All of the below seemed really strange to me, because I researched it when I wrote fakeroot-ng. To make sure, I wrote the following program:

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <sys/mman.h>
#include <unistd.h>
#include <stdio.h>

int main() {
    int fd = open("testfile", O_CREAT|O_TRUNC|O_WRONLY, 0666);
    ftruncate(fd, 4096);
    close(fd);

    fd = open("testfile", O_RDONLY);
    char *map = (char *)mmap( nullptr, 4096, PROT_READ, MAP_SHARED|MAP_FILE, fd, 0 );
    if( map==MAP_FAILED ) {
        perror("mmap failed");
        return 1;
    }

    if( mprotect(map, 4096, PROT_READ|PROT_WRITE)<0 ) {
        perror("mprotect failed");
        return 1;
    }
}

As I expected, the mprotect fails with "Permission denied". It is not possible to change the mapping to allow writing to the memory when it is a shared mapping of a file opened in read-only. I have no idea what the other resources are saying, but this is what I see.

Since the above completely describes what the loader does to an executable file, I don't see how mprotect can be used there either.

With that said, I know that debuggers often change the text segment of debugees in order to insert breakpoints. I have to admit that testing the mechanism by which that happens is on my todo list, but I have not got around to actually doing it. I am guessing that writing from the debugger switches the map mode to MAP_PRIVATE (i.e. - copy on write), which means there is no need to flush the changes to the file.

With that said, I have been unable to find a way to trigger that change by any other means, so a mapping made from a file opened in read only mode is, as far as I can tell, safe from manipulation barring the ptrace route.

Shachar

I am researching this issue and I am confused with the finding

Some articles, ex https://shanetully.com/2013/12/writing-a-self-mutating-x86_64-c-program/
state that mprotect() can change protection of executable section.

As I understanf pte entry has page protection bits set to RO so  mprotect should change pte which is loaded to MMU/TLB. Why kernel can not refuse do perform this mprotect call(). Whu we do norhave kernel config options to forbid user-space mutable code as security feature?



From the other side,  when  run-time linker or elf_loader loads the executable it uses MAP_DENYWRITE which protect executable file from being overwritten. 

But writing to  executable text  will make  page dirty and require the write-back which is disabled by MAP_DENYWRITE. (or it might be disable for other processes except current, I am not sure?)


To add to the confusion, the following quote from the LWN articlle 
https://lwn.net/Articles/422487/ about CONFIG_DEBUG_SET_MODULE_RONX 
"Marking the kernel module pages as RO and/or NX is important not only because it is consistent with how the rest of the kernel pages are handled"
  
Digging dipper I see that ARM since kernel version 4.11 has CONFIG_STRICT_KERNEL_RWX ,  and as I understand it is enforced in hardware.

But I am not sure that some variant of pte_clear(), pte_mkexec(0 can not disable it.

So let me cut to final qiestion:

Suppose I want to cut off dynamic code instrumentation, like ftrace and friends.
Is it achievable at least at ARM architecture to enforce RO+X at hardware or kernel? 

Thanks to all folks for reading till this point.

Regards
Lev



_______________________________________________
Linux-il mailing list
Linux-il@cs.huji.ac.il
http://mailman.cs.huji.ac.il/mailman/listinfo/linux-il