linux/net/ceph/messenger.c
<<
>>
Prefs
   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/*
  33 * We track the state of the socket on a given connection using
  34 * values defined below.  The transition to a new socket state is
  35 * handled by a function which verifies we aren't coming from an
  36 * unexpected state.
  37 *
  38 *      --------
  39 *      | NEW* |  transient initial state
  40 *      --------
  41 *          | con_sock_state_init()
  42 *          v
  43 *      ----------
  44 *      | CLOSED |  initialized, but no socket (and no
  45 *      ----------  TCP connection)
  46 *       ^      \
  47 *       |       \ con_sock_state_connecting()
  48 *       |        ----------------------
  49 *       |                              \
  50 *       + con_sock_state_closed()       \
  51 *       |+---------------------------    \
  52 *       | \                          \    \
  53 *       |  -----------                \    \
  54 *       |  | CLOSING |  socket event;  \    \
  55 *       |  -----------  await close     \    \
  56 *       |       ^                        \   |
  57 *       |       |                         \  |
  58 *       |       + con_sock_state_closing() \ |
  59 *       |      / \                         | |
  60 *       |     /   ---------------          | |
  61 *       |    /                   \         v v
  62 *       |   /                    --------------
  63 *       |  /    -----------------| CONNECTING |  socket created, TCP
  64 *       |  |   /                 --------------  connect initiated
  65 *       |  |   | con_sock_state_connected()
  66 *       |  |   v
  67 *      -------------
  68 *      | CONNECTED |  TCP connection established
  69 *      -------------
  70 *
  71 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
  72 */
  73
  74#define CON_SOCK_STATE_NEW              0       /* -> CLOSED */
  75#define CON_SOCK_STATE_CLOSED           1       /* -> CONNECTING */
  76#define CON_SOCK_STATE_CONNECTING       2       /* -> CONNECTED or -> CLOSING */
  77#define CON_SOCK_STATE_CONNECTED        3       /* -> CLOSING or -> CLOSED */
  78#define CON_SOCK_STATE_CLOSING          4       /* -> CLOSED */
  79
  80/*
  81 * connection states
  82 */
  83#define CON_STATE_CLOSED        1  /* -> PREOPEN */
  84#define CON_STATE_PREOPEN       2  /* -> CONNECTING, CLOSED */
  85#define CON_STATE_CONNECTING    3  /* -> NEGOTIATING, CLOSED */
  86#define CON_STATE_NEGOTIATING   4  /* -> OPEN, CLOSED */
  87#define CON_STATE_OPEN          5  /* -> STANDBY, CLOSED */
  88#define CON_STATE_STANDBY       6  /* -> PREOPEN, CLOSED */
  89
  90/*
  91 * ceph_connection flag bits
  92 */
  93#define CON_FLAG_LOSSYTX           0  /* we can close channel or drop
  94                                       * messages on errors */
  95#define CON_FLAG_KEEPALIVE_PENDING 1  /* we need to send a keepalive */
  96#define CON_FLAG_WRITE_PENDING     2  /* we have data ready to send */
  97#define CON_FLAG_SOCK_CLOSED       3  /* socket state changed to closed */
  98#define CON_FLAG_BACKOFF           4  /* need to retry queuing delayed work */
  99
 100/* static tag bytes (protocol control messages) */
 101static char tag_msg = CEPH_MSGR_TAG_MSG;
 102static char tag_ack = CEPH_MSGR_TAG_ACK;
 103static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
 104
 105#ifdef CONFIG_LOCKDEP
 106static struct lock_class_key socket_class;
 107#endif
 108
 109/*
 110 * When skipping (ignoring) a block of input we read it into a "skip
 111 * buffer," which is this many bytes in size.
 112 */
 113#define SKIP_BUF_SIZE   1024
 114
 115static void queue_con(struct ceph_connection *con);
 116static void con_work(struct work_struct *);
 117static void ceph_fault(struct ceph_connection *con);
 118
 119/*
 120 * Nicely render a sockaddr as a string.  An array of formatted
 121 * strings is used, to approximate reentrancy.
 122 */
 123#define ADDR_STR_COUNT_LOG      5       /* log2(# address strings in array) */
 124#define ADDR_STR_COUNT          (1 << ADDR_STR_COUNT_LOG)
 125#define ADDR_STR_COUNT_MASK     (ADDR_STR_COUNT - 1)
 126#define MAX_ADDR_STR_LEN        64      /* 54 is enough */
 127
 128static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
 129static atomic_t addr_str_seq = ATOMIC_INIT(0);
 130
 131static struct page *zero_page;          /* used in certain error cases */
 132
 133const char *ceph_pr_addr(const struct sockaddr_storage *ss)
 134{
 135        int i;
 136        char *s;
 137        struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
 138        struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
 139
 140        i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
 141        s = addr_str[i];
 142
 143        switch (ss->ss_family) {
 144        case AF_INET:
 145                snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
 146                         ntohs(in4->sin_port));
 147                break;
 148
 149        case AF_INET6:
 150                snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
 151                         ntohs(in6->sin6_port));
 152                break;
 153
 154        default:
 155                snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
 156                         ss->ss_family);
 157        }
 158
 159        return s;
 160}
 161EXPORT_SYMBOL(ceph_pr_addr);
 162
 163static void encode_my_addr(struct ceph_messenger *msgr)
 164{
 165        memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
 166        ceph_encode_addr(&msgr->my_enc_addr);
 167}
 168
 169/*
 170 * work queue for all reading and writing to/from the socket.
 171 */
 172static struct workqueue_struct *ceph_msgr_wq;
 173
 174void _ceph_msgr_exit(void)
 175{
 176        if (ceph_msgr_wq) {
 177                destroy_workqueue(ceph_msgr_wq);
 178                ceph_msgr_wq = NULL;
 179        }
 180
 181        BUG_ON(zero_page == NULL);
 182        kunmap(zero_page);
 183        page_cache_release(zero_page);
 184        zero_page = NULL;
 185}
 186
 187int ceph_msgr_init(void)
 188{
 189        BUG_ON(zero_page != NULL);
 190        zero_page = ZERO_PAGE(0);
 191        page_cache_get(zero_page);
 192
 193        ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
 194        if (ceph_msgr_wq)
 195                return 0;
 196
 197        pr_err("msgr_init failed to create workqueue\n");
 198        _ceph_msgr_exit();
 199
 200        return -ENOMEM;
 201}
 202EXPORT_SYMBOL(ceph_msgr_init);
 203
 204void ceph_msgr_exit(void)
 205{
 206        BUG_ON(ceph_msgr_wq == NULL);
 207
 208        _ceph_msgr_exit();
 209}
 210EXPORT_SYMBOL(ceph_msgr_exit);
 211
 212void ceph_msgr_flush(void)
 213{
 214        flush_workqueue(ceph_msgr_wq);
 215}
 216EXPORT_SYMBOL(ceph_msgr_flush);
 217
 218/* Connection socket state transition functions */
 219
 220static void con_sock_state_init(struct ceph_connection *con)
 221{
 222        int old_state;
 223
 224        old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
 225        if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
 226                printk("%s: unexpected old state %d\n", __func__, old_state);
 227        dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 228             CON_SOCK_STATE_CLOSED);
 229}
 230
 231static void con_sock_state_connecting(struct ceph_connection *con)
 232{
 233        int old_state;
 234
 235        old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
 236        if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
 237                printk("%s: unexpected old state %d\n", __func__, old_state);
 238        dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 239             CON_SOCK_STATE_CONNECTING);
 240}
 241
 242static void con_sock_state_connected(struct ceph_connection *con)
 243{
 244        int old_state;
 245
 246        old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
 247        if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
 248                printk("%s: unexpected old state %d\n", __func__, old_state);
 249        dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 250             CON_SOCK_STATE_CONNECTED);
 251}
 252
 253static void con_sock_state_closing(struct ceph_connection *con)
 254{
 255        int old_state;
 256
 257        old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
 258        if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
 259                        old_state != CON_SOCK_STATE_CONNECTED &&
 260                        old_state != CON_SOCK_STATE_CLOSING))
 261                printk("%s: unexpected old state %d\n", __func__, old_state);
 262        dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 263             CON_SOCK_STATE_CLOSING);
 264}
 265
 266static void con_sock_state_closed(struct ceph_connection *con)
 267{
 268        int old_state;
 269
 270        old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
 271        if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
 272                    old_state != CON_SOCK_STATE_CLOSING &&
 273                    old_state != CON_SOCK_STATE_CONNECTING &&
 274                    old_state != CON_SOCK_STATE_CLOSED))
 275                printk("%s: unexpected old state %d\n", __func__, old_state);
 276        dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
 277             CON_SOCK_STATE_CLOSED);
 278}
 279
 280/*
 281 * socket callback functions
 282 */
 283
 284/* data available on socket, or listen socket received a connect */
 285static void ceph_sock_data_ready(struct sock *sk, int count_unused)
 286{
 287        struct ceph_connection *con = sk->sk_user_data;
 288        if (atomic_read(&con->msgr->stopping)) {
 289                return;
 290        }
 291
 292        if (sk->sk_state != TCP_CLOSE_WAIT) {
 293                dout("%s on %p state = %lu, queueing work\n", __func__,
 294                     con, con->state);
 295                queue_con(con);
 296        }
 297}
 298
 299/* socket has buffer space for writing */
 300static void ceph_sock_write_space(struct sock *sk)
 301{
 302        struct ceph_connection *con = sk->sk_user_data;
 303
 304        /* only queue to workqueue if there is data we want to write,
 305         * and there is sufficient space in the socket buffer to accept
 306         * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
 307         * doesn't get called again until try_write() fills the socket
 308         * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
 309         * and net/core/stream.c:sk_stream_write_space().
 310         */
 311        if (test_bit(CON_FLAG_WRITE_PENDING, &con->flags)) {
 312                if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 313                        dout("%s %p queueing write work\n", __func__, con);
 314                        clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 315                        queue_con(con);
 316                }
 317        } else {
 318                dout("%s %p nothing to write\n", __func__, con);
 319        }
 320}
 321
 322/* socket's state has changed */
 323static void ceph_sock_state_change(struct sock *sk)
 324{
 325        struct ceph_connection *con = sk->sk_user_data;
 326
 327        dout("%s %p state = %lu sk_state = %u\n", __func__,
 328             con, con->state, sk->sk_state);
 329
 330        switch (sk->sk_state) {
 331        case TCP_CLOSE:
 332                dout("%s TCP_CLOSE\n", __func__);
 333        case TCP_CLOSE_WAIT:
 334                dout("%s TCP_CLOSE_WAIT\n", __func__);
 335                con_sock_state_closing(con);
 336                set_bit(CON_FLAG_SOCK_CLOSED, &con->flags);
 337                queue_con(con);
 338                break;
 339        case TCP_ESTABLISHED:
 340                dout("%s TCP_ESTABLISHED\n", __func__);
 341                con_sock_state_connected(con);
 342                queue_con(con);
 343                break;
 344        default:        /* Everything else is uninteresting */
 345                break;
 346        }
 347}
 348
 349/*
 350 * set up socket callbacks
 351 */
 352static void set_sock_callbacks(struct socket *sock,
 353                               struct ceph_connection *con)
 354{
 355        struct sock *sk = sock->sk;
 356        sk->sk_user_data = con;
 357        sk->sk_data_ready = ceph_sock_data_ready;
 358        sk->sk_write_space = ceph_sock_write_space;
 359        sk->sk_state_change = ceph_sock_state_change;
 360}
 361
 362
 363/*
 364 * socket helpers
 365 */
 366
 367/*
 368 * initiate connection to a remote socket.
 369 */
 370static int ceph_tcp_connect(struct ceph_connection *con)
 371{
 372        struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
 373        struct socket *sock;
 374        int ret;
 375
 376        BUG_ON(con->sock);
 377        ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
 378                               IPPROTO_TCP, &sock);
 379        if (ret)
 380                return ret;
 381        sock->sk->sk_allocation = GFP_NOFS;
 382
 383#ifdef CONFIG_LOCKDEP
 384        lockdep_set_class(&sock->sk->sk_lock, &socket_class);
 385#endif
 386
 387        set_sock_callbacks(sock, con);
 388
 389        dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
 390
 391        con_sock_state_connecting(con);
 392        ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
 393                                 O_NONBLOCK);
 394        if (ret == -EINPROGRESS) {
 395                dout("connect %s EINPROGRESS sk_state = %u\n",
 396                     ceph_pr_addr(&con->peer_addr.in_addr),
 397                     sock->sk->sk_state);
 398        } else if (ret < 0) {
 399                pr_err("connect %s error %d\n",
 400                       ceph_pr_addr(&con->peer_addr.in_addr), ret);
 401                sock_release(sock);
 402                con->error_msg = "connect error";
 403
 404                return ret;
 405        }
 406        con->sock = sock;
 407        return 0;
 408}
 409
 410static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
 411{
 412        struct kvec iov = {buf, len};
 413        struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
 414        int r;
 415
 416        r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
 417        if (r == -EAGAIN)
 418                r = 0;
 419        return r;
 420}
 421
 422/*
 423 * write something.  @more is true if caller will be sending more data
 424 * shortly.
 425 */
 426static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
 427                     size_t kvlen, size_t len, int more)
 428{
 429        struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
 430        int r;
 431
 432        if (more)
 433                msg.msg_flags |= MSG_MORE;
 434        else
 435                msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
 436
 437        r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
 438        if (r == -EAGAIN)
 439                r = 0;
 440        return r;
 441}
 442
 443static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
 444                     int offset, size_t size, int more)
 445{
 446        int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
 447        int ret;
 448
 449        ret = kernel_sendpage(sock, page, offset, size, flags);
 450        if (ret == -EAGAIN)
 451                ret = 0;
 452
 453        return ret;
 454}
 455
 456
 457/*
 458 * Shutdown/close the socket for the given connection.
 459 */
 460static int con_close_socket(struct ceph_connection *con)
 461{
 462        int rc = 0;
 463
 464        dout("con_close_socket on %p sock %p\n", con, con->sock);
 465        if (con->sock) {
 466                rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
 467                sock_release(con->sock);
 468                con->sock = NULL;
 469        }
 470
 471        /*
 472         * Forcibly clear the SOCK_CLOSED flag.  It gets set
 473         * independent of the connection mutex, and we could have
 474         * received a socket close event before we had the chance to
 475         * shut the socket down.
 476         */
 477        clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags);
 478
 479        con_sock_state_closed(con);
 480        return rc;
 481}
 482
 483/*
 484 * Reset a connection.  Discard all incoming and outgoing messages
 485 * and clear *_seq state.
 486 */
 487static void ceph_msg_remove(struct ceph_msg *msg)
 488{
 489        list_del_init(&msg->list_head);
 490        BUG_ON(msg->con == NULL);
 491        msg->con->ops->put(msg->con);
 492        msg->con = NULL;
 493
 494        ceph_msg_put(msg);
 495}
 496static void ceph_msg_remove_list(struct list_head *head)
 497{
 498        while (!list_empty(head)) {
 499                struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
 500                                                        list_head);
 501                ceph_msg_remove(msg);
 502        }
 503}
 504
 505static void reset_connection(struct ceph_connection *con)
 506{
 507        /* reset connection, out_queue, msg_ and connect_seq */
 508        /* discard existing out_queue and msg_seq */
 509        ceph_msg_remove_list(&con->out_queue);
 510        ceph_msg_remove_list(&con->out_sent);
 511
 512        if (con->in_msg) {
 513                BUG_ON(con->in_msg->con != con);
 514                con->in_msg->con = NULL;
 515                ceph_msg_put(con->in_msg);
 516                con->in_msg = NULL;
 517                con->ops->put(con);
 518        }
 519
 520        con->connect_seq = 0;
 521        con->out_seq = 0;
 522        if (con->out_msg) {
 523                ceph_msg_put(con->out_msg);
 524                con->out_msg = NULL;
 525        }
 526        con->in_seq = 0;
 527        con->in_seq_acked = 0;
 528}
 529
 530/*
 531 * mark a peer down.  drop any open connections.
 532 */
 533void ceph_con_close(struct ceph_connection *con)
 534{
 535        mutex_lock(&con->mutex);
 536        dout("con_close %p peer %s\n", con,
 537             ceph_pr_addr(&con->peer_addr.in_addr));
 538        con->state = CON_STATE_CLOSED;
 539
 540        clear_bit(CON_FLAG_LOSSYTX, &con->flags); /* so we retry next connect */
 541        clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags);
 542        clear_bit(CON_FLAG_WRITE_PENDING, &con->flags);
 543        clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags);
 544        clear_bit(CON_FLAG_BACKOFF, &con->flags);
 545
 546        reset_connection(con);
 547        con->peer_global_seq = 0;
 548        cancel_delayed_work(&con->work);
 549        con_close_socket(con);
 550        mutex_unlock(&con->mutex);
 551}
 552EXPORT_SYMBOL(ceph_con_close);
 553
 554/*
 555 * Reopen a closed connection, with a new peer address.
 556 */
 557void ceph_con_open(struct ceph_connection *con,
 558                   __u8 entity_type, __u64 entity_num,
 559                   struct ceph_entity_addr *addr)
 560{
 561        mutex_lock(&con->mutex);
 562        dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
 563
 564        BUG_ON(con->state != CON_STATE_CLOSED);
 565        con->state = CON_STATE_PREOPEN;
 566
 567        con->peer_name.type = (__u8) entity_type;
 568        con->peer_name.num = cpu_to_le64(entity_num);
 569
 570        memcpy(&con->peer_addr, addr, sizeof(*addr));
 571        con->delay = 0;      /* reset backoff memory */
 572        mutex_unlock(&con->mutex);
 573        queue_con(con);
 574}
 575EXPORT_SYMBOL(ceph_con_open);
 576
 577/*
 578 * return true if this connection ever successfully opened
 579 */
 580bool ceph_con_opened(struct ceph_connection *con)
 581{
 582        return con->connect_seq > 0;
 583}
 584
 585/*
 586 * initialize a new connection.
 587 */
 588void ceph_con_init(struct ceph_connection *con, void *private,
 589        const struct ceph_connection_operations *ops,
 590        struct ceph_messenger *msgr)
 591{
 592        dout("con_init %p\n", con);
 593        memset(con, 0, sizeof(*con));
 594        con->private = private;
 595        con->ops = ops;
 596        con->msgr = msgr;
 597
 598        con_sock_state_init(con);
 599
 600        mutex_init(&con->mutex);
 601        INIT_LIST_HEAD(&con->out_queue);
 602        INIT_LIST_HEAD(&con->out_sent);
 603        INIT_DELAYED_WORK(&con->work, con_work);
 604
 605        con->state = CON_STATE_CLOSED;
 606}
 607EXPORT_SYMBOL(ceph_con_init);
 608
 609
 610/*
 611 * We maintain a global counter to order connection attempts.  Get
 612 * a unique seq greater than @gt.
 613 */
 614static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
 615{
 616        u32 ret;
 617
 618        spin_lock(&msgr->global_seq_lock);
 619        if (msgr->global_seq < gt)
 620                msgr->global_seq = gt;
 621        ret = ++msgr->global_seq;
 622        spin_unlock(&msgr->global_seq_lock);
 623        return ret;
 624}
 625
 626static void con_out_kvec_reset(struct ceph_connection *con)
 627{
 628        con->out_kvec_left = 0;
 629        con->out_kvec_bytes = 0;
 630        con->out_kvec_cur = &con->out_kvec[0];
 631}
 632
 633static void con_out_kvec_add(struct ceph_connection *con,
 634                                size_t size, void *data)
 635{
 636        int index;
 637
 638        index = con->out_kvec_left;
 639        BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
 640
 641        con->out_kvec[index].iov_len = size;
 642        con->out_kvec[index].iov_base = data;
 643        con->out_kvec_left++;
 644        con->out_kvec_bytes += size;
 645}
 646
 647#ifdef CONFIG_BLOCK
 648static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
 649{
 650        if (!bio) {
 651                *iter = NULL;
 652                *seg = 0;
 653                return;
 654        }
 655        *iter = bio;
 656        *seg = bio->bi_idx;
 657}
 658
 659static void iter_bio_next(struct bio **bio_iter, int *seg)
 660{
 661        if (*bio_iter == NULL)
 662                return;
 663
 664        BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
 665
 666        (*seg)++;
 667        if (*seg == (*bio_iter)->bi_vcnt)
 668                init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
 669}
 670#endif
 671
 672static void prepare_write_message_data(struct ceph_connection *con)
 673{
 674        struct ceph_msg *msg = con->out_msg;
 675
 676        BUG_ON(!msg);
 677        BUG_ON(!msg->hdr.data_len);
 678
 679        /* initialize page iterator */
 680        con->out_msg_pos.page = 0;
 681        if (msg->pages)
 682                con->out_msg_pos.page_pos = msg->page_alignment;
 683        else
 684                con->out_msg_pos.page_pos = 0;
 685#ifdef CONFIG_BLOCK
 686        if (msg->bio)
 687                init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
 688#endif
 689        con->out_msg_pos.data_pos = 0;
 690        con->out_msg_pos.did_page_crc = false;
 691        con->out_more = 1;  /* data + footer will follow */
 692}
 693
 694/*
 695 * Prepare footer for currently outgoing message, and finish things
 696 * off.  Assumes out_kvec* are already valid.. we just add on to the end.
 697 */
 698static void prepare_write_message_footer(struct ceph_connection *con)
 699{
 700        struct ceph_msg *m = con->out_msg;
 701        int v = con->out_kvec_left;
 702
 703        m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
 704
 705        dout("prepare_write_message_footer %p\n", con);
 706        con->out_kvec_is_msg = true;
 707        con->out_kvec[v].iov_base = &m->footer;
 708        con->out_kvec[v].iov_len = sizeof(m->footer);
 709        con->out_kvec_bytes += sizeof(m->footer);
 710        con->out_kvec_left++;
 711        con->out_more = m->more_to_follow;
 712        con->out_msg_done = true;
 713}
 714
 715/*
 716 * Prepare headers for the next outgoing message.
 717 */
 718static void prepare_write_message(struct ceph_connection *con)
 719{
 720        struct ceph_msg *m;
 721        u32 crc;
 722
 723        con_out_kvec_reset(con);
 724        con->out_kvec_is_msg = true;
 725        con->out_msg_done = false;
 726
 727        /* Sneak an ack in there first?  If we can get it into the same
 728         * TCP packet that's a good thing. */
 729        if (con->in_seq > con->in_seq_acked) {
 730                con->in_seq_acked = con->in_seq;
 731                con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
 732                con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
 733                con_out_kvec_add(con, sizeof (con->out_temp_ack),
 734                        &con->out_temp_ack);
 735        }
 736
 737        BUG_ON(list_empty(&con->out_queue));
 738        m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
 739        con->out_msg = m;
 740        BUG_ON(m->con != con);
 741
 742        /* put message on sent list */
 743        ceph_msg_get(m);
 744        list_move_tail(&m->list_head, &con->out_sent);
 745
 746        /*
 747         * only assign outgoing seq # if we haven't sent this message
 748         * yet.  if it is requeued, resend with it's original seq.
 749         */
 750        if (m->needs_out_seq) {
 751                m->hdr.seq = cpu_to_le64(++con->out_seq);
 752                m->needs_out_seq = false;
 753        }
 754#ifdef CONFIG_BLOCK
 755        else
 756                m->bio_iter = NULL;
 757#endif
 758
 759        dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
 760             m, con->out_seq, le16_to_cpu(m->hdr.type),
 761             le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
 762             le32_to_cpu(m->hdr.data_len),
 763             m->nr_pages);
 764        BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
 765
 766        /* tag + hdr + front + middle */
 767        con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
 768        con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
 769        con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
 770
 771        if (m->middle)
 772                con_out_kvec_add(con, m->middle->vec.iov_len,
 773                        m->middle->vec.iov_base);
 774
 775        /* fill in crc (except data pages), footer */
 776        crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
 777        con->out_msg->hdr.crc = cpu_to_le32(crc);
 778        con->out_msg->footer.flags = 0;
 779
 780        crc = crc32c(0, m->front.iov_base, m->front.iov_len);
 781        con->out_msg->footer.front_crc = cpu_to_le32(crc);
 782        if (m->middle) {
 783                crc = crc32c(0, m->middle->vec.iov_base,
 784                                m->middle->vec.iov_len);
 785                con->out_msg->footer.middle_crc = cpu_to_le32(crc);
 786        } else
 787                con->out_msg->footer.middle_crc = 0;
 788        dout("%s front_crc %u middle_crc %u\n", __func__,
 789             le32_to_cpu(con->out_msg->footer.front_crc),
 790             le32_to_cpu(con->out_msg->footer.middle_crc));
 791
 792        /* is there a data payload? */
 793        con->out_msg->footer.data_crc = 0;
 794        if (m->hdr.data_len)
 795                prepare_write_message_data(con);
 796        else
 797                /* no, queue up footer too and be done */
 798                prepare_write_message_footer(con);
 799
 800        set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
 801}
 802
 803/*
 804 * Prepare an ack.
 805 */
 806static void prepare_write_ack(struct ceph_connection *con)
 807{
 808        dout("prepare_write_ack %p %llu -> %llu\n", con,
 809             con->in_seq_acked, con->in_seq);
 810        con->in_seq_acked = con->in_seq;
 811
 812        con_out_kvec_reset(con);
 813
 814        con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
 815
 816        con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
 817        con_out_kvec_add(con, sizeof (con->out_temp_ack),
 818                                &con->out_temp_ack);
 819
 820        con->out_more = 1;  /* more will follow.. eventually.. */
 821        set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
 822}
 823
 824/*
 825 * Prepare to write keepalive byte.
 826 */
 827static void prepare_write_keepalive(struct ceph_connection *con)
 828{
 829        dout("prepare_write_keepalive %p\n", con);
 830        con_out_kvec_reset(con);
 831        con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
 832        set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
 833}
 834
 835/*
 836 * Connection negotiation.
 837 */
 838
 839static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
 840                                                int *auth_proto)
 841{
 842        struct ceph_auth_handshake *auth;
 843
 844        if (!con->ops->get_authorizer) {
 845                con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
 846                con->out_connect.authorizer_len = 0;
 847                return NULL;
 848        }
 849
 850        /* Can't hold the mutex while getting authorizer */
 851        mutex_unlock(&con->mutex);
 852        auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
 853        mutex_lock(&con->mutex);
 854
 855        if (IS_ERR(auth))
 856                return auth;
 857        if (con->state != CON_STATE_NEGOTIATING)
 858                return ERR_PTR(-EAGAIN);
 859
 860        con->auth_reply_buf = auth->authorizer_reply_buf;
 861        con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
 862        return auth;
 863}
 864
 865/*
 866 * We connected to a peer and are saying hello.
 867 */
 868static void prepare_write_banner(struct ceph_connection *con)
 869{
 870        con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
 871        con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
 872                                        &con->msgr->my_enc_addr);
 873
 874        con->out_more = 0;
 875        set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
 876}
 877
 878static int prepare_write_connect(struct ceph_connection *con)
 879{
 880        unsigned int global_seq = get_global_seq(con->msgr, 0);
 881        int proto;
 882        int auth_proto;
 883        struct ceph_auth_handshake *auth;
 884
 885        switch (con->peer_name.type) {
 886        case CEPH_ENTITY_TYPE_MON:
 887                proto = CEPH_MONC_PROTOCOL;
 888                break;
 889        case CEPH_ENTITY_TYPE_OSD:
 890                proto = CEPH_OSDC_PROTOCOL;
 891                break;
 892        case CEPH_ENTITY_TYPE_MDS:
 893                proto = CEPH_MDSC_PROTOCOL;
 894                break;
 895        default:
 896                BUG();
 897        }
 898
 899        dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
 900             con->connect_seq, global_seq, proto);
 901
 902        con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
 903        con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
 904        con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
 905        con->out_connect.global_seq = cpu_to_le32(global_seq);
 906        con->out_connect.protocol_version = cpu_to_le32(proto);
 907        con->out_connect.flags = 0;
 908
 909        auth_proto = CEPH_AUTH_UNKNOWN;
 910        auth = get_connect_authorizer(con, &auth_proto);
 911        if (IS_ERR(auth))
 912                return PTR_ERR(auth);
 913
 914        con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
 915        con->out_connect.authorizer_len = auth ?
 916                cpu_to_le32(auth->authorizer_buf_len) : 0;
 917
 918        con_out_kvec_add(con, sizeof (con->out_connect),
 919                                        &con->out_connect);
 920        if (auth && auth->authorizer_buf_len)
 921                con_out_kvec_add(con, auth->authorizer_buf_len,
 922                                        auth->authorizer_buf);
 923
 924        con->out_more = 0;
 925        set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
 926
 927        return 0;
 928}
 929
 930/*
 931 * write as much of pending kvecs to the socket as we can.
 932 *  1 -> done
 933 *  0 -> socket full, but more to do
 934 * <0 -> error
 935 */
 936static int write_partial_kvec(struct ceph_connection *con)
 937{
 938        int ret;
 939
 940        dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
 941        while (con->out_kvec_bytes > 0) {
 942                ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
 943                                       con->out_kvec_left, con->out_kvec_bytes,
 944                                       con->out_more);
 945                if (ret <= 0)
 946                        goto out;
 947                con->out_kvec_bytes -= ret;
 948                if (con->out_kvec_bytes == 0)
 949                        break;            /* done */
 950
 951                /* account for full iov entries consumed */
 952                while (ret >= con->out_kvec_cur->iov_len) {
 953                        BUG_ON(!con->out_kvec_left);
 954                        ret -= con->out_kvec_cur->iov_len;
 955                        con->out_kvec_cur++;
 956                        con->out_kvec_left--;
 957                }
 958                /* and for a partially-consumed entry */
 959                if (ret) {
 960                        con->out_kvec_cur->iov_len -= ret;
 961                        con->out_kvec_cur->iov_base += ret;
 962                }
 963        }
 964        con->out_kvec_left = 0;
 965        con->out_kvec_is_msg = false;
 966        ret = 1;
 967out:
 968        dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
 969             con->out_kvec_bytes, con->out_kvec_left, ret);
 970        return ret;  /* done! */
 971}
 972
 973static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
 974                        size_t len, size_t sent, bool in_trail)
 975{
 976        struct ceph_msg *msg = con->out_msg;
 977
 978        BUG_ON(!msg);
 979        BUG_ON(!sent);
 980
 981        con->out_msg_pos.data_pos += sent;
 982        con->out_msg_pos.page_pos += sent;
 983        if (sent < len)
 984                return;
 985
 986        BUG_ON(sent != len);
 987        con->out_msg_pos.page_pos = 0;
 988        con->out_msg_pos.page++;
 989        con->out_msg_pos.did_page_crc = false;
 990        if (in_trail)
 991                list_move_tail(&page->lru,
 992                               &msg->trail->head);
 993        else if (msg->pagelist)
 994                list_move_tail(&page->lru,
 995                               &msg->pagelist->head);
 996#ifdef CONFIG_BLOCK
 997        else if (msg->bio)
 998                iter_bio_next(&msg->bio_iter, &msg->bio_seg);
 999#endif
1000}
1001
1002/*
1003 * Write as much message data payload as we can.  If we finish, queue
1004 * up the footer.
1005 *  1 -> done, footer is now queued in out_kvec[].
1006 *  0 -> socket full, but more to do
1007 * <0 -> error
1008 */
1009static int write_partial_msg_pages(struct ceph_connection *con)
1010{
1011        struct ceph_msg *msg = con->out_msg;
1012        unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1013        size_t len;
1014        bool do_datacrc = !con->msgr->nocrc;
1015        int ret;
1016        int total_max_write;
1017        bool in_trail = false;
1018        const size_t trail_len = (msg->trail ? msg->trail->length : 0);
1019        const size_t trail_off = data_len - trail_len;
1020
1021        dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1022             con, msg, con->out_msg_pos.page, msg->nr_pages,
1023             con->out_msg_pos.page_pos);
1024
1025        /*
1026         * Iterate through each page that contains data to be
1027         * written, and send as much as possible for each.
1028         *
1029         * If we are calculating the data crc (the default), we will
1030         * need to map the page.  If we have no pages, they have
1031         * been revoked, so use the zero page.
1032         */
1033        while (data_len > con->out_msg_pos.data_pos) {
1034                struct page *page = NULL;
1035                int max_write = PAGE_SIZE;
1036                int bio_offset = 0;
1037
1038                in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off;
1039                if (!in_trail)
1040                        total_max_write = trail_off - con->out_msg_pos.data_pos;
1041
1042                if (in_trail) {
1043                        total_max_write = data_len - con->out_msg_pos.data_pos;
1044
1045                        page = list_first_entry(&msg->trail->head,
1046                                                struct page, lru);
1047                } else if (msg->pages) {
1048                        page = msg->pages[con->out_msg_pos.page];
1049                } else if (msg->pagelist) {
1050                        page = list_first_entry(&msg->pagelist->head,
1051                                                struct page, lru);
1052#ifdef CONFIG_BLOCK
1053                } else if (msg->bio) {
1054                        struct bio_vec *bv;
1055
1056                        bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
1057                        page = bv->bv_page;
1058                        bio_offset = bv->bv_offset;
1059                        max_write = bv->bv_len;
1060#endif
1061                } else {
1062                        page = zero_page;
1063                }
1064                len = min_t(int, max_write - con->out_msg_pos.page_pos,
1065                            total_max_write);
1066
1067                if (do_datacrc && !con->out_msg_pos.did_page_crc) {
1068                        void *base;
1069                        u32 crc = le32_to_cpu(msg->footer.data_crc);
1070                        char *kaddr;
1071
1072                        kaddr = kmap(page);
1073                        BUG_ON(kaddr == NULL);
1074                        base = kaddr + con->out_msg_pos.page_pos + bio_offset;
1075                        crc = crc32c(crc, base, len);
1076                        kunmap(page);
1077                        msg->footer.data_crc = cpu_to_le32(crc);
1078                        con->out_msg_pos.did_page_crc = true;
1079                }
1080                ret = ceph_tcp_sendpage(con->sock, page,
1081                                      con->out_msg_pos.page_pos + bio_offset,
1082                                      len, 1);
1083                if (ret <= 0)
1084                        goto out;
1085
1086                out_msg_pos_next(con, page, len, (size_t) ret, in_trail);
1087        }
1088
1089        dout("write_partial_msg_pages %p msg %p done\n", con, msg);
1090
1091        /* prepare and queue up footer, too */
1092        if (!do_datacrc)
1093                msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1094        con_out_kvec_reset(con);
1095        prepare_write_message_footer(con);
1096        ret = 1;
1097out:
1098        return ret;
1099}
1100
1101/*
1102 * write some zeros
1103 */
1104static int write_partial_skip(struct ceph_connection *con)
1105{
1106        int ret;
1107
1108        while (con->out_skip > 0) {
1109                size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1110
1111                ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1);
1112                if (ret <= 0)
1113                        goto out;
1114                con->out_skip -= ret;
1115        }
1116        ret = 1;
1117out:
1118        return ret;
1119}
1120
1121/*
1122 * Prepare to read connection handshake, or an ack.
1123 */
1124static void prepare_read_banner(struct ceph_connection *con)
1125{
1126        dout("prepare_read_banner %p\n", con);
1127        con->in_base_pos = 0;
1128}
1129
1130static void prepare_read_connect(struct ceph_connection *con)
1131{
1132        dout("prepare_read_connect %p\n", con);
1133        con->in_base_pos = 0;
1134}
1135
1136static void prepare_read_ack(struct ceph_connection *con)
1137{
1138        dout("prepare_read_ack %p\n", con);
1139        con->in_base_pos = 0;
1140}
1141
1142static void prepare_read_tag(struct ceph_connection *con)
1143{
1144        dout("prepare_read_tag %p\n", con);
1145        con->in_base_pos = 0;
1146        con->in_tag = CEPH_MSGR_TAG_READY;
1147}
1148
1149/*
1150 * Prepare to read a message.
1151 */
1152static int prepare_read_message(struct ceph_connection *con)
1153{
1154        dout("prepare_read_message %p\n", con);
1155        BUG_ON(con->in_msg != NULL);
1156        con->in_base_pos = 0;
1157        con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1158        return 0;
1159}
1160
1161
1162static int read_partial(struct ceph_connection *con,
1163                        int end, int size, void *object)
1164{
1165        while (con->in_base_pos < end) {
1166                int left = end - con->in_base_pos;
1167                int have = size - left;
1168                int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1169                if (ret <= 0)
1170                        return ret;
1171                con->in_base_pos += ret;
1172        }
1173        return 1;
1174}
1175
1176
1177/*
1178 * Read all or part of the connect-side handshake on a new connection
1179 */
1180static int read_partial_banner(struct ceph_connection *con)
1181{
1182        int size;
1183        int end;
1184        int ret;
1185
1186        dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1187
1188        /* peer's banner */
1189        size = strlen(CEPH_BANNER);
1190        end = size;
1191        ret = read_partial(con, end, size, con->in_banner);
1192        if (ret <= 0)
1193                goto out;
1194
1195        size = sizeof (con->actual_peer_addr);
1196        end += size;
1197        ret = read_partial(con, end, size, &con->actual_peer_addr);
1198        if (ret <= 0)
1199                goto out;
1200
1201        size = sizeof (con->peer_addr_for_me);
1202        end += size;
1203        ret = read_partial(con, end, size, &con->peer_addr_for_me);
1204        if (ret <= 0)
1205                goto out;
1206
1207out:
1208        return ret;
1209}
1210
1211static int read_partial_connect(struct ceph_connection *con)
1212{
1213        int size;
1214        int end;
1215        int ret;
1216
1217        dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1218
1219        size = sizeof (con->in_reply);
1220        end = size;
1221        ret = read_partial(con, end, size, &con->in_reply);
1222        if (ret <= 0)
1223                goto out;
1224
1225        size = le32_to_cpu(con->in_reply.authorizer_len);
1226        end += size;
1227        ret = read_partial(con, end, size, con->auth_reply_buf);
1228        if (ret <= 0)
1229                goto out;
1230
1231        dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1232             con, (int)con->in_reply.tag,
1233             le32_to_cpu(con->in_reply.connect_seq),
1234             le32_to_cpu(con->in_reply.global_seq));
1235out:
1236        return ret;
1237
1238}
1239
1240/*
1241 * Verify the hello banner looks okay.
1242 */
1243static int verify_hello(struct ceph_connection *con)
1244{
1245        if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1246                pr_err("connect to %s got bad banner\n",
1247                       ceph_pr_addr(&con->peer_addr.in_addr));
1248                con->error_msg = "protocol error, bad banner";
1249                return -1;
1250        }
1251        return 0;
1252}
1253
1254static bool addr_is_blank(struct sockaddr_storage *ss)
1255{
1256        switch (ss->ss_family) {
1257        case AF_INET:
1258                return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1259        case AF_INET6:
1260                return
1261                     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1262                     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1263                     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1264                     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1265        }
1266        return false;
1267}
1268
1269static int addr_port(struct sockaddr_storage *ss)
1270{
1271        switch (ss->ss_family) {
1272        case AF_INET:
1273                return ntohs(((struct sockaddr_in *)ss)->sin_port);
1274        case AF_INET6:
1275                return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1276        }
1277        return 0;
1278}
1279
1280static void addr_set_port(struct sockaddr_storage *ss, int p)
1281{
1282        switch (ss->ss_family) {
1283        case AF_INET:
1284                ((struct sockaddr_in *)ss)->sin_port = htons(p);
1285                break;
1286        case AF_INET6:
1287                ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1288                break;
1289        }
1290}
1291
1292/*
1293 * Unlike other *_pton function semantics, zero indicates success.
1294 */
1295static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1296                char delim, const char **ipend)
1297{
1298        struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1299        struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1300
1301        memset(ss, 0, sizeof(*ss));
1302
1303        if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1304                ss->ss_family = AF_INET;
1305                return 0;
1306        }
1307
1308        if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1309                ss->ss_family = AF_INET6;
1310                return 0;
1311        }
1312
1313        return -EINVAL;
1314}
1315
1316/*
1317 * Extract hostname string and resolve using kernel DNS facility.
1318 */
1319#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1320static int ceph_dns_resolve_name(const char *name, size_t namelen,
1321                struct sockaddr_storage *ss, char delim, const char **ipend)
1322{
1323        const char *end, *delim_p;
1324        char *colon_p, *ip_addr = NULL;
1325        int ip_len, ret;
1326
1327        /*
1328         * The end of the hostname occurs immediately preceding the delimiter or
1329         * the port marker (':') where the delimiter takes precedence.
1330         */
1331        delim_p = memchr(name, delim, namelen);
1332        colon_p = memchr(name, ':', namelen);
1333
1334        if (delim_p && colon_p)
1335                end = delim_p < colon_p ? delim_p : colon_p;
1336        else if (!delim_p && colon_p)
1337                end = colon_p;
1338        else {
1339                end = delim_p;
1340                if (!end) /* case: hostname:/ */
1341                        end = name + namelen;
1342        }
1343
1344        if (end <= name)
1345                return -EINVAL;
1346
1347        /* do dns_resolve upcall */
1348        ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1349        if (ip_len > 0)
1350                ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1351        else
1352                ret = -ESRCH;
1353
1354        kfree(ip_addr);
1355
1356        *ipend = end;
1357
1358        pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1359                        ret, ret ? "failed" : ceph_pr_addr(ss));
1360
1361        return ret;
1362}
1363#else
1364static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1365                struct sockaddr_storage *ss, char delim, const char **ipend)
1366{
1367        return -EINVAL;
1368}
1369#endif
1370
1371/*
1372 * Parse a server name (IP or hostname). If a valid IP address is not found
1373 * then try to extract a hostname to resolve using userspace DNS upcall.
1374 */
1375static int ceph_parse_server_name(const char *name, size_t namelen,
1376                        struct sockaddr_storage *ss, char delim, const char **ipend)
1377{
1378        int ret;
1379
1380        ret = ceph_pton(name, namelen, ss, delim, ipend);
1381        if (ret)
1382                ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1383
1384        return ret;
1385}
1386
1387/*
1388 * Parse an ip[:port] list into an addr array.  Use the default
1389 * monitor port if a port isn't specified.
1390 */
1391int ceph_parse_ips(const char *c, const char *end,
1392                   struct ceph_entity_addr *addr,
1393                   int max_count, int *count)
1394{
1395        int i, ret = -EINVAL;
1396        const char *p = c;
1397
1398        dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1399        for (i = 0; i < max_count; i++) {
1400                const char *ipend;
1401                struct sockaddr_storage *ss = &addr[i].in_addr;
1402                int port;
1403                char delim = ',';
1404
1405                if (*p == '[') {
1406                        delim = ']';
1407                        p++;
1408                }
1409
1410                ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1411                if (ret)
1412                        goto bad;
1413                ret = -EINVAL;
1414
1415                p = ipend;
1416
1417                if (delim == ']') {
1418                        if (*p != ']') {
1419                                dout("missing matching ']'\n");
1420                                goto bad;
1421                        }
1422                        p++;
1423                }
1424
1425                /* port? */
1426                if (p < end && *p == ':') {
1427                        port = 0;
1428                        p++;
1429                        while (p < end && *p >= '0' && *p <= '9') {
1430                                port = (port * 10) + (*p - '0');
1431                                p++;
1432                        }
1433                        if (port > 65535 || port == 0)
1434                                goto bad;
1435                } else {
1436                        port = CEPH_MON_PORT;
1437                }
1438
1439                addr_set_port(ss, port);
1440
1441                dout("parse_ips got %s\n", ceph_pr_addr(ss));
1442
1443                if (p == end)
1444                        break;
1445                if (*p != ',')
1446                        goto bad;
1447                p++;
1448        }
1449
1450        if (p != end)
1451                goto bad;
1452
1453        if (count)
1454                *count = i + 1;
1455        return 0;
1456
1457bad:
1458        pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1459        return ret;
1460}
1461EXPORT_SYMBOL(ceph_parse_ips);
1462
1463static int process_banner(struct ceph_connection *con)
1464{
1465        dout("process_banner on %p\n", con);
1466
1467        if (verify_hello(con) < 0)
1468                return -1;
1469
1470        ceph_decode_addr(&con->actual_peer_addr);
1471        ceph_decode_addr(&con->peer_addr_for_me);
1472
1473        /*
1474         * Make sure the other end is who we wanted.  note that the other
1475         * end may not yet know their ip address, so if it's 0.0.0.0, give
1476         * them the benefit of the doubt.
1477         */
1478        if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1479                   sizeof(con->peer_addr)) != 0 &&
1480            !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1481              con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1482                pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1483                           ceph_pr_addr(&con->peer_addr.in_addr),
1484                           (int)le32_to_cpu(con->peer_addr.nonce),
1485                           ceph_pr_addr(&con->actual_peer_addr.in_addr),
1486                           (int)le32_to_cpu(con->actual_peer_addr.nonce));
1487                con->error_msg = "wrong peer at address";
1488                return -1;
1489        }
1490
1491        /*
1492         * did we learn our address?
1493         */
1494        if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1495                int port = addr_port(&con->msgr->inst.addr.in_addr);
1496
1497                memcpy(&con->msgr->inst.addr.in_addr,
1498                       &con->peer_addr_for_me.in_addr,
1499                       sizeof(con->peer_addr_for_me.in_addr));
1500                addr_set_port(&con->msgr->inst.addr.in_addr, port);
1501                encode_my_addr(con->msgr);
1502                dout("process_banner learned my addr is %s\n",
1503                     ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1504        }
1505
1506        return 0;
1507}
1508
1509static void fail_protocol(struct ceph_connection *con)
1510{
1511        reset_connection(con);
1512        BUG_ON(con->state != CON_STATE_NEGOTIATING);
1513        con->state = CON_STATE_CLOSED;
1514}
1515
1516static int process_connect(struct ceph_connection *con)
1517{
1518        u64 sup_feat = con->msgr->supported_features;
1519        u64 req_feat = con->msgr->required_features;
1520        u64 server_feat = le64_to_cpu(con->in_reply.features);
1521        int ret;
1522
1523        dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1524
1525        switch (con->in_reply.tag) {
1526        case CEPH_MSGR_TAG_FEATURES:
1527                pr_err("%s%lld %s feature set mismatch,"
1528                       " my %llx < server's %llx, missing %llx\n",
1529                       ENTITY_NAME(con->peer_name),
1530                       ceph_pr_addr(&con->peer_addr.in_addr),
1531                       sup_feat, server_feat, server_feat & ~sup_feat);
1532                con->error_msg = "missing required protocol features";
1533                fail_protocol(con);
1534                return -1;
1535
1536        case CEPH_MSGR_TAG_BADPROTOVER:
1537                pr_err("%s%lld %s protocol version mismatch,"
1538                       " my %d != server's %d\n",
1539                       ENTITY_NAME(con->peer_name),
1540                       ceph_pr_addr(&con->peer_addr.in_addr),
1541                       le32_to_cpu(con->out_connect.protocol_version),
1542                       le32_to_cpu(con->in_reply.protocol_version));
1543                con->error_msg = "protocol version mismatch";
1544                fail_protocol(con);
1545                return -1;
1546
1547        case CEPH_MSGR_TAG_BADAUTHORIZER:
1548                con->auth_retry++;
1549                dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1550                     con->auth_retry);
1551                if (con->auth_retry == 2) {
1552                        con->error_msg = "connect authorization failure";
1553                        return -1;
1554                }
1555                con->auth_retry = 1;
1556                con_out_kvec_reset(con);
1557                ret = prepare_write_connect(con);
1558                if (ret < 0)
1559                        return ret;
1560                prepare_read_connect(con);
1561                break;
1562
1563        case CEPH_MSGR_TAG_RESETSESSION:
1564                /*
1565                 * If we connected with a large connect_seq but the peer
1566                 * has no record of a session with us (no connection, or
1567                 * connect_seq == 0), they will send RESETSESION to indicate
1568                 * that they must have reset their session, and may have
1569                 * dropped messages.
1570                 */
1571                dout("process_connect got RESET peer seq %u\n",
1572                     le32_to_cpu(con->in_reply.connect_seq));
1573                pr_err("%s%lld %s connection reset\n",
1574                       ENTITY_NAME(con->peer_name),
1575                       ceph_pr_addr(&con->peer_addr.in_addr));
1576                reset_connection(con);
1577                con_out_kvec_reset(con);
1578                ret = prepare_write_connect(con);
1579                if (ret < 0)
1580                        return ret;
1581                prepare_read_connect(con);
1582
1583                /* Tell ceph about it. */
1584                mutex_unlock(&con->mutex);
1585                pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1586                if (con->ops->peer_reset)
1587                        con->ops->peer_reset(con);
1588                mutex_lock(&con->mutex);
1589                if (con->state != CON_STATE_NEGOTIATING)
1590                        return -EAGAIN;
1591                break;
1592
1593        case CEPH_MSGR_TAG_RETRY_SESSION:
1594                /*
1595                 * If we sent a smaller connect_seq than the peer has, try
1596                 * again with a larger value.
1597                 */
1598                dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1599                     le32_to_cpu(con->out_connect.connect_seq),
1600                     le32_to_cpu(con->in_reply.connect_seq));
1601                con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
1602                con_out_kvec_reset(con);
1603                ret = prepare_write_connect(con);
1604                if (ret < 0)
1605                        return ret;
1606                prepare_read_connect(con);
1607                break;
1608
1609        case CEPH_MSGR_TAG_RETRY_GLOBAL:
1610                /*
1611                 * If we sent a smaller global_seq than the peer has, try
1612                 * again with a larger value.
1613                 */
1614                dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1615                     con->peer_global_seq,
1616                     le32_to_cpu(con->in_reply.global_seq));
1617                get_global_seq(con->msgr,
1618                               le32_to_cpu(con->in_reply.global_seq));
1619                con_out_kvec_reset(con);
1620                ret = prepare_write_connect(con);
1621                if (ret < 0)
1622                        return ret;
1623                prepare_read_connect(con);
1624                break;
1625
1626        case CEPH_MSGR_TAG_READY:
1627                if (req_feat & ~server_feat) {
1628                        pr_err("%s%lld %s protocol feature mismatch,"
1629                               " my required %llx > server's %llx, need %llx\n",
1630                               ENTITY_NAME(con->peer_name),
1631                               ceph_pr_addr(&con->peer_addr.in_addr),
1632                               req_feat, server_feat, req_feat & ~server_feat);
1633                        con->error_msg = "missing required protocol features";
1634                        fail_protocol(con);
1635                        return -1;
1636                }
1637
1638                BUG_ON(con->state != CON_STATE_NEGOTIATING);
1639                con->state = CON_STATE_OPEN;
1640
1641                con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1642                con->connect_seq++;
1643                con->peer_features = server_feat;
1644                dout("process_connect got READY gseq %d cseq %d (%d)\n",
1645                     con->peer_global_seq,
1646                     le32_to_cpu(con->in_reply.connect_seq),
1647                     con->connect_seq);
1648                WARN_ON(con->connect_seq !=
1649                        le32_to_cpu(con->in_reply.connect_seq));
1650
1651                if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1652                        set_bit(CON_FLAG_LOSSYTX, &con->flags);
1653
1654                con->delay = 0;      /* reset backoff memory */
1655
1656                prepare_read_tag(con);
1657                break;
1658
1659        case CEPH_MSGR_TAG_WAIT:
1660                /*
1661                 * If there is a connection race (we are opening
1662                 * connections to each other), one of us may just have
1663                 * to WAIT.  This shouldn't happen if we are the
1664                 * client.
1665                 */
1666                pr_err("process_connect got WAIT as client\n");
1667                con->error_msg = "protocol error, got WAIT as client";
1668                return -1;
1669
1670        default:
1671                pr_err("connect protocol error, will retry\n");
1672                con->error_msg = "protocol error, garbage tag during connect";
1673                return -1;
1674        }
1675        return 0;
1676}
1677
1678
1679/*
1680 * read (part of) an ack
1681 */
1682static int read_partial_ack(struct ceph_connection *con)
1683{
1684        int size = sizeof (con->in_temp_ack);
1685        int end = size;
1686
1687        return read_partial(con, end, size, &con->in_temp_ack);
1688}
1689
1690
1691/*
1692 * We can finally discard anything that's been acked.
1693 */
1694static void process_ack(struct ceph_connection *con)
1695{
1696        struct ceph_msg *m;
1697        u64 ack = le64_to_cpu(con->in_temp_ack);
1698        u64 seq;
1699
1700        while (!list_empty(&con->out_sent)) {
1701                m = list_first_entry(&con->out_sent, struct ceph_msg,
1702                                     list_head);
1703                seq = le64_to_cpu(m->hdr.seq);
1704                if (seq > ack)
1705                        break;
1706                dout("got ack for seq %llu type %d at %p\n", seq,
1707                     le16_to_cpu(m->hdr.type), m);
1708                m->ack_stamp = jiffies;
1709                ceph_msg_remove(m);
1710        }
1711        prepare_read_tag(con);
1712}
1713
1714
1715
1716
1717static int read_partial_message_section(struct ceph_connection *con,
1718                                        struct kvec *section,
1719                                        unsigned int sec_len, u32 *crc)
1720{
1721        int ret, left;
1722
1723        BUG_ON(!section);
1724
1725        while (section->iov_len < sec_len) {
1726                BUG_ON(section->iov_base == NULL);
1727                left = sec_len - section->iov_len;
1728                ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1729                                       section->iov_len, left);
1730                if (ret <= 0)
1731                        return ret;
1732                section->iov_len += ret;
1733        }
1734        if (section->iov_len == sec_len)
1735                *crc = crc32c(0, section->iov_base, section->iov_len);
1736
1737        return 1;
1738}
1739
1740static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
1741
1742static int read_partial_message_pages(struct ceph_connection *con,
1743                                      struct page **pages,
1744                                      unsigned int data_len, bool do_datacrc)
1745{
1746        void *p;
1747        int ret;
1748        int left;
1749
1750        left = min((int)(data_len - con->in_msg_pos.data_pos),
1751                   (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1752        /* (page) data */
1753        BUG_ON(pages == NULL);
1754        p = kmap(pages[con->in_msg_pos.page]);
1755        ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1756                               left);
1757        if (ret > 0 && do_datacrc)
1758                con->in_data_crc =
1759                        crc32c(con->in_data_crc,
1760                                  p + con->in_msg_pos.page_pos, ret);
1761        kunmap(pages[con->in_msg_pos.page]);
1762        if (ret <= 0)
1763                return ret;
1764        con->in_msg_pos.data_pos += ret;
1765        con->in_msg_pos.page_pos += ret;
1766        if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1767                con->in_msg_pos.page_pos = 0;
1768                con->in_msg_pos.page++;
1769        }
1770
1771        return ret;
1772}
1773
1774#ifdef CONFIG_BLOCK
1775static int read_partial_message_bio(struct ceph_connection *con,
1776                                    struct bio **bio_iter, int *bio_seg,
1777                                    unsigned int data_len, bool do_datacrc)
1778{
1779        struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1780        void *p;
1781        int ret, left;
1782
1783        left = min((int)(data_len - con->in_msg_pos.data_pos),
1784                   (int)(bv->bv_len - con->in_msg_pos.page_pos));
1785
1786        p = kmap(bv->bv_page) + bv->bv_offset;
1787
1788        ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1789                               left);
1790        if (ret > 0 && do_datacrc)
1791                con->in_data_crc =
1792                        crc32c(con->in_data_crc,
1793                                  p + con->in_msg_pos.page_pos, ret);
1794        kunmap(bv->bv_page);
1795        if (ret <= 0)
1796                return ret;
1797        con->in_msg_pos.data_pos += ret;
1798        con->in_msg_pos.page_pos += ret;
1799        if (con->in_msg_pos.page_pos == bv->bv_len) {
1800                con->in_msg_pos.page_pos = 0;
1801                iter_bio_next(bio_iter, bio_seg);
1802        }
1803
1804        return ret;
1805}
1806#endif
1807
1808/*
1809 * read (part of) a message.
1810 */
1811static int read_partial_message(struct ceph_connection *con)
1812{
1813        struct ceph_msg *m = con->in_msg;
1814        int size;
1815        int end;
1816        int ret;
1817        unsigned int front_len, middle_len, data_len;
1818        bool do_datacrc = !con->msgr->nocrc;
1819        u64 seq;
1820        u32 crc;
1821
1822        dout("read_partial_message con %p msg %p\n", con, m);
1823
1824        /* header */
1825        size = sizeof (con->in_hdr);
1826        end = size;
1827        ret = read_partial(con, end, size, &con->in_hdr);
1828        if (ret <= 0)
1829                return ret;
1830
1831        crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1832        if (cpu_to_le32(crc) != con->in_hdr.crc) {
1833                pr_err("read_partial_message bad hdr "
1834                       " crc %u != expected %u\n",
1835                       crc, con->in_hdr.crc);
1836                return -EBADMSG;
1837        }
1838
1839        front_len = le32_to_cpu(con->in_hdr.front_len);
1840        if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1841                return -EIO;
1842        middle_len = le32_to_cpu(con->in_hdr.middle_len);
1843        if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1844                return -EIO;
1845        data_len = le32_to_cpu(con->in_hdr.data_len);
1846        if (data_len > CEPH_MSG_MAX_DATA_LEN)
1847                return -EIO;
1848
1849        /* verify seq# */
1850        seq = le64_to_cpu(con->in_hdr.seq);
1851        if ((s64)seq - (s64)con->in_seq < 1) {
1852                pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1853                        ENTITY_NAME(con->peer_name),
1854                        ceph_pr_addr(&con->peer_addr.in_addr),
1855                        seq, con->in_seq + 1);
1856                con->in_base_pos = -front_len - middle_len - data_len -
1857                        sizeof(m->footer);
1858                con->in_tag = CEPH_MSGR_TAG_READY;
1859                return 0;
1860        } else if ((s64)seq - (s64)con->in_seq > 1) {
1861                pr_err("read_partial_message bad seq %lld expected %lld\n",
1862                       seq, con->in_seq + 1);
1863                con->error_msg = "bad message sequence # for incoming message";
1864                return -EBADMSG;
1865        }
1866
1867        /* allocate message? */
1868        if (!con->in_msg) {
1869                int skip = 0;
1870
1871                dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1872                     con->in_hdr.front_len, con->in_hdr.data_len);
1873                ret = ceph_con_in_msg_alloc(con, &skip);
1874                if (ret < 0)
1875                        return ret;
1876                if (skip) {
1877                        /* skip this message */
1878                        dout("alloc_msg said skip message\n");
1879                        BUG_ON(con->in_msg);
1880                        con->in_base_pos = -front_len - middle_len - data_len -
1881                                sizeof(m->footer);
1882                        con->in_tag = CEPH_MSGR_TAG_READY;
1883                        con->in_seq++;
1884                        return 0;
1885                }
1886
1887                BUG_ON(!con->in_msg);
1888                BUG_ON(con->in_msg->con != con);
1889                m = con->in_msg;
1890                m->front.iov_len = 0;    /* haven't read it yet */
1891                if (m->middle)
1892                        m->middle->vec.iov_len = 0;
1893
1894                con->in_msg_pos.page = 0;
1895                if (m->pages)
1896                        con->in_msg_pos.page_pos = m->page_alignment;
1897                else
1898                        con->in_msg_pos.page_pos = 0;
1899                con->in_msg_pos.data_pos = 0;
1900
1901#ifdef CONFIG_BLOCK
1902                if (m->bio)
1903                        init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1904#endif
1905        }
1906
1907        /* front */
1908        ret = read_partial_message_section(con, &m->front, front_len,
1909                                           &con->in_front_crc);
1910        if (ret <= 0)
1911                return ret;
1912
1913        /* middle */
1914        if (m->middle) {
1915                ret = read_partial_message_section(con, &m->middle->vec,
1916                                                   middle_len,
1917                                                   &con->in_middle_crc);
1918                if (ret <= 0)
1919                        return ret;
1920        }
1921
1922        /* (page) data */
1923        while (con->in_msg_pos.data_pos < data_len) {
1924                if (m->pages) {
1925                        ret = read_partial_message_pages(con, m->pages,
1926                                                 data_len, do_datacrc);
1927                        if (ret <= 0)
1928                                return ret;
1929#ifdef CONFIG_BLOCK
1930                } else if (m->bio) {
1931                        BUG_ON(!m->bio_iter);
1932                        ret = read_partial_message_bio(con,
1933                                                 &m->bio_iter, &m->bio_seg,
1934                                                 data_len, do_datacrc);
1935                        if (ret <= 0)
1936                                return ret;
1937#endif
1938                } else {
1939                        BUG_ON(1);
1940                }
1941        }
1942
1943        /* footer */
1944        size = sizeof (m->footer);
1945        end += size;
1946        ret = read_partial(con, end, size, &m->footer);
1947        if (ret <= 0)
1948                return ret;
1949
1950        dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1951             m, front_len, m->footer.front_crc, middle_len,
1952             m->footer.middle_crc, data_len, m->footer.data_crc);
1953
1954        /* crc ok? */
1955        if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1956                pr_err("read_partial_message %p front crc %u != exp. %u\n",
1957                       m, con->in_front_crc, m->footer.front_crc);
1958                return -EBADMSG;
1959        }
1960        if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1961                pr_err("read_partial_message %p middle crc %u != exp %u\n",
1962                       m, con->in_middle_crc, m->footer.middle_crc);
1963                return -EBADMSG;
1964        }
1965        if (do_datacrc &&
1966            (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1967            con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1968                pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1969                       con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1970                return -EBADMSG;
1971        }
1972
1973        return 1; /* done! */
1974}
1975
1976/*
1977 * Process message.  This happens in the worker thread.  The callback should
1978 * be careful not to do anything that waits on other incoming messages or it
1979 * may deadlock.
1980 */
1981static void process_message(struct ceph_connection *con)
1982{
1983        struct ceph_msg *msg;
1984
1985        BUG_ON(con->in_msg->con != con);
1986        con->in_msg->con = NULL;
1987        msg = con->in_msg;
1988        con->in_msg = NULL;
1989        con->ops->put(con);
1990
1991        /* if first message, set peer_name */
1992        if (con->peer_name.type == 0)
1993                con->peer_name = msg->hdr.src;
1994
1995        con->in_seq++;
1996        mutex_unlock(&con->mutex);
1997
1998        dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1999             msg, le64_to_cpu(msg->hdr.seq),
2000             ENTITY_NAME(msg->hdr.src),
2001             le16_to_cpu(msg->hdr.type),
2002             ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2003             le32_to_cpu(msg->hdr.front_len),
2004             le32_to_cpu(msg->hdr.data_len),
2005             con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2006        con->ops->dispatch(con, msg);
2007
2008        mutex_lock(&con->mutex);
2009}
2010
2011
2012/*
2013 * Write something to the socket.  Called in a worker thread when the
2014 * socket appears to be writeable and we have something ready to send.
2015 */
2016static int try_write(struct ceph_connection *con)
2017{
2018        int ret = 1;
2019
2020        dout("try_write start %p state %lu\n", con, con->state);
2021
2022more:
2023        dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2024
2025        /* open the socket first? */
2026        if (con->state == CON_STATE_PREOPEN) {
2027                BUG_ON(con->sock);
2028                con->state = CON_STATE_CONNECTING;
2029
2030                con_out_kvec_reset(con);
2031                prepare_write_banner(con);
2032                prepare_read_banner(con);
2033
2034                BUG_ON(con->in_msg);
2035                con->in_tag = CEPH_MSGR_TAG_READY;
2036                dout("try_write initiating connect on %p new state %lu\n",
2037                     con, con->state);
2038                ret = ceph_tcp_connect(con);
2039                if (ret < 0) {
2040                        con->error_msg = "connect error";
2041                        goto out;
2042                }
2043        }
2044
2045more_kvec:
2046        /* kvec data queued? */
2047        if (con->out_skip) {
2048                ret = write_partial_skip(con);
2049                if (ret <= 0)
2050                        goto out;
2051        }
2052        if (con->out_kvec_left) {
2053                ret = write_partial_kvec(con);
2054                if (ret <= 0)
2055                        goto out;
2056        }
2057
2058        /* msg pages? */
2059        if (con->out_msg) {
2060                if (con->out_msg_done) {
2061                        ceph_msg_put(con->out_msg);
2062                        con->out_msg = NULL;   /* we're done with this one */
2063                        goto do_next;
2064                }
2065
2066                ret = write_partial_msg_pages(con);
2067                if (ret == 1)
2068                        goto more_kvec;  /* we need to send the footer, too! */
2069                if (ret == 0)
2070                        goto out;
2071                if (ret < 0) {
2072                        dout("try_write write_partial_msg_pages err %d\n",
2073                             ret);
2074                        goto out;
2075                }
2076        }
2077
2078do_next:
2079        if (con->state == CON_STATE_OPEN) {
2080                /* is anything else pending? */
2081                if (!list_empty(&con->out_queue)) {
2082                        prepare_write_message(con);
2083                        goto more;
2084                }
2085                if (con->in_seq > con->in_seq_acked) {
2086                        prepare_write_ack(con);
2087                        goto more;
2088                }
2089                if (test_and_clear_bit(CON_FLAG_KEEPALIVE_PENDING,
2090                                       &con->flags)) {
2091                        prepare_write_keepalive(con);
2092                        goto more;
2093                }
2094        }
2095
2096        /* Nothing to do! */
2097        clear_bit(CON_FLAG_WRITE_PENDING, &con->flags);
2098        dout("try_write nothing else to write.\n");
2099        ret = 0;
2100out:
2101        dout("try_write done on %p ret %d\n", con, ret);
2102        return ret;
2103}
2104
2105
2106
2107/*
2108 * Read what we can from the socket.
2109 */
2110static int try_read(struct ceph_connection *con)
2111{
2112        int ret = -1;
2113
2114more:
2115        dout("try_read start on %p state %lu\n", con, con->state);
2116        if (con->state != CON_STATE_CONNECTING &&
2117            con->state != CON_STATE_NEGOTIATING &&
2118            con->state != CON_STATE_OPEN)
2119                return 0;
2120
2121        BUG_ON(!con->sock);
2122
2123        dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2124             con->in_base_pos);
2125
2126        if (con->state == CON_STATE_CONNECTING) {
2127                dout("try_read connecting\n");
2128                ret = read_partial_banner(con);
2129                if (ret <= 0)
2130                        goto out;
2131                ret = process_banner(con);
2132                if (ret < 0)
2133                        goto out;
2134
2135                BUG_ON(con->state != CON_STATE_CONNECTING);
2136                con->state = CON_STATE_NEGOTIATING;
2137
2138                /*
2139                 * Received banner is good, exchange connection info.
2140                 * Do not reset out_kvec, as sending our banner raced
2141                 * with receiving peer banner after connect completed.
2142                 */
2143                ret = prepare_write_connect(con);
2144                if (ret < 0)
2145                        goto out;
2146                prepare_read_connect(con);
2147
2148                /* Send connection info before awaiting response */
2149                goto out;
2150        }
2151
2152        if (con->state == CON_STATE_NEGOTIATING) {
2153                dout("try_read negotiating\n");
2154                ret = read_partial_connect(con);
2155                if (ret <= 0)
2156                        goto out;
2157                ret = process_connect(con);
2158                if (ret < 0)
2159                        goto out;
2160                goto more;
2161        }
2162
2163        BUG_ON(con->state != CON_STATE_OPEN);
2164
2165        if (con->in_base_pos < 0) {
2166                /*
2167                 * skipping + discarding content.
2168                 *
2169                 * FIXME: there must be a better way to do this!
2170                 */
2171                static char buf[SKIP_BUF_SIZE];
2172                int skip = min((int) sizeof (buf), -con->in_base_pos);
2173
2174                dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2175                ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2176                if (ret <= 0)
2177                        goto out;
2178                con->in_base_pos += ret;
2179                if (con->in_base_pos)
2180                        goto more;
2181        }
2182        if (con->in_tag == CEPH_MSGR_TAG_READY) {
2183                /*
2184                 * what's next?
2185                 */
2186                ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2187                if (ret <= 0)
2188                        goto out;
2189                dout("try_read got tag %d\n", (int)con->in_tag);
2190                switch (con->in_tag) {
2191                case CEPH_MSGR_TAG_MSG:
2192                        prepare_read_message(con);
2193                        break;
2194                case CEPH_MSGR_TAG_ACK:
2195                        prepare_read_ack(con);
2196                        break;
2197                case CEPH_MSGR_TAG_CLOSE:
2198                        con_close_socket(con);
2199                        con->state = CON_STATE_CLOSED;
2200                        goto out;
2201                default:
2202                        goto bad_tag;
2203                }
2204        }
2205        if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2206                ret = read_partial_message(con);
2207                if (ret <= 0) {
2208                        switch (ret) {
2209                        case -EBADMSG:
2210                                con->error_msg = "bad crc";
2211                                ret = -EIO;
2212                                break;
2213                        case -EIO:
2214                                con->error_msg = "io error";
2215                                break;
2216                        }
2217                        goto out;
2218                }
2219                if (con->in_tag == CEPH_MSGR_TAG_READY)
2220                        goto more;
2221                process_message(con);
2222                if (con->state == CON_STATE_OPEN)
2223                        prepare_read_tag(con);
2224                goto more;
2225        }
2226        if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2227                ret = read_partial_ack(con);
2228                if (ret <= 0)
2229                        goto out;
2230                process_ack(con);
2231                goto more;
2232        }
2233
2234out:
2235        dout("try_read done on %p ret %d\n", con, ret);
2236        return ret;
2237
2238bad_tag:
2239        pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2240        con->error_msg = "protocol error, garbage tag";
2241        ret = -1;
2242        goto out;
2243}
2244
2245
2246/*
2247 * Atomically queue work on a connection.  Bump @con reference to
2248 * avoid races with connection teardown.
2249 */
2250static void queue_con(struct ceph_connection *con)
2251{
2252        if (!con->ops->get(con)) {
2253                dout("queue_con %p ref count 0\n", con);
2254                return;
2255        }
2256
2257        if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
2258                dout("queue_con %p - already queued\n", con);
2259                con->ops->put(con);
2260        } else {
2261                dout("queue_con %p\n", con);
2262        }
2263}
2264
2265/*
2266 * Do some work on a connection.  Drop a connection ref when we're done.
2267 */
2268static void con_work(struct work_struct *work)
2269{
2270        struct ceph_connection *con = container_of(work, struct ceph_connection,
2271                                                   work.work);
2272        int ret;
2273
2274        mutex_lock(&con->mutex);
2275restart:
2276        if (test_and_clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags)) {
2277                switch (con->state) {
2278                case CON_STATE_CONNECTING:
2279                        con->error_msg = "connection failed";
2280                        break;
2281                case CON_STATE_NEGOTIATING:
2282                        con->error_msg = "negotiation failed";
2283                        break;
2284                case CON_STATE_OPEN:
2285                        con->error_msg = "socket closed";
2286                        break;
2287                default:
2288                        dout("unrecognized con state %d\n", (int)con->state);
2289                        con->error_msg = "unrecognized con state";
2290                        BUG();
2291                }
2292                goto fault;
2293        }
2294
2295        if (test_and_clear_bit(CON_FLAG_BACKOFF, &con->flags)) {
2296                dout("con_work %p backing off\n", con);
2297                if (queue_delayed_work(ceph_msgr_wq, &con->work,
2298                                       round_jiffies_relative(con->delay))) {
2299                        dout("con_work %p backoff %lu\n", con, con->delay);
2300                        mutex_unlock(&con->mutex);
2301                        return;
2302                } else {
2303                        dout("con_work %p FAILED to back off %lu\n", con,
2304                             con->delay);
2305                        set_bit(CON_FLAG_BACKOFF, &con->flags);
2306                }
2307                goto done;
2308        }
2309
2310        if (con->state == CON_STATE_STANDBY) {
2311                dout("con_work %p STANDBY\n", con);
2312                goto done;
2313        }
2314        if (con->state == CON_STATE_CLOSED) {
2315                dout("con_work %p CLOSED\n", con);
2316                BUG_ON(con->sock);
2317                goto done;
2318        }
2319        if (con->state == CON_STATE_PREOPEN) {
2320                dout("con_work OPENING\n");
2321                BUG_ON(con->sock);
2322        }
2323
2324        ret = try_read(con);
2325        if (ret == -EAGAIN)
2326                goto restart;
2327        if (ret < 0) {
2328                con->error_msg = "socket error on read";
2329                goto fault;
2330        }
2331
2332        ret = try_write(con);
2333        if (ret == -EAGAIN)
2334                goto restart;
2335        if (ret < 0) {
2336                con->error_msg = "socket error on write";
2337                goto fault;
2338        }
2339
2340done:
2341        mutex_unlock(&con->mutex);
2342done_unlocked:
2343        con->ops->put(con);
2344        return;
2345
2346fault:
2347        ceph_fault(con);     /* error/fault path */
2348        goto done_unlocked;
2349}
2350
2351
2352/*
2353 * Generic error/fault handler.  A retry mechanism is used with
2354 * exponential backoff
2355 */
2356static void ceph_fault(struct ceph_connection *con)
2357        __releases(con->mutex)
2358{
2359        pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2360               ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2361        dout("fault %p state %lu to peer %s\n",
2362             con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2363
2364        BUG_ON(con->state != CON_STATE_CONNECTING &&
2365               con->state != CON_STATE_NEGOTIATING &&
2366               con->state != CON_STATE_OPEN);
2367
2368        con_close_socket(con);
2369
2370        if (test_bit(CON_FLAG_LOSSYTX, &con->flags)) {
2371                dout("fault on LOSSYTX channel, marking CLOSED\n");
2372                con->state = CON_STATE_CLOSED;
2373                goto out_unlock;
2374        }
2375
2376        if (con->in_msg) {
2377                BUG_ON(con->in_msg->con != con);
2378                con->in_msg->con = NULL;
2379                ceph_msg_put(con->in_msg);
2380                con->in_msg = NULL;
2381                con->ops->put(con);
2382        }
2383
2384        /* Requeue anything that hasn't been acked */
2385        list_splice_init(&con->out_sent, &con->out_queue);
2386
2387        /* If there are no messages queued or keepalive pending, place
2388         * the connection in a STANDBY state */
2389        if (list_empty(&con->out_queue) &&
2390            !test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags)) {
2391                dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2392                clear_bit(CON_FLAG_WRITE_PENDING, &con->flags);
2393                con->state = CON_STATE_STANDBY;
2394        } else {
2395                /* retry after a delay. */
2396                con->state = CON_STATE_PREOPEN;
2397                if (con->delay == 0)
2398                        con->delay = BASE_DELAY_INTERVAL;
2399                else if (con->delay < MAX_DELAY_INTERVAL)
2400                        con->delay *= 2;
2401                con->ops->get(con);
2402                if (queue_delayed_work(ceph_msgr_wq, &con->work,
2403                                       round_jiffies_relative(con->delay))) {
2404                        dout("fault queued %p delay %lu\n", con, con->delay);
2405                } else {
2406                        con->ops->put(con);
2407                        dout("fault failed to queue %p delay %lu, backoff\n",
2408                             con, con->delay);
2409                        /*
2410                         * In many cases we see a socket state change
2411                         * while con_work is running and end up
2412                         * queuing (non-delayed) work, such that we
2413                         * can't backoff with a delay.  Set a flag so
2414                         * that when con_work restarts we schedule the
2415                         * delay then.
2416                         */
2417                        set_bit(CON_FLAG_BACKOFF, &con->flags);
2418                }
2419        }
2420
2421out_unlock:
2422        mutex_unlock(&con->mutex);
2423        /*
2424         * in case we faulted due to authentication, invalidate our
2425         * current tickets so that we can get new ones.
2426         */
2427        if (con->auth_retry && con->ops->invalidate_authorizer) {
2428                dout("calling invalidate_authorizer()\n");
2429                con->ops->invalidate_authorizer(con);
2430        }
2431
2432        if (con->ops->fault)
2433                con->ops->fault(con);
2434}
2435
2436
2437
2438/*
2439 * initialize a new messenger instance
2440 */
2441void ceph_messenger_init(struct ceph_messenger *msgr,
2442                        struct ceph_entity_addr *myaddr,
2443                        u32 supported_features,
2444                        u32 required_features,
2445                        bool nocrc)
2446{
2447        msgr->supported_features = supported_features;
2448        msgr->required_features = required_features;
2449
2450        spin_lock_init(&msgr->global_seq_lock);
2451
2452        if (myaddr)
2453                msgr->inst.addr = *myaddr;
2454
2455        /* select a random nonce */
2456        msgr->inst.addr.type = 0;
2457        get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2458        encode_my_addr(msgr);
2459        msgr->nocrc = nocrc;
2460
2461        atomic_set(&msgr->stopping, 0);
2462
2463        dout("%s %p\n", __func__, msgr);
2464}
2465EXPORT_SYMBOL(ceph_messenger_init);
2466
2467static void clear_standby(struct ceph_connection *con)
2468{
2469        /* come back from STANDBY? */
2470        if (con->state == CON_STATE_STANDBY) {
2471                dout("clear_standby %p and ++connect_seq\n", con);
2472                con->state = CON_STATE_PREOPEN;
2473                con->connect_seq++;
2474                WARN_ON(test_bit(CON_FLAG_WRITE_PENDING, &con->flags));
2475                WARN_ON(test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags));
2476        }
2477}
2478
2479/*
2480 * Queue up an outgoing message on the given connection.
2481 */
2482void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2483{
2484        /* set src+dst */
2485        msg->hdr.src = con->msgr->inst.name;
2486        BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2487        msg->needs_out_seq = true;
2488
2489        mutex_lock(&con->mutex);
2490
2491        if (con->state == CON_STATE_CLOSED) {
2492                dout("con_send %p closed, dropping %p\n", con, msg);
2493                ceph_msg_put(msg);
2494                mutex_unlock(&con->mutex);
2495                return;
2496        }
2497
2498        BUG_ON(msg->con != NULL);
2499        msg->con = con->ops->get(con);
2500        BUG_ON(msg->con == NULL);
2501
2502        BUG_ON(!list_empty(&msg->list_head));
2503        list_add_tail(&msg->list_head, &con->out_queue);
2504        dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2505             ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2506             ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2507             le32_to_cpu(msg->hdr.front_len),
2508             le32_to_cpu(msg->hdr.middle_len),
2509             le32_to_cpu(msg->hdr.data_len));
2510
2511        clear_standby(con);
2512        mutex_unlock(&con->mutex);
2513
2514        /* if there wasn't anything waiting to send before, queue
2515         * new work */
2516        if (test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0)
2517                queue_con(con);
2518}
2519EXPORT_SYMBOL(ceph_con_send);
2520
2521/*
2522 * Revoke a message that was previously queued for send
2523 */
2524void ceph_msg_revoke(struct ceph_msg *msg)
2525{
2526        struct ceph_connection *con = msg->con;
2527
2528        if (!con)
2529                return;         /* Message not in our possession */
2530
2531        mutex_lock(&con->mutex);
2532        if (!list_empty(&msg->list_head)) {
2533                dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2534                list_del_init(&msg->list_head);
2535                BUG_ON(msg->con == NULL);
2536                msg->con->ops->put(msg->con);
2537                msg->con = NULL;
2538                msg->hdr.seq = 0;
2539
2540                ceph_msg_put(msg);
2541        }
2542        if (con->out_msg == msg) {
2543                dout("%s %p msg %p - was sending\n", __func__, con, msg);
2544                con->out_msg = NULL;
2545                if (con->out_kvec_is_msg) {
2546                        con->out_skip = con->out_kvec_bytes;
2547                        con->out_kvec_is_msg = false;
2548                }
2549                msg->hdr.seq = 0;
2550
2551                ceph_msg_put(msg);
2552        }
2553        mutex_unlock(&con->mutex);
2554}
2555
2556/*
2557 * Revoke a message that we may be reading data into
2558 */
2559void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2560{
2561        struct ceph_connection *con;
2562
2563        BUG_ON(msg == NULL);
2564        if (!msg->con) {
2565                dout("%s msg %p null con\n", __func__, msg);
2566
2567                return;         /* Message not in our possession */
2568        }
2569
2570        con = msg->con;
2571        mutex_lock(&con->mutex);
2572        if (con->in_msg == msg) {
2573                unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2574                unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2575                unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2576
2577                /* skip rest of message */
2578                dout("%s %p msg %p revoked\n", __func__, con, msg);
2579                con->in_base_pos = con->in_base_pos -
2580                                sizeof(struct ceph_msg_header) -
2581                                front_len -
2582                                middle_len -
2583                                data_len -
2584                                sizeof(struct ceph_msg_footer);
2585                ceph_msg_put(con->in_msg);
2586                con->in_msg = NULL;
2587                con->in_tag = CEPH_MSGR_TAG_READY;
2588                con->in_seq++;
2589        } else {
2590                dout("%s %p in_msg %p msg %p no-op\n",
2591                     __func__, con, con->in_msg, msg);
2592        }
2593        mutex_unlock(&con->mutex);
2594}
2595
2596/*
2597 * Queue a keepalive byte to ensure the tcp connection is alive.
2598 */
2599void ceph_con_keepalive(struct ceph_connection *con)
2600{
2601        dout("con_keepalive %p\n", con);
2602        mutex_lock(&con->mutex);
2603        clear_standby(con);
2604        mutex_unlock(&con->mutex);
2605        if (test_and_set_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags) == 0 &&
2606            test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0)
2607                queue_con(con);
2608}
2609EXPORT_SYMBOL(ceph_con_keepalive);
2610
2611
2612/*
2613 * construct a new message with given type, size
2614 * the new msg has a ref count of 1.
2615 */
2616struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2617                              bool can_fail)
2618{
2619        struct ceph_msg *m;
2620
2621        m = kmalloc(sizeof(*m), flags);
2622        if (m == NULL)
2623                goto out;
2624        kref_init(&m->kref);
2625
2626        m->con = NULL;
2627        INIT_LIST_HEAD(&m->list_head);
2628
2629        m->hdr.tid = 0;
2630        m->hdr.type = cpu_to_le16(type);
2631        m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2632        m->hdr.version = 0;
2633        m->hdr.front_len = cpu_to_le32(front_len);
2634        m->hdr.middle_len = 0;
2635        m->hdr.data_len = 0;
2636        m->hdr.data_off = 0;
2637        m->hdr.reserved = 0;
2638        m->footer.front_crc = 0;
2639        m->footer.middle_crc = 0;
2640        m->footer.data_crc = 0;
2641        m->footer.flags = 0;
2642        m->front_max = front_len;
2643        m->front_is_vmalloc = false;
2644        m->more_to_follow = false;
2645        m->ack_stamp = 0;
2646        m->pool = NULL;
2647
2648        /* middle */
2649        m->middle = NULL;
2650
2651        /* data */
2652        m->nr_pages = 0;
2653        m->page_alignment = 0;
2654        m->pages = NULL;
2655        m->pagelist = NULL;
2656        m->bio = NULL;
2657        m->bio_iter = NULL;
2658        m->bio_seg = 0;
2659        m->trail = NULL;
2660
2661        /* front */
2662        if (front_len) {
2663                if (front_len > PAGE_CACHE_SIZE) {
2664                        m->front.iov_base = __vmalloc(front_len, flags,
2665                                                      PAGE_KERNEL);
2666                        m->front_is_vmalloc = true;
2667                } else {
2668                        m->front.iov_base = kmalloc(front_len, flags);
2669                }
2670                if (m->front.iov_base == NULL) {
2671                        dout("ceph_msg_new can't allocate %d bytes\n",
2672                             front_len);
2673                        goto out2;
2674                }
2675        } else {
2676                m->front.iov_base = NULL;
2677        }
2678        m->front.iov_len = front_len;
2679
2680        dout("ceph_msg_new %p front %d\n", m, front_len);
2681        return m;
2682
2683out2:
2684        ceph_msg_put(m);
2685out:
2686        if (!can_fail) {
2687                pr_err("msg_new can't create type %d front %d\n", type,
2688                       front_len);
2689                WARN_ON(1);
2690        } else {
2691                dout("msg_new can't create type %d front %d\n", type,
2692                     front_len);
2693        }
2694        return NULL;
2695}
2696EXPORT_SYMBOL(ceph_msg_new);
2697
2698/*
2699 * Allocate "middle" portion of a message, if it is needed and wasn't
2700 * allocated by alloc_msg.  This allows us to read a small fixed-size
2701 * per-type header in the front and then gracefully fail (i.e.,
2702 * propagate the error to the caller based on info in the front) when
2703 * the middle is too large.
2704 */
2705static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2706{
2707        int type = le16_to_cpu(msg->hdr.type);
2708        int middle_len = le32_to_cpu(msg->hdr.middle_len);
2709
2710        dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2711             ceph_msg_type_name(type), middle_len);
2712        BUG_ON(!middle_len);
2713        BUG_ON(msg->middle);
2714
2715        msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2716        if (!msg->middle)
2717                return -ENOMEM;
2718        return 0;
2719}
2720
2721/*
2722 * Allocate a message for receiving an incoming message on a
2723 * connection, and save the result in con->in_msg.  Uses the
2724 * connection's private alloc_msg op if available.
2725 *
2726 * Returns 0 on success, or a negative error code.
2727 *
2728 * On success, if we set *skip = 1:
2729 *  - the next message should be skipped and ignored.
2730 *  - con->in_msg == NULL
2731 * or if we set *skip = 0:
2732 *  - con->in_msg is non-null.
2733 * On error (ENOMEM, EAGAIN, ...),
2734 *  - con->in_msg == NULL
2735 */
2736static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2737{
2738        struct ceph_msg_header *hdr = &con->in_hdr;
2739        int type = le16_to_cpu(hdr->type);
2740        int front_len = le32_to_cpu(hdr->front_len);
2741        int middle_len = le32_to_cpu(hdr->middle_len);
2742        int ret = 0;
2743
2744        BUG_ON(con->in_msg != NULL);
2745
2746        if (con->ops->alloc_msg) {
2747                struct ceph_msg *msg;
2748
2749                mutex_unlock(&con->mutex);
2750                msg = con->ops->alloc_msg(con, hdr, skip);
2751                mutex_lock(&con->mutex);
2752                if (con->state != CON_STATE_OPEN) {
2753                        if (msg)
2754                                ceph_msg_put(msg);
2755                        return -EAGAIN;
2756                }
2757                con->in_msg = msg;
2758                if (con->in_msg) {
2759                        con->in_msg->con = con->ops->get(con);
2760                        BUG_ON(con->in_msg->con == NULL);
2761                }
2762                if (*skip) {
2763                        con->in_msg = NULL;
2764                        return 0;
2765                }
2766                if (!con->in_msg) {
2767                        con->error_msg =
2768                                "error allocating memory for incoming message";
2769                        return -ENOMEM;
2770                }
2771        }
2772        if (!con->in_msg) {
2773                con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2774                if (!con->in_msg) {
2775                        pr_err("unable to allocate msg type %d len %d\n",
2776                               type, front_len);
2777                        return -ENOMEM;
2778                }
2779                con->in_msg->con = con->ops->get(con);
2780                BUG_ON(con->in_msg->con == NULL);
2781                con->in_msg->page_alignment = le16_to_cpu(hdr->data_off);
2782        }
2783        memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2784
2785        if (middle_len && !con->in_msg->middle) {
2786                ret = ceph_alloc_middle(con, con->in_msg);
2787                if (ret < 0) {
2788                        ceph_msg_put(con->in_msg);
2789                        con->in_msg = NULL;
2790                }
2791        }
2792
2793        return ret;
2794}
2795
2796
2797/*
2798 * Free a generically kmalloc'd message.
2799 */
2800void ceph_msg_kfree(struct ceph_msg *m)
2801{
2802        dout("msg_kfree %p\n", m);
2803        if (m->front_is_vmalloc)
2804                vfree(m->front.iov_base);
2805        else
2806                kfree(m->front.iov_base);
2807        kfree(m);
2808}
2809
2810/*
2811 * Drop a msg ref.  Destroy as needed.
2812 */
2813void ceph_msg_last_put(struct kref *kref)
2814{
2815        struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2816
2817        dout("ceph_msg_put last one on %p\n", m);
2818        WARN_ON(!list_empty(&m->list_head));
2819
2820        /* drop middle, data, if any */
2821        if (m->middle) {
2822                ceph_buffer_put(m->middle);
2823                m->middle = NULL;
2824        }
2825        m->nr_pages = 0;
2826        m->pages = NULL;
2827
2828        if (m->pagelist) {
2829                ceph_pagelist_release(m->pagelist);
2830                kfree(m->pagelist);
2831                m->pagelist = NULL;
2832        }
2833
2834        m->trail = NULL;
2835
2836        if (m->pool)
2837                ceph_msgpool_put(m->pool, m);
2838        else
2839                ceph_msg_kfree(m);
2840}
2841EXPORT_SYMBOL(ceph_msg_last_put);
2842
2843void ceph_msg_dump(struct ceph_msg *msg)
2844{
2845        pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2846                 msg->front_max, msg->nr_pages);
2847        print_hex_dump(KERN_DEBUG, "header: ",
2848                       DUMP_PREFIX_OFFSET, 16, 1,
2849                       &msg->hdr, sizeof(msg->hdr), true);
2850        print_hex_dump(KERN_DEBUG, " front: ",
2851                       DUMP_PREFIX_OFFSET, 16, 1,
2852                       msg->front.iov_base, msg->front.iov_len, true);
2853        if (msg->middle)
2854                print_hex_dump(KERN_DEBUG, "middle: ",
2855                               DUMP_PREFIX_OFFSET, 16, 1,
2856                               msg->middle->vec.iov_base,
2857                               msg->middle->vec.iov_len, true);
2858        print_hex_dump(KERN_DEBUG, "footer: ",
2859                       DUMP_PREFIX_OFFSET, 16, 1,
2860                       &msg->footer, sizeof(msg->footer), true);
2861}
2862EXPORT_SYMBOL(ceph_msg_dump);
2863
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.