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