xref: /freebsd/sys/kern/sys_process.c (revision 069ac184)
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/param.h>
35 #include <sys/systm.h>
36 #include <sys/ktr.h>
37 #include <sys/limits.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/reg.h>
41 #include <sys/syscallsubr.h>
42 #include <sys/sysent.h>
43 #include <sys/sysproto.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/vnode.h>
47 #include <sys/ptrace.h>
48 #include <sys/rwlock.h>
49 #include <sys/sx.h>
50 #include <sys/malloc.h>
51 #include <sys/signalvar.h>
52 #include <sys/caprights.h>
53 #include <sys/filedesc.h>
54 
55 #include <security/audit/audit.h>
56 
57 #include <vm/vm.h>
58 #include <vm/pmap.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_map.h>
61 #include <vm/vm_kern.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_page.h>
64 #include <vm/vm_param.h>
65 
66 #ifdef COMPAT_FREEBSD32
67 #include <sys/procfs.h>
68 #endif
69 
70 /* Assert it's safe to unlock a process, e.g. to allocate working memory */
71 #define	PROC_ASSERT_TRACEREQ(p)	MPASS(((p)->p_flag2 & P2_PTRACEREQ) != 0)
72 
73 /*
74  * Functions implemented using PROC_ACTION():
75  *
76  * proc_read_regs(proc, regs)
77  *	Get the current user-visible register set from the process
78  *	and copy it into the regs structure (<machine/reg.h>).
79  *	The process is stopped at the time read_regs is called.
80  *
81  * proc_write_regs(proc, regs)
82  *	Update the current register set from the passed in regs
83  *	structure.  Take care to avoid clobbering special CPU
84  *	registers or privileged bits in the PSL.
85  *	Depending on the architecture this may have fix-up work to do,
86  *	especially if the IAR or PCW are modified.
87  *	The process is stopped at the time write_regs is called.
88  *
89  * proc_read_fpregs, proc_write_fpregs
90  *	deal with the floating point register set, otherwise as above.
91  *
92  * proc_read_dbregs, proc_write_dbregs
93  *	deal with the processor debug register set, otherwise as above.
94  *
95  * proc_sstep(proc)
96  *	Arrange for the process to trap after executing a single instruction.
97  */
98 
99 #define	PROC_ACTION(action) do {					\
100 	int error;							\
101 									\
102 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
103 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
104 		error = EIO;						\
105 	else								\
106 		error = (action);					\
107 	return (error);							\
108 } while (0)
109 
110 int
111 proc_read_regs(struct thread *td, struct reg *regs)
112 {
113 
114 	PROC_ACTION(fill_regs(td, regs));
115 }
116 
117 int
118 proc_write_regs(struct thread *td, struct reg *regs)
119 {
120 
121 	PROC_ACTION(set_regs(td, regs));
122 }
123 
124 int
125 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
126 {
127 
128 	PROC_ACTION(fill_dbregs(td, dbregs));
129 }
130 
131 int
132 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
133 {
134 
135 	PROC_ACTION(set_dbregs(td, dbregs));
136 }
137 
138 /*
139  * Ptrace doesn't support fpregs at all, and there are no security holes
140  * or translations for fpregs, so we can just copy them.
141  */
142 int
143 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
144 {
145 
146 	PROC_ACTION(fill_fpregs(td, fpregs));
147 }
148 
149 int
150 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
151 {
152 
153 	PROC_ACTION(set_fpregs(td, fpregs));
154 }
155 
156 static struct regset *
157 proc_find_regset(struct thread *td, int note)
158 {
159 	struct regset **regsetp, **regset_end, *regset;
160 	struct sysentvec *sv;
161 
162 	sv = td->td_proc->p_sysent;
163 	regsetp = sv->sv_regset_begin;
164 	if (regsetp == NULL)
165 		return (NULL);
166 	regset_end = sv->sv_regset_end;
167 	MPASS(regset_end != NULL);
168 	for (; regsetp < regset_end; regsetp++) {
169 		regset = *regsetp;
170 		if (regset->note != note)
171 			continue;
172 
173 		return (regset);
174 	}
175 
176 	return (NULL);
177 }
178 
179 static int
180 proc_read_regset(struct thread *td, int note, struct iovec *iov)
181 {
182 	struct regset *regset;
183 	struct proc *p;
184 	void *buf;
185 	size_t size;
186 	int error;
187 
188 	regset = proc_find_regset(td, note);
189 	if (regset == NULL)
190 		return (EINVAL);
191 
192 	if (regset->get == NULL)
193 		return (EINVAL);
194 
195 	size = regset->size;
196 	/*
197 	 * The regset is dynamically sized, e.g. the size could change
198 	 * depending on the hardware, or may have a per-thread size.
199 	 */
200 	if (size == 0) {
201 		if (!regset->get(regset, td, NULL, &size))
202 			return (EINVAL);
203 	}
204 
205 	if (iov->iov_base == NULL) {
206 		iov->iov_len = size;
207 		if (iov->iov_len == 0)
208 			return (EINVAL);
209 
210 		return (0);
211 	}
212 
213 	/* The length is wrong, return an error */
214 	if (iov->iov_len != size)
215 		return (EINVAL);
216 
217 	error = 0;
218 	p = td->td_proc;
219 
220 	/* Drop the proc lock while allocating the temp buffer */
221 	PROC_ASSERT_TRACEREQ(p);
222 	PROC_UNLOCK(p);
223 	buf = malloc(size, M_TEMP, M_WAITOK);
224 	PROC_LOCK(p);
225 
226 	if (!regset->get(regset, td, buf, &size)) {
227 		error = EINVAL;
228 	} else {
229 		KASSERT(size == regset->size || regset->size == 0,
230 		    ("%s: Getter function changed the size", __func__));
231 
232 		iov->iov_len = size;
233 		PROC_UNLOCK(p);
234 		error = copyout(buf, iov->iov_base, size);
235 		PROC_LOCK(p);
236 	}
237 
238 	free(buf, M_TEMP);
239 
240 	return (error);
241 }
242 
243 static int
244 proc_write_regset(struct thread *td, int note, struct iovec *iov)
245 {
246 	struct regset *regset;
247 	struct proc *p;
248 	void *buf;
249 	size_t size;
250 	int error;
251 
252 	regset = proc_find_regset(td, note);
253 	if (regset == NULL)
254 		return (EINVAL);
255 
256 	size = regset->size;
257 	/*
258 	 * The regset is dynamically sized, e.g. the size could change
259 	 * depending on the hardware, or may have a per-thread size.
260 	 */
261 	if (size == 0) {
262 		if (!regset->get(regset, td, NULL, &size))
263 			return (EINVAL);
264 	}
265 
266 	/* The length is wrong, return an error */
267 	if (iov->iov_len != size)
268 		return (EINVAL);
269 
270 	if (regset->set == NULL)
271 		return (EINVAL);
272 
273 	p = td->td_proc;
274 
275 	/* Drop the proc lock while allocating the temp buffer */
276 	PROC_ASSERT_TRACEREQ(p);
277 	PROC_UNLOCK(p);
278 	buf = malloc(size, M_TEMP, M_WAITOK);
279 	error = copyin(iov->iov_base, buf, size);
280 	PROC_LOCK(p);
281 
282 	if (error == 0) {
283 		if (!regset->set(regset, td, buf, size)) {
284 			error = EINVAL;
285 		}
286 	}
287 
288 	free(buf, M_TEMP);
289 
290 	return (error);
291 }
292 
293 #ifdef COMPAT_FREEBSD32
294 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
295 int
296 proc_read_regs32(struct thread *td, struct reg32 *regs32)
297 {
298 
299 	PROC_ACTION(fill_regs32(td, regs32));
300 }
301 
302 int
303 proc_write_regs32(struct thread *td, struct reg32 *regs32)
304 {
305 
306 	PROC_ACTION(set_regs32(td, regs32));
307 }
308 
309 int
310 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
311 {
312 
313 	PROC_ACTION(fill_dbregs32(td, dbregs32));
314 }
315 
316 int
317 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
318 {
319 
320 	PROC_ACTION(set_dbregs32(td, dbregs32));
321 }
322 
323 int
324 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
325 {
326 
327 	PROC_ACTION(fill_fpregs32(td, fpregs32));
328 }
329 
330 int
331 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
332 {
333 
334 	PROC_ACTION(set_fpregs32(td, fpregs32));
335 }
336 #endif
337 
338 int
339 proc_sstep(struct thread *td)
340 {
341 
342 	PROC_ACTION(ptrace_single_step(td));
343 }
344 
345 int
346 proc_rwmem(struct proc *p, struct uio *uio)
347 {
348 	vm_map_t map;
349 	vm_offset_t pageno;		/* page number */
350 	vm_prot_t reqprot;
351 	int error, fault_flags, page_offset, writing;
352 
353 	/*
354 	 * Make sure that the process' vmspace remains live.
355 	 */
356 	if (p != curproc)
357 		PROC_ASSERT_HELD(p);
358 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
359 
360 	/*
361 	 * The map we want...
362 	 */
363 	map = &p->p_vmspace->vm_map;
364 
365 	/*
366 	 * If we are writing, then we request vm_fault() to create a private
367 	 * copy of each page.  Since these copies will not be writeable by the
368 	 * process, we must explicity request that they be dirtied.
369 	 */
370 	writing = uio->uio_rw == UIO_WRITE;
371 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
372 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
373 
374 	/*
375 	 * Only map in one page at a time.  We don't have to, but it
376 	 * makes things easier.  This way is trivial - right?
377 	 */
378 	do {
379 		vm_offset_t uva;
380 		u_int len;
381 		vm_page_t m;
382 
383 		uva = (vm_offset_t)uio->uio_offset;
384 
385 		/*
386 		 * Get the page number of this segment.
387 		 */
388 		pageno = trunc_page(uva);
389 		page_offset = uva - pageno;
390 
391 		/*
392 		 * How many bytes to copy
393 		 */
394 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
395 
396 		/*
397 		 * Fault and hold the page on behalf of the process.
398 		 */
399 		error = vm_fault(map, pageno, reqprot, fault_flags, &m);
400 		if (error != KERN_SUCCESS) {
401 			if (error == KERN_RESOURCE_SHORTAGE)
402 				error = ENOMEM;
403 			else
404 				error = EFAULT;
405 			break;
406 		}
407 
408 		/*
409 		 * Now do the i/o move.
410 		 */
411 		error = uiomove_fromphys(&m, page_offset, len, uio);
412 
413 		/* Make the I-cache coherent for breakpoints. */
414 		if (writing && error == 0) {
415 			vm_map_lock_read(map);
416 			if (vm_map_check_protection(map, pageno, pageno +
417 			    PAGE_SIZE, VM_PROT_EXECUTE))
418 				vm_sync_icache(map, uva, len);
419 			vm_map_unlock_read(map);
420 		}
421 
422 		/*
423 		 * Release the page.
424 		 */
425 		vm_page_unwire(m, PQ_ACTIVE);
426 
427 	} while (error == 0 && uio->uio_resid > 0);
428 
429 	return (error);
430 }
431 
432 static ssize_t
433 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
434     size_t len, enum uio_rw rw)
435 {
436 	struct iovec iov;
437 	struct uio uio;
438 	ssize_t slen;
439 
440 	MPASS(len < SSIZE_MAX);
441 	slen = (ssize_t)len;
442 
443 	iov.iov_base = (caddr_t)buf;
444 	iov.iov_len = len;
445 	uio.uio_iov = &iov;
446 	uio.uio_iovcnt = 1;
447 	uio.uio_offset = va;
448 	uio.uio_resid = slen;
449 	uio.uio_segflg = UIO_SYSSPACE;
450 	uio.uio_rw = rw;
451 	uio.uio_td = td;
452 	proc_rwmem(p, &uio);
453 	if (uio.uio_resid == slen)
454 		return (-1);
455 	return (slen - uio.uio_resid);
456 }
457 
458 ssize_t
459 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
460     size_t len)
461 {
462 
463 	return (proc_iop(td, p, va, buf, len, UIO_READ));
464 }
465 
466 ssize_t
467 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
468     size_t len)
469 {
470 
471 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
472 }
473 
474 static int
475 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
476 {
477 	struct vattr vattr;
478 	vm_map_t map;
479 	vm_map_entry_t entry;
480 	vm_object_t obj, tobj, lobj;
481 	struct vmspace *vm;
482 	struct vnode *vp;
483 	char *freepath, *fullpath;
484 	u_int pathlen;
485 	int error, index;
486 
487 	error = 0;
488 	obj = NULL;
489 
490 	vm = vmspace_acquire_ref(p);
491 	map = &vm->vm_map;
492 	vm_map_lock_read(map);
493 
494 	do {
495 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
496 		    ("Submap in map header"));
497 		index = 0;
498 		VM_MAP_ENTRY_FOREACH(entry, map) {
499 			if (index >= pve->pve_entry &&
500 			    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
501 				break;
502 			index++;
503 		}
504 		if (index < pve->pve_entry) {
505 			error = EINVAL;
506 			break;
507 		}
508 		if (entry == &map->header) {
509 			error = ENOENT;
510 			break;
511 		}
512 
513 		/* We got an entry. */
514 		pve->pve_entry = index + 1;
515 		pve->pve_timestamp = map->timestamp;
516 		pve->pve_start = entry->start;
517 		pve->pve_end = entry->end - 1;
518 		pve->pve_offset = entry->offset;
519 		pve->pve_prot = entry->protection;
520 
521 		/* Backing object's path needed? */
522 		if (pve->pve_pathlen == 0)
523 			break;
524 
525 		pathlen = pve->pve_pathlen;
526 		pve->pve_pathlen = 0;
527 
528 		obj = entry->object.vm_object;
529 		if (obj != NULL)
530 			VM_OBJECT_RLOCK(obj);
531 	} while (0);
532 
533 	vm_map_unlock_read(map);
534 
535 	pve->pve_fsid = VNOVAL;
536 	pve->pve_fileid = VNOVAL;
537 
538 	if (error == 0 && obj != NULL) {
539 		lobj = obj;
540 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
541 			if (tobj != obj)
542 				VM_OBJECT_RLOCK(tobj);
543 			if (lobj != obj)
544 				VM_OBJECT_RUNLOCK(lobj);
545 			lobj = tobj;
546 			pve->pve_offset += tobj->backing_object_offset;
547 		}
548 		vp = vm_object_vnode(lobj);
549 		if (vp != NULL)
550 			vref(vp);
551 		if (lobj != obj)
552 			VM_OBJECT_RUNLOCK(lobj);
553 		VM_OBJECT_RUNLOCK(obj);
554 
555 		if (vp != NULL) {
556 			freepath = NULL;
557 			fullpath = NULL;
558 			vn_fullpath(vp, &fullpath, &freepath);
559 			vn_lock(vp, LK_SHARED | LK_RETRY);
560 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
561 				pve->pve_fileid = vattr.va_fileid;
562 				pve->pve_fsid = vattr.va_fsid;
563 			}
564 			vput(vp);
565 
566 			if (fullpath != NULL) {
567 				pve->pve_pathlen = strlen(fullpath) + 1;
568 				if (pve->pve_pathlen <= pathlen) {
569 					error = copyout(fullpath, pve->pve_path,
570 					    pve->pve_pathlen);
571 				} else
572 					error = ENAMETOOLONG;
573 			}
574 			if (freepath != NULL)
575 				free(freepath, M_TEMP);
576 		}
577 	}
578 	vmspace_free(vm);
579 	if (error == 0)
580 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
581 		    p->p_pid, pve->pve_entry, pve->pve_start);
582 
583 	return (error);
584 }
585 
586 /*
587  * Process debugging system call.
588  */
589 #ifndef _SYS_SYSPROTO_H_
590 struct ptrace_args {
591 	int	req;
592 	pid_t	pid;
593 	caddr_t	addr;
594 	int	data;
595 };
596 #endif
597 
598 int
599 sys_ptrace(struct thread *td, struct ptrace_args *uap)
600 {
601 	/*
602 	 * XXX this obfuscation is to reduce stack usage, but the register
603 	 * structs may be too large to put on the stack anyway.
604 	 */
605 	union {
606 		struct ptrace_io_desc piod;
607 		struct ptrace_lwpinfo pl;
608 		struct ptrace_vm_entry pve;
609 		struct ptrace_coredump pc;
610 		struct ptrace_sc_remote sr;
611 		struct dbreg dbreg;
612 		struct fpreg fpreg;
613 		struct reg reg;
614 		struct iovec vec;
615 		syscallarg_t args[nitems(td->td_sa.args)];
616 		struct ptrace_sc_ret psr;
617 		int ptevents;
618 	} r;
619 	syscallarg_t pscr_args[nitems(td->td_sa.args)];
620 	void *addr;
621 	int error;
622 
623 	if (!allow_ptrace)
624 		return (ENOSYS);
625 	error = 0;
626 
627 	AUDIT_ARG_PID(uap->pid);
628 	AUDIT_ARG_CMD(uap->req);
629 	AUDIT_ARG_VALUE(uap->data);
630 	addr = &r;
631 	switch (uap->req) {
632 	case PT_GET_EVENT_MASK:
633 	case PT_LWPINFO:
634 	case PT_GET_SC_ARGS:
635 	case PT_GET_SC_RET:
636 		break;
637 	case PT_GETREGS:
638 		bzero(&r.reg, sizeof(r.reg));
639 		break;
640 	case PT_GETFPREGS:
641 		bzero(&r.fpreg, sizeof(r.fpreg));
642 		break;
643 	case PT_GETDBREGS:
644 		bzero(&r.dbreg, sizeof(r.dbreg));
645 		break;
646 	case PT_GETREGSET:
647 	case PT_SETREGSET:
648 		error = copyin(uap->addr, &r.vec, sizeof(r.vec));
649 		break;
650 	case PT_SETREGS:
651 		error = copyin(uap->addr, &r.reg, sizeof(r.reg));
652 		break;
653 	case PT_SETFPREGS:
654 		error = copyin(uap->addr, &r.fpreg, sizeof(r.fpreg));
655 		break;
656 	case PT_SETDBREGS:
657 		error = copyin(uap->addr, &r.dbreg, sizeof(r.dbreg));
658 		break;
659 	case PT_SET_EVENT_MASK:
660 		if (uap->data != sizeof(r.ptevents))
661 			error = EINVAL;
662 		else
663 			error = copyin(uap->addr, &r.ptevents, uap->data);
664 		break;
665 	case PT_IO:
666 		error = copyin(uap->addr, &r.piod, sizeof(r.piod));
667 		break;
668 	case PT_VM_ENTRY:
669 		error = copyin(uap->addr, &r.pve, sizeof(r.pve));
670 		break;
671 	case PT_COREDUMP:
672 		if (uap->data != sizeof(r.pc))
673 			error = EINVAL;
674 		else
675 			error = copyin(uap->addr, &r.pc, uap->data);
676 		break;
677 	case PT_SC_REMOTE:
678 		if (uap->data != sizeof(r.sr)) {
679 			error = EINVAL;
680 			break;
681 		}
682 		error = copyin(uap->addr, &r.sr, uap->data);
683 		if (error != 0)
684 			break;
685 		if (r.sr.pscr_nargs > nitems(td->td_sa.args)) {
686 			error = EINVAL;
687 			break;
688 		}
689 		error = copyin(r.sr.pscr_args, pscr_args,
690 		    sizeof(u_long) * r.sr.pscr_nargs);
691 		if (error != 0)
692 			break;
693 		r.sr.pscr_args = pscr_args;
694 		break;
695 	default:
696 		addr = uap->addr;
697 		break;
698 	}
699 	if (error)
700 		return (error);
701 
702 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
703 	if (error)
704 		return (error);
705 
706 	switch (uap->req) {
707 	case PT_VM_ENTRY:
708 		error = copyout(&r.pve, uap->addr, sizeof(r.pve));
709 		break;
710 	case PT_IO:
711 		error = copyout(&r.piod, uap->addr, sizeof(r.piod));
712 		break;
713 	case PT_GETREGS:
714 		error = copyout(&r.reg, uap->addr, sizeof(r.reg));
715 		break;
716 	case PT_GETFPREGS:
717 		error = copyout(&r.fpreg, uap->addr, sizeof(r.fpreg));
718 		break;
719 	case PT_GETDBREGS:
720 		error = copyout(&r.dbreg, uap->addr, sizeof(r.dbreg));
721 		break;
722 	case PT_GETREGSET:
723 		error = copyout(&r.vec, uap->addr, sizeof(r.vec));
724 		break;
725 	case PT_GET_EVENT_MASK:
726 		/* NB: The size in uap->data is validated in kern_ptrace(). */
727 		error = copyout(&r.ptevents, uap->addr, uap->data);
728 		break;
729 	case PT_LWPINFO:
730 		/* NB: The size in uap->data is validated in kern_ptrace(). */
731 		error = copyout(&r.pl, uap->addr, uap->data);
732 		break;
733 	case PT_GET_SC_ARGS:
734 		error = copyout(r.args, uap->addr, MIN(uap->data,
735 		    sizeof(r.args)));
736 		break;
737 	case PT_GET_SC_RET:
738 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
739 		    sizeof(r.psr)));
740 		break;
741 	case PT_SC_REMOTE:
742 		error = copyout(&r.sr.pscr_ret, uap->addr +
743 		    offsetof(struct ptrace_sc_remote, pscr_ret),
744 		    sizeof(r.sr.pscr_ret));
745 		break;
746 	}
747 
748 	return (error);
749 }
750 
751 #ifdef COMPAT_FREEBSD32
752 /*
753  *   PROC_READ(regs, td2, addr);
754  * becomes either:
755  *   proc_read_regs(td2, addr);
756  * or
757  *   proc_read_regs32(td2, addr);
758  * .. except this is done at runtime.  There is an additional
759  * complication in that PROC_WRITE disallows 32 bit consumers
760  * from writing to 64 bit address space targets.
761  */
762 #define	PROC_READ(w, t, a)	wrap32 ? \
763 	proc_read_ ## w ## 32(t, a) : \
764 	proc_read_ ## w (t, a)
765 #define	PROC_WRITE(w, t, a)	wrap32 ? \
766 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
767 	proc_write_ ## w (t, a)
768 #else
769 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
770 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
771 #endif
772 
773 void
774 proc_set_traced(struct proc *p, bool stop)
775 {
776 
777 	sx_assert(&proctree_lock, SX_XLOCKED);
778 	PROC_LOCK_ASSERT(p, MA_OWNED);
779 	p->p_flag |= P_TRACED;
780 	if (stop)
781 		p->p_flag2 |= P2_PTRACE_FSTP;
782 	p->p_ptevents = PTRACE_DEFAULT;
783 }
784 
785 void
786 ptrace_unsuspend(struct proc *p)
787 {
788 	PROC_LOCK_ASSERT(p, MA_OWNED);
789 
790 	PROC_SLOCK(p);
791 	p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
792 	thread_unsuspend(p);
793 	PROC_SUNLOCK(p);
794 	itimer_proc_continue(p);
795 	kqtimer_proc_continue(p);
796 }
797 
798 static int
799 proc_can_ptrace(struct thread *td, struct proc *p)
800 {
801 	int error;
802 
803 	PROC_LOCK_ASSERT(p, MA_OWNED);
804 
805 	if ((p->p_flag & P_WEXIT) != 0)
806 		return (ESRCH);
807 
808 	if ((error = p_cansee(td, p)) != 0)
809 		return (error);
810 	if ((error = p_candebug(td, p)) != 0)
811 		return (error);
812 
813 	/* not being traced... */
814 	if ((p->p_flag & P_TRACED) == 0)
815 		return (EPERM);
816 
817 	/* not being traced by YOU */
818 	if (p->p_pptr != td->td_proc)
819 		return (EBUSY);
820 
821 	/* not currently stopped */
822 	if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
823 	    p->p_suspcount != p->p_numthreads  ||
824 	    (p->p_flag & P_WAITED) == 0)
825 		return (EBUSY);
826 
827 	return (0);
828 }
829 
830 static struct thread *
831 ptrace_sel_coredump_thread(struct proc *p)
832 {
833 	struct thread *td2;
834 
835 	PROC_LOCK_ASSERT(p, MA_OWNED);
836 	MPASS((p->p_flag & P_STOPPED_TRACE) != 0);
837 
838 	FOREACH_THREAD_IN_PROC(p, td2) {
839 		if ((td2->td_dbgflags & TDB_SSWITCH) != 0)
840 			return (td2);
841 	}
842 	return (NULL);
843 }
844 
845 int
846 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
847 {
848 	struct iovec iov;
849 	struct uio uio;
850 	struct proc *curp, *p, *pp;
851 	struct thread *td2 = NULL, *td3;
852 	struct ptrace_io_desc *piod = NULL;
853 	struct ptrace_lwpinfo *pl;
854 	struct ptrace_sc_ret *psr;
855 	struct ptrace_sc_remote *pscr;
856 	struct file *fp;
857 	struct ptrace_coredump *pc;
858 	struct thr_coredump_req *tcq;
859 	struct thr_syscall_req *tsr;
860 	int error, num, tmp;
861 	lwpid_t tid = 0, *buf;
862 #ifdef COMPAT_FREEBSD32
863 	int wrap32 = 0, safe = 0;
864 #endif
865 	bool proctree_locked, p2_req_set;
866 
867 	curp = td->td_proc;
868 	proctree_locked = false;
869 	p2_req_set = false;
870 
871 	/* Lock proctree before locking the process. */
872 	switch (req) {
873 	case PT_TRACE_ME:
874 	case PT_ATTACH:
875 	case PT_STEP:
876 	case PT_CONTINUE:
877 	case PT_TO_SCE:
878 	case PT_TO_SCX:
879 	case PT_SYSCALL:
880 	case PT_FOLLOW_FORK:
881 	case PT_LWP_EVENTS:
882 	case PT_GET_EVENT_MASK:
883 	case PT_SET_EVENT_MASK:
884 	case PT_DETACH:
885 	case PT_GET_SC_ARGS:
886 		sx_xlock(&proctree_lock);
887 		proctree_locked = true;
888 		break;
889 	default:
890 		break;
891 	}
892 
893 	if (req == PT_TRACE_ME) {
894 		p = td->td_proc;
895 		PROC_LOCK(p);
896 	} else {
897 		if (pid <= PID_MAX) {
898 			if ((p = pfind(pid)) == NULL) {
899 				if (proctree_locked)
900 					sx_xunlock(&proctree_lock);
901 				return (ESRCH);
902 			}
903 		} else {
904 			td2 = tdfind(pid, -1);
905 			if (td2 == NULL) {
906 				if (proctree_locked)
907 					sx_xunlock(&proctree_lock);
908 				return (ESRCH);
909 			}
910 			p = td2->td_proc;
911 			tid = pid;
912 			pid = p->p_pid;
913 		}
914 	}
915 	AUDIT_ARG_PROCESS(p);
916 
917 	if ((p->p_flag & P_WEXIT) != 0) {
918 		error = ESRCH;
919 		goto fail;
920 	}
921 	if ((error = p_cansee(td, p)) != 0)
922 		goto fail;
923 
924 	if ((error = p_candebug(td, p)) != 0)
925 		goto fail;
926 
927 	/*
928 	 * System processes can't be debugged.
929 	 */
930 	if ((p->p_flag & P_SYSTEM) != 0) {
931 		error = EINVAL;
932 		goto fail;
933 	}
934 
935 	if (tid == 0) {
936 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
937 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
938 			td2 = p->p_xthread;
939 		} else {
940 			td2 = FIRST_THREAD_IN_PROC(p);
941 		}
942 		tid = td2->td_tid;
943 	}
944 
945 #ifdef COMPAT_FREEBSD32
946 	/*
947 	 * Test if we're a 32 bit client and what the target is.
948 	 * Set the wrap controls accordingly.
949 	 */
950 	if (SV_CURPROC_FLAG(SV_ILP32)) {
951 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
952 			safe = 1;
953 		wrap32 = 1;
954 	}
955 #endif
956 	/*
957 	 * Permissions check
958 	 */
959 	switch (req) {
960 	case PT_TRACE_ME:
961 		/*
962 		 * Always legal, when there is a parent process which
963 		 * could trace us.  Otherwise, reject.
964 		 */
965 		if ((p->p_flag & P_TRACED) != 0) {
966 			error = EBUSY;
967 			goto fail;
968 		}
969 		if (p->p_pptr == initproc) {
970 			error = EPERM;
971 			goto fail;
972 		}
973 		break;
974 
975 	case PT_ATTACH:
976 		/* Self */
977 		if (p == td->td_proc) {
978 			error = EINVAL;
979 			goto fail;
980 		}
981 
982 		/* Already traced */
983 		if (p->p_flag & P_TRACED) {
984 			error = EBUSY;
985 			goto fail;
986 		}
987 
988 		/* Can't trace an ancestor if you're being traced. */
989 		if (curp->p_flag & P_TRACED) {
990 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
991 				if (pp == p) {
992 					error = EINVAL;
993 					goto fail;
994 				}
995 			}
996 		}
997 
998 		/* OK */
999 		break;
1000 
1001 	case PT_CLEARSTEP:
1002 		/* Allow thread to clear single step for itself */
1003 		if (td->td_tid == tid)
1004 			break;
1005 
1006 		/* FALLTHROUGH */
1007 	default:
1008 		/*
1009 		 * Check for ptrace eligibility before waiting for
1010 		 * holds to drain.
1011 		 */
1012 		error = proc_can_ptrace(td, p);
1013 		if (error != 0)
1014 			goto fail;
1015 
1016 		/*
1017 		 * Block parallel ptrace requests.  Most important, do
1018 		 * not allow other thread in debugger to continue the
1019 		 * debuggee until coredump finished.
1020 		 */
1021 		while ((p->p_flag2 & P2_PTRACEREQ) != 0) {
1022 			if (proctree_locked)
1023 				sx_xunlock(&proctree_lock);
1024 			error = msleep(&p->p_flag2, &p->p_mtx, PPAUSE | PCATCH |
1025 			    (proctree_locked ? PDROP : 0), "pptrace", 0);
1026 			if (proctree_locked) {
1027 				sx_xlock(&proctree_lock);
1028 				PROC_LOCK(p);
1029 			}
1030 			if (error == 0 && td2->td_proc != p)
1031 				error = ESRCH;
1032 			if (error == 0)
1033 				error = proc_can_ptrace(td, p);
1034 			if (error != 0)
1035 				goto fail;
1036 		}
1037 
1038 		/* Ok */
1039 		break;
1040 	}
1041 
1042 	/*
1043 	 * Keep this process around and request parallel ptrace()
1044 	 * request to wait until we finish this request.
1045 	 */
1046 	MPASS((p->p_flag2 & P2_PTRACEREQ) == 0);
1047 	p->p_flag2 |= P2_PTRACEREQ;
1048 	p2_req_set = true;
1049 	_PHOLD(p);
1050 
1051 	/*
1052 	 * Actually do the requests
1053 	 */
1054 
1055 	td->td_retval[0] = 0;
1056 
1057 	switch (req) {
1058 	case PT_TRACE_ME:
1059 		/* set my trace flag and "owner" so it can read/write me */
1060 		proc_set_traced(p, false);
1061 		if (p->p_flag & P_PPWAIT)
1062 			p->p_flag |= P_PPTRACE;
1063 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
1064 		break;
1065 
1066 	case PT_ATTACH:
1067 		/* security check done above */
1068 		/*
1069 		 * It would be nice if the tracing relationship was separate
1070 		 * from the parent relationship but that would require
1071 		 * another set of links in the proc struct or for "wait"
1072 		 * to scan the entire proc table.  To make life easier,
1073 		 * we just re-parent the process we're trying to trace.
1074 		 * The old parent is remembered so we can put things back
1075 		 * on a "detach".
1076 		 */
1077 		proc_set_traced(p, true);
1078 		proc_reparent(p, td->td_proc, false);
1079 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
1080 		    p->p_oppid);
1081 
1082 		sx_xunlock(&proctree_lock);
1083 		proctree_locked = false;
1084 		MPASS(p->p_xthread == NULL);
1085 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
1086 
1087 		/*
1088 		 * If already stopped due to a stop signal, clear the
1089 		 * existing stop before triggering a traced SIGSTOP.
1090 		 */
1091 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
1092 			PROC_SLOCK(p);
1093 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
1094 			thread_unsuspend(p);
1095 			PROC_SUNLOCK(p);
1096 		}
1097 
1098 		kern_psignal(p, SIGSTOP);
1099 		break;
1100 
1101 	case PT_CLEARSTEP:
1102 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
1103 		    p->p_pid);
1104 		error = ptrace_clear_single_step(td2);
1105 		break;
1106 
1107 	case PT_SETSTEP:
1108 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
1109 		    p->p_pid);
1110 		error = ptrace_single_step(td2);
1111 		break;
1112 
1113 	case PT_SUSPEND:
1114 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
1115 		    p->p_pid);
1116 		td2->td_dbgflags |= TDB_SUSPEND;
1117 		ast_sched(td2, TDA_SUSPEND);
1118 		break;
1119 
1120 	case PT_RESUME:
1121 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1122 		    p->p_pid);
1123 		td2->td_dbgflags &= ~TDB_SUSPEND;
1124 		break;
1125 
1126 	case PT_FOLLOW_FORK:
1127 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1128 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1129 		    data ? "enabled" : "disabled");
1130 		if (data)
1131 			p->p_ptevents |= PTRACE_FORK;
1132 		else
1133 			p->p_ptevents &= ~PTRACE_FORK;
1134 		break;
1135 
1136 	case PT_LWP_EVENTS:
1137 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1138 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1139 		    data ? "enabled" : "disabled");
1140 		if (data)
1141 			p->p_ptevents |= PTRACE_LWP;
1142 		else
1143 			p->p_ptevents &= ~PTRACE_LWP;
1144 		break;
1145 
1146 	case PT_GET_EVENT_MASK:
1147 		if (data != sizeof(p->p_ptevents)) {
1148 			error = EINVAL;
1149 			break;
1150 		}
1151 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1152 		    p->p_ptevents);
1153 		*(int *)addr = p->p_ptevents;
1154 		break;
1155 
1156 	case PT_SET_EVENT_MASK:
1157 		if (data != sizeof(p->p_ptevents)) {
1158 			error = EINVAL;
1159 			break;
1160 		}
1161 		tmp = *(int *)addr;
1162 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1163 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1164 			error = EINVAL;
1165 			break;
1166 		}
1167 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1168 		    p->p_pid, p->p_ptevents, tmp);
1169 		p->p_ptevents = tmp;
1170 		break;
1171 
1172 	case PT_GET_SC_ARGS:
1173 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1174 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1175 #ifdef COMPAT_FREEBSD32
1176 		    || (wrap32 && !safe)
1177 #endif
1178 		    ) {
1179 			error = EINVAL;
1180 			break;
1181 		}
1182 		bzero(addr, sizeof(td2->td_sa.args));
1183 		/* See the explanation in linux_ptrace_get_syscall_info(). */
1184 		bcopy(td2->td_sa.args, addr, SV_PROC_ABI(td->td_proc) ==
1185 		    SV_ABI_LINUX ? sizeof(td2->td_sa.args) :
1186 		    td2->td_sa.callp->sy_narg * sizeof(syscallarg_t));
1187 		break;
1188 
1189 	case PT_GET_SC_RET:
1190 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1191 #ifdef COMPAT_FREEBSD32
1192 		    || (wrap32 && !safe)
1193 #endif
1194 		    ) {
1195 			error = EINVAL;
1196 			break;
1197 		}
1198 		psr = addr;
1199 		bzero(psr, sizeof(*psr));
1200 		psr->sr_error = td2->td_errno;
1201 		if (psr->sr_error == 0) {
1202 			psr->sr_retval[0] = td2->td_retval[0];
1203 			psr->sr_retval[1] = td2->td_retval[1];
1204 		}
1205 		CTR4(KTR_PTRACE,
1206 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1207 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1208 		    psr->sr_retval[1]);
1209 		break;
1210 
1211 	case PT_STEP:
1212 	case PT_CONTINUE:
1213 	case PT_TO_SCE:
1214 	case PT_TO_SCX:
1215 	case PT_SYSCALL:
1216 	case PT_DETACH:
1217 		/* Zero means do not send any signal */
1218 		if (data < 0 || data > _SIG_MAXSIG) {
1219 			error = EINVAL;
1220 			break;
1221 		}
1222 
1223 		switch (req) {
1224 		case PT_STEP:
1225 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1226 			    td2->td_tid, p->p_pid, data);
1227 			error = ptrace_single_step(td2);
1228 			if (error)
1229 				goto out;
1230 			break;
1231 		case PT_CONTINUE:
1232 		case PT_TO_SCE:
1233 		case PT_TO_SCX:
1234 		case PT_SYSCALL:
1235 			if (addr != (void *)1) {
1236 				error = ptrace_set_pc(td2,
1237 				    (u_long)(uintfptr_t)addr);
1238 				if (error)
1239 					goto out;
1240 			}
1241 			switch (req) {
1242 			case PT_TO_SCE:
1243 				p->p_ptevents |= PTRACE_SCE;
1244 				CTR4(KTR_PTRACE,
1245 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1246 				    p->p_pid, p->p_ptevents,
1247 				    (u_long)(uintfptr_t)addr, data);
1248 				break;
1249 			case PT_TO_SCX:
1250 				p->p_ptevents |= PTRACE_SCX;
1251 				CTR4(KTR_PTRACE,
1252 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1253 				    p->p_pid, p->p_ptevents,
1254 				    (u_long)(uintfptr_t)addr, data);
1255 				break;
1256 			case PT_SYSCALL:
1257 				p->p_ptevents |= PTRACE_SYSCALL;
1258 				CTR4(KTR_PTRACE,
1259 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1260 				    p->p_pid, p->p_ptevents,
1261 				    (u_long)(uintfptr_t)addr, data);
1262 				break;
1263 			case PT_CONTINUE:
1264 				CTR3(KTR_PTRACE,
1265 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1266 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1267 				break;
1268 			}
1269 			break;
1270 		case PT_DETACH:
1271 			/*
1272 			 * Clear P_TRACED before reparenting
1273 			 * a detached process back to its original
1274 			 * parent.  Otherwise the debugee will be set
1275 			 * as an orphan of the debugger.
1276 			 */
1277 			p->p_flag &= ~(P_TRACED | P_WAITED);
1278 
1279 			/*
1280 			 * Reset the process parent.
1281 			 */
1282 			if (p->p_oppid != p->p_pptr->p_pid) {
1283 				PROC_LOCK(p->p_pptr);
1284 				sigqueue_take(p->p_ksi);
1285 				PROC_UNLOCK(p->p_pptr);
1286 
1287 				pp = proc_realparent(p);
1288 				proc_reparent(p, pp, false);
1289 				if (pp == initproc)
1290 					p->p_sigparent = SIGCHLD;
1291 				CTR3(KTR_PTRACE,
1292 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1293 				    p->p_pid, pp->p_pid, data);
1294 			} else {
1295 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1296 				    p->p_pid, data);
1297 			}
1298 
1299 			p->p_ptevents = 0;
1300 			FOREACH_THREAD_IN_PROC(p, td3) {
1301 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1302 					sigqueue_delete(&td3->td_sigqueue,
1303 					    SIGSTOP);
1304 				}
1305 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1306 				    TDB_SUSPEND | TDB_BORN);
1307 			}
1308 
1309 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1310 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1311 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1312 			}
1313 
1314 			/* should we send SIGCHLD? */
1315 			/* childproc_continued(p); */
1316 			break;
1317 		}
1318 
1319 		sx_xunlock(&proctree_lock);
1320 		proctree_locked = false;
1321 
1322 	sendsig:
1323 		MPASS(!proctree_locked);
1324 
1325 		/*
1326 		 * Clear the pending event for the thread that just
1327 		 * reported its event (p_xthread).  This may not be
1328 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1329 		 * the debugger is resuming a different thread.
1330 		 *
1331 		 * Deliver any pending signal via the reporting thread.
1332 		 */
1333 		MPASS(p->p_xthread != NULL);
1334 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1335 		p->p_xthread->td_xsig = data;
1336 		p->p_xthread = NULL;
1337 		p->p_xsig = data;
1338 
1339 		/*
1340 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1341 		 * always works immediately, even if another thread is
1342 		 * unsuspended first and attempts to handle a
1343 		 * different signal or if the POSIX.1b style signal
1344 		 * queue cannot accommodate any new signals.
1345 		 */
1346 		if (data == SIGKILL)
1347 			proc_wkilled(p);
1348 
1349 		/*
1350 		 * Unsuspend all threads.  To leave a thread
1351 		 * suspended, use PT_SUSPEND to suspend it before
1352 		 * continuing the process.
1353 		 */
1354 		ptrace_unsuspend(p);
1355 		break;
1356 
1357 	case PT_WRITE_I:
1358 	case PT_WRITE_D:
1359 		td2->td_dbgflags |= TDB_USERWR;
1360 		PROC_UNLOCK(p);
1361 		error = 0;
1362 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1363 		    sizeof(int)) != sizeof(int))
1364 			error = ENOMEM;
1365 		else
1366 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1367 			    p->p_pid, addr, data);
1368 		PROC_LOCK(p);
1369 		break;
1370 
1371 	case PT_READ_I:
1372 	case PT_READ_D:
1373 		PROC_UNLOCK(p);
1374 		error = tmp = 0;
1375 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1376 		    sizeof(int)) != sizeof(int))
1377 			error = ENOMEM;
1378 		else
1379 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1380 			    p->p_pid, addr, tmp);
1381 		td->td_retval[0] = tmp;
1382 		PROC_LOCK(p);
1383 		break;
1384 
1385 	case PT_IO:
1386 		piod = addr;
1387 		iov.iov_base = piod->piod_addr;
1388 		iov.iov_len = piod->piod_len;
1389 		uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1390 		uio.uio_resid = piod->piod_len;
1391 		uio.uio_iov = &iov;
1392 		uio.uio_iovcnt = 1;
1393 		uio.uio_segflg = UIO_USERSPACE;
1394 		uio.uio_td = td;
1395 		switch (piod->piod_op) {
1396 		case PIOD_READ_D:
1397 		case PIOD_READ_I:
1398 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1399 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1400 			uio.uio_rw = UIO_READ;
1401 			break;
1402 		case PIOD_WRITE_D:
1403 		case PIOD_WRITE_I:
1404 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1405 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1406 			td2->td_dbgflags |= TDB_USERWR;
1407 			uio.uio_rw = UIO_WRITE;
1408 			break;
1409 		default:
1410 			error = EINVAL;
1411 			goto out;
1412 		}
1413 		PROC_UNLOCK(p);
1414 		error = proc_rwmem(p, &uio);
1415 		piod->piod_len -= uio.uio_resid;
1416 		PROC_LOCK(p);
1417 		break;
1418 
1419 	case PT_KILL:
1420 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1421 		data = SIGKILL;
1422 		goto sendsig;	/* in PT_CONTINUE above */
1423 
1424 	case PT_SETREGS:
1425 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1426 		    p->p_pid);
1427 		td2->td_dbgflags |= TDB_USERWR;
1428 		error = PROC_WRITE(regs, td2, addr);
1429 		break;
1430 
1431 	case PT_GETREGS:
1432 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1433 		    p->p_pid);
1434 		error = PROC_READ(regs, td2, addr);
1435 		break;
1436 
1437 	case PT_SETFPREGS:
1438 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1439 		    p->p_pid);
1440 		td2->td_dbgflags |= TDB_USERWR;
1441 		error = PROC_WRITE(fpregs, td2, addr);
1442 		break;
1443 
1444 	case PT_GETFPREGS:
1445 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1446 		    p->p_pid);
1447 		error = PROC_READ(fpregs, td2, addr);
1448 		break;
1449 
1450 	case PT_SETDBREGS:
1451 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1452 		    p->p_pid);
1453 		td2->td_dbgflags |= TDB_USERWR;
1454 		error = PROC_WRITE(dbregs, td2, addr);
1455 		break;
1456 
1457 	case PT_GETDBREGS:
1458 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1459 		    p->p_pid);
1460 		error = PROC_READ(dbregs, td2, addr);
1461 		break;
1462 
1463 	case PT_SETREGSET:
1464 		CTR2(KTR_PTRACE, "PT_SETREGSET: tid %d (pid %d)", td2->td_tid,
1465 		    p->p_pid);
1466 		error = proc_write_regset(td2, data, addr);
1467 		break;
1468 
1469 	case PT_GETREGSET:
1470 		CTR2(KTR_PTRACE, "PT_GETREGSET: tid %d (pid %d)", td2->td_tid,
1471 		    p->p_pid);
1472 		error = proc_read_regset(td2, data, addr);
1473 		break;
1474 
1475 	case PT_LWPINFO:
1476 		if (data <= 0 || data > sizeof(*pl)) {
1477 			error = EINVAL;
1478 			break;
1479 		}
1480 		pl = addr;
1481 		bzero(pl, sizeof(*pl));
1482 		pl->pl_lwpid = td2->td_tid;
1483 		pl->pl_event = PL_EVENT_NONE;
1484 		pl->pl_flags = 0;
1485 		if (td2->td_dbgflags & TDB_XSIG) {
1486 			pl->pl_event = PL_EVENT_SIGNAL;
1487 			if (td2->td_si.si_signo != 0 &&
1488 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1489 			    + sizeof(pl->pl_siginfo)){
1490 				pl->pl_flags |= PL_FLAG_SI;
1491 				pl->pl_siginfo = td2->td_si;
1492 			}
1493 		}
1494 		if (td2->td_dbgflags & TDB_SCE)
1495 			pl->pl_flags |= PL_FLAG_SCE;
1496 		else if (td2->td_dbgflags & TDB_SCX)
1497 			pl->pl_flags |= PL_FLAG_SCX;
1498 		if (td2->td_dbgflags & TDB_EXEC)
1499 			pl->pl_flags |= PL_FLAG_EXEC;
1500 		if (td2->td_dbgflags & TDB_FORK) {
1501 			pl->pl_flags |= PL_FLAG_FORKED;
1502 			pl->pl_child_pid = td2->td_dbg_forked;
1503 			if (td2->td_dbgflags & TDB_VFORK)
1504 				pl->pl_flags |= PL_FLAG_VFORKED;
1505 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1506 		    TDB_VFORK)
1507 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1508 		if (td2->td_dbgflags & TDB_CHILD)
1509 			pl->pl_flags |= PL_FLAG_CHILD;
1510 		if (td2->td_dbgflags & TDB_BORN)
1511 			pl->pl_flags |= PL_FLAG_BORN;
1512 		if (td2->td_dbgflags & TDB_EXIT)
1513 			pl->pl_flags |= PL_FLAG_EXITED;
1514 		pl->pl_sigmask = td2->td_sigmask;
1515 		pl->pl_siglist = td2->td_siglist;
1516 		strcpy(pl->pl_tdname, td2->td_name);
1517 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1518 			pl->pl_syscall_code = td2->td_sa.code;
1519 			pl->pl_syscall_narg = td2->td_sa.callp->sy_narg;
1520 		} else {
1521 			pl->pl_syscall_code = 0;
1522 			pl->pl_syscall_narg = 0;
1523 		}
1524 		CTR6(KTR_PTRACE,
1525     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1526 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1527 		    pl->pl_child_pid, pl->pl_syscall_code);
1528 		break;
1529 
1530 	case PT_GETNUMLWPS:
1531 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1532 		    p->p_numthreads);
1533 		td->td_retval[0] = p->p_numthreads;
1534 		break;
1535 
1536 	case PT_GETLWPLIST:
1537 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1538 		    p->p_pid, data, p->p_numthreads);
1539 		if (data <= 0) {
1540 			error = EINVAL;
1541 			break;
1542 		}
1543 		num = imin(p->p_numthreads, data);
1544 		PROC_UNLOCK(p);
1545 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1546 		tmp = 0;
1547 		PROC_LOCK(p);
1548 		FOREACH_THREAD_IN_PROC(p, td2) {
1549 			if (tmp >= num)
1550 				break;
1551 			buf[tmp++] = td2->td_tid;
1552 		}
1553 		PROC_UNLOCK(p);
1554 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1555 		free(buf, M_TEMP);
1556 		if (!error)
1557 			td->td_retval[0] = tmp;
1558 		PROC_LOCK(p);
1559 		break;
1560 
1561 	case PT_VM_TIMESTAMP:
1562 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1563 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1564 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1565 		break;
1566 
1567 	case PT_VM_ENTRY:
1568 		PROC_UNLOCK(p);
1569 		error = ptrace_vm_entry(td, p, addr);
1570 		PROC_LOCK(p);
1571 		break;
1572 
1573 	case PT_COREDUMP:
1574 		pc = addr;
1575 		CTR2(KTR_PTRACE, "PT_COREDUMP: pid %d, fd %d",
1576 		    p->p_pid, pc->pc_fd);
1577 
1578 		if ((pc->pc_flags & ~(PC_COMPRESS | PC_ALL)) != 0) {
1579 			error = EINVAL;
1580 			break;
1581 		}
1582 		PROC_UNLOCK(p);
1583 
1584 		tcq = malloc(sizeof(*tcq), M_TEMP, M_WAITOK | M_ZERO);
1585 		fp = NULL;
1586 		error = fget_write(td, pc->pc_fd, &cap_write_rights, &fp);
1587 		if (error != 0)
1588 			goto coredump_cleanup_nofp;
1589 		if (fp->f_type != DTYPE_VNODE || fp->f_vnode->v_type != VREG) {
1590 			error = EPIPE;
1591 			goto coredump_cleanup;
1592 		}
1593 
1594 		PROC_LOCK(p);
1595 		error = proc_can_ptrace(td, p);
1596 		if (error != 0)
1597 			goto coredump_cleanup_locked;
1598 
1599 		td2 = ptrace_sel_coredump_thread(p);
1600 		if (td2 == NULL) {
1601 			error = EBUSY;
1602 			goto coredump_cleanup_locked;
1603 		}
1604 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1605 		    TDB_SCREMOTEREQ)) == 0,
1606 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1607 
1608 		tcq->tc_vp = fp->f_vnode;
1609 		tcq->tc_limit = pc->pc_limit == 0 ? OFF_MAX : pc->pc_limit;
1610 		tcq->tc_flags = SVC_PT_COREDUMP;
1611 		if ((pc->pc_flags & PC_COMPRESS) == 0)
1612 			tcq->tc_flags |= SVC_NOCOMPRESS;
1613 		if ((pc->pc_flags & PC_ALL) != 0)
1614 			tcq->tc_flags |= SVC_ALL;
1615 		td2->td_remotereq = tcq;
1616 		td2->td_dbgflags |= TDB_COREDUMPREQ;
1617 		thread_run_flash(td2);
1618 		while ((td2->td_dbgflags & TDB_COREDUMPREQ) != 0)
1619 			msleep(p, &p->p_mtx, PPAUSE, "crdmp", 0);
1620 		error = tcq->tc_error;
1621 coredump_cleanup_locked:
1622 		PROC_UNLOCK(p);
1623 coredump_cleanup:
1624 		fdrop(fp, td);
1625 coredump_cleanup_nofp:
1626 		free(tcq, M_TEMP);
1627 		PROC_LOCK(p);
1628 		break;
1629 
1630 	case PT_SC_REMOTE:
1631 		pscr = addr;
1632 		CTR2(KTR_PTRACE, "PT_SC_REMOTE: pid %d, syscall %d",
1633 		    p->p_pid, pscr->pscr_syscall);
1634 		if ((td2->td_dbgflags & TDB_BOUNDARY) == 0) {
1635 			error = EBUSY;
1636 			break;
1637 		}
1638 		PROC_UNLOCK(p);
1639 		MPASS(pscr->pscr_nargs <= nitems(td->td_sa.args));
1640 
1641 		tsr = malloc(sizeof(struct thr_syscall_req), M_TEMP,
1642 		    M_WAITOK | M_ZERO);
1643 
1644 		tsr->ts_sa.code = pscr->pscr_syscall;
1645 		tsr->ts_nargs = pscr->pscr_nargs;
1646 		memcpy(&tsr->ts_sa.args, pscr->pscr_args,
1647 		    sizeof(syscallarg_t) * tsr->ts_nargs);
1648 
1649 		PROC_LOCK(p);
1650 		error = proc_can_ptrace(td, p);
1651 		if (error != 0) {
1652 			free(tsr, M_TEMP);
1653 			break;
1654 		}
1655 		if (td2->td_proc != p) {
1656 			free(tsr, M_TEMP);
1657 			error = ESRCH;
1658 			break;
1659 		}
1660 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1661 		    TDB_SCREMOTEREQ)) == 0,
1662 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1663 
1664 		td2->td_remotereq = tsr;
1665 		td2->td_dbgflags |= TDB_SCREMOTEREQ;
1666 		thread_run_flash(td2);
1667 		while ((td2->td_dbgflags & TDB_SCREMOTEREQ) != 0)
1668 			msleep(p, &p->p_mtx, PPAUSE, "pscrx", 0);
1669 		error = 0;
1670 		memcpy(&pscr->pscr_ret, &tsr->ts_ret, sizeof(tsr->ts_ret));
1671 		free(tsr, M_TEMP);
1672 		break;
1673 
1674 	default:
1675 #ifdef __HAVE_PTRACE_MACHDEP
1676 		if (req >= PT_FIRSTMACH) {
1677 			PROC_UNLOCK(p);
1678 			error = cpu_ptrace(td2, req, addr, data);
1679 			PROC_LOCK(p);
1680 		} else
1681 #endif
1682 			/* Unknown request. */
1683 			error = EINVAL;
1684 		break;
1685 	}
1686 out:
1687 	/* Drop our hold on this process now that the request has completed. */
1688 	_PRELE(p);
1689 fail:
1690 	if (p2_req_set) {
1691 		if ((p->p_flag2 & P2_PTRACEREQ) != 0)
1692 			wakeup(&p->p_flag2);
1693 		p->p_flag2 &= ~P2_PTRACEREQ;
1694 	}
1695 	PROC_UNLOCK(p);
1696 	if (proctree_locked)
1697 		sx_xunlock(&proctree_lock);
1698 	return (error);
1699 }
1700 #undef PROC_READ
1701 #undef PROC_WRITE
1702