xref: /freebsd/sys/kern/sys_process.c (revision 13ec1e3155c7e9bf037b12af186351b7fa9b9450)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 1994, Sean Eric Fagan
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by Sean Eric Fagan.
18  * 4. The name of the author may not be used to endorse or promote products
19  *    derived from this software without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/ktr.h>
40 #include <sys/limits.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/reg.h>
44 #include <sys/syscallsubr.h>
45 #include <sys/sysent.h>
46 #include <sys/sysproto.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/vnode.h>
50 #include <sys/ptrace.h>
51 #include <sys/rwlock.h>
52 #include <sys/sx.h>
53 #include <sys/malloc.h>
54 #include <sys/signalvar.h>
55 #include <sys/caprights.h>
56 #include <sys/filedesc.h>
57 
58 #include <security/audit/audit.h>
59 
60 #include <vm/vm.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_extern.h>
63 #include <vm/vm_map.h>
64 #include <vm/vm_kern.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_param.h>
68 
69 #ifdef COMPAT_FREEBSD32
70 #include <sys/procfs.h>
71 #endif
72 
73 /* Assert it's safe to unlock a process, e.g. to allocate working memory */
74 #define	PROC_ASSERT_TRACEREQ(p)	MPASS(((p)->p_flag2 & P2_PTRACEREQ) != 0)
75 
76 /*
77  * Functions implemented using PROC_ACTION():
78  *
79  * proc_read_regs(proc, regs)
80  *	Get the current user-visible register set from the process
81  *	and copy it into the regs structure (<machine/reg.h>).
82  *	The process is stopped at the time read_regs is called.
83  *
84  * proc_write_regs(proc, regs)
85  *	Update the current register set from the passed in regs
86  *	structure.  Take care to avoid clobbering special CPU
87  *	registers or privileged bits in the PSL.
88  *	Depending on the architecture this may have fix-up work to do,
89  *	especially if the IAR or PCW are modified.
90  *	The process is stopped at the time write_regs is called.
91  *
92  * proc_read_fpregs, proc_write_fpregs
93  *	deal with the floating point register set, otherwise as above.
94  *
95  * proc_read_dbregs, proc_write_dbregs
96  *	deal with the processor debug register set, otherwise as above.
97  *
98  * proc_sstep(proc)
99  *	Arrange for the process to trap after executing a single instruction.
100  */
101 
102 #define	PROC_ACTION(action) do {					\
103 	int error;							\
104 									\
105 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
106 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
107 		error = EIO;						\
108 	else								\
109 		error = (action);					\
110 	return (error);							\
111 } while (0)
112 
113 int
114 proc_read_regs(struct thread *td, struct reg *regs)
115 {
116 
117 	PROC_ACTION(fill_regs(td, regs));
118 }
119 
120 int
121 proc_write_regs(struct thread *td, struct reg *regs)
122 {
123 
124 	PROC_ACTION(set_regs(td, regs));
125 }
126 
127 int
128 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
129 {
130 
131 	PROC_ACTION(fill_dbregs(td, dbregs));
132 }
133 
134 int
135 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
136 {
137 
138 	PROC_ACTION(set_dbregs(td, dbregs));
139 }
140 
141 /*
142  * Ptrace doesn't support fpregs at all, and there are no security holes
143  * or translations for fpregs, so we can just copy them.
144  */
145 int
146 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
147 {
148 
149 	PROC_ACTION(fill_fpregs(td, fpregs));
150 }
151 
152 int
153 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
154 {
155 
156 	PROC_ACTION(set_fpregs(td, fpregs));
157 }
158 
159 static struct regset *
160 proc_find_regset(struct thread *td, int note)
161 {
162 	struct regset **regsetp, **regset_end, *regset;
163 	struct sysentvec *sv;
164 
165 	sv = td->td_proc->p_sysent;
166 	regsetp = sv->sv_regset_begin;
167 	if (regsetp == NULL)
168 		return (NULL);
169 	regset_end = sv->sv_regset_end;
170 	MPASS(regset_end != NULL);
171 	for (; regsetp < regset_end; regsetp++) {
172 		regset = *regsetp;
173 		if (regset->note != note)
174 			continue;
175 
176 		return (regset);
177 	}
178 
179 	return (NULL);
180 }
181 
182 static int
183 proc_read_regset(struct thread *td, int note, struct iovec *iov)
184 {
185 	struct regset *regset;
186 	struct proc *p;
187 	void *buf;
188 	size_t size;
189 	int error;
190 
191 	regset = proc_find_regset(td, note);
192 	if (regset == NULL)
193 		return (EINVAL);
194 
195 	if (iov->iov_base == NULL) {
196 		iov->iov_len = regset->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 != regset->size)
205 		return (EINVAL);
206 
207 	if (regset->get == NULL)
208 		return (EINVAL);
209 
210 	error = 0;
211 	size = regset->size;
212 	p = td->td_proc;
213 
214 	/* Drop the proc lock while allocating the temp buffer */
215 	PROC_ASSERT_TRACEREQ(p);
216 	PROC_UNLOCK(p);
217 	buf = malloc(size, M_TEMP, M_WAITOK);
218 	PROC_LOCK(p);
219 
220 	if (!regset->get(regset, td, buf, &size)) {
221 		error = EINVAL;
222 	} else {
223 		KASSERT(size == regset->size,
224 		    ("%s: Getter function changed the size", __func__));
225 
226 		iov->iov_len = size;
227 		PROC_UNLOCK(p);
228 		error = copyout(buf, iov->iov_base, size);
229 		PROC_LOCK(p);
230 	}
231 
232 	free(buf, M_TEMP);
233 
234 	return (error);
235 }
236 
237 static int
238 proc_write_regset(struct thread *td, int note, struct iovec *iov)
239 {
240 	struct regset *regset;
241 	struct proc *p;
242 	void *buf;
243 	size_t size;
244 	int error;
245 
246 	regset = proc_find_regset(td, note);
247 	if (regset == NULL)
248 		return (EINVAL);
249 
250 	/* The length is wrong, return an error */
251 	if (iov->iov_len != regset->size)
252 		return (EINVAL);
253 
254 	if (regset->set == NULL)
255 		return (EINVAL);
256 
257 	size = regset->size;
258 	p = td->td_proc;
259 
260 	/* Drop the proc lock while allocating the temp buffer */
261 	PROC_ASSERT_TRACEREQ(p);
262 	PROC_UNLOCK(p);
263 	buf = malloc(size, M_TEMP, M_WAITOK);
264 	error = copyin(iov->iov_base, buf, size);
265 	PROC_LOCK(p);
266 
267 	if (error == 0) {
268 		if (!regset->set(regset, td, buf, size)) {
269 			error = EINVAL;
270 		}
271 	}
272 
273 	free(buf, M_TEMP);
274 
275 	return (error);
276 }
277 
278 #ifdef COMPAT_FREEBSD32
279 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
280 int
281 proc_read_regs32(struct thread *td, struct reg32 *regs32)
282 {
283 
284 	PROC_ACTION(fill_regs32(td, regs32));
285 }
286 
287 int
288 proc_write_regs32(struct thread *td, struct reg32 *regs32)
289 {
290 
291 	PROC_ACTION(set_regs32(td, regs32));
292 }
293 
294 int
295 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
296 {
297 
298 	PROC_ACTION(fill_dbregs32(td, dbregs32));
299 }
300 
301 int
302 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
303 {
304 
305 	PROC_ACTION(set_dbregs32(td, dbregs32));
306 }
307 
308 int
309 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
310 {
311 
312 	PROC_ACTION(fill_fpregs32(td, fpregs32));
313 }
314 
315 int
316 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
317 {
318 
319 	PROC_ACTION(set_fpregs32(td, fpregs32));
320 }
321 #endif
322 
323 int
324 proc_sstep(struct thread *td)
325 {
326 
327 	PROC_ACTION(ptrace_single_step(td));
328 }
329 
330 int
331 proc_rwmem(struct proc *p, struct uio *uio)
332 {
333 	vm_map_t map;
334 	vm_offset_t pageno;		/* page number */
335 	vm_prot_t reqprot;
336 	int error, fault_flags, page_offset, writing;
337 
338 	/*
339 	 * Assert that someone has locked this vmspace.  (Should be
340 	 * curthread but we can't assert that.)  This keeps the process
341 	 * from exiting out from under us until this operation completes.
342 	 */
343 	PROC_ASSERT_HELD(p);
344 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
345 
346 	/*
347 	 * The map we want...
348 	 */
349 	map = &p->p_vmspace->vm_map;
350 
351 	/*
352 	 * If we are writing, then we request vm_fault() to create a private
353 	 * copy of each page.  Since these copies will not be writeable by the
354 	 * process, we must explicity request that they be dirtied.
355 	 */
356 	writing = uio->uio_rw == UIO_WRITE;
357 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
358 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
359 
360 	/*
361 	 * Only map in one page at a time.  We don't have to, but it
362 	 * makes things easier.  This way is trivial - right?
363 	 */
364 	do {
365 		vm_offset_t uva;
366 		u_int len;
367 		vm_page_t m;
368 
369 		uva = (vm_offset_t)uio->uio_offset;
370 
371 		/*
372 		 * Get the page number of this segment.
373 		 */
374 		pageno = trunc_page(uva);
375 		page_offset = uva - pageno;
376 
377 		/*
378 		 * How many bytes to copy
379 		 */
380 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
381 
382 		/*
383 		 * Fault and hold the page on behalf of the process.
384 		 */
385 		error = vm_fault(map, pageno, reqprot, fault_flags, &m);
386 		if (error != KERN_SUCCESS) {
387 			if (error == KERN_RESOURCE_SHORTAGE)
388 				error = ENOMEM;
389 			else
390 				error = EFAULT;
391 			break;
392 		}
393 
394 		/*
395 		 * Now do the i/o move.
396 		 */
397 		error = uiomove_fromphys(&m, page_offset, len, uio);
398 
399 		/* Make the I-cache coherent for breakpoints. */
400 		if (writing && error == 0) {
401 			vm_map_lock_read(map);
402 			if (vm_map_check_protection(map, pageno, pageno +
403 			    PAGE_SIZE, VM_PROT_EXECUTE))
404 				vm_sync_icache(map, uva, len);
405 			vm_map_unlock_read(map);
406 		}
407 
408 		/*
409 		 * Release the page.
410 		 */
411 		vm_page_unwire(m, PQ_ACTIVE);
412 
413 	} while (error == 0 && uio->uio_resid > 0);
414 
415 	return (error);
416 }
417 
418 static ssize_t
419 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
420     size_t len, enum uio_rw rw)
421 {
422 	struct iovec iov;
423 	struct uio uio;
424 	ssize_t slen;
425 
426 	MPASS(len < SSIZE_MAX);
427 	slen = (ssize_t)len;
428 
429 	iov.iov_base = (caddr_t)buf;
430 	iov.iov_len = len;
431 	uio.uio_iov = &iov;
432 	uio.uio_iovcnt = 1;
433 	uio.uio_offset = va;
434 	uio.uio_resid = slen;
435 	uio.uio_segflg = UIO_SYSSPACE;
436 	uio.uio_rw = rw;
437 	uio.uio_td = td;
438 	proc_rwmem(p, &uio);
439 	if (uio.uio_resid == slen)
440 		return (-1);
441 	return (slen - uio.uio_resid);
442 }
443 
444 ssize_t
445 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
446     size_t len)
447 {
448 
449 	return (proc_iop(td, p, va, buf, len, UIO_READ));
450 }
451 
452 ssize_t
453 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
454     size_t len)
455 {
456 
457 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
458 }
459 
460 static int
461 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
462 {
463 	struct vattr vattr;
464 	vm_map_t map;
465 	vm_map_entry_t entry;
466 	vm_object_t obj, tobj, lobj;
467 	struct vmspace *vm;
468 	struct vnode *vp;
469 	char *freepath, *fullpath;
470 	u_int pathlen;
471 	int error, index;
472 
473 	error = 0;
474 	obj = NULL;
475 
476 	vm = vmspace_acquire_ref(p);
477 	map = &vm->vm_map;
478 	vm_map_lock_read(map);
479 
480 	do {
481 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
482 		    ("Submap in map header"));
483 		index = 0;
484 		VM_MAP_ENTRY_FOREACH(entry, map) {
485 			if (index >= pve->pve_entry &&
486 			    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
487 				break;
488 			index++;
489 		}
490 		if (index < pve->pve_entry) {
491 			error = EINVAL;
492 			break;
493 		}
494 		if (entry == &map->header) {
495 			error = ENOENT;
496 			break;
497 		}
498 
499 		/* We got an entry. */
500 		pve->pve_entry = index + 1;
501 		pve->pve_timestamp = map->timestamp;
502 		pve->pve_start = entry->start;
503 		pve->pve_end = entry->end - 1;
504 		pve->pve_offset = entry->offset;
505 		pve->pve_prot = entry->protection;
506 
507 		/* Backing object's path needed? */
508 		if (pve->pve_pathlen == 0)
509 			break;
510 
511 		pathlen = pve->pve_pathlen;
512 		pve->pve_pathlen = 0;
513 
514 		obj = entry->object.vm_object;
515 		if (obj != NULL)
516 			VM_OBJECT_RLOCK(obj);
517 	} while (0);
518 
519 	vm_map_unlock_read(map);
520 
521 	pve->pve_fsid = VNOVAL;
522 	pve->pve_fileid = VNOVAL;
523 
524 	if (error == 0 && obj != NULL) {
525 		lobj = obj;
526 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
527 			if (tobj != obj)
528 				VM_OBJECT_RLOCK(tobj);
529 			if (lobj != obj)
530 				VM_OBJECT_RUNLOCK(lobj);
531 			lobj = tobj;
532 			pve->pve_offset += tobj->backing_object_offset;
533 		}
534 		vp = vm_object_vnode(lobj);
535 		if (vp != NULL)
536 			vref(vp);
537 		if (lobj != obj)
538 			VM_OBJECT_RUNLOCK(lobj);
539 		VM_OBJECT_RUNLOCK(obj);
540 
541 		if (vp != NULL) {
542 			freepath = NULL;
543 			fullpath = NULL;
544 			vn_fullpath(vp, &fullpath, &freepath);
545 			vn_lock(vp, LK_SHARED | LK_RETRY);
546 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
547 				pve->pve_fileid = vattr.va_fileid;
548 				pve->pve_fsid = vattr.va_fsid;
549 			}
550 			vput(vp);
551 
552 			if (fullpath != NULL) {
553 				pve->pve_pathlen = strlen(fullpath) + 1;
554 				if (pve->pve_pathlen <= pathlen) {
555 					error = copyout(fullpath, pve->pve_path,
556 					    pve->pve_pathlen);
557 				} else
558 					error = ENAMETOOLONG;
559 			}
560 			if (freepath != NULL)
561 				free(freepath, M_TEMP);
562 		}
563 	}
564 	vmspace_free(vm);
565 	if (error == 0)
566 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
567 		    p->p_pid, pve->pve_entry, pve->pve_start);
568 
569 	return (error);
570 }
571 
572 /*
573  * Process debugging system call.
574  */
575 #ifndef _SYS_SYSPROTO_H_
576 struct ptrace_args {
577 	int	req;
578 	pid_t	pid;
579 	caddr_t	addr;
580 	int	data;
581 };
582 #endif
583 
584 int
585 sys_ptrace(struct thread *td, struct ptrace_args *uap)
586 {
587 	/*
588 	 * XXX this obfuscation is to reduce stack usage, but the register
589 	 * structs may be too large to put on the stack anyway.
590 	 */
591 	union {
592 		struct ptrace_io_desc piod;
593 		struct ptrace_lwpinfo pl;
594 		struct ptrace_vm_entry pve;
595 		struct ptrace_coredump pc;
596 		struct dbreg dbreg;
597 		struct fpreg fpreg;
598 		struct reg reg;
599 		struct iovec vec;
600 		char args[sizeof(td->td_sa.args)];
601 		struct ptrace_sc_ret psr;
602 		int ptevents;
603 	} r;
604 	void *addr;
605 	int error;
606 
607 	if (!allow_ptrace)
608 		return (ENOSYS);
609 	error = 0;
610 
611 	AUDIT_ARG_PID(uap->pid);
612 	AUDIT_ARG_CMD(uap->req);
613 	AUDIT_ARG_VALUE(uap->data);
614 	addr = &r;
615 	switch (uap->req) {
616 	case PT_GET_EVENT_MASK:
617 	case PT_LWPINFO:
618 	case PT_GET_SC_ARGS:
619 	case PT_GET_SC_RET:
620 		break;
621 	case PT_GETREGS:
622 		bzero(&r.reg, sizeof(r.reg));
623 		break;
624 	case PT_GETFPREGS:
625 		bzero(&r.fpreg, sizeof(r.fpreg));
626 		break;
627 	case PT_GETDBREGS:
628 		bzero(&r.dbreg, sizeof(r.dbreg));
629 		break;
630 	case PT_GETREGSET:
631 	case PT_SETREGSET:
632 		error = copyin(uap->addr, &r.vec, sizeof(r.vec));
633 		break;
634 	case PT_SETREGS:
635 		error = copyin(uap->addr, &r.reg, sizeof(r.reg));
636 		break;
637 	case PT_SETFPREGS:
638 		error = copyin(uap->addr, &r.fpreg, sizeof(r.fpreg));
639 		break;
640 	case PT_SETDBREGS:
641 		error = copyin(uap->addr, &r.dbreg, sizeof(r.dbreg));
642 		break;
643 	case PT_SET_EVENT_MASK:
644 		if (uap->data != sizeof(r.ptevents))
645 			error = EINVAL;
646 		else
647 			error = copyin(uap->addr, &r.ptevents, uap->data);
648 		break;
649 	case PT_IO:
650 		error = copyin(uap->addr, &r.piod, sizeof(r.piod));
651 		break;
652 	case PT_VM_ENTRY:
653 		error = copyin(uap->addr, &r.pve, sizeof(r.pve));
654 		break;
655 	case PT_COREDUMP:
656 		if (uap->data != sizeof(r.pc))
657 			error = EINVAL;
658 		else
659 			error = copyin(uap->addr, &r.pc, uap->data);
660 		break;
661 	default:
662 		addr = uap->addr;
663 		break;
664 	}
665 	if (error)
666 		return (error);
667 
668 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
669 	if (error)
670 		return (error);
671 
672 	switch (uap->req) {
673 	case PT_VM_ENTRY:
674 		error = copyout(&r.pve, uap->addr, sizeof(r.pve));
675 		break;
676 	case PT_IO:
677 		error = copyout(&r.piod, uap->addr, sizeof(r.piod));
678 		break;
679 	case PT_GETREGS:
680 		error = copyout(&r.reg, uap->addr, sizeof(r.reg));
681 		break;
682 	case PT_GETFPREGS:
683 		error = copyout(&r.fpreg, uap->addr, sizeof(r.fpreg));
684 		break;
685 	case PT_GETDBREGS:
686 		error = copyout(&r.dbreg, uap->addr, sizeof(r.dbreg));
687 		break;
688 	case PT_GETREGSET:
689 		error = copyout(&r.vec, uap->addr, sizeof(r.vec));
690 		break;
691 	case PT_GET_EVENT_MASK:
692 		/* NB: The size in uap->data is validated in kern_ptrace(). */
693 		error = copyout(&r.ptevents, uap->addr, uap->data);
694 		break;
695 	case PT_LWPINFO:
696 		/* NB: The size in uap->data is validated in kern_ptrace(). */
697 		error = copyout(&r.pl, uap->addr, uap->data);
698 		break;
699 	case PT_GET_SC_ARGS:
700 		error = copyout(r.args, uap->addr, MIN(uap->data,
701 		    sizeof(r.args)));
702 		break;
703 	case PT_GET_SC_RET:
704 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
705 		    sizeof(r.psr)));
706 		break;
707 	}
708 
709 	return (error);
710 }
711 
712 #ifdef COMPAT_FREEBSD32
713 /*
714  *   PROC_READ(regs, td2, addr);
715  * becomes either:
716  *   proc_read_regs(td2, addr);
717  * or
718  *   proc_read_regs32(td2, addr);
719  * .. except this is done at runtime.  There is an additional
720  * complication in that PROC_WRITE disallows 32 bit consumers
721  * from writing to 64 bit address space targets.
722  */
723 #define	PROC_READ(w, t, a)	wrap32 ? \
724 	proc_read_ ## w ## 32(t, a) : \
725 	proc_read_ ## w (t, a)
726 #define	PROC_WRITE(w, t, a)	wrap32 ? \
727 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
728 	proc_write_ ## w (t, a)
729 #else
730 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
731 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
732 #endif
733 
734 void
735 proc_set_traced(struct proc *p, bool stop)
736 {
737 
738 	sx_assert(&proctree_lock, SX_XLOCKED);
739 	PROC_LOCK_ASSERT(p, MA_OWNED);
740 	p->p_flag |= P_TRACED;
741 	if (stop)
742 		p->p_flag2 |= P2_PTRACE_FSTP;
743 	p->p_ptevents = PTRACE_DEFAULT;
744 }
745 
746 void
747 ptrace_unsuspend(struct proc *p)
748 {
749 	PROC_LOCK_ASSERT(p, MA_OWNED);
750 
751 	PROC_SLOCK(p);
752 	p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
753 	thread_unsuspend(p);
754 	PROC_SUNLOCK(p);
755 	itimer_proc_continue(p);
756 	kqtimer_proc_continue(p);
757 }
758 
759 static int
760 proc_can_ptrace(struct thread *td, struct proc *p)
761 {
762 	int error;
763 
764 	PROC_LOCK_ASSERT(p, MA_OWNED);
765 
766 	if ((p->p_flag & P_WEXIT) != 0)
767 		return (ESRCH);
768 
769 	if ((error = p_cansee(td, p)) != 0)
770 		return (error);
771 	if ((error = p_candebug(td, p)) != 0)
772 		return (error);
773 
774 	/* not being traced... */
775 	if ((p->p_flag & P_TRACED) == 0)
776 		return (EPERM);
777 
778 	/* not being traced by YOU */
779 	if (p->p_pptr != td->td_proc)
780 		return (EBUSY);
781 
782 	/* not currently stopped */
783 	if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
784 	    p->p_suspcount != p->p_numthreads  ||
785 	    (p->p_flag & P_WAITED) == 0)
786 		return (EBUSY);
787 
788 	return (0);
789 }
790 
791 static struct thread *
792 ptrace_sel_coredump_thread(struct proc *p)
793 {
794 	struct thread *td2;
795 
796 	PROC_LOCK_ASSERT(p, MA_OWNED);
797 	MPASS((p->p_flag & P_STOPPED_TRACE) != 0);
798 
799 	FOREACH_THREAD_IN_PROC(p, td2) {
800 		if ((td2->td_dbgflags & TDB_SSWITCH) != 0)
801 			return (td2);
802 	}
803 	return (NULL);
804 }
805 
806 int
807 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
808 {
809 	struct iovec iov;
810 	struct uio uio;
811 	struct proc *curp, *p, *pp;
812 	struct thread *td2 = NULL, *td3;
813 	struct ptrace_io_desc *piod = NULL;
814 	struct ptrace_lwpinfo *pl;
815 	struct ptrace_sc_ret *psr;
816 	struct file *fp;
817 	struct ptrace_coredump *pc;
818 	struct thr_coredump_req *tcq;
819 	int error, num, tmp;
820 	lwpid_t tid = 0, *buf;
821 #ifdef COMPAT_FREEBSD32
822 	int wrap32 = 0, safe = 0;
823 #endif
824 	bool proctree_locked, p2_req_set;
825 
826 	curp = td->td_proc;
827 	proctree_locked = false;
828 	p2_req_set = false;
829 
830 	/* Lock proctree before locking the process. */
831 	switch (req) {
832 	case PT_TRACE_ME:
833 	case PT_ATTACH:
834 	case PT_STEP:
835 	case PT_CONTINUE:
836 	case PT_TO_SCE:
837 	case PT_TO_SCX:
838 	case PT_SYSCALL:
839 	case PT_FOLLOW_FORK:
840 	case PT_LWP_EVENTS:
841 	case PT_GET_EVENT_MASK:
842 	case PT_SET_EVENT_MASK:
843 	case PT_DETACH:
844 	case PT_GET_SC_ARGS:
845 		sx_xlock(&proctree_lock);
846 		proctree_locked = true;
847 		break;
848 	default:
849 		break;
850 	}
851 
852 	if (req == PT_TRACE_ME) {
853 		p = td->td_proc;
854 		PROC_LOCK(p);
855 	} else {
856 		if (pid <= PID_MAX) {
857 			if ((p = pfind(pid)) == NULL) {
858 				if (proctree_locked)
859 					sx_xunlock(&proctree_lock);
860 				return (ESRCH);
861 			}
862 		} else {
863 			td2 = tdfind(pid, -1);
864 			if (td2 == NULL) {
865 				if (proctree_locked)
866 					sx_xunlock(&proctree_lock);
867 				return (ESRCH);
868 			}
869 			p = td2->td_proc;
870 			tid = pid;
871 			pid = p->p_pid;
872 		}
873 	}
874 	AUDIT_ARG_PROCESS(p);
875 
876 	if ((p->p_flag & P_WEXIT) != 0) {
877 		error = ESRCH;
878 		goto fail;
879 	}
880 	if ((error = p_cansee(td, p)) != 0)
881 		goto fail;
882 
883 	if ((error = p_candebug(td, p)) != 0)
884 		goto fail;
885 
886 	/*
887 	 * System processes can't be debugged.
888 	 */
889 	if ((p->p_flag & P_SYSTEM) != 0) {
890 		error = EINVAL;
891 		goto fail;
892 	}
893 
894 	if (tid == 0) {
895 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
896 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
897 			td2 = p->p_xthread;
898 		} else {
899 			td2 = FIRST_THREAD_IN_PROC(p);
900 		}
901 		tid = td2->td_tid;
902 	}
903 
904 #ifdef COMPAT_FREEBSD32
905 	/*
906 	 * Test if we're a 32 bit client and what the target is.
907 	 * Set the wrap controls accordingly.
908 	 */
909 	if (SV_CURPROC_FLAG(SV_ILP32)) {
910 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
911 			safe = 1;
912 		wrap32 = 1;
913 	}
914 #endif
915 	/*
916 	 * Permissions check
917 	 */
918 	switch (req) {
919 	case PT_TRACE_ME:
920 		/*
921 		 * Always legal, when there is a parent process which
922 		 * could trace us.  Otherwise, reject.
923 		 */
924 		if ((p->p_flag & P_TRACED) != 0) {
925 			error = EBUSY;
926 			goto fail;
927 		}
928 		if (p->p_pptr == initproc) {
929 			error = EPERM;
930 			goto fail;
931 		}
932 		break;
933 
934 	case PT_ATTACH:
935 		/* Self */
936 		if (p == td->td_proc) {
937 			error = EINVAL;
938 			goto fail;
939 		}
940 
941 		/* Already traced */
942 		if (p->p_flag & P_TRACED) {
943 			error = EBUSY;
944 			goto fail;
945 		}
946 
947 		/* Can't trace an ancestor if you're being traced. */
948 		if (curp->p_flag & P_TRACED) {
949 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
950 				if (pp == p) {
951 					error = EINVAL;
952 					goto fail;
953 				}
954 			}
955 		}
956 
957 		/* OK */
958 		break;
959 
960 	case PT_CLEARSTEP:
961 		/* Allow thread to clear single step for itself */
962 		if (td->td_tid == tid)
963 			break;
964 
965 		/* FALLTHROUGH */
966 	default:
967 		/*
968 		 * Check for ptrace eligibility before waiting for
969 		 * holds to drain.
970 		 */
971 		error = proc_can_ptrace(td, p);
972 		if (error != 0)
973 			goto fail;
974 
975 		/*
976 		 * Block parallel ptrace requests.  Most important, do
977 		 * not allow other thread in debugger to continue the
978 		 * debuggee until coredump finished.
979 		 */
980 		while ((p->p_flag2 & P2_PTRACEREQ) != 0) {
981 			if (proctree_locked)
982 				sx_xunlock(&proctree_lock);
983 			error = msleep(&p->p_flag2, &p->p_mtx, PPAUSE | PCATCH |
984 			    (proctree_locked ? PDROP : 0), "pptrace", 0);
985 			if (proctree_locked) {
986 				sx_xlock(&proctree_lock);
987 				PROC_LOCK(p);
988 			}
989 			if (error == 0 && td2->td_proc != p)
990 				error = ESRCH;
991 			if (error == 0)
992 				error = proc_can_ptrace(td, p);
993 			if (error != 0)
994 				goto fail;
995 		}
996 
997 		/* Ok */
998 		break;
999 	}
1000 
1001 	/*
1002 	 * Keep this process around and request parallel ptrace()
1003 	 * request to wait until we finish this request.
1004 	 */
1005 	MPASS((p->p_flag2 & P2_PTRACEREQ) == 0);
1006 	p->p_flag2 |= P2_PTRACEREQ;
1007 	p2_req_set = true;
1008 	_PHOLD(p);
1009 
1010 	/*
1011 	 * Actually do the requests
1012 	 */
1013 
1014 	td->td_retval[0] = 0;
1015 
1016 	switch (req) {
1017 	case PT_TRACE_ME:
1018 		/* set my trace flag and "owner" so it can read/write me */
1019 		proc_set_traced(p, false);
1020 		if (p->p_flag & P_PPWAIT)
1021 			p->p_flag |= P_PPTRACE;
1022 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
1023 		break;
1024 
1025 	case PT_ATTACH:
1026 		/* security check done above */
1027 		/*
1028 		 * It would be nice if the tracing relationship was separate
1029 		 * from the parent relationship but that would require
1030 		 * another set of links in the proc struct or for "wait"
1031 		 * to scan the entire proc table.  To make life easier,
1032 		 * we just re-parent the process we're trying to trace.
1033 		 * The old parent is remembered so we can put things back
1034 		 * on a "detach".
1035 		 */
1036 		proc_set_traced(p, true);
1037 		proc_reparent(p, td->td_proc, false);
1038 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
1039 		    p->p_oppid);
1040 
1041 		sx_xunlock(&proctree_lock);
1042 		proctree_locked = false;
1043 		MPASS(p->p_xthread == NULL);
1044 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
1045 
1046 		/*
1047 		 * If already stopped due to a stop signal, clear the
1048 		 * existing stop before triggering a traced SIGSTOP.
1049 		 */
1050 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
1051 			PROC_SLOCK(p);
1052 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
1053 			thread_unsuspend(p);
1054 			PROC_SUNLOCK(p);
1055 		}
1056 
1057 		kern_psignal(p, SIGSTOP);
1058 		break;
1059 
1060 	case PT_CLEARSTEP:
1061 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
1062 		    p->p_pid);
1063 		error = ptrace_clear_single_step(td2);
1064 		break;
1065 
1066 	case PT_SETSTEP:
1067 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
1068 		    p->p_pid);
1069 		error = ptrace_single_step(td2);
1070 		break;
1071 
1072 	case PT_SUSPEND:
1073 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
1074 		    p->p_pid);
1075 		td2->td_dbgflags |= TDB_SUSPEND;
1076 		thread_lock(td2);
1077 		td2->td_flags |= TDF_NEEDSUSPCHK;
1078 		thread_unlock(td2);
1079 		break;
1080 
1081 	case PT_RESUME:
1082 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1083 		    p->p_pid);
1084 		td2->td_dbgflags &= ~TDB_SUSPEND;
1085 		break;
1086 
1087 	case PT_FOLLOW_FORK:
1088 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1089 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1090 		    data ? "enabled" : "disabled");
1091 		if (data)
1092 			p->p_ptevents |= PTRACE_FORK;
1093 		else
1094 			p->p_ptevents &= ~PTRACE_FORK;
1095 		break;
1096 
1097 	case PT_LWP_EVENTS:
1098 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1099 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1100 		    data ? "enabled" : "disabled");
1101 		if (data)
1102 			p->p_ptevents |= PTRACE_LWP;
1103 		else
1104 			p->p_ptevents &= ~PTRACE_LWP;
1105 		break;
1106 
1107 	case PT_GET_EVENT_MASK:
1108 		if (data != sizeof(p->p_ptevents)) {
1109 			error = EINVAL;
1110 			break;
1111 		}
1112 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1113 		    p->p_ptevents);
1114 		*(int *)addr = p->p_ptevents;
1115 		break;
1116 
1117 	case PT_SET_EVENT_MASK:
1118 		if (data != sizeof(p->p_ptevents)) {
1119 			error = EINVAL;
1120 			break;
1121 		}
1122 		tmp = *(int *)addr;
1123 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1124 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1125 			error = EINVAL;
1126 			break;
1127 		}
1128 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1129 		    p->p_pid, p->p_ptevents, tmp);
1130 		p->p_ptevents = tmp;
1131 		break;
1132 
1133 	case PT_GET_SC_ARGS:
1134 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1135 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1136 #ifdef COMPAT_FREEBSD32
1137 		    || (wrap32 && !safe)
1138 #endif
1139 		    ) {
1140 			error = EINVAL;
1141 			break;
1142 		}
1143 		bzero(addr, sizeof(td2->td_sa.args));
1144 		/* See the explanation in linux_ptrace_get_syscall_info(). */
1145 		bcopy(td2->td_sa.args, addr, SV_PROC_ABI(td->td_proc) ==
1146 		    SV_ABI_LINUX ? sizeof(td2->td_sa.args) :
1147 		    td2->td_sa.callp->sy_narg * sizeof(register_t));
1148 		break;
1149 
1150 	case PT_GET_SC_RET:
1151 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1152 #ifdef COMPAT_FREEBSD32
1153 		    || (wrap32 && !safe)
1154 #endif
1155 		    ) {
1156 			error = EINVAL;
1157 			break;
1158 		}
1159 		psr = addr;
1160 		bzero(psr, sizeof(*psr));
1161 		psr->sr_error = td2->td_errno;
1162 		if (psr->sr_error == 0) {
1163 			psr->sr_retval[0] = td2->td_retval[0];
1164 			psr->sr_retval[1] = td2->td_retval[1];
1165 		}
1166 		CTR4(KTR_PTRACE,
1167 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1168 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1169 		    psr->sr_retval[1]);
1170 		break;
1171 
1172 	case PT_STEP:
1173 	case PT_CONTINUE:
1174 	case PT_TO_SCE:
1175 	case PT_TO_SCX:
1176 	case PT_SYSCALL:
1177 	case PT_DETACH:
1178 		/* Zero means do not send any signal */
1179 		if (data < 0 || data > _SIG_MAXSIG) {
1180 			error = EINVAL;
1181 			break;
1182 		}
1183 
1184 		switch (req) {
1185 		case PT_STEP:
1186 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1187 			    td2->td_tid, p->p_pid, data);
1188 			error = ptrace_single_step(td2);
1189 			if (error)
1190 				goto out;
1191 			break;
1192 		case PT_CONTINUE:
1193 		case PT_TO_SCE:
1194 		case PT_TO_SCX:
1195 		case PT_SYSCALL:
1196 			if (addr != (void *)1) {
1197 				error = ptrace_set_pc(td2,
1198 				    (u_long)(uintfptr_t)addr);
1199 				if (error)
1200 					goto out;
1201 			}
1202 			switch (req) {
1203 			case PT_TO_SCE:
1204 				p->p_ptevents |= PTRACE_SCE;
1205 				CTR4(KTR_PTRACE,
1206 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1207 				    p->p_pid, p->p_ptevents,
1208 				    (u_long)(uintfptr_t)addr, data);
1209 				break;
1210 			case PT_TO_SCX:
1211 				p->p_ptevents |= PTRACE_SCX;
1212 				CTR4(KTR_PTRACE,
1213 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1214 				    p->p_pid, p->p_ptevents,
1215 				    (u_long)(uintfptr_t)addr, data);
1216 				break;
1217 			case PT_SYSCALL:
1218 				p->p_ptevents |= PTRACE_SYSCALL;
1219 				CTR4(KTR_PTRACE,
1220 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1221 				    p->p_pid, p->p_ptevents,
1222 				    (u_long)(uintfptr_t)addr, data);
1223 				break;
1224 			case PT_CONTINUE:
1225 				CTR3(KTR_PTRACE,
1226 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1227 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1228 				break;
1229 			}
1230 			break;
1231 		case PT_DETACH:
1232 			/*
1233 			 * Clear P_TRACED before reparenting
1234 			 * a detached process back to its original
1235 			 * parent.  Otherwise the debugee will be set
1236 			 * as an orphan of the debugger.
1237 			 */
1238 			p->p_flag &= ~(P_TRACED | P_WAITED);
1239 
1240 			/*
1241 			 * Reset the process parent.
1242 			 */
1243 			if (p->p_oppid != p->p_pptr->p_pid) {
1244 				PROC_LOCK(p->p_pptr);
1245 				sigqueue_take(p->p_ksi);
1246 				PROC_UNLOCK(p->p_pptr);
1247 
1248 				pp = proc_realparent(p);
1249 				proc_reparent(p, pp, false);
1250 				if (pp == initproc)
1251 					p->p_sigparent = SIGCHLD;
1252 				CTR3(KTR_PTRACE,
1253 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1254 				    p->p_pid, pp->p_pid, data);
1255 			} else {
1256 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1257 				    p->p_pid, data);
1258 			}
1259 
1260 			p->p_ptevents = 0;
1261 			FOREACH_THREAD_IN_PROC(p, td3) {
1262 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1263 					sigqueue_delete(&td3->td_sigqueue,
1264 					    SIGSTOP);
1265 				}
1266 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1267 				    TDB_SUSPEND);
1268 			}
1269 
1270 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1271 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1272 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1273 			}
1274 
1275 			/* should we send SIGCHLD? */
1276 			/* childproc_continued(p); */
1277 			break;
1278 		}
1279 
1280 		sx_xunlock(&proctree_lock);
1281 		proctree_locked = false;
1282 
1283 	sendsig:
1284 		MPASS(!proctree_locked);
1285 
1286 		/*
1287 		 * Clear the pending event for the thread that just
1288 		 * reported its event (p_xthread).  This may not be
1289 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1290 		 * the debugger is resuming a different thread.
1291 		 *
1292 		 * Deliver any pending signal via the reporting thread.
1293 		 */
1294 		MPASS(p->p_xthread != NULL);
1295 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1296 		p->p_xthread->td_xsig = data;
1297 		p->p_xthread = NULL;
1298 		p->p_xsig = data;
1299 
1300 		/*
1301 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1302 		 * always works immediately, even if another thread is
1303 		 * unsuspended first and attempts to handle a
1304 		 * different signal or if the POSIX.1b style signal
1305 		 * queue cannot accommodate any new signals.
1306 		 */
1307 		if (data == SIGKILL)
1308 			proc_wkilled(p);
1309 
1310 		/*
1311 		 * Unsuspend all threads.  To leave a thread
1312 		 * suspended, use PT_SUSPEND to suspend it before
1313 		 * continuing the process.
1314 		 */
1315 		ptrace_unsuspend(p);
1316 		break;
1317 
1318 	case PT_WRITE_I:
1319 	case PT_WRITE_D:
1320 		td2->td_dbgflags |= TDB_USERWR;
1321 		PROC_UNLOCK(p);
1322 		error = 0;
1323 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1324 		    sizeof(int)) != sizeof(int))
1325 			error = ENOMEM;
1326 		else
1327 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1328 			    p->p_pid, addr, data);
1329 		PROC_LOCK(p);
1330 		break;
1331 
1332 	case PT_READ_I:
1333 	case PT_READ_D:
1334 		PROC_UNLOCK(p);
1335 		error = tmp = 0;
1336 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1337 		    sizeof(int)) != sizeof(int))
1338 			error = ENOMEM;
1339 		else
1340 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1341 			    p->p_pid, addr, tmp);
1342 		td->td_retval[0] = tmp;
1343 		PROC_LOCK(p);
1344 		break;
1345 
1346 	case PT_IO:
1347 		piod = addr;
1348 		iov.iov_base = piod->piod_addr;
1349 		iov.iov_len = piod->piod_len;
1350 		uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1351 		uio.uio_resid = piod->piod_len;
1352 		uio.uio_iov = &iov;
1353 		uio.uio_iovcnt = 1;
1354 		uio.uio_segflg = UIO_USERSPACE;
1355 		uio.uio_td = td;
1356 		switch (piod->piod_op) {
1357 		case PIOD_READ_D:
1358 		case PIOD_READ_I:
1359 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1360 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1361 			uio.uio_rw = UIO_READ;
1362 			break;
1363 		case PIOD_WRITE_D:
1364 		case PIOD_WRITE_I:
1365 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1366 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1367 			td2->td_dbgflags |= TDB_USERWR;
1368 			uio.uio_rw = UIO_WRITE;
1369 			break;
1370 		default:
1371 			error = EINVAL;
1372 			goto out;
1373 		}
1374 		PROC_UNLOCK(p);
1375 		error = proc_rwmem(p, &uio);
1376 		piod->piod_len -= uio.uio_resid;
1377 		PROC_LOCK(p);
1378 		break;
1379 
1380 	case PT_KILL:
1381 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1382 		data = SIGKILL;
1383 		goto sendsig;	/* in PT_CONTINUE above */
1384 
1385 	case PT_SETREGS:
1386 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1387 		    p->p_pid);
1388 		td2->td_dbgflags |= TDB_USERWR;
1389 		error = PROC_WRITE(regs, td2, addr);
1390 		break;
1391 
1392 	case PT_GETREGS:
1393 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1394 		    p->p_pid);
1395 		error = PROC_READ(regs, td2, addr);
1396 		break;
1397 
1398 	case PT_SETFPREGS:
1399 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1400 		    p->p_pid);
1401 		td2->td_dbgflags |= TDB_USERWR;
1402 		error = PROC_WRITE(fpregs, td2, addr);
1403 		break;
1404 
1405 	case PT_GETFPREGS:
1406 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1407 		    p->p_pid);
1408 		error = PROC_READ(fpregs, td2, addr);
1409 		break;
1410 
1411 	case PT_SETDBREGS:
1412 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1413 		    p->p_pid);
1414 		td2->td_dbgflags |= TDB_USERWR;
1415 		error = PROC_WRITE(dbregs, td2, addr);
1416 		break;
1417 
1418 	case PT_GETDBREGS:
1419 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1420 		    p->p_pid);
1421 		error = PROC_READ(dbregs, td2, addr);
1422 		break;
1423 
1424 	case PT_SETREGSET:
1425 		CTR2(KTR_PTRACE, "PT_SETREGSET: tid %d (pid %d)", td2->td_tid,
1426 		    p->p_pid);
1427 		error = proc_write_regset(td2, data, addr);
1428 		break;
1429 
1430 	case PT_GETREGSET:
1431 		CTR2(KTR_PTRACE, "PT_GETREGSET: tid %d (pid %d)", td2->td_tid,
1432 		    p->p_pid);
1433 		error = proc_read_regset(td2, data, addr);
1434 		break;
1435 
1436 	case PT_LWPINFO:
1437 		if (data <= 0 || data > sizeof(*pl)) {
1438 			error = EINVAL;
1439 			break;
1440 		}
1441 		pl = addr;
1442 		bzero(pl, sizeof(*pl));
1443 		pl->pl_lwpid = td2->td_tid;
1444 		pl->pl_event = PL_EVENT_NONE;
1445 		pl->pl_flags = 0;
1446 		if (td2->td_dbgflags & TDB_XSIG) {
1447 			pl->pl_event = PL_EVENT_SIGNAL;
1448 			if (td2->td_si.si_signo != 0 &&
1449 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1450 			    + sizeof(pl->pl_siginfo)){
1451 				pl->pl_flags |= PL_FLAG_SI;
1452 				pl->pl_siginfo = td2->td_si;
1453 			}
1454 		}
1455 		if (td2->td_dbgflags & TDB_SCE)
1456 			pl->pl_flags |= PL_FLAG_SCE;
1457 		else if (td2->td_dbgflags & TDB_SCX)
1458 			pl->pl_flags |= PL_FLAG_SCX;
1459 		if (td2->td_dbgflags & TDB_EXEC)
1460 			pl->pl_flags |= PL_FLAG_EXEC;
1461 		if (td2->td_dbgflags & TDB_FORK) {
1462 			pl->pl_flags |= PL_FLAG_FORKED;
1463 			pl->pl_child_pid = td2->td_dbg_forked;
1464 			if (td2->td_dbgflags & TDB_VFORK)
1465 				pl->pl_flags |= PL_FLAG_VFORKED;
1466 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1467 		    TDB_VFORK)
1468 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1469 		if (td2->td_dbgflags & TDB_CHILD)
1470 			pl->pl_flags |= PL_FLAG_CHILD;
1471 		if (td2->td_dbgflags & TDB_BORN)
1472 			pl->pl_flags |= PL_FLAG_BORN;
1473 		if (td2->td_dbgflags & TDB_EXIT)
1474 			pl->pl_flags |= PL_FLAG_EXITED;
1475 		pl->pl_sigmask = td2->td_sigmask;
1476 		pl->pl_siglist = td2->td_siglist;
1477 		strcpy(pl->pl_tdname, td2->td_name);
1478 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1479 			pl->pl_syscall_code = td2->td_sa.code;
1480 			pl->pl_syscall_narg = td2->td_sa.callp->sy_narg;
1481 		} else {
1482 			pl->pl_syscall_code = 0;
1483 			pl->pl_syscall_narg = 0;
1484 		}
1485 		CTR6(KTR_PTRACE,
1486     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1487 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1488 		    pl->pl_child_pid, pl->pl_syscall_code);
1489 		break;
1490 
1491 	case PT_GETNUMLWPS:
1492 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1493 		    p->p_numthreads);
1494 		td->td_retval[0] = p->p_numthreads;
1495 		break;
1496 
1497 	case PT_GETLWPLIST:
1498 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1499 		    p->p_pid, data, p->p_numthreads);
1500 		if (data <= 0) {
1501 			error = EINVAL;
1502 			break;
1503 		}
1504 		num = imin(p->p_numthreads, data);
1505 		PROC_UNLOCK(p);
1506 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1507 		tmp = 0;
1508 		PROC_LOCK(p);
1509 		FOREACH_THREAD_IN_PROC(p, td2) {
1510 			if (tmp >= num)
1511 				break;
1512 			buf[tmp++] = td2->td_tid;
1513 		}
1514 		PROC_UNLOCK(p);
1515 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1516 		free(buf, M_TEMP);
1517 		if (!error)
1518 			td->td_retval[0] = tmp;
1519 		PROC_LOCK(p);
1520 		break;
1521 
1522 	case PT_VM_TIMESTAMP:
1523 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1524 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1525 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1526 		break;
1527 
1528 	case PT_VM_ENTRY:
1529 		PROC_UNLOCK(p);
1530 		error = ptrace_vm_entry(td, p, addr);
1531 		PROC_LOCK(p);
1532 		break;
1533 
1534 	case PT_COREDUMP:
1535 		pc = addr;
1536 		CTR2(KTR_PTRACE, "PT_COREDUMP: pid %d, fd %d",
1537 		    p->p_pid, pc->pc_fd);
1538 
1539 		if ((pc->pc_flags & ~(PC_COMPRESS | PC_ALL)) != 0) {
1540 			error = EINVAL;
1541 			break;
1542 		}
1543 		PROC_UNLOCK(p);
1544 
1545 		tcq = malloc(sizeof(*tcq), M_TEMP, M_WAITOK | M_ZERO);
1546 		fp = NULL;
1547 		error = fget_write(td, pc->pc_fd, &cap_write_rights, &fp);
1548 		if (error != 0)
1549 			goto coredump_cleanup_nofp;
1550 		if (fp->f_type != DTYPE_VNODE || fp->f_vnode->v_type != VREG) {
1551 			error = EPIPE;
1552 			goto coredump_cleanup;
1553 		}
1554 
1555 		PROC_LOCK(p);
1556 		error = proc_can_ptrace(td, p);
1557 		if (error != 0)
1558 			goto coredump_cleanup_locked;
1559 
1560 		td2 = ptrace_sel_coredump_thread(p);
1561 		if (td2 == NULL) {
1562 			error = EBUSY;
1563 			goto coredump_cleanup_locked;
1564 		}
1565 		KASSERT((td2->td_dbgflags & TDB_COREDUMPRQ) == 0,
1566 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1567 
1568 		tcq->tc_vp = fp->f_vnode;
1569 		tcq->tc_limit = pc->pc_limit == 0 ? OFF_MAX : pc->pc_limit;
1570 		tcq->tc_flags = SVC_PT_COREDUMP;
1571 		if ((pc->pc_flags & PC_COMPRESS) == 0)
1572 			tcq->tc_flags |= SVC_NOCOMPRESS;
1573 		if ((pc->pc_flags & PC_ALL) != 0)
1574 			tcq->tc_flags |= SVC_ALL;
1575 		td2->td_coredump = tcq;
1576 		td2->td_dbgflags |= TDB_COREDUMPRQ;
1577 		thread_run_flash(td2);
1578 		while ((td2->td_dbgflags & TDB_COREDUMPRQ) != 0)
1579 			msleep(p, &p->p_mtx, PPAUSE, "crdmp", 0);
1580 		error = tcq->tc_error;
1581 coredump_cleanup_locked:
1582 		PROC_UNLOCK(p);
1583 coredump_cleanup:
1584 		fdrop(fp, td);
1585 coredump_cleanup_nofp:
1586 		free(tcq, M_TEMP);
1587 		PROC_LOCK(p);
1588 		break;
1589 
1590 	default:
1591 #ifdef __HAVE_PTRACE_MACHDEP
1592 		if (req >= PT_FIRSTMACH) {
1593 			PROC_UNLOCK(p);
1594 			error = cpu_ptrace(td2, req, addr, data);
1595 			PROC_LOCK(p);
1596 		} else
1597 #endif
1598 			/* Unknown request. */
1599 			error = EINVAL;
1600 		break;
1601 	}
1602 out:
1603 	/* Drop our hold on this process now that the request has completed. */
1604 	_PRELE(p);
1605 fail:
1606 	if (p2_req_set) {
1607 		if ((p->p_flag2 & P2_PTRACEREQ) != 0)
1608 			wakeup(&p->p_flag2);
1609 		p->p_flag2 &= ~P2_PTRACEREQ;
1610 	}
1611 	PROC_UNLOCK(p);
1612 	if (proctree_locked)
1613 		sx_xunlock(&proctree_lock);
1614 	return (error);
1615 }
1616 #undef PROC_READ
1617 #undef PROC_WRITE
1618