Merge branch 'BPF array map fixes and improvements'

Andrii Nakryiko says:

====================

Fix 32-bit overflow in value pointer calculations in BPF array map. And then
raise obsolete limit on array map value size. Add selftest making sure this is
working as intended.

v1->v2:
- fix broken patch #1 (no mask_index use in helper, as stated in commit
message; and add missing semicolon).
====================

Acked-by: Yonghong Song <[e-mail]>
Signed-off-by: Alexei Starovoitov <[e-mail]>

kernel/bpf/arraymap.c

···
 	    attr->map_flags & BPF_F_PRESERVE_ELEMS)
 		return -EINVAL;
 
-	if (attr->value_size > KMALLOC_MAX_SIZE)
-		/* if value_size is bigger, the user space won't be able to
-		 * access the elements.
-		 */
+	/* avoid overflow on round_up(map->value_size) */
+	if (attr->value_size > INT_MAX)
 		return -E2BIG;
 
 	return 0;
···
 	return &array->map;
 }
 
+static void *array_map_elem_ptr(struct bpf_array* array, u32 index)
+{
+	return array->value + (u64)array->elem_size * index;
+}
+
 /* Called from syscall or from eBPF program */
 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
 {
···
 	if (unlikely(index >= array->map.max_entries))
 		return NULL;
 
-	return array->value + array->elem_size * (index & array->index_mask);
+	return array->value + (u64)array->elem_size * (index & array->index_mask);
 }
 
 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
···
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	struct bpf_insn *insn = insn_buf;
-	u32 elem_size = round_up(map->value_size, 8);
+	u32 elem_size = array->elem_size;
 	const int ret = BPF_REG_0;
 	const int map_ptr = BPF_REG_1;
 	const int index = BPF_REG_2;
···
 	 * access 'value_size' of them, so copying rounded areas
 	 * will not leak any kernel data
 	 */
-	size = round_up(map->value_size, 8);
+	size = array->elem_size;
 	rcu_read_lock();
 	pptr = array->pptrs[index & array->index_mask];
 	for_each_possible_cpu(cpu) {
···
 		       value, map->value_size);
 	} else {
 		val = array->value +
-			array->elem_size * (index & array->index_mask);
+			(u64)array->elem_size * (index & array->index_mask);
 		if (map_flags & BPF_F_LOCK)
 			copy_map_value_locked(map, val, value, false);
 		else
···
 	 * returned or zeros which were zero-filled by percpu_alloc,
 	 * so no kernel data leaks possible
 	 */
-	size = round_up(map->value_size, 8);
+	size = array->elem_size;
 	rcu_read_lock();
 	pptr = array->pptrs[index & array->index_mask];
 	for_each_possible_cpu(cpu) {
···
 		return;
 
 	for (i = 0; i < array->map.max_entries; i++)
-		bpf_timer_cancel_and_free(array->value + array->elem_size * i +
-					  map->timer_off);
+		bpf_timer_cancel_and_free(array_map_elem_ptr(array, i) + map->timer_off);
 }
 
 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
···
 
 	if (map_value_has_kptrs(map)) {
 		for (i = 0; i < array->map.max_entries; i++)
-			bpf_map_free_kptrs(map, array->value + array->elem_size * i);
+			bpf_map_free_kptrs(map, array_map_elem_ptr(array, i));
 		bpf_map_free_kptr_off_tab(map);
 	}
 
···
 	index = info->index & array->index_mask;
 	if (info->percpu_value_buf)
 	       return array->pptrs[index];
-	return array->value + array->elem_size * index;
+	return array_map_elem_ptr(array, index);
 }
 
 static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
···
 	index = info->index & array->index_mask;
 	if (info->percpu_value_buf)
 	       return array->pptrs[index];
-	return array->value + array->elem_size * index;
+	return array_map_elem_ptr(array, index);
 }
 
 static int __bpf_array_map_seq_show(struct seq_file *seq, void *v)
···
 	struct bpf_iter_seq_array_map_info *info = seq->private;
 	struct bpf_iter__bpf_map_elem ctx = {};
 	struct bpf_map *map = info->map;
+	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	struct bpf_iter_meta meta;
 	struct bpf_prog *prog;
 	int off = 0, cpu = 0;
···
 			ctx.value = v;
 		} else {
 			pptr = v;
-			size = round_up(map->value_size, 8);
+			size = array->elem_size;
 			for_each_possible_cpu(cpu) {
 				bpf_long_memcpy(info->percpu_value_buf + off,
 						per_cpu_ptr(pptr, cpu),
···
 {
 	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
 	struct bpf_map *map = aux->map;
+	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	void *value_buf;
 	u32 buf_size;
 
 	if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
-		buf_size = round_up(map->value_size, 8) * num_possible_cpus();
+		buf_size = array->elem_size * num_possible_cpus();
 		value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
 		if (!value_buf)
 			return -ENOMEM;
···
 		if (is_percpu)
 			val = this_cpu_ptr(array->pptrs[i]);
 		else
-			val = array->value + array->elem_size * i;
+			val = array_map_elem_ptr(array, i);
 		num_elems++;
 		key = i;
 		ret = callback_fn((u64)(long)map, (u64)(long)&key,
···
 				   struct bpf_insn *insn_buf)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
-	u32 elem_size = round_up(map->value_size, 8);
+	u32 elem_size = array->elem_size;
 	struct bpf_insn *insn = insn_buf;
 	const int ret = BPF_REG_0;
 	const int map_ptr = BPF_REG_1;

tools/testing/selftests/bpf/prog_tests/skeleton.c

···
 
 	ASSERT_EQ(skel->bss->out_mostly_var, 123, "out_mostly_var");
 
+	ASSERT_EQ(bss->huge_arr[ARRAY_SIZE(bss->huge_arr) - 1], 123, "huge_arr");
+
 	elf_bytes = test_skeleton__elf_bytes(&elf_bytes_sz);
 	ASSERT_OK_PTR(elf_bytes, "elf_bytes");
 	ASSERT_GE(elf_bytes_sz, 0, "elf_bytes_sz");

tools/testing/selftests/bpf/progs/test_skeleton.c

···
 int read_mostly_var __read_mostly;
 int out_mostly_var;
 
+char huge_arr[16 * 1024 * 1024];
+
 SEC("raw_tp/sys_enter")
 int handler(const void *ctx)
 {
···
 		out_dynarr[i] = in_dynarr[i];
 
 	out_mostly_var = read_mostly_var;
+
+	huge_arr[sizeof(huge_arr) - 1] = 123;
 
 	return 0;
 }