fs/ceph/messenger.h

   1 #ifndef __FS_CEPH_MESSENGER_H
   2 #define __FS_CEPH_MESSENGER_H
   3
   4 #include <linux/mutex.h>
   5 #include <linux/net.h>
   6 #include <linux/radix-tree.h>
   7 #include <linux/uio.h>
   8 #include <linux/version.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         /* protocol version mismatch */
  30         void (*bad_proto) (struct ceph_connection *con);
  31
  32         /* there was some error on the socket (disconnect, whatever) */
  33         void (*fault) (struct ceph_connection *con);
  34
  35         /* a remote host as terminated a message exchange session, and messages
  36          * we sent (or they tried to send us) may be lost. */
  37         void (*peer_reset) (struct ceph_connection *con);
  38
  39         struct ceph_msg * (*alloc_msg) (struct ceph_connection *con,
  40                                         struct ceph_msg_header *hdr);
  41         int (*alloc_middle) (struct ceph_connection *con,
  42                              struct ceph_msg *msg);
  43         /* an incoming message has a data payload; tell me what pages I
  44          * should read the data into. */
  45         int (*prepare_pages) (struct ceph_connection *con, struct ceph_msg *m,
  46                               int want);
  47 };
  48
  49 extern const char *ceph_name_type_str(int t);
  50
  51 /* use format string %s%d */
  52 #define ENTITY_NAME(n) ceph_name_type_str((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
  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  */
  74 struct 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         struct list_head list_head;
  82         atomic_t nref;
  83         bool front_is_vmalloc;
  84         bool more_to_follow;
  85         int front_max;
  86
  87         struct ceph_msgpool *pool;
  88 };
  89
  90 struct ceph_msg_pos {
  91         int page, page_pos;  /* which page; offset in page */
  92         int data_pos;        /* offset in data payload */
  93         int did_page_crc;    /* true if we've calculated crc for current page */
  94 };
  95
  96 /* ceph connection fault delay defaults, for exponential backoff */
  97 #define BASE_DELAY_INTERVAL     (HZ/2)
  98 #define MAX_DELAY_INTERVAL      (5 * 60 * HZ)
  99
 100 /*
 101  * ceph_connection state bit flags
 102  *
 103  * QUEUED and BUSY are used together to ensure that only a single
 104  * thread is currently opening, reading or writing data to the socket.
 105  */
 106 #define LOSSYTX         0  /* we can close channel or drop messages on errors */
 107 #define LOSSYRX         1  /* peer may reset/drop messages */
 108 #define CONNECTING      2
 109 #define KEEPALIVE_PENDING      3
 110 #define WRITE_PENDING   4  /* we have data ready to send */
 111 #define QUEUED          5  /* there is work queued on this connection */
 112 #define BUSY            6  /* work is being done */
 113 #define STANDBY         8  /* no outgoing messages, socket closed.  we keep
 114                             * the ceph_connection around to maintain shared
 115                             * state with the peer. */
 116 #define CLOSED          10 /* we've closed the connection */
 117 #define SOCK_CLOSED     11 /* socket state changed to closed */
 118 #define REGISTERED      12 /* connection appears in con_tree */
 119 #define OPENING         13 /* open connection w/ (possibly new) peer */
 120 #define DEAD            14 /* dead, about to kfree */
 121
 122 /*
 123  * A single connection with another host.
 124  *
 125  * We maintain a queue of outgoing messages, and some session state to
 126  * ensure that we can preserve the lossless, ordered delivery of
 127  * messages in the case of a TCP disconnect.
 128  */
 129 struct ceph_connection {
 130         void *private;
 131         atomic_t nref;
 132
 133         const struct ceph_connection_operations *ops;
 134
 135         struct ceph_messenger *msgr;
 136         struct socket *sock;
 137         unsigned long state;    /* connection state (see flags above) */
 138         const char *error_msg;  /* error message, if any */
 139
 140         struct ceph_entity_addr peer_addr; /* peer address */
 141         struct ceph_entity_name peer_name; /* peer name */
 142         struct ceph_entity_addr peer_addr_for_me;
 143         u32 connect_seq;      /* identify the most recent connection
 144                                  attempt for this connection, client */
 145         u32 peer_global_seq;  /* peer's global seq for this connection */
 146
 147         /* out queue */
 148         struct mutex out_mutex;
 149         struct list_head out_queue;
 150         struct list_head out_sent;   /* sending or sent but unacked */
 151         u64 out_seq;                 /* last message queued for send */
 152         u64 out_seq_sent;            /* last message sent */
 153         bool out_keepalive_pending;
 154
 155         u64 in_seq, in_seq_acked;  /* last message received, acked */
 156
 157         /* connection negotiation temps */
 158         char in_banner[CEPH_BANNER_MAX_LEN];
 159         union {
 160                 struct {  /* outgoing connection */
 161                         struct ceph_msg_connect out_connect;
 162                         struct ceph_msg_connect_reply in_reply;
 163                 };
 164                 struct {  /* incoming */
 165                         struct ceph_msg_connect in_connect;
 166                         struct ceph_msg_connect_reply out_reply;
 167                 };
 168         };
 169         struct ceph_entity_addr actual_peer_addr;
 170
 171         /* message out temps */
 172         struct ceph_msg *out_msg;        /* sending message (== tail of
 173                                             out_sent) */
 174         struct ceph_msg_pos out_msg_pos;
 175
 176         struct kvec out_kvec[8],         /* sending header/footer data */
 177                 *out_kvec_cur;
 178         int out_kvec_left;   /* kvec's left in out_kvec */
 179         int out_skip;        /* skip this many bytes */
 180         int out_kvec_bytes;  /* total bytes left */
 181         bool out_kvec_is_msg; /* kvec refers to out_msg */
 182         int out_more;        /* there is more data after the kvecs */
 183         __le64 out_temp_ack; /* for writing an ack */
 184
 185         /* message in temps */
 186         struct ceph_msg_header in_hdr;
 187         struct ceph_msg *in_msg;
 188         struct ceph_msg_pos in_msg_pos;
 189         u32 in_front_crc, in_middle_crc, in_data_crc;  /* calculated crc */
 190
 191         char in_tag;         /* protocol control byte */
 192         int in_base_pos;     /* bytes read */
 193         __le64 in_temp_ack;  /* for reading an ack */
 194
 195         struct delayed_work work;           /* send|recv work */
 196         unsigned long       delay;          /* current delay interval */
 197 };
 198
 199
 200 extern const char *pr_addr(const struct sockaddr_storage *ss);
 201 extern int ceph_parse_ips(const char *c, const char *end,
 202                           struct ceph_entity_addr *addr,
 203                           int max_count, int *count);
 204
 205
 206 extern int ceph_msgr_init(void);
 207 extern void ceph_msgr_exit(void);
 208
 209 extern struct ceph_messenger *ceph_messenger_create(
 210         struct ceph_entity_addr *myaddr);
 211 extern void ceph_messenger_destroy(struct ceph_messenger *);
 212
 213 extern void ceph_con_init(struct ceph_messenger *msgr,
 214                           struct ceph_connection *con);
 215 extern void ceph_con_shutdown(struct ceph_connection *con);
 216 extern void ceph_con_open(struct ceph_connection *con,
 217                           struct ceph_entity_addr *addr);
 218 extern void ceph_con_close(struct ceph_connection *con);
 219 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
 220 extern void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg);
 221 extern void ceph_con_keepalive(struct ceph_connection *con);
 222 extern struct ceph_connection *ceph_con_get(struct ceph_connection *con);
 223 extern void ceph_con_put(struct ceph_connection *con);
 224
 225 extern struct ceph_msg *ceph_msg_new(int type, int front_len,
 226                                      int page_len, int page_off,
 227                                      struct page **pages);
 228 extern void ceph_msg_kfree(struct ceph_msg *m);
 229
 230 extern struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
 231                                        struct ceph_msg_header *hdr);
 232 extern int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg);
 233
 234
 235 static inline struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
 236 {
 237         dout("ceph_msg_get %p %d -> %d\n", msg, atomic_read(&msg->nref),
 238              atomic_read(&msg->nref)+1);
 239         atomic_inc(&msg->nref);
 240         return msg;
 241 }
 242 extern void ceph_msg_put(struct ceph_msg *msg);
 243
 244 #endif