sfc: protect filter table against use-after-free

If MCDI timeouts are encountered during efx_ef10_filter_table_remove(),
an FLR will be queued, but efx->filter_state will still be kfree()d.
The queued FLR will then call efx_ef10_filter_table_restore(), which
will try to use efx->filter_state. This previously caused a panic.
This patch adds an rwsem to protect the existence of efx->filter_state,
separately from the spinlock protecting its contents. Users which can
race against efx_ef10_filter_table_remove() should down_read this rwsem.

Signed-off-by: Shradha Shah <[e-mail]>
Signed-off-by: David S. Miller <[e-mail]>

drivers/net/ethernet/sfc/ef10.c

···
 	return rc;
 }
 
+/* Caller must hold efx->filter_sem for read if race against
+ * efx_ef10_filter_table_remove() is possible
+ */
 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
 {
 	struct efx_ef10_filter_table *table = efx->filter_state;
···
 	bool failed = false;
 	int rc;
 
+	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
+
 	if (!nic_data->must_restore_filters)
+		return;
+
+	if (!table)
 		return;
 
 	spin_lock_bh(&efx->filter_lock);
···
 		nic_data->must_restore_filters = false;
 }
 
+/* Caller must hold efx->filter_sem for write */
 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
 {
 	struct efx_ef10_filter_table *table = efx->filter_state;
···
 	struct efx_filter_spec *spec;
 	unsigned int filter_idx;
 	int rc;
+
+	efx->filter_state = NULL;
+	if (!table)
+		return;
 
 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
···
 	kfree(table);
 }
 
+/* Caller must hold efx->filter_sem for read if race against
+ * efx_ef10_filter_table_remove() is possible
+ */
 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
 {
 	struct efx_ef10_filter_table *table = efx->filter_state;
···
 	int i, n, rc;
 
 	if (!efx_dev_registered(efx))
+		return;
+
+	if (!table)
 		return;
 
 	/* Mark old filters that may need to be removed */

drivers/net/ethernet/sfc/efx.c

···
 
 static void efx_fini_port(struct efx_nic *efx);
 
+/* We assume that efx->type->reconfigure_mac will always try to sync RX
+ * filters and therefore needs to read-lock the filter table against freeing
+ */
+void efx_mac_reconfigure(struct efx_nic *efx)
+{
+	down_read(&efx->filter_sem);
+	efx->type->reconfigure_mac(efx);
+	up_read(&efx->filter_sem);
+}
+
 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
  * the MAC appropriately. All other PHY configuration changes are pushed
  * through phy_op->set_settings(), and pushed asynchronously to the MAC
···
 
 	mutex_lock(&efx->mac_lock);
 	if (efx->port_enabled)
-		efx->type->reconfigure_mac(efx);
+		efx_mac_reconfigure(efx);
 	mutex_unlock(&efx->mac_lock);
 }
 
···
 
 	/* Reconfigure the MAC before creating dma queues (required for
 	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
-	efx->type->reconfigure_mac(efx);
+	efx_mac_reconfigure(efx);
 
 	/* Ensure the PHY advertises the correct flow control settings */
 	rc = efx->phy_op->reconfigure(efx);
···
 	efx->port_enabled = true;
 
 	/* Ensure MAC ingress/egress is enabled */
-	efx->type->reconfigure_mac(efx);
+	efx_mac_reconfigure(efx);
 
 	mutex_unlock(&efx->mac_lock);
 }
···
 	int rc;
 
 	spin_lock_init(&efx->filter_lock);
-
+	init_rwsem(&efx->filter_sem);
+	down_write(&efx->filter_sem);
 	rc = efx->type->filter_table_probe(efx);
 	if (rc)
-		return rc;
+		goto out_unlock;
 
 #ifdef CONFIG_RFS_ACCEL
 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
···
 					   GFP_KERNEL);
 		if (!efx->rps_flow_id) {
 			efx->type->filter_table_remove(efx);
-			return -ENOMEM;
+			rc = -ENOMEM;
+			goto out_unlock;
 		}
 	}
 #endif
-
-	return 0;
+out_unlock:
+	up_write(&efx->filter_sem);
+	return rc;
 }
 
 static void efx_remove_filters(struct efx_nic *efx)
···
 #ifdef CONFIG_RFS_ACCEL
 	kfree(efx->rps_flow_id);
 #endif
+	down_write(&efx->filter_sem);
 	efx->type->filter_table_remove(efx);
+	up_write(&efx->filter_sem);
 }
 
 static void efx_restore_filters(struct efx_nic *efx)
 {
+	down_read(&efx->filter_sem);
 	efx->type->filter_table_restore(efx);
+	up_read(&efx->filter_sem);
 }
 
 /**************************************************************************
···
 
 	mutex_lock(&efx->mac_lock);
 	net_dev->mtu = new_mtu;
-	efx->type->reconfigure_mac(efx);
+	efx_mac_reconfigure(efx);
 	mutex_unlock(&efx->mac_lock);
 
 	efx_start_all(efx);
···
 
 	/* Reconfigure the MAC */
 	mutex_lock(&efx->mac_lock);
-	efx->type->reconfigure_mac(efx);
+	efx_mac_reconfigure(efx);
 	mutex_unlock(&efx->mac_lock);
 
 	return 0;
···
 			   " VFs may not function\n", rc);
 #endif
 
+	down_read(&efx->filter_sem);
 	efx_restore_filters(efx);
+	up_read(&efx->filter_sem);
 	if (efx->type->sriov_reset)
 		efx->type->sriov_reset(efx);
 

drivers/net/ethernet/sfc/efx.h

···
 
 /* Filters */
 
+void efx_mac_reconfigure(struct efx_nic *efx);
+
 /**
  * efx_filter_insert_filter - add or replace a filter
  * @efx: NIC in which to insert the filter

drivers/net/ethernet/sfc/ethtool.c

···
 	/* Reconfigure the MAC. The PHY *may* generate a link state change event
 	 * if the user just changed the advertised capabilities, but there's no
 	 * harm doing this twice */
-	efx->type->reconfigure_mac(efx);
+	efx_mac_reconfigure(efx);
 
 out:
 	mutex_unlock(&efx->mac_lock);

drivers/net/ethernet/sfc/net_driver.h

···
 #include <linux/highmem.h>
 #include <linux/workqueue.h>
 #include <linux/mutex.h>
+#include <linux/rwsem.h>
 #include <linux/vmalloc.h>
 #include <linux/i2c.h>
 #include <linux/mtd/mtd.h>
···
  * @loopback_mode: Loopback status
  * @loopback_modes: Supported loopback mode bitmask
  * @loopback_selftest: Offline self-test private state
- * @filter_lock: Filter table lock
+ * @filter_sem: Filter table rw_semaphore, for freeing the table
+ * @filter_lock: Filter table lock, for mere content changes
  * @filter_state: Architecture-dependent filter table state
  * @rps_flow_id: Flow IDs of filters allocated for accelerated RFS,
  *	indexed by filter ID
···
 
 	void *loopback_selftest;
 
+	struct rw_semaphore filter_sem;
 	spinlock_t filter_lock;
 	void *filter_state;
 #ifdef CONFIG_RFS_ACCEL