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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
39 * $FreeBSD: releng/5.1/sys/kern/kern_fork.c 114983 2003-05-13 20:36:02Z jhb $
40 */
41
42 #include "opt_ktrace.h"
43 #include "opt_mac.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/filedesc.h>
50 #include <sys/kernel.h>
51 #include <sys/sysctl.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.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/vnode.h>
61 #include <sys/acct.h>
62 #include <sys/mac.h>
63 #include <sys/ktr.h>
64 #include <sys/ktrace.h>
65 #include <sys/kthread.h>
66 #include <sys/unistd.h>
67 #include <sys/jail.h>
68 #include <sys/sx.h>
69
70 #include <vm/vm.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_extern.h>
74 #include <vm/uma.h>
75
76 #include <sys/vmmeter.h>
77 #include <sys/user.h>
78 #include <machine/critical.h>
79
80 #ifndef _SYS_SYSPROTO_H_
81 struct fork_args {
82 int dummy;
83 };
84 #endif
85
86 static int forksleep; /* Place for fork1() to sleep on. */
87
88 /*
89 * MPSAFE
90 */
91 /* ARGSUSED */
92 int
93 fork(td, uap)
94 struct thread *td;
95 struct fork_args *uap;
96 {
97 int error;
98 struct proc *p2;
99
100 error = fork1(td, RFFDG | RFPROC, 0, &p2);
101 if (error == 0) {
102 td->td_retval[0] = p2->p_pid;
103 td->td_retval[1] = 0;
104 }
105 return error;
106 }
107
108 /*
109 * MPSAFE
110 */
111 /* ARGSUSED */
112 int
113 vfork(td, uap)
114 struct thread *td;
115 struct vfork_args *uap;
116 {
117 int error;
118 struct proc *p2;
119
120 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
121 if (error == 0) {
122 td->td_retval[0] = p2->p_pid;
123 td->td_retval[1] = 0;
124 }
125 return error;
126 }
127
128 /*
129 * MPSAFE
130 */
131 int
132 rfork(td, uap)
133 struct thread *td;
134 struct rfork_args *uap;
135 {
136 int error;
137 struct proc *p2;
138
139 /* Don't allow kernel only flags. */
140 if ((uap->flags & RFKERNELONLY) != 0)
141 return (EINVAL);
142 /*
143 * Don't allow sharing of file descriptor table unless
144 * RFTHREAD flag is supplied
145 */
146 if ((uap->flags & (RFPROC | RFTHREAD | RFFDG | RFCFDG)) ==
147 RFPROC)
148 return(EINVAL);
149 error = fork1(td, uap->flags, 0, &p2);
150 if (error == 0) {
151 td->td_retval[0] = p2 ? p2->p_pid : 0;
152 td->td_retval[1] = 0;
153 }
154 return error;
155 }
156
157
158 int nprocs = 1; /* process 0 */
159 int lastpid = 0;
160 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
161 "Last used PID");
162
163 /*
164 * Random component to lastpid generation. We mix in a random factor to make
165 * it a little harder to predict. We sanity check the modulus value to avoid
166 * doing it in critical paths. Don't let it be too small or we pointlessly
167 * waste randomness entropy, and don't let it be impossibly large. Using a
168 * modulus that is too big causes a LOT more process table scans and slows
169 * down fork processing as the pidchecked caching is defeated.
170 */
171 static int randompid = 0;
172
173 static int
174 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
175 {
176 int error, pid;
177
178 sysctl_wire_old_buffer(req, sizeof(int));
179 sx_xlock(&allproc_lock);
180 pid = randompid;
181 error = sysctl_handle_int(oidp, &pid, 0, req);
182 if (error == 0 && req->newptr != NULL) {
183 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
184 pid = PID_MAX - 100;
185 else if (pid < 2) /* NOP */
186 pid = 0;
187 else if (pid < 100) /* Make it reasonable */
188 pid = 100;
189 randompid = pid;
190 }
191 sx_xunlock(&allproc_lock);
192 return (error);
193 }
194
195 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
196 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
197
198 int
199 fork1(td, flags, pages, procp)
200 struct thread *td; /* parent proc */
201 int flags;
202 int pages;
203 struct proc **procp; /* child proc */
204 {
205 struct proc *p2, *pptr;
206 uid_t uid;
207 struct proc *newproc;
208 int trypid;
209 int ok;
210 static int pidchecked = 0;
211 struct filedesc *fd;
212 struct proc *p1 = td->td_proc;
213 struct thread *td2;
214 struct kse *ke2;
215 struct ksegrp *kg2;
216 struct sigacts *newsigacts;
217 int error;
218
219 /* Can't copy and clear */
220 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
221 return (EINVAL);
222
223 mtx_lock(&Giant);
224 /*
225 * Here we don't create a new process, but we divorce
226 * certain parts of a process from itself.
227 */
228 if ((flags & RFPROC) == 0) {
229 vm_forkproc(td, NULL, NULL, flags);
230
231 /*
232 * Close all file descriptors.
233 */
234 if (flags & RFCFDG) {
235 struct filedesc *fdtmp;
236 fdtmp = fdinit(td->td_proc->p_fd);
237 fdfree(td);
238 p1->p_fd = fdtmp;
239 }
240
241 /*
242 * Unshare file descriptors (from parent.)
243 */
244 if (flags & RFFDG) {
245 FILEDESC_LOCK(p1->p_fd);
246 if (p1->p_fd->fd_refcnt > 1) {
247 struct filedesc *newfd;
248
249 newfd = fdcopy(td->td_proc->p_fd);
250 FILEDESC_UNLOCK(p1->p_fd);
251 fdfree(td);
252 p1->p_fd = newfd;
253 } else
254 FILEDESC_UNLOCK(p1->p_fd);
255 }
256 mtx_unlock(&Giant);
257 *procp = NULL;
258 return (0);
259 }
260
261 /*
262 * Note 1:1 allows for forking with one thread coming out on the
263 * other side with the expectation that the process is about to
264 * exec.
265 */
266 if (p1->p_flag & P_THREADED) {
267 /*
268 * Idle the other threads for a second.
269 * Since the user space is copied, it must remain stable.
270 * In addition, all threads (from the user perspective)
271 * need to either be suspended or in the kernel,
272 * where they will try restart in the parent and will
273 * be aborted in the child.
274 */
275 PROC_LOCK(p1);
276 if (thread_single(SINGLE_NO_EXIT)) {
277 /* Abort.. someone else is single threading before us */
278 PROC_UNLOCK(p1);
279 mtx_unlock(&Giant);
280 return (ERESTART);
281 }
282 PROC_UNLOCK(p1);
283 /*
284 * All other activity in this process
285 * is now suspended at the user boundary,
286 * (or other safe places if we think of any).
287 */
288 }
289
290 /* Allocate new proc. */
291 newproc = uma_zalloc(proc_zone, M_WAITOK);
292 #ifdef MAC
293 mac_init_proc(newproc);
294 #endif
295
296 /*
297 * Although process entries are dynamically created, we still keep
298 * a global limit on the maximum number we will create. Don't allow
299 * a nonprivileged user to use the last ten processes; don't let root
300 * exceed the limit. The variable nprocs is the current number of
301 * processes, maxproc is the limit.
302 */
303 sx_xlock(&allproc_lock);
304 uid = td->td_ucred->cr_ruid;
305 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
306 error = EAGAIN;
307 goto fail;
308 }
309
310 /*
311 * Increment the count of procs running with this uid. Don't allow
312 * a nonprivileged user to exceed their current limit.
313 */
314 PROC_LOCK(p1);
315 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
316 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
317 PROC_UNLOCK(p1);
318 if (!ok) {
319 error = EAGAIN;
320 goto fail;
321 }
322
323 /*
324 * Increment the nprocs resource before blocking can occur. There
325 * are hard-limits as to the number of processes that can run.
326 */
327 nprocs++;
328
329 /*
330 * Find an unused process ID. We remember a range of unused IDs
331 * ready to use (from lastpid+1 through pidchecked-1).
332 *
333 * If RFHIGHPID is set (used during system boot), do not allocate
334 * low-numbered pids.
335 */
336 trypid = lastpid + 1;
337 if (flags & RFHIGHPID) {
338 if (trypid < 10) {
339 trypid = 10;
340 }
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 PROC_LOCK(p2);
370 while (p2->p_pid == trypid ||
371 p2->p_pgrp->pg_id == trypid ||
372 p2->p_session->s_sid == trypid) {
373 trypid++;
374 if (trypid >= pidchecked) {
375 PROC_UNLOCK(p2);
376 goto retry;
377 }
378 }
379 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
380 pidchecked = p2->p_pid;
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->s_sid > trypid &&
385 pidchecked > p2->p_session->s_sid)
386 pidchecked = p2->p_session->s_sid;
387 PROC_UNLOCK(p2);
388 }
389 if (!doingzomb) {
390 doingzomb = 1;
391 p2 = LIST_FIRST(&zombproc);
392 goto again;
393 }
394 }
395
396 /*
397 * RFHIGHPID does not mess with the lastpid counter during boot.
398 */
399 if (flags & RFHIGHPID)
400 pidchecked = 0;
401 else
402 lastpid = trypid;
403
404 p2 = newproc;
405 p2->p_state = PRS_NEW; /* protect against others */
406 p2->p_pid = trypid;
407 LIST_INSERT_HEAD(&allproc, p2, p_list);
408 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
409 sx_xunlock(&allproc_lock);
410
411 /*
412 * Malloc things while we don't hold any locks.
413 */
414 if (flags & RFSIGSHARE)
415 newsigacts = NULL;
416 else
417 newsigacts = sigacts_alloc();
418
419 /*
420 * Copy filedesc.
421 */
422 if (flags & RFCFDG)
423 fd = fdinit(td->td_proc->p_fd);
424 else if (flags & RFFDG) {
425 FILEDESC_LOCK(p1->p_fd);
426 fd = fdcopy(td->td_proc->p_fd);
427 FILEDESC_UNLOCK(p1->p_fd);
428 } else
429 fd = fdshare(p1->p_fd);
430
431 /*
432 * Make a proc table entry for the new process.
433 * Start by zeroing the section of proc that is zero-initialized,
434 * then copy the section that is copied directly from the parent.
435 */
436 td2 = FIRST_THREAD_IN_PROC(p2);
437 kg2 = FIRST_KSEGRP_IN_PROC(p2);
438 ke2 = FIRST_KSE_IN_KSEGRP(kg2);
439
440 /* Allocate and switch to an alternate kstack if specified */
441 if (pages != 0)
442 pmap_new_altkstack(td2, pages);
443
444 PROC_LOCK(p2);
445 PROC_LOCK(p1);
446
447 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start))
448
449 bzero(&p2->p_startzero,
450 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero));
451 bzero(&ke2->ke_startzero,
452 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero));
453 bzero(&td2->td_startzero,
454 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero));
455 bzero(&kg2->kg_startzero,
456 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero));
457
458 bcopy(&p1->p_startcopy, &p2->p_startcopy,
459 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy));
460 bcopy(&td->td_startcopy, &td2->td_startcopy,
461 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy));
462 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
463 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy));
464 #undef RANGEOF
465
466 /* Set up the thread as an active thread (as if runnable). */
467 ke2->ke_state = KES_THREAD;
468 ke2->ke_thread = td2;
469 td2->td_kse = ke2;
470
471 /*
472 * Duplicate sub-structures as needed.
473 * Increase reference counts on shared objects.
474 * The p_stats substruct is set in vm_forkproc.
475 */
476 p2->p_flag = 0;
477 if (p1->p_flag & P_PROFIL)
478 startprofclock(p2);
479 mtx_lock_spin(&sched_lock);
480 p2->p_sflag = PS_INMEM;
481 /*
482 * Allow the scheduler to adjust the priority of the child and
483 * parent while we hold the sched_lock.
484 */
485 sched_fork(p1, p2);
486
487 mtx_unlock_spin(&sched_lock);
488 p2->p_ucred = crhold(td->td_ucred);
489 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */
490
491 pargs_hold(p2->p_args);
492
493 if (flags & RFSIGSHARE) {
494 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
495 } else {
496 sigacts_copy(newsigacts, p1->p_sigacts);
497 p2->p_sigacts = newsigacts;
498 }
499 if (flags & RFLINUXTHPN)
500 p2->p_sigparent = SIGUSR1;
501 else
502 p2->p_sigparent = SIGCHLD;
503
504 /* Bump references to the text vnode (for procfs) */
505 p2->p_textvp = p1->p_textvp;
506 if (p2->p_textvp)
507 VREF(p2->p_textvp);
508 p2->p_fd = fd;
509 PROC_UNLOCK(p1);
510 PROC_UNLOCK(p2);
511
512 /*
513 * p_limit is copy-on-write, bump refcnt,
514 */
515 p2->p_limit = p1->p_limit;
516 p2->p_limit->p_refcnt++;
517
518 /*
519 * Setup linkage for kernel based threading
520 */
521 if((flags & RFTHREAD) != 0) {
522 mtx_lock(&ppeers_lock);
523 p2->p_peers = p1->p_peers;
524 p1->p_peers = p2;
525 p2->p_leader = p1->p_leader;
526 mtx_unlock(&ppeers_lock);
527 PROC_LOCK(p1->p_leader);
528 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
529 PROC_UNLOCK(p1->p_leader);
530 /*
531 * The task leader is exiting, so process p1 is
532 * going to be killed shortly. Since p1 obviously
533 * isn't dead yet, we know that the leader is either
534 * sending SIGKILL's to all the processes in this
535 * task or is sleeping waiting for all the peers to
536 * exit. We let p1 complete the fork, but we need
537 * to go ahead and kill the new process p2 since
538 * the task leader may not get a chance to send
539 * SIGKILL to it. We leave it on the list so that
540 * the task leader will wait for this new process
541 * to commit suicide.
542 */
543 PROC_LOCK(p2);
544 psignal(p2, SIGKILL);
545 PROC_UNLOCK(p2);
546 } else
547 PROC_UNLOCK(p1->p_leader);
548 } else {
549 p2->p_peers = NULL;
550 p2->p_leader = p2;
551 }
552
553 sx_xlock(&proctree_lock);
554 PGRP_LOCK(p1->p_pgrp);
555 PROC_LOCK(p2);
556 PROC_LOCK(p1);
557
558 /*
559 * Preserve some more flags in subprocess. P_PROFIL has already
560 * been preserved.
561 */
562 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
563 SESS_LOCK(p1->p_session);
564 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
565 p2->p_flag |= P_CONTROLT;
566 SESS_UNLOCK(p1->p_session);
567 if (flags & RFPPWAIT)
568 p2->p_flag |= P_PPWAIT;
569
570 LIST_INSERT_AFTER(p1, p2, p_pglist);
571 PGRP_UNLOCK(p1->p_pgrp);
572 LIST_INIT(&p2->p_children);
573
574 callout_init(&p2->p_itcallout, 1);
575
576 #ifdef KTRACE
577 /*
578 * Copy traceflag and tracefile if enabled.
579 */
580 mtx_lock(&ktrace_mtx);
581 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
582 if (p1->p_traceflag & KTRFAC_INHERIT) {
583 p2->p_traceflag = p1->p_traceflag;
584 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
585 VREF(p2->p_tracevp);
586 KASSERT(p1->p_tracecred != NULL,
587 ("ktrace vnode with no cred"));
588 p2->p_tracecred = crhold(p1->p_tracecred);
589 }
590 }
591 mtx_unlock(&ktrace_mtx);
592 #endif
593
594 /*
595 * If PF_FORK is set, the child process inherits the
596 * procfs ioctl flags from its parent.
597 */
598 if (p1->p_pfsflags & PF_FORK) {
599 p2->p_stops = p1->p_stops;
600 p2->p_pfsflags = p1->p_pfsflags;
601 }
602
603 /*
604 * This begins the section where we must prevent the parent
605 * from being swapped.
606 */
607 _PHOLD(p1);
608 PROC_UNLOCK(p1);
609
610 /*
611 * Attach the new process to its parent.
612 *
613 * If RFNOWAIT is set, the newly created process becomes a child
614 * of init. This effectively disassociates the child from the
615 * parent.
616 */
617 if (flags & RFNOWAIT)
618 pptr = initproc;
619 else
620 pptr = p1;
621 p2->p_pptr = pptr;
622 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
623 sx_xunlock(&proctree_lock);
624
625 /* Inform accounting that we have forked. */
626 p2->p_acflag = AFORK;
627 PROC_UNLOCK(p2);
628
629 /*
630 * Finish creating the child process. It will return via a different
631 * execution path later. (ie: directly into user mode)
632 */
633 vm_forkproc(td, p2, td2, flags);
634
635 if (flags == (RFFDG | RFPROC)) {
636 cnt.v_forks++;
637 cnt.v_forkpages += p2->p_vmspace->vm_dsize +
638 p2->p_vmspace->vm_ssize;
639 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
640 cnt.v_vforks++;
641 cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
642 p2->p_vmspace->vm_ssize;
643 } else if (p1 == &proc0) {
644 cnt.v_kthreads++;
645 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
646 p2->p_vmspace->vm_ssize;
647 } else {
648 cnt.v_rforks++;
649 cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
650 p2->p_vmspace->vm_ssize;
651 }
652
653 /*
654 * Both processes are set up, now check if any loadable modules want
655 * to adjust anything.
656 * What if they have an error? XXX
657 */
658 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
659
660 /*
661 * If RFSTOPPED not requested, make child runnable and add to
662 * run queue.
663 */
664 microuptime(&p2->p_stats->p_start);
665 if ((flags & RFSTOPPED) == 0) {
666 mtx_lock_spin(&sched_lock);
667 p2->p_state = PRS_NORMAL;
668 TD_SET_CAN_RUN(td2);
669 setrunqueue(td2);
670 mtx_unlock_spin(&sched_lock);
671 }
672
673 /*
674 * Now can be swapped.
675 */
676 PROC_LOCK(p1);
677 _PRELE(p1);
678
679 /*
680 * tell any interested parties about the new process
681 */
682 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
683 PROC_UNLOCK(p1);
684
685 /*
686 * Preserve synchronization semantics of vfork. If waiting for
687 * child to exec or exit, set P_PPWAIT on child, and sleep on our
688 * proc (in case of exit).
689 */
690 PROC_LOCK(p2);
691 while (p2->p_flag & P_PPWAIT)
692 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
693 PROC_UNLOCK(p2);
694
695 /*
696 * If other threads are waiting, let them continue now
697 */
698 if (p1->p_flag & P_THREADED) {
699 PROC_LOCK(p1);
700 thread_single_end();
701 PROC_UNLOCK(p1);
702 }
703
704 /*
705 * Return child proc pointer to parent.
706 */
707 mtx_unlock(&Giant);
708 *procp = p2;
709 return (0);
710 fail:
711 sx_xunlock(&allproc_lock);
712 uma_zfree(proc_zone, newproc);
713 if (p1->p_flag & P_THREADED) {
714 PROC_LOCK(p1);
715 thread_single_end();
716 PROC_UNLOCK(p1);
717 }
718 tsleep(&forksleep, PUSER, "fork", hz / 2);
719 mtx_unlock(&Giant);
720 return (error);
721 }
722
723 /*
724 * Handle the return of a child process from fork1(). This function
725 * is called from the MD fork_trampoline() entry point.
726 */
727 void
728 fork_exit(callout, arg, frame)
729 void (*callout)(void *, struct trapframe *);
730 void *arg;
731 struct trapframe *frame;
732 {
733 struct thread *td;
734 struct proc *p;
735
736 if ((td = PCPU_GET(deadthread))) {
737 PCPU_SET(deadthread, NULL);
738 thread_stash(td);
739 }
740 td = curthread;
741 p = td->td_proc;
742 td->td_oncpu = PCPU_GET(cpuid);
743 p->p_state = PRS_NORMAL;
744 /*
745 * Finish setting up thread glue. We need to initialize
746 * the thread into a td_critnest=1 state. Some platforms
747 * may have already partially or fully initialized td_critnest
748 * and/or td_md.md_savecrit (when applciable).
749 *
750 * see <arch>/<arch>/critical.c
751 */
752 sched_lock.mtx_lock = (uintptr_t)td;
753 sched_lock.mtx_recurse = 0;
754 cpu_critical_fork_exit();
755 CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid,
756 p->p_comm);
757 if (PCPU_GET(switchtime.sec) == 0)
758 binuptime(PCPU_PTR(switchtime));
759 PCPU_SET(switchticks, ticks);
760 mtx_unlock_spin(&sched_lock);
761
762 /*
763 * cpu_set_fork_handler intercepts this function call to
764 * have this call a non-return function to stay in kernel mode.
765 * initproc has its own fork handler, but it does return.
766 */
767 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
768 callout(arg, frame);
769
770 /*
771 * Check if a kernel thread misbehaved and returned from its main
772 * function.
773 */
774 PROC_LOCK(p);
775 if (p->p_flag & P_KTHREAD) {
776 PROC_UNLOCK(p);
777 mtx_lock(&Giant);
778 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
779 p->p_comm, p->p_pid);
780 kthread_exit(0);
781 }
782 PROC_UNLOCK(p);
783 #ifdef DIAGNOSTIC
784 cred_free_thread(td);
785 #endif
786 mtx_assert(&Giant, MA_NOTOWNED);
787 }
788
789 /*
790 * Simplified back end of syscall(), used when returning from fork()
791 * directly into user mode. Giant is not held on entry, and must not
792 * be held on return. This function is passed in to fork_exit() as the
793 * first parameter and is called when returning to a new userland process.
794 */
795 void
796 fork_return(td, frame)
797 struct thread *td;
798 struct trapframe *frame;
799 {
800
801 userret(td, frame, 0);
802 #ifdef KTRACE
803 if (KTRPOINT(td, KTR_SYSRET))
804 ktrsysret(SYS_fork, 0, 0);
805 #endif
806 mtx_assert(&Giant, MA_NOTOWNED);
807 }
Cache object: 3a252b69ecbadf983bdd1e2f8919d53b
|