/*
- * Copyright (c) 2000-2011 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2013 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <sys/signal.h>
#include <sys/aio_kern.h>
#include <sys/sysproto.h>
+#include <sys/persona.h>
+#include <sys/reason.h>
#if SYSV_SHM
#include <sys/shm_internal.h> /* shmexec() */
#endif
#include <sys/spawn_internal.h>
#include <sys/process_policy.h>
#include <sys/codesign.h>
+#include <sys/random.h>
#include <crypto/sha1.h>
#include <libkern/libkern.h>
#include <kern/assert.h>
#include <kern/task.h>
#include <kern/coalition.h>
+#include <kern/policy_internal.h>
+#include <kern/kalloc.h>
+
+#include <os/log.h>
#if CONFIG_MACF
-#include <security/mac.h>
+#include <security/mac_framework.h>
#include <security/mac_mach_internal.h>
#endif
#if CONFIG_DTRACE
/* Do not include dtrace.h, it redefines kmem_[alloc/free] */
-extern void (*dtrace_fasttrap_exec_ptr)(proc_t);
+extern void dtrace_proc_exec(proc_t);
extern void (*dtrace_proc_waitfor_exec_ptr)(proc_t);
-extern void (*dtrace_helpers_cleanup)(proc_t);
-extern void dtrace_lazy_dofs_destroy(proc_t);
/*
* Since dtrace_proc_waitfor_exec_ptr can be added/removed in dtrace_subr.c,
#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, int in_exec);
void vfork_exit(proc_t p, int rv);
extern void proc_apply_task_networkbg_internal(proc_t, thread_t);
+extern void task_set_did_exec_flag(task_t task);
+extern void task_clear_exec_copy_flag(task_t task);
+proc_t proc_exec_switch_task(proc_t p, task_t old_task, task_t new_task, thread_t new_thread);
+boolean_t task_is_active(task_t);
+boolean_t thread_is_active(thread_t thread);
+void thread_copy_resource_info(thread_t dst_thread, thread_t src_thread);
+void *ipc_importance_exec_switch_task(task_t old_task, task_t new_task);
+extern void ipc_importance_release(void *elem);
/*
* Mach things for which prototypes are unavailable from Mach headers
( ( (user_addr_t)(addr) + (val) - 1) \
& ~((val) - 1) )
+ /* Platform Code Exec Logging */
+static int platform_exec_logging = 0;
+
+SYSCTL_DECL(_security_mac);
+
+SYSCTL_INT(_security_mac, OID_AUTO, platform_exec_logging, CTLFLAG_RW, &platform_exec_logging, 0,
+ "log cdhashes for all platform binary executions");
+
+static os_log_t peLog = OS_LOG_DEFAULT;
+
struct image_params; /* Forward */
static int exec_activate_image(struct image_params *imgp);
static int exec_copyout_strings(struct image_params *imgp, user_addr_t *stackp);
static int execargs_free(struct image_params *imgp);
static int exec_check_permissions(struct image_params *imgp);
static int exec_extract_strings(struct image_params *imgp);
-static int exec_add_apple_strings(struct image_params *imgp);
+static int exec_add_apple_strings(struct image_params *imgp, const load_result_t *load_result);
static int exec_handle_sugid(struct image_params *imgp);
static int sugid_scripts = 0;
SYSCTL_INT (_kern, OID_AUTO, sugid_scripts, CTLFLAG_RW | CTLFLAG_LOCKED, &sugid_scripts, 0, "");
static void exec_resettextvp(proc_t, struct image_params *);
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_port_actions(struct image_params *imgp, 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, uint64_t psa_darwin_role,
ipc_port_t * portwatch_ports, int portwatch_count);
char *ihp;
char *line_startp, *line_endp;
char *interp;
- proc_t p;
- struct fileproc *fp;
- int fd;
- int error;
/*
* Make sure it's a shell script. If we've already redirected
*interp++ = *ihp;
*interp = '\0';
+#if !SECURE_KERNEL
/*
- * If we have a SUID oder SGID script, create a file descriptor
+ * If we have an SUID or SGID script, create a file descriptor
* from the vnode and pass /dev/fd/%d instead of the actual
* path name so that the script does not get opened twice
*/
if (imgp->ip_origvattr->va_mode & (VSUID | VSGID)) {
+ proc_t p;
+ struct fileproc *fp;
+ int fd;
+ int error;
+
p = vfs_context_proc(imgp->ip_vfs_context);
error = falloc(p, &fp, &fd, imgp->ip_vfs_context);
if (error)
imgp->ip_interp_sugid_fd = fd;
}
+#endif
return (-3);
}
if (imgp->ip_origcputype != 0) {
/* Fat header previously matched, don't allow another fat file inside */
- return (-1);
+ error = -1; /* not claimed */
+ goto bad;
}
/* Make sure it's a fat binary */
return (error);
}
+static int
+activate_exec_state(task_t task, proc_t p, thread_t thread, load_result_t *result)
+{
+ int ret;
+
+ task_set_dyld_info(task, MACH_VM_MIN_ADDRESS, 0);
+ if (result->is64bit) {
+ task_set_64bit(task, TRUE);
+ OSBitOrAtomic(P_LP64, &p->p_flag);
+ } else {
+ task_set_64bit(task, FALSE);
+ OSBitAndAtomic(~((uint32_t)P_LP64), &p->p_flag);
+ }
+
+ 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;
+}
+
+
+/*
+ * Set p->p_comm and p->p_name to the name passed to exec
+ */
+static void
+set_proc_name(struct image_params *imgp, proc_t p)
+{
+ int 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';
+}
+
+static uint64_t get_va_fsid(struct vnode_attr *vap)
+{
+ if (VATTR_IS_SUPPORTED(vap, va_fsid64)) {
+ return *(uint64_t *)&vap->va_fsid64;
+ } else {
+ return vap->va_fsid;
+ }
+}
+
/*
* exec_mach_imgact
*
thread_t thread;
struct uthread *uthread;
vm_map_t old_map = VM_MAP_NULL;
- vm_map_t map;
+ vm_map_t map = VM_MAP_NULL;
load_return_t lret;
load_result_t load_result;
struct _posix_spawnattr *psa = NULL;
int spawn = (imgp->ip_flags & IMGPF_SPAWN);
int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC);
- int p_name_len;
+ int exec = (imgp->ip_flags & IMGPF_EXEC);
+ os_reason_t exec_failure_reason = OS_REASON_NULL;
/*
* make sure it's a Mach-O 1.0 or Mach-O 2.0 binary; the difference
if (error)
goto bad;
- error = exec_add_apple_strings(imgp);
- if (error)
- goto bad;
-
AUDIT_ARG(argv, imgp->ip_startargv, imgp->ip_argc,
imgp->ip_endargv - imgp->ip_startargv);
AUDIT_ARG(envv, imgp->ip_endargv, imgp->ip_envc,
* obtained indirectly from the image_params vfs_context_t, is the
* new child process.
*/
- if (vfexec || spawn) {
- if (vfexec) {
- 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;
- }
+ if (vfexec) {
+ imgp->ip_new_thread = fork_create_child(task, NULL, p, FALSE, (imgp->ip_flags & IMGPF_IS_64BIT), FALSE);
+ /* task and thread ref returned, will be released in __mac_execve */
+ if (imgp->ip_new_thread == NULL) {
+ error = ENOMEM;
+ goto bad;
}
-
- /* reset local idea of thread, uthread, task */
- thread = imgp->ip_new_thread;
- uthread = get_bsdthread_info(thread);
- task = new_task = get_threadtask(thread);
- map = get_task_map(task);
- } else {
- map = VM_MAP_NULL;
}
- /*
- * We set these flags here; this is OK, since if we fail after
- * this point, we have already destroyed the parent process anyway.
- */
- task_set_dyld_info(task, MACH_VM_MIN_ADDRESS, 0);
- if (imgp->ip_flags & IMGPF_IS_64BIT) {
- task_set_64bit(task, TRUE);
- OSBitOrAtomic(P_LP64, &p->p_flag);
- } else {
- task_set_64bit(task, FALSE);
- OSBitAndAtomic(~((uint32_t)P_LP64), &p->p_flag);
- }
+
+ /* reset local idea of thread, uthread, task */
+ thread = imgp->ip_new_thread;
+ uthread = get_bsdthread_info(thread);
+ task = new_task = get_threadtask(thread);
/*
* Load the Mach-O file.
/*
* 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);
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_BAD_MACHO, 0, 0);
+ if (lret == LOAD_BADMACHO_UPX) {
+ /* set anything that might be useful in the crash report */
+ set_proc_name(imgp, p);
+
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_UPX);
+ exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
+ exec_failure_reason->osr_flags |= OS_REASON_FLAG_CONSISTENT_FAILURE;
+ } else if (lret == LOAD_BADARCH_X86) {
+ /* set anything that might be useful in the crash report */
+ set_proc_name(imgp, p);
+
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_NO32EXEC);
+ exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
+ exec_failure_reason->osr_flags |= OS_REASON_FLAG_CONSISTENT_FAILURE;
+ } else {
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_BAD_MACHO);
+ }
+
goto badtoolate;
}
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_RESTRICT|CS_ENFORCEMENT|CS_REQUIRE_LV|CS_DYLD_PLATFORM|
+ (CS_VALID|CS_SIGNED|CS_DEV_CODE|
+ CS_HARD|CS_KILL|CS_RESTRICT|CS_ENFORCEMENT|CS_REQUIRE_LV|
+ CS_ENTITLEMENTS_VALIDATED|CS_DYLD_PLATFORM|
+ CS_ENTITLEMENT_FLAGS|
CS_EXEC_SET_HARD|CS_EXEC_SET_KILL|CS_EXEC_SET_ENFORCEMENT);
} else {
imgp->ip_csflags &= ~CS_VALID;
imgp->ip_csflags |= CS_KILL;
if (p->p_csflags & CS_EXEC_SET_ENFORCEMENT)
imgp->ip_csflags |= CS_ENFORCEMENT;
- if (p->p_csflags & CS_EXEC_SET_INSTALLER)
- imgp->ip_csflags |= CS_INSTALLER;
-
+ if (p->p_csflags & CS_EXEC_INHERIT_SIP) {
+ if (p->p_csflags & CS_INSTALLER)
+ imgp->ip_csflags |= CS_INSTALLER;
+ if (p->p_csflags & CS_DATAVAULT_CONTROLLER)
+ imgp->ip_csflags |= CS_DATAVAULT_CONTROLLER;
+ if (p->p_csflags & CS_NVRAM_UNRESTRICTED)
+ imgp->ip_csflags |= CS_NVRAM_UNRESTRICTED;
+ }
/*
* 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, load_result.is64bit, (void *)p->p_fd->fd_rdir, cpu_type());
+
/*
* Close file descriptors which specify close-on-exec.
*/
- fdexec(p, psa != NULL ? psa->psa_flags : 0);
+ fdexec(p, psa != NULL ? psa->psa_flags : 0, exec);
/*
* deal with set[ug]id.
*/
error = exec_handle_sugid(imgp);
if (error) {
+ vm_map_deallocate(map);
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_SUGID_FAILURE, 0, 0);
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_SUGID_FAILURE);
goto badtoolate;
- }
+ }
+
+ /*
+ * Commit to new map.
+ *
+ * Swap the new map for the old for target task, 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.
+ *
+ * The map needs to be set on the target task which is different
+ * than current task, thus swap_task_map is used instead of
+ * vm_map_switch.
+ */
+ old_map = swap_task_map(task, thread, map);
+ vm_map_deallocate(old_map);
+ old_map = NULL;
+
+ lret = activate_exec_state(task, p, thread, &load_result);
+ if (lret != KERN_SUCCESS) {
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_ACTV_THREADSTATE, 0, 0);
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_ACTV_THREADSTATE);
+ goto badtoolate;
+ }
/*
* deal with voucher on exec-calling thread.
&load_result,
p) != KERN_SUCCESS) {
error = load_return_to_errno(LOAD_NOSPACE);
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_STACK_ALLOC, 0, 0);
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_STACK_ALLOC);
goto badtoolate;
}
- if (vfexec || spawn) {
- old_map = vm_map_switch(get_task_map(task));
+ error = exec_add_apple_strings(imgp, &load_result);
+ if (error) {
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_APPLE_STRING_INIT, 0, 0);
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_APPLE_STRING_INIT);
+ goto badtoolate;
}
+ /* Switch to target task's map to copy out strings */
+ old_map = vm_map_switch(get_task_map(task));
+
if (load_result.unixproc) {
user_addr_t ap;
ap = p->user_stack;
error = exec_copyout_strings(imgp, &ap);
if (error) {
- if (vfexec || spawn)
- vm_map_switch(old_map);
+ vm_map_switch(old_map);
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_STRINGS, 0, 0);
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_STRINGS);
goto badtoolate;
}
/* 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;
error = copyoutptr(load_result.mach_header, ap, new_ptr_size);
if (error) {
- if (vfexec || spawn)
- vm_map_switch(old_map);
+ vm_map_switch(old_map);
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_DYNLINKER, 0, 0);
+ exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_DYNLINKER);
goto badtoolate;
}
task_set_dyld_info(task, load_result.all_image_info_addr,
/* Avoid immediate VM faults back into kernel */
exec_prefault_data(p, imgp, &load_result);
- if (vfexec || spawn) {
- vm_map_switch(old_map);
- }
- /* Set the entry point */
- thread_setentrypoint(thread, load_result.entry_point);
+ vm_map_switch(old_map);
/* Stop profiling */
stopprofclock(p);
*/
p->p_acflag &= ~AFORK;
+ set_proc_name(imgp, p);
+
+#if CONFIG_SECLUDED_MEMORY
+ if (secluded_for_apps &&
+ load_result.platform_binary) {
+ if (strncmp(p->p_name,
+ "Camera",
+ sizeof (p->p_name)) == 0) {
+ task_set_could_use_secluded_mem(task, TRUE);
+ } else {
+ task_set_could_use_secluded_mem(task, FALSE);
+ }
+ if (strncmp(p->p_name,
+ "mediaserverd",
+ sizeof (p->p_name)) == 0) {
+ task_set_could_also_use_secluded_mem(task, TRUE);
+ }
+ }
+#endif /* CONFIG_SECLUDED_MEMORY */
+
+ pal_dbg_set_task_name(task);
+
/*
- * Set p->p_comm and p->p_name to the name passed to exec
+ * The load result will have already been munged by AMFI to include the
+ * platform binary flag if boot-args dictated it (AMFI will mark anything
+ * that doesn't go through the upcall path as a platform binary if its
+ * enforcement is disabled).
*/
- 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 (load_result.platform_binary) {
+ if (cs_debug) {
+ printf("setting platform binary on task: pid = %d\n", p->p_pid);
+ }
- 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';
+ /*
+ * We must use 'task' here because the proc's task has not yet been
+ * switched to the new one.
+ */
+ task_set_platform_binary(task, TRUE);
+ } else {
+ if (cs_debug) {
+ printf("clearing platform binary on task: pid = %d\n", p->p_pid);
+ }
- pal_dbg_set_task_name( p->task );
+ task_set_platform_binary(task, FALSE);
+ }
#if DEVELOPMENT || DEBUG
- /*
+ /*
* Update the pid an proc name for importance base if any
*/
- task_importance_update_owner_info(p->task);
+ task_importance_update_owner_info(task);
#endif
memcpy(&p->p_uuid[0], &load_result.uuid[0], sizeof(p->p_uuid));
-// <rdar://6598155> dtrace code cleanup needed
#if CONFIG_DTRACE
- /*
- * Invalidate any predicate evaluation already cached for this thread by DTrace.
- * That's because we've just stored to p_comm and DTrace refers to that when it
- * evaluates the "execname" special variable. uid and gid may have changed as well.
- */
- dtrace_set_thread_predcache(current_thread(), 0);
-
- /*
- * Free any outstanding lazy dof entries. It is imperative we
- * always call dtrace_lazy_dofs_destroy, rather than null check
- * and call if !NULL. If we NULL test, during lazy dof faulting
- * we can race with the faulting code and proceed from here to
- * beyond the helpers cleanup. The lazy dof faulting will then
- * install new helpers which no longer belong to this process!
- */
- dtrace_lazy_dofs_destroy(p);
-
-
- /*
- * Clean up any DTrace helpers for the process.
- */
- if (p->p_dtrace_helpers != NULL && dtrace_helpers_cleanup) {
- (*dtrace_helpers_cleanup)(p);
- }
-
- /*
- * Cleanup the DTrace provider associated with this process.
- */
- proc_lock(p);
- if (p->p_dtrace_probes && dtrace_fasttrap_exec_ptr) {
- (*dtrace_fasttrap_exec_ptr)(p);
- }
- proc_unlock(p);
+ dtrace_proc_exec(p);
#endif
if (kdebug_enable) {
- long dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4;
+ long args[4] = {};
+
+ uintptr_t fsid = 0, fileid = 0;
+ if (imgp->ip_vattr) {
+ uint64_t fsid64 = get_va_fsid(imgp->ip_vattr);
+ fsid = fsid64;
+ fileid = imgp->ip_vattr->va_fileid;
+ // check for (unexpected) overflow and trace zero in that case
+ if (fsid != fsid64 || fileid != imgp->ip_vattr->va_fileid) {
+ fsid = fileid = 0;
+ }
+ }
+ KERNEL_DEBUG_CONSTANT_IST1(TRACE_DATA_EXEC, p->p_pid, fsid, fileid, 0,
+ (uintptr_t)thread_tid(thread));
/*
* Collect the pathname for tracing
*/
- kdbg_trace_string(p, &dbg_arg1, &dbg_arg2, &dbg_arg3, &dbg_arg4);
-
- if (vfexec || spawn) {
- KERNEL_DEBUG_CONSTANT1(TRACE_DATA_EXEC | DBG_FUNC_NONE,
- p->p_pid ,0,0,0, (uintptr_t)thread_tid(thread));
- KERNEL_DEBUG_CONSTANT1(TRACE_STRING_EXEC | DBG_FUNC_NONE,
- dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4, (uintptr_t)thread_tid(thread));
- } else {
- KERNEL_DEBUG_CONSTANT(TRACE_DATA_EXEC | DBG_FUNC_NONE,
- p->p_pid ,0,0,0,0);
- KERNEL_DEBUG_CONSTANT(TRACE_STRING_EXEC | DBG_FUNC_NONE,
- dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4, 0);
- }
+ kdbg_trace_string(p, &args[0], &args[1], &args[2], &args[3]);
+ KERNEL_DEBUG_CONSTANT_IST1(TRACE_STRING_EXEC, args[0], args[1],
+ args[2], args[3], (uintptr_t)thread_tid(thread));
}
/*
proc_lock(p);
p->p_stat = SSTOP;
proc_unlock(p);
- (void) task_suspend_internal(p->task);
+ (void) task_suspend_internal(task);
}
}
/* Don't allow child process to execute any instructions */
if (!spawn) {
if (vfexec) {
- psignal_vfork(p, new_task, thread, SIGKILL);
+ assert(exec_failure_reason != OS_REASON_NULL);
+ psignal_vfork_with_reason(p, new_task, thread, SIGKILL, exec_failure_reason);
+ exec_failure_reason = OS_REASON_NULL;
} else {
- psignal(p, SIGKILL);
+ assert(exec_failure_reason != OS_REASON_NULL);
+ psignal_with_reason(p, SIGKILL, exec_failure_reason);
+ exec_failure_reason = OS_REASON_NULL;
+
+ if (exec) {
+ /* Terminate the exec copy task */
+ task_terminate_internal(task);
+ }
}
/* We can't stop this system call at this point, so just pretend we succeeded */
error = 0;
+ } else {
+ os_reason_free(exec_failure_reason);
+ exec_failure_reason = OS_REASON_NULL;
}
done:
- if (!spawn) {
- /* 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);
- }
-
- /* Drop extra references for cases where we don't expect the caller to clean up */
- if (vfexec || (spawn && error == 0)) {
- task_deallocate(new_task);
- thread_deallocate(thread);
+ if (load_result.threadstate) {
+ kfree(load_result.threadstate, load_result.threadstate_sz);
+ load_result.threadstate = NULL;
}
bad:
+ /* If we hit this, we likely would have leaked an exit reason */
+ assert(exec_failure_reason == OS_REASON_NULL);
return(error);
}
*
* Description: Iterate through the available image activators, and activate
* the image associated with the imgp structure. We start with
- * the
+ * the activator for Mach-o binaries followed by that for Fat binaries
+ * for Interpreter scripts.
*
* Parameters: struct image_params * Image parameter block
*
* 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)
*/
proc_lock(p);
if (p->p_lflag & P_LEXIT) {
+ error = EDEADLK;
proc_unlock(p);
goto bad_notrans;
}
KAUTH_FILEOP_EXEC,
(uintptr_t)ndp->ni_vp, 0);
}
-
bad:
proc_transend(p, 0);
case POSIX_SPAWN_PROC_TYPE_APP_DEFAULT:
apptype = TASK_APPTYPE_APP_DEFAULT;
break;
+#if !CONFIG_EMBEDDED
case POSIX_SPAWN_PROC_TYPE_APP_TAL:
apptype = TASK_APPTYPE_APP_TAL;
break;
+#endif /* !CONFIG_EMBEDDED */
default:
apptype = TASK_APPTYPE_NONE;
/* TODO: Should an invalid value here fail the spawn? */
* and/or audit_session_spawnjoin for the current task.
*
* Parameters: struct image_params * Image parameter block
- * short psa_flags posix spawn attribute flags
*
* Returns: 0 Success
* EINVAL Failure
* ENOTSUP Illegal posix_spawn attr flag was set
*/
static errno_t
-exec_handle_port_actions(struct image_params *imgp, short psa_flags, boolean_t * portwatch_present, ipc_port_t * portwatch_ports)
+exec_handle_port_actions(struct image_params *imgp, boolean_t * portwatch_present,
+ ipc_port_t * portwatch_ports)
{
_posix_spawn_port_actions_t pacts = imgp->ip_px_spa;
+#if CONFIG_AUDIT
proc_t p = vfs_context_proc(imgp->ip_vfs_context);
+#endif
_ps_port_action_t *act = NULL;
- task_t task = p->task;
+ task_t task = get_threadtask(imgp->ip_new_thread);
ipc_port_t port = NULL;
errno_t ret = 0;
int i;
+ kern_return_t kr;
*portwatch_present = FALSE;
for (i = 0; i < pacts->pspa_count; i++) {
act = &pacts->pspa_actions[i];
- if (ipc_object_copyin(get_task_ipcspace(current_task()),
- act->new_port, MACH_MSG_TYPE_COPY_SEND,
- (ipc_object_t *) &port) != KERN_SUCCESS) {
- ret = EINVAL;
- goto done;
+ if (MACH_PORT_VALID(act->new_port)) {
+ kr = ipc_object_copyin(get_task_ipcspace(current_task()),
+ act->new_port, MACH_MSG_TYPE_COPY_SEND,
+ (ipc_object_t *) &port);
+
+ if (kr != KERN_SUCCESS) {
+ ret = EINVAL;
+ goto done;
+ }
+ } else {
+ /* it's NULL or DEAD */
+ port = CAST_MACH_NAME_TO_PORT(act->new_port);
}
switch (act->port_type) {
case PSPA_SPECIAL:
- /* Only allowed when not under vfork */
- if (!(psa_flags & POSIX_SPAWN_SETEXEC))
- ret = ENOTSUP;
- else if (task_set_special_port(task,
- act->which, port) != KERN_SUCCESS)
+ kr = task_set_special_port(task, act->which, port);
+
+ if (kr != KERN_SUCCESS)
ret = EINVAL;
break;
case PSPA_EXCEPTION:
- /* Only allowed when not under vfork */
- if (!(psa_flags & POSIX_SPAWN_SETEXEC))
- ret = ENOTSUP;
- else if (task_set_exception_ports(task,
- act->mask, port, act->behavior,
- act->flavor) != KERN_SUCCESS)
+ kr = task_set_exception_ports(task, act->mask, port,
+ act->behavior, act->flavor);
+ if (kr != KERN_SUCCESS)
ret = EINVAL;
break;
#if CONFIG_AUDIT
case PSPA_AU_SESSION:
- ret = audit_session_spawnjoin(p, port);
+ ret = audit_session_spawnjoin(p, task, port);
+ if (ret) {
+ /* audit_session_spawnjoin() has already dropped the reference in case of error. */
+ goto done;
+ }
+
break;
#endif
case PSPA_IMP_WATCHPORTS:
- if (portwatch_ports != NULL) {
+ if (portwatch_ports != NULL && IPC_PORT_VALID(port)) {
*portwatch_present = TRUE;
/* hold on to this till end of spawn */
portwatch_ports[i] = port;
- ret = 0;
- } else
+ } else {
ipc_port_release_send(port);
+ }
+
break;
default:
ret = EINVAL;
break;
}
- /* action failed, so release port resources */
-
- if (ret) {
+ if (ret) {
+ /* action failed, so release port resources */
ipc_port_release_send(port);
break;
}
if ((error = copyin(px_args->mac_extensions, psmx, px_args->mac_extensions_size)) != 0)
goto bad;
- if (PS_MAC_EXTENSIONS_SIZE(psmx->psmx_count) > px_args->mac_extensions_size) {
+ size_t extsize = PS_MAC_EXTENSIONS_SIZE(psmx->psmx_count);
+ if (extsize == 0 || extsize > px_args->mac_extensions_size) {
error = EINVAL;
goto bad;
}
}
#endif
-void
-proc_set_return_wait(proc_t p)
+#if CONFIG_PERSONAS
+static int spawn_validate_persona(struct _posix_spawn_persona_info *px_persona)
{
- proc_lock(p);
- p->p_lflag |= P_LRETURNWAIT;
- proc_unlock(p);
-}
+ int error = 0;
+ struct persona *persona = NULL;
+ int verify = px_persona->pspi_flags & POSIX_SPAWN_PERSONA_FLAGS_VERIFY;
-void
-proc_clear_return_wait(proc_t p, thread_t child_thread)
-{
- proc_lock(p);
+ /*
+ * TODO: rdar://problem/19981151
+ * Add entitlement check!
+ */
+ if (!kauth_cred_issuser(kauth_cred_get()))
+ return EPERM;
- p->p_lflag &= ~P_LRETURNWAIT;
- if (p->p_lflag & P_LRETURNWAITER) {
- wakeup(&p->p_lflag);
+ persona = persona_lookup(px_persona->pspi_id);
+ if (!persona) {
+ error = ESRCH;
+ goto out;
}
- proc_unlock(p);
+ if (verify) {
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_UID) {
+ if (px_persona->pspi_uid != persona_get_uid(persona)) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GID) {
+ if (px_persona->pspi_gid != persona_get_gid(persona)) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GROUPS) {
+ unsigned ngroups = 0;
+ gid_t groups[NGROUPS_MAX];
+
+ if (persona_get_groups(persona, &ngroups, groups,
+ px_persona->pspi_ngroups) != 0) {
+ error = EINVAL;
+ goto out;
+ }
+ if (ngroups != px_persona->pspi_ngroups) {
+ error = EINVAL;
+ goto out;
+ }
+ while (ngroups--) {
+ if (px_persona->pspi_groups[ngroups] != groups[ngroups]) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ if (px_persona->pspi_gmuid != persona_get_gmuid(persona)) {
+ error = EINVAL;
+ goto out;
+ }
+ }
+ }
+
+out:
+ if (persona)
+ persona_put(persona);
- (void)thread_resume(child_thread);
+ return error;
}
-void
-proc_wait_to_return()
+static int spawn_persona_adopt(proc_t p, struct _posix_spawn_persona_info *px_persona)
{
- proc_t p;
+ int ret;
+ kauth_cred_t cred;
+ struct persona *persona = NULL;
+ int override = !!(px_persona->pspi_flags & POSIX_SPAWN_PERSONA_FLAGS_OVERRIDE);
- p = current_proc();
- proc_lock(p);
+ 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;
- 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;
+ cred = persona_get_cred(persona);
+ if (!cred) {
+ ret = EINVAL;
+ goto out;
}
- proc_unlock(p);
- thread_bootstrap_return();
+ 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
/*
* posix_spawn
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);
+ vm_size_t px_sa_offset = offsetof(struct _posix_spawnattr, psa_ports);
+ task_t new_task = NULL;
+ boolean_t should_release_proc_ref = FALSE;
+ void *inherit = NULL;
+#if CONFIG_PERSONAS
+ struct _posix_spawn_persona_info *px_persona = NULL;
+#endif
/*
* Allocate a big chunk for locals instead of using stack since these
imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT : IMGPF_NONE);
imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
imgp->ip_mac_return = 0;
- imgp->ip_reserved = NULL;
+ imgp->ip_px_persona = NULL;
+ imgp->ip_cs_error = OS_REASON_NULL;
if (uap->adesc != USER_ADDR_NULL) {
if(is_64) {
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.reserved = 0;
- px_args.reserved_size = 0;
+ 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;
if (px_args.file_actions_size != 0) {
/* Limit file_actions to allowed number of open files */
int maxfa = (p->p_limit ? p->p_rlimit[RLIMIT_NOFILE].rlim_cur : NOFILE);
+ size_t maxfa_size = PSF_ACTIONS_SIZE(maxfa);
if (px_args.file_actions_size < PSF_ACTIONS_SIZE(1) ||
- px_args.file_actions_size > PSF_ACTIONS_SIZE(maxfa)) {
+ maxfa_size == 0 || px_args.file_actions_size > maxfa_size) {
error = EINVAL;
goto bad;
}
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) {
+ size_t psfsize = PSF_ACTIONS_SIZE(px_sfap->psfa_act_count);
+ if (psfsize == 0 || psfsize != px_args.file_actions_size) {
error = EINVAL;
goto bad;
}
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) {
+ size_t pasize = PS_PORT_ACTIONS_SIZE(px_spap->pspa_count);
+ if (pasize == 0 || pasize != px_args.port_actions_size) {
error = EINVAL;
goto bad;
}
}
+#if CONFIG_PERSONAS
+ /* copy in the persona info */
+ if (px_args.persona_info_size != 0 && px_args.persona_info != 0) {
+ /* for now, we need the exact same struct in user space */
+ if (px_args.persona_info_size != sizeof(*px_persona)) {
+ error = ERANGE;
+ goto bad;
+ }
+
+ MALLOC(px_persona, struct _posix_spawn_persona_info *, px_args.persona_info_size, M_TEMP, M_WAITOK|M_ZERO);
+ if (px_persona == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
+ imgp->ip_px_persona = px_persona;
+ if ((error = copyin(px_args.persona_info, px_persona,
+ px_args.persona_info_size)) != 0)
+ goto bad;
+ if ((error = spawn_validate_persona(px_persona)) != 0)
+ goto bad;
+ }
+#endif
#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)
do_fork1:
#endif /* CONFIG_COALITIONS */
+ /*
+ * note that this will implicitly inherit the
+ * caller's persona (if it exists)
+ */
error = fork1(p, &imgp->ip_new_thread, PROC_CREATE_SPAWN, coal);
+ /* returns a thread and task reference */
+ if (error == 0) {
+ new_task = get_threadtask(imgp->ip_new_thread);
+ }
#if CONFIG_COALITIONS
/* set the roles of this task within each given coalition */
if (error == 0) {
- kr = coalitions_set_roles(coal, get_threadtask(imgp->ip_new_thread), coal_role);
+ kr = coalitions_set_roles(coal, new_task, coal_role);
if (kr != KERN_SUCCESS)
error = EINVAL;
+ if (kdebug_debugid_enabled(MACHDBG_CODE(DBG_MACH_COALITION,
+ MACH_COALITION_ADOPT))) {
+ for (i = 0; i < COALITION_NUM_TYPES; i++) {
+ if (coal[i] != COALITION_NULL) {
+ /*
+ * On 32-bit targets, uniqueid
+ * will get truncated to 32 bits
+ */
+ KDBG_RELEASE(MACHDBG_CODE(
+ DBG_MACH_COALITION,
+ MACH_COALITION_ADOPT),
+ coalition_id(coal[i]),
+ get_task_uniqueid(new_task));
+ }
+ }
+ }
}
/* drop our references and activations - fork1() now holds them */
imgp->ip_flags |= IMGPF_SPAWN; /* spawn w/o exec */
spawn_no_exec = TRUE; /* used in later tests */
- }
-
- if (spawn_no_exec) {
- p = (proc_t)get_bsdthreadtask_info(imgp->ip_new_thread);
-
+#if CONFIG_PERSONAS
/*
- * We had to wait until this point before firing the
- * proc:::create probe, otherwise p would not point to the
- * child process.
+ * 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.
*/
- 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() */
+#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 */
+ } else {
+ /*
+ * For execve case, create a new task and thread
+ * which points to current_proc. The current_proc will point
+ * to the new task after image activation and proc ref drain.
+ *
+ * proc (current_proc) <----- old_task (current_task)
+ * ^ | ^
+ * | | |
+ * | ----------------------------------
+ * |
+ * --------- new_task (task marked as TF_EXEC_COPY)
+ *
+ * After image activation, the proc will point to the new task
+ * and would look like following.
+ *
+ * proc (current_proc) <----- old_task (current_task, marked as TPF_DID_EXEC)
+ * ^ |
+ * | |
+ * | ----------> new_task
+ * | |
+ * -----------------
+ *
+ * During exec any transition from new_task -> proc is fine, but don't allow
+ * transition from proc->task, since it will modify old_task.
+ */
+ imgp->ip_new_thread = fork_create_child(current_task(),
+ NULL, p, FALSE, p->p_flag & P_LP64, TRUE);
+ /* task and thread ref returned by fork_create_child */
+ if (imgp->ip_new_thread == NULL) {
+ error = ENOMEM;
+ goto bad;
+ }
- /*
- * 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;
+ new_task = get_threadtask(imgp->ip_new_thread);
+ imgp->ip_flags |= IMGPF_EXEC;
+ }
+
+ if (spawn_no_exec) {
+ p = (proc_t)get_bsdthreadtask_info(imgp->ip_new_thread);
+
+ /*
+ * We had to wait until this point before firing the
+ * proc:::create probe, otherwise p would not point to the
+ * child process.
+ */
+ DTRACE_PROC1(create, proc_t, p);
+ }
+ assert(p != NULL);
+
+ context.vc_thread = imgp->ip_new_thread;
+ context.vc_ucred = p->p_ucred; /* XXX must NOT be kauth_cred_get() */
/*
* Post fdcopy(), pre exec_handle_sugid() - this is where we want
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)
+ if ((error = exec_handle_port_actions(imgp, &portwatch_present, portwatch_ports)) != 0)
goto bad;
if (portwatch_present == FALSE && portwatch_ports != NULL) {
* 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.
+ * Modifications to p_ucred must be guarded using the
+ * proc's ucred lock. This prevents others from accessing
+ * a garbage credential.
*/
- if (px_sa.psa_flags & POSIX_SPAWN_RESETIDS) {
- kauth_cred_t my_cred = p->p_ucred;
+ while (px_sa.psa_flags & POSIX_SPAWN_RESETIDS) {
+ kauth_cred_t my_cred = kauth_cred_proc_ref(p);
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);
+
+ if (my_cred == my_new_cred) {
+ kauth_cred_unref(&my_cred);
+ break;
}
+
+ /* update cred on proc */
+ proc_ucred_lock(p);
+
+ if (p->p_ucred != my_cred) {
+ proc_ucred_unlock(p);
+ kauth_cred_unref(&my_new_cred);
+ continue;
+ }
+
+ /* donate cred reference on my_new_cred to p->p_ucred */
+ p->p_ucred = my_new_cred;
+ PROC_UPDATE_CREDS_ONPROC(p);
+ proc_ucred_unlock(p);
+
+ /* drop additional reference that was taken on the previous cred */
+ kauth_cred_unref(&my_cred);
}
+#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.
OSBitOrAtomic(P_DISABLE_ASLR, &p->p_flag);
#endif /* !SECURE_KERNEL */
+ /* Randomize high bits of ASLR slide */
+ if (px_sa.psa_flags & _POSIX_SPAWN_HIGH_BITS_ASLR)
+ imgp->ip_flags |= IMGPF_HIGH_BITS_ASLR;
+
/*
* Forcibly disallow execution from data pages for the spawned process
* even if it would otherwise be permitted by the architecture default.
*/
error = exec_activate_image(imgp);
+ if (error == 0 && !spawn_no_exec) {
+ p = proc_exec_switch_task(p, current_task(), new_task, imgp->ip_new_thread);
+ /* proc ref returned */
+ should_release_proc_ref = TRUE;
+ }
+
if (error == 0) {
/* process completed the exec */
exec_done = TRUE;
* 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);
- }
+ thread_t child_thread = imgp->ip_new_thread;
+ uthread_t child_uthread = get_bsdthread_info(child_thread);
/*
* Mask a list of signals, instead of them being unmasked, if
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);
+ for (sig = 1; sig < NSIG; sig++)
+ if (px_sa.psa_sigdefault & (1 << (sig-1))) {
+ error = setsigvec(p, child_thread, sig, &vec, spawn_no_exec);
}
}
if (error == 0) {
/* reset delay idle sleep status if set */
+#if !CONFIG_EMBEDDED
if ((p->p_flag & P_DELAYIDLESLEEP) == P_DELAYIDLESLEEP)
OSBitAndAtomic(~((uint32_t)P_DELAYIDLESLEEP), &p->p_flag);
+#endif /* !CONFIG_EMBEDDED */
/* upon successful spawn, re/set the proc control state */
if (imgp->ip_px_sa != NULL) {
switch (px_sa.psa_pcontrol) {
}
exec_resettextvp(p, imgp);
-#if CONFIG_MEMORYSTATUS && CONFIG_JETSAM
+#if CONFIG_MEMORYSTATUS
/* Has jetsam attributes? */
if (imgp->ip_px_sa != NULL && (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_SET)) {
/*
* With 2-level high-water-mark support, POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND is no
* longer relevant, as background limits are described via the inactive limit slots.
- * 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));
+ px_sa.psa_memlimit_inactive, FALSE);
} else {
memorystatus_update(p, px_sa.psa_priority, 0,
(px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY),
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));
+ (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_MEMLIMIT_INACTIVE_FATAL));
}
}
-#endif /* CONFIG_MEMORYSTATUS && CONFIG_JETSAM*/
+#endif /* CONFIG_MEMORYSTATUS */
}
/*
/* 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 if (error == 0) {
- /* reset the importance attribute from our previous life */
- task_importance_reset(p->task);
+ }
+
+ /* flag exec has occurred, notify only if it has not failed due to FP Key error */
+ if (!error && ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0))
+ proc_knote(p, NOTE_EXEC);
+
+
+ if (error == 0) {
+ /*
+ * We need to initialize the bank context behind the protection of
+ * the proc_trans lock to prevent a race with exit. We can't do this during
+ * exec_activate_image because task_bank_init checks entitlements that
+ * aren't loaded until subsequent calls (including exec_resettextvp).
+ */
+ error = proc_transstart(p, 0, 0);
- /* reset atm context from task */
- task_atm_reset(p->task);
+ if (error == 0) {
+ task_bank_init(get_threadtask(imgp->ip_new_thread));
+ proc_transend(p, 0);
+ }
+ }
+
+ /* Inherit task role from old task to new task for exec */
+ if (error == 0 && !spawn_no_exec) {
+ proc_inherit_task_role(get_threadtask(imgp->ip_new_thread), current_task());
}
/*
portwatch_ports, portwatch_count);
}
- /* Apply the main thread qos */
+ /*
+ * Need to transfer pending watch port boosts to the new task while still making
+ * sure that the old task remains in the importance linkage. Create an importance
+ * linkage from old task to new task, then switch the task importance base
+ * of old task and new task. After the switch the port watch boost will be
+ * boosting the new task and new task will be donating importance to old task.
+ */
+ if (error == 0 && task_did_exec(current_task())) {
+ inherit = ipc_importance_exec_switch_task(current_task(), get_threadtask(imgp->ip_new_thread));
+ }
+
if (error == 0) {
- thread_t main_thread = (imgp->ip_new_thread != NULL) ? imgp->ip_new_thread : current_thread();
+ /* Apply the main thread qos */
+ thread_t main_thread = imgp->ip_new_thread;
+ task_set_main_thread_qos(get_threadtask(imgp->ip_new_thread), main_thread);
- task_set_main_thread_qos(p->task, main_thread);
+#if CONFIG_MACF
+ /*
+ * Processes with the MAP_JIT entitlement are permitted to have
+ * a jumbo-size map.
+ */
+ if (mac_proc_check_map_anon(p, 0, 0, 0, MAP_JIT, NULL) == 0) {
+ vm_map_set_jumbo(get_task_map(p->task));
+ }
+#endif /* CONFIG_MACF */
}
/*
if (imgp != NULL && spawn_no_exec && (p->p_lflag & P_LTRACED)) {
psignal_vfork(p, p->task, imgp->ip_new_thread, SIGTRAP);
}
+
+ if (error == 0 && !spawn_no_exec)
+ KDBG(BSDDBG_CODE(DBG_BSD_PROC,BSD_PROC_EXEC),
+ p->p_pid);
}
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);
mac_cred_label_free(imgp->ip_execlabelp);
if (imgp->ip_scriptlabelp)
mac_vnode_label_free(imgp->ip_scriptlabelp);
+ if (imgp->ip_cs_error != OS_REASON_NULL) {
+ os_reason_free(imgp->ip_cs_error);
+ imgp->ip_cs_error = OS_REASON_NULL;
+ }
#endif
}
if (error) {
DTRACE_PROC1(exec__failure, int, error);
} else {
- DTRACE_PROC(exec__success);
+ dtrace_thread_didexec(imgp->ip_new_thread);
}
}
- if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL)
+ if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
(*dtrace_proc_waitfor_hook)(p);
+ }
#endif
+ /*
+ * clear bsd_info from old task if it did exec.
+ */
+ if (task_did_exec(current_task())) {
+ set_bsdtask_info(current_task(), NULL);
+ }
+
+ /* clear bsd_info from new task and terminate it if exec failed */
+ if (new_task != NULL && task_is_exec_copy(new_task)) {
+ set_bsdtask_info(new_task, NULL);
+ task_terminate_internal(new_task);
+ }
/* Return to both the parent and the child? */
if (imgp != NULL && spawn_no_exec) {
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 {
-
- /*
- * 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.
- */
- proc_clear_return_wait(p, imgp->ip_new_thread);
}
}
+
+ /*
+ * Do not terminate the current task, if proc_exec_switch_task did not
+ * switch the tasks, terminating the current task without the switch would
+ * result in loosing the SIGKILL status.
+ */
+ if (task_did_exec(current_task())) {
+ /* Terminate the current task, since exec will start in new task */
+ task_terminate_internal(current_task());
+ }
+
+ /* Release the thread ref returned by fork_create_child/fork1 */
+ if (imgp != NULL && imgp->ip_new_thread) {
+ /* wake up the new thread */
+ task_clear_return_wait(get_threadtask(imgp->ip_new_thread));
+ thread_deallocate(imgp->ip_new_thread);
+ imgp->ip_new_thread = NULL;
+ }
+
+ /* Release the ref returned by fork_create_child/fork1 */
+ if (new_task) {
+ task_deallocate(new_task);
+ new_task = NULL;
+ }
+
+ if (should_release_proc_ref) {
+ proc_rele(p);
+ }
+
if (bufp != NULL) {
FREE(bufp, M_TEMP);
}
+
+ if (inherit != NULL) {
+ ipc_importance_release(inherit);
+ }
return(error);
}
+/*
+ * proc_exec_switch_task
+ *
+ * Parameters: p proc
+ * old_task task before exec
+ * new_task task after exec
+ * new_thread thread in new task
+ *
+ * Returns: proc.
+ *
+ * Note: The function will switch the task pointer of proc
+ * from old task to new task. The switch needs to happen
+ * after draining all proc refs and inside a proc translock.
+ * In the case of failure to switch the task, which might happen
+ * if the process received a SIGKILL or jetsam killed it, it will make
+ * sure that the new tasks terminates. User proc ref returned
+ * to caller.
+ *
+ * This function is called after point of no return, in the case
+ * failure to switch, it will terminate the new task and swallow the
+ * error and let the terminated process complete exec and die.
+ */
+proc_t
+proc_exec_switch_task(proc_t p, task_t old_task, task_t new_task, thread_t new_thread)
+{
+ int error = 0;
+ boolean_t task_active;
+ boolean_t proc_active;
+ boolean_t thread_active;
+ thread_t old_thread = current_thread();
+
+ /*
+ * Switch the task pointer of proc to new task.
+ * Before switching the task, wait for proc_refdrain.
+ * After the switch happens, the proc can disappear,
+ * take a ref before it disappears. Waiting for
+ * proc_refdrain in exec will block all other threads
+ * trying to take a proc ref, boost the current thread
+ * to avoid priority inversion.
+ */
+ thread_set_exec_promotion(old_thread);
+ p = proc_refdrain_with_refwait(p, TRUE);
+ /* extra proc ref returned to the caller */
+
+ assert(get_threadtask(new_thread) == new_task);
+ task_active = task_is_active(new_task);
+
+ /* Take the proc_translock to change the task ptr */
+ proc_lock(p);
+ proc_active = !(p->p_lflag & P_LEXIT);
+
+ /* Check if the current thread is not aborted due to SIGKILL */
+ thread_active = thread_is_active(old_thread);
+
+ /*
+ * Do not switch the task if the new task or proc is already terminated
+ * as a result of error in exec past point of no return
+ */
+ if (proc_active && task_active && thread_active) {
+ error = proc_transstart(p, 1, 0);
+ if (error == 0) {
+ uthread_t new_uthread = get_bsdthread_info(new_thread);
+ uthread_t old_uthread = get_bsdthread_info(current_thread());
+
+ /*
+ * bsd_info of old_task will get cleared in execve and posix_spawn
+ * after firing exec-success/error dtrace probe.
+ */
+ p->task = new_task;
+
+ /* Clear dispatchqueue and workloop ast offset */
+ p->p_dispatchqueue_offset = 0;
+ p->p_dispatchqueue_serialno_offset = 0;
+ p->p_return_to_kernel_offset = 0;
+
+ /* Copy the signal state, dtrace state and set bsd ast on new thread */
+ act_set_astbsd(new_thread);
+ new_uthread->uu_siglist = old_uthread->uu_siglist;
+ new_uthread->uu_sigwait = old_uthread->uu_sigwait;
+ new_uthread->uu_sigmask = old_uthread->uu_sigmask;
+ new_uthread->uu_oldmask = old_uthread->uu_oldmask;
+ new_uthread->uu_vforkmask = old_uthread->uu_vforkmask;
+ new_uthread->uu_exit_reason = old_uthread->uu_exit_reason;
+#if CONFIG_DTRACE
+ new_uthread->t_dtrace_sig = old_uthread->t_dtrace_sig;
+ new_uthread->t_dtrace_stop = old_uthread->t_dtrace_stop;
+ new_uthread->t_dtrace_resumepid = old_uthread->t_dtrace_resumepid;
+ assert(new_uthread->t_dtrace_scratch == NULL);
+ new_uthread->t_dtrace_scratch = old_uthread->t_dtrace_scratch;
+
+ old_uthread->t_dtrace_sig = 0;
+ old_uthread->t_dtrace_stop = 0;
+ old_uthread->t_dtrace_resumepid = 0;
+ old_uthread->t_dtrace_scratch = NULL;
+#endif
+ /* Copy the resource accounting info */
+ thread_copy_resource_info(new_thread, current_thread());
+
+ /* Clear the exit reason and signal state on old thread */
+ old_uthread->uu_exit_reason = NULL;
+ old_uthread->uu_siglist = 0;
+
+ /* Add the new uthread to proc uthlist and remove the old one */
+ TAILQ_INSERT_TAIL(&p->p_uthlist, new_uthread, uu_list);
+ TAILQ_REMOVE(&p->p_uthlist, old_uthread, uu_list);
+
+ task_set_did_exec_flag(old_task);
+ task_clear_exec_copy_flag(new_task);
+
+ task_copy_fields_for_exec(new_task, old_task);
+
+ proc_transend(p, 1);
+ }
+ }
+
+ proc_unlock(p);
+ proc_refwake(p);
+ thread_clear_exec_promotion(old_thread);
+
+ if (error != 0 || !task_active || !proc_active || !thread_active) {
+ task_terminate_internal(new_task);
+ }
+
+ return p;
+}
/*
* execve
int is_64 = IS_64BIT_PROCESS(p);
struct vfs_context context;
struct uthread *uthread;
+ task_t new_task = NULL;
+ boolean_t should_release_proc_ref = FALSE;
+ boolean_t exec_done = FALSE;
+ boolean_t in_vfexec = FALSE;
+ void *inherit = NULL;
context.vc_thread = current_thread();
context.vc_ucred = kauth_cred_proc_ref(p); /* XXX must NOT be kauth_cred_get() */
imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT : IMGPF_NONE) | ((p->p_flag & P_DISABLE_ASLR) ? IMGPF_DISABLE_ASLR : IMGPF_NONE);
imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
imgp->ip_mac_return = 0;
-
- uthread = get_bsdthread_info(current_thread());
- if (uthread->uu_flag & UT_VFORK) {
- imgp->ip_flags |= IMGPF_VFORK_EXEC;
- }
+ imgp->ip_cs_error = OS_REASON_NULL;
#if CONFIG_MACF
if (uap->mac_p != USER_ADDR_NULL) {
}
}
#endif
+ uthread = get_bsdthread_info(current_thread());
+ if (uthread->uu_flag & UT_VFORK) {
+ imgp->ip_flags |= IMGPF_VFORK_EXEC;
+ in_vfexec = TRUE;
+ } else {
+ imgp->ip_flags |= IMGPF_EXEC;
+
+ /*
+ * For execve case, create a new task and thread
+ * which points to current_proc. The current_proc will point
+ * to the new task after image activation and proc ref drain.
+ *
+ * proc (current_proc) <----- old_task (current_task)
+ * ^ | ^
+ * | | |
+ * | ----------------------------------
+ * |
+ * --------- new_task (task marked as TF_EXEC_COPY)
+ *
+ * After image activation, the proc will point to the new task
+ * and would look like following.
+ *
+ * proc (current_proc) <----- old_task (current_task, marked as TPF_DID_EXEC)
+ * ^ |
+ * | |
+ * | ----------> new_task
+ * | |
+ * -----------------
+ *
+ * During exec any transition from new_task -> proc is fine, but don't allow
+ * transition from proc->task, since it will modify old_task.
+ */
+ imgp->ip_new_thread = fork_create_child(current_task(),
+ NULL, p, FALSE, p->p_flag & P_LP64, TRUE);
+ /* task and thread ref returned by fork_create_child */
+ if (imgp->ip_new_thread == NULL) {
+ error = ENOMEM;
+ goto exit_with_error;
+ }
+
+ new_task = get_threadtask(imgp->ip_new_thread);
+ context.vc_thread = imgp->ip_new_thread;
+ }
error = exec_activate_image(imgp);
+ /* thread and task ref returned for vfexec case */
+
+ if (imgp->ip_new_thread != NULL) {
+ /*
+ * task reference might be returned by exec_activate_image
+ * for vfexec.
+ */
+ new_task = get_threadtask(imgp->ip_new_thread);
+ }
+
+ if (!error && !in_vfexec) {
+ p = proc_exec_switch_task(p, current_task(), new_task, imgp->ip_new_thread);
+ /* proc ref returned */
+ should_release_proc_ref = TRUE;
+ }
kauth_cred_unref(&context.vc_ucred);
if (error == -1)
error = ENOEXEC;
- if (error == 0) {
+ if (!error) {
+ exec_done = TRUE;
+ assert(imgp->ip_new_thread != NULL);
+
exec_resettextvp(p, imgp);
error = check_for_signature(p, imgp);
- }
+ }
+
+ /* flag exec has occurred, notify only if it has not failed due to FP Key error */
+ if (exec_done && ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0))
+ proc_knote(p, NOTE_EXEC);
+
if (imgp->ip_vp != NULLVP)
vnode_put(imgp->ip_vp);
if (imgp->ip_scriptvp != NULLVP)
if (imgp->ip_scriptlabelp)
mac_vnode_label_free(imgp->ip_scriptlabelp);
#endif
+ if (imgp->ip_cs_error != OS_REASON_NULL) {
+ os_reason_free(imgp->ip_cs_error);
+ imgp->ip_cs_error = OS_REASON_NULL;
+ }
+
if (!error) {
+ /*
+ * We need to initialize the bank context behind the protection of
+ * the proc_trans lock to prevent a race with exit. We can't do this during
+ * exec_activate_image because task_bank_init checks entitlements that
+ * aren't loaded until subsequent calls (including exec_resettextvp).
+ */
+ error = proc_transstart(p, 0, 0);
+ }
+
+ if (!error) {
+ task_bank_init(get_threadtask(imgp->ip_new_thread));
+ proc_transend(p, 0);
+
/* 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();
+ /* Inherit task role from old task to new task for exec */
+ if (!in_vfexec) {
+ proc_inherit_task_role(get_threadtask(imgp->ip_new_thread), current_task());
+ }
- task_set_main_thread_qos(p->task, main_thread);
+ thread_t main_thread = imgp->ip_new_thread;
- /* reset task importance */
- task_importance_reset(p->task);
+ task_set_main_thread_qos(new_task, main_thread);
- /* reset atm context from task */
- task_atm_reset(p->task);
+#if CONFIG_MACF
+ /*
+ * Processes with the MAP_JIT entitlement are permitted to have
+ * a jumbo-size map.
+ */
+ if (mac_proc_check_map_anon(p, 0, 0, 0, MAP_JIT, NULL) == 0) {
+ vm_map_set_jumbo(get_task_map(new_task));
+ }
+#endif /* CONFIG_MACF */
- DTRACE_PROC(exec__success);
#if CONFIG_DTRACE
+ dtrace_thread_didexec(imgp->ip_new_thread);
+
if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL)
(*dtrace_proc_waitfor_hook)(p);
#endif
- if (imgp->ip_flags & IMGPF_VFORK_EXEC) {
+ if (in_vfexec) {
vfork_return(p, retval, p->p_pid);
- proc_clear_return_wait(p, imgp->ip_new_thread);
}
} else {
DTRACE_PROC1(exec__failure, int, error);
}
exit_with_error:
+
+ /*
+ * clear bsd_info from old task if it did exec.
+ */
+ if (task_did_exec(current_task())) {
+ set_bsdtask_info(current_task(), NULL);
+ }
+
+ /* clear bsd_info from new task and terminate it if exec failed */
+ if (new_task != NULL && task_is_exec_copy(new_task)) {
+ set_bsdtask_info(new_task, NULL);
+ task_terminate_internal(new_task);
+ }
+
+ /*
+ * Need to transfer pending watch port boosts to the new task while still making
+ * sure that the old task remains in the importance linkage. Create an importance
+ * linkage from old task to new task, then switch the task importance base
+ * of old task and new task. After the switch the port watch boost will be
+ * boosting the new task and new task will be donating importance to old task.
+ */
+ if (error == 0 && task_did_exec(current_task())) {
+ inherit = ipc_importance_exec_switch_task(current_task(), get_threadtask(imgp->ip_new_thread));
+ }
+
+ if (imgp != NULL) {
+ /*
+ * Do not terminate the current task, if proc_exec_switch_task did not
+ * switch the tasks, terminating the current task without the switch would
+ * result in loosing the SIGKILL status.
+ */
+ if (task_did_exec(current_task())) {
+ /* Terminate the current task, since exec will start in new task */
+ task_terminate_internal(current_task());
+ }
+
+ /* Release the thread ref returned by fork_create_child */
+ if (imgp->ip_new_thread) {
+ /* wake up the new exec thread */
+ task_clear_return_wait(get_threadtask(imgp->ip_new_thread));
+ thread_deallocate(imgp->ip_new_thread);
+ imgp->ip_new_thread = NULL;
+ }
+ }
+
+ /* Release the ref returned by fork_create_child */
+ if (new_task) {
+ task_deallocate(new_task);
+ new_task = NULL;
+ }
+
+ if (should_release_proc_ref) {
+ proc_rele(p);
+ }
+
if (bufp != NULL) {
FREE(bufp, M_TEMP);
}
+
+ if (inherit != NULL) {
+ ipc_importance_release(inherit);
+ }
return(error);
}
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
#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
- */
+
+#define MAIN_STACK_VALUES 4
+#define MAIN_STACK_KEY "main_stack="
+
+#define FSID_KEY "executable_file="
+#define DYLD_FSID_KEY "dyld_file="
+#define CDHASH_KEY "executable_cdhash="
+
+#define FSID_MAX_STRING "0x1234567890abcdef,0x1234567890abcdef"
+
+#define HEX_STR_LEN 18 // 64-bit hex value "0x0123456701234567"
+
static int
-exec_add_apple_strings(struct image_params *imgp)
+exec_add_entropy_key(struct image_params *imgp,
+ const char *key,
+ int values,
+ boolean_t embedNUL)
{
- int i, error;
- int new_ptr_size=4;
- char guard[19];
- char guard_vec[strlen(GUARD_KEY) + 19 * GUARD_VALUES + 1];
+ const int limit = 8;
+ uint64_t entropy[limit];
+ char str[strlen(key) + (HEX_STR_LEN + 1) * limit + 1];
+ if (values > limit) {
+ values = limit;
+ }
- char entropy[19];
- char entropy_vec[strlen(ENTROPY_KEY) + 19 * ENTROPY_VALUES + 1];
+ read_random(entropy, sizeof(entropy[0]) * values);
- 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);
+ if (embedNUL) {
+ entropy[0] &= ~(0xffull << 8);
}
+ int len = snprintf(str, sizeof(str), "%s0x%llx", key, entropy[0]);
+ int remaining = sizeof(str) - len;
+ for (int i = 1; i < values && remaining > 0; ++i) {
+ int start = sizeof(str) - remaining;
+ len = snprintf(&str[start], remaining, ",0x%llx", entropy[i]);
+ remaining -= len;
+ }
+
+ return exec_add_user_string(imgp, CAST_USER_ADDR_T(str), UIO_SYSSPACE, FALSE);
+}
+
+/*
+ * Build up the contents of the apple[] string vector
+ */
+static int
+exec_add_apple_strings(struct image_params *imgp,
+ const load_result_t *load_result)
+{
+ int error;
+ int img_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT) ? 8 : 4;
+
/* 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++;
+ {
+ char pfz_string[strlen(PFZ_KEY) + HEX_STR_LEN + 1];
+
+ if (img_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);
+ }
+ 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)
+ const char *nano_string = NANO_ENGAGE_KEY;
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(nano_string), UIO_SYSSPACE, FALSE);
+ if (error){
goto bad;
+ }
imgp->ip_applec++;
}
}
* (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)
+ error = exec_add_entropy_key(imgp, GUARD_KEY, GUARD_VALUES, TRUE);
+ 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)
+ error = exec_add_entropy_key(imgp, ENTROPY_KEY, ENTROPY_VALUES, FALSE);
+ if (error) {
goto bad;
+ }
imgp->ip_applec++;
+ /*
+ * Add MAIN_STACK_KEY: Supplies the address and size of the main thread's
+ * stack if it was allocated by the kernel.
+ *
+ * The guard page is not included in this stack size as libpthread
+ * expects to add it back in after receiving this value.
+ */
+ if (load_result->unixproc) {
+ char stack_string[strlen(MAIN_STACK_KEY) + (HEX_STR_LEN + 1) * MAIN_STACK_VALUES + 1];
+ snprintf(stack_string, sizeof(stack_string),
+ MAIN_STACK_KEY "0x%llx,0x%llx,0x%llx,0x%llx",
+ (uint64_t)load_result->user_stack,
+ (uint64_t)load_result->user_stack_size,
+ (uint64_t)load_result->user_stack_alloc,
+ (uint64_t)load_result->user_stack_alloc_size);
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(stack_string), UIO_SYSSPACE, FALSE);
+ if (error) {
+ goto bad;
+ }
+ imgp->ip_applec++;
+ }
+
+ if (imgp->ip_vattr) {
+ uint64_t fsid = get_va_fsid(imgp->ip_vattr);
+ uint64_t fsobjid = imgp->ip_vattr->va_fileid;
+
+ char fsid_string[strlen(FSID_KEY) + strlen(FSID_MAX_STRING) + 1];
+ snprintf(fsid_string, sizeof(fsid_string),
+ FSID_KEY "0x%llx,0x%llx", fsid, fsobjid);
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(fsid_string), UIO_SYSSPACE, FALSE);
+ if (error) {
+ goto bad;
+ }
+ imgp->ip_applec++;
+ }
+
+ if (imgp->ip_dyld_fsid || imgp->ip_dyld_fsobjid ) {
+ char fsid_string[strlen(DYLD_FSID_KEY) + strlen(FSID_MAX_STRING) + 1];
+ snprintf(fsid_string, sizeof(fsid_string),
+ DYLD_FSID_KEY "0x%llx,0x%llx", imgp->ip_dyld_fsid, imgp->ip_dyld_fsobjid);
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(fsid_string), UIO_SYSSPACE, FALSE);
+ if (error) {
+ goto bad;
+ }
+ imgp->ip_applec++;
+ }
+
+ uint8_t cdhash[SHA1_RESULTLEN];
+ int cdhash_errror = ubc_cs_getcdhash(imgp->ip_vp, imgp->ip_arch_offset, cdhash);
+ if (cdhash_errror == 0) {
+ char hash_string[strlen(CDHASH_KEY) + 2*SHA1_RESULTLEN + 1];
+ strncpy(hash_string, CDHASH_KEY, sizeof(hash_string));
+ char *p = hash_string + sizeof(CDHASH_KEY) - 1;
+ for (int i = 0; i < SHA1_RESULTLEN; i++) {
+ snprintf(p, 3, "%02x", (int) cdhash[i]);
+ p += 2;
+ }
+ error = exec_add_user_string(imgp, CAST_USER_ADDR_T(hash_string), 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) {
+ while (imgp->ip_strspace % img_ptr_size != 0) {
*imgp->ip_strendp++ = '\0';
imgp->ip_strspace--;
}
VATTR_WANTED(vap, va_gid);
VATTR_WANTED(vap, va_mode);
VATTR_WANTED(vap, va_fsid);
+ VATTR_WANTED(vap, va_fsid64);
VATTR_WANTED(vap, va_fileid);
VATTR_WANTED(vap, va_data_size);
if ((error = vnode_getattr(vp, vap, imgp->ip_vfs_context)) != 0)
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);
+ kauth_cred_t cred = vfs_context_ucred(imgp->ip_vfs_context);
+ kauth_cred_t my_cred, my_new_cred;
int i;
int leave_sugid_clear = 0;
int mac_reset_ipc = 0;
int error = 0;
+ task_t task = NULL;
#if CONFIG_MACF
int mac_transition, disjoint_cred = 0;
int label_update_return = 0;
handle_mac_transition:
#endif
+#if !SECURE_KERNEL
/*
* Replace the credential with a copy of itself if euid or
* egid change.
* membership resolution, then dropping their
* effective privilege to that of the desired
* final credential state.
+ *
+ * Modifications to p_ucred must be guarded using the
+ * proc's ucred lock. This prevents others from accessing
+ * a garbage credential.
*/
- 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);
+ while (imgp->ip_origvattr->va_mode & VSUID) {
+ my_cred = kauth_cred_proc_ref(p);
+ my_new_cred = kauth_cred_setresuid(my_cred, KAUTH_UID_NONE, imgp->ip_origvattr->va_uid, imgp->ip_origvattr->va_uid, KAUTH_UID_NONE);
+
+ if (my_new_cred == my_cred) {
+ kauth_cred_unref(&my_cred);
+ break;
+ }
+
/* update cred on proc */
+ proc_ucred_lock(p);
+
+ if (p->p_ucred != my_cred) {
+ proc_ucred_unlock(p);
+ kauth_cred_unref(&my_new_cred);
+ continue;
+ }
+
+ /* donate cred reference on my_new_cred to p->p_ucred */
+ p->p_ucred = my_new_cred;
PROC_UPDATE_CREDS_ONPROC(p);
+ proc_ucred_unlock(p);
+
+ /* drop additional reference that was taken on the previous cred */
+ kauth_cred_unref(&my_cred);
+
+ break;
}
- 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);
+
+ while (imgp->ip_origvattr->va_mode & VSGID) {
+ my_cred = kauth_cred_proc_ref(p);
+ my_new_cred = kauth_cred_setresgid(my_cred, KAUTH_GID_NONE, imgp->ip_origvattr->va_gid, imgp->ip_origvattr->va_gid);
+
+ if (my_new_cred == my_cred) {
+ kauth_cred_unref(&my_cred);
+ break;
+ }
+
/* update cred on proc */
+ proc_ucred_lock(p);
+
+ if (p->p_ucred != my_cred) {
+ proc_ucred_unlock(p);
+ kauth_cred_unref(&my_new_cred);
+ continue;
+ }
+
+ /* donate cred reference on my_new_cred to p->p_ucred */
+ p->p_ucred = my_new_cred;
PROC_UPDATE_CREDS_ONPROC(p);
+ proc_ucred_unlock(p);
+
+ /* drop additional reference that was taken on the previous cred */
+ kauth_cred_unref(&my_cred);
+
+ break;
}
+#endif /* !SECURE_KERNEL */
#if CONFIG_MACF
/*
* 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) {
/*
* a setuid exec to be able to access/control the
* task/thread after.
*/
- ipc_task_reset(p->task);
+ ipc_task_reset((imgp->ip_new_thread != NULL) ?
+ get_threadtask(imgp->ip_new_thread) : p->task);
ipc_thread_reset((imgp->ip_new_thread != NULL) ?
imgp->ip_new_thread : current_thread());
}
MALLOC(ndp, struct nameidata *, sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO);
if (ndp == NULL) {
+ fp_free(p, indx, fp);
error = ENOMEM;
break;
}
if ((error = vn_open(ndp, flag, 0)) != 0) {
fp_free(p, indx, fp);
+ FREE(ndp, M_TEMP);
break;
}
/*
* Implement the semantic where the effective user and group become
* the saved user and group in exec'ed programs.
+ *
+ * Modifications to p_ucred must be guarded using the
+ * proc's ucred lock. This prevents others from accessing
+ * a garbage credential.
*/
- 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);
-
+ for (;;) {
+ my_cred = kauth_cred_proc_ref(p);
+ my_new_cred = kauth_cred_setsvuidgid(my_cred, kauth_cred_getuid(my_cred), kauth_cred_getgid(my_cred));
+
+ if (my_new_cred == my_cred) {
+ kauth_cred_unref(&my_cred);
+ break;
+ }
+
+ /* update cred on proc */
+ proc_ucred_lock(p);
+
+ if (p->p_ucred != my_cred) {
+ proc_ucred_unlock(p);
+ kauth_cred_unref(&my_new_cred);
+ continue;
+ }
+
+ /* donate cred reference on my_new_cred to p->p_ucred */
+ p->p_ucred = my_new_cred;
+ PROC_UPDATE_CREDS_ONPROC(p);
+ proc_ucred_unlock(p);
+
+ /* drop additional reference that was taken on the previous cred */
+ kauth_cred_unref(&my_cred);
+
+ break;
+ }
+
+
/* Update the process' identity version and set the security token */
p->p_idversion++;
- set_security_token(p);
+
+ if (imgp->ip_new_thread != NULL) {
+ task = get_threadtask(imgp->ip_new_thread);
+ } else {
+ task = p->task;
+ }
+ set_security_token_task_internal(p, task);
return(error);
}
p->user_stack = user_stack;
proc_unlock(p);
- if (!load_result->prog_allocated_stack) {
+ if (load_result->user_stack_alloc_size > 0) {
/*
* Allocate enough space for the maximum stack size we
* will ever authorize and an extra page to act as
* vm_initial_limit_stack takes care of the extra guard page.
* Otherwise we must allocate it ourselves.
*/
-
- size = mach_vm_round_page(load_result->user_stack_size);
- if (load_result->prog_stack_size)
- size += PAGE_SIZE;
+ if (mach_vm_round_page_overflow(load_result->user_stack_alloc_size, &size)) {
+ return KERN_INVALID_ARGUMENT;
+ }
addr = mach_vm_trunc_page(load_result->user_stack - size);
- kr = mach_vm_allocate(map, &addr, size,
- VM_MAKE_TAG(VM_MEMORY_STACK) |
- VM_FLAGS_FIXED);
+ kr = mach_vm_allocate_kernel(map, &addr, size,
+ VM_FLAGS_FIXED, VM_MEMORY_STACK);
if (kr != KERN_SUCCESS) {
- /* If can't allocate at default location, try anywhere */
+ // Can't allocate at default location, try anywhere
addr = 0;
- kr = mach_vm_allocate(map, &addr, size,
- VM_MAKE_TAG(VM_MEMORY_STACK) |
- VM_FLAGS_ANYWHERE);
- if (kr != KERN_SUCCESS)
+ kr = mach_vm_allocate_kernel(map, &addr, size,
+ VM_FLAGS_ANYWHERE, VM_MEMORY_STACK);
+ if (kr != KERN_SUCCESS) {
return kr;
+ }
user_stack = addr + size;
load_result->user_stack = user_stack;
proc_unlock(p);
}
+ load_result->user_stack_alloc = addr;
+
/*
* And prevent access to what's above the current stack
* size limit for this process.
*/
- prot_addr = addr;
- if (load_result->prog_stack_size)
+ if (load_result->user_stack_size == 0) {
+ load_result->user_stack_size = unix_stack_size(p);
+ prot_size = mach_vm_trunc_page(size - load_result->user_stack_size);
+ } else {
prot_size = PAGE_SIZE;
- else
- prot_size = mach_vm_trunc_page(size - unix_stack_size(p));
+ }
+
+ prot_addr = addr;
kr = mach_vm_protect(map,
- prot_addr,
- prot_size,
- FALSE,
- VM_PROT_NONE);
+ prot_addr,
+ prot_size,
+ FALSE,
+ VM_PROT_NONE);
if (kr != KERN_SUCCESS) {
- (void) mach_vm_deallocate(map, addr, size);
+ (void)mach_vm_deallocate(map, addr, size);
return kr;
}
}
* 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.
+ * The address map of the first manufactured process matches the
+ * word width of the kernel. Once the self-exec completes, the
+ * initproc might be different.
*/
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];
+ int error;
struct execve_args init_exec_args;
+ user_addr_t argv0 = USER_ADDR_NULL, argv1 = USER_ADDR_NULL;
/*
* Validate inputs and pre-conditions
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);
+ argv0 = scratch_addr;
+ error = copyout(path, argv0, path_length);
+ if (error)
+ return error;
- argv[argc++] = (uint32_t)scratch_addr;
- scratch_addr = USER_ADDR_ALIGN(scratch_addr + path_length, 16);
+ scratch_addr = USER_ADDR_ALIGN(scratch_addr + path_length, sizeof(user_addr_t));
/*
* Put out first (and only) argument, similarly.
const char *init_args = "-s";
size_t init_args_length = strlen(init_args)+1;
- copyout(init_args, scratch_addr, init_args_length);
+ argv1 = scratch_addr;
+ error = copyout(init_args, argv1, init_args_length);
+ if (error)
+ return error;
- argv[argc++] = (uint32_t)scratch_addr;
- scratch_addr = USER_ADDR_ALIGN(scratch_addr + init_args_length, 16);
+ scratch_addr = USER_ADDR_ALIGN(scratch_addr + init_args_length, sizeof(user_addr_t));
}
- /*
- * Null-end the argument list
- */
- argv[argc] = 0;
-
- /*
- * Copy out the argument list.
- */
- (void) copyout(argv, scratch_addr, sizeof(argv));
+ if (proc_is64bit(p)) {
+ user64_addr_t argv64bit[3];
+
+ argv64bit[0] = argv0;
+ argv64bit[1] = argv1;
+ argv64bit[2] = USER_ADDR_NULL;
+
+ error = copyout(argv64bit, scratch_addr, sizeof(argv64bit));
+ if (error)
+ return error;
+ } else {
+ user32_addr_t argv32bit[3];
+
+ argv32bit[0] = (user32_addr_t)argv0;
+ argv32bit[1] = (user32_addr_t)argv1;
+ argv32bit[2] = USER_ADDR_NULL;
+
+ error = copyout(argv32bit, scratch_addr, sizeof(argv32bit));
+ if (error)
+ return error;
+ }
/*
* Set up argument block for fake call to execve.
*/
- init_exec_args.fname = CAST_USER_ADDR_T(argv[0]);
+ init_exec_args.fname = argv0;
init_exec_args.argp = scratch_addr;
init_exec_args.envp = USER_ADDR_NULL;
"/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",
* 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
{
uint32_t i;
int error;
- vm_offset_t scratch_addr = VM_MIN_ADDRESS;
+ vm_map_t map = current_map();
+ mach_vm_offset_t scratch_addr = 0;
+ mach_vm_size_t map_page_size = vm_map_page_size(map);
- (void) vm_allocate(current_map(), &scratch_addr, PAGE_SIZE, VM_FLAGS_ANYWHERE);
-#if CONFIG_MEMORYSTATUS && CONFIG_JETSAM
+ (void) mach_vm_allocate_kernel(map, &scratch_addr, map_page_size, VM_FLAGS_ANYWHERE, VM_KERN_MEMORY_NONE);
+#if CONFIG_MEMORYSTATUS
(void) memorystatus_init_at_boot_snapshot();
-#endif /* CONFIG_MEMORYSTATUS && CONFIG_JETSAM */
+#endif /* CONFIG_MEMORYSTATUS */
#if DEBUG || DEVELOPMENT
/* Check for boot-arg suffix first */
(strcmp(launchd_suffix, "release") == 0));
if (is_release_suffix) {
- error = load_init_program_at_path(p, CAST_USER_ADDR_T(scratch_addr), "/sbin/launchd");
+ printf("load_init_program: attempting to load /sbin/launchd\n");
+ error = load_init_program_at_path(p, (user_addr_t)scratch_addr, "/sbin/launchd");
if (!error)
return;
strlcpy(launchd_path, "/usr/local/sbin/launchd.", sizeof(launchd_path));
strlcat(launchd_path, launchd_suffix, sizeof(launchd_path));
- /* 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)) {
+ printf("load_init_program: attempting to load %s\n", launchd_path);
+ error = load_init_program_at_path(p, (user_addr_t)scratch_addr, launchd_path);
+ if (!error) {
return;
+ } else {
+ printf("load_init_program: failed loading %s: errno %d\n", launchd_path, error);
}
}
}
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)
+ printf("load_init_program: attempting to load %s\n", init_programs[i]);
+ error = load_init_program_at_path(p, (user_addr_t)scratch_addr, init_programs[i]);
+ if (!error) {
return;
+ } else {
+ printf("load_init_program: failed loading %s: errno %d\n", init_programs[i], error);
+ }
}
panic("Process 1 exec of %s failed, errno %d", ((i == 0) ? "<null>" : init_programs[i-1]), error);
* EIO An I/O error occurred
* EBADEXEC The executable is corrupt/unknown
*/
-static int
+static int
load_return_to_errno(load_return_t lrtn)
{
switch (lrtn) {
case LOAD_SUCCESS:
return 0;
case LOAD_BADARCH:
+ case LOAD_BADARCH_X86:
return EBADARCH;
case LOAD_BADMACHO:
+ case LOAD_BADMACHO_UPX:
return EBADMACHO;
case LOAD_SHLIB:
return ESHLIBVERS;
static kern_return_t
execargs_purgeable_allocate(char **execarg_address) {
- kern_return_t kr = vm_allocate(bsd_pageable_map, (vm_offset_t *)execarg_address, BSD_PAGEABLE_SIZE_PER_EXEC, VM_FLAGS_ANYWHERE | VM_FLAGS_PURGABLE);
+ kern_return_t kr = vm_allocate_kernel(bsd_pageable_map, (vm_offset_t *)execarg_address, BSD_PAGEABLE_SIZE_PER_EXEC, VM_FLAGS_ANYWHERE | VM_FLAGS_PURGABLE, VM_KERN_MEMORY_NONE);
assert(kr == KERN_SUCCESS);
return kr;
}
}
-/*
- * If the process is not signed or if it contains entitlements, we
- * need to communicate through the task_access_port to taskgated.
- *
- * taskgated will provide a detached code signature if present, and
- * will enforce any restrictions on entitlements.
- */
+// Includes the 0-byte (therefore "SIZE" instead of "LEN").
+static const size_t CS_CDHASH_STRING_SIZE = CS_CDHASH_LEN * 2 + 1;
-static boolean_t
-taskgated_required(proc_t p, boolean_t *require_success)
-{
- size_t length;
- 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;
- }
-
- error = cs_entitlements_blob_get(p, &blob, &length);
- if (error == 0 && blob != NULL) {
- /*
- * fatal on the desktop when entitlements are present,
- * unless we started in single-user mode
- */
- if ((boothowto & RB_SINGLE) == 0)
- *require_success = TRUE;
- /*
- * Allow initproc to run without causing taskgated to launch
- */
- if (p == initproc) {
- *require_success = FALSE;
- return FALSE;
- }
+static void cdhash_to_string(char str[CS_CDHASH_STRING_SIZE], uint8_t const * const cdhash) {
+ static char const nibble[] = "0123456789abcdef";
- if (cs_debug) printf("taskgated required for: %s\n", p->p_name);
-
- return TRUE;
- }
-
- *require_success = FALSE;
- return FALSE;
+ /* Apparently still the safest way to get a hex representation
+ * of binary data.
+ * xnu's printf routines have %*D/%20D in theory, but "not really", see:
+ * <rdar://problem/33328859> confusion around %*D/%nD in printf
+ */
+ for (int i = 0; i < CS_CDHASH_LEN; ++i) {
+ str[i*2] = nibble[(cdhash[i] & 0xf0) >> 4];
+ str[i*2+1] = nibble[cdhash[i] & 0x0f];
+ }
+ str[CS_CDHASH_STRING_SIZE - 1] = 0;
}
/*
kern_return_t kr = KERN_FAILURE;
int error = EACCES;
boolean_t unexpected_failure = FALSE;
- unsigned char hash[SHA1_RESULTLEN];
+ struct cs_blob *csb;
boolean_t require_success = FALSE;
int spawn = (imgp->ip_flags & IMGPF_SPAWN);
int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC);
+ os_reason_t signature_failure_reason = OS_REASON_NULL;
/*
* Override inherited code signing flags with the
if(p->p_csflags & (CS_HARD|CS_KILL)) {
vm_map_switch_protect(get_task_map(p->task), TRUE);
}
-
+
/*
* image activation may be failed due to policy
* which is unexpected but security framework does not
* approve of exec, kill and return immediately.
*/
if (imgp->ip_mac_return != 0) {
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_SECURITY_POLICY, 0, 0);
+ signature_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_SECURITY_POLICY);
error = imgp->ip_mac_return;
unexpected_failure = TRUE;
goto done;
}
- /* check if callout to taskgated is needed */
- if (!taskgated_required(p, &require_success)) {
+ if (imgp->ip_cs_error != OS_REASON_NULL) {
+ signature_failure_reason = imgp->ip_cs_error;
+ imgp->ip_cs_error = OS_REASON_NULL;
+ error = EACCES;
+ goto done;
+ }
+
+ /* If the code signature came through the image activation path, we skip the
+ * taskgated / externally attached path. */
+ if (imgp->ip_csflags & CS_SIGNED) {
error = 0;
goto done;
}
+ /* The rest of the code is for signatures that either already have been externally
+ * attached (likely, but not necessarily by a previous run through the taskgated
+ * path), or that will now be attached by taskgated. */
+
kr = task_get_task_access_port(p->task, &port);
if (KERN_SUCCESS != kr || !IPC_PORT_VALID(port)) {
error = 0;
- if (require_success)
+ if (require_success) {
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASK_ACCESS_PORT, 0, 0);
+ signature_failure_reason = os_reason_create(OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASK_ACCESS_PORT);
error = EACCES;
+ }
goto done;
}
break;
case KERN_FAILURE:
error = EACCES;
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASKGATED_INVALID_SIG, 0, 0);
+ signature_failure_reason = os_reason_create(OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASKGATED_INVALID_SIG);
goto done;
default:
error = EACCES;
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+ p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_TASKGATED_OTHER, 0, 0);
+ signature_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_TASKGATED_OTHER);
unexpected_failure = TRUE;
goto done;
}
/* Only do this if exec_resettextvp() did not fail */
if (p->p_textvp != NULLVP) {
- /*
- * If there's a new code directory, mark this process
- * as signed.
- */
- if (0 == ubc_cs_getcdhash(p->p_textvp, p->p_textoff, hash)) {
- proc_lock(p);
- p->p_csflags |= CS_VALID;
- proc_unlock(p);
+ csb = ubc_cs_blob_get(p->p_textvp, -1, p->p_textoff);
+
+ if (csb != NULL) {
+ /* As the enforcement we can do here is very limited, we only allow things that
+ * are the only reason why this code path still exists:
+ * Adhoc signed non-platform binaries without special cs_flags and without any
+ * entitlements (unrestricted ones still pass AMFI). */
+ if (
+ /* Revalidate the blob if necessary through bumped generation count. */
+ (ubc_cs_generation_check(p->p_textvp) == 0 ||
+ ubc_cs_blob_revalidate(p->p_textvp, csb, imgp, 0) == 0) &&
+ /* Only CS_ADHOC, no CS_KILL, CS_HARD etc. */
+ (csb->csb_flags & CS_ALLOWED_MACHO) == CS_ADHOC &&
+ /* If it has a CMS blob, it's not adhoc. The CS_ADHOC flag can lie. */
+ csblob_find_blob_bytes((const uint8_t *)csb->csb_mem_kaddr, csb->csb_mem_size,
+ CSSLOT_SIGNATURESLOT,
+ CSMAGIC_BLOBWRAPPER) == NULL &&
+ /* It could still be in a trust cache (unlikely with CS_ADHOC), or a magic path. */
+ csb->csb_platform_binary == 0 &&
+ /* No entitlements, not even unrestricted ones. */
+ csb->csb_entitlements_blob == NULL) {
+
+ proc_lock(p);
+ p->p_csflags |= CS_SIGNED | CS_VALID;
+ proc_unlock(p);
+
+ } else {
+ uint8_t cdhash[CS_CDHASH_LEN];
+ char cdhash_string[CS_CDHASH_STRING_SIZE];
+ proc_getcdhash(p, cdhash);
+ cdhash_to_string(cdhash_string, cdhash);
+ printf("ignoring detached code signature on '%s' with cdhash '%s' "
+ "because it is invalid, or not a simple adhoc signature.\n",
+ p->p_name, cdhash_string);
+ }
+
}
}
done:
- if (0 != error) {
+ if (0 == error) {
+ /* The process's code signature related properties are
+ * fully set up, so this is an opportune moment to log
+ * platform binary execution, if desired. */
+ if (platform_exec_logging != 0 && csproc_get_platform_binary(p)) {
+ uint8_t cdhash[CS_CDHASH_LEN];
+ char cdhash_string[CS_CDHASH_STRING_SIZE];
+ proc_getcdhash(p, cdhash);
+ cdhash_to_string(cdhash_string, cdhash);
+
+ os_log(peLog, "CS Platform Exec Logging: Executing platform signed binary "
+ "'%s' with cdhash %s\n", p->p_name, cdhash_string);
+ }
+ } else {
if (!unexpected_failure)
p->p_csflags |= CS_KILLED;
/* make very sure execution fails */
if (vfexec || spawn) {
- psignal_vfork(p, p->task, imgp->ip_new_thread, SIGKILL);
+ assert(signature_failure_reason != OS_REASON_NULL);
+ psignal_vfork_with_reason(p, p->task, imgp->ip_new_thread,
+ SIGKILL, signature_failure_reason);
+ signature_failure_reason = OS_REASON_NULL;
error = 0;
} else {
- psignal(p, SIGKILL);
+ assert(signature_failure_reason != OS_REASON_NULL);
+ psignal_with_reason(p, SIGKILL, signature_failure_reason);
+ signature_failure_reason = OS_REASON_NULL;
}
}
+
+ /* If we hit this, we likely would have leaked an exit reason */
+ assert(signature_failure_reason == OS_REASON_NULL);
return error;
}
vm_map_trunc_page(load_result->entry_point,
vm_map_page_mask(current_map())),
VM_PROT_READ | VM_PROT_EXECUTE,
- FALSE,
+ FALSE, VM_KERN_MEMORY_NONE,
THREAD_UNINT, NULL, 0);
if (imgp->ip_flags & IMGPF_IS_64BIT) {
vm_map_trunc_page(load_result->all_image_info_addr,
vm_map_page_mask(current_map())),
VM_PROT_READ | VM_PROT_WRITE,
- FALSE,
+ FALSE, VM_KERN_MEMORY_NONE,
THREAD_UNINT, NULL, 0);
if ((load_result->all_image_info_addr & PAGE_MASK) + expected_all_image_infos_size > PAGE_SIZE) {
/* all_image_infos straddles a page */
vm_map_trunc_page(load_result->all_image_info_addr + expected_all_image_infos_size - 1,
vm_map_page_mask(current_map())),
VM_PROT_READ | VM_PROT_WRITE,
- FALSE,
+ FALSE, VM_KERN_MEMORY_NONE,
THREAD_UNINT, NULL, 0);
}
ret = copyin(load_result->all_image_info_addr,
&all_image_infos,
expected_all_image_infos_size);
- if (ret == 0 && all_image_infos.infos32.version >= 9) {
+ if (ret == 0 && all_image_infos.infos32.version >= DYLD_ALL_IMAGE_INFOS_ADDRESS_MINIMUM_VERSION) {
user_addr_t notification_address;
user_addr_t dyld_image_address;
vm_map_trunc_page(notification_address + dyld_slide_amount,
vm_map_page_mask(current_map())),
VM_PROT_READ | VM_PROT_EXECUTE,
- FALSE,
+ FALSE, VM_KERN_MEMORY_NONE,
THREAD_UNINT, NULL, 0);
vm_fault(current_map(),
vm_map_trunc_page(dyld_image_address + dyld_slide_amount,
vm_map_page_mask(current_map())),
VM_PROT_READ | VM_PROT_EXECUTE,
- FALSE,
+ FALSE, VM_KERN_MEMORY_NONE,
THREAD_UNINT, NULL, 0);
vm_fault(current_map(),
vm_map_trunc_page(dyld_version_address + dyld_slide_amount,
vm_map_page_mask(current_map())),
VM_PROT_READ,
- FALSE,
+ FALSE, VM_KERN_MEMORY_NONE,
THREAD_UNINT, NULL, 0);
vm_fault(current_map(),
vm_map_trunc_page(dyld_all_image_infos_address + dyld_slide_amount,
vm_map_page_mask(current_map())),
VM_PROT_READ | VM_PROT_WRITE,
- FALSE,
+ FALSE, VM_KERN_MEMORY_NONE,
THREAD_UNINT, NULL, 0);
}
}
projects / apple / xnu.git / blobdiff