xref: /freebsd/sys/kern/sys_process.c (revision e4f6a1bfa31a2299612bbf9dae402bb38d38b1df)
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 "opt_compat.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.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_vm_entry32 {
81 	int		pve_entry;
82 	int		pve_timestamp;
83 	uint32_t	pve_start;
84 	uint32_t	pve_end;
85 	uint32_t	pve_offset;
86 	u_int		pve_prot;
87 	u_int		pve_pathlen;
88 	int32_t		pve_fileid;
89 	u_int		pve_fsid;
90 	uint32_t	pve_path;
91 };
92 #endif
93 
94 /*
95  * Functions implemented using PROC_ACTION():
96  *
97  * proc_read_regs(proc, regs)
98  *	Get the current user-visible register set from the process
99  *	and copy it into the regs structure (<machine/reg.h>).
100  *	The process is stopped at the time read_regs is called.
101  *
102  * proc_write_regs(proc, regs)
103  *	Update the current register set from the passed in regs
104  *	structure.  Take care to avoid clobbering special CPU
105  *	registers or privileged bits in the PSL.
106  *	Depending on the architecture this may have fix-up work to do,
107  *	especially if the IAR or PCW are modified.
108  *	The process is stopped at the time write_regs is called.
109  *
110  * proc_read_fpregs, proc_write_fpregs
111  *	deal with the floating point register set, otherwise as above.
112  *
113  * proc_read_dbregs, proc_write_dbregs
114  *	deal with the processor debug register set, otherwise as above.
115  *
116  * proc_sstep(proc)
117  *	Arrange for the process to trap after executing a single instruction.
118  */
119 
120 #define	PROC_ACTION(action) do {					\
121 	int error;							\
122 									\
123 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
124 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
125 		error = EIO;						\
126 	else								\
127 		error = (action);					\
128 	return (error);							\
129 } while(0)
130 
131 int
132 proc_read_regs(struct thread *td, struct reg *regs)
133 {
134 
135 	PROC_ACTION(fill_regs(td, regs));
136 }
137 
138 int
139 proc_write_regs(struct thread *td, struct reg *regs)
140 {
141 
142 	PROC_ACTION(set_regs(td, regs));
143 }
144 
145 int
146 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
147 {
148 
149 	PROC_ACTION(fill_dbregs(td, dbregs));
150 }
151 
152 int
153 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
154 {
155 
156 	PROC_ACTION(set_dbregs(td, dbregs));
157 }
158 
159 /*
160  * Ptrace doesn't support fpregs at all, and there are no security holes
161  * or translations for fpregs, so we can just copy them.
162  */
163 int
164 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
165 {
166 
167 	PROC_ACTION(fill_fpregs(td, fpregs));
168 }
169 
170 int
171 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
172 {
173 
174 	PROC_ACTION(set_fpregs(td, fpregs));
175 }
176 
177 #ifdef COMPAT_FREEBSD32
178 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
179 int
180 proc_read_regs32(struct thread *td, struct reg32 *regs32)
181 {
182 
183 	PROC_ACTION(fill_regs32(td, regs32));
184 }
185 
186 int
187 proc_write_regs32(struct thread *td, struct reg32 *regs32)
188 {
189 
190 	PROC_ACTION(set_regs32(td, regs32));
191 }
192 
193 int
194 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
195 {
196 
197 	PROC_ACTION(fill_dbregs32(td, dbregs32));
198 }
199 
200 int
201 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
202 {
203 
204 	PROC_ACTION(set_dbregs32(td, dbregs32));
205 }
206 
207 int
208 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
209 {
210 
211 	PROC_ACTION(fill_fpregs32(td, fpregs32));
212 }
213 
214 int
215 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
216 {
217 
218 	PROC_ACTION(set_fpregs32(td, fpregs32));
219 }
220 #endif
221 
222 int
223 proc_sstep(struct thread *td)
224 {
225 
226 	PROC_ACTION(ptrace_single_step(td));
227 }
228 
229 int
230 proc_rwmem(struct proc *p, struct uio *uio)
231 {
232 	vm_map_t map;
233 	vm_offset_t pageno;		/* page number */
234 	vm_prot_t reqprot;
235 	int error, fault_flags, page_offset, writing;
236 
237 	/*
238 	 * Assert that someone has locked this vmspace.  (Should be
239 	 * curthread but we can't assert that.)  This keeps the process
240 	 * from exiting out from under us until this operation completes.
241 	 */
242 	PROC_ASSERT_HELD(p);
243 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
244 
245 	/*
246 	 * The map we want...
247 	 */
248 	map = &p->p_vmspace->vm_map;
249 
250 	/*
251 	 * If we are writing, then we request vm_fault() to create a private
252 	 * copy of each page.  Since these copies will not be writeable by the
253 	 * process, we must explicity request that they be dirtied.
254 	 */
255 	writing = uio->uio_rw == UIO_WRITE;
256 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
257 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
258 
259 	/*
260 	 * Only map in one page at a time.  We don't have to, but it
261 	 * makes things easier.  This way is trivial - right?
262 	 */
263 	do {
264 		vm_offset_t uva;
265 		u_int len;
266 		vm_page_t m;
267 
268 		uva = (vm_offset_t)uio->uio_offset;
269 
270 		/*
271 		 * Get the page number of this segment.
272 		 */
273 		pageno = trunc_page(uva);
274 		page_offset = uva - pageno;
275 
276 		/*
277 		 * How many bytes to copy
278 		 */
279 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
280 
281 		/*
282 		 * Fault and hold the page on behalf of the process.
283 		 */
284 		error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
285 		if (error != KERN_SUCCESS) {
286 			if (error == KERN_RESOURCE_SHORTAGE)
287 				error = ENOMEM;
288 			else
289 				error = EFAULT;
290 			break;
291 		}
292 
293 		/*
294 		 * Now do the i/o move.
295 		 */
296 		error = uiomove_fromphys(&m, page_offset, len, uio);
297 
298 		/* Make the I-cache coherent for breakpoints. */
299 		if (writing && error == 0) {
300 			vm_map_lock_read(map);
301 			if (vm_map_check_protection(map, pageno, pageno +
302 			    PAGE_SIZE, VM_PROT_EXECUTE))
303 				vm_sync_icache(map, uva, len);
304 			vm_map_unlock_read(map);
305 		}
306 
307 		/*
308 		 * Release the page.
309 		 */
310 		vm_page_lock(m);
311 		vm_page_unhold(m);
312 		vm_page_unlock(m);
313 
314 	} while (error == 0 && uio->uio_resid > 0);
315 
316 	return (error);
317 }
318 
319 static ssize_t
320 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
321     size_t len, enum uio_rw rw)
322 {
323 	struct iovec iov;
324 	struct uio uio;
325 	ssize_t slen;
326 	int error;
327 
328 	MPASS(len < SSIZE_MAX);
329 	slen = (ssize_t)len;
330 
331 	iov.iov_base = (caddr_t)buf;
332 	iov.iov_len = len;
333 	uio.uio_iov = &iov;
334 	uio.uio_iovcnt = 1;
335 	uio.uio_offset = va;
336 	uio.uio_resid = slen;
337 	uio.uio_segflg = UIO_SYSSPACE;
338 	uio.uio_rw = rw;
339 	uio.uio_td = td;
340 	error = proc_rwmem(p, &uio);
341 	if (uio.uio_resid == slen)
342 		return (-1);
343 	return (slen - uio.uio_resid);
344 }
345 
346 ssize_t
347 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
348     size_t len)
349 {
350 
351 	return (proc_iop(td, p, va, buf, len, UIO_READ));
352 }
353 
354 ssize_t
355 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
356     size_t len)
357 {
358 
359 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
360 }
361 
362 static int
363 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
364 {
365 	struct vattr vattr;
366 	vm_map_t map;
367 	vm_map_entry_t entry;
368 	vm_object_t obj, tobj, lobj;
369 	struct vmspace *vm;
370 	struct vnode *vp;
371 	char *freepath, *fullpath;
372 	u_int pathlen;
373 	int error, index;
374 
375 	error = 0;
376 	obj = NULL;
377 
378 	vm = vmspace_acquire_ref(p);
379 	map = &vm->vm_map;
380 	vm_map_lock_read(map);
381 
382 	do {
383 		entry = map->header.next;
384 		index = 0;
385 		while (index < pve->pve_entry && entry != &map->header) {
386 			entry = entry->next;
387 			index++;
388 		}
389 		if (index != pve->pve_entry) {
390 			error = EINVAL;
391 			break;
392 		}
393 		while (entry != &map->header &&
394 		    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
395 			entry = entry->next;
396 			index++;
397 		}
398 		if (entry == &map->header) {
399 			error = ENOENT;
400 			break;
401 		}
402 
403 		/* We got an entry. */
404 		pve->pve_entry = index + 1;
405 		pve->pve_timestamp = map->timestamp;
406 		pve->pve_start = entry->start;
407 		pve->pve_end = entry->end - 1;
408 		pve->pve_offset = entry->offset;
409 		pve->pve_prot = entry->protection;
410 
411 		/* Backing object's path needed? */
412 		if (pve->pve_pathlen == 0)
413 			break;
414 
415 		pathlen = pve->pve_pathlen;
416 		pve->pve_pathlen = 0;
417 
418 		obj = entry->object.vm_object;
419 		if (obj != NULL)
420 			VM_OBJECT_RLOCK(obj);
421 	} while (0);
422 
423 	vm_map_unlock_read(map);
424 
425 	pve->pve_fsid = VNOVAL;
426 	pve->pve_fileid = VNOVAL;
427 
428 	if (error == 0 && obj != NULL) {
429 		lobj = obj;
430 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
431 			if (tobj != obj)
432 				VM_OBJECT_RLOCK(tobj);
433 			if (lobj != obj)
434 				VM_OBJECT_RUNLOCK(lobj);
435 			lobj = tobj;
436 			pve->pve_offset += tobj->backing_object_offset;
437 		}
438 		vp = vm_object_vnode(lobj);
439 		if (vp != NULL)
440 			vref(vp);
441 		if (lobj != obj)
442 			VM_OBJECT_RUNLOCK(lobj);
443 		VM_OBJECT_RUNLOCK(obj);
444 
445 		if (vp != NULL) {
446 			freepath = NULL;
447 			fullpath = NULL;
448 			vn_fullpath(td, vp, &fullpath, &freepath);
449 			vn_lock(vp, LK_SHARED | LK_RETRY);
450 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
451 				pve->pve_fileid = vattr.va_fileid;
452 				pve->pve_fsid = vattr.va_fsid;
453 			}
454 			vput(vp);
455 
456 			if (fullpath != NULL) {
457 				pve->pve_pathlen = strlen(fullpath) + 1;
458 				if (pve->pve_pathlen <= pathlen) {
459 					error = copyout(fullpath, pve->pve_path,
460 					    pve->pve_pathlen);
461 				} else
462 					error = ENAMETOOLONG;
463 			}
464 			if (freepath != NULL)
465 				free(freepath, M_TEMP);
466 		}
467 	}
468 	vmspace_free(vm);
469 	if (error == 0)
470 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
471 		    p->p_pid, pve->pve_entry, pve->pve_start);
472 
473 	return (error);
474 }
475 
476 #ifdef COMPAT_FREEBSD32
477 static int
478 ptrace_vm_entry32(struct thread *td, struct proc *p,
479     struct ptrace_vm_entry32 *pve32)
480 {
481 	struct ptrace_vm_entry pve;
482 	int error;
483 
484 	pve.pve_entry = pve32->pve_entry;
485 	pve.pve_pathlen = pve32->pve_pathlen;
486 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
487 
488 	error = ptrace_vm_entry(td, p, &pve);
489 	if (error == 0) {
490 		pve32->pve_entry = pve.pve_entry;
491 		pve32->pve_timestamp = pve.pve_timestamp;
492 		pve32->pve_start = pve.pve_start;
493 		pve32->pve_end = pve.pve_end;
494 		pve32->pve_offset = pve.pve_offset;
495 		pve32->pve_prot = pve.pve_prot;
496 		pve32->pve_fileid = pve.pve_fileid;
497 		pve32->pve_fsid = pve.pve_fsid;
498 	}
499 
500 	pve32->pve_pathlen = pve.pve_pathlen;
501 	return (error);
502 }
503 
504 static void
505 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
506     struct ptrace_lwpinfo32 *pl32)
507 {
508 
509 	bzero(pl32, sizeof(*pl32));
510 	pl32->pl_lwpid = pl->pl_lwpid;
511 	pl32->pl_event = pl->pl_event;
512 	pl32->pl_flags = pl->pl_flags;
513 	pl32->pl_sigmask = pl->pl_sigmask;
514 	pl32->pl_siglist = pl->pl_siglist;
515 	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
516 	strcpy(pl32->pl_tdname, pl->pl_tdname);
517 	pl32->pl_child_pid = pl->pl_child_pid;
518 	pl32->pl_syscall_code = pl->pl_syscall_code;
519 	pl32->pl_syscall_narg = pl->pl_syscall_narg;
520 }
521 #endif /* COMPAT_FREEBSD32 */
522 
523 /*
524  * Process debugging system call.
525  */
526 #ifndef _SYS_SYSPROTO_H_
527 struct ptrace_args {
528 	int	req;
529 	pid_t	pid;
530 	caddr_t	addr;
531 	int	data;
532 };
533 #endif
534 
535 #ifdef COMPAT_FREEBSD32
536 /*
537  * This CPP subterfuge is to try and reduce the number of ifdefs in
538  * the body of the code.
539  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
540  * becomes either:
541  *   copyin(uap->addr, &r.reg, sizeof r.reg);
542  * or
543  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
544  * .. except this is done at runtime.
545  */
546 #define	COPYIN(u, k, s)		wrap32 ? \
547 	copyin(u, k ## 32, s ## 32) : \
548 	copyin(u, k, s)
549 #define	COPYOUT(k, u, s)	wrap32 ? \
550 	copyout(k ## 32, u, s ## 32) : \
551 	copyout(k, u, s)
552 #else
553 #define	COPYIN(u, k, s)		copyin(u, k, s)
554 #define	COPYOUT(k, u, s)	copyout(k, u, s)
555 #endif
556 int
557 sys_ptrace(struct thread *td, struct ptrace_args *uap)
558 {
559 	/*
560 	 * XXX this obfuscation is to reduce stack usage, but the register
561 	 * structs may be too large to put on the stack anyway.
562 	 */
563 	union {
564 		struct ptrace_io_desc piod;
565 		struct ptrace_lwpinfo pl;
566 		struct ptrace_vm_entry pve;
567 		struct dbreg dbreg;
568 		struct fpreg fpreg;
569 		struct reg reg;
570 #ifdef COMPAT_FREEBSD32
571 		struct dbreg32 dbreg32;
572 		struct fpreg32 fpreg32;
573 		struct reg32 reg32;
574 		struct ptrace_io_desc32 piod32;
575 		struct ptrace_lwpinfo32 pl32;
576 		struct ptrace_vm_entry32 pve32;
577 #endif
578 		char args[nitems(td->td_sa.args) * sizeof(register_t)];
579 		int ptevents;
580 	} r;
581 	void *addr;
582 	int error = 0;
583 #ifdef COMPAT_FREEBSD32
584 	int wrap32 = 0;
585 
586 	if (SV_CURPROC_FLAG(SV_ILP32))
587 		wrap32 = 1;
588 #endif
589 	AUDIT_ARG_PID(uap->pid);
590 	AUDIT_ARG_CMD(uap->req);
591 	AUDIT_ARG_VALUE(uap->data);
592 	addr = &r;
593 	switch (uap->req) {
594 	case PT_GET_EVENT_MASK:
595 	case PT_GETREGS:
596 	case PT_GETFPREGS:
597 	case PT_GETDBREGS:
598 	case PT_LWPINFO:
599 	case PT_GET_SC_ARGS:
600 		break;
601 	case PT_SETREGS:
602 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
603 		break;
604 	case PT_SETFPREGS:
605 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
606 		break;
607 	case PT_SETDBREGS:
608 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
609 		break;
610 	case PT_SET_EVENT_MASK:
611 		if (uap->data != sizeof(r.ptevents))
612 			error = EINVAL;
613 		else
614 			error = copyin(uap->addr, &r.ptevents, uap->data);
615 		break;
616 	case PT_IO:
617 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
618 		break;
619 	case PT_VM_ENTRY:
620 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
621 		break;
622 	default:
623 		addr = uap->addr;
624 		break;
625 	}
626 	if (error)
627 		return (error);
628 
629 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
630 	if (error)
631 		return (error);
632 
633 	switch (uap->req) {
634 	case PT_VM_ENTRY:
635 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
636 		break;
637 	case PT_IO:
638 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
639 		break;
640 	case PT_GETREGS:
641 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
642 		break;
643 	case PT_GETFPREGS:
644 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
645 		break;
646 	case PT_GETDBREGS:
647 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
648 		break;
649 	case PT_GET_EVENT_MASK:
650 		/* NB: The size in uap->data is validated in kern_ptrace(). */
651 		error = copyout(&r.ptevents, uap->addr, uap->data);
652 		break;
653 	case PT_LWPINFO:
654 		/* NB: The size in uap->data is validated in kern_ptrace(). */
655 		error = copyout(&r.pl, uap->addr, uap->data);
656 		break;
657 	case PT_GET_SC_ARGS:
658 		error = copyout(r.args, uap->addr, MIN(uap->data,
659 		    sizeof(r.args)));
660 		break;
661 	}
662 
663 	return (error);
664 }
665 #undef COPYIN
666 #undef COPYOUT
667 
668 #ifdef COMPAT_FREEBSD32
669 /*
670  *   PROC_READ(regs, td2, addr);
671  * becomes either:
672  *   proc_read_regs(td2, addr);
673  * or
674  *   proc_read_regs32(td2, addr);
675  * .. except this is done at runtime.  There is an additional
676  * complication in that PROC_WRITE disallows 32 bit consumers
677  * from writing to 64 bit address space targets.
678  */
679 #define	PROC_READ(w, t, a)	wrap32 ? \
680 	proc_read_ ## w ## 32(t, a) : \
681 	proc_read_ ## w (t, a)
682 #define	PROC_WRITE(w, t, a)	wrap32 ? \
683 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
684 	proc_write_ ## w (t, a)
685 #else
686 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
687 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
688 #endif
689 
690 void
691 proc_set_traced(struct proc *p, bool stop)
692 {
693 
694 	PROC_LOCK_ASSERT(p, MA_OWNED);
695 	p->p_flag |= P_TRACED;
696 	if (stop)
697 		p->p_flag2 |= P2_PTRACE_FSTP;
698 	p->p_ptevents = PTRACE_DEFAULT;
699 	p->p_oppid = p->p_pptr->p_pid;
700 }
701 
702 int
703 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
704 {
705 	struct iovec iov;
706 	struct uio uio;
707 	struct proc *curp, *p, *pp;
708 	struct thread *td2 = NULL, *td3;
709 	struct ptrace_io_desc *piod = NULL;
710 	struct ptrace_lwpinfo *pl;
711 	int error, num, tmp;
712 	int proctree_locked = 0;
713 	lwpid_t tid = 0, *buf;
714 #ifdef COMPAT_FREEBSD32
715 	int wrap32 = 0, safe = 0;
716 	struct ptrace_io_desc32 *piod32 = NULL;
717 	struct ptrace_lwpinfo32 *pl32 = NULL;
718 	struct ptrace_lwpinfo plr;
719 #endif
720 
721 	curp = td->td_proc;
722 
723 	/* Lock proctree before locking the process. */
724 	switch (req) {
725 	case PT_TRACE_ME:
726 	case PT_ATTACH:
727 	case PT_STEP:
728 	case PT_CONTINUE:
729 	case PT_TO_SCE:
730 	case PT_TO_SCX:
731 	case PT_SYSCALL:
732 	case PT_FOLLOW_FORK:
733 	case PT_LWP_EVENTS:
734 	case PT_GET_EVENT_MASK:
735 	case PT_SET_EVENT_MASK:
736 	case PT_DETACH:
737 	case PT_GET_SC_ARGS:
738 		sx_xlock(&proctree_lock);
739 		proctree_locked = 1;
740 		break;
741 	default:
742 		break;
743 	}
744 
745 	if (req == PT_TRACE_ME) {
746 		p = td->td_proc;
747 		PROC_LOCK(p);
748 	} else {
749 		if (pid <= PID_MAX) {
750 			if ((p = pfind(pid)) == NULL) {
751 				if (proctree_locked)
752 					sx_xunlock(&proctree_lock);
753 				return (ESRCH);
754 			}
755 		} else {
756 			td2 = tdfind(pid, -1);
757 			if (td2 == NULL) {
758 				if (proctree_locked)
759 					sx_xunlock(&proctree_lock);
760 				return (ESRCH);
761 			}
762 			p = td2->td_proc;
763 			tid = pid;
764 			pid = p->p_pid;
765 		}
766 	}
767 	AUDIT_ARG_PROCESS(p);
768 
769 	if ((p->p_flag & P_WEXIT) != 0) {
770 		error = ESRCH;
771 		goto fail;
772 	}
773 	if ((error = p_cansee(td, p)) != 0)
774 		goto fail;
775 
776 	if ((error = p_candebug(td, p)) != 0)
777 		goto fail;
778 
779 	/*
780 	 * System processes can't be debugged.
781 	 */
782 	if ((p->p_flag & P_SYSTEM) != 0) {
783 		error = EINVAL;
784 		goto fail;
785 	}
786 
787 	if (tid == 0) {
788 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
789 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
790 			td2 = p->p_xthread;
791 		} else {
792 			td2 = FIRST_THREAD_IN_PROC(p);
793 		}
794 		tid = td2->td_tid;
795 	}
796 
797 #ifdef COMPAT_FREEBSD32
798 	/*
799 	 * Test if we're a 32 bit client and what the target is.
800 	 * Set the wrap controls accordingly.
801 	 */
802 	if (SV_CURPROC_FLAG(SV_ILP32)) {
803 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
804 			safe = 1;
805 		wrap32 = 1;
806 	}
807 #endif
808 	/*
809 	 * Permissions check
810 	 */
811 	switch (req) {
812 	case PT_TRACE_ME:
813 		/*
814 		 * Always legal, when there is a parent process which
815 		 * could trace us.  Otherwise, reject.
816 		 */
817 		if ((p->p_flag & P_TRACED) != 0) {
818 			error = EBUSY;
819 			goto fail;
820 		}
821 		if (p->p_pptr == initproc) {
822 			error = EPERM;
823 			goto fail;
824 		}
825 		break;
826 
827 	case PT_ATTACH:
828 		/* Self */
829 		if (p == td->td_proc) {
830 			error = EINVAL;
831 			goto fail;
832 		}
833 
834 		/* Already traced */
835 		if (p->p_flag & P_TRACED) {
836 			error = EBUSY;
837 			goto fail;
838 		}
839 
840 		/* Can't trace an ancestor if you're being traced. */
841 		if (curp->p_flag & P_TRACED) {
842 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
843 				if (pp == p) {
844 					error = EINVAL;
845 					goto fail;
846 				}
847 			}
848 		}
849 
850 
851 		/* OK */
852 		break;
853 
854 	case PT_CLEARSTEP:
855 		/* Allow thread to clear single step for itself */
856 		if (td->td_tid == tid)
857 			break;
858 
859 		/* FALLTHROUGH */
860 	default:
861 		/* not being traced... */
862 		if ((p->p_flag & P_TRACED) == 0) {
863 			error = EPERM;
864 			goto fail;
865 		}
866 
867 		/* not being traced by YOU */
868 		if (p->p_pptr != td->td_proc) {
869 			error = EBUSY;
870 			goto fail;
871 		}
872 
873 		/* not currently stopped */
874 		if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
875 		    p->p_suspcount != p->p_numthreads  ||
876 		    (p->p_flag & P_WAITED) == 0) {
877 			error = EBUSY;
878 			goto fail;
879 		}
880 
881 		/* OK */
882 		break;
883 	}
884 
885 	/* Keep this process around until we finish this request. */
886 	_PHOLD(p);
887 
888 #ifdef FIX_SSTEP
889 	/*
890 	 * Single step fixup ala procfs
891 	 */
892 	FIX_SSTEP(td2);
893 #endif
894 
895 	/*
896 	 * Actually do the requests
897 	 */
898 
899 	td->td_retval[0] = 0;
900 
901 	switch (req) {
902 	case PT_TRACE_ME:
903 		/* set my trace flag and "owner" so it can read/write me */
904 		proc_set_traced(p, false);
905 		if (p->p_flag & P_PPWAIT)
906 			p->p_flag |= P_PPTRACE;
907 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
908 		break;
909 
910 	case PT_ATTACH:
911 		/* security check done above */
912 		/*
913 		 * It would be nice if the tracing relationship was separate
914 		 * from the parent relationship but that would require
915 		 * another set of links in the proc struct or for "wait"
916 		 * to scan the entire proc table.  To make life easier,
917 		 * we just re-parent the process we're trying to trace.
918 		 * The old parent is remembered so we can put things back
919 		 * on a "detach".
920 		 */
921 		proc_set_traced(p, true);
922 		if (p->p_pptr != td->td_proc) {
923 			proc_reparent(p, td->td_proc);
924 		}
925 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
926 		    p->p_oppid);
927 
928 		sx_xunlock(&proctree_lock);
929 		proctree_locked = 0;
930 		MPASS(p->p_xthread == NULL);
931 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
932 
933 		/*
934 		 * If already stopped due to a stop signal, clear the
935 		 * existing stop before triggering a traced SIGSTOP.
936 		 */
937 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
938 			PROC_SLOCK(p);
939 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
940 			thread_unsuspend(p);
941 			PROC_SUNLOCK(p);
942 		}
943 
944 		kern_psignal(p, SIGSTOP);
945 		break;
946 
947 	case PT_CLEARSTEP:
948 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
949 		    p->p_pid);
950 		error = ptrace_clear_single_step(td2);
951 		break;
952 
953 	case PT_SETSTEP:
954 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
955 		    p->p_pid);
956 		error = ptrace_single_step(td2);
957 		break;
958 
959 	case PT_SUSPEND:
960 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
961 		    p->p_pid);
962 		td2->td_dbgflags |= TDB_SUSPEND;
963 		thread_lock(td2);
964 		td2->td_flags |= TDF_NEEDSUSPCHK;
965 		thread_unlock(td2);
966 		break;
967 
968 	case PT_RESUME:
969 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
970 		    p->p_pid);
971 		td2->td_dbgflags &= ~TDB_SUSPEND;
972 		break;
973 
974 	case PT_FOLLOW_FORK:
975 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
976 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
977 		    data ? "enabled" : "disabled");
978 		if (data)
979 			p->p_ptevents |= PTRACE_FORK;
980 		else
981 			p->p_ptevents &= ~PTRACE_FORK;
982 		break;
983 
984 	case PT_LWP_EVENTS:
985 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
986 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
987 		    data ? "enabled" : "disabled");
988 		if (data)
989 			p->p_ptevents |= PTRACE_LWP;
990 		else
991 			p->p_ptevents &= ~PTRACE_LWP;
992 		break;
993 
994 	case PT_GET_EVENT_MASK:
995 		if (data != sizeof(p->p_ptevents)) {
996 			error = EINVAL;
997 			break;
998 		}
999 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1000 		    p->p_ptevents);
1001 		*(int *)addr = p->p_ptevents;
1002 		break;
1003 
1004 	case PT_SET_EVENT_MASK:
1005 		if (data != sizeof(p->p_ptevents)) {
1006 			error = EINVAL;
1007 			break;
1008 		}
1009 		tmp = *(int *)addr;
1010 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1011 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1012 			error = EINVAL;
1013 			break;
1014 		}
1015 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1016 		    p->p_pid, p->p_ptevents, tmp);
1017 		p->p_ptevents = tmp;
1018 		break;
1019 
1020 	case PT_GET_SC_ARGS:
1021 		CTR1(KTR_PTRACE, "PT_GET_SC_ARGS: pid %d", p->p_pid);
1022 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0
1023 #ifdef COMPAT_FREEBSD32
1024 		    || (wrap32 && !safe)
1025 #endif
1026 		    ) {
1027 			error = EINVAL;
1028 			break;
1029 		}
1030 		bzero(addr, sizeof(td2->td_sa.args));
1031 #ifdef COMPAT_FREEBSD32
1032 		if (wrap32)
1033 			for (num = 0; num < nitems(td2->td_sa.args); num++)
1034 				((uint32_t *)addr)[num] = (uint32_t)
1035 				    td2->td_sa.args[num];
1036 		else
1037 #endif
1038 			bcopy(td2->td_sa.args, addr, td2->td_sa.narg *
1039 			    sizeof(register_t));
1040 		break;
1041 
1042 	case PT_STEP:
1043 	case PT_CONTINUE:
1044 	case PT_TO_SCE:
1045 	case PT_TO_SCX:
1046 	case PT_SYSCALL:
1047 	case PT_DETACH:
1048 		/* Zero means do not send any signal */
1049 		if (data < 0 || data > _SIG_MAXSIG) {
1050 			error = EINVAL;
1051 			break;
1052 		}
1053 
1054 		switch (req) {
1055 		case PT_STEP:
1056 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1057 			    td2->td_tid, p->p_pid, data);
1058 			error = ptrace_single_step(td2);
1059 			if (error)
1060 				goto out;
1061 			break;
1062 		case PT_CONTINUE:
1063 		case PT_TO_SCE:
1064 		case PT_TO_SCX:
1065 		case PT_SYSCALL:
1066 			if (addr != (void *)1) {
1067 				error = ptrace_set_pc(td2,
1068 				    (u_long)(uintfptr_t)addr);
1069 				if (error)
1070 					goto out;
1071 			}
1072 			switch (req) {
1073 			case PT_TO_SCE:
1074 				p->p_ptevents |= PTRACE_SCE;
1075 				CTR4(KTR_PTRACE,
1076 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1077 				    p->p_pid, p->p_ptevents,
1078 				    (u_long)(uintfptr_t)addr, data);
1079 				break;
1080 			case PT_TO_SCX:
1081 				p->p_ptevents |= PTRACE_SCX;
1082 				CTR4(KTR_PTRACE,
1083 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1084 				    p->p_pid, p->p_ptevents,
1085 				    (u_long)(uintfptr_t)addr, data);
1086 				break;
1087 			case PT_SYSCALL:
1088 				p->p_ptevents |= PTRACE_SYSCALL;
1089 				CTR4(KTR_PTRACE,
1090 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1091 				    p->p_pid, p->p_ptevents,
1092 				    (u_long)(uintfptr_t)addr, data);
1093 				break;
1094 			case PT_CONTINUE:
1095 				CTR3(KTR_PTRACE,
1096 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1097 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1098 				break;
1099 			}
1100 			break;
1101 		case PT_DETACH:
1102 			/*
1103 			 * Reset the process parent.
1104 			 *
1105 			 * NB: This clears P_TRACED before reparenting
1106 			 * a detached process back to its original
1107 			 * parent.  Otherwise the debugee will be set
1108 			 * as an orphan of the debugger.
1109 			 */
1110 			p->p_flag &= ~(P_TRACED | P_WAITED);
1111 			if (p->p_oppid != p->p_pptr->p_pid) {
1112 				PROC_LOCK(p->p_pptr);
1113 				sigqueue_take(p->p_ksi);
1114 				PROC_UNLOCK(p->p_pptr);
1115 
1116 				pp = proc_realparent(p);
1117 				proc_reparent(p, pp);
1118 				if (pp == initproc)
1119 					p->p_sigparent = SIGCHLD;
1120 				CTR3(KTR_PTRACE,
1121 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1122 				    p->p_pid, pp->p_pid, data);
1123 			} else
1124 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1125 				    p->p_pid, data);
1126 			p->p_oppid = 0;
1127 			p->p_ptevents = 0;
1128 			FOREACH_THREAD_IN_PROC(p, td3) {
1129 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1130 					sigqueue_delete(&td3->td_sigqueue,
1131 					    SIGSTOP);
1132 				}
1133 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1134 				    TDB_SUSPEND);
1135 			}
1136 
1137 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1138 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1139 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1140 			}
1141 
1142 			/* should we send SIGCHLD? */
1143 			/* childproc_continued(p); */
1144 			break;
1145 		}
1146 
1147 		sx_xunlock(&proctree_lock);
1148 		proctree_locked = 0;
1149 
1150 	sendsig:
1151 		MPASS(proctree_locked == 0);
1152 
1153 		/*
1154 		 * Clear the pending event for the thread that just
1155 		 * reported its event (p_xthread).  This may not be
1156 		 * the thread passed to PT_CONTINUE, PT_STEP, etc. if
1157 		 * the debugger is resuming a different thread.
1158 		 *
1159 		 * Deliver any pending signal via the reporting thread.
1160 		 */
1161 		MPASS(p->p_xthread != NULL);
1162 		p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1163 		p->p_xthread->td_xsig = data;
1164 		p->p_xthread = NULL;
1165 		p->p_xsig = data;
1166 
1167 		/*
1168 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1169 		 * always works immediately, even if another thread is
1170 		 * unsuspended first and attempts to handle a
1171 		 * different signal or if the POSIX.1b style signal
1172 		 * queue cannot accommodate any new signals.
1173 		 */
1174 		if (data == SIGKILL)
1175 			p->p_flag |= P_WKILLED;
1176 
1177 		/*
1178 		 * Unsuspend all threads.  To leave a thread
1179 		 * suspended, use PT_SUSPEND to suspend it before
1180 		 * continuing the process.
1181 		 */
1182 		PROC_SLOCK(p);
1183 		p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
1184 		thread_unsuspend(p);
1185 		PROC_SUNLOCK(p);
1186 		break;
1187 
1188 	case PT_WRITE_I:
1189 	case PT_WRITE_D:
1190 		td2->td_dbgflags |= TDB_USERWR;
1191 		PROC_UNLOCK(p);
1192 		error = 0;
1193 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1194 		    sizeof(int)) != sizeof(int))
1195 			error = ENOMEM;
1196 		else
1197 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1198 			    p->p_pid, addr, data);
1199 		PROC_LOCK(p);
1200 		break;
1201 
1202 	case PT_READ_I:
1203 	case PT_READ_D:
1204 		PROC_UNLOCK(p);
1205 		error = tmp = 0;
1206 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1207 		    sizeof(int)) != sizeof(int))
1208 			error = ENOMEM;
1209 		else
1210 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1211 			    p->p_pid, addr, tmp);
1212 		td->td_retval[0] = tmp;
1213 		PROC_LOCK(p);
1214 		break;
1215 
1216 	case PT_IO:
1217 #ifdef COMPAT_FREEBSD32
1218 		if (wrap32) {
1219 			piod32 = addr;
1220 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1221 			iov.iov_len = piod32->piod_len;
1222 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1223 			uio.uio_resid = piod32->piod_len;
1224 		} else
1225 #endif
1226 		{
1227 			piod = addr;
1228 			iov.iov_base = piod->piod_addr;
1229 			iov.iov_len = piod->piod_len;
1230 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1231 			uio.uio_resid = piod->piod_len;
1232 		}
1233 		uio.uio_iov = &iov;
1234 		uio.uio_iovcnt = 1;
1235 		uio.uio_segflg = UIO_USERSPACE;
1236 		uio.uio_td = td;
1237 #ifdef COMPAT_FREEBSD32
1238 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1239 #else
1240 		tmp = piod->piod_op;
1241 #endif
1242 		switch (tmp) {
1243 		case PIOD_READ_D:
1244 		case PIOD_READ_I:
1245 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1246 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1247 			uio.uio_rw = UIO_READ;
1248 			break;
1249 		case PIOD_WRITE_D:
1250 		case PIOD_WRITE_I:
1251 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1252 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1253 			td2->td_dbgflags |= TDB_USERWR;
1254 			uio.uio_rw = UIO_WRITE;
1255 			break;
1256 		default:
1257 			error = EINVAL;
1258 			goto out;
1259 		}
1260 		PROC_UNLOCK(p);
1261 		error = proc_rwmem(p, &uio);
1262 #ifdef COMPAT_FREEBSD32
1263 		if (wrap32)
1264 			piod32->piod_len -= uio.uio_resid;
1265 		else
1266 #endif
1267 			piod->piod_len -= uio.uio_resid;
1268 		PROC_LOCK(p);
1269 		break;
1270 
1271 	case PT_KILL:
1272 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1273 		data = SIGKILL;
1274 		goto sendsig;	/* in PT_CONTINUE above */
1275 
1276 	case PT_SETREGS:
1277 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1278 		    p->p_pid);
1279 		td2->td_dbgflags |= TDB_USERWR;
1280 		error = PROC_WRITE(regs, td2, addr);
1281 		break;
1282 
1283 	case PT_GETREGS:
1284 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1285 		    p->p_pid);
1286 		error = PROC_READ(regs, td2, addr);
1287 		break;
1288 
1289 	case PT_SETFPREGS:
1290 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1291 		    p->p_pid);
1292 		td2->td_dbgflags |= TDB_USERWR;
1293 		error = PROC_WRITE(fpregs, td2, addr);
1294 		break;
1295 
1296 	case PT_GETFPREGS:
1297 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1298 		    p->p_pid);
1299 		error = PROC_READ(fpregs, td2, addr);
1300 		break;
1301 
1302 	case PT_SETDBREGS:
1303 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1304 		    p->p_pid);
1305 		td2->td_dbgflags |= TDB_USERWR;
1306 		error = PROC_WRITE(dbregs, td2, addr);
1307 		break;
1308 
1309 	case PT_GETDBREGS:
1310 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1311 		    p->p_pid);
1312 		error = PROC_READ(dbregs, td2, addr);
1313 		break;
1314 
1315 	case PT_LWPINFO:
1316 		if (data <= 0 ||
1317 #ifdef COMPAT_FREEBSD32
1318 		    (!wrap32 && data > sizeof(*pl)) ||
1319 		    (wrap32 && data > sizeof(*pl32))) {
1320 #else
1321 		    data > sizeof(*pl)) {
1322 #endif
1323 			error = EINVAL;
1324 			break;
1325 		}
1326 #ifdef COMPAT_FREEBSD32
1327 		if (wrap32) {
1328 			pl = &plr;
1329 			pl32 = addr;
1330 		} else
1331 #endif
1332 		pl = addr;
1333 		bzero(pl, sizeof(*pl));
1334 		pl->pl_lwpid = td2->td_tid;
1335 		pl->pl_event = PL_EVENT_NONE;
1336 		pl->pl_flags = 0;
1337 		if (td2->td_dbgflags & TDB_XSIG) {
1338 			pl->pl_event = PL_EVENT_SIGNAL;
1339 			if (td2->td_si.si_signo != 0 &&
1340 #ifdef COMPAT_FREEBSD32
1341 			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1342 			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1343 			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1344 			    pl_siginfo) + sizeof(struct siginfo32)))
1345 #else
1346 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1347 			    + sizeof(pl->pl_siginfo)
1348 #endif
1349 			){
1350 				pl->pl_flags |= PL_FLAG_SI;
1351 				pl->pl_siginfo = td2->td_si;
1352 			}
1353 		}
1354 		if (td2->td_dbgflags & TDB_SCE)
1355 			pl->pl_flags |= PL_FLAG_SCE;
1356 		else if (td2->td_dbgflags & TDB_SCX)
1357 			pl->pl_flags |= PL_FLAG_SCX;
1358 		if (td2->td_dbgflags & TDB_EXEC)
1359 			pl->pl_flags |= PL_FLAG_EXEC;
1360 		if (td2->td_dbgflags & TDB_FORK) {
1361 			pl->pl_flags |= PL_FLAG_FORKED;
1362 			pl->pl_child_pid = td2->td_dbg_forked;
1363 			if (td2->td_dbgflags & TDB_VFORK)
1364 				pl->pl_flags |= PL_FLAG_VFORKED;
1365 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1366 		    TDB_VFORK)
1367 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1368 		if (td2->td_dbgflags & TDB_CHILD)
1369 			pl->pl_flags |= PL_FLAG_CHILD;
1370 		if (td2->td_dbgflags & TDB_BORN)
1371 			pl->pl_flags |= PL_FLAG_BORN;
1372 		if (td2->td_dbgflags & TDB_EXIT)
1373 			pl->pl_flags |= PL_FLAG_EXITED;
1374 		pl->pl_sigmask = td2->td_sigmask;
1375 		pl->pl_siglist = td2->td_siglist;
1376 		strcpy(pl->pl_tdname, td2->td_name);
1377 		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1378 			pl->pl_syscall_code = td2->td_sa.code;
1379 			pl->pl_syscall_narg = td2->td_sa.narg;
1380 		} else {
1381 			pl->pl_syscall_code = 0;
1382 			pl->pl_syscall_narg = 0;
1383 		}
1384 #ifdef COMPAT_FREEBSD32
1385 		if (wrap32)
1386 			ptrace_lwpinfo_to32(pl, pl32);
1387 #endif
1388 		CTR6(KTR_PTRACE,
1389     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1390 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1391 		    pl->pl_child_pid, pl->pl_syscall_code);
1392 		break;
1393 
1394 	case PT_GETNUMLWPS:
1395 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1396 		    p->p_numthreads);
1397 		td->td_retval[0] = p->p_numthreads;
1398 		break;
1399 
1400 	case PT_GETLWPLIST:
1401 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1402 		    p->p_pid, data, p->p_numthreads);
1403 		if (data <= 0) {
1404 			error = EINVAL;
1405 			break;
1406 		}
1407 		num = imin(p->p_numthreads, data);
1408 		PROC_UNLOCK(p);
1409 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1410 		tmp = 0;
1411 		PROC_LOCK(p);
1412 		FOREACH_THREAD_IN_PROC(p, td2) {
1413 			if (tmp >= num)
1414 				break;
1415 			buf[tmp++] = td2->td_tid;
1416 		}
1417 		PROC_UNLOCK(p);
1418 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1419 		free(buf, M_TEMP);
1420 		if (!error)
1421 			td->td_retval[0] = tmp;
1422 		PROC_LOCK(p);
1423 		break;
1424 
1425 	case PT_VM_TIMESTAMP:
1426 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1427 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1428 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1429 		break;
1430 
1431 	case PT_VM_ENTRY:
1432 		PROC_UNLOCK(p);
1433 #ifdef COMPAT_FREEBSD32
1434 		if (wrap32)
1435 			error = ptrace_vm_entry32(td, p, addr);
1436 		else
1437 #endif
1438 		error = ptrace_vm_entry(td, p, addr);
1439 		PROC_LOCK(p);
1440 		break;
1441 
1442 	default:
1443 #ifdef __HAVE_PTRACE_MACHDEP
1444 		if (req >= PT_FIRSTMACH) {
1445 			PROC_UNLOCK(p);
1446 			error = cpu_ptrace(td2, req, addr, data);
1447 			PROC_LOCK(p);
1448 		} else
1449 #endif
1450 			/* Unknown request. */
1451 			error = EINVAL;
1452 		break;
1453 	}
1454 
1455 out:
1456 	/* Drop our hold on this process now that the request has completed. */
1457 	_PRELE(p);
1458 fail:
1459 	PROC_UNLOCK(p);
1460 	if (proctree_locked)
1461 		sx_xunlock(&proctree_lock);
1462 	return (error);
1463 }
1464 #undef PROC_READ
1465 #undef PROC_WRITE
1466 
1467 /*
1468  * Stop a process because of a debugging event;
1469  * stay stopped until p->p_step is cleared
1470  * (cleared by PIOCCONT in procfs).
1471  */
1472 void
1473 stopevent(struct proc *p, unsigned int event, unsigned int val)
1474 {
1475 
1476 	PROC_LOCK_ASSERT(p, MA_OWNED);
1477 	p->p_step = 1;
1478 	CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1479 	    val);
1480 	do {
1481 		if (event != S_EXIT)
1482 			p->p_xsig = val;
1483 		p->p_xthread = NULL;
1484 		p->p_stype = event;	/* Which event caused the stop? */
1485 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1486 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1487 	} while (p->p_step);
1488 }
1489