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, ITER_DEST, &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         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
  45         int r;
  46
  47         BUG_ON(page_offset + length > PAGE_SIZE);
  48         bvec_set_page(&bvec, page, length, page_offset);
  49         iov_iter_bvec(&msg.msg_iter, ITER_DEST, &bvec, 1, length);
  50         r = sock_recvmsg(sock, &msg, msg.msg_flags);
  51         if (r == -EAGAIN)
  52                 r = 0;
  53         return r;
  54 }
  55
  56 /*
  57  * write something.  @more is true if caller will be sending more data
  58  * shortly.
  59  */
  60 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
  61                             size_t kvlen, size_t len, bool more)
  62 {
  63         struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
  64         int r;
  65
  66         if (more)
  67                 msg.msg_flags |= MSG_MORE;
  68         else
  69                 msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
  70
  71         r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
  72         if (r == -EAGAIN)
  73                 r = 0;
  74         return r;
  75 }
  76
  77 /*
  78  * @more: MSG_MORE or 0.
  79  */
  80 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
  81                              int offset, size_t size, int more)
  82 {
  83         struct msghdr msg = {
  84                 .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL | more,
  85         };
  86         struct bio_vec bvec;
  87         int ret;
  88
  89         /*
  90          * MSG_SPLICE_PAGES cannot properly handle pages with page_count == 0,
  91          * we need to fall back to sendmsg if that's the case.
  92          *
  93          * Same goes for slab pages: skb_can_coalesce() allows
  94          * coalescing neighboring slab objects into a single frag which
  95          * triggers one of hardened usercopy checks.
  96          */
  97         if (sendpage_ok(page))
  98                 msg.msg_flags |= MSG_SPLICE_PAGES;
  99
 100         bvec_set_page(&bvec, page, size, offset);
 101         iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);
 102
 103         ret = sock_sendmsg(sock, &msg);
 104         if (ret == -EAGAIN)
 105                 ret = 0;
 106
 107         return ret;
 108 }
 109
 110 static void con_out_kvec_reset(struct ceph_connection *con)
 111 {
 112         BUG_ON(con->v1.out_skip);
 113
 114         con->v1.out_kvec_left = 0;
 115         con->v1.out_kvec_bytes = 0;
 116         con->v1.out_kvec_cur = &con->v1.out_kvec[0];
 117 }
 118
 119 static void con_out_kvec_add(struct ceph_connection *con,
 120                                 size_t size, void *data)
 121 {
 122         int index = con->v1.out_kvec_left;
 123
 124         BUG_ON(con->v1.out_skip);
 125         BUG_ON(index >= ARRAY_SIZE(con->v1.out_kvec));
 126
 127         con->v1.out_kvec[index].iov_len = size;
 128         con->v1.out_kvec[index].iov_base = data;
 129         con->v1.out_kvec_left++;
 130         con->v1.out_kvec_bytes += size;
 131 }
 132
 133 /*
 134  * Chop off a kvec from the end.  Return residual number of bytes for
 135  * that kvec, i.e. how many bytes would have been written if the kvec
 136  * hadn't been nuked.
 137  */
 138 static int con_out_kvec_skip(struct ceph_connection *con)
 139 {
 140         int skip = 0;
 141
 142         if (con->v1.out_kvec_bytes > 0) {
 143                 skip = con->v1.out_kvec_cur[con->v1.out_kvec_left - 1].iov_len;
 144                 BUG_ON(con->v1.out_kvec_bytes < skip);
 145                 BUG_ON(!con->v1.out_kvec_left);
 146                 con->v1.out_kvec_bytes -= skip;
 147                 con->v1.out_kvec_left--;
 148         }
 149
 150         return skip;
 151 }
 152
 153 static size_t sizeof_footer(struct ceph_connection *con)
 154 {
 155         return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
 156             sizeof(struct ceph_msg_footer) :
 157             sizeof(struct ceph_msg_footer_old);
 158 }
 159
 160 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
 161 {
 162         /* Initialize data cursor if it's not a sparse read */
 163         if (!msg->sparse_read)
 164                 ceph_msg_data_cursor_init(&msg->cursor, msg, data_len);
 165 }
 166
 167 /*
 168  * Prepare footer for currently outgoing message, and finish things
 169  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
 170  */
 171 static void prepare_write_message_footer(struct ceph_connection *con)
 172 {
 173         struct ceph_msg *m = con->out_msg;
 174
 175         m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
 176
 177         dout("prepare_write_message_footer %p\n", con);
 178         con_out_kvec_add(con, sizeof_footer(con), &m->footer);
 179         if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
 180                 if (con->ops->sign_message)
 181                         con->ops->sign_message(m);
 182                 else
 183                         m->footer.sig = 0;
 184         } else {
 185                 m->old_footer.flags = m->footer.flags;
 186         }
 187         con->v1.out_more = m->more_to_follow;
 188         con->v1.out_msg_done = true;
 189 }
 190
 191 /*
 192  * Prepare headers for the next outgoing message.
 193  */
 194 static void prepare_write_message(struct ceph_connection *con)
 195 {
 196         struct ceph_msg *m;
 197         u32 crc;
 198
 199         con_out_kvec_reset(con);
 200         con->v1.out_msg_done = false;
 201
 202         /* Sneak an ack in there first?  If we can get it into the same
 203          * TCP packet that's a good thing. */
 204         if (con->in_seq > con->in_seq_acked) {
 205                 con->in_seq_acked = con->in_seq;
 206                 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
 207                 con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
 208                 con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
 209                         &con->v1.out_temp_ack);
 210         }
 211
 212         ceph_con_get_out_msg(con);
 213         m = con->out_msg;
 214
 215         dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
 216              m, con->out_seq, le16_to_cpu(m->hdr.type),
 217              le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
 218              m->data_length);
 219         WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
 220         WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
 221
 222         /* tag + hdr + front + middle */
 223         con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
 224         con_out_kvec_add(con, sizeof(con->v1.out_hdr), &con->v1.out_hdr);
 225         con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
 226
 227         if (m->middle)
 228                 con_out_kvec_add(con, m->middle->vec.iov_len,
 229                         m->middle->vec.iov_base);
 230
 231         /* fill in hdr crc and finalize hdr */
 232         crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
 233         con->out_msg->hdr.crc = cpu_to_le32(crc);
 234         memcpy(&con->v1.out_hdr, &con->out_msg->hdr, sizeof(con->v1.out_hdr));
 235
 236         /* fill in front and middle crc, footer */
 237         crc = crc32c(0, m->front.iov_base, m->front.iov_len);
 238         con->out_msg->footer.front_crc = cpu_to_le32(crc);
 239         if (m->middle) {
 240                 crc = crc32c(0, m->middle->vec.iov_base,
 241                                 m->middle->vec.iov_len);
 242                 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
 243         } else
 244                 con->out_msg->footer.middle_crc = 0;
 245         dout("%s front_crc %u middle_crc %u\n", __func__,
 246              le32_to_cpu(con->out_msg->footer.front_crc),
 247              le32_to_cpu(con->out_msg->footer.middle_crc));
 248         con->out_msg->footer.flags = 0;
 249
 250         /* is there a data payload? */
 251         con->out_msg->footer.data_crc = 0;
 252         if (m->data_length) {
 253                 prepare_message_data(con->out_msg, m->data_length);
 254                 con->v1.out_more = 1;  /* data + footer will follow */
 255         } else {
 256                 /* no, queue up footer too and be done */
 257                 prepare_write_message_footer(con);
 258         }
 259
 260         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 261 }
 262
 263 /*
 264  * Prepare an ack.
 265  */
 266 static void prepare_write_ack(struct ceph_connection *con)
 267 {
 268         dout("prepare_write_ack %p %llu -> %llu\n", con,
 269              con->in_seq_acked, con->in_seq);
 270         con->in_seq_acked = con->in_seq;
 271
 272         con_out_kvec_reset(con);
 273
 274         con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
 275
 276         con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
 277         con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
 278                          &con->v1.out_temp_ack);
 279
 280         con->v1.out_more = 1;  /* more will follow.. eventually.. */
 281         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 282 }
 283
 284 /*
 285  * Prepare to share the seq during handshake
 286  */
 287 static void prepare_write_seq(struct ceph_connection *con)
 288 {
 289         dout("prepare_write_seq %p %llu -> %llu\n", con,
 290              con->in_seq_acked, con->in_seq);
 291         con->in_seq_acked = con->in_seq;
 292
 293         con_out_kvec_reset(con);
 294
 295         con->v1.out_temp_ack = cpu_to_le64(con->in_seq_acked);
 296         con_out_kvec_add(con, sizeof(con->v1.out_temp_ack),
 297                          &con->v1.out_temp_ack);
 298
 299         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 300 }
 301
 302 /*
 303  * Prepare to write keepalive byte.
 304  */
 305 static void prepare_write_keepalive(struct ceph_connection *con)
 306 {
 307         dout("prepare_write_keepalive %p\n", con);
 308         con_out_kvec_reset(con);
 309         if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
 310                 struct timespec64 now;
 311
 312                 ktime_get_real_ts64(&now);
 313                 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
 314                 ceph_encode_timespec64(&con->v1.out_temp_keepalive2, &now);
 315                 con_out_kvec_add(con, sizeof(con->v1.out_temp_keepalive2),
 316                                  &con->v1.out_temp_keepalive2);
 317         } else {
 318                 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
 319         }
 320         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 321 }
 322
 323 /*
 324  * Connection negotiation.
 325  */
 326
 327 static int get_connect_authorizer(struct ceph_connection *con)
 328 {
 329         struct ceph_auth_handshake *auth;
 330         int auth_proto;
 331
 332         if (!con->ops->get_authorizer) {
 333                 con->v1.auth = NULL;
 334                 con->v1.out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
 335                 con->v1.out_connect.authorizer_len = 0;
 336                 return 0;
 337         }
 338
 339         auth = con->ops->get_authorizer(con, &auth_proto, con->v1.auth_retry);
 340         if (IS_ERR(auth))
 341                 return PTR_ERR(auth);
 342
 343         con->v1.auth = auth;
 344         con->v1.out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
 345         con->v1.out_connect.authorizer_len =
 346                 cpu_to_le32(auth->authorizer_buf_len);
 347         return 0;
 348 }
 349
 350 /*
 351  * We connected to a peer and are saying hello.
 352  */
 353 static void prepare_write_banner(struct ceph_connection *con)
 354 {
 355         con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
 356         con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
 357                                         &con->msgr->my_enc_addr);
 358
 359         con->v1.out_more = 0;
 360         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 361 }
 362
 363 static void __prepare_write_connect(struct ceph_connection *con)
 364 {
 365         con_out_kvec_add(con, sizeof(con->v1.out_connect),
 366                          &con->v1.out_connect);
 367         if (con->v1.auth)
 368                 con_out_kvec_add(con, con->v1.auth->authorizer_buf_len,
 369                                  con->v1.auth->authorizer_buf);
 370
 371         con->v1.out_more = 0;
 372         ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
 373 }
 374
 375 static int prepare_write_connect(struct ceph_connection *con)
 376 {
 377         unsigned int global_seq = ceph_get_global_seq(con->msgr, 0);
 378         int proto;
 379         int ret;
 380
 381         switch (con->peer_name.type) {
 382         case CEPH_ENTITY_TYPE_MON:
 383                 proto = CEPH_MONC_PROTOCOL;
 384                 break;
 385         case CEPH_ENTITY_TYPE_OSD:
 386                 proto = CEPH_OSDC_PROTOCOL;
 387                 break;
 388         case CEPH_ENTITY_TYPE_MDS:
 389                 proto = CEPH_MDSC_PROTOCOL;
 390                 break;
 391         default:
 392                 BUG();
 393         }
 394
 395         dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
 396              con->v1.connect_seq, global_seq, proto);
 397
 398         con->v1.out_connect.features =
 399                 cpu_to_le64(from_msgr(con->msgr)->supported_features);
 400         con->v1.out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
 401         con->v1.out_connect.connect_seq = cpu_to_le32(con->v1.connect_seq);
 402         con->v1.out_connect.global_seq = cpu_to_le32(global_seq);
 403         con->v1.out_connect.protocol_version = cpu_to_le32(proto);
 404         con->v1.out_connect.flags = 0;
 405
 406         ret = get_connect_authorizer(con);
 407         if (ret)
 408                 return ret;
 409
 410         __prepare_write_connect(con);
 411         return 0;
 412 }
 413
 414 /*
 415  * write as much of pending kvecs to the socket as we can.
 416  *  1 -> done
 417  *  0 -> socket full, but more to do
 418  * <0 -> error
 419  */
 420 static int write_partial_kvec(struct ceph_connection *con)
 421 {
 422         int ret;
 423
 424         dout("write_partial_kvec %p %d left\n", con, con->v1.out_kvec_bytes);
 425         while (con->v1.out_kvec_bytes > 0) {
 426                 ret = ceph_tcp_sendmsg(con->sock, con->v1.out_kvec_cur,
 427                                        con->v1.out_kvec_left,
 428                                        con->v1.out_kvec_bytes,
 429                                        con->v1.out_more);
 430                 if (ret <= 0)
 431                         goto out;
 432                 con->v1.out_kvec_bytes -= ret;
 433                 if (!con->v1.out_kvec_bytes)
 434                         break;            /* done */
 435
 436                 /* account for full iov entries consumed */
 437                 while (ret >= con->v1.out_kvec_cur->iov_len) {
 438                         BUG_ON(!con->v1.out_kvec_left);
 439                         ret -= con->v1.out_kvec_cur->iov_len;
 440                         con->v1.out_kvec_cur++;
 441                         con->v1.out_kvec_left--;
 442                 }
 443                 /* and for a partially-consumed entry */
 444                 if (ret) {
 445                         con->v1.out_kvec_cur->iov_len -= ret;
 446                         con->v1.out_kvec_cur->iov_base += ret;
 447                 }
 448         }
 449         con->v1.out_kvec_left = 0;
 450         ret = 1;
 451 out:
 452         dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
 453              con->v1.out_kvec_bytes, con->v1.out_kvec_left, ret);
 454         return ret;  /* done! */
 455 }
 456
 457 /*
 458  * Write as much message data payload as we can.  If we finish, queue
 459  * up the footer.
 460  *  1 -> done, footer is now queued in out_kvec[].
 461  *  0 -> socket full, but more to do
 462  * <0 -> error
 463  */
 464 static int write_partial_message_data(struct ceph_connection *con)
 465 {
 466         struct ceph_msg *msg = con->out_msg;
 467         struct ceph_msg_data_cursor *cursor = &msg->cursor;
 468         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
 469         u32 crc;
 470
 471         dout("%s %p msg %p\n", __func__, con, msg);
 472
 473         if (!msg->num_data_items)
 474                 return -EINVAL;
 475
 476         /*
 477          * Iterate through each page that contains data to be
 478          * written, and send as much as possible for each.
 479          *
 480          * If we are calculating the data crc (the default), we will
 481          * need to map the page.  If we have no pages, they have
 482          * been revoked, so use the zero page.
 483          */
 484         crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
 485         while (cursor->total_resid) {
 486                 struct page *page;
 487                 size_t page_offset;
 488                 size_t length;
 489                 int ret;
 490
 491                 if (!cursor->resid) {
 492                         ceph_msg_data_advance(cursor, 0);
 493                         continue;
 494                 }
 495
 496                 page = ceph_msg_data_next(cursor, &page_offset, &length);
 497                 ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
 498                                         MSG_MORE);
 499                 if (ret <= 0) {
 500                         if (do_datacrc)
 501                                 msg->footer.data_crc = cpu_to_le32(crc);
 502
 503                         return ret;
 504                 }
 505                 if (do_datacrc && cursor->need_crc)
 506                         crc = ceph_crc32c_page(crc, page, page_offset, length);
 507                 ceph_msg_data_advance(cursor, (size_t)ret);
 508         }
 509
 510         dout("%s %p msg %p done\n", __func__, con, msg);
 511
 512         /* prepare and queue up footer, too */
 513         if (do_datacrc)
 514                 msg->footer.data_crc = cpu_to_le32(crc);
 515         else
 516                 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
 517         con_out_kvec_reset(con);
 518         prepare_write_message_footer(con);
 519
 520         return 1;       /* must return > 0 to indicate success */
 521 }
 522
 523 /*
 524  * write some zeros
 525  */
 526 static int write_partial_skip(struct ceph_connection *con)
 527 {
 528         int ret;
 529
 530         dout("%s %p %d left\n", __func__, con, con->v1.out_skip);
 531         while (con->v1.out_skip > 0) {
 532                 size_t size = min(con->v1.out_skip, (int)PAGE_SIZE);
 533
 534                 ret = ceph_tcp_sendpage(con->sock, ceph_zero_page, 0, size,
 535                                         MSG_MORE);
 536                 if (ret <= 0)
 537                         goto out;
 538                 con->v1.out_skip -= ret;
 539         }
 540         ret = 1;
 541 out:
 542         return ret;
 543 }
 544
 545 /*
 546  * Prepare to read connection handshake, or an ack.
 547  */
 548 static void prepare_read_banner(struct ceph_connection *con)
 549 {
 550         dout("prepare_read_banner %p\n", con);
 551         con->v1.in_base_pos = 0;
 552 }
 553
 554 static void prepare_read_connect(struct ceph_connection *con)
 555 {
 556         dout("prepare_read_connect %p\n", con);
 557         con->v1.in_base_pos = 0;
 558 }
 559
 560 static void prepare_read_ack(struct ceph_connection *con)
 561 {
 562         dout("prepare_read_ack %p\n", con);
 563         con->v1.in_base_pos = 0;
 564 }
 565
 566 static void prepare_read_seq(struct ceph_connection *con)
 567 {
 568         dout("prepare_read_seq %p\n", con);
 569         con->v1.in_base_pos = 0;
 570         con->v1.in_tag = CEPH_MSGR_TAG_SEQ;
 571 }
 572
 573 static void prepare_read_tag(struct ceph_connection *con)
 574 {
 575         dout("prepare_read_tag %p\n", con);
 576         con->v1.in_base_pos = 0;
 577         con->v1.in_tag = CEPH_MSGR_TAG_READY;
 578 }
 579
 580 static void prepare_read_keepalive_ack(struct ceph_connection *con)
 581 {
 582         dout("prepare_read_keepalive_ack %p\n", con);
 583         con->v1.in_base_pos = 0;
 584 }
 585
 586 /*
 587  * Prepare to read a message.
 588  */
 589 static int prepare_read_message(struct ceph_connection *con)
 590 {
 591         dout("prepare_read_message %p\n", con);
 592         BUG_ON(con->in_msg != NULL);
 593         con->v1.in_base_pos = 0;
 594         con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
 595         return 0;
 596 }
 597
 598 static int read_partial(struct ceph_connection *con,
 599                         int end, int size, void *object)
 600 {
 601         while (con->v1.in_base_pos < end) {
 602                 int left = end - con->v1.in_base_pos;
 603                 int have = size - left;
 604                 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
 605                 if (ret <= 0)
 606                         return ret;
 607                 con->v1.in_base_pos += ret;
 608         }
 609         return 1;
 610 }
 611
 612 /*
 613  * Read all or part of the connect-side handshake on a new connection
 614  */
 615 static int read_partial_banner(struct ceph_connection *con)
 616 {
 617         int size;
 618         int end;
 619         int ret;
 620
 621         dout("read_partial_banner %p at %d\n", con, con->v1.in_base_pos);
 622
 623         /* peer's banner */
 624         size = strlen(CEPH_BANNER);
 625         end = size;
 626         ret = read_partial(con, end, size, con->v1.in_banner);
 627         if (ret <= 0)
 628                 goto out;
 629
 630         size = sizeof(con->v1.actual_peer_addr);
 631         end += size;
 632         ret = read_partial(con, end, size, &con->v1.actual_peer_addr);
 633         if (ret <= 0)
 634                 goto out;
 635         ceph_decode_banner_addr(&con->v1.actual_peer_addr);
 636
 637         size = sizeof(con->v1.peer_addr_for_me);
 638         end += size;
 639         ret = read_partial(con, end, size, &con->v1.peer_addr_for_me);
 640         if (ret <= 0)
 641                 goto out;
 642         ceph_decode_banner_addr(&con->v1.peer_addr_for_me);
 643
 644 out:
 645         return ret;
 646 }
 647
 648 static int read_partial_connect(struct ceph_connection *con)
 649 {
 650         int size;
 651         int end;
 652         int ret;
 653
 654         dout("read_partial_connect %p at %d\n", con, con->v1.in_base_pos);
 655
 656         size = sizeof(con->v1.in_reply);
 657         end = size;
 658         ret = read_partial(con, end, size, &con->v1.in_reply);
 659         if (ret <= 0)
 660                 goto out;
 661
 662         if (con->v1.auth) {
 663                 size = le32_to_cpu(con->v1.in_reply.authorizer_len);
 664                 if (size > con->v1.auth->authorizer_reply_buf_len) {
 665                         pr_err("authorizer reply too big: %d > %zu\n", size,
 666                                con->v1.auth->authorizer_reply_buf_len);
 667                         ret = -EINVAL;
 668                         goto out;
 669                 }
 670
 671                 end += size;
 672                 ret = read_partial(con, end, size,
 673                                    con->v1.auth->authorizer_reply_buf);
 674                 if (ret <= 0)
 675                         goto out;
 676         }
 677
 678         dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
 679              con, con->v1.in_reply.tag,
 680              le32_to_cpu(con->v1.in_reply.connect_seq),
 681              le32_to_cpu(con->v1.in_reply.global_seq));
 682 out:
 683         return ret;
 684 }
 685
 686 /*
 687  * Verify the hello banner looks okay.
 688  */
 689 static int verify_hello(struct ceph_connection *con)
 690 {
 691         if (memcmp(con->v1.in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
 692                 pr_err("connect to %s got bad banner\n",
 693                        ceph_pr_addr(&con->peer_addr));
 694                 con->error_msg = "protocol error, bad banner";
 695                 return -1;
 696         }
 697         return 0;
 698 }
 699
 700 static int process_banner(struct ceph_connection *con)
 701 {
 702         struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
 703
 704         dout("process_banner on %p\n", con);
 705
 706         if (verify_hello(con) < 0)
 707                 return -1;
 708
 709         /*
 710          * Make sure the other end is who we wanted.  note that the other
 711          * end may not yet know their ip address, so if it's 0.0.0.0, give
 712          * them the benefit of the doubt.
 713          */
 714         if (memcmp(&con->peer_addr, &con->v1.actual_peer_addr,
 715                    sizeof(con->peer_addr)) != 0 &&
 716             !(ceph_addr_is_blank(&con->v1.actual_peer_addr) &&
 717               con->v1.actual_peer_addr.nonce == con->peer_addr.nonce)) {
 718                 pr_warn("wrong peer, want %s/%u, got %s/%u\n",
 719                         ceph_pr_addr(&con->peer_addr),
 720                         le32_to_cpu(con->peer_addr.nonce),
 721                         ceph_pr_addr(&con->v1.actual_peer_addr),
 722                         le32_to_cpu(con->v1.actual_peer_addr.nonce));
 723                 con->error_msg = "wrong peer at address";
 724                 return -1;
 725         }
 726
 727         /*
 728          * did we learn our address?
 729          */
 730         if (ceph_addr_is_blank(my_addr)) {
 731                 memcpy(&my_addr->in_addr,
 732                        &con->v1.peer_addr_for_me.in_addr,
 733                        sizeof(con->v1.peer_addr_for_me.in_addr));
 734                 ceph_addr_set_port(my_addr, 0);
 735                 ceph_encode_my_addr(con->msgr);
 736                 dout("process_banner learned my addr is %s\n",
 737                      ceph_pr_addr(my_addr));
 738         }
 739
 740         return 0;
 741 }
 742
 743 static int process_connect(struct ceph_connection *con)
 744 {
 745         u64 sup_feat = from_msgr(con->msgr)->supported_features;
 746         u64 req_feat = from_msgr(con->msgr)->required_features;
 747         u64 server_feat = le64_to_cpu(con->v1.in_reply.features);
 748         int ret;
 749
 750         dout("process_connect on %p tag %d\n", con, con->v1.in_tag);
 751
 752         if (con->v1.auth) {
 753                 int len = le32_to_cpu(con->v1.in_reply.authorizer_len);
 754
 755                 /*
 756                  * Any connection that defines ->get_authorizer()
 757                  * should also define ->add_authorizer_challenge() and
 758                  * ->verify_authorizer_reply().
 759                  *
 760                  * See get_connect_authorizer().
 761                  */
 762                 if (con->v1.in_reply.tag ==
 763                                 CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
 764                         ret = con->ops->add_authorizer_challenge(
 765                                 con, con->v1.auth->authorizer_reply_buf, len);
 766                         if (ret < 0)
 767                                 return ret;
 768
 769                         con_out_kvec_reset(con);
 770                         __prepare_write_connect(con);
 771                         prepare_read_connect(con);
 772                         return 0;
 773                 }
 774
 775                 if (len) {
 776                         ret = con->ops->verify_authorizer_reply(con);
 777                         if (ret < 0) {
 778                                 con->error_msg = "bad authorize reply";
 779                                 return ret;
 780                         }
 781                 }
 782         }
 783
 784         switch (con->v1.in_reply.tag) {
 785         case CEPH_MSGR_TAG_FEATURES:
 786                 pr_err("%s%lld %s feature set mismatch,"
 787                        " my %llx < server's %llx, missing %llx\n",
 788                        ENTITY_NAME(con->peer_name),
 789                        ceph_pr_addr(&con->peer_addr),
 790                        sup_feat, server_feat, server_feat & ~sup_feat);
 791                 con->error_msg = "missing required protocol features";
 792                 return -1;
 793
 794         case CEPH_MSGR_TAG_BADPROTOVER:
 795                 pr_err("%s%lld %s protocol version mismatch,"
 796                        " my %d != server's %d\n",
 797                        ENTITY_NAME(con->peer_name),
 798                        ceph_pr_addr(&con->peer_addr),
 799                        le32_to_cpu(con->v1.out_connect.protocol_version),
 800                        le32_to_cpu(con->v1.in_reply.protocol_version));
 801                 con->error_msg = "protocol version mismatch";
 802                 return -1;
 803
 804         case CEPH_MSGR_TAG_BADAUTHORIZER:
 805                 con->v1.auth_retry++;
 806                 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
 807                      con->v1.auth_retry);
 808                 if (con->v1.auth_retry == 2) {
 809                         con->error_msg = "connect authorization failure";
 810                         return -1;
 811                 }
 812                 con_out_kvec_reset(con);
 813                 ret = prepare_write_connect(con);
 814                 if (ret < 0)
 815                         return ret;
 816                 prepare_read_connect(con);
 817                 break;
 818
 819         case CEPH_MSGR_TAG_RESETSESSION:
 820                 /*
 821                  * If we connected with a large connect_seq but the peer
 822                  * has no record of a session with us (no connection, or
 823                  * connect_seq == 0), they will send RESETSESION to indicate
 824                  * that they must have reset their session, and may have
 825                  * dropped messages.
 826                  */
 827                 dout("process_connect got RESET peer seq %u\n",
 828                      le32_to_cpu(con->v1.in_reply.connect_seq));
 829                 pr_info("%s%lld %s session reset\n",
 830                         ENTITY_NAME(con->peer_name),
 831                         ceph_pr_addr(&con->peer_addr));
 832                 ceph_con_reset_session(con);
 833                 con_out_kvec_reset(con);
 834                 ret = prepare_write_connect(con);
 835                 if (ret < 0)
 836                         return ret;
 837                 prepare_read_connect(con);
 838
 839                 /* Tell ceph about it. */
 840                 mutex_unlock(&con->mutex);
 841                 if (con->ops->peer_reset)
 842                         con->ops->peer_reset(con);
 843                 mutex_lock(&con->mutex);
 844                 if (con->state != CEPH_CON_S_V1_CONNECT_MSG)
 845                         return -EAGAIN;
 846                 break;
 847
 848         case CEPH_MSGR_TAG_RETRY_SESSION:
 849                 /*
 850                  * If we sent a smaller connect_seq than the peer has, try
 851                  * again with a larger value.
 852                  */
 853                 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
 854                      le32_to_cpu(con->v1.out_connect.connect_seq),
 855                      le32_to_cpu(con->v1.in_reply.connect_seq));
 856                 con->v1.connect_seq = le32_to_cpu(con->v1.in_reply.connect_seq);
 857                 con_out_kvec_reset(con);
 858                 ret = prepare_write_connect(con);
 859                 if (ret < 0)
 860                         return ret;
 861                 prepare_read_connect(con);
 862                 break;
 863
 864         case CEPH_MSGR_TAG_RETRY_GLOBAL:
 865                 /*
 866                  * If we sent a smaller global_seq than the peer has, try
 867                  * again with a larger value.
 868                  */
 869                 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
 870                      con->v1.peer_global_seq,
 871                      le32_to_cpu(con->v1.in_reply.global_seq));
 872                 ceph_get_global_seq(con->msgr,
 873                                     le32_to_cpu(con->v1.in_reply.global_seq));
 874                 con_out_kvec_reset(con);
 875                 ret = prepare_write_connect(con);
 876                 if (ret < 0)
 877                         return ret;
 878                 prepare_read_connect(con);
 879                 break;
 880
 881         case CEPH_MSGR_TAG_SEQ:
 882         case CEPH_MSGR_TAG_READY:
 883                 if (req_feat & ~server_feat) {
 884                         pr_err("%s%lld %s protocol feature mismatch,"
 885                                " my required %llx > server's %llx, need %llx\n",
 886                                ENTITY_NAME(con->peer_name),
 887                                ceph_pr_addr(&con->peer_addr),
 888                                req_feat, server_feat, req_feat & ~server_feat);
 889                         con->error_msg = "missing required protocol features";
 890                         return -1;
 891                 }
 892
 893                 WARN_ON(con->state != CEPH_CON_S_V1_CONNECT_MSG);
 894                 con->state = CEPH_CON_S_OPEN;
 895                 con->v1.auth_retry = 0;    /* we authenticated; clear flag */
 896                 con->v1.peer_global_seq =
 897                         le32_to_cpu(con->v1.in_reply.global_seq);
 898                 con->v1.connect_seq++;
 899                 con->peer_features = server_feat;
 900                 dout("process_connect got READY gseq %d cseq %d (%d)\n",
 901                      con->v1.peer_global_seq,
 902                      le32_to_cpu(con->v1.in_reply.connect_seq),
 903                      con->v1.connect_seq);
 904                 WARN_ON(con->v1.connect_seq !=
 905                         le32_to_cpu(con->v1.in_reply.connect_seq));
 906
 907                 if (con->v1.in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
 908                         ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
 909
 910                 con->delay = 0;      /* reset backoff memory */
 911
 912                 if (con->v1.in_reply.tag == CEPH_MSGR_TAG_SEQ) {
 913                         prepare_write_seq(con);
 914                         prepare_read_seq(con);
 915                 } else {
 916                         prepare_read_tag(con);
 917                 }
 918                 break;
 919
 920         case CEPH_MSGR_TAG_WAIT:
 921                 /*
 922                  * If there is a connection race (we are opening
 923                  * connections to each other), one of us may just have
 924                  * to WAIT.  This shouldn't happen if we are the
 925                  * client.
 926                  */
 927                 con->error_msg = "protocol error, got WAIT as client";
 928                 return -1;
 929
 930         default:
 931                 con->error_msg = "protocol error, garbage tag during connect";
 932                 return -1;
 933         }
 934         return 0;
 935 }
 936
 937 /*
 938  * read (part of) an ack
 939  */
 940 static int read_partial_ack(struct ceph_connection *con)
 941 {
 942         int size = sizeof(con->v1.in_temp_ack);
 943         int end = size;
 944
 945         return read_partial(con, end, size, &con->v1.in_temp_ack);
 946 }
 947
 948 /*
 949  * We can finally discard anything that's been acked.
 950  */
 951 static void process_ack(struct ceph_connection *con)
 952 {
 953         u64 ack = le64_to_cpu(con->v1.in_temp_ack);
 954
 955         if (con->v1.in_tag == CEPH_MSGR_TAG_ACK)
 956                 ceph_con_discard_sent(con, ack);
 957         else
 958                 ceph_con_discard_requeued(con, ack);
 959
 960         prepare_read_tag(con);
 961 }
 962
 963 static int read_partial_message_chunk(struct ceph_connection *con,
 964                                       struct kvec *section,
 965                                       unsigned int sec_len, u32 *crc)
 966 {
 967         int ret, left;
 968
 969         BUG_ON(!section);
 970
 971         while (section->iov_len < sec_len) {
 972                 BUG_ON(section->iov_base == NULL);
 973                 left = sec_len - section->iov_len;
 974                 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
 975                                        section->iov_len, left);
 976                 if (ret <= 0)
 977                         return ret;
 978                 section->iov_len += ret;
 979         }
 980         if (section->iov_len == sec_len)
 981                 *crc = crc32c(*crc, section->iov_base, section->iov_len);
 982
 983         return 1;
 984 }
 985
 986 static inline int read_partial_message_section(struct ceph_connection *con,
 987                                                struct kvec *section,
 988                                                unsigned int sec_len, u32 *crc)
 989 {
 990         *crc = 0;
 991         return read_partial_message_chunk(con, section, sec_len, crc);
 992 }
 993
 994 static int read_sparse_msg_extent(struct ceph_connection *con, u32 *crc)
 995 {
 996         struct ceph_msg_data_cursor *cursor = &con->in_msg->cursor;
 997         bool do_bounce = ceph_test_opt(from_msgr(con->msgr), RXBOUNCE);
 998
 999         if (do_bounce && unlikely(!con->bounce_page)) {
1000                 con->bounce_page = alloc_page(GFP_NOIO);
1001                 if (!con->bounce_page) {
1002                         pr_err("failed to allocate bounce page\n");
1003                         return -ENOMEM;
1004                 }
1005         }
1006
1007         while (cursor->sr_resid > 0) {
1008                 struct page *page, *rpage;
1009                 size_t off, len;
1010                 int ret;
1011
1012                 page = ceph_msg_data_next(cursor, &off, &len);
1013                 rpage = do_bounce ? con->bounce_page : page;
1014
1015                 /* clamp to what remains in extent */
1016                 len = min_t(int, len, cursor->sr_resid);
1017                 ret = ceph_tcp_recvpage(con->sock, rpage, (int)off, len);
1018                 if (ret <= 0)
1019                         return ret;
1020                 *crc = ceph_crc32c_page(*crc, rpage, off, ret);
1021                 ceph_msg_data_advance(cursor, (size_t)ret);
1022                 cursor->sr_resid -= ret;
1023                 if (do_bounce)
1024                         memcpy_page(page, off, rpage, off, ret);
1025         }
1026         return 1;
1027 }
1028
1029 static int read_sparse_msg_data(struct ceph_connection *con)
1030 {
1031         struct ceph_msg_data_cursor *cursor = &con->in_msg->cursor;
1032         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1033         u32 crc = 0;
1034         int ret = 1;
1035
1036         if (do_datacrc)
1037                 crc = con->in_data_crc;
1038
1039         do {
1040                 if (con->v1.in_sr_kvec.iov_base)
1041                         ret = read_partial_message_chunk(con,
1042                                                          &con->v1.in_sr_kvec,
1043                                                          con->v1.in_sr_len,
1044                                                          &crc);
1045                 else if (cursor->sr_resid > 0)
1046                         ret = read_sparse_msg_extent(con, &crc);
1047
1048                 if (ret <= 0) {
1049                         if (do_datacrc)
1050                                 con->in_data_crc = crc;
1051                         return ret;
1052                 }
1053
1054                 memset(&con->v1.in_sr_kvec, 0, sizeof(con->v1.in_sr_kvec));
1055                 ret = con->ops->sparse_read(con, cursor,
1056                                 (char **)&con->v1.in_sr_kvec.iov_base);
1057                 con->v1.in_sr_len = ret;
1058         } while (ret > 0);
1059
1060         if (do_datacrc)
1061                 con->in_data_crc = crc;
1062
1063         return ret < 0 ? ret : 1;  /* must return > 0 to indicate success */
1064 }
1065
1066 static int read_partial_msg_data(struct ceph_connection *con)
1067 {
1068         struct ceph_msg_data_cursor *cursor = &con->in_msg->cursor;
1069         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1070         struct page *page;
1071         size_t page_offset;
1072         size_t length;
1073         u32 crc = 0;
1074         int ret;
1075
1076         if (do_datacrc)
1077                 crc = con->in_data_crc;
1078         while (cursor->total_resid) {
1079                 if (!cursor->resid) {
1080                         ceph_msg_data_advance(cursor, 0);
1081                         continue;
1082                 }
1083
1084                 page = ceph_msg_data_next(cursor, &page_offset, &length);
1085                 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
1086                 if (ret <= 0) {
1087                         if (do_datacrc)
1088                                 con->in_data_crc = crc;
1089
1090                         return ret;
1091                 }
1092
1093                 if (do_datacrc)
1094                         crc = ceph_crc32c_page(crc, page, page_offset, ret);
1095                 ceph_msg_data_advance(cursor, (size_t)ret);
1096         }
1097         if (do_datacrc)
1098                 con->in_data_crc = crc;
1099
1100         return 1;       /* must return > 0 to indicate success */
1101 }
1102
1103 static int read_partial_msg_data_bounce(struct ceph_connection *con)
1104 {
1105         struct ceph_msg_data_cursor *cursor = &con->in_msg->cursor;
1106         struct page *page;
1107         size_t off, len;
1108         u32 crc;
1109         int ret;
1110
1111         if (unlikely(!con->bounce_page)) {
1112                 con->bounce_page = alloc_page(GFP_NOIO);
1113                 if (!con->bounce_page) {
1114                         pr_err("failed to allocate bounce page\n");
1115                         return -ENOMEM;
1116                 }
1117         }
1118
1119         crc = con->in_data_crc;
1120         while (cursor->total_resid) {
1121                 if (!cursor->resid) {
1122                         ceph_msg_data_advance(cursor, 0);
1123                         continue;
1124                 }
1125
1126                 page = ceph_msg_data_next(cursor, &off, &len);
1127                 ret = ceph_tcp_recvpage(con->sock, con->bounce_page, 0, len);
1128                 if (ret <= 0) {
1129                         con->in_data_crc = crc;
1130                         return ret;
1131                 }
1132
1133                 crc = crc32c(crc, page_address(con->bounce_page), ret);
1134                 memcpy_to_page(page, off, page_address(con->bounce_page), ret);
1135
1136                 ceph_msg_data_advance(cursor, ret);
1137         }
1138         con->in_data_crc = crc;
1139
1140         return 1;       /* must return > 0 to indicate success */
1141 }
1142
1143 /*
1144  * read (part of) a message.
1145  */
1146 static int read_partial_message(struct ceph_connection *con)
1147 {
1148         struct ceph_msg *m = con->in_msg;
1149         int size;
1150         int end;
1151         int ret;
1152         unsigned int front_len, middle_len, data_len;
1153         bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1154         bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
1155         u64 seq;
1156         u32 crc;
1157
1158         dout("read_partial_message con %p msg %p\n", con, m);
1159
1160         /* header */
1161         size = sizeof(con->v1.in_hdr);
1162         end = size;
1163         ret = read_partial(con, end, size, &con->v1.in_hdr);
1164         if (ret <= 0)
1165                 return ret;
1166
1167         crc = crc32c(0, &con->v1.in_hdr, offsetof(struct ceph_msg_header, crc));
1168         if (cpu_to_le32(crc) != con->v1.in_hdr.crc) {
1169                 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
1170                        crc, con->v1.in_hdr.crc);
1171                 return -EBADMSG;
1172         }
1173
1174         front_len = le32_to_cpu(con->v1.in_hdr.front_len);
1175         if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1176                 return -EIO;
1177         middle_len = le32_to_cpu(con->v1.in_hdr.middle_len);
1178         if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
1179                 return -EIO;
1180         data_len = le32_to_cpu(con->v1.in_hdr.data_len);
1181         if (data_len > CEPH_MSG_MAX_DATA_LEN)
1182                 return -EIO;
1183
1184         /* verify seq# */
1185         seq = le64_to_cpu(con->v1.in_hdr.seq);
1186         if ((s64)seq - (s64)con->in_seq < 1) {
1187                 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1188                         ENTITY_NAME(con->peer_name),
1189                         ceph_pr_addr(&con->peer_addr),
1190                         seq, con->in_seq + 1);
1191                 con->v1.in_base_pos = -front_len - middle_len - data_len -
1192                                       sizeof_footer(con);
1193                 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1194                 return 1;
1195         } else if ((s64)seq - (s64)con->in_seq > 1) {
1196                 pr_err("read_partial_message bad seq %lld expected %lld\n",
1197                        seq, con->in_seq + 1);
1198                 con->error_msg = "bad message sequence # for incoming message";
1199                 return -EBADE;
1200         }
1201
1202         /* allocate message? */
1203         if (!con->in_msg) {
1204                 int skip = 0;
1205
1206                 dout("got hdr type %d front %d data %d\n", con->v1.in_hdr.type,
1207                      front_len, data_len);
1208                 ret = ceph_con_in_msg_alloc(con, &con->v1.in_hdr, &skip);
1209                 if (ret < 0)
1210                         return ret;
1211
1212                 BUG_ON((!con->in_msg) ^ skip);
1213                 if (skip) {
1214                         /* skip this message */
1215                         dout("alloc_msg said skip message\n");
1216                         con->v1.in_base_pos = -front_len - middle_len -
1217                                               data_len - sizeof_footer(con);
1218                         con->v1.in_tag = CEPH_MSGR_TAG_READY;
1219                         con->in_seq++;
1220                         return 1;
1221                 }
1222
1223                 BUG_ON(!con->in_msg);
1224                 BUG_ON(con->in_msg->con != con);
1225                 m = con->in_msg;
1226                 m->front.iov_len = 0;    /* haven't read it yet */
1227                 if (m->middle)
1228                         m->middle->vec.iov_len = 0;
1229
1230                 /* prepare for data payload, if any */
1231
1232                 if (data_len)
1233                         prepare_message_data(con->in_msg, data_len);
1234         }
1235
1236         /* front */
1237         ret = read_partial_message_section(con, &m->front, front_len,
1238                                            &con->in_front_crc);
1239         if (ret <= 0)
1240                 return ret;
1241
1242         /* middle */
1243         if (m->middle) {
1244                 ret = read_partial_message_section(con, &m->middle->vec,
1245                                                    middle_len,
1246                                                    &con->in_middle_crc);
1247                 if (ret <= 0)
1248                         return ret;
1249         }
1250
1251         /* (page) data */
1252         if (data_len) {
1253                 if (!m->num_data_items)
1254                         return -EIO;
1255
1256                 if (m->sparse_read)
1257                         ret = read_sparse_msg_data(con);
1258                 else if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE))
1259                         ret = read_partial_msg_data_bounce(con);
1260                 else
1261                         ret = read_partial_msg_data(con);
1262                 if (ret <= 0)
1263                         return ret;
1264         }
1265
1266         /* footer */
1267         size = sizeof_footer(con);
1268         end += size;
1269         ret = read_partial(con, end, size, &m->footer);
1270         if (ret <= 0)
1271                 return ret;
1272
1273         if (!need_sign) {
1274                 m->footer.flags = m->old_footer.flags;
1275                 m->footer.sig = 0;
1276         }
1277
1278         dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1279              m, front_len, m->footer.front_crc, middle_len,
1280              m->footer.middle_crc, data_len, m->footer.data_crc);
1281
1282         /* crc ok? */
1283         if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1284                 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1285                        m, con->in_front_crc, m->footer.front_crc);
1286                 return -EBADMSG;
1287         }
1288         if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1289                 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1290                        m, con->in_middle_crc, m->footer.middle_crc);
1291                 return -EBADMSG;
1292         }
1293         if (do_datacrc &&
1294             (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1295             con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1296                 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1297                        con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1298                 return -EBADMSG;
1299         }
1300
1301         if (need_sign && con->ops->check_message_signature &&
1302             con->ops->check_message_signature(m)) {
1303                 pr_err("read_partial_message %p signature check failed\n", m);
1304                 return -EBADMSG;
1305         }
1306
1307         return 1; /* done! */
1308 }
1309
1310 static int read_keepalive_ack(struct ceph_connection *con)
1311 {
1312         struct ceph_timespec ceph_ts;
1313         size_t size = sizeof(ceph_ts);
1314         int ret = read_partial(con, size, size, &ceph_ts);
1315         if (ret <= 0)
1316                 return ret;
1317         ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
1318         prepare_read_tag(con);
1319         return 1;
1320 }
1321
1322 /*
1323  * Read what we can from the socket.
1324  */
1325 int ceph_con_v1_try_read(struct ceph_connection *con)
1326 {
1327         int ret = -1;
1328
1329 more:
1330         dout("try_read start %p state %d\n", con, con->state);
1331         if (con->state != CEPH_CON_S_V1_BANNER &&
1332             con->state != CEPH_CON_S_V1_CONNECT_MSG &&
1333             con->state != CEPH_CON_S_OPEN)
1334                 return 0;
1335
1336         BUG_ON(!con->sock);
1337
1338         dout("try_read tag %d in_base_pos %d\n", con->v1.in_tag,
1339              con->v1.in_base_pos);
1340
1341         if (con->state == CEPH_CON_S_V1_BANNER) {
1342                 ret = read_partial_banner(con);
1343                 if (ret <= 0)
1344                         goto out;
1345                 ret = process_banner(con);
1346                 if (ret < 0)
1347                         goto out;
1348
1349                 con->state = CEPH_CON_S_V1_CONNECT_MSG;
1350
1351                 /*
1352                  * Received banner is good, exchange connection info.
1353                  * Do not reset out_kvec, as sending our banner raced
1354                  * with receiving peer banner after connect completed.
1355                  */
1356                 ret = prepare_write_connect(con);
1357                 if (ret < 0)
1358                         goto out;
1359                 prepare_read_connect(con);
1360
1361                 /* Send connection info before awaiting response */
1362                 goto out;
1363         }
1364
1365         if (con->state == CEPH_CON_S_V1_CONNECT_MSG) {
1366                 ret = read_partial_connect(con);
1367                 if (ret <= 0)
1368                         goto out;
1369                 ret = process_connect(con);
1370                 if (ret < 0)
1371                         goto out;
1372                 goto more;
1373         }
1374
1375         WARN_ON(con->state != CEPH_CON_S_OPEN);
1376
1377         if (con->v1.in_base_pos < 0) {
1378                 /*
1379                  * skipping + discarding content.
1380                  */
1381                 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->v1.in_base_pos);
1382                 if (ret <= 0)
1383                         goto out;
1384                 dout("skipped %d / %d bytes\n", ret, -con->v1.in_base_pos);
1385                 con->v1.in_base_pos += ret;
1386                 if (con->v1.in_base_pos)
1387                         goto more;
1388         }
1389         if (con->v1.in_tag == CEPH_MSGR_TAG_READY) {
1390                 /*
1391                  * what's next?
1392                  */
1393                 ret = ceph_tcp_recvmsg(con->sock, &con->v1.in_tag, 1);
1394                 if (ret <= 0)
1395                         goto out;
1396                 dout("try_read got tag %d\n", con->v1.in_tag);
1397                 switch (con->v1.in_tag) {
1398                 case CEPH_MSGR_TAG_MSG:
1399                         prepare_read_message(con);
1400                         break;
1401                 case CEPH_MSGR_TAG_ACK:
1402                         prepare_read_ack(con);
1403                         break;
1404                 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
1405                         prepare_read_keepalive_ack(con);
1406                         break;
1407                 case CEPH_MSGR_TAG_CLOSE:
1408                         ceph_con_close_socket(con);
1409                         con->state = CEPH_CON_S_CLOSED;
1410                         goto out;
1411                 default:
1412                         goto bad_tag;
1413                 }
1414         }
1415         if (con->v1.in_tag == CEPH_MSGR_TAG_MSG) {
1416                 ret = read_partial_message(con);
1417                 if (ret <= 0) {
1418                         switch (ret) {
1419                         case -EBADMSG:
1420                                 con->error_msg = "bad crc/signature";
1421                                 fallthrough;
1422                         case -EBADE:
1423                                 ret = -EIO;
1424                                 break;
1425                         case -EIO:
1426                                 con->error_msg = "io error";
1427                                 break;
1428                         }
1429                         goto out;
1430                 }
1431                 if (con->v1.in_tag == CEPH_MSGR_TAG_READY)
1432                         goto more;
1433                 ceph_con_process_message(con);
1434                 if (con->state == CEPH_CON_S_OPEN)
1435                         prepare_read_tag(con);
1436                 goto more;
1437         }
1438         if (con->v1.in_tag == CEPH_MSGR_TAG_ACK ||
1439             con->v1.in_tag == CEPH_MSGR_TAG_SEQ) {
1440                 /*
1441                  * the final handshake seq exchange is semantically
1442                  * equivalent to an ACK
1443                  */
1444                 ret = read_partial_ack(con);
1445                 if (ret <= 0)
1446                         goto out;
1447                 process_ack(con);
1448                 goto more;
1449         }
1450         if (con->v1.in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
1451                 ret = read_keepalive_ack(con);
1452                 if (ret <= 0)
1453                         goto out;
1454                 goto more;
1455         }
1456
1457 out:
1458         dout("try_read done on %p ret %d\n", con, ret);
1459         return ret;
1460
1461 bad_tag:
1462         pr_err("try_read bad tag %d\n", con->v1.in_tag);
1463         con->error_msg = "protocol error, garbage tag";
1464         ret = -1;
1465         goto out;
1466 }
1467
1468 /*
1469  * Write something to the socket.  Called in a worker thread when the
1470  * socket appears to be writeable and we have something ready to send.
1471  */
1472 int ceph_con_v1_try_write(struct ceph_connection *con)
1473 {
1474         int ret = 1;
1475
1476         dout("try_write start %p state %d\n", con, con->state);
1477         if (con->state != CEPH_CON_S_PREOPEN &&
1478             con->state != CEPH_CON_S_V1_BANNER &&
1479             con->state != CEPH_CON_S_V1_CONNECT_MSG &&
1480             con->state != CEPH_CON_S_OPEN)
1481                 return 0;
1482
1483         /* open the socket first? */
1484         if (con->state == CEPH_CON_S_PREOPEN) {
1485                 BUG_ON(con->sock);
1486                 con->state = CEPH_CON_S_V1_BANNER;
1487
1488                 con_out_kvec_reset(con);
1489                 prepare_write_banner(con);
1490                 prepare_read_banner(con);
1491
1492                 BUG_ON(con->in_msg);
1493                 con->v1.in_tag = CEPH_MSGR_TAG_READY;
1494                 dout("try_write initiating connect on %p new state %d\n",
1495                      con, con->state);
1496                 ret = ceph_tcp_connect(con);
1497                 if (ret < 0) {
1498                         con->error_msg = "connect error";
1499                         goto out;
1500                 }
1501         }
1502
1503 more:
1504         dout("try_write out_kvec_bytes %d\n", con->v1.out_kvec_bytes);
1505         BUG_ON(!con->sock);
1506
1507         /* kvec data queued? */
1508         if (con->v1.out_kvec_left) {
1509                 ret = write_partial_kvec(con);
1510                 if (ret <= 0)
1511                         goto out;
1512         }
1513         if (con->v1.out_skip) {
1514                 ret = write_partial_skip(con);
1515                 if (ret <= 0)
1516                         goto out;
1517         }
1518
1519         /* msg pages? */
1520         if (con->out_msg) {
1521                 if (con->v1.out_msg_done) {
1522                         ceph_msg_put(con->out_msg);
1523                         con->out_msg = NULL;   /* we're done with this one */
1524                         goto do_next;
1525                 }
1526
1527                 ret = write_partial_message_data(con);
1528                 if (ret == 1)
1529                         goto more;  /* we need to send the footer, too! */
1530                 if (ret == 0)
1531                         goto out;
1532                 if (ret < 0) {
1533                         dout("try_write write_partial_message_data err %d\n",
1534                              ret);
1535                         goto out;
1536                 }
1537         }
1538
1539 do_next:
1540         if (con->state == CEPH_CON_S_OPEN) {
1541                 if (ceph_con_flag_test_and_clear(con,
1542                                 CEPH_CON_F_KEEPALIVE_PENDING)) {
1543                         prepare_write_keepalive(con);
1544                         goto more;
1545                 }
1546                 /* is anything else pending? */
1547                 if (!list_empty(&con->out_queue)) {
1548                         prepare_write_message(con);
1549                         goto more;
1550                 }
1551                 if (con->in_seq > con->in_seq_acked) {
1552                         prepare_write_ack(con);
1553                         goto more;
1554                 }
1555         }
1556
1557         /* Nothing to do! */
1558         ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1559         dout("try_write nothing else to write.\n");
1560         ret = 0;
1561 out:
1562         dout("try_write done on %p ret %d\n", con, ret);
1563         return ret;
1564 }
1565
1566 void ceph_con_v1_revoke(struct ceph_connection *con)
1567 {
1568         struct ceph_msg *msg = con->out_msg;
1569
1570         WARN_ON(con->v1.out_skip);
1571         /* footer */
1572         if (con->v1.out_msg_done) {
1573                 con->v1.out_skip += con_out_kvec_skip(con);
1574         } else {
1575                 WARN_ON(!msg->data_length);
1576                 con->v1.out_skip += sizeof_footer(con);
1577         }
1578         /* data, middle, front */
1579         if (msg->data_length)
1580                 con->v1.out_skip += msg->cursor.total_resid;
1581         if (msg->middle)
1582                 con->v1.out_skip += con_out_kvec_skip(con);
1583         con->v1.out_skip += con_out_kvec_skip(con);
1584
1585         dout("%s con %p out_kvec_bytes %d out_skip %d\n", __func__, con,
1586              con->v1.out_kvec_bytes, con->v1.out_skip);
1587 }
1588
1589 void ceph_con_v1_revoke_incoming(struct ceph_connection *con)
1590 {
1591         unsigned int front_len = le32_to_cpu(con->v1.in_hdr.front_len);
1592         unsigned int middle_len = le32_to_cpu(con->v1.in_hdr.middle_len);
1593         unsigned int data_len = le32_to_cpu(con->v1.in_hdr.data_len);
1594
1595         /* skip rest of message */
1596         con->v1.in_base_pos = con->v1.in_base_pos -
1597                         sizeof(struct ceph_msg_header) -
1598                         front_len -
1599                         middle_len -
1600                         data_len -
1601                         sizeof(struct ceph_msg_footer);
1602
1603         con->v1.in_tag = CEPH_MSGR_TAG_READY;
1604         con->in_seq++;
1605
1606         dout("%s con %p in_base_pos %d\n", __func__, con, con->v1.in_base_pos);
1607 }
1608
1609 bool ceph_con_v1_opened(struct ceph_connection *con)
1610 {
1611         return con->v1.connect_seq;
1612 }
1613
1614 void ceph_con_v1_reset_session(struct ceph_connection *con)
1615 {
1616         con->v1.connect_seq = 0;
1617         con->v1.peer_global_seq = 0;
1618 }
1619
1620 void ceph_con_v1_reset_protocol(struct ceph_connection *con)
1621 {
1622         con->v1.out_skip = 0;
1623 }