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;
From:   Zoltan Karcagi <zkr7432@gmail.com>
Subject: BUG: "Stale file handle" error when connecting to ARM server
To:     linux-nfs@vger.kernel.org
Message-ID: <5bcd51ef-9ffb-2650-108f-8d7b04beb655@gmail.com>
Date:   Fri, 26 Jul 2019 13:45:34 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.8.0
MIME-Version: 1.0
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 7bit
Content-Language: en-US
Sender: linux-nfs-owner@vger.kernel.org
Precedence: bulk

Hi,

Upgrading the nfs-utils package from 2.3.4 to 2.4.1 on my nas device
running Arch Linux ARM (alarm) broke the nfs server functionality.
On all of my clients, mount fails with a "Stale file handle" error. On the
server, I get this (side note: the reported version number is wrong):

Jul 08 10:28:50 nas rpc.mountd[19752]: Version 2.3.4 starting
Jul 08 10:30:08 nas rpc.mountd[19752]: auth_unix_ip: inbuf 'nfsd <redacted valid ipv6 address>'
Jul 08 10:30:08 nas rpc.mountd[19752]: auth_unix_ip: client 0x492c50 'zero.local'
Jul 08 10:30:08 nas rpc.mountd[19752]: nfsd_fh: inbuf 'zero.local 1 \x00000000'
Jul 08 10:30:08 nas rpc.mountd[19752]: nfsd_fh: found 0x49b698 path /srv/nfs
Jul 08 10:30:08 nas rpc.mountd[19752]: nfsd_export: inbuf 'zero.local /srv/nfs/tmp2'
Jul 08 10:30:09 nas rpc.mountd[19752]: nfsd_export: found 0x499ed0 path /srv/nfs/tmp2
Jul 08 10:30:09 nas rpc.mountd[19752]: nfsd_fh: inbuf 'zero.local 7 \x0100140000000000ae18d6965c0a40f78701c770897a4fc>
Jul 08 10:30:09 nas rpc.mountd[19752]: nfsd_fh: found (nil) path (null)

Analysis:

Consider this code snippet from utils/mountd/cache.c:
 627 static bool match_fsid(struct parsed_fsid *parsed, nfs_export *exp, char *path)
 628 {
 629         struct stat stb;
 630         int type;
 631         char u[16];
 632
 633         if (nfsd_path_stat(path, &stb) != 0)
 634                 return false;

Variable stb gets defined, then gets filled by nfs_path_stat(), which is
implemented in support/misc/nfsd_path.c. At least on alarm, definition of
struct stat in stat.h depends on __USE_FILE_OFFSET64, which comes from
config.h if defined. This requires config.h to be included before stat.h.

The include order is right in cache.c, however, it's reversed in 
nfsd_path.c. This causes the data returned by nfs_path_stat() to be in a
different structure than expected, and that's what eventually causes the
error.    

Proposed solution:

The following patch fixes those occurrences where the include order
between config.h and stat.h is wrong, by moving config.h to the top.


Regards,
Zoltan Karcagi