xref: /freebsd/sys/kern/sys_process.c (revision 3157ba21)
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 
68 struct ptrace_io_desc32 {
69 	int		piod_op;
70 	u_int32_t	piod_offs;
71 	u_int32_t	piod_addr;
72 	u_int32_t	piod_len;
73 };
74 
75 struct ptrace_vm_entry32 {
76 	int		pve_entry;
77 	int		pve_timestamp;
78 	uint32_t	pve_start;
79 	uint32_t	pve_end;
80 	uint32_t	pve_offset;
81 	u_int		pve_prot;
82 	u_int		pve_pathlen;
83 	int32_t		pve_fileid;
84 	u_int		pve_fsid;
85 	uint32_t	pve_path;
86 };
87 
88 #endif
89 
90 /*
91  * Functions implemented using PROC_ACTION():
92  *
93  * proc_read_regs(proc, regs)
94  *	Get the current user-visible register set from the process
95  *	and copy it into the regs structure (<machine/reg.h>).
96  *	The process is stopped at the time read_regs is called.
97  *
98  * proc_write_regs(proc, regs)
99  *	Update the current register set from the passed in regs
100  *	structure.  Take care to avoid clobbering special CPU
101  *	registers or privileged bits in the PSL.
102  *	Depending on the architecture this may have fix-up work to do,
103  *	especially if the IAR or PCW are modified.
104  *	The process is stopped at the time write_regs is called.
105  *
106  * proc_read_fpregs, proc_write_fpregs
107  *	deal with the floating point register set, otherwise as above.
108  *
109  * proc_read_dbregs, proc_write_dbregs
110  *	deal with the processor debug register set, otherwise as above.
111  *
112  * proc_sstep(proc)
113  *	Arrange for the process to trap after executing a single instruction.
114  */
115 
116 #define	PROC_ACTION(action) do {					\
117 	int error;							\
118 									\
119 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
120 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
121 		error = EIO;						\
122 	else								\
123 		error = (action);					\
124 	return (error);							\
125 } while(0)
126 
127 int
128 proc_read_regs(struct thread *td, struct reg *regs)
129 {
130 
131 	PROC_ACTION(fill_regs(td, regs));
132 }
133 
134 int
135 proc_write_regs(struct thread *td, struct reg *regs)
136 {
137 
138 	PROC_ACTION(set_regs(td, regs));
139 }
140 
141 int
142 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
143 {
144 
145 	PROC_ACTION(fill_dbregs(td, dbregs));
146 }
147 
148 int
149 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
150 {
151 
152 	PROC_ACTION(set_dbregs(td, dbregs));
153 }
154 
155 /*
156  * Ptrace doesn't support fpregs at all, and there are no security holes
157  * or translations for fpregs, so we can just copy them.
158  */
159 int
160 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
161 {
162 
163 	PROC_ACTION(fill_fpregs(td, fpregs));
164 }
165 
166 int
167 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
168 {
169 
170 	PROC_ACTION(set_fpregs(td, fpregs));
171 }
172 
173 #ifdef COMPAT_FREEBSD32
174 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
175 int
176 proc_read_regs32(struct thread *td, struct reg32 *regs32)
177 {
178 
179 	PROC_ACTION(fill_regs32(td, regs32));
180 }
181 
182 int
183 proc_write_regs32(struct thread *td, struct reg32 *regs32)
184 {
185 
186 	PROC_ACTION(set_regs32(td, regs32));
187 }
188 
189 int
190 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
191 {
192 
193 	PROC_ACTION(fill_dbregs32(td, dbregs32));
194 }
195 
196 int
197 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
198 {
199 
200 	PROC_ACTION(set_dbregs32(td, dbregs32));
201 }
202 
203 int
204 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
205 {
206 
207 	PROC_ACTION(fill_fpregs32(td, fpregs32));
208 }
209 
210 int
211 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
212 {
213 
214 	PROC_ACTION(set_fpregs32(td, fpregs32));
215 }
216 #endif
217 
218 int
219 proc_sstep(struct thread *td)
220 {
221 
222 	PROC_ACTION(ptrace_single_step(td));
223 }
224 
225 int
226 proc_rwmem(struct proc *p, struct uio *uio)
227 {
228 	vm_map_t map;
229 	vm_object_t backing_object, object;
230 	vm_offset_t pageno;		/* page number */
231 	vm_prot_t reqprot;
232 	int error, writing;
233 
234 	/*
235 	 * Assert that someone has locked this vmspace.  (Should be
236 	 * curthread but we can't assert that.)  This keeps the process
237 	 * from exiting out from under us until this operation completes.
238 	 */
239 	KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
240 	    p, p->p_pid));
241 
242 	/*
243 	 * The map we want...
244 	 */
245 	map = &p->p_vmspace->vm_map;
246 
247 	writing = uio->uio_rw == UIO_WRITE;
248 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
249 
250 	/*
251 	 * Only map in one page at a time.  We don't have to, but it
252 	 * makes things easier.  This way is trivial - right?
253 	 */
254 	do {
255 		vm_map_t tmap;
256 		vm_offset_t uva;
257 		int page_offset;		/* offset into page */
258 		vm_map_entry_t out_entry;
259 		vm_prot_t out_prot;
260 		boolean_t wired;
261 		vm_pindex_t pindex;
262 		u_int len;
263 		vm_page_t m;
264 
265 		object = NULL;
266 
267 		uva = (vm_offset_t)uio->uio_offset;
268 
269 		/*
270 		 * Get the page number of this segment.
271 		 */
272 		pageno = trunc_page(uva);
273 		page_offset = uva - pageno;
274 
275 		/*
276 		 * How many bytes to copy
277 		 */
278 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
279 
280 		/*
281 		 * Fault the page on behalf of the process
282 		 */
283 		error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
284 		if (error) {
285 			if (error == KERN_RESOURCE_SHORTAGE)
286 				error = ENOMEM;
287 			else
288 				error = EFAULT;
289 			break;
290 		}
291 
292 		/*
293 		 * Now we need to get the page.  out_entry and wired
294 		 * aren't used.  One would think the vm code
295 		 * would be a *bit* nicer...  We use tmap because
296 		 * vm_map_lookup() can change the map argument.
297 		 */
298 		tmap = map;
299 		error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
300 		    &object, &pindex, &out_prot, &wired);
301 		if (error) {
302 			error = EFAULT;
303 			break;
304 		}
305 		VM_OBJECT_LOCK(object);
306 		while ((m = vm_page_lookup(object, pindex)) == NULL &&
307 		    !writing &&
308 		    (backing_object = object->backing_object) != NULL) {
309 			/*
310 			 * Allow fallback to backing objects if we are reading.
311 			 */
312 			VM_OBJECT_LOCK(backing_object);
313 			pindex += OFF_TO_IDX(object->backing_object_offset);
314 			VM_OBJECT_UNLOCK(object);
315 			object = backing_object;
316 		}
317 		if (writing && m != NULL) {
318 			vm_page_dirty(m);
319 			vm_pager_page_unswapped(m);
320 		}
321 		VM_OBJECT_UNLOCK(object);
322 		if (m == NULL) {
323 			vm_map_lookup_done(tmap, out_entry);
324 			error = EFAULT;
325 			break;
326 		}
327 
328 		/*
329 		 * Hold the page in memory.
330 		 */
331 		vm_page_lock(m);
332 		vm_page_hold(m);
333 		vm_page_unlock(m);
334 
335 		/*
336 		 * We're done with tmap now.
337 		 */
338 		vm_map_lookup_done(tmap, out_entry);
339 
340 		/*
341 		 * Now do the i/o move.
342 		 */
343 		error = uiomove_fromphys(&m, page_offset, len, uio);
344 
345 		/* Make the I-cache coherent for breakpoints. */
346 		if (!error && writing && (out_prot & VM_PROT_EXECUTE))
347 			vm_sync_icache(map, uva, len);
348 
349 		/*
350 		 * Release the page.
351 		 */
352 		vm_page_lock(m);
353 		vm_page_unhold(m);
354 		vm_page_unlock(m);
355 
356 	} while (error == 0 && uio->uio_resid > 0);
357 
358 	return (error);
359 }
360 
361 static int
362 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
363 {
364 	struct vattr vattr;
365 	vm_map_t map;
366 	vm_map_entry_t entry;
367 	vm_object_t obj, tobj, lobj;
368 	struct vmspace *vm;
369 	struct vnode *vp;
370 	char *freepath, *fullpath;
371 	u_int pathlen;
372 	int error, index, vfslocked;
373 
374 	error = 0;
375 	obj = NULL;
376 
377 	vm = vmspace_acquire_ref(p);
378 	map = &vm->vm_map;
379 	vm_map_lock_read(map);
380 
381 	do {
382 		entry = map->header.next;
383 		index = 0;
384 		while (index < pve->pve_entry && entry != &map->header) {
385 			entry = entry->next;
386 			index++;
387 		}
388 		if (index != pve->pve_entry) {
389 			error = EINVAL;
390 			break;
391 		}
392 		while (entry != &map->header &&
393 		    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
394 			entry = entry->next;
395 			index++;
396 		}
397 		if (entry == &map->header) {
398 			error = ENOENT;
399 			break;
400 		}
401 
402 		/* We got an entry. */
403 		pve->pve_entry = index + 1;
404 		pve->pve_timestamp = map->timestamp;
405 		pve->pve_start = entry->start;
406 		pve->pve_end = entry->end - 1;
407 		pve->pve_offset = entry->offset;
408 		pve->pve_prot = entry->protection;
409 
410 		/* Backing object's path needed? */
411 		if (pve->pve_pathlen == 0)
412 			break;
413 
414 		pathlen = pve->pve_pathlen;
415 		pve->pve_pathlen = 0;
416 
417 		obj = entry->object.vm_object;
418 		if (obj != NULL)
419 			VM_OBJECT_LOCK(obj);
420 	} while (0);
421 
422 	vm_map_unlock_read(map);
423 	vmspace_free(vm);
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_LOCK(tobj);
433 			if (lobj != obj)
434 				VM_OBJECT_UNLOCK(lobj);
435 			lobj = tobj;
436 			pve->pve_offset += tobj->backing_object_offset;
437 		}
438 		vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
439 		if (vp != NULL)
440 			vref(vp);
441 		if (lobj != obj)
442 			VM_OBJECT_UNLOCK(lobj);
443 		VM_OBJECT_UNLOCK(obj);
444 
445 		if (vp != NULL) {
446 			freepath = NULL;
447 			fullpath = NULL;
448 			vn_fullpath(td, vp, &fullpath, &freepath);
449 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
450 			vn_lock(vp, LK_SHARED | LK_RETRY);
451 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
452 				pve->pve_fileid = vattr.va_fileid;
453 				pve->pve_fsid = vattr.va_fsid;
454 			}
455 			vput(vp);
456 			VFS_UNLOCK_GIANT(vfslocked);
457 
458 			if (fullpath != NULL) {
459 				pve->pve_pathlen = strlen(fullpath) + 1;
460 				if (pve->pve_pathlen <= pathlen) {
461 					error = copyout(fullpath, pve->pve_path,
462 					    pve->pve_pathlen);
463 				} else
464 					error = ENAMETOOLONG;
465 			}
466 			if (freepath != NULL)
467 				free(freepath, M_TEMP);
468 		}
469 	}
470 
471 	return (error);
472 }
473 
474 #ifdef COMPAT_FREEBSD32
475 static int
476 ptrace_vm_entry32(struct thread *td, struct proc *p,
477     struct ptrace_vm_entry32 *pve32)
478 {
479 	struct ptrace_vm_entry pve;
480 	int error;
481 
482 	pve.pve_entry = pve32->pve_entry;
483 	pve.pve_pathlen = pve32->pve_pathlen;
484 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
485 
486 	error = ptrace_vm_entry(td, p, &pve);
487 	if (error == 0) {
488 		pve32->pve_entry = pve.pve_entry;
489 		pve32->pve_timestamp = pve.pve_timestamp;
490 		pve32->pve_start = pve.pve_start;
491 		pve32->pve_end = pve.pve_end;
492 		pve32->pve_offset = pve.pve_offset;
493 		pve32->pve_prot = pve.pve_prot;
494 		pve32->pve_fileid = pve.pve_fileid;
495 		pve32->pve_fsid = pve.pve_fsid;
496 	}
497 
498 	pve32->pve_pathlen = pve.pve_pathlen;
499 	return (error);
500 }
501 #endif /* COMPAT_FREEBSD32 */
502 
503 /*
504  * Process debugging system call.
505  */
506 #ifndef _SYS_SYSPROTO_H_
507 struct ptrace_args {
508 	int	req;
509 	pid_t	pid;
510 	caddr_t	addr;
511 	int	data;
512 };
513 #endif
514 
515 #ifdef COMPAT_FREEBSD32
516 /*
517  * This CPP subterfuge is to try and reduce the number of ifdefs in
518  * the body of the code.
519  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
520  * becomes either:
521  *   copyin(uap->addr, &r.reg, sizeof r.reg);
522  * or
523  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
524  * .. except this is done at runtime.
525  */
526 #define	COPYIN(u, k, s)		wrap32 ? \
527 	copyin(u, k ## 32, s ## 32) : \
528 	copyin(u, k, s)
529 #define	COPYOUT(k, u, s)	wrap32 ? \
530 	copyout(k ## 32, u, s ## 32) : \
531 	copyout(k, u, s)
532 #else
533 #define	COPYIN(u, k, s)		copyin(u, k, s)
534 #define	COPYOUT(k, u, s)	copyout(k, u, s)
535 #endif
536 int
537 ptrace(struct thread *td, struct ptrace_args *uap)
538 {
539 	/*
540 	 * XXX this obfuscation is to reduce stack usage, but the register
541 	 * structs may be too large to put on the stack anyway.
542 	 */
543 	union {
544 		struct ptrace_io_desc piod;
545 		struct ptrace_lwpinfo pl;
546 		struct ptrace_vm_entry pve;
547 		struct dbreg dbreg;
548 		struct fpreg fpreg;
549 		struct reg reg;
550 #ifdef COMPAT_FREEBSD32
551 		struct dbreg32 dbreg32;
552 		struct fpreg32 fpreg32;
553 		struct reg32 reg32;
554 		struct ptrace_io_desc32 piod32;
555 		struct ptrace_vm_entry32 pve32;
556 #endif
557 	} r;
558 	void *addr;
559 	int error = 0;
560 #ifdef COMPAT_FREEBSD32
561 	int wrap32 = 0;
562 
563 	if (SV_CURPROC_FLAG(SV_ILP32))
564 		wrap32 = 1;
565 #endif
566 	AUDIT_ARG_PID(uap->pid);
567 	AUDIT_ARG_CMD(uap->req);
568 	AUDIT_ARG_VALUE(uap->data);
569 	addr = &r;
570 	switch (uap->req) {
571 	case PT_GETREGS:
572 	case PT_GETFPREGS:
573 	case PT_GETDBREGS:
574 	case PT_LWPINFO:
575 		break;
576 	case PT_SETREGS:
577 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
578 		break;
579 	case PT_SETFPREGS:
580 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
581 		break;
582 	case PT_SETDBREGS:
583 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
584 		break;
585 	case PT_IO:
586 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
587 		break;
588 	case PT_VM_ENTRY:
589 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
590 		break;
591 	default:
592 		addr = uap->addr;
593 		break;
594 	}
595 	if (error)
596 		return (error);
597 
598 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
599 	if (error)
600 		return (error);
601 
602 	switch (uap->req) {
603 	case PT_VM_ENTRY:
604 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
605 		break;
606 	case PT_IO:
607 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
608 		break;
609 	case PT_GETREGS:
610 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
611 		break;
612 	case PT_GETFPREGS:
613 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
614 		break;
615 	case PT_GETDBREGS:
616 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
617 		break;
618 	case PT_LWPINFO:
619 		error = copyout(&r.pl, uap->addr, uap->data);
620 		break;
621 	}
622 
623 	return (error);
624 }
625 #undef COPYIN
626 #undef COPYOUT
627 
628 #ifdef COMPAT_FREEBSD32
629 /*
630  *   PROC_READ(regs, td2, addr);
631  * becomes either:
632  *   proc_read_regs(td2, addr);
633  * or
634  *   proc_read_regs32(td2, addr);
635  * .. except this is done at runtime.  There is an additional
636  * complication in that PROC_WRITE disallows 32 bit consumers
637  * from writing to 64 bit address space targets.
638  */
639 #define	PROC_READ(w, t, a)	wrap32 ? \
640 	proc_read_ ## w ## 32(t, a) : \
641 	proc_read_ ## w (t, a)
642 #define	PROC_WRITE(w, t, a)	wrap32 ? \
643 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
644 	proc_write_ ## w (t, a)
645 #else
646 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
647 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
648 #endif
649 
650 int
651 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
652 {
653 	struct iovec iov;
654 	struct uio uio;
655 	struct proc *curp, *p, *pp;
656 	struct thread *td2 = NULL;
657 	struct ptrace_io_desc *piod = NULL;
658 	struct ptrace_lwpinfo *pl;
659 	int error, write, tmp, num;
660 	int proctree_locked = 0;
661 	lwpid_t tid = 0, *buf;
662 #ifdef COMPAT_FREEBSD32
663 	int wrap32 = 0, safe = 0;
664 	struct ptrace_io_desc32 *piod32 = NULL;
665 #endif
666 
667 	curp = td->td_proc;
668 
669 	/* Lock proctree before locking the process. */
670 	switch (req) {
671 	case PT_TRACE_ME:
672 	case PT_ATTACH:
673 	case PT_STEP:
674 	case PT_CONTINUE:
675 	case PT_TO_SCE:
676 	case PT_TO_SCX:
677 	case PT_SYSCALL:
678 	case PT_DETACH:
679 		sx_xlock(&proctree_lock);
680 		proctree_locked = 1;
681 		break;
682 	default:
683 		break;
684 	}
685 
686 	write = 0;
687 	if (req == PT_TRACE_ME) {
688 		p = td->td_proc;
689 		PROC_LOCK(p);
690 	} else {
691 		if (pid <= PID_MAX) {
692 			if ((p = pfind(pid)) == NULL) {
693 				if (proctree_locked)
694 					sx_xunlock(&proctree_lock);
695 				return (ESRCH);
696 			}
697 		} else {
698 			/* this is slow, should be optimized */
699 			sx_slock(&allproc_lock);
700 			FOREACH_PROC_IN_SYSTEM(p) {
701 				PROC_LOCK(p);
702 				FOREACH_THREAD_IN_PROC(p, td2) {
703 					if (td2->td_tid == pid)
704 						break;
705 				}
706 				if (td2 != NULL)
707 					break; /* proc lock held */
708 				PROC_UNLOCK(p);
709 			}
710 			sx_sunlock(&allproc_lock);
711 			if (p == NULL) {
712 				if (proctree_locked)
713 					sx_xunlock(&proctree_lock);
714 				return (ESRCH);
715 			}
716 			tid = pid;
717 			pid = p->p_pid;
718 		}
719 	}
720 	AUDIT_ARG_PROCESS(p);
721 
722 	if ((p->p_flag & P_WEXIT) != 0) {
723 		error = ESRCH;
724 		goto fail;
725 	}
726 	if ((error = p_cansee(td, p)) != 0)
727 		goto fail;
728 
729 	if ((error = p_candebug(td, p)) != 0)
730 		goto fail;
731 
732 	/*
733 	 * System processes can't be debugged.
734 	 */
735 	if ((p->p_flag & P_SYSTEM) != 0) {
736 		error = EINVAL;
737 		goto fail;
738 	}
739 
740 	if (tid == 0) {
741 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
742 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
743 			td2 = p->p_xthread;
744 		} else {
745 			td2 = FIRST_THREAD_IN_PROC(p);
746 		}
747 		tid = td2->td_tid;
748 	}
749 
750 #ifdef COMPAT_FREEBSD32
751 	/*
752 	 * Test if we're a 32 bit client and what the target is.
753 	 * Set the wrap controls accordingly.
754 	 */
755 	if (SV_CURPROC_FLAG(SV_ILP32)) {
756 		if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
757 			safe = 1;
758 		wrap32 = 1;
759 	}
760 #endif
761 	/*
762 	 * Permissions check
763 	 */
764 	switch (req) {
765 	case PT_TRACE_ME:
766 		/* Always legal. */
767 		break;
768 
769 	case PT_ATTACH:
770 		/* Self */
771 		if (p->p_pid == td->td_proc->p_pid) {
772 			error = EINVAL;
773 			goto fail;
774 		}
775 
776 		/* Already traced */
777 		if (p->p_flag & P_TRACED) {
778 			error = EBUSY;
779 			goto fail;
780 		}
781 
782 		/* Can't trace an ancestor if you're being traced. */
783 		if (curp->p_flag & P_TRACED) {
784 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
785 				if (pp == p) {
786 					error = EINVAL;
787 					goto fail;
788 				}
789 			}
790 		}
791 
792 
793 		/* OK */
794 		break;
795 
796 	case PT_CLEARSTEP:
797 		/* Allow thread to clear single step for itself */
798 		if (td->td_tid == tid)
799 			break;
800 
801 		/* FALLTHROUGH */
802 	default:
803 		/* not being traced... */
804 		if ((p->p_flag & P_TRACED) == 0) {
805 			error = EPERM;
806 			goto fail;
807 		}
808 
809 		/* not being traced by YOU */
810 		if (p->p_pptr != td->td_proc) {
811 			error = EBUSY;
812 			goto fail;
813 		}
814 
815 		/* not currently stopped */
816 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
817 		    p->p_suspcount != p->p_numthreads  ||
818 		    (p->p_flag & P_WAITED) == 0) {
819 			error = EBUSY;
820 			goto fail;
821 		}
822 
823 		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
824 			static int count = 0;
825 			if (count++ == 0)
826 				printf("P_STOPPED_TRACE not set.\n");
827 		}
828 
829 		/* OK */
830 		break;
831 	}
832 
833 	/* Keep this process around until we finish this request. */
834 	_PHOLD(p);
835 
836 #ifdef FIX_SSTEP
837 	/*
838 	 * Single step fixup ala procfs
839 	 */
840 	FIX_SSTEP(td2);
841 #endif
842 
843 	/*
844 	 * Actually do the requests
845 	 */
846 
847 	td->td_retval[0] = 0;
848 
849 	switch (req) {
850 	case PT_TRACE_ME:
851 		/* set my trace flag and "owner" so it can read/write me */
852 		p->p_flag |= P_TRACED;
853 		p->p_oppid = p->p_pptr->p_pid;
854 		break;
855 
856 	case PT_ATTACH:
857 		/* security check done above */
858 		p->p_flag |= P_TRACED;
859 		p->p_oppid = p->p_pptr->p_pid;
860 		if (p->p_pptr != td->td_proc)
861 			proc_reparent(p, td->td_proc);
862 		data = SIGSTOP;
863 		goto sendsig;	/* in PT_CONTINUE below */
864 
865 	case PT_CLEARSTEP:
866 		error = ptrace_clear_single_step(td2);
867 		break;
868 
869 	case PT_SETSTEP:
870 		error = ptrace_single_step(td2);
871 		break;
872 
873 	case PT_SUSPEND:
874 		td2->td_dbgflags |= TDB_SUSPEND;
875 		thread_lock(td2);
876 		td2->td_flags |= TDF_NEEDSUSPCHK;
877 		thread_unlock(td2);
878 		break;
879 
880 	case PT_RESUME:
881 		td2->td_dbgflags &= ~TDB_SUSPEND;
882 		break;
883 
884 	case PT_STEP:
885 	case PT_CONTINUE:
886 	case PT_TO_SCE:
887 	case PT_TO_SCX:
888 	case PT_SYSCALL:
889 	case PT_DETACH:
890 		/* Zero means do not send any signal */
891 		if (data < 0 || data > _SIG_MAXSIG) {
892 			error = EINVAL;
893 			break;
894 		}
895 
896 		switch (req) {
897 		case PT_STEP:
898 			error = ptrace_single_step(td2);
899 			if (error)
900 				goto out;
901 			break;
902 		case PT_CONTINUE:
903 		case PT_TO_SCE:
904 		case PT_TO_SCX:
905 		case PT_SYSCALL:
906 			if (addr != (void *)1) {
907 				error = ptrace_set_pc(td2,
908 				    (u_long)(uintfptr_t)addr);
909 				if (error)
910 					goto out;
911 			}
912 			switch (req) {
913 			case PT_TO_SCE:
914 				p->p_stops |= S_PT_SCE;
915 				break;
916 			case PT_TO_SCX:
917 				p->p_stops |= S_PT_SCX;
918 				break;
919 			case PT_SYSCALL:
920 				p->p_stops |= S_PT_SCE | S_PT_SCX;
921 				break;
922 			}
923 			break;
924 		case PT_DETACH:
925 			/* reset process parent */
926 			if (p->p_oppid != p->p_pptr->p_pid) {
927 				struct proc *pp;
928 
929 				PROC_LOCK(p->p_pptr);
930 				sigqueue_take(p->p_ksi);
931 				PROC_UNLOCK(p->p_pptr);
932 
933 				PROC_UNLOCK(p);
934 				pp = pfind(p->p_oppid);
935 				if (pp == NULL)
936 					pp = initproc;
937 				else
938 					PROC_UNLOCK(pp);
939 				PROC_LOCK(p);
940 				proc_reparent(p, pp);
941 				if (pp == initproc)
942 					p->p_sigparent = SIGCHLD;
943 			}
944 			p->p_flag &= ~(P_TRACED | P_WAITED);
945 			p->p_oppid = 0;
946 
947 			/* should we send SIGCHLD? */
948 			/* childproc_continued(p); */
949 			break;
950 		}
951 
952 	sendsig:
953 		if (proctree_locked) {
954 			sx_xunlock(&proctree_lock);
955 			proctree_locked = 0;
956 		}
957 		p->p_xstat = data;
958 		p->p_xthread = NULL;
959 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
960 			/* deliver or queue signal */
961 			td2->td_dbgflags &= ~TDB_XSIG;
962 			td2->td_xsig = data;
963 
964 			if (req == PT_DETACH) {
965 				struct thread *td3;
966 				FOREACH_THREAD_IN_PROC(p, td3) {
967 					td3->td_dbgflags &= ~TDB_SUSPEND;
968 				}
969 			}
970 			/*
971 			 * unsuspend all threads, to not let a thread run,
972 			 * you should use PT_SUSPEND to suspend it before
973 			 * continuing process.
974 			 */
975 			PROC_SLOCK(p);
976 			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
977 			thread_unsuspend(p);
978 			PROC_SUNLOCK(p);
979 		} else {
980 			if (data)
981 				psignal(p, data);
982 		}
983 		break;
984 
985 	case PT_WRITE_I:
986 	case PT_WRITE_D:
987 		td2->td_dbgflags |= TDB_USERWR;
988 		write = 1;
989 		/* FALLTHROUGH */
990 	case PT_READ_I:
991 	case PT_READ_D:
992 		PROC_UNLOCK(p);
993 		tmp = 0;
994 		/* write = 0 set above */
995 		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
996 		iov.iov_len = sizeof(int);
997 		uio.uio_iov = &iov;
998 		uio.uio_iovcnt = 1;
999 		uio.uio_offset = (off_t)(uintptr_t)addr;
1000 		uio.uio_resid = sizeof(int);
1001 		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
1002 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1003 		uio.uio_td = td;
1004 		error = proc_rwmem(p, &uio);
1005 		if (uio.uio_resid != 0) {
1006 			/*
1007 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
1008 			 * so I think write() can bogusly return 0.
1009 			 * XXX what happens for short writes?  We don't want
1010 			 * to write partial data.
1011 			 * XXX proc_rwmem() returns EPERM for other invalid
1012 			 * addresses.  Convert this to EINVAL.  Does this
1013 			 * clobber returns of EPERM for other reasons?
1014 			 */
1015 			if (error == 0 || error == ENOSPC || error == EPERM)
1016 				error = EINVAL;	/* EOF */
1017 		}
1018 		if (!write)
1019 			td->td_retval[0] = tmp;
1020 		PROC_LOCK(p);
1021 		break;
1022 
1023 	case PT_IO:
1024 #ifdef COMPAT_FREEBSD32
1025 		if (wrap32) {
1026 			piod32 = addr;
1027 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1028 			iov.iov_len = piod32->piod_len;
1029 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1030 			uio.uio_resid = piod32->piod_len;
1031 		} else
1032 #endif
1033 		{
1034 			piod = addr;
1035 			iov.iov_base = piod->piod_addr;
1036 			iov.iov_len = piod->piod_len;
1037 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1038 			uio.uio_resid = piod->piod_len;
1039 		}
1040 		uio.uio_iov = &iov;
1041 		uio.uio_iovcnt = 1;
1042 		uio.uio_segflg = UIO_USERSPACE;
1043 		uio.uio_td = td;
1044 #ifdef COMPAT_FREEBSD32
1045 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1046 #else
1047 		tmp = piod->piod_op;
1048 #endif
1049 		switch (tmp) {
1050 		case PIOD_READ_D:
1051 		case PIOD_READ_I:
1052 			uio.uio_rw = UIO_READ;
1053 			break;
1054 		case PIOD_WRITE_D:
1055 		case PIOD_WRITE_I:
1056 			td2->td_dbgflags |= TDB_USERWR;
1057 			uio.uio_rw = UIO_WRITE;
1058 			break;
1059 		default:
1060 			error = EINVAL;
1061 			goto out;
1062 		}
1063 		PROC_UNLOCK(p);
1064 		error = proc_rwmem(p, &uio);
1065 #ifdef COMPAT_FREEBSD32
1066 		if (wrap32)
1067 			piod32->piod_len -= uio.uio_resid;
1068 		else
1069 #endif
1070 			piod->piod_len -= uio.uio_resid;
1071 		PROC_LOCK(p);
1072 		break;
1073 
1074 	case PT_KILL:
1075 		data = SIGKILL;
1076 		goto sendsig;	/* in PT_CONTINUE above */
1077 
1078 	case PT_SETREGS:
1079 		td2->td_dbgflags |= TDB_USERWR;
1080 		error = PROC_WRITE(regs, td2, addr);
1081 		break;
1082 
1083 	case PT_GETREGS:
1084 		error = PROC_READ(regs, td2, addr);
1085 		break;
1086 
1087 	case PT_SETFPREGS:
1088 		td2->td_dbgflags |= TDB_USERWR;
1089 		error = PROC_WRITE(fpregs, td2, addr);
1090 		break;
1091 
1092 	case PT_GETFPREGS:
1093 		error = PROC_READ(fpregs, td2, addr);
1094 		break;
1095 
1096 	case PT_SETDBREGS:
1097 		td2->td_dbgflags |= TDB_USERWR;
1098 		error = PROC_WRITE(dbregs, td2, addr);
1099 		break;
1100 
1101 	case PT_GETDBREGS:
1102 		error = PROC_READ(dbregs, td2, addr);
1103 		break;
1104 
1105 	case PT_LWPINFO:
1106 		if (data <= 0 || data > sizeof(*pl)) {
1107 			error = EINVAL;
1108 			break;
1109 		}
1110 		pl = addr;
1111 		pl->pl_lwpid = td2->td_tid;
1112 		if (td2->td_dbgflags & TDB_XSIG)
1113 			pl->pl_event = PL_EVENT_SIGNAL;
1114 		pl->pl_flags = 0;
1115 		if (td2->td_dbgflags & TDB_SCE)
1116 			pl->pl_flags |= PL_FLAG_SCE;
1117 		else if (td2->td_dbgflags & TDB_SCX)
1118 			pl->pl_flags |= PL_FLAG_SCX;
1119 		if (td2->td_dbgflags & TDB_EXEC)
1120 			pl->pl_flags |= PL_FLAG_EXEC;
1121 		pl->pl_sigmask = td2->td_sigmask;
1122 		pl->pl_siglist = td2->td_siglist;
1123 		break;
1124 
1125 	case PT_GETNUMLWPS:
1126 		td->td_retval[0] = p->p_numthreads;
1127 		break;
1128 
1129 	case PT_GETLWPLIST:
1130 		if (data <= 0) {
1131 			error = EINVAL;
1132 			break;
1133 		}
1134 		num = imin(p->p_numthreads, data);
1135 		PROC_UNLOCK(p);
1136 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1137 		tmp = 0;
1138 		PROC_LOCK(p);
1139 		FOREACH_THREAD_IN_PROC(p, td2) {
1140 			if (tmp >= num)
1141 				break;
1142 			buf[tmp++] = td2->td_tid;
1143 		}
1144 		PROC_UNLOCK(p);
1145 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1146 		free(buf, M_TEMP);
1147 		if (!error)
1148 			td->td_retval[0] = tmp;
1149 		PROC_LOCK(p);
1150 		break;
1151 
1152 	case PT_VM_TIMESTAMP:
1153 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1154 		break;
1155 
1156 	case PT_VM_ENTRY:
1157 		PROC_UNLOCK(p);
1158 #ifdef COMPAT_FREEBSD32
1159 		if (wrap32)
1160 			error = ptrace_vm_entry32(td, p, addr);
1161 		else
1162 #endif
1163 		error = ptrace_vm_entry(td, p, addr);
1164 		PROC_LOCK(p);
1165 		break;
1166 
1167 	default:
1168 #ifdef __HAVE_PTRACE_MACHDEP
1169 		if (req >= PT_FIRSTMACH) {
1170 			PROC_UNLOCK(p);
1171 			error = cpu_ptrace(td2, req, addr, data);
1172 			PROC_LOCK(p);
1173 		} else
1174 #endif
1175 			/* Unknown request. */
1176 			error = EINVAL;
1177 		break;
1178 	}
1179 
1180 out:
1181 	/* Drop our hold on this process now that the request has completed. */
1182 	_PRELE(p);
1183 fail:
1184 	PROC_UNLOCK(p);
1185 	if (proctree_locked)
1186 		sx_xunlock(&proctree_lock);
1187 	return (error);
1188 }
1189 #undef PROC_READ
1190 #undef PROC_WRITE
1191 
1192 /*
1193  * Stop a process because of a debugging event;
1194  * stay stopped until p->p_step is cleared
1195  * (cleared by PIOCCONT in procfs).
1196  */
1197 void
1198 stopevent(struct proc *p, unsigned int event, unsigned int val)
1199 {
1200 
1201 	PROC_LOCK_ASSERT(p, MA_OWNED);
1202 	p->p_step = 1;
1203 	do {
1204 		p->p_xstat = val;
1205 		p->p_xthread = NULL;
1206 		p->p_stype = event;	/* Which event caused the stop? */
1207 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1208 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1209 	} while (p->p_step);
1210 }
1211