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.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
39 * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $
40 * $DragonFly: src/sys/kern/kern_fork.c,v 1.48 2006/05/23 20:35:10 dillon Exp $
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/vnode.h>
56 #include <sys/ktrace.h>
57 #include <sys/unistd.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_zone.h>
68 #include <sys/vmmeter.h>
70 #include <sys/thread2.h>
72 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
75 * These are the stuctures used to create a callout list for things to do
76 * when forking a process
80 TAILQ_ENTRY(forklist) next;
83 TAILQ_HEAD(forklist_head, forklist);
84 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
86 int forksleep; /* Place for fork1() to sleep on. */
90 fork(struct fork_args *uap)
92 struct lwp *lp = curthread->td_lwp;
96 error = fork1(lp, RFFDG | RFPROC, &p2);
98 start_forked_proc(lp, p2);
99 uap->sysmsg_fds[0] = p2->p_pid;
100 uap->sysmsg_fds[1] = 0;
107 vfork(struct vfork_args *uap)
109 struct lwp *lp = curthread->td_lwp;
113 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
115 start_forked_proc(lp, p2);
116 uap->sysmsg_fds[0] = p2->p_pid;
117 uap->sysmsg_fds[1] = 0;
123 * Handle rforks. An rfork may (1) operate on the current process without
124 * creating a new, (2) create a new process that shared the current process's
125 * vmspace, signals, and/or descriptors, or (3) create a new process that does
126 * not share these things (normal fork).
128 * Note that we only call start_forked_proc() if a new process is actually
131 * rfork { int flags }
134 rfork(struct rfork_args *uap)
136 struct lwp *lp = curthread->td_lwp;
140 if ((uap->flags & RFKERNELONLY) != 0)
143 error = fork1(lp, uap->flags, &p2);
146 start_forked_proc(lp, p2);
147 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
148 uap->sysmsg_fds[1] = 0;
154 int nprocs = 1; /* process 0 */
155 static int nextpid = 0;
158 * Random component to nextpid 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.
165 static int randompid = 0;
168 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
173 error = sysctl_handle_int(oidp, &pid, 0, req);
174 if (error || !req->newptr)
176 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
178 else if (pid < 2) /* NOP */
180 else if (pid < 100) /* Make it reasonable */
186 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
187 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
190 fork1(struct lwp *lp1, int flags, struct proc **procp)
192 struct proc *p1 = lp1->lwp_proc;
193 struct proc *p2, *pptr;
196 struct proc *newproc;
198 static int curfail = 0, pidchecked = 0;
199 static struct timeval lastfail;
201 struct filedesc_to_leader *fdtol;
203 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
207 * Here we don't create a new process, but we divorce
208 * certain parts of a process from itself.
210 if ((flags & RFPROC) == 0) {
212 vm_fork(p1, 0, flags);
215 * Close all file descriptors.
217 if (flags & RFCFDG) {
218 struct filedesc *fdtmp;
225 * Unshare file descriptors (from parent.)
228 if (p1->p_fd->fd_refcnt > 1) {
229 struct filedesc *newfd;
240 * Although process entries are dynamically created, we still keep
241 * a global limit on the maximum number we will create. Don't allow
242 * a nonprivileged user to use the last ten processes; don't let root
243 * exceed the limit. The variable nprocs is the current number of
244 * processes, maxproc is the limit.
246 uid = p1->p_ucred->cr_ruid;
247 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
248 if (ppsratecheck(&lastfail, &curfail, 1))
249 printf("maxproc limit exceeded by uid %d, please "
250 "see tuning(7) and login.conf(5).\n", uid);
251 tsleep(&forksleep, 0, "fork", hz / 2);
255 * Increment the nprocs resource before blocking can occur. There
256 * are hard-limits as to the number of processes that can run.
261 * Increment the count of procs running with this uid. Don't allow
262 * a nonprivileged user to exceed their current limit.
264 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
265 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
268 * Back out the process count
271 if (ppsratecheck(&lastfail, &curfail, 1))
272 printf("maxproc limit exceeded by uid %d, please "
273 "see tuning(7) and login.conf(5).\n", uid);
274 tsleep(&forksleep, 0, "fork", hz / 2);
278 /* Allocate new proc. */
279 newproc = zalloc(proc_zone);
282 * Setup linkage for kernel based threading XXX lwp
284 if (flags & RFTHREAD) {
285 newproc->p_peers = p1->p_peers;
286 p1->p_peers = newproc;
287 newproc->p_leader = p1->p_leader;
289 newproc->p_peers = NULL;
290 newproc->p_leader = newproc;
293 newproc->p_wakeup = 0;
294 newproc->p_vmspace = NULL;
295 newproc->p_numposixlocks = 0;
296 newproc->p_emuldata = NULL;
297 TAILQ_INIT(&newproc->p_lwp.lwp_sysmsgq);
298 LIST_INIT(&newproc->p_lwps);
301 lp2 = &newproc->p_lwp;
302 lp2->lwp_proc = newproc;
304 LIST_INSERT_HEAD(&newproc->p_lwps, lp2, lwp_list);
305 newproc->p_nthreads = 1;
306 newproc->p_nstopped = 0;
307 newproc->p_lasttid = 0;
310 * Find an unused process ID. We remember a range of unused IDs
311 * ready to use (from nextpid+1 through pidchecked-1).
315 nextpid += arc4random() % randompid;
318 * If the process ID prototype has wrapped around,
319 * restart somewhat above 0, as the low-numbered procs
320 * tend to include daemons that don't exit.
322 if (nextpid >= PID_MAX) {
323 nextpid = nextpid % PID_MAX;
328 if (nextpid >= pidchecked) {
331 pidchecked = PID_MAX;
333 * Scan the active and zombie procs to check whether this pid
334 * is in use. Remember the lowest pid that's greater
335 * than nextpid, so we can avoid checking for a while.
337 p2 = LIST_FIRST(&allproc);
339 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
340 while (p2->p_pid == nextpid ||
341 p2->p_pgrp->pg_id == nextpid ||
342 p2->p_session->s_sid == nextpid) {
344 if (nextpid >= pidchecked)
347 if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
348 pidchecked = p2->p_pid;
349 if (p2->p_pgrp->pg_id > nextpid &&
350 pidchecked > p2->p_pgrp->pg_id)
351 pidchecked = p2->p_pgrp->pg_id;
352 if (p2->p_session->s_sid > nextpid &&
353 pidchecked > p2->p_session->s_sid)
354 pidchecked = p2->p_session->s_sid;
358 p2 = LIST_FIRST(&zombproc);
364 p2->p_stat = SIDL; /* protect against others */
366 LIST_INSERT_HEAD(&allproc, p2, p_list);
367 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
370 * Make a proc table entry for the new process.
371 * Start by zeroing the section of proc that is zero-initialized,
372 * then copy the section that is copied directly from the parent.
374 bzero(&p2->p_startzero,
375 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
376 bzero(&lp2->lwp_startzero,
377 (unsigned) ((caddr_t)&lp2->lwp_endzero -
378 (caddr_t)&lp2->lwp_startzero));
379 bcopy(&p1->p_startcopy, &p2->p_startcopy,
380 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
381 bcopy(&p1->p_lwp.lwp_startcopy, &lp2->lwp_startcopy,
382 (unsigned) ((caddr_t)&lp2->lwp_endcopy -
383 (caddr_t)&lp2->lwp_startcopy));
385 p2->p_aioinfo = NULL;
388 * Duplicate sub-structures as needed.
389 * Increase reference counts on shared objects.
390 * The p_stats and p_sigacts substructs are set in vm_fork.
391 * p_lock is in the copy area and must be cleared.
395 if (p1->p_flag & P_PROFIL)
397 p2->p_ucred = crhold(p1->p_ucred);
399 if (jailed(p2->p_ucred))
400 p2->p_flag |= P_JAILED;
403 p2->p_args->ar_ref++;
405 if (flags & RFSIGSHARE) {
406 p2->p_procsig = p1->p_procsig;
407 p2->p_procsig->ps_refcnt++;
408 if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
409 struct sigacts *newsigacts;
411 /* Create the shared sigacts structure */
412 MALLOC(newsigacts, struct sigacts *,
413 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
416 * Set p_sigacts to the new shared structure.
417 * Note that this is updating p1->p_sigacts at the
418 * same time, since p_sigacts is just a pointer to
419 * the shared p_procsig->ps_sigacts.
421 p2->p_sigacts = newsigacts;
422 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
423 sizeof(*p2->p_sigacts));
424 *p2->p_sigacts = p1->p_addr->u_sigacts;
428 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
429 M_SUBPROC, M_WAITOK);
430 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
431 p2->p_procsig->ps_refcnt = 1;
432 p2->p_sigacts = NULL; /* finished in vm_fork() */
434 if (flags & RFLINUXTHPN)
435 p2->p_sigparent = SIGUSR1;
437 p2->p_sigparent = SIGCHLD;
439 /* bump references to the text vnode (for procfs) */
440 p2->p_textvp = p1->p_textvp;
444 if (flags & RFCFDG) {
445 p2->p_fd = fdinit(p1);
447 } else if (flags & RFFDG) {
448 p2->p_fd = fdcopy(p1);
451 p2->p_fd = fdshare(p1);
452 if (p1->p_fdtol == NULL)
454 filedesc_to_leader_alloc(NULL,
456 if ((flags & RFTHREAD) != 0) {
458 * Shared file descriptor table and
459 * shared process leaders.
462 fdtol->fdl_refcount++;
465 * Shared file descriptor table, and
466 * different process leaders
468 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
472 p2->p_limit = plimit_fork(p1->p_limit);
475 * Preserve some more flags in subprocess. P_PROFIL has already
478 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
479 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
480 p2->p_flag |= P_CONTROLT;
481 if (flags & RFPPWAIT)
482 p2->p_flag |= P_PPWAIT;
485 * Once we are on a pglist we may receive signals. XXX we might
486 * race a ^C being sent to the process group by not receiving it
487 * at all prior to this line.
489 LIST_INSERT_AFTER(p1, p2, p_pglist);
492 * Attach the new process to its parent.
494 * If RFNOWAIT is set, the newly created process becomes a child
495 * of init. This effectively disassociates the child from the
498 if (flags & RFNOWAIT)
503 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
504 LIST_INIT(&p2->p_children);
505 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
506 callout_init(&p2->p_ithandle);
510 * Copy traceflag and tracefile if enabled. If not inherited,
511 * these were zeroed above but we still could have a trace race
512 * so make sure p2's p_tracenode is NULL.
514 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
515 p2->p_traceflag = p1->p_traceflag;
516 p2->p_tracenode = ktrinherit(p1->p_tracenode);
521 * Inherit the scheduler and initialize scheduler-related fields.
522 * Set cpbase to the last timeout that occured (not the upcoming
525 p2->p_usched = p1->p_usched;
526 lp2->lwp_cpbase = mycpu->gd_schedclock.time -
527 mycpu->gd_schedclock.periodic;
528 p2->p_usched->heuristic_forking(&p1->p_lwp, lp2);
531 * This begins the section where we must prevent the parent
532 * from being swapped.
537 * Finish creating the child process. It will return via a different
538 * execution path later. (ie: directly into user mode)
540 vm_fork(p1, p2, flags);
541 caps_fork(p1, p2, flags);
543 if (flags == (RFFDG | RFPROC)) {
544 mycpu->gd_cnt.v_forks++;
545 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
546 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
547 mycpu->gd_cnt.v_vforks++;
548 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
549 } else if (p1 == &proc0) {
550 mycpu->gd_cnt.v_kthreads++;
551 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
553 mycpu->gd_cnt.v_rforks++;
554 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
558 * Both processes are set up, now check if any loadable modules want
559 * to adjust anything.
560 * What if they have an error? XXX
562 TAILQ_FOREACH(ep, &fork_list, next) {
563 (*ep->function)(p1, p2, flags);
567 * Set the start time. Note that the process is not runnable. The
568 * caller is responsible for making it runnable.
570 microtime(&p2->p_start);
571 p2->p_acflag = AFORK;
574 * tell any interested parties about the new process
576 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
579 * Return child proc pointer to parent.
586 * The next two functionms are general routines to handle adding/deleting
587 * items on the fork callout list.
590 * Take the arguments given and put them onto the fork callout list,
591 * However first make sure that it's not already there.
592 * Returns 0 on success or a standard error number.
595 at_fork(forklist_fn function)
600 /* let the programmer know if he's been stupid */
601 if (rm_at_fork(function)) {
602 printf("WARNING: fork callout entry (%p) already present\n",
606 ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
607 ep->function = function;
608 TAILQ_INSERT_TAIL(&fork_list, ep, next);
613 * Scan the exit callout list for the given item and remove it..
614 * Returns the number of items removed (0 or 1)
617 rm_at_fork(forklist_fn function)
621 TAILQ_FOREACH(ep, &fork_list, next) {
622 if (ep->function == function) {
623 TAILQ_REMOVE(&fork_list, ep, next);
632 * Add a forked process to the run queue after any remaining setup, such
633 * as setting the fork handler, has been completed.
636 start_forked_proc(struct lwp *lp1, struct proc *p2)
640 KKASSERT(p2 != NULL && p2->p_nthreads == 1);
642 lp2 = LIST_FIRST(&p2->p_lwps);
645 * Move from SIDL to RUN queue, and activate the process's thread.
646 * Activation of the thread effectively makes the process "a"
647 * current process, so we do not setrunqueue().
649 * YYY setrunqueue works here but we should clean up the trampoline
650 * code so we just schedule the LWKT thread and let the trampoline
651 * deal with the userland scheduler on return to userland.
653 KASSERT(p2->p_stat == SIDL,
654 ("cannot start forked process, bad status: %p", p2));
655 p2->p_usched->resetpriority(lp2);
658 p2->p_usched->setrunqueue(lp2);
662 * Now can be swapped.
664 PRELE(lp1->lwp_proc);
667 * Preserve synchronization semantics of vfork. If waiting for
668 * child to exec or exit, set P_PPWAIT on child, and sleep on our
669 * proc (in case of exit).
671 while (p2->p_flag & P_PPWAIT)
672 tsleep(lp1->lwp_proc, 0, "ppwait", 0);