xref: /freebsd/sys/kern/sys_process.c (revision 7f08176ee8b672feebd8a12572b43dddcb88046e)
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 	char	pl_tdname[MAXCOMLEN + 1];	/* LWP name. */
97 };
98 
99 #endif
100 
101 /*
102  * Functions implemented using PROC_ACTION():
103  *
104  * proc_read_regs(proc, regs)
105  *	Get the current user-visible register set from the process
106  *	and copy it into the regs structure (<machine/reg.h>).
107  *	The process is stopped at the time read_regs is called.
108  *
109  * proc_write_regs(proc, regs)
110  *	Update the current register set from the passed in regs
111  *	structure.  Take care to avoid clobbering special CPU
112  *	registers or privileged bits in the PSL.
113  *	Depending on the architecture this may have fix-up work to do,
114  *	especially if the IAR or PCW are modified.
115  *	The process is stopped at the time write_regs is called.
116  *
117  * proc_read_fpregs, proc_write_fpregs
118  *	deal with the floating point register set, otherwise as above.
119  *
120  * proc_read_dbregs, proc_write_dbregs
121  *	deal with the processor debug register set, otherwise as above.
122  *
123  * proc_sstep(proc)
124  *	Arrange for the process to trap after executing a single instruction.
125  */
126 
127 #define	PROC_ACTION(action) do {					\
128 	int error;							\
129 									\
130 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
131 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
132 		error = EIO;						\
133 	else								\
134 		error = (action);					\
135 	return (error);							\
136 } while(0)
137 
138 int
139 proc_read_regs(struct thread *td, struct reg *regs)
140 {
141 
142 	PROC_ACTION(fill_regs(td, regs));
143 }
144 
145 int
146 proc_write_regs(struct thread *td, struct reg *regs)
147 {
148 
149 	PROC_ACTION(set_regs(td, regs));
150 }
151 
152 int
153 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
154 {
155 
156 	PROC_ACTION(fill_dbregs(td, dbregs));
157 }
158 
159 int
160 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
161 {
162 
163 	PROC_ACTION(set_dbregs(td, dbregs));
164 }
165 
166 /*
167  * Ptrace doesn't support fpregs at all, and there are no security holes
168  * or translations for fpregs, so we can just copy them.
169  */
170 int
171 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
172 {
173 
174 	PROC_ACTION(fill_fpregs(td, fpregs));
175 }
176 
177 int
178 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
179 {
180 
181 	PROC_ACTION(set_fpregs(td, fpregs));
182 }
183 
184 #ifdef COMPAT_FREEBSD32
185 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
186 int
187 proc_read_regs32(struct thread *td, struct reg32 *regs32)
188 {
189 
190 	PROC_ACTION(fill_regs32(td, regs32));
191 }
192 
193 int
194 proc_write_regs32(struct thread *td, struct reg32 *regs32)
195 {
196 
197 	PROC_ACTION(set_regs32(td, regs32));
198 }
199 
200 int
201 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
202 {
203 
204 	PROC_ACTION(fill_dbregs32(td, dbregs32));
205 }
206 
207 int
208 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
209 {
210 
211 	PROC_ACTION(set_dbregs32(td, dbregs32));
212 }
213 
214 int
215 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
216 {
217 
218 	PROC_ACTION(fill_fpregs32(td, fpregs32));
219 }
220 
221 int
222 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
223 {
224 
225 	PROC_ACTION(set_fpregs32(td, fpregs32));
226 }
227 #endif
228 
229 int
230 proc_sstep(struct thread *td)
231 {
232 
233 	PROC_ACTION(ptrace_single_step(td));
234 }
235 
236 int
237 proc_rwmem(struct proc *p, struct uio *uio)
238 {
239 	vm_map_t map;
240 	vm_object_t backing_object, object;
241 	vm_offset_t pageno;		/* page number */
242 	vm_prot_t reqprot;
243 	int error, writing;
244 
245 	/*
246 	 * Assert that someone has locked this vmspace.  (Should be
247 	 * curthread but we can't assert that.)  This keeps the process
248 	 * from exiting out from under us until this operation completes.
249 	 */
250 	KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
251 	    p, p->p_pid));
252 
253 	/*
254 	 * The map we want...
255 	 */
256 	map = &p->p_vmspace->vm_map;
257 
258 	writing = uio->uio_rw == UIO_WRITE;
259 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
260 
261 	/*
262 	 * Only map in one page at a time.  We don't have to, but it
263 	 * makes things easier.  This way is trivial - right?
264 	 */
265 	do {
266 		vm_map_t tmap;
267 		vm_offset_t uva;
268 		int page_offset;		/* offset into page */
269 		vm_map_entry_t out_entry;
270 		vm_prot_t out_prot;
271 		boolean_t wired;
272 		vm_pindex_t pindex;
273 		u_int len;
274 		vm_page_t m;
275 
276 		object = NULL;
277 
278 		uva = (vm_offset_t)uio->uio_offset;
279 
280 		/*
281 		 * Get the page number of this segment.
282 		 */
283 		pageno = trunc_page(uva);
284 		page_offset = uva - pageno;
285 
286 		/*
287 		 * How many bytes to copy
288 		 */
289 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
290 
291 		/*
292 		 * Fault the page on behalf of the process
293 		 */
294 		error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
295 		if (error) {
296 			if (error == KERN_RESOURCE_SHORTAGE)
297 				error = ENOMEM;
298 			else
299 				error = EFAULT;
300 			break;
301 		}
302 
303 		/*
304 		 * Now we need to get the page.  out_entry and wired
305 		 * aren't used.  One would think the vm code
306 		 * would be a *bit* nicer...  We use tmap because
307 		 * vm_map_lookup() can change the map argument.
308 		 */
309 		tmap = map;
310 		error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
311 		    &object, &pindex, &out_prot, &wired);
312 		if (error) {
313 			error = EFAULT;
314 			break;
315 		}
316 		VM_OBJECT_LOCK(object);
317 		while ((m = vm_page_lookup(object, pindex)) == NULL &&
318 		    !writing &&
319 		    (backing_object = object->backing_object) != NULL) {
320 			/*
321 			 * Allow fallback to backing objects if we are reading.
322 			 */
323 			VM_OBJECT_LOCK(backing_object);
324 			pindex += OFF_TO_IDX(object->backing_object_offset);
325 			VM_OBJECT_UNLOCK(object);
326 			object = backing_object;
327 		}
328 		if (writing && m != NULL) {
329 			vm_page_dirty(m);
330 			vm_pager_page_unswapped(m);
331 		}
332 		VM_OBJECT_UNLOCK(object);
333 		if (m == NULL) {
334 			vm_map_lookup_done(tmap, out_entry);
335 			error = EFAULT;
336 			break;
337 		}
338 
339 		/*
340 		 * Hold the page in memory.
341 		 */
342 		vm_page_lock(m);
343 		vm_page_hold(m);
344 		vm_page_unlock(m);
345 
346 		/*
347 		 * We're done with tmap now.
348 		 */
349 		vm_map_lookup_done(tmap, out_entry);
350 
351 		/*
352 		 * Now do the i/o move.
353 		 */
354 		error = uiomove_fromphys(&m, page_offset, len, uio);
355 
356 		/* Make the I-cache coherent for breakpoints. */
357 		if (!error && writing && (out_prot & VM_PROT_EXECUTE))
358 			vm_sync_icache(map, uva, len);
359 
360 		/*
361 		 * Release the page.
362 		 */
363 		vm_page_lock(m);
364 		vm_page_unhold(m);
365 		vm_page_unlock(m);
366 
367 	} while (error == 0 && uio->uio_resid > 0);
368 
369 	return (error);
370 }
371 
372 static int
373 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
374 {
375 	struct vattr vattr;
376 	vm_map_t map;
377 	vm_map_entry_t entry;
378 	vm_object_t obj, tobj, lobj;
379 	struct vmspace *vm;
380 	struct vnode *vp;
381 	char *freepath, *fullpath;
382 	u_int pathlen;
383 	int error, index, vfslocked;
384 
385 	error = 0;
386 	obj = NULL;
387 
388 	vm = vmspace_acquire_ref(p);
389 	map = &vm->vm_map;
390 	vm_map_lock_read(map);
391 
392 	do {
393 		entry = map->header.next;
394 		index = 0;
395 		while (index < pve->pve_entry && entry != &map->header) {
396 			entry = entry->next;
397 			index++;
398 		}
399 		if (index != pve->pve_entry) {
400 			error = EINVAL;
401 			break;
402 		}
403 		while (entry != &map->header &&
404 		    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
405 			entry = entry->next;
406 			index++;
407 		}
408 		if (entry == &map->header) {
409 			error = ENOENT;
410 			break;
411 		}
412 
413 		/* We got an entry. */
414 		pve->pve_entry = index + 1;
415 		pve->pve_timestamp = map->timestamp;
416 		pve->pve_start = entry->start;
417 		pve->pve_end = entry->end - 1;
418 		pve->pve_offset = entry->offset;
419 		pve->pve_prot = entry->protection;
420 
421 		/* Backing object's path needed? */
422 		if (pve->pve_pathlen == 0)
423 			break;
424 
425 		pathlen = pve->pve_pathlen;
426 		pve->pve_pathlen = 0;
427 
428 		obj = entry->object.vm_object;
429 		if (obj != NULL)
430 			VM_OBJECT_LOCK(obj);
431 	} while (0);
432 
433 	vm_map_unlock_read(map);
434 	vmspace_free(vm);
435 
436 	pve->pve_fsid = VNOVAL;
437 	pve->pve_fileid = VNOVAL;
438 
439 	if (error == 0 && obj != NULL) {
440 		lobj = obj;
441 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
442 			if (tobj != obj)
443 				VM_OBJECT_LOCK(tobj);
444 			if (lobj != obj)
445 				VM_OBJECT_UNLOCK(lobj);
446 			lobj = tobj;
447 			pve->pve_offset += tobj->backing_object_offset;
448 		}
449 		vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
450 		if (vp != NULL)
451 			vref(vp);
452 		if (lobj != obj)
453 			VM_OBJECT_UNLOCK(lobj);
454 		VM_OBJECT_UNLOCK(obj);
455 
456 		if (vp != NULL) {
457 			freepath = NULL;
458 			fullpath = NULL;
459 			vn_fullpath(td, vp, &fullpath, &freepath);
460 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
461 			vn_lock(vp, LK_SHARED | LK_RETRY);
462 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
463 				pve->pve_fileid = vattr.va_fileid;
464 				pve->pve_fsid = vattr.va_fsid;
465 			}
466 			vput(vp);
467 			VFS_UNLOCK_GIANT(vfslocked);
468 
469 			if (fullpath != NULL) {
470 				pve->pve_pathlen = strlen(fullpath) + 1;
471 				if (pve->pve_pathlen <= pathlen) {
472 					error = copyout(fullpath, pve->pve_path,
473 					    pve->pve_pathlen);
474 				} else
475 					error = ENAMETOOLONG;
476 			}
477 			if (freepath != NULL)
478 				free(freepath, M_TEMP);
479 		}
480 	}
481 
482 	return (error);
483 }
484 
485 #ifdef COMPAT_FREEBSD32
486 static int
487 ptrace_vm_entry32(struct thread *td, struct proc *p,
488     struct ptrace_vm_entry32 *pve32)
489 {
490 	struct ptrace_vm_entry pve;
491 	int error;
492 
493 	pve.pve_entry = pve32->pve_entry;
494 	pve.pve_pathlen = pve32->pve_pathlen;
495 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
496 
497 	error = ptrace_vm_entry(td, p, &pve);
498 	if (error == 0) {
499 		pve32->pve_entry = pve.pve_entry;
500 		pve32->pve_timestamp = pve.pve_timestamp;
501 		pve32->pve_start = pve.pve_start;
502 		pve32->pve_end = pve.pve_end;
503 		pve32->pve_offset = pve.pve_offset;
504 		pve32->pve_prot = pve.pve_prot;
505 		pve32->pve_fileid = pve.pve_fileid;
506 		pve32->pve_fsid = pve.pve_fsid;
507 	}
508 
509 	pve32->pve_pathlen = pve.pve_pathlen;
510 	return (error);
511 }
512 
513 static void
514 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
515     struct ptrace_lwpinfo32 *pl32)
516 {
517 
518 	pl32->pl_lwpid = pl->pl_lwpid;
519 	pl32->pl_event = pl->pl_event;
520 	pl32->pl_flags = pl->pl_flags;
521 	pl32->pl_sigmask = pl->pl_sigmask;
522 	pl32->pl_siglist = pl->pl_siglist;
523 	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
524 	strcpy(pl32->pl_tdname, pl->pl_tdname);
525 }
526 #endif /* COMPAT_FREEBSD32 */
527 
528 /*
529  * Process debugging system call.
530  */
531 #ifndef _SYS_SYSPROTO_H_
532 struct ptrace_args {
533 	int	req;
534 	pid_t	pid;
535 	caddr_t	addr;
536 	int	data;
537 };
538 #endif
539 
540 #ifdef COMPAT_FREEBSD32
541 /*
542  * This CPP subterfuge is to try and reduce the number of ifdefs in
543  * the body of the code.
544  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
545  * becomes either:
546  *   copyin(uap->addr, &r.reg, sizeof r.reg);
547  * or
548  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
549  * .. except this is done at runtime.
550  */
551 #define	COPYIN(u, k, s)		wrap32 ? \
552 	copyin(u, k ## 32, s ## 32) : \
553 	copyin(u, k, s)
554 #define	COPYOUT(k, u, s)	wrap32 ? \
555 	copyout(k ## 32, u, s ## 32) : \
556 	copyout(k, u, s)
557 #else
558 #define	COPYIN(u, k, s)		copyin(u, k, s)
559 #define	COPYOUT(k, u, s)	copyout(k, u, s)
560 #endif
561 int
562 ptrace(struct thread *td, struct ptrace_args *uap)
563 {
564 	/*
565 	 * XXX this obfuscation is to reduce stack usage, but the register
566 	 * structs may be too large to put on the stack anyway.
567 	 */
568 	union {
569 		struct ptrace_io_desc piod;
570 		struct ptrace_lwpinfo pl;
571 		struct ptrace_vm_entry pve;
572 		struct dbreg dbreg;
573 		struct fpreg fpreg;
574 		struct reg reg;
575 #ifdef COMPAT_FREEBSD32
576 		struct dbreg32 dbreg32;
577 		struct fpreg32 fpreg32;
578 		struct reg32 reg32;
579 		struct ptrace_io_desc32 piod32;
580 		struct ptrace_lwpinfo32 pl32;
581 		struct ptrace_vm_entry32 pve32;
582 #endif
583 	} r;
584 	void *addr;
585 	int error = 0;
586 #ifdef COMPAT_FREEBSD32
587 	int wrap32 = 0;
588 
589 	if (SV_CURPROC_FLAG(SV_ILP32))
590 		wrap32 = 1;
591 #endif
592 	AUDIT_ARG_PID(uap->pid);
593 	AUDIT_ARG_CMD(uap->req);
594 	AUDIT_ARG_VALUE(uap->data);
595 	addr = &r;
596 	switch (uap->req) {
597 	case PT_GETREGS:
598 	case PT_GETFPREGS:
599 	case PT_GETDBREGS:
600 	case PT_LWPINFO:
601 		break;
602 	case PT_SETREGS:
603 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
604 		break;
605 	case PT_SETFPREGS:
606 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
607 		break;
608 	case PT_SETDBREGS:
609 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
610 		break;
611 	case PT_IO:
612 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
613 		break;
614 	case PT_VM_ENTRY:
615 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
616 		break;
617 	default:
618 		addr = uap->addr;
619 		break;
620 	}
621 	if (error)
622 		return (error);
623 
624 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
625 	if (error)
626 		return (error);
627 
628 	switch (uap->req) {
629 	case PT_VM_ENTRY:
630 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
631 		break;
632 	case PT_IO:
633 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
634 		break;
635 	case PT_GETREGS:
636 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
637 		break;
638 	case PT_GETFPREGS:
639 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
640 		break;
641 	case PT_GETDBREGS:
642 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
643 		break;
644 	case PT_LWPINFO:
645 		error = copyout(&r.pl, uap->addr, uap->data);
646 		break;
647 	}
648 
649 	return (error);
650 }
651 #undef COPYIN
652 #undef COPYOUT
653 
654 #ifdef COMPAT_FREEBSD32
655 /*
656  *   PROC_READ(regs, td2, addr);
657  * becomes either:
658  *   proc_read_regs(td2, addr);
659  * or
660  *   proc_read_regs32(td2, addr);
661  * .. except this is done at runtime.  There is an additional
662  * complication in that PROC_WRITE disallows 32 bit consumers
663  * from writing to 64 bit address space targets.
664  */
665 #define	PROC_READ(w, t, a)	wrap32 ? \
666 	proc_read_ ## w ## 32(t, a) : \
667 	proc_read_ ## w (t, a)
668 #define	PROC_WRITE(w, t, a)	wrap32 ? \
669 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
670 	proc_write_ ## w (t, a)
671 #else
672 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
673 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
674 #endif
675 
676 int
677 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
678 {
679 	struct iovec iov;
680 	struct uio uio;
681 	struct proc *curp, *p, *pp;
682 	struct thread *td2 = NULL;
683 	struct ptrace_io_desc *piod = NULL;
684 	struct ptrace_lwpinfo *pl;
685 	int error, write, tmp, num;
686 	int proctree_locked = 0;
687 	lwpid_t tid = 0, *buf;
688 #ifdef COMPAT_FREEBSD32
689 	int wrap32 = 0, safe = 0;
690 	struct ptrace_io_desc32 *piod32 = NULL;
691 	struct ptrace_lwpinfo32 *pl32 = NULL;
692 	struct ptrace_lwpinfo plr;
693 #endif
694 
695 	curp = td->td_proc;
696 
697 	/* Lock proctree before locking the process. */
698 	switch (req) {
699 	case PT_TRACE_ME:
700 	case PT_ATTACH:
701 	case PT_STEP:
702 	case PT_CONTINUE:
703 	case PT_TO_SCE:
704 	case PT_TO_SCX:
705 	case PT_SYSCALL:
706 	case PT_DETACH:
707 		sx_xlock(&proctree_lock);
708 		proctree_locked = 1;
709 		break;
710 	default:
711 		break;
712 	}
713 
714 	write = 0;
715 	if (req == PT_TRACE_ME) {
716 		p = td->td_proc;
717 		PROC_LOCK(p);
718 	} else {
719 		if (pid <= PID_MAX) {
720 			if ((p = pfind(pid)) == NULL) {
721 				if (proctree_locked)
722 					sx_xunlock(&proctree_lock);
723 				return (ESRCH);
724 			}
725 		} else {
726 			td2 = tdfind(pid, -1);
727 			if (td2 == NULL) {
728 				if (proctree_locked)
729 					sx_xunlock(&proctree_lock);
730 				return (ESRCH);
731 			}
732 			p = td2->td_proc;
733 			tid = pid;
734 			pid = p->p_pid;
735 		}
736 	}
737 	AUDIT_ARG_PROCESS(p);
738 
739 	if ((p->p_flag & P_WEXIT) != 0) {
740 		error = ESRCH;
741 		goto fail;
742 	}
743 	if ((error = p_cansee(td, p)) != 0)
744 		goto fail;
745 
746 	if ((error = p_candebug(td, p)) != 0)
747 		goto fail;
748 
749 	/*
750 	 * System processes can't be debugged.
751 	 */
752 	if ((p->p_flag & P_SYSTEM) != 0) {
753 		error = EINVAL;
754 		goto fail;
755 	}
756 
757 	if (tid == 0) {
758 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
759 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
760 			td2 = p->p_xthread;
761 		} else {
762 			td2 = FIRST_THREAD_IN_PROC(p);
763 		}
764 		tid = td2->td_tid;
765 	}
766 
767 #ifdef COMPAT_FREEBSD32
768 	/*
769 	 * Test if we're a 32 bit client and what the target is.
770 	 * Set the wrap controls accordingly.
771 	 */
772 	if (SV_CURPROC_FLAG(SV_ILP32)) {
773 		if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
774 			safe = 1;
775 		wrap32 = 1;
776 	}
777 #endif
778 	/*
779 	 * Permissions check
780 	 */
781 	switch (req) {
782 	case PT_TRACE_ME:
783 		/* Always legal. */
784 		break;
785 
786 	case PT_ATTACH:
787 		/* Self */
788 		if (p->p_pid == td->td_proc->p_pid) {
789 			error = EINVAL;
790 			goto fail;
791 		}
792 
793 		/* Already traced */
794 		if (p->p_flag & P_TRACED) {
795 			error = EBUSY;
796 			goto fail;
797 		}
798 
799 		/* Can't trace an ancestor if you're being traced. */
800 		if (curp->p_flag & P_TRACED) {
801 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
802 				if (pp == p) {
803 					error = EINVAL;
804 					goto fail;
805 				}
806 			}
807 		}
808 
809 
810 		/* OK */
811 		break;
812 
813 	case PT_CLEARSTEP:
814 		/* Allow thread to clear single step for itself */
815 		if (td->td_tid == tid)
816 			break;
817 
818 		/* FALLTHROUGH */
819 	default:
820 		/* not being traced... */
821 		if ((p->p_flag & P_TRACED) == 0) {
822 			error = EPERM;
823 			goto fail;
824 		}
825 
826 		/* not being traced by YOU */
827 		if (p->p_pptr != td->td_proc) {
828 			error = EBUSY;
829 			goto fail;
830 		}
831 
832 		/* not currently stopped */
833 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
834 		    p->p_suspcount != p->p_numthreads  ||
835 		    (p->p_flag & P_WAITED) == 0) {
836 			error = EBUSY;
837 			goto fail;
838 		}
839 
840 		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
841 			static int count = 0;
842 			if (count++ == 0)
843 				printf("P_STOPPED_TRACE not set.\n");
844 		}
845 
846 		/* OK */
847 		break;
848 	}
849 
850 	/* Keep this process around until we finish this request. */
851 	_PHOLD(p);
852 
853 #ifdef FIX_SSTEP
854 	/*
855 	 * Single step fixup ala procfs
856 	 */
857 	FIX_SSTEP(td2);
858 #endif
859 
860 	/*
861 	 * Actually do the requests
862 	 */
863 
864 	td->td_retval[0] = 0;
865 
866 	switch (req) {
867 	case PT_TRACE_ME:
868 		/* set my trace flag and "owner" so it can read/write me */
869 		p->p_flag |= P_TRACED;
870 		p->p_oppid = p->p_pptr->p_pid;
871 		break;
872 
873 	case PT_ATTACH:
874 		/* security check done above */
875 		p->p_flag |= P_TRACED;
876 		p->p_oppid = p->p_pptr->p_pid;
877 		if (p->p_pptr != td->td_proc)
878 			proc_reparent(p, td->td_proc);
879 		data = SIGSTOP;
880 		goto sendsig;	/* in PT_CONTINUE below */
881 
882 	case PT_CLEARSTEP:
883 		error = ptrace_clear_single_step(td2);
884 		break;
885 
886 	case PT_SETSTEP:
887 		error = ptrace_single_step(td2);
888 		break;
889 
890 	case PT_SUSPEND:
891 		td2->td_dbgflags |= TDB_SUSPEND;
892 		thread_lock(td2);
893 		td2->td_flags |= TDF_NEEDSUSPCHK;
894 		thread_unlock(td2);
895 		break;
896 
897 	case PT_RESUME:
898 		td2->td_dbgflags &= ~TDB_SUSPEND;
899 		break;
900 
901 	case PT_STEP:
902 	case PT_CONTINUE:
903 	case PT_TO_SCE:
904 	case PT_TO_SCX:
905 	case PT_SYSCALL:
906 	case PT_DETACH:
907 		/* Zero means do not send any signal */
908 		if (data < 0 || data > _SIG_MAXSIG) {
909 			error = EINVAL;
910 			break;
911 		}
912 
913 		switch (req) {
914 		case PT_STEP:
915 			error = ptrace_single_step(td2);
916 			if (error)
917 				goto out;
918 			break;
919 		case PT_CONTINUE:
920 		case PT_TO_SCE:
921 		case PT_TO_SCX:
922 		case PT_SYSCALL:
923 			if (addr != (void *)1) {
924 				error = ptrace_set_pc(td2,
925 				    (u_long)(uintfptr_t)addr);
926 				if (error)
927 					goto out;
928 			}
929 			switch (req) {
930 			case PT_TO_SCE:
931 				p->p_stops |= S_PT_SCE;
932 				break;
933 			case PT_TO_SCX:
934 				p->p_stops |= S_PT_SCX;
935 				break;
936 			case PT_SYSCALL:
937 				p->p_stops |= S_PT_SCE | S_PT_SCX;
938 				break;
939 			}
940 			break;
941 		case PT_DETACH:
942 			/* reset process parent */
943 			if (p->p_oppid != p->p_pptr->p_pid) {
944 				struct proc *pp;
945 
946 				PROC_LOCK(p->p_pptr);
947 				sigqueue_take(p->p_ksi);
948 				PROC_UNLOCK(p->p_pptr);
949 
950 				PROC_UNLOCK(p);
951 				pp = pfind(p->p_oppid);
952 				if (pp == NULL)
953 					pp = initproc;
954 				else
955 					PROC_UNLOCK(pp);
956 				PROC_LOCK(p);
957 				proc_reparent(p, pp);
958 				if (pp == initproc)
959 					p->p_sigparent = SIGCHLD;
960 			}
961 			p->p_flag &= ~(P_TRACED | P_WAITED);
962 			p->p_oppid = 0;
963 
964 			/* should we send SIGCHLD? */
965 			/* childproc_continued(p); */
966 			break;
967 		}
968 
969 	sendsig:
970 		if (proctree_locked) {
971 			sx_xunlock(&proctree_lock);
972 			proctree_locked = 0;
973 		}
974 		p->p_xstat = data;
975 		p->p_xthread = NULL;
976 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
977 			/* deliver or queue signal */
978 			td2->td_dbgflags &= ~TDB_XSIG;
979 			td2->td_xsig = data;
980 
981 			if (req == PT_DETACH) {
982 				struct thread *td3;
983 				FOREACH_THREAD_IN_PROC(p, td3) {
984 					td3->td_dbgflags &= ~TDB_SUSPEND;
985 				}
986 			}
987 			/*
988 			 * unsuspend all threads, to not let a thread run,
989 			 * you should use PT_SUSPEND to suspend it before
990 			 * continuing process.
991 			 */
992 			PROC_SLOCK(p);
993 			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
994 			thread_unsuspend(p);
995 			PROC_SUNLOCK(p);
996 		} else {
997 			if (data)
998 				psignal(p, data);
999 		}
1000 		break;
1001 
1002 	case PT_WRITE_I:
1003 	case PT_WRITE_D:
1004 		td2->td_dbgflags |= TDB_USERWR;
1005 		write = 1;
1006 		/* FALLTHROUGH */
1007 	case PT_READ_I:
1008 	case PT_READ_D:
1009 		PROC_UNLOCK(p);
1010 		tmp = 0;
1011 		/* write = 0 set above */
1012 		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1013 		iov.iov_len = sizeof(int);
1014 		uio.uio_iov = &iov;
1015 		uio.uio_iovcnt = 1;
1016 		uio.uio_offset = (off_t)(uintptr_t)addr;
1017 		uio.uio_resid = sizeof(int);
1018 		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
1019 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1020 		uio.uio_td = td;
1021 		error = proc_rwmem(p, &uio);
1022 		if (uio.uio_resid != 0) {
1023 			/*
1024 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
1025 			 * so I think write() can bogusly return 0.
1026 			 * XXX what happens for short writes?  We don't want
1027 			 * to write partial data.
1028 			 * XXX proc_rwmem() returns EPERM for other invalid
1029 			 * addresses.  Convert this to EINVAL.  Does this
1030 			 * clobber returns of EPERM for other reasons?
1031 			 */
1032 			if (error == 0 || error == ENOSPC || error == EPERM)
1033 				error = EINVAL;	/* EOF */
1034 		}
1035 		if (!write)
1036 			td->td_retval[0] = tmp;
1037 		PROC_LOCK(p);
1038 		break;
1039 
1040 	case PT_IO:
1041 #ifdef COMPAT_FREEBSD32
1042 		if (wrap32) {
1043 			piod32 = addr;
1044 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1045 			iov.iov_len = piod32->piod_len;
1046 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1047 			uio.uio_resid = piod32->piod_len;
1048 		} else
1049 #endif
1050 		{
1051 			piod = addr;
1052 			iov.iov_base = piod->piod_addr;
1053 			iov.iov_len = piod->piod_len;
1054 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1055 			uio.uio_resid = piod->piod_len;
1056 		}
1057 		uio.uio_iov = &iov;
1058 		uio.uio_iovcnt = 1;
1059 		uio.uio_segflg = UIO_USERSPACE;
1060 		uio.uio_td = td;
1061 #ifdef COMPAT_FREEBSD32
1062 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1063 #else
1064 		tmp = piod->piod_op;
1065 #endif
1066 		switch (tmp) {
1067 		case PIOD_READ_D:
1068 		case PIOD_READ_I:
1069 			uio.uio_rw = UIO_READ;
1070 			break;
1071 		case PIOD_WRITE_D:
1072 		case PIOD_WRITE_I:
1073 			td2->td_dbgflags |= TDB_USERWR;
1074 			uio.uio_rw = UIO_WRITE;
1075 			break;
1076 		default:
1077 			error = EINVAL;
1078 			goto out;
1079 		}
1080 		PROC_UNLOCK(p);
1081 		error = proc_rwmem(p, &uio);
1082 #ifdef COMPAT_FREEBSD32
1083 		if (wrap32)
1084 			piod32->piod_len -= uio.uio_resid;
1085 		else
1086 #endif
1087 			piod->piod_len -= uio.uio_resid;
1088 		PROC_LOCK(p);
1089 		break;
1090 
1091 	case PT_KILL:
1092 		data = SIGKILL;
1093 		goto sendsig;	/* in PT_CONTINUE above */
1094 
1095 	case PT_SETREGS:
1096 		td2->td_dbgflags |= TDB_USERWR;
1097 		error = PROC_WRITE(regs, td2, addr);
1098 		break;
1099 
1100 	case PT_GETREGS:
1101 		error = PROC_READ(regs, td2, addr);
1102 		break;
1103 
1104 	case PT_SETFPREGS:
1105 		td2->td_dbgflags |= TDB_USERWR;
1106 		error = PROC_WRITE(fpregs, td2, addr);
1107 		break;
1108 
1109 	case PT_GETFPREGS:
1110 		error = PROC_READ(fpregs, td2, addr);
1111 		break;
1112 
1113 	case PT_SETDBREGS:
1114 		td2->td_dbgflags |= TDB_USERWR;
1115 		error = PROC_WRITE(dbregs, td2, addr);
1116 		break;
1117 
1118 	case PT_GETDBREGS:
1119 		error = PROC_READ(dbregs, td2, addr);
1120 		break;
1121 
1122 	case PT_LWPINFO:
1123 		if (data <= 0 ||
1124 #ifdef COMPAT_FREEBSD32
1125 		    (!wrap32 && data > sizeof(*pl)) ||
1126 		    (wrap32 && data > sizeof(*pl32))) {
1127 #else
1128 		    data > sizeof(*pl)) {
1129 #endif
1130 			error = EINVAL;
1131 			break;
1132 		}
1133 #ifdef COMPAT_FREEBSD32
1134 		if (wrap32) {
1135 			pl = &plr;
1136 			pl32 = addr;
1137 		} else
1138 #endif
1139 		pl = addr;
1140 		pl->pl_lwpid = td2->td_tid;
1141 		pl->pl_flags = 0;
1142 		if (td2->td_dbgflags & TDB_XSIG) {
1143 			pl->pl_event = PL_EVENT_SIGNAL;
1144 			if (td2->td_dbgksi.ksi_signo != 0 &&
1145 #ifdef COMPAT_FREEBSD32
1146 			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1147 			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1148 			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1149 			    pl_siginfo) + sizeof(struct siginfo32)))
1150 #else
1151 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1152 			    + sizeof(pl->pl_siginfo)
1153 #endif
1154 			){
1155 				pl->pl_flags |= PL_FLAG_SI;
1156 				pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1157 			}
1158 		}
1159 		if ((pl->pl_flags & PL_FLAG_SI) == 0)
1160 			bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1161 		if (td2->td_dbgflags & TDB_SCE)
1162 			pl->pl_flags |= PL_FLAG_SCE;
1163 		else if (td2->td_dbgflags & TDB_SCX)
1164 			pl->pl_flags |= PL_FLAG_SCX;
1165 		if (td2->td_dbgflags & TDB_EXEC)
1166 			pl->pl_flags |= PL_FLAG_EXEC;
1167 		pl->pl_sigmask = td2->td_sigmask;
1168 		pl->pl_siglist = td2->td_siglist;
1169 		strcpy(pl->pl_tdname, td2->td_name);
1170 #ifdef COMPAT_FREEBSD32
1171 		if (wrap32)
1172 			ptrace_lwpinfo_to32(pl, pl32);
1173 #endif
1174 		break;
1175 
1176 	case PT_GETNUMLWPS:
1177 		td->td_retval[0] = p->p_numthreads;
1178 		break;
1179 
1180 	case PT_GETLWPLIST:
1181 		if (data <= 0) {
1182 			error = EINVAL;
1183 			break;
1184 		}
1185 		num = imin(p->p_numthreads, data);
1186 		PROC_UNLOCK(p);
1187 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1188 		tmp = 0;
1189 		PROC_LOCK(p);
1190 		FOREACH_THREAD_IN_PROC(p, td2) {
1191 			if (tmp >= num)
1192 				break;
1193 			buf[tmp++] = td2->td_tid;
1194 		}
1195 		PROC_UNLOCK(p);
1196 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1197 		free(buf, M_TEMP);
1198 		if (!error)
1199 			td->td_retval[0] = tmp;
1200 		PROC_LOCK(p);
1201 		break;
1202 
1203 	case PT_VM_TIMESTAMP:
1204 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1205 		break;
1206 
1207 	case PT_VM_ENTRY:
1208 		PROC_UNLOCK(p);
1209 #ifdef COMPAT_FREEBSD32
1210 		if (wrap32)
1211 			error = ptrace_vm_entry32(td, p, addr);
1212 		else
1213 #endif
1214 		error = ptrace_vm_entry(td, p, addr);
1215 		PROC_LOCK(p);
1216 		break;
1217 
1218 	default:
1219 #ifdef __HAVE_PTRACE_MACHDEP
1220 		if (req >= PT_FIRSTMACH) {
1221 			PROC_UNLOCK(p);
1222 			error = cpu_ptrace(td2, req, addr, data);
1223 			PROC_LOCK(p);
1224 		} else
1225 #endif
1226 			/* Unknown request. */
1227 			error = EINVAL;
1228 		break;
1229 	}
1230 
1231 out:
1232 	/* Drop our hold on this process now that the request has completed. */
1233 	_PRELE(p);
1234 fail:
1235 	PROC_UNLOCK(p);
1236 	if (proctree_locked)
1237 		sx_xunlock(&proctree_lock);
1238 	return (error);
1239 }
1240 #undef PROC_READ
1241 #undef PROC_WRITE
1242 
1243 /*
1244  * Stop a process because of a debugging event;
1245  * stay stopped until p->p_step is cleared
1246  * (cleared by PIOCCONT in procfs).
1247  */
1248 void
1249 stopevent(struct proc *p, unsigned int event, unsigned int val)
1250 {
1251 
1252 	PROC_LOCK_ASSERT(p, MA_OWNED);
1253 	p->p_step = 1;
1254 	do {
1255 		p->p_xstat = val;
1256 		p->p_xthread = NULL;
1257 		p->p_stype = event;	/* Which event caused the stop? */
1258 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1259 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1260 	} while (p->p_step);
1261 }
1262