[cap-talk] C-like Capability language

William Pearson wil.pearson at gmail.com
Thu Jul 31 16:58:22 CDT 2008


Some people may remember me from my previous post. Quick recap I want
to do capability based security in a VM designed for real-time systems
which uses an agoric system to allocate system resources. This uses a
judge (generally human, potentially automated) that distributes credit
to the programs that do well to allow them to buy the computational
resources, rather than using real money.  So that the system would
come to be controlled by programs that do whatever the judge valued.
Also I am interested in allowing mobile code down to the
scheduler/device driver level, so a bit further down than the classic
agoric system.

As such I am trying to design a C-like language, fairly low level but
abstracting a few things away. Basically the syntax would be C
(structs and all), with added syntax to deal with capabilities and
domains. So not an object capability model, but it still might be
interesting.

There would be also be an assembler code that could be inlined for
very low level stuff. The VM would enforce all the capabilities. The
language constructs would just be to make it explicit.

domain d;
//All code compiled after this declaration before another such
declaration would be considered in the same domain. It has no meaning
between compilations, you can't compile domains seperately

typedef struct bank
{
  int cash;
  int * array;
} struct;

bank mybank;
mybank.array = malloc(8*sizeof(int));
int temp;
bank% bankcap= @bank;
// % specifies this is a capability to a bank (leaving it in a
specially defined NULL harmless state if undefined)
// @ creates a capability to an existing object in the same way that &
gets a pointer (under the hood it would create a read only capability,
with a length equal to the sizeof bank )

bankcap-]cash = 0;
// -] acts as the -> operator in C but for capabilities. This would
cause an error as by default capabilities to structs are read only,
capabilities to function execute only etc.
int temp = bankcap-]array[1] ;
// also not allowed as it is implicitly dereferencing a pointer;


rescind(bankcap.lock, bankcap.key)
// capabilities would have two parts locks and keys. Keys would be the
part duplicated and passed around. You can also have multiple locks
for the same object. This functions would mean that the key would no
longer be valid for the lock. If you create two capabilities to the
same function they would have different locks and keys, if you copy a
capability it would get the same lock and a copy of the key. This
allows the lock to be changed, so the key is no longer valid.

There will be lots of functions that programs can only use on
capabilities with locks defined in their own domain.

Has something similar been done before (a google of cap-talk suggested
some thought had been put into it for EROS), and does anyone have any
better ideas for syntax other than slavishly copying the C syntax?

  Will Pearson


More information about the cap-talk mailing list