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: releng/6.1/sys/kern/kern_fork.c 158179 2006-04-30 16:44:43Z cvs2svn $");
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/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>
67 #include <sys/signalvar.h>
68
69 #include <vm/vm.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_extern.h>
73 #include <vm/uma.h>
74
75
76 #ifndef _SYS_SYSPROTO_H_
77 struct fork_args {
78 int dummy;
79 };
80 #endif
81
82 static int forksleep; /* Place for fork1() to sleep on. */
83
84 /*
85 * MPSAFE
86 */
87 /* ARGSUSED */
88 int
89 fork(td, uap)
90 struct thread *td;
91 struct fork_args *uap;
92 {
93 int error;
94 struct proc *p2;
95
96 error = fork1(td, RFFDG | RFPROC, 0, &p2);
97 if (error == 0) {
98 td->td_retval[0] = p2->p_pid;
99 td->td_retval[1] = 0;
100 }
101 return (error);
102 }
103
104 /*
105 * MPSAFE
106 */
107 /* ARGSUSED */
108 int
109 vfork(td, uap)
110 struct thread *td;
111 struct vfork_args *uap;
112 {
113 int error;
114 struct proc *p2;
115
116 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
117 if (error == 0) {
118 td->td_retval[0] = p2->p_pid;
119 td->td_retval[1] = 0;
120 }
121 return (error);
122 }
123
124 /*
125 * MPSAFE
126 */
127 int
128 rfork(td, uap)
129 struct thread *td;
130 struct rfork_args *uap;
131 {
132 struct proc *p2;
133 int error;
134
135 /* Don't allow kernel-only flags. */
136 if ((uap->flags & RFKERNELONLY) != 0)
137 return (EINVAL);
138
139 error = fork1(td, uap->flags, 0, &p2);
140 if (error == 0) {
141 td->td_retval[0] = p2 ? p2->p_pid : 0;
142 td->td_retval[1] = 0;
143 }
144 return (error);
145 }
146
147 int nprocs = 1; /* process 0 */
148 int lastpid = 0;
149 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
150 "Last used PID");
151
152 /*
153 * Random component to lastpid generation. We mix in a random factor to make
154 * it a little harder to predict. We sanity check the modulus value to avoid
155 * doing it in critical paths. Don't let it be too small or we pointlessly
156 * waste randomness entropy, and don't let it be impossibly large. Using a
157 * modulus that is too big causes a LOT more process table scans and slows
158 * down fork processing as the pidchecked caching is defeated.
159 */
160 static int randompid = 0;
161
162 static int
163 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
164 {
165 int error, pid;
166
167 error = sysctl_wire_old_buffer(req, sizeof(int));
168 if (error != 0)
169 return(error);
170 sx_xlock(&allproc_lock);
171 pid = randompid;
172 error = sysctl_handle_int(oidp, &pid, 0, req);
173 if (error == 0 && req->newptr != NULL) {
174 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
175 pid = PID_MAX - 100;
176 else if (pid < 2) /* NOP */
177 pid = 0;
178 else if (pid < 100) /* Make it reasonable */
179 pid = 100;
180 randompid = pid;
181 }
182 sx_xunlock(&allproc_lock);
183 return (error);
184 }
185
186 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
187 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
188
189 int
190 fork1(td, flags, pages, procp)
191 struct thread *td;
192 int flags;
193 int pages;
194 struct proc **procp;
195 {
196 struct proc *p1, *p2, *pptr;
197 uid_t uid;
198 struct proc *newproc;
199 int ok, trypid;
200 static int curfail, pidchecked = 0;
201 static struct timeval lastfail;
202 struct filedesc *fd;
203 struct filedesc_to_leader *fdtol;
204 struct thread *td2;
205 struct ksegrp *kg2;
206 struct sigacts *newsigacts;
207 int error;
208
209 /* Can't copy and clear. */
210 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
211 return (EINVAL);
212
213 p1 = td->td_proc;
214
215 /*
216 * Here we don't create a new process, but we divorce
217 * certain parts of a process from itself.
218 */
219 if ((flags & RFPROC) == 0) {
220 if ((p1->p_flag & P_HADTHREADS) &&
221 (flags & (RFCFDG | RFFDG))) {
222 PROC_LOCK(p1);
223 if (thread_single(SINGLE_BOUNDARY)) {
224 PROC_UNLOCK(p1);
225 return (ERESTART);
226 }
227 PROC_UNLOCK(p1);
228 }
229
230 vm_forkproc(td, NULL, NULL, flags);
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 fdunshare(p1, td);
246
247 if ((p1->p_flag & P_HADTHREADS) &&
248 (flags & (RFCFDG | RFFDG))) {
249 PROC_LOCK(p1);
250 thread_single_end();
251 PROC_UNLOCK(p1);
252 }
253 *procp = NULL;
254 return (0);
255 }
256
257 /*
258 * Note 1:1 allows for forking with one thread coming out on the
259 * other side with the expectation that the process is about to
260 * exec.
261 */
262 if (p1->p_flag & P_HADTHREADS) {
263 /*
264 * Idle the other threads for a second.
265 * Since the user space is copied, it must remain stable.
266 * In addition, all threads (from the user perspective)
267 * need to either be suspended or in the kernel,
268 * where they will try restart in the parent and will
269 * be aborted in the child.
270 */
271 PROC_LOCK(p1);
272 if (thread_single(SINGLE_NO_EXIT)) {
273 /* Abort. Someone else is single threading before us. */
274 PROC_UNLOCK(p1);
275 return (ERESTART);
276 }
277 PROC_UNLOCK(p1);
278 /*
279 * All other activity in this process
280 * is now suspended at the user boundary,
281 * (or other safe places if we think of any).
282 */
283 }
284
285 /* Allocate new proc. */
286 newproc = uma_zalloc(proc_zone, M_WAITOK);
287 #ifdef MAC
288 mac_init_proc(newproc);
289 #endif
290 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
291
292 /* We have to lock the process tree while we look for a pid. */
293 sx_slock(&proctree_lock);
294
295 /*
296 * Although process entries are dynamically created, we still keep
297 * a global limit on the maximum number we will create. Don't allow
298 * a nonprivileged user to use the last ten processes; don't let root
299 * exceed the limit. The variable nprocs is the current number of
300 * processes, maxproc is the limit.
301 */
302 sx_xlock(&allproc_lock);
303 uid = td->td_ucred->cr_ruid;
304 if ((nprocs >= maxproc - 10 &&
305 suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
306 nprocs >= maxproc) {
307 error = EAGAIN;
308 goto fail;
309 }
310
311 /*
312 * Increment the count of procs running with this uid. Don't allow
313 * a nonprivileged user to exceed their current limit.
314 */
315 PROC_LOCK(p1);
316 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
317 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0);
318 PROC_UNLOCK(p1);
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 PROC_LOCK(p2);
370 while (p2->p_pid == trypid ||
371 (p2->p_pgrp != NULL &&
372 (p2->p_pgrp->pg_id == trypid ||
373 (p2->p_session != NULL &&
374 p2->p_session->s_sid == trypid)))) {
375 trypid++;
376 if (trypid >= pidchecked) {
377 PROC_UNLOCK(p2);
378 goto retry;
379 }
380 }
381 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
382 pidchecked = p2->p_pid;
383 if (p2->p_pgrp != NULL) {
384 if (p2->p_pgrp->pg_id > trypid &&
385 pidchecked > p2->p_pgrp->pg_id)
386 pidchecked = p2->p_pgrp->pg_id;
387 if (p2->p_session != NULL &&
388 p2->p_session->s_sid > trypid &&
389 pidchecked > p2->p_session->s_sid)
390 pidchecked = p2->p_session->s_sid;
391 }
392 PROC_UNLOCK(p2);
393 }
394 if (!doingzomb) {
395 doingzomb = 1;
396 p2 = LIST_FIRST(&zombproc);
397 goto again;
398 }
399 }
400 sx_sunlock(&proctree_lock);
401
402 /*
403 * RFHIGHPID does not mess with the lastpid counter during boot.
404 */
405 if (flags & RFHIGHPID)
406 pidchecked = 0;
407 else
408 lastpid = trypid;
409
410 p2 = newproc;
411 p2->p_state = PRS_NEW; /* protect against others */
412 p2->p_pid = trypid;
413 LIST_INSERT_HEAD(&allproc, p2, p_list);
414 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
415 sx_xunlock(&allproc_lock);
416
417 /*
418 * Malloc things while we don't hold any locks.
419 */
420 if (flags & RFSIGSHARE)
421 newsigacts = NULL;
422 else
423 newsigacts = sigacts_alloc();
424
425 /*
426 * Copy filedesc.
427 */
428 if (flags & RFCFDG) {
429 fd = fdinit(p1->p_fd);
430 fdtol = NULL;
431 } else if (flags & RFFDG) {
432 fd = fdcopy(p1->p_fd);
433 fdtol = NULL;
434 } else {
435 fd = fdshare(p1->p_fd);
436 if (p1->p_fdtol == NULL)
437 p1->p_fdtol =
438 filedesc_to_leader_alloc(NULL,
439 NULL,
440 p1->p_leader);
441 if ((flags & RFTHREAD) != 0) {
442 /*
443 * Shared file descriptor table and
444 * shared process leaders.
445 */
446 fdtol = p1->p_fdtol;
447 FILEDESC_LOCK_FAST(p1->p_fd);
448 fdtol->fdl_refcount++;
449 FILEDESC_UNLOCK_FAST(p1->p_fd);
450 } else {
451 /*
452 * Shared file descriptor table, and
453 * different process leaders
454 */
455 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
456 p1->p_fd,
457 p2);
458 }
459 }
460 /*
461 * Make a proc table entry for the new process.
462 * Start by zeroing the section of proc that is zero-initialized,
463 * then copy the section that is copied directly from the parent.
464 */
465 td2 = FIRST_THREAD_IN_PROC(p2);
466 kg2 = FIRST_KSEGRP_IN_PROC(p2);
467
468 /* Allocate and switch to an alternate kstack if specified. */
469 if (pages != 0)
470 vm_thread_new_altkstack(td2, pages);
471
472 PROC_LOCK(p2);
473 PROC_LOCK(p1);
474
475 bzero(&p2->p_startzero,
476 __rangeof(struct proc, p_startzero, p_endzero));
477 bzero(&td2->td_startzero,
478 __rangeof(struct thread, td_startzero, td_endzero));
479 bzero(&kg2->kg_startzero,
480 __rangeof(struct ksegrp, kg_startzero, kg_endzero));
481
482 bcopy(&p1->p_startcopy, &p2->p_startcopy,
483 __rangeof(struct proc, p_startcopy, p_endcopy));
484 bcopy(&td->td_startcopy, &td2->td_startcopy,
485 __rangeof(struct thread, td_startcopy, td_endcopy));
486 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
487 __rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
488
489 td2->td_sigstk = td->td_sigstk;
490 td2->td_sigmask = td->td_sigmask;
491
492 /*
493 * Duplicate sub-structures as needed.
494 * Increase reference counts on shared objects.
495 */
496 p2->p_flag = 0;
497 if (p1->p_flag & P_PROFIL)
498 startprofclock(p2);
499 mtx_lock_spin(&sched_lock);
500 p2->p_sflag = PS_INMEM;
501 /*
502 * Allow the scheduler to adjust the priority of the child and
503 * parent while we hold the sched_lock.
504 */
505 sched_fork(td, td2);
506
507 mtx_unlock_spin(&sched_lock);
508 p2->p_ucred = crhold(td->td_ucred);
509 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */
510
511 pargs_hold(p2->p_args);
512
513 if (flags & RFSIGSHARE) {
514 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
515 } else {
516 sigacts_copy(newsigacts, p1->p_sigacts);
517 p2->p_sigacts = newsigacts;
518 }
519 if (flags & RFLINUXTHPN)
520 p2->p_sigparent = SIGUSR1;
521 else
522 p2->p_sigparent = SIGCHLD;
523
524 p2->p_textvp = p1->p_textvp;
525 p2->p_fd = fd;
526 p2->p_fdtol = fdtol;
527
528 /*
529 * p_limit is copy-on-write. Bump its refcount.
530 */
531 p2->p_limit = lim_hold(p1->p_limit);
532
533 pstats_fork(p1->p_stats, p2->p_stats);
534
535 PROC_UNLOCK(p1);
536 PROC_UNLOCK(p2);
537
538 /* Bump references to the text vnode (for procfs) */
539 if (p2->p_textvp)
540 vref(p2->p_textvp);
541
542 /*
543 * Set up linkage for kernel based threading.
544 */
545 if ((flags & RFTHREAD) != 0) {
546 mtx_lock(&ppeers_lock);
547 p2->p_peers = p1->p_peers;
548 p1->p_peers = p2;
549 p2->p_leader = p1->p_leader;
550 mtx_unlock(&ppeers_lock);
551 PROC_LOCK(p1->p_leader);
552 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
553 PROC_UNLOCK(p1->p_leader);
554 /*
555 * The task leader is exiting, so process p1 is
556 * going to be killed shortly. Since p1 obviously
557 * isn't dead yet, we know that the leader is either
558 * sending SIGKILL's to all the processes in this
559 * task or is sleeping waiting for all the peers to
560 * exit. We let p1 complete the fork, but we need
561 * to go ahead and kill the new process p2 since
562 * the task leader may not get a chance to send
563 * SIGKILL to it. We leave it on the list so that
564 * the task leader will wait for this new process
565 * to commit suicide.
566 */
567 PROC_LOCK(p2);
568 psignal(p2, SIGKILL);
569 PROC_UNLOCK(p2);
570 } else
571 PROC_UNLOCK(p1->p_leader);
572 } else {
573 p2->p_peers = NULL;
574 p2->p_leader = p2;
575 }
576
577 sx_xlock(&proctree_lock);
578 PGRP_LOCK(p1->p_pgrp);
579 PROC_LOCK(p2);
580 PROC_LOCK(p1);
581
582 /*
583 * Preserve some more flags in subprocess. P_PROFIL has already
584 * been preserved.
585 */
586 p2->p_flag |= p1->p_flag & P_SUGID;
587 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
588 SESS_LOCK(p1->p_session);
589 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
590 p2->p_flag |= P_CONTROLT;
591 SESS_UNLOCK(p1->p_session);
592 if (flags & RFPPWAIT)
593 p2->p_flag |= P_PPWAIT;
594
595 p2->p_pgrp = p1->p_pgrp;
596 LIST_INSERT_AFTER(p1, p2, p_pglist);
597 PGRP_UNLOCK(p1->p_pgrp);
598 LIST_INIT(&p2->p_children);
599
600 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
601
602 #ifdef KTRACE
603 /*
604 * Copy traceflag and tracefile if enabled.
605 */
606 mtx_lock(&ktrace_mtx);
607 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
608 if (p1->p_traceflag & KTRFAC_INHERIT) {
609 p2->p_traceflag = p1->p_traceflag;
610 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
611 VREF(p2->p_tracevp);
612 KASSERT(p1->p_tracecred != NULL,
613 ("ktrace vnode with no cred"));
614 p2->p_tracecred = crhold(p1->p_tracecred);
615 }
616 }
617 mtx_unlock(&ktrace_mtx);
618 #endif
619
620 /*
621 * If PF_FORK is set, the child process inherits the
622 * procfs ioctl flags from its parent.
623 */
624 if (p1->p_pfsflags & PF_FORK) {
625 p2->p_stops = p1->p_stops;
626 p2->p_pfsflags = p1->p_pfsflags;
627 }
628
629 /*
630 * This begins the section where we must prevent the parent
631 * from being swapped.
632 */
633 _PHOLD(p1);
634 PROC_UNLOCK(p1);
635
636 /*
637 * Attach the new process to its parent.
638 *
639 * If RFNOWAIT is set, the newly created process becomes a child
640 * of init. This effectively disassociates the child from the
641 * parent.
642 */
643 if (flags & RFNOWAIT)
644 pptr = initproc;
645 else
646 pptr = p1;
647 p2->p_pptr = pptr;
648 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
649 sx_xunlock(&proctree_lock);
650
651 /* Inform accounting that we have forked. */
652 p2->p_acflag = AFORK;
653 PROC_UNLOCK(p2);
654
655 /*
656 * Finish creating the child process. It will return via a different
657 * execution path later. (ie: directly into user mode)
658 */
659 vm_forkproc(td, p2, td2, flags);
660
661 if (flags == (RFFDG | RFPROC)) {
662 atomic_add_int(&cnt.v_forks, 1);
663 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize +
664 p2->p_vmspace->vm_ssize);
665 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
666 atomic_add_int(&cnt.v_vforks, 1);
667 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
668 p2->p_vmspace->vm_ssize);
669 } else if (p1 == &proc0) {
670 atomic_add_int(&cnt.v_kthreads, 1);
671 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
672 p2->p_vmspace->vm_ssize);
673 } else {
674 atomic_add_int(&cnt.v_rforks, 1);
675 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
676 p2->p_vmspace->vm_ssize);
677 }
678
679 /*
680 * Both processes are set up, now check if any loadable modules want
681 * to adjust anything.
682 * What if they have an error? XXX
683 */
684 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
685
686 /*
687 * Set the child start time and mark the process as being complete.
688 */
689 microuptime(&p2->p_stats->p_start);
690 mtx_lock_spin(&sched_lock);
691 p2->p_state = PRS_NORMAL;
692
693 /*
694 * If RFSTOPPED not requested, make child runnable and add to
695 * run queue.
696 */
697 if ((flags & RFSTOPPED) == 0) {
698 TD_SET_CAN_RUN(td2);
699 setrunqueue(td2, SRQ_BORING);
700 }
701 mtx_unlock_spin(&sched_lock);
702
703 /*
704 * Now can be swapped.
705 */
706 PROC_LOCK(p1);
707 _PRELE(p1);
708
709 /*
710 * Tell any interested parties about the new process.
711 */
712 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
713
714 PROC_UNLOCK(p1);
715
716 /*
717 * Preserve synchronization semantics of vfork. If waiting for
718 * child to exec or exit, set P_PPWAIT on child, and sleep on our
719 * proc (in case of exit).
720 */
721 PROC_LOCK(p2);
722 while (p2->p_flag & P_PPWAIT)
723 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
724 PROC_UNLOCK(p2);
725
726 /*
727 * If other threads are waiting, let them continue now.
728 */
729 if (p1->p_flag & P_HADTHREADS) {
730 PROC_LOCK(p1);
731 thread_single_end();
732 PROC_UNLOCK(p1);
733 }
734
735 /*
736 * Return child proc pointer to parent.
737 */
738 *procp = p2;
739 return (0);
740 fail:
741 sx_sunlock(&proctree_lock);
742 if (ppsratecheck(&lastfail, &curfail, 1))
743 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
744 uid);
745 sx_xunlock(&allproc_lock);
746 #ifdef MAC
747 mac_destroy_proc(newproc);
748 #endif
749 uma_zfree(proc_zone, newproc);
750 if (p1->p_flag & P_HADTHREADS) {
751 PROC_LOCK(p1);
752 thread_single_end();
753 PROC_UNLOCK(p1);
754 }
755 tsleep(&forksleep, PUSER, "fork", hz / 2);
756 return (error);
757 }
758
759 /*
760 * Handle the return of a child process from fork1(). This function
761 * is called from the MD fork_trampoline() entry point.
762 */
763 void
764 fork_exit(callout, arg, frame)
765 void (*callout)(void *, struct trapframe *);
766 void *arg;
767 struct trapframe *frame;
768 {
769 struct proc *p;
770 struct thread *td;
771
772 /*
773 * Finish setting up thread glue so that it begins execution in a
774 * non-nested critical section with sched_lock held but not recursed.
775 */
776 td = curthread;
777 p = td->td_proc;
778 td->td_oncpu = PCPU_GET(cpuid);
779 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
780
781 sched_lock.mtx_lock = (uintptr_t)td;
782 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
783 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
784 td, td->td_sched, p->p_pid, p->p_comm);
785
786 /*
787 * Processes normally resume in mi_switch() after being
788 * cpu_switch()'ed to, but when children start up they arrive here
789 * instead, so we must do much the same things as mi_switch() would.
790 */
791
792 if ((td = PCPU_GET(deadthread))) {
793 PCPU_SET(deadthread, NULL);
794 thread_stash(td);
795 }
796 td = curthread;
797 mtx_unlock_spin(&sched_lock);
798
799 /*
800 * cpu_set_fork_handler intercepts this function call to
801 * have this call a non-return function to stay in kernel mode.
802 * initproc has its own fork handler, but it does return.
803 */
804 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
805 callout(arg, frame);
806
807 /*
808 * Check if a kernel thread misbehaved and returned from its main
809 * function.
810 */
811 PROC_LOCK(p);
812 if (p->p_flag & P_KTHREAD) {
813 PROC_UNLOCK(p);
814 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
815 p->p_comm, p->p_pid);
816 kthread_exit(0);
817 }
818 PROC_UNLOCK(p);
819 mtx_assert(&Giant, MA_NOTOWNED);
820 }
821
822 /*
823 * Simplified back end of syscall(), used when returning from fork()
824 * directly into user mode. Giant is not held on entry, and must not
825 * be held on return. This function is passed in to fork_exit() as the
826 * first parameter and is called when returning to a new userland process.
827 */
828 void
829 fork_return(td, frame)
830 struct thread *td;
831 struct trapframe *frame;
832 {
833
834 userret(td, frame, 0);
835 #ifdef KTRACE
836 if (KTRPOINT(td, KTR_SYSRET))
837 ktrsysret(SYS_fork, 0, 0);
838 #endif
839 mtx_assert(&Giant, MA_NOTOWNED);
840 }
Cache object: f4de9b208454ac4be1fca6e44cf5387d
|