sys_process.c revision 289780
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 289780 2015-10-23 01:27:44Z 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	int	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	vmspace_free(vm);
393
394	pve->pve_fsid = VNOVAL;
395	pve->pve_fileid = VNOVAL;
396
397	if (error == 0 && obj != NULL) {
398		lobj = obj;
399		for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
400			if (tobj != obj)
401				VM_OBJECT_RLOCK(tobj);
402			if (lobj != obj)
403				VM_OBJECT_RUNLOCK(lobj);
404			lobj = tobj;
405			pve->pve_offset += tobj->backing_object_offset;
406		}
407		vp = vm_object_vnode(lobj);
408		if (vp != NULL)
409			vref(vp);
410		if (lobj != obj)
411			VM_OBJECT_RUNLOCK(lobj);
412		VM_OBJECT_RUNLOCK(obj);
413
414		if (vp != NULL) {
415			freepath = NULL;
416			fullpath = NULL;
417			vn_fullpath(td, vp, &fullpath, &freepath);
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
425			if (fullpath != NULL) {
426				pve->pve_pathlen = strlen(fullpath) + 1;
427				if (pve->pve_pathlen <= pathlen) {
428					error = copyout(fullpath, pve->pve_path,
429					    pve->pve_pathlen);
430				} else
431					error = ENAMETOOLONG;
432			}
433			if (freepath != NULL)
434				free(freepath, M_TEMP);
435		}
436	}
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_DETACH:
670		sx_xlock(&proctree_lock);
671		proctree_locked = 1;
672		break;
673	default:
674		break;
675	}
676
677	write = 0;
678	if (req == PT_TRACE_ME) {
679		p = td->td_proc;
680		PROC_LOCK(p);
681	} else {
682		if (pid <= PID_MAX) {
683			if ((p = pfind(pid)) == NULL) {
684				if (proctree_locked)
685					sx_xunlock(&proctree_lock);
686				return (ESRCH);
687			}
688		} else {
689			td2 = tdfind(pid, -1);
690			if (td2 == NULL) {
691				if (proctree_locked)
692					sx_xunlock(&proctree_lock);
693				return (ESRCH);
694			}
695			p = td2->td_proc;
696			tid = pid;
697			pid = p->p_pid;
698		}
699	}
700	AUDIT_ARG_PROCESS(p);
701
702	if ((p->p_flag & P_WEXIT) != 0) {
703		error = ESRCH;
704		goto fail;
705	}
706	if ((error = p_cansee(td, p)) != 0)
707		goto fail;
708
709	if ((error = p_candebug(td, p)) != 0)
710		goto fail;
711
712	/*
713	 * System processes can't be debugged.
714	 */
715	if ((p->p_flag & P_SYSTEM) != 0) {
716		error = EINVAL;
717		goto fail;
718	}
719
720	if (tid == 0) {
721		if ((p->p_flag & P_STOPPED_TRACE) != 0) {
722			KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
723			td2 = p->p_xthread;
724		} else {
725			td2 = FIRST_THREAD_IN_PROC(p);
726		}
727		tid = td2->td_tid;
728	}
729
730#ifdef COMPAT_FREEBSD32
731	/*
732	 * Test if we're a 32 bit client and what the target is.
733	 * Set the wrap controls accordingly.
734	 */
735	if (SV_CURPROC_FLAG(SV_ILP32)) {
736		if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
737			safe = 1;
738		wrap32 = 1;
739	}
740#endif
741	/*
742	 * Permissions check
743	 */
744	switch (req) {
745	case PT_TRACE_ME:
746		/* Always legal. */
747		break;
748
749	case PT_ATTACH:
750		/* Self */
751		if (p->p_pid == td->td_proc->p_pid) {
752			error = EINVAL;
753			goto fail;
754		}
755
756		/* Already traced */
757		if (p->p_flag & P_TRACED) {
758			error = EBUSY;
759			goto fail;
760		}
761
762		/* Can't trace an ancestor if you're being traced. */
763		if (curp->p_flag & P_TRACED) {
764			for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
765				if (pp == p) {
766					error = EINVAL;
767					goto fail;
768				}
769			}
770		}
771
772
773		/* OK */
774		break;
775
776	case PT_CLEARSTEP:
777		/* Allow thread to clear single step for itself */
778		if (td->td_tid == tid)
779			break;
780
781		/* FALLTHROUGH */
782	default:
783		/* not being traced... */
784		if ((p->p_flag & P_TRACED) == 0) {
785			error = EPERM;
786			goto fail;
787		}
788
789		/* not being traced by YOU */
790		if (p->p_pptr != td->td_proc) {
791			error = EBUSY;
792			goto fail;
793		}
794
795		/* not currently stopped */
796		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
797		    p->p_suspcount != p->p_numthreads  ||
798		    (p->p_flag & P_WAITED) == 0) {
799			error = EBUSY;
800			goto fail;
801		}
802
803		if ((p->p_flag & P_STOPPED_TRACE) == 0) {
804			static int count = 0;
805			if (count++ == 0)
806				printf("P_STOPPED_TRACE not set.\n");
807		}
808
809		/* OK */
810		break;
811	}
812
813	/* Keep this process around until we finish this request. */
814	_PHOLD(p);
815
816#ifdef FIX_SSTEP
817	/*
818	 * Single step fixup ala procfs
819	 */
820	FIX_SSTEP(td2);
821#endif
822
823	/*
824	 * Actually do the requests
825	 */
826
827	td->td_retval[0] = 0;
828
829	switch (req) {
830	case PT_TRACE_ME:
831		/* set my trace flag and "owner" so it can read/write me */
832		p->p_flag |= P_TRACED;
833		if (p->p_flag & P_PPWAIT)
834			p->p_flag |= P_PPTRACE;
835		p->p_oppid = p->p_pptr->p_pid;
836		CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
837		break;
838
839	case PT_ATTACH:
840		/* security check done above */
841		/*
842		 * It would be nice if the tracing relationship was separate
843		 * from the parent relationship but that would require
844		 * another set of links in the proc struct or for "wait"
845		 * to scan the entire proc table.  To make life easier,
846		 * we just re-parent the process we're trying to trace.
847		 * The old parent is remembered so we can put things back
848		 * on a "detach".
849		 */
850		p->p_flag |= P_TRACED;
851		p->p_oppid = p->p_pptr->p_pid;
852		if (p->p_pptr != td->td_proc) {
853			proc_reparent(p, td->td_proc);
854		}
855		data = SIGSTOP;
856		CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
857		    p->p_oppid);
858		goto sendsig;	/* in PT_CONTINUE below */
859
860	case PT_CLEARSTEP:
861		CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
862		    p->p_pid);
863		error = ptrace_clear_single_step(td2);
864		break;
865
866	case PT_SETSTEP:
867		CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
868		    p->p_pid);
869		error = ptrace_single_step(td2);
870		break;
871
872	case PT_SUSPEND:
873		CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
874		    p->p_pid);
875		td2->td_dbgflags |= TDB_SUSPEND;
876		thread_lock(td2);
877		td2->td_flags |= TDF_NEEDSUSPCHK;
878		thread_unlock(td2);
879		break;
880
881	case PT_RESUME:
882		CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
883		    p->p_pid);
884		td2->td_dbgflags &= ~TDB_SUSPEND;
885		break;
886
887	case PT_FOLLOW_FORK:
888		CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
889		    p->p_flag & P_FOLLOWFORK ? "enabled" : "disabled",
890		    data ? "enabled" : "disabled");
891		if (data)
892			p->p_flag |= P_FOLLOWFORK;
893		else
894			p->p_flag &= ~P_FOLLOWFORK;
895		break;
896
897	case PT_STEP:
898	case PT_CONTINUE:
899	case PT_TO_SCE:
900	case PT_TO_SCX:
901	case PT_SYSCALL:
902	case PT_DETACH:
903		/* Zero means do not send any signal */
904		if (data < 0 || data > _SIG_MAXSIG) {
905			error = EINVAL;
906			break;
907		}
908
909		switch (req) {
910		case PT_STEP:
911			CTR2(KTR_PTRACE, "PT_STEP: tid %d (pid %d)",
912			    td2->td_tid, p->p_pid);
913			error = ptrace_single_step(td2);
914			if (error)
915				goto out;
916			break;
917		case PT_CONTINUE:
918		case PT_TO_SCE:
919		case PT_TO_SCX:
920		case PT_SYSCALL:
921			if (addr != (void *)1) {
922				error = ptrace_set_pc(td2,
923				    (u_long)(uintfptr_t)addr);
924				if (error)
925					goto out;
926			}
927			switch (req) {
928			case PT_TO_SCE:
929				p->p_stops |= S_PT_SCE;
930				CTR2(KTR_PTRACE,
931				    "PT_TO_SCE: pid %d, stops = %#x", p->p_pid,
932				    p->p_stops);
933				break;
934			case PT_TO_SCX:
935				p->p_stops |= S_PT_SCX;
936				CTR2(KTR_PTRACE,
937				    "PT_TO_SCX: pid %d, stops = %#x", p->p_pid,
938				    p->p_stops);
939				break;
940			case PT_SYSCALL:
941				p->p_stops |= S_PT_SCE | S_PT_SCX;
942				CTR2(KTR_PTRACE,
943				    "PT_SYSCALL: pid %d, stops = %#x", p->p_pid,
944				    p->p_stops);
945				break;
946			case PT_CONTINUE:
947				CTR1(KTR_PTRACE,
948				    "PT_CONTINUE: pid %d", p->p_pid);
949				break;
950			}
951			break;
952		case PT_DETACH:
953			/*
954			 * Reset the process parent.
955			 *
956			 * NB: This clears P_TRACED before reparenting
957			 * a detached process back to its original
958			 * parent.  Otherwise the debugee will be set
959			 * as an orphan of the debugger.
960			 */
961			p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
962			if (p->p_oppid != p->p_pptr->p_pid) {
963				PROC_LOCK(p->p_pptr);
964				sigqueue_take(p->p_ksi);
965				PROC_UNLOCK(p->p_pptr);
966
967				pp = proc_realparent(p);
968				proc_reparent(p, pp);
969				if (pp == initproc)
970					p->p_sigparent = SIGCHLD;
971				CTR2(KTR_PTRACE,
972				    "PT_DETACH: pid %d reparented to pid %d",
973				    p->p_pid, pp->p_pid);
974			} else
975				CTR1(KTR_PTRACE, "PT_DETACH: pid %d", p->p_pid);
976			p->p_oppid = 0;
977			p->p_stops = 0;
978
979			/* should we send SIGCHLD? */
980			/* childproc_continued(p); */
981			break;
982		}
983
984	sendsig:
985		if (proctree_locked) {
986			sx_xunlock(&proctree_lock);
987			proctree_locked = 0;
988		}
989		p->p_xstat = data;
990		p->p_xthread = NULL;
991		if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
992			/* deliver or queue signal */
993			td2->td_dbgflags &= ~TDB_XSIG;
994			td2->td_xsig = data;
995
996			if (req == PT_DETACH) {
997				FOREACH_THREAD_IN_PROC(p, td3)
998					td3->td_dbgflags &= ~TDB_SUSPEND;
999			}
1000			/*
1001			 * unsuspend all threads, to not let a thread run,
1002			 * you should use PT_SUSPEND to suspend it before
1003			 * continuing process.
1004			 */
1005			PROC_SLOCK(p);
1006			p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
1007			thread_unsuspend(p);
1008			PROC_SUNLOCK(p);
1009			if (req == PT_ATTACH)
1010				kern_psignal(p, data);
1011		} else {
1012			if (data)
1013				kern_psignal(p, data);
1014		}
1015		break;
1016
1017	case PT_WRITE_I:
1018	case PT_WRITE_D:
1019		td2->td_dbgflags |= TDB_USERWR;
1020		write = 1;
1021		/* FALLTHROUGH */
1022	case PT_READ_I:
1023	case PT_READ_D:
1024		PROC_UNLOCK(p);
1025		tmp = 0;
1026		/* write = 0 set above */
1027		iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1028		iov.iov_len = sizeof(int);
1029		uio.uio_iov = &iov;
1030		uio.uio_iovcnt = 1;
1031		uio.uio_offset = (off_t)(uintptr_t)addr;
1032		uio.uio_resid = sizeof(int);
1033		uio.uio_segflg = UIO_SYSSPACE;	/* i.e.: the uap */
1034		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1035		uio.uio_td = td;
1036		error = proc_rwmem(p, &uio);
1037		if (uio.uio_resid != 0) {
1038			/*
1039			 * XXX proc_rwmem() doesn't currently return ENOSPC,
1040			 * so I think write() can bogusly return 0.
1041			 * XXX what happens for short writes?  We don't want
1042			 * to write partial data.
1043			 * XXX proc_rwmem() returns EPERM for other invalid
1044			 * addresses.  Convert this to EINVAL.  Does this
1045			 * clobber returns of EPERM for other reasons?
1046			 */
1047			if (error == 0 || error == ENOSPC || error == EPERM)
1048				error = EINVAL;	/* EOF */
1049		}
1050		if (!write)
1051			td->td_retval[0] = tmp;
1052		if (error == 0) {
1053			if (write)
1054				CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1055				    p->p_pid, addr, data);
1056			else
1057				CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1058				    p->p_pid, addr, tmp);
1059		}
1060		PROC_LOCK(p);
1061		break;
1062
1063	case PT_IO:
1064#ifdef COMPAT_FREEBSD32
1065		if (wrap32) {
1066			piod32 = addr;
1067			iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1068			iov.iov_len = piod32->piod_len;
1069			uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1070			uio.uio_resid = piod32->piod_len;
1071		} else
1072#endif
1073		{
1074			piod = addr;
1075			iov.iov_base = piod->piod_addr;
1076			iov.iov_len = piod->piod_len;
1077			uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1078			uio.uio_resid = piod->piod_len;
1079		}
1080		uio.uio_iov = &iov;
1081		uio.uio_iovcnt = 1;
1082		uio.uio_segflg = UIO_USERSPACE;
1083		uio.uio_td = td;
1084#ifdef COMPAT_FREEBSD32
1085		tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1086#else
1087		tmp = piod->piod_op;
1088#endif
1089		switch (tmp) {
1090		case PIOD_READ_D:
1091		case PIOD_READ_I:
1092			CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1093			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1094			uio.uio_rw = UIO_READ;
1095			break;
1096		case PIOD_WRITE_D:
1097		case PIOD_WRITE_I:
1098			CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1099			    p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1100			td2->td_dbgflags |= TDB_USERWR;
1101			uio.uio_rw = UIO_WRITE;
1102			break;
1103		default:
1104			error = EINVAL;
1105			goto out;
1106		}
1107		PROC_UNLOCK(p);
1108		error = proc_rwmem(p, &uio);
1109#ifdef COMPAT_FREEBSD32
1110		if (wrap32)
1111			piod32->piod_len -= uio.uio_resid;
1112		else
1113#endif
1114			piod->piod_len -= uio.uio_resid;
1115		PROC_LOCK(p);
1116		break;
1117
1118	case PT_KILL:
1119		CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1120		data = SIGKILL;
1121		goto sendsig;	/* in PT_CONTINUE above */
1122
1123	case PT_SETREGS:
1124		CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1125		    p->p_pid);
1126		td2->td_dbgflags |= TDB_USERWR;
1127		error = PROC_WRITE(regs, td2, addr);
1128		break;
1129
1130	case PT_GETREGS:
1131		CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1132		    p->p_pid);
1133		error = PROC_READ(regs, td2, addr);
1134		break;
1135
1136	case PT_SETFPREGS:
1137		CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1138		    p->p_pid);
1139		td2->td_dbgflags |= TDB_USERWR;
1140		error = PROC_WRITE(fpregs, td2, addr);
1141		break;
1142
1143	case PT_GETFPREGS:
1144		CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1145		    p->p_pid);
1146		error = PROC_READ(fpregs, td2, addr);
1147		break;
1148
1149	case PT_SETDBREGS:
1150		CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1151		    p->p_pid);
1152		td2->td_dbgflags |= TDB_USERWR;
1153		error = PROC_WRITE(dbregs, td2, addr);
1154		break;
1155
1156	case PT_GETDBREGS:
1157		CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1158		    p->p_pid);
1159		error = PROC_READ(dbregs, td2, addr);
1160		break;
1161
1162	case PT_LWPINFO:
1163		if (data <= 0 ||
1164#ifdef COMPAT_FREEBSD32
1165		    (!wrap32 && data > sizeof(*pl)) ||
1166		    (wrap32 && data > sizeof(*pl32))) {
1167#else
1168		    data > sizeof(*pl)) {
1169#endif
1170			error = EINVAL;
1171			break;
1172		}
1173#ifdef COMPAT_FREEBSD32
1174		if (wrap32) {
1175			pl = &plr;
1176			pl32 = addr;
1177		} else
1178#endif
1179		pl = addr;
1180		pl->pl_lwpid = td2->td_tid;
1181		pl->pl_event = PL_EVENT_NONE;
1182		pl->pl_flags = 0;
1183		if (td2->td_dbgflags & TDB_XSIG) {
1184			pl->pl_event = PL_EVENT_SIGNAL;
1185			if (td2->td_dbgksi.ksi_signo != 0 &&
1186#ifdef COMPAT_FREEBSD32
1187			    ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1188			    pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1189			    (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1190			    pl_siginfo) + sizeof(struct siginfo32)))
1191#else
1192			    data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1193			    + sizeof(pl->pl_siginfo)
1194#endif
1195			){
1196				pl->pl_flags |= PL_FLAG_SI;
1197				pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1198			}
1199		}
1200		if ((pl->pl_flags & PL_FLAG_SI) == 0)
1201			bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1202		if (td2->td_dbgflags & TDB_SCE)
1203			pl->pl_flags |= PL_FLAG_SCE;
1204		else if (td2->td_dbgflags & TDB_SCX)
1205			pl->pl_flags |= PL_FLAG_SCX;
1206		if (td2->td_dbgflags & TDB_EXEC)
1207			pl->pl_flags |= PL_FLAG_EXEC;
1208		if (td2->td_dbgflags & TDB_FORK) {
1209			pl->pl_flags |= PL_FLAG_FORKED;
1210			pl->pl_child_pid = td2->td_dbg_forked;
1211		}
1212		if (td2->td_dbgflags & TDB_CHILD)
1213			pl->pl_flags |= PL_FLAG_CHILD;
1214		pl->pl_sigmask = td2->td_sigmask;
1215		pl->pl_siglist = td2->td_siglist;
1216		strcpy(pl->pl_tdname, td2->td_name);
1217		if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1218			pl->pl_syscall_code = td2->td_dbg_sc_code;
1219			pl->pl_syscall_narg = td2->td_dbg_sc_narg;
1220		} else {
1221			pl->pl_syscall_code = 0;
1222			pl->pl_syscall_narg = 0;
1223		}
1224#ifdef COMPAT_FREEBSD32
1225		if (wrap32)
1226			ptrace_lwpinfo_to32(pl, pl32);
1227#endif
1228		CTR5(KTR_PTRACE,
1229	    "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d",
1230		    td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1231		    pl->pl_child_pid);
1232		break;
1233
1234	case PT_GETNUMLWPS:
1235		CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1236		    p->p_numthreads);
1237		td->td_retval[0] = p->p_numthreads;
1238		break;
1239
1240	case PT_GETLWPLIST:
1241		CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1242		    p->p_pid, data, p->p_numthreads);
1243		if (data <= 0) {
1244			error = EINVAL;
1245			break;
1246		}
1247		num = imin(p->p_numthreads, data);
1248		PROC_UNLOCK(p);
1249		buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1250		tmp = 0;
1251		PROC_LOCK(p);
1252		FOREACH_THREAD_IN_PROC(p, td2) {
1253			if (tmp >= num)
1254				break;
1255			buf[tmp++] = td2->td_tid;
1256		}
1257		PROC_UNLOCK(p);
1258		error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1259		free(buf, M_TEMP);
1260		if (!error)
1261			td->td_retval[0] = tmp;
1262		PROC_LOCK(p);
1263		break;
1264
1265	case PT_VM_TIMESTAMP:
1266		CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1267		    p->p_pid, p->p_vmspace->vm_map.timestamp);
1268		td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1269		break;
1270
1271	case PT_VM_ENTRY:
1272		PROC_UNLOCK(p);
1273#ifdef COMPAT_FREEBSD32
1274		if (wrap32)
1275			error = ptrace_vm_entry32(td, p, addr);
1276		else
1277#endif
1278		error = ptrace_vm_entry(td, p, addr);
1279		PROC_LOCK(p);
1280		break;
1281
1282	default:
1283#ifdef __HAVE_PTRACE_MACHDEP
1284		if (req >= PT_FIRSTMACH) {
1285			PROC_UNLOCK(p);
1286			error = cpu_ptrace(td2, req, addr, data);
1287			PROC_LOCK(p);
1288		} else
1289#endif
1290			/* Unknown request. */
1291			error = EINVAL;
1292		break;
1293	}
1294
1295out:
1296	/* Drop our hold on this process now that the request has completed. */
1297	_PRELE(p);
1298fail:
1299	PROC_UNLOCK(p);
1300	if (proctree_locked)
1301		sx_xunlock(&proctree_lock);
1302	return (error);
1303}
1304#undef PROC_READ
1305#undef PROC_WRITE
1306
1307/*
1308 * Stop a process because of a debugging event;
1309 * stay stopped until p->p_step is cleared
1310 * (cleared by PIOCCONT in procfs).
1311 */
1312void
1313stopevent(struct proc *p, unsigned int event, unsigned int val)
1314{
1315
1316	PROC_LOCK_ASSERT(p, MA_OWNED);
1317	p->p_step = 1;
1318	CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1319	    val);
1320	do {
1321		p->p_xstat = val;
1322		p->p_xthread = NULL;
1323		p->p_stype = event;	/* Which event caused the stop? */
1324		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
1325		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1326	} while (p->p_step);
1327}
1328