src/common/runas.c

   1 /*
   2  * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
   3  * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  * Copyright (C) 2019 Jérémie Galarneau <jeremie.galarneau@efficios.com>
   5  *
   6  * SPDX-License-Identifier: GPL-2.0-only
   7  *
   8  */
   9
  10 #define _LGPL_SOURCE
  11 #include <assert.h>
  12 #include <fcntl.h>
  13 #include <grp.h>
  14 #include <limits.h>
  15 #include <pwd.h>
  16 #include <sched.h>
  17 #include <signal.h>
  18 #include <stdio.h>
  19 #include <stdlib.h>
  20 #include <string.h>
  21 #include <sys/stat.h>
  22 #include <sys/types.h>
  23 #include <sys/wait.h>
  24 #include <unistd.h>
  25
  26 #include <common/lttng-kernel.h>
  27 #include <common/common.h>
  28 #include <common/utils.h>
  29 #include <common/compat/getenv.h>
  30 #include <common/compat/prctl.h>
  31 #include <common/unix.h>
  32 #include <common/defaults.h>
  33 #include <common/lttng-elf.h>
  34
  35 #include <lttng/constant.h>
  36
  37 #include "runas.h"
  38
  39 #define GETPW_BUFFER_FALLBACK_SIZE 4096
  40
  41 struct run_as_data;
  42 struct run_as_ret;
  43 typedef int (*run_as_fct)(struct run_as_data *data, struct run_as_ret *ret_value);
  44
  45 enum run_as_cmd {
  46         RUN_AS_MKDIR,
  47         RUN_AS_MKDIRAT,
  48         RUN_AS_MKDIR_RECURSIVE,
  49         RUN_AS_MKDIRAT_RECURSIVE,
  50         RUN_AS_OPEN,
  51         RUN_AS_OPENAT,
  52         RUN_AS_UNLINK,
  53         RUN_AS_UNLINKAT,
  54         RUN_AS_RMDIR,
  55         RUN_AS_RMDIRAT,
  56         RUN_AS_RMDIR_RECURSIVE,
  57         RUN_AS_RMDIRAT_RECURSIVE,
  58         RUN_AS_RENAME,
  59         RUN_AS_RENAMEAT,
  60         RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET,
  61         RUN_AS_EXTRACT_SDT_PROBE_OFFSETS,
  62 };
  63
  64 struct run_as_mkdir_data {
  65         int dirfd;
  66         char path[LTTNG_PATH_MAX];
  67         mode_t mode;
  68 } LTTNG_PACKED;
  69
  70 struct run_as_open_data {
  71         int dirfd;
  72         char path[LTTNG_PATH_MAX];
  73         int flags;
  74         mode_t mode;
  75 } LTTNG_PACKED;
  76
  77 struct run_as_unlink_data {
  78         int dirfd;
  79         char path[LTTNG_PATH_MAX];
  80 } LTTNG_PACKED;
  81
  82 struct run_as_rmdir_data {
  83         int dirfd;
  84         char path[LTTNG_PATH_MAX];
  85         int flags; /* enum lttng_directory_handle_rmdir_recursive_flags */
  86 } LTTNG_PACKED;
  87
  88 struct run_as_extract_elf_symbol_offset_data {
  89         int fd;
  90         char function[LTTNG_SYMBOL_NAME_LEN];
  91 } LTTNG_PACKED;
  92
  93 struct run_as_extract_sdt_probe_offsets_data {
  94         int fd;
  95         char probe_name[LTTNG_SYMBOL_NAME_LEN];
  96         char provider_name[LTTNG_SYMBOL_NAME_LEN];
  97 } LTTNG_PACKED;
  98
  99 struct run_as_rename_data {
 100         /*
 101          * [0] = old_dirfd
 102          * [1] = new_dirfd
 103          */
 104         int dirfds[2];
 105         char old_path[LTTNG_PATH_MAX];
 106         char new_path[LTTNG_PATH_MAX];
 107 } LTTNG_PACKED;
 108
 109 struct run_as_open_ret {
 110         int fd;
 111 } LTTNG_PACKED;
 112
 113 struct run_as_extract_elf_symbol_offset_ret {
 114         uint64_t offset;
 115 } LTTNG_PACKED;
 116
 117 struct run_as_extract_sdt_probe_offsets_ret {
 118         uint32_t num_offset;
 119         uint64_t offsets[LTTNG_KERNEL_MAX_UPROBE_NUM];
 120 } LTTNG_PACKED;
 121
 122 struct run_as_data {
 123         enum run_as_cmd cmd;
 124         union {
 125                 struct run_as_mkdir_data mkdir;
 126                 struct run_as_open_data open;
 127                 struct run_as_unlink_data unlink;
 128                 struct run_as_rmdir_data rmdir;
 129                 struct run_as_rename_data rename;
 130                 struct run_as_extract_elf_symbol_offset_data extract_elf_symbol_offset;
 131                 struct run_as_extract_sdt_probe_offsets_data extract_sdt_probe_offsets;
 132         } u;
 133         uid_t uid;
 134         gid_t gid;
 135 } LTTNG_PACKED;
 136
 137 /*
 138  * The run_as_ret structure holds the returned value and status of the command.
 139  *
 140  * The `u` union field holds the return value of the command; in most cases it
 141  * represents the success or the failure of the command. In more complex
 142  * commands, it holds a computed value.
 143  *
 144  * The _errno field is the errno recorded after the execution of the command.
 145  *
 146  * The _error fields is used the signify that return status of the command. For
 147  * simple commands returning `int` the _error field will be the same as the
 148  * ret_int field. In complex commands, it signify the success or failure of the
 149  * command.
 150  *
 151  */
 152 struct run_as_ret {
 153         union {
 154                 int ret;
 155                 struct run_as_open_ret open;
 156                 struct run_as_extract_elf_symbol_offset_ret extract_elf_symbol_offset;
 157                 struct run_as_extract_sdt_probe_offsets_ret extract_sdt_probe_offsets;
 158         } u;
 159         int _errno;
 160         bool _error;
 161 } LTTNG_PACKED;
 162
 163 #define COMMAND_IN_FDS(data_ptr) ({                                     \
 164         int *fds = NULL;                                                \
 165         if (command_properties[data_ptr->cmd].in_fds_offset != -1) {    \
 166                 fds = (int *) ((char *) data_ptr + command_properties[data_ptr->cmd].in_fds_offset); \
 167         }                                                               \
 168         fds;                                                            \
 169 })
 170
 171 #define COMMAND_OUT_FDS(cmd, ret_ptr) ({                                \
 172         int *fds = NULL;                                                \
 173         if (command_properties[cmd].out_fds_offset != -1) {             \
 174                 fds = (int *) ((char *) ret_ptr + command_properties[cmd].out_fds_offset); \
 175         }                                                               \
 176         fds;                                                            \
 177 })
 178
 179 #define COMMAND_IN_FD_COUNT(data_ptr) ({                \
 180         command_properties[data_ptr->cmd].in_fd_count;  \
 181 })
 182
 183 #define COMMAND_OUT_FD_COUNT(cmd) ({            \
 184         command_properties[cmd].out_fd_count;   \
 185 })
 186
 187 #define COMMAND_USE_CWD_FD(data_ptr) command_properties[data_ptr->cmd].use_cwd_fd
 188
 189 struct run_as_command_properties {
 190         /* Set to -1 when not applicable. */
 191         ptrdiff_t in_fds_offset, out_fds_offset;
 192         unsigned int in_fd_count, out_fd_count;
 193         bool use_cwd_fd;
 194 };
 195
 196 static const struct run_as_command_properties command_properties[] = {
 197         [RUN_AS_MKDIR] = {
 198                 .in_fds_offset = offsetof(struct run_as_data, u.mkdir.dirfd),
 199                 .in_fd_count = 1,
 200                 .out_fds_offset = -1,
 201                 .out_fd_count = 0,
 202                 .use_cwd_fd = true,
 203         },
 204         [RUN_AS_MKDIRAT] = {
 205                 .in_fds_offset = offsetof(struct run_as_data, u.mkdir.dirfd),
 206                 .in_fd_count = 1,
 207                 .out_fds_offset = -1,
 208                 .out_fd_count = 0,
 209                 .use_cwd_fd = false,
 210         },
 211         [RUN_AS_MKDIR_RECURSIVE] = {
 212                 .in_fds_offset = offsetof(struct run_as_data, u.mkdir.dirfd),
 213                 .in_fd_count = 1,
 214                 .out_fds_offset = -1,
 215                 .out_fd_count = 0,
 216                 .use_cwd_fd = true,
 217         },
 218         [RUN_AS_MKDIRAT_RECURSIVE] = {
 219                 .in_fds_offset = offsetof(struct run_as_data, u.mkdir.dirfd),
 220                 .in_fd_count = 1,
 221                 .out_fds_offset = -1,
 222                 .out_fd_count = 0,
 223                 .use_cwd_fd = false,
 224         },
 225         [RUN_AS_OPEN] = {
 226                 .in_fds_offset = offsetof(struct run_as_data, u.open.dirfd),
 227                 .in_fd_count = 1,
 228                 .out_fds_offset = offsetof(struct run_as_ret, u.open.fd),
 229                 .out_fd_count = 1,
 230                 .use_cwd_fd = true,
 231         },
 232         [RUN_AS_OPENAT] = {
 233                 .in_fds_offset = offsetof(struct run_as_data, u.open.dirfd),
 234                 .in_fd_count = 1,
 235                 .out_fds_offset = offsetof(struct run_as_ret, u.open.fd),
 236                 .out_fd_count = 1,
 237                 .use_cwd_fd = false,
 238         },
 239         [RUN_AS_UNLINK] = {
 240                 .in_fds_offset = offsetof(struct run_as_data, u.unlink.dirfd),
 241                 .in_fd_count = 1,
 242                 .out_fds_offset = -1,
 243                 .out_fd_count = 0,
 244                 .use_cwd_fd = true,
 245         },
 246         [RUN_AS_UNLINKAT] = {
 247                 .in_fds_offset = offsetof(struct run_as_data, u.unlink.dirfd),
 248                 .in_fd_count = 1,
 249                 .out_fds_offset = -1,
 250                 .out_fd_count = 0,
 251                 .use_cwd_fd = false,
 252         },
 253         [RUN_AS_RMDIR_RECURSIVE] = {
 254                 .in_fds_offset = offsetof(struct run_as_data, u.rmdir.dirfd),
 255                 .in_fd_count = 1,
 256                 .out_fds_offset = -1,
 257                 .out_fd_count = 0,
 258                 .use_cwd_fd = true,
 259         },
 260         [RUN_AS_RMDIRAT_RECURSIVE] = {
 261                 .in_fds_offset = offsetof(struct run_as_data, u.rmdir.dirfd),
 262                 .in_fd_count = 1,
 263                 .out_fds_offset = -1,
 264                 .out_fd_count = 0,
 265                 .use_cwd_fd = false,
 266         },
 267         [RUN_AS_RMDIR] = {
 268                 .in_fds_offset = offsetof(struct run_as_data, u.rmdir.dirfd),
 269                 .in_fd_count = 1,
 270                 .out_fds_offset = -1,
 271                 .out_fd_count = 0,
 272                 .use_cwd_fd = true,
 273         },
 274         [RUN_AS_RMDIRAT] = {
 275                 .in_fds_offset = offsetof(struct run_as_data, u.rmdir.dirfd),
 276                 .in_fd_count = 1,
 277                 .out_fds_offset = -1,
 278                 .out_fd_count = 0,
 279                 .use_cwd_fd = false,
 280         },
 281         [RUN_AS_RENAME] = {
 282                 .in_fds_offset = offsetof(struct run_as_data, u.rename.dirfds),
 283                 .in_fd_count = 2,
 284                 .out_fds_offset = -1,
 285                 .out_fd_count = 0,
 286                 .use_cwd_fd = true,
 287         },
 288         [RUN_AS_RENAMEAT] = {
 289                 .in_fds_offset = offsetof(struct run_as_data, u.rename.dirfds),
 290                 .in_fd_count = 2,
 291                 .out_fds_offset = -1,
 292                 .out_fd_count = 0,
 293                 .use_cwd_fd = false,
 294         },
 295         [RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET] = {
 296                 .in_fds_offset = offsetof(struct run_as_data,
 297                                 u.extract_elf_symbol_offset.fd),
 298                 .in_fd_count = 1,
 299                 .out_fds_offset = -1,
 300                 .out_fd_count = 0,
 301                 .use_cwd_fd = false,
 302         },
 303         [RUN_AS_EXTRACT_SDT_PROBE_OFFSETS] = {
 304                 .in_fds_offset = offsetof(struct run_as_data,
 305                                 u.extract_sdt_probe_offsets.fd),
 306                 .in_fd_count = 1,
 307                 .out_fds_offset = -1,
 308                 .out_fd_count = 0,
 309                 .use_cwd_fd = false,
 310         },
 311 };
 312
 313 struct run_as_worker {
 314         pid_t pid;      /* Worker PID. */
 315         int sockpair[2];
 316         char *procname;
 317 };
 318
 319 /* Single global worker per process (for now). */
 320 static struct run_as_worker *global_worker;
 321 /* Lock protecting the worker. */
 322 static pthread_mutex_t worker_lock = PTHREAD_MUTEX_INITIALIZER;
 323
 324 #ifdef VALGRIND
 325 static
 326 int use_clone(void)
 327 {
 328         return 0;
 329 }
 330 #else
 331 static
 332 int use_clone(void)
 333 {
 334         return !lttng_secure_getenv("LTTNG_DEBUG_NOCLONE");
 335 }
 336 #endif
 337
 338 /*
 339  * Create recursively directory using the FULL path.
 340  */
 341 static
 342 int _mkdirat_recursive(struct run_as_data *data, struct run_as_ret *ret_value)
 343 {
 344         const char *path;
 345         mode_t mode;
 346         struct lttng_directory_handle *handle;
 347
 348         path = data->u.mkdir.path;
 349         mode = data->u.mkdir.mode;
 350
 351         handle = lttng_directory_handle_create_from_dirfd(data->u.mkdir.dirfd);
 352         if (!handle) {
 353                 ret_value->_errno = errno;
 354                 ret_value->_error = true;
 355                 ret_value->u.ret = -1;
 356                 goto end;
 357         }
 358         /* Ownership of dirfd is transferred to the handle. */
 359         data->u.mkdir.dirfd = -1;
 360         /* Safe to call as we have transitioned to the requested uid/gid. */
 361         ret_value->u.ret = lttng_directory_handle_create_subdirectory_recursive(
 362                         handle, path, mode);
 363         ret_value->_errno = errno;
 364         ret_value->_error = (ret_value->u.ret) ? true : false;
 365         lttng_directory_handle_put(handle);
 366 end:
 367         return ret_value->u.ret;
 368 }
 369
 370 static
 371 int _mkdirat(struct run_as_data *data, struct run_as_ret *ret_value)
 372 {
 373         const char *path;
 374         mode_t mode;
 375         struct lttng_directory_handle *handle;
 376
 377         path = data->u.mkdir.path;
 378         mode = data->u.mkdir.mode;
 379
 380         handle = lttng_directory_handle_create_from_dirfd(data->u.mkdir.dirfd);
 381         if (!handle) {
 382                 ret_value->u.ret = -1;
 383                 ret_value->_errno = errno;
 384                 ret_value->_error = true;
 385                 goto end;
 386         }
 387         /* Ownership of dirfd is transferred to the handle. */
 388         data->u.mkdir.dirfd = -1;
 389         /* Safe to call as we have transitioned to the requested uid/gid. */
 390         ret_value->u.ret = lttng_directory_handle_create_subdirectory(
 391                         handle, path, mode);
 392         ret_value->_errno = errno;
 393         ret_value->_error = (ret_value->u.ret) ? true : false;
 394         lttng_directory_handle_put(handle);
 395 end:
 396         return ret_value->u.ret;
 397 }
 398
 399 static
 400 int _open(struct run_as_data *data, struct run_as_ret *ret_value)
 401 {
 402         int fd;
 403         struct lttng_directory_handle *handle;
 404
 405         handle = lttng_directory_handle_create_from_dirfd(data->u.open.dirfd);
 406         if (!handle) {
 407                 ret_value->_errno = errno;
 408                 ret_value->_error = true;
 409                 ret_value->u.ret = -1;
 410                 goto end;
 411         }
 412         /* Ownership of dirfd is transferred to the handle. */
 413         data->u.open.dirfd = -1;
 414
 415         fd = lttng_directory_handle_open_file(handle,
 416                         data->u.open.path, data->u.open.flags,
 417                         data->u.open.mode);
 418         if (fd < 0) {
 419                 ret_value->u.ret = -1;
 420                 ret_value->u.open.fd = -1;
 421         } else {
 422                 ret_value->u.ret = 0;
 423                 ret_value->u.open.fd = fd;
 424         }
 425
 426         ret_value->_errno = errno;
 427         ret_value->_error = fd < 0;
 428         lttng_directory_handle_put(handle);
 429 end:
 430         return ret_value->u.ret;
 431 }
 432
 433 static
 434 int _unlink(struct run_as_data *data, struct run_as_ret *ret_value)
 435 {
 436         struct lttng_directory_handle *handle;
 437
 438         handle = lttng_directory_handle_create_from_dirfd(data->u.unlink.dirfd);
 439         if (!handle) {
 440                 ret_value->u.ret = -1;
 441                 ret_value->_errno = errno;
 442                 ret_value->_error = true;
 443                 goto end;
 444         }
 445
 446         /* Ownership of dirfd is transferred to the handle. */
 447         data->u.unlink.dirfd = -1;
 448
 449         ret_value->u.ret = lttng_directory_handle_unlink_file(handle,
 450                         data->u.unlink.path);
 451         ret_value->_errno = errno;
 452         ret_value->_error = (ret_value->u.ret) ? true : false;
 453         lttng_directory_handle_put(handle);
 454 end:
 455         return ret_value->u.ret;
 456 }
 457
 458 static
 459 int _rmdir(struct run_as_data *data, struct run_as_ret *ret_value)
 460 {
 461         struct lttng_directory_handle *handle;
 462
 463         handle = lttng_directory_handle_create_from_dirfd(data->u.rmdir.dirfd);
 464         if (!handle) {
 465                 ret_value->u.ret = -1;
 466                 ret_value->_errno = errno;
 467                 ret_value->_error = true;
 468                 goto end;
 469         }
 470
 471         /* Ownership of dirfd is transferred to the handle. */
 472         data->u.rmdir.dirfd = -1;
 473
 474         ret_value->u.ret = lttng_directory_handle_remove_subdirectory(
 475                         handle, data->u.rmdir.path);
 476         ret_value->_errno = errno;
 477         ret_value->_error = (ret_value->u.ret) ? true : false;
 478         lttng_directory_handle_put(handle);
 479 end:
 480         return ret_value->u.ret;
 481 }
 482
 483 static
 484 int _rmdir_recursive(struct run_as_data *data, struct run_as_ret *ret_value)
 485 {
 486         struct lttng_directory_handle *handle;
 487
 488         handle = lttng_directory_handle_create_from_dirfd(data->u.rmdir.dirfd);
 489         if (!handle) {
 490                 ret_value->u.ret = -1;
 491                 ret_value->_errno = errno;
 492                 ret_value->_error = true;
 493                 goto end;
 494         }
 495
 496         /* Ownership of dirfd is transferred to the handle. */
 497         data->u.rmdir.dirfd = -1;
 498
 499         ret_value->u.ret = lttng_directory_handle_remove_subdirectory_recursive(
 500                         handle, data->u.rmdir.path, data->u.rmdir.flags);
 501         ret_value->_errno = errno;
 502         ret_value->_error = (ret_value->u.ret) ? true : false;
 503         lttng_directory_handle_put(handle);
 504 end:
 505         return ret_value->u.ret;
 506 }
 507
 508 static
 509 int _rename(struct run_as_data *data, struct run_as_ret *ret_value)
 510 {
 511         const char *old_path, *new_path;
 512         struct lttng_directory_handle *old_handle = NULL, *new_handle = NULL;
 513
 514         old_path = data->u.rename.old_path;
 515         new_path = data->u.rename.new_path;
 516
 517         old_handle = lttng_directory_handle_create_from_dirfd(
 518                         data->u.rename.dirfds[0]);
 519         if (!old_handle) {
 520                 ret_value->u.ret = -1;
 521                 goto end;
 522         }
 523         new_handle = lttng_directory_handle_create_from_dirfd(
 524                         data->u.rename.dirfds[1]);
 525         if (!new_handle) {
 526                 ret_value->u.ret = -1;
 527                 goto end;
 528         }
 529
 530         /* Ownership of dirfds are transferred to the handles. */
 531         data->u.rename.dirfds[0] = data->u.rename.dirfds[1] = -1;
 532
 533         /* Safe to call as we have transitioned to the requested uid/gid. */
 534         ret_value->u.ret = lttng_directory_handle_rename(
 535                         old_handle, old_path, new_handle, new_path);
 536 end:
 537         lttng_directory_handle_put(old_handle);
 538         lttng_directory_handle_put(new_handle);
 539         ret_value->_errno = errno;
 540         ret_value->_error = (ret_value->u.ret) ? true : false;
 541         return ret_value->u.ret;
 542 }
 543
 544 #ifdef HAVE_ELF_H
 545 static
 546 int _extract_elf_symbol_offset(struct run_as_data *data,
 547                 struct run_as_ret *ret_value)
 548 {
 549         int ret = 0;
 550         uint64_t offset;
 551
 552         ret_value->_error = false;
 553         ret = lttng_elf_get_symbol_offset(data->u.extract_elf_symbol_offset.fd,
 554                          data->u.extract_elf_symbol_offset.function,
 555                          &offset);
 556         if (ret) {
 557                 DBG("Failed to extract ELF function offset");
 558                 ret_value->_error = true;
 559         }
 560         ret_value->u.extract_elf_symbol_offset.offset = offset;
 561
 562         return ret;
 563 }
 564
 565 static
 566 int _extract_sdt_probe_offsets(struct run_as_data *data,
 567                 struct run_as_ret *ret_value)
 568 {
 569         int ret = 0;
 570         uint64_t *offsets = NULL;
 571         uint32_t num_offset;
 572
 573         ret_value->_error = false;
 574
 575         /* On success, this call allocates the offsets paramater. */
 576         ret = lttng_elf_get_sdt_probe_offsets(
 577                         data->u.extract_sdt_probe_offsets.fd,
 578                         data->u.extract_sdt_probe_offsets.provider_name,
 579                         data->u.extract_sdt_probe_offsets.probe_name,
 580                         &offsets, &num_offset);
 581
 582         if (ret) {
 583                 DBG("Failed to extract SDT probe offsets");
 584                 ret_value->_error = true;
 585                 goto end;
 586         }
 587
 588         if (num_offset <= 0 || num_offset > LTTNG_KERNEL_MAX_UPROBE_NUM) {
 589                 DBG("Wrong number of probes.");
 590                 ret = -1;
 591                 ret_value->_error = true;
 592                 goto free_offset;
 593         }
 594
 595         /* Copy the content of the offsets array to the ret struct. */
 596         memcpy(ret_value->u.extract_sdt_probe_offsets.offsets,
 597                         offsets, num_offset * sizeof(uint64_t));
 598
 599         ret_value->u.extract_sdt_probe_offsets.num_offset = num_offset;
 600
 601 free_offset:
 602         free(offsets);
 603 end:
 604         return ret;
 605 }
 606 #else
 607 static
 608 int _extract_elf_symbol_offset(struct run_as_data *data,
 609                 struct run_as_ret *ret_value)
 610 {
 611         ERR("Unimplemented runas command RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET");
 612         return -1;
 613 }
 614
 615 static
 616 int _extract_sdt_probe_offsets(struct run_as_data *data,
 617                 struct run_as_ret *ret_value)
 618 {
 619         ERR("Unimplemented runas command RUN_AS_EXTRACT_SDT_PROBE_OFFSETS");
 620         return -1;
 621 }
 622 #endif
 623
 624 static
 625 run_as_fct run_as_enum_to_fct(enum run_as_cmd cmd)
 626 {
 627         switch (cmd) {
 628         case RUN_AS_MKDIR:
 629         case RUN_AS_MKDIRAT:
 630                 return _mkdirat;
 631         case RUN_AS_MKDIR_RECURSIVE:
 632         case RUN_AS_MKDIRAT_RECURSIVE:
 633                 return _mkdirat_recursive;
 634         case RUN_AS_OPEN:
 635         case RUN_AS_OPENAT:
 636                 return _open;
 637         case RUN_AS_UNLINK:
 638         case RUN_AS_UNLINKAT:
 639                 return _unlink;
 640         case RUN_AS_RMDIR:
 641         case RUN_AS_RMDIRAT:
 642                 return _rmdir;
 643         case RUN_AS_RMDIR_RECURSIVE:
 644         case RUN_AS_RMDIRAT_RECURSIVE:
 645                 return _rmdir_recursive;
 646         case RUN_AS_RENAME:
 647         case RUN_AS_RENAMEAT:
 648                 return _rename;
 649         case RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET:
 650                 return _extract_elf_symbol_offset;
 651         case RUN_AS_EXTRACT_SDT_PROBE_OFFSETS:
 652                 return _extract_sdt_probe_offsets;
 653         default:
 654                 ERR("Unknown command %d", (int) cmd);
 655                 return NULL;
 656         }
 657 }
 658
 659 static
 660 int do_send_fds(int sock, const int *fds, unsigned int fd_count)
 661 {
 662         ssize_t len;
 663         unsigned int i;
 664
 665         for (i = 0; i < fd_count; i++) {
 666                 if (fds[i] < 0) {
 667                         ERR("Attempt to send invalid file descriptor to master (fd = %i)",
 668                                         fds[i]);
 669                         /* Return 0 as this is not a fatal error. */
 670                         return 0;
 671                 }
 672         }
 673
 674         len = lttcomm_send_fds_unix_sock(sock, fds, fd_count);
 675         return len < 0 ? -1 : 0;
 676 }
 677
 678 static
 679 int do_recv_fds(int sock, int *fds, unsigned int fd_count)
 680 {
 681         int ret = 0;
 682         unsigned int i;
 683         ssize_t len;
 684
 685         len = lttcomm_recv_fds_unix_sock(sock, fds, fd_count);
 686         if (len == 0) {
 687                 ret = -1;
 688                 goto end;
 689         } else if (len < 0) {
 690                 PERROR("Failed to receive file descriptors from socket");
 691                 ret = -1;
 692                 goto end;
 693         }
 694
 695         for (i = 0; i < fd_count; i++) {
 696                 if (fds[i] < 0) {
 697                         ERR("Invalid file descriptor received from worker (fd = %i)", fds[i]);
 698                         /* Return 0 as this is not a fatal error. */
 699                 }
 700         }
 701 end:
 702         return ret;
 703 }
 704
 705 static
 706 int send_fds_to_worker(const struct run_as_worker *worker,
 707                 const struct run_as_data *data)
 708 {
 709         int ret = 0;
 710         unsigned int i;
 711
 712         if (COMMAND_USE_CWD_FD(data) || COMMAND_IN_FD_COUNT(data) == 0) {
 713                 goto end;
 714         }
 715
 716         for (i = 0; i < COMMAND_IN_FD_COUNT(data); i++) {
 717                 if (COMMAND_IN_FDS(data)[i] < 0) {
 718                         ERR("Refusing to send invalid fd to worker (fd = %i)",
 719                                         COMMAND_IN_FDS(data)[i]);
 720                         ret = -1;
 721                         goto end;
 722                 }
 723         }
 724
 725         ret = do_send_fds(worker->sockpair[0], COMMAND_IN_FDS(data),
 726                         COMMAND_IN_FD_COUNT(data));
 727         if (ret < 0) {
 728                 PERROR("Failed to send file descriptor to run-as worker");
 729                 ret = -1;
 730                 goto end;
 731         }
 732 end:
 733         return ret;
 734 }
 735
 736 static
 737 int send_fds_to_master(struct run_as_worker *worker, enum run_as_cmd cmd,
 738                 struct run_as_ret *run_as_ret)
 739 {
 740         int ret = 0;
 741         unsigned int i;
 742
 743         if (COMMAND_OUT_FD_COUNT(cmd) == 0) {
 744                 goto end;
 745         }
 746
 747         ret = do_send_fds(worker->sockpair[1], COMMAND_OUT_FDS(cmd, run_as_ret),
 748                         COMMAND_OUT_FD_COUNT(cmd));
 749         if (ret < 0) {
 750                 PERROR("Failed to send file descriptor to master process");
 751                 goto end;
 752         }
 753
 754         for (i = 0; i < COMMAND_OUT_FD_COUNT(cmd); i++) {
 755                 int ret_close = close(COMMAND_OUT_FDS(cmd, run_as_ret)[i]);
 756
 757                 if (ret_close < 0) {
 758                         PERROR("Failed to close result file descriptor");
 759                 }
 760         }
 761 end:
 762         return ret;
 763 }
 764
 765 static
 766 int recv_fds_from_worker(const struct run_as_worker *worker, enum run_as_cmd cmd,
 767                 struct run_as_ret *run_as_ret)
 768 {
 769         int ret = 0;
 770
 771         if (COMMAND_OUT_FD_COUNT(cmd) == 0) {
 772                 goto end;
 773         }
 774
 775         ret = do_recv_fds(worker->sockpair[0], COMMAND_OUT_FDS(cmd, run_as_ret),
 776                         COMMAND_OUT_FD_COUNT(cmd));
 777         if (ret < 0) {
 778                 PERROR("Failed to receive file descriptor from run-as worker");
 779                 ret = -1;
 780         }
 781 end:
 782         return ret;
 783 }
 784
 785 static
 786 int recv_fds_from_master(struct run_as_worker *worker, struct run_as_data *data)
 787 {
 788         int ret = 0;
 789
 790         if (COMMAND_USE_CWD_FD(data)) {
 791                 unsigned int i;
 792
 793                 for (i = 0; i < COMMAND_IN_FD_COUNT(data); i++) {
 794                         COMMAND_IN_FDS(data)[i] = AT_FDCWD;
 795                 }
 796                 goto end;
 797         }
 798
 799         ret = do_recv_fds(worker->sockpair[1], COMMAND_IN_FDS(data),
 800                         COMMAND_IN_FD_COUNT(data));
 801         if (ret < 0) {
 802                 PERROR("Failed to receive file descriptors from master process");
 803                 ret = -1;
 804         }
 805 end:
 806         return ret;
 807 }
 808
 809 static
 810 int cleanup_received_fds(struct run_as_data *data)
 811 {
 812         int ret = 0, i;
 813
 814         for (i = 0; i < COMMAND_IN_FD_COUNT(data); i++) {
 815                 if (COMMAND_IN_FDS(data)[i] == -1) {
 816                         continue;
 817                 }
 818                 ret = close(COMMAND_IN_FDS(data)[i]);
 819                 if (ret) {
 820                         PERROR("Failed to close file descriptor received fd in run-as worker");
 821                         goto end;
 822                 }
 823         }
 824 end:
 825         return ret;
 826 }
 827
 828 static int get_user_infos_from_uid(
 829                 uid_t uid, char **username, gid_t *primary_gid)
 830 {
 831         int ret;
 832         char *buf = NULL;
 833         size_t buf_size;
 834         struct passwd pwd;
 835         struct passwd *result = NULL;
 836
 837         /* Fetch the max size for the temporary buffer. */
 838         errno = 0;
 839         buf_size = sysconf(_SC_GETPW_R_SIZE_MAX);
 840         if (buf_size < 0) {
 841                 if (errno != 0) {
 842                         PERROR("Failed to query _SC_GETPW_R_SIZE_MAX");
 843                         goto error;
 844                 }
 845
 846                 /* Limit is indeterminate. */
 847                 WARN("Failed to query _SC_GETPW_R_SIZE_MAX as it is "
 848                         "indeterminate; falling back to default buffer size");
 849                 buf_size = GETPW_BUFFER_FALLBACK_SIZE;
 850         }
 851
 852         buf = zmalloc(buf_size);
 853         if (buf == NULL) {
 854                 PERROR("Failed to allocate buffer to get password file entries");
 855                 goto error;
 856         }
 857
 858         ret = getpwuid_r(uid, &pwd, buf, buf_size, &result);
 859         if (ret < 0) {
 860                 PERROR("Failed to get user information for user:  uid = %d",
 861                                 (int) uid);
 862                 goto error;
 863         }
 864
 865         if (result == NULL) {
 866                 ERR("Failed to find user information in password entries: uid = %d",
 867                                 (int) uid);
 868                 ret = -1;
 869                 goto error;
 870         }
 871
 872         *username = strdup(result->pw_name);
 873         if (*username == NULL) {
 874                 PERROR("Failed to copy user name");
 875                 goto error;
 876         }
 877
 878         *primary_gid = result->pw_gid;
 879
 880 end:
 881         free(buf);
 882         return ret;
 883 error:
 884         *username = NULL;
 885         *primary_gid = -1;
 886         ret = -1;
 887         goto end;
 888 }
 889
 890 static int demote_creds(
 891                 uid_t prev_uid, gid_t prev_gid, uid_t new_uid, gid_t new_gid)
 892 {
 893         int ret = 0;
 894         gid_t primary_gid;
 895         char *username = NULL;
 896
 897         /* Change the group id. */
 898         if (prev_gid != new_gid) {
 899                 ret = setegid(new_gid);
 900                 if (ret < 0) {
 901                         PERROR("Failed to set effective group id: new_gid = %d",
 902                                         (int) new_gid);
 903                         goto end;
 904                 }
 905         }
 906
 907         /* Change the user id. */
 908         if (prev_uid != new_uid) {
 909                 ret = get_user_infos_from_uid(new_uid, &username, &primary_gid);
 910                 if (ret < 0) {
 911                         goto end;
 912                 }
 913
 914                 /*
 915                  * Initialize the supplementary group access list.
 916                  *
 917                  * This is needed to handle cases where the supplementary groups
 918                  * of the user the process is demoting-to would give it access
 919                  * to a given file/folder, but not it's primary group.
 920                  *
 921                  * e.g
 922                  *   username: User1
 923                  *   Primary Group: User1
 924                  *   Secondary group: Disk, Network
 925                  *
 926                  *   mkdir inside the following directory must work since User1
 927                  *   is part of the Network group.
 928                  *
 929                  *   drwxrwx--- 2 root Network 4096 Jul 23 17:17 /tmp/my_folder/
 930                  *
 931                  *
 932                  * The order of the following initgroups and seteuid calls is
 933                  * important here;
 934                  * Only a root process or one with CAP_SETGID capability can
 935                  * call the the initgroups() function. We must initialize the
 936                  * supplementary groups before we change the effective
 937                  * UID to a less-privileged user.
 938                  */
 939                 ret = initgroups(username, primary_gid);
 940                 if (ret < 0) {
 941                         PERROR("Failed to init the supplementary group access list: "
 942                                         "username = `%s`, primary gid = %d", username,
 943                                         (int) primary_gid);
 944                         goto end;
 945                 }
 946
 947                 ret = seteuid(new_uid);
 948                 if (ret < 0) {
 949                         PERROR("Failed to set effective user id: new_uid = %d",
 950                                         (int) new_uid);
 951                         goto end;
 952                 }
 953         }
 954 end:
 955         free(username);
 956         return ret;
 957 }
 958
 959 static int promote_creds(
 960                 uid_t prev_uid, gid_t prev_gid, uid_t new_uid, gid_t new_gid)
 961 {
 962         int ret = 0;
 963         gid_t primary_gid;
 964         char *username = NULL;
 965
 966         /* Change the group id. */
 967         if (prev_gid != new_gid) {
 968                 ret = setegid(new_gid);
 969                 if (ret < 0) {
 970                         PERROR("Failed to set effective group id: new_gid = %d",
 971                                         (int) new_gid);
 972                         goto end;
 973                 }
 974         }
 975
 976         /* Change the user id. */
 977         if (prev_uid != new_uid) {
 978                 ret = get_user_infos_from_uid(new_uid, &username, &primary_gid);
 979                 if (ret < 0) {
 980                         goto end;
 981                 }
 982
 983                 /*
 984                  * seteuid call must be done before the initgroups call because
 985                  * we need to be privileged (CAP_SETGID) to call initgroups().
 986                  */
 987                 ret = seteuid(new_uid);
 988                 if (ret < 0) {
 989                         PERROR("Failed to set effective user id: new_uid = %d",
 990                                         (int) new_uid);
 991                         goto end;
 992                 }
 993
 994                 /*
 995                  * Initialize the supplementary group access list.
 996                  *
 997                  * There is a possibility the groups we set in the following
 998                  * initgroups() call are not exactly the same as the ones we
 999                  * had when we originally demoted. This can happen if the
1000                  * /etc/group file is modified after the runas process is
1001                  * forked. This is very unlikely.
1002                  */
1003                 ret = initgroups(username, primary_gid);
1004                 if (ret < 0) {
1005                         PERROR("Failed to init the supplementary group access "
1006                                         "list: username = `%s`, primary gid = %d",
1007                                         username, (int) primary_gid)
1008                         goto end;
1009                 }
1010         }
1011 end:
1012         free(username);
1013         return ret;
1014 }
1015
1016 /*
1017  * Return < 0 on error, 0 if OK, 1 on hangup.
1018  */
1019 static
1020 int handle_one_cmd(struct run_as_worker *worker)
1021 {
1022         int ret = 0, promote_ret;
1023         struct run_as_data data = {};
1024         ssize_t readlen, writelen;
1025         struct run_as_ret sendret = {};
1026         run_as_fct cmd;
1027         uid_t prev_ruid;
1028         gid_t prev_rgid;
1029
1030         /*
1031          * Stage 1: Receive run_as_data struct from the master.
1032          * The structure contains the command type and all the parameters needed for
1033          * its execution
1034          */
1035         readlen = lttcomm_recv_unix_sock(worker->sockpair[1], &data,
1036                         sizeof(data));
1037         if (readlen == 0) {
1038                 /* hang up */
1039                 ret = 1;
1040                 goto end;
1041         }
1042         if (readlen < sizeof(data)) {
1043                 PERROR("lttcomm_recv_unix_sock error");
1044                 ret = -1;
1045                 goto end;
1046         }
1047
1048         cmd = run_as_enum_to_fct(data.cmd);
1049         if (!cmd) {
1050                 ret = -1;
1051                 goto end;
1052         }
1053
1054         /*
1055          * Stage 2: Receive file descriptor from master.
1056          * Some commands need a file descriptor as input so if it's needed we
1057          * receive the fd using the Unix socket.
1058          */
1059         ret = recv_fds_from_master(worker, &data);
1060         if (ret < 0) {
1061                 PERROR("recv_fd_from_master error");
1062                 ret = -1;
1063                 goto end;
1064         }
1065
1066         prev_ruid = getuid();
1067         prev_rgid = getgid();
1068
1069         ret = demote_creds(prev_ruid, prev_rgid, data.uid, data.gid);
1070         if (ret < 0) {
1071                 goto write_return;
1072         }
1073
1074         /*
1075          * Also set umask to 0 for mkdir executable bit.
1076          */
1077         umask(0);
1078
1079         /*
1080          * Stage 3: Execute the command
1081          */
1082         ret = (*cmd)(&data, &sendret);
1083         if (ret < 0) {
1084                 DBG("Execution of command returned an error");
1085         }
1086
1087 write_return:
1088         ret = cleanup_received_fds(&data);
1089         if (ret < 0) {
1090                 ERR("Error cleaning up FD");
1091                 goto promote_back;
1092         }
1093
1094         /*
1095          * Stage 4: Send run_as_ret structure to the master.
1096          * This structure contain the return value of the command and the errno.
1097          */
1098         writelen = lttcomm_send_unix_sock(worker->sockpair[1], &sendret,
1099                         sizeof(sendret));
1100         if (writelen < sizeof(sendret)) {
1101                 PERROR("lttcomm_send_unix_sock error");
1102                 ret = -1;
1103                 goto promote_back;
1104         }
1105
1106         /*
1107          * Stage 5: Send resulting file descriptors to the master.
1108          */
1109         ret = send_fds_to_master(worker, data.cmd, &sendret);
1110         if (ret < 0) {
1111                 DBG("Sending FD to master returned an error");
1112         }
1113
1114         ret = 0;
1115
1116 promote_back:
1117         /* Return to previous uid/gid. */
1118         promote_ret = promote_creds(data.uid, data.gid, prev_ruid, prev_rgid);
1119         if (promote_ret < 0) {
1120                 ERR("Failed to promote back to the initial credentials");
1121         }
1122
1123 end:
1124         return ret;
1125 }
1126
1127 static
1128 int run_as_worker(struct run_as_worker *worker)
1129 {
1130         int ret;
1131         ssize_t writelen;
1132         struct run_as_ret sendret;
1133         size_t proc_orig_len;
1134
1135         /*
1136          * Initialize worker. Set a different process cmdline.
1137          */
1138         proc_orig_len = strlen(worker->procname);
1139         memset(worker->procname, 0, proc_orig_len);
1140         strncpy(worker->procname, DEFAULT_RUN_AS_WORKER_NAME, proc_orig_len);
1141
1142         ret = lttng_prctl(PR_SET_NAME,
1143                         (unsigned long) DEFAULT_RUN_AS_WORKER_NAME, 0, 0, 0);
1144         if (ret && ret != -ENOSYS) {
1145                 /* Don't fail as this is not essential. */
1146                 PERROR("prctl PR_SET_NAME");
1147         }
1148
1149         memset(&sendret, 0, sizeof(sendret));
1150
1151         writelen = lttcomm_send_unix_sock(worker->sockpair[1], &sendret,
1152                         sizeof(sendret));
1153         if (writelen < sizeof(sendret)) {
1154                 PERROR("lttcomm_send_unix_sock error");
1155                 ret = EXIT_FAILURE;
1156                 goto end;
1157         }
1158
1159         for (;;) {
1160                 ret = handle_one_cmd(worker);
1161                 if (ret < 0) {
1162                         ret = EXIT_FAILURE;
1163                         goto end;
1164                 } else if (ret > 0) {
1165                         break;
1166                 } else {
1167                         continue;       /* Next command. */
1168                 }
1169         }
1170         ret = EXIT_SUCCESS;
1171 end:
1172         return ret;
1173 }
1174
1175 static
1176 int run_as_cmd(struct run_as_worker *worker,
1177                 enum run_as_cmd cmd,
1178                 struct run_as_data *data,
1179                 struct run_as_ret *ret_value,
1180                 uid_t uid, gid_t gid)
1181 {
1182         int ret = 0;
1183         ssize_t readlen, writelen;
1184
1185         /*
1186          * If we are non-root, we can only deal with our own uid.
1187          */
1188         if (geteuid() != 0) {
1189                 if (uid != geteuid()) {
1190                         ret = -1;
1191                         ret_value->_errno = EPERM;
1192                         ERR("Client (%d)/Server (%d) UID mismatch (and sessiond is not root)",
1193                                 (int) uid, (int) geteuid());
1194                         goto end;
1195                 }
1196         }
1197
1198         data->cmd = cmd;
1199         data->uid = uid;
1200         data->gid = gid;
1201
1202         /*
1203          * Stage 1: Send the run_as_data struct to the worker process
1204          */
1205         writelen = lttcomm_send_unix_sock(worker->sockpair[0], data,
1206                         sizeof(*data));
1207         if (writelen < sizeof(*data)) {
1208                 PERROR("Error writing message to run_as");
1209                 ret = -1;
1210                 ret_value->_errno = EIO;
1211                 goto end;
1212         }
1213
1214         /*
1215          * Stage 2: Send file descriptor to the worker process if needed
1216          */
1217         ret = send_fds_to_worker(worker, data);
1218         if (ret) {
1219                 PERROR("do_send_fd error");
1220                 ret = -1;
1221                 ret_value->_errno = EIO;
1222                 goto end;
1223         }
1224
1225         /*
1226          * Stage 3: Wait for the execution of the command
1227          */
1228
1229         /*
1230          * Stage 4: Receive the run_as_ret struct containing the return value and
1231          * errno
1232          */
1233         readlen = lttcomm_recv_unix_sock(worker->sockpair[0], ret_value,
1234                         sizeof(*ret_value));
1235         if (!readlen) {
1236                 ERR("Run-as worker has hung-up during run_as_cmd");
1237                 ret = -1;
1238                 ret_value->_errno = EIO;
1239                 goto end;
1240         } else if (readlen < sizeof(*ret_value)) {
1241                 PERROR("Error reading response from run_as");
1242                 ret = -1;
1243                 ret_value->_errno = errno;
1244                 goto end;
1245         }
1246
1247         if (ret_value->_error) {
1248                 /* Skip stage 5 on error as there will be no fd to receive. */
1249                 goto end;
1250         }
1251
1252         /*
1253          * Stage 5: Receive file descriptor if needed
1254          */
1255         ret = recv_fds_from_worker(worker, cmd, ret_value);
1256         if (ret < 0) {
1257                 ERR("Error receiving fd");
1258                 ret = -1;
1259                 ret_value->_errno = EIO;
1260         }
1261
1262 end:
1263         return ret;
1264 }
1265
1266 /*
1267  * This is for debugging ONLY and should not be considered secure.
1268  */
1269 static
1270 int run_as_noworker(enum run_as_cmd cmd,
1271                 struct run_as_data *data, struct run_as_ret *ret_value,
1272                 uid_t uid, gid_t gid)
1273 {
1274         int ret, saved_errno;
1275         mode_t old_mask;
1276         run_as_fct fct;
1277
1278         fct = run_as_enum_to_fct(cmd);
1279         if (!fct) {
1280                 errno = -ENOSYS;
1281                 ret = -1;
1282                 goto end;
1283         }
1284         old_mask = umask(0);
1285         ret = fct(data, ret_value);
1286         saved_errno = ret_value->_errno;
1287         umask(old_mask);
1288         errno = saved_errno;
1289 end:
1290         return ret;
1291 }
1292
1293 static
1294 int reset_sighandler(void)
1295 {
1296         int sig;
1297
1298         DBG("Resetting run_as worker signal handlers to default");
1299         for (sig = 1; sig <= 31; sig++) {
1300                 (void) signal(sig, SIG_DFL);
1301         }
1302         return 0;
1303 }
1304
1305 static
1306 void worker_sighandler(int sig)
1307 {
1308         const char *signame;
1309
1310         /*
1311          * The worker will inherit its parent's signals since they are part of
1312          * the same process group. However, in the case of SIGINT and SIGTERM,
1313          * we want to give the worker a chance to teardown gracefully when its
1314          * parent closes the command socket.
1315          */
1316         switch (sig) {
1317         case SIGINT:
1318                 signame = "SIGINT";
1319                 break;
1320         case SIGTERM:
1321                 signame = "SIGTERM";
1322                 break;
1323         default:
1324                 signame = NULL;
1325         }
1326
1327         if (signame) {
1328                 DBG("run_as worker received signal %s", signame);
1329         } else {
1330                 DBG("run_as_worker received signal %d", sig);
1331         }
1332 }
1333
1334 static
1335 int set_worker_sighandlers(void)
1336 {
1337         int ret = 0;
1338         sigset_t sigset;
1339         struct sigaction sa;
1340
1341         if ((ret = sigemptyset(&sigset)) < 0) {
1342                 PERROR("sigemptyset");
1343                 goto end;
1344         }
1345
1346         sa.sa_handler = worker_sighandler;
1347         sa.sa_mask = sigset;
1348         sa.sa_flags = 0;
1349         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
1350                 PERROR("sigaction SIGINT");
1351                 goto end;
1352         }
1353
1354         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
1355                 PERROR("sigaction SIGTERM");
1356                 goto end;
1357         }
1358
1359         DBG("run_as signal handler set for SIGTERM and SIGINT");
1360 end:
1361         return ret;
1362 }
1363
1364 static
1365 int run_as_create_worker_no_lock(const char *procname,
1366                 post_fork_cleanup_cb clean_up_func,
1367                 void *clean_up_user_data)
1368 {
1369         pid_t pid;
1370         int i, ret = 0;
1371         ssize_t readlen;
1372         struct run_as_ret recvret;
1373         struct run_as_worker *worker;
1374
1375         assert(!global_worker);
1376         if (!use_clone()) {
1377                 /*
1378                  * Don't initialize a worker, all run_as tasks will be performed
1379                  * in the current process.
1380                  */
1381                 ret = 0;
1382                 goto end;
1383         }
1384         worker = zmalloc(sizeof(*worker));
1385         if (!worker) {
1386                 ret = -ENOMEM;
1387                 goto end;
1388         }
1389         worker->procname = strdup(procname);
1390         if (!worker->procname) {
1391                 ret = -ENOMEM;
1392                 goto error_procname_alloc;
1393         }
1394         /* Create unix socket. */
1395         if (lttcomm_create_anon_unix_socketpair(worker->sockpair) < 0) {
1396                 ret = -1;
1397                 goto error_sock;
1398         }
1399
1400         /* Fork worker. */
1401         pid = fork();
1402         if (pid < 0) {
1403                 PERROR("fork");
1404                 ret = -1;
1405                 goto error_fork;
1406         } else if (pid == 0) {
1407                 /* Child */
1408
1409                 reset_sighandler();
1410
1411                 set_worker_sighandlers();
1412                 if (clean_up_func) {
1413                         if (clean_up_func(clean_up_user_data) < 0) {
1414                                 ERR("Run-as post-fork clean-up failed, exiting.");
1415                                 exit(EXIT_FAILURE);
1416                         }
1417                 }
1418
1419                 /* Just close, no shutdown. */
1420                 if (close(worker->sockpair[0])) {
1421                         PERROR("close");
1422                         exit(EXIT_FAILURE);
1423                 }
1424
1425                 /*
1426                  * Close all FDs aside from STDIN, STDOUT, STDERR and sockpair[1]
1427                  * Sockpair[1] is used as a control channel with the master
1428                  */
1429                 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
1430                         if (i != worker->sockpair[1]) {
1431                                 (void) close(i);
1432                         }
1433                 }
1434
1435                 worker->sockpair[0] = -1;
1436                 ret = run_as_worker(worker);
1437                 if (lttcomm_close_unix_sock(worker->sockpair[1])) {
1438                         PERROR("close");
1439                         ret = -1;
1440                 }
1441                 worker->sockpair[1] = -1;
1442                 free(worker->procname);
1443                 free(worker);
1444                 LOG(ret ? PRINT_ERR : PRINT_DBG, "run_as worker exiting (ret = %d)", ret);
1445                 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
1446         } else {
1447                 /* Parent */
1448
1449                 /* Just close, no shutdown. */
1450                 if (close(worker->sockpair[1])) {
1451                         PERROR("close");
1452                         ret = -1;
1453                         goto error_fork;
1454                 }
1455                 worker->sockpair[1] = -1;
1456                 worker->pid = pid;
1457                 /* Wait for worker to become ready. */
1458                 readlen = lttcomm_recv_unix_sock(worker->sockpair[0],
1459                                 &recvret, sizeof(recvret));
1460                 if (readlen < sizeof(recvret)) {
1461                         ERR("readlen: %zd", readlen);
1462                         PERROR("Error reading response from run_as at creation");
1463                         ret = -1;
1464                         goto error_fork;
1465                 }
1466                 global_worker = worker;
1467         }
1468 end:
1469         return ret;
1470
1471         /* Error handling. */
1472 error_fork:
1473         for (i = 0; i < 2; i++) {
1474                 if (worker->sockpair[i] < 0) {
1475                         continue;
1476                 }
1477                 if (lttcomm_close_unix_sock(worker->sockpair[i])) {
1478                         PERROR("close");
1479                 }
1480                 worker->sockpair[i] = -1;
1481         }
1482 error_sock:
1483         free(worker->procname);
1484 error_procname_alloc:
1485         free(worker);
1486         return ret;
1487 }
1488
1489 static
1490 void run_as_destroy_worker_no_lock(void)
1491 {
1492         struct run_as_worker *worker = global_worker;
1493
1494         DBG("Destroying run_as worker");
1495         if (!worker) {
1496                 return;
1497         }
1498         /* Close unix socket */
1499         DBG("Closing run_as worker socket");
1500         if (lttcomm_close_unix_sock(worker->sockpair[0])) {
1501                 PERROR("close");
1502         }
1503         worker->sockpair[0] = -1;
1504         /* Wait for worker. */
1505         for (;;) {
1506                 int status;
1507                 pid_t wait_ret;
1508
1509                 wait_ret = waitpid(worker->pid, &status, 0);
1510                 if (wait_ret < 0) {
1511                         if (errno == EINTR) {
1512                                 continue;
1513                         }
1514                         PERROR("waitpid");
1515                         break;
1516                 }
1517
1518                 if (WIFEXITED(status)) {
1519                         LOG(WEXITSTATUS(status) == 0 ? PRINT_DBG : PRINT_ERR,
1520                                         DEFAULT_RUN_AS_WORKER_NAME " terminated with status code %d",
1521                                         WEXITSTATUS(status));
1522                         break;
1523                 } else if (WIFSIGNALED(status)) {
1524                         ERR(DEFAULT_RUN_AS_WORKER_NAME " was killed by signal %d",
1525                                         WTERMSIG(status));
1526                         break;
1527                 }
1528         }
1529         free(worker->procname);
1530         free(worker);
1531         global_worker = NULL;
1532 }
1533
1534 static
1535 int run_as_restart_worker(struct run_as_worker *worker)
1536 {
1537         int ret = 0;
1538         char *procname = NULL;
1539
1540         procname = worker->procname;
1541
1542         /* Close socket to run_as worker process and clean up the zombie process */
1543         run_as_destroy_worker_no_lock();
1544
1545         /* Create a new run_as worker process*/
1546         ret = run_as_create_worker_no_lock(procname, NULL, NULL);
1547         if (ret < 0 ) {
1548                 ERR("Restarting the worker process failed");
1549                 ret = -1;
1550                 goto err;
1551         }
1552 err:
1553         return ret;
1554 }
1555
1556 static
1557 int run_as(enum run_as_cmd cmd, struct run_as_data *data,
1558                    struct run_as_ret *ret_value, uid_t uid, gid_t gid)
1559 {
1560         int ret, saved_errno;
1561
1562         pthread_mutex_lock(&worker_lock);
1563         if (use_clone()) {
1564                 DBG("Using run_as worker");
1565
1566                 assert(global_worker);
1567
1568                 ret = run_as_cmd(global_worker, cmd, data, ret_value, uid, gid);
1569                 saved_errno = ret_value->_errno;
1570
1571                 /*
1572                  * If the worker thread crashed the errno is set to EIO. we log
1573                  * the error and  start a new worker process.
1574                  */
1575                 if (ret == -1 && saved_errno == EIO) {
1576                         DBG("Socket closed unexpectedly... "
1577                                         "Restarting the worker process");
1578                         ret = run_as_restart_worker(global_worker);
1579                         if (ret == -1) {
1580                                 ERR("Failed to restart worker process.");
1581                                 goto err;
1582                         }
1583                 }
1584         } else {
1585                 DBG("Using run_as without worker");
1586                 ret = run_as_noworker(cmd, data, ret_value, uid, gid);
1587         }
1588 err:
1589         pthread_mutex_unlock(&worker_lock);
1590         return ret;
1591 }
1592
1593 LTTNG_HIDDEN
1594 int run_as_mkdir_recursive(const char *path, mode_t mode, uid_t uid, gid_t gid)
1595 {
1596         return run_as_mkdirat_recursive(AT_FDCWD, path, mode, uid, gid);
1597 }
1598
1599 LTTNG_HIDDEN
1600 int run_as_mkdirat_recursive(int dirfd, const char *path, mode_t mode,
1601                 uid_t uid, gid_t gid)
1602 {
1603         int ret;
1604         struct run_as_data data = {};
1605         struct run_as_ret run_as_ret = {};
1606
1607         DBG3("mkdirat() recursive fd = %d%s, path = %s, mode = %d, uid = %d, gid = %d",
1608                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1609                         path, (int) mode, (int) uid, (int) gid);
1610         ret = lttng_strncpy(data.u.mkdir.path, path,
1611                         sizeof(data.u.mkdir.path));
1612         if (ret) {
1613                 ERR("Failed to copy path argument of mkdirat recursive command");
1614                 goto error;
1615         }
1616         data.u.mkdir.path[sizeof(data.u.mkdir.path) - 1] = '\0';
1617         data.u.mkdir.mode = mode;
1618         data.u.mkdir.dirfd = dirfd;
1619         run_as(dirfd == AT_FDCWD ? RUN_AS_MKDIR_RECURSIVE : RUN_AS_MKDIRAT_RECURSIVE,
1620                         &data, &run_as_ret, uid, gid);
1621         errno = run_as_ret._errno;
1622         ret = run_as_ret.u.ret;
1623 error:
1624         return ret;
1625 }
1626
1627 LTTNG_HIDDEN
1628 int run_as_mkdir(const char *path, mode_t mode, uid_t uid, gid_t gid)
1629 {
1630         return run_as_mkdirat(AT_FDCWD, path, mode, uid, gid);
1631 }
1632
1633 LTTNG_HIDDEN
1634 int run_as_mkdirat(int dirfd, const char *path, mode_t mode,
1635                 uid_t uid, gid_t gid)
1636 {
1637         int ret;
1638         struct run_as_data data = {};
1639         struct run_as_ret run_as_ret = {};
1640
1641         DBG3("mkdirat() recursive fd = %d%s, path = %s, mode = %d, uid = %d, gid = %d",
1642                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1643                         path, (int) mode, (int) uid, (int) gid);
1644         ret = lttng_strncpy(data.u.mkdir.path, path,
1645                         sizeof(data.u.mkdir.path));
1646         if (ret) {
1647                 ERR("Failed to copy path argument of mkdirat command");
1648                 goto error;
1649         }
1650         data.u.mkdir.path[sizeof(data.u.mkdir.path) - 1] = '\0';
1651         data.u.mkdir.mode = mode;
1652         data.u.mkdir.dirfd = dirfd;
1653         run_as(dirfd == AT_FDCWD ? RUN_AS_MKDIR : RUN_AS_MKDIRAT,
1654                         &data, &run_as_ret, uid, gid);
1655         errno = run_as_ret._errno;
1656         ret = run_as_ret.u.ret;
1657 error:
1658         return ret;
1659 }
1660
1661 LTTNG_HIDDEN
1662 int run_as_open(const char *path, int flags, mode_t mode, uid_t uid,
1663                 gid_t gid)
1664 {
1665         return run_as_openat(AT_FDCWD, path, flags, mode, uid, gid);
1666 }
1667
1668 LTTNG_HIDDEN
1669 int run_as_openat(int dirfd, const char *path, int flags, mode_t mode,
1670                 uid_t uid, gid_t gid)
1671 {
1672         int ret;
1673         struct run_as_data data = {};
1674         struct run_as_ret run_as_ret = {};
1675
1676         DBG3("openat() fd = %d%s, path = %s, flags = %X, mode = %d, uid %d, gid %d",
1677                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1678                         path, flags, (int) mode, (int) uid, (int) gid);
1679         ret = lttng_strncpy(data.u.open.path, path, sizeof(data.u.open.path));
1680         if (ret) {
1681                 ERR("Failed to copy path argument of open command");
1682                 goto error;
1683         }
1684         data.u.open.flags = flags;
1685         data.u.open.mode = mode;
1686         data.u.open.dirfd = dirfd;
1687         run_as(dirfd == AT_FDCWD ? RUN_AS_OPEN : RUN_AS_OPENAT,
1688                         &data, &run_as_ret, uid, gid);
1689         errno = run_as_ret._errno;
1690         ret = run_as_ret.u.ret < 0 ? run_as_ret.u.ret :
1691                         run_as_ret.u.open.fd;
1692 error:
1693         return ret;
1694 }
1695
1696 LTTNG_HIDDEN
1697 int run_as_unlink(const char *path, uid_t uid, gid_t gid)
1698 {
1699         return run_as_unlinkat(AT_FDCWD, path, uid, gid);
1700 }
1701
1702 LTTNG_HIDDEN
1703 int run_as_unlinkat(int dirfd, const char *path, uid_t uid, gid_t gid)
1704 {
1705         int ret;
1706         struct run_as_data data = {};
1707         struct run_as_ret run_as_ret = {};
1708
1709         DBG3("unlinkat() fd = %d%s, path = %s, uid = %d, gid = %d",
1710                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1711                         path, (int) uid, (int) gid);
1712         ret = lttng_strncpy(data.u.unlink.path, path,
1713                         sizeof(data.u.unlink.path));
1714         if (ret) {
1715                 goto error;
1716         }
1717         data.u.unlink.dirfd = dirfd;
1718         run_as(dirfd == AT_FDCWD ? RUN_AS_UNLINK : RUN_AS_UNLINKAT, &data,
1719                         &run_as_ret, uid, gid);
1720         errno = run_as_ret._errno;
1721         ret = run_as_ret.u.ret;
1722 error:
1723         return ret;
1724 }
1725
1726 LTTNG_HIDDEN
1727 int run_as_rmdir(const char *path, uid_t uid, gid_t gid)
1728 {
1729         return run_as_rmdirat(AT_FDCWD, path, uid, gid);
1730 }
1731
1732 LTTNG_HIDDEN
1733 int run_as_rmdirat(int dirfd, const char *path, uid_t uid, gid_t gid)
1734 {
1735         int ret;
1736         struct run_as_data data = {};
1737         struct run_as_ret run_as_ret = {};
1738
1739         DBG3("rmdirat() fd = %d%s, path = %s, uid = %d, gid = %d",
1740                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1741                         path, (int) uid, (int) gid);
1742         ret = lttng_strncpy(data.u.rmdir.path, path,
1743                         sizeof(data.u.rmdir.path));
1744         if (ret) {
1745                 goto error;
1746         }
1747         data.u.rmdir.dirfd = dirfd;
1748         run_as(dirfd == AT_FDCWD ? RUN_AS_RMDIR : RUN_AS_RMDIRAT, &data,
1749                         &run_as_ret, uid, gid);
1750         errno = run_as_ret._errno;
1751         ret = run_as_ret.u.ret;
1752 error:
1753         return ret;
1754 }
1755
1756 LTTNG_HIDDEN
1757 int run_as_rmdir_recursive(const char *path, uid_t uid, gid_t gid, int flags)
1758 {
1759         return run_as_rmdirat_recursive(AT_FDCWD, path, uid, gid, flags);
1760 }
1761
1762 LTTNG_HIDDEN
1763 int run_as_rmdirat_recursive(int dirfd, const char *path, uid_t uid, gid_t gid, int flags)
1764 {
1765         int ret;
1766         struct run_as_data data = {};
1767         struct run_as_ret run_as_ret = {};
1768
1769         DBG3("rmdirat() recursive fd = %d%s, path = %s, uid = %d, gid = %d",
1770                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1771                         path, (int) uid, (int) gid);
1772         ret = lttng_strncpy(data.u.rmdir.path, path,
1773                         sizeof(data.u.rmdir.path));
1774         if (ret) {
1775                 goto error;
1776         }
1777         data.u.rmdir.dirfd = dirfd;
1778         data.u.rmdir.flags = flags;
1779         run_as(dirfd == AT_FDCWD ? RUN_AS_RMDIR_RECURSIVE : RUN_AS_RMDIRAT_RECURSIVE,
1780                         &data, &run_as_ret, uid, gid);
1781         errno = run_as_ret._errno;
1782         ret = run_as_ret.u.ret;
1783 error:
1784         return ret;
1785 }
1786
1787 LTTNG_HIDDEN
1788 int run_as_rename(const char *old, const char *new, uid_t uid, gid_t gid)
1789 {
1790         return run_as_renameat(AT_FDCWD, old, AT_FDCWD, new, uid, gid);
1791 }
1792
1793 LTTNG_HIDDEN
1794 int run_as_renameat(int old_dirfd, const char *old_name,
1795                 int new_dirfd, const char *new_name, uid_t uid, gid_t gid)
1796 {
1797         int ret;
1798         struct run_as_data data = {};
1799         struct run_as_ret run_as_ret = {};
1800
1801         DBG3("renameat() old_dirfd = %d%s, old_name = %s, new_dirfd = %d%s, new_name = %s, uid = %d, gid = %d",
1802                         old_dirfd, old_dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1803                         old_name,
1804                         new_dirfd, new_dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1805                         new_name, (int) uid, (int) gid);
1806         ret = lttng_strncpy(data.u.rename.old_path, old_name,
1807                         sizeof(data.u.rename.old_path));
1808         if (ret) {
1809                 goto error;
1810         }
1811         ret = lttng_strncpy(data.u.rename.new_path, new_name,
1812                         sizeof(data.u.rename.new_path));
1813         if (ret) {
1814                 goto error;
1815         }
1816
1817         data.u.rename.dirfds[0] = old_dirfd;
1818         data.u.rename.dirfds[1] = new_dirfd;
1819         run_as(old_dirfd == AT_FDCWD && new_dirfd == AT_FDCWD ?
1820                         RUN_AS_RENAME : RUN_AS_RENAMEAT,
1821                         &data, &run_as_ret, uid, gid);
1822         errno = run_as_ret._errno;
1823         ret = run_as_ret.u.ret;
1824 error:
1825         return ret;
1826 }
1827
1828 LTTNG_HIDDEN
1829 int run_as_extract_elf_symbol_offset(int fd, const char* function,
1830                 uid_t uid, gid_t gid, uint64_t *offset)
1831 {
1832         int ret;
1833         struct run_as_data data = {};
1834         struct run_as_ret run_as_ret = {};
1835
1836         DBG3("extract_elf_symbol_offset() on fd=%d and function=%s "
1837                         "with for uid %d and gid %d", fd, function,
1838                         (int) uid, (int) gid);
1839
1840         data.u.extract_elf_symbol_offset.fd = fd;
1841
1842         strncpy(data.u.extract_elf_symbol_offset.function, function, LTTNG_SYMBOL_NAME_LEN - 1);
1843         data.u.extract_elf_symbol_offset.function[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1844         ret = lttng_strncpy(data.u.extract_elf_symbol_offset.function,
1845                         function,
1846                         sizeof(data.u.extract_elf_symbol_offset.function));
1847         if (ret) {
1848                 goto error;
1849         }
1850
1851         run_as(RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET, &data, &run_as_ret, uid, gid);
1852         errno = run_as_ret._errno;
1853         if (run_as_ret._error) {
1854                 ret = -1;
1855                 goto error;
1856         }
1857
1858         *offset = run_as_ret.u.extract_elf_symbol_offset.offset;
1859 error:
1860         return ret;
1861 }
1862
1863 LTTNG_HIDDEN
1864 int run_as_extract_sdt_probe_offsets(int fd, const char* provider_name,
1865                 const char* probe_name, uid_t uid, gid_t gid,
1866                 uint64_t **offsets, uint32_t *num_offset)
1867 {
1868         int ret;
1869         struct run_as_data data = {};
1870         struct run_as_ret run_as_ret = {};
1871
1872         DBG3("extract_sdt_probe_offsets() on fd=%d, probe_name=%s and "
1873                         "provider_name=%s with for uid %d and gid %d", fd,
1874                         probe_name, provider_name, (int) uid, (int) gid);
1875
1876         data.u.extract_sdt_probe_offsets.fd = fd;
1877
1878         ret = lttng_strncpy(data.u.extract_sdt_probe_offsets.probe_name, probe_name,
1879                         sizeof(data.u.extract_sdt_probe_offsets.probe_name));
1880         if (ret) {
1881                 goto error;
1882         }
1883         ret = lttng_strncpy(data.u.extract_sdt_probe_offsets.provider_name,
1884                         provider_name,
1885                         sizeof(data.u.extract_sdt_probe_offsets.provider_name));
1886         if (ret) {
1887                 goto error;
1888         }
1889
1890         run_as(RUN_AS_EXTRACT_SDT_PROBE_OFFSETS, &data, &run_as_ret, uid, gid);
1891         errno = run_as_ret._errno;
1892         if (run_as_ret._error) {
1893                 ret = -1;
1894                 goto error;
1895         }
1896
1897         *num_offset = run_as_ret.u.extract_sdt_probe_offsets.num_offset;
1898         *offsets = zmalloc(*num_offset * sizeof(uint64_t));
1899         if (!*offsets) {
1900                 ret = -ENOMEM;
1901                 goto error;
1902         }
1903
1904         memcpy(*offsets, run_as_ret.u.extract_sdt_probe_offsets.offsets,
1905                         *num_offset * sizeof(uint64_t));
1906 error:
1907         return ret;
1908 }
1909
1910 LTTNG_HIDDEN
1911 int run_as_create_worker(const char *procname,
1912                 post_fork_cleanup_cb clean_up_func,
1913                 void *clean_up_user_data)
1914 {
1915         int ret;
1916
1917         pthread_mutex_lock(&worker_lock);
1918         ret = run_as_create_worker_no_lock(procname, clean_up_func,
1919                         clean_up_user_data);
1920         pthread_mutex_unlock(&worker_lock);
1921         return ret;
1922 }
1923
1924 LTTNG_HIDDEN
1925 void run_as_destroy_worker(void)
1926 {
1927         pthread_mutex_lock(&worker_lock);
1928         run_as_destroy_worker_no_lock();
1929         pthread_mutex_unlock(&worker_lock);
1930 }