Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 30 Aug 2004 11:21:47 -0700
From:      Julian Elischer <julian@elischer.org>
To:        Gleb Smirnoff <glebius@freebsd.org>
Cc:        FreeBSD current mailing list <current@freebsd.org>
Subject:   Re: mutex Giant not owned at /usr/src/sys/kern/vfs_vnops.c:120
Message-ID:  <4133703B.4050601@elischer.org>
In-Reply-To: <20040830144041.GA33772@cell.sick.ru>
References:  <20040830095729.GD30701@cell.sick.ru> <Pine.NEB.3.96L.1040830102627.18639D-100000@fledge.watson.org> <20040830144041.GA33772@cell.sick.ru>

next in thread | previous in thread | raw e-mail | index | archive | help
One option is to remove the functionality of making netgraph load kernel 
modules..
 I think this is outside the scope of netgraph really....


Gleb Smirnoff wrote:

>On Mon, Aug 30, 2004 at 10:35:05AM -0400, Robert Watson wrote:
>R> > B> > This causes Giant to be acquired in the event we enter the linker code
>R> > B> > (and hence VFS code) via netgraph ngc_send().  It should be safe in this
>R> > B> > context as we enter protocol send routines without mutexes held (i.e., why
>R> > B> > we're also able to do blocking memory allocation here.)
>R> > B> 
>R> > B> please commit.
>R> > 
>R> > I think Giant should be acquired in linker_load_module(), and this way
>R> > we will prevent this panic in other codepaths. For example in
>R> > vfs_mount.c, when vfs will be Giant-free. Have I missed something?
>R> 
>R> Well, one of the primary reasons the linker needs Giant here is its use of
>R> VFS.  I think for now I'd like to acquire Giant in netgraph so as to
>R> expose the use of Giant in that piece of the network stack in the calling
>R> code.  We might want to add GIANT_REQUIRED assertions in the linker code
>R> to make sure we trigger assertions even if the linker doesn't hit VFS to
>R> make sure potentially hitting Giant is caught.
>
>OK. Then commit it pls.
>
>  
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4133703B.4050601>