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/workqueue.h>
11 #include <linux/ceph/types.h>
12 #include <linux/ceph/buffer.h>
15 struct ceph_connection;
18 * Ceph defines these callbacks for handling connection events.
20 struct ceph_connection_operations {
21 struct ceph_connection *(*get)(struct ceph_connection *);
22 void (*put)(struct ceph_connection *);
24 /* handle an incoming message. */
25 void (*dispatch) (struct ceph_connection *con, struct ceph_msg *m);
27 /* authorize an outgoing connection */
28 struct ceph_auth_handshake *(*get_authorizer) (
29 struct ceph_connection *con,
30 int *proto, int force_new);
31 int (*verify_authorizer_reply) (struct ceph_connection *con, int len);
32 int (*invalidate_authorizer)(struct ceph_connection *con);
34 /* there was some error on the socket (disconnect, whatever) */
35 void (*fault) (struct ceph_connection *con);
37 /* a remote host as terminated a message exchange session, and messages
38 * we sent (or they tried to send us) may be lost. */
39 void (*peer_reset) (struct ceph_connection *con);
41 struct ceph_msg * (*alloc_msg) (struct ceph_connection *con,
42 struct ceph_msg_header *hdr,
46 /* use format string %s%d */
47 #define ENTITY_NAME(n) ceph_entity_type_name((n).type), le64_to_cpu((n).num)
49 struct ceph_messenger {
50 struct ceph_entity_inst inst; /* my name+address */
51 struct ceph_entity_addr my_enc_addr;
57 * the global_seq counts connections i (attempt to) initiate
58 * in order to disambiguate certain connect race conditions.
61 spinlock_t global_seq_lock;
63 u32 supported_features;
64 u32 required_features;
67 enum ceph_msg_data_type {
68 CEPH_MSG_DATA_NONE, /* message contains no data payload */
69 CEPH_MSG_DATA_PAGES, /* data source/destination is a page array */
70 CEPH_MSG_DATA_PAGELIST, /* data source/destination is a pagelist */
72 CEPH_MSG_DATA_BIO, /* data source/destination is a bio list */
73 #endif /* CONFIG_BLOCK */
76 static __inline__ bool ceph_msg_data_type_valid(enum ceph_msg_data_type type)
79 case CEPH_MSG_DATA_NONE:
80 case CEPH_MSG_DATA_PAGES:
81 case CEPH_MSG_DATA_PAGELIST:
83 case CEPH_MSG_DATA_BIO:
84 #endif /* CONFIG_BLOCK */
91 struct ceph_msg_data {
92 struct list_head links; /* ceph_msg->data */
93 enum ceph_msg_data_type type;
100 #endif /* CONFIG_BLOCK */
102 struct page **pages; /* NOT OWNER. */
103 size_t length; /* total # bytes */
104 unsigned int alignment; /* first page */
106 struct ceph_pagelist *pagelist;
110 struct ceph_msg_data_cursor {
111 size_t total_resid; /* across all data items */
112 struct list_head *data_head; /* = &ceph_msg->data */
114 struct ceph_msg_data *data; /* current data item */
115 size_t resid; /* bytes not yet consumed */
116 bool last_piece; /* current is last piece */
117 bool need_crc; /* crc update needed */
121 struct bio *bio; /* bio from list */
122 unsigned int vector_index; /* vector from bio */
123 unsigned int vector_offset; /* bytes from vector */
125 #endif /* CONFIG_BLOCK */
127 unsigned int page_offset; /* offset in page */
128 unsigned short page_index; /* index in array */
129 unsigned short page_count; /* pages in array */
131 struct { /* pagelist */
132 struct page *page; /* page from list */
133 size_t offset; /* bytes from list */
139 * a single message. it contains a header (src, dest, message type, etc.),
140 * footer (crc values, mainly), a "front" message body, and possibly a
141 * data payload (stored in some number of pages).
144 struct ceph_msg_header hdr; /* header */
145 struct ceph_msg_footer footer; /* footer */
146 struct kvec front; /* unaligned blobs of message */
147 struct ceph_buffer *middle;
150 struct list_head data;
151 struct ceph_msg_data_cursor cursor;
153 struct ceph_connection *con;
154 struct list_head list_head; /* links for connection lists */
157 bool front_is_vmalloc;
161 unsigned long ack_stamp; /* tx: when we were acked */
163 struct ceph_msgpool *pool;
166 /* ceph connection fault delay defaults, for exponential backoff */
167 #define BASE_DELAY_INTERVAL (HZ/2)
168 #define MAX_DELAY_INTERVAL (5 * 60 * HZ)
171 * A single connection with another host.
173 * We maintain a queue of outgoing messages, and some session state to
174 * ensure that we can preserve the lossless, ordered delivery of
175 * messages in the case of a TCP disconnect.
177 struct ceph_connection {
180 const struct ceph_connection_operations *ops;
182 struct ceph_messenger *msgr;
186 struct ceph_entity_addr peer_addr; /* peer address */
187 struct ceph_entity_addr peer_addr_for_me;
191 const char *error_msg; /* error message, if any */
193 struct ceph_entity_name peer_name; /* peer name */
195 unsigned peer_features;
196 u32 connect_seq; /* identify the most recent connection
197 attempt for this connection, client */
198 u32 peer_global_seq; /* peer's global seq for this connection */
200 int auth_retry; /* true if we need a newer authorizer */
201 void *auth_reply_buf; /* where to put the authorizer reply */
202 int auth_reply_buf_len;
207 struct list_head out_queue;
208 struct list_head out_sent; /* sending or sent but unacked */
209 u64 out_seq; /* last message queued for send */
211 u64 in_seq, in_seq_acked; /* last message received, acked */
213 /* connection negotiation temps */
214 char in_banner[CEPH_BANNER_MAX_LEN];
215 struct ceph_msg_connect out_connect;
216 struct ceph_msg_connect_reply in_reply;
217 struct ceph_entity_addr actual_peer_addr;
219 /* message out temps */
220 struct ceph_msg *out_msg; /* sending message (== tail of
224 struct kvec out_kvec[8], /* sending header/footer data */
226 int out_kvec_left; /* kvec's left in out_kvec */
227 int out_skip; /* skip this many bytes */
228 int out_kvec_bytes; /* total bytes left */
229 bool out_kvec_is_msg; /* kvec refers to out_msg */
230 int out_more; /* there is more data after the kvecs */
231 __le64 out_temp_ack; /* for writing an ack */
233 /* message in temps */
234 struct ceph_msg_header in_hdr;
235 struct ceph_msg *in_msg;
236 u32 in_front_crc, in_middle_crc, in_data_crc; /* calculated crc */
238 char in_tag; /* protocol control byte */
239 int in_base_pos; /* bytes read */
240 __le64 in_temp_ack; /* for reading an ack */
242 struct delayed_work work; /* send|recv work */
243 unsigned long delay; /* current delay interval */
247 extern const char *ceph_pr_addr(const struct sockaddr_storage *ss);
248 extern int ceph_parse_ips(const char *c, const char *end,
249 struct ceph_entity_addr *addr,
250 int max_count, int *count);
253 extern int ceph_msgr_init(void);
254 extern void ceph_msgr_exit(void);
255 extern void ceph_msgr_flush(void);
257 extern void ceph_messenger_init(struct ceph_messenger *msgr,
258 struct ceph_entity_addr *myaddr,
259 u32 supported_features,
260 u32 required_features,
263 extern void ceph_con_init(struct ceph_connection *con, void *private,
264 const struct ceph_connection_operations *ops,
265 struct ceph_messenger *msgr);
266 extern void ceph_con_open(struct ceph_connection *con,
267 __u8 entity_type, __u64 entity_num,
268 struct ceph_entity_addr *addr);
269 extern bool ceph_con_opened(struct ceph_connection *con);
270 extern void ceph_con_close(struct ceph_connection *con);
271 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
273 extern void ceph_msg_revoke(struct ceph_msg *msg);
274 extern void ceph_msg_revoke_incoming(struct ceph_msg *msg);
276 extern void ceph_con_keepalive(struct ceph_connection *con);
278 extern void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
279 size_t length, size_t alignment);
280 extern void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
281 struct ceph_pagelist *pagelist);
283 extern void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
285 #endif /* CONFIG_BLOCK */
287 extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
289 extern void ceph_msg_kfree(struct ceph_msg *m);
292 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
294 kref_get(&msg->kref);
297 extern void ceph_msg_last_put(struct kref *kref);
298 static inline void ceph_msg_put(struct ceph_msg *msg)
300 kref_put(&msg->kref, ceph_msg_last_put);
303 extern void ceph_msg_dump(struct ceph_msg *msg);