kconsumerd/kconsumerd.c

   1 /*
   2  * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
   3  *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License
   7  * as published by the Free Software Foundation; either version 2
   8  * of the License, or (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  18  */
  19
  20 #define _GNU_SOURCE
  21 #include <fcntl.h>
  22 #include <getopt.h>
  23 #include <grp.h>
  24 #include <limits.h>
  25 #include <pthread.h>
  26 #include <signal.h>
  27 #include <stdio.h>
  28 #include <stdlib.h>
  29 #include <string.h>
  30 #include <sys/ipc.h>
  31 #include <sys/shm.h>
  32 #include <sys/socket.h>
  33 #include <sys/stat.h>
  34 #include <sys/types.h>
  35 #include <urcu/list.h>
  36 #include <poll.h>
  37 #include <unistd.h>
  38 #include <sys/mman.h>
  39
  40 #include "lttngerr.h"
  41 #include "libkernelctl.h"
  42 #include "liblttsessiondcomm.h"
  43 #include "kconsumerd.h"
  44
  45 /* Init the list of FDs */
  46 static struct ltt_kconsumerd_fd_list kconsumerd_fd_list = {
  47         .head = CDS_LIST_HEAD_INIT(kconsumerd_fd_list.head),
  48 };
  49
  50 /* Number of element for the list below. */
  51 static unsigned int fds_count;
  52
  53 /* If the local array of FDs needs update in the poll function */
  54 static unsigned int update_fd_array = 1;
  55
  56 /* lock the fd array and structures */
  57 static pthread_mutex_t kconsumerd_lock_fds;
  58
  59 /* the two threads (receive fd and poll) */
  60 static pthread_t threads[2];
  61
  62 /* communication with splice */
  63 static int thread_pipe[2];
  64
  65 /* pipe to wake the poll thread when necessary */
  66 static int poll_pipe[2];
  67
  68 /* socket to communicate errors with sessiond */
  69 static int error_socket = -1;
  70
  71 /* to count the number of time the user pressed ctrl+c */
  72 static int sigintcount = 0;
  73
  74 /* flag to inform the polling thread to quit when all fd hung up */
  75 static int quit = 0;
  76
  77 /* Argument variables */
  78 int opt_quiet;
  79 int opt_verbose;
  80 static int opt_daemon;
  81 static const char *progname;
  82 static char command_sock_path[PATH_MAX]; /* Global command socket path */
  83 static char error_sock_path[PATH_MAX]; /* Global error path */
  84
  85 /*
  86  * del_fd
  87  *
  88  * Remove a fd from the global list protected by a mutex
  89  */
  90 static void del_fd(struct ltt_kconsumerd_fd *lcf)
  91 {
  92         DBG("Removing %d", lcf->consumerd_fd);
  93         pthread_mutex_lock(&kconsumerd_lock_fds);
  94         cds_list_del(&lcf->list);
  95         if (fds_count > 0) {
  96                 fds_count--;
  97                 DBG("Removed ltt_kconsumerd_fd");
  98                 if (lcf != NULL) {
  99                         close(lcf->out_fd);
 100                         close(lcf->consumerd_fd);
 101                         free(lcf);
 102                         lcf = NULL;
 103                 }
 104         }
 105         pthread_mutex_unlock(&kconsumerd_lock_fds);
 106 }
 107
 108 /*
 109  *  cleanup
 110  *
 111  *  Cleanup the daemon's socket on exit
 112  */
 113 static void cleanup()
 114 {
 115         struct ltt_kconsumerd_fd *iter;
 116
 117         /* remove the socket file */
 118         unlink(command_sock_path);
 119
 120         /* unblock the threads */
 121         WARN("Terminating the threads before exiting");
 122         pthread_cancel(threads[0]);
 123         pthread_cancel(threads[1]);
 124
 125         /* close all outfd */
 126         cds_list_for_each_entry(iter, &kconsumerd_fd_list.head, list) {
 127                 del_fd(iter);
 128         }
 129 }
 130
 131 /*
 132  * send_error
 133  *
 134  * send return code to ltt-sessiond
 135  */
 136 static int send_error(enum lttcomm_return_code cmd)
 137 {
 138         if (error_socket > 0) {
 139                 return lttcomm_send_unix_sock(error_socket, &cmd,
 140                                 sizeof(enum lttcomm_sessiond_command));
 141         } else {
 142                 return 0;
 143         }
 144 }
 145
 146 /*
 147  * add_fd
 148  *
 149  * Add a fd to the global list protected by a mutex
 150  */
 151 static int add_fd(struct lttcomm_kconsumerd_msg *buf, int consumerd_fd)
 152 {
 153         struct ltt_kconsumerd_fd *tmp_fd;
 154         int ret;
 155
 156         tmp_fd = malloc(sizeof(struct ltt_kconsumerd_fd));
 157         tmp_fd->sessiond_fd = buf->fd;
 158         tmp_fd->consumerd_fd = consumerd_fd;
 159         tmp_fd->state = buf->state;
 160         tmp_fd->max_sb_size = buf->max_sb_size;
 161         strncpy(tmp_fd->path_name, buf->path_name, PATH_MAX);
 162
 163         /* Opening the tracefile in write mode */
 164         DBG("Opening %s for writing", tmp_fd->path_name);
 165         ret = open(tmp_fd->path_name,
 166                         O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO);
 167         if (ret < 0) {
 168                 ERR("Opening %s", tmp_fd->path_name);
 169                 perror("open");
 170                 goto end;
 171         }
 172         tmp_fd->out_fd = ret;
 173         tmp_fd->out_fd_offset = 0;
 174
 175         DBG("Adding %s (%d, %d, %d)", tmp_fd->path_name,
 176                         tmp_fd->sessiond_fd, tmp_fd->consumerd_fd, tmp_fd->out_fd);
 177
 178         pthread_mutex_lock(&kconsumerd_lock_fds);
 179         cds_list_add(&tmp_fd->list, &kconsumerd_fd_list.head);
 180         fds_count++;
 181         pthread_mutex_unlock(&kconsumerd_lock_fds);
 182
 183 end:
 184         return ret;
 185 }
 186
 187
 188 /*
 189  *  sighandler
 190  *
 191  *  Signal handler for the daemon
 192  */
 193 static void sighandler(int sig)
 194 {
 195         if (sig == SIGINT && sigintcount++ == 0) {
 196                 DBG("ignoring first SIGINT");
 197                 return;
 198         }
 199
 200         cleanup();
 201
 202         return;
 203 }
 204
 205 /*
 206  *  set_signal_handler
 207  *
 208  *  Setup signal handler for :
 209  *      SIGINT, SIGTERM, SIGPIPE
 210  */
 211 static int set_signal_handler(void)
 212 {
 213         int ret = 0;
 214         struct sigaction sa;
 215         sigset_t sigset;
 216
 217         if ((ret = sigemptyset(&sigset)) < 0) {
 218                 perror("sigemptyset");
 219                 return ret;
 220         }
 221
 222         sa.sa_handler = sighandler;
 223         sa.sa_mask = sigset;
 224         sa.sa_flags = 0;
 225         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
 226                 perror("sigaction");
 227                 return ret;
 228         }
 229
 230         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
 231                 perror("sigaction");
 232                 return ret;
 233         }
 234
 235         if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
 236                 perror("sigaction");
 237                 return ret;
 238         }
 239
 240         return ret;
 241 }
 242
 243 /*
 244  * on_read_subbuffer_mmap
 245  *
 246  * mmap the ring buffer, read it and write the data to the tracefile.
 247  * Returns the number of bytes written
 248  */
 249 static int on_read_subbuffer_mmap(struct ltt_kconsumerd_fd *kconsumerd_fd,
 250                 unsigned long len)
 251 {
 252         unsigned long mmap_len;
 253         unsigned long mmap_offset;
 254         unsigned long padded_len;
 255         unsigned long padding_len;
 256         char *mmap_base;
 257         char *padding = NULL;
 258         long ret = 0;
 259         off_t orig_offset = kconsumerd_fd->out_fd_offset;
 260         int fd = kconsumerd_fd->consumerd_fd;
 261         int outfd = kconsumerd_fd->out_fd;
 262
 263         /* get the padded subbuffer size to know the padding required */
 264         ret = kernctl_get_padded_subbuf_size(fd, &padded_len);
 265         if (ret != 0) {
 266                 ret = errno;
 267                 perror("kernctl_get_padded_subbuf_size");
 268                 goto end;
 269         }
 270         padding_len = padded_len - len;
 271         padding = malloc(padding_len * sizeof(char));
 272         memset(padding, '\0', padding_len);
 273
 274         /* get the len of the mmap region */
 275         ret = kernctl_get_mmap_len(fd, &mmap_len);
 276         if (ret != 0) {
 277                 ret = errno;
 278                 perror("kernctl_get_mmap_len");
 279                 goto end;
 280         }
 281
 282         /* get the offset inside the fd to mmap */
 283         ret = kernctl_get_mmap_read_offset(fd, &mmap_offset);
 284         if (ret != 0) {
 285                 ret = errno;
 286                 perror("kernctl_get_mmap_read_offset");
 287                 goto end;
 288         }
 289
 290         mmap_base = mmap(NULL, mmap_len, PROT_READ, MAP_PRIVATE, fd, mmap_offset);
 291         if (mmap_base == MAP_FAILED) {
 292                 perror("Error mmaping");
 293                 ret = -1;
 294                 goto end;
 295         }
 296
 297         while (len > 0) {
 298                 ret = write(outfd, mmap_base, len);
 299                 if (ret >= len) {
 300                         len = 0;
 301                 } else if (ret < 0) {
 302                         ret = errno;
 303                         perror("Error in file write");
 304                         goto end;
 305                 }
 306                 /* This won't block, but will start writeout asynchronously */
 307                 sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret,
 308                                 SYNC_FILE_RANGE_WRITE);
 309                 kconsumerd_fd->out_fd_offset += ret;
 310         }
 311
 312         /* once all the data is written, write the padding to disk */
 313         ret = write(outfd, padding, padding_len);
 314         if (ret < 0) {
 315                 ret = errno;
 316                 perror("Error writing padding to file");
 317                 goto end;
 318         }
 319
 320         /*
 321          * This does a blocking write-and-wait on any page that belongs to the
 322          * subbuffer prior to the one we just wrote.
 323          * Don't care about error values, as these are just hints and ways to
 324          * limit the amount of page cache used.
 325          */
 326         if (orig_offset >= kconsumerd_fd->max_sb_size) {
 327                 sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size,
 328                                 kconsumerd_fd->max_sb_size,
 329                                 SYNC_FILE_RANGE_WAIT_BEFORE
 330                                 | SYNC_FILE_RANGE_WRITE
 331                                 | SYNC_FILE_RANGE_WAIT_AFTER);
 332                 /*
 333                  * Give hints to the kernel about how we access the file:
 334                  * POSIX_FADV_DONTNEED : we won't re-access data in a near
 335                  * future after we write it.
 336                  * We need to call fadvise again after the file grows because
 337                  * the kernel does not seem to apply fadvise to non-existing
 338                  * parts of the file.
 339                  * Call fadvise _after_ having waited for the page writeback to
 340                  * complete because the dirty page writeback semantic is not
 341                  * well defined. So it can be expected to lead to lower
 342                  * throughput in streaming.
 343                  */
 344                 posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size,
 345                                 kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED);
 346         }
 347         goto end;
 348
 349 end:
 350         if (padding != NULL) {
 351                 free(padding);
 352         }
 353         return ret;
 354 }
 355
 356 /*
 357  * on_read_subbuffer
 358  *
 359  * Splice the data from the ring buffer to the tracefile.
 360  * Returns the number of bytes spliced
 361  */
 362 static int on_read_subbuffer(struct ltt_kconsumerd_fd *kconsumerd_fd,
 363                 unsigned long len)
 364 {
 365         long ret = 0;
 366         loff_t offset = 0;
 367         off_t orig_offset = kconsumerd_fd->out_fd_offset;
 368         int fd = kconsumerd_fd->consumerd_fd;
 369         int outfd = kconsumerd_fd->out_fd;
 370
 371         while (len > 0) {
 372                 DBG("splice chan to pipe offset %lu (fd : %d)",
 373                                 (unsigned long)offset, fd);
 374                 ret = splice(fd, &offset, thread_pipe[1], NULL, len,
 375                                 SPLICE_F_MOVE | SPLICE_F_MORE);
 376                 DBG("splice chan to pipe ret %ld", ret);
 377                 if (ret < 0) {
 378                         ret = errno;
 379                         perror("Error in relay splice");
 380                         goto splice_error;
 381                 }
 382
 383                 ret = splice(thread_pipe[0], NULL, outfd, NULL, ret,
 384                                 SPLICE_F_MOVE | SPLICE_F_MORE);
 385                 DBG("splice pipe to file %ld", ret);
 386                 if (ret < 0) {
 387                         ret = errno;
 388                         perror("Error in file splice");
 389                         goto splice_error;
 390                 }
 391                 if (ret >= len) {
 392                         len = 0;
 393                 }
 394                 /* This won't block, but will start writeout asynchronously */
 395                 sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret,
 396                                 SYNC_FILE_RANGE_WRITE);
 397                 kconsumerd_fd->out_fd_offset += ret;
 398         }
 399
 400         /*
 401          * This does a blocking write-and-wait on any page that belongs to the
 402          * subbuffer prior to the one we just wrote.
 403          * Don't care about error values, as these are just hints and ways to
 404          * limit the amount of page cache used.
 405          */
 406         if (orig_offset >= kconsumerd_fd->max_sb_size) {
 407                 sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size,
 408                                 kconsumerd_fd->max_sb_size,
 409                                 SYNC_FILE_RANGE_WAIT_BEFORE
 410                                 | SYNC_FILE_RANGE_WRITE
 411                                 | SYNC_FILE_RANGE_WAIT_AFTER);
 412                 /*
 413                  * Give hints to the kernel about how we access the file:
 414                  * POSIX_FADV_DONTNEED : we won't re-access data in a near
 415                  * future after we write it.
 416                  * We need to call fadvise again after the file grows because
 417                  * the kernel does not seem to apply fadvise to non-existing
 418                  * parts of the file.
 419                  * Call fadvise _after_ having waited for the page writeback to
 420                  * complete because the dirty page writeback semantic is not
 421                  * well defined. So it can be expected to lead to lower
 422                  * throughput in streaming.
 423                  */
 424                 posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size,
 425                                 kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED);
 426         }
 427         goto end;
 428
 429 splice_error:
 430         /* send the appropriate error description to sessiond */
 431         switch(ret) {
 432         case EBADF:
 433                 send_error(KCONSUMERD_SPLICE_EBADF);
 434                 break;
 435         case EINVAL:
 436                 send_error(KCONSUMERD_SPLICE_EINVAL);
 437                 break;
 438         case ENOMEM:
 439                 send_error(KCONSUMERD_SPLICE_ENOMEM);
 440                 break;
 441         case ESPIPE:
 442                 send_error(KCONSUMERD_SPLICE_ESPIPE);
 443                 break;
 444         }
 445
 446 end:
 447         return ret;
 448 }
 449
 450 /*
 451  * read_subbuffer
 452  *
 453  * Consume data on a file descriptor and write it on a trace file
 454  */
 455 static int read_subbuffer(struct ltt_kconsumerd_fd *kconsumerd_fd)
 456 {
 457         unsigned long len;
 458         int err;
 459         long ret = 0;
 460         int infd = kconsumerd_fd->consumerd_fd;
 461
 462         DBG("In read_subbuffer (infd : %d)", infd);
 463         /* Get the next subbuffer */
 464         err = kernctl_get_next_subbuf(infd);
 465         if (err != 0) {
 466                 ret = errno;
 467                 perror("Reserving sub buffer failed (everything is normal, "
 468                                 "it is due to concurrency)");
 469                 goto end;
 470         }
 471
 472         if (DEFAULT_CHANNEL_OUTPUT == LTTNG_KERNEL_SPLICE) {
 473                 /* read the whole subbuffer */
 474                 err = kernctl_get_padded_subbuf_size(infd, &len);
 475                 if (err != 0) {
 476                         ret = errno;
 477                         perror("Getting sub-buffer len failed.");
 478                         goto end;
 479                 }
 480
 481                 /* splice the subbuffer to the tracefile */
 482                 ret = on_read_subbuffer(kconsumerd_fd, len);
 483                 if (ret < 0) {
 484                         /*
 485                          * display the error but continue processing to try
 486                          * to release the subbuffer
 487                          */
 488                         ERR("Error splicing to tracefile");
 489                 }
 490         } else if (DEFAULT_CHANNEL_OUTPUT == LTTNG_KERNEL_MMAP) {
 491                 /* read the used subbuffer size */
 492                 err = kernctl_get_subbuf_size(infd, &len);
 493                 if (err != 0) {
 494                         ret = errno;
 495                         perror("Getting sub-buffer len failed.");
 496                         goto end;
 497                 }
 498
 499                 /* write the subbuffer to the tracefile */
 500                 ret = on_read_subbuffer_mmap(kconsumerd_fd, len);
 501                 if (ret < 0) {
 502                         /*
 503                          * display the error but continue processing to try
 504                          * to release the subbuffer
 505                          */
 506                         ERR("Error writing to tracefile");
 507                 }
 508         } else {
 509                 ERR("Unknown output method");
 510                 ret = -1;
 511                 goto end;
 512         }
 513
 514         err = kernctl_put_next_subbuf(infd);
 515         if (err != 0) {
 516                 ret = errno;
 517                 if (errno == EFAULT) {
 518                         perror("Error in unreserving sub buffer\n");
 519                 } else if (errno == EIO) {
 520                         /* Should never happen with newer LTTng versions */
 521                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 522                 }
 523                 goto end;
 524         }
 525
 526 end:
 527         return ret;
 528 }
 529
 530 /*
 531  * change_fd_state
 532  *
 533  * Update a fd according to what we just received
 534  */
 535 static void change_fd_state(int sessiond_fd,
 536                 enum kconsumerd_fd_state state)
 537 {
 538         struct ltt_kconsumerd_fd *iter;
 539         cds_list_for_each_entry(iter, &kconsumerd_fd_list.head, list) {
 540                 if (iter->sessiond_fd == sessiond_fd) {
 541                         iter->state = state;
 542                         break;
 543                 }
 544         }
 545 }
 546
 547 /*
 548  * consumerd_recv_fd
 549  *
 550  * Receives an array of file descriptors and the associated
 551  * structures describing each fd (path name).
 552  * Returns the size of received data
 553  */
 554 static int consumerd_recv_fd(int sfd, int size,
 555                 enum kconsumerd_command cmd_type)
 556 {
 557         struct msghdr msg;
 558         struct iovec iov[1];
 559         int ret = 0, i, tmp2;
 560         struct cmsghdr *cmsg;
 561         int nb_fd;
 562         char recv_fd[CMSG_SPACE(sizeof(int))];
 563         struct lttcomm_kconsumerd_msg lkm;
 564
 565         /* the number of fds we are about to receive */
 566         nb_fd = size / sizeof(struct lttcomm_kconsumerd_msg);
 567
 568         for (i = 0; i < nb_fd; i++) {
 569                 memset(&msg, 0, sizeof(msg));
 570
 571                 /* Prepare to receive the structures */
 572                 iov[0].iov_base = &lkm;
 573                 iov[0].iov_len = sizeof(lkm);
 574                 msg.msg_iov = iov;
 575                 msg.msg_iovlen = 1;
 576
 577                 msg.msg_control = recv_fd;
 578                 msg.msg_controllen = sizeof(recv_fd);
 579
 580                 DBG("Waiting to receive fd");
 581                 if ((ret = recvmsg(sfd, &msg, 0)) < 0) {
 582                         perror("recvmsg");
 583                         continue;
 584                 }
 585
 586                 if (ret != (size / nb_fd)) {
 587                         ERR("Received only %d, expected %d", ret, size);
 588                         send_error(KCONSUMERD_ERROR_RECV_FD);
 589                         goto end;
 590                 }
 591
 592                 cmsg = CMSG_FIRSTHDR(&msg);
 593                 if (!cmsg) {
 594                         ERR("Invalid control message header");
 595                         ret = -1;
 596                         send_error(KCONSUMERD_ERROR_RECV_FD);
 597                         goto end;
 598                 }
 599
 600                 /* if we received fds */
 601                 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
 602                         switch (cmd_type) {
 603                         case ADD_STREAM:
 604                                 DBG("add_fd %s (%d)", lkm.path_name, (CMSG_DATA(cmsg)[0]));
 605                                 ret = add_fd(&lkm, (CMSG_DATA(cmsg)[0]));
 606                                 if (ret < 0) {
 607                                         send_error(KCONSUMERD_OUTFD_ERROR);
 608                                         goto end;
 609                                 }
 610                                 break;
 611                         case UPDATE_STREAM:
 612                                 change_fd_state(lkm.fd, lkm.state);
 613                                 break;
 614                         default:
 615                                 break;
 616                         }
 617                         /* flag to tell the polling thread to update its fd array */
 618                         update_fd_array = 1;
 619                         /* signal the poll thread */
 620                         tmp2 = write(poll_pipe[1], "4", 1);
 621                 } else {
 622                         ERR("Didn't received any fd");
 623                         send_error(KCONSUMERD_ERROR_RECV_FD);
 624                         ret = -1;
 625                         goto end;
 626                 }
 627         }
 628
 629 end:
 630         DBG("consumerd_recv_fd thread exiting");
 631         return ret;
 632 }
 633
 634 /*
 635  *  thread_receive_fds
 636  *
 637  *  This thread listens on the consumerd socket and
 638  *  receives the file descriptors from ltt-sessiond
 639  */
 640 static void *thread_receive_fds(void *data)
 641 {
 642         int sock, client_socket, ret;
 643         struct lttcomm_kconsumerd_header tmp;
 644
 645         DBG("Creating command socket %s", command_sock_path);
 646         unlink(command_sock_path);
 647         client_socket = lttcomm_create_unix_sock(command_sock_path);
 648         if (client_socket < 0) {
 649                 ERR("Cannot create command socket");
 650                 goto end;
 651         }
 652
 653         ret = lttcomm_listen_unix_sock(client_socket);
 654         if (ret < 0) {
 655                 goto end;
 656         }
 657
 658         DBG("Sending ready command to ltt-sessiond");
 659         ret = send_error(KCONSUMERD_COMMAND_SOCK_READY);
 660         if (ret < 0) {
 661                 ERR("Error sending ready command to ltt-sessiond");
 662                 goto end;
 663         }
 664
 665         /* Blocking call, waiting for transmission */
 666         sock = lttcomm_accept_unix_sock(client_socket);
 667         if (sock <= 0) {
 668                 WARN("On accept");
 669                 goto end;
 670         }
 671         while (1) {
 672                 /* We first get the number of fd we are about to receive */
 673                 ret = lttcomm_recv_unix_sock(sock, &tmp,
 674                                 sizeof(struct lttcomm_kconsumerd_header));
 675                 if (ret <= 0) {
 676                         ERR("Communication interrupted on command socket");
 677                         goto end;
 678                 }
 679                 if (tmp.cmd_type == STOP) {
 680                         DBG("Received STOP command");
 681                         goto end;
 682                 }
 683                 /* we received a command to add or update fds */
 684                 ret = consumerd_recv_fd(sock, tmp.payload_size, tmp.cmd_type);
 685                 if (ret <= 0) {
 686                         ERR("Receiving the FD, exiting");
 687                         goto end;
 688                 }
 689         }
 690
 691 end:
 692         DBG("thread_receive_fds exiting");
 693         quit = 1;
 694         ret = write(poll_pipe[1], "4", 1);
 695         if (ret < 0) {
 696                 perror("poll pipe write");
 697         }
 698         return NULL;
 699 }
 700
 701 /*
 702  * update_poll_array
 703  *
 704  * Allocate the pollfd structure and the local view of the out fds
 705  * to avoid doing a lookup in the linked list and concurrency issues
 706  * when writing is needed.
 707  * Returns the number of fds in the structures
 708  */
 709 static int update_poll_array(struct pollfd **pollfd,
 710                 struct ltt_kconsumerd_fd **local_kconsumerd_fd)
 711 {
 712         struct ltt_kconsumerd_fd *iter;
 713         int i = 0;
 714
 715
 716         DBG("Updating poll fd array");
 717         pthread_mutex_lock(&kconsumerd_lock_fds);
 718
 719         cds_list_for_each_entry(iter, &kconsumerd_fd_list.head, list) {
 720                 DBG("Inside for each");
 721                 if (iter->state == ACTIVE_FD) {
 722                         DBG("Active FD %d", iter->consumerd_fd);
 723                         (*pollfd)[i].fd = iter->consumerd_fd;
 724                         (*pollfd)[i].events = POLLIN | POLLPRI;
 725                         local_kconsumerd_fd[i] = iter;
 726                         i++;
 727                 }
 728         }
 729         /*
 730          * insert the poll_pipe at the end of the array and don't increment i
 731          * so nb_fd is the number of real FD
 732          */
 733         (*pollfd)[i].fd = poll_pipe[0];
 734         (*pollfd)[i].events = POLLIN;
 735
 736         update_fd_array = 0;
 737         pthread_mutex_unlock(&kconsumerd_lock_fds);
 738         return i;
 739
 740 }
 741
 742 /*
 743  *  thread_poll_fds
 744  *
 745  *  This thread polls the fds in the ltt_fd_list to consume the data
 746  *  and write it to tracefile if necessary.
 747  */
 748 static void *thread_poll_fds(void *data)
 749 {
 750         int num_rdy, num_hup, high_prio, ret, i;
 751         struct pollfd *pollfd = NULL;
 752         /* local view of the fds */
 753         struct ltt_kconsumerd_fd **local_kconsumerd_fd = NULL;
 754         /* local view of fds_count */
 755         int nb_fd = 0;
 756         char tmp;
 757         int tmp2;
 758
 759         ret = pipe(thread_pipe);
 760         if (ret < 0) {
 761                 perror("Error creating pipe");
 762                 goto end;
 763         }
 764
 765         local_kconsumerd_fd = malloc(sizeof(struct ltt_kconsumerd_fd));
 766
 767         while (1) {
 768                 high_prio = 0;
 769                 num_hup = 0;
 770
 771                 /*
 772                  * the ltt_fd_list has been updated, we need to update our
 773                  * local array as well
 774                  */
 775                 if (update_fd_array == 1) {
 776                         if (pollfd != NULL) {
 777                                 free(pollfd);
 778                                 pollfd = NULL;
 779                         }
 780                         if (local_kconsumerd_fd != NULL) {
 781                                 free(local_kconsumerd_fd);
 782                                 local_kconsumerd_fd = NULL;
 783                         }
 784                         /* allocate for all fds + 1 for the poll_pipe */
 785                         pollfd = malloc((fds_count + 1) * sizeof(struct pollfd));
 786                         if (pollfd == NULL) {
 787                                 perror("pollfd malloc");
 788                                 goto end;
 789                         }
 790                         /* allocate for all fds + 1 for the poll_pipe */
 791                         local_kconsumerd_fd = malloc((fds_count + 1) * sizeof(struct ltt_kconsumerd_fd));
 792                         if (local_kconsumerd_fd == NULL) {
 793                                 perror("local_kconsumerd_fd malloc");
 794                                 goto end;
 795                         }
 796
 797                         ret = update_poll_array(&pollfd, local_kconsumerd_fd);
 798                         if (ret < 0) {
 799                                 ERR("Error in allocating pollfd or local_outfds");
 800                                 send_error(KCONSUMERD_POLL_ERROR);
 801                                 goto end;
 802                         }
 803                         nb_fd = ret;
 804                 }
 805
 806                 /* poll on the array of fds */
 807                 DBG("polling on %d fd", nb_fd + 1);
 808                 num_rdy = poll(pollfd, nb_fd + 1, -1);
 809                 DBG("poll num_rdy : %d", num_rdy);
 810                 if (num_rdy == -1) {
 811                         perror("Poll error");
 812                         send_error(KCONSUMERD_POLL_ERROR);
 813                         goto end;
 814                 }
 815
 816                 /* No FDs and quit, cleanup the thread */
 817                 if (nb_fd == 0 && quit == 1) {
 818                         goto end;
 819                 }
 820
 821                 /*
 822                  * if only the poll_pipe triggered poll to return just return to the
 823                  * beginning of the loop to update the array
 824                  */
 825                 if (num_rdy == 1 && pollfd[nb_fd].revents == POLLIN) {
 826                         DBG("poll_pipe wake up");
 827                         tmp2 = read(poll_pipe[0], &tmp, 1);
 828                         continue;
 829                 }
 830
 831                 /* Take care of high priority channels first. */
 832                 for (i = 0; i < nb_fd; i++) {
 833                         switch(pollfd[i].revents) {
 834                         case POLLERR:
 835                                 ERR("Error returned in polling fd %d.", pollfd[i].fd);
 836                                 del_fd(local_kconsumerd_fd[i]);
 837                                 update_fd_array = 1;
 838                                 num_hup++;
 839                                 break;
 840                         case POLLHUP:
 841                                 ERR("Polling fd %d tells it has hung up.", pollfd[i].fd);
 842                                 del_fd(local_kconsumerd_fd[i]);
 843                                 update_fd_array = 1;
 844                                 num_hup++;
 845                                 break;
 846                         case POLLNVAL:
 847                                 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
 848                                 del_fd(local_kconsumerd_fd[i]);
 849                                 update_fd_array = 1;
 850                                 num_hup++;
 851                                 break;
 852                         case POLLPRI:
 853                                 DBG("Urgent read on fd %d", pollfd[i].fd);
 854                                 high_prio = 1;
 855                                 ret = read_subbuffer(local_kconsumerd_fd[i]);
 856                                 /* it's ok to have an unavailable sub-buffer (FIXME : is it ?) */
 857                                 if (ret == EAGAIN) {
 858                                         ret = 0;
 859                                 }
 860                                 break;
 861                         }
 862                 }
 863
 864                 /* If every buffer FD has hung up, we end the read loop here */
 865                 if (nb_fd > 0 && num_hup == nb_fd) {
 866                         DBG("every buffer FD has hung up\n");
 867                         if (quit == 1) {
 868                                 goto end;
 869                         }
 870                         continue;
 871                 }
 872
 873                 /* Take care of low priority channels. */
 874                 if (high_prio == 0) {
 875                         for (i = 0; i < nb_fd; i++) {
 876                                 if (pollfd[i].revents == POLLIN) {
 877                                         DBG("Normal read on fd %d", pollfd[i].fd);
 878                                         ret = read_subbuffer(local_kconsumerd_fd[i]);
 879                                         /* it's ok to have an unavailable subbuffer (FIXME : is it ?) */
 880                                         if (ret == EAGAIN) {
 881                                                 ret = 0;
 882                                         }
 883                                 }
 884                         }
 885                 }
 886         }
 887 end:
 888         DBG("polling thread exiting");
 889         if (pollfd != NULL) {
 890                 free(pollfd);
 891                 pollfd = NULL;
 892         }
 893         if (local_kconsumerd_fd != NULL) {
 894                 free(local_kconsumerd_fd);
 895                 local_kconsumerd_fd = NULL;
 896         }
 897         cleanup();
 898         return NULL;
 899 }
 900
 901 /*
 902  * usage function on stderr
 903  */
 904 static void usage(void)
 905 {
 906         fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
 907         fprintf(stderr, "  -h, --help                         "
 908                         "Display this usage.\n");
 909         fprintf(stderr, "  -c, --kconsumerd-cmd-sock PATH     "
 910                         "Specify path for the command socket\n");
 911         fprintf(stderr, "  -e, --kconsumerd-err-sock PATH     "
 912                         "Specify path for the error socket\n");
 913         fprintf(stderr, "  -d, --daemonize                    "
 914                         "Start as a daemon.\n");
 915         fprintf(stderr, "  -q, --quiet                        "
 916                         "No output at all.\n");
 917         fprintf(stderr, "  -v, --verbose                      "
 918                         "Verbose mode. Activate DBG() macro.\n");
 919         fprintf(stderr, "  -V, --version                      "
 920                         "Show version number.\n");
 921 }
 922
 923 /*
 924  * daemon argument parsing
 925  */
 926 static void parse_args(int argc, char **argv)
 927 {
 928         int c;
 929
 930         static struct option long_options[] = {
 931                 { "kconsumerd-cmd-sock", 1, 0, 'c' },
 932                 { "kconsumerd-err-sock", 1, 0, 'e' },
 933                 { "daemonize", 0, 0, 'd' },
 934                 { "help", 0, 0, 'h' },
 935                 { "quiet", 0, 0, 'q' },
 936                 { "verbose", 0, 0, 'v' },
 937                 { "version", 0, 0, 'V' },
 938                 { NULL, 0, 0, 0 }
 939         };
 940
 941         while (1) {
 942                 int option_index = 0;
 943                 c = getopt_long(argc, argv, "dhqvV" "c:e:", long_options, &option_index);
 944                 if (c == -1) {
 945                         break;
 946                 }
 947
 948                 switch (c) {
 949                 case 0:
 950                         fprintf(stderr, "option %s", long_options[option_index].name);
 951                         if (optarg) {
 952                                 fprintf(stderr, " with arg %s\n", optarg);
 953                         }
 954                         break;
 955                 case 'c':
 956                         snprintf(command_sock_path, PATH_MAX, "%s", optarg);
 957                         break;
 958                 case 'e':
 959                         snprintf(error_sock_path, PATH_MAX, "%s", optarg);
 960                         break;
 961                 case 'd':
 962                         opt_daemon = 1;
 963                         break;
 964                 case 'h':
 965                         usage();
 966                         exit(EXIT_FAILURE);
 967                 case 'q':
 968                         opt_quiet = 1;
 969                         break;
 970                 case 'v':
 971                         opt_verbose = 1;
 972                         break;
 973                 case 'V':
 974                         fprintf(stdout, "%s\n", VERSION);
 975                         exit(EXIT_SUCCESS);
 976                 default:
 977                         usage();
 978                         exit(EXIT_FAILURE);
 979                 }
 980         }
 981 }
 982
 983
 984 /*
 985  * main
 986  */
 987 int main(int argc, char **argv)
 988 {
 989         int i;
 990         int ret = 0;
 991         void *status;
 992
 993         /* Parse arguments */
 994         progname = argv[0];
 995         parse_args(argc, argv);
 996
 997         /* Daemonize */
 998         if (opt_daemon) {
 999                 ret = daemon(0, 0);
1000                 if (ret < 0) {
1001                         perror("daemon");
1002                         goto error;
1003                 }
1004         }
1005
1006         if (strlen(command_sock_path) == 0) {
1007                 snprintf(command_sock_path, PATH_MAX,
1008                                 KCONSUMERD_CMD_SOCK_PATH);
1009         }
1010         if (strlen(error_sock_path) == 0) {
1011                 snprintf(error_sock_path, PATH_MAX,
1012                                 KCONSUMERD_ERR_SOCK_PATH);
1013         }
1014
1015         if (set_signal_handler() < 0) {
1016                 goto error;
1017         }
1018
1019         /* create the pipe to wake to polling thread when needed */
1020         ret = pipe(poll_pipe);
1021         if (ret < 0) {
1022                 perror("Error creating poll pipe");
1023                 goto end;
1024         }
1025
1026         /* Connect to the socket created by ltt-sessiond to report errors */
1027         DBG("Connecting to error socket %s", error_sock_path);
1028         error_socket = lttcomm_connect_unix_sock(error_sock_path);
1029         /* not a fatal error, but all communication with ltt-sessiond will fail */
1030         if (error_socket < 0) {
1031                 WARN("Cannot connect to error socket, is ltt-sessiond started ?");
1032         }
1033
1034         /* Create the thread to manage the receive of fd */
1035         ret = pthread_create(&threads[0], NULL, thread_receive_fds, (void *) NULL);
1036         if (ret != 0) {
1037                 perror("pthread_create");
1038                 goto error;
1039         }
1040
1041         /* Create thread to manage the polling/writing of traces */
1042         ret = pthread_create(&threads[1], NULL, thread_poll_fds, (void *) NULL);
1043         if (ret != 0) {
1044                 perror("pthread_create");
1045                 goto error;
1046         }
1047
1048         for (i = 0; i < 2; i++) {
1049                 ret = pthread_join(threads[i], &status);
1050                 if (ret != 0) {
1051                         perror("pthread_join");
1052                         goto error;
1053                 }
1054         }
1055         ret = EXIT_SUCCESS;
1056         send_error(KCONSUMERD_EXIT_SUCCESS);
1057         goto end;
1058
1059 error:
1060         ret = EXIT_FAILURE;
1061         send_error(KCONSUMERD_EXIT_FAILURE);
1062
1063 end:
1064         cleanup();
1065
1066         return ret;
1067 }