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