#include <sys/signal.h>
#include <sys/aio_kern.h>
#include <sys/sysproto.h>
+#include <sys/persona.h>
#if SYSV_SHM
#include <sys/shm_internal.h> /* shmexec() */
#endif
#include <kern/assert.h>
#include <kern/task.h>
#include <kern/coalition.h>
+#include <kern/kalloc.h>
#if CONFIG_MACF
#include <security/mac.h>
#endif
/* support for child creation in exec after vfork */
-thread_t fork_create_child(task_t parent_task, coalition_t parent_coalition, proc_t child_proc, int inherit_memory, int is64bit);
+thread_t fork_create_child(task_t parent_task, coalition_t *parent_coalition, proc_t child_proc, int inherit_memory, int is64bit);
void vfork_exit(proc_t p, int rv);
-int setsigvec(proc_t, thread_t, int, struct __kern_sigaction *, boolean_t in_sigstart);
extern void proc_apply_task_networkbg_internal(proc_t, thread_t);
/*
#endif
extern struct savearea *get_user_regs(thread_t);
+extern kern_return_t machine_thread_neon_state_initialize(thread_t thread);
__attribute__((noinline)) int __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__(mach_port_t task_access_port, int32_t new_pid);
extern vm_map_t bsd_pageable_map;
extern const struct fileops vnops;
-#define ROUND_PTR(type, addr) \
- (type *)( ( (uintptr_t)(addr) + 16 - 1) \
- & ~(16 - 1) )
+#define USER_ADDR_ALIGN(addr, val) \
+ ( ( (user_addr_t)(addr) + (val) - 1) \
+ & ~((val) - 1) )
struct image_params; /* Forward */
static int exec_activate_image(struct image_params *imgp);
static int check_for_signature(proc_t, struct image_params *);
static void exec_prefault_data(proc_t, struct image_params *, load_result_t *);
static errno_t exec_handle_port_actions(struct image_params *imgp, short psa_flags, boolean_t * portwatch_present, ipc_port_t * portwatch_ports);
-static errno_t exec_handle_spawnattr_policy(proc_t p, int psa_apptype, uint64_t psa_qos_clamp,
+static errno_t exec_handle_spawnattr_policy(proc_t p, int psa_apptype, uint64_t psa_qos_clamp, uint64_t psa_darwin_role,
ipc_port_t * portwatch_ports, int portwatch_count);
/*
return error;
}
+/*
+ * dyld is now passed the executable path as a getenv-like variable
+ * in the same fashion as the stack_guard and malloc_entropy keys.
+ */
+#define EXECUTABLE_KEY "executable_path="
+
/*
* exec_save_path
*
* unacceptable for dyld.
*/
static int
-exec_save_path(struct image_params *imgp, user_addr_t path, int seg)
+exec_save_path(struct image_params *imgp, user_addr_t path, int seg, const char **excpath)
{
int error;
- size_t len;
+ size_t len;
char *kpath;
+ // imgp->ip_strings can come out of a cache, so we need to obliterate the
+ // old path.
+ memset(imgp->ip_strings, '\0', strlen(EXECUTABLE_KEY) + MAXPATHLEN);
+
len = MIN(MAXPATHLEN, imgp->ip_strspace);
switch(seg) {
case UIO_USERSPACE32:
case UIO_USERSPACE64: /* Same for copyin()... */
- error = copyinstr(path, imgp->ip_strings, len, &len);
+ error = copyinstr(path, imgp->ip_strings + strlen(EXECUTABLE_KEY), len, &len);
break;
case UIO_SYSSPACE:
kpath = CAST_DOWN(char *,path); /* SAFE */
- error = copystr(kpath, imgp->ip_strings, len, &len);
+ error = copystr(kpath, imgp->ip_strings + strlen(EXECUTABLE_KEY), len, &len);
break;
default:
error = EFAULT;
}
if (!error) {
+ bcopy(EXECUTABLE_KEY, imgp->ip_strings, strlen(EXECUTABLE_KEY));
+ len += strlen(EXECUTABLE_KEY);
+
imgp->ip_strendp += len;
imgp->ip_strspace -= len;
+
+ if (excpath) {
+ *excpath = imgp->ip_strings + strlen(EXECUTABLE_KEY);
+ }
}
return(error);
exec_reset_save_path(imgp);
exec_save_path(imgp, CAST_USER_ADDR_T(imgp->ip_interp_buffer),
- UIO_SYSSPACE);
+ UIO_SYSSPACE, NULL);
/* Copy the entire interpreter + args for later processing into argv[] */
interp = imgp->ip_interp_buffer;
return (error);
}
+static int
+activate_thread_state(thread_t thread, load_result_t *result)
+{
+ int ret;
+
+ ret = thread_state_initialize(thread);
+ if (ret != KERN_SUCCESS) {
+ return ret;
+ }
+
+
+ if (result->threadstate) {
+ uint32_t *ts = result->threadstate;
+ uint32_t total_size = result->threadstate_sz;
+
+ while (total_size > 0) {
+ uint32_t flavor = *ts++;
+ uint32_t size = *ts++;
+
+ ret = thread_setstatus(thread, flavor, (thread_state_t)ts, size);
+ if (ret) {
+ return ret;
+ }
+ ts += size;
+ total_size -= (size + 2) * sizeof(uint32_t);
+ }
+ }
+
+ thread_setentrypoint(thread, result->entry_point);
+
+ return KERN_SUCCESS;
+}
+
+
/*
* exec_mach_imgact
*
struct _posix_spawnattr *psa = NULL;
int spawn = (imgp->ip_flags & IMGPF_SPAWN);
int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC);
+ int p_name_len;
/*
* make sure it's a Mach-O 1.0 or Mach-O 2.0 binary; the difference
*/
if (vfexec || spawn) {
if (vfexec) {
- imgp->ip_new_thread = fork_create_child(task, COALITION_NULL, p, FALSE, (imgp->ip_flags & IMGPF_IS_64BIT));
+ imgp->ip_new_thread = fork_create_child(task, NULL, p, FALSE, (imgp->ip_flags & IMGPF_IS_64BIT));
if (imgp->ip_new_thread == NULL) {
error = ENOMEM;
goto bad;
/*
* Actually load the image file we previously decided to load.
*/
- lret = load_machfile(imgp, mach_header, thread, map, &load_result);
+ lret = load_machfile(imgp, mach_header, thread, &map, &load_result);
if (lret != LOAD_SUCCESS) {
error = load_return_to_errno(lret);
p->p_cpusubtype = imgp->ip_origcpusubtype;
proc_unlock(p);
- vm_map_set_user_wire_limit(get_task_map(task), p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
+ vm_map_set_user_wire_limit(map, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
/*
* Set code-signing flags if this binary is signed, or if parent has
if (load_result.csflags & CS_VALID) {
imgp->ip_csflags |= load_result.csflags &
(CS_VALID|
- CS_HARD|CS_KILL|CS_ENFORCEMENT|CS_REQUIRE_LV|CS_DYLD_PLATFORM|
+ CS_HARD|CS_KILL|CS_RESTRICT|CS_ENFORCEMENT|CS_REQUIRE_LV|CS_DYLD_PLATFORM|
CS_EXEC_SET_HARD|CS_EXEC_SET_KILL|CS_EXEC_SET_ENFORCEMENT);
} else {
imgp->ip_csflags &= ~CS_VALID;
if (p->p_csflags & CS_EXEC_SET_INSTALLER)
imgp->ip_csflags |= CS_INSTALLER;
-
/*
* Set up the system reserved areas in the new address space.
*/
- vm_map_exec(get_task_map(task),
- task,
- (void *) p->p_fd->fd_rdir,
- cpu_type());
-
+ vm_map_exec(map, task, (void *)p->p_fd->fd_rdir, cpu_type());
+
/*
* Close file descriptors which specify close-on-exec.
*/
*/
error = exec_handle_sugid(imgp);
if (error) {
+ if (spawn || !vfexec) {
+ vm_map_deallocate(map);
+ }
goto badtoolate;
- }
+ }
+
+ /*
+ * Commit to new map.
+ *
+ * Swap the new map for the old, which consumes our new map reference but
+ * each leaves us responsible for the old_map reference. That lets us get
+ * off the pmap associated with it, and then we can release it.
+ */
+ if (!vfexec) {
+ old_map = swap_task_map(task, thread, map, !spawn);
+ vm_map_deallocate(old_map);
+ }
+
+ lret = activate_thread_state(thread, &load_result);
+ if (lret != KERN_SUCCESS) {
+ goto badtoolate;
+ }
/*
* deal with voucher on exec-calling thread.
/* Set the stack */
thread_setuserstack(thread, ap);
}
-
+
if (load_result.dynlinker) {
uint64_t ap;
int new_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4;
if (vfexec || spawn) {
vm_map_switch(old_map);
}
- /* Set the entry point */
- thread_setentrypoint(thread, load_result.entry_point);
/* Stop profiling */
stopprofclock(p);
* Remember file name for accounting.
*/
p->p_acflag &= ~AFORK;
- /* If the translated name isn't NULL, then we want to use
- * that translated name as the name we show as the "real" name.
- * Otherwise, use the name passed into exec.
+
+ /*
+ * Set p->p_comm and p->p_name to the name passed to exec
*/
- if (0 != imgp->ip_p_comm[0]) {
- bcopy((caddr_t)imgp->ip_p_comm, (caddr_t)p->p_comm,
- sizeof(p->p_comm));
- } else {
- if (imgp->ip_ndp->ni_cnd.cn_namelen > MAXCOMLEN)
- imgp->ip_ndp->ni_cnd.cn_namelen = MAXCOMLEN;
- bcopy((caddr_t)imgp->ip_ndp->ni_cnd.cn_nameptr, (caddr_t)p->p_comm,
- (unsigned)imgp->ip_ndp->ni_cnd.cn_namelen);
- p->p_comm[imgp->ip_ndp->ni_cnd.cn_namelen] = '\0';
- }
+ p_name_len = sizeof(p->p_name) - 1;
+ if(imgp->ip_ndp->ni_cnd.cn_namelen > p_name_len)
+ imgp->ip_ndp->ni_cnd.cn_namelen = p_name_len;
+ bcopy((caddr_t)imgp->ip_ndp->ni_cnd.cn_nameptr, (caddr_t)p->p_name,
+ (unsigned)imgp->ip_ndp->ni_cnd.cn_namelen);
+ p->p_name[imgp->ip_ndp->ni_cnd.cn_namelen] = '\0';
+
+ if (imgp->ip_ndp->ni_cnd.cn_namelen > MAXCOMLEN)
+ imgp->ip_ndp->ni_cnd.cn_namelen = MAXCOMLEN;
+ bcopy((caddr_t)imgp->ip_ndp->ni_cnd.cn_nameptr, (caddr_t)p->p_comm,
+ (unsigned)imgp->ip_ndp->ni_cnd.cn_namelen);
+ p->p_comm[imgp->ip_ndp->ni_cnd.cn_namelen] = '\0';
pal_dbg_set_task_name( p->task );
}
}
- /*
- * Ensure the 'translated' and 'affinity' flags are cleared, since we
- * no longer run PowerPC binaries.
- */
- OSBitAndAtomic(~((uint32_t)(P_TRANSLATED | P_AFFINITY)), &p->p_flag);
-
/*
* If posix_spawned with the START_SUSPENDED flag, stop the
* process before it runs.
proc_lock(p);
p->p_stat = SSTOP;
proc_unlock(p);
- (void) task_suspend(p->task);
+ (void) task_suspend_internal(p->task);
}
}
thread_deallocate(thread);
}
+ if (load_result.threadstate) {
+ kfree(load_result.threadstate, load_result.threadstate_sz);
+ load_result.threadstate = NULL;
+ }
+
bad:
return(error);
}
* namei:???
* vn_rdwr:??? [anything vn_rdwr can return]
* <ex_imgact>:??? [anything an imgact can return]
+ * EDEADLK Process is being terminated
*/
static int
exec_activate_image(struct image_params *imgp)
{
struct nameidata *ndp = NULL;
+ const char *excpath;
int error;
int resid;
int once = 1; /* save SGUID-ness for interpreted files */
if (error)
goto bad_notrans;
- error = exec_save_path(imgp, imgp->ip_user_fname, imgp->ip_seg);
+ error = exec_save_path(imgp, imgp->ip_user_fname, imgp->ip_seg, &excpath);
if (error) {
goto bad_notrans;
}
- /* Use imgp->ip_strings, which contains the copyin-ed exec path */
- DTRACE_PROC1(exec, uintptr_t, imgp->ip_strings);
+ /* Use excpath, which contains the copyin-ed exec path */
+ DTRACE_PROC1(exec, uintptr_t, excpath);
MALLOC(ndp, struct nameidata *, sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO);
if (ndp == NULL) {
}
NDINIT(ndp, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF | AUDITVNPATH1,
- UIO_SYSSPACE, CAST_USER_ADDR_T(imgp->ip_strings), imgp->ip_vfs_context);
+ UIO_SYSSPACE, CAST_USER_ADDR_T(excpath), imgp->ip_vfs_context);
again:
error = namei(ndp);
*/
proc_lock(p);
if (p->p_lflag & P_LEXIT) {
+ error = EDEADLK;
proc_unlock(p);
goto bad_notrans;
}
imgp->ip_vp = NULL; /* already put */
imgp->ip_ndp = NULL; /* already nameidone */
- /* Use imgp->ip_strings, which exec_shell_imgact reset to the interpreter */
+ /* Use excpath, which exec_shell_imgact reset to the interpreter */
NDINIT(ndp, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF,
- UIO_SYSSPACE, CAST_USER_ADDR_T(imgp->ip_strings), imgp->ip_vfs_context);
+ UIO_SYSSPACE, CAST_USER_ADDR_T(excpath), imgp->ip_vfs_context);
proc_transend(p, 0);
goto again;
(uintptr_t)ndp->ni_vp, 0);
}
+ if (error == 0) {
+ /*
+ * Reset atm context from task
+ */
+ task_atm_reset(p->task);
+
+ /*
+ * Reset old bank context from task
+ */
+ task_bank_reset(p->task);
+ }
bad:
proc_transend(p, 0);
* Returns: 0 Success
*/
static errno_t
-exec_handle_spawnattr_policy(proc_t p, int psa_apptype, uint64_t psa_qos_clamp,
+exec_handle_spawnattr_policy(proc_t p, int psa_apptype, uint64_t psa_qos_clamp, uint64_t psa_darwin_role,
ipc_port_t * portwatch_ports, int portwatch_count)
{
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;
}
}
- if (psa_apptype != TASK_APPTYPE_NONE || qos_clamp != THREAD_QOS_UNSPECIFIED) {
- proc_set_task_spawnpolicy(p->task, apptype, qos_clamp,
+ 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) {
+ proc_set_task_spawnpolicy(p->task, apptype, qos_clamp, role,
portwatch_ports, portwatch_count);
}
}
#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;
+
+ /*
+ * TODO: rdar://problem/19981151
+ * Add entitlement check!
+ */
+ if (!kauth_cred_issuser(kauth_cred_get()))
+ return EPERM;
+
+ 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) {
+ int 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 != (int)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
+
+void
+proc_set_return_wait(proc_t p)
+{
+ proc_lock(p);
+ p->p_lflag |= P_LRETURNWAIT;
+ proc_unlock(p);
+}
+
+void
+proc_clear_return_wait(proc_t p, thread_t child_thread)
+{
+ proc_lock(p);
+
+ p->p_lflag &= ~P_LRETURNWAIT;
+ if (p->p_lflag & P_LRETURNWAITER) {
+ wakeup(&p->p_lflag);
+ }
+
+ proc_unlock(p);
+
+ (void)thread_resume(child_thread);
+}
+
+void
+proc_wait_to_return()
+{
+ proc_t p;
+
+ p = current_proc();
+ proc_lock(p);
+
+ if (p->p_lflag & P_LRETURNWAIT) {
+ p->p_lflag |= P_LRETURNWAITER;
+ do {
+ msleep(&p->p_lflag, &p->p_mlock, 0,
+ "thread_check_setup_complete", NULL);
+ } while (p->p_lflag & P_LRETURNWAIT);
+ p->p_lflag &= ~P_LRETURNWAITER;
+ }
+
+ proc_unlock(p);
+ thread_bootstrap_return();
+}
+
/*
* posix_spawn
*
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;
int portwatch_count = 0;
ipc_port_t * portwatch_ports = NULL;
vm_size_t px_sa_offset = offsetof(struct _posix_spawnattr, psa_ports);
+#if CONFIG_PERSONAS
+ struct _posix_spawn_persona_info *px_persona = NULL;
+#endif
/*
* Allocate a big chunk for locals instead of using stack since these
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;
+ imgp->ip_px_persona = NULL;
if (uap->adesc != USER_ADDR_NULL) {
if(is_64) {
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);
}
if (error)
goto bad;
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
#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)
*/
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;
+ /* Set the new task's coalition, if it is requested. */
+ coalition_t coal[COALITION_NUM_TYPES] = { COALITION_NULL };
#if CONFIG_COALITIONS
- if (imgp->ip_px_sa) {
- uint64_t cid = px_sa.psa_coalitionid;
+ 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) {
-#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
+ /*
+ * 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);
- if (error != 0) {
- if (coal != COALITION_NULL) {
#if CONFIG_COALITIONS
- coalition_remove_active(coal);
- coalition_release(coal);
+ /* set the roles of this task within each given coalition */
+ if (error == 0) {
+ kr = coalitions_set_roles(coal, get_threadtask(imgp->ip_new_thread), coal_role);
+ if (kr != KERN_SUCCESS)
+ error = EINVAL;
+ }
+
+ /* 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 */
- if (coal != COALITION_NULL) {
-#if CONFIG_COALITIONS
- coalition_remove_active(coal);
- coalition_release(coal);
-#endif /* CONFIG_COALITIONS */
+#if CONFIG_PERSONAS
+ /*
+ * If the parent isn't in a persona (launchd), and
+ * hasn't specified a new persona for the process,
+ * then we'll put the process into the system persona
+ *
+ * TODO: this will have to be re-worked because as of
+ * now, without any launchd adoption, the resulting
+ * xpcproxy process will not have sufficient
+ * privileges to setuid/gid.
+ */
+#if 0
+ if (!proc_has_persona(p) && imgp->ip_px_persona == NULL) {
+ MALLOC(px_persona, struct _posix_spawn_persona_info *,
+ sizeof(*px_persona), M_TEMP, M_WAITOK|M_ZERO);
+ if (px_persona == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+ px_persona->pspi_id = persona_get_id(g_system_persona);
+ imgp->ip_px_persona = px_persona;
}
+#endif /* 0 */
+#endif /* CONFIG_PERSONAS */
}
if (spawn_no_exec) {
}
}
+#if CONFIG_PERSONAS
+ if (spawn_no_exec && imgp->ip_px_persona != NULL) {
+ /*
+ * 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; embedded allows for a
- * boot-arg (-disable_aslr) to deal with this (which itself is
- * only honored on DEVELOPMENT or DEBUG builds of xnu).
+ *
+ * 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 */
/*
* Forcibly disallow execution from data pages for the spawned process
* 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,
+ px_sa.psa_cpumonitor_percent,
+ px_sa.psa_cpumonitor_interval * NSEC_PER_SEC,
0, TRUE);
}
}
#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));
+ /*
+ * 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.
+ * At the kernel layer, the flag is ignored.
+ *
+ * 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.
+ * The kernel layer will flag this mapping.
+ */
+ if (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND) {
+ memorystatus_update(p, px_sa.psa_priority, 0,
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY),
+ TRUE,
+ -1, TRUE,
+ px_sa.psa_memlimit_inactive, FALSE,
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND));
+ } else {
+ memorystatus_update(p, px_sa.psa_priority, 0,
+ (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),
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND));
+ }
+
}
-#endif
+#endif /* CONFIG_MEMORYSTATUS && CONFIG_JETSAM*/
}
/*
} else if (error == 0) {
/* reset the importance attribute from our previous life */
task_importance_reset(p->task);
+ }
- /* reset atm context from task */
- task_atm_reset(p->task);
+ 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(p->task);
+ proc_transend(p, 0);
+ }
}
+
/*
* Apply the spawnattr policy, apptype (which primes the task for importance donation),
* and bind any portwatch ports to the new task.
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,
+ exec_handle_spawnattr_policy(p, psa->psa_apptype, psa->psa_qos_clamp, psa->psa_darwin_role,
portwatch_ports, portwatch_count);
}
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 CONFIG_MACF
if (imgp->ip_px_smpx != NULL)
spawn_free_macpolicyinfo(imgp->ip_px_smpx);
p->exit_thread = current_thread();
proc_unlock(p);
exit1(p, 1, (int *)NULL);
+ proc_clear_return_wait(p, imgp->ip_new_thread);
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);
+ proc_clear_return_wait(p, imgp->ip_new_thread);
}
} else {
* queue references on them, so we should be fine
* with the delayed resume of the thread here.
*/
- (void)thread_resume(imgp->ip_new_thread);
+ proc_clear_return_wait(p, imgp->ip_new_thread);
}
}
if (bufp != NULL) {
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;
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;
if (imgp->ip_scriptlabelp)
mac_vnode_label_free(imgp->ip_scriptlabelp);
#endif
+
+ 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(p->task);
+ proc_transend(p, 0);
+ }
+ }
+
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();
+ 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 (imgp->ip_flags & IMGPF_VFORK_EXEC) {
vfork_return(p, retval, p->p_pid);
- (void)thread_resume(imgp->ip_new_thread);
+ proc_clear_return_wait(p, imgp->ip_new_thread);
}
} else {
DTRACE_PROC1(exec__failure, int, error);
* modifying any others sharing it.
*/
if (mac_transition) {
+ /*
+ * This hook may generate upcalls that require
+ * importance donation from the kernel.
+ * (23925818)
+ */
+ thread_t thread = current_thread();
+ thread_enable_send_importance(thread, TRUE);
kauth_proc_label_update_execve(p,
imgp->ip_vfs_context,
imgp->ip_vp,
imgp->ip_px_smpx,
&disjoint_cred, /* will be non zero if disjoint */
&label_update_return);
+ thread_enable_send_importance(thread, FALSE);
if (disjoint_cred) {
/*
#include <sys/reboot.h>
+/*
+ * load_init_program_at_path
+ *
+ * Description: Load the "init" program; in most cases, this will be "launchd"
+ *
+ * Parameters: p Process to call execve() to create
+ * the "init" program
+ * scratch_addr Page in p, scratch space
+ * path NULL terminated path
+ *
+ * Returns: KERN_SUCCESS Success
+ * !KERN_SUCCESS See execve/mac_execve for error codes
+ *
+ * Notes: The process that is passed in is the first manufactured
+ * process on the system, and gets here via bsd_ast() firing
+ * for the first time. This is done to ensure that bsd_init()
+ * has run to completion.
+ *
+ * The address map of the first manufactured process is 32 bit.
+ * WHEN this becomes 64b, this code will fail; it needs to be
+ * made 64b capable.
+ */
+static int
+load_init_program_at_path(proc_t p, user_addr_t scratch_addr, const char* path)
+{
+ uint32_t argv[3];
+ uint32_t argc = 0;
+ int retval[2];
+ struct execve_args init_exec_args;
+
+ /*
+ * Validate inputs and pre-conditions
+ */
+ assert(p);
+ assert(scratch_addr);
+ assert(path);
+
+ if (IS_64BIT_PROCESS(p)) {
+ panic("Init against 64b primordial proc not implemented");
+ }
+
+ /*
+ * Copy out program name.
+ */
+ size_t path_length = strlen(path) + 1;
+ (void) copyout(path, scratch_addr, path_length);
+
+ argv[argc++] = (uint32_t)scratch_addr;
+ scratch_addr = USER_ADDR_ALIGN(scratch_addr + path_length, 16);
+
+ /*
+ * Put out first (and only) argument, similarly.
+ * Assumes everything fits in a page as allocated above.
+ */
+ if (boothowto & RB_SINGLE) {
+ const char *init_args = "-s";
+ size_t init_args_length = strlen(init_args)+1;
+
+ copyout(init_args, scratch_addr, init_args_length);
+
+ argv[argc++] = (uint32_t)scratch_addr;
+ scratch_addr = USER_ADDR_ALIGN(scratch_addr + init_args_length, 16);
+ }
+
+ /*
+ * Null-end the argument list
+ */
+ argv[argc] = 0;
+
+ /*
+ * Copy out the argument list.
+ */
+ (void) copyout(argv, scratch_addr, sizeof(argv));
+
+ /*
+ * Set up argument block for fake call to execve.
+ */
+ init_exec_args.fname = CAST_USER_ADDR_T(argv[0]);
+ init_exec_args.argp = scratch_addr;
+ init_exec_args.envp = USER_ADDR_NULL;
+
+ /*
+ * So that init task is set with uid,gid 0 token
+ */
+ set_security_token(p);
+
+ return execve(p, &init_exec_args, retval);
+}
+
static const char * init_programs[] = {
+#if DEBUG
+ "/usr/local/sbin/launchd.debug",
+#endif
#if DEVELOPMENT || DEBUG
+ /* Remove DEBUG conditional when <rdar://problem/17931977> is fixed */
"/usr/local/sbin/launchd.development",
#endif
"/sbin/launchd",
* process on the system, and gets here via bsd_ast() firing
* for the first time. This is done to ensure that bsd_init()
* has run to completion.
+ *
+ * In DEBUG & DEVELOPMENT builds, the launchdsuffix boot-arg
+ * may be used to select a specific launchd executable. As with
+ * the kcsuffix boot-arg, setting launchdsuffix to "" or "release"
+ * will force /sbin/launchd to be selected.
+ *
+ * The DEBUG kernel will continue to check for a .development
+ * version until <rdar://problem/17931977> is fixed.
+ *
+ * Search order by build:
+ *
+ * DEBUG DEVELOPMENT RELEASE PATH
+ * ----------------------------------------------------------------------------------
+ * 1 1 NA /usr/local/sbin/launchd.$LAUNCHDSUFFIX
+ * 2 NA NA /usr/local/sbin/launchd.debug
+ * 3 2 NA /usr/local/sbin/launchd.development
+ * 4 3 1 /sbin/launchd
*/
void
load_init_program(proc_t p)
{
- vm_offset_t init_addr, addr;
- int argc;
- uint32_t argv[3];
- unsigned int i;
- int error;
- int retval[2];
- const char *init_program_name;
- struct execve_args init_exec_args;
-
- init_addr = VM_MIN_ADDRESS;
- (void) vm_allocate(current_map(), &init_addr, PAGE_SIZE, VM_FLAGS_ANYWHERE);
- if (init_addr == 0)
- init_addr++;
-
- for (i = 0; i < sizeof(init_programs)/sizeof(init_programs[0]); i++) {
-
- init_program_name = init_programs[i];
- addr = init_addr;
- argc = 0;
-
- /*
- * Copy out program name.
- */
- (void) copyout(init_program_name, CAST_USER_ADDR_T(addr), strlen(init_program_name)+1);
-
- argv[argc++] = (uint32_t)addr;
- addr += strlen(init_program_name)+1;
- addr = (vm_offset_t)ROUND_PTR(char, addr);
-
- /*
- * Put out first (and only) argument, similarly.
- * Assumes everything fits in a page as allocated above.
- */
- if (boothowto & RB_SINGLE) {
- const char *init_args = "-s";
+ uint32_t i;
+ int error;
+ vm_offset_t scratch_addr = VM_MIN_ADDRESS;
- copyout(init_args, CAST_USER_ADDR_T(addr), strlen(init_args)+1);
+ (void) vm_allocate(current_map(), &scratch_addr, PAGE_SIZE, VM_FLAGS_ANYWHERE);
+#if CONFIG_MEMORYSTATUS && CONFIG_JETSAM
+ (void) memorystatus_init_at_boot_snapshot();
+#endif /* CONFIG_MEMORYSTATUS && CONFIG_JETSAM */
+
+#if DEBUG || DEVELOPMENT
+ /* Check for boot-arg suffix first */
+ char launchd_suffix[64];
+ if (PE_parse_boot_argn("launchdsuffix", launchd_suffix, sizeof(launchd_suffix))) {
+ char launchd_path[128];
+ boolean_t is_release_suffix = ((launchd_suffix[0] == 0) ||
+ (strcmp(launchd_suffix, "release") == 0));
+
+ if (is_release_suffix) {
+ error = load_init_program_at_path(p, CAST_USER_ADDR_T(scratch_addr), "/sbin/launchd");
+ if (!error)
+ return;
+
+ panic("Process 1 exec of launchd.release failed, errno %d", error);
+ } else {
+ strlcpy(launchd_path, "/usr/local/sbin/launchd.", sizeof(launchd_path));
+ strlcat(launchd_path, launchd_suffix, sizeof(launchd_path));
- argv[argc++] = (uint32_t)addr;
- addr += strlen(init_args)+1;
- addr = (vm_offset_t)ROUND_PTR(char, addr);
+ /* All the error data is lost in the loop below, don't
+ * attempt to save it. */
+ if (!load_init_program_at_path(p, CAST_USER_ADDR_T(scratch_addr), launchd_path)) {
+ return;
+ }
}
+ }
+#endif
- /*
- * Null-end the argument list
- */
- argv[argc] = 0;
-
- /*
- * Copy out the argument list.
- */
- (void) copyout(argv, CAST_USER_ADDR_T(addr), sizeof(argv));
-
- /*
- * Set up argument block for fake call to execve.
- */
- init_exec_args.fname = CAST_USER_ADDR_T(argv[0]);
- init_exec_args.argp = CAST_USER_ADDR_T((char **)addr);
- init_exec_args.envp = CAST_USER_ADDR_T(0);
-
- /*
- * So that init task is set with uid,gid 0 token
- */
- set_security_token(p);
-
- error = execve(p, &init_exec_args, retval);
+ error = ENOENT;
+ for (i = 0; i < sizeof(init_programs)/sizeof(init_programs[0]); i++) {
+ error = load_init_program_at_path(p, CAST_USER_ADDR_T(scratch_addr), init_programs[i]);
if (!error)
return;
}
- panic("Process 1 exec of %s failed, errno %d", init_program_name, error);
+ panic("Process 1 exec of %s failed, errno %d", ((i == 0) ? "<null>" : init_programs[i-1]), error);
}
/*
void *blob;
int error;
+ if (cs_debug > 2)
+ csvnode_print_debug(p->p_textvp);
+
+ const int can_skip_taskgated = csproc_get_platform_binary(p) && !csproc_get_platform_path(p);
+ if (can_skip_taskgated) {
+ if (cs_debug) printf("taskgated not required for: %s\n", p->p_name);
+ *require_success = FALSE;
+ return FALSE;
+ }
+
if ((p->p_csflags & CS_VALID) == 0) {
*require_success = FALSE;
return TRUE;
return FALSE;
}
+ if (cs_debug) printf("taskgated required for: %s\n", p->p_name);
+
return TRUE;
}
*require_success = FALSE;
- return 0;
+ return FALSE;
}
/*