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