FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_fork.c
1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_ktrace.h"
41 #include "opt_mac.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/eventhandler.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/kthread.h>
50 #include <sys/sysctl.h>
51 #include <sys/lock.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/pioctl.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/syscall.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
62 #include <sys/acct.h>
63 #include <sys/ktr.h>
64 #include <sys/ktrace.h>
65 #include <sys/unistd.h>
66 #include <sys/sx.h>
67 #include <sys/signalvar.h>
68
69 #include <security/audit/audit.h>
70 #include <security/mac/mac_framework.h>
71
72 #include <vm/vm.h>
73 #include <vm/pmap.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_extern.h>
76 #include <vm/uma.h>
77
78
79 #ifndef _SYS_SYSPROTO_H_
80 struct fork_args {
81 int dummy;
82 };
83 #endif
84
85 /* ARGSUSED */
86 int
87 fork(td, uap)
88 struct thread *td;
89 struct fork_args *uap;
90 {
91 int error;
92 struct proc *p2;
93
94 error = fork1(td, RFFDG | RFPROC, 0, &p2);
95 if (error == 0) {
96 td->td_retval[0] = p2->p_pid;
97 td->td_retval[1] = 0;
98 }
99 return (error);
100 }
101
102 /* ARGSUSED */
103 int
104 vfork(td, uap)
105 struct thread *td;
106 struct vfork_args *uap;
107 {
108 int error;
109 struct proc *p2;
110
111 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
112 if (error == 0) {
113 td->td_retval[0] = p2->p_pid;
114 td->td_retval[1] = 0;
115 }
116 return (error);
117 }
118
119 int
120 rfork(td, uap)
121 struct thread *td;
122 struct rfork_args *uap;
123 {
124 struct proc *p2;
125 int error;
126
127 /* Don't allow kernel-only flags. */
128 if ((uap->flags & RFKERNELONLY) != 0)
129 return (EINVAL);
130
131 AUDIT_ARG(fflags, uap->flags);
132 error = fork1(td, uap->flags, 0, &p2);
133 if (error == 0) {
134 td->td_retval[0] = p2 ? p2->p_pid : 0;
135 td->td_retval[1] = 0;
136 }
137 return (error);
138 }
139
140 int nprocs = 1; /* process 0 */
141 int lastpid = 0;
142 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
143 "Last used PID");
144
145 /*
146 * Random component to lastpid generation. We mix in a random factor to make
147 * it a little harder to predict. We sanity check the modulus value to avoid
148 * doing it in critical paths. Don't let it be too small or we pointlessly
149 * waste randomness entropy, and don't let it be impossibly large. Using a
150 * modulus that is too big causes a LOT more process table scans and slows
151 * down fork processing as the pidchecked caching is defeated.
152 */
153 static int randompid = 0;
154
155 static int
156 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
157 {
158 int error, pid;
159
160 error = sysctl_wire_old_buffer(req, sizeof(int));
161 if (error != 0)
162 return(error);
163 sx_xlock(&allproc_lock);
164 pid = randompid;
165 error = sysctl_handle_int(oidp, &pid, 0, req);
166 if (error == 0 && req->newptr != NULL) {
167 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
168 pid = PID_MAX - 100;
169 else if (pid < 2) /* NOP */
170 pid = 0;
171 else if (pid < 100) /* Make it reasonable */
172 pid = 100;
173 randompid = pid;
174 }
175 sx_xunlock(&allproc_lock);
176 return (error);
177 }
178
179 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
180 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
181
182 int
183 fork1(td, flags, pages, procp)
184 struct thread *td;
185 int flags;
186 int pages;
187 struct proc **procp;
188 {
189 struct proc *p1, *p2, *pptr;
190 struct proc *newproc;
191 int ok, trypid;
192 static int curfail, pidchecked = 0;
193 static struct timeval lastfail;
194 struct filedesc *fd;
195 struct filedesc_to_leader *fdtol;
196 struct thread *td2;
197 struct sigacts *newsigacts;
198 struct vmspace *vm2;
199 int error;
200
201 /* Can't copy and clear. */
202 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
203 return (EINVAL);
204
205 p1 = td->td_proc;
206
207 /*
208 * Here we don't create a new process, but we divorce
209 * certain parts of a process from itself.
210 */
211 if ((flags & RFPROC) == 0) {
212 if ((p1->p_flag & P_HADTHREADS) &&
213 (flags & (RFCFDG | RFFDG))) {
214 PROC_LOCK(p1);
215 if (thread_single(SINGLE_BOUNDARY)) {
216 PROC_UNLOCK(p1);
217 return (ERESTART);
218 }
219 PROC_UNLOCK(p1);
220 }
221
222 error = vm_forkproc(td, NULL, NULL, NULL, flags);
223 if (error)
224 goto norfproc_fail;
225
226 /*
227 * Close all file descriptors.
228 */
229 if (flags & RFCFDG) {
230 struct filedesc *fdtmp;
231 fdtmp = fdinit(td->td_proc->p_fd);
232 fdfree(td);
233 p1->p_fd = fdtmp;
234 }
235
236 /*
237 * Unshare file descriptors (from parent).
238 */
239 if (flags & RFFDG)
240 fdunshare(p1, td);
241
242 norfproc_fail:
243 if ((p1->p_flag & P_HADTHREADS) &&
244 (flags & (RFCFDG | RFFDG))) {
245 PROC_LOCK(p1);
246 thread_single_end();
247 PROC_UNLOCK(p1);
248 }
249 *procp = NULL;
250 return (error);
251 }
252
253 /* Allocate new proc. */
254 newproc = uma_zalloc(proc_zone, M_WAITOK);
255 if (TAILQ_EMPTY(&newproc->p_threads)) {
256 td2 = thread_alloc();
257 if (td2 == NULL) {
258 error = ENOMEM;
259 goto fail1;
260 }
261 proc_linkup(newproc, td2);
262 sched_newproc(newproc, td2);
263 } else
264 td2 = FIRST_THREAD_IN_PROC(newproc);
265
266 /* Allocate and switch to an alternate kstack if specified. */
267 if (pages != 0) {
268 if (!vm_thread_new_altkstack(td2, pages)) {
269 error = ENOMEM;
270 goto fail1;
271 }
272 }
273 if ((flags & RFMEM) == 0) {
274 vm2 = vmspace_fork(p1->p_vmspace);
275 if (vm2 == NULL) {
276 error = ENOMEM;
277 goto fail1;
278 }
279 } else
280 vm2 = NULL;
281 #ifdef MAC
282 mac_init_proc(newproc);
283 #endif
284 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
285 STAILQ_INIT(&newproc->p_ktr);
286
287 /* We have to lock the process tree while we look for a pid. */
288 sx_slock(&proctree_lock);
289
290 /*
291 * Although process entries are dynamically created, we still keep
292 * a global limit on the maximum number we will create. Don't allow
293 * a nonprivileged user to use the last ten processes; don't let root
294 * exceed the limit. The variable nprocs is the current number of
295 * processes, maxproc is the limit.
296 */
297 sx_xlock(&allproc_lock);
298 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
299 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
300 error = EAGAIN;
301 goto fail;
302 }
303
304 /*
305 * Increment the count of procs running with this uid. Don't allow
306 * a nonprivileged user to exceed their current limit.
307 *
308 * XXXRW: Can we avoid privilege here if it's not needed?
309 */
310 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
311 if (error == 0)
312 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
313 else {
314 PROC_LOCK(p1);
315 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
316 lim_cur(p1, RLIMIT_NPROC));
317 PROC_UNLOCK(p1);
318 }
319 if (!ok) {
320 error = EAGAIN;
321 goto fail;
322 }
323
324 /*
325 * Increment the nprocs resource before blocking can occur. There
326 * are hard-limits as to the number of processes that can run.
327 */
328 nprocs++;
329
330 /*
331 * Find an unused process ID. We remember a range of unused IDs
332 * ready to use (from lastpid+1 through pidchecked-1).
333 *
334 * If RFHIGHPID is set (used during system boot), do not allocate
335 * low-numbered pids.
336 */
337 trypid = lastpid + 1;
338 if (flags & RFHIGHPID) {
339 if (trypid < 10)
340 trypid = 10;
341 } else {
342 if (randompid)
343 trypid += arc4random() % randompid;
344 }
345 retry:
346 /*
347 * If the process ID prototype has wrapped around,
348 * restart somewhat above 0, as the low-numbered procs
349 * tend to include daemons that don't exit.
350 */
351 if (trypid >= PID_MAX) {
352 trypid = trypid % PID_MAX;
353 if (trypid < 100)
354 trypid += 100;
355 pidchecked = 0;
356 }
357 if (trypid >= pidchecked) {
358 int doingzomb = 0;
359
360 pidchecked = PID_MAX;
361 /*
362 * Scan the active and zombie procs to check whether this pid
363 * is in use. Remember the lowest pid that's greater
364 * than trypid, so we can avoid checking for a while.
365 */
366 p2 = LIST_FIRST(&allproc);
367 again:
368 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
369 while (p2->p_pid == trypid ||
370 (p2->p_pgrp != NULL &&
371 (p2->p_pgrp->pg_id == trypid ||
372 (p2->p_session != NULL &&
373 p2->p_session->s_sid == trypid)))) {
374 trypid++;
375 if (trypid >= pidchecked)
376 goto retry;
377 }
378 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
379 pidchecked = p2->p_pid;
380 if (p2->p_pgrp != NULL) {
381 if (p2->p_pgrp->pg_id > trypid &&
382 pidchecked > p2->p_pgrp->pg_id)
383 pidchecked = p2->p_pgrp->pg_id;
384 if (p2->p_session != NULL &&
385 p2->p_session->s_sid > trypid &&
386 pidchecked > p2->p_session->s_sid)
387 pidchecked = p2->p_session->s_sid;
388 }
389 }
390 if (!doingzomb) {
391 doingzomb = 1;
392 p2 = LIST_FIRST(&zombproc);
393 goto again;
394 }
395 }
396 sx_sunlock(&proctree_lock);
397
398 /*
399 * RFHIGHPID does not mess with the lastpid counter during boot.
400 */
401 if (flags & RFHIGHPID)
402 pidchecked = 0;
403 else
404 lastpid = trypid;
405
406 p2 = newproc;
407 p2->p_state = PRS_NEW; /* protect against others */
408 p2->p_pid = trypid;
409 /*
410 * Allow the scheduler to initialize the child.
411 */
412 thread_lock(td);
413 sched_fork(td, td2);
414 thread_unlock(td);
415 AUDIT_ARG(pid, p2->p_pid);
416 LIST_INSERT_HEAD(&allproc, p2, p_list);
417 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
418
419 PROC_LOCK(p2);
420 PROC_LOCK(p1);
421
422 sx_xunlock(&allproc_lock);
423
424 bcopy(&p1->p_startcopy, &p2->p_startcopy,
425 __rangeof(struct proc, p_startcopy, p_endcopy));
426 PROC_UNLOCK(p1);
427
428 bzero(&p2->p_startzero,
429 __rangeof(struct proc, p_startzero, p_endzero));
430
431 p2->p_ucred = crhold(td->td_ucred);
432 PROC_UNLOCK(p2);
433
434 /*
435 * Malloc things while we don't hold any locks.
436 */
437 if (flags & RFSIGSHARE)
438 newsigacts = NULL;
439 else
440 newsigacts = sigacts_alloc();
441
442 /*
443 * Copy filedesc.
444 */
445 if (flags & RFCFDG) {
446 fd = fdinit(p1->p_fd);
447 fdtol = NULL;
448 } else if (flags & RFFDG) {
449 fd = fdcopy(p1->p_fd);
450 fdtol = NULL;
451 } else {
452 fd = fdshare(p1->p_fd);
453 if (p1->p_fdtol == NULL)
454 p1->p_fdtol =
455 filedesc_to_leader_alloc(NULL,
456 NULL,
457 p1->p_leader);
458 if ((flags & RFTHREAD) != 0) {
459 /*
460 * Shared file descriptor table and
461 * shared process leaders.
462 */
463 fdtol = p1->p_fdtol;
464 FILEDESC_XLOCK(p1->p_fd);
465 fdtol->fdl_refcount++;
466 FILEDESC_XUNLOCK(p1->p_fd);
467 } else {
468 /*
469 * Shared file descriptor table, and
470 * different process leaders
471 */
472 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
473 p1->p_fd,
474 p2);
475 }
476 }
477 /*
478 * Make a proc table entry for the new process.
479 * Start by zeroing the section of proc that is zero-initialized,
480 * then copy the section that is copied directly from the parent.
481 */
482
483 PROC_LOCK(p2);
484 PROC_LOCK(p1);
485
486 bzero(&td2->td_startzero,
487 __rangeof(struct thread, td_startzero, td_endzero));
488
489 bcopy(&td->td_startcopy, &td2->td_startcopy,
490 __rangeof(struct thread, td_startcopy, td_endcopy));
491
492 td2->td_sigstk = td->td_sigstk;
493 td2->td_sigmask = td->td_sigmask;
494 td2->td_flags = TDF_INMEM;
495
496 /*
497 * Duplicate sub-structures as needed.
498 * Increase reference counts on shared objects.
499 */
500 p2->p_flag = P_INMEM;
501 p2->p_swtick = ticks;
502 if (p1->p_flag & P_PROFIL)
503 startprofclock(p2);
504 td2->td_ucred = crhold(p2->p_ucred);
505 pargs_hold(p2->p_args);
506
507 if (flags & RFSIGSHARE) {
508 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
509 } else {
510 sigacts_copy(newsigacts, p1->p_sigacts);
511 p2->p_sigacts = newsigacts;
512 }
513 if (flags & RFLINUXTHPN)
514 p2->p_sigparent = SIGUSR1;
515 else
516 p2->p_sigparent = SIGCHLD;
517
518 p2->p_textvp = p1->p_textvp;
519 p2->p_fd = fd;
520 p2->p_fdtol = fdtol;
521
522 /*
523 * p_limit is copy-on-write. Bump its refcount.
524 */
525 lim_fork(p1, p2);
526
527 pstats_fork(p1->p_stats, p2->p_stats);
528
529 PROC_UNLOCK(p1);
530 PROC_UNLOCK(p2);
531
532 /* Bump references to the text vnode (for procfs) */
533 if (p2->p_textvp)
534 vref(p2->p_textvp);
535
536 /*
537 * Set up linkage for kernel based threading.
538 */
539 if ((flags & RFTHREAD) != 0) {
540 mtx_lock(&ppeers_lock);
541 p2->p_peers = p1->p_peers;
542 p1->p_peers = p2;
543 p2->p_leader = p1->p_leader;
544 mtx_unlock(&ppeers_lock);
545 PROC_LOCK(p1->p_leader);
546 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
547 PROC_UNLOCK(p1->p_leader);
548 /*
549 * The task leader is exiting, so process p1 is
550 * going to be killed shortly. Since p1 obviously
551 * isn't dead yet, we know that the leader is either
552 * sending SIGKILL's to all the processes in this
553 * task or is sleeping waiting for all the peers to
554 * exit. We let p1 complete the fork, but we need
555 * to go ahead and kill the new process p2 since
556 * the task leader may not get a chance to send
557 * SIGKILL to it. We leave it on the list so that
558 * the task leader will wait for this new process
559 * to commit suicide.
560 */
561 PROC_LOCK(p2);
562 psignal(p2, SIGKILL);
563 PROC_UNLOCK(p2);
564 } else
565 PROC_UNLOCK(p1->p_leader);
566 } else {
567 p2->p_peers = NULL;
568 p2->p_leader = p2;
569 }
570
571 sx_xlock(&proctree_lock);
572 PGRP_LOCK(p1->p_pgrp);
573 PROC_LOCK(p2);
574 PROC_LOCK(p1);
575
576 /*
577 * Preserve some more flags in subprocess. P_PROFIL has already
578 * been preserved.
579 */
580 p2->p_flag |= p1->p_flag & P_SUGID;
581 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
582 SESS_LOCK(p1->p_session);
583 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
584 p2->p_flag |= P_CONTROLT;
585 SESS_UNLOCK(p1->p_session);
586 if (flags & RFPPWAIT)
587 p2->p_flag |= P_PPWAIT;
588
589 p2->p_pgrp = p1->p_pgrp;
590 LIST_INSERT_AFTER(p1, p2, p_pglist);
591 PGRP_UNLOCK(p1->p_pgrp);
592 LIST_INIT(&p2->p_children);
593
594 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
595
596 #ifdef KTRACE
597 /*
598 * Copy traceflag and tracefile if enabled.
599 */
600 mtx_lock(&ktrace_mtx);
601 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
602 if (p1->p_traceflag & KTRFAC_INHERIT) {
603 p2->p_traceflag = p1->p_traceflag;
604 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
605 VREF(p2->p_tracevp);
606 KASSERT(p1->p_tracecred != NULL,
607 ("ktrace vnode with no cred"));
608 p2->p_tracecred = crhold(p1->p_tracecred);
609 }
610 }
611 mtx_unlock(&ktrace_mtx);
612 #endif
613
614 /*
615 * If PF_FORK is set, the child process inherits the
616 * procfs ioctl flags from its parent.
617 */
618 if (p1->p_pfsflags & PF_FORK) {
619 p2->p_stops = p1->p_stops;
620 p2->p_pfsflags = p1->p_pfsflags;
621 }
622
623 /*
624 * This begins the section where we must prevent the parent
625 * from being swapped.
626 */
627 _PHOLD(p1);
628 PROC_UNLOCK(p1);
629
630 /*
631 * Attach the new process to its parent.
632 *
633 * If RFNOWAIT is set, the newly created process becomes a child
634 * of init. This effectively disassociates the child from the
635 * parent.
636 */
637 if (flags & RFNOWAIT)
638 pptr = initproc;
639 else
640 pptr = p1;
641 p2->p_pptr = pptr;
642 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
643 sx_xunlock(&proctree_lock);
644
645 /* Inform accounting that we have forked. */
646 p2->p_acflag = AFORK;
647 PROC_UNLOCK(p2);
648
649 /*
650 * Finish creating the child process. It will return via a different
651 * execution path later. (ie: directly into user mode)
652 */
653 vm_forkproc(td, p2, td2, vm2, flags);
654
655 if (flags == (RFFDG | RFPROC)) {
656 PCPU_INC(cnt.v_forks);
657 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
658 p2->p_vmspace->vm_ssize);
659 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
660 PCPU_INC(cnt.v_vforks);
661 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
662 p2->p_vmspace->vm_ssize);
663 } else if (p1 == &proc0) {
664 PCPU_INC(cnt.v_kthreads);
665 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
666 p2->p_vmspace->vm_ssize);
667 } else {
668 PCPU_INC(cnt.v_rforks);
669 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
670 p2->p_vmspace->vm_ssize);
671 }
672
673 /*
674 * Both processes are set up, now check if any loadable modules want
675 * to adjust anything.
676 * What if they have an error? XXX
677 */
678 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
679
680 /*
681 * Set the child start time and mark the process as being complete.
682 */
683 microuptime(&p2->p_stats->p_start);
684 PROC_SLOCK(p2);
685 p2->p_state = PRS_NORMAL;
686 PROC_SUNLOCK(p2);
687
688 /*
689 * If RFSTOPPED not requested, make child runnable and add to
690 * run queue.
691 */
692 if ((flags & RFSTOPPED) == 0) {
693 thread_lock(td2);
694 TD_SET_CAN_RUN(td2);
695 sched_add(td2, SRQ_BORING);
696 thread_unlock(td2);
697 }
698
699 /*
700 * Now can be swapped.
701 */
702 PROC_LOCK(p1);
703 _PRELE(p1);
704
705 /*
706 * Tell any interested parties about the new process.
707 */
708 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
709
710 PROC_UNLOCK(p1);
711
712 /*
713 * Preserve synchronization semantics of vfork. If waiting for
714 * child to exec or exit, set P_PPWAIT on child, and sleep on our
715 * proc (in case of exit).
716 */
717 PROC_LOCK(p2);
718 while (p2->p_flag & P_PPWAIT)
719 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
720 PROC_UNLOCK(p2);
721
722 /*
723 * Return child proc pointer to parent.
724 */
725 *procp = p2;
726 return (0);
727 fail:
728 sx_sunlock(&proctree_lock);
729 if (ppsratecheck(&lastfail, &curfail, 1))
730 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
731 td->td_ucred->cr_ruid);
732 sx_xunlock(&allproc_lock);
733 #ifdef MAC
734 mac_destroy_proc(newproc);
735 #endif
736 fail1:
737 uma_zfree(proc_zone, newproc);
738 pause("fork", hz / 2);
739 return (error);
740 }
741
742 /*
743 * Handle the return of a child process from fork1(). This function
744 * is called from the MD fork_trampoline() entry point.
745 */
746 void
747 fork_exit(callout, arg, frame)
748 void (*callout)(void *, struct trapframe *);
749 void *arg;
750 struct trapframe *frame;
751 {
752 struct proc *p;
753 struct thread *td;
754 struct thread *dtd;
755
756 td = curthread;
757 p = td->td_proc;
758 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
759
760 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
761 td, td->td_sched, p->p_pid, p->p_comm);
762
763 sched_fork_exit(td);
764 /*
765 * Processes normally resume in mi_switch() after being
766 * cpu_switch()'ed to, but when children start up they arrive here
767 * instead, so we must do much the same things as mi_switch() would.
768 */
769 if ((dtd = PCPU_GET(deadthread))) {
770 PCPU_SET(deadthread, NULL);
771 thread_stash(dtd);
772 }
773 thread_unlock(td);
774
775 /*
776 * cpu_set_fork_handler intercepts this function call to
777 * have this call a non-return function to stay in kernel mode.
778 * initproc has its own fork handler, but it does return.
779 */
780 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
781 callout(arg, frame);
782
783 /*
784 * Check if a kernel thread misbehaved and returned from its main
785 * function.
786 */
787 if (p->p_flag & P_KTHREAD) {
788 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
789 p->p_comm, p->p_pid);
790 kthread_exit(0);
791 }
792 mtx_assert(&Giant, MA_NOTOWNED);
793
794 EVENTHANDLER_INVOKE(schedtail, p);
795 }
796
797 /*
798 * Simplified back end of syscall(), used when returning from fork()
799 * directly into user mode. Giant is not held on entry, and must not
800 * be held on return. This function is passed in to fork_exit() as the
801 * first parameter and is called when returning to a new userland process.
802 */
803 void
804 fork_return(td, frame)
805 struct thread *td;
806 struct trapframe *frame;
807 {
808
809 userret(td, frame);
810 #ifdef KTRACE
811 if (KTRPOINT(td, KTR_SYSRET))
812 ktrsysret(SYS_fork, 0, 0);
813 #endif
814 mtx_assert(&Giant, MA_NOTOWNED);
815 }
Cache object: 2ef8844e8580fb1773537e74a6def1a6
|