readme

基于组件方式开发操作系统的OSKIT源代码

DPF version 2.0. 

	The current system is a rewrite of the dpf system described in
Engler & Kaashoek, SIGCOMM 96.  For simplicity, its language has been
pared to resemble that of pathfinder (Bailey et al OSDI 95).  There is
no support for fragmentation.  If the language is too spare,  or you
need fragmentation (or the implementation is slow on some style of
usage you care about) please email [e-mail].  In particular
this version of dpf has been optimized for insertion overhead rather
than demux speed (the current system demux's about 2x slower than the
system in the paper).   There are modifications that can be done to 
speed this up, but I've been deferring until there is a
stated need: if you care about demux speed let me know and I'll fix
it.  More details can be found in the OVERVIEW file.  The official
DPF url is:
		http://www.pdos.lcs.mit.edu/~engler/dpf.html

-------------------------------------------------------------
To build the system:
	get dpf
	untar in its own directory

	get vcode
		http://www.pdos.lcs.mit.edu/~engler/latest-vcode-release.tar.Z
	make the system and symbolically link the vcode source directory
		(vcode-src) as 'vcode'' in dpf source directory:
			ln -s ``path-to-vcode''/vcode-src vcode

	make clean
	make depend
	make

To test, cd into the tests directory and type:
	make
	dpf  ( will spit out a bunch text; should not assert fail)
	driver -f 100 -d 100 (should not assert fail)


-----------------------------------------------------------
Using the system


The interface to construct a filter is in dpf.h; example uses are
in dpf-test.c and in xlate.c (xlate.c translates old dpf filters
to new ones).  

Public interface:
	int dpf_insert(struct dpf_ir *filter);
		install filter in the current dpf trie; returns the
		filter id or < 0 on error.

	int dpf_delete(unsigned pid);
		delete filter pid from filter trie.  Returns < 0 on
		failure.

	int (*dpf_iptr)(unsigned char *msg, unsigned nbytes);
		demult message (msg, of length nbytes); returns
		claiming filter id or 0 if all filters reject.

A simple filter:
        /* 
         * Create filter to check that the 8th byte in msg is 16 and
         * the short at offset 10 is 0x1234.
         */
	#include <dpf.h>
	#include <demand.h>
        int main(void) {
                struct dpf_ir filter;  /* struct to hold filter. */
                uint8 msg[128];
                int pid, res;

                dpf_init(&filter);  /* must be done before use */
                dpf_eq8(&filter, 8, 16);
                dpf_eq16(&filter, 10, 0x1234);

                /* insert filter */
                if((pid = dpf_insert(&filter)) < 0)
                        fatal(Insertion failed);

		/* print out disassembled code. */
                dpf_output();

                /* create message */
                *(uint8 *) ((uint8 *)msg + 8) = 16;
                *(uint16 *) ((uint8 *)msg + 10) = 0x1234;

                /* test it */
                if((res = dpf_iptr(msg, sizeof msg)) != pid) {
                        printf("pid  = %d, res = %d\n", pid, res );
                        fatal(bogus classification);
                }

                /* delete filter */
                if(dpf_delete(pid) < 0)
                        fatal(Delete failed);

		return 0;
        }       


NOTE:
	The current system has debugging turned on.  For decent speed,
	uncomment out the optimization flags (-O3 -DNDEBUG) in the
	dpf makefile.

To build a standalone system for testing filters, simply include
dpf.h in your application and link against dpf.lib.a.

Including DPF in a kernel is similar (note that it requires qsort, malloc
and free).

An example use of dpf to demux a message:
	if((filter_id = dpf_iptr(msg, nbytes)))
		/* matched */
		give message to connection filter_id for processing.
	else
		/* did not match */
		discard message.

Useful site-specific configuration values for minimum message size,
maximum number of active filters, and other goodies are defined in
dpf-config.h

Limitations:
	+ The test suite assumes little-endian (MIPS, Alpha).  
	+ The new version of DPF requires some vcode modifications 
		(currently supported on only MIPS and Alpha).  
If you need other architectures email me.

Since this is a preliminary release I request that you please notify us
if you are making a comparision to other systems.  The various
performance bugs in this release are readily fixed.