GNU bug report logs - #14618
cp: erroneous sparse copy command results in empty file

Previous Next

Full log

View this message in rfc822 format

From: Pádraig Brady <P <at> draigBrady.com>
To: Koen De Wit <koen.de.wit <at> oracle.com>
Cc: 14618 <at> debbugs.gnu.org
Subject: bug#14618: Erroneous sparse copy command results in empty file
Date: Fri, 14 Jun 2013 21:03:52 +0100

On 06/14/2013 12:35 PM, Koen De Wit wrote:
> When trying to create a sparse copy of a file on a different filesystem you get an "Invalid cross-device link" error, as expected. However, an empty file is created instead of the sparse copy:

As a side note, sparse copying is a separate thing to reflinking.

> # mkfs.btrfs -f /dev/sde1
> # mkfs.btrfs -f /dev/sde2
> # mount /dev/sde1 /mnt/part1
> # mount /dev/sde2 /mnt/part2
> # echo "abc" > /mnt/part1/testfile
> # ./cp --reflink /mnt/part1/testfile /mnt/part2/testcopy
> ./cp: failed to clone ‘/mnt/part2/testcopy’ from ‘/mnt/part1/testfile’: Invalid cross-device link
> # ls -la /mnt/part1/testfile
> -rw-r--r--. 1 root root 4 Jun 14 14:49 /mnt/part1/testfile
> # ls -la /mnt/part2/testcopy
> -rw-r--r--. 1 root root 0 Jun 14 14:49 /mnt/part2/testcopy
> 
> The same occurs when trying to create a sparse copy on a filesystem that does not support sparse copies:
> 
> # mkfs.ext3 /dev/sde1
> # mount /dev/sde1 /mnt/part1
> # echo "abc" > /mnt/part1/testfile
> # ./cp --reflink /mnt/part1/testfile /mnt/part1/testcopy
> ./cp: failed to clone ‘/mnt/part1/testcopy’ from ‘/mnt/part1/testfile’: Inappropriate ioctl for device
> # ls -la /mnt/part1
> (...)
> -rw-r--r--. 1 root root 0 Jun 14 14:58 testcopy
> -rw-r--r--. 1 root root 4 Jun 14 14:58 testfile
> 
> If creating a sparse copy fails, no file should be created.
> 
> ln treats this error in the right way:
> 
> # ln /mnt/part1/testfile /mnt/part2/testlink
> ln: creating hard link `/mnt/part2/testlink' => `/mnt/part1/testfile': Invalid cross-device link
> # ls -la /mnt/part2/testlink
> ls: cannot access /mnt/part2/testlink: No such file or directory
> 
> Tested with coreutils today's git HEAD and kernel 3.9.5.

This was discussed previously:
http://lists.gnu.org/archive/html/coreutils/2013-03/msg00056.html
Summary of that is that it may be best handled outside of cp.

cheers,
Pádraig.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.