diff --git a/pythonbpf/allocation_pass.py b/pythonbpf/allocation_pass.py
index 10ff7b2f..e332c848 100644
--- a/pythonbpf/allocation_pass.py
+++ b/pythonbpf/allocation_pass.py
@@ -6,7 +6,15 @@
 from pythonbpf.helper import HelperHandlerRegistry
 from pythonbpf.vmlinux_parser.dependency_node import Field
 from .expr import VmlinuxHandlerRegistry
-from pythonbpf.type_deducer import ctypes_to_ir, is_ctypes, IntTy, signedness, byte_size
+from pythonbpf.type_deducer import (
+    ctypes_to_ir,
+    is_ctypes,
+    IntTy,
+    signedness,
+    byte_size,
+    PktPtrTy,
+)
+from pythonbpf.expr.packet_pointer import packet_type
 from pythonbpf.expr.type_inference import infer_int_type
 from pythonbpf.maps import BPFMapType
 
@@ -301,7 +309,10 @@ def _allocate_for_binop(builder, var_name, rval, local_sym_tab, compilation_cont
     inferred = infer_int_type(rval, local_sym_tab, compilation_context)
     if inferred is None:
         logger.debug(f"Could not infer a type for {var_name}, assuming signed i64")
-    ir_type = IntTy(64, signedness(inferred) if inferred is not None else True)
+    if isinstance(inferred, PktPtrTy):
+        ir_type = inferred  # data + 14 is still a packet pointer
+    else:
+        ir_type = IntTy(64, signedness(inferred) if inferred is not None else True)
     var = builder.alloca(ir_type, name=var_name)
     var.align = 8
     local_sym_tab[var_name] = LocalSymbol(var, ir_type)
@@ -401,6 +412,13 @@ def _allocate_for_attribute(
             # Same discriminator handle_vmlinux_struct_field uses: a context
             # argument has no alloca of its own.
             is_context_field = local_sym_tab[struct_var].var is None
+            pkt_ty = packet_type(rval, local_sym_tab)
+            if pkt_ty is not None:
+                # A packet-pointer field: a 64-bit slot that carries the kind.
+                var = _allocate_with_type(builder, var_name, pkt_ty)
+                local_sym_tab[var_name] = LocalSymbol(var, pkt_ty)
+                logger.info(f"Pre-allocated {var_name} as {pkt_ty.describe()}")
+                return
             if not VmlinuxHandlerRegistry.has_field(vmlinux_struct_name, field_name):
                 logger.error(
                     f"Field '{field_name}' not found in vmlinux struct '{vmlinux_struct_name}'"
diff --git a/pythonbpf/expr/expr_pass.py b/pythonbpf/expr/expr_pass.py
index b9984b83..939e23e3 100644
--- a/pythonbpf/expr/expr_pass.py
+++ b/pythonbpf/expr/expr_pass.py
@@ -9,6 +9,7 @@
     field_int_type,
     is_ctypes,
     IntTy,
+    PktPtrTy,
     int_literal_type,
     signedness,
 )
@@ -24,6 +25,7 @@
     get_base_type_and_depth,
 )
 from .vmlinux_registry import VmlinuxHandlerRegistry
+from .packet_pointer import MAX_PACKET_OFF, packet_type
 from ..vmlinux_parser.dependency_node import Field
 
 logger: Logger = logging.getLogger(__name__)
@@ -122,6 +124,11 @@ def _handle_attribute_expr(
                     expr, local_sym_tab, None, builder
                 )
                 if vmlinux_result is not None:
+                    pkt_ty = packet_type(expr, local_sym_tab)
+                    if pkt_ty is not None:
+                        # A packet-pointer field: the value is the loaded
+                        # pointer, and the descriptor says so from here on.
+                        return vmlinux_result[0], pkt_ty
                     return vmlinux_result
                 else:
                     raise RuntimeError("Vmlinux struct did not process successfully")
@@ -199,6 +206,8 @@ def _descriptor(val, ty):
     value unless the descriptor is itself an integer type, sign from the
     descriptor (an IntTy, a vmlinux Field, or plain -> signed). None when the
     value is not an integer at all."""
+    if isinstance(ty, PktPtrTy):
+        return ty  # a packet pointer keeps its kind, never re-ranked
     if isinstance(ty, ir.IntType):
         return IntTy(ty.width, signedness(ty))
     field = field_int_type(ty)
@@ -291,6 +300,10 @@ def _handle_binary_op_impl(func, compilation_context, rval, builder, local_sym_t
     right, right_ty = get_typed_operand(
         func, compilation_context, rval.right, builder, local_sym_tab
     )
+    if isinstance(left_ty, PktPtrTy) or isinstance(right_ty, PktPtrTy):
+        # Before the usual arithmetic conversions, which would re-rank the
+        # pointer as an ordinary integer and drop its kind.
+        return _packet_pointer_binop(builder, rval, left, left_ty, right, right_ty)
     result_ty = usual_arithmetic_conversions(left_ty, right_ty)
     logger.info(
         f"binop {type(op).__name__}: {left_ty.describe()} x {right_ty.describe()} "
@@ -302,6 +315,50 @@ def _handle_binary_op_impl(func, compilation_context, rval, builder, local_sym_t
     return canonicalise(builder, result, result_ty), result_ty
 
 
+def _packet_pointer_binop(builder, rval, left, left_ty, right, right_ty):
+    """A binary operation with a packet-pointer operand (see packet_pointer.py):
+    64-bit, never ranked as an ordinary integer, with the offset taken as an
+    unsigned 16-bit value so the verifier can track the packet range. The
+    result of pointer +/- offset is a pointer of the same kind; pointer -
+    pointer is a plain 64-bit length."""
+    op = rval.op
+    where = f"line {rval.lineno}: {ast.unparse(rval)}"
+    left_pkt = isinstance(left_ty, PktPtrTy)
+    right_pkt = isinstance(right_ty, PktPtrTy)
+    if left_pkt and right_pkt:
+        if isinstance(op, ast.Sub):
+            # data_end - data: a length, an ordinary 64-bit scalar.
+            return builder.sub(left, right), IntTy(64, False)
+        raise SyntaxError(
+            f"only subtraction is defined between packet pointers ({where})"
+        )
+    if not isinstance(op, (ast.Add, ast.Sub)):
+        raise SyntaxError(
+            f"only + and - with an offset are allowed on a packet pointer ({where})"
+        )
+    if right_pkt and isinstance(op, ast.Sub):
+        raise SyntaxError(f"cannot subtract a packet pointer from an offset ({where})")
+
+    ptr, ptr_ty, off = (left, left_ty, right) if left_pkt else (right, right_ty, left)
+    if ptr_ty.kind == "pkt_end":
+        # The verifier only compares against the end of the packet.
+        raise SyntaxError(
+            f"a packet-end pointer can only be compared, not offset ({where})"
+        )
+    if isinstance(off, ir.Constant) and isinstance(off.constant, int):
+        if not 0 <= off.constant <= MAX_PACKET_OFF:
+            raise SyntaxError(
+                f"packet offset {off.constant} is outside 0..{MAX_PACKET_OFF} "
+                f"({where}); subtract instead of adding a negative offset"
+            )
+    # The offset as u16, widened with zero-extension: a value the verifier can
+    # bound to [0, 0xffff].
+    off = builder.trunc(off, ir.IntType(16)) if off.type.width > 16 else off
+    off = builder.zext(off, ir.IntType(64)) if off.type.width < 64 else off
+    result = builder.add(ptr, off) if isinstance(op, ast.Add) else builder.sub(ptr, off)
+    return result, PktPtrTy(ptr_ty.kind)
+
+
 def _handle_binary_op(
     func,
     compilation_context,
@@ -411,6 +468,13 @@ def _handle_compare(func, compilation_context, builder, cond, local_sym_tab):
     lhs, lhs_ty = lhs
     rhs, rhs_ty = rhs
     lhs_desc, rhs_desc = _descriptor(lhs, lhs_ty), _descriptor(rhs, rhs_ty)
+    if isinstance(lhs_desc, PktPtrTy) or isinstance(rhs_desc, PktPtrTy):
+        # A packet bounds check (data + n > data_end): the verifier accepts
+        # pointer comparisons only at 64 bits, and they are unsigned.
+        u64 = IntTy(64, False)
+        lhs = convert(builder, lhs, lhs_desc, u64)
+        rhs = convert(builder, rhs, rhs_desc, u64)
+        return handle_comparator(func, builder, cond.ops[0], lhs, rhs, signed=False)
     if lhs_desc is not None and rhs_desc is not None:
         # Both integers: compare in the promoted type, which also picks the
         # signed or unsigned predicate (u64 > s64 is an unsigned compare in C).
diff --git a/pythonbpf/expr/packet_pointer.py b/pythonbpf/expr/packet_pointer.py
new file mode 100644
index 00000000..fe0fa2ff
--- /dev/null
+++ b/pythonbpf/expr/packet_pointer.py
@@ -0,0 +1,68 @@
+"""Packet-pointer context fields.
+
+A few context fields are declared u32 in C but are packet pointers to the
+verifier: it rewrites the context load into a 64-bit pointer load. C's
+integer rules (which rank them as u32) therefore produce code the verifier
+rejects ("32-bit pointer arithmetic prohibited"); C programs dodge that by
+casting through (void *)(long). The compiler special-cases them instead, so
+users never need the cast:
+
+- the field is used at 64 bits, as the pointer it is;
+- an offset added to or subtracted from it is taken as an unsigned 16-bit
+  value (truncated, then zero-extended), because the verifier only tracks a
+  packet range whose offset stays within MAX_PACKET_OFF (0xffff);
+- pointer - pointer (e.g. data_end - data) is an ordinary 64-bit length.
+
+The field read gives its value a PktPtrTy descriptor (a 64-bit unsigned
+IntTy tagged with the verifier's kind), and the rules key on that descriptor,
+so they follow the pointer through locals (`d = ctx.data`), through
+`d + 14`, and into comparisons (`ctx.data + 34 > ctx.data_end` is a 64-bit
+unsigned compare). packet-end pointers take no offset at all.
+
+Storing a packet pointer into a narrower slot truncates it, as C does, and the
+result is an ordinary integer: the verifier forbids 32-bit *arithmetic* on a
+packet pointer, not a 32-bit store of one (upstream's
+cgroup_skb_direct_packet_access stores data_end into a __u32 global). Keeping
+the pointer when a local is declared narrower is a job for allocation, which
+could give such a slot the widest type ever stored into it.
+"""
+
+import ast
+
+from pythonbpf.type_deducer import PktPtrTy
+
+# Context struct -> {field: packet kind}. Taken from the verifier's
+# is_valid_access callbacks, restricted to the fields vmlinux declares as
+# 32-bit integers.
+PACKET_POINTER_FIELDS = {
+    "struct_xdp_md": {"data": "pkt", "data_meta": "pkt_meta", "data_end": "pkt_end"},
+    "struct___sk_buff": {
+        "data": "pkt",
+        "data_meta": "pkt_meta",
+        "data_end": "pkt_end",
+    },
+}
+
+MAX_PACKET_OFF = 0xFFFF
+
+
+def packet_kind(expr, local_sym_tab) -> "str | None":
+    """The packet kind of `ctx.<field>` when ctx is a context parameter of one
+    of the structs above and the field is one of its packet-pointer fields;
+    None otherwise. This is the only place a PktPtrTy originates: from here it
+    travels as a descriptor, through locals, binop results and comparisons."""
+    if not (isinstance(expr, ast.Attribute) and isinstance(expr.value, ast.Name)):
+        return None
+    sym = (local_sym_tab or {}).get(expr.value.id)
+    if sym is None:
+        return None
+    meta = sym.metadata
+    struct_name = meta if isinstance(meta, str) else getattr(meta, "__name__", None)
+    if not isinstance(struct_name, str):
+        return None
+    return PACKET_POINTER_FIELDS.get(struct_name, {}).get(expr.attr)
+
+
+def packet_type(expr, local_sym_tab) -> "PktPtrTy | None":
+    kind = packet_kind(expr, local_sym_tab)
+    return PktPtrTy(kind) if kind is not None else None
diff --git a/pythonbpf/expr/type_inference.py b/pythonbpf/expr/type_inference.py
index 9a0c56b8..89dcf8df 100644
--- a/pythonbpf/expr/type_inference.py
+++ b/pythonbpf/expr/type_inference.py
@@ -13,6 +13,7 @@
 from llvmlite import ir
 
 from pythonbpf.type_deducer import (
+    PktPtrTy,
     IntTy,
     ctypes_to_ir,
     is_ctypes,
@@ -21,9 +22,12 @@
 )
 from .operators import usual_arithmetic_conversions
 from .vmlinux_registry import VmlinuxHandlerRegistry
+from .packet_pointer import packet_type
 
 
 def _as_intty(ty):
+    if isinstance(ty, PktPtrTy):
+        return ty
     if isinstance(ty, ir.IntType):
         return IntTy(ty.width, signedness(ty))
     return None
@@ -48,6 +52,10 @@ def infer_int_type(expr, local_sym_tab, compilation_context):
     if isinstance(expr, ast.BinOp):
         left = infer_int_type(expr.left, local_sym_tab, compilation_context)
         right = infer_int_type(expr.right, local_sym_tab, compilation_context)
+        if isinstance(left, PktPtrTy) and isinstance(right, PktPtrTy):
+            return IntTy(64, False)  # pointer - pointer: a length
+        if isinstance(left, PktPtrTy) or isinstance(right, PktPtrTy):
+            return left if isinstance(left, PktPtrTy) else right
         if left is None or right is None:
             return None
         return usual_arithmetic_conversions(left, right)
@@ -83,6 +91,9 @@ def infer_int_type(expr, local_sym_tab, compilation_context):
         return None
 
     if isinstance(expr, ast.Attribute) and isinstance(expr.value, ast.Name):
+        pkt_ty = packet_type(expr, local_sym_tab)
+        if pkt_ty is not None:
+            return pkt_ty
         base = local_sym_tab.get(expr.value.id)
         if base is None:
             return None
diff --git a/pythonbpf/type_deducer.py b/pythonbpf/type_deducer.py
index 1854b3da..d9d39ad9 100644
--- a/pythonbpf/type_deducer.py
+++ b/pythonbpf/type_deducer.py
@@ -41,6 +41,33 @@ def describe(self) -> str:
         return f"{'i' if self.signed else 'u'}{self.width}"
 
 
+PACKET_KINDS = ("pkt", "pkt_meta", "pkt_end")
+
+
+class PktPtrTy(IntTy):
+    """A packet pointer: 64-bit, unsigned, and tagged with the verifier's kind
+    for it (packet data, packet metadata, or packet end). It is an IntTy so
+    every integer path accepts it; the few sites that must treat it as a
+    pointer check isinstance(ty, PktPtrTy) before the integer rules apply.
+    See pythonbpf/expr/packet_pointer.py for where kinds come from and the
+    rules applied to them."""
+
+    def __new__(cls, kind: str):
+        return IntTy.__new__(cls, 64, False)
+
+    def __init__(self, kind: str):
+        if kind not in PACKET_KINDS:
+            raise ValueError(f"unknown packet pointer kind {kind!r}")
+        IntTy.__init__(self, 64, False)
+        self.kind = kind
+
+    def __getnewargs__(self):  # type: ignore[override]
+        return (self.kind,)
+
+    def describe(self) -> str:
+        return f"{self.kind}*"
+
+
 def byte_size(ty) -> int:
     """Bytes an integer type occupies in memory: its width rounded up to whole
     bytes, so a 1-bit bool still takes one byte, as C's _Bool does."""
diff --git a/tests/failing_tests/vmlinux/ctx_data_end_offset.py b/tests/failing_tests/vmlinux/ctx_data_end_offset.py
new file mode 100644
index 00000000..238fe577
--- /dev/null
+++ b/tests/failing_tests/vmlinux/ctx_data_end_offset.py
@@ -0,0 +1,21 @@
+# data_end can only be compared against; the verifier rejects arithmetic on
+# it, so the compiler refuses it.
+from ctypes import c_int64, c_uint32  # noqa: F401
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    e = ctx.data_end + 1
+    return c_int64(e)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/failing_tests/vmlinux/ctx_data_offset_too_big.py b/tests/failing_tests/vmlinux/ctx_data_offset_too_big.py
new file mode 100644
index 00000000..32dc4a36
--- /dev/null
+++ b/tests/failing_tests/vmlinux/ctx_data_offset_too_big.py
@@ -0,0 +1,21 @@
+# 70000 cannot be a packet offset: the verifier tracks packet ranges only
+# up to 0xffff. Rejected at compile time rather than silently wrapped.
+from ctypes import c_int64
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    d = ctx.data + 70000
+    return c_int64(d)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_bounds_check.py b/tests/passing_tests/vmlinux/ctx_bounds_check.py
new file mode 100644
index 00000000..4e91357e
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_bounds_check.py
@@ -0,0 +1,25 @@
+# The canonical XDP bounds check, through locals: the packet-pointer type
+# travels with data and data_end, so data + 34 > data_end is a 64-bit
+# unsigned comparison, which is what the verifier needs to prove the range.
+from ctypes import c_int64, c_uint32  # noqa: F401
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    data = ctx.data
+    data_end = ctx.data_end
+    if data + 34 > data_end:
+        return c_int64(1)
+    return c_int64(2)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_bounds_check_direct.py b/tests/passing_tests/vmlinux/ctx_bounds_check_direct.py
new file mode 100644
index 00000000..cac2cbd2
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_bounds_check_direct.py
@@ -0,0 +1,21 @@
+# The same bounds check written on the fields directly.
+from ctypes import c_int64, c_uint32  # noqa: F401
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    if ctx.data + 14 > ctx.data_end:
+        return c_int64(1)
+    return c_int64(2)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_data_arith.py b/tests/passing_tests/vmlinux/ctx_data_arith.py
new file mode 100644
index 00000000..3006a41a
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_data_arith.py
@@ -0,0 +1,26 @@
+# xdp_md.data is declared u32, but the verifier treats it as a packet pointer
+# and prohibits 32-bit arithmetic on it. C's integer rules would rank
+# ctx.data - k as a u32 subtraction (w0 += -1, rejected with "R0 32-bit
+# pointer arithmetic prohibited"); the compiler special-cases packet-pointer
+# fields instead (pythonbpf/expr/packet_pointer.py): 64-bit arithmetic, the
+# offset taken as u16, so this is r0 += -1 and no cast is needed.
+from ctypes import c_int32, c_int64
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    k = c_int32(1)
+    d = ctx.data - k
+    return c_int64(d)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_data_length.py b/tests/passing_tests/vmlinux/ctx_data_length.py
new file mode 100644
index 00000000..6d96ba79
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_data_length.py
@@ -0,0 +1,21 @@
+# Pointer minus pointer is a length: an ordinary 64-bit subtraction, with no
+# 16-bit clamp on either side.
+from ctypes import c_int64
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    n = ctx.data_end - ctx.data
+    return c_int64(n)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_data_meta_offset.py b/tests/passing_tests/vmlinux/ctx_data_meta_offset.py
new file mode 100644
index 00000000..05b7c272
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_data_meta_offset.py
@@ -0,0 +1,20 @@
+# data_meta is a packet-metadata pointer; offsets are allowed on it too.
+from ctypes import c_int64, c_uint32  # noqa: F401
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    m = ctx.data_meta + 4
+    return c_int64(m)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_data_narrow_store.py b/tests/passing_tests/vmlinux/ctx_data_narrow_store.py
new file mode 100644
index 00000000..ef7c6ae8
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_data_narrow_store.py
@@ -0,0 +1,25 @@
+# A packet pointer stored into a 32-bit local is truncated, as in C
+# (`__u32 x = ctx->data`), and is an ordinary integer from then on. The
+# verifier forbids 32-bit arithmetic on a packet pointer, not a narrow store
+# of one; upstream's cgroup_skb_direct_packet_access does the same with a
+# __u32 global.
+from ctypes import c_int64, c_uint32  # noqa: F401
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    x = c_uint32(0)
+    x = ctx.data
+    return c_int64(x)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_data_plus_one.py b/tests/passing_tests/vmlinux/ctx_data_plus_one.py
new file mode 100644
index 00000000..9faa235c
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_data_plus_one.py
@@ -0,0 +1,21 @@
+# Adding a constant to a packet pointer is 64-bit pointer arithmetic
+# (r0 += 1), not the u32 addition C's integer rules would give xdp_md.data.
+from ctypes import c_int64
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    d = ctx.data + 1
+    return c_int64(d)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_data_runtime_offset.py b/tests/passing_tests/vmlinux/ctx_data_runtime_offset.py
new file mode 100644
index 00000000..4adde4b2
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_data_runtime_offset.py
@@ -0,0 +1,24 @@
+# A runtime offset added to a packet pointer is taken as an unsigned 16-bit
+# value (masked to 0xffff, zero-extended), so the verifier can bound the
+# packet range; the addition itself is 64-bit.
+from ctypes import c_int64
+from pythonbpf import bpf, section, bpfglobal, compile
+from pythonbpf.helper import random
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    off = random()
+    d = ctx.data + off
+    return c_int64(d)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/passing_tests/vmlinux/ctx_data_via_local.py b/tests/passing_tests/vmlinux/ctx_data_via_local.py
new file mode 100644
index 00000000..1e2d36d7
--- /dev/null
+++ b/tests/passing_tests/vmlinux/ctx_data_via_local.py
@@ -0,0 +1,22 @@
+# A local copied from a packet-pointer field is still a packet pointer: its
+# slot is 64-bit and d + 14 is 64-bit pointer arithmetic.
+from ctypes import c_int64, c_uint32  # noqa: F401
+from pythonbpf import bpf, section, bpfglobal, compile
+from vmlinux import struct_xdp_md
+
+
+@bpf
+@section("xdp")
+def prog(ctx: struct_xdp_md) -> c_int64:
+    data = ctx.data
+    eth = data + 14
+    return c_int64(eth)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()
diff --git a/tests/test_config.toml b/tests/test_config.toml
index ada07a15..eb680d65 100644
--- a/tests/test_config.toml
+++ b/tests/test_config.toml
@@ -64,6 +64,11 @@
 "failing_tests/loops/for_map_items.py" = {reason = "for/while loops not implemented: no sugar over bpf_for_each_map_elem()-style map iteration exists yet", level = "ir"}
 
 
+# A packet offset has to fit the verifier's 16-bit packet range.
+"failing_tests/vmlinux/ctx_data_offset_too_big.py" = {reason = "A constant packet offset outside 0..65535 is a compile error (the verifier tracks packet ranges up to MAX_PACKET_OFF)", level = "ir"}
+"failing_tests/vmlinux/ctx_data_end_offset.py" = {reason = "A packet-end pointer can only be compared, not offset (the verifier rejects arithmetic on it)", level = "ir"}
+
+
 "kernel_selftest_equivalent/ringbuf/reserve_submit_discard.py" = {reason = "RingBuffer reserve/typed record/discard workflow is planned but not implemented yet", level = "ir"}
 
 "kernel_selftest_equivalent/vmlinux/perf_skip.py" = {reason = "Nested struct field access (ctx.regs.ip) not supported; one level such as ctx.sample_period works", level = "ir"}
diff --git a/tests/test_signedness_ir.py b/tests/test_signedness_ir.py
index 96b003e4..666880c2 100644
--- a/tests/test_signedness_ir.py
+++ b/tests/test_signedness_ir.py
@@ -68,6 +68,38 @@
         [r"\bsub i64", r"trunc i64 .* to i32", r"zext i32 .* to i64"],
         [],
     ),
+    # Packet-pointer fields: 64-bit arithmetic, never the field's u32 rank,
+    # offsets as u16 (trunc to i16, zext); pointer - pointer is a plain sub.
+    # No computed value is cut to 32 bits (constant stores such as
+    # k = c_int32(1) legitimately are).
+    "vmlinux/ctx_data_arith.py": (
+        [r"\bsub i64"],
+        [r"trunc i64 %.* to i32"],
+    ),
+    "vmlinux/ctx_data_plus_one.py": (
+        [r"\badd i64"],
+        [r"trunc i64 %.* to i32"],
+    ),
+    "vmlinux/ctx_data_runtime_offset.py": (
+        [r"trunc i64 .* to i16", r"zext i16 .* to i64", r"\badd i64"],
+        [r"trunc i64 %.* to i32"],
+    ),
+    "vmlinux/ctx_data_length.py": (
+        [r"\bsub i64"],
+        [r"to i16", r"trunc i64 %.* to i32"],
+    ),
+    "vmlinux/ctx_bounds_check.py": (
+        [r"icmp ugt i64", r"\badd i64"],
+        [r"trunc i64 %.* to i32", r"icmp \w+ i32"],
+    ),
+    "vmlinux/ctx_bounds_check_direct.py": (
+        [r"icmp ugt i64"],
+        [r"trunc i64 %.* to i32", r"icmp \w+ i32"],
+    ),
+    "vmlinux/ctx_data_via_local.py": (
+        [r"\badd i64"],
+        [r"trunc i64 %.* to i32"],
+    ),
     # c_uint16(0) declares a 16-bit slot, so a u32 ctx field stored into it
     # is truncated to 16 bits (C: __u16 queue = ctx->rx_queue_index).
     "vmlinux/ctx_field_narrow_store.py": (