Message-ID: <556C73AE.4090900@oracle.com>
Date: Mon, 01 Jun 2015 18:01:02 +0300
From: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
MIME-Version: 1.0
To: linux-nfs@vger.kernel.org
CC: Vasily Isaenko <vasily.isaenko@oracle.com>,
        "SHUANG.QIU" <shuang.qiu@oracle.com>
Subject: nfsd: EACCES vs EPERM on utime()/utimes() calls
Content-Type: text/plain; charset=utf-8; format=flowed
Sender: linux-nfs-owner@vger.kernel.org

Hello.

As the man page for utime/utimes states [1], EPERM is returned if the 
second argument of utime/utimes is not NULL and:
  * the caller's effective user id does not match the owner of the file
  * the caller does not have write access to the file
  * the caller is not privileged

However, I don't see this behavior with NFS, I see EACCES is generated 
instead.

Please, consider this test:

#include <stdio.h>
#include <sys/types.h>
#include <sys/time.h>
#include <utime.h>

int main(int argc, char *argv[])
{
	struct utimbuf u = { .actime = 0, .modtime = 0 };
	struct timeval tv = { .tv_sec = 0, .tv_usec = 0 };

	if (utime(argv[1], &u))
		perror("utime() failed");

	if (utimes(argv[1], &tv))
		perror("utimes() failed");
	
	return 0;
}

In my environment the kernel is 4.1.0-rc6 x86_64, and there are 2 NFS 
mounted directories:
127.0.0.1:/opt/export/disk0 on /mnt/lin type nfs (rw,vers=3,addr=127.0.0.1)
192.168.0.12:/export/bla on /mnt/sol type nfs (rw,vers=3,addr=192.168.0.12)

/mnt/sol is from Solaris 11.2 x86_64. /opt/export/disk0 is handled by 
the in-kernel nfs server.

Execution of the above test program gives:

* the server is Linux -> EACCES is generated:

[stas@ol6-x64 tmp]$ ls -l /mnt/lin/test_file
-rw-r--r-- 1 root root 0 Jun  1 17:28 /mnt/lin/test_file
[stas@ol6-x64 tmp]$ strace -e utime,utimes ./utime_test /mnt/lin/test_file
utime("/mnt/lin/test_file", [0, 0])     = -1 EACCES (Permission denied)
utime() failed: Permission denied
utimes("/mnt/lin/test_file", {{0, 0}, {0, 0}}) = -1 EACCES (Permission 
denied)
utimes() failed: Permission denied

* the server is Solaris 11.2 -> EPERM is generated

[stas@ol6-x64 tmp]$ ls -l /mnt/sol/test_file
-rw-r--r--+ 1 root root 0 Jun  1  2015 /mnt/sol/test_file
[stas@ol6-x64 tmp]$ strace -e utime,utimes ./utime_test /mnt/sol/test_file
utime("/mnt/sol/test_file", [0, 0])     = -1 EPERM (Operation not permitted)
utime() failed: Operation not permitted
utimes("/mnt/sol/test_file", {{0, 0}, {0, 0}}) = -1 EPERM (Operation not 
permitted)
utimes() failed: Operation not permitted

* on a local ext4 file system EPERM is generated:

[stas@ol6-x64 tmp]$ ls -l /tmp/test_file
-rw-r--r-- 1 root root 0 Jun  1 17:51 /tmp/test_file
[stas@ol6-x64 tmp]$ strace -e utime,utimes ./utime_test /tmp/test_file
utime("/tmp/test_file", [0, 0])         = -1 EPERM (Operation not permitted)
utime() failed: Operation not permitted
utimes("/tmp/test_file", {{0, 0}, {0, 0}}) = -1 EPERM (Operation not 
permitted)
utimes() failed: Operation not permitted

Plus EPERM is generated when the NFS server is FreeBSD 9.1.

Could anybody, clarify, if the described behavior a bug in the Linux NFS 
server implementation or not?

Thank you.

[1] http://man7.org/linux/man-pages/man2/utime.2.html

PS: this all was found using utime06, utimes01 LTP test cases.