1 #ifndef __FS_CEPH_MESSENGER_H
2 #define __FS_CEPH_MESSENGER_H
4 #include <linux/mutex.h>
6 #include <linux/radix-tree.h>
8 #include <linux/version.h>
9 #include <linux/workqueue.h>
15 struct ceph_connection
;
17 extern struct workqueue_struct
*ceph_msgr_wq
; /* receive work queue */
20 * Ceph defines these callbacks for handling connection events.
22 struct ceph_connection_operations
{
23 struct ceph_connection
*(*get
)(struct ceph_connection
*);
24 void (*put
)(struct ceph_connection
*);
26 /* handle an incoming message. */
27 void (*dispatch
) (struct ceph_connection
*con
, struct ceph_msg
*m
);
29 /* protocol version mismatch */
30 void (*bad_proto
) (struct ceph_connection
*con
);
32 /* there was some error on the socket (disconnect, whatever) */
33 void (*fault
) (struct ceph_connection
*con
);
35 /* a remote host as terminated a message exchange session, and messages
36 * we sent (or they tried to send us) may be lost. */
37 void (*peer_reset
) (struct ceph_connection
*con
);
39 struct ceph_msg
* (*alloc_msg
) (struct ceph_connection
*con
,
40 struct ceph_msg_header
*hdr
);
41 int (*alloc_middle
) (struct ceph_connection
*con
,
42 struct ceph_msg
*msg
);
43 /* an incoming message has a data payload; tell me what pages I
44 * should read the data into. */
45 int (*prepare_pages
) (struct ceph_connection
*con
, struct ceph_msg
*m
,
49 extern const char *ceph_name_type_str(int t
);
51 /* use format string %s%d */
52 #define ENTITY_NAME(n) ceph_name_type_str((n).type), le64_to_cpu((n).num)
54 struct ceph_messenger
{
55 struct ceph_entity_inst inst
; /* my name+address */
56 struct ceph_entity_addr my_enc_addr
;
57 struct page
*zero_page
; /* used in certain error cases */
62 * the global_seq counts connections i (attempt to) initiate
63 * in order to disambiguate certain connect race conditions.
66 spinlock_t global_seq_lock
;
70 * a single message. it contains a header (src, dest, message type, etc.),
71 * footer (crc values, mainly), a "front" message body, and possibly a
72 * data payload (stored in some number of pages).
75 struct ceph_msg_header hdr
; /* header */
76 struct ceph_msg_footer footer
; /* footer */
77 struct kvec front
; /* unaligned blobs of message */
78 struct ceph_buffer
*middle
;
79 struct page
**pages
; /* data payload. NOT OWNER. */
80 unsigned nr_pages
; /* size of page array */
81 struct list_head list_head
;
83 bool front_is_vmalloc
;
87 struct ceph_msgpool
*pool
;
91 int page
, page_pos
; /* which page; offset in page */
92 int data_pos
; /* offset in data payload */
93 int did_page_crc
; /* true if we've calculated crc for current page */
96 /* ceph connection fault delay defaults, for exponential backoff */
97 #define BASE_DELAY_INTERVAL (HZ/2)
98 #define MAX_DELAY_INTERVAL (5 * 60 * HZ)
101 * ceph_connection state bit flags
103 * QUEUED and BUSY are used together to ensure that only a single
104 * thread is currently opening, reading or writing data to the socket.
106 #define LOSSYTX 0 /* we can close channel or drop messages on errors */
107 #define LOSSYRX 1 /* peer may reset/drop messages */
109 #define KEEPALIVE_PENDING 3
110 #define WRITE_PENDING 4 /* we have data ready to send */
111 #define QUEUED 5 /* there is work queued on this connection */
112 #define BUSY 6 /* work is being done */
113 #define STANDBY 8 /* no outgoing messages, socket closed. we keep
114 * the ceph_connection around to maintain shared
115 * state with the peer. */
116 #define CLOSED 10 /* we've closed the connection */
117 #define SOCK_CLOSED 11 /* socket state changed to closed */
118 #define REGISTERED 12 /* connection appears in con_tree */
119 #define OPENING 13 /* open connection w/ (possibly new) peer */
120 #define DEAD 14 /* dead, about to kfree */
123 * A single connection with another host.
125 * We maintain a queue of outgoing messages, and some session state to
126 * ensure that we can preserve the lossless, ordered delivery of
127 * messages in the case of a TCP disconnect.
129 struct ceph_connection
{
133 const struct ceph_connection_operations
*ops
;
135 struct ceph_messenger
*msgr
;
137 unsigned long state
; /* connection state (see flags above) */
138 const char *error_msg
; /* error message, if any */
140 struct ceph_entity_addr peer_addr
; /* peer address */
141 struct ceph_entity_name peer_name
; /* peer name */
142 struct ceph_entity_addr peer_addr_for_me
;
143 u32 connect_seq
; /* identify the most recent connection
144 attempt for this connection, client */
145 u32 peer_global_seq
; /* peer's global seq for this connection */
148 struct mutex out_mutex
;
149 struct list_head out_queue
;
150 struct list_head out_sent
; /* sending or sent but unacked */
151 u64 out_seq
; /* last message queued for send */
152 u64 out_seq_sent
; /* last message sent */
153 bool out_keepalive_pending
;
155 u64 in_seq
, in_seq_acked
; /* last message received, acked */
157 /* connection negotiation temps */
158 char in_banner
[CEPH_BANNER_MAX_LEN
];
160 struct { /* outgoing connection */
161 struct ceph_msg_connect out_connect
;
162 struct ceph_msg_connect_reply in_reply
;
164 struct { /* incoming */
165 struct ceph_msg_connect in_connect
;
166 struct ceph_msg_connect_reply out_reply
;
169 struct ceph_entity_addr actual_peer_addr
;
171 /* message out temps */
172 struct ceph_msg
*out_msg
; /* sending message (== tail of
174 struct ceph_msg_pos out_msg_pos
;
176 struct kvec out_kvec
[8], /* sending header/footer data */
178 int out_kvec_left
; /* kvec's left in out_kvec */
179 int out_skip
; /* skip this many bytes */
180 int out_kvec_bytes
; /* total bytes left */
181 bool out_kvec_is_msg
; /* kvec refers to out_msg */
182 int out_more
; /* there is more data after the kvecs */
183 __le64 out_temp_ack
; /* for writing an ack */
185 /* message in temps */
186 struct ceph_msg_header in_hdr
;
187 struct ceph_msg
*in_msg
;
188 struct ceph_msg_pos in_msg_pos
;
189 u32 in_front_crc
, in_middle_crc
, in_data_crc
; /* calculated crc */
191 char in_tag
; /* protocol control byte */
192 int in_base_pos
; /* bytes read */
193 __le64 in_temp_ack
; /* for reading an ack */
195 struct delayed_work work
; /* send|recv work */
196 unsigned long delay
; /* current delay interval */
200 extern const char *pr_addr(const struct sockaddr_storage
*ss
);
201 extern int ceph_parse_ips(const char *c
, const char *end
,
202 struct ceph_entity_addr
*addr
,
203 int max_count
, int *count
);
206 extern int ceph_msgr_init(void);
207 extern void ceph_msgr_exit(void);
209 extern struct ceph_messenger
*ceph_messenger_create(
210 struct ceph_entity_addr
*myaddr
);
211 extern void ceph_messenger_destroy(struct ceph_messenger
*);
213 extern void ceph_con_init(struct ceph_messenger
*msgr
,
214 struct ceph_connection
*con
);
215 extern void ceph_con_shutdown(struct ceph_connection
*con
);
216 extern void ceph_con_open(struct ceph_connection
*con
,
217 struct ceph_entity_addr
*addr
);
218 extern void ceph_con_close(struct ceph_connection
*con
);
219 extern void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
);
220 extern void ceph_con_revoke(struct ceph_connection
*con
, struct ceph_msg
*msg
);
221 extern void ceph_con_keepalive(struct ceph_connection
*con
);
222 extern struct ceph_connection
*ceph_con_get(struct ceph_connection
*con
);
223 extern void ceph_con_put(struct ceph_connection
*con
);
225 extern struct ceph_msg
*ceph_msg_new(int type
, int front_len
,
226 int page_len
, int page_off
,
227 struct page
**pages
);
228 extern void ceph_msg_kfree(struct ceph_msg
*m
);
230 extern struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
231 struct ceph_msg_header
*hdr
);
232 extern int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
);
235 static inline struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
237 dout("ceph_msg_get %p %d -> %d\n", msg
, atomic_read(&msg
->nref
),
238 atomic_read(&msg
->nref
)+1);
239 atomic_inc(&msg
->nref
);
242 extern void ceph_msg_put(struct ceph_msg
*msg
);