Re: [PATCH 3/3] libmount/utils: Use binary search to compare pseudofs

[Dave Reisner <d@falconindy.com>]

On Sun, May 20, 2012 at 06:05:12PM +0200, Davidlohr Bueso wrote:
> On Sun, 2012-05-20 at 10:51 -0400, Dave Reisner wrote:
> > On Sun, May 20, 2012 at 04:06:00PM +0200, Davidlohr Bueso wrote:
> > > On Sat, 2012-05-19 at 20:39 -0400, Dave Reisner wrote:
> > > > While we're at it, expand this list of known pseudofs types.
> > > > 
> > > > Signed-off-by: Dave Reisner <dreisner@archlinux.org>
> > > > ---
> > > >  libmount/src/utils.c |   68 +++++++++++++++++++++++++++++++-------------------
> > > >  1 file changed, 43 insertions(+), 25 deletions(-)
> > > > 
> > > > diff --git a/libmount/src/utils.c b/libmount/src/utils.c
> > > > index b824edc..77557f2 100644
> > > > --- a/libmount/src/utils.c
> > > > +++ b/libmount/src/utils.c
> > > > @@ -72,6 +72,15 @@ int mnt_parse_offset(const char *str, size_t len, uintmax_t *res)
> > > >  	return rc;
> > > >  }
> > > >  
> > > > +/* used as a callback by bsearch in mnt_fstype_is_pseudofs() */
> > > > +static int fstype_cmp(const void *v1, const void *v2)
> > > > +{
> > > > +	const char *s1 = *(const char **)v1;
> > > > +	const char *s2 = *(const char **)v2;
> > > > +
> > > > +	return strcmp(s1, s2);
> > > > +}
> > > > +
> > > >  /* returns basename and keeps dirname in the @path, if @path is "/" (root)
> > > >   * then returns empty string */
> > > >  static char *stripoff_last_component(char *path)
> > > > @@ -214,31 +223,40 @@ char *mnt_unmangle(const char *str)
> > > >   */
> > > >  int mnt_fstype_is_pseudofs(const char *type)
> > > >  {
> > > > -	if (!type)
> > > > -		return 0;
> > > > -	if (strcmp(type, "none")  == 0 ||
> > > > -	    strcmp(type, "proc")  == 0 ||
> > > > -	    strcmp(type, "tmpfs") == 0 ||
> > > > -	    strcmp(type, "sysfs") == 0 ||
> > > > -	    strcmp(type, "autofs") == 0 ||
> > > > -	    strcmp(type, "devpts") == 0||
> > > > -	    strcmp(type, "cgroup") == 0 ||
> > > > -	    strcmp(type, "devtmpfs") == 0 ||
> > > > -	    strcmp(type, "devfs") == 0 ||
> > > > -	    strcmp(type, "dlmfs") == 0 ||
> > > > -	    strcmp(type, "cpuset") == 0 ||
> > > > -	    strcmp(type, "configfs") == 0 ||
> > > > -	    strcmp(type, "securityfs") == 0 ||
> > > > -	    strcmp(type, "hugetlbfs") == 0 ||
> > > > -	    strcmp(type, "rpc_pipefs") == 0 ||
> > > > -	    strcmp(type, "fusectl") == 0 ||
> > > > -	    strcmp(type, "mqueue") == 0 ||
> > > > -	    strcmp(type, "binfmt_misc") == 0 ||
> > > > -	    strcmp(type, "fuse.gvfs-fuse-daemon") == 0 ||
> > > > -	    strcmp(type, "debugfs") == 0 ||
> > > > -	    strcmp(type, "spufs") == 0)
> > > > -		return 1;
> > > > -	return 0;
> > > > +	static const char *pseudofs[] = {
> > > > +		"anon_inodefs",
> > > > +		"autofs",
> > > > +		"bdev",
> > > > +		"binfmt_misc",
> > > > +		"cgroup",
> > > > +		"configfs",
> > > > +		"cpuset",
> > > > +		"debugfs",
> > > > +		"devfs",
> > > > +		"devpts",
> > > > +		"devtmpfs",
> > > > +		"dlmfs",
> > > > +		"fuse.gvfs-fuse-daemon",
> > > > +		"fusectl",
> > > > +		"hugetlbfs",
> > > > +		"mqueue",
> > > > +		"nfsd",
> > > > +		"none",
> > > > +		"pipefs",
> > > > +		"proc",
> > > > +		"pstore",
> > > > +		"ramfs",
> > > > +		"rootfs",
> > > > +		"rpc_pipefs",
> > > > +		"securityfs",
> > > > +		"sockfs",
> > > > +		"spufs",
> > > > +		"sysfs",
> > > > +		"tmpfs"
> > > > +	};
> > > > +
> > > 
> > > What about applying sort (qsort(3)) before searching here? This would
> > > correct any manual string sorting that pseudofs[] imposes to the
> > > developer.
> > > 
> > 
> > Hmm, so the overhead from this seems to be negligible, but given that
> > this is running as part of a library, wouldn't we need a reentrant
> > version of qsort? qsort_r was added to glibc in 2.8, but I'm not
> > familiar with how portable this would be to other libc's.
> 
> I normally worry about reentrancy when dealing with interrupt handling,
> like signals. Also consider that pseudofs[] isn't global.
> 

qsort_r takes an extra param to the callback to hold an element while
its being swapped. The reentrancy of the function isn't guaranteed just
because it operates on a non-global var.

Regardless, I'm not really in favor of adding this. I think it's reasonable
to push a little responsibility on the developer and preemptively ensure
that this is sorted. Perhaps add a comment in the function would be
sufficient: "make sure this is sorted or Karel will come after you".

d

> > 
> > > > +	return !(bsearch(&type, pseudofs, ARRAY_SIZE(pseudofs),
> > > > +				sizeof(char*), fstype_cmp) == NULL);
> > > >  }
> > > >  
> > > >  /**
> > > 
> > > 
> > > --
> > > To unsubscribe from this list: send the line "unsubscribe util-linux" in
> > > the body of a message to majordomo@vger.kernel.org
> > > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > 
> 
>