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/priv.h>
45#include <sys/proc.h>
46#include <sys/procctl.h>
47#include <sys/vnode.h>
48#include <sys/ptrace.h>
49#include <sys/sx.h>
50#include <sys/malloc.h>
51#include <sys/signalvar.h>
52
53#include <machine/reg.h>
54
55#include <security/audit/audit.h>
56
57#include <vm/vm.h>
58#include <vm/pmap.h>
59#include <vm/vm_extern.h>
60#include <vm/vm_map.h>
61#include <vm/vm_kern.h>
62#include <vm/vm_object.h>
63#include <vm/vm_page.h>
64#include <vm/vm_pager.h>
65#include <vm/vm_param.h>
66
67#ifdef COMPAT_FREEBSD32
68#include <sys/procfs.h>
69#include <compat/freebsd32/freebsd32_signal.h>
70
71struct ptrace_io_desc32 {
72	int		piod_op;
73	uint32_t	piod_offs;
74	uint32_t	piod_addr;
75	uint32_t	piod_len;
76};
77
78struct ptrace_vm_entry32 {
79	int		pve_entry;
80	int		pve_timestamp;
81	uint32_t	pve_start;
82	uint32_t	pve_end;
83	uint32_t	pve_offset;
84	u_int		pve_prot;
85	u_int		pve_pathlen;
86	int32_t		pve_fileid;
87	u_int		pve_fsid;
88	uint32_t	pve_path;
89};
90
91struct ptrace_lwpinfo32 {
92	lwpid_t	pl_lwpid;	/* LWP described. */
93	int	pl_event;	/* Event that stopped the LWP. */
94	int	pl_flags;	/* LWP flags. */
95	sigset_t	pl_sigmask;	/* LWP signal mask */
96	sigset_t	pl_siglist;	/* LWP pending signal */
97	struct siginfo32 pl_siginfo;	/* siginfo for signal */
98	char	pl_tdname[MAXCOMLEN + 1];	/* LWP name. */
99	int	pl_child_pid;		/* New child pid */
100};
101
102#endif
103
104/*
105 * Functions implemented using PROC_ACTION():
106 *
107 * proc_read_regs(proc, regs)
108 *	Get the current user-visible register set from the process
109 *	and copy it into the regs structure (<machine/reg.h>).
110 *	The process is stopped at the time read_regs is called.
111 *
112 * proc_write_regs(proc, regs)
113 *	Update the current register set from the passed in regs
114 *	structure.  Take care to avoid clobbering special CPU
115 *	registers or privileged bits in the PSL.
116 *	Depending on the architecture this may have fix-up work to do,
117 *	especially if the IAR or PCW are modified.
118 *	The process is stopped at the time write_regs is called.
119 *
120 * proc_read_fpregs, proc_write_fpregs
121 *	deal with the floating point register set, otherwise as above.
122 *
123 * proc_read_dbregs, proc_write_dbregs
124 *	deal with the processor debug register set, otherwise as above.
125 *
126 * proc_sstep(proc)
127 *	Arrange for the process to trap after executing a single instruction.
128 */
129
130#define	PROC_ACTION(action) do {					\
131	int error;							\
132									\
133	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
134	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
135		error = EIO;						\
136	else								\
137		error = (action);					\
138	return (error);							\
139} while(0)
140
141int
142proc_read_regs(struct thread *td, struct reg *regs)
143{
144
145	PROC_ACTION(fill_regs(td, regs));
146}
147
148int
149proc_write_regs(struct thread *td, struct reg *regs)
150{
151
152	PROC_ACTION(set_regs(td, regs));
153}
154
155int
156proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
157{
158
159	PROC_ACTION(fill_dbregs(td, dbregs));
160}
161
162int
163proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
164{
165
166	PROC_ACTION(set_dbregs(td, dbregs));
167}
168
169/*
170 * Ptrace doesn't support fpregs at all, and there are no security holes
171 * or translations for fpregs, so we can just copy them.
172 */
173int
174proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
175{
176
177	PROC_ACTION(fill_fpregs(td, fpregs));
178}
179
180int
181proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
182{
183
184	PROC_ACTION(set_fpregs(td, fpregs));
185}
186
187#ifdef COMPAT_FREEBSD32
188/* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
189int
190proc_read_regs32(struct thread *td, struct reg32 *regs32)
191{
192
193	PROC_ACTION(fill_regs32(td, regs32));
194}
195
196int
197proc_write_regs32(struct thread *td, struct reg32 *regs32)
198{
199
200	PROC_ACTION(set_regs32(td, regs32));
201}
202
203int
204proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
205{
206
207	PROC_ACTION(fill_dbregs32(td, dbregs32));
208}
209
210int
211proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
212{
213
214	PROC_ACTION(set_dbregs32(td, dbregs32));
215}
216
217int
218proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
219{
220
221	PROC_ACTION(fill_fpregs32(td, fpregs32));
222}
223
224int
225proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
226{
227
228	PROC_ACTION(set_fpregs32(td, fpregs32));
229}
230#endif
231
232int
233proc_sstep(struct thread *td)
234{
235
236	PROC_ACTION(ptrace_single_step(td));
237}
238
239int
240proc_rwmem(struct proc *p, struct uio *uio)
241{
242	vm_map_t map;
243	vm_offset_t pageno;		/* page number */
244	vm_prot_t reqprot;
245	int error, fault_flags, page_offset, writing;
246
247	/*
248	 * Assert that someone has locked this vmspace.  (Should be
249	 * curthread but we can't assert that.)  This keeps the process
250	 * from exiting out from under us until this operation completes.
251	 */
252	KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
253	    p, p->p_pid));
254
255	/*
256	 * The map we want...
257	 */
258	map = &p->p_vmspace->vm_map;
259
260	/*
261	 * If we are writing, then we request vm_fault() to create a private
262	 * copy of each page.  Since these copies will not be writeable by the
263	 * process, we must explicity request that they be dirtied.
264	 */
265	writing = uio->uio_rw == UIO_WRITE;
266	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
267	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
268
269	/*
270	 * Only map in one page at a time.  We don't have to, but it
271	 * makes things easier.  This way is trivial - right?
272	 */
273	do {
274		vm_offset_t uva;
275		u_int len;
276		vm_page_t m;
277
278		uva = (vm_offset_t)uio->uio_offset;
279
280		/*
281		 * Get the page number of this segment.
282		 */
283		pageno = trunc_page(uva);
284		page_offset = uva - pageno;
285
286		/*
287		 * How many bytes to copy
288		 */
289		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
290
291		/*
292		 * Fault and hold the page on behalf of the process.
293		 */
294		error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
295		if (error != KERN_SUCCESS) {
296			if (error == KERN_RESOURCE_SHORTAGE)
297				error = ENOMEM;
298			else
299				error = EFAULT;
300			break;
301		}
302
303		/*
304		 * Now do the i/o move.
305		 */
306		error = uiomove_fromphys(&m, page_offset, len, uio);
307
308		/* Make the I-cache coherent for breakpoints. */
309		if (writing && error == 0) {
310			vm_map_lock_read(map);
311			if (vm_map_check_protection(map, pageno, pageno +
312			    PAGE_SIZE, VM_PROT_EXECUTE))
313				vm_sync_icache(map, uva, len);
314			vm_map_unlock_read(map);
315		}
316
317		/*
318		 * Release the page.
319		 */
320		vm_page_lock(m);
321		vm_page_unhold(m);
322		vm_page_unlock(m);
323
324	} while (error == 0 && uio->uio_resid > 0);
325
326	return (error);
327}
328
329static int
330ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
331{
332	struct vattr vattr;
333	vm_map_t map;
334	vm_map_entry_t entry;
335	vm_object_t obj, tobj, lobj;
336	struct vmspace *vm;
337	struct vnode *vp;
338	char *freepath, *fullpath;
339	u_int pathlen;
340	int error, index, vfslocked;
341
342	error = 0;
343	obj = NULL;
344
345	vm = vmspace_acquire_ref(p);
346	map = &vm->vm_map;
347	vm_map_lock_read(map);
348
349	do {
350		entry = map->header.next;
351		index = 0;
352		while (index < pve->pve_entry && entry != &map->header) {
353			entry = entry->next;
354			index++;
355		}
356		if (index != pve->pve_entry) {
357			error = EINVAL;
358			break;
359		}
360		while (entry != &map->header &&
361		    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
362			entry = entry->next;
363			index++;
364		}
365		if (entry == &map->header) {
366			error = ENOENT;
367			break;
368		}
369
370		/* We got an entry. */
371		pve->pve_entry = index + 1;
372		pve->pve_timestamp = map->timestamp;
373		pve->pve_start = entry->start;
374		pve->pve_end = entry->end - 1;
375		pve->pve_offset = entry->offset;
376		pve->pve_prot = entry->protection;
377
378		/* Backing object's path needed? */
379		if (pve->pve_pathlen == 0)
380			break;
381
382		pathlen = pve->pve_pathlen;
383		pve->pve_pathlen = 0;
384
385		obj = entry->object.vm_object;
386		if (obj != NULL)
387			VM_OBJECT_LOCK(obj);
388	} while (0);
389
390	vm_map_unlock_read(map);
391	vmspace_free(vm);
392
393	pve->pve_fsid = VNOVAL;
394	pve->pve_fileid = VNOVAL;
395
396	if (error == 0 && obj != NULL) {
397		lobj = obj;
398		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
399			if (tobj != obj)
400				VM_OBJECT_LOCK(tobj);
401			if (lobj != obj)
402				VM_OBJECT_UNLOCK(lobj);
403			lobj = tobj;
404			pve->pve_offset += tobj->backing_object_offset;
405		}
406		vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
407		if (vp != NULL)
408			vref(vp);
409		if (lobj != obj)
410			VM_OBJECT_UNLOCK(lobj);
411		VM_OBJECT_UNLOCK(obj);
412
413		if (vp != NULL) {
414			freepath = NULL;
415			fullpath = NULL;
416			vn_fullpath(td, vp, &fullpath, &freepath);
417			vfslocked = VFS_LOCK_GIANT(vp->v_mount);
418			vn_lock(vp, LK_SHARED | LK_RETRY);
419			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
420				pve->pve_fileid = vattr.va_fileid;
421				pve->pve_fsid = vattr.va_fsid;
422			}
423			vput(vp);
424			VFS_UNLOCK_GIANT(vfslocked);
425
426			if (fullpath != NULL) {
427				pve->pve_pathlen = strlen(fullpath) + 1;
428				if (pve->pve_pathlen <= pathlen) {
429					error = copyout(fullpath, pve->pve_path,
430					    pve->pve_pathlen);
431				} else
432					error = ENAMETOOLONG;
433			}
434			if (freepath != NULL)
435				free(freepath, M_TEMP);
436		}
437	}
438
439	return (error);
440}
441
442#ifdef COMPAT_FREEBSD32
443static int
444ptrace_vm_entry32(struct thread *td, struct proc *p,
445    struct ptrace_vm_entry32 *pve32)
446{
447	struct ptrace_vm_entry pve;
448	int error;
449
450	pve.pve_entry = pve32->pve_entry;
451	pve.pve_pathlen = pve32->pve_pathlen;
452	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
453
454	error = ptrace_vm_entry(td, p, &pve);
455	if (error == 0) {
456		pve32->pve_entry = pve.pve_entry;
457		pve32->pve_timestamp = pve.pve_timestamp;
458		pve32->pve_start = pve.pve_start;
459		pve32->pve_end = pve.pve_end;
460		pve32->pve_offset = pve.pve_offset;
461		pve32->pve_prot = pve.pve_prot;
462		pve32->pve_fileid = pve.pve_fileid;
463		pve32->pve_fsid = pve.pve_fsid;
464	}
465
466	pve32->pve_pathlen = pve.pve_pathlen;
467	return (error);
468}
469
470static void
471ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
472    struct ptrace_lwpinfo32 *pl32)
473{
474
475	pl32->pl_lwpid = pl->pl_lwpid;
476	pl32->pl_event = pl->pl_event;
477	pl32->pl_flags = pl->pl_flags;
478	pl32->pl_sigmask = pl->pl_sigmask;
479	pl32->pl_siglist = pl->pl_siglist;
480	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
481	strcpy(pl32->pl_tdname, pl->pl_tdname);
482	pl32->pl_child_pid = pl->pl_child_pid;
483}
484#endif /* COMPAT_FREEBSD32 */
485
486/*
487 * Process debugging system call.
488 */
489#ifndef _SYS_SYSPROTO_H_
490struct ptrace_args {
491	int	req;
492	pid_t	pid;
493	caddr_t	addr;
494	int	data;
495};
496#endif
497
498#ifdef COMPAT_FREEBSD32
499/*
500 * This CPP subterfuge is to try and reduce the number of ifdefs in
501 * the body of the code.
502 *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
503 * becomes either:
504 *   copyin(uap->addr, &r.reg, sizeof r.reg);
505 * or
506 *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
507 * .. except this is done at runtime.
508 */
509#define	COPYIN(u, k, s)		wrap32 ? \
510	copyin(u, k ## 32, s ## 32) : \
511	copyin(u, k, s)
512#define	COPYOUT(k, u, s)	wrap32 ? \
513	copyout(k ## 32, u, s ## 32) : \
514	copyout(k, u, s)
515#else
516#define	COPYIN(u, k, s)		copyin(u, k, s)
517#define	COPYOUT(k, u, s)	copyout(k, u, s)
518#endif
519int
520sys_ptrace(struct thread *td, struct ptrace_args *uap)
521{
522	/*
523	 * XXX this obfuscation is to reduce stack usage, but the register
524	 * structs may be too large to put on the stack anyway.
525	 */
526	union {
527		struct ptrace_io_desc piod;
528		struct ptrace_lwpinfo pl;
529		struct ptrace_vm_entry pve;
530		struct dbreg dbreg;
531		struct fpreg fpreg;
532		struct reg reg;
533#ifdef COMPAT_FREEBSD32
534		struct dbreg32 dbreg32;
535		struct fpreg32 fpreg32;
536		struct reg32 reg32;
537		struct ptrace_io_desc32 piod32;
538		struct ptrace_lwpinfo32 pl32;
539		struct ptrace_vm_entry32 pve32;
540#endif
541	} r;
542	void *addr;
543	int error = 0;
544#ifdef COMPAT_FREEBSD32
545	int wrap32 = 0;
546
547	if (SV_CURPROC_FLAG(SV_ILP32))
548		wrap32 = 1;
549#endif
550	AUDIT_ARG_PID(uap->pid);
551	AUDIT_ARG_CMD(uap->req);
552	AUDIT_ARG_VALUE(uap->data);
553	addr = &r;
554	switch (uap->req) {
555	case PT_GETREGS:
556	case PT_GETFPREGS:
557	case PT_GETDBREGS:
558	case PT_LWPINFO:
559		break;
560	case PT_SETREGS:
561		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
562		break;
563	case PT_SETFPREGS:
564		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
565		break;
566	case PT_SETDBREGS:
567		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
568		break;
569	case PT_IO:
570		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
571		break;
572	case PT_VM_ENTRY:
573		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
574		break;
575	default:
576		addr = uap->addr;
577		break;
578	}
579	if (error)
580		return (error);
581
582	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
583	if (error)
584		return (error);
585
586	switch (uap->req) {
587	case PT_VM_ENTRY:
588		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
589		break;
590	case PT_IO:
591		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
592		break;
593	case PT_GETREGS:
594		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
595		break;
596	case PT_GETFPREGS:
597		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
598		break;
599	case PT_GETDBREGS:
600		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
601		break;
602	case PT_LWPINFO:
603		error = copyout(&r.pl, uap->addr, uap->data);
604		break;
605	}
606
607	return (error);
608}
609#undef COPYIN
610#undef COPYOUT
611
612#ifdef COMPAT_FREEBSD32
613/*
614 *   PROC_READ(regs, td2, addr);
615 * becomes either:
616 *   proc_read_regs(td2, addr);
617 * or
618 *   proc_read_regs32(td2, addr);
619 * .. except this is done at runtime.  There is an additional
620 * complication in that PROC_WRITE disallows 32 bit consumers
621 * from writing to 64 bit address space targets.
622 */
623#define	PROC_READ(w, t, a)	wrap32 ? \
624	proc_read_ ## w ## 32(t, a) : \
625	proc_read_ ## w (t, a)
626#define	PROC_WRITE(w, t, a)	wrap32 ? \
627	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
628	proc_write_ ## w (t, a)
629#else
630#define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
631#define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
632#endif
633
634int
635kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
636{
637	struct iovec iov;
638	struct uio uio;
639	struct proc *curp, *p, *pp;
640	struct thread *td2 = NULL;
641	struct ptrace_io_desc *piod = NULL;
642	struct ptrace_lwpinfo *pl;
643	int error, write, tmp, num;
644	int proctree_locked = 0;
645	lwpid_t tid = 0, *buf;
646#ifdef COMPAT_FREEBSD32
647	int wrap32 = 0, safe = 0;
648	struct ptrace_io_desc32 *piod32 = NULL;
649	struct ptrace_lwpinfo32 *pl32 = NULL;
650	struct ptrace_lwpinfo plr;
651#endif
652
653	curp = td->td_proc;
654
655	/* Lock proctree before locking the process. */
656	switch (req) {
657	case PT_TRACE_ME:
658	case PT_ATTACH:
659	case PT_STEP:
660	case PT_CONTINUE:
661	case PT_TO_SCE:
662	case PT_TO_SCX:
663	case PT_SYSCALL:
664	case PT_FOLLOW_FORK:
665	case PT_DETACH:
666		sx_xlock(&proctree_lock);
667		proctree_locked = 1;
668		break;
669	default:
670		break;
671	}
672
673	write = 0;
674	if (req == PT_TRACE_ME) {
675		p = td->td_proc;
676		PROC_LOCK(p);
677	} else {
678		if (pid <= PID_MAX) {
679			if ((p = pfind(pid)) == NULL) {
680				if (proctree_locked)
681					sx_xunlock(&proctree_lock);
682				return (ESRCH);
683			}
684		} else {
685			td2 = tdfind(pid, -1);
686			if (td2 == NULL) {
687				if (proctree_locked)
688					sx_xunlock(&proctree_lock);
689				return (ESRCH);
690			}
691			p = td2->td_proc;
692			tid = pid;
693			pid = p->p_pid;
694		}
695	}
696	AUDIT_ARG_PROCESS(p);
697
698	if ((p->p_flag & P_WEXIT) != 0) {
699		error = ESRCH;
700		goto fail;
701	}
702	if ((error = p_cansee(td, p)) != 0)
703		goto fail;
704
705	if ((error = p_candebug(td, p)) != 0)
706		goto fail;
707
708	/*
709	 * System processes can't be debugged.
710	 */
711	if ((p->p_flag & P_SYSTEM) != 0) {
712		error = EINVAL;
713		goto fail;
714	}
715
716	if (tid == 0) {
717		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
718			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
719			td2 = p->p_xthread;
720		} else {
721			td2 = FIRST_THREAD_IN_PROC(p);
722		}
723		tid = td2->td_tid;
724	}
725
726#ifdef COMPAT_FREEBSD32
727	/*
728	 * Test if we're a 32 bit client and what the target is.
729	 * Set the wrap controls accordingly.
730	 */
731	if (SV_CURPROC_FLAG(SV_ILP32)) {
732		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
733			safe = 1;
734		wrap32 = 1;
735	}
736#endif
737	/*
738	 * Permissions check
739	 */
740	switch (req) {
741	case PT_TRACE_ME:
742		/* Always legal. */
743		break;
744
745	case PT_ATTACH:
746		/* Self */
747		if (p->p_pid == td->td_proc->p_pid) {
748			error = EINVAL;
749			goto fail;
750		}
751
752		/* Already traced */
753		if (p->p_flag & P_TRACED) {
754			error = EBUSY;
755			goto fail;
756		}
757
758		/* Can't trace an ancestor if you're being traced. */
759		if (curp->p_flag & P_TRACED) {
760			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
761				if (pp == p) {
762					error = EINVAL;
763					goto fail;
764				}
765			}
766		}
767
768
769		/* OK */
770		break;
771
772	case PT_CLEARSTEP:
773		/* Allow thread to clear single step for itself */
774		if (td->td_tid == tid)
775			break;
776
777		/* FALLTHROUGH */
778	default:
779		/* not being traced... */
780		if ((p->p_flag & P_TRACED) == 0) {
781			error = EPERM;
782			goto fail;
783		}
784
785		/* not being traced by YOU */
786		if (p->p_pptr != td->td_proc) {
787			error = EBUSY;
788			goto fail;
789		}
790
791		/* not currently stopped */
792		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
793		    p->p_suspcount != p->p_numthreads  ||
794		    (p->p_flag & P_WAITED) == 0) {
795			error = EBUSY;
796			goto fail;
797		}
798
799		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
800			static int count = 0;
801			if (count++ == 0)
802				printf("P_STOPPED_TRACE not set.\n");
803		}
804
805		/* OK */
806		break;
807	}
808
809	/* Keep this process around until we finish this request. */
810	_PHOLD(p);
811
812#ifdef FIX_SSTEP
813	/*
814	 * Single step fixup ala procfs
815	 */
816	FIX_SSTEP(td2);
817#endif
818
819	/*
820	 * Actually do the requests
821	 */
822
823	td->td_retval[0] = 0;
824
825	switch (req) {
826	case PT_TRACE_ME:
827		/* set my trace flag and "owner" so it can read/write me */
828		p->p_flag |= P_TRACED;
829		if (p->p_flag & P_PPWAIT)
830			p->p_flag |= P_PPTRACE;
831		p->p_oppid = p->p_pptr->p_pid;
832		break;
833
834	case PT_ATTACH:
835		/* security check done above */
836		/*
837		 * It would be nice if the tracing relationship was separate
838		 * from the parent relationship but that would require
839		 * another set of links in the proc struct or for "wait"
840		 * to scan the entire proc table.  To make life easier,
841		 * we just re-parent the process we're trying to trace.
842		 * The old parent is remembered so we can put things back
843		 * on a "detach".
844		 */
845		p->p_flag |= P_TRACED;
846		p->p_oppid = p->p_pptr->p_pid;
847		if (p->p_pptr != td->td_proc) {
848			proc_reparent(p, td->td_proc);
849		}
850		data = SIGSTOP;
851		goto sendsig;	/* in PT_CONTINUE below */
852
853	case PT_CLEARSTEP:
854		error = ptrace_clear_single_step(td2);
855		break;
856
857	case PT_SETSTEP:
858		error = ptrace_single_step(td2);
859		break;
860
861	case PT_SUSPEND:
862		td2->td_dbgflags |= TDB_SUSPEND;
863		thread_lock(td2);
864		td2->td_flags |= TDF_NEEDSUSPCHK;
865		thread_unlock(td2);
866		break;
867
868	case PT_RESUME:
869		td2->td_dbgflags &= ~TDB_SUSPEND;
870		break;
871
872	case PT_FOLLOW_FORK:
873		if (data)
874			p->p_flag |= P_FOLLOWFORK;
875		else
876			p->p_flag &= ~P_FOLLOWFORK;
877		break;
878
879	case PT_STEP:
880	case PT_CONTINUE:
881	case PT_TO_SCE:
882	case PT_TO_SCX:
883	case PT_SYSCALL:
884	case PT_DETACH:
885		/* Zero means do not send any signal */
886		if (data < 0 || data > _SIG_MAXSIG) {
887			error = EINVAL;
888			break;
889		}
890
891		switch (req) {
892		case PT_STEP:
893			error = ptrace_single_step(td2);
894			if (error)
895				goto out;
896			break;
897		case PT_CONTINUE:
898		case PT_TO_SCE:
899		case PT_TO_SCX:
900		case PT_SYSCALL:
901			if (addr != (void *)1) {
902				error = ptrace_set_pc(td2,
903				    (u_long)(uintfptr_t)addr);
904				if (error)
905					goto out;
906			}
907			switch (req) {
908			case PT_TO_SCE:
909				p->p_stops |= S_PT_SCE;
910				break;
911			case PT_TO_SCX:
912				p->p_stops |= S_PT_SCX;
913				break;
914			case PT_SYSCALL:
915				p->p_stops |= S_PT_SCE | S_PT_SCX;
916				break;
917			}
918			break;
919		case PT_DETACH:
920			/* reset process parent */
921			if (p->p_oppid != p->p_pptr->p_pid) {
922				struct proc *pp;
923
924				PROC_LOCK(p->p_pptr);
925				sigqueue_take(p->p_ksi);
926				PROC_UNLOCK(p->p_pptr);
927
928				PROC_UNLOCK(p);
929				pp = pfind(p->p_oppid);
930				if (pp == NULL)
931					pp = initproc;
932				else
933					PROC_UNLOCK(pp);
934				PROC_LOCK(p);
935				proc_reparent(p, pp);
936				if (pp == initproc)
937					p->p_sigparent = SIGCHLD;
938			}
939			p->p_oppid = 0;
940			p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
941
942			/* should we send SIGCHLD? */
943			/* childproc_continued(p); */
944			break;
945		}
946
947	sendsig:
948		if (proctree_locked) {
949			sx_xunlock(&proctree_lock);
950			proctree_locked = 0;
951		}
952		p->p_xstat = data;
953		p->p_xthread = NULL;
954		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
955			/* deliver or queue signal */
956			td2->td_dbgflags &= ~TDB_XSIG;
957			td2->td_xsig = data;
958
959			if (req == PT_DETACH) {
960				struct thread *td3;
961				FOREACH_THREAD_IN_PROC(p, td3) {
962					td3->td_dbgflags &= ~TDB_SUSPEND;
963				}
964			}
965			/*
966			 * unsuspend all threads, to not let a thread run,
967			 * you should use PT_SUSPEND to suspend it before
968			 * continuing process.
969			 */
970			PROC_SLOCK(p);
971			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
972			thread_unsuspend(p);
973			PROC_SUNLOCK(p);
974		} else {
975			if (data)
976				kern_psignal(p, data);
977		}
978		break;
979
980	case PT_WRITE_I:
981	case PT_WRITE_D:
982		td2->td_dbgflags |= TDB_USERWR;
983		write = 1;
984		/* FALLTHROUGH */
985	case PT_READ_I:
986	case PT_READ_D:
987		PROC_UNLOCK(p);
988		tmp = 0;
989		/* write = 0 set above */
990		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
991		iov.iov_len = sizeof(int);
992		uio.uio_iov = &iov;
993		uio.uio_iovcnt = 1;
994		uio.uio_offset = (off_t)(uintptr_t)addr;
995		uio.uio_resid = sizeof(int);
996		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
997		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
998		uio.uio_td = td;
999		error = proc_rwmem(p, &uio);
1000		if (uio.uio_resid != 0) {
1001			/*
1002			 * XXX proc_rwmem() doesn't currently return ENOSPC,
1003			 * so I think write() can bogusly return 0.
1004			 * XXX what happens for short writes?  We don't want
1005			 * to write partial data.
1006			 * XXX proc_rwmem() returns EPERM for other invalid
1007			 * addresses.  Convert this to EINVAL.  Does this
1008			 * clobber returns of EPERM for other reasons?
1009			 */
1010			if (error == 0 || error == ENOSPC || error == EPERM)
1011				error = EINVAL;	/* EOF */
1012		}
1013		if (!write)
1014			td->td_retval[0] = tmp;
1015		PROC_LOCK(p);
1016		break;
1017
1018	case PT_IO:
1019#ifdef COMPAT_FREEBSD32
1020		if (wrap32) {
1021			piod32 = addr;
1022			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1023			iov.iov_len = piod32->piod_len;
1024			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1025			uio.uio_resid = piod32->piod_len;
1026		} else
1027#endif
1028		{
1029			piod = addr;
1030			iov.iov_base = piod->piod_addr;
1031			iov.iov_len = piod->piod_len;
1032			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1033			uio.uio_resid = piod->piod_len;
1034		}
1035		uio.uio_iov = &iov;
1036		uio.uio_iovcnt = 1;
1037		uio.uio_segflg = UIO_USERSPACE;
1038		uio.uio_td = td;
1039#ifdef COMPAT_FREEBSD32
1040		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1041#else
1042		tmp = piod->piod_op;
1043#endif
1044		switch (tmp) {
1045		case PIOD_READ_D:
1046		case PIOD_READ_I:
1047			uio.uio_rw = UIO_READ;
1048			break;
1049		case PIOD_WRITE_D:
1050		case PIOD_WRITE_I:
1051			td2->td_dbgflags |= TDB_USERWR;
1052			uio.uio_rw = UIO_WRITE;
1053			break;
1054		default:
1055			error = EINVAL;
1056			goto out;
1057		}
1058		PROC_UNLOCK(p);
1059		error = proc_rwmem(p, &uio);
1060#ifdef COMPAT_FREEBSD32
1061		if (wrap32)
1062			piod32->piod_len -= uio.uio_resid;
1063		else
1064#endif
1065			piod->piod_len -= uio.uio_resid;
1066		PROC_LOCK(p);
1067		break;
1068
1069	case PT_KILL:
1070		data = SIGKILL;
1071		goto sendsig;	/* in PT_CONTINUE above */
1072
1073	case PT_SETREGS:
1074		td2->td_dbgflags |= TDB_USERWR;
1075		error = PROC_WRITE(regs, td2, addr);
1076		break;
1077
1078	case PT_GETREGS:
1079		error = PROC_READ(regs, td2, addr);
1080		break;
1081
1082	case PT_SETFPREGS:
1083		td2->td_dbgflags |= TDB_USERWR;
1084		error = PROC_WRITE(fpregs, td2, addr);
1085		break;
1086
1087	case PT_GETFPREGS:
1088		error = PROC_READ(fpregs, td2, addr);
1089		break;
1090
1091	case PT_SETDBREGS:
1092		td2->td_dbgflags |= TDB_USERWR;
1093		error = PROC_WRITE(dbregs, td2, addr);
1094		break;
1095
1096	case PT_GETDBREGS:
1097		error = PROC_READ(dbregs, td2, addr);
1098		break;
1099
1100	case PT_LWPINFO:
1101		if (data <= 0 ||
1102#ifdef COMPAT_FREEBSD32
1103		    (!wrap32 && data > sizeof(*pl)) ||
1104		    (wrap32 && data > sizeof(*pl32))) {
1105#else
1106		    data > sizeof(*pl)) {
1107#endif
1108			error = EINVAL;
1109			break;
1110		}
1111#ifdef COMPAT_FREEBSD32
1112		if (wrap32) {
1113			pl = &plr;
1114			pl32 = addr;
1115		} else
1116#endif
1117		pl = addr;
1118		pl->pl_lwpid = td2->td_tid;
1119		pl->pl_event = PL_EVENT_NONE;
1120		pl->pl_flags = 0;
1121		if (td2->td_dbgflags & TDB_XSIG) {
1122			pl->pl_event = PL_EVENT_SIGNAL;
1123			if (td2->td_dbgksi.ksi_signo != 0 &&
1124#ifdef COMPAT_FREEBSD32
1125			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1126			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1127			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1128			    pl_siginfo) + sizeof(struct siginfo32)))
1129#else
1130			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1131			    + sizeof(pl->pl_siginfo)
1132#endif
1133			){
1134				pl->pl_flags |= PL_FLAG_SI;
1135				pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1136			}
1137		}
1138		if ((pl->pl_flags & PL_FLAG_SI) == 0)
1139			bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1140		if (td2->td_dbgflags & TDB_SCE)
1141			pl->pl_flags |= PL_FLAG_SCE;
1142		else if (td2->td_dbgflags & TDB_SCX)
1143			pl->pl_flags |= PL_FLAG_SCX;
1144		if (td2->td_dbgflags & TDB_EXEC)
1145			pl->pl_flags |= PL_FLAG_EXEC;
1146		if (td2->td_dbgflags & TDB_FORK) {
1147			pl->pl_flags |= PL_FLAG_FORKED;
1148			pl->pl_child_pid = td2->td_dbg_forked;
1149		}
1150		if (td2->td_dbgflags & TDB_CHILD)
1151			pl->pl_flags |= PL_FLAG_CHILD;
1152		pl->pl_sigmask = td2->td_sigmask;
1153		pl->pl_siglist = td2->td_siglist;
1154		strcpy(pl->pl_tdname, td2->td_name);
1155#ifdef COMPAT_FREEBSD32
1156		if (wrap32)
1157			ptrace_lwpinfo_to32(pl, pl32);
1158#endif
1159		break;
1160
1161	case PT_GETNUMLWPS:
1162		td->td_retval[0] = p->p_numthreads;
1163		break;
1164
1165	case PT_GETLWPLIST:
1166		if (data <= 0) {
1167			error = EINVAL;
1168			break;
1169		}
1170		num = imin(p->p_numthreads, data);
1171		PROC_UNLOCK(p);
1172		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1173		tmp = 0;
1174		PROC_LOCK(p);
1175		FOREACH_THREAD_IN_PROC(p, td2) {
1176			if (tmp >= num)
1177				break;
1178			buf[tmp++] = td2->td_tid;
1179		}
1180		PROC_UNLOCK(p);
1181		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1182		free(buf, M_TEMP);
1183		if (!error)
1184			td->td_retval[0] = tmp;
1185		PROC_LOCK(p);
1186		break;
1187
1188	case PT_VM_TIMESTAMP:
1189		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1190		break;
1191
1192	case PT_VM_ENTRY:
1193		PROC_UNLOCK(p);
1194#ifdef COMPAT_FREEBSD32
1195		if (wrap32)
1196			error = ptrace_vm_entry32(td, p, addr);
1197		else
1198#endif
1199		error = ptrace_vm_entry(td, p, addr);
1200		PROC_LOCK(p);
1201		break;
1202
1203	default:
1204#ifdef __HAVE_PTRACE_MACHDEP
1205		if (req >= PT_FIRSTMACH) {
1206			PROC_UNLOCK(p);
1207			error = cpu_ptrace(td2, req, addr, data);
1208			PROC_LOCK(p);
1209		} else
1210#endif
1211			/* Unknown request. */
1212			error = EINVAL;
1213		break;
1214	}
1215
1216out:
1217	/* Drop our hold on this process now that the request has completed. */
1218	_PRELE(p);
1219fail:
1220	PROC_UNLOCK(p);
1221	if (proctree_locked)
1222		sx_xunlock(&proctree_lock);
1223	return (error);
1224}
1225#undef PROC_READ
1226#undef PROC_WRITE
1227
1228/*
1229 * Stop a process because of a debugging event;
1230 * stay stopped until p->p_step is cleared
1231 * (cleared by PIOCCONT in procfs).
1232 */
1233void
1234stopevent(struct proc *p, unsigned int event, unsigned int val)
1235{
1236
1237	PROC_LOCK_ASSERT(p, MA_OWNED);
1238	p->p_step = 1;
1239	do {
1240		p->p_xstat = val;
1241		p->p_xthread = NULL;
1242		p->p_stype = event;	/* Which event caused the stop? */
1243		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1244		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1245	} while (p->p_step);
1246}
1247
1248static int
1249protect_setchild(struct thread *td, struct proc *p, int flags)
1250{
1251
1252	PROC_LOCK_ASSERT(p, MA_OWNED);
1253	if (p->p_flag & P_SYSTEM || p_cansee(td, p) != 0)
1254		return (0);
1255	if (flags & PPROT_SET) {
1256		p->p_flag |= P_PROTECTED;
1257		if (flags & PPROT_INHERIT)
1258			p->p_flag2 |= P2_INHERIT_PROTECTED;
1259	} else {
1260		p->p_flag &= ~P_PROTECTED;
1261		p->p_flag2 &= ~P2_INHERIT_PROTECTED;
1262	}
1263	return (1);
1264}
1265
1266static int
1267protect_setchildren(struct thread *td, struct proc *top, int flags)
1268{
1269	struct proc *p;
1270	int ret;
1271
1272	p = top;
1273	ret = 0;
1274	sx_assert(&proctree_lock, SX_LOCKED);
1275	for (;;) {
1276		ret |= protect_setchild(td, p, flags);
1277		PROC_UNLOCK(p);
1278		/*
1279		 * If this process has children, descend to them next,
1280		 * otherwise do any siblings, and if done with this level,
1281		 * follow back up the tree (but not past top).
1282		 */
1283		if (!LIST_EMPTY(&p->p_children))
1284			p = LIST_FIRST(&p->p_children);
1285		else for (;;) {
1286			if (p == top) {
1287				PROC_LOCK(p);
1288				return (ret);
1289			}
1290			if (LIST_NEXT(p, p_sibling)) {
1291				p = LIST_NEXT(p, p_sibling);
1292				break;
1293			}
1294			p = p->p_pptr;
1295		}
1296		PROC_LOCK(p);
1297	}
1298}
1299
1300static int
1301protect_set(struct thread *td, struct proc *p, int flags)
1302{
1303	int error, ret;
1304
1305	switch (PPROT_OP(flags)) {
1306	case PPROT_SET:
1307	case PPROT_CLEAR:
1308		break;
1309	default:
1310		return (EINVAL);
1311	}
1312
1313	if ((PPROT_FLAGS(flags) & ~(PPROT_DESCEND | PPROT_INHERIT)) != 0)
1314		return (EINVAL);
1315
1316	error = priv_check(td, PRIV_VM_MADV_PROTECT);
1317	if (error)
1318		return (error);
1319
1320	if (flags & PPROT_DESCEND)
1321		ret = protect_setchildren(td, p, flags);
1322	else
1323		ret = protect_setchild(td, p, flags);
1324	if (ret == 0)
1325		return (EPERM);
1326	return (0);
1327}
1328
1329#ifndef _SYS_SYSPROTO_H_
1330struct procctl_args {
1331	idtype_t idtype;
1332	id_t	id;
1333	int	com;
1334	void	*data;
1335};
1336#endif
1337/* ARGSUSED */
1338int
1339sys_procctl(struct thread *td, struct procctl_args *uap)
1340{
1341	int error, flags;
1342	void *data;
1343
1344	switch (uap->com) {
1345	case PROC_SPROTECT:
1346		error = copyin(uap->data, &flags, sizeof(flags));
1347		if (error)
1348			return (error);
1349		data = &flags;
1350		break;
1351	default:
1352		return (EINVAL);
1353	}
1354
1355	return (kern_procctl(td, uap->idtype, uap->id, uap->com, data));
1356}
1357
1358static int
1359kern_procctl_single(struct thread *td, struct proc *p, int com, void *data)
1360{
1361
1362	PROC_LOCK_ASSERT(p, MA_OWNED);
1363	switch (com) {
1364	case PROC_SPROTECT:
1365		return (protect_set(td, p, *(int *)data));
1366	default:
1367		return (EINVAL);
1368	}
1369}
1370
1371int
1372kern_procctl(struct thread *td, idtype_t idtype, id_t id, int com, void *data)
1373{
1374	struct pgrp *pg;
1375	struct proc *p;
1376	int error, first_error, ok;
1377
1378	sx_slock(&proctree_lock);
1379	switch (idtype) {
1380	case P_PID:
1381		p = pfind(id);
1382		if (p == NULL) {
1383			error = ESRCH;
1384			break;
1385		}
1386		if (p->p_state == PRS_NEW)
1387			error = ESRCH;
1388		else
1389			error = p_cansee(td, p);
1390		if (error == 0)
1391			error = kern_procctl_single(td, p, com, data);
1392		PROC_UNLOCK(p);
1393		break;
1394	case P_PGID:
1395		/*
1396		 * Attempt to apply the operation to all members of the
1397		 * group.  Ignore processes in the group that can't be
1398		 * seen.  Ignore errors so long as at least one process is
1399		 * able to complete the request successfully.
1400		 */
1401		pg = pgfind(id);
1402		if (pg == NULL) {
1403			error = ESRCH;
1404			break;
1405		}
1406		PGRP_UNLOCK(pg);
1407		ok = 0;
1408		first_error = 0;
1409		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1410			PROC_LOCK(p);
1411			if (p->p_state == PRS_NEW || p_cansee(td, p) != 0) {
1412				PROC_UNLOCK(p);
1413				continue;
1414			}
1415			error = kern_procctl_single(td, p, com, data);
1416			PROC_UNLOCK(p);
1417			if (error == 0)
1418				ok = 1;
1419			else if (first_error == 0)
1420				first_error = error;
1421		}
1422		if (ok)
1423			error = 0;
1424		else if (first_error != 0)
1425			error = first_error;
1426		else
1427			/*
1428			 * Was not able to see any processes in the
1429			 * process group.
1430			 */
1431			error = ESRCH;
1432		break;
1433	default:
1434		error = EINVAL;
1435		break;
1436	}
1437	sx_sunlock(&proctree_lock);
1438	return (error);
1439}
1440