Date: Wed, 6 Aug 2008 16:39:50 +0300 From: Kostik Belousov <kostikbel@gmail.com> To: Rick Macklem <rmacklem@uoguelph.ca> Cc: freebsd-fs@freebsd.org Subject: Re: doing vfs_hash_get when vnode locked Message-ID: <20080806133950.GN97161@deviant.kiev.zoral.com.ua> In-Reply-To: <Pine.GSO.4.63.0808051638410.29561@muncher.cs.uoguelph.ca> References: <Pine.GSO.4.63.0808041657200.3482@muncher.cs.uoguelph.ca> <20080805083229.GB97161@deviant.kiev.zoral.com.ua> <Pine.GSO.4.63.0808051052350.27663@muncher.cs.uoguelph.ca> <20080805153221.GG97161@deviant.kiev.zoral.com.ua> <Pine.GSO.4.63.0808051242110.23305@muncher.cs.uoguelph.ca> <20080805165114.GH97161@deviant.kiev.zoral.com.ua> <Pine.GSO.4.63.0808051342290.2620@muncher.cs.uoguelph.ca> <20080805194341.GI97161@deviant.kiev.zoral.com.ua> <Pine.GSO.4.63.0808051638410.29561@muncher.cs.uoguelph.ca>
next in thread | previous in thread | raw e-mail | index | archive | help
--IiH5iyEqjK6f5lUg Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Tue, Aug 05, 2008 at 04:58:30PM -0400, Rick Macklem wrote: >=20 >=20 > On Tue, 5 Aug 2008, Kostik Belousov wrote: >=20 > [stuff snipped] > >>Ok, I just spent a few minutes snooping around in vfs_subr.c and I think > >>I see the problem. vget() has called vholdl() and then > >>v_upgrade_usecount(), which has incremented the usecount and taken the > >>vnode off the free list. This appears to prevent vgonel() from being > >>called on it for most cases, but there is still the case in vflush() > >>where the FORCECLOSE flag is set. > >Yes, exactly. > > > [more stuff snipped] > > > >But, what guarantees that the vnode would not be reclaimed before/under > >your vref() it ? For instance, what if the vnode is locked due to reclaim > >being in progress ? > > > So long as I never do a vflush() with FORCECLOSE, I can't see anywhere=20 > that will vgonel() it once I have gotten it via vget(). (v_usecount > incremented and not on the vnode freelist) >=20 > The way I just coded it is: > - the function that does the vfs_hash_get() without LK_EXCLUSIVE just > fails if MNTK_UNMOUNTF is set. > - my nfs_close just returns when MNTK_UNMOUNTF is set. > - my nfs_unmount() doesn't set FORCECLOSE on the vflush() but instead > sleeps and retries a bunch of times if vflush() fails for MNT_FORCE. > - my nfs_unmount() and other code (mostly based on the vanilla FreeBSD > client makes requests all fail with EINTR when MNTK_UNMOUNTF is set). You still has the race where the MNTK_UNMOUNTF is set after you check returned false, isn't it ? BTW, is your fs marked as mpsafe ? >=20 > I think this should work for a forced unmount, since once requests all > fail and the recovery also fails, I think vflush() will work without > the FORCECLOSE flag. >=20 > As far as I can see, since I'm not vflush()'ng with FORCECLOSE, then > nothing will vgonel() the vnode until it has been vrele()'d. (If there > is a case other than vflush() with FORCECLOSE that will vgone() it when > it is not on the freelist and has a v_usecount > 0, then I'll need to > handle that as well, but I can't see one.) Yes, ATM it should be safe, since only vflush() does reclamation for the vnodes with usecount > 0. On the other hand, I believe our VFS never makes a guarantee that this is the only location of the call. --IiH5iyEqjK6f5lUg Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.9 (FreeBSD) iEYEARECAAYFAkiZqaYACgkQC3+MBN1Mb4hdHACdFRKyFhrhmRqHxBWYnk3Ka2id s7MAoNsuFDa+MYMXykCQ8APpEDsrX4A/ =FMcN -----END PGP SIGNATURE----- --IiH5iyEqjK6f5lUg--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20080806133950.GN97161>