xref: /freebsd/sys/kern/sys_process.c (revision ae2bac960e47ce620c0d2d9ce2068670ea9f8aa6)
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 int
57 proc_rwmem(struct proc *p, struct uio *uio)
58 {
59 	struct vmspace *vm;
60 	vm_map_t map;
61 	vm_object_t object = NULL;
62 	vm_offset_t pageno = 0;		/* page number */
63 	vm_prot_t reqprot;
64 	vm_offset_t kva;
65 	int error;
66 	int writing;
67 
68 	GIANT_REQUIRED;
69 
70 	/*
71 	 * if the vmspace is in the midst of being deallocated or the
72 	 * process is exiting, don't try to grab anything.  The page table
73 	 * usage in that process can be messed up.
74 	 */
75 	vm = p->p_vmspace;
76 	if ((p->p_flag & P_WEXIT))
77 		return (EFAULT);
78 	if (vm->vm_refcnt < 1)
79 		return (EFAULT);
80 	++vm->vm_refcnt;
81 	/*
82 	 * The map we want...
83 	 */
84 	map = &vm->vm_map;
85 
86 	writing = uio->uio_rw == UIO_WRITE;
87 	reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
88 	    VM_PROT_READ;
89 
90 	kva = kmem_alloc_pageable(kernel_map, PAGE_SIZE);
91 
92 	/*
93 	 * Only map in one page at a time.  We don't have to, but it
94 	 * makes things easier.  This way is trivial - right?
95 	 */
96 	do {
97 		vm_map_t tmap;
98 		vm_offset_t uva;
99 		int page_offset;		/* offset into page */
100 		vm_map_entry_t out_entry;
101 		vm_prot_t out_prot;
102 		boolean_t wired;
103 		vm_pindex_t pindex;
104 		u_int len;
105 		vm_page_t m;
106 
107 		object = NULL;
108 
109 		uva = (vm_offset_t)uio->uio_offset;
110 
111 		/*
112 		 * Get the page number of this segment.
113 		 */
114 		pageno = trunc_page(uva);
115 		page_offset = uva - pageno;
116 
117 		/*
118 		 * How many bytes to copy
119 		 */
120 		len = min(PAGE_SIZE - page_offset, uio->uio_resid);
121 
122 		/*
123 		 * Fault the page on behalf of the process
124 		 */
125 		error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
126 		if (error) {
127 			error = EFAULT;
128 			break;
129 		}
130 
131 		/*
132 		 * Now we need to get the page.  out_entry, out_prot, wired,
133 		 * and single_use aren't used.  One would think the vm code
134 		 * would be a *bit* nicer...  We use tmap because
135 		 * vm_map_lookup() can change the map argument.
136 		 */
137 		tmap = map;
138 		error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
139 		    &object, &pindex, &out_prot, &wired);
140 
141 		if (error) {
142 			error = EFAULT;
143 
144 			/*
145 			 * Make sure that there is no residue in 'object' from
146 			 * an error return on vm_map_lookup.
147 			 */
148 			object = NULL;
149 
150 			break;
151 		}
152 
153 		m = vm_page_lookup(object, pindex);
154 
155 		/* Allow fallback to backing objects if we are reading */
156 
157 		while (m == NULL && !writing && object->backing_object) {
158 
159 			pindex += OFF_TO_IDX(object->backing_object_offset);
160 			object = object->backing_object;
161 
162 			m = vm_page_lookup(object, pindex);
163 		}
164 
165 		if (m == NULL) {
166 			error = EFAULT;
167 
168 			/*
169 			 * Make sure that there is no residue in 'object' from
170 			 * an error return on vm_map_lookup.
171 			 */
172 			object = NULL;
173 
174 			vm_map_lookup_done(tmap, out_entry);
175 
176 			break;
177 		}
178 
179 		/*
180 		 * Wire the page into memory
181 		 */
182 		vm_page_wire(m);
183 
184 		/*
185 		 * We're done with tmap now.
186 		 * But reference the object first, so that we won't loose
187 		 * it.
188 		 */
189 		vm_object_reference(object);
190 		vm_map_lookup_done(tmap, out_entry);
191 
192 		pmap_kenter(kva, VM_PAGE_TO_PHYS(m));
193 
194 		/*
195 		 * Now do the i/o move.
196 		 */
197 		error = uiomove((caddr_t)(kva + page_offset), len, uio);
198 
199 		pmap_kremove(kva);
200 
201 		/*
202 		 * release the page and the object
203 		 */
204 		vm_page_unwire(m, 1);
205 		vm_object_deallocate(object);
206 
207 		object = NULL;
208 
209 	} while (error == 0 && uio->uio_resid > 0);
210 
211 	if (object)
212 		vm_object_deallocate(object);
213 
214 	kmem_free(kernel_map, kva, PAGE_SIZE);
215 	vmspace_free(vm);
216 	return (error);
217 }
218 
219 /*
220  * Process debugging system call.
221  */
222 #ifndef _SYS_SYSPROTO_H_
223 struct ptrace_args {
224 	int	req;
225 	pid_t	pid;
226 	caddr_t	addr;
227 	int	data;
228 };
229 #endif
230 
231 int
232 ptrace(struct thread *td, struct ptrace_args *uap)
233 {
234 	struct proc *curp = td->td_proc;
235 	struct proc *p;
236 	struct iovec iov;
237 	struct uio uio;
238 	union {
239 		struct reg	reg;
240 		struct dbreg	dbreg;
241 		struct fpreg	fpreg;
242 	} r;
243 	int error = 0;
244 	int write;
245 
246 	write = 0;
247 	if (uap->req == PT_TRACE_ME) {
248 		p = curp;
249 		PROC_LOCK(p);
250 	} else {
251 		if ((p = pfind(uap->pid)) == NULL)
252 			return (ESRCH);
253 	}
254 	if (p_cansee(curp, p)) {
255 		PROC_UNLOCK(p);
256 		return (ESRCH);
257 	}
258 
259 	if ((error = p_candebug(curp, p)) != 0) {
260 		PROC_UNLOCK(p);
261 		return (error);
262 	}
263 
264 	/*
265 	 * Don't debug system processes!
266 	 */
267 	if ((p->p_flag & P_SYSTEM) != 0) {
268 		PROC_UNLOCK(p);
269 		return (EINVAL);
270 	}
271 
272 	/*
273 	 * Permissions check
274 	 */
275 	switch (uap->req) {
276 	case PT_TRACE_ME:
277 		/* Always legal. */
278 		break;
279 
280 	case PT_ATTACH:
281 		/* Self */
282 		if (p->p_pid == curp->p_pid) {
283 			PROC_UNLOCK(p);
284 			return (EINVAL);
285 		}
286 
287 		/* Already traced */
288 		if (p->p_flag & P_TRACED) {
289 			PROC_UNLOCK(p);
290 			return (EBUSY);
291 		}
292 
293 		/* OK */
294 		break;
295 
296 	case PT_READ_I:
297 	case PT_READ_D:
298 	case PT_WRITE_I:
299 	case PT_WRITE_D:
300 	case PT_CONTINUE:
301 	case PT_KILL:
302 	case PT_STEP:
303 	case PT_DETACH:
304 #ifdef PT_GETREGS
305 	case PT_GETREGS:
306 #endif
307 #ifdef PT_SETREGS
308 	case PT_SETREGS:
309 #endif
310 #ifdef PT_GETFPREGS
311 	case PT_GETFPREGS:
312 #endif
313 #ifdef PT_SETFPREGS
314 	case PT_SETFPREGS:
315 #endif
316 #ifdef PT_GETDBREGS
317 	case PT_GETDBREGS:
318 #endif
319 #ifdef PT_SETDBREGS
320 	case PT_SETDBREGS:
321 #endif
322 		/* not being traced... */
323 		if ((p->p_flag & P_TRACED) == 0) {
324 			PROC_UNLOCK(p);
325 			return (EPERM);
326 		}
327 
328 		/* not being traced by YOU */
329 		if (p->p_pptr != curp) {
330 			PROC_UNLOCK(p);
331 			return (EBUSY);
332 		}
333 
334 		/* not currently stopped */
335 		mtx_lock_spin(&sched_lock);
336 		if (p->p_stat != SSTOP || (p->p_flag & P_WAITED) == 0) {
337 			mtx_unlock_spin(&sched_lock);
338 			PROC_UNLOCK(p);
339 			return (EBUSY);
340 		}
341 		mtx_unlock_spin(&sched_lock);
342 
343 		/* OK */
344 		break;
345 
346 	default:
347 		PROC_UNLOCK(p);
348 		return (EINVAL);
349 	}
350 
351 	PROC_UNLOCK(p);
352 #ifdef FIX_SSTEP
353 	/*
354 	 * Single step fixup ala procfs
355 	 */
356 	FIX_SSTEP(&p->p_thread);	/* XXXKSE */
357 #endif
358 
359 	/*
360 	 * Actually do the requests
361 	 */
362 
363 	td->td_retval[0] = 0;
364 
365 	switch (uap->req) {
366 	case PT_TRACE_ME:
367 		/* set my trace flag and "owner" so it can read/write me */
368 		sx_xlock(&proctree_lock);
369 		PROC_LOCK(p);
370 		p->p_flag |= P_TRACED;
371 		p->p_oppid = p->p_pptr->p_pid;
372 		PROC_UNLOCK(p);
373 		sx_xunlock(&proctree_lock);
374 		return (0);
375 
376 	case PT_ATTACH:
377 		/* security check done above */
378 		sx_xlock(&proctree_lock);
379 		PROC_LOCK(p);
380 		p->p_flag |= P_TRACED;
381 		p->p_oppid = p->p_pptr->p_pid;
382 		if (p->p_pptr != curp)
383 			proc_reparent(p, curp);
384 		PROC_UNLOCK(p);
385 		sx_xunlock(&proctree_lock);
386 		uap->data = SIGSTOP;
387 		goto sendsig;	/* in PT_CONTINUE below */
388 
389 	case PT_STEP:
390 	case PT_CONTINUE:
391 	case PT_DETACH:
392 		if ((uap->req != PT_STEP) && ((unsigned)uap->data >= NSIG))
393 			return (EINVAL);
394 
395 		PHOLD(p);
396 
397 		if (uap->req == PT_STEP) {
398 			if ((error = ptrace_single_step(&p->p_thread))) {
399 				PRELE(p);
400 				return (error);
401 			}
402 		}
403 
404 		if (uap->addr != (caddr_t)1) {
405 			fill_kinfo_proc(p, &p->p_uarea->u_kproc);
406 			if ((error = ptrace_set_pc(&p->p_thread,
407 			    (u_long)(uintfptr_t)uap->addr))) {
408 				PRELE(p);
409 				return (error);
410 			}
411 		}
412 		PRELE(p);
413 
414 		if (uap->req == PT_DETACH) {
415 			/* reset process parent */
416 			sx_xlock(&proctree_lock);
417 			if (p->p_oppid != p->p_pptr->p_pid) {
418 				struct proc *pp;
419 
420 				pp = pfind(p->p_oppid);
421 				if (pp != NULL)
422 					PROC_UNLOCK(pp);
423 				else
424 					pp = initproc;
425 				PROC_LOCK(p);
426 				proc_reparent(p, pp);
427 			} else
428 				PROC_LOCK(p);
429 			p->p_flag &= ~(P_TRACED | P_WAITED);
430 			p->p_oppid = 0;
431 
432 			PROC_UNLOCK(p);
433 			sx_xunlock(&proctree_lock);
434 
435 			/* should we send SIGCHLD? */
436 
437 		}
438 
439 	sendsig:
440 		/* deliver or queue signal */
441 		PROC_LOCK(p);
442 		mtx_lock_spin(&sched_lock);
443 		if (p->p_stat == SSTOP) {
444 			p->p_xstat = uap->data;
445 			setrunnable(&p->p_thread); /* XXXKSE */
446 			mtx_unlock_spin(&sched_lock);
447 		} else {
448 			mtx_unlock_spin(&sched_lock);
449 			if (uap->data)
450 				psignal(p, uap->data);
451 
452 		}
453 		PROC_UNLOCK(p);
454 		return (0);
455 
456 	case PT_WRITE_I:
457 	case PT_WRITE_D:
458 		write = 1;
459 		/* fallthrough */
460 	case PT_READ_I:
461 	case PT_READ_D:
462 		/* write = 0 set above */
463 		iov.iov_base = write ? (caddr_t)&uap->data :
464 		    (caddr_t)td->td_retval;
465 		iov.iov_len = sizeof(int);
466 		uio.uio_iov = &iov;
467 		uio.uio_iovcnt = 1;
468 		uio.uio_offset = (off_t)(uintptr_t)uap->addr;
469 		uio.uio_resid = sizeof(int);
470 		uio.uio_segflg = UIO_SYSSPACE;	/* ie: the uap */
471 		uio.uio_rw = write ? UIO_WRITE : UIO_READ;
472 		uio.uio_td = td;
473 		error = proc_rwmem(p, &uio);
474 		if (uio.uio_resid != 0) {
475 			/*
476 			 * XXX proc_rwmem() doesn't currently return ENOSPC,
477 			 * so I think write() can bogusly return 0.
478 			 * XXX what happens for short writes?  We don't want
479 			 * to write partial data.
480 			 * XXX proc_rwmem() returns EPERM for other invalid
481 			 * addresses.  Convert this to EINVAL.  Does this
482 			 * clobber returns of EPERM for other reasons?
483 			 */
484 			if (error == 0 || error == ENOSPC || error == EPERM)
485 				error = EINVAL;	/* EOF */
486 		}
487 		return (error);
488 
489 	case PT_KILL:
490 		uap->data = SIGKILL;
491 		goto sendsig;	/* in PT_CONTINUE above */
492 
493 #ifdef PT_SETREGS
494 	case PT_SETREGS:
495 		error = copyin(uap->addr, &r.reg, sizeof r.reg);
496 		if (error == 0) {
497 			PHOLD(p);
498 			error = procfs_write_regs(&p->p_thread, &r.reg);
499 			PRELE(p);
500 		}
501 		return (error);
502 #endif /* PT_SETREGS */
503 
504 #ifdef PT_GETREGS
505 	case PT_GETREGS:
506 		PHOLD(p);
507 		error = procfs_read_regs(&p->p_thread, &r.reg);
508 		PRELE(p);
509 		if (error == 0)
510 			error = copyout(&r.reg, uap->addr, sizeof r.reg);
511 		return (error);
512 #endif /* PT_SETREGS */
513 
514 #ifdef PT_SETFPREGS
515 	case PT_SETFPREGS:
516 		error = copyin(uap->addr, &r.fpreg, sizeof r.fpreg);
517 		if (error == 0) {
518 			PHOLD(p);
519 			error = procfs_write_fpregs(&p->p_thread, &r.fpreg);
520 			PRELE(p);
521 		}
522 		return (error);
523 #endif /* PT_SETFPREGS */
524 
525 #ifdef PT_GETFPREGS
526 	case PT_GETFPREGS:
527 		PHOLD(p);
528 		error = procfs_read_fpregs(&p->p_thread, &r.fpreg);
529 		PRELE(p);
530 		if (error == 0)
531 			error = copyout(&r.fpreg, uap->addr, sizeof r.fpreg);
532 		return (error);
533 #endif /* PT_SETFPREGS */
534 
535 #ifdef PT_SETDBREGS
536 	case PT_SETDBREGS:
537 		error = copyin(uap->addr, &r.dbreg, sizeof r.dbreg);
538 		if (error == 0) {
539 			PHOLD(p);
540 			error = procfs_write_dbregs(&p->p_thread, &r.dbreg);
541 			PRELE(p);
542 		}
543 		return (error);
544 #endif /* PT_SETDBREGS */
545 
546 #ifdef PT_GETDBREGS
547 	case PT_GETDBREGS:
548 		PHOLD(p);
549 		error = procfs_read_dbregs(&p->p_thread, &r.dbreg);
550 		PRELE(p);
551 		if (error == 0)
552 			error = copyout(&r.dbreg, uap->addr, sizeof r.dbreg);
553 		return (error);
554 #endif /* PT_SETDBREGS */
555 
556 	default:
557 		KASSERT(0, ("unreachable code\n"));
558 		break;
559 	}
560 
561 	KASSERT(0, ("unreachable code\n"));
562 	return (0);
563 }
564 
565 int
566 trace_req(struct proc *p)
567 {
568 	return (1);
569 }
570 
571 /*
572  * stopevent()
573  * Stop a process because of a debugging event;
574  * stay stopped until p->p_step is cleared
575  * (cleared by PIOCCONT in procfs).
576  *
577  * Must be called with the proc struct mutex held.
578  */
579 
580 void
581 stopevent(struct proc *p, unsigned int event, unsigned int val)
582 {
583 
584 	PROC_LOCK_ASSERT(p, MA_OWNED | MA_NOTRECURSED);
585 	p->p_step = 1;
586 
587 	do {
588 		p->p_xstat = val;
589 		p->p_stype = event;	/* Which event caused the stop? */
590 		wakeup(&p->p_stype);	/* Wake up any PIOCWAIT'ing procs */
591 		msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
592 	} while (p->p_step);
593 }
594