xref: /freebsd/sys/compat/linux/linux_fork.c (revision f39bffc62c1395bde25d152c7f68fdf7cbaab414) 
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/racct.h>
44 #include <sys/sched.h>
45 #include <sys/syscallsubr.h>
46 #include <sys/sx.h>
47 #include <sys/umtx.h>
48 #include <sys/unistd.h>
49 #include <sys/wait.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_map.h>
54 
55 #ifdef COMPAT_LINUX32
56 #include <machine/../linux32/linux.h>
57 #include <machine/../linux32/linux32_proto.h>
58 #else
59 #include <machine/../linux/linux.h>
60 #include <machine/../linux/linux_proto.h>
61 #endif
62 #include <compat/linux/linux_emul.h>
63 #include <compat/linux/linux_futex.h>
64 #include <compat/linux/linux_misc.h>
65 #include <compat/linux/linux_util.h>
66 
67 #ifdef LINUX_LEGACY_SYSCALLS
68 int
69 linux_fork(struct thread *td, struct linux_fork_args *args)
70 {
71 	struct fork_req fr;
72 	int error;
73 	struct proc *p2;
74 	struct thread *td2;
75 
76 #ifdef DEBUG
77 	if (ldebug(fork))
78 		printf(ARGS(fork, ""));
79 #endif
80 
81 	bzero(&fr, sizeof(fr));
82 	fr.fr_flags = RFFDG | RFPROC | RFSTOPPED;
83 	fr.fr_procp = &p2;
84 	if ((error = fork1(td, &fr)) != 0)
85 		return (error);
86 
87 	td2 = FIRST_THREAD_IN_PROC(p2);
88 
89 	linux_proc_init(td, td2, 0);
90 
91 	td->td_retval[0] = p2->p_pid;
92 
93 	/*
94 	 * Make this runnable after we are finished with it.
95 	 */
96 	thread_lock(td2);
97 	TD_SET_CAN_RUN(td2);
98 	sched_add(td2, SRQ_BORING);
99 	thread_unlock(td2);
100 
101 	return (0);
102 }
103 
104 int
105 linux_vfork(struct thread *td, struct linux_vfork_args *args)
106 {
107 	struct fork_req fr;
108 	int error;
109 	struct proc *p2;
110 	struct thread *td2;
111 
112 #ifdef DEBUG
113 	if (ldebug(vfork))
114 		printf(ARGS(vfork, ""));
115 #endif
116 
117 	bzero(&fr, sizeof(fr));
118 	fr.fr_flags = RFFDG | RFPROC | RFMEM | RFPPWAIT | RFSTOPPED;
119 	fr.fr_procp = &p2;
120 	if ((error = fork1(td, &fr)) != 0)
121 		return (error);
122 
123 	td2 = FIRST_THREAD_IN_PROC(p2);
124 
125 	linux_proc_init(td, td2, 0);
126 
127 	td->td_retval[0] = p2->p_pid;
128 
129 	/*
130 	 * Make this runnable after we are finished with it.
131 	 */
132 	thread_lock(td2);
133 	TD_SET_CAN_RUN(td2);
134 	sched_add(td2, SRQ_BORING);
135 	thread_unlock(td2);
136 
137 	return (0);
138 }
139 #endif
140 
141 static int
142 linux_clone_proc(struct thread *td, struct linux_clone_args *args)
143 {
144 	struct fork_req fr;
145 	int error, ff = RFPROC | RFSTOPPED;
146 	struct proc *p2;
147 	struct thread *td2;
148 	int exit_signal;
149 	struct linux_emuldata *em;
150 
151 #ifdef DEBUG
152 	if (ldebug(clone)) {
153 		printf(ARGS(clone, "flags %x, stack %p, parent tid: %p, "
154 		    "child tid: %p"), (unsigned)args->flags,
155 		    args->stack, args->parent_tidptr, args->child_tidptr);
156 	}
157 #endif
158 
159 	exit_signal = args->flags & 0x000000ff;
160 	if (LINUX_SIG_VALID(exit_signal)) {
161 		exit_signal = linux_to_bsd_signal(exit_signal);
162 	} else if (exit_signal != 0)
163 		return (EINVAL);
164 
165 	if (args->flags & LINUX_CLONE_VM)
166 		ff |= RFMEM;
167 	if (args->flags & LINUX_CLONE_SIGHAND)
168 		ff |= RFSIGSHARE;
169 	/*
170 	 * XXX: In Linux, sharing of fs info (chroot/cwd/umask)
171 	 * and open files is independent.  In FreeBSD, its in one
172 	 * structure but in reality it does not cause any problems
173 	 * because both of these flags are usually set together.
174 	 */
175 	if (!(args->flags & (LINUX_CLONE_FILES | LINUX_CLONE_FS)))
176 		ff |= RFFDG;
177 
178 	if (args->flags & LINUX_CLONE_PARENT_SETTID)
179 		if (args->parent_tidptr == NULL)
180 			return (EINVAL);
181 
182 	if (args->flags & LINUX_CLONE_VFORK)
183 		ff |= RFPPWAIT;
184 
185 	bzero(&fr, sizeof(fr));
186 	fr.fr_flags = ff;
187 	fr.fr_procp = &p2;
188 	error = fork1(td, &fr);
189 	if (error)
190 		return (error);
191 
192 	td2 = FIRST_THREAD_IN_PROC(p2);
193 
194 	/* create the emuldata */
195 	linux_proc_init(td, td2, args->flags);
196 
197 	em = em_find(td2);
198 	KASSERT(em != NULL, ("clone_proc: emuldata not found.\n"));
199 
200 	if (args->flags & LINUX_CLONE_CHILD_SETTID)
201 		em->child_set_tid = args->child_tidptr;
202 	else
203 		em->child_set_tid = NULL;
204 
205 	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
206 		em->child_clear_tid = args->child_tidptr;
207 	else
208 		em->child_clear_tid = NULL;
209 
210 	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
211 		error = copyout(&p2->p_pid, args->parent_tidptr,
212 		    sizeof(p2->p_pid));
213 		if (error)
214 			printf(LMSG("copyout failed!"));
215 	}
216 
217 	PROC_LOCK(p2);
218 	p2->p_sigparent = exit_signal;
219 	PROC_UNLOCK(p2);
220 	/*
221 	 * In a case of stack = NULL, we are supposed to COW calling process
222 	 * stack. This is what normal fork() does, so we just keep tf_rsp arg
223 	 * intact.
224 	 */
225 	linux_set_upcall_kse(td2, PTROUT(args->stack));
226 
227 	if (args->flags & LINUX_CLONE_SETTLS)
228 		linux_set_cloned_tls(td2, args->tls);
229 
230 	/*
231 	 * If CLONE_PARENT is set, then the parent of the new process will be
232 	 * the same as that of the calling process.
233 	 */
234 	if (args->flags & LINUX_CLONE_PARENT) {
235 		sx_xlock(&proctree_lock);
236 		PROC_LOCK(p2);
237 		proc_reparent(p2, td->td_proc->p_pptr);
238 		PROC_UNLOCK(p2);
239 		sx_xunlock(&proctree_lock);
240 	}
241 
242 #ifdef DEBUG
243 	if (ldebug(clone))
244 		printf(LMSG("clone: successful rfork to %d, "
245 		    "stack %p sig = %d"), (int)p2->p_pid, args->stack,
246 		    exit_signal);
247 #endif
248 
249 	/*
250 	 * Make this runnable after we are finished with it.
251 	 */
252 	thread_lock(td2);
253 	TD_SET_CAN_RUN(td2);
254 	sched_add(td2, SRQ_BORING);
255 	thread_unlock(td2);
256 
257 	td->td_retval[0] = p2->p_pid;
258 
259 	return (0);
260 }
261 
262 static int
263 linux_clone_thread(struct thread *td, struct linux_clone_args *args)
264 {
265 	struct linux_emuldata *em;
266 	struct thread *newtd;
267 	struct proc *p;
268 	int error;
269 
270 #ifdef DEBUG
271 	if (ldebug(clone)) {
272 		printf(ARGS(clone, "thread: flags %x, stack %p, parent tid: %p, "
273 		    "child tid: %p"), (unsigned)args->flags,
274 		    args->stack, args->parent_tidptr, args->child_tidptr);
275 	}
276 #endif
277 
278 	LINUX_CTR4(clone_thread, "thread(%d) flags %x ptid %p ctid %p",
279 	    td->td_tid, (unsigned)args->flags,
280 	    args->parent_tidptr, args->child_tidptr);
281 
282 	if (args->flags & LINUX_CLONE_PARENT_SETTID)
283 		if (args->parent_tidptr == NULL)
284 			return (EINVAL);
285 
286 	/* Threads should be created with own stack */
287 	if (args->stack == NULL)
288 		return (EINVAL);
289 
290 	p = td->td_proc;
291 
292 #ifdef RACCT
293 	if (racct_enable) {
294 		PROC_LOCK(p);
295 		error = racct_add(p, RACCT_NTHR, 1);
296 		PROC_UNLOCK(p);
297 		if (error != 0)
298 			return (EPROCLIM);
299 	}
300 #endif
301 
302 	/* Initialize our td */
303 	error = kern_thr_alloc(p, 0, &newtd);
304 	if (error)
305 		goto fail;
306 
307 	cpu_copy_thread(newtd, td);
308 
309 	bzero(&newtd->td_startzero,
310 	    __rangeof(struct thread, td_startzero, td_endzero));
311 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
312 	    __rangeof(struct thread, td_startcopy, td_endcopy));
313 
314 	newtd->td_proc = p;
315 	thread_cow_get(newtd, td);
316 
317 	/* create the emuldata */
318 	linux_proc_init(td, newtd, args->flags);
319 
320 	em = em_find(newtd);
321 	KASSERT(em != NULL, ("clone_thread: emuldata not found.\n"));
322 
323 	if (args->flags & LINUX_CLONE_SETTLS)
324 		linux_set_cloned_tls(newtd, args->tls);
325 
326 	if (args->flags & LINUX_CLONE_CHILD_SETTID)
327 		em->child_set_tid = args->child_tidptr;
328 	else
329 		em->child_set_tid = NULL;
330 
331 	if (args->flags & LINUX_CLONE_CHILD_CLEARTID)
332 		em->child_clear_tid = args->child_tidptr;
333 	else
334 		em->child_clear_tid = NULL;
335 
336 	cpu_thread_clean(newtd);
337 
338 	linux_set_upcall_kse(newtd, PTROUT(args->stack));
339 
340 	PROC_LOCK(p);
341 	p->p_flag |= P_HADTHREADS;
342 	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
343 
344 	if (args->flags & LINUX_CLONE_PARENT)
345 		thread_link(newtd, p->p_pptr);
346 	else
347 		thread_link(newtd, p);
348 
349 	thread_lock(td);
350 	/* let the scheduler know about these things. */
351 	sched_fork_thread(td, newtd);
352 	thread_unlock(td);
353 	if (P_SHOULDSTOP(p))
354 		newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
355 	PROC_UNLOCK(p);
356 
357 	tidhash_add(newtd);
358 
359 #ifdef DEBUG
360 	if (ldebug(clone))
361 		printf(ARGS(clone, "successful clone to %d, stack %p"),
362 		(int)newtd->td_tid, args->stack);
363 #endif
364 
365 	LINUX_CTR2(clone_thread, "thread(%d) successful clone to %d",
366 	    td->td_tid, newtd->td_tid);
367 
368 	if (args->flags & LINUX_CLONE_PARENT_SETTID) {
369 		error = copyout(&newtd->td_tid, args->parent_tidptr,
370 		    sizeof(newtd->td_tid));
371 		if (error)
372 			printf(LMSG("clone_thread: copyout failed!"));
373 	}
374 
375 	/*
376 	 * Make this runnable after we are finished with it.
377 	 */
378 	thread_lock(newtd);
379 	TD_SET_CAN_RUN(newtd);
380 	sched_add(newtd, SRQ_BORING);
381 	thread_unlock(newtd);
382 
383 	td->td_retval[0] = newtd->td_tid;
384 
385 	return (0);
386 
387 fail:
388 #ifdef RACCT
389 	if (racct_enable) {
390 		PROC_LOCK(p);
391 		racct_sub(p, RACCT_NTHR, 1);
392 		PROC_UNLOCK(p);
393 	}
394 #endif
395 	return (error);
396 }
397 
398 int
399 linux_clone(struct thread *td, struct linux_clone_args *args)
400 {
401 
402 	if (args->flags & LINUX_CLONE_THREAD)
403 		return (linux_clone_thread(td, args));
404 	else
405 		return (linux_clone_proc(td, args));
406 }
407 
408 int
409 linux_exit(struct thread *td, struct linux_exit_args *args)
410 {
411 	struct linux_emuldata *em;
412 
413 	em = em_find(td);
414 	KASSERT(em != NULL, ("exit: emuldata not found.\n"));
415 
416 	LINUX_CTR2(exit, "thread(%d) (%d)", em->em_tid, args->rval);
417 
418 	umtx_thread_exit(td);
419 
420 	linux_thread_detach(td);
421 
422 	/*
423 	 * XXX. When the last two threads of a process
424 	 * exit via pthread_exit() try thr_exit() first.
425 	 */
426 	kern_thr_exit(td);
427 	exit1(td, args->rval, 0);
428 		/* NOTREACHED */
429 }
430 
431 int
432 linux_set_tid_address(struct thread *td, struct linux_set_tid_address_args *args)
433 {
434 	struct linux_emuldata *em;
435 
436 	em = em_find(td);
437 	KASSERT(em != NULL, ("set_tid_address: emuldata not found.\n"));
438 
439 	em->child_clear_tid = args->tidptr;
440 
441 	td->td_retval[0] = em->em_tid;
442 
443 	LINUX_CTR3(set_tid_address, "tidptr(%d) %p, returns %d",
444 	    em->em_tid, args->tidptr, td->td_retval[0]);
445 
446 	return (0);
447 }
448 
449 void
450 linux_thread_detach(struct thread *td)
451 {
452 	struct linux_sys_futex_args cup;
453 	struct linux_emuldata *em;
454 	int *child_clear_tid;
455 	int error;
456 
457 	em = em_find(td);
458 	KASSERT(em != NULL, ("thread_detach: emuldata not found.\n"));
459 
460 	LINUX_CTR1(thread_detach, "thread(%d)", em->em_tid);
461 
462 	release_futexes(td, em);
463 
464 	child_clear_tid = em->child_clear_tid;
465 
466 	if (child_clear_tid != NULL) {
467 
468 		LINUX_CTR2(thread_detach, "thread(%d) %p",
469 		    em->em_tid, child_clear_tid);
470 
471 		error = suword32(child_clear_tid, 0);
472 		if (error != 0)
473 			return;
474 
475 		cup.uaddr = child_clear_tid;
476 		cup.op = LINUX_FUTEX_WAKE;
477 		cup.val = 1;		/* wake one */
478 		cup.timeout = NULL;
479 		cup.uaddr2 = NULL;
480 		cup.val3 = 0;
481 		error = linux_sys_futex(td, &cup);
482 		/*
483 		 * this cannot happen at the moment and if this happens it
484 		 * probably means there is a user space bug
485 		 */
486 		if (error != 0)
487 			linux_msg(td, "futex stuff in thread_detach failed.");
488 	}
489 }
490