Date: Mon, 19 Jul 2004 11:23:51 +0400 From: Gleb Smirnoff <glebius@freebsd.org> To: Julian Elischer <julian@elischer.org> Cc: Robert Watson <rwatson@freebsd.org> Subject: Re: linker_load_module(NULL, "modname", ...) from thread with no user process Message-ID: <20040719072351.GC45598@cell.sick.ru> In-Reply-To: <40FB533D.2080208@elischer.org> References: <Pine.NEB.3.96L.1040719003147.37108w-100000@fledge.watson.org> <40FB533D.2080208@elischer.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, Jul 18, 2004 at 09:51:09PM -0700, Julian Elischer wrote: J> >> there is problem when linker_load_module() is called from a kernel J> >>thread with no associated user process, and it asks to load module by J> >>name, not by filename. With such parameters it requires looking through J> >>device.hints file. And vn_open() assumes that J> >>ndp->ni_cnd->cn_thread->td_proc is valid. J> >> J> >>Any ideas how to solve this? J> >> J> >> J> > J> >Generally speaking, attempting to perform file I/O from an interrupt J> >thread or software interrupt is a really bad idea. There are a number of J> >reasons this is the case, not least that lookups and file operations occur J> >in the context of a process with a root directory, current working J> >directory, etc, and that a network swi or ithread doesn't have said J> >context (and may execute before that's available). Also, stalling the J> >netisr or an ithread on disk I/O seems to be a bad idea as well, not to J> >mention the NFS root file system case. So the question would seem to be J> >"Can we avoid it entirely?". I'm not quite sure what the answer here is, J> >but most similar cases I know of involve an asynchronous upcall message to J> >user space to load the module, or it being pushed from user space to J> >kernel without an upcall. Vis., devd loading a module in response to a J> >device event, etc. Adopting something more like that would help to avoid J> >this situation. I've seen similar reports a couple of times in the past, J> >and each time it worries me :-). J> > J> > J> Probably the onl thing to do is to refuse to try an load the modules if J> you are not running J> in the context of a process.. J> the question is: J> J> "What were you doing when this happenned?" Run mpd as PPPoE server. Ensure that ng_tee is not loaded, yet. Connect to this server. I have hit this panic with ngctl also, but can't remember hot to reproduce it. What is the correct way to determine whether we are in interrupt thread or not? -- Totus tuus, Glebius. GLEBIUS-RIPN GLEB-RIPE
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040719072351.GC45598>