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> | |
31b8006e SW |
9 | #include <linux/workqueue.h> |
10 | ||
11 | #include "types.h" | |
12 | #include "buffer.h" | |
13 | ||
14 | struct ceph_msg; | |
15 | struct ceph_connection; | |
16 | ||
31b8006e SW |
17 | /* |
18 | * Ceph defines these callbacks for handling connection events. | |
19 | */ | |
20 | struct ceph_connection_operations { | |
21 | struct ceph_connection *(*get)(struct ceph_connection *); | |
22 | void (*put)(struct ceph_connection *); | |
23 | ||
24 | /* handle an incoming message. */ | |
25 | void (*dispatch) (struct ceph_connection *con, struct ceph_msg *m); | |
26 | ||
4e7a5dcd | 27 | /* authorize an outgoing connection */ |
a3530df3 AE |
28 | struct ceph_auth_handshake *(*get_authorizer) ( |
29 | struct ceph_connection *con, | |
8f43fb53 | 30 | int *proto, int force_new); |
4e7a5dcd | 31 | int (*verify_authorizer_reply) (struct ceph_connection *con, int len); |
9bd2e6f8 | 32 | int (*invalidate_authorizer)(struct ceph_connection *con); |
4e7a5dcd | 33 | |
31b8006e SW |
34 | /* there was some error on the socket (disconnect, whatever) */ |
35 | void (*fault) (struct ceph_connection *con); | |
36 | ||
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); | |
40 | ||
41 | struct ceph_msg * (*alloc_msg) (struct ceph_connection *con, | |
2450418c YS |
42 | struct ceph_msg_header *hdr, |
43 | int *skip); | |
31b8006e SW |
44 | }; |
45 | ||
31b8006e | 46 | /* use format string %s%d */ |
ca9d93a2 | 47 | #define ENTITY_NAME(n) ceph_entity_type_name((n).type), le64_to_cpu((n).num) |
31b8006e SW |
48 | |
49 | struct ceph_messenger { | |
50 | struct ceph_entity_inst inst; /* my name+address */ | |
63f2d211 | 51 | struct ceph_entity_addr my_enc_addr; |
31b8006e SW |
52 | |
53 | bool nocrc; | |
54 | ||
55 | /* | |
56 | * the global_seq counts connections i (attempt to) initiate | |
57 | * in order to disambiguate certain connect race conditions. | |
58 | */ | |
59 | u32 global_seq; | |
60 | spinlock_t global_seq_lock; | |
3d14c5d2 YS |
61 | |
62 | u32 supported_features; | |
63 | u32 required_features; | |
31b8006e SW |
64 | }; |
65 | ||
66 | /* | |
67 | * a single message. it contains a header (src, dest, message type, etc.), | |
68 | * footer (crc values, mainly), a "front" message body, and possibly a | |
69 | * data payload (stored in some number of pages). | |
70 | */ | |
71 | struct ceph_msg { | |
72 | struct ceph_msg_header hdr; /* header */ | |
73 | struct ceph_msg_footer footer; /* footer */ | |
74 | struct kvec front; /* unaligned blobs of message */ | |
75 | struct ceph_buffer *middle; | |
76 | struct page **pages; /* data payload. NOT OWNER. */ | |
77 | unsigned nr_pages; /* size of page array */ | |
c5c6b19d | 78 | unsigned page_alignment; /* io offset in first page */ |
58bb3b37 | 79 | struct ceph_pagelist *pagelist; /* instead of pages */ |
31b8006e | 80 | struct list_head list_head; |
c2e552e7 | 81 | struct kref kref; |
68b4476b YS |
82 | struct bio *bio; /* instead of pages/pagelist */ |
83 | struct bio *bio_iter; /* bio iterator */ | |
84 | int bio_seg; /* current bio segment */ | |
85 | struct ceph_pagelist *trail; /* the trailing part of the data */ | |
31b8006e SW |
86 | bool front_is_vmalloc; |
87 | bool more_to_follow; | |
e84346b7 | 88 | bool needs_out_seq; |
31b8006e | 89 | int front_max; |
4cf9d544 | 90 | unsigned long ack_stamp; /* tx: when we were acked */ |
31b8006e SW |
91 | |
92 | struct ceph_msgpool *pool; | |
93 | }; | |
94 | ||
95 | struct ceph_msg_pos { | |
96 | int page, page_pos; /* which page; offset in page */ | |
97 | int data_pos; /* offset in data payload */ | |
bca064d2 | 98 | bool did_page_crc; /* true if we've calculated crc for current page */ |
31b8006e SW |
99 | }; |
100 | ||
101 | /* ceph connection fault delay defaults, for exponential backoff */ | |
102 | #define BASE_DELAY_INTERVAL (HZ/2) | |
103 | #define MAX_DELAY_INTERVAL (5 * 60 * HZ) | |
104 | ||
105 | /* | |
928443cd | 106 | * ceph_connection flag bits |
31b8006e | 107 | */ |
928443cd | 108 | |
31b8006e | 109 | #define LOSSYTX 0 /* we can close channel or drop messages on errors */ |
31b8006e SW |
110 | #define KEEPALIVE_PENDING 3 |
111 | #define WRITE_PENDING 4 /* we have data ready to send */ | |
928443cd AE |
112 | #define SOCK_CLOSED 11 /* socket state changed to closed */ |
113 | #define BACKOFF 15 | |
114 | ||
115 | /* | |
116 | * ceph_connection states | |
117 | */ | |
118 | #define CONNECTING 1 | |
119 | #define NEGOTIATING 2 | |
31b8006e SW |
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 */ | |
31b8006e | 124 | #define OPENING 13 /* open connection w/ (possibly new) peer */ |
31b8006e SW |
125 | |
126 | /* | |
127 | * A single connection with another host. | |
128 | * | |
129 | * We maintain a queue of outgoing messages, and some session state to | |
130 | * ensure that we can preserve the lossless, ordered delivery of | |
131 | * messages in the case of a TCP disconnect. | |
132 | */ | |
133 | struct ceph_connection { | |
134 | void *private; | |
135 | atomic_t nref; | |
136 | ||
137 | const struct ceph_connection_operations *ops; | |
138 | ||
139 | struct ceph_messenger *msgr; | |
ce2c8903 AE |
140 | |
141 | atomic_t sock_state; | |
31b8006e | 142 | struct socket *sock; |
ce2c8903 AE |
143 | struct ceph_entity_addr peer_addr; /* peer address */ |
144 | struct ceph_entity_addr peer_addr_for_me; | |
145 | ||
928443cd AE |
146 | unsigned long flags; |
147 | unsigned long state; | |
31b8006e SW |
148 | const char *error_msg; /* error message, if any */ |
149 | ||
31b8006e | 150 | struct ceph_entity_name peer_name; /* peer name */ |
ce2c8903 | 151 | |
aba558e2 | 152 | unsigned peer_features; |
31b8006e SW |
153 | u32 connect_seq; /* identify the most recent connection |
154 | attempt for this connection, client */ | |
155 | u32 peer_global_seq; /* peer's global seq for this connection */ | |
156 | ||
4e7a5dcd SW |
157 | int auth_retry; /* true if we need a newer authorizer */ |
158 | void *auth_reply_buf; /* where to put the authorizer reply */ | |
159 | int auth_reply_buf_len; | |
160 | ||
ec302645 SW |
161 | struct mutex mutex; |
162 | ||
31b8006e | 163 | /* out queue */ |
31b8006e SW |
164 | struct list_head out_queue; |
165 | struct list_head out_sent; /* sending or sent but unacked */ | |
166 | u64 out_seq; /* last message queued for send */ | |
31b8006e SW |
167 | |
168 | u64 in_seq, in_seq_acked; /* last message received, acked */ | |
169 | ||
170 | /* connection negotiation temps */ | |
171 | char in_banner[CEPH_BANNER_MAX_LEN]; | |
172 | union { | |
173 | struct { /* outgoing connection */ | |
174 | struct ceph_msg_connect out_connect; | |
175 | struct ceph_msg_connect_reply in_reply; | |
176 | }; | |
177 | struct { /* incoming */ | |
178 | struct ceph_msg_connect in_connect; | |
179 | struct ceph_msg_connect_reply out_reply; | |
180 | }; | |
181 | }; | |
182 | struct ceph_entity_addr actual_peer_addr; | |
183 | ||
184 | /* message out temps */ | |
185 | struct ceph_msg *out_msg; /* sending message (== tail of | |
186 | out_sent) */ | |
c86a2930 | 187 | bool out_msg_done; |
31b8006e SW |
188 | struct ceph_msg_pos out_msg_pos; |
189 | ||
190 | struct kvec out_kvec[8], /* sending header/footer data */ | |
191 | *out_kvec_cur; | |
192 | int out_kvec_left; /* kvec's left in out_kvec */ | |
193 | int out_skip; /* skip this many bytes */ | |
194 | int out_kvec_bytes; /* total bytes left */ | |
195 | bool out_kvec_is_msg; /* kvec refers to out_msg */ | |
196 | int out_more; /* there is more data after the kvecs */ | |
197 | __le64 out_temp_ack; /* for writing an ack */ | |
198 | ||
199 | /* message in temps */ | |
200 | struct ceph_msg_header in_hdr; | |
201 | struct ceph_msg *in_msg; | |
202 | struct ceph_msg_pos in_msg_pos; | |
203 | u32 in_front_crc, in_middle_crc, in_data_crc; /* calculated crc */ | |
204 | ||
205 | char in_tag; /* protocol control byte */ | |
206 | int in_base_pos; /* bytes read */ | |
207 | __le64 in_temp_ack; /* for reading an ack */ | |
208 | ||
209 | struct delayed_work work; /* send|recv work */ | |
210 | unsigned long delay; /* current delay interval */ | |
211 | }; | |
212 | ||
213 | ||
3d14c5d2 | 214 | extern const char *ceph_pr_addr(const struct sockaddr_storage *ss); |
31b8006e SW |
215 | extern int ceph_parse_ips(const char *c, const char *end, |
216 | struct ceph_entity_addr *addr, | |
217 | int max_count, int *count); | |
218 | ||
219 | ||
220 | extern int ceph_msgr_init(void); | |
221 | extern void ceph_msgr_exit(void); | |
a922d38f | 222 | extern void ceph_msgr_flush(void); |
31b8006e | 223 | |
15d9882c AE |
224 | extern void ceph_messenger_init(struct ceph_messenger *msgr, |
225 | struct ceph_entity_addr *myaddr, | |
226 | u32 supported_features, | |
227 | u32 required_features, | |
228 | bool nocrc); | |
31b8006e SW |
229 | |
230 | extern void ceph_con_init(struct ceph_messenger *msgr, | |
231 | struct ceph_connection *con); | |
31b8006e SW |
232 | extern void ceph_con_open(struct ceph_connection *con, |
233 | struct ceph_entity_addr *addr); | |
87b315a5 | 234 | extern bool ceph_con_opened(struct ceph_connection *con); |
31b8006e SW |
235 | extern void ceph_con_close(struct ceph_connection *con); |
236 | extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg); | |
237 | extern void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg); | |
0d59ab81 YS |
238 | extern void ceph_con_revoke_message(struct ceph_connection *con, |
239 | struct ceph_msg *msg); | |
31b8006e SW |
240 | extern void ceph_con_keepalive(struct ceph_connection *con); |
241 | extern struct ceph_connection *ceph_con_get(struct ceph_connection *con); | |
242 | extern void ceph_con_put(struct ceph_connection *con); | |
243 | ||
b61c2763 SW |
244 | extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags, |
245 | bool can_fail); | |
31b8006e SW |
246 | extern void ceph_msg_kfree(struct ceph_msg *m); |
247 | ||
31b8006e SW |
248 | |
249 | static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) | |
250 | { | |
c2e552e7 | 251 | kref_get(&msg->kref); |
31b8006e SW |
252 | return msg; |
253 | } | |
c2e552e7 SW |
254 | extern void ceph_msg_last_put(struct kref *kref); |
255 | static inline void ceph_msg_put(struct ceph_msg *msg) | |
256 | { | |
257 | kref_put(&msg->kref, ceph_msg_last_put); | |
258 | } | |
31b8006e | 259 | |
9ec7cab1 SW |
260 | extern void ceph_msg_dump(struct ceph_msg *msg); |
261 | ||
31b8006e | 262 | #endif |