The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/netinet/ip_dummynet.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * Copyright (c) 1998-2002 Luigi Rizzo, Universita` di Pisa
    3  * Portions Copyright (c) 2000 Akamba Corp.
    4  * All rights reserved
    5  *
    6  * Redistribution and use in source and binary forms, with or without
    7  * modification, are permitted provided that the following conditions
    8  * are met:
    9  * 1. Redistributions of source code must retain the above copyright
   10  *    notice, this list of conditions and the following disclaimer.
   11  * 2. Redistributions in binary form must reproduce the above copyright
   12  *    notice, this list of conditions and the following disclaimer in the
   13  *    documentation and/or other materials provided with the distribution.
   14  *
   15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   25  * SUCH DAMAGE.
   26  *
   27  * $FreeBSD: releng/5.2/sys/netinet/ip_dummynet.c 123169 2003-12-06 09:01:25Z emax $
   28  */
   29 
   30 #define DUMMYNET_DEBUG
   31 
   32 /*
   33  * This module implements IP dummynet, a bandwidth limiter/delay emulator
   34  * used in conjunction with the ipfw package.
   35  * Description of the data structures used is in ip_dummynet.h
   36  * Here you mainly find the following blocks of code:
   37  *  + variable declarations;
   38  *  + heap management functions;
   39  *  + scheduler and dummynet functions;
   40  *  + configuration and initialization.
   41  *
   42  * NOTA BENE: critical sections are protected by the "dummynet lock".
   43  *
   44  * Most important Changes:
   45  *
   46  * 011004: KLDable
   47  * 010124: Fixed WF2Q behaviour
   48  * 010122: Fixed spl protection.
   49  * 000601: WF2Q support
   50  * 000106: large rewrite, use heaps to handle very many pipes.
   51  * 980513:      initial release
   52  *
   53  * include files marked with XXX are probably not needed
   54  */
   55 
   56 #include <sys/param.h>
   57 #include <sys/systm.h>
   58 #include <sys/malloc.h>
   59 #include <sys/mbuf.h>
   60 #include <sys/kernel.h>
   61 #include <sys/module.h>
   62 #include <sys/proc.h>
   63 #include <sys/socket.h>
   64 #include <sys/socketvar.h>
   65 #include <sys/time.h>
   66 #include <sys/sysctl.h>
   67 #include <net/if.h>
   68 #include <net/route.h>
   69 #include <netinet/in.h>
   70 #include <netinet/in_systm.h>
   71 #include <netinet/in_var.h>
   72 #include <netinet/ip.h>
   73 #include <netinet/ip_fw.h>
   74 #include <netinet/ip_dummynet.h>
   75 #include <netinet/ip_var.h>
   76 
   77 #include <netinet/if_ether.h> /* for struct arpcom */
   78 #include <net/bridge.h>
   79 
   80 /*
   81  * We keep a private variable for the simulation time, but we could
   82  * probably use an existing one ("softticks" in sys/kern/kern_timer.c)
   83  */
   84 static dn_key curr_time = 0 ; /* current simulation time */
   85 
   86 static int dn_hash_size = 64 ;  /* default hash size */
   87 
   88 /* statistics on number of queue searches and search steps */
   89 static int searches, search_steps ;
   90 static int pipe_expire = 1 ;   /* expire queue if empty */
   91 static int dn_max_ratio = 16 ; /* max queues/buckets ratio */
   92 
   93 static int red_lookup_depth = 256;      /* RED - default lookup table depth */
   94 static int red_avg_pkt_size = 512;      /* RED - default medium packet size */
   95 static int red_max_pkt_size = 1500;     /* RED - default max packet size */
   96 
   97 /*
   98  * Three heaps contain queues and pipes that the scheduler handles:
   99  *
  100  * ready_heap contains all dn_flow_queue related to fixed-rate pipes.
  101  *
  102  * wfq_ready_heap contains the pipes associated with WF2Q flows
  103  *
  104  * extract_heap contains pipes associated with delay lines.
  105  *
  106  */
  107 
  108 MALLOC_DEFINE(M_DUMMYNET, "dummynet", "dummynet heap");
  109 
  110 static struct dn_heap ready_heap, extract_heap, wfq_ready_heap ;
  111 
  112 static int heap_init(struct dn_heap *h, int size) ;
  113 static int heap_insert (struct dn_heap *h, dn_key key1, void *p);
  114 static void heap_extract(struct dn_heap *h, void *obj);
  115 
  116 static void transmit_event(struct dn_pipe *pipe);
  117 static void ready_event(struct dn_flow_queue *q);
  118 
  119 static struct dn_pipe *all_pipes = NULL ;       /* list of all pipes */
  120 static struct dn_flow_set *all_flow_sets = NULL ;/* list of all flow_sets */
  121 
  122 static struct callout dn_timeout;
  123 
  124 #ifdef SYSCTL_NODE
  125 SYSCTL_NODE(_net_inet_ip, OID_AUTO, dummynet,
  126                 CTLFLAG_RW, 0, "Dummynet");
  127 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, hash_size,
  128             CTLFLAG_RW, &dn_hash_size, 0, "Default hash table size");
  129 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, curr_time,
  130             CTLFLAG_RD, &curr_time, 0, "Current tick");
  131 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, ready_heap,
  132             CTLFLAG_RD, &ready_heap.size, 0, "Size of ready heap");
  133 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, extract_heap,
  134             CTLFLAG_RD, &extract_heap.size, 0, "Size of extract heap");
  135 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, searches,
  136             CTLFLAG_RD, &searches, 0, "Number of queue searches");
  137 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, search_steps,
  138             CTLFLAG_RD, &search_steps, 0, "Number of queue search steps");
  139 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, expire,
  140             CTLFLAG_RW, &pipe_expire, 0, "Expire queue if empty");
  141 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, max_chain_len,
  142             CTLFLAG_RW, &dn_max_ratio, 0,
  143         "Max ratio between dynamic queues and buckets");
  144 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_lookup_depth,
  145         CTLFLAG_RD, &red_lookup_depth, 0, "Depth of RED lookup table");
  146 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_avg_pkt_size,
  147         CTLFLAG_RD, &red_avg_pkt_size, 0, "RED Medium packet size");
  148 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, red_max_pkt_size,
  149         CTLFLAG_RD, &red_max_pkt_size, 0, "RED Max packet size");
  150 #endif
  151 
  152 #ifdef DUMMYNET_DEBUG
  153 int     dummynet_debug = 0;
  154 #ifdef SYSCTL_NODE
  155 SYSCTL_INT(_net_inet_ip_dummynet, OID_AUTO, debug, CTLFLAG_RW, &dummynet_debug,
  156             0, "control debugging printfs");
  157 #endif
  158 #define DPRINTF(X)      if (dummynet_debug) printf X
  159 #else
  160 #define DPRINTF(X)
  161 #endif
  162 
  163 static struct mtx dummynet_mtx;
  164 /*
  165  * NB: Recursion is needed to deal with re-entry via ICMP.  That is,
  166  *     a packet may be dispatched via ip_input from dummynet_io and
  167  *     re-enter through ip_output.  Yech.
  168  */
  169 #define DUMMYNET_LOCK_INIT() \
  170         mtx_init(&dummynet_mtx, "dummynet", NULL, MTX_DEF | MTX_RECURSE)
  171 #define DUMMYNET_LOCK_DESTROY() mtx_destroy(&dummynet_mtx)
  172 #define DUMMYNET_LOCK()         mtx_lock(&dummynet_mtx)
  173 #define DUMMYNET_UNLOCK()       mtx_unlock(&dummynet_mtx)
  174 #define DUMMYNET_LOCK_ASSERT()  mtx_assert(&dummynet_mtx, MA_OWNED)
  175 
  176 static int config_pipe(struct dn_pipe *p);
  177 static int ip_dn_ctl(struct sockopt *sopt);
  178 
  179 static void rt_unref(struct rtentry *, const char *);
  180 static void dummynet(void *);
  181 static void dummynet_flush(void);
  182 void dummynet_drain(void);
  183 static ip_dn_io_t dummynet_io;
  184 static void dn_rule_delete(void *);
  185 
  186 int if_tx_rdy(struct ifnet *ifp);
  187 
  188 static void
  189 rt_unref(struct rtentry *rt, const char *where)
  190 {
  191     if (rt == NULL)
  192         return ;
  193     RT_LOCK(rt);
  194     if (rt->rt_refcnt <= 0) {
  195         printf("dummynet: warning, refcnt now %ld, decreasing (%s)\n",
  196             rt->rt_refcnt, where);
  197     }
  198     RTFREE_LOCKED(rt);
  199 }
  200 
  201 /*
  202  * Heap management functions.
  203  *
  204  * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2.
  205  * Some macros help finding parent/children so we can optimize them.
  206  *
  207  * heap_init() is called to expand the heap when needed.
  208  * Increment size in blocks of 16 entries.
  209  * XXX failure to allocate a new element is a pretty bad failure
  210  * as we basically stall a whole queue forever!!
  211  * Returns 1 on error, 0 on success
  212  */
  213 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 )
  214 #define HEAP_LEFT(x) ( 2*(x) + 1 )
  215 #define HEAP_IS_LEFT(x) ( (x) & 1 )
  216 #define HEAP_RIGHT(x) ( 2*(x) + 2 )
  217 #define HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; }
  218 #define HEAP_INCREMENT  15
  219 
  220 static int
  221 heap_init(struct dn_heap *h, int new_size)
  222 {
  223     struct dn_heap_entry *p;
  224 
  225     if (h->size >= new_size ) {
  226         printf("dummynet: %s, Bogus call, have %d want %d\n", __func__,
  227                 h->size, new_size);
  228         return 0 ;
  229     }
  230     new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT ;
  231     p = malloc(new_size * sizeof(*p), M_DUMMYNET, M_NOWAIT);
  232     if (p == NULL) {
  233         printf("dummynet: %s, resize %d failed\n", __func__, new_size );
  234         return 1 ; /* error */
  235     }
  236     if (h->size > 0) {
  237         bcopy(h->p, p, h->size * sizeof(*p) );
  238         free(h->p, M_DUMMYNET);
  239     }
  240     h->p = p ;
  241     h->size = new_size ;
  242     return 0 ;
  243 }
  244 
  245 /*
  246  * Insert element in heap. Normally, p != NULL, we insert p in
  247  * a new position and bubble up. If p == NULL, then the element is
  248  * already in place, and key is the position where to start the
  249  * bubble-up.
  250  * Returns 1 on failure (cannot allocate new heap entry)
  251  *
  252  * If offset > 0 the position (index, int) of the element in the heap is
  253  * also stored in the element itself at the given offset in bytes.
  254  */
  255 #define SET_OFFSET(heap, node) \
  256     if (heap->offset > 0) \
  257             *((int *)((char *)(heap->p[node].object) + heap->offset)) = node ;
  258 /*
  259  * RESET_OFFSET is used for sanity checks. It sets offset to an invalid value.
  260  */
  261 #define RESET_OFFSET(heap, node) \
  262     if (heap->offset > 0) \
  263             *((int *)((char *)(heap->p[node].object) + heap->offset)) = -1 ;
  264 static int
  265 heap_insert(struct dn_heap *h, dn_key key1, void *p)
  266 {
  267     int son = h->elements ;
  268 
  269     if (p == NULL)      /* data already there, set starting point */
  270         son = key1 ;
  271     else {              /* insert new element at the end, possibly resize */
  272         son = h->elements ;
  273         if (son == h->size) /* need resize... */
  274             if (heap_init(h, h->elements+1) )
  275                 return 1 ; /* failure... */
  276         h->p[son].object = p ;
  277         h->p[son].key = key1 ;
  278         h->elements++ ;
  279     }
  280     while (son > 0) {                           /* bubble up */
  281         int father = HEAP_FATHER(son) ;
  282         struct dn_heap_entry tmp  ;
  283 
  284         if (DN_KEY_LT( h->p[father].key, h->p[son].key ) )
  285             break ; /* found right position */
  286         /* son smaller than father, swap and repeat */
  287         HEAP_SWAP(h->p[son], h->p[father], tmp) ;
  288         SET_OFFSET(h, son);
  289         son = father ;
  290     }
  291     SET_OFFSET(h, son);
  292     return 0 ;
  293 }
  294 
  295 /*
  296  * remove top element from heap, or obj if obj != NULL
  297  */
  298 static void
  299 heap_extract(struct dn_heap *h, void *obj)
  300 {
  301     int child, father, max = h->elements - 1 ;
  302 
  303     if (max < 0) {
  304         printf("dummynet: warning, extract from empty heap 0x%p\n", h);
  305         return ;
  306     }
  307     father = 0 ; /* default: move up smallest child */
  308     if (obj != NULL) { /* extract specific element, index is at offset */
  309         if (h->offset <= 0)
  310             panic("dummynet: heap_extract from middle not supported on this heap!!!\n");
  311         father = *((int *)((char *)obj + h->offset)) ;
  312         if (father < 0 || father >= h->elements) {
  313             printf("dummynet: heap_extract, father %d out of bound 0..%d\n",
  314                 father, h->elements);
  315             panic("dummynet: heap_extract");
  316         }
  317     }
  318     RESET_OFFSET(h, father);
  319     child = HEAP_LEFT(father) ;         /* left child */
  320     while (child <= max) {              /* valid entry */
  321         if (child != max && DN_KEY_LT(h->p[child+1].key, h->p[child].key) )
  322             child = child+1 ;           /* take right child, otherwise left */
  323         h->p[father] = h->p[child] ;
  324         SET_OFFSET(h, father);
  325         father = child ;
  326         child = HEAP_LEFT(child) ;   /* left child for next loop */
  327     }
  328     h->elements-- ;
  329     if (father != max) {
  330         /*
  331          * Fill hole with last entry and bubble up, reusing the insert code
  332          */
  333         h->p[father] = h->p[max] ;
  334         heap_insert(h, father, NULL); /* this one cannot fail */
  335     }
  336 }
  337 
  338 #if 0
  339 /*
  340  * change object position and update references
  341  * XXX this one is never used!
  342  */
  343 static void
  344 heap_move(struct dn_heap *h, dn_key new_key, void *object)
  345 {
  346     int temp;
  347     int i ;
  348     int max = h->elements-1 ;
  349     struct dn_heap_entry buf ;
  350 
  351     if (h->offset <= 0)
  352         panic("cannot move items on this heap");
  353 
  354     i = *((int *)((char *)object + h->offset));
  355     if (DN_KEY_LT(new_key, h->p[i].key) ) { /* must move up */
  356         h->p[i].key = new_key ;
  357         for (; i>0 && DN_KEY_LT(new_key, h->p[(temp = HEAP_FATHER(i))].key) ;
  358                  i = temp ) { /* bubble up */
  359             HEAP_SWAP(h->p[i], h->p[temp], buf) ;
  360             SET_OFFSET(h, i);
  361         }
  362     } else {            /* must move down */
  363         h->p[i].key = new_key ;
  364         while ( (temp = HEAP_LEFT(i)) <= max ) { /* found left child */
  365             if ((temp != max) && DN_KEY_GT(h->p[temp].key, h->p[temp+1].key))
  366                 temp++ ; /* select child with min key */
  367             if (DN_KEY_GT(new_key, h->p[temp].key)) { /* go down */
  368                 HEAP_SWAP(h->p[i], h->p[temp], buf) ;
  369                 SET_OFFSET(h, i);
  370             } else
  371                 break ;
  372             i = temp ;
  373         }
  374     }
  375     SET_OFFSET(h, i);
  376 }
  377 #endif /* heap_move, unused */
  378 
  379 /*
  380  * heapify() will reorganize data inside an array to maintain the
  381  * heap property. It is needed when we delete a bunch of entries.
  382  */
  383 static void
  384 heapify(struct dn_heap *h)
  385 {
  386     int i ;
  387 
  388     for (i = 0 ; i < h->elements ; i++ )
  389         heap_insert(h, i , NULL) ;
  390 }
  391 
  392 /*
  393  * cleanup the heap and free data structure
  394  */
  395 static void
  396 heap_free(struct dn_heap *h)
  397 {
  398     if (h->size >0 )
  399         free(h->p, M_DUMMYNET);
  400     bzero(h, sizeof(*h) );
  401 }
  402 
  403 /*
  404  * --- end of heap management functions ---
  405  */
  406 
  407 /*
  408  * Scheduler functions:
  409  *
  410  * transmit_event() is called when the delay-line needs to enter
  411  * the scheduler, either because of existing pkts getting ready,
  412  * or new packets entering the queue. The event handled is the delivery
  413  * time of the packet.
  414  *
  415  * ready_event() does something similar with fixed-rate queues, and the
  416  * event handled is the finish time of the head pkt.
  417  *
  418  * wfq_ready_event() does something similar with WF2Q queues, and the
  419  * event handled is the start time of the head pkt.
  420  *
  421  * In all cases, we make sure that the data structures are consistent
  422  * before passing pkts out, because this might trigger recursive
  423  * invocations of the procedures.
  424  */
  425 static void
  426 transmit_event(struct dn_pipe *pipe)
  427 {
  428     struct dn_pkt *pkt ;
  429 
  430     DUMMYNET_LOCK_ASSERT();
  431 
  432     while ( (pkt = pipe->head) && DN_KEY_LEQ(pkt->output_time, curr_time) ) {
  433         /*
  434          * first unlink, then call procedures, since ip_input() can invoke
  435          * ip_output() and viceversa, thus causing nested calls
  436          */
  437         pipe->head = DN_NEXT(pkt) ;
  438 
  439         /* XXX: drop the lock for now to avoid LOR's */
  440         DUMMYNET_UNLOCK();
  441         /*
  442          * The actual mbuf is preceded by a struct dn_pkt, resembling an mbuf
  443          * (NOT A REAL one, just a small block of malloc'ed memory) with
  444          *     m_type = MT_TAG, m_flags = PACKET_TAG_DUMMYNET
  445          *     dn_m (m_next) = actual mbuf to be processed by ip_input/output
  446          * and some other fields.
  447          * The block IS FREED HERE because it contains parameters passed
  448          * to the called routine.
  449          */
  450         switch (pkt->dn_dir) {
  451         case DN_TO_IP_OUT:
  452             (void)ip_output((struct mbuf *)pkt, NULL, NULL, 0, NULL, NULL);
  453             rt_unref (pkt->ro.ro_rt, __func__) ;
  454             break ;
  455 
  456         case DN_TO_IP_IN :
  457             ip_input((struct mbuf *)pkt) ;
  458             break ;
  459 
  460         case DN_TO_BDG_FWD :
  461             if (!BDG_LOADED) {
  462                 /* somebody unloaded the bridge module. Drop pkt */
  463                 /* XXX rate limit */
  464                 printf("dummynet: dropping bridged packet trapped in pipe\n");
  465                 m_freem(pkt->dn_m);
  466                 break;
  467             } /* fallthrough */
  468         case DN_TO_ETH_DEMUX:
  469             {
  470                 struct mbuf *m = (struct mbuf *)pkt ;
  471 
  472                 if (pkt->dn_m->m_len < ETHER_HDR_LEN &&
  473                     (pkt->dn_m = m_pullup(pkt->dn_m, ETHER_HDR_LEN)) == NULL) {
  474                     printf("dummynet/bridge: pullup fail, dropping pkt\n");
  475                     break;
  476                 }
  477                 /*
  478                  * bdg_forward() wants a pointer to the pseudo-mbuf-header, but
  479                  * on return it will supply the pointer to the actual packet
  480                  * (originally pkt->dn_m, but could be something else now) if
  481                  * it has not consumed it.
  482                  */
  483                 if (pkt->dn_dir == DN_TO_BDG_FWD) {
  484                     m = bdg_forward_ptr(m, pkt->ifp);
  485                     if (m)
  486                         m_freem(m);
  487                 } else
  488                     ether_demux(NULL, m); /* which consumes the mbuf */
  489             }
  490             break ;
  491         case DN_TO_ETH_OUT:
  492             ether_output_frame(pkt->ifp, (struct mbuf *)pkt);
  493             break;
  494 
  495         default:
  496             printf("dummynet: bad switch %d!\n", pkt->dn_dir);
  497             m_freem(pkt->dn_m);
  498             break ;
  499         }
  500         free(pkt, M_DUMMYNET);
  501         DUMMYNET_LOCK();
  502     }
  503     /* if there are leftover packets, put into the heap for next event */
  504     if ( (pkt = pipe->head) )
  505          heap_insert(&extract_heap, pkt->output_time, pipe ) ;
  506     /* XXX should check errors on heap_insert, by draining the
  507      * whole pipe p and hoping in the future we are more successful
  508      */
  509 }
  510 
  511 /*
  512  * the following macro computes how many ticks we have to wait
  513  * before being able to transmit a packet. The credit is taken from
  514  * either a pipe (WF2Q) or a flow_queue (per-flow queueing)
  515  */
  516 #define SET_TICKS(pkt, q, p)    \
  517     (pkt->dn_m->m_pkthdr.len*8*hz - (q)->numbytes + p->bandwidth - 1 ) / \
  518             p->bandwidth ;
  519 
  520 /*
  521  * extract pkt from queue, compute output time (could be now)
  522  * and put into delay line (p_queue)
  523  */
  524 static void
  525 move_pkt(struct dn_pkt *pkt, struct dn_flow_queue *q,
  526         struct dn_pipe *p, int len)
  527 {
  528     q->head = DN_NEXT(pkt) ;
  529     q->len-- ;
  530     q->len_bytes -= len ;
  531 
  532     pkt->output_time = curr_time + p->delay ;
  533 
  534     if (p->head == NULL)
  535         p->head = pkt;
  536     else
  537         DN_NEXT(p->tail) = pkt;
  538     p->tail = pkt;
  539     DN_NEXT(p->tail) = NULL;
  540 }
  541 
  542 /*
  543  * ready_event() is invoked every time the queue must enter the
  544  * scheduler, either because the first packet arrives, or because
  545  * a previously scheduled event fired.
  546  * On invokation, drain as many pkts as possible (could be 0) and then
  547  * if there are leftover packets reinsert the pkt in the scheduler.
  548  */
  549 static void
  550 ready_event(struct dn_flow_queue *q)
  551 {
  552     struct dn_pkt *pkt;
  553     struct dn_pipe *p = q->fs->pipe ;
  554     int p_was_empty ;
  555 
  556     DUMMYNET_LOCK_ASSERT();
  557 
  558     if (p == NULL) {
  559         printf("dummynet: ready_event- pipe is gone\n");
  560         return ;
  561     }
  562     p_was_empty = (p->head == NULL) ;
  563 
  564     /*
  565      * schedule fixed-rate queues linked to this pipe:
  566      * Account for the bw accumulated since last scheduling, then
  567      * drain as many pkts as allowed by q->numbytes and move to
  568      * the delay line (in p) computing output time.
  569      * bandwidth==0 (no limit) means we can drain the whole queue,
  570      * setting len_scaled = 0 does the job.
  571      */
  572     q->numbytes += ( curr_time - q->sched_time ) * p->bandwidth;
  573     while ( (pkt = q->head) != NULL ) {
  574         int len = pkt->dn_m->m_pkthdr.len;
  575         int len_scaled = p->bandwidth ? len*8*hz : 0 ;
  576         if (len_scaled > q->numbytes )
  577             break ;
  578         q->numbytes -= len_scaled ;
  579         move_pkt(pkt, q, p, len);
  580     }
  581     /*
  582      * If we have more packets queued, schedule next ready event
  583      * (can only occur when bandwidth != 0, otherwise we would have
  584      * flushed the whole queue in the previous loop).
  585      * To this purpose we record the current time and compute how many
  586      * ticks to go for the finish time of the packet.
  587      */
  588     if ( (pkt = q->head) != NULL ) { /* this implies bandwidth != 0 */
  589         dn_key t = SET_TICKS(pkt, q, p); /* ticks i have to wait */
  590         q->sched_time = curr_time ;
  591         heap_insert(&ready_heap, curr_time + t, (void *)q );
  592         /* XXX should check errors on heap_insert, and drain the whole
  593          * queue on error hoping next time we are luckier.
  594          */
  595     } else {    /* RED needs to know when the queue becomes empty */
  596         q->q_time = curr_time;
  597         q->numbytes = 0;
  598     }
  599     /*
  600      * If the delay line was empty call transmit_event(p) now.
  601      * Otherwise, the scheduler will take care of it.
  602      */
  603     if (p_was_empty)
  604         transmit_event(p);
  605 }
  606 
  607 /*
  608  * Called when we can transmit packets on WF2Q queues. Take pkts out of
  609  * the queues at their start time, and enqueue into the delay line.
  610  * Packets are drained until p->numbytes < 0. As long as
  611  * len_scaled >= p->numbytes, the packet goes into the delay line
  612  * with a deadline p->delay. For the last packet, if p->numbytes<0,
  613  * there is an additional delay.
  614  */
  615 static void
  616 ready_event_wfq(struct dn_pipe *p)
  617 {
  618     int p_was_empty = (p->head == NULL) ;
  619     struct dn_heap *sch = &(p->scheduler_heap);
  620     struct dn_heap *neh = &(p->not_eligible_heap) ;
  621 
  622     DUMMYNET_LOCK_ASSERT();
  623 
  624     if (p->if_name[0] == 0) /* tx clock is simulated */
  625         p->numbytes += ( curr_time - p->sched_time ) * p->bandwidth;
  626     else { /* tx clock is for real, the ifq must be empty or this is a NOP */
  627         if (p->ifp && p->ifp->if_snd.ifq_head != NULL)
  628             return ;
  629         else {
  630             DPRINTF(("dummynet: pipe %d ready from %s --\n",
  631                 p->pipe_nr, p->if_name));
  632         }
  633     }
  634 
  635     /*
  636      * While we have backlogged traffic AND credit, we need to do
  637      * something on the queue.
  638      */
  639     while ( p->numbytes >=0 && (sch->elements>0 || neh->elements >0) ) {
  640         if (sch->elements > 0) { /* have some eligible pkts to send out */
  641             struct dn_flow_queue *q = sch->p[0].object ;
  642             struct dn_pkt *pkt = q->head;
  643             struct dn_flow_set *fs = q->fs;
  644             u_int64_t len = pkt->dn_m->m_pkthdr.len;
  645             int len_scaled = p->bandwidth ? len*8*hz : 0 ;
  646 
  647             heap_extract(sch, NULL); /* remove queue from heap */
  648             p->numbytes -= len_scaled ;
  649             move_pkt(pkt, q, p, len);
  650 
  651             p->V += (len<<MY_M) / p->sum ; /* update V */
  652             q->S = q->F ; /* update start time */
  653             if (q->len == 0) { /* Flow not backlogged any more */
  654                 fs->backlogged-- ;
  655                 heap_insert(&(p->idle_heap), q->F, q);
  656             } else { /* still backlogged */
  657                 /*
  658                  * update F and position in backlogged queue, then
  659                  * put flow in not_eligible_heap (we will fix this later).
  660                  */
  661                 len = (q->head)->dn_m->m_pkthdr.len;
  662                 q->F += (len<<MY_M)/(u_int64_t) fs->weight ;
  663                 if (DN_KEY_LEQ(q->S, p->V))
  664                     heap_insert(neh, q->S, q);
  665                 else
  666                     heap_insert(sch, q->F, q);
  667             }
  668         }
  669         /*
  670          * now compute V = max(V, min(S_i)). Remember that all elements in sch
  671          * have by definition S_i <= V so if sch is not empty, V is surely
  672          * the max and we must not update it. Conversely, if sch is empty
  673          * we only need to look at neh.
  674          */
  675         if (sch->elements == 0 && neh->elements > 0)
  676             p->V = MAX64 ( p->V, neh->p[0].key );
  677         /* move from neh to sch any packets that have become eligible */
  678         while (neh->elements > 0 && DN_KEY_LEQ(neh->p[0].key, p->V) ) {
  679             struct dn_flow_queue *q = neh->p[0].object ;
  680             heap_extract(neh, NULL);
  681             heap_insert(sch, q->F, q);
  682         }
  683 
  684         if (p->if_name[0] != '\0') {/* tx clock is from a real thing */
  685             p->numbytes = -1 ; /* mark not ready for I/O */
  686             break ;
  687         }
  688     }
  689     if (sch->elements == 0 && neh->elements == 0 && p->numbytes >= 0
  690             && p->idle_heap.elements > 0) {
  691         /*
  692          * no traffic and no events scheduled. We can get rid of idle-heap.
  693          */
  694         int i ;
  695 
  696         for (i = 0 ; i < p->idle_heap.elements ; i++) {
  697             struct dn_flow_queue *q = p->idle_heap.p[i].object ;
  698 
  699             q->F = 0 ;
  700             q->S = q->F + 1 ;
  701         }
  702         p->sum = 0 ;
  703         p->V = 0 ;
  704         p->idle_heap.elements = 0 ;
  705     }
  706     /*
  707      * If we are getting clocks from dummynet (not a real interface) and
  708      * If we are under credit, schedule the next ready event.
  709      * Also fix the delivery time of the last packet.
  710      */
  711     if (p->if_name[0]==0 && p->numbytes < 0) { /* this implies bandwidth >0 */
  712         dn_key t=0 ; /* number of ticks i have to wait */
  713 
  714         if (p->bandwidth > 0)
  715             t = ( p->bandwidth -1 - p->numbytes) / p->bandwidth ;
  716         p->tail->output_time += t ;
  717         p->sched_time = curr_time ;
  718         heap_insert(&wfq_ready_heap, curr_time + t, (void *)p);
  719         /* XXX should check errors on heap_insert, and drain the whole
  720          * queue on error hoping next time we are luckier.
  721          */
  722     }
  723     /*
  724      * If the delay line was empty call transmit_event(p) now.
  725      * Otherwise, the scheduler will take care of it.
  726      */
  727     if (p_was_empty)
  728         transmit_event(p);
  729 }
  730 
  731 /*
  732  * This is called once per tick, or HZ times per second. It is used to
  733  * increment the current tick counter and schedule expired events.
  734  */
  735 static void
  736 dummynet(void * __unused unused)
  737 {
  738     void *p ; /* generic parameter to handler */
  739     struct dn_heap *h ;
  740     struct dn_heap *heaps[3];
  741     int i;
  742     struct dn_pipe *pe ;
  743 
  744     heaps[0] = &ready_heap ;            /* fixed-rate queues */
  745     heaps[1] = &wfq_ready_heap ;        /* wfq queues */
  746     heaps[2] = &extract_heap ;          /* delay line */
  747 
  748     DUMMYNET_LOCK();
  749     curr_time++ ;
  750     for (i=0; i < 3 ; i++) {
  751         h = heaps[i];
  752         while (h->elements > 0 && DN_KEY_LEQ(h->p[0].key, curr_time) ) {
  753             if (h->p[0].key > curr_time)
  754                 printf("dummynet: warning, heap %d is %d ticks late\n",
  755                     i, (int)(curr_time - h->p[0].key));
  756             p = h->p[0].object ; /* store a copy before heap_extract */
  757             heap_extract(h, NULL); /* need to extract before processing */
  758             if (i == 0)
  759                 ready_event(p) ;
  760             else if (i == 1) {
  761                 struct dn_pipe *pipe = p;
  762                 if (pipe->if_name[0] != '\0')
  763                     printf("dummynet: bad ready_event_wfq for pipe %s\n",
  764                         pipe->if_name);
  765                 else
  766                     ready_event_wfq(p) ;
  767             } else
  768                 transmit_event(p);
  769         }
  770     }
  771     /* sweep pipes trying to expire idle flow_queues */
  772     for (pe = all_pipes; pe ; pe = pe->next )
  773         if (pe->idle_heap.elements > 0 &&
  774                 DN_KEY_LT(pe->idle_heap.p[0].key, pe->V) ) {
  775             struct dn_flow_queue *q = pe->idle_heap.p[0].object ;
  776 
  777             heap_extract(&(pe->idle_heap), NULL);
  778             q->S = q->F + 1 ; /* mark timestamp as invalid */
  779             pe->sum -= q->fs->weight ;
  780         }
  781     DUMMYNET_UNLOCK();
  782 
  783     callout_reset(&dn_timeout, 1, dummynet, NULL);
  784 }
  785 
  786 /*
  787  * called by an interface when tx_rdy occurs.
  788  */
  789 int
  790 if_tx_rdy(struct ifnet *ifp)
  791 {
  792     struct dn_pipe *p;
  793 
  794     DUMMYNET_LOCK();
  795     for (p = all_pipes; p ; p = p->next )
  796         if (p->ifp == ifp)
  797             break ;
  798     if (p == NULL) {
  799         for (p = all_pipes; p ; p = p->next )
  800             if (!strcmp(p->if_name, ifp->if_xname) ) {
  801                 p->ifp = ifp ;
  802                 DPRINTF(("dummynet: ++ tx rdy from %s (now found)\n",
  803                         ifp->if_xname));
  804                 break ;
  805             }
  806     }
  807     if (p != NULL) {
  808         DPRINTF(("dummynet: ++ tx rdy from %s - qlen %d\n", ifp->if_xname,
  809                 ifp->if_snd.ifq_len));
  810         p->numbytes = 0 ; /* mark ready for I/O */
  811         ready_event_wfq(p);
  812     }
  813     DUMMYNET_UNLOCK();
  814 
  815     return 0;
  816 }
  817 
  818 /*
  819  * Unconditionally expire empty queues in case of shortage.
  820  * Returns the number of queues freed.
  821  */
  822 static int
  823 expire_queues(struct dn_flow_set *fs)
  824 {
  825     struct dn_flow_queue *q, *prev ;
  826     int i, initial_elements = fs->rq_elements ;
  827 
  828     if (fs->last_expired == time_second)
  829         return 0 ;
  830     fs->last_expired = time_second ;
  831     for (i = 0 ; i <= fs->rq_size ; i++) /* last one is overflow */
  832         for (prev=NULL, q = fs->rq[i] ; q != NULL ; )
  833             if (q->head != NULL || q->S != q->F+1) {
  834                 prev = q ;
  835                 q = q->next ;
  836             } else { /* entry is idle, expire it */
  837                 struct dn_flow_queue *old_q = q ;
  838 
  839                 if (prev != NULL)
  840                     prev->next = q = q->next ;
  841                 else
  842                     fs->rq[i] = q = q->next ;
  843                 fs->rq_elements-- ;
  844                 free(old_q, M_DUMMYNET);
  845             }
  846     return initial_elements - fs->rq_elements ;
  847 }
  848 
  849 /*
  850  * If room, create a new queue and put at head of slot i;
  851  * otherwise, create or use the default queue.
  852  */
  853 static struct dn_flow_queue *
  854 create_queue(struct dn_flow_set *fs, int i)
  855 {
  856     struct dn_flow_queue *q ;
  857 
  858     if (fs->rq_elements > fs->rq_size * dn_max_ratio &&
  859             expire_queues(fs) == 0) {
  860         /*
  861          * No way to get room, use or create overflow queue.
  862          */
  863         i = fs->rq_size ;
  864         if ( fs->rq[i] != NULL )
  865             return fs->rq[i] ;
  866     }
  867     q = malloc(sizeof(*q), M_DUMMYNET, M_NOWAIT | M_ZERO);
  868     if (q == NULL) {
  869         printf("dummynet: sorry, cannot allocate queue for new flow\n");
  870         return NULL ;
  871     }
  872     q->fs = fs ;
  873     q->hash_slot = i ;
  874     q->next = fs->rq[i] ;
  875     q->S = q->F + 1;   /* hack - mark timestamp as invalid */
  876     fs->rq[i] = q ;
  877     fs->rq_elements++ ;
  878     return q ;
  879 }
  880 
  881 /*
  882  * Given a flow_set and a pkt in last_pkt, find a matching queue
  883  * after appropriate masking. The queue is moved to front
  884  * so that further searches take less time.
  885  */
  886 static struct dn_flow_queue *
  887 find_queue(struct dn_flow_set *fs, struct ipfw_flow_id *id)
  888 {
  889     int i = 0 ; /* we need i and q for new allocations */
  890     struct dn_flow_queue *q, *prev;
  891 
  892     if ( !(fs->flags_fs & DN_HAVE_FLOW_MASK) )
  893         q = fs->rq[0] ;
  894     else {
  895         /* first, do the masking */
  896         id->dst_ip &= fs->flow_mask.dst_ip ;
  897         id->src_ip &= fs->flow_mask.src_ip ;
  898         id->dst_port &= fs->flow_mask.dst_port ;
  899         id->src_port &= fs->flow_mask.src_port ;
  900         id->proto &= fs->flow_mask.proto ;
  901         id->flags = 0 ; /* we don't care about this one */
  902         /* then, hash function */
  903         i = ( (id->dst_ip) & 0xffff ) ^
  904             ( (id->dst_ip >> 15) & 0xffff ) ^
  905             ( (id->src_ip << 1) & 0xffff ) ^
  906             ( (id->src_ip >> 16 ) & 0xffff ) ^
  907             (id->dst_port << 1) ^ (id->src_port) ^
  908             (id->proto );
  909         i = i % fs->rq_size ;
  910         /* finally, scan the current list for a match */
  911         searches++ ;
  912         for (prev=NULL, q = fs->rq[i] ; q ; ) {
  913             search_steps++;
  914             if (id->dst_ip == q->id.dst_ip &&
  915                     id->src_ip == q->id.src_ip &&
  916                     id->dst_port == q->id.dst_port &&
  917                     id->src_port == q->id.src_port &&
  918                     id->proto == q->id.proto &&
  919                     id->flags == q->id.flags)
  920                 break ; /* found */
  921             else if (pipe_expire && q->head == NULL && q->S == q->F+1 ) {
  922                 /* entry is idle and not in any heap, expire it */
  923                 struct dn_flow_queue *old_q = q ;
  924 
  925                 if (prev != NULL)
  926                     prev->next = q = q->next ;
  927                 else
  928                     fs->rq[i] = q = q->next ;
  929                 fs->rq_elements-- ;
  930                 free(old_q, M_DUMMYNET);
  931                 continue ;
  932             }
  933             prev = q ;
  934             q = q->next ;
  935         }
  936         if (q && prev != NULL) { /* found and not in front */
  937             prev->next = q->next ;
  938             q->next = fs->rq[i] ;
  939             fs->rq[i] = q ;
  940         }
  941     }
  942     if (q == NULL) { /* no match, need to allocate a new entry */
  943         q = create_queue(fs, i);
  944         if (q != NULL)
  945         q->id = *id ;
  946     }
  947     return q ;
  948 }
  949 
  950 static int
  951 red_drops(struct dn_flow_set *fs, struct dn_flow_queue *q, int len)
  952 {
  953     /*
  954      * RED algorithm
  955      *
  956      * RED calculates the average queue size (avg) using a low-pass filter
  957      * with an exponential weighted (w_q) moving average:
  958      *  avg  <-  (1-w_q) * avg + w_q * q_size
  959      * where q_size is the queue length (measured in bytes or * packets).
  960      *
  961      * If q_size == 0, we compute the idle time for the link, and set
  962      *  avg = (1 - w_q)^(idle/s)
  963      * where s is the time needed for transmitting a medium-sized packet.
  964      *
  965      * Now, if avg < min_th the packet is enqueued.
  966      * If avg > max_th the packet is dropped. Otherwise, the packet is
  967      * dropped with probability P function of avg.
  968      *
  969      */
  970 
  971     int64_t p_b = 0;
  972     /* queue in bytes or packets ? */
  973     u_int q_size = (fs->flags_fs & DN_QSIZE_IS_BYTES) ? q->len_bytes : q->len;
  974 
  975     DPRINTF(("\ndummynet: %d q: %2u ", (int) curr_time, q_size));
  976 
  977     /* average queue size estimation */
  978     if (q_size != 0) {
  979         /*
  980          * queue is not empty, avg <- avg + (q_size - avg) * w_q
  981          */
  982         int diff = SCALE(q_size) - q->avg;
  983         int64_t v = SCALE_MUL((int64_t) diff, (int64_t) fs->w_q);
  984 
  985         q->avg += (int) v;
  986     } else {
  987         /*
  988          * queue is empty, find for how long the queue has been
  989          * empty and use a lookup table for computing
  990          * (1 - * w_q)^(idle_time/s) where s is the time to send a
  991          * (small) packet.
  992          * XXX check wraps...
  993          */
  994         if (q->avg) {
  995             u_int t = (curr_time - q->q_time) / fs->lookup_step;
  996 
  997             q->avg = (t < fs->lookup_depth) ?
  998                     SCALE_MUL(q->avg, fs->w_q_lookup[t]) : 0;
  999         }
 1000     }
 1001     DPRINTF(("dummynet: avg: %u ", SCALE_VAL(q->avg)));
 1002 
 1003     /* should i drop ? */
 1004 
 1005     if (q->avg < fs->min_th) {
 1006         q->count = -1;
 1007         return 0; /* accept packet ; */
 1008     }
 1009     if (q->avg >= fs->max_th) { /* average queue >=  max threshold */
 1010         if (fs->flags_fs & DN_IS_GENTLE_RED) {
 1011             /*
 1012              * According to Gentle-RED, if avg is greater than max_th the
 1013              * packet is dropped with a probability
 1014              *  p_b = c_3 * avg - c_4
 1015              * where c_3 = (1 - max_p) / max_th, and c_4 = 1 - 2 * max_p
 1016              */
 1017             p_b = SCALE_MUL((int64_t) fs->c_3, (int64_t) q->avg) - fs->c_4;
 1018         } else {
 1019             q->count = -1;
 1020             DPRINTF(("dummynet: - drop"));
 1021             return 1 ;
 1022         }
 1023     } else if (q->avg > fs->min_th) {
 1024         /*
 1025          * we compute p_b using the linear dropping function p_b = c_1 *
 1026          * avg - c_2, where c_1 = max_p / (max_th - min_th), and c_2 =
 1027          * max_p * min_th / (max_th - min_th)
 1028          */
 1029         p_b = SCALE_MUL((int64_t) fs->c_1, (int64_t) q->avg) - fs->c_2;
 1030     }
 1031     if (fs->flags_fs & DN_QSIZE_IS_BYTES)
 1032         p_b = (p_b * len) / fs->max_pkt_size;
 1033     if (++q->count == 0)
 1034         q->random = random() & 0xffff;
 1035     else {
 1036         /*
 1037          * q->count counts packets arrived since last drop, so a greater
 1038          * value of q->count means a greater packet drop probability.
 1039          */
 1040         if (SCALE_MUL(p_b, SCALE((int64_t) q->count)) > q->random) {
 1041             q->count = 0;
 1042             DPRINTF(("dummynet: - red drop"));
 1043             /* after a drop we calculate a new random value */
 1044             q->random = random() & 0xffff;
 1045             return 1;    /* drop */
 1046         }
 1047     }
 1048     /* end of RED algorithm */
 1049     return 0 ; /* accept */
 1050 }
 1051 
 1052 static __inline
 1053 struct dn_flow_set *
 1054 locate_flowset(int pipe_nr, struct ip_fw *rule)
 1055 {
 1056 #if IPFW2
 1057     struct dn_flow_set *fs;
 1058     ipfw_insn *cmd = rule->cmd + rule->act_ofs;
 1059 
 1060     if (cmd->opcode == O_LOG)
 1061         cmd += F_LEN(cmd);
 1062 #ifdef __i386__
 1063     fs = ((ipfw_insn_pipe *)cmd)->pipe_ptr;
 1064 #else
 1065     bcopy(& ((ipfw_insn_pipe *)cmd)->pipe_ptr, &fs, sizeof(fs));
 1066 #endif
 1067 
 1068     if (fs != NULL)
 1069         return fs;
 1070 
 1071     if (cmd->opcode == O_QUEUE)
 1072 #else /* !IPFW2 */
 1073     struct dn_flow_set *fs = NULL ;
 1074 
 1075     if ( (rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_QUEUE )
 1076 #endif /* !IPFW2 */
 1077         for (fs=all_flow_sets; fs && fs->fs_nr != pipe_nr; fs=fs->next)
 1078             ;
 1079     else {
 1080         struct dn_pipe *p1;
 1081         for (p1 = all_pipes; p1 && p1->pipe_nr != pipe_nr; p1 = p1->next)
 1082             ;
 1083         if (p1 != NULL)
 1084             fs = &(p1->fs) ;
 1085     }
 1086     /* record for the future */
 1087 #if IPFW2
 1088 #ifdef __i386__
 1089     ((ipfw_insn_pipe *)cmd)->pipe_ptr = fs;
 1090 #else
 1091     bcopy(&fs, & ((ipfw_insn_pipe *)cmd)->pipe_ptr, sizeof(fs));
 1092 #endif
 1093 #else
 1094     if (fs != NULL)
 1095         rule->pipe_ptr = fs;
 1096 #endif
 1097     return fs ;
 1098 }
 1099 
 1100 /*
 1101  * dummynet hook for packets. Below 'pipe' is a pipe or a queue
 1102  * depending on whether WF2Q or fixed bw is used.
 1103  *
 1104  * pipe_nr      pipe or queue the packet is destined for.
 1105  * dir          where shall we send the packet after dummynet.
 1106  * m            the mbuf with the packet
 1107  * ifp          the 'ifp' parameter from the caller.
 1108  *              NULL in ip_input, destination interface in ip_output,
 1109  *              real_dst in bdg_forward
 1110  * ro           route parameter (only used in ip_output, NULL otherwise)
 1111  * dst          destination address, only used by ip_output
 1112  * rule         matching rule, in case of multiple passes
 1113  * flags        flags from the caller, only used in ip_output
 1114  *
 1115  */
 1116 static int
 1117 dummynet_io(struct mbuf *m, int pipe_nr, int dir, struct ip_fw_args *fwa)
 1118 {
 1119     struct dn_pkt *pkt;
 1120     struct dn_flow_set *fs;
 1121     struct dn_pipe *pipe ;
 1122     u_int64_t len = m->m_pkthdr.len ;
 1123     struct dn_flow_queue *q = NULL ;
 1124     int is_pipe;
 1125 #if IPFW2
 1126     ipfw_insn *cmd = fwa->rule->cmd + fwa->rule->act_ofs;
 1127 
 1128     if (cmd->opcode == O_LOG)
 1129         cmd += F_LEN(cmd);
 1130     is_pipe = (cmd->opcode == O_PIPE);
 1131 #else
 1132     is_pipe = (fwa->rule->fw_flg & IP_FW_F_COMMAND) == IP_FW_F_PIPE;
 1133 #endif
 1134 
 1135     pipe_nr &= 0xffff ;
 1136 
 1137     DUMMYNET_LOCK();
 1138     /*
 1139      * This is a dummynet rule, so we expect an O_PIPE or O_QUEUE rule.
 1140      */
 1141     fs = locate_flowset(pipe_nr, fwa->rule);
 1142     if (fs == NULL)
 1143         goto dropit ;   /* this queue/pipe does not exist! */
 1144     pipe = fs->pipe ;
 1145     if (pipe == NULL) { /* must be a queue, try find a matching pipe */
 1146         for (pipe = all_pipes; pipe && pipe->pipe_nr != fs->parent_nr;
 1147                  pipe = pipe->next)
 1148             ;
 1149         if (pipe != NULL)
 1150             fs->pipe = pipe ;
 1151         else {
 1152             printf("dummynet: no pipe %d for queue %d, drop pkt\n",
 1153                 fs->parent_nr, fs->fs_nr);
 1154             goto dropit ;
 1155         }
 1156     }
 1157     q = find_queue(fs, &(fwa->f_id));
 1158     if ( q == NULL )
 1159         goto dropit ;           /* cannot allocate queue                */
 1160     /*
 1161      * update statistics, then check reasons to drop pkt
 1162      */
 1163     q->tot_bytes += len ;
 1164     q->tot_pkts++ ;
 1165     if ( fs->plr && random() < fs->plr )
 1166         goto dropit ;           /* random pkt drop                      */
 1167     if ( fs->flags_fs & DN_QSIZE_IS_BYTES) {
 1168         if (q->len_bytes > fs->qsize)
 1169             goto dropit ;       /* queue size overflow                  */
 1170     } else {
 1171         if (q->len >= fs->qsize)
 1172             goto dropit ;       /* queue count overflow                 */
 1173     }
 1174     if ( fs->flags_fs & DN_IS_RED && red_drops(fs, q, len) )
 1175         goto dropit ;
 1176 
 1177     /* XXX expensive to zero, see if we can remove it*/
 1178     pkt = (struct dn_pkt *)malloc(sizeof (*pkt), M_DUMMYNET, M_NOWAIT|M_ZERO);
 1179     if ( pkt == NULL )
 1180         goto dropit ;           /* cannot allocate packet header        */
 1181     /* ok, i can handle the pkt now... */
 1182     /* build and enqueue packet + parameters */
 1183     pkt->hdr.mh_type = MT_TAG;
 1184     pkt->hdr.mh_flags = PACKET_TAG_DUMMYNET;
 1185     pkt->rule = fwa->rule ;
 1186     DN_NEXT(pkt) = NULL;
 1187     pkt->dn_m = m;
 1188     pkt->dn_dir = dir ;
 1189 
 1190     pkt->ifp = fwa->oif;
 1191     if (dir == DN_TO_IP_OUT) {
 1192         /*
 1193          * We need to copy *ro because for ICMP pkts (and maybe others)
 1194          * the caller passed a pointer into the stack; dst might also be
 1195          * a pointer into *ro so it needs to be updated.
 1196          */
 1197         pkt->ro = *(fwa->ro);
 1198         if (pkt->ro.ro_rt) {
 1199             RT_LOCK(pkt->ro.ro_rt);
 1200             RT_ADDREF(pkt->ro.ro_rt) ;
 1201             RT_UNLOCK(pkt->ro.ro_rt);
 1202         }
 1203         if (fwa->dst == (struct sockaddr_in *)&fwa->ro->ro_dst) /* dst points into ro */
 1204             fwa->dst = (struct sockaddr_in *)&(pkt->ro.ro_dst) ;
 1205         pkt->dn_dst = fwa->dst;
 1206         pkt->flags = fwa->flags;
 1207     }
 1208     if (q->head == NULL)
 1209         q->head = pkt;
 1210     else
 1211         DN_NEXT(q->tail) = pkt;
 1212     q->tail = pkt;
 1213     q->len++;
 1214     q->len_bytes += len ;
 1215 
 1216     if ( q->head != pkt )       /* flow was not idle, we are done */
 1217         goto done;
 1218     /*
 1219      * If we reach this point the flow was previously idle, so we need
 1220      * to schedule it. This involves different actions for fixed-rate or
 1221      * WF2Q queues.
 1222      */
 1223     if (is_pipe) {
 1224         /*
 1225          * Fixed-rate queue: just insert into the ready_heap.
 1226          */
 1227         dn_key t = 0 ;
 1228         if (pipe->bandwidth)
 1229             t = SET_TICKS(pkt, q, pipe);
 1230         q->sched_time = curr_time ;
 1231         if (t == 0)     /* must process it now */
 1232             ready_event( q );
 1233         else
 1234             heap_insert(&ready_heap, curr_time + t , q );
 1235     } else {
 1236         /*
 1237          * WF2Q. First, compute start time S: if the flow was idle (S=F+1)
 1238          * set S to the virtual time V for the controlling pipe, and update
 1239          * the sum of weights for the pipe; otherwise, remove flow from
 1240          * idle_heap and set S to max(F,V).
 1241          * Second, compute finish time F = S + len/weight.
 1242          * Third, if pipe was idle, update V=max(S, V).
 1243          * Fourth, count one more backlogged flow.
 1244          */
 1245         if (DN_KEY_GT(q->S, q->F)) { /* means timestamps are invalid */
 1246             q->S = pipe->V ;
 1247             pipe->sum += fs->weight ; /* add weight of new queue */
 1248         } else {
 1249             heap_extract(&(pipe->idle_heap), q);
 1250             q->S = MAX64(q->F, pipe->V ) ;
 1251         }
 1252         q->F = q->S + ( len<<MY_M )/(u_int64_t) fs->weight;
 1253 
 1254         if (pipe->not_eligible_heap.elements == 0 &&
 1255                 pipe->scheduler_heap.elements == 0)
 1256             pipe->V = MAX64 ( q->S, pipe->V );
 1257         fs->backlogged++ ;
 1258         /*
 1259          * Look at eligibility. A flow is not eligibile if S>V (when
 1260          * this happens, it means that there is some other flow already
 1261          * scheduled for the same pipe, so the scheduler_heap cannot be
 1262          * empty). If the flow is not eligible we just store it in the
 1263          * not_eligible_heap. Otherwise, we store in the scheduler_heap
 1264          * and possibly invoke ready_event_wfq() right now if there is
 1265          * leftover credit.
 1266          * Note that for all flows in scheduler_heap (SCH), S_i <= V,
 1267          * and for all flows in not_eligible_heap (NEH), S_i > V .
 1268          * So when we need to compute max( V, min(S_i) ) forall i in SCH+NEH,
 1269          * we only need to look into NEH.
 1270          */
 1271         if (DN_KEY_GT(q->S, pipe->V) ) { /* not eligible */
 1272             if (pipe->scheduler_heap.elements == 0)
 1273                 printf("dummynet: ++ ouch! not eligible but empty scheduler!\n");
 1274             heap_insert(&(pipe->not_eligible_heap), q->S, q);
 1275         } else {
 1276             heap_insert(&(pipe->scheduler_heap), q->F, q);
 1277             if (pipe->numbytes >= 0) { /* pipe is idle */
 1278                 if (pipe->scheduler_heap.elements != 1)
 1279                     printf("dummynet: OUCH! pipe should have been idle!\n");
 1280                 DPRINTF(("dummynet: waking up pipe %d at %d\n",
 1281                         pipe->pipe_nr, (int)(q->F >> MY_M)));
 1282                 pipe->sched_time = curr_time ;
 1283                 ready_event_wfq(pipe);
 1284             }
 1285         }
 1286     }
 1287 done:
 1288     DUMMYNET_UNLOCK();
 1289     return 0;
 1290 
 1291 dropit:
 1292     if (q)
 1293         q->drops++ ;
 1294     DUMMYNET_UNLOCK();
 1295     m_freem(m);
 1296     return ( (fs && (fs->flags_fs & DN_NOERROR)) ? 0 : ENOBUFS);
 1297 }
 1298 
 1299 /*
 1300  * Below, the rt_unref is only needed when (pkt->dn_dir == DN_TO_IP_OUT)
 1301  * Doing this would probably save us the initial bzero of dn_pkt
 1302  */
 1303 #define DN_FREE_PKT(pkt)        {               \
 1304         struct dn_pkt *n = pkt ;                \
 1305         rt_unref ( n->ro.ro_rt, __func__ ) ;    \
 1306         m_freem(n->dn_m);                       \
 1307         pkt = DN_NEXT(n) ;                      \
 1308         free(n, M_DUMMYNET) ;   }
 1309 
 1310 /*
 1311  * Dispose all packets and flow_queues on a flow_set.
 1312  * If all=1, also remove red lookup table and other storage,
 1313  * including the descriptor itself.
 1314  * For the one in dn_pipe MUST also cleanup ready_heap...
 1315  */
 1316 static void
 1317 purge_flow_set(struct dn_flow_set *fs, int all)
 1318 {
 1319     struct dn_pkt *pkt ;
 1320     struct dn_flow_queue *q, *qn ;
 1321     int i ;
 1322 
 1323     DUMMYNET_LOCK_ASSERT();
 1324 
 1325     for (i = 0 ; i <= fs->rq_size ; i++ ) {
 1326         for (q = fs->rq[i] ; q ; q = qn ) {
 1327             for (pkt = q->head ; pkt ; )
 1328                 DN_FREE_PKT(pkt) ;
 1329             qn = q->next ;
 1330             free(q, M_DUMMYNET);
 1331         }
 1332         fs->rq[i] = NULL ;
 1333     }
 1334     fs->rq_elements = 0 ;
 1335     if (all) {
 1336         /* RED - free lookup table */
 1337         if (fs->w_q_lookup)
 1338             free(fs->w_q_lookup, M_DUMMYNET);
 1339         if (fs->rq)
 1340             free(fs->rq, M_DUMMYNET);
 1341         /* if this fs is not part of a pipe, free it */
 1342         if (fs->pipe && fs != &(fs->pipe->fs) )
 1343             free(fs, M_DUMMYNET);
 1344     }
 1345 }
 1346 
 1347 /*
 1348  * Dispose all packets queued on a pipe (not a flow_set).
 1349  * Also free all resources associated to a pipe, which is about
 1350  * to be deleted.
 1351  */
 1352 static void
 1353 purge_pipe(struct dn_pipe *pipe)
 1354 {
 1355     struct dn_pkt *pkt ;
 1356 
 1357     purge_flow_set( &(pipe->fs), 1 );
 1358 
 1359     for (pkt = pipe->head ; pkt ; )
 1360         DN_FREE_PKT(pkt) ;
 1361 
 1362     heap_free( &(pipe->scheduler_heap) );
 1363     heap_free( &(pipe->not_eligible_heap) );
 1364     heap_free( &(pipe->idle_heap) );
 1365 }
 1366 
 1367 /*
 1368  * Delete all pipes and heaps returning memory. Must also
 1369  * remove references from all ipfw rules to all pipes.
 1370  */
 1371 static void
 1372 dummynet_flush()
 1373 {
 1374     struct dn_pipe *curr_p, *p ;
 1375     struct dn_flow_set *fs, *curr_fs;
 1376 
 1377     DUMMYNET_LOCK();
 1378     /* remove all references to pipes ...*/
 1379     flush_pipe_ptrs(NULL);
 1380     /* prevent future matches... */
 1381     p = all_pipes ;
 1382     all_pipes = NULL ;
 1383     fs = all_flow_sets ;
 1384     all_flow_sets = NULL ;
 1385     /* and free heaps so we don't have unwanted events */
 1386     heap_free(&ready_heap);
 1387     heap_free(&wfq_ready_heap);
 1388     heap_free(&extract_heap);
 1389 
 1390     /*
 1391      * Now purge all queued pkts and delete all pipes
 1392      */
 1393     /* scan and purge all flow_sets. */
 1394     for ( ; fs ; ) {
 1395         curr_fs = fs ;
 1396         fs = fs->next ;
 1397         purge_flow_set(curr_fs, 1);
 1398     }
 1399     for ( ; p ; ) {
 1400         purge_pipe(p);
 1401         curr_p = p ;
 1402         p = p->next ;
 1403         free(curr_p, M_DUMMYNET);
 1404     }
 1405     DUMMYNET_UNLOCK();
 1406 }
 1407 
 1408 
 1409 extern struct ip_fw *ip_fw_default_rule ;
 1410 static void
 1411 dn_rule_delete_fs(struct dn_flow_set *fs, void *r)
 1412 {
 1413     int i ;
 1414     struct dn_flow_queue *q ;
 1415     struct dn_pkt *pkt ;
 1416 
 1417     for (i = 0 ; i <= fs->rq_size ; i++) /* last one is ovflow */
 1418         for (q = fs->rq[i] ; q ; q = q->next )
 1419             for (pkt = q->head ; pkt ; pkt = DN_NEXT(pkt) )
 1420                 if (pkt->rule == r)
 1421                     pkt->rule = ip_fw_default_rule ;
 1422 }
 1423 /*
 1424  * when a firewall rule is deleted, scan all queues and remove the flow-id
 1425  * from packets matching this rule.
 1426  */
 1427 void
 1428 dn_rule_delete(void *r)
 1429 {
 1430     struct dn_pipe *p ;
 1431     struct dn_pkt *pkt ;
 1432     struct dn_flow_set *fs ;
 1433 
 1434     DUMMYNET_LOCK();
 1435     /*
 1436      * If the rule references a queue (dn_flow_set), then scan
 1437      * the flow set, otherwise scan pipes. Should do either, but doing
 1438      * both does not harm.
 1439      */
 1440     for ( fs = all_flow_sets ; fs ; fs = fs->next )
 1441         dn_rule_delete_fs(fs, r);
 1442     for ( p = all_pipes ; p ; p = p->next ) {
 1443         fs = &(p->fs) ;
 1444         dn_rule_delete_fs(fs, r);
 1445         for (pkt = p->head ; pkt ; pkt = DN_NEXT(pkt) )
 1446             if (pkt->rule == r)
 1447                 pkt->rule = ip_fw_default_rule ;
 1448     }
 1449     DUMMYNET_UNLOCK();
 1450 }
 1451 
 1452 /*
 1453  * setup RED parameters
 1454  */
 1455 static int
 1456 config_red(struct dn_flow_set *p, struct dn_flow_set * x)
 1457 {
 1458     int i;
 1459 
 1460     x->w_q = p->w_q;
 1461     x->min_th = SCALE(p->min_th);
 1462     x->max_th = SCALE(p->max_th);
 1463     x->max_p = p->max_p;
 1464 
 1465     x->c_1 = p->max_p / (p->max_th - p->min_th);
 1466     x->c_2 = SCALE_MUL(x->c_1, SCALE(p->min_th));
 1467     if (x->flags_fs & DN_IS_GENTLE_RED) {
 1468         x->c_3 = (SCALE(1) - p->max_p) / p->max_th;
 1469         x->c_4 = (SCALE(1) - 2 * p->max_p);
 1470     }
 1471 
 1472     /* if the lookup table already exist, free and create it again */
 1473     if (x->w_q_lookup) {
 1474         free(x->w_q_lookup, M_DUMMYNET);
 1475         x->w_q_lookup = NULL ;
 1476     }
 1477     if (red_lookup_depth == 0) {
 1478         printf("\ndummynet: net.inet.ip.dummynet.red_lookup_depth must be > 0\n");
 1479         free(x, M_DUMMYNET);
 1480         return EINVAL;
 1481     }
 1482     x->lookup_depth = red_lookup_depth;
 1483     x->w_q_lookup = (u_int *) malloc(x->lookup_depth * sizeof(int),
 1484             M_DUMMYNET, M_NOWAIT);
 1485     if (x->w_q_lookup == NULL) {
 1486         printf("dummynet: sorry, cannot allocate red lookup table\n");
 1487         free(x, M_DUMMYNET);
 1488         return ENOSPC;
 1489     }
 1490 
 1491     /* fill the lookup table with (1 - w_q)^x */
 1492     x->lookup_step = p->lookup_step ;
 1493     x->lookup_weight = p->lookup_weight ;
 1494     x->w_q_lookup[0] = SCALE(1) - x->w_q;
 1495     for (i = 1; i < x->lookup_depth; i++)
 1496         x->w_q_lookup[i] = SCALE_MUL(x->w_q_lookup[i - 1], x->lookup_weight);
 1497     if (red_avg_pkt_size < 1)
 1498         red_avg_pkt_size = 512 ;
 1499     x->avg_pkt_size = red_avg_pkt_size ;
 1500     if (red_max_pkt_size < 1)
 1501         red_max_pkt_size = 1500 ;
 1502     x->max_pkt_size = red_max_pkt_size ;
 1503     return 0 ;
 1504 }
 1505 
 1506 static int
 1507 alloc_hash(struct dn_flow_set *x, struct dn_flow_set *pfs)
 1508 {
 1509     if (x->flags_fs & DN_HAVE_FLOW_MASK) {     /* allocate some slots */
 1510         int l = pfs->rq_size;
 1511 
 1512         if (l == 0)
 1513             l = dn_hash_size;
 1514         if (l < 4)
 1515             l = 4;
 1516         else if (l > DN_MAX_HASH_SIZE)
 1517             l = DN_MAX_HASH_SIZE;
 1518         x->rq_size = l;
 1519     } else                  /* one is enough for null mask */
 1520         x->rq_size = 1;
 1521     x->rq = malloc((1 + x->rq_size) * sizeof(struct dn_flow_queue *),
 1522             M_DUMMYNET, M_NOWAIT | M_ZERO);
 1523     if (x->rq == NULL) {
 1524         printf("dummynet: sorry, cannot allocate queue\n");
 1525         return ENOSPC;
 1526     }
 1527     x->rq_elements = 0;
 1528     return 0 ;
 1529 }
 1530 
 1531 static void
 1532 set_fs_parms(struct dn_flow_set *x, struct dn_flow_set *src)
 1533 {
 1534     x->flags_fs = src->flags_fs;
 1535     x->qsize = src->qsize;
 1536     x->plr = src->plr;
 1537     x->flow_mask = src->flow_mask;
 1538     if (x->flags_fs & DN_QSIZE_IS_BYTES) {
 1539         if (x->qsize > 1024*1024)
 1540             x->qsize = 1024*1024 ;
 1541     } else {
 1542         if (x->qsize == 0)
 1543             x->qsize = 50 ;
 1544         if (x->qsize > 100)
 1545             x->qsize = 50 ;
 1546     }
 1547     /* configuring RED */
 1548     if ( x->flags_fs & DN_IS_RED )
 1549         config_red(src, x) ;    /* XXX should check errors */
 1550 }
 1551 
 1552 /*
 1553  * setup pipe or queue parameters.
 1554  */
 1555 
 1556 static int
 1557 config_pipe(struct dn_pipe *p)
 1558 {
 1559     int i, r;
 1560     struct dn_flow_set *pfs = &(p->fs);
 1561     struct dn_flow_queue *q;
 1562 
 1563     /*
 1564      * The config program passes parameters as follows:
 1565      * bw = bits/second (0 means no limits),
 1566      * delay = ms, must be translated into ticks.
 1567      * qsize = slots/bytes
 1568      */
 1569     p->delay = ( p->delay * hz ) / 1000 ;
 1570     /* We need either a pipe number or a flow_set number */
 1571     if (p->pipe_nr == 0 && pfs->fs_nr == 0)
 1572         return EINVAL ;
 1573     if (p->pipe_nr != 0 && pfs->fs_nr != 0)
 1574         return EINVAL ;
 1575     if (p->pipe_nr != 0) { /* this is a pipe */
 1576         struct dn_pipe *x, *a, *b;
 1577 
 1578         DUMMYNET_LOCK();
 1579         /* locate pipe */
 1580         for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ;
 1581                  a = b , b = b->next) ;
 1582 
 1583         if (b == NULL || b->pipe_nr != p->pipe_nr) { /* new pipe */
 1584             x = malloc(sizeof(struct dn_pipe), M_DUMMYNET, M_NOWAIT | M_ZERO);
 1585             if (x == NULL) {
 1586                 printf("dummynet: no memory for new pipe\n");
 1587                 return ENOSPC;
 1588             }
 1589             x->pipe_nr = p->pipe_nr;
 1590             x->fs.pipe = x ;
 1591             /* idle_heap is the only one from which we extract from the middle.
 1592              */
 1593             x->idle_heap.size = x->idle_heap.elements = 0 ;
 1594             x->idle_heap.offset=OFFSET_OF(struct dn_flow_queue, heap_pos);
 1595         } else {
 1596             x = b;
 1597             /* Flush accumulated credit for all queues */
 1598             for (i = 0; i <= x->fs.rq_size; i++)
 1599                 for (q = x->fs.rq[i]; q; q = q->next)
 1600                     q->numbytes = 0;
 1601         }
 1602 
 1603         x->bandwidth = p->bandwidth ;
 1604         x->numbytes = 0; /* just in case... */
 1605         bcopy(p->if_name, x->if_name, sizeof(p->if_name) );
 1606         x->ifp = NULL ; /* reset interface ptr */
 1607         x->delay = p->delay ;
 1608         set_fs_parms(&(x->fs), pfs);
 1609 
 1610 
 1611         if ( x->fs.rq == NULL ) { /* a new pipe */
 1612             r = alloc_hash(&(x->fs), pfs) ;
 1613             if (r) {
 1614                 DUMMYNET_UNLOCK();
 1615                 free(x, M_DUMMYNET);
 1616                 return r ;
 1617             }
 1618             x->next = b ;
 1619             if (a == NULL)
 1620                 all_pipes = x ;
 1621             else
 1622                 a->next = x ;
 1623         }
 1624         DUMMYNET_UNLOCK();
 1625     } else { /* config queue */
 1626         struct dn_flow_set *x, *a, *b ;
 1627 
 1628         DUMMYNET_LOCK();
 1629         /* locate flow_set */
 1630         for (a=NULL, b=all_flow_sets ; b && b->fs_nr < pfs->fs_nr ;
 1631                  a = b , b = b->next) ;
 1632 
 1633         if (b == NULL || b->fs_nr != pfs->fs_nr) { /* new  */
 1634             if (pfs->parent_nr == 0)    /* need link to a pipe */
 1635                 return EINVAL ;
 1636             x = malloc(sizeof(struct dn_flow_set), M_DUMMYNET, M_NOWAIT|M_ZERO);
 1637             if (x == NULL) {
 1638                 DUMMYNET_UNLOCK();
 1639                 printf("dummynet: no memory for new flow_set\n");
 1640                 return ENOSPC;
 1641             }
 1642             x->fs_nr = pfs->fs_nr;
 1643             x->parent_nr = pfs->parent_nr;
 1644             x->weight = pfs->weight ;
 1645             if (x->weight == 0)
 1646                 x->weight = 1 ;
 1647             else if (x->weight > 100)
 1648                 x->weight = 100 ;
 1649         } else {
 1650             /* Change parent pipe not allowed; must delete and recreate */
 1651             if (pfs->parent_nr != 0 && b->parent_nr != pfs->parent_nr)
 1652                 return EINVAL ;
 1653             x = b;
 1654         }
 1655         set_fs_parms(x, pfs);
 1656 
 1657         if ( x->rq == NULL ) { /* a new flow_set */
 1658             r = alloc_hash(x, pfs) ;
 1659             if (r) {
 1660                 DUMMYNET_UNLOCK();
 1661                 free(x, M_DUMMYNET);
 1662                 return r ;
 1663             }
 1664             x->next = b;
 1665             if (a == NULL)
 1666                 all_flow_sets = x;
 1667             else
 1668                 a->next = x;
 1669         }
 1670         DUMMYNET_UNLOCK();
 1671     }
 1672     return 0 ;
 1673 }
 1674 
 1675 /*
 1676  * Helper function to remove from a heap queues which are linked to
 1677  * a flow_set about to be deleted.
 1678  */
 1679 static void
 1680 fs_remove_from_heap(struct dn_heap *h, struct dn_flow_set *fs)
 1681 {
 1682     int i = 0, found = 0 ;
 1683     for (; i < h->elements ;)
 1684         if ( ((struct dn_flow_queue *)h->p[i].object)->fs == fs) {
 1685             h->elements-- ;
 1686             h->p[i] = h->p[h->elements] ;
 1687             found++ ;
 1688         } else
 1689             i++ ;
 1690     if (found)
 1691         heapify(h);
 1692 }
 1693 
 1694 /*
 1695  * helper function to remove a pipe from a heap (can be there at most once)
 1696  */
 1697 static void
 1698 pipe_remove_from_heap(struct dn_heap *h, struct dn_pipe *p)
 1699 {
 1700     if (h->elements > 0) {
 1701         int i = 0 ;
 1702         for (i=0; i < h->elements ; i++ ) {
 1703             if (h->p[i].object == p) { /* found it */
 1704                 h->elements-- ;
 1705                 h->p[i] = h->p[h->elements] ;
 1706                 heapify(h);
 1707                 break ;
 1708             }
 1709         }
 1710     }
 1711 }
 1712 
 1713 /*
 1714  * drain all queues. Called in case of severe mbuf shortage.
 1715  */
 1716 void
 1717 dummynet_drain()
 1718 {
 1719     struct dn_flow_set *fs;
 1720     struct dn_pipe *p;
 1721     struct dn_pkt *pkt;
 1722 
 1723     DUMMYNET_LOCK_ASSERT();
 1724 
 1725     heap_free(&ready_heap);
 1726     heap_free(&wfq_ready_heap);
 1727     heap_free(&extract_heap);
 1728     /* remove all references to this pipe from flow_sets */
 1729     for (fs = all_flow_sets; fs; fs= fs->next )
 1730         purge_flow_set(fs, 0);
 1731 
 1732     for (p = all_pipes; p; p= p->next ) {
 1733         purge_flow_set(&(p->fs), 0);
 1734         for (pkt = p->head ; pkt ; )
 1735             DN_FREE_PKT(pkt) ;
 1736         p->head = p->tail = NULL ;
 1737     }
 1738 }
 1739 
 1740 /*
 1741  * Fully delete a pipe or a queue, cleaning up associated info.
 1742  */
 1743 static int
 1744 delete_pipe(struct dn_pipe *p)
 1745 {
 1746     if (p->pipe_nr == 0 && p->fs.fs_nr == 0)
 1747         return EINVAL ;
 1748     if (p->pipe_nr != 0 && p->fs.fs_nr != 0)
 1749         return EINVAL ;
 1750     if (p->pipe_nr != 0) { /* this is an old-style pipe */
 1751         struct dn_pipe *a, *b;
 1752         struct dn_flow_set *fs;
 1753 
 1754         DUMMYNET_LOCK();
 1755         /* locate pipe */
 1756         for (a = NULL , b = all_pipes ; b && b->pipe_nr < p->pipe_nr ;
 1757                  a = b , b = b->next) ;
 1758         if (b == NULL || (b->pipe_nr != p->pipe_nr) ) {
 1759             DUMMYNET_UNLOCK();
 1760             return EINVAL ; /* not found */
 1761         }
 1762 
 1763         /* unlink from list of pipes */
 1764         if (a == NULL)
 1765             all_pipes = b->next ;
 1766         else
 1767             a->next = b->next ;
 1768         /* remove references to this pipe from the ip_fw rules. */
 1769         flush_pipe_ptrs(&(b->fs));
 1770 
 1771         /* remove all references to this pipe from flow_sets */
 1772         for (fs = all_flow_sets; fs; fs= fs->next )
 1773             if (fs->pipe == b) {
 1774                 printf("dummynet: ++ ref to pipe %d from fs %d\n",
 1775                         p->pipe_nr, fs->fs_nr);
 1776                 fs->pipe = NULL ;
 1777                 purge_flow_set(fs, 0);
 1778             }
 1779         fs_remove_from_heap(&ready_heap, &(b->fs));
 1780         purge_pipe(b);  /* remove all data associated to this pipe */
 1781         /* remove reference to here from extract_heap and wfq_ready_heap */
 1782         pipe_remove_from_heap(&extract_heap, b);
 1783         pipe_remove_from_heap(&wfq_ready_heap, b);
 1784         DUMMYNET_UNLOCK();
 1785 
 1786         free(b, M_DUMMYNET);
 1787     } else { /* this is a WF2Q queue (dn_flow_set) */
 1788         struct dn_flow_set *a, *b;
 1789 
 1790         DUMMYNET_LOCK();
 1791         /* locate set */
 1792         for (a = NULL, b = all_flow_sets ; b && b->fs_nr < p->fs.fs_nr ;
 1793                  a = b , b = b->next) ;
 1794         if (b == NULL || (b->fs_nr != p->fs.fs_nr) ) {
 1795             DUMMYNET_UNLOCK();
 1796             return EINVAL ; /* not found */
 1797         }
 1798 
 1799         if (a == NULL)
 1800             all_flow_sets = b->next ;
 1801         else
 1802             a->next = b->next ;
 1803         /* remove references to this flow_set from the ip_fw rules. */
 1804         flush_pipe_ptrs(b);
 1805 
 1806         if (b->pipe != NULL) {
 1807             /* Update total weight on parent pipe and cleanup parent heaps */
 1808             b->pipe->sum -= b->weight * b->backlogged ;
 1809             fs_remove_from_heap(&(b->pipe->not_eligible_heap), b);
 1810             fs_remove_from_heap(&(b->pipe->scheduler_heap), b);
 1811 #if 1   /* XXX should i remove from idle_heap as well ? */
 1812             fs_remove_from_heap(&(b->pipe->idle_heap), b);
 1813 #endif
 1814         }
 1815         purge_flow_set(b, 1);
 1816         DUMMYNET_UNLOCK();
 1817     }
 1818     return 0 ;
 1819 }
 1820 
 1821 /*
 1822  * helper function used to copy data from kernel in DUMMYNET_GET
 1823  */
 1824 static char *
 1825 dn_copy_set(struct dn_flow_set *set, char *bp)
 1826 {
 1827     int i, copied = 0 ;
 1828     struct dn_flow_queue *q, *qp = (struct dn_flow_queue *)bp;
 1829 
 1830     DUMMYNET_LOCK_ASSERT();
 1831 
 1832     for (i = 0 ; i <= set->rq_size ; i++)
 1833         for (q = set->rq[i] ; q ; q = q->next, qp++ ) {
 1834             if (q->hash_slot != i)
 1835                 printf("dummynet: ++ at %d: wrong slot (have %d, "
 1836                     "should be %d)\n", copied, q->hash_slot, i);
 1837             if (q->fs != set)
 1838                 printf("dummynet: ++ at %d: wrong fs ptr (have %p, should be %p)\n",
 1839                         i, q->fs, set);
 1840             copied++ ;
 1841             bcopy(q, qp, sizeof( *q ) );
 1842             /* cleanup pointers */
 1843             qp->next = NULL ;
 1844             qp->head = qp->tail = NULL ;
 1845             qp->fs = NULL ;
 1846         }
 1847     if (copied != set->rq_elements)
 1848         printf("dummynet: ++ wrong count, have %d should be %d\n",
 1849             copied, set->rq_elements);
 1850     return (char *)qp ;
 1851 }
 1852 
 1853 static int
 1854 dummynet_get(struct sockopt *sopt)
 1855 {
 1856     char *buf, *bp ; /* bp is the "copy-pointer" */
 1857     size_t size ;
 1858     struct dn_flow_set *set ;
 1859     struct dn_pipe *p ;
 1860     int error=0 ;
 1861 
 1862     /* XXX lock held too long */
 1863     DUMMYNET_LOCK();
 1864     /*
 1865      * compute size of data structures: list of pipes and flow_sets.
 1866      */
 1867     for (p = all_pipes, size = 0 ; p ; p = p->next )
 1868         size += sizeof( *p ) +
 1869             p->fs.rq_elements * sizeof(struct dn_flow_queue);
 1870     for (set = all_flow_sets ; set ; set = set->next )
 1871         size += sizeof ( *set ) +
 1872             set->rq_elements * sizeof(struct dn_flow_queue);
 1873     buf = malloc(size, M_TEMP, M_NOWAIT);
 1874     if (buf == 0) {
 1875         DUMMYNET_UNLOCK();
 1876         return ENOBUFS ;
 1877     }
 1878     for (p = all_pipes, bp = buf ; p ; p = p->next ) {
 1879         struct dn_pipe *pipe_bp = (struct dn_pipe *)bp ;
 1880 
 1881         /*
 1882          * copy pipe descriptor into *bp, convert delay back to ms,
 1883          * then copy the flow_set descriptor(s) one at a time.
 1884          * After each flow_set, copy the queue descriptor it owns.
 1885          */
 1886         bcopy(p, bp, sizeof( *p ) );
 1887         pipe_bp->delay = (pipe_bp->delay * 1000) / hz ;
 1888         /*
 1889          * XXX the following is a hack based on ->next being the
 1890          * first field in dn_pipe and dn_flow_set. The correct
 1891          * solution would be to move the dn_flow_set to the beginning
 1892          * of struct dn_pipe.
 1893          */
 1894         pipe_bp->next = (struct dn_pipe *)DN_IS_PIPE ;
 1895         /* clean pointers */
 1896         pipe_bp->head = pipe_bp->tail = NULL ;
 1897         pipe_bp->fs.next = NULL ;
 1898         pipe_bp->fs.pipe = NULL ;
 1899         pipe_bp->fs.rq = NULL ;
 1900 
 1901         bp += sizeof( *p ) ;
 1902         bp = dn_copy_set( &(p->fs), bp );
 1903     }
 1904     for (set = all_flow_sets ; set ; set = set->next ) {
 1905         struct dn_flow_set *fs_bp = (struct dn_flow_set *)bp ;
 1906         bcopy(set, bp, sizeof( *set ) );
 1907         /* XXX same hack as above */
 1908         fs_bp->next = (struct dn_flow_set *)DN_IS_QUEUE ;
 1909         fs_bp->pipe = NULL ;
 1910         fs_bp->rq = NULL ;
 1911         bp += sizeof( *set ) ;
 1912         bp = dn_copy_set( set, bp );
 1913     }
 1914     DUMMYNET_UNLOCK();
 1915 
 1916     error = sooptcopyout(sopt, buf, size);
 1917     free(buf, M_TEMP);
 1918     return error ;
 1919 }
 1920 
 1921 /*
 1922  * Handler for the various dummynet socket options (get, flush, config, del)
 1923  */
 1924 static int
 1925 ip_dn_ctl(struct sockopt *sopt)
 1926 {
 1927     int error = 0 ;
 1928     struct dn_pipe *p, tmp_pipe;
 1929 
 1930     /* Disallow sets in really-really secure mode. */
 1931     if (sopt->sopt_dir == SOPT_SET) {
 1932 #if __FreeBSD_version >= 500034
 1933         error =  securelevel_ge(sopt->sopt_td->td_ucred, 3);
 1934         if (error)
 1935             return (error);
 1936 #else
 1937         if (securelevel >= 3)
 1938             return (EPERM);
 1939 #endif
 1940     }
 1941 
 1942     switch (sopt->sopt_name) {
 1943     default :
 1944         printf("dummynet: -- unknown option %d", sopt->sopt_name);
 1945         return EINVAL ;
 1946 
 1947     case IP_DUMMYNET_GET :
 1948         error = dummynet_get(sopt);
 1949         break ;
 1950 
 1951     case IP_DUMMYNET_FLUSH :
 1952         dummynet_flush() ;
 1953         break ;
 1954 
 1955     case IP_DUMMYNET_CONFIGURE :
 1956         p = &tmp_pipe ;
 1957         error = sooptcopyin(sopt, p, sizeof *p, sizeof *p);
 1958         if (error)
 1959             break ;
 1960         error = config_pipe(p);
 1961         break ;
 1962 
 1963     case IP_DUMMYNET_DEL :      /* remove a pipe or queue */
 1964         p = &tmp_pipe ;
 1965         error = sooptcopyin(sopt, p, sizeof *p, sizeof *p);
 1966         if (error)
 1967             break ;
 1968 
 1969         error = delete_pipe(p);
 1970         break ;
 1971     }
 1972     return error ;
 1973 }
 1974 
 1975 static void
 1976 ip_dn_init(void)
 1977 {
 1978     if (bootverbose)
 1979             printf("DUMMYNET initialized (011031)\n");
 1980 
 1981     DUMMYNET_LOCK_INIT();
 1982 
 1983     all_pipes = NULL ;
 1984     all_flow_sets = NULL ;
 1985     ready_heap.size = ready_heap.elements = 0 ;
 1986     ready_heap.offset = 0 ;
 1987 
 1988     wfq_ready_heap.size = wfq_ready_heap.elements = 0 ;
 1989     wfq_ready_heap.offset = 0 ;
 1990 
 1991     extract_heap.size = extract_heap.elements = 0 ;
 1992     extract_heap.offset = 0 ;
 1993 
 1994     ip_dn_ctl_ptr = ip_dn_ctl;
 1995     ip_dn_io_ptr = dummynet_io;
 1996     ip_dn_ruledel_ptr = dn_rule_delete;
 1997 
 1998     callout_init(&dn_timeout, debug_mpsafenet ? CALLOUT_MPSAFE : 0);
 1999     callout_reset(&dn_timeout, 1, dummynet, NULL);
 2000 }
 2001 
 2002 #ifdef KLD_MODULE
 2003 static void
 2004 ip_dn_destroy(void)
 2005 {
 2006     ip_dn_ctl_ptr = NULL;
 2007     ip_dn_io_ptr = NULL;
 2008     ip_dn_ruledel_ptr = NULL;
 2009 
 2010     callout_stop(&dn_timeout);
 2011     dummynet_flush();
 2012 
 2013     DUMMYNET_LOCK_DESTROY();
 2014 }
 2015 #endif /* KLD_MODULE */
 2016 
 2017 static int
 2018 dummynet_modevent(module_t mod, int type, void *data)
 2019 {
 2020         switch (type) {
 2021         case MOD_LOAD:
 2022                 if (DUMMYNET_LOADED) {
 2023                     printf("DUMMYNET already loaded\n");
 2024                     return EEXIST ;
 2025                 }
 2026                 ip_dn_init();
 2027                 break;
 2028 
 2029         case MOD_UNLOAD:
 2030 #if !defined(KLD_MODULE)
 2031                 printf("dummynet statically compiled, cannot unload\n");
 2032                 return EINVAL ;
 2033 #else
 2034                 ip_dn_destroy();
 2035 #endif
 2036                 break ;
 2037         default:
 2038                 break ;
 2039         }
 2040         return 0 ;
 2041 }
 2042 
 2043 static moduledata_t dummynet_mod = {
 2044         "dummynet",
 2045         dummynet_modevent,
 2046         NULL
 2047 };
 2048 DECLARE_MODULE(dummynet, dummynet_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 2049 MODULE_DEPEND(dummynet, ipfw, 1, 1, 1);
 2050 MODULE_VERSION(dummynet, 1);

Cache object: e4e88bd1b5fd3aabbaea77cc96f216bf


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.