ia64/xen-unstable

view freebsd-5.3-xen-sparse/kern/kern_fork.c @ 7127:805ee053e61f

VIRQs and IPIs on VCPU#0 are automatically re-bound on save/restore.
Signed-off-by: Keir Fraser <keir@xensource.com>
author kaf24@firebug.cl.cam.ac.uk
date Thu Sep 29 13:05:43 2005 +0100 (2005-09-29)
parents a01199a95070
children
line source
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 */
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: src/sys/kern/kern_fork.c,v 1.234.2.4 2004/09/18 04:11:35 julian Exp $");
40 #include "opt_ktrace.h"
41 #include "opt_mac.h"
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/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/resourcevar.h>
57 #include <sys/sched.h>
58 #include <sys/syscall.h>
59 #include <sys/vmmeter.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/unistd.h>
66 #include <sys/sx.h>
68 #include <vm/vm.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_extern.h>
72 #include <vm/uma.h>
74 #include <sys/user.h>
75 #include <machine/critical.h>
77 #ifndef _SYS_SYSPROTO_H_
78 struct fork_args {
79 int dummy;
80 };
81 #endif
83 static int forksleep; /* Place for fork1() to sleep on. */
85 /*
86 * MPSAFE
87 */
88 /* ARGSUSED */
89 int
90 fork(td, uap)
91 struct thread *td;
92 struct fork_args *uap;
93 {
94 int error;
95 struct proc *p2;
97 error = fork1(td, RFFDG | RFPROC, 0, &p2);
98 if (error == 0) {
99 td->td_retval[0] = p2->p_pid;
100 td->td_retval[1] = 0;
101 }
102 return (error);
103 }
105 /*
106 * MPSAFE
107 */
108 /* ARGSUSED */
109 int
110 vfork(td, uap)
111 struct thread *td;
112 struct vfork_args *uap;
113 {
114 int error;
115 struct proc *p2;
117 error = fork1(td, RFFDG | RFPROC /* | RFPPWAIT | RFMEM */, 0, &p2);
118 if (error == 0) {
119 td->td_retval[0] = p2->p_pid;
120 td->td_retval[1] = 0;
121 }
122 return (error);
123 }
125 /*
126 * MPSAFE
127 */
128 int
129 rfork(td, uap)
130 struct thread *td;
131 struct rfork_args *uap;
132 {
133 struct proc *p2;
134 int error;
136 /* Don't allow kernel-only flags. */
137 if ((uap->flags & RFKERNELONLY) != 0)
138 return (EINVAL);
140 error = fork1(td, uap->flags, 0, &p2);
141 if (error == 0) {
142 td->td_retval[0] = p2 ? p2->p_pid : 0;
143 td->td_retval[1] = 0;
144 }
145 return (error);
146 }
148 int nprocs = 1; /* process 0 */
149 int lastpid = 0;
150 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
151 "Last used PID");
153 /*
154 * Random component to lastpid generation. We mix in a random factor to make
155 * it a little harder to predict. We sanity check the modulus value to avoid
156 * doing it in critical paths. Don't let it be too small or we pointlessly
157 * waste randomness entropy, and don't let it be impossibly large. Using a
158 * modulus that is too big causes a LOT more process table scans and slows
159 * down fork processing as the pidchecked caching is defeated.
160 */
161 static int randompid = 0;
163 static int
164 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
165 {
166 int error, pid;
168 error = sysctl_wire_old_buffer(req, sizeof(int));
169 if (error != 0)
170 return(error);
171 sx_xlock(&allproc_lock);
172 pid = randompid;
173 error = sysctl_handle_int(oidp, &pid, 0, req);
174 if (error == 0 && req->newptr != NULL) {
175 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
176 pid = PID_MAX - 100;
177 else if (pid < 2) /* NOP */
178 pid = 0;
179 else if (pid < 100) /* Make it reasonable */
180 pid = 100;
181 randompid = pid;
182 }
183 sx_xunlock(&allproc_lock);
184 return (error);
185 }
187 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
188 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
190 int
191 fork1(td, flags, pages, procp)
192 struct thread *td;
193 int flags;
194 int pages;
195 struct proc **procp;
196 {
197 struct proc *p1, *p2, *pptr;
198 uid_t uid;
199 struct proc *newproc;
200 int ok, trypid;
201 static int curfail, pidchecked = 0;
202 static struct timeval lastfail;
203 struct filedesc *fd;
204 struct filedesc_to_leader *fdtol;
205 struct thread *td2;
206 struct ksegrp *kg2;
207 struct sigacts *newsigacts;
208 int error;
210 /* Can't copy and clear. */
211 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
212 return (EINVAL);
214 p1 = td->td_proc;
216 /*
217 * Here we don't create a new process, but we divorce
218 * certain parts of a process from itself.
219 */
220 if ((flags & RFPROC) == 0) {
221 mtx_lock(&Giant);
222 vm_forkproc(td, NULL, NULL, flags);
223 mtx_unlock(&Giant);
225 /*
226 * Close all file descriptors.
227 */
228 if (flags & RFCFDG) {
229 struct filedesc *fdtmp;
230 FILEDESC_LOCK(td->td_proc->p_fd);
231 fdtmp = fdinit(td->td_proc->p_fd);
232 FILEDESC_UNLOCK(td->td_proc->p_fd);
233 fdfree(td);
234 p1->p_fd = fdtmp;
235 }
237 /*
238 * Unshare file descriptors (from parent).
239 */
240 if (flags & RFFDG) {
241 FILEDESC_LOCK(p1->p_fd);
242 if (p1->p_fd->fd_refcnt > 1) {
243 struct filedesc *newfd;
245 newfd = fdcopy(td->td_proc->p_fd);
246 FILEDESC_UNLOCK(p1->p_fd);
247 fdfree(td);
248 p1->p_fd = newfd;
249 } else
250 FILEDESC_UNLOCK(p1->p_fd);
251 }
252 *procp = NULL;
253 return (0);
254 }
256 /*
257 * Note 1:1 allows for forking with one thread coming out on the
258 * other side with the expectation that the process is about to
259 * exec.
260 */
261 if (p1->p_flag & P_HADTHREADS) {
262 /*
263 * Idle the other threads for a second.
264 * Since the user space is copied, it must remain stable.
265 * In addition, all threads (from the user perspective)
266 * need to either be suspended or in the kernel,
267 * where they will try restart in the parent and will
268 * be aborted in the child.
269 */
270 PROC_LOCK(p1);
271 if (thread_single(SINGLE_NO_EXIT)) {
272 /* Abort. Someone else is single threading before us. */
273 PROC_UNLOCK(p1);
274 return (ERESTART);
275 }
276 PROC_UNLOCK(p1);
277 /*
278 * All other activity in this process
279 * is now suspended at the user boundary,
280 * (or other safe places if we think of any).
281 */
282 }
284 /* Allocate new proc. */
285 newproc = uma_zalloc(proc_zone, M_WAITOK);
286 #ifdef MAC
287 mac_init_proc(newproc);
288 #endif
289 knlist_init(&newproc->p_klist, &newproc->p_mtx);
291 /* We have to lock the process tree while we look for a pid. */
292 sx_slock(&proctree_lock);
294 /*
295 * Although process entries are dynamically created, we still keep
296 * a global limit on the maximum number we will create. Don't allow
297 * a nonprivileged user to use the last ten processes; don't let root
298 * exceed the limit. The variable nprocs is the current number of
299 * processes, maxproc is the limit.
300 */
301 sx_xlock(&allproc_lock);
302 uid = td->td_ucred->cr_ruid;
303 if ((nprocs >= maxproc - 10 &&
304 suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
305 nprocs >= maxproc) {
306 error = EAGAIN;
307 goto fail;
308 }
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) ? lim_cur(p1, RLIMIT_NPROC) : 0);
317 PROC_UNLOCK(p1);
318 if (!ok) {
319 error = EAGAIN;
320 goto fail;
321 }
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++;
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 } else {
341 if (randompid)
342 trypid += arc4random() % randompid;
343 }
344 retry:
345 /*
346 * If the process ID prototype has wrapped around,
347 * restart somewhat above 0, as the low-numbered procs
348 * tend to include daemons that don't exit.
349 */
350 if (trypid >= PID_MAX) {
351 trypid = trypid % PID_MAX;
352 if (trypid < 100)
353 trypid += 100;
354 pidchecked = 0;
355 }
356 if (trypid >= pidchecked) {
357 int doingzomb = 0;
359 pidchecked = PID_MAX;
360 /*
361 * Scan the active and zombie procs to check whether this pid
362 * is in use. Remember the lowest pid that's greater
363 * than trypid, so we can avoid checking for a while.
364 */
365 p2 = LIST_FIRST(&allproc);
366 again:
367 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
368 PROC_LOCK(p2);
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 PROC_UNLOCK(p2);
377 goto retry;
378 }
379 }
380 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
381 pidchecked = p2->p_pid;
382 if (p2->p_pgrp != NULL) {
383 if (p2->p_pgrp->pg_id > trypid &&
384 pidchecked > p2->p_pgrp->pg_id)
385 pidchecked = p2->p_pgrp->pg_id;
386 if (p2->p_session != NULL &&
387 p2->p_session->s_sid > trypid &&
388 pidchecked > p2->p_session->s_sid)
389 pidchecked = p2->p_session->s_sid;
390 }
391 PROC_UNLOCK(p2);
392 }
393 if (!doingzomb) {
394 doingzomb = 1;
395 p2 = LIST_FIRST(&zombproc);
396 goto again;
397 }
398 }
399 sx_sunlock(&proctree_lock);
401 /*
402 * RFHIGHPID does not mess with the lastpid counter during boot.
403 */
404 if (flags & RFHIGHPID)
405 pidchecked = 0;
406 else
407 lastpid = trypid;
409 p2 = newproc;
410 p2->p_state = PRS_NEW; /* protect against others */
411 p2->p_pid = trypid;
412 LIST_INSERT_HEAD(&allproc, p2, p_list);
413 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
414 sx_xunlock(&allproc_lock);
416 /*
417 * Malloc things while we don't hold any locks.
418 */
419 if (flags & RFSIGSHARE)
420 newsigacts = NULL;
421 else
422 newsigacts = sigacts_alloc();
424 /*
425 * Copy filedesc.
426 */
427 if (flags & RFCFDG) {
428 FILEDESC_LOCK(td->td_proc->p_fd);
429 fd = fdinit(td->td_proc->p_fd);
430 FILEDESC_UNLOCK(td->td_proc->p_fd);
431 fdtol = NULL;
432 } else if (flags & RFFDG) {
433 FILEDESC_LOCK(p1->p_fd);
434 fd = fdcopy(td->td_proc->p_fd);
435 FILEDESC_UNLOCK(p1->p_fd);
436 fdtol = NULL;
437 } else {
438 fd = fdshare(p1->p_fd);
439 if (p1->p_fdtol == NULL)
440 p1->p_fdtol =
441 filedesc_to_leader_alloc(NULL,
442 NULL,
443 p1->p_leader);
444 if ((flags & RFTHREAD) != 0) {
445 /*
446 * Shared file descriptor table and
447 * shared process leaders.
448 */
449 fdtol = p1->p_fdtol;
450 FILEDESC_LOCK(p1->p_fd);
451 fdtol->fdl_refcount++;
452 FILEDESC_UNLOCK(p1->p_fd);
453 } else {
454 /*
455 * Shared file descriptor table, and
456 * different process leaders
457 */
458 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
459 p1->p_fd,
460 p2);
461 }
462 }
463 /*
464 * Make a proc table entry for the new process.
465 * Start by zeroing the section of proc that is zero-initialized,
466 * then copy the section that is copied directly from the parent.
467 */
468 td2 = FIRST_THREAD_IN_PROC(p2);
469 kg2 = FIRST_KSEGRP_IN_PROC(p2);
471 /* Allocate and switch to an alternate kstack if specified. */
472 if (pages != 0)
473 vm_thread_new_altkstack(td2, pages);
475 PROC_LOCK(p2);
476 PROC_LOCK(p1);
478 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start))
480 bzero(&p2->p_startzero,
481 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero));
482 bzero(&td2->td_startzero,
483 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero));
484 bzero(&kg2->kg_startzero,
485 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero));
487 bcopy(&p1->p_startcopy, &p2->p_startcopy,
488 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy));
489 bcopy(&td->td_startcopy, &td2->td_startcopy,
490 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy));
491 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
492 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy));
493 #undef RANGEOF
495 td2->td_sigstk = td->td_sigstk;
497 /*
498 * Duplicate sub-structures as needed.
499 * Increase reference counts on shared objects.
500 * The p_stats substruct is set in vm_forkproc.
501 */
502 p2->p_flag = 0;
503 if (p1->p_flag & P_PROFIL)
504 startprofclock(p2);
505 mtx_lock_spin(&sched_lock);
506 p2->p_sflag = PS_INMEM;
507 /*
508 * Allow the scheduler to adjust the priority of the child and
509 * parent while we hold the sched_lock.
510 */
511 sched_fork(td, td2);
513 mtx_unlock_spin(&sched_lock);
514 p2->p_ucred = crhold(td->td_ucred);
515 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */
517 pargs_hold(p2->p_args);
519 if (flags & RFSIGSHARE) {
520 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
521 } else {
522 sigacts_copy(newsigacts, p1->p_sigacts);
523 p2->p_sigacts = newsigacts;
524 }
525 if (flags & RFLINUXTHPN)
526 p2->p_sigparent = SIGUSR1;
527 else
528 p2->p_sigparent = SIGCHLD;
530 p2->p_textvp = p1->p_textvp;
531 p2->p_fd = fd;
532 p2->p_fdtol = fdtol;
534 /*
535 * p_limit is copy-on-write. Bump its refcount.
536 */
537 p2->p_limit = lim_hold(p1->p_limit);
538 PROC_UNLOCK(p1);
539 PROC_UNLOCK(p2);
541 /* Bump references to the text vnode (for procfs) */
542 if (p2->p_textvp)
543 vref(p2->p_textvp);
545 /*
546 * Set up linkage for kernel based threading.
547 */
548 if ((flags & RFTHREAD) != 0) {
549 mtx_lock(&ppeers_lock);
550 p2->p_peers = p1->p_peers;
551 p1->p_peers = p2;
552 p2->p_leader = p1->p_leader;
553 mtx_unlock(&ppeers_lock);
554 PROC_LOCK(p1->p_leader);
555 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
556 PROC_UNLOCK(p1->p_leader);
557 /*
558 * The task leader is exiting, so process p1 is
559 * going to be killed shortly. Since p1 obviously
560 * isn't dead yet, we know that the leader is either
561 * sending SIGKILL's to all the processes in this
562 * task or is sleeping waiting for all the peers to
563 * exit. We let p1 complete the fork, but we need
564 * to go ahead and kill the new process p2 since
565 * the task leader may not get a chance to send
566 * SIGKILL to it. We leave it on the list so that
567 * the task leader will wait for this new process
568 * to commit suicide.
569 */
570 PROC_LOCK(p2);
571 psignal(p2, SIGKILL);
572 PROC_UNLOCK(p2);
573 } else
574 PROC_UNLOCK(p1->p_leader);
575 } else {
576 p2->p_peers = NULL;
577 p2->p_leader = p2;
578 }
580 sx_xlock(&proctree_lock);
581 PGRP_LOCK(p1->p_pgrp);
582 PROC_LOCK(p2);
583 PROC_LOCK(p1);
585 /*
586 * Preserve some more flags in subprocess. P_PROFIL has already
587 * been preserved.
588 */
589 p2->p_flag |= p1->p_flag & P_SUGID;
590 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
591 SESS_LOCK(p1->p_session);
592 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
593 p2->p_flag |= P_CONTROLT;
594 SESS_UNLOCK(p1->p_session);
595 if (flags & RFPPWAIT)
596 p2->p_flag |= P_PPWAIT;
598 p2->p_pgrp = p1->p_pgrp;
599 LIST_INSERT_AFTER(p1, p2, p_pglist);
600 PGRP_UNLOCK(p1->p_pgrp);
601 LIST_INIT(&p2->p_children);
603 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
605 #ifdef KTRACE
606 /*
607 * Copy traceflag and tracefile if enabled.
608 */
609 mtx_lock(&ktrace_mtx);
610 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
611 if (p1->p_traceflag & KTRFAC_INHERIT) {
612 p2->p_traceflag = p1->p_traceflag;
613 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
614 VREF(p2->p_tracevp);
615 KASSERT(p1->p_tracecred != NULL,
616 ("ktrace vnode with no cred"));
617 p2->p_tracecred = crhold(p1->p_tracecred);
618 }
619 }
620 mtx_unlock(&ktrace_mtx);
621 #endif
623 /*
624 * If PF_FORK is set, the child process inherits the
625 * procfs ioctl flags from its parent.
626 */
627 if (p1->p_pfsflags & PF_FORK) {
628 p2->p_stops = p1->p_stops;
629 p2->p_pfsflags = p1->p_pfsflags;
630 }
632 /*
633 * This begins the section where we must prevent the parent
634 * from being swapped.
635 */
636 _PHOLD(p1);
637 PROC_UNLOCK(p1);
639 /*
640 * Attach the new process to its parent.
641 *
642 * If RFNOWAIT is set, the newly created process becomes a child
643 * of init. This effectively disassociates the child from the
644 * parent.
645 */
646 if (flags & RFNOWAIT)
647 pptr = initproc;
648 else
649 pptr = p1;
650 p2->p_pptr = pptr;
651 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
652 sx_xunlock(&proctree_lock);
654 /* Inform accounting that we have forked. */
655 p2->p_acflag = AFORK;
656 PROC_UNLOCK(p2);
658 /*
659 * Finish creating the child process. It will return via a different
660 * execution path later. (ie: directly into user mode)
661 */
662 mtx_lock(&Giant);
663 vm_forkproc(td, p2, td2, flags);
665 if (flags == (RFFDG | RFPROC)) {
666 cnt.v_forks++;
667 cnt.v_forkpages += p2->p_vmspace->vm_dsize +
668 p2->p_vmspace->vm_ssize;
669 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
670 cnt.v_vforks++;
671 cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
672 p2->p_vmspace->vm_ssize;
673 } else if (p1 == &proc0) {
674 cnt.v_kthreads++;
675 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
676 p2->p_vmspace->vm_ssize;
677 } else {
678 cnt.v_rforks++;
679 cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
680 p2->p_vmspace->vm_ssize;
681 }
682 mtx_unlock(&Giant);
684 /*
685 * Both processes are set up, now check if any loadable modules want
686 * to adjust anything.
687 * What if they have an error? XXX
688 */
689 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
691 /*
692 * Set the child start time and mark the process as being complete.
693 */
694 microuptime(&p2->p_stats->p_start);
695 mtx_lock_spin(&sched_lock);
696 p2->p_state = PRS_NORMAL;
698 /*
699 * If RFSTOPPED not requested, make child runnable and add to
700 * run queue.
701 */
702 if ((flags & RFSTOPPED) == 0) {
703 TD_SET_CAN_RUN(td2);
704 setrunqueue(td2, SRQ_BORING);
705 }
706 mtx_unlock_spin(&sched_lock);
708 /*
709 * Now can be swapped.
710 */
711 PROC_LOCK(p1);
712 _PRELE(p1);
714 /*
715 * Tell any interested parties about the new process.
716 */
717 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
719 PROC_UNLOCK(p1);
721 /*
722 * Preserve synchronization semantics of vfork. If waiting for
723 * child to exec or exit, set P_PPWAIT on child, and sleep on our
724 * proc (in case of exit).
725 */
726 PROC_LOCK(p2);
727 while (p2->p_flag & P_PPWAIT)
728 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
729 PROC_UNLOCK(p2);
731 /*
732 * If other threads are waiting, let them continue now.
733 */
734 if (p1->p_flag & P_HADTHREADS) {
735 PROC_LOCK(p1);
736 thread_single_end();
737 PROC_UNLOCK(p1);
738 }
740 /*
741 * Return child proc pointer to parent.
742 */
743 *procp = p2;
744 return (0);
745 fail:
746 sx_sunlock(&proctree_lock);
747 if (ppsratecheck(&lastfail, &curfail, 1))
748 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
749 uid);
750 sx_xunlock(&allproc_lock);
751 #ifdef MAC
752 mac_destroy_proc(newproc);
753 #endif
754 uma_zfree(proc_zone, newproc);
755 if (p1->p_flag & P_HADTHREADS) {
756 PROC_LOCK(p1);
757 thread_single_end();
758 PROC_UNLOCK(p1);
759 }
760 tsleep(&forksleep, PUSER, "fork", hz / 2);
761 return (error);
762 }
764 /*
765 * Handle the return of a child process from fork1(). This function
766 * is called from the MD fork_trampoline() entry point.
767 */
768 void
769 fork_exit(callout, arg, frame)
770 void (*callout)(void *, struct trapframe *);
771 void *arg;
772 struct trapframe *frame;
773 {
774 struct proc *p;
775 struct thread *td;
777 /*
778 * Finish setting up thread glue so that it begins execution in a
779 * non-nested critical section with sched_lock held but not recursed.
780 */
781 td = curthread;
782 p = td->td_proc;
783 td->td_oncpu = PCPU_GET(cpuid);
784 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
786 sched_lock.mtx_lock = (uintptr_t)td;
787 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
788 cpu_critical_fork_exit();
789 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
790 td, td->td_sched, p->p_pid, p->p_comm);
792 /*
793 * Processes normally resume in mi_switch() after being
794 * cpu_switch()'ed to, but when children start up they arrive here
795 * instead, so we must do much the same things as mi_switch() would.
796 */
798 if ((td = PCPU_GET(deadthread))) {
799 PCPU_SET(deadthread, NULL);
800 thread_stash(td);
801 }
802 td = curthread;
803 mtx_unlock_spin(&sched_lock);
805 /*
806 * cpu_set_fork_handler intercepts this function call to
807 * have this call a non-return function to stay in kernel mode.
808 * initproc has its own fork handler, but it does return.
809 */
810 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
811 callout(arg, frame);
813 /*
814 * Check if a kernel thread misbehaved and returned from its main
815 * function.
816 */
817 PROC_LOCK(p);
818 if (p->p_flag & P_KTHREAD) {
819 PROC_UNLOCK(p);
820 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
821 p->p_comm, p->p_pid);
822 kthread_exit(0);
823 }
824 PROC_UNLOCK(p);
825 mtx_assert(&Giant, MA_NOTOWNED);
826 }
828 /*
829 * Simplified back end of syscall(), used when returning from fork()
830 * directly into user mode. Giant is not held on entry, and must not
831 * be held on return. This function is passed in to fork_exit() as the
832 * first parameter and is called when returning to a new userland process.
833 */
834 void
835 fork_return(td, frame)
836 struct thread *td;
837 struct trapframe *frame;
838 {
840 userret(td, frame, 0);
841 #ifdef KTRACE
842 if (KTRPOINT(td, KTR_SYSRET))
843 ktrsysret(SYS_fork, 0, 0);
844 #endif
845 mtx_assert(&Giant, MA_NOTOWNED);
846 }