xref: /freebsd/sys/kern/sys_process.c (revision 643ac419fafba89f5adda0e0ea75b538727453fb)
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 	 * Make sure that the process' vmspace remains live.
340 	 */
341 	if (p != curproc)
342 		PROC_ASSERT_HELD(p);
343 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
344 
345 	/*
346 	 * The map we want...
347 	 */
348 	map = &p->p_vmspace->vm_map;
349 
350 	/*
351 	 * If we are writing, then we request vm_fault() to create a private
352 	 * copy of each page.  Since these copies will not be writeable by the
353 	 * process, we must explicity request that they be dirtied.
354 	 */
355 	writing = uio->uio_rw == UIO_WRITE;
356 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
357 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
358 
359 	/*
360 	 * Only map in one page at a time.  We don't have to, but it
361 	 * makes things easier.  This way is trivial - right?
362 	 */
363 	do {
364 		vm_offset_t uva;
365 		u_int len;
366 		vm_page_t m;
367 
368 		uva = (vm_offset_t)uio->uio_offset;
369 
370 		/*
371 		 * Get the page number of this segment.
372 		 */
373 		pageno = trunc_page(uva);
374 		page_offset = uva - pageno;
375 
376 		/*
377 		 * How many bytes to copy
378 		 */
379 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
380 
381 		/*
382 		 * Fault and hold the page on behalf of the process.
383 		 */
384 		error = vm_fault(map, pageno, reqprot, fault_flags, &m);
385 		if (error != KERN_SUCCESS) {
386 			if (error == KERN_RESOURCE_SHORTAGE)
387 				error = ENOMEM;
388 			else
389 				error = EFAULT;
390 			break;
391 		}
392 
393 		/*
394 		 * Now do the i/o move.
395 		 */
396 		error = uiomove_fromphys(&m, page_offset, len, uio);
397 
398 		/* Make the I-cache coherent for breakpoints. */
399 		if (writing && error == 0) {
400 			vm_map_lock_read(map);
401 			if (vm_map_check_protection(map, pageno, pageno +
402 			    PAGE_SIZE, VM_PROT_EXECUTE))
403 				vm_sync_icache(map, uva, len);
404 			vm_map_unlock_read(map);
405 		}
406 
407 		/*
408 		 * Release the page.
409 		 */
410 		vm_page_unwire(m, PQ_ACTIVE);
411 
412 	} while (error == 0 && uio->uio_resid > 0);
413 
414 	return (error);
415 }
416 
417 static ssize_t
418 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
419     size_t len, enum uio_rw rw)
420 {
421 	struct iovec iov;
422 	struct uio uio;
423 	ssize_t slen;
424 
425 	MPASS(len < SSIZE_MAX);
426 	slen = (ssize_t)len;
427 
428 	iov.iov_base = (caddr_t)buf;
429 	iov.iov_len = len;
430 	uio.uio_iov = &iov;
431 	uio.uio_iovcnt = 1;
432 	uio.uio_offset = va;
433 	uio.uio_resid = slen;
434 	uio.uio_segflg = UIO_SYSSPACE;
435 	uio.uio_rw = rw;
436 	uio.uio_td = td;
437 	proc_rwmem(p, &uio);
438 	if (uio.uio_resid == slen)
439 		return (-1);
440 	return (slen - uio.uio_resid);
441 }
442 
443 ssize_t
444 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
445     size_t len)
446 {
447 
448 	return (proc_iop(td, p, va, buf, len, UIO_READ));
449 }
450 
451 ssize_t
452 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
453     size_t len)
454 {
455 
456 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
457 }
458 
459 static int
460 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
461 {
462 	struct vattr vattr;
463 	vm_map_t map;
464 	vm_map_entry_t entry;
465 	vm_object_t obj, tobj, lobj;
466 	struct vmspace *vm;
467 	struct vnode *vp;
468 	char *freepath, *fullpath;
469 	u_int pathlen;
470 	int error, index;
471 
472 	error = 0;
473 	obj = NULL;
474 
475 	vm = vmspace_acquire_ref(p);
476 	map = &vm->vm_map;
477 	vm_map_lock_read(map);
478 
479 	do {
480 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
481 		    ("Submap in map header"));
482 		index = 0;
483 		VM_MAP_ENTRY_FOREACH(entry, map) {
484 			if (index >= pve->pve_entry &&
485 			    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
486 				break;
487 			index++;
488 		}
489 		if (index < pve->pve_entry) {
490 			error = EINVAL;
491 			break;
492 		}
493 		if (entry == &map->header) {
494 			error = ENOENT;
495 			break;
496 		}
497 
498 		/* We got an entry. */
499 		pve->pve_entry = index + 1;
500 		pve->pve_timestamp = map->timestamp;
501 		pve->pve_start = entry->start;
502 		pve->pve_end = entry->end - 1;
503 		pve->pve_offset = entry->offset;
504 		pve->pve_prot = entry->protection;
505 
506 		/* Backing object's path needed? */
507 		if (pve->pve_pathlen == 0)
508 			break;
509 
510 		pathlen = pve->pve_pathlen;
511 		pve->pve_pathlen = 0;
512 
513 		obj = entry->object.vm_object;
514 		if (obj != NULL)
515 			VM_OBJECT_RLOCK(obj);
516 	} while (0);
517 
518 	vm_map_unlock_read(map);
519 
520 	pve->pve_fsid = VNOVAL;
521 	pve->pve_fileid = VNOVAL;
522 
523 	if (error == 0 && obj != NULL) {
524 		lobj = obj;
525 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
526 			if (tobj != obj)
527 				VM_OBJECT_RLOCK(tobj);
528 			if (lobj != obj)
529 				VM_OBJECT_RUNLOCK(lobj);
530 			lobj = tobj;
531 			pve->pve_offset += tobj->backing_object_offset;
532 		}
533 		vp = vm_object_vnode(lobj);
534 		if (vp != NULL)
535 			vref(vp);
536 		if (lobj != obj)
537 			VM_OBJECT_RUNLOCK(lobj);
538 		VM_OBJECT_RUNLOCK(obj);
539 
540 		if (vp != NULL) {
541 			freepath = NULL;
542 			fullpath = NULL;
543 			vn_fullpath(vp, &fullpath, &freepath);
544 			vn_lock(vp, LK_SHARED | LK_RETRY);
545 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
546 				pve->pve_fileid = vattr.va_fileid;
547 				pve->pve_fsid = vattr.va_fsid;
548 			}
549 			vput(vp);
550 
551 			if (fullpath != NULL) {
552 				pve->pve_pathlen = strlen(fullpath) + 1;
553 				if (pve->pve_pathlen <= pathlen) {
554 					error = copyout(fullpath, pve->pve_path,
555 					    pve->pve_pathlen);
556 				} else
557 					error = ENAMETOOLONG;
558 			}
559 			if (freepath != NULL)
560 				free(freepath, M_TEMP);
561 		}
562 	}
563 	vmspace_free(vm);
564 	if (error == 0)
565 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
566 		    p->p_pid, pve->pve_entry, pve->pve_start);
567 
568 	return (error);
569 }
570 
571 /*
572  * Process debugging system call.
573  */
574 #ifndef _SYS_SYSPROTO_H_
575 struct ptrace_args {
576 	int	req;
577 	pid_t	pid;
578 	caddr_t	addr;
579 	int	data;
580 };
581 #endif
582 
583 int
584 sys_ptrace(struct thread *td, struct ptrace_args *uap)
585 {
586 	/*
587 	 * XXX this obfuscation is to reduce stack usage, but the register
588 	 * structs may be too large to put on the stack anyway.
589 	 */
590 	union {
591 		struct ptrace_io_desc piod;
592 		struct ptrace_lwpinfo pl;
593 		struct ptrace_vm_entry pve;
594 		struct ptrace_coredump pc;
595 		struct dbreg dbreg;
596 		struct fpreg fpreg;
597 		struct reg reg;
598 		struct iovec vec;
599 		syscallarg_t args[nitems(td->td_sa.args)];
600 		struct ptrace_sc_ret psr;
601 		int ptevents;
602 	} r;
603 	void *addr;
604 	int error;
605 
606 	if (!allow_ptrace)
607 		return (ENOSYS);
608 	error = 0;
609 
610 	AUDIT_ARG_PID(uap->pid);
611 	AUDIT_ARG_CMD(uap->req);
612 	AUDIT_ARG_VALUE(uap->data);
613 	addr = &r;
614 	switch (uap->req) {
615 	case PT_GET_EVENT_MASK:
616 	case PT_LWPINFO:
617 	case PT_GET_SC_ARGS:
618 	case PT_GET_SC_RET:
619 		break;
620 	case PT_GETREGS:
621 		bzero(&r.reg, sizeof(r.reg));
622 		break;
623 	case PT_GETFPREGS:
624 		bzero(&r.fpreg, sizeof(r.fpreg));
625 		break;
626 	case PT_GETDBREGS:
627 		bzero(&r.dbreg, sizeof(r.dbreg));
628 		break;
629 	case PT_GETREGSET:
630 	case PT_SETREGSET:
631 		error = copyin(uap->addr, &r.vec, sizeof(r.vec));
632 		break;
633 	case PT_SETREGS:
634 		error = copyin(uap->addr, &r.reg, sizeof(r.reg));
635 		break;
636 	case PT_SETFPREGS:
637 		error = copyin(uap->addr, &r.fpreg, sizeof(r.fpreg));
638 		break;
639 	case PT_SETDBREGS:
640 		error = copyin(uap->addr, &r.dbreg, sizeof(r.dbreg));
641 		break;
642 	case PT_SET_EVENT_MASK:
643 		if (uap->data != sizeof(r.ptevents))
644 			error = EINVAL;
645 		else
646 			error = copyin(uap->addr, &r.ptevents, uap->data);
647 		break;
648 	case PT_IO:
649 		error = copyin(uap->addr, &r.piod, sizeof(r.piod));
650 		break;
651 	case PT_VM_ENTRY:
652 		error = copyin(uap->addr, &r.pve, sizeof(r.pve));
653 		break;
654 	case PT_COREDUMP:
655 		if (uap->data != sizeof(r.pc))
656 			error = EINVAL;
657 		else
658 			error = copyin(uap->addr, &r.pc, uap->data);
659 		break;
660 	default:
661 		addr = uap->addr;
662 		break;
663 	}
664 	if (error)
665 		return (error);
666 
667 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
668 	if (error)
669 		return (error);
670 
671 	switch (uap->req) {
672 	case PT_VM_ENTRY:
673 		error = copyout(&r.pve, uap->addr, sizeof(r.pve));
674 		break;
675 	case PT_IO:
676 		error = copyout(&r.piod, uap->addr, sizeof(r.piod));
677 		break;
678 	case PT_GETREGS:
679 		error = copyout(&r.reg, uap->addr, sizeof(r.reg));
680 		break;
681 	case PT_GETFPREGS:
682 		error = copyout(&r.fpreg, uap->addr, sizeof(r.fpreg));
683 		break;
684 	case PT_GETDBREGS:
685 		error = copyout(&r.dbreg, uap->addr, sizeof(r.dbreg));
686 		break;
687 	case PT_GETREGSET:
688 		error = copyout(&r.vec, uap->addr, sizeof(r.vec));
689 		break;
690 	case PT_GET_EVENT_MASK:
691 		/* NB: The size in uap->data is validated in kern_ptrace(). */
692 		error = copyout(&r.ptevents, uap->addr, uap->data);
693 		break;
694 	case PT_LWPINFO:
695 		/* NB: The size in uap->data is validated in kern_ptrace(). */
696 		error = copyout(&r.pl, uap->addr, uap->data);
697 		break;
698 	case PT_GET_SC_ARGS:
699 		error = copyout(r.args, uap->addr, MIN(uap->data,
700 		    sizeof(r.args)));
701 		break;
702 	case PT_GET_SC_RET:
703 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
704 		    sizeof(r.psr)));
705 		break;
706 	}
707 
708 	return (error);
709 }
710 
711 #ifdef COMPAT_FREEBSD32
712 /*
713  *   PROC_READ(regs, td2, addr);
714  * becomes either:
715  *   proc_read_regs(td2, addr);
716  * or
717  *   proc_read_regs32(td2, addr);
718  * .. except this is done at runtime.  There is an additional
719  * complication in that PROC_WRITE disallows 32 bit consumers
720  * from writing to 64 bit address space targets.
721  */
722 #define	PROC_READ(w, t, a)	wrap32 ? \
723 	proc_read_ ## w ## 32(t, a) : \
724 	proc_read_ ## w (t, a)
725 #define	PROC_WRITE(w, t, a)	wrap32 ? \
726 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
727 	proc_write_ ## w (t, a)
728 #else
729 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
730 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
731 #endif
732 
733 void
734 proc_set_traced(struct proc *p, bool stop)
735 {
736 
737 	sx_assert(&proctree_lock, SX_XLOCKED);
738 	PROC_LOCK_ASSERT(p, MA_OWNED);
739 	p->p_flag |= P_TRACED;
740 	if (stop)
741 		p->p_flag2 |= P2_PTRACE_FSTP;
742 	p->p_ptevents = PTRACE_DEFAULT;
743 }
744 
745 void
746 ptrace_unsuspend(struct proc *p)
747 {
748 	PROC_LOCK_ASSERT(p, MA_OWNED);
749 
750 	PROC_SLOCK(p);
751 	p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
752 	thread_unsuspend(p);
753 	PROC_SUNLOCK(p);
754 	itimer_proc_continue(p);
755 	kqtimer_proc_continue(p);
756 }
757 
758 static int
759 proc_can_ptrace(struct thread *td, struct proc *p)
760 {
761 	int error;
762 
763 	PROC_LOCK_ASSERT(p, MA_OWNED);
764 
765 	if ((p->p_flag & P_WEXIT) != 0)
766 		return (ESRCH);
767 
768 	if ((error = p_cansee(td, p)) != 0)
769 		return (error);
770 	if ((error = p_candebug(td, p)) != 0)
771 		return (error);
772 
773 	/* not being traced... */
774 	if ((p->p_flag & P_TRACED) == 0)
775 		return (EPERM);
776 
777 	/* not being traced by YOU */
778 	if (p->p_pptr != td->td_proc)
779 		return (EBUSY);
780 
781 	/* not currently stopped */
782 	if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
783 	    p->p_suspcount != p->p_numthreads  ||
784 	    (p->p_flag & P_WAITED) == 0)
785 		return (EBUSY);
786 
787 	return (0);
788 }
789 
790 static struct thread *
791 ptrace_sel_coredump_thread(struct proc *p)
792 {
793 	struct thread *td2;
794 
795 	PROC_LOCK_ASSERT(p, MA_OWNED);
796 	MPASS((p->p_flag & P_STOPPED_TRACE) != 0);
797 
798 	FOREACH_THREAD_IN_PROC(p, td2) {
799 		if ((td2->td_dbgflags & TDB_SSWITCH) != 0)
800 			return (td2);
801 	}
802 	return (NULL);
803 }
804 
805 int
806 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
807 {
808 	struct iovec iov;
809 	struct uio uio;
810 	struct proc *curp, *p, *pp;
811 	struct thread *td2 = NULL, *td3;
812 	struct ptrace_io_desc *piod = NULL;
813 	struct ptrace_lwpinfo *pl;
814 	struct ptrace_sc_ret *psr;
815 	struct file *fp;
816 	struct ptrace_coredump *pc;
817 	struct thr_coredump_req *tcq;
818 	int error, num, tmp;
819 	lwpid_t tid = 0, *buf;
820 #ifdef COMPAT_FREEBSD32
821 	int wrap32 = 0, safe = 0;
822 #endif
823 	bool proctree_locked, p2_req_set;
824 
825 	curp = td->td_proc;
826 	proctree_locked = false;
827 	p2_req_set = false;
828 
829 	/* Lock proctree before locking the process. */
830 	switch (req) {
831 	case PT_TRACE_ME:
832 	case PT_ATTACH:
833 	case PT_STEP:
834 	case PT_CONTINUE:
835 	case PT_TO_SCE:
836 	case PT_TO_SCX:
837 	case PT_SYSCALL:
838 	case PT_FOLLOW_FORK:
839 	case PT_LWP_EVENTS:
840 	case PT_GET_EVENT_MASK:
841 	case PT_SET_EVENT_MASK:
842 	case PT_DETACH:
843 	case PT_GET_SC_ARGS:
844 		sx_xlock(&proctree_lock);
845 		proctree_locked = true;
846 		break;
847 	default:
848 		break;
849 	}
850 
851 	if (req == PT_TRACE_ME) {
852 		p = td->td_proc;
853 		PROC_LOCK(p);
854 	} else {
855 		if (pid <= PID_MAX) {
856 			if ((p = pfind(pid)) == NULL) {
857 				if (proctree_locked)
858 					sx_xunlock(&proctree_lock);
859 				return (ESRCH);
860 			}
861 		} else {
862 			td2 = tdfind(pid, -1);
863 			if (td2 == NULL) {
864 				if (proctree_locked)
865 					sx_xunlock(&proctree_lock);
866 				return (ESRCH);
867 			}
868 			p = td2->td_proc;
869 			tid = pid;
870 			pid = p->p_pid;
871 		}
872 	}
873 	AUDIT_ARG_PROCESS(p);
874 
875 	if ((p->p_flag & P_WEXIT) != 0) {
876 		error = ESRCH;
877 		goto fail;
878 	}
879 	if ((error = p_cansee(td, p)) != 0)
880 		goto fail;
881 
882 	if ((error = p_candebug(td, p)) != 0)
883 		goto fail;
884 
885 	/*
886 	 * System processes can't be debugged.
887 	 */
888 	if ((p->p_flag & P_SYSTEM) != 0) {
889 		error = EINVAL;
890 		goto fail;
891 	}
892 
893 	if (tid == 0) {
894 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
895 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
896 			td2 = p->p_xthread;
897 		} else {
898 			td2 = FIRST_THREAD_IN_PROC(p);
899 		}
900 		tid = td2->td_tid;
901 	}
902 
903 #ifdef COMPAT_FREEBSD32
904 	/*
905 	 * Test if we're a 32 bit client and what the target is.
906 	 * Set the wrap controls accordingly.
907 	 */
908 	if (SV_CURPROC_FLAG(SV_ILP32)) {
909 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
910 			safe = 1;
911 		wrap32 = 1;
912 	}
913 #endif
914 	/*
915 	 * Permissions check
916 	 */
917 	switch (req) {
918 	case PT_TRACE_ME:
919 		/*
920 		 * Always legal, when there is a parent process which
921 		 * could trace us.  Otherwise, reject.
922 		 */
923 		if ((p->p_flag & P_TRACED) != 0) {
924 			error = EBUSY;
925 			goto fail;
926 		}
927 		if (p->p_pptr == initproc) {
928 			error = EPERM;
929 			goto fail;
930 		}
931 		break;
932 
933 	case PT_ATTACH:
934 		/* Self */
935 		if (p == td->td_proc) {
936 			error = EINVAL;
937 			goto fail;
938 		}
939 
940 		/* Already traced */
941 		if (p->p_flag & P_TRACED) {
942 			error = EBUSY;
943 			goto fail;
944 		}
945 
946 		/* Can't trace an ancestor if you're being traced. */
947 		if (curp->p_flag & P_TRACED) {
948 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
949 				if (pp == p) {
950 					error = EINVAL;
951 					goto fail;
952 				}
953 			}
954 		}
955 
956 		/* OK */
957 		break;
958 
959 	case PT_CLEARSTEP:
960 		/* Allow thread to clear single step for itself */
961 		if (td->td_tid == tid)
962 			break;
963 
964 		/* FALLTHROUGH */
965 	default:
966 		/*
967 		 * Check for ptrace eligibility before waiting for
968 		 * holds to drain.
969 		 */
970 		error = proc_can_ptrace(td, p);
971 		if (error != 0)
972 			goto fail;
973 
974 		/*
975 		 * Block parallel ptrace requests.  Most important, do
976 		 * not allow other thread in debugger to continue the
977 		 * debuggee until coredump finished.
978 		 */
979 		while ((p->p_flag2 & P2_PTRACEREQ) != 0) {
980 			if (proctree_locked)
981 				sx_xunlock(&proctree_lock);
982 			error = msleep(&p->p_flag2, &p->p_mtx, PPAUSE | PCATCH |
983 			    (proctree_locked ? PDROP : 0), "pptrace", 0);
984 			if (proctree_locked) {
985 				sx_xlock(&proctree_lock);
986 				PROC_LOCK(p);
987 			}
988 			if (error == 0 && td2->td_proc != p)
989 				error = ESRCH;
990 			if (error == 0)
991 				error = proc_can_ptrace(td, p);
992 			if (error != 0)
993 				goto fail;
994 		}
995 
996 		/* Ok */
997 		break;
998 	}
999 
1000 	/*
1001 	 * Keep this process around and request parallel ptrace()
1002 	 * request to wait until we finish this request.
1003 	 */
1004 	MPASS((p->p_flag2 & P2_PTRACEREQ) == 0);
1005 	p->p_flag2 |= P2_PTRACEREQ;
1006 	p2_req_set = true;
1007 	_PHOLD(p);
1008 
1009 	/*
1010 	 * Actually do the requests
1011 	 */
1012 
1013 	td->td_retval[0] = 0;
1014 
1015 	switch (req) {
1016 	case PT_TRACE_ME:
1017 		/* set my trace flag and "owner" so it can read/write me */
1018 		proc_set_traced(p, false);
1019 		if (p->p_flag & P_PPWAIT)
1020 			p->p_flag |= P_PPTRACE;
1021 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
1022 		break;
1023 
1024 	case PT_ATTACH:
1025 		/* security check done above */
1026 		/*
1027 		 * It would be nice if the tracing relationship was separate
1028 		 * from the parent relationship but that would require
1029 		 * another set of links in the proc struct or for "wait"
1030 		 * to scan the entire proc table.  To make life easier,
1031 		 * we just re-parent the process we're trying to trace.
1032 		 * The old parent is remembered so we can put things back
1033 		 * on a "detach".
1034 		 */
1035 		proc_set_traced(p, true);
1036 		proc_reparent(p, td->td_proc, false);
1037 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
1038 		    p->p_oppid);
1039 
1040 		sx_xunlock(&proctree_lock);
1041 		proctree_locked = false;
1042 		MPASS(p->p_xthread == NULL);
1043 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
1044 
1045 		/*
1046 		 * If already stopped due to a stop signal, clear the
1047 		 * existing stop before triggering a traced SIGSTOP.
1048 		 */
1049 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
1050 			PROC_SLOCK(p);
1051 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
1052 			thread_unsuspend(p);
1053 			PROC_SUNLOCK(p);
1054 		}
1055 
1056 		kern_psignal(p, SIGSTOP);
1057 		break;
1058 
1059 	case PT_CLEARSTEP:
1060 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
1061 		    p->p_pid);
1062 		error = ptrace_clear_single_step(td2);
1063 		break;
1064 
1065 	case PT_SETSTEP:
1066 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
1067 		    p->p_pid);
1068 		error = ptrace_single_step(td2);
1069 		break;
1070 
1071 	case PT_SUSPEND:
1072 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
1073 		    p->p_pid);
1074 		td2->td_dbgflags |= TDB_SUSPEND;
1075 		thread_lock(td2);
1076 		td2->td_flags |= TDF_NEEDSUSPCHK;
1077 		thread_unlock(td2);
1078 		break;
1079 
1080 	case PT_RESUME:
1081 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1082 		    p->p_pid);
1083 		td2->td_dbgflags &= ~TDB_SUSPEND;
1084 		break;
1085 
1086 	case PT_FOLLOW_FORK:
1087 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1088 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1089 		    data ? "enabled" : "disabled");
1090 		if (data)
1091 			p->p_ptevents |= PTRACE_FORK;
1092 		else
1093 			p->p_ptevents &= ~PTRACE_FORK;
1094 		break;
1095 
1096 	case PT_LWP_EVENTS:
1097 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1098 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1099 		    data ? "enabled" : "disabled");
1100 		if (data)
1101 			p->p_ptevents |= PTRACE_LWP;
1102 		else
1103 			p->p_ptevents &= ~PTRACE_LWP;
1104 		break;
1105 
1106 	case PT_GET_EVENT_MASK:
1107 		if (data != sizeof(p->p_ptevents)) {
1108 			error = EINVAL;
1109 			break;
1110 		}
1111 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1112 		    p->p_ptevents);
1113 		*(int *)addr = p->p_ptevents;
1114 		break;
1115 
1116 	case PT_SET_EVENT_MASK:
1117 		if (data != sizeof(p->p_ptevents)) {
1118 			error = EINVAL;
1119 			break;
1120 		}
1121 		tmp = *(int *)addr;
1122 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1123 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1124 			error = EINVAL;
1125 			break;
1126 		}
1127 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1128 		    p->p_pid, p->p_ptevents, tmp);
1129 		p->p_ptevents = tmp;
1130 		break;
1131 
1132 	case PT_GET_SC_ARGS:
1133 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1134 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1135 #ifdef COMPAT_FREEBSD32
1136 		    || (wrap32 && !safe)
1137 #endif
1138 		    ) {
1139 			error = EINVAL;
1140 			break;
1141 		}
1142 		bzero(addr, sizeof(td2->td_sa.args));
1143 		/* See the explanation in linux_ptrace_get_syscall_info(). */
1144 		bcopy(td2->td_sa.args, addr, SV_PROC_ABI(td->td_proc) ==
1145 		    SV_ABI_LINUX ? sizeof(td2->td_sa.args) :
1146 		    td2->td_sa.callp->sy_narg * sizeof(syscallarg_t));
1147 		break;
1148 
1149 	case PT_GET_SC_RET:
1150 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1151 #ifdef COMPAT_FREEBSD32
1152 		    || (wrap32 && !safe)
1153 #endif
1154 		    ) {
1155 			error = EINVAL;
1156 			break;
1157 		}
1158 		psr = addr;
1159 		bzero(psr, sizeof(*psr));
1160 		psr->sr_error = td2->td_errno;
1161 		if (psr->sr_error == 0) {
1162 			psr->sr_retval[0] = td2->td_retval[0];
1163 			psr->sr_retval[1] = td2->td_retval[1];
1164 		}
1165 		CTR4(KTR_PTRACE,
1166 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1167 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1168 		    psr->sr_retval[1]);
1169 		break;
1170 
1171 	case PT_STEP:
1172 	case PT_CONTINUE:
1173 	case PT_TO_SCE:
1174 	case PT_TO_SCX:
1175 	case PT_SYSCALL:
1176 	case PT_DETACH:
1177 		/* Zero means do not send any signal */
1178 		if (data < 0 || data > _SIG_MAXSIG) {
1179 			error = EINVAL;
1180 			break;
1181 		}
1182 
1183 		switch (req) {
1184 		case PT_STEP:
1185 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1186 			    td2->td_tid, p->p_pid, data);
1187 			error = ptrace_single_step(td2);
1188 			if (error)
1189 				goto out;
1190 			break;
1191 		case PT_CONTINUE:
1192 		case PT_TO_SCE:
1193 		case PT_TO_SCX:
1194 		case PT_SYSCALL:
1195 			if (addr != (void *)1) {
1196 				error = ptrace_set_pc(td2,
1197 				    (u_long)(uintfptr_t)addr);
1198 				if (error)
1199 					goto out;
1200 			}
1201 			switch (req) {
1202 			case PT_TO_SCE:
1203 				p->p_ptevents |= PTRACE_SCE;
1204 				CTR4(KTR_PTRACE,
1205 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1206 				    p->p_pid, p->p_ptevents,
1207 				    (u_long)(uintfptr_t)addr, data);
1208 				break;
1209 			case PT_TO_SCX:
1210 				p->p_ptevents |= PTRACE_SCX;
1211 				CTR4(KTR_PTRACE,
1212 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1213 				    p->p_pid, p->p_ptevents,
1214 				    (u_long)(uintfptr_t)addr, data);
1215 				break;
1216 			case PT_SYSCALL:
1217 				p->p_ptevents |= PTRACE_SYSCALL;
1218 				CTR4(KTR_PTRACE,
1219 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1220 				    p->p_pid, p->p_ptevents,
1221 				    (u_long)(uintfptr_t)addr, data);
1222 				break;
1223 			case PT_CONTINUE:
1224 				CTR3(KTR_PTRACE,
1225 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1226 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1227 				break;
1228 			}
1229 			break;
1230 		case PT_DETACH:
1231 			/*
1232 			 * Clear P_TRACED before reparenting
1233 			 * a detached process back to its original
1234 			 * parent.  Otherwise the debugee will be set
1235 			 * as an orphan of the debugger.
1236 			 */
1237 			p->p_flag &= ~(P_TRACED | P_WAITED);
1238 
1239 			/*
1240 			 * Reset the process parent.
1241 			 */
1242 			if (p->p_oppid != p->p_pptr->p_pid) {
1243 				PROC_LOCK(p->p_pptr);
1244 				sigqueue_take(p->p_ksi);
1245 				PROC_UNLOCK(p->p_pptr);
1246 
1247 				pp = proc_realparent(p);
1248 				proc_reparent(p, pp, false);
1249 				if (pp == initproc)
1250 					p->p_sigparent = SIGCHLD;
1251 				CTR3(KTR_PTRACE,
1252 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1253 				    p->p_pid, pp->p_pid, data);
1254 			} else {
1255 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1256 				    p->p_pid, data);
1257 			}
1258 
1259 			p->p_ptevents = 0;
1260 			FOREACH_THREAD_IN_PROC(p, td3) {
1261 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1262 					sigqueue_delete(&td3->td_sigqueue,
1263 					    SIGSTOP);
1264 				}
1265 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1266 				    TDB_SUSPEND);
1267 			}
1268 
1269 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1270 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1271 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1272 			}
1273 
1274 			/* should we send SIGCHLD? */
1275 			/* childproc_continued(p); */
1276 			break;
1277 		}
1278 
1279 		sx_xunlock(&proctree_lock);
1280 		proctree_locked = false;
1281 
1282 	sendsig:
1283 		MPASS(!proctree_locked);
1284 
1285 		/*
1286 		 * Clear the pending event for the thread that just
1287 		 * reported its event (p_xthread).  This may not be
1288 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1289 		 * the debugger is resuming a different thread.
1290 		 *
1291 		 * Deliver any pending signal via the reporting thread.
1292 		 */
1293 		MPASS(p->p_xthread != NULL);
1294 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1295 		p->p_xthread->td_xsig = data;
1296 		p->p_xthread = NULL;
1297 		p->p_xsig = data;
1298 
1299 		/*
1300 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1301 		 * always works immediately, even if another thread is
1302 		 * unsuspended first and attempts to handle a
1303 		 * different signal or if the POSIX.1b style signal
1304 		 * queue cannot accommodate any new signals.
1305 		 */
1306 		if (data == SIGKILL)
1307 			proc_wkilled(p);
1308 
1309 		/*
1310 		 * Unsuspend all threads.  To leave a thread
1311 		 * suspended, use PT_SUSPEND to suspend it before
1312 		 * continuing the process.
1313 		 */
1314 		ptrace_unsuspend(p);
1315 		break;
1316 
1317 	case PT_WRITE_I:
1318 	case PT_WRITE_D:
1319 		td2->td_dbgflags |= TDB_USERWR;
1320 		PROC_UNLOCK(p);
1321 		error = 0;
1322 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1323 		    sizeof(int)) != sizeof(int))
1324 			error = ENOMEM;
1325 		else
1326 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1327 			    p->p_pid, addr, data);
1328 		PROC_LOCK(p);
1329 		break;
1330 
1331 	case PT_READ_I:
1332 	case PT_READ_D:
1333 		PROC_UNLOCK(p);
1334 		error = tmp = 0;
1335 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1336 		    sizeof(int)) != sizeof(int))
1337 			error = ENOMEM;
1338 		else
1339 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1340 			    p->p_pid, addr, tmp);
1341 		td->td_retval[0] = tmp;
1342 		PROC_LOCK(p);
1343 		break;
1344 
1345 	case PT_IO:
1346 		piod = addr;
1347 		iov.iov_base = piod->piod_addr;
1348 		iov.iov_len = piod->piod_len;
1349 		uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1350 		uio.uio_resid = piod->piod_len;
1351 		uio.uio_iov = &iov;
1352 		uio.uio_iovcnt = 1;
1353 		uio.uio_segflg = UIO_USERSPACE;
1354 		uio.uio_td = td;
1355 		switch (piod->piod_op) {
1356 		case PIOD_READ_D:
1357 		case PIOD_READ_I:
1358 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1359 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1360 			uio.uio_rw = UIO_READ;
1361 			break;
1362 		case PIOD_WRITE_D:
1363 		case PIOD_WRITE_I:
1364 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1365 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1366 			td2->td_dbgflags |= TDB_USERWR;
1367 			uio.uio_rw = UIO_WRITE;
1368 			break;
1369 		default:
1370 			error = EINVAL;
1371 			goto out;
1372 		}
1373 		PROC_UNLOCK(p);
1374 		error = proc_rwmem(p, &uio);
1375 		piod->piod_len -= uio.uio_resid;
1376 		PROC_LOCK(p);
1377 		break;
1378 
1379 	case PT_KILL:
1380 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1381 		data = SIGKILL;
1382 		goto sendsig;	/* in PT_CONTINUE above */
1383 
1384 	case PT_SETREGS:
1385 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1386 		    p->p_pid);
1387 		td2->td_dbgflags |= TDB_USERWR;
1388 		error = PROC_WRITE(regs, td2, addr);
1389 		break;
1390 
1391 	case PT_GETREGS:
1392 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1393 		    p->p_pid);
1394 		error = PROC_READ(regs, td2, addr);
1395 		break;
1396 
1397 	case PT_SETFPREGS:
1398 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1399 		    p->p_pid);
1400 		td2->td_dbgflags |= TDB_USERWR;
1401 		error = PROC_WRITE(fpregs, td2, addr);
1402 		break;
1403 
1404 	case PT_GETFPREGS:
1405 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1406 		    p->p_pid);
1407 		error = PROC_READ(fpregs, td2, addr);
1408 		break;
1409 
1410 	case PT_SETDBREGS:
1411 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1412 		    p->p_pid);
1413 		td2->td_dbgflags |= TDB_USERWR;
1414 		error = PROC_WRITE(dbregs, td2, addr);
1415 		break;
1416 
1417 	case PT_GETDBREGS:
1418 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1419 		    p->p_pid);
1420 		error = PROC_READ(dbregs, td2, addr);
1421 		break;
1422 
1423 	case PT_SETREGSET:
1424 		CTR2(KTR_PTRACE, "PT_SETREGSET: tid %d (pid %d)", td2->td_tid,
1425 		    p->p_pid);
1426 		error = proc_write_regset(td2, data, addr);
1427 		break;
1428 
1429 	case PT_GETREGSET:
1430 		CTR2(KTR_PTRACE, "PT_GETREGSET: tid %d (pid %d)", td2->td_tid,
1431 		    p->p_pid);
1432 		error = proc_read_regset(td2, data, addr);
1433 		break;
1434 
1435 	case PT_LWPINFO:
1436 		if (data <= 0 || data > sizeof(*pl)) {
1437 			error = EINVAL;
1438 			break;
1439 		}
1440 		pl = addr;
1441 		bzero(pl, sizeof(*pl));
1442 		pl->pl_lwpid = td2->td_tid;
1443 		pl->pl_event = PL_EVENT_NONE;
1444 		pl->pl_flags = 0;
1445 		if (td2->td_dbgflags & TDB_XSIG) {
1446 			pl->pl_event = PL_EVENT_SIGNAL;
1447 			if (td2->td_si.si_signo != 0 &&
1448 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1449 			    + sizeof(pl->pl_siginfo)){
1450 				pl->pl_flags |= PL_FLAG_SI;
1451 				pl->pl_siginfo = td2->td_si;
1452 			}
1453 		}
1454 		if (td2->td_dbgflags & TDB_SCE)
1455 			pl->pl_flags |= PL_FLAG_SCE;
1456 		else if (td2->td_dbgflags & TDB_SCX)
1457 			pl->pl_flags |= PL_FLAG_SCX;
1458 		if (td2->td_dbgflags & TDB_EXEC)
1459 			pl->pl_flags |= PL_FLAG_EXEC;
1460 		if (td2->td_dbgflags & TDB_FORK) {
1461 			pl->pl_flags |= PL_FLAG_FORKED;
1462 			pl->pl_child_pid = td2->td_dbg_forked;
1463 			if (td2->td_dbgflags & TDB_VFORK)
1464 				pl->pl_flags |= PL_FLAG_VFORKED;
1465 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1466 		    TDB_VFORK)
1467 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1468 		if (td2->td_dbgflags & TDB_CHILD)
1469 			pl->pl_flags |= PL_FLAG_CHILD;
1470 		if (td2->td_dbgflags & TDB_BORN)
1471 			pl->pl_flags |= PL_FLAG_BORN;
1472 		if (td2->td_dbgflags & TDB_EXIT)
1473 			pl->pl_flags |= PL_FLAG_EXITED;
1474 		pl->pl_sigmask = td2->td_sigmask;
1475 		pl->pl_siglist = td2->td_siglist;
1476 		strcpy(pl->pl_tdname, td2->td_name);
1477 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1478 			pl->pl_syscall_code = td2->td_sa.code;
1479 			pl->pl_syscall_narg = td2->td_sa.callp->sy_narg;
1480 		} else {
1481 			pl->pl_syscall_code = 0;
1482 			pl->pl_syscall_narg = 0;
1483 		}
1484 		CTR6(KTR_PTRACE,
1485     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1486 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1487 		    pl->pl_child_pid, pl->pl_syscall_code);
1488 		break;
1489 
1490 	case PT_GETNUMLWPS:
1491 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1492 		    p->p_numthreads);
1493 		td->td_retval[0] = p->p_numthreads;
1494 		break;
1495 
1496 	case PT_GETLWPLIST:
1497 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1498 		    p->p_pid, data, p->p_numthreads);
1499 		if (data <= 0) {
1500 			error = EINVAL;
1501 			break;
1502 		}
1503 		num = imin(p->p_numthreads, data);
1504 		PROC_UNLOCK(p);
1505 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1506 		tmp = 0;
1507 		PROC_LOCK(p);
1508 		FOREACH_THREAD_IN_PROC(p, td2) {
1509 			if (tmp >= num)
1510 				break;
1511 			buf[tmp++] = td2->td_tid;
1512 		}
1513 		PROC_UNLOCK(p);
1514 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1515 		free(buf, M_TEMP);
1516 		if (!error)
1517 			td->td_retval[0] = tmp;
1518 		PROC_LOCK(p);
1519 		break;
1520 
1521 	case PT_VM_TIMESTAMP:
1522 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1523 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1524 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1525 		break;
1526 
1527 	case PT_VM_ENTRY:
1528 		PROC_UNLOCK(p);
1529 		error = ptrace_vm_entry(td, p, addr);
1530 		PROC_LOCK(p);
1531 		break;
1532 
1533 	case PT_COREDUMP:
1534 		pc = addr;
1535 		CTR2(KTR_PTRACE, "PT_COREDUMP: pid %d, fd %d",
1536 		    p->p_pid, pc->pc_fd);
1537 
1538 		if ((pc->pc_flags & ~(PC_COMPRESS | PC_ALL)) != 0) {
1539 			error = EINVAL;
1540 			break;
1541 		}
1542 		PROC_UNLOCK(p);
1543 
1544 		tcq = malloc(sizeof(*tcq), M_TEMP, M_WAITOK | M_ZERO);
1545 		fp = NULL;
1546 		error = fget_write(td, pc->pc_fd, &cap_write_rights, &fp);
1547 		if (error != 0)
1548 			goto coredump_cleanup_nofp;
1549 		if (fp->f_type != DTYPE_VNODE || fp->f_vnode->v_type != VREG) {
1550 			error = EPIPE;
1551 			goto coredump_cleanup;
1552 		}
1553 
1554 		PROC_LOCK(p);
1555 		error = proc_can_ptrace(td, p);
1556 		if (error != 0)
1557 			goto coredump_cleanup_locked;
1558 
1559 		td2 = ptrace_sel_coredump_thread(p);
1560 		if (td2 == NULL) {
1561 			error = EBUSY;
1562 			goto coredump_cleanup_locked;
1563 		}
1564 		KASSERT((td2->td_dbgflags & TDB_COREDUMPRQ) == 0,
1565 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1566 
1567 		tcq->tc_vp = fp->f_vnode;
1568 		tcq->tc_limit = pc->pc_limit == 0 ? OFF_MAX : pc->pc_limit;
1569 		tcq->tc_flags = SVC_PT_COREDUMP;
1570 		if ((pc->pc_flags & PC_COMPRESS) == 0)
1571 			tcq->tc_flags |= SVC_NOCOMPRESS;
1572 		if ((pc->pc_flags & PC_ALL) != 0)
1573 			tcq->tc_flags |= SVC_ALL;
1574 		td2->td_coredump = tcq;
1575 		td2->td_dbgflags |= TDB_COREDUMPRQ;
1576 		thread_run_flash(td2);
1577 		while ((td2->td_dbgflags & TDB_COREDUMPRQ) != 0)
1578 			msleep(p, &p->p_mtx, PPAUSE, "crdmp", 0);
1579 		error = tcq->tc_error;
1580 coredump_cleanup_locked:
1581 		PROC_UNLOCK(p);
1582 coredump_cleanup:
1583 		fdrop(fp, td);
1584 coredump_cleanup_nofp:
1585 		free(tcq, M_TEMP);
1586 		PROC_LOCK(p);
1587 		break;
1588 
1589 	default:
1590 #ifdef __HAVE_PTRACE_MACHDEP
1591 		if (req >= PT_FIRSTMACH) {
1592 			PROC_UNLOCK(p);
1593 			error = cpu_ptrace(td2, req, addr, data);
1594 			PROC_LOCK(p);
1595 		} else
1596 #endif
1597 			/* Unknown request. */
1598 			error = EINVAL;
1599 		break;
1600 	}
1601 out:
1602 	/* Drop our hold on this process now that the request has completed. */
1603 	_PRELE(p);
1604 fail:
1605 	if (p2_req_set) {
1606 		if ((p->p_flag2 & P2_PTRACEREQ) != 0)
1607 			wakeup(&p->p_flag2);
1608 		p->p_flag2 &= ~P2_PTRACEREQ;
1609 	}
1610 	PROC_UNLOCK(p);
1611 	if (proctree_locked)
1612 		sx_xunlock(&proctree_lock);
1613 	return (error);
1614 }
1615 #undef PROC_READ
1616 #undef PROC_WRITE
1617