xref: /freebsd/sys/kern/sys_process.c (revision 17d6c636720d00f77e5d098daf4c278f89d84f7b)
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  * $FreeBSD$
32  */
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/sysproto.h>
39 #include <sys/proc.h>
40 #include <sys/vnode.h>
41 #include <sys/ptrace.h>
42 #include <sys/sx.h>
43 #include <sys/user.h>
44 
45 #include <machine/reg.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_param.h>
49 #include <vm/pmap.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_kern.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_page.h>
55 
56 #define	PROC_REG_ACTION(name, action, type)				\
57 int									\
58 proc_##name##_##type##s(struct thread *td, struct type *regs)		\
59 {									\
60 	int error;							\
61 									\
62 	mtx_lock_spin(&sched_lock);					\
63 	error = (action##_##type##s(td, regs));				\
64 	mtx_unlock_spin(&sched_lock);					\
65 	return (error);							\
66 }
67 
68 PROC_REG_ACTION(read, fill, reg);
69 PROC_REG_ACTION(write, set, reg);
70 PROC_REG_ACTION(read, fill, dbreg);
71 PROC_REG_ACTION(write, set, dbreg);
72 PROC_REG_ACTION(read, fill, fpreg);
73 PROC_REG_ACTION(write, set, fpreg);
74 
75 int
76 proc_sstep(struct thread *td)
77 {
78 	int error;
79 
80 	mtx_lock_spin(&sched_lock);
81 	error = ptrace_single_step(td);
82 	mtx_unlock_spin(&sched_lock);
83 	return (error);
84 }
85 
86 int
87 proc_rwmem(struct proc *p, struct uio *uio)
88 {
89 	struct vmspace *vm;
90 	vm_map_t map;
91 	vm_object_t object = NULL;
92 	vm_offset_t pageno = 0;		/* page number */
93 	vm_prot_t reqprot;
94 	vm_offset_t kva;
95 	int error;
96 	int writing;
97 
98 	GIANT_REQUIRED;
99 
100 	/*
101 	 * if the vmspace is in the midst of being deallocated or the
102 	 * process is exiting, don't try to grab anything.  The page table
103 	 * usage in that process can be messed up.
104 	 */
105 	vm = p->p_vmspace;
106 	if ((p->p_flag & P_WEXIT))
107 		return (EFAULT);
108 	if (vm->vm_refcnt < 1)
109 		return (EFAULT);
110 	++vm->vm_refcnt;
111 	/*
112 	 * The map we want...
113 	 */
114 	map = &vm->vm_map;
115 
116 	writing = uio->uio_rw == UIO_WRITE;
117 	reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
118 	    VM_PROT_READ;
119 
120 	kva = kmem_alloc_pageable(kernel_map, PAGE_SIZE);
121 
122 	/*
123 	 * Only map in one page at a time.  We don't have to, but it
124 	 * makes things easier.  This way is trivial - right?
125 	 */
126 	do {
127 		vm_map_t tmap;
128 		vm_offset_t uva;
129 		int page_offset;		/* offset into page */
130 		vm_map_entry_t out_entry;
131 		vm_prot_t out_prot;
132 		boolean_t wired;
133 		vm_pindex_t pindex;
134 		u_int len;
135 		vm_page_t m;
136 
137 		object = NULL;
138 
139 		uva = (vm_offset_t)uio->uio_offset;
140 
141 		/*
142 		 * Get the page number of this segment.
143 		 */
144 		pageno = trunc_page(uva);
145 		page_offset = uva - pageno;
146 
147 		/*
148 		 * How many bytes to copy
149 		 */
150 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
151 
152 		/*
153 		 * Fault the page on behalf of the process
154 		 */
155 		error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
156 		if (error) {
157 			error = EFAULT;
158 			break;
159 		}
160 
161 		/*
162 		 * Now we need to get the page.  out_entry, out_prot, wired,
163 		 * and single_use aren't used.  One would think the vm code
164 		 * would be a *bit* nicer...  We use tmap because
165 		 * vm_map_lookup() can change the map argument.
166 		 */
167 		tmap = map;
168 		error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
169 		    &object, &pindex, &out_prot, &wired);
170 
171 		if (error) {
172 			error = EFAULT;
173 
174 			/*
175 			 * Make sure that there is no residue in 'object' from
176 			 * an error return on vm_map_lookup.
177 			 */
178 			object = NULL;
179 
180 			break;
181 		}
182 
183 		m = vm_page_lookup(object, pindex);
184 
185 		/* Allow fallback to backing objects if we are reading */
186 
187 		while (m == NULL && !writing && object->backing_object) {
188 
189 			pindex += OFF_TO_IDX(object->backing_object_offset);
190 			object = object->backing_object;
191 
192 			m = vm_page_lookup(object, pindex);
193 		}
194 
195 		if (m == NULL) {
196 			error = EFAULT;
197 
198 			/*
199 			 * Make sure that there is no residue in 'object' from
200 			 * an error return on vm_map_lookup.
201 			 */
202 			object = NULL;
203 
204 			vm_map_lookup_done(tmap, out_entry);
205 
206 			break;
207 		}
208 
209 		/*
210 		 * Wire the page into memory
211 		 */
212 		vm_page_wire(m);
213 
214 		/*
215 		 * We're done with tmap now.
216 		 * But reference the object first, so that we won't loose
217 		 * it.
218 		 */
219 		vm_object_reference(object);
220 		vm_map_lookup_done(tmap, out_entry);
221 
222 		pmap_kenter(kva, VM_PAGE_TO_PHYS(m));
223 
224 		/*
225 		 * Now do the i/o move.
226 		 */
227 		error = uiomove((caddr_t)(kva + page_offset), len, uio);
228 
229 		pmap_kremove(kva);
230 
231 		/*
232 		 * release the page and the object
233 		 */
234 		vm_page_unwire(m, 1);
235 		vm_object_deallocate(object);
236 
237 		object = NULL;
238 
239 	} while (error == 0 && uio->uio_resid > 0);
240 
241 	if (object)
242 		vm_object_deallocate(object);
243 
244 	kmem_free(kernel_map, kva, PAGE_SIZE);
245 	vmspace_free(vm);
246 	return (error);
247 }
248 
249 /*
250  * Process debugging system call.
251  */
252 #ifndef _SYS_SYSPROTO_H_
253 struct ptrace_args {
254 	int	req;
255 	pid_t	pid;
256 	caddr_t	addr;
257 	int	data;
258 };
259 #endif
260 
261 int
262 ptrace(struct thread *td, struct ptrace_args *uap)
263 {
264 	struct proc *curp = td->td_proc;
265 	struct proc *p;
266 	struct iovec iov;
267 	struct uio uio;
268 	union {
269 		struct reg	reg;
270 		struct dbreg	dbreg;
271 		struct fpreg	fpreg;
272 	} r;
273 	int error = 0;
274 	int write;
275 
276 	write = 0;
277 	if (uap->req == PT_TRACE_ME) {
278 		p = curp;
279 		PROC_LOCK(p);
280 	} else {
281 		if ((p = pfind(uap->pid)) == NULL)
282 			return (ESRCH);
283 	}
284 	if (p_cansee(curp, p)) {
285 		PROC_UNLOCK(p);
286 		return (ESRCH);
287 	}
288 
289 	if ((error = p_candebug(curp, p)) != 0) {
290 		PROC_UNLOCK(p);
291 		return (error);
292 	}
293 
294 	/*
295 	 * Don't debug system processes!
296 	 */
297 	if ((p->p_flag & P_SYSTEM) != 0) {
298 		PROC_UNLOCK(p);
299 		return (EINVAL);
300 	}
301 
302 	/*
303 	 * Permissions check
304 	 */
305 	switch (uap->req) {
306 	case PT_TRACE_ME:
307 		/* Always legal. */
308 		break;
309 
310 	case PT_ATTACH:
311 		/* Self */
312 		if (p->p_pid == curp->p_pid) {
313 			PROC_UNLOCK(p);
314 			return (EINVAL);
315 		}
316 
317 		/* Already traced */
318 		if (p->p_flag & P_TRACED) {
319 			PROC_UNLOCK(p);
320 			return (EBUSY);
321 		}
322 
323 		/* OK */
324 		break;
325 
326 	case PT_READ_I:
327 	case PT_READ_D:
328 	case PT_WRITE_I:
329 	case PT_WRITE_D:
330 	case PT_CONTINUE:
331 	case PT_KILL:
332 	case PT_STEP:
333 	case PT_DETACH:
334 #ifdef PT_GETREGS
335 	case PT_GETREGS:
336 #endif
337 #ifdef PT_SETREGS
338 	case PT_SETREGS:
339 #endif
340 #ifdef PT_GETFPREGS
341 	case PT_GETFPREGS:
342 #endif
343 #ifdef PT_SETFPREGS
344 	case PT_SETFPREGS:
345 #endif
346 #ifdef PT_GETDBREGS
347 	case PT_GETDBREGS:
348 #endif
349 #ifdef PT_SETDBREGS
350 	case PT_SETDBREGS:
351 #endif
352 		/* not being traced... */
353 		if ((p->p_flag & P_TRACED) == 0) {
354 			PROC_UNLOCK(p);
355 			return (EPERM);
356 		}
357 
358 		/* not being traced by YOU */
359 		if (p->p_pptr != curp) {
360 			PROC_UNLOCK(p);
361 			return (EBUSY);
362 		}
363 
364 		/* not currently stopped */
365 		mtx_lock_spin(&sched_lock);
366 		if (p->p_stat != SSTOP || (p->p_flag & P_WAITED) == 0) {
367 			mtx_unlock_spin(&sched_lock);
368 			PROC_UNLOCK(p);
369 			return (EBUSY);
370 		}
371 		mtx_unlock_spin(&sched_lock);
372 
373 		/* OK */
374 		break;
375 
376 	default:
377 		PROC_UNLOCK(p);
378 		return (EINVAL);
379 	}
380 
381 	PROC_UNLOCK(p);
382 #ifdef FIX_SSTEP
383 	/*
384 	 * Single step fixup ala procfs
385 	 */
386 	FIX_SSTEP(&p->p_thread);	/* XXXKSE */
387 #endif
388 
389 	/*
390 	 * Actually do the requests
391 	 */
392 
393 	td->td_retval[0] = 0;
394 
395 	switch (uap->req) {
396 	case PT_TRACE_ME:
397 		/* set my trace flag and "owner" so it can read/write me */
398 		sx_xlock(&proctree_lock);
399 		PROC_LOCK(p);
400 		p->p_flag |= P_TRACED;
401 		p->p_oppid = p->p_pptr->p_pid;
402 		PROC_UNLOCK(p);
403 		sx_xunlock(&proctree_lock);
404 		return (0);
405 
406 	case PT_ATTACH:
407 		/* security check done above */
408 		sx_xlock(&proctree_lock);
409 		PROC_LOCK(p);
410 		p->p_flag |= P_TRACED;
411 		p->p_oppid = p->p_pptr->p_pid;
412 		if (p->p_pptr != curp)
413 			proc_reparent(p, curp);
414 		PROC_UNLOCK(p);
415 		sx_xunlock(&proctree_lock);
416 		uap->data = SIGSTOP;
417 		goto sendsig;	/* in PT_CONTINUE below */
418 
419 	case PT_STEP:
420 	case PT_CONTINUE:
421 	case PT_DETACH:
422 		if ((uap->req != PT_STEP) && ((unsigned)uap->data >= NSIG))
423 			return (EINVAL);
424 
425 		PHOLD(p);
426 
427 		if (uap->req == PT_STEP) {
428 			if ((error = ptrace_single_step(&p->p_thread))) {
429 				PRELE(p);
430 				return (error);
431 			}
432 		}
433 
434 		if (uap->addr != (caddr_t)1) {
435 			fill_kinfo_proc(p, &p->p_uarea->u_kproc);
436 			if ((error = ptrace_set_pc(&p->p_thread,
437 			    (u_long)(uintfptr_t)uap->addr))) {
438 				PRELE(p);
439 				return (error);
440 			}
441 		}
442 		PRELE(p);
443 
444 		if (uap->req == PT_DETACH) {
445 			/* reset process parent */
446 			sx_xlock(&proctree_lock);
447 			if (p->p_oppid != p->p_pptr->p_pid) {
448 				struct proc *pp;
449 
450 				pp = pfind(p->p_oppid);
451 				if (pp != NULL)
452 					PROC_UNLOCK(pp);
453 				else
454 					pp = initproc;
455 				PROC_LOCK(p);
456 				proc_reparent(p, pp);
457 			} else
458 				PROC_LOCK(p);
459 			p->p_flag &= ~(P_TRACED | P_WAITED);
460 			p->p_oppid = 0;
461 
462 			PROC_UNLOCK(p);
463 			sx_xunlock(&proctree_lock);
464 
465 			/* should we send SIGCHLD? */
466 
467 		}
468 
469 	sendsig:
470 		/* deliver or queue signal */
471 		PROC_LOCK(p);
472 		mtx_lock_spin(&sched_lock);
473 		if (p->p_stat == SSTOP) {
474 			p->p_xstat = uap->data;
475 			setrunnable(&p->p_thread); /* XXXKSE */
476 			mtx_unlock_spin(&sched_lock);
477 		} else {
478 			mtx_unlock_spin(&sched_lock);
479 			if (uap->data)
480 				psignal(p, uap->data);
481 
482 		}
483 		PROC_UNLOCK(p);
484 		return (0);
485 
486 	case PT_WRITE_I:
487 	case PT_WRITE_D:
488 		write = 1;
489 		/* fallthrough */
490 	case PT_READ_I:
491 	case PT_READ_D:
492 		/* write = 0 set above */
493 		iov.iov_base = write ? (caddr_t)&uap->data :
494 		    (caddr_t)td->td_retval;
495 		iov.iov_len = sizeof(int);
496 		uio.uio_iov = &iov;
497 		uio.uio_iovcnt = 1;
498 		uio.uio_offset = (off_t)(uintptr_t)uap->addr;
499 		uio.uio_resid = sizeof(int);
500 		uio.uio_segflg = UIO_SYSSPACE;	/* ie: the uap */
501 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
502 		uio.uio_td = td;
503 		error = proc_rwmem(p, &uio);
504 		if (uio.uio_resid != 0) {
505 			/*
506 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
507 			 * so I think write() can bogusly return 0.
508 			 * XXX what happens for short writes?  We don't want
509 			 * to write partial data.
510 			 * XXX proc_rwmem() returns EPERM for other invalid
511 			 * addresses.  Convert this to EINVAL.  Does this
512 			 * clobber returns of EPERM for other reasons?
513 			 */
514 			if (error == 0 || error == ENOSPC || error == EPERM)
515 				error = EINVAL;	/* EOF */
516 		}
517 		return (error);
518 
519 	case PT_KILL:
520 		uap->data = SIGKILL;
521 		goto sendsig;	/* in PT_CONTINUE above */
522 
523 #ifdef PT_SETREGS
524 	case PT_SETREGS:
525 		error = copyin(uap->addr, &r.reg, sizeof r.reg);
526 		if (error == 0) {
527 			PHOLD(p);
528 			error = proc_write_regs(&p->p_thread, &r.reg);
529 			PRELE(p);
530 		}
531 		return (error);
532 #endif /* PT_SETREGS */
533 
534 #ifdef PT_GETREGS
535 	case PT_GETREGS:
536 		PHOLD(p);
537 		error = proc_read_regs(&p->p_thread, &r.reg);
538 		PRELE(p);
539 		if (error == 0)
540 			error = copyout(&r.reg, uap->addr, sizeof r.reg);
541 		return (error);
542 #endif /* PT_SETREGS */
543 
544 #ifdef PT_SETFPREGS
545 	case PT_SETFPREGS:
546 		error = copyin(uap->addr, &r.fpreg, sizeof r.fpreg);
547 		if (error == 0) {
548 			PHOLD(p);
549 			error = proc_write_fpregs(&p->p_thread, &r.fpreg);
550 			PRELE(p);
551 		}
552 		return (error);
553 #endif /* PT_SETFPREGS */
554 
555 #ifdef PT_GETFPREGS
556 	case PT_GETFPREGS:
557 		PHOLD(p);
558 		error = proc_read_fpregs(&p->p_thread, &r.fpreg);
559 		PRELE(p);
560 		if (error == 0)
561 			error = copyout(&r.fpreg, uap->addr, sizeof r.fpreg);
562 		return (error);
563 #endif /* PT_SETFPREGS */
564 
565 #ifdef PT_SETDBREGS
566 	case PT_SETDBREGS:
567 		error = copyin(uap->addr, &r.dbreg, sizeof r.dbreg);
568 		if (error == 0) {
569 			PHOLD(p);
570 			error = proc_write_dbregs(&p->p_thread, &r.dbreg);
571 			PRELE(p);
572 		}
573 		return (error);
574 #endif /* PT_SETDBREGS */
575 
576 #ifdef PT_GETDBREGS
577 	case PT_GETDBREGS:
578 		PHOLD(p);
579 		error = proc_read_dbregs(&p->p_thread, &r.dbreg);
580 		PRELE(p);
581 		if (error == 0)
582 			error = copyout(&r.dbreg, uap->addr, sizeof r.dbreg);
583 		return (error);
584 #endif /* PT_SETDBREGS */
585 
586 	default:
587 		KASSERT(0, ("unreachable code\n"));
588 		break;
589 	}
590 
591 	KASSERT(0, ("unreachable code\n"));
592 	return (0);
593 }
594 
595 int
596 trace_req(struct proc *p)
597 {
598 	return (1);
599 }
600 
601 /*
602  * stopevent()
603  * Stop a process because of a debugging event;
604  * stay stopped until p->p_step is cleared
605  * (cleared by PIOCCONT in procfs).
606  *
607  * Must be called with the proc struct mutex held.
608  */
609 
610 void
611 stopevent(struct proc *p, unsigned int event, unsigned int val)
612 {
613 
614 	PROC_LOCK_ASSERT(p, MA_OWNED | MA_NOTRECURSED);
615 	p->p_step = 1;
616 
617 	do {
618 		p->p_xstat = val;
619 		p->p_stype = event;	/* Which event caused the stop? */
620 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
621 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
622 	} while (p->p_step);
623 }
624