lilo/raid?

lilo/raid?

[Roy Sigurd Karlsbakk]

hi

still - sorry if this is OT - I'm just too close to tear my hair or head off 
or something

The documentation everywhere, including the lilo 22.3.1 sample conf ffile 
tells me "use boot = /dev/md0", but lilo, when run, just tells me 

Fatal: Filesystem would be destroyed by LILO boot sector: /dev/md0

Please help

roy

-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.

Re: lilo/raid?

[Bill Davidsen]

On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:

> still - sorry if this is OT - I'm just too close to tear my hair or head off 
> or something
> 
> The documentation everywhere, including the lilo 22.3.1 sample conf ffile 
> tells me "use boot = /dev/md0", but lilo, when run, just tells me 
> 
> Fatal: Filesystem would be destroyed by LILO boot sector: /dev/md0

I saw something like that when someone had made a raid device by hand and
used hda and hdb instead of hda1 and hdb1.

-- 
bill davidsen <davidsen@tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.

Re: lilo/raid?

[Andries Brouwer]

On Mon, Jul 01, 2002 at 04:04:58PM +0200, Roy Sigurd Karlsbakk wrote:

> still - sorry if this is OT - I'm just too close to tear my hair or head off 
> or something
> 
> The documentation everywhere, including the lilo 22.3.1 sample conf ffile 
> tells me "use boot = /dev/md0", but lilo, when run, just tells me 
> 
> Fatal: Filesystem would be destroyed by LILO boot sector: /dev/md0
> 
> Please help

But (i) why don't you ask the maintainer of LILO,
(ii) why don't you read the source?

If I grep for "Filesystem would be destroyed" in the LILO sources
I find it in lilo-22.3.1/bsect.c:

    ireloc = part_nowrite(boot_dev);
    if (ireloc>1) {
        die("Filesystem would be destroyed by LILO boot sector: %s", boot_dev);
    }

So, some routine part_nowrite() returned a value larger than 1.
This routine lives in partition.c and does

int part_nowrite(char* device)
{
    int fd;
    BOOT_SECTOR bs;

    int ret=0;  /* say ok, unless we recognize a problem partition */

    if ((fd = open(device, O_RDONLY)) < 0) pdie("part_nowrite check:");
    if (read(fd, bs.sector, sizeof(bs)) != SECTOR_SIZE) pdie("part_nowrite: rea\
d:");

/* check for XFS */
    if (!strncmp("XFSB", bs.sector, 4)) ret=2;

/* check for NTFS */
    else if (   !strncmp("NTFS", bs.par_d.system, 4)
                || strstr(bs.sector,"NTLDR")  ) ret=2;

/* check for SWAP -- last check, as 'bs' is overwritten */
    else if (*(long*)bs.sector == 0xFFFFFFFEUL) {
        if (lseek(fd, (PAGE_SIZE)-SECTOR_SIZE, SEEK_SET) != (PAGE_SIZE)-SECTOR_\
SIZE)
            pdie("part_nowrite lseek:");
        if (SECTOR_SIZE != read(fd, bs.sector, sizeof(bs)) ) pdie("part_nowrite\
 swap check:");
        if (!strncmp(bs.sector+SECTOR_SIZE-10,"SWAPSPACE2",10)
            || !strncmp(bs.sector+SECTOR_SIZE-10,"SWAP-SPACE",10) ) ret=2;
    }

/* check for override (-F) command line flag */
    if (ret==2 && force_fs) {
        fprintf(errstd, "WARNING: '-F' override used. Filesystem on  %s  may be\
 destroyed.\n", device);
        ret=0;
    }

    return ret;
}

So, it looks like (1) you can use the -F flag and lilo will go ahead anyway,
(2) lilo thinks it recognizes XFS or NTFS or swap space on the boot sector
of your device.

It must be easy for you to investigate what you have in reality, and
determine whether there is any potential problem.

Andries

Re: lilo/raid?

[Zwane Mwaikambo]

On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:

> LABEL=/                 /                       ext3    defaults        1 1
> /dev/md2                /tmp                    ext3    defaults        1 2
> /dev/md3                /var                    jfs     defaults        1 2
> /dev/md4                /data                   jfs     defaults        1 2
> /dev/md1                swap                    swap    defaults        0 0

One small thing, you do know that you can interleave swap?

Regards,
	Zwane

-- 
http://function.linuxpower.ca
		

Re: lilo/raid?

[Roy Sigurd Karlsbakk]

On Monday 01 July 2002 17:52, Bill Davidsen wrote:
> On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:
> > still - sorry if this is OT - I'm just too close to tear my hair or head
> > off or something
> >
> > The documentation everywhere, including the lilo 22.3.1 sample conf ffile
> > tells me "use boot = /dev/md0", but lilo, when run, just tells me
> >
> > Fatal: Filesystem would be destroyed by LILO boot sector: /dev/md0
>
> I saw something like that when someone had made a raid device by hand and
> used hda and hdb instead of hda1 and hdb1.

problem is: lilo does not seem to install at all with hd[ab]1 given. only 
hdm(!!!), and then it just goes LI

See dmesg log at http://karlsbakk.net/bugs/ for more info

thanks for all help

roy

-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.

Re: lilo/raid?

[Fabio Massimo Di Nitto]

Roy Sigurd Karlsbakk wrote:

>hi
>
>still - sorry if this is OT - I'm just too close to tear my hair or head off 
>or something
>
>The documentation everywhere, including the lilo 22.3.1 sample conf ffile 
>tells me "use boot = /dev/md0", but lilo, when run, just tells me 
>
>Fatal: Filesystem would be destroyed by LILO boot sector: /dev/md0
>
>Please help
>
>roy
>
>  
>
Hi Roy,
            I have some machines running root on raid but I always have
boot=/dev/hda and a second copy of lilo.con with boot pointing to /dev/hdb
or whatever drive it was.
What is really importat for you anyway is to be able to boot again if one of
the disk is failing. The raid will ensure you to keep the root but I 
always keep
a boot floppy for emergencies. Not all the BIOS supports a boot from 
/dev/hdb
(I used hda/b just as an example...).
At end if you loose a disk you will be always able to change it and boot 
again
to rebuilt it.

Fabio

Re: lilo/raid?

[Roy Sigurd Karlsbakk]

> What is really importat for you anyway is to be able to boot again if one
> of the disk is failing. 

My main concern now is to be able to boot the bloody server from something 
apart from a floppy!

-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.

Re: lilo/raid?

[Roy Sigurd Karlsbakk]

> So, it looks like (1) you can use the -F flag and lilo will go ahead
> anyway, (2) lilo thinks it recognizes XFS or NTFS or swap space on the boot
> sector of your device.

Thanks
I added some debug output in part_nowrite(), and - you're right - it thinks 
it's NTFS. More info below

fstab:
LABEL=/                 /                       ext3    defaults        1 1
/dev/md2                /tmp                    ext3    defaults        1 2
/dev/md3                /var                    jfs     defaults        1 2
/dev/md4                /data                   jfs     defaults        1 2
/dev/md1                swap                    swap    defaults        0 0
none                    /dev/pts                devpts  gid=5,mode=620  0 0
none                    /proc                   proc    defaults        0 0
none                    /dev/shm                tmpfs   defaults        0 0
/dev/fd0                /mnt/floppy             auto    noauto,owner,kudzu 0 0

/proc/partitions:
   9     0    4096448 md0 0 0 0 0 0 0 0 0 0 0 0
   9     1    2048128 md1 0 0 0 0 0 0 0 0 0 0 0
   9     2    2048192 md2 0 0 0 0 0 0 0 0 0 0 0
   9     3   40958720 md3 0 0 0 0 0 0 0 0 0 0 0
   9     4 1631383040 md4 0 0 0 0 0 0 0 0 0 0 0
  89     0  120627360 hdo 52 144 632 400 524 106 4368 130 -2 534050 41882082
  89     1    4095976 hdo1 40 35 360 230 517 6 4136 80 0 260 310
  89     2  116531352 hdo2 11 106 264 160 7 100 232 50 0 110 210
  89    64  120627360 hdp 7 3 56 100 2 0 16 0 -2 534050 41881682
  89    65    4095976 hdp1 3 0 24 40 1 0 8 0 0 40 40
  89    66  116531352 hdp2 3 0 24 50 1 0 8 0 0 50 50
  88     0  120627360 hdm 23 10 192 230 259 0 2072 80 -2 534060 41881822
  88     1    4096543 hdm1 16 0 128 160 257 0 2056 80 0 220 240
  88     2  116527477 hdm2 6 7 56 60 2 0 16 0 0 60 60
  88    64  120627360 hdn 22 10 184 140 13 0 104 10 -2 534070 41881682
  88    65    4096543 hdn1 18 7 152 110 12 0 96 10 0 120 120
  88    66  116527477 hdn2 3 0 24 30 1 0 8 0 0 30 30
  57     0  120627360 hdk 44 53 488 400 342 29 2824 180 -2 534070 41882072
  57     1    4095976 hdk1 36 31 392 330 341 29 2816 180 0 420 510
  57     2  116531352 hdk2 7 19 88 60 1 0 8 0 0 60 60
  57    64  120627360 hdl 25 159 488 260 177 155 1696 110 -2 534040 41881862
  57    65    4095976 hdl1 19 62 264 170 174 61 1496 90 0 240 260
  57    66  116531352 hdl2 5 94 216 90 3 94 200 20 0 80 110
  56     0  120627360 hdi 32 14 296 240 129 12 1080 70 -2 534080 41881772
  56     1    4095976 hdi1 28 11 264 200 128 12 1072 70 0 240 270
  56     2  116531352 hdi2 3 0 24 40 1 0 8 0 0 40 40
  56    64  120627360 hdj 39 68 424 270 324 62 2704 230 -2 534030 41881972
  56    65    4095976 hdj1 32 62 368 210 323 62 2696 230 0 350 440
  56    66  116531352 hdj2 6 3 48 50 1 0 8 0 0 50 50
  34     0  120627360 hdg 37 7 328 360 392 2 3152 130 -2 534110 41881922
  34     1    4095976 hdg1 32 4 288 300 390 2 3136 130 0 350 430
  34     2  116531352 hdg2 4 0 32 60 2 0 16 0 0 60 60
  34    64  120627360 hdh 26 5 224 260 330 2 2656 160 -2 534060 41881872
  34    65    4095976 hdh1 22 2 192 190 329 2 2648 160 0 300 350
  34    66  116531352 hdh2 3 0 24 60 1 0 8 0 0 60 60
  33     0  120627360 hde 36 33 312 400 15 7 128 30 -2 534130 41881842
  33     1    4095976 hde1 32 30 280 350 14 7 120 30 0 140 380
  33     2  116531352 hde2 3 0 24 40 1 0 8 0 0 40 40
  33    64  120627360 hdf 39 11 376 350 863 5 6944 480 -2 534090 41882252
  33    65    4095976 hdf1 35 8 344 280 862 5 6936 480 0 630 760
  33    66  116531352 hdf2 3 0 24 60 1 0 8 0 0 60 60
  22     0  120627360 hdc 20 12 154 100 20 27 360 10 -2 534140 41881502
  22     1    2048224 hdc1 4 0 32 0 1 0 8 0 0 0 0
  22     2    2048256 hdc2 12 9 90 50 18 27 344 10 0 60 60
  22     3  116530848 hdc3 3 0 24 50 1 0 8 0 0 50 50
  22    64  120627360 hdd 22 68 264 150 20 27 360 10 -2 534140 41881552
  22    65    2048224 hdd1 5 0 40 20 1 0 8 0 0 20 20
  22    66    2048256 hdd2 13 65 192 70 18 27 344 10 0 80 80
  22    67  116530848 hdd3 3 0 24 50 1 0 8 0 0 50 50
   3     0  120627360 hda 972 1726 21338 3630 502 1142 13192 1210 -2 533970 
41886202
   3     1    4096543 hda1 953 1694 21170 3540 497 1136 13152 1210 0 3550 4750
   3     2  116527477 hda2 18 29 160 90 5 6 40 0 0 90 90
   3    64  120627360 hdb 844 1990 22648 3410 498 1136 13160 2740 -2 533950 
41887522
   3    65    4096543 hdb1 840 1987 22616 3350 497 1136 13152 2740 0 3340 6090
   3    66  116527477 hdb2 3 0 24 50 1 0 8 0 0 50 50


-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.

Re: lilo/raid?

[Bill Davidsen]

On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:

> On Monday 01 July 2002 17:52, Bill Davidsen wrote:
> > On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:
> > > still - sorry if this is OT - I'm just too close to tear my hair or head
> > > off or something
> > >
> > > The documentation everywhere, including the lilo 22.3.1 sample conf ffile
> > > tells me "use boot = /dev/md0", but lilo, when run, just tells me
> > >
> > > Fatal: Filesystem would be destroyed by LILO boot sector: /dev/md0
> >
> > I saw something like that when someone had made a raid device by hand and
> > used hda and hdb instead of hda1 and hdb1.
> 
> problem is: lilo does not seem to install at all with hd[ab]1 given. only 
> hdm(!!!), and then it just goes LI
> 
> See dmesg log at http://karlsbakk.net/bugs/ for more info

No, what I had in mind was that in the raid definition, instead of a
partition such as /dev/hda1, the physical driver was given, like /dev/hda.
That gave the message you mention. In any case you seem to have identified
the problem, so this is probably not relevant.

-- 
bill davidsen <davidsen@tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.

Re: lilo/raid?

[Helge Hafting]

Zwane Mwaikambo wrote:
> 
> On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:
> 
> > LABEL=/                 /                       ext3    defaults        1 1
> > /dev/md2                /tmp                    ext3    defaults        1 2
> > /dev/md3                /var                    jfs     defaults        1 2
> > /dev/md4                /data                   jfs     defaults        1 2
> > /dev/md1                swap                    swap    defaults        0 0
> 
> One small thing, you do know that you can interleave swap?

There are sometimes reasons not to do that.
Heavy swapping may be caused by attempts to cache 
massive io on some fs.  You better not have swap
on that heavily accessed spindle - because then
everything ends up waiting on that io.

Keeping swap somewhere else means other programs
just wait a little for swap - undisturbed by the massive
io also going on.

Helge Hafting

Re: lilo/raid?

[Zwane Mwaikambo]

On Tue, 2 Jul 2002, Helge Hafting wrote:

> > > /dev/md1                swap                    swap    defaults        0 0
> > 
> > One small thing, you do know that you can interleave swap?
> 
> There are sometimes reasons not to do that.
> Heavy swapping may be caused by attempts to cache 
> massive io on some fs.  You better not have swap
> on that heavily accessed spindle - because then
> everything ends up waiting on that io.
> 
> Keeping swap somewhere else means other programs
> just wait a little for swap - undisturbed by the massive
> io also going on.

True, but what i meant was that instead of creating a RAID device to swap 
to, he could have just interleaved normal swap partitions and gotten the 
same effect.

Regards,
	Zwane Mwaikambo

-- 
http://function.linuxpower.ca
		

Re: lilo/raid?

[jw schultz]

On Tue, Jul 02, 2002 at 11:12:03AM +0200, Zwane Mwaikambo wrote:
> On Tue, 2 Jul 2002, Helge Hafting wrote:
> 
> > > > /dev/md1                swap                    swap    defaults        0 0
> > > 
> > > One small thing, you do know that you can interleave swap?
> > 
> > There are sometimes reasons not to do that.
> > Heavy swapping may be caused by attempts to cache 
> > massive io on some fs.  You better not have swap
> > on that heavily accessed spindle - because then
> > everything ends up waiting on that io.
> > 
> > Keeping swap somewhere else means other programs
> > just wait a little for swap - undisturbed by the massive
> > io also going on.
> 
> True, but what i meant was that instead of creating a RAID device to swap 
> to, he could have just interleaved normal swap partitions and gotten the 
> same effect.

Not a config i would recommend.  While spreading swap over
multiple spindles that way is good for speed it multiplies the
likelihood that a disk failure will down the system.  If you
think a dead filesystem is a mess, just watch what happens
when swap goes dead.

Much better to put swap on several RAID-1 volumes if
you want to spread it around.  Disks just aren't that
expensive anymore.  RAID-5 though would be a bad move since
swap gets an order of magnitude more writes than reads.  On
account of the heavy write tendencies i lean toward
so-called hardware RAID for swap in order to offload the PCI
buss.

-- 
________________________________________________________________
	J.W. Schultz            Pegasystems Technologies
	email address:		jw@pegasys.ws

		Remember Cernan and Schmitt

Re: lilo/raid?

[Jakob Oestergaard]

On Tue, Jul 02, 2002 at 11:12:03AM +0200, Zwane Mwaikambo wrote:
> On Tue, 2 Jul 2002, Helge Hafting wrote:
> 
> > > > /dev/md1                swap                    swap    defaults        0 0
> > > 
> > > One small thing, you do know that you can interleave swap?
> > 
> > There are sometimes reasons not to do that.
> > Heavy swapping may be caused by attempts to cache 
> > massive io on some fs.  You better not have swap
> > on that heavily accessed spindle - because then
> > everything ends up waiting on that io.
> > 
> > Keeping swap somewhere else means other programs
> > just wait a little for swap - undisturbed by the massive
> > io also going on.
> 
> True, but what i meant was that instead of creating a RAID device to swap 
> to, he could have just interleaved normal swap partitions and gotten the 
> same effect.

If it is a RAID-1 device, there are very good reasons for creating a
RAID device for the swapping  :)

But other than that, you are right. There is no reason for creating a
RAID-0 device for swapping.

-- 
................................................................
:   jakob@unthought.net   : And I see the elder races,         :
:.........................: putrid forms of man                :
:   Jakob Østergaard      : See him rise and claim the earth,  :
:        OZ9ABN           : his downfall is at hand.           :
:.........................:............{Konkhra}...............:

Re: lilo/raid?

[Roy Sigurd Karlsbakk]

On Monday 01 July 2002 17:59, Zwane Mwaikambo wrote:
> On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:
> > LABEL=/                 /                       ext3    defaults        1
> > 1 /dev/md2                /tmp                    ext3    defaults       
> > 1 2 /dev/md3                /var                    jfs     defaults     
> >   1 2 /dev/md4                /data                   jfs     defaults   
> >     1 2 /dev/md1                swap                    swap    defaults 
> >       0 0
>
> One small thing, you do know that you can interleave swap?

What does that mean?

-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.

Re: lilo/raid?

[Roy Sigurd Karlsbakk]

On Tuesday 02 July 2002 10:16, Helge Hafting wrote:
> Zwane Mwaikambo wrote:
> > On Mon, 1 Jul 2002, Roy Sigurd Karlsbakk wrote:
> > > LABEL=/                 /                       ext3    defaults       
> > > 1 1 /dev/md2                /tmp                    ext3    defaults   
> > >     1 2 /dev/md3                /var                    jfs    
> > > defaults        1 2 /dev/md4                /data                   jfs
> > >     defaults        1 2 /dev/md1                swap                   
> > > swap    defaults        0 0
> >
> > One small thing, you do know that you can interleave swap?
>
> There are sometimes reasons not to do that.
> Heavy swapping may be caused by attempts to cache
> massive io on some fs.  
<snip/>

I've seen it. 1GB of swap for caching heavy downloads (se earlier thread 'VM 
fsckup' or somehting)

What is the reason of using swap for cache buffers?????

roy

-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.

Re: lilo/raid?

[Helge Hafting]

Roy Sigurd Karlsbakk wrote:

> I've seen it. 1GB of swap for caching heavy downloads (se earlier thread 'VM
> fsckup' or somehting)
> 
> What is the reason of using swap for cache buffers?????

To be precise - swap is never used _for_ cache buffers - you'll
never see file contents in the swap partition, perhaps with
the exception of tmpfs stuff.

But aggressive caching may indeed push other stuff into swap,
typically little-used program memory.

The balance is probably off, but the feature is a good one.
The point is to minimize disk waiting by ensuring that
often-used stuff is in memory, possibly pushing seldom-used
stuff into swap.

Some workload results in some files being accessed a lot,
such as the homepage files for a webserver.  Caching them
makes sense then, even if some little-used sleeping
program is pushed into swap.  It is used less than those
often-used files, so recovering it from swap when it
eventually is needed results in less total waiting
than what you get if every web server access hits the disk.

The balance being off have several explanations:
1. People don't _expect_ to wait for paging activity, while
   they expect having to wait for the disk.
2. Selecting wich pages to keep and which to swap (or
   not cache if it is a data file) is hard.
   The reverse mapping VM will limit the unnecessary swapping,
   making the best decision is sometimes impossible without
   it, or at least much harder.  
3. The problem is still hard, even with rmap.  That's because
   we don't know how files and executables will be used in
   the future.  All we have to make the decision is statistics
   about previous use.  
   Allowing applications to give the kernel hints may help,
   but giving good hints may be hard, and the potential
   for abuse is there.  (App gives unrealistic hints - 
   and seems more snappy than the competition.  And lots
   of other processes suffer.)  This is avoidable by only
   allowing negative hints, i.e. "Don't bother caching this"

You probably agree that it makes sense to swap out
program initialization code that won't be used
again once the program is up and running.  The problem is
that the VM system can't identify such code other than
by the fact that it is a long time since the last use.

Helge Hafting

Re: lilo/raid?

[Roy Sigurd Karlsbakk]

> > What is the reason of using swap for cache buffers?????
>
> To be precise - swap is never used _for_ cache buffers - you'll
> never see file contents in the swap partition, perhaps with
> the exception of tmpfs stuff.
>
> But aggressive caching may indeed push other stuff into swap,
> typically little-used program memory.

ok.
tell me, then

When having an http-server-of-choice (tried several), I start downloading 
10-50 files at 4Mbps. After some time, the server OOMs. The only processes 
running are syslog, nfs daemons (idle) and the web server. This happens 
without swap or with swap (1gig swap - fills up, and the server dies).

My last thread about it was "[BUG] 2.4 VM sucks. Again". After a rather 
experimental patch my akpm, the problem was solved.

<snip>

roy
-- 
Roy Sigurd Karlsbakk, Datavaktmester

Computers are like air conditioners.
They stop working when you open Windows.