net/ceph/messenger.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/crc32c.h>
   5 #include <linux/ctype.h>
   6 #include <linux/highmem.h>
   7 #include <linux/inet.h>
   8 #include <linux/kthread.h>
   9 #include <linux/net.h>
  10 #include <linux/nsproxy.h>
  11 #include <linux/sched/mm.h>
  12 #include <linux/slab.h>
  13 #include <linux/socket.h>
  14 #include <linux/string.h>
  15 #ifdef  CONFIG_BLOCK
  16 #include <linux/bio.h>
  17 #endif  /* CONFIG_BLOCK */
  18 #include <linux/dns_resolver.h>
  19 #include <net/tcp.h>
  20
  21 #include <linux/ceph/ceph_features.h>
  22 #include <linux/ceph/libceph.h>
  23 #include <linux/ceph/messenger.h>
  24 #include <linux/ceph/decode.h>
  25 #include <linux/ceph/pagelist.h>
  26 #include <linux/export.h>
  27
  28 /*
  29  * Ceph uses the messenger to exchange ceph_msg messages with other
  30  * hosts in the system.  The messenger provides ordered and reliable
  31  * delivery.  We tolerate TCP disconnects by reconnecting (with
  32  * exponential backoff) in the case of a fault (disconnection, bad
  33  * crc, protocol error).  Acks allow sent messages to be discarded by
  34  * the sender.
  35  */
  36
  37 /*
  38  * We track the state of the socket on a given connection using
  39  * values defined below.  The transition to a new socket state is
  40  * handled by a function which verifies we aren't coming from an
  41  * unexpected state.
  42  *
  43  *      --------
  44  *      | NEW* |  transient initial state
  45  *      --------
  46  *          | con_sock_state_init()
  47  *          v
  48  *      ----------
  49  *      | CLOSED |  initialized, but no socket (and no
  50  *      ----------  TCP connection)
  51  *       ^      \
  52  *       |       \ con_sock_state_connecting()
  53  *       |        ----------------------
  54  *       |                              \
  55  *       + con_sock_state_closed()       \
  56  *       |+---------------------------    \
  57  *       | \                          \    \
  58  *       |  -----------                \    \
  59  *       |  | CLOSING |  socket event;  \    \
  60  *       |  -----------  await close     \    \
  61  *       |       ^                        \   |
  62  *       |       |                         \  |
  63  *       |       + con_sock_state_closing() \ |
  64  *       |      / \                         | |
  65  *       |     /   ---------------          | |
  66  *       |    /                   \         v v
  67  *       |   /                    --------------
  68  *       |  /    -----------------| CONNECTING |  socket created, TCP
  69  *       |  |   /                 --------------  connect initiated
  70  *       |  |   | con_sock_state_connected()
  71  *       |  |   v
  72  *      -------------
  73  *      | CONNECTED |  TCP connection established
  74  *      -------------
  75  *
  76  * State values for ceph_connection->sock_state; NEW is assumed to be 0.
  77  */
  78
  79 #define CON_SOCK_STATE_NEW              0       /* -> CLOSED */
  80 #define CON_SOCK_STATE_CLOSED           1       /* -> CONNECTING */
  81 #define CON_SOCK_STATE_CONNECTING       2       /* -> CONNECTED or -> CLOSING */
  82 #define CON_SOCK_STATE_CONNECTED        3       /* -> CLOSING or -> CLOSED */
  83 #define CON_SOCK_STATE_CLOSING          4       /* -> CLOSED */
  84
  85 /*
  86  * connection states
  87  */
  88 #define CON_STATE_CLOSED        1  /* -> PREOPEN */
  89 #define CON_STATE_PREOPEN       2  /* -> CONNECTING, CLOSED */
  90 #define CON_STATE_CONNECTING    3  /* -> NEGOTIATING, CLOSED */
  91 #define CON_STATE_NEGOTIATING   4  /* -> OPEN, CLOSED */
  92 #define CON_STATE_OPEN          5  /* -> STANDBY, CLOSED */
  93 #define CON_STATE_STANDBY       6  /* -> PREOPEN, CLOSED */
  94
  95 /*
  96  * ceph_connection flag bits
  97  */
  98 #define CON_FLAG_LOSSYTX           0  /* we can close channel or drop
  99                                        * messages on errors */
 100 #define CON_FLAG_KEEPALIVE_PENDING 1  /* we need to send a keepalive */
 101 #define CON_FLAG_WRITE_PENDING     2  /* we have data ready to send */
 102 #define CON_FLAG_SOCK_CLOSED       3  /* socket state changed to closed */
 103 #define CON_FLAG_BACKOFF           4  /* need to retry queuing delayed work */
 104
 105 static bool con_flag_valid(unsigned long con_flag)
 106 {
 107         switch (con_flag) {
 108         case CON_FLAG_LOSSYTX:
 109         case CON_FLAG_KEEPALIVE_PENDING:
 110         case CON_FLAG_WRITE_PENDING:
 111         case CON_FLAG_SOCK_CLOSED:
 112         case CON_FLAG_BACKOFF:
 113                 return true;
 114         default:
 115                 return false;
 116         }
 117 }
 118
 119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
 120 {
 121         BUG_ON(!con_flag_valid(con_flag));
 122
 123         clear_bit(con_flag, &con->flags);
 124 }
 125
 126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
 127 {
 128         BUG_ON(!con_flag_valid(con_flag));
 129
 130         set_bit(con_flag, &con->flags);
 131 }
 132
 133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
 134 {
 135         BUG_ON(!con_flag_valid(con_flag));
 136
 137         return test_bit(con_flag, &con->flags);
 138 }
 139
 140 static bool con_flag_test_and_clear(struct ceph_connection *con,
 141                                         unsigned long con_flag)
 142 {
 143         BUG_ON(!con_flag_valid(con_flag));
 144
 145         return test_and_clear_bit(con_flag, &con->flags);
 146 }
 147
 148 static bool con_flag_test_and_set(struct ceph_connection *con,
 149                                         unsigned long con_flag)
 150 {
 151         BUG_ON(!con_flag_valid(con_flag));
 152
 153         return test_and_set_bit(con_flag, &con->flags);
 154 }
 155
 156 /* Slab caches for frequently-allocated structures */
 157
 158 static struct kmem_cache        *ceph_msg_cache;
 159
 160 /* static tag bytes (protocol control messages) */
 161 static char tag_msg = CEPH_MSGR_TAG_MSG;
 162 static char tag_ack = CEPH_MSGR_TAG_ACK;
 163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
 164 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
 165
 166 #ifdef CONFIG_LOCKDEP
 167 static struct lock_class_key socket_class;
 168 #endif
 169
 170 static void queue_con(struct ceph_connection *con);
 171 static void cancel_con(struct ceph_connection *con);
 172 static void ceph_con_workfn(struct work_struct *);
 173 static void con_fault(struct ceph_connection *con);
 174
 175 /*
 176  * Nicely render a sockaddr as a string.  An array of formatted
 177  * strings is used, to approximate reentrancy.
 178  */
 179 #define ADDR_STR_COUNT_LOG      5       /* log2(# address strings in array) */
 180 #define ADDR_STR_COUNT          (1 << ADDR_STR_COUNT_LOG)
 181 #define ADDR_STR_COUNT_MASK     (ADDR_STR_COUNT - 1)
 182 #define MAX_ADDR_STR_LEN        64      /* 54 is enough */
 183
 184 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
 185 static atomic_t addr_str_seq = ATOMIC_INIT(0);
 186
 187 static struct page *zero_page;          /* used in certain error cases */
 188
 189 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
 190 {
 191         int i;
 192         char *s;
 193         struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
 194         struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
 195
 196         i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
 197         s = addr_str[i];
 198
 199         switch (ss->ss_family) {
 200         case AF_INET:
 201                 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
 202                          ntohs(in4->sin_port));
 203                 break;
 204
 205         case AF_INET6:
 206                 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
 207                          ntohs(in6->sin6_port));
 208                 break;
 209
 210         default:
 211                 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
 212                          ss->ss_family);
 213         }
 214
 215         return s;
 216 }
 217 EXPORT_SYMBOL(ceph_pr_addr);
 218
 219 static void encode_my_addr(struct ceph_messenger *msgr)
 220 {
 221         memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
 222         ceph_encode_addr(&msgr->my_enc_addr);
 223 }
 224
 225 /*
 226  * work queue for all reading and writing to/from the socket.
 227  */
 228 static struct workqueue_struct *ceph_msgr_wq;
 229
 230 static int ceph_msgr_slab_init(void)
 231 {
 232         BUG_ON(ceph_msg_cache);
 233         ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
 234         if (!ceph_msg_cache)
 235                 return -ENOMEM;
 236
 237         return 0;
 238 }
 239
 240 static void ceph_msgr_slab_exit(void)
 241 {
 242         BUG_ON(!ceph_msg_cache);
 243         kmem_cache_destroy(ceph_msg_cache);
 244         ceph_msg_cache = NULL;
 245 }
 246
 247 static void _ceph_msgr_exit(void)
 248 {
 249         if (ceph_msgr_wq) {
 250                 destroy_workqueue(ceph_msgr_wq);
 251                 ceph_msgr_wq = NULL;
 252         }
 253
 254         BUG_ON(zero_page == NULL);
 255         put_page(zero_page);
 256         zero_page = NULL;
 257
 258         ceph_msgr_slab_exit();
 259 }
 260
 261 int __init ceph_msgr_init(void)
 262 {
 263         if (ceph_msgr_slab_init())
 264                 return -ENOMEM;
 265
 266         BUG_ON(zero_page != NULL);
 267         zero_page = ZERO_PAGE(0);
 268         get_page(zero_page);
 269
 270         /*
 271          * The number of active work items is limited by the number of
 272          * connections, so leave @max_active at default.
 273          */
 274         ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
 275         if (ceph_msgr_wq)
 276                 return 0;
 277
 278         pr_err("msgr_init failed to create workqueue\n");
 279         _ceph_msgr_exit();
 280
 281         return -ENOMEM;
 282 }
 283
 284 void ceph_msgr_exit(void)
 285 {
 286         BUG_ON(ceph_msgr_wq == NULL);
 287
 288         _ceph_msgr_exit();
 289 }
 290
 291 void ceph_msgr_flush(void)
 292 {
 293         flush_workqueue(ceph_msgr_wq);
 294 }
 295 EXPORT_SYMBOL(ceph_msgr_flush);
 296
 297 /* Connection socket state transition functions */
 298
 299 static void con_sock_state_init(struct ceph_connection *con)
 300 {
 301         int old_state;
 302
 303         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
 304         if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
 305                 printk("%s: unexpected old state %d\n", __func__, old_state);
 306         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 307              CON_SOCK_STATE_CLOSED);
 308 }
 309
 310 static void con_sock_state_connecting(struct ceph_connection *con)
 311 {
 312         int old_state;
 313
 314         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
 315         if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
 316                 printk("%s: unexpected old state %d\n", __func__, old_state);
 317         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 318              CON_SOCK_STATE_CONNECTING);
 319 }
 320
 321 static void con_sock_state_connected(struct ceph_connection *con)
 322 {
 323         int old_state;
 324
 325         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
 326         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
 327                 printk("%s: unexpected old state %d\n", __func__, old_state);
 328         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 329              CON_SOCK_STATE_CONNECTED);
 330 }
 331
 332 static void con_sock_state_closing(struct ceph_connection *con)
 333 {
 334         int old_state;
 335
 336         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
 337         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
 338                         old_state != CON_SOCK_STATE_CONNECTED &&
 339                         old_state != CON_SOCK_STATE_CLOSING))
 340                 printk("%s: unexpected old state %d\n", __func__, old_state);
 341         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 342              CON_SOCK_STATE_CLOSING);
 343 }
 344
 345 static void con_sock_state_closed(struct ceph_connection *con)
 346 {
 347         int old_state;
 348
 349         old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
 350         if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
 351                     old_state != CON_SOCK_STATE_CLOSING &&
 352                     old_state != CON_SOCK_STATE_CONNECTING &&
 353                     old_state != CON_SOCK_STATE_CLOSED))
 354                 printk("%s: unexpected old state %d\n", __func__, old_state);
 355         dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 356              CON_SOCK_STATE_CLOSED);
 357 }
 358
 359 /*
 360  * socket callback functions
 361  */
 362
 363 /* data available on socket, or listen socket received a connect */
 364 static void ceph_sock_data_ready(struct sock *sk)
 365 {
 366         struct ceph_connection *con = sk->sk_user_data;
 367         if (atomic_read(&con->msgr->stopping)) {
 368                 return;
 369         }
 370
 371         if (sk->sk_state != TCP_CLOSE_WAIT) {
 372                 dout("%s on %p state = %lu, queueing work\n", __func__,
 373                      con, con->state);
 374                 queue_con(con);
 375         }
 376 }
 377
 378 /* socket has buffer space for writing */
 379 static void ceph_sock_write_space(struct sock *sk)
 380 {
 381         struct ceph_connection *con = sk->sk_user_data;
 382
 383         /* only queue to workqueue if there is data we want to write,
 384          * and there is sufficient space in the socket buffer to accept
 385          * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
 386          * doesn't get called again until try_write() fills the socket
 387          * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
 388          * and net/core/stream.c:sk_stream_write_space().
 389          */
 390         if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
 391                 if (sk_stream_is_writeable(sk)) {
 392                         dout("%s %p queueing write work\n", __func__, con);
 393                         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 394                         queue_con(con);
 395                 }
 396         } else {
 397                 dout("%s %p nothing to write\n", __func__, con);
 398         }
 399 }
 400
 401 /* socket's state has changed */
 402 static void ceph_sock_state_change(struct sock *sk)
 403 {
 404         struct ceph_connection *con = sk->sk_user_data;
 405
 406         dout("%s %p state = %lu sk_state = %u\n", __func__,
 407              con, con->state, sk->sk_state);
 408
 409         switch (sk->sk_state) {
 410         case TCP_CLOSE:
 411                 dout("%s TCP_CLOSE\n", __func__);
 412                 /* fall through */
 413         case TCP_CLOSE_WAIT:
 414                 dout("%s TCP_CLOSE_WAIT\n", __func__);
 415                 con_sock_state_closing(con);
 416                 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
 417                 queue_con(con);
 418                 break;
 419         case TCP_ESTABLISHED:
 420                 dout("%s TCP_ESTABLISHED\n", __func__);
 421                 con_sock_state_connected(con);
 422                 queue_con(con);
 423                 break;
 424         default:        /* Everything else is uninteresting */
 425                 break;
 426         }
 427 }
 428
 429 /*
 430  * set up socket callbacks
 431  */
 432 static void set_sock_callbacks(struct socket *sock,
 433                                struct ceph_connection *con)
 434 {
 435         struct sock *sk = sock->sk;
 436         sk->sk_user_data = con;
 437         sk->sk_data_ready = ceph_sock_data_ready;
 438         sk->sk_write_space = ceph_sock_write_space;
 439         sk->sk_state_change = ceph_sock_state_change;
 440 }
 441
 442
 443 /*
 444  * socket helpers
 445  */
 446
 447 /*
 448  * initiate connection to a remote socket.
 449  */
 450 static int ceph_tcp_connect(struct ceph_connection *con)
 451 {
 452         struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
 453         struct socket *sock;
 454         unsigned int noio_flag;
 455         int ret;
 456
 457         BUG_ON(con->sock);
 458
 459         /* sock_create_kern() allocates with GFP_KERNEL */
 460         noio_flag = memalloc_noio_save();
 461         ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
 462                                SOCK_STREAM, IPPROTO_TCP, &sock);
 463         memalloc_noio_restore(noio_flag);
 464         if (ret)
 465                 return ret;
 466         sock->sk->sk_allocation = GFP_NOFS;
 467
 468 #ifdef CONFIG_LOCKDEP
 469         lockdep_set_class(&sock->sk->sk_lock, &socket_class);
 470 #endif
 471
 472         set_sock_callbacks(sock, con);
 473
 474         dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
 475
 476         con_sock_state_connecting(con);
 477         ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss),
 478                                  O_NONBLOCK);
 479         if (ret == -EINPROGRESS) {
 480                 dout("connect %s EINPROGRESS sk_state = %u\n",
 481                      ceph_pr_addr(&con->peer_addr.in_addr),
 482                      sock->sk->sk_state);
 483         } else if (ret < 0) {
 484                 pr_err("connect %s error %d\n",
 485                        ceph_pr_addr(&con->peer_addr.in_addr), ret);
 486                 sock_release(sock);
 487                 return ret;
 488         }
 489
 490         if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
 491                 int optval = 1;
 492
 493                 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
 494                                         (char *)&optval, sizeof(optval));
 495                 if (ret)
 496                         pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
 497                                ret);
 498         }
 499
 500         con->sock = sock;
 501         return 0;
 502 }
 503
 504 /*
 505  * If @buf is NULL, discard up to @len bytes.
 506  */
 507 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
 508 {
 509         struct kvec iov = {buf, len};
 510         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
 511         int r;
 512
 513         if (!buf)
 514                 msg.msg_flags |= MSG_TRUNC;
 515
 516         iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
 517         r = sock_recvmsg(sock, &msg, msg.msg_flags);
 518         if (r == -EAGAIN)
 519                 r = 0;
 520         return r;
 521 }
 522
 523 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
 524                      int page_offset, size_t length)
 525 {
 526         struct bio_vec bvec = {
 527                 .bv_page = page,
 528                 .bv_offset = page_offset,
 529                 .bv_len = length
 530         };
 531         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
 532         int r;
 533
 534         BUG_ON(page_offset + length > PAGE_SIZE);
 535         iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
 536         r = sock_recvmsg(sock, &msg, msg.msg_flags);
 537         if (r == -EAGAIN)
 538                 r = 0;
 539         return r;
 540 }
 541
 542 /*
 543  * write something.  @more is true if caller will be sending more data
 544  * shortly.
 545  */
 546 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
 547                             size_t kvlen, size_t len, bool more)
 548 {
 549         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
 550         int r;
 551
 552         if (more)
 553                 msg.msg_flags |= MSG_MORE;
 554         else
 555                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
 556
 557         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
 558         if (r == -EAGAIN)
 559                 r = 0;
 560         return r;
 561 }
 562
 563 /*
 564  * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
 565  */
 566 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
 567                              int offset, size_t size, int more)
 568 {
 569         ssize_t (*sendpage)(struct socket *sock, struct page *page,
 570                             int offset, size_t size, int flags);
 571         int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
 572         int ret;
 573
 574         /*
 575          * sendpage cannot properly handle pages with page_count == 0,
 576          * we need to fall back to sendmsg if that's the case.
 577          *
 578          * Same goes for slab pages: skb_can_coalesce() allows
 579          * coalescing neighboring slab objects into a single frag which
 580          * triggers one of hardened usercopy checks.
 581          */
 582         if (page_count(page) >= 1 && !PageSlab(page))
 583                 sendpage = sock->ops->sendpage;
 584         else
 585                 sendpage = sock_no_sendpage;
 586
 587         ret = sendpage(sock, page, offset, size, flags);
 588         if (ret == -EAGAIN)
 589                 ret = 0;
 590
 591         return ret;
 592 }
 593
 594 /*
 595  * Shutdown/close the socket for the given connection.
 596  */
 597 static int con_close_socket(struct ceph_connection *con)
 598 {
 599         int rc = 0;
 600
 601         dout("con_close_socket on %p sock %p\n", con, con->sock);
 602         if (con->sock) {
 603                 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
 604                 sock_release(con->sock);
 605                 con->sock = NULL;
 606         }
 607
 608         /*
 609          * Forcibly clear the SOCK_CLOSED flag.  It gets set
 610          * independent of the connection mutex, and we could have
 611          * received a socket close event before we had the chance to
 612          * shut the socket down.
 613          */
 614         con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
 615
 616         con_sock_state_closed(con);
 617         return rc;
 618 }
 619
 620 /*
 621  * Reset a connection.  Discard all incoming and outgoing messages
 622  * and clear *_seq state.
 623  */
 624 static void ceph_msg_remove(struct ceph_msg *msg)
 625 {
 626         list_del_init(&msg->list_head);
 627
 628         ceph_msg_put(msg);
 629 }
 630 static void ceph_msg_remove_list(struct list_head *head)
 631 {
 632         while (!list_empty(head)) {
 633                 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
 634                                                         list_head);
 635                 ceph_msg_remove(msg);
 636         }
 637 }
 638
 639 static void reset_connection(struct ceph_connection *con)
 640 {
 641         /* reset connection, out_queue, msg_ and connect_seq */
 642         /* discard existing out_queue and msg_seq */
 643         dout("reset_connection %p\n", con);
 644         ceph_msg_remove_list(&con->out_queue);
 645         ceph_msg_remove_list(&con->out_sent);
 646
 647         if (con->in_msg) {
 648                 BUG_ON(con->in_msg->con != con);
 649                 ceph_msg_put(con->in_msg);
 650                 con->in_msg = NULL;
 651         }
 652
 653         con->connect_seq = 0;
 654         con->out_seq = 0;
 655         if (con->out_msg) {
 656                 BUG_ON(con->out_msg->con != con);
 657                 ceph_msg_put(con->out_msg);
 658                 con->out_msg = NULL;
 659         }
 660         con->in_seq = 0;
 661         con->in_seq_acked = 0;
 662
 663         con->out_skip = 0;
 664 }
 665
 666 /*
 667  * mark a peer down.  drop any open connections.
 668  */
 669 void ceph_con_close(struct ceph_connection *con)
 670 {
 671         mutex_lock(&con->mutex);
 672         dout("con_close %p peer %s\n", con,
 673              ceph_pr_addr(&con->peer_addr.in_addr));
 674         con->state = CON_STATE_CLOSED;
 675
 676         con_flag_clear(con, CON_FLAG_LOSSYTX);  /* so we retry next connect */
 677         con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
 678         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
 679         con_flag_clear(con, CON_FLAG_BACKOFF);
 680
 681         reset_connection(con);
 682         con->peer_global_seq = 0;
 683         cancel_con(con);
 684         con_close_socket(con);
 685         mutex_unlock(&con->mutex);
 686 }
 687 EXPORT_SYMBOL(ceph_con_close);
 688
 689 /*
 690  * Reopen a closed connection, with a new peer address.
 691  */
 692 void ceph_con_open(struct ceph_connection *con,
 693                    __u8 entity_type, __u64 entity_num,
 694                    struct ceph_entity_addr *addr)
 695 {
 696         mutex_lock(&con->mutex);
 697         dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
 698
 699         WARN_ON(con->state != CON_STATE_CLOSED);
 700         con->state = CON_STATE_PREOPEN;
 701
 702         con->peer_name.type = (__u8) entity_type;
 703         con->peer_name.num = cpu_to_le64(entity_num);
 704
 705         memcpy(&con->peer_addr, addr, sizeof(*addr));
 706         con->delay = 0;      /* reset backoff memory */
 707         mutex_unlock(&con->mutex);
 708         queue_con(con);
 709 }
 710 EXPORT_SYMBOL(ceph_con_open);
 711
 712 /*
 713  * return true if this connection ever successfully opened
 714  */
 715 bool ceph_con_opened(struct ceph_connection *con)
 716 {
 717         return con->connect_seq > 0;
 718 }
 719
 720 /*
 721  * initialize a new connection.
 722  */
 723 void ceph_con_init(struct ceph_connection *con, void *private,
 724         const struct ceph_connection_operations *ops,
 725         struct ceph_messenger *msgr)
 726 {
 727         dout("con_init %p\n", con);
 728         memset(con, 0, sizeof(*con));
 729         con->private = private;
 730         con->ops = ops;
 731         con->msgr = msgr;
 732
 733         con_sock_state_init(con);
 734
 735         mutex_init(&con->mutex);
 736         INIT_LIST_HEAD(&con->out_queue);
 737         INIT_LIST_HEAD(&con->out_sent);
 738         INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
 739
 740         con->state = CON_STATE_CLOSED;
 741 }
 742 EXPORT_SYMBOL(ceph_con_init);
 743
 744
 745 /*
 746  * We maintain a global counter to order connection attempts.  Get
 747  * a unique seq greater than @gt.
 748  */
 749 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
 750 {
 751         u32 ret;
 752
 753         spin_lock(&msgr->global_seq_lock);
 754         if (msgr->global_seq < gt)
 755                 msgr->global_seq = gt;
 756         ret = ++msgr->global_seq;
 757         spin_unlock(&msgr->global_seq_lock);
 758         return ret;
 759 }
 760
 761 static void con_out_kvec_reset(struct ceph_connection *con)
 762 {
 763         BUG_ON(con->out_skip);
 764
 765         con->out_kvec_left = 0;
 766         con->out_kvec_bytes = 0;
 767         con->out_kvec_cur = &con->out_kvec[0];
 768 }
 769
 770 static void con_out_kvec_add(struct ceph_connection *con,
 771                                 size_t size, void *data)
 772 {
 773         int index = con->out_kvec_left;
 774
 775         BUG_ON(con->out_skip);
 776         BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
 777
 778         con->out_kvec[index].iov_len = size;
 779         con->out_kvec[index].iov_base = data;
 780         con->out_kvec_left++;
 781         con->out_kvec_bytes += size;
 782 }
 783
 784 /*
 785  * Chop off a kvec from the end.  Return residual number of bytes for
 786  * that kvec, i.e. how many bytes would have been written if the kvec
 787  * hadn't been nuked.
 788  */
 789 static int con_out_kvec_skip(struct ceph_connection *con)
 790 {
 791         int off = con->out_kvec_cur - con->out_kvec;
 792         int skip = 0;
 793
 794         if (con->out_kvec_bytes > 0) {
 795                 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
 796                 BUG_ON(con->out_kvec_bytes < skip);
 797                 BUG_ON(!con->out_kvec_left);
 798                 con->out_kvec_bytes -= skip;
 799                 con->out_kvec_left--;
 800         }
 801
 802         return skip;
 803 }
 804
 805 #ifdef CONFIG_BLOCK
 806
 807 /*
 808  * For a bio data item, a piece is whatever remains of the next
 809  * entry in the current bio iovec, or the first entry in the next
 810  * bio in the list.
 811  */
 812 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
 813                                         size_t length)
 814 {
 815         struct ceph_msg_data *data = cursor->data;
 816         struct ceph_bio_iter *it = &cursor->bio_iter;
 817
 818         cursor->resid = min_t(size_t, length, data->bio_length);
 819         *it = data->bio_pos;
 820         if (cursor->resid < it->iter.bi_size)
 821                 it->iter.bi_size = cursor->resid;
 822
 823         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
 824         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
 825 }
 826
 827 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
 828                                                 size_t *page_offset,
 829                                                 size_t *length)
 830 {
 831         struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
 832                                            cursor->bio_iter.iter);
 833
 834         *page_offset = bv.bv_offset;
 835         *length = bv.bv_len;
 836         return bv.bv_page;
 837 }
 838
 839 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
 840                                         size_t bytes)
 841 {
 842         struct ceph_bio_iter *it = &cursor->bio_iter;
 843         struct page *page = bio_iter_page(it->bio, it->iter);
 844
 845         BUG_ON(bytes > cursor->resid);
 846         BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
 847         cursor->resid -= bytes;
 848         bio_advance_iter(it->bio, &it->iter, bytes);
 849
 850         if (!cursor->resid) {
 851                 BUG_ON(!cursor->last_piece);
 852                 return false;   /* no more data */
 853         }
 854
 855         if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
 856                        page == bio_iter_page(it->bio, it->iter)))
 857                 return false;   /* more bytes to process in this segment */
 858
 859         if (!it->iter.bi_size) {
 860                 it->bio = it->bio->bi_next;
 861                 it->iter = it->bio->bi_iter;
 862                 if (cursor->resid < it->iter.bi_size)
 863                         it->iter.bi_size = cursor->resid;
 864         }
 865
 866         BUG_ON(cursor->last_piece);
 867         BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
 868         cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
 869         return true;
 870 }
 871 #endif /* CONFIG_BLOCK */
 872
 873 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
 874                                         size_t length)
 875 {
 876         struct ceph_msg_data *data = cursor->data;
 877         struct bio_vec *bvecs = data->bvec_pos.bvecs;
 878
 879         cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
 880         cursor->bvec_iter = data->bvec_pos.iter;
 881         cursor->bvec_iter.bi_size = cursor->resid;
 882
 883         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
 884         cursor->last_piece =
 885             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
 886 }
 887
 888 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
 889                                                 size_t *page_offset,
 890                                                 size_t *length)
 891 {
 892         struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
 893                                            cursor->bvec_iter);
 894
 895         *page_offset = bv.bv_offset;
 896         *length = bv.bv_len;
 897         return bv.bv_page;
 898 }
 899
 900 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
 901                                         size_t bytes)
 902 {
 903         struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
 904         struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
 905
 906         BUG_ON(bytes > cursor->resid);
 907         BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
 908         cursor->resid -= bytes;
 909         bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
 910
 911         if (!cursor->resid) {
 912                 BUG_ON(!cursor->last_piece);
 913                 return false;   /* no more data */
 914         }
 915
 916         if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
 917                        page == bvec_iter_page(bvecs, cursor->bvec_iter)))
 918                 return false;   /* more bytes to process in this segment */
 919
 920         BUG_ON(cursor->last_piece);
 921         BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
 922         cursor->last_piece =
 923             cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
 924         return true;
 925 }
 926
 927 /*
 928  * For a page array, a piece comes from the first page in the array
 929  * that has not already been fully consumed.
 930  */
 931 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
 932                                         size_t length)
 933 {
 934         struct ceph_msg_data *data = cursor->data;
 935         int page_count;
 936
 937         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
 938
 939         BUG_ON(!data->pages);
 940         BUG_ON(!data->length);
 941
 942         cursor->resid = min(length, data->length);
 943         page_count = calc_pages_for(data->alignment, (u64)data->length);
 944         cursor->page_offset = data->alignment & ~PAGE_MASK;
 945         cursor->page_index = 0;
 946         BUG_ON(page_count > (int)USHRT_MAX);
 947         cursor->page_count = (unsigned short)page_count;
 948         BUG_ON(length > SIZE_MAX - cursor->page_offset);
 949         cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
 950 }
 951
 952 static struct page *
 953 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
 954                                         size_t *page_offset, size_t *length)
 955 {
 956         struct ceph_msg_data *data = cursor->data;
 957
 958         BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
 959
 960         BUG_ON(cursor->page_index >= cursor->page_count);
 961         BUG_ON(cursor->page_offset >= PAGE_SIZE);
 962
 963         *page_offset = cursor->page_offset;
 964         if (cursor->last_piece)
 965                 *length = cursor->resid;
 966         else
 967                 *length = PAGE_SIZE - *page_offset;
 968
 969         return data->pages[cursor->page_index];
 970 }
 971
 972 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
 973                                                 size_t bytes)
 974 {
 975         BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
 976
 977         BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
 978
 979         /* Advance the cursor page offset */
 980
 981         cursor->resid -= bytes;
 982         cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
 983         if (!bytes || cursor->page_offset)
 984                 return false;   /* more bytes to process in the current page */
 985
 986         if (!cursor->resid)
 987                 return false;   /* no more data */
 988
 989         /* Move on to the next page; offset is already at 0 */
 990
 991         BUG_ON(cursor->page_index >= cursor->page_count);
 992         cursor->page_index++;
 993         cursor->last_piece = cursor->resid <= PAGE_SIZE;
 994
 995         return true;
 996 }
 997
 998 /*
 999  * For a pagelist, a piece is whatever remains to be consumed in the
1000  * first page in the list, or the front of the next page.
1001  */
1002 static void
1003 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1004                                         size_t length)
1005 {
1006         struct ceph_msg_data *data = cursor->data;
1007         struct ceph_pagelist *pagelist;
1008         struct page *page;
1009
1010         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1011
1012         pagelist = data->pagelist;
1013         BUG_ON(!pagelist);
1014
1015         if (!length)
1016                 return;         /* pagelist can be assigned but empty */
1017
1018         BUG_ON(list_empty(&pagelist->head));
1019         page = list_first_entry(&pagelist->head, struct page, lru);
1020
1021         cursor->resid = min(length, pagelist->length);
1022         cursor->page = page;
1023         cursor->offset = 0;
1024         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1025 }
1026
1027 static struct page *
1028 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1029                                 size_t *page_offset, size_t *length)
1030 {
1031         struct ceph_msg_data *data = cursor->data;
1032         struct ceph_pagelist *pagelist;
1033
1034         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1035
1036         pagelist = data->pagelist;
1037         BUG_ON(!pagelist);
1038
1039         BUG_ON(!cursor->page);
1040         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1041
1042         /* offset of first page in pagelist is always 0 */
1043         *page_offset = cursor->offset & ~PAGE_MASK;
1044         if (cursor->last_piece)
1045                 *length = cursor->resid;
1046         else
1047                 *length = PAGE_SIZE - *page_offset;
1048
1049         return cursor->page;
1050 }
1051
1052 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1053                                                 size_t bytes)
1054 {
1055         struct ceph_msg_data *data = cursor->data;
1056         struct ceph_pagelist *pagelist;
1057
1058         BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1059
1060         pagelist = data->pagelist;
1061         BUG_ON(!pagelist);
1062
1063         BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1064         BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1065
1066         /* Advance the cursor offset */
1067
1068         cursor->resid -= bytes;
1069         cursor->offset += bytes;
1070         /* offset of first page in pagelist is always 0 */
1071         if (!bytes || cursor->offset & ~PAGE_MASK)
1072                 return false;   /* more bytes to process in the current page */
1073
1074         if (!cursor->resid)
1075                 return false;   /* no more data */
1076
1077         /* Move on to the next page */
1078
1079         BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1080         cursor->page = list_next_entry(cursor->page, lru);
1081         cursor->last_piece = cursor->resid <= PAGE_SIZE;
1082
1083         return true;
1084 }
1085
1086 /*
1087  * Message data is handled (sent or received) in pieces, where each
1088  * piece resides on a single page.  The network layer might not
1089  * consume an entire piece at once.  A data item's cursor keeps
1090  * track of which piece is next to process and how much remains to
1091  * be processed in that piece.  It also tracks whether the current
1092  * piece is the last one in the data item.
1093  */
1094 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1095 {
1096         size_t length = cursor->total_resid;
1097
1098         switch (cursor->data->type) {
1099         case CEPH_MSG_DATA_PAGELIST:
1100                 ceph_msg_data_pagelist_cursor_init(cursor, length);
1101                 break;
1102         case CEPH_MSG_DATA_PAGES:
1103                 ceph_msg_data_pages_cursor_init(cursor, length);
1104                 break;
1105 #ifdef CONFIG_BLOCK
1106         case CEPH_MSG_DATA_BIO:
1107                 ceph_msg_data_bio_cursor_init(cursor, length);
1108                 break;
1109 #endif /* CONFIG_BLOCK */
1110         case CEPH_MSG_DATA_BVECS:
1111                 ceph_msg_data_bvecs_cursor_init(cursor, length);
1112                 break;
1113         case CEPH_MSG_DATA_NONE:
1114         default:
1115                 /* BUG(); */
1116                 break;
1117         }
1118         cursor->need_crc = true;
1119 }
1120
1121 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1122 {
1123         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1124
1125         BUG_ON(!length);
1126         BUG_ON(length > msg->data_length);
1127         BUG_ON(!msg->num_data_items);
1128
1129         cursor->total_resid = length;
1130         cursor->data = msg->data;
1131
1132         __ceph_msg_data_cursor_init(cursor);
1133 }
1134
1135 /*
1136  * Return the page containing the next piece to process for a given
1137  * data item, and supply the page offset and length of that piece.
1138  * Indicate whether this is the last piece in this data item.
1139  */
1140 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1141                                         size_t *page_offset, size_t *length,
1142                                         bool *last_piece)
1143 {
1144         struct page *page;
1145
1146         switch (cursor->data->type) {
1147         case CEPH_MSG_DATA_PAGELIST:
1148                 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1149                 break;
1150         case CEPH_MSG_DATA_PAGES:
1151                 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1152                 break;
1153 #ifdef CONFIG_BLOCK
1154         case CEPH_MSG_DATA_BIO:
1155                 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1156                 break;
1157 #endif /* CONFIG_BLOCK */
1158         case CEPH_MSG_DATA_BVECS:
1159                 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1160                 break;
1161         case CEPH_MSG_DATA_NONE:
1162         default:
1163                 page = NULL;
1164                 break;
1165         }
1166
1167         BUG_ON(!page);
1168         BUG_ON(*page_offset + *length > PAGE_SIZE);
1169         BUG_ON(!*length);
1170         BUG_ON(*length > cursor->resid);
1171         if (last_piece)
1172                 *last_piece = cursor->last_piece;
1173
1174         return page;
1175 }
1176
1177 /*
1178  * Returns true if the result moves the cursor on to the next piece
1179  * of the data item.
1180  */
1181 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1182                                   size_t bytes)
1183 {
1184         bool new_piece;
1185
1186         BUG_ON(bytes > cursor->resid);
1187         switch (cursor->data->type) {
1188         case CEPH_MSG_DATA_PAGELIST:
1189                 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1190                 break;
1191         case CEPH_MSG_DATA_PAGES:
1192                 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1193                 break;
1194 #ifdef CONFIG_BLOCK
1195         case CEPH_MSG_DATA_BIO:
1196                 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1197                 break;
1198 #endif /* CONFIG_BLOCK */
1199         case CEPH_MSG_DATA_BVECS:
1200                 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1201                 break;
1202         case CEPH_MSG_DATA_NONE:
1203         default:
1204                 BUG();
1205                 break;
1206         }
1207         cursor->total_resid -= bytes;
1208
1209         if (!cursor->resid && cursor->total_resid) {
1210                 WARN_ON(!cursor->last_piece);
1211                 cursor->data++;
1212                 __ceph_msg_data_cursor_init(cursor);
1213                 new_piece = true;
1214         }
1215         cursor->need_crc = new_piece;
1216 }
1217
1218 static size_t sizeof_footer(struct ceph_connection *con)
1219 {
1220         return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1221             sizeof(struct ceph_msg_footer) :
1222             sizeof(struct ceph_msg_footer_old);
1223 }
1224
1225 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1226 {
1227         /* Initialize data cursor */
1228
1229         ceph_msg_data_cursor_init(msg, (size_t)data_len);
1230 }
1231
1232 /*
1233  * Prepare footer for currently outgoing message, and finish things
1234  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1235  */
1236 static void prepare_write_message_footer(struct ceph_connection *con)
1237 {
1238         struct ceph_msg *m = con->out_msg;
1239
1240         m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1241
1242         dout("prepare_write_message_footer %p\n", con);
1243         con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1244         if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1245                 if (con->ops->sign_message)
1246                         con->ops->sign_message(m);
1247                 else
1248                         m->footer.sig = 0;
1249         } else {
1250                 m->old_footer.flags = m->footer.flags;
1251         }
1252         con->out_more = m->more_to_follow;
1253         con->out_msg_done = true;
1254 }
1255
1256 /*
1257  * Prepare headers for the next outgoing message.
1258  */
1259 static void prepare_write_message(struct ceph_connection *con)
1260 {
1261         struct ceph_msg *m;
1262         u32 crc;
1263
1264         con_out_kvec_reset(con);
1265         con->out_msg_done = false;
1266
1267         /* Sneak an ack in there first?  If we can get it into the same
1268          * TCP packet that's a good thing. */
1269         if (con->in_seq > con->in_seq_acked) {
1270                 con->in_seq_acked = con->in_seq;
1271                 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1272                 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1273                 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1274                         &con->out_temp_ack);
1275         }
1276
1277         BUG_ON(list_empty(&con->out_queue));
1278         m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1279         con->out_msg = m;
1280         BUG_ON(m->con != con);
1281
1282         /* put message on sent list */
1283         ceph_msg_get(m);
1284         list_move_tail(&m->list_head, &con->out_sent);
1285
1286         /*
1287          * only assign outgoing seq # if we haven't sent this message
1288          * yet.  if it is requeued, resend with it's original seq.
1289          */
1290         if (m->needs_out_seq) {
1291                 m->hdr.seq = cpu_to_le64(++con->out_seq);
1292                 m->needs_out_seq = false;
1293
1294                 if (con->ops->reencode_message)
1295                         con->ops->reencode_message(m);
1296         }
1297
1298         dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1299              m, con->out_seq, le16_to_cpu(m->hdr.type),
1300              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1301              m->data_length);
1302         WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1303         WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1304
1305         /* tag + hdr + front + middle */
1306         con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1307         con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1308         con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1309
1310         if (m->middle)
1311                 con_out_kvec_add(con, m->middle->vec.iov_len,
1312                         m->middle->vec.iov_base);
1313
1314         /* fill in hdr crc and finalize hdr */
1315         crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1316         con->out_msg->hdr.crc = cpu_to_le32(crc);
1317         memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1318
1319         /* fill in front and middle crc, footer */
1320         crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1321         con->out_msg->footer.front_crc = cpu_to_le32(crc);
1322         if (m->middle) {
1323                 crc = crc32c(0, m->middle->vec.iov_base,
1324                                 m->middle->vec.iov_len);
1325                 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1326         } else
1327                 con->out_msg->footer.middle_crc = 0;
1328         dout("%s front_crc %u middle_crc %u\n", __func__,
1329              le32_to_cpu(con->out_msg->footer.front_crc),
1330              le32_to_cpu(con->out_msg->footer.middle_crc));
1331         con->out_msg->footer.flags = 0;
1332
1333         /* is there a data payload? */
1334         con->out_msg->footer.data_crc = 0;
1335         if (m->data_length) {
1336                 prepare_message_data(con->out_msg, m->data_length);
1337                 con->out_more = 1;  /* data + footer will follow */
1338         } else {
1339                 /* no, queue up footer too and be done */
1340                 prepare_write_message_footer(con);
1341         }
1342
1343         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1344 }
1345
1346 /*
1347  * Prepare an ack.
1348  */
1349 static void prepare_write_ack(struct ceph_connection *con)
1350 {
1351         dout("prepare_write_ack %p %llu -> %llu\n", con,
1352              con->in_seq_acked, con->in_seq);
1353         con->in_seq_acked = con->in_seq;
1354
1355         con_out_kvec_reset(con);
1356
1357         con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1358
1359         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1360         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1361                                 &con->out_temp_ack);
1362
1363         con->out_more = 1;  /* more will follow.. eventually.. */
1364         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1365 }
1366
1367 /*
1368  * Prepare to share the seq during handshake
1369  */
1370 static void prepare_write_seq(struct ceph_connection *con)
1371 {
1372         dout("prepare_write_seq %p %llu -> %llu\n", con,
1373              con->in_seq_acked, con->in_seq);
1374         con->in_seq_acked = con->in_seq;
1375
1376         con_out_kvec_reset(con);
1377
1378         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1379         con_out_kvec_add(con, sizeof (con->out_temp_ack),
1380                          &con->out_temp_ack);
1381
1382         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1383 }
1384
1385 /*
1386  * Prepare to write keepalive byte.
1387  */
1388 static void prepare_write_keepalive(struct ceph_connection *con)
1389 {
1390         dout("prepare_write_keepalive %p\n", con);
1391         con_out_kvec_reset(con);
1392         if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1393                 struct timespec64 now;
1394
1395                 ktime_get_real_ts64(&now);
1396                 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1397                 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
1398                 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1399                                  &con->out_temp_keepalive2);
1400         } else {
1401                 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1402         }
1403         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1404 }
1405
1406 /*
1407  * Connection negotiation.
1408  */
1409
1410 static int get_connect_authorizer(struct ceph_connection *con)
1411 {
1412         struct ceph_auth_handshake *auth;
1413         int auth_proto;
1414
1415         if (!con->ops->get_authorizer) {
1416                 con->auth = NULL;
1417                 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1418                 con->out_connect.authorizer_len = 0;
1419                 return 0;
1420         }
1421
1422         auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1423         if (IS_ERR(auth))
1424                 return PTR_ERR(auth);
1425
1426         con->auth = auth;
1427         con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1428         con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1429         return 0;
1430 }
1431
1432 /*
1433  * We connected to a peer and are saying hello.
1434  */
1435 static void prepare_write_banner(struct ceph_connection *con)
1436 {
1437         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1438         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1439                                         &con->msgr->my_enc_addr);
1440
1441         con->out_more = 0;
1442         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1443 }
1444
1445 static void __prepare_write_connect(struct ceph_connection *con)
1446 {
1447         con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1448         if (con->auth)
1449                 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1450                                  con->auth->authorizer_buf);
1451
1452         con->out_more = 0;
1453         con_flag_set(con, CON_FLAG_WRITE_PENDING);
1454 }
1455
1456 static int prepare_write_connect(struct ceph_connection *con)
1457 {
1458         unsigned int global_seq = get_global_seq(con->msgr, 0);
1459         int proto;
1460         int ret;
1461
1462         switch (con->peer_name.type) {
1463         case CEPH_ENTITY_TYPE_MON:
1464                 proto = CEPH_MONC_PROTOCOL;
1465                 break;
1466         case CEPH_ENTITY_TYPE_OSD:
1467                 proto = CEPH_OSDC_PROTOCOL;
1468                 break;
1469         case CEPH_ENTITY_TYPE_MDS:
1470                 proto = CEPH_MDSC_PROTOCOL;
1471                 break;
1472         default:
1473                 BUG();
1474         }
1475
1476         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1477              con->connect_seq, global_seq, proto);
1478
1479         con->out_connect.features =
1480             cpu_to_le64(from_msgr(con->msgr)->supported_features);
1481         con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1482         con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1483         con->out_connect.global_seq = cpu_to_le32(global_seq);
1484         con->out_connect.protocol_version = cpu_to_le32(proto);
1485         con->out_connect.flags = 0;
1486
1487         ret = get_connect_authorizer(con);
1488         if (ret)
1489                 return ret;
1490
1491         __prepare_write_connect(con);
1492         return 0;
1493 }
1494
1495 /*
1496  * write as much of pending kvecs to the socket as we can.
1497  *  1 -> done
1498  *  0 -> socket full, but more to do
1499  * <0 -> error
1500  */
1501 static int write_partial_kvec(struct ceph_connection *con)
1502 {
1503         int ret;
1504
1505         dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1506         while (con->out_kvec_bytes > 0) {
1507                 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1508                                        con->out_kvec_left, con->out_kvec_bytes,
1509                                        con->out_more);
1510                 if (ret <= 0)
1511                         goto out;
1512                 con->out_kvec_bytes -= ret;
1513                 if (con->out_kvec_bytes == 0)
1514                         break;            /* done */
1515
1516                 /* account for full iov entries consumed */
1517                 while (ret >= con->out_kvec_cur->iov_len) {
1518                         BUG_ON(!con->out_kvec_left);
1519                         ret -= con->out_kvec_cur->iov_len;
1520                         con->out_kvec_cur++;
1521                         con->out_kvec_left--;
1522                 }
1523                 /* and for a partially-consumed entry */
1524                 if (ret) {
1525                         con->out_kvec_cur->iov_len -= ret;
1526                         con->out_kvec_cur->iov_base += ret;
1527                 }
1528         }
1529         con->out_kvec_left = 0;
1530         ret = 1;
1531 out:
1532         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1533              con->out_kvec_bytes, con->out_kvec_left, ret);
1534         return ret;  /* done! */
1535 }
1536
1537 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1538                                 unsigned int page_offset,
1539                                 unsigned int length)
1540 {
1541         char *kaddr;
1542
1543         kaddr = kmap(page);
1544         BUG_ON(kaddr == NULL);
1545         crc = crc32c(crc, kaddr + page_offset, length);
1546         kunmap(page);
1547
1548         return crc;
1549 }
1550 /*
1551  * Write as much message data payload as we can.  If we finish, queue
1552  * up the footer.
1553  *  1 -> done, footer is now queued in out_kvec[].
1554  *  0 -> socket full, but more to do
1555  * <0 -> error
1556  */
1557 static int write_partial_message_data(struct ceph_connection *con)
1558 {
1559         struct ceph_msg *msg = con->out_msg;
1560         struct ceph_msg_data_cursor *cursor = &msg->cursor;
1561         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1562         int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1563         u32 crc;
1564
1565         dout("%s %p msg %p\n", __func__, con, msg);
1566
1567         if (!msg->num_data_items)
1568                 return -EINVAL;
1569
1570         /*
1571          * Iterate through each page that contains data to be
1572          * written, and send as much as possible for each.
1573          *
1574          * If we are calculating the data crc (the default), we will
1575          * need to map the page.  If we have no pages, they have
1576          * been revoked, so use the zero page.
1577          */
1578         crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1579         while (cursor->total_resid) {
1580                 struct page *page;
1581                 size_t page_offset;
1582                 size_t length;
1583                 int ret;
1584
1585                 if (!cursor->resid) {
1586                         ceph_msg_data_advance(cursor, 0);
1587                         continue;
1588                 }
1589
1590                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
1591                 if (length == cursor->total_resid)
1592                         more = MSG_MORE;
1593                 ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
1594                                         more);
1595                 if (ret <= 0) {
1596                         if (do_datacrc)
1597                                 msg->footer.data_crc = cpu_to_le32(crc);
1598
1599                         return ret;
1600                 }
1601                 if (do_datacrc && cursor->need_crc)
1602                         crc = ceph_crc32c_page(crc, page, page_offset, length);
1603                 ceph_msg_data_advance(cursor, (size_t)ret);
1604         }
1605
1606         dout("%s %p msg %p done\n", __func__, con, msg);
1607
1608         /* prepare and queue up footer, too */
1609         if (do_datacrc)
1610                 msg->footer.data_crc = cpu_to_le32(crc);
1611         else
1612                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1613         con_out_kvec_reset(con);
1614         prepare_write_message_footer(con);
1615
1616         return 1;       /* must return > 0 to indicate success */
1617 }
1618
1619 /*
1620  * write some zeros
1621  */
1622 static int write_partial_skip(struct ceph_connection *con)
1623 {
1624         int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1625         int ret;
1626
1627         dout("%s %p %d left\n", __func__, con, con->out_skip);
1628         while (con->out_skip > 0) {
1629                 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1630
1631                 if (size == con->out_skip)
1632                         more = MSG_MORE;
1633                 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, more);
1634                 if (ret <= 0)
1635                         goto out;
1636                 con->out_skip -= ret;
1637         }
1638         ret = 1;
1639 out:
1640         return ret;
1641 }
1642
1643 /*
1644  * Prepare to read connection handshake, or an ack.
1645  */
1646 static void prepare_read_banner(struct ceph_connection *con)
1647 {
1648         dout("prepare_read_banner %p\n", con);
1649         con->in_base_pos = 0;
1650 }
1651
1652 static void prepare_read_connect(struct ceph_connection *con)
1653 {
1654         dout("prepare_read_connect %p\n", con);
1655         con->in_base_pos = 0;
1656 }
1657
1658 static void prepare_read_ack(struct ceph_connection *con)
1659 {
1660         dout("prepare_read_ack %p\n", con);
1661         con->in_base_pos = 0;
1662 }
1663
1664 static void prepare_read_seq(struct ceph_connection *con)
1665 {
1666         dout("prepare_read_seq %p\n", con);
1667         con->in_base_pos = 0;
1668         con->in_tag = CEPH_MSGR_TAG_SEQ;
1669 }
1670
1671 static void prepare_read_tag(struct ceph_connection *con)
1672 {
1673         dout("prepare_read_tag %p\n", con);
1674         con->in_base_pos = 0;
1675         con->in_tag = CEPH_MSGR_TAG_READY;
1676 }
1677
1678 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1679 {
1680         dout("prepare_read_keepalive_ack %p\n", con);
1681         con->in_base_pos = 0;
1682 }
1683
1684 /*
1685  * Prepare to read a message.
1686  */
1687 static int prepare_read_message(struct ceph_connection *con)
1688 {
1689         dout("prepare_read_message %p\n", con);
1690         BUG_ON(con->in_msg != NULL);
1691         con->in_base_pos = 0;
1692         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1693         return 0;
1694 }
1695
1696
1697 static int read_partial(struct ceph_connection *con,
1698                         int end, int size, void *object)
1699 {
1700         while (con->in_base_pos < end) {
1701                 int left = end - con->in_base_pos;
1702                 int have = size - left;
1703                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1704                 if (ret <= 0)
1705                         return ret;
1706                 con->in_base_pos += ret;
1707         }
1708         return 1;
1709 }
1710
1711
1712 /*
1713  * Read all or part of the connect-side handshake on a new connection
1714  */
1715 static int read_partial_banner(struct ceph_connection *con)
1716 {
1717         int size;
1718         int end;
1719         int ret;
1720
1721         dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1722
1723         /* peer's banner */
1724         size = strlen(CEPH_BANNER);
1725         end = size;
1726         ret = read_partial(con, end, size, con->in_banner);
1727         if (ret <= 0)
1728                 goto out;
1729
1730         size = sizeof (con->actual_peer_addr);
1731         end += size;
1732         ret = read_partial(con, end, size, &con->actual_peer_addr);
1733         if (ret <= 0)
1734                 goto out;
1735
1736         size = sizeof (con->peer_addr_for_me);
1737         end += size;
1738         ret = read_partial(con, end, size, &con->peer_addr_for_me);
1739         if (ret <= 0)
1740                 goto out;
1741
1742 out:
1743         return ret;
1744 }
1745
1746 static int read_partial_connect(struct ceph_connection *con)
1747 {
1748         int size;
1749         int end;
1750         int ret;
1751
1752         dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1753
1754         size = sizeof (con->in_reply);
1755         end = size;
1756         ret = read_partial(con, end, size, &con->in_reply);
1757         if (ret <= 0)
1758                 goto out;
1759
1760         if (con->auth) {
1761                 size = le32_to_cpu(con->in_reply.authorizer_len);
1762                 if (size > con->auth->authorizer_reply_buf_len) {
1763                         pr_err("authorizer reply too big: %d > %zu\n", size,
1764                                con->auth->authorizer_reply_buf_len);
1765                         ret = -EINVAL;
1766                         goto out;
1767                 }
1768
1769                 end += size;
1770                 ret = read_partial(con, end, size,
1771                                    con->auth->authorizer_reply_buf);
1772                 if (ret <= 0)
1773                         goto out;
1774         }
1775
1776         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1777              con, (int)con->in_reply.tag,
1778              le32_to_cpu(con->in_reply.connect_seq),
1779              le32_to_cpu(con->in_reply.global_seq));
1780 out:
1781         return ret;
1782 }
1783
1784 /*
1785  * Verify the hello banner looks okay.
1786  */
1787 static int verify_hello(struct ceph_connection *con)
1788 {
1789         if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1790                 pr_err("connect to %s got bad banner\n",
1791                        ceph_pr_addr(&con->peer_addr.in_addr));
1792                 con->error_msg = "protocol error, bad banner";
1793                 return -1;
1794         }
1795         return 0;
1796 }
1797
1798 static bool addr_is_blank(struct ceph_entity_addr *addr)
1799 {
1800         struct sockaddr_storage ss = addr->in_addr; /* align */
1801         struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1802         struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1803
1804         switch (ss.ss_family) {
1805         case AF_INET:
1806                 return addr4->s_addr == htonl(INADDR_ANY);
1807         case AF_INET6:
1808                 return ipv6_addr_any(addr6);
1809         default:
1810                 return true;
1811         }
1812 }
1813
1814 static int addr_port(struct ceph_entity_addr *addr)
1815 {
1816         switch (get_unaligned(&addr->in_addr.ss_family)) {
1817         case AF_INET:
1818                 return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1819         case AF_INET6:
1820                 return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1821         }
1822         return 0;
1823 }
1824
1825 static void addr_set_port(struct ceph_entity_addr *addr, int p)
1826 {
1827         switch (get_unaligned(&addr->in_addr.ss_family)) {
1828         case AF_INET:
1829                 put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1830                 break;
1831         case AF_INET6:
1832                 put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1833                 break;
1834         }
1835 }
1836
1837 /*
1838  * Unlike other *_pton function semantics, zero indicates success.
1839  */
1840 static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1841                 char delim, const char **ipend)
1842 {
1843         memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1844
1845         if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1846                 put_unaligned(AF_INET, &addr->in_addr.ss_family);
1847                 return 0;
1848         }
1849
1850         if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1851                 put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1852                 return 0;
1853         }
1854
1855         return -EINVAL;
1856 }
1857
1858 /*
1859  * Extract hostname string and resolve using kernel DNS facility.
1860  */
1861 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1862 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1863                 struct ceph_entity_addr *addr, char delim, const char **ipend)
1864 {
1865         const char *end, *delim_p;
1866         char *colon_p, *ip_addr = NULL;
1867         int ip_len, ret;
1868
1869         /*
1870          * The end of the hostname occurs immediately preceding the delimiter or
1871          * the port marker (':') where the delimiter takes precedence.
1872          */
1873         delim_p = memchr(name, delim, namelen);
1874         colon_p = memchr(name, ':', namelen);
1875
1876         if (delim_p && colon_p)
1877                 end = delim_p < colon_p ? delim_p : colon_p;
1878         else if (!delim_p && colon_p)
1879                 end = colon_p;
1880         else {
1881                 end = delim_p;
1882                 if (!end) /* case: hostname:/ */
1883                         end = name + namelen;
1884         }
1885
1886         if (end <= name)
1887                 return -EINVAL;
1888
1889         /* do dns_resolve upcall */
1890         ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1891         if (ip_len > 0)
1892                 ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1893         else
1894                 ret = -ESRCH;
1895
1896         kfree(ip_addr);
1897
1898         *ipend = end;
1899
1900         pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1901                         ret, ret ? "failed" : ceph_pr_addr(&addr->in_addr));
1902
1903         return ret;
1904 }
1905 #else
1906 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1907                 struct ceph_entity_addr *addr, char delim, const char **ipend)
1908 {
1909         return -EINVAL;
1910 }
1911 #endif
1912
1913 /*
1914  * Parse a server name (IP or hostname). If a valid IP address is not found
1915  * then try to extract a hostname to resolve using userspace DNS upcall.
1916  */
1917 static int ceph_parse_server_name(const char *name, size_t namelen,
1918                 struct ceph_entity_addr *addr, char delim, const char **ipend)
1919 {
1920         int ret;
1921
1922         ret = ceph_pton(name, namelen, addr, delim, ipend);
1923         if (ret)
1924                 ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1925
1926         return ret;
1927 }
1928
1929 /*
1930  * Parse an ip[:port] list into an addr array.  Use the default
1931  * monitor port if a port isn't specified.
1932  */
1933 int ceph_parse_ips(const char *c, const char *end,
1934                    struct ceph_entity_addr *addr,
1935                    int max_count, int *count)
1936 {
1937         int i, ret = -EINVAL;
1938         const char *p = c;
1939
1940         dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1941         for (i = 0; i < max_count; i++) {
1942                 const char *ipend;
1943                 int port;
1944                 char delim = ',';
1945
1946                 if (*p == '[') {
1947                         delim = ']';
1948                         p++;
1949                 }
1950
1951                 ret = ceph_parse_server_name(p, end - p, &addr[i], delim, &ipend);
1952                 if (ret)
1953                         goto bad;
1954                 ret = -EINVAL;
1955
1956                 p = ipend;
1957
1958                 if (delim == ']') {
1959                         if (*p != ']') {
1960                                 dout("missing matching ']'\n");
1961                                 goto bad;
1962                         }
1963                         p++;
1964                 }
1965
1966                 /* port? */
1967                 if (p < end && *p == ':') {
1968                         port = 0;
1969                         p++;
1970                         while (p < end && *p >= '0' && *p <= '9') {
1971                                 port = (port * 10) + (*p - '0');
1972                                 p++;
1973                         }
1974                         if (port == 0)
1975                                 port = CEPH_MON_PORT;
1976                         else if (port > 65535)
1977                                 goto bad;
1978                 } else {
1979                         port = CEPH_MON_PORT;
1980                 }
1981
1982                 addr_set_port(&addr[i], port);
1983
1984                 dout("parse_ips got %s\n", ceph_pr_addr(&addr[i].in_addr));
1985
1986                 if (p == end)
1987                         break;
1988                 if (*p != ',')
1989                         goto bad;
1990                 p++;
1991         }
1992
1993         if (p != end)
1994                 goto bad;
1995
1996         if (count)
1997                 *count = i + 1;
1998         return 0;
1999
2000 bad:
2001         pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2002         return ret;
2003 }
2004 EXPORT_SYMBOL(ceph_parse_ips);
2005
2006 static int process_banner(struct ceph_connection *con)
2007 {
2008         dout("process_banner on %p\n", con);
2009
2010         if (verify_hello(con) < 0)
2011                 return -1;
2012
2013         ceph_decode_addr(&con->actual_peer_addr);
2014         ceph_decode_addr(&con->peer_addr_for_me);
2015
2016         /*
2017          * Make sure the other end is who we wanted.  note that the other
2018          * end may not yet know their ip address, so if it's 0.0.0.0, give
2019          * them the benefit of the doubt.
2020          */
2021         if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2022                    sizeof(con->peer_addr)) != 0 &&
2023             !(addr_is_blank(&con->actual_peer_addr) &&
2024               con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2025                 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2026                         ceph_pr_addr(&con->peer_addr.in_addr),
2027                         (int)le32_to_cpu(con->peer_addr.nonce),
2028                         ceph_pr_addr(&con->actual_peer_addr.in_addr),
2029                         (int)le32_to_cpu(con->actual_peer_addr.nonce));
2030                 con->error_msg = "wrong peer at address";
2031                 return -1;
2032         }
2033
2034         /*
2035          * did we learn our address?
2036          */
2037         if (addr_is_blank(&con->msgr->inst.addr)) {
2038                 int port = addr_port(&con->msgr->inst.addr);
2039
2040                 memcpy(&con->msgr->inst.addr.in_addr,
2041                        &con->peer_addr_for_me.in_addr,
2042                        sizeof(con->peer_addr_for_me.in_addr));
2043                 addr_set_port(&con->msgr->inst.addr, port);
2044                 encode_my_addr(con->msgr);
2045                 dout("process_banner learned my addr is %s\n",
2046                      ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2047         }
2048
2049         return 0;
2050 }
2051
2052 static int process_connect(struct ceph_connection *con)
2053 {
2054         u64 sup_feat = from_msgr(con->msgr)->supported_features;
2055         u64 req_feat = from_msgr(con->msgr)->required_features;
2056         u64 server_feat = le64_to_cpu(con->in_reply.features);
2057         int ret;
2058
2059         dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2060
2061         if (con->auth) {
2062                 int len = le32_to_cpu(con->in_reply.authorizer_len);
2063
2064                 /*
2065                  * Any connection that defines ->get_authorizer()
2066                  * should also define ->add_authorizer_challenge() and
2067                  * ->verify_authorizer_reply().
2068                  *
2069                  * See get_connect_authorizer().
2070                  */
2071                 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2072                         ret = con->ops->add_authorizer_challenge(
2073                                     con, con->auth->authorizer_reply_buf, len);
2074                         if (ret < 0)
2075                                 return ret;
2076
2077                         con_out_kvec_reset(con);
2078                         __prepare_write_connect(con);
2079                         prepare_read_connect(con);
2080                         return 0;
2081                 }
2082
2083                 if (len) {
2084                         ret = con->ops->verify_authorizer_reply(con);
2085                         if (ret < 0) {
2086                                 con->error_msg = "bad authorize reply";
2087                                 return ret;
2088                         }
2089                 }
2090         }
2091
2092         switch (con->in_reply.tag) {
2093         case CEPH_MSGR_TAG_FEATURES:
2094                 pr_err("%s%lld %s feature set mismatch,"
2095                        " my %llx < server's %llx, missing %llx\n",
2096                        ENTITY_NAME(con->peer_name),
2097                        ceph_pr_addr(&con->peer_addr.in_addr),
2098                        sup_feat, server_feat, server_feat & ~sup_feat);
2099                 con->error_msg = "missing required protocol features";
2100                 reset_connection(con);
2101                 return -1;
2102
2103         case CEPH_MSGR_TAG_BADPROTOVER:
2104                 pr_err("%s%lld %s protocol version mismatch,"
2105                        " my %d != server's %d\n",
2106                        ENTITY_NAME(con->peer_name),
2107                        ceph_pr_addr(&con->peer_addr.in_addr),
2108                        le32_to_cpu(con->out_connect.protocol_version),
2109                        le32_to_cpu(con->in_reply.protocol_version));
2110                 con->error_msg = "protocol version mismatch";
2111                 reset_connection(con);
2112                 return -1;
2113
2114         case CEPH_MSGR_TAG_BADAUTHORIZER:
2115                 con->auth_retry++;
2116                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2117                      con->auth_retry);
2118                 if (con->auth_retry == 2) {
2119                         con->error_msg = "connect authorization failure";
2120                         return -1;
2121                 }
2122                 con_out_kvec_reset(con);
2123                 ret = prepare_write_connect(con);
2124                 if (ret < 0)
2125                         return ret;
2126                 prepare_read_connect(con);
2127                 break;
2128
2129         case CEPH_MSGR_TAG_RESETSESSION:
2130                 /*
2131                  * If we connected with a large connect_seq but the peer
2132                  * has no record of a session with us (no connection, or
2133                  * connect_seq == 0), they will send RESETSESION to indicate
2134                  * that they must have reset their session, and may have
2135                  * dropped messages.
2136                  */
2137                 dout("process_connect got RESET peer seq %u\n",
2138                      le32_to_cpu(con->in_reply.connect_seq));
2139                 pr_err("%s%lld %s connection reset\n",
2140                        ENTITY_NAME(con->peer_name),
2141                        ceph_pr_addr(&con->peer_addr.in_addr));
2142                 reset_connection(con);
2143                 con_out_kvec_reset(con);
2144                 ret = prepare_write_connect(con);
2145                 if (ret < 0)
2146                         return ret;
2147                 prepare_read_connect(con);
2148
2149                 /* Tell ceph about it. */
2150                 mutex_unlock(&con->mutex);
2151                 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2152                 if (con->ops->peer_reset)
2153                         con->ops->peer_reset(con);
2154                 mutex_lock(&con->mutex);
2155                 if (con->state != CON_STATE_NEGOTIATING)
2156                         return -EAGAIN;
2157                 break;
2158
2159         case CEPH_MSGR_TAG_RETRY_SESSION:
2160                 /*
2161                  * If we sent a smaller connect_seq than the peer has, try
2162                  * again with a larger value.
2163                  */
2164                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2165                      le32_to_cpu(con->out_connect.connect_seq),
2166                      le32_to_cpu(con->in_reply.connect_seq));
2167                 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2168                 con_out_kvec_reset(con);
2169                 ret = prepare_write_connect(con);
2170                 if (ret < 0)
2171                         return ret;
2172                 prepare_read_connect(con);
2173                 break;
2174
2175         case CEPH_MSGR_TAG_RETRY_GLOBAL:
2176                 /*
2177                  * If we sent a smaller global_seq than the peer has, try
2178                  * again with a larger value.
2179                  */
2180                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2181                      con->peer_global_seq,
2182                      le32_to_cpu(con->in_reply.global_seq));
2183                 get_global_seq(con->msgr,
2184                                le32_to_cpu(con->in_reply.global_seq));
2185                 con_out_kvec_reset(con);
2186                 ret = prepare_write_connect(con);
2187                 if (ret < 0)
2188                         return ret;
2189                 prepare_read_connect(con);
2190                 break;
2191
2192         case CEPH_MSGR_TAG_SEQ:
2193         case CEPH_MSGR_TAG_READY:
2194                 if (req_feat & ~server_feat) {
2195                         pr_err("%s%lld %s protocol feature mismatch,"
2196                                " my required %llx > server's %llx, need %llx\n",
2197                                ENTITY_NAME(con->peer_name),
2198                                ceph_pr_addr(&con->peer_addr.in_addr),
2199                                req_feat, server_feat, req_feat & ~server_feat);
2200                         con->error_msg = "missing required protocol features";
2201                         reset_connection(con);
2202                         return -1;
2203                 }
2204
2205                 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2206                 con->state = CON_STATE_OPEN;
2207                 con->auth_retry = 0;    /* we authenticated; clear flag */
2208                 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2209                 con->connect_seq++;
2210                 con->peer_features = server_feat;
2211                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2212                      con->peer_global_seq,
2213                      le32_to_cpu(con->in_reply.connect_seq),
2214                      con->connect_seq);
2215                 WARN_ON(con->connect_seq !=
2216                         le32_to_cpu(con->in_reply.connect_seq));
2217
2218                 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2219                         con_flag_set(con, CON_FLAG_LOSSYTX);
2220
2221                 con->delay = 0;      /* reset backoff memory */
2222
2223                 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2224                         prepare_write_seq(con);
2225                         prepare_read_seq(con);
2226                 } else {
2227                         prepare_read_tag(con);
2228                 }
2229                 break;
2230
2231         case CEPH_MSGR_TAG_WAIT:
2232                 /*
2233                  * If there is a connection race (we are opening
2234                  * connections to each other), one of us may just have
2235                  * to WAIT.  This shouldn't happen if we are the
2236                  * client.
2237                  */
2238                 con->error_msg = "protocol error, got WAIT as client";
2239                 return -1;
2240
2241         default:
2242                 con->error_msg = "protocol error, garbage tag during connect";
2243                 return -1;
2244         }
2245         return 0;
2246 }
2247
2248
2249 /*
2250  * read (part of) an ack
2251  */
2252 static int read_partial_ack(struct ceph_connection *con)
2253 {
2254         int size = sizeof (con->in_temp_ack);
2255         int end = size;
2256
2257         return read_partial(con, end, size, &con->in_temp_ack);
2258 }
2259
2260 /*
2261  * We can finally discard anything that's been acked.
2262  */
2263 static void process_ack(struct ceph_connection *con)
2264 {
2265         struct ceph_msg *m;
2266         u64 ack = le64_to_cpu(con->in_temp_ack);
2267         u64 seq;
2268         bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2269         struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2270
2271         /*
2272          * In the reconnect case, con_fault() has requeued messages
2273          * in out_sent. We should cleanup old messages according to
2274          * the reconnect seq.
2275          */
2276         while (!list_empty(list)) {
2277                 m = list_first_entry(list, struct ceph_msg, list_head);
2278                 if (reconnect && m->needs_out_seq)
2279                         break;
2280                 seq = le64_to_cpu(m->hdr.seq);
2281                 if (seq > ack)
2282                         break;
2283                 dout("got ack for seq %llu type %d at %p\n", seq,
2284                      le16_to_cpu(m->hdr.type), m);
2285                 m->ack_stamp = jiffies;
2286                 ceph_msg_remove(m);
2287         }
2288
2289         prepare_read_tag(con);
2290 }
2291
2292
2293 static int read_partial_message_section(struct ceph_connection *con,
2294                                         struct kvec *section,
2295                                         unsigned int sec_len, u32 *crc)
2296 {
2297         int ret, left;
2298
2299         BUG_ON(!section);
2300
2301         while (section->iov_len < sec_len) {
2302                 BUG_ON(section->iov_base == NULL);
2303                 left = sec_len - section->iov_len;
2304                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2305                                        section->iov_len, left);
2306                 if (ret <= 0)
2307                         return ret;
2308                 section->iov_len += ret;
2309         }
2310         if (section->iov_len == sec_len)
2311                 *crc = crc32c(0, section->iov_base, section->iov_len);
2312
2313         return 1;
2314 }
2315
2316 static int read_partial_msg_data(struct ceph_connection *con)
2317 {
2318         struct ceph_msg *msg = con->in_msg;
2319         struct ceph_msg_data_cursor *cursor = &msg->cursor;
2320         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2321         struct page *page;
2322         size_t page_offset;
2323         size_t length;
2324         u32 crc = 0;
2325         int ret;
2326
2327         if (!msg->num_data_items)
2328                 return -EIO;
2329
2330         if (do_datacrc)
2331                 crc = con->in_data_crc;
2332         while (cursor->total_resid) {
2333                 if (!cursor->resid) {
2334                         ceph_msg_data_advance(cursor, 0);
2335                         continue;
2336                 }
2337
2338                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2339                 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2340                 if (ret <= 0) {
2341                         if (do_datacrc)
2342                                 con->in_data_crc = crc;
2343
2344                         return ret;
2345                 }
2346
2347                 if (do_datacrc)
2348                         crc = ceph_crc32c_page(crc, page, page_offset, ret);
2349                 ceph_msg_data_advance(cursor, (size_t)ret);
2350         }
2351         if (do_datacrc)
2352                 con->in_data_crc = crc;
2353
2354         return 1;       /* must return > 0 to indicate success */
2355 }
2356
2357 /*
2358  * read (part of) a message.
2359  */
2360 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2361
2362 static int read_partial_message(struct ceph_connection *con)
2363 {
2364         struct ceph_msg *m = con->in_msg;
2365         int size;
2366         int end;
2367         int ret;
2368         unsigned int front_len, middle_len, data_len;
2369         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2370         bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2371         u64 seq;
2372         u32 crc;
2373
2374         dout("read_partial_message con %p msg %p\n", con, m);
2375
2376         /* header */
2377         size = sizeof (con->in_hdr);
2378         end = size;
2379         ret = read_partial(con, end, size, &con->in_hdr);
2380         if (ret <= 0)
2381                 return ret;
2382
2383         crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2384         if (cpu_to_le32(crc) != con->in_hdr.crc) {
2385                 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2386                        crc, con->in_hdr.crc);
2387                 return -EBADMSG;
2388         }
2389
2390         front_len = le32_to_cpu(con->in_hdr.front_len);
2391         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2392                 return -EIO;
2393         middle_len = le32_to_cpu(con->in_hdr.middle_len);
2394         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2395                 return -EIO;
2396         data_len = le32_to_cpu(con->in_hdr.data_len);
2397         if (data_len > CEPH_MSG_MAX_DATA_LEN)
2398                 return -EIO;
2399
2400         /* verify seq# */
2401         seq = le64_to_cpu(con->in_hdr.seq);
2402         if ((s64)seq - (s64)con->in_seq < 1) {
2403                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2404                         ENTITY_NAME(con->peer_name),
2405                         ceph_pr_addr(&con->peer_addr.in_addr),
2406                         seq, con->in_seq + 1);
2407                 con->in_base_pos = -front_len - middle_len - data_len -
2408                         sizeof_footer(con);
2409                 con->in_tag = CEPH_MSGR_TAG_READY;
2410                 return 1;
2411         } else if ((s64)seq - (s64)con->in_seq > 1) {
2412                 pr_err("read_partial_message bad seq %lld expected %lld\n",
2413                        seq, con->in_seq + 1);
2414                 con->error_msg = "bad message sequence # for incoming message";
2415                 return -EBADE;
2416         }
2417
2418         /* allocate message? */
2419         if (!con->in_msg) {
2420                 int skip = 0;
2421
2422                 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2423                      front_len, data_len);
2424                 ret = ceph_con_in_msg_alloc(con, &skip);
2425                 if (ret < 0)
2426                         return ret;
2427
2428                 BUG_ON(!con->in_msg ^ skip);
2429                 if (skip) {
2430                         /* skip this message */
2431                         dout("alloc_msg said skip message\n");
2432                         con->in_base_pos = -front_len - middle_len - data_len -
2433                                 sizeof_footer(con);
2434                         con->in_tag = CEPH_MSGR_TAG_READY;
2435                         con->in_seq++;
2436                         return 1;
2437                 }
2438
2439                 BUG_ON(!con->in_msg);
2440                 BUG_ON(con->in_msg->con != con);
2441                 m = con->in_msg;
2442                 m->front.iov_len = 0;    /* haven't read it yet */
2443                 if (m->middle)
2444                         m->middle->vec.iov_len = 0;
2445
2446                 /* prepare for data payload, if any */
2447
2448                 if (data_len)
2449                         prepare_message_data(con->in_msg, data_len);
2450         }
2451
2452         /* front */
2453         ret = read_partial_message_section(con, &m->front, front_len,
2454                                            &con->in_front_crc);
2455         if (ret <= 0)
2456                 return ret;
2457
2458         /* middle */
2459         if (m->middle) {
2460                 ret = read_partial_message_section(con, &m->middle->vec,
2461                                                    middle_len,
2462                                                    &con->in_middle_crc);
2463                 if (ret <= 0)
2464                         return ret;
2465         }
2466
2467         /* (page) data */
2468         if (data_len) {
2469                 ret = read_partial_msg_data(con);
2470                 if (ret <= 0)
2471                         return ret;
2472         }
2473
2474         /* footer */
2475         size = sizeof_footer(con);
2476         end += size;
2477         ret = read_partial(con, end, size, &m->footer);
2478         if (ret <= 0)
2479                 return ret;
2480
2481         if (!need_sign) {
2482                 m->footer.flags = m->old_footer.flags;
2483                 m->footer.sig = 0;
2484         }
2485
2486         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2487              m, front_len, m->footer.front_crc, middle_len,
2488              m->footer.middle_crc, data_len, m->footer.data_crc);
2489
2490         /* crc ok? */
2491         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2492                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2493                        m, con->in_front_crc, m->footer.front_crc);
2494                 return -EBADMSG;
2495         }
2496         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2497                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2498                        m, con->in_middle_crc, m->footer.middle_crc);
2499                 return -EBADMSG;
2500         }
2501         if (do_datacrc &&
2502             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2503             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2504                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2505                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2506                 return -EBADMSG;
2507         }
2508
2509         if (need_sign && con->ops->check_message_signature &&
2510             con->ops->check_message_signature(m)) {
2511                 pr_err("read_partial_message %p signature check failed\n", m);
2512                 return -EBADMSG;
2513         }
2514
2515         return 1; /* done! */
2516 }
2517
2518 /*
2519  * Process message.  This happens in the worker thread.  The callback should
2520  * be careful not to do anything that waits on other incoming messages or it
2521  * may deadlock.
2522  */
2523 static void process_message(struct ceph_connection *con)
2524 {
2525         struct ceph_msg *msg = con->in_msg;
2526
2527         BUG_ON(con->in_msg->con != con);
2528         con->in_msg = NULL;
2529
2530         /* if first message, set peer_name */
2531         if (con->peer_name.type == 0)
2532                 con->peer_name = msg->hdr.src;
2533
2534         con->in_seq++;
2535         mutex_unlock(&con->mutex);
2536
2537         dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2538              msg, le64_to_cpu(msg->hdr.seq),
2539              ENTITY_NAME(msg->hdr.src),
2540              le16_to_cpu(msg->hdr.type),
2541              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2542              le32_to_cpu(msg->hdr.front_len),
2543              le32_to_cpu(msg->hdr.data_len),
2544              con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2545         con->ops->dispatch(con, msg);
2546
2547         mutex_lock(&con->mutex);
2548 }
2549
2550 static int read_keepalive_ack(struct ceph_connection *con)
2551 {
2552         struct ceph_timespec ceph_ts;
2553         size_t size = sizeof(ceph_ts);
2554         int ret = read_partial(con, size, size, &ceph_ts);
2555         if (ret <= 0)
2556                 return ret;
2557         ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
2558         prepare_read_tag(con);
2559         return 1;
2560 }
2561
2562 /*
2563  * Write something to the socket.  Called in a worker thread when the
2564  * socket appears to be writeable and we have something ready to send.
2565  */
2566 static int try_write(struct ceph_connection *con)
2567 {
2568         int ret = 1;
2569
2570         dout("try_write start %p state %lu\n", con, con->state);
2571         if (con->state != CON_STATE_PREOPEN &&
2572             con->state != CON_STATE_CONNECTING &&
2573             con->state != CON_STATE_NEGOTIATING &&
2574             con->state != CON_STATE_OPEN)
2575                 return 0;
2576
2577         /* open the socket first? */
2578         if (con->state == CON_STATE_PREOPEN) {
2579                 BUG_ON(con->sock);
2580                 con->state = CON_STATE_CONNECTING;
2581
2582                 con_out_kvec_reset(con);
2583                 prepare_write_banner(con);
2584                 prepare_read_banner(con);
2585
2586                 BUG_ON(con->in_msg);
2587                 con->in_tag = CEPH_MSGR_TAG_READY;
2588                 dout("try_write initiating connect on %p new state %lu\n",
2589                      con, con->state);
2590                 ret = ceph_tcp_connect(con);
2591                 if (ret < 0) {
2592                         con->error_msg = "connect error";
2593                         goto out;
2594                 }
2595         }
2596
2597 more:
2598         dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2599         BUG_ON(!con->sock);
2600
2601         /* kvec data queued? */
2602         if (con->out_kvec_left) {
2603                 ret = write_partial_kvec(con);
2604                 if (ret <= 0)
2605                         goto out;
2606         }
2607         if (con->out_skip) {
2608                 ret = write_partial_skip(con);
2609                 if (ret <= 0)
2610                         goto out;
2611         }
2612
2613         /* msg pages? */
2614         if (con->out_msg) {
2615                 if (con->out_msg_done) {
2616                         ceph_msg_put(con->out_msg);
2617                         con->out_msg = NULL;   /* we're done with this one */
2618                         goto do_next;
2619                 }
2620
2621                 ret = write_partial_message_data(con);
2622                 if (ret == 1)
2623                         goto more;  /* we need to send the footer, too! */
2624                 if (ret == 0)
2625                         goto out;
2626                 if (ret < 0) {
2627                         dout("try_write write_partial_message_data err %d\n",
2628                              ret);
2629                         goto out;
2630                 }
2631         }
2632
2633 do_next:
2634         if (con->state == CON_STATE_OPEN) {
2635                 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2636                         prepare_write_keepalive(con);
2637                         goto more;
2638                 }
2639                 /* is anything else pending? */
2640                 if (!list_empty(&con->out_queue)) {
2641                         prepare_write_message(con);
2642                         goto more;
2643                 }
2644                 if (con->in_seq > con->in_seq_acked) {
2645                         prepare_write_ack(con);
2646                         goto more;
2647                 }
2648         }
2649
2650         /* Nothing to do! */
2651         con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2652         dout("try_write nothing else to write.\n");
2653         ret = 0;
2654 out:
2655         dout("try_write done on %p ret %d\n", con, ret);
2656         return ret;
2657 }
2658
2659 /*
2660  * Read what we can from the socket.
2661  */
2662 static int try_read(struct ceph_connection *con)
2663 {
2664         int ret = -1;
2665
2666 more:
2667         dout("try_read start on %p state %lu\n", con, con->state);
2668         if (con->state != CON_STATE_CONNECTING &&
2669             con->state != CON_STATE_NEGOTIATING &&
2670             con->state != CON_STATE_OPEN)
2671                 return 0;
2672
2673         BUG_ON(!con->sock);
2674
2675         dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2676              con->in_base_pos);
2677
2678         if (con->state == CON_STATE_CONNECTING) {
2679                 dout("try_read connecting\n");
2680                 ret = read_partial_banner(con);
2681                 if (ret <= 0)
2682                         goto out;
2683                 ret = process_banner(con);
2684                 if (ret < 0)
2685                         goto out;
2686
2687                 con->state = CON_STATE_NEGOTIATING;
2688
2689                 /*
2690                  * Received banner is good, exchange connection info.
2691                  * Do not reset out_kvec, as sending our banner raced
2692                  * with receiving peer banner after connect completed.
2693                  */
2694                 ret = prepare_write_connect(con);
2695                 if (ret < 0)
2696                         goto out;
2697                 prepare_read_connect(con);
2698
2699                 /* Send connection info before awaiting response */
2700                 goto out;
2701         }
2702
2703         if (con->state == CON_STATE_NEGOTIATING) {
2704                 dout("try_read negotiating\n");
2705                 ret = read_partial_connect(con);
2706                 if (ret <= 0)
2707                         goto out;
2708                 ret = process_connect(con);
2709                 if (ret < 0)
2710                         goto out;
2711                 goto more;
2712         }
2713
2714         WARN_ON(con->state != CON_STATE_OPEN);
2715
2716         if (con->in_base_pos < 0) {
2717                 /*
2718                  * skipping + discarding content.
2719                  */
2720                 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
2721                 if (ret <= 0)
2722                         goto out;
2723                 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
2724                 con->in_base_pos += ret;
2725                 if (con->in_base_pos)
2726                         goto more;
2727         }
2728         if (con->in_tag == CEPH_MSGR_TAG_READY) {
2729                 /*
2730                  * what's next?
2731                  */
2732                 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2733                 if (ret <= 0)
2734                         goto out;
2735                 dout("try_read got tag %d\n", (int)con->in_tag);
2736                 switch (con->in_tag) {
2737                 case CEPH_MSGR_TAG_MSG:
2738                         prepare_read_message(con);
2739                         break;
2740                 case CEPH_MSGR_TAG_ACK:
2741                         prepare_read_ack(con);
2742                         break;
2743                 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2744                         prepare_read_keepalive_ack(con);
2745                         break;
2746                 case CEPH_MSGR_TAG_CLOSE:
2747                         con_close_socket(con);
2748                         con->state = CON_STATE_CLOSED;
2749                         goto out;
2750                 default:
2751                         goto bad_tag;
2752                 }
2753         }
2754         if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2755                 ret = read_partial_message(con);
2756                 if (ret <= 0) {
2757                         switch (ret) {
2758                         case -EBADMSG:
2759                                 con->error_msg = "bad crc/signature";
2760                                 /* fall through */
2761                         case -EBADE:
2762                                 ret = -EIO;
2763                                 break;
2764                         case -EIO:
2765                                 con->error_msg = "io error";
2766                                 break;
2767                         }
2768                         goto out;
2769                 }
2770                 if (con->in_tag == CEPH_MSGR_TAG_READY)
2771                         goto more;
2772                 process_message(con);
2773                 if (con->state == CON_STATE_OPEN)
2774                         prepare_read_tag(con);
2775                 goto more;
2776         }
2777         if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2778             con->in_tag == CEPH_MSGR_TAG_SEQ) {
2779                 /*
2780                  * the final handshake seq exchange is semantically
2781                  * equivalent to an ACK
2782                  */
2783                 ret = read_partial_ack(con);
2784                 if (ret <= 0)
2785                         goto out;
2786                 process_ack(con);
2787                 goto more;
2788         }
2789         if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2790                 ret = read_keepalive_ack(con);
2791                 if (ret <= 0)
2792                         goto out;
2793                 goto more;
2794         }
2795
2796 out:
2797         dout("try_read done on %p ret %d\n", con, ret);
2798         return ret;
2799
2800 bad_tag:
2801         pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2802         con->error_msg = "protocol error, garbage tag";
2803         ret = -1;
2804         goto out;
2805 }
2806
2807
2808 /*
2809  * Atomically queue work on a connection after the specified delay.
2810  * Bump @con reference to avoid races with connection teardown.
2811  * Returns 0 if work was queued, or an error code otherwise.
2812  */
2813 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2814 {
2815         if (!con->ops->get(con)) {
2816                 dout("%s %p ref count 0\n", __func__, con);
2817                 return -ENOENT;
2818         }
2819
2820         if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2821                 dout("%s %p - already queued\n", __func__, con);
2822                 con->ops->put(con);
2823                 return -EBUSY;
2824         }
2825
2826         dout("%s %p %lu\n", __func__, con, delay);
2827         return 0;
2828 }
2829
2830 static void queue_con(struct ceph_connection *con)
2831 {
2832         (void) queue_con_delay(con, 0);
2833 }
2834
2835 static void cancel_con(struct ceph_connection *con)
2836 {
2837         if (cancel_delayed_work(&con->work)) {
2838                 dout("%s %p\n", __func__, con);
2839                 con->ops->put(con);
2840         }
2841 }
2842
2843 static bool con_sock_closed(struct ceph_connection *con)
2844 {
2845         if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2846                 return false;
2847
2848 #define CASE(x)                                                         \
2849         case CON_STATE_ ## x:                                           \
2850                 con->error_msg = "socket closed (con state " #x ")";    \
2851                 break;
2852
2853         switch (con->state) {
2854         CASE(CLOSED);
2855         CASE(PREOPEN);
2856         CASE(CONNECTING);
2857         CASE(NEGOTIATING);
2858         CASE(OPEN);
2859         CASE(STANDBY);
2860         default:
2861                 pr_warn("%s con %p unrecognized state %lu\n",
2862                         __func__, con, con->state);
2863                 con->error_msg = "unrecognized con state";
2864                 BUG();
2865                 break;
2866         }
2867 #undef CASE
2868
2869         return true;
2870 }
2871
2872 static bool con_backoff(struct ceph_connection *con)
2873 {
2874         int ret;
2875
2876         if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2877                 return false;
2878
2879         ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2880         if (ret) {
2881                 dout("%s: con %p FAILED to back off %lu\n", __func__,
2882                         con, con->delay);
2883                 BUG_ON(ret == -ENOENT);
2884                 con_flag_set(con, CON_FLAG_BACKOFF);
2885         }
2886
2887         return true;
2888 }
2889
2890 /* Finish fault handling; con->mutex must *not* be held here */
2891
2892 static void con_fault_finish(struct ceph_connection *con)
2893 {
2894         dout("%s %p\n", __func__, con);
2895
2896         /*
2897          * in case we faulted due to authentication, invalidate our
2898          * current tickets so that we can get new ones.
2899          */
2900         if (con->auth_retry) {
2901                 dout("auth_retry %d, invalidating\n", con->auth_retry);
2902                 if (con->ops->invalidate_authorizer)
2903                         con->ops->invalidate_authorizer(con);
2904                 con->auth_retry = 0;
2905         }
2906
2907         if (con->ops->fault)
2908                 con->ops->fault(con);
2909 }
2910
2911 /*
2912  * Do some work on a connection.  Drop a connection ref when we're done.
2913  */
2914 static void ceph_con_workfn(struct work_struct *work)
2915 {
2916         struct ceph_connection *con = container_of(work, struct ceph_connection,
2917                                                    work.work);
2918         bool fault;
2919
2920         mutex_lock(&con->mutex);
2921         while (true) {
2922                 int ret;
2923
2924                 if ((fault = con_sock_closed(con))) {
2925                         dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2926                         break;
2927                 }
2928                 if (con_backoff(con)) {
2929                         dout("%s: con %p BACKOFF\n", __func__, con);
2930                         break;
2931                 }
2932                 if (con->state == CON_STATE_STANDBY) {
2933                         dout("%s: con %p STANDBY\n", __func__, con);
2934                         break;
2935                 }
2936                 if (con->state == CON_STATE_CLOSED) {
2937                         dout("%s: con %p CLOSED\n", __func__, con);
2938                         BUG_ON(con->sock);
2939                         break;
2940                 }
2941                 if (con->state == CON_STATE_PREOPEN) {
2942                         dout("%s: con %p PREOPEN\n", __func__, con);
2943                         BUG_ON(con->sock);
2944                 }
2945
2946                 ret = try_read(con);
2947                 if (ret < 0) {
2948                         if (ret == -EAGAIN)
2949                                 continue;
2950                         if (!con->error_msg)
2951                                 con->error_msg = "socket error on read";
2952                         fault = true;
2953                         break;
2954                 }
2955
2956                 ret = try_write(con);
2957                 if (ret < 0) {
2958                         if (ret == -EAGAIN)
2959                                 continue;
2960                         if (!con->error_msg)
2961                                 con->error_msg = "socket error on write";
2962                         fault = true;
2963                 }
2964
2965                 break;  /* If we make it to here, we're done */
2966         }
2967         if (fault)
2968                 con_fault(con);
2969         mutex_unlock(&con->mutex);
2970
2971         if (fault)
2972                 con_fault_finish(con);
2973
2974         con->ops->put(con);
2975 }
2976
2977 /*
2978  * Generic error/fault handler.  A retry mechanism is used with
2979  * exponential backoff
2980  */
2981 static void con_fault(struct ceph_connection *con)
2982 {
2983         dout("fault %p state %lu to peer %s\n",
2984              con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2985
2986         pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2987                 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2988         con->error_msg = NULL;
2989
2990         WARN_ON(con->state != CON_STATE_CONNECTING &&
2991                con->state != CON_STATE_NEGOTIATING &&
2992                con->state != CON_STATE_OPEN);
2993
2994         con_close_socket(con);
2995
2996         if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2997                 dout("fault on LOSSYTX channel, marking CLOSED\n");
2998                 con->state = CON_STATE_CLOSED;
2999                 return;
3000         }
3001
3002         if (con->in_msg) {
3003                 BUG_ON(con->in_msg->con != con);
3004                 ceph_msg_put(con->in_msg);
3005                 con->in_msg = NULL;
3006         }
3007
3008         /* Requeue anything that hasn't been acked */
3009         list_splice_init(&con->out_sent, &con->out_queue);
3010
3011         /* If there are no messages queued or keepalive pending, place
3012          * the connection in a STANDBY state */
3013         if (list_empty(&con->out_queue) &&
3014             !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3015                 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3016                 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3017                 con->state = CON_STATE_STANDBY;
3018         } else {
3019                 /* retry after a delay. */
3020                 con->state = CON_STATE_PREOPEN;
3021                 if (con->delay == 0)
3022                         con->delay = BASE_DELAY_INTERVAL;
3023                 else if (con->delay < MAX_DELAY_INTERVAL)
3024                         con->delay *= 2;
3025                 con_flag_set(con, CON_FLAG_BACKOFF);
3026                 queue_con(con);
3027         }
3028 }
3029
3030
3031
3032 /*
3033  * initialize a new messenger instance
3034  */
3035 void ceph_messenger_init(struct ceph_messenger *msgr,
3036                          struct ceph_entity_addr *myaddr)
3037 {
3038         spin_lock_init(&msgr->global_seq_lock);
3039
3040         if (myaddr)
3041                 msgr->inst.addr = *myaddr;
3042
3043         /* select a random nonce */
3044         msgr->inst.addr.type = 0;
3045         get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3046         encode_my_addr(msgr);
3047
3048         atomic_set(&msgr->stopping, 0);
3049         write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3050
3051         dout("%s %p\n", __func__, msgr);
3052 }
3053 EXPORT_SYMBOL(ceph_messenger_init);
3054
3055 void ceph_messenger_fini(struct ceph_messenger *msgr)
3056 {
3057         put_net(read_pnet(&msgr->net));
3058 }
3059 EXPORT_SYMBOL(ceph_messenger_fini);
3060
3061 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3062 {
3063         if (msg->con)
3064                 msg->con->ops->put(msg->con);
3065
3066         msg->con = con ? con->ops->get(con) : NULL;
3067         BUG_ON(msg->con != con);
3068 }
3069
3070 static void clear_standby(struct ceph_connection *con)
3071 {
3072         /* come back from STANDBY? */
3073         if (con->state == CON_STATE_STANDBY) {
3074                 dout("clear_standby %p and ++connect_seq\n", con);
3075                 con->state = CON_STATE_PREOPEN;
3076                 con->connect_seq++;
3077                 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3078                 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3079         }
3080 }
3081
3082 /*
3083  * Queue up an outgoing message on the given connection.
3084  */
3085 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3086 {
3087         /* set src+dst */
3088         msg->hdr.src = con->msgr->inst.name;
3089         BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3090         msg->needs_out_seq = true;
3091
3092         mutex_lock(&con->mutex);
3093
3094         if (con->state == CON_STATE_CLOSED) {
3095                 dout("con_send %p closed, dropping %p\n", con, msg);
3096                 ceph_msg_put(msg);
3097                 mutex_unlock(&con->mutex);
3098                 return;
3099         }
3100
3101         msg_con_set(msg, con);
3102
3103         BUG_ON(!list_empty(&msg->list_head));
3104         list_add_tail(&msg->list_head, &con->out_queue);
3105         dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3106              ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3107              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3108              le32_to_cpu(msg->hdr.front_len),
3109              le32_to_cpu(msg->hdr.middle_len),
3110              le32_to_cpu(msg->hdr.data_len));
3111
3112         clear_standby(con);
3113         mutex_unlock(&con->mutex);
3114
3115         /* if there wasn't anything waiting to send before, queue
3116          * new work */
3117         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3118                 queue_con(con);
3119 }
3120 EXPORT_SYMBOL(ceph_con_send);
3121
3122 /*
3123  * Revoke a message that was previously queued for send
3124  */
3125 void ceph_msg_revoke(struct ceph_msg *msg)
3126 {
3127         struct ceph_connection *con = msg->con;
3128
3129         if (!con) {
3130                 dout("%s msg %p null con\n", __func__, msg);
3131                 return;         /* Message not in our possession */
3132         }
3133
3134         mutex_lock(&con->mutex);
3135         if (!list_empty(&msg->list_head)) {
3136                 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3137                 list_del_init(&msg->list_head);
3138                 msg->hdr.seq = 0;
3139
3140                 ceph_msg_put(msg);
3141         }
3142         if (con->out_msg == msg) {
3143                 BUG_ON(con->out_skip);
3144                 /* footer */
3145                 if (con->out_msg_done) {
3146                         con->out_skip += con_out_kvec_skip(con);
3147                 } else {
3148                         BUG_ON(!msg->data_length);
3149                         con->out_skip += sizeof_footer(con);
3150                 }
3151                 /* data, middle, front */
3152                 if (msg->data_length)
3153                         con->out_skip += msg->cursor.total_resid;
3154                 if (msg->middle)
3155                         con->out_skip += con_out_kvec_skip(con);
3156                 con->out_skip += con_out_kvec_skip(con);
3157
3158                 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3159                      __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3160                 msg->hdr.seq = 0;
3161                 con->out_msg = NULL;
3162                 ceph_msg_put(msg);
3163         }
3164
3165         mutex_unlock(&con->mutex);
3166 }
3167
3168 /*
3169  * Revoke a message that we may be reading data into
3170  */
3171 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3172 {
3173         struct ceph_connection *con = msg->con;
3174
3175         if (!con) {
3176                 dout("%s msg %p null con\n", __func__, msg);
3177                 return;         /* Message not in our possession */
3178         }
3179
3180         mutex_lock(&con->mutex);
3181         if (con->in_msg == msg) {
3182                 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3183                 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3184                 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3185
3186                 /* skip rest of message */
3187                 dout("%s %p msg %p revoked\n", __func__, con, msg);
3188                 con->in_base_pos = con->in_base_pos -
3189                                 sizeof(struct ceph_msg_header) -
3190                                 front_len -
3191                                 middle_len -
3192                                 data_len -
3193                                 sizeof(struct ceph_msg_footer);
3194                 ceph_msg_put(con->in_msg);
3195                 con->in_msg = NULL;
3196                 con->in_tag = CEPH_MSGR_TAG_READY;
3197                 con->in_seq++;
3198         } else {
3199                 dout("%s %p in_msg %p msg %p no-op\n",
3200                      __func__, con, con->in_msg, msg);
3201         }
3202         mutex_unlock(&con->mutex);
3203 }
3204
3205 /*
3206  * Queue a keepalive byte to ensure the tcp connection is alive.
3207  */
3208 void ceph_con_keepalive(struct ceph_connection *con)
3209 {
3210         dout("con_keepalive %p\n", con);
3211         mutex_lock(&con->mutex);
3212         clear_standby(con);
3213         con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
3214         mutex_unlock(&con->mutex);
3215
3216         if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3217                 queue_con(con);
3218 }
3219 EXPORT_SYMBOL(ceph_con_keepalive);
3220
3221 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3222                                unsigned long interval)
3223 {
3224         if (interval > 0 &&
3225             (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3226                 struct timespec64 now;
3227                 struct timespec64 ts;
3228                 ktime_get_real_ts64(&now);
3229                 jiffies_to_timespec64(interval, &ts);
3230                 ts = timespec64_add(con->last_keepalive_ack, ts);
3231                 return timespec64_compare(&now, &ts) >= 0;
3232         }
3233         return false;
3234 }
3235
3236 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
3237 {
3238         BUG_ON(msg->num_data_items >= msg->max_data_items);
3239         return &msg->data[msg->num_data_items++];
3240 }
3241
3242 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3243 {
3244         if (data->type == CEPH_MSG_DATA_PAGELIST)
3245                 ceph_pagelist_release(data->pagelist);
3246 }
3247
3248 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3249                 size_t length, size_t alignment)
3250 {
3251         struct ceph_msg_data *data;
3252
3253         BUG_ON(!pages);
3254         BUG_ON(!length);
3255
3256         data = ceph_msg_data_add(msg);
3257         data->type = CEPH_MSG_DATA_PAGES;
3258         data->pages = pages;
3259         data->length = length;
3260         data->alignment = alignment & ~PAGE_MASK;
3261
3262         msg->data_length += length;
3263 }
3264 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3265
3266 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3267                                 struct ceph_pagelist *pagelist)
3268 {
3269         struct ceph_msg_data *data;
3270
3271         BUG_ON(!pagelist);
3272         BUG_ON(!pagelist->length);
3273
3274         data = ceph_msg_data_add(msg);
3275         data->type = CEPH_MSG_DATA_PAGELIST;
3276         refcount_inc(&pagelist->refcnt);
3277         data->pagelist = pagelist;
3278
3279         msg->data_length += pagelist->length;
3280 }
3281 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3282
3283 #ifdef  CONFIG_BLOCK
3284 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3285                            u32 length)
3286 {
3287         struct ceph_msg_data *data;
3288
3289         data = ceph_msg_data_add(msg);
3290         data->type = CEPH_MSG_DATA_BIO;
3291         data->bio_pos = *bio_pos;
3292         data->bio_length = length;
3293
3294         msg->data_length += length;
3295 }
3296 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3297 #endif  /* CONFIG_BLOCK */
3298
3299 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3300                              struct ceph_bvec_iter *bvec_pos)
3301 {
3302         struct ceph_msg_data *data;
3303
3304         data = ceph_msg_data_add(msg);
3305         data->type = CEPH_MSG_DATA_BVECS;
3306         data->bvec_pos = *bvec_pos;
3307
3308         msg->data_length += bvec_pos->iter.bi_size;
3309 }
3310 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3311
3312 /*
3313  * construct a new message with given type, size
3314  * the new msg has a ref count of 1.
3315  */
3316 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
3317                                gfp_t flags, bool can_fail)
3318 {
3319         struct ceph_msg *m;
3320
3321         m = kmem_cache_zalloc(ceph_msg_cache, flags);
3322         if (m == NULL)
3323                 goto out;
3324
3325         m->hdr.type = cpu_to_le16(type);
3326         m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3327         m->hdr.front_len = cpu_to_le32(front_len);
3328
3329         INIT_LIST_HEAD(&m->list_head);
3330         kref_init(&m->kref);
3331
3332         /* front */
3333         if (front_len) {
3334                 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3335                 if (m->front.iov_base == NULL) {
3336                         dout("ceph_msg_new can't allocate %d bytes\n",
3337                              front_len);
3338                         goto out2;
3339                 }
3340         } else {
3341                 m->front.iov_base = NULL;
3342         }
3343         m->front_alloc_len = m->front.iov_len = front_len;
3344
3345         if (max_data_items) {
3346                 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
3347                                         flags);
3348                 if (!m->data)
3349                         goto out2;
3350
3351                 m->max_data_items = max_data_items;
3352         }
3353
3354         dout("ceph_msg_new %p front %d\n", m, front_len);
3355         return m;
3356
3357 out2:
3358         ceph_msg_put(m);
3359 out:
3360         if (!can_fail) {
3361                 pr_err("msg_new can't create type %d front %d\n", type,
3362                        front_len);
3363                 WARN_ON(1);
3364         } else {
3365                 dout("msg_new can't create type %d front %d\n", type,
3366                      front_len);
3367         }
3368         return NULL;
3369 }
3370 EXPORT_SYMBOL(ceph_msg_new2);
3371
3372 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3373                               bool can_fail)
3374 {
3375         return ceph_msg_new2(type, front_len, 0, flags, can_fail);
3376 }
3377 EXPORT_SYMBOL(ceph_msg_new);
3378
3379 /*
3380  * Allocate "middle" portion of a message, if it is needed and wasn't
3381  * allocated by alloc_msg.  This allows us to read a small fixed-size
3382  * per-type header in the front and then gracefully fail (i.e.,
3383  * propagate the error to the caller based on info in the front) when
3384  * the middle is too large.
3385  */
3386 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3387 {
3388         int type = le16_to_cpu(msg->hdr.type);
3389         int middle_len = le32_to_cpu(msg->hdr.middle_len);
3390
3391         dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3392              ceph_msg_type_name(type), middle_len);
3393         BUG_ON(!middle_len);
3394         BUG_ON(msg->middle);
3395
3396         msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3397         if (!msg->middle)
3398                 return -ENOMEM;
3399         return 0;
3400 }
3401
3402 /*
3403  * Allocate a message for receiving an incoming message on a
3404  * connection, and save the result in con->in_msg.  Uses the
3405  * connection's private alloc_msg op if available.
3406  *
3407  * Returns 0 on success, or a negative error code.
3408  *
3409  * On success, if we set *skip = 1:
3410  *  - the next message should be skipped and ignored.
3411  *  - con->in_msg == NULL
3412  * or if we set *skip = 0:
3413  *  - con->in_msg is non-null.
3414  * On error (ENOMEM, EAGAIN, ...),
3415  *  - con->in_msg == NULL
3416  */
3417 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3418 {
3419         struct ceph_msg_header *hdr = &con->in_hdr;
3420         int middle_len = le32_to_cpu(hdr->middle_len);
3421         struct ceph_msg *msg;
3422         int ret = 0;
3423
3424         BUG_ON(con->in_msg != NULL);
3425         BUG_ON(!con->ops->alloc_msg);
3426
3427         mutex_unlock(&con->mutex);
3428         msg = con->ops->alloc_msg(con, hdr, skip);
3429         mutex_lock(&con->mutex);
3430         if (con->state != CON_STATE_OPEN) {
3431                 if (msg)
3432                         ceph_msg_put(msg);
3433                 return -EAGAIN;
3434         }
3435         if (msg) {
3436                 BUG_ON(*skip);
3437                 msg_con_set(msg, con);
3438                 con->in_msg = msg;
3439         } else {
3440                 /*
3441                  * Null message pointer means either we should skip
3442                  * this message or we couldn't allocate memory.  The
3443                  * former is not an error.
3444                  */
3445                 if (*skip)
3446                         return 0;
3447
3448                 con->error_msg = "error allocating memory for incoming message";
3449                 return -ENOMEM;
3450         }
3451         memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3452
3453         if (middle_len && !con->in_msg->middle) {
3454                 ret = ceph_alloc_middle(con, con->in_msg);
3455                 if (ret < 0) {
3456                         ceph_msg_put(con->in_msg);
3457                         con->in_msg = NULL;
3458                 }
3459         }
3460
3461         return ret;
3462 }
3463
3464
3465 /*
3466  * Free a generically kmalloc'd message.
3467  */
3468 static void ceph_msg_free(struct ceph_msg *m)
3469 {
3470         dout("%s %p\n", __func__, m);
3471         kvfree(m->front.iov_base);
3472         kfree(m->data);
3473         kmem_cache_free(ceph_msg_cache, m);
3474 }
3475
3476 static void ceph_msg_release(struct kref *kref)
3477 {
3478         struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3479         int i;
3480
3481         dout("%s %p\n", __func__, m);
3482         WARN_ON(!list_empty(&m->list_head));
3483
3484         msg_con_set(m, NULL);
3485
3486         /* drop middle, data, if any */
3487         if (m->middle) {
3488                 ceph_buffer_put(m->middle);
3489                 m->middle = NULL;
3490         }
3491
3492         for (i = 0; i < m->num_data_items; i++)
3493                 ceph_msg_data_destroy(&m->data[i]);
3494
3495         if (m->pool)
3496                 ceph_msgpool_put(m->pool, m);
3497         else
3498                 ceph_msg_free(m);
3499 }
3500
3501 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3502 {
3503         dout("%s %p (was %d)\n", __func__, msg,
3504              kref_read(&msg->kref));
3505         kref_get(&msg->kref);
3506         return msg;
3507 }
3508 EXPORT_SYMBOL(ceph_msg_get);
3509
3510 void ceph_msg_put(struct ceph_msg *msg)
3511 {
3512         dout("%s %p (was %d)\n", __func__, msg,
3513              kref_read(&msg->kref));
3514         kref_put(&msg->kref, ceph_msg_release);
3515 }
3516 EXPORT_SYMBOL(ceph_msg_put);
3517
3518 void ceph_msg_dump(struct ceph_msg *msg)
3519 {
3520         pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3521                  msg->front_alloc_len, msg->data_length);
3522         print_hex_dump(KERN_DEBUG, "header: ",
3523                        DUMP_PREFIX_OFFSET, 16, 1,
3524                        &msg->hdr, sizeof(msg->hdr), true);
3525         print_hex_dump(KERN_DEBUG, " front: ",
3526                        DUMP_PREFIX_OFFSET, 16, 1,
3527                        msg->front.iov_base, msg->front.iov_len, true);
3528         if (msg->middle)
3529                 print_hex_dump(KERN_DEBUG, "middle: ",
3530                                DUMP_PREFIX_OFFSET, 16, 1,
3531                                msg->middle->vec.iov_base,
3532                                msg->middle->vec.iov_len, true);
3533         print_hex_dump(KERN_DEBUG, "footer: ",
3534                        DUMP_PREFIX_OFFSET, 16, 1,
3535                        &msg->footer, sizeof(msg->footer), true);
3536 }
3537 EXPORT_SYMBOL(ceph_msg_dump);