]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/bsd_kern.c
xnu-3789.51.2.tar.gz
[apple/xnu.git] / osfmk / kern / bsd_kern.c
index 4f6aa60aff69faf627614a7223ec7cc5ff7d662f..e9e601433aa796ddadef5dcd832a838968f5c0f8 100644 (file)
@@ -1,16 +1,19 @@
 /*
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
  *
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 #include <mach/mach_types.h>
  */
 #include <mach/mach_types.h>
-#include <kern/queue.h>
-#include <kern/ast.h>
+#include <mach/machine/vm_param.h>
+#include <mach/task.h>
+
+#include <kern/kern_types.h>
+#include <kern/ledger.h>
+#include <kern/processor.h>
 #include <kern/thread.h>
 #include <kern/thread.h>
-#include <kern/thread_act.h>
 #include <kern/task.h>
 #include <kern/spl.h>
 #include <kern/task.h>
 #include <kern/spl.h>
-#include <kern/lock.h>
-#include <vm/vm_map.h>
-#include <vm/pmap.h>
+#include <kern/ast.h>
 #include <ipc/ipc_port.h>
 #include <ipc/ipc_object.h>
 #include <ipc/ipc_port.h>
 #include <ipc/ipc_object.h>
+#include <vm/vm_map.h>
+#include <vm/vm_kern.h>
+#include <vm/pmap.h>
+#include <vm/vm_protos.h> /* last */
+#include <sys/resource.h>
+#include <sys/signal.h>
 
 #undef thread_should_halt
 
 #undef thread_should_halt
-#undef ipc_port_release
-
-decl_simple_lock_data(extern,reaper_lock)
-extern queue_head_t           reaper_queue;
 
 /* BSD KERN COMPONENT INTERFACE */
 
 
 /* BSD KERN COMPONENT INTERFACE */
 
-task_t bsd_init_task = TASK_NULL;
-char   init_task_failure_data[1024];
+extern unsigned int not_in_kdp; /* Skip acquiring locks if we're in kdp */
  
  
-thread_act_t get_firstthread(task_t);
-vm_map_t  get_task_map(task_t);
-ipc_space_t  get_task_ipcspace(task_t);
-boolean_t is_kerneltask(task_t);
-boolean_t is_thread_idle(thread_t);
-boolean_t is_thread_running(thread_act_t);
-thread_shuttle_t getshuttle_thread( thread_act_t);
-thread_act_t getact_thread( thread_shuttle_t);
-vm_offset_t get_map_min( vm_map_t);
-vm_offset_t get_map_max( vm_map_t);
+thread_t get_firstthread(task_t);
 int get_task_userstop(task_t);
 int get_task_userstop(task_t);
-int get_thread_userstop(thread_act_t);
-boolean_t thread_should_abort(thread_shuttle_t);
+int get_thread_userstop(thread_t);
 boolean_t current_thread_aborted(void);
 boolean_t current_thread_aborted(void);
-void task_act_iterate_wth_args(task_t, void(*)(thread_act_t, void *), void *);
-void ipc_port_release(ipc_port_t);
-boolean_t is_thread_active(thread_t);
-kern_return_t get_thread_waitresult(thread_t);
-vm_size_t get_vmmap_size(vm_map_t);
-int get_vmmap_entries(vm_map_t);
-int  get_task_numacts(task_t);
-thread_act_t get_firstthread(task_t task);
-kern_return_t get_signalact(task_t , thread_act_t *, thread_t *, int);
-void astbsd_on(void);
+void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
+kern_return_t get_signalact(task_t , thread_t *, int);
+int fill_task_rusage(task_t task, rusage_info_current *ri);
+int fill_task_io_rusage(task_t task, rusage_info_current *ri);
+int fill_task_qos_rusage(task_t task, rusage_info_current *ri);
+void fill_task_billed_usage(task_t task, rusage_info_current *ri);
+void task_bsdtask_kill(task_t);
+
+extern uint64_t get_dispatchqueue_serialno_offset_from_proc(void *p);
+extern uint64_t proc_uniqueid(void *p);
+
+#if MACH_BSD
+extern void psignal(void *, int);
+#endif
 
 /*
  *
 
 /*
  *
@@ -79,6 +78,21 @@ void  *get_bsdtask_info(task_t t)
        return(t->bsd_info);
 }
 
        return(t->bsd_info);
 }
 
+void task_bsdtask_kill(task_t t)
+{
+       void * bsd_info = get_bsdtask_info(t);
+       if (bsd_info != NULL) {
+               psignal(bsd_info, SIGKILL);
+       }
+}
+/*
+ *
+ */
+void *get_bsdthreadtask_info(thread_t th)
+{
+       return(th->task != TASK_NULL ? th->task->bsd_info : NULL);
+}
+
 /*
  *
  */
 /*
  *
  */
@@ -90,137 +104,168 @@ void set_bsdtask_info(task_t t,void * v)
 /*
  *
  */
 /*
  *
  */
-void *get_bsdthread_info(thread_act_t th)
+void *get_bsdthread_info(thread_t th)
 {
        return(th->uthread);
 }
 
 {
        return(th->uthread);
 }
 
+/*
+ * XXX
+ */
+int get_thread_lock_count(thread_t th);                /* forced forward */
+int get_thread_lock_count(thread_t th)
+{
+       return(th->mutex_count);
+}
+
 /*
  * XXX: wait for BSD to  fix signal code
  * Until then, we cannot block here.  We know the task
  * can't go away, so we make sure it is still active after
  * retrieving the first thread for extra safety.
  */
 /*
  * XXX: wait for BSD to  fix signal code
  * Until then, we cannot block here.  We know the task
  * can't go away, so we make sure it is still active after
  * retrieving the first thread for extra safety.
  */
-thread_act_t get_firstthread(task_t task)
+thread_t get_firstthread(task_t task)
 {
 {
-       thread_act_t    thr_act;
+       thread_t        thread = (thread_t)(void *)queue_first(&task->threads);
+
+       if (queue_end(&task->threads, (queue_entry_t)thread))
+               thread = THREAD_NULL;
 
 
-       thr_act = (thread_act_t)queue_first(&task->thr_acts);
-       if (thr_act == (thread_act_t)&task->thr_acts)
-               thr_act = THR_ACT_NULL;
        if (!task->active)
        if (!task->active)
-               return(THR_ACT_NULL);
-       return(thr_act);
+               return (THREAD_NULL);
+
+       return (thread);
 }
 
 }
 
-kern_return_t get_signalact(task_t task,thread_act_t * thact, thread_t * thshut, int setast)
+kern_return_t
+get_signalact(
+       task_t          task,
+       thread_t        *result_out,
+       int                     setast)
 {
 {
-
-        thread_act_t inc;
-        thread_act_t ninc;
-        thread_act_t thr_act;
-       thread_t        th;
+       kern_return_t   result = KERN_SUCCESS;
+       thread_t                inc, thread = THREAD_NULL;
 
        task_lock(task);
 
        task_lock(task);
+
        if (!task->active) {
                task_unlock(task);
        if (!task->active) {
                task_unlock(task);
-               return(KERN_FAILURE);
-       }
-
-        thr_act = THR_ACT_NULL;
-        for (inc  = (thread_act_t)queue_first(&task->thr_acts);
-             inc != (thread_act_t)&task->thr_acts;
-             inc  = ninc) {
-                th = act_lock_thread(inc);
-                if ((inc->active)  && 
-                    ((th->state & (TH_ABORT|TH_ABORT_SAFELY)) != TH_ABORT)) {
-                    thr_act = inc;
-                   break;
-                }
-                act_unlock_thread(inc);
-                ninc = (thread_act_t)queue_next(&inc->thr_acts);
-        }
-out:
-        if (thact) 
-                *thact = thr_act;
 
 
-        if (thshut)
-                *thshut = thr_act? thr_act->thread: THREAD_NULL ;
-        if (thr_act) {
-                if (setast)
-                    act_set_astbsd(thr_act);
+               return (KERN_FAILURE);
+       }
+
+       for (inc  = (thread_t)(void *)queue_first(&task->threads);
+                       !queue_end(&task->threads, (queue_entry_t)inc); ) {
+               thread_mtx_lock(inc);
+               if (inc->active &&
+                               (inc->sched_flags & TH_SFLAG_ABORTED_MASK) != TH_SFLAG_ABORT) {
+                       thread = inc;
+                       break;
+               }
+               thread_mtx_unlock(inc);
+
+               inc = (thread_t)(void *)queue_next(&inc->task_threads);
+       }
+
+       if (result_out) 
+               *result_out = thread;
+
+       if (thread) {
+               if (setast)
+                       act_set_astbsd(thread);
+
+               thread_mtx_unlock(thread);
+       }
+       else
+               result = KERN_FAILURE;
 
 
-                act_unlock_thread(thr_act);
-        }
        task_unlock(task);
 
        task_unlock(task);
 
-        if (thr_act) 
-            return(KERN_SUCCESS);
-        else 
-            return(KERN_FAILURE);
+       return (result);
 }
 
 
 }
 
 
-kern_return_t check_actforsig(task_t task, thread_act_t thact, thread_t * thshut, int setast)
+kern_return_t
+check_actforsig(
+       task_t                  task,
+       thread_t                thread,
+       int                             setast)
 {
 {
-
-        thread_act_t inc;
-        thread_act_t ninc;
-        thread_act_t thr_act;
-               thread_t        th;
-               int found=0;
+       kern_return_t   result = KERN_FAILURE;
+       thread_t                inc;
 
        task_lock(task);
 
        task_lock(task);
+
        if (!task->active) {
                task_unlock(task);
        if (!task->active) {
                task_unlock(task);
-               return(KERN_FAILURE);
-       }
-
-        thr_act = THR_ACT_NULL;
-        for (inc  = (thread_act_t)queue_first(&task->thr_acts);
-             inc != (thread_act_t)&task->thr_acts;
-             inc  = ninc) {
-
-                               if (inc != thact) {
-                       ninc = (thread_act_t)queue_next(&inc->thr_acts);
-                                               continue;
-                               }
-                th = act_lock_thread(inc);
-                if ((inc->active)  && 
-                    ((th->state & (TH_ABORT|TH_ABORT_SAFELY)) != TH_ABORT)) {
-                                       found = 1;
-                    thr_act = inc;
-                   break;
-                }
-                act_unlock_thread(inc);
-                /* ninc = (thread_act_t)queue_next(&inc->thr_acts); */
+
+               return (KERN_FAILURE);
+       }
+
+       for (inc  = (thread_t)(void *)queue_first(&task->threads);
+                       !queue_end(&task->threads, (queue_entry_t)inc); ) {
+               if (inc == thread) {
+                       thread_mtx_lock(inc);
+
+                       if (inc->active  && 
+                                       (inc->sched_flags & TH_SFLAG_ABORTED_MASK) != TH_SFLAG_ABORT) {
+                               result = KERN_SUCCESS;
                                break;
                                break;
-        }
-out:
-               if (found) {
-               if (thshut)
-                       *thshut = thr_act? thr_act->thread: THREAD_NULL ;
-            if (setast)
-                               act_set_astbsd(thr_act);
-
-           act_unlock_thread(thr_act);
-        }
-               task_unlock(task);
+                       }
+
+                       thread_mtx_unlock(inc);
+                       break;
+               }
+
+               inc = (thread_t)(void *)queue_next(&inc->task_threads);
+       }
 
 
-        if (found) 
-            return(KERN_SUCCESS);
-        else 
-            return(KERN_FAILURE);
+       if (result == KERN_SUCCESS) {
+               if (setast)
+                       act_set_astbsd(thread);
+
+               thread_mtx_unlock(thread);
+       }
+
+       task_unlock(task);
+
+       return (result);
+}
+
+ledger_t  get_task_ledger(task_t t)
+{
+       return(t->ledger);
 }
 
 /*
 }
 
 /*
- *
+ * This is only safe to call from a thread executing in
+ * in the task's context or if the task is locked. Otherwise,
+ * the map could be switched for the task (and freed) before
+ * we go to return it here.
  */
 vm_map_t  get_task_map(task_t t)
 {
        return(t->map);
 }
 
  */
 vm_map_t  get_task_map(task_t t)
 {
        return(t->map);
 }
 
+vm_map_t  get_task_map_reference(task_t t)
+{
+       vm_map_t m;
+
+       if (t == NULL)
+               return VM_MAP_NULL;
+
+       task_lock(t);
+       if (!t->active) {
+               task_unlock(t);
+               return VM_MAP_NULL;
+       }
+       m = t->map;
+       vm_map_reference_swap(m);
+       task_unlock(t);
+       return m;
+}
+
 /*
  *
  */
 /*
  *
  */
@@ -229,47 +274,76 @@ ipc_space_t  get_task_ipcspace(task_t t)
        return(t->itk_space);
 }
 
        return(t->itk_space);
 }
 
+int get_task_numactivethreads(task_t task)
+{
+       thread_t        inc;
+       int num_active_thr=0;
+       task_lock(task);
+
+       for (inc  = (thread_t)(void *)queue_first(&task->threads);
+                       !queue_end(&task->threads, (queue_entry_t)inc); inc = (thread_t)(void *)queue_next(&inc->task_threads)) 
+       {
+               if(inc->active)
+                       num_active_thr++;
+       }
+       task_unlock(task);
+       return num_active_thr;
+}
+
 int  get_task_numacts(task_t t)
 {
 int  get_task_numacts(task_t t)
 {
-       return(t->thr_act_count);
+       return(t->thread_count);
+}
+
+/* does this machine need  64bit register set for signal handler */
+int is_64signalregset(void)
+{
+       if (task_has_64BitData(current_task())) {
+               return(1);
+       }
+
+       return(0);
 }
 
 /*
 }
 
 /*
- * Reset the current task's map by taking a reference
- * on the new map.  The old map reference is returned.
+ * Swap in a new map for the task/thread pair; the old map reference is
+ * returned. Also does a pmap switch if thread provided is current thread.
  */
 vm_map_t
  */
 vm_map_t
-swap_task_map(task_t task,vm_map_t map)
+swap_task_map(task_t task, thread_t thread, vm_map_t map)
 {
        vm_map_t old_map;
 {
        vm_map_t old_map;
+       boolean_t doswitch = (thread == current_thread()) ? TRUE : FALSE;
+
+       if (task != thread->task)
+               panic("swap_task_map");
 
 
-       vm_map_reference(map);
        task_lock(task);
        task_lock(task);
+       mp_disable_preemption();
+
        old_map = task->map;
        old_map = task->map;
-       task->map = map;
+       thread->map = task->map = map;
+       vm_commit_pagezero_status(map);
+
+       if (doswitch) {
+               pmap_switch(map->pmap);
+       }
+       mp_enable_preemption();
        task_unlock(task);
        task_unlock(task);
-       return old_map;
-}
 
 
-/*
- * Reset the current act map.
- * The caller donates us a reference to the new map
- * and we donote our reference to the old map to him.
- */
-vm_map_t
-swap_act_map(thread_act_t thr_act,vm_map_t map)
-{
-       vm_map_t old_map;
+#if (defined(__i386__) || defined(__x86_64__)) && NCOPY_WINDOWS > 0
+       inval_copy_windows(thread);
+#endif
 
 
-       act_lock(thr_act);
-       old_map = thr_act->map;
-       thr_act->map = map;
-       act_unlock(thr_act);
        return old_map;
 }
 
 /*
  *
        return old_map;
 }
 
 /*
  *
+ * This is only safe to call from a thread executing in
+ * in the task's context or if the task is locked. Otherwise,
+ * the map could be switched for the task (and freed) before
+ * we go to return it here.
  */
 pmap_t  get_task_pmap(task_t t)
 {
  */
 pmap_t  get_task_pmap(task_t t)
 {
@@ -279,66 +353,230 @@ pmap_t  get_task_pmap(task_t t)
 /*
  *
  */
 /*
  *
  */
-pmap_t  get_map_pmap(vm_map_t map)
+uint64_t get_task_resident_size(task_t task) 
 {
 {
-       return(map->pmap);
+       vm_map_t map;
+       
+       map = (task == kernel_task) ? kernel_map: task->map;
+       return((uint64_t)pmap_resident_count(map->pmap) * PAGE_SIZE_64);
 }
 }
-/*
- *
- */
-task_t get_threadtask(thread_act_t th)
+
+uint64_t get_task_compressed(task_t task) 
 {
 {
-       return(th->task);
+       vm_map_t map;
+       
+       map = (task == kernel_task) ? kernel_map: task->map;
+       return((uint64_t)pmap_compressed(map->pmap) * PAGE_SIZE_64);
+}
+
+uint64_t get_task_resident_max(task_t task) 
+{
+       vm_map_t map;
+       
+       map = (task == kernel_task) ? kernel_map: task->map;
+       return((uint64_t)pmap_resident_max(map->pmap) * PAGE_SIZE_64);
 }
 
 }
 
+uint64_t get_task_purgeable_size(task_t task) 
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+       uint64_t volatile_size = 0;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.purgeable_volatile, &credit, &debit);
+       if (ret != KERN_SUCCESS) {
+               return 0;
+       }
+
+       volatile_size += (credit - debit);
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.purgeable_volatile_compressed, &credit, &debit);
+       if (ret != KERN_SUCCESS) {
+               return 0;
+       }
+
+       volatile_size += (credit - debit);
+
+       return volatile_size;
+}
 
 /*
  *
  */
 
 /*
  *
  */
-boolean_t is_thread_idle(thread_t th)
-{
-       return((th->state & TH_IDLE) == TH_IDLE);
+uint64_t get_task_phys_footprint(task_t task) 
+{      
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+       
+       ret = ledger_get_entries(task->ledger, task_ledgers.phys_footprint, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
 }
 
 /*
  *
  */
 }
 
 /*
  *
  */
-boolean_t is_thread_running(thread_act_t thact)
-{
-       thread_t th = thact->thread;
-       return((th->state & TH_RUN) == TH_RUN);
+uint64_t get_task_phys_footprint_max(task_t task) 
+{      
+       kern_return_t ret;
+       ledger_amount_t max;
+       
+       ret = ledger_get_maximum(task->ledger, task_ledgers.phys_footprint, &max);
+       if (KERN_SUCCESS == ret) {
+               return max;
+       }
+
+       return 0;
 }
 
 /*
  *
  */
 }
 
 /*
  *
  */
-thread_shuttle_t
-getshuttle_thread(
-       thread_act_t    th)
+uint64_t get_task_phys_footprint_limit(task_t task)
 {
 {
-#ifdef DEBUG
-       assert(th->thread);
-#endif
-       return(th->thread);
+       kern_return_t ret;
+       ledger_amount_t max;
+
+       ret = ledger_get_limit(task->ledger, task_ledgers.phys_footprint, &max);
+       if (KERN_SUCCESS == ret) {
+               return max;
+       }
+
+       return 0;
+}
+
+uint64_t get_task_internal(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.internal, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_internal_compressed(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.internal_compressed, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_purgeable_nonvolatile(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.purgeable_nonvolatile, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_purgeable_nonvolatile_compressed(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.purgeable_nonvolatile_compressed, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_alternate_accounting(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.alternate_accounting, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_alternate_accounting_compressed(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.alternate_accounting_compressed, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_page_table(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.page_table, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_iokit_mapped(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+
+       ret = ledger_get_entries(task->ledger, task_ledgers.iokit_mapped, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
+}
+
+uint64_t get_task_cpu_time(task_t task)
+{
+       kern_return_t ret;
+       ledger_amount_t credit, debit;
+       
+       ret = ledger_get_entries(task->ledger, task_ledgers.cpu_time, &credit, &debit);
+       if (KERN_SUCCESS == ret) {
+               return (credit - debit);
+       }
+
+       return 0;
 }
 
 /*
  *
  */
 }
 
 /*
  *
  */
-thread_act_t
-getact_thread(
-       thread_shuttle_t        th)
+task_t get_threadtask(thread_t th)
 {
 {
-#ifdef DEBUG
-       assert(th->top_act);
-#endif
-       return(th->top_act);
+       return(th->task);
 }
 
 /*
  *
  */
 }
 
 /*
  *
  */
-vm_offset_t
+vm_map_offset_t
 get_map_min(
        vm_map_t        map)
 {
 get_map_min(
        vm_map_t        map)
 {
@@ -348,20 +586,22 @@ get_map_min(
 /*
  *
  */
 /*
  *
  */
-vm_offset_t
+vm_map_offset_t
 get_map_max(
        vm_map_t        map)
 {
        return(vm_map_max(map));
 }
 get_map_max(
        vm_map_t        map)
 {
        return(vm_map_max(map));
 }
-vm_size_t
+vm_map_size_t
 get_vmmap_size(
        vm_map_t        map)
 {
        return(map->size);
 }
 
 get_vmmap_size(
        vm_map_t        map)
 {
        return(map->size);
 }
 
-int
+#if CONFIG_COREDUMP
+
+static int
 get_vmsubmap_entries(
        vm_map_t        map,
        vm_object_offset_t      start,
 get_vmsubmap_entries(
        vm_map_t        map,
        vm_object_offset_t      start,
@@ -370,7 +610,8 @@ get_vmsubmap_entries(
        int     total_entries = 0;
        vm_map_entry_t  entry;
 
        int     total_entries = 0;
        vm_map_entry_t  entry;
 
-       vm_map_lock(map);
+       if (not_in_kdp)
+         vm_map_lock(map);
        entry = vm_map_first_entry(map);
        while((entry != vm_map_to_entry(map)) && (entry->vme_start < start)) {
                entry = entry->vme_next;
        entry = vm_map_first_entry(map);
        while((entry != vm_map_to_entry(map)) && (entry->vme_start < start)) {
                entry = entry->vme_next;
@@ -379,16 +620,18 @@ get_vmsubmap_entries(
        while((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
                if(entry->is_sub_map) {
                        total_entries +=        
        while((entry != vm_map_to_entry(map)) && (entry->vme_start < end)) {
                if(entry->is_sub_map) {
                        total_entries +=        
-                               get_vmsubmap_entries(entry->object.sub_map, 
-                                       entry->offset, 
-                                       entry->offset + 
-                                       (entry->vme_end - entry->vme_start));
+                               get_vmsubmap_entries(VME_SUBMAP(entry), 
+                                                    VME_OFFSET(entry), 
+                                                    (VME_OFFSET(entry) + 
+                                                     entry->vme_end -
+                                                     entry->vme_start));
                } else {
                        total_entries += 1;
                }
                entry = entry->vme_next;
        }
                } else {
                        total_entries += 1;
                }
                entry = entry->vme_next;
        }
-       vm_map_unlock(map);
+       if (not_in_kdp)
+         vm_map_unlock(map);
        return(total_entries);
 }
 
        return(total_entries);
 }
 
@@ -399,24 +642,28 @@ get_vmmap_entries(
        int     total_entries = 0;
        vm_map_entry_t  entry;
 
        int     total_entries = 0;
        vm_map_entry_t  entry;
 
-       vm_map_lock(map);
+       if (not_in_kdp)
+         vm_map_lock(map);
        entry = vm_map_first_entry(map);
 
        while(entry != vm_map_to_entry(map)) {
                if(entry->is_sub_map) {
                        total_entries +=        
        entry = vm_map_first_entry(map);
 
        while(entry != vm_map_to_entry(map)) {
                if(entry->is_sub_map) {
                        total_entries +=        
-                               get_vmsubmap_entries(entry->object.sub_map, 
-                                       entry->offset, 
-                                       entry->offset + 
-                                       (entry->vme_end - entry->vme_start));
+                               get_vmsubmap_entries(VME_SUBMAP(entry), 
+                                                    VME_OFFSET(entry),
+                                                    (VME_OFFSET(entry) + 
+                                                     entry->vme_end -
+                                                     entry->vme_start));
                } else {
                        total_entries += 1;
                }
                entry = entry->vme_next;
        }
                } else {
                        total_entries += 1;
                }
                entry = entry->vme_next;
        }
-       vm_map_unlock(map);
+       if (not_in_kdp)
+         vm_map_unlock(map);
        return(total_entries);
 }
        return(total_entries);
 }
+#endif /* CONFIG_COREDUMP */
 
 /*
  *
 
 /*
  *
@@ -436,20 +683,39 @@ get_task_userstop(
  */
 int
 get_thread_userstop(
  */
 int
 get_thread_userstop(
-       thread_act_t th)
+       thread_t th)
 {
        return(th->user_stop_count);
 }
 
 {
        return(th->user_stop_count);
 }
 
+/*
+ *
+ */
+boolean_t
+get_task_pidsuspended(
+       task_t task)
+{
+    return (task->pidsuspended);
+}
+
+/*
+ *
+ */
+boolean_t 
+get_task_frozen(
+       task_t task)
+{
+    return (task->frozen);   
+}
+
 /*
  *
  */
 boolean_t
 thread_should_abort(
 /*
  *
  */
 boolean_t
 thread_should_abort(
-       thread_shuttle_t th)
+       thread_t th)
 {
 {
-       return(!th->top_act || !th->top_act->active || 
-              (th->state & (TH_ABORT|TH_ABORT_SAFELY)) == TH_ABORT);
+       return ((th->sched_flags & TH_SFLAG_ABORTED_MASK) == TH_SFLAG_ABORT);
 }
 
 /*
 }
 
 /*
@@ -467,15 +733,14 @@ current_thread_aborted (
        thread_t th = current_thread();
        spl_t s;
 
        thread_t th = current_thread();
        spl_t s;
 
-       if (!th->top_act || 
-               ((th->state & (TH_ABORT|TH_ABORT_SAFELY)) == TH_ABORT &&
-                th->interrupt_level != THREAD_UNINT))
+       if ((th->sched_flags & TH_SFLAG_ABORTED_MASK) == TH_SFLAG_ABORT &&
+                       (th->options & TH_OPT_INTMASK) != THREAD_UNINT)
                return (TRUE);
                return (TRUE);
-       if (th->state & TH_ABORT_SAFELY) {
+       if (th->sched_flags & TH_SFLAG_ABORTSAFELY) {
                s = splsched();
                thread_lock(th);
                s = splsched();
                thread_lock(th);
-               if (th->state & TH_ABORT_SAFELY)
-                       th->state &= ~(TH_ABORT|TH_ABORT_SAFELY);
+               if (th->sched_flags & TH_SFLAG_ABORTSAFELY)
+                       th->sched_flags &= ~TH_SFLAG_ABORTED_MASK;
                thread_unlock(th);
                splx(s);
        }
                thread_unlock(th);
                splx(s);
        }
@@ -487,49 +752,311 @@ current_thread_aborted (
  */
 void
 task_act_iterate_wth_args(
  */
 void
 task_act_iterate_wth_args(
-       task_t task,
-       void (*func_callback)(thread_act_t, void *),
-       void *func_arg)
+       task_t                  task,
+       void                    (*func_callback)(thread_t, void *),
+       void                    *func_arg)
 {
 {
-        thread_act_t inc, ninc;
+       thread_t        inc;
 
        task_lock(task);
 
        task_lock(task);
-        for (inc  = (thread_act_t)queue_first(&task->thr_acts);
-             inc != (thread_act_t)&task->thr_acts;
-             inc  = ninc) {
-                ninc = (thread_act_t)queue_next(&inc->thr_acts);
-                (void) (*func_callback)(inc, func_arg);
-        }
+
+       for (inc  = (thread_t)(void *)queue_first(&task->threads);
+                       !queue_end(&task->threads, (queue_entry_t)inc); ) {
+               (void) (*func_callback)(inc, func_arg);
+               inc = (thread_t)(void *)queue_next(&inc->task_threads);
+       }
+
        task_unlock(task);
 }
 
        task_unlock(task);
 }
 
+
+#include <sys/bsdtask_info.h>
+
 void
 void
-ipc_port_release(
-       ipc_port_t port)
+fill_taskprocinfo(task_t task, struct proc_taskinfo_internal * ptinfo)
 {
 {
-       ipc_object_release(&(port)->ip_object);
+       vm_map_t map;
+       task_absolutetime_info_data_t   tinfo;
+       thread_t thread;
+       uint32_t cswitch = 0, numrunning = 0;
+       uint32_t syscalls_unix = 0;
+       uint32_t syscalls_mach = 0;
+
+       task_lock(task);
+
+       map = (task == kernel_task)? kernel_map: task->map;
+
+       ptinfo->pti_virtual_size  = map->size;
+       ptinfo->pti_resident_size =
+               (mach_vm_size_t)(pmap_resident_count(map->pmap))
+               * PAGE_SIZE_64;
+
+       ptinfo->pti_policy = ((task != kernel_task)?
+                                          POLICY_TIMESHARE: POLICY_RR);
+
+       tinfo.threads_user = tinfo.threads_system = 0;
+       tinfo.total_user = task->total_user_time;
+       tinfo.total_system = task->total_system_time;
+
+       queue_iterate(&task->threads, thread, thread_t, task_threads) {
+               uint64_t    tval;
+               spl_t x;
+
+               if (thread->options & TH_OPT_IDLE_THREAD)
+                       continue;
+
+               x = splsched();
+               thread_lock(thread);
+
+               if ((thread->state & TH_RUN) == TH_RUN)
+                       numrunning++;
+               cswitch += thread->c_switch;
+               tval = timer_grab(&thread->user_timer);
+               tinfo.threads_user += tval;
+               tinfo.total_user += tval;
+
+               tval = timer_grab(&thread->system_timer);
+
+               if (thread->precise_user_kernel_time) {
+                       tinfo.threads_system += tval;
+                       tinfo.total_system += tval;
+               } else {
+                       /* system_timer may represent either sys or user */
+                       tinfo.threads_user += tval;
+                       tinfo.total_user += tval;
+               }
+
+               syscalls_unix += thread->syscalls_unix;
+               syscalls_mach += thread->syscalls_mach;
+
+               thread_unlock(thread);
+               splx(x);
+       }
+
+       ptinfo->pti_total_system = tinfo.total_system;
+       ptinfo->pti_total_user = tinfo.total_user;
+       ptinfo->pti_threads_system = tinfo.threads_system;
+       ptinfo->pti_threads_user = tinfo.threads_user;
+       
+       ptinfo->pti_faults = task->faults;
+       ptinfo->pti_pageins = task->pageins;
+       ptinfo->pti_cow_faults = task->cow_faults;
+       ptinfo->pti_messages_sent = task->messages_sent;
+       ptinfo->pti_messages_received = task->messages_received;
+       ptinfo->pti_syscalls_mach = task->syscalls_mach + syscalls_mach;
+       ptinfo->pti_syscalls_unix = task->syscalls_unix + syscalls_unix;
+       ptinfo->pti_csw = task->c_switch + cswitch;
+       ptinfo->pti_threadnum = task->thread_count;
+       ptinfo->pti_numrunning = numrunning;
+       ptinfo->pti_priority = task->priority;
+
+       task_unlock(task);
 }
 
 }
 
-boolean_t
-is_thread_active(
-       thread_shuttle_t th)
+int 
+fill_taskthreadinfo(task_t task, uint64_t thaddr, int thuniqueid, struct proc_threadinfo_internal * ptinfo, void * vpp, int *vidp)
 {
 {
-       return(th->active);
+       thread_t  thact;
+       int err=0;
+       mach_msg_type_number_t count;
+       thread_basic_info_data_t basic_info;
+       kern_return_t kret;
+       uint64_t addr = 0;
+
+       task_lock(task);
+
+       for (thact  = (thread_t)(void *)queue_first(&task->threads);
+                       !queue_end(&task->threads, (queue_entry_t)thact); ) {
+               addr = (thuniqueid==0)?thact->machine.cthread_self: thact->thread_id;
+               if (addr == thaddr)
+               {
+               
+                       count = THREAD_BASIC_INFO_COUNT;
+                       if ((kret = thread_info_internal(thact, THREAD_BASIC_INFO, (thread_info_t)&basic_info, &count)) != KERN_SUCCESS) {
+                               err = 1;
+                               goto out;       
+                       }
+                       ptinfo->pth_user_time = ((basic_info.user_time.seconds * (integer_t)NSEC_PER_SEC) + (basic_info.user_time.microseconds * (integer_t)NSEC_PER_USEC));
+                       ptinfo->pth_system_time = ((basic_info.system_time.seconds * (integer_t)NSEC_PER_SEC) + (basic_info.system_time.microseconds * (integer_t)NSEC_PER_USEC));
+
+                       ptinfo->pth_cpu_usage = basic_info.cpu_usage;
+                       ptinfo->pth_policy = basic_info.policy;
+                       ptinfo->pth_run_state = basic_info.run_state;
+                       ptinfo->pth_flags = basic_info.flags;
+                       ptinfo->pth_sleep_time = basic_info.sleep_time;
+                       ptinfo->pth_curpri = thact->sched_pri;
+                       ptinfo->pth_priority = thact->base_pri;
+                       ptinfo->pth_maxpriority = thact->max_priority;
+                       
+                       if ((vpp != NULL) && (thact->uthread != NULL)) 
+                               bsd_threadcdir(thact->uthread, vpp, vidp);
+                       bsd_getthreadname(thact->uthread,ptinfo->pth_name);
+                       err = 0;
+                       goto out; 
+               }
+               thact = (thread_t)(void *)queue_next(&thact->task_threads);
+       }
+       err = 1;
+
+out:
+       task_unlock(task);
+       return(err);
 }
 
 }
 
-kern_return_t
-get_thread_waitresult(
-       thread_shuttle_t th)
+int
+fill_taskthreadlist(task_t task, void * buffer, int thcount)
 {
 {
-       return(th->wait_result);
+       int numthr=0;
+       thread_t thact;
+       uint64_t * uptr;
+       uint64_t  thaddr;
+
+       uptr = (uint64_t *)buffer;
+
+       task_lock(task);
+
+       for (thact  = (thread_t)(void *)queue_first(&task->threads);
+                       !queue_end(&task->threads, (queue_entry_t)thact); ) {
+               thaddr = thact->machine.cthread_self;
+               *uptr++ = thaddr;
+               numthr++;
+               if (numthr >= thcount)
+                       goto out;
+               thact = (thread_t)(void *)queue_next(&thact->task_threads);
+       }
+
+out:
+       task_unlock(task);
+       return (int)(numthr * sizeof(uint64_t));
+       
+}
+
+int
+get_numthreads(task_t task)
+{
+       return(task->thread_count);
+}
+
+/*
+ * Gather the various pieces of info about the designated task, 
+ * and collect it all into a single rusage_info.
+ */
+int
+fill_task_rusage(task_t task, rusage_info_current *ri)
+{
+       struct task_power_info powerinfo;
+
+       assert(task != TASK_NULL);
+       task_lock(task);
+
+       task_power_info_locked(task, &powerinfo, NULL, NULL);
+       ri->ri_pkg_idle_wkups = powerinfo.task_platform_idle_wakeups;
+       ri->ri_interrupt_wkups = powerinfo.task_interrupt_wakeups;
+       ri->ri_user_time = powerinfo.total_user;
+       ri->ri_system_time = powerinfo.total_system;
+
+       ledger_get_balance(task->ledger, task_ledgers.phys_footprint,
+                          (ledger_amount_t *)&ri->ri_phys_footprint);
+       ledger_get_balance(task->ledger, task_ledgers.phys_mem,
+                          (ledger_amount_t *)&ri->ri_resident_size);
+       ledger_get_balance(task->ledger, task_ledgers.wired_mem,
+                          (ledger_amount_t *)&ri->ri_wired_size);
+
+       ri->ri_pageins = task->pageins;
+
+       task_unlock(task);
+       return (0);
 }
 
 void
 }
 
 void
-astbsd_on(void)
+fill_task_billed_usage(task_t task __unused, rusage_info_current *ri)
 {
 {
-       boolean_t       reenable;
+#if CONFIG_BANK
+       ri->ri_billed_system_time = bank_billed_time_safe(task);
+       ri->ri_serviced_system_time = bank_serviced_time_safe(task);
+#else
+       ri->ri_billed_system_time = 0;
+       ri->ri_serviced_system_time = 0;
+#endif
+}
+
+int
+fill_task_io_rusage(task_t task, rusage_info_current *ri)
+{
+       assert(task != TASK_NULL);
+       task_lock(task);
+
+       if (task->task_io_stats) {
+               ri->ri_diskio_bytesread = task->task_io_stats->disk_reads.size;
+               ri->ri_diskio_byteswritten = (task->task_io_stats->total_io.size - task->task_io_stats->disk_reads.size);
+       } else {
+               /* I/O Stats unavailable */
+               ri->ri_diskio_bytesread = 0;
+               ri->ri_diskio_byteswritten = 0;
+       }
+       task_unlock(task);
+       return (0);
+}
+
+int
+fill_task_qos_rusage(task_t task, rusage_info_current *ri)
+{
+       thread_t thread;
+
+       assert(task != TASK_NULL);
+       task_lock(task);
+
+       /* Rollup Qos time of all the threads to task */
+       queue_iterate(&task->threads, thread, thread_t, task_threads) {
+               if (thread->options & TH_OPT_IDLE_THREAD)
+                       continue;
+
+               thread_update_qos_cpu_time(thread);
+       }
+       ri->ri_cpu_time_qos_default = task->cpu_time_qos_stats.cpu_time_qos_default;
+       ri->ri_cpu_time_qos_maintenance = task->cpu_time_qos_stats.cpu_time_qos_maintenance;
+       ri->ri_cpu_time_qos_background = task->cpu_time_qos_stats.cpu_time_qos_background;
+       ri->ri_cpu_time_qos_utility = task->cpu_time_qos_stats.cpu_time_qos_utility;
+       ri->ri_cpu_time_qos_legacy = task->cpu_time_qos_stats.cpu_time_qos_legacy;
+       ri->ri_cpu_time_qos_user_initiated = task->cpu_time_qos_stats.cpu_time_qos_user_initiated;
+       ri->ri_cpu_time_qos_user_interactive = task->cpu_time_qos_stats.cpu_time_qos_user_interactive;
+
+       task_unlock(task);
+       return (0);
+}
+
+uint64_t
+get_task_dispatchqueue_serialno_offset(task_t task)
+{
+       uint64_t dq_serialno_offset = 0;
+
+       if (task->bsd_info) {
+               dq_serialno_offset = get_dispatchqueue_serialno_offset_from_proc(task->bsd_info);
+       }
+
+       return dq_serialno_offset;
+}
+
+uint64_t
+get_task_uniqueid(task_t task)
+{
+       if (task->bsd_info) {
+               return proc_uniqueid(task->bsd_info);
+       } else {
+               return UINT64_MAX;
+       }
+}
 
 
-       reenable = ml_set_interrupts_enabled(FALSE);
-       ast_on_fast(AST_BSD);
-       (void)ml_set_interrupts_enabled(reenable);
+#if CONFIG_MACF
+struct label *
+get_task_crash_label(task_t task)
+{
+       return task->crash_label;
 }
 }
+
+void
+set_task_crash_label(task_t task, struct label *label)
+{
+       task->crash_label = label;
+}
+#endif