xref: /freebsd/sys/kern/sys_process.c (revision 82aca8ef0dd73a09e46271158f822a7e49082cd0)
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/systm.h>
35 #include <sys/caprights.h>
36 #include <sys/filedesc.h>
37 #include <sys/imgact.h>
38 #include <sys/ktr.h>
39 #include <sys/limits.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mman.h>
43 #include <sys/mutex.h>
44 #include <sys/priv.h>
45 #include <sys/proc.h>
46 #include <sys/ptrace.h>
47 #include <sys/reg.h>
48 #include <sys/rwlock.h>
49 #include <sys/signalvar.h>
50 #include <sys/sleepqueue.h>
51 #include <sys/sx.h>
52 #include <sys/syscallsubr.h>
53 #include <sys/sysent.h>
54 #include <sys/sysproto.h>
55 #include <sys/vnode.h>
56 
57 #include <security/audit/audit.h>
58 
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_extern.h>
62 #include <vm/vm_map.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_param.h>
67 
68 #ifdef COMPAT_FREEBSD32
69 #include <sys/procfs.h>
70 #endif
71 
72 /* Assert it's safe to unlock a process, e.g. to allocate working memory */
73 #define	PROC_ASSERT_TRACEREQ(p)	MPASS(((p)->p_flag2 & P2_PTRACEREQ) != 0)
74 
75 /*
76  * Functions implemented below:
77  *
78  * proc_read_regs(proc, regs)
79  *	Get the current user-visible register set from the process
80  *	and copy it into the regs structure (<machine/reg.h>).
81  *	The process is stopped at the time read_regs is called.
82  *
83  * proc_write_regs(proc, regs)
84  *	Update the current register set from the passed in regs
85  *	structure.  Take care to avoid clobbering special CPU
86  *	registers or privileged bits in the PSL.
87  *	Depending on the architecture this may have fix-up work to do,
88  *	especially if the IAR or PCW are modified.
89  *	The process is stopped at the time write_regs is called.
90  *
91  * proc_read_fpregs, proc_write_fpregs
92  *	deal with the floating point register set, otherwise as above.
93  *
94  * proc_read_dbregs, proc_write_dbregs
95  *	deal with the processor debug register set, otherwise as above.
96  *
97  * proc_sstep(proc)
98  *	Arrange for the process to trap after executing a single instruction.
99  */
100 
101 int
102 proc_read_regs(struct thread *td, struct reg *regs)
103 {
104 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
105 	return (fill_regs(td, regs));
106 }
107 
108 int
109 proc_write_regs(struct thread *td, struct reg *regs)
110 {
111 	int error;
112 
113 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
114 	error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
115 	if (error != 0)
116 		return (error);
117 	return (set_regs(td, regs));
118 }
119 
120 int
121 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
122 {
123 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
124 	return (fill_dbregs(td, dbregs));
125 }
126 
127 int
128 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
129 {
130 	int error;
131 
132 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
133 	error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
134 	if (error != 0)
135 		return (error);
136 	return (set_dbregs(td, dbregs));
137 }
138 
139 /*
140  * Ptrace doesn't support fpregs at all, and there are no security holes
141  * or translations for fpregs, so we can just copy them.
142  */
143 int
144 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
145 {
146 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
147 	return (fill_fpregs(td, fpregs));
148 }
149 
150 int
151 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
152 {
153 	int error;
154 
155 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
156 	error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
157 	if (error != 0)
158 		return (error);
159 	return (set_fpregs(td, fpregs));
160 }
161 
162 static struct regset *
163 proc_find_regset(struct thread *td, int note)
164 {
165 	struct regset **regsetp, **regset_end, *regset;
166 	struct sysentvec *sv;
167 
168 	sv = td->td_proc->p_sysent;
169 	regsetp = sv->sv_regset_begin;
170 	if (regsetp == NULL)
171 		return (NULL);
172 	regset_end = sv->sv_regset_end;
173 	MPASS(regset_end != NULL);
174 	for (; regsetp < regset_end; regsetp++) {
175 		regset = *regsetp;
176 		if (regset->note != note)
177 			continue;
178 
179 		return (regset);
180 	}
181 
182 	return (NULL);
183 }
184 
185 static int
186 proc_read_regset(struct thread *td, int note, struct iovec *iov)
187 {
188 	struct regset *regset;
189 	struct proc *p;
190 	void *buf;
191 	size_t size;
192 	int error;
193 
194 	regset = proc_find_regset(td, note);
195 	if (regset == NULL)
196 		return (EINVAL);
197 
198 	if (regset->get == NULL)
199 		return (EINVAL);
200 
201 	size = regset->size;
202 	/*
203 	 * The regset is dynamically sized, e.g. the size could change
204 	 * depending on the hardware, or may have a per-thread size.
205 	 */
206 	if (size == 0) {
207 		if (!regset->get(regset, td, NULL, &size))
208 			return (EINVAL);
209 	}
210 
211 	if (iov->iov_base == NULL) {
212 		iov->iov_len = size;
213 		if (iov->iov_len == 0)
214 			return (EINVAL);
215 
216 		return (0);
217 	}
218 
219 	/* The length is wrong, return an error */
220 	if (iov->iov_len != size)
221 		return (EINVAL);
222 
223 	error = 0;
224 	p = td->td_proc;
225 
226 	/* Drop the proc lock while allocating the temp buffer */
227 	PROC_ASSERT_TRACEREQ(p);
228 	PROC_UNLOCK(p);
229 	buf = malloc(size, M_TEMP, M_WAITOK);
230 	PROC_LOCK(p);
231 
232 	if (!regset->get(regset, td, buf, &size)) {
233 		error = EINVAL;
234 	} else {
235 		KASSERT(size == regset->size || regset->size == 0,
236 		    ("%s: Getter function changed the size", __func__));
237 
238 		iov->iov_len = size;
239 		PROC_UNLOCK(p);
240 		error = copyout(buf, iov->iov_base, size);
241 		PROC_LOCK(p);
242 	}
243 
244 	free(buf, M_TEMP);
245 
246 	return (error);
247 }
248 
249 static int
250 proc_write_regset(struct thread *td, int note, struct iovec *iov)
251 {
252 	struct regset *regset;
253 	struct proc *p;
254 	void *buf;
255 	size_t size;
256 	int error;
257 
258 	regset = proc_find_regset(td, note);
259 	if (regset == NULL)
260 		return (EINVAL);
261 
262 	size = regset->size;
263 	/*
264 	 * The regset is dynamically sized, e.g. the size could change
265 	 * depending on the hardware, or may have a per-thread size.
266 	 */
267 	if (size == 0) {
268 		if (!regset->get(regset, td, NULL, &size))
269 			return (EINVAL);
270 	}
271 
272 	/* The length is wrong, return an error */
273 	if (iov->iov_len != size)
274 		return (EINVAL);
275 
276 	if (regset->set == NULL)
277 		return (EINVAL);
278 
279 	error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
280 	if (error != 0)
281 		return (error);
282 
283 	p = td->td_proc;
284 
285 	/* Drop the proc lock while allocating the temp buffer */
286 	PROC_ASSERT_TRACEREQ(p);
287 	PROC_UNLOCK(p);
288 	buf = malloc(size, M_TEMP, M_WAITOK);
289 	error = copyin(iov->iov_base, buf, size);
290 	PROC_LOCK(p);
291 
292 	if (error == 0) {
293 		if (!regset->set(regset, td, buf, size)) {
294 			error = EINVAL;
295 		}
296 	}
297 
298 	free(buf, M_TEMP);
299 
300 	return (error);
301 }
302 
303 #ifdef COMPAT_FREEBSD32
304 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
305 int
306 proc_read_regs32(struct thread *td, struct reg32 *regs32)
307 {
308 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
309 	return (fill_regs32(td, regs32));
310 }
311 
312 int
313 proc_write_regs32(struct thread *td, struct reg32 *regs32)
314 {
315 	int error;
316 
317 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
318 	error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
319 	if (error != 0)
320 		return (error);
321 	return (set_regs32(td, regs32));
322 }
323 
324 int
325 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
326 {
327 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
328 	return (fill_dbregs32(td, dbregs32));
329 }
330 
331 int
332 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
333 {
334 	int error;
335 
336 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
337 	error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
338 	if (error != 0)
339 		return (error);
340 	return (set_dbregs32(td, dbregs32));
341 }
342 
343 int
344 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
345 {
346 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
347 	return (fill_fpregs32(td, fpregs32));
348 }
349 
350 int
351 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
352 {
353 	int error;
354 
355 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
356 	error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
357 	if (error != 0)
358 		return (error);
359 	return (set_fpregs32(td, fpregs32));
360 }
361 #endif
362 
363 int
364 proc_sstep(struct thread *td)
365 {
366 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
367 	return (ptrace_single_step(td));
368 }
369 
370 static int
371 proc_vmspace_check_access(struct thread *td, struct proc *p, int flags)
372 {
373 	PROC_ASSERT_HELD(p);
374 	if ((flags & PRVM_CHECK_DEBUG) != 0)
375 		return (p_candebug(td, p));
376 	if ((flags & PRVM_CHECK_VISIBILITY) != 0)
377 		return (p_cansee(td, p));
378 	return (0);
379 }
380 
381 int
382 proc_vmspace_ref(struct thread *td, struct proc *p, int flags,
383     struct vmspace **vmp)
384 {
385 	struct vmspace *vm;
386 	int error;
387 
388 	MPASS((flags & ~(PRVM_BLOCK_EXEC | PRVM_CHECK_VISIBILITY |
389 	    PRVM_CHECK_DEBUG)) == 0);
390 	MPASS((flags & (PRVM_CHECK_VISIBILITY | PRVM_CHECK_DEBUG)) !=
391 	    (PRVM_CHECK_VISIBILITY | PRVM_CHECK_DEBUG));
392 
393 	PROC_LOCK(p);
394 	if (p != td->td_proc) {
395 		PROC_ASSERT_HELD(p);
396 
397 		/*
398 		 * Make sure that the vmspace doesn't switch out from
399 		 * under us.
400 		 */
401 		if ((flags & PRVM_BLOCK_EXEC) != 0) {
402 			for (;;) {
403 				if (!execve_block(td, p)) {
404 					PROC_LOCK(p);
405 					continue;
406 				}
407 				error = proc_vmspace_check_access(td, p, flags);
408 				if (error != 0) {
409 					execve_unblock(td, p);
410 					PROC_UNLOCK(p);
411 					return (error);
412 				}
413 				break;
414 			}
415 		} else {
416 			error = proc_vmspace_check_access(td, p, flags);
417 			if (error != 0) {
418 				PROC_UNLOCK(p);
419 				return (error);
420 			}
421 		}
422 	}
423 	vm = vmspace_acquire_ref(p);
424 	if (vm == NULL) {
425 		if (p != td->td_proc && (flags & PRVM_BLOCK_EXEC) != 0)
426 			execve_unblock(td, p);
427 		PROC_UNLOCK(p);
428 		return (ESRCH);
429 	}
430 	PROC_UNLOCK(p);
431 	*vmp = vm;
432 	return (0);
433 }
434 
435 void
436 proc_vmspace_unref(struct thread *td, struct proc *p, int flags,
437     struct vmspace *vm)
438 {
439 	vmspace_free(vm);
440 	if (p != td->td_proc && (flags & PRVM_BLOCK_EXEC) != 0) {
441 		PROC_LOCK(p);
442 		PROC_ASSERT_HELD(p);
443 		execve_unblock(td, p);
444 		PROC_UNLOCK(p);
445 	}
446 }
447 
448 static int
449 vmspace_rwmem(struct vmspace *vm, struct uio *uio)
450 {
451 	vm_map_t map;
452 	vm_offset_t pageno;		/* page number */
453 	vm_prot_t reqprot;
454 	int error, fault_flags, page_offset, writing;
455 
456 	map = &vm->vm_map;
457 
458 	/*
459 	 * If we are writing, then we request vm_fault() to create a private
460 	 * copy of each page.  Since these copies will not be writeable by the
461 	 * process, we must explicitly request that they be dirtied.
462 	 */
463 	writing = uio->uio_rw == UIO_WRITE;
464 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
465 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
466 
467 	if (writing) {
468 		error = priv_check(curthread, PRIV_PROC_MEM_WRITE);
469 		if (error != 0)
470 			goto out;
471 	}
472 
473 	/*
474 	 * Only map in one page at a time.  We don't have to, but it
475 	 * makes things easier.  This way is trivial - right?
476 	 */
477 	do {
478 		vm_offset_t uva;
479 		u_int len;
480 		vm_page_t m;
481 
482 		uva = (vm_offset_t)uio->uio_offset;
483 
484 		/*
485 		 * Get the page number of this segment.
486 		 */
487 		pageno = trunc_page(uva);
488 		page_offset = uva - pageno;
489 
490 		/*
491 		 * How many bytes to copy
492 		 */
493 		len = MIN(PAGE_SIZE - page_offset, uio->uio_resid);
494 
495 		/*
496 		 * Fault and hold the page on behalf of the process.
497 		 */
498 		error = vm_fault(map, pageno, reqprot, fault_flags, &m);
499 		if (error != KERN_SUCCESS) {
500 			if (error == KERN_RESOURCE_SHORTAGE)
501 				error = ENOMEM;
502 			else
503 				error = EFAULT;
504 			break;
505 		}
506 
507 		/*
508 		 * Now do the i/o move.
509 		 */
510 		error = uiomove_fromphys(&m, page_offset, len, uio);
511 
512 		/* Make the I-cache coherent for breakpoints. */
513 		if (writing && error == 0) {
514 			vm_map_lock_read(map);
515 			if (vm_map_check_protection(map, pageno, pageno +
516 			    PAGE_SIZE, VM_PROT_EXECUTE))
517 				vm_sync_icache(map, uva, len);
518 			vm_map_unlock_read(map);
519 		}
520 
521 		/*
522 		 * Release the page.
523 		 */
524 		vm_page_unwire(m, PQ_ACTIVE);
525 
526 	} while (error == 0 && uio->uio_resid > 0);
527 
528 out:
529 	return (error);
530 }
531 
532 int
533 proc_rwmem(struct proc *p, struct uio *uio, int flags)
534 {
535 	struct vmspace *vm;
536 	struct thread *td;
537 	int error;
538 
539 	td = curthread;
540 	error = proc_vmspace_ref(td, p, flags, &vm);
541 	if (error != 0)
542 		return (error);
543 	error = vmspace_rwmem(vm, uio);
544 	proc_vmspace_unref(td, p, flags, vm);
545 	return (error);
546 }
547 
548 ssize_t
549 vmspace_iop(struct thread *td, struct vmspace *vm, vm_offset_t va, void *buf,
550     size_t len, enum uio_rw rw)
551 {
552 	struct iovec iov;
553 	struct uio uio;
554 	ssize_t slen;
555 	int error;
556 
557 	MPASS(len < SSIZE_MAX);
558 	slen = (ssize_t)len;
559 
560 	iov.iov_base = (caddr_t)buf;
561 	iov.iov_len = len;
562 	uio.uio_iov = &iov;
563 	uio.uio_iovcnt = 1;
564 	uio.uio_offset = va;
565 	uio.uio_resid = slen;
566 	uio.uio_segflg = UIO_SYSSPACE;
567 	uio.uio_rw = rw;
568 	uio.uio_td = td;
569 	error = vmspace_rwmem(vm, &uio);
570 	if (error != 0 || uio.uio_resid == slen)
571 		return (-1);
572 	return (slen - uio.uio_resid);
573 }
574 
575 ssize_t
576 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
577     size_t len)
578 {
579 
580 	return (vmspace_iop(td, p->p_vmspace, va, buf, len, UIO_READ));
581 }
582 
583 ssize_t
584 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
585     size_t len)
586 {
587 
588 	return (vmspace_iop(td, p->p_vmspace, va, buf, len, UIO_WRITE));
589 }
590 
591 static int
592 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
593 {
594 	struct vattr vattr;
595 	vm_map_t map;
596 	vm_map_entry_t entry;
597 	vm_object_t obj, tobj, lobj;
598 	struct vmspace *vm;
599 	struct vnode *vp;
600 	char *freepath, *fullpath;
601 	u_int pathlen;
602 	int error, index;
603 
604 	error = 0;
605 	obj = NULL;
606 
607 	vm = vmspace_acquire_ref(p);
608 	map = &vm->vm_map;
609 	vm_map_lock_read(map);
610 
611 	do {
612 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
613 		    ("Submap in map header"));
614 		index = 0;
615 		VM_MAP_ENTRY_FOREACH(entry, map) {
616 			if (index >= pve->pve_entry &&
617 			    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
618 				break;
619 			index++;
620 		}
621 		if (index < pve->pve_entry) {
622 			error = EINVAL;
623 			break;
624 		}
625 		if (entry == &map->header) {
626 			error = ENOENT;
627 			break;
628 		}
629 
630 		/* We got an entry. */
631 		pve->pve_entry = index + 1;
632 		pve->pve_timestamp = map->timestamp;
633 		pve->pve_start = entry->start;
634 		pve->pve_end = entry->end - 1;
635 		pve->pve_offset = entry->offset;
636 		pve->pve_prot = entry->protection |
637 		    PROT_MAX(entry->max_protection);
638 
639 		/* Backing object's path needed? */
640 		if (pve->pve_pathlen == 0)
641 			break;
642 
643 		pathlen = pve->pve_pathlen;
644 		pve->pve_pathlen = 0;
645 
646 		obj = entry->object.vm_object;
647 		if (obj != NULL)
648 			VM_OBJECT_RLOCK(obj);
649 	} while (0);
650 
651 	vm_map_unlock_read(map);
652 
653 	pve->pve_fsid = VNOVAL;
654 	pve->pve_fileid = VNOVAL;
655 
656 	if (error == 0 && obj != NULL) {
657 		lobj = obj;
658 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
659 			if (tobj != obj)
660 				VM_OBJECT_RLOCK(tobj);
661 			if (lobj != obj)
662 				VM_OBJECT_RUNLOCK(lobj);
663 			lobj = tobj;
664 			pve->pve_offset += tobj->backing_object_offset;
665 		}
666 		vp = vm_object_vnode(lobj);
667 		if (vp != NULL)
668 			vref(vp);
669 		if (lobj != obj)
670 			VM_OBJECT_RUNLOCK(lobj);
671 		VM_OBJECT_RUNLOCK(obj);
672 
673 		if (vp != NULL) {
674 			freepath = NULL;
675 			fullpath = NULL;
676 			vn_fullpath(vp, &fullpath, &freepath);
677 			vn_lock(vp, LK_SHARED | LK_RETRY);
678 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
679 				pve->pve_fileid = vattr.va_fileid;
680 				pve->pve_fsid = vattr.va_fsid;
681 			}
682 			vput(vp);
683 
684 			if (fullpath != NULL) {
685 				pve->pve_pathlen = strlen(fullpath) + 1;
686 				if (pve->pve_pathlen <= pathlen) {
687 					error = copyout(fullpath, pve->pve_path,
688 					    pve->pve_pathlen);
689 				} else
690 					error = ENAMETOOLONG;
691 			}
692 			if (freepath != NULL)
693 				free(freepath, M_TEMP);
694 		}
695 	}
696 	vmspace_free(vm);
697 	if (error == 0)
698 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
699 		    p->p_pid, pve->pve_entry, pve->pve_start);
700 
701 	return (error);
702 }
703 
704 /*
705  * Process debugging system call.
706  */
707 #ifndef _SYS_SYSPROTO_H_
708 struct ptrace_args {
709 	int	req;
710 	pid_t	pid;
711 	caddr_t	addr;
712 	int	data;
713 };
714 #endif
715 
716 int
717 sys_ptrace(struct thread *td, struct ptrace_args *uap)
718 {
719 	/*
720 	 * XXX this obfuscation is to reduce stack usage, but the register
721 	 * structs may be too large to put on the stack anyway.
722 	 */
723 	union {
724 		struct ptrace_io_desc piod;
725 		struct ptrace_lwpinfo pl;
726 		struct ptrace_vm_entry pve;
727 		struct ptrace_coredump pc;
728 		struct ptrace_sc_remote sr;
729 		struct dbreg dbreg;
730 		struct fpreg fpreg;
731 		struct reg reg;
732 		struct iovec vec;
733 		syscallarg_t args[nitems(td->td_sa.args)];
734 		struct ptrace_sc_ret psr;
735 		int ptevents;
736 	} r;
737 	syscallarg_t pscr_args[nitems(td->td_sa.args)];
738 	void *addr;
739 	int error;
740 
741 	if (!allow_ptrace)
742 		return (ENOSYS);
743 	error = 0;
744 
745 	AUDIT_ARG_PID(uap->pid);
746 	AUDIT_ARG_CMD(uap->req);
747 	AUDIT_ARG_VALUE(uap->data);
748 	addr = &r;
749 	switch (uap->req) {
750 	case PT_GET_EVENT_MASK:
751 	case PT_LWPINFO:
752 	case PT_GET_SC_ARGS:
753 	case PT_GET_SC_RET:
754 		break;
755 	case PT_SET_SC_RET:
756 		if (uap->data != sizeof(r.psr))
757 			error = EINVAL;
758 		else
759 			error = copyin(uap->addr, &r.psr, sizeof(r.psr));
760 		break;
761 	case PT_GETREGS:
762 		bzero(&r.reg, sizeof(r.reg));
763 		break;
764 	case PT_GETFPREGS:
765 		bzero(&r.fpreg, sizeof(r.fpreg));
766 		break;
767 	case PT_GETDBREGS:
768 		bzero(&r.dbreg, sizeof(r.dbreg));
769 		break;
770 	case PT_GETREGSET:
771 	case PT_SETREGSET:
772 		error = copyin(uap->addr, &r.vec, sizeof(r.vec));
773 		break;
774 	case PT_SETREGS:
775 		error = copyin(uap->addr, &r.reg, sizeof(r.reg));
776 		break;
777 	case PT_SETFPREGS:
778 		error = copyin(uap->addr, &r.fpreg, sizeof(r.fpreg));
779 		break;
780 	case PT_SETDBREGS:
781 		error = copyin(uap->addr, &r.dbreg, sizeof(r.dbreg));
782 		break;
783 	case PT_SET_EVENT_MASK:
784 		if (uap->data != sizeof(r.ptevents))
785 			error = EINVAL;
786 		else
787 			error = copyin(uap->addr, &r.ptevents, uap->data);
788 		break;
789 	case PT_IO:
790 		error = copyin(uap->addr, &r.piod, sizeof(r.piod));
791 		break;
792 	case PT_VM_ENTRY:
793 		error = copyin(uap->addr, &r.pve, sizeof(r.pve));
794 		break;
795 	case PT_COREDUMP:
796 		if (uap->data != sizeof(r.pc))
797 			error = EINVAL;
798 		else
799 			error = copyin(uap->addr, &r.pc, uap->data);
800 		break;
801 	case PT_SC_REMOTE:
802 		if (uap->data != sizeof(r.sr)) {
803 			error = EINVAL;
804 			break;
805 		}
806 		error = copyin(uap->addr, &r.sr, uap->data);
807 		if (error != 0)
808 			break;
809 		if (r.sr.pscr_nargs > nitems(td->td_sa.args)) {
810 			error = EINVAL;
811 			break;
812 		}
813 		error = copyin(r.sr.pscr_args, pscr_args,
814 		    sizeof(u_long) * r.sr.pscr_nargs);
815 		if (error != 0)
816 			break;
817 		r.sr.pscr_args = pscr_args;
818 		break;
819 	case PTINTERNAL_FIRST ... PTINTERNAL_LAST:
820 		error = EINVAL;
821 		break;
822 	default:
823 		addr = uap->addr;
824 		break;
825 	}
826 	if (error != 0)
827 		return (error);
828 
829 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
830 	if (error != 0)
831 		return (error);
832 
833 	switch (uap->req) {
834 	case PT_VM_ENTRY:
835 		error = copyout(&r.pve, uap->addr, sizeof(r.pve));
836 		break;
837 	case PT_IO:
838 		error = copyout(&r.piod, uap->addr, sizeof(r.piod));
839 		break;
840 	case PT_GETREGS:
841 		error = copyout(&r.reg, uap->addr, sizeof(r.reg));
842 		break;
843 	case PT_GETFPREGS:
844 		error = copyout(&r.fpreg, uap->addr, sizeof(r.fpreg));
845 		break;
846 	case PT_GETDBREGS:
847 		error = copyout(&r.dbreg, uap->addr, sizeof(r.dbreg));
848 		break;
849 	case PT_GETREGSET:
850 		error = copyout(&r.vec, uap->addr, sizeof(r.vec));
851 		break;
852 	case PT_GET_EVENT_MASK:
853 		/* NB: The size in uap->data is validated in kern_ptrace(). */
854 		error = copyout(&r.ptevents, uap->addr, uap->data);
855 		break;
856 	case PT_LWPINFO:
857 		/* NB: The size in uap->data is validated in kern_ptrace(). */
858 		error = copyout(&r.pl, uap->addr, uap->data);
859 		break;
860 	case PT_GET_SC_ARGS:
861 		error = copyout(r.args, uap->addr, MIN(uap->data,
862 		    sizeof(r.args)));
863 		break;
864 	case PT_GET_SC_RET:
865 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
866 		    sizeof(r.psr)));
867 		break;
868 	case PT_SC_REMOTE:
869 		error = copyout(&r.sr.pscr_ret, uap->addr +
870 		    offsetof(struct ptrace_sc_remote, pscr_ret),
871 		    sizeof(r.sr.pscr_ret));
872 		break;
873 	}
874 
875 	return (error);
876 }
877 
878 #ifdef COMPAT_FREEBSD32
879 /*
880  *   PROC_READ(regs, td2, addr);
881  * becomes either:
882  *   proc_read_regs(td2, addr);
883  * or
884  *   proc_read_regs32(td2, addr);
885  * .. except this is done at runtime.  There is an additional
886  * complication in that PROC_WRITE disallows 32 bit consumers
887  * from writing to 64 bit address space targets.
888  */
889 #define	PROC_READ(w, t, a)	wrap32 ? \
890 	proc_read_ ## w ## 32(t, a) : \
891 	proc_read_ ## w (t, a)
892 #define	PROC_WRITE(w, t, a)	wrap32 ? \
893 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
894 	proc_write_ ## w (t, a)
895 #else
896 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
897 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
898 #endif
899 
900 void
901 proc_set_traced(struct proc *p, bool stop)
902 {
903 
904 	sx_assert(&proctree_lock, SX_XLOCKED);
905 	PROC_LOCK_ASSERT(p, MA_OWNED);
906 	p->p_flag |= P_TRACED;
907 	if (stop)
908 		p->p_flag2 |= P2_PTRACE_FSTP;
909 	p->p_ptevents = PTRACE_DEFAULT;
910 }
911 
912 void
913 ptrace_unsuspend(struct proc *p)
914 {
915 	PROC_LOCK_ASSERT(p, MA_OWNED);
916 
917 	PROC_SLOCK(p);
918 	p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
919 	thread_unsuspend(p);
920 	PROC_SUNLOCK(p);
921 	itimer_proc_continue(p);
922 	kqtimer_proc_continue(p);
923 }
924 
925 static int
926 proc_can_ptrace(struct thread *td, struct proc *p)
927 {
928 	int error;
929 
930 	PROC_LOCK_ASSERT(p, MA_OWNED);
931 
932 	if ((p->p_flag & P_WEXIT) != 0)
933 		return (ESRCH);
934 
935 	if ((error = p_cansee(td, p)) != 0)
936 		return (error);
937 	if ((error = p_candebug(td, p)) != 0)
938 		return (error);
939 
940 	/* not being traced... */
941 	if ((p->p_flag & P_TRACED) == 0)
942 		return (EPERM);
943 
944 	/* not being traced by YOU */
945 	if (p->p_pptr != td->td_proc)
946 		return (EBUSY);
947 
948 	/* not currently stopped */
949 	if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
950 	    p->p_suspcount != p->p_numthreads  ||
951 	    (p->p_flag & P_WAITED) == 0)
952 		return (EBUSY);
953 
954 	return (0);
955 }
956 
957 static struct thread *
958 ptrace_sel_coredump_thread(struct proc *p)
959 {
960 	struct thread *td2;
961 
962 	PROC_LOCK_ASSERT(p, MA_OWNED);
963 	MPASS((p->p_flag & P_STOPPED_TRACE) != 0);
964 
965 	FOREACH_THREAD_IN_PROC(p, td2) {
966 		if ((td2->td_dbgflags & TDB_SSWITCH) != 0)
967 			return (td2);
968 	}
969 	return (NULL);
970 }
971 
972 int
973 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
974 {
975 	struct iovec iov;
976 	struct uio uio;
977 	struct proc *curp, *p, *pp;
978 	struct thread *td2 = NULL, *td3;
979 	struct ptrace_io_desc *piod = NULL;
980 	struct ptrace_lwpinfo *pl;
981 	struct ptrace_sc_ret *psr;
982 	struct ptrace_sc_remote *pscr;
983 	struct file *fp;
984 	struct ptrace_coredump *pc;
985 	struct thr_coredump_req *tcq;
986 	struct thr_syscall_req *tsr;
987 	int error, num, tmp;
988 	lwpid_t tid = 0, *buf;
989 #ifdef COMPAT_FREEBSD32
990 	int wrap32 = 0, safe = 0;
991 #endif
992 	bool proctree_locked, p2_req_set;
993 
994 	curp = td->td_proc;
995 	proctree_locked = false;
996 	p2_req_set = false;
997 
998 	/* Lock proctree before locking the process. */
999 	switch (req) {
1000 	case PT_TRACE_ME:
1001 	case PT_ATTACH:
1002 	case PT_STEP:
1003 	case PT_CONTINUE:
1004 	case PT_TO_SCE:
1005 	case PT_TO_SCX:
1006 	case PT_SYSCALL:
1007 	case PT_FOLLOW_FORK:
1008 	case PT_LWP_EVENTS:
1009 	case PT_GET_EVENT_MASK:
1010 	case PT_SET_EVENT_MASK:
1011 	case PT_DETACH:
1012 	case PT_GET_SC_ARGS:
1013 		sx_xlock(&proctree_lock);
1014 		proctree_locked = true;
1015 		break;
1016 	default:
1017 		break;
1018 	}
1019 
1020 	if (req == PT_TRACE_ME) {
1021 		p = td->td_proc;
1022 		PROC_LOCK(p);
1023 	} else {
1024 		if (pid <= PID_MAX) {
1025 			if ((p = pfind(pid)) == NULL) {
1026 				if (proctree_locked)
1027 					sx_xunlock(&proctree_lock);
1028 				return (ESRCH);
1029 			}
1030 		} else {
1031 			td2 = tdfind(pid, -1);
1032 			if (td2 == NULL) {
1033 				if (proctree_locked)
1034 					sx_xunlock(&proctree_lock);
1035 				return (ESRCH);
1036 			}
1037 			p = td2->td_proc;
1038 			tid = pid;
1039 			pid = p->p_pid;
1040 		}
1041 	}
1042 	AUDIT_ARG_PROCESS(p);
1043 
1044 	if ((p->p_flag & P_WEXIT) != 0) {
1045 		error = ESRCH;
1046 		goto fail;
1047 	}
1048 	if ((error = p_cansee(td, p)) != 0)
1049 		goto fail;
1050 
1051 	if ((error = p_candebug(td, p)) != 0)
1052 		goto fail;
1053 
1054 	/*
1055 	 * System processes can't be debugged.
1056 	 */
1057 	if ((p->p_flag & P_SYSTEM) != 0) {
1058 		error = EINVAL;
1059 		goto fail;
1060 	}
1061 
1062 	if (tid == 0) {
1063 		if ((p->p_flag & P_STOPPED_TRACE) != 0)
1064 			td2 = p->p_xthread;
1065 		if (td2 == NULL)
1066 			td2 = FIRST_THREAD_IN_PROC(p);
1067 		tid = td2->td_tid;
1068 	}
1069 
1070 #ifdef COMPAT_FREEBSD32
1071 	/*
1072 	 * Test if we're a 32 bit client and what the target is.
1073 	 * Set the wrap controls accordingly.
1074 	 */
1075 	if (SV_CURPROC_FLAG(SV_ILP32)) {
1076 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
1077 			safe = 1;
1078 		wrap32 = 1;
1079 	}
1080 #endif
1081 	/*
1082 	 * Permissions check
1083 	 */
1084 	switch (req) {
1085 	case PT_TRACE_ME:
1086 		/*
1087 		 * Always legal, when there is a parent process which
1088 		 * could trace us.  Otherwise, reject.
1089 		 */
1090 		if ((p->p_flag & P_TRACED) != 0) {
1091 			error = EBUSY;
1092 			goto fail;
1093 		}
1094 		if (p->p_pptr == initproc) {
1095 			error = EPERM;
1096 			goto fail;
1097 		}
1098 		break;
1099 
1100 	case PT_ATTACH:
1101 		/* Self */
1102 		if (p == td->td_proc) {
1103 			error = EINVAL;
1104 			goto fail;
1105 		}
1106 
1107 		/* Already traced */
1108 		if (p->p_flag & P_TRACED) {
1109 			error = EBUSY;
1110 			goto fail;
1111 		}
1112 
1113 		/* Can't trace an ancestor if you're being traced. */
1114 		if (curp->p_flag & P_TRACED) {
1115 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
1116 				if (pp == p) {
1117 					error = EINVAL;
1118 					goto fail;
1119 				}
1120 			}
1121 		}
1122 
1123 		/* OK */
1124 		break;
1125 
1126 	case PT_CLEARSTEP:
1127 		/* Allow thread to clear single step for itself */
1128 		if (td->td_tid == tid)
1129 			break;
1130 
1131 		/* FALLTHROUGH */
1132 	default:
1133 		/*
1134 		 * Check for ptrace eligibility before waiting for
1135 		 * holds to drain.
1136 		 */
1137 		error = proc_can_ptrace(td, p);
1138 		if (error != 0)
1139 			goto fail;
1140 
1141 		/*
1142 		 * Block parallel ptrace requests.  Most important, do
1143 		 * not allow other thread in debugger to continue the
1144 		 * debuggee until coredump finished.
1145 		 */
1146 		while ((p->p_flag2 & P2_PTRACEREQ) != 0) {
1147 			if (proctree_locked)
1148 				sx_xunlock(&proctree_lock);
1149 			error = msleep(&p->p_flag2, &p->p_mtx, PPAUSE | PCATCH |
1150 			    (proctree_locked ? PDROP : 0), "pptrace", 0);
1151 			if (proctree_locked) {
1152 				sx_xlock(&proctree_lock);
1153 				PROC_LOCK(p);
1154 			}
1155 			if (error == 0 && td2->td_proc != p)
1156 				error = ESRCH;
1157 			if (error == 0)
1158 				error = proc_can_ptrace(td, p);
1159 			if (error != 0)
1160 				goto fail;
1161 		}
1162 
1163 		/* Ok */
1164 		break;
1165 	}
1166 
1167 	/*
1168 	 * Keep this process around and request parallel ptrace()
1169 	 * request to wait until we finish this request.
1170 	 */
1171 	MPASS((p->p_flag2 & P2_PTRACEREQ) == 0);
1172 	p->p_flag2 |= P2_PTRACEREQ;
1173 	p2_req_set = true;
1174 	_PHOLD(p);
1175 
1176 	/*
1177 	 * Actually do the requests
1178 	 */
1179 
1180 	td->td_retval[0] = 0;
1181 
1182 	switch (req) {
1183 	case PT_TRACE_ME:
1184 		/* set my trace flag and "owner" so it can read/write me */
1185 		proc_set_traced(p, false);
1186 		if (p->p_flag & P_PPWAIT)
1187 			p->p_flag |= P_PPTRACE;
1188 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
1189 		break;
1190 
1191 	case PT_ATTACH:
1192 		/* security check done above */
1193 		/*
1194 		 * It would be nice if the tracing relationship was separate
1195 		 * from the parent relationship but that would require
1196 		 * another set of links in the proc struct or for "wait"
1197 		 * to scan the entire proc table.  To make life easier,
1198 		 * we just re-parent the process we're trying to trace.
1199 		 * The old parent is remembered so we can put things back
1200 		 * on a "detach".
1201 		 */
1202 		proc_set_traced(p, true);
1203 		proc_reparent(p, td->td_proc, false);
1204 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
1205 		    p->p_oppid);
1206 
1207 		sx_xunlock(&proctree_lock);
1208 		proctree_locked = false;
1209 		MPASS(p->p_xthread == NULL);
1210 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
1211 
1212 		/*
1213 		 * If already stopped due to a stop signal, clear the
1214 		 * existing stop before triggering a traced SIGSTOP.
1215 		 */
1216 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
1217 			PROC_SLOCK(p);
1218 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
1219 			thread_unsuspend(p);
1220 			PROC_SUNLOCK(p);
1221 		}
1222 
1223 		kern_psignal(p, SIGSTOP);
1224 		break;
1225 
1226 	case PT_CLEARSTEP:
1227 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
1228 		    p->p_pid);
1229 		error = ptrace_clear_single_step(td2);
1230 		break;
1231 
1232 	case PT_SETSTEP:
1233 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
1234 		    p->p_pid);
1235 		error = ptrace_single_step(td2);
1236 		break;
1237 
1238 	case PT_SUSPEND:
1239 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
1240 		    p->p_pid);
1241 		td2->td_dbgflags |= TDB_SUSPEND;
1242 		ast_sched(td2, TDA_SUSPEND);
1243 		break;
1244 
1245 	case PT_RESUME:
1246 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1247 		    p->p_pid);
1248 		td2->td_dbgflags &= ~TDB_SUSPEND;
1249 		break;
1250 
1251 	case PT_FOLLOW_FORK:
1252 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1253 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1254 		    data ? "enabled" : "disabled");
1255 		if (data)
1256 			p->p_ptevents |= PTRACE_FORK;
1257 		else
1258 			p->p_ptevents &= ~PTRACE_FORK;
1259 		break;
1260 
1261 	case PT_LWP_EVENTS:
1262 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1263 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1264 		    data ? "enabled" : "disabled");
1265 		if (data)
1266 			p->p_ptevents |= PTRACE_LWP;
1267 		else
1268 			p->p_ptevents &= ~PTRACE_LWP;
1269 		break;
1270 
1271 	case PT_GET_EVENT_MASK:
1272 		if (data != sizeof(p->p_ptevents)) {
1273 			error = EINVAL;
1274 			break;
1275 		}
1276 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1277 		    p->p_ptevents);
1278 		*(int *)addr = p->p_ptevents;
1279 		break;
1280 
1281 	case PT_SET_EVENT_MASK:
1282 		if (data != sizeof(p->p_ptevents)) {
1283 			error = EINVAL;
1284 			break;
1285 		}
1286 		tmp = *(int *)addr;
1287 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1288 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1289 			error = EINVAL;
1290 			break;
1291 		}
1292 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1293 		    p->p_pid, p->p_ptevents, tmp);
1294 		p->p_ptevents = tmp;
1295 		break;
1296 
1297 	case PT_GET_SC_ARGS:
1298 	case PTLINUX_GET_SC_ARGS:
1299 		CTR2(KTR_PTRACE, "%s: pid %d", req == PT_GET_SC_ARGS ?
1300 		    "PT_GET_SC_ARGS" : "PT_LINUX_GET_SC_ARGS", p->p_pid);
1301 		if (((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0 &&
1302 		     td2->td_sa.code == 0)
1303 #ifdef COMPAT_FREEBSD32
1304 		    || (wrap32 && !safe)
1305 #endif
1306 		    ) {
1307 			error = EINVAL;
1308 			break;
1309 		}
1310 		if (req == PT_GET_SC_ARGS) {
1311 			bzero(addr, sizeof(td2->td_sa.args));
1312 			bcopy(td2->td_sa.args, addr, td2->td_sa.callp->sy_narg *
1313 			    sizeof(syscallarg_t));
1314 		} else {
1315 			/*
1316 			 * Emulate a Linux bug which which strace(1) depends on:
1317 			 * at initialization it tests whether ptrace works by
1318 			 * calling close(2), or some other single-argument
1319 			 * syscall, _with six arguments_, and then verifies
1320 			 * whether it can fetch them all using this API;
1321 			 * otherwise it bails out.
1322 			 */
1323 			bcopy(td2->td_sa.args, addr, 6 * sizeof(syscallarg_t));
1324 		}
1325 		break;
1326 
1327 	case PT_GET_SC_RET:
1328 		if ((td2->td_dbgflags & TDB_SCX) == 0
1329 #ifdef COMPAT_FREEBSD32
1330 		    || (wrap32 && !safe)
1331 #endif
1332 		    ) {
1333 			error = EINVAL;
1334 			break;
1335 		}
1336 		psr = addr;
1337 		bzero(psr, sizeof(*psr));
1338 		psr->sr_error = td2->td_errno;
1339 		if (psr->sr_error == 0) {
1340 			psr->sr_retval[0] = td2->td_retval[0];
1341 			psr->sr_retval[1] = td2->td_retval[1];
1342 		}
1343 		CTR4(KTR_PTRACE,
1344 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1345 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1346 		    psr->sr_retval[1]);
1347 		break;
1348 
1349 	case PT_SET_SC_RET:
1350 		if ((td2->td_dbgflags & TDB_SCE) == 0
1351 #ifdef COMPAT_FREEBSD32
1352 		    || (wrap32 && !safe)
1353 #endif
1354 		    ) {
1355 			error = EINVAL;
1356 			break;
1357 		}
1358 		psr = addr;
1359 		td2->td_errno = psr->sr_error;
1360 		if (td2->td_errno == 0) {
1361 			td2->td_retval[0] = psr->sr_retval[0];
1362 			td2->td_retval[1] = psr->sr_retval[1];
1363 		}
1364 		td2->td_dbgflags |= TDB_SET_SC_RET;
1365 		break;
1366 
1367 	case PT_STEP:
1368 	case PT_CONTINUE:
1369 	case PT_TO_SCE:
1370 	case PT_TO_SCX:
1371 	case PT_SYSCALL:
1372 	case PT_DETACH:
1373 		/* Zero means do not send any signal */
1374 		if (data < 0 || data > _SIG_MAXSIG) {
1375 			error = EINVAL;
1376 			break;
1377 		}
1378 
1379 		switch (req) {
1380 		case PT_STEP:
1381 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1382 			    td2->td_tid, p->p_pid, data);
1383 			error = ptrace_single_step(td2);
1384 			if (error != 0)
1385 				goto out;
1386 			break;
1387 		case PT_CONTINUE:
1388 		case PT_TO_SCE:
1389 		case PT_TO_SCX:
1390 		case PT_SYSCALL:
1391 			if (addr != (void *)1) {
1392 				error = ptrace_set_pc(td2,
1393 				    (u_long)(uintfptr_t)addr);
1394 				if (error != 0)
1395 					goto out;
1396 				td2->td_dbgflags |= TDB_USERWR;
1397 			}
1398 			switch (req) {
1399 			case PT_TO_SCE:
1400 				p->p_ptevents |= PTRACE_SCE;
1401 				CTR4(KTR_PTRACE,
1402 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1403 				    p->p_pid, p->p_ptevents,
1404 				    (u_long)(uintfptr_t)addr, data);
1405 				break;
1406 			case PT_TO_SCX:
1407 				p->p_ptevents |= PTRACE_SCX;
1408 				CTR4(KTR_PTRACE,
1409 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1410 				    p->p_pid, p->p_ptevents,
1411 				    (u_long)(uintfptr_t)addr, data);
1412 				break;
1413 			case PT_SYSCALL:
1414 				p->p_ptevents |= PTRACE_SYSCALL;
1415 				CTR4(KTR_PTRACE,
1416 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1417 				    p->p_pid, p->p_ptevents,
1418 				    (u_long)(uintfptr_t)addr, data);
1419 				break;
1420 			case PT_CONTINUE:
1421 				CTR3(KTR_PTRACE,
1422 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1423 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1424 				break;
1425 			}
1426 			break;
1427 		case PT_DETACH:
1428 			/*
1429 			 * Clear P_TRACED before reparenting
1430 			 * a detached process back to its original
1431 			 * parent.  Otherwise the debugee will be set
1432 			 * as an orphan of the debugger.
1433 			 */
1434 			p->p_flag &= ~(P_TRACED | P_WAITED);
1435 
1436 			/*
1437 			 * Reset the process parent.
1438 			 */
1439 			if (p->p_oppid != p->p_pptr->p_pid) {
1440 				PROC_LOCK(p->p_pptr);
1441 				sigqueue_take(p->p_ksi);
1442 				PROC_UNLOCK(p->p_pptr);
1443 
1444 				pp = proc_realparent(p);
1445 				proc_reparent(p, pp, false);
1446 				if (pp == initproc)
1447 					p->p_sigparent = SIGCHLD;
1448 				CTR3(KTR_PTRACE,
1449 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1450 				    p->p_pid, pp->p_pid, data);
1451 			} else {
1452 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1453 				    p->p_pid, data);
1454 			}
1455 
1456 			p->p_ptevents = 0;
1457 			FOREACH_THREAD_IN_PROC(p, td3) {
1458 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1459 					sigqueue_delete(&td3->td_sigqueue,
1460 					    SIGSTOP);
1461 				}
1462 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1463 				    TDB_SUSPEND | TDB_BORN);
1464 			}
1465 
1466 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1467 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1468 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1469 			}
1470 
1471 			/*
1472 			 * Send SIGCHLD and wakeup the parent as needed.  It
1473 			 * may be the case that they had stopped the child
1474 			 * before it got ptraced, and now they're in the middle
1475 			 * of a wait(2) for it to continue.
1476 			 */
1477 			PROC_LOCK(p->p_pptr);
1478 			childproc_continued(p);
1479 			PROC_UNLOCK(p->p_pptr);
1480 			break;
1481 		}
1482 
1483 		sx_xunlock(&proctree_lock);
1484 		proctree_locked = false;
1485 
1486 	sendsig:
1487 		MPASS(!proctree_locked);
1488 
1489 		/*
1490 		 * Clear the pending event for the thread that just
1491 		 * reported its event (p_xthread), if any.  This may
1492 		 * not be the thread passed to PT_CONTINUE, PT_STEP,
1493 		 * etc. if the debugger is resuming a different
1494 		 * thread.  There might be no reporting thread if
1495 		 * the process was just attached.
1496 		 *
1497 		 * Deliver any pending signal via the reporting thread.
1498 		 */
1499 		if (p->p_xthread != NULL) {
1500 			p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1501 			p->p_xthread->td_xsig = data;
1502 			p->p_xthread = NULL;
1503 		}
1504 		p->p_xsig = data;
1505 
1506 		/*
1507 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1508 		 * always works immediately, even if another thread is
1509 		 * unsuspended first and attempts to handle a
1510 		 * different signal or if the POSIX.1b style signal
1511 		 * queue cannot accommodate any new signals.
1512 		 */
1513 		if (data == SIGKILL)
1514 			proc_wkilled(p);
1515 
1516 		/*
1517 		 * If the PT_CONTINUE-like operation is attempted on
1518 		 * the thread on sleepq, this is possible only after
1519 		 * the transparent PT_ATTACH.  In this case, if the
1520 		 * caller modified the thread state, e.g. by writing
1521 		 * register file or specifying the pc, make the thread
1522 		 * xstopped by waking it up.
1523 		 */
1524 		if ((td2->td_dbgflags & TDB_USERWR) != 0 &&
1525 		    pt_attach_transparent) {
1526 			thread_lock(td2);
1527 			if (TD_ON_SLEEPQ(td2) &&
1528 			    (td2->td_flags & TDF_SINTR) != 0) {
1529 				td2->td_dbgflags &= ~TDB_USERWR;
1530 				sleepq_abort(td2, EINTR);
1531 			} else {
1532 				thread_unlock(td2);
1533 			}
1534 		}
1535 
1536 		/*
1537 		 * Unsuspend all threads.  To leave a thread
1538 		 * suspended, use PT_SUSPEND to suspend it before
1539 		 * continuing the process.
1540 		 */
1541 		ptrace_unsuspend(p);
1542 		break;
1543 
1544 	case PT_WRITE_I:
1545 	case PT_WRITE_D:
1546 		td2->td_dbgflags |= TDB_USERWR;
1547 		PROC_UNLOCK(p);
1548 		error = 0;
1549 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1550 		    sizeof(int)) != sizeof(int))
1551 			error = ENOMEM;
1552 		else
1553 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1554 			    p->p_pid, addr, data);
1555 		PROC_LOCK(p);
1556 		break;
1557 
1558 	case PT_READ_I:
1559 	case PT_READ_D:
1560 		PROC_UNLOCK(p);
1561 		error = tmp = 0;
1562 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1563 		    sizeof(int)) != sizeof(int))
1564 			error = ENOMEM;
1565 		else
1566 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1567 			    p->p_pid, addr, tmp);
1568 		td->td_retval[0] = tmp;
1569 		PROC_LOCK(p);
1570 		break;
1571 
1572 	case PT_IO:
1573 		piod = addr;
1574 		if (piod->piod_len > SSIZE_MAX) {
1575 			error = EINVAL;
1576 			goto out;
1577 		}
1578 		iov.iov_base = piod->piod_addr;
1579 		iov.iov_len = piod->piod_len;
1580 		uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1581 		uio.uio_resid = piod->piod_len;
1582 		uio.uio_iov = &iov;
1583 		uio.uio_iovcnt = 1;
1584 		uio.uio_segflg = UIO_USERSPACE;
1585 		uio.uio_td = td;
1586 		switch (piod->piod_op) {
1587 		case PIOD_READ_D:
1588 		case PIOD_READ_I:
1589 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1590 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1591 			uio.uio_rw = UIO_READ;
1592 			break;
1593 		case PIOD_WRITE_D:
1594 		case PIOD_WRITE_I:
1595 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1596 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1597 			td2->td_dbgflags |= TDB_USERWR;
1598 			uio.uio_rw = UIO_WRITE;
1599 			break;
1600 		default:
1601 			error = EINVAL;
1602 			goto out;
1603 		}
1604 		PROC_UNLOCK(p);
1605 		error = proc_rwmem(p, &uio, 0);
1606 		piod->piod_len -= uio.uio_resid;
1607 		PROC_LOCK(p);
1608 		break;
1609 
1610 	case PT_KILL:
1611 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1612 		data = SIGKILL;
1613 		goto sendsig;	/* in PT_CONTINUE above */
1614 
1615 	case PT_SETREGS:
1616 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1617 		    p->p_pid);
1618 		td2->td_dbgflags |= TDB_USERWR;
1619 		error = PROC_WRITE(regs, td2, addr);
1620 		break;
1621 
1622 	case PT_GETREGS:
1623 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1624 		    p->p_pid);
1625 		error = PROC_READ(regs, td2, addr);
1626 		break;
1627 
1628 	case PT_SETFPREGS:
1629 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1630 		    p->p_pid);
1631 		td2->td_dbgflags |= TDB_USERWR;
1632 		error = PROC_WRITE(fpregs, td2, addr);
1633 		break;
1634 
1635 	case PT_GETFPREGS:
1636 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1637 		    p->p_pid);
1638 		error = PROC_READ(fpregs, td2, addr);
1639 		break;
1640 
1641 	case PT_SETDBREGS:
1642 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1643 		    p->p_pid);
1644 		td2->td_dbgflags |= TDB_USERWR;
1645 		error = PROC_WRITE(dbregs, td2, addr);
1646 		break;
1647 
1648 	case PT_GETDBREGS:
1649 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1650 		    p->p_pid);
1651 		error = PROC_READ(dbregs, td2, addr);
1652 		break;
1653 
1654 	case PT_SETREGSET:
1655 		CTR2(KTR_PTRACE, "PT_SETREGSET: tid %d (pid %d)", td2->td_tid,
1656 		    p->p_pid);
1657 		error = proc_write_regset(td2, data, addr);
1658 		break;
1659 
1660 	case PT_GETREGSET:
1661 		CTR2(KTR_PTRACE, "PT_GETREGSET: tid %d (pid %d)", td2->td_tid,
1662 		    p->p_pid);
1663 		error = proc_read_regset(td2, data, addr);
1664 		break;
1665 
1666 	case PT_LWPINFO:
1667 		if (data <= 0 || data > sizeof(*pl)) {
1668 			error = EINVAL;
1669 			break;
1670 		}
1671 		pl = addr;
1672 		bzero(pl, sizeof(*pl));
1673 		pl->pl_lwpid = td2->td_tid;
1674 		pl->pl_event = PL_EVENT_NONE;
1675 		pl->pl_flags = 0;
1676 		if (td2->td_dbgflags & TDB_XSIG) {
1677 			pl->pl_event = PL_EVENT_SIGNAL;
1678 			if (td2->td_si.si_signo != 0 &&
1679 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1680 			    + sizeof(pl->pl_siginfo)){
1681 				pl->pl_flags |= PL_FLAG_SI;
1682 				pl->pl_siginfo = td2->td_si;
1683 			}
1684 		}
1685 		if (td2->td_dbgflags & TDB_SCE)
1686 			pl->pl_flags |= PL_FLAG_SCE;
1687 		else if (td2->td_dbgflags & TDB_SCX)
1688 			pl->pl_flags |= PL_FLAG_SCX;
1689 		if (td2->td_dbgflags & TDB_EXEC)
1690 			pl->pl_flags |= PL_FLAG_EXEC;
1691 		if (td2->td_dbgflags & TDB_FORK) {
1692 			pl->pl_flags |= PL_FLAG_FORKED;
1693 			pl->pl_child_pid = td2->td_dbg_forked;
1694 			if (td2->td_dbgflags & TDB_VFORK)
1695 				pl->pl_flags |= PL_FLAG_VFORKED;
1696 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1697 		    TDB_VFORK)
1698 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1699 		if (td2->td_dbgflags & TDB_CHILD)
1700 			pl->pl_flags |= PL_FLAG_CHILD;
1701 		if (td2->td_dbgflags & TDB_BORN)
1702 			pl->pl_flags |= PL_FLAG_BORN;
1703 		if (td2->td_dbgflags & TDB_EXIT)
1704 			pl->pl_flags |= PL_FLAG_EXITED;
1705 		pl->pl_sigmask = td2->td_sigmask;
1706 		pl->pl_siglist = td2->td_siglist;
1707 		strcpy(pl->pl_tdname, td2->td_name);
1708 		if (td2->td_sa.code != 0) {
1709 			pl->pl_syscall_code = td2->td_sa.code;
1710 			pl->pl_syscall_narg = td2->td_sa.callp->sy_narg;
1711 		}
1712 		CTR6(KTR_PTRACE,
1713     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1714 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1715 		    pl->pl_child_pid, pl->pl_syscall_code);
1716 		break;
1717 
1718 	case PT_GETNUMLWPS:
1719 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1720 		    p->p_numthreads);
1721 		td->td_retval[0] = p->p_numthreads;
1722 		break;
1723 
1724 	case PT_GETLWPLIST:
1725 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1726 		    p->p_pid, data, p->p_numthreads);
1727 		if (data <= 0) {
1728 			error = EINVAL;
1729 			break;
1730 		}
1731 		num = imin(p->p_numthreads, data);
1732 		PROC_UNLOCK(p);
1733 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1734 		tmp = 0;
1735 		PROC_LOCK(p);
1736 		FOREACH_THREAD_IN_PROC(p, td2) {
1737 			if (tmp >= num)
1738 				break;
1739 			buf[tmp++] = td2->td_tid;
1740 		}
1741 		PROC_UNLOCK(p);
1742 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1743 		free(buf, M_TEMP);
1744 		if (!error)
1745 			td->td_retval[0] = tmp;
1746 		PROC_LOCK(p);
1747 		break;
1748 
1749 	case PT_VM_TIMESTAMP:
1750 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1751 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1752 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1753 		break;
1754 
1755 	case PT_VM_ENTRY:
1756 		PROC_UNLOCK(p);
1757 		error = ptrace_vm_entry(td, p, addr);
1758 		PROC_LOCK(p);
1759 		break;
1760 
1761 	case PT_COREDUMP:
1762 		pc = addr;
1763 		CTR2(KTR_PTRACE, "PT_COREDUMP: pid %d, fd %d",
1764 		    p->p_pid, pc->pc_fd);
1765 
1766 		if ((pc->pc_flags & ~(PC_COMPRESS | PC_ALL)) != 0) {
1767 			error = EINVAL;
1768 			break;
1769 		}
1770 		PROC_UNLOCK(p);
1771 
1772 		tcq = malloc(sizeof(*tcq), M_TEMP, M_WAITOK | M_ZERO);
1773 		fp = NULL;
1774 		error = fget_write(td, pc->pc_fd, &cap_write_rights, &fp);
1775 		if (error != 0)
1776 			goto coredump_cleanup_nofp;
1777 		if (fp->f_type != DTYPE_VNODE || fp->f_vnode->v_type != VREG) {
1778 			error = EPIPE;
1779 			goto coredump_cleanup;
1780 		}
1781 
1782 		PROC_LOCK(p);
1783 		error = proc_can_ptrace(td, p);
1784 		if (error != 0)
1785 			goto coredump_cleanup_locked;
1786 
1787 		td2 = ptrace_sel_coredump_thread(p);
1788 		if (td2 == NULL) {
1789 			error = EBUSY;
1790 			goto coredump_cleanup_locked;
1791 		}
1792 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1793 		    TDB_SCREMOTEREQ)) == 0,
1794 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1795 
1796 		tcq->tc_vp = fp->f_vnode;
1797 		tcq->tc_limit = pc->pc_limit == 0 ? OFF_MAX : pc->pc_limit;
1798 		tcq->tc_flags = SVC_PT_COREDUMP;
1799 		if ((pc->pc_flags & PC_COMPRESS) == 0)
1800 			tcq->tc_flags |= SVC_NOCOMPRESS;
1801 		if ((pc->pc_flags & PC_ALL) != 0)
1802 			tcq->tc_flags |= SVC_ALL;
1803 		td2->td_remotereq = tcq;
1804 		td2->td_dbgflags |= TDB_COREDUMPREQ;
1805 		thread_run_flash(td2);
1806 		while ((td2->td_dbgflags & TDB_COREDUMPREQ) != 0)
1807 			msleep(p, &p->p_mtx, PPAUSE, "crdmp", 0);
1808 		error = tcq->tc_error;
1809 coredump_cleanup_locked:
1810 		PROC_UNLOCK(p);
1811 coredump_cleanup:
1812 		fdrop(fp, td);
1813 coredump_cleanup_nofp:
1814 		free(tcq, M_TEMP);
1815 		PROC_LOCK(p);
1816 		break;
1817 
1818 	case PT_SC_REMOTE:
1819 		pscr = addr;
1820 		CTR2(KTR_PTRACE, "PT_SC_REMOTE: pid %d, syscall %d",
1821 		    p->p_pid, pscr->pscr_syscall);
1822 		if ((td2->td_dbgflags & TDB_BOUNDARY) == 0) {
1823 			error = EBUSY;
1824 			break;
1825 		}
1826 		PROC_UNLOCK(p);
1827 		MPASS(pscr->pscr_nargs <= nitems(td->td_sa.args));
1828 
1829 		tsr = malloc(sizeof(struct thr_syscall_req), M_TEMP,
1830 		    M_WAITOK | M_ZERO);
1831 
1832 		tsr->ts_sa.code = pscr->pscr_syscall;
1833 		tsr->ts_nargs = pscr->pscr_nargs;
1834 		memcpy(&tsr->ts_sa.args, pscr->pscr_args,
1835 		    sizeof(syscallarg_t) * tsr->ts_nargs);
1836 
1837 		PROC_LOCK(p);
1838 		error = proc_can_ptrace(td, p);
1839 		if (error != 0) {
1840 			free(tsr, M_TEMP);
1841 			break;
1842 		}
1843 		if (td2->td_proc != p) {
1844 			free(tsr, M_TEMP);
1845 			error = ESRCH;
1846 			break;
1847 		}
1848 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1849 		    TDB_SCREMOTEREQ)) == 0,
1850 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1851 
1852 		td2->td_remotereq = tsr;
1853 		td2->td_dbgflags |= TDB_SCREMOTEREQ;
1854 		thread_run_flash(td2);
1855 		while ((td2->td_dbgflags & TDB_SCREMOTEREQ) != 0)
1856 			msleep(p, &p->p_mtx, PPAUSE, "pscrx", 0);
1857 		error = 0;
1858 		memcpy(&pscr->pscr_ret, &tsr->ts_ret, sizeof(tsr->ts_ret));
1859 		free(tsr, M_TEMP);
1860 		break;
1861 
1862 	default:
1863 #ifdef __HAVE_PTRACE_MACHDEP
1864 		if (req >= PT_FIRSTMACH) {
1865 			PROC_UNLOCK(p);
1866 			error = cpu_ptrace(td2, req, addr, data);
1867 			PROC_LOCK(p);
1868 		} else
1869 #endif
1870 			/* Unknown request. */
1871 			error = EINVAL;
1872 		break;
1873 	}
1874 out:
1875 	/* Drop our hold on this process now that the request has completed. */
1876 	_PRELE(p);
1877 fail:
1878 	if (p2_req_set) {
1879 		if ((p->p_flag2 & P2_PTRACEREQ) != 0)
1880 			wakeup(&p->p_flag2);
1881 		p->p_flag2 &= ~P2_PTRACEREQ;
1882 	}
1883 	PROC_UNLOCK(p);
1884 	if (proctree_locked)
1885 		sx_xunlock(&proctree_lock);
1886 	return (error);
1887 }
1888 #undef PROC_READ
1889 #undef PROC_WRITE
1890