On Thu, Sep 21, 2017 at 03:04:05PM -0400, Alan Stern wrote: > On Thu, 21 Sep 2017, Andrey Konovalov wrote: > > > On Thu, Sep 21, 2017 at 6:10 PM, Greg Kroah-Hartman > > <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: > > > On Thu, Sep 21, 2017 at 05:39:05PM +0200, Andrey Konovalov wrote: > > >> Hi! > > >> > > >> I've got the following report while fuzzing the kernel with syzkaller. > > >> > > >> On commit ebb2c2437d8008d46796902ff390653822af6cc4 (Sep 18). > > >> > > >> The issue occurs when we iterate over interface altsettings, but I > > >> don't see the driver doing anything wrong. I might be missing > > >> something, or this might be an issue in USB core altsettings parsing. > > > > > > > > > Any chance you happen to have the descriptors that you were feeding into > > > the kernel at this crash? That might help in figuring out what "went > > > wrong". > > > > Here's the data that I feed into dummy_udc: > > > > 00 00 00 00 09 02 12 00 01 34 05 80 07 09 04 6e > > 09 00 08 06 62 00 12 01 05 00 cb f7 71 83 04 00 > > 05 00 ab f6 07 81 08 01 > > > > Also attaching a C reproducer (requires dummy_hcd and gadgetfs) and my .config. > > Why do your reproducers use an mmap'ed array for their data instead of > a plain old statically allocated array? > > Anyway, this turns out to be a genuine (and subtle!) bug in the uas > driver. The uas_find_uas_alt_setting() routine in uas-detect.h returns > a bAlternateSetting value, but then the uas_use_uas_driver() routine > uses this value as an index to the altsetting array -- which it isn't. > > Normally this doesn't cause any problems because the the entries in the > array have bAlternateSetting values 0, 1, etc., so the value is equal > to the index. But in your fuzzed case, that wasn't true. > > The patch below fixes this bug, by returning a pointer to the > alt-setting entry instead of either the value or the index. Pointers > are less subject to misinterpretation. Ugh, messy, nice catch and fix, I'll go queue it up now, thanks for resolving this. greg k-h
