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.h

Version: -  FREEBSD  -  FREEBSD11  -  FREEBSD10  -  FREEBSD9  -  FREEBSD92  -  FREEBSD91  -  FREEBSD90  -  FREEBSD8  -  FREEBSD82  -  FREEBSD81  -  FREEBSD80  -  FREEBSD7  -  FREEBSD74  -  FREEBSD73  -  FREEBSD72  -  FREEBSD71  -  FREEBSD70  -  FREEBSD6  -  FREEBSD64  -  FREEBSD63  -  FREEBSD62  -  FREEBSD61  -  FREEBSD60  -  FREEBSD5  -  FREEBSD55  -  FREEBSD54  -  FREEBSD53  -  FREEBSD52  -  FREEBSD51  -  FREEBSD50  -  FREEBSD4  -  FREEBSD3  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
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/7.4/sys/netinet/ip_dummynet.h 178486 2008-04-25 10:29:26Z oleg $
   28  */
   29 
   30 #ifndef _IP_DUMMYNET_H
   31 #define _IP_DUMMYNET_H
   32 
   33 /*
   34  * Definition of dummynet data structures. In the structures, I decided
   35  * not to use the macros in <sys/queue.h> in the hope of making the code
   36  * easier to port to other architectures. The type of lists and queue we
   37  * use here is pretty simple anyways.
   38  */
   39 
   40 /*
   41  * We start with a heap, which is used in the scheduler to decide when
   42  * to transmit packets etc.
   43  *
   44  * The key for the heap is used for two different values:
   45  *
   46  * 1. timer ticks- max 10K/second, so 32 bits are enough;
   47  *
   48  * 2. virtual times. These increase in steps of len/x, where len is the
   49  *    packet length, and x is either the weight of the flow, or the
   50  *    sum of all weights.
   51  *    If we limit to max 1000 flows and a max weight of 100, then
   52  *    x needs 17 bits. The packet size is 16 bits, so we can easily
   53  *    overflow if we do not allow errors.
   54  * So we use a key "dn_key" which is 64 bits. Some macros are used to
   55  * compare key values and handle wraparounds.
   56  * MAX64 returns the largest of two key values.
   57  * MY_M is used as a shift count when doing fixed point arithmetic
   58  * (a better name would be useful...).
   59  */
   60 typedef u_int64_t dn_key ;      /* sorting key */
   61 #define DN_KEY_LT(a,b)     ((int64_t)((a)-(b)) < 0)
   62 #define DN_KEY_LEQ(a,b)    ((int64_t)((a)-(b)) <= 0)
   63 #define DN_KEY_GT(a,b)     ((int64_t)((a)-(b)) > 0)
   64 #define DN_KEY_GEQ(a,b)    ((int64_t)((a)-(b)) >= 0)
   65 #define MAX64(x,y)  (( (int64_t) ( (y)-(x) )) > 0 ) ? (y) : (x)
   66 #define MY_M    16 /* number of left shift to obtain a larger precision */
   67 
   68 /*
   69  * XXX With this scaling, max 1000 flows, max weight 100, 1Gbit/s, the
   70  * virtual time wraps every 15 days.
   71  */
   72 
   73 
   74 /*
   75  * The maximum hash table size for queues.  This value must be a power
   76  * of 2.
   77  */
   78 #define DN_MAX_HASH_SIZE 65536
   79 
   80 /*
   81  * A heap entry is made of a key and a pointer to the actual
   82  * object stored in the heap.
   83  * The heap is an array of dn_heap_entry entries, dynamically allocated.
   84  * Current size is "size", with "elements" actually in use.
   85  * The heap normally supports only ordered insert and extract from the top.
   86  * If we want to extract an object from the middle of the heap, we
   87  * have to know where the object itself is located in the heap (or we
   88  * need to scan the whole array). To this purpose, an object has a
   89  * field (int) which contains the index of the object itself into the
   90  * heap. When the object is moved, the field must also be updated.
   91  * The offset of the index in the object is stored in the 'offset'
   92  * field in the heap descriptor. The assumption is that this offset
   93  * is non-zero if we want to support extract from the middle.
   94  */
   95 struct dn_heap_entry {
   96     dn_key key ;        /* sorting key. Topmost element is smallest one */
   97     void *object ;      /* object pointer */
   98 } ;
   99 
  100 struct dn_heap {
  101     int size ;
  102     int elements ;
  103     int offset ; /* XXX if > 0 this is the offset of direct ptr to obj */
  104     struct dn_heap_entry *p ;   /* really an array of "size" entries */
  105 } ;
  106 
  107 #ifdef _KERNEL
  108 /*
  109  * Packets processed by dummynet have an mbuf tag associated with
  110  * them that carries their dummynet state.  This is used within
  111  * the dummynet code as well as outside when checking for special
  112  * processing requirements.
  113  */
  114 struct dn_pkt_tag {
  115     struct ip_fw *rule;         /* matching rule */
  116     int dn_dir;                 /* action when packet comes out. */
  117 #define DN_TO_IP_OUT    1
  118 #define DN_TO_IP_IN     2
  119 /* Obsolete: #define DN_TO_BDG_FWD      3 */
  120 #define DN_TO_ETH_DEMUX 4
  121 #define DN_TO_ETH_OUT   5
  122 #define DN_TO_IP6_IN    6
  123 #define DN_TO_IP6_OUT   7
  124 #define DN_TO_IFB_FWD   8
  125 
  126     dn_key output_time;         /* when the pkt is due for delivery     */
  127     struct ifnet *ifp;          /* interface, for ip_output             */
  128     struct _ip6dn_args ip6opt;  /* XXX ipv6 options                     */
  129 };
  130 #endif /* _KERNEL */
  131 
  132 /*
  133  * Overall structure of dummynet (with WF2Q+):
  134 
  135 In dummynet, packets are selected with the firewall rules, and passed
  136 to two different objects: PIPE or QUEUE.
  137 
  138 A QUEUE is just a queue with configurable size and queue management
  139 policy. It is also associated with a mask (to discriminate among
  140 different flows), a weight (used to give different shares of the
  141 bandwidth to different flows) and a "pipe", which essentially
  142 supplies the transmit clock for all queues associated with that
  143 pipe.
  144 
  145 A PIPE emulates a fixed-bandwidth link, whose bandwidth is
  146 configurable.  The "clock" for a pipe can come from either an
  147 internal timer, or from the transmit interrupt of an interface.
  148 A pipe is also associated with one (or more, if masks are used)
  149 queue, where all packets for that pipe are stored.
  150 
  151 The bandwidth available on the pipe is shared by the queues
  152 associated with that pipe (only one in case the packet is sent
  153 to a PIPE) according to the WF2Q+ scheduling algorithm and the
  154 configured weights.
  155 
  156 In general, incoming packets are stored in the appropriate queue,
  157 which is then placed into one of a few heaps managed by a scheduler
  158 to decide when the packet should be extracted.
  159 The scheduler (a function called dummynet()) is run at every timer
  160 tick, and grabs queues from the head of the heaps when they are
  161 ready for processing.
  162 
  163 There are three data structures definining a pipe and associated queues:
  164 
  165  + dn_pipe, which contains the main configuration parameters related
  166    to delay and bandwidth;
  167  + dn_flow_set, which contains WF2Q+ configuration, flow
  168    masks, plr and RED configuration;
  169  + dn_flow_queue, which is the per-flow queue (containing the packets)
  170 
  171 Multiple dn_flow_set can be linked to the same pipe, and multiple
  172 dn_flow_queue can be linked to the same dn_flow_set.
  173 All data structures are linked in a linear list which is used for
  174 housekeeping purposes.
  175 
  176 During configuration, we create and initialize the dn_flow_set
  177 and dn_pipe structures (a dn_pipe also contains a dn_flow_set).
  178 
  179 At runtime: packets are sent to the appropriate dn_flow_set (either
  180 WFQ ones, or the one embedded in the dn_pipe for fixed-rate flows),
  181 which in turn dispatches them to the appropriate dn_flow_queue
  182 (created dynamically according to the masks).
  183 
  184 The transmit clock for fixed rate flows (ready_event()) selects the
  185 dn_flow_queue to be used to transmit the next packet. For WF2Q,
  186 wfq_ready_event() extract a pipe which in turn selects the right
  187 flow using a number of heaps defined into the pipe itself.
  188 
  189  *
  190  */
  191 
  192 /*
  193  * per flow queue. This contains the flow identifier, the queue
  194  * of packets, counters, and parameters used to support both RED and
  195  * WF2Q+.
  196  *
  197  * A dn_flow_queue is created and initialized whenever a packet for
  198  * a new flow arrives.
  199  */
  200 struct dn_flow_queue {
  201     struct dn_flow_queue *next ;
  202     struct ipfw_flow_id id ;
  203 
  204     struct mbuf *head, *tail ;  /* queue of packets */
  205     u_int len ;
  206     u_int len_bytes ;
  207     u_long numbytes ;           /* credit for transmission (dynamic queues) */
  208 
  209     u_int64_t tot_pkts ;        /* statistics counters  */
  210     u_int64_t tot_bytes ;
  211     u_int32_t drops ;
  212 
  213     int hash_slot ;             /* debugging/diagnostic */
  214 
  215     /* RED parameters */
  216     int avg ;                   /* average queue length est. (scaled) */
  217     int count ;                 /* arrivals since last RED drop */
  218     int random ;                /* random value (scaled) */
  219     u_int32_t q_time ;          /* start of queue idle time */
  220 
  221     /* WF2Q+ support */
  222     struct dn_flow_set *fs ;    /* parent flow set */
  223     int heap_pos ;              /* position (index) of struct in heap */
  224     dn_key sched_time ;         /* current time when queue enters ready_heap */
  225 
  226     dn_key S,F ;                /* start time, finish time */
  227     /*
  228      * Setting F < S means the timestamp is invalid. We only need
  229      * to test this when the queue is empty.
  230      */
  231 } ;
  232 
  233 /*
  234  * flow_set descriptor. Contains the "template" parameters for the
  235  * queue configuration, and pointers to the hash table of dn_flow_queue's.
  236  *
  237  * The hash table is an array of lists -- we identify the slot by
  238  * hashing the flow-id, then scan the list looking for a match.
  239  * The size of the hash table (buckets) is configurable on a per-queue
  240  * basis.
  241  *
  242  * A dn_flow_set is created whenever a new queue or pipe is created (in the
  243  * latter case, the structure is located inside the struct dn_pipe).
  244  */
  245 struct dn_flow_set {
  246     SLIST_ENTRY(dn_flow_set)    next;   /* linked list in a hash slot */
  247 
  248     u_short fs_nr ;             /* flow_set number       */
  249     u_short flags_fs;
  250 #define DN_HAVE_FLOW_MASK       0x0001
  251 #define DN_IS_RED               0x0002
  252 #define DN_IS_GENTLE_RED        0x0004
  253 #define DN_QSIZE_IS_BYTES       0x0008  /* queue size is measured in bytes */
  254 #define DN_NOERROR              0x0010  /* do not report ENOBUFS on drops  */
  255 #define DN_IS_PIPE              0x4000
  256 #define DN_IS_QUEUE             0x8000
  257 
  258     struct dn_pipe *pipe ;      /* pointer to parent pipe */
  259     u_short parent_nr ;         /* parent pipe#, 0 if local to a pipe */
  260 
  261     int weight ;                /* WFQ queue weight */
  262     int qsize ;                 /* queue size in slots or bytes */
  263     int plr ;                   /* pkt loss rate (2^31-1 means 100%) */
  264 
  265     struct ipfw_flow_id flow_mask ;
  266 
  267     /* hash table of queues onto this flow_set */
  268     int rq_size ;               /* number of slots */
  269     int rq_elements ;           /* active elements */
  270     struct dn_flow_queue **rq;  /* array of rq_size entries */
  271 
  272     u_int32_t last_expired ;    /* do not expire too frequently */
  273     int backlogged ;            /* #active queues for this flowset */
  274 
  275         /* RED parameters */
  276 #define SCALE_RED               16
  277 #define SCALE(x)                ( (x) << SCALE_RED )
  278 #define SCALE_VAL(x)            ( (x) >> SCALE_RED )
  279 #define SCALE_MUL(x,y)          ( ( (x) * (y) ) >> SCALE_RED )
  280     int w_q ;                   /* queue weight (scaled) */
  281     int max_th ;                /* maximum threshold for queue (scaled) */
  282     int min_th ;                /* minimum threshold for queue (scaled) */
  283     int max_p ;                 /* maximum value for p_b (scaled) */
  284     u_int c_1 ;                 /* max_p/(max_th-min_th) (scaled) */
  285     u_int c_2 ;                 /* max_p*min_th/(max_th-min_th) (scaled) */
  286     u_int c_3 ;                 /* for GRED, (1-max_p)/max_th (scaled) */
  287     u_int c_4 ;                 /* for GRED, 1 - 2*max_p (scaled) */
  288     u_int * w_q_lookup ;        /* lookup table for computing (1-w_q)^t */
  289     u_int lookup_depth ;        /* depth of lookup table */
  290     int lookup_step ;           /* granularity inside the lookup table */
  291     int lookup_weight ;         /* equal to (1-w_q)^t / (1-w_q)^(t+1) */
  292     int avg_pkt_size ;          /* medium packet size */
  293     int max_pkt_size ;          /* max packet size */
  294 };
  295 SLIST_HEAD(dn_flow_set_head, dn_flow_set);
  296 
  297 /*
  298  * Pipe descriptor. Contains global parameters, delay-line queue,
  299  * and the flow_set used for fixed-rate queues.
  300  *
  301  * For WF2Q+ support it also has 3 heaps holding dn_flow_queue:
  302  *   not_eligible_heap, for queues whose start time is higher
  303  *      than the virtual time. Sorted by start time.
  304  *   scheduler_heap, for queues eligible for scheduling. Sorted by
  305  *      finish time.
  306  *   idle_heap, all flows that are idle and can be removed. We
  307  *      do that on each tick so we do not slow down too much
  308  *      operations during forwarding.
  309  *
  310  */
  311 struct dn_pipe {                /* a pipe */
  312     SLIST_ENTRY(dn_pipe)        next;   /* linked list in a hash slot */
  313 
  314     int pipe_nr ;               /* number       */
  315     int bandwidth;              /* really, bytes/tick.  */
  316     int delay ;                 /* really, ticks        */
  317 
  318     struct      mbuf *head, *tail ;     /* packets in delay line */
  319 
  320     /* WF2Q+ */
  321     struct dn_heap scheduler_heap ; /* top extract - key Finish time*/
  322     struct dn_heap not_eligible_heap; /* top extract- key Start time */
  323     struct dn_heap idle_heap ; /* random extract - key Start=Finish time */
  324 
  325     dn_key V ;                  /* virtual time */
  326     int sum;                    /* sum of weights of all active sessions */
  327     int numbytes;               /* bits I can transmit (more or less). */
  328 
  329     dn_key sched_time ;         /* time pipe was scheduled in ready_heap */
  330 
  331     /*
  332      * When the tx clock come from an interface (if_name[0] != '\0'), its name
  333      * is stored below, whereas the ifp is filled when the rule is configured.
  334      */
  335     char if_name[IFNAMSIZ];
  336     struct ifnet *ifp ;
  337     int ready ; /* set if ifp != NULL and we got a signal from it */
  338 
  339     struct dn_flow_set fs ; /* used with fixed-rate flows */
  340 };
  341 SLIST_HEAD(dn_pipe_head, dn_pipe);
  342 
  343 #ifdef _KERNEL
  344 typedef int ip_dn_ctl_t(struct sockopt *); /* raw_ip.c */
  345 typedef void ip_dn_ruledel_t(void *); /* ip_fw.c */
  346 typedef int ip_dn_io_t(struct mbuf **m, int dir, struct ip_fw_args *fwa);
  347 extern  ip_dn_ctl_t *ip_dn_ctl_ptr;
  348 extern  ip_dn_ruledel_t *ip_dn_ruledel_ptr;
  349 extern  ip_dn_io_t *ip_dn_io_ptr;
  350 #define DUMMYNET_LOADED (ip_dn_io_ptr != NULL)
  351 
  352 /*
  353  * Return the IPFW rule associated with the dummynet tag; if any.
  354  * Make sure that the dummynet tag is not reused by lower layers.
  355  */
  356 static __inline struct ip_fw *
  357 ip_dn_claim_rule(struct mbuf *m)
  358 {
  359         struct m_tag *mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
  360         if (mtag != NULL) {
  361                 mtag->m_tag_id = PACKET_TAG_NONE;
  362                 return (((struct dn_pkt_tag *)(mtag+1))->rule);
  363         } else
  364                 return (NULL);
  365 }
  366 #endif
  367 #endif /* _IP_DUMMYNET_H */

Cache object: 678d4244980035a61031338708ffe4bf


[ 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.