[v2,net-next,03/23] net: add dev_hold_track() and dev_put_track() helpers

Commit Message

Eric Dumazet Dec. 3, 2021, 2:46 a.m. UTC

From: Eric Dumazet <[e-mail]>

They should replace dev_hold() and dev_put().

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

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

Signed-off-by: Eric Dumazet <[e-mail]>
---
 include/linux/netdevice.h | 44 +++++++++++++++++++++++++++++++++++++++
 net/Kconfig               |  8 +++++++
 net/core/dev.c            |  3 +++
 3 files changed, 55 insertions(+)

Patch

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 65117f01d5f2a9c9e815b6c967d5e9e4c94af0ae..daeda6ae8d6520c38ff0ce18bac2dc957940de5d 100644
--- 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;
@@ -2178,6 +2185,7 @@  struct net_device {
 #else
 	refcount_t		dev_refcnt;
 #endif
+	struct ref_tracker_dir	refcnt_tracker;
 
 	struct list_head	link_watch_list;
 
@@ -3805,6 +3813,7 @@  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)
 {
@@ -3822,6 +3831,7 @@  static inline void dev_put(struct net_device *dev)
  *	@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)
 {
@@ -3834,6 +3844,40 @@  static inline void dev_hold(struct net_device *dev)
 	}
 }
 
+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.
diff --git a/net/Kconfig b/net/Kconfig
index 8a1f9d0287de3c32040eee03b60114c6e6d150bc..0b665e60b53490f44eeda1e5506d4e125ef4c53a 100644
--- a/net/Kconfig
+++ b/net/Kconfig
@@ -253,6 +253,14 @@  config PCPU_DEV_REFCNT
 	  network device refcount are using per cpu variables if this option is set.
 	  This can be forced to N to detect underflows (with a performance drop).
 
+config NET_DEV_REFCNT_TRACKER
+	bool "Enable tracking in dev_put_track() and dev_hold_track()"
+	select REF_TRACKER
+	default n
+	help
+	  Enable debugging feature to track leaked device references.
+	  This adds memory and cpu costs.
+
 config RPS
 	bool
 	depends on SMP && SYSFS
diff --git a/net/core/dev.c b/net/core/dev.c
index aba8acc1238c2213b9fc0c47c93568f731f375e5..1740d6cfe86b58359cceaec7ee9cc015a3843723 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -9864,6 +9864,7 @@  static void netdev_wait_allrefs(struct net_device *dev)
 			       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;
 		}
 	}
@@ -10154,6 +10155,7 @@  struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
 	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)
@@ -10270,6 +10272,7 @@  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;