xref: /freebsd/sys/compat/linux/linux_fork.c (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e) 
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 "opt_compat.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/imgact.h>
39 #include <sys/ktr.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/proc.h>
43 #include <sys/ptrace.h>
44 #include <sys/racct.h>
45 #include <sys/sched.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sx.h>
48 #include <sys/umtxvar.h>
49 #include <sys/unistd.h>
50 #include <sys/wait.h>
51 
52 #include <vm/vm.h>
53 #include <vm/pmap.h>
54 #include <vm/vm_map.h>
55 
56 #ifdef COMPAT_LINUX32
57 #include <machine/../linux32/linux.h>
58 #include <machine/../linux32/linux32_proto.h>
59 #else
60 #include <machine/../linux/linux.h>
61 #include <machine/../linux/linux_proto.h>
62 #endif
63 #include <compat/linux/linux.h>
64 #include <compat/linux/linux_emul.h>
65 #include <compat/linux/linux_fork.h>
66 #include <compat/linux/linux_futex.h>
67 #include <compat/linux/linux_mib.h>
68 #include <compat/linux/linux_misc.h>
69 #include <compat/linux/linux_util.h>
70 
71 #ifdef LINUX_LEGACY_SYSCALLS
72 int
73 linux_fork(struct thread *td, struct linux_fork_args *args)
74 {
75 	struct fork_req fr;
76 	int error;
77 	struct proc *p2;
78 	struct thread *td2;
79 
80 	bzero(&fr, sizeof(fr));
81 	fr.fr_flags = RFFDG | RFPROC | RFSTOPPED;
82 	fr.fr_procp = &p2;
83 	if ((error = fork1(td, &fr)) != 0)
84 		return (error);
85 
86 	td2 = FIRST_THREAD_IN_PROC(p2);
87 
88 	linux_proc_init(td, td2, false);
89 
90 	td->td_retval[0] = p2->p_pid;
91 
92 	/*
93 	 * Make this runnable after we are finished with it.
94 	 */
95 	thread_lock(td2);
96 	TD_SET_CAN_RUN(td2);
97 	sched_add(td2, SRQ_BORING);
98 
99 	return (0);
100 }
101 
102 int
103 linux_vfork(struct thread *td, struct linux_vfork_args *args)
104 {
105 	struct fork_req fr;
106 	int error;
107 	struct proc *p2;
108 	struct thread *td2;
109 
110 	bzero(&fr, sizeof(fr));
111 	fr.fr_flags = RFFDG | RFPROC | RFMEM | RFPPWAIT | RFSTOPPED;
112 	fr.fr_procp = &p2;
113 	if ((error = fork1(td, &fr)) != 0)
114 		return (error);
115 
116 	td2 = FIRST_THREAD_IN_PROC(p2);
117 
118 	linux_proc_init(td, td2, false);
119 
120 	td->td_retval[0] = p2->p_pid;
121 
122 	/*
123 	 * Make this runnable after we are finished with it.
124 	 */
125 	thread_lock(td2);
126 	TD_SET_CAN_RUN(td2);
127 	sched_add(td2, SRQ_BORING);
128 
129 	return (0);
130 }
131 #endif
132 
133 static int
134 linux_clone_proc(struct thread *td, struct l_clone_args *args)
135 {
136 	struct fork_req fr;
137 	int error, ff, f2;
138 	struct proc *p2;
139 	struct thread *td2;
140 	int exit_signal;
141 	struct linux_emuldata *em;
142 
143 	f2 = 0;
144 	ff = RFPROC | RFSTOPPED;
145 	if (LINUX_SIG_VALID(args->exit_signal)) {
146 		exit_signal = linux_to_bsd_signal(args->exit_signal);
147 	} else if (args->exit_signal != 0)
148 		return (EINVAL);
149 	else
150 		exit_signal = 0;
151 
152 	if (args->flags & LINUX_CLONE_VM)
153 		ff |= RFMEM;
154 	if (args->flags & LINUX_CLONE_SIGHAND)
155 		ff |= RFSIGSHARE;
156 	if ((args->flags & LINUX_CLONE_CLEAR_SIGHAND) != 0)
157 		f2 |= FR2_DROPSIG_CAUGHT;
158 	if (args->flags & LINUX_CLONE_FILES) {
159 		if (!(args->flags & LINUX_CLONE_FS))
160 			f2 |= FR2_SHARE_PATHS;
161 	} else {
162 		ff |= RFFDG;
163 		if (args->flags & LINUX_CLONE_FS)
164 			f2 |= FR2_SHARE_PATHS;
165 	}
166 
167 	if (args->flags & LINUX_CLONE_PARENT_SETTID)
168 		if (args->parent_tid == NULL)
169 			return (EINVAL);
170 
171 	if (args->flags & LINUX_CLONE_VFORK)
172 		ff |= RFPPWAIT;
173 
174 	bzero(&fr, sizeof(fr));
175 	fr.fr_flags = ff;
176 	fr.fr_flags2 = f2;
177 	fr.fr_procp = &p2;
178 	error = fork1(td, &fr);
179 	if (error)
180 		return (error);
181 
182 	td2 = FIRST_THREAD_IN_PROC(p2);
183 
184 	/* create the emuldata */
185 	linux_proc_init(td, td2, false);
186 
187 	em = em_find(td2);
188 	KASSERT(em != NULL, ("clone_proc: emuldata not found.\n"));
189 
190 	if (args->flags & LINUX_CLONE_CHILD_SETTID)
191 		em->child_set_tid = args->child_tid;
192 	else
193 		em->child_set_tid = NULL;
194 
195 	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
196 		em->child_clear_tid = args->child_tid;
197 	else
198 		em->child_clear_tid = NULL;
199 
200 	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
201 		error = copyout(&p2->p_pid, args->parent_tid,
202 		    sizeof(p2->p_pid));
203 		if (error)
204 			linux_msg(td, "copyout p_pid failed!");
205 	}
206 
207 	PROC_LOCK(p2);
208 	p2->p_sigparent = exit_signal;
209 	PROC_UNLOCK(p2);
210 	/*
211 	 * In a case of stack = NULL, we are supposed to COW calling process
212 	 * stack. This is what normal fork() does, so we just keep tf_rsp arg
213 	 * intact.
214 	 */
215 	linux_set_upcall(td2, args->stack);
216 
217 	if (args->flags & LINUX_CLONE_SETTLS)
218 		linux_set_cloned_tls(td2, PTRIN(args->tls));
219 
220 	/*
221 	 * If CLONE_PARENT is set, then the parent of the new process will be
222 	 * the same as that of the calling process.
223 	 */
224 	if (args->flags & LINUX_CLONE_PARENT) {
225 		sx_xlock(&proctree_lock);
226 		PROC_LOCK(p2);
227 		proc_reparent(p2, td->td_proc->p_pptr, true);
228 		PROC_UNLOCK(p2);
229 		sx_xunlock(&proctree_lock);
230 	}
231 
232 	/*
233 	 * Make this runnable after we are finished with it.
234 	 */
235 	thread_lock(td2);
236 	TD_SET_CAN_RUN(td2);
237 	sched_add(td2, SRQ_BORING);
238 
239 	td->td_retval[0] = p2->p_pid;
240 
241 	return (0);
242 }
243 
244 static int
245 linux_clone_thread(struct thread *td, struct l_clone_args *args)
246 {
247 	struct linux_emuldata *em;
248 	struct thread *newtd;
249 	struct proc *p;
250 	int error;
251 
252 	LINUX_CTR4(clone_thread, "thread(%d) flags %x ptid %p ctid %p",
253 	    td->td_tid, (unsigned)args->flags,
254 	    args->parent_tid, args->child_tid);
255 
256 	if ((args->flags & LINUX_CLONE_PARENT) != 0)
257 		return (EINVAL);
258 	if (args->flags & LINUX_CLONE_PARENT_SETTID)
259 		if (args->parent_tid == NULL)
260 			return (EINVAL);
261 
262 	/* Threads should be created with own stack */
263 	if (PTRIN(args->stack) == NULL)
264 		return (EINVAL);
265 
266 	p = td->td_proc;
267 
268 #ifdef RACCT
269 	if (racct_enable) {
270 		PROC_LOCK(p);
271 		error = racct_add(p, RACCT_NTHR, 1);
272 		PROC_UNLOCK(p);
273 		if (error != 0)
274 			return (EPROCLIM);
275 	}
276 #endif
277 
278 	/* Initialize our td */
279 	error = kern_thr_alloc(p, 0, &newtd);
280 	if (error)
281 		goto fail;
282 
283 	bzero(&newtd->td_startzero,
284 	    __rangeof(struct thread, td_startzero, td_endzero));
285 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
286 	    __rangeof(struct thread, td_startcopy, td_endcopy));
287 
288 	newtd->td_proc = p;
289 	thread_cow_get(newtd, td);
290 
291 	cpu_copy_thread(newtd, td);
292 
293 	/* create the emuldata */
294 	linux_proc_init(td, newtd, true);
295 
296 	em = em_find(newtd);
297 	KASSERT(em != NULL, ("clone_thread: emuldata not found.\n"));
298 
299 	if (args->flags & LINUX_CLONE_SETTLS)
300 		linux_set_cloned_tls(newtd, PTRIN(args->tls));
301 
302 	if (args->flags & LINUX_CLONE_CHILD_SETTID)
303 		em->child_set_tid = args->child_tid;
304 	else
305 		em->child_set_tid = NULL;
306 
307 	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
308 		em->child_clear_tid = args->child_tid;
309 	else
310 		em->child_clear_tid = NULL;
311 
312 	cpu_thread_clean(newtd);
313 
314 	linux_set_upcall(newtd, args->stack);
315 
316 	PROC_LOCK(p);
317 	p->p_flag |= P_HADTHREADS;
318 	thread_link(newtd, p);
319 	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
320 
321 	thread_lock(td);
322 	/* let the scheduler know about these things. */
323 	sched_fork_thread(td, newtd);
324 	thread_unlock(td);
325 	if (P_SHOULDSTOP(p))
326 		ast_sched(newtd, TDA_SUSPEND);
327 
328 	if (p->p_ptevents & PTRACE_LWP)
329 		newtd->td_dbgflags |= TDB_BORN;
330 	PROC_UNLOCK(p);
331 
332 	tidhash_add(newtd);
333 
334 	LINUX_CTR2(clone_thread, "thread(%d) successful clone to %d",
335 	    td->td_tid, newtd->td_tid);
336 
337 	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
338 		error = copyout(&newtd->td_tid, args->parent_tid,
339 		    sizeof(newtd->td_tid));
340 		if (error)
341 			linux_msg(td, "clone_thread: copyout td_tid failed!");
342 	}
343 
344 	/*
345 	 * Make this runnable after we are finished with it.
346 	 */
347 	thread_lock(newtd);
348 	TD_SET_CAN_RUN(newtd);
349 	sched_add(newtd, SRQ_BORING);
350 
351 	td->td_retval[0] = newtd->td_tid;
352 
353 	return (0);
354 
355 fail:
356 #ifdef RACCT
357 	if (racct_enable) {
358 		PROC_LOCK(p);
359 		racct_sub(p, RACCT_NTHR, 1);
360 		PROC_UNLOCK(p);
361 	}
362 #endif
363 	return (error);
364 }
365 
366 int
367 linux_clone(struct thread *td, struct linux_clone_args *args)
368 {
369 	struct l_clone_args ca = {
370 		.flags = (lower_32_bits(args->flags) & ~LINUX_CSIGNAL),
371 		.child_tid = args->child_tidptr,
372 		.parent_tid = args->parent_tidptr,
373 		.exit_signal = (lower_32_bits(args->flags) & LINUX_CSIGNAL),
374 		.stack = args->stack,
375 		.tls = args->tls,
376 	};
377 
378 	if (args->flags & LINUX_CLONE_THREAD)
379 		return (linux_clone_thread(td, &ca));
380 	else
381 		return (linux_clone_proc(td, &ca));
382 }
383 
384 
385 static int
386 linux_clone3_args_valid(struct l_user_clone_args *uca)
387 {
388 
389 	/* Verify that no unknown flags are passed along. */
390 	if ((uca->flags & ~(LINUX_CLONE_LEGACY_FLAGS |
391 	    LINUX_CLONE_CLEAR_SIGHAND | LINUX_CLONE_INTO_CGROUP)) != 0)
392 		return (EINVAL);
393 	if ((uca->flags & (LINUX_CLONE_DETACHED | LINUX_CSIGNAL)) != 0)
394 		return (EINVAL);
395 
396 	if ((uca->flags & (LINUX_CLONE_SIGHAND | LINUX_CLONE_CLEAR_SIGHAND)) ==
397 	    (LINUX_CLONE_SIGHAND | LINUX_CLONE_CLEAR_SIGHAND))
398 		return (EINVAL);
399 	if ((uca->flags & (LINUX_CLONE_THREAD | LINUX_CLONE_PARENT)) != 0 &&
400 	    uca->exit_signal != 0)
401 		return (EINVAL);
402 
403 	/* We don't support set_tid, only validate input. */
404 	if (uca->set_tid_size > LINUX_MAX_PID_NS_LEVEL)
405 		return (EINVAL);
406 	if (uca->set_tid == 0 && uca->set_tid_size > 0)
407 		return (EINVAL);
408 	if (uca->set_tid != 0 && uca->set_tid_size == 0)
409 		return (EINVAL);
410 
411 	if (uca->stack == 0 && uca->stack_size > 0)
412 		return (EINVAL);
413 	if (uca->stack != 0 && uca->stack_size == 0)
414 		return (EINVAL);
415 
416 	/* Verify that higher 32bits of exit_signal are unset. */
417 	if ((uca->exit_signal & ~(uint64_t)LINUX_CSIGNAL) != 0)
418 		return (EINVAL);
419 
420 	/* Verify that no unsupported flags are passed along. */
421 	if ((uca->flags & LINUX_CLONE_NEWTIME) != 0) {
422 		LINUX_RATELIMIT_MSG("unsupported clone3 option CLONE_NEWTIME");
423 		return (ENOSYS);
424 	}
425 	if ((uca->flags & LINUX_CLONE_INTO_CGROUP) != 0) {
426 		LINUX_RATELIMIT_MSG("unsupported clone3 option CLONE_INTO_CGROUP");
427 		return (ENOSYS);
428 	}
429 	if (uca->set_tid != 0 || uca->set_tid_size != 0) {
430 		LINUX_RATELIMIT_MSG("unsupported clone3 set_tid");
431 		return (ENOSYS);
432 	}
433 
434 	return (0);
435 }
436 
437 int
438 linux_clone3(struct thread *td, struct linux_clone3_args *args)
439 {
440 	struct l_user_clone_args *uca;
441 	struct l_clone_args *ca;
442 	size_t size;
443 	int error;
444 
445 	if (args->usize > PAGE_SIZE)
446 		return (E2BIG);
447 	if (args->usize < LINUX_CLONE_ARGS_SIZE_VER0)
448 		return (EINVAL);
449 
450 	/*
451 	 * usize can be less than size of struct clone_args, to avoid using
452 	 * of uninitialized data of struct clone_args, allocate at least
453 	 * sizeof(struct clone_args) storage and zero it.
454 	 */
455 	size = max(args->usize, sizeof(*uca));
456 	uca = malloc(size, M_LINUX, M_WAITOK | M_ZERO);
457 	error = copyin(args->uargs, uca, args->usize);
458 	if (error != 0)
459 		goto out;
460 	error = linux_clone3_args_valid(uca);
461 	if (error != 0)
462 		goto out;
463 	ca = malloc(sizeof(*ca), M_LINUX, M_WAITOK | M_ZERO);
464 	ca->flags = uca->flags;
465 	ca->child_tid = PTRIN(uca->child_tid);
466 	ca->parent_tid = PTRIN(uca->parent_tid);
467 	ca->exit_signal = uca->exit_signal;
468 	ca->stack = uca->stack + uca->stack_size;
469 	ca->stack_size = uca->stack_size;
470 	ca->tls = uca->tls;
471 
472 	if ((ca->flags & LINUX_CLONE_THREAD) != 0)
473 		error = linux_clone_thread(td, ca);
474 	else
475 		error = linux_clone_proc(td, ca);
476 	free(ca, M_LINUX);
477 out:
478 	free(uca, M_LINUX);
479 	return (error);
480 }
481 
482 int
483 linux_exit(struct thread *td, struct linux_exit_args *args)
484 {
485 	struct linux_emuldata *em __diagused;
486 
487 	em = em_find(td);
488 	KASSERT(em != NULL, ("exit: emuldata not found.\n"));
489 
490 	LINUX_CTR2(exit, "thread(%d) (%d)", em->em_tid, args->rval);
491 
492 	linux_thread_detach(td);
493 
494 	/*
495 	 * XXX. When the last two threads of a process
496 	 * exit via pthread_exit() try thr_exit() first.
497 	 */
498 	kern_thr_exit(td);
499 	exit1(td, args->rval, 0);
500 		/* NOTREACHED */
501 }
502 
503 int
504 linux_set_tid_address(struct thread *td, struct linux_set_tid_address_args *args)
505 {
506 	struct linux_emuldata *em;
507 
508 	em = em_find(td);
509 	KASSERT(em != NULL, ("set_tid_address: emuldata not found.\n"));
510 
511 	em->child_clear_tid = args->tidptr;
512 
513 	td->td_retval[0] = em->em_tid;
514 
515 	LINUX_CTR3(set_tid_address, "tidptr(%d) %p, returns %d",
516 	    em->em_tid, args->tidptr, td->td_retval[0]);
517 
518 	return (0);
519 }
520 
521 void
522 linux_thread_detach(struct thread *td)
523 {
524 	struct linux_emuldata *em;
525 	int *child_clear_tid;
526 	int error;
527 
528 	em = em_find(td);
529 	KASSERT(em != NULL, ("thread_detach: emuldata not found.\n"));
530 
531 	LINUX_CTR1(thread_detach, "thread(%d)", em->em_tid);
532 
533 	release_futexes(td, em);
534 
535 	child_clear_tid = em->child_clear_tid;
536 
537 	if (child_clear_tid != NULL) {
538 		LINUX_CTR2(thread_detach, "thread(%d) %p",
539 		    em->em_tid, child_clear_tid);
540 
541 		error = suword32(child_clear_tid, 0);
542 		if (error != 0)
543 			return;
544 
545 		error = futex_wake(td, child_clear_tid, 1, false);
546 		/*
547 		 * this cannot happen at the moment and if this happens it
548 		 * probably means there is a user space bug
549 		 */
550 		if (error != 0)
551 			linux_msg(td, "futex stuff in thread_detach failed.");
552 	}
553 
554 	/*
555 	 * Do not rely on the robust list which is maintained by userspace,
556 	 * cleanup remaining pi (if any) after release_futexes anyway.
557 	 */
558 	umtx_thread_exit(td);
559 }
560