usb panic

Doing "kldunload ums; kldload ums" while a mouse was connected (to
trigger a devd event without having to physically disconnect and
reconnect the mouse) triggered the following panic:

kernel: type 12 trap, code=0
Stopped at      strncpy+0x14:   movb    0(%edx),%al
db> where
strncpy(c66a6524,0,10,c1745504,c66a6400) at strncpy+0x14
usbd_fill_deviceinfo(c5f92900,c66a6400,1,0,c05b707e) at usbd_fill_deviceinfo+0x121
usbioctl(c05ffe20,c1745504,c66a6400,1,c64aa690) at usbioctl+0x223
spec_ioctl(ec00ab88,ec00ac34,c04f4c0f,ec00ab88,c05f7bc0) at spec_ioctl+0xf2
spec_vnoperate(ec00ab88) at spec_vnoperate+0x13
vn_ioctl(c63f3aa0,c1745504,c66a6400,c66fb080,c64aa690) at vn_ioctl+0x17f
ioctl(c64aa690,ec00ad14,3,1,282) at ioctl+0x37c
syscall(2f,2f,2f,6,0) at syscall+0x22b
Xint0x80_syscall() at Xint0x80_syscall+0x1d
--- syscall (54, FreeBSD ELF32, ioctl), eip = 0x880b7a17, esp = 0xbfbfe2ac, ebp = 0xbfbfe458 ---

as usual, dumps are broken, but the code at least looks like this:

(gdb) l *(usbd_fill_deviceinfo+0x121)
0x33d5 is in usbd_fill_deviceinfo (/usr/src/sys/dev/usb/usb_subr.c:1282).
1277            if (dev->subdevs != NULL) {
1278                    for (i = 0; dev->subdevs[i] &&
1279                                 i < USB_MAX_DEVNAMES; i++) {
1280                            strncpy(di->udi_devnames[i], USBDEVPTRNAME(dev->subdevs[i]),
1281                                    USB_MAX_DEVNAMELEN);
1282                            di->udi_devnames[i][USB_MAX_DEVNAMELEN-1] = '\0';
1283                    }
1284            } else {
1285                    i = 0;
1286            }

so dev->subdevs[i] is not NULL, but it does not have a name since
USBDEVPTRNAME(dev->subdevs[i]) is NULL.  Looks like better locking and
invariants are required; it shouldn't be possible (IMHO) for that code
to stumble across an incomplete bdev.

(BTW, we should use strlcpy() here rather than strncpy())

DES
-- 
Dag-Erling Smørgrav - des_at_des.no