1/* $NetBSD: kern_fork.c,v 1.196 2016/11/04 18:14:04 christos Exp $ */
2
3/*-
4 * Copyright (c) 1999, 2001, 2004, 2006, 2007, 2008 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9 * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33/*
34 * Copyright (c) 1982, 1986, 1989, 1991, 1993
35 * The Regents of the University of California. All rights reserved.
36 * (c) UNIX System Laboratories, Inc.
37 * All or some portions of this file are derived from material licensed
38 * to the University of California by American Telephone and Telegraph
39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40 * the permission of UNIX System Laboratories, Inc.
41 *
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
44 * are met:
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)kern_fork.c 8.8 (Berkeley) 2/14/95
67 */
68
69#include <sys/cdefs.h>
70__KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.196 2016/11/04 18:14:04 christos Exp $");
71
72#include "opt_ktrace.h"
73#include "opt_dtrace.h"
74
75#include <sys/param.h>
76#include <sys/systm.h>
77#include <sys/filedesc.h>
78#include <sys/kernel.h>
79#include <sys/pool.h>
80#include <sys/mount.h>
81#include <sys/proc.h>
82#include <sys/ras.h>
83#include <sys/resourcevar.h>
84#include <sys/vnode.h>
85#include <sys/file.h>
86#include <sys/acct.h>
87#include <sys/ktrace.h>
88#include <sys/sched.h>
89#include <sys/signalvar.h>
90#include <sys/kauth.h>
91#include <sys/atomic.h>
92#include <sys/syscallargs.h>
93#include <sys/uidinfo.h>
94#include <sys/sdt.h>
95#include <sys/ptrace.h>
96
97#include <uvm/uvm_extern.h>
98
99/*
100 * DTrace SDT provider definitions
101 */
102SDT_PROVIDER_DECLARE(proc);
103SDT_PROBE_DEFINE3(proc, kernel, , create,
104 "struct proc *", /* new process */
105 "struct proc *", /* parent process */
106 "int" /* flags */);
107
108u_int nprocs __cacheline_aligned = 1; /* process 0 */
109
110/*
111 * Number of ticks to sleep if fork() would fail due to process hitting
112 * limits. Exported in miliseconds to userland via sysctl.
113 */
114int forkfsleep = 0;
115
116int
117sys_fork(struct lwp *l, const void *v, register_t *retval)
118{
119
120 return fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL);
121}
122
123/*
124 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
125 * Address space is not shared, but parent is blocked until child exit.
126 */
127int
128sys_vfork(struct lwp *l, const void *v, register_t *retval)
129{
130
131 return fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
132 retval, NULL);
133}
134
135/*
136 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
137 * semantics. Address space is shared, and parent is blocked until child exit.
138 */
139int
140sys___vfork14(struct lwp *l, const void *v, register_t *retval)
141{
142
143 return fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
144 NULL, NULL, retval, NULL);
145}
146
147/*
148 * Linux-compatible __clone(2) system call.
149 */
150int
151sys___clone(struct lwp *l, const struct sys___clone_args *uap,
152 register_t *retval)
153{
154 /* {
155 syscallarg(int) flags;
156 syscallarg(void *) stack;
157 } */
158 int flags, sig;
159
160 /*
161 * We don't support the CLONE_PID or CLONE_PTRACE flags.
162 */
163 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
164 return EINVAL;
165
166 /*
167 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
168 */
169 if (SCARG(uap, flags) & CLONE_SIGHAND
170 && (SCARG(uap, flags) & CLONE_VM) == 0)
171 return EINVAL;
172
173 flags = 0;
174
175 if (SCARG(uap, flags) & CLONE_VM)
176 flags |= FORK_SHAREVM;
177 if (SCARG(uap, flags) & CLONE_FS)
178 flags |= FORK_SHARECWD;
179 if (SCARG(uap, flags) & CLONE_FILES)
180 flags |= FORK_SHAREFILES;
181 if (SCARG(uap, flags) & CLONE_SIGHAND)
182 flags |= FORK_SHARESIGS;
183 if (SCARG(uap, flags) & CLONE_VFORK)
184 flags |= FORK_PPWAIT;
185
186 sig = SCARG(uap, flags) & CLONE_CSIGNAL;
187 if (sig < 0 || sig >= _NSIG)
188 return EINVAL;
189
190 /*
191 * Note that the Linux API does not provide a portable way of
192 * specifying the stack area; the caller must know if the stack
193 * grows up or down. So, we pass a stack size of 0, so that the
194 * code that makes this adjustment is a noop.
195 */
196 return fork1(l, flags, sig, SCARG(uap, stack), 0,
197 NULL, NULL, retval, NULL);
198}
199
200/*
201 * Print the 'table full' message once per 10 seconds.
202 */
203static struct timeval fork_tfmrate = { 10, 0 };
204
205/*
206 * General fork call. Note that another LWP in the process may call exec()
207 * or exit() while we are forking. It's safe to continue here, because
208 * neither operation will complete until all LWPs have exited the process.
209 */
210int
211fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
212 void (*func)(void *), void *arg, register_t *retval,
213 struct proc **rnewprocp)
214{
215 struct proc *p1, *p2, *parent;
216 struct plimit *p1_lim;
217 uid_t uid;
218 struct lwp *l2;
219 int count;
220 vaddr_t uaddr;
221 int tnprocs;
222 int tracefork;
223 int error = 0;
224
225 p1 = l1->l_proc;
226 uid = kauth_cred_getuid(l1->l_cred);
227 tnprocs = atomic_inc_uint_nv(&nprocs);
228
229 /*
230 * Although process entries are dynamically created, we still keep
231 * a global limit on the maximum number we will create.
232 */
233 if (__predict_false(tnprocs >= maxproc))
234 error = -1;
235 else
236 error = kauth_authorize_process(l1->l_cred,
237 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
238
239 if (error) {
240 static struct timeval lasttfm;
241 atomic_dec_uint(&nprocs);
242 if (ratecheck(&lasttfm, &fork_tfmrate))
243 tablefull("proc", "increase kern.maxproc or NPROC");
244 if (forkfsleep)
245 kpause("forkmx", false, forkfsleep, NULL);
246 return EAGAIN;
247 }
248
249 /*
250 * Enforce limits.
251 */
252 count = chgproccnt(uid, 1);
253 if (__predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
254 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
255 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
256 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0) {
257 (void)chgproccnt(uid, -1);
258 atomic_dec_uint(&nprocs);
259 if (forkfsleep)
260 kpause("forkulim", false, forkfsleep, NULL);
261 return EAGAIN;
262 }
263 }
264
265 /*
266 * Allocate virtual address space for the U-area now, while it
267 * is still easy to abort the fork operation if we're out of
268 * kernel virtual address space.
269 */
270 uaddr = uvm_uarea_alloc();
271 if (__predict_false(uaddr == 0)) {
272 (void)chgproccnt(uid, -1);
273 atomic_dec_uint(&nprocs);
274 return ENOMEM;
275 }
276
277 /*
278 * We are now committed to the fork. From here on, we may
279 * block on resources, but resource allocation may NOT fail.
280 */
281
282 /* Allocate new proc. */
283 p2 = proc_alloc();
284
285 /*
286 * Make a proc table entry for the new process.
287 * Start by zeroing the section of proc that is zero-initialized,
288 * then copy the section that is copied directly from the parent.
289 */
290 memset(&p2->p_startzero, 0,
291 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
292 memcpy(&p2->p_startcopy, &p1->p_startcopy,
293 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
294
295 TAILQ_INIT(&p2->p_sigpend.sp_info);
296
297 LIST_INIT(&p2->p_lwps);
298 LIST_INIT(&p2->p_sigwaiters);
299
300 /*
301 * Duplicate sub-structures as needed.
302 * Increase reference counts on shared objects.
303 * Inherit flags we want to keep. The flags related to SIGCHLD
304 * handling are important in order to keep a consistent behaviour
305 * for the child after the fork. If we are a 32-bit process, the
306 * child will be too.
307 */
308 p2->p_flag =
309 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
310 p2->p_emul = p1->p_emul;
311 p2->p_execsw = p1->p_execsw;
312
313 if (flags & FORK_SYSTEM) {
314 /*
315 * Mark it as a system process. Set P_NOCLDWAIT so that
316 * children are reparented to init(8) when they exit.
317 * init(8) can easily wait them out for us.
318 */
319 p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT);
320 }
321
322 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
323 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
324 rw_init(&p2->p_reflock);
325 cv_init(&p2->p_waitcv, "wait");
326 cv_init(&p2->p_lwpcv, "lwpwait");
327
328 /*
329 * Share a lock between the processes if they are to share signal
330 * state: we must synchronize access to it.
331 */
332 if (flags & FORK_SHARESIGS) {
333 p2->p_lock = p1->p_lock;
334 mutex_obj_hold(p1->p_lock);
335 } else
336 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
337
338 kauth_proc_fork(p1, p2);
339
340 p2->p_raslist = NULL;
341#if defined(__HAVE_RAS)
342 ras_fork(p1, p2);
343#endif
344
345 /* bump references to the text vnode (for procfs) */
346 p2->p_textvp = p1->p_textvp;
347 if (p2->p_textvp)
348 vref(p2->p_textvp);
349
350 if (flags & FORK_SHAREFILES)
351 fd_share(p2);
352 else if (flags & FORK_CLEANFILES)
353 p2->p_fd = fd_init(NULL);
354 else
355 p2->p_fd = fd_copy();
356
357 /* XXX racy */
358 p2->p_mqueue_cnt = p1->p_mqueue_cnt;
359
360 if (flags & FORK_SHARECWD)
361 cwdshare(p2);
362 else
363 p2->p_cwdi = cwdinit();
364
365 /*
366 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
367 * we just need increase pl_refcnt.
368 */
369 p1_lim = p1->p_limit;
370 if (!p1_lim->pl_writeable) {
371 lim_addref(p1_lim);
372 p2->p_limit = p1_lim;
373 } else {
374 p2->p_limit = lim_copy(p1_lim);
375 }
376
377 if (flags & FORK_PPWAIT) {
378 /* Mark ourselves as waiting for a child. */
379 l1->l_pflag |= LP_VFORKWAIT;
380 p2->p_lflag = PL_PPWAIT;
381 p2->p_vforklwp = l1;
382 } else {
383 p2->p_lflag = 0;
384 }
385 p2->p_sflag = 0;
386 p2->p_slflag = 0;
387 parent = (flags & FORK_NOWAIT) ? initproc : p1;
388 p2->p_pptr = parent;
389 p2->p_ppid = parent->p_pid;
390 LIST_INIT(&p2->p_children);
391
392 p2->p_aio = NULL;
393
394#ifdef KTRACE
395 /*
396 * Copy traceflag and tracefile if enabled.
397 * If not inherited, these were zeroed above.
398 */
399 if (p1->p_traceflag & KTRFAC_INHERIT) {
400 mutex_enter(&ktrace_lock);
401 p2->p_traceflag = p1->p_traceflag;
402 if ((p2->p_tracep = p1->p_tracep) != NULL)
403 ktradref(p2);
404 mutex_exit(&ktrace_lock);
405 }
406#endif
407
408 /*
409 * Create signal actions for the child process.
410 */
411 p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS);
412 mutex_enter(p1->p_lock);
413 p2->p_sflag |=
414 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
415 sched_proc_fork(p1, p2);
416 mutex_exit(p1->p_lock);
417
418 p2->p_stflag = p1->p_stflag;
419
420 /*
421 * p_stats.
422 * Copy parts of p_stats, and zero out the rest.
423 */
424 p2->p_stats = pstatscopy(p1->p_stats);
425
426 /*
427 * Set up the new process address space.
428 */
429 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false);
430
431 /*
432 * Finish creating the child process.
433 * It will return through a different path later.
434 */
435 lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0,
436 stack, stacksize, (func != NULL) ? func : child_return, arg, &l2,
437 l1->l_class);
438
439 /*
440 * Inherit l_private from the parent.
441 * Note that we cannot use lwp_setprivate() here since that
442 * also sets the CPU TLS register, which is incorrect if the
443 * process has changed that without letting the kernel know.
444 */
445 l2->l_private = l1->l_private;
446
447 /*
448 * If emulation has a process fork hook, call it now.
449 */
450 if (p2->p_emul->e_proc_fork)
451 (*p2->p_emul->e_proc_fork)(p2, l1, flags);
452
453 /*
454 * ...and finally, any other random fork hooks that subsystems
455 * might have registered.
456 */
457 doforkhooks(p2, p1);
458
459 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
460
461 /*
462 * It's now safe for the scheduler and other processes to see the
463 * child process.
464 */
465 mutex_enter(proc_lock);
466
467 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
468 p2->p_lflag |= PL_CONTROLT;
469
470 LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
471 p2->p_exitsig = exitsig; /* signal for parent on exit */
472
473 /*
474 * We don't want to tracefork vfork()ed processes because they
475 * will not receive the SIGTRAP until it is too late.
476 */
477 tracefork = (p1->p_slflag & (PSL_TRACEFORK|PSL_TRACED)) ==
478 (PSL_TRACEFORK|PSL_TRACED) && (flags && FORK_PPWAIT) == 0;
479 if (tracefork) {
480 proc_changeparent(p2, p1->p_pptr);
481 /*
482 * Set ptrace status.
483 */
484 p1->p_fpid = p2->p_pid;
485 p2->p_fpid = p1->p_pid;
486 }
487
488 LIST_INSERT_AFTER(p1, p2, p_pglist);
489 LIST_INSERT_HEAD(&allproc, p2, p_list);
490
491 p2->p_trace_enabled = trace_is_enabled(p2);
492#ifdef __HAVE_SYSCALL_INTERN
493 (*p2->p_emul->e_syscall_intern)(p2);
494#endif
495
496 /*
497 * Update stats now that we know the fork was successful.
498 */
499 uvmexp.forks++;
500 if (flags & FORK_PPWAIT)
501 uvmexp.forks_ppwait++;
502 if (flags & FORK_SHAREVM)
503 uvmexp.forks_sharevm++;
504
505 /*
506 * Pass a pointer to the new process to the caller.
507 */
508 if (rnewprocp != NULL)
509 *rnewprocp = p2;
510
511 if (ktrpoint(KTR_EMUL))
512 p2->p_traceflag |= KTRFAC_TRC_EMUL;
513
514 /*
515 * Notify any interested parties about the new process.
516 */
517 if (!SLIST_EMPTY(&p1->p_klist)) {
518 mutex_exit(proc_lock);
519 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
520 mutex_enter(proc_lock);
521 }
522
523 /*
524 * Make child runnable, set start time, and add to run queue except
525 * if the parent requested the child to start in SSTOP state.
526 */
527 mutex_enter(p2->p_lock);
528
529 /*
530 * Start profiling.
531 */
532 if ((p2->p_stflag & PST_PROFIL) != 0) {
533 mutex_spin_enter(&p2->p_stmutex);
534 startprofclock(p2);
535 mutex_spin_exit(&p2->p_stmutex);
536 }
537
538 getmicrotime(&p2->p_stats->p_start);
539 p2->p_acflag = AFORK;
540 lwp_lock(l2);
541 KASSERT(p2->p_nrlwps == 1);
542 if (p2->p_sflag & PS_STOPFORK) {
543 struct schedstate_percpu *spc = &l2->l_cpu->ci_schedstate;
544 p2->p_nrlwps = 0;
545 p2->p_stat = SSTOP;
546 p2->p_waited = 0;
547 p1->p_nstopchild++;
548 l2->l_stat = LSSTOP;
549 KASSERT(l2->l_wchan == NULL);
550 lwp_unlock_to(l2, spc->spc_lwplock);
551 } else {
552 p2->p_nrlwps = 1;
553 p2->p_stat = SACTIVE;
554 l2->l_stat = LSRUN;
555 sched_enqueue(l2, false);
556 lwp_unlock(l2);
557 }
558
559 /*
560 * Return child pid to parent process,
561 * marking us as parent via retval[1].
562 */
563 if (retval != NULL) {
564 retval[0] = p2->p_pid;
565 retval[1] = 0;
566 }
567 mutex_exit(p2->p_lock);
568
569 /*
570 * Preserve synchronization semantics of vfork. If waiting for
571 * child to exec or exit, sleep until it clears LP_VFORKWAIT.
572 */
573#if 0
574 while (l1->l_pflag & LP_VFORKWAIT) {
575 cv_wait(&l1->l_waitcv, proc_lock);
576 }
577#else
578 while (p2->p_lflag & PL_PPWAIT)
579 cv_wait(&p1->p_waitcv, proc_lock);
580#endif
581
582 /*
583 * Let the parent know that we are tracing its child.
584 */
585 if (tracefork) {
586 ksiginfo_t ksi;
587
588 KSI_INIT_EMPTY(&ksi);
589 ksi.ksi_signo = SIGTRAP;
590 ksi.ksi_lid = l1->l_lid;
591 kpsignal(p1, &ksi, NULL);
592 }
593 mutex_exit(proc_lock);
594
595 return 0;
596}
597