[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Wed, 06 Oct 2021 14:57:55 +0200
From: Pablo Correa Gomez <ablocorrea@...mail.com>
To: Rich Felker <dalias@...c.org>
Cc: musl@...ts.openwall.com
Subject: Re: newlocale: Segmentation fault when locale input is NULL
Thank you very much to both for the detailed explanation. I really
appreciate you took the time to explain why musl behaviour should
remain the way it is. I will proceed and fix the bug in GNOME.
> In particular, here it seems to have found a bug -- what could the
> application have possibly meant by passing a null pointer there? Did
> it actually intend the behavior of ""? Or of "C"? Or if the intent
> was
> to have this mean "don't use a context-local locale", why pass the
> pointer to newlocale and process the error (which could include a
> number of other errors you'd certainly want to treat differently)
> rather than checking for null and taking a different code path before
> calling newlocale?
Very valid point. Definitely something worth to ask the maintainers.
Best and thank you again,
Pablo.
On Wed, 2021-10-06 at 08:29 -0400, Rich Felker wrote:
> On Wed, Oct 06, 2021 at 11:31:29AM +0200, Pablo Correa Gomez wrote:
> > Dear musl maintainers,
> >
> > While doing some work in GNOME control center for postmarketos, we
> > bumped into a segmentation fault which is also present in GNOME in
> > Alpine[1].
> >
> > After doing some degugging, I figured out that the reason is that,
> > through GNOME desktop[2], there is a call to newlocale, where they
> > end
> > up calling it with a NULL argument.
> >
> > newlocale(LC_CTYPE, NULL, (locale_t)0);
> >
> > In this case, "name" is passed to __get_locale in
> > src/locale/newlocale.c:27 and then dereferenced in
> > src/locale/locale_map.c:43, causing a segmentation fault.
> >
> > In the case of glibc, this is not an issue, as per the
> > documentation[3]
> > they consider it an error:
> >
> > EINVAL locale is NULL.
> >
> > Unfortunately, this is a difference in the implementation between
> > glibc
> > and musl, maybe due to the fact that the standard[4] in not clear
> > in
> > this point:
> >
> >
> > The newlocale() function may fail if:
> >
> > [EINVAL]
> > The locale argument is not a valid string pointer.
>
> This is specifically documented as a "may fail", not a "shall fail",
> i.e. it's not guaranteed to happen. It comes from POSIX, and is an
> instance of a weird pattern the committee tried to fix (but missed
> some places it seems) where they wrote "may fail"s for conditions
> that
> already have undefined behavior (here, use of an invalid pointer) in
> which case EINVAL would already be allowed as a side effect of the UB
> without any further specification. (The same thing created a lot of
> confusion in the past about use of pthread_t values past the end of
> their lifetime.)
>
> > My personal believe is that adding a NULL pointer check in musl is
> > very
> > simple and might help not only GNOME desktop, but maybe also other
> > projects in the future. This is the reason why I brought the issue
> > here
> > first instead of directly patching GNOME desktop. If you believe
> > that
> > musl behaviour should remain the way it is, please let me know and
> > I
> > will send MRs for upstream and Alpine's GNOME desktop. I am not
> > subscribed to the mailing list, so I would appreciate if I am CC'ed
> > in
> > any response.
>
> The musl behavior should remain the way it is. My text on the
> rationale actually made it into the glibc wiki some years back:
>
> https://sourceware.org/glibc/wiki/Style_and_Conventions#Invalid_pointers
>
> "The GNU C library considers it a QoI feature not to mask user
> bugs by detecting invalid pointers and returning EINVAL (unless
> the API is standardized and says it does that). If passing a bad
> pointer has undefined behavior, it is far more useful in the long
> run if it crashes quickly rather than diagnosing an error that is
> probably ignored by the flaky caller.
>
> If you're going to check for NULL pointer arguments where you
> have
> not entered into a contract to accept and interpret them, do so
> with an assert, not a conditional error return. This way the bugs
> in the caller will be immediately detected and can be fixed, and
> it makes it easy to disable the overhead in production builds.
> The
> assert can be valuable as code documentation. However, a segfault
> from dereferencing the NULL pointer is just as effective for
> debugging. If you return an error code to a caller which has
> already proven itself buggy, the most likely result is that the
> caller will ignore the error, and bad things will happen much
> later down the line when the original cause of the error has
> become difficult or impossible to track down. Why is it
> reasonable
> to assume the caller will ignore the error you return? Because
> the
> caller already ignored the error return of malloc or fopen or
> some
> other library-specific allocation function which returned NULL to
> indicate an error."
>
> In particular, here it seems to have found a bug -- what could the
> application have possibly meant by passing a null pointer there? Did
> it actually intend the behavior of ""? Or of "C"? Or if the intent
> was
> to have this mean "don't use a context-local locale", why pass the
> pointer to newlocale and process the error (which could include a
> number of other errors you'd certainly want to treat differently)
> rather than checking for null and taking a different code path before
> calling newlocale?
>
> Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
