+ * Parameters: proc_t p process to apply attributes to
+ * int psa_apptype posix spawn attribute apptype
+ *
+ * Returns: 0 Success
+ */
+static errno_t
+exec_handle_spawnattr_policy(proc_t p, thread_t thread, int psa_apptype, uint64_t psa_qos_clamp,
+ uint64_t psa_darwin_role, struct exec_port_actions *port_actions)
+{
+ int apptype = TASK_APPTYPE_NONE;
+ int qos_clamp = THREAD_QOS_UNSPECIFIED;
+ int role = TASK_UNSPECIFIED;
+
+ if ((psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK) != 0) {
+ int proctype = psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK;
+
+ switch (proctype) {
+ case POSIX_SPAWN_PROC_TYPE_DAEMON_INTERACTIVE:
+ apptype = TASK_APPTYPE_DAEMON_INTERACTIVE;
+ break;
+ case POSIX_SPAWN_PROC_TYPE_DAEMON_STANDARD:
+ apptype = TASK_APPTYPE_DAEMON_STANDARD;
+ break;
+ case POSIX_SPAWN_PROC_TYPE_DAEMON_ADAPTIVE:
+ apptype = TASK_APPTYPE_DAEMON_ADAPTIVE;
+ break;
+ case POSIX_SPAWN_PROC_TYPE_DAEMON_BACKGROUND:
+ apptype = TASK_APPTYPE_DAEMON_BACKGROUND;
+ break;
+ case POSIX_SPAWN_PROC_TYPE_APP_DEFAULT:
+ apptype = TASK_APPTYPE_APP_DEFAULT;
+ break;
+#if !CONFIG_EMBEDDED
+ case POSIX_SPAWN_PROC_TYPE_APP_TAL:
+ apptype = TASK_APPTYPE_APP_TAL;
+ break;
+#endif /* !CONFIG_EMBEDDED */
+ case POSIX_SPAWN_PROC_TYPE_DRIVER:
+ apptype = TASK_APPTYPE_DRIVER;
+ break;
+ default:
+ apptype = TASK_APPTYPE_NONE;
+ /* TODO: Should an invalid value here fail the spawn? */
+ break;
+ }
+ }
+
+ if (psa_qos_clamp != POSIX_SPAWN_PROC_CLAMP_NONE) {
+ switch (psa_qos_clamp) {
+ case POSIX_SPAWN_PROC_CLAMP_UTILITY:
+ qos_clamp = THREAD_QOS_UTILITY;
+ break;
+ case POSIX_SPAWN_PROC_CLAMP_BACKGROUND:
+ qos_clamp = THREAD_QOS_BACKGROUND;
+ break;
+ case POSIX_SPAWN_PROC_CLAMP_MAINTENANCE:
+ qos_clamp = THREAD_QOS_MAINTENANCE;
+ break;
+ default:
+ qos_clamp = THREAD_QOS_UNSPECIFIED;
+ /* TODO: Should an invalid value here fail the spawn? */
+ break;
+ }
+ }
+
+ if (psa_darwin_role != PRIO_DARWIN_ROLE_DEFAULT) {
+ proc_darwin_role_to_task_role(psa_darwin_role, &role);
+ }
+
+ if (apptype != TASK_APPTYPE_NONE ||
+ qos_clamp != THREAD_QOS_UNSPECIFIED ||
+ role != TASK_UNSPECIFIED ||
+ port_actions->portwatch_count) {
+ proc_set_task_spawnpolicy(p->task, thread, apptype, qos_clamp, role,
+ port_actions->portwatch_array, port_actions->portwatch_count);
+ }
+
+ if (port_actions->registered_count) {
+ if (mach_ports_register(p->task, port_actions->registered_array,
+ port_actions->registered_count)) {
+ return EINVAL;
+ }
+ /* mach_ports_register() consumed the array */
+ port_actions->registered_array = NULL;
+ port_actions->registered_count = 0;
+ }
+
+ return 0;
+}
+
+static void
+exec_port_actions_destroy(struct exec_port_actions *port_actions)
+{
+ if (port_actions->portwatch_array) {
+ for (uint32_t i = 0; i < port_actions->portwatch_count; i++) {
+ ipc_port_t port = NULL;
+ if ((port = port_actions->portwatch_array[i]) != NULL) {
+ ipc_port_release_send(port);
+ }
+ }
+ kfree(port_actions->portwatch_array,
+ port_actions->portwatch_count * sizeof(ipc_port_t *));
+ }
+
+ if (port_actions->registered_array) {
+ for (uint32_t i = 0; i < port_actions->registered_count; i++) {
+ ipc_port_t port = NULL;
+ if ((port = port_actions->registered_array[i]) != NULL) {
+ ipc_port_release_send(port);
+ }
+ }
+ kfree(port_actions->registered_array,
+ port_actions->registered_count * sizeof(ipc_port_t *));
+ }
+}
+
+/*
+ * exec_handle_port_actions
+ *
+ * Description: Go through the _posix_port_actions_t contents,
+ * calling task_set_special_port, task_set_exception_ports
+ * and/or audit_session_spawnjoin for the current task.
+ *
+ * Parameters: struct image_params * Image parameter block
+ *
+ * Returns: 0 Success
+ * EINVAL Failure
+ * ENOTSUP Illegal posix_spawn attr flag was set
+ */
+static errno_t
+exec_handle_port_actions(struct image_params *imgp,
+ struct exec_port_actions *actions)
+{
+ _posix_spawn_port_actions_t pacts = imgp->ip_px_spa;
+#if CONFIG_AUDIT
+ proc_t p = vfs_context_proc(imgp->ip_vfs_context);
+#endif
+ _ps_port_action_t *act = NULL;
+ task_t task = get_threadtask(imgp->ip_new_thread);
+ ipc_port_t port = NULL;
+ errno_t ret = 0;
+ int i, portwatch_i = 0, registered_i = 0;
+ kern_return_t kr;
+ boolean_t task_has_watchport_boost = task_has_watchports(current_task());
+ boolean_t in_exec = (imgp->ip_flags & IMGPF_EXEC);
+ boolean_t suid_cred_specified = FALSE;
+
+ for (i = 0; i < pacts->pspa_count; i++) {
+ act = &pacts->pspa_actions[i];
+
+ switch (act->port_type) {
+ case PSPA_SPECIAL:
+ case PSPA_EXCEPTION:
+#if CONFIG_AUDIT
+ case PSPA_AU_SESSION:
+#endif
+ break;
+ case PSPA_IMP_WATCHPORTS:
+ if (++actions->portwatch_count > TASK_MAX_WATCHPORT_COUNT) {
+ ret = EINVAL;
+ goto done;
+ }
+ break;
+ case PSPA_REGISTERED_PORTS:
+ if (++actions->registered_count > TASK_PORT_REGISTER_MAX) {
+ ret = EINVAL;
+ goto done;
+ }
+ break;
+
+ case PSPA_SUID_CRED:
+ /* Only a single suid credential can be specified. */
+ if (suid_cred_specified) {
+ ret = EINVAL;
+ goto done;
+ }
+ suid_cred_specified = TRUE;
+ break;
+
+ default:
+ ret = EINVAL;
+ goto done;
+ }
+ }
+
+ if (actions->portwatch_count) {
+ if (in_exec && task_has_watchport_boost) {
+ ret = EINVAL;
+ goto done;
+ }
+ actions->portwatch_array =
+ kalloc(sizeof(ipc_port_t *) * actions->portwatch_count);
+ if (actions->portwatch_array == NULL) {
+ ret = ENOMEM;
+ goto done;
+ }
+ bzero(actions->portwatch_array,
+ sizeof(ipc_port_t *) * actions->portwatch_count);
+ }
+
+ if (actions->registered_count) {
+ actions->registered_array =
+ kalloc(sizeof(ipc_port_t *) * actions->registered_count);
+ if (actions->registered_array == NULL) {
+ ret = ENOMEM;
+ goto done;
+ }
+ bzero(actions->registered_array,
+ sizeof(ipc_port_t *) * actions->registered_count);
+ }
+
+ for (i = 0; i < pacts->pspa_count; i++) {
+ act = &pacts->pspa_actions[i];
+
+ if (MACH_PORT_VALID(act->new_port)) {
+ kr = ipc_object_copyin(get_task_ipcspace(current_task()),
+ act->new_port, MACH_MSG_TYPE_COPY_SEND,
+ (ipc_object_t *) &port, 0, NULL, IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND);
+
+ if (kr != KERN_SUCCESS) {
+ ret = EINVAL;
+ goto done;
+ }
+ } else {
+ /* it's NULL or DEAD */
+ port = CAST_MACH_NAME_TO_PORT(act->new_port);
+ }
+
+ switch (act->port_type) {
+ case PSPA_SPECIAL:
+ kr = task_set_special_port(task, act->which, port);
+
+ if (kr != KERN_SUCCESS) {
+ ret = EINVAL;
+ }
+ break;
+
+ case PSPA_EXCEPTION:
+ kr = task_set_exception_ports(task, act->mask, port,
+ act->behavior, act->flavor);
+ if (kr != KERN_SUCCESS) {
+ ret = EINVAL;
+ }
+ break;
+#if CONFIG_AUDIT
+ case PSPA_AU_SESSION:
+ ret = audit_session_spawnjoin(p, task, port);
+ if (ret) {
+ /* audit_session_spawnjoin() has already dropped the reference in case of error. */
+ goto done;
+ }
+
+ break;
+#endif
+ case PSPA_IMP_WATCHPORTS:
+ if (actions->portwatch_array) {
+ /* hold on to this till end of spawn */
+ actions->portwatch_array[portwatch_i++] = port;
+ } else {
+ ipc_port_release_send(port);
+ }
+ break;
+ case PSPA_REGISTERED_PORTS:
+ /* hold on to this till end of spawn */
+ actions->registered_array[registered_i++] = port;
+ break;
+
+ case PSPA_SUID_CRED:
+ imgp->ip_sc_port = port;
+ break;
+
+ default:
+ ret = EINVAL;
+ break;
+ }
+
+ if (ret) {
+ /* action failed, so release port resources */
+ ipc_port_release_send(port);
+ break;
+ }
+ }
+
+done:
+ if (0 != ret) {
+ DTRACE_PROC1(spawn__port__failure, mach_port_name_t, act->new_port);
+ }
+ return ret;
+}
+
+/*
+ * exec_handle_file_actions
+ *
+ * Description: Go through the _posix_file_actions_t contents applying the
+ * open, close, and dup2 operations to the open file table for
+ * the current process.
+ *
+ * Parameters: struct image_params * Image parameter block
+ *
+ * Returns: 0 Success
+ * ???
+ *
+ * Note: Actions are applied in the order specified, with the credential
+ * of the parent process. This is done to permit the parent
+ * process to utilize POSIX_SPAWN_RESETIDS to drop privilege in
+ * the child following operations the child may in fact not be
+ * normally permitted to perform.
+ */
+static int
+exec_handle_file_actions(struct image_params *imgp, short psa_flags)
+{
+ int error = 0;
+ int action;
+ proc_t p = vfs_context_proc(imgp->ip_vfs_context);
+ _posix_spawn_file_actions_t px_sfap = imgp->ip_px_sfa;
+ int ival[2]; /* dummy retval for system calls) */
+#if CONFIG_AUDIT
+ struct uthread *uthread = get_bsdthread_info(current_thread());
+#endif
+
+ for (action = 0; action < px_sfap->psfa_act_count; action++) {
+ _psfa_action_t *psfa = &px_sfap->psfa_act_acts[action];
+
+ switch (psfa->psfaa_type) {
+ case PSFA_OPEN: {
+ /*
+ * Open is different, in that it requires the use of
+ * a path argument, which is normally copied in from
+ * user space; because of this, we have to support an
+ * open from kernel space that passes an address space
+ * context of UIO_SYSSPACE, and casts the address
+ * argument to a user_addr_t.
+ */
+ char *bufp = NULL;
+ struct vnode_attr *vap;
+ struct nameidata *ndp;
+ int mode = psfa->psfaa_openargs.psfao_mode;
+ struct dup2_args dup2a;
+ struct close_nocancel_args ca;
+ int origfd;
+
+ MALLOC(bufp, char *, sizeof(*vap) + sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO);
+ if (bufp == NULL) {
+ error = ENOMEM;
+ break;
+ }
+
+ vap = (struct vnode_attr *) bufp;
+ ndp = (struct nameidata *) (bufp + sizeof(*vap));
+
+ VATTR_INIT(vap);
+ /* Mask off all but regular access permissions */
+ mode = ((mode & ~p->p_fd->fd_cmask) & ALLPERMS) & ~S_ISTXT;
+ VATTR_SET(vap, va_mode, mode & ACCESSPERMS);
+
+ AUDIT_SUBCALL_ENTER(OPEN, p, uthread);
+
+ NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW | AUDITVNPATH1, UIO_SYSSPACE,
+ CAST_USER_ADDR_T(psfa->psfaa_openargs.psfao_path),
+ imgp->ip_vfs_context);
+
+ error = open1(imgp->ip_vfs_context,
+ ndp,
+ psfa->psfaa_openargs.psfao_oflag,
+ vap,
+ fileproc_alloc_init, NULL,
+ ival);
+
+ FREE(bufp, M_TEMP);
+
+ AUDIT_SUBCALL_EXIT(uthread, error);
+
+ /*
+ * If there's an error, or we get the right fd by
+ * accident, then drop out here. This is easier than
+ * reworking all the open code to preallocate fd
+ * slots, and internally taking one as an argument.
+ */
+ if (error || ival[0] == psfa->psfaa_filedes) {
+ break;
+ }
+
+ origfd = ival[0];
+ /*
+ * If we didn't fall out from an error, we ended up
+ * with the wrong fd; so now we've got to try to dup2
+ * it to the right one.
+ */
+ dup2a.from = origfd;
+ dup2a.to = psfa->psfaa_filedes;
+
+ /*
+ * The dup2() system call implementation sets
+ * ival to newfd in the success case, but we
+ * can ignore that, since if we didn't get the
+ * fd we wanted, the error will stop us.
+ */
+ AUDIT_SUBCALL_ENTER(DUP2, p, uthread);
+ error = dup2(p, &dup2a, ival);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ if (error) {
+ break;
+ }
+
+ /*
+ * Finally, close the original fd.
+ */
+ ca.fd = origfd;
+
+ AUDIT_SUBCALL_ENTER(CLOSE, p, uthread);
+ error = close_nocancel(p, &ca, ival);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ }
+ break;
+
+ case PSFA_DUP2: {
+ struct dup2_args dup2a;
+
+ dup2a.from = psfa->psfaa_filedes;
+ dup2a.to = psfa->psfaa_dup2args.psfad_newfiledes;
+
+ /*
+ * The dup2() system call implementation sets
+ * ival to newfd in the success case, but we
+ * can ignore that, since if we didn't get the
+ * fd we wanted, the error will stop us.
+ */
+ AUDIT_SUBCALL_ENTER(DUP2, p, uthread);
+ error = dup2(p, &dup2a, ival);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ }
+ break;
+
+ case PSFA_FILEPORT_DUP2: {
+ ipc_port_t port;
+ kern_return_t kr;
+ struct dup2_args dup2a;
+ struct close_nocancel_args ca;
+
+ if (!MACH_PORT_VALID(psfa->psfaa_fileport)) {
+ error = EINVAL;
+ break;
+ }
+
+ kr = ipc_object_copyin(get_task_ipcspace(current_task()),
+ psfa->psfaa_fileport, MACH_MSG_TYPE_COPY_SEND,
+ (ipc_object_t *) &port, 0, NULL, IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND);
+
+ if (kr != KERN_SUCCESS) {
+ error = EINVAL;
+ break;
+ }
+
+ error = fileport_makefd_internal(p, port, 0, ival);
+
+ if (IPC_PORT_NULL != port) {
+ ipc_port_release_send(port);
+ }
+
+ if (error || ival[0] == psfa->psfaa_dup2args.psfad_newfiledes) {
+ break;
+ }
+
+ dup2a.from = ca.fd = ival[0];
+ dup2a.to = psfa->psfaa_dup2args.psfad_newfiledes;
+ AUDIT_SUBCALL_ENTER(DUP2, p, uthread);
+ error = dup2(p, &dup2a, ival);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ if (error) {
+ break;
+ }
+
+ AUDIT_SUBCALL_ENTER(CLOSE, p, uthread);
+ error = close_nocancel(p, &ca, ival);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ }
+ break;
+
+ case PSFA_CLOSE: {
+ struct close_nocancel_args ca;
+
+ ca.fd = psfa->psfaa_filedes;
+
+ AUDIT_SUBCALL_ENTER(CLOSE, p, uthread);
+ error = close_nocancel(p, &ca, ival);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ }
+ break;
+
+ case PSFA_INHERIT: {
+ struct fcntl_nocancel_args fcntla;
+
+ /*
+ * Check to see if the descriptor exists, and
+ * ensure it's -not- marked as close-on-exec.
+ *
+ * Attempting to "inherit" a guarded fd will
+ * result in a error.
+ */
+ fcntla.fd = psfa->psfaa_filedes;
+ fcntla.cmd = F_GETFD;
+ if ((error = fcntl_nocancel(p, &fcntla, ival)) != 0) {
+ break;
+ }
+
+ if ((ival[0] & FD_CLOEXEC) == FD_CLOEXEC) {
+ fcntla.fd = psfa->psfaa_filedes;
+ fcntla.cmd = F_SETFD;
+ fcntla.arg = ival[0] & ~FD_CLOEXEC;
+ error = fcntl_nocancel(p, &fcntla, ival);
+ }
+ }
+ break;
+
+ case PSFA_CHDIR: {
+ /*
+ * Chdir is different, in that it requires the use of
+ * a path argument, which is normally copied in from
+ * user space; because of this, we have to support a
+ * chdir from kernel space that passes an address space
+ * context of UIO_SYSSPACE, and casts the address
+ * argument to a user_addr_t.
+ */
+ struct nameidata nd;
+
+ AUDIT_SUBCALL_ENTER(CHDIR, p, uthread);
+ NDINIT(&nd, LOOKUP, OP_CHDIR, FOLLOW | AUDITVNPATH1, UIO_SYSSPACE,
+ CAST_USER_ADDR_T(psfa->psfaa_chdirargs.psfac_path),
+ imgp->ip_vfs_context);
+
+ error = chdir_internal(p, imgp->ip_vfs_context, &nd, 0);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ }
+ break;
+
+ case PSFA_FCHDIR: {
+ struct fchdir_args fchdira;
+
+ fchdira.fd = psfa->psfaa_filedes;
+
+ AUDIT_SUBCALL_ENTER(FCHDIR, p, uthread);
+ error = fchdir(p, &fchdira, ival);
+ AUDIT_SUBCALL_EXIT(uthread, error);
+ }
+ break;
+
+ default:
+ error = EINVAL;
+ break;
+ }
+
+ /* All file actions failures are considered fatal, per POSIX */
+
+ if (error) {
+ if (PSFA_OPEN == psfa->psfaa_type) {
+ DTRACE_PROC1(spawn__open__failure, uintptr_t,
+ psfa->psfaa_openargs.psfao_path);
+ } else {
+ DTRACE_PROC1(spawn__fd__failure, int, psfa->psfaa_filedes);
+ }
+ break;
+ }
+ }
+
+ if (error != 0 || (psa_flags & POSIX_SPAWN_CLOEXEC_DEFAULT) == 0) {
+ return error;
+ }
+
+ /*
+ * If POSIX_SPAWN_CLOEXEC_DEFAULT is set, behave (during
+ * this spawn only) as if "close on exec" is the default
+ * disposition of all pre-existing file descriptors. In this case,
+ * the list of file descriptors mentioned in the file actions
+ * are the only ones that can be inherited, so mark them now.
+ *
+ * The actual closing part comes later, in fdexec().
+ */
+ proc_fdlock(p);
+ for (action = 0; action < px_sfap->psfa_act_count; action++) {
+ _psfa_action_t *psfa = &px_sfap->psfa_act_acts[action];
+ int fd = psfa->psfaa_filedes;
+
+ switch (psfa->psfaa_type) {
+ case PSFA_DUP2:
+ case PSFA_FILEPORT_DUP2:
+ fd = psfa->psfaa_dup2args.psfad_newfiledes;
+ /*FALLTHROUGH*/
+ case PSFA_OPEN:
+ case PSFA_INHERIT:
+ *fdflags(p, fd) |= UF_INHERIT;
+ break;
+
+ case PSFA_CLOSE:
+ case PSFA_CHDIR:
+ case PSFA_FCHDIR:
+ /*
+ * Although PSFA_FCHDIR does have a file descriptor, it is not
+ * *creating* one, thus we do not automatically mark it for
+ * inheritance under POSIX_SPAWN_CLOEXEC_DEFAULT. A client that
+ * wishes it to be inherited should use the PSFA_INHERIT action
+ * explicitly.
+ */
+ break;
+ }
+ }
+ proc_fdunlock(p);
+
+ return 0;
+}
+
+#if CONFIG_MACF
+/*
+ * exec_spawnattr_getmacpolicyinfo
+ */
+void *
+exec_spawnattr_getmacpolicyinfo(const void *macextensions, const char *policyname, size_t *lenp)
+{
+ const struct _posix_spawn_mac_policy_extensions *psmx = macextensions;
+ int i;
+
+ if (psmx == NULL) {
+ return NULL;
+ }
+
+ for (i = 0; i < psmx->psmx_count; i++) {
+ const _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[i];
+ if (strncmp(extension->policyname, policyname, sizeof(extension->policyname)) == 0) {
+ if (lenp != NULL) {
+ *lenp = extension->datalen;
+ }
+ return extension->datap;
+ }
+ }
+
+ if (lenp != NULL) {
+ *lenp = 0;
+ }
+ return NULL;
+}
+
+static int
+spawn_copyin_macpolicyinfo(const struct user__posix_spawn_args_desc *px_args, _posix_spawn_mac_policy_extensions_t *psmxp)
+{
+ _posix_spawn_mac_policy_extensions_t psmx = NULL;
+ int error = 0;
+ int copycnt = 0;
+ int i = 0;
+
+ *psmxp = NULL;
+
+ if (px_args->mac_extensions_size < PS_MAC_EXTENSIONS_SIZE(1) ||
+ px_args->mac_extensions_size > PAGE_SIZE) {
+ error = EINVAL;
+ goto bad;
+ }
+
+ MALLOC(psmx, _posix_spawn_mac_policy_extensions_t, px_args->mac_extensions_size, M_TEMP, M_WAITOK);
+ if ((error = copyin(px_args->mac_extensions, psmx, px_args->mac_extensions_size)) != 0) {
+ goto bad;
+ }
+
+ size_t extsize = PS_MAC_EXTENSIONS_SIZE(psmx->psmx_count);
+ if (extsize == 0 || extsize > px_args->mac_extensions_size) {
+ error = EINVAL;
+ goto bad;
+ }
+
+ for (i = 0; i < psmx->psmx_count; i++) {
+ _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[i];
+ if (extension->datalen == 0 || extension->datalen > PAGE_SIZE) {
+ error = EINVAL;
+ goto bad;
+ }
+ }
+
+ for (copycnt = 0; copycnt < psmx->psmx_count; copycnt++) {
+ _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[copycnt];
+ void *data = NULL;
+
+ MALLOC(data, void *, extension->datalen, M_TEMP, M_WAITOK);
+ if ((error = copyin(extension->data, data, extension->datalen)) != 0) {
+ FREE(data, M_TEMP);
+ goto bad;
+ }
+ extension->datap = data;
+ }
+
+ *psmxp = psmx;
+ return 0;
+
+bad:
+ if (psmx != NULL) {
+ for (i = 0; i < copycnt; i++) {
+ FREE(psmx->psmx_extensions[i].datap, M_TEMP);
+ }
+ FREE(psmx, M_TEMP);
+ }
+ return error;
+}
+
+static void
+spawn_free_macpolicyinfo(_posix_spawn_mac_policy_extensions_t psmx)
+{
+ int i;
+
+ if (psmx == NULL) {
+ return;
+ }
+ for (i = 0; i < psmx->psmx_count; i++) {
+ FREE(psmx->psmx_extensions[i].datap, M_TEMP);
+ }
+ FREE(psmx, M_TEMP);
+}
+#endif /* CONFIG_MACF */
+
+#if CONFIG_COALITIONS
+static inline void
+spawn_coalitions_release_all(coalition_t coal[COALITION_NUM_TYPES])
+{
+ for (int c = 0; c < COALITION_NUM_TYPES; c++) {
+ if (coal[c]) {
+ coalition_remove_active(coal[c]);
+ coalition_release(coal[c]);
+ }
+ }
+}
+#endif
+
+#if CONFIG_PERSONAS
+static int
+spawn_validate_persona(struct _posix_spawn_persona_info *px_persona)
+{
+ int error = 0;
+ struct persona *persona = NULL;
+ int verify = px_persona->pspi_flags & POSIX_SPAWN_PERSONA_FLAGS_VERIFY;
+
+ if (!IOTaskHasEntitlement(current_task(), PERSONA_MGMT_ENTITLEMENT)) {
+ return EPERM;
+ }
+
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GROUPS) {
+ if (px_persona->pspi_ngroups > NGROUPS_MAX) {
+ return EINVAL;
+ }
+ }
+
+ persona = persona_lookup(px_persona->pspi_id);
+ if (!persona) {
+ error = ESRCH;
+ goto out;
+ }
+
+ if (verify) {
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_UID) {
+ if (px_persona->pspi_uid != persona_get_uid(persona)) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GID) {
+ if (px_persona->pspi_gid != persona_get_gid(persona)) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GROUPS) {
+ unsigned ngroups = 0;
+ gid_t groups[NGROUPS_MAX];
+
+ if (persona_get_groups(persona, &ngroups, groups,
+ px_persona->pspi_ngroups) != 0) {
+ error = EINVAL;
+ goto out;
+ }
+ if (ngroups != px_persona->pspi_ngroups) {
+ error = EINVAL;
+ goto out;
+ }
+ while (ngroups--) {
+ if (px_persona->pspi_groups[ngroups] != groups[ngroups]) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ if (px_persona->pspi_gmuid != persona_get_gmuid(persona)) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ }
+
+out:
+ if (persona) {
+ persona_put(persona);
+ }
+
+ return error;
+}
+
+static int
+spawn_persona_adopt(proc_t p, struct _posix_spawn_persona_info *px_persona)
+{
+ int ret;
+ kauth_cred_t cred;
+ struct persona *persona = NULL;
+ int override = !!(px_persona->pspi_flags & POSIX_SPAWN_PERSONA_FLAGS_OVERRIDE);
+
+ if (!override) {
+ return persona_proc_adopt_id(p, px_persona->pspi_id, NULL);
+ }
+
+ /*
+ * we want to spawn into the given persona, but we want to override
+ * the kauth with a different UID/GID combo
+ */
+ persona = persona_lookup(px_persona->pspi_id);
+ if (!persona) {
+ return ESRCH;
+ }
+
+ cred = persona_get_cred(persona);
+ if (!cred) {
+ ret = EINVAL;
+ goto out;
+ }
+
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_UID) {
+ cred = kauth_cred_setresuid(cred,
+ px_persona->pspi_uid,
+ px_persona->pspi_uid,
+ px_persona->pspi_uid,
+ KAUTH_UID_NONE);
+ }
+
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GID) {
+ cred = kauth_cred_setresgid(cred,
+ px_persona->pspi_gid,
+ px_persona->pspi_gid,
+ px_persona->pspi_gid);
+ }
+
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GROUPS) {
+ cred = kauth_cred_setgroups(cred,
+ px_persona->pspi_groups,
+ px_persona->pspi_ngroups,
+ px_persona->pspi_gmuid);
+ }
+
+ ret = persona_proc_adopt(p, persona, cred);
+
+out:
+ persona_put(persona);
+ return ret;
+}
+#endif
+
+#if __arm64__
+extern int legacy_footprint_entitlement_mode;
+static inline void
+proc_legacy_footprint_entitled(proc_t p, task_t task, const char *caller)
+{
+#pragma unused(p, caller)
+ boolean_t legacy_footprint_entitled;
+
+ switch (legacy_footprint_entitlement_mode) {
+ case LEGACY_FOOTPRINT_ENTITLEMENT_IGNORE:
+ /* the entitlement is ignored */
+ break;
+ case LEGACY_FOOTPRINT_ENTITLEMENT_IOS11_ACCT:
+ /* the entitlement grants iOS11 legacy accounting */
+ legacy_footprint_entitled = IOTaskHasEntitlement(task,
+ "com.apple.private.memory.legacy_footprint");
+ if (legacy_footprint_entitled) {
+ task_set_legacy_footprint(task);
+ }
+ break;
+ case LEGACY_FOOTPRINT_ENTITLEMENT_LIMIT_INCREASE:
+ /* the entitlement grants a footprint limit increase */
+ legacy_footprint_entitled = IOTaskHasEntitlement(task,
+ "com.apple.private.memory.legacy_footprint");
+ if (legacy_footprint_entitled) {
+ task_set_extra_footprint_limit(task);
+ }
+ break;
+ default:
+ break;
+ }
+}
+
+static inline void
+proc_ios13extended_footprint_entitled(proc_t p, task_t task, const char *caller)
+{
+#pragma unused(p, caller)
+ boolean_t ios13extended_footprint_entitled;
+
+ /* the entitlement grants a footprint limit increase */
+ ios13extended_footprint_entitled = IOTaskHasEntitlement(task,
+ "com.apple.developer.memory.ios13extended_footprint");
+ if (ios13extended_footprint_entitled) {
+ task_set_ios13extended_footprint_limit(task);
+ }
+}
+#endif /* __arm64__ */
+
+/*
+ * Apply a modification on the proc's kauth cred until it converges.
+ *
+ * `update` consumes its argument to return a new kauth cred.
+ */
+static void
+apply_kauth_cred_update(proc_t p,
+ kauth_cred_t (^update)(kauth_cred_t orig_cred))
+{
+ kauth_cred_t my_cred, my_new_cred;
+
+ my_cred = kauth_cred_proc_ref(p);
+ for (;;) {
+ my_new_cred = update(my_cred);
+ if (my_cred == my_new_cred) {
+ kauth_cred_unref(&my_new_cred);
+ break;
+ }
+
+ /* try update cred on proc */
+ proc_ucred_lock(p);
+
+ if (p->p_ucred == my_cred) {
+ /* base pointer didn't change, donate our ref */
+ p->p_ucred = my_new_cred;
+ PROC_UPDATE_CREDS_ONPROC(p);
+ proc_ucred_unlock(p);
+
+ /* drop p->p_ucred reference */
+ kauth_cred_unref(&my_cred);
+ break;
+ }
+
+ /* base pointer changed, retry */
+ my_cred = p->p_ucred;
+ kauth_cred_ref(my_cred);
+ proc_ucred_unlock(p);
+
+ kauth_cred_unref(&my_new_cred);
+ }
+}
+
+static int
+spawn_posix_cred_adopt(proc_t p,
+ struct _posix_spawn_posix_cred_info *px_pcred_info)
+{
+ int error = 0;
+
+ if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_GID) {
+ struct setgid_args args = {
+ .gid = px_pcred_info->pspci_gid,
+ };
+ error = setgid(p, &args, NULL);
+ if (error) {
+ return error;
+ }
+ }
+
+ if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_GROUPS) {
+ error = setgroups_internal(p,
+ px_pcred_info->pspci_ngroups,
+ px_pcred_info->pspci_groups,
+ px_pcred_info->pspci_gmuid);
+ if (error) {
+ return error;
+ }
+ }
+
+ if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_UID) {
+ struct setuid_args args = {
+ .uid = px_pcred_info->pspci_uid,
+ };
+ error = setuid(p, &args, NULL);
+ if (error) {
+ return error;
+ }
+ }
+ return 0;
+}
+
+/*
+ * posix_spawn
+ *
+ * Parameters: uap->pid Pointer to pid return area
+ * uap->fname File name to exec
+ * uap->argp Argument list
+ * uap->envp Environment list
+ *
+ * Returns: 0 Success
+ * EINVAL Invalid argument
+ * ENOTSUP Not supported
+ * ENOEXEC Executable file format error
+ * exec_activate_image:EINVAL Invalid argument
+ * exec_activate_image:EACCES Permission denied
+ * exec_activate_image:EINTR Interrupted function
+ * exec_activate_image:ENOMEM Not enough space
+ * exec_activate_image:EFAULT Bad address
+ * exec_activate_image:ENAMETOOLONG Filename too long
+ * exec_activate_image:ENOEXEC Executable file format error
+ * exec_activate_image:ETXTBSY Text file busy [misuse of error code]
+ * exec_activate_image:EAUTH Image decryption failed
+ * exec_activate_image:EBADEXEC The executable is corrupt/unknown
+ * exec_activate_image:???
+ * mac_execve_enter:???
+ *
+ * TODO: Expect to need __mac_posix_spawn() at some point...
+ * Handle posix_spawnattr_t
+ * Handle posix_spawn_file_actions_t
+ */
+int
+posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval)
+{
+ proc_t p = ap; /* quiet bogus GCC vfork() warning */
+ user_addr_t pid = uap->pid;
+ int ival[2]; /* dummy retval for setpgid() */
+ char *bufp = NULL;
+ struct image_params *imgp;
+ struct vnode_attr *vap;
+ struct vnode_attr *origvap;
+ struct uthread *uthread = 0; /* compiler complains if not set to 0*/
+ int error, sig;
+ int is_64 = IS_64BIT_PROCESS(p);
+ struct vfs_context context;
+ struct user__posix_spawn_args_desc px_args;
+ struct _posix_spawnattr px_sa;
+ _posix_spawn_file_actions_t px_sfap = NULL;
+ _posix_spawn_port_actions_t px_spap = NULL;
+ struct __kern_sigaction vec;
+ boolean_t spawn_no_exec = FALSE;
+ boolean_t proc_transit_set = TRUE;
+ boolean_t exec_done = FALSE;
+ struct exec_port_actions port_actions = { };
+ vm_size_t px_sa_offset = offsetof(struct _posix_spawnattr, psa_ports);
+ task_t old_task = current_task();
+ task_t new_task = NULL;
+ boolean_t should_release_proc_ref = FALSE;
+ void *inherit = NULL;
+#if CONFIG_PERSONAS
+ struct _posix_spawn_persona_info *px_persona = NULL;
+#endif
+ struct _posix_spawn_posix_cred_info *px_pcred_info = NULL;
+
+ /*
+ * Allocate a big chunk for locals instead of using stack since these
+ * structures are pretty big.
+ */
+ MALLOC(bufp, char *, (sizeof(*imgp) + sizeof(*vap) + sizeof(*origvap)), M_TEMP, M_WAITOK | M_ZERO);
+ imgp = (struct image_params *) bufp;
+ if (bufp == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+ vap = (struct vnode_attr *) (bufp + sizeof(*imgp));
+ origvap = (struct vnode_attr *) (bufp + sizeof(*imgp) + sizeof(*vap));
+
+ /* Initialize the common data in the image_params structure */
+ imgp->ip_user_fname = uap->path;
+ imgp->ip_user_argv = uap->argv;
+ imgp->ip_user_envv = uap->envp;
+ imgp->ip_vattr = vap;
+ imgp->ip_origvattr = origvap;
+ imgp->ip_vfs_context = &context;
+ imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT_ADDR : IMGPF_NONE);
+ imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
+ imgp->ip_mac_return = 0;
+ imgp->ip_px_persona = NULL;
+ imgp->ip_px_pcred_info = NULL;
+ imgp->ip_cs_error = OS_REASON_NULL;
+ imgp->ip_simulator_binary = IMGPF_SB_DEFAULT;
+
+ if (uap->adesc != USER_ADDR_NULL) {
+ if (is_64) {
+ error = copyin(uap->adesc, &px_args, sizeof(px_args));
+ } else {
+ struct user32__posix_spawn_args_desc px_args32;
+
+ error = copyin(uap->adesc, &px_args32, sizeof(px_args32));
+
+ /*
+ * Convert arguments descriptor from external 32 bit
+ * representation to internal 64 bit representation
+ */
+ px_args.attr_size = px_args32.attr_size;
+ px_args.attrp = CAST_USER_ADDR_T(px_args32.attrp);
+ px_args.file_actions_size = px_args32.file_actions_size;
+ px_args.file_actions = CAST_USER_ADDR_T(px_args32.file_actions);
+ px_args.port_actions_size = px_args32.port_actions_size;
+ px_args.port_actions = CAST_USER_ADDR_T(px_args32.port_actions);
+ px_args.mac_extensions_size = px_args32.mac_extensions_size;
+ px_args.mac_extensions = CAST_USER_ADDR_T(px_args32.mac_extensions);
+ px_args.coal_info_size = px_args32.coal_info_size;
+ px_args.coal_info = CAST_USER_ADDR_T(px_args32.coal_info);
+ px_args.persona_info_size = px_args32.persona_info_size;
+ px_args.persona_info = CAST_USER_ADDR_T(px_args32.persona_info);
+ px_args.posix_cred_info_size = px_args32.posix_cred_info_size;
+ px_args.posix_cred_info = CAST_USER_ADDR_T(px_args32.posix_cred_info);
+ }
+ if (error) {
+ goto bad;
+ }
+
+ if (px_args.attr_size != 0) {
+ /*
+ * We are not copying the port_actions pointer,
+ * because we already have it from px_args.
+ * This is a bit fragile: <rdar://problem/16427422>
+ */
+
+ if ((error = copyin(px_args.attrp, &px_sa, px_sa_offset)) != 0) {
+ goto bad;
+ }
+
+ bzero((void *)((unsigned long) &px_sa + px_sa_offset), sizeof(px_sa) - px_sa_offset );
+
+ imgp->ip_px_sa = &px_sa;
+ }
+ if (px_args.file_actions_size != 0) {
+ /* Limit file_actions to allowed number of open files */
+ int maxfa = (p->p_limit ? p->p_rlimit[RLIMIT_NOFILE].rlim_cur : NOFILE);
+ size_t maxfa_size = PSF_ACTIONS_SIZE(maxfa);
+ if (px_args.file_actions_size < PSF_ACTIONS_SIZE(1) ||
+ maxfa_size == 0 || px_args.file_actions_size > maxfa_size) {
+ error = EINVAL;
+ goto bad;
+ }
+ MALLOC(px_sfap, _posix_spawn_file_actions_t, px_args.file_actions_size, M_TEMP, M_WAITOK);
+ if (px_sfap == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+ imgp->ip_px_sfa = px_sfap;
+
+ if ((error = copyin(px_args.file_actions, px_sfap,
+ px_args.file_actions_size)) != 0) {
+ goto bad;
+ }
+
+ /* Verify that the action count matches the struct size */
+ size_t psfsize = PSF_ACTIONS_SIZE(px_sfap->psfa_act_count);
+ if (psfsize == 0 || psfsize != px_args.file_actions_size) {
+ error = EINVAL;
+ goto bad;
+ }
+ }
+ if (px_args.port_actions_size != 0) {
+ /* Limit port_actions to one page of data */
+ if (px_args.port_actions_size < PS_PORT_ACTIONS_SIZE(1) ||
+ px_args.port_actions_size > PAGE_SIZE) {
+ error = EINVAL;
+ goto bad;
+ }
+
+ MALLOC(px_spap, _posix_spawn_port_actions_t,
+ px_args.port_actions_size, M_TEMP, M_WAITOK);
+ if (px_spap == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+ imgp->ip_px_spa = px_spap;
+
+ if ((error = copyin(px_args.port_actions, px_spap,
+ px_args.port_actions_size)) != 0) {
+ goto bad;
+ }
+
+ /* Verify that the action count matches the struct size */
+ size_t pasize = PS_PORT_ACTIONS_SIZE(px_spap->pspa_count);
+ if (pasize == 0 || pasize != px_args.port_actions_size) {
+ error = EINVAL;
+ goto bad;
+ }
+ }
+#if CONFIG_PERSONAS
+ /* copy in the persona info */
+ if (px_args.persona_info_size != 0 && px_args.persona_info != 0) {
+ /* for now, we need the exact same struct in user space */
+ if (px_args.persona_info_size != sizeof(*px_persona)) {
+ error = ERANGE;
+ goto bad;
+ }
+
+ MALLOC(px_persona, struct _posix_spawn_persona_info *, px_args.persona_info_size, M_TEMP, M_WAITOK | M_ZERO);
+ if (px_persona == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+ imgp->ip_px_persona = px_persona;
+
+ if ((error = copyin(px_args.persona_info, px_persona,
+ px_args.persona_info_size)) != 0) {
+ goto bad;
+ }
+ if ((error = spawn_validate_persona(px_persona)) != 0) {
+ goto bad;
+ }
+ }
+#endif
+ /* copy in the posix cred info */
+ if (px_args.posix_cred_info_size != 0 && px_args.posix_cred_info != 0) {
+ /* for now, we need the exact same struct in user space */
+ if (px_args.posix_cred_info_size != sizeof(*px_pcred_info)) {
+ error = ERANGE;
+ goto bad;
+ }
+
+ if (!kauth_cred_issuser(kauth_cred_get())) {
+ error = EPERM;
+ goto bad;
+ }
+
+ MALLOC(px_pcred_info, struct _posix_spawn_posix_cred_info *,
+ px_args.posix_cred_info_size, M_TEMP, M_WAITOK | M_ZERO);
+ if (px_pcred_info == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+ imgp->ip_px_pcred_info = px_pcred_info;
+
+ if ((error = copyin(px_args.posix_cred_info, px_pcred_info,
+ px_args.posix_cred_info_size)) != 0) {
+ goto bad;
+ }
+
+ if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_GROUPS) {
+ if (px_pcred_info->pspci_ngroups > NGROUPS_MAX) {
+ error = EINVAL;
+ goto bad;
+ }
+ }
+ }
+#if CONFIG_MACF
+ if (px_args.mac_extensions_size != 0) {
+ if ((error = spawn_copyin_macpolicyinfo(&px_args, (_posix_spawn_mac_policy_extensions_t *)&imgp->ip_px_smpx)) != 0) {
+ goto bad;
+ }
+ }
+#endif /* CONFIG_MACF */
+ }
+
+ /* set uthread to parent */
+ uthread = get_bsdthread_info(current_thread());
+
+ /*
+ * <rdar://6640530>; this does not result in a behaviour change
+ * relative to Leopard, so there should not be any existing code
+ * which depends on it.
+ */
+ if (uthread->uu_flag & UT_VFORK) {
+ error = EINVAL;
+ goto bad;
+ }
+
+ if (imgp->ip_px_sa != NULL) {
+ struct _posix_spawnattr *psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
+ if ((error = exec_validate_spawnattr_policy(psa->psa_apptype)) != 0) {
+ goto bad;
+ }
+ }
+
+ /*
+ * If we don't have the extension flag that turns "posix_spawn()"
+ * into "execve() with options", then we will be creating a new
+ * process which does not inherit memory from the parent process,
+ * which is one of the most expensive things about using fork()
+ * and execve().
+ */
+ if (imgp->ip_px_sa == NULL || !(px_sa.psa_flags & POSIX_SPAWN_SETEXEC)) {
+ /* Set the new task's coalition, if it is requested. */
+ coalition_t coal[COALITION_NUM_TYPES] = { COALITION_NULL };
+#if CONFIG_COALITIONS
+ int i, ncoals;
+ kern_return_t kr = KERN_SUCCESS;
+ struct _posix_spawn_coalition_info coal_info;
+ int coal_role[COALITION_NUM_TYPES];
+
+ if (imgp->ip_px_sa == NULL || !px_args.coal_info) {
+ goto do_fork1;
+ }
+
+ memset(&coal_info, 0, sizeof(coal_info));
+
+ if (px_args.coal_info_size > sizeof(coal_info)) {
+ px_args.coal_info_size = sizeof(coal_info);
+ }
+ error = copyin(px_args.coal_info,
+ &coal_info, px_args.coal_info_size);
+ if (error != 0) {
+ goto bad;
+ }
+
+ ncoals = 0;
+ for (i = 0; i < COALITION_NUM_TYPES; i++) {
+ uint64_t cid = coal_info.psci_info[i].psci_id;
+ if (cid != 0) {
+ /*
+ * don't allow tasks which are not in a
+ * privileged coalition to spawn processes
+ * into coalitions other than their own
+ */
+ if (!task_is_in_privileged_coalition(p->task, i)) {
+ coal_dbg("ERROR: %d not in privilegd "
+ "coalition of type %d",
+ p->p_pid, i);
+ spawn_coalitions_release_all(coal);
+ error = EPERM;
+ goto bad;
+ }
+
+ coal_dbg("searching for coalition id:%llu", cid);
+ /*
+ * take a reference and activation on the
+ * coalition to guard against free-while-spawn
+ * races
+ */
+ coal[i] = coalition_find_and_activate_by_id(cid);
+ if (coal[i] == COALITION_NULL) {
+ coal_dbg("could not find coalition id:%llu "
+ "(perhaps it has been terminated or reaped)", cid);
+ /*
+ * release any other coalition's we
+ * may have a reference to
+ */
+ spawn_coalitions_release_all(coal);
+ error = ESRCH;
+ goto bad;
+ }
+ if (coalition_type(coal[i]) != i) {
+ coal_dbg("coalition with id:%lld is not of type:%d"
+ " (it's type:%d)", cid, i, coalition_type(coal[i]));
+ error = ESRCH;
+ goto bad;
+ }
+ coal_role[i] = coal_info.psci_info[i].psci_role;
+ ncoals++;
+ }
+ }
+ if (ncoals < COALITION_NUM_TYPES) {
+ /*
+ * If the user is attempting to spawn into a subset of
+ * the known coalition types, then make sure they have
+ * _at_least_ specified a resource coalition. If not,
+ * the following fork1() call will implicitly force an
+ * inheritance from 'p' and won't actually spawn the
+ * new task into the coalitions the user specified.
+ * (also the call to coalitions_set_roles will panic)
+ */
+ if (coal[COALITION_TYPE_RESOURCE] == COALITION_NULL) {
+ spawn_coalitions_release_all(coal);
+ error = EINVAL;
+ goto bad;
+ }
+ }
+do_fork1:
+#endif /* CONFIG_COALITIONS */
+
+ /*
+ * note that this will implicitly inherit the
+ * caller's persona (if it exists)
+ */
+ error = fork1(p, &imgp->ip_new_thread, PROC_CREATE_SPAWN, coal);
+ /* returns a thread and task reference */
+
+ if (error == 0) {
+ new_task = get_threadtask(imgp->ip_new_thread);
+ }
+#if CONFIG_COALITIONS
+ /* set the roles of this task within each given coalition */
+ if (error == 0) {
+ kr = coalitions_set_roles(coal, new_task, coal_role);
+ if (kr != KERN_SUCCESS) {
+ error = EINVAL;
+ }
+ if (kdebug_debugid_enabled(MACHDBG_CODE(DBG_MACH_COALITION,
+ MACH_COALITION_ADOPT))) {
+ for (i = 0; i < COALITION_NUM_TYPES; i++) {
+ if (coal[i] != COALITION_NULL) {
+ /*
+ * On 32-bit targets, uniqueid
+ * will get truncated to 32 bits
+ */
+ KDBG_RELEASE(MACHDBG_CODE(
+ DBG_MACH_COALITION,
+ MACH_COALITION_ADOPT),
+ coalition_id(coal[i]),
+ get_task_uniqueid(new_task));
+ }
+ }
+ }
+ }
+
+ /* drop our references and activations - fork1() now holds them */
+ spawn_coalitions_release_all(coal);
+#endif /* CONFIG_COALITIONS */
+ if (error != 0) {
+ goto bad;
+ }
+ imgp->ip_flags |= IMGPF_SPAWN; /* spawn w/o exec */
+ spawn_no_exec = TRUE; /* used in later tests */
+ } else {
+ /*
+ * For execve case, create a new task and thread
+ * which points to current_proc. The current_proc will point
+ * to the new task after image activation and proc ref drain.
+ *
+ * proc (current_proc) <----- old_task (current_task)
+ * ^ | ^
+ * | | |
+ * | ----------------------------------
+ * |
+ * --------- new_task (task marked as TF_EXEC_COPY)
+ *
+ * After image activation, the proc will point to the new task
+ * and would look like following.
+ *
+ * proc (current_proc) <----- old_task (current_task, marked as TPF_DID_EXEC)
+ * ^ |
+ * | |
+ * | ----------> new_task
+ * | |
+ * -----------------
+ *
+ * During exec any transition from new_task -> proc is fine, but don't allow
+ * transition from proc->task, since it will modify old_task.
+ */
+ imgp->ip_new_thread = fork_create_child(old_task,
+ NULL,
+ p,
+ FALSE,
+ p->p_flag & P_LP64,
+ task_get_64bit_data(old_task),
+ TRUE);
+ /* task and thread ref returned by fork_create_child */
+ if (imgp->ip_new_thread == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+
+ new_task = get_threadtask(imgp->ip_new_thread);
+ imgp->ip_flags |= IMGPF_EXEC;
+ }
+
+ if (spawn_no_exec) {
+ p = (proc_t)get_bsdthreadtask_info(imgp->ip_new_thread);
+
+ /*
+ * We had to wait until this point before firing the
+ * proc:::create probe, otherwise p would not point to the
+ * child process.
+ */
+ DTRACE_PROC1(create, proc_t, p);
+ }
+ assert(p != NULL);
+
+ context.vc_thread = imgp->ip_new_thread;
+ context.vc_ucred = p->p_ucred; /* XXX must NOT be kauth_cred_get() */
+
+ /*
+ * Post fdcopy(), pre exec_handle_sugid() - this is where we want
+ * to handle the file_actions. Since vfork() also ends up setting
+ * us into the parent process group, and saved off the signal flags,
+ * this is also where we want to handle the spawn flags.
+ */
+
+ /* Has spawn file actions? */
+ if (imgp->ip_px_sfa != NULL) {
+ /*
+ * The POSIX_SPAWN_CLOEXEC_DEFAULT flag
+ * is handled in exec_handle_file_actions().
+ */
+#if CONFIG_AUDIT
+ /*
+ * The file actions auditing can overwrite the upath of
+ * AUE_POSIX_SPAWN audit record. Save the audit record.
+ */
+ struct kaudit_record *save_uu_ar = uthread->uu_ar;
+ uthread->uu_ar = NULL;
+#endif
+ error = exec_handle_file_actions(imgp,
+ imgp->ip_px_sa != NULL ? px_sa.psa_flags : 0);
+#if CONFIG_AUDIT
+ /* Restore the AUE_POSIX_SPAWN audit record. */
+ uthread->uu_ar = save_uu_ar;
+#endif
+ if (error != 0) {
+ goto bad;
+ }
+ }
+
+ /* Has spawn port actions? */
+ if (imgp->ip_px_spa != NULL) {
+#if CONFIG_AUDIT
+ /*
+ * Do the same for the port actions as we did for the file
+ * actions. Save the AUE_POSIX_SPAWN audit record.
+ */
+ struct kaudit_record *save_uu_ar = uthread->uu_ar;
+ uthread->uu_ar = NULL;
+#endif
+ error = exec_handle_port_actions(imgp, &port_actions);
+#if CONFIG_AUDIT
+ /* Restore the AUE_POSIX_SPAWN audit record. */
+ uthread->uu_ar = save_uu_ar;
+#endif
+ if (error != 0) {
+ goto bad;
+ }
+ }
+
+ /* Has spawn attr? */
+ if (imgp->ip_px_sa != NULL) {
+ /*
+ * Reset UID/GID to parent's RUID/RGID; This works only
+ * because the operation occurs *after* the vfork() and
+ * before the call to exec_handle_sugid() by the image
+ * activator called from exec_activate_image(). POSIX
+ * requires that any setuid/setgid bits on the process
+ * image will take precedence over the spawn attributes
+ * (re)setting them.
+ *
+ * Modifications to p_ucred must be guarded using the
+ * proc's ucred lock. This prevents others from accessing
+ * a garbage credential.
+ */
+ if (px_sa.psa_flags & POSIX_SPAWN_RESETIDS) {
+ apply_kauth_cred_update(p, ^kauth_cred_t (kauth_cred_t my_cred){
+ return kauth_cred_setuidgid(my_cred,
+ kauth_cred_getruid(my_cred),
+ kauth_cred_getrgid(my_cred));
+ });
+ }
+
+ if (imgp->ip_px_pcred_info) {
+ if (!spawn_no_exec) {
+ error = ENOTSUP;
+ goto bad;
+ }
+
+ error = spawn_posix_cred_adopt(p, imgp->ip_px_pcred_info);
+ if (error != 0) {
+ goto bad;
+ }
+ }
+
+#if CONFIG_PERSONAS
+ if (imgp->ip_px_persona != NULL) {
+ if (!spawn_no_exec) {
+ error = ENOTSUP;
+ goto bad;
+ }
+
+ /*
+ * If we were asked to spawn a process into a new persona,
+ * do the credential switch now (which may override the UID/GID
+ * inherit done just above). It's important to do this switch
+ * before image activation both for reasons stated above, and
+ * to ensure that the new persona has access to the image/file
+ * being executed.
+ */
+ error = spawn_persona_adopt(p, imgp->ip_px_persona);
+ if (error != 0) {
+ goto bad;
+ }
+ }
+#endif /* CONFIG_PERSONAS */
+#if !SECURE_KERNEL
+ /*
+ * Disable ASLR for the spawned process.
+ *
+ * But only do so if we are not embedded + RELEASE.
+ * While embedded allows for a boot-arg (-disable_aslr)
+ * to deal with this (which itself is only honored on
+ * DEVELOPMENT or DEBUG builds of xnu), it is often
+ * useful or necessary to disable ASLR on a per-process
+ * basis for unit testing and debugging.
+ */
+ if (px_sa.psa_flags & _POSIX_SPAWN_DISABLE_ASLR) {
+ OSBitOrAtomic(P_DISABLE_ASLR, &p->p_flag);
+ }
+#endif /* !SECURE_KERNEL */
+
+ /* Randomize high bits of ASLR slide */
+ if (px_sa.psa_flags & _POSIX_SPAWN_HIGH_BITS_ASLR) {
+ imgp->ip_flags |= IMGPF_HIGH_BITS_ASLR;
+ }
+
+#if !SECURE_KERNEL
+ /*
+ * Forcibly disallow execution from data pages for the spawned process
+ * even if it would otherwise be permitted by the architecture default.
+ */
+ if (px_sa.psa_flags & _POSIX_SPAWN_ALLOW_DATA_EXEC) {
+ imgp->ip_flags |= IMGPF_ALLOW_DATA_EXEC;
+ }
+#endif /* !SECURE_KERNEL */
+
+ if ((px_sa.psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK) ==
+ POSIX_SPAWN_PROC_TYPE_DRIVER) {
+ imgp->ip_flags |= IMGPF_DRIVER;
+ }
+ }
+
+ /*
+ * Disable ASLR during image activation. This occurs either if the
+ * _POSIX_SPAWN_DISABLE_ASLR attribute was found above or if
+ * P_DISABLE_ASLR was inherited from the parent process.
+ */
+ if (p->p_flag & P_DISABLE_ASLR) {
+ imgp->ip_flags |= IMGPF_DISABLE_ASLR;
+ }
+
+ /*
+ * Clear transition flag so we won't hang if exec_activate_image() causes
+ * an automount (and launchd does a proc sysctl to service it).
+ *
+ * <rdar://problem/6848672>, <rdar://problem/5959568>.
+ */
+ if (spawn_no_exec) {
+ proc_transend(p, 0);
+ proc_transit_set = 0;
+ }
+
+#if MAC_SPAWN /* XXX */
+ if (uap->mac_p != USER_ADDR_NULL) {
+ error = mac_execve_enter(uap->mac_p, imgp);
+ if (error) {
+ goto bad;
+ }
+ }
+#endif
+
+ /*
+ * Activate the image
+ */
+ error = exec_activate_image(imgp);
+#if defined(HAS_APPLE_PAC)
+ ml_task_set_disable_user_jop(new_task, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
+ ml_thread_set_disable_user_jop(imgp->ip_new_thread, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
+#endif
+
+ if (error == 0 && !spawn_no_exec) {
+ p = proc_exec_switch_task(p, old_task, new_task, imgp->ip_new_thread);
+ /* proc ref returned */
+ should_release_proc_ref = TRUE;
+
+ /*
+ * Need to transfer pending watch port boosts to the new task while still making
+ * sure that the old task remains in the importance linkage. Create an importance
+ * linkage from old task to new task, then switch the task importance base
+ * of old task and new task. After the switch the port watch boost will be
+ * boosting the new task and new task will be donating importance to old task.
+ */
+ inherit = ipc_importance_exec_switch_task(old_task, new_task);
+ }
+
+ if (error == 0) {
+ /* process completed the exec */
+ exec_done = TRUE;
+ } else if (error == -1) {
+ /* Image not claimed by any activator? */
+ error = ENOEXEC;
+ }
+
+ if (!error && imgp->ip_px_sa != NULL) {
+ thread_t child_thread = imgp->ip_new_thread;
+ uthread_t child_uthread = get_bsdthread_info(child_thread);
+
+ /*
+ * Because of POSIX_SPAWN_SETEXEC, we need to handle this after image
+ * activation, else when image activation fails (before the point of no
+ * return) would leave the parent process in a modified state.
+ */
+ if (px_sa.psa_flags & POSIX_SPAWN_SETPGROUP) {
+ struct setpgid_args spga;
+ spga.pid = p->p_pid;
+ spga.pgid = px_sa.psa_pgroup;
+ /*
+ * Effectively, call setpgid() system call; works
+ * because there are no pointer arguments.
+ */
+ if ((error = setpgid(p, &spga, ival)) != 0) {
+ goto bad;
+ }
+ }
+
+ if (px_sa.psa_flags & POSIX_SPAWN_SETSID) {
+ error = setsid_internal(p);
+ if (error != 0) {
+ goto bad;
+ }
+ }
+
+ /*
+ * If we have a spawn attr, and it contains signal related flags,
+ * the we need to process them in the "context" of the new child
+ * process, so we have to process it following image activation,
+ * prior to making the thread runnable in user space. This is
+ * necessitated by some signal information being per-thread rather
+ * than per-process, and we don't have the new allocation in hand
+ * until after the image is activated.
+ */
+
+ /*
+ * Mask a list of signals, instead of them being unmasked, if
+ * they were unmasked in the parent; note that some signals
+ * are not maskable.
+ */
+ if (px_sa.psa_flags & POSIX_SPAWN_SETSIGMASK) {
+ child_uthread->uu_sigmask = (px_sa.psa_sigmask & ~sigcantmask);
+ }
+ /*
+ * Default a list of signals instead of ignoring them, if
+ * they were ignored in the parent. Note that we pass
+ * spawn_no_exec to setsigvec() to indicate that we called
+ * fork1() and therefore do not need to call proc_signalstart()
+ * internally.
+ */
+ if (px_sa.psa_flags & POSIX_SPAWN_SETSIGDEF) {
+ vec.sa_handler = SIG_DFL;
+ vec.sa_tramp = 0;
+ vec.sa_mask = 0;
+ vec.sa_flags = 0;
+ for (sig = 1; sig < NSIG; sig++) {
+ if (px_sa.psa_sigdefault & (1 << (sig - 1))) {
+ error = setsigvec(p, child_thread, sig, &vec, spawn_no_exec);
+ }
+ }
+ }
+
+ /*
+ * Activate the CPU usage monitor, if requested. This is done via a task-wide, per-thread CPU
+ * usage limit, which will generate a resource exceeded exception if any one thread exceeds the
+ * limit.
+ *
+ * Userland gives us interval in seconds, and the kernel SPI expects nanoseconds.
+ */
+ if (px_sa.psa_cpumonitor_percent != 0) {
+ /*
+ * Always treat a CPU monitor activation coming from spawn as entitled. Requiring
+ * an entitlement to configure the monitor a certain way seems silly, since
+ * whomever is turning it on could just as easily choose not to do so.
+ */
+ error = proc_set_task_ruse_cpu(p->task,
+ TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC,
+ px_sa.psa_cpumonitor_percent,
+ px_sa.psa_cpumonitor_interval * NSEC_PER_SEC,
+ 0, TRUE);
+ }
+
+
+ if (px_pcred_info &&
+ (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_LOGIN)) {
+ /*
+ * setlogin() must happen after setsid()
+ */
+ setlogin_internal(p, px_pcred_info->pspci_login);
+ }
+ }
+
+bad:
+
+ if (error == 0) {
+ /* reset delay idle sleep status if set */
+#if !CONFIG_EMBEDDED
+ if ((p->p_flag & P_DELAYIDLESLEEP) == P_DELAYIDLESLEEP) {
+ OSBitAndAtomic(~((uint32_t)P_DELAYIDLESLEEP), &p->p_flag);
+ }
+#endif /* !CONFIG_EMBEDDED */
+ /* upon successful spawn, re/set the proc control state */
+ if (imgp->ip_px_sa != NULL) {
+ switch (px_sa.psa_pcontrol) {
+ case POSIX_SPAWN_PCONTROL_THROTTLE:
+ p->p_pcaction = P_PCTHROTTLE;
+ break;
+ case POSIX_SPAWN_PCONTROL_SUSPEND:
+ p->p_pcaction = P_PCSUSP;
+ break;
+ case POSIX_SPAWN_PCONTROL_KILL:
+ p->p_pcaction = P_PCKILL;
+ break;
+ case POSIX_SPAWN_PCONTROL_NONE:
+ default:
+ p->p_pcaction = 0;
+ break;
+ }
+ ;
+ }
+ exec_resettextvp(p, imgp);
+
+#if CONFIG_MEMORYSTATUS
+ /* Set jetsam priority for DriverKit processes */
+ if (px_sa.psa_apptype == POSIX_SPAWN_PROC_TYPE_DRIVER) {
+ px_sa.psa_priority = JETSAM_PRIORITY_DRIVER_APPLE;
+ }
+
+ /* Has jetsam attributes? */
+ if (imgp->ip_px_sa != NULL && (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_SET)) {
+ /*
+ * With 2-level high-water-mark support, POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND is no
+ * longer relevant, as background limits are described via the inactive limit slots.
+ *
+ * That said, however, if the POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND is passed in,
+ * we attempt to mimic previous behavior by forcing the BG limit data into the
+ * inactive/non-fatal mode and force the active slots to hold system_wide/fatal mode.
+ */
+
+ if (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND) {
+ memorystatus_update(p, px_sa.psa_priority, 0, FALSE, /* assertion priority */
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY),
+ TRUE,
+ -1, TRUE,
+ px_sa.psa_memlimit_inactive, FALSE);
+ } else {
+ memorystatus_update(p, px_sa.psa_priority, 0, FALSE, /* assertion priority */
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY),
+ TRUE,
+ px_sa.psa_memlimit_active,
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_MEMLIMIT_ACTIVE_FATAL),
+ px_sa.psa_memlimit_inactive,
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_MEMLIMIT_INACTIVE_FATAL));
+ }
+ }
+
+ /* Has jetsam relaunch behavior? */
+ if (imgp->ip_px_sa != NULL && (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_MASK)) {
+ /*
+ * Launchd has passed in data indicating the behavior of this process in response to jetsam.
+ * This data would be used by the jetsam subsystem to determine the position and protection
+ * offered to this process on dirty -> clean transitions.
+ */
+ int relaunch_flags = P_MEMSTAT_RELAUNCH_UNKNOWN;
+ switch (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_MASK) {
+ case POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_LOW:
+ relaunch_flags = P_MEMSTAT_RELAUNCH_LOW;
+ break;
+ case POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_MED:
+ relaunch_flags = P_MEMSTAT_RELAUNCH_MED;
+ break;
+ case POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_HIGH:
+ relaunch_flags = P_MEMSTAT_RELAUNCH_HIGH;
+ break;
+ default:
+ break;
+ }
+ memorystatus_relaunch_flags_update(p, relaunch_flags);
+ }
+
+#endif /* CONFIG_MEMORYSTATUS */
+ if (imgp->ip_px_sa != NULL && px_sa.psa_thread_limit > 0) {
+ task_set_thread_limit(new_task, (uint16_t)px_sa.psa_thread_limit);
+ }
+ }
+
+ /*
+ * If we successfully called fork1(), we always need to do this;
+ * we identify this case by noting the IMGPF_SPAWN flag. This is
+ * because we come back from that call with signals blocked in the
+ * child, and we have to unblock them, but we want to wait until
+ * after we've performed any spawn actions. This has to happen
+ * before check_for_signature(), which uses psignal.
+ */
+ if (spawn_no_exec) {
+ if (proc_transit_set) {
+ proc_transend(p, 0);
+ }
+
+ /*
+ * Drop the signal lock on the child which was taken on our
+ * behalf by forkproc()/cloneproc() to prevent signals being
+ * received by the child in a partially constructed state.
+ */
+ proc_signalend(p, 0);
+
+ /* flag the 'fork' has occurred */
+ proc_knote(p->p_pptr, NOTE_FORK | p->p_pid);
+ }
+
+ /* flag exec has occurred, notify only if it has not failed due to FP Key error */
+ if (!error && ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0)) {
+ proc_knote(p, NOTE_EXEC);
+ }
+
+
+ if (error == 0) {
+ /*
+ * We need to initialize the bank context behind the protection of
+ * the proc_trans lock to prevent a race with exit. We can't do this during
+ * exec_activate_image because task_bank_init checks entitlements that
+ * aren't loaded until subsequent calls (including exec_resettextvp).
+ */
+ error = proc_transstart(p, 0, 0);
+
+ if (error == 0) {
+ task_bank_init(new_task);
+ proc_transend(p, 0);
+ }
+
+#if __arm64__
+ proc_legacy_footprint_entitled(p, new_task, __FUNCTION__);
+ proc_ios13extended_footprint_entitled(p, new_task, __FUNCTION__);
+#endif /* __arm64__ */
+ }
+
+ /* Inherit task role from old task to new task for exec */
+ if (error == 0 && !spawn_no_exec) {
+ proc_inherit_task_role(new_task, old_task);
+ }
+
+#if CONFIG_ARCADE
+ if (error == 0) {
+ /*
+ * Check to see if we need to trigger an arcade upcall AST now
+ * that the vnode has been reset on the task.
+ */
+ arcade_prepare(new_task, imgp->ip_new_thread);
+ }
+#endif /* CONFIG_ARCADE */
+
+ /* Clear the initial wait on the thread before handling spawn policy */
+ if (imgp && imgp->ip_new_thread) {
+ task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_INITIAL_WAIT);
+ }
+
+ /*
+ * Apply the spawnattr policy, apptype (which primes the task for importance donation),
+ * and bind any portwatch ports to the new task.
+ * This must be done after the exec so that the child's thread is ready,
+ * and after the in transit state has been released, because priority is
+ * dropped here so we need to be prepared for a potentially long preemption interval
+ *
+ * TODO: Consider splitting this up into separate phases
+ */
+ if (error == 0 && imgp->ip_px_sa != NULL) {
+ struct _posix_spawnattr *psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
+
+ error = exec_handle_spawnattr_policy(p, imgp->ip_new_thread, psa->psa_apptype, psa->psa_qos_clamp,
+ psa->psa_darwin_role, &port_actions);
+ }
+
+ /* Transfer the turnstile watchport boost to new task if in exec */
+ if (error == 0 && !spawn_no_exec) {
+ task_transfer_turnstile_watchports(old_task, new_task, imgp->ip_new_thread);
+ }
+
+ /*
+ * Apply the requested maximum address.
+ */
+ if (error == 0 && imgp->ip_px_sa != NULL) {
+ struct _posix_spawnattr *psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
+
+ if (psa->psa_max_addr) {
+ vm_map_set_max_addr(get_task_map(new_task), psa->psa_max_addr);
+ }
+ }
+
+ if (error == 0) {
+ /* Apply the main thread qos */
+ thread_t main_thread = imgp->ip_new_thread;
+ task_set_main_thread_qos(new_task, main_thread);
+
+#if CONFIG_MACF
+ /*
+ * Processes with the MAP_JIT entitlement are permitted to have
+ * a jumbo-size map.
+ */
+ if (mac_proc_check_map_anon(p, 0, 0, 0, MAP_JIT, NULL) == 0) {
+ vm_map_set_jumbo(get_task_map(new_task));
+ vm_map_set_jit_entitled(get_task_map(new_task));
+ }
+#endif /* CONFIG_MACF */
+ }
+
+ /*
+ * Release any ports we kept around for binding to the new task
+ * We need to release the rights even if the posix_spawn has failed.
+ */
+ if (imgp->ip_px_spa != NULL) {
+ exec_port_actions_destroy(&port_actions);
+ }
+
+ /*
+ * We have to delay operations which might throw a signal until after
+ * the signals have been unblocked; however, we want that to happen
+ * after exec_resettextvp() so that the textvp is correct when they
+ * fire.
+ */
+ if (error == 0) {
+ error = check_for_signature(p, imgp);
+
+ /*
+ * Pay for our earlier safety; deliver the delayed signals from
+ * the incomplete spawn process now that it's complete.
+ */
+ if (imgp != NULL && spawn_no_exec && (p->p_lflag & P_LTRACED)) {
+ psignal_vfork(p, p->task, imgp->ip_new_thread, SIGTRAP);
+ }
+
+ if (error == 0 && !spawn_no_exec) {
+ KDBG(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXEC),
+ p->p_pid);
+ }
+ }
+
+
+ if (imgp != NULL) {
+ if (imgp->ip_vp) {
+ vnode_put(imgp->ip_vp);
+ }
+ if (imgp->ip_scriptvp) {
+ vnode_put(imgp->ip_scriptvp);
+ }
+ if (imgp->ip_strings) {
+ execargs_free(imgp);
+ }
+ if (imgp->ip_px_sfa != NULL) {
+ FREE(imgp->ip_px_sfa, M_TEMP);
+ }
+ if (imgp->ip_px_spa != NULL) {
+ FREE(imgp->ip_px_spa, M_TEMP);
+ }
+#if CONFIG_PERSONAS
+ if (imgp->ip_px_persona != NULL) {
+ FREE(imgp->ip_px_persona, M_TEMP);
+ }
+#endif
+ if (imgp->ip_px_pcred_info != NULL) {
+ FREE(imgp->ip_px_pcred_info, M_TEMP);
+ }
+#if CONFIG_MACF
+ if (imgp->ip_px_smpx != NULL) {
+ spawn_free_macpolicyinfo(imgp->ip_px_smpx);
+ }
+ if (imgp->ip_execlabelp) {
+ mac_cred_label_free(imgp->ip_execlabelp);
+ }
+ if (imgp->ip_scriptlabelp) {
+ mac_vnode_label_free(imgp->ip_scriptlabelp);
+ }
+ if (imgp->ip_cs_error != OS_REASON_NULL) {
+ os_reason_free(imgp->ip_cs_error);
+ imgp->ip_cs_error = OS_REASON_NULL;
+ }
+#endif
+ if (imgp->ip_sc_port != NULL) {
+ ipc_port_release_send(imgp->ip_sc_port);
+ imgp->ip_sc_port = NULL;
+ }
+ }
+
+#if CONFIG_DTRACE
+ if (spawn_no_exec) {
+ /*
+ * In the original DTrace reference implementation,
+ * posix_spawn() was a libc routine that just
+ * did vfork(2) then exec(2). Thus the proc::: probes
+ * are very fork/exec oriented. The details of this
+ * in-kernel implementation of posix_spawn() is different
+ * (while producing the same process-observable effects)
+ * particularly w.r.t. errors, and which thread/process
+ * is constructing what on behalf of whom.
+ */
+ if (error) {
+ DTRACE_PROC1(spawn__failure, int, error);
+ } else {
+ DTRACE_PROC(spawn__success);
+ /*
+ * Some DTrace scripts, e.g. newproc.d in
+ * /usr/bin, rely on the the 'exec-success'
+ * probe being fired in the child after the
+ * new process image has been constructed
+ * in order to determine the associated pid.
+ *
+ * So, even though the parent built the image
+ * here, for compatibility, mark the new thread
+ * so 'exec-success' fires on it as it leaves
+ * the kernel.
+ */
+ dtrace_thread_didexec(imgp->ip_new_thread);
+ }
+ } else {
+ if (error) {
+ DTRACE_PROC1(exec__failure, int, error);
+ } else {
+ dtrace_thread_didexec(imgp->ip_new_thread);
+ }
+ }
+
+ if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
+ (*dtrace_proc_waitfor_hook)(p);
+ }
+#endif
+
+#if CONFIG_AUDIT
+ if (!error && AUDIT_ENABLED() && p) {
+ /* Add the CDHash of the new process to the audit record */
+ uint8_t *cdhash = cs_get_cdhash(p);
+ if (cdhash) {
+ AUDIT_ARG(data, cdhash, sizeof(uint8_t), CS_CDHASH_LEN);
+ }
+ }
+#endif
+
+ /*
+ * clear bsd_info from old task if it did exec.
+ */
+ if (task_did_exec(old_task)) {
+ set_bsdtask_info(old_task, NULL);
+ }
+
+ /* clear bsd_info from new task and terminate it if exec failed */
+ if (new_task != NULL && task_is_exec_copy(new_task)) {
+ set_bsdtask_info(new_task, NULL);
+ task_terminate_internal(new_task);
+ }
+
+ /* Return to both the parent and the child? */
+ if (imgp != NULL && spawn_no_exec) {
+ /*
+ * If the parent wants the pid, copy it out
+ */
+ if (pid != USER_ADDR_NULL) {
+ _Static_assert(sizeof(p->p_pid) == 4, "posix_spawn() assumes a 32-bit pid_t");
+ bool aligned = (pid & 3) == 0;
+ if (aligned) {
+ (void)copyout_atomic32(p->p_pid, pid);
+ } else {
+ (void)suword(pid, p->p_pid);
+ }
+ }
+ retval[0] = error;
+
+ /*
+ * If we had an error, perform an internal reap ; this is
+ * entirely safe, as we have a real process backing us.
+ */
+ if (error) {
+ proc_list_lock();
+ p->p_listflag |= P_LIST_DEADPARENT;
+ proc_list_unlock();
+ proc_lock(p);
+ /* make sure no one else has killed it off... */
+ if (p->p_stat != SZOMB && p->exit_thread == NULL) {
+ p->exit_thread = current_thread();
+ proc_unlock(p);
+ exit1(p, 1, (int *)NULL);
+ } else {
+ /* someone is doing it for us; just skip it */
+ proc_unlock(p);
+ }
+ }
+ }
+
+ /*
+ * Do not terminate the current task, if proc_exec_switch_task did not
+ * switch the tasks, terminating the current task without the switch would
+ * result in loosing the SIGKILL status.
+ */
+ if (task_did_exec(old_task)) {
+ /* Terminate the current task, since exec will start in new task */
+ task_terminate_internal(old_task);
+ }
+
+ /* Release the thread ref returned by fork_create_child/fork1 */
+ if (imgp != NULL && imgp->ip_new_thread) {
+ /* wake up the new thread */
+ task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_FINAL_WAIT);
+ thread_deallocate(imgp->ip_new_thread);
+ imgp->ip_new_thread = NULL;
+ }
+
+ /* Release the ref returned by fork_create_child/fork1 */
+ if (new_task) {
+ task_deallocate(new_task);
+ new_task = NULL;
+ }
+
+ if (should_release_proc_ref) {
+ proc_rele(p);
+ }
+
+ if (bufp != NULL) {
+ FREE(bufp, M_TEMP);
+ }
+
+ if (inherit != NULL) {
+ ipc_importance_release(inherit);
+ }
+
+ return error;
+}
+
+/*
+ * proc_exec_switch_task
+ *
+ * Parameters: p proc
+ * old_task task before exec
+ * new_task task after exec
+ * new_thread thread in new task
+ *
+ * Returns: proc.
+ *
+ * Note: The function will switch the task pointer of proc
+ * from old task to new task. The switch needs to happen
+ * after draining all proc refs and inside a proc translock.
+ * In the case of failure to switch the task, which might happen
+ * if the process received a SIGKILL or jetsam killed it, it will make
+ * sure that the new tasks terminates. User proc ref returned
+ * to caller.
+ *
+ * This function is called after point of no return, in the case
+ * failure to switch, it will terminate the new task and swallow the
+ * error and let the terminated process complete exec and die.
+ */
+proc_t
+proc_exec_switch_task(proc_t p, task_t old_task, task_t new_task, thread_t new_thread)
+{
+ int error = 0;
+ boolean_t task_active;
+ boolean_t proc_active;
+ boolean_t thread_active;
+ thread_t old_thread = current_thread();
+
+ /*
+ * Switch the task pointer of proc to new task.
+ * Before switching the task, wait for proc_refdrain.
+ * After the switch happens, the proc can disappear,
+ * take a ref before it disappears. Waiting for
+ * proc_refdrain in exec will block all other threads
+ * trying to take a proc ref, boost the current thread
+ * to avoid priority inversion.
+ */
+ thread_set_exec_promotion(old_thread);
+ p = proc_refdrain_with_refwait(p, TRUE);
+ /* extra proc ref returned to the caller */
+
+ assert(get_threadtask(new_thread) == new_task);
+ task_active = task_is_active(new_task);
+
+ /* Take the proc_translock to change the task ptr */
+ proc_lock(p);
+ proc_active = !(p->p_lflag & P_LEXIT);
+
+ /* Check if the current thread is not aborted due to SIGKILL */
+ thread_active = thread_is_active(old_thread);
+
+ /*
+ * Do not switch the task if the new task or proc is already terminated
+ * as a result of error in exec past point of no return
+ */
+ if (proc_active && task_active && thread_active) {
+ error = proc_transstart(p, 1, 0);
+ if (error == 0) {
+ uthread_t new_uthread = get_bsdthread_info(new_thread);
+ uthread_t old_uthread = get_bsdthread_info(current_thread());
+
+ /*
+ * bsd_info of old_task will get cleared in execve and posix_spawn
+ * after firing exec-success/error dtrace probe.
+ */
+ p->task = new_task;
+
+ /* Clear dispatchqueue and workloop ast offset */
+ p->p_dispatchqueue_offset = 0;
+ p->p_dispatchqueue_serialno_offset = 0;
+ p->p_dispatchqueue_label_offset = 0;
+ p->p_return_to_kernel_offset = 0;
+
+ /* Copy the signal state, dtrace state and set bsd ast on new thread */
+ act_set_astbsd(new_thread);
+ new_uthread->uu_siglist = old_uthread->uu_siglist;
+ new_uthread->uu_sigwait = old_uthread->uu_sigwait;
+ new_uthread->uu_sigmask = old_uthread->uu_sigmask;
+ new_uthread->uu_oldmask = old_uthread->uu_oldmask;
+ new_uthread->uu_vforkmask = old_uthread->uu_vforkmask;
+ new_uthread->uu_exit_reason = old_uthread->uu_exit_reason;
+#if CONFIG_DTRACE
+ new_uthread->t_dtrace_sig = old_uthread->t_dtrace_sig;
+ new_uthread->t_dtrace_stop = old_uthread->t_dtrace_stop;
+ new_uthread->t_dtrace_resumepid = old_uthread->t_dtrace_resumepid;
+ assert(new_uthread->t_dtrace_scratch == NULL);
+ new_uthread->t_dtrace_scratch = old_uthread->t_dtrace_scratch;
+
+ old_uthread->t_dtrace_sig = 0;
+ old_uthread->t_dtrace_stop = 0;
+ old_uthread->t_dtrace_resumepid = 0;
+ old_uthread->t_dtrace_scratch = NULL;
+#endif
+ /* Copy the resource accounting info */
+ thread_copy_resource_info(new_thread, current_thread());
+
+ /* Clear the exit reason and signal state on old thread */
+ old_uthread->uu_exit_reason = NULL;
+ old_uthread->uu_siglist = 0;
+
+ /* Add the new uthread to proc uthlist and remove the old one */
+ TAILQ_INSERT_TAIL(&p->p_uthlist, new_uthread, uu_list);
+ TAILQ_REMOVE(&p->p_uthlist, old_uthread, uu_list);
+
+ task_set_did_exec_flag(old_task);
+ task_clear_exec_copy_flag(new_task);
+
+ task_copy_fields_for_exec(new_task, old_task);
+
+ proc_transend(p, 1);
+ }
+ }
+
+ proc_unlock(p);
+ proc_refwake(p);
+ thread_clear_exec_promotion(old_thread);
+
+ if (error != 0 || !task_active || !proc_active || !thread_active) {
+ task_terminate_internal(new_task);
+ }
+
+ return p;
+}
+
+/*
+ * execve
+ *
+ * Parameters: uap->fname File name to exec
+ * uap->argp Argument list
+ * uap->envp Environment list
+ *
+ * Returns: 0 Success
+ * __mac_execve:EINVAL Invalid argument
+ * __mac_execve:ENOTSUP Invalid argument
+ * __mac_execve:EACCES Permission denied
+ * __mac_execve:EINTR Interrupted function
+ * __mac_execve:ENOMEM Not enough space
+ * __mac_execve:EFAULT Bad address
+ * __mac_execve:ENAMETOOLONG Filename too long
+ * __mac_execve:ENOEXEC Executable file format error
+ * __mac_execve:ETXTBSY Text file busy [misuse of error code]
+ * __mac_execve:???
+ *
+ * TODO: Dynamic linker header address on stack is copied via suword()
+ */
+/* ARGSUSED */
+int
+execve(proc_t p, struct execve_args *uap, int32_t *retval)
+{
+ struct __mac_execve_args muap;
+ int err;
+
+ memoryshot(VM_EXECVE, DBG_FUNC_NONE);
+
+ muap.fname = uap->fname;
+ muap.argp = uap->argp;
+ muap.envp = uap->envp;
+ muap.mac_p = USER_ADDR_NULL;
+ err = __mac_execve(p, &muap, retval);
+
+ return err;
+}
+
+/*
+ * __mac_execve
+ *
+ * Parameters: uap->fname File name to exec
+ * uap->argp Argument list
+ * uap->envp Environment list
+ * uap->mac_p MAC label supplied by caller
+ *
+ * Returns: 0 Success
+ * EINVAL Invalid argument
+ * ENOTSUP Not supported
+ * ENOEXEC Executable file format error
+ * exec_activate_image:EINVAL Invalid argument
+ * exec_activate_image:EACCES Permission denied
+ * exec_activate_image:EINTR Interrupted function
+ * exec_activate_image:ENOMEM Not enough space
+ * exec_activate_image:EFAULT Bad address
+ * exec_activate_image:ENAMETOOLONG Filename too long
+ * exec_activate_image:ENOEXEC Executable file format error
+ * exec_activate_image:ETXTBSY Text file busy [misuse of error code]
+ * exec_activate_image:EBADEXEC The executable is corrupt/unknown
+ * exec_activate_image:???
+ * mac_execve_enter:???
+ *
+ * TODO: Dynamic linker header address on stack is copied via suword()
+ */
+int
+__mac_execve(proc_t p, struct __mac_execve_args *uap, int32_t *retval)
+{
+ char *bufp = NULL;
+ struct image_params *imgp;
+ struct vnode_attr *vap;
+ struct vnode_attr *origvap;
+ int error;
+ int is_64 = IS_64BIT_PROCESS(p);
+ struct vfs_context context;
+ struct uthread *uthread;
+ task_t old_task = current_task();
+ task_t new_task = NULL;
+ boolean_t should_release_proc_ref = FALSE;
+ boolean_t exec_done = FALSE;
+ boolean_t in_vfexec = FALSE;
+ void *inherit = NULL;
+
+ context.vc_thread = current_thread();
+ context.vc_ucred = kauth_cred_proc_ref(p); /* XXX must NOT be kauth_cred_get() */
+
+ /* Allocate a big chunk for locals instead of using stack since these
+ * structures a pretty big.
+ */
+ MALLOC(bufp, char *, (sizeof(*imgp) + sizeof(*vap) + sizeof(*origvap)), M_TEMP, M_WAITOK | M_ZERO);
+ imgp = (struct image_params *) bufp;
+ if (bufp == NULL) {
+ error = ENOMEM;
+ goto exit_with_error;
+ }
+ vap = (struct vnode_attr *) (bufp + sizeof(*imgp));
+ origvap = (struct vnode_attr *) (bufp + sizeof(*imgp) + sizeof(*vap));
+
+ /* Initialize the common data in the image_params structure */
+ imgp->ip_user_fname = uap->fname;
+ imgp->ip_user_argv = uap->argp;
+ imgp->ip_user_envv = uap->envp;
+ imgp->ip_vattr = vap;
+ imgp->ip_origvattr = origvap;
+ imgp->ip_vfs_context = &context;
+ imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT_ADDR : IMGPF_NONE) | ((p->p_flag & P_DISABLE_ASLR) ? IMGPF_DISABLE_ASLR : IMGPF_NONE);
+ imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
+ imgp->ip_mac_return = 0;
+ imgp->ip_cs_error = OS_REASON_NULL;
+ imgp->ip_simulator_binary = IMGPF_SB_DEFAULT;
+
+#if CONFIG_MACF
+ if (uap->mac_p != USER_ADDR_NULL) {
+ error = mac_execve_enter(uap->mac_p, imgp);
+ if (error) {
+ kauth_cred_unref(&context.vc_ucred);
+ goto exit_with_error;
+ }
+ }
+#endif
+ uthread = get_bsdthread_info(current_thread());
+ if (uthread->uu_flag & UT_VFORK) {
+ imgp->ip_flags |= IMGPF_VFORK_EXEC;
+ in_vfexec = TRUE;
+ } else {
+ imgp->ip_flags |= IMGPF_EXEC;
+
+ /*
+ * For execve case, create a new task and thread
+ * which points to current_proc. The current_proc will point
+ * to the new task after image activation and proc ref drain.
+ *
+ * proc (current_proc) <----- old_task (current_task)
+ * ^ | ^
+ * | | |
+ * | ----------------------------------
+ * |
+ * --------- new_task (task marked as TF_EXEC_COPY)
+ *
+ * After image activation, the proc will point to the new task
+ * and would look like following.
+ *
+ * proc (current_proc) <----- old_task (current_task, marked as TPF_DID_EXEC)
+ * ^ |
+ * | |
+ * | ----------> new_task
+ * | |
+ * -----------------
+ *
+ * During exec any transition from new_task -> proc is fine, but don't allow
+ * transition from proc->task, since it will modify old_task.
+ */
+ imgp->ip_new_thread = fork_create_child(old_task,
+ NULL,
+ p,
+ FALSE,
+ p->p_flag & P_LP64,
+ task_get_64bit_data(old_task),
+ TRUE);
+ /* task and thread ref returned by fork_create_child */
+ if (imgp->ip_new_thread == NULL) {
+ error = ENOMEM;
+ goto exit_with_error;
+ }
+
+ new_task = get_threadtask(imgp->ip_new_thread);
+ context.vc_thread = imgp->ip_new_thread;
+ }
+
+ error = exec_activate_image(imgp);
+ /* thread and task ref returned for vfexec case */
+
+ if (imgp->ip_new_thread != NULL) {
+ /*
+ * task reference might be returned by exec_activate_image
+ * for vfexec.
+ */
+ new_task = get_threadtask(imgp->ip_new_thread);
+#if defined(HAS_APPLE_PAC)
+ ml_task_set_disable_user_jop(new_task, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
+ ml_thread_set_disable_user_jop(imgp->ip_new_thread, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
+#endif
+ }
+
+ if (!error && !in_vfexec) {
+ p = proc_exec_switch_task(p, old_task, new_task, imgp->ip_new_thread);
+ /* proc ref returned */
+ should_release_proc_ref = TRUE;
+
+ /*
+ * Need to transfer pending watch port boosts to the new task while still making
+ * sure that the old task remains in the importance linkage. Create an importance
+ * linkage from old task to new task, then switch the task importance base
+ * of old task and new task. After the switch the port watch boost will be
+ * boosting the new task and new task will be donating importance to old task.
+ */
+ inherit = ipc_importance_exec_switch_task(old_task, new_task);
+ }
+
+ kauth_cred_unref(&context.vc_ucred);
+
+ /* Image not claimed by any activator? */
+ if (error == -1) {
+ error = ENOEXEC;
+ }
+
+ if (!error) {
+ exec_done = TRUE;
+ assert(imgp->ip_new_thread != NULL);
+
+ exec_resettextvp(p, imgp);
+ error = check_for_signature(p, imgp);
+ }
+
+ /* flag exec has occurred, notify only if it has not failed due to FP Key error */
+ if (exec_done && ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0)) {
+ proc_knote(p, NOTE_EXEC);
+ }
+
+ if (imgp->ip_vp != NULLVP) {
+ vnode_put(imgp->ip_vp);
+ }
+ if (imgp->ip_scriptvp != NULLVP) {
+ vnode_put(imgp->ip_scriptvp);
+ }
+ if (imgp->ip_strings) {
+ execargs_free(imgp);
+ }
+#if CONFIG_MACF
+ if (imgp->ip_execlabelp) {
+ mac_cred_label_free(imgp->ip_execlabelp);
+ }
+ if (imgp->ip_scriptlabelp) {
+ mac_vnode_label_free(imgp->ip_scriptlabelp);
+ }
+#endif
+ if (imgp->ip_cs_error != OS_REASON_NULL) {
+ os_reason_free(imgp->ip_cs_error);
+ imgp->ip_cs_error = OS_REASON_NULL;
+ }
+
+ if (!error) {
+ /*
+ * We need to initialize the bank context behind the protection of
+ * the proc_trans lock to prevent a race with exit. We can't do this during
+ * exec_activate_image because task_bank_init checks entitlements that
+ * aren't loaded until subsequent calls (including exec_resettextvp).
+ */
+ error = proc_transstart(p, 0, 0);
+ }
+
+ if (!error) {
+ task_bank_init(new_task);
+ proc_transend(p, 0);
+
+#if __arm64__
+ proc_legacy_footprint_entitled(p, new_task, __FUNCTION__);
+ proc_ios13extended_footprint_entitled(p, new_task, __FUNCTION__);
+#endif /* __arm64__ */
+
+ /* Sever any extant thread affinity */
+ thread_affinity_exec(current_thread());
+
+ /* Inherit task role from old task to new task for exec */
+ if (!in_vfexec) {
+ proc_inherit_task_role(new_task, old_task);
+ }
+
+ thread_t main_thread = imgp->ip_new_thread;
+
+ task_set_main_thread_qos(new_task, main_thread);
+
+#if CONFIG_ARCADE
+ /*
+ * Check to see if we need to trigger an arcade upcall AST now
+ * that the vnode has been reset on the task.
+ */
+ arcade_prepare(new_task, imgp->ip_new_thread);
+#endif /* CONFIG_ARCADE */
+
+#if CONFIG_MACF
+ /*
+ * Processes with the MAP_JIT entitlement are permitted to have
+ * a jumbo-size map.
+ */
+ if (mac_proc_check_map_anon(p, 0, 0, 0, MAP_JIT, NULL) == 0) {
+ vm_map_set_jumbo(get_task_map(new_task));
+ vm_map_set_jit_entitled(get_task_map(new_task));
+ }
+#endif /* CONFIG_MACF */
+
+ if (vm_darkwake_mode == TRUE) {
+ /*
+ * This process is being launched when the system
+ * is in darkwake. So mark it specially. This will
+ * cause all its pages to be entered in the background Q.
+ */
+ task_set_darkwake_mode(new_task, vm_darkwake_mode);
+ }
+
+#if CONFIG_DTRACE
+ dtrace_thread_didexec(imgp->ip_new_thread);
+
+ if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
+ (*dtrace_proc_waitfor_hook)(p);
+ }
+#endif
+
+#if CONFIG_AUDIT
+ if (!error && AUDIT_ENABLED() && p) {
+ /* Add the CDHash of the new process to the audit record */
+ uint8_t *cdhash = cs_get_cdhash(p);
+ if (cdhash) {
+ AUDIT_ARG(data, cdhash, sizeof(uint8_t), CS_CDHASH_LEN);
+ }
+ }
+#endif
+
+ if (in_vfexec) {
+ vfork_return(p, retval, p->p_pid);
+ }
+ } else {
+ DTRACE_PROC1(exec__failure, int, error);
+ }
+
+exit_with_error:
+
+ /*
+ * clear bsd_info from old task if it did exec.
+ */
+ if (task_did_exec(old_task)) {
+ set_bsdtask_info(old_task, NULL);
+ }
+
+ /* clear bsd_info from new task and terminate it if exec failed */
+ if (new_task != NULL && task_is_exec_copy(new_task)) {
+ set_bsdtask_info(new_task, NULL);
+ task_terminate_internal(new_task);
+ }
+
+ if (imgp != NULL) {
+ /* Clear the initial wait on the thread transferring watchports */
+ if (imgp->ip_new_thread) {
+ task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_INITIAL_WAIT);
+ }
+
+ /* Transfer the watchport boost to new task */
+ if (!error && !in_vfexec) {
+ task_transfer_turnstile_watchports(old_task,
+ new_task, imgp->ip_new_thread);
+ }
+ /*
+ * Do not terminate the current task, if proc_exec_switch_task did not
+ * switch the tasks, terminating the current task without the switch would
+ * result in loosing the SIGKILL status.
+ */
+ if (task_did_exec(old_task)) {
+ /* Terminate the current task, since exec will start in new task */
+ task_terminate_internal(old_task);
+ }
+
+ /* Release the thread ref returned by fork_create_child */
+ if (imgp->ip_new_thread) {
+ /* wake up the new exec thread */
+ task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_FINAL_WAIT);
+ thread_deallocate(imgp->ip_new_thread);
+ imgp->ip_new_thread = NULL;
+ }
+ }
+
+ /* Release the ref returned by fork_create_child */
+ if (new_task) {
+ task_deallocate(new_task);
+ new_task = NULL;
+ }
+
+ if (should_release_proc_ref) {
+ proc_rele(p);
+ }
+
+ if (bufp != NULL) {
+ FREE(bufp, M_TEMP);
+ }
+
+ if (inherit != NULL) {
+ ipc_importance_release(inherit);
+ }
+
+ return error;
+}
+
+
+/*
+ * copyinptr
+ *
+ * Description: Copy a pointer in from user space to a user_addr_t in kernel
+ * space, based on 32/64 bitness of the user space
+ *
+ * Parameters: froma User space address
+ * toptr Address of kernel space user_addr_t
+ * ptr_size 4/8, based on 'froma' address space
+ *
+ * Returns: 0 Success
+ * EFAULT Bad 'froma'
+ *
+ * Implicit returns:
+ * *ptr_size Modified
+ */
+static int
+copyinptr(user_addr_t froma, user_addr_t *toptr, int ptr_size)
+{
+ int error;
+
+ if (ptr_size == 4) {
+ /* 64 bit value containing 32 bit address */
+ unsigned int i = 0;
+
+ error = copyin(froma, &i, 4);
+ *toptr = CAST_USER_ADDR_T(i); /* SAFE */
+ } else {
+ error = copyin(froma, toptr, 8);
+ }
+ return error;
+}
+
+
+/*
+ * copyoutptr
+ *
+ * Description: Copy a pointer out from a user_addr_t in kernel space to
+ * user space, based on 32/64 bitness of the user space
+ *
+ * Parameters: ua User space address to copy to
+ * ptr Address of kernel space user_addr_t
+ * ptr_size 4/8, based on 'ua' address space
+ *
+ * Returns: 0 Success
+ * EFAULT Bad 'ua'
+ *
+ */
+static int
+copyoutptr(user_addr_t ua, user_addr_t ptr, int ptr_size)
+{
+ int error;
+
+ if (ptr_size == 4) {
+ /* 64 bit value containing 32 bit address */
+ unsigned int i = CAST_DOWN_EXPLICIT(unsigned int, ua); /* SAFE */
+
+ error = copyout(&i, ptr, 4);
+ } else {
+ error = copyout(&ua, ptr, 8);
+ }
+ return error;
+}
+
+
+/*
+ * exec_copyout_strings
+ *
+ * Copy out the strings segment to user space. The strings segment is put
+ * on a preinitialized stack frame.
+ *
+ * Parameters: struct image_params * the image parameter block
+ * int * a pointer to the stack offset variable
+ *
+ * Returns: 0 Success
+ * !0 Faiure: errno
+ *
+ * Implicit returns:
+ * (*stackp) The stack offset, modified
+ *
+ * Note: The strings segment layout is backward, from the beginning
+ * of the top of the stack to consume the minimal amount of
+ * space possible; the returned stack pointer points to the
+ * end of the area consumed (stacks grow downward).