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/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 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 int (*invalidate_authorizer)(struct ceph_connection *con); 36 37 /* 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, 48 struct ceph_msg_header *hdr, 49 int *skip); 50 }; 51 52 /* use format string %s%d */ 53 #define ENTITY_NAME(n) ceph_entity_type_name((n).type), le64_to_cpu((n).num) 54 55 struct ceph_messenger { 56 struct ceph_entity_inst inst; /* my name+address */ 57 struct ceph_entity_addr my_enc_addr; 58 struct page *zero_page; /* used in certain error cases */ 59 60 bool nocrc; 61 62 /* 63 * the global_seq counts connections i (attempt to) initiate 64 * in order to disambiguate certain connect race conditions. 65 */ 66 u32 global_seq; 67 spinlock_t global_seq_lock; 68 69 u32 supported_features; 70 u32 required_features; 71 }; 72 73 /* 74 * a single message. it contains a header (src, dest, message type, etc.), 75 * footer (crc values, mainly), a "front" message body, and possibly a 76 * data payload (stored in some number of pages). 77 */ 78 struct ceph_msg { 79 struct ceph_msg_header hdr; /* header */ 80 struct ceph_msg_footer footer; /* footer */ 81 struct kvec front; /* unaligned blobs of message */ 82 struct ceph_buffer *middle; 83 struct page **pages; /* data payload. NOT OWNER. */ 84 unsigned nr_pages; /* size of page array */ 85 unsigned page_alignment; /* io offset in first page */ 86 struct ceph_pagelist *pagelist; /* instead of pages */ 87 struct list_head list_head; 88 struct kref kref; 89 struct bio *bio; /* instead of pages/pagelist */ 90 struct bio *bio_iter; /* bio iterator */ 91 int bio_seg; /* current bio segment */ 92 struct ceph_pagelist *trail; /* the trailing part of the data */ 93 bool front_is_vmalloc; 94 bool more_to_follow; 95 bool needs_out_seq; 96 int front_max; 97 98 struct ceph_msgpool *pool; 99 }; 100 101 struct ceph_msg_pos { 102 int page, page_pos; /* which page; offset in page */ 103 int data_pos; /* offset in data payload */ 104 int did_page_crc; /* true if we've calculated crc for current page */ 105 }; 106 107 /* ceph connection fault delay defaults, for exponential backoff */ 108 #define BASE_DELAY_INTERVAL (HZ/2) 109 #define MAX_DELAY_INTERVAL (5 * 60 * HZ) 110 111 /* 112 * ceph_connection state bit flags 113 * 114 * QUEUED and BUSY are used together to ensure that only a single 115 * thread is currently opening, reading or writing data to the socket. 116 */ 117 #define LOSSYTX 0 /* we can close channel or drop messages on errors */ 118 #define CONNECTING 1 119 #define NEGOTIATING 2 120 #define KEEPALIVE_PENDING 3 121 #define WRITE_PENDING 4 /* we have data ready to send */ 122 #define QUEUED 5 /* there is work queued on this connection */ 123 #define BUSY 6 /* work is being done */ 124 #define STANDBY 8 /* no outgoing messages, socket closed. we keep 125 * the ceph_connection around to maintain shared 126 * state with the peer. */ 127 #define CLOSED 10 /* we've closed the connection */ 128 #define SOCK_CLOSED 11 /* socket state changed to closed */ 129 #define OPENING 13 /* open connection w/ (possibly new) peer */ 130 #define DEAD 14 /* dead, about to kfree */ 131 132 /* 133 * A single connection with another host. 134 * 135 * We maintain a queue of outgoing messages, and some session state to 136 * ensure that we can preserve the lossless, ordered delivery of 137 * messages in the case of a TCP disconnect. 138 */ 139 struct ceph_connection { 140 void *private; 141 atomic_t nref; 142 143 const struct ceph_connection_operations *ops; 144 145 struct ceph_messenger *msgr; 146 struct socket *sock; 147 unsigned long state; /* connection state (see flags above) */ 148 const char *error_msg; /* error message, if any */ 149 150 struct ceph_entity_addr peer_addr; /* peer address */ 151 struct ceph_entity_name peer_name; /* peer name */ 152 struct ceph_entity_addr peer_addr_for_me; 153 unsigned peer_features; 154 u32 connect_seq; /* identify the most recent connection 155 attempt for this connection, client */ 156 u32 peer_global_seq; /* peer's global seq for this connection */ 157 158 int auth_retry; /* true if we need a newer authorizer */ 159 void *auth_reply_buf; /* where to put the authorizer reply */ 160 int auth_reply_buf_len; 161 162 struct mutex mutex; 163 164 /* out queue */ 165 struct list_head out_queue; 166 struct list_head out_sent; /* sending or sent but unacked */ 167 u64 out_seq; /* last message queued for send */ 168 bool out_keepalive_pending; 169 170 u64 in_seq, in_seq_acked; /* last message received, acked */ 171 172 /* connection negotiation temps */ 173 char in_banner[CEPH_BANNER_MAX_LEN]; 174 union { 175 struct { /* outgoing connection */ 176 struct ceph_msg_connect out_connect; 177 struct ceph_msg_connect_reply in_reply; 178 }; 179 struct { /* incoming */ 180 struct ceph_msg_connect in_connect; 181 struct ceph_msg_connect_reply out_reply; 182 }; 183 }; 184 struct ceph_entity_addr actual_peer_addr; 185 186 /* message out temps */ 187 struct ceph_msg *out_msg; /* sending message (== tail of 188 out_sent) */ 189 bool out_msg_done; 190 struct ceph_msg_pos out_msg_pos; 191 192 struct kvec out_kvec[8], /* sending header/footer data */ 193 *out_kvec_cur; 194 int out_kvec_left; /* kvec's left in out_kvec */ 195 int out_skip; /* skip this many bytes */ 196 int out_kvec_bytes; /* total bytes left */ 197 bool out_kvec_is_msg; /* kvec refers to out_msg */ 198 int out_more; /* there is more data after the kvecs */ 199 __le64 out_temp_ack; /* for writing an ack */ 200 201 /* message in temps */ 202 struct ceph_msg_header in_hdr; 203 struct ceph_msg *in_msg; 204 struct ceph_msg_pos in_msg_pos; 205 u32 in_front_crc, in_middle_crc, in_data_crc; /* calculated crc */ 206 207 char in_tag; /* protocol control byte */ 208 int in_base_pos; /* bytes read */ 209 __le64 in_temp_ack; /* for reading an ack */ 210 211 struct delayed_work work; /* send|recv work */ 212 unsigned long delay; /* current delay interval */ 213 }; 214 215 216 extern const char *ceph_pr_addr(const struct sockaddr_storage *ss); 217 extern int ceph_parse_ips(const char *c, const char *end, 218 struct ceph_entity_addr *addr, 219 int max_count, int *count); 220 221 222 extern int ceph_msgr_init(void); 223 extern void ceph_msgr_exit(void); 224 extern void ceph_msgr_flush(void); 225 226 extern struct ceph_messenger *ceph_messenger_create( 227 struct ceph_entity_addr *myaddr, 228 u32 features, u32 required); 229 extern void ceph_messenger_destroy(struct ceph_messenger *); 230 231 extern void ceph_con_init(struct ceph_messenger *msgr, 232 struct ceph_connection *con); 233 extern void ceph_con_open(struct ceph_connection *con, 234 struct ceph_entity_addr *addr); 235 extern bool ceph_con_opened(struct ceph_connection *con); 236 extern void ceph_con_close(struct ceph_connection *con); 237 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg); 238 extern void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg); 239 extern void ceph_con_revoke_message(struct ceph_connection *con, 240 struct ceph_msg *msg); 241 extern void ceph_con_keepalive(struct ceph_connection *con); 242 extern struct ceph_connection *ceph_con_get(struct ceph_connection *con); 243 extern void ceph_con_put(struct ceph_connection *con); 244 245 extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags); 246 extern void ceph_msg_kfree(struct ceph_msg *m); 247 248 249 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg) 250 { 251 kref_get(&msg->kref); 252 return msg; 253 } 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 } 259 260 extern void ceph_msg_dump(struct ceph_msg *msg); 261 262 #endif 263