fs/ceph/messenger.c

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