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         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         if (fd_count < 0) {
 457                 ret = -LTTNG_ERR_INVALID;
 458                 goto end;
 459         }
 460
 461         /*
 462          * Prepare a contiguous array of file descriptors to send them.
 463          *
 464          * Note that the reference to each fd is released during the iteration;
 465          * we're just getting the numerical value of the fds to conform to the
 466          * syscall's interface. We rely on the fact that "view" must remain
 467          * valid for the duration of the call and that the underlying payload
 468          * owns a reference to the fd_handles.
 469          */
 470         for (i = 0; i < fd_count; i++) {
 471                 struct fd_handle *handle =
 472                                 lttng_payload_view_pop_fd_handle(view);
 473                 const int raw_fd = fd_handle_get_fd(handle);
 474                 const int add_ret = lttng_dynamic_array_add_element(
 475                                 &raw_fds, &raw_fd);
 476
 477                 fd_handle_put(handle);
 478                 if (add_ret) {
 479                         ret = -LTTNG_ERR_NOMEM;
 480                         goto end;
 481                 }
 482         }
 483
 484         if (blocking) {
 485                 ret = lttcomm_send_fds_unix_sock(sock,
 486                                 (const int *) raw_fds.buffer.data, fd_count);
 487         } else {
 488                 ret = lttcomm_send_fds_unix_sock_non_block(sock,
 489                                 (const int *) raw_fds.buffer.data, fd_count);
 490         }
 491
 492 end:
 493         lttng_dynamic_array_reset(&raw_fds);
 494         return ret;
 495 }
 496
 497 LTTNG_HIDDEN
 498 ssize_t lttcomm_send_payload_view_fds_unix_sock(int sock,
 499                 struct lttng_payload_view *view)
 500 {
 501         return _lttcomm_send_payload_view_fds_unix_sock(sock, view, true);
 502 }
 503
 504 LTTNG_HIDDEN
 505 ssize_t lttcomm_send_payload_view_fds_unix_sock_non_block(int sock,
 506                 struct lttng_payload_view *view)
 507 {
 508         return _lttcomm_send_payload_view_fds_unix_sock(sock, view, false);
 509 }
 510
 511 /*
 512  * Send a message accompanied by fd(s) over a unix socket.
 513  * Only use for non blocking socket.
 514  *
 515  * Returns the size of data sent, or negative error value.
 516  */
 517 LTTNG_HIDDEN
 518 ssize_t lttcomm_send_fds_unix_sock_non_block(int sock, const int *fds, size_t nb_fd)
 519 {
 520         struct msghdr msg;
 521         struct cmsghdr *cmptr;
 522         struct iovec iov[1];
 523         ssize_t ret = -1;
 524         unsigned int sizeof_fds = nb_fd * sizeof(int);
 525         char tmp[CMSG_SPACE(sizeof_fds)];
 526         char dummy = 0;
 527
 528         memset(&msg, 0, sizeof(msg));
 529         memset(tmp, 0, sizeof(tmp));
 530
 531         if (nb_fd > LTTCOMM_MAX_SEND_FDS)
 532                 return -EINVAL;
 533
 534         msg.msg_control = (caddr_t)tmp;
 535         msg.msg_controllen = CMSG_LEN(sizeof_fds);
 536
 537         cmptr = CMSG_FIRSTHDR(&msg);
 538         if (!cmptr) {
 539                 return -1;
 540         }
 541
 542         cmptr->cmsg_level = SOL_SOCKET;
 543         cmptr->cmsg_type = SCM_RIGHTS;
 544         cmptr->cmsg_len = CMSG_LEN(sizeof_fds);
 545         memcpy(CMSG_DATA(cmptr), fds, sizeof_fds);
 546         /* Sum of the length of all control messages in the buffer: */
 547         msg.msg_controllen = cmptr->cmsg_len;
 548
 549         iov[0].iov_base = &dummy;
 550         iov[0].iov_len = 1;
 551         msg.msg_iov = iov;
 552         msg.msg_iovlen = 1;
 553
 554 retry:
 555         ret = sendmsg(sock, &msg, 0);
 556         if (ret < 0) {
 557                 if (errno == EINTR) {
 558                         goto retry;
 559                 } else {
 560                         /*
 561                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 562                          */
 563                         if (errno == EAGAIN || errno == EWOULDBLOCK) {
 564                                 /*
 565                                  * This can happen in non blocking mode.
 566                                  * Nothing was sent.
 567                                  */
 568                                 ret = 0;
 569                                 goto end;
 570                         }
 571
 572                         if (errno == EPIPE) {
 573                                 /* Expected error, pass error to caller */
 574                                 DBG3("EPIPE on sendmsg");
 575                                 ret = -1;
 576                                 goto end;
 577                         }
 578
 579                         /* Unexpected error */
 580                         PERROR("sendmsg");
 581                         ret = -1;
 582                         goto end;
 583                 }
 584         }
 585
 586 end:
 587         return ret;
 588 }
 589
 590 /*
 591  * Recv a message accompanied by fd(s) from a unix socket.
 592  *
 593  * Returns the size of received data, or negative error value.
 594  *
 595  * Expect at most "nb_fd" file descriptors. Returns the number of fd
 596  * actually received in nb_fd.
 597  */
 598 LTTNG_HIDDEN
 599 ssize_t lttcomm_recv_fds_unix_sock(int sock, int *fds, size_t nb_fd)
 600 {
 601         struct iovec iov[1];
 602         ssize_t ret = 0;
 603         struct cmsghdr *cmsg;
 604         size_t sizeof_fds = nb_fd * sizeof(int);
 605
 606 #ifdef __linux__
 607 /* Account for the struct ucred cmsg in the buffer size */
 608 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds) + CMSG_SPACE(sizeof(struct ucred))
 609 #else
 610 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds)
 611 #endif /* __linux__ */
 612
 613         char recv_buf[LTTNG_SOCK_RECV_FDS_BUF_SIZE];
 614         struct msghdr msg;
 615         char dummy;
 616
 617         memset(&msg, 0, sizeof(msg));
 618
 619         /* Prepare to receive the structures */
 620         iov[0].iov_base = &dummy;
 621         iov[0].iov_len = 1;
 622         msg.msg_iov = iov;
 623         msg.msg_iovlen = 1;
 624
 625         cmsg = (struct cmsghdr *) recv_buf;
 626         cmsg->cmsg_len = CMSG_LEN(sizeof_fds);
 627         cmsg->cmsg_level = SOL_SOCKET;
 628         cmsg->cmsg_type = SCM_RIGHTS;
 629
 630         msg.msg_control = cmsg;
 631         msg.msg_controllen = CMSG_LEN(sizeof(recv_buf));
 632         msg.msg_flags = 0;
 633
 634 retry:
 635         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 636         if (ret < 0) {
 637                 if (errno == EINTR) {
 638                         goto retry;
 639                 } else {
 640                         /* We consider EPIPE and EAGAIN as expected. */
 641                         if (!lttng_opt_quiet &&
 642                                         (errno != EPIPE && errno != EAGAIN)) {
 643                                 PERROR("recvmsg");
 644                         }
 645                         goto end;
 646                 }
 647         }
 648
 649         if (ret != 1) {
 650                 fprintf(stderr, "Error: Received %zd bytes, expected %d\n",
 651                                 ret, 1);
 652                 goto end;
 653         }
 654
 655         if (msg.msg_flags & MSG_CTRUNC) {
 656                 fprintf(stderr, "Error: Control message truncated.\n");
 657                 ret = -1;
 658                 goto end;
 659         }
 660
 661         /*
 662          * If the socket was configured with SO_PASSCRED, the kernel will add a
 663          * control message (cmsg) to the ancillary data of the unix socket. We
 664          * need to expect a cmsg of the SCM_CREDENTIALS as the first control
 665          * message.
 666          */
 667         for (cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 668                 if (cmsg->cmsg_level != SOL_SOCKET) {
 669                         fprintf(stderr, "Error: The socket needs to be of type SOL_SOCKET\n");
 670                         ret = -1;
 671                         goto end;
 672                 }
 673                 if (cmsg->cmsg_type == SCM_RIGHTS) {
 674                         /*
 675                          * We found the controle message for file descriptors,
 676                          * now copy the fds to the fds ptr and return success.
 677                          */
 678                         if (cmsg->cmsg_len != CMSG_LEN(sizeof_fds)) {
 679                                 fprintf(stderr, "Error: Received %zu bytes of"
 680                                         "ancillary data for FDs, expected %zu\n",
 681                                         (size_t) cmsg->cmsg_len,
 682                                         (size_t) CMSG_LEN(sizeof_fds));
 683                                 ret = -1;
 684                                 goto end;
 685                         }
 686                         memcpy(fds, CMSG_DATA(cmsg), sizeof_fds);
 687                         ret = sizeof_fds;
 688                         goto end;
 689                 }
 690 #ifdef __linux__
 691                 if (cmsg->cmsg_type == SCM_CREDENTIALS) {
 692                         /*
 693                          * Expect credentials to be sent when expecting fds even
 694                          * if no credential were include in the send(). The
 695                          * kernel adds them...
 696                          */
 697                         ret = -1;
 698                 }
 699 #endif /* __linux__ */
 700         }
 701 end:
 702         return ret;
 703 }
 704
 705 static
 706 void close_raw_fd(void *ptr)
 707 {
 708         const int raw_fd = *((const int *) ptr);
 709
 710         if (raw_fd >= 0) {
 711                 const int ret = close(raw_fd);
 712
 713                 if (ret) {
 714                         PERROR("Failed to close file descriptor %d", raw_fd);
 715                 }
 716         }
 717 }
 718
 719 static
 720 enum lttng_error_code add_fds_to_payload(struct lttng_dynamic_array *raw_fds,
 721                 struct lttng_payload *payload)
 722 {
 723         int i;
 724         enum lttng_error_code ret_code = LTTNG_OK;
 725         const int fd_count = lttng_dynamic_array_get_count(raw_fds);
 726
 727         for (i = 0; i < fd_count; i++) {
 728                 int ret;
 729                 struct fd_handle *handle;
 730                 int *raw_fd = (int *) lttng_dynamic_array_get_element(
 731                         raw_fds, i);
 732
 733                 handle = fd_handle_create(*raw_fd);
 734                 if (!handle) {
 735                         ret_code = LTTNG_ERR_NOMEM;
 736                         goto end;
 737                 }
 738
 739                 /* FD ownership transferred to the handle. */
 740                 *raw_fd = -1;
 741
 742                 ret = lttng_payload_push_fd_handle(payload, handle);
 743                 fd_handle_put(handle);
 744                 if (ret) {
 745                         ret_code = LTTNG_ERR_NOMEM;
 746                         goto end;
 747                 }
 748         }
 749
 750 end:
 751         return ret_code;
 752 }
 753
 754 static
 755 ssize_t _lttcomm_recv_payload_fds_unix_sock(int sock, size_t nb_fd,
 756                 struct lttng_payload *payload, bool blocking)
 757 {
 758         enum lttng_error_code add_ret;
 759         ssize_t ret;
 760         struct lttng_dynamic_array raw_fds;
 761
 762         lttng_dynamic_array_init(&raw_fds, sizeof(int), close_raw_fd);
 763         ret = lttng_dynamic_array_set_count(&raw_fds, nb_fd);
 764         if (ret) {
 765                 ret = -LTTNG_ERR_NOMEM;
 766                 goto end;
 767         }
 768
 769         if (blocking) {
 770                 ret = lttcomm_recv_fds_unix_sock(
 771                                 sock, (int *) raw_fds.buffer.data, nb_fd);
 772         } else {
 773                 ret = lttcomm_recv_fds_unix_sock_non_block(
 774                                 sock, (int *) raw_fds.buffer.data, nb_fd);
 775         }
 776
 777         if (ret < 0) {
 778                 goto end;
 779         }
 780
 781         add_ret = add_fds_to_payload(&raw_fds, payload);
 782         if (add_ret != LTTNG_OK) {
 783                 ret = - (int) add_ret;
 784                 goto end;
 785         }
 786
 787 end:
 788         lttng_dynamic_array_reset(&raw_fds);
 789         return ret;
 790 }
 791
 792 LTTNG_HIDDEN
 793 ssize_t lttcomm_recv_payload_fds_unix_sock(int sock, size_t nb_fd,
 794                            struct lttng_payload *payload)
 795 {
 796         return _lttcomm_recv_payload_fds_unix_sock(sock, nb_fd, payload, true);
 797 }
 798
 799 LTTNG_HIDDEN
 800 ssize_t lttcomm_recv_payload_fds_unix_sock_non_block(int sock, size_t nb_fd,
 801                            struct lttng_payload *payload)
 802 {
 803         return _lttcomm_recv_payload_fds_unix_sock(sock, nb_fd, payload, false);
 804 }
 805
 806 /*
 807  * Recv a message accompanied by fd(s) from a non-blocking unix socket.
 808  * Only use with non-blocking sockets.
 809  *
 810  * Returns the size of received data, or negative error value.
 811  *
 812  * Expect at most "nb_fd" file descriptors.
 813  *
 814  * Note that based on our comprehension, partial reception of fds is not
 815  * possible since the FDs are actually in the control message. It is all or
 816  * nothing, still the sender side can send the wrong number of fds.
 817  */
 818 LTTNG_HIDDEN
 819 ssize_t lttcomm_recv_fds_unix_sock_non_block(int sock, int *fds, size_t nb_fd)
 820 {
 821         struct iovec iov[1];
 822         ssize_t ret = 0;
 823         struct cmsghdr *cmsg;
 824         size_t sizeof_fds = nb_fd * sizeof(int);
 825
 826 #ifdef __linux__
 827 /* Account for the struct ucred cmsg in the buffer size */
 828 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds) + CMSG_SPACE(sizeof(struct ucred))
 829 #else
 830 #define LTTNG_SOCK_RECV_FDS_BUF_SIZE CMSG_SPACE(sizeof_fds)
 831 #endif /* __linux__ */
 832
 833         char recv_buf[LTTNG_SOCK_RECV_FDS_BUF_SIZE];
 834         struct msghdr msg;
 835         char dummy;
 836
 837         memset(&msg, 0, sizeof(msg));
 838
 839         /* Prepare to receive the structures */
 840         iov[0].iov_base = &dummy;
 841         iov[0].iov_len = 1;
 842         msg.msg_iov = iov;
 843         msg.msg_iovlen = 1;
 844
 845         cmsg = (struct cmsghdr *) recv_buf;
 846         cmsg->cmsg_len = CMSG_LEN(sizeof_fds);
 847         cmsg->cmsg_level = SOL_SOCKET;
 848         cmsg->cmsg_type = SCM_RIGHTS;
 849
 850         msg.msg_control = cmsg;
 851         msg.msg_controllen = CMSG_LEN(sizeof(recv_buf));
 852         msg.msg_flags = 0;
 853
 854 retry:
 855         ret = lttng_recvmsg_nosigpipe(sock, &msg);
 856         if (ret < 0) {
 857                 if (errno == EINTR) {
 858                         goto retry;
 859                 } else {
 860                         /*
 861                          * We consider EPIPE and EAGAIN/EWOULDBLOCK as expected.
 862                          */
 863                         if (errno == EAGAIN || errno == EWOULDBLOCK) {
 864                                 /*
 865                                  * This can happen in non blocking mode.
 866                                  * Nothing was recv.
 867                                  */
 868                                 ret = 0;
 869                                 goto end;
 870                         }
 871
 872                         if (errno == EPIPE) {
 873                                 /* Expected error, pass error to caller */
 874                                 DBG3("EPIPE on recvmsg");
 875                                 ret = -1;
 876                                 goto end;
 877                         }
 878
 879                         /* Unexpected error */
 880                         PERROR("recvmsg");
 881                         ret = -1;
 882                         goto end;
 883                 }
 884         }
 885
 886         if (ret != 1) {
 887                 fprintf(stderr, "Error: Received %zd bytes, expected %d\n",
 888                                 ret, 1);
 889                 goto end;
 890         }
 891
 892         if (msg.msg_flags & MSG_CTRUNC) {
 893                 fprintf(stderr, "Error: Control message truncated.\n");
 894                 ret = -1;
 895                 goto end;
 896         }
 897
 898         /*
 899          * If the socket was configured with SO_PASSCRED, the kernel will add a
 900          * control message (cmsg) to the ancillary data of the unix socket. We
 901          * need to expect a cmsg of the SCM_CREDENTIALS as the first control
 902          * message.
 903          */
 904         for (cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 905                 if (cmsg->cmsg_level != SOL_SOCKET) {
 906                         fprintf(stderr, "Error: The socket needs to be of type SOL_SOCKET\n");
 907                         ret = -1;
 908                         goto end;
 909                 }
 910                 if (cmsg->cmsg_type == SCM_RIGHTS) {
 911                         /*
 912                          * We found the controle message for file descriptors,
 913                          * now copy the fds to the fds ptr and return success.
 914                          */
 915                         if (cmsg->cmsg_len != CMSG_LEN(sizeof_fds)) {
 916                                 fprintf(stderr, "Error: Received %zu bytes of"
 917                                         "ancillary data for FDs, expected %zu\n",
 918                                         (size_t) cmsg->cmsg_len,
 919                                         (size_t) CMSG_LEN(sizeof_fds));
 920                                 ret = -1;
 921                                 goto end;
 922                         }
 923                         memcpy(fds, CMSG_DATA(cmsg), sizeof_fds);
 924                         ret = sizeof_fds;
 925                         goto end;
 926                 }
 927 #ifdef __linux__
 928                 if (cmsg->cmsg_type == SCM_CREDENTIALS) {
 929                         /*
 930                          * Expect credentials to be sent when expecting fds even
 931                          * if no credential were include in the send(). The
 932                          * kernel adds them...
 933                          */
 934                         ret = -1;
 935                 }
 936 #endif /* __linux__ */
 937         }
 938 end:
 939         return ret;
 940 }
 941
 942 /*
 943  * Send a message with credentials over a unix socket.
 944  *
 945  * Returns the size of data sent, or negative error value.
 946  */
 947 LTTNG_HIDDEN
 948 ssize_t lttcomm_send_creds_unix_sock(int sock, const void *buf, size_t len)
 949 {
 950         struct msghdr msg;
 951         struct iovec iov[1];
 952         ssize_t ret = -1;
 953 #ifdef __linux__
 954         struct cmsghdr *cmptr;
 955         size_t sizeof_cred = sizeof(lttng_sock_cred);
 956         char anc_buf[CMSG_SPACE(sizeof_cred)];
 957         lttng_sock_cred *creds;
 958
 959         memset(anc_buf, 0, CMSG_SPACE(sizeof_cred) * sizeof(char));
 960 #endif /* __linux__ */
 961
 962         memset(&msg, 0, sizeof(msg));
 963
 964         iov[0].iov_base = (void *) buf;
 965         iov[0].iov_len = len;
 966         msg.msg_iov = iov;
 967         msg.msg_iovlen = 1;
 968
 969 #ifdef __linux__
 970         msg.msg_control = (caddr_t) anc_buf;
 971         msg.msg_controllen = CMSG_LEN(sizeof_cred);
 972
 973         cmptr = CMSG_FIRSTHDR(&msg);
 974         if (!cmptr) {
 975                 return -1;
 976         }
 977         cmptr->cmsg_level = SOL_SOCKET;
 978         cmptr->cmsg_type = LTTNG_SOCK_CREDS;
 979         cmptr->cmsg_len = CMSG_LEN(sizeof_cred);
 980
 981         creds = (lttng_sock_cred*) CMSG_DATA(cmptr);
 982
 983         LTTNG_SOCK_SET_UID_CRED(creds, geteuid());
 984         LTTNG_SOCK_SET_GID_CRED(creds, getegid());
 985         LTTNG_SOCK_SET_PID_CRED(creds, getpid());
 986 #endif /* __linux__ */
 987
 988         do {
 989                 ret = sendmsg(sock, &msg, 0);
 990         } while (ret < 0 && errno == EINTR);
 991         if (ret < 0) {
 992                 /*
 993                  * Only warn about EPIPE when quiet mode is deactivated.
 994                  * We consider EPIPE as expected.
 995                  */
 996                 if (errno != EPIPE || !lttng_opt_quiet) {
 997                         PERROR("sendmsg");
 998                 }
 999         }
1000         return ret;
1001 }
1002
1003 /*
1004  * Recv a message accompanied with credentials from a unix socket.
1005  *
1006  * Returns the size of received data, or negative error value.
1007  */
1008 LTTNG_HIDDEN
1009 ssize_t lttcomm_recv_creds_unix_sock(int sock, void *buf, size_t len,
1010                 lttng_sock_cred *creds)
1011 {
1012         struct msghdr msg;
1013         struct iovec iov[1];
1014         ssize_t ret;
1015         size_t len_last;
1016 #ifdef __linux__
1017         struct cmsghdr *cmptr;
1018         size_t sizeof_cred = sizeof(lttng_sock_cred);
1019         char anc_buf[CMSG_SPACE(sizeof_cred)];
1020 #endif  /* __linux__ */
1021
1022         memset(&msg, 0, sizeof(msg));
1023
1024         /* Not allowed */
1025         if (creds == NULL) {
1026                 ret = -1;
1027                 goto end;
1028         }
1029
1030         /* Prepare to receive the structures */
1031         iov[0].iov_base = buf;
1032         iov[0].iov_len = len;
1033         msg.msg_iov = iov;
1034         msg.msg_iovlen = 1;
1035
1036 #ifdef __linux__
1037         msg.msg_control = anc_buf;
1038         msg.msg_controllen = sizeof(anc_buf);
1039 #endif /* __linux__ */
1040
1041         do {
1042                 len_last = iov[0].iov_len;
1043                 ret = recvmsg(sock, &msg, 0);
1044                 if (ret > 0) {
1045                         iov[0].iov_base += ret;
1046                         iov[0].iov_len -= ret;
1047                         assert(ret <= len_last);
1048                 }
1049         } while ((ret > 0 && ret < len_last) || (ret < 0 && errno == EINTR));
1050         if (ret < 0) {
1051                 PERROR("recvmsg fds");
1052                 goto end;
1053         } else if (ret > 0) {
1054                 ret = len;
1055         }
1056         /* Else ret = 0 meaning an orderly shutdown. */
1057
1058 #ifdef __linux__
1059         if (msg.msg_flags & MSG_CTRUNC) {
1060                 fprintf(stderr, "Error: Control message truncated.\n");
1061                 ret = -1;
1062                 goto end;
1063         }
1064
1065         cmptr = CMSG_FIRSTHDR(&msg);
1066         if (cmptr == NULL) {
1067                 fprintf(stderr, "Error: Invalid control message header\n");
1068                 ret = -1;
1069                 goto end;
1070         }
1071
1072         if (cmptr->cmsg_level != SOL_SOCKET ||
1073                         cmptr->cmsg_type != LTTNG_SOCK_CREDS) {
1074                 fprintf(stderr, "Didn't received any credentials\n");
1075                 ret = -1;
1076                 goto end;
1077         }
1078
1079         if (cmptr->cmsg_len != CMSG_LEN(sizeof_cred)) {
1080                 fprintf(stderr, "Error: Received %zu bytes of ancillary data, expected %zu\n",
1081                                 (size_t) cmptr->cmsg_len, (size_t) CMSG_LEN(sizeof_cred));
1082                 ret = -1;
1083                 goto end;
1084         }
1085
1086         memcpy(creds, CMSG_DATA(cmptr), sizeof_cred);
1087 #elif (defined(__FreeBSD__) || defined(__CYGWIN__) || defined(__sun__) || defined(__APPLE__))
1088         {
1089                 int peer_ret;
1090
1091                 peer_ret = getpeereid(sock, &creds->uid, &creds->gid);
1092                 if (peer_ret != 0) {
1093                         return peer_ret;
1094                 }
1095         }
1096 #else
1097 #error "Please implement credential support for your OS."
1098 #endif  /* __linux__ */
1099
1100 end:
1101         return ret;
1102 }
1103
1104 /*
1105  * Set socket option to use credentials passing.
1106  */
1107 #ifdef __linux__
1108 LTTNG_HIDDEN
1109 int lttcomm_setsockopt_creds_unix_sock(int sock)
1110 {
1111         int ret, on = 1;
1112
1113         /* Set socket for credentials retrieval */
1114         ret = setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &on, sizeof(on));
1115         if (ret < 0) {
1116                 PERROR("setsockopt creds unix sock");
1117         }
1118         return ret;
1119 }
1120 #elif (defined(__FreeBSD__) || defined(__CYGWIN__) || defined(__sun__) || defined(__APPLE__))
1121 LTTNG_HIDDEN
1122 int lttcomm_setsockopt_creds_unix_sock(int sock)
1123 {
1124         return 0;
1125 }
1126 #else
1127 #error "Please implement credential support for your OS."
1128 #endif /* __linux__ */