2008-12-25 23:14:44

by Khem Raj

[permalink] [raw]
Subject: [patch] hid2hci check usb_init return value

Hi

Working on a system which did not have a USB in it. I was getting a
segfault with hid2hci. The problem tracked down was usb_init () was
failing and we did not check for it and carried on with execution.
As a result the next call to libusb failed because we passed in NULL for
ctx because during usb_init () this should have been malloc'ed if all
have gone correctly but in case of error usb_init () free's up the
allocated memory and hence the segfault.

We should check for return value of usb_init () call and exit
immediately if it fails. Tested on 4.24 release.

OK ?

Thanks

-Khem



Attachments:
(No filename) (597.00 B)
hid2hci_usb_init.patch (1.08 kB)
Download all attachments

2008-12-30 06:26:56

by Khem Raj

[permalink] [raw]
Subject: Re: [patch] hid2hci check usb_init return value

On (29/12/08 13:12), Marcel Holtmann wrote:
> Hi Khem,
>
> > Working on a system which did not have a USB in it. I was getting a
> > segfault with hid2hci. The problem tracked down was usb_init () was
> > failing and we did not check for it and carried on with execution.
> > As a result the next call to libusb failed because we passed in NULL for
> > ctx because during usb_init () this should have been malloc'ed if all
> > have gone correctly but in case of error usb_init () free's up the
> > allocated memory and hence the segfault.
> >
> > We should check for return value of usb_init () call and exit
> > immediately if it fails. Tested on 4.24 release.
>
> the problem is that usb_init() doesn't return any errors on the versions
> that I have here:
>
> ii libusb-0.1-4 2:0.1.12-12 userspace USB programming library
> ii libusb-dev 2:0.1.12-12 userspace USB programming library
>
> Applying your patch would break complication on multiple platforms.
What you have is libusb-0.1 which has older API I suppose. There is
libusb-compat which can be used as a wrapper over libusb1 to give
libusb0 like interfaces.

Actually I was using libusb-1.0.0 API as mentioned here
http://libusb.sourceforge.net/api-1.0/group__lib.html
that also means that the patch I attached should have been using
libusb_init() instead.

Here is updated patch

Thanks

-Khem


Attachments:
(No filename) (1.35 kB)
hid2hci_usb_init.patch (1.08 kB)
Download all attachments

2008-12-29 12:12:31

by Marcel Holtmann

[permalink] [raw]
Subject: Re: [patch] hid2hci check usb_init return value

Hi Khem,

> Working on a system which did not have a USB in it. I was getting a
> segfault with hid2hci. The problem tracked down was usb_init () was
> failing and we did not check for it and carried on with execution.
> As a result the next call to libusb failed because we passed in NULL for
> ctx because during usb_init () this should have been malloc'ed if all
> have gone correctly but in case of error usb_init () free's up the
> allocated memory and hence the segfault.
>
> We should check for return value of usb_init () call and exit
> immediately if it fails. Tested on 4.24 release.

the problem is that usb_init() doesn't return any errors on the versions
that I have here:

ii libusb-0.1-4 2:0.1.12-12 userspace USB programming library
ii libusb-dev 2:0.1.12-12 userspace USB programming library

Applying your patch would break complication on multiple platforms.

Regards

Marcel



2009-01-06 18:21:04

by Marcel Holtmann

[permalink] [raw]
Subject: Re: [patch] hid2hci check usb_init return value

Hi Khem,

> > > > > Working on a system which did not have a USB in it. I was getting a
> > > > > segfault with hid2hci. The problem tracked down was usb_init () was
> > > > > failing and we did not check for it and carried on with execution.
> > > > > As a result the next call to libusb failed because we passed in NULL for
> > > > > ctx because during usb_init () this should have been malloc'ed if all
> > > > > have gone correctly but in case of error usb_init () free's up the
> > > > > allocated memory and hence the segfault.
> > > > >
> > > > > We should check for return value of usb_init () call and exit
> > > > > immediately if it fails. Tested on 4.24 release.
> > > >
> > > > the problem is that usb_init() doesn't return any errors on the versions
> > > > that I have here:
> > > >
> > > > ii libusb-0.1-4 2:0.1.12-12 userspace USB programming library
> > > > ii libusb-dev 2:0.1.12-12 userspace USB programming library
> > > >
> > > > Applying your patch would break complication on multiple platforms.
> > > What you have is libusb-0.1 which has older API I suppose. There is
> > > libusb-compat which can be used as a wrapper over libusb1 to give
> > > libusb0 like interfaces.
> > >
> > > Actually I was using libusb-1.0.0 API as mentioned here
> > > http://libusb.sourceforge.net/api-1.0/group__lib.html
> > > that also means that the patch I attached should have been using
> > > libusb_init() instead.
> > >
> > > Here is updated patch
> >
> > this still breaks with a system that has libusb-0.1 installed like the
> > current Ubuntu 8.10 does. So either we switch to libusb-1.0 completely
> > or we leave it as it is.
>
> Moving forward depending on libusb-1.0 seems a better idea to me.

feel free to submit patches for all utilities with USB support. Then we
can talk about switching.

Regards

Marcel



2009-01-06 18:17:20

by Khem Raj

[permalink] [raw]
Subject: Re: [patch] hid2hci check usb_init return value

On (06/01/09 03:54), Marcel Holtmann wrote:
> Hi Khem,
>
> > > > Working on a system which did not have a USB in it. I was getting a
> > > > segfault with hid2hci. The problem tracked down was usb_init () was
> > > > failing and we did not check for it and carried on with execution.
> > > > As a result the next call to libusb failed because we passed in NULL for
> > > > ctx because during usb_init () this should have been malloc'ed if all
> > > > have gone correctly but in case of error usb_init () free's up the
> > > > allocated memory and hence the segfault.
> > > >
> > > > We should check for return value of usb_init () call and exit
> > > > immediately if it fails. Tested on 4.24 release.
> > >
> > > the problem is that usb_init() doesn't return any errors on the versions
> > > that I have here:
> > >
> > > ii libusb-0.1-4 2:0.1.12-12 userspace USB programming library
> > > ii libusb-dev 2:0.1.12-12 userspace USB programming library
> > >
> > > Applying your patch would break complication on multiple platforms.
> > What you have is libusb-0.1 which has older API I suppose. There is
> > libusb-compat which can be used as a wrapper over libusb1 to give
> > libusb0 like interfaces.
> >
> > Actually I was using libusb-1.0.0 API as mentioned here
> > http://libusb.sourceforge.net/api-1.0/group__lib.html
> > that also means that the patch I attached should have been using
> > libusb_init() instead.
> >
> > Here is updated patch
>
> this still breaks with a system that has libusb-0.1 installed like the
> current Ubuntu 8.10 does. So either we switch to libusb-1.0 completely
> or we leave it as it is.

Moving forward depending on libusb-1.0 seems a better idea to me.

Thanks

-Khem

2009-01-06 05:19:14

by Marcel Holtmann

[permalink] [raw]
Subject: Re: [patch] hid2hci check usb_init return value

Hi David,

> >> What you have is libusb-0.1 which has older API I suppose. There is
> >> libusb-compat which can be used as a wrapper over libusb1 to give
> >> libusb0 like interfaces.
> >>
> >> Actually I was using libusb-1.0.0 API as mentioned here
> >> http://libusb.sourceforge.net/api-1.0/group__lib.html
> >> that also means that the patch I attached should have been using
> >> libusb_init() instead.
> >>
> >> Here is updated patch
> >>
> >
> > this still breaks with a system that has libusb-0.1 installed like the
> > current Ubuntu 8.10 does. So either we switch to libusb-1.0 completely
> > or we leave it as it is.
> >
>
> This looks like a job for autoconf?

it could be, but I am only doing either libusb-0.1 or libusb-1.0. We can
have support for both, but these compile time options are a bad thing
and I don't want them.

Regards

Marcel



2009-01-06 03:02:40

by David Sainty

[permalink] [raw]
Subject: Re: [patch] hid2hci check usb_init return value

Marcel Holtmann wrote:
>> What you have is libusb-0.1 which has older API I suppose. There is
>> libusb-compat which can be used as a wrapper over libusb1 to give
>> libusb0 like interfaces.
>>
>> Actually I was using libusb-1.0.0 API as mentioned here
>> http://libusb.sourceforge.net/api-1.0/group__lib.html
>> that also means that the patch I attached should have been using
>> libusb_init() instead.
>>
>> Here is updated patch
>>
>
> this still breaks with a system that has libusb-0.1 installed like the
> current Ubuntu 8.10 does. So either we switch to libusb-1.0 completely
> or we leave it as it is.
>

This looks like a job for autoconf?

2009-01-06 02:54:28

by Marcel Holtmann

[permalink] [raw]
Subject: Re: [patch] hid2hci check usb_init return value

Hi Khem,

> > > Working on a system which did not have a USB in it. I was getting a
> > > segfault with hid2hci. The problem tracked down was usb_init () was
> > > failing and we did not check for it and carried on with execution.
> > > As a result the next call to libusb failed because we passed in NULL for
> > > ctx because during usb_init () this should have been malloc'ed if all
> > > have gone correctly but in case of error usb_init () free's up the
> > > allocated memory and hence the segfault.
> > >
> > > We should check for return value of usb_init () call and exit
> > > immediately if it fails. Tested on 4.24 release.
> >
> > the problem is that usb_init() doesn't return any errors on the versions
> > that I have here:
> >
> > ii libusb-0.1-4 2:0.1.12-12 userspace USB programming library
> > ii libusb-dev 2:0.1.12-12 userspace USB programming library
> >
> > Applying your patch would break complication on multiple platforms.
> What you have is libusb-0.1 which has older API I suppose. There is
> libusb-compat which can be used as a wrapper over libusb1 to give
> libusb0 like interfaces.
>
> Actually I was using libusb-1.0.0 API as mentioned here
> http://libusb.sourceforge.net/api-1.0/group__lib.html
> that also means that the patch I attached should have been using
> libusb_init() instead.
>
> Here is updated patch

this still breaks with a system that has libusb-0.1 installed like the
current Ubuntu 8.10 does. So either we switch to libusb-1.0 completely
or we leave it as it is.

Regards

Marcel