[thirdparty/systemd.git] / src / nspawn / nspawn-bind-user.c

/* SPDX-License-Identifier: LGPL-2.1-or-later */

#include "chase-symlinks.h"
#include "fd-util.h"
#include "fileio.h"
#include "format-util.h"
#include "nspawn-bind-user.h"
#include "nspawn.h"
#include "path-util.h"
#include "user-util.h"
#include "userdb.h"

static int check_etc_passwd_collisions(
                const char *directory,
                const char *name,
                uid_t uid) {

        _cleanup_fclose_ FILE *f = NULL;
        int r;

        assert(directory);
        assert(name || uid_is_valid(uid));

        r = chase_symlinks_and_fopen_unlocked("/etc/passwd", directory, CHASE_PREFIX_ROOT, "re", NULL, &f);
        if (r == -ENOENT)
                return 0; /* no user database? then no user, hence no collision */
        if (r < 0)
                return log_error_errno(r, "Failed to open /etc/passwd of container: %m");

        for (;;) {
                struct passwd *pw;

                r = fgetpwent_sane(f, &pw);
                if (r < 0)
                        return log_error_errno(r, "Failed to iterate through /etc/passwd of container: %m");
                if (r == 0) /* EOF */
                        return 0; /* no collision */

                if (name && streq_ptr(pw->pw_name, name))
                        return 1; /* name collision */
                if (uid_is_valid(uid) && pw->pw_uid == uid)
                        return 1; /* UID collision */
        }
}

static int check_etc_group_collisions(
                const char *directory,
                const char *name,
                gid_t gid) {

        _cleanup_fclose_ FILE *f = NULL;
        int r;

        assert(directory);
        assert(name || gid_is_valid(gid));

        r = chase_symlinks_and_fopen_unlocked("/etc/group", directory, CHASE_PREFIX_ROOT, "re", NULL, &f);
        if (r == -ENOENT)
                return 0; /* no group database? then no group, hence no collision */
        if (r < 0)
                return log_error_errno(r, "Failed to open /etc/group of container: %m");

        for (;;) {
                struct group *gr;

                r = fgetgrent_sane(f, &gr);
                if (r < 0)
                        return log_error_errno(r, "Failed to iterate through /etc/group of container: %m");
                if (r == 0)
                        return 0; /* no collision */

                if (name && streq_ptr(gr->gr_name, name))
                        return 1; /* name collision */
                if (gid_is_valid(gid) && gr->gr_gid == gid)
                        return 1; /* gid collision */
        }
}

static int convert_user(
                const char *directory,
                UserRecord *u,
                GroupRecord *g,
                uid_t allocate_uid,
                UserRecord **ret_converted_user,
                GroupRecord **ret_converted_group) {

        _cleanup_(group_record_unrefp) GroupRecord *converted_group = NULL;
        _cleanup_(user_record_unrefp) UserRecord *converted_user = NULL;
        _cleanup_free_ char *h = NULL;
        JsonVariant *p, *hp = NULL;
        int r;

        assert(u);
        assert(g);
        assert(u->gid == g->gid);

        r = check_etc_passwd_collisions(directory, u->user_name, UID_INVALID);
        if (r < 0)
                return r;
        if (r > 0)
                return log_error_errno(SYNTHETIC_ERRNO(EBUSY),
                                       "Sorry, the user '%s' already exists in the container.", u->user_name);

        r = check_etc_group_collisions(directory, g->group_name, GID_INVALID);
        if (r < 0)
                return r;
        if (r > 0)
                return log_error_errno(SYNTHETIC_ERRNO(EBUSY),
                                       "Sorry, the group '%s' already exists in the container.", g->group_name);

        h = path_join("/run/host/home/", u->user_name);
        if (!h)
                return log_oom();

        /* Acquire the source hashed password array as-is, so that it retains the JSON_VARIANT_SENSITIVE flag */
        p = json_variant_by_key(u->json, "privileged");
        if (p)
                hp = json_variant_by_key(p, "hashedPassword");

        r = user_record_build(
                        &converted_user,
                        JSON_BUILD_OBJECT(
                                        JSON_BUILD_PAIR("userName", JSON_BUILD_STRING(u->user_name)),
                                        JSON_BUILD_PAIR("uid", JSON_BUILD_UNSIGNED(allocate_uid)),
                                        JSON_BUILD_PAIR("gid", JSON_BUILD_UNSIGNED(allocate_uid)),
                                        JSON_BUILD_PAIR_CONDITION(u->disposition >= 0, "disposition", JSON_BUILD_STRING(user_disposition_to_string(u->disposition))),
                                        JSON_BUILD_PAIR("homeDirectory", JSON_BUILD_STRING(h)),
                                        JSON_BUILD_PAIR("service", JSON_BUILD_CONST_STRING("io.systemd.NSpawn")),
                                        JSON_BUILD_PAIR_CONDITION(!strv_isempty(u->hashed_password), "privileged", JSON_BUILD_OBJECT(
                                                                                  JSON_BUILD_PAIR("hashedPassword", JSON_BUILD_VARIANT(hp))))));
        if (r < 0)
                return log_error_errno(r, "Failed to build container user record: %m");

        r = group_record_build(
                        &converted_group,
                        JSON_BUILD_OBJECT(
                                        JSON_BUILD_PAIR("groupName", JSON_BUILD_STRING(g->group_name)),
                                        JSON_BUILD_PAIR("gid", JSON_BUILD_UNSIGNED(allocate_uid)),
                                        JSON_BUILD_PAIR_CONDITION(g->disposition >= 0, "disposition", JSON_BUILD_STRING(user_disposition_to_string(g->disposition))),
                                        JSON_BUILD_PAIR("service", JSON_BUILD_CONST_STRING("io.systemd.NSpawn"))));
        if (r < 0)
                return log_error_errno(r, "Failed to build container group record: %m");

        *ret_converted_user = TAKE_PTR(converted_user);
        *ret_converted_group = TAKE_PTR(converted_group);

        return 0;
}

static int find_free_uid(const char *directory, uid_t max_uid, uid_t *current_uid) {
        int r;

        assert(directory);
        assert(current_uid);

        for (;; (*current_uid) ++) {
                if (*current_uid > MAP_UID_MAX || *current_uid > max_uid)
                        return log_error_errno(
                                        SYNTHETIC_ERRNO(EBUSY),
                                        "No suitable available UID in range " UID_FMT "…" UID_FMT " in container detected, can't map user.",
                                        MAP_UID_MIN, MAP_UID_MAX);

                r = check_etc_passwd_collisions(directory, NULL, *current_uid);
                if (r < 0)
                        return r;
                if (r > 0) /* already used */
                        continue;

                /* We want to use the UID also as GID, hence check for it in /etc/group too */
                r = check_etc_group_collisions(directory, NULL, (gid_t) *current_uid);
                if (r <= 0)
                        return r;
        }
}

BindUserContext* bind_user_context_free(BindUserContext *c) {
        if (!c)
                return NULL;

        assert(c->n_data == 0 || c->data);

        for (size_t i = 0; i < c->n_data; i++) {
                user_record_unref(c->data[i].host_user);
                group_record_unref(c->data[i].host_group);
                user_record_unref(c->data[i].payload_user);
                group_record_unref(c->data[i].payload_group);
        }

        return mfree(c);
}

int bind_user_prepare(
                const char *directory,
                char **bind_user,
                uid_t uid_shift,
                uid_t uid_range,
                CustomMount **custom_mounts,
                size_t *n_custom_mounts,
                BindUserContext **ret) {

        _cleanup_(bind_user_context_freep) BindUserContext *c = NULL;
        uid_t current_uid = MAP_UID_MIN;
        int r;

        assert(custom_mounts);
        assert(n_custom_mounts);
        assert(ret);

        /* This resolves the users specified in 'bind_user', generates a minimalized JSON user + group record
         * for it to stick in the container, allocates a UID/GID for it, and updates the custom mount table,
         * to include an appropriate bind mount mapping.
         *
         * This extends the passed custom_mounts/n_custom_mounts with the home directories, and allocates a
         * new BindUserContext for the user records */

        if (strv_isempty(bind_user)) {
                *ret = NULL;
                return 0;
        }

        c = new0(BindUserContext, 1);
        if (!c)
                return log_oom();

        STRV_FOREACH(n, bind_user) {
                _cleanup_(user_record_unrefp) UserRecord *u = NULL, *cu = NULL;
                _cleanup_(group_record_unrefp) GroupRecord *g = NULL, *cg = NULL;
                _cleanup_free_ char *sm = NULL, *sd = NULL;
                CustomMount *cm;

                r = userdb_by_name(*n, USERDB_DONT_SYNTHESIZE, &u);
                if (r < 0)
                        return log_error_errno(r, "Failed to resolve user '%s': %m", *n);

                /* For now, let's refuse mapping the root/nobody users explicitly. The records we generate
                 * are strictly additive, nss-systemd is typically placed last in /etc/nsswitch.conf. Thus
                 * even if we wanted, we couldn't override the root or nobody user records. Note we also
                 * check for name conflicts in /etc/passwd + /etc/group later on, which would usually filter
                 * out root/nobody too, hence these checks might appear redundant — but they actually are
                 * not, as we want to support environments where /etc/passwd and /etc/group are non-existent,
                 * and the user/group databases fully synthesized at runtime. Moreover, the name of the
                 * user/group name of the "nobody" account differs between distros, hence a check by numeric
                 * UID is safer. */
                if (u->uid == 0 || streq(u->user_name, "root"))
                        return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Mapping 'root' user not supported, sorry.");
                if (u->uid == UID_NOBODY || STR_IN_SET(u->user_name, NOBODY_USER_NAME, "nobody"))
                        return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Mapping 'nobody' user not supported, sorry.");

                if (u->uid >= uid_shift && u->uid < uid_shift + uid_range)
                        return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID of user '%s' to map is already in container UID range, refusing.", u->user_name);

                r = groupdb_by_gid(u->gid, USERDB_DONT_SYNTHESIZE, &g);
                if (r < 0)
                        return log_error_errno(r, "Failed to resolve group of user '%s': %m", u->user_name);

                if (g->gid >= uid_shift && g->gid < uid_shift + uid_range)
                        return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "GID of group '%s' to map is already in container GID range, refusing.", g->group_name);

                /* We want to synthesize exactly one user + group from the host into the container. This only
                 * makes sense if the user on the host has its own private group. We can't reasonably check
                 * this, so we just check of the name of user and group match.
                 *
                 * One of these days we might want to support users in a shared/common group too, but it's
                 * not clear to me how this would have to be mapped, precisely given that the common group
                 * probably already exists in the container. */
                if (!streq(u->user_name, g->group_name))
                        return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
                                               "Sorry, mapping users without private groups is currently not supported.");

                r = find_free_uid(directory, uid_range, &current_uid);
                if (r < 0)
                        return r;

                r = convert_user(directory, u, g, current_uid, &cu, &cg);
                if (r < 0)
                        return r;

                if (!GREEDY_REALLOC(c->data, c->n_data + 1))
                        return log_oom();

                sm = strdup(u->home_directory);
                if (!sm)
                        return log_oom();

                sd = strdup(cu->home_directory);
                if (!sd)
                        return log_oom();

                cm = reallocarray(*custom_mounts, sizeof(CustomMount), *n_custom_mounts + 1);
                if (!cm)
                        return log_oom();

                *custom_mounts = cm;

                (*custom_mounts)[(*n_custom_mounts)++] = (CustomMount) {
                        .type = CUSTOM_MOUNT_BIND,
                        .source = TAKE_PTR(sm),
                        .destination = TAKE_PTR(sd),
                };

                c->data[c->n_data++] = (BindUserData) {
                        .host_user = TAKE_PTR(u),
                        .host_group = TAKE_PTR(g),
                        .payload_user = TAKE_PTR(cu),
                        .payload_group = TAKE_PTR(cg),
                };

                current_uid++;
        }

        *ret = TAKE_PTR(c);
        return 1;
}

static int write_and_symlink(
                const char *root,
                JsonVariant *v,
                const char *name,
                uid_t uid,
                const char *suffix,
                WriteStringFileFlags extra_flags) {

        _cleanup_free_ char *j = NULL, *f = NULL, *p = NULL, *q = NULL;
        int r;

        assert(root);
        assert(v);
        assert(name);
        assert(uid_is_valid(uid));
        assert(suffix);

        r = json_variant_format(v, JSON_FORMAT_NEWLINE, &j);
        if (r < 0)
                return log_error_errno(r, "Failed to format user record JSON: %m");

        f = strjoin(name, suffix);
        if (!f)
                return log_oom();

        p = path_join(root, "/run/host/userdb/", f);
        if (!p)
                return log_oom();

        if (asprintf(&q, "%s/run/host/userdb/" UID_FMT "%s", root, uid, suffix) < 0)
                return log_oom();

        if (symlink(f, q) < 0)
                return log_error_errno(errno, "Failed to create symlink '%s': %m", q);

        r = userns_lchown(q, 0, 0);
        if (r < 0)
                return log_error_errno(r, "Failed to adjust access mode of '%s': %m", q);

        r = write_string_file(p, j, WRITE_STRING_FILE_CREATE|extra_flags);
        if (r < 0)
                return log_error_errno(r, "Failed to write %s: %m", p);

        r = userns_lchown(p, 0, 0);
        if (r < 0)
                return log_error_errno(r, "Failed to adjust access mode of '%s': %m", p);

        return 0;
}

int bind_user_setup(
                const BindUserContext *c,
                const char *root) {

        static const UserRecordLoadFlags strip_flags = /* Removes privileged info */
                USER_RECORD_REQUIRE_REGULAR|
                USER_RECORD_STRIP_PRIVILEGED|
                USER_RECORD_ALLOW_PER_MACHINE|
                USER_RECORD_ALLOW_BINDING|
                USER_RECORD_ALLOW_SIGNATURE|
                USER_RECORD_PERMISSIVE;
        static const UserRecordLoadFlags shadow_flags = /* Extracts privileged info */
                USER_RECORD_STRIP_REGULAR|
                USER_RECORD_ALLOW_PRIVILEGED|
                USER_RECORD_STRIP_PER_MACHINE|
                USER_RECORD_STRIP_BINDING|
                USER_RECORD_STRIP_SIGNATURE|
                USER_RECORD_EMPTY_OK|
                USER_RECORD_PERMISSIVE;
        int r;

        assert(root);

        if (!c || c->n_data == 0)
                return 0;

        r = userns_mkdir(root, "/run/host", 0755, 0, 0);
        if (r < 0)
                return log_error_errno(r, "Failed to create /run/host: %m");

        r = userns_mkdir(root, "/run/host/home", 0755, 0, 0);
        if (r < 0)
                return log_error_errno(r, "Failed to create /run/host/userdb: %m");

        r = userns_mkdir(root, "/run/host/userdb", 0755, 0, 0);
        if (r < 0)
                return log_error_errno(r, "Failed to create /run/host/userdb: %m");

        for (size_t i = 0; i < c->n_data; i++) {
                _cleanup_(group_record_unrefp) GroupRecord *stripped_group = NULL, *shadow_group = NULL;
                _cleanup_(user_record_unrefp) UserRecord *stripped_user = NULL, *shadow_user = NULL;
                const BindUserData *d = c->data + i;

                /* First, write shadow (i.e. privileged) data for group record */
                r = group_record_clone(d->payload_group, shadow_flags, &shadow_group);
                if (r < 0)
                        return log_error_errno(r, "Failed to extract privileged information from group record: %m");

                if (!json_variant_is_blank_object(shadow_group->json)) {
                        r = write_and_symlink(
                                        root,
                                        shadow_group->json,
                                        d->payload_group->group_name,
                                        d->payload_group->gid,
                                        ".group-privileged",
                                        WRITE_STRING_FILE_MODE_0600);
                        if (r < 0)
                                return r;
                }

                /* Second, write main part of group record. */
                r = group_record_clone(d->payload_group, strip_flags, &stripped_group);
                if (r < 0)
                        return log_error_errno(r, "Failed to strip privileged information from group record: %m");

                r = write_and_symlink(
                                root,
                                stripped_group->json,
                                d->payload_group->group_name,
                                d->payload_group->gid,
                                ".group",
                                0);
                if (r < 0)
                        return r;

                /* Third, write out user shadow data. i.e. extract privileged info from user record */
                r = user_record_clone(d->payload_user, shadow_flags, &shadow_user);
                if (r < 0)
                        return log_error_errno(r, "Failed to extract privileged information from user record: %m");

                if (!json_variant_is_blank_object(shadow_user->json)) {
                        r = write_and_symlink(
                                        root,
                                        shadow_user->json,
                                        d->payload_user->user_name,
                                        d->payload_user->uid,
                                        ".user-privileged",
                                        WRITE_STRING_FILE_MODE_0600);
                        if (r < 0)
                                return r;
                }

                /* Finally write out the main part of the user record */
                r = user_record_clone(d->payload_user, strip_flags, &stripped_user);
                if (r < 0)
                        return log_error_errno(r, "Failed to strip privileged information from user record: %m");

                r = write_and_symlink(
                                root,
                                stripped_user->json,
                                d->payload_user->user_name,
                                d->payload_user->uid,
                                ".user",
                                0);
                if (r < 0)
                        return r;
        }

        return 1;
}
Commit	Line	Data
2f893044 LP	1	/* SPDX-License-Identifier: LGPL-2.1-or-later */
2f893044 LP	2
f4351959	3	#include "chase-symlinks.h"
2f893044 LP	4	#include "fd-util.h"
	5	#include "fileio.h"
	6	#include "format-util.h"
2f893044 LP	7	#include "nspawn-bind-user.h"
	8	#include "nspawn.h"
	9	#include "path-util.h"
	10	#include "user-util.h"
	11	#include "userdb.h"
	12
2f893044 LP	13	static int check_etc_passwd_collisions(
	14	const char *directory,
	15	const char *name,
	16	uid_t uid) {
	17
	18	_cleanup_fclose_ FILE *f = NULL;
	19	int r;
	20
	21	assert(directory);
	22	assert(name \|\| uid_is_valid(uid));
	23
01bebba3	24	r = chase_symlinks_and_fopen_unlocked("/etc/passwd", directory, CHASE_PREFIX_ROOT, "re", NULL, &f);
2f893044 LP	25	if (r == -ENOENT)
	26	return 0; /* no user database? then no user, hence no collision */
	27	if (r < 0)
	28	return log_error_errno(r, "Failed to open /etc/passwd of container: %m");
	29
	30	for (;;) {
	31	struct passwd *pw;
	32
	33	r = fgetpwent_sane(f, &pw);
	34	if (r < 0)
	35	return log_error_errno(r, "Failed to iterate through /etc/passwd of container: %m");
	36	if (r == 0) /* EOF */
	37	return 0; /* no collision */
	38
	39	if (name && streq_ptr(pw->pw_name, name))
	40	return 1; /* name collision */
	41	if (uid_is_valid(uid) && pw->pw_uid == uid)
	42	return 1; /* UID collision */
	43	}
	44	}
	45
	46	static int check_etc_group_collisions(
	47	const char *directory,
	48	const char *name,
	49	gid_t gid) {
	50
	51	_cleanup_fclose_ FILE *f = NULL;
	52	int r;
	53
	54	assert(directory);
	55	assert(name \|\| gid_is_valid(gid));
	56
01bebba3	57	r = chase_symlinks_and_fopen_unlocked("/etc/group", directory, CHASE_PREFIX_ROOT, "re", NULL, &f);
2f893044 LP	58	if (r == -ENOENT)
	59	return 0; /* no group database? then no group, hence no collision */
	60	if (r < 0)
	61	return log_error_errno(r, "Failed to open /etc/group of container: %m");
	62
	63	for (;;) {
	64	struct group *gr;
	65
	66	r = fgetgrent_sane(f, &gr);
	67	if (r < 0)
	68	return log_error_errno(r, "Failed to iterate through /etc/group of container: %m");
	69	if (r == 0)
	70	return 0; /* no collision */
	71
	72	if (name && streq_ptr(gr->gr_name, name))
	73	return 1; /* name collision */
	74	if (gid_is_valid(gid) && gr->gr_gid == gid)
	75	return 1; /* gid collision */
	76	}
	77	}
	78
	79	static int convert_user(
	80	const char *directory,
	81	UserRecord *u,
	82	GroupRecord *g,
	83	uid_t allocate_uid,
	84	UserRecord **ret_converted_user,
	85	GroupRecord **ret_converted_group) {
	86
	87	_cleanup_(group_record_unrefp) GroupRecord *converted_group = NULL;
	88	_cleanup_(user_record_unrefp) UserRecord *converted_user = NULL;
	89	_cleanup_free_ char *h = NULL;
	90	JsonVariant p, hp = NULL;
	91	int r;
	92
	93	assert(u);
	94	assert(g);
	95	assert(u->gid == g->gid);
	96
	97	r = check_etc_passwd_collisions(directory, u->user_name, UID_INVALID);
	98	if (r < 0)
	99	return r;
	100	if (r > 0)
	101	return log_error_errno(SYNTHETIC_ERRNO(EBUSY),
	102	"Sorry, the user '%s' already exists in the container.", u->user_name);
	103
	104	r = check_etc_group_collisions(directory, g->group_name, GID_INVALID);
	105	if (r < 0)
	106	return r;
	107	if (r > 0)
	108	return log_error_errno(SYNTHETIC_ERRNO(EBUSY),
	109	"Sorry, the group '%s' already exists in the container.", g->group_name);
	110
	111	h = path_join("/run/host/home/", u->user_name);
	112	if (!h)
	113	return log_oom();
	114
	115	/* Acquire the source hashed password array as-is, so that it retains the JSON_VARIANT_SENSITIVE flag */
	116	p = json_variant_by_key(u->json, "privileged");
	117	if (p)
	118	hp = json_variant_by_key(p, "hashedPassword");
	119
	120	r = user_record_build(
	121	&converted_user,
122	JSON_BUILD_OBJECT(
123	JSON_BUILD_PAIR("userName", JSON_BUILD_STRING(u->user_name)),
124	JSON_BUILD_PAIR("uid", JSON_BUILD_UNSIGNED(allocate_uid)),
125	JSON_BUILD_PAIR("gid", JSON_BUILD_UNSIGNED(allocate_uid)),
126	JSON_BUILD_PAIR_CONDITION(u->disposition >= 0, "disposition", JSON_BUILD_STRING(user_disposition_to_string(u->disposition))),
127	JSON_BUILD_PAIR("homeDirectory", JSON_BUILD_STRING(h)),
0cdf6b14	128	JSON_BUILD_PAIR("service", JSON_BUILD_CONST_STRING("io.systemd.NSpawn")),
2f893044 LP	129	JSON_BUILD_PAIR_CONDITION(!strv_isempty(u->hashed_password), "privileged", JSON_BUILD_OBJECT(
	130	JSON_BUILD_PAIR("hashedPassword", JSON_BUILD_VARIANT(hp))))));
	131	if (r < 0)
	132	return log_error_errno(r, "Failed to build container user record: %m");
	133
	134	r = group_record_build(
	135	&converted_group,
	136	JSON_BUILD_OBJECT(
	137	JSON_BUILD_PAIR("groupName", JSON_BUILD_STRING(g->group_name)),
	138	JSON_BUILD_PAIR("gid", JSON_BUILD_UNSIGNED(allocate_uid)),
	139	JSON_BUILD_PAIR_CONDITION(g->disposition >= 0, "disposition", JSON_BUILD_STRING(user_disposition_to_string(g->disposition))),
0cdf6b14	140	JSON_BUILD_PAIR("service", JSON_BUILD_CONST_STRING("io.systemd.NSpawn"))));
2f893044 LP	141	if (r < 0)
	142	return log_error_errno(r, "Failed to build container group record: %m");
	143
	144	*ret_converted_user = TAKE_PTR(converted_user);
	145	*ret_converted_group = TAKE_PTR(converted_group);
	146
	147	return 0;
	148	}
	149
	150	static int find_free_uid(const char directory, uid_t max_uid, uid_t current_uid) {
	151	int r;
	152
	153	assert(directory);
	154	assert(current_uid);
	155
	156	for (;; (*current_uid) ++) {
76ef5d04	157	if (current_uid > MAP_UID_MAX \|\| current_uid > max_uid)
2f893044 LP	158	return log_error_errno(
	159	SYNTHETIC_ERRNO(EBUSY),
	160	"No suitable available UID in range " UID_FMT "…" UID_FMT " in container detected, can't map user.",
76ef5d04	161	MAP_UID_MIN, MAP_UID_MAX);
2f893044 LP	162
	163	r = check_etc_passwd_collisions(directory, NULL, *current_uid);
	164	if (r < 0)
	165	return r;
	166	if (r > 0) /* already used */
	167	continue;
	168
	169	/* We want to use the UID also as GID, hence check for it in /etc/group too */
	170	r = check_etc_group_collisions(directory, NULL, (gid_t) *current_uid);
2f092762	171	if (r <= 0)
2f893044	172	return r;
2f893044 LP	173	}
	174	}
	175
	176	BindUserContext* bind_user_context_free(BindUserContext *c) {
	177	if (!c)
	178	return NULL;
	179
	180	assert(c->n_data == 0 \|\| c->data);
	181
	182	for (size_t i = 0; i < c->n_data; i++) {
	183	user_record_unref(c->data[i].host_user);
	184	group_record_unref(c->data[i].host_group);
	185	user_record_unref(c->data[i].payload_user);
	186	group_record_unref(c->data[i].payload_group);
	187	}
	188
	189	return mfree(c);
	190	}
	191
	192	int bind_user_prepare(
	193	const char *directory,
	194	char **bind_user,
	195	uid_t uid_shift,
	196	uid_t uid_range,
	197	CustomMount **custom_mounts,
	198	size_t *n_custom_mounts,
	199	BindUserContext **ret) {
	200
	201	_cleanup_(bind_user_context_freep) BindUserContext *c = NULL;
76ef5d04	202	uid_t current_uid = MAP_UID_MIN;
2f893044 LP	203	int r;
	204
	205	assert(custom_mounts);
	206	assert(n_custom_mounts);
	207	assert(ret);
	208
	209	/* This resolves the users specified in 'bind_user', generates a minimalized JSON user + group record
	210	* for it to stick in the container, allocates a UID/GID for it, and updates the custom mount table,
	211	* to include an appropriate bind mount mapping.
	212	*
	213	* This extends the passed custom_mounts/n_custom_mounts with the home directories, and allocates a
	214	* new BindUserContext for the user records */
	215
	216	if (strv_isempty(bind_user)) {
	217	*ret = NULL;
	218	return 0;
	219	}
	220
	221	c = new0(BindUserContext, 1);
	222	if (!c)
	223	return log_oom();
	224
	225	STRV_FOREACH(n, bind_user) {
	226	_cleanup_(user_record_unrefp) UserRecord u = NULL, cu = NULL;
	227	_cleanup_(group_record_unrefp) GroupRecord g = NULL, cg = NULL;
	228	_cleanup_free_ char sm = NULL, sd = NULL;
	229	CustomMount *cm;
	230
	231	r = userdb_by_name(*n, USERDB_DONT_SYNTHESIZE, &u);
	232	if (r < 0)
	233	return log_error_errno(r, "Failed to resolve user '%s': %m", *n);
	234
	235	/* For now, let's refuse mapping the root/nobody users explicitly. The records we generate
	236	* are strictly additive, nss-systemd is typically placed last in /etc/nsswitch.conf. Thus
	237	* even if we wanted, we couldn't override the root or nobody user records. Note we also
	238	* check for name conflicts in /etc/passwd + /etc/group later on, which would usually filter
	239	* out root/nobody too, hence these checks might appear redundant — but they actually are
	240	* not, as we want to support environments where /etc/passwd and /etc/group are non-existent,
	241	* and the user/group databases fully synthesized at runtime. Moreover, the name of the
	242	* user/group name of the "nobody" account differs between distros, hence a check by numeric
	243	* UID is safer. */
	244	if (u->uid == 0 \|\| streq(u->user_name, "root"))
	245	return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Mapping 'root' user not supported, sorry.");
	246	if (u->uid == UID_NOBODY \|\| STR_IN_SET(u->user_name, NOBODY_USER_NAME, "nobody"))
	247	return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Mapping 'nobody' user not supported, sorry.");
	248
	249	if (u->uid >= uid_shift && u->uid < uid_shift + uid_range)
	250	return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID of user '%s' to map is already in container UID range, refusing.", u->user_name);
	251
	252	r = groupdb_by_gid(u->gid, USERDB_DONT_SYNTHESIZE, &g);
	253	if (r < 0)
	254	return log_error_errno(r, "Failed to resolve group of user '%s': %m", u->user_name);
	255
	256	if (g->gid >= uid_shift && g->gid < uid_shift + uid_range)
	257	return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "GID of group '%s' to map is already in container GID range, refusing.", g->group_name);
	258
	259	/* We want to synthesize exactly one user + group from the host into the container. This only
	260	* makes sense if the user on the host has its own private group. We can't reasonably check
	261	* this, so we just check of the name of user and group match.
	262	*
	263	* One of these days we might want to support users in a shared/common group too, but it's
	264	* not clear to me how this would have to be mapped, precisely given that the common group
	265	* probably already exists in the container. */
	266	if (!streq(u->user_name, g->group_name))
267	return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
268	"Sorry, mapping users without private groups is currently not supported.");
269
270	r = find_free_uid(directory, uid_range, &current_uid);
271	if (r < 0)
272	return r;
273
274	r = convert_user(directory, u, g, current_uid, &cu, &cg);
275	if (r < 0)
276	return r;
277
354dadb3	278	if (!GREEDY_REALLOC(c->data, c->n_data + 1))
2f893044 LP	279	return log_oom();
	280
	281	sm = strdup(u->home_directory);
	282	if (!sm)
	283	return log_oom();
	284
	285	sd = strdup(cu->home_directory);
	286	if (!sd)
	287	return log_oom();
	288
	289	cm = reallocarray(custom_mounts, sizeof(CustomMount), n_custom_mounts + 1);
	290	if (!cm)
	291	return log_oom();
	292
	293	*custom_mounts = cm;
	294
	295	(custom_mounts)[(n_custom_mounts)++] = (CustomMount) {
	296	.type = CUSTOM_MOUNT_BIND,
	297	.source = TAKE_PTR(sm),
	298	.destination = TAKE_PTR(sd),
	299	};
	300
	301	c->data[c->n_data++] = (BindUserData) {
	302	.host_user = TAKE_PTR(u),
	303	.host_group = TAKE_PTR(g),
	304	.payload_user = TAKE_PTR(cu),
	305	.payload_group = TAKE_PTR(cg),
	306	};
	307
	308	current_uid++;
	309	}
	310
	311	*ret = TAKE_PTR(c);
	312	return 1;
	313	}
	314
	315	static int write_and_symlink(
	316	const char *root,
	317	JsonVariant *v,
	318	const char *name,
	319	uid_t uid,
	320	const char *suffix,
	321	WriteStringFileFlags extra_flags) {
	322
	323	_cleanup_free_ char j = NULL, f = NULL, p = NULL, q = NULL;
	324	int r;
	325
	326	assert(root);
	327	assert(v);
	328	assert(name);
	329	assert(uid_is_valid(uid));
	330	assert(suffix);
	331
	332	r = json_variant_format(v, JSON_FORMAT_NEWLINE, &j);
	333	if (r < 0)
	334	return log_error_errno(r, "Failed to format user record JSON: %m");
	335
	336	f = strjoin(name, suffix);
	337	if (!f)
	338	return log_oom();
	339
	340	p = path_join(root, "/run/host/userdb/", f);
	341	if (!p)
	342	return log_oom();
343
344	if (asprintf(&q, "%s/run/host/userdb/" UID_FMT "%s", root, uid, suffix) < 0)
345	return log_oom();
346
347	if (symlink(f, q) < 0)
348	return log_error_errno(errno, "Failed to create symlink '%s': %m", q);
349
350	r = userns_lchown(q, 0, 0);
351	if (r < 0)
352	return log_error_errno(r, "Failed to adjust access mode of '%s': %m", q);
353
354	r = write_string_file(p, j, WRITE_STRING_FILE_CREATE\|extra_flags);
355	if (r < 0)
356	return log_error_errno(r, "Failed to write %s: %m", p);
357
358	r = userns_lchown(p, 0, 0);
359	if (r < 0)
360	return log_error_errno(r, "Failed to adjust access mode of '%s': %m", p);
361
362	return 0;
363	}
364
365	int bind_user_setup(
366	const BindUserContext *c,
367	const char *root) {
368
369	static const UserRecordLoadFlags strip_flags = /* Removes privileged info */
370	USER_RECORD_REQUIRE_REGULAR\|
371	USER_RECORD_STRIP_PRIVILEGED\|
372	USER_RECORD_ALLOW_PER_MACHINE\|
373	USER_RECORD_ALLOW_BINDING\|
bfc0cc1a LP	374	USER_RECORD_ALLOW_SIGNATURE\|
bfc0cc1a LP	375	USER_RECORD_PERMISSIVE;
2f893044 LP	376	static const UserRecordLoadFlags shadow_flags = /* Extracts privileged info */
	377	USER_RECORD_STRIP_REGULAR\|
	378	USER_RECORD_ALLOW_PRIVILEGED\|
	379	USER_RECORD_STRIP_PER_MACHINE\|
	380	USER_RECORD_STRIP_BINDING\|
	381	USER_RECORD_STRIP_SIGNATURE\|
bfc0cc1a LP	382	USER_RECORD_EMPTY_OK\|
bfc0cc1a LP	383	USER_RECORD_PERMISSIVE;
2f893044 LP	384	int r;
	385
	386	assert(root);
	387
	388	if (!c \|\| c->n_data == 0)
	389	return 0;
	390
	391	r = userns_mkdir(root, "/run/host", 0755, 0, 0);
	392	if (r < 0)
	393	return log_error_errno(r, "Failed to create /run/host: %m");
	394
	395	r = userns_mkdir(root, "/run/host/home", 0755, 0, 0);
	396	if (r < 0)
	397	return log_error_errno(r, "Failed to create /run/host/userdb: %m");
	398
	399	r = userns_mkdir(root, "/run/host/userdb", 0755, 0, 0);
	400	if (r < 0)
	401	return log_error_errno(r, "Failed to create /run/host/userdb: %m");
	402
	403	for (size_t i = 0; i < c->n_data; i++) {
	404	_cleanup_(group_record_unrefp) GroupRecord stripped_group = NULL, shadow_group = NULL;
	405	_cleanup_(user_record_unrefp) UserRecord stripped_user = NULL, shadow_user = NULL;
	406	const BindUserData *d = c->data + i;
	407
	408	/* First, write shadow (i.e. privileged) data for group record */
	409	r = group_record_clone(d->payload_group, shadow_flags, &shadow_group);
	410	if (r < 0)
	411	return log_error_errno(r, "Failed to extract privileged information from group record: %m");
	412
	413	if (!json_variant_is_blank_object(shadow_group->json)) {
	414	r = write_and_symlink(
	415	root,
	416	shadow_group->json,
	417	d->payload_group->group_name,
	418	d->payload_group->gid,
	419	".group-privileged",
	420	WRITE_STRING_FILE_MODE_0600);
	421	if (r < 0)
	422	return r;
	423	}
	424
	425	/* Second, write main part of group record. */
	426	r = group_record_clone(d->payload_group, strip_flags, &stripped_group);
	427	if (r < 0)
	428	return log_error_errno(r, "Failed to strip privileged information from group record: %m");
	429
	430	r = write_and_symlink(
	431	root,
	432	stripped_group->json,
	433	d->payload_group->group_name,
	434	d->payload_group->gid,
	435	".group",
	436	0);
	437	if (r < 0)
	438	return r;
	439
	440	/* Third, write out user shadow data. i.e. extract privileged info from user record */
	441	r = user_record_clone(d->payload_user, shadow_flags, &shadow_user);
	442	if (r < 0)
	443	return log_error_errno(r, "Failed to extract privileged information from user record: %m");
	444
	445	if (!json_variant_is_blank_object(shadow_user->json)) {
	446	r = write_and_symlink(
	447	root,
448	shadow_user->json,
449	d->payload_user->user_name,
450	d->payload_user->uid,
451	".user-privileged",
452	WRITE_STRING_FILE_MODE_0600);
453	if (r < 0)
454	return r;
455	}
456
457	/* Finally write out the main part of the user record */
458	r = user_record_clone(d->payload_user, strip_flags, &stripped_user);
459	if (r < 0)
460	return log_error_errno(r, "Failed to strip privileged information from user record: %m");
461
462	r = write_and_symlink(
463	root,
464	stripped_user->json,
465	d->payload_user->user_name,
466	d->payload_user->uid,
467	".user",
468	0);
469	if (r < 0)
470	return r;
471	}
472
473	return 1;
474	}