[PATCH bpf-next 06/13] bpf: implement bpf iterator for array maps

From: Yonghong Song <yhs@fb.com>
To: <bpf@vger.kernel.org>, <netdev@vger.kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>, <kernel-team@fb.com>,
	Martin KaFai Lau <kafai@fb.com>
Subject: [PATCH bpf-next 06/13] bpf: implement bpf iterator for array maps
Date: Mon, 13 Jul 2020 09:17:46 -0700	[thread overview]
Message-ID: <20200713161746.3077087-1-yhs@fb.com> (raw)
In-Reply-To: <20200713161739.3076283-1-yhs@fb.com>

The bpf iterators for array and percpu array
are implemented. Similar to hash maps, for percpu
array map, bpf program will receive values
from all cpus.

Signed-off-by: Yonghong Song <yhs@fb.com>
---
 kernel/bpf/arraymap.c | 140 ++++++++++++++++++++++++++++++++++++++++++
 kernel/bpf/map_iter.c |   6 +-
 2 files changed, 144 insertions(+), 2 deletions(-)

diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
index c66e8273fccd..e855bafb3a59 100644
--- a/kernel/bpf/arraymap.c
+++ b/kernel/bpf/arraymap.c
@@ -487,6 +487,144 @@ static int array_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
 				   vma->vm_pgoff + pgoff);
 }
 
+struct bpf_iter_seq_array_map_info {
+	struct bpf_map *map;
+	void *percpu_value_buf;
+	u32 index;
+};
+
+static void *bpf_array_map_seq_start(struct seq_file *seq, loff_t *pos)
+{
+	struct bpf_iter_seq_array_map_info *info = seq->private;
+	struct bpf_map *map = info->map;
+	struct bpf_array *array;
+	u32 index;
+
+	if (info->index >= map->max_entries)
+		return NULL;
+
+	if (*pos == 0)
+		++*pos;
+	array = container_of(map, struct bpf_array, map);
+	index = info->index & array->index_mask;
+	if (info->percpu_value_buf)
+	       return array->pptrs[index];
+	return array->value + array->elem_size * index;
+}
+
+static void *bpf_array_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+	struct bpf_iter_seq_array_map_info *info = seq->private;
+	struct bpf_map *map = info->map;
+	struct bpf_array *array;
+	u32 index;
+
+	++*pos;
+	++info->index;
+	if (info->index >= map->max_entries)
+		return NULL;
+
+	array = container_of(map, struct bpf_array, map);
+	index = info->index & array->index_mask;
+	if (info->percpu_value_buf)
+	       return array->pptrs[index];
+	return array->value + array->elem_size * 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_iter_meta meta;
+	struct bpf_array *array;
+	struct bpf_prog *prog;
+	int off = 0, cpu = 0;
+	void __percpu **pptr;
+	u32 size;
+
+	meta.seq = seq;
+	prog = bpf_iter_get_info(&meta, v == NULL);
+	if (!prog)
+		return 0;
+
+	ctx.meta = &meta;
+	ctx.map = info->map;
+	if (v) {
+		ctx.key = &info->index;
+
+		array = container_of(map, struct bpf_array, map);
+		if (!info->percpu_value_buf) {
+			ctx.value = v;
+		} else {
+			pptr = v;
+			size = round_up(map->value_size, 8);
+			for_each_possible_cpu(cpu) {
+				bpf_long_memcpy(info->percpu_value_buf + off,
+						per_cpu_ptr(pptr, cpu),
+						size);
+				off += size;
+			}
+			ctx.value = info->percpu_value_buf;
+		}
+	}
+
+	return bpf_iter_run_prog(prog, &ctx);
+}
+
+static int bpf_array_map_seq_show(struct seq_file *seq, void *v)
+{
+	return __bpf_array_map_seq_show(seq, v);
+}
+
+static void bpf_array_map_seq_stop(struct seq_file *seq, void *v)
+{
+	if (!v)
+		(void)__bpf_array_map_seq_show(seq, NULL);
+}
+
+static int bpf_iter_init_array_map(void *priv_data,
+				   struct bpf_iter_aux_info *aux)
+{
+	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
+	struct bpf_map *map = aux->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();
+		value_buf = kmalloc(buf_size, GFP_USER | __GFP_NOWARN);
+		if (!value_buf)
+			return -ENOMEM;
+
+		seq_info->percpu_value_buf = value_buf;
+	}
+
+	seq_info->map = map;
+	return 0;
+}
+
+static void bpf_iter_fini_array_map(void *priv_data)
+{
+	struct bpf_iter_seq_array_map_info *seq_info = priv_data;
+
+	kfree(seq_info->percpu_value_buf);
+}
+
+static const struct seq_operations bpf_array_map_seq_ops = {
+	.start	= bpf_array_map_seq_start,
+	.next	= bpf_array_map_seq_next,
+	.stop	= bpf_array_map_seq_stop,
+	.show	= bpf_array_map_seq_show,
+};
+
+static const struct bpf_iter_seq_info iter_seq_info = {
+	.seq_ops		= &bpf_array_map_seq_ops,
+	.init_seq_private	= bpf_iter_init_array_map,
+	.fini_seq_private	= bpf_iter_fini_array_map,
+	.seq_priv_size		= sizeof(struct bpf_iter_seq_array_map_info),
+};
+
 static int array_map_btf_id;
 const struct bpf_map_ops array_map_ops = {
 	.map_alloc_check = array_map_alloc_check,
@@ -506,6 +644,7 @@ const struct bpf_map_ops array_map_ops = {
 	.map_update_batch = generic_map_update_batch,
 	.map_btf_name = "bpf_array",
 	.map_btf_id = &array_map_btf_id,
+	.iter_seq_info = &iter_seq_info,
 };
 
 static int percpu_array_map_btf_id;
@@ -521,6 +660,7 @@ const struct bpf_map_ops percpu_array_map_ops = {
 	.map_check_btf = array_map_check_btf,
 	.map_btf_name = "bpf_array",
 	.map_btf_id = &percpu_array_map_btf_id,
+	.iter_seq_info = &iter_seq_info,
 };
 
 static int fd_array_map_alloc_check(union bpf_attr *attr)
diff --git a/kernel/bpf/map_iter.c b/kernel/bpf/map_iter.c
index 2988244853d1..5729d78166e3 100644
--- a/kernel/bpf/map_iter.c
+++ b/kernel/bpf/map_iter.c
@@ -106,10 +106,12 @@ static int bpf_iter_check_map(struct bpf_prog *prog,
 	bool is_percpu = false;
 
 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
-	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH)
+	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
+	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 		is_percpu = true;
 	else if (map->map_type != BPF_MAP_TYPE_HASH &&
-		 map->map_type != BPF_MAP_TYPE_LRU_HASH)
+		 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
+		 map->map_type != BPF_MAP_TYPE_ARRAY)
 		return -EINVAL;
 
 	key_acc_size = prog->aux->max_rdonly_access[0];
-- 
2.24.1

