1 #ifndef __FS_CEPH_MESSENGER_H
2 #define __FS_CEPH_MESSENGER_H
3
4 #include <linux/kref.h>
5 #include <linux/mutex.h>
6 #include <linux/net.h>
7 #include <linux/radix-tree.h>
8 #include <linux/uio.h>
9 #include <linux/workqueue.h>
10
11 #include "types.h"
12 #include "buffer.h"
13
14 struct ceph_msg;
15 struct ceph_connection;
16
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
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);
33
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,
42 struct ceph_msg_header *hdr,
43 int *skip);
44 };
45
46 /* use format string %s%d */
47 #define ENTITY_NAME(n) ceph_entity_type_name((n).type), le64_to_cpu((n).num)
48
49 struct ceph_messenger {
50 struct ceph_entity_inst inst; /* my name+address */
51 struct ceph_entity_addr my_enc_addr;
52
53 atomic_t stopping;
54 bool nocrc;
55
56 /*
57 * the global_seq counts connections i (attempt to) initiate
58 * in order to disambiguate certain connect race conditions.
59 */
60 u32 global_seq;
61 spinlock_t global_seq_lock;
62
63 u32 supported_features;
64 u32 required_features;
65 };
66
67 /*
68 * a single message. it contains a header (src, dest, message type, etc.),
69 * footer (crc values, mainly), a "front" message body, and possibly a
70 * data payload (stored in some number of pages).
71 */
72 struct ceph_msg {
73 struct ceph_msg_header hdr; /* header */
74 struct ceph_msg_footer footer; /* footer */
75 struct kvec front; /* unaligned blobs of message */
76 struct ceph_buffer *middle;
77 struct page **pages; /* data payload. NOT OWNER. */
78 unsigned nr_pages; /* size of page array */
79 unsigned page_alignment; /* io offset in first page */
80 struct ceph_pagelist *pagelist; /* instead of pages */
81
82 struct ceph_connection *con;
83 struct list_head list_head;
84
85 struct kref kref;
86 struct bio *bio; /* instead of pages/pagelist */
87 struct bio *bio_iter; /* bio iterator */
88 int bio_seg; /* current bio segment */
89 struct ceph_pagelist *trail; /* the trailing part of the data */
90 bool front_is_vmalloc;
91 bool more_to_follow;
92 bool needs_out_seq;
93 int front_max;
94 unsigned long ack_stamp; /* tx: when we were acked */
95
96 struct ceph_msgpool *pool;
97 };
98
99 struct ceph_msg_pos {
100 int page, page_pos; /* which page; offset in page */
101 int data_pos; /* offset in data payload */
102 bool did_page_crc; /* true if we've calculated crc for current page */
103 };
104
105 /* ceph connection fault delay defaults, for exponential backoff */
106 #define BASE_DELAY_INTERVAL (HZ/2)
107 #define MAX_DELAY_INTERVAL (5 * 60 * HZ)
108
109 /*
110 * A single connection with another host.
111 *
112 * We maintain a queue of outgoing messages, and some session state to
113 * ensure that we can preserve the lossless, ordered delivery of
114 * messages in the case of a TCP disconnect.
115 */
116 struct ceph_connection {
117 void *private;
118
119 const struct ceph_connection_operations *ops;
120
121 struct ceph_messenger *msgr;
122
123 atomic_t sock_state;
124 struct socket *sock;
125 struct ceph_entity_addr peer_addr; /* peer address */
126 struct ceph_entity_addr peer_addr_for_me;
127
128 unsigned long flags;
129 unsigned long state;
130 const char *error_msg; /* error message, if any */
131
132 struct ceph_entity_name peer_name; /* peer name */
133
134 unsigned peer_features;
135 u32 connect_seq; /* identify the most recent connection
136 attempt for this connection, client */
137 u32 peer_global_seq; /* peer's global seq for this connection */
138
139 int auth_retry; /* true if we need a newer authorizer */
140 void *auth_reply_buf; /* where to put the authorizer reply */
141 int auth_reply_buf_len;
142
143 struct mutex mutex;
144
145 /* out queue */
146 struct list_head out_queue;
147 struct list_head out_sent; /* sending or sent but unacked */
148 u64 out_seq; /* last message queued for send */
149
150 u64 in_seq, in_seq_acked; /* last message received, acked */
151
152 /* connection negotiation temps */
153 char in_banner[CEPH_BANNER_MAX_LEN];
154 struct ceph_msg_connect out_connect;
155 struct ceph_msg_connect_reply in_reply;
156 struct ceph_entity_addr actual_peer_addr;
157
158 /* message out temps */
159 struct ceph_msg *out_msg; /* sending message (== tail of
160 out_sent) */
161 bool out_msg_done;
162 struct ceph_msg_pos out_msg_pos;
163
164 struct kvec out_kvec[8], /* sending header/footer data */
165 *out_kvec_cur;
166 int out_kvec_left; /* kvec's left in out_kvec */
167 int out_skip; /* skip this many bytes */
168 int out_kvec_bytes; /* total bytes left */
169 bool out_kvec_is_msg; /* kvec refers to out_msg */
170 int out_more; /* there is more data after the kvecs */
171 __le64 out_temp_ack; /* for writing an ack */
172
173 /* message in temps */
174 struct ceph_msg_header in_hdr;
175 struct ceph_msg *in_msg;
176 struct ceph_msg_pos in_msg_pos;
177 u32 in_front_crc, in_middle_crc, in_data_crc; /* calculated crc */
178
179 char in_tag; /* protocol control byte */
180 int in_base_pos; /* bytes read */
181 __le64 in_temp_ack; /* for reading an ack */
182
183 struct delayed_work work; /* send|recv work */
184 unsigned long delay; /* current delay interval */
185 };
186
187
188 extern const char *ceph_pr_addr(const struct sockaddr_storage *ss);
189 extern int ceph_parse_ips(const char *c, const char *end,
190 struct ceph_entity_addr *addr,
191 int max_count, int *count);
192
193
194 extern int ceph_msgr_init(void);
195 extern void ceph_msgr_exit(void);
196 extern void ceph_msgr_flush(void);
197
198 extern void ceph_messenger_init(struct ceph_messenger *msgr,
199 struct ceph_entity_addr *myaddr,
200 u32 supported_features,
201 u32 required_features,
202 bool nocrc);
203
204 extern void ceph_con_init(struct ceph_connection *con, void *private,
205 const struct ceph_connection_operations *ops,
206 struct ceph_messenger *msgr);
207 extern void ceph_con_open(struct ceph_connection *con,
208 __u8 entity_type, __u64 entity_num,
209 struct ceph_entity_addr *addr);
210 extern bool ceph_con_opened(struct ceph_connection *con);
211 extern void ceph_con_close(struct ceph_connection *con);
212 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
213
214 extern void ceph_msg_revoke(struct ceph_msg *msg);
215 extern void ceph_msg_revoke_incoming(struct ceph_msg *msg);
216
217 extern void ceph_con_keepalive(struct ceph_connection *con);
218
219 extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
220 bool can_fail);
221 extern void ceph_msg_kfree(struct ceph_msg *m);
222
223
ceph_msg_get(struct ceph_msg * msg)224 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
225 {
226 kref_get(&msg->kref);
227 return msg;
228 }
229 extern void ceph_msg_last_put(struct kref *kref);
ceph_msg_put(struct ceph_msg * msg)230 static inline void ceph_msg_put(struct ceph_msg *msg)
231 {
232 kref_put(&msg->kref, ceph_msg_last_put);
233 }
234
235 extern void ceph_msg_dump(struct ceph_msg *msg);
236
237 #endif
238