ip_dummynet.h 16 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
/*-
 * Copyright (c) 1998-2002, 2006 Luigi Rizzo, Universita` di Pisa
 * Portions Copyright (c) 2000 Akamba Corp.
 * All rights reserved
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 * $FreeBSD: src/sys/netinet/ip_dummynet.h,v 1.32.2.1 2005/01/31 23:26:35 imp Exp $
 */

#ifndef _IP_DUMMYNET_H
#define _IP_DUMMYNET_H

/*
 * Definition of dummynet data structures. In the structures, I decided
 * not to use the macros in <sys/queue.h> in the hope of making the code
 * easier to port to other architectures. The type of lists and queue we
 * use here is pretty simple anyways.
 */

/*
 * We start with a heap, which is used in the scheduler to decide when
 * to transmit packets etc.
 *
 * The key for the heap is used for two different values:
 *
 * 1. timer ticks- max 10K/second, so 32 bits are enough;
 *
 * 2. virtual times. These increase in steps of len/x, where len is the
 *    packet length, and x is either the weight of the flow, or the
 *    sum of all weights.
 *    If we limit to max 1000 flows and a max weight of 100, then
 *    x needs 17 bits. The packet size is 16 bits, so we can easily
 *    overflow if we do not allow errors.
 * So we use a key "dn_key" which is 64 bits. Some macros are used to
 * compare key values and handle wraparounds.
 * MAX64 returns the largest of two key values.
 * MY_M is used as a shift count when doing fixed point arithmetic
 * (a better name would be useful...).
 */
typedef u_int64_t dn_key ;      /* sorting key */
#define DN_KEY_LT(a,b)     ((int64_t)((a)-(b)) < 0)
#define DN_KEY_LEQ(a,b)    ((int64_t)((a)-(b)) <= 0)
#define DN_KEY_GT(a,b)     ((int64_t)((a)-(b)) > 0)
#define DN_KEY_GEQ(a,b)    ((int64_t)((a)-(b)) >= 0)
#define MAX64(x,y)  (( (int64_t) ( (y)-(x) )) > 0 ) ? (y) : (x)
#define MY_M	16 /* number of left shift to obtain a larger precision */

/*
 * XXX With this scaling, max 1000 flows, max weight 100, 1Gbit/s, the
 * virtual time wraps every 15 days.
 */

/*
 * The OFFSET_OF macro is used to return the offset of a field within
 * a structure. It is used by the heap management routines.
 */
#define OFFSET_OF(type, field) ((int)&( ((type *)0)->field) )

/*
 * The maximum hash table size for queues.  This value must be a power
 * of 2.
 */
#define DN_MAX_HASH_SIZE 65536

/*
 * A heap entry is made of a key and a pointer to the actual
 * object stored in the heap.
 * The heap is an array of dn_heap_entry entries, dynamically allocated.
 * Current size is "size", with "elements" actually in use.
 * The heap normally supports only ordered insert and extract from the top.
 * If we want to extract an object from the middle of the heap, we
 * have to know where the object itself is located in the heap (or we
 * need to scan the whole array). To this purpose, an object has a
 * field (int) which contains the index of the object itself into the
 * heap. When the object is moved, the field must also be updated.
 * The offset of the index in the object is stored in the 'offset'
 * field in the heap descriptor. The assumption is that this offset
 * is non-zero if we want to support extract from the middle.
 */
struct dn_heap_entry {
    dn_key key ;	/* sorting key. Topmost element is smallest one */
    void *object ;	/* object pointer */
} ;

struct dn_heap {
    int size ;
    int elements ;
    int offset ; /* XXX if > 0 this is the offset of direct ptr to obj */
    struct dn_heap_entry *p ;	/* really an array of "size" entries */
} ;

#ifdef _KERNEL
/*
 * Packets processed by dummynet have an mbuf tag associated with
 * them that carries their dummynet state.  This is used within
 * the dummynet code as well as outside when checking for special
 * processing requirements.
 */
struct dn_pkt_tag {
    struct ip_fw *rule;		/* matching rule */
    int dn_dir;			/* action when packet comes out. */
#define DN_TO_IP_OUT	1
#define DN_TO_IP_IN	2
#define DN_TO_BDG_FWD	3
#define DN_TO_ETH_DEMUX	4
#define DN_TO_ETH_OUT	5

    dn_key input_time;		/* when the pkt entered dummynet	*/
    dn_key output_time;		/* when the pkt is due for delivery	*/
    struct ifnet *ifp;		/* interface, for ip_output		*/
    int flags ;			/* flags, for ip_output (IPv6 ?)	*/
    int isDummy;
};
#endif /* _KERNEL */

/*
 * Overall structure of dummynet (with WF2Q+):

In dummynet, packets are selected with the firewall rules, and passed
to two different objects: PIPE or QUEUE.

A QUEUE is just a queue with configurable size and queue management
policy. It is also associated with a mask (to discriminate among
different flows), a weight (used to give different shares of the
bandwidth to different flows) and a "pipe", which essentially
supplies the transmit clock for all queues associated with that
pipe.

A PIPE emulates a fixed-bandwidth link, whose bandwidth is
configurable.  The "clock" for a pipe can come from either an
internal timer, or from the transmit interrupt of an interface.
A pipe is also associated with one (or more, if masks are used)
queue, where all packets for that pipe are stored.

The bandwidth available on the pipe is shared by the queues
associated with that pipe (only one in case the packet is sent
to a PIPE) according to the WF2Q+ scheduling algorithm and the
configured weights.

In general, incoming packets are stored in the appropriate queue,
which is then placed into one of a few heaps managed by a scheduler
to decide when the packet should be extracted.
The scheduler (a function called dummynet()) is run at every timer
tick, and grabs queues from the head of the heaps when they are
ready for processing.

There are three data structures definining a pipe and associated queues:

 + dn_pipe, which contains the main configuration parameters related
   to delay and bandwidth;
 + dn_flow_set, which contains WF2Q+ configuration, flow
   masks, plr and RED configuration;
 + dn_flow_queue, which is the per-flow queue (containing the packets)

Multiple dn_flow_set can be linked to the same pipe, and multiple
dn_flow_queue can be linked to the same dn_flow_set.
All data structures are linked in a linear list which is used for
housekeeping purposes.

During configuration, we create and initialize the dn_flow_set
and dn_pipe structures (a dn_pipe also contains a dn_flow_set).

At runtime: packets are sent to the appropriate dn_flow_set (either
WFQ ones, or the one embedded in the dn_pipe for fixed-rate flows),
which in turn dispatches them to the appropriate dn_flow_queue
(created dynamically according to the masks).

The transmit clock for fixed rate flows (ready_event()) selects the
dn_flow_queue to be used to transmit the next packet. For WF2Q,
wfq_ready_event() extract a pipe which in turn selects the right
flow using a number of heaps defined into the pipe itself.

 *
 */

/*
 * per flow queue. This contains the flow identifier, the queue
 * of packets, counters, and parameters used to support both RED and
 * WF2Q+.
 *
 * A dn_flow_queue is created and initialized whenever a packet for
 * a new flow arrives.
 */
struct dn_flow_queue {
    struct dn_flow_queue *next ;
    struct ipfw_flow_id id ;

    struct mbuf *head, *tail ;	/* queue of packets */
    u_int len ;
    u_int len_bytes ;
    u_long numbytes ;		/* credit for transmission (dynamic queues) */

    u_int64_t tot_pkts ;	/* statistics counters	*/
    u_int64_t tot_bytes ;
    u_int32_t drops ;

    int hash_slot ;		/* debugging/diagnostic */

    /* RED parameters */
    int avg ;                   /* average queue length est. (scaled) */
    int count ;                 /* arrivals since last RED drop */
    int random ;                /* random value (scaled) */
    u_int32_t q_time ;          /* start of queue idle time */

    /* WF2Q+ support */
    struct dn_flow_set *fs ;	/* parent flow set */
    int heap_pos ;		/* position (index) of struct in heap */
    dn_key sched_time ;		/* current time when queue enters ready_heap */

    dn_key S,F ;		/* start time, finish time */
    /*
     * Setting F < S means the timestamp is invalid. We only need
     * to test this when the queue is empty.
     */
} ;

/*
 * flow_set descriptor. Contains the "template" parameters for the
 * queue configuration, and pointers to the hash table of dn_flow_queue's.
 *
 * The hash table is an array of lists -- we identify the slot by
 * hashing the flow-id, then scan the list looking for a match.
 * The size of the hash table (buckets) is configurable on a per-queue
 * basis.
 *
 * A dn_flow_set is created whenever a new queue or pipe is created (in the
 * latter case, the structure is located inside the struct dn_pipe).
 */
struct dn_flow_set {
    struct dn_flow_set *next; /* next flow set in all_flow_sets list */

    u_short fs_nr ;             /* flow_set number       */
    u_short flags_fs;
#define DN_HAVE_FLOW_MASK	0x0001
#define DN_IS_RED		0x0002
#define DN_IS_GENTLE_RED	0x0004
#define DN_QSIZE_IS_BYTES	0x0008	/* queue size is measured in bytes */
#define DN_NOERROR		0x0010	/* do not report ENOBUFS on drops  */
#define DN_IS_PIPE		0x4000
#define DN_IS_QUEUE		0x8000

    struct dn_pipe *pipe ;	/* pointer to parent pipe */
    u_short parent_nr ;		/* parent pipe#, 0 if local to a pipe */

    int weight ;		/* WFQ queue weight */
    int qsize ;			/* queue size in slots or bytes */
    int plr ;			/* pkt loss rate (2^31-1 means 100%) */

    struct ipfw_flow_id flow_mask ;

    /* hash table of queues onto this flow_set */
    int rq_size ;		/* number of slots */
    int rq_elements ;		/* active elements */
    struct dn_flow_queue **rq;	/* array of rq_size entries */

    u_int32_t last_expired ;	/* do not expire too frequently */
    int backlogged ;		/* #active queues for this flowset */

        /* RED parameters */
#define SCALE_RED               16
#define SCALE(x)                ( (x) << SCALE_RED )
#define SCALE_VAL(x)            ( (x) >> SCALE_RED )
#define SCALE_MUL(x,y)          ( ( (x) * (y) ) >> SCALE_RED )
    int w_q ;			/* queue weight (scaled) */
    int max_th ;		/* maximum threshold for queue (scaled) */
    int min_th ;		/* minimum threshold for queue (scaled) */
    int max_p ;			/* maximum value for p_b (scaled) */
    u_int c_1 ;			/* max_p/(max_th-min_th) (scaled) */
    u_int c_2 ;			/* max_p*min_th/(max_th-min_th) (scaled) */
    u_int c_3 ;			/* for GRED, (1-max_p)/max_th (scaled) */
    u_int c_4 ;			/* for GRED, 1 - 2*max_p (scaled) */
    u_int * w_q_lookup ;	/* lookup table for computing (1-w_q)^t */
    u_int lookup_depth ;	/* depth of lookup table */
    int lookup_step ;		/* granularity inside the lookup table */
    int lookup_weight ;		/* equal to (1-w_q)^t / (1-w_q)^(t+1) */
    int avg_pkt_size ;		/* medium packet size */
    int max_pkt_size ;		/* max packet size */
} ;


/* distribution types */
#define DN_DIST_CONST_TIME	0x01
#define DN_DIST_CONST_RATE	0x02
#define DN_DIST_UNIFORM		0x04
#define DN_DIST_POISSON		0x08
#define DN_DIST_TABLE_RANDOM	0x10
#define DN_DIST_TABLE_DETERM	0x20

#define DN_TABLE_DIST \
	(DN_DIST_TABLE_RANDOM|DN_DIST_TABLE_DETERM|DN_DIST_POISSON)
#define DN_CONST_DIST (DN_DIST_CONST_RATE|DN_DIST_CONST_TIME)


/*
 * Pipe descriptor. Contains global parameters, delay-line queue,
 * and the flow_set used for fixed-rate queues.
 *
 * For WF2Q+ support it also has 3 heaps holding dn_flow_queue:
 *   not_eligible_heap, for queues whose start time is higher
 *	than the virtual time. Sorted by start time.
 *   scheduler_heap, for queues eligible for scheduling. Sorted by
 *	finish time.
 *   idle_heap, all flows that are idle and can be removed. We
 *	do that on each tick so we do not slow down too much
 *	operations during forwarding.
 *
 */

/* delay, bandwidth, loss parameters for dn_pipe */

struct dn_delay {
	int	delay ;			/* really, ticks	*/
	int	dist;		/* distribution type */
	int	mean;
	int	stddev;
	int    *quantum;	/* how frequently to recalculate delay */
	int	qentries;	/* entries in quantum table */
	int	quantum_expire;	/* its expiration date  */
	int    *table;		/* table of possible values */
	int	entries;	/* entries in table */ 
	int	tablepos;	/* current pos in table for deterministic */
};

struct dn_bw {
	int	bandwidth;		/* bits/sec */
	int	dist;			/* distribution type */
	int	mean;
	int	stddev;
	int    *quantum;	/* how frequently to recalculate bw */
	int	qentries;	/* entries in quantum table */
	int	quantum_expire;	/* its expiration date  */
	int    *table;		/* table of possible values */
	int	entries;	/* entries in table */
	int	tablepos;	/* current pos in table for deterministic */

    //pramod-CHANGES
    int backfill;
};

struct dn_loss {
	int	plr ;		/* pkt loss rate (2^31-1 means 100%) */
	int	dist;		/* distribution type */
	int	nextdroptime;	/* time to drop pkt at head of queue */
	int	maxinq;		/* max time left in queue */
	int	mean;
	int	stddev;
	int    *quantum;	/* how frequently to recalc loss rate */
	int	qentries;	/* entries in quantum table */
	int	quantum_expire;	/* its expiration date	*/
	int    *table;		/* table of possible values */
	int	entries;	/* entries in table */
	int	tablepos;	/* current pos in table for deterministic */
};

struct dn_pipe {		/* a pipe */
    struct dn_pipe *next ;

    int	pipe_nr ;		/* number	*/
    struct dn_delay delay;
    struct dn_bw bandwidth;
    struct dn_loss loss;

    struct	mbuf *head, *tail ;	/* packets in delay line */

    /* WF2Q+ */
    struct dn_heap scheduler_heap ; /* top extract - key Finish time*/
    struct dn_heap not_eligible_heap; /* top extract- key Start time */
    struct dn_heap idle_heap ; /* random extract - key Start=Finish time */

    dn_key V ;			/* virtual time */
    int sum;			/* sum of weights of all active sessions */
    int numbytes;		/* bits I can transmit (more or less). */

    u_int32_t drops ;  /* statistics of packet dropped from this pipe ....Code changed */

    dn_key sched_time ;		/* time pipe was scheduled in ready_heap */

    /*
     * When the tx clock come from an interface (if_name[0] != '\0'), its name
     * is stored below, whereas the ifp is filled when the rule is configured.
     */
    char if_name[IFNAMSIZ];
    struct ifnet *ifp ;
    int ready ; /* set if ifp != NULL and we got a signal from it */

    struct dn_flow_set fs ; /* used with fixed-rate flows */

    //pramod-CHANGES:
    int enableBackFill;
    int startBackFill;
    int outStandingBackFillBits;
    dn_key lastFillTick;
};

#ifdef _KERNEL
typedef	int ip_dn_ctl_t(struct sockopt *); /* raw_ip.c */
typedef	void ip_dn_ruledel_t(void *); /* ip_fw.c */
typedef	int ip_dn_io_t(struct mbuf *m, int pipe_nr, int dir,
	struct ip_fw_args *fwa);
extern	ip_dn_ctl_t *ip_dn_ctl_ptr;
extern	ip_dn_ruledel_t *ip_dn_ruledel_ptr;
extern	ip_dn_io_t *ip_dn_io_ptr;
#define	DUMMYNET_LOADED	(ip_dn_io_ptr != NULL)

/*
 * Return the IPFW rule associated with the dummynet tag; if any.
 * Make sure that the dummynet tag is not reused by lower layers.
 */
static __inline struct ip_fw *
ip_dn_claim_rule(struct mbuf *m)
{
	struct m_tag *mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
	if (mtag != NULL) {
		mtag->m_tag_id = PACKET_TAG_NONE;
		return (((struct dn_pkt_tag *)(mtag+1))->rule);
	} else
		return (NULL);
}
#endif
#endif /* _IP_DUMMYNET_H */