2 * Copyright (c) 2003 Peter Wemm.
3 * Copyright (c) 1990 The Regents of the University of California.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. Neither the name of the University nor the names of its contributors
15 * may be used to endorse or promote products derived from this software
16 * without specific prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include "opt_capsicum.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/capsicum.h>
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
47 #include <sys/sysproto.h>
52 #include <vm/vm_kern.h> /* for kernel_map */
53 #include <vm/vm_extern.h>
55 #include <machine/frame.h>
56 #include <machine/md_var.h>
57 #include <machine/pcb.h>
58 #include <machine/specialreg.h>
59 #include <machine/sysarch.h>
60 #include <machine/tss.h>
61 #include <machine/vmparam.h>
63 #include <security/audit/audit.h>
67 int max_ldt_segment = 1024;
68 SYSCTL_INT(_machdep, OID_AUTO, max_ldt_segment, CTLFLAG_RDTUN,
70 "Maximum number of allowed LDT segments in the single address space");
73 max_ldt_segment_init(void *arg __unused)
76 if (max_ldt_segment <= 0)
78 if (max_ldt_segment > MAX_LD)
79 max_ldt_segment = MAX_LD;
81 SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL);
85 static void set_user_ldt_rv(struct vmspace *vmsp);
88 static void user_ldt_derefl(struct proc_ldt *pldt);
90 #ifndef _SYS_SYSPROTO_H_
98 sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space)
100 struct i386_ldt_args *largs, la;
101 struct user_segment_descriptor *lp;
105 * XXXKIB check that the BSM generation code knows to encode
108 AUDIT_ARG_CMD(uap->op);
109 if (uap_space == UIO_USERSPACE) {
110 error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args));
115 largs = (struct i386_ldt_args *)uap->parms;
119 error = amd64_get_ldt(td, largs);
122 if (largs->descs != NULL && largs->num > max_ldt_segment)
124 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
125 if (largs->descs != NULL) {
126 lp = malloc(largs->num * sizeof(struct
127 user_segment_descriptor), M_TEMP, M_WAITOK);
128 error = copyin(largs->descs, lp, largs->num *
129 sizeof(struct user_segment_descriptor));
131 error = amd64_set_ldt(td, largs, lp);
134 error = amd64_set_ldt(td, largs, NULL);
142 update_gdt_gsbase(struct thread *td, uint32_t base)
144 struct user_segment_descriptor *sd;
148 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
150 sd = PCPU_GET(gs32p);
151 sd->sd_lobase = base & 0xffffff;
152 sd->sd_hibase = (base >> 24) & 0xff;
157 update_gdt_fsbase(struct thread *td, uint32_t base)
159 struct user_segment_descriptor *sd;
163 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
165 sd = PCPU_GET(fs32p);
166 sd->sd_lobase = base & 0xffffff;
167 sd->sd_hibase = (base >> 24) & 0xff;
174 register struct sysarch_args *uap;
177 struct pcb *pcb = curthread->td_pcb;
180 struct i386_ioperm_args iargs;
181 struct i386_get_xfpustate i386xfpu;
182 struct amd64_get_xfpustate a64xfpu;
184 #ifdef CAPABILITY_MODE
186 * When adding new operations, add a new case statement here to
187 * explicitly indicate whether or not the operation is safe to
188 * perform in capability mode.
190 if (IN_CAPABILITY_MODE(td)) {
194 case I386_GET_IOPERM:
195 case I386_GET_FSBASE:
196 case I386_SET_FSBASE:
197 case I386_GET_GSBASE:
198 case I386_SET_GSBASE:
199 case I386_GET_XFPUSTATE:
200 case AMD64_GET_FSBASE:
201 case AMD64_SET_FSBASE:
202 case AMD64_GET_GSBASE:
203 case AMD64_SET_GSBASE:
204 case AMD64_GET_XFPUSTATE:
207 case I386_SET_IOPERM:
210 if (KTRPOINT(td, KTR_CAPFAIL))
211 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
218 if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
219 return (sysarch_ldt(td, uap, UIO_USERSPACE));
221 * XXXKIB check that the BSM generation code knows to encode
224 AUDIT_ARG_CMD(uap->op);
226 case I386_GET_IOPERM:
227 case I386_SET_IOPERM:
228 if ((error = copyin(uap->parms, &iargs,
229 sizeof(struct i386_ioperm_args))) != 0)
232 case I386_GET_XFPUSTATE:
233 if ((error = copyin(uap->parms, &i386xfpu,
234 sizeof(struct i386_get_xfpustate))) != 0)
236 a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
237 a64xfpu.len = i386xfpu.len;
239 case AMD64_GET_XFPUSTATE:
240 if ((error = copyin(uap->parms, &a64xfpu,
241 sizeof(struct amd64_get_xfpustate))) != 0)
249 case I386_GET_IOPERM:
250 error = amd64_get_ioperm(td, &iargs);
252 error = copyout(&iargs, uap->parms,
253 sizeof(struct i386_ioperm_args));
255 case I386_SET_IOPERM:
256 error = amd64_set_ioperm(td, &iargs);
258 case I386_GET_FSBASE:
259 i386base = pcb->pcb_fsbase;
260 error = copyout(&i386base, uap->parms, sizeof(i386base));
262 case I386_SET_FSBASE:
263 error = copyin(uap->parms, &i386base, sizeof(i386base));
265 pcb->pcb_fsbase = i386base;
266 td->td_frame->tf_fs = _ufssel;
267 update_gdt_fsbase(td, i386base);
270 case I386_GET_GSBASE:
271 i386base = pcb->pcb_gsbase;
272 error = copyout(&i386base, uap->parms, sizeof(i386base));
274 case I386_SET_GSBASE:
275 error = copyin(uap->parms, &i386base, sizeof(i386base));
277 pcb->pcb_gsbase = i386base;
278 td->td_frame->tf_gs = _ugssel;
279 update_gdt_gsbase(td, i386base);
282 case AMD64_GET_FSBASE:
283 error = copyout(&pcb->pcb_fsbase, uap->parms, sizeof(pcb->pcb_fsbase));
286 case AMD64_SET_FSBASE:
287 error = copyin(uap->parms, &a64base, sizeof(a64base));
289 if (a64base < VM_MAXUSER_ADDRESS) {
290 pcb->pcb_fsbase = a64base;
291 set_pcb_flags(pcb, PCB_FULL_IRET);
292 td->td_frame->tf_fs = _ufssel;
298 case AMD64_GET_GSBASE:
299 error = copyout(&pcb->pcb_gsbase, uap->parms, sizeof(pcb->pcb_gsbase));
302 case AMD64_SET_GSBASE:
303 error = copyin(uap->parms, &a64base, sizeof(a64base));
305 if (a64base < VM_MAXUSER_ADDRESS) {
306 pcb->pcb_gsbase = a64base;
307 set_pcb_flags(pcb, PCB_FULL_IRET);
308 td->td_frame->tf_gs = _ugssel;
314 case I386_GET_XFPUSTATE:
315 case AMD64_GET_XFPUSTATE:
316 if (a64xfpu.len > cpu_max_ext_state_size -
317 sizeof(struct savefpu))
320 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
321 a64xfpu.addr, a64xfpu.len);
332 amd64_set_ioperm(td, uap)
334 struct i386_ioperm_args *uap;
337 struct amd64tss *tssp;
338 struct system_segment_descriptor *tss_sd;
343 if ((error = priv_check(td, PRIV_IO)) != 0)
345 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
347 if (uap->start > uap->start + uap->length ||
348 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
353 * While this is restricted to root, we should probably figure out
354 * whether any other driver is using this i/o address, as so not to
355 * cause confusion. This probably requires a global 'usage registry'.
358 if (pcb->pcb_tssp == NULL) {
359 tssp = (struct amd64tss *)kmem_malloc(kernel_arena,
360 ctob(IOPAGES+1), M_WAITOK);
361 iomap = (char *)&tssp[1];
362 memset(iomap, 0xff, IOPERM_BITMAP_SIZE);
364 /* Takes care of tss_rsp0. */
365 memcpy(tssp, &common_tss[PCPU_GET(cpuid)],
366 sizeof(struct amd64tss));
367 tssp->tss_iobase = sizeof(*tssp);
368 pcb->pcb_tssp = tssp;
369 tss_sd = PCPU_GET(tss);
370 tss_sd->sd_lobase = (u_long)tssp & 0xffffff;
371 tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful;
372 tss_sd->sd_type = SDT_SYSTSS;
373 ltr(GSEL(GPROC0_SEL, SEL_KPL));
374 PCPU_SET(tssp, tssp);
377 iomap = (char *)&pcb->pcb_tssp[1];
378 for (i = uap->start; i < uap->start + uap->length; i++) {
380 iomap[i >> 3] &= ~(1 << (i & 7));
382 iomap[i >> 3] |= (1 << (i & 7));
388 amd64_get_ioperm(td, uap)
390 struct i386_ioperm_args *uap;
395 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
397 if (td->td_pcb->pcb_tssp == NULL) {
402 iomap = (char *)&td->td_pcb->pcb_tssp[1];
405 state = (iomap[i >> 3] >> (i & 7)) & 1;
406 uap->enable = !state;
409 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
410 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
420 * Update the GDT entry pointing to the LDT to point to the LDT of the
424 set_user_ldt(struct mdproc *mdp)
428 *PCPU_GET(ldt) = mdp->md_ldt_sd;
429 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
436 set_user_ldt_rv(struct vmspace *vmsp)
441 if (vmsp != td->td_proc->p_vmspace)
444 set_user_ldt(&td->td_proc->p_md);
450 user_ldt_alloc(struct proc *p, int force)
452 struct proc_ldt *pldt, *new_ldt;
454 struct soft_segment_descriptor sldt;
456 mtx_assert(&dt_lock, MA_OWNED);
458 if (!force && mdp->md_ldt != NULL)
459 return (mdp->md_ldt);
460 mtx_unlock(&dt_lock);
461 new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
462 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
463 max_ldt_segment * sizeof(struct user_segment_descriptor),
465 new_ldt->ldt_refcnt = 1;
466 sldt.ssd_base = (uint64_t)new_ldt->ldt_base;
467 sldt.ssd_limit = max_ldt_segment *
468 sizeof(struct user_segment_descriptor) - 1;
469 sldt.ssd_type = SDT_SYSLDT;
470 sldt.ssd_dpl = SEL_KPL;
477 if (pldt != NULL && !force) {
478 kmem_free(kernel_arena, (vm_offset_t)new_ldt->ldt_base,
479 max_ldt_segment * sizeof(struct user_segment_descriptor));
480 free(new_ldt, M_SUBPROC);
485 bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
486 sizeof(struct user_segment_descriptor));
487 user_ldt_derefl(pldt);
489 ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
490 atomic_store_rel_ptr((volatile uintptr_t *)&mdp->md_ldt,
495 return (mdp->md_ldt);
499 user_ldt_free(struct thread *td)
501 struct proc *p = td->td_proc;
502 struct mdproc *mdp = &p->p_md;
503 struct proc_ldt *pldt;
505 mtx_assert(&dt_lock, MA_OWNED);
506 if ((pldt = mdp->md_ldt) == NULL) {
507 mtx_unlock(&dt_lock);
512 bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
514 lldt(GSEL(GNULL_SEL, SEL_KPL));
515 user_ldt_deref(pldt);
519 user_ldt_derefl(struct proc_ldt *pldt)
522 if (--pldt->ldt_refcnt == 0) {
523 kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
524 max_ldt_segment * sizeof(struct user_segment_descriptor));
525 free(pldt, M_SUBPROC);
530 user_ldt_deref(struct proc_ldt *pldt)
533 mtx_assert(&dt_lock, MA_OWNED);
534 user_ldt_derefl(pldt);
535 mtx_unlock(&dt_lock);
539 * Note for the authors of compat layers (linux, etc): copyout() in
540 * the function below is not a problem since it presents data in
541 * arch-specific format (i.e. i386-specific in this case), not in
542 * the OS-specific one.
545 amd64_get_ldt(td, uap)
547 struct i386_ldt_args *uap;
550 struct proc_ldt *pldt;
552 struct user_segment_descriptor *lp;
555 printf("amd64_get_ldt: start=%d num=%d descs=%p\n",
556 uap->start, uap->num, (void *)uap->descs);
559 if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
560 lp = &((struct user_segment_descriptor *)(pldt->ldt_base))
562 num = min(uap->num, max_ldt_segment);
566 if ((uap->start > (unsigned int)max_ldt_segment) ||
567 ((unsigned int)num > (unsigned int)max_ldt_segment) ||
568 ((unsigned int)(uap->start + num) > (unsigned int)max_ldt_segment))
571 error = copyout(lp, uap->descs, num *
572 sizeof(struct user_segment_descriptor));
574 td->td_retval[0] = num;
580 amd64_set_ldt(td, uap, descs)
582 struct i386_ldt_args *uap;
583 struct user_segment_descriptor *descs;
586 unsigned int largest_ld, i;
587 struct mdproc *mdp = &td->td_proc->p_md;
588 struct proc_ldt *pldt;
589 struct user_segment_descriptor *dp;
593 printf("amd64_set_ldt: start=%d num=%d descs=%p\n",
594 uap->start, uap->num, (void *)uap->descs);
597 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
600 /* Free descriptors */
601 if (uap->start == 0 && uap->num == 0)
602 uap->num = max_ldt_segment;
605 if ((pldt = mdp->md_ldt) == NULL ||
606 uap->start >= max_ldt_segment)
608 largest_ld = uap->start + uap->num;
609 if (largest_ld > max_ldt_segment)
610 largest_ld = max_ldt_segment;
611 i = largest_ld - uap->start;
613 bzero(&((struct user_segment_descriptor *)(pldt->ldt_base))
614 [uap->start], sizeof(struct user_segment_descriptor) * i);
615 mtx_unlock(&dt_lock);
619 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
620 /* verify range of descriptors to modify */
621 largest_ld = uap->start + uap->num;
622 if (uap->start >= max_ldt_segment ||
623 largest_ld > max_ldt_segment)
627 /* Check descriptors for access violations */
628 for (i = 0; i < uap->num; i++) {
631 switch (dp->sd_type) {
632 case SDT_SYSNULL: /* system null */
650 /* I can't think of any reason to allow a user proc
651 * to create a segment of these types. They are
657 /* memory segment types */
658 case SDT_MEMEC: /* memory execute only conforming */
659 case SDT_MEMEAC: /* memory execute only accessed conforming */
660 case SDT_MEMERC: /* memory execute read conforming */
661 case SDT_MEMERAC: /* memory execute read accessed conforming */
662 /* Must be "present" if executable and conforming. */
666 case SDT_MEMRO: /* memory read only */
667 case SDT_MEMROA: /* memory read only accessed */
668 case SDT_MEMRW: /* memory read write */
669 case SDT_MEMRWA: /* memory read write accessed */
670 case SDT_MEMROD: /* memory read only expand dwn limit */
671 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
672 case SDT_MEMRWD: /* memory read write expand dwn limit */
673 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
674 case SDT_MEME: /* memory execute only */
675 case SDT_MEMEA: /* memory execute only accessed */
676 case SDT_MEMER: /* memory execute read */
677 case SDT_MEMERA: /* memory execute read accessed */
684 /* Only user (ring-3) descriptors may be present. */
685 if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
689 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
690 /* Allocate a free slot */
692 pldt = user_ldt_alloc(p, 0);
694 mtx_unlock(&dt_lock);
699 * start scanning a bit up to leave room for NVidia and
700 * Wine, which still user the "Blat" method of allocation.
703 dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
704 for (; i < max_ldt_segment; ++i, ++dp) {
705 if (dp->sd_type == SDT_SYSNULL)
708 if (i >= max_ldt_segment) {
709 mtx_unlock(&dt_lock);
713 error = amd64_set_ldt_data(td, i, 1, descs);
714 mtx_unlock(&dt_lock);
716 largest_ld = uap->start + uap->num;
717 if (largest_ld > max_ldt_segment)
720 if (user_ldt_alloc(p, 0) != NULL) {
721 error = amd64_set_ldt_data(td, uap->start, uap->num,
724 mtx_unlock(&dt_lock);
727 td->td_retval[0] = uap->start;
732 amd64_set_ldt_data(struct thread *td, int start, int num,
733 struct user_segment_descriptor *descs)
735 struct mdproc *mdp = &td->td_proc->p_md;
736 struct proc_ldt *pldt = mdp->md_ldt;
738 mtx_assert(&dt_lock, MA_OWNED);
742 &((struct user_segment_descriptor *)(pldt->ldt_base))[start],
743 num * sizeof(struct user_segment_descriptor));