xref: /freebsd-14-stable/sys/amd64/amd64/sys_machdep.c (revision 73f2b9f05a6b613e9aa6829101d63ec4e5a2d200)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2003 Peter Wemm.
5  * Copyright (c) 1990 The Regents of the University of California.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from: @(#)sys_machdep.c	5.5 (Berkeley) 1/19/91
33  */
34 
35 #include <sys/cdefs.h>
36 #include "opt_capsicum.h"
37 #include "opt_ktrace.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/capsicum.h>
42 #include <sys/kernel.h>
43 #include <sys/ktrace.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mutex.h>
47 #include <sys/pcpu.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/smp.h>
51 #include <sys/sysent.h>
52 #include <sys/sysproto.h>
53 #include <sys/uio.h>
54 
55 #include <vm/vm.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_kern.h>		/* for kernel_map */
58 #include <vm/vm_map.h>
59 #include <vm/vm_extern.h>
60 
61 #include <machine/frame.h>
62 #include <machine/md_var.h>
63 #include <machine/pcb.h>
64 #include <machine/specialreg.h>
65 #include <machine/sysarch.h>
66 #include <machine/tss.h>
67 #include <machine/vmparam.h>
68 
69 #include <security/audit/audit.h>
70 
71 static void user_ldt_deref(struct proc_ldt *pldt);
72 static void user_ldt_derefl(struct proc_ldt *pldt);
73 
74 #define	MAX_LD		8192
75 
76 int max_ldt_segment = 512;
77 SYSCTL_INT(_machdep, OID_AUTO, max_ldt_segment, CTLFLAG_RDTUN,
78     &max_ldt_segment, 0,
79     "Maximum number of allowed LDT segments in the single address space");
80 
81 static void
max_ldt_segment_init(void * arg __unused)82 max_ldt_segment_init(void *arg __unused)
83 {
84 
85 	if (max_ldt_segment <= 0)
86 		max_ldt_segment = 1;
87 	if (max_ldt_segment > MAX_LD)
88 		max_ldt_segment = MAX_LD;
89 }
90 SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL);
91 
92 #ifndef _SYS_SYSPROTO_H_
93 struct sysarch_args {
94 	int op;
95 	char *parms;
96 };
97 #endif
98 
99 int
sysarch_ldt(struct thread * td,struct sysarch_args * uap,int uap_space)100 sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space)
101 {
102 	struct i386_ldt_args *largs, la;
103 	struct user_segment_descriptor *lp;
104 	int error = 0;
105 
106 	/*
107 	 * XXXKIB check that the BSM generation code knows to encode
108 	 * the op argument.
109 	 */
110 	AUDIT_ARG_CMD(uap->op);
111 	if (uap_space == UIO_USERSPACE) {
112 		error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args));
113 		if (error != 0)
114 			return (error);
115 		largs = &la;
116 	} else
117 		largs = (struct i386_ldt_args *)uap->parms;
118 
119 	switch (uap->op) {
120 	case I386_GET_LDT:
121 		error = amd64_get_ldt(td, largs);
122 		break;
123 	case I386_SET_LDT:
124 		if (largs->descs != NULL && largs->num > max_ldt_segment)
125 			return (EINVAL);
126 		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
127 		if (largs->descs != NULL) {
128 			lp = malloc(largs->num * sizeof(struct
129 			    user_segment_descriptor), M_TEMP, M_WAITOK);
130 			error = copyin(largs->descs, lp, largs->num *
131 			    sizeof(struct user_segment_descriptor));
132 			if (error == 0)
133 				error = amd64_set_ldt(td, largs, lp);
134 			free(lp, M_TEMP);
135 		} else {
136 			error = amd64_set_ldt(td, largs, NULL);
137 		}
138 		break;
139 	}
140 	return (error);
141 }
142 
143 void
update_gdt_gsbase(struct thread * td,uint32_t base)144 update_gdt_gsbase(struct thread *td, uint32_t base)
145 {
146 	struct user_segment_descriptor *sd;
147 
148 	if (td != curthread)
149 		return;
150 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
151 	critical_enter();
152 	sd = PCPU_GET(gs32p);
153 	sd->sd_lobase = base & 0xffffff;
154 	sd->sd_hibase = (base >> 24) & 0xff;
155 	critical_exit();
156 }
157 
158 void
update_gdt_fsbase(struct thread * td,uint32_t base)159 update_gdt_fsbase(struct thread *td, uint32_t base)
160 {
161 	struct user_segment_descriptor *sd;
162 
163 	if (td != curthread)
164 		return;
165 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
166 	critical_enter();
167 	sd = PCPU_GET(fs32p);
168 	sd->sd_lobase = base & 0xffffff;
169 	sd->sd_hibase = (base >> 24) & 0xff;
170 	critical_exit();
171 }
172 
173 int
sysarch(struct thread * td,struct sysarch_args * uap)174 sysarch(struct thread *td, struct sysarch_args *uap)
175 {
176 	struct pcb *pcb;
177 	struct vm_map *map;
178 	uint32_t i386base;
179 	uint64_t a64base;
180 	struct i386_ioperm_args iargs;
181 	struct i386_get_xfpustate i386xfpu;
182 	struct i386_set_pkru i386pkru;
183 	struct amd64_get_xfpustate a64xfpu;
184 	struct amd64_set_pkru a64pkru;
185 	int error;
186 
187 #ifdef CAPABILITY_MODE
188 	/*
189 	 * When adding new operations, add a new case statement here to
190 	 * explicitly indicate whether or not the operation is safe to
191 	 * perform in capability mode.
192 	 */
193 	switch (uap->op) {
194 	case I386_GET_LDT:
195 	case I386_SET_LDT:
196 	case I386_GET_IOPERM:
197 	case I386_GET_FSBASE:
198 	case I386_SET_FSBASE:
199 	case I386_GET_GSBASE:
200 	case I386_SET_GSBASE:
201 	case I386_GET_XFPUSTATE:
202 	case I386_SET_PKRU:
203 	case I386_CLEAR_PKRU:
204 	case AMD64_GET_FSBASE:
205 	case AMD64_SET_FSBASE:
206 	case AMD64_GET_GSBASE:
207 	case AMD64_SET_GSBASE:
208 	case AMD64_GET_XFPUSTATE:
209 	case AMD64_SET_PKRU:
210 	case AMD64_CLEAR_PKRU:
211 		break;
212 
213 	case I386_SET_IOPERM:
214 	default:
215 		if (CAP_TRACING(td))
216 			ktrcapfail(CAPFAIL_SYSCALL, &uap->op);
217 		if (IN_CAPABILITY_MODE(td))
218 			return (ECAPMODE);
219 		break;
220 	}
221 #endif
222 
223 	if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
224 		return (sysarch_ldt(td, uap, UIO_USERSPACE));
225 
226 	error = 0;
227 	pcb = td->td_pcb;
228 
229 	/*
230 	 * XXXKIB check that the BSM generation code knows to encode
231 	 * the op argument.
232 	 */
233 	AUDIT_ARG_CMD(uap->op);
234 	switch (uap->op) {
235 	case I386_GET_IOPERM:
236 	case I386_SET_IOPERM:
237 		if ((error = copyin(uap->parms, &iargs,
238 		    sizeof(struct i386_ioperm_args))) != 0)
239 			return (error);
240 		break;
241 	case I386_GET_XFPUSTATE:
242 		if ((error = copyin(uap->parms, &i386xfpu,
243 		    sizeof(struct i386_get_xfpustate))) != 0)
244 			return (error);
245 		a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
246 		a64xfpu.len = i386xfpu.len;
247 		break;
248 	case I386_SET_PKRU:
249 	case I386_CLEAR_PKRU:
250 		if ((error = copyin(uap->parms, &i386pkru,
251 		    sizeof(struct i386_set_pkru))) != 0)
252 			return (error);
253 		a64pkru.addr = (void *)(uintptr_t)i386pkru.addr;
254 		a64pkru.len = i386pkru.len;
255 		a64pkru.keyidx = i386pkru.keyidx;
256 		a64pkru.flags = i386pkru.flags;
257 		break;
258 	case AMD64_GET_XFPUSTATE:
259 		if ((error = copyin(uap->parms, &a64xfpu,
260 		    sizeof(struct amd64_get_xfpustate))) != 0)
261 			return (error);
262 		break;
263 	case AMD64_SET_PKRU:
264 	case AMD64_CLEAR_PKRU:
265 		if ((error = copyin(uap->parms, &a64pkru,
266 		    sizeof(struct amd64_set_pkru))) != 0)
267 			return (error);
268 		break;
269 	default:
270 		break;
271 	}
272 
273 	switch (uap->op) {
274 	case I386_GET_IOPERM:
275 		error = amd64_get_ioperm(td, &iargs);
276 		if (error == 0)
277 			error = copyout(&iargs, uap->parms,
278 			    sizeof(struct i386_ioperm_args));
279 		break;
280 	case I386_SET_IOPERM:
281 		error = amd64_set_ioperm(td, &iargs);
282 		break;
283 	case I386_GET_FSBASE:
284 		update_pcb_bases(pcb);
285 		i386base = pcb->pcb_fsbase;
286 		error = copyout(&i386base, uap->parms, sizeof(i386base));
287 		break;
288 	case I386_SET_FSBASE:
289 		error = copyin(uap->parms, &i386base, sizeof(i386base));
290 		if (error == 0) {
291 			set_pcb_flags(pcb, PCB_FULL_IRET);
292 			pcb->pcb_fsbase = i386base;
293 			td->td_frame->tf_fs = _ufssel;
294 			update_gdt_fsbase(td, i386base);
295 		}
296 		break;
297 	case I386_GET_GSBASE:
298 		update_pcb_bases(pcb);
299 		i386base = pcb->pcb_gsbase;
300 		error = copyout(&i386base, uap->parms, sizeof(i386base));
301 		break;
302 	case I386_SET_GSBASE:
303 		error = copyin(uap->parms, &i386base, sizeof(i386base));
304 		if (error == 0) {
305 			set_pcb_flags(pcb, PCB_FULL_IRET);
306 			pcb->pcb_gsbase = i386base;
307 			td->td_frame->tf_gs = _ugssel;
308 			update_gdt_gsbase(td, i386base);
309 		}
310 		break;
311 	case AMD64_GET_FSBASE:
312 		update_pcb_bases(pcb);
313 		error = copyout(&pcb->pcb_fsbase, uap->parms,
314 		    sizeof(pcb->pcb_fsbase));
315 		break;
316 
317 	case AMD64_SET_FSBASE:
318 		error = copyin(uap->parms, &a64base, sizeof(a64base));
319 		if (error == 0) {
320 			if (a64base < curproc->p_sysent->sv_maxuser) {
321 				set_pcb_flags(pcb, PCB_FULL_IRET);
322 				pcb->pcb_fsbase = a64base;
323 				td->td_frame->tf_fs = _ufssel;
324 			} else
325 				error = EINVAL;
326 		}
327 		break;
328 
329 	case AMD64_GET_GSBASE:
330 		update_pcb_bases(pcb);
331 		error = copyout(&pcb->pcb_gsbase, uap->parms,
332 		    sizeof(pcb->pcb_gsbase));
333 		break;
334 
335 	case AMD64_SET_GSBASE:
336 		error = copyin(uap->parms, &a64base, sizeof(a64base));
337 		if (error == 0) {
338 			if (a64base < curproc->p_sysent->sv_maxuser) {
339 				set_pcb_flags(pcb, PCB_FULL_IRET);
340 				pcb->pcb_gsbase = a64base;
341 				td->td_frame->tf_gs = _ugssel;
342 			} else
343 				error = EINVAL;
344 		}
345 		break;
346 
347 	case I386_GET_XFPUSTATE:
348 	case AMD64_GET_XFPUSTATE:
349 		if (a64xfpu.len > cpu_max_ext_state_size -
350 		    sizeof(struct savefpu))
351 			return (EINVAL);
352 		fpugetregs(td);
353 		error = copyout((char *)(get_pcb_user_save_td(td) + 1),
354 		    a64xfpu.addr, a64xfpu.len);
355 		break;
356 
357 	case I386_SET_PKRU:
358 	case AMD64_SET_PKRU:
359 		/*
360 		 * Read-lock the map to synchronize with parallel
361 		 * pmap_vmspace_copy() on fork.
362 		 */
363 		map = &td->td_proc->p_vmspace->vm_map;
364 		vm_map_lock_read(map);
365 		error = pmap_pkru_set(PCPU_GET(curpmap),
366 		    (vm_offset_t)a64pkru.addr, (vm_offset_t)a64pkru.addr +
367 		    a64pkru.len, a64pkru.keyidx, a64pkru.flags);
368 		vm_map_unlock_read(map);
369 		break;
370 
371 	case I386_CLEAR_PKRU:
372 	case AMD64_CLEAR_PKRU:
373 		if (a64pkru.flags != 0 || a64pkru.keyidx != 0) {
374 			error = EINVAL;
375 			break;
376 		}
377 		map = &td->td_proc->p_vmspace->vm_map;
378 		vm_map_lock_read(map);
379 		error = pmap_pkru_clear(PCPU_GET(curpmap),
380 		    (vm_offset_t)a64pkru.addr,
381 		    (vm_offset_t)a64pkru.addr + a64pkru.len);
382 		vm_map_unlock_read(map);
383 		break;
384 
385 	default:
386 		error = EINVAL;
387 		break;
388 	}
389 	return (error);
390 }
391 
392 int
amd64_set_ioperm(struct thread * td,struct i386_ioperm_args * uap)393 amd64_set_ioperm(struct thread *td, struct i386_ioperm_args *uap)
394 {
395 	char *iomap;
396 	struct amd64tss *tssp;
397 	struct system_segment_descriptor *tss_sd;
398 	struct pcb *pcb;
399 	u_int i;
400 	int error;
401 
402 	if ((error = priv_check(td, PRIV_IO)) != 0)
403 		return (error);
404 	if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
405 		return (error);
406 	if (uap->start > uap->start + uap->length ||
407 	    uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
408 		return (EINVAL);
409 
410 	/*
411 	 * XXX
412 	 * While this is restricted to root, we should probably figure out
413 	 * whether any other driver is using this i/o address, as so not to
414 	 * cause confusion.  This probably requires a global 'usage registry'.
415 	 */
416 	pcb = td->td_pcb;
417 	if (pcb->pcb_tssp == NULL) {
418 		tssp = kmem_malloc(ctob(IOPAGES + 1), M_WAITOK);
419 		pmap_pti_add_kva((vm_offset_t)tssp, (vm_offset_t)tssp +
420 		    ctob(IOPAGES + 1), false);
421 		iomap = (char *)&tssp[1];
422 		memset(iomap, 0xff, IOPERM_BITMAP_SIZE);
423 		critical_enter();
424 		/* Takes care of tss_rsp0. */
425 		memcpy(tssp, PCPU_PTR(common_tss), sizeof(struct amd64tss));
426 		tssp->tss_iobase = sizeof(*tssp);
427 		pcb->pcb_tssp = tssp;
428 		tss_sd = PCPU_GET(tss);
429 		tss_sd->sd_lobase = (u_long)tssp & 0xffffff;
430 		tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful;
431 		tss_sd->sd_type = SDT_SYSTSS;
432 		ltr(GSEL(GPROC0_SEL, SEL_KPL));
433 		PCPU_SET(tssp, tssp);
434 		critical_exit();
435 	} else
436 		iomap = (char *)&pcb->pcb_tssp[1];
437 	for (i = uap->start; i < uap->start + uap->length; i++) {
438 		if (uap->enable)
439 			iomap[i >> 3] &= ~(1 << (i & 7));
440 		else
441 			iomap[i >> 3] |= (1 << (i & 7));
442 	}
443 	return (error);
444 }
445 
446 int
amd64_get_ioperm(struct thread * td,struct i386_ioperm_args * uap)447 amd64_get_ioperm(struct thread *td, struct i386_ioperm_args *uap)
448 {
449 	int i, state;
450 	char *iomap;
451 
452 	if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
453 		return (EINVAL);
454 	if (td->td_pcb->pcb_tssp == NULL) {
455 		uap->length = 0;
456 		goto done;
457 	}
458 
459 	iomap = (char *)&td->td_pcb->pcb_tssp[1];
460 
461 	i = uap->start;
462 	state = (iomap[i >> 3] >> (i & 7)) & 1;
463 	uap->enable = !state;
464 	uap->length = 1;
465 
466 	for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
467 		if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
468 			break;
469 		uap->length++;
470 	}
471 
472 done:
473 	return (0);
474 }
475 
476 /*
477  * Update the GDT entry pointing to the LDT to point to the LDT of the
478  * current process.
479  */
480 static void
set_user_ldt(struct mdproc * mdp)481 set_user_ldt(struct mdproc *mdp)
482 {
483 
484 	*PCPU_GET(ldt) = mdp->md_ldt_sd;
485 	lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
486 }
487 
488 static void
set_user_ldt_rv(void * arg)489 set_user_ldt_rv(void *arg)
490 {
491 	struct proc *orig, *target;
492 	struct proc_ldt *ldt;
493 
494 	orig = arg;
495 	target = curthread->td_proc;
496 
497 	ldt = (void *)atomic_load_acq_ptr((uintptr_t *)&orig->p_md.md_ldt);
498 	if (target->p_md.md_ldt != ldt)
499 		return;
500 
501 	set_user_ldt(&target->p_md);
502 }
503 
504 struct proc_ldt *
user_ldt_alloc(struct proc * p,int force)505 user_ldt_alloc(struct proc *p, int force)
506 {
507 	struct proc_ldt *pldt, *new_ldt;
508 	struct mdproc *mdp;
509 	struct soft_segment_descriptor sldt;
510 	vm_offset_t sva;
511 	vm_size_t sz;
512 
513 	mtx_assert(&dt_lock, MA_OWNED);
514 	mdp = &p->p_md;
515 	if (!force && mdp->md_ldt != NULL)
516 		return (mdp->md_ldt);
517 	mtx_unlock(&dt_lock);
518 	new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
519 	sz = max_ldt_segment * sizeof(struct user_segment_descriptor);
520 	new_ldt->ldt_base = kmem_malloc(sz, M_WAITOK | M_ZERO);
521 	sva = (uintptr_t)new_ldt->ldt_base;
522 	pmap_pti_add_kva(sva, sva + sz, false);
523 	new_ldt->ldt_refcnt = 1;
524 	sldt.ssd_base = sva;
525 	sldt.ssd_limit = sz - 1;
526 	sldt.ssd_type = SDT_SYSLDT;
527 	sldt.ssd_dpl = SEL_KPL;
528 	sldt.ssd_p = 1;
529 	sldt.ssd_long = 0;
530 	sldt.ssd_def32 = 0;
531 	sldt.ssd_gran = 0;
532 	mtx_lock(&dt_lock);
533 	pldt = mdp->md_ldt;
534 	if (pldt != NULL && !force) {
535 		pmap_pti_remove_kva(sva, sva + sz);
536 		kmem_free(new_ldt->ldt_base, sz);
537 		free(new_ldt, M_SUBPROC);
538 		return (pldt);
539 	}
540 
541 	if (pldt != NULL) {
542 		bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
543 		    sizeof(struct user_segment_descriptor));
544 		user_ldt_derefl(pldt);
545 	}
546 	critical_enter();
547 	ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
548 	atomic_thread_fence_rel();
549 	mdp->md_ldt = new_ldt;
550 	critical_exit();
551 	smp_rendezvous(NULL, set_user_ldt_rv, NULL, p);
552 
553 	return (mdp->md_ldt);
554 }
555 
556 void
user_ldt_free(struct thread * td)557 user_ldt_free(struct thread *td)
558 {
559 	struct proc *p = td->td_proc;
560 	struct mdproc *mdp = &p->p_md;
561 	struct proc_ldt *pldt;
562 
563 	mtx_lock(&dt_lock);
564 	if ((pldt = mdp->md_ldt) == NULL) {
565 		mtx_unlock(&dt_lock);
566 		return;
567 	}
568 
569 	critical_enter();
570 	mdp->md_ldt = NULL;
571 	atomic_thread_fence_rel();
572 	bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
573 	if (td == curthread)
574 		lldt(GSEL(GNULL_SEL, SEL_KPL));
575 	critical_exit();
576 	user_ldt_deref(pldt);
577 }
578 
579 static void
user_ldt_derefl(struct proc_ldt * pldt)580 user_ldt_derefl(struct proc_ldt *pldt)
581 {
582 	vm_offset_t sva;
583 	vm_size_t sz;
584 
585 	if (--pldt->ldt_refcnt == 0) {
586 		sva = (vm_offset_t)pldt->ldt_base;
587 		sz = max_ldt_segment * sizeof(struct user_segment_descriptor);
588 		pmap_pti_remove_kva(sva, sva + sz);
589 		kmem_free(pldt->ldt_base, sz);
590 		free(pldt, M_SUBPROC);
591 	}
592 }
593 
594 static void
user_ldt_deref(struct proc_ldt * pldt)595 user_ldt_deref(struct proc_ldt *pldt)
596 {
597 
598 	mtx_assert(&dt_lock, MA_OWNED);
599 	user_ldt_derefl(pldt);
600 	mtx_unlock(&dt_lock);
601 }
602 
603 /*
604  * Note for the authors of compat layers (linux, etc): copyout() in
605  * the function below is not a problem since it presents data in
606  * arch-specific format (i.e. i386-specific in this case), not in
607  * the OS-specific one.
608  */
609 int
amd64_get_ldt(struct thread * td,struct i386_ldt_args * uap)610 amd64_get_ldt(struct thread *td, struct i386_ldt_args *uap)
611 {
612 	struct proc_ldt *pldt;
613 	struct user_segment_descriptor *lp;
614 	uint64_t *data;
615 	u_int i, num;
616 	int error;
617 
618 #ifdef	DEBUG
619 	printf("amd64_get_ldt: start=%u num=%u descs=%p\n",
620 	    uap->start, uap->num, (void *)uap->descs);
621 #endif
622 
623 	pldt = td->td_proc->p_md.md_ldt;
624 	if (pldt == NULL || uap->start >= max_ldt_segment || uap->num == 0) {
625 		td->td_retval[0] = 0;
626 		return (0);
627 	}
628 	num = min(uap->num, max_ldt_segment - uap->start);
629 	lp = &((struct user_segment_descriptor *)(pldt->ldt_base))[uap->start];
630 	data = malloc(num * sizeof(struct user_segment_descriptor), M_TEMP,
631 	    M_WAITOK);
632 	mtx_lock(&dt_lock);
633 	for (i = 0; i < num; i++)
634 		data[i] = ((volatile uint64_t *)lp)[i];
635 	mtx_unlock(&dt_lock);
636 	error = copyout(data, uap->descs, num *
637 	    sizeof(struct user_segment_descriptor));
638 	free(data, M_TEMP);
639 	if (error == 0)
640 		td->td_retval[0] = num;
641 	return (error);
642 }
643 
644 int
amd64_set_ldt(struct thread * td,struct i386_ldt_args * uap,struct user_segment_descriptor * descs)645 amd64_set_ldt(struct thread *td, struct i386_ldt_args *uap,
646     struct user_segment_descriptor *descs)
647 {
648 	struct mdproc *mdp;
649 	struct proc_ldt *pldt;
650 	struct user_segment_descriptor *dp;
651 	struct proc *p;
652 	u_int largest_ld, i;
653 	int error;
654 
655 #ifdef	DEBUG
656 	printf("amd64_set_ldt: start=%u num=%u descs=%p\n",
657 	    uap->start, uap->num, (void *)uap->descs);
658 #endif
659 	mdp = &td->td_proc->p_md;
660 	error = 0;
661 
662 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
663 	p = td->td_proc;
664 	if (descs == NULL) {
665 		/* Free descriptors */
666 		if (uap->start == 0 && uap->num == 0)
667 			uap->num = max_ldt_segment;
668 		if (uap->num == 0)
669 			return (EINVAL);
670 		if ((pldt = mdp->md_ldt) == NULL ||
671 		    uap->start >= max_ldt_segment)
672 			return (0);
673 		largest_ld = uap->start + uap->num;
674 		if (largest_ld > max_ldt_segment)
675 			largest_ld = max_ldt_segment;
676 		if (largest_ld < uap->start)
677 			return (EINVAL);
678 		mtx_lock(&dt_lock);
679 		for (i = uap->start; i < largest_ld; i++)
680 			((volatile uint64_t *)(pldt->ldt_base))[i] = 0;
681 		mtx_unlock(&dt_lock);
682 		return (0);
683 	}
684 
685 	if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
686 		/* verify range of descriptors to modify */
687 		largest_ld = uap->start + uap->num;
688 		if (uap->start >= max_ldt_segment ||
689 		    largest_ld > max_ldt_segment ||
690 		    largest_ld < uap->start)
691 			return (EINVAL);
692 	}
693 
694 	/* Check descriptors for access violations */
695 	for (i = 0; i < uap->num; i++) {
696 		dp = &descs[i];
697 
698 		switch (dp->sd_type) {
699 		case SDT_SYSNULL:	/* system null */
700 			dp->sd_p = 0;
701 			break;
702 		case SDT_SYS286TSS:
703 		case SDT_SYSLDT:
704 		case SDT_SYS286BSY:
705 		case SDT_SYS286CGT:
706 		case SDT_SYSTASKGT:
707 		case SDT_SYS286IGT:
708 		case SDT_SYS286TGT:
709 		case SDT_SYSNULL2:
710 		case SDT_SYSTSS:
711 		case SDT_SYSNULL3:
712 		case SDT_SYSBSY:
713 		case SDT_SYSCGT:
714 		case SDT_SYSNULL4:
715 		case SDT_SYSIGT:
716 		case SDT_SYSTGT:
717 			return (EACCES);
718 
719 		/* memory segment types */
720 		case SDT_MEMEC:   /* memory execute only conforming */
721 		case SDT_MEMEAC:  /* memory execute only accessed conforming */
722 		case SDT_MEMERC:  /* memory execute read conforming */
723 		case SDT_MEMERAC: /* memory execute read accessed conforming */
724 			 /* Must be "present" if executable and conforming. */
725 			if (dp->sd_p == 0)
726 				return (EACCES);
727 			break;
728 		case SDT_MEMRO:   /* memory read only */
729 		case SDT_MEMROA:  /* memory read only accessed */
730 		case SDT_MEMRW:   /* memory read write */
731 		case SDT_MEMRWA:  /* memory read write accessed */
732 		case SDT_MEMROD:  /* memory read only expand dwn limit */
733 		case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
734 		case SDT_MEMRWD:  /* memory read write expand dwn limit */
735 		case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
736 		case SDT_MEME:    /* memory execute only */
737 		case SDT_MEMEA:   /* memory execute only accessed */
738 		case SDT_MEMER:   /* memory execute read */
739 		case SDT_MEMERA:  /* memory execute read accessed */
740 			break;
741 		default:
742 			return(EINVAL);
743 		}
744 
745 		/* Only user (ring-3) descriptors may be present. */
746 		if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
747 			return (EACCES);
748 	}
749 
750 	if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
751 		/* Allocate a free slot */
752 		mtx_lock(&dt_lock);
753 		pldt = user_ldt_alloc(p, 0);
754 		if (pldt == NULL) {
755 			mtx_unlock(&dt_lock);
756 			return (ENOMEM);
757 		}
758 
759 		/*
760 		 * start scanning a bit up to leave room for NVidia and
761 		 * Wine, which still user the "Blat" method of allocation.
762 		 */
763 		i = 16;
764 		dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
765 		for (; i < max_ldt_segment; ++i, ++dp) {
766 			if (dp->sd_type == SDT_SYSNULL)
767 				break;
768 		}
769 		if (i >= max_ldt_segment) {
770 			mtx_unlock(&dt_lock);
771 			return (ENOSPC);
772 		}
773 		uap->start = i;
774 		error = amd64_set_ldt_data(td, i, 1, descs);
775 		mtx_unlock(&dt_lock);
776 	} else {
777 		largest_ld = uap->start + uap->num;
778 		if (largest_ld > max_ldt_segment)
779 			return (EINVAL);
780 		mtx_lock(&dt_lock);
781 		if (user_ldt_alloc(p, 0) != NULL) {
782 			error = amd64_set_ldt_data(td, uap->start, uap->num,
783 			    descs);
784 		}
785 		mtx_unlock(&dt_lock);
786 	}
787 	if (error == 0)
788 		td->td_retval[0] = uap->start;
789 	return (error);
790 }
791 
792 int
amd64_set_ldt_data(struct thread * td,int start,int num,struct user_segment_descriptor * descs)793 amd64_set_ldt_data(struct thread *td, int start, int num,
794     struct user_segment_descriptor *descs)
795 {
796 	struct mdproc *mdp;
797 	struct proc_ldt *pldt;
798 	volatile uint64_t *dst, *src;
799 	int i;
800 
801 	mtx_assert(&dt_lock, MA_OWNED);
802 
803 	mdp = &td->td_proc->p_md;
804 	pldt = mdp->md_ldt;
805 	dst = (volatile uint64_t *)(pldt->ldt_base);
806 	src = (volatile uint64_t *)descs;
807 	for (i = 0; i < num; i++)
808 		dst[start + i] = src[i];
809 	return (0);
810 }
811