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("$FreeBSD$");
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/bitstring.h>
48 #include <sys/sysproto.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
52 #include <sys/jail.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/sysctl.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mutex.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/procdesc.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 /* ARGSUSED */
102 int
103 sys_fork(struct thread *td, struct fork_args *uap)
104 {
105 struct fork_req fr;
106 int error, pid;
107
108 bzero(&fr, sizeof(fr));
109 fr.fr_flags = RFFDG | RFPROC;
110 fr.fr_pidp = &pid;
111 error = fork1(td, &fr);
112 if (error == 0) {
113 td->td_retval[0] = pid;
114 td->td_retval[1] = 0;
115 }
116 return (error);
117 }
118
119 /* ARGUSED */
120 int
121 sys_pdfork(struct thread *td, struct pdfork_args *uap)
122 {
123 struct fork_req fr;
124 int error, fd, pid;
125
126 bzero(&fr, sizeof(fr));
127 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
128 fr.fr_pidp = &pid;
129 fr.fr_pd_fd = &fd;
130 fr.fr_pd_flags = uap->flags;
131 AUDIT_ARG_FFLAGS(uap->flags);
132 /*
133 * It is necessary to return fd by reference because 0 is a valid file
134 * descriptor number, and the child needs to be able to distinguish
135 * itself from the parent using the return value.
136 */
137 error = fork1(td, &fr);
138 if (error == 0) {
139 td->td_retval[0] = pid;
140 td->td_retval[1] = 0;
141 error = copyout(&fd, uap->fdp, sizeof(fd));
142 }
143 return (error);
144 }
145
146 /* ARGSUSED */
147 int
148 sys_vfork(struct thread *td, struct vfork_args *uap)
149 {
150 struct fork_req fr;
151 int error, pid;
152
153 bzero(&fr, sizeof(fr));
154 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
155 fr.fr_pidp = &pid;
156 error = fork1(td, &fr);
157 if (error == 0) {
158 td->td_retval[0] = pid;
159 td->td_retval[1] = 0;
160 }
161 return (error);
162 }
163
164 int
165 sys_rfork(struct thread *td, struct rfork_args *uap)
166 {
167 struct fork_req fr;
168 int error, pid;
169
170 /* Don't allow kernel-only flags. */
171 if ((uap->flags & RFKERNELONLY) != 0)
172 return (EINVAL);
173 /* RFSPAWN must not appear with others */
174 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
175 return (EINVAL);
176
177 AUDIT_ARG_FFLAGS(uap->flags);
178 bzero(&fr, sizeof(fr));
179 if ((uap->flags & RFSPAWN) != 0) {
180 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
181 fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
182 } else {
183 fr.fr_flags = uap->flags;
184 }
185 fr.fr_pidp = &pid;
186 error = fork1(td, &fr);
187 if (error == 0) {
188 td->td_retval[0] = pid;
189 td->td_retval[1] = 0;
190 }
191 return (error);
192 }
193
194 int __exclusive_cache_line nprocs = 1; /* process 0 */
195 int lastpid = 0;
196 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
197 "Last used PID");
198
199 /*
200 * Random component to lastpid generation. We mix in a random factor to make
201 * it a little harder to predict. We sanity check the modulus value to avoid
202 * doing it in critical paths. Don't let it be too small or we pointlessly
203 * waste randomness entropy, and don't let it be impossibly large. Using a
204 * modulus that is too big causes a LOT more process table scans and slows
205 * down fork processing as the pidchecked caching is defeated.
206 */
207 static int randompid = 0;
208
209 static int
210 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
211 {
212 int error, pid;
213
214 error = sysctl_wire_old_buffer(req, sizeof(int));
215 if (error != 0)
216 return(error);
217 sx_xlock(&allproc_lock);
218 pid = randompid;
219 error = sysctl_handle_int(oidp, &pid, 0, req);
220 if (error == 0 && req->newptr != NULL) {
221 if (pid == 0)
222 randompid = 0;
223 else if (pid == 1)
224 /* generate a random PID modulus between 100 and 1123 */
225 randompid = 100 + arc4random() % 1024;
226 else if (pid < 0 || pid > pid_max - 100)
227 /* out of range */
228 randompid = pid_max - 100;
229 else if (pid < 100)
230 /* Make it reasonable */
231 randompid = 100;
232 else
233 randompid = pid;
234 }
235 sx_xunlock(&allproc_lock);
236 return (error);
237 }
238
239 SYSCTL_PROC(_kern, OID_AUTO, randompid,
240 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
241 sysctl_kern_randompid, "I",
242 "Random PID modulus. Special values: 0: disable, 1: choose random value");
243
244 extern bitstr_t proc_id_pidmap;
245 extern bitstr_t proc_id_grpidmap;
246 extern bitstr_t proc_id_sessidmap;
247 extern bitstr_t proc_id_reapmap;
248
249 /*
250 * Find an unused process ID
251 *
252 * If RFHIGHPID is set (used during system boot), do not allocate
253 * low-numbered pids.
254 */
255 static int
256 fork_findpid(int flags)
257 {
258 pid_t result;
259 int trypid, random;
260
261 /*
262 * Avoid calling arc4random with procid_lock held.
263 */
264 random = 0;
265 if (__predict_false(randompid))
266 random = arc4random() % randompid;
267
268 mtx_lock(&procid_lock);
269
270 trypid = lastpid + 1;
271 if (flags & RFHIGHPID) {
272 if (trypid < 10)
273 trypid = 10;
274 } else {
275 trypid += random;
276 }
277 retry:
278 if (trypid >= pid_max)
279 trypid = 2;
280
281 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
282 if (result == -1) {
283 KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
284 trypid = 2;
285 goto retry;
286 }
287 if (bit_test(&proc_id_grpidmap, result) ||
288 bit_test(&proc_id_sessidmap, result) ||
289 bit_test(&proc_id_reapmap, result)) {
290 trypid = result + 1;
291 goto retry;
292 }
293
294 /*
295 * RFHIGHPID does not mess with the lastpid counter during boot.
296 */
297 if ((flags & RFHIGHPID) == 0)
298 lastpid = result;
299
300 bit_set(&proc_id_pidmap, result);
301 mtx_unlock(&procid_lock);
302
303 return (result);
304 }
305
306 static int
307 fork_norfproc(struct thread *td, int flags)
308 {
309 int error;
310 struct proc *p1;
311
312 KASSERT((flags & RFPROC) == 0,
313 ("fork_norfproc called with RFPROC set"));
314 p1 = td->td_proc;
315
316 /*
317 * Quiesce other threads if necessary. If RFMEM is not specified we
318 * must ensure that other threads do not concurrently create a second
319 * process sharing the vmspace, see vmspace_unshare().
320 */
321 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
322 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
323 PROC_LOCK(p1);
324 if (thread_single(p1, SINGLE_BOUNDARY)) {
325 PROC_UNLOCK(p1);
326 return (ERESTART);
327 }
328 PROC_UNLOCK(p1);
329 }
330
331 error = vm_forkproc(td, NULL, NULL, NULL, flags);
332 if (error)
333 goto fail;
334
335 /*
336 * Close all file descriptors.
337 */
338 if (flags & RFCFDG) {
339 struct filedesc *fdtmp;
340 struct pwddesc *pdtmp;
341 pdtmp = pdinit(td->td_proc->p_pd, false);
342 fdtmp = fdinit(td->td_proc->p_fd, false, NULL);
343 pdescfree(td);
344 fdescfree(td);
345 p1->p_fd = fdtmp;
346 p1->p_pd = pdtmp;
347 }
348
349 /*
350 * Unshare file descriptors (from parent).
351 */
352 if (flags & RFFDG) {
353 fdunshare(td);
354 pdunshare(td);
355 }
356
357 fail:
358 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
359 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
360 PROC_LOCK(p1);
361 thread_single_end(p1, SINGLE_BOUNDARY);
362 PROC_UNLOCK(p1);
363 }
364 return (error);
365 }
366
367 static void
368 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
369 struct vmspace *vm2, struct file *fp_procdesc)
370 {
371 struct proc *p1, *pptr;
372 struct filedesc *fd;
373 struct filedesc_to_leader *fdtol;
374 struct pwddesc *pd;
375 struct sigacts *newsigacts;
376
377 p1 = td->td_proc;
378
379 PROC_LOCK(p1);
380 bcopy(&p1->p_startcopy, &p2->p_startcopy,
381 __rangeof(struct proc, p_startcopy, p_endcopy));
382 pargs_hold(p2->p_args);
383 PROC_UNLOCK(p1);
384
385 bzero(&p2->p_startzero,
386 __rangeof(struct proc, p_startzero, p_endzero));
387
388 /* Tell the prison that we exist. */
389 prison_proc_hold(p2->p_ucred->cr_prison);
390
391 p2->p_state = PRS_NEW; /* protect against others */
392 p2->p_pid = fork_findpid(fr->fr_flags);
393 AUDIT_ARG_PID(p2->p_pid);
394
395 sx_xlock(&allproc_lock);
396 LIST_INSERT_HEAD(&allproc, p2, p_list);
397 allproc_gen++;
398 sx_xunlock(&allproc_lock);
399
400 sx_xlock(PIDHASHLOCK(p2->p_pid));
401 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
402 sx_xunlock(PIDHASHLOCK(p2->p_pid));
403
404 tidhash_add(td2);
405
406 /*
407 * Malloc things while we don't hold any locks.
408 */
409 if (fr->fr_flags & RFSIGSHARE)
410 newsigacts = NULL;
411 else
412 newsigacts = sigacts_alloc();
413
414 /*
415 * Copy filedesc.
416 */
417 if (fr->fr_flags & RFCFDG) {
418 pd = pdinit(p1->p_pd, false);
419 fd = fdinit(p1->p_fd, false, NULL);
420 fdtol = NULL;
421 } else if (fr->fr_flags & RFFDG) {
422 if (fr->fr_flags2 & FR2_SHARE_PATHS)
423 pd = pdshare(p1->p_pd);
424 else
425 pd = pdcopy(p1->p_pd);
426 fd = fdcopy(p1->p_fd);
427 fdtol = NULL;
428 } else {
429 if (fr->fr_flags2 & FR2_SHARE_PATHS)
430 pd = pdcopy(p1->p_pd);
431 else
432 pd = pdshare(p1->p_pd);
433 fd = fdshare(p1->p_fd);
434 if (p1->p_fdtol == NULL)
435 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
436 p1->p_leader);
437 if ((fr->fr_flags & RFTHREAD) != 0) {
438 /*
439 * Shared file descriptor table, and shared
440 * process leaders.
441 */
442 fdtol = p1->p_fdtol;
443 FILEDESC_XLOCK(p1->p_fd);
444 fdtol->fdl_refcount++;
445 FILEDESC_XUNLOCK(p1->p_fd);
446 } else {
447 /*
448 * Shared file descriptor table, and different
449 * process leaders.
450 */
451 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
452 p1->p_fd, p2);
453 }
454 }
455 /*
456 * Make a proc table entry for the new process.
457 * Start by zeroing the section of proc that is zero-initialized,
458 * then copy the section that is copied directly from the parent.
459 */
460
461 PROC_LOCK(p2);
462 PROC_LOCK(p1);
463
464 bzero(&td2->td_startzero,
465 __rangeof(struct thread, td_startzero, td_endzero));
466
467 bcopy(&td->td_startcopy, &td2->td_startcopy,
468 __rangeof(struct thread, td_startcopy, td_endcopy));
469
470 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
471 td2->td_sigstk = td->td_sigstk;
472 td2->td_flags = TDF_INMEM;
473 td2->td_lend_user_pri = PRI_MAX;
474
475 #ifdef VIMAGE
476 td2->td_vnet = NULL;
477 td2->td_vnet_lpush = NULL;
478 #endif
479
480 /*
481 * Allow the scheduler to initialize the child.
482 */
483 thread_lock(td);
484 sched_fork(td, td2);
485 thread_unlock(td);
486
487 /*
488 * Duplicate sub-structures as needed.
489 * Increase reference counts on shared objects.
490 */
491 p2->p_flag = P_INMEM;
492 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
493 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
494 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
495 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS |
496 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
497 p2->p_swtick = ticks;
498 if (p1->p_flag & P_PROFIL)
499 startprofclock(p2);
500
501 if (fr->fr_flags & RFSIGSHARE) {
502 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
503 } else {
504 sigacts_copy(newsigacts, p1->p_sigacts);
505 p2->p_sigacts = newsigacts;
506 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
507 mtx_lock(&p2->p_sigacts->ps_mtx);
508 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
509 sig_drop_caught(p2);
510 if ((fr->fr_flags2 & FR2_KPROC) != 0)
511 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
512 mtx_unlock(&p2->p_sigacts->ps_mtx);
513 }
514 }
515
516 if (fr->fr_flags & RFTSIGZMB)
517 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
518 else if (fr->fr_flags & RFLINUXTHPN)
519 p2->p_sigparent = SIGUSR1;
520 else
521 p2->p_sigparent = SIGCHLD;
522
523 if ((fr->fr_flags2 & FR2_KPROC) != 0) {
524 p2->p_flag |= P_SYSTEM | P_KPROC;
525 td2->td_pflags |= TDP_KTHREAD;
526 }
527
528 p2->p_textvp = p1->p_textvp;
529 p2->p_textdvp = p1->p_textdvp;
530 p2->p_fd = fd;
531 p2->p_fdtol = fdtol;
532 p2->p_pd = pd;
533
534 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
535 p2->p_flag |= P_PROTECTED;
536 p2->p_flag2 |= P2_INHERIT_PROTECTED;
537 }
538
539 /*
540 * p_limit is copy-on-write. Bump its refcount.
541 */
542 lim_fork(p1, p2);
543
544 thread_cow_get_proc(td2, p2);
545
546 pstats_fork(p1->p_stats, p2->p_stats);
547
548 PROC_UNLOCK(p1);
549 PROC_UNLOCK(p2);
550
551 /*
552 * Bump references to the text vnode and directory, and copy
553 * the hardlink name.
554 */
555 if (p2->p_textvp != NULL)
556 vrefact(p2->p_textvp);
557 if (p2->p_textdvp != NULL)
558 vrefact(p2->p_textdvp);
559 p2->p_binname = p1->p_binname == NULL ? NULL :
560 strdup(p1->p_binname, M_PARGS);
561
562 /*
563 * Set up linkage for kernel based threading.
564 */
565 if ((fr->fr_flags & RFTHREAD) != 0) {
566 mtx_lock(&ppeers_lock);
567 p2->p_peers = p1->p_peers;
568 p1->p_peers = p2;
569 p2->p_leader = p1->p_leader;
570 mtx_unlock(&ppeers_lock);
571 PROC_LOCK(p1->p_leader);
572 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
573 PROC_UNLOCK(p1->p_leader);
574 /*
575 * The task leader is exiting, so process p1 is
576 * going to be killed shortly. Since p1 obviously
577 * isn't dead yet, we know that the leader is either
578 * sending SIGKILL's to all the processes in this
579 * task or is sleeping waiting for all the peers to
580 * exit. We let p1 complete the fork, but we need
581 * to go ahead and kill the new process p2 since
582 * the task leader may not get a chance to send
583 * SIGKILL to it. We leave it on the list so that
584 * the task leader will wait for this new process
585 * to commit suicide.
586 */
587 PROC_LOCK(p2);
588 kern_psignal(p2, SIGKILL);
589 PROC_UNLOCK(p2);
590 } else
591 PROC_UNLOCK(p1->p_leader);
592 } else {
593 p2->p_peers = NULL;
594 p2->p_leader = p2;
595 }
596
597 sx_xlock(&proctree_lock);
598 PGRP_LOCK(p1->p_pgrp);
599 PROC_LOCK(p2);
600 PROC_LOCK(p1);
601
602 /*
603 * Preserve some more flags in subprocess. P_PROFIL has already
604 * been preserved.
605 */
606 p2->p_flag |= p1->p_flag & P_SUGID;
607 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK |
608 TDP_SIGFASTBLOCK)) | TDP_FORKING;
609 SESS_LOCK(p1->p_session);
610 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
611 p2->p_flag |= P_CONTROLT;
612 SESS_UNLOCK(p1->p_session);
613 if (fr->fr_flags & RFPPWAIT)
614 p2->p_flag |= P_PPWAIT;
615
616 p2->p_pgrp = p1->p_pgrp;
617 LIST_INSERT_AFTER(p1, p2, p_pglist);
618 PGRP_UNLOCK(p1->p_pgrp);
619 LIST_INIT(&p2->p_children);
620 LIST_INIT(&p2->p_orphans);
621
622 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
623 TAILQ_INIT(&p2->p_kqtim_stop);
624
625 /*
626 * This begins the section where we must prevent the parent
627 * from being swapped.
628 */
629 _PHOLD(p1);
630 PROC_UNLOCK(p1);
631
632 /*
633 * Attach the new process to its parent.
634 *
635 * If RFNOWAIT is set, the newly created process becomes a child
636 * of init. This effectively disassociates the child from the
637 * parent.
638 */
639 if ((fr->fr_flags & RFNOWAIT) != 0) {
640 pptr = p1->p_reaper;
641 p2->p_reaper = pptr;
642 } else {
643 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
644 p1 : p1->p_reaper;
645 pptr = p1;
646 }
647 p2->p_pptr = pptr;
648 p2->p_oppid = pptr->p_pid;
649 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
650 LIST_INIT(&p2->p_reaplist);
651 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
652 if (p2->p_reaper == p1 && p1 != initproc) {
653 p2->p_reapsubtree = p2->p_pid;
654 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
655 }
656 sx_xunlock(&proctree_lock);
657
658 /* Inform accounting that we have forked. */
659 p2->p_acflag = AFORK;
660 PROC_UNLOCK(p2);
661
662 #ifdef KTRACE
663 ktrprocfork(p1, p2);
664 #endif
665
666 /*
667 * Finish creating the child process. It will return via a different
668 * execution path later. (ie: directly into user mode)
669 */
670 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
671
672 if (fr->fr_flags == (RFFDG | RFPROC)) {
673 VM_CNT_INC(v_forks);
674 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
675 p2->p_vmspace->vm_ssize);
676 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
677 VM_CNT_INC(v_vforks);
678 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
679 p2->p_vmspace->vm_ssize);
680 } else if (p1 == &proc0) {
681 VM_CNT_INC(v_kthreads);
682 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
683 p2->p_vmspace->vm_ssize);
684 } else {
685 VM_CNT_INC(v_rforks);
686 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
687 p2->p_vmspace->vm_ssize);
688 }
689
690 /*
691 * Associate the process descriptor with the process before anything
692 * can happen that might cause that process to need the descriptor.
693 * However, don't do this until after fork(2) can no longer fail.
694 */
695 if (fr->fr_flags & RFPROCDESC)
696 procdesc_new(p2, fr->fr_pd_flags);
697
698 /*
699 * Both processes are set up, now check if any loadable modules want
700 * to adjust anything.
701 */
702 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
703
704 /*
705 * Set the child start time and mark the process as being complete.
706 */
707 PROC_LOCK(p2);
708 PROC_LOCK(p1);
709 microuptime(&p2->p_stats->p_start);
710 PROC_SLOCK(p2);
711 p2->p_state = PRS_NORMAL;
712 PROC_SUNLOCK(p2);
713
714 #ifdef KDTRACE_HOOKS
715 /*
716 * Tell the DTrace fasttrap provider about the new process so that any
717 * tracepoints inherited from the parent can be removed. We have to do
718 * this only after p_state is PRS_NORMAL since the fasttrap module will
719 * use pfind() later on.
720 */
721 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
722 dtrace_fasttrap_fork(p1, p2);
723 #endif
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 * Tell any interested parties about the new process.
733 */
734 knote_fork(p1->p_klist, p2->p_pid);
735
736 /*
737 * Now can be swapped.
738 */
739 _PRELE(p1);
740 PROC_UNLOCK(p1);
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, false);
778 }
779 PROC_UNLOCK(p2);
780 sx_xunlock(&proctree_lock);
781 }
782
783 racct_proc_fork_done(p2);
784
785 if ((fr->fr_flags & RFSTOPPED) == 0) {
786 if (fr->fr_pidp != NULL)
787 *fr->fr_pidp = p2->p_pid;
788 /*
789 * If RFSTOPPED not requested, make child runnable and
790 * add to run queue.
791 */
792 thread_lock(td2);
793 TD_SET_CAN_RUN(td2);
794 sched_add(td2, SRQ_BORING);
795 } else {
796 *fr->fr_procp = p2;
797 }
798 }
799
800 void
801 fork_rfppwait(struct thread *td)
802 {
803 struct proc *p, *p2;
804
805 MPASS(td->td_pflags & TDP_RFPPWAIT);
806
807 p = td->td_proc;
808 /*
809 * Preserve synchronization semantics of vfork. If
810 * waiting for child to exec or exit, fork set
811 * P_PPWAIT on child, and there we sleep on our proc
812 * (in case of exit).
813 *
814 * Do it after the ptracestop() above is finished, to
815 * not block our debugger until child execs or exits
816 * to finish vfork wait.
817 */
818 td->td_pflags &= ~TDP_RFPPWAIT;
819 p2 = td->td_rfppwait_p;
820 again:
821 PROC_LOCK(p2);
822 while (p2->p_flag & P_PPWAIT) {
823 PROC_LOCK(p);
824 if (thread_suspend_check_needed()) {
825 PROC_UNLOCK(p2);
826 thread_suspend_check(0);
827 PROC_UNLOCK(p);
828 goto again;
829 } else {
830 PROC_UNLOCK(p);
831 }
832 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
833 }
834 PROC_UNLOCK(p2);
835
836 if (td->td_dbgflags & TDB_VFORK) {
837 PROC_LOCK(p);
838 if (p->p_ptevents & PTRACE_VFORK)
839 ptracestop(td, SIGTRAP, NULL);
840 td->td_dbgflags &= ~TDB_VFORK;
841 PROC_UNLOCK(p);
842 }
843 }
844
845 int
846 fork1(struct thread *td, struct fork_req *fr)
847 {
848 struct proc *p1, *newproc;
849 struct thread *td2;
850 struct vmspace *vm2;
851 struct ucred *cred;
852 struct file *fp_procdesc;
853 vm_ooffset_t mem_charged;
854 int error, nprocs_new;
855 static int curfail;
856 static struct timeval lastfail;
857 int flags, pages;
858
859 flags = fr->fr_flags;
860 pages = fr->fr_pages;
861
862 if ((flags & RFSTOPPED) != 0)
863 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
864 else
865 MPASS(fr->fr_procp == NULL);
866
867 /* Check for the undefined or unimplemented flags. */
868 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
869 return (EINVAL);
870
871 /* Signal value requires RFTSIGZMB. */
872 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
873 return (EINVAL);
874
875 /* Can't copy and clear. */
876 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
877 return (EINVAL);
878
879 /* Check the validity of the signal number. */
880 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
881 return (EINVAL);
882
883 if ((flags & RFPROCDESC) != 0) {
884 /* Can't not create a process yet get a process descriptor. */
885 if ((flags & RFPROC) == 0)
886 return (EINVAL);
887
888 /* Must provide a place to put a procdesc if creating one. */
889 if (fr->fr_pd_fd == NULL)
890 return (EINVAL);
891
892 /* Check if we are using supported flags. */
893 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
894 return (EINVAL);
895 }
896
897 p1 = td->td_proc;
898
899 /*
900 * Here we don't create a new process, but we divorce
901 * certain parts of a process from itself.
902 */
903 if ((flags & RFPROC) == 0) {
904 if (fr->fr_procp != NULL)
905 *fr->fr_procp = NULL;
906 else if (fr->fr_pidp != NULL)
907 *fr->fr_pidp = 0;
908 return (fork_norfproc(td, flags));
909 }
910
911 fp_procdesc = NULL;
912 newproc = NULL;
913 vm2 = NULL;
914
915 /*
916 * Increment the nprocs resource before allocations occur.
917 * Although process entries are dynamically created, we still
918 * keep a global limit on the maximum number we will
919 * create. There are hard-limits as to the number of processes
920 * that can run, established by the KVA and memory usage for
921 * the process data.
922 *
923 * Don't allow a nonprivileged user to use the last ten
924 * processes; don't let root exceed the limit.
925 */
926 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
927 if (nprocs_new >= maxproc - 10) {
928 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
929 nprocs_new >= maxproc) {
930 error = EAGAIN;
931 sx_xlock(&allproc_lock);
932 if (ppsratecheck(&lastfail, &curfail, 1)) {
933 printf("maxproc limit exceeded by uid %u "
934 "(pid %d); see tuning(7) and "
935 "login.conf(5)\n",
936 td->td_ucred->cr_ruid, p1->p_pid);
937 }
938 sx_xunlock(&allproc_lock);
939 goto fail2;
940 }
941 }
942
943 /*
944 * If required, create a process descriptor in the parent first; we
945 * will abandon it if something goes wrong. We don't finit() until
946 * later.
947 */
948 if (flags & RFPROCDESC) {
949 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
950 fr->fr_pd_flags, fr->fr_pd_fcaps);
951 if (error != 0)
952 goto fail2;
953 AUDIT_ARG_FD(*fr->fr_pd_fd);
954 }
955
956 mem_charged = 0;
957 if (pages == 0)
958 pages = kstack_pages;
959 /* Allocate new proc. */
960 newproc = uma_zalloc(proc_zone, M_WAITOK);
961 td2 = FIRST_THREAD_IN_PROC(newproc);
962 if (td2 == NULL) {
963 td2 = thread_alloc(pages);
964 if (td2 == NULL) {
965 error = ENOMEM;
966 goto fail2;
967 }
968 proc_linkup(newproc, td2);
969 } else {
970 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
971 if (td2->td_kstack != 0)
972 vm_thread_dispose(td2);
973 if (!thread_alloc_stack(td2, pages)) {
974 error = ENOMEM;
975 goto fail2;
976 }
977 }
978 }
979
980 if ((flags & RFMEM) == 0) {
981 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
982 if (vm2 == NULL) {
983 error = ENOMEM;
984 goto fail2;
985 }
986 if (!swap_reserve(mem_charged)) {
987 /*
988 * The swap reservation failed. The accounting
989 * from the entries of the copied vm2 will be
990 * subtracted in vmspace_free(), so force the
991 * reservation there.
992 */
993 swap_reserve_force(mem_charged);
994 error = ENOMEM;
995 goto fail2;
996 }
997 } else
998 vm2 = NULL;
999
1000 /*
1001 * XXX: This is ugly; when we copy resource usage, we need to bump
1002 * per-cred resource counters.
1003 */
1004 proc_set_cred_init(newproc, td->td_ucred);
1005
1006 /*
1007 * Initialize resource accounting for the child process.
1008 */
1009 error = racct_proc_fork(p1, newproc);
1010 if (error != 0) {
1011 error = EAGAIN;
1012 goto fail1;
1013 }
1014
1015 #ifdef MAC
1016 mac_proc_init(newproc);
1017 #endif
1018 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1019 STAILQ_INIT(&newproc->p_ktr);
1020
1021 /*
1022 * Increment the count of procs running with this uid. Don't allow
1023 * a nonprivileged user to exceed their current limit.
1024 */
1025 cred = td->td_ucred;
1026 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1027 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1028 goto fail0;
1029 chgproccnt(cred->cr_ruidinfo, 1, 0);
1030 }
1031
1032 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1033 return (0);
1034 fail0:
1035 error = EAGAIN;
1036 #ifdef MAC
1037 mac_proc_destroy(newproc);
1038 #endif
1039 racct_proc_exit(newproc);
1040 fail1:
1041 proc_unset_cred(newproc);
1042 fail2:
1043 if (vm2 != NULL)
1044 vmspace_free(vm2);
1045 uma_zfree(proc_zone, newproc);
1046 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1047 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1048 fdrop(fp_procdesc, td);
1049 }
1050 atomic_add_int(&nprocs, -1);
1051 pause("fork", hz / 2);
1052 return (error);
1053 }
1054
1055 /*
1056 * Handle the return of a child process from fork1(). This function
1057 * is called from the MD fork_trampoline() entry point.
1058 */
1059 void
1060 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1061 struct trapframe *frame)
1062 {
1063 struct proc *p;
1064 struct thread *td;
1065 struct thread *dtd;
1066
1067 td = curthread;
1068 p = td->td_proc;
1069 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1070
1071 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1072 td, td_get_sched(td), p->p_pid, td->td_name);
1073
1074 sched_fork_exit(td);
1075 /*
1076 * Processes normally resume in mi_switch() after being
1077 * cpu_switch()'ed to, but when children start up they arrive here
1078 * instead, so we must do much the same things as mi_switch() would.
1079 */
1080 if ((dtd = PCPU_GET(deadthread))) {
1081 PCPU_SET(deadthread, NULL);
1082 thread_stash(dtd);
1083 }
1084 thread_unlock(td);
1085
1086 /*
1087 * cpu_fork_kthread_handler intercepts this function call to
1088 * have this call a non-return function to stay in kernel mode.
1089 * initproc has its own fork handler, but it does return.
1090 */
1091 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1092 callout(arg, frame);
1093
1094 /*
1095 * Check if a kernel thread misbehaved and returned from its main
1096 * function.
1097 */
1098 if (p->p_flag & P_KPROC) {
1099 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1100 td->td_name, p->p_pid);
1101 kthread_exit();
1102 }
1103 mtx_assert(&Giant, MA_NOTOWNED);
1104
1105 if (p->p_sysent->sv_schedtail != NULL)
1106 (p->p_sysent->sv_schedtail)(td);
1107 td->td_pflags &= ~TDP_FORKING;
1108 }
1109
1110 /*
1111 * Simplified back end of syscall(), used when returning from fork()
1112 * directly into user mode. This function is passed in to fork_exit()
1113 * as the first parameter and is called when returning to a new
1114 * userland process.
1115 */
1116 void
1117 fork_return(struct thread *td, struct trapframe *frame)
1118 {
1119 struct proc *p;
1120
1121 p = td->td_proc;
1122 if (td->td_dbgflags & TDB_STOPATFORK) {
1123 PROC_LOCK(p);
1124 if ((p->p_flag & P_TRACED) != 0) {
1125 /*
1126 * Inform the debugger if one is still present.
1127 */
1128 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1129 ptracestop(td, SIGSTOP, NULL);
1130 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1131 } else {
1132 /*
1133 * ... otherwise clear the request.
1134 */
1135 td->td_dbgflags &= ~TDB_STOPATFORK;
1136 }
1137 PROC_UNLOCK(p);
1138 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1139 /*
1140 * This is the start of a new thread in a traced
1141 * process. Report a system call exit event.
1142 */
1143 PROC_LOCK(p);
1144 td->td_dbgflags |= TDB_SCX;
1145 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1146 (td->td_dbgflags & TDB_BORN) != 0)
1147 ptracestop(td, SIGTRAP, NULL);
1148 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1149 PROC_UNLOCK(p);
1150 }
1151
1152 /*
1153 * If the prison was killed mid-fork, die along with it.
1154 */
1155 if (!prison_isalive(td->td_ucred->cr_prison))
1156 exit1(td, 0, SIGKILL);
1157
1158 userret(td, frame);
1159
1160 #ifdef KTRACE
1161 if (KTRPOINT(td, KTR_SYSRET))
1162 ktrsysret(SYS_fork, 0, 0);
1163 #endif
1164 }
Cache object: 71e3a0002cc25febcc9e55406358aab3
|