FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_fork.c
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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_kstack_pages.h"
43 #include "opt_procdesc.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/fcntl.h>
50 #include <sys/filedesc.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/kthread.h>
54 #include <sys/sysctl.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mutex.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procdesc.h>
61 #include <sys/pioctl.h>
62 #include <sys/ptrace.h>
63 #include <sys/racct.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sched.h>
66 #include <sys/syscall.h>
67 #include <sys/vmmeter.h>
68 #include <sys/vnode.h>
69 #include <sys/acct.h>
70 #include <sys/ktr.h>
71 #include <sys/ktrace.h>
72 #include <sys/unistd.h>
73 #include <sys/sdt.h>
74 #include <sys/sx.h>
75 #include <sys/sysent.h>
76 #include <sys/signalvar.h>
77
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80
81 #include <vm/vm.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_extern.h>
85 #include <vm/uma.h>
86
87 #ifdef KDTRACE_HOOKS
88 #include <sys/dtrace_bsd.h>
89 dtrace_fork_func_t dtrace_fasttrap_fork;
90 #endif
91
92 SDT_PROVIDER_DECLARE(proc);
93 SDT_PROBE_DEFINE3(proc, kernel, , create, "struct proc *",
94 "struct proc *", "int");
95
96 #ifndef _SYS_SYSPROTO_H_
97 struct fork_args {
98 int dummy;
99 };
100 #endif
101
102 /* ARGSUSED */
103 int
104 sys_fork(struct thread *td, struct fork_args *uap)
105 {
106 int error;
107 struct proc *p2;
108
109 error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0);
110 if (error == 0) {
111 td->td_retval[0] = p2->p_pid;
112 td->td_retval[1] = 0;
113 }
114 return (error);
115 }
116
117 /* ARGUSED */
118 int
119 sys_pdfork(td, uap)
120 struct thread *td;
121 struct pdfork_args *uap;
122 {
123 #ifdef PROCDESC
124 int error, fd;
125 struct proc *p2;
126
127 /*
128 * It is necessary to return fd by reference because 0 is a valid file
129 * descriptor number, and the child needs to be able to distinguish
130 * itself from the parent using the return value.
131 */
132 error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2,
133 &fd, uap->flags);
134 if (error == 0) {
135 td->td_retval[0] = p2->p_pid;
136 td->td_retval[1] = 0;
137 error = copyout(&fd, uap->fdp, sizeof(fd));
138 }
139 return (error);
140 #else
141 return (ENOSYS);
142 #endif
143 }
144
145 /* ARGSUSED */
146 int
147 sys_vfork(struct thread *td, struct vfork_args *uap)
148 {
149 int error, flags;
150 struct proc *p2;
151
152 #ifdef XEN
153 flags = RFFDG | RFPROC; /* validate that this is still an issue */
154 #else
155 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
156 #endif
157 error = fork1(td, flags, 0, &p2, NULL, 0);
158 if (error == 0) {
159 td->td_retval[0] = p2->p_pid;
160 td->td_retval[1] = 0;
161 }
162 return (error);
163 }
164
165 int
166 sys_rfork(struct thread *td, struct rfork_args *uap)
167 {
168 struct proc *p2;
169 int error;
170
171 /* Don't allow kernel-only flags. */
172 if ((uap->flags & RFKERNELONLY) != 0)
173 return (EINVAL);
174
175 AUDIT_ARG_FFLAGS(uap->flags);
176 error = fork1(td, uap->flags, 0, &p2, NULL, 0);
177 if (error == 0) {
178 td->td_retval[0] = p2 ? p2->p_pid : 0;
179 td->td_retval[1] = 0;
180 }
181 return (error);
182 }
183
184 int nprocs = 1; /* process 0 */
185 int lastpid = 0;
186 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
187 "Last used PID");
188
189 /*
190 * Random component to lastpid generation. We mix in a random factor to make
191 * it a little harder to predict. We sanity check the modulus value to avoid
192 * doing it in critical paths. Don't let it be too small or we pointlessly
193 * waste randomness entropy, and don't let it be impossibly large. Using a
194 * modulus that is too big causes a LOT more process table scans and slows
195 * down fork processing as the pidchecked caching is defeated.
196 */
197 static int randompid = 0;
198
199 static int
200 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
201 {
202 int error, pid;
203
204 error = sysctl_wire_old_buffer(req, sizeof(int));
205 if (error != 0)
206 return(error);
207 sx_xlock(&allproc_lock);
208 pid = randompid;
209 error = sysctl_handle_int(oidp, &pid, 0, req);
210 if (error == 0 && req->newptr != NULL) {
211 if (pid < 0 || pid > pid_max - 100) /* out of range */
212 pid = pid_max - 100;
213 else if (pid < 2) /* NOP */
214 pid = 0;
215 else if (pid < 100) /* Make it reasonable */
216 pid = 100;
217 randompid = pid;
218 }
219 sx_xunlock(&allproc_lock);
220 return (error);
221 }
222
223 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
224 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
225
226 static int
227 fork_findpid(int flags)
228 {
229 struct proc *p;
230 int trypid;
231 static int pidchecked = 0;
232
233 /*
234 * Requires allproc_lock in order to iterate over the list
235 * of processes, and proctree_lock to access p_pgrp.
236 */
237 sx_assert(&allproc_lock, SX_LOCKED);
238 sx_assert(&proctree_lock, SX_LOCKED);
239
240 /*
241 * Find an unused process ID. We remember a range of unused IDs
242 * ready to use (from lastpid+1 through pidchecked-1).
243 *
244 * If RFHIGHPID is set (used during system boot), do not allocate
245 * low-numbered pids.
246 */
247 trypid = lastpid + 1;
248 if (flags & RFHIGHPID) {
249 if (trypid < 10)
250 trypid = 10;
251 } else {
252 if (randompid)
253 trypid += arc4random() % randompid;
254 }
255 retry:
256 /*
257 * If the process ID prototype has wrapped around,
258 * restart somewhat above 0, as the low-numbered procs
259 * tend to include daemons that don't exit.
260 */
261 if (trypid >= pid_max) {
262 trypid = trypid % pid_max;
263 if (trypid < 100)
264 trypid += 100;
265 pidchecked = 0;
266 }
267 if (trypid >= pidchecked) {
268 int doingzomb = 0;
269
270 pidchecked = PID_MAX;
271 /*
272 * Scan the active and zombie procs to check whether this pid
273 * is in use. Remember the lowest pid that's greater
274 * than trypid, so we can avoid checking for a while.
275 */
276 p = LIST_FIRST(&allproc);
277 again:
278 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
279 while (p->p_pid == trypid ||
280 (p->p_pgrp != NULL &&
281 (p->p_pgrp->pg_id == trypid ||
282 (p->p_session != NULL &&
283 p->p_session->s_sid == trypid)))) {
284 trypid++;
285 if (trypid >= pidchecked)
286 goto retry;
287 }
288 if (p->p_pid > trypid && pidchecked > p->p_pid)
289 pidchecked = p->p_pid;
290 if (p->p_pgrp != NULL) {
291 if (p->p_pgrp->pg_id > trypid &&
292 pidchecked > p->p_pgrp->pg_id)
293 pidchecked = p->p_pgrp->pg_id;
294 if (p->p_session != NULL &&
295 p->p_session->s_sid > trypid &&
296 pidchecked > p->p_session->s_sid)
297 pidchecked = p->p_session->s_sid;
298 }
299 }
300 if (!doingzomb) {
301 doingzomb = 1;
302 p = LIST_FIRST(&zombproc);
303 goto again;
304 }
305 }
306
307 /*
308 * RFHIGHPID does not mess with the lastpid counter during boot.
309 */
310 if (flags & RFHIGHPID)
311 pidchecked = 0;
312 else
313 lastpid = trypid;
314
315 return (trypid);
316 }
317
318 static int
319 fork_norfproc(struct thread *td, int flags)
320 {
321 int error;
322 struct proc *p1;
323
324 KASSERT((flags & RFPROC) == 0,
325 ("fork_norfproc called with RFPROC set"));
326 p1 = td->td_proc;
327
328 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
329 (flags & (RFCFDG | RFFDG))) {
330 PROC_LOCK(p1);
331 if (thread_single(SINGLE_BOUNDARY)) {
332 PROC_UNLOCK(p1);
333 return (ERESTART);
334 }
335 PROC_UNLOCK(p1);
336 }
337
338 error = vm_forkproc(td, NULL, NULL, NULL, flags);
339 if (error)
340 goto fail;
341
342 /*
343 * Close all file descriptors.
344 */
345 if (flags & RFCFDG) {
346 struct filedesc *fdtmp;
347 fdtmp = fdinit(td->td_proc->p_fd);
348 fdfree(td);
349 p1->p_fd = fdtmp;
350 }
351
352 /*
353 * Unshare file descriptors (from parent).
354 */
355 if (flags & RFFDG)
356 fdunshare(p1, td);
357
358 fail:
359 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
360 (flags & (RFCFDG | RFFDG))) {
361 PROC_LOCK(p1);
362 thread_single_end();
363 PROC_UNLOCK(p1);
364 }
365 return (error);
366 }
367
368 static void
369 do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
370 struct vmspace *vm2, int pdflags)
371 {
372 struct proc *p1, *pptr;
373 int p2_held, trypid;
374 struct filedesc *fd;
375 struct filedesc_to_leader *fdtol;
376 struct sigacts *newsigacts;
377
378 sx_assert(&proctree_lock, SX_SLOCKED);
379 sx_assert(&allproc_lock, SX_XLOCKED);
380
381 p2_held = 0;
382 p1 = td->td_proc;
383
384 /*
385 * Increment the nprocs resource before blocking can occur. There
386 * are hard-limits as to the number of processes that can run.
387 */
388 nprocs++;
389
390 trypid = fork_findpid(flags);
391
392 sx_sunlock(&proctree_lock);
393
394 p2->p_state = PRS_NEW; /* protect against others */
395 p2->p_pid = trypid;
396 AUDIT_ARG_PID(p2->p_pid);
397 LIST_INSERT_HEAD(&allproc, p2, p_list);
398 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
399 tidhash_add(td2);
400 PROC_LOCK(p2);
401 PROC_LOCK(p1);
402
403 sx_xunlock(&allproc_lock);
404
405 bcopy(&p1->p_startcopy, &p2->p_startcopy,
406 __rangeof(struct proc, p_startcopy, p_endcopy));
407 pargs_hold(p2->p_args);
408 PROC_UNLOCK(p1);
409
410 bzero(&p2->p_startzero,
411 __rangeof(struct proc, p_startzero, p_endzero));
412 p2->p_treeflag = 0;
413
414 p2->p_ucred = crhold(td->td_ucred);
415
416 /* Tell the prison that we exist. */
417 prison_proc_hold(p2->p_ucred->cr_prison);
418
419 PROC_UNLOCK(p2);
420
421 /*
422 * Malloc things while we don't hold any locks.
423 */
424 if (flags & RFSIGSHARE)
425 newsigacts = NULL;
426 else
427 newsigacts = sigacts_alloc();
428
429 /*
430 * Copy filedesc.
431 */
432 if (flags & RFCFDG) {
433 fd = fdinit(p1->p_fd);
434 fdtol = NULL;
435 } else if (flags & RFFDG) {
436 fd = fdcopy(p1->p_fd);
437 fdtol = NULL;
438 } else {
439 fd = fdshare(p1->p_fd);
440 if (p1->p_fdtol == NULL)
441 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
442 p1->p_leader);
443 if ((flags & RFTHREAD) != 0) {
444 /*
445 * Shared file descriptor table, and shared
446 * process leaders.
447 */
448 fdtol = p1->p_fdtol;
449 FILEDESC_XLOCK(p1->p_fd);
450 fdtol->fdl_refcount++;
451 FILEDESC_XUNLOCK(p1->p_fd);
452 } else {
453 /*
454 * Shared file descriptor table, and different
455 * process leaders.
456 */
457 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
458 p1->p_fd, p2);
459 }
460 }
461 /*
462 * Make a proc table entry for the new process.
463 * Start by zeroing the section of proc that is zero-initialized,
464 * then copy the section that is copied directly from the parent.
465 */
466
467 PROC_LOCK(p2);
468 PROC_LOCK(p1);
469
470 bzero(&td2->td_startzero,
471 __rangeof(struct thread, td_startzero, td_endzero));
472
473 bcopy(&td->td_startcopy, &td2->td_startcopy,
474 __rangeof(struct thread, td_startcopy, td_endcopy));
475
476 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
477 td2->td_sigstk = td->td_sigstk;
478 td2->td_sigmask = td->td_sigmask;
479 td2->td_flags = TDF_INMEM;
480 td2->td_lend_user_pri = PRI_MAX;
481 td2->td_dbg_sc_code = td->td_dbg_sc_code;
482 td2->td_dbg_sc_narg = td->td_dbg_sc_narg;
483
484 #ifdef VIMAGE
485 td2->td_vnet = NULL;
486 td2->td_vnet_lpush = NULL;
487 #endif
488
489 /*
490 * Allow the scheduler to initialize the child.
491 */
492 thread_lock(td);
493 sched_fork(td, td2);
494 thread_unlock(td);
495
496 /*
497 * Duplicate sub-structures as needed.
498 * Increase reference counts on shared objects.
499 */
500 p2->p_flag = P_INMEM;
501 p2->p_flag2 = 0;
502 p2->p_swtick = ticks;
503 if (p1->p_flag & P_PROFIL)
504 startprofclock(p2);
505 td2->td_ucred = crhold(p2->p_ucred);
506
507 if (flags & RFSIGSHARE) {
508 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
509 } else {
510 sigacts_copy(newsigacts, p1->p_sigacts);
511 p2->p_sigacts = newsigacts;
512 }
513
514 if (flags & RFTSIGZMB)
515 p2->p_sigparent = RFTSIGNUM(flags);
516 else if (flags & RFLINUXTHPN)
517 p2->p_sigparent = SIGUSR1;
518 else
519 p2->p_sigparent = SIGCHLD;
520
521 p2->p_textvp = p1->p_textvp;
522 p2->p_fd = fd;
523 p2->p_fdtol = fdtol;
524
525 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
526 p2->p_flag |= P_PROTECTED;
527 p2->p_flag2 |= P2_INHERIT_PROTECTED;
528 }
529
530 /*
531 * p_limit is copy-on-write. Bump its refcount.
532 */
533 lim_fork(p1, p2);
534
535 pstats_fork(p1->p_stats, p2->p_stats);
536
537 PROC_UNLOCK(p1);
538 PROC_UNLOCK(p2);
539
540 /* Bump references to the text vnode (for procfs). */
541 if (p2->p_textvp)
542 vref(p2->p_textvp);
543
544 /*
545 * Set up linkage for kernel based threading.
546 */
547 if ((flags & RFTHREAD) != 0) {
548 mtx_lock(&ppeers_lock);
549 p2->p_peers = p1->p_peers;
550 p1->p_peers = p2;
551 p2->p_leader = p1->p_leader;
552 mtx_unlock(&ppeers_lock);
553 PROC_LOCK(p1->p_leader);
554 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
555 PROC_UNLOCK(p1->p_leader);
556 /*
557 * The task leader is exiting, so process p1 is
558 * going to be killed shortly. Since p1 obviously
559 * isn't dead yet, we know that the leader is either
560 * sending SIGKILL's to all the processes in this
561 * task or is sleeping waiting for all the peers to
562 * exit. We let p1 complete the fork, but we need
563 * to go ahead and kill the new process p2 since
564 * the task leader may not get a chance to send
565 * SIGKILL to it. We leave it on the list so that
566 * the task leader will wait for this new process
567 * to commit suicide.
568 */
569 PROC_LOCK(p2);
570 kern_psignal(p2, SIGKILL);
571 PROC_UNLOCK(p2);
572 } else
573 PROC_UNLOCK(p1->p_leader);
574 } else {
575 p2->p_peers = NULL;
576 p2->p_leader = p2;
577 }
578
579 sx_xlock(&proctree_lock);
580 PGRP_LOCK(p1->p_pgrp);
581 PROC_LOCK(p2);
582 PROC_LOCK(p1);
583
584 /*
585 * Preserve some more flags in subprocess. P_PROFIL has already
586 * been preserved.
587 */
588 p2->p_flag |= p1->p_flag & P_SUGID;
589 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
590 SESS_LOCK(p1->p_session);
591 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
592 p2->p_flag |= P_CONTROLT;
593 SESS_UNLOCK(p1->p_session);
594 if (flags & RFPPWAIT)
595 p2->p_flag |= P_PPWAIT;
596
597 p2->p_pgrp = p1->p_pgrp;
598 LIST_INSERT_AFTER(p1, p2, p_pglist);
599 PGRP_UNLOCK(p1->p_pgrp);
600 LIST_INIT(&p2->p_children);
601 LIST_INIT(&p2->p_orphans);
602
603 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
604
605 /*
606 * If PF_FORK is set, the child process inherits the
607 * procfs ioctl flags from its parent.
608 */
609 if (p1->p_pfsflags & PF_FORK) {
610 p2->p_stops = p1->p_stops;
611 p2->p_pfsflags = p1->p_pfsflags;
612 }
613
614 /*
615 * This begins the section where we must prevent the parent
616 * from being swapped.
617 */
618 _PHOLD(p1);
619 PROC_UNLOCK(p1);
620
621 /*
622 * Attach the new process to its parent.
623 *
624 * If RFNOWAIT is set, the newly created process becomes a child
625 * of init. This effectively disassociates the child from the
626 * parent.
627 */
628 if (flags & RFNOWAIT)
629 pptr = initproc;
630 else
631 pptr = p1;
632 p2->p_pptr = pptr;
633 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
634 sx_xunlock(&proctree_lock);
635
636 /* Inform accounting that we have forked. */
637 p2->p_acflag = AFORK;
638 PROC_UNLOCK(p2);
639
640 #ifdef KTRACE
641 ktrprocfork(p1, p2);
642 #endif
643
644 /*
645 * Finish creating the child process. It will return via a different
646 * execution path later. (ie: directly into user mode)
647 */
648 vm_forkproc(td, p2, td2, vm2, flags);
649
650 if (flags == (RFFDG | RFPROC)) {
651 PCPU_INC(cnt.v_forks);
652 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
653 p2->p_vmspace->vm_ssize);
654 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
655 PCPU_INC(cnt.v_vforks);
656 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
657 p2->p_vmspace->vm_ssize);
658 } else if (p1 == &proc0) {
659 PCPU_INC(cnt.v_kthreads);
660 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
661 p2->p_vmspace->vm_ssize);
662 } else {
663 PCPU_INC(cnt.v_rforks);
664 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
665 p2->p_vmspace->vm_ssize);
666 }
667
668 #ifdef PROCDESC
669 /*
670 * Associate the process descriptor with the process before anything
671 * can happen that might cause that process to need the descriptor.
672 * However, don't do this until after fork(2) can no longer fail.
673 */
674 if (flags & RFPROCDESC)
675 procdesc_new(p2, pdflags);
676 #endif
677
678 /*
679 * Both processes are set up, now check if any loadable modules want
680 * to adjust anything.
681 */
682 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
683
684 /*
685 * Set the child start time and mark the process as being complete.
686 */
687 PROC_LOCK(p2);
688 PROC_LOCK(p1);
689 microuptime(&p2->p_stats->p_start);
690 PROC_SLOCK(p2);
691 p2->p_state = PRS_NORMAL;
692 PROC_SUNLOCK(p2);
693
694 #ifdef KDTRACE_HOOKS
695 /*
696 * Tell the DTrace fasttrap provider about the new process so that any
697 * tracepoints inherited from the parent can be removed. We have to do
698 * this only after p_state is PRS_NORMAL since the fasttrap module will
699 * use pfind() later on.
700 */
701 if ((flags & RFMEM) == 0 && dtrace_fasttrap_fork)
702 dtrace_fasttrap_fork(p1, p2);
703 #endif
704 if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED |
705 P_FOLLOWFORK)) {
706 /*
707 * Arrange for debugger to receive the fork event.
708 *
709 * We can report PL_FLAG_FORKED regardless of
710 * P_FOLLOWFORK settings, but it does not make a sense
711 * for runaway child.
712 */
713 td->td_dbgflags |= TDB_FORK;
714 td->td_dbg_forked = p2->p_pid;
715 td2->td_dbgflags |= TDB_STOPATFORK;
716 _PHOLD(p2);
717 p2_held = 1;
718 }
719 if (flags & RFPPWAIT) {
720 td->td_pflags |= TDP_RFPPWAIT;
721 td->td_rfppwait_p = p2;
722 }
723 PROC_UNLOCK(p2);
724 if ((flags & RFSTOPPED) == 0) {
725 /*
726 * If RFSTOPPED not requested, make child runnable and
727 * add to run queue.
728 */
729 thread_lock(td2);
730 TD_SET_CAN_RUN(td2);
731 sched_add(td2, SRQ_BORING);
732 thread_unlock(td2);
733 }
734
735 /*
736 * Now can be swapped.
737 */
738 _PRELE(p1);
739 PROC_UNLOCK(p1);
740
741 /*
742 * Tell any interested parties about the new process.
743 */
744 knote_fork(&p1->p_klist, p2->p_pid);
745 SDT_PROBE3(proc, kernel, , create, p2, p1, flags);
746
747 /*
748 * Wait until debugger is attached to child.
749 */
750 PROC_LOCK(p2);
751 while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
752 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
753 if (p2_held)
754 _PRELE(p2);
755 PROC_UNLOCK(p2);
756 }
757
758 int
759 fork1(struct thread *td, int flags, int pages, struct proc **procp,
760 int *procdescp, int pdflags)
761 {
762 struct proc *p1;
763 struct proc *newproc;
764 int ok;
765 struct thread *td2;
766 struct vmspace *vm2;
767 vm_ooffset_t mem_charged;
768 int error;
769 static int curfail;
770 static struct timeval lastfail;
771 #ifdef PROCDESC
772 struct file *fp_procdesc = NULL;
773 #endif
774
775 /* Check for the undefined or unimplemented flags. */
776 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
777 return (EINVAL);
778
779 /* Signal value requires RFTSIGZMB. */
780 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
781 return (EINVAL);
782
783 /* Can't copy and clear. */
784 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
785 return (EINVAL);
786
787 /* Check the validity of the signal number. */
788 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
789 return (EINVAL);
790
791 #ifdef PROCDESC
792 if ((flags & RFPROCDESC) != 0) {
793 /* Can't not create a process yet get a process descriptor. */
794 if ((flags & RFPROC) == 0)
795 return (EINVAL);
796
797 /* Must provide a place to put a procdesc if creating one. */
798 if (procdescp == NULL)
799 return (EINVAL);
800 }
801 #endif
802
803 p1 = td->td_proc;
804
805 /*
806 * Here we don't create a new process, but we divorce
807 * certain parts of a process from itself.
808 */
809 if ((flags & RFPROC) == 0) {
810 *procp = NULL;
811 return (fork_norfproc(td, flags));
812 }
813
814 #ifdef PROCDESC
815 /*
816 * If required, create a process descriptor in the parent first; we
817 * will abandon it if something goes wrong. We don't finit() until
818 * later.
819 */
820 if (flags & RFPROCDESC) {
821 error = falloc(td, &fp_procdesc, procdescp, 0);
822 if (error != 0)
823 return (error);
824 }
825 #endif
826
827 mem_charged = 0;
828 vm2 = NULL;
829 if (pages == 0)
830 pages = KSTACK_PAGES;
831 /* Allocate new proc. */
832 newproc = uma_zalloc(proc_zone, M_WAITOK);
833 td2 = FIRST_THREAD_IN_PROC(newproc);
834 if (td2 == NULL) {
835 td2 = thread_alloc(pages);
836 if (td2 == NULL) {
837 error = ENOMEM;
838 goto fail1;
839 }
840 proc_linkup(newproc, td2);
841 } else {
842 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
843 if (td2->td_kstack != 0)
844 vm_thread_dispose(td2);
845 if (!thread_alloc_stack(td2, pages)) {
846 error = ENOMEM;
847 goto fail1;
848 }
849 }
850 }
851
852 if ((flags & RFMEM) == 0) {
853 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
854 if (vm2 == NULL) {
855 error = ENOMEM;
856 goto fail1;
857 }
858 if (!swap_reserve(mem_charged)) {
859 /*
860 * The swap reservation failed. The accounting
861 * from the entries of the copied vm2 will be
862 * substracted in vmspace_free(), so force the
863 * reservation there.
864 */
865 swap_reserve_force(mem_charged);
866 error = ENOMEM;
867 goto fail1;
868 }
869 } else
870 vm2 = NULL;
871
872 /*
873 * XXX: This is ugly; when we copy resource usage, we need to bump
874 * per-cred resource counters.
875 */
876 newproc->p_ucred = p1->p_ucred;
877
878 /*
879 * Initialize resource accounting for the child process.
880 */
881 error = racct_proc_fork(p1, newproc);
882 if (error != 0) {
883 error = EAGAIN;
884 goto fail1;
885 }
886
887 #ifdef MAC
888 mac_proc_init(newproc);
889 #endif
890 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
891 STAILQ_INIT(&newproc->p_ktr);
892
893 /* We have to lock the process tree while we look for a pid. */
894 sx_slock(&proctree_lock);
895
896 /*
897 * Although process entries are dynamically created, we still keep
898 * a global limit on the maximum number we will create. Don't allow
899 * a nonprivileged user to use the last ten processes; don't let root
900 * exceed the limit. The variable nprocs is the current number of
901 * processes, maxproc is the limit.
902 */
903 sx_xlock(&allproc_lock);
904 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
905 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
906 error = EAGAIN;
907 goto fail;
908 }
909
910 /*
911 * Increment the count of procs running with this uid. Don't allow
912 * a nonprivileged user to exceed their current limit.
913 *
914 * XXXRW: Can we avoid privilege here if it's not needed?
915 */
916 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
917 if (error == 0)
918 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
919 else {
920 PROC_LOCK(p1);
921 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
922 lim_cur(p1, RLIMIT_NPROC));
923 PROC_UNLOCK(p1);
924 }
925 if (ok) {
926 do_fork(td, flags, newproc, td2, vm2, pdflags);
927
928 /*
929 * Return child proc pointer to parent.
930 */
931 *procp = newproc;
932 #ifdef PROCDESC
933 if (flags & RFPROCDESC)
934 procdesc_finit(newproc->p_procdesc, fp_procdesc);
935 #endif
936 racct_proc_fork_done(newproc);
937 return (0);
938 }
939
940 error = EAGAIN;
941 fail:
942 sx_sunlock(&proctree_lock);
943 if (ppsratecheck(&lastfail, &curfail, 1))
944 printf("maxproc limit exceeded by uid %u (pid %d); see tuning(7) and login.conf(5)\n",
945 td->td_ucred->cr_ruid, p1->p_pid);
946 sx_xunlock(&allproc_lock);
947 #ifdef MAC
948 mac_proc_destroy(newproc);
949 #endif
950 racct_proc_exit(newproc);
951 fail1:
952 if (vm2 != NULL)
953 vmspace_free(vm2);
954 uma_zfree(proc_zone, newproc);
955 #ifdef PROCDESC
956 if (((flags & RFPROCDESC) != 0) && (fp_procdesc != NULL))
957 fdrop(fp_procdesc, td);
958 #endif
959 pause("fork", hz / 2);
960 return (error);
961 }
962
963 /*
964 * Handle the return of a child process from fork1(). This function
965 * is called from the MD fork_trampoline() entry point.
966 */
967 void
968 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
969 struct trapframe *frame)
970 {
971 struct proc *p;
972 struct thread *td;
973 struct thread *dtd;
974
975 td = curthread;
976 p = td->td_proc;
977 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
978
979 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
980 td, td->td_sched, p->p_pid, td->td_name);
981
982 sched_fork_exit(td);
983 /*
984 * Processes normally resume in mi_switch() after being
985 * cpu_switch()'ed to, but when children start up they arrive here
986 * instead, so we must do much the same things as mi_switch() would.
987 */
988 if ((dtd = PCPU_GET(deadthread))) {
989 PCPU_SET(deadthread, NULL);
990 thread_stash(dtd);
991 }
992 thread_unlock(td);
993
994 /*
995 * cpu_set_fork_handler intercepts this function call to
996 * have this call a non-return function to stay in kernel mode.
997 * initproc has its own fork handler, but it does return.
998 */
999 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1000 callout(arg, frame);
1001
1002 /*
1003 * Check if a kernel thread misbehaved and returned from its main
1004 * function.
1005 */
1006 if (p->p_flag & P_KTHREAD) {
1007 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1008 td->td_name, p->p_pid);
1009 kthread_exit();
1010 }
1011 mtx_assert(&Giant, MA_NOTOWNED);
1012
1013 if (p->p_sysent->sv_schedtail != NULL)
1014 (p->p_sysent->sv_schedtail)(td);
1015 }
1016
1017 /*
1018 * Simplified back end of syscall(), used when returning from fork()
1019 * directly into user mode. Giant is not held on entry, and must not
1020 * be held on return. This function is passed in to fork_exit() as the
1021 * first parameter and is called when returning to a new userland process.
1022 */
1023 void
1024 fork_return(struct thread *td, struct trapframe *frame)
1025 {
1026 struct proc *p, *dbg;
1027
1028 p = td->td_proc;
1029 if (td->td_dbgflags & TDB_STOPATFORK) {
1030 sx_xlock(&proctree_lock);
1031 PROC_LOCK(p);
1032 if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) ==
1033 (P_TRACED | P_FOLLOWFORK)) {
1034 /*
1035 * If debugger still wants auto-attach for the
1036 * parent's children, do it now.
1037 */
1038 dbg = p->p_pptr->p_pptr;
1039 p->p_flag |= P_TRACED;
1040 p->p_oppid = p->p_pptr->p_pid;
1041 CTR2(KTR_PTRACE,
1042 "fork_return: attaching to new child pid %d: oppid %d",
1043 p->p_pid, p->p_oppid);
1044 proc_reparent(p, dbg);
1045 sx_xunlock(&proctree_lock);
1046 td->td_dbgflags |= TDB_CHILD | TDB_SCX;
1047 ptracestop(td, SIGSTOP);
1048 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1049 } else {
1050 /*
1051 * ... otherwise clear the request.
1052 */
1053 sx_xunlock(&proctree_lock);
1054 td->td_dbgflags &= ~TDB_STOPATFORK;
1055 cv_broadcast(&p->p_dbgwait);
1056 }
1057 PROC_UNLOCK(p);
1058 } else if (p->p_flag & P_TRACED) {
1059 /*
1060 * This is the start of a new thread in a traced
1061 * process. Report a system call exit event.
1062 */
1063 PROC_LOCK(p);
1064 td->td_dbgflags |= TDB_SCX;
1065 _STOPEVENT(p, S_SCX, td->td_dbg_sc_code);
1066 if ((p->p_stops & S_PT_SCX) != 0)
1067 ptracestop(td, SIGTRAP);
1068 td->td_dbgflags &= ~TDB_SCX;
1069 PROC_UNLOCK(p);
1070 }
1071
1072 userret(td, frame);
1073
1074 #ifdef KTRACE
1075 if (KTRPOINT(td, KTR_SYSRET))
1076 ktrsysret(SYS_fork, 0, 0);
1077 #endif
1078 mtx_assert(&Giant, MA_NOTOWNED);
1079 }
Cache object: c5174f55c53b29f20dddf91a3a13d661
|