]>
Commit | Line | Data |
---|---|---|
31b8006e SW |
1 | #ifndef __FS_CEPH_MESSENGER_H |
2 | #define __FS_CEPH_MESSENGER_H | |
3 | ||
c2e552e7 | 4 | #include <linux/kref.h> |
31b8006e SW |
5 | #include <linux/mutex.h> |
6 | #include <linux/net.h> | |
7 | #include <linux/radix-tree.h> | |
8 | #include <linux/uio.h> | |
9 | #include <linux/version.h> | |
10 | #include <linux/workqueue.h> | |
11 | ||
12 | #include "types.h" | |
13 | #include "buffer.h" | |
14 | ||
15 | struct ceph_msg; | |
16 | struct ceph_connection; | |
17 | ||
18 | extern struct workqueue_struct *ceph_msgr_wq; /* receive work queue */ | |
19 | ||
20 | /* | |
21 | * Ceph defines these callbacks for handling connection events. | |
22 | */ | |
23 | struct ceph_connection_operations { | |
24 | struct ceph_connection *(*get)(struct ceph_connection *); | |
25 | void (*put)(struct ceph_connection *); | |
26 | ||
27 | /* handle an incoming message. */ | |
28 | void (*dispatch) (struct ceph_connection *con, struct ceph_msg *m); | |
29 | ||
4e7a5dcd SW |
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); | |
35 | ||
31b8006e SW |
36 | /* protocol version mismatch */ |
37 | void (*bad_proto) (struct ceph_connection *con); | |
38 | ||
39 | /* there was some error on the socket (disconnect, whatever) */ | |
40 | void (*fault) (struct ceph_connection *con); | |
41 | ||
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); | |
45 | ||
46 | struct ceph_msg * (*alloc_msg) (struct ceph_connection *con, | |
2450418c YS |
47 | struct ceph_msg_header *hdr, |
48 | int *skip); | |
31b8006e SW |
49 | }; |
50 | ||
51 | extern const char *ceph_name_type_str(int t); | |
52 | ||
53 | /* use format string %s%d */ | |
54 | #define ENTITY_NAME(n) ceph_name_type_str((n).type), le64_to_cpu((n).num) | |
55 | ||
56 | struct ceph_messenger { | |
57 | struct ceph_entity_inst inst; /* my name+address */ | |
63f2d211 | 58 | struct ceph_entity_addr my_enc_addr; |
31b8006e SW |
59 | struct page *zero_page; /* used in certain error cases */ |
60 | ||
61 | bool nocrc; | |
62 | ||
63 | /* | |
64 | * the global_seq counts connections i (attempt to) initiate | |
65 | * in order to disambiguate certain connect race conditions. | |
66 | */ | |
67 | u32 global_seq; | |
68 | spinlock_t global_seq_lock; | |
69 | }; | |
70 | ||
71 | /* | |
72 | * a single message. it contains a header (src, dest, message type, etc.), | |
73 | * footer (crc values, mainly), a "front" message body, and possibly a | |
74 | * data payload (stored in some number of pages). | |
75 | */ | |
76 | struct ceph_msg { | |
77 | struct ceph_msg_header hdr; /* header */ | |
78 | struct ceph_msg_footer footer; /* footer */ | |
79 | struct kvec front; /* unaligned blobs of message */ | |
80 | struct ceph_buffer *middle; | |
81 | struct page **pages; /* data payload. NOT OWNER. */ | |
82 | unsigned nr_pages; /* size of page array */ | |
58bb3b37 | 83 | struct ceph_pagelist *pagelist; /* instead of pages */ |
31b8006e | 84 | struct list_head list_head; |
c2e552e7 | 85 | struct kref kref; |
31b8006e SW |
86 | bool front_is_vmalloc; |
87 | bool more_to_follow; | |
88 | int front_max; | |
89 | ||
90 | struct ceph_msgpool *pool; | |
91 | }; | |
92 | ||
93 | struct ceph_msg_pos { | |
94 | int page, page_pos; /* which page; offset in page */ | |
95 | int data_pos; /* offset in data payload */ | |
96 | int did_page_crc; /* true if we've calculated crc for current page */ | |
97 | }; | |
98 | ||
99 | /* ceph connection fault delay defaults, for exponential backoff */ | |
100 | #define BASE_DELAY_INTERVAL (HZ/2) | |
101 | #define MAX_DELAY_INTERVAL (5 * 60 * HZ) | |
102 | ||
103 | /* | |
104 | * ceph_connection state bit flags | |
105 | * | |
106 | * QUEUED and BUSY are used together to ensure that only a single | |
107 | * thread is currently opening, reading or writing data to the socket. | |
108 | */ | |
109 | #define LOSSYTX 0 /* we can close channel or drop messages on errors */ | |
eed0ef2c SW |
110 | #define CONNECTING 1 |
111 | #define NEGOTIATING 2 | |
31b8006e SW |
112 | #define KEEPALIVE_PENDING 3 |
113 | #define WRITE_PENDING 4 /* we have data ready to send */ | |
114 | #define QUEUED 5 /* there is work queued on this connection */ | |
115 | #define BUSY 6 /* work is being done */ | |
116 | #define STANDBY 8 /* no outgoing messages, socket closed. we keep | |
117 | * the ceph_connection around to maintain shared | |
118 | * state with the peer. */ | |
119 | #define CLOSED 10 /* we've closed the connection */ | |
120 | #define SOCK_CLOSED 11 /* socket state changed to closed */ | |
121 | #define REGISTERED 12 /* connection appears in con_tree */ | |
122 | #define OPENING 13 /* open connection w/ (possibly new) peer */ | |
123 | #define DEAD 14 /* dead, about to kfree */ | |
124 | ||
125 | /* | |
126 | * A single connection with another host. | |
127 | * | |
128 | * We maintain a queue of outgoing messages, and some session state to | |
129 | * ensure that we can preserve the lossless, ordered delivery of | |
130 | * messages in the case of a TCP disconnect. | |
131 | */ | |
132 | struct ceph_connection { | |
133 | void *private; | |
134 | atomic_t nref; | |
135 | ||
136 | const struct ceph_connection_operations *ops; | |
137 | ||
138 | struct ceph_messenger *msgr; | |
139 | struct socket *sock; | |
140 | unsigned long state; /* connection state (see flags above) */ | |
141 | const char *error_msg; /* error message, if any */ | |
142 | ||
143 | struct ceph_entity_addr peer_addr; /* peer address */ | |
144 | struct ceph_entity_name peer_name; /* peer name */ | |
145 | struct ceph_entity_addr peer_addr_for_me; | |
146 | u32 connect_seq; /* identify the most recent connection | |
147 | attempt for this connection, client */ | |
148 | u32 peer_global_seq; /* peer's global seq for this connection */ | |
149 | ||
4e7a5dcd SW |
150 | int auth_retry; /* true if we need a newer authorizer */ |
151 | void *auth_reply_buf; /* where to put the authorizer reply */ | |
152 | int auth_reply_buf_len; | |
153 | ||
ec302645 SW |
154 | struct mutex mutex; |
155 | ||
31b8006e | 156 | /* out queue */ |
31b8006e SW |
157 | struct list_head out_queue; |
158 | struct list_head out_sent; /* sending or sent but unacked */ | |
159 | u64 out_seq; /* last message queued for send */ | |
160 | u64 out_seq_sent; /* last message sent */ | |
161 | bool out_keepalive_pending; | |
162 | ||
163 | u64 in_seq, in_seq_acked; /* last message received, acked */ | |
164 | ||
165 | /* connection negotiation temps */ | |
166 | char in_banner[CEPH_BANNER_MAX_LEN]; | |
167 | union { | |
168 | struct { /* outgoing connection */ | |
169 | struct ceph_msg_connect out_connect; | |
170 | struct ceph_msg_connect_reply in_reply; | |
171 | }; | |
172 | struct { /* incoming */ | |
173 | struct ceph_msg_connect in_connect; | |
174 | struct ceph_msg_connect_reply out_reply; | |
175 | }; | |
176 | }; | |
177 | struct ceph_entity_addr actual_peer_addr; | |
178 | ||
179 | /* message out temps */ | |
180 | struct ceph_msg *out_msg; /* sending message (== tail of | |
181 | out_sent) */ | |
c86a2930 | 182 | bool out_msg_done; |
31b8006e SW |
183 | struct ceph_msg_pos out_msg_pos; |
184 | ||
185 | struct kvec out_kvec[8], /* sending header/footer data */ | |
186 | *out_kvec_cur; | |
187 | int out_kvec_left; /* kvec's left in out_kvec */ | |
188 | int out_skip; /* skip this many bytes */ | |
189 | int out_kvec_bytes; /* total bytes left */ | |
190 | bool out_kvec_is_msg; /* kvec refers to out_msg */ | |
191 | int out_more; /* there is more data after the kvecs */ | |
192 | __le64 out_temp_ack; /* for writing an ack */ | |
193 | ||
194 | /* message in temps */ | |
195 | struct ceph_msg_header in_hdr; | |
196 | struct ceph_msg *in_msg; | |
197 | struct ceph_msg_pos in_msg_pos; | |
198 | u32 in_front_crc, in_middle_crc, in_data_crc; /* calculated crc */ | |
199 | ||
200 | char in_tag; /* protocol control byte */ | |
201 | int in_base_pos; /* bytes read */ | |
202 | __le64 in_temp_ack; /* for reading an ack */ | |
203 | ||
204 | struct delayed_work work; /* send|recv work */ | |
205 | unsigned long delay; /* current delay interval */ | |
206 | }; | |
207 | ||
208 | ||
209 | extern const char *pr_addr(const struct sockaddr_storage *ss); | |
210 | extern int ceph_parse_ips(const char *c, const char *end, | |
211 | struct ceph_entity_addr *addr, | |
212 | int max_count, int *count); | |
213 | ||
214 | ||
215 | extern int ceph_msgr_init(void); | |
216 | extern void ceph_msgr_exit(void); | |
217 | ||
218 | extern struct ceph_messenger *ceph_messenger_create( | |
219 | struct ceph_entity_addr *myaddr); | |
220 | extern void ceph_messenger_destroy(struct ceph_messenger *); | |
221 | ||
222 | extern void ceph_con_init(struct ceph_messenger *msgr, | |
223 | struct ceph_connection *con); | |
31b8006e SW |
224 | extern void ceph_con_open(struct ceph_connection *con, |
225 | struct ceph_entity_addr *addr); | |
226 | extern void ceph_con_close(struct ceph_connection *con); | |
227 | extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg); | |
228 | extern void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg); | |
350b1c32 SW |
229 | extern void ceph_con_revoke_pages(struct ceph_connection *con, |
230 | struct page **pages); | |
31b8006e SW |
231 | extern void ceph_con_keepalive(struct ceph_connection *con); |
232 | extern struct ceph_connection *ceph_con_get(struct ceph_connection *con); | |
233 | extern void ceph_con_put(struct ceph_connection *con); | |
234 | ||
235 | extern struct ceph_msg *ceph_msg_new(int type, int front_len, | |
236 | int page_len, int page_off, | |
237 | struct page **pages); | |
238 | extern void ceph_msg_kfree(struct ceph_msg *m); | |
239 | ||
31b8006e SW |
240 | |
241 | static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) | |
242 | { | |
c2e552e7 | 243 | kref_get(&msg->kref); |
31b8006e SW |
244 | return msg; |
245 | } | |
c2e552e7 SW |
246 | extern void ceph_msg_last_put(struct kref *kref); |
247 | static inline void ceph_msg_put(struct ceph_msg *msg) | |
248 | { | |
249 | kref_put(&msg->kref, ceph_msg_last_put); | |
250 | } | |
31b8006e | 251 | |
9ec7cab1 SW |
252 | extern void ceph_msg_dump(struct ceph_msg *msg); |
253 | ||
31b8006e | 254 | #endif |