xref: /freebsd/sys/kern/sys_process.c (revision 036d2e814bf0f5d88ffb4b24c159320894541757)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 1994, Sean Eric Fagan
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by Sean Eric Fagan.
18  * 4. The name of the author may not be used to endorse or promote products
19  *    derived from this software without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/ktr.h>
40 #include <sys/limits.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/syscallsubr.h>
44 #include <sys/sysent.h>
45 #include <sys/sysproto.h>
46 #include <sys/pioctl.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/vnode.h>
50 #include <sys/ptrace.h>
51 #include <sys/rwlock.h>
52 #include <sys/sx.h>
53 #include <sys/malloc.h>
54 #include <sys/signalvar.h>
55 
56 #include <machine/reg.h>
57 
58 #include <security/audit/audit.h>
59 
60 #include <vm/vm.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_extern.h>
63 #include <vm/vm_map.h>
64 #include <vm/vm_kern.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_param.h>
68 
69 #ifdef COMPAT_FREEBSD32
70 #include <sys/procfs.h>
71 #include <compat/freebsd32/freebsd32_signal.h>
72 
73 struct ptrace_io_desc32 {
74 	int		piod_op;
75 	uint32_t	piod_offs;
76 	uint32_t	piod_addr;
77 	uint32_t	piod_len;
78 };
79 
80 struct ptrace_sc_ret32 {
81 	uint32_t	sr_retval[2];
82 	int		sr_error;
83 };
84 
85 struct ptrace_vm_entry32 {
86 	int		pve_entry;
87 	int		pve_timestamp;
88 	uint32_t	pve_start;
89 	uint32_t	pve_end;
90 	uint32_t	pve_offset;
91 	u_int		pve_prot;
92 	u_int		pve_pathlen;
93 	int32_t		pve_fileid;
94 	u_int		pve_fsid;
95 	uint32_t	pve_path;
96 };
97 #endif
98 
99 /*
100  * Functions implemented using PROC_ACTION():
101  *
102  * proc_read_regs(proc, regs)
103  *	Get the current user-visible register set from the process
104  *	and copy it into the regs structure (<machine/reg.h>).
105  *	The process is stopped at the time read_regs is called.
106  *
107  * proc_write_regs(proc, regs)
108  *	Update the current register set from the passed in regs
109  *	structure.  Take care to avoid clobbering special CPU
110  *	registers or privileged bits in the PSL.
111  *	Depending on the architecture this may have fix-up work to do,
112  *	especially if the IAR or PCW are modified.
113  *	The process is stopped at the time write_regs is called.
114  *
115  * proc_read_fpregs, proc_write_fpregs
116  *	deal with the floating point register set, otherwise as above.
117  *
118  * proc_read_dbregs, proc_write_dbregs
119  *	deal with the processor debug register set, otherwise as above.
120  *
121  * proc_sstep(proc)
122  *	Arrange for the process to trap after executing a single instruction.
123  */
124 
125 #define	PROC_ACTION(action) do {					\
126 	int error;							\
127 									\
128 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
129 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
130 		error = EIO;						\
131 	else								\
132 		error = (action);					\
133 	return (error);							\
134 } while(0)
135 
136 int
137 proc_read_regs(struct thread *td, struct reg *regs)
138 {
139 
140 	PROC_ACTION(fill_regs(td, regs));
141 }
142 
143 int
144 proc_write_regs(struct thread *td, struct reg *regs)
145 {
146 
147 	PROC_ACTION(set_regs(td, regs));
148 }
149 
150 int
151 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
152 {
153 
154 	PROC_ACTION(fill_dbregs(td, dbregs));
155 }
156 
157 int
158 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
159 {
160 
161 	PROC_ACTION(set_dbregs(td, dbregs));
162 }
163 
164 /*
165  * Ptrace doesn't support fpregs at all, and there are no security holes
166  * or translations for fpregs, so we can just copy them.
167  */
168 int
169 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
170 {
171 
172 	PROC_ACTION(fill_fpregs(td, fpregs));
173 }
174 
175 int
176 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
177 {
178 
179 	PROC_ACTION(set_fpregs(td, fpregs));
180 }
181 
182 #ifdef COMPAT_FREEBSD32
183 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
184 int
185 proc_read_regs32(struct thread *td, struct reg32 *regs32)
186 {
187 
188 	PROC_ACTION(fill_regs32(td, regs32));
189 }
190 
191 int
192 proc_write_regs32(struct thread *td, struct reg32 *regs32)
193 {
194 
195 	PROC_ACTION(set_regs32(td, regs32));
196 }
197 
198 int
199 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
200 {
201 
202 	PROC_ACTION(fill_dbregs32(td, dbregs32));
203 }
204 
205 int
206 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
207 {
208 
209 	PROC_ACTION(set_dbregs32(td, dbregs32));
210 }
211 
212 int
213 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
214 {
215 
216 	PROC_ACTION(fill_fpregs32(td, fpregs32));
217 }
218 
219 int
220 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
221 {
222 
223 	PROC_ACTION(set_fpregs32(td, fpregs32));
224 }
225 #endif
226 
227 int
228 proc_sstep(struct thread *td)
229 {
230 
231 	PROC_ACTION(ptrace_single_step(td));
232 }
233 
234 int
235 proc_rwmem(struct proc *p, struct uio *uio)
236 {
237 	vm_map_t map;
238 	vm_offset_t pageno;		/* page number */
239 	vm_prot_t reqprot;
240 	int error, fault_flags, page_offset, writing;
241 
242 	/*
243 	 * Assert that someone has locked this vmspace.  (Should be
244 	 * curthread but we can't assert that.)  This keeps the process
245 	 * from exiting out from under us until this operation completes.
246 	 */
247 	PROC_ASSERT_HELD(p);
248 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
249 
250 	/*
251 	 * The map we want...
252 	 */
253 	map = &p->p_vmspace->vm_map;
254 
255 	/*
256 	 * If we are writing, then we request vm_fault() to create a private
257 	 * copy of each page.  Since these copies will not be writeable by the
258 	 * process, we must explicity request that they be dirtied.
259 	 */
260 	writing = uio->uio_rw == UIO_WRITE;
261 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
262 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
263 
264 	/*
265 	 * Only map in one page at a time.  We don't have to, but it
266 	 * makes things easier.  This way is trivial - right?
267 	 */
268 	do {
269 		vm_offset_t uva;
270 		u_int len;
271 		vm_page_t m;
272 
273 		uva = (vm_offset_t)uio->uio_offset;
274 
275 		/*
276 		 * Get the page number of this segment.
277 		 */
278 		pageno = trunc_page(uva);
279 		page_offset = uva - pageno;
280 
281 		/*
282 		 * How many bytes to copy
283 		 */
284 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
285 
286 		/*
287 		 * Fault and hold the page on behalf of the process.
288 		 */
289 		error = vm_fault(map, pageno, reqprot, fault_flags, &m);
290 		if (error != KERN_SUCCESS) {
291 			if (error == KERN_RESOURCE_SHORTAGE)
292 				error = ENOMEM;
293 			else
294 				error = EFAULT;
295 			break;
296 		}
297 
298 		/*
299 		 * Now do the i/o move.
300 		 */
301 		error = uiomove_fromphys(&m, page_offset, len, uio);
302 
303 		/* Make the I-cache coherent for breakpoints. */
304 		if (writing && error == 0) {
305 			vm_map_lock_read(map);
306 			if (vm_map_check_protection(map, pageno, pageno +
307 			    PAGE_SIZE, VM_PROT_EXECUTE))
308 				vm_sync_icache(map, uva, len);
309 			vm_map_unlock_read(map);
310 		}
311 
312 		/*
313 		 * Release the page.
314 		 */
315 		vm_page_unwire(m, PQ_ACTIVE);
316 
317 	} while (error == 0 && uio->uio_resid > 0);
318 
319 	return (error);
320 }
321 
322 static ssize_t
323 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
324     size_t len, enum uio_rw rw)
325 {
326 	struct iovec iov;
327 	struct uio uio;
328 	ssize_t slen;
329 
330 	MPASS(len < SSIZE_MAX);
331 	slen = (ssize_t)len;
332 
333 	iov.iov_base = (caddr_t)buf;
334 	iov.iov_len = len;
335 	uio.uio_iov = &iov;
336 	uio.uio_iovcnt = 1;
337 	uio.uio_offset = va;
338 	uio.uio_resid = slen;
339 	uio.uio_segflg = UIO_SYSSPACE;
340 	uio.uio_rw = rw;
341 	uio.uio_td = td;
342 	proc_rwmem(p, &uio);
343 	if (uio.uio_resid == slen)
344 		return (-1);
345 	return (slen - uio.uio_resid);
346 }
347 
348 ssize_t
349 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
350     size_t len)
351 {
352 
353 	return (proc_iop(td, p, va, buf, len, UIO_READ));
354 }
355 
356 ssize_t
357 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
358     size_t len)
359 {
360 
361 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
362 }
363 
364 static int
365 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
366 {
367 	struct vattr vattr;
368 	vm_map_t map;
369 	vm_map_entry_t entry;
370 	vm_object_t obj, tobj, lobj;
371 	struct vmspace *vm;
372 	struct vnode *vp;
373 	char *freepath, *fullpath;
374 	u_int pathlen;
375 	int error, index;
376 
377 	error = 0;
378 	obj = NULL;
379 
380 	vm = vmspace_acquire_ref(p);
381 	map = &vm->vm_map;
382 	vm_map_lock_read(map);
383 
384 	do {
385 		entry = map->header.next;
386 		index = 0;
387 		while (index < pve->pve_entry && entry != &map->header) {
388 			entry = entry->next;
389 			index++;
390 		}
391 		if (index != pve->pve_entry) {
392 			error = EINVAL;
393 			break;
394 		}
395 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
396 		    ("Submap in map header"));
397 		while ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
398 			entry = entry->next;
399 			index++;
400 		}
401 		if (entry == &map->header) {
402 			error = ENOENT;
403 			break;
404 		}
405 
406 		/* We got an entry. */
407 		pve->pve_entry = index + 1;
408 		pve->pve_timestamp = map->timestamp;
409 		pve->pve_start = entry->start;
410 		pve->pve_end = entry->end - 1;
411 		pve->pve_offset = entry->offset;
412 		pve->pve_prot = entry->protection;
413 
414 		/* Backing object's path needed? */
415 		if (pve->pve_pathlen == 0)
416 			break;
417 
418 		pathlen = pve->pve_pathlen;
419 		pve->pve_pathlen = 0;
420 
421 		obj = entry->object.vm_object;
422 		if (obj != NULL)
423 			VM_OBJECT_RLOCK(obj);
424 	} while (0);
425 
426 	vm_map_unlock_read(map);
427 
428 	pve->pve_fsid = VNOVAL;
429 	pve->pve_fileid = VNOVAL;
430 
431 	if (error == 0 && obj != NULL) {
432 		lobj = obj;
433 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
434 			if (tobj != obj)
435 				VM_OBJECT_RLOCK(tobj);
436 			if (lobj != obj)
437 				VM_OBJECT_RUNLOCK(lobj);
438 			lobj = tobj;
439 			pve->pve_offset += tobj->backing_object_offset;
440 		}
441 		vp = vm_object_vnode(lobj);
442 		if (vp != NULL)
443 			vref(vp);
444 		if (lobj != obj)
445 			VM_OBJECT_RUNLOCK(lobj);
446 		VM_OBJECT_RUNLOCK(obj);
447 
448 		if (vp != NULL) {
449 			freepath = NULL;
450 			fullpath = NULL;
451 			vn_fullpath(td, vp, &fullpath, &freepath);
452 			vn_lock(vp, LK_SHARED | LK_RETRY);
453 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
454 				pve->pve_fileid = vattr.va_fileid;
455 				pve->pve_fsid = vattr.va_fsid;
456 			}
457 			vput(vp);
458 
459 			if (fullpath != NULL) {
460 				pve->pve_pathlen = strlen(fullpath) + 1;
461 				if (pve->pve_pathlen <= pathlen) {
462 					error = copyout(fullpath, pve->pve_path,
463 					    pve->pve_pathlen);
464 				} else
465 					error = ENAMETOOLONG;
466 			}
467 			if (freepath != NULL)
468 				free(freepath, M_TEMP);
469 		}
470 	}
471 	vmspace_free(vm);
472 	if (error == 0)
473 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
474 		    p->p_pid, pve->pve_entry, pve->pve_start);
475 
476 	return (error);
477 }
478 
479 #ifdef COMPAT_FREEBSD32
480 static int
481 ptrace_vm_entry32(struct thread *td, struct proc *p,
482     struct ptrace_vm_entry32 *pve32)
483 {
484 	struct ptrace_vm_entry pve;
485 	int error;
486 
487 	pve.pve_entry = pve32->pve_entry;
488 	pve.pve_pathlen = pve32->pve_pathlen;
489 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
490 
491 	error = ptrace_vm_entry(td, p, &pve);
492 	if (error == 0) {
493 		pve32->pve_entry = pve.pve_entry;
494 		pve32->pve_timestamp = pve.pve_timestamp;
495 		pve32->pve_start = pve.pve_start;
496 		pve32->pve_end = pve.pve_end;
497 		pve32->pve_offset = pve.pve_offset;
498 		pve32->pve_prot = pve.pve_prot;
499 		pve32->pve_fileid = pve.pve_fileid;
500 		pve32->pve_fsid = pve.pve_fsid;
501 	}
502 
503 	pve32->pve_pathlen = pve.pve_pathlen;
504 	return (error);
505 }
506 
507 static void
508 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
509     struct ptrace_lwpinfo32 *pl32)
510 {
511 
512 	bzero(pl32, sizeof(*pl32));
513 	pl32->pl_lwpid = pl->pl_lwpid;
514 	pl32->pl_event = pl->pl_event;
515 	pl32->pl_flags = pl->pl_flags;
516 	pl32->pl_sigmask = pl->pl_sigmask;
517 	pl32->pl_siglist = pl->pl_siglist;
518 	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
519 	strcpy(pl32->pl_tdname, pl->pl_tdname);
520 	pl32->pl_child_pid = pl->pl_child_pid;
521 	pl32->pl_syscall_code = pl->pl_syscall_code;
522 	pl32->pl_syscall_narg = pl->pl_syscall_narg;
523 }
524 
525 static void
526 ptrace_sc_ret_to32(const struct ptrace_sc_ret *psr,
527     struct ptrace_sc_ret32 *psr32)
528 {
529 
530 	bzero(psr32, sizeof(*psr32));
531 	psr32->sr_retval[0] = psr->sr_retval[0];
532 	psr32->sr_retval[1] = psr->sr_retval[1];
533 	psr32->sr_error = psr->sr_error;
534 }
535 #endif /* COMPAT_FREEBSD32 */
536 
537 /*
538  * Process debugging system call.
539  */
540 #ifndef _SYS_SYSPROTO_H_
541 struct ptrace_args {
542 	int	req;
543 	pid_t	pid;
544 	caddr_t	addr;
545 	int	data;
546 };
547 #endif
548 
549 #ifdef COMPAT_FREEBSD32
550 /*
551  * This CPP subterfuge is to try and reduce the number of ifdefs in
552  * the body of the code.
553  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
554  * becomes either:
555  *   copyin(uap->addr, &r.reg, sizeof r.reg);
556  * or
557  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
558  * .. except this is done at runtime.
559  */
560 #define	BZERO(a, s)		wrap32 ? \
561 	bzero(a ## 32, s ## 32) : \
562 	bzero(a, s)
563 #define	COPYIN(u, k, s)		wrap32 ? \
564 	copyin(u, k ## 32, s ## 32) : \
565 	copyin(u, k, s)
566 #define	COPYOUT(k, u, s)	wrap32 ? \
567 	copyout(k ## 32, u, s ## 32) : \
568 	copyout(k, u, s)
569 #else
570 #define	BZERO(a, s)		bzero(a, s)
571 #define	COPYIN(u, k, s)		copyin(u, k, s)
572 #define	COPYOUT(k, u, s)	copyout(k, u, s)
573 #endif
574 int
575 sys_ptrace(struct thread *td, struct ptrace_args *uap)
576 {
577 	/*
578 	 * XXX this obfuscation is to reduce stack usage, but the register
579 	 * structs may be too large to put on the stack anyway.
580 	 */
581 	union {
582 		struct ptrace_io_desc piod;
583 		struct ptrace_lwpinfo pl;
584 		struct ptrace_vm_entry pve;
585 		struct dbreg dbreg;
586 		struct fpreg fpreg;
587 		struct reg reg;
588 #ifdef COMPAT_FREEBSD32
589 		struct dbreg32 dbreg32;
590 		struct fpreg32 fpreg32;
591 		struct reg32 reg32;
592 		struct ptrace_io_desc32 piod32;
593 		struct ptrace_lwpinfo32 pl32;
594 		struct ptrace_vm_entry32 pve32;
595 #endif
596 		char args[sizeof(td->td_sa.args)];
597 		struct ptrace_sc_ret psr;
598 		int ptevents;
599 	} r;
600 	void *addr;
601 	int error = 0;
602 #ifdef COMPAT_FREEBSD32
603 	int wrap32 = 0;
604 
605 	if (SV_CURPROC_FLAG(SV_ILP32))
606 		wrap32 = 1;
607 #endif
608 	AUDIT_ARG_PID(uap->pid);
609 	AUDIT_ARG_CMD(uap->req);
610 	AUDIT_ARG_VALUE(uap->data);
611 	addr = &r;
612 	switch (uap->req) {
613 	case PT_GET_EVENT_MASK:
614 	case PT_LWPINFO:
615 	case PT_GET_SC_ARGS:
616 	case PT_GET_SC_RET:
617 		break;
618 	case PT_GETREGS:
619 		BZERO(&r.reg, sizeof r.reg);
620 		break;
621 	case PT_GETFPREGS:
622 		BZERO(&r.fpreg, sizeof r.fpreg);
623 		break;
624 	case PT_GETDBREGS:
625 		BZERO(&r.dbreg, sizeof r.dbreg);
626 		break;
627 	case PT_SETREGS:
628 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
629 		break;
630 	case PT_SETFPREGS:
631 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
632 		break;
633 	case PT_SETDBREGS:
634 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
635 		break;
636 	case PT_SET_EVENT_MASK:
637 		if (uap->data != sizeof(r.ptevents))
638 			error = EINVAL;
639 		else
640 			error = copyin(uap->addr, &r.ptevents, uap->data);
641 		break;
642 	case PT_IO:
643 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
644 		break;
645 	case PT_VM_ENTRY:
646 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
647 		break;
648 	default:
649 		addr = uap->addr;
650 		break;
651 	}
652 	if (error)
653 		return (error);
654 
655 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
656 	if (error)
657 		return (error);
658 
659 	switch (uap->req) {
660 	case PT_VM_ENTRY:
661 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
662 		break;
663 	case PT_IO:
664 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
665 		break;
666 	case PT_GETREGS:
667 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
668 		break;
669 	case PT_GETFPREGS:
670 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
671 		break;
672 	case PT_GETDBREGS:
673 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
674 		break;
675 	case PT_GET_EVENT_MASK:
676 		/* NB: The size in uap->data is validated in kern_ptrace(). */
677 		error = copyout(&r.ptevents, uap->addr, uap->data);
678 		break;
679 	case PT_LWPINFO:
680 		/* NB: The size in uap->data is validated in kern_ptrace(). */
681 		error = copyout(&r.pl, uap->addr, uap->data);
682 		break;
683 	case PT_GET_SC_ARGS:
684 		error = copyout(r.args, uap->addr, MIN(uap->data,
685 		    sizeof(r.args)));
686 		break;
687 	case PT_GET_SC_RET:
688 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
689 		    sizeof(r.psr)));
690 		break;
691 	}
692 
693 	return (error);
694 }
695 #undef COPYIN
696 #undef COPYOUT
697 #undef BZERO
698 
699 #ifdef COMPAT_FREEBSD32
700 /*
701  *   PROC_READ(regs, td2, addr);
702  * becomes either:
703  *   proc_read_regs(td2, addr);
704  * or
705  *   proc_read_regs32(td2, addr);
706  * .. except this is done at runtime.  There is an additional
707  * complication in that PROC_WRITE disallows 32 bit consumers
708  * from writing to 64 bit address space targets.
709  */
710 #define	PROC_READ(w, t, a)	wrap32 ? \
711 	proc_read_ ## w ## 32(t, a) : \
712 	proc_read_ ## w (t, a)
713 #define	PROC_WRITE(w, t, a)	wrap32 ? \
714 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
715 	proc_write_ ## w (t, a)
716 #else
717 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
718 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
719 #endif
720 
721 void
722 proc_set_traced(struct proc *p, bool stop)
723 {
724 
725 	sx_assert(&proctree_lock, SX_XLOCKED);
726 	PROC_LOCK_ASSERT(p, MA_OWNED);
727 	p->p_flag |= P_TRACED;
728 	if (stop)
729 		p->p_flag2 |= P2_PTRACE_FSTP;
730 	p->p_ptevents = PTRACE_DEFAULT;
731 }
732 
733 int
734 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
735 {
736 	struct iovec iov;
737 	struct uio uio;
738 	struct proc *curp, *p, *pp;
739 	struct thread *td2 = NULL, *td3;
740 	struct ptrace_io_desc *piod = NULL;
741 	struct ptrace_lwpinfo *pl;
742 	struct ptrace_sc_ret *psr;
743 	int error, num, tmp;
744 	int proctree_locked = 0;
745 	lwpid_t tid = 0, *buf;
746 #ifdef COMPAT_FREEBSD32
747 	int wrap32 = 0, safe = 0;
748 	struct ptrace_io_desc32 *piod32 = NULL;
749 	struct ptrace_lwpinfo32 *pl32 = NULL;
750 	struct ptrace_sc_ret32 *psr32 = NULL;
751 	union {
752 		struct ptrace_lwpinfo pl;
753 		struct ptrace_sc_ret psr;
754 	} r;
755 #endif
756 
757 	curp = td->td_proc;
758 
759 	/* Lock proctree before locking the process. */
760 	switch (req) {
761 	case PT_TRACE_ME:
762 	case PT_ATTACH:
763 	case PT_STEP:
764 	case PT_CONTINUE:
765 	case PT_TO_SCE:
766 	case PT_TO_SCX:
767 	case PT_SYSCALL:
768 	case PT_FOLLOW_FORK:
769 	case PT_LWP_EVENTS:
770 	case PT_GET_EVENT_MASK:
771 	case PT_SET_EVENT_MASK:
772 	case PT_DETACH:
773 	case PT_GET_SC_ARGS:
774 		sx_xlock(&proctree_lock);
775 		proctree_locked = 1;
776 		break;
777 	default:
778 		break;
779 	}
780 
781 	if (req == PT_TRACE_ME) {
782 		p = td->td_proc;
783 		PROC_LOCK(p);
784 	} else {
785 		if (pid <= PID_MAX) {
786 			if ((p = pfind(pid)) == NULL) {
787 				if (proctree_locked)
788 					sx_xunlock(&proctree_lock);
789 				return (ESRCH);
790 			}
791 		} else {
792 			td2 = tdfind(pid, -1);
793 			if (td2 == NULL) {
794 				if (proctree_locked)
795 					sx_xunlock(&proctree_lock);
796 				return (ESRCH);
797 			}
798 			p = td2->td_proc;
799 			tid = pid;
800 			pid = p->p_pid;
801 		}
802 	}
803 	AUDIT_ARG_PROCESS(p);
804 
805 	if ((p->p_flag & P_WEXIT) != 0) {
806 		error = ESRCH;
807 		goto fail;
808 	}
809 	if ((error = p_cansee(td, p)) != 0)
810 		goto fail;
811 
812 	if ((error = p_candebug(td, p)) != 0)
813 		goto fail;
814 
815 	/*
816 	 * System processes can't be debugged.
817 	 */
818 	if ((p->p_flag & P_SYSTEM) != 0) {
819 		error = EINVAL;
820 		goto fail;
821 	}
822 
823 	if (tid == 0) {
824 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
825 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
826 			td2 = p->p_xthread;
827 		} else {
828 			td2 = FIRST_THREAD_IN_PROC(p);
829 		}
830 		tid = td2->td_tid;
831 	}
832 
833 #ifdef COMPAT_FREEBSD32
834 	/*
835 	 * Test if we're a 32 bit client and what the target is.
836 	 * Set the wrap controls accordingly.
837 	 */
838 	if (SV_CURPROC_FLAG(SV_ILP32)) {
839 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
840 			safe = 1;
841 		wrap32 = 1;
842 	}
843 #endif
844 	/*
845 	 * Permissions check
846 	 */
847 	switch (req) {
848 	case PT_TRACE_ME:
849 		/*
850 		 * Always legal, when there is a parent process which
851 		 * could trace us.  Otherwise, reject.
852 		 */
853 		if ((p->p_flag & P_TRACED) != 0) {
854 			error = EBUSY;
855 			goto fail;
856 		}
857 		if (p->p_pptr == initproc) {
858 			error = EPERM;
859 			goto fail;
860 		}
861 		break;
862 
863 	case PT_ATTACH:
864 		/* Self */
865 		if (p == td->td_proc) {
866 			error = EINVAL;
867 			goto fail;
868 		}
869 
870 		/* Already traced */
871 		if (p->p_flag & P_TRACED) {
872 			error = EBUSY;
873 			goto fail;
874 		}
875 
876 		/* Can't trace an ancestor if you're being traced. */
877 		if (curp->p_flag & P_TRACED) {
878 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
879 				if (pp == p) {
880 					error = EINVAL;
881 					goto fail;
882 				}
883 			}
884 		}
885 
886 
887 		/* OK */
888 		break;
889 
890 	case PT_CLEARSTEP:
891 		/* Allow thread to clear single step for itself */
892 		if (td->td_tid == tid)
893 			break;
894 
895 		/* FALLTHROUGH */
896 	default:
897 		/* not being traced... */
898 		if ((p->p_flag & P_TRACED) == 0) {
899 			error = EPERM;
900 			goto fail;
901 		}
902 
903 		/* not being traced by YOU */
904 		if (p->p_pptr != td->td_proc) {
905 			error = EBUSY;
906 			goto fail;
907 		}
908 
909 		/* not currently stopped */
910 		if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
911 		    p->p_suspcount != p->p_numthreads  ||
912 		    (p->p_flag & P_WAITED) == 0) {
913 			error = EBUSY;
914 			goto fail;
915 		}
916 
917 		/* OK */
918 		break;
919 	}
920 
921 	/* Keep this process around until we finish this request. */
922 	_PHOLD(p);
923 
924 #ifdef FIX_SSTEP
925 	/*
926 	 * Single step fixup ala procfs
927 	 */
928 	FIX_SSTEP(td2);
929 #endif
930 
931 	/*
932 	 * Actually do the requests
933 	 */
934 
935 	td->td_retval[0] = 0;
936 
937 	switch (req) {
938 	case PT_TRACE_ME:
939 		/* set my trace flag and "owner" so it can read/write me */
940 		proc_set_traced(p, false);
941 		if (p->p_flag & P_PPWAIT)
942 			p->p_flag |= P_PPTRACE;
943 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
944 		break;
945 
946 	case PT_ATTACH:
947 		/* security check done above */
948 		/*
949 		 * It would be nice if the tracing relationship was separate
950 		 * from the parent relationship but that would require
951 		 * another set of links in the proc struct or for "wait"
952 		 * to scan the entire proc table.  To make life easier,
953 		 * we just re-parent the process we're trying to trace.
954 		 * The old parent is remembered so we can put things back
955 		 * on a "detach".
956 		 */
957 		proc_set_traced(p, true);
958 		proc_reparent(p, td->td_proc, false);
959 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
960 		    p->p_oppid);
961 
962 		sx_xunlock(&proctree_lock);
963 		proctree_locked = 0;
964 		MPASS(p->p_xthread == NULL);
965 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
966 
967 		/*
968 		 * If already stopped due to a stop signal, clear the
969 		 * existing stop before triggering a traced SIGSTOP.
970 		 */
971 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
972 			PROC_SLOCK(p);
973 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
974 			thread_unsuspend(p);
975 			PROC_SUNLOCK(p);
976 		}
977 
978 		kern_psignal(p, SIGSTOP);
979 		break;
980 
981 	case PT_CLEARSTEP:
982 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
983 		    p->p_pid);
984 		error = ptrace_clear_single_step(td2);
985 		break;
986 
987 	case PT_SETSTEP:
988 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
989 		    p->p_pid);
990 		error = ptrace_single_step(td2);
991 		break;
992 
993 	case PT_SUSPEND:
994 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
995 		    p->p_pid);
996 		td2->td_dbgflags |= TDB_SUSPEND;
997 		thread_lock(td2);
998 		td2->td_flags |= TDF_NEEDSUSPCHK;
999 		thread_unlock(td2);
1000 		break;
1001 
1002 	case PT_RESUME:
1003 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1004 		    p->p_pid);
1005 		td2->td_dbgflags &= ~TDB_SUSPEND;
1006 		break;
1007 
1008 	case PT_FOLLOW_FORK:
1009 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1010 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1011 		    data ? "enabled" : "disabled");
1012 		if (data)
1013 			p->p_ptevents |= PTRACE_FORK;
1014 		else
1015 			p->p_ptevents &= ~PTRACE_FORK;
1016 		break;
1017 
1018 	case PT_LWP_EVENTS:
1019 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1020 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1021 		    data ? "enabled" : "disabled");
1022 		if (data)
1023 			p->p_ptevents |= PTRACE_LWP;
1024 		else
1025 			p->p_ptevents &= ~PTRACE_LWP;
1026 		break;
1027 
1028 	case PT_GET_EVENT_MASK:
1029 		if (data != sizeof(p->p_ptevents)) {
1030 			error = EINVAL;
1031 			break;
1032 		}
1033 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1034 		    p->p_ptevents);
1035 		*(int *)addr = p->p_ptevents;
1036 		break;
1037 
1038 	case PT_SET_EVENT_MASK:
1039 		if (data != sizeof(p->p_ptevents)) {
1040 			error = EINVAL;
1041 			break;
1042 		}
1043 		tmp = *(int *)addr;
1044 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1045 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1046 			error = EINVAL;
1047 			break;
1048 		}
1049 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1050 		    p->p_pid, p->p_ptevents, tmp);
1051 		p->p_ptevents = tmp;
1052 		break;
1053 
1054 	case PT_GET_SC_ARGS:
1055 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1056 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1057 #ifdef COMPAT_FREEBSD32
1058 		    || (wrap32 && !safe)
1059 #endif
1060 		    ) {
1061 			error = EINVAL;
1062 			break;
1063 		}
1064 		bzero(addr, sizeof(td2->td_sa.args));
1065 #ifdef COMPAT_FREEBSD32
1066 		if (wrap32)
1067 			for (num = 0; num < nitems(td2->td_sa.args); num++)
1068 				((uint32_t *)addr)[num] = (uint32_t)
1069 				    td2->td_sa.args[num];
1070 		else
1071 #endif
1072 			bcopy(td2->td_sa.args, addr, td2->td_sa.narg *
1073 			    sizeof(register_t));
1074 		break;
1075 
1076 	case PT_GET_SC_RET:
1077 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1078 #ifdef COMPAT_FREEBSD32
1079 		    || (wrap32 && !safe)
1080 #endif
1081 		    ) {
1082 			error = EINVAL;
1083 			break;
1084 		}
1085 #ifdef COMPAT_FREEBSD32
1086 		if (wrap32) {
1087 			psr = &r.psr;
1088 			psr32 = addr;
1089 		} else
1090 #endif
1091 		psr = addr;
1092 		bzero(psr, sizeof(*psr));
1093 		psr->sr_error = td2->td_errno;
1094 		if (psr->sr_error == 0) {
1095 			psr->sr_retval[0] = td2->td_retval[0];
1096 			psr->sr_retval[1] = td2->td_retval[1];
1097 		}
1098 #ifdef COMPAT_FREEBSD32
1099 		if (wrap32)
1100 			ptrace_sc_ret_to32(psr, psr32);
1101 #endif
1102 		CTR4(KTR_PTRACE,
1103 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1104 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1105 		    psr->sr_retval[1]);
1106 		break;
1107 
1108 	case PT_STEP:
1109 	case PT_CONTINUE:
1110 	case PT_TO_SCE:
1111 	case PT_TO_SCX:
1112 	case PT_SYSCALL:
1113 	case PT_DETACH:
1114 		/* Zero means do not send any signal */
1115 		if (data < 0 || data > _SIG_MAXSIG) {
1116 			error = EINVAL;
1117 			break;
1118 		}
1119 
1120 		switch (req) {
1121 		case PT_STEP:
1122 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1123 			    td2->td_tid, p->p_pid, data);
1124 			error = ptrace_single_step(td2);
1125 			if (error)
1126 				goto out;
1127 			break;
1128 		case PT_CONTINUE:
1129 		case PT_TO_SCE:
1130 		case PT_TO_SCX:
1131 		case PT_SYSCALL:
1132 			if (addr != (void *)1) {
1133 				error = ptrace_set_pc(td2,
1134 				    (u_long)(uintfptr_t)addr);
1135 				if (error)
1136 					goto out;
1137 			}
1138 			switch (req) {
1139 			case PT_TO_SCE:
1140 				p->p_ptevents |= PTRACE_SCE;
1141 				CTR4(KTR_PTRACE,
1142 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1143 				    p->p_pid, p->p_ptevents,
1144 				    (u_long)(uintfptr_t)addr, data);
1145 				break;
1146 			case PT_TO_SCX:
1147 				p->p_ptevents |= PTRACE_SCX;
1148 				CTR4(KTR_PTRACE,
1149 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1150 				    p->p_pid, p->p_ptevents,
1151 				    (u_long)(uintfptr_t)addr, data);
1152 				break;
1153 			case PT_SYSCALL:
1154 				p->p_ptevents |= PTRACE_SYSCALL;
1155 				CTR4(KTR_PTRACE,
1156 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1157 				    p->p_pid, p->p_ptevents,
1158 				    (u_long)(uintfptr_t)addr, data);
1159 				break;
1160 			case PT_CONTINUE:
1161 				CTR3(KTR_PTRACE,
1162 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1163 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1164 				break;
1165 			}
1166 			break;
1167 		case PT_DETACH:
1168 			/*
1169 			 * Reset the process parent.
1170 			 *
1171 			 * NB: This clears P_TRACED before reparenting
1172 			 * a detached process back to its original
1173 			 * parent.  Otherwise the debugee will be set
1174 			 * as an orphan of the debugger.
1175 			 */
1176 			p->p_flag &= ~(P_TRACED | P_WAITED);
1177 			if (p->p_oppid != p->p_pptr->p_pid) {
1178 				PROC_LOCK(p->p_pptr);
1179 				sigqueue_take(p->p_ksi);
1180 				PROC_UNLOCK(p->p_pptr);
1181 
1182 				pp = proc_realparent(p);
1183 				proc_reparent(p, pp, false);
1184 				if (pp == initproc)
1185 					p->p_sigparent = SIGCHLD;
1186 				CTR3(KTR_PTRACE,
1187 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1188 				    p->p_pid, pp->p_pid, data);
1189 			} else
1190 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1191 				    p->p_pid, data);
1192 			p->p_ptevents = 0;
1193 			FOREACH_THREAD_IN_PROC(p, td3) {
1194 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1195 					sigqueue_delete(&td3->td_sigqueue,
1196 					    SIGSTOP);
1197 				}
1198 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1199 				    TDB_SUSPEND);
1200 			}
1201 
1202 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1203 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1204 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1205 			}
1206 
1207 			/* should we send SIGCHLD? */
1208 			/* childproc_continued(p); */
1209 			break;
1210 		}
1211 
1212 		sx_xunlock(&proctree_lock);
1213 		proctree_locked = 0;
1214 
1215 	sendsig:
1216 		MPASS(proctree_locked == 0);
1217 
1218 		/*
1219 		 * Clear the pending event for the thread that just
1220 		 * reported its event (p_xthread).  This may not be
1221 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1222 		 * the debugger is resuming a different thread.
1223 		 *
1224 		 * Deliver any pending signal via the reporting thread.
1225 		 */
1226 		MPASS(p->p_xthread != NULL);
1227 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1228 		p->p_xthread->td_xsig = data;
1229 		p->p_xthread = NULL;
1230 		p->p_xsig = data;
1231 
1232 		/*
1233 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1234 		 * always works immediately, even if another thread is
1235 		 * unsuspended first and attempts to handle a
1236 		 * different signal or if the POSIX.1b style signal
1237 		 * queue cannot accommodate any new signals.
1238 		 */
1239 		if (data == SIGKILL)
1240 			proc_wkilled(p);
1241 
1242 		/*
1243 		 * Unsuspend all threads.  To leave a thread
1244 		 * suspended, use PT_SUSPEND to suspend it before
1245 		 * continuing the process.
1246 		 */
1247 		PROC_SLOCK(p);
1248 		p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
1249 		thread_unsuspend(p);
1250 		PROC_SUNLOCK(p);
1251 		break;
1252 
1253 	case PT_WRITE_I:
1254 	case PT_WRITE_D:
1255 		td2->td_dbgflags |= TDB_USERWR;
1256 		PROC_UNLOCK(p);
1257 		error = 0;
1258 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1259 		    sizeof(int)) != sizeof(int))
1260 			error = ENOMEM;
1261 		else
1262 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1263 			    p->p_pid, addr, data);
1264 		PROC_LOCK(p);
1265 		break;
1266 
1267 	case PT_READ_I:
1268 	case PT_READ_D:
1269 		PROC_UNLOCK(p);
1270 		error = tmp = 0;
1271 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1272 		    sizeof(int)) != sizeof(int))
1273 			error = ENOMEM;
1274 		else
1275 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1276 			    p->p_pid, addr, tmp);
1277 		td->td_retval[0] = tmp;
1278 		PROC_LOCK(p);
1279 		break;
1280 
1281 	case PT_IO:
1282 #ifdef COMPAT_FREEBSD32
1283 		if (wrap32) {
1284 			piod32 = addr;
1285 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1286 			iov.iov_len = piod32->piod_len;
1287 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1288 			uio.uio_resid = piod32->piod_len;
1289 		} else
1290 #endif
1291 		{
1292 			piod = addr;
1293 			iov.iov_base = piod->piod_addr;
1294 			iov.iov_len = piod->piod_len;
1295 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1296 			uio.uio_resid = piod->piod_len;
1297 		}
1298 		uio.uio_iov = &iov;
1299 		uio.uio_iovcnt = 1;
1300 		uio.uio_segflg = UIO_USERSPACE;
1301 		uio.uio_td = td;
1302 #ifdef COMPAT_FREEBSD32
1303 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1304 #else
1305 		tmp = piod->piod_op;
1306 #endif
1307 		switch (tmp) {
1308 		case PIOD_READ_D:
1309 		case PIOD_READ_I:
1310 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1311 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1312 			uio.uio_rw = UIO_READ;
1313 			break;
1314 		case PIOD_WRITE_D:
1315 		case PIOD_WRITE_I:
1316 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1317 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1318 			td2->td_dbgflags |= TDB_USERWR;
1319 			uio.uio_rw = UIO_WRITE;
1320 			break;
1321 		default:
1322 			error = EINVAL;
1323 			goto out;
1324 		}
1325 		PROC_UNLOCK(p);
1326 		error = proc_rwmem(p, &uio);
1327 #ifdef COMPAT_FREEBSD32
1328 		if (wrap32)
1329 			piod32->piod_len -= uio.uio_resid;
1330 		else
1331 #endif
1332 			piod->piod_len -= uio.uio_resid;
1333 		PROC_LOCK(p);
1334 		break;
1335 
1336 	case PT_KILL:
1337 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1338 		data = SIGKILL;
1339 		goto sendsig;	/* in PT_CONTINUE above */
1340 
1341 	case PT_SETREGS:
1342 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1343 		    p->p_pid);
1344 		td2->td_dbgflags |= TDB_USERWR;
1345 		error = PROC_WRITE(regs, td2, addr);
1346 		break;
1347 
1348 	case PT_GETREGS:
1349 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1350 		    p->p_pid);
1351 		error = PROC_READ(regs, td2, addr);
1352 		break;
1353 
1354 	case PT_SETFPREGS:
1355 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1356 		    p->p_pid);
1357 		td2->td_dbgflags |= TDB_USERWR;
1358 		error = PROC_WRITE(fpregs, td2, addr);
1359 		break;
1360 
1361 	case PT_GETFPREGS:
1362 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1363 		    p->p_pid);
1364 		error = PROC_READ(fpregs, td2, addr);
1365 		break;
1366 
1367 	case PT_SETDBREGS:
1368 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1369 		    p->p_pid);
1370 		td2->td_dbgflags |= TDB_USERWR;
1371 		error = PROC_WRITE(dbregs, td2, addr);
1372 		break;
1373 
1374 	case PT_GETDBREGS:
1375 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1376 		    p->p_pid);
1377 		error = PROC_READ(dbregs, td2, addr);
1378 		break;
1379 
1380 	case PT_LWPINFO:
1381 		if (data <= 0 ||
1382 #ifdef COMPAT_FREEBSD32
1383 		    (!wrap32 && data > sizeof(*pl)) ||
1384 		    (wrap32 && data > sizeof(*pl32))) {
1385 #else
1386 		    data > sizeof(*pl)) {
1387 #endif
1388 			error = EINVAL;
1389 			break;
1390 		}
1391 #ifdef COMPAT_FREEBSD32
1392 		if (wrap32) {
1393 			pl = &r.pl;
1394 			pl32 = addr;
1395 		} else
1396 #endif
1397 		pl = addr;
1398 		bzero(pl, sizeof(*pl));
1399 		pl->pl_lwpid = td2->td_tid;
1400 		pl->pl_event = PL_EVENT_NONE;
1401 		pl->pl_flags = 0;
1402 		if (td2->td_dbgflags & TDB_XSIG) {
1403 			pl->pl_event = PL_EVENT_SIGNAL;
1404 			if (td2->td_si.si_signo != 0 &&
1405 #ifdef COMPAT_FREEBSD32
1406 			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1407 			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1408 			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1409 			    pl_siginfo) + sizeof(struct siginfo32)))
1410 #else
1411 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1412 			    + sizeof(pl->pl_siginfo)
1413 #endif
1414 			){
1415 				pl->pl_flags |= PL_FLAG_SI;
1416 				pl->pl_siginfo = td2->td_si;
1417 			}
1418 		}
1419 		if (td2->td_dbgflags & TDB_SCE)
1420 			pl->pl_flags |= PL_FLAG_SCE;
1421 		else if (td2->td_dbgflags & TDB_SCX)
1422 			pl->pl_flags |= PL_FLAG_SCX;
1423 		if (td2->td_dbgflags & TDB_EXEC)
1424 			pl->pl_flags |= PL_FLAG_EXEC;
1425 		if (td2->td_dbgflags & TDB_FORK) {
1426 			pl->pl_flags |= PL_FLAG_FORKED;
1427 			pl->pl_child_pid = td2->td_dbg_forked;
1428 			if (td2->td_dbgflags & TDB_VFORK)
1429 				pl->pl_flags |= PL_FLAG_VFORKED;
1430 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1431 		    TDB_VFORK)
1432 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1433 		if (td2->td_dbgflags & TDB_CHILD)
1434 			pl->pl_flags |= PL_FLAG_CHILD;
1435 		if (td2->td_dbgflags & TDB_BORN)
1436 			pl->pl_flags |= PL_FLAG_BORN;
1437 		if (td2->td_dbgflags & TDB_EXIT)
1438 			pl->pl_flags |= PL_FLAG_EXITED;
1439 		pl->pl_sigmask = td2->td_sigmask;
1440 		pl->pl_siglist = td2->td_siglist;
1441 		strcpy(pl->pl_tdname, td2->td_name);
1442 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1443 			pl->pl_syscall_code = td2->td_sa.code;
1444 			pl->pl_syscall_narg = td2->td_sa.narg;
1445 		} else {
1446 			pl->pl_syscall_code = 0;
1447 			pl->pl_syscall_narg = 0;
1448 		}
1449 #ifdef COMPAT_FREEBSD32
1450 		if (wrap32)
1451 			ptrace_lwpinfo_to32(pl, pl32);
1452 #endif
1453 		CTR6(KTR_PTRACE,
1454     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1455 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1456 		    pl->pl_child_pid, pl->pl_syscall_code);
1457 		break;
1458 
1459 	case PT_GETNUMLWPS:
1460 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1461 		    p->p_numthreads);
1462 		td->td_retval[0] = p->p_numthreads;
1463 		break;
1464 
1465 	case PT_GETLWPLIST:
1466 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1467 		    p->p_pid, data, p->p_numthreads);
1468 		if (data <= 0) {
1469 			error = EINVAL;
1470 			break;
1471 		}
1472 		num = imin(p->p_numthreads, data);
1473 		PROC_UNLOCK(p);
1474 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1475 		tmp = 0;
1476 		PROC_LOCK(p);
1477 		FOREACH_THREAD_IN_PROC(p, td2) {
1478 			if (tmp >= num)
1479 				break;
1480 			buf[tmp++] = td2->td_tid;
1481 		}
1482 		PROC_UNLOCK(p);
1483 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1484 		free(buf, M_TEMP);
1485 		if (!error)
1486 			td->td_retval[0] = tmp;
1487 		PROC_LOCK(p);
1488 		break;
1489 
1490 	case PT_VM_TIMESTAMP:
1491 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1492 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1493 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1494 		break;
1495 
1496 	case PT_VM_ENTRY:
1497 		PROC_UNLOCK(p);
1498 #ifdef COMPAT_FREEBSD32
1499 		if (wrap32)
1500 			error = ptrace_vm_entry32(td, p, addr);
1501 		else
1502 #endif
1503 		error = ptrace_vm_entry(td, p, addr);
1504 		PROC_LOCK(p);
1505 		break;
1506 
1507 	default:
1508 #ifdef __HAVE_PTRACE_MACHDEP
1509 		if (req >= PT_FIRSTMACH) {
1510 			PROC_UNLOCK(p);
1511 			error = cpu_ptrace(td2, req, addr, data);
1512 			PROC_LOCK(p);
1513 		} else
1514 #endif
1515 			/* Unknown request. */
1516 			error = EINVAL;
1517 		break;
1518 	}
1519 
1520 out:
1521 	/* Drop our hold on this process now that the request has completed. */
1522 	_PRELE(p);
1523 fail:
1524 	PROC_UNLOCK(p);
1525 	if (proctree_locked)
1526 		sx_xunlock(&proctree_lock);
1527 	return (error);
1528 }
1529 #undef PROC_READ
1530 #undef PROC_WRITE
1531 
1532 /*
1533  * Stop a process because of a debugging event;
1534  * stay stopped until p->p_step is cleared
1535  * (cleared by PIOCCONT in procfs).
1536  */
1537 void
1538 stopevent(struct proc *p, unsigned int event, unsigned int val)
1539 {
1540 
1541 	PROC_LOCK_ASSERT(p, MA_OWNED);
1542 	p->p_step = 1;
1543 	CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1544 	    val);
1545 	do {
1546 		if (event != S_EXIT)
1547 			p->p_xsig = val;
1548 		p->p_xthread = NULL;
1549 		p->p_stype = event;	/* Which event caused the stop? */
1550 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1551 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1552 	} while (p->p_step);
1553 }
1554