1 #ifndef __FS_CEPH_MESSENGER_H
2 #define __FS_CEPH_MESSENGER_H
4 #include <linux/kref.h>
5 #include <linux/mutex.h>
7 #include <linux/radix-tree.h>
9 #include <linux/version.h>
10 #include <linux/workqueue.h>
16 struct ceph_connection;
18 extern struct workqueue_struct *ceph_msgr_wq; /* receive work queue */
21 * Ceph defines these callbacks for handling connection events.
23 struct ceph_connection_operations {
24 struct ceph_connection *(*get)(struct ceph_connection *);
25 void (*put)(struct ceph_connection *);
27 /* handle an incoming message. */
28 void (*dispatch) (struct ceph_connection *con, struct ceph_msg *m);
30 /* authorize an outgoing connection */
31 int (*get_authorizer) (struct ceph_connection *con,
32 void **buf, int *len, int *proto,
33 void **reply_buf, int *reply_len, int force_new);
34 int (*verify_authorizer_reply) (struct ceph_connection *con, int len);
36 /* protocol version mismatch */
37 void (*bad_proto) (struct ceph_connection *con);
39 /* there was some error on the socket (disconnect, whatever) */
40 void (*fault) (struct ceph_connection *con);
42 /* a remote host as terminated a message exchange session, and messages
43 * we sent (or they tried to send us) may be lost. */
44 void (*peer_reset) (struct ceph_connection *con);
46 struct ceph_msg * (*alloc_msg) (struct ceph_connection *con,
47 struct ceph_msg_header *hdr);
48 int (*alloc_middle) (struct ceph_connection *con,
49 struct ceph_msg *msg);
50 /* an incoming message has a data payload; tell me what pages I
51 * should read the data into. */
52 int (*prepare_pages) (struct ceph_connection *con, struct ceph_msg *m,
56 extern const char *ceph_name_type_str(int t);
58 /* use format string %s%d */
59 #define ENTITY_NAME(n) ceph_name_type_str((n).type), le64_to_cpu((n).num)
61 struct ceph_messenger {
62 struct ceph_entity_inst inst; /* my name+address */
63 struct ceph_entity_addr my_enc_addr;
64 struct page *zero_page; /* used in certain error cases */
69 * the global_seq counts connections i (attempt to) initiate
70 * in order to disambiguate certain connect race conditions.
73 spinlock_t global_seq_lock;
77 * a single message. it contains a header (src, dest, message type, etc.),
78 * footer (crc values, mainly), a "front" message body, and possibly a
79 * data payload (stored in some number of pages).
82 struct ceph_msg_header hdr; /* header */
83 struct ceph_msg_footer footer; /* footer */
84 struct kvec front; /* unaligned blobs of message */
85 struct ceph_buffer *middle;
86 struct page **pages; /* data payload. NOT OWNER. */
87 unsigned nr_pages; /* size of page array */
88 struct list_head list_head;
90 bool front_is_vmalloc;
94 struct ceph_msgpool *pool;
98 int page, page_pos; /* which page; offset in page */
99 int data_pos; /* offset in data payload */
100 int did_page_crc; /* true if we've calculated crc for current page */
103 /* ceph connection fault delay defaults, for exponential backoff */
104 #define BASE_DELAY_INTERVAL (HZ/2)
105 #define MAX_DELAY_INTERVAL (5 * 60 * HZ)
108 * ceph_connection state bit flags
110 * QUEUED and BUSY are used together to ensure that only a single
111 * thread is currently opening, reading or writing data to the socket.
113 #define LOSSYTX 0 /* we can close channel or drop messages on errors */
115 #define NEGOTIATING 2
116 #define KEEPALIVE_PENDING 3
117 #define WRITE_PENDING 4 /* we have data ready to send */
118 #define QUEUED 5 /* there is work queued on this connection */
119 #define BUSY 6 /* work is being done */
120 #define STANDBY 8 /* no outgoing messages, socket closed. we keep
121 * the ceph_connection around to maintain shared
122 * state with the peer. */
123 #define CLOSED 10 /* we've closed the connection */
124 #define SOCK_CLOSED 11 /* socket state changed to closed */
125 #define REGISTERED 12 /* connection appears in con_tree */
126 #define OPENING 13 /* open connection w/ (possibly new) peer */
127 #define DEAD 14 /* dead, about to kfree */
130 * A single connection with another host.
132 * We maintain a queue of outgoing messages, and some session state to
133 * ensure that we can preserve the lossless, ordered delivery of
134 * messages in the case of a TCP disconnect.
136 struct ceph_connection {
140 const struct ceph_connection_operations *ops;
142 struct ceph_messenger *msgr;
144 unsigned long state; /* connection state (see flags above) */
145 const char *error_msg; /* error message, if any */
147 struct ceph_entity_addr peer_addr; /* peer address */
148 struct ceph_entity_name peer_name; /* peer name */
149 struct ceph_entity_addr peer_addr_for_me;
150 u32 connect_seq; /* identify the most recent connection
151 attempt for this connection, client */
152 u32 peer_global_seq; /* peer's global seq for this connection */
154 int auth_retry; /* true if we need a newer authorizer */
155 void *auth_reply_buf; /* where to put the authorizer reply */
156 int auth_reply_buf_len;
161 struct list_head out_queue;
162 struct list_head out_sent; /* sending or sent but unacked */
163 u64 out_seq; /* last message queued for send */
164 u64 out_seq_sent; /* last message sent */
165 bool out_keepalive_pending;
167 u64 in_seq, in_seq_acked; /* last message received, acked */
169 /* connection negotiation temps */
170 char in_banner[CEPH_BANNER_MAX_LEN];
172 struct { /* outgoing connection */
173 struct ceph_msg_connect out_connect;
174 struct ceph_msg_connect_reply in_reply;
176 struct { /* incoming */
177 struct ceph_msg_connect in_connect;
178 struct ceph_msg_connect_reply out_reply;
181 struct ceph_entity_addr actual_peer_addr;
183 /* message out temps */
184 struct ceph_msg *out_msg; /* sending message (== tail of
187 struct ceph_msg_pos out_msg_pos;
189 struct kvec out_kvec[8], /* sending header/footer data */
191 int out_kvec_left; /* kvec's left in out_kvec */
192 int out_skip; /* skip this many bytes */
193 int out_kvec_bytes; /* total bytes left */
194 bool out_kvec_is_msg; /* kvec refers to out_msg */
195 int out_more; /* there is more data after the kvecs */
196 __le64 out_temp_ack; /* for writing an ack */
198 /* message in temps */
199 struct ceph_msg_header in_hdr;
200 struct ceph_msg *in_msg;
201 struct ceph_msg_pos in_msg_pos;
202 u32 in_front_crc, in_middle_crc, in_data_crc; /* calculated crc */
204 char in_tag; /* protocol control byte */
205 int in_base_pos; /* bytes read */
206 __le64 in_temp_ack; /* for reading an ack */
208 struct delayed_work work; /* send|recv work */
209 unsigned long delay; /* current delay interval */
213 extern const char *pr_addr(const struct sockaddr_storage *ss);
214 extern int ceph_parse_ips(const char *c, const char *end,
215 struct ceph_entity_addr *addr,
216 int max_count, int *count);
219 extern int ceph_msgr_init(void);
220 extern void ceph_msgr_exit(void);
222 extern struct ceph_messenger *ceph_messenger_create(
223 struct ceph_entity_addr *myaddr);
224 extern void ceph_messenger_destroy(struct ceph_messenger *);
226 extern void ceph_con_init(struct ceph_messenger *msgr,
227 struct ceph_connection *con);
228 extern void ceph_con_open(struct ceph_connection *con,
229 struct ceph_entity_addr *addr);
230 extern void ceph_con_close(struct ceph_connection *con);
231 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
232 extern void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg);
233 extern void ceph_con_keepalive(struct ceph_connection *con);
234 extern struct ceph_connection *ceph_con_get(struct ceph_connection *con);
235 extern void ceph_con_put(struct ceph_connection *con);
237 extern struct ceph_msg *ceph_msg_new(int type, int front_len,
238 int page_len, int page_off,
239 struct page **pages);
240 extern void ceph_msg_kfree(struct ceph_msg *m);
242 extern struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
243 struct ceph_msg_header *hdr);
244 extern int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg);
247 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
249 kref_get(&msg->kref);
252 extern void ceph_msg_last_put(struct kref *kref);
253 static inline void ceph_msg_put(struct ceph_msg *msg)
255 kref_put(&msg->kref, ceph_msg_last_put);
258 extern void ceph_msg_dump(struct ceph_msg *msg);