Re: [9fans] 9vx and create (...DMEXCL)

Re: [9fans] 9vx and create (...DMEXCL)

[erik quanstrom]

> Is create (2) with DMEXCL mode supported under 9vx?

it seems you've already got the answer ...

> If 9vx maps Plan9 permissions directly to Linux permissions this is
> understandable. But is there any workaround?

it's up to the fileserver to implement DMEXCL, and #Z doesn't.
using a different fileserver (e.g. fossil or cwfs, ken's fs) would solve
the problem.  you could also modify ldevfs-posix.c to accept DMEXCL
and map that to O_EXCL|O_CREAT which may actually do the
right thing.

- erik

Re: [9fans] 9vx and create (...DMEXCL)

[Charles Forsyth]

>you could also modify ldevfs-posix.c to accept DMEXCL
>and map that to O_EXCL|O_CREAT which may actually do the
>right thing.

no, it won't: OEXCL, perhaps, but not DMEXCL.
i'd have thought it should be used in fscreate, though,
to help chan.c implement OEXCL correctly. fscreate should
fail if the file exists, but perhaps that happens another way.

Re: [9fans] 9vx and create (...DMEXCL)

[Dmitry Golubovsky]

Hi,

OK, thanks for the clue.

Unfortunately, this (devfs-posix.c in 9vx):

	if(perm & ~(DMDIR|0777))
		error(Ebadarg);

generates the error, and the values of DMEXCL and DMAPPEND do not seem
to fit into Unix permissions. So these won't work in 9vx as long as
the host filesystem is used*.

------------------------------------------------
* Unless the host FS is ext?fs and the kernel supports extended
attributes which may require specific patches

On Dec 22, 4:50 am, fors...@terzarima.net (Charles Forsyth) wrote:

> no, it won't: OEXCL, perhaps, but not DMEXCL.
> i'd have thought it should be used in fscreate, though,
> to help chan.c implement OEXCL correctly. fscreate should
> fail if the file exists, but perhaps that happens another way.

Dmitry.

Re: [9fans] 9vx and create (...DMEXCL)

[Russ Cox]

> no, it won't: OEXCL, perhaps, but not DMEXCL.
> i'd have thought it should be used in fscreate, though,
> to help chan.c implement OEXCL correctly. fscreate should
> fail if the file exists, but perhaps that happens another way.

yes, there's a bug.
fscreate should or in O_EXCL unconditionally here
to implement the correct 9p create semantics:

		if((fd = open(path, mm|O_CREAT, 0666)) < 0)
			oserror();

the or'ing in is unconditional because create(5)
always fails when the target already exists.
the OEXCL in create(2) is a kernel-only flag
(not part of 9p), like OCEXEC.

russ

Re: [9fans] 9vx and create (...DMEXCL)

[erik quanstrom]

> i'd have thought it should be used in fscreate, though,
> to help chan.c implement OEXCL correctly. fscreate should
> fail if the file exists, but perhaps that happens another way.

it already fail on unexpected bits.  DMEXCL is unexpected.

- erik

[9fans] 9vx and create (...DMEXCL)

[Dimitry Golubovsky]

Hi,

Is create (2) with DMEXCL mode supported under 9vx?

I am trying to run webcookie, and am getting the following:


term% webcookies
webcookies: readjar: cannot acquire jar lock:
'/usr/dima/lib/L.webcookies' bad arg in system call
term%

coming likely from this code in webcookies.c:

=============================================

	for(i=0; i<50; i++){
		if((fd = create(jar->lockfile, OWRITE, DMEXCL|0666)) < 0){
			sleep(100);
			continue;
		}
		break;
	}
	if(fd < 0){
		if(debug)
			fprint(2, "open %s: %r", jar->lockfile);
		werrstr("cannot acquire jar lock: %r");
		return -1;
	}

=============================================

If 9vx maps Plan9 permissions directly to Linux permissions this is
understandable. But is there any workaround?

Thanks.


--
Dimitry Golubovsky

Anywhere on the Web