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