On Thu, 2019-12-12 at 10:07 +0800, Ian Kent wrote:
> On Wed, 2019-12-11 at 05:54 -0500, Doug Nazar wrote:
> > On 2019-12-09 23:49, Ian Kent wrote:
> > > I also need to work out why you don't get caught by the negative
> > > map entry check that's meant to prevent multiple retries for a
> > > failing map entry for a configured time.
> >
> > Sorry, I should have been more explicit. The several minute wait
> > was
> > to
> > exceed the negative cache timeout. That part was working fine.
> >
> > > > And even the entry delete below it should be ok because it will
> > > > just lookup (aka. run the program map again to get the map
> > > > entry)
> > > > and then update the multi-mount during the entry parse.
> > > >
> > > > So while the change above isn't strictly the way this should be
> > > > handled it probably should be ok.
> > > >
> > > > I haven't worked out how to handle it immediately after the
> > > > fail
> > > > just yet but the change above probably should be kept as part
> > > > of
> > > > that as well, not sure yet.
> > > >
> > > > Ian
> >
> > I did that based on my greps, that seems to be a fairly common
> > check.
> > However, it kind of felt wrong, in the sense that the 2nd attempt,
> > shouldn't depend on any previous status. I was just having trouble
> > trying to figure out the lifetime rules for the various
> > fields/states.
>
> Yes, the root of a multi-mount has all sorts of special case handling
> in a number of places.
I think this is what should be done.
Could you check and see if it behaves as required please, of course
the CHANGELOG hunk needs to be dropped.
autofs-5.1.6 - fix program map multi-mount lookup after mount fail
From: Ian Kent <raven@xxxxxxxxxx>
For the case of a singleton multi-mount program map lookup following
a mount fail (and the negative timeout has passed) the lookup key is
what's expected for an indirect map key but the the root offset map
entry already exists. This causes autofs to think it has an incorrect
lookup key and it fails the lookup when it should take the opptunity
to delete and update the cache entry since it's not actually in use
yet.
Also, if a check is done to see if the lookup is for the root offset,
deleting the entry fails because it contains an offset so it doesn't
get updated.
Signed-off-by: Ian Kent <raven@xxxxxxxxxx>
---
CHANGELOG | 1 +
modules/lookup_program.c | 7 +++++--
2 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/CHANGELOG b/CHANGELOG
index 981a0333..3c784d34 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -1,6 +1,7 @@
xx/xx/2020 autofs-5.1.7
- make bind mounts propagation slave by default.
- update ldap READMEs and schema definitions.
+- fix program map multi-mount lookup after mount fail.
07/10/2019 autofs-5.1.6
- support strictexpire mount option.
diff --git a/modules/lookup_program.c b/modules/lookup_program.c
index fcb1af74..ca209488 100644
--- a/modules/lookup_program.c
+++ b/modules/lookup_program.c
@@ -646,7 +646,7 @@ int lookup_mount(struct autofs_point *ap, const char *name, int name_len, void *
name_len, ent, ctxt->parse->context);
goto out_free;
} else {
- if (me->multi) {
+ if (me->multi && me->multi != me) {
cache_unlock(mc);
warn(ap->logopt, MODPREFIX
"unexpected lookup for active multi-mount"
@@ -656,8 +656,11 @@ int lookup_mount(struct autofs_point *ap, const char *name, int name_len, void *
cache_unlock(mc);
cache_writelock(mc);
me = cache_lookup_distinct(mc, name);
- if (me)
+ if (me) {
+ if (me->multi)
+ cache_delete_offset_list(mc, name);
cache_delete(mc, name);
+ }
cache_unlock(mc);
}
}
