Message ID | 20171011103646.11879-5-tiwai@suse.de (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Wed, Oct 11, 2017 at 12:36:41PM +0200, Takashi Iwai wrote: > As syzkaller spotted, currently line6 drivers submit a URB with the > fixed EP without checking whether it's actually available, which may > result in a kernel warning like: > usb 1-1: BOGUS urb xfer, pipe 3 != type 1 > ------------[ cut here ]------------ > WARNING: CPU: 0 PID: 24 at drivers/usb/core/urb.c:449 > usb_submit_urb+0xf8a/0x11d0 > Modules linked in: > CPU: 0 PID: 24 Comm: kworker/0:1 Not tainted 4.14.0-rc2-42613-g1488251d1a98 #238 > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011 > Workqueue: usb_hub_wq hub_event > Call Trace: > line6_start_listen+0x55f/0x9e0 sound/usb/line6/driver.c:82 > line6_init_cap_control sound/usb/line6/driver.c:690 > line6_probe+0x7c9/0x1310 sound/usb/line6/driver.c:764 > podhd_probe+0x64/0x70 sound/usb/line6/podhd.c:474 > usb_probe_interface+0x35d/0x8e0 drivers/usb/core/driver.c:361 > .... > > This patch adds a sanity check of validity of EPs at the device > initialization phase for avoiding the call with an invalid EP. > > Reported-by: Andrey Konovalov <andreyknvl@google.com> > Tested-by: Andrey Konovalov <andreyknvl@google.com> > Signed-off-by: Takashi Iwai <tiwai@suse.de> > --- > sound/usb/line6/driver.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/sound/usb/line6/driver.c b/sound/usb/line6/driver.c > index 0ff5a7d2e19f..0da6f68761e3 100644 > --- a/sound/usb/line6/driver.c > +++ b/sound/usb/line6/driver.c > @@ -78,6 +78,13 @@ static int line6_start_listen(struct usb_line6 *line6) > line6->buffer_listen, LINE6_BUFSIZE_LISTEN, > line6_data_received, line6); > } > + > + /* sanity checks of EP before actually submitting */ > + if (usb_urb_ep_type_check(line6->urb_listen)) { > + dev_err(line6->ifcdev, "invalid control EP\n"); > + return -EINVAL; > + } You're now doing this check twice (here and in usb_submit_urb) every time this URB is submitted rather just checking it once in probe. Seems like a quick band-aid to me. > + > line6->urb_listen->actual_length = 0; > err = usb_submit_urb(line6->urb_listen, GFP_ATOMIC); > return err; Johan
On Wed, 11 Oct 2017 16:28:37 +0200, Johan Hovold wrote: > > On Wed, Oct 11, 2017 at 12:36:41PM +0200, Takashi Iwai wrote: > > As syzkaller spotted, currently line6 drivers submit a URB with the > > fixed EP without checking whether it's actually available, which may > > result in a kernel warning like: > > usb 1-1: BOGUS urb xfer, pipe 3 != type 1 > > ------------[ cut here ]------------ > > WARNING: CPU: 0 PID: 24 at drivers/usb/core/urb.c:449 > > usb_submit_urb+0xf8a/0x11d0 > > Modules linked in: > > CPU: 0 PID: 24 Comm: kworker/0:1 Not tainted 4.14.0-rc2-42613-g1488251d1a98 #238 > > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011 > > Workqueue: usb_hub_wq hub_event > > Call Trace: > > line6_start_listen+0x55f/0x9e0 sound/usb/line6/driver.c:82 > > line6_init_cap_control sound/usb/line6/driver.c:690 > > line6_probe+0x7c9/0x1310 sound/usb/line6/driver.c:764 > > podhd_probe+0x64/0x70 sound/usb/line6/podhd.c:474 > > usb_probe_interface+0x35d/0x8e0 drivers/usb/core/driver.c:361 > > .... > > > > This patch adds a sanity check of validity of EPs at the device > > initialization phase for avoiding the call with an invalid EP. > > > > Reported-by: Andrey Konovalov <andreyknvl@google.com> > > Tested-by: Andrey Konovalov <andreyknvl@google.com> > > Signed-off-by: Takashi Iwai <tiwai@suse.de> > > --- > > sound/usb/line6/driver.c | 7 +++++++ > > 1 file changed, 7 insertions(+) > > > > diff --git a/sound/usb/line6/driver.c b/sound/usb/line6/driver.c > > index 0ff5a7d2e19f..0da6f68761e3 100644 > > --- a/sound/usb/line6/driver.c > > +++ b/sound/usb/line6/driver.c > > @@ -78,6 +78,13 @@ static int line6_start_listen(struct usb_line6 *line6) > > line6->buffer_listen, LINE6_BUFSIZE_LISTEN, > > line6_data_received, line6); > > } > > + > > + /* sanity checks of EP before actually submitting */ > > + if (usb_urb_ep_type_check(line6->urb_listen)) { > > + dev_err(line6->ifcdev, "invalid control EP\n"); > > + return -EINVAL; > > + } > > You're now doing this check twice (here and in usb_submit_urb) every > time this URB is submitted rather just checking it once in probe. > > Seems like a quick band-aid to me. Yes, it is :) We may split line6_start_listen() to two functions, one for initializing URBs and another for submission, but it'll be more codes and the check is fairly cheap. So again it's a question whether easier to read or to get minor optimization. thanks, Takashi
On Wed, Oct 11, 2017 at 04:45:16PM +0200, Takashi Iwai wrote: > On Wed, 11 Oct 2017 16:28:37 +0200, > Johan Hovold wrote: > > > > On Wed, Oct 11, 2017 at 12:36:41PM +0200, Takashi Iwai wrote: > > > As syzkaller spotted, currently line6 drivers submit a URB with the > > > fixed EP without checking whether it's actually available, which may > > > result in a kernel warning like: > > > usb 1-1: BOGUS urb xfer, pipe 3 != type 1 > > > ------------[ cut here ]------------ > > > WARNING: CPU: 0 PID: 24 at drivers/usb/core/urb.c:449 > > > usb_submit_urb+0xf8a/0x11d0 > > > Modules linked in: > > > CPU: 0 PID: 24 Comm: kworker/0:1 Not tainted 4.14.0-rc2-42613-g1488251d1a98 #238 > > > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011 > > > Workqueue: usb_hub_wq hub_event > > > Call Trace: > > > line6_start_listen+0x55f/0x9e0 sound/usb/line6/driver.c:82 > > > line6_init_cap_control sound/usb/line6/driver.c:690 > > > line6_probe+0x7c9/0x1310 sound/usb/line6/driver.c:764 > > > podhd_probe+0x64/0x70 sound/usb/line6/podhd.c:474 > > > usb_probe_interface+0x35d/0x8e0 drivers/usb/core/driver.c:361 > > > .... > > > > > > This patch adds a sanity check of validity of EPs at the device > > > initialization phase for avoiding the call with an invalid EP. > > > > > > Reported-by: Andrey Konovalov <andreyknvl@google.com> > > > Tested-by: Andrey Konovalov <andreyknvl@google.com> > > > Signed-off-by: Takashi Iwai <tiwai@suse.de> > > > --- > > > sound/usb/line6/driver.c | 7 +++++++ > > > 1 file changed, 7 insertions(+) > > > > > > diff --git a/sound/usb/line6/driver.c b/sound/usb/line6/driver.c > > > index 0ff5a7d2e19f..0da6f68761e3 100644 > > > --- a/sound/usb/line6/driver.c > > > +++ b/sound/usb/line6/driver.c > > > @@ -78,6 +78,13 @@ static int line6_start_listen(struct usb_line6 *line6) > > > line6->buffer_listen, LINE6_BUFSIZE_LISTEN, > > > line6_data_received, line6); > > > } > > > + > > > + /* sanity checks of EP before actually submitting */ > > > + if (usb_urb_ep_type_check(line6->urb_listen)) { > > > + dev_err(line6->ifcdev, "invalid control EP\n"); > > > + return -EINVAL; > > > + } > > > > You're now doing this check twice (here and in usb_submit_urb) every > > time this URB is submitted rather just checking it once in probe. > > > > Seems like a quick band-aid to me. > > Yes, it is :) > > We may split line6_start_listen() to two functions, one for > initializing URBs and another for submission, but it'll be more codes > and the check is fairly cheap. So again it's a question whether > easier to read or to get minor optimization. I'd say it's also a question about long-term maintainability. Those quick band-aids tend to keep piling up over time. Johan
On Wed, 11 Oct 2017 16:52:09 +0200, Johan Hovold wrote: > > On Wed, Oct 11, 2017 at 04:45:16PM +0200, Takashi Iwai wrote: > > On Wed, 11 Oct 2017 16:28:37 +0200, > > Johan Hovold wrote: > > > > > > On Wed, Oct 11, 2017 at 12:36:41PM +0200, Takashi Iwai wrote: > > > > As syzkaller spotted, currently line6 drivers submit a URB with the > > > > fixed EP without checking whether it's actually available, which may > > > > result in a kernel warning like: > > > > usb 1-1: BOGUS urb xfer, pipe 3 != type 1 > > > > ------------[ cut here ]------------ > > > > WARNING: CPU: 0 PID: 24 at drivers/usb/core/urb.c:449 > > > > usb_submit_urb+0xf8a/0x11d0 > > > > Modules linked in: > > > > CPU: 0 PID: 24 Comm: kworker/0:1 Not tainted 4.14.0-rc2-42613-g1488251d1a98 #238 > > > > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011 > > > > Workqueue: usb_hub_wq hub_event > > > > Call Trace: > > > > line6_start_listen+0x55f/0x9e0 sound/usb/line6/driver.c:82 > > > > line6_init_cap_control sound/usb/line6/driver.c:690 > > > > line6_probe+0x7c9/0x1310 sound/usb/line6/driver.c:764 > > > > podhd_probe+0x64/0x70 sound/usb/line6/podhd.c:474 > > > > usb_probe_interface+0x35d/0x8e0 drivers/usb/core/driver.c:361 > > > > .... > > > > > > > > This patch adds a sanity check of validity of EPs at the device > > > > initialization phase for avoiding the call with an invalid EP. > > > > > > > > Reported-by: Andrey Konovalov <andreyknvl@google.com> > > > > Tested-by: Andrey Konovalov <andreyknvl@google.com> > > > > Signed-off-by: Takashi Iwai <tiwai@suse.de> > > > > --- > > > > sound/usb/line6/driver.c | 7 +++++++ > > > > 1 file changed, 7 insertions(+) > > > > > > > > diff --git a/sound/usb/line6/driver.c b/sound/usb/line6/driver.c > > > > index 0ff5a7d2e19f..0da6f68761e3 100644 > > > > --- a/sound/usb/line6/driver.c > > > > +++ b/sound/usb/line6/driver.c > > > > @@ -78,6 +78,13 @@ static int line6_start_listen(struct usb_line6 *line6) > > > > line6->buffer_listen, LINE6_BUFSIZE_LISTEN, > > > > line6_data_received, line6); > > > > } > > > > + > > > > + /* sanity checks of EP before actually submitting */ > > > > + if (usb_urb_ep_type_check(line6->urb_listen)) { > > > > + dev_err(line6->ifcdev, "invalid control EP\n"); > > > > + return -EINVAL; > > > > + } > > > > > > You're now doing this check twice (here and in usb_submit_urb) every > > > time this URB is submitted rather just checking it once in probe. > > > > > > Seems like a quick band-aid to me. > > > > Yes, it is :) > > > > We may split line6_start_listen() to two functions, one for > > initializing URBs and another for submission, but it'll be more codes > > and the check is fairly cheap. So again it's a question whether > > easier to read or to get minor optimization. > > I'd say it's also a question about long-term maintainability. Those > quick band-aids tend to keep piling up over time. The codes we've touched are mostly for device-specific drivers or quirk codes, and they are basically finished ones, so I guess it's a kind of once-off fix. Let's see. thanks, Takashi
diff --git a/sound/usb/line6/driver.c b/sound/usb/line6/driver.c index 0ff5a7d2e19f..0da6f68761e3 100644 --- a/sound/usb/line6/driver.c +++ b/sound/usb/line6/driver.c @@ -78,6 +78,13 @@ static int line6_start_listen(struct usb_line6 *line6) line6->buffer_listen, LINE6_BUFSIZE_LISTEN, line6_data_received, line6); } + + /* sanity checks of EP before actually submitting */ + if (usb_urb_ep_type_check(line6->urb_listen)) { + dev_err(line6->ifcdev, "invalid control EP\n"); + return -EINVAL; + } + line6->urb_listen->actual_length = 0; err = usb_submit_urb(line6->urb_listen, GFP_ATOMIC); return err;