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