large file problem in rh7.1

[bogstad at pobox.com (Bill Bogstad)]

>...
>The sysadmin at work seemed to think it might be some funky command char
>sitting in the file. He suggested using truss when trying to cp the file.
>He's a Solaris guy :) but we found out we could use strace in rh. The
>command we used was:

That's an interesting theory, but I think that he has misinterpreted
the strace.  Let me provide an alternate explanation for the output
you are seeing.  Unfortunately, I still can't figure out why it's
failing...

>strace cp ali.mp3 alinew.mp3
>
>The last bits of the strace output were:
>
>read(3, "6\212o\7\204(fw\36\315\367O\336e\255E\321a\27ppU\313\201"...,
>4096) = 4096
>write(4, "6\212o\7\204(fw\36\315\367O\336e\255E\321a\27ppU\313\201"...,
>4096) = 4096

successful read/write of 4096 bytes from ali.mp3 to alinew.mp3

>read(3, "\23.\tN\n\16!\6\253\270T\262\30\241\325]\363\362\307\f"..., 4096)
>= 4096
>write(4, "\23.\tN\n\16!\6\253\270T\262\30\241\325]\363\362\307\f"...,
>4096) = 4096

successful read/write of 4096 bytes from ali.mp3 to alinew.mp3

>read(3, 0xbfffe420, 4096)               = -1 EIO (Input/output error)

failed read of ali.mp3.  Errno is set to EIO which is a generic IO
error.

>open("/usr/share/locale/en_US/LC_MESSAGES/fileutils.mo", O_RDONLY) = -1
>ENOENT (No such file or directory)
>open("/usr/share/locale/en/LC_MESSAGES/fileutils.mo", O_RDONLY) = -1
>ENOENT (No such file or directory)

Attempts to open locale/language specific message files to get
appropriate error message fails.

>write(2, "cp: ", 4cp: )                     = 4
>write(2, "reading `ali.mp3\'", 17reading `ali.mp3')      = 17

cp falls back to default builtin message.

>open("/usr/share/locale/en_US/LC_MESSAGES/libc.mo", O_RDONLY) = -1 ENOENT
>(No such file or directory)
>open("/usr/share/locale/en/LC_MESSAGES/libc.mo", O_RDONLY) = -1 ENOENT (No
>such file or directory)

Attempts to open locale/language specific message file to get
appropriate error message fails.

>write(2, ": Input/output error", 20: Input/output error)    = 20
>write(2, "\n", 1
>)                       = 1

cp falls back to default builtin message.

(BTW, you are getting strace output and error output from the cp
command itself fixed together here and in the previous "write(2 ..."
so that's a little confusing.  You might try using the '-o' option if
you use strace in the future.)

>close(4)                                = 0
>close(3)                                = 0
>_exit(1)                                = ?

>
>(Preceded by many more read/write statements.) The sysadmin seemed to
>think that some char was somehow sending a message to open the file
>/usr/share/locale/en_US/LC_MESSAGES/fileutils.mo. I don't have this file,
>so things were getting confused.

The important thing is the failed read which returned EIO.  Everything
else is a cascade as a result of this.  Unfortunately, strace isn't
telling me anything useful.  You say that syslog isn't recording
anything useful either.  This leaves me confused.  On the chance that
syslog is for some reason losing messages, you might try reading the
entire disk partition and see if that works.  Something like:

dd bs=100k < /dev/hda5 > /dev/null

Eventually dd should end with a report that it read N+M blocks where N
is the number of blocks and M is the number of bytes in the final
partial block.  If the number of blocks doesn't match up closely with
the results from fsck

>/: 104153/320640 files (1.5% non-contiguous), 496291/640584 blocks

then you might still have a hardware problem.

			  Good Luck,
			  Bill Bogstad
			  bogstad at pobox.com