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 1.32 2005/01/31 22:29:59 joerg 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>
71 static MALLOC_DEFINE(M_ATFORK
, "atfork", "atfork callback");
74 * These are the stuctures used to create a callout list for things to do
75 * when forking a process
79 TAILQ_ENTRY(forklist
) next
;
82 TAILQ_HEAD(forklist_head
, forklist
);
83 static struct forklist_head fork_list
= TAILQ_HEAD_INITIALIZER(fork_list
);
85 int forksleep
; /* Place for fork1() to sleep on. */
89 fork(struct fork_args
*uap
)
91 struct proc
*p
= curproc
;
95 error
= fork1(p
, RFFDG
| RFPROC
, &p2
);
97 start_forked_proc(p
, p2
);
98 uap
->sysmsg_fds
[0] = p2
->p_pid
;
99 uap
->sysmsg_fds
[1] = 0;
106 vfork(struct vfork_args
*uap
)
108 struct proc
*p
= curproc
;
112 error
= fork1(p
, RFFDG
| RFPROC
| RFPPWAIT
| RFMEM
, &p2
);
114 start_forked_proc(p
, p2
);
115 uap
->sysmsg_fds
[0] = p2
->p_pid
;
116 uap
->sysmsg_fds
[1] = 0;
122 * Handle rforks. An rfork may (1) operate on the current process without
123 * creating a new, (2) create a new process that shared the current process's
124 * vmspace, signals, and/or descriptors, or (3) create a new process that does
125 * not share these things (normal fork).
127 * Note that we only call start_forked_proc() if a new process is actually
130 * rfork { int flags }
133 rfork(struct rfork_args
*uap
)
135 struct proc
*p
= curproc
;
139 if ((uap
->flags
& RFKERNELONLY
) != 0)
142 error
= fork1(p
, uap
->flags
, &p2
);
145 start_forked_proc(p
, p2
);
146 uap
->sysmsg_fds
[0] = p2
? p2
->p_pid
: 0;
147 uap
->sysmsg_fds
[1] = 0;
153 int nprocs
= 1; /* process 0 */
154 static int nextpid
= 0;
157 * Random component to nextpid generation. We mix in a random factor to make
158 * it a little harder to predict. We sanity check the modulus value to avoid
159 * doing it in critical paths. Don't let it be too small or we pointlessly
160 * waste randomness entropy, and don't let it be impossibly large. Using a
161 * modulus that is too big causes a LOT more process table scans and slows
162 * down fork processing as the pidchecked caching is defeated.
164 static int randompid
= 0;
167 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS
)
172 error
= sysctl_handle_int(oidp
, &pid
, 0, req
);
173 if (error
|| !req
->newptr
)
175 if (pid
< 0 || pid
> PID_MAX
- 100) /* out of range */
177 else if (pid
< 2) /* NOP */
179 else if (pid
< 100) /* Make it reasonable */
185 SYSCTL_PROC(_kern
, OID_AUTO
, randompid
, CTLTYPE_INT
|CTLFLAG_RW
,
186 0, 0, sysctl_kern_randompid
, "I", "Random PID modulus");
189 fork1(struct proc
*p1
, int flags
, struct proc
**procp
)
191 struct proc
*p2
, *pptr
;
193 struct proc
*newproc
;
195 static int curfail
= 0, pidchecked
= 0;
196 static struct timeval lastfail
;
198 struct filedesc_to_leader
*fdtol
;
200 if ((flags
& (RFFDG
|RFCFDG
)) == (RFFDG
|RFCFDG
))
204 * Here we don't create a new process, but we divorce
205 * certain parts of a process from itself.
207 if ((flags
& RFPROC
) == 0) {
209 vm_fork(p1
, 0, flags
);
212 * Close all file descriptors.
214 if (flags
& RFCFDG
) {
215 struct filedesc
*fdtmp
;
222 * Unshare file descriptors (from parent.)
225 if (p1
->p_fd
->fd_refcnt
> 1) {
226 struct filedesc
*newfd
;
237 * Although process entries are dynamically created, we still keep
238 * a global limit on the maximum number we will create. Don't allow
239 * a nonprivileged user to use the last ten processes; don't let root
240 * exceed the limit. The variable nprocs is the current number of
241 * processes, maxproc is the limit.
243 uid
= p1
->p_ucred
->cr_ruid
;
244 if ((nprocs
>= maxproc
- 10 && uid
!= 0) || nprocs
>= maxproc
) {
245 if (ppsratecheck(&lastfail
, &curfail
, 1))
246 printf("maxproc limit exceeded by uid %d, please "
247 "see tuning(7) and login.conf(5).\n", uid
);
248 tsleep(&forksleep
, 0, "fork", hz
/ 2);
252 * Increment the nprocs resource before blocking can occur. There
253 * are hard-limits as to the number of processes that can run.
258 * Increment the count of procs running with this uid. Don't allow
259 * a nonprivileged user to exceed their current limit.
261 ok
= chgproccnt(p1
->p_ucred
->cr_ruidinfo
, 1,
262 (uid
!= 0) ? p1
->p_rlimit
[RLIMIT_NPROC
].rlim_cur
: 0);
265 * Back out the process count
268 if (ppsratecheck(&lastfail
, &curfail
, 1))
269 printf("maxproc limit exceeded by uid %d, please "
270 "see tuning(7) and login.conf(5).\n", uid
);
271 tsleep(&forksleep
, 0, "fork", hz
/ 2);
275 /* Allocate new proc. */
276 newproc
= zalloc(proc_zone
);
279 * Setup linkage for kernel based threading
281 if ((flags
& RFTHREAD
) != 0) {
282 newproc
->p_peers
= p1
->p_peers
;
283 p1
->p_peers
= newproc
;
284 newproc
->p_leader
= p1
->p_leader
;
286 newproc
->p_peers
= 0;
287 newproc
->p_leader
= newproc
;
290 newproc
->p_wakeup
= 0;
291 newproc
->p_vmspace
= NULL
;
292 TAILQ_INIT(&newproc
->p_sysmsgq
);
295 * Find an unused process ID. We remember a range of unused IDs
296 * ready to use (from nextpid+1 through pidchecked-1).
300 nextpid
+= arc4random() % randompid
;
303 * If the process ID prototype has wrapped around,
304 * restart somewhat above 0, as the low-numbered procs
305 * tend to include daemons that don't exit.
307 if (nextpid
>= PID_MAX
) {
308 nextpid
= nextpid
% PID_MAX
;
313 if (nextpid
>= pidchecked
) {
316 pidchecked
= PID_MAX
;
318 * Scan the active and zombie procs to check whether this pid
319 * is in use. Remember the lowest pid that's greater
320 * than nextpid, so we can avoid checking for a while.
322 p2
= LIST_FIRST(&allproc
);
324 for (; p2
!= 0; p2
= LIST_NEXT(p2
, p_list
)) {
325 while (p2
->p_pid
== nextpid
||
326 p2
->p_pgrp
->pg_id
== nextpid
||
327 p2
->p_session
->s_sid
== nextpid
) {
329 if (nextpid
>= pidchecked
)
332 if (p2
->p_pid
> nextpid
&& pidchecked
> p2
->p_pid
)
333 pidchecked
= p2
->p_pid
;
334 if (p2
->p_pgrp
->pg_id
> nextpid
&&
335 pidchecked
> p2
->p_pgrp
->pg_id
)
336 pidchecked
= p2
->p_pgrp
->pg_id
;
337 if (p2
->p_session
->s_sid
> nextpid
&&
338 pidchecked
> p2
->p_session
->s_sid
)
339 pidchecked
= p2
->p_session
->s_sid
;
343 p2
= LIST_FIRST(&zombproc
);
349 p2
->p_stat
= SIDL
; /* protect against others */
351 LIST_INSERT_HEAD(&allproc
, p2
, p_list
);
352 LIST_INSERT_HEAD(PIDHASH(p2
->p_pid
), p2
, p_hash
);
355 * Make a proc table entry for the new process.
356 * Start by zeroing the section of proc that is zero-initialized,
357 * then copy the section that is copied directly from the parent.
359 bzero(&p2
->p_startzero
,
360 (unsigned) ((caddr_t
)&p2
->p_endzero
- (caddr_t
)&p2
->p_startzero
));
361 bcopy(&p1
->p_startcopy
, &p2
->p_startcopy
,
362 (unsigned) ((caddr_t
)&p2
->p_endcopy
- (caddr_t
)&p2
->p_startcopy
));
364 p2
->p_aioinfo
= NULL
;
367 * Duplicate sub-structures as needed.
368 * Increase reference counts on shared objects.
369 * The p_stats and p_sigacts substructs are set in vm_fork.
371 p2
->p_flag
= P_INMEM
;
372 if (p1
->p_flag
& P_PROFIL
)
374 p2
->p_ucred
= crhold(p1
->p_ucred
);
376 if (jailed(p2
->p_ucred
))
377 p2
->p_flag
|= P_JAILED
;
380 p2
->p_args
->ar_ref
++;
382 if (flags
& RFSIGSHARE
) {
383 p2
->p_procsig
= p1
->p_procsig
;
384 p2
->p_procsig
->ps_refcnt
++;
385 if (p1
->p_sigacts
== &p1
->p_addr
->u_sigacts
) {
386 struct sigacts
*newsigacts
;
389 /* Create the shared sigacts structure */
390 MALLOC(newsigacts
, struct sigacts
*,
391 sizeof(struct sigacts
), M_SUBPROC
, M_WAITOK
);
394 * Set p_sigacts to the new shared structure.
395 * Note that this is updating p1->p_sigacts at the
396 * same time, since p_sigacts is just a pointer to
397 * the shared p_procsig->ps_sigacts.
399 p2
->p_sigacts
= newsigacts
;
400 bcopy(&p1
->p_addr
->u_sigacts
, p2
->p_sigacts
,
401 sizeof(*p2
->p_sigacts
));
402 *p2
->p_sigacts
= p1
->p_addr
->u_sigacts
;
406 MALLOC(p2
->p_procsig
, struct procsig
*, sizeof(struct procsig
),
407 M_SUBPROC
, M_WAITOK
);
408 bcopy(p1
->p_procsig
, p2
->p_procsig
, sizeof(*p2
->p_procsig
));
409 p2
->p_procsig
->ps_refcnt
= 1;
410 p2
->p_sigacts
= NULL
; /* finished in vm_fork() */
412 if (flags
& RFLINUXTHPN
)
413 p2
->p_sigparent
= SIGUSR1
;
415 p2
->p_sigparent
= SIGCHLD
;
417 /* bump references to the text vnode (for procfs) */
418 p2
->p_textvp
= p1
->p_textvp
;
422 if (flags
& RFCFDG
) {
423 p2
->p_fd
= fdinit(p1
);
425 } else if (flags
& RFFDG
) {
426 p2
->p_fd
= fdcopy(p1
);
429 p2
->p_fd
= fdshare(p1
);
430 if (p1
->p_fdtol
== NULL
)
432 filedesc_to_leader_alloc(NULL
,
434 if ((flags
& RFTHREAD
) != 0) {
436 * Shared file descriptor table and
437 * shared process leaders.
440 fdtol
->fdl_refcount
++;
443 * Shared file descriptor table, and
444 * different process leaders
446 fdtol
= filedesc_to_leader_alloc(p1
->p_fdtol
, p2
);
452 * If p_limit is still copy-on-write, bump refcnt,
453 * otherwise get a copy that won't be modified.
454 * (If PL_SHAREMOD is clear, the structure is shared
457 if (p1
->p_limit
->p_lflags
& PL_SHAREMOD
) {
458 p2
->p_limit
= limcopy(p1
->p_limit
);
460 p2
->p_limit
= p1
->p_limit
;
461 p2
->p_limit
->p_refcnt
++;
465 * Preserve some more flags in subprocess. P_PROFIL has already
468 p2
->p_flag
|= p1
->p_flag
& (P_SUGID
| P_ALTSTACK
);
469 if (p1
->p_session
->s_ttyvp
!= NULL
&& p1
->p_flag
& P_CONTROLT
)
470 p2
->p_flag
|= P_CONTROLT
;
471 if (flags
& RFPPWAIT
)
472 p2
->p_flag
|= P_PPWAIT
;
475 * Once we are on a pglist we may receive signals. XXX we might
476 * race a ^C being sent to the process group by not receiving it
477 * at all prior to this line.
479 LIST_INSERT_AFTER(p1
, p2
, p_pglist
);
482 * Attach the new process to its parent.
484 * If RFNOWAIT is set, the newly created process becomes a child
485 * of init. This effectively disassociates the child from the
488 if (flags
& RFNOWAIT
)
493 LIST_INSERT_HEAD(&pptr
->p_children
, p2
, p_sibling
);
494 LIST_INIT(&p2
->p_children
);
495 varsymset_init(&p2
->p_varsymset
, &p1
->p_varsymset
);
496 callout_init(&p2
->p_ithandle
);
500 * Copy traceflag and tracefile if enabled. If not inherited,
501 * these were zeroed above but we still could have a trace race
502 * so make sure p2's p_tracep is NULL.
504 if ((p1
->p_traceflag
& KTRFAC_INHERIT
) && p2
->p_tracep
== NULL
) {
505 p2
->p_traceflag
= p1
->p_traceflag
;
506 if ((p2
->p_tracep
= p1
->p_tracep
) != NULL
)
512 * Give the child process an estcpu skewed towards the batch side
513 * of the parent. This prevents batch programs from glitching
514 * interactive programs when they are first started. If the child
515 * is not a batch program it's priority will be corrected by the
518 * The interactivity model always starts at 0 (par value).
520 p2
->p_estcpu_fork
= p2
->p_estcpu
=
521 ESTCPULIM(p1
->p_estcpu
+ ESTCPURAMP
);
522 p2
->p_interactive
= 0;
525 * This begins the section where we must prevent the parent
526 * from being swapped.
531 * Finish creating the child process. It will return via a different
532 * execution path later. (ie: directly into user mode)
534 vm_fork(p1
, p2
, flags
);
535 caps_fork(p1
, p2
, flags
);
537 if (flags
== (RFFDG
| RFPROC
)) {
538 mycpu
->gd_cnt
.v_forks
++;
539 mycpu
->gd_cnt
.v_forkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
540 } else if (flags
== (RFFDG
| RFPROC
| RFPPWAIT
| RFMEM
)) {
541 mycpu
->gd_cnt
.v_vforks
++;
542 mycpu
->gd_cnt
.v_vforkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
543 } else if (p1
== &proc0
) {
544 mycpu
->gd_cnt
.v_kthreads
++;
545 mycpu
->gd_cnt
.v_kthreadpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
547 mycpu
->gd_cnt
.v_rforks
++;
548 mycpu
->gd_cnt
.v_rforkpages
+= p2
->p_vmspace
->vm_dsize
+ p2
->p_vmspace
->vm_ssize
;
552 * Both processes are set up, now check if any loadable modules want
553 * to adjust anything.
554 * What if they have an error? XXX
556 TAILQ_FOREACH(ep
, &fork_list
, next
) {
557 (*ep
->function
)(p1
, p2
, flags
);
561 * Make child runnable and add to run queue.
563 microtime(&p2
->p_thread
->td_start
);
564 p2
->p_acflag
= AFORK
;
567 * tell any interested parties about the new process
569 KNOTE(&p1
->p_klist
, NOTE_FORK
| p2
->p_pid
);
572 * Return child proc pointer to parent.
579 * The next two functionms are general routines to handle adding/deleting
580 * items on the fork callout list.
583 * Take the arguments given and put them onto the fork callout list,
584 * However first make sure that it's not already there.
585 * Returns 0 on success or a standard error number.
588 at_fork(forklist_fn function
)
593 /* let the programmer know if he's been stupid */
594 if (rm_at_fork(function
)) {
595 printf("WARNING: fork callout entry (%p) already present\n",
599 ep
= malloc(sizeof(*ep
), M_ATFORK
, M_WAITOK
|M_ZERO
);
600 ep
->function
= function
;
601 TAILQ_INSERT_TAIL(&fork_list
, ep
, next
);
606 * Scan the exit callout list for the given item and remove it..
607 * Returns the number of items removed (0 or 1)
610 rm_at_fork(forklist_fn function
)
614 TAILQ_FOREACH(ep
, &fork_list
, next
) {
615 if (ep
->function
== function
) {
616 TAILQ_REMOVE(&fork_list
, ep
, next
);
625 * Add a forked process to the run queue after any remaining setup, such
626 * as setting the fork handler, has been completed.
629 start_forked_proc(struct proc
*p1
, struct proc
*p2
)
632 * Move from SIDL to RUN queue, and activate the process's thread.
633 * Activation of the thread effectively makes the process "a"
634 * current process, so we do not setrunqueue().
636 * YYY setrunqueue works here but we should clean up the trampoline
637 * code so we just schedule the LWKT thread and let the trampoline
638 * deal with the userland scheduler on return to userland.
640 KASSERT(p2
&& p2
->p_stat
== SIDL
,
641 ("cannot start forked process, bad status: %p", p2
));
649 * Now can be swapped.
654 * Preserve synchronization semantics of vfork. If waiting for
655 * child to exec or exit, set P_PPWAIT on child, and sleep on our
656 * proc (in case of exit).
658 while (p2
->p_flag
& P_PPWAIT
)
659 tsleep(p1
, 0, "ppwait", 0);