2.6.1: data corrupton when recieving files > 1GB over network
Hello,
When I transfer files to my linux 2.6.1 box their content changes (tested via md5 sums).
I transfered a 1,8 GB (mpeg) file serveral times to this machine by using either pure-ftpd (upload) or wget (download) on that machine. I got a different md5sum each time. Same problem with a 1,4 GB (zip) file, but *not* with a 0,7 GB (mpeg) file.
When I boot the machine with Knoppix 3.2 (Linux 2.4.21-xfs) and upload the 1,8 GB file to it's ftpd (same target harddisk/partition/directory), the file is ok.
When I dupplicate the correctly recieved file with dd or cp under Linux 2.6.1 there is no corruption, too.
I don't know what tools I should use to determine at what positions these corruptions start and how much is corrupted. But I think about the first 1 GB is transfered correctly (diff needs some time before it says "Binary files test-2.6.mpeg and test-2.4.mpeg differ")
Kernel is built without module support.
[ Some lines from dmesg ]
Linux version 2.6.1 (stob@netbrake) (gcc version 2.95.4 20011002 (Debian prerelease)) #5 Sat Jan 10 01:40:00 CET 2004
CPU: Intel Pentium III (Katmai) stepping 02
agpgart: Detected VIA Apollo Pro 133 chipset
eth0: RealTek RTL8139 at 0xe3818000, 00:00:21:d5:a6:48, IRQ 10
eth0: Identified 8139 chip type 'RTL-8139B'
hda: Maxtor 98196H8, ATA DISK drive
hda: max request size: 128KiB
hda: 160086528 sectors (81964 MB) w/2048KiB Cache, CHS=65535/16/63, UDMA(33)
hda: hda1
EXT3 FS on hda1, internal journal
[ Output of scripts/ver_linux ]
Linux netbrake 2.6.1 #5 Sat Jan 10 01:40:00 CET 2004 i686 unknown
Gnu C 2.95.4 Gnu make 3.80 util-linux 2.12 mount 2.12 module-init-tools implemented e2fsprogs 1.34 Linux C Library 2.2.5 Dynamic linker (ldd) 2.2.5 Procps 3.1.15 Net-tools 1.60 Console-tools 0.2.3 Sh-utils 2.0.11
Quoted 4 lines · Hans Spath
I don't know what tools I should use to determine at what positions these corruptions start and how much is corrupted. But I think about the first 1 GB is transfered correctly (diff needs some time before it says "Binary files test-2.6.mpeg and test-2.4.mpeg differ")
Try "cmp -l".
Signature · 2 lines
Måns Rullgård [email protected]
I found a strange behavior with md5sum and nfs.
I've a nfs server (mistur) under debian woody 2.4.18, and a client (vaka) under debian sid 2.6.1-mm2 or 2.4.18-bf2.4 (onlu for test)
Linux vaka 2.6.1-mm2 #1 Sat Jan 10 19:16:02 CET 2004 i686 GNU/Linux Linux vaka 2.4.18-bf2.4 #1 Son Apr 14 09:53:28 CEST 2002 i686 GNU/Linux
Linux mistur
2.4.18-1-686 #4 Sat Nov 29 10:18:26 EST 2003 i686 unknown
vaka : dpkg 1.10.18
mistur : dpkg 1.9.21
nfs export :
path on client -> path on server/mnt/multimedia -> /mnt/multimedia/mnt/tmp -> /tmp
I've done some test with md5sum on the iso of woody 3.0r0 CD1
same file : /mnt/multimedia/iso
- mistur local (ext3) 2.4.18-1-686 :
ca679b3b8e28be00b98b007611384958 /mnt/multimedia/iso/woody-cd1.iso
- vaka remote (nfs) 2.4.18-bf2.4 :
ea34f974bdcfb2a678a97afb1fb4077d /mnt/multimedia/iso/woody-cd1.iso
- vaka remote (nfs) 2.6.1-mm2 :
ca679b3b8e28be00b98b007611384958 /mnt/multimedia/iso/woody-cd1.iso 2798b3e2b97ca8082049c9207c291ebb /mnt/multimedia/iso/woody-cd1.iso
copy on vaka in /usr/src/test/ using nfs
- vaka local (ext3) 2.4.18-bf2.4 :
287ffb166b1e32a75ad106ceffb1dc3f /usr/src/test/woody-cd1.iso
- vaka local (ext3) 2.6.1-mm2 :
287ffb166b1e32a75ad106ceffb1dc3f /usr/src/test/woody-cd1.iso
copy on mistur in /mnt/tmp using nfs
- vaka remote (nfs) 2.4.18-bf2.4 :
102fbde335166ccd980f672370fc3a4b /mnt/tmp/test/woody-cd1.iso
- vaka remote (nfs) 2.6.1-mm2 :
bd52d79225a863e484c2d1815b5c5b7c /mnt/tmp/test/woody-cd1.iso
- mistur local (ext3) 2.4.18-1-686 :
287ffb166b1e32a75ad106ceffb1dc3f /tmp/test/woody-cd1.iso
all copy was done under 2.6.1-mm2 kernel
I could do more test, all under 2.4.X and all under 2.6.X
- copy with nfs, with ftp,...
- using différent versiob of md5sum
but, not enouph time now
anyway seem to have a problem
Yoann
Quoted 42 lines · Hans Spath
Hello,
When I transfer files to my linux 2.6.1 box their content changes (tested via md5 sums).
I transfered a 1,8 GB (mpeg) file serveral times to this machine by using either pure-ftpd (upload) or wget (download) on that machine. I got a different md5sum each time. Same problem with a 1,4 GB (zip) file, but *not* with a 0,7 GB (mpeg) file.
When I boot the machine with Knoppix 3.2 (Linux 2.4.21-xfs) and upload the 1,8 GB file to it's ftpd (same target harddisk/partition/directory), the file is ok.
When I dupplicate the correctly recieved file with dd or cp under Linux 2.6.1 there is no corruption, too.
I don't know what tools I should use to determine at what positions these corruptions start and how much is corrupted. But I think about the first 1 GB is transfered correctly (diff needs some time before it says "Binary files test-2.6.mpeg and test-2.4.mpeg differ")
Kernel is built without module support.
[ Some lines from dmesg ]
Linux version 2.6.1 (stob@netbrake) (gcc version 2.95.4 20011002 (Debian prerelease)) #5 Sat Jan 10 01:40:00 CET 2004
CPU: Intel Pentium III (Katmai) stepping 02
agpgart: Detected VIA Apollo Pro 133 chipset
eth0: RealTek RTL8139 at0xe3818000, 00:00:21:d5:a6:48, IRQ 10
eth0: Identified 8139 chip type 'RTL-8139B'
hda: Maxtor 98196H8, ATA DISK drive
hda: max request size: 128KiB
hda: 160086528 sectors (81964 MB) w/2048KiB Cache, CHS=65535/16/63, UDMA(33)
hda: hda1
EXT3 FS on hda1, internal journal[ Output of
scripts/ver_linux]
Linux netbrake 2.6.1 #5 Sat Jan 10 01:40:00 CET 2004 i686 unknownGnu C 2.95.4 Gnu make 3.80 util-linux 2.12 mount 2.12 module-init-tools implemented e2fsprogs 1.34 Linux C Library 2.2.5 Dynamic linker (ldd) 2.2.5 Procps 3.1.15 Net-tools 1.60 Console-tools 0.2.3 Sh-utils 2.0.11
På su , 11/01/2004 klokka 15:16, skreiv yoann:
Quoted 11 lines
same file :
/mnt/multimedia/iso- mistur local (ext3) 2.4.18-1-686 :
ca679b3b8e28be00b98b007611384958 /mnt/multimedia/iso/woody-cd1.iso- vaka remote (nfs) 2.4.18-bf2.4 :
ea34f974bdcfb2a678a97afb1fb4077d /mnt/multimedia/iso/woody-cd1.iso- vaka remote (nfs) 2.6.1-mm2 :
ca679b3b8e28be00b98b007611384958 /mnt/multimedia/iso/woody-cd1.iso 2798b3e2b97ca8082049c9207c291ebb /mnt/multimedia/iso/woody-cd1.iso
Have you tried running memtest86 on these machines?
Also, is all this being done using the same 2.4.18-1 server on 'mistur', or are you also using a server on 'vaka'? If the former, have you tried changing servers? (BTW: is this the kernel server, or are you using the userland nfs-server daemon).
Cheers,
Trond
Quoted 5 lines
Quoted 5 lines
same file :
/mnt/multimedia/iso- mistur local (ext3) 2.4.18-1-686 :
ca679b3b8e28be00b98b007611384958/mnt/multimedia/iso/woody-cd1.iso
Quoted 8 lines
Quoted 6 lines
- vaka remote (nfs) 2.4.18-bf2.4 :
ea34f974bdcfb2a678a97afb1fb4077d/mnt/multimedia/iso/woody-cd1.iso- vaka remote (nfs) 2.6.1-mm2 :
ca679b3b8e28be00b98b007611384958/mnt/multimedia/iso/woody-cd1.iso2798b3e2b97ca8082049c9207c291ebb/mnt/multimedia/iso/woody-cd1.isoHave you tried running memtest86 on these machines?
I will... but I've not problem and using a lot the 384Mo of RAM
Quoted 2 lines
Also, is all this being done using the same 2.4.18-1 server on 'mistur', or are you also using a server on 'vaka'?
unsig the same 2.4.18 on mistur and vaka stays nfs client
Quoted 2 lines
If the former, have you tried changing servers? (BTW: is this the kernel server, or are you using the userland nfs-server daemon).
kernelland
I haven't try in userland
Quoted 2 lines
Cheers,
Trond
Yoann
På su , 11/01/2004 klokka 16:03, skreiv yoann:
Quoted 6 lines
Quoted 2 lines
If the former, have you tried changing servers? (BTW: is this the kernel server, or are you using the userland nfs-server daemon).
kernelland
I haven't try in userland
Try swapping the 2.4.18 kernel on mistur for something more recent. AFAICS that server is the common element in your test.
Cheers,
Trond
Quoted 3 lines · On Sun, Jan 11, 2004 at 05:36:32PM +0100, I wrote:
When I transfer files to my linux 2.6.1 box their content changes (tested via md5 sums).
[no problem with Knoppix 3.2 (Linux 2.4.21-xfs)]
I've finally found the problem. It was a *hardware* problem.
The problem occured only with DMA activated for harddisks and the Knoppix kernel doesn't do this by default.
(It was a bit strange, because it happend only when writing to disk while having high network traffic, even after switching NIC and IDE controler.)
I've exchanged the mainboard and now everything is working perfectly.
Sorry for wasting your time, folks.