net/ceph/messenger.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/crc32c.h>
   4 #include <linux/ctype.h>
   5 #include <linux/highmem.h>
   6 #include <linux/inet.h>
   7 #include <linux/kthread.h>
   8 #include <linux/net.h>
   9 #include <linux/slab.h>
  10 #include <linux/socket.h>
  11 #include <linux/string.h>
  12 #include <linux/bio.h>
  13 #include <linux/blkdev.h>
  14 #include <linux/dns_resolver.h>
  15 #include <net/tcp.h>
  16
  17 #include <linux/ceph/libceph.h>
  18 #include <linux/ceph/messenger.h>
  19 #include <linux/ceph/decode.h>
  20 #include <linux/ceph/pagelist.h>
  21 #include <linux/export.h>
  22
  23 /*
  24  * Ceph uses the messenger to exchange ceph_msg messages with other
  25  * hosts in the system.  The messenger provides ordered and reliable
  26  * delivery.  We tolerate TCP disconnects by reconnecting (with
  27  * exponential backoff) in the case of a fault (disconnection, bad
  28  * crc, protocol error).  Acks allow sent messages to be discarded by
  29  * the sender.
  30  */
  31
  32 /* static tag bytes (protocol control messages) */
  33 static char tag_msg = CEPH_MSGR_TAG_MSG;
  34 static char tag_ack = CEPH_MSGR_TAG_ACK;
  35 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
  36
  37 #ifdef CONFIG_LOCKDEP
  38 static struct lock_class_key socket_class;
  39 #endif
  40
  41
  42 static void queue_con(struct ceph_connection *con);
  43 static void con_work(struct work_struct *);
  44 static void ceph_fault(struct ceph_connection *con);
  45
  46 /*
  47  * nicely render a sockaddr as a string.
  48  */
  49 #define MAX_ADDR_STR 20
  50 #define MAX_ADDR_STR_LEN 60
  51 static char addr_str[MAX_ADDR_STR][MAX_ADDR_STR_LEN];
  52 static DEFINE_SPINLOCK(addr_str_lock);
  53 static int last_addr_str;
  54
  55 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
  56 {
  57         int i;
  58         char *s;
  59         struct sockaddr_in *in4 = (void *)ss;
  60         struct sockaddr_in6 *in6 = (void *)ss;
  61
  62         spin_lock(&addr_str_lock);
  63         i = last_addr_str++;
  64         if (last_addr_str == MAX_ADDR_STR)
  65                 last_addr_str = 0;
  66         spin_unlock(&addr_str_lock);
  67         s = addr_str[i];
  68
  69         switch (ss->ss_family) {
  70         case AF_INET:
  71                 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%u", &in4->sin_addr,
  72                          (unsigned int)ntohs(in4->sin_port));
  73                 break;
  74
  75         case AF_INET6:
  76                 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%u", &in6->sin6_addr,
  77                          (unsigned int)ntohs(in6->sin6_port));
  78                 break;
  79
  80         default:
  81                 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %d)",
  82                          (int)ss->ss_family);
  83         }
  84
  85         return s;
  86 }
  87 EXPORT_SYMBOL(ceph_pr_addr);
  88
  89 static void encode_my_addr(struct ceph_messenger *msgr)
  90 {
  91         memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
  92         ceph_encode_addr(&msgr->my_enc_addr);
  93 }
  94
  95 /*
  96  * work queue for all reading and writing to/from the socket.
  97  */
  98 struct workqueue_struct *ceph_msgr_wq;
  99
 100 int ceph_msgr_init(void)
 101 {
 102         /*
 103          * The number of active work items is limited by the number of
 104          * connections, so leave @max_active at default.
 105          */
 106         ceph_msgr_wq = alloc_workqueue("ceph-msgr",
 107                                        WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
 108         if (!ceph_msgr_wq) {
 109                 pr_err("msgr_init failed to create workqueue\n");
 110                 return -ENOMEM;
 111         }
 112         return 0;
 113 }
 114 EXPORT_SYMBOL(ceph_msgr_init);
 115
 116 void ceph_msgr_exit(void)
 117 {
 118         destroy_workqueue(ceph_msgr_wq);
 119 }
 120 EXPORT_SYMBOL(ceph_msgr_exit);
 121
 122 void ceph_msgr_flush(void)
 123 {
 124         flush_workqueue(ceph_msgr_wq);
 125 }
 126 EXPORT_SYMBOL(ceph_msgr_flush);
 127
 128
 129 /*
 130  * socket callback functions
 131  */
 132
 133 /* data available on socket, or listen socket received a connect */
 134 static void ceph_data_ready(struct sock *sk, int count_unused)
 135 {
 136         struct ceph_connection *con =
 137                 (struct ceph_connection *)sk->sk_user_data;
 138         if (sk->sk_state != TCP_CLOSE_WAIT) {
 139                 dout("ceph_data_ready on %p state = %lu, queueing work\n",
 140                      con, con->state);
 141                 queue_con(con);
 142         }
 143 }
 144
 145 /* socket has buffer space for writing */
 146 static void ceph_write_space(struct sock *sk)
 147 {
 148         struct ceph_connection *con =
 149                 (struct ceph_connection *)sk->sk_user_data;
 150
 151         /* only queue to workqueue if there is data we want to write. */
 152         if (test_bit(WRITE_PENDING, &con->state)) {
 153                 dout("ceph_write_space %p queueing write work\n", con);
 154                 queue_con(con);
 155         } else {
 156                 dout("ceph_write_space %p nothing to write\n", con);
 157         }
 158
 159         /* since we have our own write_space, clear the SOCK_NOSPACE flag */
 160         clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 161 }
 162
 163 /* socket's state has changed */
 164 static void ceph_state_change(struct sock *sk)
 165 {
 166         struct ceph_connection *con =
 167                 (struct ceph_connection *)sk->sk_user_data;
 168
 169         dout("ceph_state_change %p state = %lu sk_state = %u\n",
 170              con, con->state, sk->sk_state);
 171
 172         if (test_bit(CLOSED, &con->state))
 173                 return;
 174
 175         switch (sk->sk_state) {
 176         case TCP_CLOSE:
 177                 dout("ceph_state_change TCP_CLOSE\n");
 178         case TCP_CLOSE_WAIT:
 179                 dout("ceph_state_change TCP_CLOSE_WAIT\n");
 180                 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
 181                         if (test_bit(CONNECTING, &con->state))
 182                                 con->error_msg = "connection failed";
 183                         else
 184                                 con->error_msg = "socket closed";
 185                         queue_con(con);
 186                 }
 187                 break;
 188         case TCP_ESTABLISHED:
 189                 dout("ceph_state_change TCP_ESTABLISHED\n");
 190                 queue_con(con);
 191                 break;
 192         }
 193 }
 194
 195 /*
 196  * set up socket callbacks
 197  */
 198 static void set_sock_callbacks(struct socket *sock,
 199                                struct ceph_connection *con)
 200 {
 201         struct sock *sk = sock->sk;
 202         sk->sk_user_data = (void *)con;
 203         sk->sk_data_ready = ceph_data_ready;
 204         sk->sk_write_space = ceph_write_space;
 205         sk->sk_state_change = ceph_state_change;
 206 }
 207
 208
 209 /*
 210  * socket helpers
 211  */
 212
 213 /*
 214  * initiate connection to a remote socket.
 215  */
 216 static struct socket *ceph_tcp_connect(struct ceph_connection *con)
 217 {
 218         struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
 219         struct socket *sock;
 220         int ret;
 221
 222         BUG_ON(con->sock);
 223         ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
 224                                IPPROTO_TCP, &sock);
 225         if (ret)
 226                 return ERR_PTR(ret);
 227         con->sock = sock;
 228         sock->sk->sk_allocation = GFP_NOFS;
 229
 230 #ifdef CONFIG_LOCKDEP
 231         lockdep_set_class(&sock->sk->sk_lock, &socket_class);
 232 #endif
 233
 234         set_sock_callbacks(sock, con);
 235
 236         dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
 237
 238         ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
 239                                  O_NONBLOCK);
 240         if (ret == -EINPROGRESS) {
 241                 dout("connect %s EINPROGRESS sk_state = %u\n",
 242                      ceph_pr_addr(&con->peer_addr.in_addr),
 243                      sock->sk->sk_state);
 244                 ret = 0;
 245         }
 246         if (ret < 0) {
 247                 pr_err("connect %s error %d\n",
 248                        ceph_pr_addr(&con->peer_addr.in_addr), ret);
 249                 sock_release(sock);
 250                 con->sock = NULL;
 251                 con->error_msg = "connect error";
 252         }
 253
 254         if (ret < 0)
 255                 return ERR_PTR(ret);
 256         return sock;
 257 }
 258
 259 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
 260 {
 261         struct kvec iov = {buf, len};
 262         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
 263         int r;
 264
 265         r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
 266         if (r == -EAGAIN)
 267                 r = 0;
 268         return r;
 269 }
 270
 271 /*
 272  * write something.  @more is true if caller will be sending more data
 273  * shortly.
 274  */
 275 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
 276                      size_t kvlen, size_t len, int more)
 277 {
 278         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
 279         int r;
 280
 281         if (more)
 282                 msg.msg_flags |= MSG_MORE;
 283         else
 284                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
 285
 286         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
 287         if (r == -EAGAIN)
 288                 r = 0;
 289         return r;
 290 }
 291
 292 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
 293                      int offset, size_t size, bool more)
 294 {
 295         int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
 296         int ret;
 297
 298         ret = kernel_sendpage(sock, page, offset, size, flags);
 299         if (ret == -EAGAIN)
 300                 ret = 0;
 301
 302         return ret;
 303 }
 304
 305 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
 306                      int offset, size_t size, bool more)
 307 {
 308         int ret;
 309         struct kvec iov;
 310
 311         /* sendpage cannot properly handle pages with page_count == 0,
 312          * we need to fallback to sendmsg if that's the case */
 313         if (page_count(page) >= 1)
 314                 return __ceph_tcp_sendpage(sock, page, offset, size, more);
 315
 316         iov.iov_base = kmap(page) + offset;
 317         iov.iov_len = size;
 318         ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
 319         kunmap(page);
 320
 321         return ret;
 322 }
 323
 324 /*
 325  * Shutdown/close the socket for the given connection.
 326  */
 327 static int con_close_socket(struct ceph_connection *con)
 328 {
 329         int rc;
 330
 331         dout("con_close_socket on %p sock %p\n", con, con->sock);
 332         if (!con->sock)
 333                 return 0;
 334         set_bit(SOCK_CLOSED, &con->state);
 335         rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
 336         sock_release(con->sock);
 337         con->sock = NULL;
 338         clear_bit(SOCK_CLOSED, &con->state);
 339         return rc;
 340 }
 341
 342 /*
 343  * Reset a connection.  Discard all incoming and outgoing messages
 344  * and clear *_seq state.
 345  */
 346 static void ceph_msg_remove(struct ceph_msg *msg)
 347 {
 348         list_del_init(&msg->list_head);
 349         ceph_msg_put(msg);
 350 }
 351 static void ceph_msg_remove_list(struct list_head *head)
 352 {
 353         while (!list_empty(head)) {
 354                 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
 355                                                         list_head);
 356                 ceph_msg_remove(msg);
 357         }
 358 }
 359
 360 static void reset_connection(struct ceph_connection *con)
 361 {
 362         /* reset connection, out_queue, msg_ and connect_seq */
 363         /* discard existing out_queue and msg_seq */
 364         ceph_msg_remove_list(&con->out_queue);
 365         ceph_msg_remove_list(&con->out_sent);
 366
 367         if (con->in_msg) {
 368                 ceph_msg_put(con->in_msg);
 369                 con->in_msg = NULL;
 370         }
 371
 372         con->connect_seq = 0;
 373         con->out_seq = 0;
 374         if (con->out_msg) {
 375                 ceph_msg_put(con->out_msg);
 376                 con->out_msg = NULL;
 377         }
 378         con->in_seq = 0;
 379         con->in_seq_acked = 0;
 380 }
 381
 382 /*
 383  * mark a peer down.  drop any open connections.
 384  */
 385 void ceph_con_close(struct ceph_connection *con)
 386 {
 387         dout("con_close %p peer %s\n", con,
 388              ceph_pr_addr(&con->peer_addr.in_addr));
 389         set_bit(CLOSED, &con->state);  /* in case there's queued work */
 390         clear_bit(STANDBY, &con->state);  /* avoid connect_seq bump */
 391         clear_bit(LOSSYTX, &con->state);  /* so we retry next connect */
 392         clear_bit(KEEPALIVE_PENDING, &con->state);
 393         clear_bit(WRITE_PENDING, &con->state);
 394         mutex_lock(&con->mutex);
 395         reset_connection(con);
 396         con->peer_global_seq = 0;
 397         cancel_delayed_work(&con->work);
 398         mutex_unlock(&con->mutex);
 399         queue_con(con);
 400 }
 401 EXPORT_SYMBOL(ceph_con_close);
 402
 403 /*
 404  * Reopen a closed connection, with a new peer address.
 405  */
 406 void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
 407 {
 408         dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
 409         set_bit(OPENING, &con->state);
 410         clear_bit(CLOSED, &con->state);
 411         memcpy(&con->peer_addr, addr, sizeof(*addr));
 412         con->delay = 0;      /* reset backoff memory */
 413         queue_con(con);
 414 }
 415 EXPORT_SYMBOL(ceph_con_open);
 416
 417 /*
 418  * return true if this connection ever successfully opened
 419  */
 420 bool ceph_con_opened(struct ceph_connection *con)
 421 {
 422         return con->connect_seq > 0;
 423 }
 424
 425 /*
 426  * generic get/put
 427  */
 428 struct ceph_connection *ceph_con_get(struct ceph_connection *con)
 429 {
 430         dout("con_get %p nref = %d -> %d\n", con,
 431              atomic_read(&con->nref), atomic_read(&con->nref) + 1);
 432         if (atomic_inc_not_zero(&con->nref))
 433                 return con;
 434         return NULL;
 435 }
 436
 437 void ceph_con_put(struct ceph_connection *con)
 438 {
 439         dout("con_put %p nref = %d -> %d\n", con,
 440              atomic_read(&con->nref), atomic_read(&con->nref) - 1);
 441         BUG_ON(atomic_read(&con->nref) == 0);
 442         if (atomic_dec_and_test(&con->nref)) {
 443                 BUG_ON(con->sock);
 444                 kfree(con);
 445         }
 446 }
 447
 448 /*
 449  * initialize a new connection.
 450  */
 451 void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
 452 {
 453         dout("con_init %p\n", con);
 454         memset(con, 0, sizeof(*con));
 455         atomic_set(&con->nref, 1);
 456         con->msgr = msgr;
 457         mutex_init(&con->mutex);
 458         INIT_LIST_HEAD(&con->out_queue);
 459         INIT_LIST_HEAD(&con->out_sent);
 460         INIT_DELAYED_WORK(&con->work, con_work);
 461 }
 462 EXPORT_SYMBOL(ceph_con_init);
 463
 464
 465 /*
 466  * We maintain a global counter to order connection attempts.  Get
 467  * a unique seq greater than @gt.
 468  */
 469 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
 470 {
 471         u32 ret;
 472
 473         spin_lock(&msgr->global_seq_lock);
 474         if (msgr->global_seq < gt)
 475                 msgr->global_seq = gt;
 476         ret = ++msgr->global_seq;
 477         spin_unlock(&msgr->global_seq_lock);
 478         return ret;
 479 }
 480
 481
 482 /*
 483  * Prepare footer for currently outgoing message, and finish things
 484  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
 485  */
 486 static void prepare_write_message_footer(struct ceph_connection *con, int v)
 487 {
 488         struct ceph_msg *m = con->out_msg;
 489
 490         dout("prepare_write_message_footer %p\n", con);
 491         con->out_kvec_is_msg = true;
 492         con->out_kvec[v].iov_base = &m->footer;
 493         con->out_kvec[v].iov_len = sizeof(m->footer);
 494         con->out_kvec_bytes += sizeof(m->footer);
 495         con->out_kvec_left++;
 496         con->out_more = m->more_to_follow;
 497         con->out_msg_done = true;
 498 }
 499
 500 /*
 501  * Prepare headers for the next outgoing message.
 502  */
 503 static void prepare_write_message(struct ceph_connection *con)
 504 {
 505         struct ceph_msg *m;
 506         int v = 0;
 507
 508         con->out_kvec_bytes = 0;
 509         con->out_kvec_is_msg = true;
 510         con->out_msg_done = false;
 511
 512         /* Sneak an ack in there first?  If we can get it into the same
 513          * TCP packet that's a good thing. */
 514         if (con->in_seq > con->in_seq_acked) {
 515                 con->in_seq_acked = con->in_seq;
 516                 con->out_kvec[v].iov_base = &tag_ack;
 517                 con->out_kvec[v++].iov_len = 1;
 518                 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
 519                 con->out_kvec[v].iov_base = &con->out_temp_ack;
 520                 con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack);
 521                 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
 522         }
 523
 524         m = list_first_entry(&con->out_queue,
 525                        struct ceph_msg, list_head);
 526         con->out_msg = m;
 527
 528         /* put message on sent list */
 529         ceph_msg_get(m);
 530         list_move_tail(&m->list_head, &con->out_sent);
 531
 532         /*
 533          * only assign outgoing seq # if we haven't sent this message
 534          * yet.  if it is requeued, resend with it's original seq.
 535          */
 536         if (m->needs_out_seq) {
 537                 m->hdr.seq = cpu_to_le64(++con->out_seq);
 538                 m->needs_out_seq = false;
 539         }
 540
 541         dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
 542              m, con->out_seq, le16_to_cpu(m->hdr.type),
 543              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
 544              le32_to_cpu(m->hdr.data_len),
 545              m->nr_pages);
 546         BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
 547
 548         /* tag + hdr + front + middle */
 549         con->out_kvec[v].iov_base = &tag_msg;
 550         con->out_kvec[v++].iov_len = 1;
 551         con->out_kvec[v].iov_base = &m->hdr;
 552         con->out_kvec[v++].iov_len = sizeof(m->hdr);
 553         con->out_kvec[v++] = m->front;
 554         if (m->middle)
 555                 con->out_kvec[v++] = m->middle->vec;
 556         con->out_kvec_left = v;
 557         con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len +
 558                 (m->middle ? m->middle->vec.iov_len : 0);
 559         con->out_kvec_cur = con->out_kvec;
 560
 561         /* fill in crc (except data pages), footer */
 562         con->out_msg->hdr.crc =
 563                 cpu_to_le32(crc32c(0, (void *)&m->hdr,
 564                                       sizeof(m->hdr) - sizeof(m->hdr.crc)));
 565         con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
 566         con->out_msg->footer.front_crc =
 567                 cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len));
 568         if (m->middle)
 569                 con->out_msg->footer.middle_crc =
 570                         cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
 571                                            m->middle->vec.iov_len));
 572         else
 573                 con->out_msg->footer.middle_crc = 0;
 574         con->out_msg->footer.data_crc = 0;
 575         dout("prepare_write_message front_crc %u data_crc %u\n",
 576              le32_to_cpu(con->out_msg->footer.front_crc),
 577              le32_to_cpu(con->out_msg->footer.middle_crc));
 578
 579         /* is there a data payload? */
 580         if (le32_to_cpu(m->hdr.data_len) > 0) {
 581                 /* initialize page iterator */
 582                 con->out_msg_pos.page = 0;
 583                 if (m->pages)
 584                         con->out_msg_pos.page_pos = m->page_alignment;
 585                 else
 586                         con->out_msg_pos.page_pos = 0;
 587                 con->out_msg_pos.data_pos = 0;
 588                 con->out_msg_pos.did_page_crc = 0;
 589                 con->out_more = 1;  /* data + footer will follow */
 590         } else {
 591                 /* no, queue up footer too and be done */
 592                 prepare_write_message_footer(con, v);
 593         }
 594
 595         set_bit(WRITE_PENDING, &con->state);
 596 }
 597
 598 /*
 599  * Prepare an ack.
 600  */
 601 static void prepare_write_ack(struct ceph_connection *con)
 602 {
 603         dout("prepare_write_ack %p %llu -> %llu\n", con,
 604              con->in_seq_acked, con->in_seq);
 605         con->in_seq_acked = con->in_seq;
 606
 607         con->out_kvec[0].iov_base = &tag_ack;
 608         con->out_kvec[0].iov_len = 1;
 609         con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
 610         con->out_kvec[1].iov_base = &con->out_temp_ack;
 611         con->out_kvec[1].iov_len = sizeof(con->out_temp_ack);
 612         con->out_kvec_left = 2;
 613         con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
 614         con->out_kvec_cur = con->out_kvec;
 615         con->out_more = 1;  /* more will follow.. eventually.. */
 616         set_bit(WRITE_PENDING, &con->state);
 617 }
 618
 619 /*
 620  * Prepare to write keepalive byte.
 621  */
 622 static void prepare_write_keepalive(struct ceph_connection *con)
 623 {
 624         dout("prepare_write_keepalive %p\n", con);
 625         con->out_kvec[0].iov_base = &tag_keepalive;
 626         con->out_kvec[0].iov_len = 1;
 627         con->out_kvec_left = 1;
 628         con->out_kvec_bytes = 1;
 629         con->out_kvec_cur = con->out_kvec;
 630         set_bit(WRITE_PENDING, &con->state);
 631 }
 632
 633 /*
 634  * Connection negotiation.
 635  */
 636
 637 static int prepare_connect_authorizer(struct ceph_connection *con)
 638 {
 639         void *auth_buf;
 640         int auth_len = 0;
 641         int auth_protocol = 0;
 642
 643         mutex_unlock(&con->mutex);
 644         if (con->ops->get_authorizer)
 645                 con->ops->get_authorizer(con, &auth_buf, &auth_len,
 646                                          &auth_protocol, &con->auth_reply_buf,
 647                                          &con->auth_reply_buf_len,
 648                                          con->auth_retry);
 649         mutex_lock(&con->mutex);
 650
 651         if (test_bit(CLOSED, &con->state) ||
 652             test_bit(OPENING, &con->state))
 653                 return -EAGAIN;
 654
 655         con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
 656         con->out_connect.authorizer_len = cpu_to_le32(auth_len);
 657
 658         if (auth_len) {
 659                 con->out_kvec[con->out_kvec_left].iov_base = auth_buf;
 660                 con->out_kvec[con->out_kvec_left].iov_len = auth_len;
 661                 con->out_kvec_left++;
 662                 con->out_kvec_bytes += auth_len;
 663         }
 664         return 0;
 665 }
 666
 667 /*
 668  * We connected to a peer and are saying hello.
 669  */
 670 static void prepare_write_banner(struct ceph_messenger *msgr,
 671                                  struct ceph_connection *con)
 672 {
 673         int len = strlen(CEPH_BANNER);
 674
 675         con->out_kvec[0].iov_base = CEPH_BANNER;
 676         con->out_kvec[0].iov_len = len;
 677         con->out_kvec[1].iov_base = &msgr->my_enc_addr;
 678         con->out_kvec[1].iov_len = sizeof(msgr->my_enc_addr);
 679         con->out_kvec_left = 2;
 680         con->out_kvec_bytes = len + sizeof(msgr->my_enc_addr);
 681         con->out_kvec_cur = con->out_kvec;
 682         con->out_more = 0;
 683         set_bit(WRITE_PENDING, &con->state);
 684 }
 685
 686 static int prepare_write_connect(struct ceph_messenger *msgr,
 687                                  struct ceph_connection *con,
 688                                  int after_banner)
 689 {
 690         unsigned global_seq = get_global_seq(con->msgr, 0);
 691         int proto;
 692
 693         switch (con->peer_name.type) {
 694         case CEPH_ENTITY_TYPE_MON:
 695                 proto = CEPH_MONC_PROTOCOL;
 696                 break;
 697         case CEPH_ENTITY_TYPE_OSD:
 698                 proto = CEPH_OSDC_PROTOCOL;
 699                 break;
 700         case CEPH_ENTITY_TYPE_MDS:
 701                 proto = CEPH_MDSC_PROTOCOL;
 702                 break;
 703         default:
 704                 BUG();
 705         }
 706
 707         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
 708              con->connect_seq, global_seq, proto);
 709
 710         con->out_connect.features = cpu_to_le64(msgr->supported_features);
 711         con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
 712         con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
 713         con->out_connect.global_seq = cpu_to_le32(global_seq);
 714         con->out_connect.protocol_version = cpu_to_le32(proto);
 715         con->out_connect.flags = 0;
 716
 717         if (!after_banner) {
 718                 con->out_kvec_left = 0;
 719                 con->out_kvec_bytes = 0;
 720         }
 721         con->out_kvec[con->out_kvec_left].iov_base = &con->out_connect;
 722         con->out_kvec[con->out_kvec_left].iov_len = sizeof(con->out_connect);
 723         con->out_kvec_left++;
 724         con->out_kvec_bytes += sizeof(con->out_connect);
 725         con->out_kvec_cur = con->out_kvec;
 726         con->out_more = 0;
 727         set_bit(WRITE_PENDING, &con->state);
 728
 729         return prepare_connect_authorizer(con);
 730 }
 731
 732
 733 /*
 734  * write as much of pending kvecs to the socket as we can.
 735  *  1 -> done
 736  *  0 -> socket full, but more to do
 737  * <0 -> error
 738  */
 739 static int write_partial_kvec(struct ceph_connection *con)
 740 {
 741         int ret;
 742
 743         dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
 744         while (con->out_kvec_bytes > 0) {
 745                 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
 746                                        con->out_kvec_left, con->out_kvec_bytes,
 747                                        con->out_more);
 748                 if (ret <= 0)
 749                         goto out;
 750                 con->out_kvec_bytes -= ret;
 751                 if (con->out_kvec_bytes == 0)
 752                         break;            /* done */
 753                 while (ret > 0) {
 754                         if (ret >= con->out_kvec_cur->iov_len) {
 755                                 ret -= con->out_kvec_cur->iov_len;
 756                                 con->out_kvec_cur++;
 757                                 con->out_kvec_left--;
 758                         } else {
 759                                 con->out_kvec_cur->iov_len -= ret;
 760                                 con->out_kvec_cur->iov_base += ret;
 761                                 ret = 0;
 762                                 break;
 763                         }
 764                 }
 765         }
 766         con->out_kvec_left = 0;
 767         con->out_kvec_is_msg = false;
 768         ret = 1;
 769 out:
 770         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
 771              con->out_kvec_bytes, con->out_kvec_left, ret);
 772         return ret;  /* done! */
 773 }
 774
 775 #ifdef CONFIG_BLOCK
 776 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
 777 {
 778         if (!bio) {
 779                 *iter = NULL;
 780                 *seg = 0;
 781                 return;
 782         }
 783         *iter = bio;
 784         *seg = bio->bi_idx;
 785 }
 786
 787 static void iter_bio_next(struct bio **bio_iter, int *seg)
 788 {
 789         if (*bio_iter == NULL)
 790                 return;
 791
 792         BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
 793
 794         (*seg)++;
 795         if (*seg == (*bio_iter)->bi_vcnt)
 796                 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
 797 }
 798 #endif
 799
 800 /*
 801  * Write as much message data payload as we can.  If we finish, queue
 802  * up the footer.
 803  *  1 -> done, footer is now queued in out_kvec[].
 804  *  0 -> socket full, but more to do
 805  * <0 -> error
 806  */
 807 static int write_partial_msg_pages(struct ceph_connection *con)
 808 {
 809         struct ceph_msg *msg = con->out_msg;
 810         unsigned data_len = le32_to_cpu(msg->hdr.data_len);
 811         size_t len;
 812         int crc = con->msgr->nocrc;
 813         int ret;
 814         int total_max_write;
 815         int in_trail = 0;
 816         size_t trail_len = (msg->trail ? msg->trail->length : 0);
 817
 818         dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
 819              con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
 820              con->out_msg_pos.page_pos);
 821
 822 #ifdef CONFIG_BLOCK
 823         if (msg->bio && !msg->bio_iter)
 824                 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
 825 #endif
 826
 827         while (data_len > con->out_msg_pos.data_pos) {
 828                 struct page *page = NULL;
 829                 void *kaddr = NULL;
 830                 int max_write = PAGE_SIZE;
 831                 int page_shift = 0;
 832
 833                 total_max_write = data_len - trail_len -
 834                         con->out_msg_pos.data_pos;
 835
 836                 /*
 837                  * if we are calculating the data crc (the default), we need
 838                  * to map the page.  if our pages[] has been revoked, use the
 839                  * zero page.
 840                  */
 841
 842                 /* have we reached the trail part of the data? */
 843                 if (con->out_msg_pos.data_pos >= data_len - trail_len) {
 844                         in_trail = 1;
 845
 846                         total_max_write = data_len - con->out_msg_pos.data_pos;
 847
 848                         page = list_first_entry(&msg->trail->head,
 849                                                 struct page, lru);
 850                         if (crc)
 851                                 kaddr = kmap(page);
 852                         max_write = PAGE_SIZE;
 853                 } else if (msg->pages) {
 854                         page = msg->pages[con->out_msg_pos.page];
 855                         if (crc)
 856                                 kaddr = kmap(page);
 857                 } else if (msg->pagelist) {
 858                         page = list_first_entry(&msg->pagelist->head,
 859                                                 struct page, lru);
 860                         if (crc)
 861                                 kaddr = kmap(page);
 862 #ifdef CONFIG_BLOCK
 863                 } else if (msg->bio) {
 864                         struct bio_vec *bv;
 865
 866                         bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
 867                         page = bv->bv_page;
 868                         page_shift = bv->bv_offset;
 869                         if (crc)
 870                                 kaddr = kmap(page) + page_shift;
 871                         max_write = bv->bv_len;
 872 #endif
 873                 } else {
 874                         page = con->msgr->zero_page;
 875                         if (crc)
 876                                 kaddr = page_address(con->msgr->zero_page);
 877                 }
 878                 len = min_t(int, max_write - con->out_msg_pos.page_pos,
 879                             total_max_write);
 880
 881                 if (crc && !con->out_msg_pos.did_page_crc) {
 882                         void *base = kaddr + con->out_msg_pos.page_pos;
 883                         u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
 884
 885                         BUG_ON(kaddr == NULL);
 886                         con->out_msg->footer.data_crc =
 887                                 cpu_to_le32(crc32c(tmpcrc, base, len));
 888                         con->out_msg_pos.did_page_crc = 1;
 889                 }
 890                 ret = ceph_tcp_sendpage(con->sock, page,
 891                                       con->out_msg_pos.page_pos + page_shift,
 892                                       len, 1);
 893
 894                 if (crc &&
 895                     (msg->pages || msg->pagelist || msg->bio || in_trail))
 896                         kunmap(page);
 897
 898                 if (ret <= 0)
 899                         goto out;
 900
 901                 con->out_msg_pos.data_pos += ret;
 902                 con->out_msg_pos.page_pos += ret;
 903                 if (ret == len) {
 904                         con->out_msg_pos.page_pos = 0;
 905                         con->out_msg_pos.page++;
 906                         con->out_msg_pos.did_page_crc = 0;
 907                         if (in_trail)
 908                                 list_move_tail(&page->lru,
 909                                                &msg->trail->head);
 910                         else if (msg->pagelist)
 911                                 list_move_tail(&page->lru,
 912                                                &msg->pagelist->head);
 913 #ifdef CONFIG_BLOCK
 914                         else if (msg->bio)
 915                                 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
 916 #endif
 917                 }
 918         }
 919
 920         dout("write_partial_msg_pages %p msg %p done\n", con, msg);
 921
 922         /* prepare and queue up footer, too */
 923         if (!crc)
 924                 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
 925         con->out_kvec_bytes = 0;
 926         con->out_kvec_left = 0;
 927         con->out_kvec_cur = con->out_kvec;
 928         prepare_write_message_footer(con, 0);
 929         ret = 1;
 930 out:
 931         return ret;
 932 }
 933
 934 /*
 935  * write some zeros
 936  */
 937 static int write_partial_skip(struct ceph_connection *con)
 938 {
 939         int ret;
 940
 941         while (con->out_skip > 0) {
 942                 struct kvec iov = {
 943                         .iov_base = page_address(con->msgr->zero_page),
 944                         .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
 945                 };
 946
 947                 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
 948                 if (ret <= 0)
 949                         goto out;
 950                 con->out_skip -= ret;
 951         }
 952         ret = 1;
 953 out:
 954         return ret;
 955 }
 956
 957 /*
 958  * Prepare to read connection handshake, or an ack.
 959  */
 960 static void prepare_read_banner(struct ceph_connection *con)
 961 {
 962         dout("prepare_read_banner %p\n", con);
 963         con->in_base_pos = 0;
 964 }
 965
 966 static void prepare_read_connect(struct ceph_connection *con)
 967 {
 968         dout("prepare_read_connect %p\n", con);
 969         con->in_base_pos = 0;
 970 }
 971
 972 static void prepare_read_ack(struct ceph_connection *con)
 973 {
 974         dout("prepare_read_ack %p\n", con);
 975         con->in_base_pos = 0;
 976 }
 977
 978 static void prepare_read_tag(struct ceph_connection *con)
 979 {
 980         dout("prepare_read_tag %p\n", con);
 981         con->in_base_pos = 0;
 982         con->in_tag = CEPH_MSGR_TAG_READY;
 983 }
 984
 985 /*
 986  * Prepare to read a message.
 987  */
 988 static int prepare_read_message(struct ceph_connection *con)
 989 {
 990         dout("prepare_read_message %p\n", con);
 991         BUG_ON(con->in_msg != NULL);
 992         con->in_base_pos = 0;
 993         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
 994         return 0;
 995 }
 996
 997
 998 static int read_partial(struct ceph_connection *con,
 999                         int *to, int size, void *object)
1000 {
1001         *to += size;
1002         while (con->in_base_pos < *to) {
1003                 int left = *to - con->in_base_pos;
1004                 int have = size - left;
1005                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1006                 if (ret <= 0)
1007                         return ret;
1008                 con->in_base_pos += ret;
1009         }
1010         return 1;
1011 }
1012
1013
1014 /*
1015  * Read all or part of the connect-side handshake on a new connection
1016  */
1017 static int read_partial_banner(struct ceph_connection *con)
1018 {
1019         int ret, to = 0;
1020
1021         dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1022
1023         /* peer's banner */
1024         ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
1025         if (ret <= 0)
1026                 goto out;
1027         ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
1028                            &con->actual_peer_addr);
1029         if (ret <= 0)
1030                 goto out;
1031         ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
1032                            &con->peer_addr_for_me);
1033         if (ret <= 0)
1034                 goto out;
1035 out:
1036         return ret;
1037 }
1038
1039 static int read_partial_connect(struct ceph_connection *con)
1040 {
1041         int ret, to = 0;
1042
1043         dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1044
1045         ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
1046         if (ret <= 0)
1047                 goto out;
1048         ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
1049                            con->auth_reply_buf);
1050         if (ret <= 0)
1051                 goto out;
1052
1053         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1054              con, (int)con->in_reply.tag,
1055              le32_to_cpu(con->in_reply.connect_seq),
1056              le32_to_cpu(con->in_reply.global_seq));
1057 out:
1058         return ret;
1059
1060 }
1061
1062 /*
1063  * Verify the hello banner looks okay.
1064  */
1065 static int verify_hello(struct ceph_connection *con)
1066 {
1067         if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1068                 pr_err("connect to %s got bad banner\n",
1069                        ceph_pr_addr(&con->peer_addr.in_addr));
1070                 con->error_msg = "protocol error, bad banner";
1071                 return -1;
1072         }
1073         return 0;
1074 }
1075
1076 static bool addr_is_blank(struct sockaddr_storage *ss)
1077 {
1078         switch (ss->ss_family) {
1079         case AF_INET:
1080                 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1081         case AF_INET6:
1082                 return
1083                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1084                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1085                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1086                      ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1087         }
1088         return false;
1089 }
1090
1091 static int addr_port(struct sockaddr_storage *ss)
1092 {
1093         switch (ss->ss_family) {
1094         case AF_INET:
1095                 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1096         case AF_INET6:
1097                 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1098         }
1099         return 0;
1100 }
1101
1102 static void addr_set_port(struct sockaddr_storage *ss, int p)
1103 {
1104         switch (ss->ss_family) {
1105         case AF_INET:
1106                 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1107                 break;
1108         case AF_INET6:
1109                 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1110                 break;
1111         }
1112 }
1113
1114 /*
1115  * Unlike other *_pton function semantics, zero indicates success.
1116  */
1117 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1118                 char delim, const char **ipend)
1119 {
1120         struct sockaddr_in *in4 = (void *)ss;
1121         struct sockaddr_in6 *in6 = (void *)ss;
1122
1123         memset(ss, 0, sizeof(*ss));
1124
1125         if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1126                 ss->ss_family = AF_INET;
1127                 return 0;
1128         }
1129
1130         if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1131                 ss->ss_family = AF_INET6;
1132                 return 0;
1133         }
1134
1135         return -EINVAL;
1136 }
1137
1138 /*
1139  * Extract hostname string and resolve using kernel DNS facility.
1140  */
1141 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1142 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1143                 struct sockaddr_storage *ss, char delim, const char **ipend)
1144 {
1145         const char *end, *delim_p;
1146         char *colon_p, *ip_addr = NULL;
1147         int ip_len, ret;
1148
1149         /*
1150          * The end of the hostname occurs immediately preceding the delimiter or
1151          * the port marker (':') where the delimiter takes precedence.
1152          */
1153         delim_p = memchr(name, delim, namelen);
1154         colon_p = memchr(name, ':', namelen);
1155
1156         if (delim_p && colon_p)
1157                 end = delim_p < colon_p ? delim_p : colon_p;
1158         else if (!delim_p && colon_p)
1159                 end = colon_p;
1160         else {
1161                 end = delim_p;
1162                 if (!end) /* case: hostname:/ */
1163                         end = name + namelen;
1164         }
1165
1166         if (end <= name)
1167                 return -EINVAL;
1168
1169         /* do dns_resolve upcall */
1170         ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1171         if (ip_len > 0)
1172                 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1173         else
1174                 ret = -ESRCH;
1175
1176         kfree(ip_addr);
1177
1178         *ipend = end;
1179
1180         pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1181                         ret, ret ? "failed" : ceph_pr_addr(ss));
1182
1183         return ret;
1184 }
1185 #else
1186 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1187                 struct sockaddr_storage *ss, char delim, const char **ipend)
1188 {
1189         return -EINVAL;
1190 }
1191 #endif
1192
1193 /*
1194  * Parse a server name (IP or hostname). If a valid IP address is not found
1195  * then try to extract a hostname to resolve using userspace DNS upcall.
1196  */
1197 static int ceph_parse_server_name(const char *name, size_t namelen,
1198                         struct sockaddr_storage *ss, char delim, const char **ipend)
1199 {
1200         int ret;
1201
1202         ret = ceph_pton(name, namelen, ss, delim, ipend);
1203         if (ret)
1204                 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1205
1206         return ret;
1207 }
1208
1209 /*
1210  * Parse an ip[:port] list into an addr array.  Use the default
1211  * monitor port if a port isn't specified.
1212  */
1213 int ceph_parse_ips(const char *c, const char *end,
1214                    struct ceph_entity_addr *addr,
1215                    int max_count, int *count)
1216 {
1217         int i, ret = -EINVAL;
1218         const char *p = c;
1219
1220         dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1221         for (i = 0; i < max_count; i++) {
1222                 const char *ipend;
1223                 struct sockaddr_storage *ss = &addr[i].in_addr;
1224                 int port;
1225                 char delim = ',';
1226
1227                 if (*p == '[') {
1228                         delim = ']';
1229                         p++;
1230                 }
1231
1232                 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1233                 if (ret)
1234                         goto bad;
1235                 ret = -EINVAL;
1236
1237                 p = ipend;
1238
1239                 if (delim == ']') {
1240                         if (*p != ']') {
1241                                 dout("missing matching ']'\n");
1242                                 goto bad;
1243                         }
1244                         p++;
1245                 }
1246
1247                 /* port? */
1248                 if (p < end && *p == ':') {
1249                         port = 0;
1250                         p++;
1251                         while (p < end && *p >= '0' && *p <= '9') {
1252                                 port = (port * 10) + (*p - '0');
1253                                 p++;
1254                         }
1255                         if (port > 65535 || port == 0)
1256                                 goto bad;
1257                 } else {
1258                         port = CEPH_MON_PORT;
1259                 }
1260
1261                 addr_set_port(ss, port);
1262
1263                 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1264
1265                 if (p == end)
1266                         break;
1267                 if (*p != ',')
1268                         goto bad;
1269                 p++;
1270         }
1271
1272         if (p != end)
1273                 goto bad;
1274
1275         if (count)
1276                 *count = i + 1;
1277         return 0;
1278
1279 bad:
1280         pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1281         return ret;
1282 }
1283 EXPORT_SYMBOL(ceph_parse_ips);
1284
1285 static int process_banner(struct ceph_connection *con)
1286 {
1287         dout("process_banner on %p\n", con);
1288
1289         if (verify_hello(con) < 0)
1290                 return -1;
1291
1292         ceph_decode_addr(&con->actual_peer_addr);
1293         ceph_decode_addr(&con->peer_addr_for_me);
1294
1295         /*
1296          * Make sure the other end is who we wanted.  note that the other
1297          * end may not yet know their ip address, so if it's 0.0.0.0, give
1298          * them the benefit of the doubt.
1299          */
1300         if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1301                    sizeof(con->peer_addr)) != 0 &&
1302             !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1303               con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1304                 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1305                            ceph_pr_addr(&con->peer_addr.in_addr),
1306                            (int)le32_to_cpu(con->peer_addr.nonce),
1307                            ceph_pr_addr(&con->actual_peer_addr.in_addr),
1308                            (int)le32_to_cpu(con->actual_peer_addr.nonce));
1309                 con->error_msg = "wrong peer at address";
1310                 return -1;
1311         }
1312
1313         /*
1314          * did we learn our address?
1315          */
1316         if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1317                 int port = addr_port(&con->msgr->inst.addr.in_addr);
1318
1319                 memcpy(&con->msgr->inst.addr.in_addr,
1320                        &con->peer_addr_for_me.in_addr,
1321                        sizeof(con->peer_addr_for_me.in_addr));
1322                 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1323                 encode_my_addr(con->msgr);
1324                 dout("process_banner learned my addr is %s\n",
1325                      ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1326         }
1327
1328         set_bit(NEGOTIATING, &con->state);
1329         prepare_read_connect(con);
1330         return 0;
1331 }
1332
1333 static void fail_protocol(struct ceph_connection *con)
1334 {
1335         reset_connection(con);
1336         set_bit(CLOSED, &con->state);  /* in case there's queued work */
1337
1338         mutex_unlock(&con->mutex);
1339         if (con->ops->bad_proto)
1340                 con->ops->bad_proto(con);
1341         mutex_lock(&con->mutex);
1342 }
1343
1344 static int process_connect(struct ceph_connection *con)
1345 {
1346         u64 sup_feat = con->msgr->supported_features;
1347         u64 req_feat = con->msgr->required_features;
1348         u64 server_feat = le64_to_cpu(con->in_reply.features);
1349         int ret;
1350
1351         dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1352
1353         if (con->auth_reply_buf) {
1354                 /*
1355                  * Any connection that defines ->get_authorizer()
1356                  * should also define ->verify_authorizer_reply().
1357                  * See get_connect_authorizer().
1358                  */
1359                 ret = con->ops->verify_authorizer_reply(con, 0);
1360                 if (ret < 0) {
1361                         con->error_msg = "bad authorize reply";
1362                         return ret;
1363                 }
1364         }
1365
1366         switch (con->in_reply.tag) {
1367         case CEPH_MSGR_TAG_FEATURES:
1368                 pr_err("%s%lld %s feature set mismatch,"
1369                        " my %llx < server's %llx, missing %llx\n",
1370                        ENTITY_NAME(con->peer_name),
1371                        ceph_pr_addr(&con->peer_addr.in_addr),
1372                        sup_feat, server_feat, server_feat & ~sup_feat);
1373                 con->error_msg = "missing required protocol features";
1374                 fail_protocol(con);
1375                 return -1;
1376
1377         case CEPH_MSGR_TAG_BADPROTOVER:
1378                 pr_err("%s%lld %s protocol version mismatch,"
1379                        " my %d != server's %d\n",
1380                        ENTITY_NAME(con->peer_name),
1381                        ceph_pr_addr(&con->peer_addr.in_addr),
1382                        le32_to_cpu(con->out_connect.protocol_version),
1383                        le32_to_cpu(con->in_reply.protocol_version));
1384                 con->error_msg = "protocol version mismatch";
1385                 fail_protocol(con);
1386                 return -1;
1387
1388         case CEPH_MSGR_TAG_BADAUTHORIZER:
1389                 con->auth_retry++;
1390                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1391                      con->auth_retry);
1392                 if (con->auth_retry == 2) {
1393                         con->error_msg = "connect authorization failure";
1394                         return -1;
1395                 }
1396                 con->auth_retry = 1;
1397                 ret = prepare_write_connect(con->msgr, con, 0);
1398                 if (ret < 0)
1399                         return ret;
1400                 prepare_read_connect(con);
1401                 break;
1402
1403         case CEPH_MSGR_TAG_RESETSESSION:
1404                 /*
1405                  * If we connected with a large connect_seq but the peer
1406                  * has no record of a session with us (no connection, or
1407                  * connect_seq == 0), they will send RESETSESION to indicate
1408                  * that they must have reset their session, and may have
1409                  * dropped messages.
1410                  */
1411                 dout("process_connect got RESET peer seq %u\n",
1412                      le32_to_cpu(con->in_connect.connect_seq));
1413                 pr_err("%s%lld %s connection reset\n",
1414                        ENTITY_NAME(con->peer_name),
1415                        ceph_pr_addr(&con->peer_addr.in_addr));
1416                 reset_connection(con);
1417                 prepare_write_connect(con->msgr, con, 0);
1418                 prepare_read_connect(con);
1419
1420                 /* Tell ceph about it. */
1421                 mutex_unlock(&con->mutex);
1422                 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1423                 if (con->ops->peer_reset)
1424                         con->ops->peer_reset(con);
1425                 mutex_lock(&con->mutex);
1426                 if (test_bit(CLOSED, &con->state) ||
1427                     test_bit(OPENING, &con->state))
1428                         return -EAGAIN;
1429                 break;
1430
1431         case CEPH_MSGR_TAG_RETRY_SESSION:
1432                 /*
1433                  * If we sent a smaller connect_seq than the peer has, try
1434                  * again with a larger value.
1435                  */
1436                 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1437                      le32_to_cpu(con->out_connect.connect_seq),
1438                      le32_to_cpu(con->in_connect.connect_seq));
1439                 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
1440                 prepare_write_connect(con->msgr, con, 0);
1441                 prepare_read_connect(con);
1442                 break;
1443
1444         case CEPH_MSGR_TAG_RETRY_GLOBAL:
1445                 /*
1446                  * If we sent a smaller global_seq than the peer has, try
1447                  * again with a larger value.
1448                  */
1449                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1450                      con->peer_global_seq,
1451                      le32_to_cpu(con->in_connect.global_seq));
1452                 get_global_seq(con->msgr,
1453                                le32_to_cpu(con->in_connect.global_seq));
1454                 prepare_write_connect(con->msgr, con, 0);
1455                 prepare_read_connect(con);
1456                 break;
1457
1458         case CEPH_MSGR_TAG_READY:
1459                 if (req_feat & ~server_feat) {
1460                         pr_err("%s%lld %s protocol feature mismatch,"
1461                                " my required %llx > server's %llx, need %llx\n",
1462                                ENTITY_NAME(con->peer_name),
1463                                ceph_pr_addr(&con->peer_addr.in_addr),
1464                                req_feat, server_feat, req_feat & ~server_feat);
1465                         con->error_msg = "missing required protocol features";
1466                         fail_protocol(con);
1467                         return -1;
1468                 }
1469                 clear_bit(CONNECTING, &con->state);
1470                 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1471                 con->connect_seq++;
1472                 con->peer_features = server_feat;
1473                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1474                      con->peer_global_seq,
1475                      le32_to_cpu(con->in_reply.connect_seq),
1476                      con->connect_seq);
1477                 WARN_ON(con->connect_seq !=
1478                         le32_to_cpu(con->in_reply.connect_seq));
1479
1480                 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1481                         set_bit(LOSSYTX, &con->state);
1482
1483                 prepare_read_tag(con);
1484                 break;
1485
1486         case CEPH_MSGR_TAG_WAIT:
1487                 /*
1488                  * If there is a connection race (we are opening
1489                  * connections to each other), one of us may just have
1490                  * to WAIT.  This shouldn't happen if we are the
1491                  * client.
1492                  */
1493                 pr_err("process_connect got WAIT as client\n");
1494                 con->error_msg = "protocol error, got WAIT as client";
1495                 return -1;
1496
1497         default:
1498                 pr_err("connect protocol error, will retry\n");
1499                 con->error_msg = "protocol error, garbage tag during connect";
1500                 return -1;
1501         }
1502         return 0;
1503 }
1504
1505
1506 /*
1507  * read (part of) an ack
1508  */
1509 static int read_partial_ack(struct ceph_connection *con)
1510 {
1511         int to = 0;
1512
1513         return read_partial(con, &to, sizeof(con->in_temp_ack),
1514                             &con->in_temp_ack);
1515 }
1516
1517
1518 /*
1519  * We can finally discard anything that's been acked.
1520  */
1521 static void process_ack(struct ceph_connection *con)
1522 {
1523         struct ceph_msg *m;
1524         u64 ack = le64_to_cpu(con->in_temp_ack);
1525         u64 seq;
1526
1527         while (!list_empty(&con->out_sent)) {
1528                 m = list_first_entry(&con->out_sent, struct ceph_msg,
1529                                      list_head);
1530                 seq = le64_to_cpu(m->hdr.seq);
1531                 if (seq > ack)
1532                         break;
1533                 dout("got ack for seq %llu type %d at %p\n", seq,
1534                      le16_to_cpu(m->hdr.type), m);
1535                 m->ack_stamp = jiffies;
1536                 ceph_msg_remove(m);
1537         }
1538         prepare_read_tag(con);
1539 }
1540
1541
1542
1543
1544 static int read_partial_message_section(struct ceph_connection *con,
1545                                         struct kvec *section,
1546                                         unsigned int sec_len, u32 *crc)
1547 {
1548         int ret, left;
1549
1550         BUG_ON(!section);
1551
1552         while (section->iov_len < sec_len) {
1553                 BUG_ON(section->iov_base == NULL);
1554                 left = sec_len - section->iov_len;
1555                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1556                                        section->iov_len, left);
1557                 if (ret <= 0)
1558                         return ret;
1559                 section->iov_len += ret;
1560                 if (section->iov_len == sec_len)
1561                         *crc = crc32c(0, section->iov_base,
1562                                       section->iov_len);
1563         }
1564
1565         return 1;
1566 }
1567
1568 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1569                                 struct ceph_msg_header *hdr,
1570                                 int *skip);
1571
1572
1573 static int read_partial_message_pages(struct ceph_connection *con,
1574                                       struct page **pages,
1575                                       unsigned data_len, int datacrc)
1576 {
1577         void *p;
1578         int ret;
1579         int left;
1580
1581         left = min((int)(data_len - con->in_msg_pos.data_pos),
1582                    (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1583         /* (page) data */
1584         BUG_ON(pages == NULL);
1585         p = kmap(pages[con->in_msg_pos.page]);
1586         ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1587                                left);
1588         if (ret > 0 && datacrc)
1589                 con->in_data_crc =
1590                         crc32c(con->in_data_crc,
1591                                   p + con->in_msg_pos.page_pos, ret);
1592         kunmap(pages[con->in_msg_pos.page]);
1593         if (ret <= 0)
1594                 return ret;
1595         con->in_msg_pos.data_pos += ret;
1596         con->in_msg_pos.page_pos += ret;
1597         if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1598                 con->in_msg_pos.page_pos = 0;
1599                 con->in_msg_pos.page++;
1600         }
1601
1602         return ret;
1603 }
1604
1605 #ifdef CONFIG_BLOCK
1606 static int read_partial_message_bio(struct ceph_connection *con,
1607                                     struct bio **bio_iter, int *bio_seg,
1608                                     unsigned data_len, int datacrc)
1609 {
1610         struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1611         void *p;
1612         int ret, left;
1613
1614         if (IS_ERR(bv))
1615                 return PTR_ERR(bv);
1616
1617         left = min((int)(data_len - con->in_msg_pos.data_pos),
1618                    (int)(bv->bv_len - con->in_msg_pos.page_pos));
1619
1620         p = kmap(bv->bv_page) + bv->bv_offset;
1621
1622         ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1623                                left);
1624         if (ret > 0 && datacrc)
1625                 con->in_data_crc =
1626                         crc32c(con->in_data_crc,
1627                                   p + con->in_msg_pos.page_pos, ret);
1628         kunmap(bv->bv_page);
1629         if (ret <= 0)
1630                 return ret;
1631         con->in_msg_pos.data_pos += ret;
1632         con->in_msg_pos.page_pos += ret;
1633         if (con->in_msg_pos.page_pos == bv->bv_len) {
1634                 con->in_msg_pos.page_pos = 0;
1635                 iter_bio_next(bio_iter, bio_seg);
1636         }
1637
1638         return ret;
1639 }
1640 #endif
1641
1642 /*
1643  * read (part of) a message.
1644  */
1645 static int read_partial_message(struct ceph_connection *con)
1646 {
1647         struct ceph_msg *m = con->in_msg;
1648         int ret;
1649         int to, left;
1650         unsigned front_len, middle_len, data_len;
1651         int datacrc = con->msgr->nocrc;
1652         int skip;
1653         u64 seq;
1654
1655         dout("read_partial_message con %p msg %p\n", con, m);
1656
1657         /* header */
1658         while (con->in_base_pos < sizeof(con->in_hdr)) {
1659                 left = sizeof(con->in_hdr) - con->in_base_pos;
1660                 ret = ceph_tcp_recvmsg(con->sock,
1661                                        (char *)&con->in_hdr + con->in_base_pos,
1662                                        left);
1663                 if (ret <= 0)
1664                         return ret;
1665                 con->in_base_pos += ret;
1666                 if (con->in_base_pos == sizeof(con->in_hdr)) {
1667                         u32 crc = crc32c(0, (void *)&con->in_hdr,
1668                                  sizeof(con->in_hdr) - sizeof(con->in_hdr.crc));
1669                         if (crc != le32_to_cpu(con->in_hdr.crc)) {
1670                                 pr_err("read_partial_message bad hdr "
1671                                        " crc %u != expected %u\n",
1672                                        crc, con->in_hdr.crc);
1673                                 return -EBADMSG;
1674                         }
1675                 }
1676         }
1677         front_len = le32_to_cpu(con->in_hdr.front_len);
1678         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1679                 return -EIO;
1680         middle_len = le32_to_cpu(con->in_hdr.middle_len);
1681         if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1682                 return -EIO;
1683         data_len = le32_to_cpu(con->in_hdr.data_len);
1684         if (data_len > CEPH_MSG_MAX_DATA_LEN)
1685                 return -EIO;
1686
1687         /* verify seq# */
1688         seq = le64_to_cpu(con->in_hdr.seq);
1689         if ((s64)seq - (s64)con->in_seq < 1) {
1690                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1691                         ENTITY_NAME(con->peer_name),
1692                         ceph_pr_addr(&con->peer_addr.in_addr),
1693                         seq, con->in_seq + 1);
1694                 con->in_base_pos = -front_len - middle_len - data_len -
1695                         sizeof(m->footer);
1696                 con->in_tag = CEPH_MSGR_TAG_READY;
1697                 return 0;
1698         } else if ((s64)seq - (s64)con->in_seq > 1) {
1699                 pr_err("read_partial_message bad seq %lld expected %lld\n",
1700                        seq, con->in_seq + 1);
1701                 con->error_msg = "bad message sequence # for incoming message";
1702                 return -EBADMSG;
1703         }
1704
1705         /* allocate message? */
1706         if (!con->in_msg) {
1707                 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1708                      con->in_hdr.front_len, con->in_hdr.data_len);
1709                 skip = 0;
1710                 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1711                 if (skip) {
1712                         /* skip this message */
1713                         dout("alloc_msg said skip message\n");
1714                         BUG_ON(con->in_msg);
1715                         con->in_base_pos = -front_len - middle_len - data_len -
1716                                 sizeof(m->footer);
1717                         con->in_tag = CEPH_MSGR_TAG_READY;
1718                         con->in_seq++;
1719                         return 0;
1720                 }
1721                 if (!con->in_msg) {
1722                         con->error_msg =
1723                                 "error allocating memory for incoming message";
1724                         return -ENOMEM;
1725                 }
1726                 m = con->in_msg;
1727                 m->front.iov_len = 0;    /* haven't read it yet */
1728                 if (m->middle)
1729                         m->middle->vec.iov_len = 0;
1730
1731                 con->in_msg_pos.page = 0;
1732                 if (m->pages)
1733                         con->in_msg_pos.page_pos = m->page_alignment;
1734                 else
1735                         con->in_msg_pos.page_pos = 0;
1736                 con->in_msg_pos.data_pos = 0;
1737         }
1738
1739         /* front */
1740         ret = read_partial_message_section(con, &m->front, front_len,
1741                                            &con->in_front_crc);
1742         if (ret <= 0)
1743                 return ret;
1744
1745         /* middle */
1746         if (m->middle) {
1747                 ret = read_partial_message_section(con, &m->middle->vec,
1748                                                    middle_len,
1749                                                    &con->in_middle_crc);
1750                 if (ret <= 0)
1751                         return ret;
1752         }
1753 #ifdef CONFIG_BLOCK
1754         if (m->bio && !m->bio_iter)
1755                 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1756 #endif
1757
1758         /* (page) data */
1759         while (con->in_msg_pos.data_pos < data_len) {
1760                 if (m->pages) {
1761                         ret = read_partial_message_pages(con, m->pages,
1762                                                  data_len, datacrc);
1763                         if (ret <= 0)
1764                                 return ret;
1765 #ifdef CONFIG_BLOCK
1766                 } else if (m->bio) {
1767
1768                         ret = read_partial_message_bio(con,
1769                                                  &m->bio_iter, &m->bio_seg,
1770                                                  data_len, datacrc);
1771                         if (ret <= 0)
1772                                 return ret;
1773 #endif
1774                 } else {
1775                         BUG_ON(1);
1776                 }
1777         }
1778
1779         /* footer */
1780         to = sizeof(m->hdr) + sizeof(m->footer);
1781         while (con->in_base_pos < to) {
1782                 left = to - con->in_base_pos;
1783                 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1784                                        (con->in_base_pos - sizeof(m->hdr)),
1785                                        left);
1786                 if (ret <= 0)
1787                         return ret;
1788                 con->in_base_pos += ret;
1789         }
1790         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1791              m, front_len, m->footer.front_crc, middle_len,
1792              m->footer.middle_crc, data_len, m->footer.data_crc);
1793
1794         /* crc ok? */
1795         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1796                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1797                        m, con->in_front_crc, m->footer.front_crc);
1798                 return -EBADMSG;
1799         }
1800         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1801                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1802                        m, con->in_middle_crc, m->footer.middle_crc);
1803                 return -EBADMSG;
1804         }
1805         if (datacrc &&
1806             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1807             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1808                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1809                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1810                 return -EBADMSG;
1811         }
1812
1813         return 1; /* done! */
1814 }
1815
1816 /*
1817  * Process message.  This happens in the worker thread.  The callback should
1818  * be careful not to do anything that waits on other incoming messages or it
1819  * may deadlock.
1820  */
1821 static void process_message(struct ceph_connection *con)
1822 {
1823         struct ceph_msg *msg;
1824
1825         msg = con->in_msg;
1826         con->in_msg = NULL;
1827
1828         /* if first message, set peer_name */
1829         if (con->peer_name.type == 0)
1830                 con->peer_name = msg->hdr.src;
1831
1832         con->in_seq++;
1833         mutex_unlock(&con->mutex);
1834
1835         dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1836              msg, le64_to_cpu(msg->hdr.seq),
1837              ENTITY_NAME(msg->hdr.src),
1838              le16_to_cpu(msg->hdr.type),
1839              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1840              le32_to_cpu(msg->hdr.front_len),
1841              le32_to_cpu(msg->hdr.data_len),
1842              con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1843         con->ops->dispatch(con, msg);
1844
1845         mutex_lock(&con->mutex);
1846         prepare_read_tag(con);
1847 }
1848
1849
1850 /*
1851  * Write something to the socket.  Called in a worker thread when the
1852  * socket appears to be writeable and we have something ready to send.
1853  */
1854 static int try_write(struct ceph_connection *con)
1855 {
1856         struct ceph_messenger *msgr = con->msgr;
1857         int ret = 1;
1858
1859         dout("try_write start %p state %lu nref %d\n", con, con->state,
1860              atomic_read(&con->nref));
1861
1862 more:
1863         dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1864
1865         /* open the socket first? */
1866         if (con->sock == NULL) {
1867                 prepare_write_banner(msgr, con);
1868                 prepare_write_connect(msgr, con, 1);
1869                 prepare_read_banner(con);
1870                 set_bit(CONNECTING, &con->state);
1871                 clear_bit(NEGOTIATING, &con->state);
1872
1873                 BUG_ON(con->in_msg);
1874                 con->in_tag = CEPH_MSGR_TAG_READY;
1875                 dout("try_write initiating connect on %p new state %lu\n",
1876                      con, con->state);
1877                 con->sock = ceph_tcp_connect(con);
1878                 if (IS_ERR(con->sock)) {
1879                         con->sock = NULL;
1880                         con->error_msg = "connect error";
1881                         ret = -1;
1882                         goto out;
1883                 }
1884         }
1885
1886 more_kvec:
1887         /* kvec data queued? */
1888         if (con->out_skip) {
1889                 ret = write_partial_skip(con);
1890                 if (ret <= 0)
1891                         goto out;
1892         }
1893         if (con->out_kvec_left) {
1894                 ret = write_partial_kvec(con);
1895                 if (ret <= 0)
1896                         goto out;
1897         }
1898
1899         /* msg pages? */
1900         if (con->out_msg) {
1901                 if (con->out_msg_done) {
1902                         ceph_msg_put(con->out_msg);
1903                         con->out_msg = NULL;   /* we're done with this one */
1904                         goto do_next;
1905                 }
1906
1907                 ret = write_partial_msg_pages(con);
1908                 if (ret == 1)
1909                         goto more_kvec;  /* we need to send the footer, too! */
1910                 if (ret == 0)
1911                         goto out;
1912                 if (ret < 0) {
1913                         dout("try_write write_partial_msg_pages err %d\n",
1914                              ret);
1915                         goto out;
1916                 }
1917         }
1918
1919 do_next:
1920         if (!test_bit(CONNECTING, &con->state)) {
1921                 /* is anything else pending? */
1922                 if (!list_empty(&con->out_queue)) {
1923                         prepare_write_message(con);
1924                         goto more;
1925                 }
1926                 if (con->in_seq > con->in_seq_acked) {
1927                         prepare_write_ack(con);
1928                         goto more;
1929                 }
1930                 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1931                         prepare_write_keepalive(con);
1932                         goto more;
1933                 }
1934         }
1935
1936         /* Nothing to do! */
1937         clear_bit(WRITE_PENDING, &con->state);
1938         dout("try_write nothing else to write.\n");
1939         ret = 0;
1940 out:
1941         dout("try_write done on %p ret %d\n", con, ret);
1942         return ret;
1943 }
1944
1945
1946
1947 /*
1948  * Read what we can from the socket.
1949  */
1950 static int try_read(struct ceph_connection *con)
1951 {
1952         int ret = -1;
1953
1954         if (!con->sock)
1955                 return 0;
1956
1957         if (test_bit(STANDBY, &con->state))
1958                 return 0;
1959
1960         dout("try_read start on %p\n", con);
1961
1962 more:
1963         dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1964              con->in_base_pos);
1965
1966         /*
1967          * process_connect and process_message drop and re-take
1968          * con->mutex.  make sure we handle a racing close or reopen.
1969          */
1970         if (test_bit(CLOSED, &con->state) ||
1971             test_bit(OPENING, &con->state)) {
1972                 ret = -EAGAIN;
1973                 goto out;
1974         }
1975
1976         if (test_bit(CONNECTING, &con->state)) {
1977                 if (!test_bit(NEGOTIATING, &con->state)) {
1978                         dout("try_read connecting\n");
1979                         ret = read_partial_banner(con);
1980                         if (ret <= 0)
1981                                 goto out;
1982                         ret = process_banner(con);
1983                         if (ret < 0)
1984                                 goto out;
1985                 }
1986                 ret = read_partial_connect(con);
1987                 if (ret <= 0)
1988                         goto out;
1989                 ret = process_connect(con);
1990                 if (ret < 0)
1991                         goto out;
1992                 goto more;
1993         }
1994
1995         if (con->in_base_pos < 0) {
1996                 /*
1997                  * skipping + discarding content.
1998                  *
1999                  * FIXME: there must be a better way to do this!
2000                  */
2001                 static char buf[1024];
2002                 int skip = min(1024, -con->in_base_pos);
2003                 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2004                 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2005                 if (ret <= 0)
2006                         goto out;
2007                 con->in_base_pos += ret;
2008                 if (con->in_base_pos)
2009                         goto more;
2010         }
2011         if (con->in_tag == CEPH_MSGR_TAG_READY) {
2012                 /*
2013                  * what's next?
2014                  */
2015                 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2016                 if (ret <= 0)
2017                         goto out;
2018                 dout("try_read got tag %d\n", (int)con->in_tag);
2019                 switch (con->in_tag) {
2020                 case CEPH_MSGR_TAG_MSG:
2021                         prepare_read_message(con);
2022                         break;
2023                 case CEPH_MSGR_TAG_ACK:
2024                         prepare_read_ack(con);
2025                         break;
2026                 case CEPH_MSGR_TAG_CLOSE:
2027                         set_bit(CLOSED, &con->state);   /* fixme */
2028                         goto out;
2029                 default:
2030                         goto bad_tag;
2031                 }
2032         }
2033         if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2034                 ret = read_partial_message(con);
2035                 if (ret <= 0) {
2036                         switch (ret) {
2037                         case -EBADMSG:
2038                                 con->error_msg = "bad crc";
2039                                 ret = -EIO;
2040                                 break;
2041                         case -EIO:
2042                                 con->error_msg = "io error";
2043                                 break;
2044                         }
2045                         goto out;
2046                 }
2047                 if (con->in_tag == CEPH_MSGR_TAG_READY)
2048                         goto more;
2049                 process_message(con);
2050                 goto more;
2051         }
2052         if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2053                 ret = read_partial_ack(con);
2054                 if (ret <= 0)
2055                         goto out;
2056                 process_ack(con);
2057                 goto more;
2058         }
2059
2060 out:
2061         dout("try_read done on %p ret %d\n", con, ret);
2062         return ret;
2063
2064 bad_tag:
2065         pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2066         con->error_msg = "protocol error, garbage tag";
2067         ret = -1;
2068         goto out;
2069 }
2070
2071
2072 /*
2073  * Atomically queue work on a connection.  Bump @con reference to
2074  * avoid races with connection teardown.
2075  */
2076 static void queue_con(struct ceph_connection *con)
2077 {
2078         if (test_bit(DEAD, &con->state)) {
2079                 dout("queue_con %p ignoring: DEAD\n",
2080                      con);
2081                 return;
2082         }
2083
2084         if (!con->ops->get(con)) {
2085                 dout("queue_con %p ref count 0\n", con);
2086                 return;
2087         }
2088
2089         if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
2090                 dout("queue_con %p - already queued\n", con);
2091                 con->ops->put(con);
2092         } else {
2093                 dout("queue_con %p\n", con);
2094         }
2095 }
2096
2097 /*
2098  * Do some work on a connection.  Drop a connection ref when we're done.
2099  */
2100 static void con_work(struct work_struct *work)
2101 {
2102         struct ceph_connection *con = container_of(work, struct ceph_connection,
2103                                                    work.work);
2104         int ret;
2105
2106         mutex_lock(&con->mutex);
2107 restart:
2108         if (test_and_clear_bit(BACKOFF, &con->state)) {
2109                 dout("con_work %p backing off\n", con);
2110                 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2111                                        round_jiffies_relative(con->delay))) {
2112                         dout("con_work %p backoff %lu\n", con, con->delay);
2113                         mutex_unlock(&con->mutex);
2114                         return;
2115                 } else {
2116                         con->ops->put(con);
2117                         dout("con_work %p FAILED to back off %lu\n", con,
2118                              con->delay);
2119                 }
2120         }
2121
2122         if (test_bit(STANDBY, &con->state)) {
2123                 dout("con_work %p STANDBY\n", con);
2124                 goto done;
2125         }
2126         if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
2127                 dout("con_work CLOSED\n");
2128                 con_close_socket(con);
2129                 goto done;
2130         }
2131         if (test_and_clear_bit(OPENING, &con->state)) {
2132                 /* reopen w/ new peer */
2133                 dout("con_work OPENING\n");
2134                 con_close_socket(con);
2135         }
2136
2137         if (test_and_clear_bit(SOCK_CLOSED, &con->state))
2138                 goto fault;
2139
2140         ret = try_read(con);
2141         if (ret == -EAGAIN)
2142                 goto restart;
2143         if (ret < 0)
2144                 goto fault;
2145
2146         ret = try_write(con);
2147         if (ret == -EAGAIN)
2148                 goto restart;
2149         if (ret < 0)
2150                 goto fault;
2151
2152 done:
2153         mutex_unlock(&con->mutex);
2154 done_unlocked:
2155         con->ops->put(con);
2156         return;
2157
2158 fault:
2159         mutex_unlock(&con->mutex);
2160         ceph_fault(con);     /* error/fault path */
2161         goto done_unlocked;
2162 }
2163
2164
2165 /*
2166  * Generic error/fault handler.  A retry mechanism is used with
2167  * exponential backoff
2168  */
2169 static void ceph_fault(struct ceph_connection *con)
2170 {
2171         pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2172                ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2173         dout("fault %p state %lu to peer %s\n",
2174              con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2175
2176         if (test_bit(LOSSYTX, &con->state)) {
2177                 dout("fault on LOSSYTX channel\n");
2178                 goto out;
2179         }
2180
2181         mutex_lock(&con->mutex);
2182         if (test_bit(CLOSED, &con->state))
2183                 goto out_unlock;
2184
2185         con_close_socket(con);
2186
2187         if (con->in_msg) {
2188                 ceph_msg_put(con->in_msg);
2189                 con->in_msg = NULL;
2190         }
2191
2192         /* Requeue anything that hasn't been acked */
2193         list_splice_init(&con->out_sent, &con->out_queue);
2194
2195         /* If there are no messages queued or keepalive pending, place
2196          * the connection in a STANDBY state */
2197         if (list_empty(&con->out_queue) &&
2198             !test_bit(KEEPALIVE_PENDING, &con->state)) {
2199                 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2200                 clear_bit(WRITE_PENDING, &con->state);
2201                 set_bit(STANDBY, &con->state);
2202         } else {
2203                 /* retry after a delay. */
2204                 if (con->delay == 0)
2205                         con->delay = BASE_DELAY_INTERVAL;
2206                 else if (con->delay < MAX_DELAY_INTERVAL)
2207                         con->delay *= 2;
2208                 con->ops->get(con);
2209                 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2210                                        round_jiffies_relative(con->delay))) {
2211                         dout("fault queued %p delay %lu\n", con, con->delay);
2212                 } else {
2213                         con->ops->put(con);
2214                         dout("fault failed to queue %p delay %lu, backoff\n",
2215                              con, con->delay);
2216                         /*
2217                          * In many cases we see a socket state change
2218                          * while con_work is running and end up
2219                          * queuing (non-delayed) work, such that we
2220                          * can't backoff with a delay.  Set a flag so
2221                          * that when con_work restarts we schedule the
2222                          * delay then.
2223                          */
2224                         set_bit(BACKOFF, &con->state);
2225                 }
2226         }
2227
2228 out_unlock:
2229         mutex_unlock(&con->mutex);
2230 out:
2231         /*
2232          * in case we faulted due to authentication, invalidate our
2233          * current tickets so that we can get new ones.
2234          */
2235         if (con->auth_retry && con->ops->invalidate_authorizer) {
2236                 dout("calling invalidate_authorizer()\n");
2237                 con->ops->invalidate_authorizer(con);
2238         }
2239
2240         if (con->ops->fault)
2241                 con->ops->fault(con);
2242 }
2243
2244
2245
2246 /*
2247  * create a new messenger instance
2248  */
2249 struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr,
2250                                              u32 supported_features,
2251                                              u32 required_features)
2252 {
2253         struct ceph_messenger *msgr;
2254
2255         msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
2256         if (msgr == NULL)
2257                 return ERR_PTR(-ENOMEM);
2258
2259         msgr->supported_features = supported_features;
2260         msgr->required_features = required_features;
2261
2262         spin_lock_init(&msgr->global_seq_lock);
2263
2264         /* the zero page is needed if a request is "canceled" while the message
2265          * is being written over the socket */
2266         msgr->zero_page = __page_cache_alloc(GFP_KERNEL | __GFP_ZERO);
2267         if (!msgr->zero_page) {
2268                 kfree(msgr);
2269                 return ERR_PTR(-ENOMEM);
2270         }
2271         kmap(msgr->zero_page);
2272
2273         if (myaddr)
2274                 msgr->inst.addr = *myaddr;
2275
2276         /* select a random nonce */
2277         msgr->inst.addr.type = 0;
2278         get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2279         encode_my_addr(msgr);
2280
2281         dout("messenger_create %p\n", msgr);
2282         return msgr;
2283 }
2284 EXPORT_SYMBOL(ceph_messenger_create);
2285
2286 void ceph_messenger_destroy(struct ceph_messenger *msgr)
2287 {
2288         dout("destroy %p\n", msgr);
2289         kunmap(msgr->zero_page);
2290         __free_page(msgr->zero_page);
2291         kfree(msgr);
2292         dout("destroyed messenger %p\n", msgr);
2293 }
2294 EXPORT_SYMBOL(ceph_messenger_destroy);
2295
2296 static void clear_standby(struct ceph_connection *con)
2297 {
2298         /* come back from STANDBY? */
2299         if (test_and_clear_bit(STANDBY, &con->state)) {
2300                 mutex_lock(&con->mutex);
2301                 dout("clear_standby %p and ++connect_seq\n", con);
2302                 con->connect_seq++;
2303                 WARN_ON(test_bit(WRITE_PENDING, &con->state));
2304                 WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state));
2305                 mutex_unlock(&con->mutex);
2306         }
2307 }
2308
2309 /*
2310  * Queue up an outgoing message on the given connection.
2311  */
2312 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2313 {
2314         if (test_bit(CLOSED, &con->state)) {
2315                 dout("con_send %p closed, dropping %p\n", con, msg);
2316                 ceph_msg_put(msg);
2317                 return;
2318         }
2319
2320         /* set src+dst */
2321         msg->hdr.src = con->msgr->inst.name;
2322
2323         BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2324
2325         msg->needs_out_seq = true;
2326
2327         /* queue */
2328         mutex_lock(&con->mutex);
2329         BUG_ON(!list_empty(&msg->list_head));
2330         list_add_tail(&msg->list_head, &con->out_queue);
2331         dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2332              ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2333              ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2334              le32_to_cpu(msg->hdr.front_len),
2335              le32_to_cpu(msg->hdr.middle_len),
2336              le32_to_cpu(msg->hdr.data_len));
2337         mutex_unlock(&con->mutex);
2338
2339         /* if there wasn't anything waiting to send before, queue
2340          * new work */
2341         clear_standby(con);
2342         if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2343                 queue_con(con);
2344 }
2345 EXPORT_SYMBOL(ceph_con_send);
2346
2347 /*
2348  * Revoke a message that was previously queued for send
2349  */
2350 void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
2351 {
2352         mutex_lock(&con->mutex);
2353         if (!list_empty(&msg->list_head)) {
2354                 dout("con_revoke %p msg %p - was on queue\n", con, msg);
2355                 list_del_init(&msg->list_head);
2356                 ceph_msg_put(msg);
2357                 msg->hdr.seq = 0;
2358         }
2359         if (con->out_msg == msg) {
2360                 dout("con_revoke %p msg %p - was sending\n", con, msg);
2361                 con->out_msg = NULL;
2362                 if (con->out_kvec_is_msg) {
2363                         con->out_skip = con->out_kvec_bytes;
2364                         con->out_kvec_is_msg = false;
2365                 }
2366                 ceph_msg_put(msg);
2367                 msg->hdr.seq = 0;
2368         }
2369         mutex_unlock(&con->mutex);
2370 }
2371
2372 /*
2373  * Revoke a message that we may be reading data into
2374  */
2375 void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
2376 {
2377         mutex_lock(&con->mutex);
2378         if (con->in_msg && con->in_msg == msg) {
2379                 unsigned front_len = le32_to_cpu(con->in_hdr.front_len);
2380                 unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len);
2381                 unsigned data_len = le32_to_cpu(con->in_hdr.data_len);
2382
2383                 /* skip rest of message */
2384                 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
2385                         con->in_base_pos = con->in_base_pos -
2386                                 sizeof(struct ceph_msg_header) -
2387                                 front_len -
2388                                 middle_len -
2389                                 data_len -
2390                                 sizeof(struct ceph_msg_footer);
2391                 ceph_msg_put(con->in_msg);
2392                 con->in_msg = NULL;
2393                 con->in_tag = CEPH_MSGR_TAG_READY;
2394                 con->in_seq++;
2395         } else {
2396                 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2397                      con, con->in_msg, msg);
2398         }
2399         mutex_unlock(&con->mutex);
2400 }
2401
2402 /*
2403  * Queue a keepalive byte to ensure the tcp connection is alive.
2404  */
2405 void ceph_con_keepalive(struct ceph_connection *con)
2406 {
2407         dout("con_keepalive %p\n", con);
2408         clear_standby(con);
2409         if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2410             test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2411                 queue_con(con);
2412 }
2413 EXPORT_SYMBOL(ceph_con_keepalive);
2414
2415
2416 /*
2417  * construct a new message with given type, size
2418  * the new msg has a ref count of 1.
2419  */
2420 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2421                               bool can_fail)
2422 {
2423         struct ceph_msg *m;
2424
2425         m = kmalloc(sizeof(*m), flags);
2426         if (m == NULL)
2427                 goto out;
2428         kref_init(&m->kref);
2429         INIT_LIST_HEAD(&m->list_head);
2430
2431         m->hdr.tid = 0;
2432         m->hdr.type = cpu_to_le16(type);
2433         m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2434         m->hdr.version = 0;
2435         m->hdr.front_len = cpu_to_le32(front_len);
2436         m->hdr.middle_len = 0;
2437         m->hdr.data_len = 0;
2438         m->hdr.data_off = 0;
2439         m->hdr.reserved = 0;
2440         m->footer.front_crc = 0;
2441         m->footer.middle_crc = 0;
2442         m->footer.data_crc = 0;
2443         m->footer.flags = 0;
2444         m->front_alloc_len = front_len;
2445         m->front_is_vmalloc = false;
2446         m->more_to_follow = false;
2447         m->ack_stamp = 0;
2448         m->pool = NULL;
2449
2450         /* middle */
2451         m->middle = NULL;
2452
2453         /* data */
2454         m->nr_pages = 0;
2455         m->page_alignment = 0;
2456         m->pages = NULL;
2457         m->pagelist = NULL;
2458         m->bio = NULL;
2459         m->bio_iter = NULL;
2460         m->bio_seg = 0;
2461         m->trail = NULL;
2462
2463         /* front */
2464         if (front_len) {
2465                 if (front_len > PAGE_CACHE_SIZE) {
2466                         m->front.iov_base = __vmalloc(front_len, flags,
2467                                                       PAGE_KERNEL);
2468                         m->front_is_vmalloc = true;
2469                 } else {
2470                         m->front.iov_base = kmalloc(front_len, flags);
2471                 }
2472                 if (m->front.iov_base == NULL) {
2473                         dout("ceph_msg_new can't allocate %d bytes\n",
2474                              front_len);
2475                         goto out2;
2476                 }
2477         } else {
2478                 m->front.iov_base = NULL;
2479         }
2480         m->front.iov_len = front_len;
2481
2482         dout("ceph_msg_new %p front %d\n", m, front_len);
2483         return m;
2484
2485 out2:
2486         ceph_msg_put(m);
2487 out:
2488         if (!can_fail) {
2489                 pr_err("msg_new can't create type %d front %d\n", type,
2490                        front_len);
2491                 WARN_ON(1);
2492         } else {
2493                 dout("msg_new can't create type %d front %d\n", type,
2494                      front_len);
2495         }
2496         return NULL;
2497 }
2498 EXPORT_SYMBOL(ceph_msg_new);
2499
2500 /*
2501  * Allocate "middle" portion of a message, if it is needed and wasn't
2502  * allocated by alloc_msg.  This allows us to read a small fixed-size
2503  * per-type header in the front and then gracefully fail (i.e.,
2504  * propagate the error to the caller based on info in the front) when
2505  * the middle is too large.
2506  */
2507 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2508 {
2509         int type = le16_to_cpu(msg->hdr.type);
2510         int middle_len = le32_to_cpu(msg->hdr.middle_len);
2511
2512         dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2513              ceph_msg_type_name(type), middle_len);
2514         BUG_ON(!middle_len);
2515         BUG_ON(msg->middle);
2516
2517         msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2518         if (!msg->middle)
2519                 return -ENOMEM;
2520         return 0;
2521 }
2522
2523 /*
2524  * Generic message allocator, for incoming messages.
2525  */
2526 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2527                                 struct ceph_msg_header *hdr,
2528                                 int *skip)
2529 {
2530         int type = le16_to_cpu(hdr->type);
2531         int front_len = le32_to_cpu(hdr->front_len);
2532         int middle_len = le32_to_cpu(hdr->middle_len);
2533         struct ceph_msg *msg = NULL;
2534         int ret;
2535
2536         if (con->ops->alloc_msg) {
2537                 mutex_unlock(&con->mutex);
2538                 msg = con->ops->alloc_msg(con, hdr, skip);
2539                 mutex_lock(&con->mutex);
2540                 if (!msg || *skip)
2541                         return NULL;
2542         }
2543         if (!msg) {
2544                 *skip = 0;
2545                 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2546                 if (!msg) {
2547                         pr_err("unable to allocate msg type %d len %d\n",
2548                                type, front_len);
2549                         return NULL;
2550                 }
2551                 msg->page_alignment = le16_to_cpu(hdr->data_off);
2552         }
2553         memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2554
2555         if (middle_len && !msg->middle) {
2556                 ret = ceph_alloc_middle(con, msg);
2557                 if (ret < 0) {
2558                         ceph_msg_put(msg);
2559                         return NULL;
2560                 }
2561         }
2562
2563         return msg;
2564 }
2565
2566
2567 /*
2568  * Free a generically kmalloc'd message.
2569  */
2570 void ceph_msg_kfree(struct ceph_msg *m)
2571 {
2572         dout("msg_kfree %p\n", m);
2573         if (m->front_is_vmalloc)
2574                 vfree(m->front.iov_base);
2575         else
2576                 kfree(m->front.iov_base);
2577         kfree(m);
2578 }
2579
2580 /*
2581  * Drop a msg ref.  Destroy as needed.
2582  */
2583 void ceph_msg_last_put(struct kref *kref)
2584 {
2585         struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2586
2587         dout("ceph_msg_put last one on %p\n", m);
2588         WARN_ON(!list_empty(&m->list_head));
2589
2590         /* drop middle, data, if any */
2591         if (m->middle) {
2592                 ceph_buffer_put(m->middle);
2593                 m->middle = NULL;
2594         }
2595         m->nr_pages = 0;
2596         m->pages = NULL;
2597
2598         if (m->pagelist) {
2599                 ceph_pagelist_release(m->pagelist);
2600                 kfree(m->pagelist);
2601                 m->pagelist = NULL;
2602         }
2603
2604         m->trail = NULL;
2605
2606         if (m->pool)
2607                 ceph_msgpool_put(m->pool, m);
2608         else
2609                 ceph_msg_kfree(m);
2610 }
2611 EXPORT_SYMBOL(ceph_msg_last_put);
2612
2613 void ceph_msg_dump(struct ceph_msg *msg)
2614 {
2615         pr_debug("msg_dump %p (front_alloc_len %d nr_pages %d)\n", msg,
2616                  msg->front_alloc_len, msg->nr_pages);
2617         print_hex_dump(KERN_DEBUG, "header: ",
2618                        DUMP_PREFIX_OFFSET, 16, 1,
2619                        &msg->hdr, sizeof(msg->hdr), true);
2620         print_hex_dump(KERN_DEBUG, " front: ",
2621                        DUMP_PREFIX_OFFSET, 16, 1,
2622                        msg->front.iov_base, msg->front.iov_len, true);
2623         if (msg->middle)
2624                 print_hex_dump(KERN_DEBUG, "middle: ",
2625                                DUMP_PREFIX_OFFSET, 16, 1,
2626                                msg->middle->vec.iov_base,
2627                                msg->middle->vec.iov_len, true);
2628         print_hex_dump(KERN_DEBUG, "footer: ",
2629                        DUMP_PREFIX_OFFSET, 16, 1,
2630                        &msg->footer, sizeof(msg->footer), true);
2631 }
2632 EXPORT_SYMBOL(ceph_msg_dump);