Commit 6410d3a5 authored by Pramod R Sanaga's avatar Pramod R Sanaga

Changes to Dummynet in FBSD-410 and its corresponding delay-agent. These modifications

add the BACKFILL parameter to the default delay node kernel. The new Image ID is FBSD410-DEL-DNETMOD.
parent dd6fec7a
Modifications to Dummynet & delay-agent to add Backfill:
The files in the current directory (ip_dummynet.c/.h & dummynet_callback.c) are for
the delay node with FBSD-5.4 ( FBSD54-DNODE-PUBSUB image ).
Compilation instrctions:
a) Kernel:
1) Start with the FreeBSD source of the FBSD54-DNODE-PUBSUB image. Copy the ip_dummynet.h file
to /usr/include/netinet/ip_dummynet.h.
2) Copy ip_dummynet.h and ip_dummynet.c to kernel_src/sys/netinet/
3) Compile the kernel.
4) The delay node Image ID with these modifications is FBSD54-DNODE-DNETMOD.
b) Delay agent:
After step(1) above, overwrite the callback.c in the delay-agent cvs source with the file
dummynet_callback.c & run make.
Steps to build a FBSD-4.10 Delay kernel with Backfill:
The required files are in the fbsd410/kernel & fbsd410/delay-agent directories. For the kernel
follow the same steps as above(starting with the source of FBSD410-STD).
For the delay agent, overwrite the cvs files main.h & callback.c
of the cvs source with the modified versions and run make.
The delay node Image ID with these modifications is FBSD410-DEL-DNETMOD.
Pramod ( )
* Copyright (c) 2000-2003 University of Utah and the Flux Group.
* All rights reserved.
* agent-main.h --
* Delay node agent data structures
#ifndef __AGENT_MAIN_H
#define __AGENT_MAIN_H
/* for setsockopt and stuff */
#include <sys/param.h>
#include <sys/mbuf.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/time.h>
#include <sys/wait.h>
#include <paths.h>
#include <ctype.h>
#include <err.h>
#include <errno.h>
#include <grp.h>
#include <limits.h>
#include <netdb.h>
#include <pwd.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <string.h>
#include <unistd.h>
#include <sysexits.h>
#include <net/if.h>
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/ip_icmp.h>
#include <netinet/ip_fw.h>
#include <net/route.h> /* def. of struct route */
#include <netinet/ip_dummynet.h>
#include <netinet/tcp.h>
#include <arpa/inet.h>
/* for setsockopt and stuff */
#include <stdio.h>
#include <unistd.h>
#include "event.h"
#include "tbdefs.h"
#include "log.h"
#if 0
#include "tbdb.h"
#if 0
/* these are defined in testbed/lib/libtb/tbdb.h, will include
that directory later and then throw out these defines
#define MAX_LINE_LENGTH 512
#define MAX_LINKS 256 /* Virtual interfaces */
/*************************USER DEFINED TYPES********************************/
/* flag bits which indicate pipe parameters*/
#define PIPE_QSIZE_IN_BYTES 0x0001
#define PIPE_Q_IS_RED 0x0002
#define PIPE_Q_IS_GRED 0x0004
#define PIPE_HAS_FLOW_MASK 0x0008
/* Parameters for RED and gentle RED*/
typedef struct {
/* from ip_dummynet.h*/
double w_q ; /* queue weight */
int max_th ; /* maximum threshold for queue (scaled) */
int min_th ; /* minimum threshold for queue (scaled) */
double max_p ; /* maximum value for p_b */
} structRed_params;
/* link status*/
typedef enum {
LINK_UP = 0,
/* Pipe parameter structures*/
typedef struct {
int delay; /* pipe delay*/
int bw; /* pipe bw*/
int backfill; /*pramod-CHANGES, -add backfill to the pipe*/
double plr; /* queue loss rate*/
int q_size; /* queuq size in slots/bytes*/
structRed_params red_gred_params; /* red/gred params*/
struct ipfw_flow_id id ; /* flow mask of the pipe*/
int buckets; /* number of buckets*/
int n_qs; /* number of dynamic queues */
u_short flags_p;
}structpipe_params, *structpipe_params_t;
/* This structure maps the linkname (eg. link0 ) to the physical
interfaces(eg. fxp0) and to the pipe numbers created at the start of
the experiment.
Later as events are recd. this table tells which pipe configuration should
be modified
typedef struct {
char *linkname; /*link0, link1 etc*/
int islan; /* 1 if a lan, 0 if a duplex link */
int numpipes; /* 1 if a simplex pipe, 2 if a duplex pipe */
char *interfaces[2];/* fxp0, fxp1 etc*/
char *vnodes[2]; /* nodeA, nodeB*/
char linkvnodes[2][256]; /* link0-nodeA, link0-nodeB*/
char *linktype; /*simplex, duplex */
int pipes[2]; /* array of pipe numbers*/
structpipe_params params[2]; /* params for the two pipes*/
enumlinkstat stat; /* link status : UP/DOWN*/
}structlink_map, * structlink_map_t;
/*************************USER DEFINED TYPES********************************/
/******************************Function prototypes******************************/
void usage(char *);
void fill_tuple(address_tuple_t);
void agent_callback(event_handle_t handle,
event_notification_t notification, void *data);
void handle_pipes (char *objname, char *eventtype, event_notification_t
,event_handle_t, int);
int checkevent (char *);
void handle_link_up(char * linkname, int l_index);
void handle_link_down(char * linkname, int l_index);
void handle_link_modify(char * linkname, int l_index,
event_handle_t handle,
event_notification_t notification);
int get_link_params(int l_index);
void get_flowset_params(struct dn_flow_set*, int, int);
void get_queue_params(struct dn_flow_set*,int, int);
void set_link_params(int l_index, int blackhole, int);
int get_new_link_params(int l_index, event_handle_t handle,
event_notification_t notification,
int *);
void dump_link_map();
int get_link_info();
/******************************Function prototypes******************************/
#endif /*__AGENT_MAIN_H*/
* Copyright (c) 1998-2002 Luigi Rizzo, Universita` di Pisa
* Portions Copyright (c) 2000 Akamba Corp.
* All rights reserved
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* $FreeBSD: src/sys/netinet/ip_dummynet.h,v 2003/05/13 09:31:06 maxim Exp $
#ifndef _IP_DUMMYNET_H
#define _IP_DUMMYNET_H
* Definition of dummynet data structures. In the structures, I decided
* not to use the macros in <sys/queue.h> in the hope of making the code
* easier to port to other architectures. The type of lists and queue we
* use here is pretty simple anyways.
* We start with a heap, which is used in the scheduler to decide when
* to transmit packets etc.
* The key for the heap is used for two different values:
* 1. timer ticks- max 10K/second, so 32 bits are enough;
* 2. virtual times. These increase in steps of len/x, where len is the
* packet length, and x is either the weight of the flow, or the
* sum of all weights.
* If we limit to max 1000 flows and a max weight of 100, then
* x needs 17 bits. The packet size is 16 bits, so we can easily
* overflow if we do not allow errors.
* So we use a key "dn_key" which is 64 bits. Some macros are used to
* compare key values and handle wraparounds.
* MAX64 returns the largest of two key values.
* MY_M is used as a shift count when doing fixed point arithmetic
* (a better name would be useful...).
typedef u_int64_t dn_key ; /* sorting key */
#define DN_KEY_LT(a,b) ((int64_t)((a)-(b)) < 0)
#define DN_KEY_LEQ(a,b) ((int64_t)((a)-(b)) <= 0)
#define DN_KEY_GT(a,b) ((int64_t)((a)-(b)) > 0)
#define DN_KEY_GEQ(a,b) ((int64_t)((a)-(b)) >= 0)
#define MAX64(x,y) (( (int64_t) ( (y)-(x) )) > 0 ) ? (y) : (x)
#define MY_M 16 /* number of left shift to obtain a larger precision */
* XXX With this scaling, max 1000 flows, max weight 100, 1Gbit/s, the
* virtual time wraps every 15 days.
* The OFFSET_OF macro is used to return the offset of a field within
* a structure. It is used by the heap management routines.
#define OFFSET_OF(type, field) ((int)&( ((type *)0)->field) )
* The maximum hash table size for queues. This value must be a power
* of 2.
#define DN_MAX_HASH_SIZE 65536
* A heap entry is made of a key and a pointer to the actual
* object stored in the heap.
* The heap is an array of dn_heap_entry entries, dynamically allocated.
* Current size is "size", with "elements" actually in use.
* The heap normally supports only ordered insert and extract from the top.
* If we want to extract an object from the middle of the heap, we
* have to know where the object itself is located in the heap (or we
* need to scan the whole array). To this purpose, an object has a
* field (int) which contains the index of the object itself into the
* heap. When the object is moved, the field must also be updated.
* The offset of the index in the object is stored in the 'offset'
* field in the heap descriptor. The assumption is that this offset
* is non-zero if we want to support extract from the middle.
struct dn_heap_entry {
dn_key key ; /* sorting key. Topmost element is smallest one */
void *object ; /* object pointer */
} ;
struct dn_heap {
int size ;
int elements ;
int offset ; /* XXX if > 0 this is the offset of direct ptr to obj */
struct dn_heap_entry *p ; /* really an array of "size" entries */
} ;
* struct dn_pkt identifies a packet in the dummynet queue, but
* is also used to tag packets passed back to the various destinations
* (ip_input(), ip_output(), bdg_forward() and so on).
* As such the first part of the structure must be a struct m_hdr,
* followed by dummynet-specific parameters. The m_hdr must be
* initialized with
* mh_type = MT_TAG;
* mh_next = <pointer to the actual mbuf>
* mh_nextpkt, mh_data are free for dummynet use (mh_nextpkt is used to
* build a linked list of packets in a dummynet queue).
struct dn_pkt {
struct m_hdr hdr ;
#define DN_NEXT(x) (struct dn_pkt *)(x)->hdr.mh_nextpkt
#define dn_m hdr.mh_next /* packet to be forwarded */
struct ip_fw *rule; /* matching rule */
int dn_dir; /* action when packet comes out. */
#define DN_TO_IP_OUT 1
#define DN_TO_IP_IN 2
#define DN_TO_BDG_FWD 3
#define DN_TO_ETH_DEMUX 4
#define DN_TO_ETH_OUT 5
dn_key output_time; /* when the pkt is due for delivery */
struct ifnet *ifp; /* interface, for ip_output */
struct sockaddr_in *dn_dst ;
struct route ro; /* route, for ip_output. MUST COPY */
int flags ; /* flags, for ip_output (IPv6 ?) */
/* pramod-CHANGES */
int isDummy;
* Overall structure of dummynet (with WF2Q+):
In dummynet, packets are selected with the firewall rules, and passed
to two different objects: PIPE or QUEUE.
A QUEUE is just a queue with configurable size and queue management
policy. It is also associated with a mask (to discriminate among
different flows), a weight (used to give different shares of the
bandwidth to different flows) and a "pipe", which essentially
supplies the transmit clock for all queues associated with that
A PIPE emulates a fixed-bandwidth link, whose bandwidth is
configurable. The "clock" for a pipe can come from either an
internal timer, or from the transmit interrupt of an interface.
A pipe is also associated with one (or more, if masks are used)
queue, where all packets for that pipe are stored.
The bandwidth available on the pipe is shared by the queues
associated with that pipe (only one in case the packet is sent
to a PIPE) according to the WF2Q+ scheduling algorithm and the
configured weights.
In general, incoming packets are stored in the appropriate queue,
which is then placed into one of a few heaps managed by a scheduler
to decide when the packet should be extracted.
The scheduler (a function called dummynet()) is run at every timer
tick, and grabs queues from the head of the heaps when they are
ready for processing.
There are three data structures definining a pipe and associated queues:
+ dn_pipe, which contains the main configuration parameters related
to delay and bandwidth;
+ dn_flow_set, which contains WF2Q+ configuration, flow
masks, plr and RED configuration;
+ dn_flow_queue, which is the per-flow queue (containing the packets)
Multiple dn_flow_set can be linked to the same pipe, and multiple
dn_flow_queue can be linked to the same dn_flow_set.
All data structures are linked in a linear list which is used for
housekeeping purposes.
During configuration, we create and initialize the dn_flow_set
and dn_pipe structures (a dn_pipe also contains a dn_flow_set).
At runtime: packets are sent to the appropriate dn_flow_set (either
WFQ ones, or the one embedded in the dn_pipe for fixed-rate flows),
which in turn dispatches them to the appropriate dn_flow_queue
(created dynamically according to the masks).
The transmit clock for fixed rate flows (ready_event()) selects the
dn_flow_queue to be used to transmit the next packet. For WF2Q,
wfq_ready_event() extract a pipe which in turn selects the right
flow using a number of heaps defined into the pipe itself.
* per flow queue. This contains the flow identifier, the queue
* of packets, counters, and parameters used to support both RED and
* WF2Q+.
* A dn_flow_queue is created and initialized whenever a packet for
* a new flow arrives.
struct dn_flow_queue {
struct dn_flow_queue *next ;
struct ipfw_flow_id id ;
struct dn_pkt *head, *tail ; /* queue of packets */
u_int len ;
u_int len_bytes ;
u_long numbytes ; /* credit for transmission (dynamic queues) */
u_int64_t tot_pkts ; /* statistics counters */
u_int64_t tot_bytes ;
u_int32_t drops ;
int hash_slot ; /* debugging/diagnostic */
/* RED parameters */
int avg ; /* average queue length est. (scaled) */
int count ; /* arrivals since last RED drop */
int random ; /* random value (scaled) */
u_int32_t q_time ; /* start of queue idle time */
/* WF2Q+ support */
struct dn_flow_set *fs ; /* parent flow set */
int heap_pos ; /* position (index) of struct in heap */
dn_key sched_time ; /* current time when queue enters ready_heap */
dn_key S,F ; /* start time, finish time */
* Setting F < S means the timestamp is invalid. We only need
* to test this when the queue is empty.
} ;
* flow_set descriptor. Contains the "template" parameters for the
* queue configuration, and pointers to the hash table of dn_flow_queue's.
* The hash table is an array of lists -- we identify the slot by
* hashing the flow-id, then scan the list looking for a match.
* The size of the hash table (buckets) is configurable on a per-queue
* basis.
* A dn_flow_set is created whenever a new queue or pipe is created (in the
* latter case, the structure is located inside the struct dn_pipe).
struct dn_flow_set {
struct dn_flow_set *next; /* next flow set in all_flow_sets list */
u_short fs_nr ; /* flow_set number */
u_short flags_fs;
#define DN_HAVE_FLOW_MASK 0x0001
#define DN_IS_RED 0x0002
#define DN_IS_GENTLE_RED 0x0004
#define DN_QSIZE_IS_BYTES 0x0008 /* queue size is measured in bytes */
#define DN_NOERROR 0x0010 /* do not report ENOBUFS on drops */
#define DN_IS_PIPE 0x4000
#define DN_IS_QUEUE 0x8000
struct dn_pipe *pipe ; /* pointer to parent pipe */
u_short parent_nr ; /* parent pipe#, 0 if local to a pipe */
int weight ; /* WFQ queue weight */
int qsize ; /* queue size in slots or bytes */
int plr ; /* pkt loss rate (2^31-1 means 100%) */
struct ipfw_flow_id flow_mask ;
/* hash table of queues onto this flow_set */
int rq_size ; /* number of slots */
int rq_elements ; /* active elements */
struct dn_flow_queue **rq; /* array of rq_size entries */
u_int32_t last_expired ; /* do not expire too frequently */
int backlogged ; /* #active queues for this flowset */
/* RED parameters */
#define SCALE_RED 16
#define SCALE(x) ( (x) << SCALE_RED )
#define SCALE_VAL(x) ( (x) >> SCALE_RED )
#define SCALE_MUL(x,y) ( ( (x) * (y) ) >> SCALE_RED )
int w_q ; /* queue weight (scaled) */
int max_th ; /* maximum threshold for queue (scaled) */
int min_th ; /* minimum threshold for queue (scaled) */
int max_p ; /* maximum value for p_b (scaled) */
u_int c_1 ; /* max_p/(max_th-min_th) (scaled) */
u_int c_2 ; /* max_p*min_th/(max_th-min_th) (scaled) */
u_int c_3 ; /* for GRED, (1-max_p)/max_th (scaled) */
u_int c_4 ; /* for GRED, 1 - 2*max_p (scaled) */
u_int * w_q_lookup ; /* lookup table for computing (1-w_q)^t */
u_int lookup_depth ; /* depth of lookup table */
int lookup_step ; /* granularity inside the lookup table */
int lookup_weight ; /* equal to (1-w_q)^t / (1-w_q)^(t+1) */
int avg_pkt_size ; /* medium packet size */
int max_pkt_size ; /* max packet size */
} ;
* Pipe descriptor. Contains global parameters, delay-line queue,
* and the flow_set used for fixed-rate queues.
* For WF2Q+ support it also has 3 heaps holding dn_flow_queue:
* not_eligible_heap, for queues whose start time is higher
* than the virtual time. Sorted by start time.
* scheduler_heap, for queues eligible for scheduling. Sorted by
* finish time.
* idle_heap, all flows that are idle and can be removed. We
* do that on each tick so we do not slow down too much
* operations during forwarding.
struct dn_pipe { /* a pipe */
struct dn_pipe *next ;
int pipe_nr ; /* number */
int bandwidth; /* really, bytes/tick. */
int delay ; /* really, ticks */
struct dn_pkt *head, *tail ; /* packets in delay line */
/* WF2Q+ */
struct dn_heap scheduler_heap ; /* top extract - key Finish time*/
struct dn_heap not_eligible_heap; /* top extract- key Start time */
struct dn_heap idle_heap ; /* random extract - key Start=Finish time */
dn_key V ; /* virtual time */
int sum; /* sum of weights of all active sessions */
int numbytes; /* bits I can transmit (more or less). */
dn_key sched_time ; /* time pipe was scheduled in ready_heap */
* When the tx clock come from an interface (if_name[0] != '\0'), its name
* is stored below, whereas the ifp is filled when the rule is configured.
char if_name[IFNAMSIZ];
struct ifnet *ifp ;
int ready ; /* set if ifp != NULL and we got a signal from it */
struct dn_flow_set fs ; /* used with fixed-rate flows */
/* pramod - CHANGES: */
long backfill;
int enableBackFill;
int startBackFill;
int outStandingBackFillBits;
dn_key lastFillTick;
#ifdef _KERNEL
typedef int ip_dn_ctl_t(struct sockopt *); /* raw_ip.c */
typedef void ip_dn_ruledel_t(void *); /* ip_fw.c */
typedef int ip_dn_io_t(struct mbuf *m, int pipe_nr, int dir,
struct ip_fw_args *fwa);
extern ip_dn_ctl_t *ip_dn_ctl_ptr;
extern ip_dn_ruledel_t *ip_dn_ruledel_ptr;
extern ip_dn_io_t *ip_dn_io_ptr;
#define DUMMYNET_LOADED (ip_dn_io_ptr != NULL)
#endif /* _IP_DUMMYNET_H */
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment