Merge branch 'bpf-btf-id'

Martin KaFai Lau says:

====================
This series introduces BTF ID which is exposed through
the new BPF_BTF_GET_FD_BY_ID cmd, new "struct bpf_btf_info"
and new members in the "struct bpf_map_info".

Please see individual patch for details.
====================
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

include/linux/btf.h

@@ -44,5 +44,7 @@ const struct btf_type *btf_type_id_size(const struct btf *btf,
 					u32 *ret_size);
 void btf_type_seq_show(const struct btf *btf, u32 type_id, void *obj,
 		       struct seq_file *m);
+int btf_get_fd_by_id(u32 id);
+u32 btf_id(const struct btf *btf);
 
 #endif

include/uapi/linux/bpf.h

@@ -96,6 +96,7 @@ enum bpf_cmd {
 	BPF_PROG_QUERY,
 	BPF_RAW_TRACEPOINT_OPEN,
 	BPF_BTF_LOAD,
+	BPF_BTF_GET_FD_BY_ID,
 };
 
 enum bpf_map_type {
@@ -344,6 +345,7 @@ union bpf_attr {
 			__u32		start_id;
 			__u32		prog_id;
 			__u32		map_id;
+			__u32		btf_id;
 		};
 		__u32		next_id;
 		__u32		open_flags;
@@ -2130,6 +2132,15 @@ struct bpf_map_info {
 	__u32 ifindex;
 	__u64 netns_dev;
 	__u64 netns_ino;
+	__u32 btf_id;
+	__u32 btf_key_id;
+	__u32 btf_value_id;
+} __attribute__((aligned(8)));
+
+struct bpf_btf_info {
+	__aligned_u64 btf;
+	__u32 btf_size;
+	__u32 id;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

kernel/bpf/btf.c

@@ -11,6 +11,7 @@
 #include <linux/file.h>
 #include <linux/uaccess.h>
 #include <linux/kernel.h>
+#include <linux/idr.h>
 #include <linux/bpf_verifier.h>
 #include <linux/btf.h>
 
@@ -179,6 +180,9 @@
 	     i < btf_type_vlen(struct_type);				\
 	     i++, member++)
 
+static DEFINE_IDR(btf_idr);
+static DEFINE_SPINLOCK(btf_idr_lock);
+
 struct btf {
 	union {
 		struct btf_header *hdr;
@@ -193,6 +197,8 @@ struct btf {
 	u32 types_size;
 	u32 data_size;
 	refcount_t refcnt;
+	u32 id;
+	struct rcu_head rcu;
 };
 
 enum verifier_phase {
@@ -598,6 +604,42 @@ static int btf_add_type(struct btf_verifier_env *env, struct btf_type *t)
 	return 0;
 }
 
+static int btf_alloc_id(struct btf *btf)
+{
+	int id;
+
+	idr_preload(GFP_KERNEL);
+	spin_lock_bh(&btf_idr_lock);
+	id = idr_alloc_cyclic(&btf_idr, btf, 1, INT_MAX, GFP_ATOMIC);
+	if (id > 0)
+		btf->id = id;
+	spin_unlock_bh(&btf_idr_lock);
+	idr_preload_end();
+
+	if (WARN_ON_ONCE(!id))
+		return -ENOSPC;
+
+	return id > 0 ? 0 : id;
+}
+
+static void btf_free_id(struct btf *btf)
+{
+	unsigned long flags;
+
+	/*
+	 * In map-in-map, calling map_delete_elem() on outer
+	 * map will call bpf_map_put on the inner map.
+	 * It will then eventually call btf_free_id()
+	 * on the inner map.  Some of the map_delete_elem()
+	 * implementation may have irq disabled, so
+	 * we need to use the _irqsave() version instead
+	 * of the _bh() version.
+	 */
+	spin_lock_irqsave(&btf_idr_lock, flags);
+	idr_remove(&btf_idr, btf->id);
+	spin_unlock_irqrestore(&btf_idr_lock, flags);
+}
+
 static void btf_free(struct btf *btf)
 {
 	kvfree(btf->types);
@@ -607,15 +649,19 @@ static void btf_free(struct btf *btf)
 	kfree(btf);
 }
 
-static void btf_get(struct btf *btf)
+static void btf_free_rcu(struct rcu_head *rcu)
 {
-	refcount_inc(&btf->refcnt);
+	struct btf *btf = container_of(rcu, struct btf, rcu);
+
+	btf_free(btf);
 }
 
 void btf_put(struct btf *btf)
 {
-	if (btf && refcount_dec_and_test(&btf->refcnt))
-		btf_free(btf);
+	if (btf && refcount_dec_and_test(&btf->refcnt)) {
+		btf_free_id(btf);
+		call_rcu(&btf->rcu, btf_free_rcu);
+	}
 }
 
 static int env_resolve_init(struct btf_verifier_env *env)
@@ -1977,7 +2023,7 @@ static struct btf *btf_parse(void __user *btf_data, u32 btf_data_size,
 
 	if (!err) {
 		btf_verifier_env_free(env);
-		btf_get(btf);
+		refcount_set(&btf->refcnt, 1);
 		return btf;
 	}
 
@@ -2006,10 +2052,15 @@ const struct file_operations btf_fops = {
 	.release	= btf_release,
 };
 
+static int __btf_new_fd(struct btf *btf)
+{
+	return anon_inode_getfd("btf", &btf_fops, btf, O_RDONLY | O_CLOEXEC);
+}
+
 int btf_new_fd(const union bpf_attr *attr)
 {
 	struct btf *btf;
-	int fd;
+	int ret;
 
 	btf = btf_parse(u64_to_user_ptr(attr->btf),
 			attr->btf_size, attr->btf_log_level,
@@ -2018,12 +2069,23 @@ int btf_new_fd(const union bpf_attr *attr)
 	if (IS_ERR(btf))
 		return PTR_ERR(btf);
 
-	fd = anon_inode_getfd("btf", &btf_fops, btf,
-			      O_RDONLY | O_CLOEXEC);
-	if (fd < 0)
+	ret = btf_alloc_id(btf);
+	if (ret) {
+		btf_free(btf);
+		return ret;
+	}
+
+	/*
+	 * The BTF ID is published to the userspace.
+	 * All BTF free must go through call_rcu() from
+	 * now on (i.e. free by calling btf_put()).
+	 */
+
+	ret = __btf_new_fd(btf);
+	if (ret < 0)
 		btf_put(btf);
 
-	return fd;
+	return ret;
 }
 
 struct btf *btf_get_by_fd(int fd)
@@ -2042,7 +2104,7 @@ struct btf *btf_get_by_fd(int fd)
 	}
 
 	btf = f.file->private_data;
-	btf_get(btf);
+	refcount_inc(&btf->refcnt);
 	fdput(f);
 
 	return btf;
@@ -2052,13 +2114,55 @@ int btf_get_info_by_fd(const struct btf *btf,
 		       const union bpf_attr *attr,
 		       union bpf_attr __user *uattr)
 {
-	void __user *udata = u64_to_user_ptr(attr->info.info);
-	u32 copy_len = min_t(u32, btf->data_size,
-			     attr->info.info_len);
+	struct bpf_btf_info __user *uinfo;
+	struct bpf_btf_info info = {};
+	u32 info_copy, btf_copy;
+	void __user *ubtf;
+	u32 uinfo_len;
+
+	uinfo = u64_to_user_ptr(attr->info.info);
+	uinfo_len = attr->info.info_len;
 
-	if (copy_to_user(udata, btf->data, copy_len) ||
-	    put_user(btf->data_size, &uattr->info.info_len))
+	info_copy = min_t(u32, uinfo_len, sizeof(info));
+	if (copy_from_user(&info, uinfo, info_copy))
+		return -EFAULT;
+
+	info.id = btf->id;
+	ubtf = u64_to_user_ptr(info.btf);
+	btf_copy = min_t(u32, btf->data_size, info.btf_size);
+	if (copy_to_user(ubtf, btf->data, btf_copy))
+		return -EFAULT;
+	info.btf_size = btf->data_size;
+
+	if (copy_to_user(uinfo, &info, info_copy) ||
+	    put_user(info_copy, &uattr->info.info_len))
 		return -EFAULT;
 
 	return 0;
 }
+
+int btf_get_fd_by_id(u32 id)
+{
+	struct btf *btf;
+	int fd;
+
+	rcu_read_lock();
+	btf = idr_find(&btf_idr, id);
+	if (!btf || !refcount_inc_not_zero(&btf->refcnt))
+		btf = ERR_PTR(-ENOENT);
+	rcu_read_unlock();
+
+	if (IS_ERR(btf))
+		return PTR_ERR(btf);
+
+	fd = __btf_new_fd(btf);
+	if (fd < 0)
+		btf_put(btf);
+
+	return fd;
+}
+
+u32 btf_id(const struct btf *btf)
+{
+	return btf->id;
+}

kernel/bpf/syscall.c

@@ -252,7 +252,6 @@ static void bpf_map_free_deferred(struct work_struct *work)
 
 	bpf_map_uncharge_memlock(map);
 	security_bpf_map_free(map);
-	btf_put(map->btf);
 	/* implementation dependent freeing */
 	map->ops->map_free(map);
 }
@@ -273,6 +272,7 @@ static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 	if (atomic_dec_and_test(&map->refcnt)) {
 		/* bpf_map_free_id() must be called first */
 		bpf_map_free_id(map, do_idr_lock);
+		btf_put(map->btf);
 		INIT_WORK(&map->work, bpf_map_free_deferred);
 		schedule_work(&map->work);
 	}
@@ -2002,6 +2002,12 @@ static int bpf_map_get_info_by_fd(struct bpf_map *map,
 	info.map_flags = map->map_flags;
 	memcpy(info.name, map->name, sizeof(map->name));
 
+	if (map->btf) {
+		info.btf_id = btf_id(map->btf);
+		info.btf_key_id = map->btf_key_id;
+		info.btf_value_id = map->btf_value_id;
+	}
+
 	if (bpf_map_is_dev_bound(map)) {
 		err = bpf_map_offload_info_fill(&info, map);
 		if (err)
@@ -2015,6 +2021,21 @@ static int bpf_map_get_info_by_fd(struct bpf_map *map,
 	return 0;
 }
 
+static int bpf_btf_get_info_by_fd(struct btf *btf,
+				  const union bpf_attr *attr,
+				  union bpf_attr __user *uattr)
+{
+	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
+	u32 info_len = attr->info.info_len;
+	int err;
+
+	err = check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
+	if (err)
+		return err;
+
+	return btf_get_info_by_fd(btf, attr, uattr);
+}
+
 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
 
 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
@@ -2038,7 +2059,7 @@ static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
 					     uattr);
 	else if (f.file->f_op == &btf_fops)
-		err = btf_get_info_by_fd(f.file->private_data, attr, uattr);
+		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
 	else
 		err = -EINVAL;
 
@@ -2059,6 +2080,19 @@ static int bpf_btf_load(const union bpf_attr *attr)
 	return btf_new_fd(attr);
 }
 
+#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
+
+static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
+{
+	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
+		return -EINVAL;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	return btf_get_fd_by_id(attr->btf_id);
+}
+
 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
 {
 	union bpf_attr attr = {};
@@ -2142,6 +2176,9 @@ SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, siz
 	case BPF_BTF_LOAD:
 		err = bpf_btf_load(&attr);
 		break;
+	case BPF_BTF_GET_FD_BY_ID:
+		err = bpf_btf_get_fd_by_id(&attr);
+		break;
 	default:
 		err = -EINVAL;
 		break;

tools/include/uapi/linux/bpf.h

@@ -96,6 +96,7 @@ enum bpf_cmd {
 	BPF_PROG_QUERY,
 	BPF_RAW_TRACEPOINT_OPEN,
 	BPF_BTF_LOAD,
+	BPF_BTF_GET_FD_BY_ID,
 };
 
 enum bpf_map_type {
@@ -343,6 +344,7 @@ union bpf_attr {
 			__u32		start_id;
 			__u32		prog_id;
 			__u32		map_id;
+			__u32		btf_id;
 		};
 		__u32		next_id;
 		__u32		open_flags;
@@ -2129,6 +2131,15 @@ struct bpf_map_info {
 	__u32 ifindex;
 	__u64 netns_dev;
 	__u64 netns_ino;
+	__u32 btf_id;
+	__u32 btf_key_id;
+	__u32 btf_value_id;
+} __attribute__((aligned(8)));
+
+struct bpf_btf_info {
+	__aligned_u64 btf;
+	__u32 btf_size;
+	__u32 id;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

tools/lib/bpf/bpf.c

@@ -458,6 +458,16 @@ int bpf_map_get_fd_by_id(__u32 id)
 	return sys_bpf(BPF_MAP_GET_FD_BY_ID, &attr, sizeof(attr));
 }
 
+int bpf_btf_get_fd_by_id(__u32 id)
+{
+	union bpf_attr attr;
+
+	bzero(&attr, sizeof(attr));
+	attr.btf_id = id;
+
+	return sys_bpf(BPF_BTF_GET_FD_BY_ID, &attr, sizeof(attr));
+}
+
 int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len)
 {
 	union bpf_attr attr;

tools/lib/bpf/bpf.h

@@ -98,6 +98,7 @@ int bpf_prog_get_next_id(__u32 start_id, __u32 *next_id);
 int bpf_map_get_next_id(__u32 start_id, __u32 *next_id);
 int bpf_prog_get_fd_by_id(__u32 id);
 int bpf_map_get_fd_by_id(__u32 id);
+int bpf_btf_get_fd_by_id(__u32 id);
 int bpf_obj_get_info_by_fd(int prog_fd, void *info, __u32 *info_len);
 int bpf_prog_query(int target_fd, enum bpf_attach_type type, __u32 query_flags,
 		   __u32 *attach_flags, __u32 *prog_ids, __u32 *prog_cnt);