GNU bug report logs - #32393
coreutils-8.30 'mv' error for file on libfuse mounted shares

Previous Next

To add a comment to this bug, you must first unarchive it, by sending
a message to control AT debbugs.gnu.org, with unarchive 32393 in the body.
You can then email your comments to 32393 AT debbugs.gnu.org in the normal way.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: John Stanley <jpsinthemix <at> verizon.net>
To: bug-coreutils <at> gnu.org
Subject: coreutils-8.30 'mv' error for file on libfuse mounted shares
Date: Wed, 8 Aug 2018 00:30:55 -0400

Hi,
With coreutils-8.30, I'm getting a strange error for the 'mv' command 
when used to rename a file mounted as a 'fuse' share:

$ mv ~/mnt/sshfs/status-2018080{6,7}
mv: cannot move '/home/john/mnt/sshfs/status-20180806' to 
'/home/john/mnt/sshfs/status-20180807': Numerical result out of range

$ mv ~/mnt/sshfs/qq ~/mnt/sshfs/qqq
mv: cannot move '/home/john/mnt/sshfs/qq' to '/home/john/mnt/sshfs/qqq': 
Numerical result out of range

This does not occur for coreutils-8.29, and appears to be associated 
with the SYS_renameat2 syscall added in version 8.30. It also does not 
occur for 'local' file renaming.

The error "Numerical result out of range" looks like it may be coming 
from the network stack via 'fuse.'

I realize that this error may be coming from a number of places other 
than coreutils 'mv' (e.g., the kernel/fuse area, sshfs, or libfuse, but 
I'm wondering if anyone intimately familiar with the SYS_renameat2 
rework in coreutils-8.30 'mv' might have any ideas on how to pin this 
one done.

*** My System Info:
  x86_64-pc-linux-gnu  (This is a freshly re-built source-based system)
  linux-headers.4.17.13_x86_64_smp1, gcc.8.2.0, glibc.2.27, 
binutils.2.31.1, coreutils.8.30

thanks much,
John

Message #8 received at 32393 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: John Stanley <jpsinthemix <at> verizon.net>, 32393 <at> debbugs.gnu.org
Subject: Re: bug#32393: coreutils-8.30 'mv' error for file on libfuse mounted
 shares
Date: Tue, 7 Aug 2018 23:24:14 -0700

John Stanley wrote:
> I'm wondering if anyone intimately familiar with the SYS_renameat2 rework in 
> coreutils-8.30 'mv' might have any ideas on how to pin this one done.

Sure, run this:

strace mv ~/mnt/sshfs/status-2018080{6,7}

and look at the strace output.

Message #11 received at 32393 <at> debbugs.gnu.org (full text, mbox):

From: John Stanley <jpsinthemix <at> verizon.net>
To: Paul Eggert <eggert <at> cs.ucla.edu>, 32393 <at> debbugs.gnu.org
Subject: Re: bug#32393: coreutils-8.30 'mv' error for file on libfuse mounted
 shares
Date: Wed, 8 Aug 2018 04:21:11 -0400

[Message part 1 (text/plain, inline)]

Already did..should've included it. It's not particularly revealing to 
me though (I've attached the trace). The syscall goes as:

renameat2(AT_FDCWD, "/home/john/mnt/sshfs/status-20180806", AT_FDCWD, 
"/home/john/mnt/sshfs/status-20180807", RENAME_NOREPLACE) = -1 ERANGE 
(Numerical result out of range)

Also, I've run some a quick 'n dirty code to simply do the rename 
syscall which gives the same result.
But, if in this code I simply invoke glibc's  renameat() (which is 
renameat2() less the "flags" arg), I do not get any errors.  So perhaps 
the issue is related to the RENAME_NOREPLACE flag. In fact, with my test 
code, if I do the syscall renameat2 with flags = 0, then also no error.

Note that in my test cases the source and destinations are regular files 
and the destination does not exist. If the destination does exist, then 
I get a different error:

 mv  ~/mnt/sshfs/status-20180806 ~/mnt/sshfs/status-20180807
error: 17 File exists

which, of course, does not occur for "local" files.



Paul Eggert wrote:
> John Stanley wrote:
>> I'm wondering if anyone intimately familiar with the SYS_renameat2 
>> rework in coreutils-8.30 'mv' might have any ideas on how to pin this 
>> one done.
>
> Sure, run this:
>
> strace mv ~/mnt/sshfs/status-2018080{6,7}
>
> and look at the strace output.
>

[mv.mtrace (text/plain, attachment)]

Message #14 received at 32393 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: John Stanley <jpsinthemix <at> verizon.net>, 32393 <at> debbugs.gnu.org
Subject: Re: bug#32393: coreutils-8.30 'mv' error for file on libfuse mounted
 shares
Date: Wed, 8 Aug 2018 02:31:54 -0700

John Stanley wrote:
> perhaps the issue is related to the RENAME_NOREPLACE flag
My guess is that you have a fuse callback that's returning a positive value for 
an error number. It's supposed to return negative values. See:

https://sourceforge.net/p/fuse/mailman/message/35616558/

If my guess is right, you need to fix your fuse callback. I suggest starting by 
looking at this recent fix to sshfs:

https://github.com/libfuse/sshfs/commit/6480b66bd64a2d2f58db6fc0a3320f4464f4669e

The bottom line is that if you're playing with bleeding-edge and buggy 
filesystems, that's the place to look for bugs.

Message #17 received at 32393 <at> debbugs.gnu.org (full text, mbox):

From: John Stanley <jpsinthemix <at> verizon.net>
To: Paul Eggert <eggert <at> cs.ucla.edu>, 32393 <at> debbugs.gnu.org
Subject: Re: bug#32393: coreutils-8.30 'mv' error for file on libfuse mounted
 shares
Date: Wed, 8 Aug 2018 12:35:22 -0400

Can't thank you enough.. the return value addressed in the github link 
is indeed the culprit.

Paul Eggert wrote:
> John Stanley wrote:
>> perhaps the issue is related to the RENAME_NOREPLACE flag
> My guess is that you have a fuse callback that's returning a positive 
> value for an error number. It's supposed to return negative values. See:
>
> https://sourceforge.net/p/fuse/mailman/message/35616558/
>
> If my guess is right, you need to fix your fuse callback. I suggest 
> starting by looking at this recent fix to sshfs:
>
> https://github.com/libfuse/sshfs/commit/6480b66bd64a2d2f58db6fc0a3320f4464f4669e 
>
>
> The bottom line is that if you're playing with bleeding-edge and buggy 
> filesystems, that's the place to look for bugs.
>

Message #22 received at 32393-done <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: John Stanley <jpsinthemix <at> verizon.net>, 32393-done <at> debbugs.gnu.org
Subject: Re: bug#32393: coreutils-8.30 'mv' error for file on libfuse mounted
 shares
Date: Wed, 8 Aug 2018 14:18:36 -0700

John Stanley wrote:
> the return value addressed in the github link is indeed 
> the culprit.

Thanks for checking; closing the bug.

Previous Next

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