> The problem is that once you assume that, you've completely
> lost control over what happens when the program's behavior is
> undefined due to memory lifetime bugs in the application (UAF/DF/etc) 

Yes, but that's exactly the contract between the C language and the programmer: to leave everything in the hands of people for maximum control and efficiency.

If we need more "protection", then we can just use other more advanced (and inefficient) languages.

Also, because of the enormous flexibility that C gives the programmer (and also hugely destructive when used incorrectly). So I'm afraid that the various checks we do at such a high cost may also be "better than nothing" to prevent bugs?
 
Of course, these tradeoffs are indeed a delicate matter. It's hard to have a definitive answer.

--

   Best Regards
  BaiYang
  baiyang@gmail.com
  http://i.baiy.cn
**** < END OF EMAIL > **** 
 
 
From: Rich Felker
Date: 2022-09-20 20:16
To: baiyang
CC: musl
Subject: Re: Re: [musl] The heap memory performance (malloc/free/realloc) is significantly degraded in musl 1.2 (compared to 1.1)
On Tue, Sep 20, 2022 at 01:56:12PM +0800, baiyang wrote:
> > //     This multi-threaded access to the pagemap is safe for fairly
> > //     subtle reasons.  We basically assume that when an object X is
> > //     allocated by thread A and deallocated by thread B, there must
> > //     have been appropriate synchronization in the handoff of object
> > //     X from thread A to thread B.
>
> Thanks for your information.
> I feel this assumption is very reasonable: you can't have one thread
> doing "free(p)" while another thread is accessing the block pointed
> to by p without any synchronization mechanism at the same time.
 
That's a correct assumption given that the program's behavior is
defined. The problem is that once you assume that, you've completely
lost control over what happens when the program's behavior is
undefined due to memory lifetime bugs in the application (UAF/DF/etc)
and an allocator that does this will necessarily allow bugs in the
lifetimes of objects that aren't inherently exploitable on the
application level (don't contain pointers that will be written through
that might clobber other application state, etc.) to be used to gain
control over the allocator state and eventually gain control over the
flow of execution through manipulating other allocated objects that
are attack vectors.
 
A hardened allocator like mallocng or hardened_malloc does not make
assumptions about the validity of data reachable for clobbering
through application bugs that haven't already yielded a high level of
control to the attacker. Instead, the metadata assumed to be valid is
at "secret locations" outside the area where application data is
stored that are intended not to be determinable without already having
some strong level of control over execution flow. And on top of that,
it's cross-validated as much as possible.
 
We also do validation where we can of the "in-band" data that is
easily reachable by overflows in application buffers, etc. This both
blocks a lot of potential exploits, and catches application bugs that
could be exploitable before they turn into exploits, by having them
crash on developers'/packagers' systems before they're deployed and
getting the root cause investigated and fixed.
 
Rich