On Mon, Feb 14, 2022 at 03:34:42PM +0800, Dongliang Mu wrote:

On Sat, Feb 12, 2022 at 9:50 AM Alan Stern [off-list ref] wrote:

quoted

Syzbot identified a refcount leak in the hid-elo driver:

BUG: memory leak
unreferenced object 0xffff88810d49e800 (size 2048):
  comm "kworker/1:1", pid 25, jiffies 4294954629 (age 16.460s)
  hex dump (first 32 bytes):
    ff ff ff ff 31 00 00 00 00 00 00 00 00 00 00 00  ....1...........
    00 00 00 00 00 00 00 00 00 00 00 00 03 00 00 00  ................
  backtrace:
    [<ffffffff82c87a62>] kmalloc include/linux/slab.h:581 [inline]
    [<ffffffff82c87a62>] kzalloc include/linux/slab.h:715 [inline]
    [<ffffffff82c87a62>] usb_alloc_dev+0x32/0x450 drivers/usb/core/usb.c:582
    [<ffffffff82c91a47>] hub_port_connect drivers/usb/core/hub.c:5260 [inline]
    [<ffffffff82c91a47>] hub_port_connect_change drivers/usb/core/hub.c:5502 [inline]
    [<ffffffff82c91a47>] port_event drivers/usb/core/hub.c:5660 [inline]
    [<ffffffff82c91a47>] hub_event+0x1097/0x21a0 drivers/usb/core/hub.c:5742
    [<ffffffff8126c3ef>] process_one_work+0x2bf/0x600 kernel/workqueue.c:2307
    [<ffffffff8126ccd9>] worker_thread+0x59/0x5b0 kernel/workqueue.c:2454
    [<ffffffff81276765>] kthread+0x125/0x160 kernel/kthread.c:377
    [<ffffffff810022ff>] ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295

Not shown in the bug report but present in the console log:

[  182.014764][ T3257] elo 0003:04E7:0030.0006: item fetching failed at offset 0/1
[  182.022255][ T3257] elo 0003:04E7:0030.0006: parse failed
[  182.027904][ T3257] elo: probe of 0003:04E7:0030.0006 failed with error -22
[  182.214767][ T3257] usb 1-1: USB disconnect, device number 7
[  188.090199][ T3604] kmemleak: 3 new suspected memory leaks (see /sys/kernel/debug/kmemleak)
BUG: memory leak

which points to hid-elo as the buggy driver.

The leak is caused by elo_probe() failing to release the reference it
holds to the struct usb_device in its failure pathway.  In the end the
driver doesn't need to take this reference at all, because the

Hi Alan,

My patch "[PATCH] hid: elo: fix memory leak in elo_probe" is merged
several weeks ago.

Really?  It still isn't in Linus's tree as of 5.17-rc4.  I would expect 
a bug fix to go upstream as soon as possible.

However, I fix this bug by modifying the error handling code in
elo_probe. If you think the refcount is not necessary, maybe a new
patch to remove the refcount is better.

The refcount was added less than a year ago by Salah Triki in commit 
fbf42729d0e9 ("HID: elo: update the reference count of the usb device 
structure"), but the commit message doesn't explain why it is 
necessary.  There certainly isn't any obvious reason for it; the driver 
doesn't release any references after elo_remove() returns and we know 
that the usb_device structure won't be deallocated before the driver 
gets unbound.

Alan Stern