Thoughts on TMPFS no longer being considered "highly experimental"

Peter Holm peter at holm.cc
Fri Jun 24 16:20:06 UTC 2011


On Fri, Jun 24, 2011 at 05:50:43PM +0300, Kostik Belousov wrote:
> On Fri, Jun 24, 2011 at 03:21:05PM +0200, Peter Holm wrote:
> > On Fri, Jun 24, 2011 at 02:06:27PM +0300, Kostik Belousov wrote:
> > > On Fri, Jun 24, 2011 at 12:30:16PM +0200, Peter Holm wrote:
> > > > On Thu, Jun 23, 2011 at 11:21:53PM +0300, Kostik Belousov wrote:
> > > > > On Thu, Jun 23, 2011 at 09:31:09AM -0700, David O'Brien wrote:
> > > > > > Does anyone object to this patch?
> > > > > > 
> > > > > > David Wolfskill and I have run TMPFS on a number of machines for two
> > > > > > years with no problems.
> > > > > > 
> > > > > > I may have missed something, but I'm not aware of any serious PRs on
> > > > > > TMPFS either.
> > > > > > 
> > > > > > 
> > > > > > Index: tmpfs_vfsops.c
> > > > > > ===================================================================
> > > > > > --- tmpfs_vfsops.c	(revision 221113)
> > > > > > +++ tmpfs_vfsops.c	(working copy)
> > > > > > @@ -155,9 +155,6 @@ tmpfs_mount(struct mount *mp)
> > > > > >  		return EOPNOTSUPP;
> > > > > >  	}
> > > > > >  
> > > > > > -	printf("WARNING: TMPFS is considered to be a highly experimental "
> > > > > > -	    "feature in FreeBSD.\n");
> > > > > > -
> > > > > >  	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
> > > > > >  	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
> > > > > >  	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
> > > > > 
> > > > > The things I am aware of:
> > > > > - there is a races on the lookup. They were papered over in r212305,
> > > > > but the bug was not really fixed, AFAIR.
> > > > > 
> > > > > - the tmpfs does double-buffering for the mapped vnodes. This is quite
> > > > > insulting for the memory-backed fs, isn't it ? I have a patch, but it is
> > > > > still under review.
> > > > > 
> > > > > - I believe Peter Holm has more test cases that fails with tmpfs. He
> > > > > would have more details. I somewhat remember some panic on execve(2) the
> > > > > binary located on tmpfs.
> > > > > 
> > > > 
> > > > I ran the TMPFS tests I have and so far I only spotted the mmap(2)
> > > > problem:
> > > > 
> > > > http://people.freebsd.org/~pho/stress/log/tmpfs/
> > > It would be indeed good if the issue was the only remaining problem.
> > 
> > Well, more testing is needed for sure.
> > 
> > > The deadlock in tmpfs6.txt is caused by doing copyin() while having
> > > a page busied. This should be fixed indirectly by the patch to
> > > avoid double-buffering, I uploaded the latest version at
> > > http://people.freebsd.org/~kib/misc/tmpfs.5.patch
> > > 
> > > > 
> > > > > Removing the warning will not make the issues coming away.
> > > > 
> > 
> > This doesn't compile:
> > 
> > ===> tmpfs (all)
> > cc -O2 -pipe -fno-strict-aliasing -Werror -D_KERNEL -DKLD_MODULE -nostdinc   -DHAVE_KERNEL_OPTION_HEADERS -include /usr/src/sys/i386/compile/PHO/opt_global.h -I. -I@ -I@/contrib/altq
> > -finline-limit=8000 --param inline-unit-growth=100 --param large-function-growth=1000 -fno-common -g -I/usr/src/sys/i386/compile/PHO  -mno-align-long-strings -mpreferred-stack-boundary=2 -mno-sse
> > -mno-mmx -msoft-float -ffreestanding -fstack-protector -std=iso9899:1999 -fstack-protector -Wall -Wredundant-decls -Wnested-externs -Wstrict-prototypes  -Wmissing-prototypes -Wpointer-arith -Winline
> > -Wcast-qual  -Wundef -Wno-pointer-sign -fformat-extensions  -Wmissing-include-dirs -fdiagnostics-show-option -c /usr/src/sys/modules/tmpfs/../../fs/tmpfs/tmpfs_subr.c
> > cc1: warnings being treated as errors
> > /usr/src/sys/modules/tmpfs/../../fs/tmpfs/tmpfs_subr.c: In function 'tmpfs_reg_resize':
> > /usr/src/sys/modules/tmpfs/../../fs/tmpfs/tmpfs_subr.c:911: warning: 'uobj' is used uninitialized in this function
> > *** Error code 1
> 
> Yes, the patch has rotten. Please try
> http://people.freebsd.org/~kib/misc/tmpfs.6.patch

Got a "panic: Not a vnode object" quite fast:

http://people.freebsd.org/~pho/stress/log/kostik441.txt

- Peter


More information about the freebsd-current mailing list