IS_NOCMTIME and setting of ctime and mtime on remote servers

IS_NOCMTIME and setting of ctime and mtime on remote servers

[Steve French]

NFS is the only place that sets NOCMTIME on inodes in its fhget routine 
IIRC.

What is the exact intent of this?  Does it stay set (so mtime and ctime 
updates are never sent to the server) or does it get reset somewhere (I 
did not see where nfs turned it off so presumably even explicit sets of 
the time are ignored by default?)?

I realize that some users would prefer that an fs never set the file 
times to the remote server (at least for the cifs and nfs cases I have 
heard this in past years) as the server will set them implicitly anyway 
(and it can help performance).  For cached files the local times in the 
client's version of the inode could be used indefinately (at least while 
the file is cacheable on the client, continuing to hold the caching 
token/oplock).

I saw a thread on bsd mailing list from a few years ago in which some 
users were asking for a mount flag (nocmtime) to turn off ctime and 
mtime updates  for much the same reason.

Re: IS_NOCMTIME and setting of ctime and mtime on remote servers

[Miklos Szeredi]

> NFS is the only place that sets NOCMTIME on inodes in its fhget routine 
> IIRC.

FUSE too.

> What is the exact intent of this?  Does it stay set (so mtime and ctime 
> updates are never sent to the server) or does it get reset somewhere (I 
> did not see where nfs turned it off so presumably even explicit sets of 
> the time are ignored by default?)?

I don't think it gets reset.

> I realize that some users would prefer that an fs never set the file 
> times to the remote server (at least for the cifs and nfs cases I have 
> heard this in past years) as the server will set them implicitly anyway 
> (and it can help performance).  For cached files the local times in the 
> client's version of the inode could be used indefinately (at least while 
> the file is cacheable on the client, continuing to hold the caching 
> token/oplock).

It's not just performance: for network filsystems you usually want to
set the file times according to the server clock.  That way there'd be
some consitency even if the client's clock is out of sync.

Miklos

Re: IS_NOCMTIME and setting of ctime and mtime on remote servers

[Trond Myklebust]

må den 29.08.2005 Klokka 12:16 (-0500) skreiv Steve French:
> NFS is the only place that sets NOCMTIME on inodes in its fhget routine 
> IIRC.
> 
> What is the exact intent of this?  Does it stay set (so mtime and ctime 
> updates are never sent to the server) or does it get reset somewhere (I 
> did not see where nfs turned it off so presumably even explicit sets of 
> the time are ignored by default?)?

NOCMTIME just turns off the inode_update_time() crapola. It has nothing
to do with explicit sets of the time which are done through the utimes()
syscall.

IOW: it turns off all those unwanted VFS updates of mtime/ctime which
would otherwise screw up data and metadata cache revalidation.

Cheers,
  Trond

-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html