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
  19 #include <common/common.h>
  20 #include <common/compat/errno.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         assert(sock);
 185         assert(buf);
 186         assert(len > 0);
 187
 188         memset(&msg, 0, sizeof(msg));
 189
 190         iov[0].iov_base = buf;
 191         iov[0].iov_len = len;
 192         msg.msg_iov = iov;
 193         msg.msg_iovlen = 1;
 194
 195         do {
 196                 len_last = iov[0].iov_len;
 197                 ret = lttng_recvmsg_nosigpipe(sock, &msg);
 198                 if (ret > 0) {
 199                         iov[0].iov_base += ret;
 200                         iov[0].iov_len -= ret;
 201                         assert(ret <= len_last);
 202                 }
 203         } while ((ret > 0 && ret < len_last) || (ret < 0 && errno == EINTR));
 204         if (ret < 0) {
 205                 PERROR("recvmsg");
 206         } else if (ret > 0) {
 207                 ret = len;
 208         }
 209         /* Else ret = 0 meaning an orderly shutdown. */
 210
 211         return ret;
 212 }
 213
 214 /*
 215  * Receive data of size len in put that data into the buf param. Using recvmsg
 216  * API. Only use with sockets set in non-blocking mode.
 217  *
 218  * NOTE: EPIPE errors are NOT reported. This call expects the socket to be in a
 219  * poll set. The poll loop will handle the EPIPE original cause.
 220  *
 221  * Return the size of received data.
 222  */
 223 LTTNG_HIDDEN
 224 ssize_t lttcomm_recv_unix_sock_non_block(int sock, void *buf, size_t len)
 225 {
 226         struct msghdr msg;
 227         struct iovec iov[1];
 228         ssize_t ret;
 229
 230         assert(sock);
 231         assert(buf);
 232         assert(len > 0);
 233
 234         memset(&msg, 0, sizeof(msg));
 235
 236         iov[0].iov_base = buf;
 237         iov[0].iov_len = len;
 238         msg.msg_iov = iov;
 239         msg.msg_iovlen = 1;
 240
 241 retry:
 242         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 243         if (ret < 0) {
 244                 if (errno == EINTR) {
 245                         goto retry;
 246                 } else {
 247                         /*
 248                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 249                          */
 250                         if (errno == EAGAIN || errno == EWOULDBLOCK ||
 251                                         errno == EPIPE) {
 252                                 /*
 253                                  * Nothing was recv.
 254                                  */
 255                                 ret = 0;
 256                                 goto end;
 257                         }
 258
 259                         /* Unexpected error */
 260                         PERROR("recvmsg");
 261                         ret = -1;
 262                         goto end;
 263                 }
 264         }
 265
 266 end:
 267         return ret;
 268 }
 269
 270 /*
 271  * Send buf data of size len. Using sendmsg API.
 272  *
 273  * Return the size of sent data.
 274  */
 275 LTTNG_HIDDEN
 276 ssize_t lttcomm_send_unix_sock(int sock, const void *buf, size_t len)
 277 {
 278         struct msghdr msg;
 279         struct iovec iov[1];
 280         ssize_t ret;
 281
 282         assert(sock);
 283         assert(buf);
 284         assert(len > 0);
 285
 286         memset(&msg, 0, sizeof(msg));
 287
 288         iov[0].iov_base = (void *) buf;
 289         iov[0].iov_len = len;
 290         msg.msg_iov = iov;
 291         msg.msg_iovlen = 1;
 292
 293         while (iov[0].iov_len) {
 294                 ret = sendmsg(sock, &msg, 0);
 295                 if (ret < 0) {
 296                         if (errno == EINTR) {
 297                                 continue;
 298                         } else {
 299                                 /*
 300                                  * Only warn about EPIPE when quiet mode is
 301                                  * deactivated.
 302                                  * We consider EPIPE as expected.
 303                                  */
 304                                 if (errno != EPIPE || !lttng_opt_quiet) {
 305                                         PERROR("sendmsg");
 306                                 }
 307                                 goto end;
 308                         }
 309                 }
 310                 iov[0].iov_len -= ret;
 311                 iov[0].iov_base += ret;
 312         }
 313         ret = len;
 314 end:
 315         return ret;
 316 }
 317
 318 /*
 319  * Send buf data of size len. Using sendmsg API.
 320  * Only use with non-blocking sockets. The difference with the blocking version
 321  * of the function is that this one does not retry to send on partial sends,
 322  * except if the interruption was caused by a signal (EINTR).
 323  *
 324  * NOTE: EPIPE errors are NOT reported. This call expects the socket to be in a
 325  * poll set. The poll loop will handle the EPIPE original cause.
 326  *
 327  * Return the size of sent data.
 328  */
 329 LTTNG_HIDDEN
 330 ssize_t lttcomm_send_unix_sock_non_block(int sock, const void *buf, size_t len)
 331 {
 332         struct msghdr msg;
 333         struct iovec iov[1];
 334         ssize_t ret;
 335
 336         assert(sock);
 337         assert(buf);
 338         assert(len > 0);
 339
 340         memset(&msg, 0, sizeof(msg));
 341
 342         iov[0].iov_base = (void *) buf;
 343         iov[0].iov_len = len;
 344         msg.msg_iov = iov;
 345         msg.msg_iovlen = 1;
 346
 347 retry:
 348         ret = sendmsg(sock, &msg, 0);
 349         if (ret < 0) {
 350                 if (errno == EINTR) {
 351                         goto retry;
 352                 } else {
 353                         /*
 354                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 355                          */
 356                         if (errno == EAGAIN || errno == EWOULDBLOCK ||
 357                                         errno == EPIPE) {
 358                                 /*
 359                                  * This can happen in non blocking mode.
 360                                  * Nothing was sent.
 361                                  */
 362                                 ret = 0;
 363                                 goto end;
 364                         }
 365
 366                         /* Unexpected error */
 367                         PERROR("sendmsg");
 368                         ret = -1;
 369                         goto end;
 370                 }
 371         }
 372 end:
 373         return ret;
 374 }
 375
 376 /*
 377  * Shutdown cleanly a unix socket.
 378  */
 379 LTTNG_HIDDEN
 380 int lttcomm_close_unix_sock(int sock)
 381 {
 382         int ret, closeret;
 383
 384         /* Shutdown receptions and transmissions */
 385         ret = shutdown(sock, SHUT_RDWR);
 386         if (ret < 0) {
 387                 PERROR("shutdown");
 388         }
 389
 390         closeret = close(sock);
 391         if (closeret) {
 392                 PERROR("close");
 393         }
 394
 395         return ret;
 396 }
 397
 398 /*
 399  * Send a message accompanied by fd(s) over a unix socket.
 400  *
 401  * Returns the size of data sent, or negative error value.
 402  */
 403 LTTNG_HIDDEN
 404 ssize_t lttcomm_send_fds_unix_sock(int sock, const int *fds, size_t nb_fd)
 405 {
 406         struct msghdr msg;
 407         struct cmsghdr *cmptr;
 408         struct iovec iov[1];
 409         ssize_t ret = -1;
 410         unsigned int sizeof_fds = nb_fd * sizeof(int);
 411         char tmp[CMSG_SPACE(sizeof_fds)];
 412         char dummy = 0;
 413
 414         assert(sock);
 415         assert(fds);
 416         assert(nb_fd > 0);
 417
 418         memset(&msg, 0, sizeof(msg));
 419         memset(tmp, 0, sizeof(tmp));
 420
 421         if (nb_fd > LTTCOMM_MAX_SEND_FDS)
 422                 return -EINVAL;
 423
 424         msg.msg_control = (caddr_t)tmp;
 425         msg.msg_controllen = CMSG_LEN(sizeof_fds);
 426
 427         cmptr = CMSG_FIRSTHDR(&msg);
 428         if (!cmptr) {
 429                 return -1;
 430         }
 431
 432         cmptr->cmsg_level = SOL_SOCKET;
 433         cmptr->cmsg_type = SCM_RIGHTS;
 434         cmptr->cmsg_len = CMSG_LEN(sizeof_fds);
 435         memcpy(CMSG_DATA(cmptr), fds, sizeof_fds);
 436         /* Sum of the length of all control messages in the buffer: */
 437         msg.msg_controllen = cmptr->cmsg_len;
 438
 439         iov[0].iov_base = &dummy;
 440         iov[0].iov_len = 1;
 441         msg.msg_iov = iov;
 442         msg.msg_iovlen = 1;
 443
 444         do {
 445                 ret = sendmsg(sock, &msg, 0);
 446         } while (ret < 0 && errno == EINTR);
 447         if (ret < 0) {
 448                 /*
 449                  * Only warn about EPIPE when quiet mode is deactivated.
 450                  * We consider EPIPE as expected.
 451                  */
 452                 if (errno != EPIPE || !lttng_opt_quiet) {
 453                         PERROR("sendmsg");
 454                 }
 455         }
 456         return ret;
 457 }
 458
 459 /*
 460  * Send the fd(s) of a payload view over a unix socket.
 461  *
 462  * Returns the size of data sent, or negative error value.
 463  */
 464 static
 465 ssize_t _lttcomm_send_payload_view_fds_unix_sock(int sock,
 466                 struct lttng_payload_view *view,
 467                 bool blocking)
 468 {
 469         int i;
 470         ssize_t ret;
 471         struct lttng_dynamic_array raw_fds;
 472         const int fd_count = lttng_payload_view_get_fd_handle_count(view);
 473
 474         lttng_dynamic_array_init(&raw_fds, sizeof(int), NULL);
 475
 476         if (fd_count < 0) {
 477                 ret = -LTTNG_ERR_INVALID;
 478                 goto end;
 479         }
 480
 481         /*
 482          * Prepare a contiguous array of file descriptors to send them.
 483          *
 484          * Note that the reference to each fd is released during the iteration;
 485          * we're just getting the numerical value of the fds to conform to the
 486          * syscall's interface. We rely on the fact that "view" must remain
 487          * valid for the duration of the call and that the underlying payload
 488          * owns a reference to the fd_handles.
 489          */
 490         for (i = 0; i < fd_count; i++) {
 491                 struct fd_handle *handle =
 492                                 lttng_payload_view_pop_fd_handle(view);
 493                 const int raw_fd = fd_handle_get_fd(handle);
 494                 const int add_ret = lttng_dynamic_array_add_element(
 495                                 &raw_fds, &raw_fd);
 496
 497                 fd_handle_put(handle);
 498                 if (add_ret) {
 499                         ret = -LTTNG_ERR_NOMEM;
 500                         goto end;
 501                 }
 502         }
 503
 504         if (blocking) {
 505                 ret = lttcomm_send_fds_unix_sock(sock,
 506                                 (const int *) raw_fds.buffer.data, fd_count);
 507         } else {
 508                 ret = lttcomm_send_fds_unix_sock_non_block(sock,
 509                                 (const int *) raw_fds.buffer.data, fd_count);
 510         }
 511
 512 end:
 513         lttng_dynamic_array_reset(&raw_fds);
 514         return ret;
 515 }
 516
 517 LTTNG_HIDDEN
 518 ssize_t lttcomm_send_payload_view_fds_unix_sock(int sock,
 519                 struct lttng_payload_view *view)
 520 {
 521         return _lttcomm_send_payload_view_fds_unix_sock(sock, view, true);
 522 }
 523
 524 LTTNG_HIDDEN
 525 ssize_t lttcomm_send_payload_view_fds_unix_sock_non_block(int sock,
 526                 struct lttng_payload_view *view)
 527 {
 528         return _lttcomm_send_payload_view_fds_unix_sock(sock, view, false);
 529 }
 530
 531 /*
 532  * Send a message accompanied by fd(s) over a unix socket.
 533  * Only use for non blocking socket.
 534  *
 535  * Returns the size of data sent, or negative error value.
 536  */
 537 LTTNG_HIDDEN
 538 ssize_t lttcomm_send_fds_unix_sock_non_block(int sock, const int *fds, size_t nb_fd)
 539 {
 540         struct msghdr msg;
 541         struct cmsghdr *cmptr;
 542         struct iovec iov[1];
 543         ssize_t ret = -1;
 544         unsigned int sizeof_fds = nb_fd * sizeof(int);
 545         char tmp[CMSG_SPACE(sizeof_fds)];
 546         char dummy = 0;
 547
 548         assert(sock);
 549         assert(fds);
 550         assert(nb_fd > 0);
 551
 552         memset(&msg, 0, sizeof(msg));
 553         memset(tmp, 0, sizeof(tmp));
 554
 555         if (nb_fd > LTTCOMM_MAX_SEND_FDS)
 556                 return -EINVAL;
 557
 558         msg.msg_control = (caddr_t)tmp;
 559         msg.msg_controllen = CMSG_LEN(sizeof_fds);
 560
 561         cmptr = CMSG_FIRSTHDR(&msg);
 562         if (!cmptr) {
 563                 return -1;
 564         }
 565
 566         cmptr->cmsg_level = SOL_SOCKET;
 567         cmptr->cmsg_type = SCM_RIGHTS;
 568         cmptr->cmsg_len = CMSG_LEN(sizeof_fds);
 569         memcpy(CMSG_DATA(cmptr), fds, sizeof_fds);
 570         /* Sum of the length of all control messages in the buffer: */
 571         msg.msg_controllen = cmptr->cmsg_len;
 572
 573         iov[0].iov_base = &dummy;
 574         iov[0].iov_len = 1;
 575         msg.msg_iov = iov;
 576         msg.msg_iovlen = 1;
 577
 578 retry:
 579         ret = sendmsg(sock, &msg, 0);
 580         if (ret < 0) {
 581                 if (errno == EINTR) {
 582                         goto retry;
 583                 } else {
 584                         /*
 585                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 586                          */
 587                         if (errno == EAGAIN || errno == EWOULDBLOCK) {
 588                                 /*
 589                                  * This can happen in non blocking mode.
 590                                  * Nothing was sent.
 591                                  */
 592                                 ret = 0;
 593                                 goto end;
 594                         }
 595
 596                         if (errno == EPIPE) {
 597                                 /* Expected error, pass error to caller */
 598                                 DBG3("EPIPE on sendmsg");
 599                                 ret = -1;
 600                                 goto end;
 601                         }
 602
 603                         /* Unexpected error */
 604                         PERROR("sendmsg");
 605                         ret = -1;
 606                         goto end;
 607                 }
 608         }
 609
 610 end:
 611         return ret;
 612 }
 613
 614 /*
 615  * Recv a message accompanied by fd(s) from a unix socket.
 616  *
 617  * Returns the size of received data, or negative error value.
 618  *
 619  * Expect at most "nb_fd" file descriptors. Returns the number of fd
 620  * actually received in nb_fd.
 621  */
 622 LTTNG_HIDDEN
 623 ssize_t lttcomm_recv_fds_unix_sock(int sock, int *fds, size_t nb_fd)
 624 {
 625         struct iovec iov[1];
 626         ssize_t ret = 0;
 627         struct cmsghdr *cmsg;
 628         size_t sizeof_fds = nb_fd * sizeof(int);
 629
 630 #ifdef __linux__
 631 /* Account for the struct ucred cmsg in the buffer size */
 632 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds) + CMSG_SPACE(sizeof(struct ucred))
 633 #else
 634 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds)
 635 #endif /* __linux__ */
 636
 637         char recv_buf[LTTNG_SOCK_RECV_FDS_BUF_SIZE];
 638         struct msghdr msg;
 639         char dummy;
 640
 641         assert(sock);
 642         assert(fds);
 643         assert(nb_fd > 0);
 644
 645         memset(&msg, 0, sizeof(msg));
 646
 647         /* Prepare to receive the structures */
 648         iov[0].iov_base = &dummy;
 649         iov[0].iov_len = 1;
 650         msg.msg_iov = iov;
 651         msg.msg_iovlen = 1;
 652
 653         cmsg = (struct cmsghdr *) recv_buf;
 654         cmsg->cmsg_len = CMSG_LEN(sizeof_fds);
 655         cmsg->cmsg_level = SOL_SOCKET;
 656         cmsg->cmsg_type = SCM_RIGHTS;
 657
 658         msg.msg_control = cmsg;
 659         msg.msg_controllen = CMSG_LEN(sizeof(recv_buf));
 660         msg.msg_flags = 0;
 661
 662 retry:
 663         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 664         if (ret < 0) {
 665                 if (errno == EINTR) {
 666                         goto retry;
 667                 } else {
 668                         /* We consider EPIPE and EAGAIN as expected. */
 669                         if (!lttng_opt_quiet &&
 670                                         (errno != EPIPE && errno != EAGAIN)) {
 671                                 PERROR("recvmsg");
 672                         }
 673                         goto end;
 674                 }
 675         }
 676
 677         if (ret != 1) {
 678                 fprintf(stderr, "Error: Received %zd bytes, expected %d\n",
 679                                 ret, 1);
 680                 goto end;
 681         }
 682
 683         if (msg.msg_flags & MSG_CTRUNC) {
 684                 fprintf(stderr, "Error: Control message truncated.\n");
 685                 ret = -1;
 686                 goto end;
 687         }
 688
 689         /*
 690          * If the socket was configured with SO_PASSCRED, the kernel will add a
 691          * control message (cmsg) to the ancillary data of the unix socket. We
 692          * need to expect a cmsg of the SCM_CREDENTIALS as the first control
 693          * message.
 694          */
 695         for (cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 696                 if (cmsg->cmsg_level != SOL_SOCKET) {
 697                         fprintf(stderr, "Error: The socket needs to be of type SOL_SOCKET\n");
 698                         ret = -1;
 699                         goto end;
 700                 }
 701                 if (cmsg->cmsg_type == SCM_RIGHTS) {
 702                         /*
 703                          * We found the controle message for file descriptors,
 704                          * now copy the fds to the fds ptr and return success.
 705                          */
 706                         if (cmsg->cmsg_len != CMSG_LEN(sizeof_fds)) {
 707                                 fprintf(stderr, "Error: Received %zu bytes of"
 708                                         "ancillary data for FDs, expected %zu\n",
 709                                         (size_t) cmsg->cmsg_len,
 710                                         (size_t) CMSG_LEN(sizeof_fds));
 711                                 ret = -1;
 712                                 goto end;
 713                         }
 714                         memcpy(fds, CMSG_DATA(cmsg), sizeof_fds);
 715                         ret = sizeof_fds;
 716                         goto end;
 717                 }
 718 #ifdef __linux__
 719                 if (cmsg->cmsg_type == SCM_CREDENTIALS) {
 720                         /*
 721                          * Expect credentials to be sent when expecting fds even
 722                          * if no credential were include in the send(). The
 723                          * kernel adds them...
 724                          */
 725                         ret = -1;
 726                 }
 727 #endif /* __linux__ */
 728         }
 729 end:
 730         return ret;
 731 }
 732
 733 static
 734 void close_raw_fd(void *ptr)
 735 {
 736         const int raw_fd = *((const int *) ptr);
 737
 738         if (raw_fd >= 0) {
 739                 const int ret = close(raw_fd);
 740
 741                 if (ret) {
 742                         PERROR("Failed to close file descriptor %d", raw_fd);
 743                 }
 744         }
 745 }
 746
 747 static
 748 enum lttng_error_code add_fds_to_payload(struct lttng_dynamic_array *raw_fds,
 749                 struct lttng_payload *payload)
 750 {
 751         int i;
 752         enum lttng_error_code ret_code = LTTNG_OK;
 753         const int fd_count = lttng_dynamic_array_get_count(raw_fds);
 754
 755         for (i = 0; i < fd_count; i++) {
 756                 int ret;
 757                 struct fd_handle *handle;
 758                 int *raw_fd = (int *) lttng_dynamic_array_get_element(
 759                         raw_fds, i);
 760
 761                 assert(*raw_fd != -1);
 762
 763                 handle = fd_handle_create(*raw_fd);
 764                 if (!handle) {
 765                         ret_code = LTTNG_ERR_NOMEM;
 766                         goto end;
 767                 }
 768
 769                 /* FD ownership transferred to the handle. */
 770                 *raw_fd = -1;
 771
 772                 ret = lttng_payload_push_fd_handle(payload, handle);
 773                 fd_handle_put(handle);
 774                 if (ret) {
 775                         ret_code = LTTNG_ERR_NOMEM;
 776                         goto end;
 777                 }
 778         }
 779
 780 end:
 781         return ret_code;
 782 }
 783
 784 static
 785 ssize_t _lttcomm_recv_payload_fds_unix_sock(int sock, size_t nb_fd,
 786                 struct lttng_payload *payload, bool blocking)
 787 {
 788         int i = 0;
 789         enum lttng_error_code add_ret;
 790         ssize_t ret;
 791         int default_value = -1;
 792         struct lttng_dynamic_array raw_fds;
 793
 794         assert(sock);
 795         assert(payload);
 796         assert(nb_fd > 0);
 797
 798         lttng_dynamic_array_init(&raw_fds, sizeof(int), close_raw_fd);
 799
 800         for (i = 0; i < nb_fd; i++) {
 801                 if (lttng_dynamic_array_add_element(&raw_fds, &default_value)) {
 802                         ret = -LTTNG_ERR_NOMEM;
 803                         goto end;
 804                 }
 805         }
 806
 807         if (blocking) {
 808                 ret = lttcomm_recv_fds_unix_sock(
 809                                 sock, (int *) raw_fds.buffer.data, nb_fd);
 810         } else {
 811                 ret = lttcomm_recv_fds_unix_sock_non_block(
 812                                 sock, (int *) raw_fds.buffer.data, nb_fd);
 813         }
 814
 815         if (ret <= 0) {
 816                 goto end;
 817         }
 818
 819         add_ret = add_fds_to_payload(&raw_fds, payload);
 820         if (add_ret != LTTNG_OK) {
 821                 ret = - (int) add_ret;
 822                 goto end;
 823         }
 824
 825 end:
 826         lttng_dynamic_array_reset(&raw_fds);
 827         return ret;
 828 }
 829
 830 LTTNG_HIDDEN
 831 ssize_t lttcomm_recv_payload_fds_unix_sock(int sock, size_t nb_fd,
 832                            struct lttng_payload *payload)
 833 {
 834         return _lttcomm_recv_payload_fds_unix_sock(sock, nb_fd, payload, true);
 835 }
 836
 837 LTTNG_HIDDEN
 838 ssize_t lttcomm_recv_payload_fds_unix_sock_non_block(int sock, size_t nb_fd,
 839                            struct lttng_payload *payload)
 840 {
 841         return _lttcomm_recv_payload_fds_unix_sock(sock, nb_fd, payload, false);
 842 }
 843
 844 /*
 845  * Recv a message accompanied by fd(s) from a non-blocking unix socket.
 846  * Only use with non-blocking sockets.
 847  *
 848  * Returns the size of received data, or negative error value.
 849  *
 850  * Expect at most "nb_fd" file descriptors.
 851  *
 852  * Note that based on our comprehension, partial reception of fds is not
 853  * possible since the FDs are actually in the control message. It is all or
 854  * nothing, still the sender side can send the wrong number of fds.
 855  */
 856 LTTNG_HIDDEN
 857 ssize_t lttcomm_recv_fds_unix_sock_non_block(int sock, int *fds, size_t nb_fd)
 858 {
 859         struct iovec iov[1];
 860         ssize_t ret = 0;
 861         struct cmsghdr *cmsg;
 862         size_t sizeof_fds = nb_fd * sizeof(int);
 863
 864         assert(sock);
 865         assert(fds);
 866         assert(nb_fd > 0);
 867
 868 #ifdef __linux__
 869 /* Account for the struct ucred cmsg in the buffer size */
 870 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds) + CMSG_SPACE(sizeof(struct ucred))
 871 #else
 872 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds)
 873 #endif /* __linux__ */
 874
 875         char recv_buf[LTTNG_SOCK_RECV_FDS_BUF_SIZE];
 876         struct msghdr msg;
 877         char dummy;
 878
 879         memset(&msg, 0, sizeof(msg));
 880
 881         /* Prepare to receive the structures */
 882         iov[0].iov_base = &dummy;
 883         iov[0].iov_len = 1;
 884         msg.msg_iov = iov;
 885         msg.msg_iovlen = 1;
 886
 887         cmsg = (struct cmsghdr *) recv_buf;
 888         cmsg->cmsg_len = CMSG_LEN(sizeof_fds);
 889         cmsg->cmsg_level = SOL_SOCKET;
 890         cmsg->cmsg_type = SCM_RIGHTS;
 891
 892         msg.msg_control = cmsg;
 893         msg.msg_controllen = CMSG_LEN(sizeof(recv_buf));
 894         msg.msg_flags = 0;
 895
 896 retry:
 897         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 898         if (ret < 0) {
 899                 if (errno == EINTR) {
 900                         goto retry;
 901                 } else {
 902                         /*
 903                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 904                          */
 905                         if (errno == EAGAIN || errno == EWOULDBLOCK) {
 906                                 /*
 907                                  * This can happen in non blocking mode.
 908                                  * Nothing was recv.
 909                                  */
 910                                 ret = 0;
 911                                 goto end;
 912                         }
 913
 914                         if (errno == EPIPE) {
 915                                 /* Expected error, pass error to caller */
 916                                 DBG3("EPIPE on recvmsg");
 917                                 ret = -1;
 918                                 goto end;
 919                         }
 920
 921                         /* Unexpected error */
 922                         PERROR("recvmsg");
 923                         ret = -1;
 924                         goto end;
 925                 }
 926         }
 927
 928         if (ret != 1) {
 929                 fprintf(stderr, "Error: Received %zd bytes, expected %d\n",
 930                                 ret, 1);
 931                 goto end;
 932         }
 933
 934         if (msg.msg_flags & MSG_CTRUNC) {
 935                 fprintf(stderr, "Error: Control message truncated.\n");
 936                 ret = -1;
 937                 goto end;
 938         }
 939
 940         /*
 941          * If the socket was configured with SO_PASSCRED, the kernel will add a
 942          * control message (cmsg) to the ancillary data of the unix socket. We
 943          * need to expect a cmsg of the SCM_CREDENTIALS as the first control
 944          * message.
 945          */
 946         for (cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 947                 if (cmsg->cmsg_level != SOL_SOCKET) {
 948                         fprintf(stderr, "Error: The socket needs to be of type SOL_SOCKET\n");
 949                         ret = -1;
 950                         goto end;
 951                 }
 952                 if (cmsg->cmsg_type == SCM_RIGHTS) {
 953                         /*
 954                          * We found the controle message for file descriptors,
 955                          * now copy the fds to the fds ptr and return success.
 956                          */
 957                         if (cmsg->cmsg_len != CMSG_LEN(sizeof_fds)) {
 958                                 fprintf(stderr, "Error: Received %zu bytes of"
 959                                         "ancillary data for FDs, expected %zu\n",
 960                                         (size_t) cmsg->cmsg_len,
 961                                         (size_t) CMSG_LEN(sizeof_fds));
 962                                 ret = -1;
 963                                 goto end;
 964                         }
 965                         memcpy(fds, CMSG_DATA(cmsg), sizeof_fds);
 966                         ret = sizeof_fds;
 967                         goto end;
 968                 }
 969 #ifdef __linux__
 970                 if (cmsg->cmsg_type == SCM_CREDENTIALS) {
 971                         /*
 972                          * Expect credentials to be sent when expecting fds even
 973                          * if no credential were include in the send(). The
 974                          * kernel adds them...
 975                          */
 976                         ret = -1;
 977                 }
 978 #endif /* __linux__ */
 979         }
 980 end:
 981         return ret;
 982 }
 983
 984 /*
 985  * Send a message with credentials over a unix socket.
 986  *
 987  * Returns the size of data sent, or negative error value.
 988  */
 989 LTTNG_HIDDEN
 990 ssize_t lttcomm_send_creds_unix_sock(int sock, const void *buf, size_t len)
 991 {
 992         struct msghdr msg;
 993         struct iovec iov[1];
 994         ssize_t ret = -1;
 995 #if defined(__linux__) || defined(__CYGWIN__)
 996         struct cmsghdr *cmptr;
 997         size_t sizeof_cred = sizeof(lttng_sock_cred);
 998         char anc_buf[CMSG_SPACE(sizeof_cred)];
 999         lttng_sock_cred *creds;
1000
1001         memset(anc_buf, 0, CMSG_SPACE(sizeof_cred) * sizeof(char));
1002 #endif /* __linux__, __CYGWIN__ */
1003
1004         memset(&msg, 0, sizeof(msg));
1005
1006         assert(sock);
1007         assert(buf);
1008         assert(len > 0);
1009
1010         iov[0].iov_base = (void *) buf;
1011         iov[0].iov_len = len;
1012         msg.msg_iov = iov;
1013         msg.msg_iovlen = 1;
1014
1015 #if defined(__linux__) || defined(__CYGWIN__)
1016         msg.msg_control = (caddr_t) anc_buf;
1017         msg.msg_controllen = CMSG_LEN(sizeof_cred);
1018
1019         cmptr = CMSG_FIRSTHDR(&msg);
1020         if (!cmptr) {
1021                 return -1;
1022         }
1023         cmptr->cmsg_level = SOL_SOCKET;
1024         cmptr->cmsg_type = LTTNG_SOCK_CREDS;
1025         cmptr->cmsg_len = CMSG_LEN(sizeof_cred);
1026
1027         creds = (lttng_sock_cred*) CMSG_DATA(cmptr);
1028
1029         LTTNG_SOCK_SET_UID_CRED(creds, geteuid());
1030         LTTNG_SOCK_SET_GID_CRED(creds, getegid());
1031         LTTNG_SOCK_SET_PID_CRED(creds, getpid());
1032 #endif /* __linux__, __CYGWIN__ */
1033
1034         do {
1035                 ret = sendmsg(sock, &msg, 0);
1036         } while (ret < 0 && errno == EINTR);
1037         if (ret < 0) {
1038                 /*
1039                  * Only warn about EPIPE when quiet mode is deactivated.
1040                  * We consider EPIPE as expected.
1041                  */
1042                 if (errno != EPIPE || !lttng_opt_quiet) {
1043                         PERROR("sendmsg");
1044                 }
1045         }
1046         return ret;
1047 }
1048
1049 /*
1050  * Recv a message accompanied with credentials from a unix socket.
1051  *
1052  * Returns the size of received data, or negative error value.
1053  */
1054 LTTNG_HIDDEN
1055 ssize_t lttcomm_recv_creds_unix_sock(int sock, void *buf, size_t len,
1056                 lttng_sock_cred *creds)
1057 {
1058         struct msghdr msg;
1059         struct iovec iov[1];
1060         ssize_t ret;
1061         size_t len_last;
1062 #if defined(__linux__) || defined(__CYGWIN__)
1063         struct cmsghdr *cmptr;
1064         size_t sizeof_cred = sizeof(lttng_sock_cred);
1065         char anc_buf[CMSG_SPACE(sizeof_cred)];
1066 #endif  /* __linux__, __CYGWIN__ */
1067
1068         assert(sock);
1069         assert(buf);
1070         assert(len > 0);
1071         assert(creds);
1072
1073         memset(&msg, 0, sizeof(msg));
1074
1075         /* Prepare to receive the structures */
1076         iov[0].iov_base = buf;
1077         iov[0].iov_len = len;
1078         msg.msg_iov = iov;
1079         msg.msg_iovlen = 1;
1080
1081 #if defined(__linux__) || defined(__CYGWIN__)
1082         msg.msg_control = anc_buf;
1083         msg.msg_controllen = sizeof(anc_buf);
1084 #endif /* __linux__, __CYGWIN__ */
1085
1086         do {
1087                 len_last = iov[0].iov_len;
1088                 ret = recvmsg(sock, &msg, 0);
1089                 if (ret > 0) {
1090                         iov[0].iov_base += ret;
1091                         iov[0].iov_len -= ret;
1092                         assert(ret <= len_last);
1093                 }
1094         } while ((ret > 0 && ret < len_last) || (ret < 0 && errno == EINTR));
1095         if (ret < 0) {
1096                 PERROR("recvmsg fds");
1097                 goto end;
1098         } else if (ret > 0) {
1099                 ret = len;
1100         }
1101         /* Else ret = 0 meaning an orderly shutdown. */
1102
1103 #if defined(__linux__) || defined(__CYGWIN__)
1104         if (msg.msg_flags & MSG_CTRUNC) {
1105                 fprintf(stderr, "Error: Control message truncated.\n");
1106                 ret = -1;
1107                 goto end;
1108         }
1109
1110         cmptr = CMSG_FIRSTHDR(&msg);
1111         if (cmptr == NULL) {
1112                 fprintf(stderr, "Error: Invalid control message header\n");
1113                 ret = -1;
1114                 goto end;
1115         }
1116
1117         if (cmptr->cmsg_level != SOL_SOCKET ||
1118                         cmptr->cmsg_type != LTTNG_SOCK_CREDS) {
1119                 fprintf(stderr, "Didn't received any credentials\n");
1120                 ret = -1;
1121                 goto end;
1122         }
1123
1124         if (cmptr->cmsg_len != CMSG_LEN(sizeof_cred)) {
1125                 fprintf(stderr, "Error: Received %zu bytes of ancillary data, expected %zu\n",
1126                                 (size_t) cmptr->cmsg_len, (size_t) CMSG_LEN(sizeof_cred));
1127                 ret = -1;
1128                 goto end;
1129         }
1130
1131         memcpy(creds, CMSG_DATA(cmptr), sizeof_cred);
1132 #elif (defined(__FreeBSD__) || defined(__sun__) || defined(__APPLE__))
1133         if (lttng_get_unix_socket_peer_creds(sock, creds)) {
1134                 fprintf(stderr, "ARG\n");
1135                 ret = -1;
1136                 goto end;
1137         }
1138 #else
1139 #error "Please implement credential support for your OS."
1140 #endif  /* __linux__, __CYGWIN__ */
1141
1142 end:
1143         return ret;
1144 }
1145
1146 /*
1147  * Set socket option to use credentials passing.
1148  */
1149 #if defined(__linux__) || defined(__CYGWIN__)
1150 LTTNG_HIDDEN
1151 int lttcomm_setsockopt_creds_unix_sock(int sock)
1152 {
1153         int ret, on = 1;
1154
1155         /* Set socket for credentials retrieval */
1156         ret = setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &on, sizeof(on));
1157         if (ret < 0) {
1158                 PERROR("setsockopt creds unix sock");
1159         }
1160         return ret;
1161 }
1162 #elif (defined(__FreeBSD__) || defined(__sun__) || defined(__APPLE__))
1163 LTTNG_HIDDEN
1164 int lttcomm_setsockopt_creds_unix_sock(int sock)
1165 {
1166         return 0;
1167 }
1168 #else
1169 #error "Please implement credential support for your OS."
1170 #endif /* __linux__ */