[Toybox] [PATCH] Use LOOP_CLR_FD instead of a constant.

[Rob Landley]

On 03/29/2015 12:18 PM, enh wrote:
> On Sun, Mar 29, 2015 at 9:27 AM, Rich Felker <dalias at libc.org> wrote:
>> I agree with chopping out linux/*.h includes especially since they
>> tend to randomly break by introducing conflicts with libc headers.
>> Here it seems your motive is more that you want to avoid making the
>> command gratuitously Linux-dependent. However I'm not sure if it's
>> safe to send an 'unknown' ioctl to a random block device you just
>> opened. Worst case, the number could mean something completely
>> different like "format the drive".

Do you really see that as likely?

I did check in Linux. This ioctl was allocated early, back when the high
bytes were a letter for the subsystem, and it's not used for anything
else in the tree. Nobody checks what kind of device it is when calling
losetup -d on a block device, so "random format the drive when you do
that" would be a problem there too. (Maybe not as big of one, but still
the same category.)

> yeah, that's what worried me, seeing this on the same day that Darwin
> support was mentioned... it seems safer to fail to compile than to
> send random ioctls.

I've been holding off on adding infrastructure until there _were_ ports,
because I avoid infrastructure in search of a user. But having
portability.h define TOYBOX_LINUX or TOYBOX_BSD isn't out of the question.

(More likely I'd just factor out a loopdev_detach() function and move it
to portability.c.)

>> As far as I know Linux only has a
>> small known number of 'overloaded' ioctl numbers (ones with different
>> meanings for different device types) and this isn't one of them, but I
>> don't know whether the number is used for different things on other
>> systems, so in the absence of that information it might be best to
>> skip making the ioctl on non-Linux targets.

I checked what linux is doing. I can't check what non-linux is doing
because I don't have a mac or ios development environment. I'm just
responding to external interest there. :)

(I note that ifconfig also cares what the ioctl numbers work out to,
because it's masking against them to determine which of the command
table branches to go down.)

Rob

 1427666458.0