Skip to contentSkip to conversation

new patches on ftp.namesys.com: 2.4.15-pre7

REISERFS TESTING Nikita Danilov 6 participants 8 messages Nov 21, 2001 – Nov 22, 2001 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 5 lines · Dieter NЭtzel

Sorry Nikita,

but kernel 2.4.15-pre7 + preempt + ReiserFS A-N do _NOT_ boot for me.
I've tried it with "old" and "new" (current) N-inode-attrs.patch. But that doesn't matter.

Quoted 10 lines

[-]
IP: routing cache hash table of 8192 buckets, 64Kbytes
TCP: Hash tables configured (established 262144 bind 65536)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
reiserfs: checking transaction log (device 08:03) ... Using r5 hash to sort names
ReiserFS version 3.6.25
VFS: Mounted root (reiserfs filesystem) readonly. Freeing unused kernel memory: 208k freed
"Warning: unable to open an initial console."

N-inode-attrs.patch uses previously unused field in reiserfs on-disk inode structure to store inode attributes. It seems that in some cases this field actually contains garbage. It may happen that you have got immutable bit for your console device this way.

As a work around try to boot with ext2, do

lsattr /dev/console to check, and run

chattr -R -SAadiscu /reiserfs-mount-point

for each reiserfs file system to clear all attributes. Yes, it's silly.

Quoted 8 lines

SysRq: Show Regs

Pid: 1, comm: init
EIP: 0023 : [<0804c842>] CPU: 0 ESP: 002b:bffffe10 EFLAGS: 00010246 Not tainted
EAX: ffffffff EBX: bfffff00 ECX: 0000000d EDX: ffffffff
ESI: bffffef4 EDI: 00000002 EBP: bffffe78 DS: 002b ES: 002b
CR0: 8005003b CR2: 080609c0 CR3: 27adc000 CR4: 000002d0

Quoted 12 lines
Call Trace:

SysRq: Show State

init, keventd, ksoftirqd_CPU, kswapd, bdflush, kupdated, scsi_eh_0, kreiserfsd

Thanks,
Dieter

Signature · 3 lines
Dieter NЭtzel
Graduate Student, Computer Science
@home: [email protected]

Nikita.

#2 lore raw
In reply to a message not in the archive
Quoted 16 lines · Nikita Danilov
Quoted 4 lines · Dieter NJtzel

but kernel 2.4.15-pre7 + preempt + ReiserFS A-N do _NOT_ boot for me. I've tried it with "old" and "new" (current) N-inode-attrs.patch. But that doesn't matter.

Quoted 10 lines

[-]
IP: routing cache hash table of 8192 buckets, 64Kbytes
TCP: Hash tables configured (established 262144 bind 65536)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
reiserfs: checking transaction log (device 08:03) ... Using r5 hash to sort names
ReiserFS version 3.6.25
VFS: Mounted root (reiserfs filesystem) readonly. Freeing unused kernel memory: 208k freed
"Warning: unable to open an initial console."

Quoted 4 lines

N-inode-attrs.patch uses previously unused field in reiserfs on-disk inode structure to store inode attributes. It seems that in some cases this field actually contains garbage. It may happen that you have got immutable bit for your console device this way.

Hmm, this may be a kernel bug also, in a way. I don't know if ext2 allows you to set attributes on char/block special files, but if it does, then the "immutable" attribute should _probably_ apply to changing the device inode, rather than writing to the device itself.

In any case, it is also a bad thing to leave garbage in unused parts of on-disk data structs for just this reason, so mkreiserfs should zero everything that is unused inside allocated structs (and the kernel too, because reiserfs allocates inode tables dynamically, right?).

Cheers, Andreas
--
Andreas Dilger
http://sourceforge.net/projects/ext2resize/
http://www-mddsp.enel.ucalgary.ca/People/adilger/

#3 lore raw
Quoted 27 lines · Andreas Dilger
Quoted 16 lines · Nikita Danilov
Quoted 4 lines · Dieter NJtzel

but kernel 2.4.15-pre7 + preempt + ReiserFS A-N do _NOT_ boot for me. I've tried it with "old" and "new" (current) N-inode-attrs.patch. But that doesn't matter.

Quoted 10 lines

[-]
IP: routing cache hash table of 8192 buckets, 64Kbytes
TCP: Hash tables configured (established 262144 bind 65536)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
reiserfs: checking transaction log (device 08:03) ... Using r5 hash to sort names
ReiserFS version 3.6.25
VFS: Mounted root (reiserfs filesystem) readonly. Freeing unused kernel memory: 208k freed
"Warning: unable to open an initial console."

Quoted 4 lines

N-inode-attrs.patch uses previously unused field in reiserfs on-disk inode structure to store inode attributes. It seems that in some cases this field actually contains garbage. It may happen that you have got immutable bit for your console device this way.

Hmm, this may be a kernel bug also, in a way. I don't know if ext2 allows you to set attributes on char/block special files, but if it does, then the "immutable" attribute should _probably_ apply to changing the device inode, rather than writing to the device itself.

Quoted 4 lines

In any case, it is also a bad thing to leave garbage in unused parts of on-disk data structs for just this reason, so mkreiserfs should zero everything that is unused inside allocated structs (and the kernel too, because reiserfs allocates inode tables dynamically, right?).

Yes, it's right, but currently we have what we have currently. I am going to extend inode-attrs.patch and add new mount option "noattrs". With it ioctls to set and get attributes will continue to work, but attributes themselves will not have any effect. Then, one can boot with "rootflags=noattrs" and read-write root, clear all attributes by chattr -R and remount root.

I put new version of the patch in the same place, Dieter, can you please try it?

Quoted 2 lines

Cheers, Andreas

Nikita.

Quoted 4 lines

--
Andreas Dilger
http://sourceforge.net/projects/ext2resize/
http://www-mddsp.enel.ucalgary.ca/People/adilger/

#4 lore raw
In reply to a message not in the archive

Am Mittwoch, 21. November 2001 11:02 schrieb Nikita Danilov:

Quoted 9 lines

Yes, it's right, but currently we have what we have currently. I am going to extend inode-attrs.patch and add new mount option "noattrs". With it ioctls to set and get attributes will continue to work, but attributes themselves will not have any effect. Then, one can boot with "rootflags=noattrs" and read-write root, clear all attributes by chattr -R and remount root.

I put new version of the patch in the same place, Dieter, can you please try it?

So here it goes.
You made one mistake in your README file:

[-]
chattr -R -SAadiscu /reiserfs-mount-point

for each reiserfs file system. Said garbage on root file-system can prevent system from booting (for example spurious immutable flag on /dev/console) to work around this, new mount option "noattrs" is provided. With it chattr and lsattr still work, but attributes don't have any actual effect. So, boot system with
"rootflags=noattr", clear all attributes and reboot or remount root file system.
[-]

You've forgotten the little "s" but had it right in your post.

After I've 'booted into single user mode with "linux rootflags=noattrs" I could change my archives but _NOT_ /dev/console the right way. After the second reboot I got the warning and kernel hang, again. Something wrong with your patch or my installation?

SunWave1 /# l /dev/console
crw------- 1 root root 5, 1 Nov 21 11:01 /dev/console
SunWave1 /# lsattr /dev/console
lsattr: Invalid argument While reading flags on /dev/console

Two other FS in /dev have some problems, too. But I think that is right.

SunWave1 /# chattr -R -SAadiscu /dev/
chattr: Inappropriate ioctl for device while reading flags on /dev//pts
chattr: Inappropriate ioctl for device while reading flags on /dev//shm

All the files in /proc show the same.

Here comes a little positive example, I think.

SunWave1 /# lsattr /tmp/
----------X-j /tmp//xwlog
--------B-X-j /tmp//kde-nuetzel
------------- /tmp//dcop6SKhoM
---------D-Ej /tmp//YaST2.tdir
----------XEj /tmp//ksocket-root
--------B-X-j /tmp//mcop-nuetzel
--------B--Ej /tmp//kde-root
--------B-X-j /tmp//ksocket-nuetzel
--------B--Ej /tmp//INSTALL
--------BD-Ej /tmp//modules.conf.-
---------DX-j /tmp//vi.recover
--------BD-Ej /tmp//mcop-root
------------j /tmp//lost+found
--------BD-Ej /tmp//modules.conf

What next?

Thanks,
Dieter

PS Have you read about the latest ACL discussion on LKML?

#5 lore raw
Quoted 27 lines · Dieter NЭtzel

Am Mittwoch, 21. November 2001 11:02 schrieb Nikita Danilov:

Quoted 9 lines

Yes, it's right, but currently we have what we have currently. I am going to extend inode-attrs.patch and add new mount option "noattrs". With it ioctls to set and get attributes will continue to work, but attributes themselves will not have any effect. Then, one can boot with "rootflags=noattrs" and read-write root, clear all attributes by chattr -R and remount root.

I put new version of the patch in the same place, Dieter, can you please try it?

So here it goes.
You made one mistake in your README file:

[-]
chattr -R -SAadiscu /reiserfs-mount-point

for each reiserfs file system. Said garbage on root file-system can prevent system from booting (for example spurious immutable flag on /dev/console) to work around this, new mount option "noattrs" is provided. With it chattr and lsattr still work, but attributes don't have any actual effect. So, boot system with
"rootflags=noattr", clear all attributes and reboot or remount root file system.
[-]

Quoted 6 lines

You've forgotten the little "s" but had it right in your post.

After I've 'booted into single user mode with "linux rootflags=noattrs" I could change my archives but _NOT_ /dev/console the right way. After the second reboot I got the warning and kernel hang, again. Something wrong with your patch or my installation?

Something is wrong with my head, it looks. Well, the simplest way to get it working is

chattr -SAadiscu /dev
rm /dev/console
mknod /dev/console c 5 1
chmod 700 /dev/console

this way /dev/console will be recreated and attributes of parent directory "/dev" will be inherited.

Quoted 32 lines

SunWave1 /# l /dev/console
crw------- 1 root root 5, 1 Nov 21 11:01 /dev/console
SunWave1 /# lsattr /dev/console
lsattr: Invalid argument While reading flags on /dev/console

Two other FS in /dev have some problems, too. But I think that is right.

SunWave1 /# chattr -R -SAadiscu /dev/
chattr: Inappropriate ioctl for device while reading flags on /dev//pts
chattr: Inappropriate ioctl for device while reading flags on /dev//shm

All the files in /proc show the same.

Here comes a little positive example, I think.

SunWave1 /# lsattr /tmp/
----------X-j /tmp//xwlog
--------B-X-j /tmp//kde-nuetzel
------------- /tmp//dcop6SKhoM
---------D-Ej /tmp//YaST2.tdir
----------XEj /tmp//ksocket-root
--------B-X-j /tmp//mcop-nuetzel
--------B--Ej /tmp//kde-root
--------B-X-j /tmp//ksocket-nuetzel
--------B--Ej /tmp//INSTALL
--------BD-Ej /tmp//modules.conf.-
---------DX-j /tmp//vi.recover
--------BD-Ej /tmp//mcop-root
------------j /tmp//lost+found
--------BD-Ej /tmp//modules.conf

Quoted 6 lines

What next?

Thanks,
Dieter

PS Have you read about the latest ACL discussion on LKML?

Yes, but Hans thinks we shouldn't do ACL reiserfs 3.x and concentrate on v4 in stead. You can try to persuade him though. Same for extended attributes.

Nikita.

#6 lore raw
In reply to Dieter Nützel
Quoted 9 lines · Dieter Nützel

Am Mittwoch, 21. November 2001 11:02 schrieb Nikita Danilov:

Quoted 6 lines

Yes, it's right, but currently we have what we have currently. I am going to extend inode-attrs.patch and add new mount option "noattrs". With it ioctls to set and get attributes will continue to work, but attributes themselves will not have any effect. Then, one can boot with "rootflags=noattrs" and read-write root, clear all attributes by chattr -R and remount root.

Quoted 3 lines

After I've 'booted into single user mode with "linux rootflags=noattrs" I
could change my archives but _NOT_ /dev/console the right way. After the
second reboot I got the warning and kernel hang, again. Something wrong with your patch or my installation?

Erm, while disabling set and get as a nice idea, you only need to disable get. Disabling set will mean you 'cannot' reset the file that is causing the problem, because chattr won't actually be able to do what you want.

I haven't looked at the code that chattr uses either, but you may want to check at HOW it sets/resets values.

eg: Does it read in the value, change the bit and write it out again, or does it just change that bit? If it reads and get is disabled, well then you would want to make sure you set the 'full value' that needs to be there. Not a bug so much as something that just needs to be documented.

AMC Enterprises P/L    - Stuart Young
First Floor            - Network and Systems Admin
3 Chesterville Rd      - [email protected]
Cheltenham Vic 3192    - Ph:  (03) 9584-2700

http://www.amc.com.au/ - Fax: (03) 9584-2755

#7 lore raw
In reply to a message not in the archive
Quoted 6 lines · Nikita Danilov
Quoted 1 line · Dieter N�tzel

PS Have you read about the latest ACL discussion on LKML?

Yes, but Hans thinks we shouldn't do ACL reiserfs 3.x and concentrate on v4 in stead. You can try to persuade him though. Same for extended attributes.

I started to write ACL support based on Andreas Gr�nbacher's patches at
http://acl.bestbits.at a few months ago. Since there was no decided upon format, I stopped development so that I wouldn't leave non-standard filesystems in my wake. Once the ACL/EA format/API is finalized, I'll probably continue development on it.

On a more technical note, adding a new item type, such as an extended attribute item will cause problems for implementations that don't know about them. An idea kicked around a few months ago was to simply change the version string so that implementations that don't know about ACL support can't mount the new version. "can't mount" could conceivably be changed to "can only mount read-only", but certainly not read-write, since non-solid items need special handling (every item but stat data is non-solid).

This creates the following scenario:
* Versions that don't know about the new version string will refuse to mount the fs.
* Versions that know about the new version, but don't implement support will be read-only. This is because items get moved during balancing, even to un-related items. Knowledge of the inside non-solid items is required to move them around. So, even if there is a single ACL on the system, in some deep directory, it could affect balancing (cause a panic).
* Versions that know about the new version, and implement support will be read-write.

I see something along the lines of EXT2's compatibility matrix happening here, but due to the inherent incompatibilities, this may not be a prime candidate for 2.4 inclusion.

-Jeff

Signature · 3 lines
Jeff Mahoney           |   "Bill Gates is a monocle and a Persian cat away
[email protected]         |    from being the villain in a James Bond movie."
[email protected]      |                   -- Dennis Miller
#8 lore raw
In reply to a message not in the archive
Quoted 4 lines · Andreas Dilger

In any case, it is also a bad thing to leave garbage in unused parts of on-disk data structs for just this reason, so mkreiserfs should zero everything that is unused inside allocated structs (and the kernel too, because reiserfs allocates inode tables dynamically, right?).

Also reiserfsck? It would be nice if it cleaned too...

Ed Tomlinson