On Wed, Feb 29, 2012 at 09:36:02AM -0500, John Baldwin wrote: > On Wednesday, February 29, 2012 8:25:07 am Konstantin Belousov wrote: > > On Wed, Feb 29, 2012 at 02:37:25PM +0200, Mikolaj Golub wrote: > > > > > > On Wed, 29 Feb 2012 12:03:00 +0000 Robert N. M. Watson wrote: > > > > > > RNMW> I think the monitoring aspect of the patch is fine. > > > > > > RNMW> The bit I was worried about was external umask changes. This can > > > cause > > > RNMW> race conditions for applications that manage their umask -- for > > > RNMW> example, bsdtar, if I recall correctly. It's one thing to use a > > > RNMW> debugger to force an application to change its umask -- the > > > developer > > > RNMW> needs to know they are changing application behaviour. But > > > exposing a > > > RNMW> feature that can lead to correct applications but incorrect > > > results is > > > RNMW> a risky thing to do, hence my objection. > > > > > > RNMW> I think given the other objections, it would be wise to remove > > > write > > > RNMW> access to process umasks, but retain read access for procstat > > > (which is > > > RNMW> quite useful, I agree). > > > > > > I still don't see why having a sysctl RW is worse than asking users to run > > > something like in the attach when they need to change umask for another > > > process, but ok, if people don't like RW I will remove it. > > > > > What is done is attach is much worse then the sysctl, just because > > debugger attach often causes spurious EINTR, indeed seriously disrupting > > applications, as opposed to some uncertain damage that could be done in > > theory. > > kgdb doesn't though, and presumably for umask you would change it via kgdb, so > from the running process' perspective it would look the same as changing it > via > sysctl.
Right, but an idea of the change was to allow to do this for somebody who does not know how to perform it in kgdb. Not to mention that kgdb -w is risky, e.g. because filedesc might have changed under kgdb, so you would write over freed memory.
pgpGLKsCBYRqg.pgp
Description: PGP signature
