src/core/execute.h

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