1/*-
2 * Copyright (c) 2003 Peter Wemm.
3 * Copyright (c) 1990 The Regents of the University of California.
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 4. Neither the name of the University nor the names of its contributors
15 *    may be used to endorse or promote products derived from this software
16 *    without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 *
30 *	from: @(#)sys_machdep.c	5.5 (Berkeley) 1/19/91
31 */
32
33#include <sys/cdefs.h>
34__FBSDID("$FreeBSD: stable/10/sys/amd64/amd64/sys_machdep.c 307940 2016-10-25 17:16:08Z glebius $");
35
36#include "opt_capsicum.h"
37
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/capsicum.h>
41#include <sys/kernel.h>
42#include <sys/lock.h>
43#include <sys/malloc.h>
44#include <sys/mutex.h>
45#include <sys/priv.h>
46#include <sys/proc.h>
47#include <sys/sysproto.h>
48#include <sys/uio.h>
49
50#include <vm/vm.h>
51#include <vm/pmap.h>
52#include <vm/vm_kern.h>		/* for kernel_map */
53#include <vm/vm_extern.h>
54
55#include <machine/frame.h>
56#include <machine/md_var.h>
57#include <machine/pcb.h>
58#include <machine/specialreg.h>
59#include <machine/sysarch.h>
60#include <machine/tss.h>
61#include <machine/vmparam.h>
62
63#include <security/audit/audit.h>
64
65#define	MAX_LD		8192
66
67int max_ldt_segment = 1024;
68SYSCTL_INT(_machdep, OID_AUTO, max_ldt_segment, CTLFLAG_RDTUN,
69    &max_ldt_segment, 0,
70    "Maximum number of allowed LDT segments in the single address space");
71
72static void
73max_ldt_segment_init(void *arg __unused)
74{
75
76	TUNABLE_INT_FETCH("machdep.max_ldt_segment", &max_ldt_segment);
77	if (max_ldt_segment <= 0)
78		max_ldt_segment = 1;
79	if (max_ldt_segment > MAX_LD)
80		max_ldt_segment = MAX_LD;
81}
82SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL);
83
84#ifdef notyet
85#ifdef SMP
86static void set_user_ldt_rv(struct vmspace *vmsp);
87#endif
88#endif
89static void user_ldt_derefl(struct proc_ldt *pldt);
90
91#ifndef _SYS_SYSPROTO_H_
92struct sysarch_args {
93	int op;
94	char *parms;
95};
96#endif
97
98int
99sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space)
100{
101	struct i386_ldt_args *largs, la;
102	struct user_segment_descriptor *lp;
103	int error = 0;
104
105	/*
106	 * XXXKIB check that the BSM generation code knows to encode
107	 * the op argument.
108	 */
109	AUDIT_ARG_CMD(uap->op);
110	if (uap_space == UIO_USERSPACE) {
111		error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args));
112		if (error != 0)
113			return (error);
114		largs = &la;
115	} else
116		largs = (struct i386_ldt_args *)uap->parms;
117
118	switch (uap->op) {
119	case I386_GET_LDT:
120		error = amd64_get_ldt(td, largs);
121		break;
122	case I386_SET_LDT:
123		if (largs->descs != NULL && largs->num > max_ldt_segment)
124			return (EINVAL);
125		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
126		if (largs->descs != NULL) {
127			lp = malloc(largs->num * sizeof(struct
128			    user_segment_descriptor), M_TEMP, M_WAITOK);
129			error = copyin(largs->descs, lp, largs->num *
130			    sizeof(struct user_segment_descriptor));
131			if (error == 0)
132				error = amd64_set_ldt(td, largs, lp);
133			free(lp, M_TEMP);
134		} else {
135			error = amd64_set_ldt(td, largs, NULL);
136		}
137		break;
138	}
139	return (error);
140}
141
142void
143update_gdt_gsbase(struct thread *td, uint32_t base)
144{
145	struct user_segment_descriptor *sd;
146
147	if (td != curthread)
148		return;
149	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
150	critical_enter();
151	sd = PCPU_GET(gs32p);
152	sd->sd_lobase = base & 0xffffff;
153	sd->sd_hibase = (base >> 24) & 0xff;
154	critical_exit();
155}
156
157void
158update_gdt_fsbase(struct thread *td, uint32_t base)
159{
160	struct user_segment_descriptor *sd;
161
162	if (td != curthread)
163		return;
164	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
165	critical_enter();
166	sd = PCPU_GET(fs32p);
167	sd->sd_lobase = base & 0xffffff;
168	sd->sd_hibase = (base >> 24) & 0xff;
169	critical_exit();
170}
171
172int
173sysarch(td, uap)
174	struct thread *td;
175	register struct sysarch_args *uap;
176{
177	int error = 0;
178	struct pcb *pcb = curthread->td_pcb;
179	uint32_t i386base;
180	uint64_t a64base;
181	struct i386_ioperm_args iargs;
182	struct i386_get_xfpustate i386xfpu;
183	struct amd64_get_xfpustate a64xfpu;
184
185#ifdef CAPABILITY_MODE
186	/*
187	 * When adding new operations, add a new case statement here to
188	 * explicitly indicate whether or not the operation is safe to
189	 * perform in capability mode.
190	 */
191	if (IN_CAPABILITY_MODE(td)) {
192		switch (uap->op) {
193		case I386_GET_LDT:
194		case I386_SET_LDT:
195		case I386_GET_IOPERM:
196		case I386_GET_FSBASE:
197		case I386_SET_FSBASE:
198		case I386_GET_GSBASE:
199		case I386_SET_GSBASE:
200		case I386_GET_XFPUSTATE:
201		case AMD64_GET_FSBASE:
202		case AMD64_SET_FSBASE:
203		case AMD64_GET_GSBASE:
204		case AMD64_SET_GSBASE:
205		case AMD64_GET_XFPUSTATE:
206			break;
207
208		case I386_SET_IOPERM:
209		default:
210#ifdef KTRACE
211			if (KTRPOINT(td, KTR_CAPFAIL))
212				ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
213#endif
214			return (ECAPMODE);
215		}
216	}
217#endif
218
219	if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
220		return (sysarch_ldt(td, uap, UIO_USERSPACE));
221	/*
222	 * XXXKIB check that the BSM generation code knows to encode
223	 * the op argument.
224	 */
225	AUDIT_ARG_CMD(uap->op);
226	switch (uap->op) {
227	case I386_GET_IOPERM:
228	case I386_SET_IOPERM:
229		if ((error = copyin(uap->parms, &iargs,
230		    sizeof(struct i386_ioperm_args))) != 0)
231			return (error);
232		break;
233	case I386_GET_XFPUSTATE:
234		if ((error = copyin(uap->parms, &i386xfpu,
235		    sizeof(struct i386_get_xfpustate))) != 0)
236			return (error);
237		a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
238		a64xfpu.len = i386xfpu.len;
239		break;
240	case AMD64_GET_XFPUSTATE:
241		if ((error = copyin(uap->parms, &a64xfpu,
242		    sizeof(struct amd64_get_xfpustate))) != 0)
243			return (error);
244		break;
245	default:
246		break;
247	}
248
249	switch (uap->op) {
250	case I386_GET_IOPERM:
251		error = amd64_get_ioperm(td, &iargs);
252		if (error == 0)
253			error = copyout(&iargs, uap->parms,
254			    sizeof(struct i386_ioperm_args));
255		break;
256	case I386_SET_IOPERM:
257		error = amd64_set_ioperm(td, &iargs);
258		break;
259	case I386_GET_FSBASE:
260		i386base = pcb->pcb_fsbase;
261		error = copyout(&i386base, uap->parms, sizeof(i386base));
262		break;
263	case I386_SET_FSBASE:
264		error = copyin(uap->parms, &i386base, sizeof(i386base));
265		if (!error) {
266			pcb->pcb_fsbase = i386base;
267			td->td_frame->tf_fs = _ufssel;
268			update_gdt_fsbase(td, i386base);
269		}
270		break;
271	case I386_GET_GSBASE:
272		i386base = pcb->pcb_gsbase;
273		error = copyout(&i386base, uap->parms, sizeof(i386base));
274		break;
275	case I386_SET_GSBASE:
276		error = copyin(uap->parms, &i386base, sizeof(i386base));
277		if (!error) {
278			pcb->pcb_gsbase = i386base;
279			td->td_frame->tf_gs = _ugssel;
280			update_gdt_gsbase(td, i386base);
281		}
282		break;
283	case AMD64_GET_FSBASE:
284		error = copyout(&pcb->pcb_fsbase, uap->parms, sizeof(pcb->pcb_fsbase));
285		break;
286
287	case AMD64_SET_FSBASE:
288		error = copyin(uap->parms, &a64base, sizeof(a64base));
289		if (!error) {
290			if (a64base < VM_MAXUSER_ADDRESS) {
291				pcb->pcb_fsbase = a64base;
292				set_pcb_flags(pcb, PCB_FULL_IRET);
293				td->td_frame->tf_fs = _ufssel;
294			} else
295				error = EINVAL;
296		}
297		break;
298
299	case AMD64_GET_GSBASE:
300		error = copyout(&pcb->pcb_gsbase, uap->parms, sizeof(pcb->pcb_gsbase));
301		break;
302
303	case AMD64_SET_GSBASE:
304		error = copyin(uap->parms, &a64base, sizeof(a64base));
305		if (!error) {
306			if (a64base < VM_MAXUSER_ADDRESS) {
307				pcb->pcb_gsbase = a64base;
308				set_pcb_flags(pcb, PCB_FULL_IRET);
309				td->td_frame->tf_gs = _ugssel;
310			} else
311				error = EINVAL;
312		}
313		break;
314
315	case I386_GET_XFPUSTATE:
316	case AMD64_GET_XFPUSTATE:
317		if (a64xfpu.len > cpu_max_ext_state_size -
318		    sizeof(struct savefpu))
319			return (EINVAL);
320		fpugetregs(td);
321		error = copyout((char *)(get_pcb_user_save_td(td) + 1),
322		    a64xfpu.addr, a64xfpu.len);
323		break;
324
325	default:
326		error = EINVAL;
327		break;
328	}
329	return (error);
330}
331
332int
333amd64_set_ioperm(td, uap)
334	struct thread *td;
335	struct i386_ioperm_args *uap;
336{
337	char *iomap;
338	struct amd64tss *tssp;
339	struct system_segment_descriptor *tss_sd;
340	u_long *addr;
341	struct pcb *pcb;
342	u_int i;
343	int error;
344
345	if ((error = priv_check(td, PRIV_IO)) != 0)
346		return (error);
347	if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
348		return (error);
349	if (uap->start > uap->start + uap->length ||
350	    uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
351		return (EINVAL);
352
353	/*
354	 * XXX
355	 * While this is restricted to root, we should probably figure out
356	 * whether any other driver is using this i/o address, as so not to
357	 * cause confusion.  This probably requires a global 'usage registry'.
358	 */
359	pcb = td->td_pcb;
360	if (pcb->pcb_tssp == NULL) {
361		tssp = (struct amd64tss *)kmem_malloc(kernel_arena,
362		    ctob(IOPAGES+1), M_WAITOK);
363		if (tssp == NULL)
364			return (ENOMEM);
365		iomap = (char *)&tssp[1];
366		addr = (u_long *)iomap;
367		for (i = 0; i < (ctob(IOPAGES) + 1) / sizeof(u_long); i++)
368			*addr++ = ~0;
369		critical_enter();
370		/* Takes care of tss_rsp0. */
371		memcpy(tssp, &common_tss[PCPU_GET(cpuid)],
372		    sizeof(struct amd64tss));
373		tssp->tss_iobase = sizeof(*tssp);
374		pcb->pcb_tssp = tssp;
375		tss_sd = PCPU_GET(tss);
376		tss_sd->sd_lobase = (u_long)tssp & 0xffffff;
377		tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful;
378		tss_sd->sd_type = SDT_SYSTSS;
379		ltr(GSEL(GPROC0_SEL, SEL_KPL));
380		PCPU_SET(tssp, tssp);
381		critical_exit();
382	} else
383		iomap = (char *)&pcb->pcb_tssp[1];
384	for (i = uap->start; i < uap->start + uap->length; i++) {
385		if (uap->enable)
386			iomap[i >> 3] &= ~(1 << (i & 7));
387		else
388			iomap[i >> 3] |= (1 << (i & 7));
389	}
390	return (error);
391}
392
393int
394amd64_get_ioperm(td, uap)
395	struct thread *td;
396	struct i386_ioperm_args *uap;
397{
398	int i, state;
399	char *iomap;
400
401	if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
402		return (EINVAL);
403	if (td->td_pcb->pcb_tssp == NULL) {
404		uap->length = 0;
405		goto done;
406	}
407
408	iomap = (char *)&td->td_pcb->pcb_tssp[1];
409
410	i = uap->start;
411	state = (iomap[i >> 3] >> (i & 7)) & 1;
412	uap->enable = !state;
413	uap->length = 1;
414
415	for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
416		if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
417			break;
418		uap->length++;
419	}
420
421done:
422	return (0);
423}
424
425/*
426 * Update the GDT entry pointing to the LDT to point to the LDT of the
427 * current process.
428 */
429void
430set_user_ldt(struct mdproc *mdp)
431{
432
433	critical_enter();
434	*PCPU_GET(ldt) = mdp->md_ldt_sd;
435	lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
436	critical_exit();
437}
438
439#ifdef notyet
440#ifdef SMP
441static void
442set_user_ldt_rv(struct vmspace *vmsp)
443{
444	struct thread *td;
445
446	td = curthread;
447	if (vmsp != td->td_proc->p_vmspace)
448		return;
449
450	set_user_ldt(&td->td_proc->p_md);
451}
452#endif
453#endif
454
455struct proc_ldt *
456user_ldt_alloc(struct proc *p, int force)
457{
458	struct proc_ldt *pldt, *new_ldt;
459	struct mdproc *mdp;
460	struct soft_segment_descriptor sldt;
461
462	mtx_assert(&dt_lock, MA_OWNED);
463	mdp = &p->p_md;
464	if (!force && mdp->md_ldt != NULL)
465		return (mdp->md_ldt);
466	mtx_unlock(&dt_lock);
467	new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
468	new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
469	     max_ldt_segment * sizeof(struct user_segment_descriptor),
470	     M_WAITOK | M_ZERO);
471	if (new_ldt->ldt_base == NULL) {
472		FREE(new_ldt, M_SUBPROC);
473		mtx_lock(&dt_lock);
474		return (NULL);
475	}
476	new_ldt->ldt_refcnt = 1;
477	sldt.ssd_base = (uint64_t)new_ldt->ldt_base;
478	sldt.ssd_limit = max_ldt_segment *
479	    sizeof(struct user_segment_descriptor) - 1;
480	sldt.ssd_type = SDT_SYSLDT;
481	sldt.ssd_dpl = SEL_KPL;
482	sldt.ssd_p = 1;
483	sldt.ssd_long = 0;
484	sldt.ssd_def32 = 0;
485	sldt.ssd_gran = 0;
486	mtx_lock(&dt_lock);
487	pldt = mdp->md_ldt;
488	if (pldt != NULL && !force) {
489		kmem_free(kernel_arena, (vm_offset_t)new_ldt->ldt_base,
490		    max_ldt_segment * sizeof(struct user_segment_descriptor));
491		free(new_ldt, M_SUBPROC);
492		return (pldt);
493	}
494
495	if (pldt != NULL) {
496		bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
497		    sizeof(struct user_segment_descriptor));
498		user_ldt_derefl(pldt);
499	}
500	ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
501	atomic_store_rel_ptr((volatile uintptr_t *)&mdp->md_ldt,
502	    (uintptr_t)new_ldt);
503	if (p == curproc)
504		set_user_ldt(mdp);
505
506	return (mdp->md_ldt);
507}
508
509void
510user_ldt_free(struct thread *td)
511{
512	struct proc *p = td->td_proc;
513	struct mdproc *mdp = &p->p_md;
514	struct proc_ldt *pldt;
515
516	mtx_assert(&dt_lock, MA_OWNED);
517	if ((pldt = mdp->md_ldt) == NULL) {
518		mtx_unlock(&dt_lock);
519		return;
520	}
521
522	mdp->md_ldt = NULL;
523	bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
524	if (td == curthread)
525		lldt(GSEL(GNULL_SEL, SEL_KPL));
526	user_ldt_deref(pldt);
527}
528
529static void
530user_ldt_derefl(struct proc_ldt *pldt)
531{
532
533	if (--pldt->ldt_refcnt == 0) {
534		kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
535		    max_ldt_segment * sizeof(struct user_segment_descriptor));
536		free(pldt, M_SUBPROC);
537	}
538}
539
540void
541user_ldt_deref(struct proc_ldt *pldt)
542{
543
544	mtx_assert(&dt_lock, MA_OWNED);
545	user_ldt_derefl(pldt);
546	mtx_unlock(&dt_lock);
547}
548
549/*
550 * Note for the authors of compat layers (linux, etc): copyout() in
551 * the function below is not a problem since it presents data in
552 * arch-specific format (i.e. i386-specific in this case), not in
553 * the OS-specific one.
554 */
555int
556amd64_get_ldt(td, uap)
557	struct thread *td;
558	struct i386_ldt_args *uap;
559{
560	int error = 0;
561	struct proc_ldt *pldt;
562	int num;
563	struct user_segment_descriptor *lp;
564
565#ifdef	DEBUG
566	printf("amd64_get_ldt: start=%d num=%d descs=%p\n",
567	    uap->start, uap->num, (void *)uap->descs);
568#endif
569
570	if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
571		lp = &((struct user_segment_descriptor *)(pldt->ldt_base))
572		    [uap->start];
573		num = min(uap->num, max_ldt_segment);
574	} else
575		return (EINVAL);
576
577	if ((uap->start > (unsigned int)max_ldt_segment) ||
578	    ((unsigned int)num > (unsigned int)max_ldt_segment) ||
579	    ((unsigned int)(uap->start + num) > (unsigned int)max_ldt_segment))
580		return(EINVAL);
581
582	error = copyout(lp, uap->descs, num *
583	    sizeof(struct user_segment_descriptor));
584	if (!error)
585		td->td_retval[0] = num;
586
587	return(error);
588}
589
590int
591amd64_set_ldt(td, uap, descs)
592	struct thread *td;
593	struct i386_ldt_args *uap;
594	struct user_segment_descriptor *descs;
595{
596	int error = 0;
597	unsigned int largest_ld, i;
598	struct mdproc *mdp = &td->td_proc->p_md;
599	struct proc_ldt *pldt;
600	struct user_segment_descriptor *dp;
601	struct proc *p;
602
603#ifdef	DEBUG
604	printf("amd64_set_ldt: start=%d num=%d descs=%p\n",
605	    uap->start, uap->num, (void *)uap->descs);
606#endif
607
608	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
609	p = td->td_proc;
610	if (descs == NULL) {
611		/* Free descriptors */
612		if (uap->start == 0 && uap->num == 0)
613			uap->num = max_ldt_segment;
614		if (uap->num == 0)
615			return (EINVAL);
616		if ((pldt = mdp->md_ldt) == NULL ||
617		    uap->start >= max_ldt_segment)
618			return (0);
619		largest_ld = uap->start + uap->num;
620		if (largest_ld > max_ldt_segment)
621			largest_ld = max_ldt_segment;
622		if (largest_ld < uap->start)
623			return (EINVAL);
624		i = largest_ld - uap->start;
625		mtx_lock(&dt_lock);
626		bzero(&((struct user_segment_descriptor *)(pldt->ldt_base))
627		    [uap->start], sizeof(struct user_segment_descriptor) * i);
628		mtx_unlock(&dt_lock);
629		return (0);
630	}
631
632	if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
633		/* verify range of descriptors to modify */
634		largest_ld = uap->start + uap->num;
635		if (uap->start >= max_ldt_segment ||
636		    largest_ld > max_ldt_segment ||
637		    largest_ld < uap->start)
638			return (EINVAL);
639	}
640
641	/* Check descriptors for access violations */
642	for (i = 0; i < uap->num; i++) {
643		dp = &descs[i];
644
645		switch (dp->sd_type) {
646		case SDT_SYSNULL:	/* system null */
647			dp->sd_p = 0;
648			break;
649		case SDT_SYS286TSS:
650		case SDT_SYSLDT:
651		case SDT_SYS286BSY:
652		case SDT_SYS286CGT:
653		case SDT_SYSTASKGT:
654		case SDT_SYS286IGT:
655		case SDT_SYS286TGT:
656		case SDT_SYSNULL2:
657		case SDT_SYSTSS:
658		case SDT_SYSNULL3:
659		case SDT_SYSBSY:
660		case SDT_SYSCGT:
661		case SDT_SYSNULL4:
662		case SDT_SYSIGT:
663		case SDT_SYSTGT:
664			/* I can't think of any reason to allow a user proc
665			 * to create a segment of these types.  They are
666			 * for OS use only.
667			 */
668			return (EACCES);
669			/*NOTREACHED*/
670
671		/* memory segment types */
672		case SDT_MEMEC:   /* memory execute only conforming */
673		case SDT_MEMEAC:  /* memory execute only accessed conforming */
674		case SDT_MEMERC:  /* memory execute read conforming */
675		case SDT_MEMERAC: /* memory execute read accessed conforming */
676			 /* Must be "present" if executable and conforming. */
677			if (dp->sd_p == 0)
678				return (EACCES);
679			break;
680		case SDT_MEMRO:   /* memory read only */
681		case SDT_MEMROA:  /* memory read only accessed */
682		case SDT_MEMRW:   /* memory read write */
683		case SDT_MEMRWA:  /* memory read write accessed */
684		case SDT_MEMROD:  /* memory read only expand dwn limit */
685		case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
686		case SDT_MEMRWD:  /* memory read write expand dwn limit */
687		case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
688		case SDT_MEME:    /* memory execute only */
689		case SDT_MEMEA:   /* memory execute only accessed */
690		case SDT_MEMER:   /* memory execute read */
691		case SDT_MEMERA:  /* memory execute read accessed */
692			break;
693		default:
694			return(EINVAL);
695			/*NOTREACHED*/
696		}
697
698		/* Only user (ring-3) descriptors may be present. */
699		if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
700			return (EACCES);
701	}
702
703	if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
704		/* Allocate a free slot */
705		mtx_lock(&dt_lock);
706		pldt = user_ldt_alloc(p, 0);
707		if (pldt == NULL) {
708			mtx_unlock(&dt_lock);
709			return (ENOMEM);
710		}
711
712		/*
713		 * start scanning a bit up to leave room for NVidia and
714		 * Wine, which still user the "Blat" method of allocation.
715		 */
716		i = 16;
717		dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
718		for (; i < max_ldt_segment; ++i, ++dp) {
719			if (dp->sd_type == SDT_SYSNULL)
720				break;
721		}
722		if (i >= max_ldt_segment) {
723			mtx_unlock(&dt_lock);
724			return (ENOSPC);
725		}
726		uap->start = i;
727		error = amd64_set_ldt_data(td, i, 1, descs);
728		mtx_unlock(&dt_lock);
729	} else {
730		largest_ld = uap->start + uap->num;
731		if (largest_ld > max_ldt_segment)
732			return (EINVAL);
733		mtx_lock(&dt_lock);
734		if (user_ldt_alloc(p, 0) != NULL) {
735			error = amd64_set_ldt_data(td, uap->start, uap->num,
736			    descs);
737		}
738		mtx_unlock(&dt_lock);
739	}
740	if (error == 0)
741		td->td_retval[0] = uap->start;
742	return (error);
743}
744
745int
746amd64_set_ldt_data(struct thread *td, int start, int num,
747    struct user_segment_descriptor *descs)
748{
749	struct mdproc *mdp = &td->td_proc->p_md;
750	struct proc_ldt *pldt = mdp->md_ldt;
751
752	mtx_assert(&dt_lock, MA_OWNED);
753
754	/* Fill in range */
755	bcopy(descs,
756	    &((struct user_segment_descriptor *)(pldt->ldt_base))[start],
757	    num * sizeof(struct user_segment_descriptor));
758	return (0);
759}
760