Kernel: Demonstrate setpgid confusion bug

Technically, this can be 'exploited' to set the pgid of an exploiting process
to a near-arbitrary new pgid. This can cause conflicts when assigning future pgids,
destroys the session-boundary, and might confuse future pgid-to-session lookups.

In practice, I can't come up with a way that this causes actual harm.
This commit is contained in:
Ben Wiederhake 2020-08-08 21:13:07 +02:00 committed by Andreas Kling
parent cc6d5242d1
commit 7ab90e1a13
Notes: sideshowbarker 2024-07-19 04:06:58 +09:00

View File

@ -0,0 +1,296 @@
/*
* Copyright (c) Ben Wiederhake <BenWiederhake.GitHub@gmx.de>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include <AK/Assertions.h>
#include <AK/LogStream.h>
#include <fcntl.h>
#include <serenity.h>
#include <stdio.h>
#include <unistd.h>
/*
* Bug:
* A process can join a process group across sessions if both process groups
* do not have a leader (anymore). This can be used to join a session
* illegitimately. (Or, more harmlessly, to change the own PGID to an unused
* but arbitrary one, for example the PGID 0xDEADBEEF or the one that's going
* to be your program's session ID in the short-term future.)
*
* So what needs to happen:
* - There is session SA
* - There is session SB
* - There is a Process Group PGA in SA
* - There is a Process Group PGB in SB
* - PGA does not have a leader
* - PGB does not have a leader
* - There is a Process PA2 in PGA
* - There is a Process PB2 in PGB
* - PA2 calls setpgid(0, PGB)
* - Now PA2 and PB2 are in the same processgroup, but not in the same session. WHAAAAT! :^)
*
* Here's how to demonstrate the bug:
* - Time 0: PX forks into PA1
* - Time 1: PA1 creates a new session (SA) and pgrp (PGA)
* - Time 2: PA1 forks into PA2
* - Time 3: PA1 dies (PGA now has no leader)
* Note: PA2 never dies. Too much hassle.
* - Time 4: PX forks into PB1
* - Time 5: PB1 creates a new session (SB) and pgrp (PGB)
* - Time 6: PB1 forks into PB2
* - Time 7: PB1 dies (PGB now has no leader)
* - Time 8: PB2 calls pgrp(0, PGA)
* Note: PB2 writes "1" (exploit successfull) or "0" (bug is fixed) to a pipe
* - Time 9: If PX hasn't received any message yet through the pipe, it declares the test as failed (for lack of knowledge). Otherwise, it outputs accordingly.
*/
static constexpr useconds_t STEP_SIZE = 1100000;
static void fork_into(void (*fn)(void*), void* arg)
{
const pid_t rc = fork();
if (rc < 0) {
perror("fork");
exit(1);
}
if (rc > 0) {
const int disown_rc = disown(rc);
if (disown_rc < 0) {
perror("disown");
dbg() << "This might cause PA1 to remain in the Zombie state, "
"and thus in the process list, meaning the leader is "
"still 'alive' for the purpose of lookup.";
}
return;
}
fn(arg);
dbg() << "child finished (?)";
exit(1);
}
static void sleep_steps(useconds_t steps)
{
const int rc = usleep(steps * STEP_SIZE);
if (rc < 0) {
perror("usleep");
ASSERT_NOT_REACHED();
}
}
static void run_pa1(void*);
static void run_pa2(void*);
static void run_pb1(void*);
static void run_pb2(void*);
int main(int, char**)
{
// This entire function is the entirety of process PX.
// Time 0: PX forks into PA1
int fds[2];
// Serenity doesn't support O_NONBLOCK for pipes yet, so
// sadly the test will hang if something goes wrong.
if (pipe2(fds, 0) < 0) {
perror("pipe");
exit(1);
}
dbg() << "PX starts with SID=" << getsid(0) << ", PGID=" << getpgid(0) << ", PID=" << getpid() << ".";
dbg() << "PX forks into PA1";
fork_into(run_pa1, nullptr);
sleep_steps(4);
// Time 4: PX forks into PB1
dbg() << "PX forks into PB1";
fork_into(run_pb1, &(fds[1]));
sleep_steps(5);
// Time 9: If PX hasn't received any message yet through the pipe, it declares
// the test as failed (for lack of knowledge). Otherwise, it outputs accordingly.
dbg() << "PX reads from pipe";
unsigned char buf = 42;
ssize_t rc = read(fds[0], &buf, 1);
if (rc == 0) {
// In fact, we only reach this branch when *all* processes have died,
// including this one. So … should be unreachable.
printf("DOUBLE FAIL: pipe is closed, but we still have it open.\n"
"See debug log, some process probably crashed.\n");
exit(1);
}
if (rc < 0) {
if (errno == EAGAIN) {
printf("FAIL: pipe has no data. See debug log, some process os probably hanging.\n");
} else {
perror("read (unknown)");
}
exit(1);
}
ASSERT(rc == 1);
if (buf == 0) {
printf("PASS\n");
return 0;
}
if (buf == 1) {
printf("FAIL (exploit successful)\n");
return 1;
}
printf("FAIL, for some reason %c\n", buf);
return 1;
}
static void run_pa1(void*)
{
// Time 0: PX forks into PA1
sleep_steps(1);
// Time 1: PA1 creates a new session (SA) and pgrp (PGA)
dbg() << "PA1 starts with SID=" << getsid(0) << ", PGID=" << getpgid(0) << ", PID=" << getpid() << ".";
dbg() << "PA1 calls setsid()";
int rc = setsid();
if (rc < 0) {
perror("setsid (PA)");
ASSERT_NOT_REACHED();
}
dbg() << "PA1 did setsid() -> PGA=" << rc << ", SA=" << getsid(0) << ", yay!";
sleep_steps(1);
// Time 2: PA1 forks into PA2
dbg() << "PA1 forks into PA2";
fork_into(run_pa2, nullptr);
sleep_steps(1);
// Time 3: PA1 dies (PGA now has no leader)
dbg() << "PA1 dies. You should see a 'Reaped unparented process' "
"message with my ID next, OR THIS TEST IS MEANINGLESS "
"(see fork_into()).";
exit(0);
}
static void run_pa2(void*)
{
// Time 2: PA1 forks into PA2
dbg() << "PA2 starts with SID=" << getsid(0) << ", PGID=" << getpgid(0) << ", PID=" << getpid() << ".";
sleep_steps(18);
// pa_2 never *does* anything.
dbg() << "PA2 dies from boredom.";
exit(1);
}
static void run_pb1(void* pipe_fd_ptr)
{
// Time 4: PX forks into PB1
sleep_steps(1);
// Time 5: PB1 creates a new session (SB) and pgrp (PGB)
dbg() << "PB1 starts with SID=" << getsid(0) << ", PGID=" << getpgid(0) << ", PID=" << getpid() << ".";
dbg() << "PB1 calls setsid()";
int rc = setsid();
if (rc < 0) {
perror("setsid (PB)");
ASSERT_NOT_REACHED();
}
dbg() << "PB1 did setsid() -> PGB=" << rc << ", SB=" << getsid(0) << ", yay!";
sleep_steps(1);
// Time 6: PB1 forks into PB2
dbg() << "PB1 forks into PB2";
fork_into(run_pb2, pipe_fd_ptr);
sleep_steps(1);
// Time 7: PB1 dies (PGB now has no leader)
dbg() << "PB1 dies. You should see a 'Reaped unparented process' "
"message with my ID next, OR THIS TEST IS MEANINGLESS "
"(see fork_into()).";
exit(0);
}
static void simulate_sid_from_pgid(pid_t pgid)
{
pid_t rc = getpgid(pgid); // Same confusion as in the Kernel
int saved_errno = errno;
if (rc < 0 && saved_errno == ESRCH) {
dbg() << "The old get_sid_from_pgid(" << pgid << ") would return -1";
} else if (rc >= 0) {
dbg() << "FAIL: Process " << pgid << " still exists?! PGID is " << rc << ".";
} else {
perror("pgid (probably fail)");
}
}
static void run_pb2(void* pipe_fd_ptr)
{
// Time 6: PB1 forks into PB2
sleep_steps(2);
// Time 8: PB2 calls pgrp(0, PGA)
// Note: PB2 writes "1" (exploit successfull) or "0" (bug is fixed) to a pipe
dbg() << "PB2 starts with SID=" << getsid(0) << ", PGID=" << getpgid(0) << ", PID=" << getpid() << ".";
dbg() << "PB2 calls pgrp(0, PGA)";
int pga = getpid() - 3;
dbg() << "PB2: Actually, what is PGA? I guess it's " << pga << "?";
simulate_sid_from_pgid(pga);
int rc = setpgid(0, pga);
unsigned char to_write = 123;
if (rc == 0) {
dbg() << "PB2: setgpid SUCCESSFUL! CHANGED PGROUP!";
to_write = 1;
} else {
ASSERT(rc == -1);
switch (errno) {
case EACCES:
dbg() << "PB2: Failed with EACCES. Huh?!";
to_write = 101;
break;
case EINVAL:
dbg() << "PB2: Failed with EINVAL. Huh?!";
to_write = 102;
break;
case ESRCH:
dbg() << "PB2: Failed with ESRCH. Huh?!";
to_write = 103;
break;
case EPERM:
dbg() << "PB2: Failed with EPERM. Aww, no exploit today :^)";
to_write = 0;
break;
default:
dbg() << "PB2: Failed with errno=" << errno << "?!";
perror("setpgid");
to_write = 104;
break;
}
}
dbg() << "PB2 ends with SID=" << getsid(0) << ", PGID=" << getpgid(0) << ", PID=" << getpid() << ".";
int* pipe_fd = static_cast<int*>(pipe_fd_ptr);
ASSERT(*pipe_fd);
rc = write(*pipe_fd, &to_write, 1);
if (rc != 1) {
dbg() << "Wrote only " << rc << " bytes instead of 1?!";
exit(1);
}
exit(0);
}