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 52MALLOC_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 80SLIST_HEAD(dn_schk_head, dn_schk); 81SLIST_HEAD(dn_sch_inst_head, dn_sch_inst); 82SLIST_HEAD(dn_fsk_head, dn_fsk); 83SLIST_HEAD(dn_queue_head, dn_queue); 84SLIST_HEAD(dn_alg_head, dn_alg); 85 86#ifdef NEW_AQM 87SLIST_HEAD(dn_aqm_head, dn_aqm); /* for new AQMs */ 88#endif 89 90struct mq { /* a basic queue of packets*/ 91 struct mbuf *head, *tail; 92 int count; 93}; 94 95static inline void 96set_oid(struct dn_id *o, int type, int len) 97{ 98 o->type = type; 99 o->len = len; 100 o->subtype = 0; 101} 102 103/* 104 * configuration and data for a dummynet instance 105 * 106 * When a configuration is modified from userland, 'id' is incremented 107 * so we can use the value to check for stale pointers. 108 */ 109struct dn_parms { 110 uint32_t id; /* configuration version */ 111 112 /* defaults (sysctl-accessible) */ 113 int red_lookup_depth; 114 int red_avg_pkt_size; 115 int red_max_pkt_size; 116 int hash_size; 117 int max_hash_size; 118 long byte_limit; /* max queue sizes */ 119 long slot_limit; 120 121 int io_fast; 122 int debug; 123 124 /* timekeeping */ 125 struct timeval prev_t; /* last time dummynet_tick ran */ 126 struct dn_heap evheap; /* scheduled events */ 127 128 long tick_last; /* Last tick duration (usec). */ 129 long tick_delta; /* Last vs standard tick diff (usec). */ 130 long tick_delta_sum; /* Accumulated tick difference (usec).*/ 131 long tick_adjustment; /* Tick adjustments done. */ 132 long tick_lost; /* Lost(coalesced) ticks number. */ 133 /* Adjusted vs non-adjusted curr_time difference (ticks). */ 134 long tick_diff; 135 136 /* counters of objects -- used for reporting space */ 137 int schk_count; 138 int si_count; 139 int fsk_count; 140 int queue_count; 141 142 /* packet counters */ 143 unsigned long io_pkt; 144 unsigned long io_pkt_fast; 145 unsigned long io_pkt_drop; 146 147 /* ticks and other stuff */ 148 uint64_t curr_time; 149 /* flowsets and schedulers are in hash tables, with 'hash_size' 150 * buckets. fshash is looked up at every packet arrival 151 * so better be generous if we expect many entries. 152 */ 153 struct dn_ht *fshash; 154 struct dn_ht *schedhash; 155 /* list of flowsets without a scheduler -- use sch_chain */ 156 struct dn_fsk_head fsu; /* list of unlinked flowsets */ 157 158 /* Store the fs/sch to scan when draining. The value is the 159 * bucket number of the hash table. Expire can be disabled 160 * with net.inet.ip.dummynet.expire=0, or it happens every 161 * expire ticks. 162 **/ 163 int drain_fs; 164 int drain_sch; 165 uint32_t expire; 166 uint32_t expire_cycle; /* tick count */ 167 168 int init_done; 169 170#ifdef _KERNEL 171 /* 172 * This file is normally used in the kernel, unless we do 173 * some userland tests, in which case we do not need a mtx. 174 * uh_mtx arbitrates between system calls and also 175 * protects fshash, schedhash and fsunlinked. 176 * These structures are readonly for the lower half. 177 * bh_mtx protects all other structures which may be 178 * modified upon packet arrivals 179 */ 180#if defined( __linux__ ) || defined( _WIN32 ) 181 spinlock_t uh_mtx; 182 spinlock_t bh_mtx; 183#else 184 struct mtx uh_mtx; 185 struct mtx bh_mtx; 186#endif 187 188#endif /* _KERNEL */ 189}; 190 191/* 192 * Delay line, contains all packets on output from a link. 193 * Every scheduler instance has one. 194 */ 195struct delay_line { 196 struct dn_id oid; 197 struct dn_sch_inst *si; 198 struct mq mq; 199}; 200 201/* 202 * The kernel side of a flowset. It is linked in a hash table 203 * of flowsets, and in a list of children of their parent scheduler. 204 * qht is either the queue or (if HAVE_MASK) a hash table queues. 205 * Note that the mask to use is the (flow_mask|sched_mask), which 206 * changes as we attach/detach schedulers. So we store it here. 207 * 208 * XXX If we want to add scheduler-specific parameters, we need to 209 * put them in external storage because the scheduler may not be 210 * available when the fsk is created. 211 */ 212struct dn_fsk { /* kernel side of a flowset */ 213 struct dn_fs fs; 214 SLIST_ENTRY(dn_fsk) fsk_next; /* hash chain for fshash */ 215 216 struct ipfw_flow_id fsk_mask; 217 218 /* qht is a hash table of queues, or just a single queue 219 * a bit in fs.flags tells us which one 220 */ 221 struct dn_ht *qht; 222 struct dn_schk *sched; /* Sched we are linked to */ 223 SLIST_ENTRY(dn_fsk) sch_chain; /* list of fsk attached to sched */ 224 225 /* bucket index used by drain routine to drain queues for this 226 * flowset 227 */ 228 int drain_bucket; 229 /* Parameter realted to RED / GRED */ 230 /* original values are in dn_fs*/ 231 int w_q ; /* queue weight (scaled) */ 232 int max_th ; /* maximum threshold for queue (scaled) */ 233 int min_th ; /* minimum threshold for queue (scaled) */ 234 int max_p ; /* maximum value for p_b (scaled) */ 235 236 u_int c_1 ; /* max_p/(max_th-min_th) (scaled) */ 237 u_int c_2 ; /* max_p*min_th/(max_th-min_th) (scaled) */ 238 u_int c_3 ; /* for GRED, (1-max_p)/max_th (scaled) */ 239 u_int c_4 ; /* for GRED, 1 - 2*max_p (scaled) */ 240 u_int * w_q_lookup ; /* lookup table for computing (1-w_q)^t */ 241 u_int lookup_depth ; /* depth of lookup table */ 242 int lookup_step ; /* granularity inside the lookup table */ 243 int lookup_weight ; /* equal to (1-w_q)^t / (1-w_q)^(t+1) */ 244 int avg_pkt_size ; /* medium packet size */ 245 int max_pkt_size ; /* max packet size */ 246#ifdef NEW_AQM 247 struct dn_aqm *aqmfp; /* Pointer to AQM functions */ 248 void *aqmcfg; /* configuration parameters for AQM */ 249#endif 250}; 251 252/* 253 * A queue is created as a child of a flowset unless it belongs to 254 * a !MULTIQUEUE scheduler. It is normally in a hash table in the 255 * flowset. fs always points to the parent flowset. 256 * si normally points to the sch_inst, unless the flowset has been 257 * detached from the scheduler -- in this case si == NULL and we 258 * should not enqueue. 259 */ 260struct dn_queue { 261 struct dn_flow ni; /* oid, flow_id, stats */ 262 struct mq mq; /* packets queue */ 263 struct dn_sch_inst *_si; /* owner scheduler instance */ 264 SLIST_ENTRY(dn_queue) q_next; /* hash chain list for qht */ 265 struct dn_fsk *fs; /* parent flowset. */ 266 267 /* RED parameters */ 268 int avg; /* average queue length est. (scaled) */ 269 int count; /* arrivals since last RED drop */ 270 int random; /* random value (scaled) */ 271 uint64_t q_time; /* start of queue idle time */ 272#ifdef NEW_AQM 273 void *aqm_status; /* per-queue status variables*/ 274#endif 275 276}; 277 278/* 279 * The kernel side of a scheduler. Contains the userland config, 280 * a link, pointer to extra config arguments from command line, 281 * kernel flags, and a pointer to the scheduler methods. 282 * It is stored in a hash table, and holds a list of all 283 * flowsets and scheduler instances. 284 * XXX sch must be at the beginning, see schk_hash(). 285 */ 286struct dn_schk { 287 struct dn_sch sch; 288 struct dn_alg *fp; /* Pointer to scheduler functions */ 289 struct dn_link link; /* The link, embedded */ 290 struct dn_profile *profile; /* delay profile, if any */ 291 struct dn_id *cfg; /* extra config arguments */ 292 293 SLIST_ENTRY(dn_schk) schk_next; /* hash chain for schedhash */ 294 295 struct dn_fsk_head fsk_list; /* all fsk linked to me */ 296 struct dn_fsk *fs; /* Flowset for !MULTIQUEUE */ 297 298 /* bucket index used by the drain routine to drain the scheduler 299 * instance for this flowset. 300 */ 301 int drain_bucket; 302 303 /* Hash table of all instances (through sch.sched_mask) 304 * or single instance if no mask. Always valid. 305 */ 306 struct dn_ht *siht; 307}; 308 309/* 310 * Scheduler instance. 311 * Contains variables and all queues relative to a this instance. 312 * This struct is created a runtime. 313 */ 314struct dn_sch_inst { 315 struct dn_flow ni; /* oid, flowid and stats */ 316 SLIST_ENTRY(dn_sch_inst) si_next; /* hash chain for siht */ 317 struct delay_line dline; 318 struct dn_schk *sched; /* the template */ 319 int kflags; /* DN_ACTIVE */ 320 321 int64_t credit; /* bits I can transmit (more or less). */ 322 uint64_t sched_time; /* time link was scheduled in ready_heap */ 323 uint64_t idle_time; /* start of scheduler instance idle time */ 324 325 /* q_count is the number of queues that this instance is using. 326 * The counter is incremented or decremented when 327 * a reference from the queue is created or deleted. 328 * It is used to make sure that a scheduler instance can be safely 329 * deleted by the drain routine. See notes below. 330 */ 331 int q_count; 332 333}; 334 335/* 336 * NOTE about object drain. 337 * The system will automatically (XXX check when) drain queues and 338 * scheduler instances when they are idle. 339 * A queue is idle when it has no packets; an instance is idle when 340 * it is not in the evheap heap, and the corresponding delay line is empty. 341 * A queue can be safely deleted when it is idle because of the scheduler 342 * function xxx_free_queue() will remove any references to it. 343 * An instance can be only deleted when no queues reference it. To be sure 344 * of that, a counter (q_count) stores the number of queues that are pointing 345 * to the instance. 346 * 347 * XXX 348 * Order of scan: 349 * - take all flowset in a bucket for the flowset hash table 350 * - take all queues in a bucket for the flowset 351 * - increment the queue bucket 352 * - scan next flowset bucket 353 * Nothing is done if a bucket contains no entries. 354 * 355 * The same schema is used for sceduler instances 356 */ 357 358/* kernel-side flags. Linux has DN_DELETE in fcntl.h 359 */ 360enum { 361 /* 1 and 2 are reserved for the SCAN flags */ 362 DN_DESTROY = 0x0004, /* destroy */ 363 DN_DELETE_FS = 0x0008, /* destroy flowset */ 364 DN_DETACH = 0x0010, 365 DN_ACTIVE = 0x0020, /* object is in evheap */ 366 DN_F_DLINE = 0x0040, /* object is a delay line */ 367 DN_DEL_SAFE = 0x0080, /* delete a queue only if no longer needed 368 * by scheduler */ 369 DN_QHT_IS_Q = 0x0100, /* in flowset, qht is a single queue */ 370}; 371 372/* 373 * Packets processed by dummynet have an mbuf tag associated with 374 * them that carries their dummynet state. 375 * Outside dummynet, only the 'rule' field is relevant, and it must 376 * be at the beginning of the structure. 377 */ 378struct dn_pkt_tag { 379 struct ipfw_rule_ref rule; /* matching rule */ 380 381 /* second part, dummynet specific */ 382 int dn_dir; /* action when packet comes out.*/ 383 /* see ip_fw_private.h */ 384 uint64_t output_time; /* when the pkt is due for delivery*/ 385 struct ifnet *ifp; /* interface, for ip_output */ 386 struct _ip6dn_args ip6opt; /* XXX ipv6 options */ 387 uint16_t iphdr_off; /* IP header offset for mtodo() */ 388}; 389 390/* 391 * Possible values for dn_dir. XXXGL: this needs to be reviewed 392 * and converted to same values ip_fw_args.flags use. 393 */ 394enum { 395 DIR_OUT = 0, 396 DIR_IN = 1, 397 DIR_FWD = 2, 398 DIR_DROP = 3, 399 PROTO_LAYER2 = 0x4, /* set for layer 2 */ 400 PROTO_IPV4 = 0x08, 401 PROTO_IPV6 = 0x10, 402 PROTO_IFB = 0x0c, /* layer2 + ifbridge */ 403}; 404 405//extern struct dn_parms V_dn_cfg; 406VNET_DECLARE(struct dn_parms, dn_cfg); 407#define V_dn_cfg VNET(dn_cfg) 408 409int dummynet_io(struct mbuf **, struct ip_fw_args *); 410void dummynet_task(void *context, int pending); 411void dn_reschedule(void); 412struct dn_pkt_tag * dn_tag_get(struct mbuf *m); 413 414struct dn_queue *ipdn_q_find(struct dn_fsk *, struct dn_sch_inst *, 415 struct ipfw_flow_id *); 416struct dn_sch_inst *ipdn_si_find(struct dn_schk *, struct ipfw_flow_id *); 417 418/* 419 * copy_range is a template for requests for ranges of pipes/queues/scheds. 420 * The number of ranges is variable and can be derived by o.len. 421 * As a default, we use a small number of entries so that the struct 422 * fits easily on the stack and is sufficient for most common requests. 423 */ 424#define DEFAULT_RANGES 5 425struct copy_range { 426 struct dn_id o; 427 uint32_t r[ 2 * DEFAULT_RANGES ]; 428}; 429 430struct copy_args { 431 char **start; 432 char *end; 433 int flags; 434 int type; 435 struct copy_range *extra; /* extra filtering */ 436}; 437 438struct sockopt; 439int ip_dummynet_compat(struct sockopt *sopt); 440int dummynet_get(struct sockopt *sopt, void **compat); 441int dn_c_copy_q (void *_ni, void *arg); 442int dn_c_copy_pipe(struct dn_schk *s, struct copy_args *a, int nq); 443int dn_c_copy_fs(struct dn_fsk *f, struct copy_args *a, int nq); 444int dn_compat_copy_queue(struct copy_args *a, void *_o); 445int dn_compat_copy_pipe(struct copy_args *a, void *_o); 446int copy_data_helper_compat(void *_o, void *_arg); 447int dn_compat_calc_size(void); 448int do_config(void *p, int l); 449 450/* function to drain idle object */ 451void dn_drain_scheduler(void); 452void dn_drain_queue(void); 453 454#ifdef NEW_AQM 455int ecn_mark(struct mbuf* m); 456 457/* moved from ip_dn_io.c to here to be available for AQMs modules*/ 458static inline void 459mq_append(struct mq *q, struct mbuf *m) 460{ 461#ifdef USERSPACE 462 // buffers from netmap need to be copied 463 // XXX note that the routine is not expected to fail 464 ND("append %p to %p", m, q); 465 if (m->m_flags & M_STACK) { 466 struct mbuf *m_new; 467 void *p; 468 int l, ofs; 469 470 ofs = m->m_data - m->__m_extbuf; 471 // XXX allocate 472 MGETHDR(m_new, M_NOWAIT, MT_DATA); 473 ND("*** WARNING, volatile buf %p ext %p %d dofs %d m_new %p", 474 m, m->__m_extbuf, m->__m_extlen, ofs, m_new); 475 p = m_new->__m_extbuf; /* new pointer */ 476 l = m_new->__m_extlen; /* new len */ 477 if (l <= m->__m_extlen) { 478 panic("extlen too large"); 479 } 480 481 *m_new = *m; // copy 482 m_new->m_flags &= ~M_STACK; 483 m_new->__m_extbuf = p; // point to new buffer 484 _pkt_copy(m->__m_extbuf, p, m->__m_extlen); 485 m_new->m_data = p + ofs; 486 m = m_new; 487 } 488#endif /* USERSPACE */ 489 if (q->head == NULL) 490 q->head = m; 491 else 492 q->tail->m_nextpkt = m; 493 q->count++; 494 q->tail = m; 495 m->m_nextpkt = NULL; 496} 497#endif /* NEW_AQM */ 498 499#endif /* _IP_DN_PRIVATE_H */ 500