xen: allow limiting the amount of duplicated pending xenstore watches

Xenstore watches received are queued in a list and processed in a
deferred thread. Such queuing was done without any checking, so a
guest could potentially trigger a resource starvation against the
FreeBSD kernel if such kernel is watching any user-controlled xenstore
path.

Allowing limiting the amount of pending events a watch can accumulate
to prevent a remote guest from triggering this resource starvation
issue.

For the PV device backends and frontends this limitation is only
applied to the other end /state node, which is limited to 1 pending
event, the rest of the watched paths can still have unlimited pending
watches because they are either local or controlled by a privileged
domain.

The xenstore user-space device gets special treatment as it's not
possible for the kernel to know whether the paths being watched by
user-space processes are controlled by a guest domain. For this reason
watches set by the xenstore user-space device are limited to 1000
pending events. Note this can be modified using the
max_pending_watch_events sysctl of the device.

This is XSA-349.

Sponsored by:	Citrix Systems R&D
MFC after:	3 days

(cherry picked from commit 4e4e43dc9e1afc863670a031cc5cc75eb5e668d6)

Note the xenstore user-space device part of this backport is dropped,
as in stable/11 the device doesn't support setting up watches.

(cherry picked from commit d9bd043f93df1a31ef16d2198d720a0a0831357f)

Approved by:	so
Security:	XSA-349, CVE-2020-29568

6 files changed, 51 insertions, 4 deletions

diff --git a/sys/dev/xen/balloon/balloon.c b/sys/dev/xen/balloon/balloon.c
index aefa5cdb9250..87163000d4a8 100644
--- a/sys/dev/xen/balloon/balloon.c
+++ b/sys/dev/xen/balloon/balloon.c
@@ -310,7 +310,8 @@ set_new_target(unsigned long target)
 
 static struct xs_watch target_watch =
 {
-	.node = "memory/target"
+	.node = "memory/target",
+	.max_pending = 1,
 };
 
 /* React to a change in the target key */
diff --git a/sys/dev/xen/blkback/blkback.c b/sys/dev/xen/blkback/blkback.c
index a5f9cc101582..4b6eeb0cf62f 100644
--- a/sys/dev/xen/blkback/blkback.c
+++ b/sys/dev/xen/blkback/blkback.c
@@ -3767,6 +3767,12 @@ xbb_attach(device_t dev)
 	xbb->hotplug_watch.callback = xbb_attach_disk;
 	KASSERT(xbb->hotplug_watch.node == NULL, ("watch node already setup"));
 	xbb->hotplug_watch.node = strdup(sbuf_data(watch_path), M_XENBLOCKBACK);
+	/*
+	 * We don't care about the path updated, just about the value changes
+	 * on that single node, hence there's no need to queue more that one
+	 * event.
+	 */
+	xbb->hotplug_watch.max_pending = 1;
 	sbuf_delete(watch_path);
 	error = xs_register_watch(&xbb->hotplug_watch);
 	if (error != 0) {
diff --git a/sys/dev/xen/control/control.c b/sys/dev/xen/control/control.c
index d5aa97e03b73..e1739419826b 100644
--- a/sys/dev/xen/control/control.c
+++ b/sys/dev/xen/control/control.c
@@ -432,6 +432,12 @@ xctrl_attach(device_t dev)
 	xctrl->xctrl_watch.node = "control/shutdown";
 	xctrl->xctrl_watch.callback = xctrl_on_watch_event;
 	xctrl->xctrl_watch.callback_data = (uintptr_t)xctrl;
+	/*
+	 * We don't care about the path updated, just about the value changes
+	 * on that single node, hence there's no need to queue more that one
+	 * event.
+	 */
+	xctrl->xctrl_watch.max_pending = 1;
 	xs_register_watch(&xctrl->xctrl_watch);
 
 	if (xen_pv_domain())
diff --git a/sys/dev/xen/xenstore/xenstore.c b/sys/dev/xen/xenstore/xenstore.c
index 9289adfc1a79..7cba2a55f973 100644
--- a/sys/dev/xen/xenstore/xenstore.c
+++ b/sys/dev/xen/xenstore/xenstore.c
@@ -668,12 +668,17 @@ xs_process_msg(enum xsd_sockmsg_type *type)
 		mtx_lock(&xs.registered_watches_lock);
 		msg->u.watch.handle = find_watch(
 		    msg->u.watch.vec[XS_WATCH_TOKEN]);
-		if (msg->u.watch.handle != NULL) {
-			mtx_lock(&xs.watch_events_lock);
+		mtx_lock(&xs.watch_events_lock);
+		if (msg->u.watch.handle != NULL &&
+		    (!msg->u.watch.handle->max_pending ||
+		    msg->u.watch.handle->pending <
+		    msg->u.watch.handle->max_pending)) {
+			msg->u.watch.handle->pending++;
 			TAILQ_INSERT_TAIL(&xs.watch_events, msg, list);
 			wakeup(&xs.watch_events);
 			mtx_unlock(&xs.watch_events_lock);
 		} else {
+			mtx_unlock(&xs.watch_events_lock);
 			free(msg->u.watch.vec, M_XENSTORE);
 			free(msg, M_XENSTORE);
 		}
@@ -1045,8 +1050,10 @@ xenwatch_thread(void *unused)
 
 		mtx_lock(&xs.watch_events_lock);
 		msg = TAILQ_FIRST(&xs.watch_events);
-		if (msg)
+		if (msg) {
 			TAILQ_REMOVE(&xs.watch_events, msg, list);
+			msg->u.watch.handle->pending--;
+		}
 		mtx_unlock(&xs.watch_events_lock);
 
 		if (msg != NULL) {
@@ -1629,6 +1636,7 @@ xs_register_watch(struct xs_watch *watch)
 	char token[sizeof(watch) * 2 + 1];
 	int error;
 
+	watch->pending = 0;
 	sprintf(token, "%lX", (long)watch);
 
 	sx_slock(&xs.suspend_mutex);
diff --git a/sys/xen/xenbus/xenbusb.c b/sys/xen/xenbus/xenbusb.c
index 241efaa8fb4e..f191c7f62311 100644
--- a/sys/xen/xenbus/xenbusb.c
+++ b/sys/xen/xenbus/xenbusb.c
@@ -702,10 +702,21 @@ xenbusb_add_device(device_t dev, const char *type, const char *id)
 		ivars->xd_otherend_watch.node = statepath;
 		ivars->xd_otherend_watch.callback = xenbusb_otherend_watch_cb;
 		ivars->xd_otherend_watch.callback_data = (uintptr_t)ivars;
+		/*
+		 * Other end state node watch, limit to one pending event
+		 * to prevent frontends from queuing too many events that
+		 * could cause resource starvation.
+		 */
+		ivars->xd_otherend_watch.max_pending = 1;
 
 		ivars->xd_local_watch.node = ivars->xd_node;
 		ivars->xd_local_watch.callback = xenbusb_local_watch_cb;
 		ivars->xd_local_watch.callback_data = (uintptr_t)ivars;
+		/*
+		 * Watch our local path, only writable by us or a privileged
+		 * domain, no need to limit.
+		 */
+		ivars->xd_local_watch.max_pending = 0;
 
 		mtx_lock(&xbs->xbs_lock);
 		xbs->xbs_connecting_children++;
@@ -764,6 +775,12 @@ xenbusb_attach(device_t dev, char *bus_node, u_int id_components)
 	xbs->xbs_device_watch.node = bus_node;
 	xbs->xbs_device_watch.callback = xenbusb_devices_changed;
 	xbs->xbs_device_watch.callback_data = (uintptr_t)xbs;
+	/*
+	 * Allow for unlimited pending watches, as those are local paths
+	 * either controlled by the guest or only writable by privileged
+	 * domains.
+	 */
+	xbs->xbs_device_watch.max_pending = 0;
 
 	TASK_INIT(&xbs->xbs_probe_children, 0, xenbusb_probe_children_cb, dev);
 
diff --git a/sys/xen/xenstore/xenstorevar.h b/sys/xen/xenstore/xenstorevar.h
index 4c612b4b6881..e58d4e1d5721 100644
--- a/sys/xen/xenstore/xenstorevar.h
+++ b/sys/xen/xenstore/xenstorevar.h
@@ -72,6 +72,15 @@ struct xs_watch
 
 	/* Callback client data untouched by the XenStore watch mechanism. */
 	uintptr_t callback_data;
+
+	/* Maximum number of pending watch events to be delivered. */
+	unsigned int max_pending;
+
+	/*
+	 * Private counter used by xenstore to keep track of the pending
+	 * watches. Protected by xs.watch_events_lock.
+	 */
+	unsigned int pending;
 };
 LIST_HEAD(xs_watch_list, xs_watch);