35
36#include <sys/ctype.h>
37#include <sys/unistd.h>
38#include <sys/param.h>
39#include <sys/types.h>
40#include <sys/errno.h>
41#include <sys/systm.h>
42#include <sys/malloc.h>
43#include <sys/lock.h>
44#include <sys/mutex.h>
45
46#include <sys/callout.h>
47#include <sys/kdb.h>
48#include <sys/kernel.h>
49#include <sys/proc.h>
50#include <sys/condvar.h>
51#include <sys/kthread.h>
52#include <sys/module.h>
53#include <sys/smp.h>
54#include <sys/sched.h>
55#include <sys/sysctl.h>
56
57#include <machine/atomic.h>
58#include <machine/bus.h>
59#include <machine/stdarg.h>
60#include <machine/resource.h>
61
62#include <sys/bus.h>
63#include <sys/rman.h>
64
65#include <vm/vm.h>
66#include <vm/vm_param.h>
67#include <vm/pmap.h>
68#include <vm/uma.h>
69#include <vm/vm_kern.h>
70#include <vm/vm_map.h>
71#include <vm/vm_extern.h>
72
73#include <compat/ndis/pe_var.h>
74#include <compat/ndis/cfg_var.h>
75#include <compat/ndis/resource_var.h>
76#include <compat/ndis/ntoskrnl_var.h>
77#include <compat/ndis/hal_var.h>
78#include <compat/ndis/ndis_var.h>
79
80#ifdef NTOSKRNL_DEBUG_TIMERS
81static int sysctl_show_timers(SYSCTL_HANDLER_ARGS);
82
83SYSCTL_PROC(_debug, OID_AUTO, ntoskrnl_timers, CTLTYPE_INT | CTLFLAG_RW,
84 NULL, 0, sysctl_show_timers, "I",
85 "Show ntoskrnl timer stats");
86#endif
87
88struct kdpc_queue {
89 list_entry kq_disp;
90 struct thread *kq_td;
91 int kq_cpu;
92 int kq_exit;
93 int kq_running;
94 kspin_lock kq_lock;
95 nt_kevent kq_proc;
96 nt_kevent kq_done;
97};
98
99typedef struct kdpc_queue kdpc_queue;
100
101struct wb_ext {
102 struct cv we_cv;
103 struct thread *we_td;
104};
105
106typedef struct wb_ext wb_ext;
107
108#define NTOSKRNL_TIMEOUTS 256
109#ifdef NTOSKRNL_DEBUG_TIMERS
110static uint64_t ntoskrnl_timer_fires;
111static uint64_t ntoskrnl_timer_sets;
112static uint64_t ntoskrnl_timer_reloads;
113static uint64_t ntoskrnl_timer_cancels;
114#endif
115
116struct callout_entry {
117 struct callout ce_callout;
118 list_entry ce_list;
119};
120
121typedef struct callout_entry callout_entry;
122
123static struct list_entry ntoskrnl_calllist;
124static struct mtx ntoskrnl_calllock;
125struct kuser_shared_data kuser_shared_data;
126
127static struct list_entry ntoskrnl_intlist;
128static kspin_lock ntoskrnl_intlock;
129
130static uint8_t RtlEqualUnicodeString(unicode_string *,
131 unicode_string *, uint8_t);
132static void RtlCopyString(ansi_string *, const ansi_string *);
133static void RtlCopyUnicodeString(unicode_string *,
134 unicode_string *);
135static irp *IoBuildSynchronousFsdRequest(uint32_t, device_object *,
136 void *, uint32_t, uint64_t *, nt_kevent *, io_status_block *);
137static irp *IoBuildAsynchronousFsdRequest(uint32_t,
138 device_object *, void *, uint32_t, uint64_t *, io_status_block *);
139static irp *IoBuildDeviceIoControlRequest(uint32_t,
140 device_object *, void *, uint32_t, void *, uint32_t,
141 uint8_t, nt_kevent *, io_status_block *);
142static irp *IoAllocateIrp(uint8_t, uint8_t);
143static void IoReuseIrp(irp *, uint32_t);
144static void IoFreeIrp(irp *);
145static void IoInitializeIrp(irp *, uint16_t, uint8_t);
146static irp *IoMakeAssociatedIrp(irp *, uint8_t);
147static uint32_t KeWaitForMultipleObjects(uint32_t,
148 nt_dispatch_header **, uint32_t, uint32_t, uint32_t, uint8_t,
149 int64_t *, wait_block *);
150static void ntoskrnl_waittest(nt_dispatch_header *, uint32_t);
151static void ntoskrnl_satisfy_wait(nt_dispatch_header *, struct thread *);
152static void ntoskrnl_satisfy_multiple_waits(wait_block *);
153static int ntoskrnl_is_signalled(nt_dispatch_header *, struct thread *);
154static void ntoskrnl_insert_timer(ktimer *, int);
155static void ntoskrnl_remove_timer(ktimer *);
156#ifdef NTOSKRNL_DEBUG_TIMERS
157static void ntoskrnl_show_timers(void);
158#endif
159static void ntoskrnl_timercall(void *);
160static void ntoskrnl_dpc_thread(void *);
161static void ntoskrnl_destroy_dpc_threads(void);
162static void ntoskrnl_destroy_workitem_threads(void);
163static void ntoskrnl_workitem_thread(void *);
164static void ntoskrnl_workitem(device_object *, void *);
165static void ntoskrnl_unicode_to_ascii(uint16_t *, char *, int);
166static void ntoskrnl_ascii_to_unicode(char *, uint16_t *, int);
167static uint8_t ntoskrnl_insert_dpc(list_entry *, kdpc *);
168static void WRITE_REGISTER_USHORT(uint16_t *, uint16_t);
169static uint16_t READ_REGISTER_USHORT(uint16_t *);
170static void WRITE_REGISTER_ULONG(uint32_t *, uint32_t);
171static uint32_t READ_REGISTER_ULONG(uint32_t *);
172static void WRITE_REGISTER_UCHAR(uint8_t *, uint8_t);
173static uint8_t READ_REGISTER_UCHAR(uint8_t *);
174static int64_t _allmul(int64_t, int64_t);
175static int64_t _alldiv(int64_t, int64_t);
176static int64_t _allrem(int64_t, int64_t);
177static int64_t _allshr(int64_t, uint8_t);
178static int64_t _allshl(int64_t, uint8_t);
179static uint64_t _aullmul(uint64_t, uint64_t);
180static uint64_t _aulldiv(uint64_t, uint64_t);
181static uint64_t _aullrem(uint64_t, uint64_t);
182static uint64_t _aullshr(uint64_t, uint8_t);
183static uint64_t _aullshl(uint64_t, uint8_t);
184static slist_entry *ntoskrnl_pushsl(slist_header *, slist_entry *);
185static void InitializeSListHead(slist_header *);
186static slist_entry *ntoskrnl_popsl(slist_header *);
187static void ExFreePoolWithTag(void *, uint32_t);
188static void ExInitializePagedLookasideList(paged_lookaside_list *,
189 lookaside_alloc_func *, lookaside_free_func *,
190 uint32_t, size_t, uint32_t, uint16_t);
191static void ExDeletePagedLookasideList(paged_lookaside_list *);
192static void ExInitializeNPagedLookasideList(npaged_lookaside_list *,
193 lookaside_alloc_func *, lookaside_free_func *,
194 uint32_t, size_t, uint32_t, uint16_t);
195static void ExDeleteNPagedLookasideList(npaged_lookaside_list *);
196static slist_entry
197 *ExInterlockedPushEntrySList(slist_header *,
198 slist_entry *, kspin_lock *);
199static slist_entry
200 *ExInterlockedPopEntrySList(slist_header *, kspin_lock *);
201static uint32_t InterlockedIncrement(volatile uint32_t *);
202static uint32_t InterlockedDecrement(volatile uint32_t *);
203static void ExInterlockedAddLargeStatistic(uint64_t *, uint32_t);
204static void *MmAllocateContiguousMemory(uint32_t, uint64_t);
205static void *MmAllocateContiguousMemorySpecifyCache(uint32_t,
206 uint64_t, uint64_t, uint64_t, enum nt_caching_type);
207static void MmFreeContiguousMemory(void *);
208static void MmFreeContiguousMemorySpecifyCache(void *, uint32_t,
209 enum nt_caching_type);
210static uint32_t MmSizeOfMdl(void *, size_t);
211static void *MmMapLockedPages(mdl *, uint8_t);
212static void *MmMapLockedPagesSpecifyCache(mdl *,
213 uint8_t, uint32_t, void *, uint32_t, uint32_t);
214static void MmUnmapLockedPages(void *, mdl *);
215static device_t ntoskrnl_finddev(device_t, uint64_t, struct resource **);
216static void RtlZeroMemory(void *, size_t);
217static void RtlSecureZeroMemory(void *, size_t);
218static void RtlFillMemory(void *, size_t, uint8_t);
219static void RtlMoveMemory(void *, const void *, size_t);
220static ndis_status RtlCharToInteger(const char *, uint32_t, uint32_t *);
221static void RtlCopyMemory(void *, const void *, size_t);
222static size_t RtlCompareMemory(const void *, const void *, size_t);
223static ndis_status RtlUnicodeStringToInteger(unicode_string *,
224 uint32_t, uint32_t *);
225static int atoi (const char *);
226static long atol (const char *);
227static int rand(void);
228static void srand(unsigned int);
229static void KeQuerySystemTime(uint64_t *);
230static uint32_t KeTickCount(void);
231static uint8_t IoIsWdmVersionAvailable(uint8_t, uint8_t);
232static int32_t IoOpenDeviceRegistryKey(struct device_object *, uint32_t,
233 uint32_t, void **);
234static void ntoskrnl_thrfunc(void *);
235static ndis_status PsCreateSystemThread(ndis_handle *,
236 uint32_t, void *, ndis_handle, void *, void *, void *);
237static ndis_status PsTerminateSystemThread(ndis_status);
238static ndis_status IoGetDeviceObjectPointer(unicode_string *,
239 uint32_t, void *, device_object *);
240static ndis_status IoGetDeviceProperty(device_object *, uint32_t,
241 uint32_t, void *, uint32_t *);
242static void KeInitializeMutex(kmutant *, uint32_t);
243static uint32_t KeReleaseMutex(kmutant *, uint8_t);
244static uint32_t KeReadStateMutex(kmutant *);
245static ndis_status ObReferenceObjectByHandle(ndis_handle,
246 uint32_t, void *, uint8_t, void **, void **);
247static void ObfDereferenceObject(void *);
248static uint32_t ZwClose(ndis_handle);
249static uint32_t WmiQueryTraceInformation(uint32_t, void *, uint32_t,
250 uint32_t, void *);
251static uint32_t WmiTraceMessage(uint64_t, uint32_t, void *, uint16_t, ...);
252static uint32_t IoWMIRegistrationControl(device_object *, uint32_t);
253static void *ntoskrnl_memset(void *, int, size_t);
254static void *ntoskrnl_memmove(void *, void *, size_t);
255static void *ntoskrnl_memchr(void *, unsigned char, size_t);
256static char *ntoskrnl_strstr(char *, char *);
257static char *ntoskrnl_strncat(char *, char *, size_t);
258static int ntoskrnl_toupper(int);
259static int ntoskrnl_tolower(int);
260static funcptr ntoskrnl_findwrap(funcptr);
261static uint32_t DbgPrint(char *, ...);
262static void DbgBreakPoint(void);
263static void KeBugCheckEx(uint32_t, u_long, u_long, u_long, u_long);
264static int32_t KeDelayExecutionThread(uint8_t, uint8_t, int64_t *);
265static int32_t KeSetPriorityThread(struct thread *, int32_t);
266static void dummy(void);
267
268static struct mtx ntoskrnl_dispatchlock;
269static struct mtx ntoskrnl_interlock;
270static kspin_lock ntoskrnl_cancellock;
271static int ntoskrnl_kth = 0;
272static struct nt_objref_head ntoskrnl_reflist;
273static uma_zone_t mdl_zone;
274static uma_zone_t iw_zone;
275static struct kdpc_queue *kq_queues;
276static struct kdpc_queue *wq_queues;
277static int wq_idx = 0;
278
279int
280ntoskrnl_libinit()
281{
282 image_patch_table *patch;
283 int error;
284 struct proc *p;
285 kdpc_queue *kq;
286 callout_entry *e;
287 int i;
288
289 mtx_init(&ntoskrnl_dispatchlock,
290 "ntoskrnl dispatch lock", MTX_NDIS_LOCK, MTX_DEF|MTX_RECURSE);
291 mtx_init(&ntoskrnl_interlock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
292 KeInitializeSpinLock(&ntoskrnl_cancellock);
293 KeInitializeSpinLock(&ntoskrnl_intlock);
294 TAILQ_INIT(&ntoskrnl_reflist);
295
296 InitializeListHead(&ntoskrnl_calllist);
297 InitializeListHead(&ntoskrnl_intlist);
298 mtx_init(&ntoskrnl_calllock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
299
300 kq_queues = ExAllocatePoolWithTag(NonPagedPool,
301#ifdef NTOSKRNL_MULTIPLE_DPCS
302 sizeof(kdpc_queue) * mp_ncpus, 0);
303#else
304 sizeof(kdpc_queue), 0);
305#endif
306
307 if (kq_queues == NULL)
308 return (ENOMEM);
309
310 wq_queues = ExAllocatePoolWithTag(NonPagedPool,
311 sizeof(kdpc_queue) * WORKITEM_THREADS, 0);
312
313 if (wq_queues == NULL)
314 return (ENOMEM);
315
316#ifdef NTOSKRNL_MULTIPLE_DPCS
317 bzero((char *)kq_queues, sizeof(kdpc_queue) * mp_ncpus);
318#else
319 bzero((char *)kq_queues, sizeof(kdpc_queue));
320#endif
321 bzero((char *)wq_queues, sizeof(kdpc_queue) * WORKITEM_THREADS);
322
323 /*
324 * Launch the DPC threads.
325 */
326
327#ifdef NTOSKRNL_MULTIPLE_DPCS
328 for (i = 0; i < mp_ncpus; i++) {
329#else
330 for (i = 0; i < 1; i++) {
331#endif
332 kq = kq_queues + i;
333 kq->kq_cpu = i;
334 error = kproc_create(ntoskrnl_dpc_thread, kq, &p,
335 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows DPC %d", i);
336 if (error)
337 panic("failed to launch DPC thread");
338 }
339
340 /*
341 * Launch the workitem threads.
342 */
343
344 for (i = 0; i < WORKITEM_THREADS; i++) {
345 kq = wq_queues + i;
346 error = kproc_create(ntoskrnl_workitem_thread, kq, &p,
347 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Workitem %d", i);
348 if (error)
349 panic("failed to launch workitem thread");
350 }
351
352 patch = ntoskrnl_functbl;
353 while (patch->ipt_func != NULL) {
354 windrv_wrap((funcptr)patch->ipt_func,
355 (funcptr *)&patch->ipt_wrap,
356 patch->ipt_argcnt, patch->ipt_ftype);
357 patch++;
358 }
359
360 for (i = 0; i < NTOSKRNL_TIMEOUTS; i++) {
361 e = ExAllocatePoolWithTag(NonPagedPool,
362 sizeof(callout_entry), 0);
363 if (e == NULL)
364 panic("failed to allocate timeouts");
365 mtx_lock_spin(&ntoskrnl_calllock);
366 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
367 mtx_unlock_spin(&ntoskrnl_calllock);
368 }
369
370 /*
371 * MDLs are supposed to be variable size (they describe
372 * buffers containing some number of pages, but we don't
373 * know ahead of time how many pages that will be). But
374 * always allocating them off the heap is very slow. As
375 * a compromise, we create an MDL UMA zone big enough to
376 * handle any buffer requiring up to 16 pages, and we
377 * use those for any MDLs for buffers of 16 pages or less
378 * in size. For buffers larger than that (which we assume
379 * will be few and far between, we allocate the MDLs off
380 * the heap.
381 */
382
383 mdl_zone = uma_zcreate("Windows MDL", MDL_ZONE_SIZE,
384 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
385
386 iw_zone = uma_zcreate("Windows WorkItem", sizeof(io_workitem),
387 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
388
389 return (0);
390}
391
392int
393ntoskrnl_libfini()
394{
395 image_patch_table *patch;
396 callout_entry *e;
397 list_entry *l;
398
399 patch = ntoskrnl_functbl;
400 while (patch->ipt_func != NULL) {
401 windrv_unwrap(patch->ipt_wrap);
402 patch++;
403 }
404
405 /* Stop the workitem queues. */
406 ntoskrnl_destroy_workitem_threads();
407 /* Stop the DPC queues. */
408 ntoskrnl_destroy_dpc_threads();
409
410 ExFreePool(kq_queues);
411 ExFreePool(wq_queues);
412
413 uma_zdestroy(mdl_zone);
414 uma_zdestroy(iw_zone);
415
416 mtx_lock_spin(&ntoskrnl_calllock);
417 while(!IsListEmpty(&ntoskrnl_calllist)) {
418 l = RemoveHeadList(&ntoskrnl_calllist);
419 e = CONTAINING_RECORD(l, callout_entry, ce_list);
420 mtx_unlock_spin(&ntoskrnl_calllock);
421 ExFreePool(e);
422 mtx_lock_spin(&ntoskrnl_calllock);
423 }
424 mtx_unlock_spin(&ntoskrnl_calllock);
425
426 mtx_destroy(&ntoskrnl_dispatchlock);
427 mtx_destroy(&ntoskrnl_interlock);
428 mtx_destroy(&ntoskrnl_calllock);
429
430 return (0);
431}
432
433/*
434 * We need to be able to reference this externally from the wrapper;
435 * GCC only generates a local implementation of memset.
436 */
437static void *
438ntoskrnl_memset(buf, ch, size)
439 void *buf;
440 int ch;
441 size_t size;
442{
443 return (memset(buf, ch, size));
444}
445
446static void *
447ntoskrnl_memmove(dst, src, size)
448 void *src;
449 void *dst;
450 size_t size;
451{
452 bcopy(src, dst, size);
453 return (dst);
454}
455
456static void *
457ntoskrnl_memchr(void *buf, unsigned char ch, size_t len)
458{
459 if (len != 0) {
460 unsigned char *p = buf;
461
462 do {
463 if (*p++ == ch)
464 return (p - 1);
465 } while (--len != 0);
466 }
467 return (NULL);
468}
469
470static char *
471ntoskrnl_strstr(s, find)
472 char *s, *find;
473{
474 char c, sc;
475 size_t len;
476
477 if ((c = *find++) != 0) {
478 len = strlen(find);
479 do {
480 do {
481 if ((sc = *s++) == 0)
482 return (NULL);
483 } while (sc != c);
484 } while (strncmp(s, find, len) != 0);
485 s--;
486 }
487 return ((char *)s);
488}
489
490/* Taken from libc */
491static char *
492ntoskrnl_strncat(dst, src, n)
493 char *dst;
494 char *src;
495 size_t n;
496{
497 if (n != 0) {
498 char *d = dst;
499 const char *s = src;
500
501 while (*d != 0)
502 d++;
503 do {
504 if ((*d = *s++) == 0)
505 break;
506 d++;
507 } while (--n != 0);
508 *d = 0;
509 }
510 return (dst);
511}
512
513static int
514ntoskrnl_toupper(c)
515 int c;
516{
517 return (toupper(c));
518}
519
520static int
521ntoskrnl_tolower(c)
522 int c;
523{
524 return (tolower(c));
525}
526
527static uint8_t
528RtlEqualUnicodeString(unicode_string *str1, unicode_string *str2,
529 uint8_t caseinsensitive)
530{
531 int i;
532
533 if (str1->us_len != str2->us_len)
534 return (FALSE);
535
536 for (i = 0; i < str1->us_len; i++) {
537 if (caseinsensitive == TRUE) {
538 if (toupper((char)(str1->us_buf[i] & 0xFF)) !=
539 toupper((char)(str2->us_buf[i] & 0xFF)))
540 return (FALSE);
541 } else {
542 if (str1->us_buf[i] != str2->us_buf[i])
543 return (FALSE);
544 }
545 }
546
547 return (TRUE);
548}
549
550static void
551RtlCopyString(dst, src)
552 ansi_string *dst;
553 const ansi_string *src;
554{
555 if (src != NULL && src->as_buf != NULL && dst->as_buf != NULL) {
556 dst->as_len = min(src->as_len, dst->as_maxlen);
557 memcpy(dst->as_buf, src->as_buf, dst->as_len);
558 if (dst->as_len < dst->as_maxlen)
559 dst->as_buf[dst->as_len] = 0;
560 } else
561 dst->as_len = 0;
562}
563
564static void
565RtlCopyUnicodeString(dest, src)
566 unicode_string *dest;
567 unicode_string *src;
568{
569
570 if (dest->us_maxlen >= src->us_len)
571 dest->us_len = src->us_len;
572 else
573 dest->us_len = dest->us_maxlen;
574 memcpy(dest->us_buf, src->us_buf, dest->us_len);
575}
576
577static void
578ntoskrnl_ascii_to_unicode(ascii, unicode, len)
579 char *ascii;
580 uint16_t *unicode;
581 int len;
582{
583 int i;
584 uint16_t *ustr;
585
586 ustr = unicode;
587 for (i = 0; i < len; i++) {
588 *ustr = (uint16_t)ascii[i];
589 ustr++;
590 }
591}
592
593static void
594ntoskrnl_unicode_to_ascii(unicode, ascii, len)
595 uint16_t *unicode;
596 char *ascii;
597 int len;
598{
599 int i;
600 uint8_t *astr;
601
602 astr = ascii;
603 for (i = 0; i < len / 2; i++) {
604 *astr = (uint8_t)unicode[i];
605 astr++;
606 }
607}
608
609uint32_t
610RtlUnicodeStringToAnsiString(ansi_string *dest, unicode_string *src, uint8_t allocate)
611{
612 if (dest == NULL || src == NULL)
613 return (STATUS_INVALID_PARAMETER);
614
615 dest->as_len = src->us_len / 2;
616 if (dest->as_maxlen < dest->as_len)
617 dest->as_len = dest->as_maxlen;
618
619 if (allocate == TRUE) {
620 dest->as_buf = ExAllocatePoolWithTag(NonPagedPool,
621 (src->us_len / 2) + 1, 0);
622 if (dest->as_buf == NULL)
623 return (STATUS_INSUFFICIENT_RESOURCES);
624 dest->as_len = dest->as_maxlen = src->us_len / 2;
625 } else {
626 dest->as_len = src->us_len / 2; /* XXX */
627 if (dest->as_maxlen < dest->as_len)
628 dest->as_len = dest->as_maxlen;
629 }
630
631 ntoskrnl_unicode_to_ascii(src->us_buf, dest->as_buf,
632 dest->as_len * 2);
633
634 return (STATUS_SUCCESS);
635}
636
637uint32_t
638RtlAnsiStringToUnicodeString(unicode_string *dest, ansi_string *src,
639 uint8_t allocate)
640{
641 if (dest == NULL || src == NULL)
642 return (STATUS_INVALID_PARAMETER);
643
644 if (allocate == TRUE) {
645 dest->us_buf = ExAllocatePoolWithTag(NonPagedPool,
646 src->as_len * 2, 0);
647 if (dest->us_buf == NULL)
648 return (STATUS_INSUFFICIENT_RESOURCES);
649 dest->us_len = dest->us_maxlen = strlen(src->as_buf) * 2;
650 } else {
651 dest->us_len = src->as_len * 2; /* XXX */
652 if (dest->us_maxlen < dest->us_len)
653 dest->us_len = dest->us_maxlen;
654 }
655
656 ntoskrnl_ascii_to_unicode(src->as_buf, dest->us_buf,
657 dest->us_len / 2);
658
659 return (STATUS_SUCCESS);
660}
661
662void *
663ExAllocatePoolWithTag(pooltype, len, tag)
664 uint32_t pooltype;
665 size_t len;
666 uint32_t tag;
667{
668 void *buf;
669
670 buf = malloc(len, M_DEVBUF, M_NOWAIT|M_ZERO);
671 if (buf == NULL)
672 return (NULL);
673
674 return (buf);
675}
676
677static void
678ExFreePoolWithTag(buf, tag)
679 void *buf;
680 uint32_t tag;
681{
682 ExFreePool(buf);
683}
684
685void
686ExFreePool(buf)
687 void *buf;
688{
689 free(buf, M_DEVBUF);
690}
691
692uint32_t
693IoAllocateDriverObjectExtension(drv, clid, extlen, ext)
694 driver_object *drv;
695 void *clid;
696 uint32_t extlen;
697 void **ext;
698{
699 custom_extension *ce;
700
701 ce = ExAllocatePoolWithTag(NonPagedPool, sizeof(custom_extension)
702 + extlen, 0);
703
704 if (ce == NULL)
705 return (STATUS_INSUFFICIENT_RESOURCES);
706
707 ce->ce_clid = clid;
708 InsertTailList((&drv->dro_driverext->dre_usrext), (&ce->ce_list));
709
710 *ext = (void *)(ce + 1);
711
712 return (STATUS_SUCCESS);
713}
714
715void *
716IoGetDriverObjectExtension(drv, clid)
717 driver_object *drv;
718 void *clid;
719{
720 list_entry *e;
721 custom_extension *ce;
722
723 /*
724 * Sanity check. Our dummy bus drivers don't have
725 * any driver extentions.
726 */
727
728 if (drv->dro_driverext == NULL)
729 return (NULL);
730
731 e = drv->dro_driverext->dre_usrext.nle_flink;
732 while (e != &drv->dro_driverext->dre_usrext) {
733 ce = (custom_extension *)e;
734 if (ce->ce_clid == clid)
735 return ((void *)(ce + 1));
736 e = e->nle_flink;
737 }
738
739 return (NULL);
740}
741
742
743uint32_t
744IoCreateDevice(driver_object *drv, uint32_t devextlen, unicode_string *devname,
745 uint32_t devtype, uint32_t devchars, uint8_t exclusive,
746 device_object **newdev)
747{
748 device_object *dev;
749
750 dev = ExAllocatePoolWithTag(NonPagedPool, sizeof(device_object), 0);
751 if (dev == NULL)
752 return (STATUS_INSUFFICIENT_RESOURCES);
753
754 dev->do_type = devtype;
755 dev->do_drvobj = drv;
756 dev->do_currirp = NULL;
757 dev->do_flags = 0;
758
759 if (devextlen) {
760 dev->do_devext = ExAllocatePoolWithTag(NonPagedPool,
761 devextlen, 0);
762
763 if (dev->do_devext == NULL) {
764 ExFreePool(dev);
765 return (STATUS_INSUFFICIENT_RESOURCES);
766 }
767
768 bzero(dev->do_devext, devextlen);
769 } else
770 dev->do_devext = NULL;
771
772 dev->do_size = sizeof(device_object) + devextlen;
773 dev->do_refcnt = 1;
774 dev->do_attacheddev = NULL;
775 dev->do_nextdev = NULL;
776 dev->do_devtype = devtype;
777 dev->do_stacksize = 1;
778 dev->do_alignreq = 1;
779 dev->do_characteristics = devchars;
780 dev->do_iotimer = NULL;
781 KeInitializeEvent(&dev->do_devlock, EVENT_TYPE_SYNC, TRUE);
782
783 /*
784 * Vpd is used for disk/tape devices,
785 * but we don't support those. (Yet.)
786 */
787 dev->do_vpb = NULL;
788
789 dev->do_devobj_ext = ExAllocatePoolWithTag(NonPagedPool,
790 sizeof(devobj_extension), 0);
791
792 if (dev->do_devobj_ext == NULL) {
793 if (dev->do_devext != NULL)
794 ExFreePool(dev->do_devext);
795 ExFreePool(dev);
796 return (STATUS_INSUFFICIENT_RESOURCES);
797 }
798
799 dev->do_devobj_ext->dve_type = 0;
800 dev->do_devobj_ext->dve_size = sizeof(devobj_extension);
801 dev->do_devobj_ext->dve_devobj = dev;
802
803 /*
804 * Attach this device to the driver object's list
805 * of devices. Note: this is not the same as attaching
806 * the device to the device stack. The driver's AddDevice
807 * routine must explicitly call IoAddDeviceToDeviceStack()
808 * to do that.
809 */
810
811 if (drv->dro_devobj == NULL) {
812 drv->dro_devobj = dev;
813 dev->do_nextdev = NULL;
814 } else {
815 dev->do_nextdev = drv->dro_devobj;
816 drv->dro_devobj = dev;
817 }
818
819 *newdev = dev;
820
821 return (STATUS_SUCCESS);
822}
823
824void
825IoDeleteDevice(dev)
826 device_object *dev;
827{
828 device_object *prev;
829
830 if (dev == NULL)
831 return;
832
833 if (dev->do_devobj_ext != NULL)
834 ExFreePool(dev->do_devobj_ext);
835
836 if (dev->do_devext != NULL)
837 ExFreePool(dev->do_devext);
838
839 /* Unlink the device from the driver's device list. */
840
841 prev = dev->do_drvobj->dro_devobj;
842 if (prev == dev)
843 dev->do_drvobj->dro_devobj = dev->do_nextdev;
844 else {
845 while (prev->do_nextdev != dev)
846 prev = prev->do_nextdev;
847 prev->do_nextdev = dev->do_nextdev;
848 }
849
850 ExFreePool(dev);
851}
852
853device_object *
854IoGetAttachedDevice(dev)
855 device_object *dev;
856{
857 device_object *d;
858
859 if (dev == NULL)
860 return (NULL);
861
862 d = dev;
863
864 while (d->do_attacheddev != NULL)
865 d = d->do_attacheddev;
866
867 return (d);
868}
869
870static irp *
871IoBuildSynchronousFsdRequest(func, dobj, buf, len, off, event, status)
872 uint32_t func;
873 device_object *dobj;
874 void *buf;
875 uint32_t len;
876 uint64_t *off;
877 nt_kevent *event;
878 io_status_block *status;
879{
880 irp *ip;
881
882 ip = IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status);
883 if (ip == NULL)
884 return (NULL);
885 ip->irp_usrevent = event;
886
887 return (ip);
888}
889
890static irp *
891IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status)
892 uint32_t func;
893 device_object *dobj;
894 void *buf;
895 uint32_t len;
896 uint64_t *off;
897 io_status_block *status;
898{
899 irp *ip;
900 io_stack_location *sl;
901
902 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
903 if (ip == NULL)
904 return (NULL);
905
906 ip->irp_usriostat = status;
907 ip->irp_tail.irp_overlay.irp_thread = NULL;
908
909 sl = IoGetNextIrpStackLocation(ip);
910 sl->isl_major = func;
911 sl->isl_minor = 0;
912 sl->isl_flags = 0;
913 sl->isl_ctl = 0;
914 sl->isl_devobj = dobj;
915 sl->isl_fileobj = NULL;
916 sl->isl_completionfunc = NULL;
917
918 ip->irp_userbuf = buf;
919
920 if (dobj->do_flags & DO_BUFFERED_IO) {
921 ip->irp_assoc.irp_sysbuf =
922 ExAllocatePoolWithTag(NonPagedPool, len, 0);
923 if (ip->irp_assoc.irp_sysbuf == NULL) {
924 IoFreeIrp(ip);
925 return (NULL);
926 }
927 bcopy(buf, ip->irp_assoc.irp_sysbuf, len);
928 }
929
930 if (dobj->do_flags & DO_DIRECT_IO) {
931 ip->irp_mdl = IoAllocateMdl(buf, len, FALSE, FALSE, ip);
932 if (ip->irp_mdl == NULL) {
933 if (ip->irp_assoc.irp_sysbuf != NULL)
934 ExFreePool(ip->irp_assoc.irp_sysbuf);
935 IoFreeIrp(ip);
936 return (NULL);
937 }
938 ip->irp_userbuf = NULL;
939 ip->irp_assoc.irp_sysbuf = NULL;
940 }
941
942 if (func == IRP_MJ_READ) {
943 sl->isl_parameters.isl_read.isl_len = len;
944 if (off != NULL)
945 sl->isl_parameters.isl_read.isl_byteoff = *off;
946 else
947 sl->isl_parameters.isl_read.isl_byteoff = 0;
948 }
949
950 if (func == IRP_MJ_WRITE) {
951 sl->isl_parameters.isl_write.isl_len = len;
952 if (off != NULL)
953 sl->isl_parameters.isl_write.isl_byteoff = *off;
954 else
955 sl->isl_parameters.isl_write.isl_byteoff = 0;
956 }
957
958 return (ip);
959}
960
961static irp *
962IoBuildDeviceIoControlRequest(uint32_t iocode, device_object *dobj, void *ibuf,
963 uint32_t ilen, void *obuf, uint32_t olen, uint8_t isinternal,
964 nt_kevent *event, io_status_block *status)
965{
966 irp *ip;
967 io_stack_location *sl;
968 uint32_t buflen;
969
970 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
971 if (ip == NULL)
972 return (NULL);
973 ip->irp_usrevent = event;
974 ip->irp_usriostat = status;
975 ip->irp_tail.irp_overlay.irp_thread = NULL;
976
977 sl = IoGetNextIrpStackLocation(ip);
978 sl->isl_major = isinternal == TRUE ?
979 IRP_MJ_INTERNAL_DEVICE_CONTROL : IRP_MJ_DEVICE_CONTROL;
980 sl->isl_minor = 0;
981 sl->isl_flags = 0;
982 sl->isl_ctl = 0;
983 sl->isl_devobj = dobj;
984 sl->isl_fileobj = NULL;
985 sl->isl_completionfunc = NULL;
986 sl->isl_parameters.isl_ioctl.isl_iocode = iocode;
987 sl->isl_parameters.isl_ioctl.isl_ibuflen = ilen;
988 sl->isl_parameters.isl_ioctl.isl_obuflen = olen;
989
990 switch(IO_METHOD(iocode)) {
991 case METHOD_BUFFERED:
992 if (ilen > olen)
993 buflen = ilen;
994 else
995 buflen = olen;
996 if (buflen) {
997 ip->irp_assoc.irp_sysbuf =
998 ExAllocatePoolWithTag(NonPagedPool, buflen, 0);
999 if (ip->irp_assoc.irp_sysbuf == NULL) {
1000 IoFreeIrp(ip);
1001 return (NULL);
1002 }
1003 }
1004 if (ilen && ibuf != NULL) {
1005 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1006 bzero((char *)ip->irp_assoc.irp_sysbuf + ilen,
1007 buflen - ilen);
1008 } else
1009 bzero(ip->irp_assoc.irp_sysbuf, ilen);
1010 ip->irp_userbuf = obuf;
1011 break;
1012 case METHOD_IN_DIRECT:
1013 case METHOD_OUT_DIRECT:
1014 if (ilen && ibuf != NULL) {
1015 ip->irp_assoc.irp_sysbuf =
1016 ExAllocatePoolWithTag(NonPagedPool, ilen, 0);
1017 if (ip->irp_assoc.irp_sysbuf == NULL) {
1018 IoFreeIrp(ip);
1019 return (NULL);
1020 }
1021 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1022 }
1023 if (olen && obuf != NULL) {
1024 ip->irp_mdl = IoAllocateMdl(obuf, olen,
1025 FALSE, FALSE, ip);
1026 /*
1027 * Normally we would MmProbeAndLockPages()
1028 * here, but we don't have to in our
1029 * imlementation.
1030 */
1031 }
1032 break;
1033 case METHOD_NEITHER:
1034 ip->irp_userbuf = obuf;
1035 sl->isl_parameters.isl_ioctl.isl_type3ibuf = ibuf;
1036 break;
1037 default:
1038 break;
1039 }
1040
1041 /*
1042 * Ideally, we should associate this IRP with the calling
1043 * thread here.
1044 */
1045
1046 return (ip);
1047}
1048
1049static irp *
1050IoAllocateIrp(uint8_t stsize, uint8_t chargequota)
1051{
1052 irp *i;
1053
1054 i = ExAllocatePoolWithTag(NonPagedPool, IoSizeOfIrp(stsize), 0);
1055 if (i == NULL)
1056 return (NULL);
1057
1058 IoInitializeIrp(i, IoSizeOfIrp(stsize), stsize);
1059
1060 return (i);
1061}
1062
1063static irp *
1064IoMakeAssociatedIrp(irp *ip, uint8_t stsize)
1065{
1066 irp *associrp;
1067
1068 associrp = IoAllocateIrp(stsize, FALSE);
1069 if (associrp == NULL)
1070 return (NULL);
1071
1072 mtx_lock(&ntoskrnl_dispatchlock);
1073 associrp->irp_flags |= IRP_ASSOCIATED_IRP;
1074 associrp->irp_tail.irp_overlay.irp_thread =
1075 ip->irp_tail.irp_overlay.irp_thread;
1076 associrp->irp_assoc.irp_master = ip;
1077 mtx_unlock(&ntoskrnl_dispatchlock);
1078
1079 return (associrp);
1080}
1081
1082static void
1083IoFreeIrp(ip)
1084 irp *ip;
1085{
1086 ExFreePool(ip);
1087}
1088
1089static void
1090IoInitializeIrp(irp *io, uint16_t psize, uint8_t ssize)
1091{
1092 bzero((char *)io, IoSizeOfIrp(ssize));
1093 io->irp_size = psize;
1094 io->irp_stackcnt = ssize;
1095 io->irp_currentstackloc = ssize;
1096 InitializeListHead(&io->irp_thlist);
1097 io->irp_tail.irp_overlay.irp_csl =
1098 (io_stack_location *)(io + 1) + ssize;
1099}
1100
1101static void
1102IoReuseIrp(ip, status)
1103 irp *ip;
1104 uint32_t status;
1105{
1106 uint8_t allocflags;
1107
1108 allocflags = ip->irp_allocflags;
1109 IoInitializeIrp(ip, ip->irp_size, ip->irp_stackcnt);
1110 ip->irp_iostat.isb_status = status;
1111 ip->irp_allocflags = allocflags;
1112}
1113
1114void
1115IoAcquireCancelSpinLock(uint8_t *irql)
1116{
1117 KeAcquireSpinLock(&ntoskrnl_cancellock, irql);
1118}
1119
1120void
1121IoReleaseCancelSpinLock(uint8_t irql)
1122{
1123 KeReleaseSpinLock(&ntoskrnl_cancellock, irql);
1124}
1125
1126uint8_t
1127IoCancelIrp(irp *ip)
1128{
1129 cancel_func cfunc;
1130 uint8_t cancelirql;
1131
1132 IoAcquireCancelSpinLock(&cancelirql);
1133 cfunc = IoSetCancelRoutine(ip, NULL);
1134 ip->irp_cancel = TRUE;
1135 if (cfunc == NULL) {
1136 IoReleaseCancelSpinLock(cancelirql);
1137 return (FALSE);
1138 }
1139 ip->irp_cancelirql = cancelirql;
1140 MSCALL2(cfunc, IoGetCurrentIrpStackLocation(ip)->isl_devobj, ip);
1141 return (uint8_t)IoSetCancelValue(ip, TRUE);
1142}
1143
1144uint32_t
1145IofCallDriver(dobj, ip)
1146 device_object *dobj;
1147 irp *ip;
1148{
1149 driver_object *drvobj;
1150 io_stack_location *sl;
1151 uint32_t status;
1152 driver_dispatch disp;
1153
1154 drvobj = dobj->do_drvobj;
1155
1156 if (ip->irp_currentstackloc <= 0)
1157 panic("IoCallDriver(): out of stack locations");
1158
1159 IoSetNextIrpStackLocation(ip);
1160 sl = IoGetCurrentIrpStackLocation(ip);
1161
1162 sl->isl_devobj = dobj;
1163
1164 disp = drvobj->dro_dispatch[sl->isl_major];
1165 status = MSCALL2(disp, dobj, ip);
1166
1167 return (status);
1168}
1169
1170void
1171IofCompleteRequest(irp *ip, uint8_t prioboost)
1172{
1173 uint32_t status;
1174 device_object *dobj;
1175 io_stack_location *sl;
1176 completion_func cf;
1177
1178 KASSERT(ip->irp_iostat.isb_status != STATUS_PENDING,
1179 ("incorrect IRP(%p) status (STATUS_PENDING)", ip));
1180
1181 sl = IoGetCurrentIrpStackLocation(ip);
1182 IoSkipCurrentIrpStackLocation(ip);
1183
1184 do {
1185 if (sl->isl_ctl & SL_PENDING_RETURNED)
1186 ip->irp_pendingreturned = TRUE;
1187
1188 if (ip->irp_currentstackloc != (ip->irp_stackcnt + 1))
1189 dobj = IoGetCurrentIrpStackLocation(ip)->isl_devobj;
1190 else
1191 dobj = NULL;
1192
1193 if (sl->isl_completionfunc != NULL &&
1194 ((ip->irp_iostat.isb_status == STATUS_SUCCESS &&
1195 sl->isl_ctl & SL_INVOKE_ON_SUCCESS) ||
1196 (ip->irp_iostat.isb_status != STATUS_SUCCESS &&
1197 sl->isl_ctl & SL_INVOKE_ON_ERROR) ||
1198 (ip->irp_cancel == TRUE &&
1199 sl->isl_ctl & SL_INVOKE_ON_CANCEL))) {
1200 cf = sl->isl_completionfunc;
1201 status = MSCALL3(cf, dobj, ip, sl->isl_completionctx);
1202 if (status == STATUS_MORE_PROCESSING_REQUIRED)
1203 return;
1204 } else {
1205 if ((ip->irp_currentstackloc <= ip->irp_stackcnt) &&
1206 (ip->irp_pendingreturned == TRUE))
1207 IoMarkIrpPending(ip);
1208 }
1209
1210 /* move to the next. */
1211 IoSkipCurrentIrpStackLocation(ip);
1212 sl++;
1213 } while (ip->irp_currentstackloc <= (ip->irp_stackcnt + 1));
1214
1215 if (ip->irp_usriostat != NULL)
1216 *ip->irp_usriostat = ip->irp_iostat;
1217 if (ip->irp_usrevent != NULL)
1218 KeSetEvent(ip->irp_usrevent, prioboost, FALSE);
1219
1220 /* Handle any associated IRPs. */
1221
1222 if (ip->irp_flags & IRP_ASSOCIATED_IRP) {
1223 uint32_t masterirpcnt;
1224 irp *masterirp;
1225 mdl *m;
1226
1227 masterirp = ip->irp_assoc.irp_master;
1228 masterirpcnt =
1229 InterlockedDecrement(&masterirp->irp_assoc.irp_irpcnt);
1230
1231 while ((m = ip->irp_mdl) != NULL) {
1232 ip->irp_mdl = m->mdl_next;
1233 IoFreeMdl(m);
1234 }
1235 IoFreeIrp(ip);
1236 if (masterirpcnt == 0)
1237 IoCompleteRequest(masterirp, IO_NO_INCREMENT);
1238 return;
1239 }
1240
1241 /* With any luck, these conditions will never arise. */
1242
1243 if (ip->irp_flags & IRP_PAGING_IO) {
1244 if (ip->irp_mdl != NULL)
1245 IoFreeMdl(ip->irp_mdl);
1246 IoFreeIrp(ip);
1247 }
1248}
1249
1250void
1251ntoskrnl_intr(arg)
1252 void *arg;
1253{
1254 kinterrupt *iobj;
1255 uint8_t irql;
1256 uint8_t claimed;
1257 list_entry *l;
1258
1259 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1260 l = ntoskrnl_intlist.nle_flink;
1261 while (l != &ntoskrnl_intlist) {
1262 iobj = CONTAINING_RECORD(l, kinterrupt, ki_list);
1263 claimed = MSCALL2(iobj->ki_svcfunc, iobj, iobj->ki_svcctx);
1264 if (claimed == TRUE)
1265 break;
1266 l = l->nle_flink;
1267 }
1268 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1269}
1270
1271uint8_t
1272KeAcquireInterruptSpinLock(iobj)
1273 kinterrupt *iobj;
1274{
1275 uint8_t irql;
1276 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1277 return (irql);
1278}
1279
1280void
1281KeReleaseInterruptSpinLock(kinterrupt *iobj, uint8_t irql)
1282{
1283 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1284}
1285
1286uint8_t
1287KeSynchronizeExecution(iobj, syncfunc, syncctx)
1288 kinterrupt *iobj;
1289 void *syncfunc;
1290 void *syncctx;
1291{
1292 uint8_t irql;
1293
1294 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1295 MSCALL1(syncfunc, syncctx);
1296 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1297
1298 return (TRUE);
1299}
1300
1301/*
1302 * IoConnectInterrupt() is passed only the interrupt vector and
1303 * irql that a device wants to use, but no device-specific tag
1304 * of any kind. This conflicts rather badly with FreeBSD's
1305 * bus_setup_intr(), which needs the device_t for the device
1306 * requesting interrupt delivery. In order to bypass this
1307 * inconsistency, we implement a second level of interrupt
1308 * dispatching on top of bus_setup_intr(). All devices use
1309 * ntoskrnl_intr() as their ISR, and any device requesting
1310 * interrupts will be registered with ntoskrnl_intr()'s interrupt
1311 * dispatch list. When an interrupt arrives, we walk the list
1312 * and invoke all the registered ISRs. This effectively makes all
1313 * interrupts shared, but it's the only way to duplicate the
1314 * semantics of IoConnectInterrupt() and IoDisconnectInterrupt() properly.
1315 */
1316
1317uint32_t
1318IoConnectInterrupt(kinterrupt **iobj, void *svcfunc, void *svcctx,
1319 kspin_lock *lock, uint32_t vector, uint8_t irql, uint8_t syncirql,
1320 uint8_t imode, uint8_t shared, uint32_t affinity, uint8_t savefloat)
1321{
1322 uint8_t curirql;
1323
1324 *iobj = ExAllocatePoolWithTag(NonPagedPool, sizeof(kinterrupt), 0);
1325 if (*iobj == NULL)
1326 return (STATUS_INSUFFICIENT_RESOURCES);
1327
1328 (*iobj)->ki_svcfunc = svcfunc;
1329 (*iobj)->ki_svcctx = svcctx;
1330
1331 if (lock == NULL) {
1332 KeInitializeSpinLock(&(*iobj)->ki_lock_priv);
1333 (*iobj)->ki_lock = &(*iobj)->ki_lock_priv;
1334 } else
1335 (*iobj)->ki_lock = lock;
1336
1337 KeAcquireSpinLock(&ntoskrnl_intlock, &curirql);
1338 InsertHeadList((&ntoskrnl_intlist), (&(*iobj)->ki_list));
1339 KeReleaseSpinLock(&ntoskrnl_intlock, curirql);
1340
1341 return (STATUS_SUCCESS);
1342}
1343
1344void
1345IoDisconnectInterrupt(iobj)
1346 kinterrupt *iobj;
1347{
1348 uint8_t irql;
1349
1350 if (iobj == NULL)
1351 return;
1352
1353 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1354 RemoveEntryList((&iobj->ki_list));
1355 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1356
1357 ExFreePool(iobj);
1358}
1359
1360device_object *
1361IoAttachDeviceToDeviceStack(src, dst)
1362 device_object *src;
1363 device_object *dst;
1364{
1365 device_object *attached;
1366
1367 mtx_lock(&ntoskrnl_dispatchlock);
1368 attached = IoGetAttachedDevice(dst);
1369 attached->do_attacheddev = src;
1370 src->do_attacheddev = NULL;
1371 src->do_stacksize = attached->do_stacksize + 1;
1372 mtx_unlock(&ntoskrnl_dispatchlock);
1373
1374 return (attached);
1375}
1376
1377void
1378IoDetachDevice(topdev)
1379 device_object *topdev;
1380{
1381 device_object *tail;
1382
1383 mtx_lock(&ntoskrnl_dispatchlock);
1384
1385 /* First, break the chain. */
1386 tail = topdev->do_attacheddev;
1387 if (tail == NULL) {
1388 mtx_unlock(&ntoskrnl_dispatchlock);
1389 return;
1390 }
1391 topdev->do_attacheddev = tail->do_attacheddev;
1392 topdev->do_refcnt--;
1393
1394 /* Now reduce the stacksize count for the takm_il objects. */
1395
1396 tail = topdev->do_attacheddev;
1397 while (tail != NULL) {
1398 tail->do_stacksize--;
1399 tail = tail->do_attacheddev;
1400 }
1401
1402 mtx_unlock(&ntoskrnl_dispatchlock);
1403}
1404
1405/*
1406 * For the most part, an object is considered signalled if
1407 * dh_sigstate == TRUE. The exception is for mutant objects
1408 * (mutexes), where the logic works like this:
1409 *
1410 * - If the thread already owns the object and sigstate is
1411 * less than or equal to 0, then the object is considered
1412 * signalled (recursive acquisition).
1413 * - If dh_sigstate == 1, the object is also considered
1414 * signalled.
1415 */
1416
1417static int
1418ntoskrnl_is_signalled(obj, td)
1419 nt_dispatch_header *obj;
1420 struct thread *td;
1421{
1422 kmutant *km;
1423
1424 if (obj->dh_type == DISP_TYPE_MUTANT) {
1425 km = (kmutant *)obj;
1426 if ((obj->dh_sigstate <= 0 && km->km_ownerthread == td) ||
1427 obj->dh_sigstate == 1)
1428 return (TRUE);
1429 return (FALSE);
1430 }
1431
1432 if (obj->dh_sigstate > 0)
1433 return (TRUE);
1434 return (FALSE);
1435}
1436
1437static void
1438ntoskrnl_satisfy_wait(obj, td)
1439 nt_dispatch_header *obj;
1440 struct thread *td;
1441{
1442 kmutant *km;
1443
1444 switch (obj->dh_type) {
1445 case DISP_TYPE_MUTANT:
1446 km = (struct kmutant *)obj;
1447 obj->dh_sigstate--;
1448 /*
1449 * If sigstate reaches 0, the mutex is now
1450 * non-signalled (the new thread owns it).
1451 */
1452 if (obj->dh_sigstate == 0) {
1453 km->km_ownerthread = td;
1454 if (km->km_abandoned == TRUE)
1455 km->km_abandoned = FALSE;
1456 }
1457 break;
1458 /* Synchronization objects get reset to unsignalled. */
1459 case DISP_TYPE_SYNCHRONIZATION_EVENT:
1460 case DISP_TYPE_SYNCHRONIZATION_TIMER:
1461 obj->dh_sigstate = 0;
1462 break;
1463 case DISP_TYPE_SEMAPHORE:
1464 obj->dh_sigstate--;
1465 break;
1466 default:
1467 break;
1468 }
1469}
1470
1471static void
1472ntoskrnl_satisfy_multiple_waits(wb)
1473 wait_block *wb;
1474{
1475 wait_block *cur;
1476 struct thread *td;
1477
1478 cur = wb;
1479 td = wb->wb_kthread;
1480
1481 do {
1482 ntoskrnl_satisfy_wait(wb->wb_object, td);
1483 cur->wb_awakened = TRUE;
1484 cur = cur->wb_next;
1485 } while (cur != wb);
1486}
1487
1488/* Always called with dispatcher lock held. */
1489static void
1490ntoskrnl_waittest(obj, increment)
1491 nt_dispatch_header *obj;
1492 uint32_t increment;
1493{
1494 wait_block *w, *next;
1495 list_entry *e;
1496 struct thread *td;
1497 wb_ext *we;
1498 int satisfied;
1499
1500 /*
1501 * Once an object has been signalled, we walk its list of
1502 * wait blocks. If a wait block can be awakened, then satisfy
1503 * waits as necessary and wake the thread.
1504 *
1505 * The rules work like this:
1506 *
1507 * If a wait block is marked as WAITTYPE_ANY, then
1508 * we can satisfy the wait conditions on the current
1509 * object and wake the thread right away. Satisfying
1510 * the wait also has the effect of breaking us out
1511 * of the search loop.
1512 *
1513 * If the object is marked as WAITTYLE_ALL, then the
1514 * wait block will be part of a circularly linked
1515 * list of wait blocks belonging to a waiting thread
1516 * that's sleeping in KeWaitForMultipleObjects(). In
1517 * order to wake the thread, all the objects in the
1518 * wait list must be in the signalled state. If they
1519 * are, we then satisfy all of them and wake the
1520 * thread.
1521 *
1522 */
1523
1524 e = obj->dh_waitlisthead.nle_flink;
1525
1526 while (e != &obj->dh_waitlisthead && obj->dh_sigstate > 0) {
1527 w = CONTAINING_RECORD(e, wait_block, wb_waitlist);
1528 we = w->wb_ext;
1529 td = we->we_td;
1530 satisfied = FALSE;
1531 if (w->wb_waittype == WAITTYPE_ANY) {
1532 /*
1533 * Thread can be awakened if
1534 * any wait is satisfied.
1535 */
1536 ntoskrnl_satisfy_wait(obj, td);
1537 satisfied = TRUE;
1538 w->wb_awakened = TRUE;
1539 } else {
1540 /*
1541 * Thread can only be woken up
1542 * if all waits are satisfied.
1543 * If the thread is waiting on multiple
1544 * objects, they should all be linked
1545 * through the wb_next pointers in the
1546 * wait blocks.
1547 */
1548 satisfied = TRUE;
1549 next = w->wb_next;
1550 while (next != w) {
1551 if (ntoskrnl_is_signalled(obj, td) == FALSE) {
1552 satisfied = FALSE;
1553 break;
1554 }
1555 next = next->wb_next;
1556 }
1557 ntoskrnl_satisfy_multiple_waits(w);
1558 }
1559
1560 if (satisfied == TRUE)
1561 cv_broadcastpri(&we->we_cv,
1562 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
1563 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
1564
1565 e = e->nle_flink;
1566 }
1567}
1568
1569/*
1570 * Return the number of 100 nanosecond intervals since
1571 * January 1, 1601. (?!?!)
1572 */
1573void
1574ntoskrnl_time(tval)
1575 uint64_t *tval;
1576{
1577 struct timespec ts;
1578
1579 nanotime(&ts);
1580 *tval = (uint64_t)ts.tv_nsec / 100 + (uint64_t)ts.tv_sec * 10000000 +
1581 11644473600 * 10000000; /* 100ns ticks from 1601 to 1970 */
1582}
1583
1584static void
1585KeQuerySystemTime(current_time)
1586 uint64_t *current_time;
1587{
1588 ntoskrnl_time(current_time);
1589}
1590
1591static uint32_t
1592KeTickCount(void)
1593{
1594 struct timeval tv;
1595 getmicrouptime(&tv);
1596 return tvtohz(&tv);
1597}
1598
1599
1600/*
1601 * KeWaitForSingleObject() is a tricky beast, because it can be used
1602 * with several different object types: semaphores, timers, events,
1603 * mutexes and threads. Semaphores don't appear very often, but the
1604 * other object types are quite common. KeWaitForSingleObject() is
1605 * what's normally used to acquire a mutex, and it can be used to
1606 * wait for a thread termination.
1607 *
1608 * The Windows NDIS API is implemented in terms of Windows kernel
1609 * primitives, and some of the object manipulation is duplicated in
1610 * NDIS. For example, NDIS has timers and events, which are actually
1611 * Windows kevents and ktimers. Now, you're supposed to only use the
1612 * NDIS variants of these objects within the confines of the NDIS API,
1613 * but there are some naughty developers out there who will use
1614 * KeWaitForSingleObject() on NDIS timer and event objects, so we
1615 * have to support that as well. Conseqently, our NDIS timer and event
1616 * code has to be closely tied into our ntoskrnl timer and event code,
1617 * just as it is in Windows.
1618 *
1619 * KeWaitForSingleObject() may do different things for different kinds
1620 * of objects:
1621 *
1622 * - For events, we check if the event has been signalled. If the
1623 * event is already in the signalled state, we just return immediately,
1624 * otherwise we wait for it to be set to the signalled state by someone
1625 * else calling KeSetEvent(). Events can be either synchronization or
1626 * notification events.
1627 *
1628 * - For timers, if the timer has already fired and the timer is in
1629 * the signalled state, we just return, otherwise we wait on the
1630 * timer. Unlike an event, timers get signalled automatically when
1631 * they expire rather than someone having to trip them manually.
1632 * Timers initialized with KeInitializeTimer() are always notification
1633 * events: KeInitializeTimerEx() lets you initialize a timer as
1634 * either a notification or synchronization event.
1635 *
1636 * - For mutexes, we try to acquire the mutex and if we can't, we wait
1637 * on the mutex until it's available and then grab it. When a mutex is
1638 * released, it enters the signalled state, which wakes up one of the
1639 * threads waiting to acquire it. Mutexes are always synchronization
1640 * events.
1641 *
1642 * - For threads, the only thing we do is wait until the thread object
1643 * enters a signalled state, which occurs when the thread terminates.
1644 * Threads are always notification events.
1645 *
1646 * A notification event wakes up all threads waiting on an object. A
1647 * synchronization event wakes up just one. Also, a synchronization event
1648 * is auto-clearing, which means we automatically set the event back to
1649 * the non-signalled state once the wakeup is done.
1650 */
1651
1652uint32_t
1653KeWaitForSingleObject(void *arg, uint32_t reason, uint32_t mode,
1654 uint8_t alertable, int64_t *duetime)
1655{
1656 wait_block w;
1657 struct thread *td = curthread;
1658 struct timeval tv;
1659 int error = 0;
1660 uint64_t curtime;
1661 wb_ext we;
1662 nt_dispatch_header *obj;
1663
1664 obj = arg;
1665
1666 if (obj == NULL)
1667 return (STATUS_INVALID_PARAMETER);
1668
1669 mtx_lock(&ntoskrnl_dispatchlock);
1670
1671 cv_init(&we.we_cv, "KeWFS");
1672 we.we_td = td;
1673
1674 /*
1675 * Check to see if this object is already signalled,
1676 * and just return without waiting if it is.
1677 */
1678 if (ntoskrnl_is_signalled(obj, td) == TRUE) {
1679 /* Sanity check the signal state value. */
1680 if (obj->dh_sigstate != INT32_MIN) {
1681 ntoskrnl_satisfy_wait(obj, curthread);
1682 mtx_unlock(&ntoskrnl_dispatchlock);
1683 return (STATUS_SUCCESS);
1684 } else {
1685 /*
1686 * There's a limit to how many times we can
1687 * recursively acquire a mutant. If we hit
1688 * the limit, something is very wrong.
1689 */
1690 if (obj->dh_type == DISP_TYPE_MUTANT) {
1691 mtx_unlock(&ntoskrnl_dispatchlock);
1692 panic("mutant limit exceeded");
1693 }
1694 }
1695 }
1696
1697 bzero((char *)&w, sizeof(wait_block));
1698 w.wb_object = obj;
1699 w.wb_ext = &we;
1700 w.wb_waittype = WAITTYPE_ANY;
1701 w.wb_next = &w;
1702 w.wb_waitkey = 0;
1703 w.wb_awakened = FALSE;
1704 w.wb_oldpri = td->td_priority;
1705
1706 InsertTailList((&obj->dh_waitlisthead), (&w.wb_waitlist));
1707
1708 /*
1709 * The timeout value is specified in 100 nanosecond units
1710 * and can be a positive or negative number. If it's positive,
1711 * then the duetime is absolute, and we need to convert it
1712 * to an absolute offset relative to now in order to use it.
1713 * If it's negative, then the duetime is relative and we
1714 * just have to convert the units.
1715 */
1716
1717 if (duetime != NULL) {
1718 if (*duetime < 0) {
1719 tv.tv_sec = - (*duetime) / 10000000;
1720 tv.tv_usec = (- (*duetime) / 10) -
1721 (tv.tv_sec * 1000000);
1722 } else {
1723 ntoskrnl_time(&curtime);
1724 if (*duetime < curtime)
1725 tv.tv_sec = tv.tv_usec = 0;
1726 else {
1727 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1728 tv.tv_usec = ((*duetime) - curtime) / 10 -
1729 (tv.tv_sec * 1000000);
1730 }
1731 }
1732 }
1733
1734 if (duetime == NULL)
1735 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1736 else
1737 error = cv_timedwait(&we.we_cv,
1738 &ntoskrnl_dispatchlock, tvtohz(&tv));
1739
1740 RemoveEntryList(&w.wb_waitlist);
1741
1742 cv_destroy(&we.we_cv);
1743
1744 /* We timed out. Leave the object alone and return status. */
1745
1746 if (error == EWOULDBLOCK) {
1747 mtx_unlock(&ntoskrnl_dispatchlock);
1748 return (STATUS_TIMEOUT);
1749 }
1750
1751 mtx_unlock(&ntoskrnl_dispatchlock);
1752
1753 return (STATUS_SUCCESS);
1754/*
1755 return (KeWaitForMultipleObjects(1, &obj, WAITTYPE_ALL, reason,
1756 mode, alertable, duetime, &w));
1757*/
1758}
1759
1760static uint32_t
1761KeWaitForMultipleObjects(uint32_t cnt, nt_dispatch_header *obj[], uint32_t wtype,
1762 uint32_t reason, uint32_t mode, uint8_t alertable, int64_t *duetime,
1763 wait_block *wb_array)
1764{
1765 struct thread *td = curthread;
1766 wait_block *whead, *w;
1767 wait_block _wb_array[MAX_WAIT_OBJECTS];
1768 nt_dispatch_header *cur;
1769 struct timeval tv;
1770 int i, wcnt = 0, error = 0;
1771 uint64_t curtime;
1772 struct timespec t1, t2;
1773 uint32_t status = STATUS_SUCCESS;
1774 wb_ext we;
1775
1776 if (cnt > MAX_WAIT_OBJECTS)
1777 return (STATUS_INVALID_PARAMETER);
1778 if (cnt > THREAD_WAIT_OBJECTS && wb_array == NULL)
1779 return (STATUS_INVALID_PARAMETER);
1780
1781 mtx_lock(&ntoskrnl_dispatchlock);
1782
1783 cv_init(&we.we_cv, "KeWFM");
1784 we.we_td = td;
1785
1786 if (wb_array == NULL)
1787 whead = _wb_array;
1788 else
1789 whead = wb_array;
1790
1791 bzero((char *)whead, sizeof(wait_block) * cnt);
1792
1793 /* First pass: see if we can satisfy any waits immediately. */
1794
1795 wcnt = 0;
1796 w = whead;
1797
1798 for (i = 0; i < cnt; i++) {
1799 InsertTailList((&obj[i]->dh_waitlisthead),
1800 (&w->wb_waitlist));
1801 w->wb_ext = &we;
1802 w->wb_object = obj[i];
1803 w->wb_waittype = wtype;
1804 w->wb_waitkey = i;
1805 w->wb_awakened = FALSE;
1806 w->wb_oldpri = td->td_priority;
1807 w->wb_next = w + 1;
1808 w++;
1809 wcnt++;
1810 if (ntoskrnl_is_signalled(obj[i], td)) {
1811 /*
1812 * There's a limit to how many times
1813 * we can recursively acquire a mutant.
1814 * If we hit the limit, something
1815 * is very wrong.
1816 */
1817 if (obj[i]->dh_sigstate == INT32_MIN &&
1818 obj[i]->dh_type == DISP_TYPE_MUTANT) {
1819 mtx_unlock(&ntoskrnl_dispatchlock);
1820 panic("mutant limit exceeded");
1821 }
1822
1823 /*
1824 * If this is a WAITTYPE_ANY wait, then
1825 * satisfy the waited object and exit
1826 * right now.
1827 */
1828
1829 if (wtype == WAITTYPE_ANY) {
1830 ntoskrnl_satisfy_wait(obj[i], td);
1831 status = STATUS_WAIT_0 + i;
1832 goto wait_done;
1833 } else {
1834 w--;
1835 wcnt--;
1836 w->wb_object = NULL;
1837 RemoveEntryList(&w->wb_waitlist);
1838 }
1839 }
1840 }
1841
1842 /*
1843 * If this is a WAITTYPE_ALL wait and all objects are
1844 * already signalled, satisfy the waits and exit now.
1845 */
1846
1847 if (wtype == WAITTYPE_ALL && wcnt == 0) {
1848 for (i = 0; i < cnt; i++)
1849 ntoskrnl_satisfy_wait(obj[i], td);
1850 status = STATUS_SUCCESS;
1851 goto wait_done;
1852 }
1853
1854 /*
1855 * Create a circular waitblock list. The waitcount
1856 * must always be non-zero when we get here.
1857 */
1858
1859 (w - 1)->wb_next = whead;
1860
1861 /* Wait on any objects that aren't yet signalled. */
1862
1863 /* Calculate timeout, if any. */
1864
1865 if (duetime != NULL) {
1866 if (*duetime < 0) {
1867 tv.tv_sec = - (*duetime) / 10000000;
1868 tv.tv_usec = (- (*duetime) / 10) -
1869 (tv.tv_sec * 1000000);
1870 } else {
1871 ntoskrnl_time(&curtime);
1872 if (*duetime < curtime)
1873 tv.tv_sec = tv.tv_usec = 0;
1874 else {
1875 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1876 tv.tv_usec = ((*duetime) - curtime) / 10 -
1877 (tv.tv_sec * 1000000);
1878 }
1879 }
1880 }
1881
1882 while (wcnt) {
1883 nanotime(&t1);
1884
1885 if (duetime == NULL)
1886 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1887 else
1888 error = cv_timedwait(&we.we_cv,
1889 &ntoskrnl_dispatchlock, tvtohz(&tv));
1890
1891 /* Wait with timeout expired. */
1892
1893 if (error) {
1894 status = STATUS_TIMEOUT;
1895 goto wait_done;
1896 }
1897
1898 nanotime(&t2);
1899
1900 /* See what's been signalled. */
1901
1902 w = whead;
1903 do {
1904 cur = w->wb_object;
1905 if (ntoskrnl_is_signalled(cur, td) == TRUE ||
1906 w->wb_awakened == TRUE) {
1907 /* Sanity check the signal state value. */
1908 if (cur->dh_sigstate == INT32_MIN &&
1909 cur->dh_type == DISP_TYPE_MUTANT) {
1910 mtx_unlock(&ntoskrnl_dispatchlock);
1911 panic("mutant limit exceeded");
1912 }
1913 wcnt--;
1914 if (wtype == WAITTYPE_ANY) {
1915 status = w->wb_waitkey &
1916 STATUS_WAIT_0;
1917 goto wait_done;
1918 }
1919 }
1920 w = w->wb_next;
1921 } while (w != whead);
1922
1923 /*
1924 * If all objects have been signalled, or if this
1925 * is a WAITTYPE_ANY wait and we were woke up by
1926 * someone, we can bail.
1927 */
1928
1929 if (wcnt == 0) {
1930 status = STATUS_SUCCESS;
1931 goto wait_done;
1932 }
1933
1934 /*
1935 * If this is WAITTYPE_ALL wait, and there's still
1936 * objects that haven't been signalled, deduct the
1937 * time that's elapsed so far from the timeout and
1938 * wait again (or continue waiting indefinitely if
1939 * there's no timeout).
1940 */
1941
1942 if (duetime != NULL) {
1943 tv.tv_sec -= (t2.tv_sec - t1.tv_sec);
1944 tv.tv_usec -= (t2.tv_nsec - t1.tv_nsec) / 1000;
1945 }
1946 }
1947
1948
1949wait_done:
1950
1951 cv_destroy(&we.we_cv);
1952
1953 for (i = 0; i < cnt; i++) {
1954 if (whead[i].wb_object != NULL)
1955 RemoveEntryList(&whead[i].wb_waitlist);
1956
1957 }
1958 mtx_unlock(&ntoskrnl_dispatchlock);
1959
1960 return (status);
1961}
1962
1963static void
1964WRITE_REGISTER_USHORT(uint16_t *reg, uint16_t val)
1965{
1966 bus_space_write_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1967}
1968
1969static uint16_t
1970READ_REGISTER_USHORT(reg)
1971 uint16_t *reg;
1972{
1973 return (bus_space_read_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1974}
1975
1976static void
1977WRITE_REGISTER_ULONG(reg, val)
1978 uint32_t *reg;
1979 uint32_t val;
1980{
1981 bus_space_write_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1982}
1983
1984static uint32_t
1985READ_REGISTER_ULONG(reg)
1986 uint32_t *reg;
1987{
1988 return (bus_space_read_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1989}
1990
1991static uint8_t
1992READ_REGISTER_UCHAR(uint8_t *reg)
1993{
1994 return (bus_space_read_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1995}
1996
1997static void
1998WRITE_REGISTER_UCHAR(uint8_t *reg, uint8_t val)
1999{
2000 bus_space_write_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
2001}
2002
2003static int64_t
2004_allmul(a, b)
2005 int64_t a;
2006 int64_t b;
2007{
2008 return (a * b);
2009}
2010
2011static int64_t
2012_alldiv(a, b)
2013 int64_t a;
2014 int64_t b;
2015{
2016 return (a / b);
2017}
2018
2019static int64_t
2020_allrem(a, b)
2021 int64_t a;
2022 int64_t b;
2023{
2024 return (a % b);
2025}
2026
2027static uint64_t
2028_aullmul(a, b)
2029 uint64_t a;
2030 uint64_t b;
2031{
2032 return (a * b);
2033}
2034
2035static uint64_t
2036_aulldiv(a, b)
2037 uint64_t a;
2038 uint64_t b;
2039{
2040 return (a / b);
2041}
2042
2043static uint64_t
2044_aullrem(a, b)
2045 uint64_t a;
2046 uint64_t b;
2047{
2048 return (a % b);
2049}
2050
2051static int64_t
2052_allshl(int64_t a, uint8_t b)
2053{
2054 return (a << b);
2055}
2056
2057static uint64_t
2058_aullshl(uint64_t a, uint8_t b)
2059{
2060 return (a << b);
2061}
2062
2063static int64_t
2064_allshr(int64_t a, uint8_t b)
2065{
2066 return (a >> b);
2067}
2068
2069static uint64_t
2070_aullshr(uint64_t a, uint8_t b)
2071{
2072 return (a >> b);
2073}
2074
2075static slist_entry *
2076ntoskrnl_pushsl(head, entry)
2077 slist_header *head;
2078 slist_entry *entry;
2079{
2080 slist_entry *oldhead;
2081
2082 oldhead = head->slh_list.slh_next;
2083 entry->sl_next = head->slh_list.slh_next;
2084 head->slh_list.slh_next = entry;
2085 head->slh_list.slh_depth++;
2086 head->slh_list.slh_seq++;
2087
2088 return (oldhead);
2089}
2090
2091static void
2092InitializeSListHead(head)
2093 slist_header *head;
2094{
2095 memset(head, 0, sizeof(*head));
2096}
2097
2098static slist_entry *
2099ntoskrnl_popsl(head)
2100 slist_header *head;
2101{
2102 slist_entry *first;
2103
2104 first = head->slh_list.slh_next;
2105 if (first != NULL) {
2106 head->slh_list.slh_next = first->sl_next;
2107 head->slh_list.slh_depth--;
2108 head->slh_list.slh_seq++;
2109 }
2110
2111 return (first);
2112}
2113
2114/*
2115 * We need this to make lookaside lists work for amd64.
2116 * We pass a pointer to ExAllocatePoolWithTag() the lookaside
2117 * list structure. For amd64 to work right, this has to be a
2118 * pointer to the wrapped version of the routine, not the
2119 * original. Letting the Windows driver invoke the original
2120 * function directly will result in a convention calling
2121 * mismatch and a pretty crash. On x86, this effectively
2122 * becomes a no-op since ipt_func and ipt_wrap are the same.
2123 */
2124
2125static funcptr
2126ntoskrnl_findwrap(func)
2127 funcptr func;
2128{
2129 image_patch_table *patch;
2130
2131 patch = ntoskrnl_functbl;
2132 while (patch->ipt_func != NULL) {
2133 if ((funcptr)patch->ipt_func == func)
2134 return ((funcptr)patch->ipt_wrap);
2135 patch++;
2136 }
2137
2138 return (NULL);
2139}
2140
2141static void
2142ExInitializePagedLookasideList(paged_lookaside_list *lookaside,
2143 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2144 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2145{
2146 bzero((char *)lookaside, sizeof(paged_lookaside_list));
2147
2148 if (size < sizeof(slist_entry))
2149 lookaside->nll_l.gl_size = sizeof(slist_entry);
2150 else
2151 lookaside->nll_l.gl_size = size;
2152 lookaside->nll_l.gl_tag = tag;
2153 if (allocfunc == NULL)
2154 lookaside->nll_l.gl_allocfunc =
2155 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2156 else
2157 lookaside->nll_l.gl_allocfunc = allocfunc;
2158
2159 if (freefunc == NULL)
2160 lookaside->nll_l.gl_freefunc =
2161 ntoskrnl_findwrap((funcptr)ExFreePool);
2162 else
2163 lookaside->nll_l.gl_freefunc = freefunc;
2164
2165#ifdef __i386__
2166 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2167#endif
2168
2169 lookaside->nll_l.gl_type = NonPagedPool;
2170 lookaside->nll_l.gl_depth = depth;
2171 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2172}
2173
2174static void
2175ExDeletePagedLookasideList(lookaside)
2176 paged_lookaside_list *lookaside;
2177{
2178 void *buf;
2179 void (*freefunc)(void *);
2180
2181 freefunc = lookaside->nll_l.gl_freefunc;
2182 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2183 MSCALL1(freefunc, buf);
2184}
2185
2186static void
2187ExInitializeNPagedLookasideList(npaged_lookaside_list *lookaside,
2188 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2189 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2190{
2191 bzero((char *)lookaside, sizeof(npaged_lookaside_list));
2192
2193 if (size < sizeof(slist_entry))
2194 lookaside->nll_l.gl_size = sizeof(slist_entry);
2195 else
2196 lookaside->nll_l.gl_size = size;
2197 lookaside->nll_l.gl_tag = tag;
2198 if (allocfunc == NULL)
2199 lookaside->nll_l.gl_allocfunc =
2200 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2201 else
2202 lookaside->nll_l.gl_allocfunc = allocfunc;
2203
2204 if (freefunc == NULL)
2205 lookaside->nll_l.gl_freefunc =
2206 ntoskrnl_findwrap((funcptr)ExFreePool);
2207 else
2208 lookaside->nll_l.gl_freefunc = freefunc;
2209
2210#ifdef __i386__
2211 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2212#endif
2213
2214 lookaside->nll_l.gl_type = NonPagedPool;
2215 lookaside->nll_l.gl_depth = depth;
2216 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2217}
2218
2219static void
2220ExDeleteNPagedLookasideList(lookaside)
2221 npaged_lookaside_list *lookaside;
2222{
2223 void *buf;
2224 void (*freefunc)(void *);
2225
2226 freefunc = lookaside->nll_l.gl_freefunc;
2227 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2228 MSCALL1(freefunc, buf);
2229}
2230
2231slist_entry *
2232InterlockedPushEntrySList(head, entry)
2233 slist_header *head;
2234 slist_entry *entry;
2235{
2236 slist_entry *oldhead;
2237
2238 mtx_lock_spin(&ntoskrnl_interlock);
2239 oldhead = ntoskrnl_pushsl(head, entry);
2240 mtx_unlock_spin(&ntoskrnl_interlock);
2241
2242 return (oldhead);
2243}
2244
2245slist_entry *
2246InterlockedPopEntrySList(head)
2247 slist_header *head;
2248{
2249 slist_entry *first;
2250
2251 mtx_lock_spin(&ntoskrnl_interlock);
2252 first = ntoskrnl_popsl(head);
2253 mtx_unlock_spin(&ntoskrnl_interlock);
2254
2255 return (first);
2256}
2257
2258static slist_entry *
2259ExInterlockedPushEntrySList(head, entry, lock)
2260 slist_header *head;
2261 slist_entry *entry;
2262 kspin_lock *lock;
2263{
2264 return (InterlockedPushEntrySList(head, entry));
2265}
2266
2267static slist_entry *
2268ExInterlockedPopEntrySList(head, lock)
2269 slist_header *head;
2270 kspin_lock *lock;
2271{
2272 return (InterlockedPopEntrySList(head));
2273}
2274
2275uint16_t
2276ExQueryDepthSList(head)
2277 slist_header *head;
2278{
2279 uint16_t depth;
2280
2281 mtx_lock_spin(&ntoskrnl_interlock);
2282 depth = head->slh_list.slh_depth;
2283 mtx_unlock_spin(&ntoskrnl_interlock);
2284
2285 return (depth);
2286}
2287
2288void
2289KeInitializeSpinLock(lock)
2290 kspin_lock *lock;
2291{
2292 *lock = 0;
2293}
2294
2295#ifdef __i386__
2296void
2297KefAcquireSpinLockAtDpcLevel(lock)
2298 kspin_lock *lock;
2299{
2300#ifdef NTOSKRNL_DEBUG_SPINLOCKS
2301 int i = 0;
2302#endif
2303
2304 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0) {
2305 /* sit and spin */;
2306#ifdef NTOSKRNL_DEBUG_SPINLOCKS
2307 i++;
2308 if (i > 200000000)
2309 panic("DEADLOCK!");
2310#endif
2311 }
2312}
2313
2314void
2315KefReleaseSpinLockFromDpcLevel(lock)
2316 kspin_lock *lock;
2317{
2318 atomic_store_rel_int((volatile u_int *)lock, 0);
2319}
2320
2321uint8_t
2322KeAcquireSpinLockRaiseToDpc(kspin_lock *lock)
2323{
2324 uint8_t oldirql;
2325
2326 if (KeGetCurrentIrql() > DISPATCH_LEVEL)
2327 panic("IRQL_NOT_LESS_THAN_OR_EQUAL");
2328
2329 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
2330 KeAcquireSpinLockAtDpcLevel(lock);
2331
2332 return (oldirql);
2333}
2334#else
2335void
2336KeAcquireSpinLockAtDpcLevel(kspin_lock *lock)
2337{
2338 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0)
2339 /* sit and spin */;
2340}
2341
2342void
2343KeReleaseSpinLockFromDpcLevel(kspin_lock *lock)
2344{
2345 atomic_store_rel_int((volatile u_int *)lock, 0);
2346}
2347#endif /* __i386__ */
2348
2349uintptr_t
2350InterlockedExchange(dst, val)
2351 volatile uint32_t *dst;
2352 uintptr_t val;
2353{
2354 uintptr_t r;
2355
2356 mtx_lock_spin(&ntoskrnl_interlock);
2357 r = *dst;
2358 *dst = val;
2359 mtx_unlock_spin(&ntoskrnl_interlock);
2360
2361 return (r);
2362}
2363
2364static uint32_t
2365InterlockedIncrement(addend)
2366 volatile uint32_t *addend;
2367{
2368 atomic_add_long((volatile u_long *)addend, 1);
2369 return (*addend);
2370}
2371
2372static uint32_t
2373InterlockedDecrement(addend)
2374 volatile uint32_t *addend;
2375{
2376 atomic_subtract_long((volatile u_long *)addend, 1);
2377 return (*addend);
2378}
2379
2380static void
2381ExInterlockedAddLargeStatistic(addend, inc)
2382 uint64_t *addend;
2383 uint32_t inc;
2384{
2385 mtx_lock_spin(&ntoskrnl_interlock);
2386 *addend += inc;
2387 mtx_unlock_spin(&ntoskrnl_interlock);
2388};
2389
2390mdl *
2391IoAllocateMdl(void *vaddr, uint32_t len, uint8_t secondarybuf,
2392 uint8_t chargequota, irp *iopkt)
2393{
2394 mdl *m;
2395 int zone = 0;
2396
2397 if (MmSizeOfMdl(vaddr, len) > MDL_ZONE_SIZE)
2398 m = ExAllocatePoolWithTag(NonPagedPool,
2399 MmSizeOfMdl(vaddr, len), 0);
2400 else {
2401 m = uma_zalloc(mdl_zone, M_NOWAIT | M_ZERO);
2402 zone++;
2403 }
2404
2405 if (m == NULL)
2406 return (NULL);
2407
2408 MmInitializeMdl(m, vaddr, len);
2409
2410 /*
2411 * MmInitializMdl() clears the flags field, so we
2412 * have to set this here. If the MDL came from the
2413 * MDL UMA zone, tag it so we can release it to
2414 * the right place later.
2415 */
2416 if (zone)
2417 m->mdl_flags = MDL_ZONE_ALLOCED;
2418
2419 if (iopkt != NULL) {
2420 if (secondarybuf == TRUE) {
2421 mdl *last;
2422 last = iopkt->irp_mdl;
2423 while (last->mdl_next != NULL)
2424 last = last->mdl_next;
2425 last->mdl_next = m;
2426 } else {
2427 if (iopkt->irp_mdl != NULL)
2428 panic("leaking an MDL in IoAllocateMdl()");
2429 iopkt->irp_mdl = m;
2430 }
2431 }
2432
2433 return (m);
2434}
2435
2436void
2437IoFreeMdl(m)
2438 mdl *m;
2439{
2440 if (m == NULL)
2441 return;
2442
2443 if (m->mdl_flags & MDL_ZONE_ALLOCED)
2444 uma_zfree(mdl_zone, m);
2445 else
2446 ExFreePool(m);
2447}
2448
2449static void *
2450MmAllocateContiguousMemory(size, highest)
2451 uint32_t size;
2452 uint64_t highest;
2453{
2454 void *addr;
2455 size_t pagelength = roundup(size, PAGE_SIZE);
2456
2457 addr = ExAllocatePoolWithTag(NonPagedPool, pagelength, 0);
2458
2459 return (addr);
2460}
2461
2462static void *
2463MmAllocateContiguousMemorySpecifyCache(size, lowest, highest,
2464 boundary, cachetype)
2465 uint32_t size;
2466 uint64_t lowest;
2467 uint64_t highest;
2468 uint64_t boundary;
2469 enum nt_caching_type cachetype;
2470{
2471 vm_memattr_t memattr;
2472 void *ret;
2473
2474 switch (cachetype) {
2475 case MmNonCached:
2476 memattr = VM_MEMATTR_UNCACHEABLE;
2477 break;
2478 case MmWriteCombined:
2479 memattr = VM_MEMATTR_WRITE_COMBINING;
2480 break;
2481 case MmNonCachedUnordered:
2482 memattr = VM_MEMATTR_UNCACHEABLE;
2483 break;
2484 case MmCached:
2485 case MmHardwareCoherentCached:
2486 case MmUSWCCached:
2487 default:
2488 memattr = VM_MEMATTR_DEFAULT;
2489 break;
2490 }
2491
2492 ret = (void *)kmem_alloc_contig(kernel_arena, size, M_ZERO | M_NOWAIT,
2493 lowest, highest, PAGE_SIZE, boundary, memattr);
2494 if (ret != NULL)
2495 malloc_type_allocated(M_DEVBUF, round_page(size));
2496 return (ret);
2497}
2498
2499static void
2500MmFreeContiguousMemory(base)
2501 void *base;
2502{
2503 ExFreePool(base);
2504}
2505
2506static void
2507MmFreeContiguousMemorySpecifyCache(base, size, cachetype)
2508 void *base;
2509 uint32_t size;
2510 enum nt_caching_type cachetype;
2511{
2512 contigfree(base, size, M_DEVBUF);
2513}
2514
2515static uint32_t
2516MmSizeOfMdl(vaddr, len)
2517 void *vaddr;
2518 size_t len;
2519{
2520 uint32_t l;
2521
2522 l = sizeof(struct mdl) +
2523 (sizeof(vm_offset_t *) * SPAN_PAGES(vaddr, len));
2524
2525 return (l);
2526}
2527
2528/*
2529 * The Microsoft documentation says this routine fills in the
2530 * page array of an MDL with the _physical_ page addresses that
2531 * comprise the buffer, but we don't really want to do that here.
2532 * Instead, we just fill in the page array with the kernel virtual
2533 * addresses of the buffers.
2534 */
2535void
2536MmBuildMdlForNonPagedPool(m)
2537 mdl *m;
2538{
2539 vm_offset_t *mdl_pages;
2540 int pagecnt, i;
2541
2542 pagecnt = SPAN_PAGES(m->mdl_byteoffset, m->mdl_bytecount);
2543
2544 if (pagecnt > (m->mdl_size - sizeof(mdl)) / sizeof(vm_offset_t *))
2545 panic("not enough pages in MDL to describe buffer");
2546
2547 mdl_pages = MmGetMdlPfnArray(m);
2548
2549 for (i = 0; i < pagecnt; i++)
2550 *mdl_pages = (vm_offset_t)m->mdl_startva + (i * PAGE_SIZE);
2551
2552 m->mdl_flags |= MDL_SOURCE_IS_NONPAGED_POOL;
2553 m->mdl_mappedsystemva = MmGetMdlVirtualAddress(m);
2554}
2555
2556static void *
2557MmMapLockedPages(mdl *buf, uint8_t accessmode)
2558{
2559 buf->mdl_flags |= MDL_MAPPED_TO_SYSTEM_VA;
2560 return (MmGetMdlVirtualAddress(buf));
2561}
2562
2563static void *
2564MmMapLockedPagesSpecifyCache(mdl *buf, uint8_t accessmode, uint32_t cachetype,
2565 void *vaddr, uint32_t bugcheck, uint32_t prio)
2566{
2567 return (MmMapLockedPages(buf, accessmode));
2568}
2569
2570static void
2571MmUnmapLockedPages(vaddr, buf)
2572 void *vaddr;
2573 mdl *buf;
2574{
2575 buf->mdl_flags &= ~MDL_MAPPED_TO_SYSTEM_VA;
2576}
2577
2578/*
2579 * This function has a problem in that it will break if you
2580 * compile this module without PAE and try to use it on a PAE
2581 * kernel. Unfortunately, there's no way around this at the
2582 * moment. It's slightly less broken that using pmap_kextract().
2583 * You'd think the virtual memory subsystem would help us out
2584 * here, but it doesn't.
2585 */
2586
2587static uint64_t
2588MmGetPhysicalAddress(void *base)
2589{
2590 return (pmap_extract(kernel_map->pmap, (vm_offset_t)base));
2591}
2592
2593void *
2594MmGetSystemRoutineAddress(ustr)
2595 unicode_string *ustr;
2596{
2597 ansi_string astr;
2598
2599 if (RtlUnicodeStringToAnsiString(&astr, ustr, TRUE))
2600 return (NULL);
2601 return (ndis_get_routine_address(ntoskrnl_functbl, astr.as_buf));
2602}
2603
2604uint8_t
2605MmIsAddressValid(vaddr)
2606 void *vaddr;
2607{
2608 if (pmap_extract(kernel_map->pmap, (vm_offset_t)vaddr))
2609 return (TRUE);
2610
2611 return (FALSE);
2612}
2613
2614void *
2615MmMapIoSpace(paddr, len, cachetype)
2616 uint64_t paddr;
2617 uint32_t len;
2618 uint32_t cachetype;
2619{
2620 devclass_t nexus_class;
2621 device_t *nexus_devs, devp;
2622 int nexus_count = 0;
2623 device_t matching_dev = NULL;
2624 struct resource *res;
2625 int i;
2626 vm_offset_t v;
2627
2628 /* There will always be at least one nexus. */
2629
2630 nexus_class = devclass_find("nexus");
2631 devclass_get_devices(nexus_class, &nexus_devs, &nexus_count);
2632
2633 for (i = 0; i < nexus_count; i++) {
2634 devp = nexus_devs[i];
2635 matching_dev = ntoskrnl_finddev(devp, paddr, &res);
2636 if (matching_dev)
2637 break;
2638 }
2639
2640 free(nexus_devs, M_TEMP);
2641
2642 if (matching_dev == NULL)
2643 return (NULL);
2644
2645 v = (vm_offset_t)rman_get_virtual(res);
2646 if (paddr > rman_get_start(res))
2647 v += paddr - rman_get_start(res);
2648
2649 return ((void *)v);
2650}
2651
2652void
2653MmUnmapIoSpace(vaddr, len)
2654 void *vaddr;
2655 size_t len;
2656{
2657}
2658
2659
2660static device_t
2661ntoskrnl_finddev(dev, paddr, res)
2662 device_t dev;
2663 uint64_t paddr;
2664 struct resource **res;
2665{
2666 device_t *children = NULL;
2667 device_t matching_dev;
2668 int childcnt;
2669 struct resource *r;
2670 struct resource_list *rl;
2671 struct resource_list_entry *rle;
2672 uint32_t flags;
2673 int i;
2674
2675 /* We only want devices that have been successfully probed. */
2676
2677 if (device_is_alive(dev) == FALSE)
2678 return (NULL);
2679
2680 rl = BUS_GET_RESOURCE_LIST(device_get_parent(dev), dev);
2681 if (rl != NULL) {
2682 STAILQ_FOREACH(rle, rl, link) {
2683 r = rle->res;
2684
2685 if (r == NULL)
2686 continue;
2687
2688 flags = rman_get_flags(r);
2689
2690 if (rle->type == SYS_RES_MEMORY &&
2691 paddr >= rman_get_start(r) &&
2692 paddr <= rman_get_end(r)) {
2693 if (!(flags & RF_ACTIVE))
2694 bus_activate_resource(dev,
2695 SYS_RES_MEMORY, 0, r);
2696 *res = r;
2697 return (dev);
2698 }
2699 }
2700 }
2701
2702 /*
2703 * If this device has children, do another
2704 * level of recursion to inspect them.
2705 */
2706
2707 device_get_children(dev, &children, &childcnt);
2708
2709 for (i = 0; i < childcnt; i++) {
2710 matching_dev = ntoskrnl_finddev(children[i], paddr, res);
2711 if (matching_dev != NULL) {
2712 free(children, M_TEMP);
2713 return (matching_dev);
2714 }
2715 }
2716
2717
2718 /* Won't somebody please think of the children! */
2719
2720 if (children != NULL)
2721 free(children, M_TEMP);
2722
2723 return (NULL);
2724}
2725
2726/*
2727 * Workitems are unlike DPCs, in that they run in a user-mode thread
2728 * context rather than at DISPATCH_LEVEL in kernel context. In our
2729 * case we run them in kernel context anyway.
2730 */
2731static void
2732ntoskrnl_workitem_thread(arg)
2733 void *arg;
2734{
2735 kdpc_queue *kq;
2736 list_entry *l;
2737 io_workitem *iw;
2738 uint8_t irql;
2739
2740 kq = arg;
2741
2742 InitializeListHead(&kq->kq_disp);
2743 kq->kq_td = curthread;
2744 kq->kq_exit = 0;
2745 KeInitializeSpinLock(&kq->kq_lock);
2746 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
2747
2748 while (1) {
2749 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
2750
2751 KeAcquireSpinLock(&kq->kq_lock, &irql);
2752
2753 if (kq->kq_exit) {
2754 kq->kq_exit = 0;
2755 KeReleaseSpinLock(&kq->kq_lock, irql);
2756 break;
2757 }
2758
2759 while (!IsListEmpty(&kq->kq_disp)) {
2760 l = RemoveHeadList(&kq->kq_disp);
2761 iw = CONTAINING_RECORD(l,
2762 io_workitem, iw_listentry);
2763 InitializeListHead((&iw->iw_listentry));
2764 if (iw->iw_func == NULL)
2765 continue;
2766 KeReleaseSpinLock(&kq->kq_lock, irql);
2767 MSCALL2(iw->iw_func, iw->iw_dobj, iw->iw_ctx);
2768 KeAcquireSpinLock(&kq->kq_lock, &irql);
2769 }
2770
2771 KeReleaseSpinLock(&kq->kq_lock, irql);
2772 }
2773
2774 kproc_exit(0);
2775 return; /* notreached */
2776}
2777
2778static ndis_status
2779RtlCharToInteger(src, base, val)
2780 const char *src;
2781 uint32_t base;
2782 uint32_t *val;
2783{
2784 int negative = 0;
2785 uint32_t res;
2786
2787 if (!src || !val)
2788 return (STATUS_ACCESS_VIOLATION);
2789 while (*src != '\0' && *src <= ' ')
2790 src++;
2791 if (*src == '+')
2792 src++;
2793 else if (*src == '-') {
2794 src++;
2795 negative = 1;
2796 }
2797 if (base == 0) {
2798 base = 10;
2799 if (*src == '0') {
2800 src++;
2801 if (*src == 'b') {
2802 base = 2;
2803 src++;
2804 } else if (*src == 'o') {
2805 base = 8;
2806 src++;
2807 } else if (*src == 'x') {
2808 base = 16;
2809 src++;
2810 }
2811 }
2812 } else if (!(base == 2 || base == 8 || base == 10 || base == 16))
2813 return (STATUS_INVALID_PARAMETER);
2814
2815 for (res = 0; *src; src++) {
2816 int v;
2817 if (isdigit(*src))
2818 v = *src - '0';
2819 else if (isxdigit(*src))
2820 v = tolower(*src) - 'a' + 10;
2821 else
2822 v = base;
2823 if (v >= base)
2824 return (STATUS_INVALID_PARAMETER);
2825 res = res * base + v;
2826 }
2827 *val = negative ? -res : res;
2828 return (STATUS_SUCCESS);
2829}
2830
2831static void
2832ntoskrnl_destroy_workitem_threads(void)
2833{
2834 kdpc_queue *kq;
2835 int i;
2836
2837 for (i = 0; i < WORKITEM_THREADS; i++) {
2838 kq = wq_queues + i;
2839 kq->kq_exit = 1;
2840 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2841 while (kq->kq_exit)
2842 tsleep(kq->kq_td->td_proc, PWAIT, "waitiw", hz/10);
2843 }
2844}
2845
2846io_workitem *
2847IoAllocateWorkItem(dobj)
2848 device_object *dobj;
2849{
2850 io_workitem *iw;
2851
2852 iw = uma_zalloc(iw_zone, M_NOWAIT);
2853 if (iw == NULL)
2854 return (NULL);
2855
2856 InitializeListHead(&iw->iw_listentry);
2857 iw->iw_dobj = dobj;
2858
2859 mtx_lock(&ntoskrnl_dispatchlock);
2860 iw->iw_idx = wq_idx;
2861 WORKIDX_INC(wq_idx);
2862 mtx_unlock(&ntoskrnl_dispatchlock);
2863
2864 return (iw);
2865}
2866
2867void
2868IoFreeWorkItem(iw)
2869 io_workitem *iw;
2870{
2871 uma_zfree(iw_zone, iw);
2872}
2873
2874void
2875IoQueueWorkItem(iw, iw_func, qtype, ctx)
2876 io_workitem *iw;
2877 io_workitem_func iw_func;
2878 uint32_t qtype;
2879 void *ctx;
2880{
2881 kdpc_queue *kq;
2882 list_entry *l;
2883 io_workitem *cur;
2884 uint8_t irql;
2885
2886 kq = wq_queues + iw->iw_idx;
2887
2888 KeAcquireSpinLock(&kq->kq_lock, &irql);
2889
2890 /*
2891 * Traverse the list and make sure this workitem hasn't
2892 * already been inserted. Queuing the same workitem
2893 * twice will hose the list but good.
2894 */
2895
2896 l = kq->kq_disp.nle_flink;
2897 while (l != &kq->kq_disp) {
2898 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2899 if (cur == iw) {
2900 /* Already queued -- do nothing. */
2901 KeReleaseSpinLock(&kq->kq_lock, irql);
2902 return;
2903 }
2904 l = l->nle_flink;
2905 }
2906
2907 iw->iw_func = iw_func;
2908 iw->iw_ctx = ctx;
2909
2910 InsertTailList((&kq->kq_disp), (&iw->iw_listentry));
2911 KeReleaseSpinLock(&kq->kq_lock, irql);
2912
2913 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2914}
2915
2916static void
2917ntoskrnl_workitem(dobj, arg)
2918 device_object *dobj;
2919 void *arg;
2920{
2921 io_workitem *iw;
2922 work_queue_item *w;
2923 work_item_func f;
2924
2925 iw = arg;
2926 w = (work_queue_item *)dobj;
2927 f = (work_item_func)w->wqi_func;
2928 uma_zfree(iw_zone, iw);
2929 MSCALL2(f, w, w->wqi_ctx);
2930}
2931
2932/*
2933 * The ExQueueWorkItem() API is deprecated in Windows XP. Microsoft
2934 * warns that it's unsafe and to use IoQueueWorkItem() instead. The
2935 * problem with ExQueueWorkItem() is that it can't guard against
2936 * the condition where a driver submits a job to the work queue and
2937 * is then unloaded before the job is able to run. IoQueueWorkItem()
2938 * acquires a reference to the device's device_object via the
2939 * object manager and retains it until after the job has completed,
2940 * which prevents the driver from being unloaded before the job
2941 * runs. (We don't currently support this behavior, though hopefully
2942 * that will change once the object manager API is fleshed out a bit.)
2943 *
2944 * Having said all that, the ExQueueWorkItem() API remains, because
2945 * there are still other parts of Windows that use it, including
2946 * NDIS itself: NdisScheduleWorkItem() calls ExQueueWorkItem().
2947 * We fake up the ExQueueWorkItem() API on top of our implementation
2948 * of IoQueueWorkItem(). Workitem thread #3 is reserved exclusively
2949 * for ExQueueWorkItem() jobs, and we pass a pointer to the work
2950 * queue item (provided by the caller) in to IoAllocateWorkItem()
2951 * instead of the device_object. We need to save this pointer so
2952 * we can apply a sanity check: as with the DPC queue and other
2953 * workitem queues, we can't allow the same work queue item to
2954 * be queued twice. If it's already pending, we silently return
2955 */
2956
2957void
2958ExQueueWorkItem(w, qtype)
2959 work_queue_item *w;
2960 uint32_t qtype;
2961{
2962 io_workitem *iw;
2963 io_workitem_func iwf;
2964 kdpc_queue *kq;
2965 list_entry *l;
2966 io_workitem *cur;
2967 uint8_t irql;
2968
2969
2970 /*
2971 * We need to do a special sanity test to make sure
2972 * the ExQueueWorkItem() API isn't used to queue
2973 * the same workitem twice. Rather than checking the
2974 * io_workitem pointer itself, we test the attached
2975 * device object, which is really a pointer to the
2976 * legacy work queue item structure.
2977 */
2978
2979 kq = wq_queues + WORKITEM_LEGACY_THREAD;
2980 KeAcquireSpinLock(&kq->kq_lock, &irql);
2981 l = kq->kq_disp.nle_flink;
2982 while (l != &kq->kq_disp) {
2983 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2984 if (cur->iw_dobj == (device_object *)w) {
2985 /* Already queued -- do nothing. */
2986 KeReleaseSpinLock(&kq->kq_lock, irql);
2987 return;
2988 }
2989 l = l->nle_flink;
2990 }
2991 KeReleaseSpinLock(&kq->kq_lock, irql);
2992
2993 iw = IoAllocateWorkItem((device_object *)w);
2994 if (iw == NULL)
2995 return;
2996
2997 iw->iw_idx = WORKITEM_LEGACY_THREAD;
2998 iwf = (io_workitem_func)ntoskrnl_findwrap((funcptr)ntoskrnl_workitem);
2999 IoQueueWorkItem(iw, iwf, qtype, iw);
3000}
3001
3002static void
3003RtlZeroMemory(dst, len)
3004 void *dst;
3005 size_t len;
3006{
3007 bzero(dst, len);
3008}
3009
3010static void
3011RtlSecureZeroMemory(dst, len)
3012 void *dst;
3013 size_t len;
3014{
3015 memset(dst, 0, len);
3016}
3017
3018static void
3019RtlFillMemory(void *dst, size_t len, uint8_t c)
3020{
3021 memset(dst, c, len);
3022}
3023
3024static void
3025RtlMoveMemory(dst, src, len)
3026 void *dst;
3027 const void *src;
3028 size_t len;
3029{
3030 memmove(dst, src, len);
3031}
3032
3033static void
3034RtlCopyMemory(dst, src, len)
3035 void *dst;
3036 const void *src;
3037 size_t len;
3038{
3039 bcopy(src, dst, len);
3040}
3041
3042static size_t
3043RtlCompareMemory(s1, s2, len)
3044 const void *s1;
3045 const void *s2;
3046 size_t len;
3047{
3048 size_t i;
3049 uint8_t *m1, *m2;
3050
3051 m1 = __DECONST(char *, s1);
3052 m2 = __DECONST(char *, s2);
3053
3054 for (i = 0; i < len && m1[i] == m2[i]; i++);
3055 return (i);
3056}
3057
3058void
3059RtlInitAnsiString(dst, src)
3060 ansi_string *dst;
3061 char *src;
3062{
3063 ansi_string *a;
3064
3065 a = dst;
3066 if (a == NULL)
3067 return;
3068 if (src == NULL) {
3069 a->as_len = a->as_maxlen = 0;
3070 a->as_buf = NULL;
3071 } else {
3072 a->as_buf = src;
3073 a->as_len = a->as_maxlen = strlen(src);
3074 }
3075}
3076
3077void
3078RtlInitUnicodeString(dst, src)
3079 unicode_string *dst;
3080 uint16_t *src;
3081{
3082 unicode_string *u;
3083 int i;
3084
3085 u = dst;
3086 if (u == NULL)
3087 return;
3088 if (src == NULL) {
3089 u->us_len = u->us_maxlen = 0;
3090 u->us_buf = NULL;
3091 } else {
3092 i = 0;
3093 while(src[i] != 0)
3094 i++;
3095 u->us_buf = src;
3096 u->us_len = u->us_maxlen = i * 2;
3097 }
3098}
3099
3100ndis_status
3101RtlUnicodeStringToInteger(ustr, base, val)
3102 unicode_string *ustr;
3103 uint32_t base;
3104 uint32_t *val;
3105{
3106 uint16_t *uchr;
3107 int len, neg = 0;
3108 char abuf[64];
3109 char *astr;
3110
3111 uchr = ustr->us_buf;
3112 len = ustr->us_len;
3113 bzero(abuf, sizeof(abuf));
3114
3115 if ((char)((*uchr) & 0xFF) == '-') {
3116 neg = 1;
3117 uchr++;
3118 len -= 2;
3119 } else if ((char)((*uchr) & 0xFF) == '+') {
3120 neg = 0;
3121 uchr++;
3122 len -= 2;
3123 }
3124
3125 if (base == 0) {
3126 if ((char)((*uchr) & 0xFF) == 'b') {
3127 base = 2;
3128 uchr++;
3129 len -= 2;
3130 } else if ((char)((*uchr) & 0xFF) == 'o') {
3131 base = 8;
3132 uchr++;
3133 len -= 2;
3134 } else if ((char)((*uchr) & 0xFF) == 'x') {
3135 base = 16;
3136 uchr++;
3137 len -= 2;
3138 } else
3139 base = 10;
3140 }
3141
3142 astr = abuf;
3143 if (neg) {
3144 strcpy(astr, "-");
3145 astr++;
3146 }
3147
3148 ntoskrnl_unicode_to_ascii(uchr, astr, len);
3149 *val = strtoul(abuf, NULL, base);
3150
3151 return (STATUS_SUCCESS);
3152}
3153
3154void
3155RtlFreeUnicodeString(ustr)
3156 unicode_string *ustr;
3157{
3158 if (ustr->us_buf == NULL)
3159 return;
3160 ExFreePool(ustr->us_buf);
3161 ustr->us_buf = NULL;
3162}
3163
3164void
3165RtlFreeAnsiString(astr)
3166 ansi_string *astr;
3167{
3168 if (astr->as_buf == NULL)
3169 return;
3170 ExFreePool(astr->as_buf);
3171 astr->as_buf = NULL;
3172}
3173
3174static int
3175atoi(str)
3176 const char *str;
3177{
3178 return (int)strtol(str, (char **)NULL, 10);
3179}
3180
3181static long
3182atol(str)
3183 const char *str;
3184{
3185 return strtol(str, (char **)NULL, 10);
3186}
3187
3188static int
3189rand(void)
3190{
3191 struct timeval tv;
3192
3193 microtime(&tv);
3194 srandom(tv.tv_usec);
3195 return ((int)random());
3196}
3197
3198static void
3199srand(seed)
3200 unsigned int seed;
3201{
3202 srandom(seed);
3203}
3204
3205static uint8_t
3206IoIsWdmVersionAvailable(uint8_t major, uint8_t minor)
3207{
3208 if (major == WDM_MAJOR && minor == WDM_MINOR_WINXP)
3209 return (TRUE);
3210 return (FALSE);
3211}
3212
3213static int32_t
3214IoOpenDeviceRegistryKey(struct device_object *devobj, uint32_t type,
3215 uint32_t mask, void **key)
3216{
3217 return (NDIS_STATUS_INVALID_DEVICE_REQUEST);
3218}
3219
3220static ndis_status
3221IoGetDeviceObjectPointer(name, reqaccess, fileobj, devobj)
3222 unicode_string *name;
3223 uint32_t reqaccess;
3224 void *fileobj;
3225 device_object *devobj;
3226{
3227 return (STATUS_SUCCESS);
3228}
3229
3230static ndis_status
3231IoGetDeviceProperty(devobj, regprop, buflen, prop, reslen)
3232 device_object *devobj;
3233 uint32_t regprop;
3234 uint32_t buflen;
3235 void *prop;
3236 uint32_t *reslen;
3237{
3238 driver_object *drv;
3239 uint16_t **name;
3240
3241 drv = devobj->do_drvobj;
3242
3243 switch (regprop) {
3244 case DEVPROP_DRIVER_KEYNAME:
3245 name = prop;
3246 *name = drv->dro_drivername.us_buf;
3247 *reslen = drv->dro_drivername.us_len;
3248 break;
3249 default:
3250 return (STATUS_INVALID_PARAMETER_2);
3251 break;
3252 }
3253
3254 return (STATUS_SUCCESS);
3255}
3256
3257static void
3258KeInitializeMutex(kmutex, level)
3259 kmutant *kmutex;
3260 uint32_t level;
3261{
3262 InitializeListHead((&kmutex->km_header.dh_waitlisthead));
3263 kmutex->km_abandoned = FALSE;
3264 kmutex->km_apcdisable = 1;
3265 kmutex->km_header.dh_sigstate = 1;
3266 kmutex->km_header.dh_type = DISP_TYPE_MUTANT;
3267 kmutex->km_header.dh_size = sizeof(kmutant) / sizeof(uint32_t);
3268 kmutex->km_ownerthread = NULL;
3269}
3270
3271static uint32_t
3272KeReleaseMutex(kmutant *kmutex, uint8_t kwait)
3273{
3274 uint32_t prevstate;
3275
3276 mtx_lock(&ntoskrnl_dispatchlock);
3277 prevstate = kmutex->km_header.dh_sigstate;
3278 if (kmutex->km_ownerthread != curthread) {
3279 mtx_unlock(&ntoskrnl_dispatchlock);
3280 return (STATUS_MUTANT_NOT_OWNED);
3281 }
3282
3283 kmutex->km_header.dh_sigstate++;
3284 kmutex->km_abandoned = FALSE;
3285
3286 if (kmutex->km_header.dh_sigstate == 1) {
3287 kmutex->km_ownerthread = NULL;
3288 ntoskrnl_waittest(&kmutex->km_header, IO_NO_INCREMENT);
3289 }
3290
3291 mtx_unlock(&ntoskrnl_dispatchlock);
3292
3293 return (prevstate);
3294}
3295
3296static uint32_t
3297KeReadStateMutex(kmutex)
3298 kmutant *kmutex;
3299{
3300 return (kmutex->km_header.dh_sigstate);
3301}
3302
3303void
3304KeInitializeEvent(nt_kevent *kevent, uint32_t type, uint8_t state)
3305{
3306 InitializeListHead((&kevent->k_header.dh_waitlisthead));
3307 kevent->k_header.dh_sigstate = state;
3308 if (type == EVENT_TYPE_NOTIFY)
3309 kevent->k_header.dh_type = DISP_TYPE_NOTIFICATION_EVENT;
3310 else
3311 kevent->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_EVENT;
3312 kevent->k_header.dh_size = sizeof(nt_kevent) / sizeof(uint32_t);
3313}
3314
3315uint32_t
3316KeResetEvent(kevent)
3317 nt_kevent *kevent;
3318{
3319 uint32_t prevstate;
3320
3321 mtx_lock(&ntoskrnl_dispatchlock);
3322 prevstate = kevent->k_header.dh_sigstate;
3323 kevent->k_header.dh_sigstate = FALSE;
3324 mtx_unlock(&ntoskrnl_dispatchlock);
3325
3326 return (prevstate);
3327}
3328
3329uint32_t
3330KeSetEvent(nt_kevent *kevent, uint32_t increment, uint8_t kwait)
3331{
3332 uint32_t prevstate;
3333 wait_block *w;
3334 nt_dispatch_header *dh;
3335 struct thread *td;
3336 wb_ext *we;
3337
3338 mtx_lock(&ntoskrnl_dispatchlock);
3339 prevstate = kevent->k_header.dh_sigstate;
3340 dh = &kevent->k_header;
3341
3342 if (IsListEmpty(&dh->dh_waitlisthead))
3343 /*
3344 * If there's nobody in the waitlist, just set
3345 * the state to signalled.
3346 */
3347 dh->dh_sigstate = 1;
3348 else {
3349 /*
3350 * Get the first waiter. If this is a synchronization
3351 * event, just wake up that one thread (don't bother
3352 * setting the state to signalled since we're supposed
3353 * to automatically clear synchronization events anyway).
3354 *
3355 * If it's a notification event, or the first
3356 * waiter is doing a WAITTYPE_ALL wait, go through
3357 * the full wait satisfaction process.
3358 */
3359 w = CONTAINING_RECORD(dh->dh_waitlisthead.nle_flink,
3360 wait_block, wb_waitlist);
3361 we = w->wb_ext;
3362 td = we->we_td;
3363 if (kevent->k_header.dh_type == DISP_TYPE_NOTIFICATION_EVENT ||
3364 w->wb_waittype == WAITTYPE_ALL) {
3365 if (prevstate == 0) {
3366 dh->dh_sigstate = 1;
3367 ntoskrnl_waittest(dh, increment);
3368 }
3369 } else {
3370 w->wb_awakened |= TRUE;
3371 cv_broadcastpri(&we->we_cv,
3372 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
3373 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
3374 }
3375 }
3376
3377 mtx_unlock(&ntoskrnl_dispatchlock);
3378
3379 return (prevstate);
3380}
3381
3382void
3383KeClearEvent(kevent)
3384 nt_kevent *kevent;
3385{
3386 kevent->k_header.dh_sigstate = FALSE;
3387}
3388
3389uint32_t
3390KeReadStateEvent(kevent)
3391 nt_kevent *kevent;
3392{
3393 return (kevent->k_header.dh_sigstate);
3394}
3395
3396/*
3397 * The object manager in Windows is responsible for managing
3398 * references and access to various types of objects, including
3399 * device_objects, events, threads, timers and so on. However,
3400 * there's a difference in the way objects are handled in user
3401 * mode versus kernel mode.
3402 *
3403 * In user mode (i.e. Win32 applications), all objects are
3404 * managed by the object manager. For example, when you create
3405 * a timer or event object, you actually end up with an
3406 * object_header (for the object manager's bookkeeping
3407 * purposes) and an object body (which contains the actual object
3408 * structure, e.g. ktimer, kevent, etc...). This allows Windows
3409 * to manage resource quotas and to enforce access restrictions
3410 * on basically every kind of system object handled by the kernel.
3411 *
3412 * However, in kernel mode, you only end up using the object
3413 * manager some of the time. For example, in a driver, you create
3414 * a timer object by simply allocating the memory for a ktimer
3415 * structure and initializing it with KeInitializeTimer(). Hence,
3416 * the timer has no object_header and no reference counting or
3417 * security/resource checks are done on it. The assumption in
3418 * this case is that if you're running in kernel mode, you know
3419 * what you're doing, and you're already at an elevated privilege
3420 * anyway.
3421 *
3422 * There are some exceptions to this. The two most important ones
3423 * for our purposes are device_objects and threads. We need to use
3424 * the object manager to do reference counting on device_objects,
3425 * and for threads, you can only get a pointer to a thread's
3426 * dispatch header by using ObReferenceObjectByHandle() on the
3427 * handle returned by PsCreateSystemThread().
3428 */
3429
3430static ndis_status
3431ObReferenceObjectByHandle(ndis_handle handle, uint32_t reqaccess, void *otype,
3432 uint8_t accessmode, void **object, void **handleinfo)
3433{
3434 nt_objref *nr;
3435
3436 nr = malloc(sizeof(nt_objref), M_DEVBUF, M_NOWAIT|M_ZERO);
3437 if (nr == NULL)
3438 return (STATUS_INSUFFICIENT_RESOURCES);
3439
3440 InitializeListHead((&nr->no_dh.dh_waitlisthead));
3441 nr->no_obj = handle;
3442 nr->no_dh.dh_type = DISP_TYPE_THREAD;
3443 nr->no_dh.dh_sigstate = 0;
3444 nr->no_dh.dh_size = (uint8_t)(sizeof(struct thread) /
3445 sizeof(uint32_t));
3446 TAILQ_INSERT_TAIL(&ntoskrnl_reflist, nr, link);
3447 *object = nr;
3448
3449 return (STATUS_SUCCESS);
3450}
3451
3452static void
3453ObfDereferenceObject(object)
3454 void *object;
3455{
3456 nt_objref *nr;
3457
3458 nr = object;
3459 TAILQ_REMOVE(&ntoskrnl_reflist, nr, link);
3460 free(nr, M_DEVBUF);
3461}
3462
3463static uint32_t
3464ZwClose(handle)
3465 ndis_handle handle;
3466{
3467 return (STATUS_SUCCESS);
3468}
3469
3470static uint32_t
3471WmiQueryTraceInformation(traceclass, traceinfo, infolen, reqlen, buf)
3472 uint32_t traceclass;
3473 void *traceinfo;
3474 uint32_t infolen;
3475 uint32_t reqlen;
3476 void *buf;
3477{
3478 return (STATUS_NOT_FOUND);
3479}
3480
3481static uint32_t
3482WmiTraceMessage(uint64_t loghandle, uint32_t messageflags,
3483 void *guid, uint16_t messagenum, ...)
3484{
3485 return (STATUS_SUCCESS);
3486}
3487
3488static uint32_t
3489IoWMIRegistrationControl(dobj, action)
3490 device_object *dobj;
3491 uint32_t action;
3492{
3493 return (STATUS_SUCCESS);
3494}
3495
3496/*
3497 * This is here just in case the thread returns without calling
3498 * PsTerminateSystemThread().
3499 */
3500static void
3501ntoskrnl_thrfunc(arg)
3502 void *arg;
3503{
3504 thread_context *thrctx;
3505 uint32_t (*tfunc)(void *);
3506 void *tctx;
3507 uint32_t rval;
3508
3509 thrctx = arg;
3510 tfunc = thrctx->tc_thrfunc;
3511 tctx = thrctx->tc_thrctx;
3512 free(thrctx, M_TEMP);
3513
3514 rval = MSCALL1(tfunc, tctx);
3515
3516 PsTerminateSystemThread(rval);
3517 return; /* notreached */
3518}
3519
3520static ndis_status
3521PsCreateSystemThread(handle, reqaccess, objattrs, phandle,
3522 clientid, thrfunc, thrctx)
3523 ndis_handle *handle;
3524 uint32_t reqaccess;
3525 void *objattrs;
3526 ndis_handle phandle;
3527 void *clientid;
3528 void *thrfunc;
3529 void *thrctx;
3530{
3531 int error;
3532 thread_context *tc;
3533 struct proc *p;
3534
3535 tc = malloc(sizeof(thread_context), M_TEMP, M_NOWAIT);
3536 if (tc == NULL)
3537 return (STATUS_INSUFFICIENT_RESOURCES);
3538
3539 tc->tc_thrctx = thrctx;
3540 tc->tc_thrfunc = thrfunc;
3541
3542 error = kproc_create(ntoskrnl_thrfunc, tc, &p,
3543 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Kthread %d", ntoskrnl_kth);
3544
3545 if (error) {
3546 free(tc, M_TEMP);
3547 return (STATUS_INSUFFICIENT_RESOURCES);
3548 }
3549
3550 *handle = p;
3551 ntoskrnl_kth++;
3552
3553 return (STATUS_SUCCESS);
3554}
3555
3556/*
3557 * In Windows, the exit of a thread is an event that you're allowed
3558 * to wait on, assuming you've obtained a reference to the thread using
3559 * ObReferenceObjectByHandle(). Unfortunately, the only way we can
3560 * simulate this behavior is to register each thread we create in a
3561 * reference list, and if someone holds a reference to us, we poke
3562 * them.
3563 */
3564static ndis_status
3565PsTerminateSystemThread(status)
3566 ndis_status status;
3567{
3568 struct nt_objref *nr;
3569
3570 mtx_lock(&ntoskrnl_dispatchlock);
3571 TAILQ_FOREACH(nr, &ntoskrnl_reflist, link) {
3572 if (nr->no_obj != curthread->td_proc)
3573 continue;
3574 nr->no_dh.dh_sigstate = 1;
3575 ntoskrnl_waittest(&nr->no_dh, IO_NO_INCREMENT);
3576 break;
3577 }
3578 mtx_unlock(&ntoskrnl_dispatchlock);
3579
3580 ntoskrnl_kth--;
3581
3582 kproc_exit(0);
3583 return (0); /* notreached */
3584}
3585
3586static uint32_t
3587DbgPrint(char *fmt, ...)
3588{
3589 va_list ap;
3590
3591 if (bootverbose) {
3592 va_start(ap, fmt);
3593 vprintf(fmt, ap);
3594 va_end(ap);
3595 }
3596
3597 return (STATUS_SUCCESS);
3598}
3599
3600static void
3601DbgBreakPoint(void)
3602{
3603
3604 kdb_enter(KDB_WHY_NDIS, "DbgBreakPoint(): breakpoint");
3605}
3606
3607static void
3608KeBugCheckEx(code, param1, param2, param3, param4)
3609 uint32_t code;
3610 u_long param1;
3611 u_long param2;
3612 u_long param3;
3613 u_long param4;
3614{
3615 panic("KeBugCheckEx: STOP 0x%X", code);
3616}
3617
3618static void
3619ntoskrnl_timercall(arg)
3620 void *arg;
3621{
3622 ktimer *timer;
3623 struct timeval tv;
3624 kdpc *dpc;
3625
3626 mtx_lock(&ntoskrnl_dispatchlock);
3627
3628 timer = arg;
3629
3630#ifdef NTOSKRNL_DEBUG_TIMERS
3631 ntoskrnl_timer_fires++;
3632#endif
3633 ntoskrnl_remove_timer(timer);
3634
3635 /*
3636 * This should never happen, but complain
3637 * if it does.
3638 */
3639
3640 if (timer->k_header.dh_inserted == FALSE) {
3641 mtx_unlock(&ntoskrnl_dispatchlock);
3642 printf("NTOS: timer %p fired even though "
3643 "it was canceled\n", timer);
3644 return;
3645 }
3646
3647 /* Mark the timer as no longer being on the timer queue. */
3648
3649 timer->k_header.dh_inserted = FALSE;
3650
3651 /* Now signal the object and satisfy any waits on it. */
3652
3653 timer->k_header.dh_sigstate = 1;
3654 ntoskrnl_waittest(&timer->k_header, IO_NO_INCREMENT);
3655
3656 /*
3657 * If this is a periodic timer, re-arm it
3658 * so it will fire again. We do this before
3659 * calling any deferred procedure calls because
3660 * it's possible the DPC might cancel the timer,
3661 * in which case it would be wrong for us to
3662 * re-arm it again afterwards.
3663 */
3664
3665 if (timer->k_period) {
3666 tv.tv_sec = 0;
3667 tv.tv_usec = timer->k_period * 1000;
3668 timer->k_header.dh_inserted = TRUE;
3669 ntoskrnl_insert_timer(timer, tvtohz(&tv));
3670#ifdef NTOSKRNL_DEBUG_TIMERS
3671 ntoskrnl_timer_reloads++;
3672#endif
3673 }
3674
3675 dpc = timer->k_dpc;
3676
3677 mtx_unlock(&ntoskrnl_dispatchlock);
3678
3679 /* If there's a DPC associated with the timer, queue it up. */
3680
3681 if (dpc != NULL)
3682 KeInsertQueueDpc(dpc, NULL, NULL);
3683}
3684
3685#ifdef NTOSKRNL_DEBUG_TIMERS
3686static int
3687sysctl_show_timers(SYSCTL_HANDLER_ARGS)
3688{
3689 int ret;
3690
3691 ret = 0;
3692 ntoskrnl_show_timers();
3693 return (sysctl_handle_int(oidp, &ret, 0, req));
3694}
3695
3696static void
3697ntoskrnl_show_timers()
3698{
3699 int i = 0;
3700 list_entry *l;
3701
3702 mtx_lock_spin(&ntoskrnl_calllock);
3703 l = ntoskrnl_calllist.nle_flink;
3704 while(l != &ntoskrnl_calllist) {
3705 i++;
3706 l = l->nle_flink;
3707 }
3708 mtx_unlock_spin(&ntoskrnl_calllock);
3709
3710 printf("\n");
3711 printf("%d timers available (out of %d)\n", i, NTOSKRNL_TIMEOUTS);
3712 printf("timer sets: %qu\n", ntoskrnl_timer_sets);
3713 printf("timer reloads: %qu\n", ntoskrnl_timer_reloads);
3714 printf("timer cancels: %qu\n", ntoskrnl_timer_cancels);
3715 printf("timer fires: %qu\n", ntoskrnl_timer_fires);
3716 printf("\n");
3717}
3718#endif
3719
3720/*
3721 * Must be called with dispatcher lock held.
3722 */
3723
3724static void
3725ntoskrnl_insert_timer(timer, ticks)
3726 ktimer *timer;
3727 int ticks;
3728{
3729 callout_entry *e;
3730 list_entry *l;
3731 struct callout *c;
3732
3733 /*
3734 * Try and allocate a timer.
3735 */
3736 mtx_lock_spin(&ntoskrnl_calllock);
3737 if (IsListEmpty(&ntoskrnl_calllist)) {
3738 mtx_unlock_spin(&ntoskrnl_calllock);
3739#ifdef NTOSKRNL_DEBUG_TIMERS
3740 ntoskrnl_show_timers();
3741#endif
3742 panic("out of timers!");
3743 }
3744 l = RemoveHeadList(&ntoskrnl_calllist);
3745 mtx_unlock_spin(&ntoskrnl_calllock);
3746
3747 e = CONTAINING_RECORD(l, callout_entry, ce_list);
3748 c = &e->ce_callout;
3749
3750 timer->k_callout = c;
3751
3752 callout_init(c, 1);
3753 callout_reset(c, ticks, ntoskrnl_timercall, timer);
3754}
3755
3756static void
3757ntoskrnl_remove_timer(timer)
3758 ktimer *timer;
3759{
3760 callout_entry *e;
3761
3762 e = (callout_entry *)timer->k_callout;
3763 callout_stop(timer->k_callout);
3764
3765 mtx_lock_spin(&ntoskrnl_calllock);
3766 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
3767 mtx_unlock_spin(&ntoskrnl_calllock);
3768}
3769
3770void
3771KeInitializeTimer(timer)
3772 ktimer *timer;
3773{
3774 if (timer == NULL)
3775 return;
3776
3777 KeInitializeTimerEx(timer, EVENT_TYPE_NOTIFY);
3778}
3779
3780void
3781KeInitializeTimerEx(timer, type)
3782 ktimer *timer;
3783 uint32_t type;
3784{
3785 if (timer == NULL)
3786 return;
3787
3788 bzero((char *)timer, sizeof(ktimer));
3789 InitializeListHead((&timer->k_header.dh_waitlisthead));
3790 timer->k_header.dh_sigstate = FALSE;
3791 timer->k_header.dh_inserted = FALSE;
3792 if (type == EVENT_TYPE_NOTIFY)
3793 timer->k_header.dh_type = DISP_TYPE_NOTIFICATION_TIMER;
3794 else
3795 timer->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_TIMER;
3796 timer->k_header.dh_size = sizeof(ktimer) / sizeof(uint32_t);
3797}
3798
3799/*
3800 * DPC subsystem. A Windows Defered Procedure Call has the following
3801 * properties:
3802 * - It runs at DISPATCH_LEVEL.
3803 * - It can have one of 3 importance values that control when it
3804 * runs relative to other DPCs in the queue.
3805 * - On SMP systems, it can be set to run on a specific processor.
3806 * In order to satisfy the last property, we create a DPC thread for
3807 * each CPU in the system and bind it to that CPU. Each thread
3808 * maintains three queues with different importance levels, which
3809 * will be processed in order from lowest to highest.
3810 *
3811 * In Windows, interrupt handlers run as DPCs. (Not to be confused
3812 * with ISRs, which run in interrupt context and can preempt DPCs.)
3813 * ISRs are given the highest importance so that they'll take
3814 * precedence over timers and other things.
3815 */
3816
3817static void
3818ntoskrnl_dpc_thread(arg)
3819 void *arg;
3820{
3821 kdpc_queue *kq;
3822 kdpc *d;
3823 list_entry *l;
3824 uint8_t irql;
3825
3826 kq = arg;
3827
3828 InitializeListHead(&kq->kq_disp);
3829 kq->kq_td = curthread;
3830 kq->kq_exit = 0;
3831 kq->kq_running = FALSE;
3832 KeInitializeSpinLock(&kq->kq_lock);
3833 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
3834 KeInitializeEvent(&kq->kq_done, EVENT_TYPE_SYNC, FALSE);
3835
3836 /*
3837 * Elevate our priority. DPCs are used to run interrupt
3838 * handlers, and they should trigger as soon as possible
3839 * once scheduled by an ISR.
3840 */
3841
3842 thread_lock(curthread);
3843#ifdef NTOSKRNL_MULTIPLE_DPCS
3844 sched_bind(curthread, kq->kq_cpu);
3845#endif
3846 sched_prio(curthread, PRI_MIN_KERN);
3847 thread_unlock(curthread);
3848
3849 while (1) {
3850 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
3851
3852 KeAcquireSpinLock(&kq->kq_lock, &irql);
3853
3854 if (kq->kq_exit) {
3855 kq->kq_exit = 0;
3856 KeReleaseSpinLock(&kq->kq_lock, irql);
3857 break;
3858 }
3859
3860 kq->kq_running = TRUE;
3861
3862 while (!IsListEmpty(&kq->kq_disp)) {
3863 l = RemoveHeadList((&kq->kq_disp));
3864 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3865 InitializeListHead((&d->k_dpclistentry));
3866 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
3867 MSCALL4(d->k_deferedfunc, d, d->k_deferredctx,
3868 d->k_sysarg1, d->k_sysarg2);
3869 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3870 }
3871
3872 kq->kq_running = FALSE;
3873
3874 KeReleaseSpinLock(&kq->kq_lock, irql);
3875
3876 KeSetEvent(&kq->kq_done, IO_NO_INCREMENT, FALSE);
3877 }
3878
3879 kproc_exit(0);
3880 return; /* notreached */
3881}
3882
3883static void
3884ntoskrnl_destroy_dpc_threads(void)
3885{
3886 kdpc_queue *kq;
3887 kdpc dpc;
3888 int i;
3889
3890 kq = kq_queues;
3891#ifdef NTOSKRNL_MULTIPLE_DPCS
3892 for (i = 0; i < mp_ncpus; i++) {
3893#else
3894 for (i = 0; i < 1; i++) {
3895#endif
3896 kq += i;
3897
3898 kq->kq_exit = 1;
3899 KeInitializeDpc(&dpc, NULL, NULL);
3900 KeSetTargetProcessorDpc(&dpc, i);
3901 KeInsertQueueDpc(&dpc, NULL, NULL);
3902 while (kq->kq_exit)
3903 tsleep(kq->kq_td->td_proc, PWAIT, "dpcw", hz/10);
3904 }
3905}
3906
3907static uint8_t
3908ntoskrnl_insert_dpc(head, dpc)
3909 list_entry *head;
3910 kdpc *dpc;
3911{
3912 list_entry *l;
3913 kdpc *d;
3914
3915 l = head->nle_flink;
3916 while (l != head) {
3917 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3918 if (d == dpc)
3919 return (FALSE);
3920 l = l->nle_flink;
3921 }
3922
3923 if (dpc->k_importance == KDPC_IMPORTANCE_LOW)
3924 InsertTailList((head), (&dpc->k_dpclistentry));
3925 else
3926 InsertHeadList((head), (&dpc->k_dpclistentry));
3927
3928 return (TRUE);
3929}
3930
3931void
3932KeInitializeDpc(dpc, dpcfunc, dpcctx)
3933 kdpc *dpc;
3934 void *dpcfunc;
3935 void *dpcctx;
3936{
3937
3938 if (dpc == NULL)
3939 return;
3940
3941 dpc->k_deferedfunc = dpcfunc;
3942 dpc->k_deferredctx = dpcctx;
3943 dpc->k_num = KDPC_CPU_DEFAULT;
3944 dpc->k_importance = KDPC_IMPORTANCE_MEDIUM;
3945 InitializeListHead((&dpc->k_dpclistentry));
3946}
3947
3948uint8_t
3949KeInsertQueueDpc(dpc, sysarg1, sysarg2)
3950 kdpc *dpc;
3951 void *sysarg1;
3952 void *sysarg2;
3953{
3954 kdpc_queue *kq;
3955 uint8_t r;
3956 uint8_t irql;
3957
3958 if (dpc == NULL)
3959 return (FALSE);
3960
3961 kq = kq_queues;
3962
3963#ifdef NTOSKRNL_MULTIPLE_DPCS
3964 KeRaiseIrql(DISPATCH_LEVEL, &irql);
3965
3966 /*
3967 * By default, the DPC is queued to run on the same CPU
3968 * that scheduled it.
3969 */
3970
3971 if (dpc->k_num == KDPC_CPU_DEFAULT)
3972 kq += curthread->td_oncpu;
3973 else
3974 kq += dpc->k_num;
3975 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3976#else
3977 KeAcquireSpinLock(&kq->kq_lock, &irql);
3978#endif
3979
3980 r = ntoskrnl_insert_dpc(&kq->kq_disp, dpc);
3981 if (r == TRUE) {
3982 dpc->k_sysarg1 = sysarg1;
3983 dpc->k_sysarg2 = sysarg2;
3984 }
3985 KeReleaseSpinLock(&kq->kq_lock, irql);
3986
3987 if (r == FALSE)
3988 return (r);
3989
3990 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
3991
3992 return (r);
3993}
3994
3995uint8_t
3996KeRemoveQueueDpc(dpc)
3997 kdpc *dpc;
3998{
3999 kdpc_queue *kq;
4000 uint8_t irql;
4001
4002 if (dpc == NULL)
4003 return (FALSE);
4004
4005#ifdef NTOSKRNL_MULTIPLE_DPCS
4006 KeRaiseIrql(DISPATCH_LEVEL, &irql);
4007
4008 kq = kq_queues + dpc->k_num;
4009
4010 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
4011#else
4012 kq = kq_queues;
4013 KeAcquireSpinLock(&kq->kq_lock, &irql);
4014#endif
4015
4016 if (dpc->k_dpclistentry.nle_flink == &dpc->k_dpclistentry) {
4017 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
4018 KeLowerIrql(irql);
4019 return (FALSE);
4020 }
4021
4022 RemoveEntryList((&dpc->k_dpclistentry));
4023 InitializeListHead((&dpc->k_dpclistentry));
4024
4025 KeReleaseSpinLock(&kq->kq_lock, irql);
4026
4027 return (TRUE);
4028}
4029
4030void
4031KeSetImportanceDpc(dpc, imp)
4032 kdpc *dpc;
4033 uint32_t imp;
4034{
4035 if (imp != KDPC_IMPORTANCE_LOW &&
4036 imp != KDPC_IMPORTANCE_MEDIUM &&
4037 imp != KDPC_IMPORTANCE_HIGH)
4038 return;
4039
4040 dpc->k_importance = (uint8_t)imp;
4041}
4042
4043void
4044KeSetTargetProcessorDpc(kdpc *dpc, uint8_t cpu)
4045{
4046 if (cpu > mp_ncpus)
4047 return;
4048
4049 dpc->k_num = cpu;
4050}
4051
4052void
4053KeFlushQueuedDpcs(void)
4054{
4055 kdpc_queue *kq;
4056 int i;
4057
4058 /*
4059 * Poke each DPC queue and wait
4060 * for them to drain.
4061 */
4062
4063#ifdef NTOSKRNL_MULTIPLE_DPCS
4064 for (i = 0; i < mp_ncpus; i++) {
4065#else
4066 for (i = 0; i < 1; i++) {
4067#endif
4068 kq = kq_queues + i;
4069 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
4070 KeWaitForSingleObject(&kq->kq_done, 0, 0, TRUE, NULL);
4071 }
4072}
4073
4074uint32_t
4075KeGetCurrentProcessorNumber(void)
4076{
4077 return ((uint32_t)curthread->td_oncpu);
4078}
4079
4080uint8_t
4081KeSetTimerEx(timer, duetime, period, dpc)
4082 ktimer *timer;
4083 int64_t duetime;
4084 uint32_t period;
4085 kdpc *dpc;
4086{
4087 struct timeval tv;
4088 uint64_t curtime;
4089 uint8_t pending;
4090
4091 if (timer == NULL)
4092 return (FALSE);
4093
4094 mtx_lock(&ntoskrnl_dispatchlock);
4095
4096 if (timer->k_header.dh_inserted == TRUE) {
4097 ntoskrnl_remove_timer(timer);
4098#ifdef NTOSKRNL_DEBUG_TIMERS
4099 ntoskrnl_timer_cancels++;
4100#endif
4101 timer->k_header.dh_inserted = FALSE;
4102 pending = TRUE;
4103 } else
4104 pending = FALSE;
4105
4106 timer->k_duetime = duetime;
4107 timer->k_period = period;
4108 timer->k_header.dh_sigstate = FALSE;
4109 timer->k_dpc = dpc;
4110
4111 if (duetime < 0) {
4112 tv.tv_sec = - (duetime) / 10000000;
4113 tv.tv_usec = (- (duetime) / 10) -
4114 (tv.tv_sec * 1000000);
4115 } else {
4116 ntoskrnl_time(&curtime);
4117 if (duetime < curtime)
4118 tv.tv_sec = tv.tv_usec = 0;
4119 else {
4120 tv.tv_sec = ((duetime) - curtime) / 10000000;
4121 tv.tv_usec = ((duetime) - curtime) / 10 -
4122 (tv.tv_sec * 1000000);
4123 }
4124 }
4125
4126 timer->k_header.dh_inserted = TRUE;
4127 ntoskrnl_insert_timer(timer, tvtohz(&tv));
4128#ifdef NTOSKRNL_DEBUG_TIMERS
4129 ntoskrnl_timer_sets++;
4130#endif
4131
4132 mtx_unlock(&ntoskrnl_dispatchlock);
4133
4134 return (pending);
4135}
4136
4137uint8_t
4138KeSetTimer(timer, duetime, dpc)
4139 ktimer *timer;
4140 int64_t duetime;
4141 kdpc *dpc;
4142{
4143 return (KeSetTimerEx(timer, duetime, 0, dpc));
4144}
4145
4146/*
4147 * The Windows DDK documentation seems to say that cancelling
4148 * a timer that has a DPC will result in the DPC also being
4149 * cancelled, but this isn't really the case.
4150 */
4151
4152uint8_t
4153KeCancelTimer(timer)
4154 ktimer *timer;
4155{
4156 uint8_t pending;
4157
4158 if (timer == NULL)
4159 return (FALSE);
4160
4161 mtx_lock(&ntoskrnl_dispatchlock);
4162
4163 pending = timer->k_header.dh_inserted;
4164
4165 if (timer->k_header.dh_inserted == TRUE) {
4166 timer->k_header.dh_inserted = FALSE;
4167 ntoskrnl_remove_timer(timer);
4168#ifdef NTOSKRNL_DEBUG_TIMERS
4169 ntoskrnl_timer_cancels++;
4170#endif
4171 }
4172
4173 mtx_unlock(&ntoskrnl_dispatchlock);
4174
4175 return (pending);
4176}
4177
4178uint8_t
4179KeReadStateTimer(timer)
4180 ktimer *timer;
4181{
4182 return (timer->k_header.dh_sigstate);
4183}
4184
4185static int32_t
4186KeDelayExecutionThread(uint8_t wait_mode, uint8_t alertable, int64_t *interval)
4187{
4188 ktimer timer;
4189
4190 if (wait_mode != 0)
4191 panic("invalid wait_mode %d", wait_mode);
4192
4193 KeInitializeTimer(&timer);
4194 KeSetTimer(&timer, *interval, NULL);
4195 KeWaitForSingleObject(&timer, 0, 0, alertable, NULL);
4196
4197 return STATUS_SUCCESS;
4198}
4199
4200static uint64_t
4201KeQueryInterruptTime(void)
4202{
4203 int ticks;
4204 struct timeval tv;
4205
4206 getmicrouptime(&tv);
4207
4208 ticks = tvtohz(&tv);
4209
| 35
36#include <sys/ctype.h>
37#include <sys/unistd.h>
38#include <sys/param.h>
39#include <sys/types.h>
40#include <sys/errno.h>
41#include <sys/systm.h>
42#include <sys/malloc.h>
43#include <sys/lock.h>
44#include <sys/mutex.h>
45
46#include <sys/callout.h>
47#include <sys/kdb.h>
48#include <sys/kernel.h>
49#include <sys/proc.h>
50#include <sys/condvar.h>
51#include <sys/kthread.h>
52#include <sys/module.h>
53#include <sys/smp.h>
54#include <sys/sched.h>
55#include <sys/sysctl.h>
56
57#include <machine/atomic.h>
58#include <machine/bus.h>
59#include <machine/stdarg.h>
60#include <machine/resource.h>
61
62#include <sys/bus.h>
63#include <sys/rman.h>
64
65#include <vm/vm.h>
66#include <vm/vm_param.h>
67#include <vm/pmap.h>
68#include <vm/uma.h>
69#include <vm/vm_kern.h>
70#include <vm/vm_map.h>
71#include <vm/vm_extern.h>
72
73#include <compat/ndis/pe_var.h>
74#include <compat/ndis/cfg_var.h>
75#include <compat/ndis/resource_var.h>
76#include <compat/ndis/ntoskrnl_var.h>
77#include <compat/ndis/hal_var.h>
78#include <compat/ndis/ndis_var.h>
79
80#ifdef NTOSKRNL_DEBUG_TIMERS
81static int sysctl_show_timers(SYSCTL_HANDLER_ARGS);
82
83SYSCTL_PROC(_debug, OID_AUTO, ntoskrnl_timers, CTLTYPE_INT | CTLFLAG_RW,
84 NULL, 0, sysctl_show_timers, "I",
85 "Show ntoskrnl timer stats");
86#endif
87
88struct kdpc_queue {
89 list_entry kq_disp;
90 struct thread *kq_td;
91 int kq_cpu;
92 int kq_exit;
93 int kq_running;
94 kspin_lock kq_lock;
95 nt_kevent kq_proc;
96 nt_kevent kq_done;
97};
98
99typedef struct kdpc_queue kdpc_queue;
100
101struct wb_ext {
102 struct cv we_cv;
103 struct thread *we_td;
104};
105
106typedef struct wb_ext wb_ext;
107
108#define NTOSKRNL_TIMEOUTS 256
109#ifdef NTOSKRNL_DEBUG_TIMERS
110static uint64_t ntoskrnl_timer_fires;
111static uint64_t ntoskrnl_timer_sets;
112static uint64_t ntoskrnl_timer_reloads;
113static uint64_t ntoskrnl_timer_cancels;
114#endif
115
116struct callout_entry {
117 struct callout ce_callout;
118 list_entry ce_list;
119};
120
121typedef struct callout_entry callout_entry;
122
123static struct list_entry ntoskrnl_calllist;
124static struct mtx ntoskrnl_calllock;
125struct kuser_shared_data kuser_shared_data;
126
127static struct list_entry ntoskrnl_intlist;
128static kspin_lock ntoskrnl_intlock;
129
130static uint8_t RtlEqualUnicodeString(unicode_string *,
131 unicode_string *, uint8_t);
132static void RtlCopyString(ansi_string *, const ansi_string *);
133static void RtlCopyUnicodeString(unicode_string *,
134 unicode_string *);
135static irp *IoBuildSynchronousFsdRequest(uint32_t, device_object *,
136 void *, uint32_t, uint64_t *, nt_kevent *, io_status_block *);
137static irp *IoBuildAsynchronousFsdRequest(uint32_t,
138 device_object *, void *, uint32_t, uint64_t *, io_status_block *);
139static irp *IoBuildDeviceIoControlRequest(uint32_t,
140 device_object *, void *, uint32_t, void *, uint32_t,
141 uint8_t, nt_kevent *, io_status_block *);
142static irp *IoAllocateIrp(uint8_t, uint8_t);
143static void IoReuseIrp(irp *, uint32_t);
144static void IoFreeIrp(irp *);
145static void IoInitializeIrp(irp *, uint16_t, uint8_t);
146static irp *IoMakeAssociatedIrp(irp *, uint8_t);
147static uint32_t KeWaitForMultipleObjects(uint32_t,
148 nt_dispatch_header **, uint32_t, uint32_t, uint32_t, uint8_t,
149 int64_t *, wait_block *);
150static void ntoskrnl_waittest(nt_dispatch_header *, uint32_t);
151static void ntoskrnl_satisfy_wait(nt_dispatch_header *, struct thread *);
152static void ntoskrnl_satisfy_multiple_waits(wait_block *);
153static int ntoskrnl_is_signalled(nt_dispatch_header *, struct thread *);
154static void ntoskrnl_insert_timer(ktimer *, int);
155static void ntoskrnl_remove_timer(ktimer *);
156#ifdef NTOSKRNL_DEBUG_TIMERS
157static void ntoskrnl_show_timers(void);
158#endif
159static void ntoskrnl_timercall(void *);
160static void ntoskrnl_dpc_thread(void *);
161static void ntoskrnl_destroy_dpc_threads(void);
162static void ntoskrnl_destroy_workitem_threads(void);
163static void ntoskrnl_workitem_thread(void *);
164static void ntoskrnl_workitem(device_object *, void *);
165static void ntoskrnl_unicode_to_ascii(uint16_t *, char *, int);
166static void ntoskrnl_ascii_to_unicode(char *, uint16_t *, int);
167static uint8_t ntoskrnl_insert_dpc(list_entry *, kdpc *);
168static void WRITE_REGISTER_USHORT(uint16_t *, uint16_t);
169static uint16_t READ_REGISTER_USHORT(uint16_t *);
170static void WRITE_REGISTER_ULONG(uint32_t *, uint32_t);
171static uint32_t READ_REGISTER_ULONG(uint32_t *);
172static void WRITE_REGISTER_UCHAR(uint8_t *, uint8_t);
173static uint8_t READ_REGISTER_UCHAR(uint8_t *);
174static int64_t _allmul(int64_t, int64_t);
175static int64_t _alldiv(int64_t, int64_t);
176static int64_t _allrem(int64_t, int64_t);
177static int64_t _allshr(int64_t, uint8_t);
178static int64_t _allshl(int64_t, uint8_t);
179static uint64_t _aullmul(uint64_t, uint64_t);
180static uint64_t _aulldiv(uint64_t, uint64_t);
181static uint64_t _aullrem(uint64_t, uint64_t);
182static uint64_t _aullshr(uint64_t, uint8_t);
183static uint64_t _aullshl(uint64_t, uint8_t);
184static slist_entry *ntoskrnl_pushsl(slist_header *, slist_entry *);
185static void InitializeSListHead(slist_header *);
186static slist_entry *ntoskrnl_popsl(slist_header *);
187static void ExFreePoolWithTag(void *, uint32_t);
188static void ExInitializePagedLookasideList(paged_lookaside_list *,
189 lookaside_alloc_func *, lookaside_free_func *,
190 uint32_t, size_t, uint32_t, uint16_t);
191static void ExDeletePagedLookasideList(paged_lookaside_list *);
192static void ExInitializeNPagedLookasideList(npaged_lookaside_list *,
193 lookaside_alloc_func *, lookaside_free_func *,
194 uint32_t, size_t, uint32_t, uint16_t);
195static void ExDeleteNPagedLookasideList(npaged_lookaside_list *);
196static slist_entry
197 *ExInterlockedPushEntrySList(slist_header *,
198 slist_entry *, kspin_lock *);
199static slist_entry
200 *ExInterlockedPopEntrySList(slist_header *, kspin_lock *);
201static uint32_t InterlockedIncrement(volatile uint32_t *);
202static uint32_t InterlockedDecrement(volatile uint32_t *);
203static void ExInterlockedAddLargeStatistic(uint64_t *, uint32_t);
204static void *MmAllocateContiguousMemory(uint32_t, uint64_t);
205static void *MmAllocateContiguousMemorySpecifyCache(uint32_t,
206 uint64_t, uint64_t, uint64_t, enum nt_caching_type);
207static void MmFreeContiguousMemory(void *);
208static void MmFreeContiguousMemorySpecifyCache(void *, uint32_t,
209 enum nt_caching_type);
210static uint32_t MmSizeOfMdl(void *, size_t);
211static void *MmMapLockedPages(mdl *, uint8_t);
212static void *MmMapLockedPagesSpecifyCache(mdl *,
213 uint8_t, uint32_t, void *, uint32_t, uint32_t);
214static void MmUnmapLockedPages(void *, mdl *);
215static device_t ntoskrnl_finddev(device_t, uint64_t, struct resource **);
216static void RtlZeroMemory(void *, size_t);
217static void RtlSecureZeroMemory(void *, size_t);
218static void RtlFillMemory(void *, size_t, uint8_t);
219static void RtlMoveMemory(void *, const void *, size_t);
220static ndis_status RtlCharToInteger(const char *, uint32_t, uint32_t *);
221static void RtlCopyMemory(void *, const void *, size_t);
222static size_t RtlCompareMemory(const void *, const void *, size_t);
223static ndis_status RtlUnicodeStringToInteger(unicode_string *,
224 uint32_t, uint32_t *);
225static int atoi (const char *);
226static long atol (const char *);
227static int rand(void);
228static void srand(unsigned int);
229static void KeQuerySystemTime(uint64_t *);
230static uint32_t KeTickCount(void);
231static uint8_t IoIsWdmVersionAvailable(uint8_t, uint8_t);
232static int32_t IoOpenDeviceRegistryKey(struct device_object *, uint32_t,
233 uint32_t, void **);
234static void ntoskrnl_thrfunc(void *);
235static ndis_status PsCreateSystemThread(ndis_handle *,
236 uint32_t, void *, ndis_handle, void *, void *, void *);
237static ndis_status PsTerminateSystemThread(ndis_status);
238static ndis_status IoGetDeviceObjectPointer(unicode_string *,
239 uint32_t, void *, device_object *);
240static ndis_status IoGetDeviceProperty(device_object *, uint32_t,
241 uint32_t, void *, uint32_t *);
242static void KeInitializeMutex(kmutant *, uint32_t);
243static uint32_t KeReleaseMutex(kmutant *, uint8_t);
244static uint32_t KeReadStateMutex(kmutant *);
245static ndis_status ObReferenceObjectByHandle(ndis_handle,
246 uint32_t, void *, uint8_t, void **, void **);
247static void ObfDereferenceObject(void *);
248static uint32_t ZwClose(ndis_handle);
249static uint32_t WmiQueryTraceInformation(uint32_t, void *, uint32_t,
250 uint32_t, void *);
251static uint32_t WmiTraceMessage(uint64_t, uint32_t, void *, uint16_t, ...);
252static uint32_t IoWMIRegistrationControl(device_object *, uint32_t);
253static void *ntoskrnl_memset(void *, int, size_t);
254static void *ntoskrnl_memmove(void *, void *, size_t);
255static void *ntoskrnl_memchr(void *, unsigned char, size_t);
256static char *ntoskrnl_strstr(char *, char *);
257static char *ntoskrnl_strncat(char *, char *, size_t);
258static int ntoskrnl_toupper(int);
259static int ntoskrnl_tolower(int);
260static funcptr ntoskrnl_findwrap(funcptr);
261static uint32_t DbgPrint(char *, ...);
262static void DbgBreakPoint(void);
263static void KeBugCheckEx(uint32_t, u_long, u_long, u_long, u_long);
264static int32_t KeDelayExecutionThread(uint8_t, uint8_t, int64_t *);
265static int32_t KeSetPriorityThread(struct thread *, int32_t);
266static void dummy(void);
267
268static struct mtx ntoskrnl_dispatchlock;
269static struct mtx ntoskrnl_interlock;
270static kspin_lock ntoskrnl_cancellock;
271static int ntoskrnl_kth = 0;
272static struct nt_objref_head ntoskrnl_reflist;
273static uma_zone_t mdl_zone;
274static uma_zone_t iw_zone;
275static struct kdpc_queue *kq_queues;
276static struct kdpc_queue *wq_queues;
277static int wq_idx = 0;
278
279int
280ntoskrnl_libinit()
281{
282 image_patch_table *patch;
283 int error;
284 struct proc *p;
285 kdpc_queue *kq;
286 callout_entry *e;
287 int i;
288
289 mtx_init(&ntoskrnl_dispatchlock,
290 "ntoskrnl dispatch lock", MTX_NDIS_LOCK, MTX_DEF|MTX_RECURSE);
291 mtx_init(&ntoskrnl_interlock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
292 KeInitializeSpinLock(&ntoskrnl_cancellock);
293 KeInitializeSpinLock(&ntoskrnl_intlock);
294 TAILQ_INIT(&ntoskrnl_reflist);
295
296 InitializeListHead(&ntoskrnl_calllist);
297 InitializeListHead(&ntoskrnl_intlist);
298 mtx_init(&ntoskrnl_calllock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
299
300 kq_queues = ExAllocatePoolWithTag(NonPagedPool,
301#ifdef NTOSKRNL_MULTIPLE_DPCS
302 sizeof(kdpc_queue) * mp_ncpus, 0);
303#else
304 sizeof(kdpc_queue), 0);
305#endif
306
307 if (kq_queues == NULL)
308 return (ENOMEM);
309
310 wq_queues = ExAllocatePoolWithTag(NonPagedPool,
311 sizeof(kdpc_queue) * WORKITEM_THREADS, 0);
312
313 if (wq_queues == NULL)
314 return (ENOMEM);
315
316#ifdef NTOSKRNL_MULTIPLE_DPCS
317 bzero((char *)kq_queues, sizeof(kdpc_queue) * mp_ncpus);
318#else
319 bzero((char *)kq_queues, sizeof(kdpc_queue));
320#endif
321 bzero((char *)wq_queues, sizeof(kdpc_queue) * WORKITEM_THREADS);
322
323 /*
324 * Launch the DPC threads.
325 */
326
327#ifdef NTOSKRNL_MULTIPLE_DPCS
328 for (i = 0; i < mp_ncpus; i++) {
329#else
330 for (i = 0; i < 1; i++) {
331#endif
332 kq = kq_queues + i;
333 kq->kq_cpu = i;
334 error = kproc_create(ntoskrnl_dpc_thread, kq, &p,
335 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows DPC %d", i);
336 if (error)
337 panic("failed to launch DPC thread");
338 }
339
340 /*
341 * Launch the workitem threads.
342 */
343
344 for (i = 0; i < WORKITEM_THREADS; i++) {
345 kq = wq_queues + i;
346 error = kproc_create(ntoskrnl_workitem_thread, kq, &p,
347 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Workitem %d", i);
348 if (error)
349 panic("failed to launch workitem thread");
350 }
351
352 patch = ntoskrnl_functbl;
353 while (patch->ipt_func != NULL) {
354 windrv_wrap((funcptr)patch->ipt_func,
355 (funcptr *)&patch->ipt_wrap,
356 patch->ipt_argcnt, patch->ipt_ftype);
357 patch++;
358 }
359
360 for (i = 0; i < NTOSKRNL_TIMEOUTS; i++) {
361 e = ExAllocatePoolWithTag(NonPagedPool,
362 sizeof(callout_entry), 0);
363 if (e == NULL)
364 panic("failed to allocate timeouts");
365 mtx_lock_spin(&ntoskrnl_calllock);
366 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
367 mtx_unlock_spin(&ntoskrnl_calllock);
368 }
369
370 /*
371 * MDLs are supposed to be variable size (they describe
372 * buffers containing some number of pages, but we don't
373 * know ahead of time how many pages that will be). But
374 * always allocating them off the heap is very slow. As
375 * a compromise, we create an MDL UMA zone big enough to
376 * handle any buffer requiring up to 16 pages, and we
377 * use those for any MDLs for buffers of 16 pages or less
378 * in size. For buffers larger than that (which we assume
379 * will be few and far between, we allocate the MDLs off
380 * the heap.
381 */
382
383 mdl_zone = uma_zcreate("Windows MDL", MDL_ZONE_SIZE,
384 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
385
386 iw_zone = uma_zcreate("Windows WorkItem", sizeof(io_workitem),
387 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
388
389 return (0);
390}
391
392int
393ntoskrnl_libfini()
394{
395 image_patch_table *patch;
396 callout_entry *e;
397 list_entry *l;
398
399 patch = ntoskrnl_functbl;
400 while (patch->ipt_func != NULL) {
401 windrv_unwrap(patch->ipt_wrap);
402 patch++;
403 }
404
405 /* Stop the workitem queues. */
406 ntoskrnl_destroy_workitem_threads();
407 /* Stop the DPC queues. */
408 ntoskrnl_destroy_dpc_threads();
409
410 ExFreePool(kq_queues);
411 ExFreePool(wq_queues);
412
413 uma_zdestroy(mdl_zone);
414 uma_zdestroy(iw_zone);
415
416 mtx_lock_spin(&ntoskrnl_calllock);
417 while(!IsListEmpty(&ntoskrnl_calllist)) {
418 l = RemoveHeadList(&ntoskrnl_calllist);
419 e = CONTAINING_RECORD(l, callout_entry, ce_list);
420 mtx_unlock_spin(&ntoskrnl_calllock);
421 ExFreePool(e);
422 mtx_lock_spin(&ntoskrnl_calllock);
423 }
424 mtx_unlock_spin(&ntoskrnl_calllock);
425
426 mtx_destroy(&ntoskrnl_dispatchlock);
427 mtx_destroy(&ntoskrnl_interlock);
428 mtx_destroy(&ntoskrnl_calllock);
429
430 return (0);
431}
432
433/*
434 * We need to be able to reference this externally from the wrapper;
435 * GCC only generates a local implementation of memset.
436 */
437static void *
438ntoskrnl_memset(buf, ch, size)
439 void *buf;
440 int ch;
441 size_t size;
442{
443 return (memset(buf, ch, size));
444}
445
446static void *
447ntoskrnl_memmove(dst, src, size)
448 void *src;
449 void *dst;
450 size_t size;
451{
452 bcopy(src, dst, size);
453 return (dst);
454}
455
456static void *
457ntoskrnl_memchr(void *buf, unsigned char ch, size_t len)
458{
459 if (len != 0) {
460 unsigned char *p = buf;
461
462 do {
463 if (*p++ == ch)
464 return (p - 1);
465 } while (--len != 0);
466 }
467 return (NULL);
468}
469
470static char *
471ntoskrnl_strstr(s, find)
472 char *s, *find;
473{
474 char c, sc;
475 size_t len;
476
477 if ((c = *find++) != 0) {
478 len = strlen(find);
479 do {
480 do {
481 if ((sc = *s++) == 0)
482 return (NULL);
483 } while (sc != c);
484 } while (strncmp(s, find, len) != 0);
485 s--;
486 }
487 return ((char *)s);
488}
489
490/* Taken from libc */
491static char *
492ntoskrnl_strncat(dst, src, n)
493 char *dst;
494 char *src;
495 size_t n;
496{
497 if (n != 0) {
498 char *d = dst;
499 const char *s = src;
500
501 while (*d != 0)
502 d++;
503 do {
504 if ((*d = *s++) == 0)
505 break;
506 d++;
507 } while (--n != 0);
508 *d = 0;
509 }
510 return (dst);
511}
512
513static int
514ntoskrnl_toupper(c)
515 int c;
516{
517 return (toupper(c));
518}
519
520static int
521ntoskrnl_tolower(c)
522 int c;
523{
524 return (tolower(c));
525}
526
527static uint8_t
528RtlEqualUnicodeString(unicode_string *str1, unicode_string *str2,
529 uint8_t caseinsensitive)
530{
531 int i;
532
533 if (str1->us_len != str2->us_len)
534 return (FALSE);
535
536 for (i = 0; i < str1->us_len; i++) {
537 if (caseinsensitive == TRUE) {
538 if (toupper((char)(str1->us_buf[i] & 0xFF)) !=
539 toupper((char)(str2->us_buf[i] & 0xFF)))
540 return (FALSE);
541 } else {
542 if (str1->us_buf[i] != str2->us_buf[i])
543 return (FALSE);
544 }
545 }
546
547 return (TRUE);
548}
549
550static void
551RtlCopyString(dst, src)
552 ansi_string *dst;
553 const ansi_string *src;
554{
555 if (src != NULL && src->as_buf != NULL && dst->as_buf != NULL) {
556 dst->as_len = min(src->as_len, dst->as_maxlen);
557 memcpy(dst->as_buf, src->as_buf, dst->as_len);
558 if (dst->as_len < dst->as_maxlen)
559 dst->as_buf[dst->as_len] = 0;
560 } else
561 dst->as_len = 0;
562}
563
564static void
565RtlCopyUnicodeString(dest, src)
566 unicode_string *dest;
567 unicode_string *src;
568{
569
570 if (dest->us_maxlen >= src->us_len)
571 dest->us_len = src->us_len;
572 else
573 dest->us_len = dest->us_maxlen;
574 memcpy(dest->us_buf, src->us_buf, dest->us_len);
575}
576
577static void
578ntoskrnl_ascii_to_unicode(ascii, unicode, len)
579 char *ascii;
580 uint16_t *unicode;
581 int len;
582{
583 int i;
584 uint16_t *ustr;
585
586 ustr = unicode;
587 for (i = 0; i < len; i++) {
588 *ustr = (uint16_t)ascii[i];
589 ustr++;
590 }
591}
592
593static void
594ntoskrnl_unicode_to_ascii(unicode, ascii, len)
595 uint16_t *unicode;
596 char *ascii;
597 int len;
598{
599 int i;
600 uint8_t *astr;
601
602 astr = ascii;
603 for (i = 0; i < len / 2; i++) {
604 *astr = (uint8_t)unicode[i];
605 astr++;
606 }
607}
608
609uint32_t
610RtlUnicodeStringToAnsiString(ansi_string *dest, unicode_string *src, uint8_t allocate)
611{
612 if (dest == NULL || src == NULL)
613 return (STATUS_INVALID_PARAMETER);
614
615 dest->as_len = src->us_len / 2;
616 if (dest->as_maxlen < dest->as_len)
617 dest->as_len = dest->as_maxlen;
618
619 if (allocate == TRUE) {
620 dest->as_buf = ExAllocatePoolWithTag(NonPagedPool,
621 (src->us_len / 2) + 1, 0);
622 if (dest->as_buf == NULL)
623 return (STATUS_INSUFFICIENT_RESOURCES);
624 dest->as_len = dest->as_maxlen = src->us_len / 2;
625 } else {
626 dest->as_len = src->us_len / 2; /* XXX */
627 if (dest->as_maxlen < dest->as_len)
628 dest->as_len = dest->as_maxlen;
629 }
630
631 ntoskrnl_unicode_to_ascii(src->us_buf, dest->as_buf,
632 dest->as_len * 2);
633
634 return (STATUS_SUCCESS);
635}
636
637uint32_t
638RtlAnsiStringToUnicodeString(unicode_string *dest, ansi_string *src,
639 uint8_t allocate)
640{
641 if (dest == NULL || src == NULL)
642 return (STATUS_INVALID_PARAMETER);
643
644 if (allocate == TRUE) {
645 dest->us_buf = ExAllocatePoolWithTag(NonPagedPool,
646 src->as_len * 2, 0);
647 if (dest->us_buf == NULL)
648 return (STATUS_INSUFFICIENT_RESOURCES);
649 dest->us_len = dest->us_maxlen = strlen(src->as_buf) * 2;
650 } else {
651 dest->us_len = src->as_len * 2; /* XXX */
652 if (dest->us_maxlen < dest->us_len)
653 dest->us_len = dest->us_maxlen;
654 }
655
656 ntoskrnl_ascii_to_unicode(src->as_buf, dest->us_buf,
657 dest->us_len / 2);
658
659 return (STATUS_SUCCESS);
660}
661
662void *
663ExAllocatePoolWithTag(pooltype, len, tag)
664 uint32_t pooltype;
665 size_t len;
666 uint32_t tag;
667{
668 void *buf;
669
670 buf = malloc(len, M_DEVBUF, M_NOWAIT|M_ZERO);
671 if (buf == NULL)
672 return (NULL);
673
674 return (buf);
675}
676
677static void
678ExFreePoolWithTag(buf, tag)
679 void *buf;
680 uint32_t tag;
681{
682 ExFreePool(buf);
683}
684
685void
686ExFreePool(buf)
687 void *buf;
688{
689 free(buf, M_DEVBUF);
690}
691
692uint32_t
693IoAllocateDriverObjectExtension(drv, clid, extlen, ext)
694 driver_object *drv;
695 void *clid;
696 uint32_t extlen;
697 void **ext;
698{
699 custom_extension *ce;
700
701 ce = ExAllocatePoolWithTag(NonPagedPool, sizeof(custom_extension)
702 + extlen, 0);
703
704 if (ce == NULL)
705 return (STATUS_INSUFFICIENT_RESOURCES);
706
707 ce->ce_clid = clid;
708 InsertTailList((&drv->dro_driverext->dre_usrext), (&ce->ce_list));
709
710 *ext = (void *)(ce + 1);
711
712 return (STATUS_SUCCESS);
713}
714
715void *
716IoGetDriverObjectExtension(drv, clid)
717 driver_object *drv;
718 void *clid;
719{
720 list_entry *e;
721 custom_extension *ce;
722
723 /*
724 * Sanity check. Our dummy bus drivers don't have
725 * any driver extentions.
726 */
727
728 if (drv->dro_driverext == NULL)
729 return (NULL);
730
731 e = drv->dro_driverext->dre_usrext.nle_flink;
732 while (e != &drv->dro_driverext->dre_usrext) {
733 ce = (custom_extension *)e;
734 if (ce->ce_clid == clid)
735 return ((void *)(ce + 1));
736 e = e->nle_flink;
737 }
738
739 return (NULL);
740}
741
742
743uint32_t
744IoCreateDevice(driver_object *drv, uint32_t devextlen, unicode_string *devname,
745 uint32_t devtype, uint32_t devchars, uint8_t exclusive,
746 device_object **newdev)
747{
748 device_object *dev;
749
750 dev = ExAllocatePoolWithTag(NonPagedPool, sizeof(device_object), 0);
751 if (dev == NULL)
752 return (STATUS_INSUFFICIENT_RESOURCES);
753
754 dev->do_type = devtype;
755 dev->do_drvobj = drv;
756 dev->do_currirp = NULL;
757 dev->do_flags = 0;
758
759 if (devextlen) {
760 dev->do_devext = ExAllocatePoolWithTag(NonPagedPool,
761 devextlen, 0);
762
763 if (dev->do_devext == NULL) {
764 ExFreePool(dev);
765 return (STATUS_INSUFFICIENT_RESOURCES);
766 }
767
768 bzero(dev->do_devext, devextlen);
769 } else
770 dev->do_devext = NULL;
771
772 dev->do_size = sizeof(device_object) + devextlen;
773 dev->do_refcnt = 1;
774 dev->do_attacheddev = NULL;
775 dev->do_nextdev = NULL;
776 dev->do_devtype = devtype;
777 dev->do_stacksize = 1;
778 dev->do_alignreq = 1;
779 dev->do_characteristics = devchars;
780 dev->do_iotimer = NULL;
781 KeInitializeEvent(&dev->do_devlock, EVENT_TYPE_SYNC, TRUE);
782
783 /*
784 * Vpd is used for disk/tape devices,
785 * but we don't support those. (Yet.)
786 */
787 dev->do_vpb = NULL;
788
789 dev->do_devobj_ext = ExAllocatePoolWithTag(NonPagedPool,
790 sizeof(devobj_extension), 0);
791
792 if (dev->do_devobj_ext == NULL) {
793 if (dev->do_devext != NULL)
794 ExFreePool(dev->do_devext);
795 ExFreePool(dev);
796 return (STATUS_INSUFFICIENT_RESOURCES);
797 }
798
799 dev->do_devobj_ext->dve_type = 0;
800 dev->do_devobj_ext->dve_size = sizeof(devobj_extension);
801 dev->do_devobj_ext->dve_devobj = dev;
802
803 /*
804 * Attach this device to the driver object's list
805 * of devices. Note: this is not the same as attaching
806 * the device to the device stack. The driver's AddDevice
807 * routine must explicitly call IoAddDeviceToDeviceStack()
808 * to do that.
809 */
810
811 if (drv->dro_devobj == NULL) {
812 drv->dro_devobj = dev;
813 dev->do_nextdev = NULL;
814 } else {
815 dev->do_nextdev = drv->dro_devobj;
816 drv->dro_devobj = dev;
817 }
818
819 *newdev = dev;
820
821 return (STATUS_SUCCESS);
822}
823
824void
825IoDeleteDevice(dev)
826 device_object *dev;
827{
828 device_object *prev;
829
830 if (dev == NULL)
831 return;
832
833 if (dev->do_devobj_ext != NULL)
834 ExFreePool(dev->do_devobj_ext);
835
836 if (dev->do_devext != NULL)
837 ExFreePool(dev->do_devext);
838
839 /* Unlink the device from the driver's device list. */
840
841 prev = dev->do_drvobj->dro_devobj;
842 if (prev == dev)
843 dev->do_drvobj->dro_devobj = dev->do_nextdev;
844 else {
845 while (prev->do_nextdev != dev)
846 prev = prev->do_nextdev;
847 prev->do_nextdev = dev->do_nextdev;
848 }
849
850 ExFreePool(dev);
851}
852
853device_object *
854IoGetAttachedDevice(dev)
855 device_object *dev;
856{
857 device_object *d;
858
859 if (dev == NULL)
860 return (NULL);
861
862 d = dev;
863
864 while (d->do_attacheddev != NULL)
865 d = d->do_attacheddev;
866
867 return (d);
868}
869
870static irp *
871IoBuildSynchronousFsdRequest(func, dobj, buf, len, off, event, status)
872 uint32_t func;
873 device_object *dobj;
874 void *buf;
875 uint32_t len;
876 uint64_t *off;
877 nt_kevent *event;
878 io_status_block *status;
879{
880 irp *ip;
881
882 ip = IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status);
883 if (ip == NULL)
884 return (NULL);
885 ip->irp_usrevent = event;
886
887 return (ip);
888}
889
890static irp *
891IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status)
892 uint32_t func;
893 device_object *dobj;
894 void *buf;
895 uint32_t len;
896 uint64_t *off;
897 io_status_block *status;
898{
899 irp *ip;
900 io_stack_location *sl;
901
902 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
903 if (ip == NULL)
904 return (NULL);
905
906 ip->irp_usriostat = status;
907 ip->irp_tail.irp_overlay.irp_thread = NULL;
908
909 sl = IoGetNextIrpStackLocation(ip);
910 sl->isl_major = func;
911 sl->isl_minor = 0;
912 sl->isl_flags = 0;
913 sl->isl_ctl = 0;
914 sl->isl_devobj = dobj;
915 sl->isl_fileobj = NULL;
916 sl->isl_completionfunc = NULL;
917
918 ip->irp_userbuf = buf;
919
920 if (dobj->do_flags & DO_BUFFERED_IO) {
921 ip->irp_assoc.irp_sysbuf =
922 ExAllocatePoolWithTag(NonPagedPool, len, 0);
923 if (ip->irp_assoc.irp_sysbuf == NULL) {
924 IoFreeIrp(ip);
925 return (NULL);
926 }
927 bcopy(buf, ip->irp_assoc.irp_sysbuf, len);
928 }
929
930 if (dobj->do_flags & DO_DIRECT_IO) {
931 ip->irp_mdl = IoAllocateMdl(buf, len, FALSE, FALSE, ip);
932 if (ip->irp_mdl == NULL) {
933 if (ip->irp_assoc.irp_sysbuf != NULL)
934 ExFreePool(ip->irp_assoc.irp_sysbuf);
935 IoFreeIrp(ip);
936 return (NULL);
937 }
938 ip->irp_userbuf = NULL;
939 ip->irp_assoc.irp_sysbuf = NULL;
940 }
941
942 if (func == IRP_MJ_READ) {
943 sl->isl_parameters.isl_read.isl_len = len;
944 if (off != NULL)
945 sl->isl_parameters.isl_read.isl_byteoff = *off;
946 else
947 sl->isl_parameters.isl_read.isl_byteoff = 0;
948 }
949
950 if (func == IRP_MJ_WRITE) {
951 sl->isl_parameters.isl_write.isl_len = len;
952 if (off != NULL)
953 sl->isl_parameters.isl_write.isl_byteoff = *off;
954 else
955 sl->isl_parameters.isl_write.isl_byteoff = 0;
956 }
957
958 return (ip);
959}
960
961static irp *
962IoBuildDeviceIoControlRequest(uint32_t iocode, device_object *dobj, void *ibuf,
963 uint32_t ilen, void *obuf, uint32_t olen, uint8_t isinternal,
964 nt_kevent *event, io_status_block *status)
965{
966 irp *ip;
967 io_stack_location *sl;
968 uint32_t buflen;
969
970 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
971 if (ip == NULL)
972 return (NULL);
973 ip->irp_usrevent = event;
974 ip->irp_usriostat = status;
975 ip->irp_tail.irp_overlay.irp_thread = NULL;
976
977 sl = IoGetNextIrpStackLocation(ip);
978 sl->isl_major = isinternal == TRUE ?
979 IRP_MJ_INTERNAL_DEVICE_CONTROL : IRP_MJ_DEVICE_CONTROL;
980 sl->isl_minor = 0;
981 sl->isl_flags = 0;
982 sl->isl_ctl = 0;
983 sl->isl_devobj = dobj;
984 sl->isl_fileobj = NULL;
985 sl->isl_completionfunc = NULL;
986 sl->isl_parameters.isl_ioctl.isl_iocode = iocode;
987 sl->isl_parameters.isl_ioctl.isl_ibuflen = ilen;
988 sl->isl_parameters.isl_ioctl.isl_obuflen = olen;
989
990 switch(IO_METHOD(iocode)) {
991 case METHOD_BUFFERED:
992 if (ilen > olen)
993 buflen = ilen;
994 else
995 buflen = olen;
996 if (buflen) {
997 ip->irp_assoc.irp_sysbuf =
998 ExAllocatePoolWithTag(NonPagedPool, buflen, 0);
999 if (ip->irp_assoc.irp_sysbuf == NULL) {
1000 IoFreeIrp(ip);
1001 return (NULL);
1002 }
1003 }
1004 if (ilen && ibuf != NULL) {
1005 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1006 bzero((char *)ip->irp_assoc.irp_sysbuf + ilen,
1007 buflen - ilen);
1008 } else
1009 bzero(ip->irp_assoc.irp_sysbuf, ilen);
1010 ip->irp_userbuf = obuf;
1011 break;
1012 case METHOD_IN_DIRECT:
1013 case METHOD_OUT_DIRECT:
1014 if (ilen && ibuf != NULL) {
1015 ip->irp_assoc.irp_sysbuf =
1016 ExAllocatePoolWithTag(NonPagedPool, ilen, 0);
1017 if (ip->irp_assoc.irp_sysbuf == NULL) {
1018 IoFreeIrp(ip);
1019 return (NULL);
1020 }
1021 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1022 }
1023 if (olen && obuf != NULL) {
1024 ip->irp_mdl = IoAllocateMdl(obuf, olen,
1025 FALSE, FALSE, ip);
1026 /*
1027 * Normally we would MmProbeAndLockPages()
1028 * here, but we don't have to in our
1029 * imlementation.
1030 */
1031 }
1032 break;
1033 case METHOD_NEITHER:
1034 ip->irp_userbuf = obuf;
1035 sl->isl_parameters.isl_ioctl.isl_type3ibuf = ibuf;
1036 break;
1037 default:
1038 break;
1039 }
1040
1041 /*
1042 * Ideally, we should associate this IRP with the calling
1043 * thread here.
1044 */
1045
1046 return (ip);
1047}
1048
1049static irp *
1050IoAllocateIrp(uint8_t stsize, uint8_t chargequota)
1051{
1052 irp *i;
1053
1054 i = ExAllocatePoolWithTag(NonPagedPool, IoSizeOfIrp(stsize), 0);
1055 if (i == NULL)
1056 return (NULL);
1057
1058 IoInitializeIrp(i, IoSizeOfIrp(stsize), stsize);
1059
1060 return (i);
1061}
1062
1063static irp *
1064IoMakeAssociatedIrp(irp *ip, uint8_t stsize)
1065{
1066 irp *associrp;
1067
1068 associrp = IoAllocateIrp(stsize, FALSE);
1069 if (associrp == NULL)
1070 return (NULL);
1071
1072 mtx_lock(&ntoskrnl_dispatchlock);
1073 associrp->irp_flags |= IRP_ASSOCIATED_IRP;
1074 associrp->irp_tail.irp_overlay.irp_thread =
1075 ip->irp_tail.irp_overlay.irp_thread;
1076 associrp->irp_assoc.irp_master = ip;
1077 mtx_unlock(&ntoskrnl_dispatchlock);
1078
1079 return (associrp);
1080}
1081
1082static void
1083IoFreeIrp(ip)
1084 irp *ip;
1085{
1086 ExFreePool(ip);
1087}
1088
1089static void
1090IoInitializeIrp(irp *io, uint16_t psize, uint8_t ssize)
1091{
1092 bzero((char *)io, IoSizeOfIrp(ssize));
1093 io->irp_size = psize;
1094 io->irp_stackcnt = ssize;
1095 io->irp_currentstackloc = ssize;
1096 InitializeListHead(&io->irp_thlist);
1097 io->irp_tail.irp_overlay.irp_csl =
1098 (io_stack_location *)(io + 1) + ssize;
1099}
1100
1101static void
1102IoReuseIrp(ip, status)
1103 irp *ip;
1104 uint32_t status;
1105{
1106 uint8_t allocflags;
1107
1108 allocflags = ip->irp_allocflags;
1109 IoInitializeIrp(ip, ip->irp_size, ip->irp_stackcnt);
1110 ip->irp_iostat.isb_status = status;
1111 ip->irp_allocflags = allocflags;
1112}
1113
1114void
1115IoAcquireCancelSpinLock(uint8_t *irql)
1116{
1117 KeAcquireSpinLock(&ntoskrnl_cancellock, irql);
1118}
1119
1120void
1121IoReleaseCancelSpinLock(uint8_t irql)
1122{
1123 KeReleaseSpinLock(&ntoskrnl_cancellock, irql);
1124}
1125
1126uint8_t
1127IoCancelIrp(irp *ip)
1128{
1129 cancel_func cfunc;
1130 uint8_t cancelirql;
1131
1132 IoAcquireCancelSpinLock(&cancelirql);
1133 cfunc = IoSetCancelRoutine(ip, NULL);
1134 ip->irp_cancel = TRUE;
1135 if (cfunc == NULL) {
1136 IoReleaseCancelSpinLock(cancelirql);
1137 return (FALSE);
1138 }
1139 ip->irp_cancelirql = cancelirql;
1140 MSCALL2(cfunc, IoGetCurrentIrpStackLocation(ip)->isl_devobj, ip);
1141 return (uint8_t)IoSetCancelValue(ip, TRUE);
1142}
1143
1144uint32_t
1145IofCallDriver(dobj, ip)
1146 device_object *dobj;
1147 irp *ip;
1148{
1149 driver_object *drvobj;
1150 io_stack_location *sl;
1151 uint32_t status;
1152 driver_dispatch disp;
1153
1154 drvobj = dobj->do_drvobj;
1155
1156 if (ip->irp_currentstackloc <= 0)
1157 panic("IoCallDriver(): out of stack locations");
1158
1159 IoSetNextIrpStackLocation(ip);
1160 sl = IoGetCurrentIrpStackLocation(ip);
1161
1162 sl->isl_devobj = dobj;
1163
1164 disp = drvobj->dro_dispatch[sl->isl_major];
1165 status = MSCALL2(disp, dobj, ip);
1166
1167 return (status);
1168}
1169
1170void
1171IofCompleteRequest(irp *ip, uint8_t prioboost)
1172{
1173 uint32_t status;
1174 device_object *dobj;
1175 io_stack_location *sl;
1176 completion_func cf;
1177
1178 KASSERT(ip->irp_iostat.isb_status != STATUS_PENDING,
1179 ("incorrect IRP(%p) status (STATUS_PENDING)", ip));
1180
1181 sl = IoGetCurrentIrpStackLocation(ip);
1182 IoSkipCurrentIrpStackLocation(ip);
1183
1184 do {
1185 if (sl->isl_ctl & SL_PENDING_RETURNED)
1186 ip->irp_pendingreturned = TRUE;
1187
1188 if (ip->irp_currentstackloc != (ip->irp_stackcnt + 1))
1189 dobj = IoGetCurrentIrpStackLocation(ip)->isl_devobj;
1190 else
1191 dobj = NULL;
1192
1193 if (sl->isl_completionfunc != NULL &&
1194 ((ip->irp_iostat.isb_status == STATUS_SUCCESS &&
1195 sl->isl_ctl & SL_INVOKE_ON_SUCCESS) ||
1196 (ip->irp_iostat.isb_status != STATUS_SUCCESS &&
1197 sl->isl_ctl & SL_INVOKE_ON_ERROR) ||
1198 (ip->irp_cancel == TRUE &&
1199 sl->isl_ctl & SL_INVOKE_ON_CANCEL))) {
1200 cf = sl->isl_completionfunc;
1201 status = MSCALL3(cf, dobj, ip, sl->isl_completionctx);
1202 if (status == STATUS_MORE_PROCESSING_REQUIRED)
1203 return;
1204 } else {
1205 if ((ip->irp_currentstackloc <= ip->irp_stackcnt) &&
1206 (ip->irp_pendingreturned == TRUE))
1207 IoMarkIrpPending(ip);
1208 }
1209
1210 /* move to the next. */
1211 IoSkipCurrentIrpStackLocation(ip);
1212 sl++;
1213 } while (ip->irp_currentstackloc <= (ip->irp_stackcnt + 1));
1214
1215 if (ip->irp_usriostat != NULL)
1216 *ip->irp_usriostat = ip->irp_iostat;
1217 if (ip->irp_usrevent != NULL)
1218 KeSetEvent(ip->irp_usrevent, prioboost, FALSE);
1219
1220 /* Handle any associated IRPs. */
1221
1222 if (ip->irp_flags & IRP_ASSOCIATED_IRP) {
1223 uint32_t masterirpcnt;
1224 irp *masterirp;
1225 mdl *m;
1226
1227 masterirp = ip->irp_assoc.irp_master;
1228 masterirpcnt =
1229 InterlockedDecrement(&masterirp->irp_assoc.irp_irpcnt);
1230
1231 while ((m = ip->irp_mdl) != NULL) {
1232 ip->irp_mdl = m->mdl_next;
1233 IoFreeMdl(m);
1234 }
1235 IoFreeIrp(ip);
1236 if (masterirpcnt == 0)
1237 IoCompleteRequest(masterirp, IO_NO_INCREMENT);
1238 return;
1239 }
1240
1241 /* With any luck, these conditions will never arise. */
1242
1243 if (ip->irp_flags & IRP_PAGING_IO) {
1244 if (ip->irp_mdl != NULL)
1245 IoFreeMdl(ip->irp_mdl);
1246 IoFreeIrp(ip);
1247 }
1248}
1249
1250void
1251ntoskrnl_intr(arg)
1252 void *arg;
1253{
1254 kinterrupt *iobj;
1255 uint8_t irql;
1256 uint8_t claimed;
1257 list_entry *l;
1258
1259 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1260 l = ntoskrnl_intlist.nle_flink;
1261 while (l != &ntoskrnl_intlist) {
1262 iobj = CONTAINING_RECORD(l, kinterrupt, ki_list);
1263 claimed = MSCALL2(iobj->ki_svcfunc, iobj, iobj->ki_svcctx);
1264 if (claimed == TRUE)
1265 break;
1266 l = l->nle_flink;
1267 }
1268 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1269}
1270
1271uint8_t
1272KeAcquireInterruptSpinLock(iobj)
1273 kinterrupt *iobj;
1274{
1275 uint8_t irql;
1276 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1277 return (irql);
1278}
1279
1280void
1281KeReleaseInterruptSpinLock(kinterrupt *iobj, uint8_t irql)
1282{
1283 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1284}
1285
1286uint8_t
1287KeSynchronizeExecution(iobj, syncfunc, syncctx)
1288 kinterrupt *iobj;
1289 void *syncfunc;
1290 void *syncctx;
1291{
1292 uint8_t irql;
1293
1294 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1295 MSCALL1(syncfunc, syncctx);
1296 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1297
1298 return (TRUE);
1299}
1300
1301/*
1302 * IoConnectInterrupt() is passed only the interrupt vector and
1303 * irql that a device wants to use, but no device-specific tag
1304 * of any kind. This conflicts rather badly with FreeBSD's
1305 * bus_setup_intr(), which needs the device_t for the device
1306 * requesting interrupt delivery. In order to bypass this
1307 * inconsistency, we implement a second level of interrupt
1308 * dispatching on top of bus_setup_intr(). All devices use
1309 * ntoskrnl_intr() as their ISR, and any device requesting
1310 * interrupts will be registered with ntoskrnl_intr()'s interrupt
1311 * dispatch list. When an interrupt arrives, we walk the list
1312 * and invoke all the registered ISRs. This effectively makes all
1313 * interrupts shared, but it's the only way to duplicate the
1314 * semantics of IoConnectInterrupt() and IoDisconnectInterrupt() properly.
1315 */
1316
1317uint32_t
1318IoConnectInterrupt(kinterrupt **iobj, void *svcfunc, void *svcctx,
1319 kspin_lock *lock, uint32_t vector, uint8_t irql, uint8_t syncirql,
1320 uint8_t imode, uint8_t shared, uint32_t affinity, uint8_t savefloat)
1321{
1322 uint8_t curirql;
1323
1324 *iobj = ExAllocatePoolWithTag(NonPagedPool, sizeof(kinterrupt), 0);
1325 if (*iobj == NULL)
1326 return (STATUS_INSUFFICIENT_RESOURCES);
1327
1328 (*iobj)->ki_svcfunc = svcfunc;
1329 (*iobj)->ki_svcctx = svcctx;
1330
1331 if (lock == NULL) {
1332 KeInitializeSpinLock(&(*iobj)->ki_lock_priv);
1333 (*iobj)->ki_lock = &(*iobj)->ki_lock_priv;
1334 } else
1335 (*iobj)->ki_lock = lock;
1336
1337 KeAcquireSpinLock(&ntoskrnl_intlock, &curirql);
1338 InsertHeadList((&ntoskrnl_intlist), (&(*iobj)->ki_list));
1339 KeReleaseSpinLock(&ntoskrnl_intlock, curirql);
1340
1341 return (STATUS_SUCCESS);
1342}
1343
1344void
1345IoDisconnectInterrupt(iobj)
1346 kinterrupt *iobj;
1347{
1348 uint8_t irql;
1349
1350 if (iobj == NULL)
1351 return;
1352
1353 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1354 RemoveEntryList((&iobj->ki_list));
1355 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1356
1357 ExFreePool(iobj);
1358}
1359
1360device_object *
1361IoAttachDeviceToDeviceStack(src, dst)
1362 device_object *src;
1363 device_object *dst;
1364{
1365 device_object *attached;
1366
1367 mtx_lock(&ntoskrnl_dispatchlock);
1368 attached = IoGetAttachedDevice(dst);
1369 attached->do_attacheddev = src;
1370 src->do_attacheddev = NULL;
1371 src->do_stacksize = attached->do_stacksize + 1;
1372 mtx_unlock(&ntoskrnl_dispatchlock);
1373
1374 return (attached);
1375}
1376
1377void
1378IoDetachDevice(topdev)
1379 device_object *topdev;
1380{
1381 device_object *tail;
1382
1383 mtx_lock(&ntoskrnl_dispatchlock);
1384
1385 /* First, break the chain. */
1386 tail = topdev->do_attacheddev;
1387 if (tail == NULL) {
1388 mtx_unlock(&ntoskrnl_dispatchlock);
1389 return;
1390 }
1391 topdev->do_attacheddev = tail->do_attacheddev;
1392 topdev->do_refcnt--;
1393
1394 /* Now reduce the stacksize count for the takm_il objects. */
1395
1396 tail = topdev->do_attacheddev;
1397 while (tail != NULL) {
1398 tail->do_stacksize--;
1399 tail = tail->do_attacheddev;
1400 }
1401
1402 mtx_unlock(&ntoskrnl_dispatchlock);
1403}
1404
1405/*
1406 * For the most part, an object is considered signalled if
1407 * dh_sigstate == TRUE. The exception is for mutant objects
1408 * (mutexes), where the logic works like this:
1409 *
1410 * - If the thread already owns the object and sigstate is
1411 * less than or equal to 0, then the object is considered
1412 * signalled (recursive acquisition).
1413 * - If dh_sigstate == 1, the object is also considered
1414 * signalled.
1415 */
1416
1417static int
1418ntoskrnl_is_signalled(obj, td)
1419 nt_dispatch_header *obj;
1420 struct thread *td;
1421{
1422 kmutant *km;
1423
1424 if (obj->dh_type == DISP_TYPE_MUTANT) {
1425 km = (kmutant *)obj;
1426 if ((obj->dh_sigstate <= 0 && km->km_ownerthread == td) ||
1427 obj->dh_sigstate == 1)
1428 return (TRUE);
1429 return (FALSE);
1430 }
1431
1432 if (obj->dh_sigstate > 0)
1433 return (TRUE);
1434 return (FALSE);
1435}
1436
1437static void
1438ntoskrnl_satisfy_wait(obj, td)
1439 nt_dispatch_header *obj;
1440 struct thread *td;
1441{
1442 kmutant *km;
1443
1444 switch (obj->dh_type) {
1445 case DISP_TYPE_MUTANT:
1446 km = (struct kmutant *)obj;
1447 obj->dh_sigstate--;
1448 /*
1449 * If sigstate reaches 0, the mutex is now
1450 * non-signalled (the new thread owns it).
1451 */
1452 if (obj->dh_sigstate == 0) {
1453 km->km_ownerthread = td;
1454 if (km->km_abandoned == TRUE)
1455 km->km_abandoned = FALSE;
1456 }
1457 break;
1458 /* Synchronization objects get reset to unsignalled. */
1459 case DISP_TYPE_SYNCHRONIZATION_EVENT:
1460 case DISP_TYPE_SYNCHRONIZATION_TIMER:
1461 obj->dh_sigstate = 0;
1462 break;
1463 case DISP_TYPE_SEMAPHORE:
1464 obj->dh_sigstate--;
1465 break;
1466 default:
1467 break;
1468 }
1469}
1470
1471static void
1472ntoskrnl_satisfy_multiple_waits(wb)
1473 wait_block *wb;
1474{
1475 wait_block *cur;
1476 struct thread *td;
1477
1478 cur = wb;
1479 td = wb->wb_kthread;
1480
1481 do {
1482 ntoskrnl_satisfy_wait(wb->wb_object, td);
1483 cur->wb_awakened = TRUE;
1484 cur = cur->wb_next;
1485 } while (cur != wb);
1486}
1487
1488/* Always called with dispatcher lock held. */
1489static void
1490ntoskrnl_waittest(obj, increment)
1491 nt_dispatch_header *obj;
1492 uint32_t increment;
1493{
1494 wait_block *w, *next;
1495 list_entry *e;
1496 struct thread *td;
1497 wb_ext *we;
1498 int satisfied;
1499
1500 /*
1501 * Once an object has been signalled, we walk its list of
1502 * wait blocks. If a wait block can be awakened, then satisfy
1503 * waits as necessary and wake the thread.
1504 *
1505 * The rules work like this:
1506 *
1507 * If a wait block is marked as WAITTYPE_ANY, then
1508 * we can satisfy the wait conditions on the current
1509 * object and wake the thread right away. Satisfying
1510 * the wait also has the effect of breaking us out
1511 * of the search loop.
1512 *
1513 * If the object is marked as WAITTYLE_ALL, then the
1514 * wait block will be part of a circularly linked
1515 * list of wait blocks belonging to a waiting thread
1516 * that's sleeping in KeWaitForMultipleObjects(). In
1517 * order to wake the thread, all the objects in the
1518 * wait list must be in the signalled state. If they
1519 * are, we then satisfy all of them and wake the
1520 * thread.
1521 *
1522 */
1523
1524 e = obj->dh_waitlisthead.nle_flink;
1525
1526 while (e != &obj->dh_waitlisthead && obj->dh_sigstate > 0) {
1527 w = CONTAINING_RECORD(e, wait_block, wb_waitlist);
1528 we = w->wb_ext;
1529 td = we->we_td;
1530 satisfied = FALSE;
1531 if (w->wb_waittype == WAITTYPE_ANY) {
1532 /*
1533 * Thread can be awakened if
1534 * any wait is satisfied.
1535 */
1536 ntoskrnl_satisfy_wait(obj, td);
1537 satisfied = TRUE;
1538 w->wb_awakened = TRUE;
1539 } else {
1540 /*
1541 * Thread can only be woken up
1542 * if all waits are satisfied.
1543 * If the thread is waiting on multiple
1544 * objects, they should all be linked
1545 * through the wb_next pointers in the
1546 * wait blocks.
1547 */
1548 satisfied = TRUE;
1549 next = w->wb_next;
1550 while (next != w) {
1551 if (ntoskrnl_is_signalled(obj, td) == FALSE) {
1552 satisfied = FALSE;
1553 break;
1554 }
1555 next = next->wb_next;
1556 }
1557 ntoskrnl_satisfy_multiple_waits(w);
1558 }
1559
1560 if (satisfied == TRUE)
1561 cv_broadcastpri(&we->we_cv,
1562 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
1563 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
1564
1565 e = e->nle_flink;
1566 }
1567}
1568
1569/*
1570 * Return the number of 100 nanosecond intervals since
1571 * January 1, 1601. (?!?!)
1572 */
1573void
1574ntoskrnl_time(tval)
1575 uint64_t *tval;
1576{
1577 struct timespec ts;
1578
1579 nanotime(&ts);
1580 *tval = (uint64_t)ts.tv_nsec / 100 + (uint64_t)ts.tv_sec * 10000000 +
1581 11644473600 * 10000000; /* 100ns ticks from 1601 to 1970 */
1582}
1583
1584static void
1585KeQuerySystemTime(current_time)
1586 uint64_t *current_time;
1587{
1588 ntoskrnl_time(current_time);
1589}
1590
1591static uint32_t
1592KeTickCount(void)
1593{
1594 struct timeval tv;
1595 getmicrouptime(&tv);
1596 return tvtohz(&tv);
1597}
1598
1599
1600/*
1601 * KeWaitForSingleObject() is a tricky beast, because it can be used
1602 * with several different object types: semaphores, timers, events,
1603 * mutexes and threads. Semaphores don't appear very often, but the
1604 * other object types are quite common. KeWaitForSingleObject() is
1605 * what's normally used to acquire a mutex, and it can be used to
1606 * wait for a thread termination.
1607 *
1608 * The Windows NDIS API is implemented in terms of Windows kernel
1609 * primitives, and some of the object manipulation is duplicated in
1610 * NDIS. For example, NDIS has timers and events, which are actually
1611 * Windows kevents and ktimers. Now, you're supposed to only use the
1612 * NDIS variants of these objects within the confines of the NDIS API,
1613 * but there are some naughty developers out there who will use
1614 * KeWaitForSingleObject() on NDIS timer and event objects, so we
1615 * have to support that as well. Conseqently, our NDIS timer and event
1616 * code has to be closely tied into our ntoskrnl timer and event code,
1617 * just as it is in Windows.
1618 *
1619 * KeWaitForSingleObject() may do different things for different kinds
1620 * of objects:
1621 *
1622 * - For events, we check if the event has been signalled. If the
1623 * event is already in the signalled state, we just return immediately,
1624 * otherwise we wait for it to be set to the signalled state by someone
1625 * else calling KeSetEvent(). Events can be either synchronization or
1626 * notification events.
1627 *
1628 * - For timers, if the timer has already fired and the timer is in
1629 * the signalled state, we just return, otherwise we wait on the
1630 * timer. Unlike an event, timers get signalled automatically when
1631 * they expire rather than someone having to trip them manually.
1632 * Timers initialized with KeInitializeTimer() are always notification
1633 * events: KeInitializeTimerEx() lets you initialize a timer as
1634 * either a notification or synchronization event.
1635 *
1636 * - For mutexes, we try to acquire the mutex and if we can't, we wait
1637 * on the mutex until it's available and then grab it. When a mutex is
1638 * released, it enters the signalled state, which wakes up one of the
1639 * threads waiting to acquire it. Mutexes are always synchronization
1640 * events.
1641 *
1642 * - For threads, the only thing we do is wait until the thread object
1643 * enters a signalled state, which occurs when the thread terminates.
1644 * Threads are always notification events.
1645 *
1646 * A notification event wakes up all threads waiting on an object. A
1647 * synchronization event wakes up just one. Also, a synchronization event
1648 * is auto-clearing, which means we automatically set the event back to
1649 * the non-signalled state once the wakeup is done.
1650 */
1651
1652uint32_t
1653KeWaitForSingleObject(void *arg, uint32_t reason, uint32_t mode,
1654 uint8_t alertable, int64_t *duetime)
1655{
1656 wait_block w;
1657 struct thread *td = curthread;
1658 struct timeval tv;
1659 int error = 0;
1660 uint64_t curtime;
1661 wb_ext we;
1662 nt_dispatch_header *obj;
1663
1664 obj = arg;
1665
1666 if (obj == NULL)
1667 return (STATUS_INVALID_PARAMETER);
1668
1669 mtx_lock(&ntoskrnl_dispatchlock);
1670
1671 cv_init(&we.we_cv, "KeWFS");
1672 we.we_td = td;
1673
1674 /*
1675 * Check to see if this object is already signalled,
1676 * and just return without waiting if it is.
1677 */
1678 if (ntoskrnl_is_signalled(obj, td) == TRUE) {
1679 /* Sanity check the signal state value. */
1680 if (obj->dh_sigstate != INT32_MIN) {
1681 ntoskrnl_satisfy_wait(obj, curthread);
1682 mtx_unlock(&ntoskrnl_dispatchlock);
1683 return (STATUS_SUCCESS);
1684 } else {
1685 /*
1686 * There's a limit to how many times we can
1687 * recursively acquire a mutant. If we hit
1688 * the limit, something is very wrong.
1689 */
1690 if (obj->dh_type == DISP_TYPE_MUTANT) {
1691 mtx_unlock(&ntoskrnl_dispatchlock);
1692 panic("mutant limit exceeded");
1693 }
1694 }
1695 }
1696
1697 bzero((char *)&w, sizeof(wait_block));
1698 w.wb_object = obj;
1699 w.wb_ext = &we;
1700 w.wb_waittype = WAITTYPE_ANY;
1701 w.wb_next = &w;
1702 w.wb_waitkey = 0;
1703 w.wb_awakened = FALSE;
1704 w.wb_oldpri = td->td_priority;
1705
1706 InsertTailList((&obj->dh_waitlisthead), (&w.wb_waitlist));
1707
1708 /*
1709 * The timeout value is specified in 100 nanosecond units
1710 * and can be a positive or negative number. If it's positive,
1711 * then the duetime is absolute, and we need to convert it
1712 * to an absolute offset relative to now in order to use it.
1713 * If it's negative, then the duetime is relative and we
1714 * just have to convert the units.
1715 */
1716
1717 if (duetime != NULL) {
1718 if (*duetime < 0) {
1719 tv.tv_sec = - (*duetime) / 10000000;
1720 tv.tv_usec = (- (*duetime) / 10) -
1721 (tv.tv_sec * 1000000);
1722 } else {
1723 ntoskrnl_time(&curtime);
1724 if (*duetime < curtime)
1725 tv.tv_sec = tv.tv_usec = 0;
1726 else {
1727 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1728 tv.tv_usec = ((*duetime) - curtime) / 10 -
1729 (tv.tv_sec * 1000000);
1730 }
1731 }
1732 }
1733
1734 if (duetime == NULL)
1735 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1736 else
1737 error = cv_timedwait(&we.we_cv,
1738 &ntoskrnl_dispatchlock, tvtohz(&tv));
1739
1740 RemoveEntryList(&w.wb_waitlist);
1741
1742 cv_destroy(&we.we_cv);
1743
1744 /* We timed out. Leave the object alone and return status. */
1745
1746 if (error == EWOULDBLOCK) {
1747 mtx_unlock(&ntoskrnl_dispatchlock);
1748 return (STATUS_TIMEOUT);
1749 }
1750
1751 mtx_unlock(&ntoskrnl_dispatchlock);
1752
1753 return (STATUS_SUCCESS);
1754/*
1755 return (KeWaitForMultipleObjects(1, &obj, WAITTYPE_ALL, reason,
1756 mode, alertable, duetime, &w));
1757*/
1758}
1759
1760static uint32_t
1761KeWaitForMultipleObjects(uint32_t cnt, nt_dispatch_header *obj[], uint32_t wtype,
1762 uint32_t reason, uint32_t mode, uint8_t alertable, int64_t *duetime,
1763 wait_block *wb_array)
1764{
1765 struct thread *td = curthread;
1766 wait_block *whead, *w;
1767 wait_block _wb_array[MAX_WAIT_OBJECTS];
1768 nt_dispatch_header *cur;
1769 struct timeval tv;
1770 int i, wcnt = 0, error = 0;
1771 uint64_t curtime;
1772 struct timespec t1, t2;
1773 uint32_t status = STATUS_SUCCESS;
1774 wb_ext we;
1775
1776 if (cnt > MAX_WAIT_OBJECTS)
1777 return (STATUS_INVALID_PARAMETER);
1778 if (cnt > THREAD_WAIT_OBJECTS && wb_array == NULL)
1779 return (STATUS_INVALID_PARAMETER);
1780
1781 mtx_lock(&ntoskrnl_dispatchlock);
1782
1783 cv_init(&we.we_cv, "KeWFM");
1784 we.we_td = td;
1785
1786 if (wb_array == NULL)
1787 whead = _wb_array;
1788 else
1789 whead = wb_array;
1790
1791 bzero((char *)whead, sizeof(wait_block) * cnt);
1792
1793 /* First pass: see if we can satisfy any waits immediately. */
1794
1795 wcnt = 0;
1796 w = whead;
1797
1798 for (i = 0; i < cnt; i++) {
1799 InsertTailList((&obj[i]->dh_waitlisthead),
1800 (&w->wb_waitlist));
1801 w->wb_ext = &we;
1802 w->wb_object = obj[i];
1803 w->wb_waittype = wtype;
1804 w->wb_waitkey = i;
1805 w->wb_awakened = FALSE;
1806 w->wb_oldpri = td->td_priority;
1807 w->wb_next = w + 1;
1808 w++;
1809 wcnt++;
1810 if (ntoskrnl_is_signalled(obj[i], td)) {
1811 /*
1812 * There's a limit to how many times
1813 * we can recursively acquire a mutant.
1814 * If we hit the limit, something
1815 * is very wrong.
1816 */
1817 if (obj[i]->dh_sigstate == INT32_MIN &&
1818 obj[i]->dh_type == DISP_TYPE_MUTANT) {
1819 mtx_unlock(&ntoskrnl_dispatchlock);
1820 panic("mutant limit exceeded");
1821 }
1822
1823 /*
1824 * If this is a WAITTYPE_ANY wait, then
1825 * satisfy the waited object and exit
1826 * right now.
1827 */
1828
1829 if (wtype == WAITTYPE_ANY) {
1830 ntoskrnl_satisfy_wait(obj[i], td);
1831 status = STATUS_WAIT_0 + i;
1832 goto wait_done;
1833 } else {
1834 w--;
1835 wcnt--;
1836 w->wb_object = NULL;
1837 RemoveEntryList(&w->wb_waitlist);
1838 }
1839 }
1840 }
1841
1842 /*
1843 * If this is a WAITTYPE_ALL wait and all objects are
1844 * already signalled, satisfy the waits and exit now.
1845 */
1846
1847 if (wtype == WAITTYPE_ALL && wcnt == 0) {
1848 for (i = 0; i < cnt; i++)
1849 ntoskrnl_satisfy_wait(obj[i], td);
1850 status = STATUS_SUCCESS;
1851 goto wait_done;
1852 }
1853
1854 /*
1855 * Create a circular waitblock list. The waitcount
1856 * must always be non-zero when we get here.
1857 */
1858
1859 (w - 1)->wb_next = whead;
1860
1861 /* Wait on any objects that aren't yet signalled. */
1862
1863 /* Calculate timeout, if any. */
1864
1865 if (duetime != NULL) {
1866 if (*duetime < 0) {
1867 tv.tv_sec = - (*duetime) / 10000000;
1868 tv.tv_usec = (- (*duetime) / 10) -
1869 (tv.tv_sec * 1000000);
1870 } else {
1871 ntoskrnl_time(&curtime);
1872 if (*duetime < curtime)
1873 tv.tv_sec = tv.tv_usec = 0;
1874 else {
1875 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1876 tv.tv_usec = ((*duetime) - curtime) / 10 -
1877 (tv.tv_sec * 1000000);
1878 }
1879 }
1880 }
1881
1882 while (wcnt) {
1883 nanotime(&t1);
1884
1885 if (duetime == NULL)
1886 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1887 else
1888 error = cv_timedwait(&we.we_cv,
1889 &ntoskrnl_dispatchlock, tvtohz(&tv));
1890
1891 /* Wait with timeout expired. */
1892
1893 if (error) {
1894 status = STATUS_TIMEOUT;
1895 goto wait_done;
1896 }
1897
1898 nanotime(&t2);
1899
1900 /* See what's been signalled. */
1901
1902 w = whead;
1903 do {
1904 cur = w->wb_object;
1905 if (ntoskrnl_is_signalled(cur, td) == TRUE ||
1906 w->wb_awakened == TRUE) {
1907 /* Sanity check the signal state value. */
1908 if (cur->dh_sigstate == INT32_MIN &&
1909 cur->dh_type == DISP_TYPE_MUTANT) {
1910 mtx_unlock(&ntoskrnl_dispatchlock);
1911 panic("mutant limit exceeded");
1912 }
1913 wcnt--;
1914 if (wtype == WAITTYPE_ANY) {
1915 status = w->wb_waitkey &
1916 STATUS_WAIT_0;
1917 goto wait_done;
1918 }
1919 }
1920 w = w->wb_next;
1921 } while (w != whead);
1922
1923 /*
1924 * If all objects have been signalled, or if this
1925 * is a WAITTYPE_ANY wait and we were woke up by
1926 * someone, we can bail.
1927 */
1928
1929 if (wcnt == 0) {
1930 status = STATUS_SUCCESS;
1931 goto wait_done;
1932 }
1933
1934 /*
1935 * If this is WAITTYPE_ALL wait, and there's still
1936 * objects that haven't been signalled, deduct the
1937 * time that's elapsed so far from the timeout and
1938 * wait again (or continue waiting indefinitely if
1939 * there's no timeout).
1940 */
1941
1942 if (duetime != NULL) {
1943 tv.tv_sec -= (t2.tv_sec - t1.tv_sec);
1944 tv.tv_usec -= (t2.tv_nsec - t1.tv_nsec) / 1000;
1945 }
1946 }
1947
1948
1949wait_done:
1950
1951 cv_destroy(&we.we_cv);
1952
1953 for (i = 0; i < cnt; i++) {
1954 if (whead[i].wb_object != NULL)
1955 RemoveEntryList(&whead[i].wb_waitlist);
1956
1957 }
1958 mtx_unlock(&ntoskrnl_dispatchlock);
1959
1960 return (status);
1961}
1962
1963static void
1964WRITE_REGISTER_USHORT(uint16_t *reg, uint16_t val)
1965{
1966 bus_space_write_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1967}
1968
1969static uint16_t
1970READ_REGISTER_USHORT(reg)
1971 uint16_t *reg;
1972{
1973 return (bus_space_read_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1974}
1975
1976static void
1977WRITE_REGISTER_ULONG(reg, val)
1978 uint32_t *reg;
1979 uint32_t val;
1980{
1981 bus_space_write_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1982}
1983
1984static uint32_t
1985READ_REGISTER_ULONG(reg)
1986 uint32_t *reg;
1987{
1988 return (bus_space_read_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1989}
1990
1991static uint8_t
1992READ_REGISTER_UCHAR(uint8_t *reg)
1993{
1994 return (bus_space_read_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1995}
1996
1997static void
1998WRITE_REGISTER_UCHAR(uint8_t *reg, uint8_t val)
1999{
2000 bus_space_write_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
2001}
2002
2003static int64_t
2004_allmul(a, b)
2005 int64_t a;
2006 int64_t b;
2007{
2008 return (a * b);
2009}
2010
2011static int64_t
2012_alldiv(a, b)
2013 int64_t a;
2014 int64_t b;
2015{
2016 return (a / b);
2017}
2018
2019static int64_t
2020_allrem(a, b)
2021 int64_t a;
2022 int64_t b;
2023{
2024 return (a % b);
2025}
2026
2027static uint64_t
2028_aullmul(a, b)
2029 uint64_t a;
2030 uint64_t b;
2031{
2032 return (a * b);
2033}
2034
2035static uint64_t
2036_aulldiv(a, b)
2037 uint64_t a;
2038 uint64_t b;
2039{
2040 return (a / b);
2041}
2042
2043static uint64_t
2044_aullrem(a, b)
2045 uint64_t a;
2046 uint64_t b;
2047{
2048 return (a % b);
2049}
2050
2051static int64_t
2052_allshl(int64_t a, uint8_t b)
2053{
2054 return (a << b);
2055}
2056
2057static uint64_t
2058_aullshl(uint64_t a, uint8_t b)
2059{
2060 return (a << b);
2061}
2062
2063static int64_t
2064_allshr(int64_t a, uint8_t b)
2065{
2066 return (a >> b);
2067}
2068
2069static uint64_t
2070_aullshr(uint64_t a, uint8_t b)
2071{
2072 return (a >> b);
2073}
2074
2075static slist_entry *
2076ntoskrnl_pushsl(head, entry)
2077 slist_header *head;
2078 slist_entry *entry;
2079{
2080 slist_entry *oldhead;
2081
2082 oldhead = head->slh_list.slh_next;
2083 entry->sl_next = head->slh_list.slh_next;
2084 head->slh_list.slh_next = entry;
2085 head->slh_list.slh_depth++;
2086 head->slh_list.slh_seq++;
2087
2088 return (oldhead);
2089}
2090
2091static void
2092InitializeSListHead(head)
2093 slist_header *head;
2094{
2095 memset(head, 0, sizeof(*head));
2096}
2097
2098static slist_entry *
2099ntoskrnl_popsl(head)
2100 slist_header *head;
2101{
2102 slist_entry *first;
2103
2104 first = head->slh_list.slh_next;
2105 if (first != NULL) {
2106 head->slh_list.slh_next = first->sl_next;
2107 head->slh_list.slh_depth--;
2108 head->slh_list.slh_seq++;
2109 }
2110
2111 return (first);
2112}
2113
2114/*
2115 * We need this to make lookaside lists work for amd64.
2116 * We pass a pointer to ExAllocatePoolWithTag() the lookaside
2117 * list structure. For amd64 to work right, this has to be a
2118 * pointer to the wrapped version of the routine, not the
2119 * original. Letting the Windows driver invoke the original
2120 * function directly will result in a convention calling
2121 * mismatch and a pretty crash. On x86, this effectively
2122 * becomes a no-op since ipt_func and ipt_wrap are the same.
2123 */
2124
2125static funcptr
2126ntoskrnl_findwrap(func)
2127 funcptr func;
2128{
2129 image_patch_table *patch;
2130
2131 patch = ntoskrnl_functbl;
2132 while (patch->ipt_func != NULL) {
2133 if ((funcptr)patch->ipt_func == func)
2134 return ((funcptr)patch->ipt_wrap);
2135 patch++;
2136 }
2137
2138 return (NULL);
2139}
2140
2141static void
2142ExInitializePagedLookasideList(paged_lookaside_list *lookaside,
2143 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2144 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2145{
2146 bzero((char *)lookaside, sizeof(paged_lookaside_list));
2147
2148 if (size < sizeof(slist_entry))
2149 lookaside->nll_l.gl_size = sizeof(slist_entry);
2150 else
2151 lookaside->nll_l.gl_size = size;
2152 lookaside->nll_l.gl_tag = tag;
2153 if (allocfunc == NULL)
2154 lookaside->nll_l.gl_allocfunc =
2155 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2156 else
2157 lookaside->nll_l.gl_allocfunc = allocfunc;
2158
2159 if (freefunc == NULL)
2160 lookaside->nll_l.gl_freefunc =
2161 ntoskrnl_findwrap((funcptr)ExFreePool);
2162 else
2163 lookaside->nll_l.gl_freefunc = freefunc;
2164
2165#ifdef __i386__
2166 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2167#endif
2168
2169 lookaside->nll_l.gl_type = NonPagedPool;
2170 lookaside->nll_l.gl_depth = depth;
2171 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2172}
2173
2174static void
2175ExDeletePagedLookasideList(lookaside)
2176 paged_lookaside_list *lookaside;
2177{
2178 void *buf;
2179 void (*freefunc)(void *);
2180
2181 freefunc = lookaside->nll_l.gl_freefunc;
2182 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2183 MSCALL1(freefunc, buf);
2184}
2185
2186static void
2187ExInitializeNPagedLookasideList(npaged_lookaside_list *lookaside,
2188 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2189 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2190{
2191 bzero((char *)lookaside, sizeof(npaged_lookaside_list));
2192
2193 if (size < sizeof(slist_entry))
2194 lookaside->nll_l.gl_size = sizeof(slist_entry);
2195 else
2196 lookaside->nll_l.gl_size = size;
2197 lookaside->nll_l.gl_tag = tag;
2198 if (allocfunc == NULL)
2199 lookaside->nll_l.gl_allocfunc =
2200 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2201 else
2202 lookaside->nll_l.gl_allocfunc = allocfunc;
2203
2204 if (freefunc == NULL)
2205 lookaside->nll_l.gl_freefunc =
2206 ntoskrnl_findwrap((funcptr)ExFreePool);
2207 else
2208 lookaside->nll_l.gl_freefunc = freefunc;
2209
2210#ifdef __i386__
2211 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2212#endif
2213
2214 lookaside->nll_l.gl_type = NonPagedPool;
2215 lookaside->nll_l.gl_depth = depth;
2216 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2217}
2218
2219static void
2220ExDeleteNPagedLookasideList(lookaside)
2221 npaged_lookaside_list *lookaside;
2222{
2223 void *buf;
2224 void (*freefunc)(void *);
2225
2226 freefunc = lookaside->nll_l.gl_freefunc;
2227 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2228 MSCALL1(freefunc, buf);
2229}
2230
2231slist_entry *
2232InterlockedPushEntrySList(head, entry)
2233 slist_header *head;
2234 slist_entry *entry;
2235{
2236 slist_entry *oldhead;
2237
2238 mtx_lock_spin(&ntoskrnl_interlock);
2239 oldhead = ntoskrnl_pushsl(head, entry);
2240 mtx_unlock_spin(&ntoskrnl_interlock);
2241
2242 return (oldhead);
2243}
2244
2245slist_entry *
2246InterlockedPopEntrySList(head)
2247 slist_header *head;
2248{
2249 slist_entry *first;
2250
2251 mtx_lock_spin(&ntoskrnl_interlock);
2252 first = ntoskrnl_popsl(head);
2253 mtx_unlock_spin(&ntoskrnl_interlock);
2254
2255 return (first);
2256}
2257
2258static slist_entry *
2259ExInterlockedPushEntrySList(head, entry, lock)
2260 slist_header *head;
2261 slist_entry *entry;
2262 kspin_lock *lock;
2263{
2264 return (InterlockedPushEntrySList(head, entry));
2265}
2266
2267static slist_entry *
2268ExInterlockedPopEntrySList(head, lock)
2269 slist_header *head;
2270 kspin_lock *lock;
2271{
2272 return (InterlockedPopEntrySList(head));
2273}
2274
2275uint16_t
2276ExQueryDepthSList(head)
2277 slist_header *head;
2278{
2279 uint16_t depth;
2280
2281 mtx_lock_spin(&ntoskrnl_interlock);
2282 depth = head->slh_list.slh_depth;
2283 mtx_unlock_spin(&ntoskrnl_interlock);
2284
2285 return (depth);
2286}
2287
2288void
2289KeInitializeSpinLock(lock)
2290 kspin_lock *lock;
2291{
2292 *lock = 0;
2293}
2294
2295#ifdef __i386__
2296void
2297KefAcquireSpinLockAtDpcLevel(lock)
2298 kspin_lock *lock;
2299{
2300#ifdef NTOSKRNL_DEBUG_SPINLOCKS
2301 int i = 0;
2302#endif
2303
2304 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0) {
2305 /* sit and spin */;
2306#ifdef NTOSKRNL_DEBUG_SPINLOCKS
2307 i++;
2308 if (i > 200000000)
2309 panic("DEADLOCK!");
2310#endif
2311 }
2312}
2313
2314void
2315KefReleaseSpinLockFromDpcLevel(lock)
2316 kspin_lock *lock;
2317{
2318 atomic_store_rel_int((volatile u_int *)lock, 0);
2319}
2320
2321uint8_t
2322KeAcquireSpinLockRaiseToDpc(kspin_lock *lock)
2323{
2324 uint8_t oldirql;
2325
2326 if (KeGetCurrentIrql() > DISPATCH_LEVEL)
2327 panic("IRQL_NOT_LESS_THAN_OR_EQUAL");
2328
2329 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
2330 KeAcquireSpinLockAtDpcLevel(lock);
2331
2332 return (oldirql);
2333}
2334#else
2335void
2336KeAcquireSpinLockAtDpcLevel(kspin_lock *lock)
2337{
2338 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0)
2339 /* sit and spin */;
2340}
2341
2342void
2343KeReleaseSpinLockFromDpcLevel(kspin_lock *lock)
2344{
2345 atomic_store_rel_int((volatile u_int *)lock, 0);
2346}
2347#endif /* __i386__ */
2348
2349uintptr_t
2350InterlockedExchange(dst, val)
2351 volatile uint32_t *dst;
2352 uintptr_t val;
2353{
2354 uintptr_t r;
2355
2356 mtx_lock_spin(&ntoskrnl_interlock);
2357 r = *dst;
2358 *dst = val;
2359 mtx_unlock_spin(&ntoskrnl_interlock);
2360
2361 return (r);
2362}
2363
2364static uint32_t
2365InterlockedIncrement(addend)
2366 volatile uint32_t *addend;
2367{
2368 atomic_add_long((volatile u_long *)addend, 1);
2369 return (*addend);
2370}
2371
2372static uint32_t
2373InterlockedDecrement(addend)
2374 volatile uint32_t *addend;
2375{
2376 atomic_subtract_long((volatile u_long *)addend, 1);
2377 return (*addend);
2378}
2379
2380static void
2381ExInterlockedAddLargeStatistic(addend, inc)
2382 uint64_t *addend;
2383 uint32_t inc;
2384{
2385 mtx_lock_spin(&ntoskrnl_interlock);
2386 *addend += inc;
2387 mtx_unlock_spin(&ntoskrnl_interlock);
2388};
2389
2390mdl *
2391IoAllocateMdl(void *vaddr, uint32_t len, uint8_t secondarybuf,
2392 uint8_t chargequota, irp *iopkt)
2393{
2394 mdl *m;
2395 int zone = 0;
2396
2397 if (MmSizeOfMdl(vaddr, len) > MDL_ZONE_SIZE)
2398 m = ExAllocatePoolWithTag(NonPagedPool,
2399 MmSizeOfMdl(vaddr, len), 0);
2400 else {
2401 m = uma_zalloc(mdl_zone, M_NOWAIT | M_ZERO);
2402 zone++;
2403 }
2404
2405 if (m == NULL)
2406 return (NULL);
2407
2408 MmInitializeMdl(m, vaddr, len);
2409
2410 /*
2411 * MmInitializMdl() clears the flags field, so we
2412 * have to set this here. If the MDL came from the
2413 * MDL UMA zone, tag it so we can release it to
2414 * the right place later.
2415 */
2416 if (zone)
2417 m->mdl_flags = MDL_ZONE_ALLOCED;
2418
2419 if (iopkt != NULL) {
2420 if (secondarybuf == TRUE) {
2421 mdl *last;
2422 last = iopkt->irp_mdl;
2423 while (last->mdl_next != NULL)
2424 last = last->mdl_next;
2425 last->mdl_next = m;
2426 } else {
2427 if (iopkt->irp_mdl != NULL)
2428 panic("leaking an MDL in IoAllocateMdl()");
2429 iopkt->irp_mdl = m;
2430 }
2431 }
2432
2433 return (m);
2434}
2435
2436void
2437IoFreeMdl(m)
2438 mdl *m;
2439{
2440 if (m == NULL)
2441 return;
2442
2443 if (m->mdl_flags & MDL_ZONE_ALLOCED)
2444 uma_zfree(mdl_zone, m);
2445 else
2446 ExFreePool(m);
2447}
2448
2449static void *
2450MmAllocateContiguousMemory(size, highest)
2451 uint32_t size;
2452 uint64_t highest;
2453{
2454 void *addr;
2455 size_t pagelength = roundup(size, PAGE_SIZE);
2456
2457 addr = ExAllocatePoolWithTag(NonPagedPool, pagelength, 0);
2458
2459 return (addr);
2460}
2461
2462static void *
2463MmAllocateContiguousMemorySpecifyCache(size, lowest, highest,
2464 boundary, cachetype)
2465 uint32_t size;
2466 uint64_t lowest;
2467 uint64_t highest;
2468 uint64_t boundary;
2469 enum nt_caching_type cachetype;
2470{
2471 vm_memattr_t memattr;
2472 void *ret;
2473
2474 switch (cachetype) {
2475 case MmNonCached:
2476 memattr = VM_MEMATTR_UNCACHEABLE;
2477 break;
2478 case MmWriteCombined:
2479 memattr = VM_MEMATTR_WRITE_COMBINING;
2480 break;
2481 case MmNonCachedUnordered:
2482 memattr = VM_MEMATTR_UNCACHEABLE;
2483 break;
2484 case MmCached:
2485 case MmHardwareCoherentCached:
2486 case MmUSWCCached:
2487 default:
2488 memattr = VM_MEMATTR_DEFAULT;
2489 break;
2490 }
2491
2492 ret = (void *)kmem_alloc_contig(kernel_arena, size, M_ZERO | M_NOWAIT,
2493 lowest, highest, PAGE_SIZE, boundary, memattr);
2494 if (ret != NULL)
2495 malloc_type_allocated(M_DEVBUF, round_page(size));
2496 return (ret);
2497}
2498
2499static void
2500MmFreeContiguousMemory(base)
2501 void *base;
2502{
2503 ExFreePool(base);
2504}
2505
2506static void
2507MmFreeContiguousMemorySpecifyCache(base, size, cachetype)
2508 void *base;
2509 uint32_t size;
2510 enum nt_caching_type cachetype;
2511{
2512 contigfree(base, size, M_DEVBUF);
2513}
2514
2515static uint32_t
2516MmSizeOfMdl(vaddr, len)
2517 void *vaddr;
2518 size_t len;
2519{
2520 uint32_t l;
2521
2522 l = sizeof(struct mdl) +
2523 (sizeof(vm_offset_t *) * SPAN_PAGES(vaddr, len));
2524
2525 return (l);
2526}
2527
2528/*
2529 * The Microsoft documentation says this routine fills in the
2530 * page array of an MDL with the _physical_ page addresses that
2531 * comprise the buffer, but we don't really want to do that here.
2532 * Instead, we just fill in the page array with the kernel virtual
2533 * addresses of the buffers.
2534 */
2535void
2536MmBuildMdlForNonPagedPool(m)
2537 mdl *m;
2538{
2539 vm_offset_t *mdl_pages;
2540 int pagecnt, i;
2541
2542 pagecnt = SPAN_PAGES(m->mdl_byteoffset, m->mdl_bytecount);
2543
2544 if (pagecnt > (m->mdl_size - sizeof(mdl)) / sizeof(vm_offset_t *))
2545 panic("not enough pages in MDL to describe buffer");
2546
2547 mdl_pages = MmGetMdlPfnArray(m);
2548
2549 for (i = 0; i < pagecnt; i++)
2550 *mdl_pages = (vm_offset_t)m->mdl_startva + (i * PAGE_SIZE);
2551
2552 m->mdl_flags |= MDL_SOURCE_IS_NONPAGED_POOL;
2553 m->mdl_mappedsystemva = MmGetMdlVirtualAddress(m);
2554}
2555
2556static void *
2557MmMapLockedPages(mdl *buf, uint8_t accessmode)
2558{
2559 buf->mdl_flags |= MDL_MAPPED_TO_SYSTEM_VA;
2560 return (MmGetMdlVirtualAddress(buf));
2561}
2562
2563static void *
2564MmMapLockedPagesSpecifyCache(mdl *buf, uint8_t accessmode, uint32_t cachetype,
2565 void *vaddr, uint32_t bugcheck, uint32_t prio)
2566{
2567 return (MmMapLockedPages(buf, accessmode));
2568}
2569
2570static void
2571MmUnmapLockedPages(vaddr, buf)
2572 void *vaddr;
2573 mdl *buf;
2574{
2575 buf->mdl_flags &= ~MDL_MAPPED_TO_SYSTEM_VA;
2576}
2577
2578/*
2579 * This function has a problem in that it will break if you
2580 * compile this module without PAE and try to use it on a PAE
2581 * kernel. Unfortunately, there's no way around this at the
2582 * moment. It's slightly less broken that using pmap_kextract().
2583 * You'd think the virtual memory subsystem would help us out
2584 * here, but it doesn't.
2585 */
2586
2587static uint64_t
2588MmGetPhysicalAddress(void *base)
2589{
2590 return (pmap_extract(kernel_map->pmap, (vm_offset_t)base));
2591}
2592
2593void *
2594MmGetSystemRoutineAddress(ustr)
2595 unicode_string *ustr;
2596{
2597 ansi_string astr;
2598
2599 if (RtlUnicodeStringToAnsiString(&astr, ustr, TRUE))
2600 return (NULL);
2601 return (ndis_get_routine_address(ntoskrnl_functbl, astr.as_buf));
2602}
2603
2604uint8_t
2605MmIsAddressValid(vaddr)
2606 void *vaddr;
2607{
2608 if (pmap_extract(kernel_map->pmap, (vm_offset_t)vaddr))
2609 return (TRUE);
2610
2611 return (FALSE);
2612}
2613
2614void *
2615MmMapIoSpace(paddr, len, cachetype)
2616 uint64_t paddr;
2617 uint32_t len;
2618 uint32_t cachetype;
2619{
2620 devclass_t nexus_class;
2621 device_t *nexus_devs, devp;
2622 int nexus_count = 0;
2623 device_t matching_dev = NULL;
2624 struct resource *res;
2625 int i;
2626 vm_offset_t v;
2627
2628 /* There will always be at least one nexus. */
2629
2630 nexus_class = devclass_find("nexus");
2631 devclass_get_devices(nexus_class, &nexus_devs, &nexus_count);
2632
2633 for (i = 0; i < nexus_count; i++) {
2634 devp = nexus_devs[i];
2635 matching_dev = ntoskrnl_finddev(devp, paddr, &res);
2636 if (matching_dev)
2637 break;
2638 }
2639
2640 free(nexus_devs, M_TEMP);
2641
2642 if (matching_dev == NULL)
2643 return (NULL);
2644
2645 v = (vm_offset_t)rman_get_virtual(res);
2646 if (paddr > rman_get_start(res))
2647 v += paddr - rman_get_start(res);
2648
2649 return ((void *)v);
2650}
2651
2652void
2653MmUnmapIoSpace(vaddr, len)
2654 void *vaddr;
2655 size_t len;
2656{
2657}
2658
2659
2660static device_t
2661ntoskrnl_finddev(dev, paddr, res)
2662 device_t dev;
2663 uint64_t paddr;
2664 struct resource **res;
2665{
2666 device_t *children = NULL;
2667 device_t matching_dev;
2668 int childcnt;
2669 struct resource *r;
2670 struct resource_list *rl;
2671 struct resource_list_entry *rle;
2672 uint32_t flags;
2673 int i;
2674
2675 /* We only want devices that have been successfully probed. */
2676
2677 if (device_is_alive(dev) == FALSE)
2678 return (NULL);
2679
2680 rl = BUS_GET_RESOURCE_LIST(device_get_parent(dev), dev);
2681 if (rl != NULL) {
2682 STAILQ_FOREACH(rle, rl, link) {
2683 r = rle->res;
2684
2685 if (r == NULL)
2686 continue;
2687
2688 flags = rman_get_flags(r);
2689
2690 if (rle->type == SYS_RES_MEMORY &&
2691 paddr >= rman_get_start(r) &&
2692 paddr <= rman_get_end(r)) {
2693 if (!(flags & RF_ACTIVE))
2694 bus_activate_resource(dev,
2695 SYS_RES_MEMORY, 0, r);
2696 *res = r;
2697 return (dev);
2698 }
2699 }
2700 }
2701
2702 /*
2703 * If this device has children, do another
2704 * level of recursion to inspect them.
2705 */
2706
2707 device_get_children(dev, &children, &childcnt);
2708
2709 for (i = 0; i < childcnt; i++) {
2710 matching_dev = ntoskrnl_finddev(children[i], paddr, res);
2711 if (matching_dev != NULL) {
2712 free(children, M_TEMP);
2713 return (matching_dev);
2714 }
2715 }
2716
2717
2718 /* Won't somebody please think of the children! */
2719
2720 if (children != NULL)
2721 free(children, M_TEMP);
2722
2723 return (NULL);
2724}
2725
2726/*
2727 * Workitems are unlike DPCs, in that they run in a user-mode thread
2728 * context rather than at DISPATCH_LEVEL in kernel context. In our
2729 * case we run them in kernel context anyway.
2730 */
2731static void
2732ntoskrnl_workitem_thread(arg)
2733 void *arg;
2734{
2735 kdpc_queue *kq;
2736 list_entry *l;
2737 io_workitem *iw;
2738 uint8_t irql;
2739
2740 kq = arg;
2741
2742 InitializeListHead(&kq->kq_disp);
2743 kq->kq_td = curthread;
2744 kq->kq_exit = 0;
2745 KeInitializeSpinLock(&kq->kq_lock);
2746 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
2747
2748 while (1) {
2749 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
2750
2751 KeAcquireSpinLock(&kq->kq_lock, &irql);
2752
2753 if (kq->kq_exit) {
2754 kq->kq_exit = 0;
2755 KeReleaseSpinLock(&kq->kq_lock, irql);
2756 break;
2757 }
2758
2759 while (!IsListEmpty(&kq->kq_disp)) {
2760 l = RemoveHeadList(&kq->kq_disp);
2761 iw = CONTAINING_RECORD(l,
2762 io_workitem, iw_listentry);
2763 InitializeListHead((&iw->iw_listentry));
2764 if (iw->iw_func == NULL)
2765 continue;
2766 KeReleaseSpinLock(&kq->kq_lock, irql);
2767 MSCALL2(iw->iw_func, iw->iw_dobj, iw->iw_ctx);
2768 KeAcquireSpinLock(&kq->kq_lock, &irql);
2769 }
2770
2771 KeReleaseSpinLock(&kq->kq_lock, irql);
2772 }
2773
2774 kproc_exit(0);
2775 return; /* notreached */
2776}
2777
2778static ndis_status
2779RtlCharToInteger(src, base, val)
2780 const char *src;
2781 uint32_t base;
2782 uint32_t *val;
2783{
2784 int negative = 0;
2785 uint32_t res;
2786
2787 if (!src || !val)
2788 return (STATUS_ACCESS_VIOLATION);
2789 while (*src != '\0' && *src <= ' ')
2790 src++;
2791 if (*src == '+')
2792 src++;
2793 else if (*src == '-') {
2794 src++;
2795 negative = 1;
2796 }
2797 if (base == 0) {
2798 base = 10;
2799 if (*src == '0') {
2800 src++;
2801 if (*src == 'b') {
2802 base = 2;
2803 src++;
2804 } else if (*src == 'o') {
2805 base = 8;
2806 src++;
2807 } else if (*src == 'x') {
2808 base = 16;
2809 src++;
2810 }
2811 }
2812 } else if (!(base == 2 || base == 8 || base == 10 || base == 16))
2813 return (STATUS_INVALID_PARAMETER);
2814
2815 for (res = 0; *src; src++) {
2816 int v;
2817 if (isdigit(*src))
2818 v = *src - '0';
2819 else if (isxdigit(*src))
2820 v = tolower(*src) - 'a' + 10;
2821 else
2822 v = base;
2823 if (v >= base)
2824 return (STATUS_INVALID_PARAMETER);
2825 res = res * base + v;
2826 }
2827 *val = negative ? -res : res;
2828 return (STATUS_SUCCESS);
2829}
2830
2831static void
2832ntoskrnl_destroy_workitem_threads(void)
2833{
2834 kdpc_queue *kq;
2835 int i;
2836
2837 for (i = 0; i < WORKITEM_THREADS; i++) {
2838 kq = wq_queues + i;
2839 kq->kq_exit = 1;
2840 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2841 while (kq->kq_exit)
2842 tsleep(kq->kq_td->td_proc, PWAIT, "waitiw", hz/10);
2843 }
2844}
2845
2846io_workitem *
2847IoAllocateWorkItem(dobj)
2848 device_object *dobj;
2849{
2850 io_workitem *iw;
2851
2852 iw = uma_zalloc(iw_zone, M_NOWAIT);
2853 if (iw == NULL)
2854 return (NULL);
2855
2856 InitializeListHead(&iw->iw_listentry);
2857 iw->iw_dobj = dobj;
2858
2859 mtx_lock(&ntoskrnl_dispatchlock);
2860 iw->iw_idx = wq_idx;
2861 WORKIDX_INC(wq_idx);
2862 mtx_unlock(&ntoskrnl_dispatchlock);
2863
2864 return (iw);
2865}
2866
2867void
2868IoFreeWorkItem(iw)
2869 io_workitem *iw;
2870{
2871 uma_zfree(iw_zone, iw);
2872}
2873
2874void
2875IoQueueWorkItem(iw, iw_func, qtype, ctx)
2876 io_workitem *iw;
2877 io_workitem_func iw_func;
2878 uint32_t qtype;
2879 void *ctx;
2880{
2881 kdpc_queue *kq;
2882 list_entry *l;
2883 io_workitem *cur;
2884 uint8_t irql;
2885
2886 kq = wq_queues + iw->iw_idx;
2887
2888 KeAcquireSpinLock(&kq->kq_lock, &irql);
2889
2890 /*
2891 * Traverse the list and make sure this workitem hasn't
2892 * already been inserted. Queuing the same workitem
2893 * twice will hose the list but good.
2894 */
2895
2896 l = kq->kq_disp.nle_flink;
2897 while (l != &kq->kq_disp) {
2898 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2899 if (cur == iw) {
2900 /* Already queued -- do nothing. */
2901 KeReleaseSpinLock(&kq->kq_lock, irql);
2902 return;
2903 }
2904 l = l->nle_flink;
2905 }
2906
2907 iw->iw_func = iw_func;
2908 iw->iw_ctx = ctx;
2909
2910 InsertTailList((&kq->kq_disp), (&iw->iw_listentry));
2911 KeReleaseSpinLock(&kq->kq_lock, irql);
2912
2913 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2914}
2915
2916static void
2917ntoskrnl_workitem(dobj, arg)
2918 device_object *dobj;
2919 void *arg;
2920{
2921 io_workitem *iw;
2922 work_queue_item *w;
2923 work_item_func f;
2924
2925 iw = arg;
2926 w = (work_queue_item *)dobj;
2927 f = (work_item_func)w->wqi_func;
2928 uma_zfree(iw_zone, iw);
2929 MSCALL2(f, w, w->wqi_ctx);
2930}
2931
2932/*
2933 * The ExQueueWorkItem() API is deprecated in Windows XP. Microsoft
2934 * warns that it's unsafe and to use IoQueueWorkItem() instead. The
2935 * problem with ExQueueWorkItem() is that it can't guard against
2936 * the condition where a driver submits a job to the work queue and
2937 * is then unloaded before the job is able to run. IoQueueWorkItem()
2938 * acquires a reference to the device's device_object via the
2939 * object manager and retains it until after the job has completed,
2940 * which prevents the driver from being unloaded before the job
2941 * runs. (We don't currently support this behavior, though hopefully
2942 * that will change once the object manager API is fleshed out a bit.)
2943 *
2944 * Having said all that, the ExQueueWorkItem() API remains, because
2945 * there are still other parts of Windows that use it, including
2946 * NDIS itself: NdisScheduleWorkItem() calls ExQueueWorkItem().
2947 * We fake up the ExQueueWorkItem() API on top of our implementation
2948 * of IoQueueWorkItem(). Workitem thread #3 is reserved exclusively
2949 * for ExQueueWorkItem() jobs, and we pass a pointer to the work
2950 * queue item (provided by the caller) in to IoAllocateWorkItem()
2951 * instead of the device_object. We need to save this pointer so
2952 * we can apply a sanity check: as with the DPC queue and other
2953 * workitem queues, we can't allow the same work queue item to
2954 * be queued twice. If it's already pending, we silently return
2955 */
2956
2957void
2958ExQueueWorkItem(w, qtype)
2959 work_queue_item *w;
2960 uint32_t qtype;
2961{
2962 io_workitem *iw;
2963 io_workitem_func iwf;
2964 kdpc_queue *kq;
2965 list_entry *l;
2966 io_workitem *cur;
2967 uint8_t irql;
2968
2969
2970 /*
2971 * We need to do a special sanity test to make sure
2972 * the ExQueueWorkItem() API isn't used to queue
2973 * the same workitem twice. Rather than checking the
2974 * io_workitem pointer itself, we test the attached
2975 * device object, which is really a pointer to the
2976 * legacy work queue item structure.
2977 */
2978
2979 kq = wq_queues + WORKITEM_LEGACY_THREAD;
2980 KeAcquireSpinLock(&kq->kq_lock, &irql);
2981 l = kq->kq_disp.nle_flink;
2982 while (l != &kq->kq_disp) {
2983 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2984 if (cur->iw_dobj == (device_object *)w) {
2985 /* Already queued -- do nothing. */
2986 KeReleaseSpinLock(&kq->kq_lock, irql);
2987 return;
2988 }
2989 l = l->nle_flink;
2990 }
2991 KeReleaseSpinLock(&kq->kq_lock, irql);
2992
2993 iw = IoAllocateWorkItem((device_object *)w);
2994 if (iw == NULL)
2995 return;
2996
2997 iw->iw_idx = WORKITEM_LEGACY_THREAD;
2998 iwf = (io_workitem_func)ntoskrnl_findwrap((funcptr)ntoskrnl_workitem);
2999 IoQueueWorkItem(iw, iwf, qtype, iw);
3000}
3001
3002static void
3003RtlZeroMemory(dst, len)
3004 void *dst;
3005 size_t len;
3006{
3007 bzero(dst, len);
3008}
3009
3010static void
3011RtlSecureZeroMemory(dst, len)
3012 void *dst;
3013 size_t len;
3014{
3015 memset(dst, 0, len);
3016}
3017
3018static void
3019RtlFillMemory(void *dst, size_t len, uint8_t c)
3020{
3021 memset(dst, c, len);
3022}
3023
3024static void
3025RtlMoveMemory(dst, src, len)
3026 void *dst;
3027 const void *src;
3028 size_t len;
3029{
3030 memmove(dst, src, len);
3031}
3032
3033static void
3034RtlCopyMemory(dst, src, len)
3035 void *dst;
3036 const void *src;
3037 size_t len;
3038{
3039 bcopy(src, dst, len);
3040}
3041
3042static size_t
3043RtlCompareMemory(s1, s2, len)
3044 const void *s1;
3045 const void *s2;
3046 size_t len;
3047{
3048 size_t i;
3049 uint8_t *m1, *m2;
3050
3051 m1 = __DECONST(char *, s1);
3052 m2 = __DECONST(char *, s2);
3053
3054 for (i = 0; i < len && m1[i] == m2[i]; i++);
3055 return (i);
3056}
3057
3058void
3059RtlInitAnsiString(dst, src)
3060 ansi_string *dst;
3061 char *src;
3062{
3063 ansi_string *a;
3064
3065 a = dst;
3066 if (a == NULL)
3067 return;
3068 if (src == NULL) {
3069 a->as_len = a->as_maxlen = 0;
3070 a->as_buf = NULL;
3071 } else {
3072 a->as_buf = src;
3073 a->as_len = a->as_maxlen = strlen(src);
3074 }
3075}
3076
3077void
3078RtlInitUnicodeString(dst, src)
3079 unicode_string *dst;
3080 uint16_t *src;
3081{
3082 unicode_string *u;
3083 int i;
3084
3085 u = dst;
3086 if (u == NULL)
3087 return;
3088 if (src == NULL) {
3089 u->us_len = u->us_maxlen = 0;
3090 u->us_buf = NULL;
3091 } else {
3092 i = 0;
3093 while(src[i] != 0)
3094 i++;
3095 u->us_buf = src;
3096 u->us_len = u->us_maxlen = i * 2;
3097 }
3098}
3099
3100ndis_status
3101RtlUnicodeStringToInteger(ustr, base, val)
3102 unicode_string *ustr;
3103 uint32_t base;
3104 uint32_t *val;
3105{
3106 uint16_t *uchr;
3107 int len, neg = 0;
3108 char abuf[64];
3109 char *astr;
3110
3111 uchr = ustr->us_buf;
3112 len = ustr->us_len;
3113 bzero(abuf, sizeof(abuf));
3114
3115 if ((char)((*uchr) & 0xFF) == '-') {
3116 neg = 1;
3117 uchr++;
3118 len -= 2;
3119 } else if ((char)((*uchr) & 0xFF) == '+') {
3120 neg = 0;
3121 uchr++;
3122 len -= 2;
3123 }
3124
3125 if (base == 0) {
3126 if ((char)((*uchr) & 0xFF) == 'b') {
3127 base = 2;
3128 uchr++;
3129 len -= 2;
3130 } else if ((char)((*uchr) & 0xFF) == 'o') {
3131 base = 8;
3132 uchr++;
3133 len -= 2;
3134 } else if ((char)((*uchr) & 0xFF) == 'x') {
3135 base = 16;
3136 uchr++;
3137 len -= 2;
3138 } else
3139 base = 10;
3140 }
3141
3142 astr = abuf;
3143 if (neg) {
3144 strcpy(astr, "-");
3145 astr++;
3146 }
3147
3148 ntoskrnl_unicode_to_ascii(uchr, astr, len);
3149 *val = strtoul(abuf, NULL, base);
3150
3151 return (STATUS_SUCCESS);
3152}
3153
3154void
3155RtlFreeUnicodeString(ustr)
3156 unicode_string *ustr;
3157{
3158 if (ustr->us_buf == NULL)
3159 return;
3160 ExFreePool(ustr->us_buf);
3161 ustr->us_buf = NULL;
3162}
3163
3164void
3165RtlFreeAnsiString(astr)
3166 ansi_string *astr;
3167{
3168 if (astr->as_buf == NULL)
3169 return;
3170 ExFreePool(astr->as_buf);
3171 astr->as_buf = NULL;
3172}
3173
3174static int
3175atoi(str)
3176 const char *str;
3177{
3178 return (int)strtol(str, (char **)NULL, 10);
3179}
3180
3181static long
3182atol(str)
3183 const char *str;
3184{
3185 return strtol(str, (char **)NULL, 10);
3186}
3187
3188static int
3189rand(void)
3190{
3191 struct timeval tv;
3192
3193 microtime(&tv);
3194 srandom(tv.tv_usec);
3195 return ((int)random());
3196}
3197
3198static void
3199srand(seed)
3200 unsigned int seed;
3201{
3202 srandom(seed);
3203}
3204
3205static uint8_t
3206IoIsWdmVersionAvailable(uint8_t major, uint8_t minor)
3207{
3208 if (major == WDM_MAJOR && minor == WDM_MINOR_WINXP)
3209 return (TRUE);
3210 return (FALSE);
3211}
3212
3213static int32_t
3214IoOpenDeviceRegistryKey(struct device_object *devobj, uint32_t type,
3215 uint32_t mask, void **key)
3216{
3217 return (NDIS_STATUS_INVALID_DEVICE_REQUEST);
3218}
3219
3220static ndis_status
3221IoGetDeviceObjectPointer(name, reqaccess, fileobj, devobj)
3222 unicode_string *name;
3223 uint32_t reqaccess;
3224 void *fileobj;
3225 device_object *devobj;
3226{
3227 return (STATUS_SUCCESS);
3228}
3229
3230static ndis_status
3231IoGetDeviceProperty(devobj, regprop, buflen, prop, reslen)
3232 device_object *devobj;
3233 uint32_t regprop;
3234 uint32_t buflen;
3235 void *prop;
3236 uint32_t *reslen;
3237{
3238 driver_object *drv;
3239 uint16_t **name;
3240
3241 drv = devobj->do_drvobj;
3242
3243 switch (regprop) {
3244 case DEVPROP_DRIVER_KEYNAME:
3245 name = prop;
3246 *name = drv->dro_drivername.us_buf;
3247 *reslen = drv->dro_drivername.us_len;
3248 break;
3249 default:
3250 return (STATUS_INVALID_PARAMETER_2);
3251 break;
3252 }
3253
3254 return (STATUS_SUCCESS);
3255}
3256
3257static void
3258KeInitializeMutex(kmutex, level)
3259 kmutant *kmutex;
3260 uint32_t level;
3261{
3262 InitializeListHead((&kmutex->km_header.dh_waitlisthead));
3263 kmutex->km_abandoned = FALSE;
3264 kmutex->km_apcdisable = 1;
3265 kmutex->km_header.dh_sigstate = 1;
3266 kmutex->km_header.dh_type = DISP_TYPE_MUTANT;
3267 kmutex->km_header.dh_size = sizeof(kmutant) / sizeof(uint32_t);
3268 kmutex->km_ownerthread = NULL;
3269}
3270
3271static uint32_t
3272KeReleaseMutex(kmutant *kmutex, uint8_t kwait)
3273{
3274 uint32_t prevstate;
3275
3276 mtx_lock(&ntoskrnl_dispatchlock);
3277 prevstate = kmutex->km_header.dh_sigstate;
3278 if (kmutex->km_ownerthread != curthread) {
3279 mtx_unlock(&ntoskrnl_dispatchlock);
3280 return (STATUS_MUTANT_NOT_OWNED);
3281 }
3282
3283 kmutex->km_header.dh_sigstate++;
3284 kmutex->km_abandoned = FALSE;
3285
3286 if (kmutex->km_header.dh_sigstate == 1) {
3287 kmutex->km_ownerthread = NULL;
3288 ntoskrnl_waittest(&kmutex->km_header, IO_NO_INCREMENT);
3289 }
3290
3291 mtx_unlock(&ntoskrnl_dispatchlock);
3292
3293 return (prevstate);
3294}
3295
3296static uint32_t
3297KeReadStateMutex(kmutex)
3298 kmutant *kmutex;
3299{
3300 return (kmutex->km_header.dh_sigstate);
3301}
3302
3303void
3304KeInitializeEvent(nt_kevent *kevent, uint32_t type, uint8_t state)
3305{
3306 InitializeListHead((&kevent->k_header.dh_waitlisthead));
3307 kevent->k_header.dh_sigstate = state;
3308 if (type == EVENT_TYPE_NOTIFY)
3309 kevent->k_header.dh_type = DISP_TYPE_NOTIFICATION_EVENT;
3310 else
3311 kevent->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_EVENT;
3312 kevent->k_header.dh_size = sizeof(nt_kevent) / sizeof(uint32_t);
3313}
3314
3315uint32_t
3316KeResetEvent(kevent)
3317 nt_kevent *kevent;
3318{
3319 uint32_t prevstate;
3320
3321 mtx_lock(&ntoskrnl_dispatchlock);
3322 prevstate = kevent->k_header.dh_sigstate;
3323 kevent->k_header.dh_sigstate = FALSE;
3324 mtx_unlock(&ntoskrnl_dispatchlock);
3325
3326 return (prevstate);
3327}
3328
3329uint32_t
3330KeSetEvent(nt_kevent *kevent, uint32_t increment, uint8_t kwait)
3331{
3332 uint32_t prevstate;
3333 wait_block *w;
3334 nt_dispatch_header *dh;
3335 struct thread *td;
3336 wb_ext *we;
3337
3338 mtx_lock(&ntoskrnl_dispatchlock);
3339 prevstate = kevent->k_header.dh_sigstate;
3340 dh = &kevent->k_header;
3341
3342 if (IsListEmpty(&dh->dh_waitlisthead))
3343 /*
3344 * If there's nobody in the waitlist, just set
3345 * the state to signalled.
3346 */
3347 dh->dh_sigstate = 1;
3348 else {
3349 /*
3350 * Get the first waiter. If this is a synchronization
3351 * event, just wake up that one thread (don't bother
3352 * setting the state to signalled since we're supposed
3353 * to automatically clear synchronization events anyway).
3354 *
3355 * If it's a notification event, or the first
3356 * waiter is doing a WAITTYPE_ALL wait, go through
3357 * the full wait satisfaction process.
3358 */
3359 w = CONTAINING_RECORD(dh->dh_waitlisthead.nle_flink,
3360 wait_block, wb_waitlist);
3361 we = w->wb_ext;
3362 td = we->we_td;
3363 if (kevent->k_header.dh_type == DISP_TYPE_NOTIFICATION_EVENT ||
3364 w->wb_waittype == WAITTYPE_ALL) {
3365 if (prevstate == 0) {
3366 dh->dh_sigstate = 1;
3367 ntoskrnl_waittest(dh, increment);
3368 }
3369 } else {
3370 w->wb_awakened |= TRUE;
3371 cv_broadcastpri(&we->we_cv,
3372 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
3373 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
3374 }
3375 }
3376
3377 mtx_unlock(&ntoskrnl_dispatchlock);
3378
3379 return (prevstate);
3380}
3381
3382void
3383KeClearEvent(kevent)
3384 nt_kevent *kevent;
3385{
3386 kevent->k_header.dh_sigstate = FALSE;
3387}
3388
3389uint32_t
3390KeReadStateEvent(kevent)
3391 nt_kevent *kevent;
3392{
3393 return (kevent->k_header.dh_sigstate);
3394}
3395
3396/*
3397 * The object manager in Windows is responsible for managing
3398 * references and access to various types of objects, including
3399 * device_objects, events, threads, timers and so on. However,
3400 * there's a difference in the way objects are handled in user
3401 * mode versus kernel mode.
3402 *
3403 * In user mode (i.e. Win32 applications), all objects are
3404 * managed by the object manager. For example, when you create
3405 * a timer or event object, you actually end up with an
3406 * object_header (for the object manager's bookkeeping
3407 * purposes) and an object body (which contains the actual object
3408 * structure, e.g. ktimer, kevent, etc...). This allows Windows
3409 * to manage resource quotas and to enforce access restrictions
3410 * on basically every kind of system object handled by the kernel.
3411 *
3412 * However, in kernel mode, you only end up using the object
3413 * manager some of the time. For example, in a driver, you create
3414 * a timer object by simply allocating the memory for a ktimer
3415 * structure and initializing it with KeInitializeTimer(). Hence,
3416 * the timer has no object_header and no reference counting or
3417 * security/resource checks are done on it. The assumption in
3418 * this case is that if you're running in kernel mode, you know
3419 * what you're doing, and you're already at an elevated privilege
3420 * anyway.
3421 *
3422 * There are some exceptions to this. The two most important ones
3423 * for our purposes are device_objects and threads. We need to use
3424 * the object manager to do reference counting on device_objects,
3425 * and for threads, you can only get a pointer to a thread's
3426 * dispatch header by using ObReferenceObjectByHandle() on the
3427 * handle returned by PsCreateSystemThread().
3428 */
3429
3430static ndis_status
3431ObReferenceObjectByHandle(ndis_handle handle, uint32_t reqaccess, void *otype,
3432 uint8_t accessmode, void **object, void **handleinfo)
3433{
3434 nt_objref *nr;
3435
3436 nr = malloc(sizeof(nt_objref), M_DEVBUF, M_NOWAIT|M_ZERO);
3437 if (nr == NULL)
3438 return (STATUS_INSUFFICIENT_RESOURCES);
3439
3440 InitializeListHead((&nr->no_dh.dh_waitlisthead));
3441 nr->no_obj = handle;
3442 nr->no_dh.dh_type = DISP_TYPE_THREAD;
3443 nr->no_dh.dh_sigstate = 0;
3444 nr->no_dh.dh_size = (uint8_t)(sizeof(struct thread) /
3445 sizeof(uint32_t));
3446 TAILQ_INSERT_TAIL(&ntoskrnl_reflist, nr, link);
3447 *object = nr;
3448
3449 return (STATUS_SUCCESS);
3450}
3451
3452static void
3453ObfDereferenceObject(object)
3454 void *object;
3455{
3456 nt_objref *nr;
3457
3458 nr = object;
3459 TAILQ_REMOVE(&ntoskrnl_reflist, nr, link);
3460 free(nr, M_DEVBUF);
3461}
3462
3463static uint32_t
3464ZwClose(handle)
3465 ndis_handle handle;
3466{
3467 return (STATUS_SUCCESS);
3468}
3469
3470static uint32_t
3471WmiQueryTraceInformation(traceclass, traceinfo, infolen, reqlen, buf)
3472 uint32_t traceclass;
3473 void *traceinfo;
3474 uint32_t infolen;
3475 uint32_t reqlen;
3476 void *buf;
3477{
3478 return (STATUS_NOT_FOUND);
3479}
3480
3481static uint32_t
3482WmiTraceMessage(uint64_t loghandle, uint32_t messageflags,
3483 void *guid, uint16_t messagenum, ...)
3484{
3485 return (STATUS_SUCCESS);
3486}
3487
3488static uint32_t
3489IoWMIRegistrationControl(dobj, action)
3490 device_object *dobj;
3491 uint32_t action;
3492{
3493 return (STATUS_SUCCESS);
3494}
3495
3496/*
3497 * This is here just in case the thread returns without calling
3498 * PsTerminateSystemThread().
3499 */
3500static void
3501ntoskrnl_thrfunc(arg)
3502 void *arg;
3503{
3504 thread_context *thrctx;
3505 uint32_t (*tfunc)(void *);
3506 void *tctx;
3507 uint32_t rval;
3508
3509 thrctx = arg;
3510 tfunc = thrctx->tc_thrfunc;
3511 tctx = thrctx->tc_thrctx;
3512 free(thrctx, M_TEMP);
3513
3514 rval = MSCALL1(tfunc, tctx);
3515
3516 PsTerminateSystemThread(rval);
3517 return; /* notreached */
3518}
3519
3520static ndis_status
3521PsCreateSystemThread(handle, reqaccess, objattrs, phandle,
3522 clientid, thrfunc, thrctx)
3523 ndis_handle *handle;
3524 uint32_t reqaccess;
3525 void *objattrs;
3526 ndis_handle phandle;
3527 void *clientid;
3528 void *thrfunc;
3529 void *thrctx;
3530{
3531 int error;
3532 thread_context *tc;
3533 struct proc *p;
3534
3535 tc = malloc(sizeof(thread_context), M_TEMP, M_NOWAIT);
3536 if (tc == NULL)
3537 return (STATUS_INSUFFICIENT_RESOURCES);
3538
3539 tc->tc_thrctx = thrctx;
3540 tc->tc_thrfunc = thrfunc;
3541
3542 error = kproc_create(ntoskrnl_thrfunc, tc, &p,
3543 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Kthread %d", ntoskrnl_kth);
3544
3545 if (error) {
3546 free(tc, M_TEMP);
3547 return (STATUS_INSUFFICIENT_RESOURCES);
3548 }
3549
3550 *handle = p;
3551 ntoskrnl_kth++;
3552
3553 return (STATUS_SUCCESS);
3554}
3555
3556/*
3557 * In Windows, the exit of a thread is an event that you're allowed
3558 * to wait on, assuming you've obtained a reference to the thread using
3559 * ObReferenceObjectByHandle(). Unfortunately, the only way we can
3560 * simulate this behavior is to register each thread we create in a
3561 * reference list, and if someone holds a reference to us, we poke
3562 * them.
3563 */
3564static ndis_status
3565PsTerminateSystemThread(status)
3566 ndis_status status;
3567{
3568 struct nt_objref *nr;
3569
3570 mtx_lock(&ntoskrnl_dispatchlock);
3571 TAILQ_FOREACH(nr, &ntoskrnl_reflist, link) {
3572 if (nr->no_obj != curthread->td_proc)
3573 continue;
3574 nr->no_dh.dh_sigstate = 1;
3575 ntoskrnl_waittest(&nr->no_dh, IO_NO_INCREMENT);
3576 break;
3577 }
3578 mtx_unlock(&ntoskrnl_dispatchlock);
3579
3580 ntoskrnl_kth--;
3581
3582 kproc_exit(0);
3583 return (0); /* notreached */
3584}
3585
3586static uint32_t
3587DbgPrint(char *fmt, ...)
3588{
3589 va_list ap;
3590
3591 if (bootverbose) {
3592 va_start(ap, fmt);
3593 vprintf(fmt, ap);
3594 va_end(ap);
3595 }
3596
3597 return (STATUS_SUCCESS);
3598}
3599
3600static void
3601DbgBreakPoint(void)
3602{
3603
3604 kdb_enter(KDB_WHY_NDIS, "DbgBreakPoint(): breakpoint");
3605}
3606
3607static void
3608KeBugCheckEx(code, param1, param2, param3, param4)
3609 uint32_t code;
3610 u_long param1;
3611 u_long param2;
3612 u_long param3;
3613 u_long param4;
3614{
3615 panic("KeBugCheckEx: STOP 0x%X", code);
3616}
3617
3618static void
3619ntoskrnl_timercall(arg)
3620 void *arg;
3621{
3622 ktimer *timer;
3623 struct timeval tv;
3624 kdpc *dpc;
3625
3626 mtx_lock(&ntoskrnl_dispatchlock);
3627
3628 timer = arg;
3629
3630#ifdef NTOSKRNL_DEBUG_TIMERS
3631 ntoskrnl_timer_fires++;
3632#endif
3633 ntoskrnl_remove_timer(timer);
3634
3635 /*
3636 * This should never happen, but complain
3637 * if it does.
3638 */
3639
3640 if (timer->k_header.dh_inserted == FALSE) {
3641 mtx_unlock(&ntoskrnl_dispatchlock);
3642 printf("NTOS: timer %p fired even though "
3643 "it was canceled\n", timer);
3644 return;
3645 }
3646
3647 /* Mark the timer as no longer being on the timer queue. */
3648
3649 timer->k_header.dh_inserted = FALSE;
3650
3651 /* Now signal the object and satisfy any waits on it. */
3652
3653 timer->k_header.dh_sigstate = 1;
3654 ntoskrnl_waittest(&timer->k_header, IO_NO_INCREMENT);
3655
3656 /*
3657 * If this is a periodic timer, re-arm it
3658 * so it will fire again. We do this before
3659 * calling any deferred procedure calls because
3660 * it's possible the DPC might cancel the timer,
3661 * in which case it would be wrong for us to
3662 * re-arm it again afterwards.
3663 */
3664
3665 if (timer->k_period) {
3666 tv.tv_sec = 0;
3667 tv.tv_usec = timer->k_period * 1000;
3668 timer->k_header.dh_inserted = TRUE;
3669 ntoskrnl_insert_timer(timer, tvtohz(&tv));
3670#ifdef NTOSKRNL_DEBUG_TIMERS
3671 ntoskrnl_timer_reloads++;
3672#endif
3673 }
3674
3675 dpc = timer->k_dpc;
3676
3677 mtx_unlock(&ntoskrnl_dispatchlock);
3678
3679 /* If there's a DPC associated with the timer, queue it up. */
3680
3681 if (dpc != NULL)
3682 KeInsertQueueDpc(dpc, NULL, NULL);
3683}
3684
3685#ifdef NTOSKRNL_DEBUG_TIMERS
3686static int
3687sysctl_show_timers(SYSCTL_HANDLER_ARGS)
3688{
3689 int ret;
3690
3691 ret = 0;
3692 ntoskrnl_show_timers();
3693 return (sysctl_handle_int(oidp, &ret, 0, req));
3694}
3695
3696static void
3697ntoskrnl_show_timers()
3698{
3699 int i = 0;
3700 list_entry *l;
3701
3702 mtx_lock_spin(&ntoskrnl_calllock);
3703 l = ntoskrnl_calllist.nle_flink;
3704 while(l != &ntoskrnl_calllist) {
3705 i++;
3706 l = l->nle_flink;
3707 }
3708 mtx_unlock_spin(&ntoskrnl_calllock);
3709
3710 printf("\n");
3711 printf("%d timers available (out of %d)\n", i, NTOSKRNL_TIMEOUTS);
3712 printf("timer sets: %qu\n", ntoskrnl_timer_sets);
3713 printf("timer reloads: %qu\n", ntoskrnl_timer_reloads);
3714 printf("timer cancels: %qu\n", ntoskrnl_timer_cancels);
3715 printf("timer fires: %qu\n", ntoskrnl_timer_fires);
3716 printf("\n");
3717}
3718#endif
3719
3720/*
3721 * Must be called with dispatcher lock held.
3722 */
3723
3724static void
3725ntoskrnl_insert_timer(timer, ticks)
3726 ktimer *timer;
3727 int ticks;
3728{
3729 callout_entry *e;
3730 list_entry *l;
3731 struct callout *c;
3732
3733 /*
3734 * Try and allocate a timer.
3735 */
3736 mtx_lock_spin(&ntoskrnl_calllock);
3737 if (IsListEmpty(&ntoskrnl_calllist)) {
3738 mtx_unlock_spin(&ntoskrnl_calllock);
3739#ifdef NTOSKRNL_DEBUG_TIMERS
3740 ntoskrnl_show_timers();
3741#endif
3742 panic("out of timers!");
3743 }
3744 l = RemoveHeadList(&ntoskrnl_calllist);
3745 mtx_unlock_spin(&ntoskrnl_calllock);
3746
3747 e = CONTAINING_RECORD(l, callout_entry, ce_list);
3748 c = &e->ce_callout;
3749
3750 timer->k_callout = c;
3751
3752 callout_init(c, 1);
3753 callout_reset(c, ticks, ntoskrnl_timercall, timer);
3754}
3755
3756static void
3757ntoskrnl_remove_timer(timer)
3758 ktimer *timer;
3759{
3760 callout_entry *e;
3761
3762 e = (callout_entry *)timer->k_callout;
3763 callout_stop(timer->k_callout);
3764
3765 mtx_lock_spin(&ntoskrnl_calllock);
3766 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
3767 mtx_unlock_spin(&ntoskrnl_calllock);
3768}
3769
3770void
3771KeInitializeTimer(timer)
3772 ktimer *timer;
3773{
3774 if (timer == NULL)
3775 return;
3776
3777 KeInitializeTimerEx(timer, EVENT_TYPE_NOTIFY);
3778}
3779
3780void
3781KeInitializeTimerEx(timer, type)
3782 ktimer *timer;
3783 uint32_t type;
3784{
3785 if (timer == NULL)
3786 return;
3787
3788 bzero((char *)timer, sizeof(ktimer));
3789 InitializeListHead((&timer->k_header.dh_waitlisthead));
3790 timer->k_header.dh_sigstate = FALSE;
3791 timer->k_header.dh_inserted = FALSE;
3792 if (type == EVENT_TYPE_NOTIFY)
3793 timer->k_header.dh_type = DISP_TYPE_NOTIFICATION_TIMER;
3794 else
3795 timer->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_TIMER;
3796 timer->k_header.dh_size = sizeof(ktimer) / sizeof(uint32_t);
3797}
3798
3799/*
3800 * DPC subsystem. A Windows Defered Procedure Call has the following
3801 * properties:
3802 * - It runs at DISPATCH_LEVEL.
3803 * - It can have one of 3 importance values that control when it
3804 * runs relative to other DPCs in the queue.
3805 * - On SMP systems, it can be set to run on a specific processor.
3806 * In order to satisfy the last property, we create a DPC thread for
3807 * each CPU in the system and bind it to that CPU. Each thread
3808 * maintains three queues with different importance levels, which
3809 * will be processed in order from lowest to highest.
3810 *
3811 * In Windows, interrupt handlers run as DPCs. (Not to be confused
3812 * with ISRs, which run in interrupt context and can preempt DPCs.)
3813 * ISRs are given the highest importance so that they'll take
3814 * precedence over timers and other things.
3815 */
3816
3817static void
3818ntoskrnl_dpc_thread(arg)
3819 void *arg;
3820{
3821 kdpc_queue *kq;
3822 kdpc *d;
3823 list_entry *l;
3824 uint8_t irql;
3825
3826 kq = arg;
3827
3828 InitializeListHead(&kq->kq_disp);
3829 kq->kq_td = curthread;
3830 kq->kq_exit = 0;
3831 kq->kq_running = FALSE;
3832 KeInitializeSpinLock(&kq->kq_lock);
3833 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
3834 KeInitializeEvent(&kq->kq_done, EVENT_TYPE_SYNC, FALSE);
3835
3836 /*
3837 * Elevate our priority. DPCs are used to run interrupt
3838 * handlers, and they should trigger as soon as possible
3839 * once scheduled by an ISR.
3840 */
3841
3842 thread_lock(curthread);
3843#ifdef NTOSKRNL_MULTIPLE_DPCS
3844 sched_bind(curthread, kq->kq_cpu);
3845#endif
3846 sched_prio(curthread, PRI_MIN_KERN);
3847 thread_unlock(curthread);
3848
3849 while (1) {
3850 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
3851
3852 KeAcquireSpinLock(&kq->kq_lock, &irql);
3853
3854 if (kq->kq_exit) {
3855 kq->kq_exit = 0;
3856 KeReleaseSpinLock(&kq->kq_lock, irql);
3857 break;
3858 }
3859
3860 kq->kq_running = TRUE;
3861
3862 while (!IsListEmpty(&kq->kq_disp)) {
3863 l = RemoveHeadList((&kq->kq_disp));
3864 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3865 InitializeListHead((&d->k_dpclistentry));
3866 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
3867 MSCALL4(d->k_deferedfunc, d, d->k_deferredctx,
3868 d->k_sysarg1, d->k_sysarg2);
3869 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3870 }
3871
3872 kq->kq_running = FALSE;
3873
3874 KeReleaseSpinLock(&kq->kq_lock, irql);
3875
3876 KeSetEvent(&kq->kq_done, IO_NO_INCREMENT, FALSE);
3877 }
3878
3879 kproc_exit(0);
3880 return; /* notreached */
3881}
3882
3883static void
3884ntoskrnl_destroy_dpc_threads(void)
3885{
3886 kdpc_queue *kq;
3887 kdpc dpc;
3888 int i;
3889
3890 kq = kq_queues;
3891#ifdef NTOSKRNL_MULTIPLE_DPCS
3892 for (i = 0; i < mp_ncpus; i++) {
3893#else
3894 for (i = 0; i < 1; i++) {
3895#endif
3896 kq += i;
3897
3898 kq->kq_exit = 1;
3899 KeInitializeDpc(&dpc, NULL, NULL);
3900 KeSetTargetProcessorDpc(&dpc, i);
3901 KeInsertQueueDpc(&dpc, NULL, NULL);
3902 while (kq->kq_exit)
3903 tsleep(kq->kq_td->td_proc, PWAIT, "dpcw", hz/10);
3904 }
3905}
3906
3907static uint8_t
3908ntoskrnl_insert_dpc(head, dpc)
3909 list_entry *head;
3910 kdpc *dpc;
3911{
3912 list_entry *l;
3913 kdpc *d;
3914
3915 l = head->nle_flink;
3916 while (l != head) {
3917 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3918 if (d == dpc)
3919 return (FALSE);
3920 l = l->nle_flink;
3921 }
3922
3923 if (dpc->k_importance == KDPC_IMPORTANCE_LOW)
3924 InsertTailList((head), (&dpc->k_dpclistentry));
3925 else
3926 InsertHeadList((head), (&dpc->k_dpclistentry));
3927
3928 return (TRUE);
3929}
3930
3931void
3932KeInitializeDpc(dpc, dpcfunc, dpcctx)
3933 kdpc *dpc;
3934 void *dpcfunc;
3935 void *dpcctx;
3936{
3937
3938 if (dpc == NULL)
3939 return;
3940
3941 dpc->k_deferedfunc = dpcfunc;
3942 dpc->k_deferredctx = dpcctx;
3943 dpc->k_num = KDPC_CPU_DEFAULT;
3944 dpc->k_importance = KDPC_IMPORTANCE_MEDIUM;
3945 InitializeListHead((&dpc->k_dpclistentry));
3946}
3947
3948uint8_t
3949KeInsertQueueDpc(dpc, sysarg1, sysarg2)
3950 kdpc *dpc;
3951 void *sysarg1;
3952 void *sysarg2;
3953{
3954 kdpc_queue *kq;
3955 uint8_t r;
3956 uint8_t irql;
3957
3958 if (dpc == NULL)
3959 return (FALSE);
3960
3961 kq = kq_queues;
3962
3963#ifdef NTOSKRNL_MULTIPLE_DPCS
3964 KeRaiseIrql(DISPATCH_LEVEL, &irql);
3965
3966 /*
3967 * By default, the DPC is queued to run on the same CPU
3968 * that scheduled it.
3969 */
3970
3971 if (dpc->k_num == KDPC_CPU_DEFAULT)
3972 kq += curthread->td_oncpu;
3973 else
3974 kq += dpc->k_num;
3975 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3976#else
3977 KeAcquireSpinLock(&kq->kq_lock, &irql);
3978#endif
3979
3980 r = ntoskrnl_insert_dpc(&kq->kq_disp, dpc);
3981 if (r == TRUE) {
3982 dpc->k_sysarg1 = sysarg1;
3983 dpc->k_sysarg2 = sysarg2;
3984 }
3985 KeReleaseSpinLock(&kq->kq_lock, irql);
3986
3987 if (r == FALSE)
3988 return (r);
3989
3990 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
3991
3992 return (r);
3993}
3994
3995uint8_t
3996KeRemoveQueueDpc(dpc)
3997 kdpc *dpc;
3998{
3999 kdpc_queue *kq;
4000 uint8_t irql;
4001
4002 if (dpc == NULL)
4003 return (FALSE);
4004
4005#ifdef NTOSKRNL_MULTIPLE_DPCS
4006 KeRaiseIrql(DISPATCH_LEVEL, &irql);
4007
4008 kq = kq_queues + dpc->k_num;
4009
4010 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
4011#else
4012 kq = kq_queues;
4013 KeAcquireSpinLock(&kq->kq_lock, &irql);
4014#endif
4015
4016 if (dpc->k_dpclistentry.nle_flink == &dpc->k_dpclistentry) {
4017 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
4018 KeLowerIrql(irql);
4019 return (FALSE);
4020 }
4021
4022 RemoveEntryList((&dpc->k_dpclistentry));
4023 InitializeListHead((&dpc->k_dpclistentry));
4024
4025 KeReleaseSpinLock(&kq->kq_lock, irql);
4026
4027 return (TRUE);
4028}
4029
4030void
4031KeSetImportanceDpc(dpc, imp)
4032 kdpc *dpc;
4033 uint32_t imp;
4034{
4035 if (imp != KDPC_IMPORTANCE_LOW &&
4036 imp != KDPC_IMPORTANCE_MEDIUM &&
4037 imp != KDPC_IMPORTANCE_HIGH)
4038 return;
4039
4040 dpc->k_importance = (uint8_t)imp;
4041}
4042
4043void
4044KeSetTargetProcessorDpc(kdpc *dpc, uint8_t cpu)
4045{
4046 if (cpu > mp_ncpus)
4047 return;
4048
4049 dpc->k_num = cpu;
4050}
4051
4052void
4053KeFlushQueuedDpcs(void)
4054{
4055 kdpc_queue *kq;
4056 int i;
4057
4058 /*
4059 * Poke each DPC queue and wait
4060 * for them to drain.
4061 */
4062
4063#ifdef NTOSKRNL_MULTIPLE_DPCS
4064 for (i = 0; i < mp_ncpus; i++) {
4065#else
4066 for (i = 0; i < 1; i++) {
4067#endif
4068 kq = kq_queues + i;
4069 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
4070 KeWaitForSingleObject(&kq->kq_done, 0, 0, TRUE, NULL);
4071 }
4072}
4073
4074uint32_t
4075KeGetCurrentProcessorNumber(void)
4076{
4077 return ((uint32_t)curthread->td_oncpu);
4078}
4079
4080uint8_t
4081KeSetTimerEx(timer, duetime, period, dpc)
4082 ktimer *timer;
4083 int64_t duetime;
4084 uint32_t period;
4085 kdpc *dpc;
4086{
4087 struct timeval tv;
4088 uint64_t curtime;
4089 uint8_t pending;
4090
4091 if (timer == NULL)
4092 return (FALSE);
4093
4094 mtx_lock(&ntoskrnl_dispatchlock);
4095
4096 if (timer->k_header.dh_inserted == TRUE) {
4097 ntoskrnl_remove_timer(timer);
4098#ifdef NTOSKRNL_DEBUG_TIMERS
4099 ntoskrnl_timer_cancels++;
4100#endif
4101 timer->k_header.dh_inserted = FALSE;
4102 pending = TRUE;
4103 } else
4104 pending = FALSE;
4105
4106 timer->k_duetime = duetime;
4107 timer->k_period = period;
4108 timer->k_header.dh_sigstate = FALSE;
4109 timer->k_dpc = dpc;
4110
4111 if (duetime < 0) {
4112 tv.tv_sec = - (duetime) / 10000000;
4113 tv.tv_usec = (- (duetime) / 10) -
4114 (tv.tv_sec * 1000000);
4115 } else {
4116 ntoskrnl_time(&curtime);
4117 if (duetime < curtime)
4118 tv.tv_sec = tv.tv_usec = 0;
4119 else {
4120 tv.tv_sec = ((duetime) - curtime) / 10000000;
4121 tv.tv_usec = ((duetime) - curtime) / 10 -
4122 (tv.tv_sec * 1000000);
4123 }
4124 }
4125
4126 timer->k_header.dh_inserted = TRUE;
4127 ntoskrnl_insert_timer(timer, tvtohz(&tv));
4128#ifdef NTOSKRNL_DEBUG_TIMERS
4129 ntoskrnl_timer_sets++;
4130#endif
4131
4132 mtx_unlock(&ntoskrnl_dispatchlock);
4133
4134 return (pending);
4135}
4136
4137uint8_t
4138KeSetTimer(timer, duetime, dpc)
4139 ktimer *timer;
4140 int64_t duetime;
4141 kdpc *dpc;
4142{
4143 return (KeSetTimerEx(timer, duetime, 0, dpc));
4144}
4145
4146/*
4147 * The Windows DDK documentation seems to say that cancelling
4148 * a timer that has a DPC will result in the DPC also being
4149 * cancelled, but this isn't really the case.
4150 */
4151
4152uint8_t
4153KeCancelTimer(timer)
4154 ktimer *timer;
4155{
4156 uint8_t pending;
4157
4158 if (timer == NULL)
4159 return (FALSE);
4160
4161 mtx_lock(&ntoskrnl_dispatchlock);
4162
4163 pending = timer->k_header.dh_inserted;
4164
4165 if (timer->k_header.dh_inserted == TRUE) {
4166 timer->k_header.dh_inserted = FALSE;
4167 ntoskrnl_remove_timer(timer);
4168#ifdef NTOSKRNL_DEBUG_TIMERS
4169 ntoskrnl_timer_cancels++;
4170#endif
4171 }
4172
4173 mtx_unlock(&ntoskrnl_dispatchlock);
4174
4175 return (pending);
4176}
4177
4178uint8_t
4179KeReadStateTimer(timer)
4180 ktimer *timer;
4181{
4182 return (timer->k_header.dh_sigstate);
4183}
4184
4185static int32_t
4186KeDelayExecutionThread(uint8_t wait_mode, uint8_t alertable, int64_t *interval)
4187{
4188 ktimer timer;
4189
4190 if (wait_mode != 0)
4191 panic("invalid wait_mode %d", wait_mode);
4192
4193 KeInitializeTimer(&timer);
4194 KeSetTimer(&timer, *interval, NULL);
4195 KeWaitForSingleObject(&timer, 0, 0, alertable, NULL);
4196
4197 return STATUS_SUCCESS;
4198}
4199
4200static uint64_t
4201KeQueryInterruptTime(void)
4202{
4203 int ticks;
4204 struct timeval tv;
4205
4206 getmicrouptime(&tv);
4207
4208 ticks = tvtohz(&tv);
4209
|
4211}
4212
4213static struct thread *
4214KeGetCurrentThread(void)
4215{
4216
4217 return curthread;
4218}
4219
4220static int32_t
4221KeSetPriorityThread(td, pri)
4222 struct thread *td;
4223 int32_t pri;
4224{
4225 int32_t old;
4226
4227 if (td == NULL)
4228 return LOW_REALTIME_PRIORITY;
4229
4230 if (td->td_priority <= PRI_MIN_KERN)
4231 old = HIGH_PRIORITY;
4232 else if (td->td_priority >= PRI_MAX_KERN)
4233 old = LOW_PRIORITY;
4234 else
4235 old = LOW_REALTIME_PRIORITY;
4236
4237 thread_lock(td);
4238 if (pri == HIGH_PRIORITY)
4239 sched_prio(td, PRI_MIN_KERN);
4240 if (pri == LOW_REALTIME_PRIORITY)
4241 sched_prio(td, PRI_MIN_KERN + (PRI_MAX_KERN - PRI_MIN_KERN) / 2);
4242 if (pri == LOW_PRIORITY)
4243 sched_prio(td, PRI_MAX_KERN);
4244 thread_unlock(td);
4245
4246 return old;
4247}
4248
4249static void
4250dummy()
4251{
4252 printf("ntoskrnl dummy called...\n");
4253}
4254
4255
4256image_patch_table ntoskrnl_functbl[] = {
4257 IMPORT_SFUNC(RtlZeroMemory, 2),
4258 IMPORT_SFUNC(RtlSecureZeroMemory, 2),
4259 IMPORT_SFUNC(RtlFillMemory, 3),
4260 IMPORT_SFUNC(RtlMoveMemory, 3),
4261 IMPORT_SFUNC(RtlCharToInteger, 3),
4262 IMPORT_SFUNC(RtlCopyMemory, 3),
4263 IMPORT_SFUNC(RtlCopyString, 2),
4264 IMPORT_SFUNC(RtlCompareMemory, 3),
4265 IMPORT_SFUNC(RtlEqualUnicodeString, 3),
4266 IMPORT_SFUNC(RtlCopyUnicodeString, 2),
4267 IMPORT_SFUNC(RtlUnicodeStringToAnsiString, 3),
4268 IMPORT_SFUNC(RtlAnsiStringToUnicodeString, 3),
4269 IMPORT_SFUNC(RtlInitAnsiString, 2),
4270 IMPORT_SFUNC_MAP(RtlInitString, RtlInitAnsiString, 2),
4271 IMPORT_SFUNC(RtlInitUnicodeString, 2),
4272 IMPORT_SFUNC(RtlFreeAnsiString, 1),
4273 IMPORT_SFUNC(RtlFreeUnicodeString, 1),
4274 IMPORT_SFUNC(RtlUnicodeStringToInteger, 3),
4275 IMPORT_CFUNC(sprintf, 0),
4276 IMPORT_CFUNC(vsprintf, 0),
4277 IMPORT_CFUNC_MAP(_snprintf, snprintf, 0),
4278 IMPORT_CFUNC_MAP(_vsnprintf, vsnprintf, 0),
4279 IMPORT_CFUNC(DbgPrint, 0),
4280 IMPORT_SFUNC(DbgBreakPoint, 0),
4281 IMPORT_SFUNC(KeBugCheckEx, 5),
4282 IMPORT_CFUNC(strncmp, 0),
4283 IMPORT_CFUNC(strcmp, 0),
4284 IMPORT_CFUNC_MAP(stricmp, strcasecmp, 0),
4285 IMPORT_CFUNC(strncpy, 0),
4286 IMPORT_CFUNC(strcpy, 0),
4287 IMPORT_CFUNC(strlen, 0),
4288 IMPORT_CFUNC_MAP(toupper, ntoskrnl_toupper, 0),
4289 IMPORT_CFUNC_MAP(tolower, ntoskrnl_tolower, 0),
4290 IMPORT_CFUNC_MAP(strstr, ntoskrnl_strstr, 0),
4291 IMPORT_CFUNC_MAP(strncat, ntoskrnl_strncat, 0),
4292 IMPORT_CFUNC_MAP(strchr, index, 0),
4293 IMPORT_CFUNC_MAP(strrchr, rindex, 0),
4294 IMPORT_CFUNC(memcpy, 0),
4295 IMPORT_CFUNC_MAP(memmove, ntoskrnl_memmove, 0),
4296 IMPORT_CFUNC_MAP(memset, ntoskrnl_memset, 0),
4297 IMPORT_CFUNC_MAP(memchr, ntoskrnl_memchr, 0),
4298 IMPORT_SFUNC(IoAllocateDriverObjectExtension, 4),
4299 IMPORT_SFUNC(IoGetDriverObjectExtension, 2),
4300 IMPORT_FFUNC(IofCallDriver, 2),
4301 IMPORT_FFUNC(IofCompleteRequest, 2),
4302 IMPORT_SFUNC(IoAcquireCancelSpinLock, 1),
4303 IMPORT_SFUNC(IoReleaseCancelSpinLock, 1),
4304 IMPORT_SFUNC(IoCancelIrp, 1),
4305 IMPORT_SFUNC(IoConnectInterrupt, 11),
4306 IMPORT_SFUNC(IoDisconnectInterrupt, 1),
4307 IMPORT_SFUNC(IoCreateDevice, 7),
4308 IMPORT_SFUNC(IoDeleteDevice, 1),
4309 IMPORT_SFUNC(IoGetAttachedDevice, 1),
4310 IMPORT_SFUNC(IoAttachDeviceToDeviceStack, 2),
4311 IMPORT_SFUNC(IoDetachDevice, 1),
4312 IMPORT_SFUNC(IoBuildSynchronousFsdRequest, 7),
4313 IMPORT_SFUNC(IoBuildAsynchronousFsdRequest, 6),
4314 IMPORT_SFUNC(IoBuildDeviceIoControlRequest, 9),
4315 IMPORT_SFUNC(IoAllocateIrp, 2),
4316 IMPORT_SFUNC(IoReuseIrp, 2),
4317 IMPORT_SFUNC(IoMakeAssociatedIrp, 2),
4318 IMPORT_SFUNC(IoFreeIrp, 1),
4319 IMPORT_SFUNC(IoInitializeIrp, 3),
4320 IMPORT_SFUNC(KeAcquireInterruptSpinLock, 1),
4321 IMPORT_SFUNC(KeReleaseInterruptSpinLock, 2),
4322 IMPORT_SFUNC(KeSynchronizeExecution, 3),
4323 IMPORT_SFUNC(KeWaitForSingleObject, 5),
4324 IMPORT_SFUNC(KeWaitForMultipleObjects, 8),
4325 IMPORT_SFUNC(_allmul, 4),
4326 IMPORT_SFUNC(_alldiv, 4),
4327 IMPORT_SFUNC(_allrem, 4),
4328 IMPORT_RFUNC(_allshr, 0),
4329 IMPORT_RFUNC(_allshl, 0),
4330 IMPORT_SFUNC(_aullmul, 4),
4331 IMPORT_SFUNC(_aulldiv, 4),
4332 IMPORT_SFUNC(_aullrem, 4),
4333 IMPORT_RFUNC(_aullshr, 0),
4334 IMPORT_RFUNC(_aullshl, 0),
4335 IMPORT_CFUNC(atoi, 0),
4336 IMPORT_CFUNC(atol, 0),
4337 IMPORT_CFUNC(rand, 0),
4338 IMPORT_CFUNC(srand, 0),
4339 IMPORT_SFUNC(WRITE_REGISTER_USHORT, 2),
4340 IMPORT_SFUNC(READ_REGISTER_USHORT, 1),
4341 IMPORT_SFUNC(WRITE_REGISTER_ULONG, 2),
4342 IMPORT_SFUNC(READ_REGISTER_ULONG, 1),
4343 IMPORT_SFUNC(READ_REGISTER_UCHAR, 1),
4344 IMPORT_SFUNC(WRITE_REGISTER_UCHAR, 2),
4345 IMPORT_SFUNC(ExInitializePagedLookasideList, 7),
4346 IMPORT_SFUNC(ExDeletePagedLookasideList, 1),
4347 IMPORT_SFUNC(ExInitializeNPagedLookasideList, 7),
4348 IMPORT_SFUNC(ExDeleteNPagedLookasideList, 1),
4349 IMPORT_FFUNC(InterlockedPopEntrySList, 1),
4350 IMPORT_FFUNC(InitializeSListHead, 1),
4351 IMPORT_FFUNC(InterlockedPushEntrySList, 2),
4352 IMPORT_SFUNC(ExQueryDepthSList, 1),
4353 IMPORT_FFUNC_MAP(ExpInterlockedPopEntrySList,
4354 InterlockedPopEntrySList, 1),
4355 IMPORT_FFUNC_MAP(ExpInterlockedPushEntrySList,
4356 InterlockedPushEntrySList, 2),
4357 IMPORT_FFUNC(ExInterlockedPopEntrySList, 2),
4358 IMPORT_FFUNC(ExInterlockedPushEntrySList, 3),
4359 IMPORT_SFUNC(ExAllocatePoolWithTag, 3),
4360 IMPORT_SFUNC(ExFreePoolWithTag, 2),
4361 IMPORT_SFUNC(ExFreePool, 1),
4362#ifdef __i386__
4363 IMPORT_FFUNC(KefAcquireSpinLockAtDpcLevel, 1),
4364 IMPORT_FFUNC(KefReleaseSpinLockFromDpcLevel,1),
4365 IMPORT_FFUNC(KeAcquireSpinLockRaiseToDpc, 1),
4366#else
4367 /*
4368 * For AMD64, we can get away with just mapping
4369 * KeAcquireSpinLockRaiseToDpc() directly to KfAcquireSpinLock()
4370 * because the calling conventions end up being the same.
4371 * On i386, we have to be careful because KfAcquireSpinLock()
4372 * is _fastcall but KeAcquireSpinLockRaiseToDpc() isn't.
4373 */
4374 IMPORT_SFUNC(KeAcquireSpinLockAtDpcLevel, 1),
4375 IMPORT_SFUNC(KeReleaseSpinLockFromDpcLevel, 1),
4376 IMPORT_SFUNC_MAP(KeAcquireSpinLockRaiseToDpc, KfAcquireSpinLock, 1),
4377#endif
4378 IMPORT_SFUNC_MAP(KeReleaseSpinLock, KfReleaseSpinLock, 1),
4379 IMPORT_FFUNC(InterlockedIncrement, 1),
4380 IMPORT_FFUNC(InterlockedDecrement, 1),
4381 IMPORT_FFUNC(InterlockedExchange, 2),
4382 IMPORT_FFUNC(ExInterlockedAddLargeStatistic, 2),
4383 IMPORT_SFUNC(IoAllocateMdl, 5),
4384 IMPORT_SFUNC(IoFreeMdl, 1),
4385 IMPORT_SFUNC(MmAllocateContiguousMemory, 2 + 1),
4386 IMPORT_SFUNC(MmAllocateContiguousMemorySpecifyCache, 5 + 3),
4387 IMPORT_SFUNC(MmFreeContiguousMemory, 1),
4388 IMPORT_SFUNC(MmFreeContiguousMemorySpecifyCache, 3),
4389 IMPORT_SFUNC(MmSizeOfMdl, 1),
4390 IMPORT_SFUNC(MmMapLockedPages, 2),
4391 IMPORT_SFUNC(MmMapLockedPagesSpecifyCache, 6),
4392 IMPORT_SFUNC(MmUnmapLockedPages, 2),
4393 IMPORT_SFUNC(MmBuildMdlForNonPagedPool, 1),
4394 IMPORT_SFUNC(MmGetPhysicalAddress, 1),
4395 IMPORT_SFUNC(MmGetSystemRoutineAddress, 1),
4396 IMPORT_SFUNC(MmIsAddressValid, 1),
4397 IMPORT_SFUNC(MmMapIoSpace, 3 + 1),
4398 IMPORT_SFUNC(MmUnmapIoSpace, 2),
4399 IMPORT_SFUNC(KeInitializeSpinLock, 1),
4400 IMPORT_SFUNC(IoIsWdmVersionAvailable, 2),
4401 IMPORT_SFUNC(IoOpenDeviceRegistryKey, 4),
4402 IMPORT_SFUNC(IoGetDeviceObjectPointer, 4),
4403 IMPORT_SFUNC(IoGetDeviceProperty, 5),
4404 IMPORT_SFUNC(IoAllocateWorkItem, 1),
4405 IMPORT_SFUNC(IoFreeWorkItem, 1),
4406 IMPORT_SFUNC(IoQueueWorkItem, 4),
4407 IMPORT_SFUNC(ExQueueWorkItem, 2),
4408 IMPORT_SFUNC(ntoskrnl_workitem, 2),
4409 IMPORT_SFUNC(KeInitializeMutex, 2),
4410 IMPORT_SFUNC(KeReleaseMutex, 2),
4411 IMPORT_SFUNC(KeReadStateMutex, 1),
4412 IMPORT_SFUNC(KeInitializeEvent, 3),
4413 IMPORT_SFUNC(KeSetEvent, 3),
4414 IMPORT_SFUNC(KeResetEvent, 1),
4415 IMPORT_SFUNC(KeClearEvent, 1),
4416 IMPORT_SFUNC(KeReadStateEvent, 1),
4417 IMPORT_SFUNC(KeInitializeTimer, 1),
4418 IMPORT_SFUNC(KeInitializeTimerEx, 2),
4419 IMPORT_SFUNC(KeSetTimer, 3),
4420 IMPORT_SFUNC(KeSetTimerEx, 4),
4421 IMPORT_SFUNC(KeCancelTimer, 1),
4422 IMPORT_SFUNC(KeReadStateTimer, 1),
4423 IMPORT_SFUNC(KeInitializeDpc, 3),
4424 IMPORT_SFUNC(KeInsertQueueDpc, 3),
4425 IMPORT_SFUNC(KeRemoveQueueDpc, 1),
4426 IMPORT_SFUNC(KeSetImportanceDpc, 2),
4427 IMPORT_SFUNC(KeSetTargetProcessorDpc, 2),
4428 IMPORT_SFUNC(KeFlushQueuedDpcs, 0),
4429 IMPORT_SFUNC(KeGetCurrentProcessorNumber, 1),
4430 IMPORT_SFUNC(ObReferenceObjectByHandle, 6),
4431 IMPORT_FFUNC(ObfDereferenceObject, 1),
4432 IMPORT_SFUNC(ZwClose, 1),
4433 IMPORT_SFUNC(PsCreateSystemThread, 7),
4434 IMPORT_SFUNC(PsTerminateSystemThread, 1),
4435 IMPORT_SFUNC(IoWMIRegistrationControl, 2),
4436 IMPORT_SFUNC(WmiQueryTraceInformation, 5),
4437 IMPORT_CFUNC(WmiTraceMessage, 0),
4438 IMPORT_SFUNC(KeQuerySystemTime, 1),
4439 IMPORT_CFUNC(KeTickCount, 0),
4440 IMPORT_SFUNC(KeDelayExecutionThread, 3),
4441 IMPORT_SFUNC(KeQueryInterruptTime, 0),
4442 IMPORT_SFUNC(KeGetCurrentThread, 0),
4443 IMPORT_SFUNC(KeSetPriorityThread, 2),
4444
4445 /*
4446 * This last entry is a catch-all for any function we haven't
4447 * implemented yet. The PE import list patching routine will
4448 * use it for any function that doesn't have an explicit match
4449 * in this table.
4450 */
4451
4452 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
4453
4454 /* End of list. */
4455
4456 { NULL, NULL, NULL }
4457};
| 4211}
4212
4213static struct thread *
4214KeGetCurrentThread(void)
4215{
4216
4217 return curthread;
4218}
4219
4220static int32_t
4221KeSetPriorityThread(td, pri)
4222 struct thread *td;
4223 int32_t pri;
4224{
4225 int32_t old;
4226
4227 if (td == NULL)
4228 return LOW_REALTIME_PRIORITY;
4229
4230 if (td->td_priority <= PRI_MIN_KERN)
4231 old = HIGH_PRIORITY;
4232 else if (td->td_priority >= PRI_MAX_KERN)
4233 old = LOW_PRIORITY;
4234 else
4235 old = LOW_REALTIME_PRIORITY;
4236
4237 thread_lock(td);
4238 if (pri == HIGH_PRIORITY)
4239 sched_prio(td, PRI_MIN_KERN);
4240 if (pri == LOW_REALTIME_PRIORITY)
4241 sched_prio(td, PRI_MIN_KERN + (PRI_MAX_KERN - PRI_MIN_KERN) / 2);
4242 if (pri == LOW_PRIORITY)
4243 sched_prio(td, PRI_MAX_KERN);
4244 thread_unlock(td);
4245
4246 return old;
4247}
4248
4249static void
4250dummy()
4251{
4252 printf("ntoskrnl dummy called...\n");
4253}
4254
4255
4256image_patch_table ntoskrnl_functbl[] = {
4257 IMPORT_SFUNC(RtlZeroMemory, 2),
4258 IMPORT_SFUNC(RtlSecureZeroMemory, 2),
4259 IMPORT_SFUNC(RtlFillMemory, 3),
4260 IMPORT_SFUNC(RtlMoveMemory, 3),
4261 IMPORT_SFUNC(RtlCharToInteger, 3),
4262 IMPORT_SFUNC(RtlCopyMemory, 3),
4263 IMPORT_SFUNC(RtlCopyString, 2),
4264 IMPORT_SFUNC(RtlCompareMemory, 3),
4265 IMPORT_SFUNC(RtlEqualUnicodeString, 3),
4266 IMPORT_SFUNC(RtlCopyUnicodeString, 2),
4267 IMPORT_SFUNC(RtlUnicodeStringToAnsiString, 3),
4268 IMPORT_SFUNC(RtlAnsiStringToUnicodeString, 3),
4269 IMPORT_SFUNC(RtlInitAnsiString, 2),
4270 IMPORT_SFUNC_MAP(RtlInitString, RtlInitAnsiString, 2),
4271 IMPORT_SFUNC(RtlInitUnicodeString, 2),
4272 IMPORT_SFUNC(RtlFreeAnsiString, 1),
4273 IMPORT_SFUNC(RtlFreeUnicodeString, 1),
4274 IMPORT_SFUNC(RtlUnicodeStringToInteger, 3),
4275 IMPORT_CFUNC(sprintf, 0),
4276 IMPORT_CFUNC(vsprintf, 0),
4277 IMPORT_CFUNC_MAP(_snprintf, snprintf, 0),
4278 IMPORT_CFUNC_MAP(_vsnprintf, vsnprintf, 0),
4279 IMPORT_CFUNC(DbgPrint, 0),
4280 IMPORT_SFUNC(DbgBreakPoint, 0),
4281 IMPORT_SFUNC(KeBugCheckEx, 5),
4282 IMPORT_CFUNC(strncmp, 0),
4283 IMPORT_CFUNC(strcmp, 0),
4284 IMPORT_CFUNC_MAP(stricmp, strcasecmp, 0),
4285 IMPORT_CFUNC(strncpy, 0),
4286 IMPORT_CFUNC(strcpy, 0),
4287 IMPORT_CFUNC(strlen, 0),
4288 IMPORT_CFUNC_MAP(toupper, ntoskrnl_toupper, 0),
4289 IMPORT_CFUNC_MAP(tolower, ntoskrnl_tolower, 0),
4290 IMPORT_CFUNC_MAP(strstr, ntoskrnl_strstr, 0),
4291 IMPORT_CFUNC_MAP(strncat, ntoskrnl_strncat, 0),
4292 IMPORT_CFUNC_MAP(strchr, index, 0),
4293 IMPORT_CFUNC_MAP(strrchr, rindex, 0),
4294 IMPORT_CFUNC(memcpy, 0),
4295 IMPORT_CFUNC_MAP(memmove, ntoskrnl_memmove, 0),
4296 IMPORT_CFUNC_MAP(memset, ntoskrnl_memset, 0),
4297 IMPORT_CFUNC_MAP(memchr, ntoskrnl_memchr, 0),
4298 IMPORT_SFUNC(IoAllocateDriverObjectExtension, 4),
4299 IMPORT_SFUNC(IoGetDriverObjectExtension, 2),
4300 IMPORT_FFUNC(IofCallDriver, 2),
4301 IMPORT_FFUNC(IofCompleteRequest, 2),
4302 IMPORT_SFUNC(IoAcquireCancelSpinLock, 1),
4303 IMPORT_SFUNC(IoReleaseCancelSpinLock, 1),
4304 IMPORT_SFUNC(IoCancelIrp, 1),
4305 IMPORT_SFUNC(IoConnectInterrupt, 11),
4306 IMPORT_SFUNC(IoDisconnectInterrupt, 1),
4307 IMPORT_SFUNC(IoCreateDevice, 7),
4308 IMPORT_SFUNC(IoDeleteDevice, 1),
4309 IMPORT_SFUNC(IoGetAttachedDevice, 1),
4310 IMPORT_SFUNC(IoAttachDeviceToDeviceStack, 2),
4311 IMPORT_SFUNC(IoDetachDevice, 1),
4312 IMPORT_SFUNC(IoBuildSynchronousFsdRequest, 7),
4313 IMPORT_SFUNC(IoBuildAsynchronousFsdRequest, 6),
4314 IMPORT_SFUNC(IoBuildDeviceIoControlRequest, 9),
4315 IMPORT_SFUNC(IoAllocateIrp, 2),
4316 IMPORT_SFUNC(IoReuseIrp, 2),
4317 IMPORT_SFUNC(IoMakeAssociatedIrp, 2),
4318 IMPORT_SFUNC(IoFreeIrp, 1),
4319 IMPORT_SFUNC(IoInitializeIrp, 3),
4320 IMPORT_SFUNC(KeAcquireInterruptSpinLock, 1),
4321 IMPORT_SFUNC(KeReleaseInterruptSpinLock, 2),
4322 IMPORT_SFUNC(KeSynchronizeExecution, 3),
4323 IMPORT_SFUNC(KeWaitForSingleObject, 5),
4324 IMPORT_SFUNC(KeWaitForMultipleObjects, 8),
4325 IMPORT_SFUNC(_allmul, 4),
4326 IMPORT_SFUNC(_alldiv, 4),
4327 IMPORT_SFUNC(_allrem, 4),
4328 IMPORT_RFUNC(_allshr, 0),
4329 IMPORT_RFUNC(_allshl, 0),
4330 IMPORT_SFUNC(_aullmul, 4),
4331 IMPORT_SFUNC(_aulldiv, 4),
4332 IMPORT_SFUNC(_aullrem, 4),
4333 IMPORT_RFUNC(_aullshr, 0),
4334 IMPORT_RFUNC(_aullshl, 0),
4335 IMPORT_CFUNC(atoi, 0),
4336 IMPORT_CFUNC(atol, 0),
4337 IMPORT_CFUNC(rand, 0),
4338 IMPORT_CFUNC(srand, 0),
4339 IMPORT_SFUNC(WRITE_REGISTER_USHORT, 2),
4340 IMPORT_SFUNC(READ_REGISTER_USHORT, 1),
4341 IMPORT_SFUNC(WRITE_REGISTER_ULONG, 2),
4342 IMPORT_SFUNC(READ_REGISTER_ULONG, 1),
4343 IMPORT_SFUNC(READ_REGISTER_UCHAR, 1),
4344 IMPORT_SFUNC(WRITE_REGISTER_UCHAR, 2),
4345 IMPORT_SFUNC(ExInitializePagedLookasideList, 7),
4346 IMPORT_SFUNC(ExDeletePagedLookasideList, 1),
4347 IMPORT_SFUNC(ExInitializeNPagedLookasideList, 7),
4348 IMPORT_SFUNC(ExDeleteNPagedLookasideList, 1),
4349 IMPORT_FFUNC(InterlockedPopEntrySList, 1),
4350 IMPORT_FFUNC(InitializeSListHead, 1),
4351 IMPORT_FFUNC(InterlockedPushEntrySList, 2),
4352 IMPORT_SFUNC(ExQueryDepthSList, 1),
4353 IMPORT_FFUNC_MAP(ExpInterlockedPopEntrySList,
4354 InterlockedPopEntrySList, 1),
4355 IMPORT_FFUNC_MAP(ExpInterlockedPushEntrySList,
4356 InterlockedPushEntrySList, 2),
4357 IMPORT_FFUNC(ExInterlockedPopEntrySList, 2),
4358 IMPORT_FFUNC(ExInterlockedPushEntrySList, 3),
4359 IMPORT_SFUNC(ExAllocatePoolWithTag, 3),
4360 IMPORT_SFUNC(ExFreePoolWithTag, 2),
4361 IMPORT_SFUNC(ExFreePool, 1),
4362#ifdef __i386__
4363 IMPORT_FFUNC(KefAcquireSpinLockAtDpcLevel, 1),
4364 IMPORT_FFUNC(KefReleaseSpinLockFromDpcLevel,1),
4365 IMPORT_FFUNC(KeAcquireSpinLockRaiseToDpc, 1),
4366#else
4367 /*
4368 * For AMD64, we can get away with just mapping
4369 * KeAcquireSpinLockRaiseToDpc() directly to KfAcquireSpinLock()
4370 * because the calling conventions end up being the same.
4371 * On i386, we have to be careful because KfAcquireSpinLock()
4372 * is _fastcall but KeAcquireSpinLockRaiseToDpc() isn't.
4373 */
4374 IMPORT_SFUNC(KeAcquireSpinLockAtDpcLevel, 1),
4375 IMPORT_SFUNC(KeReleaseSpinLockFromDpcLevel, 1),
4376 IMPORT_SFUNC_MAP(KeAcquireSpinLockRaiseToDpc, KfAcquireSpinLock, 1),
4377#endif
4378 IMPORT_SFUNC_MAP(KeReleaseSpinLock, KfReleaseSpinLock, 1),
4379 IMPORT_FFUNC(InterlockedIncrement, 1),
4380 IMPORT_FFUNC(InterlockedDecrement, 1),
4381 IMPORT_FFUNC(InterlockedExchange, 2),
4382 IMPORT_FFUNC(ExInterlockedAddLargeStatistic, 2),
4383 IMPORT_SFUNC(IoAllocateMdl, 5),
4384 IMPORT_SFUNC(IoFreeMdl, 1),
4385 IMPORT_SFUNC(MmAllocateContiguousMemory, 2 + 1),
4386 IMPORT_SFUNC(MmAllocateContiguousMemorySpecifyCache, 5 + 3),
4387 IMPORT_SFUNC(MmFreeContiguousMemory, 1),
4388 IMPORT_SFUNC(MmFreeContiguousMemorySpecifyCache, 3),
4389 IMPORT_SFUNC(MmSizeOfMdl, 1),
4390 IMPORT_SFUNC(MmMapLockedPages, 2),
4391 IMPORT_SFUNC(MmMapLockedPagesSpecifyCache, 6),
4392 IMPORT_SFUNC(MmUnmapLockedPages, 2),
4393 IMPORT_SFUNC(MmBuildMdlForNonPagedPool, 1),
4394 IMPORT_SFUNC(MmGetPhysicalAddress, 1),
4395 IMPORT_SFUNC(MmGetSystemRoutineAddress, 1),
4396 IMPORT_SFUNC(MmIsAddressValid, 1),
4397 IMPORT_SFUNC(MmMapIoSpace, 3 + 1),
4398 IMPORT_SFUNC(MmUnmapIoSpace, 2),
4399 IMPORT_SFUNC(KeInitializeSpinLock, 1),
4400 IMPORT_SFUNC(IoIsWdmVersionAvailable, 2),
4401 IMPORT_SFUNC(IoOpenDeviceRegistryKey, 4),
4402 IMPORT_SFUNC(IoGetDeviceObjectPointer, 4),
4403 IMPORT_SFUNC(IoGetDeviceProperty, 5),
4404 IMPORT_SFUNC(IoAllocateWorkItem, 1),
4405 IMPORT_SFUNC(IoFreeWorkItem, 1),
4406 IMPORT_SFUNC(IoQueueWorkItem, 4),
4407 IMPORT_SFUNC(ExQueueWorkItem, 2),
4408 IMPORT_SFUNC(ntoskrnl_workitem, 2),
4409 IMPORT_SFUNC(KeInitializeMutex, 2),
4410 IMPORT_SFUNC(KeReleaseMutex, 2),
4411 IMPORT_SFUNC(KeReadStateMutex, 1),
4412 IMPORT_SFUNC(KeInitializeEvent, 3),
4413 IMPORT_SFUNC(KeSetEvent, 3),
4414 IMPORT_SFUNC(KeResetEvent, 1),
4415 IMPORT_SFUNC(KeClearEvent, 1),
4416 IMPORT_SFUNC(KeReadStateEvent, 1),
4417 IMPORT_SFUNC(KeInitializeTimer, 1),
4418 IMPORT_SFUNC(KeInitializeTimerEx, 2),
4419 IMPORT_SFUNC(KeSetTimer, 3),
4420 IMPORT_SFUNC(KeSetTimerEx, 4),
4421 IMPORT_SFUNC(KeCancelTimer, 1),
4422 IMPORT_SFUNC(KeReadStateTimer, 1),
4423 IMPORT_SFUNC(KeInitializeDpc, 3),
4424 IMPORT_SFUNC(KeInsertQueueDpc, 3),
4425 IMPORT_SFUNC(KeRemoveQueueDpc, 1),
4426 IMPORT_SFUNC(KeSetImportanceDpc, 2),
4427 IMPORT_SFUNC(KeSetTargetProcessorDpc, 2),
4428 IMPORT_SFUNC(KeFlushQueuedDpcs, 0),
4429 IMPORT_SFUNC(KeGetCurrentProcessorNumber, 1),
4430 IMPORT_SFUNC(ObReferenceObjectByHandle, 6),
4431 IMPORT_FFUNC(ObfDereferenceObject, 1),
4432 IMPORT_SFUNC(ZwClose, 1),
4433 IMPORT_SFUNC(PsCreateSystemThread, 7),
4434 IMPORT_SFUNC(PsTerminateSystemThread, 1),
4435 IMPORT_SFUNC(IoWMIRegistrationControl, 2),
4436 IMPORT_SFUNC(WmiQueryTraceInformation, 5),
4437 IMPORT_CFUNC(WmiTraceMessage, 0),
4438 IMPORT_SFUNC(KeQuerySystemTime, 1),
4439 IMPORT_CFUNC(KeTickCount, 0),
4440 IMPORT_SFUNC(KeDelayExecutionThread, 3),
4441 IMPORT_SFUNC(KeQueryInterruptTime, 0),
4442 IMPORT_SFUNC(KeGetCurrentThread, 0),
4443 IMPORT_SFUNC(KeSetPriorityThread, 2),
4444
4445 /*
4446 * This last entry is a catch-all for any function we haven't
4447 * implemented yet. The PE import list patching routine will
4448 * use it for any function that doesn't have an explicit match
4449 * in this table.
4450 */
4451
4452 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
4453
4454 /* End of list. */
4455
4456 { NULL, NULL, NULL }
4457};
|