xref: /freebsd/sys/kern/sys_process.c (revision 2f513db7)
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 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
386 		    ("Submap in map header"));
387 		index = 0;
388 		VM_MAP_ENTRY_FOREACH(entry, map) {
389 			if (index >= pve->pve_entry &&
390 			    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
391 				break;
392 			index++;
393 		}
394 		if (index < pve->pve_entry) {
395 			error = EINVAL;
396 			break;
397 		}
398 		if (entry == &map->header) {
399 			error = ENOENT;
400 			break;
401 		}
402 
403 		/* We got an entry. */
404 		pve->pve_entry = index + 1;
405 		pve->pve_timestamp = map->timestamp;
406 		pve->pve_start = entry->start;
407 		pve->pve_end = entry->end - 1;
408 		pve->pve_offset = entry->offset;
409 		pve->pve_prot = entry->protection;
410 
411 		/* Backing object's path needed? */
412 		if (pve->pve_pathlen == 0)
413 			break;
414 
415 		pathlen = pve->pve_pathlen;
416 		pve->pve_pathlen = 0;
417 
418 		obj = entry->object.vm_object;
419 		if (obj != NULL)
420 			VM_OBJECT_RLOCK(obj);
421 	} while (0);
422 
423 	vm_map_unlock_read(map);
424 
425 	pve->pve_fsid = VNOVAL;
426 	pve->pve_fileid = VNOVAL;
427 
428 	if (error == 0 && obj != NULL) {
429 		lobj = obj;
430 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
431 			if (tobj != obj)
432 				VM_OBJECT_RLOCK(tobj);
433 			if (lobj != obj)
434 				VM_OBJECT_RUNLOCK(lobj);
435 			lobj = tobj;
436 			pve->pve_offset += tobj->backing_object_offset;
437 		}
438 		vp = vm_object_vnode(lobj);
439 		if (vp != NULL)
440 			vref(vp);
441 		if (lobj != obj)
442 			VM_OBJECT_RUNLOCK(lobj);
443 		VM_OBJECT_RUNLOCK(obj);
444 
445 		if (vp != NULL) {
446 			freepath = NULL;
447 			fullpath = NULL;
448 			vn_fullpath(td, vp, &fullpath, &freepath);
449 			vn_lock(vp, LK_SHARED | LK_RETRY);
450 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
451 				pve->pve_fileid = vattr.va_fileid;
452 				pve->pve_fsid = vattr.va_fsid;
453 			}
454 			vput(vp);
455 
456 			if (fullpath != NULL) {
457 				pve->pve_pathlen = strlen(fullpath) + 1;
458 				if (pve->pve_pathlen <= pathlen) {
459 					error = copyout(fullpath, pve->pve_path,
460 					    pve->pve_pathlen);
461 				} else
462 					error = ENAMETOOLONG;
463 			}
464 			if (freepath != NULL)
465 				free(freepath, M_TEMP);
466 		}
467 	}
468 	vmspace_free(vm);
469 	if (error == 0)
470 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
471 		    p->p_pid, pve->pve_entry, pve->pve_start);
472 
473 	return (error);
474 }
475 
476 #ifdef COMPAT_FREEBSD32
477 static int
478 ptrace_vm_entry32(struct thread *td, struct proc *p,
479     struct ptrace_vm_entry32 *pve32)
480 {
481 	struct ptrace_vm_entry pve;
482 	int error;
483 
484 	pve.pve_entry = pve32->pve_entry;
485 	pve.pve_pathlen = pve32->pve_pathlen;
486 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
487 
488 	error = ptrace_vm_entry(td, p, &pve);
489 	if (error == 0) {
490 		pve32->pve_entry = pve.pve_entry;
491 		pve32->pve_timestamp = pve.pve_timestamp;
492 		pve32->pve_start = pve.pve_start;
493 		pve32->pve_end = pve.pve_end;
494 		pve32->pve_offset = pve.pve_offset;
495 		pve32->pve_prot = pve.pve_prot;
496 		pve32->pve_fileid = pve.pve_fileid;
497 		pve32->pve_fsid = pve.pve_fsid;
498 	}
499 
500 	pve32->pve_pathlen = pve.pve_pathlen;
501 	return (error);
502 }
503 
504 static void
505 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
506     struct ptrace_lwpinfo32 *pl32)
507 {
508 
509 	bzero(pl32, sizeof(*pl32));
510 	pl32->pl_lwpid = pl->pl_lwpid;
511 	pl32->pl_event = pl->pl_event;
512 	pl32->pl_flags = pl->pl_flags;
513 	pl32->pl_sigmask = pl->pl_sigmask;
514 	pl32->pl_siglist = pl->pl_siglist;
515 	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
516 	strcpy(pl32->pl_tdname, pl->pl_tdname);
517 	pl32->pl_child_pid = pl->pl_child_pid;
518 	pl32->pl_syscall_code = pl->pl_syscall_code;
519 	pl32->pl_syscall_narg = pl->pl_syscall_narg;
520 }
521 
522 static void
523 ptrace_sc_ret_to32(const struct ptrace_sc_ret *psr,
524     struct ptrace_sc_ret32 *psr32)
525 {
526 
527 	bzero(psr32, sizeof(*psr32));
528 	psr32->sr_retval[0] = psr->sr_retval[0];
529 	psr32->sr_retval[1] = psr->sr_retval[1];
530 	psr32->sr_error = psr->sr_error;
531 }
532 #endif /* COMPAT_FREEBSD32 */
533 
534 /*
535  * Process debugging system call.
536  */
537 #ifndef _SYS_SYSPROTO_H_
538 struct ptrace_args {
539 	int	req;
540 	pid_t	pid;
541 	caddr_t	addr;
542 	int	data;
543 };
544 #endif
545 
546 #ifdef COMPAT_FREEBSD32
547 /*
548  * This CPP subterfuge is to try and reduce the number of ifdefs in
549  * the body of the code.
550  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
551  * becomes either:
552  *   copyin(uap->addr, &r.reg, sizeof r.reg);
553  * or
554  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
555  * .. except this is done at runtime.
556  */
557 #define	BZERO(a, s)		wrap32 ? \
558 	bzero(a ## 32, s ## 32) : \
559 	bzero(a, s)
560 #define	COPYIN(u, k, s)		wrap32 ? \
561 	copyin(u, k ## 32, s ## 32) : \
562 	copyin(u, k, s)
563 #define	COPYOUT(k, u, s)	wrap32 ? \
564 	copyout(k ## 32, u, s ## 32) : \
565 	copyout(k, u, s)
566 #else
567 #define	BZERO(a, s)		bzero(a, s)
568 #define	COPYIN(u, k, s)		copyin(u, k, s)
569 #define	COPYOUT(k, u, s)	copyout(k, u, s)
570 #endif
571 int
572 sys_ptrace(struct thread *td, struct ptrace_args *uap)
573 {
574 	/*
575 	 * XXX this obfuscation is to reduce stack usage, but the register
576 	 * structs may be too large to put on the stack anyway.
577 	 */
578 	union {
579 		struct ptrace_io_desc piod;
580 		struct ptrace_lwpinfo pl;
581 		struct ptrace_vm_entry pve;
582 		struct dbreg dbreg;
583 		struct fpreg fpreg;
584 		struct reg reg;
585 #ifdef COMPAT_FREEBSD32
586 		struct dbreg32 dbreg32;
587 		struct fpreg32 fpreg32;
588 		struct reg32 reg32;
589 		struct ptrace_io_desc32 piod32;
590 		struct ptrace_lwpinfo32 pl32;
591 		struct ptrace_vm_entry32 pve32;
592 #endif
593 		char args[sizeof(td->td_sa.args)];
594 		struct ptrace_sc_ret psr;
595 		int ptevents;
596 	} r;
597 	void *addr;
598 	int error = 0;
599 #ifdef COMPAT_FREEBSD32
600 	int wrap32 = 0;
601 
602 	if (SV_CURPROC_FLAG(SV_ILP32))
603 		wrap32 = 1;
604 #endif
605 	AUDIT_ARG_PID(uap->pid);
606 	AUDIT_ARG_CMD(uap->req);
607 	AUDIT_ARG_VALUE(uap->data);
608 	addr = &r;
609 	switch (uap->req) {
610 	case PT_GET_EVENT_MASK:
611 	case PT_LWPINFO:
612 	case PT_GET_SC_ARGS:
613 	case PT_GET_SC_RET:
614 		break;
615 	case PT_GETREGS:
616 		BZERO(&r.reg, sizeof r.reg);
617 		break;
618 	case PT_GETFPREGS:
619 		BZERO(&r.fpreg, sizeof r.fpreg);
620 		break;
621 	case PT_GETDBREGS:
622 		BZERO(&r.dbreg, sizeof r.dbreg);
623 		break;
624 	case PT_SETREGS:
625 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
626 		break;
627 	case PT_SETFPREGS:
628 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
629 		break;
630 	case PT_SETDBREGS:
631 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
632 		break;
633 	case PT_SET_EVENT_MASK:
634 		if (uap->data != sizeof(r.ptevents))
635 			error = EINVAL;
636 		else
637 			error = copyin(uap->addr, &r.ptevents, uap->data);
638 		break;
639 	case PT_IO:
640 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
641 		break;
642 	case PT_VM_ENTRY:
643 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
644 		break;
645 	default:
646 		addr = uap->addr;
647 		break;
648 	}
649 	if (error)
650 		return (error);
651 
652 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
653 	if (error)
654 		return (error);
655 
656 	switch (uap->req) {
657 	case PT_VM_ENTRY:
658 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
659 		break;
660 	case PT_IO:
661 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
662 		break;
663 	case PT_GETREGS:
664 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
665 		break;
666 	case PT_GETFPREGS:
667 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
668 		break;
669 	case PT_GETDBREGS:
670 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
671 		break;
672 	case PT_GET_EVENT_MASK:
673 		/* NB: The size in uap->data is validated in kern_ptrace(). */
674 		error = copyout(&r.ptevents, uap->addr, uap->data);
675 		break;
676 	case PT_LWPINFO:
677 		/* NB: The size in uap->data is validated in kern_ptrace(). */
678 		error = copyout(&r.pl, uap->addr, uap->data);
679 		break;
680 	case PT_GET_SC_ARGS:
681 		error = copyout(r.args, uap->addr, MIN(uap->data,
682 		    sizeof(r.args)));
683 		break;
684 	case PT_GET_SC_RET:
685 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
686 		    sizeof(r.psr)));
687 		break;
688 	}
689 
690 	return (error);
691 }
692 #undef COPYIN
693 #undef COPYOUT
694 #undef BZERO
695 
696 #ifdef COMPAT_FREEBSD32
697 /*
698  *   PROC_READ(regs, td2, addr);
699  * becomes either:
700  *   proc_read_regs(td2, addr);
701  * or
702  *   proc_read_regs32(td2, addr);
703  * .. except this is done at runtime.  There is an additional
704  * complication in that PROC_WRITE disallows 32 bit consumers
705  * from writing to 64 bit address space targets.
706  */
707 #define	PROC_READ(w, t, a)	wrap32 ? \
708 	proc_read_ ## w ## 32(t, a) : \
709 	proc_read_ ## w (t, a)
710 #define	PROC_WRITE(w, t, a)	wrap32 ? \
711 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
712 	proc_write_ ## w (t, a)
713 #else
714 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
715 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
716 #endif
717 
718 void
719 proc_set_traced(struct proc *p, bool stop)
720 {
721 
722 	sx_assert(&proctree_lock, SX_XLOCKED);
723 	PROC_LOCK_ASSERT(p, MA_OWNED);
724 	p->p_flag |= P_TRACED;
725 	if (stop)
726 		p->p_flag2 |= P2_PTRACE_FSTP;
727 	p->p_ptevents = PTRACE_DEFAULT;
728 }
729 
730 int
731 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
732 {
733 	struct iovec iov;
734 	struct uio uio;
735 	struct proc *curp, *p, *pp;
736 	struct thread *td2 = NULL, *td3;
737 	struct ptrace_io_desc *piod = NULL;
738 	struct ptrace_lwpinfo *pl;
739 	struct ptrace_sc_ret *psr;
740 	int error, num, tmp;
741 	int proctree_locked = 0;
742 	lwpid_t tid = 0, *buf;
743 #ifdef COMPAT_FREEBSD32
744 	int wrap32 = 0, safe = 0;
745 	struct ptrace_io_desc32 *piod32 = NULL;
746 	struct ptrace_lwpinfo32 *pl32 = NULL;
747 	struct ptrace_sc_ret32 *psr32 = NULL;
748 	union {
749 		struct ptrace_lwpinfo pl;
750 		struct ptrace_sc_ret psr;
751 	} r;
752 #endif
753 
754 	curp = td->td_proc;
755 
756 	/* Lock proctree before locking the process. */
757 	switch (req) {
758 	case PT_TRACE_ME:
759 	case PT_ATTACH:
760 	case PT_STEP:
761 	case PT_CONTINUE:
762 	case PT_TO_SCE:
763 	case PT_TO_SCX:
764 	case PT_SYSCALL:
765 	case PT_FOLLOW_FORK:
766 	case PT_LWP_EVENTS:
767 	case PT_GET_EVENT_MASK:
768 	case PT_SET_EVENT_MASK:
769 	case PT_DETACH:
770 	case PT_GET_SC_ARGS:
771 		sx_xlock(&proctree_lock);
772 		proctree_locked = 1;
773 		break;
774 	default:
775 		break;
776 	}
777 
778 	if (req == PT_TRACE_ME) {
779 		p = td->td_proc;
780 		PROC_LOCK(p);
781 	} else {
782 		if (pid <= PID_MAX) {
783 			if ((p = pfind(pid)) == NULL) {
784 				if (proctree_locked)
785 					sx_xunlock(&proctree_lock);
786 				return (ESRCH);
787 			}
788 		} else {
789 			td2 = tdfind(pid, -1);
790 			if (td2 == NULL) {
791 				if (proctree_locked)
792 					sx_xunlock(&proctree_lock);
793 				return (ESRCH);
794 			}
795 			p = td2->td_proc;
796 			tid = pid;
797 			pid = p->p_pid;
798 		}
799 	}
800 	AUDIT_ARG_PROCESS(p);
801 
802 	if ((p->p_flag & P_WEXIT) != 0) {
803 		error = ESRCH;
804 		goto fail;
805 	}
806 	if ((error = p_cansee(td, p)) != 0)
807 		goto fail;
808 
809 	if ((error = p_candebug(td, p)) != 0)
810 		goto fail;
811 
812 	/*
813 	 * System processes can't be debugged.
814 	 */
815 	if ((p->p_flag & P_SYSTEM) != 0) {
816 		error = EINVAL;
817 		goto fail;
818 	}
819 
820 	if (tid == 0) {
821 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
822 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
823 			td2 = p->p_xthread;
824 		} else {
825 			td2 = FIRST_THREAD_IN_PROC(p);
826 		}
827 		tid = td2->td_tid;
828 	}
829 
830 #ifdef COMPAT_FREEBSD32
831 	/*
832 	 * Test if we're a 32 bit client and what the target is.
833 	 * Set the wrap controls accordingly.
834 	 */
835 	if (SV_CURPROC_FLAG(SV_ILP32)) {
836 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
837 			safe = 1;
838 		wrap32 = 1;
839 	}
840 #endif
841 	/*
842 	 * Permissions check
843 	 */
844 	switch (req) {
845 	case PT_TRACE_ME:
846 		/*
847 		 * Always legal, when there is a parent process which
848 		 * could trace us.  Otherwise, reject.
849 		 */
850 		if ((p->p_flag & P_TRACED) != 0) {
851 			error = EBUSY;
852 			goto fail;
853 		}
854 		if (p->p_pptr == initproc) {
855 			error = EPERM;
856 			goto fail;
857 		}
858 		break;
859 
860 	case PT_ATTACH:
861 		/* Self */
862 		if (p == td->td_proc) {
863 			error = EINVAL;
864 			goto fail;
865 		}
866 
867 		/* Already traced */
868 		if (p->p_flag & P_TRACED) {
869 			error = EBUSY;
870 			goto fail;
871 		}
872 
873 		/* Can't trace an ancestor if you're being traced. */
874 		if (curp->p_flag & P_TRACED) {
875 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
876 				if (pp == p) {
877 					error = EINVAL;
878 					goto fail;
879 				}
880 			}
881 		}
882 
883 		/* OK */
884 		break;
885 
886 	case PT_CLEARSTEP:
887 		/* Allow thread to clear single step for itself */
888 		if (td->td_tid == tid)
889 			break;
890 
891 		/* FALLTHROUGH */
892 	default:
893 		/* not being traced... */
894 		if ((p->p_flag & P_TRACED) == 0) {
895 			error = EPERM;
896 			goto fail;
897 		}
898 
899 		/* not being traced by YOU */
900 		if (p->p_pptr != td->td_proc) {
901 			error = EBUSY;
902 			goto fail;
903 		}
904 
905 		/* not currently stopped */
906 		if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
907 		    p->p_suspcount != p->p_numthreads  ||
908 		    (p->p_flag & P_WAITED) == 0) {
909 			error = EBUSY;
910 			goto fail;
911 		}
912 
913 		/* OK */
914 		break;
915 	}
916 
917 	/* Keep this process around until we finish this request. */
918 	_PHOLD(p);
919 
920 #ifdef FIX_SSTEP
921 	/*
922 	 * Single step fixup ala procfs
923 	 */
924 	FIX_SSTEP(td2);
925 #endif
926 
927 	/*
928 	 * Actually do the requests
929 	 */
930 
931 	td->td_retval[0] = 0;
932 
933 	switch (req) {
934 	case PT_TRACE_ME:
935 		/* set my trace flag and "owner" so it can read/write me */
936 		proc_set_traced(p, false);
937 		if (p->p_flag & P_PPWAIT)
938 			p->p_flag |= P_PPTRACE;
939 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
940 		break;
941 
942 	case PT_ATTACH:
943 		/* security check done above */
944 		/*
945 		 * It would be nice if the tracing relationship was separate
946 		 * from the parent relationship but that would require
947 		 * another set of links in the proc struct or for "wait"
948 		 * to scan the entire proc table.  To make life easier,
949 		 * we just re-parent the process we're trying to trace.
950 		 * The old parent is remembered so we can put things back
951 		 * on a "detach".
952 		 */
953 		proc_set_traced(p, true);
954 		proc_reparent(p, td->td_proc, false);
955 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
956 		    p->p_oppid);
957 
958 		sx_xunlock(&proctree_lock);
959 		proctree_locked = 0;
960 		MPASS(p->p_xthread == NULL);
961 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
962 
963 		/*
964 		 * If already stopped due to a stop signal, clear the
965 		 * existing stop before triggering a traced SIGSTOP.
966 		 */
967 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
968 			PROC_SLOCK(p);
969 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
970 			thread_unsuspend(p);
971 			PROC_SUNLOCK(p);
972 		}
973 
974 		kern_psignal(p, SIGSTOP);
975 		break;
976 
977 	case PT_CLEARSTEP:
978 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
979 		    p->p_pid);
980 		error = ptrace_clear_single_step(td2);
981 		break;
982 
983 	case PT_SETSTEP:
984 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
985 		    p->p_pid);
986 		error = ptrace_single_step(td2);
987 		break;
988 
989 	case PT_SUSPEND:
990 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
991 		    p->p_pid);
992 		td2->td_dbgflags |= TDB_SUSPEND;
993 		thread_lock(td2);
994 		td2->td_flags |= TDF_NEEDSUSPCHK;
995 		thread_unlock(td2);
996 		break;
997 
998 	case PT_RESUME:
999 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1000 		    p->p_pid);
1001 		td2->td_dbgflags &= ~TDB_SUSPEND;
1002 		break;
1003 
1004 	case PT_FOLLOW_FORK:
1005 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1006 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1007 		    data ? "enabled" : "disabled");
1008 		if (data)
1009 			p->p_ptevents |= PTRACE_FORK;
1010 		else
1011 			p->p_ptevents &= ~PTRACE_FORK;
1012 		break;
1013 
1014 	case PT_LWP_EVENTS:
1015 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1016 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1017 		    data ? "enabled" : "disabled");
1018 		if (data)
1019 			p->p_ptevents |= PTRACE_LWP;
1020 		else
1021 			p->p_ptevents &= ~PTRACE_LWP;
1022 		break;
1023 
1024 	case PT_GET_EVENT_MASK:
1025 		if (data != sizeof(p->p_ptevents)) {
1026 			error = EINVAL;
1027 			break;
1028 		}
1029 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1030 		    p->p_ptevents);
1031 		*(int *)addr = p->p_ptevents;
1032 		break;
1033 
1034 	case PT_SET_EVENT_MASK:
1035 		if (data != sizeof(p->p_ptevents)) {
1036 			error = EINVAL;
1037 			break;
1038 		}
1039 		tmp = *(int *)addr;
1040 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1041 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1042 			error = EINVAL;
1043 			break;
1044 		}
1045 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1046 		    p->p_pid, p->p_ptevents, tmp);
1047 		p->p_ptevents = tmp;
1048 		break;
1049 
1050 	case PT_GET_SC_ARGS:
1051 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1052 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1053 #ifdef COMPAT_FREEBSD32
1054 		    || (wrap32 && !safe)
1055 #endif
1056 		    ) {
1057 			error = EINVAL;
1058 			break;
1059 		}
1060 		bzero(addr, sizeof(td2->td_sa.args));
1061 #ifdef COMPAT_FREEBSD32
1062 		if (wrap32)
1063 			for (num = 0; num < nitems(td2->td_sa.args); num++)
1064 				((uint32_t *)addr)[num] = (uint32_t)
1065 				    td2->td_sa.args[num];
1066 		else
1067 #endif
1068 			bcopy(td2->td_sa.args, addr, td2->td_sa.narg *
1069 			    sizeof(register_t));
1070 		break;
1071 
1072 	case PT_GET_SC_RET:
1073 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1074 #ifdef COMPAT_FREEBSD32
1075 		    || (wrap32 && !safe)
1076 #endif
1077 		    ) {
1078 			error = EINVAL;
1079 			break;
1080 		}
1081 #ifdef COMPAT_FREEBSD32
1082 		if (wrap32) {
1083 			psr = &r.psr;
1084 			psr32 = addr;
1085 		} else
1086 #endif
1087 		psr = addr;
1088 		bzero(psr, sizeof(*psr));
1089 		psr->sr_error = td2->td_errno;
1090 		if (psr->sr_error == 0) {
1091 			psr->sr_retval[0] = td2->td_retval[0];
1092 			psr->sr_retval[1] = td2->td_retval[1];
1093 		}
1094 #ifdef COMPAT_FREEBSD32
1095 		if (wrap32)
1096 			ptrace_sc_ret_to32(psr, psr32);
1097 #endif
1098 		CTR4(KTR_PTRACE,
1099 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1100 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1101 		    psr->sr_retval[1]);
1102 		break;
1103 
1104 	case PT_STEP:
1105 	case PT_CONTINUE:
1106 	case PT_TO_SCE:
1107 	case PT_TO_SCX:
1108 	case PT_SYSCALL:
1109 	case PT_DETACH:
1110 		/* Zero means do not send any signal */
1111 		if (data < 0 || data > _SIG_MAXSIG) {
1112 			error = EINVAL;
1113 			break;
1114 		}
1115 
1116 		switch (req) {
1117 		case PT_STEP:
1118 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1119 			    td2->td_tid, p->p_pid, data);
1120 			error = ptrace_single_step(td2);
1121 			if (error)
1122 				goto out;
1123 			break;
1124 		case PT_CONTINUE:
1125 		case PT_TO_SCE:
1126 		case PT_TO_SCX:
1127 		case PT_SYSCALL:
1128 			if (addr != (void *)1) {
1129 				error = ptrace_set_pc(td2,
1130 				    (u_long)(uintfptr_t)addr);
1131 				if (error)
1132 					goto out;
1133 			}
1134 			switch (req) {
1135 			case PT_TO_SCE:
1136 				p->p_ptevents |= PTRACE_SCE;
1137 				CTR4(KTR_PTRACE,
1138 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1139 				    p->p_pid, p->p_ptevents,
1140 				    (u_long)(uintfptr_t)addr, data);
1141 				break;
1142 			case PT_TO_SCX:
1143 				p->p_ptevents |= PTRACE_SCX;
1144 				CTR4(KTR_PTRACE,
1145 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1146 				    p->p_pid, p->p_ptevents,
1147 				    (u_long)(uintfptr_t)addr, data);
1148 				break;
1149 			case PT_SYSCALL:
1150 				p->p_ptevents |= PTRACE_SYSCALL;
1151 				CTR4(KTR_PTRACE,
1152 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1153 				    p->p_pid, p->p_ptevents,
1154 				    (u_long)(uintfptr_t)addr, data);
1155 				break;
1156 			case PT_CONTINUE:
1157 				CTR3(KTR_PTRACE,
1158 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1159 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1160 				break;
1161 			}
1162 			break;
1163 		case PT_DETACH:
1164 			/*
1165 			 * Reset the process parent.
1166 			 *
1167 			 * NB: This clears P_TRACED before reparenting
1168 			 * a detached process back to its original
1169 			 * parent.  Otherwise the debugee will be set
1170 			 * as an orphan of the debugger.
1171 			 */
1172 			p->p_flag &= ~(P_TRACED | P_WAITED);
1173 			if (p->p_oppid != p->p_pptr->p_pid) {
1174 				PROC_LOCK(p->p_pptr);
1175 				sigqueue_take(p->p_ksi);
1176 				PROC_UNLOCK(p->p_pptr);
1177 
1178 				pp = proc_realparent(p);
1179 				proc_reparent(p, pp, false);
1180 				if (pp == initproc)
1181 					p->p_sigparent = SIGCHLD;
1182 				CTR3(KTR_PTRACE,
1183 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1184 				    p->p_pid, pp->p_pid, data);
1185 			} else
1186 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1187 				    p->p_pid, data);
1188 			p->p_ptevents = 0;
1189 			FOREACH_THREAD_IN_PROC(p, td3) {
1190 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1191 					sigqueue_delete(&td3->td_sigqueue,
1192 					    SIGSTOP);
1193 				}
1194 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1195 				    TDB_SUSPEND);
1196 			}
1197 
1198 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1199 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1200 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1201 			}
1202 
1203 			/* should we send SIGCHLD? */
1204 			/* childproc_continued(p); */
1205 			break;
1206 		}
1207 
1208 		sx_xunlock(&proctree_lock);
1209 		proctree_locked = 0;
1210 
1211 	sendsig:
1212 		MPASS(proctree_locked == 0);
1213 
1214 		/*
1215 		 * Clear the pending event for the thread that just
1216 		 * reported its event (p_xthread).  This may not be
1217 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1218 		 * the debugger is resuming a different thread.
1219 		 *
1220 		 * Deliver any pending signal via the reporting thread.
1221 		 */
1222 		MPASS(p->p_xthread != NULL);
1223 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1224 		p->p_xthread->td_xsig = data;
1225 		p->p_xthread = NULL;
1226 		p->p_xsig = data;
1227 
1228 		/*
1229 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1230 		 * always works immediately, even if another thread is
1231 		 * unsuspended first and attempts to handle a
1232 		 * different signal or if the POSIX.1b style signal
1233 		 * queue cannot accommodate any new signals.
1234 		 */
1235 		if (data == SIGKILL)
1236 			proc_wkilled(p);
1237 
1238 		/*
1239 		 * Unsuspend all threads.  To leave a thread
1240 		 * suspended, use PT_SUSPEND to suspend it before
1241 		 * continuing the process.
1242 		 */
1243 		PROC_SLOCK(p);
1244 		p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
1245 		thread_unsuspend(p);
1246 		PROC_SUNLOCK(p);
1247 		break;
1248 
1249 	case PT_WRITE_I:
1250 	case PT_WRITE_D:
1251 		td2->td_dbgflags |= TDB_USERWR;
1252 		PROC_UNLOCK(p);
1253 		error = 0;
1254 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1255 		    sizeof(int)) != sizeof(int))
1256 			error = ENOMEM;
1257 		else
1258 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1259 			    p->p_pid, addr, data);
1260 		PROC_LOCK(p);
1261 		break;
1262 
1263 	case PT_READ_I:
1264 	case PT_READ_D:
1265 		PROC_UNLOCK(p);
1266 		error = tmp = 0;
1267 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1268 		    sizeof(int)) != sizeof(int))
1269 			error = ENOMEM;
1270 		else
1271 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1272 			    p->p_pid, addr, tmp);
1273 		td->td_retval[0] = tmp;
1274 		PROC_LOCK(p);
1275 		break;
1276 
1277 	case PT_IO:
1278 #ifdef COMPAT_FREEBSD32
1279 		if (wrap32) {
1280 			piod32 = addr;
1281 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1282 			iov.iov_len = piod32->piod_len;
1283 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1284 			uio.uio_resid = piod32->piod_len;
1285 		} else
1286 #endif
1287 		{
1288 			piod = addr;
1289 			iov.iov_base = piod->piod_addr;
1290 			iov.iov_len = piod->piod_len;
1291 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1292 			uio.uio_resid = piod->piod_len;
1293 		}
1294 		uio.uio_iov = &iov;
1295 		uio.uio_iovcnt = 1;
1296 		uio.uio_segflg = UIO_USERSPACE;
1297 		uio.uio_td = td;
1298 #ifdef COMPAT_FREEBSD32
1299 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1300 #else
1301 		tmp = piod->piod_op;
1302 #endif
1303 		switch (tmp) {
1304 		case PIOD_READ_D:
1305 		case PIOD_READ_I:
1306 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1307 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1308 			uio.uio_rw = UIO_READ;
1309 			break;
1310 		case PIOD_WRITE_D:
1311 		case PIOD_WRITE_I:
1312 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1313 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1314 			td2->td_dbgflags |= TDB_USERWR;
1315 			uio.uio_rw = UIO_WRITE;
1316 			break;
1317 		default:
1318 			error = EINVAL;
1319 			goto out;
1320 		}
1321 		PROC_UNLOCK(p);
1322 		error = proc_rwmem(p, &uio);
1323 #ifdef COMPAT_FREEBSD32
1324 		if (wrap32)
1325 			piod32->piod_len -= uio.uio_resid;
1326 		else
1327 #endif
1328 			piod->piod_len -= uio.uio_resid;
1329 		PROC_LOCK(p);
1330 		break;
1331 
1332 	case PT_KILL:
1333 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1334 		data = SIGKILL;
1335 		goto sendsig;	/* in PT_CONTINUE above */
1336 
1337 	case PT_SETREGS:
1338 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1339 		    p->p_pid);
1340 		td2->td_dbgflags |= TDB_USERWR;
1341 		error = PROC_WRITE(regs, td2, addr);
1342 		break;
1343 
1344 	case PT_GETREGS:
1345 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1346 		    p->p_pid);
1347 		error = PROC_READ(regs, td2, addr);
1348 		break;
1349 
1350 	case PT_SETFPREGS:
1351 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1352 		    p->p_pid);
1353 		td2->td_dbgflags |= TDB_USERWR;
1354 		error = PROC_WRITE(fpregs, td2, addr);
1355 		break;
1356 
1357 	case PT_GETFPREGS:
1358 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1359 		    p->p_pid);
1360 		error = PROC_READ(fpregs, td2, addr);
1361 		break;
1362 
1363 	case PT_SETDBREGS:
1364 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1365 		    p->p_pid);
1366 		td2->td_dbgflags |= TDB_USERWR;
1367 		error = PROC_WRITE(dbregs, td2, addr);
1368 		break;
1369 
1370 	case PT_GETDBREGS:
1371 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1372 		    p->p_pid);
1373 		error = PROC_READ(dbregs, td2, addr);
1374 		break;
1375 
1376 	case PT_LWPINFO:
1377 		if (data <= 0 ||
1378 #ifdef COMPAT_FREEBSD32
1379 		    (!wrap32 && data > sizeof(*pl)) ||
1380 		    (wrap32 && data > sizeof(*pl32))) {
1381 #else
1382 		    data > sizeof(*pl)) {
1383 #endif
1384 			error = EINVAL;
1385 			break;
1386 		}
1387 #ifdef COMPAT_FREEBSD32
1388 		if (wrap32) {
1389 			pl = &r.pl;
1390 			pl32 = addr;
1391 		} else
1392 #endif
1393 		pl = addr;
1394 		bzero(pl, sizeof(*pl));
1395 		pl->pl_lwpid = td2->td_tid;
1396 		pl->pl_event = PL_EVENT_NONE;
1397 		pl->pl_flags = 0;
1398 		if (td2->td_dbgflags & TDB_XSIG) {
1399 			pl->pl_event = PL_EVENT_SIGNAL;
1400 			if (td2->td_si.si_signo != 0 &&
1401 #ifdef COMPAT_FREEBSD32
1402 			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1403 			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1404 			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1405 			    pl_siginfo) + sizeof(struct siginfo32)))
1406 #else
1407 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1408 			    + sizeof(pl->pl_siginfo)
1409 #endif
1410 			){
1411 				pl->pl_flags |= PL_FLAG_SI;
1412 				pl->pl_siginfo = td2->td_si;
1413 			}
1414 		}
1415 		if (td2->td_dbgflags & TDB_SCE)
1416 			pl->pl_flags |= PL_FLAG_SCE;
1417 		else if (td2->td_dbgflags & TDB_SCX)
1418 			pl->pl_flags |= PL_FLAG_SCX;
1419 		if (td2->td_dbgflags & TDB_EXEC)
1420 			pl->pl_flags |= PL_FLAG_EXEC;
1421 		if (td2->td_dbgflags & TDB_FORK) {
1422 			pl->pl_flags |= PL_FLAG_FORKED;
1423 			pl->pl_child_pid = td2->td_dbg_forked;
1424 			if (td2->td_dbgflags & TDB_VFORK)
1425 				pl->pl_flags |= PL_FLAG_VFORKED;
1426 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1427 		    TDB_VFORK)
1428 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1429 		if (td2->td_dbgflags & TDB_CHILD)
1430 			pl->pl_flags |= PL_FLAG_CHILD;
1431 		if (td2->td_dbgflags & TDB_BORN)
1432 			pl->pl_flags |= PL_FLAG_BORN;
1433 		if (td2->td_dbgflags & TDB_EXIT)
1434 			pl->pl_flags |= PL_FLAG_EXITED;
1435 		pl->pl_sigmask = td2->td_sigmask;
1436 		pl->pl_siglist = td2->td_siglist;
1437 		strcpy(pl->pl_tdname, td2->td_name);
1438 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1439 			pl->pl_syscall_code = td2->td_sa.code;
1440 			pl->pl_syscall_narg = td2->td_sa.narg;
1441 		} else {
1442 			pl->pl_syscall_code = 0;
1443 			pl->pl_syscall_narg = 0;
1444 		}
1445 #ifdef COMPAT_FREEBSD32
1446 		if (wrap32)
1447 			ptrace_lwpinfo_to32(pl, pl32);
1448 #endif
1449 		CTR6(KTR_PTRACE,
1450     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1451 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1452 		    pl->pl_child_pid, pl->pl_syscall_code);
1453 		break;
1454 
1455 	case PT_GETNUMLWPS:
1456 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1457 		    p->p_numthreads);
1458 		td->td_retval[0] = p->p_numthreads;
1459 		break;
1460 
1461 	case PT_GETLWPLIST:
1462 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1463 		    p->p_pid, data, p->p_numthreads);
1464 		if (data <= 0) {
1465 			error = EINVAL;
1466 			break;
1467 		}
1468 		num = imin(p->p_numthreads, data);
1469 		PROC_UNLOCK(p);
1470 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1471 		tmp = 0;
1472 		PROC_LOCK(p);
1473 		FOREACH_THREAD_IN_PROC(p, td2) {
1474 			if (tmp >= num)
1475 				break;
1476 			buf[tmp++] = td2->td_tid;
1477 		}
1478 		PROC_UNLOCK(p);
1479 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1480 		free(buf, M_TEMP);
1481 		if (!error)
1482 			td->td_retval[0] = tmp;
1483 		PROC_LOCK(p);
1484 		break;
1485 
1486 	case PT_VM_TIMESTAMP:
1487 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1488 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1489 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1490 		break;
1491 
1492 	case PT_VM_ENTRY:
1493 		PROC_UNLOCK(p);
1494 #ifdef COMPAT_FREEBSD32
1495 		if (wrap32)
1496 			error = ptrace_vm_entry32(td, p, addr);
1497 		else
1498 #endif
1499 		error = ptrace_vm_entry(td, p, addr);
1500 		PROC_LOCK(p);
1501 		break;
1502 
1503 	default:
1504 #ifdef __HAVE_PTRACE_MACHDEP
1505 		if (req >= PT_FIRSTMACH) {
1506 			PROC_UNLOCK(p);
1507 			error = cpu_ptrace(td2, req, addr, data);
1508 			PROC_LOCK(p);
1509 		} else
1510 #endif
1511 			/* Unknown request. */
1512 			error = EINVAL;
1513 		break;
1514 	}
1515 
1516 out:
1517 	/* Drop our hold on this process now that the request has completed. */
1518 	_PRELE(p);
1519 fail:
1520 	PROC_UNLOCK(p);
1521 	if (proctree_locked)
1522 		sx_xunlock(&proctree_lock);
1523 	return (error);
1524 }
1525 #undef PROC_READ
1526 #undef PROC_WRITE
1527 
1528 /*
1529  * Stop a process because of a debugging event;
1530  * stay stopped until p->p_step is cleared
1531  * (cleared by PIOCCONT in procfs).
1532  */
1533 void
1534 stopevent(struct proc *p, unsigned int event, unsigned int val)
1535 {
1536 
1537 	PROC_LOCK_ASSERT(p, MA_OWNED);
1538 	p->p_step = 1;
1539 	CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1540 	    val);
1541 	do {
1542 		if (event != S_EXIT)
1543 			p->p_xsig = val;
1544 		p->p_xthread = NULL;
1545 		p->p_stype = event;	/* Which event caused the stop? */
1546 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1547 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1548 	} while (p->p_step);
1549 }
1550