Merge branch 'bpf-per-cpu-maps'

Alexei Starovoitov says:

====================
bpf: introduce per-cpu maps

We've started to use bpf to trace every packet and atomic add
instruction (event JITed) started to show up in perf profile.
The solution is to do per-cpu counters.
For PERCPU_(HASH|ARRAY) map the existing bpf_map_lookup() helper
returns per-cpu area which bpf programs can use to store and
increment the counters. The BPF_MAP_LOOKUP_ELEM syscall command
returns areas from all cpus and user process aggregates the counters.
The usage example is in patch 6. The api turned out to be very
easy to use from bpf program and from user space.
Long term we were discussing to add 'bounded loop' instruction,
so bpf programs can do aggregation within the program which may
help some use cases. Right now user space aggregation of
per-cpu counters fits the best.

This patch set is new approach for per-cpu hash and array maps.
I've reused the map tests written by Martin and Ming, but
implementation and api is new. Old discussion here:
http://thread.gmane.org/gmane.linux.kernel/2123800/focus=2126435
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 155 insertions, 11 deletions

diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
index 89ebbc4d1164f..bd3bdf2486a7b 100644
--- a/kernel/bpf/arraymap.c
+++ b/kernel/bpf/arraymap.c
@@ -17,11 +17,39 @@
 #include <linux/filter.h>
 #include <linux/perf_event.h>
 
+static void bpf_array_free_percpu(struct bpf_array *array)
+{
+	int i;
+
+	for (i = 0; i < array->map.max_entries; i++)
+		free_percpu(array->pptrs[i]);
+}
+
+static int bpf_array_alloc_percpu(struct bpf_array *array)
+{
+	void __percpu *ptr;
+	int i;
+
+	for (i = 0; i < array->map.max_entries; i++) {
+		ptr = __alloc_percpu_gfp(array->elem_size, 8,
+					 GFP_USER | __GFP_NOWARN);
+		if (!ptr) {
+			bpf_array_free_percpu(array);
+			return -ENOMEM;
+		}
+		array->pptrs[i] = ptr;
+	}
+
+	return 0;
+}
+
 /* Called from syscall */
 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
 {
+	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
 	struct bpf_array *array;
-	u32 elem_size, array_size;
+	u64 array_size;
+	u32 elem_size;
 
 	/* check sanity of attributes */
 	if (attr->max_entries == 0 || attr->key_size != 4 ||
@@ -36,12 +64,16 @@ static struct bpf_map *array_map_alloc(union bpf_attr *attr)
 
 	elem_size = round_up(attr->value_size, 8);
 
-	/* check round_up into zero and u32 overflow */
-	if (elem_size == 0 ||
-	    attr->max_entries > (U32_MAX - PAGE_SIZE - sizeof(*array)) / elem_size)
+	array_size = sizeof(*array);
+	if (percpu)
+		array_size += (u64) attr->max_entries * sizeof(void *);
+	else
+		array_size += (u64) attr->max_entries * elem_size;
+
+	/* make sure there is no u32 overflow later in round_up() */
+	if (array_size >= U32_MAX - PAGE_SIZE)
 		return ERR_PTR(-ENOMEM);
 
-	array_size = sizeof(*array) + attr->max_entries * elem_size;
 
 	/* allocate all map elements and zero-initialize them */
 	array = kzalloc(array_size, GFP_USER | __GFP_NOWARN);
@@ -52,12 +84,25 @@ static struct bpf_map *array_map_alloc(union bpf_attr *attr)
 	}
 
 	/* copy mandatory map attributes */
+	array->map.map_type = attr->map_type;
 	array->map.key_size = attr->key_size;
 	array->map.value_size = attr->value_size;
 	array->map.max_entries = attr->max_entries;
-	array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
 	array->elem_size = elem_size;
 
+	if (!percpu)
+		goto out;
+
+	array_size += (u64) attr->max_entries * elem_size * num_possible_cpus();
+
+	if (array_size >= U32_MAX - PAGE_SIZE ||
+	    elem_size > PCPU_MIN_UNIT_SIZE || bpf_array_alloc_percpu(array)) {
+		kvfree(array);
+		return ERR_PTR(-ENOMEM);
+	}
+out:
+	array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
+
 	return &array->map;
 }
 
@@ -67,12 +112,50 @@ static void *array_map_lookup_elem(struct bpf_map *map, void *key)
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	u32 index = *(u32 *)key;
 
-	if (index >= array->map.max_entries)
+	if (unlikely(index >= array->map.max_entries))
 		return NULL;
 
 	return array->value + array->elem_size * index;
 }
 
+/* Called from eBPF program */
+static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
+{
+	struct bpf_array *array = container_of(map, struct bpf_array, map);
+	u32 index = *(u32 *)key;
+
+	if (unlikely(index >= array->map.max_entries))
+		return NULL;
+
+	return this_cpu_ptr(array->pptrs[index]);
+}
+
+int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
+{
+	struct bpf_array *array = container_of(map, struct bpf_array, map);
+	u32 index = *(u32 *)key;
+	void __percpu *pptr;
+	int cpu, off = 0;
+	u32 size;
+
+	if (unlikely(index >= array->map.max_entries))
+		return -ENOENT;
+
+	/* per_cpu areas are zero-filled and bpf programs can only
+	 * access 'value_size' of them, so copying rounded areas
+	 * will not leak any kernel data
+	 */
+	size = round_up(map->value_size, 8);
+	rcu_read_lock();
+	pptr = array->pptrs[index];
+	for_each_possible_cpu(cpu) {
+		bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
+		off += size;
+	}
+	rcu_read_unlock();
+	return 0;
+}
+
 /* Called from syscall */
 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 {
@@ -99,19 +182,62 @@ static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	u32 index = *(u32 *)key;
 
-	if (map_flags > BPF_EXIST)
+	if (unlikely(map_flags > BPF_EXIST))
 		/* unknown flags */
 		return -EINVAL;
 
-	if (index >= array->map.max_entries)
+	if (unlikely(index >= array->map.max_entries))
 		/* all elements were pre-allocated, cannot insert a new one */
 		return -E2BIG;
 
-	if (map_flags == BPF_NOEXIST)
+	if (unlikely(map_flags == BPF_NOEXIST))
 		/* all elements already exist */
 		return -EEXIST;
 
-	memcpy(array->value + array->elem_size * index, value, map->value_size);
+	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
+		memcpy(this_cpu_ptr(array->pptrs[index]),
+		       value, map->value_size);
+	else
+		memcpy(array->value + array->elem_size * index,
+		       value, map->value_size);
+	return 0;
+}
+
+int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
+			    u64 map_flags)
+{
+	struct bpf_array *array = container_of(map, struct bpf_array, map);
+	u32 index = *(u32 *)key;
+	void __percpu *pptr;
+	int cpu, off = 0;
+	u32 size;
+
+	if (unlikely(map_flags > BPF_EXIST))
+		/* unknown flags */
+		return -EINVAL;
+
+	if (unlikely(index >= array->map.max_entries))
+		/* all elements were pre-allocated, cannot insert a new one */
+		return -E2BIG;
+
+	if (unlikely(map_flags == BPF_NOEXIST))
+		/* all elements already exist */
+		return -EEXIST;
+
+	/* the user space will provide round_up(value_size, 8) bytes that
+	 * will be copied into per-cpu area. bpf programs can only access
+	 * value_size of it. During lookup the same extra bytes will be
+	 * returned or zeros which were zero-filled by percpu_alloc,
+	 * so no kernel data leaks possible
+	 */
+	size = round_up(map->value_size, 8);
+	rcu_read_lock();
+	pptr = array->pptrs[index];
+	for_each_possible_cpu(cpu) {
+		bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
+		off += size;
+	}
+	rcu_read_unlock();
 	return 0;
 }
 
@@ -133,6 +259,9 @@ static void array_map_free(struct bpf_map *map)
 	 */
 	synchronize_rcu();
 
+	if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
+		bpf_array_free_percpu(array);
+
 	kvfree(array);
 }
 
@@ -150,9 +279,24 @@ static struct bpf_map_type_list array_type __read_mostly = {
 	.type = BPF_MAP_TYPE_ARRAY,
 };
 
+static const struct bpf_map_ops percpu_array_ops = {
+	.map_alloc = array_map_alloc,
+	.map_free = array_map_free,
+	.map_get_next_key = array_map_get_next_key,
+	.map_lookup_elem = percpu_array_map_lookup_elem,
+	.map_update_elem = array_map_update_elem,
+	.map_delete_elem = array_map_delete_elem,
+};
+
+static struct bpf_map_type_list percpu_array_type __read_mostly = {
+	.ops = &percpu_array_ops,
+	.type = BPF_MAP_TYPE_PERCPU_ARRAY,
+};
+
 static int __init register_array_map(void)
 {
 	bpf_register_map_type(&array_type);
+	bpf_register_map_type(&percpu_array_type);
 	return 0;
 }
 late_initcall(register_array_map);