xref: /freebsd/sys/kern/sys_process.c (revision c2b1d5a3d67455a4dc6ffae7c1c41f74e8d162b0)
1 /*-
2  * Copyright (c) 1994, Sean Eric Fagan
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Sean Eric Fagan.
16  * 4. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/syscallsubr.h>
42 #include <sys/sysent.h>
43 #include <sys/sysproto.h>
44 #include <sys/proc.h>
45 #include <sys/vnode.h>
46 #include <sys/ptrace.h>
47 #include <sys/sx.h>
48 #include <sys/malloc.h>
49 #include <sys/signalvar.h>
50 
51 #include <machine/reg.h>
52 
53 #include <security/audit/audit.h>
54 
55 #include <vm/vm.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_extern.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_kern.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_param.h>
63 
64 #ifdef COMPAT_IA32
65 #include <sys/procfs.h>
66 #include <machine/fpu.h>
67 #include <compat/ia32/ia32_reg.h>
68 
69 struct ptrace_io_desc32 {
70 	int		piod_op;
71 	u_int32_t	piod_offs;
72 	u_int32_t	piod_addr;
73 	u_int32_t	piod_len;
74 };
75 #endif
76 
77 /*
78  * Functions implemented using PROC_ACTION():
79  *
80  * proc_read_regs(proc, regs)
81  *	Get the current user-visible register set from the process
82  *	and copy it into the regs structure (<machine/reg.h>).
83  *	The process is stopped at the time read_regs is called.
84  *
85  * proc_write_regs(proc, regs)
86  *	Update the current register set from the passed in regs
87  *	structure.  Take care to avoid clobbering special CPU
88  *	registers or privileged bits in the PSL.
89  *	Depending on the architecture this may have fix-up work to do,
90  *	especially if the IAR or PCW are modified.
91  *	The process is stopped at the time write_regs is called.
92  *
93  * proc_read_fpregs, proc_write_fpregs
94  *	deal with the floating point register set, otherwise as above.
95  *
96  * proc_read_dbregs, proc_write_dbregs
97  *	deal with the processor debug register set, otherwise as above.
98  *
99  * proc_sstep(proc)
100  *	Arrange for the process to trap after executing a single instruction.
101  */
102 
103 #define	PROC_ACTION(action) do {					\
104 	int error;							\
105 									\
106 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
107 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
108 		error = EIO;						\
109 	else								\
110 		error = (action);					\
111 	return (error);							\
112 } while(0)
113 
114 int
115 proc_read_regs(struct thread *td, struct reg *regs)
116 {
117 
118 	PROC_ACTION(fill_regs(td, regs));
119 }
120 
121 int
122 proc_write_regs(struct thread *td, struct reg *regs)
123 {
124 
125 	PROC_ACTION(set_regs(td, regs));
126 }
127 
128 int
129 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
130 {
131 
132 	PROC_ACTION(fill_dbregs(td, dbregs));
133 }
134 
135 int
136 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
137 {
138 
139 	PROC_ACTION(set_dbregs(td, dbregs));
140 }
141 
142 /*
143  * Ptrace doesn't support fpregs at all, and there are no security holes
144  * or translations for fpregs, so we can just copy them.
145  */
146 int
147 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
148 {
149 
150 	PROC_ACTION(fill_fpregs(td, fpregs));
151 }
152 
153 int
154 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
155 {
156 
157 	PROC_ACTION(set_fpregs(td, fpregs));
158 }
159 
160 #ifdef COMPAT_IA32
161 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
162 int
163 proc_read_regs32(struct thread *td, struct reg32 *regs32)
164 {
165 
166 	PROC_ACTION(fill_regs32(td, regs32));
167 }
168 
169 int
170 proc_write_regs32(struct thread *td, struct reg32 *regs32)
171 {
172 
173 	PROC_ACTION(set_regs32(td, regs32));
174 }
175 
176 int
177 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
178 {
179 
180 	PROC_ACTION(fill_dbregs32(td, dbregs32));
181 }
182 
183 int
184 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
185 {
186 
187 	PROC_ACTION(set_dbregs32(td, dbregs32));
188 }
189 
190 int
191 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
192 {
193 
194 	PROC_ACTION(fill_fpregs32(td, fpregs32));
195 }
196 
197 int
198 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
199 {
200 
201 	PROC_ACTION(set_fpregs32(td, fpregs32));
202 }
203 #endif
204 
205 int
206 proc_sstep(struct thread *td)
207 {
208 
209 	PROC_ACTION(ptrace_single_step(td));
210 }
211 
212 int
213 proc_rwmem(struct proc *p, struct uio *uio)
214 {
215 	vm_map_t map;
216 	vm_object_t backing_object, object = NULL;
217 	vm_offset_t pageno = 0;		/* page number */
218 	vm_prot_t reqprot;
219 	int error, fault_flags, writing;
220 
221 	/*
222 	 * Assert that someone has locked this vmspace.  (Should be
223 	 * curthread but we can't assert that.)  This keeps the process
224 	 * from exiting out from under us until this operation completes.
225 	 */
226 	KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
227 	    p, p->p_pid));
228 
229 	/*
230 	 * The map we want...
231 	 */
232 	map = &p->p_vmspace->vm_map;
233 
234 	writing = uio->uio_rw == UIO_WRITE;
235 	reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
236 	    VM_PROT_READ;
237 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
238 
239 	/*
240 	 * Only map in one page at a time.  We don't have to, but it
241 	 * makes things easier.  This way is trivial - right?
242 	 */
243 	do {
244 		vm_map_t tmap;
245 		vm_offset_t uva;
246 		int page_offset;		/* offset into page */
247 		vm_map_entry_t out_entry;
248 		vm_prot_t out_prot;
249 		boolean_t wired;
250 		vm_pindex_t pindex;
251 		u_int len;
252 		vm_page_t m;
253 
254 		object = NULL;
255 
256 		uva = (vm_offset_t)uio->uio_offset;
257 
258 		/*
259 		 * Get the page number of this segment.
260 		 */
261 		pageno = trunc_page(uva);
262 		page_offset = uva - pageno;
263 
264 		/*
265 		 * How many bytes to copy
266 		 */
267 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
268 
269 		/*
270 		 * Fault the page on behalf of the process
271 		 */
272 		error = vm_fault(map, pageno, reqprot, fault_flags);
273 		if (error) {
274 			if (error == KERN_RESOURCE_SHORTAGE)
275 				error = ENOMEM;
276 			else
277 				error = EFAULT;
278 			break;
279 		}
280 
281 		/*
282 		 * Now we need to get the page.  out_entry, wired,
283 		 * and single_use aren't used.  One would think the vm code
284 		 * would be a *bit* nicer...  We use tmap because
285 		 * vm_map_lookup() can change the map argument.
286 		 */
287 		tmap = map;
288 		error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
289 		    &object, &pindex, &out_prot, &wired);
290 		if (error) {
291 			error = EFAULT;
292 			break;
293 		}
294 		VM_OBJECT_LOCK(object);
295 		while ((m = vm_page_lookup(object, pindex)) == NULL &&
296 		    !writing &&
297 		    (backing_object = object->backing_object) != NULL) {
298 			/*
299 			 * Allow fallback to backing objects if we are reading.
300 			 */
301 			VM_OBJECT_LOCK(backing_object);
302 			pindex += OFF_TO_IDX(object->backing_object_offset);
303 			VM_OBJECT_UNLOCK(object);
304 			object = backing_object;
305 		}
306 		VM_OBJECT_UNLOCK(object);
307 		if (m == NULL) {
308 			vm_map_lookup_done(tmap, out_entry);
309 			error = EFAULT;
310 			break;
311 		}
312 
313 		/*
314 		 * Hold the page in memory.
315 		 */
316 		vm_page_lock_queues();
317 		vm_page_hold(m);
318 		vm_page_unlock_queues();
319 
320 		/*
321 		 * We're done with tmap now.
322 		 */
323 		vm_map_lookup_done(tmap, out_entry);
324 
325 		/*
326 		 * Now do the i/o move.
327 		 */
328 		error = uiomove_fromphys(&m, page_offset, len, uio);
329 
330 		/* Make the I-cache coherent for breakpoints. */
331 		if (!error && writing && (out_prot & VM_PROT_EXECUTE))
332 			vm_sync_icache(map, uva, len);
333 
334 		/*
335 		 * Release the page.
336 		 */
337 		vm_page_lock_queues();
338 		vm_page_unhold(m);
339 		vm_page_unlock_queues();
340 
341 	} while (error == 0 && uio->uio_resid > 0);
342 
343 	return (error);
344 }
345 
346 /*
347  * Process debugging system call.
348  */
349 #ifndef _SYS_SYSPROTO_H_
350 struct ptrace_args {
351 	int	req;
352 	pid_t	pid;
353 	caddr_t	addr;
354 	int	data;
355 };
356 #endif
357 
358 #ifdef COMPAT_IA32
359 /*
360  * This CPP subterfuge is to try and reduce the number of ifdefs in
361  * the body of the code.
362  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
363  * becomes either:
364  *   copyin(uap->addr, &r.reg, sizeof r.reg);
365  * or
366  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
367  * .. except this is done at runtime.
368  */
369 #define	COPYIN(u, k, s)		wrap32 ? \
370 	copyin(u, k ## 32, s ## 32) : \
371 	copyin(u, k, s)
372 #define	COPYOUT(k, u, s)	wrap32 ? \
373 	copyout(k ## 32, u, s ## 32) : \
374 	copyout(k, u, s)
375 #else
376 #define	COPYIN(u, k, s)		copyin(u, k, s)
377 #define	COPYOUT(k, u, s)	copyout(k, u, s)
378 #endif
379 int
380 ptrace(struct thread *td, struct ptrace_args *uap)
381 {
382 	/*
383 	 * XXX this obfuscation is to reduce stack usage, but the register
384 	 * structs may be too large to put on the stack anyway.
385 	 */
386 	union {
387 		struct ptrace_io_desc piod;
388 		struct ptrace_lwpinfo pl;
389 		struct dbreg dbreg;
390 		struct fpreg fpreg;
391 		struct reg reg;
392 #ifdef COMPAT_IA32
393 		struct dbreg32 dbreg32;
394 		struct fpreg32 fpreg32;
395 		struct reg32 reg32;
396 		struct ptrace_io_desc32 piod32;
397 #endif
398 	} r;
399 	void *addr;
400 	int error = 0;
401 #ifdef COMPAT_IA32
402 	int wrap32 = 0;
403 
404 	if (SV_CURPROC_FLAG(SV_ILP32))
405 		wrap32 = 1;
406 #endif
407 	AUDIT_ARG_PID(uap->pid);
408 	AUDIT_ARG_CMD(uap->req);
409 	AUDIT_ARG_VALUE(uap->data);
410 	addr = &r;
411 	switch (uap->req) {
412 	case PT_GETREGS:
413 	case PT_GETFPREGS:
414 	case PT_GETDBREGS:
415 	case PT_LWPINFO:
416 		break;
417 	case PT_SETREGS:
418 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
419 		break;
420 	case PT_SETFPREGS:
421 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
422 		break;
423 	case PT_SETDBREGS:
424 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
425 		break;
426 	case PT_IO:
427 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
428 		break;
429 	default:
430 		addr = uap->addr;
431 		break;
432 	}
433 	if (error)
434 		return (error);
435 
436 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
437 	if (error)
438 		return (error);
439 
440 	switch (uap->req) {
441 	case PT_IO:
442 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
443 		break;
444 	case PT_GETREGS:
445 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
446 		break;
447 	case PT_GETFPREGS:
448 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
449 		break;
450 	case PT_GETDBREGS:
451 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
452 		break;
453 	case PT_LWPINFO:
454 		error = copyout(&r.pl, uap->addr, uap->data);
455 		break;
456 	}
457 
458 	return (error);
459 }
460 #undef COPYIN
461 #undef COPYOUT
462 
463 #ifdef COMPAT_IA32
464 /*
465  *   PROC_READ(regs, td2, addr);
466  * becomes either:
467  *   proc_read_regs(td2, addr);
468  * or
469  *   proc_read_regs32(td2, addr);
470  * .. except this is done at runtime.  There is an additional
471  * complication in that PROC_WRITE disallows 32 bit consumers
472  * from writing to 64 bit address space targets.
473  */
474 #define	PROC_READ(w, t, a)	wrap32 ? \
475 	proc_read_ ## w ## 32(t, a) : \
476 	proc_read_ ## w (t, a)
477 #define	PROC_WRITE(w, t, a)	wrap32 ? \
478 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
479 	proc_write_ ## w (t, a)
480 #else
481 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
482 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
483 #endif
484 
485 int
486 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
487 {
488 	struct iovec iov;
489 	struct uio uio;
490 	struct proc *curp, *p, *pp;
491 	struct thread *td2 = NULL;
492 	struct ptrace_io_desc *piod = NULL;
493 	struct ptrace_lwpinfo *pl;
494 	int error, write, tmp, num;
495 	int proctree_locked = 0;
496 	lwpid_t tid = 0, *buf;
497 #ifdef COMPAT_IA32
498 	int wrap32 = 0, safe = 0;
499 	struct ptrace_io_desc32 *piod32 = NULL;
500 #endif
501 
502 	curp = td->td_proc;
503 
504 	/* Lock proctree before locking the process. */
505 	switch (req) {
506 	case PT_TRACE_ME:
507 	case PT_ATTACH:
508 	case PT_STEP:
509 	case PT_CONTINUE:
510 	case PT_TO_SCE:
511 	case PT_TO_SCX:
512 	case PT_SYSCALL:
513 	case PT_DETACH:
514 		sx_xlock(&proctree_lock);
515 		proctree_locked = 1;
516 		break;
517 	default:
518 		break;
519 	}
520 
521 	write = 0;
522 	if (req == PT_TRACE_ME) {
523 		p = td->td_proc;
524 		PROC_LOCK(p);
525 	} else {
526 		if (pid <= PID_MAX) {
527 			if ((p = pfind(pid)) == NULL) {
528 				if (proctree_locked)
529 					sx_xunlock(&proctree_lock);
530 				return (ESRCH);
531 			}
532 		} else {
533 			/* this is slow, should be optimized */
534 			sx_slock(&allproc_lock);
535 			FOREACH_PROC_IN_SYSTEM(p) {
536 				PROC_LOCK(p);
537 				FOREACH_THREAD_IN_PROC(p, td2) {
538 					if (td2->td_tid == pid)
539 						break;
540 				}
541 				if (td2 != NULL)
542 					break; /* proc lock held */
543 				PROC_UNLOCK(p);
544 			}
545 			sx_sunlock(&allproc_lock);
546 			if (p == NULL) {
547 				if (proctree_locked)
548 					sx_xunlock(&proctree_lock);
549 				return (ESRCH);
550 			}
551 			tid = pid;
552 			pid = p->p_pid;
553 		}
554 	}
555 	AUDIT_ARG_PROCESS(p);
556 
557 	if ((p->p_flag & P_WEXIT) != 0) {
558 		error = ESRCH;
559 		goto fail;
560 	}
561 	if ((error = p_cansee(td, p)) != 0)
562 		goto fail;
563 
564 	if ((error = p_candebug(td, p)) != 0)
565 		goto fail;
566 
567 	/*
568 	 * System processes can't be debugged.
569 	 */
570 	if ((p->p_flag & P_SYSTEM) != 0) {
571 		error = EINVAL;
572 		goto fail;
573 	}
574 
575 	if (tid == 0) {
576 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
577 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
578 			td2 = p->p_xthread;
579 		} else {
580 			td2 = FIRST_THREAD_IN_PROC(p);
581 		}
582 		tid = td2->td_tid;
583 	}
584 
585 #ifdef COMPAT_IA32
586 	/*
587 	 * Test if we're a 32 bit client and what the target is.
588 	 * Set the wrap controls accordingly.
589 	 */
590 	if (SV_CURPROC_FLAG(SV_ILP32)) {
591 		if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
592 			safe = 1;
593 		wrap32 = 1;
594 	}
595 #endif
596 	/*
597 	 * Permissions check
598 	 */
599 	switch (req) {
600 	case PT_TRACE_ME:
601 		/* Always legal. */
602 		break;
603 
604 	case PT_ATTACH:
605 		/* Self */
606 		if (p->p_pid == td->td_proc->p_pid) {
607 			error = EINVAL;
608 			goto fail;
609 		}
610 
611 		/* Already traced */
612 		if (p->p_flag & P_TRACED) {
613 			error = EBUSY;
614 			goto fail;
615 		}
616 
617 		/* Can't trace an ancestor if you're being traced. */
618 		if (curp->p_flag & P_TRACED) {
619 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
620 				if (pp == p) {
621 					error = EINVAL;
622 					goto fail;
623 				}
624 			}
625 		}
626 
627 
628 		/* OK */
629 		break;
630 
631 	case PT_CLEARSTEP:
632 		/* Allow thread to clear single step for itself */
633 		if (td->td_tid == tid)
634 			break;
635 
636 		/* FALLTHROUGH */
637 	default:
638 		/* not being traced... */
639 		if ((p->p_flag & P_TRACED) == 0) {
640 			error = EPERM;
641 			goto fail;
642 		}
643 
644 		/* not being traced by YOU */
645 		if (p->p_pptr != td->td_proc) {
646 			error = EBUSY;
647 			goto fail;
648 		}
649 
650 		/* not currently stopped */
651 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
652 		    p->p_suspcount != p->p_numthreads  ||
653 		    (p->p_flag & P_WAITED) == 0) {
654 			error = EBUSY;
655 			goto fail;
656 		}
657 
658 		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
659 			static int count = 0;
660 			if (count++ == 0)
661 				printf("P_STOPPED_TRACE not set.\n");
662 		}
663 
664 		/* OK */
665 		break;
666 	}
667 
668 	/* Keep this process around until we finish this request. */
669 	_PHOLD(p);
670 
671 #ifdef FIX_SSTEP
672 	/*
673 	 * Single step fixup ala procfs
674 	 */
675 	FIX_SSTEP(td2);
676 #endif
677 
678 	/*
679 	 * Actually do the requests
680 	 */
681 
682 	td->td_retval[0] = 0;
683 
684 	switch (req) {
685 	case PT_TRACE_ME:
686 		/* set my trace flag and "owner" so it can read/write me */
687 		p->p_flag |= P_TRACED;
688 		p->p_oppid = p->p_pptr->p_pid;
689 		break;
690 
691 	case PT_ATTACH:
692 		/* security check done above */
693 		p->p_flag |= P_TRACED;
694 		p->p_oppid = p->p_pptr->p_pid;
695 		if (p->p_pptr != td->td_proc)
696 			proc_reparent(p, td->td_proc);
697 		data = SIGSTOP;
698 		goto sendsig;	/* in PT_CONTINUE below */
699 
700 	case PT_CLEARSTEP:
701 		error = ptrace_clear_single_step(td2);
702 		break;
703 
704 	case PT_SETSTEP:
705 		error = ptrace_single_step(td2);
706 		break;
707 
708 	case PT_SUSPEND:
709 		td2->td_dbgflags |= TDB_SUSPEND;
710 		thread_lock(td2);
711 		td2->td_flags |= TDF_NEEDSUSPCHK;
712 		thread_unlock(td2);
713 		break;
714 
715 	case PT_RESUME:
716 		td2->td_dbgflags &= ~TDB_SUSPEND;
717 		break;
718 
719 	case PT_STEP:
720 	case PT_CONTINUE:
721 	case PT_TO_SCE:
722 	case PT_TO_SCX:
723 	case PT_SYSCALL:
724 	case PT_DETACH:
725 		/* Zero means do not send any signal */
726 		if (data < 0 || data > _SIG_MAXSIG) {
727 			error = EINVAL;
728 			break;
729 		}
730 
731 		switch (req) {
732 		case PT_STEP:
733 			error = ptrace_single_step(td2);
734 			if (error)
735 				goto out;
736 			break;
737 		case PT_TO_SCE:
738 			p->p_stops |= S_PT_SCE;
739 			break;
740 		case PT_TO_SCX:
741 			p->p_stops |= S_PT_SCX;
742 			break;
743 		case PT_SYSCALL:
744 			p->p_stops |= S_PT_SCE | S_PT_SCX;
745 			break;
746 		}
747 
748 		if (addr != (void *)1) {
749 			error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
750 			if (error)
751 				break;
752 		}
753 
754 		if (req == PT_DETACH) {
755 			/* reset process parent */
756 			if (p->p_oppid != p->p_pptr->p_pid) {
757 				struct proc *pp;
758 
759 				PROC_LOCK(p->p_pptr);
760 				sigqueue_take(p->p_ksi);
761 				PROC_UNLOCK(p->p_pptr);
762 
763 				PROC_UNLOCK(p);
764 				pp = pfind(p->p_oppid);
765 				if (pp == NULL)
766 					pp = initproc;
767 				else
768 					PROC_UNLOCK(pp);
769 				PROC_LOCK(p);
770 				proc_reparent(p, pp);
771 				if (pp == initproc)
772 					p->p_sigparent = SIGCHLD;
773 			}
774 			p->p_flag &= ~(P_TRACED | P_WAITED);
775 			p->p_oppid = 0;
776 
777 			/* should we send SIGCHLD? */
778 			/* childproc_continued(p); */
779 		}
780 
781 	sendsig:
782 		if (proctree_locked) {
783 			sx_xunlock(&proctree_lock);
784 			proctree_locked = 0;
785 		}
786 		p->p_xstat = data;
787 		p->p_xthread = NULL;
788 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
789 			/* deliver or queue signal */
790 			td2->td_dbgflags &= ~TDB_XSIG;
791 			td2->td_xsig = data;
792 
793 			if (req == PT_DETACH) {
794 				struct thread *td3;
795 				FOREACH_THREAD_IN_PROC(p, td3) {
796 					td3->td_dbgflags &= ~TDB_SUSPEND;
797 				}
798 			}
799 			/*
800 			 * unsuspend all threads, to not let a thread run,
801 			 * you should use PT_SUSPEND to suspend it before
802 			 * continuing process.
803 			 */
804 			PROC_SLOCK(p);
805 			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
806 			thread_unsuspend(p);
807 			PROC_SUNLOCK(p);
808 		} else {
809 			if (data)
810 				psignal(p, data);
811 		}
812 		break;
813 
814 	case PT_WRITE_I:
815 	case PT_WRITE_D:
816 		write = 1;
817 		/* FALLTHROUGH */
818 	case PT_READ_I:
819 	case PT_READ_D:
820 		PROC_UNLOCK(p);
821 		tmp = 0;
822 		/* write = 0 set above */
823 		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
824 		iov.iov_len = sizeof(int);
825 		uio.uio_iov = &iov;
826 		uio.uio_iovcnt = 1;
827 		uio.uio_offset = (off_t)(uintptr_t)addr;
828 		uio.uio_resid = sizeof(int);
829 		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
830 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
831 		uio.uio_td = td;
832 		error = proc_rwmem(p, &uio);
833 		if (uio.uio_resid != 0) {
834 			/*
835 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
836 			 * so I think write() can bogusly return 0.
837 			 * XXX what happens for short writes?  We don't want
838 			 * to write partial data.
839 			 * XXX proc_rwmem() returns EPERM for other invalid
840 			 * addresses.  Convert this to EINVAL.  Does this
841 			 * clobber returns of EPERM for other reasons?
842 			 */
843 			if (error == 0 || error == ENOSPC || error == EPERM)
844 				error = EINVAL;	/* EOF */
845 		}
846 		if (!write)
847 			td->td_retval[0] = tmp;
848 		PROC_LOCK(p);
849 		break;
850 
851 	case PT_IO:
852 #ifdef COMPAT_IA32
853 		if (wrap32) {
854 			piod32 = addr;
855 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
856 			iov.iov_len = piod32->piod_len;
857 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
858 			uio.uio_resid = piod32->piod_len;
859 		} else
860 #endif
861 		{
862 			piod = addr;
863 			iov.iov_base = piod->piod_addr;
864 			iov.iov_len = piod->piod_len;
865 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
866 			uio.uio_resid = piod->piod_len;
867 		}
868 		uio.uio_iov = &iov;
869 		uio.uio_iovcnt = 1;
870 		uio.uio_segflg = UIO_USERSPACE;
871 		uio.uio_td = td;
872 #ifdef COMPAT_IA32
873 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
874 #else
875 		tmp = piod->piod_op;
876 #endif
877 		switch (tmp) {
878 		case PIOD_READ_D:
879 		case PIOD_READ_I:
880 			uio.uio_rw = UIO_READ;
881 			break;
882 		case PIOD_WRITE_D:
883 		case PIOD_WRITE_I:
884 			uio.uio_rw = UIO_WRITE;
885 			break;
886 		default:
887 			error = EINVAL;
888 			goto out;
889 		}
890 		PROC_UNLOCK(p);
891 		error = proc_rwmem(p, &uio);
892 #ifdef COMPAT_IA32
893 		if (wrap32)
894 			piod32->piod_len -= uio.uio_resid;
895 		else
896 #endif
897 			piod->piod_len -= uio.uio_resid;
898 		PROC_LOCK(p);
899 		break;
900 
901 	case PT_KILL:
902 		data = SIGKILL;
903 		goto sendsig;	/* in PT_CONTINUE above */
904 
905 	case PT_SETREGS:
906 		error = PROC_WRITE(regs, td2, addr);
907 		break;
908 
909 	case PT_GETREGS:
910 		error = PROC_READ(regs, td2, addr);
911 		break;
912 
913 	case PT_SETFPREGS:
914 		error = PROC_WRITE(fpregs, td2, addr);
915 		break;
916 
917 	case PT_GETFPREGS:
918 		error = PROC_READ(fpregs, td2, addr);
919 		break;
920 
921 	case PT_SETDBREGS:
922 		error = PROC_WRITE(dbregs, td2, addr);
923 		break;
924 
925 	case PT_GETDBREGS:
926 		error = PROC_READ(dbregs, td2, addr);
927 		break;
928 
929 	case PT_LWPINFO:
930 		if (data <= 0 || data > sizeof(*pl)) {
931 			error = EINVAL;
932 			break;
933 		}
934 		pl = addr;
935 		pl->pl_lwpid = td2->td_tid;
936 		if (td2->td_dbgflags & TDB_XSIG)
937 			pl->pl_event = PL_EVENT_SIGNAL;
938 		else
939 			pl->pl_event = 0;
940 		pl->pl_flags = 0;
941 		pl->pl_sigmask = td2->td_sigmask;
942 		pl->pl_siglist = td2->td_siglist;
943 		break;
944 
945 	case PT_GETNUMLWPS:
946 		td->td_retval[0] = p->p_numthreads;
947 		break;
948 
949 	case PT_GETLWPLIST:
950 		if (data <= 0) {
951 			error = EINVAL;
952 			break;
953 		}
954 		num = imin(p->p_numthreads, data);
955 		PROC_UNLOCK(p);
956 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
957 		tmp = 0;
958 		PROC_LOCK(p);
959 		FOREACH_THREAD_IN_PROC(p, td2) {
960 			if (tmp >= num)
961 				break;
962 			buf[tmp++] = td2->td_tid;
963 		}
964 		PROC_UNLOCK(p);
965 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
966 		free(buf, M_TEMP);
967 		if (!error)
968 			td->td_retval[0] = tmp;
969 		PROC_LOCK(p);
970 		break;
971 
972 	default:
973 #ifdef __HAVE_PTRACE_MACHDEP
974 		if (req >= PT_FIRSTMACH) {
975 			PROC_UNLOCK(p);
976 			error = cpu_ptrace(td2, req, addr, data);
977 			PROC_LOCK(p);
978 		} else
979 #endif
980 			/* Unknown request. */
981 			error = EINVAL;
982 		break;
983 	}
984 
985 out:
986 	/* Drop our hold on this process now that the request has completed. */
987 	_PRELE(p);
988 fail:
989 	PROC_UNLOCK(p);
990 	if (proctree_locked)
991 		sx_xunlock(&proctree_lock);
992 	return (error);
993 }
994 #undef PROC_READ
995 #undef PROC_WRITE
996 
997 /*
998  * Stop a process because of a debugging event;
999  * stay stopped until p->p_step is cleared
1000  * (cleared by PIOCCONT in procfs).
1001  */
1002 void
1003 stopevent(struct proc *p, unsigned int event, unsigned int val)
1004 {
1005 
1006 	PROC_LOCK_ASSERT(p, MA_OWNED);
1007 	p->p_step = 1;
1008 	do {
1009 		p->p_xstat = val;
1010 		p->p_xthread = NULL;
1011 		p->p_stype = event;	/* Which event caused the stop? */
1012 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1013 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1014 	} while (p->p_step);
1015 }
1016