FW: OT: broken/useless ansi - console driver??
Fairlight
fairlite at fairlite.com
Tue Oct 25 21:39:06 PDT 2005
Another toppie for your convenience.
Yeah, after my last post (actually during, kinda) I figured about the only
thing it could affect would be testing of exit/disconnect type code. I
mean, it's possible. But it's also irrelevant, as I agree with you that,
even if it makes no technical difference how -one- thing is shut down in
-one- circumstance, the basic principle of violating a running process's
santity (as you put it) is just plain wrong.
Ironically enough, linux 2.2 (yes, that far back, and I also am confident
from memory that this goes back far earlier--likely back to 1.0 or probably
even 0.99 from memories of triggering it while upgrading things) gets it
right:
[cobalt] [~/bin] [12:02am]: cp ftp ftp.bak
[cobalt] [~/bin] [12:02am]: cp gmon ftp
cp: cannot create regular file `ftp': Text file busy
Hey, at least we know if the linux developers actually did accidentally
purloin something of SCO's through IBM, it wasn't the crappy VM layer,
right? :) :) :)
FWIW, JPR's related assertion that it was lifted from SVR4 may actually
be correct, based on -proper- testing on Solaris, as Ken revised mine.
Solaris gives me a bus error, although no core. The program is just as
dead in the end, however, which is wrong. SCO apparently isn't alone
in this faux pas, and I really don't know how many other vendors may be
affected. It was vaguely described as being from Sun's implementation of
SVR4 from 1989. Who knows how many other vendors, if any, used that code
as a basis for their VM layer? It would be interesting to try several
modern SVR4 based systems and see how many are hosed in this respect.
I agree with you though. There's no valid reason for it to be broken the
way it is--on any platform. I likely couldn't be paid to agree with one.
At least not for less than a positive integer followed by six zeroes. :)
Yeah, I'm a stubborn SOB on principles, as you (and probably everyone)
know(s). Happily, my wife is the one person from whom I don't catch hell
for it. I count my blessings there.
Anyway, I think we can sum up and call it a day: SCO 6 (and Solaris 5.8)
are currently broken in this area. No arguments here.
I suppose I should report this to Sun now that I know about it, eh?
Actually, there is one thing I'm curious about. Ken and Jay would both
likely know more than I would about the feasibility of this line of
conjecture. Remember when I copied over the same binary on Solaris and
it didn't break? Would it be possible to inject a modified binary the
same size, but slightly modified so as to actually capture sensitive data
or the like via the modifications? Or is it doomed to fail and crash if
-any- byte in the file is different from the original image? I mean, just
crashing something is one thing. If it became a security issue, that's
entirely another.
mark->
More information about the Filepro-list
mailing list