From: Eric Dumazet <edumazet@google.com>
Date: Sat, 4 Dec 2021 20:21:57 -0800
Subject: net: add net device refcount tracker infrastructure
Patch-mainline: v5.17-rc1
Git-commit: 4d92b95ff2f95f13df9bad0b5a25a9f60e72758d
References: jsc#PED-1565

net device are refcounted. Over the years we had numerous bugs
caused by imbalanced dev_hold() and dev_put() calls.

The general idea is to be able to precisely pair each decrement with
a corresponding prior increment. Both share a cookie, basically
a pointer to private data storing stack traces.

This patch adds dev_hold_track() and dev_put_track().

To use these helpers, each data structure owning a refcount
should also use a "netdevice_tracker" to pair the hold and put.

netdevice_tracker dev_tracker;
...
dev_hold_track(dev, &dev_tracker, GFP_ATOMIC);
...
dev_put_track(dev, &dev_tracker);

Whenever a leak happens, we will get precise stack traces
of the point dev_hold_track() happened, at device dismantle phase.

We will also get a stack trace if too many dev_put_track() for the same
netdevice_tracker are attempted.

This is guarded by CONFIG_NET_DEV_REFCNT_TRACKER option.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
 include/linux/netdevice.h |   45 +++++++++++++++++++++++++++++++++++++++++++++
 lib/Kconfig.debug         |    5 +++++
 net/Kconfig.debug         |   10 ++++++++++
 net/core/dev.c            |    3 +++
 4 files changed, 63 insertions(+)
 create mode 100644 net/Kconfig.debug

--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -48,6 +48,7 @@
 #include <uapi/linux/pkt_cls.h>
 #include <linux/hashtable.h>
 #include <linux/rbtree.h>
+#include <linux/ref_tracker.h>
 
 struct netpoll_info;
 struct device;
@@ -300,6 +301,12 @@ enum netdev_state_t {
 };
 
 
+#ifdef CONFIG_NET_DEV_REFCNT_TRACKER
+typedef struct ref_tracker *netdevice_tracker;
+#else
+typedef struct {} netdevice_tracker;
+#endif
+
 struct gro_list {
 	struct list_head	list;
 	int			count;
@@ -1859,6 +1866,7 @@ enum netdev_ml_priv_type {
  *	@proto_down_reason:	reason a netdev interface is held down
  *	@pcpu_refcnt:		Number of references to this device
  *	@dev_refcnt:		Number of references to this device
+ *	@refcnt_tracker:	Tracker directory for tracked references to this device
  *	@todo_list:		Delayed register/unregister
  *	@link_watch_list:	XXX: need comments on this one
  *
@@ -2166,6 +2174,7 @@ struct net_device {
 #else
 	refcount_t		dev_refcnt;
 #endif
+	struct ref_tracker_dir	refcnt_tracker;
 
 	struct list_head	link_watch_list;
 
@@ -3793,6 +3802,7 @@ void netdev_run_todo(void);
  *	@dev: network device
  *
  * Release reference to device to allow it to be freed.
+ * Try using dev_put_track() instead.
  */
 static inline void dev_put(struct net_device *dev)
 {
@@ -3810,6 +3820,7 @@ static inline void dev_put(struct net_de
  *	@dev: network device
  *
  * Hold reference to device to keep it from being freed.
+ * Try using dev_hold_track() instead.
  */
 static inline void dev_hold(struct net_device *dev)
 {
@@ -3822,6 +3833,40 @@ static inline void dev_hold(struct net_d
 	}
 }
 
+static inline void netdev_tracker_alloc(struct net_device *dev,
+					netdevice_tracker *tracker, gfp_t gfp)
+{
+#ifdef CONFIG_NET_DEV_REFCNT_TRACKER
+	ref_tracker_alloc(&dev->refcnt_tracker, tracker, gfp);
+#endif
+}
+
+static inline void netdev_tracker_free(struct net_device *dev,
+				       netdevice_tracker *tracker)
+{
+#ifdef CONFIG_NET_DEV_REFCNT_TRACKER
+	ref_tracker_free(&dev->refcnt_tracker, tracker);
+#endif
+}
+
+static inline void dev_hold_track(struct net_device *dev,
+				  netdevice_tracker *tracker, gfp_t gfp)
+{
+	if (dev) {
+		dev_hold(dev);
+		netdev_tracker_alloc(dev, tracker, gfp);
+	}
+}
+
+static inline void dev_put_track(struct net_device *dev,
+				 netdevice_tracker *tracker)
+{
+	if (dev) {
+		netdev_tracker_free(dev, tracker);
+		dev_put(dev);
+	}
+}
+
 /* Carrier loss detection, dial on demand. The functions netif_carrier_on
  * and _off may be called from IRQ context, but it is caller
  * who is responsible for serialization of these calls.
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -596,6 +596,11 @@ config DEBUG_MISC
 	  Say Y here if you need to enable miscellaneous debug code that should
 	  be under a more specific debug option but isn't.
 
+menu "Networking Debugging"
+
+source "net/Kconfig.debug"
+
+endmenu # "Networking Debugging"
 
 menu "Memory Debugging"
 
--- /dev/null
+++ b/net/Kconfig.debug
@@ -0,0 +1,10 @@
+# SPDX-License-Identifier: GPL-2.0-only
+
+config NET_DEV_REFCNT_TRACKER
+	bool "Enable net device refcount tracking"
+	depends on DEBUG_KERNEL && STACKTRACE_SUPPORT
+	select REF_TRACKER
+	default n
+	help
+	  Enable debugging feature to track device references.
+	  This adds memory and cpu costs.
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -9863,6 +9863,7 @@ static void netdev_wait_allrefs(struct n
 			       netdev_unregister_timeout_secs * HZ)) {
 			pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
 				 dev->name, refcnt);
+			ref_tracker_dir_print(&dev->refcnt_tracker, 10);
 			warning_time = jiffies;
 		}
 	}
@@ -10153,6 +10154,7 @@ struct net_device *alloc_netdev_mqs(int
 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
 	dev->padded = (char *)dev - (char *)p;
 
+	ref_tracker_dir_init(&dev->refcnt_tracker, 128);
 #ifdef CONFIG_PCPU_DEV_REFCNT
 	dev->pcpu_refcnt = alloc_percpu(int);
 	if (!dev->pcpu_refcnt)
@@ -10269,6 +10271,7 @@ void free_netdev(struct net_device *dev)
 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
 		netif_napi_del(p);
 
+	ref_tracker_dir_exit(&dev->refcnt_tracker);
 #ifdef CONFIG_PCPU_DEV_REFCNT
 	free_percpu(dev->pcpu_refcnt);
 	dev->pcpu_refcnt = NULL;