Re: [9fans] kfs error

Re: [9fans] kfs error

[presotto]

[-- Attachment #1: Type: text/plain, Size: 501 bytes --]

forsyth is correct.  Comment from the sync/cmd process in kfs:

/*
 * process to synch blocks
 * it puts out a block/line every second
 * it waits 10 seconds if catches up.
 * in both cases, it takes about 10 seconds
 * to get up-to-date.
 *
 * it also updates the filter stats
 * and executes commands
 */

Unless you give the command 'disk/kfscmd nosync', kfs will write
back updated blocks pretty quickly.  It's still good to do a
'disk/kfscmd halt' before turning the machine off.

[-- Attachment #2: Type: message/rfc822, Size: 3601 bytes --]

[-- Attachment #2.1.1: Type: text/plain, Size: 84 bytes --]

i believe kfs does that automatically every 10 seconds
(see syncproc in main.c)


[-- Attachment #2.1.2: Type: message/rfc822, Size: 2006 bytes --]

From: "Boyd Roberts" <boyd@fr.inter.net>
To: <9fans@cse.psu.edu>
Subject: Re: [9fans] kfs error
Date: Sat, 29 Sep 2001 04:52:58 +0200
Message-ID: <0d1601c14891$da148440$a2b9c6d4@SOMA>

paranoid that i am, i added some code to the end of /rc/bin/termrc
to do a unix style update; it just does a disk/kfscmd sync every 5 mins.

Re: [9fans] kfs error

[Boyd Roberts]

> forsyth is correct.  Comment from the sync/cmd process in kfs:

ok, i'll rip it out.  i like having less code.

> 'disk/kfscmd halt' before turning the machine off.

yeah, i added a function called halt to /rc/bin/termrc too.

Re: [9fans] kfs error

[Wladimir Mutel]

Boyd Roberts <boyd@fr.inter.net> wrote:

>> 'disk/kfscmd halt' before turning the machine off.

> yeah, i added a function called halt to /rc/bin/termrc too.

	Is there a way to finish running 'rio' desktop ? 
	So that termrc could continue after 'rio' and halt the kfs ?

Re: [9fans] kfs error

[Boyd Roberts]

>> yeah, i added a function called halt to /rc/bin/termrc too.

> Is there a way to finish running 'rio' desktop ? 
> So that termrc could continue after 'rio' and halt the kfs ?

boots quickly enough so i just halt and ^t^tr it.

it's rare that i _don't_ want to run rio, although i
did log in as none to take out some kernel print's
with ed, 'cos it drives rio (well, the display and
me) nuts.

Re: [9fans] kfs error

[Wladimir Mutel]

Boyd Roberts <boyd@fr.inter.net> wrote:
>>> yeah, i added a function called halt to /rc/bin/termrc too.

>> Is there a way to finish running 'rio' desktop ? 
>> So that termrc could continue after 'rio' and halt the kfs ?

> boots quickly enough so i just halt and ^t^tr it.

	May be it is possible to finish rio by sending it a signal ?
	^t^tr looks too brutal, and it requires to do 'kfscmd halt' by hand, 
	instead of writing this command in termrc to do it after rio's finish.

[9fans] kfs error

[anothy]

yesterday i started getting errors from kfs. i get a
print on my console of "wrenread failed: i/o error",
and 'kfscmd check' tells me "xtag: p null". also,
while most of these errors return empty files or
directories (where there should be content), some
hang the read. as a result, i can't have kfs do a
more complete check, because it hangs. i've not been
able to determine any pattern for which will return
empty and which will cause the read to block. the
box experienced no trauma, and this started after
a few weeks of uptime; i left it for the day, and it
had started when i got back. the system is, for the
most part, still functional, outside of the few
affected directories. i also get two other errors
from a kfs check: several lines like:
	check: "free list": address dup 87267
one like this:
	check: "free list": xtag: p null
and then after the stats, several lines of the form
"missing: 225065", followed by a "...".

anyone got any idea what could have caused this? i've
got some fairly important data i can't get to; any
chance of recovery, or am i just SOL?
ア

Re: [9fans] kfs error

[William Josephson]

On Mon, Oct 01, 2001 at 08:26:11AM -0400, rob pike wrote:

>           The kprint file may be read to receive a copy of the data
>           written to the console by the kernel's print statements or
>           by processes writing to /dev/cons.  Only data written after

A very pleasant addition which I just backported to 3rd Ed. this
weekend.  Quite convenient.  Didn't realize it was Russ's idea, though.

 -WJ

Re: [9fans] kfs error

[rob pike]

[-- Attachment #1: Type: text/plain, Size: 708 bytes --]

The new, as yet unreleased system, includes this paragraph
in cons(3) documenting an idea by Russ Cox:

          The kprint file may be read to receive a copy of the data
          written to the console by the kernel's print statements or
          by processes writing to /dev/cons.  Only data written after
          the file is opened is available.  If the machine's console
          is a serial line, the data is sent both to the console and
          to kprint; if its console is a graphics screen, the data is
          sent either to the display or to kprint, but not both.  (It
          is advisable not to open kprint on terminals until you have
          started rio(1).)

-rob


[-- Attachment #2: Type: message/rfc822, Size: 2159 bytes --]

From: "Boyd Roberts" <boyd@fr.inter.net>
To: <9fans@cse.psu.edu>
Subject: Re: [9fans] kfs error
Date: Mon, 1 Oct 2001 12:45:17 +0200
Message-ID: <009b01c14a66$290975d0$a2b9c6d4@SOMA>

>> yeah, i added a function called halt to /rc/bin/termrc too.

> Is there a way to finish running 'rio' desktop ? 
> So that termrc could continue after 'rio' and halt the kfs ?

boots quickly enough so i just halt and ^t^tr it.

it's rare that i _don't_ want to run rio, although i
did log in as none to take out some kernel print's
with ed, 'cos it drives rio (well, the display and
me) nuts.

Re: [9fans] kfs error

[jmk]

looking at the code, the rule for kfs seems to be the max file size in
blocks is
	6 + iba + iba*iba
where
	iba = blocksize/sizeof(ulong)
so, for a 1024 byte block iba = 256 and the total number of
blocks is
	6 + 256 + 256*256 = 65798
for the more usual 4096 byte blocks we get
	6 + 1024 + 1024*1024 = 1049606
giving a max file size of 4299186176 bytes.
with a 16Kb block size the max file size becomes 274945114112.

On Sat Sep 29 23:52:36 EDT 2001, anothy@cosym.net wrote:
> aw, shoot. and i knew that (a while ago), too.
> dd to the rescue, i guess. ☹
>
> is the rule always 65798 blocks * blocksize? i
> remember seeing a table of max file sizes with
> different block sizes, and i don't think anyone
> would've bothered to type it out if it was that
> regular a pattern.
>
> thanks, jmk.
> -α.

Re: [9fans] kfs error

[anothy]

aw, shoot. and i knew that (a while ago), too.
dd to the rescue, i guess. ☹

is the rule always 65798 blocks * blocksize? i
remember seeing a table of max file sizes with
different block sizes, and i don't think anyone
would've bothered to type it out if it was that
regular a pattern.

thanks, jmk.
-α.

Re: [9fans] kfs error

[forsyth]

[-- Attachment #1: Type: text/plain, Size: 84 bytes --]

i believe kfs does that automatically every 10 seconds
(see syncproc in main.c)


[-- Attachment #2: Type: message/rfc822, Size: 2006 bytes --]

From: "Boyd Roberts" <boyd@fr.inter.net>
To: <9fans@cse.psu.edu>
Subject: Re: [9fans] kfs error
Date: Sat, 29 Sep 2001 04:52:58 +0200
Message-ID: <0d1601c14891$da148440$a2b9c6d4@SOMA>

paranoid that i am, i added some code to the end of /rc/bin/termrc
to do a unix style update; it just does a disk/kfscmd sync every 5 mins.

Re: [9fans] kfs error

[jmk]

The file is too big.

With 1K blocks and kfs the max file size is
	6 direct blocks
	+256 indirect blocks
	+256*256 double indirect blocks

	= 65798 blocks
	= 67377152 bytes

On Fri Sep 28 22:28:31 EDT 2001, anothy@cosym.net wrote:
> i'm trying to copy a ~120MB CD image from a system i'm
> cpu'd into back to my terminal. the cp fails, complaining
> "file system full". on the terminal console is printed the
> message "dnodebuf: trip indirect". kfscmd check tells me
> i've got 760780 free blocks (1K blocks), which matches
> what i think i should have.
>
> i did a kfscmd check rft, and it found one file with a bad
> xtag, but that was it. any idea what's going on? i've found
> the message in kfs (disk/kfs/dentry.c:89), but i'm not
> exactly clear on what it's doing there. can anyone save me
> the trouble of figuring out what's wrong with my fils
> system? any help, much appreciated.
> -α.
>

Re: [9fans] kfs error

[Boyd Roberts]

paranoid that i am, i added some code to the end of /rc/bin/termrc
to do a unix style update; it just does a disk/kfscmd sync every 5 mins.

[9fans] kfs error

[anothy]

i'm trying to copy a ~120MB CD image from a system i'm
cpu'd into back to my terminal. the cp fails, complaining
"file system full". on the terminal console is printed the
message "dnodebuf: trip indirect". kfscmd check tells me
i've got 760780 free blocks (1K blocks), which matches
what i think i should have.

i did a kfscmd check rft, and it found one file with a bad
xtag, but that was it. any idea what's going on? i've found
the message in kfs (disk/kfs/dentry.c:89), but i'm not
exactly clear on what it's doing there. can anyone save me
the trouble of figuring out what's wrong with my fils
system? any help, much appreciated.
-α.