[Bugs] [Bug 1746142] New: ctime: If atime is updated via utimensat syscall ctime is not getting updated

Tue Aug 27 18:24:33 UTC 2019

https://bugzilla.redhat.com/show_bug.cgi?id=1746142

            Bug ID: 1746142
           Summary: ctime: If atime is updated via utimensat syscall ctime
                    is not getting updated
           Product: GlusterFS
           Version: 7
          Hardware: x86_64
                OS: Linux
            Status: NEW
         Component: ctime
          Severity: high
          Assignee: bugs at gluster.org
          Reporter: khiremat at redhat.com
                CC: bugs at gluster.org, nchilaka at redhat.com
        Depends On: 1738786
            Blocks: 1743627, 1746138
  Target Milestone: ---
    Classification: Community

+++ This bug was initially created as a clone of Bug #1738786 +++

Description of problem:
When atime|mtime is updated via utime family of syscalls,
ctime is not updated.

Version-Release number of selected component (if applicable):
mainline

How reproducible:
Always

Steps to Reproduce:
touch /mnt/file1
stat /mnt/file1
sleep 1;
touch -m -d "2020-01-01 12:00:00" /mnt/file1
stat /mnt/file1

Actual results:
ctime is same between two stats above

Expected results:
ctime should be changed between two stats above

Additional info:

--- Additional comment from Worker Ant on 2019-08-08 07:43:19 UTC ---

REVIEW: https://review.gluster.org/23177 (ctime: Fix ctime issue with utime
family of syscalls) posted (#1) for review on master by Kotresh HR

--- Additional comment from Kotresh HR on 2019-08-20 10:49:58 UTC ---

Discussion at the patch which is worth mentioning here

Kinglong Mee:

@Kotresh HR, I cannot reproduce this problem as you description,

Steps to Reproduce:
touch /mnt/file1
stat /mnt/file1
sleep 1;
touch -m -d "2020-01-01 12:00:00" /mnt/file1
stat /mnt/file1

Actual results:
ctime is same between two stats above

Expected results:
ctime should be changed between two stats above

I test at glusterfs mount and nfs mount, all get right result as,
# sh test.sh 
  File: file1
  Size: 0               Blocks: 0          IO Block: 131072 regular empty file
Device: 30h/48d Inode: 13578024673158818984  Links: 1
Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2019-08-08 18:39:16.099467595 +0800
Modify: 2019-08-08 18:39:16.099467595 +0800
Change: 2019-08-08 18:39:16.100847925 +0800
 Birth: -
  File: file1
  Size: 0               Blocks: 0          IO Block: 131072 regular empty file
Device: 30h/48d Inode: 13578024673158818984  Links: 1
Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
Access: 2019-08-08 18:39:16.099467595 +0800
Modify: 2020-01-01 12:00:00.000000000 +0800
Change: 2019-08-08 18:39:17.126800759 +0800
 Birth: -

---------------

Sorry, this is not happening with your patch[1]. Because we don't update ctime
to mtime which was previously being done.

But with your patch [1], I think all setattrs even some internal setattrs are
updating ctime. So when file is created, all times should be same. But now it's
not.

[root at f281 glusterfs]# stat /mastermnt/file3
  File: /mastermnt/file3
  Size: 0               Blocks: 0          IO Block: 131072 regular empty file
Device: 2eh/46d Inode: 13563962387061186202  Links: 1
Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
Context: system_u:object_r:fusefs_t:s0
Access: 2019-08-08 17:44:31.341319550 +0530
Modify: 2019-08-08 17:44:31.341319550 +0530
Change: 2019-08-08 17:44:31.342550008 +0530  <<<< ctime is different
 Birth: -
[root at f281

So I was trying to fix this issue. And other thing, ideally updating
atime|mtime should update ctime with current time, which was not happening in
"posix_update_utime_in_mdata" but was happening as part of "posix_set_ctime" in
posix_setattr. 

-----------------------------
|But with your patch [1], I think all setattrs even some internal setattrs are
updating ctime. So when file is created, all times should be same. But now it's
not.
Yes, you are right. when file is created, all times should be same.
With the patch[1], those times are different.

For nfs, a create of a file, nfs client sends a create rpc, and a setattr(set
time of server).
Ganesha.nfsd gets the CLOCK_REALTIME for mtime/atime, and the utime xlator gets
the realtime for ctime, so that, we cannot gets all times same when creating
file.
I think we should let utime xlator gets the realtime for all
times(ctime/atime/mtime), ganesha.nfsd does not do that.

|With this patch, it is clean. I am inclined to take this patch in if this
solves the original nfs problem you reported. Could you please test that out
and let me know?
With this patch, the nfs problem of bad ctime is not exist now.

[1] https://review.gluster.org/#/c/23154/

--- Additional comment from Worker Ant on 2019-08-20 16:54:08 UTC ---

REVIEW: https://review.gluster.org/23177 (ctime: Fix ctime issue with utime
family of syscalls) merged (#7) on master by Amar Tumballi

Referenced Bugs:

https://bugzilla.redhat.com/show_bug.cgi?id=1738786
[Bug 1738786] ctime: If atime is updated via utimensat syscall ctime is not
getting updated
https://bugzilla.redhat.com/show_bug.cgi?id=1743627
[Bug 1743627] ctime: If atime is updated via utimensat syscall ctime is not
getting updated
https://bugzilla.redhat.com/show_bug.cgi?id=1746138
[Bug 1746138] ctime: If atime is updated via utimensat syscall ctime is not
getting updated
-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.