Date: Thu, 27 Mar 2014 15:58:19 +0100 From: Mateusz Guzik <mjguzik@gmail.com> To: David Xu <davidxu@freebsd.org> Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org, Mateusz Guzik <mjg@FreeBSD.org> Subject: Re: svn commit: r263755 - head/sys/kern Message-ID: <20140327145819.GA4730@dft-labs.eu> In-Reply-To: <5333E581.1000100@freebsd.org> References: <201403252330.s2PNUaei052956@svn.freebsd.org> <5333D70D.7050306@freebsd.org> <20140327083730.GA22942@dft-labs.eu> <5333E581.1000100@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Mar 27, 2014 at 04:46:57PM +0800, David Xu wrote: > On 2014/03/27 16:37, Mateusz Guzik wrote: > >On Thu, Mar 27, 2014 at 03:45:17PM +0800, David Xu wrote: > >>I think the async process pointer can be cleared when a process exits > >>by registering an event handler. please see attached patch. > >> > > > >Sure, but I'm not very fond of this solution. > > > >This is a rather obscure bug you wont hit unless you explicitly try, > >and even then you need root privs by default. > > > OK, but I don't like the bug exists in kernel. It is not obscure for me, > I can run "shutdown now" command, and insert a device, and then the > kernel will write garbage data into freed memory space. > Not sure what you mean. devd does not use this feature, and even if it did async_proc is cleared on close, which happens while signal delivery is still legal. That said, you are not going to encounter this bug unless you code something up to specifically trigger it. fwiw, I think we could axe this feature if there was no way to fix it without introducing a check for every process. > >As such writing a callback function which will be executed for all exiting > >processes seems unjustified for me. > > > >Ideally we would get some mechanism which would allow to register > >callbacks for events related to given entity. Then it could be used to > >provide a "call this function when process p exits", amongst other things. > > > > Yes, but the callback itself is cheap enough and is not worth to be > per-entity entry. > There is other code in the kernel which would benefit from such functionality - dev/syscons/scmouse, dev/vt/vt_core.c, aio and possibly more. As such I think this is worth pursuing. -- Mateusz Guzik <mjguzik gmail.com>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20140327145819.GA4730>