sys_process.c revision 304017
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: stable/10/sys/kern/sys_process.c 304017 2016-08-12 19:43:06Z jhb $");
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/vnode.h>
47#include <sys/ptrace.h>
48#include <sys/rwlock.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_param.h>
65
66#ifdef COMPAT_FREEBSD32
67#include <sys/procfs.h>
68#include <compat/freebsd32/freebsd32_signal.h>
69
70struct ptrace_io_desc32 {
71	int		piod_op;
72	uint32_t	piod_offs;
73	uint32_t	piod_addr;
74	uint32_t	piod_len;
75};
76
77struct ptrace_vm_entry32 {
78	int		pve_entry;
79	int		pve_timestamp;
80	uint32_t	pve_start;
81	uint32_t	pve_end;
82	uint32_t	pve_offset;
83	u_int		pve_prot;
84	u_int		pve_pathlen;
85	int32_t		pve_fileid;
86	u_int		pve_fsid;
87	uint32_t	pve_path;
88};
89
90struct ptrace_lwpinfo32 {
91	lwpid_t	pl_lwpid;	/* LWP described. */
92	int	pl_event;	/* Event that stopped the LWP. */
93	int	pl_flags;	/* LWP flags. */
94	sigset_t	pl_sigmask;	/* LWP signal mask */
95	sigset_t	pl_siglist;	/* LWP pending signal */
96	struct siginfo32 pl_siginfo;	/* siginfo for signal */
97	char	pl_tdname[MAXCOMLEN + 1];	/* LWP name. */
98	pid_t	pl_child_pid;		/* New child pid */
99	u_int		pl_syscall_code;
100	u_int		pl_syscall_narg;
101};
102
103#endif
104
105/*
106 * Functions implemented using PROC_ACTION():
107 *
108 * proc_read_regs(proc, regs)
109 *	Get the current user-visible register set from the process
110 *	and copy it into the regs structure (<machine/reg.h>).
111 *	The process is stopped at the time read_regs is called.
112 *
113 * proc_write_regs(proc, regs)
114 *	Update the current register set from the passed in regs
115 *	structure.  Take care to avoid clobbering special CPU
116 *	registers or privileged bits in the PSL.
117 *	Depending on the architecture this may have fix-up work to do,
118 *	especially if the IAR or PCW are modified.
119 *	The process is stopped at the time write_regs is called.
120 *
121 * proc_read_fpregs, proc_write_fpregs
122 *	deal with the floating point register set, otherwise as above.
123 *
124 * proc_read_dbregs, proc_write_dbregs
125 *	deal with the processor debug register set, otherwise as above.
126 *
127 * proc_sstep(proc)
128 *	Arrange for the process to trap after executing a single instruction.
129 */
130
131#define	PROC_ACTION(action) do {					\
132	int error;							\
133									\
134	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
135	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
136		error = EIO;						\
137	else								\
138		error = (action);					\
139	return (error);							\
140} while(0)
141
142int
143proc_read_regs(struct thread *td, struct reg *regs)
144{
145
146	PROC_ACTION(fill_regs(td, regs));
147}
148
149int
150proc_write_regs(struct thread *td, struct reg *regs)
151{
152
153	PROC_ACTION(set_regs(td, regs));
154}
155
156int
157proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
158{
159
160	PROC_ACTION(fill_dbregs(td, dbregs));
161}
162
163int
164proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
165{
166
167	PROC_ACTION(set_dbregs(td, dbregs));
168}
169
170/*
171 * Ptrace doesn't support fpregs at all, and there are no security holes
172 * or translations for fpregs, so we can just copy them.
173 */
174int
175proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
176{
177
178	PROC_ACTION(fill_fpregs(td, fpregs));
179}
180
181int
182proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
183{
184
185	PROC_ACTION(set_fpregs(td, fpregs));
186}
187
188#ifdef COMPAT_FREEBSD32
189/* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
190int
191proc_read_regs32(struct thread *td, struct reg32 *regs32)
192{
193
194	PROC_ACTION(fill_regs32(td, regs32));
195}
196
197int
198proc_write_regs32(struct thread *td, struct reg32 *regs32)
199{
200
201	PROC_ACTION(set_regs32(td, regs32));
202}
203
204int
205proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
206{
207
208	PROC_ACTION(fill_dbregs32(td, dbregs32));
209}
210
211int
212proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
213{
214
215	PROC_ACTION(set_dbregs32(td, dbregs32));
216}
217
218int
219proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
220{
221
222	PROC_ACTION(fill_fpregs32(td, fpregs32));
223}
224
225int
226proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
227{
228
229	PROC_ACTION(set_fpregs32(td, fpregs32));
230}
231#endif
232
233int
234proc_sstep(struct thread *td)
235{
236
237	PROC_ACTION(ptrace_single_step(td));
238}
239
240int
241proc_rwmem(struct proc *p, struct uio *uio)
242{
243	vm_map_t map;
244	vm_offset_t pageno;		/* page number */
245	vm_prot_t reqprot;
246	int error, fault_flags, page_offset, writing;
247
248	/*
249	 * Assert that someone has locked this vmspace.  (Should be
250	 * curthread but we can't assert that.)  This keeps the process
251	 * from exiting out from under us until this operation completes.
252	 */
253	KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
254	    p, p->p_pid));
255
256	/*
257	 * The map we want...
258	 */
259	map = &p->p_vmspace->vm_map;
260
261	/*
262	 * If we are writing, then we request vm_fault() to create a private
263	 * copy of each page.  Since these copies will not be writeable by the
264	 * process, we must explicity request that they be dirtied.
265	 */
266	writing = uio->uio_rw == UIO_WRITE;
267	reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
268	fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
269
270	/*
271	 * Only map in one page at a time.  We don't have to, but it
272	 * makes things easier.  This way is trivial - right?
273	 */
274	do {
275		vm_offset_t uva;
276		u_int len;
277		vm_page_t m;
278
279		uva = (vm_offset_t)uio->uio_offset;
280
281		/*
282		 * Get the page number of this segment.
283		 */
284		pageno = trunc_page(uva);
285		page_offset = uva - pageno;
286
287		/*
288		 * How many bytes to copy
289		 */
290		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
291
292		/*
293		 * Fault and hold the page on behalf of the process.
294		 */
295		error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
296		if (error != KERN_SUCCESS) {
297			if (error == KERN_RESOURCE_SHORTAGE)
298				error = ENOMEM;
299			else
300				error = EFAULT;
301			break;
302		}
303
304		/*
305		 * Now do the i/o move.
306		 */
307		error = uiomove_fromphys(&m, page_offset, len, uio);
308
309		/* Make the I-cache coherent for breakpoints. */
310		if (writing && error == 0) {
311			vm_map_lock_read(map);
312			if (vm_map_check_protection(map, pageno, pageno +
313			    PAGE_SIZE, VM_PROT_EXECUTE))
314				vm_sync_icache(map, uva, len);
315			vm_map_unlock_read(map);
316		}
317
318		/*
319		 * Release the page.
320		 */
321		vm_page_lock(m);
322		vm_page_unhold(m);
323		vm_page_unlock(m);
324
325	} while (error == 0 && uio->uio_resid > 0);
326
327	return (error);
328}
329
330static int
331ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
332{
333	struct vattr vattr;
334	vm_map_t map;
335	vm_map_entry_t entry;
336	vm_object_t obj, tobj, lobj;
337	struct vmspace *vm;
338	struct vnode *vp;
339	char *freepath, *fullpath;
340	u_int pathlen;
341	int error, index;
342
343	error = 0;
344	obj = NULL;
345
346	vm = vmspace_acquire_ref(p);
347	map = &vm->vm_map;
348	vm_map_lock_read(map);
349
350	do {
351		entry = map->header.next;
352		index = 0;
353		while (index < pve->pve_entry && entry != &map->header) {
354			entry = entry->next;
355			index++;
356		}
357		if (index != pve->pve_entry) {
358			error = EINVAL;
359			break;
360		}
361		while (entry != &map->header &&
362		    (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
363			entry = entry->next;
364			index++;
365		}
366		if (entry == &map->header) {
367			error = ENOENT;
368			break;
369		}
370
371		/* We got an entry. */
372		pve->pve_entry = index + 1;
373		pve->pve_timestamp = map->timestamp;
374		pve->pve_start = entry->start;
375		pve->pve_end = entry->end - 1;
376		pve->pve_offset = entry->offset;
377		pve->pve_prot = entry->protection;
378
379		/* Backing object's path needed? */
380		if (pve->pve_pathlen == 0)
381			break;
382
383		pathlen = pve->pve_pathlen;
384		pve->pve_pathlen = 0;
385
386		obj = entry->object.vm_object;
387		if (obj != NULL)
388			VM_OBJECT_RLOCK(obj);
389	} while (0);
390
391	vm_map_unlock_read(map);
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_RLOCK(tobj);
401			if (lobj != obj)
402				VM_OBJECT_RUNLOCK(lobj);
403			lobj = tobj;
404			pve->pve_offset += tobj->backing_object_offset;
405		}
406		vp = vm_object_vnode(lobj);
407		if (vp != NULL)
408			vref(vp);
409		if (lobj != obj)
410			VM_OBJECT_RUNLOCK(lobj);
411		VM_OBJECT_RUNLOCK(obj);
412
413		if (vp != NULL) {
414			freepath = NULL;
415			fullpath = NULL;
416			vn_fullpath(td, vp, &fullpath, &freepath);
417			vn_lock(vp, LK_SHARED | LK_RETRY);
418			if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
419				pve->pve_fileid = vattr.va_fileid;
420				pve->pve_fsid = vattr.va_fsid;
421			}
422			vput(vp);
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	vmspace_free(vm);
437	if (error == 0)
438		CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
439		    p->p_pid, pve->pve_entry, pve->pve_start);
440
441	return (error);
442}
443
444#ifdef COMPAT_FREEBSD32
445static int
446ptrace_vm_entry32(struct thread *td, struct proc *p,
447    struct ptrace_vm_entry32 *pve32)
448{
449	struct ptrace_vm_entry pve;
450	int error;
451
452	pve.pve_entry = pve32->pve_entry;
453	pve.pve_pathlen = pve32->pve_pathlen;
454	pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
455
456	error = ptrace_vm_entry(td, p, &pve);
457	if (error == 0) {
458		pve32->pve_entry = pve.pve_entry;
459		pve32->pve_timestamp = pve.pve_timestamp;
460		pve32->pve_start = pve.pve_start;
461		pve32->pve_end = pve.pve_end;
462		pve32->pve_offset = pve.pve_offset;
463		pve32->pve_prot = pve.pve_prot;
464		pve32->pve_fileid = pve.pve_fileid;
465		pve32->pve_fsid = pve.pve_fsid;
466	}
467
468	pve32->pve_pathlen = pve.pve_pathlen;
469	return (error);
470}
471
472static void
473ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
474    struct ptrace_lwpinfo32 *pl32)
475{
476
477	pl32->pl_lwpid = pl->pl_lwpid;
478	pl32->pl_event = pl->pl_event;
479	pl32->pl_flags = pl->pl_flags;
480	pl32->pl_sigmask = pl->pl_sigmask;
481	pl32->pl_siglist = pl->pl_siglist;
482	siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
483	strcpy(pl32->pl_tdname, pl->pl_tdname);
484	pl32->pl_child_pid = pl->pl_child_pid;
485	pl32->pl_syscall_code = pl->pl_syscall_code;
486	pl32->pl_syscall_narg = pl->pl_syscall_narg;
487}
488#endif /* COMPAT_FREEBSD32 */
489
490/*
491 * Process debugging system call.
492 */
493#ifndef _SYS_SYSPROTO_H_
494struct ptrace_args {
495	int	req;
496	pid_t	pid;
497	caddr_t	addr;
498	int	data;
499};
500#endif
501
502#ifdef COMPAT_FREEBSD32
503/*
504 * This CPP subterfuge is to try and reduce the number of ifdefs in
505 * the body of the code.
506 *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
507 * becomes either:
508 *   copyin(uap->addr, &r.reg, sizeof r.reg);
509 * or
510 *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
511 * .. except this is done at runtime.
512 */
513#define	COPYIN(u, k, s)		wrap32 ? \
514	copyin(u, k ## 32, s ## 32) : \
515	copyin(u, k, s)
516#define	COPYOUT(k, u, s)	wrap32 ? \
517	copyout(k ## 32, u, s ## 32) : \
518	copyout(k, u, s)
519#else
520#define	COPYIN(u, k, s)		copyin(u, k, s)
521#define	COPYOUT(k, u, s)	copyout(k, u, s)
522#endif
523int
524sys_ptrace(struct thread *td, struct ptrace_args *uap)
525{
526	/*
527	 * XXX this obfuscation is to reduce stack usage, but the register
528	 * structs may be too large to put on the stack anyway.
529	 */
530	union {
531		struct ptrace_io_desc piod;
532		struct ptrace_lwpinfo pl;
533		struct ptrace_vm_entry pve;
534		struct dbreg dbreg;
535		struct fpreg fpreg;
536		struct reg reg;
537#ifdef COMPAT_FREEBSD32
538		struct dbreg32 dbreg32;
539		struct fpreg32 fpreg32;
540		struct reg32 reg32;
541		struct ptrace_io_desc32 piod32;
542		struct ptrace_lwpinfo32 pl32;
543		struct ptrace_vm_entry32 pve32;
544#endif
545	} r;
546	void *addr;
547	int error = 0;
548#ifdef COMPAT_FREEBSD32
549	int wrap32 = 0;
550
551	if (SV_CURPROC_FLAG(SV_ILP32))
552		wrap32 = 1;
553#endif
554	AUDIT_ARG_PID(uap->pid);
555	AUDIT_ARG_CMD(uap->req);
556	AUDIT_ARG_VALUE(uap->data);
557	addr = &r;
558	switch (uap->req) {
559	case PT_GETREGS:
560	case PT_GETFPREGS:
561	case PT_GETDBREGS:
562	case PT_LWPINFO:
563		break;
564	case PT_SETREGS:
565		error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
566		break;
567	case PT_SETFPREGS:
568		error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
569		break;
570	case PT_SETDBREGS:
571		error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
572		break;
573	case PT_IO:
574		error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
575		break;
576	case PT_VM_ENTRY:
577		error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
578		break;
579	default:
580		addr = uap->addr;
581		break;
582	}
583	if (error)
584		return (error);
585
586	error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
587	if (error)
588		return (error);
589
590	switch (uap->req) {
591	case PT_VM_ENTRY:
592		error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
593		break;
594	case PT_IO:
595		error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
596		break;
597	case PT_GETREGS:
598		error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
599		break;
600	case PT_GETFPREGS:
601		error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
602		break;
603	case PT_GETDBREGS:
604		error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
605		break;
606	case PT_LWPINFO:
607		error = copyout(&r.pl, uap->addr, uap->data);
608		break;
609	}
610
611	return (error);
612}
613#undef COPYIN
614#undef COPYOUT
615
616#ifdef COMPAT_FREEBSD32
617/*
618 *   PROC_READ(regs, td2, addr);
619 * becomes either:
620 *   proc_read_regs(td2, addr);
621 * or
622 *   proc_read_regs32(td2, addr);
623 * .. except this is done at runtime.  There is an additional
624 * complication in that PROC_WRITE disallows 32 bit consumers
625 * from writing to 64 bit address space targets.
626 */
627#define	PROC_READ(w, t, a)	wrap32 ? \
628	proc_read_ ## w ## 32(t, a) : \
629	proc_read_ ## w (t, a)
630#define	PROC_WRITE(w, t, a)	wrap32 ? \
631	(safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
632	proc_write_ ## w (t, a)
633#else
634#define	PROC_READ(w, t, a)	proc_read_ ## w (t, a)
635#define	PROC_WRITE(w, t, a)	proc_write_ ## w (t, a)
636#endif
637
638int
639kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
640{
641	struct iovec iov;
642	struct uio uio;
643	struct proc *curp, *p, *pp;
644	struct thread *td2 = NULL, *td3;
645	struct ptrace_io_desc *piod = NULL;
646	struct ptrace_lwpinfo *pl;
647	int error, write, tmp, num;
648	int proctree_locked = 0;
649	lwpid_t tid = 0, *buf;
650#ifdef COMPAT_FREEBSD32
651	int wrap32 = 0, safe = 0;
652	struct ptrace_io_desc32 *piod32 = NULL;
653	struct ptrace_lwpinfo32 *pl32 = NULL;
654	struct ptrace_lwpinfo plr;
655#endif
656
657	curp = td->td_proc;
658
659	/* Lock proctree before locking the process. */
660	switch (req) {
661	case PT_TRACE_ME:
662	case PT_ATTACH:
663	case PT_STEP:
664	case PT_CONTINUE:
665	case PT_TO_SCE:
666	case PT_TO_SCX:
667	case PT_SYSCALL:
668	case PT_FOLLOW_FORK:
669	case PT_LWP_EVENTS:
670	case PT_DETACH:
671		sx_xlock(&proctree_lock);
672		proctree_locked = 1;
673		break;
674	default:
675		break;
676	}
677
678	write = 0;
679	if (req == PT_TRACE_ME) {
680		p = td->td_proc;
681		PROC_LOCK(p);
682	} else {
683		if (pid <= PID_MAX) {
684			if ((p = pfind(pid)) == NULL) {
685				if (proctree_locked)
686					sx_xunlock(&proctree_lock);
687				return (ESRCH);
688			}
689		} else {
690			td2 = tdfind(pid, -1);
691			if (td2 == NULL) {
692				if (proctree_locked)
693					sx_xunlock(&proctree_lock);
694				return (ESRCH);
695			}
696			p = td2->td_proc;
697			tid = pid;
698			pid = p->p_pid;
699		}
700	}
701	AUDIT_ARG_PROCESS(p);
702
703	if ((p->p_flag & P_WEXIT) != 0) {
704		error = ESRCH;
705		goto fail;
706	}
707	if ((error = p_cansee(td, p)) != 0)
708		goto fail;
709
710	if ((error = p_candebug(td, p)) != 0)
711		goto fail;
712
713	/*
714	 * System processes can't be debugged.
715	 */
716	if ((p->p_flag & P_SYSTEM) != 0) {
717		error = EINVAL;
718		goto fail;
719	}
720
721	if (tid == 0) {
722		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
723			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
724			td2 = p->p_xthread;
725		} else {
726			td2 = FIRST_THREAD_IN_PROC(p);
727		}
728		tid = td2->td_tid;
729	}
730
731#ifdef COMPAT_FREEBSD32
732	/*
733	 * Test if we're a 32 bit client and what the target is.
734	 * Set the wrap controls accordingly.
735	 */
736	if (SV_CURPROC_FLAG(SV_ILP32)) {
737		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
738			safe = 1;
739		wrap32 = 1;
740	}
741#endif
742	/*
743	 * Permissions check
744	 */
745	switch (req) {
746	case PT_TRACE_ME:
747		/*
748		 * Always legal, when there is a parent process which
749		 * could trace us.  Otherwise, reject.
750		 */
751		if ((p->p_flag & P_TRACED) != 0) {
752			error = EBUSY;
753			goto fail;
754		}
755		if (p->p_pptr == initproc) {
756			error = EPERM;
757			goto fail;
758		}
759		break;
760
761	case PT_ATTACH:
762		/* Self */
763		if (p == td->td_proc) {
764			error = EINVAL;
765			goto fail;
766		}
767
768		/* Already traced */
769		if (p->p_flag & P_TRACED) {
770			error = EBUSY;
771			goto fail;
772		}
773
774		/* Can't trace an ancestor if you're being traced. */
775		if (curp->p_flag & P_TRACED) {
776			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
777				if (pp == p) {
778					error = EINVAL;
779					goto fail;
780				}
781			}
782		}
783
784
785		/* OK */
786		break;
787
788	case PT_CLEARSTEP:
789		/* Allow thread to clear single step for itself */
790		if (td->td_tid == tid)
791			break;
792
793		/* FALLTHROUGH */
794	default:
795		/* not being traced... */
796		if ((p->p_flag & P_TRACED) == 0) {
797			error = EPERM;
798			goto fail;
799		}
800
801		/* not being traced by YOU */
802		if (p->p_pptr != td->td_proc) {
803			error = EBUSY;
804			goto fail;
805		}
806
807		/* not currently stopped */
808		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
809		    p->p_suspcount != p->p_numthreads  ||
810		    (p->p_flag & P_WAITED) == 0) {
811			error = EBUSY;
812			goto fail;
813		}
814
815		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
816			static int count = 0;
817			if (count++ == 0)
818				printf("P_STOPPED_TRACE not set.\n");
819		}
820
821		/* OK */
822		break;
823	}
824
825	/* Keep this process around until we finish this request. */
826	_PHOLD(p);
827
828#ifdef FIX_SSTEP
829	/*
830	 * Single step fixup ala procfs
831	 */
832	FIX_SSTEP(td2);
833#endif
834
835	/*
836	 * Actually do the requests
837	 */
838
839	td->td_retval[0] = 0;
840
841	switch (req) {
842	case PT_TRACE_ME:
843		/* set my trace flag and "owner" so it can read/write me */
844		p->p_flag |= P_TRACED;
845		if (p->p_flag & P_PPWAIT)
846			p->p_flag |= P_PPTRACE;
847		p->p_oppid = p->p_pptr->p_pid;
848		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
849		break;
850
851	case PT_ATTACH:
852		/* security check done above */
853		/*
854		 * It would be nice if the tracing relationship was separate
855		 * from the parent relationship but that would require
856		 * another set of links in the proc struct or for "wait"
857		 * to scan the entire proc table.  To make life easier,
858		 * we just re-parent the process we're trying to trace.
859		 * The old parent is remembered so we can put things back
860		 * on a "detach".
861		 */
862		p->p_flag |= P_TRACED;
863		p->p_oppid = p->p_pptr->p_pid;
864		if (p->p_pptr != td->td_proc) {
865			proc_reparent(p, td->td_proc);
866		}
867		data = SIGSTOP;
868		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
869		    p->p_oppid);
870		goto sendsig;	/* in PT_CONTINUE below */
871
872	case PT_CLEARSTEP:
873		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
874		    p->p_pid);
875		error = ptrace_clear_single_step(td2);
876		break;
877
878	case PT_SETSTEP:
879		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
880		    p->p_pid);
881		error = ptrace_single_step(td2);
882		break;
883
884	case PT_SUSPEND:
885		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
886		    p->p_pid);
887		td2->td_dbgflags |= TDB_SUSPEND;
888		thread_lock(td2);
889		td2->td_flags |= TDF_NEEDSUSPCHK;
890		thread_unlock(td2);
891		break;
892
893	case PT_RESUME:
894		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
895		    p->p_pid);
896		td2->td_dbgflags &= ~TDB_SUSPEND;
897		break;
898
899	case PT_FOLLOW_FORK:
900		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
901		    p->p_flag & P_FOLLOWFORK ? "enabled" : "disabled",
902		    data ? "enabled" : "disabled");
903		if (data)
904			p->p_flag |= P_FOLLOWFORK;
905		else
906			p->p_flag &= ~P_FOLLOWFORK;
907		break;
908
909	case PT_LWP_EVENTS:
910		CTR3(KTR_PTRACE, "PT_LWP_EVENTS: pid %d %s -> %s", p->p_pid,
911		    p->p_flag2 & P2_LWP_EVENTS ? "enabled" : "disabled",
912		    data ? "enabled" : "disabled");
913		if (data)
914			p->p_flag2 |= P2_LWP_EVENTS;
915		else
916			p->p_flag2 &= ~P2_LWP_EVENTS;
917		break;
918
919	case PT_STEP:
920	case PT_CONTINUE:
921	case PT_TO_SCE:
922	case PT_TO_SCX:
923	case PT_SYSCALL:
924	case PT_DETACH:
925		/* Zero means do not send any signal */
926		if (data < 0 || data > _SIG_MAXSIG) {
927			error = EINVAL;
928			break;
929		}
930
931		switch (req) {
932		case PT_STEP:
933			CTR2(KTR_PTRACE, "PT_STEP: tid %d (pid %d)",
934			    td2->td_tid, p->p_pid);
935			error = ptrace_single_step(td2);
936			if (error)
937				goto out;
938			break;
939		case PT_CONTINUE:
940		case PT_TO_SCE:
941		case PT_TO_SCX:
942		case PT_SYSCALL:
943			if (addr != (void *)1) {
944				error = ptrace_set_pc(td2,
945				    (u_long)(uintfptr_t)addr);
946				if (error)
947					goto out;
948			}
949			switch (req) {
950			case PT_TO_SCE:
951				p->p_stops |= S_PT_SCE;
952				CTR4(KTR_PTRACE,
953		    "PT_TO_SCE: pid %d, stops = %#x, PC = %#lx, sig = %d",
954				    p->p_pid, p->p_stops,
955				    (u_long)(uintfptr_t)addr, data);
956				break;
957			case PT_TO_SCX:
958				p->p_stops |= S_PT_SCX;
959				CTR4(KTR_PTRACE,
960		    "PT_TO_SCX: pid %d, stops = %#x, PC = %#lx, sig = %d",
961				    p->p_pid, p->p_stops,
962				    (u_long)(uintfptr_t)addr, data);
963				break;
964			case PT_SYSCALL:
965				p->p_stops |= S_PT_SCE | S_PT_SCX;
966				CTR4(KTR_PTRACE,
967		    "PT_SYSCALL: pid %d, stops = %#x, PC = %#lx, sig = %d",
968				    p->p_pid, p->p_stops,
969				    (u_long)(uintfptr_t)addr, data);
970				break;
971			case PT_CONTINUE:
972				CTR3(KTR_PTRACE,
973				    "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
974				    p->p_pid, (u_long)(uintfptr_t)addr, data);
975				break;
976			}
977			break;
978		case PT_DETACH:
979			/*
980			 * Reset the process parent.
981			 *
982			 * NB: This clears P_TRACED before reparenting
983			 * a detached process back to its original
984			 * parent.  Otherwise the debugee will be set
985			 * as an orphan of the debugger.
986			 */
987			p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
988			if (p->p_oppid != p->p_pptr->p_pid) {
989				PROC_LOCK(p->p_pptr);
990				sigqueue_take(p->p_ksi);
991				PROC_UNLOCK(p->p_pptr);
992
993				pp = proc_realparent(p);
994				proc_reparent(p, pp);
995				if (pp == initproc)
996					p->p_sigparent = SIGCHLD;
997				CTR3(KTR_PTRACE,
998			    "PT_DETACH: pid %d reparented to pid %d, sig %d",
999				    p->p_pid, pp->p_pid, data);
1000			} else
1001				CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
1002				    p->p_pid, data);
1003			p->p_oppid = 0;
1004			p->p_stops = 0;
1005
1006			/* should we send SIGCHLD? */
1007			/* childproc_continued(p); */
1008			break;
1009		}
1010
1011	sendsig:
1012		if (proctree_locked) {
1013			sx_xunlock(&proctree_lock);
1014			proctree_locked = 0;
1015		}
1016		p->p_xstat = data;
1017		p->p_xthread = NULL;
1018		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
1019			/* deliver or queue signal */
1020			td2->td_dbgflags &= ~TDB_XSIG;
1021			td2->td_xsig = data;
1022
1023			if (req == PT_DETACH) {
1024				FOREACH_THREAD_IN_PROC(p, td3)
1025					td3->td_dbgflags &= ~TDB_SUSPEND;
1026			}
1027			/*
1028			 * unsuspend all threads, to not let a thread run,
1029			 * you should use PT_SUSPEND to suspend it before
1030			 * continuing process.
1031			 */
1032			PROC_SLOCK(p);
1033			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
1034			thread_unsuspend(p);
1035			PROC_SUNLOCK(p);
1036			if (req == PT_ATTACH)
1037				kern_psignal(p, data);
1038		} else {
1039			if (data)
1040				kern_psignal(p, data);
1041		}
1042		break;
1043
1044	case PT_WRITE_I:
1045	case PT_WRITE_D:
1046		td2->td_dbgflags |= TDB_USERWR;
1047		write = 1;
1048		/* FALLTHROUGH */
1049	case PT_READ_I:
1050	case PT_READ_D:
1051		PROC_UNLOCK(p);
1052		tmp = 0;
1053		/* write = 0 set above */
1054		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1055		iov.iov_len = sizeof(int);
1056		uio.uio_iov = &iov;
1057		uio.uio_iovcnt = 1;
1058		uio.uio_offset = (off_t)(uintptr_t)addr;
1059		uio.uio_resid = sizeof(int);
1060		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
1061		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1062		uio.uio_td = td;
1063		error = proc_rwmem(p, &uio);
1064		if (uio.uio_resid != 0) {
1065			/*
1066			 * XXX proc_rwmem() doesn't currently return ENOSPC,
1067			 * so I think write() can bogusly return 0.
1068			 * XXX what happens for short writes?  We don't want
1069			 * to write partial data.
1070			 * XXX proc_rwmem() returns EPERM for other invalid
1071			 * addresses.  Convert this to EINVAL.  Does this
1072			 * clobber returns of EPERM for other reasons?
1073			 */
1074			if (error == 0 || error == ENOSPC || error == EPERM)
1075				error = EINVAL;	/* EOF */
1076		}
1077		if (!write)
1078			td->td_retval[0] = tmp;
1079		if (error == 0) {
1080			if (write)
1081				CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1082				    p->p_pid, addr, data);
1083			else
1084				CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1085				    p->p_pid, addr, tmp);
1086		}
1087		PROC_LOCK(p);
1088		break;
1089
1090	case PT_IO:
1091#ifdef COMPAT_FREEBSD32
1092		if (wrap32) {
1093			piod32 = addr;
1094			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1095			iov.iov_len = piod32->piod_len;
1096			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1097			uio.uio_resid = piod32->piod_len;
1098		} else
1099#endif
1100		{
1101			piod = addr;
1102			iov.iov_base = piod->piod_addr;
1103			iov.iov_len = piod->piod_len;
1104			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1105			uio.uio_resid = piod->piod_len;
1106		}
1107		uio.uio_iov = &iov;
1108		uio.uio_iovcnt = 1;
1109		uio.uio_segflg = UIO_USERSPACE;
1110		uio.uio_td = td;
1111#ifdef COMPAT_FREEBSD32
1112		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1113#else
1114		tmp = piod->piod_op;
1115#endif
1116		switch (tmp) {
1117		case PIOD_READ_D:
1118		case PIOD_READ_I:
1119			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1120			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1121			uio.uio_rw = UIO_READ;
1122			break;
1123		case PIOD_WRITE_D:
1124		case PIOD_WRITE_I:
1125			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1126			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1127			td2->td_dbgflags |= TDB_USERWR;
1128			uio.uio_rw = UIO_WRITE;
1129			break;
1130		default:
1131			error = EINVAL;
1132			goto out;
1133		}
1134		PROC_UNLOCK(p);
1135		error = proc_rwmem(p, &uio);
1136#ifdef COMPAT_FREEBSD32
1137		if (wrap32)
1138			piod32->piod_len -= uio.uio_resid;
1139		else
1140#endif
1141			piod->piod_len -= uio.uio_resid;
1142		PROC_LOCK(p);
1143		break;
1144
1145	case PT_KILL:
1146		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1147		data = SIGKILL;
1148		goto sendsig;	/* in PT_CONTINUE above */
1149
1150	case PT_SETREGS:
1151		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1152		    p->p_pid);
1153		td2->td_dbgflags |= TDB_USERWR;
1154		error = PROC_WRITE(regs, td2, addr);
1155		break;
1156
1157	case PT_GETREGS:
1158		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1159		    p->p_pid);
1160		error = PROC_READ(regs, td2, addr);
1161		break;
1162
1163	case PT_SETFPREGS:
1164		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1165		    p->p_pid);
1166		td2->td_dbgflags |= TDB_USERWR;
1167		error = PROC_WRITE(fpregs, td2, addr);
1168		break;
1169
1170	case PT_GETFPREGS:
1171		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1172		    p->p_pid);
1173		error = PROC_READ(fpregs, td2, addr);
1174		break;
1175
1176	case PT_SETDBREGS:
1177		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1178		    p->p_pid);
1179		td2->td_dbgflags |= TDB_USERWR;
1180		error = PROC_WRITE(dbregs, td2, addr);
1181		break;
1182
1183	case PT_GETDBREGS:
1184		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1185		    p->p_pid);
1186		error = PROC_READ(dbregs, td2, addr);
1187		break;
1188
1189	case PT_LWPINFO:
1190		if (data <= 0 ||
1191#ifdef COMPAT_FREEBSD32
1192		    (!wrap32 && data > sizeof(*pl)) ||
1193		    (wrap32 && data > sizeof(*pl32))) {
1194#else
1195		    data > sizeof(*pl)) {
1196#endif
1197			error = EINVAL;
1198			break;
1199		}
1200#ifdef COMPAT_FREEBSD32
1201		if (wrap32) {
1202			pl = &plr;
1203			pl32 = addr;
1204		} else
1205#endif
1206		pl = addr;
1207		pl->pl_lwpid = td2->td_tid;
1208		pl->pl_event = PL_EVENT_NONE;
1209		pl->pl_flags = 0;
1210		if (td2->td_dbgflags & TDB_XSIG) {
1211			pl->pl_event = PL_EVENT_SIGNAL;
1212			if (td2->td_dbgksi.ksi_signo != 0 &&
1213#ifdef COMPAT_FREEBSD32
1214			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1215			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1216			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1217			    pl_siginfo) + sizeof(struct siginfo32)))
1218#else
1219			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1220			    + sizeof(pl->pl_siginfo)
1221#endif
1222			){
1223				pl->pl_flags |= PL_FLAG_SI;
1224				pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1225			}
1226		}
1227		if ((pl->pl_flags & PL_FLAG_SI) == 0)
1228			bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1229		if (td2->td_dbgflags & TDB_SCE)
1230			pl->pl_flags |= PL_FLAG_SCE;
1231		else if (td2->td_dbgflags & TDB_SCX)
1232			pl->pl_flags |= PL_FLAG_SCX;
1233		if (td2->td_dbgflags & TDB_EXEC)
1234			pl->pl_flags |= PL_FLAG_EXEC;
1235		if (td2->td_dbgflags & TDB_FORK) {
1236			pl->pl_flags |= PL_FLAG_FORKED;
1237			pl->pl_child_pid = td2->td_dbg_forked;
1238		}
1239		if (td2->td_dbgflags & TDB_CHILD)
1240			pl->pl_flags |= PL_FLAG_CHILD;
1241		if (td2->td_dbgflags & TDB_BORN)
1242			pl->pl_flags |= PL_FLAG_BORN;
1243		if (td2->td_dbgflags & TDB_EXIT)
1244			pl->pl_flags |= PL_FLAG_EXITED;
1245		pl->pl_sigmask = td2->td_sigmask;
1246		pl->pl_siglist = td2->td_siglist;
1247		strcpy(pl->pl_tdname, td2->td_name);
1248		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1249			pl->pl_syscall_code = td2->td_dbg_sc_code;
1250			pl->pl_syscall_narg = td2->td_dbg_sc_narg;
1251		} else {
1252			pl->pl_syscall_code = 0;
1253			pl->pl_syscall_narg = 0;
1254		}
1255#ifdef COMPAT_FREEBSD32
1256		if (wrap32)
1257			ptrace_lwpinfo_to32(pl, pl32);
1258#endif
1259		CTR6(KTR_PTRACE,
1260    "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1261		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1262		    pl->pl_child_pid, pl->pl_syscall_code);
1263		break;
1264
1265	case PT_GETNUMLWPS:
1266		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1267		    p->p_numthreads);
1268		td->td_retval[0] = p->p_numthreads;
1269		break;
1270
1271	case PT_GETLWPLIST:
1272		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1273		    p->p_pid, data, p->p_numthreads);
1274		if (data <= 0) {
1275			error = EINVAL;
1276			break;
1277		}
1278		num = imin(p->p_numthreads, data);
1279		PROC_UNLOCK(p);
1280		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1281		tmp = 0;
1282		PROC_LOCK(p);
1283		FOREACH_THREAD_IN_PROC(p, td2) {
1284			if (tmp >= num)
1285				break;
1286			buf[tmp++] = td2->td_tid;
1287		}
1288		PROC_UNLOCK(p);
1289		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1290		free(buf, M_TEMP);
1291		if (!error)
1292			td->td_retval[0] = tmp;
1293		PROC_LOCK(p);
1294		break;
1295
1296	case PT_VM_TIMESTAMP:
1297		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1298		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1299		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1300		break;
1301
1302	case PT_VM_ENTRY:
1303		PROC_UNLOCK(p);
1304#ifdef COMPAT_FREEBSD32
1305		if (wrap32)
1306			error = ptrace_vm_entry32(td, p, addr);
1307		else
1308#endif
1309		error = ptrace_vm_entry(td, p, addr);
1310		PROC_LOCK(p);
1311		break;
1312
1313	default:
1314#ifdef __HAVE_PTRACE_MACHDEP
1315		if (req >= PT_FIRSTMACH) {
1316			PROC_UNLOCK(p);
1317			error = cpu_ptrace(td2, req, addr, data);
1318			PROC_LOCK(p);
1319		} else
1320#endif
1321			/* Unknown request. */
1322			error = EINVAL;
1323		break;
1324	}
1325
1326out:
1327	/* Drop our hold on this process now that the request has completed. */
1328	_PRELE(p);
1329fail:
1330	PROC_UNLOCK(p);
1331	if (proctree_locked)
1332		sx_xunlock(&proctree_lock);
1333	return (error);
1334}
1335#undef PROC_READ
1336#undef PROC_WRITE
1337
1338/*
1339 * Stop a process because of a debugging event;
1340 * stay stopped until p->p_step is cleared
1341 * (cleared by PIOCCONT in procfs).
1342 */
1343void
1344stopevent(struct proc *p, unsigned int event, unsigned int val)
1345{
1346
1347	PROC_LOCK_ASSERT(p, MA_OWNED);
1348	p->p_step = 1;
1349	CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1350	    val);
1351	do {
1352		p->p_xstat = val;
1353		p->p_xthread = NULL;
1354		p->p_stype = event;	/* Which event caused the stop? */
1355		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1356		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1357	} while (p->p_step);
1358}
1359