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 * 4. 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/capability.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 TUNABLE_INT_FETCH("machdep.max_ldt_segment", &max_ldt_segment);
77 if (max_ldt_segment <= 0)
79 if (max_ldt_segment > MAX_LD)
80 max_ldt_segment = MAX_LD;
82 SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL);
86 static void set_user_ldt_rv(struct vmspace *vmsp);
89 static void user_ldt_derefl(struct proc_ldt *pldt);
91 #ifndef _SYS_SYSPROTO_H_
99 sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space)
101 struct i386_ldt_args *largs, la;
102 struct user_segment_descriptor *lp;
106 * XXXKIB check that the BSM generation code knows to encode
109 AUDIT_ARG_CMD(uap->op);
110 if (uap_space == UIO_USERSPACE) {
111 error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args));
116 largs = (struct i386_ldt_args *)uap->parms;
120 error = amd64_get_ldt(td, largs);
123 if (largs->descs != NULL && largs->num > max_ldt_segment)
125 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
126 if (largs->descs != NULL) {
127 lp = malloc(largs->num * sizeof(struct
128 user_segment_descriptor), M_TEMP, M_WAITOK);
129 error = copyin(largs->descs, lp, largs->num *
130 sizeof(struct user_segment_descriptor));
132 error = amd64_set_ldt(td, largs, lp);
135 error = amd64_set_ldt(td, largs, NULL);
143 update_gdt_gsbase(struct thread *td, uint32_t base)
145 struct user_segment_descriptor *sd;
149 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
151 sd = PCPU_GET(gs32p);
152 sd->sd_lobase = base & 0xffffff;
153 sd->sd_hibase = (base >> 24) & 0xff;
158 update_gdt_fsbase(struct thread *td, uint32_t base)
160 struct user_segment_descriptor *sd;
164 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
166 sd = PCPU_GET(fs32p);
167 sd->sd_lobase = base & 0xffffff;
168 sd->sd_hibase = (base >> 24) & 0xff;
175 register struct sysarch_args *uap;
178 struct pcb *pcb = curthread->td_pcb;
181 struct i386_ioperm_args iargs;
182 struct i386_get_xfpustate i386xfpu;
183 struct amd64_get_xfpustate a64xfpu;
185 #ifdef CAPABILITY_MODE
187 * When adding new operations, add a new case statement here to
188 * explicitly indicate whether or not the operation is safe to
189 * perform in capability mode.
191 if (IN_CAPABILITY_MODE(td)) {
195 case I386_GET_IOPERM:
196 case I386_GET_FSBASE:
197 case I386_SET_FSBASE:
198 case I386_GET_GSBASE:
199 case I386_SET_GSBASE:
200 case I386_GET_XFPUSTATE:
201 case AMD64_GET_FSBASE:
202 case AMD64_SET_FSBASE:
203 case AMD64_GET_GSBASE:
204 case AMD64_SET_GSBASE:
205 case AMD64_GET_XFPUSTATE:
208 case I386_SET_IOPERM:
215 if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
216 return (sysarch_ldt(td, uap, UIO_USERSPACE));
218 * XXXKIB check that the BSM generation code knows to encode
221 AUDIT_ARG_CMD(uap->op);
223 case I386_GET_IOPERM:
224 case I386_SET_IOPERM:
225 if ((error = copyin(uap->parms, &iargs,
226 sizeof(struct i386_ioperm_args))) != 0)
229 case I386_GET_XFPUSTATE:
230 if ((error = copyin(uap->parms, &i386xfpu,
231 sizeof(struct i386_get_xfpustate))) != 0)
233 a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
234 a64xfpu.len = i386xfpu.len;
236 case AMD64_GET_XFPUSTATE:
237 if ((error = copyin(uap->parms, &a64xfpu,
238 sizeof(struct amd64_get_xfpustate))) != 0)
246 case I386_GET_IOPERM:
247 error = amd64_get_ioperm(td, &iargs);
249 error = copyout(&iargs, uap->parms,
250 sizeof(struct i386_ioperm_args));
252 case I386_SET_IOPERM:
253 error = amd64_set_ioperm(td, &iargs);
255 case I386_GET_FSBASE:
256 i386base = pcb->pcb_fsbase;
257 error = copyout(&i386base, uap->parms, sizeof(i386base));
259 case I386_SET_FSBASE:
260 error = copyin(uap->parms, &i386base, sizeof(i386base));
262 pcb->pcb_fsbase = i386base;
263 td->td_frame->tf_fs = _ufssel;
264 update_gdt_fsbase(td, i386base);
267 case I386_GET_GSBASE:
268 i386base = pcb->pcb_gsbase;
269 error = copyout(&i386base, uap->parms, sizeof(i386base));
271 case I386_SET_GSBASE:
272 error = copyin(uap->parms, &i386base, sizeof(i386base));
274 pcb->pcb_gsbase = i386base;
275 td->td_frame->tf_gs = _ugssel;
276 update_gdt_gsbase(td, i386base);
279 case AMD64_GET_FSBASE:
280 error = copyout(&pcb->pcb_fsbase, uap->parms, sizeof(pcb->pcb_fsbase));
283 case AMD64_SET_FSBASE:
284 error = copyin(uap->parms, &a64base, sizeof(a64base));
286 if (a64base < VM_MAXUSER_ADDRESS) {
287 pcb->pcb_fsbase = a64base;
288 set_pcb_flags(pcb, PCB_FULL_IRET);
289 td->td_frame->tf_fs = _ufssel;
295 case AMD64_GET_GSBASE:
296 error = copyout(&pcb->pcb_gsbase, uap->parms, sizeof(pcb->pcb_gsbase));
299 case AMD64_SET_GSBASE:
300 error = copyin(uap->parms, &a64base, sizeof(a64base));
302 if (a64base < VM_MAXUSER_ADDRESS) {
303 pcb->pcb_gsbase = a64base;
304 set_pcb_flags(pcb, PCB_FULL_IRET);
305 td->td_frame->tf_gs = _ugssel;
311 case I386_GET_XFPUSTATE:
312 case AMD64_GET_XFPUSTATE:
313 if (a64xfpu.len > cpu_max_ext_state_size -
314 sizeof(struct savefpu))
317 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
318 a64xfpu.addr, a64xfpu.len);
329 amd64_set_ioperm(td, uap)
331 struct i386_ioperm_args *uap;
335 struct amd64tss *tssp;
336 struct system_segment_descriptor *tss_sd;
340 if ((error = priv_check(td, PRIV_IO)) != 0)
342 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
344 if (uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
349 * While this is restricted to root, we should probably figure out
350 * whether any other driver is using this i/o address, as so not to
351 * cause confusion. This probably requires a global 'usage registry'.
354 if (pcb->pcb_tssp == NULL) {
355 tssp = (struct amd64tss *)kmem_alloc(kernel_map,
359 iomap = (char *)&tssp[1];
360 addr = (u_long *)iomap;
361 for (i = 0; i < (ctob(IOPAGES) + 1) / sizeof(u_long); i++)
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_alloc(kernel_map,
463 max_ldt_segment * sizeof(struct user_segment_descriptor));
464 if (new_ldt->ldt_base == NULL) {
465 FREE(new_ldt, M_SUBPROC);
469 new_ldt->ldt_refcnt = 1;
470 sldt.ssd_base = (uint64_t)new_ldt->ldt_base;
471 sldt.ssd_limit = max_ldt_segment *
472 sizeof(struct user_segment_descriptor) - 1;
473 sldt.ssd_type = SDT_SYSLDT;
474 sldt.ssd_dpl = SEL_KPL;
481 if (pldt != NULL && !force) {
482 kmem_free(kernel_map, (vm_offset_t)new_ldt->ldt_base,
483 max_ldt_segment * sizeof(struct user_segment_descriptor));
484 free(new_ldt, M_SUBPROC);
489 bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
490 sizeof(struct user_segment_descriptor));
491 user_ldt_derefl(pldt);
493 ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
494 atomic_store_rel_ptr((volatile uintptr_t *)&mdp->md_ldt,
499 return (mdp->md_ldt);
503 user_ldt_free(struct thread *td)
505 struct proc *p = td->td_proc;
506 struct mdproc *mdp = &p->p_md;
507 struct proc_ldt *pldt;
509 mtx_assert(&dt_lock, MA_OWNED);
510 if ((pldt = mdp->md_ldt) == NULL) {
511 mtx_unlock(&dt_lock);
516 bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
518 lldt(GSEL(GNULL_SEL, SEL_KPL));
519 user_ldt_deref(pldt);
523 user_ldt_derefl(struct proc_ldt *pldt)
526 if (--pldt->ldt_refcnt == 0) {
527 kmem_free(kernel_map, (vm_offset_t)pldt->ldt_base,
528 max_ldt_segment * sizeof(struct user_segment_descriptor));
529 free(pldt, M_SUBPROC);
534 user_ldt_deref(struct proc_ldt *pldt)
537 mtx_assert(&dt_lock, MA_OWNED);
538 user_ldt_derefl(pldt);
539 mtx_unlock(&dt_lock);
543 * Note for the authors of compat layers (linux, etc): copyout() in
544 * the function below is not a problem since it presents data in
545 * arch-specific format (i.e. i386-specific in this case), not in
546 * the OS-specific one.
549 amd64_get_ldt(td, uap)
551 struct i386_ldt_args *uap;
554 struct proc_ldt *pldt;
556 struct user_segment_descriptor *lp;
559 printf("amd64_get_ldt: start=%d num=%d descs=%p\n",
560 uap->start, uap->num, (void *)uap->descs);
563 if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
564 lp = &((struct user_segment_descriptor *)(pldt->ldt_base))
566 num = min(uap->num, max_ldt_segment);
570 if ((uap->start > (unsigned int)max_ldt_segment) ||
571 ((unsigned int)num > (unsigned int)max_ldt_segment) ||
572 ((unsigned int)(uap->start + num) > (unsigned int)max_ldt_segment))
575 error = copyout(lp, uap->descs, num *
576 sizeof(struct user_segment_descriptor));
578 td->td_retval[0] = num;
584 amd64_set_ldt(td, uap, descs)
586 struct i386_ldt_args *uap;
587 struct user_segment_descriptor *descs;
590 unsigned int largest_ld, i;
591 struct mdproc *mdp = &td->td_proc->p_md;
592 struct proc_ldt *pldt;
593 struct user_segment_descriptor *dp;
597 printf("amd64_set_ldt: start=%d num=%d descs=%p\n",
598 uap->start, uap->num, (void *)uap->descs);
601 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
604 /* Free descriptors */
605 if (uap->start == 0 && uap->num == 0)
606 uap->num = max_ldt_segment;
609 if ((pldt = mdp->md_ldt) == NULL ||
610 uap->start >= max_ldt_segment)
612 largest_ld = uap->start + uap->num;
613 if (largest_ld > max_ldt_segment)
614 largest_ld = max_ldt_segment;
615 if (largest_ld < uap->start)
617 i = largest_ld - uap->start;
619 bzero(&((struct user_segment_descriptor *)(pldt->ldt_base))
620 [uap->start], sizeof(struct user_segment_descriptor) * i);
621 mtx_unlock(&dt_lock);
625 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
626 /* verify range of descriptors to modify */
627 largest_ld = uap->start + uap->num;
628 if (uap->start >= max_ldt_segment ||
629 largest_ld > max_ldt_segment ||
630 largest_ld < uap->start)
634 /* Check descriptors for access violations */
635 for (i = 0; i < uap->num; i++) {
638 switch (dp->sd_type) {
639 case SDT_SYSNULL: /* system null */
657 /* I can't think of any reason to allow a user proc
658 * to create a segment of these types. They are
664 /* memory segment types */
665 case SDT_MEMEC: /* memory execute only conforming */
666 case SDT_MEMEAC: /* memory execute only accessed conforming */
667 case SDT_MEMERC: /* memory execute read conforming */
668 case SDT_MEMERAC: /* memory execute read accessed conforming */
669 /* Must be "present" if executable and conforming. */
673 case SDT_MEMRO: /* memory read only */
674 case SDT_MEMROA: /* memory read only accessed */
675 case SDT_MEMRW: /* memory read write */
676 case SDT_MEMRWA: /* memory read write accessed */
677 case SDT_MEMROD: /* memory read only expand dwn limit */
678 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
679 case SDT_MEMRWD: /* memory read write expand dwn limit */
680 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
681 case SDT_MEME: /* memory execute only */
682 case SDT_MEMEA: /* memory execute only accessed */
683 case SDT_MEMER: /* memory execute read */
684 case SDT_MEMERA: /* memory execute read accessed */
691 /* Only user (ring-3) descriptors may be present. */
692 if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
696 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
697 /* Allocate a free slot */
699 pldt = user_ldt_alloc(p, 0);
701 mtx_unlock(&dt_lock);
706 * start scanning a bit up to leave room for NVidia and
707 * Wine, which still user the "Blat" method of allocation.
710 dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
711 for (; i < max_ldt_segment; ++i, ++dp) {
712 if (dp->sd_type == SDT_SYSNULL)
715 if (i >= max_ldt_segment) {
716 mtx_unlock(&dt_lock);
720 error = amd64_set_ldt_data(td, i, 1, descs);
721 mtx_unlock(&dt_lock);
723 largest_ld = uap->start + uap->num;
724 if (largest_ld > max_ldt_segment)
727 if (user_ldt_alloc(p, 0) != NULL) {
728 error = amd64_set_ldt_data(td, uap->start, uap->num,
731 mtx_unlock(&dt_lock);
734 td->td_retval[0] = uap->start;
739 amd64_set_ldt_data(struct thread *td, int start, int num,
740 struct user_segment_descriptor *descs)
742 struct mdproc *mdp = &td->td_proc->p_md;
743 struct proc_ldt *pldt = mdp->md_ldt;
745 mtx_assert(&dt_lock, MA_OWNED);
749 &((struct user_segment_descriptor *)(pldt->ldt_base))[start],
750 num * sizeof(struct user_segment_descriptor));