Date: Sun, 16 Mar 2008 14:20:55 +1100 (EST) From: Bruce Evans <brde@optusnet.com.au> To: Kostik Belousov <kostikbel@gmail.com> Cc: Ed Schouten <ed@80386.nl>, FreeBSD Arch <arch@freebsd.org> Subject: Re: vgone() calling VOP_CLOSE() -> blocked threads? Message-ID: <20080316133138.J41270@delplex.bde.org> In-Reply-To: <20080315194809.GN10374@deviant.kiev.zoral.com.ua> References: <20080315124008.GF80576@hoeg.nl> <20080316015903.N39516@delplex.bde.org> <20080315194809.GN10374@deviant.kiev.zoral.com.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 15 Mar 2008, Kostik Belousov wrote: > On Sun, Mar 16, 2008 at 03:55:18AM +1100, Bruce Evans wrote: >> Other problems near here: >> - neither vfs nor drivers currently know how many threads are in a >> driver. vfs uses vp->v_rdev->si_usecount, but this doesn't quite work > This is provided by si_threadcount. > See the dev(vn)_refthread and it usage in the devfs vnops and fops. So why doesn't reovoke() use it? :-). All uses of si_usecount, which normally happen via vcount() and count_dev(), are suspect, especially the latter. vcount() is only used in revoke(), in svr4_fcntl.c to handle another revoke(), and for FreeBSD < 6 in reiserfs for an old multiple-mount check. count_dev() is only used in ata-tape.c (to decide in the same broken way as vfs if a close is the last one -- this driver uses D_TRACKCLOSE to get d_close() called on all closes. This gives it the burden of deciding whether the close is the last one, and it can't do this any better than vfs. D_TRACKCLOSE is used in a few other drivers which don't call count_dev()), in devfs_close() (to decide whether to release the controlling terminal and to decide when to call d_close()). Hmm, it seems to be not vfs but only devfs which handles last-close specially. devfs is closer to devices, so it should know how to use si_threadcount here. Hopefully si_threadcount counts threads sleeping in open or close, although si_usecount doesn't. d_close (or something) should be called to wake up these threads even if si_usecount is 0. Drivers which support sleeping in open or close must support d_close (or something) being called to forcibly end such sleeps. revoke() should forcibly end such sleeps, so it needs to check si_threadcount too. si_usecount in its current form might end up being unused, so si_threadcount could be renamed back to it. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20080316133138.J41270>