xref: /freebsd/sys/kern/sys_process.c (revision 6f3544cd7084abbadd83637993a4f41fd30e6ccd)
1 /*-
2  * Copyright (c) 1994, Sean Eric Fagan
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by Sean Eric Fagan.
16  * 4. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/syscallsubr.h>
42 #include <sys/sysent.h>
43 #include <sys/sysproto.h>
44 #include <sys/proc.h>
45 #include <sys/vnode.h>
46 #include <sys/ptrace.h>
47 #include <sys/sx.h>
48 #include <sys/malloc.h>
49 #include <sys/signalvar.h>
50 
51 #include <machine/reg.h>
52 
53 #include <security/audit/audit.h>
54 
55 #include <vm/vm.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_extern.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_kern.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pager.h>
63 #include <vm/vm_param.h>
64 
65 #ifdef COMPAT_FREEBSD32
66 #include <sys/procfs.h>
67 #include <compat/freebsd32/freebsd32_signal.h>
68 
69 struct ptrace_io_desc32 {
70 	int		piod_op;
71 	uint32_t	piod_offs;
72 	uint32_t	piod_addr;
73 	uint32_t	piod_len;
74 };
75 
76 struct ptrace_vm_entry32 {
77 	int		pve_entry;
78 	int		pve_timestamp;
79 	uint32_t	pve_start;
80 	uint32_t	pve_end;
81 	uint32_t	pve_offset;
82 	u_int		pve_prot;
83 	u_int		pve_pathlen;
84 	int32_t		pve_fileid;
85 	u_int		pve_fsid;
86 	uint32_t	pve_path;
87 };
88 
89 struct ptrace_lwpinfo32 {
90 	lwpid_t	pl_lwpid;	/* LWP described. */
91 	int	pl_event;	/* Event that stopped the LWP. */
92 	int	pl_flags;	/* LWP flags. */
93 	sigset_t	pl_sigmask;	/* LWP signal mask */
94 	sigset_t	pl_siglist;	/* LWP pending signal */
95 	struct siginfo32 pl_siginfo;	/* siginfo for signal */
96 };
97 
98 #endif
99 
100 /*
101  * Functions implemented using PROC_ACTION():
102  *
103  * proc_read_regs(proc, regs)
104  *	Get the current user-visible register set from the process
105  *	and copy it into the regs structure (<machine/reg.h>).
106  *	The process is stopped at the time read_regs is called.
107  *
108  * proc_write_regs(proc, regs)
109  *	Update the current register set from the passed in regs
110  *	structure.  Take care to avoid clobbering special CPU
111  *	registers or privileged bits in the PSL.
112  *	Depending on the architecture this may have fix-up work to do,
113  *	especially if the IAR or PCW are modified.
114  *	The process is stopped at the time write_regs is called.
115  *
116  * proc_read_fpregs, proc_write_fpregs
117  *	deal with the floating point register set, otherwise as above.
118  *
119  * proc_read_dbregs, proc_write_dbregs
120  *	deal with the processor debug register set, otherwise as above.
121  *
122  * proc_sstep(proc)
123  *	Arrange for the process to trap after executing a single instruction.
124  */
125 
126 #define	PROC_ACTION(action) do {					\
127 	int error;							\
128 									\
129 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
130 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
131 		error = EIO;						\
132 	else								\
133 		error = (action);					\
134 	return (error);							\
135 } while(0)
136 
137 int
138 proc_read_regs(struct thread *td, struct reg *regs)
139 {
140 
141 	PROC_ACTION(fill_regs(td, regs));
142 }
143 
144 int
145 proc_write_regs(struct thread *td, struct reg *regs)
146 {
147 
148 	PROC_ACTION(set_regs(td, regs));
149 }
150 
151 int
152 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
153 {
154 
155 	PROC_ACTION(fill_dbregs(td, dbregs));
156 }
157 
158 int
159 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
160 {
161 
162 	PROC_ACTION(set_dbregs(td, dbregs));
163 }
164 
165 /*
166  * Ptrace doesn't support fpregs at all, and there are no security holes
167  * or translations for fpregs, so we can just copy them.
168  */
169 int
170 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
171 {
172 
173 	PROC_ACTION(fill_fpregs(td, fpregs));
174 }
175 
176 int
177 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
178 {
179 
180 	PROC_ACTION(set_fpregs(td, fpregs));
181 }
182 
183 #ifdef COMPAT_FREEBSD32
184 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
185 int
186 proc_read_regs32(struct thread *td, struct reg32 *regs32)
187 {
188 
189 	PROC_ACTION(fill_regs32(td, regs32));
190 }
191 
192 int
193 proc_write_regs32(struct thread *td, struct reg32 *regs32)
194 {
195 
196 	PROC_ACTION(set_regs32(td, regs32));
197 }
198 
199 int
200 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
201 {
202 
203 	PROC_ACTION(fill_dbregs32(td, dbregs32));
204 }
205 
206 int
207 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
208 {
209 
210 	PROC_ACTION(set_dbregs32(td, dbregs32));
211 }
212 
213 int
214 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
215 {
216 
217 	PROC_ACTION(fill_fpregs32(td, fpregs32));
218 }
219 
220 int
221 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
222 {
223 
224 	PROC_ACTION(set_fpregs32(td, fpregs32));
225 }
226 #endif
227 
228 int
229 proc_sstep(struct thread *td)
230 {
231 
232 	PROC_ACTION(ptrace_single_step(td));
233 }
234 
235 int
236 proc_rwmem(struct proc *p, struct uio *uio)
237 {
238 	vm_map_t map;
239 	vm_object_t backing_object, object;
240 	vm_offset_t pageno;		/* page number */
241 	vm_prot_t reqprot;
242 	int error, writing;
243 
244 	/*
245 	 * Assert that someone has locked this vmspace.  (Should be
246 	 * curthread but we can't assert that.)  This keeps the process
247 	 * from exiting out from under us until this operation completes.
248 	 */
249 	KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
250 	    p, p->p_pid));
251 
252 	/*
253 	 * The map we want...
254 	 */
255 	map = &p->p_vmspace->vm_map;
256 
257 	writing = uio->uio_rw == UIO_WRITE;
258 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
259 
260 	/*
261 	 * Only map in one page at a time.  We don't have to, but it
262 	 * makes things easier.  This way is trivial - right?
263 	 */
264 	do {
265 		vm_map_t tmap;
266 		vm_offset_t uva;
267 		int page_offset;		/* offset into page */
268 		vm_map_entry_t out_entry;
269 		vm_prot_t out_prot;
270 		boolean_t wired;
271 		vm_pindex_t pindex;
272 		u_int len;
273 		vm_page_t m;
274 
275 		object = NULL;
276 
277 		uva = (vm_offset_t)uio->uio_offset;
278 
279 		/*
280 		 * Get the page number of this segment.
281 		 */
282 		pageno = trunc_page(uva);
283 		page_offset = uva - pageno;
284 
285 		/*
286 		 * How many bytes to copy
287 		 */
288 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
289 
290 		/*
291 		 * Fault the page on behalf of the process
292 		 */
293 		error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
294 		if (error) {
295 			if (error == KERN_RESOURCE_SHORTAGE)
296 				error = ENOMEM;
297 			else
298 				error = EFAULT;
299 			break;
300 		}
301 
302 		/*
303 		 * Now we need to get the page.  out_entry and wired
304 		 * aren't used.  One would think the vm code
305 		 * would be a *bit* nicer...  We use tmap because
306 		 * vm_map_lookup() can change the map argument.
307 		 */
308 		tmap = map;
309 		error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
310 		    &object, &pindex, &out_prot, &wired);
311 		if (error) {
312 			error = EFAULT;
313 			break;
314 		}
315 		VM_OBJECT_LOCK(object);
316 		while ((m = vm_page_lookup(object, pindex)) == NULL &&
317 		    !writing &&
318 		    (backing_object = object->backing_object) != NULL) {
319 			/*
320 			 * Allow fallback to backing objects if we are reading.
321 			 */
322 			VM_OBJECT_LOCK(backing_object);
323 			pindex += OFF_TO_IDX(object->backing_object_offset);
324 			VM_OBJECT_UNLOCK(object);
325 			object = backing_object;
326 		}
327 		if (writing && m != NULL) {
328 			vm_page_dirty(m);
329 			vm_pager_page_unswapped(m);
330 		}
331 		VM_OBJECT_UNLOCK(object);
332 		if (m == NULL) {
333 			vm_map_lookup_done(tmap, out_entry);
334 			error = EFAULT;
335 			break;
336 		}
337 
338 		/*
339 		 * Hold the page in memory.
340 		 */
341 		vm_page_lock(m);
342 		vm_page_hold(m);
343 		vm_page_unlock(m);
344 
345 		/*
346 		 * We're done with tmap now.
347 		 */
348 		vm_map_lookup_done(tmap, out_entry);
349 
350 		/*
351 		 * Now do the i/o move.
352 		 */
353 		error = uiomove_fromphys(&m, page_offset, len, uio);
354 
355 		/* Make the I-cache coherent for breakpoints. */
356 		if (!error && writing && (out_prot & VM_PROT_EXECUTE))
357 			vm_sync_icache(map, uva, len);
358 
359 		/*
360 		 * Release the page.
361 		 */
362 		vm_page_lock(m);
363 		vm_page_unhold(m);
364 		vm_page_unlock(m);
365 
366 	} while (error == 0 && uio->uio_resid > 0);
367 
368 	return (error);
369 }
370 
371 static int
372 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
373 {
374 	struct vattr vattr;
375 	vm_map_t map;
376 	vm_map_entry_t entry;
377 	vm_object_t obj, tobj, lobj;
378 	struct vmspace *vm;
379 	struct vnode *vp;
380 	char *freepath, *fullpath;
381 	u_int pathlen;
382 	int error, index, vfslocked;
383 
384 	error = 0;
385 	obj = NULL;
386 
387 	vm = vmspace_acquire_ref(p);
388 	map = &vm->vm_map;
389 	vm_map_lock_read(map);
390 
391 	do {
392 		entry = map->header.next;
393 		index = 0;
394 		while (index < pve->pve_entry && entry != &map->header) {
395 			entry = entry->next;
396 			index++;
397 		}
398 		if (index != pve->pve_entry) {
399 			error = EINVAL;
400 			break;
401 		}
402 		while (entry != &map->header &&
403 		    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
404 			entry = entry->next;
405 			index++;
406 		}
407 		if (entry == &map->header) {
408 			error = ENOENT;
409 			break;
410 		}
411 
412 		/* We got an entry. */
413 		pve->pve_entry = index + 1;
414 		pve->pve_timestamp = map->timestamp;
415 		pve->pve_start = entry->start;
416 		pve->pve_end = entry->end - 1;
417 		pve->pve_offset = entry->offset;
418 		pve->pve_prot = entry->protection;
419 
420 		/* Backing object's path needed? */
421 		if (pve->pve_pathlen == 0)
422 			break;
423 
424 		pathlen = pve->pve_pathlen;
425 		pve->pve_pathlen = 0;
426 
427 		obj = entry->object.vm_object;
428 		if (obj != NULL)
429 			VM_OBJECT_LOCK(obj);
430 	} while (0);
431 
432 	vm_map_unlock_read(map);
433 	vmspace_free(vm);
434 
435 	pve->pve_fsid = VNOVAL;
436 	pve->pve_fileid = VNOVAL;
437 
438 	if (error == 0 && obj != NULL) {
439 		lobj = obj;
440 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
441 			if (tobj != obj)
442 				VM_OBJECT_LOCK(tobj);
443 			if (lobj != obj)
444 				VM_OBJECT_UNLOCK(lobj);
445 			lobj = tobj;
446 			pve->pve_offset += tobj->backing_object_offset;
447 		}
448 		vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
449 		if (vp != NULL)
450 			vref(vp);
451 		if (lobj != obj)
452 			VM_OBJECT_UNLOCK(lobj);
453 		VM_OBJECT_UNLOCK(obj);
454 
455 		if (vp != NULL) {
456 			freepath = NULL;
457 			fullpath = NULL;
458 			vn_fullpath(td, vp, &fullpath, &freepath);
459 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
460 			vn_lock(vp, LK_SHARED | LK_RETRY);
461 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
462 				pve->pve_fileid = vattr.va_fileid;
463 				pve->pve_fsid = vattr.va_fsid;
464 			}
465 			vput(vp);
466 			VFS_UNLOCK_GIANT(vfslocked);
467 
468 			if (fullpath != NULL) {
469 				pve->pve_pathlen = strlen(fullpath) + 1;
470 				if (pve->pve_pathlen <= pathlen) {
471 					error = copyout(fullpath, pve->pve_path,
472 					    pve->pve_pathlen);
473 				} else
474 					error = ENAMETOOLONG;
475 			}
476 			if (freepath != NULL)
477 				free(freepath, M_TEMP);
478 		}
479 	}
480 
481 	return (error);
482 }
483 
484 #ifdef COMPAT_FREEBSD32
485 static int
486 ptrace_vm_entry32(struct thread *td, struct proc *p,
487     struct ptrace_vm_entry32 *pve32)
488 {
489 	struct ptrace_vm_entry pve;
490 	int error;
491 
492 	pve.pve_entry = pve32->pve_entry;
493 	pve.pve_pathlen = pve32->pve_pathlen;
494 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
495 
496 	error = ptrace_vm_entry(td, p, &pve);
497 	if (error == 0) {
498 		pve32->pve_entry = pve.pve_entry;
499 		pve32->pve_timestamp = pve.pve_timestamp;
500 		pve32->pve_start = pve.pve_start;
501 		pve32->pve_end = pve.pve_end;
502 		pve32->pve_offset = pve.pve_offset;
503 		pve32->pve_prot = pve.pve_prot;
504 		pve32->pve_fileid = pve.pve_fileid;
505 		pve32->pve_fsid = pve.pve_fsid;
506 	}
507 
508 	pve32->pve_pathlen = pve.pve_pathlen;
509 	return (error);
510 }
511 
512 static void
513 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
514     struct ptrace_lwpinfo32 *pl32)
515 {
516 
517 	pl32->pl_lwpid = pl->pl_lwpid;
518 	pl32->pl_event = pl->pl_event;
519 	pl32->pl_flags = pl->pl_flags;
520 	pl32->pl_sigmask = pl->pl_sigmask;
521 	pl32->pl_siglist = pl->pl_siglist;
522 	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
523 }
524 #endif /* COMPAT_FREEBSD32 */
525 
526 /*
527  * Process debugging system call.
528  */
529 #ifndef _SYS_SYSPROTO_H_
530 struct ptrace_args {
531 	int	req;
532 	pid_t	pid;
533 	caddr_t	addr;
534 	int	data;
535 };
536 #endif
537 
538 #ifdef COMPAT_FREEBSD32
539 /*
540  * This CPP subterfuge is to try and reduce the number of ifdefs in
541  * the body of the code.
542  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
543  * becomes either:
544  *   copyin(uap->addr, &r.reg, sizeof r.reg);
545  * or
546  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
547  * .. except this is done at runtime.
548  */
549 #define	COPYIN(u, k, s)		wrap32 ? \
550 	copyin(u, k ## 32, s ## 32) : \
551 	copyin(u, k, s)
552 #define	COPYOUT(k, u, s)	wrap32 ? \
553 	copyout(k ## 32, u, s ## 32) : \
554 	copyout(k, u, s)
555 #else
556 #define	COPYIN(u, k, s)		copyin(u, k, s)
557 #define	COPYOUT(k, u, s)	copyout(k, u, s)
558 #endif
559 int
560 ptrace(struct thread *td, struct ptrace_args *uap)
561 {
562 	/*
563 	 * XXX this obfuscation is to reduce stack usage, but the register
564 	 * structs may be too large to put on the stack anyway.
565 	 */
566 	union {
567 		struct ptrace_io_desc piod;
568 		struct ptrace_lwpinfo pl;
569 		struct ptrace_vm_entry pve;
570 		struct dbreg dbreg;
571 		struct fpreg fpreg;
572 		struct reg reg;
573 #ifdef COMPAT_FREEBSD32
574 		struct dbreg32 dbreg32;
575 		struct fpreg32 fpreg32;
576 		struct reg32 reg32;
577 		struct ptrace_io_desc32 piod32;
578 		struct ptrace_lwpinfo32 pl32;
579 		struct ptrace_vm_entry32 pve32;
580 #endif
581 	} r;
582 	void *addr;
583 	int error = 0;
584 #ifdef COMPAT_FREEBSD32
585 	int wrap32 = 0;
586 
587 	if (SV_CURPROC_FLAG(SV_ILP32))
588 		wrap32 = 1;
589 #endif
590 	AUDIT_ARG_PID(uap->pid);
591 	AUDIT_ARG_CMD(uap->req);
592 	AUDIT_ARG_VALUE(uap->data);
593 	addr = &r;
594 	switch (uap->req) {
595 	case PT_GETREGS:
596 	case PT_GETFPREGS:
597 	case PT_GETDBREGS:
598 	case PT_LWPINFO:
599 		break;
600 	case PT_SETREGS:
601 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
602 		break;
603 	case PT_SETFPREGS:
604 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
605 		break;
606 	case PT_SETDBREGS:
607 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
608 		break;
609 	case PT_IO:
610 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
611 		break;
612 	case PT_VM_ENTRY:
613 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
614 		break;
615 	default:
616 		addr = uap->addr;
617 		break;
618 	}
619 	if (error)
620 		return (error);
621 
622 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
623 	if (error)
624 		return (error);
625 
626 	switch (uap->req) {
627 	case PT_VM_ENTRY:
628 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
629 		break;
630 	case PT_IO:
631 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
632 		break;
633 	case PT_GETREGS:
634 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
635 		break;
636 	case PT_GETFPREGS:
637 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
638 		break;
639 	case PT_GETDBREGS:
640 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
641 		break;
642 	case PT_LWPINFO:
643 		error = copyout(&r.pl, uap->addr, uap->data);
644 		break;
645 	}
646 
647 	return (error);
648 }
649 #undef COPYIN
650 #undef COPYOUT
651 
652 #ifdef COMPAT_FREEBSD32
653 /*
654  *   PROC_READ(regs, td2, addr);
655  * becomes either:
656  *   proc_read_regs(td2, addr);
657  * or
658  *   proc_read_regs32(td2, addr);
659  * .. except this is done at runtime.  There is an additional
660  * complication in that PROC_WRITE disallows 32 bit consumers
661  * from writing to 64 bit address space targets.
662  */
663 #define	PROC_READ(w, t, a)	wrap32 ? \
664 	proc_read_ ## w ## 32(t, a) : \
665 	proc_read_ ## w (t, a)
666 #define	PROC_WRITE(w, t, a)	wrap32 ? \
667 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
668 	proc_write_ ## w (t, a)
669 #else
670 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
671 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
672 #endif
673 
674 int
675 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
676 {
677 	struct iovec iov;
678 	struct uio uio;
679 	struct proc *curp, *p, *pp;
680 	struct thread *td2 = NULL;
681 	struct ptrace_io_desc *piod = NULL;
682 	struct ptrace_lwpinfo *pl;
683 	int error, write, tmp, num;
684 	int proctree_locked = 0;
685 	lwpid_t tid = 0, *buf;
686 #ifdef COMPAT_FREEBSD32
687 	int wrap32 = 0, safe = 0;
688 	struct ptrace_io_desc32 *piod32 = NULL;
689 	struct ptrace_lwpinfo32 *pl32 = NULL;
690 	struct ptrace_lwpinfo plr;
691 #endif
692 
693 	curp = td->td_proc;
694 
695 	/* Lock proctree before locking the process. */
696 	switch (req) {
697 	case PT_TRACE_ME:
698 	case PT_ATTACH:
699 	case PT_STEP:
700 	case PT_CONTINUE:
701 	case PT_TO_SCE:
702 	case PT_TO_SCX:
703 	case PT_SYSCALL:
704 	case PT_DETACH:
705 		sx_xlock(&proctree_lock);
706 		proctree_locked = 1;
707 		break;
708 	default:
709 		break;
710 	}
711 
712 	write = 0;
713 	if (req == PT_TRACE_ME) {
714 		p = td->td_proc;
715 		PROC_LOCK(p);
716 	} else {
717 		if (pid <= PID_MAX) {
718 			if ((p = pfind(pid)) == NULL) {
719 				if (proctree_locked)
720 					sx_xunlock(&proctree_lock);
721 				return (ESRCH);
722 			}
723 		} else {
724 			td2 = tdfind(pid, -1);
725 			if (td2 == NULL) {
726 				if (proctree_locked)
727 					sx_xunlock(&proctree_lock);
728 				return (ESRCH);
729 			}
730 			p = td2->td_proc;
731 			tid = pid;
732 			pid = p->p_pid;
733 		}
734 	}
735 	AUDIT_ARG_PROCESS(p);
736 
737 	if ((p->p_flag & P_WEXIT) != 0) {
738 		error = ESRCH;
739 		goto fail;
740 	}
741 	if ((error = p_cansee(td, p)) != 0)
742 		goto fail;
743 
744 	if ((error = p_candebug(td, p)) != 0)
745 		goto fail;
746 
747 	/*
748 	 * System processes can't be debugged.
749 	 */
750 	if ((p->p_flag & P_SYSTEM) != 0) {
751 		error = EINVAL;
752 		goto fail;
753 	}
754 
755 	if (tid == 0) {
756 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
757 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
758 			td2 = p->p_xthread;
759 		} else {
760 			td2 = FIRST_THREAD_IN_PROC(p);
761 		}
762 		tid = td2->td_tid;
763 	}
764 
765 #ifdef COMPAT_FREEBSD32
766 	/*
767 	 * Test if we're a 32 bit client and what the target is.
768 	 * Set the wrap controls accordingly.
769 	 */
770 	if (SV_CURPROC_FLAG(SV_ILP32)) {
771 		if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
772 			safe = 1;
773 		wrap32 = 1;
774 	}
775 #endif
776 	/*
777 	 * Permissions check
778 	 */
779 	switch (req) {
780 	case PT_TRACE_ME:
781 		/* Always legal. */
782 		break;
783 
784 	case PT_ATTACH:
785 		/* Self */
786 		if (p->p_pid == td->td_proc->p_pid) {
787 			error = EINVAL;
788 			goto fail;
789 		}
790 
791 		/* Already traced */
792 		if (p->p_flag & P_TRACED) {
793 			error = EBUSY;
794 			goto fail;
795 		}
796 
797 		/* Can't trace an ancestor if you're being traced. */
798 		if (curp->p_flag & P_TRACED) {
799 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
800 				if (pp == p) {
801 					error = EINVAL;
802 					goto fail;
803 				}
804 			}
805 		}
806 
807 
808 		/* OK */
809 		break;
810 
811 	case PT_CLEARSTEP:
812 		/* Allow thread to clear single step for itself */
813 		if (td->td_tid == tid)
814 			break;
815 
816 		/* FALLTHROUGH */
817 	default:
818 		/* not being traced... */
819 		if ((p->p_flag & P_TRACED) == 0) {
820 			error = EPERM;
821 			goto fail;
822 		}
823 
824 		/* not being traced by YOU */
825 		if (p->p_pptr != td->td_proc) {
826 			error = EBUSY;
827 			goto fail;
828 		}
829 
830 		/* not currently stopped */
831 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
832 		    p->p_suspcount != p->p_numthreads  ||
833 		    (p->p_flag & P_WAITED) == 0) {
834 			error = EBUSY;
835 			goto fail;
836 		}
837 
838 		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
839 			static int count = 0;
840 			if (count++ == 0)
841 				printf("P_STOPPED_TRACE not set.\n");
842 		}
843 
844 		/* OK */
845 		break;
846 	}
847 
848 	/* Keep this process around until we finish this request. */
849 	_PHOLD(p);
850 
851 #ifdef FIX_SSTEP
852 	/*
853 	 * Single step fixup ala procfs
854 	 */
855 	FIX_SSTEP(td2);
856 #endif
857 
858 	/*
859 	 * Actually do the requests
860 	 */
861 
862 	td->td_retval[0] = 0;
863 
864 	switch (req) {
865 	case PT_TRACE_ME:
866 		/* set my trace flag and "owner" so it can read/write me */
867 		p->p_flag |= P_TRACED;
868 		p->p_oppid = p->p_pptr->p_pid;
869 		break;
870 
871 	case PT_ATTACH:
872 		/* security check done above */
873 		p->p_flag |= P_TRACED;
874 		p->p_oppid = p->p_pptr->p_pid;
875 		if (p->p_pptr != td->td_proc)
876 			proc_reparent(p, td->td_proc);
877 		data = SIGSTOP;
878 		goto sendsig;	/* in PT_CONTINUE below */
879 
880 	case PT_CLEARSTEP:
881 		error = ptrace_clear_single_step(td2);
882 		break;
883 
884 	case PT_SETSTEP:
885 		error = ptrace_single_step(td2);
886 		break;
887 
888 	case PT_SUSPEND:
889 		td2->td_dbgflags |= TDB_SUSPEND;
890 		thread_lock(td2);
891 		td2->td_flags |= TDF_NEEDSUSPCHK;
892 		thread_unlock(td2);
893 		break;
894 
895 	case PT_RESUME:
896 		td2->td_dbgflags &= ~TDB_SUSPEND;
897 		break;
898 
899 	case PT_STEP:
900 	case PT_CONTINUE:
901 	case PT_TO_SCE:
902 	case PT_TO_SCX:
903 	case PT_SYSCALL:
904 	case PT_DETACH:
905 		/* Zero means do not send any signal */
906 		if (data < 0 || data > _SIG_MAXSIG) {
907 			error = EINVAL;
908 			break;
909 		}
910 
911 		switch (req) {
912 		case PT_STEP:
913 			error = ptrace_single_step(td2);
914 			if (error)
915 				goto out;
916 			break;
917 		case PT_CONTINUE:
918 		case PT_TO_SCE:
919 		case PT_TO_SCX:
920 		case PT_SYSCALL:
921 			if (addr != (void *)1) {
922 				error = ptrace_set_pc(td2,
923 				    (u_long)(uintfptr_t)addr);
924 				if (error)
925 					goto out;
926 			}
927 			switch (req) {
928 			case PT_TO_SCE:
929 				p->p_stops |= S_PT_SCE;
930 				break;
931 			case PT_TO_SCX:
932 				p->p_stops |= S_PT_SCX;
933 				break;
934 			case PT_SYSCALL:
935 				p->p_stops |= S_PT_SCE | S_PT_SCX;
936 				break;
937 			}
938 			break;
939 		case PT_DETACH:
940 			/* reset process parent */
941 			if (p->p_oppid != p->p_pptr->p_pid) {
942 				struct proc *pp;
943 
944 				PROC_LOCK(p->p_pptr);
945 				sigqueue_take(p->p_ksi);
946 				PROC_UNLOCK(p->p_pptr);
947 
948 				PROC_UNLOCK(p);
949 				pp = pfind(p->p_oppid);
950 				if (pp == NULL)
951 					pp = initproc;
952 				else
953 					PROC_UNLOCK(pp);
954 				PROC_LOCK(p);
955 				proc_reparent(p, pp);
956 				if (pp == initproc)
957 					p->p_sigparent = SIGCHLD;
958 			}
959 			p->p_flag &= ~(P_TRACED | P_WAITED);
960 			p->p_oppid = 0;
961 
962 			/* should we send SIGCHLD? */
963 			/* childproc_continued(p); */
964 			break;
965 		}
966 
967 	sendsig:
968 		if (proctree_locked) {
969 			sx_xunlock(&proctree_lock);
970 			proctree_locked = 0;
971 		}
972 		p->p_xstat = data;
973 		p->p_xthread = NULL;
974 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
975 			/* deliver or queue signal */
976 			td2->td_dbgflags &= ~TDB_XSIG;
977 			td2->td_xsig = data;
978 
979 			if (req == PT_DETACH) {
980 				struct thread *td3;
981 				FOREACH_THREAD_IN_PROC(p, td3) {
982 					td3->td_dbgflags &= ~TDB_SUSPEND;
983 				}
984 			}
985 			/*
986 			 * unsuspend all threads, to not let a thread run,
987 			 * you should use PT_SUSPEND to suspend it before
988 			 * continuing process.
989 			 */
990 			PROC_SLOCK(p);
991 			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
992 			thread_unsuspend(p);
993 			PROC_SUNLOCK(p);
994 		} else {
995 			if (data)
996 				psignal(p, data);
997 		}
998 		break;
999 
1000 	case PT_WRITE_I:
1001 	case PT_WRITE_D:
1002 		td2->td_dbgflags |= TDB_USERWR;
1003 		write = 1;
1004 		/* FALLTHROUGH */
1005 	case PT_READ_I:
1006 	case PT_READ_D:
1007 		PROC_UNLOCK(p);
1008 		tmp = 0;
1009 		/* write = 0 set above */
1010 		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1011 		iov.iov_len = sizeof(int);
1012 		uio.uio_iov = &iov;
1013 		uio.uio_iovcnt = 1;
1014 		uio.uio_offset = (off_t)(uintptr_t)addr;
1015 		uio.uio_resid = sizeof(int);
1016 		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
1017 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1018 		uio.uio_td = td;
1019 		error = proc_rwmem(p, &uio);
1020 		if (uio.uio_resid != 0) {
1021 			/*
1022 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
1023 			 * so I think write() can bogusly return 0.
1024 			 * XXX what happens for short writes?  We don't want
1025 			 * to write partial data.
1026 			 * XXX proc_rwmem() returns EPERM for other invalid
1027 			 * addresses.  Convert this to EINVAL.  Does this
1028 			 * clobber returns of EPERM for other reasons?
1029 			 */
1030 			if (error == 0 || error == ENOSPC || error == EPERM)
1031 				error = EINVAL;	/* EOF */
1032 		}
1033 		if (!write)
1034 			td->td_retval[0] = tmp;
1035 		PROC_LOCK(p);
1036 		break;
1037 
1038 	case PT_IO:
1039 #ifdef COMPAT_FREEBSD32
1040 		if (wrap32) {
1041 			piod32 = addr;
1042 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1043 			iov.iov_len = piod32->piod_len;
1044 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1045 			uio.uio_resid = piod32->piod_len;
1046 		} else
1047 #endif
1048 		{
1049 			piod = addr;
1050 			iov.iov_base = piod->piod_addr;
1051 			iov.iov_len = piod->piod_len;
1052 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1053 			uio.uio_resid = piod->piod_len;
1054 		}
1055 		uio.uio_iov = &iov;
1056 		uio.uio_iovcnt = 1;
1057 		uio.uio_segflg = UIO_USERSPACE;
1058 		uio.uio_td = td;
1059 #ifdef COMPAT_FREEBSD32
1060 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1061 #else
1062 		tmp = piod->piod_op;
1063 #endif
1064 		switch (tmp) {
1065 		case PIOD_READ_D:
1066 		case PIOD_READ_I:
1067 			uio.uio_rw = UIO_READ;
1068 			break;
1069 		case PIOD_WRITE_D:
1070 		case PIOD_WRITE_I:
1071 			td2->td_dbgflags |= TDB_USERWR;
1072 			uio.uio_rw = UIO_WRITE;
1073 			break;
1074 		default:
1075 			error = EINVAL;
1076 			goto out;
1077 		}
1078 		PROC_UNLOCK(p);
1079 		error = proc_rwmem(p, &uio);
1080 #ifdef COMPAT_FREEBSD32
1081 		if (wrap32)
1082 			piod32->piod_len -= uio.uio_resid;
1083 		else
1084 #endif
1085 			piod->piod_len -= uio.uio_resid;
1086 		PROC_LOCK(p);
1087 		break;
1088 
1089 	case PT_KILL:
1090 		data = SIGKILL;
1091 		goto sendsig;	/* in PT_CONTINUE above */
1092 
1093 	case PT_SETREGS:
1094 		td2->td_dbgflags |= TDB_USERWR;
1095 		error = PROC_WRITE(regs, td2, addr);
1096 		break;
1097 
1098 	case PT_GETREGS:
1099 		error = PROC_READ(regs, td2, addr);
1100 		break;
1101 
1102 	case PT_SETFPREGS:
1103 		td2->td_dbgflags |= TDB_USERWR;
1104 		error = PROC_WRITE(fpregs, td2, addr);
1105 		break;
1106 
1107 	case PT_GETFPREGS:
1108 		error = PROC_READ(fpregs, td2, addr);
1109 		break;
1110 
1111 	case PT_SETDBREGS:
1112 		td2->td_dbgflags |= TDB_USERWR;
1113 		error = PROC_WRITE(dbregs, td2, addr);
1114 		break;
1115 
1116 	case PT_GETDBREGS:
1117 		error = PROC_READ(dbregs, td2, addr);
1118 		break;
1119 
1120 	case PT_LWPINFO:
1121 		if (data <= 0 ||
1122 #ifdef COMPAT_FREEBSD32
1123 		    (!wrap32 && data > sizeof(*pl)) ||
1124 		    (wrap32 && data > sizeof(*pl32))) {
1125 #else
1126 		    data > sizeof(*pl)) {
1127 #endif
1128 			error = EINVAL;
1129 			break;
1130 		}
1131 #ifdef COMPAT_FREEBSD32
1132 		if (wrap32) {
1133 			pl = &plr;
1134 			pl32 = addr;
1135 		} else
1136 #endif
1137 		pl = addr;
1138 		pl->pl_lwpid = td2->td_tid;
1139 		pl->pl_flags = 0;
1140 		if (td2->td_dbgflags & TDB_XSIG) {
1141 			pl->pl_event = PL_EVENT_SIGNAL;
1142 			if (td2->td_dbgksi.ksi_signo != 0 &&
1143 #ifdef COMPAT_FREEBSD32
1144 			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1145 			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1146 			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1147 			    pl_siginfo) + sizeof(struct siginfo32)))
1148 #else
1149 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1150 			    + sizeof(pl->pl_siginfo)
1151 #endif
1152 			){
1153 				pl->pl_flags |= PL_FLAG_SI;
1154 				pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1155 			}
1156 		}
1157 		if ((pl->pl_flags & PL_FLAG_SI) == 0)
1158 			bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1159 		if (td2->td_dbgflags & TDB_SCE)
1160 			pl->pl_flags |= PL_FLAG_SCE;
1161 		else if (td2->td_dbgflags & TDB_SCX)
1162 			pl->pl_flags |= PL_FLAG_SCX;
1163 		if (td2->td_dbgflags & TDB_EXEC)
1164 			pl->pl_flags |= PL_FLAG_EXEC;
1165 		pl->pl_sigmask = td2->td_sigmask;
1166 		pl->pl_siglist = td2->td_siglist;
1167 #ifdef COMPAT_FREEBSD32
1168 		if (wrap32)
1169 			ptrace_lwpinfo_to32(pl, pl32);
1170 #endif
1171 		break;
1172 
1173 	case PT_GETNUMLWPS:
1174 		td->td_retval[0] = p->p_numthreads;
1175 		break;
1176 
1177 	case PT_GETLWPLIST:
1178 		if (data <= 0) {
1179 			error = EINVAL;
1180 			break;
1181 		}
1182 		num = imin(p->p_numthreads, data);
1183 		PROC_UNLOCK(p);
1184 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1185 		tmp = 0;
1186 		PROC_LOCK(p);
1187 		FOREACH_THREAD_IN_PROC(p, td2) {
1188 			if (tmp >= num)
1189 				break;
1190 			buf[tmp++] = td2->td_tid;
1191 		}
1192 		PROC_UNLOCK(p);
1193 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1194 		free(buf, M_TEMP);
1195 		if (!error)
1196 			td->td_retval[0] = tmp;
1197 		PROC_LOCK(p);
1198 		break;
1199 
1200 	case PT_VM_TIMESTAMP:
1201 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1202 		break;
1203 
1204 	case PT_VM_ENTRY:
1205 		PROC_UNLOCK(p);
1206 #ifdef COMPAT_FREEBSD32
1207 		if (wrap32)
1208 			error = ptrace_vm_entry32(td, p, addr);
1209 		else
1210 #endif
1211 		error = ptrace_vm_entry(td, p, addr);
1212 		PROC_LOCK(p);
1213 		break;
1214 
1215 	default:
1216 #ifdef __HAVE_PTRACE_MACHDEP
1217 		if (req >= PT_FIRSTMACH) {
1218 			PROC_UNLOCK(p);
1219 			error = cpu_ptrace(td2, req, addr, data);
1220 			PROC_LOCK(p);
1221 		} else
1222 #endif
1223 			/* Unknown request. */
1224 			error = EINVAL;
1225 		break;
1226 	}
1227 
1228 out:
1229 	/* Drop our hold on this process now that the request has completed. */
1230 	_PRELE(p);
1231 fail:
1232 	PROC_UNLOCK(p);
1233 	if (proctree_locked)
1234 		sx_xunlock(&proctree_lock);
1235 	return (error);
1236 }
1237 #undef PROC_READ
1238 #undef PROC_WRITE
1239 
1240 /*
1241  * Stop a process because of a debugging event;
1242  * stay stopped until p->p_step is cleared
1243  * (cleared by PIOCCONT in procfs).
1244  */
1245 void
1246 stopevent(struct proc *p, unsigned int event, unsigned int val)
1247 {
1248 
1249 	PROC_LOCK_ASSERT(p, MA_OWNED);
1250 	p->p_step = 1;
1251 	do {
1252 		p->p_xstat = val;
1253 		p->p_xthread = NULL;
1254 		p->p_stype = event;	/* Which event caused the stop? */
1255 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1256 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1257 	} while (p->p_step);
1258 }
1259