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 1.77 2008/05/18 20:02:02 nth 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>
67 #include <sys/vmmeter.h>
68 #include <sys/thread2.h>
69 #include <sys/signal2.h>
70 #include <sys/spinlock2.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 static struct lwp
*lwp_fork(struct lwp
*, struct proc
*, int flags
);
88 int forksleep
; /* Place for fork1() to sleep on. */
91 * Red-Black tree support for LWPs
95 rb_lwp_compare(struct lwp
*lp1
, struct lwp
*lp2
)
97 if (lp1
->lwp_tid
< lp2
->lwp_tid
)
99 if (lp1
->lwp_tid
> lp2
->lwp_tid
)
104 RB_GENERATE2(lwp_rb_tree
, lwp
, u
.lwp_rbnode
, rb_lwp_compare
, lwpid_t
, lwp_tid
);
109 sys_fork(struct fork_args
*uap
)
111 struct lwp
*lp
= curthread
->td_lwp
;
115 error
= fork1(lp
, RFFDG
| RFPROC
| RFPGLOCK
, &p2
);
117 start_forked_proc(lp
, p2
);
118 uap
->sysmsg_fds
[0] = p2
->p_pid
;
119 uap
->sysmsg_fds
[1] = 0;
126 sys_vfork(struct vfork_args
*uap
)
128 struct lwp
*lp
= curthread
->td_lwp
;
132 error
= fork1(lp
, RFFDG
| RFPROC
| RFPPWAIT
| RFMEM
| RFPGLOCK
, &p2
);
134 start_forked_proc(lp
, p2
);
135 uap
->sysmsg_fds
[0] = p2
->p_pid
;
136 uap
->sysmsg_fds
[1] = 0;
142 * Handle rforks. An rfork may (1) operate on the current process without
143 * creating a new, (2) create a new process that shared the current process's
144 * vmspace, signals, and/or descriptors, or (3) create a new process that does
145 * not share these things (normal fork).
147 * Note that we only call start_forked_proc() if a new process is actually
150 * rfork { int flags }
153 sys_rfork(struct rfork_args
*uap
)
155 struct lwp
*lp
= curthread
->td_lwp
;
159 if ((uap
->flags
& RFKERNELONLY
) != 0)
162 error
= fork1(lp
, uap
->flags
| RFPGLOCK
, &p2
);
165 start_forked_proc(lp
, p2
);
166 uap
->sysmsg_fds
[0] = p2
? p2
->p_pid
: 0;
167 uap
->sysmsg_fds
[1] = 0;
173 sys_lwp_create(struct lwp_create_args
*uap
)
175 struct proc
*p
= curproc
;
177 struct lwp_params params
;
180 error
= copyin(uap
->params
, ¶ms
, sizeof(params
));
184 plimit_lwp_fork(p
); /* force exclusive access */
185 lp
= lwp_fork(curthread
->td_lwp
, p
, RFPROC
);
186 error
= cpu_prepare_lwp(lp
, ¶ms
);
187 if (params
.tid1
!= NULL
&&
188 (error
= copyout(&lp
->lwp_tid
, params
.tid1
, sizeof(lp
->lwp_tid
))))
190 if (params
.tid2
!= NULL
&&
191 (error
= copyout(&lp
->lwp_tid
, params
.tid2
, sizeof(lp
->lwp_tid
))))
195 * Now schedule the new lwp.
197 p
->p_usched
->resetpriority(lp
);
199 lp
->lwp_stat
= LSRUN
;
200 p
->p_usched
->setrunqueue(lp
);
206 lwp_rb_tree_RB_REMOVE(&p
->p_lwp_tree
, lp
);
208 /* lwp_dispose expects an exited lwp, and a held proc */
209 lp
->lwp_flag
|= LWP_WEXIT
;
210 lp
->lwp_thread
->td_flags
|= TDF_EXITING
;
217 int nprocs
= 1; /* process 0 */
220 fork1(struct lwp
*lp1
, int flags
, struct proc
**procp
)
222 struct proc
*p1
= lp1
->lwp_proc
;
223 struct proc
*p2
, *pptr
;
227 static int curfail
= 0;
228 static struct timeval lastfail
;
230 struct filedesc_to_leader
*fdtol
;
232 if ((flags
& (RFFDG
|RFCFDG
)) == (RFFDG
|RFCFDG
))
236 * Here we don't create a new process, but we divorce
237 * certain parts of a process from itself.
239 if ((flags
& RFPROC
) == 0) {
241 * This kind of stunt does not work anymore if
242 * there are native threads (lwps) running
244 if (p1
->p_nthreads
!= 1)
247 vm_fork(p1
, 0, flags
);
250 * Close all file descriptors.
252 if (flags
& RFCFDG
) {
253 struct filedesc
*fdtmp
;
259 * Unshare file descriptors (from parent.)
262 if (p1
->p_fd
->fd_refcnt
> 1) {
263 struct filedesc
*newfd
;
273 * Interlock against process group signal delivery. If signals
274 * are pending after the interlock is obtained we have to restart
275 * the system call to process the signals. If we don't the child
276 * can miss a pgsignal (such as ^C) sent during the fork.
278 * We can't use CURSIG() here because it will process any STOPs
279 * and cause the process group lock to be held indefinitely. If
280 * a STOP occurs, the fork will be restarted after the CONT.
284 if ((flags
& RFPGLOCK
) && (pgrp
= p1
->p_pgrp
) != NULL
) {
285 lockmgr(&pgrp
->pg_lock
, LK_SHARED
);
286 if (CURSIG_NOBLOCK(lp1
)) {
293 * Although process entries are dynamically created, we still keep
294 * a global limit on the maximum number we will create. Don't allow
295 * a nonprivileged user to use the last ten processes; don't let root
296 * exceed the limit. The variable nprocs is the current number of
297 * processes, maxproc is the limit.
299 uid
= p1
->p_ucred
->cr_ruid
;
300 if ((nprocs
>= maxproc
- 10 && uid
!= 0) || nprocs
>= maxproc
) {
301 if (ppsratecheck(&lastfail
, &curfail
, 1))
302 kprintf("maxproc limit exceeded by uid %d, please "
303 "see tuning(7) and login.conf(5).\n", uid
);
304 tsleep(&forksleep
, 0, "fork", hz
/ 2);
309 * Increment the nprocs resource before blocking can occur. There
310 * are hard-limits as to the number of processes that can run.
315 * Increment the count of procs running with this uid. Don't allow
316 * a nonprivileged user to exceed their current limit.
318 ok
= chgproccnt(p1
->p_ucred
->cr_ruidinfo
, 1,
319 (uid
!= 0) ? p1
->p_rlimit
[RLIMIT_NPROC
].rlim_cur
: 0);
322 * Back out the process count
325 if (ppsratecheck(&lastfail
, &curfail
, 1))
326 kprintf("maxproc limit exceeded by uid %d, please "
327 "see tuning(7) and login.conf(5).\n", uid
);
328 tsleep(&forksleep
, 0, "fork", hz
/ 2);
333 /* Allocate new proc. */
334 p2
= kmalloc(sizeof(struct proc
), M_PROC
, M_WAITOK
|M_ZERO
);
337 * Setup linkage for kernel based threading XXX lwp
339 if (flags
& RFTHREAD
) {
340 p2
->p_peers
= p1
->p_peers
;
342 p2
->p_leader
= p1
->p_leader
;
347 RB_INIT(&p2
->p_lwp_tree
);
348 spin_init(&p2
->p_spin
);
349 p2
->p_lasttid
= -1; /* first tid will be 0 */
352 * Setting the state to SIDL protects the partially initialized
353 * process once it starts getting hooked into the rest of the system.
356 proc_add_allproc(p2
);
359 * Make a proc table entry for the new process.
360 * The whole structure was zeroed above, so copy the section that is
361 * copied directly from the parent.
363 bcopy(&p1
->p_startcopy
, &p2
->p_startcopy
,
364 (unsigned) ((caddr_t
)&p2
->p_endcopy
- (caddr_t
)&p2
->p_startcopy
));
367 * Duplicate sub-structures as needed.
368 * Increase reference counts on shared objects.
370 if (p1
->p_flag
& P_PROFIL
)
372 p2
->p_ucred
= crhold(p1
->p_ucred
);
374 kprintf("Debug: p_lock race averted\n");
377 if (jailed(p2
->p_ucred
))
378 p2
->p_flag
|= P_JAILED
;
381 p2
->p_args
->ar_ref
++;
383 p2
->p_usched
= p1
->p_usched
;
385 if (flags
& RFSIGSHARE
) {
386 p2
->p_sigacts
= p1
->p_sigacts
;
387 p2
->p_sigacts
->ps_refcnt
++;
389 p2
->p_sigacts
= (struct sigacts
*)kmalloc(sizeof(*p2
->p_sigacts
),
390 M_SUBPROC
, M_WAITOK
);
391 bcopy(p1
->p_sigacts
, p2
->p_sigacts
, sizeof(*p2
->p_sigacts
));
392 p2
->p_sigacts
->ps_refcnt
= 1;
394 if (flags
& RFLINUXTHPN
)
395 p2
->p_sigparent
= SIGUSR1
;
397 p2
->p_sigparent
= SIGCHLD
;
399 /* bump references to the text vnode (for procfs) */
400 p2
->p_textvp
= p1
->p_textvp
;
405 * Handle file descriptors
407 if (flags
& RFCFDG
) {
408 p2
->p_fd
= fdinit(p1
);
410 } else if (flags
& RFFDG
) {
411 p2
->p_fd
= fdcopy(p1
);
414 p2
->p_fd
= fdshare(p1
);
415 if (p1
->p_fdtol
== NULL
)
417 filedesc_to_leader_alloc(NULL
,
419 if ((flags
& RFTHREAD
) != 0) {
421 * Shared file descriptor table and
422 * shared process leaders.
425 fdtol
->fdl_refcount
++;
428 * Shared file descriptor table, and
429 * different process leaders
431 fdtol
= filedesc_to_leader_alloc(p1
->p_fdtol
, p2
);
435 p2
->p_limit
= plimit_fork(p1
);
438 * Preserve some more flags in subprocess. P_PROFIL has already
441 p2
->p_flag
|= p1
->p_flag
& P_SUGID
;
442 if (p1
->p_session
->s_ttyvp
!= NULL
&& p1
->p_flag
& P_CONTROLT
)
443 p2
->p_flag
|= P_CONTROLT
;
444 if (flags
& RFPPWAIT
)
445 p2
->p_flag
|= P_PPWAIT
;
448 * Inherit the virtual kernel structure (allows a virtual kernel
449 * to fork to simulate multiple cpus).
452 vkernel_inherit(p1
, p2
);
455 * Once we are on a pglist we may receive signals. XXX we might
456 * race a ^C being sent to the process group by not receiving it
457 * at all prior to this line.
459 LIST_INSERT_AFTER(p1
, p2
, p_pglist
);
462 * Attach the new process to its parent.
464 * If RFNOWAIT is set, the newly created process becomes a child
465 * of init. This effectively disassociates the child from the
468 if (flags
& RFNOWAIT
)
473 LIST_INSERT_HEAD(&pptr
->p_children
, p2
, p_sibling
);
474 LIST_INIT(&p2
->p_children
);
475 varsymset_init(&p2
->p_varsymset
, &p1
->p_varsymset
);
476 callout_init(&p2
->p_ithandle
);
480 * Copy traceflag and tracefile if enabled. If not inherited,
481 * these were zeroed above but we still could have a trace race
482 * so make sure p2's p_tracenode is NULL.
484 if ((p1
->p_traceflag
& KTRFAC_INHERIT
) && p2
->p_tracenode
== NULL
) {
485 p2
->p_traceflag
= p1
->p_traceflag
;
486 p2
->p_tracenode
= ktrinherit(p1
->p_tracenode
);
491 * This begins the section where we must prevent the parent
492 * from being swapped.
494 * Gets PRELE'd in the caller in start_forked_proc().
498 vm_fork(p1
, p2
, flags
);
501 * Create the first lwp associated with the new proc.
502 * It will return via a different execution path later, directly
503 * into userland, after it was put on the runq by
504 * start_forked_proc().
506 lwp_fork(lp1
, p2
, flags
);
508 if (flags
== (RFFDG
| RFPROC
| RFPGLOCK
)) {
509 mycpu
->gd_cnt
.v_forks
++;
510 mycpu
->gd_cnt
.v_forkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
511 } else if (flags
== (RFFDG
| RFPROC
| RFPPWAIT
| RFMEM
| RFPGLOCK
)) {
512 mycpu
->gd_cnt
.v_vforks
++;
513 mycpu
->gd_cnt
.v_vforkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
514 } else if (p1
== &proc0
) {
515 mycpu
->gd_cnt
.v_kthreads
++;
516 mycpu
->gd_cnt
.v_kthreadpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
518 mycpu
->gd_cnt
.v_rforks
++;
519 mycpu
->gd_cnt
.v_rforkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
523 * Both processes are set up, now check if any loadable modules want
524 * to adjust anything.
525 * What if they have an error? XXX
527 TAILQ_FOREACH(ep
, &fork_list
, next
) {
528 (*ep
->function
)(p1
, p2
, flags
);
532 * Set the start time. Note that the process is not runnable. The
533 * caller is responsible for making it runnable.
535 microtime(&p2
->p_start
);
536 p2
->p_acflag
= AFORK
;
539 * tell any interested parties about the new process
541 KNOTE(&p1
->p_klist
, NOTE_FORK
| p2
->p_pid
);
544 * Return child proc pointer to parent.
549 lockmgr(&pgrp
->pg_lock
, LK_RELEASE
);
554 lwp_fork(struct lwp
*origlp
, struct proc
*destproc
, int flags
)
559 lp
= kmalloc(sizeof(struct lwp
), M_LWP
, M_WAITOK
|M_ZERO
);
561 lp
->lwp_proc
= destproc
;
562 lp
->lwp_vmspace
= destproc
->p_vmspace
;
563 lp
->lwp_stat
= LSRUN
;
564 bcopy(&origlp
->lwp_startcopy
, &lp
->lwp_startcopy
,
565 (unsigned) ((caddr_t
)&lp
->lwp_endcopy
-
566 (caddr_t
)&lp
->lwp_startcopy
));
567 lp
->lwp_flag
|= origlp
->lwp_flag
& LWP_ALTSTACK
;
569 * Set cpbase to the last timeout that occured (not the upcoming
572 * A critical section is required since a timer IPI can update
573 * scheduler specific data.
576 lp
->lwp_cpbase
= mycpu
->gd_schedclock
.time
-
577 mycpu
->gd_schedclock
.periodic
;
578 destproc
->p_usched
->heuristic_forking(origlp
, lp
);
580 lp
->lwp_cpumask
&= usched_mastermask
;
583 * Assign a TID to the lp. Loop until the insert succeeds (returns
586 lp
->lwp_tid
= destproc
->p_lasttid
;
588 if (++lp
->lwp_tid
< 0)
590 } while (lwp_rb_tree_RB_INSERT(&destproc
->p_lwp_tree
, lp
) != NULL
);
591 destproc
->p_lasttid
= lp
->lwp_tid
;
592 destproc
->p_nthreads
++;
594 td
= lwkt_alloc_thread(NULL
, LWKT_THREAD_STACK
, -1, 0);
596 td
->td_proc
= destproc
;
598 td
->td_switch
= cpu_heavy_switch
;
600 KKASSERT(td
->td_mpcount
== 1);
602 lwkt_setpri(td
, TDPRI_KERN_USER
);
603 lwkt_set_comm(td
, "%s", destproc
->p_comm
);
606 * cpu_fork will copy and update the pcb, set up the kernel stack,
607 * and make the child ready to run.
609 cpu_fork(origlp
, lp
, flags
);
610 caps_fork(origlp
->lwp_thread
, lp
->lwp_thread
);
616 * The next two functionms are general routines to handle adding/deleting
617 * items on the fork callout list.
620 * Take the arguments given and put them onto the fork callout list,
621 * However first make sure that it's not already there.
622 * Returns 0 on success or a standard error number.
625 at_fork(forklist_fn function
)
630 /* let the programmer know if he's been stupid */
631 if (rm_at_fork(function
)) {
632 kprintf("WARNING: fork callout entry (%p) already present\n",
636 ep
= kmalloc(sizeof(*ep
), M_ATFORK
, M_WAITOK
|M_ZERO
);
637 ep
->function
= function
;
638 TAILQ_INSERT_TAIL(&fork_list
, ep
, next
);
643 * Scan the exit callout list for the given item and remove it..
644 * Returns the number of items removed (0 or 1)
647 rm_at_fork(forklist_fn function
)
651 TAILQ_FOREACH(ep
, &fork_list
, next
) {
652 if (ep
->function
== function
) {
653 TAILQ_REMOVE(&fork_list
, ep
, next
);
662 * Add a forked process to the run queue after any remaining setup, such
663 * as setting the fork handler, has been completed.
666 start_forked_proc(struct lwp
*lp1
, struct proc
*p2
)
668 struct lwp
*lp2
= ONLY_LWP_IN_PROC(p2
);
671 * Move from SIDL to RUN queue, and activate the process's thread.
672 * Activation of the thread effectively makes the process "a"
673 * current process, so we do not setrunqueue().
675 * YYY setrunqueue works here but we should clean up the trampoline
676 * code so we just schedule the LWKT thread and let the trampoline
677 * deal with the userland scheduler on return to userland.
679 KASSERT(p2
->p_stat
== SIDL
,
680 ("cannot start forked process, bad status: %p", p2
));
681 p2
->p_usched
->resetpriority(lp2
);
683 p2
->p_stat
= SACTIVE
;
684 lp2
->lwp_stat
= LSRUN
;
685 p2
->p_usched
->setrunqueue(lp2
);
689 * Now can be swapped.
691 PRELE(lp1
->lwp_proc
);
694 * Preserve synchronization semantics of vfork. If waiting for
695 * child to exec or exit, set P_PPWAIT on child, and sleep on our
696 * proc (in case of exit).
698 while (p2
->p_flag
& P_PPWAIT
)
699 tsleep(lp1
->lwp_proc
, 0, "ppwait", 0);