Received-SPF: pass (google.com: best guess record for domain of linux-nfs-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
In-Reply-To: <20191011170709.GE19318@fieldses.org>
To:     linux-nfs@vger.kernel.org
Cc:     "J. Bruce Fields" <bfields@redhat.com>,
        Thomas Deutschmann <whissi@gentoo.org>,
        Steve Dickson <steved@redhat.com>
From:   Doug Nazar <nazard@nazar.ca>
Subject: Re: nfs-utils: v3 mounts broken due to statx() returning EINVAL
Message-ID: <dd522611-f8eb-bb94-9ec6-09a6725e8cca@nazar.ca>
Date:   Tue, 26 Nov 2019 02:22:40 -0500
User-Agent: Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:71.0) Gecko/20100101
 Thunderbird/71.0
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Content-Language: en-US
Sender: linux-nfs-owner@vger.kernel.org
Precedence: bulk

On Fri, Oct 11, 2019 at 17:07:09PM +0000, Bruce Fields wrote:
 >On Tue, Oct 08, 2019 at 10:23:56PM +0200, Thomas Deutschmann wrote:
 >> Hi,
 >>
 >> we have some user reporting that NFS v3 mounts are broken
 >> when using glibc-2.29 and linux-4.9.x (4.9.128) because
 >> statx() with mask=STATX_BASIC_STATS returns EINVAL.
 >>
 >> Looks like this isn't happening with <nfs-utils-2.4.1 or
 >> newer kernels.
 >>
 >> The following workaround was confirmed to be working:
 >>
 >> --- a/support/misc/xstat.c    2019-06-24 21:31:55.260371592 +0200
 >> +++ b/support/misc/xstat.c    2019-06-24 21:32:29.098777436 +0200
 >> @@ -47,6 +47,8 @@
 >>              statx_copy(statbuf, &stxbuf);
 >>              return 0;
 >>          }
 >> +        if (errno == EINVAL)
 >> +            errno = ENOSYS;
 >>          if (errno == ENOSYS)
 >>              statx_supported = 0;
 >>      } else
 >>
 >>
 >> Bug: https://bugs.gentoo.org/688644
 >>
 >> At the moment I have no clue whether this is kernel/glibc or
 >> nfs-utils related; if the patch is safe to apply...
 >
 >Well, sounds like nfs-utils started using statx in 2.4.1.  And just the
 >fact that varying the kernel version makes it sound like there was a
 >kernel bug causing an EINVAL return in this case, and that bug got
 >fixed.
 >
 >One way to confirm might be running mount under strace and looking for
 >that EINVAL return.

Just to provide an update on this issue, it was tracked down to glibc's 
statx emulation not supporting AT_STATX_DONT_SYNC and returning EINVAL.

So on a kernel without statx support, but a new glibc with statx 
support, nfs-utils will always fail to stat any paths.

Either this or a similar fix is required to support older kernels.

Doug