/include/linux/ceph/messenger.h

  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
 14struct ceph_msg;
 15struct ceph_connection;
 16
 17/*
 18 * Ceph defines these callbacks for handling connection events.
 19 */
 20struct 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	int (*get_authorizer) (struct ceph_connection *con,
 29			       void **buf, int *len, int *proto,
 30			       void **reply_buf, int *reply_len, int force_new);
 31	int (*verify_authorizer_reply) (struct ceph_connection *con, int len);
 32	int (*invalidate_authorizer)(struct ceph_connection *con);
 33
 34	/* protocol version mismatch */
 35	void (*bad_proto) (struct ceph_connection *con);
 36
 37	/* there was some error on the socket (disconnect, whatever) */
 38	void (*fault) (struct ceph_connection *con);
 39
 40	/* a remote host as terminated a message exchange session, and messages
 41	 * we sent (or they tried to send us) may be lost. */
 42	void (*peer_reset) (struct ceph_connection *con);
 43
 44	struct ceph_msg * (*alloc_msg) (struct ceph_connection *con,
 45					struct ceph_msg_header *hdr,
 46					int *skip);
 47};
 48
 49/* use format string %s%d */
 50#define ENTITY_NAME(n) ceph_entity_type_name((n).type), le64_to_cpu((n).num)
 51
 52struct ceph_messenger {
 53	struct ceph_entity_inst inst;    /* my name+address */
 54	struct ceph_entity_addr my_enc_addr;
 55
 56	bool nocrc;
 57
 58	/*
 59	 * the global_seq counts connections i (attempt to) initiate
 60	 * in order to disambiguate certain connect race conditions.
 61	 */
 62	u32 global_seq;
 63	spinlock_t global_seq_lock;
 64
 65	u32 supported_features;
 66	u32 required_features;
 67};
 68
 69/*
 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).
 73 */
 74struct ceph_msg {
 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	unsigned page_alignment;        /* io offset in first page */
 82	struct ceph_pagelist *pagelist; /* instead of pages */
 83	struct list_head list_head;
 84	struct kref kref;
 85	struct bio  *bio;		/* instead of pages/pagelist */
 86	struct bio  *bio_iter;		/* bio iterator */
 87	int bio_seg;			/* current bio segment */
 88	struct ceph_pagelist *trail;	/* the trailing part of the data */
 89	bool front_is_vmalloc;
 90	bool more_to_follow;
 91	bool needs_out_seq;
 92	int front_max;
 93	unsigned long ack_stamp;        /* tx: when we were acked */
 94
 95	struct ceph_msgpool *pool;
 96};
 97
 98struct ceph_msg_pos {
 99	int page, page_pos;  /* which page; offset in page */
100	int data_pos;        /* offset in data payload */
101	bool did_page_crc;   /* true if we've calculated crc for current page */
102};
103
104/* ceph connection fault delay defaults, for exponential backoff */
105#define BASE_DELAY_INTERVAL	(HZ/2)
106#define MAX_DELAY_INTERVAL	(5 * 60 * HZ)
107
108/*
109 * ceph_connection state bit flags
110 */
111#define LOSSYTX         0  /* we can close channel or drop messages on errors */
112#define CONNECTING	1
113#define NEGOTIATING	2
114#define KEEPALIVE_PENDING      3
115#define WRITE_PENDING	4  /* we have data ready to send */
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 OPENING         13 /* open connection w/ (possibly new) peer */
122#define DEAD            14 /* dead, about to kfree */
123#define BACKOFF         15
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 */
132struct 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	unsigned peer_features;
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
151	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
155	struct mutex mutex;
156
157	/* out queue */
158	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
162	u64 in_seq, in_seq_acked;  /* last message received, acked */
163
164	/* connection negotiation temps */
165	char in_banner[CEPH_BANNER_MAX_LEN];
166	union {
167		struct {  /* outgoing connection */
168			struct ceph_msg_connect out_connect;
169			struct ceph_msg_connect_reply in_reply;
170		};
171		struct {  /* incoming */
172			struct ceph_msg_connect in_connect;
173			struct ceph_msg_connect_reply out_reply;
174		};
175	};
176	struct ceph_entity_addr actual_peer_addr;
177
178	/* message out temps */
179	struct ceph_msg *out_msg;        /* sending message (== tail of
180					    out_sent) */
181	bool out_msg_done;
182	struct ceph_msg_pos out_msg_pos;
183
184	struct kvec out_kvec[8],         /* sending header/footer data */
185		*out_kvec_cur;
186	int out_kvec_left;   /* kvec's left in out_kvec */
187	int out_skip;        /* skip this many bytes */
188	int out_kvec_bytes;  /* total bytes left */
189	bool out_kvec_is_msg; /* kvec refers to out_msg */
190	int out_more;        /* there is more data after the kvecs */
191	__le64 out_temp_ack; /* for writing an ack */
192
193	/* message in temps */
194	struct ceph_msg_header in_hdr;
195	struct ceph_msg *in_msg;
196	struct ceph_msg_pos in_msg_pos;
197	u32 in_front_crc, in_middle_crc, in_data_crc;  /* calculated crc */
198
199	char in_tag;         /* protocol control byte */
200	int in_base_pos;     /* bytes read */
201	__le64 in_temp_ack;  /* for reading an ack */
202
203	struct delayed_work work;	    /* send|recv work */
204	unsigned long       delay;          /* current delay interval */
205};
206
207
208extern const char *ceph_pr_addr(const struct sockaddr_storage *ss);
209extern int ceph_parse_ips(const char *c, const char *end,
210			  struct ceph_entity_addr *addr,
211			  int max_count, int *count);
212
213
214extern int ceph_msgr_init(void);
215extern void ceph_msgr_exit(void);
216extern void ceph_msgr_flush(void);
217
218extern struct ceph_messenger *ceph_messenger_create(
219	struct ceph_entity_addr *myaddr,
220	u32 features, u32 required);
221extern void ceph_messenger_destroy(struct ceph_messenger *);
222
223extern void ceph_con_init(struct ceph_messenger *msgr,
224			  struct ceph_connection *con);
225extern void ceph_con_open(struct ceph_connection *con,
226			  struct ceph_entity_addr *addr);
227extern bool ceph_con_opened(struct ceph_connection *con);
228extern 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);
231extern void ceph_con_revoke_message(struct ceph_connection *con,
232				  struct ceph_msg *msg);
233extern 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, gfp_t flags,
238				     bool can_fail);
239extern void ceph_msg_kfree(struct ceph_msg *m);
240
241
242static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
243{
244	kref_get(&msg->kref);
245	return msg;
246}
247extern void ceph_msg_last_put(struct kref *kref);
248static inline void ceph_msg_put(struct ceph_msg *msg)
249{
250	kref_put(&msg->kref, ceph_msg_last_put);
251}
252
253extern void ceph_msg_dump(struct ceph_msg *msg);
254
255#endif