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         long raw_get_pw_buf_size;
 834         size_t get_pw_buf_size;
 835         struct passwd pwd;
 836         struct passwd *result = NULL;
 837
 838         /* Fetch the max size for the temporary buffer. */
 839         errno = 0;
 840         raw_get_pw_buf_size = sysconf(_SC_GETPW_R_SIZE_MAX);
 841         if (raw_get_pw_buf_size < 0) {
 842                 if (errno != 0) {
 843                         PERROR("Failed to query _SC_GETPW_R_SIZE_MAX");
 844                         goto error;
 845                 }
 846
 847                 /* Limit is indeterminate. */
 848                 WARN("Failed to query _SC_GETPW_R_SIZE_MAX as it is "
 849                         "indeterminate; falling back to default buffer size");
 850                 raw_get_pw_buf_size = GETPW_BUFFER_FALLBACK_SIZE;
 851         }
 852
 853         get_pw_buf_size = (size_t) raw_get_pw_buf_size;
 854
 855         buf = zmalloc(get_pw_buf_size);
 856         if (buf == NULL) {
 857                 PERROR("Failed to allocate buffer to get password file entries");
 858                 goto error;
 859         }
 860
 861         ret = getpwuid_r(uid, &pwd, buf, get_pw_buf_size, &result);
 862         if (ret < 0) {
 863                 PERROR("Failed to get user information for user:  uid = %d",
 864                                 (int) uid);
 865                 goto error;
 866         }
 867
 868         if (result == NULL) {
 869                 ERR("Failed to find user information in password entries: uid = %d",
 870                                 (int) uid);
 871                 ret = -1;
 872                 goto error;
 873         }
 874
 875         *username = strdup(result->pw_name);
 876         if (*username == NULL) {
 877                 PERROR("Failed to copy user name");
 878                 goto error;
 879         }
 880
 881         *primary_gid = result->pw_gid;
 882
 883 end:
 884         free(buf);
 885         return ret;
 886 error:
 887         *username = NULL;
 888         *primary_gid = -1;
 889         ret = -1;
 890         goto end;
 891 }
 892
 893 static int demote_creds(
 894                 uid_t prev_uid, gid_t prev_gid, uid_t new_uid, gid_t new_gid)
 895 {
 896         int ret = 0;
 897         gid_t primary_gid;
 898         char *username = NULL;
 899
 900         /* Change the group id. */
 901         if (prev_gid != new_gid) {
 902                 ret = setegid(new_gid);
 903                 if (ret < 0) {
 904                         PERROR("Failed to set effective group id: new_gid = %d",
 905                                         (int) new_gid);
 906                         goto end;
 907                 }
 908         }
 909
 910         /* Change the user id. */
 911         if (prev_uid != new_uid) {
 912                 ret = get_user_infos_from_uid(new_uid, &username, &primary_gid);
 913                 if (ret < 0) {
 914                         goto end;
 915                 }
 916
 917                 /*
 918                  * Initialize the supplementary group access list.
 919                  *
 920                  * This is needed to handle cases where the supplementary groups
 921                  * of the user the process is demoting-to would give it access
 922                  * to a given file/folder, but not it's primary group.
 923                  *
 924                  * e.g
 925                  *   username: User1
 926                  *   Primary Group: User1
 927                  *   Secondary group: Disk, Network
 928                  *
 929                  *   mkdir inside the following directory must work since User1
 930                  *   is part of the Network group.
 931                  *
 932                  *   drwxrwx--- 2 root Network 4096 Jul 23 17:17 /tmp/my_folder/
 933                  *
 934                  *
 935                  * The order of the following initgroups and seteuid calls is
 936                  * important here;
 937                  * Only a root process or one with CAP_SETGID capability can
 938                  * call the the initgroups() function. We must initialize the
 939                  * supplementary groups before we change the effective
 940                  * UID to a less-privileged user.
 941                  */
 942                 ret = initgroups(username, primary_gid);
 943                 if (ret < 0) {
 944                         PERROR("Failed to init the supplementary group access list: "
 945                                         "username = `%s`, primary gid = %d", username,
 946                                         (int) primary_gid);
 947                         goto end;
 948                 }
 949
 950                 ret = seteuid(new_uid);
 951                 if (ret < 0) {
 952                         PERROR("Failed to set effective user id: new_uid = %d",
 953                                         (int) new_uid);
 954                         goto end;
 955                 }
 956         }
 957 end:
 958         free(username);
 959         return ret;
 960 }
 961
 962 static int promote_creds(
 963                 uid_t prev_uid, gid_t prev_gid, uid_t new_uid, gid_t new_gid)
 964 {
 965         int ret = 0;
 966         gid_t primary_gid;
 967         char *username = NULL;
 968
 969         /* Change the group id. */
 970         if (prev_gid != new_gid) {
 971                 ret = setegid(new_gid);
 972                 if (ret < 0) {
 973                         PERROR("Failed to set effective group id: new_gid = %d",
 974                                         (int) new_gid);
 975                         goto end;
 976                 }
 977         }
 978
 979         /* Change the user id. */
 980         if (prev_uid != new_uid) {
 981                 ret = get_user_infos_from_uid(new_uid, &username, &primary_gid);
 982                 if (ret < 0) {
 983                         goto end;
 984                 }
 985
 986                 /*
 987                  * seteuid call must be done before the initgroups call because
 988                  * we need to be privileged (CAP_SETGID) to call initgroups().
 989                  */
 990                 ret = seteuid(new_uid);
 991                 if (ret < 0) {
 992                         PERROR("Failed to set effective user id: new_uid = %d",
 993                                         (int) new_uid);
 994                         goto end;
 995                 }
 996
 997                 /*
 998                  * Initialize the supplementary group access list.
 999                  *
1000                  * There is a possibility the groups we set in the following
1001                  * initgroups() call are not exactly the same as the ones we
1002                  * had when we originally demoted. This can happen if the
1003                  * /etc/group file is modified after the runas process is
1004                  * forked. This is very unlikely.
1005                  */
1006                 ret = initgroups(username, primary_gid);
1007                 if (ret < 0) {
1008                         PERROR("Failed to init the supplementary group access "
1009                                         "list: username = `%s`, primary gid = %d",
1010                                         username, (int) primary_gid)
1011                         goto end;
1012                 }
1013         }
1014 end:
1015         free(username);
1016         return ret;
1017 }
1018
1019 /*
1020  * Return < 0 on error, 0 if OK, 1 on hangup.
1021  */
1022 static
1023 int handle_one_cmd(struct run_as_worker *worker)
1024 {
1025         int ret = 0, promote_ret;
1026         struct run_as_data data = {};
1027         ssize_t readlen, writelen;
1028         struct run_as_ret sendret = {};
1029         run_as_fct cmd;
1030         uid_t prev_ruid;
1031         gid_t prev_rgid;
1032
1033         /*
1034          * Stage 1: Receive run_as_data struct from the master.
1035          * The structure contains the command type and all the parameters needed for
1036          * its execution
1037          */
1038         readlen = lttcomm_recv_unix_sock(worker->sockpair[1], &data,
1039                         sizeof(data));
1040         if (readlen == 0) {
1041                 /* hang up */
1042                 ret = 1;
1043                 goto end;
1044         }
1045         if (readlen < sizeof(data)) {
1046                 PERROR("lttcomm_recv_unix_sock error");
1047                 ret = -1;
1048                 goto end;
1049         }
1050
1051         cmd = run_as_enum_to_fct(data.cmd);
1052         if (!cmd) {
1053                 ret = -1;
1054                 goto end;
1055         }
1056
1057         /*
1058          * Stage 2: Receive file descriptor from master.
1059          * Some commands need a file descriptor as input so if it's needed we
1060          * receive the fd using the Unix socket.
1061          */
1062         ret = recv_fds_from_master(worker, &data);
1063         if (ret < 0) {
1064                 PERROR("recv_fd_from_master error");
1065                 ret = -1;
1066                 goto end;
1067         }
1068
1069         prev_ruid = getuid();
1070         prev_rgid = getgid();
1071
1072         ret = demote_creds(prev_ruid, prev_rgid, data.uid, data.gid);
1073         if (ret < 0) {
1074                 goto write_return;
1075         }
1076
1077         /*
1078          * Also set umask to 0 for mkdir executable bit.
1079          */
1080         umask(0);
1081
1082         /*
1083          * Stage 3: Execute the command
1084          */
1085         ret = (*cmd)(&data, &sendret);
1086         if (ret < 0) {
1087                 DBG("Execution of command returned an error");
1088         }
1089
1090 write_return:
1091         ret = cleanup_received_fds(&data);
1092         if (ret < 0) {
1093                 ERR("Error cleaning up FD");
1094                 goto promote_back;
1095         }
1096
1097         /*
1098          * Stage 4: Send run_as_ret structure to the master.
1099          * This structure contain the return value of the command and the errno.
1100          */
1101         writelen = lttcomm_send_unix_sock(worker->sockpair[1], &sendret,
1102                         sizeof(sendret));
1103         if (writelen < sizeof(sendret)) {
1104                 PERROR("lttcomm_send_unix_sock error");
1105                 ret = -1;
1106                 goto promote_back;
1107         }
1108
1109         /*
1110          * Stage 5: Send resulting file descriptors to the master.
1111          */
1112         ret = send_fds_to_master(worker, data.cmd, &sendret);
1113         if (ret < 0) {
1114                 DBG("Sending FD to master returned an error");
1115         }
1116
1117         ret = 0;
1118
1119 promote_back:
1120         /* Return to previous uid/gid. */
1121         promote_ret = promote_creds(data.uid, data.gid, prev_ruid, prev_rgid);
1122         if (promote_ret < 0) {
1123                 ERR("Failed to promote back to the initial credentials");
1124         }
1125
1126 end:
1127         return ret;
1128 }
1129
1130 static
1131 int run_as_worker(struct run_as_worker *worker)
1132 {
1133         int ret;
1134         ssize_t writelen;
1135         struct run_as_ret sendret;
1136         size_t proc_orig_len;
1137
1138         /*
1139          * Initialize worker. Set a different process cmdline.
1140          */
1141         proc_orig_len = strlen(worker->procname);
1142         memset(worker->procname, 0, proc_orig_len);
1143         strncpy(worker->procname, DEFAULT_RUN_AS_WORKER_NAME, proc_orig_len);
1144
1145         ret = lttng_prctl(PR_SET_NAME,
1146                         (unsigned long) DEFAULT_RUN_AS_WORKER_NAME, 0, 0, 0);
1147         if (ret && ret != -ENOSYS) {
1148                 /* Don't fail as this is not essential. */
1149                 PERROR("prctl PR_SET_NAME");
1150         }
1151
1152         memset(&sendret, 0, sizeof(sendret));
1153
1154         writelen = lttcomm_send_unix_sock(worker->sockpair[1], &sendret,
1155                         sizeof(sendret));
1156         if (writelen < sizeof(sendret)) {
1157                 PERROR("lttcomm_send_unix_sock error");
1158                 ret = EXIT_FAILURE;
1159                 goto end;
1160         }
1161
1162         for (;;) {
1163                 ret = handle_one_cmd(worker);
1164                 if (ret < 0) {
1165                         ret = EXIT_FAILURE;
1166                         goto end;
1167                 } else if (ret > 0) {
1168                         break;
1169                 } else {
1170                         continue;       /* Next command. */
1171                 }
1172         }
1173         ret = EXIT_SUCCESS;
1174 end:
1175         return ret;
1176 }
1177
1178 static
1179 int run_as_cmd(struct run_as_worker *worker,
1180                 enum run_as_cmd cmd,
1181                 struct run_as_data *data,
1182                 struct run_as_ret *ret_value,
1183                 uid_t uid, gid_t gid)
1184 {
1185         int ret = 0;
1186         ssize_t readlen, writelen;
1187
1188         /*
1189          * If we are non-root, we can only deal with our own uid.
1190          */
1191         if (geteuid() != 0) {
1192                 if (uid != geteuid()) {
1193                         ret = -1;
1194                         ret_value->_errno = EPERM;
1195                         ERR("Client (%d)/Server (%d) UID mismatch (and sessiond is not root)",
1196                                 (int) uid, (int) geteuid());
1197                         goto end;
1198                 }
1199         }
1200
1201         data->cmd = cmd;
1202         data->uid = uid;
1203         data->gid = gid;
1204
1205         /*
1206          * Stage 1: Send the run_as_data struct to the worker process
1207          */
1208         writelen = lttcomm_send_unix_sock(worker->sockpair[0], data,
1209                         sizeof(*data));
1210         if (writelen < sizeof(*data)) {
1211                 PERROR("Error writing message to run_as");
1212                 ret = -1;
1213                 ret_value->_errno = EIO;
1214                 goto end;
1215         }
1216
1217         /*
1218          * Stage 2: Send file descriptor to the worker process if needed
1219          */
1220         ret = send_fds_to_worker(worker, data);
1221         if (ret) {
1222                 PERROR("do_send_fd error");
1223                 ret = -1;
1224                 ret_value->_errno = EIO;
1225                 goto end;
1226         }
1227
1228         /*
1229          * Stage 3: Wait for the execution of the command
1230          */
1231
1232         /*
1233          * Stage 4: Receive the run_as_ret struct containing the return value and
1234          * errno
1235          */
1236         readlen = lttcomm_recv_unix_sock(worker->sockpair[0], ret_value,
1237                         sizeof(*ret_value));
1238         if (!readlen) {
1239                 ERR("Run-as worker has hung-up during run_as_cmd");
1240                 ret = -1;
1241                 ret_value->_errno = EIO;
1242                 goto end;
1243         } else if (readlen < sizeof(*ret_value)) {
1244                 PERROR("Error reading response from run_as");
1245                 ret = -1;
1246                 ret_value->_errno = errno;
1247                 goto end;
1248         }
1249
1250         if (ret_value->_error) {
1251                 /* Skip stage 5 on error as there will be no fd to receive. */
1252                 goto end;
1253         }
1254
1255         /*
1256          * Stage 5: Receive file descriptor if needed
1257          */
1258         ret = recv_fds_from_worker(worker, cmd, ret_value);
1259         if (ret < 0) {
1260                 ERR("Error receiving fd");
1261                 ret = -1;
1262                 ret_value->_errno = EIO;
1263         }
1264
1265 end:
1266         return ret;
1267 }
1268
1269 /*
1270  * This is for debugging ONLY and should not be considered secure.
1271  */
1272 static
1273 int run_as_noworker(enum run_as_cmd cmd,
1274                 struct run_as_data *data, struct run_as_ret *ret_value,
1275                 uid_t uid, gid_t gid)
1276 {
1277         int ret, saved_errno;
1278         mode_t old_mask;
1279         run_as_fct fct;
1280
1281         fct = run_as_enum_to_fct(cmd);
1282         if (!fct) {
1283                 errno = -ENOSYS;
1284                 ret = -1;
1285                 goto end;
1286         }
1287         old_mask = umask(0);
1288         ret = fct(data, ret_value);
1289         saved_errno = ret_value->_errno;
1290         umask(old_mask);
1291         errno = saved_errno;
1292 end:
1293         return ret;
1294 }
1295
1296 static
1297 int reset_sighandler(void)
1298 {
1299         int sig;
1300
1301         DBG("Resetting run_as worker signal handlers to default");
1302         for (sig = 1; sig <= 31; sig++) {
1303                 (void) signal(sig, SIG_DFL);
1304         }
1305         return 0;
1306 }
1307
1308 static
1309 void worker_sighandler(int sig)
1310 {
1311         const char *signame;
1312
1313         /*
1314          * The worker will inherit its parent's signals since they are part of
1315          * the same process group. However, in the case of SIGINT and SIGTERM,
1316          * we want to give the worker a chance to teardown gracefully when its
1317          * parent closes the command socket.
1318          */
1319         switch (sig) {
1320         case SIGINT:
1321                 signame = "SIGINT";
1322                 break;
1323         case SIGTERM:
1324                 signame = "SIGTERM";
1325                 break;
1326         default:
1327                 signame = NULL;
1328         }
1329
1330         if (signame) {
1331                 DBG("run_as worker received signal %s", signame);
1332         } else {
1333                 DBG("run_as_worker received signal %d", sig);
1334         }
1335 }
1336
1337 static
1338 int set_worker_sighandlers(void)
1339 {
1340         int ret = 0;
1341         sigset_t sigset;
1342         struct sigaction sa;
1343
1344         if ((ret = sigemptyset(&sigset)) < 0) {
1345                 PERROR("sigemptyset");
1346                 goto end;
1347         }
1348
1349         sa.sa_handler = worker_sighandler;
1350         sa.sa_mask = sigset;
1351         sa.sa_flags = 0;
1352         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
1353                 PERROR("sigaction SIGINT");
1354                 goto end;
1355         }
1356
1357         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
1358                 PERROR("sigaction SIGTERM");
1359                 goto end;
1360         }
1361
1362         DBG("run_as signal handler set for SIGTERM and SIGINT");
1363 end:
1364         return ret;
1365 }
1366
1367 static
1368 int run_as_create_worker_no_lock(const char *procname,
1369                 post_fork_cleanup_cb clean_up_func,
1370                 void *clean_up_user_data)
1371 {
1372         pid_t pid;
1373         int i, ret = 0;
1374         ssize_t readlen;
1375         struct run_as_ret recvret;
1376         struct run_as_worker *worker;
1377
1378         assert(!global_worker);
1379         if (!use_clone()) {
1380                 /*
1381                  * Don't initialize a worker, all run_as tasks will be performed
1382                  * in the current process.
1383                  */
1384                 ret = 0;
1385                 goto end;
1386         }
1387         worker = zmalloc(sizeof(*worker));
1388         if (!worker) {
1389                 ret = -ENOMEM;
1390                 goto end;
1391         }
1392         worker->procname = strdup(procname);
1393         if (!worker->procname) {
1394                 ret = -ENOMEM;
1395                 goto error_procname_alloc;
1396         }
1397         /* Create unix socket. */
1398         if (lttcomm_create_anon_unix_socketpair(worker->sockpair) < 0) {
1399                 ret = -1;
1400                 goto error_sock;
1401         }
1402
1403         /* Fork worker. */
1404         pid = fork();
1405         if (pid < 0) {
1406                 PERROR("fork");
1407                 ret = -1;
1408                 goto error_fork;
1409         } else if (pid == 0) {
1410                 /* Child */
1411
1412                 reset_sighandler();
1413
1414                 set_worker_sighandlers();
1415                 if (clean_up_func) {
1416                         if (clean_up_func(clean_up_user_data) < 0) {
1417                                 ERR("Run-as post-fork clean-up failed, exiting.");
1418                                 exit(EXIT_FAILURE);
1419                         }
1420                 }
1421
1422                 /* Just close, no shutdown. */
1423                 if (close(worker->sockpair[0])) {
1424                         PERROR("close");
1425                         exit(EXIT_FAILURE);
1426                 }
1427
1428                 /*
1429                  * Close all FDs aside from STDIN, STDOUT, STDERR and sockpair[1]
1430                  * Sockpair[1] is used as a control channel with the master
1431                  */
1432                 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
1433                         if (i != worker->sockpair[1]) {
1434                                 (void) close(i);
1435                         }
1436                 }
1437
1438                 worker->sockpair[0] = -1;
1439                 ret = run_as_worker(worker);
1440                 if (lttcomm_close_unix_sock(worker->sockpair[1])) {
1441                         PERROR("close");
1442                         ret = -1;
1443                 }
1444                 worker->sockpair[1] = -1;
1445                 free(worker->procname);
1446                 free(worker);
1447                 LOG(ret ? PRINT_ERR : PRINT_DBG, "run_as worker exiting (ret = %d)", ret);
1448                 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
1449         } else {
1450                 /* Parent */
1451
1452                 /* Just close, no shutdown. */
1453                 if (close(worker->sockpair[1])) {
1454                         PERROR("close");
1455                         ret = -1;
1456                         goto error_fork;
1457                 }
1458                 worker->sockpair[1] = -1;
1459                 worker->pid = pid;
1460                 /* Wait for worker to become ready. */
1461                 readlen = lttcomm_recv_unix_sock(worker->sockpair[0],
1462                                 &recvret, sizeof(recvret));
1463                 if (readlen < sizeof(recvret)) {
1464                         ERR("readlen: %zd", readlen);
1465                         PERROR("Error reading response from run_as at creation");
1466                         ret = -1;
1467                         goto error_fork;
1468                 }
1469                 global_worker = worker;
1470         }
1471 end:
1472         return ret;
1473
1474         /* Error handling. */
1475 error_fork:
1476         for (i = 0; i < 2; i++) {
1477                 if (worker->sockpair[i] < 0) {
1478                         continue;
1479                 }
1480                 if (lttcomm_close_unix_sock(worker->sockpair[i])) {
1481                         PERROR("close");
1482                 }
1483                 worker->sockpair[i] = -1;
1484         }
1485 error_sock:
1486         free(worker->procname);
1487 error_procname_alloc:
1488         free(worker);
1489         return ret;
1490 }
1491
1492 static
1493 void run_as_destroy_worker_no_lock(void)
1494 {
1495         struct run_as_worker *worker = global_worker;
1496
1497         DBG("Destroying run_as worker");
1498         if (!worker) {
1499                 return;
1500         }
1501         /* Close unix socket */
1502         DBG("Closing run_as worker socket");
1503         if (lttcomm_close_unix_sock(worker->sockpair[0])) {
1504                 PERROR("close");
1505         }
1506         worker->sockpair[0] = -1;
1507         /* Wait for worker. */
1508         for (;;) {
1509                 int status;
1510                 pid_t wait_ret;
1511
1512                 wait_ret = waitpid(worker->pid, &status, 0);
1513                 if (wait_ret < 0) {
1514                         if (errno == EINTR) {
1515                                 continue;
1516                         }
1517                         PERROR("waitpid");
1518                         break;
1519                 }
1520
1521                 if (WIFEXITED(status)) {
1522                         LOG(WEXITSTATUS(status) == 0 ? PRINT_DBG : PRINT_ERR,
1523                                         DEFAULT_RUN_AS_WORKER_NAME " terminated with status code %d",
1524                                         WEXITSTATUS(status));
1525                         break;
1526                 } else if (WIFSIGNALED(status)) {
1527                         ERR(DEFAULT_RUN_AS_WORKER_NAME " was killed by signal %d",
1528                                         WTERMSIG(status));
1529                         break;
1530                 }
1531         }
1532         free(worker->procname);
1533         free(worker);
1534         global_worker = NULL;
1535 }
1536
1537 static
1538 int run_as_restart_worker(struct run_as_worker *worker)
1539 {
1540         int ret = 0;
1541         char *procname = NULL;
1542
1543         procname = worker->procname;
1544
1545         /* Close socket to run_as worker process and clean up the zombie process */
1546         run_as_destroy_worker_no_lock();
1547
1548         /* Create a new run_as worker process*/
1549         ret = run_as_create_worker_no_lock(procname, NULL, NULL);
1550         if (ret < 0 ) {
1551                 ERR("Restarting the worker process failed");
1552                 ret = -1;
1553                 goto err;
1554         }
1555 err:
1556         return ret;
1557 }
1558
1559 static
1560 int run_as(enum run_as_cmd cmd, struct run_as_data *data,
1561                    struct run_as_ret *ret_value, uid_t uid, gid_t gid)
1562 {
1563         int ret, saved_errno;
1564
1565         pthread_mutex_lock(&worker_lock);
1566         if (use_clone()) {
1567                 DBG("Using run_as worker");
1568
1569                 assert(global_worker);
1570
1571                 ret = run_as_cmd(global_worker, cmd, data, ret_value, uid, gid);
1572                 saved_errno = ret_value->_errno;
1573
1574                 /*
1575                  * If the worker thread crashed the errno is set to EIO. we log
1576                  * the error and  start a new worker process.
1577                  */
1578                 if (ret == -1 && saved_errno == EIO) {
1579                         DBG("Socket closed unexpectedly... "
1580                                         "Restarting the worker process");
1581                         ret = run_as_restart_worker(global_worker);
1582                         if (ret == -1) {
1583                                 ERR("Failed to restart worker process.");
1584                                 goto err;
1585                         }
1586                 }
1587         } else {
1588                 DBG("Using run_as without worker");
1589                 ret = run_as_noworker(cmd, data, ret_value, uid, gid);
1590         }
1591 err:
1592         pthread_mutex_unlock(&worker_lock);
1593         return ret;
1594 }
1595
1596 LTTNG_HIDDEN
1597 int run_as_mkdir_recursive(const char *path, mode_t mode, uid_t uid, gid_t gid)
1598 {
1599         return run_as_mkdirat_recursive(AT_FDCWD, path, mode, uid, gid);
1600 }
1601
1602 LTTNG_HIDDEN
1603 int run_as_mkdirat_recursive(int dirfd, const char *path, mode_t mode,
1604                 uid_t uid, gid_t gid)
1605 {
1606         int ret;
1607         struct run_as_data data = {};
1608         struct run_as_ret run_as_ret = {};
1609
1610         DBG3("mkdirat() recursive fd = %d%s, path = %s, mode = %d, uid = %d, gid = %d",
1611                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1612                         path, (int) mode, (int) uid, (int) gid);
1613         ret = lttng_strncpy(data.u.mkdir.path, path,
1614                         sizeof(data.u.mkdir.path));
1615         if (ret) {
1616                 ERR("Failed to copy path argument of mkdirat recursive command");
1617                 goto error;
1618         }
1619         data.u.mkdir.path[sizeof(data.u.mkdir.path) - 1] = '\0';
1620         data.u.mkdir.mode = mode;
1621         data.u.mkdir.dirfd = dirfd;
1622         run_as(dirfd == AT_FDCWD ? RUN_AS_MKDIR_RECURSIVE : RUN_AS_MKDIRAT_RECURSIVE,
1623                         &data, &run_as_ret, uid, gid);
1624         errno = run_as_ret._errno;
1625         ret = run_as_ret.u.ret;
1626 error:
1627         return ret;
1628 }
1629
1630 LTTNG_HIDDEN
1631 int run_as_mkdir(const char *path, mode_t mode, uid_t uid, gid_t gid)
1632 {
1633         return run_as_mkdirat(AT_FDCWD, path, mode, uid, gid);
1634 }
1635
1636 LTTNG_HIDDEN
1637 int run_as_mkdirat(int dirfd, const char *path, mode_t mode,
1638                 uid_t uid, gid_t gid)
1639 {
1640         int ret;
1641         struct run_as_data data = {};
1642         struct run_as_ret run_as_ret = {};
1643
1644         DBG3("mkdirat() recursive fd = %d%s, path = %s, mode = %d, uid = %d, gid = %d",
1645                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1646                         path, (int) mode, (int) uid, (int) gid);
1647         ret = lttng_strncpy(data.u.mkdir.path, path,
1648                         sizeof(data.u.mkdir.path));
1649         if (ret) {
1650                 ERR("Failed to copy path argument of mkdirat command");
1651                 goto error;
1652         }
1653         data.u.mkdir.path[sizeof(data.u.mkdir.path) - 1] = '\0';
1654         data.u.mkdir.mode = mode;
1655         data.u.mkdir.dirfd = dirfd;
1656         run_as(dirfd == AT_FDCWD ? RUN_AS_MKDIR : RUN_AS_MKDIRAT,
1657                         &data, &run_as_ret, uid, gid);
1658         errno = run_as_ret._errno;
1659         ret = run_as_ret.u.ret;
1660 error:
1661         return ret;
1662 }
1663
1664 LTTNG_HIDDEN
1665 int run_as_open(const char *path, int flags, mode_t mode, uid_t uid,
1666                 gid_t gid)
1667 {
1668         return run_as_openat(AT_FDCWD, path, flags, mode, uid, gid);
1669 }
1670
1671 LTTNG_HIDDEN
1672 int run_as_openat(int dirfd, const char *path, int flags, mode_t mode,
1673                 uid_t uid, gid_t gid)
1674 {
1675         int ret;
1676         struct run_as_data data = {};
1677         struct run_as_ret run_as_ret = {};
1678
1679         DBG3("openat() fd = %d%s, path = %s, flags = %X, mode = %d, uid %d, gid %d",
1680                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1681                         path, flags, (int) mode, (int) uid, (int) gid);
1682         ret = lttng_strncpy(data.u.open.path, path, sizeof(data.u.open.path));
1683         if (ret) {
1684                 ERR("Failed to copy path argument of open command");
1685                 goto error;
1686         }
1687         data.u.open.flags = flags;
1688         data.u.open.mode = mode;
1689         data.u.open.dirfd = dirfd;
1690         run_as(dirfd == AT_FDCWD ? RUN_AS_OPEN : RUN_AS_OPENAT,
1691                         &data, &run_as_ret, uid, gid);
1692         errno = run_as_ret._errno;
1693         ret = run_as_ret.u.ret < 0 ? run_as_ret.u.ret :
1694                         run_as_ret.u.open.fd;
1695 error:
1696         return ret;
1697 }
1698
1699 LTTNG_HIDDEN
1700 int run_as_unlink(const char *path, uid_t uid, gid_t gid)
1701 {
1702         return run_as_unlinkat(AT_FDCWD, path, uid, gid);
1703 }
1704
1705 LTTNG_HIDDEN
1706 int run_as_unlinkat(int dirfd, const char *path, uid_t uid, gid_t gid)
1707 {
1708         int ret;
1709         struct run_as_data data = {};
1710         struct run_as_ret run_as_ret = {};
1711
1712         DBG3("unlinkat() fd = %d%s, path = %s, uid = %d, gid = %d",
1713                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1714                         path, (int) uid, (int) gid);
1715         ret = lttng_strncpy(data.u.unlink.path, path,
1716                         sizeof(data.u.unlink.path));
1717         if (ret) {
1718                 goto error;
1719         }
1720         data.u.unlink.dirfd = dirfd;
1721         run_as(dirfd == AT_FDCWD ? RUN_AS_UNLINK : RUN_AS_UNLINKAT, &data,
1722                         &run_as_ret, uid, gid);
1723         errno = run_as_ret._errno;
1724         ret = run_as_ret.u.ret;
1725 error:
1726         return ret;
1727 }
1728
1729 LTTNG_HIDDEN
1730 int run_as_rmdir(const char *path, uid_t uid, gid_t gid)
1731 {
1732         return run_as_rmdirat(AT_FDCWD, path, uid, gid);
1733 }
1734
1735 LTTNG_HIDDEN
1736 int run_as_rmdirat(int dirfd, const char *path, uid_t uid, gid_t gid)
1737 {
1738         int ret;
1739         struct run_as_data data = {};
1740         struct run_as_ret run_as_ret = {};
1741
1742         DBG3("rmdirat() fd = %d%s, path = %s, uid = %d, gid = %d",
1743                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1744                         path, (int) uid, (int) gid);
1745         ret = lttng_strncpy(data.u.rmdir.path, path,
1746                         sizeof(data.u.rmdir.path));
1747         if (ret) {
1748                 goto error;
1749         }
1750         data.u.rmdir.dirfd = dirfd;
1751         run_as(dirfd == AT_FDCWD ? RUN_AS_RMDIR : RUN_AS_RMDIRAT, &data,
1752                         &run_as_ret, uid, gid);
1753         errno = run_as_ret._errno;
1754         ret = run_as_ret.u.ret;
1755 error:
1756         return ret;
1757 }
1758
1759 LTTNG_HIDDEN
1760 int run_as_rmdir_recursive(const char *path, uid_t uid, gid_t gid, int flags)
1761 {
1762         return run_as_rmdirat_recursive(AT_FDCWD, path, uid, gid, flags);
1763 }
1764
1765 LTTNG_HIDDEN
1766 int run_as_rmdirat_recursive(int dirfd, const char *path, uid_t uid, gid_t gid, int flags)
1767 {
1768         int ret;
1769         struct run_as_data data = {};
1770         struct run_as_ret run_as_ret = {};
1771
1772         DBG3("rmdirat() recursive fd = %d%s, path = %s, uid = %d, gid = %d",
1773                         dirfd, dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1774                         path, (int) uid, (int) gid);
1775         ret = lttng_strncpy(data.u.rmdir.path, path,
1776                         sizeof(data.u.rmdir.path));
1777         if (ret) {
1778                 goto error;
1779         }
1780         data.u.rmdir.dirfd = dirfd;
1781         data.u.rmdir.flags = flags;
1782         run_as(dirfd == AT_FDCWD ? RUN_AS_RMDIR_RECURSIVE : RUN_AS_RMDIRAT_RECURSIVE,
1783                         &data, &run_as_ret, uid, gid);
1784         errno = run_as_ret._errno;
1785         ret = run_as_ret.u.ret;
1786 error:
1787         return ret;
1788 }
1789
1790 LTTNG_HIDDEN
1791 int run_as_rename(const char *old, const char *new, uid_t uid, gid_t gid)
1792 {
1793         return run_as_renameat(AT_FDCWD, old, AT_FDCWD, new, uid, gid);
1794 }
1795
1796 LTTNG_HIDDEN
1797 int run_as_renameat(int old_dirfd, const char *old_name,
1798                 int new_dirfd, const char *new_name, uid_t uid, gid_t gid)
1799 {
1800         int ret;
1801         struct run_as_data data = {};
1802         struct run_as_ret run_as_ret = {};
1803
1804         DBG3("renameat() old_dirfd = %d%s, old_name = %s, new_dirfd = %d%s, new_name = %s, uid = %d, gid = %d",
1805                         old_dirfd, old_dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1806                         old_name,
1807                         new_dirfd, new_dirfd == AT_FDCWD ? " (AT_FDCWD)" : "",
1808                         new_name, (int) uid, (int) gid);
1809         ret = lttng_strncpy(data.u.rename.old_path, old_name,
1810                         sizeof(data.u.rename.old_path));
1811         if (ret) {
1812                 goto error;
1813         }
1814         ret = lttng_strncpy(data.u.rename.new_path, new_name,
1815                         sizeof(data.u.rename.new_path));
1816         if (ret) {
1817                 goto error;
1818         }
1819
1820         data.u.rename.dirfds[0] = old_dirfd;
1821         data.u.rename.dirfds[1] = new_dirfd;
1822         run_as(old_dirfd == AT_FDCWD && new_dirfd == AT_FDCWD ?
1823                         RUN_AS_RENAME : RUN_AS_RENAMEAT,
1824                         &data, &run_as_ret, uid, gid);
1825         errno = run_as_ret._errno;
1826         ret = run_as_ret.u.ret;
1827 error:
1828         return ret;
1829 }
1830
1831 LTTNG_HIDDEN
1832 int run_as_extract_elf_symbol_offset(int fd, const char* function,
1833                 uid_t uid, gid_t gid, uint64_t *offset)
1834 {
1835         int ret;
1836         struct run_as_data data = {};
1837         struct run_as_ret run_as_ret = {};
1838
1839         DBG3("extract_elf_symbol_offset() on fd=%d and function=%s "
1840                         "with for uid %d and gid %d", fd, function,
1841                         (int) uid, (int) gid);
1842
1843         data.u.extract_elf_symbol_offset.fd = fd;
1844
1845         strncpy(data.u.extract_elf_symbol_offset.function, function, LTTNG_SYMBOL_NAME_LEN - 1);
1846         data.u.extract_elf_symbol_offset.function[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1847         ret = lttng_strncpy(data.u.extract_elf_symbol_offset.function,
1848                         function,
1849                         sizeof(data.u.extract_elf_symbol_offset.function));
1850         if (ret) {
1851                 goto error;
1852         }
1853
1854         run_as(RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET, &data, &run_as_ret, uid, gid);
1855         errno = run_as_ret._errno;
1856         if (run_as_ret._error) {
1857                 ret = -1;
1858                 goto error;
1859         }
1860
1861         *offset = run_as_ret.u.extract_elf_symbol_offset.offset;
1862 error:
1863         return ret;
1864 }
1865
1866 LTTNG_HIDDEN
1867 int run_as_extract_sdt_probe_offsets(int fd, const char* provider_name,
1868                 const char* probe_name, uid_t uid, gid_t gid,
1869                 uint64_t **offsets, uint32_t *num_offset)
1870 {
1871         int ret;
1872         struct run_as_data data = {};
1873         struct run_as_ret run_as_ret = {};
1874
1875         DBG3("extract_sdt_probe_offsets() on fd=%d, probe_name=%s and "
1876                         "provider_name=%s with for uid %d and gid %d", fd,
1877                         probe_name, provider_name, (int) uid, (int) gid);
1878
1879         data.u.extract_sdt_probe_offsets.fd = fd;
1880
1881         ret = lttng_strncpy(data.u.extract_sdt_probe_offsets.probe_name, probe_name,
1882                         sizeof(data.u.extract_sdt_probe_offsets.probe_name));
1883         if (ret) {
1884                 goto error;
1885         }
1886         ret = lttng_strncpy(data.u.extract_sdt_probe_offsets.provider_name,
1887                         provider_name,
1888                         sizeof(data.u.extract_sdt_probe_offsets.provider_name));
1889         if (ret) {
1890                 goto error;
1891         }
1892
1893         run_as(RUN_AS_EXTRACT_SDT_PROBE_OFFSETS, &data, &run_as_ret, uid, gid);
1894         errno = run_as_ret._errno;
1895         if (run_as_ret._error) {
1896                 ret = -1;
1897                 goto error;
1898         }
1899
1900         *num_offset = run_as_ret.u.extract_sdt_probe_offsets.num_offset;
1901         *offsets = zmalloc(*num_offset * sizeof(uint64_t));
1902         if (!*offsets) {
1903                 ret = -ENOMEM;
1904                 goto error;
1905         }
1906
1907         memcpy(*offsets, run_as_ret.u.extract_sdt_probe_offsets.offsets,
1908                         *num_offset * sizeof(uint64_t));
1909 error:
1910         return ret;
1911 }
1912
1913 LTTNG_HIDDEN
1914 int run_as_create_worker(const char *procname,
1915                 post_fork_cleanup_cb clean_up_func,
1916                 void *clean_up_user_data)
1917 {
1918         int ret;
1919
1920         pthread_mutex_lock(&worker_lock);
1921         ret = run_as_create_worker_no_lock(procname, clean_up_func,
1922                         clean_up_user_data);
1923         pthread_mutex_unlock(&worker_lock);
1924         return ret;
1925 }
1926
1927 LTTNG_HIDDEN
1928 void run_as_destroy_worker(void)
1929 {
1930         pthread_mutex_lock(&worker_lock);
1931         run_as_destroy_worker_no_lock();
1932         pthread_mutex_unlock(&worker_lock);
1933 }