FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_fork.c
1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
37 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD: releng/12.0/sys/kern/kern_fork.c 335504 2018-06-21 21:12:49Z kib $");
41
42 #include "opt_ktrace.h"
43 #include "opt_kstack_pages.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, , , create, "struct proc *", "struct proc *", "int");
94
95 #ifndef _SYS_SYSPROTO_H_
96 struct fork_args {
97 int dummy;
98 };
99 #endif
100
101 EVENTHANDLER_LIST_DECLARE(process_fork);
102
103 /* ARGSUSED */
104 int
105 sys_fork(struct thread *td, struct fork_args *uap)
106 {
107 struct fork_req fr;
108 int error, pid;
109
110 bzero(&fr, sizeof(fr));
111 fr.fr_flags = RFFDG | RFPROC;
112 fr.fr_pidp = &pid;
113 error = fork1(td, &fr);
114 if (error == 0) {
115 td->td_retval[0] = pid;
116 td->td_retval[1] = 0;
117 }
118 return (error);
119 }
120
121 /* ARGUSED */
122 int
123 sys_pdfork(struct thread *td, struct pdfork_args *uap)
124 {
125 struct fork_req fr;
126 int error, fd, pid;
127
128 bzero(&fr, sizeof(fr));
129 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
130 fr.fr_pidp = &pid;
131 fr.fr_pd_fd = &fd;
132 fr.fr_pd_flags = uap->flags;
133 /*
134 * It is necessary to return fd by reference because 0 is a valid file
135 * descriptor number, and the child needs to be able to distinguish
136 * itself from the parent using the return value.
137 */
138 error = fork1(td, &fr);
139 if (error == 0) {
140 td->td_retval[0] = pid;
141 td->td_retval[1] = 0;
142 error = copyout(&fd, uap->fdp, sizeof(fd));
143 }
144 return (error);
145 }
146
147 /* ARGSUSED */
148 int
149 sys_vfork(struct thread *td, struct vfork_args *uap)
150 {
151 struct fork_req fr;
152 int error, pid;
153
154 bzero(&fr, sizeof(fr));
155 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
156 fr.fr_pidp = &pid;
157 error = fork1(td, &fr);
158 if (error == 0) {
159 td->td_retval[0] = 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 fork_req fr;
169 int error, pid;
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 bzero(&fr, sizeof(fr));
177 fr.fr_flags = uap->flags;
178 fr.fr_pidp = &pid;
179 error = fork1(td, &fr);
180 if (error == 0) {
181 td->td_retval[0] = pid;
182 td->td_retval[1] = 0;
183 }
184 return (error);
185 }
186
187 int nprocs = 1; /* process 0 */
188 int lastpid = 0;
189 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
190 "Last used PID");
191
192 /*
193 * Random component to lastpid generation. We mix in a random factor to make
194 * it a little harder to predict. We sanity check the modulus value to avoid
195 * doing it in critical paths. Don't let it be too small or we pointlessly
196 * waste randomness entropy, and don't let it be impossibly large. Using a
197 * modulus that is too big causes a LOT more process table scans and slows
198 * down fork processing as the pidchecked caching is defeated.
199 */
200 static int randompid = 0;
201
202 static int
203 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
204 {
205 int error, pid;
206
207 error = sysctl_wire_old_buffer(req, sizeof(int));
208 if (error != 0)
209 return(error);
210 sx_xlock(&allproc_lock);
211 pid = randompid;
212 error = sysctl_handle_int(oidp, &pid, 0, req);
213 if (error == 0 && req->newptr != NULL) {
214 if (pid == 0)
215 randompid = 0;
216 else if (pid == 1)
217 /* generate a random PID modulus between 100 and 1123 */
218 randompid = 100 + arc4random() % 1024;
219 else if (pid < 0 || pid > pid_max - 100)
220 /* out of range */
221 randompid = pid_max - 100;
222 else if (pid < 100)
223 /* Make it reasonable */
224 randompid = 100;
225 else
226 randompid = pid;
227 }
228 sx_xunlock(&allproc_lock);
229 return (error);
230 }
231
232 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
233 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value");
234
235 static int
236 fork_findpid(int flags)
237 {
238 struct proc *p;
239 int trypid;
240 static int pidchecked = 0;
241
242 /*
243 * Requires allproc_lock in order to iterate over the list
244 * of processes, and proctree_lock to access p_pgrp.
245 */
246 sx_assert(&allproc_lock, SX_LOCKED);
247 sx_assert(&proctree_lock, SX_LOCKED);
248
249 /*
250 * Find an unused process ID. We remember a range of unused IDs
251 * ready to use (from lastpid+1 through pidchecked-1).
252 *
253 * If RFHIGHPID is set (used during system boot), do not allocate
254 * low-numbered pids.
255 */
256 trypid = lastpid + 1;
257 if (flags & RFHIGHPID) {
258 if (trypid < 10)
259 trypid = 10;
260 } else {
261 if (randompid)
262 trypid += arc4random() % randompid;
263 }
264 retry:
265 /*
266 * If the process ID prototype has wrapped around,
267 * restart somewhat above 0, as the low-numbered procs
268 * tend to include daemons that don't exit.
269 */
270 if (trypid >= pid_max) {
271 trypid = trypid % pid_max;
272 if (trypid < 100)
273 trypid += 100;
274 pidchecked = 0;
275 }
276 if (trypid >= pidchecked) {
277 int doingzomb = 0;
278
279 pidchecked = PID_MAX;
280 /*
281 * Scan the active and zombie procs to check whether this pid
282 * is in use. Remember the lowest pid that's greater
283 * than trypid, so we can avoid checking for a while.
284 *
285 * Avoid reuse of the process group id, session id or
286 * the reaper subtree id. Note that for process group
287 * and sessions, the amount of reserved pids is
288 * limited by process limit. For the subtree ids, the
289 * id is kept reserved only while there is a
290 * non-reaped process in the subtree, so amount of
291 * reserved pids is limited by process limit times
292 * two.
293 */
294 p = LIST_FIRST(&allproc);
295 again:
296 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
297 while (p->p_pid == trypid ||
298 p->p_reapsubtree == trypid ||
299 (p->p_pgrp != NULL &&
300 (p->p_pgrp->pg_id == trypid ||
301 (p->p_session != NULL &&
302 p->p_session->s_sid == trypid)))) {
303 trypid++;
304 if (trypid >= pidchecked)
305 goto retry;
306 }
307 if (p->p_pid > trypid && pidchecked > p->p_pid)
308 pidchecked = p->p_pid;
309 if (p->p_pgrp != NULL) {
310 if (p->p_pgrp->pg_id > trypid &&
311 pidchecked > p->p_pgrp->pg_id)
312 pidchecked = p->p_pgrp->pg_id;
313 if (p->p_session != NULL &&
314 p->p_session->s_sid > trypid &&
315 pidchecked > p->p_session->s_sid)
316 pidchecked = p->p_session->s_sid;
317 }
318 }
319 if (!doingzomb) {
320 doingzomb = 1;
321 p = LIST_FIRST(&zombproc);
322 goto again;
323 }
324 }
325
326 /*
327 * RFHIGHPID does not mess with the lastpid counter during boot.
328 */
329 if (flags & RFHIGHPID)
330 pidchecked = 0;
331 else
332 lastpid = trypid;
333
334 return (trypid);
335 }
336
337 static int
338 fork_norfproc(struct thread *td, int flags)
339 {
340 int error;
341 struct proc *p1;
342
343 KASSERT((flags & RFPROC) == 0,
344 ("fork_norfproc called with RFPROC set"));
345 p1 = td->td_proc;
346
347 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
348 (flags & (RFCFDG | RFFDG))) {
349 PROC_LOCK(p1);
350 if (thread_single(p1, SINGLE_BOUNDARY)) {
351 PROC_UNLOCK(p1);
352 return (ERESTART);
353 }
354 PROC_UNLOCK(p1);
355 }
356
357 error = vm_forkproc(td, NULL, NULL, NULL, flags);
358 if (error)
359 goto fail;
360
361 /*
362 * Close all file descriptors.
363 */
364 if (flags & RFCFDG) {
365 struct filedesc *fdtmp;
366 fdtmp = fdinit(td->td_proc->p_fd, false);
367 fdescfree(td);
368 p1->p_fd = fdtmp;
369 }
370
371 /*
372 * Unshare file descriptors (from parent).
373 */
374 if (flags & RFFDG)
375 fdunshare(td);
376
377 fail:
378 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
379 (flags & (RFCFDG | RFFDG))) {
380 PROC_LOCK(p1);
381 thread_single_end(p1, SINGLE_BOUNDARY);
382 PROC_UNLOCK(p1);
383 }
384 return (error);
385 }
386
387 static void
388 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
389 struct vmspace *vm2, struct file *fp_procdesc)
390 {
391 struct proc *p1, *pptr;
392 int trypid;
393 struct filedesc *fd;
394 struct filedesc_to_leader *fdtol;
395 struct sigacts *newsigacts;
396
397 sx_assert(&proctree_lock, SX_LOCKED);
398 sx_assert(&allproc_lock, SX_XLOCKED);
399
400 p1 = td->td_proc;
401
402 trypid = fork_findpid(fr->fr_flags);
403
404 p2->p_state = PRS_NEW; /* protect against others */
405 p2->p_pid = trypid;
406 AUDIT_ARG_PID(p2->p_pid);
407 LIST_INSERT_HEAD(&allproc, p2, p_list);
408 allproc_gen++;
409 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
410 PROC_LOCK(p2);
411 PROC_LOCK(p1);
412
413 sx_xunlock(&allproc_lock);
414 sx_xunlock(&proctree_lock);
415
416 bcopy(&p1->p_startcopy, &p2->p_startcopy,
417 __rangeof(struct proc, p_startcopy, p_endcopy));
418 pargs_hold(p2->p_args);
419
420 PROC_UNLOCK(p1);
421
422 bzero(&p2->p_startzero,
423 __rangeof(struct proc, p_startzero, p_endzero));
424
425 /* Tell the prison that we exist. */
426 prison_proc_hold(p2->p_ucred->cr_prison);
427
428 PROC_UNLOCK(p2);
429
430 tidhash_add(td2);
431
432 /*
433 * Malloc things while we don't hold any locks.
434 */
435 if (fr->fr_flags & RFSIGSHARE)
436 newsigacts = NULL;
437 else
438 newsigacts = sigacts_alloc();
439
440 /*
441 * Copy filedesc.
442 */
443 if (fr->fr_flags & RFCFDG) {
444 fd = fdinit(p1->p_fd, false);
445 fdtol = NULL;
446 } else if (fr->fr_flags & RFFDG) {
447 fd = fdcopy(p1->p_fd);
448 fdtol = NULL;
449 } else {
450 fd = fdshare(p1->p_fd);
451 if (p1->p_fdtol == NULL)
452 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
453 p1->p_leader);
454 if ((fr->fr_flags & RFTHREAD) != 0) {
455 /*
456 * Shared file descriptor table, and shared
457 * process leaders.
458 */
459 fdtol = p1->p_fdtol;
460 FILEDESC_XLOCK(p1->p_fd);
461 fdtol->fdl_refcount++;
462 FILEDESC_XUNLOCK(p1->p_fd);
463 } else {
464 /*
465 * Shared file descriptor table, and different
466 * process leaders.
467 */
468 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
469 p1->p_fd, p2);
470 }
471 }
472 /*
473 * Make a proc table entry for the new process.
474 * Start by zeroing the section of proc that is zero-initialized,
475 * then copy the section that is copied directly from the parent.
476 */
477
478 PROC_LOCK(p2);
479 PROC_LOCK(p1);
480
481 bzero(&td2->td_startzero,
482 __rangeof(struct thread, td_startzero, td_endzero));
483
484 bcopy(&td->td_startcopy, &td2->td_startcopy,
485 __rangeof(struct thread, td_startcopy, td_endcopy));
486
487 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
488 td2->td_sigstk = td->td_sigstk;
489 td2->td_flags = TDF_INMEM;
490 td2->td_lend_user_pri = PRI_MAX;
491
492 #ifdef VIMAGE
493 td2->td_vnet = NULL;
494 td2->td_vnet_lpush = NULL;
495 #endif
496
497 /*
498 * Allow the scheduler to initialize the child.
499 */
500 thread_lock(td);
501 sched_fork(td, td2);
502 thread_unlock(td);
503
504 /*
505 * Duplicate sub-structures as needed.
506 * Increase reference counts on shared objects.
507 */
508 p2->p_flag = P_INMEM;
509 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
510 p2->p_swtick = ticks;
511 if (p1->p_flag & P_PROFIL)
512 startprofclock(p2);
513
514 if (fr->fr_flags & RFSIGSHARE) {
515 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
516 } else {
517 sigacts_copy(newsigacts, p1->p_sigacts);
518 p2->p_sigacts = newsigacts;
519 }
520
521 if (fr->fr_flags & RFTSIGZMB)
522 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
523 else if (fr->fr_flags & RFLINUXTHPN)
524 p2->p_sigparent = SIGUSR1;
525 else
526 p2->p_sigparent = SIGCHLD;
527
528 p2->p_textvp = p1->p_textvp;
529 p2->p_fd = fd;
530 p2->p_fdtol = fdtol;
531
532 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
533 p2->p_flag |= P_PROTECTED;
534 p2->p_flag2 |= P2_INHERIT_PROTECTED;
535 }
536
537 /*
538 * p_limit is copy-on-write. Bump its refcount.
539 */
540 lim_fork(p1, p2);
541
542 thread_cow_get_proc(td2, p2);
543
544 pstats_fork(p1->p_stats, p2->p_stats);
545
546 PROC_UNLOCK(p1);
547 PROC_UNLOCK(p2);
548
549 /* Bump references to the text vnode (for procfs). */
550 if (p2->p_textvp)
551 vrefact(p2->p_textvp);
552
553 /*
554 * Set up linkage for kernel based threading.
555 */
556 if ((fr->fr_flags & RFTHREAD) != 0) {
557 mtx_lock(&ppeers_lock);
558 p2->p_peers = p1->p_peers;
559 p1->p_peers = p2;
560 p2->p_leader = p1->p_leader;
561 mtx_unlock(&ppeers_lock);
562 PROC_LOCK(p1->p_leader);
563 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
564 PROC_UNLOCK(p1->p_leader);
565 /*
566 * The task leader is exiting, so process p1 is
567 * going to be killed shortly. Since p1 obviously
568 * isn't dead yet, we know that the leader is either
569 * sending SIGKILL's to all the processes in this
570 * task or is sleeping waiting for all the peers to
571 * exit. We let p1 complete the fork, but we need
572 * to go ahead and kill the new process p2 since
573 * the task leader may not get a chance to send
574 * SIGKILL to it. We leave it on the list so that
575 * the task leader will wait for this new process
576 * to commit suicide.
577 */
578 PROC_LOCK(p2);
579 kern_psignal(p2, SIGKILL);
580 PROC_UNLOCK(p2);
581 } else
582 PROC_UNLOCK(p1->p_leader);
583 } else {
584 p2->p_peers = NULL;
585 p2->p_leader = p2;
586 }
587
588 sx_xlock(&proctree_lock);
589 PGRP_LOCK(p1->p_pgrp);
590 PROC_LOCK(p2);
591 PROC_LOCK(p1);
592
593 /*
594 * Preserve some more flags in subprocess. P_PROFIL has already
595 * been preserved.
596 */
597 p2->p_flag |= p1->p_flag & P_SUGID;
598 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
599 SESS_LOCK(p1->p_session);
600 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
601 p2->p_flag |= P_CONTROLT;
602 SESS_UNLOCK(p1->p_session);
603 if (fr->fr_flags & RFPPWAIT)
604 p2->p_flag |= P_PPWAIT;
605
606 p2->p_pgrp = p1->p_pgrp;
607 LIST_INSERT_AFTER(p1, p2, p_pglist);
608 PGRP_UNLOCK(p1->p_pgrp);
609 LIST_INIT(&p2->p_children);
610 LIST_INIT(&p2->p_orphans);
611
612 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
613
614 /*
615 * If PF_FORK is set, the child process inherits the
616 * procfs ioctl flags from its parent.
617 */
618 if (p1->p_pfsflags & PF_FORK) {
619 p2->p_stops = p1->p_stops;
620 p2->p_pfsflags = p1->p_pfsflags;
621 }
622
623 /*
624 * This begins the section where we must prevent the parent
625 * from being swapped.
626 */
627 _PHOLD(p1);
628 PROC_UNLOCK(p1);
629
630 /*
631 * Attach the new process to its parent.
632 *
633 * If RFNOWAIT is set, the newly created process becomes a child
634 * of init. This effectively disassociates the child from the
635 * parent.
636 */
637 if ((fr->fr_flags & RFNOWAIT) != 0) {
638 pptr = p1->p_reaper;
639 p2->p_reaper = pptr;
640 } else {
641 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
642 p1 : p1->p_reaper;
643 pptr = p1;
644 }
645 p2->p_pptr = pptr;
646 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
647 LIST_INIT(&p2->p_reaplist);
648 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
649 if (p2->p_reaper == p1)
650 p2->p_reapsubtree = p2->p_pid;
651 sx_xunlock(&proctree_lock);
652
653 /* Inform accounting that we have forked. */
654 p2->p_acflag = AFORK;
655 PROC_UNLOCK(p2);
656
657 #ifdef KTRACE
658 ktrprocfork(p1, p2);
659 #endif
660
661 /*
662 * Finish creating the child process. It will return via a different
663 * execution path later. (ie: directly into user mode)
664 */
665 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
666
667 if (fr->fr_flags == (RFFDG | RFPROC)) {
668 VM_CNT_INC(v_forks);
669 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
670 p2->p_vmspace->vm_ssize);
671 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
672 VM_CNT_INC(v_vforks);
673 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
674 p2->p_vmspace->vm_ssize);
675 } else if (p1 == &proc0) {
676 VM_CNT_INC(v_kthreads);
677 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
678 p2->p_vmspace->vm_ssize);
679 } else {
680 VM_CNT_INC(v_rforks);
681 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
682 p2->p_vmspace->vm_ssize);
683 }
684
685 /*
686 * Associate the process descriptor with the process before anything
687 * can happen that might cause that process to need the descriptor.
688 * However, don't do this until after fork(2) can no longer fail.
689 */
690 if (fr->fr_flags & RFPROCDESC)
691 procdesc_new(p2, fr->fr_pd_flags);
692
693 /*
694 * Both processes are set up, now check if any loadable modules want
695 * to adjust anything.
696 */
697 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
698
699 /*
700 * Set the child start time and mark the process as being complete.
701 */
702 PROC_LOCK(p2);
703 PROC_LOCK(p1);
704 microuptime(&p2->p_stats->p_start);
705 PROC_SLOCK(p2);
706 p2->p_state = PRS_NORMAL;
707 PROC_SUNLOCK(p2);
708
709 #ifdef KDTRACE_HOOKS
710 /*
711 * Tell the DTrace fasttrap provider about the new process so that any
712 * tracepoints inherited from the parent can be removed. We have to do
713 * this only after p_state is PRS_NORMAL since the fasttrap module will
714 * use pfind() later on.
715 */
716 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
717 dtrace_fasttrap_fork(p1, p2);
718 #endif
719 /*
720 * Hold the process so that it cannot exit after we make it runnable,
721 * but before we wait for the debugger.
722 */
723 _PHOLD(p2);
724 if (fr->fr_flags & RFPPWAIT) {
725 td->td_pflags |= TDP_RFPPWAIT;
726 td->td_rfppwait_p = p2;
727 td->td_dbgflags |= TDB_VFORK;
728 }
729 PROC_UNLOCK(p2);
730
731 /*
732 * Now can be swapped.
733 */
734 _PRELE(p1);
735 PROC_UNLOCK(p1);
736
737 /*
738 * Tell any interested parties about the new process.
739 */
740 knote_fork(p1->p_klist, p2->p_pid);
741 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
742
743 if (fr->fr_flags & RFPROCDESC) {
744 procdesc_finit(p2->p_procdesc, fp_procdesc);
745 fdrop(fp_procdesc, td);
746 }
747
748 /*
749 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
750 * synced with forks in progress so it is OK if we miss it
751 * if being set atm.
752 */
753 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
754 sx_xlock(&proctree_lock);
755 PROC_LOCK(p2);
756
757 /*
758 * p1->p_ptevents & p1->p_pptr are protected by both
759 * process and proctree locks for modifications,
760 * so owning proctree_lock allows the race-free read.
761 */
762 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
763 /*
764 * Arrange for debugger to receive the fork event.
765 *
766 * We can report PL_FLAG_FORKED regardless of
767 * P_FOLLOWFORK settings, but it does not make a sense
768 * for runaway child.
769 */
770 td->td_dbgflags |= TDB_FORK;
771 td->td_dbg_forked = p2->p_pid;
772 td2->td_dbgflags |= TDB_STOPATFORK;
773 proc_set_traced(p2, true);
774 CTR2(KTR_PTRACE,
775 "do_fork: attaching to new child pid %d: oppid %d",
776 p2->p_pid, p2->p_oppid);
777 proc_reparent(p2, p1->p_pptr);
778 }
779 PROC_UNLOCK(p2);
780 sx_xunlock(&proctree_lock);
781 }
782
783 if ((fr->fr_flags & RFSTOPPED) == 0) {
784 /*
785 * If RFSTOPPED not requested, make child runnable and
786 * add to run queue.
787 */
788 thread_lock(td2);
789 TD_SET_CAN_RUN(td2);
790 sched_add(td2, SRQ_BORING);
791 thread_unlock(td2);
792 if (fr->fr_pidp != NULL)
793 *fr->fr_pidp = p2->p_pid;
794 } else {
795 *fr->fr_procp = p2;
796 }
797
798 PROC_LOCK(p2);
799 _PRELE(p2);
800 racct_proc_fork_done(p2);
801 PROC_UNLOCK(p2);
802 }
803
804 int
805 fork1(struct thread *td, struct fork_req *fr)
806 {
807 struct proc *p1, *newproc;
808 struct thread *td2;
809 struct vmspace *vm2;
810 struct file *fp_procdesc;
811 vm_ooffset_t mem_charged;
812 int error, nprocs_new, ok;
813 static int curfail;
814 static struct timeval lastfail;
815 int flags, pages;
816
817 flags = fr->fr_flags;
818 pages = fr->fr_pages;
819
820 if ((flags & RFSTOPPED) != 0)
821 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
822 else
823 MPASS(fr->fr_procp == NULL);
824
825 /* Check for the undefined or unimplemented flags. */
826 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
827 return (EINVAL);
828
829 /* Signal value requires RFTSIGZMB. */
830 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
831 return (EINVAL);
832
833 /* Can't copy and clear. */
834 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
835 return (EINVAL);
836
837 /* Check the validity of the signal number. */
838 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
839 return (EINVAL);
840
841 if ((flags & RFPROCDESC) != 0) {
842 /* Can't not create a process yet get a process descriptor. */
843 if ((flags & RFPROC) == 0)
844 return (EINVAL);
845
846 /* Must provide a place to put a procdesc if creating one. */
847 if (fr->fr_pd_fd == NULL)
848 return (EINVAL);
849
850 /* Check if we are using supported flags. */
851 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
852 return (EINVAL);
853 }
854
855 p1 = td->td_proc;
856
857 /*
858 * Here we don't create a new process, but we divorce
859 * certain parts of a process from itself.
860 */
861 if ((flags & RFPROC) == 0) {
862 if (fr->fr_procp != NULL)
863 *fr->fr_procp = NULL;
864 else if (fr->fr_pidp != NULL)
865 *fr->fr_pidp = 0;
866 return (fork_norfproc(td, flags));
867 }
868
869 fp_procdesc = NULL;
870 newproc = NULL;
871 vm2 = NULL;
872
873 /*
874 * Increment the nprocs resource before allocations occur.
875 * Although process entries are dynamically created, we still
876 * keep a global limit on the maximum number we will
877 * create. There are hard-limits as to the number of processes
878 * that can run, established by the KVA and memory usage for
879 * the process data.
880 *
881 * Don't allow a nonprivileged user to use the last ten
882 * processes; don't let root exceed the limit.
883 */
884 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
885 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
886 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
887 error = EAGAIN;
888 sx_xlock(&allproc_lock);
889 if (ppsratecheck(&lastfail, &curfail, 1)) {
890 printf("maxproc limit exceeded by uid %u (pid %d); "
891 "see tuning(7) and login.conf(5)\n",
892 td->td_ucred->cr_ruid, p1->p_pid);
893 }
894 sx_xunlock(&allproc_lock);
895 goto fail2;
896 }
897
898 /*
899 * If required, create a process descriptor in the parent first; we
900 * will abandon it if something goes wrong. We don't finit() until
901 * later.
902 */
903 if (flags & RFPROCDESC) {
904 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
905 fr->fr_pd_flags, fr->fr_pd_fcaps);
906 if (error != 0)
907 goto fail2;
908 }
909
910 mem_charged = 0;
911 if (pages == 0)
912 pages = kstack_pages;
913 /* Allocate new proc. */
914 newproc = uma_zalloc(proc_zone, M_WAITOK);
915 td2 = FIRST_THREAD_IN_PROC(newproc);
916 if (td2 == NULL) {
917 td2 = thread_alloc(pages);
918 if (td2 == NULL) {
919 error = ENOMEM;
920 goto fail2;
921 }
922 proc_linkup(newproc, td2);
923 } else {
924 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
925 if (td2->td_kstack != 0)
926 vm_thread_dispose(td2);
927 if (!thread_alloc_stack(td2, pages)) {
928 error = ENOMEM;
929 goto fail2;
930 }
931 }
932 }
933
934 if ((flags & RFMEM) == 0) {
935 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
936 if (vm2 == NULL) {
937 error = ENOMEM;
938 goto fail2;
939 }
940 if (!swap_reserve(mem_charged)) {
941 /*
942 * The swap reservation failed. The accounting
943 * from the entries of the copied vm2 will be
944 * subtracted in vmspace_free(), so force the
945 * reservation there.
946 */
947 swap_reserve_force(mem_charged);
948 error = ENOMEM;
949 goto fail2;
950 }
951 } else
952 vm2 = NULL;
953
954 /*
955 * XXX: This is ugly; when we copy resource usage, we need to bump
956 * per-cred resource counters.
957 */
958 proc_set_cred_init(newproc, crhold(td->td_ucred));
959
960 /*
961 * Initialize resource accounting for the child process.
962 */
963 error = racct_proc_fork(p1, newproc);
964 if (error != 0) {
965 error = EAGAIN;
966 goto fail1;
967 }
968
969 #ifdef MAC
970 mac_proc_init(newproc);
971 #endif
972 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
973 STAILQ_INIT(&newproc->p_ktr);
974
975 /* We have to lock the process tree while we look for a pid. */
976 sx_xlock(&proctree_lock);
977 sx_xlock(&allproc_lock);
978
979 /*
980 * Increment the count of procs running with this uid. Don't allow
981 * a nonprivileged user to exceed their current limit.
982 *
983 * XXXRW: Can we avoid privilege here if it's not needed?
984 */
985 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
986 if (error == 0)
987 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
988 else {
989 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
990 lim_cur(td, RLIMIT_NPROC));
991 }
992 if (ok) {
993 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
994 return (0);
995 }
996
997 error = EAGAIN;
998 sx_xunlock(&allproc_lock);
999 sx_xunlock(&proctree_lock);
1000 #ifdef MAC
1001 mac_proc_destroy(newproc);
1002 #endif
1003 racct_proc_exit(newproc);
1004 fail1:
1005 crfree(newproc->p_ucred);
1006 newproc->p_ucred = NULL;
1007 fail2:
1008 if (vm2 != NULL)
1009 vmspace_free(vm2);
1010 uma_zfree(proc_zone, newproc);
1011 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1012 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1013 fdrop(fp_procdesc, td);
1014 }
1015 atomic_add_int(&nprocs, -1);
1016 pause("fork", hz / 2);
1017 return (error);
1018 }
1019
1020 /*
1021 * Handle the return of a child process from fork1(). This function
1022 * is called from the MD fork_trampoline() entry point.
1023 */
1024 void
1025 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1026 struct trapframe *frame)
1027 {
1028 struct proc *p;
1029 struct thread *td;
1030 struct thread *dtd;
1031
1032 td = curthread;
1033 p = td->td_proc;
1034 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1035
1036 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1037 td, td_get_sched(td), p->p_pid, td->td_name);
1038
1039 sched_fork_exit(td);
1040 /*
1041 * Processes normally resume in mi_switch() after being
1042 * cpu_switch()'ed to, but when children start up they arrive here
1043 * instead, so we must do much the same things as mi_switch() would.
1044 */
1045 if ((dtd = PCPU_GET(deadthread))) {
1046 PCPU_SET(deadthread, NULL);
1047 thread_stash(dtd);
1048 }
1049 thread_unlock(td);
1050
1051 /*
1052 * cpu_fork_kthread_handler intercepts this function call to
1053 * have this call a non-return function to stay in kernel mode.
1054 * initproc has its own fork handler, but it does return.
1055 */
1056 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1057 callout(arg, frame);
1058
1059 /*
1060 * Check if a kernel thread misbehaved and returned from its main
1061 * function.
1062 */
1063 if (p->p_flag & P_KPROC) {
1064 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1065 td->td_name, p->p_pid);
1066 kthread_exit();
1067 }
1068 mtx_assert(&Giant, MA_NOTOWNED);
1069
1070 if (p->p_sysent->sv_schedtail != NULL)
1071 (p->p_sysent->sv_schedtail)(td);
1072 td->td_pflags &= ~TDP_FORKING;
1073 }
1074
1075 /*
1076 * Simplified back end of syscall(), used when returning from fork()
1077 * directly into user mode. This function is passed in to fork_exit()
1078 * as the first parameter and is called when returning to a new
1079 * userland process.
1080 */
1081 void
1082 fork_return(struct thread *td, struct trapframe *frame)
1083 {
1084 struct proc *p;
1085
1086 p = td->td_proc;
1087 if (td->td_dbgflags & TDB_STOPATFORK) {
1088 PROC_LOCK(p);
1089 if ((p->p_flag & P_TRACED) != 0) {
1090 /*
1091 * Inform the debugger if one is still present.
1092 */
1093 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1094 ptracestop(td, SIGSTOP, NULL);
1095 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1096 } else {
1097 /*
1098 * ... otherwise clear the request.
1099 */
1100 td->td_dbgflags &= ~TDB_STOPATFORK;
1101 }
1102 PROC_UNLOCK(p);
1103 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1104 /*
1105 * This is the start of a new thread in a traced
1106 * process. Report a system call exit event.
1107 */
1108 PROC_LOCK(p);
1109 td->td_dbgflags |= TDB_SCX;
1110 _STOPEVENT(p, S_SCX, td->td_sa.code);
1111 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1112 (td->td_dbgflags & TDB_BORN) != 0)
1113 ptracestop(td, SIGTRAP, NULL);
1114 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1115 PROC_UNLOCK(p);
1116 }
1117
1118 userret(td, frame);
1119
1120 #ifdef KTRACE
1121 if (KTRPOINT(td, KTR_SYSRET))
1122 ktrsysret(SYS_fork, 0, 0);
1123 #endif
1124 }
Cache object: 7ab95262dba275bdc5c763970fa50c1d
|