src/common/relayd/relayd.c

   1 /*
   2  * Copyright (C) 2012 - David Goulet <dgoulet@efficios.com>
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License, version 2 only, as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 51
  15  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _GNU_SOURCE
  19 #include <assert.h>
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <string.h>
  23 #include <sys/stat.h>
  24 #include <inttypes.h>
  25
  26 #include <common/common.h>
  27 #include <common/defaults.h>
  28 #include <common/sessiond-comm/relayd.h>
  29 #include <common/index/lttng-index.h>
  30
  31 #include "relayd.h"
  32
  33 /*
  34  * Send command. Fill up the header and append the data.
  35  */
  36 static int send_command(struct lttcomm_relayd_sock *rsock,
  37                 enum lttcomm_relayd_command cmd, void *data, size_t size,
  38                 int flags)
  39 {
  40         int ret;
  41         struct lttcomm_relayd_hdr header;
  42         char *buf;
  43         uint64_t buf_size = sizeof(header);
  44
  45         if (rsock->sock.fd < 0) {
  46                 return -ECONNRESET;
  47         }
  48
  49         if (data) {
  50                 buf_size += size;
  51         }
  52
  53         buf = zmalloc(buf_size);
  54         if (buf == NULL) {
  55                 PERROR("zmalloc relayd send command buf");
  56                 ret = -1;
  57                 goto alloc_error;
  58         }
  59
  60         header.cmd = htobe32(cmd);
  61         header.data_size = htobe64(size);
  62
  63         /* Zeroed for now since not used. */
  64         header.cmd_version = 0;
  65         header.circuit_id = 0;
  66
  67         /* Prepare buffer to send. */
  68         memcpy(buf, &header, sizeof(header));
  69         if (data) {
  70                 memcpy(buf + sizeof(header), data, size);
  71         }
  72
  73         ret = rsock->sock.ops->sendmsg(&rsock->sock, buf, buf_size, flags);
  74         if (ret < 0) {
  75                 ret = -errno;
  76                 goto error;
  77         }
  78
  79         DBG3("Relayd sending command %d of size %" PRIu64, cmd, buf_size);
  80
  81 error:
  82         free(buf);
  83 alloc_error:
  84         return ret;
  85 }
  86
  87 /*
  88  * Receive reply data on socket. This MUST be call after send_command or else
  89  * could result in unexpected behavior(s).
  90  */
  91 static int recv_reply(struct lttcomm_relayd_sock *rsock, void *data, size_t size)
  92 {
  93         int ret;
  94
  95         if (rsock->sock.fd < 0) {
  96                 return -ECONNRESET;
  97         }
  98
  99         DBG3("Relayd waiting for reply of size %zu", size);
 100
 101         ret = rsock->sock.ops->recvmsg(&rsock->sock, data, size, 0);
 102         if (ret <= 0 || ret != size) {
 103                 if (ret == 0) {
 104                         /* Orderly shutdown. */
 105                         DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
 106                 } else {
 107                         DBG("Receiving reply failed on sock %d for size %zu with ret %d",
 108                                         rsock->sock.fd, size, ret);
 109                 }
 110                 /* Always return -1 here and the caller can use errno. */
 111                 ret = -1;
 112                 goto error;
 113         }
 114
 115 error:
 116         return ret;
 117 }
 118
 119 /*
 120  * Send a RELAYD_CREATE_SESSION command to the relayd with the given socket and
 121  * set session_id of the relayd if we have a successful reply from the relayd.
 122  *
 123  * On success, return 0 else a negative value which is either an errno error or
 124  * a lttng error code from the relayd.
 125  */
 126 int relayd_create_session(struct lttcomm_relayd_sock *rsock, uint64_t *session_id)
 127 {
 128         int ret;
 129         struct lttcomm_relayd_status_session reply;
 130
 131         assert(rsock);
 132         assert(session_id);
 133
 134         DBG("Relayd create session");
 135
 136         /* Send command */
 137         ret = send_command(rsock, RELAYD_CREATE_SESSION, NULL, 0, 0);
 138         if (ret < 0) {
 139                 goto error;
 140         }
 141
 142         /* Receive response */
 143         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 144         if (ret < 0) {
 145                 goto error;
 146         }
 147
 148         reply.session_id = be64toh(reply.session_id);
 149         reply.ret_code = be32toh(reply.ret_code);
 150
 151         /* Return session id or negative ret code. */
 152         if (reply.ret_code != LTTNG_OK) {
 153                 ret = -1;
 154                 ERR("Relayd create session replied error %d", reply.ret_code);
 155                 goto error;
 156         } else {
 157                 ret = 0;
 158                 *session_id = reply.session_id;
 159         }
 160
 161         DBG("Relayd session created with id %" PRIu64, reply.session_id);
 162
 163 error:
 164         return ret;
 165 }
 166
 167 /*
 168  * Add stream on the relayd and assign stream handle to the stream_id argument.
 169  *
 170  * On success return 0 else return ret_code negative value.
 171  */
 172 int relayd_add_stream(struct lttcomm_relayd_sock *rsock, const char *channel_name,
 173                 const char *pathname, uint64_t *stream_id,
 174                 uint64_t tracefile_size, uint64_t tracefile_count)
 175 {
 176         int ret;
 177         struct lttcomm_relayd_add_stream msg;
 178         struct lttcomm_relayd_add_stream_2_2 msg_2_2;
 179         struct lttcomm_relayd_status_stream reply;
 180
 181         /* Code flow error. Safety net. */
 182         assert(rsock);
 183         assert(channel_name);
 184         assert(pathname);
 185
 186         DBG("Relayd adding stream for channel name %s", channel_name);
 187
 188         /* Compat with relayd 2.1 */
 189         if (rsock->minor == 1) {
 190                 strncpy(msg.channel_name, channel_name, sizeof(msg.channel_name));
 191                 strncpy(msg.pathname, pathname, sizeof(msg.pathname));
 192
 193                 /* Send command */
 194                 ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg, sizeof(msg), 0);
 195                 if (ret < 0) {
 196                         goto error;
 197                 }
 198         } else {
 199                 /* Compat with relayd 2.2+ */
 200                 strncpy(msg_2_2.channel_name, channel_name, sizeof(msg_2_2.channel_name));
 201                 strncpy(msg_2_2.pathname, pathname, sizeof(msg_2_2.pathname));
 202                 msg_2_2.tracefile_size = htobe64(tracefile_size);
 203                 msg_2_2.tracefile_count = htobe64(tracefile_count);
 204
 205                 /* Send command */
 206                 ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg_2_2, sizeof(msg_2_2), 0);
 207                 if (ret < 0) {
 208                         goto error;
 209                 }
 210         }
 211
 212         /* Waiting for reply */
 213         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 214         if (ret < 0) {
 215                 goto error;
 216         }
 217
 218         /* Back to host bytes order. */
 219         reply.handle = be64toh(reply.handle);
 220         reply.ret_code = be32toh(reply.ret_code);
 221
 222         /* Return session id or negative ret code. */
 223         if (reply.ret_code != LTTNG_OK) {
 224                 ret = -1;
 225                 ERR("Relayd add stream replied error %d", reply.ret_code);
 226         } else {
 227                 /* Success */
 228                 ret = 0;
 229                 *stream_id = reply.handle;
 230         }
 231
 232         DBG("Relayd stream added successfully with handle %" PRIu64,
 233                 reply.handle);
 234
 235 error:
 236         return ret;
 237 }
 238
 239 /*
 240  * Check version numbers on the relayd.
 241  * If major versions are compatible, we assign minor_to_use to the
 242  * minor version of the procotol we are going to use for this session.
 243  *
 244  * Return 0 if compatible else negative value.
 245  */
 246 int relayd_version_check(struct lttcomm_relayd_sock *rsock)
 247 {
 248         int ret;
 249         struct lttcomm_relayd_version msg;
 250
 251         /* Code flow error. Safety net. */
 252         assert(rsock);
 253
 254         DBG("Relayd version check for major.minor %u.%u", rsock->major,
 255                         rsock->minor);
 256
 257         /* Prepare network byte order before transmission. */
 258         msg.major = htobe32(rsock->major);
 259         msg.minor = htobe32(rsock->minor);
 260
 261         /* Send command */
 262         ret = send_command(rsock, RELAYD_VERSION, (void *) &msg, sizeof(msg), 0);
 263         if (ret < 0) {
 264                 goto error;
 265         }
 266
 267         /* Receive response */
 268         ret = recv_reply(rsock, (void *) &msg, sizeof(msg));
 269         if (ret < 0) {
 270                 goto error;
 271         }
 272
 273         /* Set back to host bytes order */
 274         msg.major = be32toh(msg.major);
 275         msg.minor = be32toh(msg.minor);
 276
 277         /*
 278          * Only validate the major version. If the other side is higher,
 279          * communication is not possible. Only major version equal can talk to each
 280          * other. If the minor version differs, the lowest version is used by both
 281          * sides.
 282          */
 283         if (msg.major != rsock->major) {
 284                 /* Not compatible */
 285                 ret = -1;
 286                 DBG2("Relayd version is NOT compatible. Relayd version %u != %u (us)",
 287                                 msg.major, rsock->major);
 288                 goto error;
 289         }
 290
 291         /*
 292          * If the relayd's minor version is higher, it will adapt to our version so
 293          * we can continue to use the latest relayd communication data structure.
 294          * If the received minor version is higher, the relayd should adapt to us.
 295          */
 296         if (rsock->minor > msg.minor) {
 297                 rsock->minor = msg.minor;
 298         }
 299
 300         /* Version number compatible */
 301         DBG2("Relayd version is compatible, using protocol version %u.%u",
 302                         rsock->major, rsock->minor);
 303         ret = 0;
 304
 305 error:
 306         return ret;
 307 }
 308
 309 /*
 310  * Add stream on the relayd and assign stream handle to the stream_id argument.
 311  *
 312  * On success return 0 else return ret_code negative value.
 313  */
 314 int relayd_send_metadata(struct lttcomm_relayd_sock *rsock, size_t len)
 315 {
 316         int ret;
 317
 318         /* Code flow error. Safety net. */
 319         assert(rsock);
 320
 321         DBG("Relayd sending metadata of size %zu", len);
 322
 323         /* Send command */
 324         ret = send_command(rsock, RELAYD_SEND_METADATA, NULL, len, 0);
 325         if (ret < 0) {
 326                 goto error;
 327         }
 328
 329         DBG2("Relayd metadata added successfully");
 330
 331         /*
 332          * After that call, the metadata data MUST be sent to the relayd so the
 333          * receive size on the other end matches the len of the metadata packet
 334          * header. This is why we don't wait for a reply here.
 335          */
 336
 337 error:
 338         return ret;
 339 }
 340
 341 /*
 342  * Connect to relay daemon with an allocated lttcomm_relayd_sock.
 343  */
 344 int relayd_connect(struct lttcomm_relayd_sock *rsock)
 345 {
 346         /* Code flow error. Safety net. */
 347         assert(rsock);
 348
 349         if (!rsock->sock.ops) {
 350                 /*
 351                  * Attempting a connect on a non-initialized socket.
 352                  */
 353                 return -ECONNRESET;
 354         }
 355
 356         DBG3("Relayd connect ...");
 357
 358         return rsock->sock.ops->connect(&rsock->sock);
 359 }
 360
 361 /*
 362  * Close relayd socket with an allocated lttcomm_relayd_sock.
 363  *
 364  * If no socket operations are found, simply return 0 meaning that everything
 365  * is fine. Without operations, the socket can not possibly be opened or used.
 366  * This is possible if the socket was allocated but not created. However, the
 367  * caller could simply use it to store a valid file descriptor for instance
 368  * passed over a Unix socket and call this to cleanup but still without a valid
 369  * ops pointer.
 370  *
 371  * Return the close returned value. On error, a negative value is usually
 372  * returned back from close(2).
 373  */
 374 int relayd_close(struct lttcomm_relayd_sock *rsock)
 375 {
 376         int ret;
 377
 378         /* Code flow error. Safety net. */
 379         assert(rsock);
 380
 381         /* An invalid fd is fine, return success. */
 382         if (rsock->sock.fd < 0) {
 383                 ret = 0;
 384                 goto end;
 385         }
 386
 387         DBG3("Relayd closing socket %d", rsock->sock.fd);
 388
 389         if (rsock->sock.ops) {
 390                 ret = rsock->sock.ops->close(&rsock->sock);
 391         } else {
 392                 /* Default call if no specific ops found. */
 393                 ret = close(rsock->sock.fd);
 394                 if (ret < 0) {
 395                         PERROR("relayd_close default close");
 396                 }
 397         }
 398         rsock->sock.fd = -1;
 399
 400 end:
 401         return ret;
 402 }
 403
 404 /*
 405  * Send data header structure to the relayd.
 406  */
 407 int relayd_send_data_hdr(struct lttcomm_relayd_sock *rsock,
 408                 struct lttcomm_relayd_data_hdr *hdr, size_t size)
 409 {
 410         int ret;
 411
 412         /* Code flow error. Safety net. */
 413         assert(rsock);
 414         assert(hdr);
 415
 416         if (rsock->sock.fd < 0) {
 417                 return -ECONNRESET;
 418         }
 419
 420         DBG3("Relayd sending data header of size %zu", size);
 421
 422         /* Again, safety net */
 423         if (size == 0) {
 424                 size = sizeof(struct lttcomm_relayd_data_hdr);
 425         }
 426
 427         /* Only send data header. */
 428         ret = rsock->sock.ops->sendmsg(&rsock->sock, hdr, size, 0);
 429         if (ret < 0) {
 430                 ret = -errno;
 431                 goto error;
 432         }
 433
 434         /*
 435          * The data MUST be sent right after that command for the receive on the
 436          * other end to match the size in the header.
 437          */
 438
 439 error:
 440         return ret;
 441 }
 442
 443 /*
 444  * Send close stream command to the relayd.
 445  */
 446 int relayd_send_close_stream(struct lttcomm_relayd_sock *rsock, uint64_t stream_id,
 447                 uint64_t last_net_seq_num)
 448 {
 449         int ret;
 450         struct lttcomm_relayd_close_stream msg;
 451         struct lttcomm_relayd_generic_reply reply;
 452
 453         /* Code flow error. Safety net. */
 454         assert(rsock);
 455
 456         DBG("Relayd closing stream id %" PRIu64, stream_id);
 457
 458         msg.stream_id = htobe64(stream_id);
 459         msg.last_net_seq_num = htobe64(last_net_seq_num);
 460
 461         /* Send command */
 462         ret = send_command(rsock, RELAYD_CLOSE_STREAM, (void *) &msg, sizeof(msg), 0);
 463         if (ret < 0) {
 464                 goto error;
 465         }
 466
 467         /* Receive response */
 468         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 469         if (ret < 0) {
 470                 goto error;
 471         }
 472
 473         reply.ret_code = be32toh(reply.ret_code);
 474
 475         /* Return session id or negative ret code. */
 476         if (reply.ret_code != LTTNG_OK) {
 477                 ret = -1;
 478                 ERR("Relayd close stream replied error %d", reply.ret_code);
 479         } else {
 480                 /* Success */
 481                 ret = 0;
 482         }
 483
 484         DBG("Relayd close stream id %" PRIu64 " successfully", stream_id);
 485
 486 error:
 487         return ret;
 488 }
 489
 490 /*
 491  * Check for data availability for a given stream id.
 492  *
 493  * Return 0 if NOT pending, 1 if so and a negative value on error.
 494  */
 495 int relayd_data_pending(struct lttcomm_relayd_sock *rsock, uint64_t stream_id,
 496                 uint64_t last_net_seq_num)
 497 {
 498         int ret;
 499         struct lttcomm_relayd_data_pending msg;
 500         struct lttcomm_relayd_generic_reply reply;
 501
 502         /* Code flow error. Safety net. */
 503         assert(rsock);
 504
 505         DBG("Relayd data pending for stream id %" PRIu64, stream_id);
 506
 507         msg.stream_id = htobe64(stream_id);
 508         msg.last_net_seq_num = htobe64(last_net_seq_num);
 509
 510         /* Send command */
 511         ret = send_command(rsock, RELAYD_DATA_PENDING, (void *) &msg,
 512                         sizeof(msg), 0);
 513         if (ret < 0) {
 514                 goto error;
 515         }
 516
 517         /* Receive response */
 518         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 519         if (ret < 0) {
 520                 goto error;
 521         }
 522
 523         reply.ret_code = be32toh(reply.ret_code);
 524
 525         /* Return session id or negative ret code. */
 526         if (reply.ret_code >= LTTNG_OK) {
 527                 ERR("Relayd data pending replied error %d", reply.ret_code);
 528         }
 529
 530         /* At this point, the ret code is either 1 or 0 */
 531         ret = reply.ret_code;
 532
 533         DBG("Relayd data is %s pending for stream id %" PRIu64,
 534                         ret == 1 ? "" : "NOT", stream_id);
 535
 536 error:
 537         return ret;
 538 }
 539
 540 /*
 541  * Check on the relayd side for a quiescent state on the control socket.
 542  */
 543 int relayd_quiescent_control(struct lttcomm_relayd_sock *rsock,
 544                 uint64_t metadata_stream_id)
 545 {
 546         int ret;
 547         struct lttcomm_relayd_quiescent_control msg;
 548         struct lttcomm_relayd_generic_reply reply;
 549
 550         /* Code flow error. Safety net. */
 551         assert(rsock);
 552
 553         DBG("Relayd checking quiescent control state");
 554
 555         msg.stream_id = htobe64(metadata_stream_id);
 556
 557         /* Send command */
 558         ret = send_command(rsock, RELAYD_QUIESCENT_CONTROL, &msg, sizeof(msg), 0);
 559         if (ret < 0) {
 560                 goto error;
 561         }
 562
 563         /* Receive response */
 564         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 565         if (ret < 0) {
 566                 goto error;
 567         }
 568
 569         reply.ret_code = be32toh(reply.ret_code);
 570
 571         /* Return session id or negative ret code. */
 572         if (reply.ret_code != LTTNG_OK) {
 573                 ret = -1;
 574                 ERR("Relayd quiescent control replied error %d", reply.ret_code);
 575                 goto error;
 576         }
 577
 578         /* Control socket is quiescent */
 579         return 0;
 580
 581 error:
 582         return ret;
 583 }
 584
 585 /*
 586  * Begin a data pending command for a specific session id.
 587  */
 588 int relayd_begin_data_pending(struct lttcomm_relayd_sock *rsock, uint64_t id)
 589 {
 590         int ret;
 591         struct lttcomm_relayd_begin_data_pending msg;
 592         struct lttcomm_relayd_generic_reply reply;
 593
 594         /* Code flow error. Safety net. */
 595         assert(rsock);
 596
 597         DBG("Relayd begin data pending");
 598
 599         msg.session_id = htobe64(id);
 600
 601         /* Send command */
 602         ret = send_command(rsock, RELAYD_BEGIN_DATA_PENDING, &msg, sizeof(msg), 0);
 603         if (ret < 0) {
 604                 goto error;
 605         }
 606
 607         /* Receive response */
 608         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 609         if (ret < 0) {
 610                 goto error;
 611         }
 612
 613         reply.ret_code = be32toh(reply.ret_code);
 614
 615         /* Return session id or negative ret code. */
 616         if (reply.ret_code != LTTNG_OK) {
 617                 ret = -1;
 618                 ERR("Relayd begin data pending replied error %d", reply.ret_code);
 619                 goto error;
 620         }
 621
 622         return 0;
 623
 624 error:
 625         return ret;
 626 }
 627
 628 /*
 629  * End a data pending command for a specific session id.
 630  *
 631  * Return 0 on success and set is_data_inflight to 0 if no data is being
 632  * streamed or 1 if it is the case.
 633  */
 634 int relayd_end_data_pending(struct lttcomm_relayd_sock *rsock, uint64_t id,
 635                 unsigned int *is_data_inflight)
 636 {
 637         int ret;
 638         struct lttcomm_relayd_end_data_pending msg;
 639         struct lttcomm_relayd_generic_reply reply;
 640
 641         /* Code flow error. Safety net. */
 642         assert(rsock);
 643
 644         DBG("Relayd end data pending");
 645
 646         msg.session_id = htobe64(id);
 647
 648         /* Send command */
 649         ret = send_command(rsock, RELAYD_END_DATA_PENDING, &msg, sizeof(msg), 0);
 650         if (ret < 0) {
 651                 goto error;
 652         }
 653
 654         /* Receive response */
 655         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 656         if (ret < 0) {
 657                 goto error;
 658         }
 659
 660         reply.ret_code = be32toh(reply.ret_code);
 661         if (reply.ret_code < 0) {
 662                 ret = reply.ret_code;
 663                 goto error;
 664         }
 665
 666         *is_data_inflight = reply.ret_code;
 667
 668         DBG("Relayd end data pending is data inflight: %d", reply.ret_code);
 669
 670         return 0;
 671
 672 error:
 673         return ret;
 674 }
 675
 676 /*
 677  * Send index to the relayd.
 678  */
 679 int relayd_send_index(struct lttcomm_relayd_sock *rsock,
 680                 struct lttng_packet_index *index, uint64_t relay_stream_id,
 681                 uint64_t net_seq_num)
 682 {
 683         int ret;
 684         struct lttcomm_relayd_index msg;
 685         struct lttcomm_relayd_generic_reply reply;
 686
 687         /* Code flow error. Safety net. */
 688         assert(rsock);
 689
 690         if (rsock->minor < 4) {
 691                 DBG("Not sending indexes before protocol 2.4");
 692                 ret = 0;
 693                 goto error;
 694         }
 695
 696         DBG("Relayd sending index for stream ID %" PRIu64, relay_stream_id);
 697
 698         msg.relay_stream_id = htobe64(relay_stream_id);
 699         msg.net_seq_num = htobe64(net_seq_num);
 700
 701         /* The index is already in big endian. */
 702         msg.packet_size = index->packet_size;
 703         msg.content_size = index->content_size;
 704         msg.timestamp_begin = index->timestamp_begin;
 705         msg.timestamp_end = index->timestamp_end;
 706         msg.events_discarded = index->events_discarded;
 707         msg.stream_id = index->stream_id;
 708
 709         /* Send command */
 710         ret = send_command(rsock, RELAYD_SEND_INDEX, &msg, sizeof(msg), 0);
 711         if (ret < 0) {
 712                 goto error;
 713         }
 714
 715         /* Receive response */
 716         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 717         if (ret < 0) {
 718                 goto error;
 719         }
 720
 721         reply.ret_code = be32toh(reply.ret_code);
 722
 723         /* Return session id or negative ret code. */
 724         if (reply.ret_code != LTTNG_OK) {
 725                 ret = -1;
 726                 ERR("Relayd send index replied error %d", reply.ret_code);
 727         } else {
 728                 /* Success */
 729                 ret = 0;
 730         }
 731
 732 error:
 733         return ret;
 734 }