evtchn: evtchn_reset() shouldn't succeed with still-open ports

While the function closes all ports, it does so without holding any
lock, and hence racing requests may be issued causing new ports to get
opened. This would have been problematic in particular if such a newly
opened port had a port number above the new implementation limit (i.e.
when switching from FIFO to 2-level) after the reset, as prior to
"evtchn: relax port_is_valid()" this could have led to e.g.
evtchn_close()'s "BUG_ON(!port_is_valid(d2, port2))" to trigger.

Introduce a counter of active ports and check that it's (still) no
larger then the number of Xen internally used ones after obtaining the
necessary lock in evtchn_reset().

As to the access model of the new {active,xen}_evtchns fields - while
all writes get done using write_atomic(), reads ought to use
read_atomic() only when outside of a suitably locked region.

Note that as of now evtchn_bind_virq() and evtchn_bind_ipi() don't have
a need to call check_free_port().

This is part of XSA-343.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Reviewed-by: Julien Grall <jgrall@amazon.com>

diff --git a/xen/common/event_channel.c b/xen/common/event_channel.c
index efbba61558d1c5fd92d77833b1c93efc9fc11a14..258281ea88b41cbd9c1b59b31229bc657e60e38c 100644 (file)
--- a/xen/common/event_channel.c
+++ b/xen/common/event_channel.c
@@ -188,6 +188,8 @@ int evtchn_allocate_port(struct domain *d, evtchn_port_t port)
         write_atomic(&d->valid_evtchns, d->valid_evtchns + EVTCHNS_PER_BUCKET);
     }
 
+    write_atomic(&d->active_evtchns, d->active_evtchns + 1);
+
     return 0;
 }
 
@@ -211,11 +213,26 @@ static int get_free_port(struct domain *d)
     return -ENOSPC;
 }
 
+/*
+ * Check whether a port is still marked free, and if so update the domain
+ * counter accordingly.  To be used on function exit paths.
+ */
+static void check_free_port(struct domain *d, evtchn_port_t port)
+{
+    if ( port_is_valid(d, port) &&
+         evtchn_from_port(d, port)->state == ECS_FREE )
+        write_atomic(&d->active_evtchns, d->active_evtchns - 1);
+}
+
 void evtchn_free(struct domain *d, struct evtchn *chn)
 {
     /* Clear pending event to avoid unexpected behavior on re-bind. */
     evtchn_port_clear_pending(d, chn);
 
+    if ( consumer_is_xen(chn) )
+        write_atomic(&d->xen_evtchns, d->xen_evtchns - 1);
+    write_atomic(&d->active_evtchns, d->active_evtchns - 1);
+
     /* Reset binding to vcpu0 when the channel is freed. */
     chn->state          = ECS_FREE;
     chn->notify_vcpu_id = 0;
@@ -258,6 +275,7 @@ static long evtchn_alloc_unbound(evtchn_alloc_unbound_t *alloc)
     alloc->port = port;
 
  out:
+    check_free_port(d, port);
     spin_unlock(&d->event_lock);
     rcu_unlock_domain(d);
 
@@ -351,6 +369,7 @@ static long evtchn_bind_interdomain(evtchn_bind_interdomain_t *bind)
     bind->local_port = lport;
 
  out:
+    check_free_port(ld, lport);
     spin_unlock(&ld->event_lock);
     if ( ld != rd )
         spin_unlock(&rd->event_lock);
@@ -484,7 +503,7 @@ static long evtchn_bind_pirq(evtchn_bind_pirq_t *bind)
     struct domain *d = current->domain;
     struct vcpu   *v = d->vcpu[0];
     struct pirq   *info;
-    int            port, pirq = bind->pirq;
+    int            port = 0, pirq = bind->pirq;
     long           rc;
 
     if ( (pirq < 0) || (pirq >= d->nr_pirqs) )
@@ -532,6 +551,7 @@ static long evtchn_bind_pirq(evtchn_bind_pirq_t *bind)
     arch_evtchn_bind_pirq(d, pirq);
 
  out:
+    check_free_port(d, port);
     spin_unlock(&d->event_lock);
 
     return rc;
@@ -1005,10 +1025,10 @@ int evtchn_unmask(unsigned int port)
     return 0;
 }
 
-
 int evtchn_reset(struct domain *d)
 {
     unsigned int i;
+    int rc = 0;
 
     if ( d != current->domain && !d->controller_pause_count )
         return -EINVAL;
@@ -1018,7 +1038,9 @@ int evtchn_reset(struct domain *d)
 
     spin_lock(&d->event_lock);
 
-    if ( d->evtchn_fifo )
+    if ( d->active_evtchns > d->xen_evtchns )
+        rc = -EAGAIN;
+    else if ( d->evtchn_fifo )
     {
         /* Switching back to 2-level ABI. */
         evtchn_fifo_destroy(d);
@@ -1027,7 +1049,7 @@ int evtchn_reset(struct domain *d)
 
     spin_unlock(&d->event_lock);
 
-    return 0;
+    return rc;
 }
 
 static long evtchn_set_priority(const struct evtchn_set_priority *set_priority)
@@ -1213,10 +1235,9 @@ int alloc_unbound_xen_event_channel(
 
     spin_lock(&ld->event_lock);
 
-    rc = get_free_port(ld);
+    port = rc = get_free_port(ld);
     if ( rc < 0 )
         goto out;
-    port = rc;
     chn = evtchn_from_port(ld, port);
 
     rc = xsm_evtchn_unbound(XSM_TARGET, ld, chn, remote_domid);
@@ -1232,7 +1253,10 @@ int alloc_unbound_xen_event_channel(
 
     spin_unlock(&chn->lock);
 
+    write_atomic(&ld->xen_evtchns, ld->xen_evtchns + 1);
+
  out:
+    check_free_port(ld, port);
     spin_unlock(&ld->event_lock);
 
     return rc < 0 ? rc : port;
@@ -1308,6 +1332,7 @@ int evtchn_init(struct domain *d)
         return -EINVAL;
     }
     evtchn_from_port(d, 0)->state = ECS_RESERVED;
+    write_atomic(&d->active_evtchns, 0);
 
 #if MAX_VIRT_CPUS > BITS_PER_LONG
     d->poll_mask = xzalloc_array(unsigned long,
@@ -1335,6 +1360,8 @@ void evtchn_destroy(struct domain *d)
     for ( i = 0; port_is_valid(d, i); i++ )
         evtchn_close(d, i, 0);
 
+    ASSERT(!d->active_evtchns);
+
     clear_global_virq_handlers(d);
 
     evtchn_fifo_destroy(d);

diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
index 56fe0e774d3953245261615573630e5c4fae0e3c..a8ebfe6d955f7b488b0e0564c22b286f930fb364 100644 (file)
--- a/xen/include/xen/sched.h
+++ b/xen/include/xen/sched.h
@@ -345,6 +345,16 @@ struct domain
     struct evtchn  **evtchn_group[NR_EVTCHN_GROUPS]; /* all other buckets */
     unsigned int     max_evtchn_port; /* max permitted port number */
     unsigned int     valid_evtchns;   /* number of allocated event channels */
+    /*
+     * Number of in-use event channels.  Writers should use write_atomic().
+     * Readers need to use read_atomic() only when not holding event_lock.
+     */
+    unsigned int     active_evtchns;
+    /*
+     * Number of event channels used internally by Xen (not subject to
+     * EVTCHNOP_reset).  Read/write access like for active_evtchns.
+     */
+    unsigned int     xen_evtchns;
     spinlock_t       event_lock;
     const struct evtchn_port_ops *evtchn_port_ops;
     struct evtchn_fifo_domain *evtchn_fifo;