silly memory question ...

silly memory question ...

[Josh Fryman]

hi all,

i'm looking to do something stupid.  can anyone tell me why it won't
work? (is there some kind of special protection on EXE memory pages?)

i have an application, App1, that has a 64KB function (64KB of NOP)
"void funcX(void)".  App1 opens a socket to another application, App2,
which is going to feed code to App1 to stick in funcX.

that is, i've got some other set of functions i plan on feeding to App1
and they're all going to look like "funcX" to App1.  those functions are
coming from App2, which is chewing on a bunch of var-things to spit out 
just the right code for App1.

now, when i try to do this, i get Random Stuff.  sometimes, i just get
a segfault.  sometimes, it seems to work, but the memory at funcX is
hosed if i do a hex-dump compared to what i received from App2.  either
way, it isn't doing what i want...

is there some trick here to make this work?  or am i SOL?

[is there a better place to ask this question?]

thanks,

josh

Re: silly memory question ...

[Josh Fryman]

hi all,

thanks to all of you for your suggestions... it turns out this
is an ARM-specific issue, and maybe some other platforms as well.
(or maybe just my specific ARM kernel/glibc combo. ;)

it turns out x86 ignores some "modes" for pages in memory.  or
maybe ld.so doesn't bother to set them.  anyway, the code snippet 
below "fixes" the pages i need fixed to become R/W/X.

on *my* ARM, you can't default execute data pages; you can't 
write text pages; etc, etc. (my kernel is 2.4.0 with glibc 2.1.2.)

i don't know what other platforms enforce this, but hopefully 
this is as educational for those suggesting solutions as it has 
been for me. if i had specified i was running on ARM originally, 
it may have made things clearer.

thanks again,

josh

   // fix permissions on __app_code - we need R/W/X, not just R/X ... 
   // ARM-specific problem, but this fix should work on ALL platform targets

   printf("Fixing TEXT segment permissions for R/W/X....\n");
   for (i=0; i<APP_CODE_K*1024/PAGESIZE; i++)
   {
      code = (ui32*) ((((ui32)__app_code + i*PAGESIZE) - 1) & ~(PAGESIZE-1));
      if (mprotect( code, PAGESIZE, PROT_READ|PROT_WRITE|PROT_EXEC) )
      {
         printf("mprotect() on 0x%08x failed!\nerrno (%d) indicates status: ", code, errno);
	 switch (errno) 
	 {
	    case EINVAL: printf("EINVAL: not valid ptr, or not multiple of PAGESIZE (%d)\n", PAGESIZE);  
                         break;
	    case EFAULT: printf("EFAULT: memory can not be accessed.\n"); 
                         break;
	    case EACCES: printf("EACCES: memory can not be given specified access modes.\n"); 
                         break;
	    case ENOMEM: printf("ENOMEM: internal kernel structures could not be allocated.\n"); 
                         break;
	    default:     printf("??????: unknown error result.\n");
	 }
         exit( CLNT_MPROT_FAIL );
      }
   }

note:
the dummy "nop" function "__app_code()" is APP_CODE_K in size of
KBs of NOPs.

note 2:
i hope the KiB/MiB/whatever standard suggestion is never adopted 
and revoked really soon now.