OpenSolaris

Flag Day: IOCPARM_MASK changes

Date: Wed, 18 Jun 2008 07:22:58 -0700
From: "Garrett D'Amore" <Garrett.Damore at sun dot com>
To: onnv-gate at onnv dot eng dot sun dot com, on-all at sun dot com
Subject: Flag Day: IOCPARM_MASK changes

(Belated notice, as we weren't aware of the issues surrounding this
earlier.)

Build 93 contains my putback for 6711665 Widen IOCPARM_MASK to 0x1fff,
also covered under PSARC 2008/348.

It has come to my attention that there may be some consumers of the old
values which erroneously passed an argument to large to be encoded
properly in forming ioctl values under the old value with _IOR, _IOW, or
_IORW (see sys/ioccom.h).  The two cases of this known are the tun(7m)
module (SIOCGTUNPARM, SIOCSTUNPARM) and kqemu.

As a result of this change, it will be necessary to update ifconfig at
the same time as updating to build 93 or newer, particularly if tunnels
are in use (such as with IPsec or Punchin.)

I believe there are no other consumers of these ioctls than ifconfig,
and as they are planned to be removed soon as part of Clearview, I am
not planning on introducing any fixes for other consumers of the tun(7M)
ioctls.  If anyone has any evidence to the contrary, please contact me
asap, and I will endeavor to correct the situation.

If using kqemu, it will be necessary to recompile both the userland and
kernel bits at the same time.

Note that correct use of the impacted macros (i.e. when the structure to
be encoded was smaller than the maximum value of 255 bytes) was not
affected, and I am unaware of any instances other than kqemu or tun
where this is an issue.  Note also that the macros in sys/ioccom.h,
while inherited from BSD, have never been documented as part of the DDI.

    ~-- Garrett