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 _LGPL_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/compat/endian.h>
  29 #include <common/sessiond-comm/relayd.h>
  30 #include <common/index/ctf-index.h>
  31
  32 #include "relayd.h"
  33
  34 /*
  35  * Send command. Fill up the header and append the data.
  36  */
  37 static int send_command(struct lttcomm_relayd_sock *rsock,
  38                 enum lttcomm_relayd_command cmd, void *data, size_t size,
  39                 int flags)
  40 {
  41         int ret;
  42         struct lttcomm_relayd_hdr header;
  43         char *buf;
  44         uint64_t buf_size = sizeof(header);
  45
  46         if (rsock->sock.fd < 0) {
  47                 return -ECONNRESET;
  48         }
  49
  50         if (data) {
  51                 buf_size += size;
  52         }
  53
  54         buf = zmalloc(buf_size);
  55         if (buf == NULL) {
  56                 PERROR("zmalloc relayd send command buf");
  57                 ret = -1;
  58                 goto alloc_error;
  59         }
  60
  61         memset(&header, 0, sizeof(header));
  62         header.cmd = htobe32(cmd);
  63         header.data_size = htobe64(size);
  64
  65         /* Zeroed for now since not used. */
  66         header.cmd_version = 0;
  67         header.circuit_id = 0;
  68
  69         /* Prepare buffer to send. */
  70         memcpy(buf, &header, sizeof(header));
  71         if (data) {
  72                 memcpy(buf + sizeof(header), data, size);
  73         }
  74
  75         DBG3("Relayd sending command %d of size %" PRIu64, (int) cmd, buf_size);
  76         ret = rsock->sock.ops->sendmsg(&rsock->sock, buf, buf_size, flags);
  77         if (ret < 0) {
  78                 PERROR("Failed to send command %d of size %" PRIu64,
  79                                 (int) cmd, buf_size);
  80                 ret = -errno;
  81                 goto error;
  82         }
  83 error:
  84         free(buf);
  85 alloc_error:
  86         return ret;
  87 }
  88
  89 static int recv_reply_ignore(struct lttcomm_relayd_sock *rsock, size_t size)
  90 {
  91         int ret;
  92         size_t cutfoff = 128;
  93
  94         /*
  95          * To prevent ever growing size of recv_reply to ignore, if the number
  96          * of bytes we want to ignore is bigger than `cutoff`, consume half of
  97          * the cutoff. We might block on it but still, most of bytes to ignore
  98          * should already be ready to consume at this point.
  99          *
 100          * This kind of scenario can easily happen on stopped session with a
 101          * live_timer since no actual receive is done on the socket that would
 102          * discard the `ignore` portion.
 103          *
 104          * TCP guarantee in-order transmission both on send and receive so this
 105          * is safe to do.
 106          */
 107         if (rsock->bytes_to_ignore_on_recv >= cutfoff) {
 108                 size_t to_discard = cutfoff / 2;
 109
 110                 ret = rsock->sock.ops->recvmsg(&rsock->sock, NULL, to_discard, MSG_TRUNC);
 111                 if (ret <= 0 || ret != to_discard) {
 112                         if (ret == 0) {
 113                                 /* Orderly shutdown. */
 114                                 DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
 115                         } else {
 116                                 DBG("Receiving reply to discard failed on sock %d for size %zu with ret %d",
 117                                                 rsock->sock.fd, to_discard, ret);
 118                         }
 119                         ret = -1;
 120                         goto error;
 121                 }
 122
 123                 DBG("Force discard of %zu bytes for socket %d", to_discard, rsock->sock.fd);
 124                 rsock->bytes_to_ignore_on_recv -= to_discard;
 125         }
 126
 127         DBG3("Relayd ignore reply of %zu bytes for socket %d.", size, rsock->sock.fd);
 128         /* Do not wait for the current reply to be ignored */
 129         rsock->bytes_to_ignore_on_recv += size;
 130         ret = 0;
 131
 132 error:
 133         return ret;
 134 }
 135
 136 /*
 137  * Receive reply data on socket. This MUST be call after send_command or else
 138  * could result in unexpected behavior(s).
 139  */
 140 static int recv_reply(struct lttcomm_relayd_sock *rsock, void *data, size_t size)
 141 {
 142         int ret;
 143
 144         if (rsock->sock.fd < 0) {
 145                 return -ECONNRESET;
 146         }
 147
 148         /*
 149          * We have to consume the bytes that are marked to ignore.
 150          */
 151         if (rsock->bytes_to_ignore_on_recv != 0) {
 152                 ret = rsock->sock.ops->recvmsg(&rsock->sock, NULL, rsock->bytes_to_ignore_on_recv, MSG_TRUNC);
 153                 if (ret <= 0 || ret != rsock->bytes_to_ignore_on_recv) {
 154                         if (ret == 0) {
 155                                 /* Orderly shutdown. */
 156                                 DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
 157                         } else {
 158                                 DBG("Receiving reply to skip failed on sock %d for size %zu with ret %d",
 159                                                 rsock->sock.fd, rsock->bytes_to_ignore_on_recv, ret);
 160                         }
 161                         ret = -1;
 162                         goto error;
 163                 }
 164                 DBG("Discarded %zu bytes on sock %d", rsock->bytes_to_ignore_on_recv, rsock->sock.fd);
 165                 rsock->bytes_to_ignore_on_recv = 0;
 166         }
 167
 168         DBG3("Relayd waiting for reply of size %zu", size);
 169
 170         ret = rsock->sock.ops->recvmsg(&rsock->sock, data, size, 0);
 171         if (ret <= 0 || ret != size) {
 172                 if (ret == 0) {
 173                         /* Orderly shutdown. */
 174                         DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
 175                 } else {
 176                         DBG("Receiving reply failed on sock %d for size %zu with ret %d",
 177                                         rsock->sock.fd, size, ret);
 178                 }
 179                 /* Always return -1 here and the caller can use errno. */
 180                 ret = -1;
 181                 goto error;
 182         }
 183
 184 error:
 185         return ret;
 186 }
 187
 188 /*
 189  * Starting at 2.4, RELAYD_CREATE_SESSION takes additional parameters to
 190  * support the live reading capability.
 191  */
 192 static int relayd_create_session_2_4(struct lttcomm_relayd_sock *rsock,
 193                 uint64_t *session_id, char *session_name, char *hostname,
 194                 int session_live_timer, unsigned int snapshot)
 195 {
 196         int ret;
 197         struct lttcomm_relayd_create_session_2_4 msg;
 198
 199         if (lttng_strncpy(msg.session_name, session_name,
 200                         sizeof(msg.session_name))) {
 201                 ret = -1;
 202                 goto error;
 203         }
 204         if (lttng_strncpy(msg.hostname, hostname, sizeof(msg.hostname))) {
 205                 ret = -1;
 206                 goto error;
 207         }
 208         msg.live_timer = htobe32(session_live_timer);
 209         msg.snapshot = htobe32(snapshot);
 210
 211         /* Send command */
 212         ret = send_command(rsock, RELAYD_CREATE_SESSION, &msg, sizeof(msg), 0);
 213         if (ret < 0) {
 214                 goto error;
 215         }
 216
 217 error:
 218         return ret;
 219 }
 220
 221 /*
 222  * RELAYD_CREATE_SESSION from 2.1 to 2.3.
 223  */
 224 static int relayd_create_session_2_1(struct lttcomm_relayd_sock *rsock,
 225                 uint64_t *session_id)
 226 {
 227         int ret;
 228
 229         /* Send command */
 230         ret = send_command(rsock, RELAYD_CREATE_SESSION, NULL, 0, 0);
 231         if (ret < 0) {
 232                 goto error;
 233         }
 234
 235 error:
 236         return ret;
 237 }
 238
 239 /*
 240  * Send a RELAYD_CREATE_SESSION command to the relayd with the given socket and
 241  * set session_id of the relayd if we have a successful reply from the relayd.
 242  *
 243  * On success, return 0 else a negative value which is either an errno error or
 244  * a lttng error code from the relayd.
 245  */
 246 int relayd_create_session(struct lttcomm_relayd_sock *rsock, uint64_t *session_id,
 247                 char *session_name, char *hostname, int session_live_timer,
 248                 unsigned int snapshot)
 249 {
 250         int ret;
 251         struct lttcomm_relayd_status_session reply;
 252
 253         assert(rsock);
 254         assert(session_id);
 255
 256         DBG("Relayd create session");
 257
 258         switch(rsock->minor) {
 259                 case 1:
 260                 case 2:
 261                 case 3:
 262                         ret = relayd_create_session_2_1(rsock, session_id);
 263                         break;
 264                 case 4:
 265                 default:
 266                         ret = relayd_create_session_2_4(rsock, session_id, session_name,
 267                                         hostname, session_live_timer, snapshot);
 268                         break;
 269         }
 270
 271         if (ret < 0) {
 272                 goto error;
 273         }
 274
 275         /* Receive response */
 276         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 277         if (ret < 0) {
 278                 goto error;
 279         }
 280
 281         reply.session_id = be64toh(reply.session_id);
 282         reply.ret_code = be32toh(reply.ret_code);
 283
 284         /* Return session id or negative ret code. */
 285         if (reply.ret_code != LTTNG_OK) {
 286                 ret = -1;
 287                 ERR("Relayd create session replied error %d", reply.ret_code);
 288                 goto error;
 289         } else {
 290                 ret = 0;
 291                 *session_id = reply.session_id;
 292         }
 293
 294         DBG("Relayd session created with id %" PRIu64, reply.session_id);
 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_add_stream(struct lttcomm_relayd_sock *rsock, const char *channel_name,
 306                 const char *pathname, uint64_t *stream_id,
 307                 uint64_t tracefile_size, uint64_t tracefile_count)
 308 {
 309         int ret;
 310         struct lttcomm_relayd_add_stream msg;
 311         struct lttcomm_relayd_add_stream_2_2 msg_2_2;
 312         struct lttcomm_relayd_status_stream reply;
 313
 314         /* Code flow error. Safety net. */
 315         assert(rsock);
 316         assert(channel_name);
 317         assert(pathname);
 318
 319         DBG("Relayd adding stream for channel name %s", channel_name);
 320
 321         /* Compat with relayd 2.1 */
 322         if (rsock->minor == 1) {
 323                 memset(&msg, 0, sizeof(msg));
 324                 if (lttng_strncpy(msg.channel_name, channel_name,
 325                                 sizeof(msg.channel_name))) {
 326                         ret = -1;
 327                         goto error;
 328                 }
 329                 if (lttng_strncpy(msg.pathname, pathname,
 330                                 sizeof(msg.pathname))) {
 331                         ret = -1;
 332                         goto error;
 333                 }
 334
 335                 /* Send command */
 336                 ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg, sizeof(msg), 0);
 337                 if (ret < 0) {
 338                         goto error;
 339                 }
 340         } else {
 341                 memset(&msg_2_2, 0, sizeof(msg_2_2));
 342                 /* Compat with relayd 2.2+ */
 343                 if (lttng_strncpy(msg_2_2.channel_name, channel_name,
 344                                 sizeof(msg_2_2.channel_name))) {
 345                         ret = -1;
 346                         goto error;
 347                 }
 348                 if (lttng_strncpy(msg_2_2.pathname, pathname,
 349                                 sizeof(msg_2_2.pathname))) {
 350                         ret = -1;
 351                         goto error;
 352                 }
 353                 msg_2_2.tracefile_size = htobe64(tracefile_size);
 354                 msg_2_2.tracefile_count = htobe64(tracefile_count);
 355
 356                 /* Send command */
 357                 ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg_2_2, sizeof(msg_2_2), 0);
 358                 if (ret < 0) {
 359                         goto error;
 360                 }
 361         }
 362
 363         /* Waiting for reply */
 364         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 365         if (ret < 0) {
 366                 goto error;
 367         }
 368
 369         /* Back to host bytes order. */
 370         reply.handle = be64toh(reply.handle);
 371         reply.ret_code = be32toh(reply.ret_code);
 372
 373         /* Return session id or negative ret code. */
 374         if (reply.ret_code != LTTNG_OK) {
 375                 ret = -1;
 376                 ERR("Relayd add stream replied error %d", reply.ret_code);
 377         } else {
 378                 /* Success */
 379                 ret = 0;
 380                 *stream_id = reply.handle;
 381         }
 382
 383         DBG("Relayd stream added successfully with handle %" PRIu64,
 384                 reply.handle);
 385
 386 error:
 387         return ret;
 388 }
 389
 390 /*
 391  * Add stream on the relayd. Send part.
 392  *
 393  * On success return 0 else return ret_code negative value.
 394  */
 395 int relayd_add_stream_send(struct lttcomm_relayd_sock *rsock, const char *channel_name,
 396                 const char *pathname, uint64_t tracefile_size, uint64_t tracefile_count)
 397 {
 398         int ret;
 399         struct lttcomm_relayd_add_stream msg;
 400         struct lttcomm_relayd_add_stream_2_2 msg_2_2;
 401
 402         /* Code flow error. Safety net. */
 403         assert(rsock);
 404         assert(channel_name);
 405         assert(pathname);
 406
 407         DBG("Relayd adding stream for channel name %s. Part send", channel_name);
 408
 409         /* Compat with relayd 2.1 */
 410         if (rsock->minor == 1) {
 411                 memset(&msg, 0, sizeof(msg));
 412                 if (lttng_strncpy(msg.channel_name, channel_name,
 413                                 sizeof(msg.channel_name))) {
 414                         ret = -1;
 415                         goto error;
 416                 }
 417                 if (lttng_strncpy(msg.pathname, pathname,
 418                                 sizeof(msg.pathname))) {
 419                         ret = -1;
 420                         goto error;
 421                 }
 422
 423                 /* Send command */
 424                 ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg, sizeof(msg), 0);
 425                 if (ret < 0) {
 426                         goto error;
 427                 }
 428         } else {
 429                 memset(&msg_2_2, 0, sizeof(msg_2_2));
 430                 /* Compat with relayd 2.2+ */
 431                 if (lttng_strncpy(msg_2_2.channel_name, channel_name,
 432                                 sizeof(msg_2_2.channel_name))) {
 433                         ret = -1;
 434                         goto error;
 435                 }
 436                 if (lttng_strncpy(msg_2_2.pathname, pathname,
 437                                 sizeof(msg_2_2.pathname))) {
 438                         ret = -1;
 439                         goto error;
 440                 }
 441                 msg_2_2.tracefile_size = htobe64(tracefile_size);
 442                 msg_2_2.tracefile_count = htobe64(tracefile_count);
 443
 444                 /* Send command */
 445                 ret = send_command(rsock, RELAYD_ADD_STREAM, (void *) &msg_2_2, sizeof(msg_2_2), 0);
 446                 if (ret < 0) {
 447                         goto error;
 448                 }
 449         }
 450
 451         DBG("Relayd add stream sent for channel name %s.", channel_name);
 452         ret = 0;
 453
 454 error:
 455         return ret;
 456 }
 457
 458 int relayd_add_stream_rcv(struct lttcomm_relayd_sock *rsock, uint64_t *_stream_id)
 459 {
 460         int ret;
 461         struct lttcomm_relayd_status_stream reply;
 462
 463         /* Code flow error. Safety net. */
 464         assert(rsock);
 465
 466         /* Waiting for reply */
 467         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 468         if (ret < 0) {
 469                 goto error;
 470         }
 471
 472         /* Back to host bytes order. */
 473         reply.handle = be64toh(reply.handle);
 474         reply.ret_code = be32toh(reply.ret_code);
 475
 476         /* Return session id or negative ret code. */
 477         if (reply.ret_code != LTTNG_OK) {
 478                 ret = -1;
 479                 ERR("Relayd add stream replied error %d", reply.ret_code);
 480         } else {
 481                 /* Success */
 482                 ret = 0;
 483                 *_stream_id = reply.handle;
 484         }
 485
 486         DBG("Relayd stream added successfully with handle %" PRIu64,
 487                 reply.handle);
 488
 489 error:
 490         return ret;
 491 }
 492
 493 /*
 494  * Inform the relay that all the streams for the current channel has been sent.
 495  *
 496  * On success return 0 else return ret_code negative value.
 497  */
 498 int relayd_streams_sent(struct lttcomm_relayd_sock *rsock)
 499 {
 500         int ret;
 501         struct lttcomm_relayd_generic_reply reply;
 502
 503         /* Code flow error. Safety net. */
 504         assert(rsock);
 505
 506         DBG("Relayd sending streams sent.");
 507
 508         /* This feature was introduced in 2.4, ignore it for earlier versions. */
 509         if (rsock->minor < 4) {
 510                 ret = 0;
 511                 goto end;
 512         }
 513
 514         /* Send command */
 515         ret = send_command(rsock, RELAYD_STREAMS_SENT, NULL, 0, 0);
 516         if (ret < 0) {
 517                 goto error;
 518         }
 519
 520         /* Waiting for reply */
 521         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 522         if (ret < 0) {
 523                 goto error;
 524         }
 525
 526         /* Back to host bytes order. */
 527         reply.ret_code = be32toh(reply.ret_code);
 528
 529         /* Return session id or negative ret code. */
 530         if (reply.ret_code != LTTNG_OK) {
 531                 ret = -1;
 532                 ERR("Relayd streams sent replied error %d", reply.ret_code);
 533                 goto error;
 534         } else {
 535                 /* Success */
 536                 ret = 0;
 537         }
 538
 539         DBG("Relayd streams sent success");
 540
 541 error:
 542 end:
 543         return ret;
 544 }
 545
 546 /*
 547  * Check version numbers on the relayd.
 548  * If major versions are compatible, we assign minor_to_use to the
 549  * minor version of the procotol we are going to use for this session.
 550  *
 551  * Return 0 if the two daemons are compatible, LTTNG_ERR_RELAYD_VERSION_FAIL
 552  * otherwise, or a negative value on network errors.
 553  */
 554 int relayd_version_check(struct lttcomm_relayd_sock *rsock)
 555 {
 556         int ret;
 557         struct lttcomm_relayd_version msg;
 558
 559         /* Code flow error. Safety net. */
 560         assert(rsock);
 561
 562         DBG("Relayd version check for major.minor %u.%u", rsock->major,
 563                         rsock->minor);
 564
 565         memset(&msg, 0, sizeof(msg));
 566         /* Prepare network byte order before transmission. */
 567         msg.major = htobe32(rsock->major);
 568         msg.minor = htobe32(rsock->minor);
 569
 570         /* Send command */
 571         ret = send_command(rsock, RELAYD_VERSION, (void *) &msg, sizeof(msg), 0);
 572         if (ret < 0) {
 573                 goto error;
 574         }
 575
 576         /* Receive response */
 577         ret = recv_reply(rsock, (void *) &msg, sizeof(msg));
 578         if (ret < 0) {
 579                 goto error;
 580         }
 581
 582         /* Set back to host bytes order */
 583         msg.major = be32toh(msg.major);
 584         msg.minor = be32toh(msg.minor);
 585
 586         /*
 587          * Only validate the major version. If the other side is higher,
 588          * communication is not possible. Only major version equal can talk to each
 589          * other. If the minor version differs, the lowest version is used by both
 590          * sides.
 591          */
 592         if (msg.major != rsock->major) {
 593                 /* Not compatible */
 594                 ret = LTTNG_ERR_RELAYD_VERSION_FAIL;
 595                 DBG2("Relayd version is NOT compatible. Relayd version %u != %u (us)",
 596                                 msg.major, rsock->major);
 597                 goto error;
 598         }
 599
 600         /*
 601          * If the relayd's minor version is higher, it will adapt to our version so
 602          * we can continue to use the latest relayd communication data structure.
 603          * If the received minor version is higher, the relayd should adapt to us.
 604          */
 605         if (rsock->minor > msg.minor) {
 606                 rsock->minor = msg.minor;
 607         }
 608
 609         /* Version number compatible */
 610         DBG2("Relayd version is compatible, using protocol version %u.%u",
 611                         rsock->major, rsock->minor);
 612         ret = 0;
 613
 614 error:
 615         return ret;
 616 }
 617
 618 /*
 619  * Add stream on the relayd and assign stream handle to the stream_id argument.
 620  *
 621  * On success return 0 else return ret_code negative value.
 622  */
 623 int relayd_send_metadata(struct lttcomm_relayd_sock *rsock, size_t len)
 624 {
 625         int ret;
 626
 627         /* Code flow error. Safety net. */
 628         assert(rsock);
 629
 630         DBG("Relayd sending metadata of size %zu", len);
 631
 632         /* Send command */
 633         ret = send_command(rsock, RELAYD_SEND_METADATA, NULL, len, 0);
 634         if (ret < 0) {
 635                 goto error;
 636         }
 637
 638         DBG2("Relayd metadata added successfully");
 639
 640         /*
 641          * After that call, the metadata data MUST be sent to the relayd so the
 642          * receive size on the other end matches the len of the metadata packet
 643          * header. This is why we don't wait for a reply here.
 644          */
 645
 646 error:
 647         return ret;
 648 }
 649
 650 /*
 651  * Connect to relay daemon with an allocated lttcomm_relayd_sock.
 652  */
 653 int relayd_connect(struct lttcomm_relayd_sock *rsock)
 654 {
 655         /* Code flow error. Safety net. */
 656         assert(rsock);
 657
 658         if (!rsock->sock.ops) {
 659                 /*
 660                  * Attempting a connect on a non-initialized socket.
 661                  */
 662                 return -ECONNRESET;
 663         }
 664
 665         DBG3("Relayd connect ...");
 666
 667         return rsock->sock.ops->connect(&rsock->sock);
 668 }
 669
 670 /*
 671  * Close relayd socket with an allocated lttcomm_relayd_sock.
 672  *
 673  * If no socket operations are found, simply return 0 meaning that everything
 674  * is fine. Without operations, the socket can not possibly be opened or used.
 675  * This is possible if the socket was allocated but not created. However, the
 676  * caller could simply use it to store a valid file descriptor for instance
 677  * passed over a Unix socket and call this to cleanup but still without a valid
 678  * ops pointer.
 679  *
 680  * Return the close returned value. On error, a negative value is usually
 681  * returned back from close(2).
 682  */
 683 int relayd_close(struct lttcomm_relayd_sock *rsock)
 684 {
 685         int ret;
 686
 687         /* Code flow error. Safety net. */
 688         assert(rsock);
 689
 690         /* An invalid fd is fine, return success. */
 691         if (rsock->sock.fd < 0) {
 692                 ret = 0;
 693                 goto end;
 694         }
 695
 696         /*
 697          * This ensure that we do not close the socket while the lttng-relayd
 698          * expects to be able to send a response that we skipped.
 699          * While we loose some time to receive everything, this keep the
 700          * protocol intact from the point of view of lttng-relayd.
 701          */
 702         if (rsock->bytes_to_ignore_on_recv != 0) {
 703                 ret = rsock->sock.ops->recvmsg(&rsock->sock, NULL, rsock->bytes_to_ignore_on_recv, MSG_TRUNC);
 704                 if (ret <= 0 || ret != rsock->bytes_to_ignore_on_recv) {
 705                         if (ret == 0) {
 706                                 /* Orderly shutdown. */
 707                                 DBG("Socket %d has performed an orderly shutdown", rsock->sock.fd);
 708                         } else {
 709                                 DBG("Receiving reply to skip failed on sock %d for size %zu with ret %d",
 710                                                 rsock->sock.fd, rsock->bytes_to_ignore_on_recv, ret);
 711                         }
 712                 }
 713                 DBG("Discarded %zu bytes on sock %d", rsock->bytes_to_ignore_on_recv, rsock->sock.fd);
 714                 rsock->bytes_to_ignore_on_recv = 0;
 715         }
 716
 717         DBG3("Relayd closing socket %d", rsock->sock.fd);
 718
 719         if (rsock->sock.ops) {
 720                 ret = rsock->sock.ops->close(&rsock->sock);
 721         } else {
 722                 /* Default call if no specific ops found. */
 723                 ret = close(rsock->sock.fd);
 724                 if (ret < 0) {
 725                         PERROR("relayd_close default close");
 726                 }
 727         }
 728         rsock->sock.fd = -1;
 729
 730 end:
 731         return ret;
 732 }
 733
 734 /*
 735  * Send data header structure to the relayd.
 736  */
 737 int relayd_send_data_hdr(struct lttcomm_relayd_sock *rsock,
 738                 struct lttcomm_relayd_data_hdr *hdr, size_t size)
 739 {
 740         int ret;
 741
 742         /* Code flow error. Safety net. */
 743         assert(rsock);
 744         assert(hdr);
 745
 746         if (rsock->sock.fd < 0) {
 747                 return -ECONNRESET;
 748         }
 749
 750         DBG3("Relayd sending data header of size %zu", size);
 751
 752         /* Again, safety net */
 753         if (size == 0) {
 754                 size = sizeof(struct lttcomm_relayd_data_hdr);
 755         }
 756
 757         /* Only send data header. */
 758         ret = rsock->sock.ops->sendmsg(&rsock->sock, hdr, size, 0);
 759         if (ret < 0) {
 760                 ret = -errno;
 761                 goto error;
 762         }
 763
 764         /*
 765          * The data MUST be sent right after that command for the receive on the
 766          * other end to match the size in the header.
 767          */
 768
 769 error:
 770         return ret;
 771 }
 772
 773 /*
 774  * Send close stream command to the relayd.
 775  */
 776 int relayd_send_close_stream(struct lttcomm_relayd_sock *rsock, uint64_t stream_id,
 777                 uint64_t last_net_seq_num)
 778 {
 779         int ret;
 780         struct lttcomm_relayd_close_stream msg;
 781         struct lttcomm_relayd_generic_reply reply;
 782
 783         /* Code flow error. Safety net. */
 784         assert(rsock);
 785
 786         DBG("Relayd closing stream id %" PRIu64, stream_id);
 787
 788         memset(&msg, 0, sizeof(msg));
 789         msg.stream_id = htobe64(stream_id);
 790         msg.last_net_seq_num = htobe64(last_net_seq_num);
 791
 792         /* Send command */
 793         ret = send_command(rsock, RELAYD_CLOSE_STREAM, (void *) &msg, sizeof(msg), 0);
 794         if (ret < 0) {
 795                 goto error;
 796         }
 797
 798         /*
 799          * Discard response since we do not really care for it and that TCP
 800          * guarantee in-order delivery. As for error handling, there is not much
 801          * to do at this point (closing).
 802          **/
 803         ret = recv_reply_ignore(rsock, sizeof(reply));
 804         if (ret < 0) {
 805                 goto error;
 806         }
 807
 808         DBG("Relayd close stream id %" PRIu64 " successfully", stream_id);
 809
 810 error:
 811         return ret;
 812 }
 813
 814 /*
 815  * Check for data availability for a given stream id.
 816  *
 817  * Return 0 if NOT pending, 1 if so and a negative value on error.
 818  */
 819 int relayd_data_pending(struct lttcomm_relayd_sock *rsock, uint64_t stream_id,
 820                 uint64_t last_net_seq_num)
 821 {
 822         int ret;
 823         struct lttcomm_relayd_data_pending msg;
 824         struct lttcomm_relayd_generic_reply reply;
 825
 826         /* Code flow error. Safety net. */
 827         assert(rsock);
 828
 829         DBG("Relayd data pending for stream id %" PRIu64, stream_id);
 830
 831         memset(&msg, 0, sizeof(msg));
 832         msg.stream_id = htobe64(stream_id);
 833         msg.last_net_seq_num = htobe64(last_net_seq_num);
 834
 835         /* Send command */
 836         ret = send_command(rsock, RELAYD_DATA_PENDING, (void *) &msg,
 837                         sizeof(msg), 0);
 838         if (ret < 0) {
 839                 goto error;
 840         }
 841
 842         /* Receive response */
 843         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 844         if (ret < 0) {
 845                 goto error;
 846         }
 847
 848         reply.ret_code = be32toh(reply.ret_code);
 849
 850         /* Return session id or negative ret code. */
 851         if (reply.ret_code >= LTTNG_OK) {
 852                 ERR("Relayd data pending replied error %d", reply.ret_code);
 853         }
 854
 855         /* At this point, the ret code is either 1 or 0 */
 856         ret = reply.ret_code;
 857
 858         DBG("Relayd data is %s pending for stream id %" PRIu64,
 859                         ret == 1 ? "" : "NOT", stream_id);
 860
 861 error:
 862         return ret;
 863 }
 864
 865 /*
 866  * Check on the relayd side for a quiescent state on the control socket.
 867  */
 868 int relayd_quiescent_control(struct lttcomm_relayd_sock *rsock,
 869                 uint64_t metadata_stream_id)
 870 {
 871         int ret;
 872         struct lttcomm_relayd_quiescent_control msg;
 873         struct lttcomm_relayd_generic_reply reply;
 874
 875         /* Code flow error. Safety net. */
 876         assert(rsock);
 877
 878         DBG("Relayd checking quiescent control state");
 879
 880         memset(&msg, 0, sizeof(msg));
 881         msg.stream_id = htobe64(metadata_stream_id);
 882
 883         /* Send command */
 884         ret = send_command(rsock, RELAYD_QUIESCENT_CONTROL, &msg, sizeof(msg), 0);
 885         if (ret < 0) {
 886                 goto error;
 887         }
 888
 889         /* Receive response */
 890         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 891         if (ret < 0) {
 892                 goto error;
 893         }
 894
 895         reply.ret_code = be32toh(reply.ret_code);
 896
 897         /* Return session id or negative ret code. */
 898         if (reply.ret_code != LTTNG_OK) {
 899                 ret = -1;
 900                 ERR("Relayd quiescent control replied error %d", reply.ret_code);
 901                 goto error;
 902         }
 903
 904         /* Control socket is quiescent */
 905         return 0;
 906
 907 error:
 908         return ret;
 909 }
 910
 911 /*
 912  * Begin a data pending command for a specific session id.
 913  */
 914 int relayd_begin_data_pending(struct lttcomm_relayd_sock *rsock, uint64_t id)
 915 {
 916         int ret;
 917         struct lttcomm_relayd_begin_data_pending msg;
 918         struct lttcomm_relayd_generic_reply reply;
 919
 920         /* Code flow error. Safety net. */
 921         assert(rsock);
 922
 923         DBG("Relayd begin data pending");
 924
 925         memset(&msg, 0, sizeof(msg));
 926         msg.session_id = htobe64(id);
 927
 928         /* Send command */
 929         ret = send_command(rsock, RELAYD_BEGIN_DATA_PENDING, &msg, sizeof(msg), 0);
 930         if (ret < 0) {
 931                 goto error;
 932         }
 933
 934         /* Receive response */
 935         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 936         if (ret < 0) {
 937                 goto error;
 938         }
 939
 940         reply.ret_code = be32toh(reply.ret_code);
 941
 942         /* Return session id or negative ret code. */
 943         if (reply.ret_code != LTTNG_OK) {
 944                 ret = -1;
 945                 ERR("Relayd begin data pending replied error %d", reply.ret_code);
 946                 goto error;
 947         }
 948
 949         return 0;
 950
 951 error:
 952         return ret;
 953 }
 954
 955 /*
 956  * End a data pending command for a specific session id.
 957  *
 958  * Return 0 on success and set is_data_inflight to 0 if no data is being
 959  * streamed or 1 if it is the case.
 960  */
 961 int relayd_end_data_pending(struct lttcomm_relayd_sock *rsock, uint64_t id,
 962                 unsigned int *is_data_inflight)
 963 {
 964         int ret, recv_ret;
 965         struct lttcomm_relayd_end_data_pending msg;
 966         struct lttcomm_relayd_generic_reply reply;
 967
 968         /* Code flow error. Safety net. */
 969         assert(rsock);
 970
 971         DBG("Relayd end data pending");
 972
 973         memset(&msg, 0, sizeof(msg));
 974         msg.session_id = htobe64(id);
 975
 976         /* Send command */
 977         ret = send_command(rsock, RELAYD_END_DATA_PENDING, &msg, sizeof(msg), 0);
 978         if (ret < 0) {
 979                 goto error;
 980         }
 981
 982         /* Receive response */
 983         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
 984         if (ret < 0) {
 985                 goto error;
 986         }
 987
 988         recv_ret = be32toh(reply.ret_code);
 989         if (recv_ret < 0) {
 990                 ret = recv_ret;
 991                 goto error;
 992         }
 993
 994         *is_data_inflight = recv_ret;
 995
 996         DBG("Relayd end data pending is data inflight: %d", recv_ret);
 997
 998         return 0;
 999
1000 error:
1001         return ret;
1002 }
1003
1004 /*
1005  * Send index to the relayd.
1006  */
1007 int relayd_send_index(struct lttcomm_relayd_sock *rsock,
1008                 struct ctf_packet_index *index, uint64_t relay_stream_id,
1009                 uint64_t net_seq_num)
1010 {
1011         int ret;
1012         struct lttcomm_relayd_index msg;
1013         struct lttcomm_relayd_generic_reply reply;
1014
1015         /* Code flow error. Safety net. */
1016         assert(rsock);
1017
1018         if (rsock->minor < 4) {
1019                 DBG("Not sending indexes before protocol 2.4");
1020                 ret = 0;
1021                 goto error;
1022         }
1023
1024         DBG("Relayd sending index for stream ID %" PRIu64, relay_stream_id);
1025
1026         memset(&msg, 0, sizeof(msg));
1027         msg.relay_stream_id = htobe64(relay_stream_id);
1028         msg.net_seq_num = htobe64(net_seq_num);
1029
1030         /* The index is already in big endian. */
1031         msg.packet_size = index->packet_size;
1032         msg.content_size = index->content_size;
1033         msg.timestamp_begin = index->timestamp_begin;
1034         msg.timestamp_end = index->timestamp_end;
1035         msg.events_discarded = index->events_discarded;
1036         msg.stream_id = index->stream_id;
1037
1038         if (rsock->minor >= 8) {
1039                 msg.stream_instance_id = index->stream_instance_id;
1040                 msg.packet_seq_num = index->packet_seq_num;
1041         }
1042
1043         /* Send command */
1044         ret = send_command(rsock, RELAYD_SEND_INDEX, &msg,
1045                 lttcomm_relayd_index_len(lttng_to_index_major(rsock->major,
1046                                                                 rsock->minor),
1047                                 lttng_to_index_minor(rsock->major, rsock->minor)),
1048                                 0);
1049         if (ret < 0) {
1050                 goto error;
1051         }
1052
1053         /*
1054          * Ignore the response. TCP guarantee in-order arrival and the overall
1055          * protocol do not rely on hard ordering between the control and data
1056          * socket for index.
1057          * Indexes are sent either at the end of the buffer consumption or
1058          * during the live timer.
1059          */
1060         ret = recv_reply_ignore(rsock, sizeof(reply));
1061         if (ret < 0) {
1062                 goto error;
1063         }
1064
1065 error:
1066         return ret;
1067 }
1068
1069 /*
1070  * Ask the relay to reset the metadata trace file (regeneration).
1071  */
1072 int relayd_reset_metadata(struct lttcomm_relayd_sock *rsock,
1073                 uint64_t stream_id, uint64_t version)
1074 {
1075         int ret;
1076         struct lttcomm_relayd_reset_metadata msg;
1077         struct lttcomm_relayd_generic_reply reply;
1078
1079         /* Code flow error. Safety net. */
1080         assert(rsock);
1081
1082         /* Should have been prevented by the sessiond. */
1083         if (rsock->minor < 8) {
1084                 ERR("Metadata regeneration unsupported before 2.8");
1085                 ret = -1;
1086                 goto error;
1087         }
1088
1089         DBG("Relayd reset metadata stream id %" PRIu64, stream_id);
1090
1091         memset(&msg, 0, sizeof(msg));
1092         msg.stream_id = htobe64(stream_id);
1093         msg.version = htobe64(version);
1094
1095         /* Send command */
1096         ret = send_command(rsock, RELAYD_RESET_METADATA, (void *) &msg, sizeof(msg), 0);
1097         if (ret < 0) {
1098                 goto error;
1099         }
1100
1101         /* Receive response */
1102         ret = recv_reply(rsock, (void *) &reply, sizeof(reply));
1103         if (ret < 0) {
1104                 goto error;
1105         }
1106
1107         reply.ret_code = be32toh(reply.ret_code);
1108
1109         /* Return session id or negative ret code. */
1110         if (reply.ret_code != LTTNG_OK) {
1111                 ret = -1;
1112                 ERR("Relayd reset metadata replied error %d", reply.ret_code);
1113         } else {
1114                 /* Success */
1115                 ret = 0;
1116         }
1117
1118         DBG("Relayd reset metadata stream id %" PRIu64 " successfully", stream_id);
1119
1120 error:
1121         return ret;
1122 }