Re: [PATCH v3 20/23] fs: Allow copy_mount_options() to access user-space in a single pass

From: Dave Martin <Dave.Martin@arm.com>
To: Catalin Marinas <catalin.marinas@arm.com>
Cc: linux-arch@vger.kernel.org,
	Richard Earnshaw <Richard.Earnshaw@arm.com>,
	Will Deacon <will@kernel.org>,
	Szabolcs Nagy <szabolcs.nagy@arm.com>,
	Andrey Konovalov <andreyknvl@google.com>,
	Kevin Brodsky <kevin.brodsky@arm.com>,
	linux-mm@kvack.org, Alexander Viro <viro@zeniv.linux.org.uk>,
	Vincenzo Frascino <vincenzo.frascino@arm.com>,
	Peter Collingbourne <pcc@google.com>,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH v3 20/23] fs: Allow copy_mount_options() to access user-space in a single pass
Date: Mon, 4 May 2020 17:40:58 +0100	[thread overview]
Message-ID: <20200504164058.GJ30377@arm.com> (raw)
In-Reply-To: <20200429135224.GB10651@gaia>

On Wed, Apr 29, 2020 at 02:52:25PM +0100, Catalin Marinas wrote:
> On Wed, Apr 29, 2020 at 11:26:51AM +0100, Dave P Martin wrote:
> > On Tue, Apr 21, 2020 at 03:26:00PM +0100, Catalin Marinas wrote:
> > > diff --git a/arch/arm64/include/asm/uaccess.h b/arch/arm64/include/asm/uaccess.h
> > > index 32fc8061aa76..566da441eba2 100644
> > > --- a/arch/arm64/include/asm/uaccess.h
> > > +++ b/arch/arm64/include/asm/uaccess.h
> > > @@ -416,6 +416,17 @@ extern unsigned long __must_check __arch_copy_in_user(void __user *to, const voi
> > >  #define INLINE_COPY_TO_USER
> > >  #define INLINE_COPY_FROM_USER
> > >  
> > > +static inline bool arch_has_exact_copy_from_user(unsigned long n)
> > > +{
> > > +	/*
> > > +	 * copy_from_user() aligns the source pointer if the size is greater
> > > +	 * than 15. Since all the loads are naturally aligned, they can only
> > > +	 * fail on the first byte.
> > > +	 */
> > > +	return n > 15;
> > > +}
> > > +#define arch_has_exact_copy_from_user
> > 
> > Did you mean:
> > 
> > #define arch_has_exact_copy_from_user arch_has_exact_copy_from_user
> 
> Yes (and I shouldn't write patches late in the day).
> 
> > Mind you, if this expands to 1 I'd have expected copy_mount_options()
> > not to compile, so I may be missing something.
> 
> I think arch_has_exact_copy_from_user() (with the braces) is looked up
> in the function namespace, so the macro isn't expanded. So arguably the
> patch is correct but pretty dodgy ;).
> 
> I scrapped this in my second attempt in reply to Kevin.

Problem solved!

> > > diff --git a/fs/namespace.c b/fs/namespace.c
> > > index a28e4db075ed..8febc50dfc5d 100644
> > > --- a/fs/namespace.c
> > > +++ b/fs/namespace.c
> > > @@ -3025,13 +3025,16 @@ void *copy_mount_options(const void __user * data)
> > 
> > [ Is this applying a band-aid to duct tape?
> > 
> > The fs presumably knows ahead of time whether it's expecting a string or
> > a fixed-size blob for data, so I'd hope we could just DTRT rather than
> > playing SEGV roulette here.
> > 
> > This might require more refactoring than makes sense for this series
> > though. ]
> 
> That's possible but it means moving the copy from sys_mount() to the
> specific places where it has additional information (the filesystems).
> I'm not even sure it's guaranteed to be strings. If it is, we could just
> replace all this with a strncpy_from_user().

Fair enough.  I'll add it to my wishlist...

Cheers
---Dave