xref: /freebsd/sys/netpfil/ipfw/ip_dn_private.h (revision 045c8f526484cb3b97f5fd693987f4376fa43c5f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2010 Luigi Rizzo, Riccardo Panicucci, Universita` di Pisa
5  * All rights reserved
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 /*
30  * internal dummynet APIs.
31  *
32  * $FreeBSD$
33  */
34 
35 #ifndef _IP_DN_PRIVATE_H
36 #define _IP_DN_PRIVATE_H
37 
38 /* debugging support
39  * use ND() to remove debugging, D() to print a line,
40  * DX(level, ...) to print above a certain level
41  * If you redefine D() you are expected to redefine all.
42  */
43 #ifndef D
44 #define ND(fmt, ...) do {} while (0)
45 #define D1(fmt, ...) do {} while (0)
46 #define D(fmt, ...) printf("%-10s " fmt "\n",      \
47         __FUNCTION__, ## __VA_ARGS__)
48 #define DX(lev, fmt, ...) do {              \
49         if (V_dn_cfg.debug > lev) D(fmt, ## __VA_ARGS__); } while (0)
50 #endif
51 
52 MALLOC_DECLARE(M_DUMMYNET);
53 
54 #ifndef __linux__
55 #define div64(a, b)  ((int64_t)(a) / (int64_t)(b))
56 #endif
57 
58 #define DN_LOCK_INIT() do {				\
59 	mtx_init(&V_dn_cfg.uh_mtx, "dn_uh", NULL, MTX_DEF);	\
60 	mtx_init(&V_dn_cfg.bh_mtx, "dn_bh", NULL, MTX_DEF);	\
61 	} while (0)
62 #define DN_LOCK_DESTROY() do {				\
63 	mtx_destroy(&V_dn_cfg.uh_mtx);			\
64 	mtx_destroy(&V_dn_cfg.bh_mtx);			\
65 	} while (0)
66 #if 0 /* not used yet */
67 #define DN_UH_RLOCK()		mtx_lock(&V_dn_cfg.uh_mtx)
68 #define DN_UH_RUNLOCK()		mtx_unlock(&V_dn_cfg.uh_mtx)
69 #define DN_UH_WLOCK()		mtx_lock(&V_dn_cfg.uh_mtx)
70 #define DN_UH_WUNLOCK()		mtx_unlock(&V_dn_cfg.uh_mtx)
71 #define DN_UH_LOCK_ASSERT()	mtx_assert(&V_dn_cfg.uh_mtx, MA_OWNED)
72 #endif
73 
74 #define DN_BH_RLOCK()		mtx_lock(&V_dn_cfg.uh_mtx)
75 #define DN_BH_RUNLOCK()		mtx_unlock(&V_dn_cfg.uh_mtx)
76 #define DN_BH_WLOCK()		mtx_lock(&V_dn_cfg.uh_mtx)
77 #define DN_BH_WUNLOCK()		mtx_unlock(&V_dn_cfg.uh_mtx)
78 #define DN_BH_LOCK_ASSERT()	mtx_assert(&V_dn_cfg.uh_mtx, MA_OWNED)
79 
80 SLIST_HEAD(dn_fsk_head, dn_fsk);
81 
82 struct mq {	/* a basic queue of packets*/
83         struct mbuf *head, *tail;
84 	int count;
85 };
86 
87 static inline void
88 set_oid(struct dn_id *o, int type, int len)
89 {
90         o->type = type;
91         o->len = len;
92         o->subtype = 0;
93 }
94 
95 /*
96  * configuration and data for a dummynet instance
97  *
98  * When a configuration is modified from userland, 'id' is incremented
99  * so we can use the value to check for stale pointers.
100  */
101 struct dn_parms {
102 	uint32_t	id;		/* configuration version */
103 
104 	/* defaults (sysctl-accessible) */
105 	int	red_lookup_depth;
106 	int	red_avg_pkt_size;
107 	int	red_max_pkt_size;
108 	int	hash_size;
109 	int	max_hash_size;
110 	long	byte_limit;		/* max queue sizes */
111 	long	slot_limit;
112 
113 	int	io_fast;
114 	int	debug;
115 
116 	/* timekeeping */
117 	struct timeval prev_t;		/* last time dummynet_tick ran */
118 	struct dn_heap	evheap;		/* scheduled events */
119 
120 	long	tick_last;		/* Last tick duration (usec). */
121 	long	tick_delta;		/* Last vs standard tick diff (usec). */
122 	long	tick_delta_sum;	/* Accumulated tick difference (usec).*/
123 	long	tick_adjustment;	/* Tick adjustments done. */
124 	long	tick_lost;		/* Lost(coalesced) ticks number. */
125 	/* Adjusted vs non-adjusted curr_time difference (ticks). */
126 	long	tick_diff;
127 
128 	/* counters of objects -- used for reporting space */
129 	int	schk_count;
130 	int	si_count;
131 	int	fsk_count;
132 	int	queue_count;
133 
134 	/* packet counters */
135 	unsigned long	io_pkt;
136 	unsigned long	io_pkt_fast;
137 	unsigned long	io_pkt_drop;
138 
139 	/* ticks and other stuff */
140 	uint64_t	curr_time;
141 	/* flowsets and schedulers are in hash tables, with 'hash_size'
142 	 * buckets. fshash is looked up at every packet arrival
143 	 * so better be generous if we expect many entries.
144 	 */
145 	struct dn_ht	*fshash;
146 	struct dn_ht	*schedhash;
147 	/* list of flowsets without a scheduler -- use sch_chain */
148 	struct dn_fsk_head	fsu;	/* list of unlinked flowsets */
149 
150 	/* Store the fs/sch to scan when draining. The value is the
151 	 * bucket number of the hash table. Expire can be disabled
152 	 * with net.inet.ip.dummynet.expire=0, or it happens every
153 	 * expire ticks.
154 	 **/
155 	int drain_fs;
156 	int drain_sch;
157 	uint32_t expire;
158 	uint32_t expire_cycle;	/* tick count */
159 
160 	int init_done;
161 
162 #ifdef _KERNEL
163 	/*
164 	 * This file is normally used in the kernel, unless we do
165 	 * some userland tests, in which case we do not need a mtx.
166 	 * uh_mtx arbitrates between system calls and also
167 	 * protects fshash, schedhash and fsunlinked.
168 	 * These structures are readonly for the lower half.
169 	 * bh_mtx protects all other structures which may be
170 	 * modified upon packet arrivals
171 	 */
172 #if defined( __linux__ ) || defined( _WIN32 )
173 	spinlock_t uh_mtx;
174 	spinlock_t bh_mtx;
175 #else
176 	struct mtx uh_mtx;
177 	struct mtx bh_mtx;
178 #endif
179 
180 #endif /* _KERNEL */
181 };
182 
183 /*
184  * Delay line, contains all packets on output from a link.
185  * Every scheduler instance has one.
186  */
187 struct delay_line {
188 	struct dn_id oid;
189 	struct dn_sch_inst *si;
190 	struct mq mq;
191 };
192 
193 /*
194  * The kernel side of a flowset. It is linked in a hash table
195  * of flowsets, and in a list of children of their parent scheduler.
196  * qht is either the queue or (if HAVE_MASK) a hash table queues.
197  * Note that the mask to use is the (flow_mask|sched_mask), which
198  * changes as we attach/detach schedulers. So we store it here.
199  *
200  * XXX If we want to add scheduler-specific parameters, we need to
201  * put them in external storage because the scheduler may not be
202  * available when the fsk is created.
203  */
204 struct dn_fsk { /* kernel side of a flowset */
205 	struct dn_fs fs;
206 	SLIST_ENTRY(dn_fsk) fsk_next;	/* hash chain for fshash */
207 
208 	struct ipfw_flow_id fsk_mask;
209 
210 	/* qht is a hash table of queues, or just a single queue
211 	 * a bit in fs.flags tells us which one
212 	 */
213 	struct dn_ht	*qht;
214 	struct dn_schk *sched;		/* Sched we are linked to */
215 	SLIST_ENTRY(dn_fsk) sch_chain;	/* list of fsk attached to sched */
216 
217 	/* bucket index used by drain routine to drain queues for this
218 	 * flowset
219 	 */
220 	int drain_bucket;
221 	/* Parameter realted to RED / GRED */
222 	/* original values are in dn_fs*/
223 	int w_q ;		/* queue weight (scaled) */
224 	int max_th ;		/* maximum threshold for queue (scaled) */
225 	int min_th ;		/* minimum threshold for queue (scaled) */
226 	int max_p ;		/* maximum value for p_b (scaled) */
227 
228 	u_int c_1 ;		/* max_p/(max_th-min_th) (scaled) */
229 	u_int c_2 ;		/* max_p*min_th/(max_th-min_th) (scaled) */
230 	u_int c_3 ;		/* for GRED, (1-max_p)/max_th (scaled) */
231 	u_int c_4 ;		/* for GRED, 1 - 2*max_p (scaled) */
232 	u_int * w_q_lookup ;	/* lookup table for computing (1-w_q)^t */
233 	u_int lookup_depth ;	/* depth of lookup table */
234 	int lookup_step ;	/* granularity inside the lookup table */
235 	int lookup_weight ;	/* equal to (1-w_q)^t / (1-w_q)^(t+1) */
236 	int avg_pkt_size ;	/* medium packet size */
237 	int max_pkt_size ;	/* max packet size */
238 #ifdef NEW_AQM
239 	struct dn_aqm *aqmfp;	/* Pointer to AQM functions */
240 	void *aqmcfg;	/* configuration parameters for AQM */
241 #endif
242 };
243 
244 /*
245  * A queue is created as a child of a flowset unless it belongs to
246  * a !MULTIQUEUE scheduler. It is normally in a hash table in the
247  * flowset. fs always points to the parent flowset.
248  * si normally points to the sch_inst, unless the flowset has been
249  * detached from the scheduler -- in this case si == NULL and we
250  * should not enqueue.
251  */
252 struct dn_queue {
253 	struct dn_flow ni;	/* oid, flow_id, stats */
254 	struct mq mq;	/* packets queue */
255 	struct dn_sch_inst *_si;	/* owner scheduler instance */
256 	SLIST_ENTRY(dn_queue) q_next; /* hash chain list for qht */
257 	struct dn_fsk *fs;		/* parent flowset. */
258 
259 	/* RED parameters */
260 	int avg;		/* average queue length est. (scaled) */
261 	int count;		/* arrivals since last RED drop */
262 	int random;		/* random value (scaled) */
263 	uint64_t q_time;	/* start of queue idle time */
264 #ifdef NEW_AQM
265 	void *aqm_status;	/* per-queue status variables*/
266 #endif
267 
268 };
269 
270 /*
271  * The kernel side of a scheduler. Contains the userland config,
272  * a link, pointer to extra config arguments from command line,
273  * kernel flags, and a pointer to the scheduler methods.
274  * It is stored in a hash table, and holds a list of all
275  * flowsets and scheduler instances.
276  * XXX sch must be at the beginning, see schk_hash().
277  */
278 struct dn_schk {
279 	struct dn_sch sch;
280 	struct dn_alg *fp;	/* Pointer to scheduler functions */
281 	struct dn_link link;	/* The link, embedded */
282 	struct dn_profile *profile; /* delay profile, if any */
283 	struct dn_id *cfg;	/* extra config arguments */
284 
285 	SLIST_ENTRY(dn_schk) schk_next;  /* hash chain for schedhash */
286 
287 	struct dn_fsk_head fsk_list;  /* all fsk linked to me */
288 	struct dn_fsk *fs;	/* Flowset for !MULTIQUEUE */
289 
290 	/* bucket index used by the drain routine to drain the scheduler
291 	 * instance for this flowset.
292 	 */
293 	int drain_bucket;
294 
295 	/* Hash table of all instances (through sch.sched_mask)
296 	 * or single instance if no mask. Always valid.
297 	 */
298 	struct dn_ht	*siht;
299 };
300 
301 /*
302  * Scheduler instance.
303  * Contains variables and all queues relative to a this instance.
304  * This struct is created a runtime.
305  */
306 struct dn_sch_inst {
307 	struct dn_flow	ni;	/* oid, flowid and stats */
308 	SLIST_ENTRY(dn_sch_inst) si_next; /* hash chain for siht */
309 	struct delay_line dline;
310 	struct dn_schk *sched;	/* the template */
311 	int		kflags;	/* DN_ACTIVE */
312 
313 	int64_t	credit;		/* bits I can transmit (more or less). */
314 	uint64_t sched_time;	/* time link was scheduled in ready_heap */
315 	uint64_t idle_time;	/* start of scheduler instance idle time */
316 
317 	/* q_count is the number of queues that this instance is using.
318 	 * The counter is incremented or decremented when
319 	 * a reference from the queue is created or deleted.
320 	 * It is used to make sure that a scheduler instance can be safely
321 	 * deleted by the drain routine. See notes below.
322 	 */
323 	int q_count;
324 
325 };
326 
327 /*
328  * NOTE about object drain.
329  * The system will automatically (XXX check when) drain queues and
330  * scheduler instances when they are idle.
331  * A queue is idle when it has no packets; an instance is idle when
332  * it is not in the evheap heap, and the corresponding delay line is empty.
333  * A queue can be safely deleted when it is idle because of the scheduler
334  * function xxx_free_queue() will remove any references to it.
335  * An instance can be only deleted when no queues reference it. To be sure
336  * of that, a counter (q_count) stores the number of queues that are pointing
337  * to the instance.
338  *
339  * XXX
340  * Order of scan:
341  * - take all flowset in a bucket for the flowset hash table
342  * - take all queues in a bucket for the flowset
343  * - increment the queue bucket
344  * - scan next flowset bucket
345  * Nothing is done if a bucket contains no entries.
346  *
347  * The same schema is used for sceduler instances
348  */
349 
350 /* kernel-side flags. Linux has DN_DELETE in fcntl.h
351  */
352 enum {
353 	/* 1 and 2 are reserved for the SCAN flags */
354 	DN_DESTROY	= 0x0004, /* destroy */
355 	DN_DELETE_FS	= 0x0008, /* destroy flowset */
356 	DN_DETACH	= 0x0010,
357 	DN_ACTIVE	= 0x0020, /* object is in evheap */
358 	DN_F_DLINE	= 0x0040, /* object is a delay line */
359 	DN_DEL_SAFE	= 0x0080, /* delete a queue only if no longer needed
360 				   * by scheduler */
361 	DN_QHT_IS_Q	= 0x0100, /* in flowset, qht is a single queue */
362 };
363 
364 /*
365  * Packets processed by dummynet have an mbuf tag associated with
366  * them that carries their dummynet state.
367  * Outside dummynet, only the 'rule' field is relevant, and it must
368  * be at the beginning of the structure.
369  */
370 struct dn_pkt_tag {
371 	struct ipfw_rule_ref rule;	/* matching rule	*/
372 
373 	/* second part, dummynet specific */
374 	int dn_dir;		/* action when packet comes out.*/
375 				/* see ip_fw_private.h		*/
376 	uint64_t output_time;	/* when the pkt is due for delivery*/
377 	struct ifnet *ifp;	/* interface, for ip_output	*/
378 	struct _ip6dn_args ip6opt;	/* XXX ipv6 options	*/
379 	uint16_t iphdr_off;	/* IP header offset for mtodo()	*/
380 };
381 
382 /*
383  * Possible values for dn_dir. XXXGL: this needs to be reviewed
384  * and converted to same values ip_fw_args.flags use.
385  */
386 enum {
387 	DIR_OUT =	0,
388 	DIR_IN =	1,
389 	DIR_FWD =	2,
390 	DIR_DROP =	3,
391 	PROTO_LAYER2 =	0x4, /* set for layer 2 */
392 	PROTO_IPV4 =	0x08,
393 	PROTO_IPV6 =	0x10,
394 	PROTO_IFB =	0x0c, /* layer2 + ifbridge */
395 };
396 
397 //extern struct dn_parms V_dn_cfg;
398 VNET_DECLARE(struct dn_parms, dn_cfg);
399 #define V_dn_cfg	VNET(dn_cfg)
400 
401 int dummynet_io(struct mbuf **, struct ip_fw_args *);
402 void dummynet_sched_lock(void);
403 void dummynet_sched_unlock(void);
404 void dummynet_task(void *context, int pending);
405 void dn_reschedule(void);
406 struct dn_pkt_tag * dn_tag_get(struct mbuf *m);
407 
408 struct dn_queue *ipdn_q_find(struct dn_fsk *, struct dn_sch_inst *,
409         struct ipfw_flow_id *);
410 struct dn_sch_inst *ipdn_si_find(struct dn_schk *, struct ipfw_flow_id *);
411 
412 /*
413  * copy_range is a template for requests for ranges of pipes/queues/scheds.
414  * The number of ranges is variable and can be derived by o.len.
415  * As a default, we use a small number of entries so that the struct
416  * fits easily on the stack and is sufficient for most common requests.
417  */
418 #define DEFAULT_RANGES	5
419 struct copy_range {
420         struct dn_id o;
421         uint32_t	r[ 2 * DEFAULT_RANGES ];
422 };
423 
424 struct copy_args {
425 	char **start;
426 	char *end;
427 	int flags;
428 	int type;
429 	struct copy_range *extra;	/* extra filtering */
430 };
431 
432 struct sockopt;
433 int ip_dummynet_compat(struct sockopt *sopt);
434 int dummynet_get(struct sockopt *sopt, void **compat);
435 int dn_c_copy_q (void *_ni, void *arg);
436 int dn_c_copy_pipe(struct dn_schk *s, struct copy_args *a, int nq);
437 int dn_c_copy_fs(struct dn_fsk *f, struct copy_args *a, int nq);
438 int dn_compat_copy_queue(struct copy_args *a, void *_o);
439 int dn_compat_copy_pipe(struct copy_args *a, void *_o);
440 int copy_data_helper_compat(void *_o, void *_arg);
441 int dn_compat_calc_size(void);
442 int do_config(void *p, size_t l);
443 
444 /* function to drain idle object */
445 void dn_drain_scheduler(void);
446 void dn_drain_queue(void);
447 
448 #ifdef NEW_AQM
449 int ecn_mark(struct mbuf* m);
450 
451 /* moved from ip_dn_io.c to here to be available for AQMs modules*/
452 static inline void
453 mq_append(struct mq *q, struct mbuf *m)
454 {
455 #ifdef USERSPACE
456 	// buffers from netmap need to be copied
457 	// XXX note that the routine is not expected to fail
458 	ND("append %p to %p", m, q);
459 	if (m->m_flags & M_STACK) {
460 		struct mbuf *m_new;
461 		void *p;
462 		int l, ofs;
463 
464 		ofs = m->m_data - m->__m_extbuf;
465 		// XXX allocate
466 		MGETHDR(m_new, M_NOWAIT, MT_DATA);
467 		ND("*** WARNING, volatile buf %p ext %p %d dofs %d m_new %p",
468 			m, m->__m_extbuf, m->__m_extlen, ofs, m_new);
469 		p = m_new->__m_extbuf;	/* new pointer */
470 		l = m_new->__m_extlen;	/* new len */
471 		if (l <= m->__m_extlen) {
472 			panic("extlen too large");
473 		}
474 
475 		*m_new = *m;	// copy
476 		m_new->m_flags &= ~M_STACK;
477 		m_new->__m_extbuf = p; // point to new buffer
478 		_pkt_copy(m->__m_extbuf, p, m->__m_extlen);
479 		m_new->m_data = p + ofs;
480 		m = m_new;
481 	}
482 #endif /* USERSPACE */
483 	if (q->head == NULL)
484 		q->head = m;
485 	else
486 		q->tail->m_nextpkt = m;
487 	q->count++;
488 	q->tail = m;
489 	m->m_nextpkt = NULL;
490 }
491 #endif /* NEW_AQM */
492 
493 #endif /* _IP_DN_PRIVATE_H */
494