[uCsimm] Wayward Web Server or is it NFS Cache Crash or is it ???

From: Robert Tagliarini (bobtag@cftnet.com)
Date: Fri Feb 11 2000 - 11:59:44 EST


Hello All,
I'm running linux-2.0.38.1pre1-vag on a motorola ads eval board and I'm using
the userland romdisk from the cd rom. I log in with kermit and use it to crank
up ppp (modified the macro at the end of Vladimir's bbug script for my setup...
Thanks again Vladimir!). I then telnet to the board from another console and
mount an nfs directory on /htdocs (so I can debug some web pages w/o flashing
rom and so I can have some dynamic pages). I then start up a background task
that loops reading an a/d and writing an html file in /htdocs where httpd can
find it (I guess the chroot that httpd does is a good thing). This loop
runs once a second and all works great for a while and I'm excited to see
real data from my target in a web page.

The problem is that after a while (last time was about 15 mins) the system
crashes. The first indication I get of a problem is when normal commands
aren't found in the telnet console. Commands in /htdocs are found but nothing
from anywhere else. It's as if telnet were all of a sudden chroot to htdocs.
After a bit more, the system crashes hard. I can go back to kermit, stop the
ppp redirect, reconnect to the original console and what I see is repeated
trace dumps like the following
Call Trace:
Code: Frame format=3 eff addr=20041000
Stack from 002b0148:
       211e0000 0009002b 10001000 7f8a002b 01e8002b 021c1000 218c002b 017e1000
       21fc002b 017e0000 00001000 3a9a0000 0003002b 017e0000 00000000 0000002b
       10000000 81a4ffff ffffff0f aefb0000 00000000 0000ffff ffff0000 00003035
       ff0faefb 30302004 1000211e 00000009 002b1000 10007f8a 002b0252 002b0286
       1000218c 002b01e8 100021fc 002b01e8 00000000 10003a9a 00000003 002b01e8
       00000000 00000000 002b1000 000081a4 ffffffff ff0faefb 00000000 00000000

addr= is the same in all of them but the stack from address changes...

Is this another example of the NFS cache problem that Vladimir mentioned in his
message of 03 Nov 1999? Where he said:

>"...It is a known bug in NFS client/File caching, that I was asking
>about maybe a month ago, but nobody bothered. what happens is that
>files that you read over NFS are cached in RAM, and for some
>reason this cache just grows indefinitely and is never freed.
 
>Thus, I would say that currently you can never read more than
>about 6-7MB over NFS without crashing the box. No matter whether
>you use httpd or cp.
 
>Try to cp that same file to /dev/null for example.
 
>Jeff, Kenneth! Are you hearing me? ..."

Or is it some other problem perhaps with httpd and chroot?

What's the trace telling me?

Any hints would be greatly appreciated.

Thanks
Bob Tagliarini
This message resent by the ucsimm@uclinux.com list server http://www.uClinux.com/



This archive was generated by hypermail 2b30 : Sun Apr 07 2002 - 00:01:34 EST