]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netgraph/ng_pipe.c
zfs: merge openzfs/zfs@8a7407012
[FreeBSD/FreeBSD.git] / sys / netgraph / ng_pipe.c
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2004-2010 University of Zagreb
5  * Copyright (c) 2007-2008 FreeBSD Foundation
6  *
7  * This software was developed by the University of Zagreb and the
8  * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
9  * FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
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 #include <sys/param.h>
42 #include <sys/errno.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/time.h>
48
49 #include <vm/uma.h>
50
51 #include <net/vnet.h>
52
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
62 static MALLOC_DEFINE(M_NG_PIPE, "ng_pipe", "ng_pipe");
63
64 /* Packet header struct */
65 struct ngp_hdr {
66         TAILQ_ENTRY(ngp_hdr)    ngp_link;       /* next pkt in queue */
67         struct timeval          when;           /* this packet's due time */
68         struct mbuf             *m;             /* ptr to the packet data */
69 };
70 TAILQ_HEAD(p_head, ngp_hdr);
71
72 /* FIFO queue struct */
73 struct ngp_fifo {
74         TAILQ_ENTRY(ngp_fifo)   fifo_le;        /* list of active queues only */
75         struct p_head           packet_head;    /* FIFO queue head */
76         u_int32_t               hash;           /* flow signature */
77         struct timeval          vtime;          /* virtual time, for WFQ */
78         u_int32_t               rr_deficit;     /* for DRR */
79         u_int32_t               packets;        /* # of packets in this queue */
80 };
81
82 /* Per hook info */
83 struct hookinfo {
84         hook_p                  hook;
85         int                     noqueue;        /* bypass any processing */
86         TAILQ_HEAD(, ngp_fifo)  fifo_head;      /* FIFO queues */
87         TAILQ_HEAD(, ngp_hdr)   qout_head;      /* delay queue head */
88         struct timeval          qin_utime;
89         struct ng_pipe_hookcfg  cfg;
90         struct ng_pipe_hookrun  run;
91         struct ng_pipe_hookstat stats;
92         uint64_t                *ber_p;         /* loss_p(BER,psize) map */
93 };
94
95 /* Per node info */
96 struct node_priv {
97         u_int64_t               delay;
98         u_int32_t               overhead;
99         u_int32_t               header_offset;
100         struct hookinfo         lower;
101         struct hookinfo         upper;
102         struct callout          timer;
103         int                     timer_scheduled;
104 };
105 typedef struct node_priv *priv_p;
106
107 /* Macro for calculating the virtual time for packet dequeueing in WFQ */
108 #define FIFO_VTIME_SORT(plen)                                           \
109         if (hinfo->cfg.wfq && hinfo->cfg.bandwidth) {                   \
110                 ngp_f->vtime.tv_usec = now->tv_usec + ((uint64_t) (plen) \
111                         + priv->overhead ) * hinfo->run.fifo_queues *   \
112                         8000000 / hinfo->cfg.bandwidth;                 \
113                 ngp_f->vtime.tv_sec = now->tv_sec +                     \
114                         ngp_f->vtime.tv_usec / 1000000;                 \
115                 ngp_f->vtime.tv_usec = ngp_f->vtime.tv_usec % 1000000;  \
116                 TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le)       \
117                         if (ngp_f1->vtime.tv_sec > ngp_f->vtime.tv_sec || \
118                             (ngp_f1->vtime.tv_sec == ngp_f->vtime.tv_sec && \
119                             ngp_f1->vtime.tv_usec > ngp_f->vtime.tv_usec)) \
120                                 break;                                  \
121                 if (ngp_f1 == NULL)                                     \
122                         TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le); \
123                 else                                                    \
124                         TAILQ_INSERT_BEFORE(ngp_f1, ngp_f, fifo_le);    \
125         } else                                                          \
126                 TAILQ_INSERT_TAIL(&hinfo->fifo_head, ngp_f, fifo_le);   \
127
128 static void     parse_cfg(struct ng_pipe_hookcfg *, struct ng_pipe_hookcfg *,
129                         struct hookinfo *, priv_p);
130 static void     pipe_dequeue(struct hookinfo *, struct timeval *);
131 static void     ngp_callout(node_p, hook_p, void *, int);
132 static int      ngp_modevent(module_t, int, void *);
133
134 /* zone for storing ngp_hdr-s */
135 static uma_zone_t ngp_zone;
136
137 /* Netgraph methods */
138 static ng_constructor_t ngp_constructor;
139 static ng_rcvmsg_t      ngp_rcvmsg;
140 static ng_shutdown_t    ngp_shutdown;
141 static ng_newhook_t     ngp_newhook;
142 static ng_rcvdata_t     ngp_rcvdata;
143 static ng_disconnect_t  ngp_disconnect;
144
145 /* Parse type for struct ng_pipe_hookstat */
146 static const struct ng_parse_struct_field
147         ng_pipe_hookstat_type_fields[] = NG_PIPE_HOOKSTAT_INFO;
148 static const struct ng_parse_type ng_pipe_hookstat_type = {
149         &ng_parse_struct_type,
150         &ng_pipe_hookstat_type_fields
151 };
152
153 /* Parse type for struct ng_pipe_stats */
154 static const struct ng_parse_struct_field ng_pipe_stats_type_fields[] =
155         NG_PIPE_STATS_INFO(&ng_pipe_hookstat_type);
156 static const struct ng_parse_type ng_pipe_stats_type = {
157         &ng_parse_struct_type,
158         &ng_pipe_stats_type_fields
159 };
160
161 /* Parse type for struct ng_pipe_hookrun */
162 static const struct ng_parse_struct_field
163         ng_pipe_hookrun_type_fields[] = NG_PIPE_HOOKRUN_INFO;
164 static const struct ng_parse_type ng_pipe_hookrun_type = {
165         &ng_parse_struct_type,
166         &ng_pipe_hookrun_type_fields
167 };
168
169 /* Parse type for struct ng_pipe_run */
170 static const struct ng_parse_struct_field
171         ng_pipe_run_type_fields[] = NG_PIPE_RUN_INFO(&ng_pipe_hookrun_type);
172 static const struct ng_parse_type ng_pipe_run_type = {
173         &ng_parse_struct_type,
174         &ng_pipe_run_type_fields
175 };
176
177 /* Parse type for struct ng_pipe_hookcfg */
178 static const struct ng_parse_struct_field
179         ng_pipe_hookcfg_type_fields[] = NG_PIPE_HOOKCFG_INFO;
180 static const struct ng_parse_type ng_pipe_hookcfg_type = {
181         &ng_parse_struct_type,
182         &ng_pipe_hookcfg_type_fields
183 };
184
185 /* Parse type for struct ng_pipe_cfg */
186 static const struct ng_parse_struct_field
187         ng_pipe_cfg_type_fields[] = NG_PIPE_CFG_INFO(&ng_pipe_hookcfg_type);
188 static const struct ng_parse_type ng_pipe_cfg_type = {
189         &ng_parse_struct_type,
190         &ng_pipe_cfg_type_fields
191 };
192
193 /* List of commands and how to convert arguments to/from ASCII */
194 static const struct ng_cmdlist ngp_cmds[] = {
195         {
196                 .cookie =       NGM_PIPE_COOKIE,
197                 .cmd =          NGM_PIPE_GET_STATS,
198                 .name =         "getstats",
199                 .respType =      &ng_pipe_stats_type
200         },
201         {
202                 .cookie =       NGM_PIPE_COOKIE,
203                 .cmd =          NGM_PIPE_CLR_STATS,
204                 .name =         "clrstats"
205         },
206         {
207                 .cookie =       NGM_PIPE_COOKIE,
208                 .cmd =          NGM_PIPE_GETCLR_STATS,
209                 .name =         "getclrstats",
210                 .respType =     &ng_pipe_stats_type
211         },
212         {
213                 .cookie =       NGM_PIPE_COOKIE,
214                 .cmd =          NGM_PIPE_GET_RUN,
215                 .name =         "getrun",
216                 .respType =     &ng_pipe_run_type
217         },
218         {
219                 .cookie =       NGM_PIPE_COOKIE,
220                 .cmd =          NGM_PIPE_GET_CFG,
221                 .name =         "getcfg",
222                 .respType =     &ng_pipe_cfg_type
223         },
224         {
225                 .cookie =       NGM_PIPE_COOKIE,
226                 .cmd =          NGM_PIPE_SET_CFG,
227                 .name =         "setcfg",
228                 .mesgType =     &ng_pipe_cfg_type,
229         },
230         { 0 }
231 };
232
233 /* Netgraph type descriptor */
234 static struct ng_type ng_pipe_typestruct = {
235         .version =      NG_ABI_VERSION,
236         .name =         NG_PIPE_NODE_TYPE,
237         .mod_event =    ngp_modevent,
238         .constructor =  ngp_constructor,
239         .shutdown =     ngp_shutdown,
240         .rcvmsg =       ngp_rcvmsg,
241         .newhook =      ngp_newhook,
242         .rcvdata =      ngp_rcvdata,
243         .disconnect =   ngp_disconnect,
244         .cmdlist =      ngp_cmds
245 };
246 NETGRAPH_INIT(pipe, &ng_pipe_typestruct);
247
248 /* Node constructor */
249 static int
250 ngp_constructor(node_p node)
251 {
252         priv_p priv;
253
254         priv = malloc(sizeof(*priv), M_NG_PIPE, M_ZERO | M_WAITOK);
255         NG_NODE_SET_PRIVATE(node, priv);
256
257         /* Mark node as single-threaded */
258         NG_NODE_FORCE_WRITER(node);
259
260         ng_callout_init(&priv->timer);
261
262         return (0);
263 }
264
265 /* Add a hook */
266 static int
267 ngp_newhook(node_p node, hook_p hook, const char *name)
268 {
269         const priv_p priv = NG_NODE_PRIVATE(node);
270         struct hookinfo *hinfo;
271
272         if (strcmp(name, NG_PIPE_HOOK_UPPER) == 0) {
273                 bzero(&priv->upper, sizeof(priv->upper));
274                 priv->upper.hook = hook;
275                 NG_HOOK_SET_PRIVATE(hook, &priv->upper);
276         } else if (strcmp(name, NG_PIPE_HOOK_LOWER) == 0) {
277                 bzero(&priv->lower, sizeof(priv->lower));
278                 priv->lower.hook = hook;
279                 NG_HOOK_SET_PRIVATE(hook, &priv->lower);
280         } else
281                 return (EINVAL);
282
283         /* Load non-zero initial cfg values */
284         hinfo = NG_HOOK_PRIVATE(hook);
285         hinfo->cfg.qin_size_limit = 50;
286         hinfo->cfg.fifo = 1;
287         hinfo->cfg.droptail = 1;
288         TAILQ_INIT(&hinfo->fifo_head);
289         TAILQ_INIT(&hinfo->qout_head);
290         return (0);
291 }
292
293 /* Receive a control message */
294 static int
295 ngp_rcvmsg(node_p node, item_p item, hook_p lasthook)
296 {
297         const priv_p priv = NG_NODE_PRIVATE(node);
298         struct ng_mesg *resp = NULL;
299         struct ng_mesg *msg, *flow_msg;
300         struct ng_pipe_stats *stats;
301         struct ng_pipe_run *run;
302         struct ng_pipe_cfg *cfg;
303         int error = 0;
304         int prev_down, now_down, cmd;
305
306         NGI_GET_MSG(item, msg);
307         switch (msg->header.typecookie) {
308         case NGM_PIPE_COOKIE:
309                 switch (msg->header.cmd) {
310                 case NGM_PIPE_GET_STATS:
311                 case NGM_PIPE_CLR_STATS:
312                 case NGM_PIPE_GETCLR_STATS:
313                         if (msg->header.cmd != NGM_PIPE_CLR_STATS) {
314                                 NG_MKRESPONSE(resp, msg,
315                                     sizeof(*stats), M_NOWAIT);
316                                 if (resp == NULL) {
317                                         error = ENOMEM;
318                                         break;
319                                 }
320                                 stats = (struct ng_pipe_stats *) resp->data;
321                                 bcopy(&priv->upper.stats, &stats->downstream,
322                                     sizeof(stats->downstream));
323                                 bcopy(&priv->lower.stats, &stats->upstream,
324                                     sizeof(stats->upstream));
325                         }
326                         if (msg->header.cmd != NGM_PIPE_GET_STATS) {
327                                 bzero(&priv->upper.stats,
328                                     sizeof(priv->upper.stats));
329                                 bzero(&priv->lower.stats,
330                                     sizeof(priv->lower.stats));
331                         }
332                         break;
333                 case NGM_PIPE_GET_RUN:
334                         NG_MKRESPONSE(resp, msg, sizeof(*run), M_NOWAIT);
335                         if (resp == NULL) {
336                                 error = ENOMEM;
337                                 break;
338                         }
339                         run = (struct ng_pipe_run *) resp->data;
340                         bcopy(&priv->upper.run, &run->downstream,
341                                 sizeof(run->downstream));
342                         bcopy(&priv->lower.run, &run->upstream,
343                                 sizeof(run->upstream));
344                         break;
345                 case NGM_PIPE_GET_CFG:
346                         NG_MKRESPONSE(resp, msg, sizeof(*cfg), M_NOWAIT);
347                         if (resp == NULL) {
348                                 error = ENOMEM;
349                                 break;
350                         }
351                         cfg = (struct ng_pipe_cfg *) resp->data;
352                         bcopy(&priv->upper.cfg, &cfg->downstream,
353                                 sizeof(cfg->downstream));
354                         bcopy(&priv->lower.cfg, &cfg->upstream,
355                                 sizeof(cfg->upstream));
356                         cfg->delay = priv->delay;
357                         cfg->overhead = priv->overhead;
358                         cfg->header_offset = priv->header_offset;
359                         if (cfg->upstream.bandwidth ==
360                             cfg->downstream.bandwidth) {
361                                 cfg->bandwidth = cfg->upstream.bandwidth;
362                                 cfg->upstream.bandwidth = 0;
363                                 cfg->downstream.bandwidth = 0;
364                         } else
365                                 cfg->bandwidth = 0;
366                         break;
367                 case NGM_PIPE_SET_CFG:
368                         cfg = (struct ng_pipe_cfg *) msg->data;
369                         if (msg->header.arglen != sizeof(*cfg)) {
370                                 error = EINVAL;
371                                 break;
372                         }
373
374                         if (cfg->delay == -1)
375                                 priv->delay = 0;
376                         else if (cfg->delay > 0 && cfg->delay < 10000000)
377                                 priv->delay = cfg->delay;
378
379                         if (cfg->bandwidth == -1) {
380                                 priv->upper.cfg.bandwidth = 0;
381                                 priv->lower.cfg.bandwidth = 0;
382                                 priv->overhead = 0;
383                         } else if (cfg->bandwidth >= 100 &&
384                             cfg->bandwidth <= 1000000000) {
385                                 priv->upper.cfg.bandwidth = cfg->bandwidth;
386                                 priv->lower.cfg.bandwidth = cfg->bandwidth;
387                                 if (cfg->bandwidth >= 10000000)
388                                         priv->overhead = 8+4+12; /* Ethernet */
389                                 else
390                                         priv->overhead = 10; /* HDLC */
391                         }
392
393                         if (cfg->overhead == -1)
394                                 priv->overhead = 0;
395                         else if (cfg->overhead > 0 &&
396                             cfg->overhead < MAX_OHSIZE)
397                                 priv->overhead = cfg->overhead;
398
399                         if (cfg->header_offset == -1)
400                                 priv->header_offset = 0;
401                         else if (cfg->header_offset > 0 &&
402                             cfg->header_offset < 64)
403                                 priv->header_offset = cfg->header_offset;
404
405                         prev_down = priv->upper.cfg.ber == 1 ||
406                             priv->lower.cfg.ber == 1;
407                         parse_cfg(&priv->upper.cfg, &cfg->downstream,
408                             &priv->upper, priv);
409                         parse_cfg(&priv->lower.cfg, &cfg->upstream,
410                             &priv->lower, priv);
411                         now_down = priv->upper.cfg.ber == 1 ||
412                             priv->lower.cfg.ber == 1;
413
414                         if (prev_down != now_down) {
415                                 if (now_down)
416                                         cmd = NGM_LINK_IS_DOWN;
417                                 else
418                                         cmd = NGM_LINK_IS_UP;
419
420                                 if (priv->lower.hook != NULL) {
421                                         NG_MKMESSAGE(flow_msg, NGM_FLOW_COOKIE,
422                                             cmd, 0, M_NOWAIT);
423                                         if (flow_msg != NULL)
424                                                 NG_SEND_MSG_HOOK(error, node,
425                                                     flow_msg, priv->lower.hook,
426                                                     0);
427                                 }
428                                 if (priv->upper.hook != NULL) {
429                                         NG_MKMESSAGE(flow_msg, NGM_FLOW_COOKIE,
430                                             cmd, 0, M_NOWAIT);
431                                         if (flow_msg != NULL)
432                                                 NG_SEND_MSG_HOOK(error, node,
433                                                     flow_msg, priv->upper.hook,
434                                                     0);
435                                 }
436                         }
437                         break;
438                 default:
439                         error = EINVAL;
440                         break;
441                 }
442                 break;
443         default:
444                 error = EINVAL;
445                 break;
446         }
447         NG_RESPOND_MSG(error, node, item, resp);
448         NG_FREE_MSG(msg);
449
450         return (error);
451 }
452
453 static void
454 parse_cfg(struct ng_pipe_hookcfg *current, struct ng_pipe_hookcfg *new,
455         struct hookinfo *hinfo, priv_p priv)
456 {
457
458         if (new->ber == -1) {
459                 current->ber = 0;
460                 if (hinfo->ber_p) {
461                         free(hinfo->ber_p, M_NG_PIPE);
462                         hinfo->ber_p = NULL;
463                 }
464         } else if (new->ber >= 1 && new->ber <= 1000000000000) {
465                 static const uint64_t one = 0x1000000000000; /* = 2^48 */
466                 uint64_t p0, p;
467                 uint32_t fsize, i;
468
469                 if (hinfo->ber_p == NULL)
470                         hinfo->ber_p =
471                             malloc((MAX_FSIZE + MAX_OHSIZE) * sizeof(uint64_t),
472                             M_NG_PIPE, M_WAITOK);
473                 current->ber = new->ber;
474
475                 /*
476                  * For given BER and each frame size N (in bytes) calculate
477                  * the probability P_OK that the frame is clean:
478                  *
479                  * P_OK(BER,N) = (1 - 1/BER)^(N*8)
480                  *
481                  * We use a 64-bit fixed-point format with decimal point
482                  * positioned between bits 47 and 48.
483                  */
484                 p0 = one - one / new->ber;
485                 p = one;
486                 for (fsize = 0; fsize < MAX_FSIZE + MAX_OHSIZE; fsize++) {
487                         hinfo->ber_p[fsize] = p;
488                         for (i = 0; i < 8; i++)
489                                 p = (p * (p0 & 0xffff) >> 48) +
490                                     (p * ((p0 >> 16) & 0xffff) >> 32) +
491                                     (p * (p0 >> 32) >> 16);
492                 }
493         }
494
495         if (new->qin_size_limit == -1)
496                 current->qin_size_limit = 0;
497         else if (new->qin_size_limit >= 5) 
498                 current->qin_size_limit = new->qin_size_limit;
499
500         if (new->qout_size_limit == -1)
501                 current->qout_size_limit = 0;
502         else if (new->qout_size_limit >= 5)
503                 current->qout_size_limit = new->qout_size_limit;
504
505         if (new->duplicate == -1)
506                 current->duplicate = 0;
507         else if (new->duplicate > 0 && new->duplicate <= 50)
508                 current->duplicate = new->duplicate;
509
510         if (new->fifo) {
511                 current->fifo = 1;
512                 current->wfq = 0;
513                 current->drr = 0;
514         }
515
516         if (new->wfq) {
517                 current->fifo = 0;
518                 current->wfq = 1;
519                 current->drr = 0;
520         }
521
522         if (new->drr) {
523                 current->fifo = 0;
524                 current->wfq = 0;
525                 /* DRR quantum */
526                 if (new->drr >= 32)
527                         current->drr = new->drr;
528                 else
529                         current->drr = 2048;            /* default quantum */
530         }
531
532         if (new->droptail) {
533                 current->droptail = 1;
534                 current->drophead = 0;
535         }
536
537         if (new->drophead) {
538                 current->droptail = 0;
539                 current->drophead = 1;
540         }
541
542         if (new->bandwidth == -1) {
543                 current->bandwidth = 0;
544                 current->fifo = 1;
545                 current->wfq = 0;
546                 current->drr = 0;
547         } else if (new->bandwidth >= 100 && new->bandwidth <= 1000000000)
548                 current->bandwidth = new->bandwidth;
549
550         if (current->bandwidth | priv->delay | 
551             current->duplicate | current->ber)
552                 hinfo->noqueue = 0;
553         else
554                 hinfo->noqueue = 1;
555 }
556
557 /*
558  * Compute a hash signature for a packet. This function suffers from the
559  * NIH sindrome, so probably it would be wise to look around what other
560  * folks have found out to be a good and efficient IP hash function...
561  */
562 static int
563 ip_hash(struct mbuf *m, int offset)
564 {
565         u_int64_t i;
566         struct ip *ip = (struct ip *)(mtod(m, u_char *) + offset);
567
568         if (m->m_len < sizeof(struct ip) + offset ||
569             ip->ip_v != 4 || ip->ip_hl << 2 != sizeof(struct ip))
570                 return 0;
571
572         i = ((u_int64_t) ip->ip_src.s_addr ^
573             ((u_int64_t) ip->ip_src.s_addr << 13) ^
574             ((u_int64_t) ip->ip_dst.s_addr << 7) ^
575             ((u_int64_t) ip->ip_dst.s_addr << 19));
576         return (i ^ (i >> 32));
577 }
578
579 /*
580  * Receive data on a hook - both in upstream and downstream direction.
581  * We put the frame on the inbound queue, and try to initiate dequeuing
582  * sequence immediately. If inbound queue is full, discard one frame
583  * depending on dropping policy (from the head or from the tail of the
584  * queue).
585  */
586 static int
587 ngp_rcvdata(hook_p hook, item_p item)
588 {
589         struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook);
590         const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook));
591         struct timeval uuptime;
592         struct timeval *now = &uuptime;
593         struct ngp_fifo *ngp_f = NULL, *ngp_f1;
594         struct ngp_hdr *ngp_h = NULL;
595         struct mbuf *m;
596         int hash, plen;
597         int error = 0;
598
599         /*
600          * Shortcut from inbound to outbound hook when neither of
601          * bandwidth, delay, BER or duplication probability is
602          * configured, nor we have queued frames to drain.
603          */
604         if (hinfo->run.qin_frames == 0 && hinfo->run.qout_frames == 0 &&
605             hinfo->noqueue) {
606                 struct hookinfo *dest;
607                 if (hinfo == &priv->lower)
608                         dest = &priv->upper;
609                 else
610                         dest = &priv->lower;
611
612                 /* Send the frame. */
613                 plen = NGI_M(item)->m_pkthdr.len;
614                 NG_FWD_ITEM_HOOK(error, item, dest->hook);
615
616                 /* Update stats. */
617                 if (error) {
618                         hinfo->stats.out_disc_frames++;
619                         hinfo->stats.out_disc_octets += plen;
620                 } else {
621                         hinfo->stats.fwd_frames++;
622                         hinfo->stats.fwd_octets += plen;
623                 }
624
625                 return (error);
626         }
627
628         microuptime(now);
629
630         /*
631          * If this was an empty queue, update service deadline time.
632          */
633         if (hinfo->run.qin_frames == 0) {
634                 struct timeval *when = &hinfo->qin_utime;
635                 if (when->tv_sec < now->tv_sec || (when->tv_sec == now->tv_sec
636                     && when->tv_usec < now->tv_usec)) {
637                         when->tv_sec = now->tv_sec;
638                         when->tv_usec = now->tv_usec;
639                 }
640         }
641
642         /* Populate the packet header */
643         ngp_h = uma_zalloc(ngp_zone, M_NOWAIT);
644         KASSERT((ngp_h != NULL), ("ngp_h zalloc failed (1)"));
645         NGI_GET_M(item, m);
646         KASSERT(m != NULL, ("NGI_GET_M failed"));
647         ngp_h->m = m;
648         NG_FREE_ITEM(item);
649
650         if (hinfo->cfg.fifo)
651                 hash = 0;       /* all packets go into a single FIFO queue */
652         else
653                 hash = ip_hash(m, priv->header_offset);
654
655         /* Find the appropriate FIFO queue for the packet and enqueue it*/
656         TAILQ_FOREACH(ngp_f, &hinfo->fifo_head, fifo_le)
657                 if (hash == ngp_f->hash)
658                         break;
659         if (ngp_f == NULL) {
660                 ngp_f = uma_zalloc(ngp_zone, M_NOWAIT);
661                 KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (2)"));
662                 TAILQ_INIT(&ngp_f->packet_head);
663                 ngp_f->hash = hash;
664                 ngp_f->packets = 1;
665                 ngp_f->rr_deficit = hinfo->cfg.drr;     /* DRR quantum */
666                 hinfo->run.fifo_queues++;
667                 TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link);
668                 FIFO_VTIME_SORT(m->m_pkthdr.len);
669         } else {
670                 TAILQ_INSERT_TAIL(&ngp_f->packet_head, ngp_h, ngp_link);
671                 ngp_f->packets++;
672         }
673         hinfo->run.qin_frames++;
674         hinfo->run.qin_octets += m->m_pkthdr.len;
675
676         /* Discard a frame if inbound queue limit has been reached */
677         if (hinfo->run.qin_frames > hinfo->cfg.qin_size_limit) {
678                 struct mbuf *m1;
679                 int longest = 0;
680
681                 /* Find the longest queue */
682                 TAILQ_FOREACH(ngp_f1, &hinfo->fifo_head, fifo_le)
683                         if (ngp_f1->packets > longest) {
684                                 longest = ngp_f1->packets;
685                                 ngp_f = ngp_f1;
686                         }
687
688                 /* Drop a frame from the queue head/tail, depending on cfg */
689                 if (hinfo->cfg.drophead) 
690                         ngp_h = TAILQ_FIRST(&ngp_f->packet_head);
691                 else 
692                         ngp_h = TAILQ_LAST(&ngp_f->packet_head, p_head);
693                 TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
694                 m1 = ngp_h->m;
695                 uma_zfree(ngp_zone, ngp_h);
696                 hinfo->run.qin_octets -= m1->m_pkthdr.len;
697                 hinfo->stats.in_disc_octets += m1->m_pkthdr.len;
698                 m_freem(m1);
699                 if (--(ngp_f->packets) == 0) {
700                         TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
701                         uma_zfree(ngp_zone, ngp_f);
702                         hinfo->run.fifo_queues--;
703                 }
704                 hinfo->run.qin_frames--;
705                 hinfo->stats.in_disc_frames++;
706         }
707
708         /*
709          * Try to start the dequeuing process immediately.
710          */
711         pipe_dequeue(hinfo, now);
712
713         return (0);
714 }
715
716 /*
717  * Dequeueing sequence - we basically do the following:
718  *  1) Try to extract the frame from the inbound (bandwidth) queue;
719  *  2) In accordance to BER specified, discard the frame randomly;
720  *  3) If the frame survives BER, prepend it with delay info and move it
721  *     to outbound (delay) queue;
722  *  4) Loop to 2) until bandwidth quota for this timeslice is reached, or
723  *     inbound queue is flushed completely;
724  *  5) Dequeue frames from the outbound queue and send them downstream until
725  *     outbound queue is flushed completely, or the next frame in the queue
726  *     is not due to be dequeued yet
727  */
728 static void
729 pipe_dequeue(struct hookinfo *hinfo, struct timeval *now) {
730         static uint64_t rand, oldrand;
731         const node_p node = NG_HOOK_NODE(hinfo->hook);
732         const priv_p priv = NG_NODE_PRIVATE(node);
733         struct hookinfo *dest;
734         struct ngp_fifo *ngp_f, *ngp_f1;
735         struct ngp_hdr *ngp_h;
736         struct timeval *when;
737         struct mbuf *m;
738         int plen, error = 0;
739
740         /* Which one is the destination hook? */
741         if (hinfo == &priv->lower)
742                 dest = &priv->upper;
743         else
744                 dest = &priv->lower;
745
746         /* Bandwidth queue processing */
747         while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) {
748                 when = &hinfo->qin_utime;
749                 if (when->tv_sec > now->tv_sec || (when->tv_sec == now->tv_sec
750                     && when->tv_usec > now->tv_usec))
751                         break;
752
753                 ngp_h = TAILQ_FIRST(&ngp_f->packet_head);
754                 m = ngp_h->m;
755
756                 /* Deficit Round Robin (DRR) processing */
757                 if (hinfo->cfg.drr) {
758                         if (ngp_f->rr_deficit >= m->m_pkthdr.len) {
759                                 ngp_f->rr_deficit -= m->m_pkthdr.len;
760                         } else {
761                                 ngp_f->rr_deficit += hinfo->cfg.drr;
762                                 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
763                                 TAILQ_INSERT_TAIL(&hinfo->fifo_head,
764                                     ngp_f, fifo_le);
765                                 continue;
766                         }
767                 }
768
769                 /*
770                  * Either create a duplicate and pass it on, or dequeue
771                  * the original packet...
772                  */
773                 if (hinfo->cfg.duplicate &&
774                     random() % 100 <= hinfo->cfg.duplicate) {
775                         ngp_h = uma_zalloc(ngp_zone, M_NOWAIT);
776                         KASSERT(ngp_h != NULL, ("ngp_h zalloc failed (3)"));
777                         m = m_dup(m, M_NOWAIT);
778                         KASSERT(m != NULL, ("m_dup failed"));
779                         ngp_h->m = m;
780                 } else {
781                         TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
782                         hinfo->run.qin_frames--;
783                         hinfo->run.qin_octets -= m->m_pkthdr.len;
784                         ngp_f->packets--;
785                 }
786                 
787                 /* Calculate the serialization delay */
788                 if (hinfo->cfg.bandwidth) {
789                         hinfo->qin_utime.tv_usec +=
790                             ((uint64_t) m->m_pkthdr.len + priv->overhead ) *
791                             8000000 / hinfo->cfg.bandwidth;
792                         hinfo->qin_utime.tv_sec +=
793                             hinfo->qin_utime.tv_usec / 1000000;
794                         hinfo->qin_utime.tv_usec =
795                             hinfo->qin_utime.tv_usec % 1000000;
796                 }
797                 when = &ngp_h->when;
798                 when->tv_sec = hinfo->qin_utime.tv_sec;
799                 when->tv_usec = hinfo->qin_utime.tv_usec;
800
801                 /* Sort / rearrange inbound queues */
802                 if (ngp_f->packets) {
803                         if (hinfo->cfg.wfq) {
804                                 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
805                                 FIFO_VTIME_SORT(TAILQ_FIRST(
806                                     &ngp_f->packet_head)->m->m_pkthdr.len)
807                         }
808                 } else {
809                         TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
810                         uma_zfree(ngp_zone, ngp_f);
811                         hinfo->run.fifo_queues--;
812                 }
813
814                 /* Randomly discard the frame, according to BER setting */
815                 if (hinfo->cfg.ber) {
816                         oldrand = rand;
817                         rand = random();
818                         if (((oldrand ^ rand) << 17) >=
819                             hinfo->ber_p[priv->overhead + m->m_pkthdr.len]) {
820                                 hinfo->stats.out_disc_frames++;
821                                 hinfo->stats.out_disc_octets += m->m_pkthdr.len;
822                                 uma_zfree(ngp_zone, ngp_h);
823                                 m_freem(m);
824                                 continue;
825                         }
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                 when = &ngp_h->when;
852                 m = ngp_h->m;
853                 if (when->tv_sec > now->tv_sec ||
854                     (when->tv_sec == now->tv_sec &&
855                     when->tv_usec > now->tv_usec))
856                         break;
857
858                 /* Update outbound queue stats */
859                 plen = m->m_pkthdr.len;
860                 hinfo->run.qout_frames--;
861                 hinfo->run.qout_octets -= plen;
862
863                 /* Dequeue the packet from qout */
864                 TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link);
865                 uma_zfree(ngp_zone, ngp_h);
866
867                 NG_SEND_DATA(error, dest->hook, m, meta);
868                 if (error) {
869                         hinfo->stats.out_disc_frames++;
870                         hinfo->stats.out_disc_octets += plen;
871                 } else {
872                         hinfo->stats.fwd_frames++;
873                         hinfo->stats.fwd_octets += plen;
874                 }
875         }
876
877         if ((hinfo->run.qin_frames != 0 || hinfo->run.qout_frames != 0) &&
878             !priv->timer_scheduled) {
879                 ng_callout(&priv->timer, node, NULL, 1, ngp_callout, NULL, 0);
880                 priv->timer_scheduled = 1;
881         }
882 }
883
884 /*
885  * This routine is called on every clock tick.  We poll connected hooks
886  * for queued frames by calling pipe_dequeue().
887  */
888 static void
889 ngp_callout(node_p node, hook_p hook, void *arg1, int arg2)
890 {
891         const priv_p priv = NG_NODE_PRIVATE(node);
892         struct timeval now;
893
894         priv->timer_scheduled = 0;
895         microuptime(&now);
896         if (priv->upper.hook != NULL)
897                 pipe_dequeue(&priv->upper, &now);
898         if (priv->lower.hook != NULL)
899                 pipe_dequeue(&priv->lower, &now);
900 }
901
902 /*
903  * Shutdown processing
904  *
905  * This is tricky. If we have both a lower and upper hook, then we
906  * probably want to extricate ourselves and leave the two peers
907  * still linked to each other. Otherwise we should just shut down as
908  * a normal node would.
909  */
910 static int
911 ngp_shutdown(node_p node)
912 {
913         const priv_p priv = NG_NODE_PRIVATE(node);
914
915         if (priv->timer_scheduled)
916                 ng_uncallout(&priv->timer, node);
917         if (priv->lower.hook && priv->upper.hook)
918                 ng_bypass(priv->lower.hook, priv->upper.hook);
919         else {
920                 if (priv->upper.hook != NULL)
921                         ng_rmhook_self(priv->upper.hook);
922                 if (priv->lower.hook != NULL)
923                         ng_rmhook_self(priv->lower.hook);
924         }
925         NG_NODE_UNREF(node);
926         free(priv, M_NG_PIPE);
927         return (0);
928 }
929
930 /*
931  * Hook disconnection
932  */
933 static int
934 ngp_disconnect(hook_p hook)
935 {
936         struct hookinfo *const hinfo = NG_HOOK_PRIVATE(hook);
937         struct ngp_fifo *ngp_f;
938         struct ngp_hdr *ngp_h;
939
940         KASSERT(hinfo != NULL, ("%s: null info", __FUNCTION__));
941         hinfo->hook = NULL;
942
943         /* Flush all fifo queues associated with the hook */
944         while ((ngp_f = TAILQ_FIRST(&hinfo->fifo_head))) {
945                 while ((ngp_h = TAILQ_FIRST(&ngp_f->packet_head))) {
946                         TAILQ_REMOVE(&ngp_f->packet_head, ngp_h, ngp_link);
947                         m_freem(ngp_h->m);
948                         uma_zfree(ngp_zone, ngp_h);
949                 }
950                 TAILQ_REMOVE(&hinfo->fifo_head, ngp_f, fifo_le);
951                 uma_zfree(ngp_zone, ngp_f);
952         }
953
954         /* Flush the delay queue */
955         while ((ngp_h = TAILQ_FIRST(&hinfo->qout_head))) {
956                 TAILQ_REMOVE(&hinfo->qout_head, ngp_h, ngp_link);
957                 m_freem(ngp_h->m);
958                 uma_zfree(ngp_zone, ngp_h);
959         }
960
961         /* Release the packet loss probability table (BER) */
962         if (hinfo->ber_p)
963                 free(hinfo->ber_p, M_NG_PIPE);
964
965         return (0);
966 }
967
968 static int
969 ngp_modevent(module_t mod, int type, void *unused)
970 {
971         int error = 0;
972
973         switch (type) {
974         case MOD_LOAD:
975                 ngp_zone = uma_zcreate("ng_pipe", max(sizeof(struct ngp_hdr),
976                     sizeof (struct ngp_fifo)), NULL, NULL, NULL, NULL,
977                     UMA_ALIGN_PTR, 0);
978                 if (ngp_zone == NULL)
979                         panic("ng_pipe: couldn't allocate descriptor zone");
980                 break;
981         case MOD_UNLOAD:
982                 uma_zdestroy(ngp_zone);
983                 break;
984         default:
985                 error = EOPNOTSUPP;
986                 break;
987         }
988
989         return (error);
990 }