]> git.karo-electronics.de Git - karo-tx-linux.git/commit
staging: comedi: comedi_test: fix race when cancelling command
authorIan Abbott <abbotti@mev.co.uk>
Fri, 4 Jan 2013 11:33:21 +0000 (11:33 +0000)
committerBen Hutchings <ben@decadent.org.uk>
Wed, 6 Feb 2013 04:33:25 +0000 (04:33 +0000)
commit41e52c46717b0120d247d04d330f16d1501d357d
tree52e36e9ebf50b18acc00c8235390c69e8eecd8c6
parent3288bff62342281c6c6cdd2df6e6fc19afe98498
staging: comedi: comedi_test: fix race when cancelling command

commit c0729eeefdcd76db338f635162bf0739fd2c5f6f upstream.

Éric Piel reported a kernel oops in the "comedi_test" module.  It was a
NULL pointer dereference within `waveform_ai_interrupt()` (actually a
timer function) that sometimes occurred when a running asynchronous
command is cancelled (either by the `COMEDI_CANCEL` ioctl or by closing
the device file).

This seems to be a race between the caller of `waveform_ai_cancel()`
which on return from that function goes and tears down the running
command, and the timer function which uses the command.  In particular,
`async->cmd.chanlist` gets freed (and the pointer set to NULL) by
`do_become_nonbusy()` in "comedi_fops.c" but a previously scheduled
`waveform_ai_interrupt()` timer function will dereference that pointer
regardless, leading to the oops.

Fix it by replacing the `del_timer()` call in `waveform_ai_cancel()`
with `del_timer_sync()`.

Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Reported-by: Éric Piel <piel@delmic.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
drivers/staging/comedi/drivers/comedi_test.c