src/common/unix.c

   1 /*
   2  * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
   3  * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * SPDX-License-Identifier: GPL-2.0-only
   6  *
   7  */
   8
   9 #define _LGPL_SOURCE
  10 #include <assert.h>
  11 #include <limits.h>
  12 #include <stdio.h>
  13 #include <stdlib.h>
  14 #include <string.h>
  15 #include <sys/stat.h>
  16 #include <sys/types.h>
  17 #include <unistd.h>
  18 #include <errno.h>
  19
  20 #include <common/common.h>
  21 #include <common/sessiond-comm/sessiond-comm.h>
  22 #include <common/fd-handle.h>
  23
  24 #include "unix.h"
  25
  26 /*
  27  * Connect to unix socket using the path name.
  28  */
  29 LTTNG_HIDDEN
  30 int lttcomm_connect_unix_sock(const char *pathname)
  31 {
  32         struct sockaddr_un s_un;
  33         int fd, ret, closeret;
  34
  35         if (strlen(pathname) >= sizeof(s_un.sun_path)) {
  36                 ERR("unix socket address (\"%s\") is longer than the platform's limit (%zu > %zu).",
  37                                 pathname, strlen(pathname) + 1,
  38                                 sizeof(s_un.sun_path));
  39                 ret = -ENAMETOOLONG;
  40                 goto error;
  41         }
  42
  43         fd = socket(PF_UNIX, SOCK_STREAM, 0);
  44         if (fd < 0) {
  45                 PERROR("socket");
  46                 ret = fd;
  47                 goto error;
  48         }
  49
  50         memset(&s_un, 0, sizeof(s_un));
  51         s_un.sun_family = AF_UNIX;
  52         strncpy(s_un.sun_path, pathname, sizeof(s_un.sun_path));
  53         s_un.sun_path[sizeof(s_un.sun_path) - 1] = '\0';
  54
  55         ret = connect(fd, (struct sockaddr *) &s_un, sizeof(s_un));
  56         if (ret < 0) {
  57                 /*
  58                  * Don't print message on connect error, because connect is used in
  59                  * normal execution to detect if sessiond is alive.
  60                  */
  61                 goto error_connect;
  62         }
  63
  64         return fd;
  65
  66 error_connect:
  67         closeret = close(fd);
  68         if (closeret) {
  69                 PERROR("close");
  70         }
  71 error:
  72         return ret;
  73 }
  74
  75 /*
  76  * Do an accept(2) on the sock and return the new file descriptor. The socket
  77  * MUST be bind(2) before.
  78  */
  79 LTTNG_HIDDEN
  80 int lttcomm_accept_unix_sock(int sock)
  81 {
  82         int new_fd;
  83         struct sockaddr_un s_un;
  84         socklen_t len = sizeof(s_un);
  85
  86         /* Blocking call */
  87         new_fd = accept(sock, (struct sockaddr *) &s_un, &len);
  88         if (new_fd < 0) {
  89                 PERROR("accept");
  90         }
  91
  92         return new_fd;
  93 }
  94
  95 LTTNG_HIDDEN
  96 int lttcomm_create_anon_unix_socketpair(int *fds)
  97 {
  98         if (socketpair(PF_UNIX, SOCK_STREAM, 0, fds) < 0) {
  99                 PERROR("socketpair");
 100                 return -1;
 101         }
 102         return 0;
 103 }
 104
 105 /*
 106  * Creates a AF_UNIX local socket using pathname bind the socket upon creation
 107  * and return the fd.
 108  */
 109 LTTNG_HIDDEN
 110 int lttcomm_create_unix_sock(const char *pathname)
 111 {
 112         struct sockaddr_un s_un;
 113         int fd = -1;
 114         int ret = -1;
 115
 116         if (strlen(pathname) >= sizeof(s_un.sun_path)) {
 117                 ERR("unix socket address (\"%s\") is longer than the platform's limit (%zu > %zu).",
 118                                 pathname, strlen(pathname) + 1,
 119                                 sizeof(s_un.sun_path));
 120                 ret = -ENAMETOOLONG;
 121                 goto error;
 122         }
 123
 124         /* Create server socket */
 125         if ((fd = socket(PF_UNIX, SOCK_STREAM, 0)) < 0) {
 126                 PERROR("socket");
 127                 goto error;
 128         }
 129
 130         memset(&s_un, 0, sizeof(s_un));
 131         s_un.sun_family = AF_UNIX;
 132         strncpy(s_un.sun_path, pathname, sizeof(s_un.sun_path));
 133         s_un.sun_path[sizeof(s_un.sun_path) - 1] = '\0';
 134
 135         /* Unlink the old file if present */
 136         (void) unlink(pathname);
 137         ret = bind(fd, (struct sockaddr *) &s_un, sizeof(s_un));
 138         if (ret < 0) {
 139                 PERROR("bind");
 140                 goto error;
 141         }
 142
 143         return fd;
 144
 145 error:
 146         if (fd >= 0) {
 147                 if (close(fd) < 0) {
 148                         PERROR("close create unix sock");
 149                 }
 150         }
 151         return ret;
 152 }
 153
 154 /*
 155  * Make the socket listen using LTTNG_SESSIOND_COMM_MAX_LISTEN.
 156  */
 157 LTTNG_HIDDEN
 158 int lttcomm_listen_unix_sock(int sock)
 159 {
 160         int ret;
 161
 162         ret = listen(sock, LTTNG_SESSIOND_COMM_MAX_LISTEN);
 163         if (ret < 0) {
 164                 PERROR("listen");
 165         }
 166
 167         return ret;
 168 }
 169
 170 /*
 171  * Receive data of size len in put that data into the buf param. Using recvmsg
 172  * API.
 173  *
 174  * Return the size of received data.
 175  */
 176 LTTNG_HIDDEN
 177 ssize_t lttcomm_recv_unix_sock(int sock, void *buf, size_t len)
 178 {
 179         struct msghdr msg;
 180         struct iovec iov[1];
 181         ssize_t ret = -1;
 182         size_t len_last;
 183
 184         memset(&msg, 0, sizeof(msg));
 185
 186         iov[0].iov_base = buf;
 187         iov[0].iov_len = len;
 188         msg.msg_iov = iov;
 189         msg.msg_iovlen = 1;
 190
 191         do {
 192                 len_last = iov[0].iov_len;
 193                 ret = lttng_recvmsg_nosigpipe(sock, &msg);
 194                 if (ret > 0) {
 195                         iov[0].iov_base += ret;
 196                         iov[0].iov_len -= ret;
 197                         assert(ret <= len_last);
 198                 }
 199         } while ((ret > 0 && ret < len_last) || (ret < 0 && errno == EINTR));
 200         if (ret < 0) {
 201                 PERROR("recvmsg");
 202         } else if (ret > 0) {
 203                 ret = len;
 204         }
 205         /* Else ret = 0 meaning an orderly shutdown. */
 206
 207         return ret;
 208 }
 209
 210 /*
 211  * Receive data of size len in put that data into the buf param. Using recvmsg
 212  * API. Only use with sockets set in non-blocking mode.
 213  *
 214  * NOTE: EPIPE errors are NOT reported. This call expects the socket to be in a
 215  * poll set. The poll loop will handle the EPIPE original cause.
 216  *
 217  * Return the size of received data.
 218  */
 219 LTTNG_HIDDEN
 220 ssize_t lttcomm_recv_unix_sock_non_block(int sock, void *buf, size_t len)
 221 {
 222         struct msghdr msg;
 223         struct iovec iov[1];
 224         ssize_t ret;
 225
 226         memset(&msg, 0, sizeof(msg));
 227
 228         iov[0].iov_base = buf;
 229         iov[0].iov_len = len;
 230         msg.msg_iov = iov;
 231         msg.msg_iovlen = 1;
 232
 233 retry:
 234         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 235         if (ret < 0) {
 236                 if (errno == EINTR) {
 237                         goto retry;
 238                 } else {
 239                         /*
 240                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 241                          */
 242                         if (errno == EAGAIN || errno == EWOULDBLOCK ||
 243                                         errno == EPIPE) {
 244                                 /*
 245                                  * Nothing was recv.
 246                                  */
 247                                 ret = 0;
 248                                 goto end;
 249                         }
 250
 251                         /* Unexpected error */
 252                         PERROR("recvmsg");
 253                         ret = -1;
 254                         goto end;
 255                 }
 256         }
 257
 258 end:
 259         return ret;
 260 }
 261
 262 /*
 263  * Send buf data of size len. Using sendmsg API.
 264  *
 265  * Return the size of sent data.
 266  */
 267 LTTNG_HIDDEN
 268 ssize_t lttcomm_send_unix_sock(int sock, const void *buf, size_t len)
 269 {
 270         struct msghdr msg;
 271         struct iovec iov[1];
 272         ssize_t ret;
 273
 274         memset(&msg, 0, sizeof(msg));
 275
 276         iov[0].iov_base = (void *) buf;
 277         iov[0].iov_len = len;
 278         msg.msg_iov = iov;
 279         msg.msg_iovlen = 1;
 280
 281         while (iov[0].iov_len) {
 282                 ret = sendmsg(sock, &msg, 0);
 283                 if (ret < 0) {
 284                         if (errno == EINTR) {
 285                                 continue;
 286                         } else {
 287                                 /*
 288                                  * Only warn about EPIPE when quiet mode is
 289                                  * deactivated.
 290                                  * We consider EPIPE as expected.
 291                                  */
 292                                 if (errno != EPIPE || !lttng_opt_quiet) {
 293                                         PERROR("sendmsg");
 294                                 }
 295                                 goto end;
 296                         }
 297                 }
 298                 iov[0].iov_len -= ret;
 299                 iov[0].iov_base += ret;
 300         }
 301         ret = len;
 302 end:
 303         return ret;
 304 }
 305
 306 /*
 307  * Send buf data of size len. Using sendmsg API.
 308  * Only use with non-blocking sockets. The difference with the blocking version
 309  * of the function is that this one does not retry to send on partial sends,
 310  * except if the interruption was caused by a signal (EINTR).
 311  *
 312  * NOTE: EPIPE errors are NOT reported. This call expects the socket to be in a
 313  * poll set. The poll loop will handle the EPIPE original cause.
 314  *
 315  * Return the size of sent data.
 316  */
 317 LTTNG_HIDDEN
 318 ssize_t lttcomm_send_unix_sock_non_block(int sock, const void *buf, size_t len)
 319 {
 320         struct msghdr msg;
 321         struct iovec iov[1];
 322         ssize_t ret;
 323
 324         memset(&msg, 0, sizeof(msg));
 325
 326         iov[0].iov_base = (void *) buf;
 327         iov[0].iov_len = len;
 328         msg.msg_iov = iov;
 329         msg.msg_iovlen = 1;
 330
 331 retry:
 332         ret = sendmsg(sock, &msg, 0);
 333         if (ret < 0) {
 334                 if (errno == EINTR) {
 335                         goto retry;
 336                 } else {
 337                         /*
 338                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 339                          */
 340                         if (errno == EAGAIN || errno == EWOULDBLOCK ||
 341                                         errno == EPIPE) {
 342                                 /*
 343                                  * This can happen in non blocking mode.
 344                                  * Nothing was sent.
 345                                  */
 346                                 ret = 0;
 347                                 goto end;
 348                         }
 349
 350                         /* Unexpected error */
 351                         PERROR("sendmsg");
 352                         ret = -1;
 353                         goto end;
 354                 }
 355         }
 356 end:
 357         return ret;
 358 }
 359
 360 /*
 361  * Shutdown cleanly a unix socket.
 362  */
 363 LTTNG_HIDDEN
 364 int lttcomm_close_unix_sock(int sock)
 365 {
 366         int ret, closeret;
 367
 368         /* Shutdown receptions and transmissions */
 369         ret = shutdown(sock, SHUT_RDWR);
 370         if (ret < 0) {
 371                 PERROR("shutdown");
 372         }
 373
 374         closeret = close(sock);
 375         if (closeret) {
 376                 PERROR("close");
 377         }
 378
 379         return ret;
 380 }
 381
 382 /*
 383  * Send a message accompanied by fd(s) over a unix socket.
 384  *
 385  * Returns the size of data sent, or negative error value.
 386  */
 387 LTTNG_HIDDEN
 388 ssize_t lttcomm_send_fds_unix_sock(int sock, const int *fds, size_t nb_fd)
 389 {
 390         struct msghdr msg;
 391         struct cmsghdr *cmptr;
 392         struct iovec iov[1];
 393         ssize_t ret = -1;
 394         unsigned int sizeof_fds = nb_fd * sizeof(int);
 395         char tmp[CMSG_SPACE(sizeof_fds)];
 396         char dummy = 0;
 397
 398         memset(&msg, 0, sizeof(msg));
 399         memset(tmp, 0, sizeof(tmp));
 400
 401         if (nb_fd > LTTCOMM_MAX_SEND_FDS)
 402                 return -EINVAL;
 403
 404         msg.msg_control = (caddr_t)tmp;
 405         msg.msg_controllen = CMSG_LEN(sizeof_fds);
 406
 407         cmptr = CMSG_FIRSTHDR(&msg);
 408         if (!cmptr) {
 409                 return -1;
 410         }
 411
 412         cmptr->cmsg_level = SOL_SOCKET;
 413         cmptr->cmsg_type = SCM_RIGHTS;
 414         cmptr->cmsg_len = CMSG_LEN(sizeof_fds);
 415         memcpy(CMSG_DATA(cmptr), fds, sizeof_fds);
 416         /* Sum of the length of all control messages in the buffer: */
 417         msg.msg_controllen = cmptr->cmsg_len;
 418
 419         iov[0].iov_base = &dummy;
 420         iov[0].iov_len = 1;
 421         msg.msg_iov = iov;
 422         msg.msg_iovlen = 1;
 423
 424         do {
 425                 ret = sendmsg(sock, &msg, 0);
 426         } while (ret < 0 && errno == EINTR);
 427         if (ret < 0) {
 428                 /*
 429                  * Only warn about EPIPE when quiet mode is deactivated.
 430                  * We consider EPIPE as expected.
 431                  */
 432                 if (errno != EPIPE || !lttng_opt_quiet) {
 433                         PERROR("sendmsg");
 434                 }
 435         }
 436         return ret;
 437 }
 438
 439 /*
 440  * Send the fd(s) of a payload view over a unix socket.
 441  *
 442  * Returns the size of data sent, or negative error value.
 443  */
 444 static
 445 ssize_t _lttcomm_send_payload_view_fds_unix_sock(int sock,
 446                 struct lttng_payload_view *view,
 447                 bool blocking)
 448 {
 449         int i;
 450         ssize_t ret;
 451         struct lttng_dynamic_array raw_fds;
 452         const int fd_count = lttng_payload_view_get_fd_handle_count(view);
 453
 454         lttng_dynamic_array_init(&raw_fds, sizeof(int), NULL);
 455
 456         /*
 457          * Prepare a contiguous array of file descriptors to send them.
 458          *
 459          * Note that the reference to each fd is released during the iteration;
 460          * we're just getting the numerical value of the fds to conform to the
 461          * syscall's interface. We rely on the fact that "view" must remain
 462          * valid for the duration of the call and that the underlying payload
 463          * owns a reference to the fd_handles.
 464          */
 465         for (i = 0; i < fd_count; i++) {
 466                 struct fd_handle *handle =
 467                                 lttng_payload_view_pop_fd_handle(view);
 468                 const int raw_fd = fd_handle_get_fd(handle);
 469                 const int add_ret = lttng_dynamic_array_add_element(
 470                                 &raw_fds, &raw_fd);
 471
 472                 fd_handle_put(handle);
 473                 if (add_ret) {
 474                         ret = -LTTNG_ERR_NOMEM;
 475                         goto end;
 476                 }
 477         }
 478
 479         if (blocking) {
 480                 ret = lttcomm_send_fds_unix_sock(sock,
 481                                 (const int *) raw_fds.buffer.data, fd_count);
 482         } else {
 483                 ret = lttcomm_send_fds_unix_sock_non_block(sock,
 484                                 (const int *) raw_fds.buffer.data, fd_count);
 485         }
 486
 487 end:
 488         lttng_dynamic_array_reset(&raw_fds);
 489         return ret;
 490 }
 491
 492 LTTNG_HIDDEN
 493 ssize_t lttcomm_send_payload_view_fds_unix_sock(int sock,
 494                 struct lttng_payload_view *view)
 495 {
 496         return _lttcomm_send_payload_view_fds_unix_sock(sock, view, true);
 497 }
 498
 499 LTTNG_HIDDEN
 500 ssize_t lttcomm_send_payload_view_fds_unix_sock_non_block(int sock,
 501                 struct lttng_payload_view *view)
 502 {
 503         return _lttcomm_send_payload_view_fds_unix_sock(sock, view, false);
 504 }
 505
 506 /*
 507  * Send a message accompanied by fd(s) over a unix socket.
 508  * Only use for non blocking socket.
 509  *
 510  * Returns the size of data sent, or negative error value.
 511  */
 512 LTTNG_HIDDEN
 513 ssize_t lttcomm_send_fds_unix_sock_non_block(int sock, const int *fds, size_t nb_fd)
 514 {
 515         struct msghdr msg;
 516         struct cmsghdr *cmptr;
 517         struct iovec iov[1];
 518         ssize_t ret = -1;
 519         unsigned int sizeof_fds = nb_fd * sizeof(int);
 520         char tmp[CMSG_SPACE(sizeof_fds)];
 521         char dummy = 0;
 522
 523         memset(&msg, 0, sizeof(msg));
 524         memset(tmp, 0, sizeof(tmp));
 525
 526         if (nb_fd > LTTCOMM_MAX_SEND_FDS)
 527                 return -EINVAL;
 528
 529         msg.msg_control = (caddr_t)tmp;
 530         msg.msg_controllen = CMSG_LEN(sizeof_fds);
 531
 532         cmptr = CMSG_FIRSTHDR(&msg);
 533         if (!cmptr) {
 534                 return -1;
 535         }
 536
 537         cmptr->cmsg_level = SOL_SOCKET;
 538         cmptr->cmsg_type = SCM_RIGHTS;
 539         cmptr->cmsg_len = CMSG_LEN(sizeof_fds);
 540         memcpy(CMSG_DATA(cmptr), fds, sizeof_fds);
 541         /* Sum of the length of all control messages in the buffer: */
 542         msg.msg_controllen = cmptr->cmsg_len;
 543
 544         iov[0].iov_base = &dummy;
 545         iov[0].iov_len = 1;
 546         msg.msg_iov = iov;
 547         msg.msg_iovlen = 1;
 548
 549 retry:
 550         ret = sendmsg(sock, &msg, 0);
 551         if (ret < 0) {
 552                 if (errno == EINTR) {
 553                         goto retry;
 554                 } else {
 555                         /*
 556                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 557                          */
 558                         if (errno == EAGAIN || errno == EWOULDBLOCK) {
 559                                 /*
 560                                  * This can happen in non blocking mode.
 561                                  * Nothing was sent.
 562                                  */
 563                                 ret = 0;
 564                                 goto end;
 565                         }
 566
 567                         if (errno == EPIPE) {
 568                                 /* Expected error, pass error to caller */
 569                                 DBG3("EPIPE on sendmsg");
 570                                 ret = -1;
 571                                 goto end;
 572                         }
 573
 574                         /* Unexpected error */
 575                         PERROR("sendmsg");
 576                         ret = -1;
 577                         goto end;
 578                 }
 579         }
 580
 581 end:
 582         return ret;
 583 }
 584
 585 /*
 586  * Recv a message accompanied by fd(s) from a unix socket.
 587  *
 588  * Returns the size of received data, or negative error value.
 589  *
 590  * Expect at most "nb_fd" file descriptors. Returns the number of fd
 591  * actually received in nb_fd.
 592  */
 593 LTTNG_HIDDEN
 594 ssize_t lttcomm_recv_fds_unix_sock(int sock, int *fds, size_t nb_fd)
 595 {
 596         struct iovec iov[1];
 597         ssize_t ret = 0;
 598         struct cmsghdr *cmsg;
 599         size_t sizeof_fds = nb_fd * sizeof(int);
 600
 601 #ifdef __linux__
 602 /* Account for the struct ucred cmsg in the buffer size */
 603 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds) + CMSG_SPACE(sizeof(struct ucred))
 604 #else
 605 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds)
 606 #endif /* __linux__ */
 607
 608         char recv_buf[LTTNG_SOCK_RECV_FDS_BUF_SIZE];
 609         struct msghdr msg;
 610         char dummy;
 611
 612         memset(&msg, 0, sizeof(msg));
 613
 614         /* Prepare to receive the structures */
 615         iov[0].iov_base = &dummy;
 616         iov[0].iov_len = 1;
 617         msg.msg_iov = iov;
 618         msg.msg_iovlen = 1;
 619
 620         cmsg = (struct cmsghdr *) recv_buf;
 621         cmsg->cmsg_len = CMSG_LEN(sizeof_fds);
 622         cmsg->cmsg_level = SOL_SOCKET;
 623         cmsg->cmsg_type = SCM_RIGHTS;
 624
 625         msg.msg_control = cmsg;
 626         msg.msg_controllen = CMSG_LEN(sizeof(recv_buf));
 627         msg.msg_flags = 0;
 628
 629 retry:
 630         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 631         if (ret < 0) {
 632                 if (errno == EINTR) {
 633                         goto retry;
 634                 } else {
 635                         /* We consider EPIPE and EAGAIN as expected. */
 636                         if (!lttng_opt_quiet &&
 637                                         (errno != EPIPE && errno != EAGAIN)) {
 638                                 PERROR("recvmsg");
 639                         }
 640                         goto end;
 641                 }
 642         }
 643
 644         if (ret != 1) {
 645                 fprintf(stderr, "Error: Received %zd bytes, expected %d\n",
 646                                 ret, 1);
 647                 goto end;
 648         }
 649
 650         if (msg.msg_flags & MSG_CTRUNC) {
 651                 fprintf(stderr, "Error: Control message truncated.\n");
 652                 ret = -1;
 653                 goto end;
 654         }
 655
 656         /*
 657          * If the socket was configured with SO_PASSCRED, the kernel will add a
 658          * control message (cmsg) to the ancillary data of the unix socket. We
 659          * need to expect a cmsg of the SCM_CREDENTIALS as the first control
 660          * message.
 661          */
 662         for (cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 663                 if (cmsg->cmsg_level != SOL_SOCKET) {
 664                         fprintf(stderr, "Error: The socket needs to be of type SOL_SOCKET\n");
 665                         ret = -1;
 666                         goto end;
 667                 }
 668                 if (cmsg->cmsg_type == SCM_RIGHTS) {
 669                         /*
 670                          * We found the controle message for file descriptors,
 671                          * now copy the fds to the fds ptr and return success.
 672                          */
 673                         if (cmsg->cmsg_len != CMSG_LEN(sizeof_fds)) {
 674                                 fprintf(stderr, "Error: Received %zu bytes of"
 675                                         "ancillary data for FDs, expected %zu\n",
 676                                         (size_t) cmsg->cmsg_len,
 677                                         (size_t) CMSG_LEN(sizeof_fds));
 678                                 ret = -1;
 679                                 goto end;
 680                         }
 681                         memcpy(fds, CMSG_DATA(cmsg), sizeof_fds);
 682                         ret = sizeof_fds;
 683                         goto end;
 684                 }
 685 #ifdef __linux__
 686                 if (cmsg->cmsg_type == SCM_CREDENTIALS) {
 687                         /*
 688                          * Expect credentials to be sent when expecting fds even
 689                          * if no credential were include in the send(). The
 690                          * kernel adds them...
 691                          */
 692                         ret = -1;
 693                 }
 694 #endif /* __linux__ */
 695         }
 696 end:
 697         return ret;
 698 }
 699
 700 static
 701 void close_raw_fd(void *ptr)
 702 {
 703         const int raw_fd = *((const int *) ptr);
 704
 705         if (raw_fd >= 0) {
 706                 const int ret = close(raw_fd);
 707
 708                 if (ret) {
 709                         PERROR("Failed to close file descriptor %d", raw_fd);
 710                 }
 711         }
 712 }
 713
 714 static
 715 enum lttng_error_code add_fds_to_payload(struct lttng_dynamic_array *raw_fds,
 716                 struct lttng_payload *payload)
 717 {
 718         int i;
 719         enum lttng_error_code ret_code = LTTNG_OK;
 720         const int fd_count = lttng_dynamic_array_get_count(raw_fds);
 721
 722         for (i = 0; i < fd_count; i++) {
 723                 int ret;
 724                 struct fd_handle *handle;
 725                 int *raw_fd = (int *) lttng_dynamic_array_get_element(
 726                         raw_fds, i);
 727
 728                 handle = fd_handle_create(*raw_fd);
 729                 if (!handle) {
 730                         ret_code = LTTNG_ERR_NOMEM;
 731                         goto end;
 732                 }
 733
 734                 /* FD ownership transferred to the handle. */
 735                 *raw_fd = -1;
 736
 737                 ret = lttng_payload_push_fd_handle(payload, handle);
 738                 fd_handle_put(handle);
 739                 if (ret) {
 740                         ret_code = LTTNG_ERR_NOMEM;
 741                         goto end;
 742                 }
 743         }
 744
 745 end:
 746         return ret_code;
 747 }
 748
 749 static
 750 ssize_t _lttcomm_recv_payload_fds_unix_sock(int sock, size_t nb_fd,
 751                 struct lttng_payload *payload, bool blocking)
 752 {
 753         enum lttng_error_code add_ret;
 754         ssize_t ret;
 755         struct lttng_dynamic_array raw_fds;
 756
 757         lttng_dynamic_array_init(&raw_fds, sizeof(int), close_raw_fd);
 758         ret = lttng_dynamic_array_set_count(&raw_fds, nb_fd);
 759         if (ret) {
 760                 ret = -LTTNG_ERR_NOMEM;
 761                 goto end;
 762         }
 763
 764         if (blocking) {
 765                 ret = lttcomm_recv_fds_unix_sock(
 766                                 sock, (int *) raw_fds.buffer.data, nb_fd);
 767         } else {
 768                 ret = lttcomm_recv_fds_unix_sock_non_block(
 769                                 sock, (int *) raw_fds.buffer.data, nb_fd);
 770         }
 771
 772         if (ret < 0) {
 773                 goto end;
 774         }
 775
 776         add_ret = add_fds_to_payload(&raw_fds, payload);
 777         if (add_ret != LTTNG_OK) {
 778                 ret = - (int) add_ret;
 779                 goto end;
 780         }
 781
 782 end:
 783         lttng_dynamic_array_reset(&raw_fds);
 784         return ret;
 785 }
 786
 787 LTTNG_HIDDEN
 788 ssize_t lttcomm_recv_payload_fds_unix_sock(int sock, size_t nb_fd,
 789                            struct lttng_payload *payload)
 790 {
 791         return _lttcomm_recv_payload_fds_unix_sock(sock, nb_fd, payload, true);
 792 }
 793
 794 LTTNG_HIDDEN
 795 ssize_t lttcomm_recv_payload_fds_unix_sock_non_block(int sock, size_t nb_fd,
 796                            struct lttng_payload *payload)
 797 {
 798         return _lttcomm_recv_payload_fds_unix_sock(sock, nb_fd, payload, false);
 799 }
 800
 801 /*
 802  * Recv a message accompanied by fd(s) from a non-blocking unix socket.
 803  * Only use with non-blocking sockets.
 804  *
 805  * Returns the size of received data, or negative error value.
 806  *
 807  * Expect at most "nb_fd" file descriptors.
 808  *
 809  * Note that based on our comprehension, partial reception of fds is not
 810  * possible since the FDs are actually in the control message. It is all or
 811  * nothing, still the sender side can send the wrong number of fds.
 812  */
 813 LTTNG_HIDDEN
 814 ssize_t lttcomm_recv_fds_unix_sock_non_block(int sock, int *fds, size_t nb_fd)
 815 {
 816         struct iovec iov[1];
 817         ssize_t ret = 0;
 818         struct cmsghdr *cmsg;
 819         size_t sizeof_fds = nb_fd * sizeof(int);
 820
 821 #ifdef __linux__
 822 /* Account for the struct ucred cmsg in the buffer size */
 823 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds) + CMSG_SPACE(sizeof(struct ucred))
 824 #else
 825 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds)
 826 #endif /* __linux__ */
 827
 828         char recv_buf[LTTNG_SOCK_RECV_FDS_BUF_SIZE];
 829         struct msghdr msg;
 830         char dummy;
 831
 832         memset(&msg, 0, sizeof(msg));
 833
 834         /* Prepare to receive the structures */
 835         iov[0].iov_base = &dummy;
 836         iov[0].iov_len = 1;
 837         msg.msg_iov = iov;
 838         msg.msg_iovlen = 1;
 839
 840         cmsg = (struct cmsghdr *) recv_buf;
 841         cmsg->cmsg_len = CMSG_LEN(sizeof_fds);
 842         cmsg->cmsg_level = SOL_SOCKET;
 843         cmsg->cmsg_type = SCM_RIGHTS;
 844
 845         msg.msg_control = cmsg;
 846         msg.msg_controllen = CMSG_LEN(sizeof(recv_buf));
 847         msg.msg_flags = 0;
 848
 849 retry:
 850         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 851         if (ret < 0) {
 852                 if (errno == EINTR) {
 853                         goto retry;
 854                 } else {
 855                         /*
 856                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 857                          */
 858                         if (errno == EAGAIN || errno == EWOULDBLOCK) {
 859                                 /*
 860                                  * This can happen in non blocking mode.
 861                                  * Nothing was recv.
 862                                  */
 863                                 ret = 0;
 864                                 goto end;
 865                         }
 866
 867                         if (errno == EPIPE) {
 868                                 /* Expected error, pass error to caller */
 869                                 DBG3("EPIPE on recvmsg");
 870                                 ret = -1;
 871                                 goto end;
 872                         }
 873
 874                         /* Unexpected error */
 875                         PERROR("recvmsg");
 876                         ret = -1;
 877                         goto end;
 878                 }
 879         }
 880
 881         if (ret != 1) {
 882                 fprintf(stderr, "Error: Received %zd bytes, expected %d\n",
 883                                 ret, 1);
 884                 goto end;
 885         }
 886
 887         if (msg.msg_flags & MSG_CTRUNC) {
 888                 fprintf(stderr, "Error: Control message truncated.\n");
 889                 ret = -1;
 890                 goto end;
 891         }
 892
 893         /*
 894          * If the socket was configured with SO_PASSCRED, the kernel will add a
 895          * control message (cmsg) to the ancillary data of the unix socket. We
 896          * need to expect a cmsg of the SCM_CREDENTIALS as the first control
 897          * message.
 898          */
 899         for (cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 900                 if (cmsg->cmsg_level != SOL_SOCKET) {
 901                         fprintf(stderr, "Error: The socket needs to be of type SOL_SOCKET\n");
 902                         ret = -1;
 903                         goto end;
 904                 }
 905                 if (cmsg->cmsg_type == SCM_RIGHTS) {
 906                         /*
 907                          * We found the controle message for file descriptors,
 908                          * now copy the fds to the fds ptr and return success.
 909                          */
 910                         if (cmsg->cmsg_len != CMSG_LEN(sizeof_fds)) {
 911                                 fprintf(stderr, "Error: Received %zu bytes of"
 912                                         "ancillary data for FDs, expected %zu\n",
 913                                         (size_t) cmsg->cmsg_len,
 914                                         (size_t) CMSG_LEN(sizeof_fds));
 915                                 ret = -1;
 916                                 goto end;
 917                         }
 918                         memcpy(fds, CMSG_DATA(cmsg), sizeof_fds);
 919                         ret = sizeof_fds;
 920                         goto end;
 921                 }
 922 #ifdef __linux__
 923                 if (cmsg->cmsg_type == SCM_CREDENTIALS) {
 924                         /*
 925                          * Expect credentials to be sent when expecting fds even
 926                          * if no credential were include in the send(). The
 927                          * kernel adds them...
 928                          */
 929                         ret = -1;
 930                 }
 931 #endif /* __linux__ */
 932         }
 933 end:
 934         return ret;
 935 }
 936
 937 /*
 938  * Send a message with credentials over a unix socket.
 939  *
 940  * Returns the size of data sent, or negative error value.
 941  */
 942 LTTNG_HIDDEN
 943 ssize_t lttcomm_send_creds_unix_sock(int sock, const void *buf, size_t len)
 944 {
 945         struct msghdr msg;
 946         struct iovec iov[1];
 947         ssize_t ret = -1;
 948 #ifdef __linux__
 949         struct cmsghdr *cmptr;
 950         size_t sizeof_cred = sizeof(lttng_sock_cred);
 951         char anc_buf[CMSG_SPACE(sizeof_cred)];
 952         lttng_sock_cred *creds;
 953
 954         memset(anc_buf, 0, CMSG_SPACE(sizeof_cred) * sizeof(char));
 955 #endif /* __linux__ */
 956
 957         memset(&msg, 0, sizeof(msg));
 958
 959         iov[0].iov_base = (void *) buf;
 960         iov[0].iov_len = len;
 961         msg.msg_iov = iov;
 962         msg.msg_iovlen = 1;
 963
 964 #ifdef __linux__
 965         msg.msg_control = (caddr_t) anc_buf;
 966         msg.msg_controllen = CMSG_LEN(sizeof_cred);
 967
 968         cmptr = CMSG_FIRSTHDR(&msg);
 969         if (!cmptr) {
 970                 return -1;
 971         }
 972         cmptr->cmsg_level = SOL_SOCKET;
 973         cmptr->cmsg_type = LTTNG_SOCK_CREDS;
 974         cmptr->cmsg_len = CMSG_LEN(sizeof_cred);
 975
 976         creds = (lttng_sock_cred*) CMSG_DATA(cmptr);
 977
 978         LTTNG_SOCK_SET_UID_CRED(creds, geteuid());
 979         LTTNG_SOCK_SET_GID_CRED(creds, getegid());
 980         LTTNG_SOCK_SET_PID_CRED(creds, getpid());
 981 #endif /* __linux__ */
 982
 983         do {
 984                 ret = sendmsg(sock, &msg, 0);
 985         } while (ret < 0 && errno == EINTR);
 986         if (ret < 0) {
 987                 /*
 988                  * Only warn about EPIPE when quiet mode is deactivated.
 989                  * We consider EPIPE as expected.
 990                  */
 991                 if (errno != EPIPE || !lttng_opt_quiet) {
 992                         PERROR("sendmsg");
 993                 }
 994         }
 995         return ret;
 996 }
 997
 998 /*
 999  * Recv a message accompanied with credentials from a unix socket.
1000  *
1001  * Returns the size of received data, or negative error value.
1002  */
1003 LTTNG_HIDDEN
1004 ssize_t lttcomm_recv_creds_unix_sock(int sock, void *buf, size_t len,
1005                 lttng_sock_cred *creds)
1006 {
1007         struct msghdr msg;
1008         struct iovec iov[1];
1009         ssize_t ret;
1010         size_t len_last;
1011 #ifdef __linux__
1012         struct cmsghdr *cmptr;
1013         size_t sizeof_cred = sizeof(lttng_sock_cred);
1014         char anc_buf[CMSG_SPACE(sizeof_cred)];
1015 #endif  /* __linux__ */
1016
1017         memset(&msg, 0, sizeof(msg));
1018
1019         /* Not allowed */
1020         if (creds == NULL) {
1021                 ret = -1;
1022                 goto end;
1023         }
1024
1025         /* Prepare to receive the structures */
1026         iov[0].iov_base = buf;
1027         iov[0].iov_len = len;
1028         msg.msg_iov = iov;
1029         msg.msg_iovlen = 1;
1030
1031 #ifdef __linux__
1032         msg.msg_control = anc_buf;
1033         msg.msg_controllen = sizeof(anc_buf);
1034 #endif /* __linux__ */
1035
1036         do {
1037                 len_last = iov[0].iov_len;
1038                 ret = recvmsg(sock, &msg, 0);
1039                 if (ret > 0) {
1040                         iov[0].iov_base += ret;
1041                         iov[0].iov_len -= ret;
1042                         assert(ret <= len_last);
1043                 }
1044         } while ((ret > 0 && ret < len_last) || (ret < 0 && errno == EINTR));
1045         if (ret < 0) {
1046                 PERROR("recvmsg fds");
1047                 goto end;
1048         } else if (ret > 0) {
1049                 ret = len;
1050         }
1051         /* Else ret = 0 meaning an orderly shutdown. */
1052
1053 #ifdef __linux__
1054         if (msg.msg_flags & MSG_CTRUNC) {
1055                 fprintf(stderr, "Error: Control message truncated.\n");
1056                 ret = -1;
1057                 goto end;
1058         }
1059
1060         cmptr = CMSG_FIRSTHDR(&msg);
1061         if (cmptr == NULL) {
1062                 fprintf(stderr, "Error: Invalid control message header\n");
1063                 ret = -1;
1064                 goto end;
1065         }
1066
1067         if (cmptr->cmsg_level != SOL_SOCKET ||
1068                         cmptr->cmsg_type != LTTNG_SOCK_CREDS) {
1069                 fprintf(stderr, "Didn't received any credentials\n");
1070                 ret = -1;
1071                 goto end;
1072         }
1073
1074         if (cmptr->cmsg_len != CMSG_LEN(sizeof_cred)) {
1075                 fprintf(stderr, "Error: Received %zu bytes of ancillary data, expected %zu\n",
1076                                 (size_t) cmptr->cmsg_len, (size_t) CMSG_LEN(sizeof_cred));
1077                 ret = -1;
1078                 goto end;
1079         }
1080
1081         memcpy(creds, CMSG_DATA(cmptr), sizeof_cred);
1082 #elif (defined(__FreeBSD__) || defined(__CYGWIN__) || defined(__sun__) || defined(__APPLE__))
1083         {
1084                 int peer_ret;
1085
1086                 peer_ret = getpeereid(sock, &creds->uid, &creds->gid);
1087                 if (peer_ret != 0) {
1088                         return peer_ret;
1089                 }
1090         }
1091 #else
1092 #error "Please implement credential support for your OS."
1093 #endif  /* __linux__ */
1094
1095 end:
1096         return ret;
1097 }
1098
1099 /*
1100  * Set socket option to use credentials passing.
1101  */
1102 #ifdef __linux__
1103 LTTNG_HIDDEN
1104 int lttcomm_setsockopt_creds_unix_sock(int sock)
1105 {
1106         int ret, on = 1;
1107
1108         /* Set socket for credentials retrieval */
1109         ret = setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &on, sizeof(on));
1110         if (ret < 0) {
1111                 PERROR("setsockopt creds unix sock");
1112         }
1113         return ret;
1114 }
1115 #elif (defined(__FreeBSD__) || defined(__CYGWIN__) || defined(__sun__) || defined(__APPLE__))
1116 LTTNG_HIDDEN
1117 int lttcomm_setsockopt_creds_unix_sock(int sock)
1118 {
1119         return 0;
1120 }
1121 #else
1122 #error "Please implement credential support for your OS."
1123 #endif /* __linux__ */