1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include "sd-messages.h"
9 #include "alloc-util.h"
10 #include "bus-error.h"
13 #include "errno-util.h"
15 #include "extract-word.h"
18 #include "format-util.h"
20 #include "machine-dbus.h"
22 #include "mkdir-label.h"
23 #include "parse-util.h"
24 #include "path-util.h"
25 #include "process-util.h"
26 #include "serialize.h"
28 #include "stdio-util.h"
29 #include "string-table.h"
30 #include "terminal-util.h"
31 #include "tmpfile-util.h"
32 #include "unit-name.h"
33 #include "user-util.h"
36 Machine
* machine_new(Manager
*manager
, MachineClass
class, const char *name
) {
40 assert(class < _MACHINE_CLASS_MAX
);
43 /* Passing class == _MACHINE_CLASS_INVALID here is fine. It
44 * means as much as "we don't know yet", and that we'll figure
45 * it out later when loading the state file. */
51 m
->name
= strdup(name
);
55 if (class != MACHINE_HOST
) {
56 m
->state_file
= path_join("/run/systemd/machines", m
->name
);
63 if (hashmap_put(manager
->machines
, m
->name
, m
) < 0)
76 Machine
* machine_free(Machine
*m
) {
81 operation_free(m
->operations
);
84 LIST_REMOVE(gc_queue
, m
->manager
->machine_gc_queue
, m
);
86 machine_release_unit(m
);
90 (void) hashmap_remove(m
->manager
->machines
, m
->name
);
92 if (m
->manager
->host_machine
== m
)
93 m
->manager
->host_machine
= NULL
;
96 (void) hashmap_remove_value(m
->manager
->machine_leaders
, PID_TO_PTR(m
->leader
), m
);
98 sd_bus_message_unref(m
->create_message
);
103 free(m
->root_directory
);
108 int machine_save(Machine
*m
) {
109 _cleanup_free_
char *temp_path
= NULL
;
110 _cleanup_fclose_
FILE *f
= NULL
;
121 r
= mkdir_safe_label("/run/systemd/machines", 0755, 0, 0, MKDIR_WARN_MODE
);
125 r
= fopen_temporary(m
->state_file
, &f
, &temp_path
);
129 (void) fchmod(fileno(f
), 0644);
132 "# This is private data. Do not parse.\n"
137 _cleanup_free_
char *escaped
= NULL
;
139 escaped
= cescape(m
->unit
);
145 fprintf(f
, "SCOPE=%s\n", escaped
); /* We continue to call this "SCOPE=" because it is internal only, and we want to stay compatible with old files */
149 fprintf(f
, "SCOPE_JOB=%s\n", m
->scope_job
);
152 _cleanup_free_
char *escaped
= NULL
;
154 escaped
= cescape(m
->service
);
159 fprintf(f
, "SERVICE=%s\n", escaped
);
162 if (m
->root_directory
) {
163 _cleanup_free_
char *escaped
= NULL
;
165 escaped
= cescape(m
->root_directory
);
170 fprintf(f
, "ROOT=%s\n", escaped
);
173 if (!sd_id128_is_null(m
->id
))
174 fprintf(f
, "ID=" SD_ID128_FORMAT_STR
"\n", SD_ID128_FORMAT_VAL(m
->id
));
177 fprintf(f
, "LEADER="PID_FMT
"\n", m
->leader
);
179 if (m
->class != _MACHINE_CLASS_INVALID
)
180 fprintf(f
, "CLASS=%s\n", machine_class_to_string(m
->class));
182 if (dual_timestamp_is_set(&m
->timestamp
))
184 "REALTIME="USEC_FMT
"\n"
185 "MONOTONIC="USEC_FMT
"\n",
186 m
->timestamp
.realtime
,
187 m
->timestamp
.monotonic
);
189 if (m
->n_netif
> 0) {
194 for (i
= 0; i
< m
->n_netif
; i
++) {
198 fprintf(f
, "%i", m
->netif
[i
]);
204 r
= fflush_and_check(f
);
208 if (rename(temp_path
, m
->state_file
) < 0) {
216 /* Create a symlink from the unit name to the machine
217 * name, so that we can quickly find the machine for
218 * each given unit. Ignore error. */
219 sl
= strjoina("/run/systemd/machines/unit:", m
->unit
);
220 (void) symlink(m
->name
, sl
);
226 (void) unlink(m
->state_file
);
229 (void) unlink(temp_path
);
231 return log_error_errno(r
, "Failed to save machine data %s: %m", m
->state_file
);
234 static void machine_unlink(Machine
*m
) {
240 sl
= strjoina("/run/systemd/machines/unit:", m
->unit
);
245 (void) unlink(m
->state_file
);
248 int machine_load(Machine
*m
) {
249 _cleanup_free_
char *realtime
= NULL
, *monotonic
= NULL
, *id
= NULL
, *leader
= NULL
, *class = NULL
, *netif
= NULL
;
257 r
= parse_env_file(NULL
, m
->state_file
,
259 "SCOPE_JOB", &m
->scope_job
,
260 "SERVICE", &m
->service
,
261 "ROOT", &m
->root_directory
,
265 "REALTIME", &realtime
,
266 "MONOTONIC", &monotonic
,
271 return log_error_errno(r
, "Failed to read %s: %m", m
->state_file
);
274 sd_id128_from_string(id
, &m
->id
);
277 parse_pid(leader
, &m
->leader
);
282 c
= machine_class_from_string(class);
288 (void) deserialize_usec(realtime
, &m
->timestamp
.realtime
);
290 (void) deserialize_usec(monotonic
, &m
->timestamp
.monotonic
);
293 _cleanup_free_
int *ni
= NULL
;
299 _cleanup_free_
char *word
= NULL
;
301 r
= extract_first_word(&p
, &word
, NULL
, 0);
307 log_warning_errno(r
, "Failed to parse NETIF: %s", netif
);
311 r
= parse_ifindex(word
);
315 if (!GREEDY_REALLOC(ni
, nr
+ 1))
321 free_and_replace(m
->netif
, ni
);
328 static int machine_start_scope(
330 sd_bus_message
*more_properties
,
331 sd_bus_error
*error
) {
333 _cleanup_(sd_bus_message_unrefp
) sd_bus_message
*m
= NULL
, *reply
= NULL
;
334 _cleanup_free_
char *escaped
= NULL
, *unit
= NULL
;
335 const char *description
;
339 assert(machine
->leader
> 0);
340 assert(!machine
->unit
);
342 escaped
= unit_name_escape(machine
->name
);
346 unit
= strjoin("machine-", escaped
, ".scope");
350 r
= sd_bus_message_new_method_call(
351 machine
->manager
->bus
,
353 "org.freedesktop.systemd1",
354 "/org/freedesktop/systemd1",
355 "org.freedesktop.systemd1.Manager",
356 "StartTransientUnit");
360 r
= sd_bus_message_append(m
, "ss", unit
, "fail");
364 r
= sd_bus_message_open_container(m
, 'a', "(sv)");
368 r
= sd_bus_message_append(m
, "(sv)", "Slice", "s", SPECIAL_MACHINE_SLICE
);
372 description
= strjoina(machine
->class == MACHINE_VM
? "Virtual Machine " : "Container ", machine
->name
);
373 r
= sd_bus_message_append(m
, "(sv)", "Description", "s", description
);
377 r
= sd_bus_message_append(m
, "(sv)(sv)(sv)(sv)(sv)",
378 "PIDs", "au", 1, machine
->leader
,
380 "CollectMode", "s", "inactive-or-failed",
382 "TasksMax", "t", UINT64_C(16384));
386 if (more_properties
) {
387 r
= sd_bus_message_copy(m
, more_properties
, true);
392 r
= sd_bus_message_close_container(m
);
396 r
= sd_bus_message_append(m
, "a(sa(sv))", 0);
400 r
= sd_bus_call(NULL
, m
, 0, error
, &reply
);
404 machine
->unit
= TAKE_PTR(unit
);
405 machine
->referenced
= true;
408 r
= sd_bus_message_read(reply
, "o", &job
);
412 return free_and_strdup(&machine
->scope_job
, job
);
415 static int machine_ensure_scope(Machine
*m
, sd_bus_message
*properties
, sd_bus_error
*error
) {
419 assert(m
->class != MACHINE_HOST
);
422 r
= machine_start_scope(m
, properties
, error
);
424 return log_error_errno(r
, "Failed to start machine scope: %s", bus_error_message(error
, r
));
428 hashmap_put(m
->manager
->machine_units
, m
->unit
, m
);
433 int machine_start(Machine
*m
, sd_bus_message
*properties
, sd_bus_error
*error
) {
438 if (!IN_SET(m
->class, MACHINE_CONTAINER
, MACHINE_VM
))
444 r
= hashmap_put(m
->manager
->machine_leaders
, PID_TO_PTR(m
->leader
), m
);
449 r
= machine_ensure_scope(m
, properties
, error
);
454 "MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR
,
456 "LEADER="PID_FMT
, m
->leader
,
457 LOG_MESSAGE("New machine %s.", m
->name
));
459 if (!dual_timestamp_is_set(&m
->timestamp
))
460 dual_timestamp_get(&m
->timestamp
);
464 /* Save new machine data */
467 machine_send_signal(m
, true);
468 (void) manager_enqueue_nscd_cache_flush(m
->manager
);
473 int machine_stop(Machine
*m
) {
478 if (!IN_SET(m
->class, MACHINE_CONTAINER
, MACHINE_VM
))
482 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
485 r
= manager_stop_unit(m
->manager
, m
->unit
, &error
, &job
);
487 return log_error_errno(r
, "Failed to stop machine scope: %s", bus_error_message(&error
, r
));
489 free_and_replace(m
->scope_job
, job
);
495 (void) manager_enqueue_nscd_cache_flush(m
->manager
);
500 int machine_finalize(Machine
*m
) {
505 "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR
,
507 "LEADER="PID_FMT
, m
->leader
,
508 LOG_MESSAGE("Machine %s terminated.", m
->name
));
510 m
->stopping
= true; /* The machine is supposed to be going away. Don't try to kill it. */
514 machine_add_to_gc_queue(m
);
517 machine_send_signal(m
, false);
524 bool machine_may_gc(Machine
*m
, bool drop_not_started
) {
527 if (m
->class == MACHINE_HOST
)
530 if (drop_not_started
&& !m
->started
)
533 if (m
->scope_job
&& manager_job_is_active(m
->manager
, m
->scope_job
))
536 if (m
->unit
&& manager_unit_is_active(m
->manager
, m
->unit
))
542 void machine_add_to_gc_queue(Machine
*m
) {
548 LIST_PREPEND(gc_queue
, m
->manager
->machine_gc_queue
, m
);
549 m
->in_gc_queue
= true;
552 MachineState
machine_get_state(Machine
*s
) {
555 if (s
->class == MACHINE_HOST
)
556 return MACHINE_RUNNING
;
559 return MACHINE_CLOSING
;
562 return MACHINE_OPENING
;
564 return MACHINE_RUNNING
;
567 int machine_kill(Machine
*m
, KillWho who
, int signo
) {
570 if (!IN_SET(m
->class, MACHINE_VM
, MACHINE_CONTAINER
))
576 if (who
== KILL_LEADER
) /* If we shall simply kill the leader, do so directly */
577 return RET_NERRNO(kill(m
->leader
, signo
));
579 /* Otherwise, make PID 1 do it for us, for the entire cgroup */
580 return manager_kill_unit(m
->manager
, m
->unit
, signo
, NULL
);
583 int machine_openpt(Machine
*m
, int flags
, char **ret_slave
) {
590 return openpt_allocate(flags
, ret_slave
);
592 case MACHINE_CONTAINER
:
596 return openpt_allocate_in_namespace(m
->leader
, flags
, ret_slave
);
603 int machine_open_terminal(Machine
*m
, const char *path
, int mode
) {
609 return open_terminal(path
, mode
);
611 case MACHINE_CONTAINER
:
615 return open_terminal_in_namespace(m
->leader
, path
, mode
);
622 void machine_release_unit(Machine
*m
) {
629 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
632 r
= manager_unref_unit(m
->manager
, m
->unit
, &error
);
634 log_warning_errno(r
, "Failed to drop reference to machine scope, ignoring: %s",
635 bus_error_message(&error
, r
));
637 m
->referenced
= false;
640 (void) hashmap_remove(m
->manager
->machine_units
, m
->unit
);
641 m
->unit
= mfree(m
->unit
);
644 int machine_get_uid_shift(Machine
*m
, uid_t
*ret
) {
645 char p
[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t
) + 1];
646 uid_t uid_base
, uid_shift
, uid_range
;
647 gid_t gid_base
, gid_shift
, gid_range
;
648 _cleanup_fclose_
FILE *f
= NULL
;
654 /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
655 * mappings. In most cases setups should be simple like this, and administrators should only care about the
656 * basic offset a container has relative to the host. This is what this function exposes.
658 * If we encounter any more complex mappings we politely refuse this with ENXIO. */
660 if (m
->class == MACHINE_HOST
) {
665 if (m
->class != MACHINE_CONTAINER
)
668 xsprintf(p
, "/proc/" PID_FMT
"/uid_map", m
->leader
);
671 if (errno
== ENOENT
) {
672 /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
680 /* Read the first line. There's at least one. */
682 k
= fscanf(f
, UID_FMT
" " UID_FMT
" " UID_FMT
"\n", &uid_base
, &uid_shift
, &uid_range
);
685 return errno_or_else(EIO
);
690 /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
693 /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
694 if (uid_range
< UID_NOBODY
)
697 /* If there's more than one line, then we don't support this mapping. */
698 r
= safe_fgetc(f
, NULL
);
701 if (r
!= 0) /* Insist on EOF */
706 xsprintf(p
, "/proc/" PID_FMT
"/gid_map", m
->leader
);
711 /* Read the first line. There's at least one. */
713 k
= fscanf(f
, GID_FMT
" " GID_FMT
" " GID_FMT
"\n", &gid_base
, &gid_shift
, &gid_range
);
716 return errno_or_else(EIO
);
721 /* If there's more than one line, then we don't support this file. */
722 r
= safe_fgetc(f
, NULL
);
725 if (r
!= 0) /* Insist on EOF */
728 /* If the UID and GID mapping doesn't match, we don't support this mapping. */
729 if (uid_base
!= (uid_t
) gid_base
)
731 if (uid_shift
!= (uid_t
) gid_shift
)
733 if (uid_range
!= (uid_t
) gid_range
)
740 static int machine_owns_uid_internal(
742 const char *map_file
, /* "uid_map" or "gid_map" */
744 uid_t
*ret_internal_uid
) {
746 _cleanup_fclose_
FILE *f
= NULL
;
749 /* This is a generic implementation for both uids and gids, under the assumptions they have the same types and semantics. */
750 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
754 /* Checks if the specified host UID is owned by the machine, and returns the UID it maps to
755 * internally in the machine */
757 if (machine
->class != MACHINE_CONTAINER
)
760 p
= procfs_file_alloca(machine
->leader
, map_file
);
763 log_debug_errno(errno
, "Failed to open %s, ignoring.", p
);
768 uid_t uid_base
, uid_shift
, uid_range
, converted
;
772 k
= fscanf(f
, UID_FMT
" " UID_FMT
" " UID_FMT
, &uid_base
, &uid_shift
, &uid_range
);
773 if (k
< 0 && feof(f
))
777 return errno_or_else(EIO
);
782 /* The private user namespace is disabled, ignoring. */
786 if (uid
< uid_shift
|| uid
>= uid_shift
+ uid_range
)
789 converted
= (uid
- uid_shift
+ uid_base
);
790 if (!uid_is_valid(converted
))
793 if (ret_internal_uid
)
794 *ret_internal_uid
= converted
;
800 if (ret_internal_uid
)
801 *ret_internal_uid
= UID_INVALID
;
806 int machine_owns_uid(Machine
*machine
, uid_t uid
, uid_t
*ret_internal_uid
) {
807 return machine_owns_uid_internal(machine
, "uid_map", uid
, ret_internal_uid
);
810 int machine_owns_gid(Machine
*machine
, gid_t gid
, gid_t
*ret_internal_gid
) {
811 return machine_owns_uid_internal(machine
, "gid_map", (uid_t
) gid
, (uid_t
*) ret_internal_gid
);
814 static int machine_translate_uid_internal(
816 const char *map_file
, /* "uid_map" or "gid_map" */
818 uid_t
*ret_host_uid
) {
820 _cleanup_fclose_
FILE *f
= NULL
;
823 /* This is a generic implementation for both uids and gids, under the assumptions they have the same types and semantics. */
824 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
827 assert(uid_is_valid(uid
));
829 if (machine
->class != MACHINE_CONTAINER
)
832 /* Translates a machine UID into a host UID */
834 p
= procfs_file_alloca(machine
->leader
, map_file
);
840 uid_t uid_base
, uid_shift
, uid_range
, converted
;
844 k
= fscanf(f
, UID_FMT
" " UID_FMT
" " UID_FMT
, &uid_base
, &uid_shift
, &uid_range
);
845 if (k
< 0 && feof(f
))
849 return errno_or_else(EIO
);
854 if (uid
< uid_base
|| uid
>= uid_base
+ uid_range
)
857 converted
= uid
- uid_base
+ uid_shift
;
858 if (!uid_is_valid(converted
))
862 *ret_host_uid
= converted
;
869 int machine_translate_uid(Machine
*machine
, gid_t uid
, gid_t
*ret_host_uid
) {
870 return machine_translate_uid_internal(machine
, "uid_map", uid
, ret_host_uid
);
873 int machine_translate_gid(Machine
*machine
, gid_t gid
, gid_t
*ret_host_gid
) {
874 return machine_translate_uid_internal(machine
, "gid_map", (uid_t
) gid
, (uid_t
*) ret_host_gid
);
877 static const char* const machine_class_table
[_MACHINE_CLASS_MAX
] = {
878 [MACHINE_CONTAINER
] = "container",
880 [MACHINE_HOST
] = "host",
883 DEFINE_STRING_TABLE_LOOKUP(machine_class
, MachineClass
);
885 static const char* const machine_state_table
[_MACHINE_STATE_MAX
] = {
886 [MACHINE_OPENING
] = "opening",
887 [MACHINE_RUNNING
] = "running",
888 [MACHINE_CLOSING
] = "closing"
891 DEFINE_STRING_TABLE_LOOKUP(machine_state
, MachineState
);
893 static const char* const kill_who_table
[_KILL_WHO_MAX
] = {
894 [KILL_LEADER
] = "leader",
898 DEFINE_STRING_TABLE_LOOKUP(kill_who
, KillWho
);