Skip to contentSkip to conversation

i2c: core: Fix possible memleak in i2c_new_client_device()

PATCH In review Yang Yingliang 2 participants 2 messages Nov 24, 2022 – Sep 29, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 i2c: core: Fix possible memleak in i2c_new_client_device()

I got memory leak as follows when doing fault injection test:

unreferenced object 0xffff888014aec078 (size 8):
comm "xrun", pid 356, jiffies 4294910619 (age 16.332s) hex dump (first 8 bytes):

    31 2d 30 30 31 63 00 00                          1-001c..
  backtrace:
    [<00000000eb56c0a9>] __kmalloc_track_caller+0x1a6/0x300
    [<000000000b220ea3>] kvasprintf+0xad/0x140
    [<00000000b83203e5>] kvasprintf_const+0x62/0x190
    [<000000002a5eab37>] kobject_set_name_vargs+0x56/0x140
    [<00000000300ac279>] dev_set_name+0xb0/0xe0
    [<00000000b66ebd6f>] i2c_new_client_device+0x7e4/0x9a0

If device_register() returns error in i2c_new_client_device(), the name allocated by i2c_dev_set_name() need be freed. As comment of device_register() says, it should use put_device() to give up the reference in the error path.

Moving i2c_dev_set_name() to front of device_register(), so it's managed by device core, then call put_device(), when the refcount is 0, the name will be freed in kobject_cleanup() and the 'client' will be freed in i2c_client_dev_release().

Because the 'client' is freed by calling put_device(), so it don't need goto error label, release other resources before put_device() and return directly.

---
 drivers/i2c/i2c-core-base.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)
drivers/i2c/i2c-core-base.c +10 -5
index b4edf10e8fd0..161bf28f2d87 100644
@@ -954,7 +954,6 @@ i2c_new_client_device(struct i2c_adapter *adap, struct i2c_board_info const *inf
 	client->dev.fwnode = info->fwnode;
 
 	device_enable_async_suspend(&client->dev);
-	i2c_dev_set_name(adap, client, info);
 
 	if (info->swnode) {
 		status = device_add_software_node(&client->dev, info->swnode);
@@ -966,17 +965,23 @@ i2c_new_client_device(struct i2c_adapter *adap, struct i2c_board_info const *inf
 		}
 	}
 
+	i2c_dev_set_name(adap, client, info);
 	status = device_register(&client->dev);
-	if (status)
-		goto out_remove_swnode;
+	if (status) {
+		device_remove_software_node(&client->dev);
+		of_node_put(info->of_node);
+		dev_err(&adap->dev,
+			"Failed to register i2c client %s at 0x%02x (%d)\n",
+			client->name, client->addr, status);
+		put_device(&client->dev);
+		return ERR_PTR(status);
+	}
 
 	dev_dbg(&adap->dev, "client [%s] registered with bus id %s\n",
 		client->name, dev_name(&client->dev));
 
 	return client;
 
-out_remove_swnode:
-	device_remove_software_node(&client->dev);
 out_err_put_of_node:
 	of_node_put(info->of_node);
 out_err:
Signature · 1 line
2.25.1
#1 lore raw
Quoted 24 lines · Yang Yingliang

I got memory leak as follows when doing fault injection test:

unreferenced object 0xffff888014aec078 (size 8):
comm "xrun", pid 356, jiffies 4294910619 (age 16.332s) hex dump (first 8 bytes):

    31 2d 30 30 31 63 00 00                          1-001c..
  backtrace:
    [<00000000eb56c0a9>] __kmalloc_track_caller+0x1a6/0x300
    [<000000000b220ea3>] kvasprintf+0xad/0x140
    [<00000000b83203e5>] kvasprintf_const+0x62/0x190
    [<000000002a5eab37>] kobject_set_name_vargs+0x56/0x140
    [<00000000300ac279>] dev_set_name+0xb0/0xe0
    [<00000000b66ebd6f>] i2c_new_client_device+0x7e4/0x9a0

If device_register() returns error in i2c_new_client_device(), the name allocated by i2c_dev_set_name() need be freed. As comment of device_register() says, it should use put_device() to give up the reference in the error path.

Moving i2c_dev_set_name() to front of device_register(), so it's managed by device core, then call put_device(), when the refcount is 0, the name will be freed in kobject_cleanup() and the 'client' will be freed in i2c_client_dev_release().

Quoted 5 lines

Because the 'client' is freed by calling put_device(), so it don't need goto error label, release other resources before put_device() and return directly.

The report seems correct to me, yet the solution is copying code a tad too much IMO. I'll send another approach I'd suggest to spark some discussion. Thanks for the report in any case!