bpf: Allow selecting numa node during map creation

The current map creation API does not allow to provide the numa-node
preference.  The memory usually comes from where the map-creation-process
is running.  The performance is not ideal if the bpf_prog is known to
always run in a numa node different from the map-creation-process.

One of the use case is sharding on CPU to different LRU maps (i.e.
an array of LRU maps).  Here is the test result of map_perf_test on
the INNER_LRU_HASH_PREALLOC test if we force the lru map used by
CPU0 to be allocated from a remote numa node:

[ The machine has 20 cores. CPU0-9 at node 0. CPU10-19 at node 1 ]

># taskset -c 10 ./map_perf_test 512 8 1260000 8000000
5:inner_lru_hash_map_perf pre-alloc 1628380 events per sec
4:inner_lru_hash_map_perf pre-alloc 1626396 events per sec
3:inner_lru_hash_map_perf pre-alloc 1626144 events per sec
6:inner_lru_hash_map_perf pre-alloc 1621657 events per sec
2:inner_lru_hash_map_perf pre-alloc 1621534 events per sec
1:inner_lru_hash_map_perf pre-alloc 1620292 events per sec
7:inner_lru_hash_map_perf pre-alloc 1613305 events per sec
0:inner_lru_hash_map_perf pre-alloc 1239150 events per sec  #<<<

After specifying numa node:
># taskset -c 10 ./map_perf_test 512 8 1260000 8000000
5:inner_lru_hash_map_perf pre-alloc 1629627 events per sec
3:inner_lru_hash_map_perf pre-alloc 1628057 events per sec
1:inner_lru_hash_map_perf pre-alloc 1623054 events per sec
6:inner_lru_hash_map_perf pre-alloc 1616033 events per sec
2:inner_lru_hash_map_perf pre-alloc 1614630 events per sec
4:inner_lru_hash_map_perf pre-alloc 1612651 events per sec
7:inner_lru_hash_map_perf pre-alloc 1609337 events per sec
0:inner_lru_hash_map_perf pre-alloc 1619340 events per sec #<<<

This patch adds one field, numa_node, to the bpf_attr.  Since numa node 0
is a valid node, a new flag BPF_F_NUMA_NODE is also added.  The numa_node
field is honored if and only if the BPF_F_NUMA_NODE flag is set.

Numa node selection is not supported for percpu map.

This patch does not change all the kmalloc.  F.e.
'htab = kzalloc()' is not changed since the object
is small enough to stay in the cache.

Signed-off-by: Martin KaFai Lau <[email protected]>
Acked-by: Daniel Borkmann <[email protected]>
Acked-by: Alexei Starovoitov <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Author: iamkafai

Diff for: include/linux/bpf.h

@@ -51,6 +51,7 @@ struct bpf_map {
 	u32 map_flags;
 	u32 pages;
 	u32 id;
+	int numa_node;
 	struct user_struct *user;
 	const struct bpf_map_ops *ops;
 	struct work_struct work;
@@ -264,7 +265,7 @@ struct bpf_map * __must_check bpf_map_inc(struct bpf_map *map, bool uref);
 void bpf_map_put_with_uref(struct bpf_map *map);
 void bpf_map_put(struct bpf_map *map);
 int bpf_map_precharge_memlock(u32 pages);
-void *bpf_map_area_alloc(size_t size);
+void *bpf_map_area_alloc(size_t size, int numa_node);
 void bpf_map_area_free(void *base);
 
 extern int sysctl_unprivileged_bpf_disabled;
@@ -316,6 +317,13 @@ struct net_device  *__dev_map_lookup_elem(struct bpf_map *map, u32 key);
 void __dev_map_insert_ctx(struct bpf_map *map, u32 index);
 void __dev_map_flush(struct bpf_map *map);
 
+/* Return map's numa specified by userspace */
+static inline int bpf_map_attr_numa_node(const union bpf_attr *attr)
+{
+	return (attr->map_flags & BPF_F_NUMA_NODE) ?
+		attr->numa_node : NUMA_NO_NODE;
+}
+
 #else
 static inline struct bpf_prog *bpf_prog_get(u32 ufd)
 {

Diff for: include/uapi/linux/bpf.h

@@ -165,6 +165,7 @@ enum bpf_attach_type {
 #define BPF_NOEXIST	1 /* create new element if it didn't exist */
 #define BPF_EXIST	2 /* update existing element */
 
+/* flags for BPF_MAP_CREATE command */
 #define BPF_F_NO_PREALLOC	(1U << 0)
 /* Instead of having one common LRU list in the
  * BPF_MAP_TYPE_LRU_[PERCPU_]HASH map, use a percpu LRU list
@@ -173,15 +174,22 @@ enum bpf_attach_type {
  * across different LRU lists.
  */
 #define BPF_F_NO_COMMON_LRU	(1U << 1)
+/* Specify numa node during map creation */
+#define BPF_F_NUMA_NODE		(1U << 2)
 
 union bpf_attr {
 	struct { /* anonymous struct used by BPF_MAP_CREATE command */
 		__u32	map_type;	/* one of enum bpf_map_type */
 		__u32	key_size;	/* size of key in bytes */
 		__u32	value_size;	/* size of value in bytes */
 		__u32	max_entries;	/* max number of entries in a map */
-		__u32	map_flags;	/* prealloc or not */
+		__u32	map_flags;	/* BPF_MAP_CREATE related
+					 * flags defined above.
+					 */
 		__u32	inner_map_fd;	/* fd pointing to the inner map */
+		__u32	numa_node;	/* numa node (effective only if
+					 * BPF_F_NUMA_NODE is set).
+					 */
 	};
 
 	struct { /* anonymous struct used by BPF_MAP_*_ELEM commands */

Diff for: kernel/bpf/arraymap.c

@@ -49,13 +49,15 @@ static int bpf_array_alloc_percpu(struct bpf_array *array)
 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
 {
 	bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
+	int numa_node = bpf_map_attr_numa_node(attr);
 	struct bpf_array *array;
 	u64 array_size;
 	u32 elem_size;
 
 	/* check sanity of attributes */
 	if (attr->max_entries == 0 || attr->key_size != 4 ||
-	    attr->value_size == 0 || attr->map_flags)
+	    attr->value_size == 0 || attr->map_flags & ~BPF_F_NUMA_NODE ||
+	    (percpu && numa_node != NUMA_NO_NODE))
 		return ERR_PTR(-EINVAL);
 
 	if (attr->value_size > KMALLOC_MAX_SIZE)
@@ -77,7 +79,7 @@ static struct bpf_map *array_map_alloc(union bpf_attr *attr)
 		return ERR_PTR(-ENOMEM);
 
 	/* allocate all map elements and zero-initialize them */
-	array = bpf_map_area_alloc(array_size);
+	array = bpf_map_area_alloc(array_size, numa_node);
 	if (!array)
 		return ERR_PTR(-ENOMEM);
 
@@ -87,6 +89,7 @@ static struct bpf_map *array_map_alloc(union bpf_attr *attr)
 	array->map.value_size = attr->value_size;
 	array->map.max_entries = attr->max_entries;
 	array->map.map_flags = attr->map_flags;
+	array->map.numa_node = numa_node;
 	array->elem_size = elem_size;
 
 	if (!percpu)

Diff for: kernel/bpf/devmap.c

@@ -80,7 +80,7 @@ static struct bpf_map *dev_map_alloc(union bpf_attr *attr)
 
 	/* check sanity of attributes */
 	if (attr->max_entries == 0 || attr->key_size != 4 ||
-	    attr->value_size != 4 || attr->map_flags)
+	    attr->value_size != 4 || attr->map_flags & ~BPF_F_NUMA_NODE)
 		return ERR_PTR(-EINVAL);
 
 	dtab = kzalloc(sizeof(*dtab), GFP_USER);
@@ -93,6 +93,7 @@ static struct bpf_map *dev_map_alloc(union bpf_attr *attr)
 	dtab->map.value_size = attr->value_size;
 	dtab->map.max_entries = attr->max_entries;
 	dtab->map.map_flags = attr->map_flags;
+	dtab->map.numa_node = bpf_map_attr_numa_node(attr);
 
 	err = -ENOMEM;
 
@@ -119,7 +120,8 @@ static struct bpf_map *dev_map_alloc(union bpf_attr *attr)
 		goto free_dtab;
 
 	dtab->netdev_map = bpf_map_area_alloc(dtab->map.max_entries *
-					      sizeof(struct bpf_dtab_netdev *));
+					      sizeof(struct bpf_dtab_netdev *),
+					      dtab->map.numa_node);
 	if (!dtab->netdev_map)
 		goto free_dtab;
 
@@ -344,7 +346,8 @@ static int dev_map_update_elem(struct bpf_map *map, void *key, void *value,
 	if (!ifindex) {
 		dev = NULL;
 	} else {
-		dev = kmalloc(sizeof(*dev), GFP_ATOMIC | __GFP_NOWARN);
+		dev = kmalloc_node(sizeof(*dev), GFP_ATOMIC | __GFP_NOWARN,
+				   map->numa_node);
 		if (!dev)
 			return -ENOMEM;
 

Diff for: kernel/bpf/hashtab.c

@@ -18,6 +18,9 @@
 #include "bpf_lru_list.h"
 #include "map_in_map.h"
 
+#define HTAB_CREATE_FLAG_MASK \
+	(BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU | BPF_F_NUMA_NODE)
+
 struct bucket {
 	struct hlist_nulls_head head;
 	raw_spinlock_t lock;
@@ -138,7 +141,8 @@ static int prealloc_init(struct bpf_htab *htab)
 	if (!htab_is_percpu(htab) && !htab_is_lru(htab))
 		num_entries += num_possible_cpus();
 
-	htab->elems = bpf_map_area_alloc(htab->elem_size * num_entries);
+	htab->elems = bpf_map_area_alloc(htab->elem_size * num_entries,
+					 htab->map.numa_node);
 	if (!htab->elems)
 		return -ENOMEM;
 
@@ -233,6 +237,7 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 	 */
 	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
 	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
+	int numa_node = bpf_map_attr_numa_node(attr);
 	struct bpf_htab *htab;
 	int err, i;
 	u64 cost;
@@ -248,7 +253,7 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 		 */
 		return ERR_PTR(-EPERM);
 
-	if (attr->map_flags & ~(BPF_F_NO_PREALLOC | BPF_F_NO_COMMON_LRU))
+	if (attr->map_flags & ~HTAB_CREATE_FLAG_MASK)
 		/* reserved bits should not be used */
 		return ERR_PTR(-EINVAL);
 
@@ -258,6 +263,9 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 	if (lru && !prealloc)
 		return ERR_PTR(-ENOTSUPP);
 
+	if (numa_node != NUMA_NO_NODE && (percpu || percpu_lru))
+		return ERR_PTR(-EINVAL);
+
 	htab = kzalloc(sizeof(*htab), GFP_USER);
 	if (!htab)
 		return ERR_PTR(-ENOMEM);
@@ -268,6 +276,7 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 	htab->map.value_size = attr->value_size;
 	htab->map.max_entries = attr->max_entries;
 	htab->map.map_flags = attr->map_flags;
+	htab->map.numa_node = numa_node;
 
 	/* check sanity of attributes.
 	 * value_size == 0 may be allowed in the future to use map as a set
@@ -346,7 +355,8 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
 
 	err = -ENOMEM;
 	htab->buckets = bpf_map_area_alloc(htab->n_buckets *
-					   sizeof(struct bucket));
+					   sizeof(struct bucket),
+					   htab->map.numa_node);
 	if (!htab->buckets)
 		goto free_htab;
 
@@ -689,7 +699,8 @@ static struct htab_elem *alloc_htab_elem(struct bpf_htab *htab, void *key,
 				atomic_dec(&htab->count);
 				return ERR_PTR(-E2BIG);
 			}
-		l_new = kmalloc(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN);
+		l_new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
+				     htab->map.numa_node);
 		if (!l_new)
 			return ERR_PTR(-ENOMEM);
 	}

Diff for: kernel/bpf/lpm_trie.c

@@ -244,7 +244,8 @@ static struct lpm_trie_node *lpm_trie_node_alloc(const struct lpm_trie *trie,
 	if (value)
 		size += trie->map.value_size;
 
-	node = kmalloc(size, GFP_ATOMIC | __GFP_NOWARN);
+	node = kmalloc_node(size, GFP_ATOMIC | __GFP_NOWARN,
+			    trie->map.numa_node);
 	if (!node)
 		return NULL;
 
@@ -405,6 +406,8 @@ static int trie_delete_elem(struct bpf_map *map, void *key)
 #define LPM_KEY_SIZE_MAX	LPM_KEY_SIZE(LPM_DATA_SIZE_MAX)
 #define LPM_KEY_SIZE_MIN	LPM_KEY_SIZE(LPM_DATA_SIZE_MIN)
 
+#define LPM_CREATE_FLAG_MASK	(BPF_F_NO_PREALLOC | BPF_F_NUMA_NODE)
+
 static struct bpf_map *trie_alloc(union bpf_attr *attr)
 {
 	struct lpm_trie *trie;
@@ -416,7 +419,8 @@ static struct bpf_map *trie_alloc(union bpf_attr *attr)
 
 	/* check sanity of attributes */
 	if (attr->max_entries == 0 ||
-	    attr->map_flags != BPF_F_NO_PREALLOC ||
+	    !(attr->map_flags & BPF_F_NO_PREALLOC) ||
+	    attr->map_flags & ~LPM_CREATE_FLAG_MASK ||
 	    attr->key_size < LPM_KEY_SIZE_MIN ||
 	    attr->key_size > LPM_KEY_SIZE_MAX ||
 	    attr->value_size < LPM_VAL_SIZE_MIN ||
@@ -433,6 +437,7 @@ static struct bpf_map *trie_alloc(union bpf_attr *attr)
 	trie->map.value_size = attr->value_size;
 	trie->map.max_entries = attr->max_entries;
 	trie->map.map_flags = attr->map_flags;
+	trie->map.numa_node = bpf_map_attr_numa_node(attr);
 	trie->data_size = attr->key_size -
 			  offsetof(struct bpf_lpm_trie_key, data);
 	trie->max_prefixlen = trie->data_size * 8;

Diff for: kernel/bpf/sockmap.c

@@ -443,7 +443,9 @@ static struct smap_psock *smap_init_psock(struct sock *sock,
 {
 	struct smap_psock *psock;
 
-	psock = kzalloc(sizeof(struct smap_psock), GFP_ATOMIC | __GFP_NOWARN);
+	psock = kzalloc_node(sizeof(struct smap_psock),
+			     GFP_ATOMIC | __GFP_NOWARN,
+			     stab->map.numa_node);
 	if (!psock)
 		return ERR_PTR(-ENOMEM);
 
@@ -465,7 +467,7 @@ static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
 
 	/* check sanity of attributes */
 	if (attr->max_entries == 0 || attr->key_size != 4 ||
-	    attr->value_size != 4 || attr->map_flags)
+	    attr->value_size != 4 || attr->map_flags & ~BPF_F_NUMA_NODE)
 		return ERR_PTR(-EINVAL);
 
 	if (attr->value_size > KMALLOC_MAX_SIZE)
@@ -481,6 +483,7 @@ static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
 	stab->map.value_size = attr->value_size;
 	stab->map.max_entries = attr->max_entries;
 	stab->map.map_flags = attr->map_flags;
+	stab->map.numa_node = bpf_map_attr_numa_node(attr);
 
 	/* make sure page count doesn't overflow */
 	cost = (u64) stab->map.max_entries * sizeof(struct sock *);
@@ -495,7 +498,8 @@ static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
 		goto free_stab;
 
 	stab->sock_map = bpf_map_area_alloc(stab->map.max_entries *
-					    sizeof(struct sock *));
+					    sizeof(struct sock *),
+					    stab->map.numa_node);
 	if (!stab->sock_map)
 		goto free_stab;
 

Diff for: kernel/bpf/stackmap.c

@@ -31,7 +31,8 @@ static int prealloc_elems_and_freelist(struct bpf_stack_map *smap)
 	u32 elem_size = sizeof(struct stack_map_bucket) + smap->map.value_size;
 	int err;
 
-	smap->elems = bpf_map_area_alloc(elem_size * smap->map.max_entries);
+	smap->elems = bpf_map_area_alloc(elem_size * smap->map.max_entries,
+					 smap->map.numa_node);
 	if (!smap->elems)
 		return -ENOMEM;
 
@@ -59,7 +60,7 @@ static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
 	if (!capable(CAP_SYS_ADMIN))
 		return ERR_PTR(-EPERM);
 
-	if (attr->map_flags)
+	if (attr->map_flags & ~BPF_F_NUMA_NODE)
 		return ERR_PTR(-EINVAL);
 
 	/* check sanity of attributes */
@@ -75,7 +76,7 @@ static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
 	if (cost >= U32_MAX - PAGE_SIZE)
 		return ERR_PTR(-E2BIG);
 
-	smap = bpf_map_area_alloc(cost);
+	smap = bpf_map_area_alloc(cost, bpf_map_attr_numa_node(attr));
 	if (!smap)
 		return ERR_PTR(-ENOMEM);
 
@@ -91,6 +92,7 @@ static struct bpf_map *stack_map_alloc(union bpf_attr *attr)
 	smap->map.map_flags = attr->map_flags;
 	smap->n_buckets = n_buckets;
 	smap->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
+	smap->map.numa_node = bpf_map_attr_numa_node(attr);
 
 	err = bpf_map_precharge_memlock(smap->map.pages);
 	if (err)

Diff for: kernel/bpf/syscall.c

@@ -105,7 +105,7 @@ static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 	return map;
 }
 
-void *bpf_map_area_alloc(size_t size)
+void *bpf_map_area_alloc(size_t size, int numa_node)
 {
 	/* We definitely need __GFP_NORETRY, so OOM killer doesn't
 	 * trigger under memory pressure as we really just want to
@@ -115,12 +115,13 @@ void *bpf_map_area_alloc(size_t size)
 	void *area;
 
 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
-		area = kmalloc(size, GFP_USER | flags);
+		area = kmalloc_node(size, GFP_USER | flags, numa_node);
 		if (area != NULL)
 			return area;
 	}
 
-	return __vmalloc(size, GFP_KERNEL | flags, PAGE_KERNEL);
+	return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
+					   __builtin_return_address(0));
 }
 
 void bpf_map_area_free(void *area)
@@ -309,17 +310,22 @@ int bpf_map_new_fd(struct bpf_map *map)
 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
 
-#define BPF_MAP_CREATE_LAST_FIELD inner_map_fd
+#define BPF_MAP_CREATE_LAST_FIELD numa_node
 /* called via syscall */
 static int map_create(union bpf_attr *attr)
 {
+	int numa_node = bpf_map_attr_numa_node(attr);
 	struct bpf_map *map;
 	int err;
 
 	err = CHECK_ATTR(BPF_MAP_CREATE);
 	if (err)
 		return -EINVAL;
 
+	if (numa_node != NUMA_NO_NODE &&
+	    (numa_node >= nr_node_ids || !node_online(numa_node)))
+		return -EINVAL;
+
 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 	map = find_and_alloc_map(attr);
 	if (IS_ERR(map))