Inaccurate nanoseconds in timestamp for fstat when backed by CIFS file system

[eggert]

An GNU Emacs user is reporting that when using Emacs to edit a file in an encfs file system (where the encrypted version is stored via CIFS in a Microsoft server), Emacs complains that the file has changed even though it has not. We can reproduce the bug without using Emacs. Please see http://bugs.gnu.org/19607 for the Emacs bug report, in particular Message#38 (which gives a small C test program) and Message#41 (which confirms that the bug is present. The C program can be found at http://debbugs.gnu.org/cgi/bugreport.cgi?msg=38;filename=test.c;att=1;bug=19607.

The problem appears to be that encfs and/or CIFS is marking a file for update, then the program calls fstat to get the time stamp, and then afterwards encfs and/or CIFS actually updates st_mtime. POSIX does not allow this behavior; fstat should get the updated st_mtime.

[rfjakob]

This can also be reproduced on CIFS (without EncFS). Forwarded to https://bugzilla.samba.org/show_bug.cgi?id=11078

[rfjakob]

A new test case shows that EncFS does do something wrong. Code at https://gist.github.com/rfjakob/a956df2edcb69b486938 , results:

CIFS:

   Emacs should work around this POSIX-conformance bug.
   stat #1: 1422398418.561215369
   stat #2: 1422398418.561215300
   stat #3: 1422398418.561215300

CIFS + encfs:

   Emacs does not work around this POSIX-conformance bug.
   stat #1: 1422398433.735274107
   stat #2: 1422398433.735274107
   stat #3: 1422398433.735274100

CIFS + "encfs -o attr_timeout=0":

   Emacs should work around this POSIX-conformance bug.
   stat #1: 1422398527.607547169
   stat #2: 1422398527.607547100
   stat #3: 1422398527.607547100