bpf: direct packet access

Extended BPF carried over two instructions from classic to access
packet data: LD_ABS and LD_IND. They're highly optimized in JITs,
but due to their design they have to do length check for every access.
When BPF is processing 20M packets per second single LD_ABS after JIT
is consuming 3% cpu. Hence the need to optimize it further by amortizing
the cost of 'off < skb_headlen' over multiple packet accesses.
One option is to introduce two new eBPF instructions LD_ABS_DW and LD_IND_DW
with similar usage as skb_header_pointer().
The kernel part for interpreter and x64 JIT was implemented in [1], but such
new insns behave like old ld_abs and abort the program with 'return 0' if
access is beyond linear data. Such hidden control flow is hard to workaround
plus changing JITs and rolling out new llvm is incovenient.

Therefore allow cls_bpf/act_bpf program access skb->data directly:
int bpf_prog(struct __sk_buff *skb)
{
  struct iphdr *ip;

  if (skb->data + sizeof(struct iphdr) + ETH_HLEN > skb->data_end)
      /* packet too small */
      return 0;

  ip = skb->data + ETH_HLEN;

  /* access IP header fields with direct loads */
  if (ip->version != 4 || ip->saddr == 0x7f000001)
      return 1;
  [...]
}

This solution avoids introduction of new instructions. llvm stays
the same and all JITs stay the same, but verifier has to work extra hard
to prove safety of the above program.

For XDP the direct store instructions can be allowed as well.

The skb->data is NET_IP_ALIGNED, so for common cases the verifier can check
the alignment. The complex packet parsers where packet pointer is adjusted
incrementally cannot be tracked for alignment, so allow byte access in such cases
and misaligned access on architectures that define efficient_unaligned_access

[1] https://git.kernel.org/cgit/linux/kernel/git/ast/bpf.git/?h=ld_abs_dwSigned-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/uapi/linux/bpf.h

@@ -370,6 +370,8 @@ struct __sk_buff {
 	__u32 cb[5];
 	__u32 hash;
 	__u32 tc_classid;
+	__u32 data;
+	__u32 data_end;
 };
 
 struct bpf_tunnel_key {

kernel/bpf/core.c

@@ -794,6 +794,11 @@ void __weak bpf_int_jit_compile(struct bpf_prog *prog)
 {
 }
 
+bool __weak bpf_helper_changes_skb_data(void *func)
+{
+	return false;
+}
+
 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
  */

kernel/bpf/verifier.c

 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
+ * Copyright (c) 2016 Facebook
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of version 2 of the GNU General Public
@@ -136,13 +137,32 @@ enum bpf_reg_type {
 	FRAME_PTR,		 /* reg == frame_pointer */
 	PTR_TO_STACK,		 /* reg == frame_pointer + imm */
 	CONST_IMM,		 /* constant integer value */
+
+	/* PTR_TO_PACKET represents:
+	 * skb->data
+	 * skb->data + imm
+	 * skb->data + (u16) var
+	 * skb->data + (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) menas that 'imm' was added
+	 */
+	PTR_TO_PACKET,
+	PTR_TO_PACKET_END,	 /* skb->data + headlen */
 };
 
 struct reg_state {
 	enum bpf_reg_type type;
 	union {
-		/* valid when type == CONST_IMM | PTR_TO_STACK */
-		long imm;
+		/* valid when type == CONST_IMM | PTR_TO_STACK | UNKNOWN_VALUE */
+		s64 imm;
+
+		/* valid when type == PTR_TO_PACKET* */
+		struct {
+			u32 id;
+			u16 off;
+			u16 range;
+		};
 
 		/* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
 		 *   PTR_TO_MAP_VALUE_OR_NULL
@@ -247,6 +267,8 @@ static const char * const reg_type_str[] = {
 	[FRAME_PTR]		= "fp",
 	[PTR_TO_STACK]		= "fp",
 	[CONST_IMM]		= "imm",
+	[PTR_TO_PACKET]		= "pkt",
+	[PTR_TO_PACKET_END]	= "pkt_end",
 };
 
 static void print_verifier_state(struct verifier_state *state)
@@ -262,7 +284,12 @@ static void print_verifier_state(struct verifier_state *state)
 			continue;
 		verbose(" R%d=%s", i, reg_type_str[t]);
 		if (t == CONST_IMM || t == PTR_TO_STACK)
-			verbose("%ld", reg->imm);
+			verbose("%lld", reg->imm);
+		else if (t == PTR_TO_PACKET)
+			verbose("(id=%d,off=%d,r=%d)",
+				reg->id, reg->off, reg->range);
+		else if (t == UNKNOWN_VALUE && reg->imm)
+			verbose("%lld", reg->imm);
 		else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE ||
 			 t == PTR_TO_MAP_VALUE_OR_NULL)
 			verbose("(ks=%d,vs=%d)",
@@ -548,6 +575,8 @@ static bool is_spillable_regtype(enum bpf_reg_type type)
 	case PTR_TO_MAP_VALUE_OR_NULL:
 	case PTR_TO_STACK:
 	case PTR_TO_CTX:
+	case PTR_TO_PACKET:
+	case PTR_TO_PACKET_END:
 	case FRAME_PTR:
 	case CONST_PTR_TO_MAP:
 		return true;
@@ -647,6 +676,27 @@ static int check_map_access(struct verifier_env *env, u32 regno, int off,
 	return 0;
 }
 
+#define MAX_PACKET_OFF 0xffff
+
+static int check_packet_access(struct verifier_env *env, u32 regno, int off,
+			       int size)
+{
+	struct reg_state *regs = env->cur_state.regs;
+	struct reg_state *reg = &regs[regno];
+	int linear_size = (int) reg->range - (int) reg->off;
+
+	if (linear_size < 0 || linear_size >= MAX_PACKET_OFF) {
+		verbose("verifier bug\n");
+		return -EFAULT;
+	}
+	if (off < 0 || off + size > linear_size) {
+		verbose("invalid access to packet, off=%d size=%d, allowed=%d\n",
+			off, size, linear_size);
+		return -EACCES;
+	}
+	return 0;
+}
+
 /* check access to 'struct bpf_context' fields */
 static int check_ctx_access(struct verifier_env *env, int off, int size,
 			    enum bpf_access_type t)
@@ -677,6 +727,45 @@ static bool is_pointer_value(struct verifier_env *env, int regno)
 	}
 }
 
+static int check_ptr_alignment(struct verifier_env *env, struct reg_state *reg,
+			       int off, int size)
+{
+	if (reg->type != PTR_TO_PACKET) {
+		if (off % size != 0) {
+			verbose("misaligned access off %d size %d\n", off, size);
+			return -EACCES;
+		} else {
+			return 0;
+		}
+	}
+
+	switch (env->prog->type) {
+	case BPF_PROG_TYPE_SCHED_CLS:
+	case BPF_PROG_TYPE_SCHED_ACT:
+		break;
+	default:
+		verbose("verifier is misconfigured\n");
+		return -EACCES;
+	}
+
+	if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS))
+		/* misaligned access to packet is ok on x86,arm,arm64 */
+		return 0;
+
+	if (reg->id && size != 1) {
+		verbose("Unknown packet alignment. Only byte-sized access allowed\n");
+		return -EACCES;
+	}
+
+	/* skb->data is NET_IP_ALIGN-ed */
+	if ((NET_IP_ALIGN + reg->off + off) % size != 0) {
+		verbose("misaligned packet access off %d+%d+%d size %d\n",
+			NET_IP_ALIGN, reg->off, off, size);
+		return -EACCES;
+	}
+	return 0;
+}
+
 /* check whether memory at (regno + off) is accessible for t = (read | write)
  * if t==write, value_regno is a register which value is stored into memory
  * if t==read, value_regno is a register which will receive the value from memory
@@ -698,10 +787,9 @@ static int check_mem_access(struct verifier_env *env, u32 regno, int off,
 	if (size < 0)
 		return size;
 
-	if (off % size != 0) {
-		verbose("misaligned access off %d size %d\n", off, size);
-		return -EACCES;
-	}
+	err = check_ptr_alignment(env, reg, off, size);
+	if (err)
+		return err;
 
 	if (reg->type == PTR_TO_MAP_VALUE) {
 		if (t == BPF_WRITE && value_regno >= 0 &&
@@ -720,8 +808,16 @@ static int check_mem_access(struct verifier_env *env, u32 regno, int off,
 			return -EACCES;
 		}
 		err = check_ctx_access(env, off, size, t);
-		if (!err && t == BPF_READ && value_regno >= 0)
+		if (!err && t == BPF_READ && value_regno >= 0) {
 			mark_reg_unknown_value(state->regs, value_regno);
+			if (off == offsetof(struct __sk_buff, data) &&
+			    env->allow_ptr_leaks)
+				/* note that reg.[id|off|range] == 0 */
+				state->regs[value_regno].type = PTR_TO_PACKET;
+			else if (off == offsetof(struct __sk_buff, data_end) &&
+				 env->allow_ptr_leaks)
+				state->regs[value_regno].type = PTR_TO_PACKET_END;
+		}
 
 	} else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) {
 		if (off >= 0 || off < -MAX_BPF_STACK) {
@@ -739,11 +835,28 @@ static int check_mem_access(struct verifier_env *env, u32 regno, int off,
 		} else {
 			err = check_stack_read(state, off, size, value_regno);
 		}
+	} else if (state->regs[regno].type == PTR_TO_PACKET) {
+		if (t == BPF_WRITE) {
+			verbose("cannot write into packet\n");
+			return -EACCES;
+		}
+		err = check_packet_access(env, regno, off, size);
+		if (!err && t == BPF_READ && value_regno >= 0)
+			mark_reg_unknown_value(state->regs, value_regno);
 	} else {
 		verbose("R%d invalid mem access '%s'\n",
 			regno, reg_type_str[reg->type]);
 		return -EACCES;
 	}
+
+	if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks &&
+	    state->regs[value_regno].type == UNKNOWN_VALUE) {
+		/* 1 or 2 byte load zero-extends, determine the number of
+		 * zero upper bits. Not doing it fo 4 byte load, since
+		 * such values cannot be added to ptr_to_packet anyway.
+		 */
+		state->regs[value_regno].imm = 64 - size * 8;
+	}
 	return err;
 }
 
@@ -1001,6 +1114,29 @@ static int check_raw_mode(const struct bpf_func_proto *fn)
 	return count > 1 ? -EINVAL : 0;
 }
 
+static void clear_all_pkt_pointers(struct verifier_env *env)
+{
+	struct verifier_state *state = &env->cur_state;
+	struct reg_state *regs = state->regs, *reg;
+	int i;
+
+	for (i = 0; i < MAX_BPF_REG; i++)
+		if (regs[i].type == PTR_TO_PACKET ||
+		    regs[i].type == PTR_TO_PACKET_END)
+			mark_reg_unknown_value(regs, i);
+
+	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
+		if (state->stack_slot_type[i] != STACK_SPILL)
+			continue;
+		reg = &state->spilled_regs[i / BPF_REG_SIZE];
+		if (reg->type != PTR_TO_PACKET &&
+		    reg->type != PTR_TO_PACKET_END)
+			continue;
+		reg->type = UNKNOWN_VALUE;
+		reg->imm = 0;
+	}
+}
+
 static int check_call(struct verifier_env *env, int func_id)
 {
 	struct verifier_state *state = &env->cur_state;
@@ -1008,6 +1144,7 @@ static int check_call(struct verifier_env *env, int func_id)
 	struct reg_state *regs = state->regs;
 	struct reg_state *reg;
 	struct bpf_call_arg_meta meta;
+	bool changes_data;
 	int i, err;
 
 	/* find function prototype */
@@ -1030,6 +1167,8 @@ static int check_call(struct verifier_env *env, int func_id)
 		return -EINVAL;
 	}
 
+	changes_data = bpf_helper_changes_skb_data(fn->func);
+
 	memset(&meta, 0, sizeof(meta));
 
 	/* We only support one arg being in raw mode at the moment, which
@@ -1100,6 +1239,189 @@ static int check_call(struct verifier_env *env, int func_id)
 	if (err)
 		return err;
 
+	if (changes_data)
+		clear_all_pkt_pointers(env);
+	return 0;
+}
+
+static int check_packet_ptr_add(struct verifier_env *env, struct bpf_insn *insn)
+{
+	struct reg_state *regs = env->cur_state.regs;
+	struct reg_state *dst_reg = &regs[insn->dst_reg];
+	struct reg_state *src_reg = &regs[insn->src_reg];
+	s32 imm;
+
+	if (BPF_SRC(insn->code) == BPF_K) {
+		/* pkt_ptr += imm */
+		imm = insn->imm;
+
+add_imm:
+		if (imm <= 0) {
+			verbose("addition of negative constant to packet pointer is not allowed\n");
+			return -EACCES;
+		}
+		if (imm >= MAX_PACKET_OFF ||
+		    imm + dst_reg->off >= MAX_PACKET_OFF) {
+			verbose("constant %d is too large to add to packet pointer\n",
+				imm);
+			return -EACCES;
+		}
+		/* a constant was added to pkt_ptr.
+		 * Remember it while keeping the same 'id'
+		 */
+		dst_reg->off += imm;
+	} else {
+		if (src_reg->type == CONST_IMM) {
+			/* pkt_ptr += reg where reg is known constant */
+			imm = src_reg->imm;
+			goto add_imm;
+		}
+		/* disallow pkt_ptr += reg
+		 * if reg is not uknown_value with guaranteed zero upper bits
+		 * otherwise pkt_ptr may overflow and addition will become
+		 * subtraction which is not allowed
+		 */
+		if (src_reg->type != UNKNOWN_VALUE) {
+			verbose("cannot add '%s' to ptr_to_packet\n",
+				reg_type_str[src_reg->type]);
+			return -EACCES;
+		}
+		if (src_reg->imm < 48) {
+			verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n",
+				src_reg->imm);
+			return -EACCES;
+		}
+		/* dst_reg stays as pkt_ptr type and since some positive
+		 * integer value was added to the pointer, increment its 'id'
+		 */
+		dst_reg->id++;
+
+		/* something was added to pkt_ptr, set range and off to zero */
+		dst_reg->off = 0;
+		dst_reg->range = 0;
+	}
+	return 0;
+}
+
+static int evaluate_reg_alu(struct verifier_env *env, struct bpf_insn *insn)
+{
+	struct reg_state *regs = env->cur_state.regs;
+	struct reg_state *dst_reg = &regs[insn->dst_reg];
+	u8 opcode = BPF_OP(insn->code);
+	s64 imm_log2;
+
+	/* for type == UNKNOWN_VALUE:
+	 * imm > 0 -> number of zero upper bits
+	 * imm == 0 -> don't track which is the same as all bits can be non-zero
+	 */
+
+	if (BPF_SRC(insn->code) == BPF_X) {
+		struct reg_state *src_reg = &regs[insn->src_reg];
+
+		if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 &&
+		    dst_reg->imm && opcode == BPF_ADD) {
+			/* dreg += sreg
+			 * where both have zero upper bits. Adding them
+			 * can only result making one more bit non-zero
+			 * in the larger value.
+			 * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47)
+			 *     0xffff (imm=48) + 0xffff = 0x1fffe (imm=47)
+			 */
+			dst_reg->imm = min(dst_reg->imm, src_reg->imm);
+			dst_reg->imm--;
+			return 0;
+		}
+		if (src_reg->type == CONST_IMM && src_reg->imm > 0 &&
+		    dst_reg->imm && opcode == BPF_ADD) {
+			/* dreg += sreg
+			 * where dreg has zero upper bits and sreg is const.
+			 * Adding them can only result making one more bit
+			 * non-zero in the larger value.
+			 */
+			imm_log2 = __ilog2_u64((long long)src_reg->imm);
+			dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
+			dst_reg->imm--;
+			return 0;
+		}
+		/* all other cases non supported yet, just mark dst_reg */
+		dst_reg->imm = 0;
+		return 0;
+	}
+
+	/* sign extend 32-bit imm into 64-bit to make sure that
+	 * negative values occupy bit 63. Note ilog2() would have
+	 * been incorrect, since sizeof(insn->imm) == 4
+	 */
+	imm_log2 = __ilog2_u64((long long)insn->imm);
+
+	if (dst_reg->imm && opcode == BPF_LSH) {
+		/* reg <<= imm
+		 * if reg was a result of 2 byte load, then its imm == 48
+		 * which means that upper 48 bits are zero and shifting this reg
+		 * left by 4 would mean that upper 44 bits are still zero
+		 */
+		dst_reg->imm -= insn->imm;
+	} else if (dst_reg->imm && opcode == BPF_MUL) {
+		/* reg *= imm
+		 * if multiplying by 14 subtract 4
+		 * This is conservative calculation of upper zero bits.
+		 * It's not trying to special case insn->imm == 1 or 0 cases
+		 */
+		dst_reg->imm -= imm_log2 + 1;
+	} else if (opcode == BPF_AND) {
+		/* reg &= imm */
+		dst_reg->imm = 63 - imm_log2;
+	} else if (dst_reg->imm && opcode == BPF_ADD) {
+		/* reg += imm */
+		dst_reg->imm = min(dst_reg->imm, 63 - imm_log2);
+		dst_reg->imm--;
+	} else if (opcode == BPF_RSH) {
+		/* reg >>= imm
+		 * which means that after right shift, upper bits will be zero
+		 * note that verifier already checked that
+		 * 0 <= imm < 64 for shift insn
+		 */
+		dst_reg->imm += insn->imm;
+		if (unlikely(dst_reg->imm > 64))
+			/* some dumb code did:
+			 * r2 = *(u32 *)mem;
+			 * r2 >>= 32;
+			 * and all bits are zero now */
+			dst_reg->imm = 64;
+	} else {
+		/* all other alu ops, means that we don't know what will
+		 * happen to the value, mark it with unknown number of zero bits
+		 */
+		dst_reg->imm = 0;
+	}
+
+	if (dst_reg->imm < 0) {
+		/* all 64 bits of the register can contain non-zero bits
+		 * and such value cannot be added to ptr_to_packet, since it
+		 * may overflow, mark it as unknown to avoid further eval
+		 */
+		dst_reg->imm = 0;
+	}
+	return 0;
+}
+
+static int evaluate_reg_imm_alu(struct verifier_env *env, struct bpf_insn *insn)
+{
+	struct reg_state *regs = env->cur_state.regs;
+	struct reg_state *dst_reg = &regs[insn->dst_reg];
+	struct reg_state *src_reg = &regs[insn->src_reg];
+	u8 opcode = BPF_OP(insn->code);
+
+	/* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn.
+	 * Don't care about overflow or negative values, just add them
+	 */
+	if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K)
+		dst_reg->imm += insn->imm;
+	else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X &&
+		 src_reg->type == CONST_IMM)
+		dst_reg->imm += src_reg->imm;
+	else
+		mark_reg_unknown_value(regs, insn->dst_reg);
 	return 0;
 }
 
@@ -1245,6 +1567,21 @@ static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
 			dst_reg->type = PTR_TO_STACK;
 			dst_reg->imm = insn->imm;
 			return 0;
+		} else if (opcode == BPF_ADD &&
+			   BPF_CLASS(insn->code) == BPF_ALU64 &&
+			   dst_reg->type == PTR_TO_PACKET) {
+			/* ptr_to_packet += K|X */
+			return check_packet_ptr_add(env, insn);
+		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
+			   dst_reg->type == UNKNOWN_VALUE &&
+			   env->allow_ptr_leaks) {
+			/* unknown += K|X */
+			return evaluate_reg_alu(env, insn);
+		} else if (BPF_CLASS(insn->code) == BPF_ALU64 &&
+			   dst_reg->type == CONST_IMM &&
+			   env->allow_ptr_leaks) {
+			/* reg_imm += K|X */
+			return evaluate_reg_imm_alu(env, insn);
 		} else if (is_pointer_value(env, insn->dst_reg)) {
 			verbose("R%d pointer arithmetic prohibited\n",
 				insn->dst_reg);
@@ -1263,6 +1600,34 @@ static int check_alu_op(struct verifier_env *env, struct bpf_insn *insn)
 	return 0;
 }
 
+static void find_good_pkt_pointers(struct verifier_env *env,
+				   struct reg_state *dst_reg)
+{
+	struct verifier_state *state = &env->cur_state;
+	struct reg_state *regs = state->regs, *reg;
+	int i;
+	/* r2 = r3;
+	 * r2 += 8
+	 * if (r2 > pkt_end) goto somewhere
+	 * r2 == dst_reg, pkt_end == src_reg,
+	 * r2=pkt(id=n,off=8,r=0)
+	 * r3=pkt(id=n,off=0,r=0)
+	 * find register r3 and mark its range as r3=pkt(id=n,off=0,r=8)
+	 * so that range of bytes [r3, r3 + 8) is safe to access
+	 */
+	for (i = 0; i < MAX_BPF_REG; i++)
+		if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id)
+			regs[i].range = dst_reg->off;
+
+	for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) {
+		if (state->stack_slot_type[i] != STACK_SPILL)
+			continue;
+		reg = &state->spilled_regs[i / BPF_REG_SIZE];
+		if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id)
+			reg->range = dst_reg->off;
+	}
+}
+
 static int check_cond_jmp_op(struct verifier_env *env,
 			     struct bpf_insn *insn, int *insn_idx)
 {
@@ -1346,6 +1711,10 @@ static int check_cond_jmp_op(struct verifier_env *env,
 			regs[insn->dst_reg].type = CONST_IMM;
 			regs[insn->dst_reg].imm = 0;
 		}
+	} else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT &&
+		   dst_reg->type == PTR_TO_PACKET &&
+		   regs[insn->src_reg].type == PTR_TO_PACKET_END) {
+		find_good_pkt_pointers(env, dst_reg);
 	} else if (is_pointer_value(env, insn->dst_reg)) {
 		verbose("R%d pointer comparison prohibited\n", insn->dst_reg);
 		return -EACCES;
@@ -1685,6 +2054,58 @@ static int check_cfg(struct verifier_env *env)
 	return ret;
 }
 
+/* the following conditions reduce the number of explored insns
+ * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet
+ */
+static bool compare_ptrs_to_packet(struct reg_state *old, struct reg_state *cur)
+{
+	if (old->id != cur->id)
+		return false;
+
+	/* old ptr_to_packet is more conservative, since it allows smaller
+	 * range. Ex:
+	 * old(off=0,r=10) is equal to cur(off=0,r=20), because
+	 * old(off=0,r=10) means that with range=10 the verifier proceeded
+	 * further and found no issues with the program. Now we're in the same
+	 * spot with cur(off=0,r=20), so we're safe too, since anything further
+	 * will only be looking at most 10 bytes after this pointer.
+	 */
+	if (old->off == cur->off && old->range < cur->range)
+		return true;
+
+	/* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0)
+	 * since both cannot be used for packet access and safe(old)
+	 * pointer has smaller off that could be used for further
+	 * 'if (ptr > data_end)' check
+	 * Ex:
+	 * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean
+	 * that we cannot access the packet.
+	 * The safe range is:
+	 * [ptr, ptr + range - off)
+	 * so whenever off >=range, it means no safe bytes from this pointer.
+	 * When comparing old->off <= cur->off, it means that older code
+	 * went with smaller offset and that offset was later
+	 * used to figure out the safe range after 'if (ptr > data_end)' check
+	 * Say, 'old' state was explored like:
+	 * ... R3(off=0, r=0)
+	 * R4 = R3 + 20
+	 * ... now R4(off=20,r=0)  <-- here
+	 * if (R4 > data_end)
+	 * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access.
+	 * ... the code further went all the way to bpf_exit.
+	 * Now the 'cur' state at the mark 'here' has R4(off=30,r=0).
+	 * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier
+	 * goes further, such cur_R4 will give larger safe packet range after
+	 * 'if (R4 > data_end)' and all further insn were already good with r=20,
+	 * so they will be good with r=30 and we can prune the search.
+	 */
+	if (old->off <= cur->off &&
+	    old->off >= old->range && cur->off >= cur->range)
+		return true;
+
+	return false;
+}
+
 /* compare two verifier states
  *
  * all states stored in state_list are known to be valid, since
@@ -1727,6 +2148,10 @@ static bool states_equal(struct verifier_state *old, struct verifier_state *cur)
 		    (rold->type == UNKNOWN_VALUE && rcur->type != NOT_INIT))
 			continue;
 
+		if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET &&
+		    compare_ptrs_to_packet(rold, rcur))
+			continue;
+
 		return false;
 	}