1 /*-
2 * Copyright (c) 1990 The Regents of the University of California.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/10.1/sys/i386/i386/sys_machdep.c 267714 2014-06-22 08:32:31Z kib $");
34
35 #include "opt_capsicum.h"
36 #include "opt_kstack_pages.h"
37
38 #include <sys/param.h>
39 #include <sys/capability.h>
40 #include <sys/systm.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/smp.h>
47 #include <sys/sysproto.h>
48
49 #include <vm/vm.h>
50 #include <vm/pmap.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_extern.h>
53
54 #include <machine/cpu.h>
55 #include <machine/pcb.h>
56 #include <machine/pcb_ext.h>
57 #include <machine/proc.h>
58 #include <machine/sysarch.h>
59
60 #include <security/audit/audit.h>
61
62 #ifdef XEN
63 #include <machine/xen/xenfunc.h>
64
65 void i386_reset_ldt(struct proc_ldt *pldt);
66
67 void
68 i386_reset_ldt(struct proc_ldt *pldt)
69 {
70 xen_set_ldt((vm_offset_t)pldt->ldt_base, pldt->ldt_len);
71 }
72 #else
73 #define i386_reset_ldt(x)
74 #endif
75
76 #include <vm/vm_kern.h> /* for kernel_map */
77
78 #define MAX_LD 8192
79 #define LD_PER_PAGE 512
80 #define NEW_MAX_LD(num) ((num + LD_PER_PAGE) & ~(LD_PER_PAGE-1))
81 #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3)
82 #define NULL_LDT_BASE ((caddr_t)NULL)
83
84 #ifdef SMP
85 static void set_user_ldt_rv(struct vmspace *vmsp);
86 #endif
87 static int i386_set_ldt_data(struct thread *, int start, int num,
88 union descriptor *descs);
89 static int i386_ldt_grow(struct thread *td, int len);
90
91 #ifndef _SYS_SYSPROTO_H_
92 struct sysarch_args {
93 int op;
94 char *parms;
95 };
96 #endif
97
98 int
99 sysarch(td, uap)
100 struct thread *td;
101 register struct sysarch_args *uap;
102 {
103 int error;
104 union descriptor *lp;
105 union {
106 struct i386_ldt_args largs;
107 struct i386_ioperm_args iargs;
108 } kargs;
109 uint32_t base;
110 struct segment_descriptor sd, *sdp;
111
112 AUDIT_ARG_CMD(uap->op);
113
114 #ifdef CAPABILITY_MODE
115 /*
116 * When adding new operations, add a new case statement here to
117 * explicitly indicate whether or not the operation is safe to
118 * perform in capability mode.
119 */
120 if (IN_CAPABILITY_MODE(td)) {
121 switch (uap->op) {
122 case I386_GET_LDT:
123 case I386_SET_LDT:
124 case I386_GET_IOPERM:
125 case I386_GET_FSBASE:
126 case I386_SET_FSBASE:
127 case I386_GET_GSBASE:
128 case I386_SET_GSBASE:
129 break;
130
131 case I386_SET_IOPERM:
132 default:
133 #ifdef KTRACE
134 if (KTRPOINT(td, KTR_CAPFAIL))
135 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
136 #endif
137 return (ECAPMODE);
138 }
139 }
140 #endif
141
142 switch (uap->op) {
143 case I386_GET_IOPERM:
144 case I386_SET_IOPERM:
145 if ((error = copyin(uap->parms, &kargs.iargs,
146 sizeof(struct i386_ioperm_args))) != 0)
147 return (error);
148 break;
149 case I386_GET_LDT:
150 case I386_SET_LDT:
151 if ((error = copyin(uap->parms, &kargs.largs,
152 sizeof(struct i386_ldt_args))) != 0)
153 return (error);
154 if (kargs.largs.num > MAX_LD || kargs.largs.num <= 0)
155 return (EINVAL);
156 break;
157 default:
158 break;
159 }
160
161 switch(uap->op) {
162 case I386_GET_LDT:
163 error = i386_get_ldt(td, &kargs.largs);
164 break;
165 case I386_SET_LDT:
166 if (kargs.largs.descs != NULL) {
167 lp = (union descriptor *)malloc(
168 kargs.largs.num * sizeof(union descriptor),
169 M_TEMP, M_WAITOK);
170 error = copyin(kargs.largs.descs, lp,
171 kargs.largs.num * sizeof(union descriptor));
172 if (error == 0)
173 error = i386_set_ldt(td, &kargs.largs, lp);
174 free(lp, M_TEMP);
175 } else {
176 error = i386_set_ldt(td, &kargs.largs, NULL);
177 }
178 break;
179 case I386_GET_IOPERM:
180 error = i386_get_ioperm(td, &kargs.iargs);
181 if (error == 0)
182 error = copyout(&kargs.iargs, uap->parms,
183 sizeof(struct i386_ioperm_args));
184 break;
185 case I386_SET_IOPERM:
186 error = i386_set_ioperm(td, &kargs.iargs);
187 break;
188 case I386_VM86:
189 error = vm86_sysarch(td, uap->parms);
190 break;
191 case I386_GET_FSBASE:
192 sdp = &td->td_pcb->pcb_fsd;
193 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
194 error = copyout(&base, uap->parms, sizeof(base));
195 break;
196 case I386_SET_FSBASE:
197 error = copyin(uap->parms, &base, sizeof(base));
198 if (!error) {
199 /*
200 * Construct a descriptor and store it in the pcb for
201 * the next context switch. Also store it in the gdt
202 * so that the load of tf_fs into %fs will activate it
203 * at return to userland.
204 */
205 sd.sd_lobase = base & 0xffffff;
206 sd.sd_hibase = (base >> 24) & 0xff;
207 #ifdef XEN
208 /* need to do nosegneg like Linux */
209 sd.sd_lolimit = (HYPERVISOR_VIRT_START >> 12) & 0xffff;
210 #else
211 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
212 #endif
213 sd.sd_hilimit = 0xf;
214 sd.sd_type = SDT_MEMRWA;
215 sd.sd_dpl = SEL_UPL;
216 sd.sd_p = 1;
217 sd.sd_xx = 0;
218 sd.sd_def32 = 1;
219 sd.sd_gran = 1;
220 critical_enter();
221 td->td_pcb->pcb_fsd = sd;
222 #ifdef XEN
223 HYPERVISOR_update_descriptor(vtomach(&PCPU_GET(fsgs_gdt)[0]),
224 *(uint64_t *)&sd);
225 #else
226 PCPU_GET(fsgs_gdt)[0] = sd;
227 #endif
228 critical_exit();
229 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
230 }
231 break;
232 case I386_GET_GSBASE:
233 sdp = &td->td_pcb->pcb_gsd;
234 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
235 error = copyout(&base, uap->parms, sizeof(base));
236 break;
237 case I386_SET_GSBASE:
238 error = copyin(uap->parms, &base, sizeof(base));
239 if (!error) {
240 /*
241 * Construct a descriptor and store it in the pcb for
242 * the next context switch. Also store it in the gdt
243 * because we have to do a load_gs() right now.
244 */
245 sd.sd_lobase = base & 0xffffff;
246 sd.sd_hibase = (base >> 24) & 0xff;
247
248 #ifdef XEN
249 /* need to do nosegneg like Linux */
250 sd.sd_lolimit = (HYPERVISOR_VIRT_START >> 12) & 0xffff;
251 #else
252 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
253 #endif
254 sd.sd_hilimit = 0xf;
255 sd.sd_type = SDT_MEMRWA;
256 sd.sd_dpl = SEL_UPL;
257 sd.sd_p = 1;
258 sd.sd_xx = 0;
259 sd.sd_def32 = 1;
260 sd.sd_gran = 1;
261 critical_enter();
262 td->td_pcb->pcb_gsd = sd;
263 #ifdef XEN
264 HYPERVISOR_update_descriptor(vtomach(&PCPU_GET(fsgs_gdt)[1]),
265 *(uint64_t *)&sd);
266 #else
267 PCPU_GET(fsgs_gdt)[1] = sd;
268 #endif
269 critical_exit();
270 load_gs(GSEL(GUGS_SEL, SEL_UPL));
271 }
272 break;
273 default:
274 error = EINVAL;
275 break;
276 }
277 return (error);
278 }
279
280 int
281 i386_extend_pcb(struct thread *td)
282 {
283 int i, offset;
284 u_long *addr;
285 struct pcb_ext *ext;
286 struct soft_segment_descriptor ssd = {
287 0, /* segment base address (overwritten) */
288 ctob(IOPAGES + 1) - 1, /* length */
289 SDT_SYS386TSS, /* segment type */
290 0, /* priority level */
291 1, /* descriptor present */
292 0, 0,
293 0, /* default 32 size */
294 0 /* granularity */
295 };
296
297 ext = (struct pcb_ext *)kmem_malloc(kernel_arena, ctob(IOPAGES+1),
298 M_WAITOK | M_ZERO);
299 /* -16 is so we can convert a trapframe into vm86trapframe inplace */
300 ext->ext_tss.tss_esp0 = td->td_kstack + ctob(KSTACK_PAGES) -
301 sizeof(struct pcb) - 16;
302 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
303 /*
304 * The last byte of the i/o map must be followed by an 0xff byte.
305 * We arbitrarily allocate 16 bytes here, to keep the starting
306 * address on a doubleword boundary.
307 */
308 offset = PAGE_SIZE - 16;
309 ext->ext_tss.tss_ioopt =
310 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16;
311 ext->ext_iomap = (caddr_t)ext + offset;
312 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32;
313
314 addr = (u_long *)ext->ext_vm86.vm86_intmap;
315 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++)
316 *addr++ = ~0;
317
318 ssd.ssd_base = (unsigned)&ext->ext_tss;
319 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext);
320 ssdtosd(&ssd, &ext->ext_tssd);
321
322 KASSERT(td == curthread, ("giving TSS to !curthread"));
323 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!"));
324
325 /* Switch to the new TSS. */
326 critical_enter();
327 td->td_pcb->pcb_ext = ext;
328 PCPU_SET(private_tss, 1);
329 *PCPU_GET(tss_gdt) = ext->ext_tssd;
330 ltr(GSEL(GPROC0_SEL, SEL_KPL));
331 critical_exit();
332
333 return 0;
334 }
335
336 int
337 i386_set_ioperm(td, uap)
338 struct thread *td;
339 struct i386_ioperm_args *uap;
340 {
341 int i, error;
342 char *iomap;
343
344 if ((error = priv_check(td, PRIV_IO)) != 0)
345 return (error);
346 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
347 return (error);
348 /*
349 * XXX
350 * While this is restricted to root, we should probably figure out
351 * whether any other driver is using this i/o address, as so not to
352 * cause confusion. This probably requires a global 'usage registry'.
353 */
354
355 if (td->td_pcb->pcb_ext == 0)
356 if ((error = i386_extend_pcb(td)) != 0)
357 return (error);
358 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
359
360 if (uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
361 return (EINVAL);
362
363 for (i = uap->start; i < uap->start + uap->length; i++) {
364 if (uap->enable)
365 iomap[i >> 3] &= ~(1 << (i & 7));
366 else
367 iomap[i >> 3] |= (1 << (i & 7));
368 }
369 return (error);
370 }
371
372 int
373 i386_get_ioperm(td, uap)
374 struct thread *td;
375 struct i386_ioperm_args *uap;
376 {
377 int i, state;
378 char *iomap;
379
380 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
381 return (EINVAL);
382
383 if (td->td_pcb->pcb_ext == 0) {
384 uap->length = 0;
385 goto done;
386 }
387
388 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
389
390 i = uap->start;
391 state = (iomap[i >> 3] >> (i & 7)) & 1;
392 uap->enable = !state;
393 uap->length = 1;
394
395 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
396 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
397 break;
398 uap->length++;
399 }
400
401 done:
402 return (0);
403 }
404
405 /*
406 * Update the GDT entry pointing to the LDT to point to the LDT of the
407 * current process. Manage dt_lock holding/unholding autonomously.
408 */
409 void
410 set_user_ldt(struct mdproc *mdp)
411 {
412 struct proc_ldt *pldt;
413 int dtlocked;
414
415 dtlocked = 0;
416 if (!mtx_owned(&dt_lock)) {
417 mtx_lock_spin(&dt_lock);
418 dtlocked = 1;
419 }
420
421 pldt = mdp->md_ldt;
422 #ifdef XEN
423 i386_reset_ldt(pldt);
424 PCPU_SET(currentldt, (int)pldt);
425 #else
426 #ifdef SMP
427 gdt[PCPU_GET(cpuid) * NGDT + GUSERLDT_SEL].sd = pldt->ldt_sd;
428 #else
429 gdt[GUSERLDT_SEL].sd = pldt->ldt_sd;
430 #endif
431 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
432 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL));
433 #endif /* XEN */
434 if (dtlocked)
435 mtx_unlock_spin(&dt_lock);
436 }
437
438 #ifdef SMP
439 static void
440 set_user_ldt_rv(struct vmspace *vmsp)
441 {
442 struct thread *td;
443
444 td = curthread;
445 if (vmsp != td->td_proc->p_vmspace)
446 return;
447
448 set_user_ldt(&td->td_proc->p_md);
449 }
450 #endif
451
452 #ifdef XEN
453
454 /*
455 * dt_lock must be held. Returns with dt_lock held.
456 */
457 struct proc_ldt *
458 user_ldt_alloc(struct mdproc *mdp, int len)
459 {
460 struct proc_ldt *pldt, *new_ldt;
461
462 mtx_assert(&dt_lock, MA_OWNED);
463 mtx_unlock_spin(&dt_lock);
464 new_ldt = malloc(sizeof(struct proc_ldt),
465 M_SUBPROC, M_WAITOK);
466
467 new_ldt->ldt_len = len = NEW_MAX_LD(len);
468 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
469 round_page(len * sizeof(union descriptor)), M_WAITOK);
470 new_ldt->ldt_refcnt = 1;
471 new_ldt->ldt_active = 0;
472
473 mtx_lock_spin(&dt_lock);
474 if ((pldt = mdp->md_ldt)) {
475 if (len > pldt->ldt_len)
476 len = pldt->ldt_len;
477 bcopy(pldt->ldt_base, new_ldt->ldt_base,
478 len * sizeof(union descriptor));
479 } else {
480 bcopy(ldt, new_ldt->ldt_base, PAGE_SIZE);
481 }
482 mtx_unlock_spin(&dt_lock); /* XXX kill once pmap locking fixed. */
483 pmap_map_readonly(kernel_pmap, (vm_offset_t)new_ldt->ldt_base,
484 new_ldt->ldt_len*sizeof(union descriptor));
485 mtx_lock_spin(&dt_lock); /* XXX kill once pmap locking fixed. */
486 return (new_ldt);
487 }
488 #else
489 /*
490 * dt_lock must be held. Returns with dt_lock held.
491 */
492 struct proc_ldt *
493 user_ldt_alloc(struct mdproc *mdp, int len)
494 {
495 struct proc_ldt *pldt, *new_ldt;
496
497 mtx_assert(&dt_lock, MA_OWNED);
498 mtx_unlock_spin(&dt_lock);
499 new_ldt = malloc(sizeof(struct proc_ldt),
500 M_SUBPROC, M_WAITOK);
501
502 new_ldt->ldt_len = len = NEW_MAX_LD(len);
503 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
504 len * sizeof(union descriptor), M_WAITOK);
505 new_ldt->ldt_refcnt = 1;
506 new_ldt->ldt_active = 0;
507
508 mtx_lock_spin(&dt_lock);
509 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base;
510 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1;
511 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd);
512
513 if ((pldt = mdp->md_ldt) != NULL) {
514 if (len > pldt->ldt_len)
515 len = pldt->ldt_len;
516 bcopy(pldt->ldt_base, new_ldt->ldt_base,
517 len * sizeof(union descriptor));
518 } else
519 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt));
520
521 return (new_ldt);
522 }
523 #endif /* !XEN */
524
525 /*
526 * Must be called with dt_lock held. Returns with dt_lock unheld.
527 */
528 void
529 user_ldt_free(struct thread *td)
530 {
531 struct mdproc *mdp = &td->td_proc->p_md;
532 struct proc_ldt *pldt;
533
534 mtx_assert(&dt_lock, MA_OWNED);
535 if ((pldt = mdp->md_ldt) == NULL) {
536 mtx_unlock_spin(&dt_lock);
537 return;
538 }
539
540 if (td == curthread) {
541 #ifdef XEN
542 i386_reset_ldt(&default_proc_ldt);
543 PCPU_SET(currentldt, (int)&default_proc_ldt);
544 #else
545 lldt(_default_ldt);
546 PCPU_SET(currentldt, _default_ldt);
547 #endif
548 }
549
550 mdp->md_ldt = NULL;
551 user_ldt_deref(pldt);
552 }
553
554 void
555 user_ldt_deref(struct proc_ldt *pldt)
556 {
557
558 mtx_assert(&dt_lock, MA_OWNED);
559 if (--pldt->ldt_refcnt == 0) {
560 mtx_unlock_spin(&dt_lock);
561 kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
562 pldt->ldt_len * sizeof(union descriptor));
563 free(pldt, M_SUBPROC);
564 } else
565 mtx_unlock_spin(&dt_lock);
566 }
567
568 /*
569 * Note for the authors of compat layers (linux, etc): copyout() in
570 * the function below is not a problem since it presents data in
571 * arch-specific format (i.e. i386-specific in this case), not in
572 * the OS-specific one.
573 */
574 int
575 i386_get_ldt(td, uap)
576 struct thread *td;
577 struct i386_ldt_args *uap;
578 {
579 int error = 0;
580 struct proc_ldt *pldt;
581 int nldt, num;
582 union descriptor *lp;
583
584 #ifdef DEBUG
585 printf("i386_get_ldt: start=%d num=%d descs=%p\n",
586 uap->start, uap->num, (void *)uap->descs);
587 #endif
588
589 mtx_lock_spin(&dt_lock);
590 if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
591 nldt = pldt->ldt_len;
592 lp = &((union descriptor *)(pldt->ldt_base))[uap->start];
593 mtx_unlock_spin(&dt_lock);
594 num = min(uap->num, nldt);
595 } else {
596 mtx_unlock_spin(&dt_lock);
597 nldt = sizeof(ldt)/sizeof(ldt[0]);
598 num = min(uap->num, nldt);
599 lp = &ldt[uap->start];
600 }
601
602 if ((uap->start > (unsigned int)nldt) ||
603 ((unsigned int)num > (unsigned int)nldt) ||
604 ((unsigned int)(uap->start + num) > (unsigned int)nldt))
605 return(EINVAL);
606
607 error = copyout(lp, uap->descs, num * sizeof(union descriptor));
608 if (!error)
609 td->td_retval[0] = num;
610
611 return(error);
612 }
613
614 int
615 i386_set_ldt(td, uap, descs)
616 struct thread *td;
617 struct i386_ldt_args *uap;
618 union descriptor *descs;
619 {
620 int error = 0, i;
621 int largest_ld;
622 struct mdproc *mdp = &td->td_proc->p_md;
623 struct proc_ldt *pldt;
624 union descriptor *dp;
625
626 #ifdef DEBUG
627 printf("i386_set_ldt: start=%d num=%d descs=%p\n",
628 uap->start, uap->num, (void *)uap->descs);
629 #endif
630
631 if (descs == NULL) {
632 /* Free descriptors */
633 if (uap->start == 0 && uap->num == 0) {
634 /*
635 * Treat this as a special case, so userland needn't
636 * know magic number NLDT.
637 */
638 uap->start = NLDT;
639 uap->num = MAX_LD - NLDT;
640 }
641 if (uap->num == 0)
642 return (EINVAL);
643 mtx_lock_spin(&dt_lock);
644 if ((pldt = mdp->md_ldt) == NULL ||
645 uap->start >= pldt->ldt_len) {
646 mtx_unlock_spin(&dt_lock);
647 return (0);
648 }
649 largest_ld = uap->start + uap->num;
650 if (largest_ld > pldt->ldt_len)
651 largest_ld = pldt->ldt_len;
652 i = largest_ld - uap->start;
653 bzero(&((union descriptor *)(pldt->ldt_base))[uap->start],
654 sizeof(union descriptor) * i);
655 mtx_unlock_spin(&dt_lock);
656 return (0);
657 }
658
659 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
660 /* verify range of descriptors to modify */
661 largest_ld = uap->start + uap->num;
662 if (uap->start >= MAX_LD || largest_ld > MAX_LD) {
663 return (EINVAL);
664 }
665 }
666
667 /* Check descriptors for access violations */
668 for (i = 0; i < uap->num; i++) {
669 dp = &descs[i];
670
671 switch (dp->sd.sd_type) {
672 case SDT_SYSNULL: /* system null */
673 dp->sd.sd_p = 0;
674 break;
675 case SDT_SYS286TSS: /* system 286 TSS available */
676 case SDT_SYSLDT: /* system local descriptor table */
677 case SDT_SYS286BSY: /* system 286 TSS busy */
678 case SDT_SYSTASKGT: /* system task gate */
679 case SDT_SYS286IGT: /* system 286 interrupt gate */
680 case SDT_SYS286TGT: /* system 286 trap gate */
681 case SDT_SYSNULL2: /* undefined by Intel */
682 case SDT_SYS386TSS: /* system 386 TSS available */
683 case SDT_SYSNULL3: /* undefined by Intel */
684 case SDT_SYS386BSY: /* system 386 TSS busy */
685 case SDT_SYSNULL4: /* undefined by Intel */
686 case SDT_SYS386IGT: /* system 386 interrupt gate */
687 case SDT_SYS386TGT: /* system 386 trap gate */
688 case SDT_SYS286CGT: /* system 286 call gate */
689 case SDT_SYS386CGT: /* system 386 call gate */
690 /* I can't think of any reason to allow a user proc
691 * to create a segment of these types. They are
692 * for OS use only.
693 */
694 return (EACCES);
695 /*NOTREACHED*/
696
697 /* memory segment types */
698 case SDT_MEMEC: /* memory execute only conforming */
699 case SDT_MEMEAC: /* memory execute only accessed conforming */
700 case SDT_MEMERC: /* memory execute read conforming */
701 case SDT_MEMERAC: /* memory execute read accessed conforming */
702 /* Must be "present" if executable and conforming. */
703 if (dp->sd.sd_p == 0)
704 return (EACCES);
705 break;
706 case SDT_MEMRO: /* memory read only */
707 case SDT_MEMROA: /* memory read only accessed */
708 case SDT_MEMRW: /* memory read write */
709 case SDT_MEMRWA: /* memory read write accessed */
710 case SDT_MEMROD: /* memory read only expand dwn limit */
711 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
712 case SDT_MEMRWD: /* memory read write expand dwn limit */
713 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
714 case SDT_MEME: /* memory execute only */
715 case SDT_MEMEA: /* memory execute only accessed */
716 case SDT_MEMER: /* memory execute read */
717 case SDT_MEMERA: /* memory execute read accessed */
718 break;
719 default:
720 return(EINVAL);
721 /*NOTREACHED*/
722 }
723
724 /* Only user (ring-3) descriptors may be present. */
725 if ((dp->sd.sd_p != 0) && (dp->sd.sd_dpl != SEL_UPL))
726 return (EACCES);
727 }
728
729 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
730 /* Allocate a free slot */
731 mtx_lock_spin(&dt_lock);
732 if ((pldt = mdp->md_ldt) == NULL) {
733 if ((error = i386_ldt_grow(td, NLDT + 1))) {
734 mtx_unlock_spin(&dt_lock);
735 return (error);
736 }
737 pldt = mdp->md_ldt;
738 }
739 again:
740 /*
741 * start scanning a bit up to leave room for NVidia and
742 * Wine, which still user the "Blat" method of allocation.
743 */
744 dp = &((union descriptor *)(pldt->ldt_base))[NLDT];
745 for (i = NLDT; i < pldt->ldt_len; ++i) {
746 if (dp->sd.sd_type == SDT_SYSNULL)
747 break;
748 dp++;
749 }
750 if (i >= pldt->ldt_len) {
751 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) {
752 mtx_unlock_spin(&dt_lock);
753 return (error);
754 }
755 goto again;
756 }
757 uap->start = i;
758 error = i386_set_ldt_data(td, i, 1, descs);
759 mtx_unlock_spin(&dt_lock);
760 } else {
761 largest_ld = uap->start + uap->num;
762 mtx_lock_spin(&dt_lock);
763 if (!(error = i386_ldt_grow(td, largest_ld))) {
764 error = i386_set_ldt_data(td, uap->start, uap->num,
765 descs);
766 }
767 mtx_unlock_spin(&dt_lock);
768 }
769 if (error == 0)
770 td->td_retval[0] = uap->start;
771 return (error);
772 }
773 #ifdef XEN
774 static int
775 i386_set_ldt_data(struct thread *td, int start, int num,
776 union descriptor *descs)
777 {
778 struct mdproc *mdp = &td->td_proc->p_md;
779 struct proc_ldt *pldt = mdp->md_ldt;
780
781 mtx_assert(&dt_lock, MA_OWNED);
782
783 while (num) {
784 xen_update_descriptor(
785 &((union descriptor *)(pldt->ldt_base))[start],
786 descs);
787 num--;
788 start++;
789 descs++;
790 }
791 return (0);
792 }
793 #else
794 static int
795 i386_set_ldt_data(struct thread *td, int start, int num,
796 union descriptor *descs)
797 {
798 struct mdproc *mdp = &td->td_proc->p_md;
799 struct proc_ldt *pldt = mdp->md_ldt;
800
801 mtx_assert(&dt_lock, MA_OWNED);
802
803 /* Fill in range */
804 bcopy(descs,
805 &((union descriptor *)(pldt->ldt_base))[start],
806 num * sizeof(union descriptor));
807 return (0);
808 }
809 #endif /* !XEN */
810
811 static int
812 i386_ldt_grow(struct thread *td, int len)
813 {
814 struct mdproc *mdp = &td->td_proc->p_md;
815 struct proc_ldt *new_ldt, *pldt;
816 caddr_t old_ldt_base = NULL_LDT_BASE;
817 int old_ldt_len = 0;
818
819 mtx_assert(&dt_lock, MA_OWNED);
820
821 if (len > MAX_LD)
822 return (ENOMEM);
823 if (len < NLDT + 1)
824 len = NLDT + 1;
825
826 /* Allocate a user ldt. */
827 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) {
828 new_ldt = user_ldt_alloc(mdp, len);
829 if (new_ldt == NULL)
830 return (ENOMEM);
831 pldt = mdp->md_ldt;
832
833 if (pldt != NULL) {
834 if (new_ldt->ldt_len <= pldt->ldt_len) {
835 /*
836 * We just lost the race for allocation, so
837 * free the new object and return.
838 */
839 mtx_unlock_spin(&dt_lock);
840 kmem_free(kernel_arena,
841 (vm_offset_t)new_ldt->ldt_base,
842 new_ldt->ldt_len * sizeof(union descriptor));
843 free(new_ldt, M_SUBPROC);
844 mtx_lock_spin(&dt_lock);
845 return (0);
846 }
847
848 /*
849 * We have to substitute the current LDT entry for
850 * curproc with the new one since its size grew.
851 */
852 old_ldt_base = pldt->ldt_base;
853 old_ldt_len = pldt->ldt_len;
854 pldt->ldt_sd = new_ldt->ldt_sd;
855 pldt->ldt_base = new_ldt->ldt_base;
856 pldt->ldt_len = new_ldt->ldt_len;
857 } else
858 mdp->md_ldt = pldt = new_ldt;
859 #ifdef SMP
860 /*
861 * Signal other cpus to reload ldt. We need to unlock dt_lock
862 * here because other CPU will contest on it since their
863 * curthreads won't hold the lock and will block when trying
864 * to acquire it.
865 */
866 mtx_unlock_spin(&dt_lock);
867 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv,
868 NULL, td->td_proc->p_vmspace);
869 #else
870 set_user_ldt(&td->td_proc->p_md);
871 mtx_unlock_spin(&dt_lock);
872 #endif
873 if (old_ldt_base != NULL_LDT_BASE) {
874 kmem_free(kernel_arena, (vm_offset_t)old_ldt_base,
875 old_ldt_len * sizeof(union descriptor));
876 free(new_ldt, M_SUBPROC);
877 }
878 mtx_lock_spin(&dt_lock);
879 }
880 return (0);
881 }
Cache object: e27b3656fded68bd87e561cef66f2f51
|