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 <linux/ceph/types.h>
  12 #include <linux/ceph/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 #define ceph_msg_has_data(m)            ((m)->data.type != CEPH_MSG_DATA_NONE)
  68
  69 enum ceph_msg_data_type {
  70         CEPH_MSG_DATA_NONE,     /* message contains no data payload */
  71         CEPH_MSG_DATA_PAGES,    /* data source/destination is a page array */
  72         CEPH_MSG_DATA_PAGELIST, /* data source/destination is a pagelist */
  73 #ifdef CONFIG_BLOCK
  74         CEPH_MSG_DATA_BIO,      /* data source/destination is a bio list */
  75 #endif /* CONFIG_BLOCK */
  76 };
  77
  78 static __inline__ bool ceph_msg_data_type_valid(enum ceph_msg_data_type type)
  79 {
  80         switch (type) {
  81         case CEPH_MSG_DATA_NONE:
  82         case CEPH_MSG_DATA_PAGES:
  83         case CEPH_MSG_DATA_PAGELIST:
  84 #ifdef CONFIG_BLOCK
  85         case CEPH_MSG_DATA_BIO:
  86 #endif /* CONFIG_BLOCK */
  87                 return true;
  88         default:
  89                 return false;
  90         }
  91 }
  92
  93 struct ceph_msg_data_cursor {
  94         size_t          resid;          /* bytes not yet consumed */
  95         bool            last_piece;     /* now at last piece of data item */
  96         union {
  97 #ifdef CONFIG_BLOCK
  98                 struct {                                /* bio */
  99                         struct bio      *bio;           /* bio from list */
 100                         unsigned int    vector_index;   /* vector from bio */
 101                         unsigned int    vector_offset;  /* bytes from vector */
 102                 };
 103 #endif /* CONFIG_BLOCK */
 104                 struct {                                /* pages */
 105                         unsigned int    page_offset;    /* offset in page */
 106                         unsigned short  page_index;     /* index in array */
 107                         unsigned short  page_count;     /* pages in array */
 108                 };
 109                 struct {                                /* pagelist */
 110                         struct page     *page;          /* page from list */
 111                         size_t          offset;         /* bytes from list */
 112                 };
 113         };
 114 };
 115
 116 struct ceph_msg_data {
 117         enum ceph_msg_data_type         type;
 118         union {
 119 #ifdef CONFIG_BLOCK
 120                 struct bio              *bio;
 121 #endif /* CONFIG_BLOCK */
 122                 struct {
 123                         struct page     **pages;        /* NOT OWNER. */
 124                         size_t          length;         /* total # bytes */
 125                         unsigned int    alignment;      /* first page */
 126                 };
 127                 struct ceph_pagelist    *pagelist;
 128         };
 129         struct ceph_msg_data_cursor     cursor;         /* pagelist only */
 130 };
 131
 132 /*
 133  * a single message.  it contains a header (src, dest, message type, etc.),
 134  * footer (crc values, mainly), a "front" message body, and possibly a
 135  * data payload (stored in some number of pages).
 136  */
 137 struct ceph_msg {
 138         struct ceph_msg_header hdr;     /* header */
 139         struct ceph_msg_footer footer;  /* footer */
 140         struct kvec front;              /* unaligned blobs of message */
 141         struct ceph_buffer *middle;
 142
 143         /* data payload */
 144         struct ceph_msg_data    data;
 145
 146         struct ceph_connection *con;
 147         struct list_head list_head;     /* links for connection lists */
 148
 149         struct kref kref;
 150         bool front_is_vmalloc;
 151         bool more_to_follow;
 152         bool needs_out_seq;
 153         int front_max;
 154         unsigned long ack_stamp;        /* tx: when we were acked */
 155
 156         struct ceph_msgpool *pool;
 157 };
 158
 159 struct ceph_msg_pos {
 160         bool did_page_crc;   /* true if we've calculated crc for current page */
 161 };
 162
 163 /* ceph connection fault delay defaults, for exponential backoff */
 164 #define BASE_DELAY_INTERVAL     (HZ/2)
 165 #define MAX_DELAY_INTERVAL      (5 * 60 * HZ)
 166
 167 /*
 168  * A single connection with another host.
 169  *
 170  * We maintain a queue of outgoing messages, and some session state to
 171  * ensure that we can preserve the lossless, ordered delivery of
 172  * messages in the case of a TCP disconnect.
 173  */
 174 struct ceph_connection {
 175         void *private;
 176
 177         const struct ceph_connection_operations *ops;
 178
 179         struct ceph_messenger *msgr;
 180
 181         atomic_t sock_state;
 182         struct socket *sock;
 183         struct ceph_entity_addr peer_addr; /* peer address */
 184         struct ceph_entity_addr peer_addr_for_me;
 185
 186         unsigned long flags;
 187         unsigned long state;
 188         const char *error_msg;  /* error message, if any */
 189
 190         struct ceph_entity_name peer_name; /* peer name */
 191
 192         unsigned peer_features;
 193         u32 connect_seq;      /* identify the most recent connection
 194                                  attempt for this connection, client */
 195         u32 peer_global_seq;  /* peer's global seq for this connection */
 196
 197         int auth_retry;       /* true if we need a newer authorizer */
 198         void *auth_reply_buf;   /* where to put the authorizer reply */
 199         int auth_reply_buf_len;
 200
 201         struct mutex mutex;
 202
 203         /* out queue */
 204         struct list_head out_queue;
 205         struct list_head out_sent;   /* sending or sent but unacked */
 206         u64 out_seq;                 /* last message queued for send */
 207
 208         u64 in_seq, in_seq_acked;  /* last message received, acked */
 209
 210         /* connection negotiation temps */
 211         char in_banner[CEPH_BANNER_MAX_LEN];
 212         struct ceph_msg_connect out_connect;
 213         struct ceph_msg_connect_reply in_reply;
 214         struct ceph_entity_addr actual_peer_addr;
 215
 216         /* message out temps */
 217         struct ceph_msg *out_msg;        /* sending message (== tail of
 218                                             out_sent) */
 219         bool out_msg_done;
 220         struct ceph_msg_pos out_msg_pos;
 221
 222         struct kvec out_kvec[8],         /* sending header/footer data */
 223                 *out_kvec_cur;
 224         int out_kvec_left;   /* kvec's left in out_kvec */
 225         int out_skip;        /* skip this many bytes */
 226         int out_kvec_bytes;  /* total bytes left */
 227         bool out_kvec_is_msg; /* kvec refers to out_msg */
 228         int out_more;        /* there is more data after the kvecs */
 229         __le64 out_temp_ack; /* for writing an ack */
 230
 231         /* message in temps */
 232         struct ceph_msg_header in_hdr;
 233         struct ceph_msg *in_msg;
 234         struct ceph_msg_pos in_msg_pos;
 235         u32 in_front_crc, in_middle_crc, in_data_crc;  /* calculated crc */
 236
 237         char in_tag;         /* protocol control byte */
 238         int in_base_pos;     /* bytes read */
 239         __le64 in_temp_ack;  /* for reading an ack */
 240
 241         struct delayed_work work;           /* send|recv work */
 242         unsigned long       delay;          /* current delay interval */
 243 };
 244
 245
 246 extern const char *ceph_pr_addr(const struct sockaddr_storage *ss);
 247 extern int ceph_parse_ips(const char *c, const char *end,
 248                           struct ceph_entity_addr *addr,
 249                           int max_count, int *count);
 250
 251
 252 extern int ceph_msgr_init(void);
 253 extern void ceph_msgr_exit(void);
 254 extern void ceph_msgr_flush(void);
 255
 256 extern void ceph_messenger_init(struct ceph_messenger *msgr,
 257                         struct ceph_entity_addr *myaddr,
 258                         u32 supported_features,
 259                         u32 required_features,
 260                         bool nocrc);
 261
 262 extern void ceph_con_init(struct ceph_connection *con, void *private,
 263                         const struct ceph_connection_operations *ops,
 264                         struct ceph_messenger *msgr);
 265 extern void ceph_con_open(struct ceph_connection *con,
 266                           __u8 entity_type, __u64 entity_num,
 267                           struct ceph_entity_addr *addr);
 268 extern bool ceph_con_opened(struct ceph_connection *con);
 269 extern void ceph_con_close(struct ceph_connection *con);
 270 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
 271
 272 extern void ceph_msg_revoke(struct ceph_msg *msg);
 273 extern void ceph_msg_revoke_incoming(struct ceph_msg *msg);
 274
 275 extern void ceph_con_keepalive(struct ceph_connection *con);
 276
 277 extern void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
 278                                 size_t length, size_t alignment);
 279 extern void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
 280                                 struct ceph_pagelist *pagelist);
 281 extern void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio);
 282
 283 extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
 284                                      bool can_fail);
 285 extern void ceph_msg_kfree(struct ceph_msg *m);
 286
 287
 288 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
 289 {
 290         kref_get(&msg->kref);
 291         return msg;
 292 }
 293 extern void ceph_msg_last_put(struct kref *kref);
 294 static inline void ceph_msg_put(struct ceph_msg *msg)
 295 {
 296         kref_put(&msg->kref, ceph_msg_last_put);
 297 }
 298
 299 extern void ceph_msg_dump(struct ceph_msg *msg);
 300
 301 #endif