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_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_mac.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/kthread.h>
51 #include <sys/sysctl.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/pioctl.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sched.h>
60 #include <sys/syscall.h>
61 #include <sys/vmmeter.h>
62 #include <sys/vnode.h>
63 #include <sys/acct.h>
64 #include <sys/ktr.h>
65 #include <sys/ktrace.h>
66 #include <sys/unistd.h>
67 #include <sys/sdt.h>
68 #include <sys/sx.h>
69 #include <sys/signalvar.h>
70
71 #include <security/audit/audit.h>
72 #include <security/mac/mac_framework.h>
73
74 #include <vm/vm.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_extern.h>
78 #include <vm/uma.h>
79
80 #ifdef KDTRACE_HOOKS
81 #include <sys/dtrace_bsd.h>
82 dtrace_fork_func_t dtrace_fasttrap_fork;
83 #endif
84
85 SDT_PROVIDER_DECLARE(proc);
86 SDT_PROBE_DEFINE(proc, kernel, , create);
87 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
88 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
90
91 #ifndef _SYS_SYSPROTO_H_
92 struct fork_args {
93 int dummy;
94 };
95 #endif
96
97 /* ARGSUSED */
98 int
99 fork(td, uap)
100 struct thread *td;
101 struct fork_args *uap;
102 {
103 int error;
104 struct proc *p2;
105
106 error = fork1(td, RFFDG | RFPROC, 0, &p2);
107 if (error == 0) {
108 td->td_retval[0] = p2->p_pid;
109 td->td_retval[1] = 0;
110 }
111 return (error);
112 }
113
114 /* ARGSUSED */
115 int
116 vfork(td, uap)
117 struct thread *td;
118 struct vfork_args *uap;
119 {
120 int error;
121 struct proc *p2;
122
123 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
124 if (error == 0) {
125 td->td_retval[0] = p2->p_pid;
126 td->td_retval[1] = 0;
127 }
128 return (error);
129 }
130
131 int
132 rfork(td, uap)
133 struct thread *td;
134 struct rfork_args *uap;
135 {
136 struct proc *p2;
137 int error;
138
139 /* Don't allow kernel-only flags. */
140 if ((uap->flags & RFKERNELONLY) != 0)
141 return (EINVAL);
142
143 AUDIT_ARG(fflags, uap->flags);
144 error = fork1(td, uap->flags, 0, &p2);
145 if (error == 0) {
146 td->td_retval[0] = p2 ? p2->p_pid : 0;
147 td->td_retval[1] = 0;
148 }
149 return (error);
150 }
151
152 int nprocs = 1; /* process 0 */
153 int lastpid = 0;
154 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
155 "Last used PID");
156
157 /*
158 * Random component to lastpid generation. We mix in a random factor to make
159 * it a little harder to predict. We sanity check the modulus value to avoid
160 * doing it in critical paths. Don't let it be too small or we pointlessly
161 * waste randomness entropy, and don't let it be impossibly large. Using a
162 * modulus that is too big causes a LOT more process table scans and slows
163 * down fork processing as the pidchecked caching is defeated.
164 */
165 static int randompid = 0;
166
167 static int
168 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
169 {
170 int error, pid;
171
172 error = sysctl_wire_old_buffer(req, sizeof(int));
173 if (error != 0)
174 return(error);
175 sx_xlock(&allproc_lock);
176 pid = randompid;
177 error = sysctl_handle_int(oidp, &pid, 0, req);
178 if (error == 0 && req->newptr != NULL) {
179 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
180 pid = PID_MAX - 100;
181 else if (pid < 2) /* NOP */
182 pid = 0;
183 else if (pid < 100) /* Make it reasonable */
184 pid = 100;
185 randompid = pid;
186 }
187 sx_xunlock(&allproc_lock);
188 return (error);
189 }
190
191 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
192 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
193
194 int
195 fork1(td, flags, pages, procp)
196 struct thread *td;
197 int flags;
198 int pages;
199 struct proc **procp;
200 {
201 struct proc *p1, *p2, *pptr;
202 struct proc *newproc;
203 int ok, trypid;
204 static int curfail, pidchecked = 0;
205 static struct timeval lastfail;
206 struct filedesc *fd;
207 struct filedesc_to_leader *fdtol;
208 struct thread *td2;
209 struct sigacts *newsigacts;
210 struct vmspace *vm2;
211 int error;
212
213 /* Can't copy and clear. */
214 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
215 return (EINVAL);
216
217 p1 = td->td_proc;
218
219 /*
220 * Here we don't create a new process, but we divorce
221 * certain parts of a process from itself.
222 */
223 if ((flags & RFPROC) == 0) {
224 if ((p1->p_flag & P_HADTHREADS) &&
225 (flags & (RFCFDG | RFFDG))) {
226 PROC_LOCK(p1);
227 if (thread_single(SINGLE_BOUNDARY)) {
228 PROC_UNLOCK(p1);
229 return (ERESTART);
230 }
231 PROC_UNLOCK(p1);
232 }
233
234 error = vm_forkproc(td, NULL, NULL, NULL, flags);
235 if (error)
236 goto norfproc_fail;
237
238 /*
239 * Close all file descriptors.
240 */
241 if (flags & RFCFDG) {
242 struct filedesc *fdtmp;
243 fdtmp = fdinit(td->td_proc->p_fd);
244 fdfree(td);
245 p1->p_fd = fdtmp;
246 }
247
248 /*
249 * Unshare file descriptors (from parent).
250 */
251 if (flags & RFFDG)
252 fdunshare(p1, td);
253
254 norfproc_fail:
255 if ((p1->p_flag & P_HADTHREADS) &&
256 (flags & (RFCFDG | RFFDG))) {
257 PROC_LOCK(p1);
258 thread_single_end();
259 PROC_UNLOCK(p1);
260 }
261 *procp = NULL;
262 return (error);
263 }
264
265 vm2 = NULL;
266 /* Allocate new proc. */
267 newproc = uma_zalloc(proc_zone, M_WAITOK);
268 if (TAILQ_EMPTY(&newproc->p_threads)) {
269 td2 = thread_alloc();
270 if (td2 == NULL) {
271 error = ENOMEM;
272 goto fail1;
273 }
274 proc_linkup(newproc, td2);
275 sched_newproc(newproc, td2);
276 } else
277 td2 = FIRST_THREAD_IN_PROC(newproc);
278
279 /* Allocate and switch to an alternate kstack if specified. */
280 if (pages != 0) {
281 if (!vm_thread_new_altkstack(td2, pages)) {
282 error = ENOMEM;
283 goto fail1;
284 }
285 }
286 if ((flags & RFMEM) == 0) {
287 vm2 = vmspace_fork(p1->p_vmspace);
288 if (vm2 == NULL) {
289 error = ENOMEM;
290 goto fail1;
291 }
292 }
293 #ifdef MAC
294 mac_init_proc(newproc);
295 #endif
296 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
297 STAILQ_INIT(&newproc->p_ktr);
298
299 /* We have to lock the process tree while we look for a pid. */
300 sx_slock(&proctree_lock);
301
302 /*
303 * Although process entries are dynamically created, we still keep
304 * a global limit on the maximum number we will create. Don't allow
305 * a nonprivileged user to use the last ten processes; don't let root
306 * exceed the limit. The variable nprocs is the current number of
307 * processes, maxproc is the limit.
308 */
309 sx_xlock(&allproc_lock);
310 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
311 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
312 error = EAGAIN;
313 goto fail;
314 }
315
316 /*
317 * Increment the count of procs running with this uid. Don't allow
318 * a nonprivileged user to exceed their current limit.
319 *
320 * XXXRW: Can we avoid privilege here if it's not needed?
321 */
322 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
323 if (error == 0)
324 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
325 else {
326 PROC_LOCK(p1);
327 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
328 lim_cur(p1, RLIMIT_NPROC));
329 PROC_UNLOCK(p1);
330 }
331 if (!ok) {
332 error = EAGAIN;
333 goto fail;
334 }
335
336 /*
337 * Increment the nprocs resource before blocking can occur. There
338 * are hard-limits as to the number of processes that can run.
339 */
340 nprocs++;
341
342 /*
343 * Find an unused process ID. We remember a range of unused IDs
344 * ready to use (from lastpid+1 through pidchecked-1).
345 *
346 * If RFHIGHPID is set (used during system boot), do not allocate
347 * low-numbered pids.
348 */
349 trypid = lastpid + 1;
350 if (flags & RFHIGHPID) {
351 if (trypid < 10)
352 trypid = 10;
353 } else {
354 if (randompid)
355 trypid += arc4random() % randompid;
356 }
357 retry:
358 /*
359 * If the process ID prototype has wrapped around,
360 * restart somewhat above 0, as the low-numbered procs
361 * tend to include daemons that don't exit.
362 */
363 if (trypid >= PID_MAX) {
364 trypid = trypid % PID_MAX;
365 if (trypid < 100)
366 trypid += 100;
367 pidchecked = 0;
368 }
369 if (trypid >= pidchecked) {
370 int doingzomb = 0;
371
372 pidchecked = PID_MAX;
373 /*
374 * Scan the active and zombie procs to check whether this pid
375 * is in use. Remember the lowest pid that's greater
376 * than trypid, so we can avoid checking for a while.
377 */
378 p2 = LIST_FIRST(&allproc);
379 again:
380 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
381 while (p2->p_pid == trypid ||
382 (p2->p_pgrp != NULL &&
383 (p2->p_pgrp->pg_id == trypid ||
384 (p2->p_session != NULL &&
385 p2->p_session->s_sid == trypid)))) {
386 trypid++;
387 if (trypid >= pidchecked)
388 goto retry;
389 }
390 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
391 pidchecked = p2->p_pid;
392 if (p2->p_pgrp != NULL) {
393 if (p2->p_pgrp->pg_id > trypid &&
394 pidchecked > p2->p_pgrp->pg_id)
395 pidchecked = p2->p_pgrp->pg_id;
396 if (p2->p_session != NULL &&
397 p2->p_session->s_sid > trypid &&
398 pidchecked > p2->p_session->s_sid)
399 pidchecked = p2->p_session->s_sid;
400 }
401 }
402 if (!doingzomb) {
403 doingzomb = 1;
404 p2 = LIST_FIRST(&zombproc);
405 goto again;
406 }
407 }
408 sx_sunlock(&proctree_lock);
409
410 /*
411 * RFHIGHPID does not mess with the lastpid counter during boot.
412 */
413 if (flags & RFHIGHPID)
414 pidchecked = 0;
415 else
416 lastpid = trypid;
417
418 p2 = newproc;
419 p2->p_state = PRS_NEW; /* protect against others */
420 p2->p_pid = trypid;
421 /*
422 * Allow the scheduler to initialize the child.
423 */
424 thread_lock(td);
425 sched_fork(td, td2);
426 thread_unlock(td);
427 AUDIT_ARG(pid, p2->p_pid);
428 LIST_INSERT_HEAD(&allproc, p2, p_list);
429 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
430
431 PROC_LOCK(p2);
432 PROC_LOCK(p1);
433
434 sx_xunlock(&allproc_lock);
435
436 bcopy(&p1->p_startcopy, &p2->p_startcopy,
437 __rangeof(struct proc, p_startcopy, p_endcopy));
438 pargs_hold(p2->p_args);
439 PROC_UNLOCK(p1);
440
441 bzero(&p2->p_startzero,
442 __rangeof(struct proc, p_startzero, p_endzero));
443
444 p2->p_ucred = crhold(td->td_ucred);
445 PROC_UNLOCK(p2);
446
447 /*
448 * Malloc things while we don't hold any locks.
449 */
450 if (flags & RFSIGSHARE)
451 newsigacts = NULL;
452 else
453 newsigacts = sigacts_alloc();
454
455 /*
456 * Copy filedesc.
457 */
458 if (flags & RFCFDG) {
459 fd = fdinit(p1->p_fd);
460 fdtol = NULL;
461 } else if (flags & RFFDG) {
462 fd = fdcopy(p1->p_fd);
463 fdtol = NULL;
464 } else {
465 fd = fdshare(p1->p_fd);
466 if (p1->p_fdtol == NULL)
467 p1->p_fdtol =
468 filedesc_to_leader_alloc(NULL,
469 NULL,
470 p1->p_leader);
471 if ((flags & RFTHREAD) != 0) {
472 /*
473 * Shared file descriptor table and
474 * shared process leaders.
475 */
476 fdtol = p1->p_fdtol;
477 FILEDESC_XLOCK(p1->p_fd);
478 fdtol->fdl_refcount++;
479 FILEDESC_XUNLOCK(p1->p_fd);
480 } else {
481 /*
482 * Shared file descriptor table, and
483 * different process leaders
484 */
485 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
486 p1->p_fd,
487 p2);
488 }
489 }
490 /*
491 * Make a proc table entry for the new process.
492 * Start by zeroing the section of proc that is zero-initialized,
493 * then copy the section that is copied directly from the parent.
494 */
495
496 PROC_LOCK(p2);
497 PROC_LOCK(p1);
498
499 bzero(&td2->td_startzero,
500 __rangeof(struct thread, td_startzero, td_endzero));
501
502 bcopy(&td->td_startcopy, &td2->td_startcopy,
503 __rangeof(struct thread, td_startcopy, td_endcopy));
504
505 td2->td_sigstk = td->td_sigstk;
506 td2->td_sigmask = td->td_sigmask;
507 td2->td_flags = TDF_INMEM;
508
509 /*
510 * Duplicate sub-structures as needed.
511 * Increase reference counts on shared objects.
512 */
513 p2->p_flag = P_INMEM;
514 p2->p_swtick = ticks;
515 if (p1->p_flag & P_PROFIL)
516 startprofclock(p2);
517 td2->td_ucred = crhold(p2->p_ucred);
518
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;
529
530 p2->p_textvp = p1->p_textvp;
531 p2->p_fd = fd;
532 p2->p_fdtol = fdtol;
533
534 /*
535 * p_limit is copy-on-write. Bump its refcount.
536 */
537 lim_fork(p1, p2);
538
539 pstats_fork(p1->p_stats, p2->p_stats);
540
541 PROC_UNLOCK(p1);
542 PROC_UNLOCK(p2);
543
544 /* Bump references to the text vnode (for procfs) */
545 if (p2->p_textvp)
546 vref(p2->p_textvp);
547
548 /*
549 * Set up linkage for kernel based threading.
550 */
551 if ((flags & RFTHREAD) != 0) {
552 mtx_lock(&ppeers_lock);
553 p2->p_peers = p1->p_peers;
554 p1->p_peers = p2;
555 p2->p_leader = p1->p_leader;
556 mtx_unlock(&ppeers_lock);
557 PROC_LOCK(p1->p_leader);
558 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
559 PROC_UNLOCK(p1->p_leader);
560 /*
561 * The task leader is exiting, so process p1 is
562 * going to be killed shortly. Since p1 obviously
563 * isn't dead yet, we know that the leader is either
564 * sending SIGKILL's to all the processes in this
565 * task or is sleeping waiting for all the peers to
566 * exit. We let p1 complete the fork, but we need
567 * to go ahead and kill the new process p2 since
568 * the task leader may not get a chance to send
569 * SIGKILL to it. We leave it on the list so that
570 * the task leader will wait for this new process
571 * to commit suicide.
572 */
573 PROC_LOCK(p2);
574 psignal(p2, SIGKILL);
575 PROC_UNLOCK(p2);
576 } else
577 PROC_UNLOCK(p1->p_leader);
578 } else {
579 p2->p_peers = NULL;
580 p2->p_leader = p2;
581 }
582
583 sx_xlock(&proctree_lock);
584 PGRP_LOCK(p1->p_pgrp);
585 PROC_LOCK(p2);
586 PROC_LOCK(p1);
587
588 /*
589 * Preserve some more flags in subprocess. P_PROFIL has already
590 * been preserved.
591 */
592 p2->p_flag |= p1->p_flag & P_SUGID;
593 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
594 SESS_LOCK(p1->p_session);
595 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
596 p2->p_flag |= P_CONTROLT;
597 SESS_UNLOCK(p1->p_session);
598 if (flags & RFPPWAIT)
599 p2->p_flag |= P_PPWAIT;
600
601 p2->p_pgrp = p1->p_pgrp;
602 LIST_INSERT_AFTER(p1, p2, p_pglist);
603 PGRP_UNLOCK(p1->p_pgrp);
604 LIST_INIT(&p2->p_children);
605
606 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
607
608 #ifdef KTRACE
609 /*
610 * Copy traceflag and tracefile if enabled.
611 */
612 mtx_lock(&ktrace_mtx);
613 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
614 if (p1->p_traceflag & KTRFAC_INHERIT) {
615 p2->p_traceflag = p1->p_traceflag;
616 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
617 VREF(p2->p_tracevp);
618 KASSERT(p1->p_tracecred != NULL,
619 ("ktrace vnode with no cred"));
620 p2->p_tracecred = crhold(p1->p_tracecred);
621 }
622 }
623 mtx_unlock(&ktrace_mtx);
624 #endif
625
626 /*
627 * If PF_FORK is set, the child process inherits the
628 * procfs ioctl flags from its parent.
629 */
630 if (p1->p_pfsflags & PF_FORK) {
631 p2->p_stops = p1->p_stops;
632 p2->p_pfsflags = p1->p_pfsflags;
633 }
634
635 #ifdef KDTRACE_HOOKS
636 /*
637 * Tell the DTrace fasttrap provider about the new process
638 * if it has registered an interest.
639 */
640 if (dtrace_fasttrap_fork)
641 dtrace_fasttrap_fork(p1, p2);
642 #endif
643
644 /*
645 * This begins the section where we must prevent the parent
646 * from being swapped.
647 */
648 _PHOLD(p1);
649 PROC_UNLOCK(p1);
650
651 /*
652 * Attach the new process to its parent.
653 *
654 * If RFNOWAIT is set, the newly created process becomes a child
655 * of init. This effectively disassociates the child from the
656 * parent.
657 */
658 if (flags & RFNOWAIT)
659 pptr = initproc;
660 else
661 pptr = p1;
662 p2->p_pptr = pptr;
663 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
664 sx_xunlock(&proctree_lock);
665
666 /* Inform accounting that we have forked. */
667 p2->p_acflag = AFORK;
668 PROC_UNLOCK(p2);
669
670 /*
671 * Finish creating the child process. It will return via a different
672 * execution path later. (ie: directly into user mode)
673 */
674 vm_forkproc(td, p2, td2, vm2, flags);
675
676 if (flags == (RFFDG | RFPROC)) {
677 PCPU_INC(cnt.v_forks);
678 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
679 p2->p_vmspace->vm_ssize);
680 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
681 PCPU_INC(cnt.v_vforks);
682 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
683 p2->p_vmspace->vm_ssize);
684 } else if (p1 == &proc0) {
685 PCPU_INC(cnt.v_kthreads);
686 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
687 p2->p_vmspace->vm_ssize);
688 } else {
689 PCPU_INC(cnt.v_rforks);
690 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
691 p2->p_vmspace->vm_ssize);
692 }
693
694 /*
695 * Both processes are set up, now check if any loadable modules want
696 * to adjust anything.
697 * What if they have an error? XXX
698 */
699 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
700
701 /*
702 * Set the child start time and mark the process as being complete.
703 */
704 microuptime(&p2->p_stats->p_start);
705 PROC_SLOCK(p2);
706 p2->p_state = PRS_NORMAL;
707 PROC_SUNLOCK(p2);
708
709 /*
710 * If RFSTOPPED not requested, make child runnable and add to
711 * run queue.
712 */
713 if ((flags & RFSTOPPED) == 0) {
714 thread_lock(td2);
715 TD_SET_CAN_RUN(td2);
716 sched_add(td2, SRQ_BORING);
717 thread_unlock(td2);
718 }
719
720 /*
721 * Now can be swapped.
722 */
723 PROC_LOCK(p1);
724 _PRELE(p1);
725 PROC_UNLOCK(p1);
726
727 /*
728 * Tell any interested parties about the new process.
729 */
730 knote_fork(&p1->p_klist, p2->p_pid);
731 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
732
733 /*
734 * Preserve synchronization semantics of vfork. If waiting for
735 * child to exec or exit, set P_PPWAIT on child, and sleep on our
736 * proc (in case of exit).
737 */
738 PROC_LOCK(p2);
739 while (p2->p_flag & P_PPWAIT)
740 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
741 PROC_UNLOCK(p2);
742
743 /*
744 * Return child proc pointer to parent.
745 */
746 *procp = p2;
747 return (0);
748 fail:
749 sx_sunlock(&proctree_lock);
750 if (ppsratecheck(&lastfail, &curfail, 1))
751 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
752 td->td_ucred->cr_ruid);
753 sx_xunlock(&allproc_lock);
754 #ifdef MAC
755 mac_destroy_proc(newproc);
756 #endif
757 fail1:
758 if (vm2 != NULL)
759 vmspace_free(vm2);
760 uma_zfree(proc_zone, newproc);
761 pause("fork", hz / 2);
762 return (error);
763 }
764
765 /*
766 * Handle the return of a child process from fork1(). This function
767 * is called from the MD fork_trampoline() entry point.
768 */
769 void
770 fork_exit(callout, arg, frame)
771 void (*callout)(void *, struct trapframe *);
772 void *arg;
773 struct trapframe *frame;
774 {
775 struct proc *p;
776 struct thread *td;
777 struct thread *dtd;
778
779 td = curthread;
780 p = td->td_proc;
781 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
782
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 sched_fork_exit(td);
787 /*
788 * Processes normally resume in mi_switch() after being
789 * cpu_switch()'ed to, but when children start up they arrive here
790 * instead, so we must do much the same things as mi_switch() would.
791 */
792 if ((dtd = PCPU_GET(deadthread))) {
793 PCPU_SET(deadthread, NULL);
794 thread_stash(dtd);
795 }
796 thread_unlock(td);
797
798 /*
799 * cpu_set_fork_handler intercepts this function call to
800 * have this call a non-return function to stay in kernel mode.
801 * initproc has its own fork handler, but it does return.
802 */
803 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
804 callout(arg, frame);
805
806 /*
807 * Check if a kernel thread misbehaved and returned from its main
808 * function.
809 */
810 if (p->p_flag & P_KTHREAD) {
811 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
812 p->p_comm, p->p_pid);
813 kthread_exit(0);
814 }
815 mtx_assert(&Giant, MA_NOTOWNED);
816
817 EVENTHANDLER_INVOKE(schedtail, p);
818 }
819
820 /*
821 * Simplified back end of syscall(), used when returning from fork()
822 * directly into user mode. Giant is not held on entry, and must not
823 * be held on return. This function is passed in to fork_exit() as the
824 * first parameter and is called when returning to a new userland process.
825 */
826 void
827 fork_return(td, frame)
828 struct thread *td;
829 struct trapframe *frame;
830 {
831
832 userret(td, frame);
833 #ifdef KTRACE
834 if (KTRPOINT(td, KTR_SYSRET))
835 ktrsysret(SYS_fork, 0, 0);
836 #endif
837 mtx_assert(&Giant, MA_NOTOWNED);
838 }
Cache object: cbb660e5bc9361036e0ae94f22c64b9e
|