xref: /freebsd/sys/kern/sys_process.c (revision 72ae04c7334743c0fec2e8c99ac13fd8ca63f51b)
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/mman.h>
40 #include <sys/mutex.h>
41 #include <sys/reg.h>
42 #include <sys/syscallsubr.h>
43 #include <sys/sysent.h>
44 #include <sys/sysproto.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/vnode.h>
48 #include <sys/ptrace.h>
49 #include <sys/rwlock.h>
50 #include <sys/sx.h>
51 #include <sys/malloc.h>
52 #include <sys/signalvar.h>
53 #include <sys/caprights.h>
54 #include <sys/filedesc.h>
55 
56 #include <security/audit/audit.h>
57 
58 #include <vm/vm.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_extern.h>
61 #include <vm/vm_map.h>
62 #include <vm/vm_kern.h>
63 #include <vm/vm_object.h>
64 #include <vm/vm_page.h>
65 #include <vm/vm_param.h>
66 
67 #ifdef COMPAT_FREEBSD32
68 #include <sys/procfs.h>
69 #endif
70 
71 /* Assert it's safe to unlock a process, e.g. to allocate working memory */
72 #define	PROC_ASSERT_TRACEREQ(p)	MPASS(((p)->p_flag2 & P2_PTRACEREQ) != 0)
73 
74 /*
75  * Functions implemented below:
76  *
77  * proc_read_regs(proc, regs)
78  *	Get the current user-visible register set from the process
79  *	and copy it into the regs structure (<machine/reg.h>).
80  *	The process is stopped at the time read_regs is called.
81  *
82  * proc_write_regs(proc, regs)
83  *	Update the current register set from the passed in regs
84  *	structure.  Take care to avoid clobbering special CPU
85  *	registers or privileged bits in the PSL.
86  *	Depending on the architecture this may have fix-up work to do,
87  *	especially if the IAR or PCW are modified.
88  *	The process is stopped at the time write_regs is called.
89  *
90  * proc_read_fpregs, proc_write_fpregs
91  *	deal with the floating point register set, otherwise as above.
92  *
93  * proc_read_dbregs, proc_write_dbregs
94  *	deal with the processor debug register set, otherwise as above.
95  *
96  * proc_sstep(proc)
97  *	Arrange for the process to trap after executing a single instruction.
98  */
99 
100 int
101 proc_read_regs(struct thread *td, struct reg *regs)
102 {
103 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
104 	return (fill_regs(td, regs));
105 }
106 
107 int
108 proc_write_regs(struct thread *td, struct reg *regs)
109 {
110 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
111 	return (set_regs(td, regs));
112 }
113 
114 int
115 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
116 {
117 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
118 	return (fill_dbregs(td, dbregs));
119 }
120 
121 int
122 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
123 {
124 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
125 	return (set_dbregs(td, dbregs));
126 }
127 
128 /*
129  * Ptrace doesn't support fpregs at all, and there are no security holes
130  * or translations for fpregs, so we can just copy them.
131  */
132 int
133 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
134 {
135 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
136 	return (fill_fpregs(td, fpregs));
137 }
138 
139 int
140 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
141 {
142 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
143 	return (set_fpregs(td, fpregs));
144 }
145 
146 static struct regset *
147 proc_find_regset(struct thread *td, int note)
148 {
149 	struct regset **regsetp, **regset_end, *regset;
150 	struct sysentvec *sv;
151 
152 	sv = td->td_proc->p_sysent;
153 	regsetp = sv->sv_regset_begin;
154 	if (regsetp == NULL)
155 		return (NULL);
156 	regset_end = sv->sv_regset_end;
157 	MPASS(regset_end != NULL);
158 	for (; regsetp < regset_end; regsetp++) {
159 		regset = *regsetp;
160 		if (regset->note != note)
161 			continue;
162 
163 		return (regset);
164 	}
165 
166 	return (NULL);
167 }
168 
169 static int
170 proc_read_regset(struct thread *td, int note, struct iovec *iov)
171 {
172 	struct regset *regset;
173 	struct proc *p;
174 	void *buf;
175 	size_t size;
176 	int error;
177 
178 	regset = proc_find_regset(td, note);
179 	if (regset == NULL)
180 		return (EINVAL);
181 
182 	if (regset->get == NULL)
183 		return (EINVAL);
184 
185 	size = regset->size;
186 	/*
187 	 * The regset is dynamically sized, e.g. the size could change
188 	 * depending on the hardware, or may have a per-thread size.
189 	 */
190 	if (size == 0) {
191 		if (!regset->get(regset, td, NULL, &size))
192 			return (EINVAL);
193 	}
194 
195 	if (iov->iov_base == NULL) {
196 		iov->iov_len = size;
197 		if (iov->iov_len == 0)
198 			return (EINVAL);
199 
200 		return (0);
201 	}
202 
203 	/* The length is wrong, return an error */
204 	if (iov->iov_len != size)
205 		return (EINVAL);
206 
207 	error = 0;
208 	p = td->td_proc;
209 
210 	/* Drop the proc lock while allocating the temp buffer */
211 	PROC_ASSERT_TRACEREQ(p);
212 	PROC_UNLOCK(p);
213 	buf = malloc(size, M_TEMP, M_WAITOK);
214 	PROC_LOCK(p);
215 
216 	if (!regset->get(regset, td, buf, &size)) {
217 		error = EINVAL;
218 	} else {
219 		KASSERT(size == regset->size || regset->size == 0,
220 		    ("%s: Getter function changed the size", __func__));
221 
222 		iov->iov_len = size;
223 		PROC_UNLOCK(p);
224 		error = copyout(buf, iov->iov_base, size);
225 		PROC_LOCK(p);
226 	}
227 
228 	free(buf, M_TEMP);
229 
230 	return (error);
231 }
232 
233 static int
234 proc_write_regset(struct thread *td, int note, struct iovec *iov)
235 {
236 	struct regset *regset;
237 	struct proc *p;
238 	void *buf;
239 	size_t size;
240 	int error;
241 
242 	regset = proc_find_regset(td, note);
243 	if (regset == NULL)
244 		return (EINVAL);
245 
246 	size = regset->size;
247 	/*
248 	 * The regset is dynamically sized, e.g. the size could change
249 	 * depending on the hardware, or may have a per-thread size.
250 	 */
251 	if (size == 0) {
252 		if (!regset->get(regset, td, NULL, &size))
253 			return (EINVAL);
254 	}
255 
256 	/* The length is wrong, return an error */
257 	if (iov->iov_len != size)
258 		return (EINVAL);
259 
260 	if (regset->set == NULL)
261 		return (EINVAL);
262 
263 	p = td->td_proc;
264 
265 	/* Drop the proc lock while allocating the temp buffer */
266 	PROC_ASSERT_TRACEREQ(p);
267 	PROC_UNLOCK(p);
268 	buf = malloc(size, M_TEMP, M_WAITOK);
269 	error = copyin(iov->iov_base, buf, size);
270 	PROC_LOCK(p);
271 
272 	if (error == 0) {
273 		if (!regset->set(regset, td, buf, size)) {
274 			error = EINVAL;
275 		}
276 	}
277 
278 	free(buf, M_TEMP);
279 
280 	return (error);
281 }
282 
283 #ifdef COMPAT_FREEBSD32
284 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
285 int
286 proc_read_regs32(struct thread *td, struct reg32 *regs32)
287 {
288 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
289 	return (fill_regs32(td, regs32));
290 }
291 
292 int
293 proc_write_regs32(struct thread *td, struct reg32 *regs32)
294 {
295 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
296 	return (set_regs32(td, regs32));
297 }
298 
299 int
300 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
301 {
302 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
303 	return (fill_dbregs32(td, dbregs32));
304 }
305 
306 int
307 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
308 {
309 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
310 	return (set_dbregs32(td, dbregs32));
311 }
312 
313 int
314 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
315 {
316 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
317 	return (fill_fpregs32(td, fpregs32));
318 }
319 
320 int
321 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
322 {
323 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
324 	return (set_fpregs32(td, fpregs32));
325 }
326 #endif
327 
328 int
329 proc_sstep(struct thread *td)
330 {
331 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
332 	return (ptrace_single_step(td));
333 }
334 
335 int
336 proc_rwmem(struct proc *p, struct uio *uio)
337 {
338 	vm_map_t map;
339 	vm_offset_t pageno;		/* page number */
340 	vm_prot_t reqprot;
341 	int error, fault_flags, page_offset, writing;
342 
343 	/*
344 	 * Make sure that the process' vmspace remains live.
345 	 */
346 	if (p != curproc)
347 		PROC_ASSERT_HELD(p);
348 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
349 
350 	/*
351 	 * The map we want...
352 	 */
353 	map = &p->p_vmspace->vm_map;
354 
355 	/*
356 	 * If we are writing, then we request vm_fault() to create a private
357 	 * copy of each page.  Since these copies will not be writeable by the
358 	 * process, we must explicity request that they be dirtied.
359 	 */
360 	writing = uio->uio_rw == UIO_WRITE;
361 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
362 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
363 
364 	/*
365 	 * Only map in one page at a time.  We don't have to, but it
366 	 * makes things easier.  This way is trivial - right?
367 	 */
368 	do {
369 		vm_offset_t uva;
370 		u_int len;
371 		vm_page_t m;
372 
373 		uva = (vm_offset_t)uio->uio_offset;
374 
375 		/*
376 		 * Get the page number of this segment.
377 		 */
378 		pageno = trunc_page(uva);
379 		page_offset = uva - pageno;
380 
381 		/*
382 		 * How many bytes to copy
383 		 */
384 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
385 
386 		/*
387 		 * Fault and hold the page on behalf of the process.
388 		 */
389 		error = vm_fault(map, pageno, reqprot, fault_flags, &m);
390 		if (error != KERN_SUCCESS) {
391 			if (error == KERN_RESOURCE_SHORTAGE)
392 				error = ENOMEM;
393 			else
394 				error = EFAULT;
395 			break;
396 		}
397 
398 		/*
399 		 * Now do the i/o move.
400 		 */
401 		error = uiomove_fromphys(&m, page_offset, len, uio);
402 
403 		/* Make the I-cache coherent for breakpoints. */
404 		if (writing && error == 0) {
405 			vm_map_lock_read(map);
406 			if (vm_map_check_protection(map, pageno, pageno +
407 			    PAGE_SIZE, VM_PROT_EXECUTE))
408 				vm_sync_icache(map, uva, len);
409 			vm_map_unlock_read(map);
410 		}
411 
412 		/*
413 		 * Release the page.
414 		 */
415 		vm_page_unwire(m, PQ_ACTIVE);
416 
417 	} while (error == 0 && uio->uio_resid > 0);
418 
419 	return (error);
420 }
421 
422 static ssize_t
423 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
424     size_t len, enum uio_rw rw)
425 {
426 	struct iovec iov;
427 	struct uio uio;
428 	ssize_t slen;
429 
430 	MPASS(len < SSIZE_MAX);
431 	slen = (ssize_t)len;
432 
433 	iov.iov_base = (caddr_t)buf;
434 	iov.iov_len = len;
435 	uio.uio_iov = &iov;
436 	uio.uio_iovcnt = 1;
437 	uio.uio_offset = va;
438 	uio.uio_resid = slen;
439 	uio.uio_segflg = UIO_SYSSPACE;
440 	uio.uio_rw = rw;
441 	uio.uio_td = td;
442 	proc_rwmem(p, &uio);
443 	if (uio.uio_resid == slen)
444 		return (-1);
445 	return (slen - uio.uio_resid);
446 }
447 
448 ssize_t
449 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
450     size_t len)
451 {
452 
453 	return (proc_iop(td, p, va, buf, len, UIO_READ));
454 }
455 
456 ssize_t
457 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
458     size_t len)
459 {
460 
461 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
462 }
463 
464 static int
465 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
466 {
467 	struct vattr vattr;
468 	vm_map_t map;
469 	vm_map_entry_t entry;
470 	vm_object_t obj, tobj, lobj;
471 	struct vmspace *vm;
472 	struct vnode *vp;
473 	char *freepath, *fullpath;
474 	u_int pathlen;
475 	int error, index;
476 
477 	error = 0;
478 	obj = NULL;
479 
480 	vm = vmspace_acquire_ref(p);
481 	map = &vm->vm_map;
482 	vm_map_lock_read(map);
483 
484 	do {
485 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
486 		    ("Submap in map header"));
487 		index = 0;
488 		VM_MAP_ENTRY_FOREACH(entry, map) {
489 			if (index >= pve->pve_entry &&
490 			    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
491 				break;
492 			index++;
493 		}
494 		if (index < pve->pve_entry) {
495 			error = EINVAL;
496 			break;
497 		}
498 		if (entry == &map->header) {
499 			error = ENOENT;
500 			break;
501 		}
502 
503 		/* We got an entry. */
504 		pve->pve_entry = index + 1;
505 		pve->pve_timestamp = map->timestamp;
506 		pve->pve_start = entry->start;
507 		pve->pve_end = entry->end - 1;
508 		pve->pve_offset = entry->offset;
509 		pve->pve_prot = entry->protection |
510 		    PROT_MAX(entry->max_protection);
511 
512 		/* Backing object's path needed? */
513 		if (pve->pve_pathlen == 0)
514 			break;
515 
516 		pathlen = pve->pve_pathlen;
517 		pve->pve_pathlen = 0;
518 
519 		obj = entry->object.vm_object;
520 		if (obj != NULL)
521 			VM_OBJECT_RLOCK(obj);
522 	} while (0);
523 
524 	vm_map_unlock_read(map);
525 
526 	pve->pve_fsid = VNOVAL;
527 	pve->pve_fileid = VNOVAL;
528 
529 	if (error == 0 && obj != NULL) {
530 		lobj = obj;
531 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
532 			if (tobj != obj)
533 				VM_OBJECT_RLOCK(tobj);
534 			if (lobj != obj)
535 				VM_OBJECT_RUNLOCK(lobj);
536 			lobj = tobj;
537 			pve->pve_offset += tobj->backing_object_offset;
538 		}
539 		vp = vm_object_vnode(lobj);
540 		if (vp != NULL)
541 			vref(vp);
542 		if (lobj != obj)
543 			VM_OBJECT_RUNLOCK(lobj);
544 		VM_OBJECT_RUNLOCK(obj);
545 
546 		if (vp != NULL) {
547 			freepath = NULL;
548 			fullpath = NULL;
549 			vn_fullpath(vp, &fullpath, &freepath);
550 			vn_lock(vp, LK_SHARED | LK_RETRY);
551 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
552 				pve->pve_fileid = vattr.va_fileid;
553 				pve->pve_fsid = vattr.va_fsid;
554 			}
555 			vput(vp);
556 
557 			if (fullpath != NULL) {
558 				pve->pve_pathlen = strlen(fullpath) + 1;
559 				if (pve->pve_pathlen <= pathlen) {
560 					error = copyout(fullpath, pve->pve_path,
561 					    pve->pve_pathlen);
562 				} else
563 					error = ENAMETOOLONG;
564 			}
565 			if (freepath != NULL)
566 				free(freepath, M_TEMP);
567 		}
568 	}
569 	vmspace_free(vm);
570 	if (error == 0)
571 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
572 		    p->p_pid, pve->pve_entry, pve->pve_start);
573 
574 	return (error);
575 }
576 
577 /*
578  * Process debugging system call.
579  */
580 #ifndef _SYS_SYSPROTO_H_
581 struct ptrace_args {
582 	int	req;
583 	pid_t	pid;
584 	caddr_t	addr;
585 	int	data;
586 };
587 #endif
588 
589 int
590 sys_ptrace(struct thread *td, struct ptrace_args *uap)
591 {
592 	/*
593 	 * XXX this obfuscation is to reduce stack usage, but the register
594 	 * structs may be too large to put on the stack anyway.
595 	 */
596 	union {
597 		struct ptrace_io_desc piod;
598 		struct ptrace_lwpinfo pl;
599 		struct ptrace_vm_entry pve;
600 		struct ptrace_coredump pc;
601 		struct ptrace_sc_remote sr;
602 		struct dbreg dbreg;
603 		struct fpreg fpreg;
604 		struct reg reg;
605 		struct iovec vec;
606 		syscallarg_t args[nitems(td->td_sa.args)];
607 		struct ptrace_sc_ret psr;
608 		int ptevents;
609 	} r;
610 	syscallarg_t pscr_args[nitems(td->td_sa.args)];
611 	void *addr;
612 	int error;
613 
614 	if (!allow_ptrace)
615 		return (ENOSYS);
616 	error = 0;
617 
618 	AUDIT_ARG_PID(uap->pid);
619 	AUDIT_ARG_CMD(uap->req);
620 	AUDIT_ARG_VALUE(uap->data);
621 	addr = &r;
622 	switch (uap->req) {
623 	case PT_GET_EVENT_MASK:
624 	case PT_LWPINFO:
625 	case PT_GET_SC_ARGS:
626 	case PT_GET_SC_RET:
627 		break;
628 	case PT_GETREGS:
629 		bzero(&r.reg, sizeof(r.reg));
630 		break;
631 	case PT_GETFPREGS:
632 		bzero(&r.fpreg, sizeof(r.fpreg));
633 		break;
634 	case PT_GETDBREGS:
635 		bzero(&r.dbreg, sizeof(r.dbreg));
636 		break;
637 	case PT_GETREGSET:
638 	case PT_SETREGSET:
639 		error = copyin(uap->addr, &r.vec, sizeof(r.vec));
640 		break;
641 	case PT_SETREGS:
642 		error = copyin(uap->addr, &r.reg, sizeof(r.reg));
643 		break;
644 	case PT_SETFPREGS:
645 		error = copyin(uap->addr, &r.fpreg, sizeof(r.fpreg));
646 		break;
647 	case PT_SETDBREGS:
648 		error = copyin(uap->addr, &r.dbreg, sizeof(r.dbreg));
649 		break;
650 	case PT_SET_EVENT_MASK:
651 		if (uap->data != sizeof(r.ptevents))
652 			error = EINVAL;
653 		else
654 			error = copyin(uap->addr, &r.ptevents, uap->data);
655 		break;
656 	case PT_IO:
657 		error = copyin(uap->addr, &r.piod, sizeof(r.piod));
658 		break;
659 	case PT_VM_ENTRY:
660 		error = copyin(uap->addr, &r.pve, sizeof(r.pve));
661 		break;
662 	case PT_COREDUMP:
663 		if (uap->data != sizeof(r.pc))
664 			error = EINVAL;
665 		else
666 			error = copyin(uap->addr, &r.pc, uap->data);
667 		break;
668 	case PT_SC_REMOTE:
669 		if (uap->data != sizeof(r.sr)) {
670 			error = EINVAL;
671 			break;
672 		}
673 		error = copyin(uap->addr, &r.sr, uap->data);
674 		if (error != 0)
675 			break;
676 		if (r.sr.pscr_nargs > nitems(td->td_sa.args)) {
677 			error = EINVAL;
678 			break;
679 		}
680 		error = copyin(r.sr.pscr_args, pscr_args,
681 		    sizeof(u_long) * r.sr.pscr_nargs);
682 		if (error != 0)
683 			break;
684 		r.sr.pscr_args = pscr_args;
685 		break;
686 	default:
687 		addr = uap->addr;
688 		break;
689 	}
690 	if (error)
691 		return (error);
692 
693 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
694 	if (error)
695 		return (error);
696 
697 	switch (uap->req) {
698 	case PT_VM_ENTRY:
699 		error = copyout(&r.pve, uap->addr, sizeof(r.pve));
700 		break;
701 	case PT_IO:
702 		error = copyout(&r.piod, uap->addr, sizeof(r.piod));
703 		break;
704 	case PT_GETREGS:
705 		error = copyout(&r.reg, uap->addr, sizeof(r.reg));
706 		break;
707 	case PT_GETFPREGS:
708 		error = copyout(&r.fpreg, uap->addr, sizeof(r.fpreg));
709 		break;
710 	case PT_GETDBREGS:
711 		error = copyout(&r.dbreg, uap->addr, sizeof(r.dbreg));
712 		break;
713 	case PT_GETREGSET:
714 		error = copyout(&r.vec, uap->addr, sizeof(r.vec));
715 		break;
716 	case PT_GET_EVENT_MASK:
717 		/* NB: The size in uap->data is validated in kern_ptrace(). */
718 		error = copyout(&r.ptevents, uap->addr, uap->data);
719 		break;
720 	case PT_LWPINFO:
721 		/* NB: The size in uap->data is validated in kern_ptrace(). */
722 		error = copyout(&r.pl, uap->addr, uap->data);
723 		break;
724 	case PT_GET_SC_ARGS:
725 		error = copyout(r.args, uap->addr, MIN(uap->data,
726 		    sizeof(r.args)));
727 		break;
728 	case PT_GET_SC_RET:
729 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
730 		    sizeof(r.psr)));
731 		break;
732 	case PT_SC_REMOTE:
733 		error = copyout(&r.sr.pscr_ret, uap->addr +
734 		    offsetof(struct ptrace_sc_remote, pscr_ret),
735 		    sizeof(r.sr.pscr_ret));
736 		break;
737 	}
738 
739 	return (error);
740 }
741 
742 #ifdef COMPAT_FREEBSD32
743 /*
744  *   PROC_READ(regs, td2, addr);
745  * becomes either:
746  *   proc_read_regs(td2, addr);
747  * or
748  *   proc_read_regs32(td2, addr);
749  * .. except this is done at runtime.  There is an additional
750  * complication in that PROC_WRITE disallows 32 bit consumers
751  * from writing to 64 bit address space targets.
752  */
753 #define	PROC_READ(w, t, a)	wrap32 ? \
754 	proc_read_ ## w ## 32(t, a) : \
755 	proc_read_ ## w (t, a)
756 #define	PROC_WRITE(w, t, a)	wrap32 ? \
757 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
758 	proc_write_ ## w (t, a)
759 #else
760 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
761 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
762 #endif
763 
764 void
765 proc_set_traced(struct proc *p, bool stop)
766 {
767 
768 	sx_assert(&proctree_lock, SX_XLOCKED);
769 	PROC_LOCK_ASSERT(p, MA_OWNED);
770 	p->p_flag |= P_TRACED;
771 	if (stop)
772 		p->p_flag2 |= P2_PTRACE_FSTP;
773 	p->p_ptevents = PTRACE_DEFAULT;
774 }
775 
776 void
777 ptrace_unsuspend(struct proc *p)
778 {
779 	PROC_LOCK_ASSERT(p, MA_OWNED);
780 
781 	PROC_SLOCK(p);
782 	p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
783 	thread_unsuspend(p);
784 	PROC_SUNLOCK(p);
785 	itimer_proc_continue(p);
786 	kqtimer_proc_continue(p);
787 }
788 
789 static int
790 proc_can_ptrace(struct thread *td, struct proc *p)
791 {
792 	int error;
793 
794 	PROC_LOCK_ASSERT(p, MA_OWNED);
795 
796 	if ((p->p_flag & P_WEXIT) != 0)
797 		return (ESRCH);
798 
799 	if ((error = p_cansee(td, p)) != 0)
800 		return (error);
801 	if ((error = p_candebug(td, p)) != 0)
802 		return (error);
803 
804 	/* not being traced... */
805 	if ((p->p_flag & P_TRACED) == 0)
806 		return (EPERM);
807 
808 	/* not being traced by YOU */
809 	if (p->p_pptr != td->td_proc)
810 		return (EBUSY);
811 
812 	/* not currently stopped */
813 	if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
814 	    p->p_suspcount != p->p_numthreads  ||
815 	    (p->p_flag & P_WAITED) == 0)
816 		return (EBUSY);
817 
818 	return (0);
819 }
820 
821 static struct thread *
822 ptrace_sel_coredump_thread(struct proc *p)
823 {
824 	struct thread *td2;
825 
826 	PROC_LOCK_ASSERT(p, MA_OWNED);
827 	MPASS((p->p_flag & P_STOPPED_TRACE) != 0);
828 
829 	FOREACH_THREAD_IN_PROC(p, td2) {
830 		if ((td2->td_dbgflags & TDB_SSWITCH) != 0)
831 			return (td2);
832 	}
833 	return (NULL);
834 }
835 
836 int
837 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
838 {
839 	struct iovec iov;
840 	struct uio uio;
841 	struct proc *curp, *p, *pp;
842 	struct thread *td2 = NULL, *td3;
843 	struct ptrace_io_desc *piod = NULL;
844 	struct ptrace_lwpinfo *pl;
845 	struct ptrace_sc_ret *psr;
846 	struct ptrace_sc_remote *pscr;
847 	struct file *fp;
848 	struct ptrace_coredump *pc;
849 	struct thr_coredump_req *tcq;
850 	struct thr_syscall_req *tsr;
851 	int error, num, tmp;
852 	lwpid_t tid = 0, *buf;
853 #ifdef COMPAT_FREEBSD32
854 	int wrap32 = 0, safe = 0;
855 #endif
856 	bool proctree_locked, p2_req_set;
857 
858 	curp = td->td_proc;
859 	proctree_locked = false;
860 	p2_req_set = false;
861 
862 	/* Lock proctree before locking the process. */
863 	switch (req) {
864 	case PT_TRACE_ME:
865 	case PT_ATTACH:
866 	case PT_STEP:
867 	case PT_CONTINUE:
868 	case PT_TO_SCE:
869 	case PT_TO_SCX:
870 	case PT_SYSCALL:
871 	case PT_FOLLOW_FORK:
872 	case PT_LWP_EVENTS:
873 	case PT_GET_EVENT_MASK:
874 	case PT_SET_EVENT_MASK:
875 	case PT_DETACH:
876 	case PT_GET_SC_ARGS:
877 		sx_xlock(&proctree_lock);
878 		proctree_locked = true;
879 		break;
880 	default:
881 		break;
882 	}
883 
884 	if (req == PT_TRACE_ME) {
885 		p = td->td_proc;
886 		PROC_LOCK(p);
887 	} else {
888 		if (pid <= PID_MAX) {
889 			if ((p = pfind(pid)) == NULL) {
890 				if (proctree_locked)
891 					sx_xunlock(&proctree_lock);
892 				return (ESRCH);
893 			}
894 		} else {
895 			td2 = tdfind(pid, -1);
896 			if (td2 == NULL) {
897 				if (proctree_locked)
898 					sx_xunlock(&proctree_lock);
899 				return (ESRCH);
900 			}
901 			p = td2->td_proc;
902 			tid = pid;
903 			pid = p->p_pid;
904 		}
905 	}
906 	AUDIT_ARG_PROCESS(p);
907 
908 	if ((p->p_flag & P_WEXIT) != 0) {
909 		error = ESRCH;
910 		goto fail;
911 	}
912 	if ((error = p_cansee(td, p)) != 0)
913 		goto fail;
914 
915 	if ((error = p_candebug(td, p)) != 0)
916 		goto fail;
917 
918 	/*
919 	 * System processes can't be debugged.
920 	 */
921 	if ((p->p_flag & P_SYSTEM) != 0) {
922 		error = EINVAL;
923 		goto fail;
924 	}
925 
926 	if (tid == 0) {
927 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
928 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
929 			td2 = p->p_xthread;
930 		} else {
931 			td2 = FIRST_THREAD_IN_PROC(p);
932 		}
933 		tid = td2->td_tid;
934 	}
935 
936 #ifdef COMPAT_FREEBSD32
937 	/*
938 	 * Test if we're a 32 bit client and what the target is.
939 	 * Set the wrap controls accordingly.
940 	 */
941 	if (SV_CURPROC_FLAG(SV_ILP32)) {
942 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
943 			safe = 1;
944 		wrap32 = 1;
945 	}
946 #endif
947 	/*
948 	 * Permissions check
949 	 */
950 	switch (req) {
951 	case PT_TRACE_ME:
952 		/*
953 		 * Always legal, when there is a parent process which
954 		 * could trace us.  Otherwise, reject.
955 		 */
956 		if ((p->p_flag & P_TRACED) != 0) {
957 			error = EBUSY;
958 			goto fail;
959 		}
960 		if (p->p_pptr == initproc) {
961 			error = EPERM;
962 			goto fail;
963 		}
964 		break;
965 
966 	case PT_ATTACH:
967 		/* Self */
968 		if (p == td->td_proc) {
969 			error = EINVAL;
970 			goto fail;
971 		}
972 
973 		/* Already traced */
974 		if (p->p_flag & P_TRACED) {
975 			error = EBUSY;
976 			goto fail;
977 		}
978 
979 		/* Can't trace an ancestor if you're being traced. */
980 		if (curp->p_flag & P_TRACED) {
981 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
982 				if (pp == p) {
983 					error = EINVAL;
984 					goto fail;
985 				}
986 			}
987 		}
988 
989 		/* OK */
990 		break;
991 
992 	case PT_CLEARSTEP:
993 		/* Allow thread to clear single step for itself */
994 		if (td->td_tid == tid)
995 			break;
996 
997 		/* FALLTHROUGH */
998 	default:
999 		/*
1000 		 * Check for ptrace eligibility before waiting for
1001 		 * holds to drain.
1002 		 */
1003 		error = proc_can_ptrace(td, p);
1004 		if (error != 0)
1005 			goto fail;
1006 
1007 		/*
1008 		 * Block parallel ptrace requests.  Most important, do
1009 		 * not allow other thread in debugger to continue the
1010 		 * debuggee until coredump finished.
1011 		 */
1012 		while ((p->p_flag2 & P2_PTRACEREQ) != 0) {
1013 			if (proctree_locked)
1014 				sx_xunlock(&proctree_lock);
1015 			error = msleep(&p->p_flag2, &p->p_mtx, PPAUSE | PCATCH |
1016 			    (proctree_locked ? PDROP : 0), "pptrace", 0);
1017 			if (proctree_locked) {
1018 				sx_xlock(&proctree_lock);
1019 				PROC_LOCK(p);
1020 			}
1021 			if (error == 0 && td2->td_proc != p)
1022 				error = ESRCH;
1023 			if (error == 0)
1024 				error = proc_can_ptrace(td, p);
1025 			if (error != 0)
1026 				goto fail;
1027 		}
1028 
1029 		/* Ok */
1030 		break;
1031 	}
1032 
1033 	/*
1034 	 * Keep this process around and request parallel ptrace()
1035 	 * request to wait until we finish this request.
1036 	 */
1037 	MPASS((p->p_flag2 & P2_PTRACEREQ) == 0);
1038 	p->p_flag2 |= P2_PTRACEREQ;
1039 	p2_req_set = true;
1040 	_PHOLD(p);
1041 
1042 	/*
1043 	 * Actually do the requests
1044 	 */
1045 
1046 	td->td_retval[0] = 0;
1047 
1048 	switch (req) {
1049 	case PT_TRACE_ME:
1050 		/* set my trace flag and "owner" so it can read/write me */
1051 		proc_set_traced(p, false);
1052 		if (p->p_flag & P_PPWAIT)
1053 			p->p_flag |= P_PPTRACE;
1054 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
1055 		break;
1056 
1057 	case PT_ATTACH:
1058 		/* security check done above */
1059 		/*
1060 		 * It would be nice if the tracing relationship was separate
1061 		 * from the parent relationship but that would require
1062 		 * another set of links in the proc struct or for "wait"
1063 		 * to scan the entire proc table.  To make life easier,
1064 		 * we just re-parent the process we're trying to trace.
1065 		 * The old parent is remembered so we can put things back
1066 		 * on a "detach".
1067 		 */
1068 		proc_set_traced(p, true);
1069 		proc_reparent(p, td->td_proc, false);
1070 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
1071 		    p->p_oppid);
1072 
1073 		sx_xunlock(&proctree_lock);
1074 		proctree_locked = false;
1075 		MPASS(p->p_xthread == NULL);
1076 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
1077 
1078 		/*
1079 		 * If already stopped due to a stop signal, clear the
1080 		 * existing stop before triggering a traced SIGSTOP.
1081 		 */
1082 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
1083 			PROC_SLOCK(p);
1084 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
1085 			thread_unsuspend(p);
1086 			PROC_SUNLOCK(p);
1087 		}
1088 
1089 		kern_psignal(p, SIGSTOP);
1090 		break;
1091 
1092 	case PT_CLEARSTEP:
1093 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
1094 		    p->p_pid);
1095 		error = ptrace_clear_single_step(td2);
1096 		break;
1097 
1098 	case PT_SETSTEP:
1099 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
1100 		    p->p_pid);
1101 		error = ptrace_single_step(td2);
1102 		break;
1103 
1104 	case PT_SUSPEND:
1105 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
1106 		    p->p_pid);
1107 		td2->td_dbgflags |= TDB_SUSPEND;
1108 		ast_sched(td2, TDA_SUSPEND);
1109 		break;
1110 
1111 	case PT_RESUME:
1112 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1113 		    p->p_pid);
1114 		td2->td_dbgflags &= ~TDB_SUSPEND;
1115 		break;
1116 
1117 	case PT_FOLLOW_FORK:
1118 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1119 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1120 		    data ? "enabled" : "disabled");
1121 		if (data)
1122 			p->p_ptevents |= PTRACE_FORK;
1123 		else
1124 			p->p_ptevents &= ~PTRACE_FORK;
1125 		break;
1126 
1127 	case PT_LWP_EVENTS:
1128 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1129 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1130 		    data ? "enabled" : "disabled");
1131 		if (data)
1132 			p->p_ptevents |= PTRACE_LWP;
1133 		else
1134 			p->p_ptevents &= ~PTRACE_LWP;
1135 		break;
1136 
1137 	case PT_GET_EVENT_MASK:
1138 		if (data != sizeof(p->p_ptevents)) {
1139 			error = EINVAL;
1140 			break;
1141 		}
1142 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1143 		    p->p_ptevents);
1144 		*(int *)addr = p->p_ptevents;
1145 		break;
1146 
1147 	case PT_SET_EVENT_MASK:
1148 		if (data != sizeof(p->p_ptevents)) {
1149 			error = EINVAL;
1150 			break;
1151 		}
1152 		tmp = *(int *)addr;
1153 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1154 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1155 			error = EINVAL;
1156 			break;
1157 		}
1158 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1159 		    p->p_pid, p->p_ptevents, tmp);
1160 		p->p_ptevents = tmp;
1161 		break;
1162 
1163 	case PT_GET_SC_ARGS:
1164 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1165 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1166 #ifdef COMPAT_FREEBSD32
1167 		    || (wrap32 && !safe)
1168 #endif
1169 		    ) {
1170 			error = EINVAL;
1171 			break;
1172 		}
1173 		bzero(addr, sizeof(td2->td_sa.args));
1174 		/* See the explanation in linux_ptrace_get_syscall_info(). */
1175 		bcopy(td2->td_sa.args, addr, SV_PROC_ABI(td->td_proc) ==
1176 		    SV_ABI_LINUX ? sizeof(td2->td_sa.args) :
1177 		    td2->td_sa.callp->sy_narg * sizeof(syscallarg_t));
1178 		break;
1179 
1180 	case PT_GET_SC_RET:
1181 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1182 #ifdef COMPAT_FREEBSD32
1183 		    || (wrap32 && !safe)
1184 #endif
1185 		    ) {
1186 			error = EINVAL;
1187 			break;
1188 		}
1189 		psr = addr;
1190 		bzero(psr, sizeof(*psr));
1191 		psr->sr_error = td2->td_errno;
1192 		if (psr->sr_error == 0) {
1193 			psr->sr_retval[0] = td2->td_retval[0];
1194 			psr->sr_retval[1] = td2->td_retval[1];
1195 		}
1196 		CTR4(KTR_PTRACE,
1197 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1198 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1199 		    psr->sr_retval[1]);
1200 		break;
1201 
1202 	case PT_STEP:
1203 	case PT_CONTINUE:
1204 	case PT_TO_SCE:
1205 	case PT_TO_SCX:
1206 	case PT_SYSCALL:
1207 	case PT_DETACH:
1208 		/* Zero means do not send any signal */
1209 		if (data < 0 || data > _SIG_MAXSIG) {
1210 			error = EINVAL;
1211 			break;
1212 		}
1213 
1214 		switch (req) {
1215 		case PT_STEP:
1216 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1217 			    td2->td_tid, p->p_pid, data);
1218 			error = ptrace_single_step(td2);
1219 			if (error)
1220 				goto out;
1221 			break;
1222 		case PT_CONTINUE:
1223 		case PT_TO_SCE:
1224 		case PT_TO_SCX:
1225 		case PT_SYSCALL:
1226 			if (addr != (void *)1) {
1227 				error = ptrace_set_pc(td2,
1228 				    (u_long)(uintfptr_t)addr);
1229 				if (error)
1230 					goto out;
1231 			}
1232 			switch (req) {
1233 			case PT_TO_SCE:
1234 				p->p_ptevents |= PTRACE_SCE;
1235 				CTR4(KTR_PTRACE,
1236 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1237 				    p->p_pid, p->p_ptevents,
1238 				    (u_long)(uintfptr_t)addr, data);
1239 				break;
1240 			case PT_TO_SCX:
1241 				p->p_ptevents |= PTRACE_SCX;
1242 				CTR4(KTR_PTRACE,
1243 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1244 				    p->p_pid, p->p_ptevents,
1245 				    (u_long)(uintfptr_t)addr, data);
1246 				break;
1247 			case PT_SYSCALL:
1248 				p->p_ptevents |= PTRACE_SYSCALL;
1249 				CTR4(KTR_PTRACE,
1250 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1251 				    p->p_pid, p->p_ptevents,
1252 				    (u_long)(uintfptr_t)addr, data);
1253 				break;
1254 			case PT_CONTINUE:
1255 				CTR3(KTR_PTRACE,
1256 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1257 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1258 				break;
1259 			}
1260 			break;
1261 		case PT_DETACH:
1262 			/*
1263 			 * Clear P_TRACED before reparenting
1264 			 * a detached process back to its original
1265 			 * parent.  Otherwise the debugee will be set
1266 			 * as an orphan of the debugger.
1267 			 */
1268 			p->p_flag &= ~(P_TRACED | P_WAITED);
1269 
1270 			/*
1271 			 * Reset the process parent.
1272 			 */
1273 			if (p->p_oppid != p->p_pptr->p_pid) {
1274 				PROC_LOCK(p->p_pptr);
1275 				sigqueue_take(p->p_ksi);
1276 				PROC_UNLOCK(p->p_pptr);
1277 
1278 				pp = proc_realparent(p);
1279 				proc_reparent(p, pp, false);
1280 				if (pp == initproc)
1281 					p->p_sigparent = SIGCHLD;
1282 				CTR3(KTR_PTRACE,
1283 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1284 				    p->p_pid, pp->p_pid, data);
1285 			} else {
1286 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1287 				    p->p_pid, data);
1288 			}
1289 
1290 			p->p_ptevents = 0;
1291 			FOREACH_THREAD_IN_PROC(p, td3) {
1292 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1293 					sigqueue_delete(&td3->td_sigqueue,
1294 					    SIGSTOP);
1295 				}
1296 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1297 				    TDB_SUSPEND | TDB_BORN);
1298 			}
1299 
1300 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1301 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1302 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1303 			}
1304 
1305 			/* should we send SIGCHLD? */
1306 			/* childproc_continued(p); */
1307 			break;
1308 		}
1309 
1310 		sx_xunlock(&proctree_lock);
1311 		proctree_locked = false;
1312 
1313 	sendsig:
1314 		MPASS(!proctree_locked);
1315 
1316 		/*
1317 		 * Clear the pending event for the thread that just
1318 		 * reported its event (p_xthread).  This may not be
1319 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1320 		 * the debugger is resuming a different thread.
1321 		 *
1322 		 * Deliver any pending signal via the reporting thread.
1323 		 */
1324 		MPASS(p->p_xthread != NULL);
1325 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1326 		p->p_xthread->td_xsig = data;
1327 		p->p_xthread = NULL;
1328 		p->p_xsig = data;
1329 
1330 		/*
1331 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1332 		 * always works immediately, even if another thread is
1333 		 * unsuspended first and attempts to handle a
1334 		 * different signal or if the POSIX.1b style signal
1335 		 * queue cannot accommodate any new signals.
1336 		 */
1337 		if (data == SIGKILL)
1338 			proc_wkilled(p);
1339 
1340 		/*
1341 		 * Unsuspend all threads.  To leave a thread
1342 		 * suspended, use PT_SUSPEND to suspend it before
1343 		 * continuing the process.
1344 		 */
1345 		ptrace_unsuspend(p);
1346 		break;
1347 
1348 	case PT_WRITE_I:
1349 	case PT_WRITE_D:
1350 		td2->td_dbgflags |= TDB_USERWR;
1351 		PROC_UNLOCK(p);
1352 		error = 0;
1353 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1354 		    sizeof(int)) != sizeof(int))
1355 			error = ENOMEM;
1356 		else
1357 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1358 			    p->p_pid, addr, data);
1359 		PROC_LOCK(p);
1360 		break;
1361 
1362 	case PT_READ_I:
1363 	case PT_READ_D:
1364 		PROC_UNLOCK(p);
1365 		error = tmp = 0;
1366 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1367 		    sizeof(int)) != sizeof(int))
1368 			error = ENOMEM;
1369 		else
1370 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1371 			    p->p_pid, addr, tmp);
1372 		td->td_retval[0] = tmp;
1373 		PROC_LOCK(p);
1374 		break;
1375 
1376 	case PT_IO:
1377 		piod = addr;
1378 		iov.iov_base = piod->piod_addr;
1379 		iov.iov_len = piod->piod_len;
1380 		uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1381 		uio.uio_resid = piod->piod_len;
1382 		uio.uio_iov = &iov;
1383 		uio.uio_iovcnt = 1;
1384 		uio.uio_segflg = UIO_USERSPACE;
1385 		uio.uio_td = td;
1386 		switch (piod->piod_op) {
1387 		case PIOD_READ_D:
1388 		case PIOD_READ_I:
1389 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1390 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1391 			uio.uio_rw = UIO_READ;
1392 			break;
1393 		case PIOD_WRITE_D:
1394 		case PIOD_WRITE_I:
1395 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1396 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1397 			td2->td_dbgflags |= TDB_USERWR;
1398 			uio.uio_rw = UIO_WRITE;
1399 			break;
1400 		default:
1401 			error = EINVAL;
1402 			goto out;
1403 		}
1404 		PROC_UNLOCK(p);
1405 		error = proc_rwmem(p, &uio);
1406 		piod->piod_len -= uio.uio_resid;
1407 		PROC_LOCK(p);
1408 		break;
1409 
1410 	case PT_KILL:
1411 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1412 		data = SIGKILL;
1413 		goto sendsig;	/* in PT_CONTINUE above */
1414 
1415 	case PT_SETREGS:
1416 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1417 		    p->p_pid);
1418 		td2->td_dbgflags |= TDB_USERWR;
1419 		error = PROC_WRITE(regs, td2, addr);
1420 		break;
1421 
1422 	case PT_GETREGS:
1423 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1424 		    p->p_pid);
1425 		error = PROC_READ(regs, td2, addr);
1426 		break;
1427 
1428 	case PT_SETFPREGS:
1429 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1430 		    p->p_pid);
1431 		td2->td_dbgflags |= TDB_USERWR;
1432 		error = PROC_WRITE(fpregs, td2, addr);
1433 		break;
1434 
1435 	case PT_GETFPREGS:
1436 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1437 		    p->p_pid);
1438 		error = PROC_READ(fpregs, td2, addr);
1439 		break;
1440 
1441 	case PT_SETDBREGS:
1442 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1443 		    p->p_pid);
1444 		td2->td_dbgflags |= TDB_USERWR;
1445 		error = PROC_WRITE(dbregs, td2, addr);
1446 		break;
1447 
1448 	case PT_GETDBREGS:
1449 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1450 		    p->p_pid);
1451 		error = PROC_READ(dbregs, td2, addr);
1452 		break;
1453 
1454 	case PT_SETREGSET:
1455 		CTR2(KTR_PTRACE, "PT_SETREGSET: tid %d (pid %d)", td2->td_tid,
1456 		    p->p_pid);
1457 		error = proc_write_regset(td2, data, addr);
1458 		break;
1459 
1460 	case PT_GETREGSET:
1461 		CTR2(KTR_PTRACE, "PT_GETREGSET: tid %d (pid %d)", td2->td_tid,
1462 		    p->p_pid);
1463 		error = proc_read_regset(td2, data, addr);
1464 		break;
1465 
1466 	case PT_LWPINFO:
1467 		if (data <= 0 || data > sizeof(*pl)) {
1468 			error = EINVAL;
1469 			break;
1470 		}
1471 		pl = addr;
1472 		bzero(pl, sizeof(*pl));
1473 		pl->pl_lwpid = td2->td_tid;
1474 		pl->pl_event = PL_EVENT_NONE;
1475 		pl->pl_flags = 0;
1476 		if (td2->td_dbgflags & TDB_XSIG) {
1477 			pl->pl_event = PL_EVENT_SIGNAL;
1478 			if (td2->td_si.si_signo != 0 &&
1479 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1480 			    + sizeof(pl->pl_siginfo)){
1481 				pl->pl_flags |= PL_FLAG_SI;
1482 				pl->pl_siginfo = td2->td_si;
1483 			}
1484 		}
1485 		if (td2->td_dbgflags & TDB_SCE)
1486 			pl->pl_flags |= PL_FLAG_SCE;
1487 		else if (td2->td_dbgflags & TDB_SCX)
1488 			pl->pl_flags |= PL_FLAG_SCX;
1489 		if (td2->td_dbgflags & TDB_EXEC)
1490 			pl->pl_flags |= PL_FLAG_EXEC;
1491 		if (td2->td_dbgflags & TDB_FORK) {
1492 			pl->pl_flags |= PL_FLAG_FORKED;
1493 			pl->pl_child_pid = td2->td_dbg_forked;
1494 			if (td2->td_dbgflags & TDB_VFORK)
1495 				pl->pl_flags |= PL_FLAG_VFORKED;
1496 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1497 		    TDB_VFORK)
1498 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1499 		if (td2->td_dbgflags & TDB_CHILD)
1500 			pl->pl_flags |= PL_FLAG_CHILD;
1501 		if (td2->td_dbgflags & TDB_BORN)
1502 			pl->pl_flags |= PL_FLAG_BORN;
1503 		if (td2->td_dbgflags & TDB_EXIT)
1504 			pl->pl_flags |= PL_FLAG_EXITED;
1505 		pl->pl_sigmask = td2->td_sigmask;
1506 		pl->pl_siglist = td2->td_siglist;
1507 		strcpy(pl->pl_tdname, td2->td_name);
1508 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1509 			pl->pl_syscall_code = td2->td_sa.code;
1510 			pl->pl_syscall_narg = td2->td_sa.callp->sy_narg;
1511 		} else {
1512 			pl->pl_syscall_code = 0;
1513 			pl->pl_syscall_narg = 0;
1514 		}
1515 		CTR6(KTR_PTRACE,
1516     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1517 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1518 		    pl->pl_child_pid, pl->pl_syscall_code);
1519 		break;
1520 
1521 	case PT_GETNUMLWPS:
1522 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1523 		    p->p_numthreads);
1524 		td->td_retval[0] = p->p_numthreads;
1525 		break;
1526 
1527 	case PT_GETLWPLIST:
1528 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1529 		    p->p_pid, data, p->p_numthreads);
1530 		if (data <= 0) {
1531 			error = EINVAL;
1532 			break;
1533 		}
1534 		num = imin(p->p_numthreads, data);
1535 		PROC_UNLOCK(p);
1536 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1537 		tmp = 0;
1538 		PROC_LOCK(p);
1539 		FOREACH_THREAD_IN_PROC(p, td2) {
1540 			if (tmp >= num)
1541 				break;
1542 			buf[tmp++] = td2->td_tid;
1543 		}
1544 		PROC_UNLOCK(p);
1545 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1546 		free(buf, M_TEMP);
1547 		if (!error)
1548 			td->td_retval[0] = tmp;
1549 		PROC_LOCK(p);
1550 		break;
1551 
1552 	case PT_VM_TIMESTAMP:
1553 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1554 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1555 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1556 		break;
1557 
1558 	case PT_VM_ENTRY:
1559 		PROC_UNLOCK(p);
1560 		error = ptrace_vm_entry(td, p, addr);
1561 		PROC_LOCK(p);
1562 		break;
1563 
1564 	case PT_COREDUMP:
1565 		pc = addr;
1566 		CTR2(KTR_PTRACE, "PT_COREDUMP: pid %d, fd %d",
1567 		    p->p_pid, pc->pc_fd);
1568 
1569 		if ((pc->pc_flags & ~(PC_COMPRESS | PC_ALL)) != 0) {
1570 			error = EINVAL;
1571 			break;
1572 		}
1573 		PROC_UNLOCK(p);
1574 
1575 		tcq = malloc(sizeof(*tcq), M_TEMP, M_WAITOK | M_ZERO);
1576 		fp = NULL;
1577 		error = fget_write(td, pc->pc_fd, &cap_write_rights, &fp);
1578 		if (error != 0)
1579 			goto coredump_cleanup_nofp;
1580 		if (fp->f_type != DTYPE_VNODE || fp->f_vnode->v_type != VREG) {
1581 			error = EPIPE;
1582 			goto coredump_cleanup;
1583 		}
1584 
1585 		PROC_LOCK(p);
1586 		error = proc_can_ptrace(td, p);
1587 		if (error != 0)
1588 			goto coredump_cleanup_locked;
1589 
1590 		td2 = ptrace_sel_coredump_thread(p);
1591 		if (td2 == NULL) {
1592 			error = EBUSY;
1593 			goto coredump_cleanup_locked;
1594 		}
1595 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1596 		    TDB_SCREMOTEREQ)) == 0,
1597 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1598 
1599 		tcq->tc_vp = fp->f_vnode;
1600 		tcq->tc_limit = pc->pc_limit == 0 ? OFF_MAX : pc->pc_limit;
1601 		tcq->tc_flags = SVC_PT_COREDUMP;
1602 		if ((pc->pc_flags & PC_COMPRESS) == 0)
1603 			tcq->tc_flags |= SVC_NOCOMPRESS;
1604 		if ((pc->pc_flags & PC_ALL) != 0)
1605 			tcq->tc_flags |= SVC_ALL;
1606 		td2->td_remotereq = tcq;
1607 		td2->td_dbgflags |= TDB_COREDUMPREQ;
1608 		thread_run_flash(td2);
1609 		while ((td2->td_dbgflags & TDB_COREDUMPREQ) != 0)
1610 			msleep(p, &p->p_mtx, PPAUSE, "crdmp", 0);
1611 		error = tcq->tc_error;
1612 coredump_cleanup_locked:
1613 		PROC_UNLOCK(p);
1614 coredump_cleanup:
1615 		fdrop(fp, td);
1616 coredump_cleanup_nofp:
1617 		free(tcq, M_TEMP);
1618 		PROC_LOCK(p);
1619 		break;
1620 
1621 	case PT_SC_REMOTE:
1622 		pscr = addr;
1623 		CTR2(KTR_PTRACE, "PT_SC_REMOTE: pid %d, syscall %d",
1624 		    p->p_pid, pscr->pscr_syscall);
1625 		if ((td2->td_dbgflags & TDB_BOUNDARY) == 0) {
1626 			error = EBUSY;
1627 			break;
1628 		}
1629 		PROC_UNLOCK(p);
1630 		MPASS(pscr->pscr_nargs <= nitems(td->td_sa.args));
1631 
1632 		tsr = malloc(sizeof(struct thr_syscall_req), M_TEMP,
1633 		    M_WAITOK | M_ZERO);
1634 
1635 		tsr->ts_sa.code = pscr->pscr_syscall;
1636 		tsr->ts_nargs = pscr->pscr_nargs;
1637 		memcpy(&tsr->ts_sa.args, pscr->pscr_args,
1638 		    sizeof(syscallarg_t) * tsr->ts_nargs);
1639 
1640 		PROC_LOCK(p);
1641 		error = proc_can_ptrace(td, p);
1642 		if (error != 0) {
1643 			free(tsr, M_TEMP);
1644 			break;
1645 		}
1646 		if (td2->td_proc != p) {
1647 			free(tsr, M_TEMP);
1648 			error = ESRCH;
1649 			break;
1650 		}
1651 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1652 		    TDB_SCREMOTEREQ)) == 0,
1653 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1654 
1655 		td2->td_remotereq = tsr;
1656 		td2->td_dbgflags |= TDB_SCREMOTEREQ;
1657 		thread_run_flash(td2);
1658 		while ((td2->td_dbgflags & TDB_SCREMOTEREQ) != 0)
1659 			msleep(p, &p->p_mtx, PPAUSE, "pscrx", 0);
1660 		error = 0;
1661 		memcpy(&pscr->pscr_ret, &tsr->ts_ret, sizeof(tsr->ts_ret));
1662 		free(tsr, M_TEMP);
1663 		break;
1664 
1665 	default:
1666 #ifdef __HAVE_PTRACE_MACHDEP
1667 		if (req >= PT_FIRSTMACH) {
1668 			PROC_UNLOCK(p);
1669 			error = cpu_ptrace(td2, req, addr, data);
1670 			PROC_LOCK(p);
1671 		} else
1672 #endif
1673 			/* Unknown request. */
1674 			error = EINVAL;
1675 		break;
1676 	}
1677 out:
1678 	/* Drop our hold on this process now that the request has completed. */
1679 	_PRELE(p);
1680 fail:
1681 	if (p2_req_set) {
1682 		if ((p->p_flag2 & P2_PTRACEREQ) != 0)
1683 			wakeup(&p->p_flag2);
1684 		p->p_flag2 &= ~P2_PTRACEREQ;
1685 	}
1686 	PROC_UNLOCK(p);
1687 	if (proctree_locked)
1688 		sx_xunlock(&proctree_lock);
1689 	return (error);
1690 }
1691 #undef PROC_READ
1692 #undef PROC_WRITE
1693