src/core/execute.h

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2 #pragma once
   3
   4 typedef struct ExecStatus ExecStatus;
   5 typedef struct ExecCommand ExecCommand;
   6 typedef struct ExecContext ExecContext;
   7 typedef struct ExecRuntime ExecRuntime;
   8 typedef struct ExecParameters ExecParameters;
   9 typedef struct Manager Manager;
  10
  11 #include <sched.h>
  12 #include <stdbool.h>
  13 #include <stdio.h>
  14 #include <sys/capability.h>
  15
  16 #include "cgroup-util.h"
  17 #include "coredump-util.h"
  18 #include "cpu-set-util.h"
  19 #include "exec-util.h"
  20 #include "fdset.h"
  21 #include "list.h"
  22 #include "missing_resource.h"
  23 #include "namespace.h"
  24 #include "nsflags.h"
  25 #include "numa-util.h"
  26 #include "path-util.h"
  27 #include "set.h"
  28 #include "time-util.h"
  29
  30 #define EXEC_STDIN_DATA_MAX (64U*1024U*1024U)
  31
  32 typedef enum ExecUtmpMode {
  33         EXEC_UTMP_INIT,
  34         EXEC_UTMP_LOGIN,
  35         EXEC_UTMP_USER,
  36         _EXEC_UTMP_MODE_MAX,
  37         _EXEC_UTMP_MODE_INVALID = -EINVAL,
  38 } ExecUtmpMode;
  39
  40 typedef enum ExecInput {
  41         EXEC_INPUT_NULL,
  42         EXEC_INPUT_TTY,
  43         EXEC_INPUT_TTY_FORCE,
  44         EXEC_INPUT_TTY_FAIL,
  45         EXEC_INPUT_SOCKET,
  46         EXEC_INPUT_NAMED_FD,
  47         EXEC_INPUT_DATA,
  48         EXEC_INPUT_FILE,
  49         _EXEC_INPUT_MAX,
  50         _EXEC_INPUT_INVALID = -EINVAL,
  51 } ExecInput;
  52
  53 typedef enum ExecOutput {
  54         EXEC_OUTPUT_INHERIT,
  55         EXEC_OUTPUT_NULL,
  56         EXEC_OUTPUT_TTY,
  57         EXEC_OUTPUT_KMSG,
  58         EXEC_OUTPUT_KMSG_AND_CONSOLE,
  59         EXEC_OUTPUT_JOURNAL,
  60         EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
  61         EXEC_OUTPUT_SOCKET,
  62         EXEC_OUTPUT_NAMED_FD,
  63         EXEC_OUTPUT_FILE,
  64         EXEC_OUTPUT_FILE_APPEND,
  65         EXEC_OUTPUT_FILE_TRUNCATE,
  66         _EXEC_OUTPUT_MAX,
  67         _EXEC_OUTPUT_INVALID = -EINVAL,
  68 } ExecOutput;
  69
  70 typedef enum ExecPreserveMode {
  71         EXEC_PRESERVE_NO,
  72         EXEC_PRESERVE_YES,
  73         EXEC_PRESERVE_RESTART,
  74         _EXEC_PRESERVE_MODE_MAX,
  75         _EXEC_PRESERVE_MODE_INVALID = -EINVAL,
  76 } ExecPreserveMode;
  77
  78 typedef enum ExecKeyringMode {
  79         EXEC_KEYRING_INHERIT,
  80         EXEC_KEYRING_PRIVATE,
  81         EXEC_KEYRING_SHARED,
  82         _EXEC_KEYRING_MODE_MAX,
  83         _EXEC_KEYRING_MODE_INVALID = -EINVAL,
  84 } ExecKeyringMode;
  85
  86 /* Contains start and exit information about an executed command.  */
  87 struct ExecStatus {
  88         dual_timestamp start_timestamp;
  89         dual_timestamp exit_timestamp;
  90         pid_t pid;
  91         int code;     /* as in siginfo_t::si_code */
  92         int status;   /* as in siginfo_t::si_status */
  93 };
  94
  95 /* Stores information about commands we execute. Covers both configuration settings as well as runtime data. */
  96 struct ExecCommand {
  97         char *path;
  98         char **argv;
  99         ExecStatus exec_status;
 100         ExecCommandFlags flags;
 101         LIST_FIELDS(ExecCommand, command); /* useful for chaining commands */
 102 };
 103
 104 /* Encapsulates certain aspects of the runtime environment that is to be shared between multiple otherwise separate
 105  * invocations of commands. Specifically, this allows sharing of /tmp and /var/tmp data as well as network namespaces
 106  * between invocations of commands. This is a reference counted object, with one reference taken by each currently
 107  * active command invocation that wants to share this runtime. */
 108 struct ExecRuntime {
 109         unsigned n_ref;
 110
 111         Manager *manager;
 112
 113         char *id; /* Unit id of the owner */
 114
 115         char *tmp_dir;
 116         char *var_tmp_dir;
 117
 118         /* An AF_UNIX socket pair, that contains a datagram containing a file descriptor referring to the network
 119          * namespace. */
 120         int netns_storage_socket[2];
 121
 122         /* Like netns_storage_socket, but the file descriptor is referring to the IPC namespace. */
 123         int ipcns_storage_socket[2];
 124 };
 125
 126 typedef enum ExecDirectoryType {
 127         EXEC_DIRECTORY_RUNTIME = 0,
 128         EXEC_DIRECTORY_STATE,
 129         EXEC_DIRECTORY_CACHE,
 130         EXEC_DIRECTORY_LOGS,
 131         EXEC_DIRECTORY_CONFIGURATION,
 132         _EXEC_DIRECTORY_TYPE_MAX,
 133         _EXEC_DIRECTORY_TYPE_INVALID = -EINVAL,
 134 } ExecDirectoryType;
 135
 136 typedef struct ExecDirectoryItem {
 137         char *path;
 138         char **symlinks;
 139         bool only_create;
 140 } ExecDirectoryItem;
 141
 142 typedef struct ExecDirectory {
 143         mode_t mode;
 144         size_t n_items;
 145         ExecDirectoryItem *items;
 146 } ExecDirectory;
 147
 148 typedef enum ExecCleanMask {
 149         /* In case you wonder why the bitmask below doesn't use "directory" in its name: we want to keep this
 150          * generic so that .timer timestamp files can nicely be covered by this too, and similar. */
 151         EXEC_CLEAN_RUNTIME       = 1U << EXEC_DIRECTORY_RUNTIME,
 152         EXEC_CLEAN_STATE         = 1U << EXEC_DIRECTORY_STATE,
 153         EXEC_CLEAN_CACHE         = 1U << EXEC_DIRECTORY_CACHE,
 154         EXEC_CLEAN_LOGS          = 1U << EXEC_DIRECTORY_LOGS,
 155         EXEC_CLEAN_CONFIGURATION = 1U << EXEC_DIRECTORY_CONFIGURATION,
 156         EXEC_CLEAN_NONE          = 0,
 157         EXEC_CLEAN_ALL           = (1U << _EXEC_DIRECTORY_TYPE_MAX) - 1,
 158         _EXEC_CLEAN_MASK_INVALID = -EINVAL,
 159 } ExecCleanMask;
 160
 161 /* A credential configured with LoadCredential= */
 162 typedef struct ExecLoadCredential {
 163         char *id, *path;
 164         bool encrypted;
 165 } ExecLoadCredential;
 166
 167 /* A credential configured with SetCredential= */
 168 typedef struct ExecSetCredential {
 169         char *id;
 170         bool encrypted;
 171         void *data;
 172         size_t size;
 173 } ExecSetCredential;
 174
 175 /* Encodes configuration parameters applied to invoked commands. Does not carry runtime data, but only configuration
 176  * changes sourced from unit files and suchlike. ExecContext objects are usually embedded into Unit objects, and do not
 177  * change after being loaded. */
 178 struct ExecContext {
 179         char **environment;
 180         char **environment_files;
 181         char **pass_environment;
 182         char **unset_environment;
 183
 184         struct rlimit *rlimit[_RLIMIT_MAX];
 185         char *working_directory, *root_directory, *root_image, *root_verity, *root_hash_path, *root_hash_sig_path;
 186         void *root_hash, *root_hash_sig;
 187         size_t root_hash_size, root_hash_sig_size;
 188         LIST_HEAD(MountOptions, root_image_options);
 189         bool working_directory_missing_ok:1;
 190         bool working_directory_home:1;
 191
 192         bool oom_score_adjust_set:1;
 193         bool coredump_filter_set:1;
 194         bool nice_set:1;
 195         bool ioprio_set:1;
 196         bool cpu_sched_set:1;
 197         bool mount_apivfs_set:1;
 198
 199         /* This is not exposed to the user but available internally. We need it to make sure that whenever we
 200          * spawn /usr/bin/mount it is run in the same process group as us so that the autofs logic detects
 201          * that it belongs to us and we don't enter a trigger loop. */
 202         bool same_pgrp;
 203
 204         bool cpu_sched_reset_on_fork;
 205         bool non_blocking;
 206
 207         mode_t umask;
 208         int oom_score_adjust;
 209         int nice;
 210         int ioprio;
 211         int cpu_sched_policy;
 212         int cpu_sched_priority;
 213         uint64_t coredump_filter;
 214
 215         CPUSet cpu_set;
 216         NUMAPolicy numa_policy;
 217         bool cpu_affinity_from_numa;
 218
 219         ExecInput std_input;
 220         ExecOutput std_output;
 221         ExecOutput std_error;
 222         bool stdio_as_fds;
 223         char *stdio_fdname[3];
 224         char *stdio_file[3];
 225
 226         void *stdin_data;
 227         size_t stdin_data_size;
 228
 229         nsec_t timer_slack_nsec;
 230
 231         char *tty_path;
 232
 233         bool tty_reset;
 234         bool tty_vhangup;
 235         bool tty_vt_disallocate;
 236
 237         unsigned tty_rows;
 238         unsigned tty_cols;
 239
 240         bool ignore_sigpipe;
 241
 242         ExecKeyringMode keyring_mode;
 243
 244         /* Since resolving these names might involve socket
 245          * connections and we don't want to deadlock ourselves these
 246          * names are resolved on execution only and in the child
 247          * process. */
 248         char *user;
 249         char *group;
 250         char **supplementary_groups;
 251
 252         char *pam_name;
 253
 254         char *utmp_id;
 255         ExecUtmpMode utmp_mode;
 256
 257         bool no_new_privileges;
 258
 259         bool selinux_context_ignore;
 260         bool apparmor_profile_ignore;
 261         bool smack_process_label_ignore;
 262
 263         char *selinux_context;
 264         char *apparmor_profile;
 265         char *smack_process_label;
 266
 267         char **read_write_paths, **read_only_paths, **inaccessible_paths, **exec_paths, **no_exec_paths;
 268         char **exec_search_path;
 269         unsigned long mount_flags;
 270         BindMount *bind_mounts;
 271         size_t n_bind_mounts;
 272         TemporaryFileSystem *temporary_filesystems;
 273         size_t n_temporary_filesystems;
 274         MountImage *mount_images;
 275         size_t n_mount_images;
 276         MountImage *extension_images;
 277         size_t n_extension_images;
 278         char **extension_directories;
 279
 280         uint64_t capability_bounding_set;
 281         uint64_t capability_ambient_set;
 282         int secure_bits;
 283
 284         int syslog_priority;
 285         bool syslog_level_prefix;
 286         char *syslog_identifier;
 287
 288         struct iovec* log_extra_fields;
 289         size_t n_log_extra_fields;
 290         Set *log_filter_allowed_patterns;
 291         Set *log_filter_denied_patterns;
 292
 293         usec_t log_ratelimit_interval_usec;
 294         unsigned log_ratelimit_burst;
 295
 296         int log_level_max;
 297
 298         char *log_namespace;
 299
 300         ProtectProc protect_proc;  /* hidepid= */
 301         ProcSubset proc_subset;    /* subset= */
 302
 303         bool private_tmp;
 304         bool private_network;
 305         bool private_devices;
 306         bool private_users;
 307         bool private_mounts;
 308         bool private_ipc;
 309         bool protect_kernel_tunables;
 310         bool protect_kernel_modules;
 311         bool protect_kernel_logs;
 312         bool protect_clock;
 313         bool protect_control_groups;
 314         ProtectSystem protect_system;
 315         ProtectHome protect_home;
 316         bool protect_hostname;
 317         bool mount_apivfs;
 318
 319         bool dynamic_user;
 320         bool remove_ipc;
 321
 322         bool memory_deny_write_execute;
 323         bool restrict_realtime;
 324         bool restrict_suid_sgid;
 325
 326         bool lock_personality;
 327         unsigned long personality;
 328
 329         unsigned long restrict_namespaces; /* The CLONE_NEWxyz flags permitted to the unit's processes */
 330
 331         Set *restrict_filesystems;
 332         bool restrict_filesystems_allow_list:1;
 333
 334         Hashmap *syscall_filter;
 335         Set *syscall_archs;
 336         int syscall_errno;
 337         bool syscall_allow_list:1;
 338
 339         Hashmap *syscall_log;
 340         bool syscall_log_allow_list:1; /* Log listed system calls */
 341
 342         bool address_families_allow_list:1;
 343         Set *address_families;
 344
 345         char *network_namespace_path;
 346         char *ipc_namespace_path;
 347
 348         ExecDirectory directories[_EXEC_DIRECTORY_TYPE_MAX];
 349         ExecPreserveMode runtime_directory_preserve_mode;
 350         usec_t timeout_clean_usec;
 351
 352         Hashmap *set_credentials; /* output id → ExecSetCredential */
 353         Hashmap *load_credentials; /* output id → ExecLoadCredential */
 354 };
 355
 356 static inline bool exec_context_restrict_namespaces_set(const ExecContext *c) {
 357         assert(c);
 358
 359         return (c->restrict_namespaces & NAMESPACE_FLAGS_ALL) != NAMESPACE_FLAGS_ALL;
 360 }
 361
 362 static inline bool exec_context_restrict_filesystems_set(const ExecContext *c) {
 363         assert(c);
 364
 365         return c->restrict_filesystems_allow_list ||
 366           !set_isempty(c->restrict_filesystems);
 367 }
 368
 369 static inline bool exec_context_with_rootfs(const ExecContext *c) {
 370         assert(c);
 371
 372         /* Checks if RootDirectory= or RootImage= are used */
 373
 374         return !empty_or_root(c->root_directory) || c->root_image;
 375 }
 376
 377 typedef enum ExecFlags {
 378         EXEC_APPLY_SANDBOXING      = 1 << 0,
 379         EXEC_APPLY_CHROOT          = 1 << 1,
 380         EXEC_APPLY_TTY_STDIN       = 1 << 2,
 381         EXEC_PASS_LOG_UNIT         = 1 << 3, /* Whether to pass the unit name to the service's journal stream connection */
 382         EXEC_CHOWN_DIRECTORIES     = 1 << 4, /* chown() the runtime/state/cache/log directories to the user we run as, under all conditions */
 383         EXEC_NSS_DYNAMIC_BYPASS    = 1 << 5, /* Set the SYSTEMD_NSS_DYNAMIC_BYPASS environment variable, to disable nss-systemd blocking on PID 1, for use by dbus-daemon */
 384         EXEC_CGROUP_DELEGATE       = 1 << 6,
 385         EXEC_IS_CONTROL            = 1 << 7,
 386         EXEC_CONTROL_CGROUP        = 1 << 8, /* Place the process not in the indicated cgroup but in a subcgroup '/.control', but only EXEC_CGROUP_DELEGATE and EXEC_IS_CONTROL is set, too */
 387         EXEC_WRITE_CREDENTIALS     = 1 << 9, /* Set up the credential store logic */
 388
 389         /* The following are not used by execute.c, but by consumers internally */
 390         EXEC_PASS_FDS              = 1 << 10,
 391         EXEC_SETENV_RESULT         = 1 << 11,
 392         EXEC_SET_WATCHDOG          = 1 << 12,
 393         EXEC_SETENV_MONITOR_RESULT = 1 << 13, /* Pass exit status to OnFailure= and OnSuccess= dependencies. */
 394 } ExecFlags;
 395
 396 /* Parameters for a specific invocation of a command. This structure is put together right before a command is
 397  * executed. */
 398 struct ExecParameters {
 399         char **environment;
 400
 401         int *fds;
 402         char **fd_names;
 403         size_t n_socket_fds;
 404         size_t n_storage_fds;
 405
 406         ExecFlags flags;
 407         bool selinux_context_net:1;
 408
 409         CGroupMask cgroup_supported;
 410         const char *cgroup_path;
 411
 412         char **prefix;
 413         const char *received_credentials_directory;
 414         const char *received_encrypted_credentials_directory;
 415
 416         const char *confirm_spawn;
 417
 418         usec_t watchdog_usec;
 419
 420         int *idle_pipe;
 421
 422         int stdin_fd;
 423         int stdout_fd;
 424         int stderr_fd;
 425
 426         /* An fd that is closed by the execve(), and thus will result in EOF when the execve() is done */
 427         int exec_fd;
 428
 429         const char *notify_socket;
 430 };
 431
 432 #include "unit.h"
 433 #include "dynamic-user.h"
 434
 435 int exec_spawn(Unit *unit,
 436                ExecCommand *command,
 437                const ExecContext *context,
 438                const ExecParameters *exec_params,
 439                ExecRuntime *runtime,
 440                DynamicCreds *dynamic_creds,
 441                pid_t *ret);
 442
 443 void exec_command_done_array(ExecCommand *c, size_t n);
 444 ExecCommand* exec_command_free_list(ExecCommand *c);
 445 void exec_command_free_array(ExecCommand **c, size_t n);
 446 void exec_command_reset_status_array(ExecCommand *c, size_t n);
 447 void exec_command_reset_status_list_array(ExecCommand **c, size_t n);
 448 void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix);
 449 void exec_command_append_list(ExecCommand **l, ExecCommand *e);
 450 int exec_command_set(ExecCommand *c, const char *path, ...) _sentinel_;
 451 int exec_command_append(ExecCommand *c, const char *path, ...) _sentinel_;
 452
 453 void exec_context_init(ExecContext *c);
 454 void exec_context_done(ExecContext *c);
 455 void exec_context_dump(const ExecContext *c, FILE* f, const char *prefix);
 456
 457 int exec_context_destroy_runtime_directory(const ExecContext *c, const char *runtime_root);
 458 int exec_context_destroy_credentials(const ExecContext *c, const char *runtime_root, const char *unit);
 459
 460 const char* exec_context_fdname(const ExecContext *c, int fd_index);
 461
 462 bool exec_context_may_touch_console(const ExecContext *c);
 463 bool exec_context_maintains_privileges(const ExecContext *c);
 464
 465 int exec_context_get_effective_ioprio(const ExecContext *c);
 466 bool exec_context_get_effective_mount_apivfs(const ExecContext *c);
 467
 468 void exec_context_free_log_extra_fields(ExecContext *c);
 469
 470 void exec_context_revert_tty(ExecContext *c);
 471
 472 int exec_context_get_clean_directories(ExecContext *c, char **prefix, ExecCleanMask mask, char ***ret);
 473 int exec_context_get_clean_mask(ExecContext *c, ExecCleanMask *ret);
 474
 475 void exec_status_start(ExecStatus *s, pid_t pid);
 476 void exec_status_exit(ExecStatus *s, const ExecContext *context, pid_t pid, int code, int status);
 477 void exec_status_dump(const ExecStatus *s, FILE *f, const char *prefix);
 478 void exec_status_reset(ExecStatus *s);
 479
 480 int exec_runtime_acquire(Manager *m, const ExecContext *c, const char *name, bool create, ExecRuntime **ret);
 481 ExecRuntime *exec_runtime_unref(ExecRuntime *r, bool destroy);
 482
 483 int exec_runtime_serialize(const Manager *m, FILE *f, FDSet *fds);
 484 int exec_runtime_deserialize_compat(Unit *u, const char *key, const char *value, FDSet *fds);
 485 int exec_runtime_deserialize_one(Manager *m, const char *value, FDSet *fds);
 486 void exec_runtime_vacuum(Manager *m);
 487
 488 void exec_params_clear(ExecParameters *p);
 489
 490 bool exec_context_get_cpu_affinity_from_numa(const ExecContext *c);
 491
 492 ExecSetCredential *exec_set_credential_free(ExecSetCredential *sc);
 493 DEFINE_TRIVIAL_CLEANUP_FUNC(ExecSetCredential*, exec_set_credential_free);
 494
 495 ExecLoadCredential *exec_load_credential_free(ExecLoadCredential *lc);
 496 DEFINE_TRIVIAL_CLEANUP_FUNC(ExecLoadCredential*, exec_load_credential_free);
 497
 498 void exec_directory_done(ExecDirectory *d);
 499 int exec_directory_add(ExecDirectory *d, const char *path, const char *symlink);
 500 void exec_directory_sort(ExecDirectory *d);
 501
 502 extern const struct hash_ops exec_set_credential_hash_ops;
 503 extern const struct hash_ops exec_load_credential_hash_ops;
 504
 505 const char* exec_output_to_string(ExecOutput i) _const_;
 506 ExecOutput exec_output_from_string(const char *s) _pure_;
 507
 508 const char* exec_input_to_string(ExecInput i) _const_;
 509 ExecInput exec_input_from_string(const char *s) _pure_;
 510
 511 const char* exec_utmp_mode_to_string(ExecUtmpMode i) _const_;
 512 ExecUtmpMode exec_utmp_mode_from_string(const char *s) _pure_;
 513
 514 const char* exec_preserve_mode_to_string(ExecPreserveMode i) _const_;
 515 ExecPreserveMode exec_preserve_mode_from_string(const char *s) _pure_;
 516
 517 const char* exec_keyring_mode_to_string(ExecKeyringMode i) _const_;
 518 ExecKeyringMode exec_keyring_mode_from_string(const char *s) _pure_;
 519
 520 const char* exec_directory_type_to_string(ExecDirectoryType i) _const_;
 521 ExecDirectoryType exec_directory_type_from_string(const char *s) _pure_;
 522
 523 const char* exec_directory_type_symlink_to_string(ExecDirectoryType i) _const_;
 524 ExecDirectoryType exec_directory_type_symlink_from_string(const char *s) _pure_;
 525
 526 const char* exec_resource_type_to_string(ExecDirectoryType i) _const_;
 527 ExecDirectoryType exec_resource_type_from_string(const char *s) _pure_;
 528
 529 bool exec_needs_mount_namespace(const ExecContext *context, const ExecParameters *params, const ExecRuntime *runtime);