     u_short flags_fs;
176 #define DN_HAVE_FLOW_MASK	0x0001
177 #define DN_IS_PIPE		0x4000
178 #define DN_IS_QUEUE		0x8000
179 #define DN_IS_RED		0x0002
180 #define DN_IS_GENTLE_RED	0x0004
181 #define DN_QSIZE_IS_BYTES	0x0008	/* queue measured in bytes */
182 
183     struct dn_pipe *pipe ;		/* pointer to parent pipe */
184     u_short parent_nr ;		/* parent pipe#, 0 if local to a pipe */
185 
186     int weight ; /* WFQ queue weight */
187     int qsize ;		/* queue size in slots or bytes */
188     int plr ;           /* pkt loss rate (2^31-1 means 100%) */
189 
190     struct ipfw_flow_id flow_mask ;
191     /* hash table of queues onto this flow_set */
192     int rq_size ;		/* number of slots */
193     int rq_elements ;		/* active elements */
194     struct dn_flow_queue **rq;	/* array of rq_size entries */
195     u_int32_t last_expired ;	/* do not expire too frequently */
196 	/* XXX some RED parameters as well ? */
197     int backlogged ;		/* #active queues for this flowset */
198 
199         /* RED parameters */
200 #define SCALE_RED               16
201 #define SCALE(x)                ( (x) << SCALE_RED )
202 #define SCALE_VAL(x)            ( (x) >> SCALE_RED )
203 #define SCALE_MUL(x,y)          ( ( (x) * (y) ) >> SCALE_RED )
204     int w_q ;               /* queue weight (scaled) */
205     int max_th ;            /* maximum threshold for queue (scaled) */
206     int min_th ;            /* minimum threshold for queue (scaled) */
207     int max_p ;             /* maximum value for p_b (scaled) */
208     u_int c_1 ;             /* max_p/(max_th-min_th) (scaled) */
209     u_int c_2 ;             /* max_p*min_th/(max_th-min_th) (scaled) */
210     u_int c_3 ;             /* for GRED, (1-max_p)/max_th (scaled) */
211     u_int c_4 ;             /* for GRED, 1 - 2*max_p (scaled) */
212     u_int * w_q_lookup ;    /* lookup table for computing (1-w_q)^t */
213     u_int lookup_depth ;    /* depth of lookup table */
214     int lookup_step ;       /* granularity inside the lookup table */
215     int lookup_weight ;     /* equal to (1-w_q)^t / (1-w_q)^(t+1) */
216     int avg_pkt_size ;      /* medium packet size */
217     int max_pkt_size ;      /* max packet size */
218 } ;
219 
220 /*
221  * Pipe descriptor. Contains global parameters, delay-line queue.
222  *
223  * For WF2Q support it also has 3 heaps holding dn_flow_queue:
224  *   not_eligible_heap, for queues whose start time is higher
225  *	than the virtual time. Sorted by start time.
226  *   scheduler_heap, for queues eligible for scheduling. Sorted by
227  *	finish time.
228  *   backlogged_heap, all flows in the two heaps above, sorted by
229  *	start time. This is used to compute the virtual time.
230  *
231  */
232 struct dn_pipe {			/* a pipe */
233 	struct dn_pipe *next ;
234 
235     int	pipe_nr ;		/* number	*/
236 	int	bandwidth;		/* really, bytes/tick.	*/
237 	int	delay ;			/* really, ticks	*/
238 
239     struct	dn_pkt *head, *tail ;	/* packets in delay line */
240 
241     /* WF2Q+ */
242     struct dn_heap scheduler_heap ; /* top extract - key Finish time*/
243     struct dn_heap not_eligible_heap; /* top extract- key Start time */
244     struct dn_heap backlogged_heap ; /* random extract - key Start time */
245 
246     dn_key V ; /* virtual time */
247     int sum;	/* sum of weights of all active sessions */
248     int numbytes;	/* bit i can transmit (more or less). */
249 
250     dn_key sched_time ; /* first time pipe is scheduled in ready_heap */
251 
252     /* the tx clock can come from an interface. In this case, the
253      * name is below, and the pointer is filled when the rule is
254      * configured. We identify this by setting the if_name to a
255      * non-empty string.
256      */
257     char if_name[16];
258     struct ifnet *ifp ;
259     int ready ; /* set if ifp != NULL and we got a signal from it */
260 
261     struct dn_flow_set fs ; /* used with fixed-rate flows */
262 };
263 
264 #ifdef _KERNEL
265 
266 MALLOC_DECLARE(M_IPFW);
267 
268 typedef int ip_dn_ctl_t __P((struct sockopt *)) ;
269 extern ip_dn_ctl_t *ip_dn_ctl_ptr;
270 
271 void dn_rule_delete(void *r);		/* used in ip_fw.c */
272 int dummynet_io(int pipe, int dir,
273 	struct mbuf *m, struct ifnet *ifp, struct route *ro,
274 	struct sockaddr_in * dst,
275 	struct ip_fw_chain *rule, int flags);
276 #endif
277 
278 #endif /* _IP_DUMMYNET_H */
279