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 1.64 2007/02/25 23:17:12 corecode 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>
69 #include <sys/thread2.h>
70 #include <sys/signal2.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 sys_fork(struct fork_args
*uap
)
92 struct lwp
*lp
= curthread
->td_lwp
;
96 error
= fork1(lp
, RFFDG
| RFPROC
| RFPGLOCK
, &p2
);
98 start_forked_proc(lp
, p2
);
99 uap
->sysmsg_fds
[0] = p2
->p_pid
;
100 uap
->sysmsg_fds
[1] = 0;
107 sys_vfork(struct vfork_args
*uap
)
109 struct lwp
*lp
= curthread
->td_lwp
;
113 error
= fork1(lp
, RFFDG
| RFPROC
| RFPPWAIT
| RFMEM
| RFPGLOCK
, &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 sys_rfork(struct rfork_args
*uap
)
136 struct lwp
*lp
= curthread
->td_lwp
;
140 if ((uap
->flags
& RFKERNELONLY
) != 0)
143 error
= fork1(lp
, uap
->flags
| RFPGLOCK
, &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
;
165 static int curfail
= 0;
166 static struct timeval lastfail
;
168 struct filedesc_to_leader
*fdtol
;
170 if ((flags
& (RFFDG
|RFCFDG
)) == (RFFDG
|RFCFDG
))
174 * Here we don't create a new process, but we divorce
175 * certain parts of a process from itself.
177 if ((flags
& RFPROC
) == 0) {
179 vm_fork(lp1
, 0, flags
);
182 * Close all file descriptors.
184 if (flags
& RFCFDG
) {
185 struct filedesc
*fdtmp
;
192 * Unshare file descriptors (from parent.)
195 if (p1
->p_fd
->fd_refcnt
> 1) {
196 struct filedesc
*newfd
;
207 * Interlock against process group signal delivery. If signals
208 * are pending after the interlock is obtained we have to restart
209 * the system call to process the signals. If we don't the child
210 * can miss a pgsignal (such as ^C) sent during the fork.
212 * We can't use CURSIG() here because it will process any STOPs
213 * and cause the process group lock to be held indefinitely. If
214 * a STOP occurs, the fork will be restarted after the CONT.
218 if ((flags
& RFPGLOCK
) && (pgrp
= p1
->p_pgrp
) != NULL
) {
219 lockmgr(&pgrp
->pg_lock
, LK_SHARED
);
227 * Although process entries are dynamically created, we still keep
228 * a global limit on the maximum number we will create. Don't allow
229 * a nonprivileged user to use the last ten processes; don't let root
230 * exceed the limit. The variable nprocs is the current number of
231 * processes, maxproc is the limit.
233 uid
= p1
->p_ucred
->cr_ruid
;
234 if ((nprocs
>= maxproc
- 10 && uid
!= 0) || nprocs
>= maxproc
) {
235 if (ppsratecheck(&lastfail
, &curfail
, 1))
236 kprintf("maxproc limit exceeded by uid %d, please "
237 "see tuning(7) and login.conf(5).\n", uid
);
238 tsleep(&forksleep
, 0, "fork", hz
/ 2);
243 * Increment the nprocs resource before blocking can occur. There
244 * are hard-limits as to the number of processes that can run.
249 * Increment the count of procs running with this uid. Don't allow
250 * a nonprivileged user to exceed their current limit.
252 ok
= chgproccnt(p1
->p_ucred
->cr_ruidinfo
, 1,
253 (uid
!= 0) ? p1
->p_rlimit
[RLIMIT_NPROC
].rlim_cur
: 0);
256 * Back out the process count
259 if (ppsratecheck(&lastfail
, &curfail
, 1))
260 kprintf("maxproc limit exceeded by uid %d, please "
261 "see tuning(7) and login.conf(5).\n", uid
);
262 tsleep(&forksleep
, 0, "fork", hz
/ 2);
267 /* Allocate new proc. */
268 p2
= zalloc(proc_zone
);
269 lp2
= zalloc(lwp_zone
);
272 * Setup linkage for kernel based threading XXX lwp
274 if (flags
& RFTHREAD
) {
275 p2
->p_peers
= p1
->p_peers
;
277 p2
->p_leader
= p1
->p_leader
;
284 p2
->p_vmspace
= NULL
;
285 p2
->p_numposixlocks
= 0;
286 p2
->p_emuldata
= NULL
;
287 LIST_INIT(&p2
->p_lwps
);
292 LIST_INSERT_HEAD(&p2
->p_lwps
, lp2
, lwp_list
);
298 * Setting the state to SIDL protects the partially initialized
299 * process once it starts getting hooked into the rest of the system.
302 lp2
->lwp_stat
= LSRUN
; /* XXX use other state? start_forked_proc() handles this*/
303 proc_add_allproc(p2
);
306 * Make a proc table entry for the new process.
307 * Start by zeroing the section of proc that is zero-initialized,
308 * then copy the section that is copied directly from the parent.
310 bzero(&p2
->p_startzero
,
311 (unsigned) ((caddr_t
)&p2
->p_endzero
- (caddr_t
)&p2
->p_startzero
));
312 bzero(&lp2
->lwp_startzero
,
313 (unsigned) ((caddr_t
)&lp2
->lwp_endzero
-
314 (caddr_t
)&lp2
->lwp_startzero
));
315 bcopy(&p1
->p_startcopy
, &p2
->p_startcopy
,
316 (unsigned) ((caddr_t
)&p2
->p_endcopy
- (caddr_t
)&p2
->p_startcopy
));
317 bcopy(&lp1
->lwp_startcopy
, &lp2
->lwp_startcopy
,
318 (unsigned) ((caddr_t
)&lp2
->lwp_endcopy
-
319 (caddr_t
)&lp2
->lwp_startcopy
));
321 p2
->p_aioinfo
= NULL
;
324 * Duplicate sub-structures as needed.
325 * Increase reference counts on shared objects.
326 * p_lock is in the copy area and must be cleared.
331 if (p1
->p_flag
& P_PROFIL
)
333 p2
->p_ucred
= crhold(p1
->p_ucred
);
335 if (jailed(p2
->p_ucred
))
336 p2
->p_flag
|= P_JAILED
;
339 p2
->p_args
->ar_ref
++;
341 if (flags
& RFSIGSHARE
) {
342 p2
->p_sigacts
= p1
->p_sigacts
;
343 p2
->p_sigacts
->ps_refcnt
++;
345 p2
->p_sigacts
= (struct sigacts
*)kmalloc(sizeof(*p2
->p_sigacts
),
346 M_SUBPROC
, M_WAITOK
);
347 bcopy(p1
->p_sigacts
, p2
->p_sigacts
, sizeof(*p2
->p_sigacts
));
348 p2
->p_sigacts
->ps_refcnt
= 1;
350 if (flags
& RFLINUXTHPN
)
351 p2
->p_sigparent
= SIGUSR1
;
353 p2
->p_sigparent
= SIGCHLD
;
355 /* bump references to the text vnode (for procfs) */
356 p2
->p_textvp
= p1
->p_textvp
;
361 * Handle file descriptors
363 if (flags
& RFCFDG
) {
364 p2
->p_fd
= fdinit(p1
);
366 } else if (flags
& RFFDG
) {
367 p2
->p_fd
= fdcopy(p1
);
370 p2
->p_fd
= fdshare(p1
);
371 if (p1
->p_fdtol
== NULL
)
373 filedesc_to_leader_alloc(NULL
,
375 if ((flags
& RFTHREAD
) != 0) {
377 * Shared file descriptor table and
378 * shared process leaders.
381 fdtol
->fdl_refcount
++;
384 * Shared file descriptor table, and
385 * different process leaders
387 fdtol
= filedesc_to_leader_alloc(p1
->p_fdtol
, p2
);
391 p2
->p_limit
= plimit_fork(p1
->p_limit
);
394 * Preserve some more flags in subprocess. P_PROFIL has already
397 p2
->p_flag
|= p1
->p_flag
& P_SUGID
;
398 lp2
->lwp_flag
|= lp1
->lwp_flag
& LWP_ALTSTACK
;
399 if (p1
->p_session
->s_ttyvp
!= NULL
&& p1
->p_flag
& P_CONTROLT
)
400 p2
->p_flag
|= P_CONTROLT
;
401 if (flags
& RFPPWAIT
)
402 p2
->p_flag
|= P_PPWAIT
;
405 * Inherit the virtual kernel structure (allows a virtual kernel
406 * to fork to simulate multiple cpus).
408 p2
->p_vkernel
= NULL
;
410 vkernel_inherit(p1
, p2
);
413 * Once we are on a pglist we may receive signals. XXX we might
414 * race a ^C being sent to the process group by not receiving it
415 * at all prior to this line.
417 LIST_INSERT_AFTER(p1
, p2
, p_pglist
);
420 * Attach the new process to its parent.
422 * If RFNOWAIT is set, the newly created process becomes a child
423 * of init. This effectively disassociates the child from the
426 if (flags
& RFNOWAIT
)
431 LIST_INSERT_HEAD(&pptr
->p_children
, p2
, p_sibling
);
432 LIST_INIT(&p2
->p_children
);
433 varsymset_init(&p2
->p_varsymset
, &p1
->p_varsymset
);
434 callout_init(&p2
->p_ithandle
);
438 * Copy traceflag and tracefile if enabled. If not inherited,
439 * these were zeroed above but we still could have a trace race
440 * so make sure p2's p_tracenode is NULL.
442 if ((p1
->p_traceflag
& KTRFAC_INHERIT
) && p2
->p_tracenode
== NULL
) {
443 p2
->p_traceflag
= p1
->p_traceflag
;
444 p2
->p_tracenode
= ktrinherit(p1
->p_tracenode
);
449 * Inherit the scheduler and initialize scheduler-related fields.
450 * Set cpbase to the last timeout that occured (not the upcoming
453 * A critical section is required since a timer IPI can update
454 * scheduler specific data.
457 p2
->p_usched
= p1
->p_usched
;
458 lp2
->lwp_cpbase
= mycpu
->gd_schedclock
.time
-
459 mycpu
->gd_schedclock
.periodic
;
460 p2
->p_usched
->heuristic_forking(lp1
, lp2
);
464 * This begins the section where we must prevent the parent
465 * from being swapped.
470 * Finish creating the child process. It will return via a different
471 * execution path later. (ie: directly into user mode)
473 vm_fork(lp1
, p2
, flags
);
474 caps_fork(lp1
->lwp_thread
, lp2
->lwp_thread
, flags
);
476 if (flags
== (RFFDG
| RFPROC
)) {
477 mycpu
->gd_cnt
.v_forks
++;
478 mycpu
->gd_cnt
.v_forkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
479 } else if (flags
== (RFFDG
| RFPROC
| RFPPWAIT
| RFMEM
)) {
480 mycpu
->gd_cnt
.v_vforks
++;
481 mycpu
->gd_cnt
.v_vforkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
482 } else if (p1
== &proc0
) {
483 mycpu
->gd_cnt
.v_kthreads
++;
484 mycpu
->gd_cnt
.v_kthreadpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
486 mycpu
->gd_cnt
.v_rforks
++;
487 mycpu
->gd_cnt
.v_rforkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
491 * Both processes are set up, now check if any loadable modules want
492 * to adjust anything.
493 * What if they have an error? XXX
495 TAILQ_FOREACH(ep
, &fork_list
, next
) {
496 (*ep
->function
)(p1
, p2
, flags
);
500 * Set the start time. Note that the process is not runnable. The
501 * caller is responsible for making it runnable.
503 microtime(&p2
->p_start
);
504 p2
->p_acflag
= AFORK
;
507 * tell any interested parties about the new process
509 KNOTE(&p1
->p_klist
, NOTE_FORK
| p2
->p_pid
);
512 * Return child proc pointer to parent.
517 lockmgr(&pgrp
->pg_lock
, LK_RELEASE
);
522 * The next two functionms are general routines to handle adding/deleting
523 * items on the fork callout list.
526 * Take the arguments given and put them onto the fork callout list,
527 * However first make sure that it's not already there.
528 * Returns 0 on success or a standard error number.
531 at_fork(forklist_fn function
)
536 /* let the programmer know if he's been stupid */
537 if (rm_at_fork(function
)) {
538 kprintf("WARNING: fork callout entry (%p) already present\n",
542 ep
= kmalloc(sizeof(*ep
), M_ATFORK
, M_WAITOK
|M_ZERO
);
543 ep
->function
= function
;
544 TAILQ_INSERT_TAIL(&fork_list
, ep
, next
);
549 * Scan the exit callout list for the given item and remove it..
550 * Returns the number of items removed (0 or 1)
553 rm_at_fork(forklist_fn function
)
557 TAILQ_FOREACH(ep
, &fork_list
, next
) {
558 if (ep
->function
== function
) {
559 TAILQ_REMOVE(&fork_list
, ep
, next
);
568 * Add a forked process to the run queue after any remaining setup, such
569 * as setting the fork handler, has been completed.
572 start_forked_proc(struct lwp
*lp1
, struct proc
*p2
)
574 struct lwp
*lp2
= ONLY_LWP_IN_PROC(p2
);
577 * Move from SIDL to RUN queue, and activate the process's thread.
578 * Activation of the thread effectively makes the process "a"
579 * current process, so we do not setrunqueue().
581 * YYY setrunqueue works here but we should clean up the trampoline
582 * code so we just schedule the LWKT thread and let the trampoline
583 * deal with the userland scheduler on return to userland.
585 KASSERT(p2
->p_stat
== SIDL
,
586 ("cannot start forked process, bad status: %p", p2
));
587 p2
->p_usched
->resetpriority(lp2
);
589 p2
->p_stat
= SACTIVE
;
590 lp2
->lwp_stat
= LSRUN
;
591 p2
->p_usched
->setrunqueue(lp2
);
595 * Now can be swapped.
597 PRELE(lp1
->lwp_proc
);
600 * Preserve synchronization semantics of vfork. If waiting for
601 * child to exec or exit, set P_PPWAIT on child, and sleep on our
602 * proc (in case of exit).
604 while (p2
->p_flag
& P_PPWAIT
)
605 tsleep(lp1
->lwp_proc
, 0, "ppwait", 0);