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 1.51 2006/06/05 00:32:37 davidxu 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 */
157 fork1(struct lwp *lp1, int flags, struct proc **procp)
159 struct proc *p1 = lp1->lwp_proc;
160 struct proc *p2, *pptr;
164 static int curfail = 0;
165 static struct timeval lastfail;
167 struct filedesc_to_leader *fdtol;
169 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
173 * Here we don't create a new process, but we divorce
174 * certain parts of a process from itself.
176 if ((flags & RFPROC) == 0) {
178 vm_fork(p1, 0, flags);
181 * Close all file descriptors.
183 if (flags & RFCFDG) {
184 struct filedesc *fdtmp;
191 * Unshare file descriptors (from parent.)
194 if (p1->p_fd->fd_refcnt > 1) {
195 struct filedesc *newfd;
206 * Although process entries are dynamically created, we still keep
207 * a global limit on the maximum number we will create. Don't allow
208 * a nonprivileged user to use the last ten processes; don't let root
209 * exceed the limit. The variable nprocs is the current number of
210 * processes, maxproc is the limit.
212 uid = p1->p_ucred->cr_ruid;
213 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
214 if (ppsratecheck(&lastfail, &curfail, 1))
215 printf("maxproc limit exceeded by uid %d, please "
216 "see tuning(7) and login.conf(5).\n", uid);
217 tsleep(&forksleep, 0, "fork", hz / 2);
221 * Increment the nprocs resource before blocking can occur. There
222 * are hard-limits as to the number of processes that can run.
227 * Increment the count of procs running with this uid. Don't allow
228 * a nonprivileged user to exceed their current limit.
230 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
231 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
234 * Back out the process count
237 if (ppsratecheck(&lastfail, &curfail, 1))
238 printf("maxproc limit exceeded by uid %d, please "
239 "see tuning(7) and login.conf(5).\n", uid);
240 tsleep(&forksleep, 0, "fork", hz / 2);
244 /* Allocate new proc. */
245 p2 = zalloc(proc_zone);
248 * Setup linkage for kernel based threading XXX lwp
250 if (flags & RFTHREAD) {
251 p2->p_peers = p1->p_peers;
253 p2->p_leader = p1->p_leader;
260 p2->p_vmspace = NULL;
261 p2->p_numposixlocks = 0;
262 p2->p_emuldata = NULL;
263 TAILQ_INIT(&p2->p_lwp.lwp_sysmsgq);
264 LIST_INIT(&p2->p_lwps);
270 lp2->lwp_cpumask = 0xFFFFFFFF; /* XXX inherit ? */
271 LIST_INSERT_HEAD(&p2->p_lwps, lp2, lwp_list);
277 * Setting the state to SIDL protects the partially initialized
278 * process once it starts getting hooked into the rest of the system.
281 proc_add_allproc(p2);
284 * Make a proc table entry for the new process.
285 * Start by zeroing the section of proc that is zero-initialized,
286 * then copy the section that is copied directly from the parent.
288 bzero(&p2->p_startzero,
289 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
290 bzero(&lp2->lwp_startzero,
291 (unsigned) ((caddr_t)&lp2->lwp_endzero -
292 (caddr_t)&lp2->lwp_startzero));
293 bcopy(&p1->p_startcopy, &p2->p_startcopy,
294 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
295 bcopy(&p1->p_lwp.lwp_startcopy, &lp2->lwp_startcopy,
296 (unsigned) ((caddr_t)&lp2->lwp_endcopy -
297 (caddr_t)&lp2->lwp_startcopy));
299 p2->p_aioinfo = NULL;
302 * Duplicate sub-structures as needed.
303 * Increase reference counts on shared objects.
304 * The p_stats and p_sigacts substructs are set in vm_fork.
305 * p_lock is in the copy area and must be cleared.
309 if (p1->p_flag & P_PROFIL)
311 p2->p_ucred = crhold(p1->p_ucred);
313 if (jailed(p2->p_ucred))
314 p2->p_flag |= P_JAILED;
317 p2->p_args->ar_ref++;
319 if (flags & RFSIGSHARE) {
320 p2->p_procsig = p1->p_procsig;
321 p2->p_procsig->ps_refcnt++;
322 if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
323 struct sigacts *newsigacts;
325 /* Create the shared sigacts structure */
326 MALLOC(newsigacts, struct sigacts *,
327 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
330 * Set p_sigacts to the new shared structure.
331 * Note that this is updating p1->p_sigacts at the
332 * same time, since p_sigacts is just a pointer to
333 * the shared p_procsig->ps_sigacts.
335 p2->p_sigacts = newsigacts;
336 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
337 sizeof(*p2->p_sigacts));
338 *p2->p_sigacts = p1->p_addr->u_sigacts;
342 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
343 M_SUBPROC, M_WAITOK);
344 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
345 p2->p_procsig->ps_refcnt = 1;
346 p2->p_sigacts = NULL; /* finished in vm_fork() */
348 if (flags & RFLINUXTHPN)
349 p2->p_sigparent = SIGUSR1;
351 p2->p_sigparent = SIGCHLD;
353 /* bump references to the text vnode (for procfs) */
354 p2->p_textvp = p1->p_textvp;
358 if (flags & RFCFDG) {
359 p2->p_fd = fdinit(p1);
361 } else if (flags & RFFDG) {
362 p2->p_fd = fdcopy(p1);
365 p2->p_fd = fdshare(p1);
366 if (p1->p_fdtol == NULL)
368 filedesc_to_leader_alloc(NULL,
370 if ((flags & RFTHREAD) != 0) {
372 * Shared file descriptor table and
373 * shared process leaders.
376 fdtol->fdl_refcount++;
379 * Shared file descriptor table, and
380 * different process leaders
382 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
386 p2->p_limit = plimit_fork(p1->p_limit);
389 * Preserve some more flags in subprocess. P_PROFIL has already
392 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
393 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
394 p2->p_flag |= P_CONTROLT;
395 if (flags & RFPPWAIT)
396 p2->p_flag |= P_PPWAIT;
399 * Once we are on a pglist we may receive signals. XXX we might
400 * race a ^C being sent to the process group by not receiving it
401 * at all prior to this line.
403 LIST_INSERT_AFTER(p1, p2, p_pglist);
406 * Attach the new process to its parent.
408 * If RFNOWAIT is set, the newly created process becomes a child
409 * of init. This effectively disassociates the child from the
412 if (flags & RFNOWAIT)
417 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
418 LIST_INIT(&p2->p_children);
419 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
420 callout_init(&p2->p_ithandle);
424 * Copy traceflag and tracefile if enabled. If not inherited,
425 * these were zeroed above but we still could have a trace race
426 * so make sure p2's p_tracenode is NULL.
428 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
429 p2->p_traceflag = p1->p_traceflag;
430 p2->p_tracenode = ktrinherit(p1->p_tracenode);
435 * Inherit the scheduler and initialize scheduler-related fields.
436 * Set cpbase to the last timeout that occured (not the upcoming
439 * A critical section is required since a timer IPI can update
440 * scheduler specific data.
443 p2->p_usched = p1->p_usched;
444 lp2->lwp_cpbase = mycpu->gd_schedclock.time -
445 mycpu->gd_schedclock.periodic;
446 p2->p_usched->heuristic_forking(&p1->p_lwp, lp2);
450 * This begins the section where we must prevent the parent
451 * from being swapped.
456 * Finish creating the child process. It will return via a different
457 * execution path later. (ie: directly into user mode)
459 vm_fork(p1, p2, flags);
460 caps_fork(p1, p2, flags);
462 if (flags == (RFFDG | RFPROC)) {
463 mycpu->gd_cnt.v_forks++;
464 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
465 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
466 mycpu->gd_cnt.v_vforks++;
467 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
468 } else if (p1 == &proc0) {
469 mycpu->gd_cnt.v_kthreads++;
470 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
472 mycpu->gd_cnt.v_rforks++;
473 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
477 * Both processes are set up, now check if any loadable modules want
478 * to adjust anything.
479 * What if they have an error? XXX
481 TAILQ_FOREACH(ep, &fork_list, next) {
482 (*ep->function)(p1, p2, flags);
486 * Set the start time. Note that the process is not runnable. The
487 * caller is responsible for making it runnable.
489 microtime(&p2->p_start);
490 p2->p_acflag = AFORK;
493 * tell any interested parties about the new process
495 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
498 * Return child proc pointer to parent.
505 * The next two functionms are general routines to handle adding/deleting
506 * items on the fork callout list.
509 * Take the arguments given and put them onto the fork callout list,
510 * However first make sure that it's not already there.
511 * Returns 0 on success or a standard error number.
514 at_fork(forklist_fn function)
519 /* let the programmer know if he's been stupid */
520 if (rm_at_fork(function)) {
521 printf("WARNING: fork callout entry (%p) already present\n",
525 ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
526 ep->function = function;
527 TAILQ_INSERT_TAIL(&fork_list, ep, next);
532 * Scan the exit callout list for the given item and remove it..
533 * Returns the number of items removed (0 or 1)
536 rm_at_fork(forklist_fn function)
540 TAILQ_FOREACH(ep, &fork_list, next) {
541 if (ep->function == function) {
542 TAILQ_REMOVE(&fork_list, ep, next);
551 * Add a forked process to the run queue after any remaining setup, such
552 * as setting the fork handler, has been completed.
555 start_forked_proc(struct lwp *lp1, struct proc *p2)
559 KKASSERT(p2 != NULL && p2->p_nthreads == 1);
561 lp2 = LIST_FIRST(&p2->p_lwps);
564 * Move from SIDL to RUN queue, and activate the process's thread.
565 * Activation of the thread effectively makes the process "a"
566 * current process, so we do not setrunqueue().
568 * YYY setrunqueue works here but we should clean up the trampoline
569 * code so we just schedule the LWKT thread and let the trampoline
570 * deal with the userland scheduler on return to userland.
572 KASSERT(p2->p_stat == SIDL,
573 ("cannot start forked process, bad status: %p", p2));
574 p2->p_usched->resetpriority(lp2);
577 p2->p_usched->setrunqueue(lp2);
581 * Now can be swapped.
583 PRELE(lp1->lwp_proc);
586 * Preserve synchronization semantics of vfork. If waiting for
587 * child to exec or exit, set P_PPWAIT on child, and sleep on our
588 * proc (in case of exit).
590 while (p2->p_flag & P_PPWAIT)
591 tsleep(lp1->lwp_proc, 0, "ppwait", 0);