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/8.3/sys/kern/kern_fork.c 224211 2011-07-19 10:41:26Z kib $");
39
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_kstack_pages.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/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/sysctl.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mutex.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/pioctl.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
61 #include <sys/syscall.h>
62 #include <sys/vmmeter.h>
63 #include <sys/vnode.h>
64 #include <sys/acct.h>
65 #include <sys/ktr.h>
66 #include <sys/ktrace.h>
67 #include <sys/unistd.h>
68 #include <sys/sdt.h>
69 #include <sys/sx.h>
70 #include <sys/sysent.h>
71 #include <sys/signalvar.h>
72
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75
76 #include <vm/vm.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_extern.h>
80 #include <vm/uma.h>
81
82 #ifdef KDTRACE_HOOKS
83 #include <sys/dtrace_bsd.h>
84 dtrace_fork_func_t dtrace_fasttrap_fork;
85 #endif
86
87 SDT_PROVIDER_DECLARE(proc);
88 SDT_PROBE_DEFINE(proc, kernel, , create, create);
89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
90 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
91 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
92
93 #ifndef _SYS_SYSPROTO_H_
94 struct fork_args {
95 int dummy;
96 };
97 #endif
98
99 /* ARGSUSED */
100 int
101 fork(td, uap)
102 struct thread *td;
103 struct fork_args *uap;
104 {
105 int error;
106 struct proc *p2;
107
108 error = fork1(td, RFFDG | RFPROC, 0, &p2);
109 if (error == 0) {
110 td->td_retval[0] = p2->p_pid;
111 td->td_retval[1] = 0;
112 }
113 return (error);
114 }
115
116 /* ARGSUSED */
117 int
118 vfork(td, uap)
119 struct thread *td;
120 struct vfork_args *uap;
121 {
122 int error, flags;
123 struct proc *p2;
124
125 #ifdef XEN
126 flags = RFFDG | RFPROC; /* validate that this is still an issue */
127 #else
128 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
129 #endif
130 error = fork1(td, flags, 0, &p2);
131 if (error == 0) {
132 td->td_retval[0] = p2->p_pid;
133 td->td_retval[1] = 0;
134 }
135 return (error);
136 }
137
138 int
139 rfork(td, uap)
140 struct thread *td;
141 struct rfork_args *uap;
142 {
143 struct proc *p2;
144 int error;
145
146 /* Don't allow kernel-only flags. */
147 if ((uap->flags & RFKERNELONLY) != 0)
148 return (EINVAL);
149
150 AUDIT_ARG_FFLAGS(uap->flags);
151 error = fork1(td, uap->flags, 0, &p2);
152 if (error == 0) {
153 td->td_retval[0] = p2 ? p2->p_pid : 0;
154 td->td_retval[1] = 0;
155 }
156 return (error);
157 }
158
159 int nprocs = 1; /* process 0 */
160 int lastpid = 0;
161 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
162 "Last used PID");
163
164 /*
165 * Random component to lastpid generation. We mix in a random factor to make
166 * it a little harder to predict. We sanity check the modulus value to avoid
167 * doing it in critical paths. Don't let it be too small or we pointlessly
168 * waste randomness entropy, and don't let it be impossibly large. Using a
169 * modulus that is too big causes a LOT more process table scans and slows
170 * down fork processing as the pidchecked caching is defeated.
171 */
172 static int randompid = 0;
173
174 static int
175 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
176 {
177 int error, pid;
178
179 error = sysctl_wire_old_buffer(req, sizeof(int));
180 if (error != 0)
181 return(error);
182 sx_xlock(&allproc_lock);
183 pid = randompid;
184 error = sysctl_handle_int(oidp, &pid, 0, req);
185 if (error == 0 && req->newptr != NULL) {
186 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
187 pid = PID_MAX - 100;
188 else if (pid < 2) /* NOP */
189 pid = 0;
190 else if (pid < 100) /* Make it reasonable */
191 pid = 100;
192 randompid = pid;
193 }
194 sx_xunlock(&allproc_lock);
195 return (error);
196 }
197
198 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
199 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
200
201 int
202 fork1(td, flags, pages, procp)
203 struct thread *td;
204 int flags;
205 int pages;
206 struct proc **procp;
207 {
208 struct proc *p1, *p2, *pptr;
209 struct proc *newproc;
210 int ok, p2_held, trypid;
211 static int curfail, pidchecked = 0;
212 static struct timeval lastfail;
213 struct filedesc *fd;
214 struct filedesc_to_leader *fdtol;
215 struct thread *td2;
216 struct sigacts *newsigacts;
217 struct vmspace *vm2;
218 vm_ooffset_t mem_charged;
219 int error;
220
221 /* Check for the undefined or unimplemented flags. */
222 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
223 return (EINVAL);
224
225 /* Signal value requires RFTSIGZMB. */
226 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
227 return (EINVAL);
228
229 /* Can't copy and clear. */
230 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
231 return (EINVAL);
232
233 /* Check the validity of the signal number. */
234 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
235 return (EINVAL);
236
237 p2_held = 0;
238 p1 = td->td_proc;
239
240 /*
241 * Here we don't create a new process, but we divorce
242 * certain parts of a process from itself.
243 */
244 if ((flags & RFPROC) == 0) {
245 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
246 (flags & (RFCFDG | RFFDG))) {
247 PROC_LOCK(p1);
248 if (thread_single(SINGLE_BOUNDARY)) {
249 PROC_UNLOCK(p1);
250 return (ERESTART);
251 }
252 PROC_UNLOCK(p1);
253 }
254
255 error = vm_forkproc(td, NULL, NULL, NULL, flags);
256 if (error)
257 goto norfproc_fail;
258
259 /*
260 * Close all file descriptors.
261 */
262 if (flags & RFCFDG) {
263 struct filedesc *fdtmp;
264 fdtmp = fdinit(td->td_proc->p_fd);
265 fdfree(td);
266 p1->p_fd = fdtmp;
267 }
268
269 /*
270 * Unshare file descriptors (from parent).
271 */
272 if (flags & RFFDG)
273 fdunshare(p1, td);
274
275 norfproc_fail:
276 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
277 (flags & (RFCFDG | RFFDG))) {
278 PROC_LOCK(p1);
279 thread_single_end();
280 PROC_UNLOCK(p1);
281 }
282 *procp = NULL;
283 return (error);
284 }
285
286 /*
287 * XXX
288 * We did have single-threading code here
289 * however it proved un-needed and caused problems
290 */
291
292 mem_charged = 0;
293 vm2 = NULL;
294 if (pages == 0)
295 pages = KSTACK_PAGES;
296 /* Allocate new proc. */
297 newproc = uma_zalloc(proc_zone, M_WAITOK);
298 td2 = FIRST_THREAD_IN_PROC(newproc);
299 if (td2 == NULL) {
300 td2 = thread_alloc(pages);
301 if (td2 == NULL) {
302 error = ENOMEM;
303 goto fail1;
304 }
305 proc_linkup(newproc, td2);
306 } else {
307 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
308 if (td2->td_kstack != 0)
309 vm_thread_dispose(td2);
310 if (!thread_alloc_stack(td2, pages)) {
311 error = ENOMEM;
312 goto fail1;
313 }
314 }
315 }
316
317 if ((flags & RFMEM) == 0) {
318 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
319 if (vm2 == NULL) {
320 error = ENOMEM;
321 goto fail1;
322 }
323 if (!swap_reserve(mem_charged)) {
324 /*
325 * The swap reservation failed. The accounting
326 * from the entries of the copied vm2 will be
327 * substracted in vmspace_free(), so force the
328 * reservation there.
329 */
330 swap_reserve_force(mem_charged);
331 error = ENOMEM;
332 goto fail1;
333 }
334 } else
335 vm2 = NULL;
336 #ifdef MAC
337 mac_proc_init(newproc);
338 #endif
339 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
340 STAILQ_INIT(&newproc->p_ktr);
341
342 /* We have to lock the process tree while we look for a pid. */
343 sx_slock(&proctree_lock);
344
345 /*
346 * Although process entries are dynamically created, we still keep
347 * a global limit on the maximum number we will create. Don't allow
348 * a nonprivileged user to use the last ten processes; don't let root
349 * exceed the limit. The variable nprocs is the current number of
350 * processes, maxproc is the limit.
351 */
352 sx_xlock(&allproc_lock);
353 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
354 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
355 error = EAGAIN;
356 goto fail;
357 }
358
359 /*
360 * Increment the count of procs running with this uid. Don't allow
361 * a nonprivileged user to exceed their current limit.
362 *
363 * XXXRW: Can we avoid privilege here if it's not needed?
364 */
365 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
366 if (error == 0)
367 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
368 else {
369 PROC_LOCK(p1);
370 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
371 lim_cur(p1, RLIMIT_NPROC));
372 PROC_UNLOCK(p1);
373 }
374 if (!ok) {
375 error = EAGAIN;
376 goto fail;
377 }
378
379 /*
380 * Increment the nprocs resource before blocking can occur. There
381 * are hard-limits as to the number of processes that can run.
382 */
383 nprocs++;
384
385 /*
386 * Find an unused process ID. We remember a range of unused IDs
387 * ready to use (from lastpid+1 through pidchecked-1).
388 *
389 * If RFHIGHPID is set (used during system boot), do not allocate
390 * low-numbered pids.
391 */
392 trypid = lastpid + 1;
393 if (flags & RFHIGHPID) {
394 if (trypid < 10)
395 trypid = 10;
396 } else {
397 if (randompid)
398 trypid += arc4random() % randompid;
399 }
400 retry:
401 /*
402 * If the process ID prototype has wrapped around,
403 * restart somewhat above 0, as the low-numbered procs
404 * tend to include daemons that don't exit.
405 */
406 if (trypid >= PID_MAX) {
407 trypid = trypid % PID_MAX;
408 if (trypid < 100)
409 trypid += 100;
410 pidchecked = 0;
411 }
412 if (trypid >= pidchecked) {
413 int doingzomb = 0;
414
415 pidchecked = PID_MAX;
416 /*
417 * Scan the active and zombie procs to check whether this pid
418 * is in use. Remember the lowest pid that's greater
419 * than trypid, so we can avoid checking for a while.
420 */
421 p2 = LIST_FIRST(&allproc);
422 again:
423 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
424 while (p2->p_pid == trypid ||
425 (p2->p_pgrp != NULL &&
426 (p2->p_pgrp->pg_id == trypid ||
427 (p2->p_session != NULL &&
428 p2->p_session->s_sid == trypid)))) {
429 trypid++;
430 if (trypid >= pidchecked)
431 goto retry;
432 }
433 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
434 pidchecked = p2->p_pid;
435 if (p2->p_pgrp != NULL) {
436 if (p2->p_pgrp->pg_id > trypid &&
437 pidchecked > p2->p_pgrp->pg_id)
438 pidchecked = p2->p_pgrp->pg_id;
439 if (p2->p_session != NULL &&
440 p2->p_session->s_sid > trypid &&
441 pidchecked > p2->p_session->s_sid)
442 pidchecked = p2->p_session->s_sid;
443 }
444 }
445 if (!doingzomb) {
446 doingzomb = 1;
447 p2 = LIST_FIRST(&zombproc);
448 goto again;
449 }
450 }
451 sx_sunlock(&proctree_lock);
452
453 /*
454 * RFHIGHPID does not mess with the lastpid counter during boot.
455 */
456 if (flags & RFHIGHPID)
457 pidchecked = 0;
458 else
459 lastpid = trypid;
460
461 p2 = newproc;
462 p2->p_state = PRS_NEW; /* protect against others */
463 p2->p_pid = trypid;
464 AUDIT_ARG_PID(p2->p_pid);
465 LIST_INSERT_HEAD(&allproc, p2, p_list);
466 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
467
468 PROC_LOCK(p2);
469 PROC_LOCK(p1);
470
471 sx_xunlock(&allproc_lock);
472
473 bcopy(&p1->p_startcopy, &p2->p_startcopy,
474 __rangeof(struct proc, p_startcopy, p_endcopy));
475 pargs_hold(p2->p_args);
476 PROC_UNLOCK(p1);
477
478 bzero(&p2->p_startzero,
479 __rangeof(struct proc, p_startzero, p_endzero));
480
481 p2->p_ucred = crhold(td->td_ucred);
482
483 /* Tell the prison that we exist. */
484 prison_proc_hold(p2->p_ucred->cr_prison);
485
486 PROC_UNLOCK(p2);
487
488 /*
489 * Malloc things while we don't hold any locks.
490 */
491 if (flags & RFSIGSHARE)
492 newsigacts = NULL;
493 else
494 newsigacts = sigacts_alloc();
495
496 /*
497 * Copy filedesc.
498 */
499 if (flags & RFCFDG) {
500 fd = fdinit(p1->p_fd);
501 fdtol = NULL;
502 } else if (flags & RFFDG) {
503 fd = fdcopy(p1->p_fd);
504 fdtol = NULL;
505 } else {
506 fd = fdshare(p1->p_fd);
507 if (p1->p_fdtol == NULL)
508 p1->p_fdtol =
509 filedesc_to_leader_alloc(NULL,
510 NULL,
511 p1->p_leader);
512 if ((flags & RFTHREAD) != 0) {
513 /*
514 * Shared file descriptor table and
515 * shared process leaders.
516 */
517 fdtol = p1->p_fdtol;
518 FILEDESC_XLOCK(p1->p_fd);
519 fdtol->fdl_refcount++;
520 FILEDESC_XUNLOCK(p1->p_fd);
521 } else {
522 /*
523 * Shared file descriptor table, and
524 * different process leaders
525 */
526 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
527 p1->p_fd,
528 p2);
529 }
530 }
531 /*
532 * Make a proc table entry for the new process.
533 * Start by zeroing the section of proc that is zero-initialized,
534 * then copy the section that is copied directly from the parent.
535 */
536
537 PROC_LOCK(p2);
538 PROC_LOCK(p1);
539
540 bzero(&td2->td_startzero,
541 __rangeof(struct thread, td_startzero, td_endzero));
542 bzero(&td2->td_rux, sizeof(td2->td_rux));
543 td2->td_map_def_user = NULL;
544 td2->td_dbg_forked = 0;
545
546 bcopy(&td->td_startcopy, &td2->td_startcopy,
547 __rangeof(struct thread, td_startcopy, td_endcopy));
548
549 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
550 td2->td_sigstk = td->td_sigstk;
551 td2->td_sigmask = td->td_sigmask;
552 td2->td_flags = TDF_INMEM;
553
554 #ifdef VIMAGE
555 td2->td_vnet = NULL;
556 td2->td_vnet_lpush = NULL;
557 #endif
558
559 /*
560 * Allow the scheduler to initialize the child.
561 */
562 thread_lock(td);
563 sched_fork(td, td2);
564 thread_unlock(td);
565
566 /*
567 * Duplicate sub-structures as needed.
568 * Increase reference counts on shared objects.
569 */
570 p2->p_flag = P_INMEM;
571 p2->p_swtick = ticks;
572 if (p1->p_flag & P_PROFIL)
573 startprofclock(p2);
574 td2->td_ucred = crhold(p2->p_ucred);
575
576 if (flags & RFSIGSHARE) {
577 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
578 } else {
579 sigacts_copy(newsigacts, p1->p_sigacts);
580 p2->p_sigacts = newsigacts;
581 }
582
583 if (flags & RFTSIGZMB)
584 p2->p_sigparent = RFTSIGNUM(flags);
585 else if (flags & RFLINUXTHPN)
586 p2->p_sigparent = SIGUSR1;
587 else
588 p2->p_sigparent = SIGCHLD;
589
590 p2->p_textvp = p1->p_textvp;
591 p2->p_fd = fd;
592 p2->p_fdtol = fdtol;
593
594 /*
595 * p_limit is copy-on-write. Bump its refcount.
596 */
597 lim_fork(p1, p2);
598
599 pstats_fork(p1->p_stats, p2->p_stats);
600
601 PROC_UNLOCK(p1);
602 PROC_UNLOCK(p2);
603
604 /* Bump references to the text vnode (for procfs) */
605 if (p2->p_textvp)
606 vref(p2->p_textvp);
607
608 /*
609 * Set up linkage for kernel based threading.
610 */
611 if ((flags & RFTHREAD) != 0) {
612 mtx_lock(&ppeers_lock);
613 p2->p_peers = p1->p_peers;
614 p1->p_peers = p2;
615 p2->p_leader = p1->p_leader;
616 mtx_unlock(&ppeers_lock);
617 PROC_LOCK(p1->p_leader);
618 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
619 PROC_UNLOCK(p1->p_leader);
620 /*
621 * The task leader is exiting, so process p1 is
622 * going to be killed shortly. Since p1 obviously
623 * isn't dead yet, we know that the leader is either
624 * sending SIGKILL's to all the processes in this
625 * task or is sleeping waiting for all the peers to
626 * exit. We let p1 complete the fork, but we need
627 * to go ahead and kill the new process p2 since
628 * the task leader may not get a chance to send
629 * SIGKILL to it. We leave it on the list so that
630 * the task leader will wait for this new process
631 * to commit suicide.
632 */
633 PROC_LOCK(p2);
634 psignal(p2, SIGKILL);
635 PROC_UNLOCK(p2);
636 } else
637 PROC_UNLOCK(p1->p_leader);
638 } else {
639 p2->p_peers = NULL;
640 p2->p_leader = p2;
641 }
642
643 sx_xlock(&proctree_lock);
644 PGRP_LOCK(p1->p_pgrp);
645 PROC_LOCK(p2);
646 PROC_LOCK(p1);
647
648 /*
649 * Preserve some more flags in subprocess. P_PROFIL has already
650 * been preserved.
651 */
652 p2->p_flag |= p1->p_flag & P_SUGID;
653 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
654 SESS_LOCK(p1->p_session);
655 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
656 p2->p_flag |= P_CONTROLT;
657 SESS_UNLOCK(p1->p_session);
658 if (flags & RFPPWAIT)
659 p2->p_flag |= P_PPWAIT;
660
661 p2->p_pgrp = p1->p_pgrp;
662 LIST_INSERT_AFTER(p1, p2, p_pglist);
663 PGRP_UNLOCK(p1->p_pgrp);
664 LIST_INIT(&p2->p_children);
665
666 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
667
668 /*
669 * If PF_FORK is set, the child process inherits the
670 * procfs ioctl flags from its parent.
671 */
672 if (p1->p_pfsflags & PF_FORK) {
673 p2->p_stops = p1->p_stops;
674 p2->p_pfsflags = p1->p_pfsflags;
675 }
676
677 /*
678 * This begins the section where we must prevent the parent
679 * from being swapped.
680 */
681 _PHOLD(p1);
682 PROC_UNLOCK(p1);
683
684 /*
685 * Attach the new process to its parent.
686 *
687 * If RFNOWAIT is set, the newly created process becomes a child
688 * of init. This effectively disassociates the child from the
689 * parent.
690 */
691 if (flags & RFNOWAIT)
692 pptr = initproc;
693 else
694 pptr = p1;
695 p2->p_pptr = pptr;
696 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
697 sx_xunlock(&proctree_lock);
698
699 /* Inform accounting that we have forked. */
700 p2->p_acflag = AFORK;
701 PROC_UNLOCK(p2);
702
703 #ifdef KTRACE
704 ktrprocfork(p1, p2);
705 #endif
706
707 /*
708 * Finish creating the child process. It will return via a different
709 * execution path later. (ie: directly into user mode)
710 */
711 vm_forkproc(td, p2, td2, vm2, flags);
712
713 if (flags == (RFFDG | RFPROC)) {
714 PCPU_INC(cnt.v_forks);
715 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
716 p2->p_vmspace->vm_ssize);
717 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
718 PCPU_INC(cnt.v_vforks);
719 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
720 p2->p_vmspace->vm_ssize);
721 } else if (p1 == &proc0) {
722 PCPU_INC(cnt.v_kthreads);
723 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
724 p2->p_vmspace->vm_ssize);
725 } else {
726 PCPU_INC(cnt.v_rforks);
727 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
728 p2->p_vmspace->vm_ssize);
729 }
730
731 /*
732 * Both processes are set up, now check if any loadable modules want
733 * to adjust anything.
734 * What if they have an error? XXX
735 */
736 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
737
738 /*
739 * Set the child start time and mark the process as being complete.
740 */
741 PROC_LOCK(p2);
742 PROC_LOCK(p1);
743 microuptime(&p2->p_stats->p_start);
744 PROC_SLOCK(p2);
745 p2->p_state = PRS_NORMAL;
746 PROC_SUNLOCK(p2);
747
748 #ifdef KDTRACE_HOOKS
749 /*
750 * Tell the DTrace fasttrap provider about the new process
751 * if it has registered an interest. We have to do this only after
752 * p_state is PRS_NORMAL since the fasttrap module will use pfind()
753 * later on.
754 */
755 if (dtrace_fasttrap_fork)
756 dtrace_fasttrap_fork(p1, p2);
757 #endif
758 if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED |
759 P_FOLLOWFORK)) {
760 /*
761 * Arrange for debugger to receive the fork event.
762 *
763 * We can report PL_FLAG_FORKED regardless of
764 * P_FOLLOWFORK settings, but it does not make a sense
765 * for runaway child.
766 */
767 td->td_dbgflags |= TDB_FORK;
768 td->td_dbg_forked = p2->p_pid;
769 td2->td_dbgflags |= TDB_STOPATFORK;
770 _PHOLD(p2);
771 p2_held = 1;
772 }
773 PROC_UNLOCK(p2);
774 if ((flags & RFSTOPPED) == 0) {
775 /*
776 * If RFSTOPPED not requested, make child runnable and
777 * add to run queue.
778 */
779 thread_lock(td2);
780 TD_SET_CAN_RUN(td2);
781 sched_add(td2, SRQ_BORING);
782 thread_unlock(td2);
783 }
784
785 /*
786 * Now can be swapped.
787 */
788 _PRELE(p1);
789 PROC_UNLOCK(p1);
790
791 /*
792 * Tell any interested parties about the new process.
793 */
794 knote_fork(&p1->p_klist, p2->p_pid);
795 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
796
797 /*
798 * Wait until debugger is attached to child.
799 */
800 PROC_LOCK(p2);
801 while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
802 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
803 if (p2_held)
804 _PRELE(p2);
805
806 /*
807 * Preserve synchronization semantics of vfork. If waiting for
808 * child to exec or exit, set P_PPWAIT on child, and sleep on our
809 * proc (in case of exit).
810 */
811 while (p2->p_flag & P_PPWAIT)
812 cv_wait(&p2->p_pwait, &p2->p_mtx);
813 PROC_UNLOCK(p2);
814
815 /*
816 * Return child proc pointer to parent.
817 */
818 *procp = p2;
819 return (0);
820 fail:
821 sx_sunlock(&proctree_lock);
822 if (ppsratecheck(&lastfail, &curfail, 1))
823 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
824 td->td_ucred->cr_ruid);
825 sx_xunlock(&allproc_lock);
826 #ifdef MAC
827 mac_proc_destroy(newproc);
828 #endif
829 fail1:
830 if (vm2 != NULL)
831 vmspace_free(vm2);
832 uma_zfree(proc_zone, newproc);
833 pause("fork", hz / 2);
834 return (error);
835 }
836
837 /*
838 * Handle the return of a child process from fork1(). This function
839 * is called from the MD fork_trampoline() entry point.
840 */
841 void
842 fork_exit(callout, arg, frame)
843 void (*callout)(void *, struct trapframe *);
844 void *arg;
845 struct trapframe *frame;
846 {
847 struct proc *p;
848 struct thread *td;
849 struct thread *dtd;
850
851 td = curthread;
852 p = td->td_proc;
853 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
854
855 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
856 td, td->td_sched, p->p_pid, td->td_name);
857
858 sched_fork_exit(td);
859 /*
860 * Processes normally resume in mi_switch() after being
861 * cpu_switch()'ed to, but when children start up they arrive here
862 * instead, so we must do much the same things as mi_switch() would.
863 */
864 if ((dtd = PCPU_GET(deadthread))) {
865 PCPU_SET(deadthread, NULL);
866 thread_stash(dtd);
867 }
868 thread_unlock(td);
869
870 /*
871 * cpu_set_fork_handler intercepts this function call to
872 * have this call a non-return function to stay in kernel mode.
873 * initproc has its own fork handler, but it does return.
874 */
875 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
876 callout(arg, frame);
877
878 /*
879 * Check if a kernel thread misbehaved and returned from its main
880 * function.
881 */
882 if (p->p_flag & P_KTHREAD) {
883 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
884 td->td_name, p->p_pid);
885 kproc_exit(0);
886 }
887 mtx_assert(&Giant, MA_NOTOWNED);
888
889 if (p->p_sysent->sv_schedtail != NULL)
890 (p->p_sysent->sv_schedtail)(td);
891 }
892
893 /*
894 * Simplified back end of syscall(), used when returning from fork()
895 * directly into user mode. Giant is not held on entry, and must not
896 * be held on return. This function is passed in to fork_exit() as the
897 * first parameter and is called when returning to a new userland process.
898 */
899 void
900 fork_return(td, frame)
901 struct thread *td;
902 struct trapframe *frame;
903 {
904 struct proc *p, *dbg;
905
906 if (td->td_dbgflags & TDB_STOPATFORK) {
907 p = td->td_proc;
908 sx_xlock(&proctree_lock);
909 PROC_LOCK(p);
910 if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) ==
911 (P_TRACED | P_FOLLOWFORK)) {
912 /*
913 * If debugger still wants auto-attach for the
914 * parent's children, do it now.
915 */
916 dbg = p->p_pptr->p_pptr;
917 p->p_flag |= P_TRACED;
918 p->p_oppid = p->p_pptr->p_pid;
919 proc_reparent(p, dbg);
920 sx_xunlock(&proctree_lock);
921 ptracestop(td, SIGSTOP);
922 } else {
923 /*
924 * ... otherwise clear the request.
925 */
926 sx_xunlock(&proctree_lock);
927 td->td_dbgflags &= ~TDB_STOPATFORK;
928 cv_broadcast(&p->p_dbgwait);
929 }
930 PROC_UNLOCK(p);
931 }
932
933 userret(td, frame);
934
935 #ifdef KTRACE
936 if (KTRPOINT(td, KTR_SYSRET))
937 ktrsysret(SYS_fork, 0, 0);
938 #endif
939 mtx_assert(&Giant, MA_NOTOWNED);
940 }
Cache object: b8f1c8e96bc6d9ea2b0d6bba4e6bc2fa
|