[thirdparty/systemd.git] / src / machine / machine.c

/* SPDX-License-Identifier: LGPL-2.1+ */

#include <errno.h>
#include <string.h>
#include <unistd.h>
#include <stdio_ext.h>

#include "sd-messages.h"

#include "alloc-util.h"
#include "bus-error.h"
#include "bus-util.h"
#include "escape.h"
#include "extract-word.h"
#include "fd-util.h"
#include "fileio.h"
#include "format-util.h"
#include "hashmap.h"
#include "machine-dbus.h"
#include "machine.h"
#include "mkdir.h"
#include "parse-util.h"
#include "process-util.h"
#include "serialize.h"
#include "special.h"
#include "stdio-util.h"
#include "string-table.h"
#include "terminal-util.h"
#include "unit-name.h"
#include "user-util.h"
#include "util.h"

Machine* machine_new(Manager *manager, MachineClass class, const char *name) {
        Machine *m;

        assert(manager);
        assert(class < _MACHINE_CLASS_MAX);
        assert(name);

        /* Passing class == _MACHINE_CLASS_INVALID here is fine. It
         * means as much as "we don't know yet", and that we'll figure
         * it out later when loading the state file. */

        m = new0(Machine, 1);
        if (!m)
                return NULL;

        m->name = strdup(name);
        if (!m->name)
                goto fail;

        if (class != MACHINE_HOST) {
                m->state_file = strappend("/run/systemd/machines/", m->name);
                if (!m->state_file)
                        goto fail;
        }

        m->class = class;

        if (hashmap_put(manager->machines, m->name, m) < 0)
                goto fail;

        m->manager = manager;

        return m;

fail:
        free(m->state_file);
        free(m->name);
        return mfree(m);
}

void machine_free(Machine *m) {
        assert(m);

        while (m->operations)
                operation_free(m->operations);

        if (m->in_gc_queue)
                LIST_REMOVE(gc_queue, m->manager->machine_gc_queue, m);

        machine_release_unit(m);

        free(m->scope_job);

        (void) hashmap_remove(m->manager->machines, m->name);

        if (m->manager->host_machine == m)
                m->manager->host_machine = NULL;

        if (m->leader > 0)
                (void) hashmap_remove_value(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);

        sd_bus_message_unref(m->create_message);

        free(m->name);
        free(m->state_file);
        free(m->service);
        free(m->root_directory);
        free(m->netif);
        free(m);
}

int machine_save(Machine *m) {
        _cleanup_free_ char *temp_path = NULL;
        _cleanup_fclose_ FILE *f = NULL;
        int r;

        assert(m);

        if (!m->state_file)
                return 0;

        if (!m->started)
                return 0;

        r = mkdir_safe_label("/run/systemd/machines", 0755, 0, 0, MKDIR_WARN_MODE);
        if (r < 0)
                goto fail;

        r = fopen_temporary(m->state_file, &f, &temp_path);
        if (r < 0)
                goto fail;

        (void) __fsetlocking(f, FSETLOCKING_BYCALLER);
        (void) fchmod(fileno(f), 0644);

        fprintf(f,
                "# This is private data. Do not parse.\n"
                "NAME=%s\n",
                m->name);

        if (m->unit) {
                _cleanup_free_ char *escaped;

                escaped = cescape(m->unit);
                if (!escaped) {
                        r = -ENOMEM;
                        goto fail;
                }

                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 */
        }

        if (m->scope_job)
                fprintf(f, "SCOPE_JOB=%s\n", m->scope_job);

        if (m->service) {
                _cleanup_free_ char *escaped;

                escaped = cescape(m->service);
                if (!escaped) {
                        r = -ENOMEM;
                        goto fail;
                }
                fprintf(f, "SERVICE=%s\n", escaped);
        }

        if (m->root_directory) {
                _cleanup_free_ char *escaped;

                escaped = cescape(m->root_directory);
                if (!escaped) {
                        r = -ENOMEM;
                        goto fail;
                }
                fprintf(f, "ROOT=%s\n", escaped);
        }

        if (!sd_id128_is_null(m->id))
                fprintf(f, "ID=" SD_ID128_FORMAT_STR "\n", SD_ID128_FORMAT_VAL(m->id));

        if (m->leader != 0)
                fprintf(f, "LEADER="PID_FMT"\n", m->leader);

        if (m->class != _MACHINE_CLASS_INVALID)
                fprintf(f, "CLASS=%s\n", machine_class_to_string(m->class));

        if (dual_timestamp_is_set(&m->timestamp))
                fprintf(f,
                        "REALTIME="USEC_FMT"\n"
                        "MONOTONIC="USEC_FMT"\n",
                        m->timestamp.realtime,
                        m->timestamp.monotonic);

        if (m->n_netif > 0) {
                size_t i;

                fputs("NETIF=", f);

                for (i = 0; i < m->n_netif; i++) {
                        if (i != 0)
                                fputc(' ', f);

                        fprintf(f, "%i", m->netif[i]);
                }

                fputc('\n', f);
        }

        r = fflush_and_check(f);
        if (r < 0)
                goto fail;

        if (rename(temp_path, m->state_file) < 0) {
                r = -errno;
                goto fail;
        }

        if (m->unit) {
                char *sl;

                /* Create a symlink from the unit name to the machine
                 * name, so that we can quickly find the machine for
                 * each given unit. Ignore error. */
                sl = strjoina("/run/systemd/machines/unit:", m->unit);
                (void) symlink(m->name, sl);
        }

        return 0;

fail:
        (void) unlink(m->state_file);

        if (temp_path)
                (void) unlink(temp_path);

        return log_error_errno(r, "Failed to save machine data %s: %m", m->state_file);
}

static void machine_unlink(Machine *m) {
        assert(m);

        if (m->unit) {
                char *sl;

                sl = strjoina("/run/systemd/machines/unit:", m->unit);
                (void) unlink(sl);
        }

        if (m->state_file)
                (void) unlink(m->state_file);
}

int machine_load(Machine *m) {
        _cleanup_free_ char *realtime = NULL, *monotonic = NULL, *id = NULL, *leader = NULL, *class = NULL, *netif = NULL;
        int r;

        assert(m);

        if (!m->state_file)
                return 0;

        r = parse_env_file(NULL, m->state_file, NEWLINE,
                           "SCOPE",     &m->unit,
                           "SCOPE_JOB", &m->scope_job,
                           "SERVICE",   &m->service,
                           "ROOT",      &m->root_directory,
                           "ID",        &id,
                           "LEADER",    &leader,
                           "CLASS",     &class,
                           "REALTIME",  &realtime,
                           "MONOTONIC", &monotonic,
                           "NETIF",     &netif,
                           NULL);
        if (r < 0) {
                if (r == -ENOENT)
                        return 0;

                return log_error_errno(r, "Failed to read %s: %m", m->state_file);
        }

        if (id)
                sd_id128_from_string(id, &m->id);

        if (leader)
                parse_pid(leader, &m->leader);

        if (class) {
                MachineClass c;

                c = machine_class_from_string(class);
                if (c >= 0)
                        m->class = c;
        }

        if (realtime)
                (void) deserialize_usec(realtime, &m->timestamp.realtime);
        if (monotonic)
                (void) deserialize_usec(monotonic, &m->timestamp.monotonic);

        if (netif) {
                size_t allocated = 0, nr = 0;
                const char *p;
                int *ni = NULL;

                p = netif;
                for (;;) {
                        _cleanup_free_ char *word = NULL;
                        int ifi;

                        r = extract_first_word(&p, &word, NULL, 0);
                        if (r == 0)
                                break;
                        if (r == -ENOMEM)
                                return log_oom();
                        if (r < 0) {
                                log_warning_errno(r, "Failed to parse NETIF: %s", netif);
                                break;
                        }

                        if (parse_ifindex(word, &ifi) < 0)
                                continue;

                        if (!GREEDY_REALLOC(ni, allocated, nr+1)) {
                                free(ni);
                                return log_oom();
                        }

                        ni[nr++] = ifi;
                }

                free(m->netif);
                m->netif = ni;
                m->n_netif = nr;
        }

        return r;
}

static int machine_start_scope(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
        assert(m);
        assert(m->class != MACHINE_HOST);

        if (!m->unit) {
                _cleanup_free_ char *escaped = NULL, *scope = NULL;
                char *description, *job = NULL;
                int r;

                escaped = unit_name_escape(m->name);
                if (!escaped)
                        return log_oom();

                scope = strjoin("machine-", escaped, ".scope");
                if (!scope)
                        return log_oom();

                description = strjoina(m->class == MACHINE_VM ? "Virtual Machine " : "Container ", m->name);

                r = manager_start_scope(m->manager, scope, m->leader, SPECIAL_MACHINE_SLICE, description, properties, error, &job);
                if (r < 0)
                        return log_error_errno(r, "Failed to start machine scope: %s", bus_error_message(error, r));

                m->unit = TAKE_PTR(scope);
                free_and_replace(m->scope_job, job);
        }

        if (m->unit)
                hashmap_put(m->manager->machine_units, m->unit, m);

        return 0;
}

int machine_start(Machine *m, sd_bus_message *properties, sd_bus_error *error) {
        int r;

        assert(m);

        if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
                return -EOPNOTSUPP;

        if (m->started)
                return 0;

        r = hashmap_put(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
        if (r < 0)
                return r;

        /* Create cgroup */
        r = machine_start_scope(m, properties, error);
        if (r < 0)
                return r;

        log_struct(LOG_INFO,
                   "MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR,
                   "NAME=%s", m->name,
                   "LEADER="PID_FMT, m->leader,
                   LOG_MESSAGE("New machine %s.", m->name));

        if (!dual_timestamp_is_set(&m->timestamp))
                dual_timestamp_get(&m->timestamp);

        m->started = true;

        /* Save new machine data */
        machine_save(m);

        machine_send_signal(m, true);

        return 0;
}

static int machine_stop_scope(Machine *m) {
        _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
        char *job = NULL;
        int r, q;

        assert(m);
        assert(m->class != MACHINE_HOST);

        if (!m->unit)
                return 0;

        r = manager_stop_unit(m->manager, m->unit, &error, &job);
        if (r < 0) {
                log_error_errno(r, "Failed to stop machine scope: %s", bus_error_message(&error, r));
                sd_bus_error_free(&error);
        } else
                free_and_replace(m->scope_job, job);

        q = manager_unref_unit(m->manager, m->unit, &error);
        if (q < 0)
                log_warning_errno(q, "Failed to drop reference to machine scope, ignoring: %s", bus_error_message(&error, r));

        return r;
}

int machine_stop(Machine *m) {
        int r;
        assert(m);

        if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
                return -EOPNOTSUPP;

        r = machine_stop_scope(m);

        m->stopping = true;

        machine_save(m);

        return r;
}

int machine_finalize(Machine *m) {
        assert(m);

        if (m->started)
                log_struct(LOG_INFO,
                           "MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
                           "NAME=%s", m->name,
                           "LEADER="PID_FMT, m->leader,
                           LOG_MESSAGE("Machine %s terminated.", m->name));

        machine_unlink(m);
        machine_add_to_gc_queue(m);

        if (m->started) {
                machine_send_signal(m, false);
                m->started = false;
        }

        return 0;
}

bool machine_may_gc(Machine *m, bool drop_not_started) {
        assert(m);

        if (m->class == MACHINE_HOST)
                return false;

        if (drop_not_started && !m->started)
                return true;

        if (m->scope_job && manager_job_is_active(m->manager, m->scope_job))
                return false;

        if (m->unit && manager_unit_is_active(m->manager, m->unit))
                return false;

        return true;
}

void machine_add_to_gc_queue(Machine *m) {
        assert(m);

        if (m->in_gc_queue)
                return;

        LIST_PREPEND(gc_queue, m->manager->machine_gc_queue, m);
        m->in_gc_queue = true;
}

MachineState machine_get_state(Machine *s) {
        assert(s);

        if (s->class == MACHINE_HOST)
                return MACHINE_RUNNING;

        if (s->stopping)
                return MACHINE_CLOSING;

        if (s->scope_job)
                return MACHINE_OPENING;

        return MACHINE_RUNNING;
}

int machine_kill(Machine *m, KillWho who, int signo) {
        assert(m);

        if (!IN_SET(m->class, MACHINE_VM, MACHINE_CONTAINER))
                return -EOPNOTSUPP;

        if (!m->unit)
                return -ESRCH;

        if (who == KILL_LEADER) {
                /* If we shall simply kill the leader, do so directly */

                if (kill(m->leader, signo) < 0)
                        return -errno;

                return 0;
        }

        /* Otherwise, make PID 1 do it for us, for the entire cgroup */
        return manager_kill_unit(m->manager, m->unit, signo, NULL);
}

int machine_openpt(Machine *m, int flags) {
        assert(m);

        switch (m->class) {

        case MACHINE_HOST: {
                int fd;

                fd = posix_openpt(flags);
                if (fd < 0)
                        return -errno;

                if (unlockpt(fd) < 0)
                        return -errno;

                return fd;
        }

        case MACHINE_CONTAINER:
                if (m->leader <= 0)
                        return -EINVAL;

                return openpt_in_namespace(m->leader, flags);

        default:
                return -EOPNOTSUPP;
        }
}

int machine_open_terminal(Machine *m, const char *path, int mode) {
        assert(m);

        switch (m->class) {

        case MACHINE_HOST:
                return open_terminal(path, mode);

        case MACHINE_CONTAINER:
                if (m->leader <= 0)
                        return -EINVAL;

                return open_terminal_in_namespace(m->leader, path, mode);

        default:
                return -EOPNOTSUPP;
        }
}

void machine_release_unit(Machine *m) {
        assert(m);

        if (!m->unit)
                return;

        (void) hashmap_remove(m->manager->machine_units, m->unit);
        m->unit = mfree(m->unit);
}

int machine_get_uid_shift(Machine *m, uid_t *ret) {
        char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
        uid_t uid_base, uid_shift, uid_range;
        gid_t gid_base, gid_shift, gid_range;
        _cleanup_fclose_ FILE *f = NULL;
        int k;

        assert(m);
        assert(ret);

        /* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
         * mappings. In most cases setups should be simple like this, and administrators should only care about the
         * basic offset a container has relative to the host. This is what this function exposes.
         *
         * If we encounter any more complex mappings we politely refuse this with ENXIO. */

        if (m->class == MACHINE_HOST) {
                *ret = 0;
                return 0;
        }

        if (m->class != MACHINE_CONTAINER)
                return -EOPNOTSUPP;

        xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
        f = fopen(p, "re");
        if (!f) {
                if (errno == ENOENT) {
                        /* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
                        *ret = 0;
                        return 0;
                }

                return -errno;
        }

        /* Read the first line. There's at least one. */
        errno = 0;
        k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
        if (k != 3) {
                if (ferror(f))
                        return -errno;

                return -EBADMSG;
        }

        /* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
        if (uid_base != 0)
                return -ENXIO;
        /* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
        if (uid_range < UID_NOBODY)
                return -ENXIO;

        /* If there's more than one line, then we don't support this mapping. */
        if (fgetc(f) != EOF)
                return -ENXIO;

        fclose(f);

        xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
        f = fopen(p, "re");
        if (!f)
                return -errno;

        /* Read the first line. There's at least one. */
        errno = 0;
        k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
        if (k != 3) {
                if (ferror(f))
                        return -errno;

                return -EBADMSG;
        }

        /* If there's more than one line, then we don't support this file. */
        if (fgetc(f) != EOF)
                return -ENXIO;

        /* If the UID and GID mapping doesn't match, we don't support this mapping. */
        if (uid_base != (uid_t) gid_base)
                return -ENXIO;
        if (uid_shift != (uid_t) gid_shift)
                return -ENXIO;
        if (uid_range != (uid_t) gid_range)
                return -ENXIO;

        *ret = uid_shift;
        return 0;
}

static const char* const machine_class_table[_MACHINE_CLASS_MAX] = {
        [MACHINE_CONTAINER] = "container",
        [MACHINE_VM] = "vm",
        [MACHINE_HOST] = "host",
};

DEFINE_STRING_TABLE_LOOKUP(machine_class, MachineClass);

static const char* const machine_state_table[_MACHINE_STATE_MAX] = {
        [MACHINE_OPENING] = "opening",
        [MACHINE_RUNNING] = "running",
        [MACHINE_CLOSING] = "closing"
};

DEFINE_STRING_TABLE_LOOKUP(machine_state, MachineState);

static const char* const kill_who_table[_KILL_WHO_MAX] = {
        [KILL_LEADER] = "leader",
        [KILL_ALL] = "all"
};

DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);
Commit	Line	Data
53e1b683	1	/* SPDX-License-Identifier: LGPL-2.1+ */
9444b1f2	2
66cb2fde	3	#include <errno.h>
9444b1f2 LP	4	#include <string.h>
9444b1f2 LP	5	#include <unistd.h>
0d536673	6	#include <stdio_ext.h>
9444b1f2	7
c3350683	8	#include "sd-messages.h"
fb6becb4	9
b5efdb8a	10	#include "alloc-util.h"
66cb2fde LP	11	#include "bus-error.h"
66cb2fde LP	12	#include "bus-util.h"
4f5dd394	13	#include "escape.h"
cf0fbc49	14	#include "extract-word.h"
3ffd4af2	15	#include "fd-util.h"
9444b1f2	16	#include "fileio.h"
f97b34a6	17	#include "format-util.h"
66cb2fde	18	#include "hashmap.h"
4f5dd394	19	#include "machine-dbus.h"
3ffd4af2	20	#include "machine.h"
66cb2fde	21	#include "mkdir.h"
6bedfcbb	22	#include "parse-util.h"
4a0b58c4	23	#include "process-util.h"
d68c645b	24	#include "serialize.h"
9444b1f2	25	#include "special.h"
3401419b	26	#include "stdio-util.h"
8b43440b	27	#include "string-table.h"
66cb2fde	28	#include "terminal-util.h"
fb6becb4	29	#include "unit-name.h"
3a664727	30	#include "user-util.h"
66cb2fde	31	#include "util.h"
9444b1f2	32
fbe55073	33	Machine* machine_new(Manager manager, MachineClass class, const char name) {
9444b1f2 LP	34	Machine *m;
	35
	36	assert(manager);
fbe55073	37	assert(class < _MACHINE_CLASS_MAX);
9444b1f2 LP	38	assert(name);
9444b1f2 LP	39
fbe55073 LP	40	/* Passing class == _MACHINE_CLASS_INVALID here is fine. It
	41	* means as much as "we don't know yet", and that we'll figure
	42	* it out later when loading the state file. */
	43
9444b1f2 LP	44	m = new0(Machine, 1);
	45	if (!m)
	46	return NULL;
	47
	48	m->name = strdup(name);
	49	if (!m->name)
	50	goto fail;
	51
fbe55073 LP	52	if (class != MACHINE_HOST) {
	53	m->state_file = strappend("/run/systemd/machines/", m->name);
	54	if (!m->state_file)
	55	goto fail;
	56	}
	57
	58	m->class = class;
9444b1f2 LP	59
	60	if (hashmap_put(manager->machines, m->name, m) < 0)
	61	goto fail;
	62
9444b1f2 LP	63	m->manager = manager;
	64
	65	return m;
	66
	67	fail:
	68	free(m->state_file);
	69	free(m->name);
6b430fdb	70	return mfree(m);
9444b1f2 LP	71	}
	72
	73	void machine_free(Machine *m) {
	74	assert(m);
	75
0370612e	76	while (m->operations)
795c5d31	77	operation_free(m->operations);
0370612e	78
9444b1f2	79	if (m->in_gc_queue)
71fda00f	80	LIST_REMOVE(gc_queue, m->manager->machine_gc_queue, m);
9444b1f2	81
9b420b3c	82	machine_release_unit(m);
9444b1f2	83
fb6becb4 LP	84	free(m->scope_job);
fb6becb4 LP	85
9b420b3c	86	(void) hashmap_remove(m->manager->machines, m->name);
9444b1f2	87
fbe55073 LP	88	if (m->manager->host_machine == m)
	89	m->manager->host_machine = NULL;
	90
d3e84ddb	91	if (m->leader > 0)
4a0b58c4	92	(void) hashmap_remove_value(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
d3e84ddb	93
c3350683	94	sd_bus_message_unref(m->create_message);
fb6becb4	95
9444b1f2 LP	96	free(m->name);
	97	free(m->state_file);
	98	free(m->service);
9444b1f2	99	free(m->root_directory);
9b5ed6fe	100	free(m->netif);
9444b1f2 LP	101	free(m);
	102	}
	103
	104	int machine_save(Machine *m) {
	105	_cleanup_free_ char *temp_path = NULL;
	106	_cleanup_fclose_ FILE *f = NULL;
	107	int r;
	108
	109	assert(m);
fbe55073 LP	110
	111	if (!m->state_file)
	112	return 0;
9444b1f2 LP	113
	114	if (!m->started)
	115	return 0;
	116
37c1d5e9	117	r = mkdir_safe_label("/run/systemd/machines", 0755, 0, 0, MKDIR_WARN_MODE);
9444b1f2	118	if (r < 0)
dacd6cee	119	goto fail;
9444b1f2 LP	120
	121	r = fopen_temporary(m->state_file, &f, &temp_path);
	122	if (r < 0)
dacd6cee	123	goto fail;
9444b1f2	124
0d536673	125	(void) __fsetlocking(f, FSETLOCKING_BYCALLER);
dacd6cee	126	(void) fchmod(fileno(f), 0644);
9444b1f2 LP	127
	128	fprintf(f,
	129	"# This is private data. Do not parse.\n"
	130	"NAME=%s\n",
	131	m->name);
	132
ca5405bb LP	133	if (m->unit) {
	134	_cleanup_free_ char *escaped;
	135
	136	escaped = cescape(m->unit);
	137	if (!escaped) {
	138	r = -ENOMEM;
dacd6cee	139	goto fail;
ca5405bb LP	140	}
	141
	142	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 */
	143	}
fb6becb4 LP	144
	145	if (m->scope_job)
	146	fprintf(f, "SCOPE_JOB=%s\n", m->scope_job);
9444b1f2	147
ca5405bb LP	148	if (m->service) {
ca5405bb LP	149	_cleanup_free_ char *escaped;
9444b1f2	150
ca5405bb LP	151	escaped = cescape(m->service);
	152	if (!escaped) {
	153	r = -ENOMEM;
dacd6cee	154	goto fail;
ca5405bb LP	155	}
	156	fprintf(f, "SERVICE=%s\n", escaped);
	157	}
	158
	159	if (m->root_directory) {
	160	_cleanup_free_ char *escaped;
	161
	162	escaped = cescape(m->root_directory);
	163	if (!escaped) {
	164	r = -ENOMEM;
dacd6cee	165	goto fail;
ca5405bb LP	166	}
	167	fprintf(f, "ROOT=%s\n", escaped);
	168	}
9444b1f2	169
3bbaff3e	170	if (!sd_id128_is_null(m->id))
9444b1f2 LP	171	fprintf(f, "ID=" SD_ID128_FORMAT_STR "\n", SD_ID128_FORMAT_VAL(m->id));
	172
	173	if (m->leader != 0)
90b2de37	174	fprintf(f, "LEADER="PID_FMT"\n", m->leader);
9444b1f2 LP	175
	176	if (m->class != _MACHINE_CLASS_INVALID)
	177	fprintf(f, "CLASS=%s\n", machine_class_to_string(m->class));
	178
	179	if (dual_timestamp_is_set(&m->timestamp))
	180	fprintf(f,
90b2de37 ZJS	181	"REALTIME="USEC_FMT"\n"
	182	"MONOTONIC="USEC_FMT"\n",
	183	m->timestamp.realtime,
	184	m->timestamp.monotonic);
9444b1f2	185
9b5ed6fe	186	if (m->n_netif > 0) {
68e16e9c	187	size_t i;
9b5ed6fe	188
0d536673	189	fputs("NETIF=", f);
9b5ed6fe LP	190
	191	for (i = 0; i < m->n_netif; i++) {
	192	if (i != 0)
0d536673	193	fputc(' ', f);
9b5ed6fe LP	194
	195	fprintf(f, "%i", m->netif[i]);
	196	}
	197
0d536673	198	fputc('\n', f);
9b5ed6fe LP	199	}
9b5ed6fe LP	200
034753ac LP	201	r = fflush_and_check(f);
034753ac LP	202	if (r < 0)
dacd6cee	203	goto fail;
9444b1f2	204
034753ac	205	if (rename(temp_path, m->state_file) < 0) {
9444b1f2	206	r = -errno;
dacd6cee	207	goto fail;
9444b1f2 LP	208	}
9444b1f2 LP	209
89f7c846 LP	210	if (m->unit) {
	211	char *sl;
	212
	213	/* Create a symlink from the unit name to the machine
	214	* name, so that we can quickly find the machine for
e62d9b81	215	* each given unit. Ignore error. */
63c372cb	216	sl = strjoina("/run/systemd/machines/unit:", m->unit);
e62d9b81	217	(void) symlink(m->name, sl);
89f7c846 LP	218	}
89f7c846 LP	219
dacd6cee	220	return 0;
034753ac	221
dacd6cee LP	222	fail:
	223	(void) unlink(m->state_file);
	224
	225	if (temp_path)
	226	(void) unlink(temp_path);
9444b1f2	227
dacd6cee	228	return log_error_errno(r, "Failed to save machine data %s: %m", m->state_file);
9444b1f2 LP	229	}
9444b1f2 LP	230
89f7c846 LP	231	static void machine_unlink(Machine *m) {
	232	assert(m);
	233
	234	if (m->unit) {
89f7c846 LP	235	char *sl;
89f7c846 LP	236
63c372cb	237	sl = strjoina("/run/systemd/machines/unit:", m->unit);
491ac9f2	238	(void) unlink(sl);
89f7c846 LP	239	}
	240
	241	if (m->state_file)
491ac9f2	242	(void) unlink(m->state_file);
89f7c846 LP	243	}
89f7c846 LP	244
9444b1f2	245	int machine_load(Machine *m) {
9b5ed6fe	246	_cleanup_free_ char realtime = NULL, monotonic = NULL, id = NULL, leader = NULL, class = NULL, netif = NULL;
9444b1f2 LP	247	int r;
	248
	249	assert(m);
	250
fbe55073 LP	251	if (!m->state_file)
	252	return 0;
	253
1a5a177e	254	r = parse_env_file(NULL, m->state_file, NEWLINE,
89f7c846	255	"SCOPE", &m->unit,
fb6becb4	256	"SCOPE_JOB", &m->scope_job,
9444b1f2	257	"SERVICE", &m->service,
9444b1f2 LP	258	"ROOT", &m->root_directory,
	259	"ID", &id,
	260	"LEADER", &leader,
	261	"CLASS", &class,
	262	"REALTIME", &realtime,
	263	"MONOTONIC", &monotonic,
9b5ed6fe	264	"NETIF", &netif,
9444b1f2 LP	265	NULL);
	266	if (r < 0) {
	267	if (r == -ENOENT)
	268	return 0;
	269
8d3d7072	270	return log_error_errno(r, "Failed to read %s: %m", m->state_file);
9444b1f2 LP	271	}
	272
	273	if (id)
	274	sd_id128_from_string(id, &m->id);
	275
	276	if (leader)
	277	parse_pid(leader, &m->leader);
	278
	279	if (class) {
	280	MachineClass c;
	281
	282	c = machine_class_from_string(class);
	283	if (c >= 0)
	284	m->class = c;
	285	}
	286
b895a735	287	if (realtime)
d68c645b	288	(void) deserialize_usec(realtime, &m->timestamp.realtime);
b895a735	289	if (monotonic)
d68c645b	290	(void) deserialize_usec(monotonic, &m->timestamp.monotonic);
9444b1f2	291
9b5ed6fe	292	if (netif) {
75a8fd6a SS	293	size_t allocated = 0, nr = 0;
75a8fd6a SS	294	const char *p;
9b5ed6fe LP	295	int *ni = NULL;
9b5ed6fe LP	296
75a8fd6a	297	p = netif;
9ed794a3	298	for (;;) {
75a8fd6a	299	_cleanup_free_ char *word = NULL;
9b5ed6fe LP	300	int ifi;
9b5ed6fe LP	301
75a8fd6a	302	r = extract_first_word(&p, &word, NULL, 0);
75a8fd6a SS	303	if (r == 0)
75a8fd6a SS	304	break;
6a37c684	305	if (r == -ENOMEM)
52278ad3	306	return log_oom();
6a37c684	307	if (r < 0) {
52278ad3	308	log_warning_errno(r, "Failed to parse NETIF: %s", netif);
6a37c684	309	break;
52278ad3	310	}
75a8fd6a	311
6ad623a3	312	if (parse_ifindex(word, &ifi) < 0)
9b5ed6fe LP	313	continue;
	314
	315	if (!GREEDY_REALLOC(ni, allocated, nr+1)) {
	316	free(ni);
	317	return log_oom();
	318	}
	319
	320	ni[nr++] = ifi;
	321	}
	322
	323	free(m->netif);
	324	m->netif = ni;
	325	m->n_netif = nr;
	326	}
	327
9444b1f2 LP	328	return r;
	329	}
	330
c3350683	331	static int machine_start_scope(Machine m, sd_bus_message properties, sd_bus_error *error) {
9444b1f2	332	assert(m);
fbe55073	333	assert(m->class != MACHINE_HOST);
9444b1f2	334
89f7c846	335	if (!m->unit) {
354f62cf YW	336	_cleanup_free_ char escaped = NULL, scope = NULL;
	337	char description, job = NULL;
	338	int r;
9444b1f2	339
fb6becb4	340	escaped = unit_name_escape(m->name);
9444b1f2 LP	341	if (!escaped)
	342	return log_oom();
	343
605405c6	344	scope = strjoin("machine-", escaped, ".scope");
f526ab7e	345	if (!scope)
9444b1f2	346	return log_oom();
9444b1f2	347
63c372cb	348	description = strjoina(m->class == MACHINE_VM ? "Virtual Machine " : "Container ", m->name);
9444b1f2	349
c3350683	350	r = manager_start_scope(m->manager, scope, m->leader, SPECIAL_MACHINE_SLICE, description, properties, error, &job);
354f62cf YW	351	if (r < 0)
	352	return log_error_errno(r, "Failed to start machine scope: %s", bus_error_message(error, r));
	353
	354	m->unit = TAKE_PTR(scope);
	355	free_and_replace(m->scope_job, job);
9444b1f2 LP	356	}
9444b1f2 LP	357
89f7c846 LP	358	if (m->unit)
89f7c846 LP	359	hashmap_put(m->manager->machine_units, m->unit, m);
d0af76e6	360
354f62cf	361	return 0;
9444b1f2 LP	362	}
9444b1f2 LP	363
c3350683	364	int machine_start(Machine m, sd_bus_message properties, sd_bus_error *error) {
9444b1f2 LP	365	int r;
	366
	367	assert(m);
	368
fbe55073 LP	369	if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
	370	return -EOPNOTSUPP;
	371
9444b1f2 LP	372	if (m->started)
	373	return 0;
	374
4a0b58c4	375	r = hashmap_put(m->manager->machine_leaders, PID_TO_PTR(m->leader), m);
d3e84ddb LP	376	if (r < 0)
	377	return r;
	378
fb6becb4	379	/* Create cgroup */
c3350683	380	r = machine_start_scope(m, properties, error);
fb6becb4 LP	381	if (r < 0)
	382	return r;
	383
9444b1f2	384	log_struct(LOG_INFO,
2b044526	385	"MESSAGE_ID=" SD_MESSAGE_MACHINE_START_STR,
9444b1f2	386	"NAME=%s", m->name,
de0671ee	387	"LEADER="PID_FMT, m->leader,
a1230ff9	388	LOG_MESSAGE("New machine %s.", m->name));
9444b1f2	389
9444b1f2 LP	390	if (!dual_timestamp_is_set(&m->timestamp))
	391	dual_timestamp_get(&m->timestamp);
	392
	393	m->started = true;
	394
	395	/* Save new machine data */
	396	machine_save(m);
	397
	398	machine_send_signal(m, true);
	399
	400	return 0;
	401	}
	402
fb6becb4	403	static int machine_stop_scope(Machine *m) {
4afd3348	404	_cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
39883f62	405	char *job = NULL;
b92d0b4c	406	int r, q;
9444b1f2 LP	407
9444b1f2 LP	408	assert(m);
fbe55073	409	assert(m->class != MACHINE_HOST);
9444b1f2	410
89f7c846	411	if (!m->unit)
fb6becb4	412	return 0;
9444b1f2	413
c00a4c8f	414	r = manager_stop_unit(m->manager, m->unit, &error, &job);
b92d0b4c LP	415	if (r < 0) {
	416	log_error_errno(r, "Failed to stop machine scope: %s", bus_error_message(&error, r));
	417	sd_bus_error_free(&error);
	418	} else
	419	free_and_replace(m->scope_job, job);
	420
	421	q = manager_unref_unit(m->manager, m->unit, &error);
	422	if (q < 0)
	423	log_warning_errno(q, "Failed to drop reference to machine scope, ignoring: %s", bus_error_message(&error, r));
9444b1f2	424
b92d0b4c	425	return r;
9444b1f2 LP	426	}
	427
	428	int machine_stop(Machine *m) {
49f3fffd LP	429	int r;
	430	assert(m);
	431
fbe55073 LP	432	if (!IN_SET(m->class, MACHINE_CONTAINER, MACHINE_VM))
	433	return -EOPNOTSUPP;
	434
49f3fffd LP	435	r = machine_stop_scope(m);
	436
	437	m->stopping = true;
	438
	439	machine_save(m);
	440
	441	return r;
	442	}
	443
	444	int machine_finalize(Machine *m) {
9444b1f2 LP	445	assert(m);
	446
	447	if (m->started)
	448	log_struct(LOG_INFO,
2b044526	449	"MESSAGE_ID=" SD_MESSAGE_MACHINE_STOP_STR,
9444b1f2	450	"NAME=%s", m->name,
de0671ee	451	"LEADER="PID_FMT, m->leader,
a1230ff9	452	LOG_MESSAGE("Machine %s terminated.", m->name));
9444b1f2	453
89f7c846	454	machine_unlink(m);
9444b1f2 LP	455	machine_add_to_gc_queue(m);
9444b1f2 LP	456
49f3fffd	457	if (m->started) {
9444b1f2	458	machine_send_signal(m, false);
49f3fffd LP	459	m->started = false;
49f3fffd LP	460	}
9444b1f2	461
49f3fffd	462	return 0;
9444b1f2 LP	463	}
9444b1f2 LP	464
554ce41f	465	bool machine_may_gc(Machine *m, bool drop_not_started) {
9444b1f2 LP	466	assert(m);
9444b1f2 LP	467
fbe55073	468	if (m->class == MACHINE_HOST)
554ce41f	469	return false;
fbe55073	470
9444b1f2	471	if (drop_not_started && !m->started)
554ce41f	472	return true;
9444b1f2	473
c3350683	474	if (m->scope_job && manager_job_is_active(m->manager, m->scope_job))
554ce41f	475	return false;
9444b1f2	476
89f7c846	477	if (m->unit && manager_unit_is_active(m->manager, m->unit))
554ce41f	478	return false;
9444b1f2	479
554ce41f	480	return true;
9444b1f2 LP	481	}
	482
	483	void machine_add_to_gc_queue(Machine *m) {
	484	assert(m);
	485
	486	if (m->in_gc_queue)
	487	return;
	488
71fda00f	489	LIST_PREPEND(gc_queue, m->manager->machine_gc_queue, m);
9444b1f2 LP	490	m->in_gc_queue = true;
	491	}
	492
fb6becb4 LP	493	MachineState machine_get_state(Machine *s) {
fb6becb4 LP	494	assert(s);
9444b1f2	495
fbe55073 LP	496	if (s->class == MACHINE_HOST)
	497	return MACHINE_RUNNING;
	498
49f3fffd LP	499	if (s->stopping)
	500	return MACHINE_CLOSING;
	501
fb6becb4	502	if (s->scope_job)
49f3fffd	503	return MACHINE_OPENING;
9444b1f2	504
fb6becb4 LP	505	return MACHINE_RUNNING;
fb6becb4 LP	506	}
9444b1f2	507
fb6becb4 LP	508	int machine_kill(Machine *m, KillWho who, int signo) {
fb6becb4 LP	509	assert(m);
9444b1f2	510
fbe55073 LP	511	if (!IN_SET(m->class, MACHINE_VM, MACHINE_CONTAINER))
	512	return -EOPNOTSUPP;
	513
89f7c846	514	if (!m->unit)
fb6becb4	515	return -ESRCH;
9444b1f2	516
de58a50e LP	517	if (who == KILL_LEADER) {
	518	/* If we shall simply kill the leader, do so directly */
	519
	520	if (kill(m->leader, signo) < 0)
	521	return -errno;
9d685ca8 ED	522
9d685ca8 ED	523	return 0;
de58a50e LP	524	}
de58a50e LP	525
b938cb90	526	/* Otherwise, make PID 1 do it for us, for the entire cgroup */
de58a50e	527	return manager_kill_unit(m->manager, m->unit, signo, NULL);
9444b1f2 LP	528	}
9444b1f2 LP	529
fbe55073 LP	530	int machine_openpt(Machine *m, int flags) {
	531	assert(m);
	532
	533	switch (m->class) {
	534
5f430ff7 LP	535	case MACHINE_HOST: {
	536	int fd;
	537
	538	fd = posix_openpt(flags);
	539	if (fd < 0)
	540	return -errno;
	541
	542	if (unlockpt(fd) < 0)
	543	return -errno;
	544
	545	return fd;
	546	}
fbe55073 LP	547
	548	case MACHINE_CONTAINER:
	549	if (m->leader <= 0)
	550	return -EINVAL;
	551
	552	return openpt_in_namespace(m->leader, flags);
	553
	554	default:
	555	return -EOPNOTSUPP;
	556	}
	557	}
	558
40e1f4ea LP	559	int machine_open_terminal(Machine m, const char path, int mode) {
	560	assert(m);
	561
	562	switch (m->class) {
	563
	564	case MACHINE_HOST:
	565	return open_terminal(path, mode);
	566
	567	case MACHINE_CONTAINER:
	568	if (m->leader <= 0)
	569	return -EINVAL;
	570
	571	return open_terminal_in_namespace(m->leader, path, mode);
	572
	573	default:
	574	return -EOPNOTSUPP;
	575	}
	576	}
	577
9b420b3c LP	578	void machine_release_unit(Machine *m) {
	579	assert(m);
	580
	581	if (!m->unit)
	582	return;
	583
	584	(void) hashmap_remove(m->manager->machine_units, m->unit);
a1e58e8e	585	m->unit = mfree(m->unit);
9b420b3c LP	586	}
9b420b3c LP	587
3401419b	588	int machine_get_uid_shift(Machine m, uid_t ret) {
fbd0b64f	589	char p[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(pid_t) + 1];
3401419b LP	590	uid_t uid_base, uid_shift, uid_range;
	591	gid_t gid_base, gid_shift, gid_range;
	592	_cleanup_fclose_ FILE *f = NULL;
	593	int k;
	594
	595	assert(m);
	596	assert(ret);
	597
	598	/* Return the base UID/GID of the specified machine. Note that this only works for containers with simple
	599	* mappings. In most cases setups should be simple like this, and administrators should only care about the
	600	* basic offset a container has relative to the host. This is what this function exposes.
	601	*
	602	* If we encounter any more complex mappings we politely refuse this with ENXIO. */
	603
	604	if (m->class == MACHINE_HOST) {
	605	*ret = 0;
	606	return 0;
	607	}
	608
	609	if (m->class != MACHINE_CONTAINER)
	610	return -EOPNOTSUPP;
	611
	612	xsprintf(p, "/proc/" PID_FMT "/uid_map", m->leader);
	613	f = fopen(p, "re");
	614	if (!f) {
	615	if (errno == ENOENT) {
	616	/* If the file doesn't exist, user namespacing is off in the kernel, return a zero mapping hence. */
	617	*ret = 0;
	618	return 0;
	619	}
	620
	621	return -errno;
	622	}
	623
	624	/* Read the first line. There's at least one. */
	625	errno = 0;
	626	k = fscanf(f, UID_FMT " " UID_FMT " " UID_FMT "\n", &uid_base, &uid_shift, &uid_range);
	627	if (k != 3) {
	628	if (ferror(f))
	629	return -errno;
	630
	631	return -EBADMSG;
	632	}
	633
	634	/* Not a mapping starting at 0? Then it's a complex mapping we can't expose here. */
	635	if (uid_base != 0)
	636	return -ENXIO;
	637	/* Insist that at least the nobody user is mapped, everything else is weird, and hence complex, and we don't support it */
3a664727	638	if (uid_range < UID_NOBODY)
3401419b LP	639	return -ENXIO;
	640
	641	/* If there's more than one line, then we don't support this mapping. */
	642	if (fgetc(f) != EOF)
	643	return -ENXIO;
	644
	645	fclose(f);
	646
	647	xsprintf(p, "/proc/" PID_FMT "/gid_map", m->leader);
	648	f = fopen(p, "re");
	649	if (!f)
	650	return -errno;
	651
	652	/* Read the first line. There's at least one. */
	653	errno = 0;
	654	k = fscanf(f, GID_FMT " " GID_FMT " " GID_FMT "\n", &gid_base, &gid_shift, &gid_range);
	655	if (k != 3) {
	656	if (ferror(f))
	657	return -errno;
	658
	659	return -EBADMSG;
	660	}
	661
	662	/* If there's more than one line, then we don't support this file. */
	663	if (fgetc(f) != EOF)
	664	return -ENXIO;
	665
	666	/* If the UID and GID mapping doesn't match, we don't support this mapping. */
	667	if (uid_base != (uid_t) gid_base)
	668	return -ENXIO;
	669	if (uid_shift != (uid_t) gid_shift)
	670	return -ENXIO;
	671	if (uid_range != (uid_t) gid_range)
	672	return -ENXIO;
	673
	674	*ret = uid_shift;
	675	return 0;
	676	}
	677
9444b1f2 LP	678	static const char* const machine_class_table[_MACHINE_CLASS_MAX] = {
9444b1f2 LP	679	[MACHINE_CONTAINER] = "container",
fbe55073 LP	680	[MACHINE_VM] = "vm",
fbe55073 LP	681	[MACHINE_HOST] = "host",
9444b1f2 LP	682	};
	683
	684	DEFINE_STRING_TABLE_LOOKUP(machine_class, MachineClass);
fb6becb4 LP	685
	686	static const char* const machine_state_table[_MACHINE_STATE_MAX] = {
	687	[MACHINE_OPENING] = "opening",
	688	[MACHINE_RUNNING] = "running",
	689	[MACHINE_CLOSING] = "closing"
	690	};
	691
	692	DEFINE_STRING_TABLE_LOOKUP(machine_state, MachineState);
1ee306e1 LP	693
	694	static const char* const kill_who_table[_KILL_WHO_MAX] = {
	695	[KILL_LEADER] = "leader",
	696	[KILL_ALL] = "all"
	697	};
	698
	699	DEFINE_STRING_TABLE_LOOKUP(kill_who, KillWho);