bpf: Refactor RCU enforcement in the verifier.

bpf_rcu_read_lock/unlock() are only available in clang compiled kernels. Lack
of such key mechanism makes it impossible for sleepable bpf programs to use RCU
pointers.

Allow bpf_rcu_read_lock/unlock() in GCC compiled kernels (though GCC doesn't
support btf_type_tag yet) and allowlist certain field dereferences in important
data structures like tast_struct, cgroup, socket that are used by sleepable
programs either as RCU pointer or full trusted pointer (which is valid outside
of RCU CS). Use BTF_TYPE_SAFE_RCU and BTF_TYPE_SAFE_TRUSTED macros for such
tagging. They will be removed once GCC supports btf_type_tag.

With that refactor check_ptr_to_btf_access(). Make it strict in enforcing
PTR_TRUSTED and PTR_UNTRUSTED while deprecating old PTR_TO_BTF_ID without
modifier flags. There is a chance that this strict enforcement might break
existing programs (especially on GCC compiled kernels), but this cleanup has to
start sooner than later. Note PTR_TO_CTX access still yields old deprecated
PTR_TO_BTF_ID. Once it's converted to strict PTR_TRUSTED or PTR_UNTRUSTED the
kfuncs and helpers will be able to default to KF_TRUSTED_ARGS. KF_RCU will
remain as a weaker version of KF_TRUSTED_ARGS where obj refcnt could be 0.

Adjust rcu_read_lock selftest to run on gcc and clang compiled kernels.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/bpf/20230303041446.3630-7-alexei.starovoitov@gmail.com

3 files changed, 154 insertions, 80 deletions

diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index c5e1d6955491..a8cb09e5973b 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -6163,6 +6163,7 @@ static int btf_struct_walk(struct bpf_verifier_log *log, const struct btf *btf,
 	const char *tname, *mname, *tag_value;
 	u32 vlen, elem_id, mid;
 
+	*flag = 0;
 again:
 	tname = __btf_name_by_offset(btf, t->name_off);
 	if (!btf_type_is_struct(t)) {
@@ -6329,6 +6330,15 @@ error:
 		 * of this field or inside of this struct
 		 */
 		if (btf_type_is_struct(mtype)) {
+			if (BTF_INFO_KIND(mtype->info) == BTF_KIND_UNION &&
+			    btf_type_vlen(mtype) != 1)
+				/*
+				 * walking unions yields untrusted pointers
+				 * with exception of __bpf_md_ptr and other
+				 * unions with a single member
+				 */
+				*flag |= PTR_UNTRUSTED;
+
 			/* our field must be inside that union or struct */
 			t = mtype;
 
@@ -6373,7 +6383,7 @@ error:
 			stype = btf_type_skip_modifiers(btf, mtype->type, &id);
 			if (btf_type_is_struct(stype)) {
 				*next_btf_id = id;
-				*flag = tmp_flag;
+				*flag |= tmp_flag;
 				return WALK_PTR;
 			}
 		}
@@ -8357,7 +8367,7 @@ out:
 
 bool btf_nested_type_is_trusted(struct bpf_verifier_log *log,
 				const struct bpf_reg_state *reg,
-				int off)
+				int off, const char *suffix)
 {
 	struct btf *btf = reg->btf;
 	const struct btf_type *walk_type, *safe_type;
@@ -8374,7 +8384,7 @@ bool btf_nested_type_is_trusted(struct bpf_verifier_log *log,
 
 	tname = btf_name_by_offset(btf, walk_type->name_off);
 
-	ret = snprintf(safe_tname, sizeof(safe_tname), "%s__safe_fields", tname);
+	ret = snprintf(safe_tname, sizeof(safe_tname), "%s%s", tname, suffix);
 	if (ret < 0)
 		return false;
 
diff --git a/kernel/bpf/cpumask.c b/kernel/bpf/cpumask.c
index 2b3fbbfebdc5..b6587ec40f1b 100644
--- a/kernel/bpf/cpumask.c
+++ b/kernel/bpf/cpumask.c
@@ -427,26 +427,26 @@ BTF_ID_FLAGS(func, bpf_cpumask_create, KF_ACQUIRE | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_cpumask_release, KF_RELEASE | KF_TRUSTED_ARGS)
 BTF_ID_FLAGS(func, bpf_cpumask_acquire, KF_ACQUIRE | KF_TRUSTED_ARGS)
 BTF_ID_FLAGS(func, bpf_cpumask_kptr_get, KF_ACQUIRE | KF_KPTR_GET | KF_RET_NULL)
-BTF_ID_FLAGS(func, bpf_cpumask_first, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_first_zero, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_set_cpu, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_clear_cpu, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_test_cpu, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_test_and_set_cpu, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_test_and_clear_cpu, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_setall, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_clear, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_and, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_or, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_xor, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_equal, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_intersects, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_subset, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_empty, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_full, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_copy, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_any, KF_TRUSTED_ARGS)
-BTF_ID_FLAGS(func, bpf_cpumask_any_and, KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, bpf_cpumask_first, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_first_zero, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_set_cpu, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_clear_cpu, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_test_cpu, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_test_and_set_cpu, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_test_and_clear_cpu, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_setall, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_clear, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_and, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_or, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_xor, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_equal, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_intersects, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_subset, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_empty, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_full, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_copy, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_any, KF_RCU)
+BTF_ID_FLAGS(func, bpf_cpumask_any_and, KF_RCU)
 BTF_SET8_END(cpumask_kfunc_btf_ids)
 
 static const struct btf_kfunc_id_set cpumask_kfunc_set = {
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index a095055d7ef4..c2adf3c24c64 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -5073,29 +5073,76 @@ static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val)
 	return 0;
 }
 
-#define BTF_TYPE_SAFE_NESTED(__type)  __PASTE(__type, __safe_fields)
+#define BTF_TYPE_SAFE_RCU(__type)  __PASTE(__type, __safe_rcu)
+#define BTF_TYPE_SAFE_TRUSTED(__type)  __PASTE(__type, __safe_trusted)
 
-BTF_TYPE_SAFE_NESTED(struct task_struct) {
+/*
+ * Allow list few fields as RCU trusted or full trusted.
+ * This logic doesn't allow mix tagging and will be removed once GCC supports
+ * btf_type_tag.
+ */
+
+/* RCU trusted: these fields are trusted in RCU CS and never NULL */
+BTF_TYPE_SAFE_RCU(struct task_struct) {
 	const cpumask_t *cpus_ptr;
 	struct css_set __rcu *cgroups;
+	struct task_struct __rcu *real_parent;
+	struct task_struct *group_leader;
 };
 
-BTF_TYPE_SAFE_NESTED(struct css_set) {
+BTF_TYPE_SAFE_RCU(struct css_set) {
 	struct cgroup *dfl_cgrp;
 };
 
-static bool nested_ptr_is_trusted(struct bpf_verifier_env *env,
-				  struct bpf_reg_state *reg,
-				  int off)
+/* full trusted: these fields are trusted even outside of RCU CS and never NULL */
+BTF_TYPE_SAFE_TRUSTED(struct bpf_iter_meta) {
+	__bpf_md_ptr(struct seq_file *, seq);
+};
+
+BTF_TYPE_SAFE_TRUSTED(struct bpf_iter__task) {
+	__bpf_md_ptr(struct bpf_iter_meta *, meta);
+	__bpf_md_ptr(struct task_struct *, task);
+};
+
+BTF_TYPE_SAFE_TRUSTED(struct linux_binprm) {
+	struct file *file;
+};
+
+BTF_TYPE_SAFE_TRUSTED(struct file) {
+	struct inode *f_inode;
+};
+
+BTF_TYPE_SAFE_TRUSTED(struct dentry) {
+	/* no negative dentry-s in places where bpf can see it */
+	struct inode *d_inode;
+};
+
+BTF_TYPE_SAFE_TRUSTED(struct socket) {
+	struct sock *sk;
+};
+
+static bool type_is_rcu(struct bpf_verifier_env *env,
+			struct bpf_reg_state *reg,
+			int off)
 {
-	/* If its parent is not trusted, it can't regain its trusted status. */
-	if (!is_trusted_reg(reg))
-		return false;
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct task_struct));
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct css_set));
 
-	BTF_TYPE_EMIT(BTF_TYPE_SAFE_NESTED(struct task_struct));
-	BTF_TYPE_EMIT(BTF_TYPE_SAFE_NESTED(struct css_set));
+	return btf_nested_type_is_trusted(&env->log, reg, off, "__safe_rcu");
+}
 
-	return btf_nested_type_is_trusted(&env->log, reg, off);
+static bool type_is_trusted(struct bpf_verifier_env *env,
+			    struct bpf_reg_state *reg,
+			    int off)
+{
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct bpf_iter_meta));
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct bpf_iter__task));
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct linux_binprm));
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct file));
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct dentry));
+	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct socket));
+
+	return btf_nested_type_is_trusted(&env->log, reg, off, "__safe_trusted");
 }
 
 static int check_ptr_to_btf_access(struct bpf_verifier_env *env,
@@ -5181,49 +5228,58 @@ static int check_ptr_to_btf_access(struct bpf_verifier_env *env,
 	if (ret < 0)
 		return ret;
 
-	/* If this is an untrusted pointer, all pointers formed by walking it
-	 * also inherit the untrusted flag.
-	 */
-	if (type_flag(reg->type) & PTR_UNTRUSTED)
-		flag |= PTR_UNTRUSTED;
+	if (ret != PTR_TO_BTF_ID) {
+		/* just mark; */
 
-	/* By default any pointer obtained from walking a trusted pointer is no
-	 * longer trusted, unless the field being accessed has explicitly been
-	 * marked as inheriting its parent's state of trust.
-	 *
-	 * An RCU-protected pointer can also be deemed trusted if we are in an
-	 * RCU read region. This case is handled below.
-	 */
-	if (nested_ptr_is_trusted(env, reg, off)) {
-		flag |= PTR_TRUSTED;
-		/*
-		 * task->cgroups is trusted. It provides a stronger guarantee
-		 * than __rcu tag on 'cgroups' field in 'struct task_struct'.
-		 * Clear MEM_RCU in such case.
+	} else if (type_flag(reg->type) & PTR_UNTRUSTED) {
+		/* If this is an untrusted pointer, all pointers formed by walking it
+		 * also inherit the untrusted flag.
+		 */
+		flag = PTR_UNTRUSTED;
+
+	} else if (is_trusted_reg(reg) || is_rcu_reg(reg)) {
+		/* By default any pointer obtained from walking a trusted pointer is no
+		 * longer trusted, unless the field being accessed has explicitly been
+		 * marked as inheriting its parent's state of trust (either full or RCU).
+		 * For example:
+		 * 'cgroups' pointer is untrusted if task->cgroups dereference
+		 * happened in a sleepable program outside of bpf_rcu_read_lock()
+		 * section. In a non-sleepable program it's trusted while in RCU CS (aka MEM_RCU).
+		 * Note bpf_rcu_read_unlock() converts MEM_RCU pointers to PTR_UNTRUSTED.
+		 *
+		 * A regular RCU-protected pointer with __rcu tag can also be deemed
+		 * trusted if we are in an RCU CS. Such pointer can be NULL.
 		 */
-		flag &= ~MEM_RCU;
+		if (type_is_trusted(env, reg, off)) {
+			flag |= PTR_TRUSTED;
+		} else if (in_rcu_cs(env) && !type_may_be_null(reg->type)) {
+			if (type_is_rcu(env, reg, off)) {
+				/* ignore __rcu tag and mark it MEM_RCU */
+				flag |= MEM_RCU;
+			} else if (flag & MEM_RCU) {
+				/* __rcu tagged pointers can be NULL */
+				flag |= PTR_MAYBE_NULL;
+			} else if (flag & (MEM_PERCPU | MEM_USER)) {
+				/* keep as-is */
+			} else {
+				/* walking unknown pointers yields untrusted pointer */
+				flag = PTR_UNTRUSTED;
+			}
+		} else {
+			/*
+			 * If not in RCU CS or MEM_RCU pointer can be NULL then
+			 * aggressively mark as untrusted otherwise such
+			 * pointers will be plain PTR_TO_BTF_ID without flags
+			 * and will be allowed to be passed into helpers for
+			 * compat reasons.
+			 */
+			flag = PTR_UNTRUSTED;
+		}
 	} else {
+		/* Old compat. Deprecated */
 		flag &= ~PTR_TRUSTED;
 	}
 
-	if (flag & MEM_RCU) {
-		/* Mark value register as MEM_RCU only if it is protected by
-		 * bpf_rcu_read_lock() and the ptr reg is rcu or trusted. MEM_RCU
-		 * itself can already indicate trustedness inside the rcu
-		 * read lock region. Also mark rcu pointer as PTR_MAYBE_NULL since
-		 * it could be null in some cases.
-		 */
-		if (in_rcu_cs(env) && (is_trusted_reg(reg) || is_rcu_reg(reg)))
-			flag |= PTR_MAYBE_NULL;
-		else
-			flag &= ~MEM_RCU;
-	} else if (reg->type & MEM_RCU) {
-		/* ptr (reg) is marked as MEM_RCU, but the struct field is not tagged
-		 * with __rcu. Mark the flag as PTR_UNTRUSTED conservatively.
-		 */
-		flag |= PTR_UNTRUSTED;
-	}
-
 	if (atype == BPF_READ && value_regno >= 0)
 		mark_btf_ld_reg(env, regs, value_regno, ret, reg->btf, btf_id, flag);
 
@@ -10049,10 +10105,6 @@ static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 
 	rcu_lock = is_kfunc_bpf_rcu_read_lock(&meta);
 	rcu_unlock = is_kfunc_bpf_rcu_read_unlock(&meta);
-	if ((rcu_lock || rcu_unlock) && !env->rcu_tag_supported) {
-		verbose(env, "no vmlinux btf rcu tag support for kfunc %s\n", func_name);
-		return -EACCES;
-	}
 
 	if (env->cur_state->active_rcu_lock) {
 		struct bpf_func_state *state;
@@ -14911,8 +14963,22 @@ static int do_check(struct bpf_verifier_env *env)
 				 * src_reg == stack|map in some other branch.
 				 * Reject it.
 				 */
-				verbose(env, "same insn cannot be used with different pointers\n");
-				return -EINVAL;
+				if (base_type(src_reg_type) == PTR_TO_BTF_ID &&
+				    base_type(*prev_src_type) == PTR_TO_BTF_ID) {
+					/*
+					 * Have to support a use case when one path through
+					 * the program yields TRUSTED pointer while another
+					 * is UNTRUSTED. Fallback to UNTRUSTED to generate
+					 * BPF_PROBE_MEM.
+					 */
+					*prev_src_type = PTR_TO_BTF_ID | PTR_UNTRUSTED;
+				} else {
+					verbose(env,
+						"The same insn cannot be used with different pointers: %s",
+						reg_type_str(env, src_reg_type));
+					verbose(env, " != %s\n", reg_type_str(env, *prev_src_type));
+					return -EINVAL;
+				}
 			}
 
 		} else if (class == BPF_STX) {
@@ -17984,8 +18050,6 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr)
 	env->bypass_spec_v1 = bpf_bypass_spec_v1();
 	env->bypass_spec_v4 = bpf_bypass_spec_v4();
 	env->bpf_capable = bpf_capable();
-	env->rcu_tag_supported = btf_vmlinux &&
-		btf_find_by_name_kind(btf_vmlinux, "rcu", BTF_KIND_TYPE_TAG) > 0;
 
 	if (is_priv)
 		env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ;