Message ID | 311d64c6-f0e2-da42-5619-1efe46df0007@faberman.de |
---|---|
State | New |
Headers | show |
Series | usb: core: Fix possible memleak in usb_add_gadget | expand |
On Sat, Sep 04, 2021 at 05:34:29PM +0200, Florian Faber wrote: > The memory for the udc structure allocated via kzalloc in line 1295 is not > freed in the error handling code, leading to a memory leak in case of an > error. > > Singed-off-by: Florian Faber <faber@faberman.de> > > --- > drivers/usb/gadget/udc/core.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/drivers/usb/gadget/udc/core.c b/drivers/usb/gadget/udc/core.c > index 14fdf918ecfe..a1270a44855a 100644 > --- a/drivers/usb/gadget/udc/core.c > +++ b/drivers/usb/gadget/udc/core.c > @@ -1346,6 +1346,8 @@ int usb_add_gadget(struct usb_gadget *gadget) > > err_put_udc: > put_device(&udc->dev); > + kfree(udc); > + gadget->udc = NULL; > > error: > return ret; > -- > 2.33.0 > > Flo > -- > Machines can do the work, so people have time to think. Did you test this? I think you will find that you just caused a use-after-free :( Please read the documentation for device_initialize() for why this is not the correct thing to do here. thanks, greg k-h
Greg, On 9/5/21 4:56 PM, Greg KH wrote: > On Sat, Sep 04, 2021 at 05:34:29PM +0200, Florian Faber wrote: >> The memory for the udc structure allocated via kzalloc in line 1295 is not >> freed in the error handling code, leading to a memory leak in case of an >> error. >> >> Singed-off-by: Florian Faber <faber@faberman.de> >> >> --- >> drivers/usb/gadget/udc/core.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/drivers/usb/gadget/udc/core.c b/drivers/usb/gadget/udc/core.c >> index 14fdf918ecfe..a1270a44855a 100644 >> --- a/drivers/usb/gadget/udc/core.c >> +++ b/drivers/usb/gadget/udc/core.c >> @@ -1346,6 +1346,8 @@ int usb_add_gadget(struct usb_gadget *gadget) >> >> err_put_udc: >> put_device(&udc->dev); >> + kfree(udc); >> + gadget->udc = NULL; >> >> error: >> return ret; >> -- >> 2.33.0 >> >> Flo >> -- >> Machines can do the work, so people have time to think. > > Did you test this? I think you will find that you just caused a > use-after-free :( Correct, please forget about this patch. This 'leak' was found by Klocwork and seemed plausible at first oversight. Sorry for wasting your time and not checking it further. > Please read the documentation for device_initialize() for why this is > not the correct thing to do here. I know now :) It was a bit counter intuitive that two different methods are used for memory allocation and freeing. Regarding the other patch: I found the real culprit in the meantime. The UDC driver (broadcom iproc udc, out of tree) did not call composite's disconnect when VBUS is lost. Out of the three gadgets I am using, only mass storage misbehaved that badly, which led me on the wrong track there. Flo -- Machines can do the work, so people have time to think.
On Sun, Sep 05, 2021 at 05:16:36PM +0200, Florian Faber wrote: > Greg, > > On 9/5/21 4:56 PM, Greg KH wrote: > > On Sat, Sep 04, 2021 at 05:34:29PM +0200, Florian Faber wrote: > > > The memory for the udc structure allocated via kzalloc in line 1295 is not > > > freed in the error handling code, leading to a memory leak in case of an > > > error. > > > > > > Singed-off-by: Florian Faber <faber@faberman.de> > > > > > > --- > > > drivers/usb/gadget/udc/core.c | 2 ++ > > > 1 file changed, 2 insertions(+) > > > > > > diff --git a/drivers/usb/gadget/udc/core.c b/drivers/usb/gadget/udc/core.c > > > index 14fdf918ecfe..a1270a44855a 100644 > > > --- a/drivers/usb/gadget/udc/core.c > > > +++ b/drivers/usb/gadget/udc/core.c > > > @@ -1346,6 +1346,8 @@ int usb_add_gadget(struct usb_gadget *gadget) > > > > > > err_put_udc: > > > put_device(&udc->dev); > > > + kfree(udc); > > > + gadget->udc = NULL; > > > > > > error: > > > return ret; > > > -- > > > 2.33.0 > > > > > > Flo > > > -- > > > Machines can do the work, so people have time to think. > > > > Did you test this? I think you will find that you just caused a > > use-after-free :( > > Correct, please forget about this patch. > > This 'leak' was found by Klocwork and seemed plausible at first oversight. > Sorry for wasting your time and not checking it further. What is "Klockwork"? How can it miss the reference counted logic that all drivers use in the kernel? > > Please read the documentation for device_initialize() for why this is > > not the correct thing to do here. > > I know now :) It was a bit counter intuitive that two different methods are > used for memory allocation and freeing. The joy of reference counted stuff, sorry. thanks, greg k-h
On 9/5/21 6:27 PM, Greg KH wrote: > On Sun, Sep 05, 2021 at 05:16:36PM +0200, Florian Faber wrote: >> This 'leak' was found by Klocwork and seemed plausible at first oversight. >> Sorry for wasting your time and not checking it further. > > What is "Klockwork"? Some static code analysis tool. I did not know it either. > How can it miss the reference counted logic that > all drivers use in the kernel? Obviously it is not running like clockwork :) Anyway, got the problem fixed, sorry for the noise. Flo -- Machines can do the work, so people have time to think.
diff --git a/drivers/usb/gadget/udc/core.c b/drivers/usb/gadget/udc/core.c index 14fdf918ecfe..a1270a44855a 100644 --- a/drivers/usb/gadget/udc/core.c +++ b/drivers/usb/gadget/udc/core.c @@ -1346,6 +1346,8 @@ int usb_add_gadget(struct usb_gadget *gadget) err_put_udc: put_device(&udc->dev); + kfree(udc); + gadget->udc = NULL; error: return ret;