blob: a5caf91cc97178039689b1f0efc49ef42e56d981 [file] [log] [blame]
Sage Weil31b80062009-10-06 11:31:13 -07001#ifndef __FS_CEPH_MESSENGER_H
2#define __FS_CEPH_MESSENGER_H
3
Sage Weilc2e552e2009-12-07 15:55:05 -08004#include <linux/kref.h>
Sage Weil31b80062009-10-06 11:31:13 -07005#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
15struct ceph_msg;
16struct ceph_connection;
17
18extern struct workqueue_struct *ceph_msgr_wq; /* receive work queue */
19
20/*
21 * Ceph defines these callbacks for handling connection events.
22 */
23struct 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
Sage Weil4e7a5dc2009-11-18 16:19:57 -080030 /* 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);
Sage Weil9bd2e6f2010-02-02 16:21:06 -080035 int (*invalidate_authorizer)(struct ceph_connection *con);
Sage Weil4e7a5dc2009-11-18 16:19:57 -080036
Sage Weil31b80062009-10-06 11:31:13 -070037 /* protocol version mismatch */
38 void (*bad_proto) (struct ceph_connection *con);
39
40 /* there was some error on the socket (disconnect, whatever) */
41 void (*fault) (struct ceph_connection *con);
42
43 /* a remote host as terminated a message exchange session, and messages
44 * we sent (or they tried to send us) may be lost. */
45 void (*peer_reset) (struct ceph_connection *con);
46
47 struct ceph_msg * (*alloc_msg) (struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -080048 struct ceph_msg_header *hdr,
49 int *skip);
Sage Weil31b80062009-10-06 11:31:13 -070050};
51
52extern const char *ceph_name_type_str(int t);
53
54/* use format string %s%d */
55#define ENTITY_NAME(n) ceph_name_type_str((n).type), le64_to_cpu((n).num)
56
57struct ceph_messenger {
58 struct ceph_entity_inst inst; /* my name+address */
Sage Weil63f2d212009-11-03 15:17:56 -080059 struct ceph_entity_addr my_enc_addr;
Sage Weil31b80062009-10-06 11:31:13 -070060 struct page *zero_page; /* used in certain error cases */
61
62 bool nocrc;
63
64 /*
65 * the global_seq counts connections i (attempt to) initiate
66 * in order to disambiguate certain connect race conditions.
67 */
68 u32 global_seq;
69 spinlock_t global_seq_lock;
70};
71
72/*
73 * a single message. it contains a header (src, dest, message type, etc.),
74 * footer (crc values, mainly), a "front" message body, and possibly a
75 * data payload (stored in some number of pages).
76 */
77struct ceph_msg {
78 struct ceph_msg_header hdr; /* header */
79 struct ceph_msg_footer footer; /* footer */
80 struct kvec front; /* unaligned blobs of message */
81 struct ceph_buffer *middle;
82 struct page **pages; /* data payload. NOT OWNER. */
83 unsigned nr_pages; /* size of page array */
Sage Weil58bb3b32009-12-23 12:12:31 -080084 struct ceph_pagelist *pagelist; /* instead of pages */
Sage Weil31b80062009-10-06 11:31:13 -070085 struct list_head list_head;
Sage Weilc2e552e2009-12-07 15:55:05 -080086 struct kref kref;
Sage Weil31b80062009-10-06 11:31:13 -070087 bool front_is_vmalloc;
88 bool more_to_follow;
Sage Weile84346b2010-05-11 21:20:38 -070089 bool needs_out_seq;
Sage Weil31b80062009-10-06 11:31:13 -070090 int front_max;
91
92 struct ceph_msgpool *pool;
93};
94
95struct ceph_msg_pos {
96 int page, page_pos; /* which page; offset in page */
97 int data_pos; /* offset in data payload */
98 int did_page_crc; /* true if we've calculated crc for current page */
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/*
106 * ceph_connection state bit flags
107 *
108 * QUEUED and BUSY are used together to ensure that only a single
109 * thread is currently opening, reading or writing data to the socket.
110 */
111#define LOSSYTX 0 /* we can close channel or drop messages on errors */
Sage Weileed0ef22009-11-10 14:34:36 -0800112#define CONNECTING 1
113#define NEGOTIATING 2
Sage Weil31b80062009-10-06 11:31:13 -0700114#define KEEPALIVE_PENDING 3
115#define WRITE_PENDING 4 /* we have data ready to send */
116#define QUEUED 5 /* there is work queued on this connection */
117#define BUSY 6 /* work is being done */
118#define STANDBY 8 /* no outgoing messages, socket closed. we keep
119 * the ceph_connection around to maintain shared
120 * state with the peer. */
121#define CLOSED 10 /* we've closed the connection */
122#define SOCK_CLOSED 11 /* socket state changed to closed */
Sage Weil31b80062009-10-06 11:31:13 -0700123#define OPENING 13 /* open connection w/ (possibly new) peer */
124#define DEAD 14 /* dead, about to kfree */
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 */
133struct ceph_connection {
134 void *private;
135 atomic_t nref;
136
137 const struct ceph_connection_operations *ops;
138
139 struct ceph_messenger *msgr;
140 struct socket *sock;
141 unsigned long state; /* connection state (see flags above) */
142 const char *error_msg; /* error message, if any */
143
144 struct ceph_entity_addr peer_addr; /* peer address */
145 struct ceph_entity_name peer_name; /* peer name */
146 struct ceph_entity_addr peer_addr_for_me;
147 u32 connect_seq; /* identify the most recent connection
148 attempt for this connection, client */
149 u32 peer_global_seq; /* peer's global seq for this connection */
150
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800151 int auth_retry; /* true if we need a newer authorizer */
152 void *auth_reply_buf; /* where to put the authorizer reply */
153 int auth_reply_buf_len;
154
Sage Weilec302642009-12-22 10:43:42 -0800155 struct mutex mutex;
156
Sage Weil31b80062009-10-06 11:31:13 -0700157 /* out queue */
Sage Weil31b80062009-10-06 11:31:13 -0700158 struct list_head out_queue;
159 struct list_head out_sent; /* sending or sent but unacked */
160 u64 out_seq; /* last message queued for send */
161 u64 out_seq_sent; /* last message sent */
162 bool out_keepalive_pending;
163
164 u64 in_seq, in_seq_acked; /* last message received, acked */
165
166 /* connection negotiation temps */
167 char in_banner[CEPH_BANNER_MAX_LEN];
168 union {
169 struct { /* outgoing connection */
170 struct ceph_msg_connect out_connect;
171 struct ceph_msg_connect_reply in_reply;
172 };
173 struct { /* incoming */
174 struct ceph_msg_connect in_connect;
175 struct ceph_msg_connect_reply out_reply;
176 };
177 };
178 struct ceph_entity_addr actual_peer_addr;
179
180 /* message out temps */
181 struct ceph_msg *out_msg; /* sending message (== tail of
182 out_sent) */
Sage Weilc86a2932009-12-14 14:04:30 -0800183 bool out_msg_done;
Sage Weil31b80062009-10-06 11:31:13 -0700184 struct ceph_msg_pos out_msg_pos;
185
186 struct kvec out_kvec[8], /* sending header/footer data */
187 *out_kvec_cur;
188 int out_kvec_left; /* kvec's left in out_kvec */
189 int out_skip; /* skip this many bytes */
190 int out_kvec_bytes; /* total bytes left */
191 bool out_kvec_is_msg; /* kvec refers to out_msg */
192 int out_more; /* there is more data after the kvecs */
193 __le64 out_temp_ack; /* for writing an ack */
194
195 /* message in temps */
196 struct ceph_msg_header in_hdr;
197 struct ceph_msg *in_msg;
198 struct ceph_msg_pos in_msg_pos;
199 u32 in_front_crc, in_middle_crc, in_data_crc; /* calculated crc */
200
201 char in_tag; /* protocol control byte */
202 int in_base_pos; /* bytes read */
203 __le64 in_temp_ack; /* for reading an ack */
204
205 struct delayed_work work; /* send|recv work */
206 unsigned long delay; /* current delay interval */
207};
208
209
210extern const char *pr_addr(const struct sockaddr_storage *ss);
211extern int ceph_parse_ips(const char *c, const char *end,
212 struct ceph_entity_addr *addr,
213 int max_count, int *count);
214
215
216extern int ceph_msgr_init(void);
217extern void ceph_msgr_exit(void);
218
219extern struct ceph_messenger *ceph_messenger_create(
220 struct ceph_entity_addr *myaddr);
221extern void ceph_messenger_destroy(struct ceph_messenger *);
222
223extern void ceph_con_init(struct ceph_messenger *msgr,
224 struct ceph_connection *con);
Sage Weil31b80062009-10-06 11:31:13 -0700225extern void ceph_con_open(struct ceph_connection *con,
226 struct ceph_entity_addr *addr);
Sage Weil87b315a2010-03-22 14:51:18 -0700227extern bool ceph_con_opened(struct ceph_connection *con);
Sage Weil31b80062009-10-06 11:31:13 -0700228extern void ceph_con_close(struct ceph_connection *con);
229extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
230extern void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg);
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -0800231extern void ceph_con_revoke_message(struct ceph_connection *con,
232 struct ceph_msg *msg);
Sage Weil31b80062009-10-06 11:31:13 -0700233extern void ceph_con_keepalive(struct ceph_connection *con);
234extern struct ceph_connection *ceph_con_get(struct ceph_connection *con);
235extern void ceph_con_put(struct ceph_connection *con);
236
237extern struct ceph_msg *ceph_msg_new(int type, int front_len,
238 int page_len, int page_off,
239 struct page **pages);
240extern void ceph_msg_kfree(struct ceph_msg *m);
241
Sage Weil31b80062009-10-06 11:31:13 -0700242
243static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
244{
Sage Weilc2e552e2009-12-07 15:55:05 -0800245 kref_get(&msg->kref);
Sage Weil31b80062009-10-06 11:31:13 -0700246 return msg;
247}
Sage Weilc2e552e2009-12-07 15:55:05 -0800248extern void ceph_msg_last_put(struct kref *kref);
249static inline void ceph_msg_put(struct ceph_msg *msg)
250{
251 kref_put(&msg->kref, ceph_msg_last_put);
252}
Sage Weil31b80062009-10-06 11:31:13 -0700253
Sage Weil9ec7cab2009-12-14 15:13:47 -0800254extern void ceph_msg_dump(struct ceph_msg *msg);
255
Sage Weil31b80062009-10-06 11:31:13 -0700256#endif