[linux-cifs-client] SMB_COM_CLOSE and LastWriteTime

[Jeff Layton]

On Thu, 24 Apr 2008 17:26:58 +0200
Guenter Kukkukk <linux at kukkukk.com> wrote:

> Am Donnerstag, 24. April 2008 schrieb Jeff Layton:
> > On Thu, 24 Apr 2008 16:13:46 +0200
> > Guenter Kukkukk <linux at kukkukk.com> wrote:
> > 
> > > Am Donnerstag, 24. April 2008 schrieb Guenter Kukkukk:
> > > 
> > > > An additional note:
> > > > i get the following, when i use
> > > > cp -p foo /mnt/linux/cifs
> > > > cp: preserving permissions for `/mnt/linux/cifs/settime.c': Input/output error
> > > > I must specify the "noacl" mount option to avoid this.
> > > > Happens when mounting recent 3.0.x or 3.2 samba server.
> > > 
> > > Ouch - should read  
> > > 
> > > cp -p settime.c /mnt/linux/cifs
> > > cp: preserving permissions for `/mnt/linux/cifs/settime.c': Input/output error
> > > 
> > > We can leave this one alone atm - could also be a wrong server setup.
> > > Cheers, Günter
> > 
> > 
> > Agreed. This doesn't appear to be directly related to the problem with
> > setting times.
> > 
> > I tested out the patch that you sent on March 13 and it seems to fix
> > the problem of setting times against win2k3. My only concern is that it
> > looks like it rips out some of the code that did an implicit
> > open/setattr/close of a file when it's not already open. Some comments
> > seem to indicate that older servers needed the file open in order to
> > properly set the time. Do we know that this won't break against older
> > servers?
> > 
> > Thanks,
> 
> I already worked on this area of code about 2 years ago, when Steve and
> me tried to add some legacy functionality for os/2 (and win9x/me).
> Most of those additions regarding setpathinfo, setfileinfo and also the
> corresponding query functions did not really make it into the code that days...
> The biggest problem was, that the legacy servers need different info levels
> and so also the passed structures like FILE_BASIC_INFO do not match.
> 
> Not counting legacy stuff, I remember that the special handling for 
> CIFS_SES_NT4 made some sense those days, but in my recent more comprehensive
> tests i did not "hit any reason for that" anymore.
> My recent patch also works with winnt.
> Probably Steve can comment on this.
> 
> I already did some recent work to add more legacy functionality. 
> Talked to Steve at sambaxp whether we should place all/most of that
> new functions into a separate "legacy" source file. He agreed on this.
> One additional needed time conversion function could still be placed into
> netmisc.c. But more on that later ...
> 

I'm afraid I don't have a working NT4 image, but I did test this patch
against Win98se. cp -p always failed to set the mtime for me on that
host before. If I understand what wireshark is telling me,
Win98se doesn't support SET_PATH_INFO calls. It does support
SET_FILE_INFO calls, however, so this patch fixes mtime setting when
doing a cp -p to a win98se share (since the file is open we end up
using SET_FILE_INFO here).

I suspect though, that we'll probably still need the implicit
open/set_file_info/close to make win98se and other early vintage
servers work correctly (if it's deemed worthwhile to fix them).

If we care about fixing this universally, we'll some fairly
complex logic:

1) if file is open use SET_FILE_INFO
2) if file is closed and server supports it, use SET_PATH_INFO
3) if file is close and server doesn't support SET_PATH_INFO, then open
the file, use SET_FILE_INFO, and close it again.

I think that even this is probably too simple, and we might need to
special case some stuff, but I suspect this will get most cases...

-- 
Jeff Layton <jlayton at redhat.com>