+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;
+ char alt_p_comm[sizeof(p->p_comm)] = {0}; /* for PowerPC */
+ 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;
+ int portwatch_count = 0;
+ ipc_port_t * portwatch_ports = NULL;
+ vm_size_t px_sa_offset = offsetof(struct _posix_spawnattr, psa_ports);
+
+ /*
+ * 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 : IMGPF_NONE);
+ imgp->ip_p_comm = alt_p_comm; /* for PowerPC */
+ imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
+ imgp->ip_mac_return = 0;
+
+ 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);
+ }
+ 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);
+ if (px_args.file_actions_size < PSF_ACTIONS_SIZE(1) ||
+ px_args.file_actions_size > PSF_ACTIONS_SIZE(maxfa)) {
+ 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 */
+ if (PSF_ACTIONS_SIZE(px_sfap->psfa_act_count) != 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 */
+ if (PS_PORT_ACTIONS_SIZE(px_spap->pspa_count) != px_args.port_actions_size) {
+ 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 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.
+ * TODO: privilege check - 15365900
+ */
+ coalition_t coal = COALITION_NULL;
+#if CONFIG_COALITIONS
+ if (imgp->ip_px_sa) {
+ uint64_t cid = px_sa.psa_coalitionid;
+ if (cid != 0) {
+#if COALITION_DEBUG
+ printf("%s: searching for coalition ID %llu\n", __func__, cid);
+#endif
+ coal = coalition_find_and_activate_by_id(cid);
+ if (coal == COALITION_NULL) {
+#if COALITION_DEBUG
+ printf("%s: could not find coalition ID %llu (perhaps it has been terminated or reaped)\n", __func__, cid);
+#endif
+ error = ESRCH;
+ goto bad;
+ }
+ }
+ }
+#endif /* CONFIG_COALITIONS */
+
+ error = fork1(p, &imgp->ip_new_thread, PROC_CREATE_SPAWN, coal);
+
+ if (error != 0) {
+ if (coal != COALITION_NULL) {
+#if CONFIG_COALITIONS
+ coalition_remove_active(coal);
+ coalition_release(coal);
+#endif /* CONFIG_COALITIONS */
+ }
+ goto bad;
+ }
+ imgp->ip_flags |= IMGPF_SPAWN; /* spawn w/o exec */
+ spawn_no_exec = TRUE; /* used in later tests */
+
+ if (coal != COALITION_NULL) {
+#if CONFIG_COALITIONS
+ coalition_remove_active(coal);
+ coalition_release(coal);
+#endif /* CONFIG_COALITIONS */
+ }
+ }
+
+ 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);
+
+ /* By default, the thread everyone plays with is the parent */
+ context.vc_thread = current_thread();
+ context.vc_ucred = p->p_ucred; /* XXX must NOT be kauth_cred_get() */
+
+ /*
+ * However, if we're not in the setexec case, redirect the context
+ * to the newly created process instead
+ */
+ if (spawn_no_exec)
+ context.vc_thread = imgp->ip_new_thread;
+
+ /*
+ * 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 ((error = exec_handle_file_actions(imgp,
+ imgp->ip_px_sa != NULL ? px_sa.psa_flags : 0)) != 0)
+ goto bad;
+ }
+
+ /* Has spawn port actions? */
+ if (imgp->ip_px_spa != NULL) {
+ boolean_t is_adaptive = FALSE;
+ boolean_t portwatch_present = FALSE;
+
+ /* Will this process become adaptive? The apptype isn't ready yet, so we can't look there. */
+ if (imgp->ip_px_sa != NULL && px_sa.psa_apptype == POSIX_SPAWN_PROC_TYPE_DAEMON_ADAPTIVE)
+ is_adaptive = TRUE;
+
+ /*
+ * portwatch only:
+ * Allocate a place to store the ports we want to bind to the new task
+ * We can't bind them until after the apptype is set.
+ */
+ if (px_spap->pspa_count != 0 && is_adaptive) {
+ portwatch_count = px_spap->pspa_count;
+ MALLOC(portwatch_ports, ipc_port_t *, (sizeof(ipc_port_t) * portwatch_count), M_TEMP, M_WAITOK | M_ZERO);
+ } else {
+ portwatch_ports = NULL;
+ }
+
+ if ((error = exec_handle_port_actions(imgp,
+ imgp->ip_px_sa != NULL ? px_sa.psa_flags : 0, &portwatch_present, portwatch_ports)) != 0)
+ goto bad;
+
+ if (portwatch_present == FALSE && portwatch_ports != NULL) {
+ FREE(portwatch_ports, M_TEMP);
+ portwatch_ports = NULL;
+ portwatch_count = 0;
+ }
+ }
+
+ /* Has spawn attr? */
+ if (imgp->ip_px_sa != NULL) {
+ /*
+ * Set the process group ID of the child process; this has
+ * to happen before the image activation.
+ */
+ 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;
+ }
+
+ /*
+ * 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.
+ *
+ * The use of p_ucred is safe, since we are acting on the
+ * new process, and it has no threads other than the one
+ * we are creating for it.
+ */
+ if (px_sa.psa_flags & POSIX_SPAWN_RESETIDS) {
+ kauth_cred_t my_cred = p->p_ucred;
+ kauth_cred_t my_new_cred = kauth_cred_setuidgid(my_cred, kauth_cred_getruid(my_cred), kauth_cred_getrgid(my_cred));
+ if (my_new_cred != my_cred) {
+ p->p_ucred = my_new_cred;
+ /* update cred on proc */
+ PROC_UPDATE_CREDS_ONPROC(p);
+ }
+ }
+
+ /*
+ * Disable ASLR for the spawned process.
+ */
+ /*
+ * But only do so if we are not embedded; embedded allows for a
+ * boot-arg (-disable_aslr) to deal with this (which itself is
+ * only honored on DEVELOPMENT or DEBUG builds of xnu).
+ */
+ if (px_sa.psa_flags & _POSIX_SPAWN_DISABLE_ASLR)
+ OSBitOrAtomic(P_DISABLE_ASLR, &p->p_flag);
+
+ /*
+ * 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;
+ }
+
+ /*
+ * 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 (error == 0) {
+ /* process completed the exec */
+ exec_done = TRUE;
+ } else if (error == -1) {
+ /* Image not claimed by any activator? */
+ error = ENOEXEC;
+ }
+
+ /*
+ * 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.
+ */
+ if (!error && imgp->ip_px_sa != NULL) {
+ thread_t child_thread = current_thread();
+ uthread_t child_uthread = uthread;
+
+ /*
+ * If we created a new child thread, then the thread and
+ * uthread are different than the current ones; otherwise,
+ * we leave them, since we are in the exec case instead.
+ */
+ if (spawn_no_exec) {
+ child_thread = imgp->ip_new_thread;
+ child_uthread = get_bsdthread_info(child_thread);
+ }
+
+ /*
+ * 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 = 0; sig < NSIG; sig++)
+ if (px_sa.psa_sigdefault & (1 << sig)) {
+ error = setsigvec(p, child_thread, sig + 1, &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.
+ *
+ * XXX - Ignore the parameters that we get from userland. The spawnattr method of
+ * activating the monitor always gets the system default parameters. Once we have
+ * an explicit spawn SPI for configuring the defaults, we can revert this to
+ * respect the params passed in from userland.
+ */
+ error = proc_set_task_ruse_cpu(p->task,
+ TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC,
+ PROC_POLICY_CPUMON_DEFAULTS, 0,
+ 0, TRUE);
+ }
+ }
+
+bad:
+
+ if (error == 0) {
+ /* reset delay idle sleep status if set */
+ if ((p->p_flag & P_DELAYIDLESLEEP) == P_DELAYIDLESLEEP)
+ OSBitAndAtomic(~((uint32_t)P_DELAYIDLESLEEP), &p->p_flag);
+ /* 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 && CONFIG_JETSAM
+ /* Has jetsam attributes? */
+ if (imgp->ip_px_sa != NULL && (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_SET)) {
+ memorystatus_update(p, px_sa.psa_priority, 0, (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY),
+ TRUE, px_sa.psa_high_water_mark, (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND),
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_MEMLIMIT_FATAL));
+ }
+#endif
+ }
+
+ /*
+ * 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);
+ /* then flag exec has occurred */
+ /* notify only if it has not failed due to FP Key error */
+ if ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0)
+ proc_knote(p, NOTE_EXEC);
+ } else {
+ /* reset the importance attribute from our previous life */
+ task_importance_reset(p->task);
+
+ /* reset atm context from task */
+ task_atm_reset(p->task);
+ }
+
+ /*
+ * 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;
+
+ exec_handle_spawnattr_policy(p, psa->psa_apptype, psa->psa_qos_clamp,
+ portwatch_ports, portwatch_count);
+ }
+
+ /* Apply the main thread qos */
+ if (error == 0) {
+ thread_t main_thread = (imgp->ip_new_thread != NULL) ? imgp->ip_new_thread : current_thread();
+
+ task_set_main_thread_qos(p->task, main_thread);
+ }
+
+ /*
+ * 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 (portwatch_ports != NULL) {
+ for (int i = 0; i < portwatch_count; i++) {
+ ipc_port_t port = NULL;
+ if ((port = portwatch_ports[i]) != NULL) {
+ ipc_port_release_send(port);
+ }
+ }
+ FREE(portwatch_ports, M_TEMP);
+ portwatch_ports = NULL;
+ portwatch_count = 0;
+ }
+
+ /*
+ * 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 (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_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);
+#endif
+ }
+
+#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_PROC(exec__success);
+ }
+ }
+
+ if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL)
+ (*dtrace_proc_waitfor_hook)(p);
+#endif
+
+ /* 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)
+ (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);
+ if (exec_done == FALSE) {
+ task_deallocate(get_threadtask(imgp->ip_new_thread));
+ thread_deallocate(imgp->ip_new_thread);
+ }
+ } else {
+ /* someone is doing it for us; just skip it */
+ proc_unlock(p);
+ }
+ } else {
+
+ /*
+ * Return to the child
+ *
+ * Note: the image activator earlier dropped the
+ * task/thread references to the newly spawned
+ * process; this is OK, since we still have suspended
+ * queue references on them, so we should be fine
+ * with the delayed resume of the thread here.
+ */
+ (void)thread_resume(imgp->ip_new_thread);
+ }
+ }
+ if (bufp != NULL) {
+ FREE(bufp, M_TEMP);
+ }
+
+ return(error);
+}
+
+
+/*
+ * 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;
+ char alt_p_comm[sizeof(p->p_comm)] = {0}; /* for PowerPC */
+ int is_64 = IS_64BIT_PROCESS(p);
+ struct vfs_context context;
+ struct uthread *uthread;
+
+ 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 : IMGPF_NONE) | ((p->p_flag & P_DISABLE_ASLR) ? IMGPF_DISABLE_ASLR : IMGPF_NONE);
+ imgp->ip_p_comm = alt_p_comm; /* for PowerPC */
+ imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
+ imgp->ip_mac_return = 0;
+
+ uthread = get_bsdthread_info(current_thread());
+ if (uthread->uu_flag & UT_VFORK) {
+ imgp->ip_flags |= IMGPF_VFORK_EXEC;
+ }
+
+#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
+
+ error = exec_activate_image(imgp);
+
+ kauth_cred_unref(&context.vc_ucred);
+
+ /* Image not claimed by any activator? */
+ if (error == -1)
+ error = ENOEXEC;
+
+ if (error == 0) {
+ exec_resettextvp(p, imgp);
+ error = check_for_signature(p, imgp);
+ }
+ 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 (!error) {
+ /* Sever any extant thread affinity */
+ thread_affinity_exec(current_thread());
+
+ thread_t main_thread = (imgp->ip_new_thread != NULL) ? imgp->ip_new_thread : current_thread();
+
+ task_set_main_thread_qos(p->task, main_thread);
+
+ /* reset task importance */
+ task_importance_reset(p->task);
+
+ /* reset atm context from task */
+ task_atm_reset(p->task);
+
+ DTRACE_PROC(exec__success);
+
+#if CONFIG_DTRACE
+ if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL)
+ (*dtrace_proc_waitfor_hook)(p);
+#endif
+
+ if (imgp->ip_flags & IMGPF_VFORK_EXEC) {
+ vfork_return(p, retval, p->p_pid);
+ (void)thread_resume(imgp->ip_new_thread);
+ }
+ } else {
+ DTRACE_PROC1(exec__failure, int, error);
+ }
+
+exit_with_error:
+ if (bufp != NULL) {
+ FREE(bufp, M_TEMP);
+ }
+
+ 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;
+
+ 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).
+ *
+ * argc is an int; arg[i] are pointers; env[i] are pointers;
+ * the 0's are (void *)NULL's
+ *
+ * The stack frame layout is:
+ *
+ * +-------------+ <- p->user_stack
+ * | 16b |
+ * +-------------+
+ * | STRING AREA |
+ * | : |
+ * | : |
+ * | : |
+ * +- -- -- -- --+
+ * | PATH AREA |
+ * +-------------+
+ * | 0 |
+ * +-------------+
+ * | applev[n] |
+ * +-------------+
+ * :
+ * :
+ * +-------------+
+ * | applev[1] |
+ * +-------------+
+ * | exec_path / |
+ * | applev[0] |
+ * +-------------+
+ * | 0 |
+ * +-------------+
+ * | env[n] |
+ * +-------------+
+ * :
+ * :
+ * +-------------+
+ * | env[0] |
+ * +-------------+
+ * | 0 |
+ * +-------------+
+ * | arg[argc-1] |
+ * +-------------+
+ * :
+ * :
+ * +-------------+
+ * | arg[0] |
+ * +-------------+
+ * | argc |
+ * sp-> +-------------+
+ *
+ * Although technically a part of the STRING AREA, we treat the PATH AREA as
+ * a separate entity. This allows us to align the beginning of the PATH AREA
+ * to a pointer boundary so that the exec_path, env[i], and argv[i] pointers
+ * which preceed it on the stack are properly aligned.
+ */
+
+static int
+exec_copyout_strings(struct image_params *imgp, user_addr_t *stackp)
+{
+ proc_t p = vfs_context_proc(imgp->ip_vfs_context);
+ int ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4;
+ int ptr_area_size;
+ void *ptr_buffer_start, *ptr_buffer;
+ int string_size;
+
+ user_addr_t string_area; /* *argv[], *env[] */
+ user_addr_t ptr_area; /* argv[], env[], applev[] */
+ user_addr_t argc_area; /* argc */
+ user_addr_t stack;
+ int error;
+
+ unsigned i;
+ struct copyout_desc {
+ char *start_string;
+ int count;
+#if CONFIG_DTRACE
+ user_addr_t *dtrace_cookie;
+#endif
+ boolean_t null_term;
+ } descriptors[] = {
+ {
+ .start_string = imgp->ip_startargv,
+ .count = imgp->ip_argc,
+#if CONFIG_DTRACE
+ .dtrace_cookie = &p->p_dtrace_argv,
+#endif
+ .null_term = TRUE
+ },
+ {
+ .start_string = imgp->ip_endargv,
+ .count = imgp->ip_envc,
+#if CONFIG_DTRACE
+ .dtrace_cookie = &p->p_dtrace_envp,
+#endif
+ .null_term = TRUE
+ },
+ {
+ .start_string = imgp->ip_strings,
+ .count = 1,
+#if CONFIG_DTRACE
+ .dtrace_cookie = NULL,
+#endif
+ .null_term = FALSE
+ },
+ {
+ .start_string = imgp->ip_endenvv,
+ .count = imgp->ip_applec - 1, /* exec_path handled above */
+#if CONFIG_DTRACE
+ .dtrace_cookie = NULL,
+#endif
+ .null_term = TRUE
+ }
+ };
+
+ stack = *stackp;
+
+ /*
+ * All previous contributors to the string area
+ * should have aligned their sub-area
+ */
+ if (imgp->ip_strspace % ptr_size != 0) {
+ error = EINVAL;
+ goto bad;
+ }
+
+ /* Grow the stack down for the strings we've been building up */
+ string_size = imgp->ip_strendp - imgp->ip_strings;
+ stack -= string_size;
+ string_area = stack;
+
+ /*
+ * Need room for one pointer for each string, plus
+ * one for the NULLs terminating the argv, envv, and apple areas.
+ */
+ ptr_area_size = (imgp->ip_argc + imgp->ip_envc + imgp->ip_applec + 3) *
+ ptr_size;
+ stack -= ptr_area_size;
+ ptr_area = stack;
+
+ /* We'll construct all the pointer arrays in our string buffer,
+ * which we already know is aligned properly, and ip_argspace
+ * was used to verify we have enough space.
+ */
+ ptr_buffer_start = ptr_buffer = (void *)imgp->ip_strendp;
+
+ /*
+ * Need room for pointer-aligned argc slot.
+ */
+ stack -= ptr_size;
+ argc_area = stack;
+
+ /*
+ * Record the size of the arguments area so that sysctl_procargs()
+ * can return the argument area without having to parse the arguments.
+ */
+ proc_lock(p);
+ p->p_argc = imgp->ip_argc;
+ p->p_argslen = (int)(*stackp - string_area);
+ proc_unlock(p);
+
+ /* Return the initial stack address: the location of argc */
+ *stackp = stack;
+
+ /*
+ * Copy out the entire strings area.
+ */
+ error = copyout(imgp->ip_strings, string_area,
+ string_size);
+ if (error)
+ goto bad;
+
+ for (i = 0; i < sizeof(descriptors)/sizeof(descriptors[0]); i++) {
+ char *cur_string = descriptors[i].start_string;
+ int j;
+
+#if CONFIG_DTRACE
+ if (descriptors[i].dtrace_cookie) {
+ proc_lock(p);
+ *descriptors[i].dtrace_cookie = ptr_area + ((uintptr_t)ptr_buffer - (uintptr_t)ptr_buffer_start); /* dtrace convenience */
+ proc_unlock(p);
+ }
+#endif /* CONFIG_DTRACE */
+
+ /*
+ * For each segment (argv, envv, applev), copy as many pointers as requested
+ * to our pointer buffer.
+ */
+ for (j = 0; j < descriptors[i].count; j++) {
+ user_addr_t cur_address = string_area + (cur_string - imgp->ip_strings);
+
+ /* Copy out the pointer to the current string. Alignment has been verified */
+ if (ptr_size == 8) {
+ *(uint64_t *)ptr_buffer = (uint64_t)cur_address;
+ } else {
+ *(uint32_t *)ptr_buffer = (uint32_t)cur_address;
+ }
+
+ ptr_buffer = (void *)((uintptr_t)ptr_buffer + ptr_size);
+ cur_string += strlen(cur_string) + 1; /* Only a NUL between strings in the same area */
+ }
+
+ if (descriptors[i].null_term) {
+ if (ptr_size == 8) {
+ *(uint64_t *)ptr_buffer = 0ULL;
+ } else {
+ *(uint32_t *)ptr_buffer = 0;
+ }
+
+ ptr_buffer = (void *)((uintptr_t)ptr_buffer + ptr_size);
+ }
+ }
+
+ /*
+ * Copy out all our pointer arrays in bulk.
+ */
+ error = copyout(ptr_buffer_start, ptr_area,
+ ptr_area_size);
+ if (error)
+ goto bad;
+
+ /* argc (int32, stored in a ptr_size area) */
+ error = copyoutptr((user_addr_t)imgp->ip_argc, argc_area, ptr_size);
+ if (error)
+ goto bad;
+
+bad:
+ return(error);
+}
+
+
+/*
+ * exec_extract_strings
+ *
+ * Copy arguments and environment from user space into work area; we may
+ * have already copied some early arguments into the work area, and if
+ * so, any arguments opied in are appended to those already there.
+ * This function is the primary manipulator of ip_argspace, since
+ * these are the arguments the client of execve(2) knows about. After
+ * each argv[]/envv[] string is copied, we charge the string length
+ * and argv[]/envv[] pointer slot to ip_argspace, so that we can
+ * full preflight the arg list size.
+ *
+ * Parameters: struct image_params * the image parameter block
+ *
+ * Returns: 0 Success
+ * !0 Failure: errno
+ *
+ * Implicit returns;
+ * (imgp->ip_argc) Count of arguments, updated
+ * (imgp->ip_envc) Count of environment strings, updated
+ * (imgp->ip_argspace) Count of remaining of NCARGS
+ * (imgp->ip_interp_buffer) Interpreter and args (mutated in place)
+ *
+ *
+ * Note: The argument and environment vectors are user space pointers
+ * to arrays of user space pointers.
+ */
+static int
+exec_extract_strings(struct image_params *imgp)
+{
+ int error = 0;
+ int ptr_size = (imgp->ip_flags & IMGPF_WAS_64BIT) ? 8 : 4;
+ int new_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4;
+ user_addr_t argv = imgp->ip_user_argv;
+ user_addr_t envv = imgp->ip_user_envv;
+
+ /*
+ * Adjust space reserved for the path name by however much padding it
+ * needs. Doing this here since we didn't know if this would be a 32-
+ * or 64-bit process back in exec_save_path.
+ */
+ while (imgp->ip_strspace % new_ptr_size != 0) {
+ *imgp->ip_strendp++ = '\0';
+ imgp->ip_strspace--;
+ /* imgp->ip_argspace--; not counted towards exec args total */
+ }
+
+ /*
+ * From now on, we start attributing string space to ip_argspace
+ */
+ imgp->ip_startargv = imgp->ip_strendp;
+ imgp->ip_argc = 0;
+
+ if((imgp->ip_flags & IMGPF_INTERPRET) != 0) {
+ user_addr_t arg;
+ char *argstart, *ch;
+
+ /* First, the arguments in the "#!" string are tokenized and extracted. */
+ argstart = imgp->ip_interp_buffer;
+ while (argstart) {
+ ch = argstart;
+ while (*ch && !IS_WHITESPACE(*ch)) {
+ ch++;
+ }
+
+ if (*ch == '\0') {
+ /* last argument, no need to NUL-terminate */
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(argstart), UIO_SYSSPACE, TRUE);
+ argstart = NULL;
+ } else {
+ /* NUL-terminate */
+ *ch = '\0';
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(argstart), UIO_SYSSPACE, TRUE);
+
+ /*
+ * Find the next string. We know spaces at the end of the string have already
+ * been stripped.
+ */
+ argstart = ch + 1;
+ while (IS_WHITESPACE(*argstart)) {
+ argstart++;
+ }
+ }
+
+ /* Error-check, regardless of whether this is the last interpreter arg or not */
+ if (error)
+ goto bad;
+ if (imgp->ip_argspace < new_ptr_size) {
+ error = E2BIG;
+ goto bad;
+ }
+ imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */
+ imgp->ip_argc++;
+ }
+
+ if (argv != 0LL) {
+ /*
+ * If we are running an interpreter, replace the av[0] that was
+ * passed to execve() with the path name that was
+ * passed to execve() for interpreters which do not use the PATH
+ * to locate their script arguments.
+ */
+ error = copyinptr(argv, &arg, ptr_size);
+ if (error)
+ goto bad;
+ if (arg != 0LL) {
+ argv += ptr_size; /* consume without using */
+ }
+ }
+
+ if (imgp->ip_interp_sugid_fd != -1) {
+ char temp[19]; /* "/dev/fd/" + 10 digits + NUL */
+ snprintf(temp, sizeof(temp), "/dev/fd/%d", imgp->ip_interp_sugid_fd);
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(temp), UIO_SYSSPACE, TRUE);
+ } else {
+ error = exec_add_user_string(imgp, imgp->ip_user_fname, imgp->ip_seg, TRUE);
+ }
+
+ if (error)
+ goto bad;
+ if (imgp->ip_argspace < new_ptr_size) {
+ error = E2BIG;
+ goto bad;
+ }
+ imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */
+ imgp->ip_argc++;
+ }
+
+ while (argv != 0LL) {
+ user_addr_t arg;
+
+ error = copyinptr(argv, &arg, ptr_size);
+ if (error)
+ goto bad;
+
+ if (arg == 0LL) {
+ break;
+ }
+
+ argv += ptr_size;
+
+ /*
+ * av[n...] = arg[n]
+ */
+ error = exec_add_user_string(imgp, arg, imgp->ip_seg, TRUE);
+ if (error)
+ goto bad;
+ if (imgp->ip_argspace < new_ptr_size) {
+ error = E2BIG;
+ goto bad;
+ }
+ imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */
+ imgp->ip_argc++;
+ }
+
+ /* Save space for argv[] NULL terminator */
+ if (imgp->ip_argspace < new_ptr_size) {
+ error = E2BIG;
+ goto bad;
+ }
+ imgp->ip_argspace -= new_ptr_size;
+
+ /* Note where the args ends and env begins. */
+ imgp->ip_endargv = imgp->ip_strendp;
+ imgp->ip_envc = 0;
+
+ /* Now, get the environment */
+ while (envv != 0LL) {
+ user_addr_t env;
+
+ error = copyinptr(envv, &env, ptr_size);
+ if (error)
+ goto bad;
+
+ envv += ptr_size;
+ if (env == 0LL) {
+ break;
+ }
+ /*
+ * av[n...] = env[n]
+ */
+ error = exec_add_user_string(imgp, env, imgp->ip_seg, TRUE);
+ if (error)
+ goto bad;
+ if (imgp->ip_argspace < new_ptr_size) {
+ error = E2BIG;
+ goto bad;
+ }
+ imgp->ip_argspace -= new_ptr_size; /* to hold envv[] entry */
+ imgp->ip_envc++;
+ }
+
+ /* Save space for envv[] NULL terminator */
+ if (imgp->ip_argspace < new_ptr_size) {
+ error = E2BIG;
+ goto bad;
+ }
+ imgp->ip_argspace -= new_ptr_size;
+
+ /* Align the tail of the combined argv+envv area */
+ while (imgp->ip_strspace % new_ptr_size != 0) {
+ if (imgp->ip_argspace < 1) {
+ error = E2BIG;
+ goto bad;
+ }
+ *imgp->ip_strendp++ = '\0';
+ imgp->ip_strspace--;
+ imgp->ip_argspace--;
+ }
+
+ /* Note where the envv ends and applev begins. */
+ imgp->ip_endenvv = imgp->ip_strendp;
+
+ /*
+ * From now on, we are no longer charging argument
+ * space to ip_argspace.
+ */
+
+bad:
+ return error;
+}
+
+static char *
+random_hex_str(char *str, int len, boolean_t embedNUL)
+{
+ uint64_t low, high, value;
+ int idx;
+ char digit;
+
+ /* A 64-bit value will only take 16 characters, plus '0x' and NULL. */
+ if (len > 19)
+ len = 19;
+
+ /* We need enough room for at least 1 digit */
+ if (len < 4)
+ return (NULL);
+
+ low = random();
+ high = random();
+ value = high << 32 | low;
+
+ if (embedNUL) {
+ /*
+ * Zero a byte to protect against C string vulnerabilities
+ * e.g. for userland __stack_chk_guard.
+ */
+ value &= ~(0xffull << 8);
+ }
+
+ str[0] = '0';
+ str[1] = 'x';
+ for (idx = 2; idx < len - 1; idx++) {
+ digit = value & 0xf;
+ value = value >> 4;
+ if (digit < 10)
+ str[idx] = '0' + digit;
+ else
+ str[idx] = 'a' + (digit - 10);
+ }
+ str[idx] = '\0';
+ return (str);
+}
+
+/*
+ * Libc has an 8-element array set up for stack guard values. It only fills
+ * in one of those entries, and both gcc and llvm seem to use only a single
+ * 8-byte guard. Until somebody needs more than an 8-byte guard value, don't
+ * do the work to construct them.
+ */
+#define GUARD_VALUES 1
+#define GUARD_KEY "stack_guard="
+
+/*
+ * System malloc needs some entropy when it is initialized.
+ */
+#define ENTROPY_VALUES 2
+#define ENTROPY_KEY "malloc_entropy="
+
+/*
+ * System malloc engages nanozone for UIAPP.
+ */
+#define NANO_ENGAGE_KEY "MallocNanoZone=1"
+
+#define PFZ_KEY "pfz="
+extern user32_addr_t commpage_text32_location;
+extern user64_addr_t commpage_text64_location;
+/*
+ * Build up the contents of the apple[] string vector
+ */
+static int
+exec_add_apple_strings(struct image_params *imgp)
+{
+ int i, error;
+ int new_ptr_size=4;
+ char guard[19];
+ char guard_vec[strlen(GUARD_KEY) + 19 * GUARD_VALUES + 1];
+
+ char entropy[19];
+ char entropy_vec[strlen(ENTROPY_KEY) + 19 * ENTROPY_VALUES + 1];
+
+ char pfz_string[strlen(PFZ_KEY) + 16 + 4 +1];
+
+ if( imgp->ip_flags & IMGPF_IS_64BIT) {
+ new_ptr_size = 8;
+ snprintf(pfz_string, sizeof(pfz_string),PFZ_KEY "0x%llx",commpage_text64_location);
+ } else {
+ snprintf(pfz_string, sizeof(pfz_string),PFZ_KEY "0x%x",commpage_text32_location);
+ }
+
+ /* exec_save_path stored the first string */
+ imgp->ip_applec = 1;
+
+ /* adding the pfz string */
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(pfz_string),UIO_SYSSPACE,FALSE);
+ if(error)
+ goto bad;
+ imgp->ip_applec++;
+
+ /* adding the NANO_ENGAGE_KEY key */
+ if (imgp->ip_px_sa) {
+ int proc_flags = (((struct _posix_spawnattr *) imgp->ip_px_sa)->psa_flags);
+
+ if ((proc_flags & _POSIX_SPAWN_NANO_ALLOCATOR) == _POSIX_SPAWN_NANO_ALLOCATOR) {
+ char uiapp_string[strlen(NANO_ENGAGE_KEY) + 1];
+
+ snprintf(uiapp_string, sizeof(uiapp_string), NANO_ENGAGE_KEY);
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(uiapp_string),UIO_SYSSPACE,FALSE);
+ if (error)
+ goto bad;
+ imgp->ip_applec++;
+ }
+ }
+
+ /*
+ * Supply libc with a collection of random values to use when
+ * implementing -fstack-protector.
+ *
+ * (The first random string always contains an embedded NUL so that
+ * __stack_chk_guard also protects against C string vulnerabilities)
+ */
+ (void)strlcpy(guard_vec, GUARD_KEY, sizeof (guard_vec));
+ for (i = 0; i < GUARD_VALUES; i++) {
+ random_hex_str(guard, sizeof (guard), i == 0);
+ if (i)
+ (void)strlcat(guard_vec, ",", sizeof (guard_vec));
+ (void)strlcat(guard_vec, guard, sizeof (guard_vec));
+ }
+
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(guard_vec), UIO_SYSSPACE, FALSE);
+ if (error)
+ goto bad;
+ imgp->ip_applec++;
+
+ /*
+ * Supply libc with entropy for system malloc.
+ */
+ (void)strlcpy(entropy_vec, ENTROPY_KEY, sizeof(entropy_vec));
+ for (i = 0; i < ENTROPY_VALUES; i++) {
+ random_hex_str(entropy, sizeof (entropy), FALSE);
+ if (i)
+ (void)strlcat(entropy_vec, ",", sizeof (entropy_vec));
+ (void)strlcat(entropy_vec, entropy, sizeof (entropy_vec));
+ }
+
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(entropy_vec), UIO_SYSSPACE, FALSE);
+ if (error)
+ goto bad;
+ imgp->ip_applec++;
+
+ /* Align the tail of the combined applev area */
+ while (imgp->ip_strspace % new_ptr_size != 0) {
+ *imgp->ip_strendp++ = '\0';
+ imgp->ip_strspace--;
+ }
+
+bad:
+ return error;
+}
+
+#define unix_stack_size(p) (p->p_rlimit[RLIMIT_STACK].rlim_cur)
+
+/*
+ * exec_check_permissions
+ *
+ * Description: Verify that the file that is being attempted to be executed
+ * is in fact allowed to be executed based on it POSIX file
+ * permissions and other access control criteria
+ *
+ * Parameters: struct image_params * the image parameter block
+ *
+ * Returns: 0 Success
+ * EACCES Permission denied
+ * ENOEXEC Executable file format error
+ * ETXTBSY Text file busy [misuse of error code]
+ * vnode_getattr:???
+ * vnode_authorize:???
+ */
+static int
+exec_check_permissions(struct image_params *imgp)
+{
+ struct vnode *vp = imgp->ip_vp;
+ struct vnode_attr *vap = imgp->ip_vattr;
+ proc_t p = vfs_context_proc(imgp->ip_vfs_context);
+ int error;
+ kauth_action_t action;
+
+ /* Only allow execution of regular files */
+ if (!vnode_isreg(vp))
+ return (EACCES);
+
+ /* Get the file attributes that we will be using here and elsewhere */
+ VATTR_INIT(vap);
+ VATTR_WANTED(vap, va_uid);
+ VATTR_WANTED(vap, va_gid);
+ VATTR_WANTED(vap, va_mode);
+ VATTR_WANTED(vap, va_fsid);
+ VATTR_WANTED(vap, va_fileid);
+ VATTR_WANTED(vap, va_data_size);
+ if ((error = vnode_getattr(vp, vap, imgp->ip_vfs_context)) != 0)
+ return (error);
+
+ /*
+ * Ensure that at least one execute bit is on - otherwise root
+ * will always succeed, and we don't want to happen unless the
+ * file really is executable.
+ */
+ if (!vfs_authopaque(vnode_mount(vp)) && ((vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0))
+ return (EACCES);
+
+ /* Disallow zero length files */
+ if (vap->va_data_size == 0)
+ return (ENOEXEC);
+
+ imgp->ip_arch_offset = (user_size_t)0;
+ imgp->ip_arch_size = vap->va_data_size;
+
+ /* Disable setuid-ness for traced programs or if MNT_NOSUID */
+ if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_lflag & P_LTRACED))
+ vap->va_mode &= ~(VSUID | VSGID);
+
+ /*
+ * Disable _POSIX_SPAWN_ALLOW_DATA_EXEC and _POSIX_SPAWN_DISABLE_ASLR
+ * flags for setuid/setgid binaries.
+ */
+ if (vap->va_mode & (VSUID | VSGID))
+ imgp->ip_flags &= ~(IMGPF_ALLOW_DATA_EXEC | IMGPF_DISABLE_ASLR);
+
+#if CONFIG_MACF
+ error = mac_vnode_check_exec(imgp->ip_vfs_context, vp, imgp);
+ if (error)
+ return (error);
+#endif
+
+ /* Check for execute permission */
+ action = KAUTH_VNODE_EXECUTE;
+ /* Traced images must also be readable */
+ if (p->p_lflag & P_LTRACED)
+ action |= KAUTH_VNODE_READ_DATA;
+ if ((error = vnode_authorize(vp, NULL, action, imgp->ip_vfs_context)) != 0)
+ return (error);
+
+#if 0
+ /* Don't let it run if anyone had it open for writing */
+ vnode_lock(vp);
+ if (vp->v_writecount) {
+ panic("going to return ETXTBSY %x", vp);
+ vnode_unlock(vp);
+ return (ETXTBSY);
+ }
+ vnode_unlock(vp);
+#endif
+
+
+ /* XXX May want to indicate to underlying FS that vnode is open */
+
+ return (error);
+}
+
+
+/*
+ * exec_handle_sugid
+ *
+ * Initially clear the P_SUGID in the process flags; if an SUGID process is
+ * exec'ing a non-SUGID image, then this is the point of no return.
+ *
+ * If the image being activated is SUGID, then replace the credential with a
+ * copy, disable tracing (unless the tracing process is root), reset the
+ * mach task port to revoke it, set the P_SUGID bit,
+ *
+ * If the saved user and group ID will be changing, then make sure it happens
+ * to a new credential, rather than a shared one.
+ *
+ * Set the security token (this is probably obsolete, given that the token
+ * should not technically be separate from the credential itself).
+ *
+ * Parameters: struct image_params * the image parameter block
+ *
+ * Returns: void No failure indication
+ *
+ * Implicit returns:
+ * <process credential> Potentially modified/replaced
+ * <task port> Potentially revoked
+ * <process flags> P_SUGID bit potentially modified
+ * <security token> Potentially modified
+ */
+static int
+exec_handle_sugid(struct image_params *imgp)
+{
+ kauth_cred_t cred = vfs_context_ucred(imgp->ip_vfs_context);
+ proc_t p = vfs_context_proc(imgp->ip_vfs_context);
+ int i;
+ int leave_sugid_clear = 0;
+ int mac_reset_ipc = 0;
+ int error = 0;
+#if CONFIG_MACF
+ int mac_transition, disjoint_cred = 0;
+ int label_update_return = 0;
+
+ /*
+ * Determine whether a call to update the MAC label will result in the
+ * credential changing.
+ *
+ * Note: MAC policies which do not actually end up modifying
+ * the label subsequently are strongly encouraged to
+ * return 0 for this check, since a non-zero answer will
+ * slow down the exec fast path for normal binaries.
+ */
+ mac_transition = mac_cred_check_label_update_execve(
+ imgp->ip_vfs_context,
+ imgp->ip_vp,
+ imgp->ip_arch_offset,
+ imgp->ip_scriptvp,
+ imgp->ip_scriptlabelp,
+ imgp->ip_execlabelp,
+ p,
+ imgp->ip_px_smpx);
+#endif
+
+ OSBitAndAtomic(~((uint32_t)P_SUGID), &p->p_flag);
+
+ /*
+ * Order of the following is important; group checks must go last,
+ * as we use the success of the 'ismember' check combined with the
+ * failure of the explicit match to indicate that we will be setting
+ * the egid of the process even though the new process did not
+ * require VSUID/VSGID bits in order for it to set the new group as
+ * its egid.
+ *
+ * Note: Technically, by this we are implying a call to
+ * setegid() in the new process, rather than implying
+ * it used its VSGID bit to set the effective group,
+ * even though there is no code in that process to make
+ * such a call.
+ */
+ if (((imgp->ip_origvattr->va_mode & VSUID) != 0 &&
+ kauth_cred_getuid(cred) != imgp->ip_origvattr->va_uid) ||
+ ((imgp->ip_origvattr->va_mode & VSGID) != 0 &&
+ ((kauth_cred_ismember_gid(cred, imgp->ip_origvattr->va_gid, &leave_sugid_clear) || !leave_sugid_clear) ||
+ (kauth_cred_getgid(cred) != imgp->ip_origvattr->va_gid)))) {
+
+#if CONFIG_MACF
+/* label for MAC transition and neither VSUID nor VSGID */
+handle_mac_transition:
+#endif
+
+ /*
+ * Replace the credential with a copy of itself if euid or
+ * egid change.
+ *
+ * Note: setuid binaries will automatically opt out of
+ * group resolver participation as a side effect
+ * of this operation. This is an intentional
+ * part of the security model, which requires a
+ * participating credential be established by
+ * escalating privilege, setting up all other
+ * aspects of the credential including whether
+ * or not to participate in external group
+ * membership resolution, then dropping their
+ * effective privilege to that of the desired
+ * final credential state.
+ */
+ if (imgp->ip_origvattr->va_mode & VSUID) {
+ p->p_ucred = kauth_cred_setresuid(p->p_ucred, KAUTH_UID_NONE, imgp->ip_origvattr->va_uid, imgp->ip_origvattr->va_uid, KAUTH_UID_NONE);
+ /* update cred on proc */
+ PROC_UPDATE_CREDS_ONPROC(p);
+ }
+ if (imgp->ip_origvattr->va_mode & VSGID) {
+ p->p_ucred = kauth_cred_setresgid(p->p_ucred, KAUTH_GID_NONE, imgp->ip_origvattr->va_gid, imgp->ip_origvattr->va_gid);
+ /* update cred on proc */
+ PROC_UPDATE_CREDS_ONPROC(p);
+ }
+
+#if CONFIG_MACF
+ /*
+ * If a policy has indicated that it will transition the label,
+ * before making the call into the MAC policies, get a new
+ * duplicate credential, so they can modify it without
+ * modifying any others sharing it.
+ */
+ if (mac_transition) {
+ kauth_proc_label_update_execve(p,
+ imgp->ip_vfs_context,
+ imgp->ip_vp,
+ imgp->ip_arch_offset,
+ imgp->ip_scriptvp,
+ imgp->ip_scriptlabelp,
+ imgp->ip_execlabelp,
+ &imgp->ip_csflags,
+ imgp->ip_px_smpx,
+ &disjoint_cred, /* will be non zero if disjoint */
+ &label_update_return);
+
+ if (disjoint_cred) {
+ /*
+ * If updating the MAC label resulted in a
+ * disjoint credential, flag that we need to
+ * set the P_SUGID bit. This protects
+ * against debuggers being attached by an
+ * insufficiently privileged process onto the
+ * result of a transition to a more privileged
+ * credential.
+ */
+ leave_sugid_clear = 0;
+ }
+
+ imgp->ip_mac_return = label_update_return;
+ }
+
+ mac_reset_ipc = mac_proc_check_inherit_ipc_ports(p, p->p_textvp, p->p_textoff, imgp->ip_vp, imgp->ip_arch_offset, imgp->ip_scriptvp);
+
+#endif /* CONFIG_MACF */
+
+ /*
+ * If 'leave_sugid_clear' is non-zero, then we passed the
+ * VSUID and MACF checks, and successfully determined that
+ * the previous cred was a member of the VSGID group, but
+ * that it was not the default at the time of the execve,
+ * and that the post-labelling credential was not disjoint.
+ * So we don't set the P_SUGID or reset mach ports and fds
+ * on the basis of simply running this code.
+ */
+ if (mac_reset_ipc || !leave_sugid_clear) {
+ /*
+ * Have mach reset the task and thread ports.
+ * We don't want anyone who had the ports before
+ * a setuid exec to be able to access/control the
+ * task/thread after.
+ */
+ ipc_task_reset(p->task);
+ ipc_thread_reset((imgp->ip_new_thread != NULL) ?
+ imgp->ip_new_thread : current_thread());
+ }
+
+ if (!leave_sugid_clear) {
+ /*
+ * Flag the process as setuid.
+ */
+ OSBitOrAtomic(P_SUGID, &p->p_flag);
+
+ /*
+ * Radar 2261856; setuid security hole fix
+ * XXX For setuid processes, attempt to ensure that
+ * stdin, stdout, and stderr are already allocated.
+ * We do not want userland to accidentally allocate
+ * descriptors in this range which has implied meaning
+ * to libc.
+ */
+ for (i = 0; i < 3; i++) {
+
+ if (p->p_fd->fd_ofiles[i] != NULL)
+ continue;
+
+ /*
+ * Do the kernel equivalent of
+ *
+ * if i == 0
+ * (void) open("/dev/null", O_RDONLY);
+ * else
+ * (void) open("/dev/null", O_WRONLY);
+ */
+
+ struct fileproc *fp;
+ int indx;
+ int flag;
+ struct nameidata *ndp = NULL;
+
+ if (i == 0)
+ flag = FREAD;
+ else
+ flag = FWRITE;
+
+ if ((error = falloc(p,
+ &fp, &indx, imgp->ip_vfs_context)) != 0)
+ continue;
+
+ MALLOC(ndp, struct nameidata *, sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO);
+ if (ndp == NULL) {
+ error = ENOMEM;
+ break;
+ }
+
+ NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW, UIO_SYSSPACE,
+ CAST_USER_ADDR_T("/dev/null"),
+ imgp->ip_vfs_context);
+
+ if ((error = vn_open(ndp, flag, 0)) != 0) {
+ fp_free(p, indx, fp);
+ break;
+ }
+
+ struct fileglob *fg = fp->f_fglob;
+
+ fg->fg_flag = flag;
+ fg->fg_ops = &vnops;
+ fg->fg_data = ndp->ni_vp;
+
+ vnode_put(ndp->ni_vp);
+
+ proc_fdlock(p);
+ procfdtbl_releasefd(p, indx, NULL);
+ fp_drop(p, indx, fp, 1);
+ proc_fdunlock(p);
+
+ FREE(ndp, M_TEMP);
+ }
+ }
+ }
+#if CONFIG_MACF
+ else {
+ /*
+ * We are here because we were told that the MAC label will
+ * be transitioned, and the binary is not VSUID or VSGID; to
+ * deal with this case, we could either duplicate a lot of
+ * code, or we can indicate we want to default the P_SUGID
+ * bit clear and jump back up.
+ */
+ if (mac_transition) {
+ leave_sugid_clear = 1;
+ goto handle_mac_transition;
+ }
+ }
+
+#endif /* CONFIG_MACF */
+
+ /*
+ * Implement the semantic where the effective user and group become
+ * the saved user and group in exec'ed programs.
+ */
+ p->p_ucred = kauth_cred_setsvuidgid(p->p_ucred, kauth_cred_getuid(p->p_ucred), kauth_cred_getgid(p->p_ucred));
+ /* update cred on proc */
+ PROC_UPDATE_CREDS_ONPROC(p);
+
+ /* Update the process' identity version and set the security token */
+ p->p_idversion++;
+ set_security_token(p);
+
+ return(error);
+}
+
+
+/*
+ * create_unix_stack
+ *
+ * Description: Set the user stack address for the process to the provided
+ * address. If a custom stack was not set as a result of the
+ * load process (i.e. as specified by the image file for the
+ * executable), then allocate the stack in the provided map and
+ * set up appropriate guard pages for enforcing administrative
+ * limits on stack growth, if they end up being needed.
+ *
+ * Parameters: p Process to set stack on
+ * load_result Information from mach-o load commands
+ * map Address map in which to allocate the new stack
+ *
+ * Returns: KERN_SUCCESS Stack successfully created
+ * !KERN_SUCCESS Mach failure code
+ */
+static kern_return_t
+create_unix_stack(vm_map_t map, load_result_t* load_result,
+ proc_t p)
projects / apple / xnu.git / blobdiff