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
  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         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;
  61
  62         u32 supported_features;
  63         u32 required_features;
  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 */
  78         unsigned page_alignment;        /* io offset in first page */
  79         struct ceph_pagelist *pagelist; /* instead of pages */
  80
  81         struct ceph_connection *con;
  82         struct list_head list_head;
  83
  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
  98 struct 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 flag bits
 110  */
 111
 112 #define LOSSYTX         0  /* we can close channel or drop messages on errors */
 113 #define KEEPALIVE_PENDING      3
 114 #define WRITE_PENDING   4  /* we have data ready to send */
 115 #define SOCK_CLOSED     11 /* socket state changed to closed */
 116 #define BACKOFF         15
 117
 118 /*
 119  * ceph_connection states
 120  */
 121 #define CONNECTING      1
 122 #define NEGOTIATING     2
 123 #define STANDBY         8  /* no outgoing messages, socket closed.  we keep
 124                             * the ceph_connection around to maintain shared
 125                             * state with the peer. */
 126 #define CLOSED          10 /* we've closed the connection */
 127 #define OPENING         13 /* open connection w/ (possibly new) peer */
 128
 129 /*
 130  * A single connection with another host.
 131  *
 132  * We maintain a queue of outgoing messages, and some session state to
 133  * ensure that we can preserve the lossless, ordered delivery of
 134  * messages in the case of a TCP disconnect.
 135  */
 136 struct ceph_connection {
 137         void *private;
 138
 139         const struct ceph_connection_operations *ops;
 140
 141         struct ceph_messenger *msgr;
 142
 143         atomic_t sock_state;
 144         struct socket *sock;
 145         struct ceph_entity_addr peer_addr; /* peer address */
 146         struct ceph_entity_addr peer_addr_for_me;
 147
 148         unsigned long flags;
 149         unsigned long state;
 150         const char *error_msg;  /* error message, if any */
 151
 152         struct ceph_entity_name peer_name; /* peer name */
 153
 154         unsigned peer_features;
 155         u32 connect_seq;      /* identify the most recent connection
 156                                  attempt for this connection, client */
 157         u32 peer_global_seq;  /* peer's global seq for this connection */
 158
 159         int auth_retry;       /* true if we need a newer authorizer */
 160         void *auth_reply_buf;   /* where to put the authorizer reply */
 161         int auth_reply_buf_len;
 162
 163         struct mutex mutex;
 164
 165         /* out queue */
 166         struct list_head out_queue;
 167         struct list_head out_sent;   /* sending or sent but unacked */
 168         u64 out_seq;                 /* last message queued for send */
 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 void ceph_messenger_init(struct ceph_messenger *msgr,
 227                         struct ceph_entity_addr *myaddr,
 228                         u32 supported_features,
 229                         u32 required_features,
 230                         bool nocrc);
 231
 232 extern void ceph_con_init(struct ceph_connection *con, void *private,
 233                         const struct ceph_connection_operations *ops,
 234                         struct ceph_messenger *msgr, __u8 entity_type,
 235                         __u64 entity_num);
 236 extern void ceph_con_open(struct ceph_connection *con,
 237                           struct ceph_entity_addr *addr);
 238 extern bool ceph_con_opened(struct ceph_connection *con);
 239 extern void ceph_con_close(struct ceph_connection *con);
 240 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
 241
 242 extern void ceph_msg_revoke(struct ceph_msg *msg);
 243 extern void ceph_msg_revoke_incoming(struct ceph_msg *msg);
 244
 245 extern void ceph_con_keepalive(struct ceph_connection *con);
 246 extern struct ceph_connection *ceph_con_get(struct ceph_connection *con);
 247 extern void ceph_con_put(struct ceph_connection *con);
 248
 249 extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
 250                                      bool can_fail);
 251 extern void ceph_msg_kfree(struct ceph_msg *m);
 252
 253
 254 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
 255 {
 256         kref_get(&msg->kref);
 257         return msg;
 258 }
 259 extern void ceph_msg_last_put(struct kref *kref);
 260 static inline void ceph_msg_put(struct ceph_msg *msg)
 261 {
 262         kref_put(&msg->kref, ceph_msg_last_put);
 263 }
 264
 265 extern void ceph_msg_dump(struct ceph_msg *msg);
 266
 267 #endif