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);
77 #ifndef _SYS_SYSPROTO_H_
87 register struct sysarch_args *uap;
92 struct i386_ldt_args largs;
93 struct i386_ioperm_args iargs;
94 struct i386_get_xfpustate xfpu;
97 struct segment_descriptor sd, *sdp;
99 AUDIT_ARG_CMD(uap->op);
101 #ifdef CAPABILITY_MODE
103 * When adding new operations, add a new case statement here to
104 * explicitly indicate whether or not the operation is safe to
105 * perform in capability mode.
107 if (IN_CAPABILITY_MODE(td)) {
111 case I386_GET_IOPERM:
112 case I386_GET_FSBASE:
113 case I386_SET_FSBASE:
114 case I386_GET_GSBASE:
115 case I386_SET_GSBASE:
116 case I386_GET_XFPUSTATE:
119 case I386_SET_IOPERM:
122 if (KTRPOINT(td, KTR_CAPFAIL))
123 ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
131 case I386_GET_IOPERM:
132 case I386_SET_IOPERM:
133 if ((error = copyin(uap->parms, &kargs.iargs,
134 sizeof(struct i386_ioperm_args))) != 0)
139 if ((error = copyin(uap->parms, &kargs.largs,
140 sizeof(struct i386_ldt_args))) != 0)
142 if (kargs.largs.num > MAX_LD || kargs.largs.num <= 0)
145 case I386_GET_XFPUSTATE:
146 if ((error = copyin(uap->parms, &kargs.xfpu,
147 sizeof(struct i386_get_xfpustate))) != 0)
156 error = i386_get_ldt(td, &kargs.largs);
159 if (kargs.largs.descs != NULL) {
160 lp = (union descriptor *)malloc(
161 kargs.largs.num * sizeof(union descriptor),
163 error = copyin(kargs.largs.descs, lp,
164 kargs.largs.num * sizeof(union descriptor));
166 error = i386_set_ldt(td, &kargs.largs, lp);
169 error = i386_set_ldt(td, &kargs.largs, NULL);
172 case I386_GET_IOPERM:
173 error = i386_get_ioperm(td, &kargs.iargs);
175 error = copyout(&kargs.iargs, uap->parms,
176 sizeof(struct i386_ioperm_args));
178 case I386_SET_IOPERM:
179 error = i386_set_ioperm(td, &kargs.iargs);
182 error = vm86_sysarch(td, uap->parms);
184 case I386_GET_FSBASE:
185 sdp = &td->td_pcb->pcb_fsd;
186 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
187 error = copyout(&base, uap->parms, sizeof(base));
189 case I386_SET_FSBASE:
190 error = copyin(uap->parms, &base, sizeof(base));
193 * Construct a descriptor and store it in the pcb for
194 * the next context switch. Also store it in the gdt
195 * so that the load of tf_fs into %fs will activate it
196 * at return to userland.
198 sd.sd_lobase = base & 0xffffff;
199 sd.sd_hibase = (base >> 24) & 0xff;
200 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
202 sd.sd_type = SDT_MEMRWA;
209 td->td_pcb->pcb_fsd = sd;
210 PCPU_GET(fsgs_gdt)[0] = sd;
212 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
215 case I386_GET_GSBASE:
216 sdp = &td->td_pcb->pcb_gsd;
217 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
218 error = copyout(&base, uap->parms, sizeof(base));
220 case I386_SET_GSBASE:
221 error = copyin(uap->parms, &base, sizeof(base));
224 * Construct a descriptor and store it in the pcb for
225 * the next context switch. Also store it in the gdt
226 * because we have to do a load_gs() right now.
228 sd.sd_lobase = base & 0xffffff;
229 sd.sd_hibase = (base >> 24) & 0xff;
231 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
233 sd.sd_type = SDT_MEMRWA;
240 td->td_pcb->pcb_gsd = sd;
241 PCPU_GET(fsgs_gdt)[1] = sd;
243 load_gs(GSEL(GUGS_SEL, SEL_UPL));
246 case I386_GET_XFPUSTATE:
247 if (kargs.xfpu.len > cpu_max_ext_state_size -
248 sizeof(union savefpu))
251 error = copyout((char *)(get_pcb_user_save_td(td) + 1),
252 kargs.xfpu.addr, kargs.xfpu.len);
262 i386_extend_pcb(struct thread *td)
267 struct soft_segment_descriptor ssd = {
268 0, /* segment base address (overwritten) */
269 ctob(IOPAGES + 1) - 1, /* length */
270 SDT_SYS386TSS, /* segment type */
271 0, /* priority level */
272 1, /* descriptor present */
274 0, /* default 32 size */
278 ext = (struct pcb_ext *)kmem_malloc(kernel_arena, ctob(IOPAGES+1),
280 /* -16 is so we can convert a trapframe into vm86trapframe inplace */
281 ext->ext_tss.tss_esp0 = td->td_kstack + ctob(KSTACK_PAGES) -
282 sizeof(struct pcb) - 16;
283 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
285 * The last byte of the i/o map must be followed by an 0xff byte.
286 * We arbitrarily allocate 16 bytes here, to keep the starting
287 * address on a doubleword boundary.
289 offset = PAGE_SIZE - 16;
290 ext->ext_tss.tss_ioopt =
291 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16;
292 ext->ext_iomap = (caddr_t)ext + offset;
293 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32;
295 addr = (u_long *)ext->ext_vm86.vm86_intmap;
296 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++)
299 ssd.ssd_base = (unsigned)&ext->ext_tss;
300 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext);
301 ssdtosd(&ssd, &ext->ext_tssd);
303 KASSERT(td == curthread, ("giving TSS to !curthread"));
304 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!"));
306 /* Switch to the new TSS. */
308 td->td_pcb->pcb_ext = ext;
309 PCPU_SET(private_tss, 1);
310 *PCPU_GET(tss_gdt) = ext->ext_tssd;
311 ltr(GSEL(GPROC0_SEL, SEL_KPL));
318 i386_set_ioperm(td, uap)
320 struct i386_ioperm_args *uap;
325 if ((error = priv_check(td, PRIV_IO)) != 0)
327 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
331 * While this is restricted to root, we should probably figure out
332 * whether any other driver is using this i/o address, as so not to
333 * cause confusion. This probably requires a global 'usage registry'.
336 if (td->td_pcb->pcb_ext == 0)
337 if ((error = i386_extend_pcb(td)) != 0)
339 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
341 if (uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
344 for (i = uap->start; i < uap->start + uap->length; i++) {
346 iomap[i >> 3] &= ~(1 << (i & 7));
348 iomap[i >> 3] |= (1 << (i & 7));
354 i386_get_ioperm(td, uap)
356 struct i386_ioperm_args *uap;
361 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
364 if (td->td_pcb->pcb_ext == 0) {
369 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
372 state = (iomap[i >> 3] >> (i & 7)) & 1;
373 uap->enable = !state;
376 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
377 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
387 * Update the GDT entry pointing to the LDT to point to the LDT of the
388 * current process. Manage dt_lock holding/unholding autonomously.
391 set_user_ldt(struct mdproc *mdp)
393 struct proc_ldt *pldt;
397 if (!mtx_owned(&dt_lock)) {
398 mtx_lock_spin(&dt_lock);
404 gdt[PCPU_GET(cpuid) * NGDT + GUSERLDT_SEL].sd = pldt->ldt_sd;
406 gdt[GUSERLDT_SEL].sd = pldt->ldt_sd;
408 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
409 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL));
411 mtx_unlock_spin(&dt_lock);
416 set_user_ldt_rv(struct vmspace *vmsp)
421 if (vmsp != td->td_proc->p_vmspace)
424 set_user_ldt(&td->td_proc->p_md);
429 * dt_lock must be held. Returns with dt_lock held.
432 user_ldt_alloc(struct mdproc *mdp, int len)
434 struct proc_ldt *pldt, *new_ldt;
436 mtx_assert(&dt_lock, MA_OWNED);
437 mtx_unlock_spin(&dt_lock);
438 new_ldt = malloc(sizeof(struct proc_ldt),
439 M_SUBPROC, M_WAITOK);
441 new_ldt->ldt_len = len = NEW_MAX_LD(len);
442 new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
443 len * sizeof(union descriptor), M_WAITOK);
444 new_ldt->ldt_refcnt = 1;
445 new_ldt->ldt_active = 0;
447 mtx_lock_spin(&dt_lock);
448 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base;
449 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1;
450 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd);
452 if ((pldt = mdp->md_ldt) != NULL) {
453 if (len > pldt->ldt_len)
455 bcopy(pldt->ldt_base, new_ldt->ldt_base,
456 len * sizeof(union descriptor));
458 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt));
464 * Must be called with dt_lock held. Returns with dt_lock unheld.
467 user_ldt_free(struct thread *td)
469 struct mdproc *mdp = &td->td_proc->p_md;
470 struct proc_ldt *pldt;
472 mtx_assert(&dt_lock, MA_OWNED);
473 if ((pldt = mdp->md_ldt) == NULL) {
474 mtx_unlock_spin(&dt_lock);
478 if (td == curthread) {
480 PCPU_SET(currentldt, _default_ldt);
484 user_ldt_deref(pldt);
488 user_ldt_deref(struct proc_ldt *pldt)
491 mtx_assert(&dt_lock, MA_OWNED);
492 if (--pldt->ldt_refcnt == 0) {
493 mtx_unlock_spin(&dt_lock);
494 kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
495 pldt->ldt_len * sizeof(union descriptor));
496 free(pldt, M_SUBPROC);
498 mtx_unlock_spin(&dt_lock);
502 * Note for the authors of compat layers (linux, etc): copyout() in
503 * the function below is not a problem since it presents data in
504 * arch-specific format (i.e. i386-specific in this case), not in
505 * the OS-specific one.
508 i386_get_ldt(td, uap)
510 struct i386_ldt_args *uap;
513 struct proc_ldt *pldt;
515 union descriptor *lp;
518 printf("i386_get_ldt: start=%d num=%d descs=%p\n",
519 uap->start, uap->num, (void *)uap->descs);
522 mtx_lock_spin(&dt_lock);
523 if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
524 nldt = pldt->ldt_len;
525 lp = &((union descriptor *)(pldt->ldt_base))[uap->start];
526 mtx_unlock_spin(&dt_lock);
527 num = min(uap->num, nldt);
529 mtx_unlock_spin(&dt_lock);
530 nldt = sizeof(ldt)/sizeof(ldt[0]);
531 num = min(uap->num, nldt);
532 lp = &ldt[uap->start];
535 if ((uap->start > (unsigned int)nldt) ||
536 ((unsigned int)num > (unsigned int)nldt) ||
537 ((unsigned int)(uap->start + num) > (unsigned int)nldt))
540 error = copyout(lp, uap->descs, num * sizeof(union descriptor));
542 td->td_retval[0] = num;
548 i386_set_ldt(td, uap, descs)
550 struct i386_ldt_args *uap;
551 union descriptor *descs;
555 struct mdproc *mdp = &td->td_proc->p_md;
556 struct proc_ldt *pldt;
557 union descriptor *dp;
560 printf("i386_set_ldt: start=%d num=%d descs=%p\n",
561 uap->start, uap->num, (void *)uap->descs);
565 /* Free descriptors */
566 if (uap->start == 0 && uap->num == 0) {
568 * Treat this as a special case, so userland needn't
569 * know magic number NLDT.
572 uap->num = MAX_LD - NLDT;
576 mtx_lock_spin(&dt_lock);
577 if ((pldt = mdp->md_ldt) == NULL ||
578 uap->start >= pldt->ldt_len) {
579 mtx_unlock_spin(&dt_lock);
582 largest_ld = uap->start + uap->num;
583 if (largest_ld > pldt->ldt_len)
584 largest_ld = pldt->ldt_len;
585 i = largest_ld - uap->start;
586 bzero(&((union descriptor *)(pldt->ldt_base))[uap->start],
587 sizeof(union descriptor) * i);
588 mtx_unlock_spin(&dt_lock);
592 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
593 /* verify range of descriptors to modify */
594 largest_ld = uap->start + uap->num;
595 if (uap->start >= MAX_LD || largest_ld > MAX_LD) {
600 /* Check descriptors for access violations */
601 for (i = 0; i < uap->num; i++) {
604 switch (dp->sd.sd_type) {
605 case SDT_SYSNULL: /* system null */
608 case SDT_SYS286TSS: /* system 286 TSS available */
609 case SDT_SYSLDT: /* system local descriptor table */
610 case SDT_SYS286BSY: /* system 286 TSS busy */
611 case SDT_SYSTASKGT: /* system task gate */
612 case SDT_SYS286IGT: /* system 286 interrupt gate */
613 case SDT_SYS286TGT: /* system 286 trap gate */
614 case SDT_SYSNULL2: /* undefined by Intel */
615 case SDT_SYS386TSS: /* system 386 TSS available */
616 case SDT_SYSNULL3: /* undefined by Intel */
617 case SDT_SYS386BSY: /* system 386 TSS busy */
618 case SDT_SYSNULL4: /* undefined by Intel */
619 case SDT_SYS386IGT: /* system 386 interrupt gate */
620 case SDT_SYS386TGT: /* system 386 trap gate */
621 case SDT_SYS286CGT: /* system 286 call gate */
622 case SDT_SYS386CGT: /* system 386 call gate */
623 /* I can't think of any reason to allow a user proc
624 * to create a segment of these types. They are
630 /* memory segment types */
631 case SDT_MEMEC: /* memory execute only conforming */
632 case SDT_MEMEAC: /* memory execute only accessed conforming */
633 case SDT_MEMERC: /* memory execute read conforming */
634 case SDT_MEMERAC: /* memory execute read accessed conforming */
635 /* Must be "present" if executable and conforming. */
636 if (dp->sd.sd_p == 0)
639 case SDT_MEMRO: /* memory read only */
640 case SDT_MEMROA: /* memory read only accessed */
641 case SDT_MEMRW: /* memory read write */
642 case SDT_MEMRWA: /* memory read write accessed */
643 case SDT_MEMROD: /* memory read only expand dwn limit */
644 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
645 case SDT_MEMRWD: /* memory read write expand dwn limit */
646 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
647 case SDT_MEME: /* memory execute only */
648 case SDT_MEMEA: /* memory execute only accessed */
649 case SDT_MEMER: /* memory execute read */
650 case SDT_MEMERA: /* memory execute read accessed */
657 /* Only user (ring-3) descriptors may be present. */
658 if ((dp->sd.sd_p != 0) && (dp->sd.sd_dpl != SEL_UPL))
662 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
663 /* Allocate a free slot */
664 mtx_lock_spin(&dt_lock);
665 if ((pldt = mdp->md_ldt) == NULL) {
666 if ((error = i386_ldt_grow(td, NLDT + 1))) {
667 mtx_unlock_spin(&dt_lock);
674 * start scanning a bit up to leave room for NVidia and
675 * Wine, which still user the "Blat" method of allocation.
677 dp = &((union descriptor *)(pldt->ldt_base))[NLDT];
678 for (i = NLDT; i < pldt->ldt_len; ++i) {
679 if (dp->sd.sd_type == SDT_SYSNULL)
683 if (i >= pldt->ldt_len) {
684 if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) {
685 mtx_unlock_spin(&dt_lock);
691 error = i386_set_ldt_data(td, i, 1, descs);
692 mtx_unlock_spin(&dt_lock);
694 largest_ld = uap->start + uap->num;
695 mtx_lock_spin(&dt_lock);
696 if (!(error = i386_ldt_grow(td, largest_ld))) {
697 error = i386_set_ldt_data(td, uap->start, uap->num,
700 mtx_unlock_spin(&dt_lock);
703 td->td_retval[0] = uap->start;
708 i386_set_ldt_data(struct thread *td, int start, int num,
709 union descriptor *descs)
711 struct mdproc *mdp = &td->td_proc->p_md;
712 struct proc_ldt *pldt = mdp->md_ldt;
714 mtx_assert(&dt_lock, MA_OWNED);
718 &((union descriptor *)(pldt->ldt_base))[start],
719 num * sizeof(union descriptor));
724 i386_ldt_grow(struct thread *td, int len)
726 struct mdproc *mdp = &td->td_proc->p_md;
727 struct proc_ldt *new_ldt, *pldt;
728 caddr_t old_ldt_base = NULL_LDT_BASE;
731 mtx_assert(&dt_lock, MA_OWNED);
738 /* Allocate a user ldt. */
739 if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) {
740 new_ldt = user_ldt_alloc(mdp, len);
746 if (new_ldt->ldt_len <= pldt->ldt_len) {
748 * We just lost the race for allocation, so
749 * free the new object and return.
751 mtx_unlock_spin(&dt_lock);
752 kmem_free(kernel_arena,
753 (vm_offset_t)new_ldt->ldt_base,
754 new_ldt->ldt_len * sizeof(union descriptor));
755 free(new_ldt, M_SUBPROC);
756 mtx_lock_spin(&dt_lock);
761 * We have to substitute the current LDT entry for
762 * curproc with the new one since its size grew.
764 old_ldt_base = pldt->ldt_base;
765 old_ldt_len = pldt->ldt_len;
766 pldt->ldt_sd = new_ldt->ldt_sd;
767 pldt->ldt_base = new_ldt->ldt_base;
768 pldt->ldt_len = new_ldt->ldt_len;
770 mdp->md_ldt = pldt = new_ldt;
773 * Signal other cpus to reload ldt. We need to unlock dt_lock
774 * here because other CPU will contest on it since their
775 * curthreads won't hold the lock and will block when trying
778 mtx_unlock_spin(&dt_lock);
779 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv,
780 NULL, td->td_proc->p_vmspace);
782 set_user_ldt(&td->td_proc->p_md);
783 mtx_unlock_spin(&dt_lock);
785 if (old_ldt_base != NULL_LDT_BASE) {
786 kmem_free(kernel_arena, (vm_offset_t)old_ldt_base,
787 old_ldt_len * sizeof(union descriptor));
788 free(new_ldt, M_SUBPROC);
790 mtx_lock_spin(&dt_lock);