kern/72987: ipfw/dummynet pipe/queue 'queue [BYTES]KBytes (red|gred) [FLOAT]/[BYTES]/[BYTES]/[FLOAT]' parameter problem

[John Engelhart]

>Number:         72987
>Category:       kern
>Synopsis:       ipfw/dummynet pipe/queue 'queue [BYTES]KBytes (red|gred) [FLOAT]/[BYTES]/[BYTES]/[FLOAT]' parameter problem
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Fri Oct 22 00:40:29 GMT 2004
>Closed-Date:
>Last-Modified:
>Originator:     John Engelhart
>Release:        5.3-BETA7
>Organization:
>Environment:
FreeBSD new.zang.com 5.3-BETA7 FreeBSD 5.3-BETA7 #0: Sun Oct 17 00:32:33 EDT 2004     johne at new.zang.com:/usr/src/sys/i386/compile/new  i386
>Description:
While testing some ipfw/dummynet configurations, I ran in to a problem with the "queue #KBytes" with either red or gred MIN and MAX specification parameter.

I'm not sure if this is a display only bug, or if the display artifact is due to a more fundamental problem.

In summary, there are four parameters passed to either red or gred, ie "red 1.0/2/3/4.0".  Of concern is parameters 2 and 3.  When the "queue" parameter is specified, it can be specified in one of two ways: in "slots" (packets), or in total number of bytes (#KBytes).  The #2 and #3 RED parameters are either the min/max specified in number of slots, or the min/max specified in number of bytes.  This only happens when the queue parameter is specified in KBytes.

If you specify more than 2^15 as a value for red/gred MIN or MAX, no error is given that a parameter maximum has been exceeded.  However, when checking the pipe configuration with an "ipfw pipe show", the value displayed is not the value specified.  Thus, specifying a red/gred MIN or MAX > 32767 results in a "ipfw pipe show" as if the parameter were a signed 16 bit value, and not the value entered.

I haven't dug in to the code, nor done any exhastive tests, but watching some of the results I got the impression that the values being used by the kernel to do at least the MAX calculations were the truncated values, and not the specified ones.
>How-To-Repeat:
# ipfw pipe 10 config bw 320Kbit/sec queue 512KBytes gred 0.1/4096/131072/0.1
# ipfw pipe 10 show
00010: 320.000 Kbit/s    0 ms  512 KB 1 queues (1 buckets) 
          GRED w_q 0.099991 min_th 4096 max_th 0 max_p 0.099991
    mask: 0x00 0x00000000/0x0000 -> 0x00000000/0x0000

Note the "max_th 0" is not 131072.  Wrapped value:

# ipfw pipe 10 config bw 320Kbit/sec queue 512Kbytes gred 0.1/4096/32768/0.1
# ipfw pipe 10 show
00010: 320.000 Kbit/s    0 ms  512 KB 1 queues (1 buckets) 
          GRED w_q 0.099991 min_th 4096 max_th -32768 max_p 0.099991
    mask: 0x00 0x00000000/0x0000 -> 0x00000000/0x0000
BKT Prot ___Source IP/port____ ____Dest. IP/port____ Tot_pkt/bytes Pkt/Byte Drp

"max_th" now at -32768.
>Fix:
Unknown.

Not sure if this is display only, or a bigger problem.  Wanted to get it logged first, and I figure an ipfw maintainer can make a faster pass at what the real problem is than I can since it's a pretty obvious bug.      
>Release-Note:
>Audit-Trail:
>Unformatted: