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
32 #include <sys/cdefs.h>
33 #include "opt_capsicum.h"
34 #include "opt_kstack_pages.h"
35 #include "opt_ktrace.h"
36
37 #include <sys/param.h>
38 #include <sys/capsicum.h>
39 #include <sys/systm.h>
40 #include <sys/ktrace.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/atomic.h>
55 #include <machine/cpu.h>
56 #include <machine/pcb.h>
57 #include <machine/pcb_ext.h>
58 #include <machine/proc.h>
59 #include <machine/sysarch.h>
60
61 #include <security/audit/audit.h>
62
63 #include <vm/vm_kern.h> /* for kernel_map */
64
65 #define MAX_LD 8192
66 #define LD_PER_PAGE 512
67 #define NEW_MAX_LD(num) rounddown2(num + LD_PER_PAGE, LD_PER_PAGE)
68 #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3)
69 #define NULL_LDT_BASE ((caddr_t)NULL)
70
71 #ifdef SMP
72 static void set_user_ldt_rv(void *arg);
73 #endif
74 static int i386_set_ldt_data(struct thread *, int start, int num,
75 union descriptor *descs);
76 static int i386_ldt_grow(struct thread *td, int len);
77
78 void
fill_based_sd(struct segment_descriptor * sdp,uint32_t base)79 fill_based_sd(struct segment_descriptor *sdp, uint32_t base)
80 {
81
82 sdp->sd_lobase = base & 0xffffff;
83 sdp->sd_hibase = (base >> 24) & 0xff;
84 sdp->sd_lolimit = 0xffff; /* 4GB limit, wraps around */
85 sdp->sd_hilimit = 0xf;
86 sdp->sd_type = SDT_MEMRWA;
87 sdp->sd_dpl = SEL_UPL;
88 sdp->sd_p = 1;
89 sdp->sd_xx = 0;
90 sdp->sd_def32 = 1;
91 sdp->sd_gran = 1;
92 }
93
94 /*
95 * Construct special descriptors for "base" selectors. Store them in
96 * the PCB for later use by cpu_switch(). Store them in the GDT for
97 * more immediate use. The GDT entries are part of the current
98 * context. Callers must load related segment registers to complete
99 * setting up the current context.
100 */
101 void
set_fsbase(struct thread * td,uint32_t base)102 set_fsbase(struct thread *td, uint32_t base)
103 {
104 struct segment_descriptor sd;
105
106 fill_based_sd(&sd, base);
107 critical_enter();
108 td->td_pcb->pcb_fsd = sd;
109 if (td == curthread)
110 PCPU_GET(fsgs_gdt)[0] = sd;
111 critical_exit();
112 }
113
114 void
set_gsbase(struct thread * td,uint32_t base)115 set_gsbase(struct thread *td, uint32_t base)
116 {
117 struct segment_descriptor sd;
118
119 fill_based_sd(&sd, base);
120 critical_enter();
121 td->td_pcb->pcb_gsd = sd;
122 if (td == curthread)
123 PCPU_GET(fsgs_gdt)[1] = sd;
124 critical_exit();
125 }
126
127 #ifndef _SYS_SYSPROTO_H_
128 struct sysarch_args {
129 int op;
130 char *parms;
131 };
132 #endif
133
134 int
sysarch(struct thread * td,struct sysarch_args * uap)135 sysarch(struct thread *td, struct sysarch_args *uap)
136 {
137 int error;
138 union descriptor *lp;
139 union {
140 struct i386_ldt_args largs;
141 struct i386_ioperm_args iargs;
142 struct i386_get_xfpustate xfpu;
143 } kargs;
144 uint32_t base;
145 struct segment_descriptor *sdp;
146
147 AUDIT_ARG_CMD(uap->op);
148
149 #ifdef CAPABILITY_MODE
150 /*
151 * When adding new operations, add a new case statement here to
152 * explicitly indicate whether or not the operation is safe to
153 * perform in capability mode.
154 */
155 switch (uap->op) {
156 case I386_GET_LDT:
157 case I386_SET_LDT:
158 case I386_GET_IOPERM:
159 case I386_GET_FSBASE:
160 case I386_SET_FSBASE:
161 case I386_GET_GSBASE:
162 case I386_SET_GSBASE:
163 case I386_GET_XFPUSTATE:
164 break;
165
166 case I386_SET_IOPERM:
167 default:
168 if (CAP_TRACING(td))
169 ktrcapfail(CAPFAIL_SYSCALL, &uap->op);
170 if (IN_CAPABILITY_MODE(td))
171 return (ECAPMODE);
172 break;
173 }
174 #endif
175
176 switch (uap->op) {
177 case I386_GET_IOPERM:
178 case I386_SET_IOPERM:
179 if ((error = copyin(uap->parms, &kargs.iargs,
180 sizeof(struct i386_ioperm_args))) != 0)
181 return (error);
182 break;
183 case I386_GET_LDT:
184 case I386_SET_LDT:
185 if ((error = copyin(uap->parms, &kargs.largs,
186 sizeof(struct i386_ldt_args))) != 0)
187 return (error);
188 break;
189 case I386_GET_XFPUSTATE:
190 if ((error = copyin(uap->parms, &kargs.xfpu,
191 sizeof(struct i386_get_xfpustate))) != 0)
192 return (error);
193 break;
194 default:
195 break;
196 }
197
198 switch (uap->op) {
199 case I386_GET_LDT:
200 error = i386_get_ldt(td, &kargs.largs);
201 break;
202 case I386_SET_LDT:
203 if (kargs.largs.descs != NULL) {
204 if (kargs.largs.num > MAX_LD)
205 return (EINVAL);
206 lp = malloc(kargs.largs.num * sizeof(union descriptor),
207 M_TEMP, M_WAITOK);
208 error = copyin(kargs.largs.descs, lp,
209 kargs.largs.num * sizeof(union descriptor));
210 if (error == 0)
211 error = i386_set_ldt(td, &kargs.largs, lp);
212 free(lp, M_TEMP);
213 } else {
214 error = i386_set_ldt(td, &kargs.largs, NULL);
215 }
216 break;
217 case I386_GET_IOPERM:
218 error = i386_get_ioperm(td, &kargs.iargs);
219 if (error == 0)
220 error = copyout(&kargs.iargs, uap->parms,
221 sizeof(struct i386_ioperm_args));
222 break;
223 case I386_SET_IOPERM:
224 error = i386_set_ioperm(td, &kargs.iargs);
225 break;
226 case I386_VM86:
227 error = vm86_sysarch(td, uap->parms);
228 break;
229 case I386_GET_FSBASE:
230 sdp = &td->td_pcb->pcb_fsd;
231 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
232 error = copyout(&base, uap->parms, sizeof(base));
233 break;
234 case I386_SET_FSBASE:
235 error = copyin(uap->parms, &base, sizeof(base));
236 if (error == 0) {
237 /*
238 * Construct the special descriptor for fsbase
239 * and arrange for doreti to load its selector
240 * soon enough.
241 */
242 set_fsbase(td, base);
243 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
244 }
245 break;
246 case I386_GET_GSBASE:
247 sdp = &td->td_pcb->pcb_gsd;
248 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
249 error = copyout(&base, uap->parms, sizeof(base));
250 break;
251 case I386_SET_GSBASE:
252 error = copyin(uap->parms, &base, sizeof(base));
253 if (error == 0) {
254 /*
255 * Construct the special descriptor for gsbase.
256 * The selector is loaded immediately, since we
257 * normally only reload %gs on context switches.
258 */
259 set_gsbase(td, base);
260 load_gs(GSEL(GUGS_SEL, SEL_UPL));
261 }
262 break;
263 case I386_GET_XFPUSTATE:
264 if (kargs.xfpu.len > cpu_max_ext_state_size -
265 sizeof(union savefpu))
266 return (EINVAL);
267 npxgetregs(td);
268 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
269 kargs.xfpu.addr, kargs.xfpu.len);
270 break;
271 default:
272 error = EINVAL;
273 break;
274 }
275 return (error);
276 }
277
278 int
i386_extend_pcb(struct thread * td)279 i386_extend_pcb(struct thread *td)
280 {
281 int i, offset;
282 u_long *addr;
283 struct pcb_ext *ext;
284 struct soft_segment_descriptor ssd = {
285 0, /* segment base address (overwritten) */
286 ctob(IOPAGES + 1) - 1, /* length */
287 SDT_SYS386TSS, /* segment type */
288 0, /* priority level */
289 1, /* descriptor present */
290 0, 0,
291 0, /* default 32 size */
292 0 /* granularity */
293 };
294
295 ext = pmap_trm_alloc(ctob(IOPAGES + 1), M_WAITOK | M_ZERO);
296 /* -16 is so we can convert a trapframe into vm86trapframe inplace */
297 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
298 /*
299 * The last byte of the i/o map must be followed by an 0xff byte.
300 * We arbitrarily allocate 16 bytes here, to keep the starting
301 * address on a doubleword boundary.
302 */
303 offset = PAGE_SIZE - 16;
304 ext->ext_tss.tss_ioopt =
305 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16;
306 ext->ext_iomap = (caddr_t)ext + offset;
307 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32;
308
309 addr = (u_long *)ext->ext_vm86.vm86_intmap;
310 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++)
311 *addr++ = ~0;
312
313 ssd.ssd_base = (unsigned)&ext->ext_tss;
314 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext);
315 ssdtosd(&ssd, &ext->ext_tssd);
316
317 KASSERT(td == curthread, ("giving TSS to !curthread"));
318 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!"));
319
320 /* Switch to the new TSS. */
321 critical_enter();
322 ext->ext_tss.tss_esp0 = PCPU_GET(trampstk);
323 td->td_pcb->pcb_ext = ext;
324 PCPU_SET(private_tss, 1);
325 *PCPU_GET(tss_gdt) = ext->ext_tssd;
326 ltr(GSEL(GPROC0_SEL, SEL_KPL));
327 critical_exit();
328
329 return 0;
330 }
331
332 int
i386_set_ioperm(struct thread * td,struct i386_ioperm_args * uap)333 i386_set_ioperm(struct thread *td, struct i386_ioperm_args *uap)
334 {
335 char *iomap;
336 u_int i;
337 int error;
338
339 if ((error = priv_check(td, PRIV_IO)) != 0)
340 return (error);
341 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
342 return (error);
343 /*
344 * XXX
345 * While this is restricted to root, we should probably figure out
346 * whether any other driver is using this i/o address, as so not to
347 * cause confusion. This probably requires a global 'usage registry'.
348 */
349
350 if (td->td_pcb->pcb_ext == 0)
351 if ((error = i386_extend_pcb(td)) != 0)
352 return (error);
353 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
354
355 if (uap->start > uap->start + uap->length ||
356 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
357 return (EINVAL);
358
359 for (i = uap->start; i < uap->start + uap->length; i++) {
360 if (uap->enable)
361 iomap[i >> 3] &= ~(1 << (i & 7));
362 else
363 iomap[i >> 3] |= (1 << (i & 7));
364 }
365 return (error);
366 }
367
368 int
i386_get_ioperm(struct thread * td,struct i386_ioperm_args * uap)369 i386_get_ioperm(struct thread *td, struct i386_ioperm_args *uap)
370 {
371 int i, state;
372 char *iomap;
373
374 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
375 return (EINVAL);
376
377 if (td->td_pcb->pcb_ext == 0) {
378 uap->length = 0;
379 goto done;
380 }
381
382 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
383
384 i = uap->start;
385 state = (iomap[i >> 3] >> (i & 7)) & 1;
386 uap->enable = !state;
387 uap->length = 1;
388
389 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
390 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
391 break;
392 uap->length++;
393 }
394
395 done:
396 return (0);
397 }
398
399 /*
400 * Update the GDT entry pointing to the LDT to point to the LDT of the
401 * current process. Manage dt_lock holding/unholding autonomously.
402 */
403 static void
set_user_ldt_locked(struct mdproc * mdp)404 set_user_ldt_locked(struct mdproc *mdp)
405 {
406 struct proc_ldt *pldt;
407 int gdt_idx;
408
409 mtx_assert(&dt_lock, MA_OWNED);
410
411 pldt = mdp->md_ldt;
412 gdt_idx = GUSERLDT_SEL;
413 gdt_idx += PCPU_GET(cpuid) * NGDT; /* always 0 on UP */
414 gdt[gdt_idx].sd = pldt->ldt_sd;
415 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
416 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL));
417 }
418
419 void
set_user_ldt(struct mdproc * mdp)420 set_user_ldt(struct mdproc *mdp)
421 {
422
423 mtx_lock_spin(&dt_lock);
424 set_user_ldt_locked(mdp);
425 mtx_unlock_spin(&dt_lock);
426 }
427
428 #ifdef SMP
429 static void
set_user_ldt_rv(void * arg)430 set_user_ldt_rv(void *arg)
431 {
432 struct proc *p;
433
434 p = curproc;
435 if (arg == p->p_vmspace)
436 set_user_ldt(&p->p_md);
437 }
438 #endif
439
440 /*
441 * dt_lock must be held. Returns with dt_lock held.
442 */
443 struct proc_ldt *
user_ldt_alloc(struct mdproc * mdp,int len)444 user_ldt_alloc(struct mdproc *mdp, int len)
445 {
446 struct proc_ldt *pldt, *new_ldt;
447
448 mtx_assert(&dt_lock, MA_OWNED);
449 mtx_unlock_spin(&dt_lock);
450 new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
451
452 new_ldt->ldt_len = len = NEW_MAX_LD(len);
453 new_ldt->ldt_base = pmap_trm_alloc(len * sizeof(union descriptor),
454 M_WAITOK | M_ZERO);
455 new_ldt->ldt_refcnt = 1;
456 new_ldt->ldt_active = 0;
457
458 mtx_lock_spin(&dt_lock);
459 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base;
460 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1;
461 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd);
462
463 if ((pldt = mdp->md_ldt) != NULL) {
464 if (len > pldt->ldt_len)
465 len = pldt->ldt_len;
466 bcopy(pldt->ldt_base, new_ldt->ldt_base,
467 len * sizeof(union descriptor));
468 } else
469 bcopy(ldt, new_ldt->ldt_base, sizeof(union descriptor) * NLDT);
470
471 return (new_ldt);
472 }
473
474 /*
475 * Must be called with dt_lock held. Returns with dt_lock unheld.
476 */
477 void
user_ldt_free(struct thread * td)478 user_ldt_free(struct thread *td)
479 {
480 struct mdproc *mdp;
481 struct proc_ldt *pldt;
482
483 mtx_assert(&dt_lock, MA_OWNED);
484 mdp = &td->td_proc->p_md;
485 if ((pldt = mdp->md_ldt) == NULL) {
486 mtx_unlock_spin(&dt_lock);
487 return;
488 }
489
490 if (td == curthread) {
491 lldt(_default_ldt);
492 PCPU_SET(currentldt, _default_ldt);
493 }
494
495 mdp->md_ldt = NULL;
496 user_ldt_deref(pldt);
497 }
498
499 void
user_ldt_deref(struct proc_ldt * pldt)500 user_ldt_deref(struct proc_ldt *pldt)
501 {
502
503 mtx_assert(&dt_lock, MA_OWNED);
504 if (--pldt->ldt_refcnt == 0) {
505 mtx_unlock_spin(&dt_lock);
506 pmap_trm_free(pldt->ldt_base, pldt->ldt_len *
507 sizeof(union descriptor));
508 free(pldt, M_SUBPROC);
509 } else
510 mtx_unlock_spin(&dt_lock);
511 }
512
513 /*
514 * Note for the authors of compat layers (linux, etc): copyout() in
515 * the function below is not a problem since it presents data in
516 * arch-specific format (i.e. i386-specific in this case), not in
517 * the OS-specific one.
518 */
519 int
i386_get_ldt(struct thread * td,struct i386_ldt_args * uap)520 i386_get_ldt(struct thread *td, struct i386_ldt_args *uap)
521 {
522 struct proc_ldt *pldt;
523 char *data;
524 u_int nldt, num;
525 int error;
526
527 #ifdef DEBUG
528 printf("i386_get_ldt: start=%u num=%u descs=%p\n",
529 uap->start, uap->num, (void *)uap->descs);
530 #endif
531
532 num = min(uap->num, MAX_LD);
533 data = malloc(num * sizeof(union descriptor), M_TEMP, M_WAITOK);
534 mtx_lock_spin(&dt_lock);
535 pldt = td->td_proc->p_md.md_ldt;
536 nldt = pldt != NULL ? pldt->ldt_len : NLDT;
537 if (uap->start >= nldt) {
538 num = 0;
539 } else {
540 num = min(num, nldt - uap->start);
541 bcopy(pldt != NULL ?
542 &((union descriptor *)(pldt->ldt_base))[uap->start] :
543 &ldt[uap->start], data, num * sizeof(union descriptor));
544 }
545 mtx_unlock_spin(&dt_lock);
546 error = copyout(data, uap->descs, num * sizeof(union descriptor));
547 if (error == 0)
548 td->td_retval[0] = num;
549 free(data, M_TEMP);
550 return (error);
551 }
552
553 int
i386_set_ldt(struct thread * td,struct i386_ldt_args * uap,union descriptor * descs)554 i386_set_ldt(struct thread *td, struct i386_ldt_args *uap,
555 union descriptor *descs)
556 {
557 struct mdproc *mdp;
558 struct proc_ldt *pldt;
559 union descriptor *dp;
560 u_int largest_ld, i;
561 int error;
562
563 #ifdef DEBUG
564 printf("i386_set_ldt: start=%u num=%u descs=%p\n",
565 uap->start, uap->num, (void *)uap->descs);
566 #endif
567 error = 0;
568 mdp = &td->td_proc->p_md;
569
570 if (descs == NULL) {
571 /* Free descriptors */
572 if (uap->start == 0 && uap->num == 0) {
573 /*
574 * Treat this as a special case, so userland needn't
575 * know magic number NLDT.
576 */
577 uap->start = NLDT;
578 uap->num = MAX_LD - NLDT;
579 }
580 mtx_lock_spin(&dt_lock);
581 if ((pldt = mdp->md_ldt) == NULL ||
582 uap->start >= pldt->ldt_len) {
583 mtx_unlock_spin(&dt_lock);
584 return (0);
585 }
586 largest_ld = uap->start + uap->num;
587 if (largest_ld > pldt->ldt_len)
588 largest_ld = pldt->ldt_len;
589 for (i = uap->start; i < largest_ld; i++)
590 atomic_store_rel_64(&((uint64_t *)(pldt->ldt_base))[i],
591 0);
592 mtx_unlock_spin(&dt_lock);
593 return (0);
594 }
595
596 if (uap->start != LDT_AUTO_ALLOC || uap->num != 1) {
597 /* verify range of descriptors to modify */
598 largest_ld = uap->start + uap->num;
599 if (uap->start >= MAX_LD || largest_ld > MAX_LD)
600 return (EINVAL);
601 }
602
603 /* Check descriptors for access violations */
604 for (i = 0; i < uap->num; i++) {
605 dp = &descs[i];
606
607 switch (dp->sd.sd_type) {
608 case SDT_SYSNULL: /* system null */
609 dp->sd.sd_p = 0;
610 break;
611 case SDT_SYS286TSS: /* system 286 TSS available */
612 case SDT_SYSLDT: /* system local descriptor table */
613 case SDT_SYS286BSY: /* system 286 TSS busy */
614 case SDT_SYSTASKGT: /* system task gate */
615 case SDT_SYS286IGT: /* system 286 interrupt gate */
616 case SDT_SYS286TGT: /* system 286 trap gate */
617 case SDT_SYSNULL2: /* undefined by Intel */
618 case SDT_SYS386TSS: /* system 386 TSS available */
619 case SDT_SYSNULL3: /* undefined by Intel */
620 case SDT_SYS386BSY: /* system 386 TSS busy */
621 case SDT_SYSNULL4: /* undefined by Intel */
622 case SDT_SYS386IGT: /* system 386 interrupt gate */
623 case SDT_SYS386TGT: /* system 386 trap gate */
624 case SDT_SYS286CGT: /* system 286 call gate */
625 case SDT_SYS386CGT: /* system 386 call gate */
626 return (EACCES);
627
628 /* memory segment types */
629 case SDT_MEMEC: /* memory execute only conforming */
630 case SDT_MEMEAC: /* memory execute only accessed conforming */
631 case SDT_MEMERC: /* memory execute read conforming */
632 case SDT_MEMERAC: /* memory execute read accessed conforming */
633 /* Must be "present" if executable and conforming. */
634 if (dp->sd.sd_p == 0)
635 return (EACCES);
636 break;
637 case SDT_MEMRO: /* memory read only */
638 case SDT_MEMROA: /* memory read only accessed */
639 case SDT_MEMRW: /* memory read write */
640 case SDT_MEMRWA: /* memory read write accessed */
641 case SDT_MEMROD: /* memory read only expand dwn limit */
642 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
643 case SDT_MEMRWD: /* memory read write expand dwn limit */
644 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
645 case SDT_MEME: /* memory execute only */
646 case SDT_MEMEA: /* memory execute only accessed */
647 case SDT_MEMER: /* memory execute read */
648 case SDT_MEMERA: /* memory execute read accessed */
649 break;
650 default:
651 return (EINVAL);
652 }
653
654 /* Only user (ring-3) descriptors may be present. */
655 if (dp->sd.sd_p != 0 && dp->sd.sd_dpl != SEL_UPL)
656 return (EACCES);
657 }
658
659 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
660 /* Allocate a free slot */
661 mtx_lock_spin(&dt_lock);
662 if ((pldt = mdp->md_ldt) == NULL) {
663 if ((error = i386_ldt_grow(td, NLDT + 1))) {
664 mtx_unlock_spin(&dt_lock);
665 return (error);
666 }
667 pldt = mdp->md_ldt;
668 }
669 again:
670 /*
671 * start scanning a bit up to leave room for NVidia and
672 * Wine, which still user the "Blat" method of allocation.
673 */
674 dp = &((union descriptor *)(pldt->ldt_base))[NLDT];
675 for (i = NLDT; i < pldt->ldt_len; ++i) {
676 if (dp->sd.sd_type == SDT_SYSNULL)
677 break;
678 dp++;
679 }
680 if (i >= pldt->ldt_len) {
681 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) {
682 mtx_unlock_spin(&dt_lock);
683 return (error);
684 }
685 goto again;
686 }
687 uap->start = i;
688 error = i386_set_ldt_data(td, i, 1, descs);
689 mtx_unlock_spin(&dt_lock);
690 } else {
691 largest_ld = uap->start + uap->num;
692 mtx_lock_spin(&dt_lock);
693 if (!(error = i386_ldt_grow(td, largest_ld))) {
694 error = i386_set_ldt_data(td, uap->start, uap->num,
695 descs);
696 }
697 mtx_unlock_spin(&dt_lock);
698 }
699 if (error == 0)
700 td->td_retval[0] = uap->start;
701 return (error);
702 }
703
704 static int
i386_set_ldt_data(struct thread * td,int start,int num,union descriptor * descs)705 i386_set_ldt_data(struct thread *td, int start, int num,
706 union descriptor *descs)
707 {
708 struct mdproc *mdp;
709 struct proc_ldt *pldt;
710 uint64_t *dst, *src;
711 int i;
712
713 mtx_assert(&dt_lock, MA_OWNED);
714
715 mdp = &td->td_proc->p_md;
716 pldt = mdp->md_ldt;
717 dst = (uint64_t *)(pldt->ldt_base);
718 src = (uint64_t *)descs;
719
720 /*
721 * Atomic(9) is used only to get 64bit atomic store with
722 * cmpxchg8b when available. There is no op without release
723 * semantic.
724 */
725 for (i = 0; i < num; i++)
726 atomic_store_rel_64(&dst[start + i], src[i]);
727 return (0);
728 }
729
730 static int
i386_ldt_grow(struct thread * td,int len)731 i386_ldt_grow(struct thread *td, int len)
732 {
733 struct mdproc *mdp;
734 struct proc_ldt *new_ldt, *pldt;
735 caddr_t old_ldt_base;
736 int old_ldt_len;
737
738 mtx_assert(&dt_lock, MA_OWNED);
739
740 if (len > MAX_LD)
741 return (ENOMEM);
742 if (len < NLDT + 1)
743 len = NLDT + 1;
744
745 mdp = &td->td_proc->p_md;
746 old_ldt_base = NULL_LDT_BASE;
747 old_ldt_len = 0;
748
749 /* Allocate a user ldt. */
750 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) {
751 new_ldt = user_ldt_alloc(mdp, len);
752 if (new_ldt == NULL)
753 return (ENOMEM);
754 pldt = mdp->md_ldt;
755
756 if (pldt != NULL) {
757 if (new_ldt->ldt_len <= pldt->ldt_len) {
758 /*
759 * We just lost the race for allocation, so
760 * free the new object and return.
761 */
762 mtx_unlock_spin(&dt_lock);
763 pmap_trm_free(new_ldt->ldt_base,
764 new_ldt->ldt_len * sizeof(union descriptor));
765 free(new_ldt, M_SUBPROC);
766 mtx_lock_spin(&dt_lock);
767 return (0);
768 }
769
770 /*
771 * We have to substitute the current LDT entry for
772 * curproc with the new one since its size grew.
773 */
774 old_ldt_base = pldt->ldt_base;
775 old_ldt_len = pldt->ldt_len;
776 pldt->ldt_sd = new_ldt->ldt_sd;
777 pldt->ldt_base = new_ldt->ldt_base;
778 pldt->ldt_len = new_ldt->ldt_len;
779 } else
780 mdp->md_ldt = pldt = new_ldt;
781 #ifdef SMP
782 /*
783 * Signal other cpus to reload ldt. We need to unlock dt_lock
784 * here because other CPU will contest on it since their
785 * curthreads won't hold the lock and will block when trying
786 * to acquire it.
787 */
788 mtx_unlock_spin(&dt_lock);
789 smp_rendezvous(NULL, set_user_ldt_rv, NULL,
790 td->td_proc->p_vmspace);
791 #else
792 set_user_ldt_locked(&td->td_proc->p_md);
793 mtx_unlock_spin(&dt_lock);
794 #endif
795 if (old_ldt_base != NULL_LDT_BASE) {
796 pmap_trm_free(old_ldt_base, old_ldt_len *
797 sizeof(union descriptor));
798 free(new_ldt, M_SUBPROC);
799 }
800 mtx_lock_spin(&dt_lock);
801 }
802 return (0);
803 }
804