Fixes core state machine.

The core state machine was broken - it didn't free resources
in START -> DOWN transition and might freed resources after
UP -> STOP transition before protocol turned down. It leads
to deadlock on olock acquisition when lock was not freed
during previous stop.

The current behavior is that resources, allocated during
DOWN -> * transition, are freed in * -> DOWN transition,
and flushing (scheduled in UP -> *) just counteract
feeding (scheduled in * -> UP). Protocol fell down
when both flushing is done (if needed) and protocol
reports DOWN.

BTW, is thera a reason why neighbour cache item acquired
by protocol is not tracked by resource mechanism?
This commit is contained in:
Ondrej Zajicek 2008-12-08 12:24:55 +01:00
parent fbde6c3908
commit d6a836f8d6
2 changed files with 42 additions and 30 deletions

View file

@ -254,7 +254,7 @@ neigh_if_down(struct iface *i)
static inline void
neigh_prune_one(neighbor *n)
{
if (n->proto->core_state != FS_FLUSHING)
if (n->proto->proto_state != PS_DOWN)
return;
rem_node(&n->n);
if (n->iface)

View file

@ -552,6 +552,30 @@ proto_feed(void *P)
proto_feed_more(P);
}
static void
proto_schedule_flush(struct proto *p)
{
/* Need to abort feeding */
if (p->core_state == FS_FEEDING)
rt_feed_baby_abort(p);
DBG("%s: Scheduling flush\n", p->name);
p->core_state = FS_FLUSHING;
proto_relink(p);
proto_flush_hooks(p);
ev_schedule(proto_flush_event);
}
static void
proto_schedule_feed(struct proto *p)
{
DBG("%s: Scheduling meal\n", p->name);
p->core_state = FS_FEEDING;
proto_relink(p);
p->attn->hook = proto_feed;
ev_schedule(p->attn);
}
/**
* proto_notify_state - notify core about protocol state change
* @p: protocol the state of which has changed
@ -562,7 +586,9 @@ proto_feed(void *P)
* it should immediately notify the core about the change by calling
* proto_notify_state() which will write the new state to the &proto
* structure and take all the actions necessary to adapt to the new
* state.
* state. State change to PS_DOWN immediately frees resources of protocol
* and might execute start callback of protocol; therefore,
* it should be used at tail positions of protocol callbacks.
*/
void
proto_notify_state(struct proto *p, unsigned ps)
@ -574,23 +600,23 @@ proto_notify_state(struct proto *p, unsigned ps)
if (ops == ps)
return;
p->proto_state = ps;
switch (ps)
{
case PS_DOWN:
neigh_prune(); // FIXME convert neighbors to resource?
rfree(p->pool);
p->pool = NULL;
if (cs == FS_HUNGRY) /* Shutdown finished */
{
p->proto_state = ps;
proto_fell_down(p);
return; /* The protocol might have ceased to exist */
}
else if (cs == FS_FLUSHING) /* Still flushing... */
;
else
{
if (cs == FS_FEEDING) /* Need to abort feeding */
rt_feed_baby_abort(p);
goto schedule_flush; /* Need to start flushing */
}
/* Otherwise, we have something to flush... */
else if (cs != FS_FLUSHING)
proto_schedule_flush(p);
break;
case PS_START:
ASSERT(ops == PS_DOWN);
@ -599,27 +625,15 @@ proto_notify_state(struct proto *p, unsigned ps)
case PS_UP:
ASSERT(ops == PS_DOWN || ops == PS_START);
ASSERT(cs == FS_HUNGRY);
DBG("%s: Scheduling meal\n", p->name);
cs = FS_FEEDING;
p->attn->hook = proto_feed;
ev_schedule(p->attn);
proto_schedule_feed(p);
break;
case PS_STOP:
if (ops != PS_DOWN)
{
schedule_flush:
DBG("%s: Scheduling flush\n", p->name);
proto_flush_hooks(p);
cs = FS_FLUSHING;
ev_schedule(proto_flush_event);
}
if ((cs = FS_FEEDING) || (cs == FS_HAPPY))
proto_schedule_flush(p);
break;
default:
bug("Invalid state transition for %s from %s/%s to */%s", p->name, c_states[cs], p_states[ops], p_states[ps]);
}
p->proto_state = ps;
p->core_state = cs;
proto_relink(p);
}
static void
@ -628,15 +642,13 @@ proto_flush_all(void *unused UNUSED)
struct proto *p;
rt_prune_all();
neigh_prune();
while ((p = HEAD(flush_proto_list))->n.next)
{
DBG("Flushing protocol %s\n", p->name);
rfree(p->pool);
p->pool = NULL;
p->core_state = FS_HUNGRY;
proto_relink(p);
proto_fell_down(p);
if (p->proto_state == PS_DOWN)
proto_fell_down(p);
}
}