bpf: fix truncated jump targets on heavy expansions

Recently during testing, I ran into the following panic:

  [  207.892422] Internal error: Accessing user space memory outside uaccess.h routines: 96000004 [#1] SMP
  [  207.901637] Modules linked in: binfmt_misc [...]
  [  207.966530] CPU: 45 PID: 2256 Comm: test_verifier Tainted: G        W         4.17.0-rc3+ #7
  [  207.974956] Hardware name: FOXCONN R2-1221R-A4/C2U4N_MB, BIOS G31FB18A 03/31/2017
  [  207.982428] pstate: 60400005 (nZCv daif +PAN -UAO)
  [  207.987214] pc : bpf_skb_load_helper_8_no_cache+0x34/0xc0
  [  207.992603] lr : 0xffff000000bdb754
  [  207.996080] sp : ffff000013703ca0
  [  207.999384] x29: ffff000013703ca0 x28: 0000000000000001
  [  208.004688] x27: 0000000000000001 x26: 0000000000000000
  [  208.009992] x25: ffff000013703ce0 x24: ffff800fb4afcb00
  [  208.015295] x23: ffff00007d2f5038 x22: ffff00007d2f5000
  [  208.020599] x21: fffffffffeff2a6f x20: 000000000000000a
  [  208.025903] x19: ffff000009578000 x18: 0000000000000a03
  [  208.031206] x17: 0000000000000000 x16: 0000000000000000
  [  208.036510] x15: 0000ffff9de83000 x14: 0000000000000000
  [  208.041813] x13: 0000000000000000 x12: 0000000000000000
  [  208.047116] x11: 0000000000000001 x10: ffff0000089e7f18
  [  208.052419] x9 : fffffffffeff2a6f x8 : 0000000000000000
  [  208.057723] x7 : 000000000000000a x6 : 00280c6160000000
  [  208.063026] x5 : 0000000000000018 x4 : 0000000000007db6
  [  208.068329] x3 : 000000000008647a x2 : 19868179b1484500
  [  208.073632] x1 : 0000000000000000 x0 : ffff000009578c08
  [  208.078938] Process test_verifier (pid: 2256, stack limit = 0x0000000049ca7974)
  [  208.086235] Call trace:
  [  208.088672]  bpf_skb_load_helper_8_no_cache+0x34/0xc0
  [  208.093713]  0xffff000000bdb754
  [  208.096845]  bpf_test_run+0x78/0xf8
  [  208.100324]  bpf_prog_test_run_skb+0x148/0x230
  [  208.104758]  sys_bpf+0x314/0x1198
  [  208.108064]  el0_svc_naked+0x30/0x34
  [  208.111632] Code: 91302260 f9400001 f9001fa1 d2800001 (29500680)
  [  208.117717] ---[ end trace 263cb8a59b5bf29f ]---

The program itself which caused this had a long jump over the whole
instruction sequence where all of the inner instructions required
heavy expansions into multiple BPF instructions. Additionally, I also
had BPF hardening enabled which requires once more rewrites of all
constant values in order to blind them. Each time we rewrite insns,
bpf_adj_branches() would need to potentially adjust branch targets
which cross the patchlet boundary to accommodate for the additional
delta. Eventually that lead to the case where the target offset could
not fit into insn->off's upper 0x7fff limit anymore where then offset
wraps around becoming negative (in s16 universe), or vice versa
depending on the jump direction.

Therefore it becomes necessary to detect and reject any such occasions
in a generic way for native eBPF and cBPF to eBPF migrations. For
the latter we can simply check bounds in the bpf_convert_filter()'s
BPF_EMIT_JMP helper macro and bail out once we surpass limits. The
bpf_patch_insn_single() for native eBPF (and cBPF to eBPF in case
of subsequent hardening) is a bit more complex in that we need to
detect such truncations before hitting the bpf_prog_realloc(). Thus
the latter is split into an extra pass to probe problematic offsets
on the original program in order to fail early. With that in place
and carefully tested I no longer hit the panic and the rewrites are
rejected properly. The above example panic I've seen on bpf-next,
though the issue itself is generic in that a guard against this issue
in bpf seems more appropriate in this case.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

kernel/bpf/core.c

@@ -218,47 +218,84 @@ int bpf_prog_calc_tag(struct bpf_prog *fp)
 	return 0;
 }
 
-static void bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta)
+static int bpf_adj_delta_to_imm(struct bpf_insn *insn, u32 pos, u32 delta,
+				u32 curr, const bool probe_pass)
 {
+	const s64 imm_min = S32_MIN, imm_max = S32_MAX;
+	s64 imm = insn->imm;
+
+	if (curr < pos && curr + imm + 1 > pos)
+		imm += delta;
+	else if (curr > pos + delta && curr + imm + 1 <= pos + delta)
+		imm -= delta;
+	if (imm < imm_min || imm > imm_max)
+		return -ERANGE;
+	if (!probe_pass)
+		insn->imm = imm;
+	return 0;
+}
+
+static int bpf_adj_delta_to_off(struct bpf_insn *insn, u32 pos, u32 delta,
+				u32 curr, const bool probe_pass)
+{
+	const s32 off_min = S16_MIN, off_max = S16_MAX;
+	s32 off = insn->off;
+
+	if (curr < pos && curr + off + 1 > pos)
+		off += delta;
+	else if (curr > pos + delta && curr + off + 1 <= pos + delta)
+		off -= delta;
+	if (off < off_min || off > off_max)
+		return -ERANGE;
+	if (!probe_pass)
+		insn->off = off;
+	return 0;
+}
+
+static int bpf_adj_branches(struct bpf_prog *prog, u32 pos, u32 delta,
+			    const bool probe_pass)
+{
+	u32 i, insn_cnt = prog->len + (probe_pass ? delta : 0);
 	struct bpf_insn *insn = prog->insnsi;
-	u32 i, insn_cnt = prog->len;
-	bool pseudo_call;
-	u8 code;
-	int off;
+	int ret = 0;
 
 	for (i = 0; i < insn_cnt; i++, insn++) {
+		u8 code;
+
+		/* In the probing pass we still operate on the original,
+		 * unpatched image in order to check overflows before we
+		 * do any other adjustments. Therefore skip the patchlet.
+		 */
+		if (probe_pass && i == pos) {
+			i += delta + 1;
+			insn++;
+		}
 		code = insn->code;
-		if (BPF_CLASS(code) != BPF_JMP)
-			continue;
-		if (BPF_OP(code) == BPF_EXIT)
+		if (BPF_CLASS(code) != BPF_JMP ||
+		    BPF_OP(code) == BPF_EXIT)
 			continue;
+		/* Adjust offset of jmps if we cross patch boundaries. */
 		if (BPF_OP(code) == BPF_CALL) {
-			if (insn->src_reg == BPF_PSEUDO_CALL)
-				pseudo_call = true;
-			else
+			if (insn->src_reg != BPF_PSEUDO_CALL)
 				continue;
+			ret = bpf_adj_delta_to_imm(insn, pos, delta, i,
+						   probe_pass);
 		} else {
-			pseudo_call = false;
+			ret = bpf_adj_delta_to_off(insn, pos, delta, i,
+						   probe_pass);
 		}
-		off = pseudo_call ? insn->imm : insn->off;
-
-		/* Adjust offset of jmps if we cross boundaries. */
-		if (i < pos && i + off + 1 > pos)
-			off += delta;
-		else if (i > pos + delta && i + off + 1 <= pos + delta)
-			off -= delta;
-
-		if (pseudo_call)
-			insn->imm = off;
-		else
-			insn->off = off;
+		if (ret)
+			break;
 	}
+
+	return ret;
 }
 
 struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 				       const struct bpf_insn *patch, u32 len)
 {
 	u32 insn_adj_cnt, insn_rest, insn_delta = len - 1;
+	const u32 cnt_max = S16_MAX;
 	struct bpf_prog *prog_adj;
 
 	/* Since our patchlet doesn't expand the image, we're done. */
@@ -269,6 +306,15 @@ struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 
 	insn_adj_cnt = prog->len + insn_delta;
 
+	/* Reject anything that would potentially let the insn->off
+	 * target overflow when we have excessive program expansions.
+	 * We need to probe here before we do any reallocation where
+	 * we afterwards may not fail anymore.
+	 */
+	if (insn_adj_cnt > cnt_max &&
+	    bpf_adj_branches(prog, off, insn_delta, true))
+		return NULL;
+
 	/* Several new instructions need to be inserted. Make room
 	 * for them. Likely, there's no need for a new allocation as
 	 * last page could have large enough tailroom.
@@ -294,7 +340,11 @@ struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 		sizeof(*patch) * insn_rest);
 	memcpy(prog_adj->insnsi + off, patch, sizeof(*patch) * len);
 
-	bpf_adj_branches(prog_adj, off, insn_delta);
+	/* We are guaranteed to not fail at this point, otherwise
+	 * the ship has sailed to reverse to the original state. An
+	 * overflow cannot happen at this point.
+	 */
+	BUG_ON(bpf_adj_branches(prog_adj, off, insn_delta, false));
 
 	return prog_adj;
 }

net/core/filter.c

@@ -481,11 +481,18 @@ static int bpf_convert_filter(struct sock_filter *prog, int len,
 
 #define BPF_EMIT_JMP							\
 	do {								\
+		const s32 off_min = S16_MIN, off_max = S16_MAX;		\
+		s32 off;						\
+									\
 		if (target >= len || target < 0)			\
 			goto err;					\
-		insn->off = addrs ? addrs[target] - addrs[i] - 1 : 0;	\
+		off = addrs ? addrs[target] - addrs[i] - 1 : 0;		\
 		/* Adjust pc relative offset for 2nd or 3rd insn. */	\
-		insn->off -= insn - tmp_insns;				\
+		off -= insn - tmp_insns;				\
+		/* Reject anything not fitting into insn->off. */	\
+		if (off < off_min || off > off_max)			\
+			goto err;					\
+		insn->off = off;					\
 	} while (0)
 
 		case BPF_JMP | BPF_JA: