xref: /freebsd/sys/kern/sys_process.c (revision 27c9070e7e726d96ce4995cfc3f64a284fdaff2c)
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 	int	pl_child_pid;		/* New child pid */
98 };
99 
100 #endif
101 
102 /*
103  * Functions implemented using PROC_ACTION():
104  *
105  * proc_read_regs(proc, regs)
106  *	Get the current user-visible register set from the process
107  *	and copy it into the regs structure (<machine/reg.h>).
108  *	The process is stopped at the time read_regs is called.
109  *
110  * proc_write_regs(proc, regs)
111  *	Update the current register set from the passed in regs
112  *	structure.  Take care to avoid clobbering special CPU
113  *	registers or privileged bits in the PSL.
114  *	Depending on the architecture this may have fix-up work to do,
115  *	especially if the IAR or PCW are modified.
116  *	The process is stopped at the time write_regs is called.
117  *
118  * proc_read_fpregs, proc_write_fpregs
119  *	deal with the floating point register set, otherwise as above.
120  *
121  * proc_read_dbregs, proc_write_dbregs
122  *	deal with the processor debug register set, otherwise as above.
123  *
124  * proc_sstep(proc)
125  *	Arrange for the process to trap after executing a single instruction.
126  */
127 
128 #define	PROC_ACTION(action) do {					\
129 	int error;							\
130 									\
131 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
132 	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
133 		error = EIO;						\
134 	else								\
135 		error = (action);					\
136 	return (error);							\
137 } while(0)
138 
139 int
140 proc_read_regs(struct thread *td, struct reg *regs)
141 {
142 
143 	PROC_ACTION(fill_regs(td, regs));
144 }
145 
146 int
147 proc_write_regs(struct thread *td, struct reg *regs)
148 {
149 
150 	PROC_ACTION(set_regs(td, regs));
151 }
152 
153 int
154 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
155 {
156 
157 	PROC_ACTION(fill_dbregs(td, dbregs));
158 }
159 
160 int
161 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
162 {
163 
164 	PROC_ACTION(set_dbregs(td, dbregs));
165 }
166 
167 /*
168  * Ptrace doesn't support fpregs at all, and there are no security holes
169  * or translations for fpregs, so we can just copy them.
170  */
171 int
172 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
173 {
174 
175 	PROC_ACTION(fill_fpregs(td, fpregs));
176 }
177 
178 int
179 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
180 {
181 
182 	PROC_ACTION(set_fpregs(td, fpregs));
183 }
184 
185 #ifdef COMPAT_FREEBSD32
186 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
187 int
188 proc_read_regs32(struct thread *td, struct reg32 *regs32)
189 {
190 
191 	PROC_ACTION(fill_regs32(td, regs32));
192 }
193 
194 int
195 proc_write_regs32(struct thread *td, struct reg32 *regs32)
196 {
197 
198 	PROC_ACTION(set_regs32(td, regs32));
199 }
200 
201 int
202 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
203 {
204 
205 	PROC_ACTION(fill_dbregs32(td, dbregs32));
206 }
207 
208 int
209 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
210 {
211 
212 	PROC_ACTION(set_dbregs32(td, dbregs32));
213 }
214 
215 int
216 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
217 {
218 
219 	PROC_ACTION(fill_fpregs32(td, fpregs32));
220 }
221 
222 int
223 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
224 {
225 
226 	PROC_ACTION(set_fpregs32(td, fpregs32));
227 }
228 #endif
229 
230 int
231 proc_sstep(struct thread *td)
232 {
233 
234 	PROC_ACTION(ptrace_single_step(td));
235 }
236 
237 int
238 proc_rwmem(struct proc *p, struct uio *uio)
239 {
240 	vm_map_t map;
241 	vm_offset_t pageno;		/* page number */
242 	vm_prot_t reqprot;
243 	int error, fault_flags, page_offset, 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 	/*
259 	 * If we are writing, then we request vm_fault() to create a private
260 	 * copy of each page.  Since these copies will not be writeable by the
261 	 * process, we must explicity request that they be dirtied.
262 	 */
263 	writing = uio->uio_rw == UIO_WRITE;
264 	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
265 	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
266 
267 	/*
268 	 * Only map in one page at a time.  We don't have to, but it
269 	 * makes things easier.  This way is trivial - right?
270 	 */
271 	do {
272 		vm_offset_t uva;
273 		u_int len;
274 		vm_page_t m;
275 
276 		uva = (vm_offset_t)uio->uio_offset;
277 
278 		/*
279 		 * Get the page number of this segment.
280 		 */
281 		pageno = trunc_page(uva);
282 		page_offset = uva - pageno;
283 
284 		/*
285 		 * How many bytes to copy
286 		 */
287 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
288 
289 		/*
290 		 * Fault and hold the page on behalf of the process.
291 		 */
292 		error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
293 		if (error != KERN_SUCCESS) {
294 			if (error == KERN_RESOURCE_SHORTAGE)
295 				error = ENOMEM;
296 			else
297 				error = EFAULT;
298 			break;
299 		}
300 
301 		/*
302 		 * Now do the i/o move.
303 		 */
304 		error = uiomove_fromphys(&m, page_offset, len, uio);
305 
306 		/* Make the I-cache coherent for breakpoints. */
307 		if (writing && error == 0) {
308 			vm_map_lock_read(map);
309 			if (vm_map_check_protection(map, pageno, pageno +
310 			    PAGE_SIZE, VM_PROT_EXECUTE))
311 				vm_sync_icache(map, uva, len);
312 			vm_map_unlock_read(map);
313 		}
314 
315 		/*
316 		 * Release the page.
317 		 */
318 		vm_page_lock(m);
319 		vm_page_unhold(m);
320 		vm_page_unlock(m);
321 
322 	} while (error == 0 && uio->uio_resid > 0);
323 
324 	return (error);
325 }
326 
327 static int
328 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
329 {
330 	struct vattr vattr;
331 	vm_map_t map;
332 	vm_map_entry_t entry;
333 	vm_object_t obj, tobj, lobj;
334 	struct vmspace *vm;
335 	struct vnode *vp;
336 	char *freepath, *fullpath;
337 	u_int pathlen;
338 	int error, index, vfslocked;
339 
340 	error = 0;
341 	obj = NULL;
342 
343 	vm = vmspace_acquire_ref(p);
344 	map = &vm->vm_map;
345 	vm_map_lock_read(map);
346 
347 	do {
348 		entry = map->header.next;
349 		index = 0;
350 		while (index < pve->pve_entry && entry != &map->header) {
351 			entry = entry->next;
352 			index++;
353 		}
354 		if (index != pve->pve_entry) {
355 			error = EINVAL;
356 			break;
357 		}
358 		while (entry != &map->header &&
359 		    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
360 			entry = entry->next;
361 			index++;
362 		}
363 		if (entry == &map->header) {
364 			error = ENOENT;
365 			break;
366 		}
367 
368 		/* We got an entry. */
369 		pve->pve_entry = index + 1;
370 		pve->pve_timestamp = map->timestamp;
371 		pve->pve_start = entry->start;
372 		pve->pve_end = entry->end - 1;
373 		pve->pve_offset = entry->offset;
374 		pve->pve_prot = entry->protection;
375 
376 		/* Backing object's path needed? */
377 		if (pve->pve_pathlen == 0)
378 			break;
379 
380 		pathlen = pve->pve_pathlen;
381 		pve->pve_pathlen = 0;
382 
383 		obj = entry->object.vm_object;
384 		if (obj != NULL)
385 			VM_OBJECT_LOCK(obj);
386 	} while (0);
387 
388 	vm_map_unlock_read(map);
389 	vmspace_free(vm);
390 
391 	pve->pve_fsid = VNOVAL;
392 	pve->pve_fileid = VNOVAL;
393 
394 	if (error == 0 && obj != NULL) {
395 		lobj = obj;
396 		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
397 			if (tobj != obj)
398 				VM_OBJECT_LOCK(tobj);
399 			if (lobj != obj)
400 				VM_OBJECT_UNLOCK(lobj);
401 			lobj = tobj;
402 			pve->pve_offset += tobj->backing_object_offset;
403 		}
404 		vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
405 		if (vp != NULL)
406 			vref(vp);
407 		if (lobj != obj)
408 			VM_OBJECT_UNLOCK(lobj);
409 		VM_OBJECT_UNLOCK(obj);
410 
411 		if (vp != NULL) {
412 			freepath = NULL;
413 			fullpath = NULL;
414 			vn_fullpath(td, vp, &fullpath, &freepath);
415 			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
416 			vn_lock(vp, LK_SHARED | LK_RETRY);
417 			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
418 				pve->pve_fileid = vattr.va_fileid;
419 				pve->pve_fsid = vattr.va_fsid;
420 			}
421 			vput(vp);
422 			VFS_UNLOCK_GIANT(vfslocked);
423 
424 			if (fullpath != NULL) {
425 				pve->pve_pathlen = strlen(fullpath) + 1;
426 				if (pve->pve_pathlen <= pathlen) {
427 					error = copyout(fullpath, pve->pve_path,
428 					    pve->pve_pathlen);
429 				} else
430 					error = ENAMETOOLONG;
431 			}
432 			if (freepath != NULL)
433 				free(freepath, M_TEMP);
434 		}
435 	}
436 
437 	return (error);
438 }
439 
440 #ifdef COMPAT_FREEBSD32
441 static int
442 ptrace_vm_entry32(struct thread *td, struct proc *p,
443     struct ptrace_vm_entry32 *pve32)
444 {
445 	struct ptrace_vm_entry pve;
446 	int error;
447 
448 	pve.pve_entry = pve32->pve_entry;
449 	pve.pve_pathlen = pve32->pve_pathlen;
450 	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
451 
452 	error = ptrace_vm_entry(td, p, &pve);
453 	if (error == 0) {
454 		pve32->pve_entry = pve.pve_entry;
455 		pve32->pve_timestamp = pve.pve_timestamp;
456 		pve32->pve_start = pve.pve_start;
457 		pve32->pve_end = pve.pve_end;
458 		pve32->pve_offset = pve.pve_offset;
459 		pve32->pve_prot = pve.pve_prot;
460 		pve32->pve_fileid = pve.pve_fileid;
461 		pve32->pve_fsid = pve.pve_fsid;
462 	}
463 
464 	pve32->pve_pathlen = pve.pve_pathlen;
465 	return (error);
466 }
467 
468 static void
469 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
470     struct ptrace_lwpinfo32 *pl32)
471 {
472 
473 	pl32->pl_lwpid = pl->pl_lwpid;
474 	pl32->pl_event = pl->pl_event;
475 	pl32->pl_flags = pl->pl_flags;
476 	pl32->pl_sigmask = pl->pl_sigmask;
477 	pl32->pl_siglist = pl->pl_siglist;
478 	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
479 	strcpy(pl32->pl_tdname, pl->pl_tdname);
480 	pl32->pl_child_pid = pl->pl_child_pid;
481 }
482 #endif /* COMPAT_FREEBSD32 */
483 
484 /*
485  * Process debugging system call.
486  */
487 #ifndef _SYS_SYSPROTO_H_
488 struct ptrace_args {
489 	int	req;
490 	pid_t	pid;
491 	caddr_t	addr;
492 	int	data;
493 };
494 #endif
495 
496 #ifdef COMPAT_FREEBSD32
497 /*
498  * This CPP subterfuge is to try and reduce the number of ifdefs in
499  * the body of the code.
500  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
501  * becomes either:
502  *   copyin(uap->addr, &r.reg, sizeof r.reg);
503  * or
504  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
505  * .. except this is done at runtime.
506  */
507 #define	COPYIN(u, k, s)		wrap32 ? \
508 	copyin(u, k ## 32, s ## 32) : \
509 	copyin(u, k, s)
510 #define	COPYOUT(k, u, s)	wrap32 ? \
511 	copyout(k ## 32, u, s ## 32) : \
512 	copyout(k, u, s)
513 #else
514 #define	COPYIN(u, k, s)		copyin(u, k, s)
515 #define	COPYOUT(k, u, s)	copyout(k, u, s)
516 #endif
517 int
518 sys_ptrace(struct thread *td, struct ptrace_args *uap)
519 {
520 	/*
521 	 * XXX this obfuscation is to reduce stack usage, but the register
522 	 * structs may be too large to put on the stack anyway.
523 	 */
524 	union {
525 		struct ptrace_io_desc piod;
526 		struct ptrace_lwpinfo pl;
527 		struct ptrace_vm_entry pve;
528 		struct dbreg dbreg;
529 		struct fpreg fpreg;
530 		struct reg reg;
531 #ifdef COMPAT_FREEBSD32
532 		struct dbreg32 dbreg32;
533 		struct fpreg32 fpreg32;
534 		struct reg32 reg32;
535 		struct ptrace_io_desc32 piod32;
536 		struct ptrace_lwpinfo32 pl32;
537 		struct ptrace_vm_entry32 pve32;
538 #endif
539 	} r;
540 	void *addr;
541 	int error = 0;
542 #ifdef COMPAT_FREEBSD32
543 	int wrap32 = 0;
544 
545 	if (SV_CURPROC_FLAG(SV_ILP32))
546 		wrap32 = 1;
547 #endif
548 	AUDIT_ARG_PID(uap->pid);
549 	AUDIT_ARG_CMD(uap->req);
550 	AUDIT_ARG_VALUE(uap->data);
551 	addr = &r;
552 	switch (uap->req) {
553 	case PT_GETREGS:
554 	case PT_GETFPREGS:
555 	case PT_GETDBREGS:
556 	case PT_LWPINFO:
557 		break;
558 	case PT_SETREGS:
559 		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
560 		break;
561 	case PT_SETFPREGS:
562 		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
563 		break;
564 	case PT_SETDBREGS:
565 		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
566 		break;
567 	case PT_IO:
568 		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
569 		break;
570 	case PT_VM_ENTRY:
571 		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
572 		break;
573 	default:
574 		addr = uap->addr;
575 		break;
576 	}
577 	if (error)
578 		return (error);
579 
580 	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
581 	if (error)
582 		return (error);
583 
584 	switch (uap->req) {
585 	case PT_VM_ENTRY:
586 		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
587 		break;
588 	case PT_IO:
589 		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
590 		break;
591 	case PT_GETREGS:
592 		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
593 		break;
594 	case PT_GETFPREGS:
595 		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
596 		break;
597 	case PT_GETDBREGS:
598 		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
599 		break;
600 	case PT_LWPINFO:
601 		error = copyout(&r.pl, uap->addr, uap->data);
602 		break;
603 	}
604 
605 	return (error);
606 }
607 #undef COPYIN
608 #undef COPYOUT
609 
610 #ifdef COMPAT_FREEBSD32
611 /*
612  *   PROC_READ(regs, td2, addr);
613  * becomes either:
614  *   proc_read_regs(td2, addr);
615  * or
616  *   proc_read_regs32(td2, addr);
617  * .. except this is done at runtime.  There is an additional
618  * complication in that PROC_WRITE disallows 32 bit consumers
619  * from writing to 64 bit address space targets.
620  */
621 #define	PROC_READ(w, t, a)	wrap32 ? \
622 	proc_read_ ## w ## 32(t, a) : \
623 	proc_read_ ## w (t, a)
624 #define	PROC_WRITE(w, t, a)	wrap32 ? \
625 	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
626 	proc_write_ ## w (t, a)
627 #else
628 #define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
629 #define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
630 #endif
631 
632 int
633 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
634 {
635 	struct iovec iov;
636 	struct uio uio;
637 	struct proc *curp, *p, *pp;
638 	struct thread *td2 = NULL, *td3;
639 	struct ptrace_io_desc *piod = NULL;
640 	struct ptrace_lwpinfo *pl;
641 	int error, write, tmp, num;
642 	int proctree_locked = 0;
643 	lwpid_t tid = 0, *buf;
644 #ifdef COMPAT_FREEBSD32
645 	int wrap32 = 0, safe = 0;
646 	struct ptrace_io_desc32 *piod32 = NULL;
647 	struct ptrace_lwpinfo32 *pl32 = NULL;
648 	struct ptrace_lwpinfo plr;
649 #endif
650 
651 	curp = td->td_proc;
652 
653 	/* Lock proctree before locking the process. */
654 	switch (req) {
655 	case PT_TRACE_ME:
656 	case PT_ATTACH:
657 	case PT_STEP:
658 	case PT_CONTINUE:
659 	case PT_TO_SCE:
660 	case PT_TO_SCX:
661 	case PT_SYSCALL:
662 	case PT_FOLLOW_FORK:
663 	case PT_DETACH:
664 		sx_xlock(&proctree_lock);
665 		proctree_locked = 1;
666 		break;
667 	default:
668 		break;
669 	}
670 
671 	write = 0;
672 	if (req == PT_TRACE_ME) {
673 		p = td->td_proc;
674 		PROC_LOCK(p);
675 	} else {
676 		if (pid <= PID_MAX) {
677 			if ((p = pfind(pid)) == NULL) {
678 				if (proctree_locked)
679 					sx_xunlock(&proctree_lock);
680 				return (ESRCH);
681 			}
682 		} else {
683 			td2 = tdfind(pid, -1);
684 			if (td2 == NULL) {
685 				if (proctree_locked)
686 					sx_xunlock(&proctree_lock);
687 				return (ESRCH);
688 			}
689 			p = td2->td_proc;
690 			tid = pid;
691 			pid = p->p_pid;
692 		}
693 	}
694 	AUDIT_ARG_PROCESS(p);
695 
696 	if ((p->p_flag & P_WEXIT) != 0) {
697 		error = ESRCH;
698 		goto fail;
699 	}
700 	if ((error = p_cansee(td, p)) != 0)
701 		goto fail;
702 
703 	if ((error = p_candebug(td, p)) != 0)
704 		goto fail;
705 
706 	/*
707 	 * System processes can't be debugged.
708 	 */
709 	if ((p->p_flag & P_SYSTEM) != 0) {
710 		error = EINVAL;
711 		goto fail;
712 	}
713 
714 	if (tid == 0) {
715 		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
716 			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
717 			td2 = p->p_xthread;
718 		} else {
719 			td2 = FIRST_THREAD_IN_PROC(p);
720 		}
721 		tid = td2->td_tid;
722 	}
723 
724 #ifdef COMPAT_FREEBSD32
725 	/*
726 	 * Test if we're a 32 bit client and what the target is.
727 	 * Set the wrap controls accordingly.
728 	 */
729 	if (SV_CURPROC_FLAG(SV_ILP32)) {
730 		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
731 			safe = 1;
732 		wrap32 = 1;
733 	}
734 #endif
735 	/*
736 	 * Permissions check
737 	 */
738 	switch (req) {
739 	case PT_TRACE_ME:
740 		/* Always legal. */
741 		break;
742 
743 	case PT_ATTACH:
744 		/* Self */
745 		if (p->p_pid == td->td_proc->p_pid) {
746 			error = EINVAL;
747 			goto fail;
748 		}
749 
750 		/* Already traced */
751 		if (p->p_flag & P_TRACED) {
752 			error = EBUSY;
753 			goto fail;
754 		}
755 
756 		/* Can't trace an ancestor if you're being traced. */
757 		if (curp->p_flag & P_TRACED) {
758 			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
759 				if (pp == p) {
760 					error = EINVAL;
761 					goto fail;
762 				}
763 			}
764 		}
765 
766 
767 		/* OK */
768 		break;
769 
770 	case PT_CLEARSTEP:
771 		/* Allow thread to clear single step for itself */
772 		if (td->td_tid == tid)
773 			break;
774 
775 		/* FALLTHROUGH */
776 	default:
777 		/* not being traced... */
778 		if ((p->p_flag & P_TRACED) == 0) {
779 			error = EPERM;
780 			goto fail;
781 		}
782 
783 		/* not being traced by YOU */
784 		if (p->p_pptr != td->td_proc) {
785 			error = EBUSY;
786 			goto fail;
787 		}
788 
789 		/* not currently stopped */
790 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
791 		    p->p_suspcount != p->p_numthreads  ||
792 		    (p->p_flag & P_WAITED) == 0) {
793 			error = EBUSY;
794 			goto fail;
795 		}
796 
797 		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
798 			static int count = 0;
799 			if (count++ == 0)
800 				printf("P_STOPPED_TRACE not set.\n");
801 		}
802 
803 		/* OK */
804 		break;
805 	}
806 
807 	/* Keep this process around until we finish this request. */
808 	_PHOLD(p);
809 
810 #ifdef FIX_SSTEP
811 	/*
812 	 * Single step fixup ala procfs
813 	 */
814 	FIX_SSTEP(td2);
815 #endif
816 
817 	/*
818 	 * Actually do the requests
819 	 */
820 
821 	td->td_retval[0] = 0;
822 
823 	switch (req) {
824 	case PT_TRACE_ME:
825 		/* set my trace flag and "owner" so it can read/write me */
826 		p->p_flag |= P_TRACED;
827 		p->p_oppid = p->p_pptr->p_pid;
828 		break;
829 
830 	case PT_ATTACH:
831 		/* security check done above */
832 		/*
833 		 * It would be nice if the tracing relationship was separate
834 		 * from the parent relationship but that would require
835 		 * another set of links in the proc struct or for "wait"
836 		 * to scan the entire proc table.  To make life easier,
837 		 * we just re-parent the process we're trying to trace.
838 		 * The old parent is remembered so we can put things back
839 		 * on a "detach".
840 		 */
841 		p->p_flag |= P_TRACED;
842 		p->p_oppid = p->p_pptr->p_pid;
843 		if (p->p_pptr != td->td_proc) {
844 			proc_reparent(p, td->td_proc);
845 		}
846 		data = SIGSTOP;
847 		goto sendsig;	/* in PT_CONTINUE below */
848 
849 	case PT_CLEARSTEP:
850 		error = ptrace_clear_single_step(td2);
851 		break;
852 
853 	case PT_SETSTEP:
854 		error = ptrace_single_step(td2);
855 		break;
856 
857 	case PT_SUSPEND:
858 		td2->td_dbgflags |= TDB_SUSPEND;
859 		thread_lock(td2);
860 		td2->td_flags |= TDF_NEEDSUSPCHK;
861 		thread_unlock(td2);
862 		break;
863 
864 	case PT_RESUME:
865 		td2->td_dbgflags &= ~TDB_SUSPEND;
866 		break;
867 
868 	case PT_FOLLOW_FORK:
869 		if (data)
870 			p->p_flag |= P_FOLLOWFORK;
871 		else
872 			p->p_flag &= ~P_FOLLOWFORK;
873 		break;
874 
875 	case PT_STEP:
876 	case PT_CONTINUE:
877 	case PT_TO_SCE:
878 	case PT_TO_SCX:
879 	case PT_SYSCALL:
880 	case PT_DETACH:
881 		/* Zero means do not send any signal */
882 		if (data < 0 || data > _SIG_MAXSIG) {
883 			error = EINVAL;
884 			break;
885 		}
886 
887 		switch (req) {
888 		case PT_STEP:
889 			error = ptrace_single_step(td2);
890 			if (error)
891 				goto out;
892 			break;
893 		case PT_CONTINUE:
894 		case PT_TO_SCE:
895 		case PT_TO_SCX:
896 		case PT_SYSCALL:
897 			if (addr != (void *)1) {
898 				error = ptrace_set_pc(td2,
899 				    (u_long)(uintfptr_t)addr);
900 				if (error)
901 					goto out;
902 			}
903 			switch (req) {
904 			case PT_TO_SCE:
905 				p->p_stops |= S_PT_SCE;
906 				break;
907 			case PT_TO_SCX:
908 				p->p_stops |= S_PT_SCX;
909 				break;
910 			case PT_SYSCALL:
911 				p->p_stops |= S_PT_SCE | S_PT_SCX;
912 				break;
913 			}
914 			break;
915 		case PT_DETACH:
916 			/* reset process parent */
917 			if (p->p_oppid != p->p_pptr->p_pid) {
918 				struct proc *pp;
919 
920 				PROC_LOCK(p->p_pptr);
921 				sigqueue_take(p->p_ksi);
922 				PROC_UNLOCK(p->p_pptr);
923 
924 				PROC_UNLOCK(p);
925 				pp = pfind(p->p_oppid);
926 				if (pp == NULL)
927 					pp = initproc;
928 				else
929 					PROC_UNLOCK(pp);
930 				PROC_LOCK(p);
931 				proc_reparent(p, pp);
932 				if (pp == initproc)
933 					p->p_sigparent = SIGCHLD;
934 			}
935 			p->p_oppid = 0;
936 			p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
937 
938 			/* should we send SIGCHLD? */
939 			/* childproc_continued(p); */
940 			break;
941 		}
942 
943 	sendsig:
944 		if (proctree_locked) {
945 			sx_xunlock(&proctree_lock);
946 			proctree_locked = 0;
947 		}
948 		p->p_xstat = data;
949 		p->p_xthread = NULL;
950 		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
951 			/* deliver or queue signal */
952 			td2->td_dbgflags &= ~TDB_XSIG;
953 			td2->td_xsig = data;
954 
955 			if (req == PT_DETACH) {
956 				FOREACH_THREAD_IN_PROC(p, td3)
957 					td3->td_dbgflags &= ~TDB_SUSPEND;
958 			}
959 			/*
960 			 * unsuspend all threads, to not let a thread run,
961 			 * you should use PT_SUSPEND to suspend it before
962 			 * continuing process.
963 			 */
964 			PROC_SLOCK(p);
965 			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
966 			thread_unsuspend(p);
967 			PROC_SUNLOCK(p);
968 			if (req == PT_ATTACH)
969 				kern_psignal(p, data);
970 		} else {
971 			if (data)
972 				kern_psignal(p, data);
973 		}
974 		break;
975 
976 	case PT_WRITE_I:
977 	case PT_WRITE_D:
978 		td2->td_dbgflags |= TDB_USERWR;
979 		write = 1;
980 		/* FALLTHROUGH */
981 	case PT_READ_I:
982 	case PT_READ_D:
983 		PROC_UNLOCK(p);
984 		tmp = 0;
985 		/* write = 0 set above */
986 		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
987 		iov.iov_len = sizeof(int);
988 		uio.uio_iov = &iov;
989 		uio.uio_iovcnt = 1;
990 		uio.uio_offset = (off_t)(uintptr_t)addr;
991 		uio.uio_resid = sizeof(int);
992 		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
993 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
994 		uio.uio_td = td;
995 		error = proc_rwmem(p, &uio);
996 		if (uio.uio_resid != 0) {
997 			/*
998 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
999 			 * so I think write() can bogusly return 0.
1000 			 * XXX what happens for short writes?  We don't want
1001 			 * to write partial data.
1002 			 * XXX proc_rwmem() returns EPERM for other invalid
1003 			 * addresses.  Convert this to EINVAL.  Does this
1004 			 * clobber returns of EPERM for other reasons?
1005 			 */
1006 			if (error == 0 || error == ENOSPC || error == EPERM)
1007 				error = EINVAL;	/* EOF */
1008 		}
1009 		if (!write)
1010 			td->td_retval[0] = tmp;
1011 		PROC_LOCK(p);
1012 		break;
1013 
1014 	case PT_IO:
1015 #ifdef COMPAT_FREEBSD32
1016 		if (wrap32) {
1017 			piod32 = addr;
1018 			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1019 			iov.iov_len = piod32->piod_len;
1020 			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1021 			uio.uio_resid = piod32->piod_len;
1022 		} else
1023 #endif
1024 		{
1025 			piod = addr;
1026 			iov.iov_base = piod->piod_addr;
1027 			iov.iov_len = piod->piod_len;
1028 			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1029 			uio.uio_resid = piod->piod_len;
1030 		}
1031 		uio.uio_iov = &iov;
1032 		uio.uio_iovcnt = 1;
1033 		uio.uio_segflg = UIO_USERSPACE;
1034 		uio.uio_td = td;
1035 #ifdef COMPAT_FREEBSD32
1036 		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1037 #else
1038 		tmp = piod->piod_op;
1039 #endif
1040 		switch (tmp) {
1041 		case PIOD_READ_D:
1042 		case PIOD_READ_I:
1043 			uio.uio_rw = UIO_READ;
1044 			break;
1045 		case PIOD_WRITE_D:
1046 		case PIOD_WRITE_I:
1047 			td2->td_dbgflags |= TDB_USERWR;
1048 			uio.uio_rw = UIO_WRITE;
1049 			break;
1050 		default:
1051 			error = EINVAL;
1052 			goto out;
1053 		}
1054 		PROC_UNLOCK(p);
1055 		error = proc_rwmem(p, &uio);
1056 #ifdef COMPAT_FREEBSD32
1057 		if (wrap32)
1058 			piod32->piod_len -= uio.uio_resid;
1059 		else
1060 #endif
1061 			piod->piod_len -= uio.uio_resid;
1062 		PROC_LOCK(p);
1063 		break;
1064 
1065 	case PT_KILL:
1066 		data = SIGKILL;
1067 		goto sendsig;	/* in PT_CONTINUE above */
1068 
1069 	case PT_SETREGS:
1070 		td2->td_dbgflags |= TDB_USERWR;
1071 		error = PROC_WRITE(regs, td2, addr);
1072 		break;
1073 
1074 	case PT_GETREGS:
1075 		error = PROC_READ(regs, td2, addr);
1076 		break;
1077 
1078 	case PT_SETFPREGS:
1079 		td2->td_dbgflags |= TDB_USERWR;
1080 		error = PROC_WRITE(fpregs, td2, addr);
1081 		break;
1082 
1083 	case PT_GETFPREGS:
1084 		error = PROC_READ(fpregs, td2, addr);
1085 		break;
1086 
1087 	case PT_SETDBREGS:
1088 		td2->td_dbgflags |= TDB_USERWR;
1089 		error = PROC_WRITE(dbregs, td2, addr);
1090 		break;
1091 
1092 	case PT_GETDBREGS:
1093 		error = PROC_READ(dbregs, td2, addr);
1094 		break;
1095 
1096 	case PT_LWPINFO:
1097 		if (data <= 0 ||
1098 #ifdef COMPAT_FREEBSD32
1099 		    (!wrap32 && data > sizeof(*pl)) ||
1100 		    (wrap32 && data > sizeof(*pl32))) {
1101 #else
1102 		    data > sizeof(*pl)) {
1103 #endif
1104 			error = EINVAL;
1105 			break;
1106 		}
1107 #ifdef COMPAT_FREEBSD32
1108 		if (wrap32) {
1109 			pl = &plr;
1110 			pl32 = addr;
1111 		} else
1112 #endif
1113 		pl = addr;
1114 		pl->pl_lwpid = td2->td_tid;
1115 		pl->pl_event = PL_EVENT_NONE;
1116 		pl->pl_flags = 0;
1117 		if (td2->td_dbgflags & TDB_XSIG) {
1118 			pl->pl_event = PL_EVENT_SIGNAL;
1119 			if (td2->td_dbgksi.ksi_signo != 0 &&
1120 #ifdef COMPAT_FREEBSD32
1121 			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1122 			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1123 			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1124 			    pl_siginfo) + sizeof(struct siginfo32)))
1125 #else
1126 			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1127 			    + sizeof(pl->pl_siginfo)
1128 #endif
1129 			){
1130 				pl->pl_flags |= PL_FLAG_SI;
1131 				pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1132 			}
1133 		}
1134 		if ((pl->pl_flags & PL_FLAG_SI) == 0)
1135 			bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1136 		if (td2->td_dbgflags & TDB_SCE)
1137 			pl->pl_flags |= PL_FLAG_SCE;
1138 		else if (td2->td_dbgflags & TDB_SCX)
1139 			pl->pl_flags |= PL_FLAG_SCX;
1140 		if (td2->td_dbgflags & TDB_EXEC)
1141 			pl->pl_flags |= PL_FLAG_EXEC;
1142 		if (td2->td_dbgflags & TDB_FORK) {
1143 			pl->pl_flags |= PL_FLAG_FORKED;
1144 			pl->pl_child_pid = td2->td_dbg_forked;
1145 		}
1146 		if (td2->td_dbgflags & TDB_CHILD)
1147 			pl->pl_flags |= PL_FLAG_CHILD;
1148 		pl->pl_sigmask = td2->td_sigmask;
1149 		pl->pl_siglist = td2->td_siglist;
1150 		strcpy(pl->pl_tdname, td2->td_name);
1151 #ifdef COMPAT_FREEBSD32
1152 		if (wrap32)
1153 			ptrace_lwpinfo_to32(pl, pl32);
1154 #endif
1155 		break;
1156 
1157 	case PT_GETNUMLWPS:
1158 		td->td_retval[0] = p->p_numthreads;
1159 		break;
1160 
1161 	case PT_GETLWPLIST:
1162 		if (data <= 0) {
1163 			error = EINVAL;
1164 			break;
1165 		}
1166 		num = imin(p->p_numthreads, data);
1167 		PROC_UNLOCK(p);
1168 		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1169 		tmp = 0;
1170 		PROC_LOCK(p);
1171 		FOREACH_THREAD_IN_PROC(p, td2) {
1172 			if (tmp >= num)
1173 				break;
1174 			buf[tmp++] = td2->td_tid;
1175 		}
1176 		PROC_UNLOCK(p);
1177 		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1178 		free(buf, M_TEMP);
1179 		if (!error)
1180 			td->td_retval[0] = tmp;
1181 		PROC_LOCK(p);
1182 		break;
1183 
1184 	case PT_VM_TIMESTAMP:
1185 		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1186 		break;
1187 
1188 	case PT_VM_ENTRY:
1189 		PROC_UNLOCK(p);
1190 #ifdef COMPAT_FREEBSD32
1191 		if (wrap32)
1192 			error = ptrace_vm_entry32(td, p, addr);
1193 		else
1194 #endif
1195 		error = ptrace_vm_entry(td, p, addr);
1196 		PROC_LOCK(p);
1197 		break;
1198 
1199 	default:
1200 #ifdef __HAVE_PTRACE_MACHDEP
1201 		if (req >= PT_FIRSTMACH) {
1202 			PROC_UNLOCK(p);
1203 			error = cpu_ptrace(td2, req, addr, data);
1204 			PROC_LOCK(p);
1205 		} else
1206 #endif
1207 			/* Unknown request. */
1208 			error = EINVAL;
1209 		break;
1210 	}
1211 
1212 out:
1213 	/* Drop our hold on this process now that the request has completed. */
1214 	_PRELE(p);
1215 fail:
1216 	PROC_UNLOCK(p);
1217 	if (proctree_locked)
1218 		sx_xunlock(&proctree_lock);
1219 	return (error);
1220 }
1221 #undef PROC_READ
1222 #undef PROC_WRITE
1223 
1224 /*
1225  * Stop a process because of a debugging event;
1226  * stay stopped until p->p_step is cleared
1227  * (cleared by PIOCCONT in procfs).
1228  */
1229 void
1230 stopevent(struct proc *p, unsigned int event, unsigned int val)
1231 {
1232 
1233 	PROC_LOCK_ASSERT(p, MA_OWNED);
1234 	p->p_step = 1;
1235 	do {
1236 		p->p_xstat = val;
1237 		p->p_xthread = NULL;
1238 		p->p_stype = event;	/* Which event caused the stop? */
1239 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1240 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1241 	} while (p->p_step);
1242 }
1243