Date: Wed, 3 Dec 2008 14:41:37 +0200 From: Kostik Belousov <kostikbel@gmail.com> To: John Baldwin <jhb@freebsd.org> Cc: scottl@freebsd.org, current@freebsd.org Subject: Re: [PATCH] Make udf(4) MPSAFE and use shared lookups Message-ID: <20081203124137.GE3045@deviant.kiev.zoral.com.ua> In-Reply-To: <200812021828.36857.jhb@freebsd.org> References: <200811201627.58289.jhb@freebsd.org> <200811211452.02545.jhb@freebsd.org> <20081122115028.GB6408@deviant.kiev.zoral.com.ua> <200812021828.36857.jhb@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
--q/d9vTEvvdeKbPNw Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Tue, Dec 02, 2008 at 06:28:36PM -0500, John Baldwin wrote: > On Saturday 22 November 2008 06:50:28 am Kostik Belousov wrote: > > udf_vget() does insmntque() before vnode is fully initialized, allowing > > other threads to find the vnode on the mount list. This is typical for > > !MPSAFE fs, and it seems corresponding call was not marked XXX for udf. >=20 > It does the same as ufs. ufs only partially initializes the i-node (as m= uch=20 > as both cd9660 and udf do) and then exclusive locks the vnode before=20 > insmntque(). They then finish initializing the i-node (bread() the d-nod= e,=20 > for example) and finally drop the vnode lock. My bad, unjustified grumbling. >=20 > > udf_lookup for ISDOTDOT case unlocks dvp before vget'ing "..", allowing > > the same race on forced unmount as ufs (I will finally commit ufs patch > > today). The race happens for !MPSAFE code too, but it is easier to > > execute without Giant. >=20 > Every fs is going to need this workaround it seems. Would be nice if the= re=20 > was an easier way to avoid cut and pasting this code N times. Perhaps we= =20 > could make lookup() check VI_DOOMED instead? I had changed it do that at= one=20 > point, but then someone pointed me at the deadfs stuff and said that was= =20 > sufficient. The point of the patch is busying mp while parent vnode is locked, that guarantees that mp is not unmounted during whole DOTDOT traversing in vop_lookup(). The deadfs stuff works for lookup result vnode and is sufficient. The fragment that someone committed into UFS can be extracted into the vfs support routine. I doubt that it can be embedded into lookup(). The problem is that some filesystems do additional operations inside vop_lookup(). --q/d9vTEvvdeKbPNw Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.9 (FreeBSD) iEYEARECAAYFAkk2foEACgkQC3+MBN1Mb4hA4QCfbOvumkkzfUeFlQNtEddszDdq rMwAnjCmPda6vAtxmlGFrclRn/KYE4DB =ikoi -----END PGP SIGNATURE----- --q/d9vTEvvdeKbPNw--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20081203124137.GE3045>