1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1990 The Regents of the University of California.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91
32 */
33
34 #include <sys/cdefs.h>
35 #include "opt_capsicum.h"
36 #include "opt_kstack_pages.h"
37 #include "opt_ktrace.h"
38
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/systm.h>
42 #include <sys/ktrace.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mutex.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/smp.h>
49 #include <sys/sysproto.h>
50
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_map.h>
54 #include <vm/vm_extern.h>
55
56 #include <machine/atomic.h>
57 #include <machine/cpu.h>
58 #include <machine/pcb.h>
59 #include <machine/pcb_ext.h>
60 #include <machine/proc.h>
61 #include <machine/sysarch.h>
62
63 #include <security/audit/audit.h>
64
65 #include <vm/vm_kern.h> /* for kernel_map */
66
67 #define MAX_LD 8192
68 #define LD_PER_PAGE 512
69 #define NEW_MAX_LD(num) rounddown2(num + LD_PER_PAGE, LD_PER_PAGE)
70 #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3)
71 #define NULL_LDT_BASE ((caddr_t)NULL)
72
73 #ifdef SMP
74 static void set_user_ldt_rv(void *arg);
75 #endif
76 static int i386_set_ldt_data(struct thread *, int start, int num,
77 union descriptor *descs);
78 static int i386_ldt_grow(struct thread *td, int len);
79
80 void
fill_based_sd(struct segment_descriptor * sdp,uint32_t base)81 fill_based_sd(struct segment_descriptor *sdp, uint32_t base)
82 {
83
84 sdp->sd_lobase = base & 0xffffff;
85 sdp->sd_hibase = (base >> 24) & 0xff;
86 sdp->sd_lolimit = 0xffff; /* 4GB limit, wraps around */
87 sdp->sd_hilimit = 0xf;
88 sdp->sd_type = SDT_MEMRWA;
89 sdp->sd_dpl = SEL_UPL;
90 sdp->sd_p = 1;
91 sdp->sd_xx = 0;
92 sdp->sd_def32 = 1;
93 sdp->sd_gran = 1;
94 }
95
96 /*
97 * Construct special descriptors for "base" selectors. Store them in
98 * the PCB for later use by cpu_switch(). Store them in the GDT for
99 * more immediate use. The GDT entries are part of the current
100 * context. Callers must load related segment registers to complete
101 * setting up the current context.
102 */
103 void
set_fsbase(struct thread * td,uint32_t base)104 set_fsbase(struct thread *td, uint32_t base)
105 {
106 struct segment_descriptor sd;
107
108 fill_based_sd(&sd, base);
109 critical_enter();
110 td->td_pcb->pcb_fsd = sd;
111 if (td == curthread)
112 PCPU_GET(fsgs_gdt)[0] = sd;
113 critical_exit();
114 }
115
116 void
set_gsbase(struct thread * td,uint32_t base)117 set_gsbase(struct thread *td, uint32_t base)
118 {
119 struct segment_descriptor sd;
120
121 fill_based_sd(&sd, base);
122 critical_enter();
123 td->td_pcb->pcb_gsd = sd;
124 if (td == curthread)
125 PCPU_GET(fsgs_gdt)[1] = sd;
126 critical_exit();
127 }
128
129 #ifndef _SYS_SYSPROTO_H_
130 struct sysarch_args {
131 int op;
132 char *parms;
133 };
134 #endif
135
136 int
sysarch(struct thread * td,struct sysarch_args * uap)137 sysarch(struct thread *td, struct sysarch_args *uap)
138 {
139 int error;
140 union descriptor *lp;
141 union {
142 struct i386_ldt_args largs;
143 struct i386_ioperm_args iargs;
144 struct i386_get_xfpustate xfpu;
145 } kargs;
146 uint32_t base;
147 struct segment_descriptor *sdp;
148
149 AUDIT_ARG_CMD(uap->op);
150
151 #ifdef CAPABILITY_MODE
152 /*
153 * When adding new operations, add a new case statement here to
154 * explicitly indicate whether or not the operation is safe to
155 * perform in capability mode.
156 */
157 switch (uap->op) {
158 case I386_GET_LDT:
159 case I386_SET_LDT:
160 case I386_GET_IOPERM:
161 case I386_GET_FSBASE:
162 case I386_SET_FSBASE:
163 case I386_GET_GSBASE:
164 case I386_SET_GSBASE:
165 case I386_GET_XFPUSTATE:
166 break;
167
168 case I386_SET_IOPERM:
169 default:
170 if (CAP_TRACING(td))
171 ktrcapfail(CAPFAIL_SYSCALL, &uap->op);
172 if (IN_CAPABILITY_MODE(td))
173 return (ECAPMODE);
174 break;
175 }
176 #endif
177
178 switch (uap->op) {
179 case I386_GET_IOPERM:
180 case I386_SET_IOPERM:
181 if ((error = copyin(uap->parms, &kargs.iargs,
182 sizeof(struct i386_ioperm_args))) != 0)
183 return (error);
184 break;
185 case I386_GET_LDT:
186 case I386_SET_LDT:
187 if ((error = copyin(uap->parms, &kargs.largs,
188 sizeof(struct i386_ldt_args))) != 0)
189 return (error);
190 break;
191 case I386_GET_XFPUSTATE:
192 if ((error = copyin(uap->parms, &kargs.xfpu,
193 sizeof(struct i386_get_xfpustate))) != 0)
194 return (error);
195 break;
196 default:
197 break;
198 }
199
200 switch (uap->op) {
201 case I386_GET_LDT:
202 error = i386_get_ldt(td, &kargs.largs);
203 break;
204 case I386_SET_LDT:
205 if (kargs.largs.descs != NULL) {
206 if (kargs.largs.num > MAX_LD)
207 return (EINVAL);
208 lp = malloc(kargs.largs.num * sizeof(union descriptor),
209 M_TEMP, M_WAITOK);
210 error = copyin(kargs.largs.descs, lp,
211 kargs.largs.num * sizeof(union descriptor));
212 if (error == 0)
213 error = i386_set_ldt(td, &kargs.largs, lp);
214 free(lp, M_TEMP);
215 } else {
216 error = i386_set_ldt(td, &kargs.largs, NULL);
217 }
218 break;
219 case I386_GET_IOPERM:
220 error = i386_get_ioperm(td, &kargs.iargs);
221 if (error == 0)
222 error = copyout(&kargs.iargs, uap->parms,
223 sizeof(struct i386_ioperm_args));
224 break;
225 case I386_SET_IOPERM:
226 error = i386_set_ioperm(td, &kargs.iargs);
227 break;
228 case I386_VM86:
229 error = vm86_sysarch(td, uap->parms);
230 break;
231 case I386_GET_FSBASE:
232 sdp = &td->td_pcb->pcb_fsd;
233 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
234 error = copyout(&base, uap->parms, sizeof(base));
235 break;
236 case I386_SET_FSBASE:
237 error = copyin(uap->parms, &base, sizeof(base));
238 if (error == 0) {
239 /*
240 * Construct the special descriptor for fsbase
241 * and arrange for doreti to load its selector
242 * soon enough.
243 */
244 set_fsbase(td, base);
245 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
246 }
247 break;
248 case I386_GET_GSBASE:
249 sdp = &td->td_pcb->pcb_gsd;
250 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
251 error = copyout(&base, uap->parms, sizeof(base));
252 break;
253 case I386_SET_GSBASE:
254 error = copyin(uap->parms, &base, sizeof(base));
255 if (error == 0) {
256 /*
257 * Construct the special descriptor for gsbase.
258 * The selector is loaded immediately, since we
259 * normally only reload %gs on context switches.
260 */
261 set_gsbase(td, base);
262 load_gs(GSEL(GUGS_SEL, SEL_UPL));
263 }
264 break;
265 case I386_GET_XFPUSTATE:
266 if (kargs.xfpu.len > cpu_max_ext_state_size -
267 sizeof(union savefpu))
268 return (EINVAL);
269 npxgetregs(td);
270 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
271 kargs.xfpu.addr, kargs.xfpu.len);
272 break;
273 default:
274 error = EINVAL;
275 break;
276 }
277 return (error);
278 }
279
280 int
i386_extend_pcb(struct thread * td)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 = pmap_trm_alloc(ctob(IOPAGES + 1), M_WAITOK | M_ZERO);
298 /* -16 is so we can convert a trapframe into vm86trapframe inplace */
299 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
300 /*
301 * The last byte of the i/o map must be followed by an 0xff byte.
302 * We arbitrarily allocate 16 bytes here, to keep the starting
303 * address on a doubleword boundary.
304 */
305 offset = PAGE_SIZE - 16;
306 ext->ext_tss.tss_ioopt =
307 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16;
308 ext->ext_iomap = (caddr_t)ext + offset;
309 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32;
310
311 addr = (u_long *)ext->ext_vm86.vm86_intmap;
312 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++)
313 *addr++ = ~0;
314
315 ssd.ssd_base = (unsigned)&ext->ext_tss;
316 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext);
317 ssdtosd(&ssd, &ext->ext_tssd);
318
319 KASSERT(td == curthread, ("giving TSS to !curthread"));
320 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!"));
321
322 /* Switch to the new TSS. */
323 critical_enter();
324 ext->ext_tss.tss_esp0 = PCPU_GET(trampstk);
325 td->td_pcb->pcb_ext = ext;
326 PCPU_SET(private_tss, 1);
327 *PCPU_GET(tss_gdt) = ext->ext_tssd;
328 ltr(GSEL(GPROC0_SEL, SEL_KPL));
329 critical_exit();
330
331 return 0;
332 }
333
334 int
i386_set_ioperm(struct thread * td,struct i386_ioperm_args * uap)335 i386_set_ioperm(struct thread *td, struct i386_ioperm_args *uap)
336 {
337 char *iomap;
338 u_int i;
339 int error;
340
341 if ((error = priv_check(td, PRIV_IO)) != 0)
342 return (error);
343 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
344 return (error);
345 /*
346 * XXX
347 * While this is restricted to root, we should probably figure out
348 * whether any other driver is using this i/o address, as so not to
349 * cause confusion. This probably requires a global 'usage registry'.
350 */
351
352 if (td->td_pcb->pcb_ext == 0)
353 if ((error = i386_extend_pcb(td)) != 0)
354 return (error);
355 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
356
357 if (uap->start > uap->start + uap->length ||
358 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
359 return (EINVAL);
360
361 for (i = uap->start; i < uap->start + uap->length; i++) {
362 if (uap->enable)
363 iomap[i >> 3] &= ~(1 << (i & 7));
364 else
365 iomap[i >> 3] |= (1 << (i & 7));
366 }
367 return (error);
368 }
369
370 int
i386_get_ioperm(struct thread * td,struct i386_ioperm_args * uap)371 i386_get_ioperm(struct thread *td, struct i386_ioperm_args *uap)
372 {
373 int i, state;
374 char *iomap;
375
376 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
377 return (EINVAL);
378
379 if (td->td_pcb->pcb_ext == 0) {
380 uap->length = 0;
381 goto done;
382 }
383
384 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
385
386 i = uap->start;
387 state = (iomap[i >> 3] >> (i & 7)) & 1;
388 uap->enable = !state;
389 uap->length = 1;
390
391 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
392 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
393 break;
394 uap->length++;
395 }
396
397 done:
398 return (0);
399 }
400
401 /*
402 * Update the GDT entry pointing to the LDT to point to the LDT of the
403 * current process. Manage dt_lock holding/unholding autonomously.
404 */
405 static void
set_user_ldt_locked(struct mdproc * mdp)406 set_user_ldt_locked(struct mdproc *mdp)
407 {
408 struct proc_ldt *pldt;
409 int gdt_idx;
410
411 mtx_assert(&dt_lock, MA_OWNED);
412
413 pldt = mdp->md_ldt;
414 gdt_idx = GUSERLDT_SEL;
415 gdt_idx += PCPU_GET(cpuid) * NGDT; /* always 0 on UP */
416 gdt[gdt_idx].sd = pldt->ldt_sd;
417 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
418 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL));
419 }
420
421 void
set_user_ldt(struct mdproc * mdp)422 set_user_ldt(struct mdproc *mdp)
423 {
424
425 mtx_lock_spin(&dt_lock);
426 set_user_ldt_locked(mdp);
427 mtx_unlock_spin(&dt_lock);
428 }
429
430 #ifdef SMP
431 static void
set_user_ldt_rv(void * arg)432 set_user_ldt_rv(void *arg)
433 {
434 struct proc *p;
435
436 p = curproc;
437 if (arg == p->p_vmspace)
438 set_user_ldt(&p->p_md);
439 }
440 #endif
441
442 /*
443 * dt_lock must be held. Returns with dt_lock held.
444 */
445 struct proc_ldt *
user_ldt_alloc(struct mdproc * mdp,int len)446 user_ldt_alloc(struct mdproc *mdp, int len)
447 {
448 struct proc_ldt *pldt, *new_ldt;
449
450 mtx_assert(&dt_lock, MA_OWNED);
451 mtx_unlock_spin(&dt_lock);
452 new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
453
454 new_ldt->ldt_len = len = NEW_MAX_LD(len);
455 new_ldt->ldt_base = pmap_trm_alloc(len * sizeof(union descriptor),
456 M_WAITOK | M_ZERO);
457 new_ldt->ldt_refcnt = 1;
458 new_ldt->ldt_active = 0;
459
460 mtx_lock_spin(&dt_lock);
461 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base;
462 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1;
463 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd);
464
465 if ((pldt = mdp->md_ldt) != NULL) {
466 if (len > pldt->ldt_len)
467 len = pldt->ldt_len;
468 bcopy(pldt->ldt_base, new_ldt->ldt_base,
469 len * sizeof(union descriptor));
470 } else
471 bcopy(ldt, new_ldt->ldt_base, sizeof(union descriptor) * NLDT);
472
473 return (new_ldt);
474 }
475
476 /*
477 * Must be called with dt_lock held. Returns with dt_lock unheld.
478 */
479 void
user_ldt_free(struct thread * td)480 user_ldt_free(struct thread *td)
481 {
482 struct mdproc *mdp;
483 struct proc_ldt *pldt;
484
485 mtx_assert(&dt_lock, MA_OWNED);
486 mdp = &td->td_proc->p_md;
487 if ((pldt = mdp->md_ldt) == NULL) {
488 mtx_unlock_spin(&dt_lock);
489 return;
490 }
491
492 if (td == curthread) {
493 lldt(_default_ldt);
494 PCPU_SET(currentldt, _default_ldt);
495 }
496
497 mdp->md_ldt = NULL;
498 user_ldt_deref(pldt);
499 }
500
501 void
user_ldt_deref(struct proc_ldt * pldt)502 user_ldt_deref(struct proc_ldt *pldt)
503 {
504
505 mtx_assert(&dt_lock, MA_OWNED);
506 if (--pldt->ldt_refcnt == 0) {
507 mtx_unlock_spin(&dt_lock);
508 pmap_trm_free(pldt->ldt_base, pldt->ldt_len *
509 sizeof(union descriptor));
510 free(pldt, M_SUBPROC);
511 } else
512 mtx_unlock_spin(&dt_lock);
513 }
514
515 /*
516 * Note for the authors of compat layers (linux, etc): copyout() in
517 * the function below is not a problem since it presents data in
518 * arch-specific format (i.e. i386-specific in this case), not in
519 * the OS-specific one.
520 */
521 int
i386_get_ldt(struct thread * td,struct i386_ldt_args * uap)522 i386_get_ldt(struct thread *td, struct i386_ldt_args *uap)
523 {
524 struct proc_ldt *pldt;
525 char *data;
526 u_int nldt, num;
527 int error;
528
529 #ifdef DEBUG
530 printf("i386_get_ldt: start=%u num=%u descs=%p\n",
531 uap->start, uap->num, (void *)uap->descs);
532 #endif
533
534 num = min(uap->num, MAX_LD);
535 data = malloc(num * sizeof(union descriptor), M_TEMP, M_WAITOK);
536 mtx_lock_spin(&dt_lock);
537 pldt = td->td_proc->p_md.md_ldt;
538 nldt = pldt != NULL ? pldt->ldt_len : NLDT;
539 if (uap->start >= nldt) {
540 num = 0;
541 } else {
542 num = min(num, nldt - uap->start);
543 bcopy(pldt != NULL ?
544 &((union descriptor *)(pldt->ldt_base))[uap->start] :
545 &ldt[uap->start], data, num * sizeof(union descriptor));
546 }
547 mtx_unlock_spin(&dt_lock);
548 error = copyout(data, uap->descs, num * sizeof(union descriptor));
549 if (error == 0)
550 td->td_retval[0] = num;
551 free(data, M_TEMP);
552 return (error);
553 }
554
555 int
i386_set_ldt(struct thread * td,struct i386_ldt_args * uap,union descriptor * descs)556 i386_set_ldt(struct thread *td, struct i386_ldt_args *uap,
557 union descriptor *descs)
558 {
559 struct mdproc *mdp;
560 struct proc_ldt *pldt;
561 union descriptor *dp;
562 u_int largest_ld, i;
563 int error;
564
565 #ifdef DEBUG
566 printf("i386_set_ldt: start=%u num=%u descs=%p\n",
567 uap->start, uap->num, (void *)uap->descs);
568 #endif
569 error = 0;
570 mdp = &td->td_proc->p_md;
571
572 if (descs == NULL) {
573 /* Free descriptors */
574 if (uap->start == 0 && uap->num == 0) {
575 /*
576 * Treat this as a special case, so userland needn't
577 * know magic number NLDT.
578 */
579 uap->start = NLDT;
580 uap->num = MAX_LD - NLDT;
581 }
582 mtx_lock_spin(&dt_lock);
583 if ((pldt = mdp->md_ldt) == NULL ||
584 uap->start >= pldt->ldt_len) {
585 mtx_unlock_spin(&dt_lock);
586 return (0);
587 }
588 largest_ld = uap->start + uap->num;
589 if (largest_ld > pldt->ldt_len)
590 largest_ld = pldt->ldt_len;
591 for (i = uap->start; i < largest_ld; i++)
592 atomic_store_rel_64(&((uint64_t *)(pldt->ldt_base))[i],
593 0);
594 mtx_unlock_spin(&dt_lock);
595 return (0);
596 }
597
598 if (uap->start != LDT_AUTO_ALLOC || uap->num != 1) {
599 /* verify range of descriptors to modify */
600 largest_ld = uap->start + uap->num;
601 if (uap->start >= MAX_LD || largest_ld > MAX_LD)
602 return (EINVAL);
603 }
604
605 /* Check descriptors for access violations */
606 for (i = 0; i < uap->num; i++) {
607 dp = &descs[i];
608
609 switch (dp->sd.sd_type) {
610 case SDT_SYSNULL: /* system null */
611 dp->sd.sd_p = 0;
612 break;
613 case SDT_SYS286TSS: /* system 286 TSS available */
614 case SDT_SYSLDT: /* system local descriptor table */
615 case SDT_SYS286BSY: /* system 286 TSS busy */
616 case SDT_SYSTASKGT: /* system task gate */
617 case SDT_SYS286IGT: /* system 286 interrupt gate */
618 case SDT_SYS286TGT: /* system 286 trap gate */
619 case SDT_SYSNULL2: /* undefined by Intel */
620 case SDT_SYS386TSS: /* system 386 TSS available */
621 case SDT_SYSNULL3: /* undefined by Intel */
622 case SDT_SYS386BSY: /* system 386 TSS busy */
623 case SDT_SYSNULL4: /* undefined by Intel */
624 case SDT_SYS386IGT: /* system 386 interrupt gate */
625 case SDT_SYS386TGT: /* system 386 trap gate */
626 case SDT_SYS286CGT: /* system 286 call gate */
627 case SDT_SYS386CGT: /* system 386 call gate */
628 return (EACCES);
629
630 /* memory segment types */
631 case SDT_MEMEC: /* memory execute only conforming */
632 case SDT_MEMEAC: /* memory execute only accessed conforming */
633 case SDT_MEMERC: /* memory execute read conforming */
634 case SDT_MEMERAC: /* memory execute read accessed conforming */
635 /* Must be "present" if executable and conforming. */
636 if (dp->sd.sd_p == 0)
637 return (EACCES);
638 break;
639 case SDT_MEMRO: /* memory read only */
640 case SDT_MEMROA: /* memory read only accessed */
641 case SDT_MEMRW: /* memory read write */
642 case SDT_MEMRWA: /* memory read write accessed */
643 case SDT_MEMROD: /* memory read only expand dwn limit */
644 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
645 case SDT_MEMRWD: /* memory read write expand dwn limit */
646 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
647 case SDT_MEME: /* memory execute only */
648 case SDT_MEMEA: /* memory execute only accessed */
649 case SDT_MEMER: /* memory execute read */
650 case SDT_MEMERA: /* memory execute read accessed */
651 break;
652 default:
653 return (EINVAL);
654 }
655
656 /* Only user (ring-3) descriptors may be present. */
657 if (dp->sd.sd_p != 0 && dp->sd.sd_dpl != SEL_UPL)
658 return (EACCES);
659 }
660
661 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
662 /* Allocate a free slot */
663 mtx_lock_spin(&dt_lock);
664 if ((pldt = mdp->md_ldt) == NULL) {
665 if ((error = i386_ldt_grow(td, NLDT + 1))) {
666 mtx_unlock_spin(&dt_lock);
667 return (error);
668 }
669 pldt = mdp->md_ldt;
670 }
671 again:
672 /*
673 * start scanning a bit up to leave room for NVidia and
674 * Wine, which still user the "Blat" method of allocation.
675 */
676 dp = &((union descriptor *)(pldt->ldt_base))[NLDT];
677 for (i = NLDT; i < pldt->ldt_len; ++i) {
678 if (dp->sd.sd_type == SDT_SYSNULL)
679 break;
680 dp++;
681 }
682 if (i >= pldt->ldt_len) {
683 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) {
684 mtx_unlock_spin(&dt_lock);
685 return (error);
686 }
687 goto again;
688 }
689 uap->start = i;
690 error = i386_set_ldt_data(td, i, 1, descs);
691 mtx_unlock_spin(&dt_lock);
692 } else {
693 largest_ld = uap->start + uap->num;
694 mtx_lock_spin(&dt_lock);
695 if (!(error = i386_ldt_grow(td, largest_ld))) {
696 error = i386_set_ldt_data(td, uap->start, uap->num,
697 descs);
698 }
699 mtx_unlock_spin(&dt_lock);
700 }
701 if (error == 0)
702 td->td_retval[0] = uap->start;
703 return (error);
704 }
705
706 static int
i386_set_ldt_data(struct thread * td,int start,int num,union descriptor * descs)707 i386_set_ldt_data(struct thread *td, int start, int num,
708 union descriptor *descs)
709 {
710 struct mdproc *mdp;
711 struct proc_ldt *pldt;
712 uint64_t *dst, *src;
713 int i;
714
715 mtx_assert(&dt_lock, MA_OWNED);
716
717 mdp = &td->td_proc->p_md;
718 pldt = mdp->md_ldt;
719 dst = (uint64_t *)(pldt->ldt_base);
720 src = (uint64_t *)descs;
721
722 /*
723 * Atomic(9) is used only to get 64bit atomic store with
724 * cmpxchg8b when available. There is no op without release
725 * semantic.
726 */
727 for (i = 0; i < num; i++)
728 atomic_store_rel_64(&dst[start + i], src[i]);
729 return (0);
730 }
731
732 static int
i386_ldt_grow(struct thread * td,int len)733 i386_ldt_grow(struct thread *td, int len)
734 {
735 struct mdproc *mdp;
736 struct proc_ldt *new_ldt, *pldt;
737 caddr_t old_ldt_base;
738 int old_ldt_len;
739
740 mtx_assert(&dt_lock, MA_OWNED);
741
742 if (len > MAX_LD)
743 return (ENOMEM);
744 if (len < NLDT + 1)
745 len = NLDT + 1;
746
747 mdp = &td->td_proc->p_md;
748 old_ldt_base = NULL_LDT_BASE;
749 old_ldt_len = 0;
750
751 /* Allocate a user ldt. */
752 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) {
753 new_ldt = user_ldt_alloc(mdp, len);
754 if (new_ldt == NULL)
755 return (ENOMEM);
756 pldt = mdp->md_ldt;
757
758 if (pldt != NULL) {
759 if (new_ldt->ldt_len <= pldt->ldt_len) {
760 /*
761 * We just lost the race for allocation, so
762 * free the new object and return.
763 */
764 mtx_unlock_spin(&dt_lock);
765 pmap_trm_free(new_ldt->ldt_base,
766 new_ldt->ldt_len * sizeof(union descriptor));
767 free(new_ldt, M_SUBPROC);
768 mtx_lock_spin(&dt_lock);
769 return (0);
770 }
771
772 /*
773 * We have to substitute the current LDT entry for
774 * curproc with the new one since its size grew.
775 */
776 old_ldt_base = pldt->ldt_base;
777 old_ldt_len = pldt->ldt_len;
778 pldt->ldt_sd = new_ldt->ldt_sd;
779 pldt->ldt_base = new_ldt->ldt_base;
780 pldt->ldt_len = new_ldt->ldt_len;
781 } else
782 mdp->md_ldt = pldt = new_ldt;
783 #ifdef SMP
784 /*
785 * Signal other cpus to reload ldt. We need to unlock dt_lock
786 * here because other CPU will contest on it since their
787 * curthreads won't hold the lock and will block when trying
788 * to acquire it.
789 */
790 mtx_unlock_spin(&dt_lock);
791 smp_rendezvous(NULL, set_user_ldt_rv, NULL,
792 td->td_proc->p_vmspace);
793 #else
794 set_user_ldt_locked(&td->td_proc->p_md);
795 mtx_unlock_spin(&dt_lock);
796 #endif
797 if (old_ldt_base != NULL_LDT_BASE) {
798 pmap_trm_free(old_ldt_base, old_ldt_len *
799 sizeof(union descriptor));
800 free(new_ldt, M_SUBPROC);
801 }
802 mtx_lock_spin(&dt_lock);
803 }
804 return (0);
805 }
806