Re: [9fans] IDE FS failure

Re: [9fans] IDE FS failure

[Russ Cox]

Well, if you're going to debug it, you want a
simpler test case, like:

	chgrp -u sys /n/yourfs/adm/foo >/n/yourfs/adm/foo

If /adm/users doesn't exist, create it and initialize it,
and then run "users" at the console to reread the file.

Russ

Re: [9fans] IDE FS failure

[Fco.J.Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 193 bytes --]

The allow comand (and an alternate version of fsconfig(8) that
documents it) is available from the idefs tar ball at
http://plan9.escet.urjc.es

Drop me a line if you need help with this.

[-- Attachment #2: Type: message/rfc822, Size: 2513 bytes --]

From: Lucio De Re <lucio@proxima.alt.za>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] IDE FS failure
Date: Fri, 27 Sep 2002 16:09:43 +0200
Message-ID: <20020927160942.A12877@cackle.proxima.alt.za>

On Fri, Sep 27, 2002 at 09:50:30AM -0400, Russ Cox wrote:
>
> You need to enter config mode while the fs is booting
> and type "allow".  That will, for that boot only, allow
> you to change uids.  The uid isn't crucial, so you could
> initialize without setting them (as long as they're getting
> set to glenda or sys rather than none) and not worry about it.
>
Nope, the fileserver was empty and it allowed me to create files.
Allow mode was already on.

The frustrating thing is that (I think) I had it all installed
properly before, but I can't remember how I had got there.  I think
synchronising the users files did the trick.

> I think the allow command is documented somewhere, but
> I can't find it.
>
You're right.  It is missing from fsconfig(8).

++L

Re: [9fans] IDE FS failure

[Lucio De Re]

On Mon, Sep 30, 2002 at 11:48:43AM +0200, Fco.J.Ballesteros wrote:
>
> The allow comand (and an alternate version of fsconfig(8) that
> documents it) is available from the idefs tar ball at
> http://plan9.escet.urjc.es
>
> Drop me a line if you need help with this.

I'm busy moving offices, shouldn't even be sending mail, right now.
But I will get back to the problem in the next day or two.  Internet
access will be an interesting problem.  I have downloaded the latest
CD-ROM image, fortunately.

++L

Re: [9fans] IDE FS failure

[Russ Cox]

You need to enter config mode while the fs is booting
and type "allow".  That will, for that boot only, allow
you to change uids.  The uid isn't crucial, so you could
initialize without setting them (as long as they're getting
set to glenda or sys rather than none) and not worry about it.

I think the allow command is documented somewhere, but
I can't find it.

Russ

Re: [9fans] IDE FS failure

[Lucio De Re]

On Fri, Sep 27, 2002 at 09:50:30AM -0400, Russ Cox wrote:
>
> You need to enter config mode while the fs is booting
> and type "allow".  That will, for that boot only, allow
> you to change uids.  The uid isn't crucial, so you could
> initialize without setting them (as long as they're getting
> set to glenda or sys rather than none) and not worry about it.
>
Nope, the fileserver was empty and it allowed me to create files.
Allow mode was already on.

The frustrating thing is that (I think) I had it all installed
properly before, but I can't remember how I had got there.  I think
synchronising the users files did the trick.

> I think the allow command is documented somewhere, but
> I can't find it.
>
You're right.  It is missing from fsconfig(8).

++L

Re: [9fans] IDE FS failure

[Axel Belinfante]

> I think the allow command is documented somewhere, but
> I can't find it.

Also there is a patch to add ``allow'' to the list of fs commands,
so you don't have to reboot, to ``allow''.
I think it is part of the idefs patches/source at nemo's site.

Axel.

Re: [9fans] IDE FS failure

[Geoff Collyer]

[-- Attachment #1: Type: text/plain, Size: 948 bytes --]

It looks like you're right; my apologies for the bad advice.  I
remembered that all the startsb block numbers had been corrected (to
"2") in 4e but had forgotten about firstsb (and I'd initialised it in
my file servers to "2").  firstsb (if non-zero) is primarily an
optimisation and it depends on the specific jukebox and on time, so in
either /sys/src/fs/emelie/9pcfs.c or /sys/src/fs/sony/9sonyfs.c,

	conf.firstsb = 13219302;

is presumably wrong (I'd bet on 9sonyfs.c being wrong) and could be
deleted or changed to

	conf.firstsb = 0;

Also, /sys/src/fs/words should probably be amended to advise zeroing
firstsb in the spun-off subtree's 9*fs.c.

I noticed just now that conf.wcpsize is not used in 4e; it was used in
3e in /sys/src/fs/port/worm.c in wcpinit(), but worm.c is now gone.
So all those

	conf.wcpsize = 10;

lines in /sys/src/fs/*/9*fs.c and

	port/main.c:	conf.wcpsize = 1024*1024;

can be deleted.

[-- Attachment #2: Type: message/rfc822, Size: 2440 bytes --]

From: David Swasey <swasey@cs.cmu.edu>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] IDE FS failure
Date: Wed, 25 Sep 2002 09:15:11 -0400
Message-ID: <a6e7e85c0618745ab14cb05bbc3c8f91@cs.cmu.edu>

If you can compile a new fs kernel using a stand-alone machine, then
you can probably get past this particular panic.  I believe this
happens because of the line

	conf.firstsb = 13219302;

in your file server's localconfinit.  (This function is part of the
file-server-specific .c file; it may be called 9pcfs.c.)

If firstsb is non-zero, then it is the address of the first super
block to consult when performing a recovery; otherwise, the value in
the startsb array (near the top of 9pcfs.c) is used.

I suggest setting firstsb to 0 and recompiling the file server kernel.
With the new kernel, "recover main" should actually start the recovery
process.

-dave

> On Tue, Sep 24, 2002 at 04:54:03PM -0700, Geoff Collyer wrote:
>>
>> 	recover main
>> 	end
>>
> I get a different error.  I wonder if I didn't configure "archive"
> badly and somehow overlapped it onto main or the cache.  The error is
> now
>
> 	panic: fworm: rbounds 13219302

Re: [9fans] IDE FS failure

[Lucio De Re]

On Wed, Sep 25, 2002 at 03:40:52PM -0700, Geoff Collyer wrote:
>
> is presumably wrong (I'd bet on 9sonyfs.c being wrong) and could be
> deleted or changed to
>
> 	conf.firstsb = 0;
>
I'm sure the problem originated with me using the wrong disk ID
when creating the "archive" filesystem.  I made that mistake again
this morning while rebuilding the fileserver.  I appreciate that
catching such a silly error would be unprofitably hard, so I'm not
complaining.

Quite the contrary, in fact, as my error caused a different problem
to be identified.

Right now, however, I have a new problem I seem unable to resolve.
In attempting to "replica/pull" a CD image (not the most recent
one, sadly) from a stand-alone server, I seem unable to assign the
right ownerships to files (sys:sys, for the most part, but
glenda:glenda also suffers).  This to a freshly reamed FS.

I assume it has something to do with the /adm/users file(s) and
the two servers' (kfs on one end and fs on the other) view of these
files.

But I haven't found a formula that works and the process is too
slow to encourage experimentation.

What would be the right approach?  Should I rather twist the floppy
image to this task (I think I had some trouble when I first tried
it, so I returned to the KFS host) and will I not encounter
inconsistencies there anyway?

Help will be greatly appreciated.  I presume I'm not simply barking
up the wrong tree altogether.

++L

Re: [9fans] IDE FS failure

[David Swasey]

If you can compile a new fs kernel using a stand-alone machine, then
you can probably get past this particular panic.  I believe this
happens because of the line

	conf.firstsb = 13219302;

in your file server's localconfinit.  (This function is part of the
file-server-specific .c file; it may be called 9pcfs.c.)

If firstsb is non-zero, then it is the address of the first super
block to consult when performing a recovery; otherwise, the value in
the startsb array (near the top of 9pcfs.c) is used.

I suggest setting firstsb to 0 and recompiling the file server kernel.
With the new kernel, "recover main" should actually start the recovery
process.

-dave

> On Tue, Sep 24, 2002 at 04:54:03PM -0700, Geoff Collyer wrote:
>>
>> 	recover main
>> 	end
>>
> I get a different error.  I wonder if I didn't configure "archive"
> badly and somehow overlapped it onto main or the cache.  The error is
> now
>
> 	panic: fworm: rbounds 13219302

Re: [9fans] IDE FS failure

[Lucio De Re]

On Wed, Sep 25, 2002 at 09:15:11AM -0400, David Swasey wrote:
>
> If you can compile a new fs kernel using a stand-alone machine, then
> you can probably get past this particular panic.  I believe this
> happens because of the line
>
> 	conf.firstsb = 13219302;
>
> in your file server's localconfinit.  (This function is part of the
> file-server-specific .c file; it may be called 9pcfs.c.)
>
Thank you, I will give it a stab.  Please forgive me if it won't be
for a few days, the end of the month looms and I have some high
priorities to attend to.

++L

Re: [9fans] IDE FS failure

[Geoff Collyer]

The fs kernel's messages don't always identify the file system in
question.  As I touch parts of the fs kernel, I've been adding %Zs to
the prints.  It sounds like the cache of your "main" file system is
corrupt.  You might try a recover of main's cache: reboot the file
server and go into config mode, then type:

	recover main
	end

This should restore your "main" to the way it was at the last dump
(typically 5:00 AM), so you'd lose any changes made since then, but if
"main"'s cache is corrupt, you've effectively lost them already.

Re: [9fans] IDE FS failure

[Lucio De Re]

On Tue, Sep 24, 2002 at 04:54:03PM -0700, Geoff Collyer wrote:
>
> 	recover main
> 	end
>
I get a different error.  I wonder if I didn't configure "archive"
badly and somehow overlapped it onto main or the cache.  The error is
now

	panic: fworm: rbounds 13219302

> This should restore your "main" to the way it was at the last dump
> (typically 5:00 AM), so you'd lose any changes made since then, but if
> "main"'s cache is corrupt, you've effectively lost them already.

It's not serious, if possibly painful to recover.  I'm sure I have
lost some data, but nothing irreproducible.

I'll wire up an RS-232C console in the next day or two, see what I
can detect from the error messages that presently flash off the
display too quickly.

++L

[9fans] IDE FS failure

[Lucio De Re]

4th Edition Plan 9, I get

	panic: cwinit: checktag c bucket

I'm not sure what else might be of interest, I could capture it
with a serial console, I suppose.

I thought it would be the second FS (archive) which I was busy
writing to when the error occurred.  I was building a CVS repository,
but it didn't go very far.

I tried "ream"ing archive from fsconfig, which caused a short delay
after the "end", but the error recurred immediately thereafter.

Any hints appreciated.  The system can be rebuilt (probably will
need to be) but I wouldn't appreciate having to deal with the
problem regularly.

This is running on a P55, with 48Meg or RAM.

++L