1 /*- 2 * Copyright (c) 2004-2008 University of Zagreb 3 * Copyright (c) 2007-2008 FreeBSD Foundation 4 * 5 * This software was developed by the University of Zagreb and the 6 * FreeBSD Foundation under sponsorship by the Stichting NLnet and the 7 * FreeBSD Foundation. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * $FreeBSD$ 31 */ 32 33 /* 34 * This node permits simple traffic shaping by emulating bandwidth 35 * and delay, as well as random packet losses. 36 * The node has two hooks, upper and lower. Traffic flowing from upper to 37 * lower hook is referenced as downstream, and vice versa. Parameters for 38 * both directions can be set separately, except for delay. 39 */ 40 41 42 #include <sys/param.h> 43 #include <sys/errno.h> 44 #include <sys/systm.h> 45 #include <sys/kernel.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/time.h> 49 50 #include <vm/uma.h> 51 52 #include <net/vnet.h> 53 54 #include <netinet/in.h> 55 #include <netinet/in_systm.h> 56 #include <netinet/ip.h> 57 58 #include <netgraph/ng_message.h> 59 #include <netgraph/netgraph.h> 60 #include <netgraph/ng_parse.h> 61 #include <netgraph/ng_pipe.h> 62 63 static MALLOC_DEFINE(M_NG_PIPE, "ng_pipe", "ng_pipe"); 64 65 struct mtx ng_pipe_giant; 66 67 /* Packet header struct */ 68 struct ngp_hdr { 69 TAILQ_ENTRY(ngp_hdr) ngp_link; /* next pkt in queue */ 70 struct timeval when; /* this packet's due time */ 71 struct mbuf *m; /* ptr to the packet data */ 72 }; 73 TAILQ_HEAD(p_head, ngp_hdr); 74 75 /* FIFO queue struct */ 76 struct ngp_fifo { 77 TAILQ_ENTRY(ngp_fifo) fifo_le; /* list of active queues only */ 78 struct p_head packet_head; /* FIFO queue head */ 79 u_int32_t hash; /* flow signature */ 80 struct timeval vtime; /* virtual time, for WFQ */ 81 u_int32_t rr_deficit; /* for DRR */ 82 u_int32_t packets; /* # of packets in this queue */ 83 }; 84 85 /* Per hook info */ 86 struct hookinfo { 87 hook_p hook; 88 int noqueue; /* bypass any processing */ 89 TAILQ_HEAD(, ngp_fifo) fifo_head; /* FIFO queues */ 90 TAILQ_HEAD(, ngp_hdr) qout_head; /* delay queue head */ 91 LIST_ENTRY(hookinfo) active_le; /* active hooks */ 92 struct timeval qin_utime; 93 struct ng_pipe_hookcfg cfg; 94 struct ng_pipe_hookrun run; 95 struct ng_pipe_hookstat stats; 96 uint64_t *ber_p; /* loss_p(BER,psize) map */ 97 }; 98 99 /* Per node info */ 100 struct node_priv { 101 u_int64_t delay; 102 u_int32_t overhead; 103 u_int32_t header_offset; 104 struct hookinfo lower; 105 struct hookinfo upper; 106 }; 107 typedef struct node_priv *priv_p; 108 109 /* Macro for calculating the virtual time for packet dequeueing in WFQ */ 110 #define FIFO_VTIME_SORT(plen) \ 111 if (hinfo->cfg.wfq && hinfo->cfg.bandwidth) { \ 112 ngp_f->vtime.tv_usec = now->tv_usec + ((uint64_t) (plen) \ 113 + priv->overhead ) * hinfo->run.fifo_queues * \ 114 8000000 / hinfo->cfg.bandwidth; \ 115 ngp_f->vtime.tv_sec = now->tv_sec + \ 116 ngp_f->vtime.tv_usec / 1000000; \ 117 ngp_f->vtime.tv_usec = ngp_f->vtime.tv_usec % 1000000; \ 118 TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le) \ 119 if (ngp_f1->vtime.tv_sec > ngp_f->vtime.tv_sec || \ 120 (ngp_f1->vtime.tv_sec == ngp_f->vtime.tv_sec && \ 121 ngp_f1->vtime.tv_usec > ngp_f->vtime.tv_usec)) \ 122 break; \ 123 if (ngp_f1 == NULL) \ 124 TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le); \ 125 else \ 126 TAILQ_INSERT_BEFORE(ngp_f1, ngp_f, fifo_le); \ 127 } else \ 128 TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le); \ 129 130 131 static void parse_cfg(struct ng_pipe_hookcfg *, struct ng_pipe_hookcfg *, 132 struct hookinfo *, priv_p); 133 static void pipe_dequeue(struct hookinfo *, struct timeval *); 134 static void pipe_scheduler(void *); 135 static void pipe_poll(void); 136 static int ngp_modevent(module_t, int, void *); 137 138 /* linked list of active "pipe" hooks */ 139 static LIST_HEAD(, hookinfo) active_head; 140 static int active_gen_id = 0; 141 142 /* timeout handle for pipe_scheduler */ 143 static struct callout polling_timer; 144 145 /* zone for storing ngp_hdr-s */ 146 static uma_zone_t ngp_zone; 147 148 /* Netgraph methods */ 149 static ng_constructor_t ngp_constructor; 150 static ng_rcvmsg_t ngp_rcvmsg; 151 static ng_shutdown_t ngp_shutdown; 152 static ng_newhook_t ngp_newhook; 153 static ng_rcvdata_t ngp_rcvdata; 154 static ng_disconnect_t ngp_disconnect; 155 156 /* Parse type for struct ng_pipe_hookstat */ 157 static const struct ng_parse_struct_field 158 ng_pipe_hookstat_type_fields[] = NG_PIPE_HOOKSTAT_INFO; 159 static const struct ng_parse_type ng_pipe_hookstat_type = { 160 &ng_parse_struct_type, 161 &ng_pipe_hookstat_type_fields 162 }; 163 164 /* Parse type for struct ng_pipe_stats */ 165 static const struct ng_parse_struct_field ng_pipe_stats_type_fields[] = 166 NG_PIPE_STATS_INFO(&ng_pipe_hookstat_type); 167 static const struct ng_parse_type ng_pipe_stats_type = { 168 &ng_parse_struct_type, 169 &ng_pipe_stats_type_fields 170 }; 171 172 /* Parse type for struct ng_pipe_hookrun */ 173 static const struct ng_parse_struct_field 174 ng_pipe_hookrun_type_fields[] = NG_PIPE_HOOKRUN_INFO; 175 static const struct ng_parse_type ng_pipe_hookrun_type = { 176 &ng_parse_struct_type, 177 &ng_pipe_hookrun_type_fields 178 }; 179 180 /* Parse type for struct ng_pipe_run */ 181 static const struct ng_parse_struct_field 182 ng_pipe_run_type_fields[] = NG_PIPE_RUN_INFO(&ng_pipe_hookrun_type); 183 static const struct ng_parse_type ng_pipe_run_type = { 184 &ng_parse_struct_type, 185 &ng_pipe_run_type_fields 186 }; 187 188 /* Parse type for struct ng_pipe_hookcfg */ 189 static const struct ng_parse_struct_field 190 ng_pipe_hookcfg_type_fields[] = NG_PIPE_HOOKCFG_INFO; 191 static const struct ng_parse_type ng_pipe_hookcfg_type = { 192 &ng_parse_struct_type, 193 &ng_pipe_hookcfg_type_fields 194 }; 195 196 /* Parse type for struct ng_pipe_cfg */ 197 static const struct ng_parse_struct_field 198 ng_pipe_cfg_type_fields[] = NG_PIPE_CFG_INFO(&ng_pipe_hookcfg_type); 199 static const struct ng_parse_type ng_pipe_cfg_type = { 200 &ng_parse_struct_type, 201 &ng_pipe_cfg_type_fields 202 }; 203 204 /* List of commands and how to convert arguments to/from ASCII */ 205 static const struct ng_cmdlist ngp_cmds[] = { 206 { 207 .cookie = NGM_PIPE_COOKIE, 208 .cmd = NGM_PIPE_GET_STATS, 209 .name = "getstats", 210 .respType = &ng_pipe_stats_type 211 }, 212 { 213 .cookie = NGM_PIPE_COOKIE, 214 .cmd = NGM_PIPE_CLR_STATS, 215 .name = "clrstats" 216 }, 217 { 218 .cookie = NGM_PIPE_COOKIE, 219 .cmd = NGM_PIPE_GETCLR_STATS, 220 .name = "getclrstats", 221 .respType = &ng_pipe_stats_type 222 }, 223 { 224 .cookie = NGM_PIPE_COOKIE, 225 .cmd = NGM_PIPE_GET_RUN, 226 .name = "getrun", 227 .respType = &ng_pipe_run_type 228 }, 229 { 230 .cookie = NGM_PIPE_COOKIE, 231 .cmd = NGM_PIPE_GET_CFG, 232 .name = "getcfg", 233 .respType = &ng_pipe_cfg_type 234 }, 235 { 236 .cookie = NGM_PIPE_COOKIE, 237 .cmd = NGM_PIPE_SET_CFG, 238 .name = "setcfg", 239 .mesgType = &ng_pipe_cfg_type, 240 }, 241 { 0 } 242 }; 243 244 /* Netgraph type descriptor */ 245 static struct ng_type ng_pipe_typestruct = { 246 .version = NG_ABI_VERSION, 247 .name = NG_PIPE_NODE_TYPE, 248 .mod_event = ngp_modevent, 249 .constructor = ngp_constructor, 250 .shutdown = ngp_shutdown, 251 .rcvmsg = ngp_rcvmsg, 252 .newhook = ngp_newhook, 253 .rcvdata = ngp_rcvdata, 254 .disconnect = ngp_disconnect, 255 .cmdlist = ngp_cmds 256 }; 257 NETGRAPH_INIT(pipe, &ng_pipe_typestruct); 258 259 /* Node constructor */ 260 static int 261 ngp_constructor(node_p node) 262 { 263 priv_p priv; 264 265 priv = malloc(sizeof(*priv), M_NG_PIPE, M_ZERO | M_NOWAIT); 266 if (priv == NULL) 267 return (ENOMEM); 268 NG_NODE_SET_PRIVATE(node, priv); 269 270 return (0); 271 } 272 273 /* Add a hook */ 274 static int 275 ngp_newhook(node_p node, hook_p hook, const char *name) 276 { 277 const priv_p priv = NG_NODE_PRIVATE(node); 278 struct hookinfo *hinfo; 279 280 if (strcmp(name, NG_PIPE_HOOK_UPPER) == 0) { 281 bzero(&priv->upper, sizeof(priv->upper)); 282 priv->upper.hook = hook; 283 NG_HOOK_SET_PRIVATE(hook, &priv->upper); 284 } else if (strcmp(name, NG_PIPE_HOOK_LOWER) == 0) { 285 bzero(&priv->lower, sizeof(priv->lower)); 286 priv->lower.hook = hook; 287 NG_HOOK_SET_PRIVATE(hook, &priv->lower); 288 } else 289 return (EINVAL); 290 291 /* Load non-zero initial cfg values */ 292 hinfo = NG_HOOK_PRIVATE(hook); 293 hinfo->cfg.qin_size_limit = 50; 294 hinfo->cfg.fifo = 1; 295 hinfo->cfg.droptail = 1; 296 TAILQ_INIT(&hinfo->fifo_head); 297 TAILQ_INIT(&hinfo->qout_head); 298 return (0); 299 } 300 301 /* Receive a control message */ 302 static int 303 ngp_rcvmsg(node_p node, item_p item, hook_p lasthook) 304 { 305 const priv_p priv = NG_NODE_PRIVATE(node); 306 struct ng_mesg *resp = NULL; 307 struct ng_mesg *msg; 308 struct ng_pipe_stats *stats; 309 struct ng_pipe_run *run; 310 struct ng_pipe_cfg *cfg; 311 int error = 0; 312 313 mtx_lock(&ng_pipe_giant); 314 315 NGI_GET_MSG(item, msg); 316 switch (msg->header.typecookie) { 317 case NGM_PIPE_COOKIE: 318 switch (msg->header.cmd) { 319 case NGM_PIPE_GET_STATS: 320 case NGM_PIPE_CLR_STATS: 321 case NGM_PIPE_GETCLR_STATS: 322 if (msg->header.cmd != NGM_PIPE_CLR_STATS) { 323 NG_MKRESPONSE(resp, msg, 324 sizeof(*stats), M_NOWAIT); 325 if (resp == NULL) { 326 error = ENOMEM; 327 break; 328 } 329 stats = (struct ng_pipe_stats *)resp->data; 330 bcopy(&priv->upper.stats, &stats->downstream, 331 sizeof(stats->downstream)); 332 bcopy(&priv->lower.stats, &stats->upstream, 333 sizeof(stats->upstream)); 334 } 335 if (msg->header.cmd != NGM_PIPE_GET_STATS) { 336 bzero(&priv->upper.stats, 337 sizeof(priv->upper.stats)); 338 bzero(&priv->lower.stats, 339 sizeof(priv->lower.stats)); 340 } 341 break; 342 case NGM_PIPE_GET_RUN: 343 NG_MKRESPONSE(resp, msg, sizeof(*run), M_NOWAIT); 344 if (resp == NULL) { 345 error = ENOMEM; 346 break; 347 } 348 run = (struct ng_pipe_run *)resp->data; 349 bcopy(&priv->upper.run, &run->downstream, 350 sizeof(run->downstream)); 351 bcopy(&priv->lower.run, &run->upstream, 352 sizeof(run->upstream)); 353 break; 354 case NGM_PIPE_GET_CFG: 355 NG_MKRESPONSE(resp, msg, sizeof(*cfg), M_NOWAIT); 356 if (resp == NULL) { 357 error = ENOMEM; 358 break; 359 } 360 cfg = (struct ng_pipe_cfg *)resp->data; 361 bcopy(&priv->upper.cfg, &cfg->downstream, 362 sizeof(cfg->downstream)); 363 bcopy(&priv->lower.cfg, &cfg->upstream, 364 sizeof(cfg->upstream)); 365 cfg->delay = priv->delay; 366 cfg->overhead = priv->overhead; 367 cfg->header_offset = priv->header_offset; 368 if (cfg->upstream.bandwidth == 369 cfg->downstream.bandwidth) { 370 cfg->bandwidth = cfg->upstream.bandwidth; 371 cfg->upstream.bandwidth = 0; 372 cfg->downstream.bandwidth = 0; 373 } else 374 cfg->bandwidth = 0; 375 break; 376 case NGM_PIPE_SET_CFG: 377 cfg = (struct ng_pipe_cfg *)msg->data; 378 if (msg->header.arglen != sizeof(*cfg)) { 379 error = EINVAL; 380 break; 381 } 382 383 if (cfg->delay == -1) 384 priv->delay = 0; 385 else if (cfg->delay > 0 && cfg->delay < 10000000) 386 priv->delay = cfg->delay; 387 388 if (cfg->bandwidth == -1) { 389 priv->upper.cfg.bandwidth = 0; 390 priv->lower.cfg.bandwidth = 0; 391 priv->overhead = 0; 392 } else if (cfg->bandwidth >= 100 && 393 cfg->bandwidth <= 1000000000) { 394 priv->upper.cfg.bandwidth = cfg->bandwidth; 395 priv->lower.cfg.bandwidth = cfg->bandwidth; 396 if (cfg->bandwidth >= 10000000) 397 priv->overhead = 8+4+12; /* Ethernet */ 398 else 399 priv->overhead = 10; /* HDLC */ 400 } 401 402 if (cfg->overhead == -1) 403 priv->overhead = 0; 404 else if (cfg->overhead > 0 && cfg->overhead < 256) 405 priv->overhead = cfg->overhead; 406 407 if (cfg->header_offset == -1) 408 priv->header_offset = 0; 409 else if (cfg->header_offset > 0 && 410 cfg->header_offset < 64) 411 priv->header_offset = cfg->header_offset; 412 413 parse_cfg(&priv->upper.cfg, &cfg->downstream, 414 &priv->upper, priv); 415 parse_cfg(&priv->lower.cfg, &cfg->upstream, 416 &priv->lower, priv); 417 break; 418 default: 419 error = EINVAL; 420 break; 421 } 422 break; 423 default: 424 error = EINVAL; 425 break; 426 } 427 NG_RESPOND_MSG(error, node, item, resp); 428 NG_FREE_MSG(msg); 429 430 mtx_unlock(&ng_pipe_giant); 431 432 return (error); 433 } 434 435 static void 436 parse_cfg(struct ng_pipe_hookcfg *current, struct ng_pipe_hookcfg *new, 437 struct hookinfo *hinfo, priv_p priv) 438 { 439 440 if (new->ber == -1) { 441 current->ber = 0; 442 if (hinfo->ber_p) { 443 free(hinfo->ber_p, M_NG_PIPE); 444 hinfo->ber_p = NULL; 445 } 446 } else if (new->ber >= 1 && new->ber <= 1000000000000) { 447 static const uint64_t one = 0x1000000000000; /* = 2^48 */ 448 uint64_t p0, p; 449 uint32_t fsize, i; 450 451 if (hinfo->ber_p == NULL) 452 hinfo->ber_p = malloc(\ 453 (MAX_FSIZE + MAX_OHSIZE)*sizeof(uint64_t), \ 454 M_NG_PIPE, M_NOWAIT); 455 current->ber = new->ber; 456 457 /* 458 * For given BER and each frame size N (in bytes) calculate 459 * the probability P_OK that the frame is clean: 460 * 461 * P_OK(BER,N) = (1 - 1/BER)^(N*8) 462 * 463 * We use a 64-bit fixed-point format with decimal point 464 * positioned between bits 47 and 48. 465 */ 466 p0 = one - one / new->ber; 467 p = one; 468 for (fsize = 0; fsize < MAX_FSIZE + MAX_OHSIZE; fsize++) { 469 hinfo->ber_p[fsize] = p; 470 for (i=0; i<8; i++) 471 p = (p*(p0&0xffff)>>48) + \ 472 (p*((p0>>16)&0xffff)>>32) + \ 473 (p*(p0>>32)>>16); 474 } 475 } 476 477 if (new->qin_size_limit == -1) 478 current->qin_size_limit = 0; 479 else if (new->qin_size_limit >= 5) 480 current->qin_size_limit = new->qin_size_limit; 481 482 if (new->qout_size_limit == -1) 483 current->qout_size_limit = 0; 484 else if (new->qout_size_limit >= 5) 485 current->qout_size_limit = new->qout_size_limit; 486 487 if (new->duplicate == -1) 488 current->duplicate = 0; 489 else if (new->duplicate > 0 && new->duplicate <= 50) 490 current->duplicate = new->duplicate; 491 492 if (new->fifo) { 493 current->fifo = 1; 494 current->wfq = 0; 495 current->drr = 0; 496 } 497 498 if (new->wfq) { 499 current->fifo = 0; 500 current->wfq = 1; 501 current->drr = 0; 502 } 503 504 if (new->drr) { 505 current->fifo = 0; 506 current->wfq = 0; 507 /* DRR quantum */ 508 if (new->drr >= 32) 509 current->drr = new->drr; 510 else 511 current->drr = 2048; /* default quantum */ 512 } 513 514 if (new->droptail) { 515 current->droptail = 1; 516 current->drophead = 0; 517 } 518 519 if (new->drophead) { 520 current->droptail = 0; 521 current->drophead = 1; 522 } 523 524 if (new->bandwidth == -1) { 525 current->bandwidth = 0; 526 current->fifo = 1; 527 current->wfq = 0; 528 current->drr = 0; 529 } else if (new->bandwidth >= 100 && new->bandwidth <= 1000000000) 530 current->bandwidth = new->bandwidth; 531 532 if (current->bandwidth | priv->delay | 533 current->duplicate | current->ber) 534 hinfo->noqueue = 0; 535 else 536 hinfo->noqueue = 1; 537 } 538 539 /* 540 * Compute a hash signature for a packet. This function suffers from the 541 * NIH sindrome, so probably it would be wise to look around what other 542 * folks have found out to be a good and efficient IP hash function... 543 */ 544 static int 545 ip_hash(struct mbuf *m, int offset) 546 { 547 u_int64_t i; 548 struct ip *ip = (struct ip *)(mtod(m, u_char *) + offset); 549 550 if (m->m_len < sizeof(struct ip) + offset || 551 ip->ip_v != 4 || ip->ip_hl << 2 != sizeof(struct ip)) 552 return 0; 553 554 i = ((u_int64_t) ip->ip_src.s_addr ^ 555 ((u_int64_t) ip->ip_src.s_addr << 13) ^ 556 ((u_int64_t) ip->ip_dst.s_addr << 7) ^ 557 ((u_int64_t) ip->ip_dst.s_addr << 19)); 558 return (i ^ (i >> 32)); 559 } 560 561 /* 562 * Receive data on a hook - both in upstream and downstream direction. 563 * We put the frame on the inbound queue, and try to initiate dequeuing 564 * sequence immediately. If inbound queue is full, discard one frame 565 * depending on dropping policy (from the head or from the tail of the 566 * queue). 567 */ 568 static int 569 ngp_rcvdata(hook_p hook, item_p item) 570 { 571 struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook); 572 const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook)); 573 struct timeval uuptime; 574 struct timeval *now = &uuptime; 575 struct ngp_fifo *ngp_f = NULL, *ngp_f1; 576 struct ngp_hdr *ngp_h = NULL; 577 struct mbuf *m; 578 int hash; 579 int error = 0; 580 581 if (hinfo->noqueue) { 582 struct hookinfo *dest; 583 if (hinfo == &priv->lower) 584 dest = &priv->upper; 585 else 586 dest = &priv->lower; 587 NG_FWD_ITEM_HOOK(error, item, dest->hook); 588 return error; 589 } 590 591 mtx_lock(&ng_pipe_giant); 592 microuptime(now); 593 594 /* 595 * Attach us to the list of active ng_pipes if this was an empty 596 * one before, and also update the queue service deadline time. 597 */ 598 if (hinfo->run.qin_frames == 0) { 599 struct timeval *when = &hinfo->qin_utime; 600 if (when->tv_sec < now->tv_sec || (when->tv_sec == now->tv_sec 601 && when->tv_usec < now->tv_usec)) { 602 when->tv_sec = now->tv_sec; 603 when->tv_usec = now->tv_usec; 604 } 605 if (hinfo->run.qout_frames == 0) 606 LIST_INSERT_HEAD(&active_head, hinfo, active_le); 607 } 608 609 /* Populate the packet header */ 610 ngp_h = uma_zalloc(ngp_zone, M_NOWAIT); 611 KASSERT((ngp_h != NULL), ("ngp_h zalloc failed (1)")); 612 NGI_GET_M(item, m); 613 KASSERT(m != NULL, ("NGI_GET_M failed")); 614 ngp_h->m = m; 615 NG_FREE_ITEM(item); 616 617 if (hinfo->cfg.fifo) 618 hash = 0; /* all packets go into a single FIFO queue */ 619 else 620 hash = ip_hash(m, priv->header_offset); 621 622 /* Find the appropriate FIFO queue for the packet and enqueue it*/ 623 TAILQ_FOREACH(ngp_f, &hinfo->fifo_head, fifo_le) 624 if (hash == ngp_f->hash) 625 break; 626 if (ngp_f == NULL) { 627 ngp_f = uma_zalloc(ngp_zone, M_NOWAIT); 628 KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (2)")); 629 TAILQ_INIT(&ngp_f->packet_head); 630 ngp_f->hash = hash; 631 ngp_f->packets = 1; 632 ngp_f->rr_deficit = hinfo->cfg.drr; /* DRR quantum */ 633 hinfo->run.fifo_queues++; 634 TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link); 635 FIFO_VTIME_SORT(m->m_pkthdr.len); 636 } else { 637 TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link); 638 ngp_f->packets++; 639 } 640 hinfo->run.qin_frames++; 641 hinfo->run.qin_octets += m->m_pkthdr.len; 642 643 /* Discard a frame if inbound queue limit has been reached */ 644 if (hinfo->run.qin_frames > hinfo->cfg.qin_size_limit) { 645 struct mbuf *m1; 646 int longest = 0; 647 648 /* Find the longest queue */ 649 TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le) 650 if (ngp_f1->packets > longest) { 651 longest = ngp_f1->packets; 652 ngp_f = ngp_f1; 653 } 654 655 /* Drop a frame from the queue head/tail, depending on cfg */ 656 if (hinfo->cfg.drophead) 657 ngp_h = TAILQ_FIRST(&ngp_f->packet_head); 658 else 659 ngp_h = TAILQ_LAST(&ngp_f->packet_head, p_head); 660 TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link); 661 m1 = ngp_h->m; 662 uma_zfree(ngp_zone, ngp_h); 663 hinfo->run.qin_octets -= m1->m_pkthdr.len; 664 hinfo->stats.in_disc_octets += m1->m_pkthdr.len; 665 m_freem(m1); 666 if (--(ngp_f->packets) == 0) { 667 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le); 668 uma_zfree(ngp_zone, ngp_f); 669 hinfo->run.fifo_queues--; 670 } 671 hinfo->run.qin_frames--; 672 hinfo->stats.in_disc_frames++; 673 } else if (hinfo->run.qin_frames > hinfo->cfg.qin_size_limit) { 674 struct mbuf *m1; 675 int longest = 0; 676 677 /* Find the longest queue */ 678 TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le) 679 if (ngp_f1->packets > longest) { 680 longest = ngp_f1->packets; 681 ngp_f = ngp_f1; 682 } 683 684 /* Drop a frame from the queue head/tail, depending on cfg */ 685 if (hinfo->cfg.drophead) 686 ngp_h = TAILQ_FIRST(&ngp_f->packet_head); 687 else 688 ngp_h = TAILQ_LAST(&ngp_f->packet_head, p_head); 689 TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link); 690 m1 = ngp_h->m; 691 uma_zfree(ngp_zone, ngp_h); 692 hinfo->run.qin_octets -= m1->m_pkthdr.len; 693 hinfo->stats.in_disc_octets += m1->m_pkthdr.len; 694 m_freem(m1); 695 if (--(ngp_f->packets) == 0) { 696 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le); 697 uma_zfree(ngp_zone, ngp_f); 698 hinfo->run.fifo_queues--; 699 } 700 hinfo->run.qin_frames--; 701 hinfo->stats.in_disc_frames++; 702 } 703 704 /* 705 * Try to start the dequeuing process immediately. We must 706 * hold the ng_pipe_giant lock here and pipe_dequeue() will 707 * release it 708 */ 709 pipe_dequeue(hinfo, now); 710 711 return (0); 712 } 713 714 715 /* 716 * Dequeueing sequence - we basically do the following: 717 * 1) Try to extract the frame from the inbound (bandwidth) queue; 718 * 2) In accordance to BER specified, discard the frame randomly; 719 * 3) If the frame survives BER, prepend it with delay info and move it 720 * to outbound (delay) queue; 721 * 4) Loop to 2) until bandwidth quota for this timeslice is reached, or 722 * inbound queue is flushed completely; 723 * 5) Extract the first frame from the outbound queue, if it's time has 724 * come. Queue the frame for transmission on the outbound hook; 725 * 6) Loop to 5) until outbound queue is flushed completely, or the next 726 * frame in the queue is not scheduled to be dequeued yet; 727 * 7) Transimit all frames queued in 5) 728 * 729 * Note: the caller must hold the ng_pipe_giant lock; this function 730 * returns with the lock released. 731 */ 732 static void 733 pipe_dequeue(struct hookinfo *hinfo, struct timeval *now) { 734 static uint64_t rand, oldrand; 735 const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hinfo->hook)); 736 struct hookinfo *dest; 737 struct ngp_fifo *ngp_f, *ngp_f1; 738 struct ngp_hdr *ngp_h; 739 struct timeval *when; 740 struct mbuf *q_head = NULL; 741 struct mbuf *q_tail = NULL; 742 struct mbuf *m; 743 int error = 0; 744 745 /* Which one is the destination hook? */ 746 if (hinfo == &priv->lower) 747 dest = &priv->upper; 748 else 749 dest = &priv->lower; 750 751 /* Bandwidth queue processing */ 752 while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) { 753 when = &hinfo->qin_utime; 754 if (when->tv_sec > now->tv_sec || (when->tv_sec == now->tv_sec 755 && when->tv_usec > now->tv_usec)) 756 break; 757 758 ngp_h = TAILQ_FIRST(&ngp_f->packet_head); 759 m = ngp_h->m; 760 761 /* Deficit Round Robin (DRR) processing */ 762 if (hinfo->cfg.drr) { 763 if (ngp_f->rr_deficit >= m->m_pkthdr.len) { 764 ngp_f->rr_deficit -= m->m_pkthdr.len; 765 } else { 766 ngp_f->rr_deficit += hinfo->cfg.drr; 767 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le); 768 TAILQ_INSERT_TAIL(&hinfo->fifo_head, 769 ngp_f, fifo_le); 770 continue; 771 } 772 } 773 774 /* 775 * Either create a duplicate and pass it on, or dequeue 776 * the original packet... 777 */ 778 if (hinfo->cfg.duplicate && 779 random() % 100 <= hinfo->cfg.duplicate) { 780 ngp_h = uma_zalloc(ngp_zone, M_NOWAIT); 781 KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (3)")); 782 ngp_h->m = m_dup(m, M_NOWAIT); 783 KASSERT(ngp_h->m != NULL, ("m_dup failed")); 784 } else { 785 TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link); 786 hinfo->run.qin_frames--; 787 hinfo->run.qin_octets -= m->m_pkthdr.len; 788 ngp_f->packets--; 789 } 790 791 /* Calculate the serialization delay */ 792 if (hinfo->cfg.bandwidth) { 793 hinfo->qin_utime.tv_usec += ((uint64_t) m->m_pkthdr.len 794 + priv->overhead ) * 795 8000000 / hinfo->cfg.bandwidth; 796 hinfo->qin_utime.tv_sec += 797 hinfo->qin_utime.tv_usec / 1000000; 798 hinfo->qin_utime.tv_usec = 799 hinfo->qin_utime.tv_usec % 1000000; 800 } 801 when = &ngp_h->when; 802 when->tv_sec = hinfo->qin_utime.tv_sec; 803 when->tv_usec = hinfo->qin_utime.tv_usec; 804 805 /* Sort / rearrange inbound queues */ 806 if (ngp_f->packets) { 807 if (hinfo->cfg.wfq) { 808 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le); 809 FIFO_VTIME_SORT(TAILQ_FIRST( 810 &ngp_f->packet_head)->m->m_pkthdr.len) 811 } 812 } else { 813 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le); 814 uma_zfree(ngp_zone, ngp_f); 815 hinfo->run.fifo_queues--; 816 } 817 818 /* Randomly discard the frame, according to BER setting */ 819 if (hinfo->cfg.ber) { 820 oldrand = rand; 821 rand = random(); 822 if (((oldrand ^ rand) << 17) >= 823 hinfo->ber_p[priv->overhead + m->m_pkthdr.len]) { 824 hinfo->stats.out_disc_frames++; 825 hinfo->stats.out_disc_octets += m->m_pkthdr.len; 826 uma_zfree(ngp_zone, ngp_h); 827 m_freem(m); 828 continue; 829 } 830 } 831 832 /* Discard frame if outbound queue size limit exceeded */ 833 if (hinfo->cfg.qout_size_limit && 834 hinfo->run.qout_frames>=hinfo->cfg.qout_size_limit) { 835 hinfo->stats.out_disc_frames++; 836 hinfo->stats.out_disc_octets += m->m_pkthdr.len; 837 uma_zfree(ngp_zone, ngp_h); 838 m_freem(m); 839 continue; 840 } 841 842 /* Calculate the propagation delay */ 843 when->tv_usec += priv->delay; 844 when->tv_sec += when->tv_usec / 1000000; 845 when->tv_usec = when->tv_usec % 1000000; 846 847 /* Put the frame into the delay queue */ 848 TAILQ_INSERT_TAIL(&hinfo->qout_head, ngp_h, ngp_link); 849 hinfo->run.qout_frames++; 850 hinfo->run.qout_octets += m->m_pkthdr.len; 851 } 852 853 /* Delay queue processing */ 854 while ((ngp_h = TAILQ_FIRST(&hinfo->qout_head))) { 855 struct mbuf *m = ngp_h->m; 856 857 when = &ngp_h->when; 858 if (when->tv_sec > now->tv_sec || 859 (when->tv_sec == now->tv_sec && 860 when->tv_usec > now->tv_usec)) 861 break; 862 863 /* Update outbound queue stats */ 864 hinfo->stats.fwd_frames++; 865 hinfo->stats.fwd_octets += m->m_pkthdr.len; 866 hinfo->run.qout_frames--; 867 hinfo->run.qout_octets -= m->m_pkthdr.len; 868 869 /* Dequeue the packet from qout */ 870 TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link); 871 uma_zfree(ngp_zone, ngp_h); 872 873 /* Enqueue locally for sending downstream */ 874 if (q_head == NULL) 875 q_head = m; 876 if (q_tail) 877 q_tail->m_nextpkt = m; 878 q_tail = m; 879 m->m_nextpkt = NULL; 880 } 881 882 /* If both queues are empty detach us from the list of active queues */ 883 if (hinfo->run.qin_frames + hinfo->run.qout_frames == 0) { 884 LIST_REMOVE(hinfo, active_le); 885 active_gen_id++; 886 } 887 888 mtx_unlock(&ng_pipe_giant); 889 890 while ((m = q_head) != NULL) { 891 q_head = m->m_nextpkt; 892 m->m_nextpkt = NULL; 893 NG_SEND_DATA(error, dest->hook, m, meta); 894 } 895 } 896 897 898 /* 899 * This routine is called on every clock tick. We poll all nodes/hooks 900 * for queued frames by calling pipe_dequeue(). 901 */ 902 static void 903 pipe_scheduler(void *arg) 904 { 905 pipe_poll(); 906 907 /* Reschedule */ 908 callout_reset(&polling_timer, 1, &pipe_scheduler, NULL); 909 } 910 911 912 /* 913 * Traverse the list of all active hooks and attempt to dequeue 914 * some packets. Hooks with empty queues are not traversed since 915 * they are not linked into this list. 916 */ 917 static void 918 pipe_poll(void) 919 { 920 struct hookinfo *hinfo; 921 struct timeval now; 922 int old_gen_id = active_gen_id; 923 924 mtx_lock(&ng_pipe_giant); 925 microuptime(&now); 926 LIST_FOREACH(hinfo, &active_head, active_le) { 927 CURVNET_SET(NG_HOOK_NODE(hinfo->hook)->nd_vnet); 928 pipe_dequeue(hinfo, &now); 929 CURVNET_RESTORE(); 930 mtx_lock(&ng_pipe_giant); 931 if (old_gen_id != active_gen_id) { 932 /* the list was updated; restart traversing */ 933 hinfo = LIST_FIRST(&active_head); 934 if (hinfo == NULL) 935 break; 936 old_gen_id = active_gen_id; 937 continue; 938 } 939 } 940 mtx_unlock(&ng_pipe_giant); 941 } 942 943 944 /* 945 * Shutdown processing 946 * 947 * This is tricky. If we have both a lower and upper hook, then we 948 * probably want to extricate ourselves and leave the two peers 949 * still linked to each other. Otherwise we should just shut down as 950 * a normal node would. 951 */ 952 static int 953 ngp_shutdown(node_p node) 954 { 955 const priv_p priv = NG_NODE_PRIVATE(node); 956 957 if (priv->lower.hook && priv->upper.hook) 958 ng_bypass(priv->lower.hook, priv->upper.hook); 959 else { 960 if (priv->upper.hook != NULL) 961 ng_rmhook_self(priv->upper.hook); 962 if (priv->lower.hook != NULL) 963 ng_rmhook_self(priv->lower.hook); 964 } 965 NG_NODE_UNREF(node); 966 free(priv, M_NG_PIPE); 967 return (0); 968 } 969 970 971 /* 972 * Hook disconnection 973 */ 974 static int 975 ngp_disconnect(hook_p hook) 976 { 977 struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook); 978 struct ngp_fifo *ngp_f; 979 struct ngp_hdr *ngp_h; 980 int removed = 0; 981 982 mtx_lock(&ng_pipe_giant); 983 984 KASSERT(hinfo != NULL, ("%s: null info", __FUNCTION__)); 985 hinfo->hook = NULL; 986 987 /* Flush all fifo queues associated with the hook */ 988 while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) { 989 while ((ngp_h = TAILQ_FIRST(&ngp_f->packet_head))) { 990 TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link); 991 m_freem(ngp_h->m); 992 uma_zfree(ngp_zone, ngp_h); 993 removed++; 994 } 995 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le); 996 uma_zfree(ngp_zone, ngp_f); 997 } 998 999 /* Flush the delay queue */ 1000 while ((ngp_h = TAILQ_FIRST(&hinfo->qout_head))) { 1001 TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link); 1002 m_freem(ngp_h->m); 1003 uma_zfree(ngp_zone, ngp_h); 1004 removed++; 1005 } 1006 1007 /* 1008 * Both queues should be empty by now, so detach us from 1009 * the list of active queues 1010 */ 1011 if (removed) { 1012 LIST_REMOVE(hinfo, active_le); 1013 active_gen_id++; 1014 } 1015 if (hinfo->run.qin_frames + hinfo->run.qout_frames != removed) 1016 printf("Mismatch: queued=%d but removed=%d !?!", 1017 hinfo->run.qin_frames + hinfo->run.qout_frames, removed); 1018 1019 /* Release the packet loss probability table (BER) */ 1020 if (hinfo->ber_p) 1021 free(hinfo->ber_p, M_NG_PIPE); 1022 1023 mtx_unlock(&ng_pipe_giant); 1024 1025 return (0); 1026 } 1027 1028 static int 1029 ngp_modevent(module_t mod, int type, void *unused) 1030 { 1031 int error = 0; 1032 1033 switch (type) { 1034 case MOD_LOAD: 1035 ngp_zone = uma_zcreate("ng_pipe", max(sizeof(struct ngp_hdr), 1036 sizeof (struct ngp_fifo)), NULL, NULL, NULL, NULL, 1037 UMA_ALIGN_PTR, 0); 1038 if (ngp_zone == NULL) 1039 panic("ng_pipe: couldn't allocate descriptor zone"); 1040 1041 mtx_init(&ng_pipe_giant, "ng_pipe_giant", NULL, MTX_DEF); 1042 LIST_INIT(&active_head); 1043 callout_init(&polling_timer, CALLOUT_MPSAFE); 1044 callout_reset(&polling_timer, 1, &pipe_scheduler, NULL); 1045 break; 1046 case MOD_UNLOAD: 1047 callout_drain(&polling_timer); 1048 uma_zdestroy(ngp_zone); 1049 mtx_destroy(&ng_pipe_giant); 1050 break; 1051 default: 1052 error = EOPNOTSUPP; 1053 break; 1054 } 1055 1056 return (error); 1057 } 1058