xref: /freebsd/sys/kern/sys_process.c (revision b2e4da0b53ad082768b8f6f83766e030fd00d02a)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 1994, Sean Eric Fagan
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by Sean Eric Fagan.
18  * 4. The name of the author may not be used to endorse or promote products
19  *    derived from this software without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/ktr.h>
37 #include <sys/limits.h>
38 #include <sys/lock.h>
39 #include <sys/mman.h>
40 #include <sys/mutex.h>
41 #include <sys/reg.h>
42 #include <sys/sleepqueue.h>
43 #include <sys/syscallsubr.h>
44 #include <sys/sysent.h>
45 #include <sys/sysproto.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/vnode.h>
49 #include <sys/ptrace.h>
50 #include <sys/rwlock.h>
51 #include <sys/sx.h>
52 #include <sys/malloc.h>
53 #include <sys/signalvar.h>
54 #include <sys/caprights.h>
55 #include <sys/filedesc.h>
56 
57 #include <security/audit/audit.h>
58 
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_extern.h>
62 #include <vm/vm_map.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_param.h>
67 
68 #ifdef COMPAT_FREEBSD32
69 #include <sys/procfs.h>
70 #endif
71 
72 /* Assert it's safe to unlock a process, e.g. to allocate working memory */
73 #define	PROC_ASSERT_TRACEREQ(p)	MPASS(((p)->p_flag2 & P2_PTRACEREQ) != 0)
74 
75 #define PROC_PRIV_CHECK(priv) do {	 			\
76 		int _error;					\
77 		_error = priv_check(curthread, priv);	\
78 		if (_error)					\
79 			return (_error);			\
80 	} while (0)
81 
82 /*
83  * Functions implemented below:
84  *
85  * proc_read_regs(proc, regs)
86  *	Get the current user-visible register set from the process
87  *	and copy it into the regs structure (<machine/reg.h>).
88  *	The process is stopped at the time read_regs is called.
89  *
90  * proc_write_regs(proc, regs)
91  *	Update the current register set from the passed in regs
92  *	structure.  Take care to avoid clobbering special CPU
93  *	registers or privileged bits in the PSL.
94  *	Depending on the architecture this may have fix-up work to do,
95  *	especially if the IAR or PCW are modified.
96  *	The process is stopped at the time write_regs is called.
97  *
98  * proc_read_fpregs, proc_write_fpregs
99  *	deal with the floating point register set, otherwise as above.
100  *
101  * proc_read_dbregs, proc_write_dbregs
102  *	deal with the processor debug register set, otherwise as above.
103  *
104  * proc_sstep(proc)
105  *	Arrange for the process to trap after executing a single instruction.
106  */
107 
108 int
109 proc_read_regs(struct thread *td, struct reg *regs)
110 {
111 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
112 	return (fill_regs(td, regs));
113 }
114 
115 int
116 proc_write_regs(struct thread *td, struct reg *regs)
117 {
118 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
119 	PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
120 	return (set_regs(td, regs));
121 }
122 
123 int
124 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
125 {
126 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
127 	return (fill_dbregs(td, dbregs));
128 }
129 
130 int
131 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
132 {
133 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
134 	PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
135 	return (set_dbregs(td, dbregs));
136 }
137 
138 /*
139  * Ptrace doesn't support fpregs at all, and there are no security holes
140  * or translations for fpregs, so we can just copy them.
141  */
142 int
143 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
144 {
145 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
146 	return (fill_fpregs(td, fpregs));
147 }
148 
149 int
150 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
151 {
152 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
153 	PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
154 	return (set_fpregs(td, fpregs));
155 }
156 
157 static struct regset *
158 proc_find_regset(struct thread *td, int note)
159 {
160 	struct regset **regsetp, **regset_end, *regset;
161 	struct sysentvec *sv;
162 
163 	sv = td->td_proc->p_sysent;
164 	regsetp = sv->sv_regset_begin;
165 	if (regsetp == NULL)
166 		return (NULL);
167 	regset_end = sv->sv_regset_end;
168 	MPASS(regset_end != NULL);
169 	for (; regsetp < regset_end; regsetp++) {
170 		regset = *regsetp;
171 		if (regset->note != note)
172 			continue;
173 
174 		return (regset);
175 	}
176 
177 	return (NULL);
178 }
179 
180 static int
181 proc_read_regset(struct thread *td, int note, struct iovec *iov)
182 {
183 	struct regset *regset;
184 	struct proc *p;
185 	void *buf;
186 	size_t size;
187 	int error;
188 
189 	regset = proc_find_regset(td, note);
190 	if (regset == NULL)
191 		return (EINVAL);
192 
193 	if (regset->get == NULL)
194 		return (EINVAL);
195 
196 	size = regset->size;
197 	/*
198 	 * The regset is dynamically sized, e.g. the size could change
199 	 * depending on the hardware, or may have a per-thread size.
200 	 */
201 	if (size == 0) {
202 		if (!regset->get(regset, td, NULL, &size))
203 			return (EINVAL);
204 	}
205 
206 	if (iov->iov_base == NULL) {
207 		iov->iov_len = size;
208 		if (iov->iov_len == 0)
209 			return (EINVAL);
210 
211 		return (0);
212 	}
213 
214 	/* The length is wrong, return an error */
215 	if (iov->iov_len != size)
216 		return (EINVAL);
217 
218 	error = 0;
219 	p = td->td_proc;
220 
221 	/* Drop the proc lock while allocating the temp buffer */
222 	PROC_ASSERT_TRACEREQ(p);
223 	PROC_UNLOCK(p);
224 	buf = malloc(size, M_TEMP, M_WAITOK);
225 	PROC_LOCK(p);
226 
227 	if (!regset->get(regset, td, buf, &size)) {
228 		error = EINVAL;
229 	} else {
230 		KASSERT(size == regset->size || regset->size == 0,
231 		    ("%s: Getter function changed the size", __func__));
232 
233 		iov->iov_len = size;
234 		PROC_UNLOCK(p);
235 		error = copyout(buf, iov->iov_base, size);
236 		PROC_LOCK(p);
237 	}
238 
239 	free(buf, M_TEMP);
240 
241 	return (error);
242 }
243 
244 static int
245 proc_write_regset(struct thread *td, int note, struct iovec *iov)
246 {
247 	struct regset *regset;
248 	struct proc *p;
249 	void *buf;
250 	size_t size;
251 	int error;
252 
253 	regset = proc_find_regset(td, note);
254 	if (regset == NULL)
255 		return (EINVAL);
256 
257 	size = regset->size;
258 	/*
259 	 * The regset is dynamically sized, e.g. the size could change
260 	 * depending on the hardware, or may have a per-thread size.
261 	 */
262 	if (size == 0) {
263 		if (!regset->get(regset, td, NULL, &size))
264 			return (EINVAL);
265 	}
266 
267 	/* The length is wrong, return an error */
268 	if (iov->iov_len != size)
269 		return (EINVAL);
270 
271 	if (regset->set == NULL)
272 		return (EINVAL);
273 
274 	PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
275 
276 	p = td->td_proc;
277 
278 	/* Drop the proc lock while allocating the temp buffer */
279 	PROC_ASSERT_TRACEREQ(p);
280 	PROC_UNLOCK(p);
281 	buf = malloc(size, M_TEMP, M_WAITOK);
282 	error = copyin(iov->iov_base, buf, size);
283 	PROC_LOCK(p);
284 
285 	if (error == 0) {
286 		if (!regset->set(regset, td, buf, size)) {
287 			error = EINVAL;
288 		}
289 	}
290 
291 	free(buf, M_TEMP);
292 
293 	return (error);
294 }
295 
296 #ifdef COMPAT_FREEBSD32
297 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
298 int
299 proc_read_regs32(struct thread *td, struct reg32 *regs32)
300 {
301 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
302 	return (fill_regs32(td, regs32));
303 }
304 
305 int
306 proc_write_regs32(struct thread *td, struct reg32 *regs32)
307 {
308 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
309 	PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
310 	return (set_regs32(td, regs32));
311 }
312 
313 int
314 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
315 {
316 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
317 	return (fill_dbregs32(td, dbregs32));
318 }
319 
320 int
321 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
322 {
323 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
324 	PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
325 	return (set_dbregs32(td, dbregs32));
326 }
327 
328 int
329 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
330 {
331 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
332 	return (fill_fpregs32(td, fpregs32));
333 }
334 
335 int
336 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
337 {
338 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
339 	PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
340 	return (set_fpregs32(td, fpregs32));
341 }
342 #endif
343 
344 int
345 proc_sstep(struct thread *td)
346 {
347 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
348 	return (ptrace_single_step(td));
349 }
350 
351 int
352 proc_rwmem(struct proc *p, struct uio *uio)
353 {
354 	vm_map_t map;
355 	vm_offset_t pageno;		/* page number */
356 	vm_prot_t reqprot;
357 	int error, fault_flags, page_offset, writing;
358 
359 	/*
360 	 * Make sure that the process' vmspace remains live.
361 	 */
362 	if (p != curproc)
363 		PROC_ASSERT_HELD(p);
364 	PROC_LOCK_ASSERT(p, MA_NOTOWNED);
365 
366 	/*
367 	 * The map we want...
368 	 */
369 	map = &p->p_vmspace->vm_map;
370 
371 	/*
372 	 * If we are writing, then we request vm_fault() to create a private
373 	 * copy of each page.  Since these copies will not be writeable by the
374 	 * process, we must explicitly request that they be dirtied.
375 	 */
376 	writing = uio->uio_rw == UIO_WRITE;
377 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
378 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
379 
380 	if (writing) {
381 		PROC_PRIV_CHECK(PRIV_PROC_MEM_WRITE);
382 	}
383 
384 	/*
385 	 * Only map in one page at a time.  We don't have to, but it
386 	 * makes things easier.  This way is trivial - right?
387 	 */
388 	do {
389 		vm_offset_t uva;
390 		u_int len;
391 		vm_page_t m;
392 
393 		uva = (vm_offset_t)uio->uio_offset;
394 
395 		/*
396 		 * Get the page number of this segment.
397 		 */
398 		pageno = trunc_page(uva);
399 		page_offset = uva - pageno;
400 
401 		/*
402 		 * How many bytes to copy
403 		 */
404 		len = MIN(PAGE_SIZE - page_offset, uio->uio_resid);
405 
406 		/*
407 		 * Fault and hold the page on behalf of the process.
408 		 */
409 		error = vm_fault(map, pageno, reqprot, fault_flags, &m);
410 		if (error != KERN_SUCCESS) {
411 			if (error == KERN_RESOURCE_SHORTAGE)
412 				error = ENOMEM;
413 			else
414 				error = EFAULT;
415 			break;
416 		}
417 
418 		/*
419 		 * Now do the i/o move.
420 		 */
421 		error = uiomove_fromphys(&m, page_offset, len, uio);
422 
423 		/* Make the I-cache coherent for breakpoints. */
424 		if (writing && error == 0) {
425 			vm_map_lock_read(map);
426 			if (vm_map_check_protection(map, pageno, pageno +
427 			    PAGE_SIZE, VM_PROT_EXECUTE))
428 				vm_sync_icache(map, uva, len);
429 			vm_map_unlock_read(map);
430 		}
431 
432 		/*
433 		 * Release the page.
434 		 */
435 		vm_page_unwire(m, PQ_ACTIVE);
436 
437 	} while (error == 0 && uio->uio_resid > 0);
438 
439 	return (error);
440 }
441 
442 static ssize_t
443 proc_iop(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
444     size_t len, enum uio_rw rw)
445 {
446 	struct iovec iov;
447 	struct uio uio;
448 	ssize_t slen;
449 
450 	MPASS(len < SSIZE_MAX);
451 	slen = (ssize_t)len;
452 
453 	iov.iov_base = (caddr_t)buf;
454 	iov.iov_len = len;
455 	uio.uio_iov = &iov;
456 	uio.uio_iovcnt = 1;
457 	uio.uio_offset = va;
458 	uio.uio_resid = slen;
459 	uio.uio_segflg = UIO_SYSSPACE;
460 	uio.uio_rw = rw;
461 	uio.uio_td = td;
462 	proc_rwmem(p, &uio);
463 	if (uio.uio_resid == slen)
464 		return (-1);
465 	return (slen - uio.uio_resid);
466 }
467 
468 ssize_t
469 proc_readmem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
470     size_t len)
471 {
472 
473 	return (proc_iop(td, p, va, buf, len, UIO_READ));
474 }
475 
476 ssize_t
477 proc_writemem(struct thread *td, struct proc *p, vm_offset_t va, void *buf,
478     size_t len)
479 {
480 
481 	return (proc_iop(td, p, va, buf, len, UIO_WRITE));
482 }
483 
484 static int
485 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
486 {
487 	struct vattr vattr;
488 	vm_map_t map;
489 	vm_map_entry_t entry;
490 	vm_object_t obj, tobj, lobj;
491 	struct vmspace *vm;
492 	struct vnode *vp;
493 	char *freepath, *fullpath;
494 	u_int pathlen;
495 	int error, index;
496 
497 	error = 0;
498 	obj = NULL;
499 
500 	vm = vmspace_acquire_ref(p);
501 	map = &vm->vm_map;
502 	vm_map_lock_read(map);
503 
504 	do {
505 		KASSERT((map->header.eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
506 		    ("Submap in map header"));
507 		index = 0;
508 		VM_MAP_ENTRY_FOREACH(entry, map) {
509 			if (index >= pve->pve_entry &&
510 			    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
511 				break;
512 			index++;
513 		}
514 		if (index < pve->pve_entry) {
515 			error = EINVAL;
516 			break;
517 		}
518 		if (entry == &map->header) {
519 			error = ENOENT;
520 			break;
521 		}
522 
523 		/* We got an entry. */
524 		pve->pve_entry = index + 1;
525 		pve->pve_timestamp = map->timestamp;
526 		pve->pve_start = entry->start;
527 		pve->pve_end = entry->end - 1;
528 		pve->pve_offset = entry->offset;
529 		pve->pve_prot = entry->protection |
530 		    PROT_MAX(entry->max_protection);
531 
532 		/* Backing object's path needed? */
533 		if (pve->pve_pathlen == 0)
534 			break;
535 
536 		pathlen = pve->pve_pathlen;
537 		pve->pve_pathlen = 0;
538 
539 		obj = entry->object.vm_object;
540 		if (obj != NULL)
541 			VM_OBJECT_RLOCK(obj);
542 	} while (0);
543 
544 	vm_map_unlock_read(map);
545 
546 	pve->pve_fsid = VNOVAL;
547 	pve->pve_fileid = VNOVAL;
548 
549 	if (error == 0 && obj != NULL) {
550 		lobj = obj;
551 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
552 			if (tobj != obj)
553 				VM_OBJECT_RLOCK(tobj);
554 			if (lobj != obj)
555 				VM_OBJECT_RUNLOCK(lobj);
556 			lobj = tobj;
557 			pve->pve_offset += tobj->backing_object_offset;
558 		}
559 		vp = vm_object_vnode(lobj);
560 		if (vp != NULL)
561 			vref(vp);
562 		if (lobj != obj)
563 			VM_OBJECT_RUNLOCK(lobj);
564 		VM_OBJECT_RUNLOCK(obj);
565 
566 		if (vp != NULL) {
567 			freepath = NULL;
568 			fullpath = NULL;
569 			vn_fullpath(vp, &fullpath, &freepath);
570 			vn_lock(vp, LK_SHARED | LK_RETRY);
571 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
572 				pve->pve_fileid = vattr.va_fileid;
573 				pve->pve_fsid = vattr.va_fsid;
574 			}
575 			vput(vp);
576 
577 			if (fullpath != NULL) {
578 				pve->pve_pathlen = strlen(fullpath) + 1;
579 				if (pve->pve_pathlen <= pathlen) {
580 					error = copyout(fullpath, pve->pve_path,
581 					    pve->pve_pathlen);
582 				} else
583 					error = ENAMETOOLONG;
584 			}
585 			if (freepath != NULL)
586 				free(freepath, M_TEMP);
587 		}
588 	}
589 	vmspace_free(vm);
590 	if (error == 0)
591 		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
592 		    p->p_pid, pve->pve_entry, pve->pve_start);
593 
594 	return (error);
595 }
596 
597 /*
598  * Process debugging system call.
599  */
600 #ifndef _SYS_SYSPROTO_H_
601 struct ptrace_args {
602 	int	req;
603 	pid_t	pid;
604 	caddr_t	addr;
605 	int	data;
606 };
607 #endif
608 
609 int
610 sys_ptrace(struct thread *td, struct ptrace_args *uap)
611 {
612 	/*
613 	 * XXX this obfuscation is to reduce stack usage, but the register
614 	 * structs may be too large to put on the stack anyway.
615 	 */
616 	union {
617 		struct ptrace_io_desc piod;
618 		struct ptrace_lwpinfo pl;
619 		struct ptrace_vm_entry pve;
620 		struct ptrace_coredump pc;
621 		struct ptrace_sc_remote sr;
622 		struct dbreg dbreg;
623 		struct fpreg fpreg;
624 		struct reg reg;
625 		struct iovec vec;
626 		syscallarg_t args[nitems(td->td_sa.args)];
627 		struct ptrace_sc_ret psr;
628 		int ptevents;
629 	} r;
630 	syscallarg_t pscr_args[nitems(td->td_sa.args)];
631 	void *addr;
632 	int error;
633 
634 	if (!allow_ptrace)
635 		return (ENOSYS);
636 	error = 0;
637 
638 	AUDIT_ARG_PID(uap->pid);
639 	AUDIT_ARG_CMD(uap->req);
640 	AUDIT_ARG_VALUE(uap->data);
641 	addr = &r;
642 	switch (uap->req) {
643 	case PT_GET_EVENT_MASK:
644 	case PT_LWPINFO:
645 	case PT_GET_SC_ARGS:
646 	case PT_GET_SC_RET:
647 		break;
648 	case PT_GETREGS:
649 		bzero(&r.reg, sizeof(r.reg));
650 		break;
651 	case PT_GETFPREGS:
652 		bzero(&r.fpreg, sizeof(r.fpreg));
653 		break;
654 	case PT_GETDBREGS:
655 		bzero(&r.dbreg, sizeof(r.dbreg));
656 		break;
657 	case PT_GETREGSET:
658 	case PT_SETREGSET:
659 		error = copyin(uap->addr, &r.vec, sizeof(r.vec));
660 		break;
661 	case PT_SETREGS:
662 		error = copyin(uap->addr, &r.reg, sizeof(r.reg));
663 		break;
664 	case PT_SETFPREGS:
665 		error = copyin(uap->addr, &r.fpreg, sizeof(r.fpreg));
666 		break;
667 	case PT_SETDBREGS:
668 		error = copyin(uap->addr, &r.dbreg, sizeof(r.dbreg));
669 		break;
670 	case PT_SET_EVENT_MASK:
671 		if (uap->data != sizeof(r.ptevents))
672 			error = EINVAL;
673 		else
674 			error = copyin(uap->addr, &r.ptevents, uap->data);
675 		break;
676 	case PT_IO:
677 		error = copyin(uap->addr, &r.piod, sizeof(r.piod));
678 		break;
679 	case PT_VM_ENTRY:
680 		error = copyin(uap->addr, &r.pve, sizeof(r.pve));
681 		break;
682 	case PT_COREDUMP:
683 		if (uap->data != sizeof(r.pc))
684 			error = EINVAL;
685 		else
686 			error = copyin(uap->addr, &r.pc, uap->data);
687 		break;
688 	case PT_SC_REMOTE:
689 		if (uap->data != sizeof(r.sr)) {
690 			error = EINVAL;
691 			break;
692 		}
693 		error = copyin(uap->addr, &r.sr, uap->data);
694 		if (error != 0)
695 			break;
696 		if (r.sr.pscr_nargs > nitems(td->td_sa.args)) {
697 			error = EINVAL;
698 			break;
699 		}
700 		error = copyin(r.sr.pscr_args, pscr_args,
701 		    sizeof(u_long) * r.sr.pscr_nargs);
702 		if (error != 0)
703 			break;
704 		r.sr.pscr_args = pscr_args;
705 		break;
706 	case PTINTERNAL_FIRST ... PTINTERNAL_LAST:
707 		error = EINVAL;
708 		break;
709 	default:
710 		addr = uap->addr;
711 		break;
712 	}
713 	if (error)
714 		return (error);
715 
716 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
717 	if (error)
718 		return (error);
719 
720 	switch (uap->req) {
721 	case PT_VM_ENTRY:
722 		error = copyout(&r.pve, uap->addr, sizeof(r.pve));
723 		break;
724 	case PT_IO:
725 		error = copyout(&r.piod, uap->addr, sizeof(r.piod));
726 		break;
727 	case PT_GETREGS:
728 		error = copyout(&r.reg, uap->addr, sizeof(r.reg));
729 		break;
730 	case PT_GETFPREGS:
731 		error = copyout(&r.fpreg, uap->addr, sizeof(r.fpreg));
732 		break;
733 	case PT_GETDBREGS:
734 		error = copyout(&r.dbreg, uap->addr, sizeof(r.dbreg));
735 		break;
736 	case PT_GETREGSET:
737 		error = copyout(&r.vec, uap->addr, sizeof(r.vec));
738 		break;
739 	case PT_GET_EVENT_MASK:
740 		/* NB: The size in uap->data is validated in kern_ptrace(). */
741 		error = copyout(&r.ptevents, uap->addr, uap->data);
742 		break;
743 	case PT_LWPINFO:
744 		/* NB: The size in uap->data is validated in kern_ptrace(). */
745 		error = copyout(&r.pl, uap->addr, uap->data);
746 		break;
747 	case PT_GET_SC_ARGS:
748 		error = copyout(r.args, uap->addr, MIN(uap->data,
749 		    sizeof(r.args)));
750 		break;
751 	case PT_GET_SC_RET:
752 		error = copyout(&r.psr, uap->addr, MIN(uap->data,
753 		    sizeof(r.psr)));
754 		break;
755 	case PT_SC_REMOTE:
756 		error = copyout(&r.sr.pscr_ret, uap->addr +
757 		    offsetof(struct ptrace_sc_remote, pscr_ret),
758 		    sizeof(r.sr.pscr_ret));
759 		break;
760 	}
761 
762 	return (error);
763 }
764 
765 #ifdef COMPAT_FREEBSD32
766 /*
767  *   PROC_READ(regs, td2, addr);
768  * becomes either:
769  *   proc_read_regs(td2, addr);
770  * or
771  *   proc_read_regs32(td2, addr);
772  * .. except this is done at runtime.  There is an additional
773  * complication in that PROC_WRITE disallows 32 bit consumers
774  * from writing to 64 bit address space targets.
775  */
776 #define	PROC_READ(w, t, a)	wrap32 ? \
777 	proc_read_ ## w ## 32(t, a) : \
778 	proc_read_ ## w (t, a)
779 #define	PROC_WRITE(w, t, a)	wrap32 ? \
780 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
781 	proc_write_ ## w (t, a)
782 #else
783 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
784 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
785 #endif
786 
787 void
788 proc_set_traced(struct proc *p, bool stop)
789 {
790 
791 	sx_assert(&proctree_lock, SX_XLOCKED);
792 	PROC_LOCK_ASSERT(p, MA_OWNED);
793 	p->p_flag |= P_TRACED;
794 	if (stop)
795 		p->p_flag2 |= P2_PTRACE_FSTP;
796 	p->p_ptevents = PTRACE_DEFAULT;
797 }
798 
799 void
800 ptrace_unsuspend(struct proc *p)
801 {
802 	PROC_LOCK_ASSERT(p, MA_OWNED);
803 
804 	PROC_SLOCK(p);
805 	p->p_flag &= ~(P_STOPPED_TRACE | P_STOPPED_SIG | P_WAITED);
806 	thread_unsuspend(p);
807 	PROC_SUNLOCK(p);
808 	itimer_proc_continue(p);
809 	kqtimer_proc_continue(p);
810 }
811 
812 static int
813 proc_can_ptrace(struct thread *td, struct proc *p)
814 {
815 	int error;
816 
817 	PROC_LOCK_ASSERT(p, MA_OWNED);
818 
819 	if ((p->p_flag & P_WEXIT) != 0)
820 		return (ESRCH);
821 
822 	if ((error = p_cansee(td, p)) != 0)
823 		return (error);
824 	if ((error = p_candebug(td, p)) != 0)
825 		return (error);
826 
827 	/* not being traced... */
828 	if ((p->p_flag & P_TRACED) == 0)
829 		return (EPERM);
830 
831 	/* not being traced by YOU */
832 	if (p->p_pptr != td->td_proc)
833 		return (EBUSY);
834 
835 	/* not currently stopped */
836 	if ((p->p_flag & P_STOPPED_TRACE) == 0 ||
837 	    p->p_suspcount != p->p_numthreads  ||
838 	    (p->p_flag & P_WAITED) == 0)
839 		return (EBUSY);
840 
841 	return (0);
842 }
843 
844 static struct thread *
845 ptrace_sel_coredump_thread(struct proc *p)
846 {
847 	struct thread *td2;
848 
849 	PROC_LOCK_ASSERT(p, MA_OWNED);
850 	MPASS((p->p_flag & P_STOPPED_TRACE) != 0);
851 
852 	FOREACH_THREAD_IN_PROC(p, td2) {
853 		if ((td2->td_dbgflags & TDB_SSWITCH) != 0)
854 			return (td2);
855 	}
856 	return (NULL);
857 }
858 
859 int
860 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
861 {
862 	struct iovec iov;
863 	struct uio uio;
864 	struct proc *curp, *p, *pp;
865 	struct thread *td2 = NULL, *td3;
866 	struct ptrace_io_desc *piod = NULL;
867 	struct ptrace_lwpinfo *pl;
868 	struct ptrace_sc_ret *psr;
869 	struct ptrace_sc_remote *pscr;
870 	struct file *fp;
871 	struct ptrace_coredump *pc;
872 	struct thr_coredump_req *tcq;
873 	struct thr_syscall_req *tsr;
874 	int error, num, tmp;
875 	lwpid_t tid = 0, *buf;
876 #ifdef COMPAT_FREEBSD32
877 	int wrap32 = 0, safe = 0;
878 #endif
879 	bool proctree_locked, p2_req_set;
880 
881 	curp = td->td_proc;
882 	proctree_locked = false;
883 	p2_req_set = false;
884 
885 	/* Lock proctree before locking the process. */
886 	switch (req) {
887 	case PT_TRACE_ME:
888 	case PT_ATTACH:
889 	case PT_STEP:
890 	case PT_CONTINUE:
891 	case PT_TO_SCE:
892 	case PT_TO_SCX:
893 	case PT_SYSCALL:
894 	case PT_FOLLOW_FORK:
895 	case PT_LWP_EVENTS:
896 	case PT_GET_EVENT_MASK:
897 	case PT_SET_EVENT_MASK:
898 	case PT_DETACH:
899 	case PT_GET_SC_ARGS:
900 		sx_xlock(&proctree_lock);
901 		proctree_locked = true;
902 		break;
903 	default:
904 		break;
905 	}
906 
907 	if (req == PT_TRACE_ME) {
908 		p = td->td_proc;
909 		PROC_LOCK(p);
910 	} else {
911 		if (pid <= PID_MAX) {
912 			if ((p = pfind(pid)) == NULL) {
913 				if (proctree_locked)
914 					sx_xunlock(&proctree_lock);
915 				return (ESRCH);
916 			}
917 		} else {
918 			td2 = tdfind(pid, -1);
919 			if (td2 == NULL) {
920 				if (proctree_locked)
921 					sx_xunlock(&proctree_lock);
922 				return (ESRCH);
923 			}
924 			p = td2->td_proc;
925 			tid = pid;
926 			pid = p->p_pid;
927 		}
928 	}
929 	AUDIT_ARG_PROCESS(p);
930 
931 	if ((p->p_flag & P_WEXIT) != 0) {
932 		error = ESRCH;
933 		goto fail;
934 	}
935 	if ((error = p_cansee(td, p)) != 0)
936 		goto fail;
937 
938 	if ((error = p_candebug(td, p)) != 0)
939 		goto fail;
940 
941 	/*
942 	 * System processes can't be debugged.
943 	 */
944 	if ((p->p_flag & P_SYSTEM) != 0) {
945 		error = EINVAL;
946 		goto fail;
947 	}
948 
949 	if (tid == 0) {
950 		if ((p->p_flag & P_STOPPED_TRACE) != 0)
951 			td2 = p->p_xthread;
952 		if (td2 == NULL)
953 			td2 = FIRST_THREAD_IN_PROC(p);
954 		tid = td2->td_tid;
955 	}
956 
957 #ifdef COMPAT_FREEBSD32
958 	/*
959 	 * Test if we're a 32 bit client and what the target is.
960 	 * Set the wrap controls accordingly.
961 	 */
962 	if (SV_CURPROC_FLAG(SV_ILP32)) {
963 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
964 			safe = 1;
965 		wrap32 = 1;
966 	}
967 #endif
968 	/*
969 	 * Permissions check
970 	 */
971 	switch (req) {
972 	case PT_TRACE_ME:
973 		/*
974 		 * Always legal, when there is a parent process which
975 		 * could trace us.  Otherwise, reject.
976 		 */
977 		if ((p->p_flag & P_TRACED) != 0) {
978 			error = EBUSY;
979 			goto fail;
980 		}
981 		if (p->p_pptr == initproc) {
982 			error = EPERM;
983 			goto fail;
984 		}
985 		break;
986 
987 	case PT_ATTACH:
988 		/* Self */
989 		if (p == td->td_proc) {
990 			error = EINVAL;
991 			goto fail;
992 		}
993 
994 		/* Already traced */
995 		if (p->p_flag & P_TRACED) {
996 			error = EBUSY;
997 			goto fail;
998 		}
999 
1000 		/* Can't trace an ancestor if you're being traced. */
1001 		if (curp->p_flag & P_TRACED) {
1002 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
1003 				if (pp == p) {
1004 					error = EINVAL;
1005 					goto fail;
1006 				}
1007 			}
1008 		}
1009 
1010 		/* OK */
1011 		break;
1012 
1013 	case PT_CLEARSTEP:
1014 		/* Allow thread to clear single step for itself */
1015 		if (td->td_tid == tid)
1016 			break;
1017 
1018 		/* FALLTHROUGH */
1019 	default:
1020 		/*
1021 		 * Check for ptrace eligibility before waiting for
1022 		 * holds to drain.
1023 		 */
1024 		error = proc_can_ptrace(td, p);
1025 		if (error != 0)
1026 			goto fail;
1027 
1028 		/*
1029 		 * Block parallel ptrace requests.  Most important, do
1030 		 * not allow other thread in debugger to continue the
1031 		 * debuggee until coredump finished.
1032 		 */
1033 		while ((p->p_flag2 & P2_PTRACEREQ) != 0) {
1034 			if (proctree_locked)
1035 				sx_xunlock(&proctree_lock);
1036 			error = msleep(&p->p_flag2, &p->p_mtx, PPAUSE | PCATCH |
1037 			    (proctree_locked ? PDROP : 0), "pptrace", 0);
1038 			if (proctree_locked) {
1039 				sx_xlock(&proctree_lock);
1040 				PROC_LOCK(p);
1041 			}
1042 			if (error == 0 && td2->td_proc != p)
1043 				error = ESRCH;
1044 			if (error == 0)
1045 				error = proc_can_ptrace(td, p);
1046 			if (error != 0)
1047 				goto fail;
1048 		}
1049 
1050 		/* Ok */
1051 		break;
1052 	}
1053 
1054 	/*
1055 	 * Keep this process around and request parallel ptrace()
1056 	 * request to wait until we finish this request.
1057 	 */
1058 	MPASS((p->p_flag2 & P2_PTRACEREQ) == 0);
1059 	p->p_flag2 |= P2_PTRACEREQ;
1060 	p2_req_set = true;
1061 	_PHOLD(p);
1062 
1063 	/*
1064 	 * Actually do the requests
1065 	 */
1066 
1067 	td->td_retval[0] = 0;
1068 
1069 	switch (req) {
1070 	case PT_TRACE_ME:
1071 		/* set my trace flag and "owner" so it can read/write me */
1072 		proc_set_traced(p, false);
1073 		if (p->p_flag & P_PPWAIT)
1074 			p->p_flag |= P_PPTRACE;
1075 		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
1076 		break;
1077 
1078 	case PT_ATTACH:
1079 		/* security check done above */
1080 		/*
1081 		 * It would be nice if the tracing relationship was separate
1082 		 * from the parent relationship but that would require
1083 		 * another set of links in the proc struct or for "wait"
1084 		 * to scan the entire proc table.  To make life easier,
1085 		 * we just re-parent the process we're trying to trace.
1086 		 * The old parent is remembered so we can put things back
1087 		 * on a "detach".
1088 		 */
1089 		proc_set_traced(p, true);
1090 		proc_reparent(p, td->td_proc, false);
1091 		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
1092 		    p->p_oppid);
1093 
1094 		sx_xunlock(&proctree_lock);
1095 		proctree_locked = false;
1096 		MPASS(p->p_xthread == NULL);
1097 		MPASS((p->p_flag & P_STOPPED_TRACE) == 0);
1098 
1099 		/*
1100 		 * If already stopped due to a stop signal, clear the
1101 		 * existing stop before triggering a traced SIGSTOP.
1102 		 */
1103 		if ((p->p_flag & P_STOPPED_SIG) != 0) {
1104 			PROC_SLOCK(p);
1105 			p->p_flag &= ~(P_STOPPED_SIG | P_WAITED);
1106 			thread_unsuspend(p);
1107 			PROC_SUNLOCK(p);
1108 		}
1109 
1110 		kern_psignal(p, SIGSTOP);
1111 		break;
1112 
1113 	case PT_CLEARSTEP:
1114 		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
1115 		    p->p_pid);
1116 		error = ptrace_clear_single_step(td2);
1117 		break;
1118 
1119 	case PT_SETSTEP:
1120 		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
1121 		    p->p_pid);
1122 		error = ptrace_single_step(td2);
1123 		break;
1124 
1125 	case PT_SUSPEND:
1126 		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
1127 		    p->p_pid);
1128 		td2->td_dbgflags |= TDB_SUSPEND;
1129 		ast_sched(td2, TDA_SUSPEND);
1130 		break;
1131 
1132 	case PT_RESUME:
1133 		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
1134 		    p->p_pid);
1135 		td2->td_dbgflags &= ~TDB_SUSPEND;
1136 		break;
1137 
1138 	case PT_FOLLOW_FORK:
1139 		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
1140 		    p->p_ptevents & PTRACE_FORK ? "enabled" : "disabled",
1141 		    data ? "enabled" : "disabled");
1142 		if (data)
1143 			p->p_ptevents |= PTRACE_FORK;
1144 		else
1145 			p->p_ptevents &= ~PTRACE_FORK;
1146 		break;
1147 
1148 	case PT_LWP_EVENTS:
1149 		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
1150 		    p->p_ptevents & PTRACE_LWP ? "enabled" : "disabled",
1151 		    data ? "enabled" : "disabled");
1152 		if (data)
1153 			p->p_ptevents |= PTRACE_LWP;
1154 		else
1155 			p->p_ptevents &= ~PTRACE_LWP;
1156 		break;
1157 
1158 	case PT_GET_EVENT_MASK:
1159 		if (data != sizeof(p->p_ptevents)) {
1160 			error = EINVAL;
1161 			break;
1162 		}
1163 		CTR2(KTR_PTRACE, "PT_GET_EVENT_MASK: pid %d mask %#x", p->p_pid,
1164 		    p->p_ptevents);
1165 		*(int *)addr = p->p_ptevents;
1166 		break;
1167 
1168 	case PT_SET_EVENT_MASK:
1169 		if (data != sizeof(p->p_ptevents)) {
1170 			error = EINVAL;
1171 			break;
1172 		}
1173 		tmp = *(int *)addr;
1174 		if ((tmp & ~(PTRACE_EXEC | PTRACE_SCE | PTRACE_SCX |
1175 		    PTRACE_FORK | PTRACE_LWP | PTRACE_VFORK)) != 0) {
1176 			error = EINVAL;
1177 			break;
1178 		}
1179 		CTR3(KTR_PTRACE, "PT_SET_EVENT_MASK: pid %d mask %#x -> %#x",
1180 		    p->p_pid, p->p_ptevents, tmp);
1181 		p->p_ptevents = tmp;
1182 		break;
1183 
1184 	case PT_GET_SC_ARGS:
1185 	case PTLINUX_GET_SC_ARGS:
1186 		CTR2(KTR_PTRACE, "%s: pid %d", req == PT_GET_SC_ARGS ?
1187 		    "PT_GET_SC_ARGS" : "PT_LINUX_GET_SC_ARGS", p->p_pid);
1188 		if (((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) == 0 &&
1189 		     td2->td_sa.code == 0)
1190 #ifdef COMPAT_FREEBSD32
1191 		    || (wrap32 && !safe)
1192 #endif
1193 		    ) {
1194 			error = EINVAL;
1195 			break;
1196 		}
1197 		if (req == PT_GET_SC_ARGS) {
1198 			bzero(addr, sizeof(td2->td_sa.args));
1199 			bcopy(td2->td_sa.args, addr, td2->td_sa.callp->sy_narg *
1200 			    sizeof(syscallarg_t));
1201 		} else {
1202 			/*
1203 			 * Emulate a Linux bug which which strace(1) depends on:
1204 			 * at initialization it tests whether ptrace works by
1205 			 * calling close(2), or some other single-argument
1206 			 * syscall, _with six arguments_, and then verifies
1207 			 * whether it can fetch them all using this API;
1208 			 * otherwise it bails out.
1209 			 */
1210 			bcopy(td2->td_sa.args, addr, 6 * sizeof(syscallarg_t));
1211 		}
1212 		break;
1213 
1214 	case PT_GET_SC_RET:
1215 		if ((td2->td_dbgflags & (TDB_SCX)) == 0
1216 #ifdef COMPAT_FREEBSD32
1217 		    || (wrap32 && !safe)
1218 #endif
1219 		    ) {
1220 			error = EINVAL;
1221 			break;
1222 		}
1223 		psr = addr;
1224 		bzero(psr, sizeof(*psr));
1225 		psr->sr_error = td2->td_errno;
1226 		if (psr->sr_error == 0) {
1227 			psr->sr_retval[0] = td2->td_retval[0];
1228 			psr->sr_retval[1] = td2->td_retval[1];
1229 		}
1230 		CTR4(KTR_PTRACE,
1231 		    "PT_GET_SC_RET: pid %d error %d retval %#lx,%#lx",
1232 		    p->p_pid, psr->sr_error, psr->sr_retval[0],
1233 		    psr->sr_retval[1]);
1234 		break;
1235 
1236 	case PT_STEP:
1237 	case PT_CONTINUE:
1238 	case PT_TO_SCE:
1239 	case PT_TO_SCX:
1240 	case PT_SYSCALL:
1241 	case PT_DETACH:
1242 		/* Zero means do not send any signal */
1243 		if (data < 0 || data > _SIG_MAXSIG) {
1244 			error = EINVAL;
1245 			break;
1246 		}
1247 
1248 		switch (req) {
1249 		case PT_STEP:
1250 			CTR3(KTR_PTRACE, "PT_STEP: tid %d (pid %d), sig = %d",
1251 			    td2->td_tid, p->p_pid, data);
1252 			error = ptrace_single_step(td2);
1253 			if (error)
1254 				goto out;
1255 			break;
1256 		case PT_CONTINUE:
1257 		case PT_TO_SCE:
1258 		case PT_TO_SCX:
1259 		case PT_SYSCALL:
1260 			if (addr != (void *)1) {
1261 				error = ptrace_set_pc(td2,
1262 				    (u_long)(uintfptr_t)addr);
1263 				if (error)
1264 					goto out;
1265 				td2->td_dbgflags |= TDB_USERWR;
1266 			}
1267 			switch (req) {
1268 			case PT_TO_SCE:
1269 				p->p_ptevents |= PTRACE_SCE;
1270 				CTR4(KTR_PTRACE,
1271 		    "PT_TO_SCE: pid %d, events = %#x, PC = %#lx, sig = %d",
1272 				    p->p_pid, p->p_ptevents,
1273 				    (u_long)(uintfptr_t)addr, data);
1274 				break;
1275 			case PT_TO_SCX:
1276 				p->p_ptevents |= PTRACE_SCX;
1277 				CTR4(KTR_PTRACE,
1278 		    "PT_TO_SCX: pid %d, events = %#x, PC = %#lx, sig = %d",
1279 				    p->p_pid, p->p_ptevents,
1280 				    (u_long)(uintfptr_t)addr, data);
1281 				break;
1282 			case PT_SYSCALL:
1283 				p->p_ptevents |= PTRACE_SYSCALL;
1284 				CTR4(KTR_PTRACE,
1285 		    "PT_SYSCALL: pid %d, events = %#x, PC = %#lx, sig = %d",
1286 				    p->p_pid, p->p_ptevents,
1287 				    (u_long)(uintfptr_t)addr, data);
1288 				break;
1289 			case PT_CONTINUE:
1290 				CTR3(KTR_PTRACE,
1291 				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
1292 				    p->p_pid, (u_long)(uintfptr_t)addr, data);
1293 				break;
1294 			}
1295 			break;
1296 		case PT_DETACH:
1297 			/*
1298 			 * Clear P_TRACED before reparenting
1299 			 * a detached process back to its original
1300 			 * parent.  Otherwise the debugee will be set
1301 			 * as an orphan of the debugger.
1302 			 */
1303 			p->p_flag &= ~(P_TRACED | P_WAITED);
1304 
1305 			/*
1306 			 * Reset the process parent.
1307 			 */
1308 			if (p->p_oppid != p->p_pptr->p_pid) {
1309 				PROC_LOCK(p->p_pptr);
1310 				sigqueue_take(p->p_ksi);
1311 				PROC_UNLOCK(p->p_pptr);
1312 
1313 				pp = proc_realparent(p);
1314 				proc_reparent(p, pp, false);
1315 				if (pp == initproc)
1316 					p->p_sigparent = SIGCHLD;
1317 				CTR3(KTR_PTRACE,
1318 			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
1319 				    p->p_pid, pp->p_pid, data);
1320 			} else {
1321 				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1322 				    p->p_pid, data);
1323 			}
1324 
1325 			p->p_ptevents = 0;
1326 			FOREACH_THREAD_IN_PROC(p, td3) {
1327 				if ((td3->td_dbgflags & TDB_FSTP) != 0) {
1328 					sigqueue_delete(&td3->td_sigqueue,
1329 					    SIGSTOP);
1330 				}
1331 				td3->td_dbgflags &= ~(TDB_XSIG | TDB_FSTP |
1332 				    TDB_SUSPEND | TDB_BORN);
1333 			}
1334 
1335 			if ((p->p_flag2 & P2_PTRACE_FSTP) != 0) {
1336 				sigqueue_delete(&p->p_sigqueue, SIGSTOP);
1337 				p->p_flag2 &= ~P2_PTRACE_FSTP;
1338 			}
1339 
1340 			/*
1341 			 * Send SIGCHLD and wakeup the parent as needed.  It
1342 			 * may be the case that they had stopped the child
1343 			 * before it got ptraced, and now they're in the middle
1344 			 * of a wait(2) for it to continue.
1345 			 */
1346 			PROC_LOCK(p->p_pptr);
1347 			childproc_continued(p);
1348 			PROC_UNLOCK(p->p_pptr);
1349 			break;
1350 		}
1351 
1352 		sx_xunlock(&proctree_lock);
1353 		proctree_locked = false;
1354 
1355 	sendsig:
1356 		MPASS(!proctree_locked);
1357 
1358 		/*
1359 		 * Clear the pending event for the thread that just
1360 		 * reported its event (p_xthread), if any.  This may
1361 		 * not be the thread passed to PT_CONTINUE, PT_STEP,
1362 		 * etc. if the debugger is resuming a different
1363 		 * thread.  There might be no reporting thread if
1364 		 * the process was just attached.
1365 		 *
1366 		 * Deliver any pending signal via the reporting thread.
1367 		 */
1368 		if (p->p_xthread != NULL) {
1369 			p->p_xthread->td_dbgflags &= ~TDB_XSIG;
1370 			p->p_xthread->td_xsig = data;
1371 			p->p_xthread = NULL;
1372 		}
1373 		p->p_xsig = data;
1374 
1375 		/*
1376 		 * P_WKILLED is insurance that a PT_KILL/SIGKILL
1377 		 * always works immediately, even if another thread is
1378 		 * unsuspended first and attempts to handle a
1379 		 * different signal or if the POSIX.1b style signal
1380 		 * queue cannot accommodate any new signals.
1381 		 */
1382 		if (data == SIGKILL)
1383 			proc_wkilled(p);
1384 
1385 		/*
1386 		 * If the PT_CONTINUE-like operation is attempted on
1387 		 * the thread on sleepq, this is possible only after
1388 		 * the transparent PT_ATTACH.  In this case, if the
1389 		 * caller modified the thread state, e.g. by writing
1390 		 * register file or specifying the pc, make the thread
1391 		 * xstopped by waking it up.
1392 		 */
1393 		if ((td2->td_dbgflags & TDB_USERWR) != 0) {
1394 			if (pt_attach_transparent) {
1395 				thread_lock(td2);
1396 				if (TD_ON_SLEEPQ(td2) &&
1397 				    (td2->td_flags & TDF_SINTR) != 0) {
1398 					sleepq_abort(td2, EINTR);
1399 				} else {
1400 					thread_unlock(td2);
1401 				}
1402 			}
1403 			td2->td_dbgflags &= ~TDB_USERWR;
1404 		}
1405 
1406 		/*
1407 		 * Unsuspend all threads.  To leave a thread
1408 		 * suspended, use PT_SUSPEND to suspend it before
1409 		 * continuing the process.
1410 		 */
1411 		ptrace_unsuspend(p);
1412 		break;
1413 
1414 	case PT_WRITE_I:
1415 	case PT_WRITE_D:
1416 		td2->td_dbgflags |= TDB_USERWR;
1417 		PROC_UNLOCK(p);
1418 		error = 0;
1419 		if (proc_writemem(td, p, (off_t)(uintptr_t)addr, &data,
1420 		    sizeof(int)) != sizeof(int))
1421 			error = ENOMEM;
1422 		else
1423 			CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1424 			    p->p_pid, addr, data);
1425 		PROC_LOCK(p);
1426 		break;
1427 
1428 	case PT_READ_I:
1429 	case PT_READ_D:
1430 		PROC_UNLOCK(p);
1431 		error = tmp = 0;
1432 		if (proc_readmem(td, p, (off_t)(uintptr_t)addr, &tmp,
1433 		    sizeof(int)) != sizeof(int))
1434 			error = ENOMEM;
1435 		else
1436 			CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1437 			    p->p_pid, addr, tmp);
1438 		td->td_retval[0] = tmp;
1439 		PROC_LOCK(p);
1440 		break;
1441 
1442 	case PT_IO:
1443 		piod = addr;
1444 		if (piod->piod_len > SSIZE_MAX) {
1445 			error = EINVAL;
1446 			goto out;
1447 		}
1448 		iov.iov_base = piod->piod_addr;
1449 		iov.iov_len = piod->piod_len;
1450 		uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1451 		uio.uio_resid = piod->piod_len;
1452 		uio.uio_iov = &iov;
1453 		uio.uio_iovcnt = 1;
1454 		uio.uio_segflg = UIO_USERSPACE;
1455 		uio.uio_td = td;
1456 		switch (piod->piod_op) {
1457 		case PIOD_READ_D:
1458 		case PIOD_READ_I:
1459 			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1460 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1461 			uio.uio_rw = UIO_READ;
1462 			break;
1463 		case PIOD_WRITE_D:
1464 		case PIOD_WRITE_I:
1465 			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1466 			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1467 			td2->td_dbgflags |= TDB_USERWR;
1468 			uio.uio_rw = UIO_WRITE;
1469 			break;
1470 		default:
1471 			error = EINVAL;
1472 			goto out;
1473 		}
1474 		PROC_UNLOCK(p);
1475 		error = proc_rwmem(p, &uio);
1476 		piod->piod_len -= uio.uio_resid;
1477 		PROC_LOCK(p);
1478 		break;
1479 
1480 	case PT_KILL:
1481 		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1482 		data = SIGKILL;
1483 		goto sendsig;	/* in PT_CONTINUE above */
1484 
1485 	case PT_SETREGS:
1486 		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1487 		    p->p_pid);
1488 		td2->td_dbgflags |= TDB_USERWR;
1489 		error = PROC_WRITE(regs, td2, addr);
1490 		break;
1491 
1492 	case PT_GETREGS:
1493 		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1494 		    p->p_pid);
1495 		error = PROC_READ(regs, td2, addr);
1496 		break;
1497 
1498 	case PT_SETFPREGS:
1499 		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1500 		    p->p_pid);
1501 		td2->td_dbgflags |= TDB_USERWR;
1502 		error = PROC_WRITE(fpregs, td2, addr);
1503 		break;
1504 
1505 	case PT_GETFPREGS:
1506 		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1507 		    p->p_pid);
1508 		error = PROC_READ(fpregs, td2, addr);
1509 		break;
1510 
1511 	case PT_SETDBREGS:
1512 		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1513 		    p->p_pid);
1514 		td2->td_dbgflags |= TDB_USERWR;
1515 		error = PROC_WRITE(dbregs, td2, addr);
1516 		break;
1517 
1518 	case PT_GETDBREGS:
1519 		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1520 		    p->p_pid);
1521 		error = PROC_READ(dbregs, td2, addr);
1522 		break;
1523 
1524 	case PT_SETREGSET:
1525 		CTR2(KTR_PTRACE, "PT_SETREGSET: tid %d (pid %d)", td2->td_tid,
1526 		    p->p_pid);
1527 		error = proc_write_regset(td2, data, addr);
1528 		break;
1529 
1530 	case PT_GETREGSET:
1531 		CTR2(KTR_PTRACE, "PT_GETREGSET: tid %d (pid %d)", td2->td_tid,
1532 		    p->p_pid);
1533 		error = proc_read_regset(td2, data, addr);
1534 		break;
1535 
1536 	case PT_LWPINFO:
1537 		if (data <= 0 || data > sizeof(*pl)) {
1538 			error = EINVAL;
1539 			break;
1540 		}
1541 		pl = addr;
1542 		bzero(pl, sizeof(*pl));
1543 		pl->pl_lwpid = td2->td_tid;
1544 		pl->pl_event = PL_EVENT_NONE;
1545 		pl->pl_flags = 0;
1546 		if (td2->td_dbgflags & TDB_XSIG) {
1547 			pl->pl_event = PL_EVENT_SIGNAL;
1548 			if (td2->td_si.si_signo != 0 &&
1549 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1550 			    + sizeof(pl->pl_siginfo)){
1551 				pl->pl_flags |= PL_FLAG_SI;
1552 				pl->pl_siginfo = td2->td_si;
1553 			}
1554 		}
1555 		if (td2->td_dbgflags & TDB_SCE)
1556 			pl->pl_flags |= PL_FLAG_SCE;
1557 		else if (td2->td_dbgflags & TDB_SCX)
1558 			pl->pl_flags |= PL_FLAG_SCX;
1559 		if (td2->td_dbgflags & TDB_EXEC)
1560 			pl->pl_flags |= PL_FLAG_EXEC;
1561 		if (td2->td_dbgflags & TDB_FORK) {
1562 			pl->pl_flags |= PL_FLAG_FORKED;
1563 			pl->pl_child_pid = td2->td_dbg_forked;
1564 			if (td2->td_dbgflags & TDB_VFORK)
1565 				pl->pl_flags |= PL_FLAG_VFORKED;
1566 		} else if ((td2->td_dbgflags & (TDB_SCX | TDB_VFORK)) ==
1567 		    TDB_VFORK)
1568 			pl->pl_flags |= PL_FLAG_VFORK_DONE;
1569 		if (td2->td_dbgflags & TDB_CHILD)
1570 			pl->pl_flags |= PL_FLAG_CHILD;
1571 		if (td2->td_dbgflags & TDB_BORN)
1572 			pl->pl_flags |= PL_FLAG_BORN;
1573 		if (td2->td_dbgflags & TDB_EXIT)
1574 			pl->pl_flags |= PL_FLAG_EXITED;
1575 		pl->pl_sigmask = td2->td_sigmask;
1576 		pl->pl_siglist = td2->td_siglist;
1577 		strcpy(pl->pl_tdname, td2->td_name);
1578 		if (td2->td_sa.code != 0) {
1579 			pl->pl_syscall_code = td2->td_sa.code;
1580 			pl->pl_syscall_narg = td2->td_sa.callp->sy_narg;
1581 		}
1582 		CTR6(KTR_PTRACE,
1583     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1584 		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1585 		    pl->pl_child_pid, pl->pl_syscall_code);
1586 		break;
1587 
1588 	case PT_GETNUMLWPS:
1589 		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1590 		    p->p_numthreads);
1591 		td->td_retval[0] = p->p_numthreads;
1592 		break;
1593 
1594 	case PT_GETLWPLIST:
1595 		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1596 		    p->p_pid, data, p->p_numthreads);
1597 		if (data <= 0) {
1598 			error = EINVAL;
1599 			break;
1600 		}
1601 		num = imin(p->p_numthreads, data);
1602 		PROC_UNLOCK(p);
1603 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1604 		tmp = 0;
1605 		PROC_LOCK(p);
1606 		FOREACH_THREAD_IN_PROC(p, td2) {
1607 			if (tmp >= num)
1608 				break;
1609 			buf[tmp++] = td2->td_tid;
1610 		}
1611 		PROC_UNLOCK(p);
1612 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1613 		free(buf, M_TEMP);
1614 		if (!error)
1615 			td->td_retval[0] = tmp;
1616 		PROC_LOCK(p);
1617 		break;
1618 
1619 	case PT_VM_TIMESTAMP:
1620 		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1621 		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1622 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1623 		break;
1624 
1625 	case PT_VM_ENTRY:
1626 		PROC_UNLOCK(p);
1627 		error = ptrace_vm_entry(td, p, addr);
1628 		PROC_LOCK(p);
1629 		break;
1630 
1631 	case PT_COREDUMP:
1632 		pc = addr;
1633 		CTR2(KTR_PTRACE, "PT_COREDUMP: pid %d, fd %d",
1634 		    p->p_pid, pc->pc_fd);
1635 
1636 		if ((pc->pc_flags & ~(PC_COMPRESS | PC_ALL)) != 0) {
1637 			error = EINVAL;
1638 			break;
1639 		}
1640 		PROC_UNLOCK(p);
1641 
1642 		tcq = malloc(sizeof(*tcq), M_TEMP, M_WAITOK | M_ZERO);
1643 		fp = NULL;
1644 		error = fget_write(td, pc->pc_fd, &cap_write_rights, &fp);
1645 		if (error != 0)
1646 			goto coredump_cleanup_nofp;
1647 		if (fp->f_type != DTYPE_VNODE || fp->f_vnode->v_type != VREG) {
1648 			error = EPIPE;
1649 			goto coredump_cleanup;
1650 		}
1651 
1652 		PROC_LOCK(p);
1653 		error = proc_can_ptrace(td, p);
1654 		if (error != 0)
1655 			goto coredump_cleanup_locked;
1656 
1657 		td2 = ptrace_sel_coredump_thread(p);
1658 		if (td2 == NULL) {
1659 			error = EBUSY;
1660 			goto coredump_cleanup_locked;
1661 		}
1662 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1663 		    TDB_SCREMOTEREQ)) == 0,
1664 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1665 
1666 		tcq->tc_vp = fp->f_vnode;
1667 		tcq->tc_limit = pc->pc_limit == 0 ? OFF_MAX : pc->pc_limit;
1668 		tcq->tc_flags = SVC_PT_COREDUMP;
1669 		if ((pc->pc_flags & PC_COMPRESS) == 0)
1670 			tcq->tc_flags |= SVC_NOCOMPRESS;
1671 		if ((pc->pc_flags & PC_ALL) != 0)
1672 			tcq->tc_flags |= SVC_ALL;
1673 		td2->td_remotereq = tcq;
1674 		td2->td_dbgflags |= TDB_COREDUMPREQ;
1675 		thread_run_flash(td2);
1676 		while ((td2->td_dbgflags & TDB_COREDUMPREQ) != 0)
1677 			msleep(p, &p->p_mtx, PPAUSE, "crdmp", 0);
1678 		error = tcq->tc_error;
1679 coredump_cleanup_locked:
1680 		PROC_UNLOCK(p);
1681 coredump_cleanup:
1682 		fdrop(fp, td);
1683 coredump_cleanup_nofp:
1684 		free(tcq, M_TEMP);
1685 		PROC_LOCK(p);
1686 		break;
1687 
1688 	case PT_SC_REMOTE:
1689 		pscr = addr;
1690 		CTR2(KTR_PTRACE, "PT_SC_REMOTE: pid %d, syscall %d",
1691 		    p->p_pid, pscr->pscr_syscall);
1692 		if ((td2->td_dbgflags & TDB_BOUNDARY) == 0) {
1693 			error = EBUSY;
1694 			break;
1695 		}
1696 		PROC_UNLOCK(p);
1697 		MPASS(pscr->pscr_nargs <= nitems(td->td_sa.args));
1698 
1699 		tsr = malloc(sizeof(struct thr_syscall_req), M_TEMP,
1700 		    M_WAITOK | M_ZERO);
1701 
1702 		tsr->ts_sa.code = pscr->pscr_syscall;
1703 		tsr->ts_nargs = pscr->pscr_nargs;
1704 		memcpy(&tsr->ts_sa.args, pscr->pscr_args,
1705 		    sizeof(syscallarg_t) * tsr->ts_nargs);
1706 
1707 		PROC_LOCK(p);
1708 		error = proc_can_ptrace(td, p);
1709 		if (error != 0) {
1710 			free(tsr, M_TEMP);
1711 			break;
1712 		}
1713 		if (td2->td_proc != p) {
1714 			free(tsr, M_TEMP);
1715 			error = ESRCH;
1716 			break;
1717 		}
1718 		KASSERT((td2->td_dbgflags & (TDB_COREDUMPREQ |
1719 		    TDB_SCREMOTEREQ)) == 0,
1720 		    ("proc %d tid %d req coredump", p->p_pid, td2->td_tid));
1721 
1722 		td2->td_remotereq = tsr;
1723 		td2->td_dbgflags |= TDB_SCREMOTEREQ;
1724 		thread_run_flash(td2);
1725 		while ((td2->td_dbgflags & TDB_SCREMOTEREQ) != 0)
1726 			msleep(p, &p->p_mtx, PPAUSE, "pscrx", 0);
1727 		error = 0;
1728 		memcpy(&pscr->pscr_ret, &tsr->ts_ret, sizeof(tsr->ts_ret));
1729 		free(tsr, M_TEMP);
1730 		break;
1731 
1732 	default:
1733 #ifdef __HAVE_PTRACE_MACHDEP
1734 		if (req >= PT_FIRSTMACH) {
1735 			PROC_UNLOCK(p);
1736 			error = cpu_ptrace(td2, req, addr, data);
1737 			PROC_LOCK(p);
1738 		} else
1739 #endif
1740 			/* Unknown request. */
1741 			error = EINVAL;
1742 		break;
1743 	}
1744 out:
1745 	/* Drop our hold on this process now that the request has completed. */
1746 	_PRELE(p);
1747 fail:
1748 	if (p2_req_set) {
1749 		if ((p->p_flag2 & P2_PTRACEREQ) != 0)
1750 			wakeup(&p->p_flag2);
1751 		p->p_flag2 &= ~P2_PTRACEREQ;
1752 	}
1753 	PROC_UNLOCK(p);
1754 	if (proctree_locked)
1755 		sx_xunlock(&proctree_lock);
1756 	return (error);
1757 }
1758 #undef PROC_READ
1759 #undef PROC_WRITE
1760