Optimize lwkt_rwlock.c a bit
[dragonfly.git] / sys / kern / kern_fork.c
CommitLineData
984263bc
MD
1/*
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.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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.
25 *
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
36 * SUCH DAMAGE.
37 *
38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
39 * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.13 2003/06/06 20:21:32 tegge Exp $
7d0bac62 40 * $DragonFly: src/sys/kern/kern_fork.c,v 1.6 2003/06/22 04:30:42 dillon Exp $
984263bc
MD
41 */
42
43#include "opt_ktrace.h"
44
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>
52#include <sys/proc.h>
53#include <sys/resourcevar.h>
54#include <sys/vnode.h>
55#include <sys/acct.h>
56#include <sys/ktrace.h>
57#include <sys/unistd.h>
58#include <sys/jail.h>
59
60#include <vm/vm.h>
61#include <sys/lock.h>
62#include <vm/pmap.h>
63#include <vm/vm_map.h>
64#include <vm/vm_extern.h>
65#include <vm/vm_zone.h>
66
67#include <sys/vmmeter.h>
68#include <sys/user.h>
69
70static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
71
72/*
73 * These are the stuctures used to create a callout list for things to do
74 * when forking a process
75 */
76struct forklist {
77 forklist_fn function;
78 TAILQ_ENTRY(forklist) next;
79};
80
81TAILQ_HEAD(forklist_head, forklist);
82static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
83
84#ifndef _SYS_SYSPROTO_H_
85struct fork_args {
86 int dummy;
87};
88#endif
89
90int forksleep; /* Place for fork1() to sleep on. */
91
92/* ARGSUSED */
93int
94fork(p, uap)
95 struct proc *p;
96 struct fork_args *uap;
97{
98 int error;
99 struct proc *p2;
100
101 error = fork1(p, RFFDG | RFPROC, &p2);
102 if (error == 0) {
7d0bac62 103 start_forked_proc(p, p2);
984263bc
MD
104 p->p_retval[0] = p2->p_pid;
105 p->p_retval[1] = 0;
106 }
107 return error;
108}
109
110/* ARGSUSED */
111int
112vfork(p, uap)
113 struct proc *p;
114 struct vfork_args *uap;
115{
116 int error;
117 struct proc *p2;
118
119 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
120 if (error == 0) {
7d0bac62 121 start_forked_proc(p, p2);
984263bc
MD
122 p->p_retval[0] = p2->p_pid;
123 p->p_retval[1] = 0;
124 }
125 return error;
126}
127
128int
129rfork(p, uap)
130 struct proc *p;
131 struct rfork_args *uap;
132{
133 int error;
134 struct proc *p2;
135
136 error = fork1(p, uap->flags, &p2);
137 if (error == 0) {
7d0bac62 138 start_forked_proc(p, p2);
984263bc
MD
139 p->p_retval[0] = p2 ? p2->p_pid : 0;
140 p->p_retval[1] = 0;
141 }
142 return error;
143}
144
145
146int nprocs = 1; /* process 0 */
147static int nextpid = 0;
148
149/*
150 * Random component to nextpid generation. We mix in a random factor to make
151 * it a little harder to predict. We sanity check the modulus value to avoid
152 * doing it in critical paths. Don't let it be too small or we pointlessly
153 * waste randomness entropy, and don't let it be impossibly large. Using a
154 * modulus that is too big causes a LOT more process table scans and slows
155 * down fork processing as the pidchecked caching is defeated.
156 */
157static int randompid = 0;
158
159static int
160sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
161{
162 int error, pid;
163
164 pid = randompid;
165 error = sysctl_handle_int(oidp, &pid, 0, req);
166 if (error || !req->newptr)
167 return (error);
168 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
169 pid = PID_MAX - 100;
170 else if (pid < 2) /* NOP */
171 pid = 0;
172 else if (pid < 100) /* Make it reasonable */
173 pid = 100;
174 randompid = pid;
175 return (error);
176}
177
178SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
179 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
180
181int
182fork1(p1, flags, procp)
183 struct proc *p1;
184 int flags;
185 struct proc **procp;
186{
187 struct proc *p2, *pptr;
188 uid_t uid;
189 struct proc *newproc;
190 int ok;
191 static int pidchecked = 0;
192 struct forklist *ep;
193 struct filedesc_to_leader *fdtol;
194
195 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
196 return (EINVAL);
197
198 /*
199 * Here we don't create a new process, but we divorce
200 * certain parts of a process from itself.
201 */
202 if ((flags & RFPROC) == 0) {
203
204 vm_fork(p1, 0, flags);
205
206 /*
207 * Close all file descriptors.
208 */
209 if (flags & RFCFDG) {
210 struct filedesc *fdtmp;
211 fdtmp = fdinit(p1);
212 fdfree(p1);
213 p1->p_fd = fdtmp;
214 }
215
216 /*
217 * Unshare file descriptors (from parent.)
218 */
219 if (flags & RFFDG) {
220 if (p1->p_fd->fd_refcnt > 1) {
221 struct filedesc *newfd;
222 newfd = fdcopy(p1);
223 fdfree(p1);
224 p1->p_fd = newfd;
225 }
226 }
227 *procp = NULL;
228 return (0);
229 }
230
231 /*
232 * Although process entries are dynamically created, we still keep
233 * a global limit on the maximum number we will create. Don't allow
234 * a nonprivileged user to use the last ten processes; don't let root
235 * exceed the limit. The variable nprocs is the current number of
236 * processes, maxproc is the limit.
237 */
238 uid = p1->p_cred->p_ruid;
239 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
240 tsleep(&forksleep, PUSER, "fork", hz / 2);
241 return (EAGAIN);
242 }
243 /*
244 * Increment the nprocs resource before blocking can occur. There
245 * are hard-limits as to the number of processes that can run.
246 */
247 nprocs++;
248
249 /*
250 * Increment the count of procs running with this uid. Don't allow
251 * a nonprivileged user to exceed their current limit.
252 */
253 ok = chgproccnt(p1->p_cred->p_uidinfo, 1,
254 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
255 if (!ok) {
256 /*
257 * Back out the process count
258 */
259 nprocs--;
260 tsleep(&forksleep, PUSER, "fork", hz / 2);
261 return (EAGAIN);
262 }
263
264 /* Allocate new proc. */
265 newproc = zalloc(proc_zone);
266
267 /*
268 * Setup linkage for kernel based threading
269 */
270 if((flags & RFTHREAD) != 0) {
271 newproc->p_peers = p1->p_peers;
272 p1->p_peers = newproc;
273 newproc->p_leader = p1->p_leader;
274 } else {
275 newproc->p_peers = 0;
276 newproc->p_leader = newproc;
277 }
278
279 newproc->p_wakeup = 0;
984263bc
MD
280 newproc->p_vmspace = NULL;
281
282 /*
283 * Find an unused process ID. We remember a range of unused IDs
284 * ready to use (from nextpid+1 through pidchecked-1).
285 */
286 nextpid++;
287 if (randompid)
288 nextpid += arc4random() % randompid;
289retry:
290 /*
291 * If the process ID prototype has wrapped around,
292 * restart somewhat above 0, as the low-numbered procs
293 * tend to include daemons that don't exit.
294 */
295 if (nextpid >= PID_MAX) {
296 nextpid = nextpid % PID_MAX;
297 if (nextpid < 100)
298 nextpid += 100;
299 pidchecked = 0;
300 }
301 if (nextpid >= pidchecked) {
302 int doingzomb = 0;
303
304 pidchecked = PID_MAX;
305 /*
306 * Scan the active and zombie procs to check whether this pid
307 * is in use. Remember the lowest pid that's greater
308 * than nextpid, so we can avoid checking for a while.
309 */
310 p2 = LIST_FIRST(&allproc);
311again:
312 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
313 while (p2->p_pid == nextpid ||
314 p2->p_pgrp->pg_id == nextpid ||
315 p2->p_session->s_sid == nextpid) {
316 nextpid++;
317 if (nextpid >= pidchecked)
318 goto retry;
319 }
320 if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
321 pidchecked = p2->p_pid;
322 if (p2->p_pgrp->pg_id > nextpid &&
323 pidchecked > p2->p_pgrp->pg_id)
324 pidchecked = p2->p_pgrp->pg_id;
325 if (p2->p_session->s_sid > nextpid &&
326 pidchecked > p2->p_session->s_sid)
327 pidchecked = p2->p_session->s_sid;
328 }
329 if (!doingzomb) {
330 doingzomb = 1;
331 p2 = LIST_FIRST(&zombproc);
332 goto again;
333 }
334 }
335
336 p2 = newproc;
337 p2->p_stat = SIDL; /* protect against others */
338 p2->p_pid = nextpid;
339 LIST_INSERT_HEAD(&allproc, p2, p_list);
340 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
341
342 /*
343 * Make a proc table entry for the new process.
344 * Start by zeroing the section of proc that is zero-initialized,
345 * then copy the section that is copied directly from the parent.
346 */
347 bzero(&p2->p_startzero,
348 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
349 bcopy(&p1->p_startcopy, &p2->p_startcopy,
350 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
351
352 p2->p_aioinfo = NULL;
353
354 /*
355 * Duplicate sub-structures as needed.
356 * Increase reference counts on shared objects.
357 * The p_stats and p_sigacts substructs are set in vm_fork.
358 */
359 p2->p_flag = P_INMEM;
360 if (p1->p_flag & P_PROFIL)
361 startprofclock(p2);
362 MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred),
363 M_SUBPROC, M_WAITOK);
364 bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred));
365 p2->p_cred->p_refcnt = 1;
366 crhold(p1->p_ucred);
367 uihold(p1->p_cred->p_uidinfo);
368
369 if (p2->p_prison) {
370 p2->p_prison->pr_ref++;
371 p2->p_flag |= P_JAILED;
372 }
373
374 if (p2->p_args)
375 p2->p_args->ar_ref++;
376
377 if (flags & RFSIGSHARE) {
378 p2->p_procsig = p1->p_procsig;
379 p2->p_procsig->ps_refcnt++;
380 if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
381 struct sigacts *newsigacts;
382 int s;
383
384 /* Create the shared sigacts structure */
385 MALLOC(newsigacts, struct sigacts *,
386 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
387 s = splhigh();
388 /*
389 * Set p_sigacts to the new shared structure.
390 * Note that this is updating p1->p_sigacts at the
391 * same time, since p_sigacts is just a pointer to
392 * the shared p_procsig->ps_sigacts.
393 */
394 p2->p_sigacts = newsigacts;
395 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
396 sizeof(*p2->p_sigacts));
397 *p2->p_sigacts = p1->p_addr->u_sigacts;
398 splx(s);
399 }
400 } else {
401 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
402 M_SUBPROC, M_WAITOK);
403 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
404 p2->p_procsig->ps_refcnt = 1;
405 p2->p_sigacts = NULL; /* finished in vm_fork() */
406 }
407 if (flags & RFLINUXTHPN)
408 p2->p_sigparent = SIGUSR1;
409 else
410 p2->p_sigparent = SIGCHLD;
411
412 /* bump references to the text vnode (for procfs) */
413 p2->p_textvp = p1->p_textvp;
414 if (p2->p_textvp)
415 VREF(p2->p_textvp);
416
417 if (flags & RFCFDG) {
418 p2->p_fd = fdinit(p1);
419 fdtol = NULL;
420 } else if (flags & RFFDG) {
421 p2->p_fd = fdcopy(p1);
422 fdtol = NULL;
423 } else {
424 p2->p_fd = fdshare(p1);
425 if (p1->p_fdtol == NULL)
426 p1->p_fdtol =
427 filedesc_to_leader_alloc(NULL,
428 p1->p_leader);
429 if ((flags & RFTHREAD) != 0) {
430 /*
431 * Shared file descriptor table and
432 * shared process leaders.
433 */
434 fdtol = p1->p_fdtol;
435 fdtol->fdl_refcount++;
436 } else {
437 /*
438 * Shared file descriptor table, and
439 * different process leaders
440 */
441 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
442 p2);
443 }
444 }
445 p2->p_fdtol = fdtol;
446
447 /*
448 * If p_limit is still copy-on-write, bump refcnt,
449 * otherwise get a copy that won't be modified.
450 * (If PL_SHAREMOD is clear, the structure is shared
451 * copy-on-write.)
452 */
453 if (p1->p_limit->p_lflags & PL_SHAREMOD)
454 p2->p_limit = limcopy(p1->p_limit);
455 else {
456 p2->p_limit = p1->p_limit;
457 p2->p_limit->p_refcnt++;
458 }
459
460 /*
461 * Preserve some more flags in subprocess. P_PROFIL has already
462 * been preserved.
463 */
464 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
465 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
466 p2->p_flag |= P_CONTROLT;
467 if (flags & RFPPWAIT)
468 p2->p_flag |= P_PPWAIT;
469
470 LIST_INSERT_AFTER(p1, p2, p_pglist);
471
472 /*
473 * Attach the new process to its parent.
474 *
475 * If RFNOWAIT is set, the newly created process becomes a child
476 * of init. This effectively disassociates the child from the
477 * parent.
478 */
479 if (flags & RFNOWAIT)
480 pptr = initproc;
481 else
482 pptr = p1;
483 p2->p_pptr = pptr;
484 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
485 LIST_INIT(&p2->p_children);
486
487#ifdef KTRACE
488 /*
489 * Copy traceflag and tracefile if enabled. If not inherited,
490 * these were zeroed above but we still could have a trace race
491 * so make sure p2's p_tracep is NULL.
492 */
493 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
494 p2->p_traceflag = p1->p_traceflag;
495 if ((p2->p_tracep = p1->p_tracep) != NULL)
496 VREF(p2->p_tracep);
497 }
498#endif
499
500 /*
501 * set priority of child to be that of parent
502 */
503 p2->p_estcpu = p1->p_estcpu;
504
505 /*
506 * This begins the section where we must prevent the parent
507 * from being swapped.
508 */
509 PHOLD(p1);
510
511 /*
512 * Finish creating the child process. It will return via a different
513 * execution path later. (ie: directly into user mode)
514 */
515 vm_fork(p1, p2, flags);
516
517 if (flags == (RFFDG | RFPROC)) {
518 cnt.v_forks++;
519 cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
520 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
521 cnt.v_vforks++;
522 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
523 } else if (p1 == &proc0) {
524 cnt.v_kthreads++;
525 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
526 } else {
527 cnt.v_rforks++;
528 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
529 }
530
531 /*
532 * Both processes are set up, now check if any loadable modules want
533 * to adjust anything.
534 * What if they have an error? XXX
535 */
536 TAILQ_FOREACH(ep, &fork_list, next) {
537 (*ep->function)(p1, p2, flags);
538 }
539
540 /*
541 * Make child runnable and add to run queue.
542 */
543 microtime(&(p2->p_stats->p_start));
544 p2->p_acflag = AFORK;
984263bc
MD
545
546 /*
547 * tell any interested parties about the new process
548 */
549 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
550
984263bc
MD
551 /*
552 * Return child proc pointer to parent.
553 */
554 *procp = p2;
555 return (0);
556}
557
558/*
559 * The next two functionms are general routines to handle adding/deleting
560 * items on the fork callout list.
561 *
562 * at_fork():
563 * Take the arguments given and put them onto the fork callout list,
564 * However first make sure that it's not already there.
565 * Returns 0 on success or a standard error number.
566 */
567
568int
569at_fork(function)
570 forklist_fn function;
571{
572 struct forklist *ep;
573
574#ifdef INVARIANTS
575 /* let the programmer know if he's been stupid */
576 if (rm_at_fork(function))
577 printf("WARNING: fork callout entry (%p) already present\n",
578 function);
579#endif
580 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
581 if (ep == NULL)
582 return (ENOMEM);
583 ep->function = function;
584 TAILQ_INSERT_TAIL(&fork_list, ep, next);
585 return (0);
586}
587
588/*
589 * Scan the exit callout list for the given item and remove it..
590 * Returns the number of items removed (0 or 1)
591 */
592
593int
594rm_at_fork(function)
595 forklist_fn function;
596{
597 struct forklist *ep;
598
599 TAILQ_FOREACH(ep, &fork_list, next) {
600 if (ep->function == function) {
601 TAILQ_REMOVE(&fork_list, ep, next);
602 free(ep, M_ATFORK);
603 return(1);
604 }
605 }
606 return (0);
607}
7d0bac62
MD
608
609/*
610 * Add a forked process to the run queue after any remaining setup, such
611 * as setting the fork handler, has been completed.
612 */
613
614void
615start_forked_proc(struct proc *p1, struct proc *p2)
616{
617 /*
618 * Move from SIDL to RUN queue
619 */
620 KASSERT(p2->p_stat == SIDL,
621 ("cannot start forked process, bad status: %p", p2));
622 (void) splhigh();
623 p2->p_stat = SRUN;
624 setrunqueue(p2);
625 (void) spl0();
626
627 /*
628 * Now can be swapped.
629 */
630 PRELE(p1);
631
632 /*
633 * Preserve synchronization semantics of vfork. If waiting for
634 * child to exec or exit, set P_PPWAIT on child, and sleep on our
635 * proc (in case of exit).
636 */
637 while (p2->p_flag & P_PPWAIT)
638 tsleep(p1, PWAIT, "ppwait", 0);
639}
640