net/ceph/messenger_v1.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/bvec.h>
   5 #include <linux/crc32c.h>
   6 #include <linux/net.h>
   7 #include <linux/socket.h>
   8 #include <net/sock.h>
   9
  10 #include <linux/ceph/ceph_features.h>
  11 #include <linux/ceph/decode.h>
  12 #include <linux/ceph/libceph.h>
  13 #include <linux/ceph/messenger.h>
  14
  15 /* static tag bytes (protocol control messages) */
  16 static char tag_msg = CEPH_MSGR_TAG_MSG;
  17 static char tag_ack = CEPH_MSGR_TAG_ACK;
  18 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
  19 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
  20
  21 /*
  22  * If @buf is NULL, discard up to @len bytes.
  23  */
  24 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
  25 {
  26         struct kvec iov = {buf, len};
  27         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
  28         int r;
  29
  30         if (!buf)
  31                 msg.msg_flags |= MSG_TRUNC;
  32
  33         iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
  34         r = sock_recvmsg(sock, &msg, msg.msg_flags);
  35         if (r == -EAGAIN)
  36                 r = 0;
  37         return r;
  38 }
  39
  40 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
  41                      int page_offset, size_t length)
  42 {
  43         struct bio_vec bvec = {
  44                 .bv_page = page,
  45                 .bv_offset = page_offset,
  46                 .bv_len = length
  47         };
  48         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
  49         int r;
  50
  51         BUG_ON(page_offset + length > PAGE_SIZE);
  52         iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
  53         r = sock_recvmsg(sock, &msg, msg.msg_flags);
  54         if (r == -EAGAIN)
  55                 r = 0;
  56         return r;
  57 }
  58
  59 /*
  60  * write something.  @more is true if caller will be sending more data
  61  * shortly.
  62  */
  63 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
  64                             size_t kvlen, size_t len, bool more)
  65 {
  66         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
  67         int r;
  68
  69         if (more)
  70                 msg.msg_flags |= MSG_MORE;
  71         else
  72                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
  73
  74         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
  75         if (r == -EAGAIN)
  76                 r = 0;
  77         return r;
  78 }
  79
  80 /*
  81  * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
  82  */
  83 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
  84                              int offset, size_t size, int more)
  85 {
  86         ssize_t (*sendpage)(struct socket *sock, struct page *page,
  87                             int offset, size_t size, int flags);
  88         int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
  89         int ret;
  90
  91         /*
  92          * sendpage cannot properly handle pages with page_count == 0,
  93          * we need to fall back to sendmsg if that's the case.
  94          *
  95          * Same goes for slab pages: skb_can_coalesce() allows
  96          * coalescing neighboring slab objects into a single frag which
  97          * triggers one of hardened usercopy checks.
  98          */
  99         if (sendpage_ok(page))
 100                 sendpage = sock->ops->sendpage;
 101         else
 102                 sendpage = sock_no_sendpage;
 103
 104         ret = sendpage(sock, page, offset, size, flags);
 105         if (ret == -EAGAIN)
 106                 ret = 0;
 107
 108         return ret;
 109 }
 110
 111 static void con_out_kvec_reset(struct ceph_connection *con)
 112 {
 113         BUG_ON(con->v1.out_skip);
 114
 115         con->v1.out_kvec_left = 0;
 116         con->v1.out_kvec_bytes = 0;
 117         con->v1.out_kvec_cur = &con->v1.out_kvec[0];
 118 }
 119
 120 static void con_out_kvec_add(struct ceph_connection *con,
 121                                 size_t size, void *data)
 122 {
 123         int index = con->v1.out_kvec_left;
 124
 125         BUG_ON(con->v1.out_skip);
 126         BUG_ON(index >= ARRAY_SIZE(con->v1.out_kvec));
 127
 128         con->v1.out_kvec[index].iov_len = size;
 129         con->v1.out_kvec[index].iov_base = data;
 130         con->v1.out_kvec_left++;
 131         con->v1.out_kvec_bytes += size;
 132 }
 133
 134 /*
 135  * Chop off a kvec from the end.  Return residual number of bytes for
 136  * that kvec, i.e. how many bytes would have been written if the kvec
 137  * hadn't been nuked.
 138  */
 139 static int con_out_kvec_skip(struct ceph_connection *con)
 140 {
 141         int skip = 0;
 142
 143         if (con->v1.out_kvec_bytes > 0) {
 144                 skip = con->v1.out_kvec_cur[con->v1.out_kvec_left - 1].iov_len;
 145                 BUG_ON(con->v1.out_kvec_bytes < skip);
 146                 BUG_ON(!con->v1.out_kvec_left);
 147                 con->v1.out_kvec_bytes -= skip;
 148                 con->v1.out_kvec_left--;
 149         }
 150
 151         return skip;
 152 }
 153
 154 static size_t sizeof_footer(struct ceph_connection *con)
 155 {
 156         return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
 157             sizeof(struct ceph_msg_footer) :
 158             sizeof(struct ceph_msg_footer_old);
 159 }
 160
 161 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
 162 {
 163         /* Initialize data cursor */
 164
 165         ceph_msg_data_cursor_init(&msg->cursor, msg, data_len);
 166 }
 167
 168 /*
 169  * Prepare footer for currently outgoing message, and finish things
 170  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
 171  */
 172 static void prepare_write_message_footer(struct ceph_connection *con)
 173 {
 174         struct ceph_msg *m = con->out_msg;
 175
 176         m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
 177
 178         dout("prepare_write_message_footer %p\n", con);
 179         con_out_kvec_add(con, sizeof_footer(con), &m->footer);
 180         if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
 181                 if (con->ops->sign_message)
 182                         con->ops->sign_message(m);
 183                 else
 184                         m->footer.sig = 0;
 185         } else {
 186                 m->old_footer.flags = m->footer.flags;
 187         }
 188         con->v1.out_more = m->more_to_follow;
 189         con->v1.out_msg_done = true;
 190 }
 191
 192 /*
 193  * Prepare headers for the next outgoing message.
 194  */
 195 static void prepare_write_message(struct ceph_connection *con)
 196 {
 197         struct ceph_msg *m;
 198         u32 crc;
 199
 200         con_out_kvec_reset(con);
 201         con->v1.out_msg_done = false;
 202
 203         /* Sneak an ack in there first?  If we can get it into the same
 204          * TCP packet that's a good thing. */
 205         if (con->in_seq > con->in_seq_acked) {
 206                 con->in_seq_acked = con->in_seq;
 207                 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
 208                 con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
 209                 con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
 210                         &con->v1.out_temp_ack);
 211         }
 212
 213         ceph_con_get_out_msg(con);
 214         m = con->out_msg;
 215
 216         dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
 217              m, con->out_seq, le16_to_cpu(m->hdr.type),
 218              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
 219              m->data_length);
 220         WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
 221         WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
 222
 223         /* tag + hdr + front + middle */
 224         con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
 225         con_out_kvec_add(con, sizeof(con->v1.out_hdr), &con->v1.out_hdr);
 226         con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
 227
 228         if (m->middle)
 229                 con_out_kvec_add(con, m->middle->vec.iov_len,
 230                         m->middle->vec.iov_base);
 231
 232         /* fill in hdr crc and finalize hdr */
 233         crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
 234         con->out_msg->hdr.crc = cpu_to_le32(crc);
 235         memcpy(&con->v1.out_hdr, &con->out_msg->hdr, sizeof(con->v1.out_hdr));
 236
 237         /* fill in front and middle crc, footer */
 238         crc = crc32c(0, m->front.iov_base, m->front.iov_len);
 239         con->out_msg->footer.front_crc = cpu_to_le32(crc);
 240         if (m->middle) {
 241                 crc = crc32c(0, m->middle->vec.iov_base,
 242                                 m->middle->vec.iov_len);
 243                 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
 244         } else
 245                 con->out_msg->footer.middle_crc = 0;
 246         dout("%s front_crc %u middle_crc %u\n", __func__,
 247              le32_to_cpu(con->out_msg->footer.front_crc),
 248              le32_to_cpu(con->out_msg->footer.middle_crc));
 249         con->out_msg->footer.flags = 0;
 250
 251         /* is there a data payload? */
 252         con->out_msg->footer.data_crc = 0;
 253         if (m->data_length) {
 254                 prepare_message_data(con->out_msg, m->data_length);
 255                 con->v1.out_more = 1;  /* data + footer will follow */
 256         } else {
 257                 /* no, queue up footer too and be done */
 258                 prepare_write_message_footer(con);
 259         }
 260
 261         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 262 }
 263
 264 /*
 265  * Prepare an ack.
 266  */
 267 static void prepare_write_ack(struct ceph_connection *con)
 268 {
 269         dout("prepare_write_ack %p %llu -> %llu\n", con,
 270              con->in_seq_acked, con->in_seq);
 271         con->in_seq_acked = con->in_seq;
 272
 273         con_out_kvec_reset(con);
 274
 275         con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
 276
 277         con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
 278         con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
 279                          &con->v1.out_temp_ack);
 280
 281         con->v1.out_more = 1;  /* more will follow.. eventually.. */
 282         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 283 }
 284
 285 /*
 286  * Prepare to share the seq during handshake
 287  */
 288 static void prepare_write_seq(struct ceph_connection *con)
 289 {
 290         dout("prepare_write_seq %p %llu -> %llu\n", con,
 291              con->in_seq_acked, con->in_seq);
 292         con->in_seq_acked = con->in_seq;
 293
 294         con_out_kvec_reset(con);
 295
 296         con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
 297         con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
 298                          &con->v1.out_temp_ack);
 299
 300         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 301 }
 302
 303 /*
 304  * Prepare to write keepalive byte.
 305  */
 306 static void prepare_write_keepalive(struct ceph_connection *con)
 307 {
 308         dout("prepare_write_keepalive %p\n", con);
 309         con_out_kvec_reset(con);
 310         if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
 311                 struct timespec64 now;
 312
 313                 ktime_get_real_ts64(&now);
 314                 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
 315                 ceph_encode_timespec64(&con->v1.out_temp_keepalive2, &now);
 316                 con_out_kvec_add(con, sizeof(con->v1.out_temp_keepalive2),
 317                                  &con->v1.out_temp_keepalive2);
 318         } else {
 319                 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
 320         }
 321         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 322 }
 323
 324 /*
 325  * Connection negotiation.
 326  */
 327
 328 static int get_connect_authorizer(struct ceph_connection *con)
 329 {
 330         struct ceph_auth_handshake *auth;
 331         int auth_proto;
 332
 333         if (!con->ops->get_authorizer) {
 334                 con->v1.auth = NULL;
 335                 con->v1.out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
 336                 con->v1.out_connect.authorizer_len = 0;
 337                 return 0;
 338         }
 339
 340         auth = con->ops->get_authorizer(con, &auth_proto, con->v1.auth_retry);
 341         if (IS_ERR(auth))
 342                 return PTR_ERR(auth);
 343
 344         con->v1.auth = auth;
 345         con->v1.out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
 346         con->v1.out_connect.authorizer_len =
 347                 cpu_to_le32(auth->authorizer_buf_len);
 348         return 0;
 349 }
 350
 351 /*
 352  * We connected to a peer and are saying hello.
 353  */
 354 static void prepare_write_banner(struct ceph_connection *con)
 355 {
 356         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
 357         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
 358                                         &con->msgr->my_enc_addr);
 359
 360         con->v1.out_more = 0;
 361         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 362 }
 363
 364 static void __prepare_write_connect(struct ceph_connection *con)
 365 {
 366         con_out_kvec_add(con, sizeof(con->v1.out_connect),
 367                          &con->v1.out_connect);
 368         if (con->v1.auth)
 369                 con_out_kvec_add(con, con->v1.auth->authorizer_buf_len,
 370                                  con->v1.auth->authorizer_buf);
 371
 372         con->v1.out_more = 0;
 373         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 374 }
 375
 376 static int prepare_write_connect(struct ceph_connection *con)
 377 {
 378         unsigned int global_seq = ceph_get_global_seq(con->msgr, 0);
 379         int proto;
 380         int ret;
 381
 382         switch (con->peer_name.type) {
 383         case CEPH_ENTITY_TYPE_MON:
 384                 proto = CEPH_MONC_PROTOCOL;
 385                 break;
 386         case CEPH_ENTITY_TYPE_OSD:
 387                 proto = CEPH_OSDC_PROTOCOL;
 388                 break;
 389         case CEPH_ENTITY_TYPE_MDS:
 390                 proto = CEPH_MDSC_PROTOCOL;
 391                 break;
 392         default:
 393                 BUG();
 394         }
 395
 396         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
 397              con->v1.connect_seq, global_seq, proto);
 398
 399         con->v1.out_connect.features =
 400                 cpu_to_le64(from_msgr(con->msgr)->supported_features);
 401         con->v1.out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
 402         con->v1.out_connect.connect_seq = cpu_to_le32(con->v1.connect_seq);
 403         con->v1.out_connect.global_seq = cpu_to_le32(global_seq);
 404         con->v1.out_connect.protocol_version = cpu_to_le32(proto);
 405         con->v1.out_connect.flags = 0;
 406
 407         ret = get_connect_authorizer(con);
 408         if (ret)
 409                 return ret;
 410
 411         __prepare_write_connect(con);
 412         return 0;
 413 }
 414
 415 /*
 416  * write as much of pending kvecs to the socket as we can.
 417  *  1 -> done
 418  *  0 -> socket full, but more to do
 419  * <0 -> error
 420  */
 421 static int write_partial_kvec(struct ceph_connection *con)
 422 {
 423         int ret;
 424
 425         dout("write_partial_kvec %p %d left\n", con, con->v1.out_kvec_bytes);
 426         while (con->v1.out_kvec_bytes > 0) {
 427                 ret = ceph_tcp_sendmsg(con->sock, con->v1.out_kvec_cur,
 428                                        con->v1.out_kvec_left,
 429                                        con->v1.out_kvec_bytes,
 430                                        con->v1.out_more);
 431                 if (ret <= 0)
 432                         goto out;
 433                 con->v1.out_kvec_bytes -= ret;
 434                 if (!con->v1.out_kvec_bytes)
 435                         break;            /* done */
 436
 437                 /* account for full iov entries consumed */
 438                 while (ret >= con->v1.out_kvec_cur->iov_len) {
 439                         BUG_ON(!con->v1.out_kvec_left);
 440                         ret -= con->v1.out_kvec_cur->iov_len;
 441                         con->v1.out_kvec_cur++;
 442                         con->v1.out_kvec_left--;
 443                 }
 444                 /* and for a partially-consumed entry */
 445                 if (ret) {
 446                         con->v1.out_kvec_cur->iov_len -= ret;
 447                         con->v1.out_kvec_cur->iov_base += ret;
 448                 }
 449         }
 450         con->v1.out_kvec_left = 0;
 451         ret = 1;
 452 out:
 453         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
 454              con->v1.out_kvec_bytes, con->v1.out_kvec_left, ret);
 455         return ret;  /* done! */
 456 }
 457
 458 /*
 459  * Write as much message data payload as we can.  If we finish, queue
 460  * up the footer.
 461  *  1 -> done, footer is now queued in out_kvec[].
 462  *  0 -> socket full, but more to do
 463  * <0 -> error
 464  */
 465 static int write_partial_message_data(struct ceph_connection *con)
 466 {
 467         struct ceph_msg *msg = con->out_msg;
 468         struct ceph_msg_data_cursor *cursor = &msg->cursor;
 469         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
 470         int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
 471         u32 crc;
 472
 473         dout("%s %p msg %p\n", __func__, con, msg);
 474
 475         if (!msg->num_data_items)
 476                 return -EINVAL;
 477
 478         /*
 479          * Iterate through each page that contains data to be
 480          * written, and send as much as possible for each.
 481          *
 482          * If we are calculating the data crc (the default), we will
 483          * need to map the page.  If we have no pages, they have
 484          * been revoked, so use the zero page.
 485          */
 486         crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
 487         while (cursor->total_resid) {
 488                 struct page *page;
 489                 size_t page_offset;
 490                 size_t length;
 491                 int ret;
 492
 493                 if (!cursor->resid) {
 494                         ceph_msg_data_advance(cursor, 0);
 495                         continue;
 496                 }
 497
 498                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
 499                 if (length == cursor->total_resid)
 500                         more = MSG_MORE;
 501                 ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
 502                                         more);
 503                 if (ret <= 0) {
 504                         if (do_datacrc)
 505                                 msg->footer.data_crc = cpu_to_le32(crc);
 506
 507                         return ret;
 508                 }
 509                 if (do_datacrc && cursor->need_crc)
 510                         crc = ceph_crc32c_page(crc, page, page_offset, length);
 511                 ceph_msg_data_advance(cursor, (size_t)ret);
 512         }
 513
 514         dout("%s %p msg %p done\n", __func__, con, msg);
 515
 516         /* prepare and queue up footer, too */
 517         if (do_datacrc)
 518                 msg->footer.data_crc = cpu_to_le32(crc);
 519         else
 520                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
 521         con_out_kvec_reset(con);
 522         prepare_write_message_footer(con);
 523
 524         return 1;       /* must return > 0 to indicate success */
 525 }
 526
 527 /*
 528  * write some zeros
 529  */
 530 static int write_partial_skip(struct ceph_connection *con)
 531 {
 532         int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
 533         int ret;
 534
 535         dout("%s %p %d left\n", __func__, con, con->v1.out_skip);
 536         while (con->v1.out_skip > 0) {
 537                 size_t size = min(con->v1.out_skip, (int)PAGE_SIZE);
 538
 539                 if (size == con->v1.out_skip)
 540                         more = MSG_MORE;
 541                 ret = ceph_tcp_sendpage(con->sock, ceph_zero_page, 0, size,
 542                                         more);
 543                 if (ret <= 0)
 544                         goto out;
 545                 con->v1.out_skip -= ret;
 546         }
 547         ret = 1;
 548 out:
 549         return ret;
 550 }
 551
 552 /*
 553  * Prepare to read connection handshake, or an ack.
 554  */
 555 static void prepare_read_banner(struct ceph_connection *con)
 556 {
 557         dout("prepare_read_banner %p\n", con);
 558         con->v1.in_base_pos = 0;
 559 }
 560
 561 static void prepare_read_connect(struct ceph_connection *con)
 562 {
 563         dout("prepare_read_connect %p\n", con);
 564         con->v1.in_base_pos = 0;
 565 }
 566
 567 static void prepare_read_ack(struct ceph_connection *con)
 568 {
 569         dout("prepare_read_ack %p\n", con);
 570         con->v1.in_base_pos = 0;
 571 }
 572
 573 static void prepare_read_seq(struct ceph_connection *con)
 574 {
 575         dout("prepare_read_seq %p\n", con);
 576         con->v1.in_base_pos = 0;
 577         con->v1.in_tag = CEPH_MSGR_TAG_SEQ;
 578 }
 579
 580 static void prepare_read_tag(struct ceph_connection *con)
 581 {
 582         dout("prepare_read_tag %p\n", con);
 583         con->v1.in_base_pos = 0;
 584         con->v1.in_tag = CEPH_MSGR_TAG_READY;
 585 }
 586
 587 static void prepare_read_keepalive_ack(struct ceph_connection *con)
 588 {
 589         dout("prepare_read_keepalive_ack %p\n", con);
 590         con->v1.in_base_pos = 0;
 591 }
 592
 593 /*
 594  * Prepare to read a message.
 595  */
 596 static int prepare_read_message(struct ceph_connection *con)
 597 {
 598         dout("prepare_read_message %p\n", con);
 599         BUG_ON(con->in_msg != NULL);
 600         con->v1.in_base_pos = 0;
 601         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
 602         return 0;
 603 }
 604
 605 static int read_partial(struct ceph_connection *con,
 606                         int end, int size, void *object)
 607 {
 608         while (con->v1.in_base_pos < end) {
 609                 int left = end - con->v1.in_base_pos;
 610                 int have = size - left;
 611                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
 612                 if (ret <= 0)
 613                         return ret;
 614                 con->v1.in_base_pos += ret;
 615         }
 616         return 1;
 617 }
 618
 619 /*
 620  * Read all or part of the connect-side handshake on a new connection
 621  */
 622 static int read_partial_banner(struct ceph_connection *con)
 623 {
 624         int size;
 625         int end;
 626         int ret;
 627
 628         dout("read_partial_banner %p at %d\n", con, con->v1.in_base_pos);
 629
 630         /* peer's banner */
 631         size = strlen(CEPH_BANNER);
 632         end = size;
 633         ret = read_partial(con, end, size, con->v1.in_banner);
 634         if (ret <= 0)
 635                 goto out;
 636
 637         size = sizeof(con->v1.actual_peer_addr);
 638         end += size;
 639         ret = read_partial(con, end, size, &con->v1.actual_peer_addr);
 640         if (ret <= 0)
 641                 goto out;
 642         ceph_decode_banner_addr(&con->v1.actual_peer_addr);
 643
 644         size = sizeof(con->v1.peer_addr_for_me);
 645         end += size;
 646         ret = read_partial(con, end, size, &con->v1.peer_addr_for_me);
 647         if (ret <= 0)
 648                 goto out;
 649         ceph_decode_banner_addr(&con->v1.peer_addr_for_me);
 650
 651 out:
 652         return ret;
 653 }
 654
 655 static int read_partial_connect(struct ceph_connection *con)
 656 {
 657         int size;
 658         int end;
 659         int ret;
 660
 661         dout("read_partial_connect %p at %d\n", con, con->v1.in_base_pos);
 662
 663         size = sizeof(con->v1.in_reply);
 664         end = size;
 665         ret = read_partial(con, end, size, &con->v1.in_reply);
 666         if (ret <= 0)
 667                 goto out;
 668
 669         if (con->v1.auth) {
 670                 size = le32_to_cpu(con->v1.in_reply.authorizer_len);
 671                 if (size > con->v1.auth->authorizer_reply_buf_len) {
 672                         pr_err("authorizer reply too big: %d > %zu\n", size,
 673                                con->v1.auth->authorizer_reply_buf_len);
 674                         ret = -EINVAL;
 675                         goto out;
 676                 }
 677
 678                 end += size;
 679                 ret = read_partial(con, end, size,
 680                                    con->v1.auth->authorizer_reply_buf);
 681                 if (ret <= 0)
 682                         goto out;
 683         }
 684
 685         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
 686              con, con->v1.in_reply.tag,
 687              le32_to_cpu(con->v1.in_reply.connect_seq),
 688              le32_to_cpu(con->v1.in_reply.global_seq));
 689 out:
 690         return ret;
 691 }
 692
 693 /*
 694  * Verify the hello banner looks okay.
 695  */
 696 static int verify_hello(struct ceph_connection *con)
 697 {
 698         if (memcmp(con->v1.in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
 699                 pr_err("connect to %s got bad banner\n",
 700                        ceph_pr_addr(&con->peer_addr));
 701                 con->error_msg = "protocol error, bad banner";
 702                 return -1;
 703         }
 704         return 0;
 705 }
 706
 707 static int process_banner(struct ceph_connection *con)
 708 {
 709         struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
 710
 711         dout("process_banner on %p\n", con);
 712
 713         if (verify_hello(con) < 0)
 714                 return -1;
 715
 716         /*
 717          * Make sure the other end is who we wanted.  note that the other
 718          * end may not yet know their ip address, so if it's 0.0.0.0, give
 719          * them the benefit of the doubt.
 720          */
 721         if (memcmp(&con->peer_addr, &con->v1.actual_peer_addr,
 722                    sizeof(con->peer_addr)) != 0 &&
 723             !(ceph_addr_is_blank(&con->v1.actual_peer_addr) &&
 724               con->v1.actual_peer_addr.nonce == con->peer_addr.nonce)) {
 725                 pr_warn("wrong peer, want %s/%u, got %s/%u\n",
 726                         ceph_pr_addr(&con->peer_addr),
 727                         le32_to_cpu(con->peer_addr.nonce),
 728                         ceph_pr_addr(&con->v1.actual_peer_addr),
 729                         le32_to_cpu(con->v1.actual_peer_addr.nonce));
 730                 con->error_msg = "wrong peer at address";
 731                 return -1;
 732         }
 733
 734         /*
 735          * did we learn our address?
 736          */
 737         if (ceph_addr_is_blank(my_addr)) {
 738                 memcpy(&my_addr->in_addr,
 739                        &con->v1.peer_addr_for_me.in_addr,
 740                        sizeof(con->v1.peer_addr_for_me.in_addr));
 741                 ceph_addr_set_port(my_addr, 0);
 742                 ceph_encode_my_addr(con->msgr);
 743                 dout("process_banner learned my addr is %s\n",
 744                      ceph_pr_addr(my_addr));
 745         }
 746
 747         return 0;
 748 }
 749
 750 static int process_connect(struct ceph_connection *con)
 751 {
 752         u64 sup_feat = from_msgr(con->msgr)->supported_features;
 753         u64 req_feat = from_msgr(con->msgr)->required_features;
 754         u64 server_feat = le64_to_cpu(con->v1.in_reply.features);
 755         int ret;
 756
 757         dout("process_connect on %p tag %d\n", con, con->v1.in_tag);
 758
 759         if (con->v1.auth) {
 760                 int len = le32_to_cpu(con->v1.in_reply.authorizer_len);
 761
 762                 /*
 763                  * Any connection that defines ->get_authorizer()
 764                  * should also define ->add_authorizer_challenge() and
 765                  * ->verify_authorizer_reply().
 766                  *
 767                  * See get_connect_authorizer().
 768                  */
 769                 if (con->v1.in_reply.tag ==
 770                                 CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
 771                         ret = con->ops->add_authorizer_challenge(
 772                                 con, con->v1.auth->authorizer_reply_buf, len);
 773                         if (ret < 0)
 774                                 return ret;
 775
 776                         con_out_kvec_reset(con);
 777                         __prepare_write_connect(con);
 778                         prepare_read_connect(con);
 779                         return 0;
 780                 }
 781
 782                 if (len) {
 783                         ret = con->ops->verify_authorizer_reply(con);
 784                         if (ret < 0) {
 785                                 con->error_msg = "bad authorize reply";
 786                                 return ret;
 787                         }
 788                 }
 789         }
 790
 791         switch (con->v1.in_reply.tag) {
 792         case CEPH_MSGR_TAG_FEATURES:
 793                 pr_err("%s%lld %s feature set mismatch,"
 794                        " my %llx < server's %llx, missing %llx\n",
 795                        ENTITY_NAME(con->peer_name),
 796                        ceph_pr_addr(&con->peer_addr),
 797                        sup_feat, server_feat, server_feat & ~sup_feat);
 798                 con->error_msg = "missing required protocol features";
 799                 return -1;
 800
 801         case CEPH_MSGR_TAG_BADPROTOVER:
 802                 pr_err("%s%lld %s protocol version mismatch,"
 803                        " my %d != server's %d\n",
 804                        ENTITY_NAME(con->peer_name),
 805                        ceph_pr_addr(&con->peer_addr),
 806                        le32_to_cpu(con->v1.out_connect.protocol_version),
 807                        le32_to_cpu(con->v1.in_reply.protocol_version));
 808                 con->error_msg = "protocol version mismatch";
 809                 return -1;
 810
 811         case CEPH_MSGR_TAG_BADAUTHORIZER:
 812                 con->v1.auth_retry++;
 813                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
 814                      con->v1.auth_retry);
 815                 if (con->v1.auth_retry == 2) {
 816                         con->error_msg = "connect authorization failure";
 817                         return -1;
 818                 }
 819                 con_out_kvec_reset(con);
 820                 ret = prepare_write_connect(con);
 821                 if (ret < 0)
 822                         return ret;
 823                 prepare_read_connect(con);
 824                 break;
 825
 826         case CEPH_MSGR_TAG_RESETSESSION:
 827                 /*
 828                  * If we connected with a large connect_seq but the peer
 829                  * has no record of a session with us (no connection, or
 830                  * connect_seq == 0), they will send RESETSESION to indicate
 831                  * that they must have reset their session, and may have
 832                  * dropped messages.
 833                  */
 834                 dout("process_connect got RESET peer seq %u\n",
 835                      le32_to_cpu(con->v1.in_reply.connect_seq));
 836                 pr_info("%s%lld %s session reset\n",
 837                         ENTITY_NAME(con->peer_name),
 838                         ceph_pr_addr(&con->peer_addr));
 839                 ceph_con_reset_session(con);
 840                 con_out_kvec_reset(con);
 841                 ret = prepare_write_connect(con);
 842                 if (ret < 0)
 843                         return ret;
 844                 prepare_read_connect(con);
 845
 846                 /* Tell ceph about it. */
 847                 mutex_unlock(&con->mutex);
 848                 if (con->ops->peer_reset)
 849                         con->ops->peer_reset(con);
 850                 mutex_lock(&con->mutex);
 851                 if (con->state != CEPH_CON_S_V1_CONNECT_MSG)
 852                         return -EAGAIN;
 853                 break;
 854
 855         case CEPH_MSGR_TAG_RETRY_SESSION:
 856                 /*
 857                  * If we sent a smaller connect_seq than the peer has, try
 858                  * again with a larger value.
 859                  */
 860                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
 861                      le32_to_cpu(con->v1.out_connect.connect_seq),
 862                      le32_to_cpu(con->v1.in_reply.connect_seq));
 863                 con->v1.connect_seq = le32_to_cpu(con->v1.in_reply.connect_seq);
 864                 con_out_kvec_reset(con);
 865                 ret = prepare_write_connect(con);
 866                 if (ret < 0)
 867                         return ret;
 868                 prepare_read_connect(con);
 869                 break;
 870
 871         case CEPH_MSGR_TAG_RETRY_GLOBAL:
 872                 /*
 873                  * If we sent a smaller global_seq than the peer has, try
 874                  * again with a larger value.
 875                  */
 876                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
 877                      con->v1.peer_global_seq,
 878                      le32_to_cpu(con->v1.in_reply.global_seq));
 879                 ceph_get_global_seq(con->msgr,
 880                                     le32_to_cpu(con->v1.in_reply.global_seq));
 881                 con_out_kvec_reset(con);
 882                 ret = prepare_write_connect(con);
 883                 if (ret < 0)
 884                         return ret;
 885                 prepare_read_connect(con);
 886                 break;
 887
 888         case CEPH_MSGR_TAG_SEQ:
 889         case CEPH_MSGR_TAG_READY:
 890                 if (req_feat & ~server_feat) {
 891                         pr_err("%s%lld %s protocol feature mismatch,"
 892                                " my required %llx > server's %llx, need %llx\n",
 893                                ENTITY_NAME(con->peer_name),
 894                                ceph_pr_addr(&con->peer_addr),
 895                                req_feat, server_feat, req_feat & ~server_feat);
 896                         con->error_msg = "missing required protocol features";
 897                         return -1;
 898                 }
 899
 900                 WARN_ON(con->state != CEPH_CON_S_V1_CONNECT_MSG);
 901                 con->state = CEPH_CON_S_OPEN;
 902                 con->v1.auth_retry = 0;    /* we authenticated; clear flag */
 903                 con->v1.peer_global_seq =
 904                         le32_to_cpu(con->v1.in_reply.global_seq);
 905                 con->v1.connect_seq++;
 906                 con->peer_features = server_feat;
 907                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
 908                      con->v1.peer_global_seq,
 909                      le32_to_cpu(con->v1.in_reply.connect_seq),
 910                      con->v1.connect_seq);
 911                 WARN_ON(con->v1.connect_seq !=
 912                         le32_to_cpu(con->v1.in_reply.connect_seq));
 913
 914                 if (con->v1.in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
 915                         ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
 916
 917                 con->delay = 0;      /* reset backoff memory */
 918
 919                 if (con->v1.in_reply.tag == CEPH_MSGR_TAG_SEQ) {
 920                         prepare_write_seq(con);
 921                         prepare_read_seq(con);
 922                 } else {
 923                         prepare_read_tag(con);
 924                 }
 925                 break;
 926
 927         case CEPH_MSGR_TAG_WAIT:
 928                 /*
 929                  * If there is a connection race (we are opening
 930                  * connections to each other), one of us may just have
 931                  * to WAIT.  This shouldn't happen if we are the
 932                  * client.
 933                  */
 934                 con->error_msg = "protocol error, got WAIT as client";
 935                 return -1;
 936
 937         default:
 938                 con->error_msg = "protocol error, garbage tag during connect";
 939                 return -1;
 940         }
 941         return 0;
 942 }
 943
 944 /*
 945  * read (part of) an ack
 946  */
 947 static int read_partial_ack(struct ceph_connection *con)
 948 {
 949         int size = sizeof(con->v1.in_temp_ack);
 950         int end = size;
 951
 952         return read_partial(con, end, size, &con->v1.in_temp_ack);
 953 }
 954
 955 /*
 956  * We can finally discard anything that's been acked.
 957  */
 958 static void process_ack(struct ceph_connection *con)
 959 {
 960         u64 ack = le64_to_cpu(con->v1.in_temp_ack);
 961
 962         if (con->v1.in_tag == CEPH_MSGR_TAG_ACK)
 963                 ceph_con_discard_sent(con, ack);
 964         else
 965                 ceph_con_discard_requeued(con, ack);
 966
 967         prepare_read_tag(con);
 968 }
 969
 970 static int read_partial_message_section(struct ceph_connection *con,
 971                                         struct kvec *section,
 972                                         unsigned int sec_len, u32 *crc)
 973 {
 974         int ret, left;
 975
 976         BUG_ON(!section);
 977
 978         while (section->iov_len < sec_len) {
 979                 BUG_ON(section->iov_base == NULL);
 980                 left = sec_len - section->iov_len;
 981                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
 982                                        section->iov_len, left);
 983                 if (ret <= 0)
 984                         return ret;
 985                 section->iov_len += ret;
 986         }
 987         if (section->iov_len == sec_len)
 988                 *crc = crc32c(0, section->iov_base, section->iov_len);
 989
 990         return 1;
 991 }
 992
 993 static int read_partial_msg_data(struct ceph_connection *con)
 994 {
 995         struct ceph_msg *msg = con->in_msg;
 996         struct ceph_msg_data_cursor *cursor = &msg->cursor;
 997         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
 998         struct page *page;
 999         size_t page_offset;
1000         size_t length;
1001         u32 crc = 0;
1002         int ret;
1003
1004         if (!msg->num_data_items)
1005                 return -EIO;
1006
1007         if (do_datacrc)
1008                 crc = con->in_data_crc;
1009         while (cursor->total_resid) {
1010                 if (!cursor->resid) {
1011                         ceph_msg_data_advance(cursor, 0);
1012                         continue;
1013                 }
1014
1015                 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
1016                 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
1017                 if (ret <= 0) {
1018                         if (do_datacrc)
1019                                 con->in_data_crc = crc;
1020
1021                         return ret;
1022                 }
1023
1024                 if (do_datacrc)
1025                         crc = ceph_crc32c_page(crc, page, page_offset, ret);
1026                 ceph_msg_data_advance(cursor, (size_t)ret);
1027         }
1028         if (do_datacrc)
1029                 con->in_data_crc = crc;
1030
1031         return 1;       /* must return > 0 to indicate success */
1032 }
1033
1034 /*
1035  * read (part of) a message.
1036  */
1037 static int read_partial_message(struct ceph_connection *con)
1038 {
1039         struct ceph_msg *m = con->in_msg;
1040         int size;
1041         int end;
1042         int ret;
1043         unsigned int front_len, middle_len, data_len;
1044         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1045         bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
1046         u64 seq;
1047         u32 crc;
1048
1049         dout("read_partial_message con %p msg %p\n", con, m);
1050
1051         /* header */
1052         size = sizeof(con->v1.in_hdr);
1053         end = size;
1054         ret = read_partial(con, end, size, &con->v1.in_hdr);
1055         if (ret <= 0)
1056                 return ret;
1057
1058         crc = crc32c(0, &con->v1.in_hdr, offsetof(struct ceph_msg_header, crc));
1059         if (cpu_to_le32(crc) != con->v1.in_hdr.crc) {
1060                 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
1061                        crc, con->v1.in_hdr.crc);
1062                 return -EBADMSG;
1063         }
1064
1065         front_len = le32_to_cpu(con->v1.in_hdr.front_len);
1066         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1067                 return -EIO;
1068         middle_len = le32_to_cpu(con->v1.in_hdr.middle_len);
1069         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
1070                 return -EIO;
1071         data_len = le32_to_cpu(con->v1.in_hdr.data_len);
1072         if (data_len > CEPH_MSG_MAX_DATA_LEN)
1073                 return -EIO;
1074
1075         /* verify seq# */
1076         seq = le64_to_cpu(con->v1.in_hdr.seq);
1077         if ((s64)seq - (s64)con->in_seq < 1) {
1078                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1079                         ENTITY_NAME(con->peer_name),
1080                         ceph_pr_addr(&con->peer_addr),
1081                         seq, con->in_seq + 1);
1082                 con->v1.in_base_pos = -front_len - middle_len - data_len -
1083                                       sizeof_footer(con);
1084                 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1085                 return 1;
1086         } else if ((s64)seq - (s64)con->in_seq > 1) {
1087                 pr_err("read_partial_message bad seq %lld expected %lld\n",
1088                        seq, con->in_seq + 1);
1089                 con->error_msg = "bad message sequence # for incoming message";
1090                 return -EBADE;
1091         }
1092
1093         /* allocate message? */
1094         if (!con->in_msg) {
1095                 int skip = 0;
1096
1097                 dout("got hdr type %d front %d data %d\n", con->v1.in_hdr.type,
1098                      front_len, data_len);
1099                 ret = ceph_con_in_msg_alloc(con, &con->v1.in_hdr, &skip);
1100                 if (ret < 0)
1101                         return ret;
1102
1103                 BUG_ON(!con->in_msg ^ skip);
1104                 if (skip) {
1105                         /* skip this message */
1106                         dout("alloc_msg said skip message\n");
1107                         con->v1.in_base_pos = -front_len - middle_len -
1108                                               data_len - sizeof_footer(con);
1109                         con->v1.in_tag = CEPH_MSGR_TAG_READY;
1110                         con->in_seq++;
1111                         return 1;
1112                 }
1113
1114                 BUG_ON(!con->in_msg);
1115                 BUG_ON(con->in_msg->con != con);
1116                 m = con->in_msg;
1117                 m->front.iov_len = 0;    /* haven't read it yet */
1118                 if (m->middle)
1119                         m->middle->vec.iov_len = 0;
1120
1121                 /* prepare for data payload, if any */
1122
1123                 if (data_len)
1124                         prepare_message_data(con->in_msg, data_len);
1125         }
1126
1127         /* front */
1128         ret = read_partial_message_section(con, &m->front, front_len,
1129                                            &con->in_front_crc);
1130         if (ret <= 0)
1131                 return ret;
1132
1133         /* middle */
1134         if (m->middle) {
1135                 ret = read_partial_message_section(con, &m->middle->vec,
1136                                                    middle_len,
1137                                                    &con->in_middle_crc);
1138                 if (ret <= 0)
1139                         return ret;
1140         }
1141
1142         /* (page) data */
1143         if (data_len) {
1144                 ret = read_partial_msg_data(con);
1145                 if (ret <= 0)
1146                         return ret;
1147         }
1148
1149         /* footer */
1150         size = sizeof_footer(con);
1151         end += size;
1152         ret = read_partial(con, end, size, &m->footer);
1153         if (ret <= 0)
1154                 return ret;
1155
1156         if (!need_sign) {
1157                 m->footer.flags = m->old_footer.flags;
1158                 m->footer.sig = 0;
1159         }
1160
1161         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1162              m, front_len, m->footer.front_crc, middle_len,
1163              m->footer.middle_crc, data_len, m->footer.data_crc);
1164
1165         /* crc ok? */
1166         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1167                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1168                        m, con->in_front_crc, m->footer.front_crc);
1169                 return -EBADMSG;
1170         }
1171         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1172                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1173                        m, con->in_middle_crc, m->footer.middle_crc);
1174                 return -EBADMSG;
1175         }
1176         if (do_datacrc &&
1177             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1178             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1179                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1180                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1181                 return -EBADMSG;
1182         }
1183
1184         if (need_sign && con->ops->check_message_signature &&
1185             con->ops->check_message_signature(m)) {
1186                 pr_err("read_partial_message %p signature check failed\n", m);
1187                 return -EBADMSG;
1188         }
1189
1190         return 1; /* done! */
1191 }
1192
1193 static int read_keepalive_ack(struct ceph_connection *con)
1194 {
1195         struct ceph_timespec ceph_ts;
1196         size_t size = sizeof(ceph_ts);
1197         int ret = read_partial(con, size, size, &ceph_ts);
1198         if (ret <= 0)
1199                 return ret;
1200         ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
1201         prepare_read_tag(con);
1202         return 1;
1203 }
1204
1205 /*
1206  * Read what we can from the socket.
1207  */
1208 int ceph_con_v1_try_read(struct ceph_connection *con)
1209 {
1210         int ret = -1;
1211
1212 more:
1213         dout("try_read start %p state %d\n", con, con->state);
1214         if (con->state != CEPH_CON_S_V1_BANNER &&
1215             con->state != CEPH_CON_S_V1_CONNECT_MSG &&
1216             con->state != CEPH_CON_S_OPEN)
1217                 return 0;
1218
1219         BUG_ON(!con->sock);
1220
1221         dout("try_read tag %d in_base_pos %d\n", con->v1.in_tag,
1222              con->v1.in_base_pos);
1223
1224         if (con->state == CEPH_CON_S_V1_BANNER) {
1225                 ret = read_partial_banner(con);
1226                 if (ret <= 0)
1227                         goto out;
1228                 ret = process_banner(con);
1229                 if (ret < 0)
1230                         goto out;
1231
1232                 con->state = CEPH_CON_S_V1_CONNECT_MSG;
1233
1234                 /*
1235                  * Received banner is good, exchange connection info.
1236                  * Do not reset out_kvec, as sending our banner raced
1237                  * with receiving peer banner after connect completed.
1238                  */
1239                 ret = prepare_write_connect(con);
1240                 if (ret < 0)
1241                         goto out;
1242                 prepare_read_connect(con);
1243
1244                 /* Send connection info before awaiting response */
1245                 goto out;
1246         }
1247
1248         if (con->state == CEPH_CON_S_V1_CONNECT_MSG) {
1249                 ret = read_partial_connect(con);
1250                 if (ret <= 0)
1251                         goto out;
1252                 ret = process_connect(con);
1253                 if (ret < 0)
1254                         goto out;
1255                 goto more;
1256         }
1257
1258         WARN_ON(con->state != CEPH_CON_S_OPEN);
1259
1260         if (con->v1.in_base_pos < 0) {
1261                 /*
1262                  * skipping + discarding content.
1263                  */
1264                 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->v1.in_base_pos);
1265                 if (ret <= 0)
1266                         goto out;
1267                 dout("skipped %d / %d bytes\n", ret, -con->v1.in_base_pos);
1268                 con->v1.in_base_pos += ret;
1269                 if (con->v1.in_base_pos)
1270                         goto more;
1271         }
1272         if (con->v1.in_tag == CEPH_MSGR_TAG_READY) {
1273                 /*
1274                  * what's next?
1275                  */
1276                 ret = ceph_tcp_recvmsg(con->sock, &con->v1.in_tag, 1);
1277                 if (ret <= 0)
1278                         goto out;
1279                 dout("try_read got tag %d\n", con->v1.in_tag);
1280                 switch (con->v1.in_tag) {
1281                 case CEPH_MSGR_TAG_MSG:
1282                         prepare_read_message(con);
1283                         break;
1284                 case CEPH_MSGR_TAG_ACK:
1285                         prepare_read_ack(con);
1286                         break;
1287                 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
1288                         prepare_read_keepalive_ack(con);
1289                         break;
1290                 case CEPH_MSGR_TAG_CLOSE:
1291                         ceph_con_close_socket(con);
1292                         con->state = CEPH_CON_S_CLOSED;
1293                         goto out;
1294                 default:
1295                         goto bad_tag;
1296                 }
1297         }
1298         if (con->v1.in_tag == CEPH_MSGR_TAG_MSG) {
1299                 ret = read_partial_message(con);
1300                 if (ret <= 0) {
1301                         switch (ret) {
1302                         case -EBADMSG:
1303                                 con->error_msg = "bad crc/signature";
1304                                 fallthrough;
1305                         case -EBADE:
1306                                 ret = -EIO;
1307                                 break;
1308                         case -EIO:
1309                                 con->error_msg = "io error";
1310                                 break;
1311                         }
1312                         goto out;
1313                 }
1314                 if (con->v1.in_tag == CEPH_MSGR_TAG_READY)
1315                         goto more;
1316                 ceph_con_process_message(con);
1317                 if (con->state == CEPH_CON_S_OPEN)
1318                         prepare_read_tag(con);
1319                 goto more;
1320         }
1321         if (con->v1.in_tag == CEPH_MSGR_TAG_ACK ||
1322             con->v1.in_tag == CEPH_MSGR_TAG_SEQ) {
1323                 /*
1324                  * the final handshake seq exchange is semantically
1325                  * equivalent to an ACK
1326                  */
1327                 ret = read_partial_ack(con);
1328                 if (ret <= 0)
1329                         goto out;
1330                 process_ack(con);
1331                 goto more;
1332         }
1333         if (con->v1.in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
1334                 ret = read_keepalive_ack(con);
1335                 if (ret <= 0)
1336                         goto out;
1337                 goto more;
1338         }
1339
1340 out:
1341         dout("try_read done on %p ret %d\n", con, ret);
1342         return ret;
1343
1344 bad_tag:
1345         pr_err("try_read bad tag %d\n", con->v1.in_tag);
1346         con->error_msg = "protocol error, garbage tag";
1347         ret = -1;
1348         goto out;
1349 }
1350
1351 /*
1352  * Write something to the socket.  Called in a worker thread when the
1353  * socket appears to be writeable and we have something ready to send.
1354  */
1355 int ceph_con_v1_try_write(struct ceph_connection *con)
1356 {
1357         int ret = 1;
1358
1359         dout("try_write start %p state %d\n", con, con->state);
1360         if (con->state != CEPH_CON_S_PREOPEN &&
1361             con->state != CEPH_CON_S_V1_BANNER &&
1362             con->state != CEPH_CON_S_V1_CONNECT_MSG &&
1363             con->state != CEPH_CON_S_OPEN)
1364                 return 0;
1365
1366         /* open the socket first? */
1367         if (con->state == CEPH_CON_S_PREOPEN) {
1368                 BUG_ON(con->sock);
1369                 con->state = CEPH_CON_S_V1_BANNER;
1370
1371                 con_out_kvec_reset(con);
1372                 prepare_write_banner(con);
1373                 prepare_read_banner(con);
1374
1375                 BUG_ON(con->in_msg);
1376                 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1377                 dout("try_write initiating connect on %p new state %d\n",
1378                      con, con->state);
1379                 ret = ceph_tcp_connect(con);
1380                 if (ret < 0) {
1381                         con->error_msg = "connect error";
1382                         goto out;
1383                 }
1384         }
1385
1386 more:
1387         dout("try_write out_kvec_bytes %d\n", con->v1.out_kvec_bytes);
1388         BUG_ON(!con->sock);
1389
1390         /* kvec data queued? */
1391         if (con->v1.out_kvec_left) {
1392                 ret = write_partial_kvec(con);
1393                 if (ret <= 0)
1394                         goto out;
1395         }
1396         if (con->v1.out_skip) {
1397                 ret = write_partial_skip(con);
1398                 if (ret <= 0)
1399                         goto out;
1400         }
1401
1402         /* msg pages? */
1403         if (con->out_msg) {
1404                 if (con->v1.out_msg_done) {
1405                         ceph_msg_put(con->out_msg);
1406                         con->out_msg = NULL;   /* we're done with this one */
1407                         goto do_next;
1408                 }
1409
1410                 ret = write_partial_message_data(con);
1411                 if (ret == 1)
1412                         goto more;  /* we need to send the footer, too! */
1413                 if (ret == 0)
1414                         goto out;
1415                 if (ret < 0) {
1416                         dout("try_write write_partial_message_data err %d\n",
1417                              ret);
1418                         goto out;
1419                 }
1420         }
1421
1422 do_next:
1423         if (con->state == CEPH_CON_S_OPEN) {
1424                 if (ceph_con_flag_test_and_clear(con,
1425                                 CEPH_CON_F_KEEPALIVE_PENDING)) {
1426                         prepare_write_keepalive(con);
1427                         goto more;
1428                 }
1429                 /* is anything else pending? */
1430                 if (!list_empty(&con->out_queue)) {
1431                         prepare_write_message(con);
1432                         goto more;
1433                 }
1434                 if (con->in_seq > con->in_seq_acked) {
1435                         prepare_write_ack(con);
1436                         goto more;
1437                 }
1438         }
1439
1440         /* Nothing to do! */
1441         ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1442         dout("try_write nothing else to write.\n");
1443         ret = 0;
1444 out:
1445         dout("try_write done on %p ret %d\n", con, ret);
1446         return ret;
1447 }
1448
1449 void ceph_con_v1_revoke(struct ceph_connection *con)
1450 {
1451         struct ceph_msg *msg = con->out_msg;
1452
1453         WARN_ON(con->v1.out_skip);
1454         /* footer */
1455         if (con->v1.out_msg_done) {
1456                 con->v1.out_skip += con_out_kvec_skip(con);
1457         } else {
1458                 WARN_ON(!msg->data_length);
1459                 con->v1.out_skip += sizeof_footer(con);
1460         }
1461         /* data, middle, front */
1462         if (msg->data_length)
1463                 con->v1.out_skip += msg->cursor.total_resid;
1464         if (msg->middle)
1465                 con->v1.out_skip += con_out_kvec_skip(con);
1466         con->v1.out_skip += con_out_kvec_skip(con);
1467
1468         dout("%s con %p out_kvec_bytes %d out_skip %d\n", __func__, con,
1469              con->v1.out_kvec_bytes, con->v1.out_skip);
1470 }
1471
1472 void ceph_con_v1_revoke_incoming(struct ceph_connection *con)
1473 {
1474         unsigned int front_len = le32_to_cpu(con->v1.in_hdr.front_len);
1475         unsigned int middle_len = le32_to_cpu(con->v1.in_hdr.middle_len);
1476         unsigned int data_len = le32_to_cpu(con->v1.in_hdr.data_len);
1477
1478         /* skip rest of message */
1479         con->v1.in_base_pos = con->v1.in_base_pos -
1480                         sizeof(struct ceph_msg_header) -
1481                         front_len -
1482                         middle_len -
1483                         data_len -
1484                         sizeof(struct ceph_msg_footer);
1485
1486         con->v1.in_tag = CEPH_MSGR_TAG_READY;
1487         con->in_seq++;
1488
1489         dout("%s con %p in_base_pos %d\n", __func__, con, con->v1.in_base_pos);
1490 }
1491
1492 bool ceph_con_v1_opened(struct ceph_connection *con)
1493 {
1494         return con->v1.connect_seq;
1495 }
1496
1497 void ceph_con_v1_reset_session(struct ceph_connection *con)
1498 {
1499         con->v1.connect_seq = 0;
1500         con->v1.peer_global_seq = 0;
1501 }
1502
1503 void ceph_con_v1_reset_protocol(struct ceph_connection *con)
1504 {
1505         con->v1.out_skip = 0;
1506 }