Quick summary: syzbot found a memory leak in the gspca driver, apparently caused by a reference being taken but not released in a probe failure pathway. On Mon, Nov 23, 2020 at 12:48:08PM -0800, syzbot wrote: > Hello, > > syzbot has tested the proposed patch but the reproducer is still triggering an issue: > memory leak in usb_set_configuration > > BUG: memory leak > unreferenced object 0xffff8881268a1800 (size 1024): > comm "kworker/0:2", pid 3644, jiffies 4294944749 (age 12.820s) > hex dump (first 32 bytes): > 48 f3 6a 26 81 88 ff ff 48 f3 6a 26 81 88 ff ff H.j&....H.j&.... > 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ > backtrace: > [<0000000089bfe22c>] kmalloc include/linux/slab.h:552 [inline] > [<0000000089bfe22c>] kzalloc include/linux/slab.h:664 [inline] > [<0000000089bfe22c>] usb_set_configuration+0x18c/0xb90 drivers/usb/core/message.c:1987 > [<00000000207f81de>] usb_generic_driver_probe+0x8c/0xc0 drivers/usb/core/generic.c:238 > [<000000007aa490e0>] usb_probe_device+0x5c/0x140 drivers/usb/core/driver.c:293 Okay, this does seem to pinpoint the problem. gspca_dev_probe2() calls v4l2_device_register() at the start, but doesn't call v4l2_device_disconnect() upon failure. I'm not at all familiar with the design of the v4l2 subsystem. Mauro or Hans: Is this the right solution? Alan Stern #syz test: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git 4d02da97 Index: usb-devel/drivers/media/usb/gspca/gspca.c =================================================================== --- usb-devel.orig/drivers/media/usb/gspca/gspca.c +++ usb-devel/drivers/media/usb/gspca/gspca.c @@ -1575,6 +1575,7 @@ out: input_unregister_device(gspca_dev->input_dev); #endif v4l2_ctrl_handler_free(gspca_dev->vdev.ctrl_handler); + v4l2_device_disconnect(&gspca_dev->v4l2_dev); kfree(gspca_dev->usb_buf); kfree(gspca_dev); return ret;
