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 #define ceph_msg_has_pages(m) ((m)->p.type == CEPH_MSG_DATA_PAGES)
68 #define ceph_msg_has_pagelist(m) ((m)->l.type == CEPH_MSG_DATA_PAGELIST)
70 #define ceph_msg_has_bio(m) ((m)->b.type == CEPH_MSG_DATA_BIO)
71 #endif /* CONFIG_BLOCK */
72 #define ceph_msg_has_trail(m) ((m)->t.type == CEPH_MSG_DATA_PAGELIST)
74 enum ceph_msg_data_type
{
75 CEPH_MSG_DATA_NONE
, /* message contains no data payload */
76 CEPH_MSG_DATA_PAGES
, /* data source/destination is a page array */
77 CEPH_MSG_DATA_PAGELIST
, /* data source/destination is a pagelist */
79 CEPH_MSG_DATA_BIO
, /* data source/destination is a bio list */
80 #endif /* CONFIG_BLOCK */
83 static __inline__
bool ceph_msg_data_type_valid(enum ceph_msg_data_type type
)
86 case CEPH_MSG_DATA_NONE
:
87 case CEPH_MSG_DATA_PAGES
:
88 case CEPH_MSG_DATA_PAGELIST
:
90 case CEPH_MSG_DATA_BIO
:
91 #endif /* CONFIG_BLOCK */
98 struct ceph_msg_data_cursor
{
99 bool last_piece
; /* now at last piece of data item */
103 struct bio
*bio
; /* bio from list */
104 unsigned int vector_index
; /* vector from bio */
105 unsigned int vector_offset
; /* bytes from vector */
107 #endif /* CONFIG_BLOCK */
109 size_t resid
; /* bytes from array */
110 unsigned int page_offset
; /* offset in page */
111 unsigned short page_index
; /* index in array */
112 unsigned short page_count
; /* pages in array */
114 struct { /* pagelist */
115 struct page
*page
; /* page from list */
116 size_t offset
; /* bytes from list */
121 struct ceph_msg_data
{
122 enum ceph_msg_data_type type
;
126 struct bio
*bio_iter
; /* iterator */
128 unsigned int bio_seg
; /* current seg in bio */
130 #endif /* CONFIG_BLOCK */
132 struct page
**pages
; /* NOT OWNER. */
133 size_t length
; /* total # bytes */
134 unsigned int alignment
; /* first page */
136 struct ceph_pagelist
*pagelist
;
138 struct ceph_msg_data_cursor cursor
; /* pagelist only */
142 * a single message. it contains a header (src, dest, message type, etc.),
143 * footer (crc values, mainly), a "front" message body, and possibly a
144 * data payload (stored in some number of pages).
147 struct ceph_msg_header hdr
; /* header */
148 struct ceph_msg_footer footer
; /* footer */
149 struct kvec front
; /* unaligned blobs of message */
150 struct ceph_buffer
*middle
;
153 struct ceph_msg_data p
; /* pages */
154 struct ceph_msg_data l
; /* pagelist */
156 struct ceph_msg_data b
; /* bio */
157 #endif /* CONFIG_BLOCK */
158 struct ceph_msg_data t
; /* trail */
160 struct ceph_connection
*con
;
161 struct list_head list_head
; /* links for connection lists */
164 bool front_is_vmalloc
;
168 unsigned long ack_stamp
; /* tx: when we were acked */
170 struct ceph_msgpool
*pool
;
173 struct ceph_msg_pos
{
174 int page
, page_pos
; /* which page; offset in page */
175 int data_pos
; /* offset in data payload */
176 bool did_page_crc
; /* true if we've calculated crc for current page */
179 /* ceph connection fault delay defaults, for exponential backoff */
180 #define BASE_DELAY_INTERVAL (HZ/2)
181 #define MAX_DELAY_INTERVAL (5 * 60 * HZ)
184 * A single connection with another host.
186 * We maintain a queue of outgoing messages, and some session state to
187 * ensure that we can preserve the lossless, ordered delivery of
188 * messages in the case of a TCP disconnect.
190 struct ceph_connection
{
193 const struct ceph_connection_operations
*ops
;
195 struct ceph_messenger
*msgr
;
199 struct ceph_entity_addr peer_addr
; /* peer address */
200 struct ceph_entity_addr peer_addr_for_me
;
204 const char *error_msg
; /* error message, if any */
206 struct ceph_entity_name peer_name
; /* peer name */
208 unsigned peer_features
;
209 u32 connect_seq
; /* identify the most recent connection
210 attempt for this connection, client */
211 u32 peer_global_seq
; /* peer's global seq for this connection */
213 int auth_retry
; /* true if we need a newer authorizer */
214 void *auth_reply_buf
; /* where to put the authorizer reply */
215 int auth_reply_buf_len
;
220 struct list_head out_queue
;
221 struct list_head out_sent
; /* sending or sent but unacked */
222 u64 out_seq
; /* last message queued for send */
224 u64 in_seq
, in_seq_acked
; /* last message received, acked */
226 /* connection negotiation temps */
227 char in_banner
[CEPH_BANNER_MAX_LEN
];
228 struct ceph_msg_connect out_connect
;
229 struct ceph_msg_connect_reply in_reply
;
230 struct ceph_entity_addr actual_peer_addr
;
232 /* message out temps */
233 struct ceph_msg
*out_msg
; /* sending message (== tail of
236 struct ceph_msg_pos out_msg_pos
;
238 struct kvec out_kvec
[8], /* sending header/footer data */
240 int out_kvec_left
; /* kvec's left in out_kvec */
241 int out_skip
; /* skip this many bytes */
242 int out_kvec_bytes
; /* total bytes left */
243 bool out_kvec_is_msg
; /* kvec refers to out_msg */
244 int out_more
; /* there is more data after the kvecs */
245 __le64 out_temp_ack
; /* for writing an ack */
247 /* message in temps */
248 struct ceph_msg_header in_hdr
;
249 struct ceph_msg
*in_msg
;
250 struct ceph_msg_pos in_msg_pos
;
251 u32 in_front_crc
, in_middle_crc
, in_data_crc
; /* calculated crc */
253 char in_tag
; /* protocol control byte */
254 int in_base_pos
; /* bytes read */
255 __le64 in_temp_ack
; /* for reading an ack */
257 struct delayed_work work
; /* send|recv work */
258 unsigned long delay
; /* current delay interval */
262 extern const char *ceph_pr_addr(const struct sockaddr_storage
*ss
);
263 extern int ceph_parse_ips(const char *c
, const char *end
,
264 struct ceph_entity_addr
*addr
,
265 int max_count
, int *count
);
268 extern int ceph_msgr_init(void);
269 extern void ceph_msgr_exit(void);
270 extern void ceph_msgr_flush(void);
272 extern void ceph_messenger_init(struct ceph_messenger
*msgr
,
273 struct ceph_entity_addr
*myaddr
,
274 u32 supported_features
,
275 u32 required_features
,
278 extern void ceph_con_init(struct ceph_connection
*con
, void *private,
279 const struct ceph_connection_operations
*ops
,
280 struct ceph_messenger
*msgr
);
281 extern void ceph_con_open(struct ceph_connection
*con
,
282 __u8 entity_type
, __u64 entity_num
,
283 struct ceph_entity_addr
*addr
);
284 extern bool ceph_con_opened(struct ceph_connection
*con
);
285 extern void ceph_con_close(struct ceph_connection
*con
);
286 extern void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
);
288 extern void ceph_msg_revoke(struct ceph_msg
*msg
);
289 extern void ceph_msg_revoke_incoming(struct ceph_msg
*msg
);
291 extern void ceph_con_keepalive(struct ceph_connection
*con
);
293 extern void ceph_msg_data_set_pages(struct ceph_msg
*msg
, struct page
**pages
,
294 size_t length
, size_t alignment
);
295 extern void ceph_msg_data_set_pagelist(struct ceph_msg
*msg
,
296 struct ceph_pagelist
*pagelist
);
297 extern void ceph_msg_data_set_bio(struct ceph_msg
*msg
, struct bio
*bio
);
298 extern void ceph_msg_data_set_trail(struct ceph_msg
*msg
,
299 struct ceph_pagelist
*trail
);
301 extern struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
303 extern void ceph_msg_kfree(struct ceph_msg
*m
);
306 static inline struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
308 kref_get(&msg
->kref
);
311 extern void ceph_msg_last_put(struct kref
*kref
);
312 static inline void ceph_msg_put(struct ceph_msg
*msg
)
314 kref_put(&msg
->kref
, ceph_msg_last_put
);
317 extern void ceph_msg_dump(struct ceph_msg
*msg
);