On Mon, 12 Aug 2019, syzbot wrote: > Hello, > > syzbot has tested the proposed patch but the reproducer still triggered > crash: > KASAN: invalid-free in hcd_buffer_free This bug report shows that Hillf's fix isn't exactly right. > usb 5-1: USB disconnect, device number 2 > ================================================================== > BUG: KASAN: double-free or invalid-free in hcd_buffer_free+0x199/0x260 > drivers/usb/core/buffer.c:165 > > CPU: 0 PID: 1745 Comm: kworker/0:2 Not tainted 5.3.0-rc2+ #1 > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS > Google 01/01/2011 > Workqueue: usb_hub_wq hub_event > Call Trace: > __dump_stack lib/dump_stack.c:77 [inline] > dump_stack+0xca/0x13e lib/dump_stack.c:113 > print_address_description+0x6a/0x32c mm/kasan/report.c:351 > kasan_report_invalid_free+0x61/0xa0 mm/kasan/report.c:444 > __kasan_slab_free+0x162/0x180 mm/kasan/common.c:428 > slab_free_hook mm/slub.c:1423 [inline] > slab_free_freelist_hook mm/slub.c:1470 [inline] > slab_free mm/slub.c:3012 [inline] > kfree+0xe4/0x2f0 mm/slub.c:3953 > hcd_buffer_free+0x199/0x260 drivers/usb/core/buffer.c:165 > usb_free_coherent+0x67/0x80 drivers/usb/core/usb.c:932 > hid_free_buffers.isra.0+0x94/0x290 drivers/hid/usbhid/hid-core.c:964 > usbhid_stop+0x308/0x450 drivers/hid/usbhid/hid-core.c:1224 > logi_dj_remove+0x107/0x210 drivers/hid/hid-logitech-dj.c:1797 Here the double-free occurred when logi_dj_remove() called hd_hw_stop()... > hid_device_remove+0xed/0x240 drivers/hid/hid-core.c:2242 > __device_release_driver drivers/base/dd.c:1118 [inline] > device_release_driver_internal+0x206/0x4c0 drivers/base/dd.c:1151 > bus_remove_device+0x2dc/0x4a0 drivers/base/bus.c:556 > device_del+0x420/0xb10 drivers/base/core.c:2288 > hid_remove_device drivers/hid/hid-core.c:2413 [inline] > hid_destroy_device+0xe1/0x150 drivers/hid/hid-core.c:2432 > usbhid_disconnect+0xad/0xd0 drivers/hid/usbhid/hid-core.c:1414 which occurred inside usbhid_disconnect()'s call to hid_destroy_device(). But just above the call to hid_destroy_device(), Hillf's patch adds a direct call to hid_hw_stop(), which is what did the original free. So it looks like the problem here is that some paths in the original unpatched code end up calling hid_hw_stop() by way of the hid_device's driver, and other paths do not. I haven't had time to track down this difference. Maybe somebody on the mailing list already knows why it occurs. Alan Stern
