linux-kernel - Re: [PATCH v2] vfs: avoid recopying file names in getname

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20150329151012.GB1477@fixme-laptop>
Date:	Sun, 29 Mar 2015 23:10:12 +0800
From:	Boqun Feng <boqun.feng@...il.com>
To:	Richard Weinberger <richard.weinberger@...il.com>
Cc:	linux-fsdevel <linux-fsdevel@...r.kernel.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Al Viro <viro@...iv.linux.org.uk>,
	Paul Moore <pmoore@...hat.com>
Subject: Re: [PATCH v2] vfs: avoid recopying file names in getname_flags

Hi Richard,

On Sun, Mar 29, 2015 at 12:13:29PM +0200, Richard Weinberger wrote:
> On Wed, Mar 25, 2015 at 7:45 PM, Boqun Feng <boqun.feng@...il.com> wrote:
> > In the current implementation of getname_flags, a file name in the
> > user-space will be recopied if it takes more space that
> > EMBEDDED_NAME_MAX, however, at this moment, EMBEDDED_NAME_MAX bytes of
> > the file name are already copied into kernel address space, the only
> > reason why the recopy is needed is that "kname", which points to the
> > string of the file name, needs to be relocated.
> >
> > And the recopy can be avoided if we change the memory layout of the
> > `names_cachep` allocation. By putting `struct filename` at the tail of
> > the allocation instead of the head, relocation of kname is avoided.
> > Once putting the struct at the tail, each byte in the user space will be
> > copied only one time, so the recopy is avoided.
> >
> > Of course, other functions aware of the layout of the `names_cachep`
> > allocation, i.e., getname_kernel and putname also need to be modified to
> > adapt to the new layout.
> >
> > Since we change the layout of `names_cachep` allocation, in order to
> > figure out whether kname and the struct are separate, we now need to
> > check whether the file name string starts at the address
> > EMBEDDED_NAME_MAX bytes before the address of the struct.  As a result,
> > `iname`, which points the end of `struct filename`, is not needed
> > anymore.
> >
> > Signed-off-by: Boqun Feng <boqun.feng@...il.com>
> > ---
> > v1 --> v2:
> > Rebase the patch onto the for-next branch of Al's vfs repo.
> >
> >
> >  fs/namei.c         | 45 ++++++++++++++++++++++++++++-----------------
> >  include/linux/fs.h |  1 -
> >  2 files changed, 28 insertions(+), 18 deletions(-)
> >
> > diff --git a/fs/namei.c b/fs/namei.c
> > index 7a11ec1..6d04029 100644
> > --- a/fs/namei.c
> > +++ b/fs/namei.c
> > @@ -119,7 +119,7 @@
> >   * PATH_MAX includes the nul terminator --RR.
> >   */
> >
> > -#define EMBEDDED_NAME_MAX      (PATH_MAX - offsetof(struct filename, iname))
> > +#define EMBEDDED_NAME_MAX      (PATH_MAX - sizeof(struct filename))
> 
> *confused*
>  EMBEDDED_NAME_MAX is currently PATH_MAX - sizeof(struct filename)
> 
> Did you mix original and new source while creating the patch?

This patch is based on branch 'for-next' in Al's tree, in that repo
commit b2ddab0e5324aec11620666eccc4f0ff64802131 ("kill struct filename.separate")
changes EMBEDDED_NAME_MAX to (PATH_MAX - offsetof(struct filename, iname))

I put the "based-on" information in the version changing log, seems it's
better to put it in the commit log to make it clearer for reviewers?

Thanks,
Boqun Feng

> 
> -- 
> Thanks,
> //richard

Content of type "application/pgp-signature" skipped