include/linux/ceph/messenger.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef __FS_CEPH_MESSENGER_H
   3 #define __FS_CEPH_MESSENGER_H
   4
   5 #include <linux/bvec.h>
   6 #include <linux/kref.h>
   7 #include <linux/mutex.h>
   8 #include <linux/net.h>
   9 #include <linux/radix-tree.h>
  10 #include <linux/uio.h>
  11 #include <linux/workqueue.h>
  12 #include <net/net_namespace.h>
  13
  14 #include <linux/ceph/types.h>
  15 #include <linux/ceph/buffer.h>
  16
  17 struct ceph_msg;
  18 struct ceph_connection;
  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         struct ceph_auth_handshake *(*get_authorizer) (
  32                                 struct ceph_connection *con,
  33                                int *proto, int force_new);
  34         int (*add_authorizer_challenge)(struct ceph_connection *con,
  35                                         void *challenge_buf,
  36                                         int challenge_buf_len);
  37         int (*verify_authorizer_reply) (struct ceph_connection *con);
  38         int (*invalidate_authorizer)(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         void (*reencode_message) (struct ceph_msg *msg);
  52
  53         int (*sign_message) (struct ceph_msg *msg);
  54         int (*check_message_signature) (struct ceph_msg *msg);
  55 };
  56
  57 /* use format string %s%d */
  58 #define ENTITY_NAME(n) ceph_entity_type_name((n).type), le64_to_cpu((n).num)
  59
  60 struct ceph_messenger {
  61         struct ceph_entity_inst inst;    /* my name+address */
  62         struct ceph_entity_addr my_enc_addr;
  63
  64         atomic_t stopping;
  65         possible_net_t net;
  66
  67         /*
  68          * the global_seq counts connections i (attempt to) initiate
  69          * in order to disambiguate certain connect race conditions.
  70          */
  71         u32 global_seq;
  72         spinlock_t global_seq_lock;
  73 };
  74
  75 enum ceph_msg_data_type {
  76         CEPH_MSG_DATA_NONE,     /* message contains no data payload */
  77         CEPH_MSG_DATA_PAGES,    /* data source/destination is a page array */
  78         CEPH_MSG_DATA_PAGELIST, /* data source/destination is a pagelist */
  79 #ifdef CONFIG_BLOCK
  80         CEPH_MSG_DATA_BIO,      /* data source/destination is a bio list */
  81 #endif /* CONFIG_BLOCK */
  82 };
  83
  84 static __inline__ bool ceph_msg_data_type_valid(enum ceph_msg_data_type type)
  85 {
  86         switch (type) {
  87         case CEPH_MSG_DATA_NONE:
  88         case CEPH_MSG_DATA_PAGES:
  89         case CEPH_MSG_DATA_PAGELIST:
  90 #ifdef CONFIG_BLOCK
  91         case CEPH_MSG_DATA_BIO:
  92 #endif /* CONFIG_BLOCK */
  93                 return true;
  94         default:
  95                 return false;
  96         }
  97 }
  98
  99 struct ceph_msg_data {
 100         struct list_head                links;  /* ceph_msg->data */
 101         enum ceph_msg_data_type         type;
 102         union {
 103 #ifdef CONFIG_BLOCK
 104                 struct {
 105                         struct bio      *bio;
 106                         size_t          bio_length;
 107                 };
 108 #endif /* CONFIG_BLOCK */
 109                 struct {
 110                         struct page     **pages;        /* NOT OWNER. */
 111                         size_t          length;         /* total # bytes */
 112                         unsigned int    alignment;      /* first page */
 113                 };
 114                 struct ceph_pagelist    *pagelist;
 115         };
 116 };
 117
 118 struct ceph_msg_data_cursor {
 119         size_t                  total_resid;    /* across all data items */
 120         struct list_head        *data_head;     /* = &ceph_msg->data */
 121
 122         struct ceph_msg_data    *data;          /* current data item */
 123         size_t                  resid;          /* bytes not yet consumed */
 124         bool                    last_piece;     /* current is last piece */
 125         bool                    need_crc;       /* crc update needed */
 126         union {
 127 #ifdef CONFIG_BLOCK
 128                 struct {                                /* bio */
 129                         struct bio      *bio;           /* bio from list */
 130                         struct bvec_iter bvec_iter;
 131                 };
 132 #endif /* CONFIG_BLOCK */
 133                 struct {                                /* pages */
 134                         unsigned int    page_offset;    /* offset in page */
 135                         unsigned short  page_index;     /* index in array */
 136                         unsigned short  page_count;     /* pages in array */
 137                 };
 138                 struct {                                /* pagelist */
 139                         struct page     *page;          /* page from list */
 140                         size_t          offset;         /* bytes from list */
 141                 };
 142         };
 143 };
 144
 145 /*
 146  * a single message.  it contains a header (src, dest, message type, etc.),
 147  * footer (crc values, mainly), a "front" message body, and possibly a
 148  * data payload (stored in some number of pages).
 149  */
 150 struct ceph_msg {
 151         struct ceph_msg_header hdr;     /* header */
 152         union {
 153                 struct ceph_msg_footer footer;          /* footer */
 154                 struct ceph_msg_footer_old old_footer;  /* old format footer */
 155         };
 156         struct kvec front;              /* unaligned blobs of message */
 157         struct ceph_buffer *middle;
 158
 159         size_t                          data_length;
 160         struct list_head                data;
 161         struct ceph_msg_data_cursor     cursor;
 162
 163         struct ceph_connection *con;
 164         struct list_head list_head;     /* links for connection lists */
 165
 166         struct kref kref;
 167         bool more_to_follow;
 168         bool needs_out_seq;
 169         int front_alloc_len;
 170         unsigned long ack_stamp;        /* tx: when we were acked */
 171
 172         struct ceph_msgpool *pool;
 173 };
 174
 175 /* ceph connection fault delay defaults, for exponential backoff */
 176 #define BASE_DELAY_INTERVAL     (HZ/2)
 177 #define MAX_DELAY_INTERVAL      (5 * 60 * HZ)
 178
 179 /*
 180  * A single connection with another host.
 181  *
 182  * We maintain a queue of outgoing messages, and some session state to
 183  * ensure that we can preserve the lossless, ordered delivery of
 184  * messages in the case of a TCP disconnect.
 185  */
 186 struct ceph_connection {
 187         void *private;
 188
 189         const struct ceph_connection_operations *ops;
 190
 191         struct ceph_messenger *msgr;
 192
 193         atomic_t sock_state;
 194         struct socket *sock;
 195         struct ceph_entity_addr peer_addr; /* peer address */
 196         struct ceph_entity_addr peer_addr_for_me;
 197
 198         unsigned long flags;
 199         unsigned long state;
 200         const char *error_msg;  /* error message, if any */
 201
 202         struct ceph_entity_name peer_name; /* peer name */
 203
 204         u64 peer_features;
 205         u32 connect_seq;      /* identify the most recent connection
 206                                  attempt for this connection, client */
 207         u32 peer_global_seq;  /* peer's global seq for this connection */
 208
 209         struct ceph_auth_handshake *auth;
 210         int auth_retry;       /* true if we need a newer authorizer */
 211
 212         struct mutex mutex;
 213
 214         /* out queue */
 215         struct list_head out_queue;
 216         struct list_head out_sent;   /* sending or sent but unacked */
 217         u64 out_seq;                 /* last message queued for send */
 218
 219         u64 in_seq, in_seq_acked;  /* last message received, acked */
 220
 221         /* connection negotiation temps */
 222         char in_banner[CEPH_BANNER_MAX_LEN];
 223         struct ceph_msg_connect out_connect;
 224         struct ceph_msg_connect_reply in_reply;
 225         struct ceph_entity_addr actual_peer_addr;
 226
 227         /* message out temps */
 228         struct ceph_msg_header out_hdr;
 229         struct ceph_msg *out_msg;        /* sending message (== tail of
 230                                             out_sent) */
 231         bool out_msg_done;
 232
 233         struct kvec out_kvec[8],         /* sending header/footer data */
 234                 *out_kvec_cur;
 235         int out_kvec_left;   /* kvec's left in out_kvec */
 236         int out_skip;        /* skip this many bytes */
 237         int out_kvec_bytes;  /* total bytes left */
 238         int out_more;        /* there is more data after the kvecs */
 239         __le64 out_temp_ack; /* for writing an ack */
 240         struct ceph_timespec out_temp_keepalive2; /* for writing keepalive2
 241                                                      stamp */
 242
 243         /* message in temps */
 244         struct ceph_msg_header in_hdr;
 245         struct ceph_msg *in_msg;
 246         u32 in_front_crc, in_middle_crc, in_data_crc;  /* calculated crc */
 247
 248         char in_tag;         /* protocol control byte */
 249         int in_base_pos;     /* bytes read */
 250         __le64 in_temp_ack;  /* for reading an ack */
 251
 252         struct timespec last_keepalive_ack; /* keepalive2 ack stamp */
 253
 254         struct delayed_work work;           /* send|recv work */
 255         unsigned long       delay;          /* current delay interval */
 256 };
 257
 258
 259 extern const char *ceph_pr_addr(const struct sockaddr_storage *ss);
 260 extern int ceph_parse_ips(const char *c, const char *end,
 261                           struct ceph_entity_addr *addr,
 262                           int max_count, int *count);
 263
 264
 265 extern int ceph_msgr_init(void);
 266 extern void ceph_msgr_exit(void);
 267 extern void ceph_msgr_flush(void);
 268
 269 extern void ceph_messenger_init(struct ceph_messenger *msgr,
 270                                 struct ceph_entity_addr *myaddr);
 271 extern void ceph_messenger_fini(struct ceph_messenger *msgr);
 272
 273 extern void ceph_con_init(struct ceph_connection *con, void *private,
 274                         const struct ceph_connection_operations *ops,
 275                         struct ceph_messenger *msgr);
 276 extern void ceph_con_open(struct ceph_connection *con,
 277                           __u8 entity_type, __u64 entity_num,
 278                           struct ceph_entity_addr *addr);
 279 extern bool ceph_con_opened(struct ceph_connection *con);
 280 extern void ceph_con_close(struct ceph_connection *con);
 281 extern void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg);
 282
 283 extern void ceph_msg_revoke(struct ceph_msg *msg);
 284 extern void ceph_msg_revoke_incoming(struct ceph_msg *msg);
 285
 286 extern void ceph_con_keepalive(struct ceph_connection *con);
 287 extern bool ceph_con_keepalive_expired(struct ceph_connection *con,
 288                                        unsigned long interval);
 289
 290 extern void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
 291                                 size_t length, size_t alignment);
 292 extern void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
 293                                 struct ceph_pagelist *pagelist);
 294 #ifdef CONFIG_BLOCK
 295 extern void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
 296                                 size_t length);
 297 #endif /* CONFIG_BLOCK */
 298
 299 extern struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
 300                                      bool can_fail);
 301
 302 extern struct ceph_msg *ceph_msg_get(struct ceph_msg *msg);
 303 extern void ceph_msg_put(struct ceph_msg *msg);
 304
 305 extern void ceph_msg_dump(struct ceph_msg *msg);
 306
 307 #endif