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 = 512;
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;
172 sysarch(struct thread *td, struct sysarch_args *uap)
175 struct pcb *pcb = curthread->td_pcb;
178 struct i386_ioperm_args iargs;
179 struct i386_get_xfpustate i386xfpu;
180 struct amd64_get_xfpustate a64xfpu;
182 #ifdef CAPABILITY_MODE
184 * When adding new operations, add a new case statement here to
185 * explicitly indicate whether or not the operation is safe to
186 * perform in capability mode.
188 if (IN_CAPABILITY_MODE(td)) {
192 case I386_GET_IOPERM:
193 case I386_GET_FSBASE:
194 case I386_SET_FSBASE:
195 case I386_GET_GSBASE:
196 case I386_SET_GSBASE:
197 case I386_GET_XFPUSTATE:
198 case AMD64_GET_FSBASE:
199 case AMD64_SET_FSBASE:
200 case AMD64_GET_GSBASE:
201 case AMD64_SET_GSBASE:
202 case AMD64_GET_XFPUSTATE:
205 case I386_SET_IOPERM:
208 if (KTRPOINT(td, KTR_CAPFAIL))
209 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
216 if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
217 return (sysarch_ldt(td, uap, UIO_USERSPACE));
219 * XXXKIB check that the BSM generation code knows to encode
222 AUDIT_ARG_CMD(uap->op);
224 case I386_GET_IOPERM:
225 case I386_SET_IOPERM:
226 if ((error = copyin(uap->parms, &iargs,
227 sizeof(struct i386_ioperm_args))) != 0)
230 case I386_GET_XFPUSTATE:
231 if ((error = copyin(uap->parms, &i386xfpu,
232 sizeof(struct i386_get_xfpustate))) != 0)
234 a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
235 a64xfpu.len = i386xfpu.len;
237 case AMD64_GET_XFPUSTATE:
238 if ((error = copyin(uap->parms, &a64xfpu,
239 sizeof(struct amd64_get_xfpustate))) != 0)
247 case I386_GET_IOPERM:
248 error = amd64_get_ioperm(td, &iargs);
250 error = copyout(&iargs, uap->parms,
251 sizeof(struct i386_ioperm_args));
253 case I386_SET_IOPERM:
254 error = amd64_set_ioperm(td, &iargs);
256 case I386_GET_FSBASE:
257 update_pcb_bases(pcb);
258 i386base = pcb->pcb_fsbase;
259 error = copyout(&i386base, uap->parms, sizeof(i386base));
261 case I386_SET_FSBASE:
262 error = copyin(uap->parms, &i386base, sizeof(i386base));
264 set_pcb_flags(pcb, PCB_FULL_IRET);
265 pcb->pcb_fsbase = i386base;
266 td->td_frame->tf_fs = _ufssel;
267 update_gdt_fsbase(td, i386base);
270 case I386_GET_GSBASE:
271 update_pcb_bases(pcb);
272 i386base = pcb->pcb_gsbase;
273 error = copyout(&i386base, uap->parms, sizeof(i386base));
275 case I386_SET_GSBASE:
276 error = copyin(uap->parms, &i386base, sizeof(i386base));
278 set_pcb_flags(pcb, PCB_FULL_IRET);
279 pcb->pcb_gsbase = i386base;
280 td->td_frame->tf_gs = _ugssel;
281 update_gdt_gsbase(td, i386base);
284 case AMD64_GET_FSBASE:
285 update_pcb_bases(pcb);
286 error = copyout(&pcb->pcb_fsbase, uap->parms,
287 sizeof(pcb->pcb_fsbase));
290 case AMD64_SET_FSBASE:
291 error = copyin(uap->parms, &a64base, sizeof(a64base));
293 if (a64base < VM_MAXUSER_ADDRESS) {
294 set_pcb_flags(pcb, PCB_FULL_IRET);
295 pcb->pcb_fsbase = a64base;
296 td->td_frame->tf_fs = _ufssel;
302 case AMD64_GET_GSBASE:
303 update_pcb_bases(pcb);
304 error = copyout(&pcb->pcb_gsbase, uap->parms,
305 sizeof(pcb->pcb_gsbase));
308 case AMD64_SET_GSBASE:
309 error = copyin(uap->parms, &a64base, sizeof(a64base));
311 if (a64base < VM_MAXUSER_ADDRESS) {
312 set_pcb_flags(pcb, PCB_FULL_IRET);
313 pcb->pcb_gsbase = a64base;
314 td->td_frame->tf_gs = _ugssel;
320 case I386_GET_XFPUSTATE:
321 case AMD64_GET_XFPUSTATE:
322 if (a64xfpu.len > cpu_max_ext_state_size -
323 sizeof(struct savefpu))
326 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
327 a64xfpu.addr, a64xfpu.len);
338 amd64_set_ioperm(td, uap)
340 struct i386_ioperm_args *uap;
343 struct amd64tss *tssp;
344 struct system_segment_descriptor *tss_sd;
349 if ((error = priv_check(td, PRIV_IO)) != 0)
351 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
353 if (uap->start > uap->start + uap->length ||
354 uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
359 * While this is restricted to root, we should probably figure out
360 * whether any other driver is using this i/o address, as so not to
361 * cause confusion. This probably requires a global 'usage registry'.
364 if (pcb->pcb_tssp == NULL) {
365 tssp = (struct amd64tss *)kmem_malloc(kernel_arena,
366 ctob(IOPAGES+1), M_WAITOK);
367 iomap = (char *)&tssp[1];
368 memset(iomap, 0xff, IOPERM_BITMAP_SIZE);
370 /* Takes care of tss_rsp0. */
371 memcpy(tssp, &common_tss[PCPU_GET(cpuid)],
372 sizeof(struct amd64tss));
373 tssp->tss_iobase = sizeof(*tssp);
374 pcb->pcb_tssp = tssp;
375 tss_sd = PCPU_GET(tss);
376 tss_sd->sd_lobase = (u_long)tssp & 0xffffff;
377 tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful;
378 tss_sd->sd_type = SDT_SYSTSS;
379 ltr(GSEL(GPROC0_SEL, SEL_KPL));
380 PCPU_SET(tssp, tssp);
383 iomap = (char *)&pcb->pcb_tssp[1];
384 for (i = uap->start; i < uap->start + uap->length; i++) {
386 iomap[i >> 3] &= ~(1 << (i & 7));
388 iomap[i >> 3] |= (1 << (i & 7));
394 amd64_get_ioperm(td, uap)
396 struct i386_ioperm_args *uap;
401 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
403 if (td->td_pcb->pcb_tssp == NULL) {
408 iomap = (char *)&td->td_pcb->pcb_tssp[1];
411 state = (iomap[i >> 3] >> (i & 7)) & 1;
412 uap->enable = !state;
415 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
416 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
426 * Update the GDT entry pointing to the LDT to point to the LDT of the
430 set_user_ldt(struct mdproc *mdp)
434 *PCPU_GET(ldt) = mdp->md_ldt_sd;
435 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
442 set_user_ldt_rv(struct vmspace *vmsp)
447 if (vmsp != td->td_proc->p_vmspace)
450 set_user_ldt(&td->td_proc->p_md);
456 user_ldt_alloc(struct proc *p, int force)
458 struct proc_ldt *pldt, *new_ldt;
460 struct soft_segment_descriptor sldt;
462 mtx_assert(&dt_lock, MA_OWNED);
464 if (!force && mdp->md_ldt != NULL)
465 return (mdp->md_ldt);
466 mtx_unlock(&dt_lock);
467 new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
468 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
469 max_ldt_segment * sizeof(struct user_segment_descriptor),
471 new_ldt->ldt_refcnt = 1;
472 sldt.ssd_base = (uint64_t)new_ldt->ldt_base;
473 sldt.ssd_limit = max_ldt_segment *
474 sizeof(struct user_segment_descriptor) - 1;
475 sldt.ssd_type = SDT_SYSLDT;
476 sldt.ssd_dpl = SEL_KPL;
483 if (pldt != NULL && !force) {
484 kmem_free(kernel_arena, (vm_offset_t)new_ldt->ldt_base,
485 max_ldt_segment * sizeof(struct user_segment_descriptor));
486 free(new_ldt, M_SUBPROC);
491 bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
492 sizeof(struct user_segment_descriptor));
493 user_ldt_derefl(pldt);
495 ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
496 atomic_store_rel_ptr((volatile uintptr_t *)&mdp->md_ldt,
501 return (mdp->md_ldt);
505 user_ldt_free(struct thread *td)
507 struct proc *p = td->td_proc;
508 struct mdproc *mdp = &p->p_md;
509 struct proc_ldt *pldt;
511 mtx_assert(&dt_lock, MA_OWNED);
512 if ((pldt = mdp->md_ldt) == NULL) {
513 mtx_unlock(&dt_lock);
518 bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
520 lldt(GSEL(GNULL_SEL, SEL_KPL));
521 user_ldt_deref(pldt);
525 user_ldt_derefl(struct proc_ldt *pldt)
528 if (--pldt->ldt_refcnt == 0) {
529 kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
530 max_ldt_segment * sizeof(struct user_segment_descriptor));
531 free(pldt, M_SUBPROC);
536 user_ldt_deref(struct proc_ldt *pldt)
539 mtx_assert(&dt_lock, MA_OWNED);
540 user_ldt_derefl(pldt);
541 mtx_unlock(&dt_lock);
545 * Note for the authors of compat layers (linux, etc): copyout() in
546 * the function below is not a problem since it presents data in
547 * arch-specific format (i.e. i386-specific in this case), not in
548 * the OS-specific one.
551 amd64_get_ldt(struct thread *td, struct i386_ldt_args *uap)
553 struct proc_ldt *pldt;
554 struct user_segment_descriptor *lp;
560 printf("amd64_get_ldt: start=%u num=%u descs=%p\n",
561 uap->start, uap->num, (void *)uap->descs);
564 if (uap->start >= max_ldt_segment)
566 num = min(uap->num, max_ldt_segment - uap->start);
567 pldt = td->td_proc->p_md.md_ldt;
570 lp = &((struct user_segment_descriptor *)(pldt->ldt_base))[uap->start];
571 data = malloc(num * sizeof(struct user_segment_descriptor), M_TEMP,
574 for (i = 0; i < num; i++)
575 data[i] = ((volatile uint64_t *)lp)[i];
576 mtx_unlock(&dt_lock);
577 error = copyout(data, uap->descs, num *
578 sizeof(struct user_segment_descriptor));
581 td->td_retval[0] = num;
586 amd64_set_ldt(struct thread *td, struct i386_ldt_args *uap,
587 struct user_segment_descriptor *descs)
590 struct proc_ldt *pldt;
591 struct user_segment_descriptor *dp;
597 printf("amd64_set_ldt: start=%u num=%u descs=%p\n",
598 uap->start, uap->num, (void *)uap->descs);
600 mdp = &td->td_proc->p_md;
603 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
606 /* Free descriptors */
607 if (uap->start == 0 && uap->num == 0)
608 uap->num = max_ldt_segment;
611 if ((pldt = mdp->md_ldt) == NULL ||
612 uap->start >= max_ldt_segment)
614 largest_ld = uap->start + uap->num;
615 if (largest_ld > max_ldt_segment)
616 largest_ld = max_ldt_segment;
617 if (largest_ld < uap->start)
620 for (i = uap->start; i < largest_ld; i++)
621 ((volatile uint64_t *)(pldt->ldt_base))[i] = 0;
622 mtx_unlock(&dt_lock);
626 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
627 /* verify range of descriptors to modify */
628 largest_ld = uap->start + uap->num;
629 if (uap->start >= max_ldt_segment ||
630 largest_ld > max_ldt_segment ||
631 largest_ld < uap->start)
635 /* Check descriptors for access violations */
636 for (i = 0; i < uap->num; i++) {
639 switch (dp->sd_type) {
640 case SDT_SYSNULL: /* system null */
660 /* memory segment types */
661 case SDT_MEMEC: /* memory execute only conforming */
662 case SDT_MEMEAC: /* memory execute only accessed conforming */
663 case SDT_MEMERC: /* memory execute read conforming */
664 case SDT_MEMERAC: /* memory execute read accessed conforming */
665 /* Must be "present" if executable and conforming. */
669 case SDT_MEMRO: /* memory read only */
670 case SDT_MEMROA: /* memory read only accessed */
671 case SDT_MEMRW: /* memory read write */
672 case SDT_MEMRWA: /* memory read write accessed */
673 case SDT_MEMROD: /* memory read only expand dwn limit */
674 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
675 case SDT_MEMRWD: /* memory read write expand dwn limit */
676 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
677 case SDT_MEME: /* memory execute only */
678 case SDT_MEMEA: /* memory execute only accessed */
679 case SDT_MEMER: /* memory execute read */
680 case SDT_MEMERA: /* memory execute read accessed */
686 /* Only user (ring-3) descriptors may be present. */
687 if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
691 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
692 /* Allocate a free slot */
694 pldt = user_ldt_alloc(p, 0);
696 mtx_unlock(&dt_lock);
701 * start scanning a bit up to leave room for NVidia and
702 * Wine, which still user the "Blat" method of allocation.
705 dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
706 for (; i < max_ldt_segment; ++i, ++dp) {
707 if (dp->sd_type == SDT_SYSNULL)
710 if (i >= max_ldt_segment) {
711 mtx_unlock(&dt_lock);
715 error = amd64_set_ldt_data(td, i, 1, descs);
716 mtx_unlock(&dt_lock);
718 largest_ld = uap->start + uap->num;
719 if (largest_ld > max_ldt_segment)
722 if (user_ldt_alloc(p, 0) != NULL) {
723 error = amd64_set_ldt_data(td, uap->start, uap->num,
726 mtx_unlock(&dt_lock);
729 td->td_retval[0] = uap->start;
734 amd64_set_ldt_data(struct thread *td, int start, int num,
735 struct user_segment_descriptor *descs)
738 struct proc_ldt *pldt;
739 volatile uint64_t *dst, *src;
742 mtx_assert(&dt_lock, MA_OWNED);
744 mdp = &td->td_proc->p_md;
746 dst = (volatile uint64_t *)(pldt->ldt_base);
747 src = (volatile uint64_t *)descs;
748 for (i = 0; i < num; i++)
749 dst[start + i] = src[i];