bpf: Introduce support for bpf_local_irq_{save,restore}

Teach the verifier about IRQ-disabled sections through the introduction
of two new kfuncs, bpf_local_irq_save, to save IRQ state and disable
them, and bpf_local_irq_restore, to restore IRQ state and enable them
back again.

For the purposes of tracking the saved IRQ state, the verifier is taught
about a new special object on the stack of type STACK_IRQ_FLAG. This is
a 8 byte value which saves the IRQ flags which are to be passed back to
the IRQ restore kfunc.

Renumber the enums for REF_TYPE_* to simplify the check in
find_lock_state, filtering out non-lock types as they grow will become
cumbersome and is unecessary.

To track a dynamic number of IRQ-disabled regions and their associated
saved states, a new resource type RES_TYPE_IRQ is introduced, which its
state management functions: acquire_irq_state and release_irq_state,
taking advantage of the refactoring and clean ups made in earlier
commits.

One notable requirement of the kernel's IRQ save and restore API is that
they cannot happen out of order. For this purpose, when releasing reference
we keep track of the prev_id we saw with REF_TYPE_IRQ. Since reference
states are inserted in increasing order of the index, this is used to
remember the ordering of acquisitions of IRQ saved states, so that we
maintain a logical stack in acquisition order of resource identities,
and can enforce LIFO ordering when restoring IRQ state. The top of the
stack is maintained using bpf_verifier_state's active_irq_id.

To maintain the stack property when releasing reference states, we need
to modify release_reference_state to instead shift the remaining array
left using memmove instead of swapping deleted element with last that
might break the ordering. A selftest to test this subtle behavior is
added in late patches.

The logic to detect initialized and unitialized irq flag slots, marking
and unmarking is similar to how it's done for iterators. No additional
checks are needed in refsafe for REF_TYPE_IRQ, apart from the usual
check_id satisfiability check on the ref[i].id. We have to perform the
same check_ids check on state->active_irq_id as well.

To ensure we don't get assigned REF_TYPE_PTR by default after
acquire_reference_state, if someone forgets to assign the type, let's
also renumber the enum ref_state_type. This way any unassigned types
get caught by refsafe's default switch statement, don't assume
REF_TYPE_PTR by default.

The kfuncs themselves are plain wrappers over local_irq_save and
local_irq_restore macros.

Acked-by: Eduard Zingerman <[e-mail]>
Signed-off-by: Kumar Kartikeya Dwivedi <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Alexei Starovoitov <[e-mail]>

include/linux/bpf_verifier.h

···
 	 */
 	STACK_DYNPTR,
 	STACK_ITER,
+	STACK_IRQ_FLAG,
 };
 
 #define BPF_REG_SIZE 8	/* size of eBPF register in bytes */
···
 	 * default to pointer reference on zero initialization of a state.
 	 */
 	enum ref_state_type {
-		REF_TYPE_PTR = 0,
-		REF_TYPE_LOCK,
+		REF_TYPE_PTR	= 1,
+		REF_TYPE_IRQ	= 2,
+		REF_TYPE_LOCK	= 3,
 	} type;
 	/* Track each reference created with a unique id, even if the same
 	 * instruction creates the reference multiple times (eg, via CALL).
···
 	u32 acquired_refs;
 	u32 active_locks;
 	u32 active_preempt_locks;
+	u32 active_irq_id;
 	bool active_rcu_lock;
 
 	bool speculative;

kernel/bpf/helpers.c

···
 	return ret + 1;
 }
 
+/* Keep unsinged long in prototype so that kfunc is usable when emitted to
+ * vmlinux.h in BPF programs directly, but note that while in BPF prog, the
+ * unsigned long always points to 8-byte region on stack, the kernel may only
+ * read and write the 4-bytes on 32-bit.
+ */
+__bpf_kfunc void bpf_local_irq_save(unsigned long *flags__irq_flag)
+{
+	local_irq_save(*flags__irq_flag);
+}
+
+__bpf_kfunc void bpf_local_irq_restore(unsigned long *flags__irq_flag)
+{
+	local_irq_restore(*flags__irq_flag);
+}
+
 __bpf_kfunc_end_defs();
 
 BTF_KFUNCS_START(generic_btf_ids)
···
 BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
 BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
+BTF_ID_FLAGS(func, bpf_local_irq_save)
+BTF_ID_FLAGS(func, bpf_local_irq_restore)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {

kernel/bpf/log.c

···
 	[STACK_ZERO]	= '0',
 	[STACK_DYNPTR]	= 'd',
 	[STACK_ITER]	= 'i',
+	[STACK_IRQ_FLAG] = 'f'
 };
 
 static void print_liveness(struct bpf_verifier_env *env,

kernel/bpf/verifier.c

···
 	return stack_slot_obj_get_spi(env, reg, "iter", nr_slots);
 }
 
+static int irq_flag_get_spi(struct bpf_verifier_env *env, struct bpf_reg_state *reg)
+{
+	return stack_slot_obj_get_spi(env, reg, "irq_flag", 1);
+}
+
 static enum bpf_dynptr_type arg_to_dynptr_type(enum bpf_arg_type arg_type)
 {
 	switch (arg_type & DYNPTR_TYPE_FLAG_MASK) {
···
 	return 0;
 }
 
+static int acquire_irq_state(struct bpf_verifier_env *env, int insn_idx);
+static int release_irq_state(struct bpf_verifier_state *state, int id);
+
+static int mark_stack_slot_irq_flag(struct bpf_verifier_env *env,
+				     struct bpf_kfunc_call_arg_meta *meta,
+				     struct bpf_reg_state *reg, int insn_idx)
+{
+	struct bpf_func_state *state = func(env, reg);
+	struct bpf_stack_state *slot;
+	struct bpf_reg_state *st;
+	int spi, i, id;
+
+	spi = irq_flag_get_spi(env, reg);
+	if (spi < 0)
+		return spi;
+
+	id = acquire_irq_state(env, insn_idx);
+	if (id < 0)
+		return id;
+
+	slot = &state->stack[spi];
+	st = &slot->spilled_ptr;
+
+	__mark_reg_known_zero(st);
+	st->type = PTR_TO_STACK; /* we don't have dedicated reg type */
+	st->live |= REG_LIVE_WRITTEN;
+	st->ref_obj_id = id;
+
+	for (i = 0; i < BPF_REG_SIZE; i++)
+		slot->slot_type[i] = STACK_IRQ_FLAG;
+
+	mark_stack_slot_scratched(env, spi);
+	return 0;
+}
+
+static int unmark_stack_slot_irq_flag(struct bpf_verifier_env *env, struct bpf_reg_state *reg)
+{
+	struct bpf_func_state *state = func(env, reg);
+	struct bpf_stack_state *slot;
+	struct bpf_reg_state *st;
+	int spi, i, err;
+
+	spi = irq_flag_get_spi(env, reg);
+	if (spi < 0)
+		return spi;
+
+	slot = &state->stack[spi];
+	st = &slot->spilled_ptr;
+
+	err = release_irq_state(env->cur_state, st->ref_obj_id);
+	WARN_ON_ONCE(err && err != -EACCES);
+	if (err) {
+		int insn_idx = 0;
+
+		for (int i = 0; i < env->cur_state->acquired_refs; i++) {
+			if (env->cur_state->refs[i].id == env->cur_state->active_irq_id) {
+				insn_idx = env->cur_state->refs[i].insn_idx;
+				break;
+			}
+		}
+
+		verbose(env, "cannot restore irq state out of order, expected id=%d acquired at insn_idx=%d\n",
+			env->cur_state->active_irq_id, insn_idx);
+		return err;
+	}
+
+	__mark_reg_not_init(env, st);
+
+	/* see unmark_stack_slots_dynptr() for why we need to set REG_LIVE_WRITTEN */
+	st->live |= REG_LIVE_WRITTEN;
+
+	for (i = 0; i < BPF_REG_SIZE; i++)
+		slot->slot_type[i] = STACK_INVALID;
+
+	mark_stack_slot_scratched(env, spi);
+	return 0;
+}
+
+static bool is_irq_flag_reg_valid_uninit(struct bpf_verifier_env *env, struct bpf_reg_state *reg)
+{
+	struct bpf_func_state *state = func(env, reg);
+	struct bpf_stack_state *slot;
+	int spi, i;
+
+	/* For -ERANGE (i.e. spi not falling into allocated stack slots), we
+	 * will do check_mem_access to check and update stack bounds later, so
+	 * return true for that case.
+	 */
+	spi = irq_flag_get_spi(env, reg);
+	if (spi == -ERANGE)
+		return true;
+	if (spi < 0)
+		return false;
+
+	slot = &state->stack[spi];
+
+	for (i = 0; i < BPF_REG_SIZE; i++)
+		if (slot->slot_type[i] == STACK_IRQ_FLAG)
+			return false;
+	return true;
+}
+
+static int is_irq_flag_reg_valid_init(struct bpf_verifier_env *env, struct bpf_reg_state *reg)
+{
+	struct bpf_func_state *state = func(env, reg);
+	struct bpf_stack_state *slot;
+	struct bpf_reg_state *st;
+	int spi, i;
+
+	spi = irq_flag_get_spi(env, reg);
+	if (spi < 0)
+		return -EINVAL;
+
+	slot = &state->stack[spi];
+	st = &slot->spilled_ptr;
+
+	if (!st->ref_obj_id)
+		return -EINVAL;
+
+	for (i = 0; i < BPF_REG_SIZE; i++)
+		if (slot->slot_type[i] != STACK_IRQ_FLAG)
+			return -EINVAL;
+	return 0;
+}
+
 /* Check if given stack slot is "special":
  *   - spilled register state (STACK_SPILL);
  *   - dynptr state (STACK_DYNPTR);
  *   - iter state (STACK_ITER).
+ *   - irq flag state (STACK_IRQ_FLAG)
  */
 static bool is_stack_slot_special(const struct bpf_stack_state *stack)
 {
···
 	case STACK_SPILL:
 	case STACK_DYNPTR:
 	case STACK_ITER:
+	case STACK_IRQ_FLAG:
 		return true;
 	case STACK_INVALID:
 	case STACK_MISC:
···
 	dst->active_locks = src->active_locks;
 	dst->active_preempt_locks = src->active_preempt_locks;
 	dst->active_rcu_lock = src->active_rcu_lock;
+	dst->active_irq_id = src->active_irq_id;
 	return 0;
 }
 
···
 	return 0;
 }
 
+static int acquire_irq_state(struct bpf_verifier_env *env, int insn_idx)
+{
+	struct bpf_verifier_state *state = env->cur_state;
+	struct bpf_reference_state *s;
+
+	s = acquire_reference_state(env, insn_idx);
+	if (!s)
+		return -ENOMEM;
+	s->type = REF_TYPE_IRQ;
+	s->id = ++env->id_gen;
+
+	state->active_irq_id = s->id;
+	return s->id;
+}
+
 static void release_reference_state(struct bpf_verifier_state *state, int idx)
 {
 	int last_idx;
+	size_t rem;
 
+	/* IRQ state requires the relative ordering of elements remaining the
+	 * same, since it relies on the refs array to behave as a stack, so that
+	 * it can detect out-of-order IRQ restore. Hence use memmove to shift
+	 * the array instead of swapping the final element into the deleted idx.
+	 */
 	last_idx = state->acquired_refs - 1;
+	rem = state->acquired_refs - idx - 1;
 	if (last_idx && idx != last_idx)
-		memcpy(&state->refs[idx], &state->refs[last_idx], sizeof(*state->refs));
+		memmove(&state->refs[idx], &state->refs[idx + 1], sizeof(*state->refs) * rem);
 	memset(&state->refs[last_idx], 0, sizeof(*state->refs));
 	state->acquired_refs--;
 	return;
···
 	return -EINVAL;
 }
 
+static int release_irq_state(struct bpf_verifier_state *state, int id)
+{
+	u32 prev_id = 0;
+	int i;
+
+	if (id != state->active_irq_id)
+		return -EACCES;
+
+	for (i = 0; i < state->acquired_refs; i++) {
+		if (state->refs[i].type != REF_TYPE_IRQ)
+			continue;
+		if (state->refs[i].id == id) {
+			release_reference_state(state, i);
+			state->active_irq_id = prev_id;
+			return 0;
+		} else {
+			prev_id = state->refs[i].id;
+		}
+	}
+	return -EINVAL;
+}
+
 static struct bpf_reference_state *find_lock_state(struct bpf_verifier_state *state, enum ref_state_type type,
 						   int id, void *ptr)
 {
···
 	for (i = 0; i < state->acquired_refs; i++) {
 		struct bpf_reference_state *s = &state->refs[i];
 
-		if (s->type == REF_TYPE_PTR || s->type != type)
+		if (s->type != type)
 			continue;
 
 		if (s->id == id && s->ptr == ptr)
···
 			  int spi, int nr_slots)
 {
 	return mark_stack_slot_obj_read(env, reg, spi, nr_slots);
+}
+
+static int mark_irq_flag_read(struct bpf_verifier_env *env, struct bpf_reg_state *reg)
+{
+	int spi;
+
+	spi = irq_flag_get_spi(env, reg);
+	if (spi < 0)
+		return spi;
+	return mark_stack_slot_obj_read(env, reg, spi, 1);
 }
 
 /* This function is supposed to be used by the following 32-bit optimization
···
 			return -EINVAL;
 		}
 
+		if (env->cur_state->active_irq_id) {
+			verbose(env, "global function calls are not allowed with IRQs disabled,\n"
+				     "use static function instead\n");
+			return -EINVAL;
+		}
+
 		if (err) {
 			verbose(env, "Caller passes invalid args into func#%d ('%s')\n",
 				subprog, sub_name);
···
 		return err;
 	}
 
+	if (check_lock && env->cur_state->active_irq_id) {
+		verbose(env, "%s cannot be used inside bpf_local_irq_save-ed region\n", prefix);
+		return -EINVAL;
+	}
+
 	if (check_lock && env->cur_state->active_rcu_lock) {
 		verbose(env, "%s cannot be used inside bpf_rcu_read_lock-ed region\n", prefix);
 		return -EINVAL;
···
 	if (env->cur_state->active_preempt_locks) {
 		if (fn->might_sleep) {
 			verbose(env, "sleepable helper %s#%d in non-preemptible region\n",
+				func_id_name(func_id), func_id);
+			return -EINVAL;
+		}
+
+		if (in_sleepable(env) && is_storage_get_function(func_id))
+			env->insn_aux_data[insn_idx].storage_get_func_atomic = true;
+	}
+
+	if (env->cur_state->active_irq_id) {
+		if (fn->might_sleep) {
+			verbose(env, "sleepable helper %s#%d in IRQ-disabled region\n",
 				func_id_name(func_id), func_id);
 			return -EINVAL;
 		}
···
 	return btf_param_match_suffix(btf, arg, "__str");
 }
 
+static bool is_kfunc_arg_irq_flag(const struct btf *btf, const struct btf_param *arg)
+{
+	return btf_param_match_suffix(btf, arg, "__irq_flag");
+}
+
 static bool is_kfunc_arg_scalar_with_name(const struct btf *btf,
 					  const struct btf_param *arg,
 					  const char *name)
···
 	KF_ARG_PTR_TO_CONST_STR,
 	KF_ARG_PTR_TO_MAP,
 	KF_ARG_PTR_TO_WORKQUEUE,
+	KF_ARG_PTR_TO_IRQ_FLAG,
 };
 
 enum special_kfunc_type {
···
 	KF_bpf_iter_css_task_new,
 	KF_bpf_session_cookie,
 	KF_bpf_get_kmem_cache,
+	KF_bpf_local_irq_save,
+	KF_bpf_local_irq_restore,
 };
 
 BTF_SET_START(special_kfunc_set)
···
 BTF_ID_UNUSED
 #endif
 BTF_ID(func, bpf_get_kmem_cache)
+BTF_ID(func, bpf_local_irq_save)
+BTF_ID(func, bpf_local_irq_restore)
 
 static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta)
 {
···
 
 	if (is_kfunc_arg_wq(meta->btf, &args[argno]))
 		return KF_ARG_PTR_TO_WORKQUEUE;
+
+	if (is_kfunc_arg_irq_flag(meta->btf, &args[argno]))
+		return KF_ARG_PTR_TO_IRQ_FLAG;
 
 	if ((base_type(reg->type) == PTR_TO_BTF_ID || reg2btf_ids[base_type(reg->type)])) {
 		if (!btf_type_is_struct(ref_t)) {
···
 	}
 	return 0;
 }
+
+static int process_irq_flag(struct bpf_verifier_env *env, int regno,
+			     struct bpf_kfunc_call_arg_meta *meta)
+{
+	struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
+	bool irq_save;
+	int err;
+
+	if (meta->func_id == special_kfunc_list[KF_bpf_local_irq_save]) {
+		irq_save = true;
+	} else if (meta->func_id == special_kfunc_list[KF_bpf_local_irq_restore]) {
+		irq_save = false;
+	} else {
+		verbose(env, "verifier internal error: unknown irq flags kfunc\n");
+		return -EFAULT;
+	}
+
+	if (irq_save) {
+		if (!is_irq_flag_reg_valid_uninit(env, reg)) {
+			verbose(env, "expected uninitialized irq flag as arg#%d\n", regno - 1);
+			return -EINVAL;
+		}
+
+		err = check_mem_access(env, env->insn_idx, regno, 0, BPF_DW, BPF_WRITE, -1, false, false);
+		if (err)
+			return err;
+
+		err = mark_stack_slot_irq_flag(env, meta, reg, env->insn_idx);
+		if (err)
+			return err;
+	} else {
+		err = is_irq_flag_reg_valid_init(env, reg);
+		if (err) {
+			verbose(env, "expected an initialized irq flag as arg#%d\n", regno - 1);
+			return err;
+		}
+
+		err = mark_irq_flag_read(env, reg);
+		if (err)
+			return err;
+
+		err = unmark_stack_slot_irq_flag(env, reg);
+		if (err)
+			return err;
+	}
+	return 0;
+}
+
 
 static int ref_set_non_owning(struct bpf_verifier_env *env, struct bpf_reg_state *reg)
 {
···
 		case KF_ARG_PTR_TO_REFCOUNTED_KPTR:
 		case KF_ARG_PTR_TO_CONST_STR:
 		case KF_ARG_PTR_TO_WORKQUEUE:
+		case KF_ARG_PTR_TO_IRQ_FLAG:
 			break;
 		default:
 			WARN_ON_ONCE(1);
···
 			if (ret < 0)
 				return ret;
 			break;
+		case KF_ARG_PTR_TO_IRQ_FLAG:
+			if (reg->type != PTR_TO_STACK) {
+				verbose(env, "arg#%d doesn't point to an irq flag on stack\n", i);
+				return -EINVAL;
+			}
+			ret = process_irq_flag(env, regno, meta);
+			if (ret < 0)
+				return ret;
+			break;
 		}
 	}
 
···
 	} else if (preempt_enable) {
 		verbose(env, "unmatched attempt to enable preemption (kernel function %s)\n", func_name);
 		return -EINVAL;
+	}
+
+	if (env->cur_state->active_irq_id && sleepable) {
+		verbose(env, "kernel func %s is sleepable within IRQ-disabled region\n", func_name);
+		return -EACCES;
 	}
 
 	/* In case of release function, we get register number of refcounted
···
 			    !check_ids(old_reg->ref_obj_id, cur_reg->ref_obj_id, idmap))
 				return false;
 			break;
+		case STACK_IRQ_FLAG:
+			old_reg = &old->stack[spi].spilled_ptr;
+			cur_reg = &cur->stack[spi].spilled_ptr;
+			if (!check_ids(old_reg->ref_obj_id, cur_reg->ref_obj_id, idmap))
+				return false;
+			break;
 		case STACK_MISC:
 		case STACK_ZERO:
 		case STACK_INVALID:
···
 	if (old->active_rcu_lock != cur->active_rcu_lock)
 		return false;
 
+	if (!check_ids(old->active_irq_id, cur->active_irq_id, idmap))
+		return false;
+
 	for (i = 0; i < old->acquired_refs; i++) {
 		if (!check_ids(old->refs[i].id, cur->refs[i].id, idmap) ||
 		    old->refs[i].type != cur->refs[i].type)
 			return false;
 		switch (old->refs[i].type) {
 		case REF_TYPE_PTR:
+		case REF_TYPE_IRQ:
 			break;
 		case REF_TYPE_LOCK:
 			if (old->refs[i].ptr != cur->refs[i].ptr)