On Tue, Oct 22, 2019 at 6:30 AM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> On Thu, Oct 17, 2019 at 04:33:09PM +0200, Arnd Bergmann wrote:
>
> > However, looking at this again after your comment I found a rather
> > more serious bug in my new RTC_IRQP_SET handling: Any 64-bit
> > machine can now bypass the permission check for RTC_IRQP_SET by
> > calling RTC_IRQP_SET32 instead.
>
> You've lost the check on RTC_EPOCH_SET as well.
Right, originally my plan was to keep the epoch handling local to
rtc-vr41xx.c as explained in the patch description. The driver is
specific to a particular very obsolete MIPS machine that was
apparently only ever used with 32-bit kernels.
I guess it can't hurt to treat it the same as RTC_IRQP_SET32
if you prefer. Folding in this change now and adapting the
changelog text:
--- a/drivers/rtc/dev.c
+++ b/drivers/rtc/dev.c
@@ -402,6 +402,7 @@ static long rtc_dev_ioctl(struct file *file,
#ifdef CONFIG_COMPAT
#define RTC_IRQP_SET32 _IOW('p', 0x0c, __u32)
#define RTC_IRQP_READ32 _IOR('p', 0x0b, __u32)
+#define RTC_EPOCH_SET32 _IOW('p', 0x0e, __u32)
static long rtc_dev_compat_ioctl(struct file *file,
unsigned int cmd, unsigned long arg)
@@ -416,6 +417,10 @@ static long rtc_dev_compat_ioctl(struct file *file,
case RTC_IRQP_SET32:
/* arg is a plain integer, not pointer */
return rtc_dev_ioctl(file, RTC_IRQP_SET, arg);
+
+ case RTC_EPOCH_SET32:
+ /* arg is a plain integer, not pointer */
+ return rtc_dev_ioctl(file, RTC_EPOCH_SET, arg);
}
return rtc_dev_ioctl(file, cmd, (unsigned long)uarg);
diff --git a/drivers/rtc/rtc-vr41xx.c b/drivers/rtc/rtc-vr41xx.c
index 79f27de545af..c3671043ace7 100644
--- a/drivers/rtc/rtc-vr41xx.c
+++ b/drivers/rtc/rtc-vr41xx.c
@@ -69,7 +69,6 @@ static void __iomem *rtc2_base;
/* 32-bit compat for ioctls that nobody else uses */
#define RTC_EPOCH_READ32 _IOR('p', 0x0d, __u32)
-#define RTC_EPOCH_SET32 _IOW('p', 0x0e, __u32)
static unsigned long epoch = 1970; /* Jan 1 1970 00:00:00 */
@@ -187,7 +186,6 @@ static int vr41xx_rtc_ioctl(struct device *dev,
unsigned int cmd, unsigned long
#ifdef CONFIG_64BIT
case RTC_EPOCH_READ32:
return put_user(epoch, (unsigned int __user *)arg);
- case RTC_EPOCH_SET32:
#endif
case RTC_EPOCH_SET:
/* Doesn't support before 1900 */
> Another potential issue is drivers/input/misc/hp_sdc_rtc.c,
> provided that the hardware in question might possibly exist
> on hppa64 boxen - CONFIG_GSC defaults to y and it's not
> 32bit-only, so that thing is at least selectable on 64bit
> kernels.
I decided long ago not to care: that code has never compiled after
it was originally merged into the kernel in 2005:
static int hp_sdc_rtc_ioctl(struct inode *inode, struct file *file,
unsigned int cmd, unsigned long arg)
{
#if 1
return -EINVAL;
#else
...
RTC_IRQP_SET, RTC_EPOCH_SET, ...
...
#endif
}
I don't see any chance that this code is revived. If anyone wanted to
make it work, the right approach would be to use the rtc framework
and rewrite the code first.
I could send a patch to remove the dead code though if that helps.
Arnd
