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 * 3. 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: releng/11.0/sys/amd64/amd64/sys_machdep.c 307935 2016-10-25 17:11:20Z 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	if (max_ldt_segment <= 0)
77		max_ldt_segment = 1;
78	if (max_ldt_segment > MAX_LD)
79		max_ldt_segment = MAX_LD;
80}
81SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL);
82
83#ifdef notyet
84#ifdef SMP
85static void set_user_ldt_rv(struct vmspace *vmsp);
86#endif
87#endif
88static void user_ldt_derefl(struct proc_ldt *pldt);
89
90#ifndef _SYS_SYSPROTO_H_
91struct sysarch_args {
92	int op;
93	char *parms;
94};
95#endif
96
97int
98sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space)
99{
100	struct i386_ldt_args *largs, la;
101	struct user_segment_descriptor *lp;
102	int error = 0;
103
104	/*
105	 * XXXKIB check that the BSM generation code knows to encode
106	 * the op argument.
107	 */
108	AUDIT_ARG_CMD(uap->op);
109	if (uap_space == UIO_USERSPACE) {
110		error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args));
111		if (error != 0)
112			return (error);
113		largs = &la;
114	} else
115		largs = (struct i386_ldt_args *)uap->parms;
116
117	switch (uap->op) {
118	case I386_GET_LDT:
119		error = amd64_get_ldt(td, largs);
120		break;
121	case I386_SET_LDT:
122		if (largs->descs != NULL && largs->num > max_ldt_segment)
123			return (EINVAL);
124		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
125		if (largs->descs != NULL) {
126			lp = malloc(largs->num * sizeof(struct
127			    user_segment_descriptor), M_TEMP, M_WAITOK);
128			error = copyin(largs->descs, lp, largs->num *
129			    sizeof(struct user_segment_descriptor));
130			if (error == 0)
131				error = amd64_set_ldt(td, largs, lp);
132			free(lp, M_TEMP);
133		} else {
134			error = amd64_set_ldt(td, largs, NULL);
135		}
136		break;
137	}
138	return (error);
139}
140
141void
142update_gdt_gsbase(struct thread *td, uint32_t base)
143{
144	struct user_segment_descriptor *sd;
145
146	if (td != curthread)
147		return;
148	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
149	critical_enter();
150	sd = PCPU_GET(gs32p);
151	sd->sd_lobase = base & 0xffffff;
152	sd->sd_hibase = (base >> 24) & 0xff;
153	critical_exit();
154}
155
156void
157update_gdt_fsbase(struct thread *td, uint32_t base)
158{
159	struct user_segment_descriptor *sd;
160
161	if (td != curthread)
162		return;
163	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
164	critical_enter();
165	sd = PCPU_GET(fs32p);
166	sd->sd_lobase = base & 0xffffff;
167	sd->sd_hibase = (base >> 24) & 0xff;
168	critical_exit();
169}
170
171int
172sysarch(td, uap)
173	struct thread *td;
174	register struct sysarch_args *uap;
175{
176	int error = 0;
177	struct pcb *pcb = curthread->td_pcb;
178	uint32_t i386base;
179	uint64_t a64base;
180	struct i386_ioperm_args iargs;
181	struct i386_get_xfpustate i386xfpu;
182	struct amd64_get_xfpustate a64xfpu;
183
184#ifdef CAPABILITY_MODE
185	/*
186	 * When adding new operations, add a new case statement here to
187	 * explicitly indicate whether or not the operation is safe to
188	 * perform in capability mode.
189	 */
190	if (IN_CAPABILITY_MODE(td)) {
191		switch (uap->op) {
192		case I386_GET_LDT:
193		case I386_SET_LDT:
194		case I386_GET_IOPERM:
195		case I386_GET_FSBASE:
196		case I386_SET_FSBASE:
197		case I386_GET_GSBASE:
198		case I386_SET_GSBASE:
199		case I386_GET_XFPUSTATE:
200		case AMD64_GET_FSBASE:
201		case AMD64_SET_FSBASE:
202		case AMD64_GET_GSBASE:
203		case AMD64_SET_GSBASE:
204		case AMD64_GET_XFPUSTATE:
205			break;
206
207		case I386_SET_IOPERM:
208		default:
209#ifdef KTRACE
210			if (KTRPOINT(td, KTR_CAPFAIL))
211				ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
212#endif
213			return (ECAPMODE);
214		}
215	}
216#endif
217
218	if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
219		return (sysarch_ldt(td, uap, UIO_USERSPACE));
220	/*
221	 * XXXKIB check that the BSM generation code knows to encode
222	 * the op argument.
223	 */
224	AUDIT_ARG_CMD(uap->op);
225	switch (uap->op) {
226	case I386_GET_IOPERM:
227	case I386_SET_IOPERM:
228		if ((error = copyin(uap->parms, &iargs,
229		    sizeof(struct i386_ioperm_args))) != 0)
230			return (error);
231		break;
232	case I386_GET_XFPUSTATE:
233		if ((error = copyin(uap->parms, &i386xfpu,
234		    sizeof(struct i386_get_xfpustate))) != 0)
235			return (error);
236		a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
237		a64xfpu.len = i386xfpu.len;
238		break;
239	case AMD64_GET_XFPUSTATE:
240		if ((error = copyin(uap->parms, &a64xfpu,
241		    sizeof(struct amd64_get_xfpustate))) != 0)
242			return (error);
243		break;
244	default:
245		break;
246	}
247
248	switch (uap->op) {
249	case I386_GET_IOPERM:
250		error = amd64_get_ioperm(td, &iargs);
251		if (error == 0)
252			error = copyout(&iargs, uap->parms,
253			    sizeof(struct i386_ioperm_args));
254		break;
255	case I386_SET_IOPERM:
256		error = amd64_set_ioperm(td, &iargs);
257		break;
258	case I386_GET_FSBASE:
259		i386base = pcb->pcb_fsbase;
260		error = copyout(&i386base, uap->parms, sizeof(i386base));
261		break;
262	case I386_SET_FSBASE:
263		error = copyin(uap->parms, &i386base, sizeof(i386base));
264		if (!error) {
265			pcb->pcb_fsbase = i386base;
266			td->td_frame->tf_fs = _ufssel;
267			update_gdt_fsbase(td, i386base);
268		}
269		break;
270	case I386_GET_GSBASE:
271		i386base = pcb->pcb_gsbase;
272		error = copyout(&i386base, uap->parms, sizeof(i386base));
273		break;
274	case I386_SET_GSBASE:
275		error = copyin(uap->parms, &i386base, sizeof(i386base));
276		if (!error) {
277			pcb->pcb_gsbase = i386base;
278			td->td_frame->tf_gs = _ugssel;
279			update_gdt_gsbase(td, i386base);
280		}
281		break;
282	case AMD64_GET_FSBASE:
283		error = copyout(&pcb->pcb_fsbase, uap->parms, sizeof(pcb->pcb_fsbase));
284		break;
285
286	case AMD64_SET_FSBASE:
287		error = copyin(uap->parms, &a64base, sizeof(a64base));
288		if (!error) {
289			if (a64base < VM_MAXUSER_ADDRESS) {
290				pcb->pcb_fsbase = a64base;
291				set_pcb_flags(pcb, PCB_FULL_IRET);
292				td->td_frame->tf_fs = _ufssel;
293			} else
294				error = EINVAL;
295		}
296		break;
297
298	case AMD64_GET_GSBASE:
299		error = copyout(&pcb->pcb_gsbase, uap->parms, sizeof(pcb->pcb_gsbase));
300		break;
301
302	case AMD64_SET_GSBASE:
303		error = copyin(uap->parms, &a64base, sizeof(a64base));
304		if (!error) {
305			if (a64base < VM_MAXUSER_ADDRESS) {
306				pcb->pcb_gsbase = a64base;
307				set_pcb_flags(pcb, PCB_FULL_IRET);
308				td->td_frame->tf_gs = _ugssel;
309			} else
310				error = EINVAL;
311		}
312		break;
313
314	case I386_GET_XFPUSTATE:
315	case AMD64_GET_XFPUSTATE:
316		if (a64xfpu.len > cpu_max_ext_state_size -
317		    sizeof(struct savefpu))
318			return (EINVAL);
319		fpugetregs(td);
320		error = copyout((char *)(get_pcb_user_save_td(td) + 1),
321		    a64xfpu.addr, a64xfpu.len);
322		break;
323
324	default:
325		error = EINVAL;
326		break;
327	}
328	return (error);
329}
330
331int
332amd64_set_ioperm(td, uap)
333	struct thread *td;
334	struct i386_ioperm_args *uap;
335{
336	char *iomap;
337	struct amd64tss *tssp;
338	struct system_segment_descriptor *tss_sd;
339	struct pcb *pcb;
340	u_int i;
341	int error;
342
343	if ((error = priv_check(td, PRIV_IO)) != 0)
344		return (error);
345	if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
346		return (error);
347	if (uap->start > uap->start + uap->length ||
348	    uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
349		return (EINVAL);
350
351	/*
352	 * XXX
353	 * While this is restricted to root, we should probably figure out
354	 * whether any other driver is using this i/o address, as so not to
355	 * cause confusion.  This probably requires a global 'usage registry'.
356	 */
357	pcb = td->td_pcb;
358	if (pcb->pcb_tssp == NULL) {
359		tssp = (struct amd64tss *)kmem_malloc(kernel_arena,
360		    ctob(IOPAGES+1), M_WAITOK);
361		iomap = (char *)&tssp[1];
362		memset(iomap, 0xff, IOPERM_BITMAP_SIZE);
363		critical_enter();
364		/* Takes care of tss_rsp0. */
365		memcpy(tssp, &common_tss[PCPU_GET(cpuid)],
366		    sizeof(struct amd64tss));
367		tssp->tss_iobase = sizeof(*tssp);
368		pcb->pcb_tssp = tssp;
369		tss_sd = PCPU_GET(tss);
370		tss_sd->sd_lobase = (u_long)tssp & 0xffffff;
371		tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful;
372		tss_sd->sd_type = SDT_SYSTSS;
373		ltr(GSEL(GPROC0_SEL, SEL_KPL));
374		PCPU_SET(tssp, tssp);
375		critical_exit();
376	} else
377		iomap = (char *)&pcb->pcb_tssp[1];
378	for (i = uap->start; i < uap->start + uap->length; i++) {
379		if (uap->enable)
380			iomap[i >> 3] &= ~(1 << (i & 7));
381		else
382			iomap[i >> 3] |= (1 << (i & 7));
383	}
384	return (error);
385}
386
387int
388amd64_get_ioperm(td, uap)
389	struct thread *td;
390	struct i386_ioperm_args *uap;
391{
392	int i, state;
393	char *iomap;
394
395	if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
396		return (EINVAL);
397	if (td->td_pcb->pcb_tssp == NULL) {
398		uap->length = 0;
399		goto done;
400	}
401
402	iomap = (char *)&td->td_pcb->pcb_tssp[1];
403
404	i = uap->start;
405	state = (iomap[i >> 3] >> (i & 7)) & 1;
406	uap->enable = !state;
407	uap->length = 1;
408
409	for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
410		if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
411			break;
412		uap->length++;
413	}
414
415done:
416	return (0);
417}
418
419/*
420 * Update the GDT entry pointing to the LDT to point to the LDT of the
421 * current process.
422 */
423void
424set_user_ldt(struct mdproc *mdp)
425{
426
427	critical_enter();
428	*PCPU_GET(ldt) = mdp->md_ldt_sd;
429	lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
430	critical_exit();
431}
432
433#ifdef notyet
434#ifdef SMP
435static void
436set_user_ldt_rv(struct vmspace *vmsp)
437{
438	struct thread *td;
439
440	td = curthread;
441	if (vmsp != td->td_proc->p_vmspace)
442		return;
443
444	set_user_ldt(&td->td_proc->p_md);
445}
446#endif
447#endif
448
449struct proc_ldt *
450user_ldt_alloc(struct proc *p, int force)
451{
452	struct proc_ldt *pldt, *new_ldt;
453	struct mdproc *mdp;
454	struct soft_segment_descriptor sldt;
455
456	mtx_assert(&dt_lock, MA_OWNED);
457	mdp = &p->p_md;
458	if (!force && mdp->md_ldt != NULL)
459		return (mdp->md_ldt);
460	mtx_unlock(&dt_lock);
461	new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
462	new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
463	     max_ldt_segment * sizeof(struct user_segment_descriptor),
464	     M_WAITOK | M_ZERO);
465	new_ldt->ldt_refcnt = 1;
466	sldt.ssd_base = (uint64_t)new_ldt->ldt_base;
467	sldt.ssd_limit = max_ldt_segment *
468	    sizeof(struct user_segment_descriptor) - 1;
469	sldt.ssd_type = SDT_SYSLDT;
470	sldt.ssd_dpl = SEL_KPL;
471	sldt.ssd_p = 1;
472	sldt.ssd_long = 0;
473	sldt.ssd_def32 = 0;
474	sldt.ssd_gran = 0;
475	mtx_lock(&dt_lock);
476	pldt = mdp->md_ldt;
477	if (pldt != NULL && !force) {
478		kmem_free(kernel_arena, (vm_offset_t)new_ldt->ldt_base,
479		    max_ldt_segment * sizeof(struct user_segment_descriptor));
480		free(new_ldt, M_SUBPROC);
481		return (pldt);
482	}
483
484	if (pldt != NULL) {
485		bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
486		    sizeof(struct user_segment_descriptor));
487		user_ldt_derefl(pldt);
488	}
489	ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
490	atomic_store_rel_ptr((volatile uintptr_t *)&mdp->md_ldt,
491	    (uintptr_t)new_ldt);
492	if (p == curproc)
493		set_user_ldt(mdp);
494
495	return (mdp->md_ldt);
496}
497
498void
499user_ldt_free(struct thread *td)
500{
501	struct proc *p = td->td_proc;
502	struct mdproc *mdp = &p->p_md;
503	struct proc_ldt *pldt;
504
505	mtx_assert(&dt_lock, MA_OWNED);
506	if ((pldt = mdp->md_ldt) == NULL) {
507		mtx_unlock(&dt_lock);
508		return;
509	}
510
511	mdp->md_ldt = NULL;
512	bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
513	if (td == curthread)
514		lldt(GSEL(GNULL_SEL, SEL_KPL));
515	user_ldt_deref(pldt);
516}
517
518static void
519user_ldt_derefl(struct proc_ldt *pldt)
520{
521
522	if (--pldt->ldt_refcnt == 0) {
523		kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
524		    max_ldt_segment * sizeof(struct user_segment_descriptor));
525		free(pldt, M_SUBPROC);
526	}
527}
528
529void
530user_ldt_deref(struct proc_ldt *pldt)
531{
532
533	mtx_assert(&dt_lock, MA_OWNED);
534	user_ldt_derefl(pldt);
535	mtx_unlock(&dt_lock);
536}
537
538/*
539 * Note for the authors of compat layers (linux, etc): copyout() in
540 * the function below is not a problem since it presents data in
541 * arch-specific format (i.e. i386-specific in this case), not in
542 * the OS-specific one.
543 */
544int
545amd64_get_ldt(td, uap)
546	struct thread *td;
547	struct i386_ldt_args *uap;
548{
549	int error = 0;
550	struct proc_ldt *pldt;
551	int num;
552	struct user_segment_descriptor *lp;
553
554#ifdef	DEBUG
555	printf("amd64_get_ldt: start=%d num=%d descs=%p\n",
556	    uap->start, uap->num, (void *)uap->descs);
557#endif
558
559	if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
560		lp = &((struct user_segment_descriptor *)(pldt->ldt_base))
561		    [uap->start];
562		num = min(uap->num, max_ldt_segment);
563	} else
564		return (EINVAL);
565
566	if ((uap->start > (unsigned int)max_ldt_segment) ||
567	    ((unsigned int)num > (unsigned int)max_ldt_segment) ||
568	    ((unsigned int)(uap->start + num) > (unsigned int)max_ldt_segment))
569		return(EINVAL);
570
571	error = copyout(lp, uap->descs, num *
572	    sizeof(struct user_segment_descriptor));
573	if (!error)
574		td->td_retval[0] = num;
575
576	return(error);
577}
578
579int
580amd64_set_ldt(td, uap, descs)
581	struct thread *td;
582	struct i386_ldt_args *uap;
583	struct user_segment_descriptor *descs;
584{
585	int error = 0;
586	unsigned int largest_ld, i;
587	struct mdproc *mdp = &td->td_proc->p_md;
588	struct proc_ldt *pldt;
589	struct user_segment_descriptor *dp;
590	struct proc *p;
591
592#ifdef	DEBUG
593	printf("amd64_set_ldt: start=%d num=%d descs=%p\n",
594	    uap->start, uap->num, (void *)uap->descs);
595#endif
596
597	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
598	p = td->td_proc;
599	if (descs == NULL) {
600		/* Free descriptors */
601		if (uap->start == 0 && uap->num == 0)
602			uap->num = max_ldt_segment;
603		if (uap->num == 0)
604			return (EINVAL);
605		if ((pldt = mdp->md_ldt) == NULL ||
606		    uap->start >= max_ldt_segment)
607			return (0);
608		largest_ld = uap->start + uap->num;
609		if (largest_ld > max_ldt_segment)
610			largest_ld = max_ldt_segment;
611		if (largest_ld < uap->start)
612			return (EINVAL);
613		i = largest_ld - uap->start;
614		mtx_lock(&dt_lock);
615		bzero(&((struct user_segment_descriptor *)(pldt->ldt_base))
616		    [uap->start], sizeof(struct user_segment_descriptor) * i);
617		mtx_unlock(&dt_lock);
618		return (0);
619	}
620
621	if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
622		/* verify range of descriptors to modify */
623		largest_ld = uap->start + uap->num;
624		if (uap->start >= max_ldt_segment ||
625		    largest_ld > max_ldt_segment ||
626		    largest_ld < uap->start)
627			return (EINVAL);
628	}
629
630	/* Check descriptors for access violations */
631	for (i = 0; i < uap->num; i++) {
632		dp = &descs[i];
633
634		switch (dp->sd_type) {
635		case SDT_SYSNULL:	/* system null */
636			dp->sd_p = 0;
637			break;
638		case SDT_SYS286TSS:
639		case SDT_SYSLDT:
640		case SDT_SYS286BSY:
641		case SDT_SYS286CGT:
642		case SDT_SYSTASKGT:
643		case SDT_SYS286IGT:
644		case SDT_SYS286TGT:
645		case SDT_SYSNULL2:
646		case SDT_SYSTSS:
647		case SDT_SYSNULL3:
648		case SDT_SYSBSY:
649		case SDT_SYSCGT:
650		case SDT_SYSNULL4:
651		case SDT_SYSIGT:
652		case SDT_SYSTGT:
653			/* I can't think of any reason to allow a user proc
654			 * to create a segment of these types.  They are
655			 * for OS use only.
656			 */
657			return (EACCES);
658			/*NOTREACHED*/
659
660		/* memory segment types */
661		case SDT_MEMEC:   /* memory execute only conforming */
662		case SDT_MEMEAC:  /* memory execute only accessed conforming */
663		case SDT_MEMERC:  /* memory execute read conforming */
664		case SDT_MEMERAC: /* memory execute read accessed conforming */
665			 /* Must be "present" if executable and conforming. */
666			if (dp->sd_p == 0)
667				return (EACCES);
668			break;
669		case SDT_MEMRO:   /* memory read only */
670		case SDT_MEMROA:  /* memory read only accessed */
671		case SDT_MEMRW:   /* memory read write */
672		case SDT_MEMRWA:  /* memory read write accessed */
673		case SDT_MEMROD:  /* memory read only expand dwn limit */
674		case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
675		case SDT_MEMRWD:  /* memory read write expand dwn limit */
676		case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
677		case SDT_MEME:    /* memory execute only */
678		case SDT_MEMEA:   /* memory execute only accessed */
679		case SDT_MEMER:   /* memory execute read */
680		case SDT_MEMERA:  /* memory execute read accessed */
681			break;
682		default:
683			return(EINVAL);
684			/*NOTREACHED*/
685		}
686
687		/* Only user (ring-3) descriptors may be present. */
688		if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
689			return (EACCES);
690	}
691
692	if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
693		/* Allocate a free slot */
694		mtx_lock(&dt_lock);
695		pldt = user_ldt_alloc(p, 0);
696		if (pldt == NULL) {
697			mtx_unlock(&dt_lock);
698			return (ENOMEM);
699		}
700
701		/*
702		 * start scanning a bit up to leave room for NVidia and
703		 * Wine, which still user the "Blat" method of allocation.
704		 */
705		i = 16;
706		dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
707		for (; i < max_ldt_segment; ++i, ++dp) {
708			if (dp->sd_type == SDT_SYSNULL)
709				break;
710		}
711		if (i >= max_ldt_segment) {
712			mtx_unlock(&dt_lock);
713			return (ENOSPC);
714		}
715		uap->start = i;
716		error = amd64_set_ldt_data(td, i, 1, descs);
717		mtx_unlock(&dt_lock);
718	} else {
719		largest_ld = uap->start + uap->num;
720		if (largest_ld > max_ldt_segment)
721			return (EINVAL);
722		mtx_lock(&dt_lock);
723		if (user_ldt_alloc(p, 0) != NULL) {
724			error = amd64_set_ldt_data(td, uap->start, uap->num,
725			    descs);
726		}
727		mtx_unlock(&dt_lock);
728	}
729	if (error == 0)
730		td->td_retval[0] = uap->start;
731	return (error);
732}
733
734int
735amd64_set_ldt_data(struct thread *td, int start, int num,
736    struct user_segment_descriptor *descs)
737{
738	struct mdproc *mdp = &td->td_proc->p_md;
739	struct proc_ldt *pldt = mdp->md_ldt;
740
741	mtx_assert(&dt_lock, MA_OWNED);
742
743	/* Fill in range */
744	bcopy(descs,
745	    &((struct user_segment_descriptor *)(pldt->ldt_base))[start],
746	    num * sizeof(struct user_segment_descriptor));
747	return (0);
748}
749