Date: Sat, 24 Oct 2020 20:24:52 +0300 From: Konstantin Belousov <kostikbel@gmail.com> To: Mateusz Guzik <mjg@freebsd.org> Cc: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r366997 - head/sys/kern Message-ID: <20201024172452.GD2643@kib.kiev.ua> In-Reply-To: <202010241330.09ODUb6x089521@repo.freebsd.org> References: <202010241330.09ODUb6x089521@repo.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Oct 24, 2020 at 01:30:37PM +0000, Mateusz Guzik wrote: > Author: mjg > Date: Sat Oct 24 13:30:37 2020 > New Revision: 366997 > URL: https://svnweb.freebsd.org/changeset/base/366997 > > Log: > vfs: fix a race where reclaim vholds freed vnodes A description of the race in the commit message would be respectful to other readers of the code, so that we do not need to reverse-eng the change to understand what and why was fixed. > > Reported by: pho > Tested by: pho (previous version) > Fixes: r366974 ("vfs: stop taking the interlock in vnode reclaim") > > Modified: > head/sys/kern/vfs_subr.c > > Modified: head/sys/kern/vfs_subr.c > ============================================================================== > --- head/sys/kern/vfs_subr.c Sat Oct 24 13:16:10 2020 (r366996) > +++ head/sys/kern/vfs_subr.c Sat Oct 24 13:30:37 2020 (r366997) > @@ -109,6 +109,7 @@ static void syncer_shutdown(void *arg, int howto); > static int vtryrecycle(struct vnode *vp); > static void v_init_counters(struct vnode *); > static void vgonel(struct vnode *); > +static bool vhold_recycle(struct vnode *); > static void vfs_knllock(void *arg); > static void vfs_knlunlock(void *arg); > static void vfs_knl_assert_locked(void *arg); > @@ -1126,7 +1127,8 @@ restart: > goto next_iter; > } > > - vhold(vp); > + if (!vhold_recycle(vp)) > + goto next_iter; > TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); > TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); > mtx_unlock(&vnode_list_mtx); > @@ -1231,7 +1233,8 @@ restart: > if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { > continue; > } > - vhold(vp); > + if (!vhold_recycle(vp)) > + continue; > count--; > mtx_unlock(&vnode_list_mtx); > vtryrecycle(vp); > @@ -3248,13 +3251,11 @@ vholdnz(struct vnode *vp) > * However, while this is more performant, it hinders debugging by eliminating > * the previously mentioned invariant. > */ > -bool > -vhold_smr(struct vnode *vp) > +static bool __always_inline > +_vhold_cond(struct vnode *vp) > { > int count; > > - VFS_SMR_ASSERT_ENTERED(); > - > count = atomic_load_int(&vp->v_holdcnt); > for (;;) { > if (count & VHOLD_NO_SMR) { > @@ -3270,6 +3271,28 @@ vhold_smr(struct vnode *vp) > return (true); > } > } > +} > + > +bool > +vhold_smr(struct vnode *vp) > +{ > + > + VFS_SMR_ASSERT_ENTERED(); > + return (_vhold_cond(vp)); > +} > + > +/* > + * Special case for vnode recycling. > + * > + * Vnodes are present on the global list until UMA takes them out. > + * Attempts to recycle only need the relevant lock and have no use for SMR. > + */ > +static bool > +vhold_recycle(struct vnode *vp) > +{ > + > + mtx_assert(&vnode_list_mtx, MA_OWNED); > + return (_vhold_cond(vp)); > } > > static void __noinline
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20201024172452.GD2643>