xref: /freebsd/sys/compat/linux/linux_fork.c (revision bdd1243d) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2004 Tim J. Robbins
5  * Copyright (c) 2002 Doug Rabson
6  * Copyright (c) 2000 Marcel Moolenaar
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer
14  *    in this position and unchanged.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/ktr.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/proc.h>
39 #include <sys/ptrace.h>
40 #include <sys/racct.h>
41 #include <sys/sched.h>
42 #include <sys/syscallsubr.h>
43 #include <sys/sx.h>
44 #include <sys/umtxvar.h>
45 #include <sys/unistd.h>
46 
47 #include <vm/vm.h>
48 #include <vm/pmap.h>
49 #include <vm/vm_map.h>
50 
51 #ifdef COMPAT_LINUX32
52 #include <machine/../linux32/linux.h>
53 #include <machine/../linux32/linux32_proto.h>
54 #else
55 #include <machine/../linux/linux.h>
56 #include <machine/../linux/linux_proto.h>
57 #endif
58 #include <compat/linux/linux.h>
59 #include <compat/linux/linux_emul.h>
60 #include <compat/linux/linux_fork.h>
61 #include <compat/linux/linux_futex.h>
62 #include <compat/linux/linux_mib.h>
63 #include <compat/linux/linux_misc.h>
64 #include <compat/linux/linux_util.h>
65 
66 #ifdef LINUX_LEGACY_SYSCALLS
67 int
68 linux_fork(struct thread *td, struct linux_fork_args *args)
69 {
70 	struct fork_req fr;
71 	int error;
72 	struct proc *p2;
73 	struct thread *td2;
74 
75 	bzero(&fr, sizeof(fr));
76 	fr.fr_flags = RFFDG | RFPROC | RFSTOPPED;
77 	fr.fr_procp = &p2;
78 	if ((error = fork1(td, &fr)) != 0)
79 		return (error);
80 
81 	td2 = FIRST_THREAD_IN_PROC(p2);
82 
83 	linux_proc_init(td, td2, false);
84 
85 	td->td_retval[0] = p2->p_pid;
86 
87 	/*
88 	 * Make this runnable after we are finished with it.
89 	 */
90 	thread_lock(td2);
91 	TD_SET_CAN_RUN(td2);
92 	sched_add(td2, SRQ_BORING);
93 
94 	return (0);
95 }
96 
97 int
98 linux_vfork(struct thread *td, struct linux_vfork_args *args)
99 {
100 	struct fork_req fr;
101 	int error;
102 	struct proc *p2;
103 	struct thread *td2;
104 
105 	bzero(&fr, sizeof(fr));
106 	fr.fr_flags = RFFDG | RFPROC | RFMEM | RFPPWAIT | RFSTOPPED;
107 	fr.fr_procp = &p2;
108 	if ((error = fork1(td, &fr)) != 0)
109 		return (error);
110 
111 	td2 = FIRST_THREAD_IN_PROC(p2);
112 
113 	linux_proc_init(td, td2, false);
114 
115 	td->td_retval[0] = p2->p_pid;
116 
117 	/*
118 	 * Make this runnable after we are finished with it.
119 	 */
120 	thread_lock(td2);
121 	TD_SET_CAN_RUN(td2);
122 	sched_add(td2, SRQ_BORING);
123 
124 	return (0);
125 }
126 #endif
127 
128 static int
129 linux_clone_proc(struct thread *td, struct l_clone_args *args)
130 {
131 	struct fork_req fr;
132 	int error, ff, f2;
133 	struct proc *p2;
134 	struct thread *td2;
135 	int exit_signal;
136 	struct linux_emuldata *em;
137 
138 	f2 = 0;
139 	ff = RFPROC | RFSTOPPED;
140 	if (LINUX_SIG_VALID(args->exit_signal)) {
141 		exit_signal = linux_to_bsd_signal(args->exit_signal);
142 	} else if (args->exit_signal != 0)
143 		return (EINVAL);
144 	else
145 		exit_signal = 0;
146 
147 	if (args->flags & LINUX_CLONE_VM)
148 		ff |= RFMEM;
149 	if (args->flags & LINUX_CLONE_SIGHAND)
150 		ff |= RFSIGSHARE;
151 	if ((args->flags & LINUX_CLONE_CLEAR_SIGHAND) != 0)
152 		f2 |= FR2_DROPSIG_CAUGHT;
153 	if (args->flags & LINUX_CLONE_FILES) {
154 		if (!(args->flags & LINUX_CLONE_FS))
155 			f2 |= FR2_SHARE_PATHS;
156 	} else {
157 		ff |= RFFDG;
158 		if (args->flags & LINUX_CLONE_FS)
159 			f2 |= FR2_SHARE_PATHS;
160 	}
161 
162 	if (args->flags & LINUX_CLONE_PARENT_SETTID)
163 		if (args->parent_tid == NULL)
164 			return (EINVAL);
165 
166 	if (args->flags & LINUX_CLONE_VFORK)
167 		ff |= RFPPWAIT;
168 
169 	bzero(&fr, sizeof(fr));
170 	fr.fr_flags = ff;
171 	fr.fr_flags2 = f2;
172 	fr.fr_procp = &p2;
173 	error = fork1(td, &fr);
174 	if (error)
175 		return (error);
176 
177 	td2 = FIRST_THREAD_IN_PROC(p2);
178 
179 	/* create the emuldata */
180 	linux_proc_init(td, td2, false);
181 
182 	em = em_find(td2);
183 	KASSERT(em != NULL, ("clone_proc: emuldata not found.\n"));
184 
185 	if (args->flags & LINUX_CLONE_CHILD_SETTID)
186 		em->child_set_tid = args->child_tid;
187 	else
188 		em->child_set_tid = NULL;
189 
190 	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
191 		em->child_clear_tid = args->child_tid;
192 	else
193 		em->child_clear_tid = NULL;
194 
195 	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
196 		error = copyout(&p2->p_pid, args->parent_tid,
197 		    sizeof(p2->p_pid));
198 		if (error)
199 			linux_msg(td, "copyout p_pid failed!");
200 	}
201 
202 	PROC_LOCK(p2);
203 	p2->p_sigparent = exit_signal;
204 	PROC_UNLOCK(p2);
205 	/*
206 	 * In a case of stack = NULL, we are supposed to COW calling process
207 	 * stack. This is what normal fork() does, so we just keep tf_rsp arg
208 	 * intact.
209 	 */
210 	linux_set_upcall(td2, args->stack);
211 
212 	if (args->flags & LINUX_CLONE_SETTLS)
213 		linux_set_cloned_tls(td2, PTRIN(args->tls));
214 
215 	/*
216 	 * If CLONE_PARENT is set, then the parent of the new process will be
217 	 * the same as that of the calling process.
218 	 */
219 	if (args->flags & LINUX_CLONE_PARENT) {
220 		sx_xlock(&proctree_lock);
221 		PROC_LOCK(p2);
222 		proc_reparent(p2, td->td_proc->p_pptr, true);
223 		PROC_UNLOCK(p2);
224 		sx_xunlock(&proctree_lock);
225 	}
226 
227 	/*
228 	 * Make this runnable after we are finished with it.
229 	 */
230 	thread_lock(td2);
231 	TD_SET_CAN_RUN(td2);
232 	sched_add(td2, SRQ_BORING);
233 
234 	td->td_retval[0] = p2->p_pid;
235 
236 	return (0);
237 }
238 
239 static int
240 linux_clone_thread(struct thread *td, struct l_clone_args *args)
241 {
242 	struct linux_emuldata *em;
243 	struct thread *newtd;
244 	struct proc *p;
245 	int error;
246 
247 	LINUX_CTR4(clone_thread, "thread(%d) flags %x ptid %p ctid %p",
248 	    td->td_tid, (unsigned)args->flags,
249 	    args->parent_tid, args->child_tid);
250 
251 	if ((args->flags & LINUX_CLONE_PARENT) != 0)
252 		return (EINVAL);
253 	if (args->flags & LINUX_CLONE_PARENT_SETTID)
254 		if (args->parent_tid == NULL)
255 			return (EINVAL);
256 
257 	/* Threads should be created with own stack */
258 	if (PTRIN(args->stack) == NULL)
259 		return (EINVAL);
260 
261 	p = td->td_proc;
262 
263 #ifdef RACCT
264 	if (racct_enable) {
265 		PROC_LOCK(p);
266 		error = racct_add(p, RACCT_NTHR, 1);
267 		PROC_UNLOCK(p);
268 		if (error != 0)
269 			return (EPROCLIM);
270 	}
271 #endif
272 
273 	/* Initialize our td */
274 	error = kern_thr_alloc(p, 0, &newtd);
275 	if (error)
276 		goto fail;
277 
278 	bzero(&newtd->td_startzero,
279 	    __rangeof(struct thread, td_startzero, td_endzero));
280 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
281 	    __rangeof(struct thread, td_startcopy, td_endcopy));
282 
283 	newtd->td_proc = p;
284 	thread_cow_get(newtd, td);
285 
286 	cpu_copy_thread(newtd, td);
287 
288 	/* create the emuldata */
289 	linux_proc_init(td, newtd, true);
290 
291 	em = em_find(newtd);
292 	KASSERT(em != NULL, ("clone_thread: emuldata not found.\n"));
293 
294 	if (args->flags & LINUX_CLONE_SETTLS)
295 		linux_set_cloned_tls(newtd, PTRIN(args->tls));
296 
297 	if (args->flags & LINUX_CLONE_CHILD_SETTID)
298 		em->child_set_tid = args->child_tid;
299 	else
300 		em->child_set_tid = NULL;
301 
302 	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
303 		em->child_clear_tid = args->child_tid;
304 	else
305 		em->child_clear_tid = NULL;
306 
307 	cpu_thread_clean(newtd);
308 
309 	linux_set_upcall(newtd, args->stack);
310 
311 	PROC_LOCK(p);
312 	p->p_flag |= P_HADTHREADS;
313 	thread_link(newtd, p);
314 	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
315 
316 	thread_lock(td);
317 	/* let the scheduler know about these things. */
318 	sched_fork_thread(td, newtd);
319 	thread_unlock(td);
320 	if (P_SHOULDSTOP(p))
321 		ast_sched(newtd, TDA_SUSPEND);
322 
323 	if (p->p_ptevents & PTRACE_LWP)
324 		newtd->td_dbgflags |= TDB_BORN;
325 	PROC_UNLOCK(p);
326 
327 	tidhash_add(newtd);
328 
329 	LINUX_CTR2(clone_thread, "thread(%d) successful clone to %d",
330 	    td->td_tid, newtd->td_tid);
331 
332 	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
333 		error = copyout(&newtd->td_tid, args->parent_tid,
334 		    sizeof(newtd->td_tid));
335 		if (error)
336 			linux_msg(td, "clone_thread: copyout td_tid failed!");
337 	}
338 
339 	/*
340 	 * Make this runnable after we are finished with it.
341 	 */
342 	thread_lock(newtd);
343 	TD_SET_CAN_RUN(newtd);
344 	sched_add(newtd, SRQ_BORING);
345 
346 	td->td_retval[0] = newtd->td_tid;
347 
348 	return (0);
349 
350 fail:
351 #ifdef RACCT
352 	if (racct_enable) {
353 		PROC_LOCK(p);
354 		racct_sub(p, RACCT_NTHR, 1);
355 		PROC_UNLOCK(p);
356 	}
357 #endif
358 	return (error);
359 }
360 
361 int
362 linux_clone(struct thread *td, struct linux_clone_args *args)
363 {
364 	struct l_clone_args ca = {
365 		.flags = (lower_32_bits(args->flags) & ~LINUX_CSIGNAL),
366 		.child_tid = args->child_tidptr,
367 		.parent_tid = args->parent_tidptr,
368 		.exit_signal = (lower_32_bits(args->flags) & LINUX_CSIGNAL),
369 		.stack = args->stack,
370 		.tls = args->tls,
371 	};
372 
373 	if (args->flags & LINUX_CLONE_THREAD)
374 		return (linux_clone_thread(td, &ca));
375 	else
376 		return (linux_clone_proc(td, &ca));
377 }
378 
379 
380 static int
381 linux_clone3_args_valid(struct l_user_clone_args *uca)
382 {
383 
384 	/* Verify that no unknown flags are passed along. */
385 	if ((uca->flags & ~(LINUX_CLONE_LEGACY_FLAGS |
386 	    LINUX_CLONE_CLEAR_SIGHAND | LINUX_CLONE_INTO_CGROUP)) != 0)
387 		return (EINVAL);
388 	if ((uca->flags & (LINUX_CLONE_DETACHED | LINUX_CSIGNAL)) != 0)
389 		return (EINVAL);
390 
391 	if ((uca->flags & (LINUX_CLONE_SIGHAND | LINUX_CLONE_CLEAR_SIGHAND)) ==
392 	    (LINUX_CLONE_SIGHAND | LINUX_CLONE_CLEAR_SIGHAND))
393 		return (EINVAL);
394 	if ((uca->flags & (LINUX_CLONE_THREAD | LINUX_CLONE_PARENT)) != 0 &&
395 	    uca->exit_signal != 0)
396 		return (EINVAL);
397 
398 	/* We don't support set_tid, only validate input. */
399 	if (uca->set_tid_size > LINUX_MAX_PID_NS_LEVEL)
400 		return (EINVAL);
401 	if (uca->set_tid == 0 && uca->set_tid_size > 0)
402 		return (EINVAL);
403 	if (uca->set_tid != 0 && uca->set_tid_size == 0)
404 		return (EINVAL);
405 
406 	if (uca->stack == 0 && uca->stack_size > 0)
407 		return (EINVAL);
408 	if (uca->stack != 0 && uca->stack_size == 0)
409 		return (EINVAL);
410 
411 	/* Verify that higher 32bits of exit_signal are unset. */
412 	if ((uca->exit_signal & ~(uint64_t)LINUX_CSIGNAL) != 0)
413 		return (EINVAL);
414 
415 	/* Verify that no unsupported flags are passed along. */
416 	if ((uca->flags & LINUX_CLONE_NEWTIME) != 0) {
417 		LINUX_RATELIMIT_MSG("unsupported clone3 option CLONE_NEWTIME");
418 		return (ENOSYS);
419 	}
420 	if ((uca->flags & LINUX_CLONE_INTO_CGROUP) != 0) {
421 		LINUX_RATELIMIT_MSG("unsupported clone3 option CLONE_INTO_CGROUP");
422 		return (ENOSYS);
423 	}
424 	if (uca->set_tid != 0 || uca->set_tid_size != 0) {
425 		LINUX_RATELIMIT_MSG("unsupported clone3 set_tid");
426 		return (ENOSYS);
427 	}
428 
429 	return (0);
430 }
431 
432 int
433 linux_clone3(struct thread *td, struct linux_clone3_args *args)
434 {
435 	struct l_user_clone_args *uca;
436 	struct l_clone_args *ca;
437 	size_t size;
438 	int error;
439 
440 	if (args->usize > PAGE_SIZE)
441 		return (E2BIG);
442 	if (args->usize < LINUX_CLONE_ARGS_SIZE_VER0)
443 		return (EINVAL);
444 
445 	/*
446 	 * usize can be less than size of struct clone_args, to avoid using
447 	 * of uninitialized data of struct clone_args, allocate at least
448 	 * sizeof(struct clone_args) storage and zero it.
449 	 */
450 	size = max(args->usize, sizeof(*uca));
451 	uca = malloc(size, M_LINUX, M_WAITOK | M_ZERO);
452 	error = copyin(args->uargs, uca, args->usize);
453 	if (error != 0)
454 		goto out;
455 	error = linux_clone3_args_valid(uca);
456 	if (error != 0)
457 		goto out;
458 	ca = malloc(sizeof(*ca), M_LINUX, M_WAITOK | M_ZERO);
459 	ca->flags = uca->flags;
460 	ca->child_tid = PTRIN(uca->child_tid);
461 	ca->parent_tid = PTRIN(uca->parent_tid);
462 	ca->exit_signal = uca->exit_signal;
463 	ca->stack = uca->stack + uca->stack_size;
464 	ca->stack_size = uca->stack_size;
465 	ca->tls = uca->tls;
466 
467 	if ((ca->flags & LINUX_CLONE_THREAD) != 0)
468 		error = linux_clone_thread(td, ca);
469 	else
470 		error = linux_clone_proc(td, ca);
471 	free(ca, M_LINUX);
472 out:
473 	free(uca, M_LINUX);
474 	return (error);
475 }
476 
477 int
478 linux_exit(struct thread *td, struct linux_exit_args *args)
479 {
480 	struct linux_emuldata *em __diagused;
481 
482 	em = em_find(td);
483 	KASSERT(em != NULL, ("exit: emuldata not found.\n"));
484 
485 	LINUX_CTR2(exit, "thread(%d) (%d)", em->em_tid, args->rval);
486 
487 	linux_thread_detach(td);
488 
489 	/*
490 	 * XXX. When the last two threads of a process
491 	 * exit via pthread_exit() try thr_exit() first.
492 	 */
493 	kern_thr_exit(td);
494 	exit1(td, args->rval, 0);
495 		/* NOTREACHED */
496 }
497 
498 int
499 linux_set_tid_address(struct thread *td, struct linux_set_tid_address_args *args)
500 {
501 	struct linux_emuldata *em;
502 
503 	em = em_find(td);
504 	KASSERT(em != NULL, ("set_tid_address: emuldata not found.\n"));
505 
506 	em->child_clear_tid = args->tidptr;
507 
508 	td->td_retval[0] = em->em_tid;
509 
510 	LINUX_CTR3(set_tid_address, "tidptr(%d) %p, returns %d",
511 	    em->em_tid, args->tidptr, td->td_retval[0]);
512 
513 	return (0);
514 }
515 
516 void
517 linux_thread_detach(struct thread *td)
518 {
519 	struct linux_emuldata *em;
520 	int *child_clear_tid;
521 	int error;
522 
523 	em = em_find(td);
524 	KASSERT(em != NULL, ("thread_detach: emuldata not found.\n"));
525 
526 	LINUX_CTR1(thread_detach, "thread(%d)", em->em_tid);
527 
528 	release_futexes(td, em);
529 
530 	child_clear_tid = em->child_clear_tid;
531 
532 	if (child_clear_tid != NULL) {
533 		LINUX_CTR2(thread_detach, "thread(%d) %p",
534 		    em->em_tid, child_clear_tid);
535 
536 		error = suword32(child_clear_tid, 0);
537 		if (error != 0)
538 			return;
539 
540 		error = futex_wake(td, child_clear_tid, 1, false);
541 		/*
542 		 * this cannot happen at the moment and if this happens it
543 		 * probably means there is a user space bug
544 		 */
545 		if (error != 0)
546 			linux_msg(td, "futex stuff in thread_detach failed.");
547 	}
548 
549 	/*
550 	 * Do not rely on the robust list which is maintained by userspace,
551 	 * cleanup remaining pi (if any) after release_futexes anyway.
552 	 */
553 	umtx_thread_exit(td);
554 }
555