kernel - TMPFS - Features, don't sync on umount, enforce snocache on root
[dragonfly.git] / sys / kern / kern_fork.c
blob12b1fcaa9c0cbfa7254377144c54e2cbee0427ea
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.
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.
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.
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>
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>
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>
67 #include <sys/vmmeter.h>
68 #include <sys/thread2.h>
69 #include <sys/signal2.h>
70 #include <sys/spinlock2.h>
71 #include <sys/mplock2.h>
73 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
76 * These are the stuctures used to create a callout list for things to do
77 * when forking a process
79 struct forklist {
80 forklist_fn function;
81 TAILQ_ENTRY(forklist) next;
84 TAILQ_HEAD(forklist_head, forklist);
85 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
87 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
89 int forksleep; /* Place for fork1() to sleep on. */
92 * Red-Black tree support for LWPs
95 static int
96 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
98 if (lp1->lwp_tid < lp2->lwp_tid)
99 return(-1);
100 if (lp1->lwp_tid > lp2->lwp_tid)
101 return(1);
102 return(0);
105 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
108 * Fork system call
110 * MPALMOSTSAFE
113 sys_fork(struct fork_args *uap)
115 struct lwp *lp = curthread->td_lwp;
116 struct proc *p2;
117 int error;
119 get_mplock();
120 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
121 if (error == 0) {
122 start_forked_proc(lp, p2);
123 uap->sysmsg_fds[0] = p2->p_pid;
124 uap->sysmsg_fds[1] = 0;
126 rel_mplock();
127 return error;
131 * MPALMOSTSAFE
134 sys_vfork(struct vfork_args *uap)
136 struct lwp *lp = curthread->td_lwp;
137 struct proc *p2;
138 int error;
140 get_mplock();
141 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
142 if (error == 0) {
143 start_forked_proc(lp, p2);
144 uap->sysmsg_fds[0] = p2->p_pid;
145 uap->sysmsg_fds[1] = 0;
147 rel_mplock();
148 return error;
152 * Handle rforks. An rfork may (1) operate on the current process without
153 * creating a new, (2) create a new process that shared the current process's
154 * vmspace, signals, and/or descriptors, or (3) create a new process that does
155 * not share these things (normal fork).
157 * Note that we only call start_forked_proc() if a new process is actually
158 * created.
160 * rfork { int flags }
162 * MPALMOSTSAFE
165 sys_rfork(struct rfork_args *uap)
167 struct lwp *lp = curthread->td_lwp;
168 struct proc *p2;
169 int error;
171 if ((uap->flags & RFKERNELONLY) != 0)
172 return (EINVAL);
174 get_mplock();
175 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
176 if (error == 0) {
177 if (p2)
178 start_forked_proc(lp, p2);
179 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
180 uap->sysmsg_fds[1] = 0;
182 rel_mplock();
183 return error;
187 * MPALMOSTSAFE
190 sys_lwp_create(struct lwp_create_args *uap)
192 struct proc *p = curproc;
193 struct lwp *lp;
194 struct lwp_params params;
195 int error;
197 error = copyin(uap->params, &params, sizeof(params));
198 if (error)
199 goto fail2;
201 get_mplock();
202 plimit_lwp_fork(p); /* force exclusive access */
203 lp = lwp_fork(curthread->td_lwp, p, RFPROC);
204 error = cpu_prepare_lwp(lp, &params);
205 if (params.tid1 != NULL &&
206 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
207 goto fail;
208 if (params.tid2 != NULL &&
209 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
210 goto fail;
213 * Now schedule the new lwp.
215 p->p_usched->resetpriority(lp);
216 crit_enter();
217 lp->lwp_stat = LSRUN;
218 p->p_usched->setrunqueue(lp);
219 crit_exit();
220 rel_mplock();
222 return (0);
224 fail:
225 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
226 --p->p_nthreads;
227 /* lwp_dispose expects an exited lwp, and a held proc */
228 lp->lwp_flag |= LWP_WEXIT;
229 lp->lwp_thread->td_flags |= TDF_EXITING;
230 PHOLD(p);
231 lwp_dispose(lp);
232 rel_mplock();
233 fail2:
234 return (error);
237 int nprocs = 1; /* process 0 */
240 fork1(struct lwp *lp1, int flags, struct proc **procp)
242 struct proc *p1 = lp1->lwp_proc;
243 struct proc *p2, *pptr;
244 struct pgrp *pgrp;
245 uid_t uid;
246 int ok, error;
247 static int curfail = 0;
248 static struct timeval lastfail;
249 struct forklist *ep;
250 struct filedesc_to_leader *fdtol;
252 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
253 return (EINVAL);
256 * Here we don't create a new process, but we divorce
257 * certain parts of a process from itself.
259 if ((flags & RFPROC) == 0) {
261 * This kind of stunt does not work anymore if
262 * there are native threads (lwps) running
264 if (p1->p_nthreads != 1)
265 return (EINVAL);
267 vm_fork(p1, 0, flags);
270 * Close all file descriptors.
272 if (flags & RFCFDG) {
273 struct filedesc *fdtmp;
274 fdtmp = fdinit(p1);
275 fdfree(p1, fdtmp);
279 * Unshare file descriptors (from parent.)
281 if (flags & RFFDG) {
282 if (p1->p_fd->fd_refcnt > 1) {
283 struct filedesc *newfd;
284 newfd = fdcopy(p1);
285 fdfree(p1, newfd);
288 *procp = NULL;
289 return (0);
293 * Interlock against process group signal delivery. If signals
294 * are pending after the interlock is obtained we have to restart
295 * the system call to process the signals. If we don't the child
296 * can miss a pgsignal (such as ^C) sent during the fork.
298 * We can't use CURSIG() here because it will process any STOPs
299 * and cause the process group lock to be held indefinitely. If
300 * a STOP occurs, the fork will be restarted after the CONT.
302 error = 0;
303 pgrp = NULL;
304 if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) {
305 lockmgr(&pgrp->pg_lock, LK_SHARED);
306 if (CURSIG_NOBLOCK(lp1)) {
307 error = ERESTART;
308 goto done;
313 * Although process entries are dynamically created, we still keep
314 * a global limit on the maximum number we will create. Don't allow
315 * a nonprivileged user to use the last ten processes; don't let root
316 * exceed the limit. The variable nprocs is the current number of
317 * processes, maxproc is the limit.
319 uid = lp1->lwp_thread->td_ucred->cr_ruid;
320 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
321 if (ppsratecheck(&lastfail, &curfail, 1))
322 kprintf("maxproc limit exceeded by uid %d, please "
323 "see tuning(7) and login.conf(5).\n", uid);
324 tsleep(&forksleep, 0, "fork", hz / 2);
325 error = EAGAIN;
326 goto done;
329 * Increment the nprocs resource before blocking can occur. There
330 * are hard-limits as to the number of processes that can run.
332 nprocs++;
335 * Increment the count of procs running with this uid. Don't allow
336 * a nonprivileged user to exceed their current limit.
338 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
339 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
340 if (!ok) {
342 * Back out the process count
344 nprocs--;
345 if (ppsratecheck(&lastfail, &curfail, 1))
346 kprintf("maxproc limit exceeded by uid %d, please "
347 "see tuning(7) and login.conf(5).\n", uid);
348 tsleep(&forksleep, 0, "fork", hz / 2);
349 error = EAGAIN;
350 goto done;
353 /* Allocate new proc. */
354 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
357 * Setup linkage for kernel based threading XXX lwp
359 if (flags & RFTHREAD) {
360 p2->p_peers = p1->p_peers;
361 p1->p_peers = p2;
362 p2->p_leader = p1->p_leader;
363 } else {
364 p2->p_leader = p2;
367 RB_INIT(&p2->p_lwp_tree);
368 spin_init(&p2->p_spin);
369 p2->p_lasttid = -1; /* first tid will be 0 */
372 * Setting the state to SIDL protects the partially initialized
373 * process once it starts getting hooked into the rest of the system.
375 p2->p_stat = SIDL;
376 proc_add_allproc(p2);
379 * Make a proc table entry for the new process.
380 * The whole structure was zeroed above, so copy the section that is
381 * copied directly from the parent.
383 bcopy(&p1->p_startcopy, &p2->p_startcopy,
384 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
387 * Duplicate sub-structures as needed.
388 * Increase reference counts on shared objects.
390 if (p1->p_flag & P_PROFIL)
391 startprofclock(p2);
392 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
393 KKASSERT(p2->p_lock == 0);
395 if (jailed(p2->p_ucred))
396 p2->p_flag |= P_JAILED;
398 if (p2->p_args)
399 p2->p_args->ar_ref++;
401 p2->p_usched = p1->p_usched;
403 if (flags & RFSIGSHARE) {
404 p2->p_sigacts = p1->p_sigacts;
405 p2->p_sigacts->ps_refcnt++;
406 } else {
407 p2->p_sigacts = (struct sigacts *)kmalloc(sizeof(*p2->p_sigacts),
408 M_SUBPROC, M_WAITOK);
409 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
410 p2->p_sigacts->ps_refcnt = 1;
412 if (flags & RFLINUXTHPN)
413 p2->p_sigparent = SIGUSR1;
414 else
415 p2->p_sigparent = SIGCHLD;
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);
423 * Handle file descriptors
425 if (flags & RFCFDG) {
426 p2->p_fd = fdinit(p1);
427 fdtol = NULL;
428 } else if (flags & RFFDG) {
429 p2->p_fd = fdcopy(p1);
430 fdtol = NULL;
431 } else {
432 p2->p_fd = fdshare(p1);
433 if (p1->p_fdtol == NULL)
434 p1->p_fdtol =
435 filedesc_to_leader_alloc(NULL,
436 p1->p_leader);
437 if ((flags & RFTHREAD) != 0) {
439 * Shared file descriptor table and
440 * shared process leaders.
442 fdtol = p1->p_fdtol;
443 fdtol->fdl_refcount++;
444 } else {
446 * Shared file descriptor table, and
447 * different process leaders
449 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
452 p2->p_fdtol = fdtol;
453 p2->p_limit = plimit_fork(p1);
456 * Preserve some more flags in subprocess. P_PROFIL has already
457 * been preserved.
459 p2->p_flag |= p1->p_flag & P_SUGID;
460 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
461 p2->p_flag |= P_CONTROLT;
462 if (flags & RFPPWAIT)
463 p2->p_flag |= P_PPWAIT;
466 * Inherit the virtual kernel structure (allows a virtual kernel
467 * to fork to simulate multiple cpus).
469 if (p1->p_vkernel)
470 vkernel_inherit(p1, p2);
473 * Once we are on a pglist we may receive signals. XXX we might
474 * race a ^C being sent to the process group by not receiving it
475 * at all prior to this line.
477 LIST_INSERT_AFTER(p1, p2, p_pglist);
480 * Attach the new process to its parent.
482 * If RFNOWAIT is set, the newly created process becomes a child
483 * of init. This effectively disassociates the child from the
484 * parent.
486 if (flags & RFNOWAIT)
487 pptr = initproc;
488 else
489 pptr = p1;
490 p2->p_pptr = pptr;
491 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
492 LIST_INIT(&p2->p_children);
493 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
494 callout_init(&p2->p_ithandle);
496 #ifdef KTRACE
498 * Copy traceflag and tracefile if enabled. If not inherited,
499 * these were zeroed above but we still could have a trace race
500 * so make sure p2's p_tracenode is NULL.
502 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
503 p2->p_traceflag = p1->p_traceflag;
504 p2->p_tracenode = ktrinherit(p1->p_tracenode);
506 #endif
509 * This begins the section where we must prevent the parent
510 * from being swapped.
512 * Gets PRELE'd in the caller in start_forked_proc().
514 PHOLD(p1);
516 vm_fork(p1, p2, flags);
519 * Create the first lwp associated with the new proc.
520 * It will return via a different execution path later, directly
521 * into userland, after it was put on the runq by
522 * start_forked_proc().
524 lwp_fork(lp1, p2, flags);
526 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
527 mycpu->gd_cnt.v_forks++;
528 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
529 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
530 mycpu->gd_cnt.v_vforks++;
531 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
532 } else if (p1 == &proc0) {
533 mycpu->gd_cnt.v_kthreads++;
534 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
535 } else {
536 mycpu->gd_cnt.v_rforks++;
537 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
541 * Both processes are set up, now check if any loadable modules want
542 * to adjust anything.
543 * What if they have an error? XXX
545 TAILQ_FOREACH(ep, &fork_list, next) {
546 (*ep->function)(p1, p2, flags);
550 * Set the start time. Note that the process is not runnable. The
551 * caller is responsible for making it runnable.
553 microtime(&p2->p_start);
554 p2->p_acflag = AFORK;
557 * tell any interested parties about the new process
559 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
562 * Return child proc pointer to parent.
564 *procp = p2;
565 done:
566 if (pgrp)
567 lockmgr(&pgrp->pg_lock, LK_RELEASE);
568 return (error);
571 static struct lwp *
572 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
574 struct lwp *lp;
575 struct thread *td;
577 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
579 lp->lwp_proc = destproc;
580 lp->lwp_vmspace = destproc->p_vmspace;
581 lp->lwp_stat = LSRUN;
582 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
583 (unsigned) ((caddr_t)&lp->lwp_endcopy -
584 (caddr_t)&lp->lwp_startcopy));
585 lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
587 * Set cpbase to the last timeout that occured (not the upcoming
588 * timeout).
590 * A critical section is required since a timer IPI can update
591 * scheduler specific data.
593 crit_enter();
594 lp->lwp_cpbase = mycpu->gd_schedclock.time -
595 mycpu->gd_schedclock.periodic;
596 destproc->p_usched->heuristic_forking(origlp, lp);
597 crit_exit();
598 lp->lwp_cpumask &= usched_mastermask;
601 * Assign a TID to the lp. Loop until the insert succeeds (returns
602 * NULL).
604 lp->lwp_tid = destproc->p_lasttid;
605 do {
606 if (++lp->lwp_tid < 0)
607 lp->lwp_tid = 1;
608 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
609 destproc->p_lasttid = lp->lwp_tid;
610 destproc->p_nthreads++;
612 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
613 lp->lwp_thread = td;
614 td->td_proc = destproc;
615 td->td_lwp = lp;
616 td->td_switch = cpu_heavy_switch;
617 #ifdef SMP
618 KKASSERT(td->td_mpcount == 1);
619 #endif
620 lwkt_setpri(td, TDPRI_KERN_USER);
621 lwkt_set_comm(td, "%s", destproc->p_comm);
624 * cpu_fork will copy and update the pcb, set up the kernel stack,
625 * and make the child ready to run.
627 cpu_fork(origlp, lp, flags);
628 caps_fork(origlp->lwp_thread, lp->lwp_thread);
629 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
631 return (lp);
635 * The next two functionms are general routines to handle adding/deleting
636 * items on the fork callout list.
638 * at_fork():
639 * Take the arguments given and put them onto the fork callout list,
640 * However first make sure that it's not already there.
641 * Returns 0 on success or a standard error number.
644 at_fork(forklist_fn function)
646 struct forklist *ep;
648 #ifdef INVARIANTS
649 /* let the programmer know if he's been stupid */
650 if (rm_at_fork(function)) {
651 kprintf("WARNING: fork callout entry (%p) already present\n",
652 function);
654 #endif
655 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
656 ep->function = function;
657 TAILQ_INSERT_TAIL(&fork_list, ep, next);
658 return (0);
662 * Scan the exit callout list for the given item and remove it..
663 * Returns the number of items removed (0 or 1)
666 rm_at_fork(forklist_fn function)
668 struct forklist *ep;
670 TAILQ_FOREACH(ep, &fork_list, next) {
671 if (ep->function == function) {
672 TAILQ_REMOVE(&fork_list, ep, next);
673 kfree(ep, M_ATFORK);
674 return(1);
677 return (0);
681 * Add a forked process to the run queue after any remaining setup, such
682 * as setting the fork handler, has been completed.
684 void
685 start_forked_proc(struct lwp *lp1, struct proc *p2)
687 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
690 * Move from SIDL to RUN queue, and activate the process's thread.
691 * Activation of the thread effectively makes the process "a"
692 * current process, so we do not setrunqueue().
694 * YYY setrunqueue works here but we should clean up the trampoline
695 * code so we just schedule the LWKT thread and let the trampoline
696 * deal with the userland scheduler on return to userland.
698 KASSERT(p2->p_stat == SIDL,
699 ("cannot start forked process, bad status: %p", p2));
700 p2->p_usched->resetpriority(lp2);
701 crit_enter();
702 p2->p_stat = SACTIVE;
703 lp2->lwp_stat = LSRUN;
704 p2->p_usched->setrunqueue(lp2);
705 crit_exit();
708 * Now can be swapped.
710 PRELE(lp1->lwp_proc);
713 * Preserve synchronization semantics of vfork. If waiting for
714 * child to exec or exit, set P_PPWAIT on child, and sleep on our
715 * proc (in case of exit).
717 while (p2->p_flag & P_PPWAIT)
718 tsleep(lp1->lwp_proc, 0, "ppwait", 0);