Merge from vendor branch SENDMAIL:
[dragonfly.git] / sys / kern / kern_fork.c
... / ...
CommitLineData
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 1.72.2.14 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 $
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#include <sys/caps.h>
60
61#include <vm/vm.h>
62#include <sys/lock.h>
63#include <vm/pmap.h>
64#include <vm/vm_map.h>
65#include <vm/vm_extern.h>
66#include <vm/vm_zone.h>
67
68#include <sys/vmmeter.h>
69#include <sys/user.h>
70
71static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
72
73/*
74 * These are the stuctures used to create a callout list for things to do
75 * when forking a process
76 */
77struct forklist {
78 forklist_fn function;
79 TAILQ_ENTRY(forklist) next;
80};
81
82TAILQ_HEAD(forklist_head, forklist);
83static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
84
85int forksleep; /* Place for fork1() to sleep on. */
86
87/* ARGSUSED */
88int
89fork(struct fork_args *uap)
90{
91 struct proc *p = curproc;
92 struct proc *p2;
93 int error;
94
95 error = fork1(p, RFFDG | RFPROC, &p2);
96 if (error == 0) {
97 start_forked_proc(p, p2);
98 uap->sysmsg_fds[0] = p2->p_pid;
99 uap->sysmsg_fds[1] = 0;
100 }
101 return error;
102}
103
104/* ARGSUSED */
105int
106vfork(struct vfork_args *uap)
107{
108 struct proc *p = curproc;
109 struct proc *p2;
110 int error;
111
112 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
113 if (error == 0) {
114 start_forked_proc(p, p2);
115 uap->sysmsg_fds[0] = p2->p_pid;
116 uap->sysmsg_fds[1] = 0;
117 }
118 return error;
119}
120
121/*
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).
126 *
127 * Note that we only call start_forked_proc() if a new process is actually
128 * created.
129 *
130 * rfork { int flags }
131 */
132int
133rfork(struct rfork_args *uap)
134{
135 struct proc *p = curproc;
136 struct proc *p2;
137 int error;
138
139 if ((uap->flags & RFKERNELONLY) != 0)
140 return (EINVAL);
141
142 error = fork1(p, uap->flags, &p2);
143 if (error == 0) {
144 if (p2)
145 start_forked_proc(p, p2);
146 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
147 uap->sysmsg_fds[1] = 0;
148 }
149 return error;
150}
151
152
153int nprocs = 1; /* process 0 */
154static int nextpid = 0;
155
156/*
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.
163 */
164static int randompid = 0;
165
166static int
167sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
168{
169 int error, pid;
170
171 pid = randompid;
172 error = sysctl_handle_int(oidp, &pid, 0, req);
173 if (error || !req->newptr)
174 return (error);
175 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
176 pid = PID_MAX - 100;
177 else if (pid < 2) /* NOP */
178 pid = 0;
179 else if (pid < 100) /* Make it reasonable */
180 pid = 100;
181 randompid = pid;
182 return (error);
183}
184
185SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
186 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
187
188int
189fork1(struct proc *p1, int flags, struct proc **procp)
190{
191 struct proc *p2, *pptr;
192 uid_t uid;
193 struct proc *newproc;
194 int ok;
195 static int curfail = 0, pidchecked = 0;
196 static struct timeval lastfail;
197 struct forklist *ep;
198 struct filedesc_to_leader *fdtol;
199
200 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
201 return (EINVAL);
202
203 /*
204 * Here we don't create a new process, but we divorce
205 * certain parts of a process from itself.
206 */
207 if ((flags & RFPROC) == 0) {
208
209 vm_fork(p1, 0, flags);
210
211 /*
212 * Close all file descriptors.
213 */
214 if (flags & RFCFDG) {
215 struct filedesc *fdtmp;
216 fdtmp = fdinit(p1);
217 fdfree(p1);
218 p1->p_fd = fdtmp;
219 }
220
221 /*
222 * Unshare file descriptors (from parent.)
223 */
224 if (flags & RFFDG) {
225 if (p1->p_fd->fd_refcnt > 1) {
226 struct filedesc *newfd;
227 newfd = fdcopy(p1);
228 fdfree(p1);
229 p1->p_fd = newfd;
230 }
231 }
232 *procp = NULL;
233 return (0);
234 }
235
236 /*
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.
242 */
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);
249 return (EAGAIN);
250 }
251 /*
252 * Increment the nprocs resource before blocking can occur. There
253 * are hard-limits as to the number of processes that can run.
254 */
255 nprocs++;
256
257 /*
258 * Increment the count of procs running with this uid. Don't allow
259 * a nonprivileged user to exceed their current limit.
260 */
261 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
262 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
263 if (!ok) {
264 /*
265 * Back out the process count
266 */
267 nprocs--;
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);
272 return (EAGAIN);
273 }
274
275 /* Allocate new proc. */
276 newproc = zalloc(proc_zone);
277
278 /*
279 * Setup linkage for kernel based threading
280 */
281 if ((flags & RFTHREAD) != 0) {
282 newproc->p_peers = p1->p_peers;
283 p1->p_peers = newproc;
284 newproc->p_leader = p1->p_leader;
285 } else {
286 newproc->p_peers = 0;
287 newproc->p_leader = newproc;
288 }
289
290 newproc->p_wakeup = 0;
291 newproc->p_vmspace = NULL;
292 TAILQ_INIT(&newproc->p_sysmsgq);
293
294 /*
295 * Find an unused process ID. We remember a range of unused IDs
296 * ready to use (from nextpid+1 through pidchecked-1).
297 */
298 nextpid++;
299 if (randompid)
300 nextpid += arc4random() % randompid;
301retry:
302 /*
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.
306 */
307 if (nextpid >= PID_MAX) {
308 nextpid = nextpid % PID_MAX;
309 if (nextpid < 100)
310 nextpid += 100;
311 pidchecked = 0;
312 }
313 if (nextpid >= pidchecked) {
314 int doingzomb = 0;
315
316 pidchecked = PID_MAX;
317 /*
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.
321 */
322 p2 = LIST_FIRST(&allproc);
323again:
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) {
328 nextpid++;
329 if (nextpid >= pidchecked)
330 goto retry;
331 }
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;
340 }
341 if (!doingzomb) {
342 doingzomb = 1;
343 p2 = LIST_FIRST(&zombproc);
344 goto again;
345 }
346 }
347
348 p2 = newproc;
349 p2->p_stat = SIDL; /* protect against others */
350 p2->p_pid = nextpid;
351 LIST_INSERT_HEAD(&allproc, p2, p_list);
352 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
353
354 /*
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.
358 */
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));
363
364 p2->p_aioinfo = NULL;
365
366 /*
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.
370 */
371 p2->p_flag = P_INMEM;
372 if (p1->p_flag & P_PROFIL)
373 startprofclock(p2);
374 p2->p_ucred = crhold(p1->p_ucred);
375
376 if (jailed(p2->p_ucred))
377 p2->p_flag |= P_JAILED;
378
379 if (p2->p_args)
380 p2->p_args->ar_ref++;
381
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;
387 int s;
388
389 /* Create the shared sigacts structure */
390 MALLOC(newsigacts, struct sigacts *,
391 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
392 s = splhigh();
393 /*
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.
398 */
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;
403 splx(s);
404 }
405 } else {
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() */
411 }
412 if (flags & RFLINUXTHPN)
413 p2->p_sigparent = SIGUSR1;
414 else
415 p2->p_sigparent = SIGCHLD;
416
417 /* bump references to the text vnode (for procfs) */
418 p2->p_textvp = p1->p_textvp;
419 if (p2->p_textvp)
420 vref(p2->p_textvp);
421
422 if (flags & RFCFDG) {
423 p2->p_fd = fdinit(p1);
424 fdtol = NULL;
425 } else if (flags & RFFDG) {
426 p2->p_fd = fdcopy(p1);
427 fdtol = NULL;
428 } else {
429 p2->p_fd = fdshare(p1);
430 if (p1->p_fdtol == NULL)
431 p1->p_fdtol =
432 filedesc_to_leader_alloc(NULL,
433 p1->p_leader);
434 if ((flags & RFTHREAD) != 0) {
435 /*
436 * Shared file descriptor table and
437 * shared process leaders.
438 */
439 fdtol = p1->p_fdtol;
440 fdtol->fdl_refcount++;
441 } else {
442 /*
443 * Shared file descriptor table, and
444 * different process leaders
445 */
446 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
447 }
448 }
449 p2->p_fdtol = fdtol;
450
451 /*
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
455 * copy-on-write.)
456 */
457 if (p1->p_limit->p_lflags & PL_SHAREMOD) {
458 p2->p_limit = limcopy(p1->p_limit);
459 } else {
460 p2->p_limit = p1->p_limit;
461 p2->p_limit->p_refcnt++;
462 }
463
464 /*
465 * Preserve some more flags in subprocess. P_PROFIL has already
466 * been preserved.
467 */
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;
473
474 /*
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.
478 */
479 LIST_INSERT_AFTER(p1, p2, p_pglist);
480
481 /*
482 * Attach the new process to its parent.
483 *
484 * If RFNOWAIT is set, the newly created process becomes a child
485 * of init. This effectively disassociates the child from the
486 * parent.
487 */
488 if (flags & RFNOWAIT)
489 pptr = initproc;
490 else
491 pptr = p1;
492 p2->p_pptr = pptr;
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);
497
498#ifdef KTRACE
499 /*
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.
503 */
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)
507 vref(p2->p_tracep);
508 }
509#endif
510
511 /*
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
516 * scheduler.
517 *
518 * The interactivity model always starts at 0 (par value).
519 */
520 p2->p_estcpu_fork = p2->p_estcpu =
521 ESTCPULIM(p1->p_estcpu + ESTCPURAMP);
522 p2->p_interactive = 0;
523
524 /*
525 * This begins the section where we must prevent the parent
526 * from being swapped.
527 */
528 PHOLD(p1);
529
530 /*
531 * Finish creating the child process. It will return via a different
532 * execution path later. (ie: directly into user mode)
533 */
534 vm_fork(p1, p2, flags);
535 caps_fork(p1, p2, flags);
536
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;
546 } else {
547 mycpu->gd_cnt.v_rforks++;
548 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
549 }
550
551 /*
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
555 */
556 TAILQ_FOREACH(ep, &fork_list, next) {
557 (*ep->function)(p1, p2, flags);
558 }
559
560 /*
561 * Make child runnable and add to run queue.
562 */
563 microtime(&p2->p_thread->td_start);
564 p2->p_acflag = AFORK;
565
566 /*
567 * tell any interested parties about the new process
568 */
569 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
570
571 /*
572 * Return child proc pointer to parent.
573 */
574 *procp = p2;
575 return (0);
576}
577
578/*
579 * The next two functionms are general routines to handle adding/deleting
580 * items on the fork callout list.
581 *
582 * at_fork():
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.
586 */
587int
588at_fork(forklist_fn function)
589{
590 struct forklist *ep;
591
592#ifdef INVARIANTS
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",
596 function);
597 }
598#endif
599 ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
600 ep->function = function;
601 TAILQ_INSERT_TAIL(&fork_list, ep, next);
602 return (0);
603}
604
605/*
606 * Scan the exit callout list for the given item and remove it..
607 * Returns the number of items removed (0 or 1)
608 */
609int
610rm_at_fork(forklist_fn function)
611{
612 struct forklist *ep;
613
614 TAILQ_FOREACH(ep, &fork_list, next) {
615 if (ep->function == function) {
616 TAILQ_REMOVE(&fork_list, ep, next);
617 free(ep, M_ATFORK);
618 return(1);
619 }
620 }
621 return (0);
622}
623
624/*
625 * Add a forked process to the run queue after any remaining setup, such
626 * as setting the fork handler, has been completed.
627 */
628void
629start_forked_proc(struct proc *p1, struct proc *p2)
630{
631 /*
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().
635 *
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.
639 */
640 KASSERT(p2 && p2->p_stat == SIDL,
641 ("cannot start forked process, bad status: %p", p2));
642 resetpriority(p2);
643 (void) splhigh();
644 p2->p_stat = SRUN;
645 setrunqueue(p2);
646 (void) spl0();
647
648 /*
649 * Now can be swapped.
650 */
651 PRELE(p1);
652
653 /*
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).
657 */
658 while (p2->p_flag & P_PPWAIT)
659 tsleep(p1, 0, "ppwait", 0);
660}
661