X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fcore%2Fexecute.c;h=ae2a52d44a05cc7fc0cb98a8630bcfef61e157e3;hp=26cde24f150de8e337bba53e01d4c2c1ed8a2ddb;hb=ee33e53a70c0d3f9d8aeafe4b8fd84a0d5e1d8d9;hpb=5521d6985648669a65d0529bb01c7a32edc99294 diff --git a/src/core/execute.c b/src/core/execute.c index 26cde24f1..ae2a52d44 100644 --- a/src/core/execute.c +++ b/src/core/execute.c @@ -38,14 +38,27 @@ #include #include #include -#include #include +#include #include +#undef basename #ifdef HAVE_PAM #include #endif +#ifdef HAVE_SELINUX +#include +#endif + +#ifdef HAVE_SECCOMP +#include +#endif + +#ifdef HAVE_APPARMOR +#include +#endif + #include "execute.h" #include "strv.h" #include "macro.h" @@ -55,24 +68,38 @@ #include "sd-messages.h" #include "ioprio.h" #include "securebits.h" -#include "cgroup.h" #include "namespace.h" -#include "tcpwrap.h" #include "exit-status.h" #include "missing.h" #include "utmp-wtmp.h" #include "def.h" -#include "loopback-setup.h" #include "path-util.h" -#include "syscall-list.h" #include "env-util.h" #include "fileio.h" +#include "unit.h" +#include "async.h" +#include "selinux-util.h" +#include "errno-list.h" +#include "af-list.h" +#include "mkdir.h" +#include "apparmor-util.h" +#include "smack-util.h" +#include "bus-kernel.h" +#include "label.h" +#include "cap-list.h" + +#ifdef HAVE_SECCOMP +#include "seccomp-util.h" +#endif #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC) +#define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC) /* This assumes there is a 'tty' group */ #define TTY_MODE 0620 +#define SNDBUF_SIZE (8*1024*1024) + static int shift_fds(int fds[], unsigned n_fds) { int start, restart_from; @@ -99,11 +126,11 @@ static int shift_fds(int fds[], unsigned n_fds) { if ((nfd = fcntl(fds[i], F_DUPFD, i+3)) < 0) return -errno; - close_nointr_nofail(fds[i]); + safe_close(fds[i]); fds[i] = nfd; /* Hmm, the fd we wanted isn't free? Then - * let's remember that and try again from here*/ + * let's remember that and try again from here */ if (nfd != i+3 && restart_from < 0) restart_from = i; } @@ -144,7 +171,7 @@ static int flags_fds(const int fds[], unsigned n_fds, bool nonblock) { return 0; } -static const char *tty_path(const ExecContext *context) { +_pure_ static const char *tty_path(const ExecContext *context) { assert(context); if (context->tty_path) @@ -153,7 +180,7 @@ static const char *tty_path(const ExecContext *context) { return "/dev/console"; } -void exec_context_tty_reset(const ExecContext *context) { +static void exec_context_tty_reset(const ExecContext *context) { assert(context); if (context->tty_vhangup) @@ -174,29 +201,18 @@ static bool is_terminal_output(ExecOutput o) { o == EXEC_OUTPUT_JOURNAL_AND_CONSOLE; } -void exec_context_serialize(const ExecContext *context, Unit *u, FILE *f) { - assert(context); - assert(u); - assert(f); - - if (context->tmp_dir) - unit_serialize_item(u, f, "tmp-dir", context->tmp_dir); - - if (context->var_tmp_dir) - unit_serialize_item(u, f, "var-tmp-dir", context->var_tmp_dir); -} - static int open_null_as(int flags, int nfd) { int fd, r; assert(nfd >= 0); - if ((fd = open("/dev/null", flags|O_NOCTTY)) < 0) + fd = open("/dev/null", flags|O_NOCTTY); + if (fd < 0) return -errno; if (fd != nfd) { r = dup2(fd, nfd) < 0 ? -errno : nfd; - close_nointr_nofail(fd); + safe_close(fd); } else r = nfd; @@ -221,15 +237,17 @@ static int connect_logger_as(const ExecContext *context, ExecOutput output, cons r = connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path)); if (r < 0) { - close_nointr_nofail(fd); + safe_close(fd); return -errno; } if (shutdown(fd, SHUT_RD) < 0) { - close_nointr_nofail(fd); + safe_close(fd); return -errno; } + fd_inc_sndbuf(fd, SNDBUF_SIZE); + dprintf(fd, "%s\n" "%s\n" @@ -248,7 +266,7 @@ static int connect_logger_as(const ExecContext *context, ExecOutput output, cons if (fd != nfd) { r = dup2(fd, nfd) < 0 ? -errno : nfd; - close_nointr_nofail(fd); + safe_close(fd); } else r = nfd; @@ -265,7 +283,7 @@ static int open_terminal_as(const char *path, mode_t mode, int nfd) { if (fd != nfd) { r = dup2(fd, nfd) < 0 ? -errno : nfd; - close_nointr_nofail(fd); + safe_close(fd); } else r = nfd; @@ -315,17 +333,17 @@ static int setup_input(const ExecContext *context, int socket_fd, bool apply_tty case EXEC_INPUT_TTY_FAIL: { int fd, r; - if ((fd = acquire_terminal( - tty_path(context), - i == EXEC_INPUT_TTY_FAIL, - i == EXEC_INPUT_TTY_FORCE, - false, - (usec_t) -1)) < 0) + fd = acquire_terminal(tty_path(context), + i == EXEC_INPUT_TTY_FAIL, + i == EXEC_INPUT_TTY_FORCE, + false, + USEC_INFINITY); + if (fd < 0) return fd; if (fd != STDIN_FILENO) { r = dup2(fd, STDIN_FILENO) < 0 ? -errno : STDIN_FILENO; - close_nointr_nofail(fd); + safe_close(fd); } else r = STDIN_FILENO; @@ -409,12 +427,13 @@ static int setup_output(const ExecContext *context, int fileno, int socket_fd, c case EXEC_OUTPUT_JOURNAL_AND_CONSOLE: r = connect_logger_as(context, o, ident, unit_id, fileno); if (r < 0) { - log_struct_unit(LOG_CRIT, unit_id, - "MESSAGE=Failed to connect std%s of %s to the journal socket: %s", - fileno == STDOUT_FILENO ? "out" : "err", - unit_id, strerror(-r), - "ERRNO=%d", -r, - NULL); + log_unit_struct(unit_id, + LOG_CRIT, + LOG_MESSAGE("Failed to connect %s of %s to the journal socket: %s", + fileno == STDOUT_FILENO ? "stdout" : "stderr", + unit_id, strerror(-r)), + LOG_ERRNO(-r), + NULL); r = open_null_as(O_WRONLY, fileno); } return r; @@ -489,7 +508,7 @@ static int setup_confirm_stdio(int *_saved_stdin, } if (fd >= 2) - close_nointr_nofail(fd); + safe_close(fd); *_saved_stdin = saved_stdin; *_saved_stdout = saved_stdout; @@ -497,20 +516,15 @@ static int setup_confirm_stdio(int *_saved_stdin, return 0; fail: - if (saved_stdout >= 0) - close_nointr_nofail(saved_stdout); - - if (saved_stdin >= 0) - close_nointr_nofail(saved_stdin); - - if (fd >= 0) - close_nointr_nofail(fd); + safe_close(saved_stdout); + safe_close(saved_stdin); + safe_close(fd); return r; } -static int write_confirm_message(const char *format, ...) { - int fd; +_printf_(1, 2) static int write_confirm_message(const char *format, ...) { + _cleanup_close_ int fd = -1; va_list ap; assert(format); @@ -523,8 +537,6 @@ static int write_confirm_message(const char *format, ...) { vdprintf(fd, format, ap); va_end(ap); - close_nointr_nofail(fd); - return 0; } @@ -546,18 +558,15 @@ static int restore_confirm_stdio(int *saved_stdin, if (dup2(*saved_stdout, STDOUT_FILENO) < 0) r = -errno; - if (*saved_stdin >= 0) - close_nointr_nofail(*saved_stdin); - - if (*saved_stdout >= 0) - close_nointr_nofail(*saved_stdout); + safe_close(*saved_stdin); + safe_close(*saved_stdout); return r; } static int ask_for_confirmation(char *response, char **argv) { int saved_stdout = -1, saved_stdin = -1, r; - char *line; + _cleanup_free_ char *line = NULL; r = setup_confirm_stdio(&saved_stdin, &saved_stdout); if (r < 0) @@ -567,8 +576,7 @@ static int ask_for_confirmation(char *response, char **argv) { if (!line) return -ENOMEM; - r = ask(response, "yns", "Execute %s? [Yes, No, Skip] ", line); - free(line); + r = ask_char(response, "yns", "Execute %s? [Yes, No, Skip] ", line); restore_confirm_stdio(&saved_stdin, &saved_stdout); @@ -655,14 +663,13 @@ static int enforce_groups(const ExecContext *context, const char *username, gid_ } static int enforce_user(const ExecContext *context, uid_t uid) { - int r; assert(context); /* Sets (but doesn't lookup) the uid and make sure we keep the * capabilities while doing so. */ if (context->capabilities) { - cap_t d; + _cleanup_cap_free_ cap_t d = NULL; static const cap_value_t bits[] = { CAP_SETUID, /* Necessary so that we can run setresuid() below */ CAP_SETPCAP /* Necessary so that we can set PR_SET_SECUREBITS later on */ @@ -682,23 +689,16 @@ static int enforce_user(const ExecContext *context, uid_t uid) { /* Second step: set the capabilities. This will reduce * the capabilities to the minimum we need. */ - if (!(d = cap_dup(context->capabilities))) + d = cap_dup(context->capabilities); + if (!d) return -errno; if (cap_set_flag(d, CAP_EFFECTIVE, ELEMENTSOF(bits), bits, CAP_SET) < 0 || - cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) { - r = -errno; - cap_free(d); - return r; - } - - if (cap_set_proc(d) < 0) { - r = -errno; - cap_free(d); - return r; - } + cap_set_flag(d, CAP_PERMITTED, ELEMENTSOF(bits), bits, CAP_SET) < 0) + return -errno; - cap_free(d); + if (cap_set_proc(d) < 0) + return -errno; } /* Third step: actually set the uids */ @@ -746,6 +746,7 @@ static int setup_pam( char **e = NULL; bool close_session = false; pid_t pam_pid = 0, parent_pid; + int flags = 0; assert(name); assert(user); @@ -758,24 +759,33 @@ static int setup_pam( * daemon. We do things this way to ensure that the main PID * of the daemon is the one we initially fork()ed. */ - if ((pam_code = pam_start(name, user, &conv, &handle)) != PAM_SUCCESS) { + if (log_get_max_level() < LOG_PRI(LOG_DEBUG)) + flags |= PAM_SILENT; + + pam_code = pam_start(name, user, &conv, &handle); + if (pam_code != PAM_SUCCESS) { handle = NULL; goto fail; } - if (tty) - if ((pam_code = pam_set_item(handle, PAM_TTY, tty)) != PAM_SUCCESS) + if (tty) { + pam_code = pam_set_item(handle, PAM_TTY, tty); + if (pam_code != PAM_SUCCESS) goto fail; + } - if ((pam_code = pam_acct_mgmt(handle, PAM_SILENT)) != PAM_SUCCESS) + pam_code = pam_acct_mgmt(handle, flags); + if (pam_code != PAM_SUCCESS) goto fail; - if ((pam_code = pam_open_session(handle, PAM_SILENT)) != PAM_SUCCESS) + pam_code = pam_open_session(handle, flags); + if (pam_code != PAM_SUCCESS) goto fail; close_session = true; - if ((!(e = pam_getenvlist(handle)))) { + e = pam_getenvlist(handle); + if (!e) { pam_code = PAM_BUF_ERR; goto fail; } @@ -789,7 +799,8 @@ static int setup_pam( parent_pid = getpid(); - if ((pam_pid = fork()) < 0) + pam_pid = fork(); + if (pam_pid < 0) goto fail; if (pam_pid == 0) { @@ -813,7 +824,7 @@ static int setup_pam( * If this fails, ignore the error - but expect sd-pam threads * to fail to exit normally */ if (setresuid(uid, uid, uid) < 0) - log_error("Error: Failed to setresuid() in sd-pam: %s", strerror(-r)); + log_error_errno(r, "Error: Failed to setresuid() in sd-pam: %m"); /* Wait until our parent died. This will only work if * the above setresuid() succeeds, otherwise the kernel @@ -840,14 +851,16 @@ static int setup_pam( } /* If our parent died we'll end the session */ - if (getppid() != parent_pid) - if ((pam_code = pam_close_session(handle, PAM_DATA_SILENT)) != PAM_SUCCESS) + if (getppid() != parent_pid) { + pam_code = pam_close_session(handle, flags); + if (pam_code != PAM_SUCCESS) goto child_finish; + } r = 0; child_finish: - pam_end(handle, pam_code | PAM_DATA_SILENT); + pam_end(handle, pam_code | flags); _exit(r); } @@ -869,16 +882,19 @@ static int setup_pam( return 0; fail: - if (pam_code != PAM_SUCCESS) + if (pam_code != PAM_SUCCESS) { + log_error("PAM failed: %s", pam_strerror(handle, pam_code)); err = -EPERM; /* PAM errors do not map to errno */ - else + } else { + log_error_errno(errno, "PAM failed: %m"); err = -errno; + } if (handle) { if (close_session) - pam_code = pam_close_session(handle, PAM_DATA_SILENT); + pam_code = pam_close_session(handle, flags); - pam_end(handle, pam_code | PAM_DATA_SILENT); + pam_end(handle, pam_code | flags); } strv_free(e); @@ -902,7 +918,7 @@ static void rename_process_from_path(const char *path) { /* This resulting string must fit in 10 chars (i.e. the length * of "/sbin/init") to look pretty in /bin/ps */ - p = path_get_file_name(path); + p = basename(path); if (isempty(p)) { rename_process("(...)"); return; @@ -925,646 +941,931 @@ static void rename_process_from_path(const char *path) { rename_process(process_name); } -static int apply_seccomp(uint32_t *syscall_filter) { - static const struct sock_filter header[] = { - VALIDATE_ARCHITECTURE, - EXAMINE_SYSCALL - }; - static const struct sock_filter footer[] = { - _KILL_PROCESS - }; +#ifdef HAVE_SECCOMP - int i; - unsigned n; - struct sock_filter *f; - struct sock_fprog prog = {}; +static int apply_seccomp(const ExecContext *c) { + uint32_t negative_action, action; + scmp_filter_ctx *seccomp; + Iterator i; + void *id; + int r; - assert(syscall_filter); + assert(c); - /* First: count the syscalls to check for */ - for (i = 0, n = 0; i < syscall_max(); i++) - if (syscall_filter[i >> 4] & (1 << (i & 31))) - n++; + negative_action = c->syscall_errno == 0 ? SCMP_ACT_KILL : SCMP_ACT_ERRNO(c->syscall_errno); - /* Second: build the filter program from a header the syscall - * matches and the footer */ - f = alloca(sizeof(struct sock_filter) * (ELEMENTSOF(header) + 2*n + ELEMENTSOF(footer))); - memcpy(f, header, sizeof(header)); + seccomp = seccomp_init(c->syscall_whitelist ? negative_action : SCMP_ACT_ALLOW); + if (!seccomp) + return -ENOMEM; - for (i = 0, n = 0; i < syscall_max(); i++) - if (syscall_filter[i >> 4] & (1 << (i & 31))) { - struct sock_filter item[] = { - BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, INDEX_TO_SYSCALL(i), 0, 1), - BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW) - }; + if (c->syscall_archs) { - assert_cc(ELEMENTSOF(item) == 2); + SET_FOREACH(id, c->syscall_archs, i) { + r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1); + if (r == -EEXIST) + continue; + if (r < 0) + goto finish; + } - f[ELEMENTSOF(header) + 2*n] = item[0]; - f[ELEMENTSOF(header) + 2*n+1] = item[1]; + } else { + r = seccomp_add_secondary_archs(seccomp); + if (r < 0) + goto finish; + } - n++; - } + action = c->syscall_whitelist ? SCMP_ACT_ALLOW : negative_action; + SET_FOREACH(id, c->syscall_filter, i) { + r = seccomp_rule_add(seccomp, action, PTR_TO_INT(id) - 1, 0); + if (r < 0) + goto finish; + } - memcpy(f + (ELEMENTSOF(header) + 2*n), footer, sizeof(footer)); + r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0); + if (r < 0) + goto finish; - /* Third: install the filter */ - prog.len = ELEMENTSOF(header) + ELEMENTSOF(footer) + 2*n; - prog.filter = f; - if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog) < 0) - return -errno; + r = seccomp_load(seccomp); - return 0; +finish: + seccomp_release(seccomp); + return r; } -int exec_spawn(ExecCommand *command, - char **argv, - ExecContext *context, - int fds[], unsigned n_fds, - char **environment, - bool apply_permissions, - bool apply_chroot, - bool apply_tty_stdin, - bool confirm_spawn, - CGroupBonding *cgroup_bondings, - CGroupAttribute *cgroup_attributes, - const char *cgroup_suffix, - const char *unit_id, - int idle_pipe[2], - pid_t *ret) { - - pid_t pid; +static int apply_address_families(const ExecContext *c) { + scmp_filter_ctx *seccomp; + Iterator i; int r; - char *line; - int socket_fd; - char _cleanup_strv_free_ **files_env = NULL; - assert(command); - assert(context); - assert(ret); - assert(fds || n_fds <= 0); + assert(c); - if (context->std_input == EXEC_INPUT_SOCKET || - context->std_output == EXEC_OUTPUT_SOCKET || - context->std_error == EXEC_OUTPUT_SOCKET) { + seccomp = seccomp_init(SCMP_ACT_ALLOW); + if (!seccomp) + return -ENOMEM; - if (n_fds != 1) - return -EINVAL; + r = seccomp_add_secondary_archs(seccomp); + if (r < 0) + goto finish; - socket_fd = fds[0]; + if (c->address_families_whitelist) { + int af, first = 0, last = 0; + void *afp; - fds = NULL; - n_fds = 0; - } else - socket_fd = -1; + /* If this is a whitelist, we first block the address + * families that are out of range and then everything + * that is not in the set. First, we find the lowest + * and highest address family in the set. */ - r = exec_context_load_environment(context, &files_env); - if (r < 0) { - log_struct_unit(LOG_ERR, - unit_id, - "MESSAGE=Failed to load environment files: %s", strerror(-r), - "ERRNO=%d", -r, - NULL); - return r; - } + SET_FOREACH(afp, c->address_families, i) { + af = PTR_TO_INT(afp); - if (!argv) - argv = command->argv; + if (af <= 0 || af >= af_max()) + continue; - line = exec_command_line(argv); - if (!line) - return log_oom(); + if (first == 0 || af < first) + first = af; - log_struct_unit(LOG_DEBUG, - unit_id, - "EXECUTABLE=%s", command->path, - "MESSAGE=About to execute: %s", line, - NULL); - free(line); + if (last == 0 || af > last) + last = af; + } - r = cgroup_bonding_realize_list(cgroup_bondings); - if (r < 0) - return r; + assert((first == 0) == (last == 0)); - /* We must initialize the attributes in the parent, before we - fork, because we really need them initialized before making - the process a member of the group (which we do in both the - child and the parent), and we cannot really apply them twice - (due to 'append' style attributes) */ - cgroup_attribute_apply_list(cgroup_attributes, cgroup_bondings); + if (first == 0) { - if (context->private_tmp && !context->tmp_dir && !context->var_tmp_dir) { - r = setup_tmpdirs(&context->tmp_dir, &context->var_tmp_dir); - if (r < 0) - return r; - } + /* No entries in the valid range, block everything */ + r = seccomp_rule_add( + seccomp, + SCMP_ACT_ERRNO(EPROTONOSUPPORT), + SCMP_SYS(socket), + 0); + if (r < 0) + goto finish; - pid = fork(); - if (pid < 0) - return -errno; + } else { - if (pid == 0) { - int i, err; - sigset_t ss; - const char *username = NULL, *home = NULL; - uid_t uid = (uid_t) -1; - gid_t gid = (gid_t) -1; - char _cleanup_strv_free_ **our_env = NULL, **pam_env = NULL, - **final_env = NULL, **final_argv = NULL; - unsigned n_env = 0; - bool set_access = false; - - /* child */ - - rename_process_from_path(command->path); - - /* We reset exactly these signals, since they are the - * only ones we set to SIG_IGN in the main daemon. All - * others we leave untouched because we set them to - * SIG_DFL or a valid handler initially, both of which - * will be demoted to SIG_DFL. */ - default_signals(SIGNALS_CRASH_HANDLER, - SIGNALS_IGNORE, -1); - - if (context->ignore_sigpipe) - ignore_signals(SIGPIPE, -1); - - assert_se(sigemptyset(&ss) == 0); - if (sigprocmask(SIG_SETMASK, &ss, NULL) < 0) { - err = -errno; - r = EXIT_SIGNAL_MASK; - goto fail_child; - } + /* Block everything below the first entry */ + r = seccomp_rule_add( + seccomp, + SCMP_ACT_ERRNO(EPROTONOSUPPORT), + SCMP_SYS(socket), + 1, + SCMP_A0(SCMP_CMP_LT, first)); + if (r < 0) + goto finish; + + /* Block everything above the last entry */ + r = seccomp_rule_add( + seccomp, + SCMP_ACT_ERRNO(EPROTONOSUPPORT), + SCMP_SYS(socket), + 1, + SCMP_A0(SCMP_CMP_GT, last)); + if (r < 0) + goto finish; - if (idle_pipe) { - if (idle_pipe[1] >= 0) - close_nointr_nofail(idle_pipe[1]); - if (idle_pipe[0] >= 0) { - fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC); - close_nointr_nofail(idle_pipe[0]); - } - } + /* Block everything between the first and last + * entry */ + for (af = 1; af < af_max(); af++) { - /* Close sockets very early to make sure we don't - * block init reexecution because it cannot bind its - * sockets */ - log_forget_fds(); - err = close_all_fds(socket_fd >= 0 ? &socket_fd : fds, - socket_fd >= 0 ? 1 : n_fds); - if (err < 0) { - r = EXIT_FDS; - goto fail_child; - } + if (set_contains(c->address_families, INT_TO_PTR(af))) + continue; - if (!context->same_pgrp) - if (setsid() < 0) { - err = -errno; - r = EXIT_SETSID; - goto fail_child; + r = seccomp_rule_add( + seccomp, + SCMP_ACT_ERRNO(EPROTONOSUPPORT), + SCMP_SYS(socket), + 1, + SCMP_A0(SCMP_CMP_EQ, af)); + if (r < 0) + goto finish; } + } - if (context->tcpwrap_name) { - if (socket_fd >= 0) - if (!socket_tcpwrap(socket_fd, context->tcpwrap_name)) { - err = -EACCES; - r = EXIT_TCPWRAP; - goto fail_child; - } + } else { + void *af; - for (i = 0; i < (int) n_fds; i++) { - if (!socket_tcpwrap(fds[i], context->tcpwrap_name)) { - err = -EACCES; - r = EXIT_TCPWRAP; - goto fail_child; - } - } - } + /* If this is a blacklist, then generate one rule for + * each address family that are then combined in OR + * checks. */ - exec_context_tty_reset(context); + SET_FOREACH(af, c->address_families, i) { - if (confirm_spawn) { - char response; - - err = ask_for_confirmation(&response, argv); - if (err == -ETIMEDOUT) - write_confirm_message("Confirmation question timed out, assuming positive response.\n"); - else if (err < 0) - write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err)); - else if (response == 's') { - write_confirm_message("Skipping execution.\n"); - err = -ECANCELED; - r = EXIT_CONFIRM; - goto fail_child; - } else if (response == 'n') { - write_confirm_message("Failing execution.\n"); - err = r = 0; - goto fail_child; - } + r = seccomp_rule_add( + seccomp, + SCMP_ACT_ERRNO(EPROTONOSUPPORT), + SCMP_SYS(socket), + 1, + SCMP_A0(SCMP_CMP_EQ, PTR_TO_INT(af))); + if (r < 0) + goto finish; } + } - /* If a socket is connected to STDIN/STDOUT/STDERR, we - * must sure to drop O_NONBLOCK */ - if (socket_fd >= 0) - fd_nonblock(socket_fd, false); + r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0); + if (r < 0) + goto finish; - err = setup_input(context, socket_fd, apply_tty_stdin); - if (err < 0) { - r = EXIT_STDIN; - goto fail_child; - } + r = seccomp_load(seccomp); - err = setup_output(context, STDOUT_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin); - if (err < 0) { - r = EXIT_STDOUT; - goto fail_child; - } +finish: + seccomp_release(seccomp); + return r; +} - err = setup_output(context, STDERR_FILENO, socket_fd, path_get_file_name(command->path), unit_id, apply_tty_stdin); - if (err < 0) { - r = EXIT_STDERR; - goto fail_child; - } +#endif - if (cgroup_bondings) { - err = cgroup_bonding_install_list(cgroup_bondings, 0, cgroup_suffix); - if (err < 0) { - r = EXIT_CGROUP; - goto fail_child; - } - } +static void do_idle_pipe_dance(int idle_pipe[4]) { + assert(idle_pipe); - if (context->oom_score_adjust_set) { - char t[16]; - snprintf(t, sizeof(t), "%i", context->oom_score_adjust); - char_array_0(t); + safe_close(idle_pipe[1]); + safe_close(idle_pipe[2]); - if (write_string_file("/proc/self/oom_score_adj", t) < 0) { - err = -errno; - r = EXIT_OOM_ADJUST; - goto fail_child; - } - } + if (idle_pipe[0] >= 0) { + int r; - if (context->nice_set) - if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) { - err = -errno; - r = EXIT_NICE; - goto fail_child; - } + r = fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT_USEC); - if (context->cpu_sched_set) { - struct sched_param param = { - .sched_priority = context->cpu_sched_priority, - }; + if (idle_pipe[3] >= 0 && r == 0 /* timeout */) { + /* Signal systemd that we are bored and want to continue. */ + write(idle_pipe[3], "x", 1); - r = sched_setscheduler(0, - context->cpu_sched_policy | - (context->cpu_sched_reset_on_fork ? - SCHED_RESET_ON_FORK : 0), - ¶m); - if (r < 0) { - err = -errno; - r = EXIT_SETSCHEDULER; - goto fail_child; - } + /* Wait for systemd to react to the signal above. */ + fd_wait_for_event(idle_pipe[0], POLLHUP, IDLE_TIMEOUT2_USEC); } - if (context->cpuset) - if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) { - err = -errno; - r = EXIT_CPUAFFINITY; - goto fail_child; - } + safe_close(idle_pipe[0]); - if (context->ioprio_set) - if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) { - err = -errno; - r = EXIT_IOPRIO; - goto fail_child; - } + } - if (context->timer_slack_nsec != (nsec_t) -1) - if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) { - err = -errno; - r = EXIT_TIMERSLACK; - goto fail_child; - } + safe_close(idle_pipe[3]); +} - if (context->utmp_id) - utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path); +static int build_environment( + const ExecContext *c, + unsigned n_fds, + usec_t watchdog_usec, + const char *home, + const char *username, + const char *shell, + char ***ret) { - if (context->user) { - username = context->user; - err = get_user_creds(&username, &uid, &gid, &home, NULL); - if (err < 0) { - r = EXIT_USER; - goto fail_child; - } + _cleanup_strv_free_ char **our_env = NULL; + unsigned n_env = 0; + char *x; - if (is_terminal_input(context->std_input)) { - err = chown_terminal(STDIN_FILENO, uid); - if (err < 0) { - r = EXIT_STDIN; - goto fail_child; - } - } + assert(c); + assert(ret); - if (cgroup_bondings && context->control_group_modify) { - err = cgroup_bonding_set_group_access_list(cgroup_bondings, 0755, uid, gid); - if (err >= 0) - err = cgroup_bonding_set_task_access_list( - cgroup_bondings, - 0644, - uid, - gid, - context->control_group_persistent); - if (err < 0) { - r = EXIT_CGROUP; - goto fail_child; - } + our_env = new0(char*, 10); + if (!our_env) + return -ENOMEM; - set_access = true; - } - } + if (n_fds > 0) { + if (asprintf(&x, "LISTEN_PID="PID_FMT, getpid()) < 0) + return -ENOMEM; + our_env[n_env++] = x; - if (cgroup_bondings && !set_access && context->control_group_persistent >= 0) { - err = cgroup_bonding_set_task_access_list( - cgroup_bondings, - (mode_t) -1, - (uid_t) -1, - (uid_t) -1, - context->control_group_persistent); - if (err < 0) { - r = EXIT_CGROUP; - goto fail_child; - } - } + if (asprintf(&x, "LISTEN_FDS=%u", n_fds) < 0) + return -ENOMEM; + our_env[n_env++] = x; + } - if (apply_permissions) { - err = enforce_groups(context, username, gid); - if (err < 0) { - r = EXIT_GROUP; - goto fail_child; - } - } + if (watchdog_usec > 0) { + if (asprintf(&x, "WATCHDOG_PID="PID_FMT, getpid()) < 0) + return -ENOMEM; + our_env[n_env++] = x; - umask(context->umask); + if (asprintf(&x, "WATCHDOG_USEC="USEC_FMT, watchdog_usec) < 0) + return -ENOMEM; + our_env[n_env++] = x; + } -#ifdef HAVE_PAM - if (apply_permissions && context->pam_name && username) { - err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds); - if (err < 0) { - r = EXIT_PAM; - goto fail_child; - } - } -#endif - if (context->private_network) { - if (unshare(CLONE_NEWNET) < 0) { - err = -errno; - r = EXIT_NETWORK; - goto fail_child; - } + if (home) { + x = strappend("HOME=", home); + if (!x) + return -ENOMEM; + our_env[n_env++] = x; + } - loopback_setup(); - } + if (username) { + x = strappend("LOGNAME=", username); + if (!x) + return -ENOMEM; + our_env[n_env++] = x; - if (strv_length(context->read_write_dirs) > 0 || - strv_length(context->read_only_dirs) > 0 || - strv_length(context->inaccessible_dirs) > 0 || - context->mount_flags != 0 || - context->private_tmp) { - err = setup_namespace(context->read_write_dirs, - context->read_only_dirs, - context->inaccessible_dirs, - context->tmp_dir, - context->var_tmp_dir, - context->private_tmp, - context->mount_flags); - if (err < 0) { - r = EXIT_NAMESPACE; - goto fail_child; - } - } + x = strappend("USER=", username); + if (!x) + return -ENOMEM; + our_env[n_env++] = x; + } - if (apply_chroot) { - if (context->root_directory) - if (chroot(context->root_directory) < 0) { - err = -errno; - r = EXIT_CHROOT; - goto fail_child; - } + if (shell) { + x = strappend("SHELL=", shell); + if (!x) + return -ENOMEM; + our_env[n_env++] = x; + } - if (chdir(context->working_directory ? context->working_directory : "/") < 0) { - err = -errno; - r = EXIT_CHDIR; - goto fail_child; - } - } else { - char _cleanup_free_ *d = NULL; - - if (asprintf(&d, "%s/%s", - context->root_directory ? context->root_directory : "", - context->working_directory ? context->working_directory : "") < 0) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + if (is_terminal_input(c->std_input) || + c->std_output == EXEC_OUTPUT_TTY || + c->std_error == EXEC_OUTPUT_TTY || + c->tty_path) { - if (chdir(d) < 0) { - err = -errno; - r = EXIT_CHDIR; - goto fail_child; - } - } + x = strdup(default_term_for_tty(tty_path(c))); + if (!x) + return -ENOMEM; + our_env[n_env++] = x; + } - /* We repeat the fd closing here, to make sure that - * nothing is leaked from the PAM modules */ - err = close_all_fds(fds, n_fds); - if (err >= 0) - err = shift_fds(fds, n_fds); - if (err >= 0) - err = flags_fds(fds, n_fds, context->non_blocking); - if (err < 0) { - r = EXIT_FDS; - goto fail_child; - } + our_env[n_env++] = NULL; + assert(n_env <= 10); - if (apply_permissions) { + *ret = our_env; + our_env = NULL; - for (i = 0; i < RLIMIT_NLIMITS; i++) { - if (!context->rlimit[i]) - continue; + return 0; +} - if (setrlimit_closest(i, context->rlimit[i]) < 0) { - err = -errno; - r = EXIT_LIMITS; - goto fail_child; - } - } +static int exec_child(ExecCommand *command, + const ExecContext *context, + const ExecParameters *params, + ExecRuntime *runtime, + char **argv, + int socket_fd, + int *fds, unsigned n_fds, + char **files_env, + int *error) { + + _cleanup_strv_free_ char **our_env = NULL, **pam_env = NULL, **final_env = NULL, **final_argv = NULL; + _cleanup_free_ char *mac_selinux_context_net = NULL; + const char *username = NULL, *home = NULL, *shell = NULL; + unsigned n_dont_close = 0; + int dont_close[n_fds + 4]; + uid_t uid = UID_INVALID; + gid_t gid = GID_INVALID; + int i, err; - if (context->capability_bounding_set_drop) { - err = capability_bounding_set_drop(context->capability_bounding_set_drop, false); - if (err < 0) { - r = EXIT_CAPABILITIES; - goto fail_child; - } - } + assert(command); + assert(context); + assert(params); + assert(error); + + rename_process_from_path(command->path); + + /* We reset exactly these signals, since they are the + * only ones we set to SIG_IGN in the main daemon. All + * others we leave untouched because we set them to + * SIG_DFL or a valid handler initially, both of which + * will be demoted to SIG_DFL. */ + default_signals(SIGNALS_CRASH_HANDLER, + SIGNALS_IGNORE, -1); + + if (context->ignore_sigpipe) + ignore_signals(SIGPIPE, -1); + + err = reset_signal_mask(); + if (err < 0) { + *error = EXIT_SIGNAL_MASK; + return err; + } - if (context->user) { - err = enforce_user(context, uid); - if (err < 0) { - r = EXIT_USER; - goto fail_child; - } - } + if (params->idle_pipe) + do_idle_pipe_dance(params->idle_pipe); - /* PR_GET_SECUREBITS is not privileged, while - * PR_SET_SECUREBITS is. So to suppress - * potential EPERMs we'll try not to call - * PR_SET_SECUREBITS unless necessary. */ - if (prctl(PR_GET_SECUREBITS) != context->secure_bits) - if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) { - err = -errno; - r = EXIT_SECUREBITS; - goto fail_child; - } + /* Close sockets very early to make sure we don't + * block init reexecution because it cannot bind its + * sockets */ + log_forget_fds(); - if (context->capabilities) - if (cap_set_proc(context->capabilities) < 0) { - err = -errno; - r = EXIT_CAPABILITIES; - goto fail_child; - } + if (socket_fd >= 0) + dont_close[n_dont_close++] = socket_fd; + if (n_fds > 0) { + memcpy(dont_close + n_dont_close, fds, sizeof(int) * n_fds); + n_dont_close += n_fds; + } + if (params->bus_endpoint_fd >= 0) + dont_close[n_dont_close++] = params->bus_endpoint_fd; + if (runtime) { + if (runtime->netns_storage_socket[0] >= 0) + dont_close[n_dont_close++] = runtime->netns_storage_socket[0]; + if (runtime->netns_storage_socket[1] >= 0) + dont_close[n_dont_close++] = runtime->netns_storage_socket[1]; + } - if (context->no_new_privileges) - if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { - err = -errno; - r = EXIT_NO_NEW_PRIVILEGES; - goto fail_child; - } + err = close_all_fds(dont_close, n_dont_close); + if (err < 0) { + *error = EXIT_FDS; + return err; + } - if (context->syscall_filter) { - err = apply_seccomp(context->syscall_filter); - if (err < 0) { - r = EXIT_SECCOMP; - goto fail_child; - } - } + if (!context->same_pgrp) + if (setsid() < 0) { + *error = EXIT_SETSID; + return -errno; } - our_env = new0(char*, 7); - if (!our_env) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; + exec_context_tty_reset(context); + + if (params->confirm_spawn) { + char response; + + err = ask_for_confirmation(&response, argv); + if (err == -ETIMEDOUT) + write_confirm_message("Confirmation question timed out, assuming positive response.\n"); + else if (err < 0) + write_confirm_message("Couldn't ask confirmation question, assuming positive response: %s\n", strerror(-err)); + else if (response == 's') { + write_confirm_message("Skipping execution.\n"); + *error = EXIT_CONFIRM; + return -ECANCELED; + } else if (response == 'n') { + write_confirm_message("Failing execution.\n"); + *error = 0; + return 0; } + } - if (n_fds > 0) - if (asprintf(our_env + n_env++, "LISTEN_PID=%lu", (unsigned long) getpid()) < 0 || - asprintf(our_env + n_env++, "LISTEN_FDS=%u", n_fds) < 0) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + /* If a socket is connected to STDIN/STDOUT/STDERR, we + * must sure to drop O_NONBLOCK */ + if (socket_fd >= 0) + fd_nonblock(socket_fd, false); - if (home) - if (asprintf(our_env + n_env++, "HOME=%s", home) < 0) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + err = setup_input(context, socket_fd, params->apply_tty_stdin); + if (err < 0) { + *error = EXIT_STDIN; + return err; + } - if (username) - if (asprintf(our_env + n_env++, "LOGNAME=%s", username) < 0 || - asprintf(our_env + n_env++, "USER=%s", username) < 0) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + err = setup_output(context, STDOUT_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin); + if (err < 0) { + *error = EXIT_STDOUT; + return err; + } - if (is_terminal_input(context->std_input) || - context->std_output == EXEC_OUTPUT_TTY || - context->std_error == EXEC_OUTPUT_TTY) - if (!(our_env[n_env++] = strdup(default_term_for_tty(tty_path(context))))) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + err = setup_output(context, STDERR_FILENO, socket_fd, basename(command->path), params->unit_id, params->apply_tty_stdin); + if (err < 0) { + *error = EXIT_STDERR; + return err; + } - assert(n_env <= 7); - - final_env = strv_env_merge(5, - environment, - our_env, - context->environment, - files_env, - pam_env, - NULL); - if (!final_env) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; + if (params->cgroup_path) { + err = cg_attach_everywhere(params->cgroup_supported, params->cgroup_path, 0, NULL, NULL); + if (err < 0) { + *error = EXIT_CGROUP; + return err; } + } - final_argv = replace_env_argv(argv, final_env); - if (!final_argv) { - err = -ENOMEM; - r = EXIT_MEMORY; - goto fail_child; - } + if (context->oom_score_adjust_set) { + char t[16]; - final_env = strv_env_clean(final_env); - - if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) { - line = exec_command_line(final_argv); - if (line) { - log_open(); - log_struct_unit(LOG_DEBUG, - unit_id, - "EXECUTABLE=%s", command->path, - "MESSAGE=Executing: %s", line, - NULL); - log_close(); - free(line); - line = NULL; - } + snprintf(t, sizeof(t), "%i", context->oom_score_adjust); + char_array_0(t); + + if (write_string_file("/proc/self/oom_score_adj", t) < 0) { + *error = EXIT_OOM_ADJUST; + return -errno; } - execve(command->path, final_argv, final_env); - err = -errno; - r = EXIT_EXEC; + } - fail_child: - if (r != 0) { - log_open(); - log_struct(LOG_ERR, MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED), - "EXECUTABLE=%s", command->path, - "MESSAGE=Failed at step %s spawning %s: %s", - exit_status_to_string(r, EXIT_STATUS_SYSTEMD), - command->path, strerror(-err), - "ERRNO=%d", -err, - NULL); - log_close(); + if (context->nice_set) + if (setpriority(PRIO_PROCESS, 0, context->nice) < 0) { + *error = EXIT_NICE; + return -errno; } - _exit(r); - } + if (context->cpu_sched_set) { + struct sched_param param = { + .sched_priority = context->cpu_sched_priority, + }; - log_struct_unit(LOG_DEBUG, - unit_id, - "MESSAGE=Forked %s as %lu", - command->path, (unsigned long) pid, - NULL); + err = sched_setscheduler(0, + context->cpu_sched_policy | + (context->cpu_sched_reset_on_fork ? + SCHED_RESET_ON_FORK : 0), + ¶m); + if (err < 0) { + *error = EXIT_SETSCHEDULER; + return -errno; + } + } - /* We add the new process to the cgroup both in the child (so - * that we can be sure that no user code is ever executed - * outside of the cgroup) and in the parent (so that we can be - * sure that when we kill the cgroup the process will be - * killed too). */ - cgroup_bonding_install_list(cgroup_bondings, pid, cgroup_suffix); + if (context->cpuset) + if (sched_setaffinity(0, CPU_ALLOC_SIZE(context->cpuset_ncpus), context->cpuset) < 0) { + *error = EXIT_CPUAFFINITY; + return -errno; + } - exec_status_start(&command->exec_status, pid); + if (context->ioprio_set) + if (ioprio_set(IOPRIO_WHO_PROCESS, 0, context->ioprio) < 0) { + *error = EXIT_IOPRIO; + return -errno; + } + + if (context->timer_slack_nsec != NSEC_INFINITY) + if (prctl(PR_SET_TIMERSLACK, context->timer_slack_nsec) < 0) { + *error = EXIT_TIMERSLACK; + return -errno; + } + + if (context->personality != 0xffffffffUL) + if (personality(context->personality) < 0) { + *error = EXIT_PERSONALITY; + return -errno; + } + + if (context->utmp_id) + utmp_put_init_process(context->utmp_id, getpid(), getsid(0), context->tty_path); + + if (context->user) { + username = context->user; + err = get_user_creds(&username, &uid, &gid, &home, &shell); + if (err < 0) { + *error = EXIT_USER; + return err; + } + + if (is_terminal_input(context->std_input)) { + err = chown_terminal(STDIN_FILENO, uid); + if (err < 0) { + *error = EXIT_STDIN; + return err; + } + } + } + +#ifdef ENABLE_KDBUS + if (params->bus_endpoint_fd >= 0 && context->bus_endpoint) { + uid_t ep_uid = (uid == UID_INVALID) ? 0 : uid; + + err = bus_kernel_set_endpoint_policy(params->bus_endpoint_fd, ep_uid, context->bus_endpoint); + if (err < 0) { + *error = EXIT_BUS_ENDPOINT; + return err; + } + } +#endif + + /* If delegation is enabled we'll pass ownership of the cgroup + * (but only in systemd's own controller hierarchy!) to the + * user of the new process. */ + if (params->cgroup_path && context->user && params->cgroup_delegate) { + err = cg_set_task_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0644, uid, gid); + if (err < 0) { + *error = EXIT_CGROUP; + return err; + } + + + err = cg_set_group_access(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, 0755, uid, gid); + if (err < 0) { + *error = EXIT_CGROUP; + return err; + } + } + + if (!strv_isempty(context->runtime_directory) && params->runtime_prefix) { + char **rt; + + STRV_FOREACH(rt, context->runtime_directory) { + _cleanup_free_ char *p; + + p = strjoin(params->runtime_prefix, "/", *rt, NULL); + if (!p) { + *error = EXIT_RUNTIME_DIRECTORY; + return -ENOMEM; + } + + err = mkdir_safe(p, context->runtime_directory_mode, uid, gid); + if (err < 0) { + *error = EXIT_RUNTIME_DIRECTORY; + return err; + } + } + } + + if (params->apply_permissions) { + err = enforce_groups(context, username, gid); + if (err < 0) { + *error = EXIT_GROUP; + return err; + } + } + + umask(context->umask); + +#ifdef HAVE_PAM + if (params->apply_permissions && context->pam_name && username) { + err = setup_pam(context->pam_name, username, uid, context->tty_path, &pam_env, fds, n_fds); + if (err < 0) { + *error = EXIT_PAM; + return err; + } + } +#endif + + if (context->private_network && runtime && runtime->netns_storage_socket[0] >= 0) { + err = setup_netns(runtime->netns_storage_socket); + if (err < 0) { + *error = EXIT_NETWORK; + return err; + } + } + + if (!strv_isempty(context->read_write_dirs) || + !strv_isempty(context->read_only_dirs) || + !strv_isempty(context->inaccessible_dirs) || + context->mount_flags != 0 || + (context->private_tmp && runtime && (runtime->tmp_dir || runtime->var_tmp_dir)) || + params->bus_endpoint_path || + context->private_devices || + context->protect_system != PROTECT_SYSTEM_NO || + context->protect_home != PROTECT_HOME_NO) { + + char *tmp = NULL, *var = NULL; + + /* The runtime struct only contains the parent + * of the private /tmp, which is + * non-accessible to world users. Inside of it + * there's a /tmp that is sticky, and that's + * the one we want to use here. */ + + if (context->private_tmp && runtime) { + if (runtime->tmp_dir) + tmp = strappenda(runtime->tmp_dir, "/tmp"); + if (runtime->var_tmp_dir) + var = strappenda(runtime->var_tmp_dir, "/tmp"); + } + + err = setup_namespace( + context->read_write_dirs, + context->read_only_dirs, + context->inaccessible_dirs, + tmp, + var, + params->bus_endpoint_path, + context->private_devices, + context->protect_home, + context->protect_system, + context->mount_flags); + + if (err == -EPERM) + log_unit_warning_errno(params->unit_id, err, "Failed to set up file system namespace due to lack of privileges. Execution sandbox will not be in effect: %m"); + else if (err < 0) { + *error = EXIT_NAMESPACE; + return err; + } + } + + if (params->apply_chroot) { + if (context->root_directory) + if (chroot(context->root_directory) < 0) { + *error = EXIT_CHROOT; + return -errno; + } + + if (chdir(context->working_directory ? context->working_directory : "/") < 0) { + *error = EXIT_CHDIR; + return -errno; + } + } else { + _cleanup_free_ char *d = NULL; + + if (asprintf(&d, "%s/%s", + context->root_directory ? context->root_directory : "", + context->working_directory ? context->working_directory : "") < 0) { + *error = EXIT_MEMORY; + return -ENOMEM; + } + + if (chdir(d) < 0) { + *error = EXIT_CHDIR; + return -errno; + } + } + +#ifdef HAVE_SELINUX + if (params->apply_permissions && mac_selinux_use() && params->selinux_context_net && socket_fd >= 0) { + err = mac_selinux_get_child_mls_label(socket_fd, command->path, context->selinux_context, &mac_selinux_context_net); + if (err < 0) { + *error = EXIT_SELINUX_CONTEXT; + return err; + } + } +#endif + + /* We repeat the fd closing here, to make sure that + * nothing is leaked from the PAM modules. Note that + * we are more aggressive this time since socket_fd + * and the netns fds we don't need anymore. The custom + * endpoint fd was needed to upload the policy and can + * now be closed as well. */ + err = close_all_fds(fds, n_fds); + if (err >= 0) + err = shift_fds(fds, n_fds); + if (err >= 0) + err = flags_fds(fds, n_fds, context->non_blocking); + if (err < 0) { + *error = EXIT_FDS; + return err; + } + + if (params->apply_permissions) { + + for (i = 0; i < _RLIMIT_MAX; i++) { + if (!context->rlimit[i]) + continue; + + if (setrlimit_closest(i, context->rlimit[i]) < 0) { + *error = EXIT_LIMITS; + return -errno; + } + } + + if (context->capability_bounding_set_drop) { + err = capability_bounding_set_drop(context->capability_bounding_set_drop, false); + if (err < 0) { + *error = EXIT_CAPABILITIES; + return err; + } + } + +#ifdef HAVE_SMACK + if (context->smack_process_label) { + err = mac_smack_apply_pid(0, context->smack_process_label); + if (err < 0) { + *error = EXIT_SMACK_PROCESS_LABEL; + return err; + } + } +#endif + + if (context->user) { + err = enforce_user(context, uid); + if (err < 0) { + *error = EXIT_USER; + return err; + } + } + + /* PR_GET_SECUREBITS is not privileged, while + * PR_SET_SECUREBITS is. So to suppress + * potential EPERMs we'll try not to call + * PR_SET_SECUREBITS unless necessary. */ + if (prctl(PR_GET_SECUREBITS) != context->secure_bits) + if (prctl(PR_SET_SECUREBITS, context->secure_bits) < 0) { + *error = EXIT_SECUREBITS; + return -errno; + } + + if (context->capabilities) + if (cap_set_proc(context->capabilities) < 0) { + *error = EXIT_CAPABILITIES; + return -errno; + } + + if (context->no_new_privileges) + if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { + *error = EXIT_NO_NEW_PRIVILEGES; + return -errno; + } + +#ifdef HAVE_SECCOMP + if (context->address_families_whitelist || + !set_isempty(context->address_families)) { + err = apply_address_families(context); + if (err < 0) { + *error = EXIT_ADDRESS_FAMILIES; + return err; + } + } + + if (context->syscall_whitelist || + !set_isempty(context->syscall_filter) || + !set_isempty(context->syscall_archs)) { + err = apply_seccomp(context); + if (err < 0) { + *error = EXIT_SECCOMP; + return err; + } + } +#endif + +#ifdef HAVE_SELINUX + if (mac_selinux_use()) { + char *exec_context = mac_selinux_context_net ?: context->selinux_context; + + if (exec_context) { + err = setexeccon(exec_context); + if (err < 0) { + *error = EXIT_SELINUX_CONTEXT; + return err; + } + } + } +#endif + +#ifdef HAVE_APPARMOR + if (context->apparmor_profile && mac_apparmor_use()) { + err = aa_change_onexec(context->apparmor_profile); + if (err < 0 && !context->apparmor_profile_ignore) { + *error = EXIT_APPARMOR_PROFILE; + return -errno; + } + } +#endif + } + + err = build_environment(context, n_fds, params->watchdog_usec, home, username, shell, &our_env); + if (err < 0) { + *error = EXIT_MEMORY; + return err; + } + + final_env = strv_env_merge(5, + params->environment, + our_env, + context->environment, + files_env, + pam_env, + NULL); + if (!final_env) { + *error = EXIT_MEMORY; + return -ENOMEM; + } + + final_argv = replace_env_argv(argv, final_env); + if (!final_argv) { + *error = EXIT_MEMORY; + return -ENOMEM; + } + + final_env = strv_env_clean(final_env); + + if (_unlikely_(log_get_max_level() >= LOG_PRI(LOG_DEBUG))) { + _cleanup_free_ char *line; + + line = exec_command_line(final_argv); + if (line) { + log_open(); + log_unit_struct(params->unit_id, + LOG_DEBUG, + "EXECUTABLE=%s", command->path, + LOG_MESSAGE("Executing: %s", line), + NULL); + log_close(); + } + } + execve(command->path, final_argv, final_env); + *error = EXIT_EXEC; + return -errno; +} + +int exec_spawn(ExecCommand *command, + const ExecContext *context, + const ExecParameters *params, + ExecRuntime *runtime, + pid_t *ret) { + + _cleanup_strv_free_ char **files_env = NULL; + int *fds = NULL; unsigned n_fds = 0; + char *line, **argv; + int socket_fd; + pid_t pid; + int err; + + assert(command); + assert(context); + assert(ret); + assert(params); + assert(params->fds || params->n_fds <= 0); + + if (context->std_input == EXEC_INPUT_SOCKET || + context->std_output == EXEC_OUTPUT_SOCKET || + context->std_error == EXEC_OUTPUT_SOCKET) { + + if (params->n_fds != 1) + return -EINVAL; + + socket_fd = params->fds[0]; + } else { + socket_fd = -1; + fds = params->fds; + n_fds = params->n_fds; + } + + err = exec_context_load_environment(context, params->unit_id, &files_env); + if (err < 0) { + log_unit_struct(params->unit_id, + LOG_ERR, + LOG_MESSAGE("Failed to load environment files: %s", strerror(-err)), + LOG_ERRNO(-err), + NULL); + return err; + } + + argv = params->argv ?: command->argv; + + line = exec_command_line(argv); + if (!line) + return log_oom(); + + log_unit_struct(params->unit_id, + LOG_DEBUG, + "EXECUTABLE=%s", command->path, + LOG_MESSAGE("About to execute: %s", line), + NULL); + free(line); + + pid = fork(); + if (pid < 0) + return -errno; + + if (pid == 0) { + int r; + + err = exec_child(command, + context, + params, + runtime, + argv, + socket_fd, + fds, n_fds, + files_env, + &r); + if (r != 0) { + log_open(); + log_struct(LOG_ERR, + LOG_MESSAGE_ID(SD_MESSAGE_SPAWN_FAILED), + "EXECUTABLE=%s", command->path, + LOG_MESSAGE("Failed at step %s spawning %s: %s", + exit_status_to_string(r, EXIT_STATUS_SYSTEMD), + command->path, strerror(-err)), + LOG_ERRNO(-err), + NULL); + log_close(); + } + + _exit(r); + } + + log_unit_struct(params->unit_id, + LOG_DEBUG, + LOG_MESSAGE("Forked %s as "PID_FMT, + command->path, pid), + NULL); + + /* We add the new process to the cgroup both in the child (so + * that we can be sure that no user code is ever executed + * outside of the cgroup) and in the parent (so that we can be + * sure that when we kill the cgroup the process will be + * killed too). */ + if (params->cgroup_path) + cg_attach(SYSTEMD_CGROUP_CONTROLLER, params->cgroup_path, pid); + + exec_status_start(&command->exec_status, pid); *ret = pid; return 0; @@ -1578,40 +1879,13 @@ void exec_context_init(ExecContext *c) { c->cpu_sched_policy = SCHED_OTHER; c->syslog_priority = LOG_DAEMON|LOG_INFO; c->syslog_level_prefix = true; - c->control_group_persistent = -1; c->ignore_sigpipe = true; - c->timer_slack_nsec = (nsec_t) -1; -} - -void exec_context_tmp_dirs_done(ExecContext *c) { - char* dirs[] = {c->tmp_dir ? c->tmp_dir : c->var_tmp_dir, - c->tmp_dir ? c->var_tmp_dir : NULL, - NULL}; - char **dirp; - - for(dirp = dirs; *dirp; dirp++) { - char *dir; - int r; - - r = rm_rf_dangerous(*dirp, false, true, false); - dir = dirname(*dirp); - if (r < 0) - log_warning("Failed to remove content of temporary directory %s: %s", - dir, strerror(-r)); - else { - r = rmdir(dir); - if (r < 0) - log_warning("Failed to remove temporary directory %s: %s", - dir, strerror(-r)); - } - - free(*dirp); - } - - c->tmp_dir = c->var_tmp_dir = NULL; + c->timer_slack_nsec = NSEC_INFINITY; + c->personality = 0xffffffffUL; + c->runtime_directory_mode = 0755; } -void exec_context_done(ExecContext *c, bool reloading_or_reexecuting) { +void exec_context_done(ExecContext *c) { unsigned l; assert(c); @@ -1635,9 +1909,6 @@ void exec_context_done(ExecContext *c, bool reloading_or_reexecuting) { free(c->tty_path); c->tty_path = NULL; - free(c->tcpwrap_name); - c->tcpwrap_name = NULL; - free(c->syslog_identifier); c->syslog_identifier = NULL; @@ -1673,11 +1944,50 @@ void exec_context_done(ExecContext *c, bool reloading_or_reexecuting) { free(c->utmp_id); c->utmp_id = NULL; - free(c->syscall_filter); + free(c->selinux_context); + c->selinux_context = NULL; + + free(c->apparmor_profile); + c->apparmor_profile = NULL; + + set_free(c->syscall_filter); c->syscall_filter = NULL; - if (!reloading_or_reexecuting) - exec_context_tmp_dirs_done(c); + set_free(c->syscall_archs); + c->syscall_archs = NULL; + + set_free(c->address_families); + c->address_families = NULL; + + strv_free(c->runtime_directory); + c->runtime_directory = NULL; + + bus_endpoint_free(c->bus_endpoint); + c->bus_endpoint = NULL; +} + +int exec_context_destroy_runtime_directory(ExecContext *c, const char *runtime_prefix) { + char **i; + + assert(c); + + if (!runtime_prefix) + return 0; + + STRV_FOREACH(i, c->runtime_directory) { + _cleanup_free_ char *p; + + p = strjoin(runtime_prefix, "/", *i, NULL); + if (!p) + return -ENOMEM; + + /* We execute this synchronously, since we need to be + * sure this is gone when we start the service + * next. */ + rm_rf_dangerous(p, false, true, false); + } + + return 0; } void exec_command_done(ExecCommand *c) { @@ -1701,7 +2011,7 @@ void exec_command_free_list(ExecCommand *c) { ExecCommand *i; while ((i = c)) { - LIST_REMOVE(ExecCommand, command, c, i); + LIST_REMOVE(command, c, i); exec_command_done(i); free(i); } @@ -1716,7 +2026,7 @@ void exec_command_free_array(ExecCommand **c, unsigned n) { } } -int exec_context_load_environment(const ExecContext *c, char ***l) { +int exec_context_load_environment(const ExecContext *c, const char *unit_id, char ***l) { char **i, **r = NULL; assert(c); @@ -1727,7 +2037,7 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { int k; bool ignore = false; char **p; - glob_t _cleanup_globfree_ pglob = {}; + _cleanup_globfree_ glob_t pglob = {}; int count, n; fn = *i; @@ -1763,17 +2073,17 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { return -EINVAL; } for (n = 0; n < count; n++) { - k = load_env_file(pglob.gl_pathv[n], NULL, &p); + k = load_env_file(NULL, pglob.gl_pathv[n], NULL, &p); if (k < 0) { if (ignore) continue; strv_free(r); return k; - } + } /* Log invalid environment variables with filename */ - if (p) - p = strv_env_clean_log(p, pglob.gl_pathv[n]); + if (p) + p = strv_env_clean_log(p, unit_id, pglob.gl_pathv[n]); if (r == NULL) r = p; @@ -1797,8 +2107,8 @@ int exec_context_load_environment(const ExecContext *c, char ***l) { } static bool tty_may_match_dev_console(const char *tty) { - char *active = NULL, *console; - bool b; + _cleanup_free_ char *active = NULL; + char *console; if (startswith(tty, "/dev/")) tty += 5; @@ -1813,10 +2123,7 @@ static bool tty_may_match_dev_console(const char *tty) { return true; /* "tty0" means the active VC, so it may be the same sometimes */ - b = streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty)); - free(active); - - return b; + return streq(console, tty) || (streq(console, "tty0") && tty_is_vc(tty)); } bool exec_context_may_touch_console(ExecContext *ec) { @@ -1837,14 +2144,13 @@ static void strv_fprintf(FILE *f, char **l) { } void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { - char ** e; + char **e; unsigned i; assert(c); assert(f); - if (!prefix) - prefix = ""; + prefix = strempty(prefix); fprintf(f, "%sUMask: %04o\n" @@ -1852,18 +2158,20 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { "%sRootDirectory: %s\n" "%sNonBlocking: %s\n" "%sPrivateTmp: %s\n" - "%sControlGroupModify: %s\n" - "%sControlGroupPersistent: %s\n" "%sPrivateNetwork: %s\n" + "%sPrivateDevices: %s\n" + "%sProtectHome: %s\n" + "%sProtectSystem: %s\n" "%sIgnoreSIGPIPE: %s\n", prefix, c->umask, prefix, c->working_directory ? c->working_directory : "/", prefix, c->root_directory ? c->root_directory : "/", prefix, yes_no(c->non_blocking), prefix, yes_no(c->private_tmp), - prefix, yes_no(c->control_group_modify), - prefix, yes_no(c->control_group_persistent), prefix, yes_no(c->private_network), + prefix, yes_no(c->private_devices), + prefix, protect_home_to_string(c->protect_home), + prefix, protect_system_to_string(c->protect_system), prefix, yes_no(c->ignore_sigpipe)); STRV_FOREACH(e, c->environment) @@ -1872,11 +2180,6 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { STRV_FOREACH(e, c->environment_files) fprintf(f, "%sEnvironmentFile: %s\n", prefix, *e); - if (c->tcpwrap_name) - fprintf(f, - "%sTCPWrapName: %s\n", - prefix, c->tcpwrap_name); - if (c->nice_set) fprintf(f, "%sNice: %i\n", @@ -1889,30 +2192,24 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { for (i = 0; i < RLIM_NLIMITS; i++) if (c->rlimit[i]) - fprintf(f, "%s%s: %llu\n", prefix, rlimit_to_string(i), (unsigned long long) c->rlimit[i]->rlim_max); + fprintf(f, "%s%s: "RLIM_FMT"\n", + prefix, rlimit_to_string(i), c->rlimit[i]->rlim_max); if (c->ioprio_set) { - char *class_str; - int r; + _cleanup_free_ char *class_str = NULL; - r = ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str); - if (r < 0) - class_str = NULL; + ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c->ioprio), &class_str); fprintf(f, "%sIOSchedulingClass: %s\n" "%sIOPriority: %i\n", prefix, strna(class_str), prefix, (int) IOPRIO_PRIO_DATA(c->ioprio)); - free(class_str); } if (c->cpu_sched_set) { - char *policy_str; - int r; + _cleanup_free_ char *policy_str = NULL; - r = sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str); - if (r < 0) - policy_str = NULL; + sched_policy_to_string_alloc(c->cpu_sched_policy, &policy_str); fprintf(f, "%sCPUSchedulingPolicy: %s\n" "%sCPUSchedulingPriority: %i\n" @@ -1920,19 +2217,18 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { prefix, strna(policy_str), prefix, c->cpu_sched_priority, prefix, yes_no(c->cpu_sched_reset_on_fork)); - free(policy_str); } if (c->cpuset) { fprintf(f, "%sCPUAffinity:", prefix); for (i = 0; i < c->cpuset_ncpus; i++) if (CPU_ISSET_S(i, CPU_ALLOC_SIZE(c->cpuset_ncpus), c->cpuset)) - fprintf(f, " %i", i); + fprintf(f, " %u", i); fputs("\n", f); } - if (c->timer_slack_nsec != (nsec_t) -1) - fprintf(f, "%sTimerSlackNSec: %lu\n", prefix, (unsigned long)c->timer_slack_nsec); + if (c->timer_slack_nsec != NSEC_INFINITY) + fprintf(f, "%sTimerSlackNSec: "NSEC_FMT "\n", prefix, c->timer_slack_nsec); fprintf(f, "%sStandardInput: %s\n" @@ -1953,37 +2249,37 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { prefix, yes_no(c->tty_vhangup), prefix, yes_no(c->tty_vt_disallocate)); - if (c->std_output == EXEC_OUTPUT_SYSLOG || c->std_output == EXEC_OUTPUT_KMSG || c->std_output == EXEC_OUTPUT_JOURNAL || - c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE || - c->std_error == EXEC_OUTPUT_SYSLOG || c->std_error == EXEC_OUTPUT_KMSG || c->std_error == EXEC_OUTPUT_JOURNAL || - c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) { - char *fac_str, *lvl_str; - int r; + if (c->std_output == EXEC_OUTPUT_SYSLOG || + c->std_output == EXEC_OUTPUT_KMSG || + c->std_output == EXEC_OUTPUT_JOURNAL || + c->std_output == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || + c->std_output == EXEC_OUTPUT_KMSG_AND_CONSOLE || + c->std_output == EXEC_OUTPUT_JOURNAL_AND_CONSOLE || + c->std_error == EXEC_OUTPUT_SYSLOG || + c->std_error == EXEC_OUTPUT_KMSG || + c->std_error == EXEC_OUTPUT_JOURNAL || + c->std_error == EXEC_OUTPUT_SYSLOG_AND_CONSOLE || + c->std_error == EXEC_OUTPUT_KMSG_AND_CONSOLE || + c->std_error == EXEC_OUTPUT_JOURNAL_AND_CONSOLE) { - r = log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str); - if (r < 0) - fac_str = NULL; + _cleanup_free_ char *fac_str = NULL, *lvl_str = NULL; - r = log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str); - if (r < 0) - lvl_str = NULL; + log_facility_unshifted_to_string_alloc(c->syslog_priority >> 3, &fac_str); + log_level_to_string_alloc(LOG_PRI(c->syslog_priority), &lvl_str); fprintf(f, "%sSyslogFacility: %s\n" "%sSyslogLevel: %s\n", prefix, strna(fac_str), prefix, strna(lvl_str)); - free(lvl_str); - free(fac_str); } if (c->capabilities) { - char *t; - if ((t = cap_to_text(c->capabilities, NULL))) { - fprintf(f, "%sCapabilities: %s\n", - prefix, t); - cap_free(t); - } + _cleanup_cap_free_charp_ char *t; + + t = cap_to_text(c->capabilities, NULL); + if (t) + fprintf(f, "%sCapabilities: %s\n", prefix, t); } if (c->secure_bits) @@ -2001,14 +2297,8 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { fprintf(f, "%sCapabilityBoundingSet:", prefix); for (l = 0; l <= cap_last_cap(); l++) - if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) { - char *t; - - if ((t = cap_to_name(l))) { - fprintf(f, " %s", t); - cap_free(t); - } - } + if (!(c->capability_bounding_set_drop & ((uint64_t) 1ULL << (uint64_t) l))) + fprintf(f, " %s", strna(capability_to_name(l))); fputs("\n", f); } @@ -2049,6 +2339,89 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) { fprintf(f, "%sUtmpIdentifier: %s\n", prefix, c->utmp_id); + + if (c->selinux_context) + fprintf(f, + "%sSELinuxContext: %s%s\n", + prefix, c->selinux_context_ignore ? "-" : "", c->selinux_context); + + if (c->personality != 0xffffffffUL) + fprintf(f, + "%sPersonality: %s\n", + prefix, strna(personality_to_string(c->personality))); + + if (c->syscall_filter) { +#ifdef HAVE_SECCOMP + Iterator j; + void *id; + bool first = true; +#endif + + fprintf(f, + "%sSystemCallFilter: ", + prefix); + + if (!c->syscall_whitelist) + fputc('~', f); + +#ifdef HAVE_SECCOMP + SET_FOREACH(id, c->syscall_filter, j) { + _cleanup_free_ char *name = NULL; + + if (first) + first = false; + else + fputc(' ', f); + + name = seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE, PTR_TO_INT(id) - 1); + fputs(strna(name), f); + } +#endif + + fputc('\n', f); + } + + if (c->syscall_archs) { +#ifdef HAVE_SECCOMP + Iterator j; + void *id; +#endif + + fprintf(f, + "%sSystemCallArchitectures:", + prefix); + +#ifdef HAVE_SECCOMP + SET_FOREACH(id, c->syscall_archs, j) + fprintf(f, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id) - 1))); +#endif + fputc('\n', f); + } + + if (c->syscall_errno != 0) + fprintf(f, + "%sSystemCallErrorNumber: %s\n", + prefix, strna(errno_to_name(c->syscall_errno))); + + if (c->apparmor_profile) + fprintf(f, + "%sAppArmorProfile: %s%s\n", + prefix, c->apparmor_profile_ignore ? "-" : "", c->apparmor_profile); +} + +bool exec_context_maintains_privileges(ExecContext *c) { + assert(c); + + /* Returns true if the process forked off would run run under + * an unchanged UID or as root. */ + + if (!c->user) + return true; + + if (streq(c->user, "root") || streq(c->user, "0")) + return true; + + return false; } void exec_status_start(ExecStatus *s, pid_t pid) { @@ -2085,15 +2458,14 @@ void exec_status_dump(ExecStatus *s, FILE *f, const char *prefix) { assert(s); assert(f); - if (!prefix) - prefix = ""; - if (s->pid <= 0) return; + prefix = strempty(prefix); + fprintf(f, - "%sPID: %lu\n", - prefix, (unsigned long) s->pid); + "%sPID: "PID_FMT"\n", + prefix, s->pid); if (s->start_timestamp.realtime > 0) fprintf(f, @@ -2150,37 +2522,27 @@ char *exec_command_line(char **argv) { } void exec_command_dump(ExecCommand *c, FILE *f, const char *prefix) { - char *p2; + _cleanup_free_ char *cmd = NULL; const char *prefix2; - char *cmd; - assert(c); assert(f); - if (!prefix) - prefix = ""; - p2 = strappend(prefix, "\t"); - prefix2 = p2 ? p2 : prefix; + prefix = strempty(prefix); + prefix2 = strappenda(prefix, "\t"); cmd = exec_command_line(c->argv); - fprintf(f, "%sCommand Line: %s\n", prefix, cmd ? cmd : strerror(ENOMEM)); - free(cmd); - exec_status_dump(&c->exec_status, f, prefix2); - - free(p2); } void exec_command_dump_list(ExecCommand *c, FILE *f, const char *prefix) { assert(f); - if (!prefix) - prefix = ""; + prefix = strempty(prefix); LIST_FOREACH(command, c, c) exec_command_dump(c, f, prefix); @@ -2194,8 +2556,8 @@ void exec_command_append_list(ExecCommand **l, ExecCommand *e) { if (*l) { /* It's kind of important, that we keep the order here */ - LIST_FIND_TAIL(ExecCommand, command, *l, end); - LIST_INSERT_AFTER(ExecCommand, command, *l, end, e); + LIST_FIND_TAIL(command, *l, end); + LIST_INSERT_AFTER(command, *l, end, e); } else *l = e; } @@ -2214,7 +2576,8 @@ int exec_command_set(ExecCommand *c, const char *path, ...) { if (!l) return -ENOMEM; - if (!(p = strdup(path))) { + p = strdup(path); + if (!p) { strv_free(l); return -ENOMEM; } @@ -2228,6 +2591,249 @@ int exec_command_set(ExecCommand *c, const char *path, ...) { return 0; } +int exec_command_append(ExecCommand *c, const char *path, ...) { + _cleanup_strv_free_ char **l = NULL; + va_list ap; + int r; + + assert(c); + assert(path); + + va_start(ap, path); + l = strv_new_ap(path, ap); + va_end(ap); + + if (!l) + return -ENOMEM; + + r = strv_extend_strv(&c->argv, l); + if (r < 0) + return r; + + return 0; +} + + +static int exec_runtime_allocate(ExecRuntime **rt) { + + if (*rt) + return 0; + + *rt = new0(ExecRuntime, 1); + if (!*rt) + return -ENOMEM; + + (*rt)->n_ref = 1; + (*rt)->netns_storage_socket[0] = (*rt)->netns_storage_socket[1] = -1; + + return 0; +} + +int exec_runtime_make(ExecRuntime **rt, ExecContext *c, const char *id) { + int r; + + assert(rt); + assert(c); + assert(id); + + if (*rt) + return 1; + + if (!c->private_network && !c->private_tmp) + return 0; + + r = exec_runtime_allocate(rt); + if (r < 0) + return r; + + if (c->private_network && (*rt)->netns_storage_socket[0] < 0) { + if (socketpair(AF_UNIX, SOCK_DGRAM, 0, (*rt)->netns_storage_socket) < 0) + return -errno; + } + + if (c->private_tmp && !(*rt)->tmp_dir) { + r = setup_tmp_dirs(id, &(*rt)->tmp_dir, &(*rt)->var_tmp_dir); + if (r < 0) + return r; + } + + return 1; +} + +ExecRuntime *exec_runtime_ref(ExecRuntime *r) { + assert(r); + assert(r->n_ref > 0); + + r->n_ref++; + return r; +} + +ExecRuntime *exec_runtime_unref(ExecRuntime *r) { + + if (!r) + return NULL; + + assert(r->n_ref > 0); + + r->n_ref--; + if (r->n_ref <= 0) { + free(r->tmp_dir); + free(r->var_tmp_dir); + safe_close_pair(r->netns_storage_socket); + free(r); + } + + return NULL; +} + +int exec_runtime_serialize(ExecRuntime *rt, Unit *u, FILE *f, FDSet *fds) { + assert(u); + assert(f); + assert(fds); + + if (!rt) + return 0; + + if (rt->tmp_dir) + unit_serialize_item(u, f, "tmp-dir", rt->tmp_dir); + + if (rt->var_tmp_dir) + unit_serialize_item(u, f, "var-tmp-dir", rt->var_tmp_dir); + + if (rt->netns_storage_socket[0] >= 0) { + int copy; + + copy = fdset_put_dup(fds, rt->netns_storage_socket[0]); + if (copy < 0) + return copy; + + unit_serialize_item_format(u, f, "netns-socket-0", "%i", copy); + } + + if (rt->netns_storage_socket[1] >= 0) { + int copy; + + copy = fdset_put_dup(fds, rt->netns_storage_socket[1]); + if (copy < 0) + return copy; + + unit_serialize_item_format(u, f, "netns-socket-1", "%i", copy); + } + + return 0; +} + +int exec_runtime_deserialize_item(ExecRuntime **rt, Unit *u, const char *key, const char *value, FDSet *fds) { + int r; + + assert(rt); + assert(key); + assert(value); + + if (streq(key, "tmp-dir")) { + char *copy; + + r = exec_runtime_allocate(rt); + if (r < 0) + return r; + + copy = strdup(value); + if (!copy) + return log_oom(); + + free((*rt)->tmp_dir); + (*rt)->tmp_dir = copy; + + } else if (streq(key, "var-tmp-dir")) { + char *copy; + + r = exec_runtime_allocate(rt); + if (r < 0) + return r; + + copy = strdup(value); + if (!copy) + return log_oom(); + + free((*rt)->var_tmp_dir); + (*rt)->var_tmp_dir = copy; + + } else if (streq(key, "netns-socket-0")) { + int fd; + + r = exec_runtime_allocate(rt); + if (r < 0) + return r; + + if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd)) + log_unit_debug(u->id, "Failed to parse netns socket value %s", value); + else { + safe_close((*rt)->netns_storage_socket[0]); + (*rt)->netns_storage_socket[0] = fdset_remove(fds, fd); + } + } else if (streq(key, "netns-socket-1")) { + int fd; + + r = exec_runtime_allocate(rt); + if (r < 0) + return r; + + if (safe_atoi(value, &fd) < 0 || !fdset_contains(fds, fd)) + log_unit_debug(u->id, "Failed to parse netns socket value %s", value); + else { + safe_close((*rt)->netns_storage_socket[1]); + (*rt)->netns_storage_socket[1] = fdset_remove(fds, fd); + } + } else + return 0; + + return 1; +} + +static void *remove_tmpdir_thread(void *p) { + _cleanup_free_ char *path = p; + + rm_rf_dangerous(path, false, true, false); + return NULL; +} + +void exec_runtime_destroy(ExecRuntime *rt) { + int r; + + if (!rt) + return; + + /* If there are multiple users of this, let's leave the stuff around */ + if (rt->n_ref > 1) + return; + + if (rt->tmp_dir) { + log_debug("Spawning thread to nuke %s", rt->tmp_dir); + + r = asynchronous_job(remove_tmpdir_thread, rt->tmp_dir); + if (r < 0) { + log_warning_errno(r, "Failed to nuke %s: %m", rt->tmp_dir); + free(rt->tmp_dir); + } + + rt->tmp_dir = NULL; + } + + if (rt->var_tmp_dir) { + log_debug("Spawning thread to nuke %s", rt->var_tmp_dir); + + r = asynchronous_job(remove_tmpdir_thread, rt->var_tmp_dir); + if (r < 0) { + log_warning_errno(r, "Failed to nuke %s: %m", rt->var_tmp_dir); + free(rt->var_tmp_dir); + } + + rt->var_tmp_dir = NULL; + } + + safe_close_pair(rt->netns_storage_socket); +} + static const char* const exec_input_table[_EXEC_INPUT_MAX] = { [EXEC_INPUT_NULL] = "null", [EXEC_INPUT_TTY] = "tty",