xref: /freebsd/sys/kern/sys_process.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
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/syscallsubr.h>
44 #include <sys/sysent.h>
45 #include <sys/sysproto.h>
46 #include <sys/pioctl.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 
56 #include <machine/reg.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 #include <compat/freebsd32/freebsd32_signal.h>
72 
73 struct ptrace_io_desc32 {
74 	int		piod_op;
75 	uint32_t	piod_offs;
76 	uint32_t	piod_addr;
77 	uint32_t	piod_len;
78 };
79 
80 struct ptrace_sc_ret32 {
81 	uint32_t	sr_retval[2];
82 	int		sr_error;
83 };
84 
85 struct ptrace_vm_entry32 {
86 	int		pve_entry;
87 	int		pve_timestamp;
88 	uint32_t	pve_start;
89 	uint32_t	pve_end;
90 	uint32_t	pve_offset;
91 	u_int		pve_prot;
92 	u_int		pve_pathlen;
93 	int32_t		pve_fileid;
94 	u_int		pve_fsid;
95 	uint32_t	pve_path;
96 };
97 #endif
98 
99 /*
100  * Functions implemented using PROC_ACTION():
101  *
102  * proc_read_regs(proc, regs)
103  *	Get the current user-visible register set from the process
104  *	and copy it into the regs structure (<machine/reg.h>).
105  *	The process is stopped at the time read_regs is called.
106  *
107  * proc_write_regs(proc, regs)
108  *	Update the current register set from the passed in regs
109  *	structure.  Take care to avoid clobbering special CPU
110  *	registers or privileged bits in the PSL.
111  *	Depending on the architecture this may have fix-up work to do,
112  *	especially if the IAR or PCW are modified.
113  *	The process is stopped at the time write_regs is called.
114  *
115  * proc_read_fpregs, proc_write_fpregs
116  *	deal with the floating point register set, otherwise as above.
117  *
118  * proc_read_dbregs, proc_write_dbregs
119  *	deal with the processor debug register set, otherwise as above.
120  *
121  * proc_sstep(proc)
122  *	Arrange for the process to trap after executing a single instruction.
123  */
124 
125 #define	PROC_ACTION(action) do {					\
126 	int error;							\
127 									\
128 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
129 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
130 		error = EIO;						\
131 	else								\
132 		error = (action);					\
133 	return (error);							\
134 } while(0)
135 
136 int
137 proc_read_regs(struct thread *td, struct reg *regs)
138 {
139 
140 	PROC_ACTION(fill_regs(td, regs));
141 }
142 
143 int
144 proc_write_regs(struct thread *td, struct reg *regs)
145 {
146 
147 	PROC_ACTION(set_regs(td, regs));
148 }
149 
150 int
151 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
152 {
153 
154 	PROC_ACTION(fill_dbregs(td, dbregs));
155 }
156 
157 int
158 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
159 {
160 
161 	PROC_ACTION(set_dbregs(td, dbregs));
162 }
163 
164 /*
165  * Ptrace doesn't support fpregs at all, and there are no security holes
166  * or translations for fpregs, so we can just copy them.
167  */
168 int
169 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
170 {
171 
172 	PROC_ACTION(fill_fpregs(td, fpregs));
173 }
174 
175 int
176 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
177 {
178 
179 	PROC_ACTION(set_fpregs(td, fpregs));
180 }
181 
182 #ifdef COMPAT_FREEBSD32
183 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
184 int
185 proc_read_regs32(struct thread *td, struct reg32 *regs32)
186 {
187 
188 	PROC_ACTION(fill_regs32(td, regs32));
189 }
190 
191 int
192 proc_write_regs32(struct thread *td, struct reg32 *regs32)
193 {
194 
195 	PROC_ACTION(set_regs32(td, regs32));
196 }
197 
198 int
199 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
200 {
201 
202 	PROC_ACTION(fill_dbregs32(td, dbregs32));
203 }
204 
205 int
206 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
207 {
208 
209 	PROC_ACTION(set_dbregs32(td, dbregs32));
210 }
211 
212 int
213 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
214 {
215 
216 	PROC_ACTION(fill_fpregs32(td, fpregs32));
217 }
218 
219 int
220 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
221 {
222 
223 	PROC_ACTION(set_fpregs32(td, fpregs32));
224 }
225 #endif
226 
227 int
228 proc_sstep(struct thread *td)
229 {
230 
231 	PROC_ACTION(ptrace_single_step(td));
232 }
233 
234 int
235 proc_rwmem(struct proc *p, struct uio *uio)
236 {
237 	vm_map_t map;
238 	vm_offset_t pageno;		/* page number */
239 	vm_prot_t reqprot;
240 	int error, fault_flags, page_offset, writing;
241 
242 	/*
243 	 * Assert that someone has locked this vmspace.  (Should be
244 	 * curthread but we can't assert that.)  This keeps the process
245 	 * from exiting out from under us until this operation completes.
246 	 */
247 	PROC_ASSERT_HELD(p);
248 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
249 
250 	/*
251 	 * The map we want...
252 	 */
253 	map = &p->p_vmspace->vm_map;
254 
255 	/*
256 	 * If we are writing, then we request vm_fault() to create a private
257 	 * copy of each page.  Since these copies will not be writeable by the
258 	 * process, we must explicity request that they be dirtied.
259 	 */
260 	writing = uio->uio_rw == UIO_WRITE;
261 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
262 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
263 
264 	/*
265 	 * Only map in one page at a time.  We don't have to, but it
266 	 * makes things easier.  This way is trivial - right?
267 	 */
268 	do {
269 		vm_offset_t uva;
270 		u_int len;
271 		vm_page_t m;
272 
273 		uva = (vm_offset_t)uio->uio_offset;
274 
275 		/*
276 		 * Get the page number of this segment.
277 		 */
278 		pageno = trunc_page(uva);
279 		page_offset = uva - pageno;
280 
281 		/*
282 		 * How many bytes to copy
283 		 */
284 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
285 
286 		/*
287 		 * Fault and hold the page on behalf of the process.
288 		 */
289 		error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
290 		if (error != KERN_SUCCESS) {
291 			if (error == KERN_RESOURCE_SHORTAGE)
292 				error = ENOMEM;
293 			else
294 				error = EFAULT;
295 			break;
296 		}
297 
298 		/*
299 		 * Now do the i/o move.
300 		 */
301 		error = uiomove_fromphys(&m, page_offset, len, uio);
302 
303 		/* Make the I-cache coherent for breakpoints. */
304 		if (writing && error == 0) {
305 			vm_map_lock_read(map);
306 			if (vm_map_check_protection(map, pageno, pageno +
307 			    PAGE_SIZE, VM_PROT_EXECUTE))
308 				vm_sync_icache(map, uva, len);
309 			vm_map_unlock_read(map);
310 		}
311 
312 		/*
313 		 * Release the page.
314 		 */
315 		vm_page_lock(m);
316 		if (vm_page_unwire(m, PQ_ACTIVE) && m->object == NULL)
317 			vm_page_free(m);
318 		vm_page_unlock(m);
319 
320 	} while (error == 0 && uio->uio_resid > 0);
321 
322 	return (error);
323 }
324 
325 static ssize_t
326 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
327     size_t len, enum uio_rw rw)
328 {
329 	struct iovec iov;
330 	struct uio uio;
331 	ssize_t slen;
332 
333 	MPASS(len < SSIZE_MAX);
334 	slen = (ssize_t)len;
335 
336 	iov.iov_base = (caddr_t)buf;
337 	iov.iov_len = len;
338 	uio.uio_iov = &iov;
339 	uio.uio_iovcnt = 1;
340 	uio.uio_offset = va;
341 	uio.uio_resid = slen;
342 	uio.uio_segflg = UIO_SYSSPACE;
343 	uio.uio_rw = rw;
344 	uio.uio_td = td;
345 	proc_rwmem(p, &uio);
346 	if (uio.uio_resid == slen)
347 		return (-1);
348 	return (slen - uio.uio_resid);
349 }
350 
351 ssize_t
352 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
353     size_t len)
354 {
355 
356 	return (proc_iop(td, p, va, buf, len, UIO_READ));
357 }
358 
359 ssize_t
360 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
361     size_t len)
362 {
363 
364 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
365 }
366 
367 static int
368 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
369 {
370 	struct vattr vattr;
371 	vm_map_t map;
372 	vm_map_entry_t entry;
373 	vm_object_t obj, tobj, lobj;
374 	struct vmspace *vm;
375 	struct vnode *vp;
376 	char *freepath, *fullpath;
377 	u_int pathlen;
378 	int error, index;
379 
380 	error = 0;
381 	obj = NULL;
382 
383 	vm = vmspace_acquire_ref(p);
384 	map = &vm->vm_map;
385 	vm_map_lock_read(map);
386 
387 	do {
388 		entry = map->header.next;
389 		index = 0;
390 		while (index < pve->pve_entry && entry != &map->header) {
391 			entry = entry->next;
392 			index++;
393 		}
394 		if (index != pve->pve_entry) {
395 			error = EINVAL;
396 			break;
397 		}
398 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
399 		    ("Submap in map header"));
400 		while ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
401 			entry = entry->next;
402 			index++;
403 		}
404 		if (entry == &map->header) {
405 			error = ENOENT;
406 			break;
407 		}
408 
409 		/* We got an entry. */
410 		pve->pve_entry = index + 1;
411 		pve->pve_timestamp = map->timestamp;
412 		pve->pve_start = entry->start;
413 		pve->pve_end = entry->end - 1;
414 		pve->pve_offset = entry->offset;
415 		pve->pve_prot = entry->protection;
416 
417 		/* Backing object's path needed? */
418 		if (pve->pve_pathlen == 0)
419 			break;
420 
421 		pathlen = pve->pve_pathlen;
422 		pve->pve_pathlen = 0;
423 
424 		obj = entry->object.vm_object;
425 		if (obj != NULL)
426 			VM_OBJECT_RLOCK(obj);
427 	} while (0);
428 
429 	vm_map_unlock_read(map);
430 
431 	pve->pve_fsid = VNOVAL;
432 	pve->pve_fileid = VNOVAL;
433 
434 	if (error == 0 && obj != NULL) {
435 		lobj = obj;
436 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
437 			if (tobj != obj)
438 				VM_OBJECT_RLOCK(tobj);
439 			if (lobj != obj)
440 				VM_OBJECT_RUNLOCK(lobj);
441 			lobj = tobj;
442 			pve->pve_offset += tobj->backing_object_offset;
443 		}
444 		vp = vm_object_vnode(lobj);
445 		if (vp != NULL)
446 			vref(vp);
447 		if (lobj != obj)
448 			VM_OBJECT_RUNLOCK(lobj);
449 		VM_OBJECT_RUNLOCK(obj);
450 
451 		if (vp != NULL) {
452 			freepath = NULL;
453 			fullpath = NULL;
454 			vn_fullpath(td, vp, &fullpath, &freepath);
455 			vn_lock(vp, LK_SHARED | LK_RETRY);
456 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
457 				pve->pve_fileid = vattr.va_fileid;
458 				pve->pve_fsid = vattr.va_fsid;
459 			}
460 			vput(vp);
461 
462 			if (fullpath != NULL) {
463 				pve->pve_pathlen = strlen(fullpath) + 1;
464 				if (pve->pve_pathlen <= pathlen) {
465 					error = copyout(fullpath, pve->pve_path,
466 					    pve->pve_pathlen);
467 				} else
468 					error = ENAMETOOLONG;
469 			}
470 			if (freepath != NULL)
471 				free(freepath, M_TEMP);
472 		}
473 	}
474 	vmspace_free(vm);
475 	if (error == 0)
476 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
477 		    p->p_pid, pve->pve_entry, pve->pve_start);
478 
479 	return (error);
480 }
481 
482 #ifdef COMPAT_FREEBSD32
483 static int
484 ptrace_vm_entry32(struct thread *td, struct proc *p,
485     struct ptrace_vm_entry32 *pve32)
486 {
487 	struct ptrace_vm_entry pve;
488 	int error;
489 
490 	pve.pve_entry = pve32->pve_entry;
491 	pve.pve_pathlen = pve32->pve_pathlen;
492 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
493 
494 	error = ptrace_vm_entry(td, p, &pve);
495 	if (error == 0) {
496 		pve32->pve_entry = pve.pve_entry;
497 		pve32->pve_timestamp = pve.pve_timestamp;
498 		pve32->pve_start = pve.pve_start;
499 		pve32->pve_end = pve.pve_end;
500 		pve32->pve_offset = pve.pve_offset;
501 		pve32->pve_prot = pve.pve_prot;
502 		pve32->pve_fileid = pve.pve_fileid;
503 		pve32->pve_fsid = pve.pve_fsid;
504 	}
505 
506 	pve32->pve_pathlen = pve.pve_pathlen;
507 	return (error);
508 }
509 
510 static void
511 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
512     struct ptrace_lwpinfo32 *pl32)
513 {
514 
515 	bzero(pl32, sizeof(*pl32));
516 	pl32->pl_lwpid = pl->pl_lwpid;
517 	pl32->pl_event = pl->pl_event;
518 	pl32->pl_flags = pl->pl_flags;
519 	pl32->pl_sigmask = pl->pl_sigmask;
520 	pl32->pl_siglist = pl->pl_siglist;
521 	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
522 	strcpy(pl32->pl_tdname, pl->pl_tdname);
523 	pl32->pl_child_pid = pl->pl_child_pid;
524 	pl32->pl_syscall_code = pl->pl_syscall_code;
525 	pl32->pl_syscall_narg = pl->pl_syscall_narg;
526 }
527 
528 static void
529 ptrace_sc_ret_to32(const struct ptrace_sc_ret *psr,
530     struct ptrace_sc_ret32 *psr32)
531 {
532 
533 	bzero(psr32, sizeof(*psr32));
534 	psr32->sr_retval[0] = psr->sr_retval[0];
535 	psr32->sr_retval[1] = psr->sr_retval[1];
536 	psr32->sr_error = psr->sr_error;
537 }
538 #endif /* COMPAT_FREEBSD32 */
539 
540 /*
541  * Process debugging system call.
542  */
543 #ifndef _SYS_SYSPROTO_H_
544 struct ptrace_args {
545 	int	req;
546 	pid_t	pid;
547 	caddr_t	addr;
548 	int	data;
549 };
550 #endif
551 
552 #ifdef COMPAT_FREEBSD32
553 /*
554  * This CPP subterfuge is to try and reduce the number of ifdefs in
555  * the body of the code.
556  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
557  * becomes either:
558  *   copyin(uap->addr, &r.reg, sizeof r.reg);
559  * or
560  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
561  * .. except this is done at runtime.
562  */
563 #define	BZERO(a, s)		wrap32 ? \
564 	bzero(a ## 32, s ## 32) : \
565 	bzero(a, s)
566 #define	COPYIN(u, k, s)		wrap32 ? \
567 	copyin(u, k ## 32, s ## 32) : \
568 	copyin(u, k, s)
569 #define	COPYOUT(k, u, s)	wrap32 ? \
570 	copyout(k ## 32, u, s ## 32) : \
571 	copyout(k, u, s)
572 #else
573 #define	BZERO(a, s)		bzero(a, s)
574 #define	COPYIN(u, k, s)		copyin(u, k, s)
575 #define	COPYOUT(k, u, s)	copyout(k, u, s)
576 #endif
577 int
578 sys_ptrace(struct thread *td, struct ptrace_args *uap)
579 {
580 	/*
581 	 * XXX this obfuscation is to reduce stack usage, but the register
582 	 * structs may be too large to put on the stack anyway.
583 	 */
584 	union {
585 		struct ptrace_io_desc piod;
586 		struct ptrace_lwpinfo pl;
587 		struct ptrace_vm_entry pve;
588 		struct dbreg dbreg;
589 		struct fpreg fpreg;
590 		struct reg reg;
591 #ifdef COMPAT_FREEBSD32
592 		struct dbreg32 dbreg32;
593 		struct fpreg32 fpreg32;
594 		struct reg32 reg32;
595 		struct ptrace_io_desc32 piod32;
596 		struct ptrace_lwpinfo32 pl32;
597 		struct ptrace_vm_entry32 pve32;
598 #endif
599 		char args[sizeof(td->td_sa.args)];
600 		struct ptrace_sc_ret psr;
601 		int ptevents;
602 	} r;
603 	void *addr;
604 	int error = 0;
605 #ifdef COMPAT_FREEBSD32
606 	int wrap32 = 0;
607 
608 	if (SV_CURPROC_FLAG(SV_ILP32))
609 		wrap32 = 1;
610 #endif
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_SETREGS:
631 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
632 		break;
633 	case PT_SETFPREGS:
634 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
635 		break;
636 	case PT_SETDBREGS:
637 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
638 		break;
639 	case PT_SET_EVENT_MASK:
640 		if (uap->data != sizeof(r.ptevents))
641 			error = EINVAL;
642 		else
643 			error = copyin(uap->addr, &r.ptevents, uap->data);
644 		break;
645 	case PT_IO:
646 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
647 		break;
648 	case PT_VM_ENTRY:
649 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
650 		break;
651 	default:
652 		addr = uap->addr;
653 		break;
654 	}
655 	if (error)
656 		return (error);
657 
658 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
659 	if (error)
660 		return (error);
661 
662 	switch (uap->req) {
663 	case PT_VM_ENTRY:
664 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
665 		break;
666 	case PT_IO:
667 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
668 		break;
669 	case PT_GETREGS:
670 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
671 		break;
672 	case PT_GETFPREGS:
673 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
674 		break;
675 	case PT_GETDBREGS:
676 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
677 		break;
678 	case PT_GET_EVENT_MASK:
679 		/* NB: The size in uap->data is validated in kern_ptrace(). */
680 		error = copyout(&r.ptevents, uap->addr, uap->data);
681 		break;
682 	case PT_LWPINFO:
683 		/* NB: The size in uap->data is validated in kern_ptrace(). */
684 		error = copyout(&r.pl, uap->addr, uap->data);
685 		break;
686 	case PT_GET_SC_ARGS:
687 		error = copyout(r.args, uap->addr, MIN(uap->data,
688 		    sizeof(r.args)));
689 		break;
690 	case PT_GET_SC_RET:
691 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
692 		    sizeof(r.psr)));
693 		break;
694 	}
695 
696 	return (error);
697 }
698 #undef COPYIN
699 #undef COPYOUT
700 #undef BZERO
701 
702 #ifdef COMPAT_FREEBSD32
703 /*
704  *   PROC_READ(regs, td2, addr);
705  * becomes either:
706  *   proc_read_regs(td2, addr);
707  * or
708  *   proc_read_regs32(td2, addr);
709  * .. except this is done at runtime.  There is an additional
710  * complication in that PROC_WRITE disallows 32 bit consumers
711  * from writing to 64 bit address space targets.
712  */
713 #define	PROC_READ(w, t, a)	wrap32 ? \
714 	proc_read_ ## w ## 32(t, a) : \
715 	proc_read_ ## w (t, a)
716 #define	PROC_WRITE(w, t, a)	wrap32 ? \
717 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
718 	proc_write_ ## w (t, a)
719 #else
720 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
721 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
722 #endif
723 
724 void
725 proc_set_traced(struct proc *p, bool stop)
726 {
727 
728 	sx_assert(&proctree_lock, SX_XLOCKED);
729 	PROC_LOCK_ASSERT(p, MA_OWNED);
730 	p->p_flag |= P_TRACED;
731 	if (stop)
732 		p->p_flag2 |= P2_PTRACE_FSTP;
733 	p->p_ptevents = PTRACE_DEFAULT;
734 }
735 
736 int
737 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
738 {
739 	struct iovec iov;
740 	struct uio uio;
741 	struct proc *curp, *p, *pp;
742 	struct thread *td2 = NULL, *td3;
743 	struct ptrace_io_desc *piod = NULL;
744 	struct ptrace_lwpinfo *pl;
745 	struct ptrace_sc_ret *psr;
746 	int error, num, tmp;
747 	int proctree_locked = 0;
748 	lwpid_t tid = 0, *buf;
749 #ifdef COMPAT_FREEBSD32
750 	int wrap32 = 0, safe = 0;
751 	struct ptrace_io_desc32 *piod32 = NULL;
752 	struct ptrace_lwpinfo32 *pl32 = NULL;
753 	struct ptrace_sc_ret32 *psr32 = NULL;
754 	union {
755 		struct ptrace_lwpinfo pl;
756 		struct ptrace_sc_ret psr;
757 	} r;
758 #endif
759 
760 	curp = td->td_proc;
761 
762 	/* Lock proctree before locking the process. */
763 	switch (req) {
764 	case PT_TRACE_ME:
765 	case PT_ATTACH:
766 	case PT_STEP:
767 	case PT_CONTINUE:
768 	case PT_TO_SCE:
769 	case PT_TO_SCX:
770 	case PT_SYSCALL:
771 	case PT_FOLLOW_FORK:
772 	case PT_LWP_EVENTS:
773 	case PT_GET_EVENT_MASK:
774 	case PT_SET_EVENT_MASK:
775 	case PT_DETACH:
776 	case PT_GET_SC_ARGS:
777 		sx_xlock(&proctree_lock);
778 		proctree_locked = 1;
779 		break;
780 	default:
781 		break;
782 	}
783 
784 	if (req == PT_TRACE_ME) {
785 		p = td->td_proc;
786 		PROC_LOCK(p);
787 	} else {
788 		if (pid <= PID_MAX) {
789 			if ((p = pfind(pid)) == NULL) {
790 				if (proctree_locked)
791 					sx_xunlock(&proctree_lock);
792 				return (ESRCH);
793 			}
794 		} else {
795 			td2 = tdfind(pid, -1);
796 			if (td2 == NULL) {
797 				if (proctree_locked)
798 					sx_xunlock(&proctree_lock);
799 				return (ESRCH);
800 			}
801 			p = td2->td_proc;
802 			tid = pid;
803 			pid = p->p_pid;
804 		}
805 	}
806 	AUDIT_ARG_PROCESS(p);
807 
808 	if ((p->p_flag & P_WEXIT) != 0) {
809 		error = ESRCH;
810 		goto fail;
811 	}
812 	if ((error = p_cansee(td, p)) != 0)
813 		goto fail;
814 
815 	if ((error = p_candebug(td, p)) != 0)
816 		goto fail;
817 
818 	/*
819 	 * System processes can't be debugged.
820 	 */
821 	if ((p->p_flag & P_SYSTEM) != 0) {
822 		error = EINVAL;
823 		goto fail;
824 	}
825 
826 	if (tid == 0) {
827 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
828 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
829 			td2 = p->p_xthread;
830 		} else {
831 			td2 = FIRST_THREAD_IN_PROC(p);
832 		}
833 		tid = td2->td_tid;
834 	}
835 
836 #ifdef COMPAT_FREEBSD32
837 	/*
838 	 * Test if we're a 32 bit client and what the target is.
839 	 * Set the wrap controls accordingly.
840 	 */
841 	if (SV_CURPROC_FLAG(SV_ILP32)) {
842 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
843 			safe = 1;
844 		wrap32 = 1;
845 	}
846 #endif
847 	/*
848 	 * Permissions check
849 	 */
850 	switch (req) {
851 	case PT_TRACE_ME:
852 		/*
853 		 * Always legal, when there is a parent process which
854 		 * could trace us.  Otherwise, reject.
855 		 */
856 		if ((p->p_flag & P_TRACED) != 0) {
857 			error = EBUSY;
858 			goto fail;
859 		}
860 		if (p->p_pptr == initproc) {
861 			error = EPERM;
862 			goto fail;
863 		}
864 		break;
865 
866 	case PT_ATTACH:
867 		/* Self */
868 		if (p == td->td_proc) {
869 			error = EINVAL;
870 			goto fail;
871 		}
872 
873 		/* Already traced */
874 		if (p->p_flag & P_TRACED) {
875 			error = EBUSY;
876 			goto fail;
877 		}
878 
879 		/* Can't trace an ancestor if you're being traced. */
880 		if (curp->p_flag & P_TRACED) {
881 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
882 				if (pp == p) {
883 					error = EINVAL;
884 					goto fail;
885 				}
886 			}
887 		}
888 
889 
890 		/* OK */
891 		break;
892 
893 	case PT_CLEARSTEP:
894 		/* Allow thread to clear single step for itself */
895 		if (td->td_tid == tid)
896 			break;
897 
898 		/* FALLTHROUGH */
899 	default:
900 		/* not being traced... */
901 		if ((p->p_flag & P_TRACED) == 0) {
902 			error = EPERM;
903 			goto fail;
904 		}
905 
906 		/* not being traced by YOU */
907 		if (p->p_pptr != td->td_proc) {
908 			error = EBUSY;
909 			goto fail;
910 		}
911 
912 		/* not currently stopped */
913 		if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
914 		    p->p_suspcount != p->p_numthreads  ||
915 		    (p->p_flag & P_WAITED) == 0) {
916 			error = EBUSY;
917 			goto fail;
918 		}
919 
920 		/* OK */
921 		break;
922 	}
923 
924 	/* Keep this process around until we finish this request. */
925 	_PHOLD(p);
926 
927 #ifdef FIX_SSTEP
928 	/*
929 	 * Single step fixup ala procfs
930 	 */
931 	FIX_SSTEP(td2);
932 #endif
933 
934 	/*
935 	 * Actually do the requests
936 	 */
937 
938 	td->td_retval[0] = 0;
939 
940 	switch (req) {
941 	case PT_TRACE_ME:
942 		/* set my trace flag and "owner" so it can read/write me */
943 		proc_set_traced(p, false);
944 		if (p->p_flag & P_PPWAIT)
945 			p->p_flag |= P_PPTRACE;
946 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
947 		break;
948 
949 	case PT_ATTACH:
950 		/* security check done above */
951 		/*
952 		 * It would be nice if the tracing relationship was separate
953 		 * from the parent relationship but that would require
954 		 * another set of links in the proc struct or for "wait"
955 		 * to scan the entire proc table.  To make life easier,
956 		 * we just re-parent the process we're trying to trace.
957 		 * The old parent is remembered so we can put things back
958 		 * on a "detach".
959 		 */
960 		proc_set_traced(p, true);
961 		proc_reparent(p, td->td_proc, false);
962 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
963 		    p->p_oppid);
964 
965 		sx_xunlock(&proctree_lock);
966 		proctree_locked = 0;
967 		MPASS(p->p_xthread == NULL);
968 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
969 
970 		/*
971 		 * If already stopped due to a stop signal, clear the
972 		 * existing stop before triggering a traced SIGSTOP.
973 		 */
974 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
975 			PROC_SLOCK(p);
976 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
977 			thread_unsuspend(p);
978 			PROC_SUNLOCK(p);
979 		}
980 
981 		kern_psignal(p, SIGSTOP);
982 		break;
983 
984 	case PT_CLEARSTEP:
985 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
986 		    p->p_pid);
987 		error = ptrace_clear_single_step(td2);
988 		break;
989 
990 	case PT_SETSTEP:
991 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
992 		    p->p_pid);
993 		error = ptrace_single_step(td2);
994 		break;
995 
996 	case PT_SUSPEND:
997 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
998 		    p->p_pid);
999 		td2->td_dbgflags |= TDB_SUSPEND;
1000 		thread_lock(td2);
1001 		td2->td_flags |= TDF_NEEDSUSPCHK;
1002 		thread_unlock(td2);
1003 		break;
1004 
1005 	case PT_RESUME:
1006 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1007 		    p->p_pid);
1008 		td2->td_dbgflags &= ~TDB_SUSPEND;
1009 		break;
1010 
1011 	case PT_FOLLOW_FORK:
1012 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1013 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1014 		    data ? "enabled" : "disabled");
1015 		if (data)
1016 			p->p_ptevents |= PTRACE_FORK;
1017 		else
1018 			p->p_ptevents &= ~PTRACE_FORK;
1019 		break;
1020 
1021 	case PT_LWP_EVENTS:
1022 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1023 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1024 		    data ? "enabled" : "disabled");
1025 		if (data)
1026 			p->p_ptevents |= PTRACE_LWP;
1027 		else
1028 			p->p_ptevents &= ~PTRACE_LWP;
1029 		break;
1030 
1031 	case PT_GET_EVENT_MASK:
1032 		if (data != sizeof(p->p_ptevents)) {
1033 			error = EINVAL;
1034 			break;
1035 		}
1036 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1037 		    p->p_ptevents);
1038 		*(int *)addr = p->p_ptevents;
1039 		break;
1040 
1041 	case PT_SET_EVENT_MASK:
1042 		if (data != sizeof(p->p_ptevents)) {
1043 			error = EINVAL;
1044 			break;
1045 		}
1046 		tmp = *(int *)addr;
1047 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1048 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1049 			error = EINVAL;
1050 			break;
1051 		}
1052 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1053 		    p->p_pid, p->p_ptevents, tmp);
1054 		p->p_ptevents = tmp;
1055 		break;
1056 
1057 	case PT_GET_SC_ARGS:
1058 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1059 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1060 #ifdef COMPAT_FREEBSD32
1061 		    || (wrap32 && !safe)
1062 #endif
1063 		    ) {
1064 			error = EINVAL;
1065 			break;
1066 		}
1067 		bzero(addr, sizeof(td2->td_sa.args));
1068 #ifdef COMPAT_FREEBSD32
1069 		if (wrap32)
1070 			for (num = 0; num < nitems(td2->td_sa.args); num++)
1071 				((uint32_t *)addr)[num] = (uint32_t)
1072 				    td2->td_sa.args[num];
1073 		else
1074 #endif
1075 			bcopy(td2->td_sa.args, addr, td2->td_sa.narg *
1076 			    sizeof(register_t));
1077 		break;
1078 
1079 	case PT_GET_SC_RET:
1080 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1081 #ifdef COMPAT_FREEBSD32
1082 		    || (wrap32 && !safe)
1083 #endif
1084 		    ) {
1085 			error = EINVAL;
1086 			break;
1087 		}
1088 #ifdef COMPAT_FREEBSD32
1089 		if (wrap32) {
1090 			psr = &r.psr;
1091 			psr32 = addr;
1092 		} else
1093 #endif
1094 		psr = addr;
1095 		bzero(psr, sizeof(*psr));
1096 		psr->sr_error = td2->td_errno;
1097 		if (psr->sr_error == 0) {
1098 			psr->sr_retval[0] = td2->td_retval[0];
1099 			psr->sr_retval[1] = td2->td_retval[1];
1100 		}
1101 #ifdef COMPAT_FREEBSD32
1102 		if (wrap32)
1103 			ptrace_sc_ret_to32(psr, psr32);
1104 #endif
1105 		CTR4(KTR_PTRACE,
1106 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1107 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1108 		    psr->sr_retval[1]);
1109 		break;
1110 
1111 	case PT_STEP:
1112 	case PT_CONTINUE:
1113 	case PT_TO_SCE:
1114 	case PT_TO_SCX:
1115 	case PT_SYSCALL:
1116 	case PT_DETACH:
1117 		/* Zero means do not send any signal */
1118 		if (data < 0 || data > _SIG_MAXSIG) {
1119 			error = EINVAL;
1120 			break;
1121 		}
1122 
1123 		switch (req) {
1124 		case PT_STEP:
1125 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1126 			    td2->td_tid, p->p_pid, data);
1127 			error = ptrace_single_step(td2);
1128 			if (error)
1129 				goto out;
1130 			break;
1131 		case PT_CONTINUE:
1132 		case PT_TO_SCE:
1133 		case PT_TO_SCX:
1134 		case PT_SYSCALL:
1135 			if (addr != (void *)1) {
1136 				error = ptrace_set_pc(td2,
1137 				    (u_long)(uintfptr_t)addr);
1138 				if (error)
1139 					goto out;
1140 			}
1141 			switch (req) {
1142 			case PT_TO_SCE:
1143 				p->p_ptevents |= PTRACE_SCE;
1144 				CTR4(KTR_PTRACE,
1145 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1146 				    p->p_pid, p->p_ptevents,
1147 				    (u_long)(uintfptr_t)addr, data);
1148 				break;
1149 			case PT_TO_SCX:
1150 				p->p_ptevents |= PTRACE_SCX;
1151 				CTR4(KTR_PTRACE,
1152 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1153 				    p->p_pid, p->p_ptevents,
1154 				    (u_long)(uintfptr_t)addr, data);
1155 				break;
1156 			case PT_SYSCALL:
1157 				p->p_ptevents |= PTRACE_SYSCALL;
1158 				CTR4(KTR_PTRACE,
1159 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1160 				    p->p_pid, p->p_ptevents,
1161 				    (u_long)(uintfptr_t)addr, data);
1162 				break;
1163 			case PT_CONTINUE:
1164 				CTR3(KTR_PTRACE,
1165 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1166 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1167 				break;
1168 			}
1169 			break;
1170 		case PT_DETACH:
1171 			/*
1172 			 * Reset the process parent.
1173 			 *
1174 			 * NB: This clears P_TRACED before reparenting
1175 			 * a detached process back to its original
1176 			 * parent.  Otherwise the debugee will be set
1177 			 * as an orphan of the debugger.
1178 			 */
1179 			p->p_flag &= ~(P_TRACED | P_WAITED);
1180 			if (p->p_oppid != p->p_pptr->p_pid) {
1181 				PROC_LOCK(p->p_pptr);
1182 				sigqueue_take(p->p_ksi);
1183 				PROC_UNLOCK(p->p_pptr);
1184 
1185 				pp = proc_realparent(p);
1186 				proc_reparent(p, pp, false);
1187 				if (pp == initproc)
1188 					p->p_sigparent = SIGCHLD;
1189 				CTR3(KTR_PTRACE,
1190 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1191 				    p->p_pid, pp->p_pid, data);
1192 			} else
1193 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1194 				    p->p_pid, data);
1195 			p->p_ptevents = 0;
1196 			FOREACH_THREAD_IN_PROC(p, td3) {
1197 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1198 					sigqueue_delete(&td3->td_sigqueue,
1199 					    SIGSTOP);
1200 				}
1201 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1202 				    TDB_SUSPEND);
1203 			}
1204 
1205 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1206 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1207 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1208 			}
1209 
1210 			/* should we send SIGCHLD? */
1211 			/* childproc_continued(p); */
1212 			break;
1213 		}
1214 
1215 		sx_xunlock(&proctree_lock);
1216 		proctree_locked = 0;
1217 
1218 	sendsig:
1219 		MPASS(proctree_locked == 0);
1220 
1221 		/*
1222 		 * Clear the pending event for the thread that just
1223 		 * reported its event (p_xthread).  This may not be
1224 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1225 		 * the debugger is resuming a different thread.
1226 		 *
1227 		 * Deliver any pending signal via the reporting thread.
1228 		 */
1229 		MPASS(p->p_xthread != NULL);
1230 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1231 		p->p_xthread->td_xsig = data;
1232 		p->p_xthread = NULL;
1233 		p->p_xsig = data;
1234 
1235 		/*
1236 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1237 		 * always works immediately, even if another thread is
1238 		 * unsuspended first and attempts to handle a
1239 		 * different signal or if the POSIX.1b style signal
1240 		 * queue cannot accommodate any new signals.
1241 		 */
1242 		if (data == SIGKILL)
1243 			proc_wkilled(p);
1244 
1245 		/*
1246 		 * Unsuspend all threads.  To leave a thread
1247 		 * suspended, use PT_SUSPEND to suspend it before
1248 		 * continuing the process.
1249 		 */
1250 		PROC_SLOCK(p);
1251 		p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
1252 		thread_unsuspend(p);
1253 		PROC_SUNLOCK(p);
1254 		break;
1255 
1256 	case PT_WRITE_I:
1257 	case PT_WRITE_D:
1258 		td2->td_dbgflags |= TDB_USERWR;
1259 		PROC_UNLOCK(p);
1260 		error = 0;
1261 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1262 		    sizeof(int)) != sizeof(int))
1263 			error = ENOMEM;
1264 		else
1265 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1266 			    p->p_pid, addr, data);
1267 		PROC_LOCK(p);
1268 		break;
1269 
1270 	case PT_READ_I:
1271 	case PT_READ_D:
1272 		PROC_UNLOCK(p);
1273 		error = tmp = 0;
1274 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1275 		    sizeof(int)) != sizeof(int))
1276 			error = ENOMEM;
1277 		else
1278 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1279 			    p->p_pid, addr, tmp);
1280 		td->td_retval[0] = tmp;
1281 		PROC_LOCK(p);
1282 		break;
1283 
1284 	case PT_IO:
1285 #ifdef COMPAT_FREEBSD32
1286 		if (wrap32) {
1287 			piod32 = addr;
1288 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1289 			iov.iov_len = piod32->piod_len;
1290 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1291 			uio.uio_resid = piod32->piod_len;
1292 		} else
1293 #endif
1294 		{
1295 			piod = addr;
1296 			iov.iov_base = piod->piod_addr;
1297 			iov.iov_len = piod->piod_len;
1298 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1299 			uio.uio_resid = piod->piod_len;
1300 		}
1301 		uio.uio_iov = &iov;
1302 		uio.uio_iovcnt = 1;
1303 		uio.uio_segflg = UIO_USERSPACE;
1304 		uio.uio_td = td;
1305 #ifdef COMPAT_FREEBSD32
1306 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1307 #else
1308 		tmp = piod->piod_op;
1309 #endif
1310 		switch (tmp) {
1311 		case PIOD_READ_D:
1312 		case PIOD_READ_I:
1313 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1314 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1315 			uio.uio_rw = UIO_READ;
1316 			break;
1317 		case PIOD_WRITE_D:
1318 		case PIOD_WRITE_I:
1319 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1320 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1321 			td2->td_dbgflags |= TDB_USERWR;
1322 			uio.uio_rw = UIO_WRITE;
1323 			break;
1324 		default:
1325 			error = EINVAL;
1326 			goto out;
1327 		}
1328 		PROC_UNLOCK(p);
1329 		error = proc_rwmem(p, &uio);
1330 #ifdef COMPAT_FREEBSD32
1331 		if (wrap32)
1332 			piod32->piod_len -= uio.uio_resid;
1333 		else
1334 #endif
1335 			piod->piod_len -= uio.uio_resid;
1336 		PROC_LOCK(p);
1337 		break;
1338 
1339 	case PT_KILL:
1340 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1341 		data = SIGKILL;
1342 		goto sendsig;	/* in PT_CONTINUE above */
1343 
1344 	case PT_SETREGS:
1345 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1346 		    p->p_pid);
1347 		td2->td_dbgflags |= TDB_USERWR;
1348 		error = PROC_WRITE(regs, td2, addr);
1349 		break;
1350 
1351 	case PT_GETREGS:
1352 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1353 		    p->p_pid);
1354 		error = PROC_READ(regs, td2, addr);
1355 		break;
1356 
1357 	case PT_SETFPREGS:
1358 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1359 		    p->p_pid);
1360 		td2->td_dbgflags |= TDB_USERWR;
1361 		error = PROC_WRITE(fpregs, td2, addr);
1362 		break;
1363 
1364 	case PT_GETFPREGS:
1365 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1366 		    p->p_pid);
1367 		error = PROC_READ(fpregs, td2, addr);
1368 		break;
1369 
1370 	case PT_SETDBREGS:
1371 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1372 		    p->p_pid);
1373 		td2->td_dbgflags |= TDB_USERWR;
1374 		error = PROC_WRITE(dbregs, td2, addr);
1375 		break;
1376 
1377 	case PT_GETDBREGS:
1378 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1379 		    p->p_pid);
1380 		error = PROC_READ(dbregs, td2, addr);
1381 		break;
1382 
1383 	case PT_LWPINFO:
1384 		if (data <= 0 ||
1385 #ifdef COMPAT_FREEBSD32
1386 		    (!wrap32 && data > sizeof(*pl)) ||
1387 		    (wrap32 && data > sizeof(*pl32))) {
1388 #else
1389 		    data > sizeof(*pl)) {
1390 #endif
1391 			error = EINVAL;
1392 			break;
1393 		}
1394 #ifdef COMPAT_FREEBSD32
1395 		if (wrap32) {
1396 			pl = &r.pl;
1397 			pl32 = addr;
1398 		} else
1399 #endif
1400 		pl = addr;
1401 		bzero(pl, sizeof(*pl));
1402 		pl->pl_lwpid = td2->td_tid;
1403 		pl->pl_event = PL_EVENT_NONE;
1404 		pl->pl_flags = 0;
1405 		if (td2->td_dbgflags & TDB_XSIG) {
1406 			pl->pl_event = PL_EVENT_SIGNAL;
1407 			if (td2->td_si.si_signo != 0 &&
1408 #ifdef COMPAT_FREEBSD32
1409 			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1410 			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1411 			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1412 			    pl_siginfo) + sizeof(struct siginfo32)))
1413 #else
1414 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1415 			    + sizeof(pl->pl_siginfo)
1416 #endif
1417 			){
1418 				pl->pl_flags |= PL_FLAG_SI;
1419 				pl->pl_siginfo = td2->td_si;
1420 			}
1421 		}
1422 		if (td2->td_dbgflags & TDB_SCE)
1423 			pl->pl_flags |= PL_FLAG_SCE;
1424 		else if (td2->td_dbgflags & TDB_SCX)
1425 			pl->pl_flags |= PL_FLAG_SCX;
1426 		if (td2->td_dbgflags & TDB_EXEC)
1427 			pl->pl_flags |= PL_FLAG_EXEC;
1428 		if (td2->td_dbgflags & TDB_FORK) {
1429 			pl->pl_flags |= PL_FLAG_FORKED;
1430 			pl->pl_child_pid = td2->td_dbg_forked;
1431 			if (td2->td_dbgflags & TDB_VFORK)
1432 				pl->pl_flags |= PL_FLAG_VFORKED;
1433 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1434 		    TDB_VFORK)
1435 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1436 		if (td2->td_dbgflags & TDB_CHILD)
1437 			pl->pl_flags |= PL_FLAG_CHILD;
1438 		if (td2->td_dbgflags & TDB_BORN)
1439 			pl->pl_flags |= PL_FLAG_BORN;
1440 		if (td2->td_dbgflags & TDB_EXIT)
1441 			pl->pl_flags |= PL_FLAG_EXITED;
1442 		pl->pl_sigmask = td2->td_sigmask;
1443 		pl->pl_siglist = td2->td_siglist;
1444 		strcpy(pl->pl_tdname, td2->td_name);
1445 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1446 			pl->pl_syscall_code = td2->td_sa.code;
1447 			pl->pl_syscall_narg = td2->td_sa.narg;
1448 		} else {
1449 			pl->pl_syscall_code = 0;
1450 			pl->pl_syscall_narg = 0;
1451 		}
1452 #ifdef COMPAT_FREEBSD32
1453 		if (wrap32)
1454 			ptrace_lwpinfo_to32(pl, pl32);
1455 #endif
1456 		CTR6(KTR_PTRACE,
1457     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1458 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1459 		    pl->pl_child_pid, pl->pl_syscall_code);
1460 		break;
1461 
1462 	case PT_GETNUMLWPS:
1463 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1464 		    p->p_numthreads);
1465 		td->td_retval[0] = p->p_numthreads;
1466 		break;
1467 
1468 	case PT_GETLWPLIST:
1469 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1470 		    p->p_pid, data, p->p_numthreads);
1471 		if (data <= 0) {
1472 			error = EINVAL;
1473 			break;
1474 		}
1475 		num = imin(p->p_numthreads, data);
1476 		PROC_UNLOCK(p);
1477 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1478 		tmp = 0;
1479 		PROC_LOCK(p);
1480 		FOREACH_THREAD_IN_PROC(p, td2) {
1481 			if (tmp >= num)
1482 				break;
1483 			buf[tmp++] = td2->td_tid;
1484 		}
1485 		PROC_UNLOCK(p);
1486 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1487 		free(buf, M_TEMP);
1488 		if (!error)
1489 			td->td_retval[0] = tmp;
1490 		PROC_LOCK(p);
1491 		break;
1492 
1493 	case PT_VM_TIMESTAMP:
1494 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1495 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1496 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1497 		break;
1498 
1499 	case PT_VM_ENTRY:
1500 		PROC_UNLOCK(p);
1501 #ifdef COMPAT_FREEBSD32
1502 		if (wrap32)
1503 			error = ptrace_vm_entry32(td, p, addr);
1504 		else
1505 #endif
1506 		error = ptrace_vm_entry(td, p, addr);
1507 		PROC_LOCK(p);
1508 		break;
1509 
1510 	default:
1511 #ifdef __HAVE_PTRACE_MACHDEP
1512 		if (req >= PT_FIRSTMACH) {
1513 			PROC_UNLOCK(p);
1514 			error = cpu_ptrace(td2, req, addr, data);
1515 			PROC_LOCK(p);
1516 		} else
1517 #endif
1518 			/* Unknown request. */
1519 			error = EINVAL;
1520 		break;
1521 	}
1522 
1523 out:
1524 	/* Drop our hold on this process now that the request has completed. */
1525 	_PRELE(p);
1526 fail:
1527 	PROC_UNLOCK(p);
1528 	if (proctree_locked)
1529 		sx_xunlock(&proctree_lock);
1530 	return (error);
1531 }
1532 #undef PROC_READ
1533 #undef PROC_WRITE
1534 
1535 /*
1536  * Stop a process because of a debugging event;
1537  * stay stopped until p->p_step is cleared
1538  * (cleared by PIOCCONT in procfs).
1539  */
1540 void
1541 stopevent(struct proc *p, unsigned int event, unsigned int val)
1542 {
1543 
1544 	PROC_LOCK_ASSERT(p, MA_OWNED);
1545 	p->p_step = 1;
1546 	CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1547 	    val);
1548 	do {
1549 		if (event != S_EXIT)
1550 			p->p_xsig = val;
1551 		p->p_xthread = NULL;
1552 		p->p_stype = event;	/* Which event caused the stop? */
1553 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1554 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1555 	} while (p->p_step);
1556 }
1557