xref: /freebsd/sys/kern/sys_process.c (revision f0a75d274af375d15b97b830966b99a02b7db911)
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/sysproto.h>
43 #include <sys/proc.h>
44 #include <sys/vnode.h>
45 #include <sys/ptrace.h>
46 #include <sys/sx.h>
47 #include <sys/malloc.h>
48 #include <sys/signalvar.h>
49 
50 #include <machine/reg.h>
51 
52 #include <security/audit/audit.h>
53 
54 #include <vm/vm.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_extern.h>
57 #include <vm/vm_map.h>
58 #include <vm/vm_kern.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_page.h>
61 
62 #ifdef COMPAT_IA32
63 #include <sys/procfs.h>
64 #include <machine/fpu.h>
65 #include <compat/ia32/ia32_reg.h>
66 
67 extern struct sysentvec ia32_freebsd_sysvec;
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_sflag & PS_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, 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 
238 	/*
239 	 * Only map in one page at a time.  We don't have to, but it
240 	 * makes things easier.  This way is trivial - right?
241 	 */
242 	do {
243 		vm_map_t tmap;
244 		vm_offset_t uva;
245 		int page_offset;		/* offset into page */
246 		vm_map_entry_t out_entry;
247 		vm_prot_t out_prot;
248 		boolean_t wired;
249 		vm_pindex_t pindex;
250 		u_int len;
251 		vm_page_t m;
252 
253 		object = NULL;
254 
255 		uva = (vm_offset_t)uio->uio_offset;
256 
257 		/*
258 		 * Get the page number of this segment.
259 		 */
260 		pageno = trunc_page(uva);
261 		page_offset = uva - pageno;
262 
263 		/*
264 		 * How many bytes to copy
265 		 */
266 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
267 
268 		/*
269 		 * Fault the page on behalf of the process
270 		 */
271 		error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
272 		if (error) {
273 			error = EFAULT;
274 			break;
275 		}
276 
277 		/*
278 		 * Now we need to get the page.  out_entry, out_prot, wired,
279 		 * and single_use aren't used.  One would think the vm code
280 		 * would be a *bit* nicer...  We use tmap because
281 		 * vm_map_lookup() can change the map argument.
282 		 */
283 		tmap = map;
284 		error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
285 		    &object, &pindex, &out_prot, &wired);
286 		if (error) {
287 			error = EFAULT;
288 			break;
289 		}
290 		VM_OBJECT_LOCK(object);
291 		while ((m = vm_page_lookup(object, pindex)) == NULL &&
292 		    !writing &&
293 		    (backing_object = object->backing_object) != NULL) {
294 			/*
295 			 * Allow fallback to backing objects if we are reading.
296 			 */
297 			VM_OBJECT_LOCK(backing_object);
298 			pindex += OFF_TO_IDX(object->backing_object_offset);
299 			VM_OBJECT_UNLOCK(object);
300 			object = backing_object;
301 		}
302 		VM_OBJECT_UNLOCK(object);
303 		if (m == NULL) {
304 			vm_map_lookup_done(tmap, out_entry);
305 			error = EFAULT;
306 			break;
307 		}
308 
309 		/*
310 		 * Hold the page in memory.
311 		 */
312 		vm_page_lock_queues();
313 		vm_page_hold(m);
314 		vm_page_unlock_queues();
315 
316 		/*
317 		 * We're done with tmap now.
318 		 */
319 		vm_map_lookup_done(tmap, out_entry);
320 
321 		/*
322 		 * Now do the i/o move.
323 		 */
324 		error = uiomove_fromphys(&m, page_offset, len, uio);
325 
326 		/*
327 		 * Release the page.
328 		 */
329 		vm_page_lock_queues();
330 		vm_page_unhold(m);
331 		vm_page_unlock_queues();
332 
333 	} while (error == 0 && uio->uio_resid > 0);
334 
335 	return (error);
336 }
337 
338 /*
339  * Process debugging system call.
340  */
341 #ifndef _SYS_SYSPROTO_H_
342 struct ptrace_args {
343 	int	req;
344 	pid_t	pid;
345 	caddr_t	addr;
346 	int	data;
347 };
348 #endif
349 
350 #ifdef COMPAT_IA32
351 /*
352  * This CPP subterfuge is to try and reduce the number of ifdefs in
353  * the body of the code.
354  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
355  * becomes either:
356  *   copyin(uap->addr, &r.reg, sizeof r.reg);
357  * or
358  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
359  * .. except this is done at runtime.
360  */
361 #define	COPYIN(u, k, s)		wrap32 ? \
362 	copyin(u, k ## 32, s ## 32) : \
363 	copyin(u, k, s)
364 #define	COPYOUT(k, u, s)	wrap32 ? \
365 	copyout(k ## 32, u, s ## 32) : \
366 	copyout(k, u, s)
367 #else
368 #define	COPYIN(u, k, s)		copyin(u, k, s)
369 #define	COPYOUT(k, u, s)	copyout(k, u, s)
370 #endif
371 int
372 ptrace(struct thread *td, struct ptrace_args *uap)
373 {
374 	/*
375 	 * XXX this obfuscation is to reduce stack usage, but the register
376 	 * structs may be too large to put on the stack anyway.
377 	 */
378 	union {
379 		struct ptrace_io_desc piod;
380 		struct ptrace_lwpinfo pl;
381 		struct dbreg dbreg;
382 		struct fpreg fpreg;
383 		struct reg reg;
384 #ifdef COMPAT_IA32
385 		struct dbreg32 dbreg32;
386 		struct fpreg32 fpreg32;
387 		struct reg32 reg32;
388 		struct ptrace_io_desc32 piod32;
389 #endif
390 	} r;
391 	void *addr;
392 	int error = 0;
393 #ifdef COMPAT_IA32
394 	int wrap32 = 0;
395 
396 	if (td->td_proc->p_sysent == &ia32_freebsd_sysvec)
397 		wrap32 = 1;
398 #endif
399 	AUDIT_ARG(pid, uap->pid);
400 	AUDIT_ARG(cmd, uap->req);
401 	AUDIT_ARG(addr, uap->addr);
402 	AUDIT_ARG(value, uap->data);
403 	addr = &r;
404 	switch (uap->req) {
405 	case PT_GETREGS:
406 	case PT_GETFPREGS:
407 	case PT_GETDBREGS:
408 	case PT_LWPINFO:
409 		break;
410 	case PT_SETREGS:
411 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
412 		break;
413 	case PT_SETFPREGS:
414 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
415 		break;
416 	case PT_SETDBREGS:
417 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
418 		break;
419 	case PT_IO:
420 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
421 		break;
422 	default:
423 		addr = uap->addr;
424 		break;
425 	}
426 	if (error)
427 		return (error);
428 
429 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
430 	if (error)
431 		return (error);
432 
433 	switch (uap->req) {
434 	case PT_IO:
435 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
436 		break;
437 	case PT_GETREGS:
438 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
439 		break;
440 	case PT_GETFPREGS:
441 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
442 		break;
443 	case PT_GETDBREGS:
444 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
445 		break;
446 	case PT_LWPINFO:
447 		error = copyout(&r.pl, uap->addr, uap->data);
448 		break;
449 	}
450 
451 	return (error);
452 }
453 #undef COPYIN
454 #undef COPYOUT
455 
456 #ifdef COMPAT_IA32
457 /*
458  *   PROC_READ(regs, td2, addr);
459  * becomes either:
460  *   proc_read_regs(td2, addr);
461  * or
462  *   proc_read_regs32(td2, addr);
463  * .. except this is done at runtime.  There is an additional
464  * complication in that PROC_WRITE disallows 32 bit consumers
465  * from writing to 64 bit address space targets.
466  */
467 #define	PROC_READ(w, t, a)	wrap32 ? \
468 	proc_read_ ## w ## 32(t, a) : \
469 	proc_read_ ## w (t, a)
470 #define	PROC_WRITE(w, t, a)	wrap32 ? \
471 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
472 	proc_write_ ## w (t, a)
473 #else
474 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
475 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
476 #endif
477 
478 int
479 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
480 {
481 	struct iovec iov;
482 	struct uio uio;
483 	struct proc *curp, *p, *pp;
484 	struct thread *td2 = NULL;
485 	struct ptrace_io_desc *piod = NULL;
486 	struct ptrace_lwpinfo *pl;
487 	int error, write, tmp, num;
488 	int proctree_locked = 0;
489 	lwpid_t tid = 0, *buf;
490 #ifdef COMPAT_IA32
491 	int wrap32 = 0, safe = 0;
492 	struct ptrace_io_desc32 *piod32 = NULL;
493 #endif
494 
495 	curp = td->td_proc;
496 
497 	/* Lock proctree before locking the process. */
498 	switch (req) {
499 	case PT_TRACE_ME:
500 	case PT_ATTACH:
501 	case PT_STEP:
502 	case PT_CONTINUE:
503 	case PT_TO_SCE:
504 	case PT_TO_SCX:
505 	case PT_SYSCALL:
506 	case PT_DETACH:
507 		sx_xlock(&proctree_lock);
508 		proctree_locked = 1;
509 		break;
510 	default:
511 		break;
512 	}
513 
514 	write = 0;
515 	if (req == PT_TRACE_ME) {
516 		p = td->td_proc;
517 		PROC_LOCK(p);
518 	} else {
519 		if (pid <= PID_MAX) {
520 			if ((p = pfind(pid)) == NULL) {
521 				if (proctree_locked)
522 					sx_xunlock(&proctree_lock);
523 				return (ESRCH);
524 			}
525 		} else {
526 			/* this is slow, should be optimized */
527 			sx_slock(&allproc_lock);
528 			FOREACH_PROC_IN_SYSTEM(p) {
529 				PROC_LOCK(p);
530 				mtx_lock_spin(&sched_lock);
531 				FOREACH_THREAD_IN_PROC(p, td2) {
532 					if (td2->td_tid == pid)
533 						break;
534 				}
535 				mtx_unlock_spin(&sched_lock);
536 				if (td2 != NULL)
537 					break; /* proc lock held */
538 				PROC_UNLOCK(p);
539 			}
540 			sx_sunlock(&allproc_lock);
541 			if (p == NULL) {
542 				if (proctree_locked)
543 					sx_xunlock(&proctree_lock);
544 				return (ESRCH);
545 			}
546 			tid = pid;
547 			pid = p->p_pid;
548 		}
549 	}
550 	AUDIT_ARG(process, p);
551 
552 	if ((p->p_flag & P_WEXIT) != 0) {
553 		error = ESRCH;
554 		goto fail;
555 	}
556 	if ((error = p_cansee(td, p)) != 0)
557 		goto fail;
558 
559 	if ((error = p_candebug(td, p)) != 0)
560 		goto fail;
561 
562 	/*
563 	 * System processes can't be debugged.
564 	 */
565 	if ((p->p_flag & P_SYSTEM) != 0) {
566 		error = EINVAL;
567 		goto fail;
568 	}
569 
570 	if (tid == 0) {
571 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
572 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
573 			td2 = p->p_xthread;
574 		} else {
575 			td2 = FIRST_THREAD_IN_PROC(p);
576 		}
577 		tid = td2->td_tid;
578 	}
579 
580 #ifdef COMPAT_IA32
581 	/*
582 	 * Test if we're a 32 bit client and what the target is.
583 	 * Set the wrap controls accordingly.
584 	 */
585 	if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) {
586 		if (td2->td_proc->p_sysent == &ia32_freebsd_sysvec)
587 			safe = 1;
588 		wrap32 = 1;
589 	}
590 #endif
591 	/*
592 	 * Permissions check
593 	 */
594 	switch (req) {
595 	case PT_TRACE_ME:
596 		/* Always legal. */
597 		break;
598 
599 	case PT_ATTACH:
600 		/* Self */
601 		if (p->p_pid == td->td_proc->p_pid) {
602 			error = EINVAL;
603 			goto fail;
604 		}
605 
606 		/* Already traced */
607 		if (p->p_flag & P_TRACED) {
608 			error = EBUSY;
609 			goto fail;
610 		}
611 
612 		/* Can't trace an ancestor if you're being traced. */
613 		if (curp->p_flag & P_TRACED) {
614 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
615 				if (pp == p) {
616 					error = EINVAL;
617 					goto fail;
618 				}
619 			}
620 		}
621 
622 
623 		/* OK */
624 		break;
625 
626 	case PT_CLEARSTEP:
627 		/* Allow thread to clear single step for itself */
628 		if (td->td_tid == tid)
629 			break;
630 
631 		/* FALLTHROUGH */
632 	default:
633 		/* not being traced... */
634 		if ((p->p_flag & P_TRACED) == 0) {
635 			error = EPERM;
636 			goto fail;
637 		}
638 
639 		/* not being traced by YOU */
640 		if (p->p_pptr != td->td_proc) {
641 			error = EBUSY;
642 			goto fail;
643 		}
644 
645 		/* not currently stopped */
646 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
647 		    p->p_suspcount != p->p_numthreads  ||
648 		    (p->p_flag & P_WAITED) == 0) {
649 			error = EBUSY;
650 			goto fail;
651 		}
652 
653 		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
654 			static int count = 0;
655 			if (count++ == 0)
656 				printf("P_STOPPED_TRACE not set.\n");
657 		}
658 
659 		/* OK */
660 		break;
661 	}
662 
663 	/* Keep this process around until we finish this request. */
664 	_PHOLD(p);
665 
666 #ifdef FIX_SSTEP
667 	/*
668 	 * Single step fixup ala procfs
669 	 */
670 	FIX_SSTEP(td2);
671 #endif
672 
673 	/*
674 	 * Actually do the requests
675 	 */
676 
677 	td->td_retval[0] = 0;
678 
679 	switch (req) {
680 	case PT_TRACE_ME:
681 		/* set my trace flag and "owner" so it can read/write me */
682 		p->p_flag |= P_TRACED;
683 		p->p_oppid = p->p_pptr->p_pid;
684 		break;
685 
686 	case PT_ATTACH:
687 		/* security check done above */
688 		p->p_flag |= P_TRACED;
689 		p->p_oppid = p->p_pptr->p_pid;
690 		if (p->p_pptr != td->td_proc)
691 			proc_reparent(p, td->td_proc);
692 		data = SIGSTOP;
693 		goto sendsig;	/* in PT_CONTINUE below */
694 
695 	case PT_CLEARSTEP:
696 		error = ptrace_clear_single_step(td2);
697 		break;
698 
699 	case PT_SETSTEP:
700 		error = ptrace_single_step(td2);
701 		break;
702 
703 	case PT_SUSPEND:
704 		mtx_lock_spin(&sched_lock);
705 		td2->td_flags |= TDF_DBSUSPEND;
706 		mtx_unlock_spin(&sched_lock);
707 		break;
708 
709 	case PT_RESUME:
710 		mtx_lock_spin(&sched_lock);
711 		td2->td_flags &= ~TDF_DBSUSPEND;
712 		mtx_unlock_spin(&sched_lock);
713 		break;
714 
715 	case PT_STEP:
716 	case PT_CONTINUE:
717 	case PT_TO_SCE:
718 	case PT_TO_SCX:
719 	case PT_SYSCALL:
720 	case PT_DETACH:
721 		/* Zero means do not send any signal */
722 		if (data < 0 || data > _SIG_MAXSIG) {
723 			error = EINVAL;
724 			break;
725 		}
726 
727 		switch (req) {
728 		case PT_STEP:
729 			error = ptrace_single_step(td2);
730 			if (error)
731 				goto out;
732 			break;
733 		case PT_TO_SCE:
734 			p->p_stops |= S_PT_SCE;
735 			break;
736 		case PT_TO_SCX:
737 			p->p_stops |= S_PT_SCX;
738 			break;
739 		case PT_SYSCALL:
740 			p->p_stops |= S_PT_SCE | S_PT_SCX;
741 			break;
742 		}
743 
744 		if (addr != (void *)1) {
745 			error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
746 			if (error)
747 				break;
748 		}
749 
750 		if (req == PT_DETACH) {
751 			/* reset process parent */
752 			if (p->p_oppid != p->p_pptr->p_pid) {
753 				struct proc *pp;
754 
755 				PROC_LOCK(p->p_pptr);
756 				sigqueue_take(p->p_ksi);
757 				PROC_UNLOCK(p->p_pptr);
758 
759 				PROC_UNLOCK(p);
760 				pp = pfind(p->p_oppid);
761 				if (pp == NULL)
762 					pp = initproc;
763 				else
764 					PROC_UNLOCK(pp);
765 				PROC_LOCK(p);
766 				proc_reparent(p, pp);
767 				if (pp == initproc)
768 					p->p_sigparent = SIGCHLD;
769 			}
770 			p->p_flag &= ~(P_TRACED | P_WAITED);
771 			p->p_oppid = 0;
772 
773 			/* should we send SIGCHLD? */
774 			/* childproc_continued(p); */
775 		}
776 
777 	sendsig:
778 		if (proctree_locked) {
779 			sx_xunlock(&proctree_lock);
780 			proctree_locked = 0;
781 		}
782 		/* deliver or queue signal */
783 		mtx_lock_spin(&sched_lock);
784 		td2->td_flags &= ~TDF_XSIG;
785 		mtx_unlock_spin(&sched_lock);
786 		td2->td_xsig = data;
787 		p->p_xstat = data;
788 		p->p_xthread = NULL;
789 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
790 			mtx_lock_spin(&sched_lock);
791 			if (req == PT_DETACH) {
792 				struct thread *td3;
793 				FOREACH_THREAD_IN_PROC(p, td3)
794 					td3->td_flags &= ~TDF_DBSUSPEND;
795 			}
796 			/*
797 			 * unsuspend all threads, to not let a thread run,
798 			 * you should use PT_SUSPEND to suspend it before
799 			 * continuing process.
800 			 */
801 			mtx_unlock_spin(&sched_lock);
802 #ifdef KSE
803 			thread_continued(p);
804 #endif
805 			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
806 			mtx_lock_spin(&sched_lock);
807 			thread_unsuspend(p);
808 			mtx_unlock_spin(&sched_lock);
809 		}
810 
811 		if (data)
812 			psignal(p, data);
813 
814 		break;
815 
816 	case PT_WRITE_I:
817 	case PT_WRITE_D:
818 		write = 1;
819 		/* FALLTHROUGH */
820 	case PT_READ_I:
821 	case PT_READ_D:
822 		PROC_UNLOCK(p);
823 		tmp = 0;
824 		/* write = 0 set above */
825 		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
826 		iov.iov_len = sizeof(int);
827 		uio.uio_iov = &iov;
828 		uio.uio_iovcnt = 1;
829 		uio.uio_offset = (off_t)(uintptr_t)addr;
830 		uio.uio_resid = sizeof(int);
831 		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
832 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
833 		uio.uio_td = td;
834 		error = proc_rwmem(p, &uio);
835 		if (uio.uio_resid != 0) {
836 			/*
837 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
838 			 * so I think write() can bogusly return 0.
839 			 * XXX what happens for short writes?  We don't want
840 			 * to write partial data.
841 			 * XXX proc_rwmem() returns EPERM for other invalid
842 			 * addresses.  Convert this to EINVAL.  Does this
843 			 * clobber returns of EPERM for other reasons?
844 			 */
845 			if (error == 0 || error == ENOSPC || error == EPERM)
846 				error = EINVAL;	/* EOF */
847 		}
848 		if (!write)
849 			td->td_retval[0] = tmp;
850 		PROC_LOCK(p);
851 		break;
852 
853 	case PT_IO:
854 #ifdef COMPAT_IA32
855 		if (wrap32) {
856 			piod32 = addr;
857 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
858 			iov.iov_len = piod32->piod_len;
859 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
860 			uio.uio_resid = piod32->piod_len;
861 		} else
862 #endif
863 		{
864 			piod = addr;
865 			iov.iov_base = piod->piod_addr;
866 			iov.iov_len = piod->piod_len;
867 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
868 			uio.uio_resid = piod->piod_len;
869 		}
870 		uio.uio_iov = &iov;
871 		uio.uio_iovcnt = 1;
872 		uio.uio_segflg = UIO_USERSPACE;
873 		uio.uio_td = td;
874 #ifdef COMPAT_IA32
875 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
876 #else
877 		tmp = piod->piod_op;
878 #endif
879 		switch (tmp) {
880 		case PIOD_READ_D:
881 		case PIOD_READ_I:
882 			uio.uio_rw = UIO_READ;
883 			break;
884 		case PIOD_WRITE_D:
885 		case PIOD_WRITE_I:
886 			uio.uio_rw = UIO_WRITE;
887 			break;
888 		default:
889 			error = EINVAL;
890 			goto out;
891 		}
892 		PROC_UNLOCK(p);
893 		error = proc_rwmem(p, &uio);
894 #ifdef COMPAT_IA32
895 		if (wrap32)
896 			piod32->piod_len -= uio.uio_resid;
897 		else
898 #endif
899 			piod->piod_len -= uio.uio_resid;
900 		PROC_LOCK(p);
901 		break;
902 
903 	case PT_KILL:
904 		data = SIGKILL;
905 		goto sendsig;	/* in PT_CONTINUE above */
906 
907 	case PT_SETREGS:
908 		error = PROC_WRITE(regs, td2, addr);
909 		break;
910 
911 	case PT_GETREGS:
912 		error = PROC_READ(regs, td2, addr);
913 		break;
914 
915 	case PT_SETFPREGS:
916 		error = PROC_WRITE(fpregs, td2, addr);
917 		break;
918 
919 	case PT_GETFPREGS:
920 		error = PROC_READ(fpregs, td2, addr);
921 		break;
922 
923 	case PT_SETDBREGS:
924 		error = PROC_WRITE(dbregs, td2, addr);
925 		break;
926 
927 	case PT_GETDBREGS:
928 		error = PROC_READ(dbregs, td2, addr);
929 		break;
930 
931 	case PT_LWPINFO:
932 		if (data <= 0 || data > sizeof(*pl)) {
933 			error = EINVAL;
934 			break;
935 		}
936 		pl = addr;
937 		pl->pl_lwpid = td2->td_tid;
938 		if (td2->td_flags & TDF_XSIG)
939 			pl->pl_event = PL_EVENT_SIGNAL;
940 		else
941 			pl->pl_event = 0;
942 #ifdef KSE
943 		if (td2->td_pflags & TDP_SA) {
944 			pl->pl_flags = PL_FLAG_SA;
945 			if (td2->td_upcall && !TD_CAN_UNBIND(td2))
946 				pl->pl_flags |= PL_FLAG_BOUND;
947 		} else {
948 			pl->pl_flags = 0;
949 		}
950 #else
951 		pl->pl_flags = 0;
952 #endif
953 		pl->pl_sigmask = td2->td_sigmask;
954 		pl->pl_siglist = td2->td_siglist;
955 		break;
956 
957 	case PT_GETNUMLWPS:
958 		td->td_retval[0] = p->p_numthreads;
959 		break;
960 
961 	case PT_GETLWPLIST:
962 		if (data <= 0) {
963 			error = EINVAL;
964 			break;
965 		}
966 		num = imin(p->p_numthreads, data);
967 		PROC_UNLOCK(p);
968 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
969 		tmp = 0;
970 		PROC_LOCK(p);
971 		mtx_lock_spin(&sched_lock);
972 		FOREACH_THREAD_IN_PROC(p, td2) {
973 			if (tmp >= num)
974 				break;
975 			buf[tmp++] = td2->td_tid;
976 		}
977 		mtx_unlock_spin(&sched_lock);
978 		PROC_UNLOCK(p);
979 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
980 		free(buf, M_TEMP);
981 		if (!error)
982 			td->td_retval[0] = tmp;
983 		PROC_LOCK(p);
984 		break;
985 
986 	default:
987 #ifdef __HAVE_PTRACE_MACHDEP
988 		if (req >= PT_FIRSTMACH) {
989 			PROC_UNLOCK(p);
990 			error = cpu_ptrace(td2, req, addr, data);
991 			PROC_LOCK(p);
992 		} else
993 #endif
994 			/* Unknown request. */
995 			error = EINVAL;
996 		break;
997 	}
998 
999 out:
1000 	/* Drop our hold on this process now that the request has completed. */
1001 	_PRELE(p);
1002 fail:
1003 	PROC_UNLOCK(p);
1004 	if (proctree_locked)
1005 		sx_xunlock(&proctree_lock);
1006 	return (error);
1007 }
1008 #undef PROC_READ
1009 #undef PROC_WRITE
1010 
1011 /*
1012  * Stop a process because of a debugging event;
1013  * stay stopped until p->p_step is cleared
1014  * (cleared by PIOCCONT in procfs).
1015  */
1016 void
1017 stopevent(struct proc *p, unsigned int event, unsigned int val)
1018 {
1019 
1020 	PROC_LOCK_ASSERT(p, MA_OWNED);
1021 	p->p_step = 1;
1022 	do {
1023 		p->p_xstat = val;
1024 		p->p_xthread = NULL;
1025 		p->p_stype = event;	/* Which event caused the stop? */
1026 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1027 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1028 	} while (p->p_step);
1029 }
1030