X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?a=blobdiff_plain;f=src%2Fnspawn%2Fnspawn.c;h=7c3d7b89beb03890db7fc73a77f66f7206514c05;hb=f418f31d5042398344740f2d8ac4dc1c2583151c;hp=d8d0dae16426828372347a022e1d0aafb5fed414;hpb=1b9e5b126359a2a2ec37de1f94f046093abc74b8;p=elogind.git diff --git a/src/nspawn/nspawn.c b/src/nspawn/nspawn.c index d8d0dae16..7c3d7b89b 100644 --- a/src/nspawn/nspawn.c +++ b/src/nspawn/nspawn.c @@ -86,6 +86,7 @@ #include "udev-util.h" #include "blkid-util.h" #include "gpt.h" +#include "siphash24.h" #ifdef HAVE_SECCOMP #include "seccomp-util.h" @@ -879,23 +880,19 @@ static int setup_ptmx(const char *dest) { } static int setup_dev_console(const char *dest, const char *console) { + _cleanup_umask_ mode_t u; + const char *to; struct stat st; - _cleanup_free_ char *to = NULL; int r; - _cleanup_umask_ mode_t u; assert(dest); assert(console); u = umask(0000); - if (stat(console, &st) < 0) { - log_error("Failed to stat %s: %m", console); + if (stat("/dev/null", &st) < 0) { + log_error("Failed to stat /dev/null: %m"); return -errno; - - } else if (!S_ISCHR(st.st_mode)) { - log_error("/dev/console is not a char device"); - return -EIO; } r = chmod_and_chown(console, 0600, 0, 0); @@ -904,16 +901,15 @@ static int setup_dev_console(const char *dest, const char *console) { return r; } - if (asprintf(&to, "%s/dev/console", dest) < 0) - return log_oom(); - /* We need to bind mount the right tty to /dev/console since * ptys can only exist on pts file systems. To have something - * to bind mount things on we create a device node first, that - * has the right major/minor (note that the major minor - * doesn't actually matter here, since we mount it over - * anyway). */ + * to bind mount things on we create a device node first, and + * use /dev/null for that since we the cgroups device policy + * allows us to create that freely, while we cannot create + * /dev/console. (Note that the major minor doesn't actually + * matter here, since we mount it over anyway). */ + to = strappenda(dest, "/dev/console"); if (mknod(to, (st.st_mode & ~07777) | 0600, st.st_rdev) < 0) { log_error("mknod() for /dev/console failed: %m"); return -errno; @@ -1263,7 +1259,7 @@ static int register_machine(pid_t pid) { return r; } - r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 8, + r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 10, /* Allow the container to * access and create the API * device nodes, so that @@ -1282,7 +1278,18 @@ static int register_machine(pid_t pid) { * container to ever create * these device nodes. */ "/dev/pts/ptmx", "rw", - "char-pts", "rw"); + "char-pts", "rw", + /* Allow the container + * access to all kdbus + * devices. Again, the + * container cannot create + * these nodes, only use + * them. We use a pretty + * open match here, so that + * the kernel API can still + * change. */ + "char-kdbus", "rw", + "char-kdbus/*", "rw"); if (r < 0) { log_error("Failed to add device whitelist: %s", strerror(-r)); return r; @@ -1393,9 +1400,46 @@ static int reset_audit_loginuid(void) { return 0; } +#define HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2) + +static int get_mac(struct ether_addr *mac) { + int r; + + uint8_t result[8]; + size_t l, sz; + uint8_t *v; + + l = strlen(arg_machine); + sz = sizeof(sd_id128_t) + l; + v = alloca(sz); + + /* fetch some persistent data unique to the host */ + r = sd_id128_get_machine((sd_id128_t*) v); + if (r < 0) + return r; + + /* combine with some data unique (on this host) to this + * container instance */ + memcpy(v + sizeof(sd_id128_t), arg_machine, l); + + /* Let's hash the host machine ID plus the container name. We + * use a fixed, but originally randomly created hash key here. */ + siphash24(result, v, sz, HASH_KEY.bytes); + + assert_cc(ETH_ALEN <= sizeof(result)); + memcpy(mac->ether_addr_octet, result, ETH_ALEN); + + /* see eth_random_addr in the kernel */ + mac->ether_addr_octet[0] &= 0xfe; /* clear multicast bit */ + mac->ether_addr_octet[0] |= 0x02; /* set local assignment bit (IEEE802) */ + + return 0; +} + static int setup_veth(pid_t pid, char iface_name[IFNAMSIZ]) { _cleanup_rtnl_message_unref_ sd_rtnl_message *m = NULL; _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL; + struct ether_addr mac; int r; if (!arg_private_network) @@ -1410,9 +1454,14 @@ static int setup_veth(pid_t pid, char iface_name[IFNAMSIZ]) { memcpy(iface_name, "vb-", 3); else memcpy(iface_name, "ve-", 3); - strncpy(iface_name+3, arg_machine, IFNAMSIZ - 3); + r = get_mac(&mac); + if (r < 0) { + log_error("Failed to generate predictable MAC address for host0"); + return r; + } + r = sd_rtnl_open(&rtnl, 0); if (r < 0) { log_error("Failed to connect to netlink: %s", strerror(-r)); @@ -1461,6 +1510,12 @@ static int setup_veth(pid_t pid, char iface_name[IFNAMSIZ]) { return r; } + r = sd_rtnl_message_append_ether_addr(m, IFLA_ADDRESS, &mac); + if (r < 0) { + log_error("Failed to add netlink MAC address: %s", strerror(-r)); + return r; + } + r = sd_rtnl_message_append_u32(m, IFLA_NET_NS_PID, pid); if (r < 0) { log_error("Failed to add netlink namespace field: %s", strerror(-r)); @@ -1885,9 +1940,9 @@ static int setup_image(char **device_path, int *loop_nr) { static int dissect_image( int fd, - char **root_device, - char **home_device, - char **srv_device, + char **root_device, bool *root_device_rw, + char **home_device, bool *home_device_rw, + char **srv_device, bool *srv_device_rw, bool *secondary) { #ifdef HAVE_BLKID @@ -1898,6 +1953,7 @@ static int dissect_image( _cleanup_blkid_free_probe_ blkid_probe b = NULL; _cleanup_udev_unref_ struct udev *udev = NULL; struct udev_list_entry *first, *item; + bool home_rw = true, root_rw = true, secondary_root_rw = true, srv_rw = true; const char *pttype = NULL; blkid_partlist pl; struct stat st; @@ -1987,6 +2043,7 @@ static int dissect_image( udev_list_entry_foreach(item, first) { _cleanup_udev_device_unref_ struct udev_device *q; const char *stype, *node; + unsigned long long flags; sd_id128_t type_id; blkid_partition pp; dev_t qn; @@ -2017,6 +2074,10 @@ static int dissect_image( if (!pp) continue; + flags = blkid_partition_get_flags(pp); + if (flags & GPT_FLAG_NO_AUTO) + continue; + nr = blkid_partition_get_partno(pp); if (nr < 0) continue; @@ -2034,6 +2095,8 @@ static int dissect_image( continue; home_nr = nr; + home_rw = !(flags & GPT_FLAG_READ_ONLY); + free(home); home = strdup(node); if (!home) @@ -2044,6 +2107,8 @@ static int dissect_image( continue; srv_nr = nr; + srv_rw = !(flags & GPT_FLAG_READ_ONLY); + free(srv); srv = strdup(node); if (!srv) @@ -2056,6 +2121,8 @@ static int dissect_image( continue; root_nr = nr; + root_rw = !(flags & GPT_FLAG_READ_ONLY); + free(root); root = strdup(node); if (!root) @@ -2069,6 +2136,9 @@ static int dissect_image( continue; secondary_root_nr = nr; + secondary_root_rw = !(flags & GPT_FLAG_READ_ONLY); + + free(secondary_root); secondary_root = strdup(node); if (!secondary_root) @@ -2086,21 +2156,29 @@ static int dissect_image( if (root) { *root_device = root; root = NULL; + + *root_device_rw = root_rw; *secondary = false; } else if (secondary_root) { *root_device = secondary_root; secondary_root = NULL; + + *root_device_rw = secondary_root_rw; *secondary = true; } if (home) { *home_device = home; home = NULL; + + *home_device_rw = home_rw; } if (srv) { *srv_device = srv; srv = NULL; + + *srv_device_rw = srv_rw; } return 0; @@ -2110,7 +2188,7 @@ static int dissect_image( #endif } -static int mount_device(const char *what, const char *where, const char *directory) { +static int mount_device(const char *what, const char *where, const char *directory, bool rw) { #ifdef HAVE_BLKID _cleanup_blkid_free_probe_ blkid_probe b = NULL; const char *fstype, *p; @@ -2119,6 +2197,9 @@ static int mount_device(const char *what, const char *where, const char *directo assert(what); assert(where); + if (arg_read_only) + rw = false; + if (directory) p = strappenda(where, directory); else @@ -2161,7 +2242,7 @@ static int mount_device(const char *what, const char *where, const char *directo return -ENOTSUP; } - if (mount(what, p, fstype, arg_read_only ? MS_NODEV|MS_RDONLY : 0, NULL) < 0) { + if (mount(what, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), NULL) < 0) { log_error("Failed to mount %s: %m", what); return -errno; } @@ -2173,13 +2254,17 @@ static int mount_device(const char *what, const char *where, const char *directo #endif } -static int mount_devices(const char *where, const char *root_device, const char *home_device, const char *srv_device) { +static int mount_devices( + const char *where, + const char *root_device, bool root_device_rw, + const char *home_device, bool home_device_rw, + const char *srv_device, bool srv_device_rw) { int r; assert(where); if (root_device) { - r = mount_device(root_device, arg_directory, NULL); + r = mount_device(root_device, arg_directory, NULL, root_device_rw); if (r < 0) { log_error("Failed to mount root directory: %s", strerror(-r)); return r; @@ -2187,7 +2272,7 @@ static int mount_devices(const char *where, const char *root_device, const char } if (home_device) { - r = mount_device(home_device, arg_directory, "/home"); + r = mount_device(home_device, arg_directory, "/home", home_device_rw); if (r < 0) { log_error("Failed to mount home directory: %s", strerror(-r)); return r; @@ -2195,7 +2280,7 @@ static int mount_devices(const char *where, const char *root_device, const char } if (srv_device) { - r = mount_device(srv_device, arg_directory, "/srv"); + r = mount_device(srv_device, arg_directory, "/srv", srv_device_rw); if (r < 0) { log_error("Failed to mount server data directory: %s", strerror(-r)); return r; @@ -2224,10 +2309,277 @@ static void loop_remove(int nr, int *image_fd) { ioctl(control, LOOP_CTL_REMOVE, nr); } +static int spawn_getent(const char *database, const char *key, pid_t *rpid) { + int pipe_fds[2]; + pid_t pid; + + assert(database); + assert(key); + assert(rpid); + + if (pipe2(pipe_fds, O_CLOEXEC) < 0) { + log_error("Failed to allocate pipe: %m"); + return -errno; + } + + pid = fork(); + if (pid < 0) { + log_error("Failed to fork getent child: %m"); + return -errno; + } else if (pid == 0) { + int nullfd; + char *empty_env = NULL; + + if (dup3(pipe_fds[1], STDOUT_FILENO, 0) < 0) + _exit(EXIT_FAILURE); + + if (pipe_fds[0] > 2) + close_nointr_nofail(pipe_fds[0]); + if (pipe_fds[1] > 2) + close_nointr_nofail(pipe_fds[1]); + + nullfd = open("/dev/null", O_RDWR); + if (nullfd < 0) + _exit(EXIT_FAILURE); + + if (dup3(nullfd, STDIN_FILENO, 0) < 0) + _exit(EXIT_FAILURE); + + if (dup3(nullfd, STDERR_FILENO, 0) < 0) + _exit(EXIT_FAILURE); + + if (nullfd > 2) + close_nointr_nofail(nullfd); + + reset_all_signal_handlers(); + close_all_fds(NULL, 0); + + execle("/usr/bin/getent", "getent", database, key, NULL, &empty_env); + execle("/bin/getent", "getent", database, key, NULL, &empty_env); + _exit(EXIT_FAILURE); + } + + close_nointr_nofail(pipe_fds[1]); + pipe_fds[1] = -1; + + *rpid = pid; + + return pipe_fds[0]; +} + +static int change_uid_gid(char **_home) { + + _cleanup_strv_free_ char **passwd = NULL; + char line[LINE_MAX], *w, *x, *state, *u, *g, *h; + _cleanup_free_ uid_t *uids = NULL; + _cleanup_free_ char *home = NULL; + _cleanup_fclose_ FILE *f = NULL; + _cleanup_close_ int fd = -1; + unsigned n_uids = 0; + size_t sz, l; + uid_t uid; + gid_t gid; + pid_t pid; + int r; + + assert(_home); + + if (!arg_user || streq(arg_user, "root") || streq(arg_user, "0")) { + /* Reset everything fully to 0, just in case */ + + if (setgroups(0, NULL) < 0) { + log_error("setgroups() failed: %m"); + return -errno; + } + + if (setresgid(0, 0, 0) < 0) { + log_error("setregid() failed: %m"); + return -errno; + } + + if (setresuid(0, 0, 0) < 0) { + log_error("setreuid() failed: %m"); + return -errno; + } + + *_home = NULL; + return 0; + } + + /* First, get user credentials */ + fd = spawn_getent("passwd", arg_user, &pid); + if (fd < 0) + return fd; + + f = fdopen(fd, "r"); + if (!f) + return log_oom(); + fd = -1; + + if (!fgets(line, sizeof(line), f)) { + + if (!ferror(f)) { + log_error("Failed to resolve user %s.", arg_user); + return -ESRCH; + } + + log_error("Failed to read from getent: %m"); + return -errno; + } + + truncate_nl(line); + + wait_for_terminate_and_warn("getent passwd", pid); + + x = strchr(line, ':'); + if (!x) { + log_error("/etc/passwd entry has invalid user field."); + return -EIO; + } + + u = strchr(x+1, ':'); + if (!u) { + log_error("/etc/passwd entry has invalid password field."); + return -EIO; + } + + u++; + g = strchr(u, ':'); + if (!g) { + log_error("/etc/passwd entry has invalid UID field."); + return -EIO; + } + + *g = 0; + g++; + x = strchr(g, ':'); + if (!x) { + log_error("/etc/passwd entry has invalid GID field."); + return -EIO; + } + + *x = 0; + h = strchr(x+1, ':'); + if (!h) { + log_error("/etc/passwd entry has invalid GECOS field."); + return -EIO; + } + + h++; + x = strchr(h, ':'); + if (!x) { + log_error("/etc/passwd entry has invalid home directory field."); + return -EIO; + } + + *x = 0; + + r = parse_uid(u, &uid); + if (r < 0) { + log_error("Failed to parse UID of user."); + return -EIO; + } + + r = parse_gid(g, &gid); + if (r < 0) { + log_error("Failed to parse GID of user."); + return -EIO; + } + + home = strdup(h); + if (!home) + return log_oom(); + + /* Second, get group memberships */ + fd = spawn_getent("initgroups", arg_user, &pid); + if (fd < 0) + return fd; + + fclose(f); + f = fdopen(fd, "r"); + if (!f) + return log_oom(); + fd = -1; + + if (!fgets(line, sizeof(line), f)) { + if (!ferror(f)) { + log_error("Failed to resolve user %s.", arg_user); + return -ESRCH; + } + + log_error("Failed to read from getent: %m"); + return -errno; + } + + truncate_nl(line); + + wait_for_terminate_and_warn("getent initgroups", pid); + + /* Skip over the username and subsequent separator whitespace */ + x = line; + x += strcspn(x, WHITESPACE); + x += strspn(x, WHITESPACE); + + FOREACH_WORD(w, l, x, state) { + char c[l+1]; + + memcpy(c, w, l); + c[l] = 0; + + if (!GREEDY_REALLOC(uids, sz, n_uids+1)) + return log_oom(); + + r = parse_uid(c, &uids[n_uids++]); + if (r < 0) { + log_error("Failed to parse group data from getent."); + return -EIO; + } + } + + r = mkdir_parents(home, 0775); + if (r < 0) { + log_error("Failed to make home root directory: %s", strerror(-r)); + return r; + } + + r = mkdir_safe(home, 0755, uid, gid); + if (r < 0 && r != -EEXIST) { + log_error("Failed to make home directory: %s", strerror(-r)); + return r; + } + + fchown(STDIN_FILENO, uid, gid); + fchown(STDOUT_FILENO, uid, gid); + fchown(STDERR_FILENO, uid, gid); + + if (setgroups(n_uids, uids) < 0) { + log_error("Failed to set auxiliary groups: %m"); + return -errno; + } + + if (setresgid(gid, gid, gid) < 0) { + log_error("setregid() failed: %m"); + return -errno; + } + + if (setresuid(uid, uid, uid) < 0) { + log_error("setreuid() failed: %m"); + return -errno; + } + + if (_home) { + *_home = home; + home = NULL; + } + + return 0; +} + int main(int argc, char *argv[]) { _cleanup_free_ char *kdbus_domain = NULL, *device_path = NULL, *root_device = NULL, *home_device = NULL, *srv_device = NULL; - _cleanup_close_ int master = -1, kdbus_fd = -1, sync_fd = -1, image_fd = -1; + bool root_device_rw = true, home_device_rw = true, srv_device_rw = true; + _cleanup_close_ int master = -1, kdbus_fd = -1, image_fd = -1; _cleanup_close_pipe_ int kmsg_socket_pair[2] = { -1, -1 }; _cleanup_fdset_free_ FDSet *fds = NULL; int r = EXIT_FAILURE, k, n_fd_passed, loop_nr = -1; @@ -2344,7 +2696,7 @@ int main(int argc, char *argv[]) { goto finish; } - r = dissect_image(image_fd, &root_device, &home_device, &srv_device, &secondary); + r = dissect_image(image_fd, &root_device, &root_device_rw, &home_device, &home_device_rw, &srv_device, &srv_device_rw, &secondary); if (r < 0) goto finish; } @@ -2401,10 +2753,18 @@ int main(int argc, char *argv[]) { assert_se(sigprocmask(SIG_BLOCK, &mask, NULL) == 0); for (;;) { + int parent_ready_fd = -1, child_ready_fd = -1; siginfo_t status; + eventfd_t x; + + parent_ready_fd = eventfd(0, EFD_CLOEXEC); + if (parent_ready_fd < 0) { + log_error("Failed to create event fd: %m"); + goto finish; + } - sync_fd = eventfd(0, EFD_CLOEXEC); - if (sync_fd < 0) { + child_ready_fd = eventfd(0, EFD_CLOEXEC); + if (child_ready_fd < 0) { log_error("Failed to create event fd: %m"); goto finish; } @@ -2424,9 +2784,7 @@ int main(int argc, char *argv[]) { if (pid == 0) { /* child */ - const char *home = NULL; - uid_t uid = (uid_t) -1; - gid_t gid = (gid_t) -1; + _cleanup_free_ char *home = NULL; unsigned n_env = 2; const char *envp[] = { "PATH=" DEFAULT_PATH_SPLIT_USR, @@ -2441,7 +2799,6 @@ int main(int argc, char *argv[]) { NULL }; char **env_use; - eventfd_t x; envp[n_env] = strv_find_prefix(environ, "TERM="); if (envp[n_env]) @@ -2500,7 +2857,10 @@ int main(int argc, char *argv[]) { goto child_fail; } - if (mount_devices(arg_directory, root_device, home_device, srv_device) < 0) + if (mount_devices(arg_directory, + root_device, root_device_rw, + home_device, home_device_rw, + srv_device, srv_device_rw) < 0) goto child_fail; /* Turn directory into bind mount */ @@ -2559,6 +2919,13 @@ int main(int argc, char *argv[]) { if (setup_kdbus(arg_directory, kdbus_domain) < 0) goto child_fail; + /* Tell the parent that we are ready, and that + * it can cgroupify us to that we lack access + * to certain devices and resources. */ + eventfd_write(child_ready_fd, 1); + close_nointr_nofail(child_ready_fd); + child_ready_fd = -1; + if (chdir(arg_directory) < 0) { log_error("chdir(%s) failed: %m", arg_directory); goto child_fail; @@ -2589,61 +2956,9 @@ int main(int argc, char *argv[]) { goto child_fail; } - if (arg_user) { - - /* Note that this resolves user names - * inside the container, and hence - * accesses the NSS modules from the - * container and not the host. This is - * a bit weird... */ - - if (get_user_creds((const char**)&arg_user, &uid, &gid, &home, NULL) < 0) { - log_error("get_user_creds() failed: %m"); - goto child_fail; - } - - if (mkdir_parents_label(home, 0775) < 0) { - log_error("mkdir_parents_label() failed: %m"); - goto child_fail; - } - - if (mkdir_safe_label(home, 0775, uid, gid) < 0) { - log_error("mkdir_safe_label() failed: %m"); - goto child_fail; - } - - if (initgroups((const char*)arg_user, gid) < 0) { - log_error("initgroups() failed: %m"); - goto child_fail; - } - - if (setresgid(gid, gid, gid) < 0) { - log_error("setregid() failed: %m"); - goto child_fail; - } - - if (setresuid(uid, uid, uid) < 0) { - log_error("setreuid() failed: %m"); - goto child_fail; - } - } else { - /* Reset everything fully to 0, just in case */ - - if (setgroups(0, NULL) < 0) { - log_error("setgroups() failed: %m"); - goto child_fail; - } - - if (setresgid(0, 0, 0) < 0) { - log_error("setregid() failed: %m"); - goto child_fail; - } - - if (setresuid(0, 0, 0) < 0) { - log_error("setreuid() failed: %m"); - goto child_fail; - } - } + r = change_uid_gid(&home); + if (r < 0) + goto child_fail; if ((asprintf((char**)(envp + n_env++), "HOME=%s", home ? home: "/root") < 0) || (asprintf((char**)(envp + n_env++), "USER=%s", arg_user ? arg_user : "root") < 0) || @@ -2687,9 +3002,13 @@ int main(int argc, char *argv[]) { } } - eventfd_read(sync_fd, &x); - close_nointr_nofail(sync_fd); - sync_fd = -1; +#ifdef HAVE_SELINUX + if (arg_selinux_context) + if (setexeccon((security_context_t) arg_selinux_context) < 0) { + log_error("setexeccon(\"%s\") failed: %m", arg_selinux_context); + goto child_fail; + } +#endif if (!strv_isempty(arg_setenv)) { char **n; @@ -2704,11 +3023,11 @@ int main(int argc, char *argv[]) { } else env_use = (char**) envp; -#ifdef HAVE_SELINUX - if (arg_selinux_context) - if (setexeccon((security_context_t) arg_selinux_context) < 0) - log_error("setexeccon(\"%s\") failed: %m", arg_selinux_context); -#endif + /* Wait until the parent is ready with the setup, too... */ + eventfd_read(parent_ready_fd, &x); + close_nointr_nofail(parent_ready_fd); + parent_ready_fd = -1; + if (arg_boot) { char **a; size_t l; @@ -2744,6 +3063,12 @@ int main(int argc, char *argv[]) { fdset_free(fds); fds = NULL; + /* Wait until the child reported that it is ready with + * all it needs to do with priviliges. After we got + * the notification we can make the process join its + * cgroup which might limit what it can do */ + eventfd_read(child_ready_fd, &x); + r = register_machine(pid); if (r < 0) goto finish; @@ -2764,9 +3089,10 @@ int main(int argc, char *argv[]) { if (r < 0) goto finish; - eventfd_write(sync_fd, 1); - close_nointr_nofail(sync_fd); - sync_fd = -1; + /* Notify the child that the parent is ready with all + * its setup, and thtat the child can now hand over + * control to the code to run inside the container. */ + eventfd_write(parent_ready_fd, 1); k = process_pty(master, &mask, arg_boot ? pid : 0, SIGRTMIN+3); if (k < 0) {