bug#68283: ls -l issues on cifs mounted directories

Hey Bjoern,

Bjoern Voigt wrote:

> After upgrading coreutils from version 9.3 to 9.4, "ls -l" shows error 
> messages, if the files or directories are in a cifs/smb3 mounted directory.
>
> Example:
> /mnt/cifstest here is mounted with "mount -t cifs -o username=myuser 
> //192.168.1.2/all /mnt/cifstest/". The server is a Samba 4.19.2 server. 
> The kernel is version 6.1.71.

> # ls -l /mnt/cifstest/
> ls: /mnt/cifstest/: Resource temporarily unavailable
> ls: /mnt/cifstest/opt: Resource temporarily unavailable
> [...]
>
> "ls -l" from coreutils 9.3 does not show any errors.
> 
> Probably it is not just the error message. "dmesg -w" shows several 
> errors after executing "ls -l":
> 
> [17563.999109] smb2_reconnect: 3 callbacks suppressed
> [17563.999113] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.160117] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.192244] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.223303] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.256080] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.287887] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.410323] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.442095] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.474065] CIFS: VFS: reconnect tcon failed rc = -11
> [17564.569435] CIFS: VFS: reconnect tcon failed rc = -11
> 
> I debugged this problem a bit. The changes in 
> coreutils-9.4/lib/file-has-acl.c:file_has_acl are probably responsible 
> for the errors. The system call listxttr produces the error message 
> "Resource temporarily unavailable". I could not find the reasons for 
> error messages shown with "dmesg".
> 
> The listxattr.c test program in the manual page (man listxattr) shows 
> the following output:
> 
> # ./listxattr /mnt/cifstest/
> listxattr: Resource temporarily unavailable
> 
> Work-around are to downgrade to coreutils 9.3 or not to use the "-l" 
> option for ls.

Thanks for the details and analysis.

Pádraig Brady suggested this may have been a CIFS bug at the time:

> Thanks for the details.
> EAGAIN is not a documented return from listxattr,
> and it seems like the cifs client should not let that leak.
> [...]
> Perhaps something to ask at linux-cifs <at> vger.kernel.org ?

I am two years late (sorry), but I see another user reported the same
symptoms on linux-cifs a few days after your email [1]. There the
culprit was statfs and listxattr, as you noticed. There is also a GitHub
thread linked where someone who ran into the same issue noted the same
'dmesg' logs you did [2].

From the linux-cifs thread, it looks like the issue got fixed and
backported quickly. So, I don't think any changes in coreutils are
needed. I'll go ahead and close this.

Thanks again for the well written report.

Collin

[1] https://lore.kernel.org/regressions/[email protected]/
[2] https://github.com/home-assistant/operating-system/issues/3041#issuecomment-1878989390