I am guessing that sanitizers are not supported with static linking.
The AddressSanitizer document says so explicitly but the others do not.
Yet the interceptors use dlsym to set up certain system calls, so by
definition they can't support static linking.
Given that sanitizers won't work with static linking, why are static
libraries created for them?
Should there be an error check in the interceptor code for dlsym
returning nullptr? I just spent a bunch of time tracking down a problem
that turned out to be dlsym returning nullptr due to static linking.
It seems like either users should not be able to link sanitizer runtime
when linking statically or there should be a runtime check that dlsym
doesn't return nullptr. Ideally both.
-David
Saying that sanitizers "do not support static linking" seems strange because we strongly advise users to link the sanitizer libraries themselves statically. This gives the best user experience, because it puts the interceptors into the executable, so they always come first in dynamic symbol search order. What I think you are saying is that the sanitizers do not support linking statically with libc. This is probably true. It would be difficult to modify the interceptors to work when libc is linked statically, so it's probably not supported. Hopefully that explains why the sanitizer libraries are static, but they don't exactly support linking with -static. On Thu, Aug 23, 2018 at 10:04 AM David Greene via llvm-dev < llvm-dev at lists.llvm.org> wrote:> I am guessing that sanitizers are not supported with static linking. > The AddressSanitizer document says so explicitly but the others do not. > Yet the interceptors use dlsym to set up certain system calls, so by > definition they can't support static linking. > > Given that sanitizers won't work with static linking, why are static > libraries created for them? > > Should there be an error check in the interceptor code for dlsym > returning nullptr? I just spent a bunch of time tracking down a problem > that turned out to be dlsym returning nullptr due to static linking. > > It seems like either users should not be able to link sanitizer runtime > when linking statically or there should be a runtime check that dlsym > doesn't return nullptr. Ideally both. > > -David > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20180823/80aaa723/attachment.html>
Ok, that makes sense. Should there be code to test the result of dlsym
and report an error to the user? Faulting simply because one uses
--static during the link stage isn't very friendly. It would be nice to
get an error with a helpful message about linking libc dynamically.
Would there be interest in a patch to implement this?
-David
Reid Kleckner via llvm-dev <llvm-dev at lists.llvm.org> writes:
> Saying that sanitizers "do not support static linking" seems
strange
> because we strongly advise users to link the sanitizer libraries
> themselves statically. This gives the best user experience, because it
> puts the interceptors into the executable, so they always come first
> in dynamic symbol search order.
>
> What I think you are saying is that the sanitizers do not support
> linking statically with libc. This is probably true. It would be
> difficult to modify the interceptors to work when libc is linked
> statically, so it's probably not supported.
>
> Hopefully that explains why the sanitizer libraries are static, but
> they don't exactly support linking with -static.
>
> On Thu, Aug 23, 2018 at 10:04 AM David Greene via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
>
> I am guessing that sanitizers are not supported with static
> linking.
> The AddressSanitizer document says so explicitly but the others do
> not.
> Yet the interceptors use dlsym to set up certain system calls, so
> by
> definition they can't support static linking.
>
> Given that sanitizers won't work with static linking, why are
> static
> libraries created for them?
>
> Should there be an error check in the interceptor code for dlsym
> returning nullptr? I just spent a bunch of time tracking down a
> problem
> that turned out to be dlsym returning nullptr due to static
> linking.
>
> It seems like either users should not be able to link sanitizer
> runtime
> when linking statically or there should be a runtime check that
> dlsym
> doesn't return nullptr. Ideally both.
>
> -David
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev