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