libceph, ceph: implement msgr2.1 protocol (crc and secure modes)

Implement msgr2.1 wire protocol, available since nautilus 14.2.11
and octopus 15.2.5.  msgr2.0 wire protocol is not implemented -- it
has several security, integrity and robustness issues and therefore
considered deprecated.

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

6 files changed, 231 insertions, 6 deletions

diff --git a/include/linux/ceph/auth.h b/include/linux/ceph/auth.h
index 6fc058fe9efa..3fbe72ebd779 100644
--- a/include/linux/ceph/auth.h
+++ b/include/linux/ceph/auth.h
@@ -120,8 +120,12 @@ int ceph_auth_entity_name_encode(const char *name, void **p, void *end);
 
 extern int ceph_build_auth(struct ceph_auth_client *ac,
 		    void *msg_buf, size_t msg_len);
-
 extern int ceph_auth_is_authenticated(struct ceph_auth_client *ac);
+
+int __ceph_auth_get_authorizer(struct ceph_auth_client *ac,
+			       struct ceph_auth_handshake *auth,
+			       int peer_type, bool force_new,
+			       int *proto, int *pref_mode, int *fallb_mode);
 extern int ceph_auth_create_authorizer(struct ceph_auth_client *ac,
 				       int peer_type,
 				       struct ceph_auth_handshake *auth);
@@ -157,4 +161,34 @@ int ceph_auth_check_message_signature(struct ceph_auth_handshake *auth,
 		return auth->check_message_signature(auth, msg);
 	return 0;
 }
+
+int ceph_auth_get_request(struct ceph_auth_client *ac, void *buf, int buf_len);
+int ceph_auth_handle_reply_more(struct ceph_auth_client *ac, void *reply,
+				int reply_len, void *buf, int buf_len);
+int ceph_auth_handle_reply_done(struct ceph_auth_client *ac,
+				u64 global_id, void *reply, int reply_len,
+				u8 *session_key, int *session_key_len,
+				u8 *con_secret, int *con_secret_len);
+bool ceph_auth_handle_bad_method(struct ceph_auth_client *ac,
+				 int used_proto, int result,
+				 const int *allowed_protos, int proto_cnt,
+				 const int *allowed_modes, int mode_cnt);
+
+int ceph_auth_get_authorizer(struct ceph_auth_client *ac,
+			     struct ceph_auth_handshake *auth,
+			     int peer_type, void *buf, int *buf_len);
+int ceph_auth_handle_svc_reply_more(struct ceph_auth_client *ac,
+				    struct ceph_auth_handshake *auth,
+				    void *reply, int reply_len,
+				    void *buf, int *buf_len);
+int ceph_auth_handle_svc_reply_done(struct ceph_auth_client *ac,
+				    struct ceph_auth_handshake *auth,
+				    void *reply, int reply_len,
+				    u8 *session_key, int *session_key_len,
+				    u8 *con_secret, int *con_secret_len);
+bool ceph_auth_handle_bad_authorizer(struct ceph_auth_client *ac,
+				     int peer_type, int used_proto, int result,
+				     const int *allowed_protos, int proto_cnt,
+				     const int *allowed_modes, int mode_cnt);
+
 #endif
diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
index ce22d5469670..e41a811026f6 100644
--- a/include/linux/ceph/ceph_fs.h
+++ b/include/linux/ceph/ceph_fs.h
@@ -93,6 +93,10 @@ struct ceph_dir_layout {
 #define CEPH_AUTH_NONE	 	0x1
 #define CEPH_AUTH_CEPHX	 	0x2
 
+#define CEPH_AUTH_MODE_NONE		0
+#define CEPH_AUTH_MODE_AUTHORIZER	1
+#define CEPH_AUTH_MODE_MON		10
+
 /* msgr2 protocol modes */
 #define CEPH_CON_MODE_UNKNOWN	0x0
 #define CEPH_CON_MODE_CRC	0x1
diff --git a/include/linux/ceph/decode.h b/include/linux/ceph/decode.h
index 9a934e04f841..04f3ace5787b 100644
--- a/include/linux/ceph/decode.h
+++ b/include/linux/ceph/decode.h
@@ -221,6 +221,7 @@ static inline void ceph_encode_timespec64(struct ceph_timespec *tv,
 #define CEPH_ENTITY_ADDR_TYPE_NONE	0
 #define CEPH_ENTITY_ADDR_TYPE_LEGACY	__cpu_to_le32(1)
 #define CEPH_ENTITY_ADDR_TYPE_MSGR2	__cpu_to_le32(2)
+#define CEPH_ENTITY_ADDR_TYPE_ANY	__cpu_to_le32(3)
 
 static inline void ceph_encode_banner_addr(struct ceph_entity_addr *a)
 {
@@ -243,6 +244,9 @@ extern int ceph_decode_entity_addr(void **p, void *end,
 int ceph_decode_entity_addrvec(void **p, void *end, bool msgr2,
 			       struct ceph_entity_addr *addr);
 
+int ceph_entity_addr_encoding_len(const struct ceph_entity_addr *addr);
+void ceph_encode_entity_addr(void **p, const struct ceph_entity_addr *addr);
+
 /*
  * encoders
  */
diff --git a/include/linux/ceph/libceph.h b/include/linux/ceph/libceph.h
index 8765a5ad267a..eb9008bb3992 100644
--- a/include/linux/ceph/libceph.h
+++ b/include/linux/ceph/libceph.h
@@ -31,10 +31,10 @@
 #define CEPH_OPT_FSID             (1<<0)
 #define CEPH_OPT_NOSHARE          (1<<1) /* don't share client with other sbs */
 #define CEPH_OPT_MYIP             (1<<2) /* specified my ip */
-#define CEPH_OPT_NOCRC            (1<<3) /* no data crc on writes */
+#define CEPH_OPT_NOCRC            (1<<3) /* no data crc on writes (msgr1) */
 #define CEPH_OPT_NOMSGAUTH	  (1<<4) /* don't require msg signing feat */
 #define CEPH_OPT_TCP_NODELAY	  (1<<5) /* TCP_NODELAY on TCP sockets */
-#define CEPH_OPT_NOMSGSIGN	  (1<<6) /* don't sign msgs */
+#define CEPH_OPT_NOMSGSIGN	  (1<<6) /* don't sign msgs (msgr1) */
 #define CEPH_OPT_ABORT_ON_FULL	  (1<<7) /* abort w/ ENOSPC when full */
 
 #define CEPH_OPT_DEFAULT   (CEPH_OPT_TCP_NODELAY)
@@ -84,6 +84,7 @@ struct ceph_options {
 #define CEPH_MONC_HUNT_BACKOFF		2
 #define CEPH_MONC_HUNT_MAX_MULT		10
 
+#define CEPH_MSG_MAX_CONTROL_LEN (16*1024*1024)
 #define CEPH_MSG_MAX_FRONT_LEN	(16*1024*1024)
 #define CEPH_MSG_MAX_MIDDLE_LEN	(16*1024*1024)
 
@@ -152,6 +153,10 @@ struct ceph_client {
 
 #define from_msgr(ms)	container_of(ms, struct ceph_client, msgr)
 
+static inline bool ceph_msgr2(struct ceph_client *client)
+{
+	return client->options->con_modes[0] != CEPH_CON_MODE_UNKNOWN;
+}
 
 /*
  * snapshots
diff --git a/include/linux/ceph/messenger.h b/include/linux/ceph/messenger.h
index 54a64e8dfce6..0e6e9ad3c3bf 100644
--- a/include/linux/ceph/messenger.h
+++ b/include/linux/ceph/messenger.h
@@ -3,6 +3,7 @@
 #define __FS_CEPH_MESSENGER_H
 
 #include <linux/bvec.h>
+#include <linux/crypto.h>
 #include <linux/kref.h>
 #include <linux/mutex.h>
 #include <linux/net.h>
@@ -52,6 +53,23 @@ struct ceph_connection_operations {
 
 	int (*sign_message) (struct ceph_msg *msg);
 	int (*check_message_signature) (struct ceph_msg *msg);
+
+	/* msgr2 authentication exchange */
+	int (*get_auth_request)(struct ceph_connection *con,
+				void *buf, int *buf_len,
+				void **authorizer, int *authorizer_len);
+	int (*handle_auth_reply_more)(struct ceph_connection *con,
+				      void *reply, int reply_len,
+				      void *buf, int *buf_len,
+				      void **authorizer, int *authorizer_len);
+	int (*handle_auth_done)(struct ceph_connection *con,
+				u64 global_id, void *reply, int reply_len,
+				u8 *session_key, int *session_key_len,
+				u8 *con_secret, int *con_secret_len);
+	int (*handle_auth_bad_method)(struct ceph_connection *con,
+				      int used_proto, int result,
+				      const int *allowed_protos, int proto_cnt,
+				      const int *allowed_modes, int mode_cnt);
 };
 
 /* use format string %s%lld */
@@ -246,8 +264,15 @@ struct ceph_msg {
 #define CEPH_CON_S_PREOPEN		2
 #define CEPH_CON_S_V1_BANNER		3
 #define CEPH_CON_S_V1_CONNECT_MSG	4
-#define CEPH_CON_S_OPEN			5
-#define CEPH_CON_S_STANDBY		6
+#define CEPH_CON_S_V2_BANNER_PREFIX	5
+#define CEPH_CON_S_V2_BANNER_PAYLOAD	6
+#define CEPH_CON_S_V2_HELLO		7
+#define CEPH_CON_S_V2_AUTH		8
+#define CEPH_CON_S_V2_AUTH_SIGNATURE	9
+#define CEPH_CON_S_V2_SESSION_CONNECT	10
+#define CEPH_CON_S_V2_SESSION_RECONNECT	11
+#define CEPH_CON_S_OPEN			12
+#define CEPH_CON_S_STANDBY		13
 
 /*
  * ceph_connection flag bits
@@ -301,6 +326,99 @@ struct ceph_connection_v1_info {
 	u32 peer_global_seq;  /* peer's global seq for this connection */
 };
 
+#define CEPH_CRC_LEN			4
+#define CEPH_GCM_KEY_LEN		16
+#define CEPH_GCM_IV_LEN			sizeof(struct ceph_gcm_nonce)
+#define CEPH_GCM_BLOCK_LEN		16
+#define CEPH_GCM_TAG_LEN		16
+
+#define CEPH_PREAMBLE_LEN		32
+#define CEPH_PREAMBLE_INLINE_LEN	48
+#define CEPH_PREAMBLE_PLAIN_LEN		CEPH_PREAMBLE_LEN
+#define CEPH_PREAMBLE_SECURE_LEN	(CEPH_PREAMBLE_LEN +		\
+					 CEPH_PREAMBLE_INLINE_LEN +	\
+					 CEPH_GCM_TAG_LEN)
+#define CEPH_EPILOGUE_PLAIN_LEN		(1 + 3 * CEPH_CRC_LEN)
+#define CEPH_EPILOGUE_SECURE_LEN	(CEPH_GCM_BLOCK_LEN + CEPH_GCM_TAG_LEN)
+
+#define CEPH_FRAME_MAX_SEGMENT_COUNT	4
+
+struct ceph_frame_desc {
+	int fd_tag;  /* FRAME_TAG_* */
+	int fd_seg_cnt;
+	int fd_lens[CEPH_FRAME_MAX_SEGMENT_COUNT];  /* logical */
+	int fd_aligns[CEPH_FRAME_MAX_SEGMENT_COUNT];
+};
+
+struct ceph_gcm_nonce {
+	__le32 fixed;
+	__le64 counter __packed;
+};
+
+struct ceph_connection_v2_info {
+	struct iov_iter in_iter;
+	struct kvec in_kvecs[5];  /* recvmsg */
+	struct bio_vec in_bvec;  /* recvmsg (in_cursor) */
+	int in_kvec_cnt;
+	int in_state;  /* IN_S_* */
+
+	struct iov_iter out_iter;
+	struct kvec out_kvecs[8];  /* sendmsg */
+	struct bio_vec out_bvec;  /* sendpage (out_cursor, out_zero),
+				     sendmsg (out_enc_pages) */
+	int out_kvec_cnt;
+	int out_state;  /* OUT_S_* */
+
+	int out_zero;  /* # of zero bytes to send */
+	bool out_iter_sendpage;  /* use sendpage if possible */
+
+	struct ceph_frame_desc in_desc;
+	struct ceph_msg_data_cursor in_cursor;
+	struct ceph_msg_data_cursor out_cursor;
+
+	struct crypto_shash *hmac_tfm;  /* post-auth signature */
+	struct crypto_aead *gcm_tfm;  /* on-wire encryption */
+	struct aead_request *gcm_req;
+	struct crypto_wait gcm_wait;
+	struct ceph_gcm_nonce in_gcm_nonce;
+	struct ceph_gcm_nonce out_gcm_nonce;
+
+	struct page **out_enc_pages;
+	int out_enc_page_cnt;
+	int out_enc_resid;
+	int out_enc_i;
+
+	int con_mode;  /* CEPH_CON_MODE_* */
+
+	void *conn_bufs[16];
+	int conn_buf_cnt;
+
+	struct kvec in_sign_kvecs[8];
+	struct kvec out_sign_kvecs[8];
+	int in_sign_kvec_cnt;
+	int out_sign_kvec_cnt;
+
+	u64 client_cookie;
+	u64 server_cookie;
+	u64 global_seq;
+	u64 connect_seq;
+	u64 peer_global_seq;
+
+	u8 in_buf[CEPH_PREAMBLE_SECURE_LEN];
+	u8 out_buf[CEPH_PREAMBLE_SECURE_LEN];
+	struct {
+		u8 late_status;  /* FRAME_LATE_STATUS_* */
+		union {
+			struct {
+				u32 front_crc;
+				u32 middle_crc;
+				u32 data_crc;
+			} __packed;
+			u8 pad[CEPH_GCM_BLOCK_LEN - 1];
+		};
+	} out_epil;
+};
+
 /*
  * A single connection with another host.
  *
@@ -346,7 +464,10 @@ struct ceph_connection {
 	struct delayed_work work;	    /* send|recv work */
 	unsigned long       delay;          /* current delay interval */
 
-	struct ceph_connection_v1_info v1;
+	union {
+		struct ceph_connection_v1_info v1;
+		struct ceph_connection_v2_info v2;
+	};
 };
 
 extern struct page *ceph_zero_page;
@@ -397,6 +518,15 @@ bool ceph_con_v1_opened(struct ceph_connection *con);
 void ceph_con_v1_reset_session(struct ceph_connection *con);
 void ceph_con_v1_reset_protocol(struct ceph_connection *con);
 
+/* messenger_v2.c */
+int ceph_con_v2_try_read(struct ceph_connection *con);
+int ceph_con_v2_try_write(struct ceph_connection *con);
+void ceph_con_v2_revoke(struct ceph_connection *con);
+void ceph_con_v2_revoke_incoming(struct ceph_connection *con);
+bool ceph_con_v2_opened(struct ceph_connection *con);
+void ceph_con_v2_reset_session(struct ceph_connection *con);
+void ceph_con_v2_reset_protocol(struct ceph_connection *con);
+
 
 extern const char *ceph_pr_addr(const struct ceph_entity_addr *addr);
 
diff --git a/include/linux/ceph/msgr.h b/include/linux/ceph/msgr.h
index 9a897a60f20b..f5e02f6c0655 100644
--- a/include/linux/ceph/msgr.h
+++ b/include/linux/ceph/msgr.h
@@ -14,10 +14,40 @@
  * constant.
  */
 #define CEPH_BANNER "ceph v027"
+#define CEPH_BANNER_LEN 9
 #define CEPH_BANNER_MAX_LEN 30
 
 
 /*
+ * messenger V2 connection banner prefix.
+ * The full banner string should have the form: "ceph v2\n<le16>"
+ * the 2 bytes are the length of the remaining banner.
+ */
+#define CEPH_BANNER_V2 "ceph v2\n"
+#define CEPH_BANNER_V2_LEN 8
+#define CEPH_BANNER_V2_PREFIX_LEN (CEPH_BANNER_V2_LEN + sizeof(__le16))
+
+/*
+ * messenger V2 features
+ */
+#define CEPH_MSGR2_INCARNATION_1 (0ull)
+
+#define DEFINE_MSGR2_FEATURE(bit, incarnation, name)               \
+	static const uint64_t CEPH_MSGR2_FEATURE_##name = (1ULL << bit); \
+	static const uint64_t CEPH_MSGR2_FEATUREMASK_##name =            \
+			(1ULL << bit | CEPH_MSGR2_INCARNATION_##incarnation);
+
+#define HAVE_MSGR2_FEATURE(x, name) \
+	(((x) & (CEPH_MSGR2_FEATUREMASK_##name)) == (CEPH_MSGR2_FEATUREMASK_##name))
+
+DEFINE_MSGR2_FEATURE( 0, 1, REVISION_1)   // msgr2.1
+
+#define CEPH_MSGR2_SUPPORTED_FEATURES (CEPH_MSGR2_FEATURE_REVISION_1)
+
+#define CEPH_MSGR2_REQUIRED_FEATURES  (CEPH_MSGR2_FEATURE_REVISION_1)
+
+
+/*
  * Rollover-safe type and comparator for 32-bit sequence numbers.
  * Comparator returns -1, 0, or 1.
  */
@@ -158,6 +188,24 @@ struct ceph_msg_header {
 	__le32 crc;       /* header crc32c */
 } __attribute__ ((packed));
 
+struct ceph_msg_header2 {
+	__le64 seq;       /* message seq# for this session */
+	__le64 tid;       /* transaction id */
+	__le16 type;      /* message type */
+	__le16 priority;  /* priority.  higher value == higher priority */
+	__le16 version;   /* version of message encoding */
+
+	__le32 data_pre_padding_len;
+	__le16 data_off;  /* sender: include full offset;
+			     receiver: mask against ~PAGE_MASK */
+
+	__le64 ack_seq;
+	__u8 flags;
+	/* oldest code we think can decode this.  unknown if zero. */
+	__le16 compat_version;
+	__le16 reserved;
+} __attribute__ ((packed));
+
 #define CEPH_MSG_PRIO_LOW     64
 #define CEPH_MSG_PRIO_DEFAULT 127
 #define CEPH_MSG_PRIO_HIGH    196