linux-kernel - Re: [PATCH v2 1/1] efivarfs: fix statfs() on efivarfs

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] [day] [month] [year] [list]
Message-ID: <CAC_iWjL0ishwsLw+uqwLTMy3eNbaR6Tr5LZ9xUwqWG66amNOvg@mail.gmail.com>
Date:   Mon, 11 Sep 2023 11:14:09 +0300
From:   Ilias Apalodimas <ilias.apalodimas@...aro.org>
To:     Ard Biesheuvel <ardb@...nel.org>
Cc:     Heinrich Schuchardt <heinrich.schuchardt@...onical.com>,
        Anisse Astier <anisse@...ier.eu>, Jeremy Kerr <jk@...abs.org>,
        Anisse Astier <an.astier@...teo.com>,
        linux-efi@...r.kernel.org, linux-kernel@...r.kernel.org,
        Christian Brauner <brauner@...nel.org>
Subject: Re: [PATCH v2 1/1] efivarfs: fix statfs() on efivarfs

On Mon, 11 Sept 2023 at 11:06, Ard Biesheuvel <ardb@...nel.org> wrote:
>
> On Mon, 11 Sept 2023 at 10:04, Ilias Apalodimas
> <ilias.apalodimas@...aro.org> wrote:
> >
> > Hi Ard,
> >
> > On Mon, 11 Sept 2023 at 09:45, Ard Biesheuvel <ardb@...nel.org> wrote:
> > >
> > > On Sun, 10 Sept 2023 at 22:42, Heinrich Schuchardt
> > > <heinrich.schuchardt@...onical.com> wrote:
> > > >
> > > > On 9/10/23 20:53, Anisse Astier wrote:
> > > > > Hi Heinrich,
> > > > >
> > > > > On Sun, Sep 10, 2023 at 06:54:45AM +0200, Heinrich Schuchardt wrote:
> > > > >> Some firmware (notably U-Boot) provides GetVariable() and
> > > > >> GetNextVariableName() but not QueryVariableInfo().
> > > > >
> > > > >  From a quick search, it seems u-boot, does support QueryVariableInfo, is
> > > > > it on a given version ?
> > > > >
> > > > > https://elixir.bootlin.com/u-boot/v2023.07.02/source/lib/efi_loader/efi_variable.c#L391
> > > >
> > > > QueryVariableInfo() and SetVariable() are available before
> > > > ExitBootServices(), i.e. in Linux' EFI stub.
> > > >
> > > > ExitBootServices() results in calling efi_variables_boot_exit_notify()
> > > > which disables these services during the UEFI runtime.
> > > >
> > > > >
> > > > >>
> > > > >> With commit d86ff3333cb1 ("efivarfs: expose used and total size") the
> > > > >> statfs syscall was broken for such firmware.
> > > > >
> > > > > Could you be more specific ? What breaks, and what regressed ? I imagine
> > > > > it could be some scripts running df, but maybe you had something else in
> > > > > mind ?
> > > >
> > > > Some more details can be found in
> > > > https://bugs.launchpad.net/ubuntu/+source/linux-meta-riscv/+bug/2034705.
> > > >
> > > > Though EFI variables are exposed via GetVariable() and
> > > > GetNextVariableName() the efivar command refuses to display variables
> > > > when statfs() reports an error.
> > > >
> > > > >
> > > > >>
> > > > >> If QueryVariableInfo() does not exist or returns an error, just report the
> > > > >> file-system size as 0 as statfs_simple() previously did.
> > > > >
> > > > > I considered doing this [2] , but we settled on returning an error
> > > > > instead for clarity:
> > > > > https://lore.kernel.org/linux-efi/20230515-vorgaben-portrait-bb1b4255d31a@brauner/
> > > > >
> > > > > I still think it would be a good idea if necessary.
> > > >
> > > > We should never break user APIs.
> > > >
> > >
> > > Indeed.
> > >
> > > > >
> > > > > On the approach, I prefer what Ard proposed, to fall back to the old
> > > > > approach. I think the difference in block size could also be a good
> > > > > marker that something wrong is happening:
> > > > > https://lore.kernel.org/linux-efi/CAMj1kXEkNSoqG4zWfCZ8Ytte5b2SzwXggZp21Xt17Pszd-q0dg@mail.gmail.com/
> > > >
> > > > This will allow user code making assumptions based on block size:
> > > > If block size > 1, assume setting variables is possible.
> > > >
> > > > We should really avoid this.
> > > >
> > >
> > > I agree that having different block sizes depending on which code path
> > > is taken is not great. But that is the situation we are already in,
> > > given that older kernels will always report PAGE_SIZE. And actually,
> > > PAGE_SIZE does not make sense either - PAGE_SIZE could be larger than
> > > 4k on ARM for instance, so the efivarfs block size will be dependent
> > > on the page size of the kernel you happened to boot.
> > >
> > > So I think we should go with the below:
> > >
> > > --- a/fs/efivarfs/super.c
> > > +++ b/fs/efivarfs/super.c
> > > @@ -32,10 +32,16 @@ static int efivarfs_statfs(struct dentry *dentry,
> > > struct kstatfs *buf)
> > >         u64 storage_space, remaining_space, max_variable_size;
> > >         efi_status_t status;
> > >
> > > -       status = efivar_query_variable_info(attr, &storage_space,
> > > &remaining_space,
> > > -                                           &max_variable_size);
> > > -       if (status != EFI_SUCCESS)
> > > -               return efi_status_to_err(status);
> > > +       /* Some UEFI firmware does not implement QueryVariableInfo() */
> > > +       storage_space = remaining_space = 0;
> > > +       if (efi_rt_services_supported(EFI_RT_SUPPORTED_QUERY_VARIABLE_INFO)) {
> > > +               status = efivar_query_variable_info(attr, &storage_space,
> > > +                                                   &remaining_space,
> > > +                                                   &max_variable_size);
> > > +               if (status != EFI_SUCCESS && status != EFI_UNSUPPORTED)
> > > +                       pr_warn_ratelimited("query_variable_info()
> > > failed: 0x%lx\n",
> > > +                                           status);
> > > +       }
> >
> > I think this is better, but shouldn't we initialize the status
> > variable now? Or is there more code following that I am missing?
> >
>
> status is not referenced again after this.

Ah fair enough. I'd still initialize it for sanity, but with or without
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@...aro.org>