diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 982213d97668d..1bec48d9e5d90 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -2191,6 +2191,10 @@ bool btf_nested_type_is_trusted(struct bpf_verifier_log *log,
 				const struct bpf_reg_state *reg,
 				int off);
 
+bool btf_type_ids_nocast_alias(struct bpf_verifier_log *log,
+			       const struct btf *reg_btf, u32 reg_id,
+			       const struct btf *arg_btf, u32 arg_id);
+
 int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo,
 		   int relo_idx, void *insn);
 
diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index dd05b5f2c1d89..47b8cb96f2c22 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -336,6 +336,12 @@ const char *btf_type_str(const struct btf_type *t)
 /* Type name size */
 #define BTF_SHOW_NAME_SIZE		80
 
+/*
+ * The suffix of a type that indicates it cannot alias another type when
+ * comparing BTF IDs for kfunc invocations.
+ */
+#define NOCAST_ALIAS_SUFFIX		"___init"
+
 /*
  * Common data to all BTF show operations. Private show functions can add
  * their own data to a structure containing a struct btf_show and consult it
@@ -8288,3 +8294,58 @@ bool btf_nested_type_is_trusted(struct bpf_verifier_log *log,
 
 	return false;
 }
+
+bool btf_type_ids_nocast_alias(struct bpf_verifier_log *log,
+			       const struct btf *reg_btf, u32 reg_id,
+			       const struct btf *arg_btf, u32 arg_id)
+{
+	const char *reg_name, *arg_name, *search_needle;
+	const struct btf_type *reg_type, *arg_type;
+	int reg_len, arg_len, cmp_len;
+	size_t pattern_len = sizeof(NOCAST_ALIAS_SUFFIX) - sizeof(char);
+
+	reg_type = btf_type_by_id(reg_btf, reg_id);
+	if (!reg_type)
+		return false;
+
+	arg_type = btf_type_by_id(arg_btf, arg_id);
+	if (!arg_type)
+		return false;
+
+	reg_name = btf_name_by_offset(reg_btf, reg_type->name_off);
+	arg_name = btf_name_by_offset(arg_btf, arg_type->name_off);
+
+	reg_len = strlen(reg_name);
+	arg_len = strlen(arg_name);
+
+	/* Exactly one of the two type names may be suffixed with ___init, so
+	 * if the strings are the same size, they can't possibly be no-cast
+	 * aliases of one another. If you have two of the same type names, e.g.
+	 * they're both nf_conn___init, it would be improper to return true
+	 * because they are _not_ no-cast aliases, they are the same type.
+	 */
+	if (reg_len == arg_len)
+		return false;
+
+	/* Either of the two names must be the other name, suffixed with ___init. */
+	if ((reg_len != arg_len + pattern_len) &&
+	    (arg_len != reg_len + pattern_len))
+		return false;
+
+	if (reg_len < arg_len) {
+		search_needle = strstr(arg_name, NOCAST_ALIAS_SUFFIX);
+		cmp_len = reg_len;
+	} else {
+		search_needle = strstr(reg_name, NOCAST_ALIAS_SUFFIX);
+		cmp_len = arg_len;
+	}
+
+	if (!search_needle)
+		return false;
+
+	/* ___init suffix must come at the end of the name */
+	if (*(search_needle + pattern_len) != '\0')
+		return false;
+
+	return !strncmp(reg_name, arg_name, cmp_len);
+}
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index bc24bdadc1931..66ec577fcb8bd 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -8820,9 +8820,37 @@ static int process_kf_arg_ptr_to_btf_id(struct bpf_verifier_env *env,
 		reg_ref_id = *reg2btf_ids[base_type(reg->type)];
 	}
 
-	if (is_kfunc_trusted_args(meta) || (is_kfunc_release(meta) && reg->ref_obj_id))
+	/* Enforce strict type matching for calls to kfuncs that are acquiring
+	 * or releasing a reference, or are no-cast aliases. We do _not_
+	 * enforce strict matching for plain KF_TRUSTED_ARGS kfuncs by default,
+	 * as we want to enable BPF programs to pass types that are bitwise
+	 * equivalent without forcing them to explicitly cast with something
+	 * like bpf_cast_to_kern_ctx().
+	 *
+	 * For example, say we had a type like the following:
+	 *
+	 * struct bpf_cpumask {
+	 *	cpumask_t cpumask;
+	 *	refcount_t usage;
+	 * };
+	 *
+	 * Note that as specified in <linux/cpumask.h>, cpumask_t is typedef'ed
+	 * to a struct cpumask, so it would be safe to pass a struct
+	 * bpf_cpumask * to a kfunc expecting a struct cpumask *.
+	 *
+	 * The philosophy here is similar to how we allow scalars of different
+	 * types to be passed to kfuncs as long as the size is the same. The
+	 * only difference here is that we're simply allowing
+	 * btf_struct_ids_match() to walk the struct at the 0th offset, and
+	 * resolve types.
+	 */
+	if (is_kfunc_acquire(meta) ||
+	    (is_kfunc_release(meta) && reg->ref_obj_id) ||
+	    btf_type_ids_nocast_alias(&env->log, reg_btf, reg_ref_id, meta->btf, ref_id))
 		strict_type_match = true;
 
+	WARN_ON_ONCE(is_kfunc_trusted_args(meta) && reg->off);
+
 	reg_ref_t = btf_type_skip_modifiers(reg_btf, reg_ref_id, &reg_ref_id);
 	reg_ref_tname = btf_name_by_offset(reg_btf, reg_ref_t->name_off);
 	if (!btf_struct_ids_match(&env->log, reg_btf, reg_ref_id, reg->off, meta->btf, ref_id, strict_type_match)) {