[9fans] venti woes with brand new install

[9fans] venti woes with brand new install

[John Soros]

Hi everyone!
I'm trying to configure a new cpu+auth server using the instructions in the wiki, and I've crossed a problem with venti.
Here is what I did:
1, downloaded the sept. 6th iso from plan9 download site
2, installed a fossil+venti system (default setup)
3, followed the wiki instructions to set up a cpu+auth server (compiled my own 9pccpuf kernel with default config)
then I rebooted my computer (luckily set up a menu with options to start the old terminal kernel),
and got venti suicide, here is a screenshot of the crash: http://jsoros.free.fr/CIMG5212.JPG , after this the computer reboots. We did a little debugging with cinap on irc, here is how we did it:
1, copy /386/bin/venti/venti to 9fat
2, boot into the live-cd
3, run /n/9fat/venti -s -c /dev/sdC0/arenas -a 'tcp!127.1!17034' -h 'tcp!127.1!8000'
it suicides here too
4, run acid on the crashed venti, here is the output of acid: http://jsoros.free.fr/CIMG5213.JPG
Thanks for any insite on what could be wrong and what I could do to fix it.

Regards
John

Re: [9fans] venti woes with brand new install

[erik quanstrom]

as a salute to the days of typing in machine code from smalltalk
magazine.  (wasted youth!)  here is the backtrace complete with
errors introduced by a faulty connection between the image and
keyboard.

abort+0x0 /sys/src/libc/9sys/abort.c:6
freezblock(b=0x17d1f90)+0x38 /sys/src/cmd/venti/srv/zblock.c:63
freeifile(f=0xa4918)+0x12 /sys/src/cmd/venti/srv/ifile.c:67
runconfig(file=0xdfffef92,config=0xa496c)+0x8e /sys/src/cmd/venti/srv/config.c:188
	f=0xdfffef92
	ok=0x0
	line=0x0
	s=0x17cff8e
	flds=0x17cff8e
	i=0x2
	sv-0x17cfe68
	av=0x17cfe88
initventi(file=0xdfffef92,conf=0xa496c)+0x3f /sys/src/cmd/venti/srv/config.c:21
threadmain(argv=0xdfffef70,argc=0x0)+0x139) /sys/src/cmd/venti/srv/venti.c:113
	vaddr=0xdfffefa6
	haddr=0xdfffefb9
	configfile=0xdfffef92
	webroot=0x0
	mem=0x0
	icmem=0x0
	bcmem=0x0
	_argc=0xfefe0068
	_args-0x4e474
	config=0x17cfea8
[...]

could you send the output of venti/config? and what does acid
say to this

	acid: *(0x17cff8e\s)

you might also get more information by inspecting how zmagic
is corrupt, e.g. replace freezblock with something like this.
(may or may not compile.)

void
freezblock(ZBlock *b)
{
	if(b){
		if(memcmp(b->data+b->_size, zmagic, OverflowCheck) != 0){
			int i;
			uchar *s;
			char *fmt;

			s = (uchar*)(b->data+b->_size);
			fmt = "  %.2x %.2x %.2x %.2x %.2x %.2x %.2x %.2x\n";

			fprint(2, "corrupt zblock: %p b->_size=%d", b, b->_size);
			for(i = 0; i < OverflowCheck; i += 8, s += 8)
				fprint(2, fmt, s[0], s[1], s[2], s[3], s[4], s[5], s[6], s[7]);
			abort();
		}
		memset(b->data+b->_size, 0, OverflowCheck);
		free(b->free);
	}
}

- erik

> Hi everyone!
> I'm trying to configure a new cpu+auth server using the instructions in the wiki, and I've crossed a problem with venti.
> Here is what I did:
> 1, downloaded the sept. 6th iso from plan9 download site
> 2, installed a fossil+venti system (default setup)
> 3, followed the wiki instructions to set up a cpu+auth server (compiled my own 9pccpuf kernel with default config)
> then I rebooted my computer (luckily set up a menu with options to start the old terminal kernel),
> and got venti suicide, here is a screenshot of the crash: http://jsoros.free.fr/CIMG5212.JPG , after this the computer reboots. We did a little debugging with cinap on irc, here is how we did it:
> 1, copy /386/bin/venti/venti to 9fat
> 2, boot into the live-cd
> 3, run /n/9fat/venti -s -c /dev/sdC0/arenas -a 'tcp!127.1!17034' -h 'tcp!127.1!8000'
> it suicides here too
> 4, run acid on the crashed venti, here is the output of acid: http://jsoros.free.fr/CIMG5213.JPG
> Thanks for any insite on what could be wrong and what I could do to fix it.
>
> Regards
> John

Re: [9fans] venti woes with brand new install

[gabidiaz]

hello

here is what you suggested:

cid% venti/venti -r -L -d -s -c /dev/sdC0/arenas -a 'tcp!127.1!59999' -h 'tcp!127.1!8009'
T+0.0000 venti: conf...
corrupt zblock: 18042e0 b->_size=8192
  64 65 66 67 68 69 6a 6b
  6c 6d 6e 6f 70 71 72 73
  74 75 76 6b 00 00 00 00
  20 00 00 cd 22 80 01 b8
venti 1400: suicide: sys: trap: fault read addr=0x0 pc=0x0002120d
cid% acid 1400
/proc/1400/text:386 plan 9 executable

/sys/lib/acid/port
/sys/lib/acid/386
acid: lstk();
abort()+0x0 /sys/src/libc/9sys/abort.c:6
freezblock(b=0x18042e0)+0xf8 /sys/src/cmd/venti/srv/zblock.c:72
	s=0x18042e5
	fmt=0x4d292
	i=0x20
freeifile(f=0xa4960)+0x12 /sys/src/cmd/venti/srv/ifile.c:67
runconfig(file=0xdfffef92,config=0xa49b4)+0x8e /sys/src/cmd/venti/srv/config.c:188
	f=0xdfffef92
	ok=0x0
	line=0x0
	s=0x180232e
	flds=0x180232e
	i=0x2
	sv=0x18021d8
	av=0x18021f8
initventi(file=0xdfffef92,conf=0xa49b4)+0x3f /sys/src/cmd/venti/srv/config.c:21
threadmain(argv=0xdfffef6c,argc=0x0)+0x139 /sys/src/cmd/venti/srv/venti.c:113
	vaddr=0xdfffefa6
	haddr=0xdfffefb9
	configfile=0xdfffef92
	webroot=0x0
	mem=0x0
	icmem=0x0
	bcmem=0x0
	_argc=0xfefe0068
	_args=0x4e4bc
	config=0x1802218
mainlauncher(arg=0x63fd8)+0x18 /sys/src/libthread/main.c:58
launcher386(arg=0x63fd8,f=0x2f8d5)+0x10 /sys/src/libthread/386.c:10
0xfefefefe ?file?:0
acid: acid: *(0x18042e5\s)

acid: *(0x180232e\s)
icmem
acid:

the config file content is:

index main
isect /dev/sdC0/isect
arenas /dev/sdC0/arenas
bloom /dev/sdC0/bloom
mem 10M
bcmem 20M
icmem 30M

with the final \n

thanks

gabi
PD: you function compiled fine :)

> as a salute to the days of typing in machine code from smalltalk
> magazine.  (wasted youth!)  here is the backtrace complete with
> errors introduced by a faulty connection between the image and
> keyboard.
>
> abort+0x0 /sys/src/libc/9sys/abort.c:6
> freezblock(b=0x17d1f90)+0x38 /sys/src/cmd/venti/srv/zblock.c:63
> freeifile(f=0xa4918)+0x12 /sys/src/cmd/venti/srv/ifile.c:67
> runconfig(file=0xdfffef92,config=0xa496c)+0x8e /sys/src/cmd/venti/srv/config.c:188
> 	f=0xdfffef92
> 	ok=0x0
> 	line=0x0
> 	s=0x17cff8e
> 	flds=0x17cff8e
> 	i=0x2
> 	sv-0x17cfe68
> 	av=0x17cfe88
> initventi(file=0xdfffef92,conf=0xa496c)+0x3f /sys/src/cmd/venti/srv/config.c:21
> threadmain(argv=0xdfffef70,argc=0x0)+0x139) /sys/src/cmd/venti/srv/venti.c:113
> 	vaddr=0xdfffefa6
> 	haddr=0xdfffefb9
> 	configfile=0xdfffef92
> 	webroot=0x0
> 	mem=0x0
> 	icmem=0x0
> 	bcmem=0x0
> 	_argc=0xfefe0068
> 	_args-0x4e474
> 	config=0x17cfea8
> [...]
>
> could you send the output of venti/config? and what does acid
> say to this
>
> 	acid: *(0x17cff8e\s)
>
> you might also get more information by inspecting how zmagic
> is corrupt, e.g. replace freezblock with something like this.
> (may or may not compile.)
>
> void
> freezblock(ZBlock *b)
> {
> 	if(b){
> 		if(memcmp(b->data+b->_size, zmagic, OverflowCheck) != 0){
> 			int i;
> 			uchar *s;
> 			char *fmt;
>
> 			s = (uchar*)(b->data+b->_size);
> 			fmt = "  %.2x %.2x %.2x %.2x %.2x %.2x %.2x %.2x\n";
>
> 			fprint(2, "corrupt zblock: %p b->_size=%d", b, b->_size);
> 			for(i = 0; i < OverflowCheck; i += 8, s += 8)
> 				fprint(2, fmt, s[0], s[1], s[2], s[3], s[4], s[5], s[6], s[7]);
> 			abort();
> 		}
> 		memset(b->data+b->_size, 0, OverflowCheck);
> 		free(b->free);
> 	}
> }
>
> - erik
>
>> Hi everyone!
>> I'm trying to configure a new cpu+auth server using the instructions in the wiki, and I've crossed a problem with venti.
>> Here is what I did:
>> 1, downloaded the sept. 6th iso from plan9 download site
>> 2, installed a fossil+venti system (default setup)
>> 3, followed the wiki instructions to set up a cpu+auth server (compiled my own 9pccpuf kernel with default config)
>> then I rebooted my computer (luckily set up a menu with options to start the old terminal kernel),
>> and got venti suicide, here is a screenshot of the crash: http://jsoros.free.fr/CIMG5212.JPG , after this the computer reboots. We did a little debugging with cinap on irc, here is how we did it:
>> 1, copy /386/bin/venti/venti to 9fat
>> 2, boot into the live-cd
>> 3, run /n/9fat/venti -s -c /dev/sdC0/arenas -a 'tcp!127.1!17034' -h 'tcp!127.1!8000'
>> it suicides here too
>> 4, run acid on the crashed venti, here is the output of acid: http://jsoros.free.fr/CIMG5213.JPG
>> Thanks for any insite on what could be wrong and what I could do to fix it.
>>
>> Regards
>> John

Re: [9fans] venti woes with brand new install

[John Soros]

Hi, again me.
I'm wondering how I will know when I can get back to using the sources venti, a changelog of some-sort would be welcomed to see when a specific change has been made. For example this time it would be nice to know when I need not be worried about being able to boot with venti if I recompile the kernel.
Thanks for the help, the new kernel with modified venti still works fine.
Cheers
John

Re: [9fans] venti woes with brand new install

[erik quanstrom]

you can use history(1) on sources.

cpu% 9fs sources
cpu% cd /n/sources/plan9/sys/src/cmd/venti/srv
cpu% history ifile.c
Sep  8 23:03:56 EDT 2007 ifile.c 2592 [geoff]
Sep  8 23:03:56 EDT 2007 /n/sourcesdump/2007/0909/plan9/sys/src/cmd/venti/srv/ifile.c 2592 [geoff]
Apr 25 17:03:46 EDT 2007 /n/sourcesdump/2007/0908/plan9/sys/src/cmd/venti/srv/ifile.c 2327 [geoff]

- erik

> Hi, again me.
> I'm wondering how I will know when I can get back to using the sources venti, a changelog of some-sort would be welcomed to see when a specific change has been made. For example this time it would be nice to know when I need not be worried about being able to boot with venti if I recompile the kernel.
> Thanks for the help, the new kernel with modified venti still works fine.
> Cheers
> John

Re: [9fans] venti woes with brand new install

[John Soros]

True, but there are no comments at all, so I can't really know what the changes fix.
I'm thinking, as i'm not a big C hacker, that a completely different change might be made to fix the problem, and the change i'm waiting for would then never appear.
Oh well, I guess I'm expecting a lot: "comments", but at the same time, without those how should we know what's happening and why?

oh well, i guess the answer to this would be "learn C", and btw that is what i'm doing.

Regards
John


On Sun, 9 Sep 2007 08:05:06 -0400
erik quanstrom <quanstro@quanstro.net> wrote:

> you can use history(1) on sources.
>
> cpu% 9fs sources
> cpu% cd /n/sources/plan9/sys/src/cmd/venti/srv
> cpu% history ifile.c
> Sep  8 23:03:56 EDT 2007 ifile.c 2592 [geoff]
> Sep  8 23:03:56 EDT 2007 /n/sourcesdump/2007/0909/plan9/sys/src/cmd/venti/srv/ifile.c 2592 [geoff]
> Apr 25 17:03:46 EDT 2007 /n/sourcesdump/2007/0908/plan9/sys/src/cmd/venti/srv/ifile.c 2327 [geoff]
>
> - erik
>
> > Hi, again me.
> > I'm wondering how I will know when I can get back to using the sources venti, a changelog of
> > some-sort would be welcomed to see when a specific change has been made. For example this time
> > it would be nice to know when I need not be worried about being able to boot with venti if I
> > recompile the kernel. Thanks for the help, the new kernel with modified venti still works fine.
> > Cheers John
>

Re: [9fans] venti woes with brand new install

[Uriel]

Even if you know C, the reasons and purpose of a change are very often
non-obvious. Specially if you are not familiar with the code being
changed and all you want to know is if whatever bug was bothering you
is supposedly fixed or not (so you can test it and provide due
feedback).

uriel

On 9/9/07, John Soros <sorosj@gmail.com> wrote:
> True, but there are no comments at all, so I can't really know what the changes fix.
> I'm thinking, as i'm not a big C hacker, that a completely different change might be made to fix the problem, and the change i'm waiting for would then never appear.
> Oh well, I guess I'm expecting a lot: "comments", but at the same time, without those how should we know what's happening and why?
>
> oh well, i guess the answer to this would be "learn C", and btw that is what i'm doing.
>
> Regards
> John
>
>
> On Sun, 9 Sep 2007 08:05:06 -0400
> erik quanstrom <quanstro@quanstro.net> wrote:
>
> > you can use history(1) on sources.
> >
> > cpu% 9fs sources
> > cpu% cd /n/sources/plan9/sys/src/cmd/venti/srv
> > cpu% history ifile.c
> > Sep  8 23:03:56 EDT 2007 ifile.c 2592 [geoff]
> > Sep  8 23:03:56 EDT 2007 /n/sourcesdump/2007/0909/plan9/sys/src/cmd/venti/srv/ifile.c 2592 [geoff]
> > Apr 25 17:03:46 EDT 2007 /n/sourcesdump/2007/0908/plan9/sys/src/cmd/venti/srv/ifile.c 2327 [geoff]
> >
> > - erik
> >
> > > Hi, again me.
> > > I'm wondering how I will know when I can get back to using the sources venti, a changelog of
> > > some-sort would be welcomed to see when a specific change has been made. For example this time
> > > it would be nice to know when I need not be worried about being able to boot with venti if I
> > > recompile the kernel. Thanks for the help, the new kernel with modified venti still works fine.
> > > Cheers John
> >
>

Re: [9fans] venti woes with brand new install

[erik quanstrom]

> True, but there are no comments at all, so I can't really know what the changes fix.
> I'm thinking, as i'm not a big C hacker, that a completely different change might be made to fix the problem, and the change i'm waiting for would then never appear.
> Oh well, I guess I'm expecting a lot: "comments", but at the same time, without those how should we know what's happening and why?
>
> oh well, i guess the answer to this would be "learn C", and btw that is what i'm doing.

http://www.cenqua.com/commentator/

in plan 9 there's no dialog that opens up when you save a file
to allow one to comment on any changes. ;-)  history is not
revision control.  it works at the filesystem level as a complete
copy of each file is available for each day in the "dump" file
system.  thus it works on any file witha granularity of 1 day.
many changes to a file during a day show up as a single
change from day n to day n+1.

- erik

Re: [9fans] venti woes with brand new install

[Uriel]

> history is not revision control.

One doesn't need to use revision control to keep a reasonable
changelog, see for example the /CHANGES file in both Inferno and p9p
(obviously revision control can be used to maintain a changelog,
whatever that is a good or a bad idea is a matter of argument).

> it works at the filesystem level as a complete
> copy of each file is available for each day in the "dump" file
> system.  thus it works on any file witha granularity of 1 day.
> many changes to a file during a day show up as a single
> change from day n to day n+1.

Actually, fossil works with much more fine grained granularity than
one day, it is up to the user to decide how that granularity should be
and for how long such snaps should be kept. (sorry for nit picking,
you probably knew that already, just was pointing this out for whoever
might not know)

uriel

Re: [9fans] venti woes with brand new install

[SHRIZZA]

> I'm wondering how I will know when I can get back to using the
> sources venti, a changelog of some-sort would be welcomed to see
> when a specific change has been made. For example this time it
> would be nice to know when I need not be worried about being able
> to boot with venti if I recompile the kernel.

http://groups.google.com/group/plan9changes

Re: [9fans] venti woes with brand new install

[Uriel]

I created plan9changes (and try to keep it running), and its value is
limited because only a very very small percentage of changes are
submitted though patch(1).

There have been efforts in the past for someone to manually annotate
automatically generated diffs in a 'post-facto' way, nobody lasted
long doing it, which is unsurprising given that this involved the
person writing the changelog entries having to guess the reasons for
the changes (painful and errorprone) and often having to check with
whoever made the change (tedious, slow and rather neutralizing the
whole point of having a changelog).

Why this approach was considered a good idea rather than having the author
of the change write a changelog entry at the same time the change is
made/committed-to-sources is something I have never understood.

uriel

On 9/10/07, SHRIZZA <shrizza@gmail.com> wrote:
> > I'm wondering how I will know when I can get back to using the
> > sources venti, a changelog of some-sort would be welcomed to see
> > when a specific change has been made. For example this time it
> > would be nice to know when I need not be worried about being able
> > to boot with venti if I recompile the kernel.
>
> http://groups.google.com/group/plan9changes
>

Re: [9fans] venti woes with brand new install

[cinap_lenrek]

> hello

> here is what you suggested:
>
> cid% venti/venti -r -L -d -s -c /dev/sdC0/arenas -a 'tcp!127.1!59999' -h 'tcp!127.1!8009'
> T+0.0000 venti: conf...
> corrupt zblock: 18042e0 b->_size=8192
>   64 65 66 67 68 69 6a 6b
>   6c 6d 6e 6f 70 71 72 73
>   74 75 76 6b 00 00 00 00
>   20 00 00 cd 22 80 01 b8
> venti 1400: suicide: sys: trap: fault read addr=0x0 pc=0x0002120d
> cid% acid 1400
> /proc/1400/text:386 plan 9 executable

found it!

that is "defghijklmnopqrstuv..." instead of the expected "1234567890abc..."

13 bytes!

b->data is incremented (and b->len decremented)13 bytes to skip the
"venti config" magic in readifile() ifile.c:39.
and freezblock() alculates the position of the zmagic by b->data + b->_size.

maybe add a b->_data in ZBuffer or fix ifile.c?

cinap

Re: [9fans] venti woes with brand new install

[erik quanstrom]

> found it!
>
> that is "defghijklmnopqrstuv..." instead of the expected "1234567890abc..."
>
> 13 bytes!
>
> b->data is incremented (and b->len decremented)13 bytes to skip the
> "venti config" magic in readifile() ifile.c:39.
> and freezblock() alculates the position of the zmagic by b->data + b->_size.
>
> maybe add a b->_data in ZBuffer or fix ifile.c?
>
> cinap

this might not fix everything as the tail magic was corrupted, too.

- erik


n/sources/plan9//sys/src/cmd/venti/srv/ifile.c:36,43 - ifile.c:36,44
  			freepart(p);
  			return -1;
  		}
- 		b->data += m;
- 		b->len -= m;
+ 		memmove(b->data, b->data+m, 8192-m);
+ 		memset(b->data+8192-m, 0, 8192-m);
+
  		z = memchr(b->data, 0, b->len);
  		if(z)
  			b->len = z - b->data;

Re: [9fans] venti woes with brand new install

[cinap_lenrek]

[-- Attachment #1: Type: text/plain, Size: 383 bytes --]

>   64 65 66 67 68 69 6a 6b
>   6c 6d 6e 6f 70 71 72 73
>   74 75 76 6b 00 00 00 00
>   20 00 00 cd 22 80 01 b8

tailmagic corrupted?

we got:
defghijklmnopqrstuv | k\0\0\0\0 \0\0\cd\22\80\01\b8

we expect:
1234567890abcdefghijklmnopqrstuv | xyz

OverflowCheck is 32 bytes, so xyz are not included in the buffer.

i dont think that the zmagic got corrupted.

cinap

[-- Attachment #2: Type: message/rfc822, Size: 3265 bytes --]

From: erik quanstrom <quanstro@quanstro.net>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] venti woes with brand new install
Date: Sat, 8 Sep 2007 12:49:30 -0400
Message-ID: <3e90b031be3e4fc9eafb1eeadfb9f9e1@quanstro.net>

> found it!
>
> that is "defghijklmnopqrstuv..." instead of the expected "1234567890abc..."
>
> 13 bytes!
>
> b->data is incremented (and b->len decremented)13 bytes to skip the
> "venti config" magic in readifile() ifile.c:39.
> and freezblock() alculates the position of the zmagic by b->data + b->_size.
>
> maybe add a b->_data in ZBuffer or fix ifile.c?
>
> cinap

this might not fix everything as the tail magic was corrupted, too.

- erik


n/sources/plan9//sys/src/cmd/venti/srv/ifile.c:36,43 - ifile.c:36,44
  			freepart(p);
  			return -1;
  		}
- 		b->data += m;
- 		b->len -= m;
+ 		memmove(b->data, b->data+m, 8192-m);
+ 		memset(b->data+8192-m, 0, 8192-m);
+
  		z = memchr(b->data, 0, b->len);
  		if(z)
  			b->len = z - b->data;

Re: [9fans] venti woes with brand new install

[cinap_lenrek]

> n/sources/plan9//sys/src/cmd/venti/srv/ifile.c:36,43 - ifile.c:36,44
>   			freepart(p);
>   			return -1;
>   		}
> - 		b->data += m;
> - 		b->len -= m;
> + 		memmove(b->data, b->data+m, 8192-m);

/*
 * > + 		memset(b->data+8192-m, 0, 8192-m);
 */

memset(b->data + 8192-m, 0, m);

> +
>   		z = memchr(b->data, 0, b->len);
>   		if(z)
>   			b->len = z - b->data;

cinap

Re: [9fans] venti woes with brand new install

[John Soros]

re all!
Well after quite a debugging session with cinap tonight on irc, I can confirm that the patch from eric with the modifications cinap made to it works like a charm. I was able to boot the new kernel, and it's now up and running.

regards
John

On Sat, 8 Sep 2007 21:08:34 +0200
cinap_lenrek@gmx.de wrote:

> > n/sources/plan9//sys/src/cmd/venti/srv/ifile.c:36,43 - ifile.c:36,44
> >   			freepart(p);
> >   			return -1;
> >   		}
> > - 		b->data += m;
> > - 		b->len -= m;
> > + 		memmove(b->data, b->data+m, 8192-m);
>
> /*
>  * > + 		memset(b->data+8192-m, 0, 8192-m);
>  */
>
> memset(b->data + 8192-m, 0, m);
>
> > +
> >   		z = memchr(b->data, 0, b->len);
> >   		if(z)
> >   			b->len = z - b->data;
>
> cinap
>