Re: [PATCH v8 03/10] Remove __morecore and __default_morecore

[Guillaume Morin <guillaume@morinfr.org>]

On 14 Jul 14:42, Adhemerval Zanella wrote:
> I agree with Carlos and Siddhesh rationale that a dangling function
> pointer in an essential code as malloc is a security liability and it 
> is about time to get removed (as there is PoC out there that does it to 
> exploit glibc malloc).

To reiterate, I have not argued to keep the hook as is. I do understand
the valid concerns about potential exploitation with the current scheme.

> However I think we can work towards a solution to enable within glibc
> instead of pushing the support on libhugetlfs, which does *not* aim to
> be a malloc replacement but rather a way to explicit provide and
> manage large pages.
> There is some discussion last year about providing large pages support
> directly on glibc without resorting to THP [1].  The idea is to hook
> up the mmap/madvise with the required flags, which seems similar of
> what is required by libhugetls morecore() implementation [2]. So I think
> it should be feasible to add support for the required bits on glibc
> and provide a tunable to actually use it. 

That would be ideal.

> Guillaume, would this be suffice for your use cases.  The libhugetlsfs
> projects does provide more functionality than glibc scope, for instance
> a linker script to for bss/data and exe section to large parges, so I
> glibc support would be a complement to what libhugetls provides.

Yes, it would be perfect for my use case.

Guillaume.

-- 
Guillaume Morin <guillaume@morinfr.org>