linux_machdep.c revision 1.6 1 /* $NetBSD: linux_machdep.c,v 1.6 2001/03/18 11:31:44 manu Exp $ */
2
3 /*-
4 * Copyright (c) 1995, 2000, 2001 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Frank van der Linden and Emmanuel Dreyfus.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/signalvar.h>
42 #include <sys/kernel.h>
43 #include <sys/map.h>
44 #include <sys/proc.h>
45 #include <sys/user.h>
46 #include <sys/buf.h>
47 #include <sys/reboot.h>
48 #include <sys/conf.h>
49 #include <sys/exec.h>
50 #include <sys/file.h>
51 #include <sys/callout.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/msgbuf.h>
55 #include <sys/mount.h>
56 #include <sys/vnode.h>
57 #include <sys/device.h>
58 #include <sys/syscallargs.h>
59 #include <sys/filedesc.h>
60 #include <sys/exec_elf.h>
61 #include <sys/disklabel.h>
62 #include <sys/ioctl.h>
63 #include <miscfs/specfs/specdev.h>
64
65 #include <compat/linux/common/linux_types.h>
66 #include <compat/linux/common/linux_signal.h>
67 #include <compat/linux/common/linux_util.h>
68 #include <compat/linux/common/linux_ioctl.h>
69 #include <compat/linux/common/linux_hdio.h>
70 #include <compat/linux/common/linux_exec.h>
71 #include <compat/linux/common/linux_machdep.h>
72
73 #include <compat/linux/linux_syscallargs.h>
74
75 #include <machine/cpu.h>
76 #include <machine/psl.h>
77 #include <machine/reg.h>
78 #include <machine/vmparam.h>
79
80 /*
81 * To see whether wscons is configured (for virtual console ioctl calls).
82 */
83 #if defined(_KERNEL) && !defined(_LKM)
84 #include "wsdisplay.h"
85 #endif
86 #if (NWSDISPLAY > 0)
87 #include <dev/wscons/wsconsio.h>
88 #include <dev/wscons/wsdisplay_usl_io.h>
89 #if defined(_KERNEL) && !defined(_LKM)
90 #endif
91 #endif
92
93 /*
94 * Set set up registers on exec.
95 * XXX not used at the moment since in sys/kern/exec_conf, LINUX_COMPAT
96 * entry uses NetBSD's native setregs instead of linux_setregs
97 */
98 void
99 linux_setregs(p, pack, stack)
100 struct proc *p;
101 struct exec_package *pack;
102 u_long stack;
103 {
104 setregs(p, pack, stack);
105 }
106
107 /*
108 * Send an interrupt to process.
109 *
110 * Adapted from arch/powerpc/powerpc/sig_machdep.c:sendsig and
111 * compat/linux/arch/i386/linux_machdep.c:linux_sendsig
112 *
113 * XXX Does not work well yet with RT signals
114 *
115 */
116
117 void
118 linux_sendsig(catcher, sig, mask, code) /* XXX Check me */
119 sig_t catcher;
120 int sig;
121 sigset_t *mask;
122 u_long code;
123 {
124 struct proc *p = curproc;
125 struct trapframe *tf;
126 struct linux_sigregs *fp, frame;
127 struct linux_pt_regs linux_regs;
128 struct linux_sigcontext sc;
129 int onstack;
130 int i;
131
132 tf = trapframe(p);
133
134 /*
135 * Do we need to jump onto the signal stack?
136 */
137 onstack =
138 (p->p_sigctx.ps_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
139 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
140
141 /*
142 * Signal stack is broken (see at the end of linux_sigreturn), so we do
143 * not use it yet. XXX fix this.
144 */
145 onstack=0;
146
147 /*
148 * Allocate space for the signal handler context.
149 */
150 if (onstack) {
151 fp = (struct linux_sigregs *)
152 ((caddr_t)p->p_sigctx.ps_sigstk.ss_sp +
153 p->p_sigctx.ps_sigstk.ss_size);
154 } else {
155 fp = (struct linux_sigregs *)tf->fixreg[1];
156 }
157 fp = (struct linux_sigregs *)((int)(fp - 1) & ~0xf);
158
159 /*
160 * Prepare a sigcontext for later.
161 */
162 sc.lsignal = (int)native_to_linux_sig[sig];
163 sc.lhandler = (unsigned long)catcher;
164 native_to_linux_old_sigset(mask, &sc.lmask);
165 sc.lregs = (struct linux_pt_regs*)fp;
166
167 /*
168 * Setup the signal stack frame as Linux does it in
169 * arch/ppc/kernel/signal.c:setup_frame()
170 *
171 * Save register context.
172 */
173 for (i = 0; i < 32; i++)
174 linux_regs.lgpr[i] = tf->fixreg[i];
175 linux_regs.lnip = tf->srr0;
176 linux_regs.lmsr = tf->srr1;
177 linux_regs.lorig_gpr3 = tf->fixreg[3]; /* XXX Is that right? */
178 linux_regs.lctr = tf->ctr;
179 linux_regs.llink = tf->lr;
180 linux_regs.lxer = tf->xer;
181 linux_regs.lccr = tf->cr;
182 linux_regs.lmq = 0; /* Unused, 601 only */
183 linux_regs.ltrap = 0; /* XXX What is ltrap counterpart in NetBSD ? */
184 linux_regs.ldar = tf->dar;
185 linux_regs.ldsisr = tf->dsisr;
186 linux_regs.lresult = tf->exc;
187 memcpy(&frame.lgp_regs, &linux_regs, sizeof(frame.lgp_regs));
188
189 /*
190 * NetBSD does not uses the FPU in the kernel, so there is no
191 * need to save floating point register. However, Linux expects
192 * them to be saved on the stack. Therefore we just keep a
193 * gap of zero'ed data where the FP registers should be stored
194 */
195 memset(&frame.lfp_regs, 0, sizeof (frame.lfp_regs));
196
197 /*
198 * Copy Linux's signal trampoline on the user stack It should not
199 * be used, but Linux binaries might expect it to be there.
200 */
201 frame.ltramp[0] = 0x38997777; /* li r0, 0x7777 */
202 frame.ltramp[1] = 0x44000002; /* sc */
203
204 /*
205 * Move it to the user stack
206 * There is a little trick here, about the LINUX_ABIGAP: the
207 * linux_sigreg structure has a 56 int gap to support rs6000/xcoff
208 * binaries. But the Linux kernel seems to do without it, and it
209 * just skip it when building the stack frame. Hence the LINUX_ABIGAP.
210 */
211 if (copyout(&frame, fp, sizeof (frame) - LINUX_ABIGAP) != 0) {
212 /*
213 * Process has trashed its stack; give it an illegal
214 * instructoin to halt it in its tracks.
215 */
216 sigexit(p, SIGILL);
217 /* NOTREACHED */
218 }
219
220 /*
221 * adjust stack pointer after the previous data copy
222 */
223 fp = (struct linux_sigregs *)
224 ((unsigned long)fp - (sizeof (frame) - LINUX_ABIGAP));
225
226 /*
227 * "Mind the gap" Linux expects a gap here.
228 */
229 fp = (struct linux_sigregs *)
230 ((unsigned long)fp - LINUX__SIGNAL_FRAMESIZE);
231
232 /*
233 * Add a sigcontext on the stack
234 */
235 if (copyout(&sc, fp, sizeof (struct linux_sigcontext)) != 0) {
236 /*
237 * Process has trashed its stack; give it an illegal
238 * instructoin to halt it in its tracks.
239 */
240 sigexit(p, SIGILL);
241 /* NOTREACHED */
242 }
243
244 /*
245 * Here, I expected to need a stack pointer adjust after the copy.
246 * Something like this: (unsigned long)fp-=sizeof(struct sigcontext)
247 * But if we do it, the signal handler does not get its arguments as
248 * expected.
249 */
250
251 /*
252 * Set the registers according to how the Linux process expects them
253 */
254 tf->fixreg[1] = (int)fp;
255 tf->lr = (int)catcher;
256 tf->fixreg[3] = (int)sig;
257 tf->fixreg[4] = (int)&fp->lgp_regs;
258 tf->srr0 = (int)p->p_sigctx.ps_sigcode;
259
260 /*
261 * Remember that we're now on the signal stack.
262 */
263 if (onstack)
264 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK;
265 #ifdef DEBUG_LINUX
266 printf("linux_sendsig: exitting. fp=0x%lx\n",(long)fp);
267 #endif
268 }
269
270 /*
271 * System call to cleanup state after a signal
272 * has been taken. Reset signal mask and
273 * stack state from context left by sendsig (above).
274 * Return to previous pc and psl as specified by
275 * context left by sendsig. Check carefully to
276 * make sure that the user has not modified the
277 * psl to gain improper privileges or to cause
278 * a machine fault.
279 *
280 * XXX not tested
281 */
282 int
283 linux_sys_rt_sigreturn(p, v, retval)
284 struct proc *p;
285 void *v;
286 register_t *retval;
287 {
288 struct linux_sys_rt_sigreturn_args /* {
289 syscallarg(struct linux_rt_sigframe *) sfp;
290 } */ *uap = v;
291 struct linux_rt_sigframe *scp, sigframe;
292 struct trapframe *tf;
293 sigset_t mask;
294 int i;
295
296 /*
297 * The trampoline code hands us the context.
298 * It is unsafe to keep track of it ourselves, in the event that a
299 * program jumps out of a signal handler.
300 */
301 scp = SCARG(uap, sfp);
302
303 /*
304 * It seems we need a 16 bytes alignement here (it just works with it,
305 * don't ask me why
306 */
307 scp = (struct linux_rt_sigframe *)((unsigned long)scp & ~0xfUL);
308
309 /*
310 * Get the context from user stack
311 */
312 if (copyin((caddr_t)scp, &sigframe, sizeof(*scp)) != 0)
313 return (EFAULT);
314
315 /*
316 * Grab the signal mask
317 */
318 linux_to_native_sigset(&sigframe.luc.luc_sigmask, &mask);
319
320 /*
321 * Restore register context. XXX need security review
322 */
323 tf = trapframe(p);
324 #ifdef DEBUG_LINUX
325 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n",
326 (unsigned long)tf, (unsigned long)scp);
327 #endif
328
329 if ((sigframe.luc.luc_context.lregs->lmsr & PSL_USERSTATIC) !=
330 (tf->srr1 & PSL_USERSTATIC))
331 return (EINVAL);
332
333 for (i = 0; i < 32; i++)
334 tf->fixreg[i] = sigframe.luc.luc_context.lregs->lgpr[i];
335 tf->lr = sigframe.luc.luc_context.lregs->llink;
336 tf->cr = sigframe.luc.luc_context.lregs->lccr;
337 tf->xer = sigframe.luc.luc_context.lregs->lxer;
338 tf->ctr = sigframe.luc.luc_context.lregs->lctr;
339 tf->srr0 = sigframe.luc.luc_context.lregs->lnip;
340 tf->srr1 = sigframe.luc.luc_context.lregs->lmsr;
341 tf->dar = sigframe.luc.luc_context.lregs->ldar;
342 tf->dsisr = sigframe.luc.luc_context.lregs->ldsisr;
343 tf->exc = sigframe.luc.luc_context.lregs->lresult;
344
345 /*
346 * Restore signal stack.
347 *
348 * XXX cannot find the onstack information in Linux sig context.
349 * Is signal stack really supported on Linux?
350 *
351 * It seems to be supported in libc6...
352 */
353 /* if (sc.sc_onstack & SS_ONSTACK)
354 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK;
355 else */
356 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK;
357
358 return (EJUSTRETURN);
359 }
360
361
362 /*
363 * The following needs code review for potential security issues
364 */
365 int
366 linux_sys_sigreturn(p, v, retval)
367 struct proc *p;
368 void *v;
369 register_t *retval;
370 {
371 struct linux_sys_sigreturn_args /* {
372 syscallarg(struct linux_sigcontext *) scp;
373 } */ *uap = v;
374 struct linux_sigcontext *scp, context;
375 struct trapframe *tf;
376 sigset_t mask;
377 int i;
378
379 /*
380 * The trampoline code hands us the context.
381 * It is unsafe to keep track of it ourselves, in the event that a
382 * program jumps out of a signal handler.
383 */
384 scp = SCARG(uap, scp);
385
386 /*
387 * It seems we need a 16 bytes alignement here (it just works with it,
388 * don't ask me why
389 */
390 (unsigned long)scp = (unsigned long) scp & ~0xfUL;
391
392 /*
393 * Get the context from user stack
394 */
395 if (copyin(scp, &context, sizeof(struct linux_sigcontext)) != 0)
396 return (EFAULT);
397
398 /*
399 * Restore register context. XXX need security review
400 */
401 tf = trapframe(p);
402 #ifdef DEBUG_LINUX
403 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n",
404 (unsigned long)tf, (unsigned long)scp);
405 #endif
406
407 if ((context.lregs->lmsr & PSL_USERSTATIC) !=
408 (tf->srr1 & PSL_USERSTATIC))
409 return (EINVAL);
410
411 for (i = 0; i < 32; i++)
412 tf->fixreg[i] = context.lregs->lgpr[i];
413 tf->lr = context.lregs->llink;
414 tf->cr = context.lregs->lccr;
415 tf->xer = context.lregs->lxer;
416 tf->ctr = context.lregs->lctr;
417 tf->srr0 = context.lregs->lnip;
418 tf->srr1 = context.lregs->lmsr;
419 tf->dar = context.lregs->ldar;
420 tf->dsisr = context.lregs->ldsisr;
421 tf->exc = context.lregs->lresult;
422
423 /*
424 * Restore signal stack.
425 *
426 * XXX cannot find the onstack information in Linux sig context.
427 * Is signal stack really supported on Linux?
428 */
429 #if 0
430 if (sc.sc_onstack & SS_ONSTACK)
431 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK;
432 else
433 #endif
434 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK;
435
436 /* Restore signal mask. */
437 linux_old_to_native_sigset(&context.lmask, &mask);
438 (void) sigprocmask1(p, SIG_SETMASK, &mask, 0);
439
440 return (EJUSTRETURN);
441 }
442
443
444 int
445 linux_sys_modify_ldt(p, v, retval)
446 struct proc *p;
447 void *v;
448 register_t *retval;
449 {
450 /*
451 * This syscall is not implemented in Linux/PowerPC: we should not
452 * be here
453 */
454 #ifdef DEBUG_LINUX
455 printf("linux_sys_modify_ldt: should not be here.\n");
456 #endif
457 return 0;
458 }
459
460 /*
461 * major device numbers remapping
462 */
463 dev_t
464 linux_fakedev(dev)
465 dev_t dev;
466 {
467 /* XXX write me */
468 return dev;
469 }
470
471 /*
472 * We come here in a last attempt to satisfy a Linux ioctl() call
473 */
474 int
475 linux_machdepioctl(p, v, retval)
476 struct proc *p;
477 void *v;
478 register_t *retval;
479 {
480 struct linux_sys_ioctl_args /* {
481 syscallarg(int) fd;
482 syscallarg(u_long) com;
483 syscallarg(caddr_t) data;
484 } */ *uap = v;
485 struct sys_ioctl_args bia;
486 u_long com;
487
488 SCARG(&bia, fd) = SCARG(uap, fd);
489 SCARG(&bia, data) = SCARG(uap, data);
490 com = SCARG(uap, com);
491
492 switch (com) {
493 default:
494 printf("linux_machdepioctl: invalid ioctl %08lx\n", com);
495 return EINVAL;
496 }
497 SCARG(&bia, com) = com;
498 return sys_ioctl(p, &bia, retval);
499 }
500 /*
501 * Set I/O permissions for a process. Just set the maximum level
502 * right away (ignoring the argument), otherwise we would have
503 * to rely on I/O permission maps, which are not implemented.
504 */
505 int
506 linux_sys_iopl(p, v, retval)
507 struct proc *p;
508 void *v;
509 register_t *retval;
510 {
511 /*
512 * This syscall is not implemented in Linux/PowerPC: we should not be here
513 */
514 #ifdef DEBUG_LINUX
515 printf("linux_sys_iopl: should not be here.\n");
516 #endif
517 return 0;
518 }
519
520 /*
521 * See above. If a root process tries to set access to an I/O port,
522 * just let it have the whole range.
523 */
524 int
525 linux_sys_ioperm(p, v, retval)
526 struct proc *p;
527 void *v;
528 register_t *retval;
529 {
530 /*
531 * This syscall is not implemented in Linux/PowerPC: we should not be here
532 */
533 #ifdef DEBUG_LINUX
534 printf("linux_sys_ioperm: should not be here.\n");
535 #endif
536 return 0;
537 }
538
539 /*
540 * wrapper linux_sys_new_uname() -> linux_sys_uname()
541 */
542 int
543 linux_sys_new_uname(p, v, retval)
544 struct proc *p;
545 void *v;
546 register_t *retval;
547 {
548 return linux_sys_uname(p, v, retval);
549 }
550
551 /*
552 * wrapper linux_sys_new_select() -> linux_sys_select()
553 */
554 int
555 linux_sys_new_select(p, v, retval)
556 struct proc *p;
557 void *v;
558 register_t *retval;
559 {
560 return linux_sys_select(p, v, retval);
561 }
562