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 extern struct workqueue_struct *ceph_msgr_wq;       /* receive work queue */
  18
  19 /*
  20  * Ceph defines these callbacks for handling connection events.
  21  */
  22 struct ceph_connection_operations {
  23         struct ceph_connection *(*get)(struct ceph_connection *);
  24         void (*put)(struct ceph_connection *);
  25
  26         /* handle an incoming message. */
  27         void (*dispatch) (struct ceph_connection *con, struct ceph_msg *m);
  28
  29         /* authorize an outgoing connection */
  30         int (*get_authorizer) (struct ceph_connection *con,
  31                                void **buf, int *len, int *proto,
  32                                void **reply_buf, int *reply_len, int force_new);
  33         int (*verify_authorizer_reply) (struct ceph_connection *con, int len);
  34         int (*invalidate_authorizer)(struct ceph_connection *con);
  35
  36         /* protocol version mismatch */
  37         void (*bad_proto) (struct ceph_connection *con);
  38
  39         /* there was some error on the socket (disconnect, whatever) */
  40         void (*fault) (struct ceph_connection *con);
  41
  42         /* a remote host as terminated a message exchange session, and messages
  43          * we sent (or they tried to send us) may be lost. */
  44         void (*peer_reset) (struct ceph_connection *con);
  45
  46         struct ceph_msg * (*alloc_msg) (struct ceph_connection *con,
  47                                         struct ceph_msg_header *hdr,
  48                                         int *skip);
  49 };
  50
  51 /* use format string %s%d */
  52 #define ENTITY_NAME(n) ceph_entity_type_name((n).type), le64_to_cpu((n).num)
  53
  54 struct ceph_messenger {
  55         struct ceph_entity_inst inst;    /* my name+address */
  56         struct ceph_entity_addr my_enc_addr;
  57         struct page *zero_page;          /* used in certain error cases */
  58
  59         bool nocrc;
  60
  61         /*
  62          * the global_seq counts connections i (attempt to) initiate
  63          * in order to disambiguate certain connect race conditions.
  64          */
  65         u32 global_seq;
  66         spinlock_t global_seq_lock;
  67
  68         u32 supported_features;
  69         u32 required_features;
  70 };
  71
  72 /*
  73  * a single message.  it contains a header (src, dest, message type, etc.),
  74  * footer (crc values, mainly), a "front" message body, and possibly a
  75  * data payload (stored in some number of pages).
  76  */
  77 struct ceph_msg {
  78         struct ceph_msg_header hdr;     /* header */
  79         struct ceph_msg_footer footer;  /* footer */
  80         struct kvec front;              /* unaligned blobs of message */
  81         struct ceph_buffer *middle;
  82         struct page **pages;            /* data payload.  NOT OWNER. */
  83         unsigned nr_pages;              /* size of page array */
  84         unsigned page_alignment;        /* io offset in first page */
  85         struct ceph_pagelist *pagelist; /* instead of pages */
  86         struct list_head list_head;
  87         struct kref kref;
  88         struct bio  *bio;               /* instead of pages/pagelist */
  89         struct bio  *bio_iter;          /* bio iterator */
  90         int bio_seg;                    /* current bio segment */
  91         struct ceph_pagelist *trail;    /* the trailing part of the data */
  92         bool front_is_vmalloc;
  93         bool more_to_follow;
  94         bool needs_out_seq;
  95         int front_max;
  96         unsigned long ack_stamp;        /* tx: when we were acked */
  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 #define LOSSYTX         0  /* we can close channel or drop messages on errors */
 115 #define CONNECTING      1
 116 #define NEGOTIATING     2
 117 #define KEEPALIVE_PENDING      3
 118 #define WRITE_PENDING   4  /* we have data ready to send */
 119 #define STANDBY         8  /* no outgoing messages, socket closed.  we keep
 120                             * the ceph_connection around to maintain shared
 121                             * state with the peer. */
 122 #define CLOSED          10 /* we've closed the connection */
 123 #define SOCK_CLOSED     11 /* socket state changed to closed */
 124 #define OPENING         13 /* open connection w/ (possibly new) peer */
 125 #define DEAD            14 /* dead, about to kfree */
 126 #define BACKOFF         15
 127
 128 /*
 129  * A single connection with another host.
 130  *
 131  * We maintain a queue of outgoing messages, and some session state to
 132  * ensure that we can preserve the lossless, ordered delivery of
 133  * messages in the case of a TCP disconnect.
 134  */
 135 struct ceph_connection {
 136         void *private;
 137         atomic_t nref;
 138
 139         const struct ceph_connection_operations *ops;
 140
 141         struct ceph_messenger *msgr;
 142         struct socket *sock;
 143         unsigned long state;    /* connection state (see flags above) */
 144         const char *error_msg;  /* error message, if any */
 145
 146         struct ceph_entity_addr peer_addr; /* peer address */
 147         struct ceph_entity_name peer_name; /* peer name */
 148         struct ceph_entity_addr peer_addr_for_me;
 149         unsigned peer_features;
 150         u32 connect_seq;      /* identify the most recent connection
 151                                  attempt for this connection, client */
 152         u32 peer_global_seq;  /* peer's global seq for this connection */
 153
 154         int auth_retry;       /* true if we need a newer authorizer */
 155         void *auth_reply_buf;   /* where to put the authorizer reply */
 156         int auth_reply_buf_len;
 157
 158         struct mutex mutex;
 159
 160         /* out queue */
 161         struct list_head out_queue;
 162         struct list_head out_sent;   /* sending or sent but unacked */
 163         u64 out_seq;                 /* last message queued for send */
 164
 165         u64 in_seq, in_seq_acked;  /* last message received, acked */
 166
 167         /* connection negotiation temps */
 168         char in_banner[CEPH_BANNER_MAX_LEN];
 169         union {
 170                 struct {  /* outgoing connection */
 171                         struct ceph_msg_connect out_connect;
 172                         struct ceph_msg_connect_reply in_reply;
 173                 };
 174                 struct {  /* incoming */
 175                         struct ceph_msg_connect in_connect;
 176                         struct ceph_msg_connect_reply out_reply;
 177                 };
 178         };
 179         struct ceph_entity_addr actual_peer_addr;
 180
 181         /* message out temps */
 182         struct ceph_msg *out_msg;        /* sending message (== tail of
 183                                             out_sent) */
 184         bool out_msg_done;
 185         struct ceph_msg_pos out_msg_pos;
 186
 187         struct kvec out_kvec[8],         /* sending header/footer data */
 188                 *out_kvec_cur;
 189         int out_kvec_left;   /* kvec's left in out_kvec */
 190         int out_skip;        /* skip this many bytes */
 191         int out_kvec_bytes;  /* total bytes left */
 192         bool out_kvec_is_msg; /* kvec refers to out_msg */
 193         int out_more;        /* there is more data after the kvecs */
 194         __le64 out_temp_ack; /* for writing an ack */
 195
 196         /* message in temps */
 197         struct ceph_msg_header in_hdr;
 198         struct ceph_msg *in_msg;
 199         struct ceph_msg_pos in_msg_pos;
 200         u32 in_front_crc, in_middle_crc, in_data_crc;  /* calculated crc */
 201
 202         char in_tag;         /* protocol control byte */
 203         int in_base_pos;     /* bytes read */
 204         __le64 in_temp_ack;  /* for reading an ack */
 205
 206         struct delayed_work work;           /* send|recv work */
 207         unsigned long       delay;          /* current delay interval */
 208 };
 209
 210
 211 extern const char *ceph_pr_addr(const struct sockaddr_storage *ss);
 212 extern int ceph_parse_ips(const char *c, const char *end,
 213                           struct ceph_entity_addr *addr,
 214                           int max_count, int *count);
 215
 216
 217 extern int ceph_msgr_init(void);
 218 extern void ceph_msgr_exit(void);
 219 extern void ceph_msgr_flush(void);
 220
 221 extern struct ceph_messenger *ceph_messenger_create(
 222         struct ceph_entity_addr *myaddr,
 223         u32 features, u32 required);
 224 extern void ceph_messenger_destroy(struct ceph_messenger *);
 225
 226 extern void ceph_con_init(struct ceph_messenger *msgr,
 227                           struct ceph_connection *con);
 228 extern void ceph_con_open(struct ceph_connection *con,
 229                           struct ceph_entity_addr *addr);
 230 extern bool ceph_con_opened(struct ceph_connection *con);
 231 extern void ceph_con_close(struct ceph_connection *con);
 232 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
 233 extern void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg);
 234 extern void ceph_con_revoke_message(struct ceph_connection *con,
 235                                   struct ceph_msg *msg);
 236 extern void ceph_con_keepalive(struct ceph_connection *con);
 237 extern struct ceph_connection *ceph_con_get(struct ceph_connection *con);
 238 extern void ceph_con_put(struct ceph_connection *con);
 239
 240 extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
 241                                      bool can_fail);
 242 extern void ceph_msg_kfree(struct ceph_msg *m);
 243
 244
 245 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
 246 {
 247         kref_get(&msg->kref);
 248         return msg;
 249 }
 250 extern void ceph_msg_last_put(struct kref *kref);
 251 static inline void ceph_msg_put(struct ceph_msg *msg)
 252 {
 253         kref_put(&msg->kref, ceph_msg_last_put);
 254 }
 255
 256 extern void ceph_msg_dump(struct ceph_msg *msg);
 257
 258 #endif