2 * Copyright (c) 1990 The Regents of the University of California.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_capsicum.h"
36 #include "opt_kstack_pages.h"
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
40 #include <sys/systm.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
47 #include <sys/sysproto.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_extern.h>
54 #include <machine/cpu.h>
55 #include <machine/pcb.h>
56 #include <machine/pcb_ext.h>
57 #include <machine/proc.h>
58 #include <machine/sysarch.h>
60 #include <security/audit/audit.h>
62 #include <vm/vm_kern.h> /* for kernel_map */
65 #define LD_PER_PAGE 512
66 #define NEW_MAX_LD(num) ((num + LD_PER_PAGE) & ~(LD_PER_PAGE-1))
67 #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3)
68 #define NULL_LDT_BASE ((caddr_t)NULL)
71 static void set_user_ldt_rv(struct vmspace *vmsp);
73 static int i386_set_ldt_data(struct thread *, int start, int num,
74 union descriptor *descs);
75 static int i386_ldt_grow(struct thread *td, int len);
78 fill_based_sd(struct segment_descriptor *sdp, uint32_t base)
81 sdp->sd_lobase = base & 0xffffff;
82 sdp->sd_hibase = (base >> 24) & 0xff;
83 sdp->sd_lolimit = 0xffff; /* 4GB limit, wraps around */
84 sdp->sd_hilimit = 0xf;
85 sdp->sd_type = SDT_MEMRWA;
86 sdp->sd_dpl = SEL_UPL;
93 #ifndef _SYS_SYSPROTO_H_
103 register struct sysarch_args *uap;
106 union descriptor *lp;
108 struct i386_ldt_args largs;
109 struct i386_ioperm_args iargs;
110 struct i386_get_xfpustate xfpu;
113 struct segment_descriptor sd, *sdp;
115 AUDIT_ARG_CMD(uap->op);
117 #ifdef CAPABILITY_MODE
119 * When adding new operations, add a new case statement here to
120 * explicitly indicate whether or not the operation is safe to
121 * perform in capability mode.
123 if (IN_CAPABILITY_MODE(td)) {
127 case I386_GET_IOPERM:
128 case I386_GET_FSBASE:
129 case I386_SET_FSBASE:
130 case I386_GET_GSBASE:
131 case I386_SET_GSBASE:
132 case I386_GET_XFPUSTATE:
135 case I386_SET_IOPERM:
138 if (KTRPOINT(td, KTR_CAPFAIL))
139 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
147 case I386_GET_IOPERM:
148 case I386_SET_IOPERM:
149 if ((error = copyin(uap->parms, &kargs.iargs,
150 sizeof(struct i386_ioperm_args))) != 0)
155 if ((error = copyin(uap->parms, &kargs.largs,
156 sizeof(struct i386_ldt_args))) != 0)
158 if (kargs.largs.num > MAX_LD || kargs.largs.num <= 0)
161 case I386_GET_XFPUSTATE:
162 if ((error = copyin(uap->parms, &kargs.xfpu,
163 sizeof(struct i386_get_xfpustate))) != 0)
172 error = i386_get_ldt(td, &kargs.largs);
175 if (kargs.largs.descs != NULL) {
176 lp = (union descriptor *)malloc(
177 kargs.largs.num * sizeof(union descriptor),
179 error = copyin(kargs.largs.descs, lp,
180 kargs.largs.num * sizeof(union descriptor));
182 error = i386_set_ldt(td, &kargs.largs, lp);
185 error = i386_set_ldt(td, &kargs.largs, NULL);
188 case I386_GET_IOPERM:
189 error = i386_get_ioperm(td, &kargs.iargs);
191 error = copyout(&kargs.iargs, uap->parms,
192 sizeof(struct i386_ioperm_args));
194 case I386_SET_IOPERM:
195 error = i386_set_ioperm(td, &kargs.iargs);
198 error = vm86_sysarch(td, uap->parms);
200 case I386_GET_FSBASE:
201 sdp = &td->td_pcb->pcb_fsd;
202 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
203 error = copyout(&base, uap->parms, sizeof(base));
205 case I386_SET_FSBASE:
206 error = copyin(uap->parms, &base, sizeof(base));
209 * Construct a descriptor and store it in the pcb for
210 * the next context switch. Also store it in the gdt
211 * so that the load of tf_fs into %fs will activate it
212 * at return to userland.
214 fill_based_sd(&sd, base);
216 td->td_pcb->pcb_fsd = sd;
217 PCPU_GET(fsgs_gdt)[0] = sd;
219 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
222 case I386_GET_GSBASE:
223 sdp = &td->td_pcb->pcb_gsd;
224 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
225 error = copyout(&base, uap->parms, sizeof(base));
227 case I386_SET_GSBASE:
228 error = copyin(uap->parms, &base, sizeof(base));
231 * Construct a descriptor and store it in the pcb for
232 * the next context switch. Also store it in the gdt
233 * because we have to do a load_gs() right now.
235 fill_based_sd(&sd, base);
237 td->td_pcb->pcb_gsd = sd;
238 PCPU_GET(fsgs_gdt)[1] = sd;
240 load_gs(GSEL(GUGS_SEL, SEL_UPL));
243 case I386_GET_XFPUSTATE:
244 if (kargs.xfpu.len > cpu_max_ext_state_size -
245 sizeof(union savefpu))
248 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
249 kargs.xfpu.addr, kargs.xfpu.len);
259 i386_extend_pcb(struct thread *td)
264 struct soft_segment_descriptor ssd = {
265 0, /* segment base address (overwritten) */
266 ctob(IOPAGES + 1) - 1, /* length */
267 SDT_SYS386TSS, /* segment type */
268 0, /* priority level */
269 1, /* descriptor present */
271 0, /* default 32 size */
275 ext = (struct pcb_ext *)kmem_malloc(kernel_arena, ctob(IOPAGES+1),
277 /* -16 is so we can convert a trapframe into vm86trapframe inplace */
278 ext->ext_tss.tss_esp0 = (vm_offset_t)td->td_pcb - 16;
279 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
281 * The last byte of the i/o map must be followed by an 0xff byte.
282 * We arbitrarily allocate 16 bytes here, to keep the starting
283 * address on a doubleword boundary.
285 offset = PAGE_SIZE - 16;
286 ext->ext_tss.tss_ioopt =
287 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16;
288 ext->ext_iomap = (caddr_t)ext + offset;
289 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32;
291 addr = (u_long *)ext->ext_vm86.vm86_intmap;
292 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++)
295 ssd.ssd_base = (unsigned)&ext->ext_tss;
296 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext);
297 ssdtosd(&ssd, &ext->ext_tssd);
299 KASSERT(td == curthread, ("giving TSS to !curthread"));
300 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!"));
302 /* Switch to the new TSS. */
304 td->td_pcb->pcb_ext = ext;
305 PCPU_SET(private_tss, 1);
306 *PCPU_GET(tss_gdt) = ext->ext_tssd;
307 ltr(GSEL(GPROC0_SEL, SEL_KPL));
314 i386_set_ioperm(td, uap)
316 struct i386_ioperm_args *uap;
321 if ((error = priv_check(td, PRIV_IO)) != 0)
323 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
327 * While this is restricted to root, we should probably figure out
328 * whether any other driver is using this i/o address, as so not to
329 * cause confusion. This probably requires a global 'usage registry'.
332 if (td->td_pcb->pcb_ext == 0)
333 if ((error = i386_extend_pcb(td)) != 0)
335 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
337 if (uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
340 for (i = uap->start; i < uap->start + uap->length; i++) {
342 iomap[i >> 3] &= ~(1 << (i & 7));
344 iomap[i >> 3] |= (1 << (i & 7));
350 i386_get_ioperm(td, uap)
352 struct i386_ioperm_args *uap;
357 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
360 if (td->td_pcb->pcb_ext == 0) {
365 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
368 state = (iomap[i >> 3] >> (i & 7)) & 1;
369 uap->enable = !state;
372 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
373 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
383 * Update the GDT entry pointing to the LDT to point to the LDT of the
384 * current process. Manage dt_lock holding/unholding autonomously.
387 set_user_ldt(struct mdproc *mdp)
389 struct proc_ldt *pldt;
393 if (!mtx_owned(&dt_lock)) {
394 mtx_lock_spin(&dt_lock);
400 gdt[PCPU_GET(cpuid) * NGDT + GUSERLDT_SEL].sd = pldt->ldt_sd;
402 gdt[GUSERLDT_SEL].sd = pldt->ldt_sd;
404 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
405 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL));
407 mtx_unlock_spin(&dt_lock);
412 set_user_ldt_rv(struct vmspace *vmsp)
417 if (vmsp != td->td_proc->p_vmspace)
420 set_user_ldt(&td->td_proc->p_md);
425 * dt_lock must be held. Returns with dt_lock held.
428 user_ldt_alloc(struct mdproc *mdp, int len)
430 struct proc_ldt *pldt, *new_ldt;
432 mtx_assert(&dt_lock, MA_OWNED);
433 mtx_unlock_spin(&dt_lock);
434 new_ldt = malloc(sizeof(struct proc_ldt),
435 M_SUBPROC, M_WAITOK);
437 new_ldt->ldt_len = len = NEW_MAX_LD(len);
438 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
439 len * sizeof(union descriptor), M_WAITOK);
440 new_ldt->ldt_refcnt = 1;
441 new_ldt->ldt_active = 0;
443 mtx_lock_spin(&dt_lock);
444 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base;
445 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1;
446 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd);
448 if ((pldt = mdp->md_ldt) != NULL) {
449 if (len > pldt->ldt_len)
451 bcopy(pldt->ldt_base, new_ldt->ldt_base,
452 len * sizeof(union descriptor));
454 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt));
460 * Must be called with dt_lock held. Returns with dt_lock unheld.
463 user_ldt_free(struct thread *td)
465 struct mdproc *mdp = &td->td_proc->p_md;
466 struct proc_ldt *pldt;
468 mtx_assert(&dt_lock, MA_OWNED);
469 if ((pldt = mdp->md_ldt) == NULL) {
470 mtx_unlock_spin(&dt_lock);
474 if (td == curthread) {
476 PCPU_SET(currentldt, _default_ldt);
480 user_ldt_deref(pldt);
484 user_ldt_deref(struct proc_ldt *pldt)
487 mtx_assert(&dt_lock, MA_OWNED);
488 if (--pldt->ldt_refcnt == 0) {
489 mtx_unlock_spin(&dt_lock);
490 kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
491 pldt->ldt_len * sizeof(union descriptor));
492 free(pldt, M_SUBPROC);
494 mtx_unlock_spin(&dt_lock);
498 * Note for the authors of compat layers (linux, etc): copyout() in
499 * the function below is not a problem since it presents data in
500 * arch-specific format (i.e. i386-specific in this case), not in
501 * the OS-specific one.
504 i386_get_ldt(td, uap)
506 struct i386_ldt_args *uap;
509 struct proc_ldt *pldt;
511 union descriptor *lp;
514 printf("i386_get_ldt: start=%d num=%d descs=%p\n",
515 uap->start, uap->num, (void *)uap->descs);
518 mtx_lock_spin(&dt_lock);
519 if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
520 nldt = pldt->ldt_len;
521 lp = &((union descriptor *)(pldt->ldt_base))[uap->start];
522 mtx_unlock_spin(&dt_lock);
523 num = min(uap->num, nldt);
525 mtx_unlock_spin(&dt_lock);
526 nldt = sizeof(ldt)/sizeof(ldt[0]);
527 num = min(uap->num, nldt);
528 lp = &ldt[uap->start];
531 if ((uap->start > (unsigned int)nldt) ||
532 ((unsigned int)num > (unsigned int)nldt) ||
533 ((unsigned int)(uap->start + num) > (unsigned int)nldt))
536 error = copyout(lp, uap->descs, num * sizeof(union descriptor));
538 td->td_retval[0] = num;
544 i386_set_ldt(td, uap, descs)
546 struct i386_ldt_args *uap;
547 union descriptor *descs;
551 struct mdproc *mdp = &td->td_proc->p_md;
552 struct proc_ldt *pldt;
553 union descriptor *dp;
556 printf("i386_set_ldt: start=%d num=%d descs=%p\n",
557 uap->start, uap->num, (void *)uap->descs);
561 /* Free descriptors */
562 if (uap->start == 0 && uap->num == 0) {
564 * Treat this as a special case, so userland needn't
565 * know magic number NLDT.
568 uap->num = MAX_LD - NLDT;
572 mtx_lock_spin(&dt_lock);
573 if ((pldt = mdp->md_ldt) == NULL ||
574 uap->start >= pldt->ldt_len) {
575 mtx_unlock_spin(&dt_lock);
578 largest_ld = uap->start + uap->num;
579 if (largest_ld > pldt->ldt_len)
580 largest_ld = pldt->ldt_len;
581 i = largest_ld - uap->start;
582 bzero(&((union descriptor *)(pldt->ldt_base))[uap->start],
583 sizeof(union descriptor) * i);
584 mtx_unlock_spin(&dt_lock);
588 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
589 /* verify range of descriptors to modify */
590 largest_ld = uap->start + uap->num;
591 if (uap->start >= MAX_LD || largest_ld > MAX_LD) {
596 /* Check descriptors for access violations */
597 for (i = 0; i < uap->num; i++) {
600 switch (dp->sd.sd_type) {
601 case SDT_SYSNULL: /* system null */
604 case SDT_SYS286TSS: /* system 286 TSS available */
605 case SDT_SYSLDT: /* system local descriptor table */
606 case SDT_SYS286BSY: /* system 286 TSS busy */
607 case SDT_SYSTASKGT: /* system task gate */
608 case SDT_SYS286IGT: /* system 286 interrupt gate */
609 case SDT_SYS286TGT: /* system 286 trap gate */
610 case SDT_SYSNULL2: /* undefined by Intel */
611 case SDT_SYS386TSS: /* system 386 TSS available */
612 case SDT_SYSNULL3: /* undefined by Intel */
613 case SDT_SYS386BSY: /* system 386 TSS busy */
614 case SDT_SYSNULL4: /* undefined by Intel */
615 case SDT_SYS386IGT: /* system 386 interrupt gate */
616 case SDT_SYS386TGT: /* system 386 trap gate */
617 case SDT_SYS286CGT: /* system 286 call gate */
618 case SDT_SYS386CGT: /* system 386 call gate */
619 /* I can't think of any reason to allow a user proc
620 * to create a segment of these types. They are
626 /* memory segment types */
627 case SDT_MEMEC: /* memory execute only conforming */
628 case SDT_MEMEAC: /* memory execute only accessed conforming */
629 case SDT_MEMERC: /* memory execute read conforming */
630 case SDT_MEMERAC: /* memory execute read accessed conforming */
631 /* Must be "present" if executable and conforming. */
632 if (dp->sd.sd_p == 0)
635 case SDT_MEMRO: /* memory read only */
636 case SDT_MEMROA: /* memory read only accessed */
637 case SDT_MEMRW: /* memory read write */
638 case SDT_MEMRWA: /* memory read write accessed */
639 case SDT_MEMROD: /* memory read only expand dwn limit */
640 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
641 case SDT_MEMRWD: /* memory read write expand dwn limit */
642 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
643 case SDT_MEME: /* memory execute only */
644 case SDT_MEMEA: /* memory execute only accessed */
645 case SDT_MEMER: /* memory execute read */
646 case SDT_MEMERA: /* memory execute read accessed */
653 /* Only user (ring-3) descriptors may be present. */
654 if ((dp->sd.sd_p != 0) && (dp->sd.sd_dpl != SEL_UPL))
658 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
659 /* Allocate a free slot */
660 mtx_lock_spin(&dt_lock);
661 if ((pldt = mdp->md_ldt) == NULL) {
662 if ((error = i386_ldt_grow(td, NLDT + 1))) {
663 mtx_unlock_spin(&dt_lock);
670 * start scanning a bit up to leave room for NVidia and
671 * Wine, which still user the "Blat" method of allocation.
673 dp = &((union descriptor *)(pldt->ldt_base))[NLDT];
674 for (i = NLDT; i < pldt->ldt_len; ++i) {
675 if (dp->sd.sd_type == SDT_SYSNULL)
679 if (i >= pldt->ldt_len) {
680 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) {
681 mtx_unlock_spin(&dt_lock);
687 error = i386_set_ldt_data(td, i, 1, descs);
688 mtx_unlock_spin(&dt_lock);
690 largest_ld = uap->start + uap->num;
691 mtx_lock_spin(&dt_lock);
692 if (!(error = i386_ldt_grow(td, largest_ld))) {
693 error = i386_set_ldt_data(td, uap->start, uap->num,
696 mtx_unlock_spin(&dt_lock);
699 td->td_retval[0] = uap->start;
704 i386_set_ldt_data(struct thread *td, int start, int num,
705 union descriptor *descs)
707 struct mdproc *mdp = &td->td_proc->p_md;
708 struct proc_ldt *pldt = mdp->md_ldt;
710 mtx_assert(&dt_lock, MA_OWNED);
714 &((union descriptor *)(pldt->ldt_base))[start],
715 num * sizeof(union descriptor));
720 i386_ldt_grow(struct thread *td, int len)
722 struct mdproc *mdp = &td->td_proc->p_md;
723 struct proc_ldt *new_ldt, *pldt;
724 caddr_t old_ldt_base = NULL_LDT_BASE;
727 mtx_assert(&dt_lock, MA_OWNED);
734 /* Allocate a user ldt. */
735 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) {
736 new_ldt = user_ldt_alloc(mdp, len);
742 if (new_ldt->ldt_len <= pldt->ldt_len) {
744 * We just lost the race for allocation, so
745 * free the new object and return.
747 mtx_unlock_spin(&dt_lock);
748 kmem_free(kernel_arena,
749 (vm_offset_t)new_ldt->ldt_base,
750 new_ldt->ldt_len * sizeof(union descriptor));
751 free(new_ldt, M_SUBPROC);
752 mtx_lock_spin(&dt_lock);
757 * We have to substitute the current LDT entry for
758 * curproc with the new one since its size grew.
760 old_ldt_base = pldt->ldt_base;
761 old_ldt_len = pldt->ldt_len;
762 pldt->ldt_sd = new_ldt->ldt_sd;
763 pldt->ldt_base = new_ldt->ldt_base;
764 pldt->ldt_len = new_ldt->ldt_len;
766 mdp->md_ldt = pldt = new_ldt;
769 * Signal other cpus to reload ldt. We need to unlock dt_lock
770 * here because other CPU will contest on it since their
771 * curthreads won't hold the lock and will block when trying
774 mtx_unlock_spin(&dt_lock);
775 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv,
776 NULL, td->td_proc->p_vmspace);
778 set_user_ldt(&td->td_proc->p_md);
779 mtx_unlock_spin(&dt_lock);
781 if (old_ldt_base != NULL_LDT_BASE) {
782 kmem_free(kernel_arena, (vm_offset_t)old_ldt_base,
783 old_ldt_len * sizeof(union descriptor));
784 free(new_ldt, M_SUBPROC);
786 mtx_lock_spin(&dt_lock);