On Wed, 11 Feb 2015, Dave Mielke wrote: > [quoted lines by Alan Stern on 2015/02/11 at 15:22 -0500] > > >Well, it's a mystery. There are exactly two places in the usbfs code > >where a signal is sent: async_completed() and usbdev_remove(). As you > >can tell from the usbfs_snoop output, async_completed() doesn't get > >called until the URB completes. (And usbdev_remove() gets called only > >when the device is disconnected > > I just checked the snoop log again and can verify that the word remove is > nowhere in it. > > >besides which, the values you got for si_errno and si_addr were different from > >what usbdev_remove() uses.) > > The struct signalfd_siginfo has all zeros (NULL) for si_addr. Only these fields > have something in them: si_signo=SIGRTMIN (correct), si_errno=-ENOENT, > si_code=SI_ASYNCIO, si_pid, si_int, si_ptr. I'm using 64-bits so si_ptr is > bigger than si_int, but the low-order four bytes of si_ptr are the same as > what's in si_int. I suspect both are left over junk from somewhere else. > > As an aside: Perhaps usbfs could zero out the whole struct first so that we can > be sure of what it considered to be meaningful. Go ahead and modify the source code to zero out sinfo in async_completed(). That's a good security measure in any case, because the values do get passed back to the user. On the other hand, async_completed() always calls snoop(), so if you didn't see the "urb complete" line in the snoop output then the signal can't be coming from async_completed(). If you think somehow async_completed() is getting called at the wrong time, add a dump_stack() call inside the first "if (signr)" block. That will show how it got called. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
