xref: /openbsd-src/sys/kern/sys_process.c (revision bcc2e3d7508b5e9be73e81bd83a4cbe5e750bd95)
1 /*	$OpenBSD: sys_process.c,v 1.91 2022/12/21 07:59:02 claudio Exp $	*/
2 /*	$NetBSD: sys_process.c,v 1.55 1996/05/15 06:17:47 tls Exp $	*/
3 
4 /*-
5  * Copyright (c) 1994 Christopher G. Demetriou.  All rights reserved.
6  * Copyright (c) 1982, 1986, 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	from: @(#)sys_process.c	8.1 (Berkeley) 6/10/93
39  */
40 
41 /*
42  * References:
43  *	(1) Bach's "The Design of the UNIX Operating System",
44  *	(2) sys/miscfs/procfs from UCB's 4.4BSD-Lite distribution,
45  *	(3) the "4.4BSD Programmer's Reference Manual" published
46  *		by USENIX and O'Reilly & Associates.
47  * The 4.4BSD PRM does a reasonably good job of documenting what the various
48  * ptrace() requests should actually do, and its text is quoted several times
49  * in this file.
50  */
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/exec.h>
55 #include <sys/proc.h>
56 #include <sys/signalvar.h>
57 #include <sys/errno.h>
58 #include <sys/malloc.h>
59 #include <sys/ptrace.h>
60 #include <sys/uio.h>
61 #include <sys/sched.h>
62 #include <sys/exec_elf.h>
63 
64 #include <sys/mount.h>
65 #include <sys/syscallargs.h>
66 
67 #include <uvm/uvm_extern.h>
68 
69 #include <machine/reg.h>
70 
71 #ifdef PTRACE
72 
73 static inline int	process_checktracestate(struct process *_curpr,
74 			    struct process *_tr, struct proc *_t);
75 static inline struct process *process_tprfind(pid_t _tpid, struct proc **_tp);
76 
77 int	ptrace_ctrl(struct proc *, int, pid_t, caddr_t, int);
78 int	ptrace_ustate(struct proc *, int, pid_t, void *, int, register_t *);
79 int	ptrace_kstate(struct proc *, int, pid_t, void *);
80 
81 int	global_ptrace;	/* permit tracing of not children */
82 
83 
84 /*
85  * Process debugging system call.
86  */
87 int
88 sys_ptrace(struct proc *p, void *v, register_t *retval)
89 {
90 	struct sys_ptrace_args /* {
91 		syscallarg(int) req;
92 		syscallarg(pid_t) pid;
93 		syscallarg(caddr_t) addr;
94 		syscallarg(int) data;
95 	} */ *uap = v;
96 	int req = SCARG(uap, req);
97 	pid_t pid = SCARG(uap, pid);
98 	caddr_t uaddr = SCARG(uap, addr);	/* userspace */
99 	void *kaddr = NULL;			/* kernelspace */
100 	int data = SCARG(uap, data);
101 	union {
102 		struct ptrace_thread_state u_pts;
103 		struct ptrace_io_desc u_piod;
104 		struct ptrace_event u_pe;
105 		struct ptrace_state u_ps;
106 		register_t u_wcookie;
107 	} u;
108 	int size = 0;
109 	enum { NONE, IN, IN_ALLOC, OUT, OUT_ALLOC, IN_OUT } mode;
110 	int kstate = 0;
111 	int error;
112 
113 	*retval = 0;
114 
115 	/* Figure out what sort of copyin/out operations we'll do */
116 	switch (req) {
117 	case PT_TRACE_ME:
118 	case PT_CONTINUE:
119 	case PT_KILL:
120 	case PT_ATTACH:
121 	case PT_DETACH:
122 #ifdef PT_STEP
123 	case PT_STEP:
124 #endif
125 		/* control operations do no copyin/out; dispatch directly */
126 		return ptrace_ctrl(p, req, pid, uaddr, data);
127 
128 	case PT_READ_I:
129 	case PT_READ_D:
130 	case PT_WRITE_I:
131 	case PT_WRITE_D:
132 		mode = NONE;
133 		break;
134 	case PT_IO:
135 		mode = IN_OUT;
136 		size = sizeof u.u_piod;
137 		data = size;	/* suppress the data == size check */
138 		break;
139 	case PT_GET_THREAD_FIRST:
140 		mode = OUT;
141 		size = sizeof u.u_pts;
142 		kstate = 1;
143 		break;
144 	case PT_GET_THREAD_NEXT:
145 		mode = IN_OUT;
146 		size = sizeof u.u_pts;
147 		kstate = 1;
148 		break;
149 	case PT_GET_EVENT_MASK:
150 		mode = OUT;
151 		size = sizeof u.u_pe;
152 		kstate = 1;
153 		break;
154 	case PT_SET_EVENT_MASK:
155 		mode = IN;
156 		size = sizeof u.u_pe;
157 		kstate = 1;
158 		break;
159 	case PT_GET_PROCESS_STATE:
160 		mode = OUT;
161 		size = sizeof u.u_ps;
162 		kstate = 1;
163 		break;
164 	case PT_GETREGS:
165 		mode = OUT_ALLOC;
166 		size = sizeof(struct reg);
167 		break;
168 	case PT_SETREGS:
169 		mode = IN_ALLOC;
170 		size = sizeof(struct reg);
171 		break;
172 #ifdef PT_GETFPREGS
173 	case PT_GETFPREGS:
174 		mode = OUT_ALLOC;
175 		size = sizeof(struct fpreg);
176 		break;
177 #endif
178 #ifdef PT_SETFPREGS
179 	case PT_SETFPREGS:
180 		mode = IN_ALLOC;
181 		size = sizeof(struct fpreg);
182 		break;
183 #endif
184 #ifdef PT_GETXMMREGS
185 	case PT_GETXMMREGS:
186 		mode = OUT_ALLOC;
187 		size = sizeof(struct xmmregs);
188 		break;
189 #endif
190 #ifdef PT_SETXMMREGS
191 	case PT_SETXMMREGS:
192 		mode = IN_ALLOC;
193 		size = sizeof(struct xmmregs);
194 		break;
195 #endif
196 #ifdef PT_WCOOKIE
197 	case PT_WCOOKIE:
198 		mode = OUT;
199 		size = sizeof u.u_wcookie;
200 		data = size;	/* suppress the data == size check */
201 		break;
202 #endif
203 	default:
204 		return EINVAL;
205 	}
206 
207 
208 	/* Now do any copyin()s and allocations in a consistent manner */
209 	switch (mode) {
210 	case NONE:
211 		kaddr = uaddr;
212 		break;
213 	case IN:
214 	case IN_OUT:
215 	case OUT:
216 		KASSERT(size <= sizeof u);
217 		if (data != size)
218 			return EINVAL;
219 		if (mode == OUT)
220 			memset(&u, 0, size);
221 		else { /* IN or IN_OUT */
222 			if ((error = copyin(uaddr, &u, size)))
223 				return error;
224 		}
225 		kaddr = &u;
226 		break;
227 	case IN_ALLOC:
228 		kaddr = malloc(size, M_TEMP, M_WAITOK);
229 		if ((error = copyin(uaddr, kaddr, size))) {
230 			free(kaddr, M_TEMP, size);
231 			return error;
232 		}
233 		break;
234 	case OUT_ALLOC:
235 		kaddr = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
236 		break;
237 	}
238 
239 	if (kstate)
240 		error = ptrace_kstate(p, req, pid, kaddr);
241 	else
242 		error = ptrace_ustate(p, req, pid, kaddr, data, retval);
243 
244 	/* Do any copyout()s and frees */
245 	if (error == 0) {
246 		switch (mode) {
247 		case NONE:
248 		case IN:
249 		case IN_ALLOC:
250 			break;
251 		case IN_OUT:
252 		case OUT:
253 			error = copyout(&u, uaddr, size);
254 			if (req == PT_IO) {
255 				/* historically, errors here are ignored */
256 				error = 0;
257 			}
258 			break;
259 		case OUT_ALLOC:
260 			error = copyout(kaddr, uaddr, size);
261 			break;
262 		}
263 	}
264 
265 	if (mode == IN_ALLOC || mode == OUT_ALLOC)
266 		free(kaddr, M_TEMP, size);
267 	return error;
268 }
269 
270 /*
271  * ptrace control requests: attach, detach, continue, kill, single-step, etc
272  */
273 int
274 ptrace_ctrl(struct proc *p, int req, pid_t pid, caddr_t addr, int data)
275 {
276 	struct proc *t;				/* target thread */
277 	struct process *tr;			/* target process */
278 	int error = 0;
279 	int s;
280 
281 	switch (req) {
282 	case PT_TRACE_ME:
283 		/* Just set the trace flag. */
284 		tr = p->p_p;
285 		if (ISSET(tr->ps_flags, PS_TRACED))
286 			return EBUSY;
287 		atomic_setbits_int(&tr->ps_flags, PS_TRACED);
288 		tr->ps_oppid = tr->ps_pptr->ps_pid;
289 		if (tr->ps_ptstat == NULL)
290 			tr->ps_ptstat = malloc(sizeof(*tr->ps_ptstat),
291 			    M_SUBPROC, M_WAITOK);
292 		memset(tr->ps_ptstat, 0, sizeof(*tr->ps_ptstat));
293 		return 0;
294 
295 	/* calls that only operate on the PID */
296 	case PT_KILL:
297 	case PT_ATTACH:
298 	case PT_DETACH:
299 		/* Find the process we're supposed to be operating on. */
300 		if ((tr = prfind(pid)) == NULL) {
301 			error = ESRCH;
302 			goto fail;
303 		}
304 		t = TAILQ_FIRST(&tr->ps_threads);
305 		break;
306 
307 	/* calls that accept a PID or a thread ID */
308 	case PT_CONTINUE:
309 #ifdef PT_STEP
310 	case PT_STEP:
311 #endif
312 		if ((tr = process_tprfind(pid, &t)) == NULL) {
313 			error = ESRCH;
314 			goto fail;
315 		}
316 		break;
317 	}
318 
319 	/* Check permissions/state */
320 	if (req != PT_ATTACH) {
321 		/* Check that the data is a valid signal number or zero. */
322 		if (req != PT_KILL && (data < 0 || data >= NSIG)) {
323 			error = EINVAL;
324 			goto fail;
325 		}
326 
327 		/* Most operations require the target to already be traced */
328 		if ((error = process_checktracestate(p->p_p, tr, t)))
329 			goto fail;
330 
331 		/* Do single-step fixup if needed. */
332 		FIX_SSTEP(t);
333 	} else {
334 		/*
335 		 * PT_ATTACH is the opposite; you can't attach to a process if:
336 		 *	(1) it's the process that's doing the attaching,
337 		 */
338 		if (tr == p->p_p) {
339 			error = EINVAL;
340 			goto fail;
341 		}
342 
343 		/*
344 		 *	(2) it's a system process
345 		 */
346 		if (ISSET(tr->ps_flags, PS_SYSTEM)) {
347 			error = EPERM;
348 			goto fail;
349 		}
350 
351 		/*
352 		 *	(3) it's already being traced, or
353 		 */
354 		if (ISSET(tr->ps_flags, PS_TRACED)) {
355 			error = EBUSY;
356 			goto fail;
357 		}
358 
359 		/*
360 		 *	(4) it's in the middle of execve(2)
361 		 */
362 		if (ISSET(tr->ps_flags, PS_INEXEC)) {
363 			error = EAGAIN;
364 			goto fail;
365 		}
366 
367 		/*
368 		 *	(5) it's not owned by you, or the last exec
369 		 *	    gave us setuid/setgid privs (unless
370 		 *	    you're root), or...
371 		 *
372 		 *      [Note: once PS_SUGID or PS_SUGIDEXEC gets set in
373 		 *	execve(), they stay set until the process does
374 		 *	another execve().  Hence this prevents a setuid
375 		 *	process which revokes its special privileges using
376 		 *	setuid() from being traced.  This is good security.]
377 		 */
378 		if ((tr->ps_ucred->cr_ruid != p->p_ucred->cr_ruid ||
379 		    ISSET(tr->ps_flags, PS_SUGIDEXEC | PS_SUGID)) &&
380 		    (error = suser(p)) != 0)
381 			goto fail;
382 
383 		/*
384 		 * 	(5.5) it's not a child of the tracing process.
385 		 */
386 		if (global_ptrace == 0 && !inferior(tr, p->p_p) &&
387 		    (error = suser(p)) != 0)
388 			goto fail;
389 
390 		/*
391 		 *	(6) ...it's init, which controls the security level
392 		 *	    of the entire system, and the system was not
393 		 *          compiled with permanently insecure mode turned
394 		 *	    on.
395 		 */
396 		if ((tr->ps_pid == 1) && (securelevel > -1)) {
397 			error = EPERM;
398 			goto fail;
399 		}
400 
401 		/*
402 		 *	(7) it's an ancestor of the current process and
403 		 *	    not init (because that would create a loop in
404 		 *	    the process graph).
405 		 */
406 		if (tr->ps_pid != 1 && inferior(p->p_p, tr)) {
407 			error = EINVAL;
408 			goto fail;
409 		}
410 	}
411 
412 	switch (req) {
413 
414 #ifdef PT_STEP
415 	case PT_STEP:
416 		/*
417 		 * From the 4.4BSD PRM:
418 		 * "Execution continues as in request PT_CONTINUE; however
419 		 * as soon as possible after execution of at least one
420 		 * instruction, execution stops again. [ ... ]"
421 		 */
422 #endif
423 	case PT_CONTINUE:
424 		/*
425 		 * From the 4.4BSD PRM:
426 		 * "The data argument is taken as a signal number and the
427 		 * child's execution continues at location addr as if it
428 		 * incurred that signal.  Normally the signal number will
429 		 * be either 0 to indicate that the signal that caused the
430 		 * stop should be ignored, or that value fetched out of
431 		 * the process's image indicating which signal caused
432 		 * the stop.  If addr is (int *)1 then execution continues
433 		 * from where it stopped."
434 		 */
435 
436 		if (pid < THREAD_PID_OFFSET && tr->ps_single)
437 			t = tr->ps_single;
438 
439 		/* If the address parameter is not (int *)1, set the pc. */
440 		if ((int *)addr != (int *)1)
441 			if ((error = process_set_pc(t, addr)) != 0)
442 				goto fail;
443 
444 #ifdef PT_STEP
445 		/*
446 		 * Arrange for a single-step, if that's requested and possible.
447 		 */
448 		error = process_sstep(t, req == PT_STEP);
449 		if (error)
450 			goto fail;
451 #endif
452 		goto sendsig;
453 
454 	case PT_DETACH:
455 		/*
456 		 * From the 4.4BSD PRM:
457 		 * "The data argument is taken as a signal number and the
458 		 * child's execution continues at location addr as if it
459 		 * incurred that signal.  Normally the signal number will
460 		 * be either 0 to indicate that the signal that caused the
461 		 * stop should be ignored, or that value fetched out of
462 		 * the process's image indicating which signal caused
463 		 * the stop.  If addr is (int *)1 then execution continues
464 		 * from where it stopped."
465 		 */
466 
467 		if (pid < THREAD_PID_OFFSET && tr->ps_single)
468 			t = tr->ps_single;
469 
470 #ifdef PT_STEP
471 		/*
472 		 * Stop single stepping.
473 		 */
474 		error = process_sstep(t, 0);
475 		if (error)
476 			goto fail;
477 #endif
478 
479 		process_untrace(tr);
480 		atomic_clearbits_int(&tr->ps_flags, PS_WAITED);
481 
482 	sendsig:
483 		memset(tr->ps_ptstat, 0, sizeof(*tr->ps_ptstat));
484 
485 		/* Finally, deliver the requested signal (or none). */
486 		if (t->p_stat == SSTOP) {
487 			tr->ps_xsig = data;
488 			SCHED_LOCK(s);
489 			setrunnable(t);
490 			SCHED_UNLOCK(s);
491 		} else {
492 			if (data != 0)
493 				psignal(t, data);
494 		}
495 		break;
496 
497 	case PT_KILL:
498 		if (pid < THREAD_PID_OFFSET && tr->ps_single)
499 			t = tr->ps_single;
500 
501 		/* just send the process a KILL signal. */
502 		data = SIGKILL;
503 		goto sendsig;	/* in PT_CONTINUE, above. */
504 
505 	case PT_ATTACH:
506 		/*
507 		 * As was done in procfs:
508 		 * Go ahead and set the trace flag.
509 		 * Save the old parent (it's reset in
510 		 *   _DETACH, and also in kern_exit.c:wait4()
511 		 * Reparent the process so that the tracing
512 		 *   proc gets to see all the action.
513 		 * Stop the target.
514 		 */
515 		atomic_setbits_int(&tr->ps_flags, PS_TRACED);
516 		tr->ps_oppid = tr->ps_pptr->ps_pid;
517 		process_reparent(tr, p->p_p);
518 		if (tr->ps_ptstat == NULL)
519 			tr->ps_ptstat = malloc(sizeof(*tr->ps_ptstat),
520 			    M_SUBPROC, M_WAITOK);
521 		data = SIGSTOP;
522 		goto sendsig;
523 	default:
524 		KASSERTMSG(0, "%s: unhandled request %d", __func__, req);
525 		break;
526 	}
527 
528 fail:
529 	return error;
530 }
531 
532 /*
533  * ptrace kernel-state requests: thread list, event mask, process state
534  */
535 int
536 ptrace_kstate(struct proc *p, int req, pid_t pid, void *addr)
537 {
538 	struct process *tr;			/* target process */
539 	struct ptrace_event *pe = addr;
540 	int error;
541 
542 	KASSERT((p->p_flag & P_SYSTEM) == 0);
543 
544 	/* Find the process we're supposed to be operating on. */
545 	if ((tr = prfind(pid)) == NULL)
546 		return ESRCH;
547 
548 	if ((error = process_checktracestate(p->p_p, tr, NULL)))
549 		return error;
550 
551 	switch (req) {
552 	case PT_GET_THREAD_FIRST:
553 	case PT_GET_THREAD_NEXT:
554 	      {
555 		struct ptrace_thread_state *pts = addr;
556 		struct proc *t;
557 
558 		if (req == PT_GET_THREAD_NEXT) {
559 			t = tfind(pts->pts_tid - THREAD_PID_OFFSET);
560 			if (t == NULL || ISSET(t->p_flag, P_WEXIT))
561 				return ESRCH;
562 			if (t->p_p != tr)
563 				return EINVAL;
564 			t = TAILQ_NEXT(t, p_thr_link);
565 		} else {
566 			t = TAILQ_FIRST(&tr->ps_threads);
567 		}
568 
569 		if (t == NULL)
570 			pts->pts_tid = -1;
571 		else
572 			pts->pts_tid = t->p_tid + THREAD_PID_OFFSET;
573 		return 0;
574 	      }
575 	}
576 
577 	switch (req) {
578 	case PT_GET_EVENT_MASK:
579 		pe->pe_set_event = tr->ps_ptmask;
580 		break;
581 	case PT_SET_EVENT_MASK:
582 		tr->ps_ptmask = pe->pe_set_event;
583 		break;
584 	case PT_GET_PROCESS_STATE:
585 		if (tr->ps_single)
586 			tr->ps_ptstat->pe_tid =
587 			    tr->ps_single->p_tid + THREAD_PID_OFFSET;
588 		memcpy(addr, tr->ps_ptstat, sizeof *tr->ps_ptstat);
589 		break;
590 	default:
591 		KASSERTMSG(0, "%s: unhandled request %d", __func__, req);
592 		break;
593 	}
594 
595 	return 0;
596 }
597 
598 /*
599  * ptrace user-state requests: memory access, registers, stack cookie
600  */
601 int
602 ptrace_ustate(struct proc *p, int req, pid_t pid, void *addr, int data,
603     register_t *retval)
604 {
605 	struct proc *t;				/* target thread */
606 	struct process *tr;			/* target process */
607 	struct uio uio;
608 	struct iovec iov;
609 	int error, write;
610 	int temp = 0;
611 
612 	KASSERT((p->p_flag & P_SYSTEM) == 0);
613 
614 	/* Accept either PID or TID */
615 	if ((tr = process_tprfind(pid, &t)) == NULL)
616 		return ESRCH;
617 
618 	if ((error = process_checktracestate(p->p_p, tr, t)))
619 		return error;
620 
621 	FIX_SSTEP(t);
622 
623 	/* Now do the operation. */
624 	write = 0;
625 
626 	if ((error = process_checkioperm(p, tr)) != 0)
627 		return error;
628 
629 	switch (req) {
630 	case PT_WRITE_I:		/* XXX no separate I and D spaces */
631 	case PT_WRITE_D:
632 		write = 1;
633 		temp = data;
634 	case PT_READ_I:		/* XXX no separate I and D spaces */
635 	case PT_READ_D:
636 		/* write = 0 done above. */
637 		iov.iov_base = (caddr_t)&temp;
638 		iov.iov_len = sizeof(int);
639 		uio.uio_iov = &iov;
640 		uio.uio_iovcnt = 1;
641 		uio.uio_offset = (off_t)(vaddr_t)addr;
642 		uio.uio_resid = sizeof(int);
643 		uio.uio_segflg = UIO_SYSSPACE;
644 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
645 		uio.uio_procp = p;
646 		error = process_domem(p, tr, &uio, write ? PT_WRITE_I :
647 				PT_READ_I);
648 		if (write == 0)
649 			*retval = temp;
650 		return error;
651 
652 	case PT_IO:
653 	      {
654 		struct ptrace_io_desc *piod = addr;
655 
656 		iov.iov_base = piod->piod_addr;
657 		iov.iov_len = piod->piod_len;
658 		uio.uio_iov = &iov;
659 		uio.uio_iovcnt = 1;
660 		uio.uio_offset = (off_t)(vaddr_t)piod->piod_offs;
661 		uio.uio_resid = piod->piod_len;
662 		uio.uio_segflg = UIO_USERSPACE;
663 		uio.uio_procp = p;
664 		switch (piod->piod_op) {
665 		case PIOD_READ_I:
666 			req = PT_READ_I;
667 			uio.uio_rw = UIO_READ;
668 			break;
669 		case PIOD_READ_D:
670 			req = PT_READ_D;
671 			uio.uio_rw = UIO_READ;
672 			break;
673 		case PIOD_WRITE_I:
674 			req = PT_WRITE_I;
675 			uio.uio_rw = UIO_WRITE;
676 			break;
677 		case PIOD_WRITE_D:
678 			req = PT_WRITE_D;
679 			uio.uio_rw = UIO_WRITE;
680 			break;
681 		case PIOD_READ_AUXV:
682 			req = PT_READ_D;
683 			uio.uio_rw = UIO_READ;
684 			temp = ELF_AUX_WORDS * sizeof(char *);
685 			if (uio.uio_offset > temp)
686 				return EIO;
687 			if (uio.uio_resid > temp - uio.uio_offset)
688 				uio.uio_resid = temp - uio.uio_offset;
689 			piod->piod_len = iov.iov_len = uio.uio_resid;
690 			uio.uio_offset += tr->ps_auxinfo;
691 #ifdef MACHINE_STACK_GROWS_UP
692 			if (uio.uio_offset < (off_t)tr->ps_strings)
693 				return EIO;
694 #else
695 			if (uio.uio_offset > (off_t)tr->ps_strings)
696 				return EIO;
697 			if ((uio.uio_offset + uio.uio_resid) >
698 			    (off_t)tr->ps_strings)
699 				uio.uio_resid = (off_t)tr->ps_strings -
700 				    uio.uio_offset;
701 #endif
702 			break;
703 		default:
704 			return EINVAL;
705 		}
706 		error = process_domem(p, tr, &uio, req);
707 		piod->piod_len -= uio.uio_resid;
708 		return error;
709 	      }
710 
711 	case PT_SETREGS:
712 		return process_write_regs(t, addr);
713 	case PT_GETREGS:
714 		return process_read_regs(t, addr);
715 
716 #ifdef PT_SETFPREGS
717 	case PT_SETFPREGS:
718 		return process_write_fpregs(t, addr);
719 #endif
720 #ifdef PT_SETFPREGS
721 	case PT_GETFPREGS:
722 		return process_read_fpregs(t, addr);
723 #endif
724 #ifdef PT_SETXMMREGS
725 	case PT_SETXMMREGS:
726 		return process_write_xmmregs(t, addr);
727 #endif
728 #ifdef PT_SETXMMREGS
729 	case PT_GETXMMREGS:
730 		return process_read_xmmregs(t, addr);
731 #endif
732 #ifdef PT_WCOOKIE
733 	case PT_WCOOKIE:
734 		*(register_t *)addr = process_get_wcookie(t);
735 		return 0;
736 #endif
737 	default:
738 		KASSERTMSG(0, "%s: unhandled request %d", __func__, req);
739 		break;
740 	}
741 
742 	return 0;
743 }
744 
745 
746 /*
747  * Helper for doing "it could be a PID or TID" lookup.  On failure
748  * returns NULL; on success returns the selected process and sets *tp
749  * to an appropriate thread in that process.
750  */
751 static inline struct process *
752 process_tprfind(pid_t tpid, struct proc **tp)
753 {
754 	if (tpid > THREAD_PID_OFFSET) {
755 		struct proc *t = tfind(tpid - THREAD_PID_OFFSET);
756 
757 		if (t == NULL)
758 			return NULL;
759 		*tp = t;
760 		return t->p_p;
761 	} else {
762 		struct process *tr = prfind(tpid);
763 
764 		if (tr == NULL)
765 			return NULL;
766 		*tp = TAILQ_FIRST(&tr->ps_threads);
767 		return tr;
768 	}
769 }
770 
771 
772 /*
773  * Check whether 'tr' is currently traced by 'curpr' and in a state
774  * to be manipulated.  If 't' is supplied then it must be stopped and
775  * waited for.
776  */
777 static inline int
778 process_checktracestate(struct process *curpr, struct process *tr,
779     struct proc *t)
780 {
781 	/*
782 	 * You can't do what you want to the process if:
783 	 *	(1) It's not being traced at all,
784 	 */
785 	if (!ISSET(tr->ps_flags, PS_TRACED))
786 		return EPERM;
787 
788 	/*
789 	 *	(2) it's not being traced by _you_, or
790 	 */
791 	if (tr->ps_pptr != curpr)
792 		return EBUSY;
793 
794 	/*
795 	 *	(3) it's in the middle of execve(2)
796 	 */
797 	if (ISSET(tr->ps_flags, PS_INEXEC))
798 		return EAGAIN;
799 
800 	/*
801 	 *	(4) if a thread was specified and it's not currently stopped.
802 	 */
803 	if (t != NULL &&
804 	    (t->p_stat != SSTOP || !ISSET(tr->ps_flags, PS_WAITED)))
805 		return EBUSY;
806 
807 	return 0;
808 }
809 
810 
811 /*
812  * Check if a process is allowed to fiddle with the memory of another.
813  *
814  * p = tracer
815  * tr = tracee
816  *
817  * 1.  You can't attach to a process not owned by you or one that has raised
818  *     its privileges.
819  * 1a. ...unless you are root.
820  *
821  * 2.  init is always off-limits because it can control the securelevel.
822  * 2a. ...unless securelevel is permanently set to insecure.
823  *
824  * 3.  Processes that are in the process of doing an exec() are always
825  *     off-limits because of the can of worms they are. Just wait a
826  *     second.
827  */
828 int
829 process_checkioperm(struct proc *p, struct process *tr)
830 {
831 	int error;
832 
833 	if ((tr->ps_ucred->cr_ruid != p->p_ucred->cr_ruid ||
834 	    ISSET(tr->ps_flags, PS_SUGIDEXEC | PS_SUGID)) &&
835 	    (error = suser(p)) != 0)
836 		return (error);
837 
838 	if ((tr->ps_pid == 1) && (securelevel > -1))
839 		return (EPERM);
840 
841 	if (ISSET(tr->ps_flags, PS_INEXEC))
842 		return (EAGAIN);
843 
844 	return (0);
845 }
846 
847 int
848 process_domem(struct proc *curp, struct process *tr, struct uio *uio, int req)
849 {
850 	struct vmspace *vm;
851 	int error;
852 	vaddr_t addr;
853 	vsize_t len;
854 
855 	len = uio->uio_resid;
856 	if (len == 0)
857 		return 0;
858 
859 	if ((error = process_checkioperm(curp, tr)) != 0)
860 		return error;
861 
862 	vm = tr->ps_vmspace;
863 	if ((tr->ps_flags & PS_EXITING) || (vm->vm_refcnt < 1))
864 		return EFAULT;
865 	addr = uio->uio_offset;
866 
867 	uvmspace_addref(vm);
868 
869 	error = uvm_io(&vm->vm_map, uio,
870 	    (uio->uio_rw == UIO_WRITE) ? UVM_IO_FIXPROT : 0);
871 
872 	uvmspace_free(vm);
873 
874 	if (error == 0 && req == PT_WRITE_I)
875 		pmap_proc_iflush(tr, addr, len);
876 
877 	return error;
878 }
879 #endif
880