[PATCH 1/1] hwclock: Verify RTC file descriptor; use reentrant functions

[Denys Vlasenko]

On Fri, Apr 25, 2014 at 7:48 PM, Bryan Evenson <bevenson at melinkcorp.com> wrote:
> If the RTC is in use by another process or if the dev interface
> is not properly established, calling rtc_xopen will fail.

I don't see how is that possible:

int FAST_FUNC rtc_xopen(const char **default_rtc, int flags)
{
        int rtc;

        if (!*default_rtc) {
                *default_rtc = "/dev/rtc";
                rtc = open(*default_rtc, flags);
                if (rtc >= 0)
                        return rtc;
                *default_rtc = "/dev/rtc0";
                rtc = open(*default_rtc, flags);
                if (rtc >= 0)
                        return rtc;
                *default_rtc = "/dev/misc/rtc";
        }

        return xopen(*default_rtc, flags);
}

If all open()'s fail, we end up executing "return xopen(...)".
And xopen() never returns failure - it exits
with error message instead.


> Similarly, the returned pointer of ctime and localtime was
> being used without checking if the pointer was valid. One of
> these calls was causing a call to hwclock to enter an
> uninterruptable sleep when the invalid hwclock access occurred.

Did you investigate which value of time_t caused that?

> The problem can be reproduced by running two instances of the
> following shell script:
>
>  #!/bin/sh
>  while [ 1 ]; do
>    /sbin/hwclock -w -u
>    sleep 1;
>  done

Works for me with no problems with current bbox git tree.
Modifying it to run hwclock much more often:

#!/bin/sh
i=0
while true; do
        ./busybox hwclock -w -u
        echo $i $?
        : $((i++))
done

and running two instances, I get this:

...
...
11461 0
hwclock: can't open '/dev/misc/rtc': No such file or directory
11462 1
11463 0
hwclock: can't open '/dev/misc/rtc': No such file or directory
11464 1
hwclock: can't open '/dev/misc/rtc': No such file or directory
11465 1
11466 0
...
...

Looks like it doesn't hang.
It's a question whether it should be failing or retrying,
but I don't see it hanging.


> This patch verifies the RTC file descriptor is valid before it
> is used, and it switches all ctime and localtime calls to
> ctime_r localtime_r respectively.  Before applying the patch,
> running the previously described test resulted in one instance
> of the script to hang with hwclock in an uninterruptable sleep
> within seconds of starting both scripts.  After the patch, ran
> two instances of the scripts for ten minutes without permanently
> blocking access to the RTC.

Let's focus first on reproducing the problem.
Which version of busybox do you use, what .config file,
on what kernel?