tdb_check fails with tdb read-only open.

[Rusty Russell]

On Tue, 14 Sep 2010 03:15:30 am Jeremy Allison wrote:
> On Mon, Sep 13, 2010 at 05:20:50PM +0930, Rusty Russell wrote:
> > On Thu, 29 Jul 2010 07:35:13 am tridge at samba.org wrote:
> > > looks good to me - nice and simple :-)
> > 
> > And unfortunately, doesn't work.
> > 
> > I just got around to porting this to CCAN, and added a unit test for it.
> > 
> > tdb_lockall_read() fails on r/o TDBs.  So that needs to be fixed too.  You
> 
> How does this fail ? As far as I can see it
> only does F_RDLCK calls on the db - shouldn't
> that always succeed against a r/o fd ? Just trying
> to understand the error.

I missed it too; it's in tdb_allrecord_check:

	int tdb_lockall_read(struct tdb_context *tdb)
	{
		tdb_trace(tdb, "tdb_lockall_read");
		return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
	}

Here's tdb_allrecord_lock:

	int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
			       enum tdb_lock_flags flags, bool upgradable)
	{
		switch (tdb_allrecord_check(tdb, ltype, flags, upgradable)) {
		case -1:
			return -1;
		case 0:
			return 0;
		}
	...

And here's tdb_allrecord_check:

	static int tdb_allrecord_check(struct tdb_context *tdb, int ltype,
				       enum tdb_lock_flags flags, bool upgradable)
	{
		/* There are no locks on read-only dbs */
		if (tdb->read_only || tdb->traverse_read) {
			tdb->ecode = TDB_ERR_LOCK;
			return -1;
		}
	...

I'd forgotten this feature of read-only TDBs; they use no locks *at all*.

I added a unit test.  Unfortunately the unit tests are in the CCAN version
of tdb; fortunately I keep them in sync so we benefit in SAMBA.

> > Patch coming...
> 
> Doesn't your patch still leave the F_WRLCK code
> inside tdb_lock_and_recover() ?

Yes, but we can never get there (and if we did, it'd fail as expected).

Cheers,
Rusty.