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
37 #include <sys/cdefs.h>
38 #include "opt_ktrace.h"
39 #include "opt_kstack_pages.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/bitstring.h>
44 #include <sys/sysproto.h>
45 #include <sys/eventhandler.h>
46 #include <sys/fcntl.h>
47 #include <sys/filedesc.h>
48 #include <sys/jail.h>
49 #include <sys/kernel.h>
50 #include <sys/kthread.h>
51 #include <sys/sysctl.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/msan.h>
55 #include <sys/mutex.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/procdesc.h>
59 #include <sys/ptrace.h>
60 #include <sys/racct.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sched.h>
63 #include <sys/syscall.h>
64 #include <sys/vmmeter.h>
65 #include <sys/vnode.h>
66 #include <sys/acct.h>
67 #include <sys/ktr.h>
68 #include <sys/ktrace.h>
69 #include <sys/unistd.h>
70 #include <sys/sdt.h>
71 #include <sys/sx.h>
72 #include <sys/sysent.h>
73 #include <sys/signalvar.h>
74
75 #include <security/audit/audit.h>
76 #include <security/mac/mac_framework.h>
77
78 #include <vm/vm.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_extern.h>
82 #include <vm/uma.h>
83
84 #ifdef KDTRACE_HOOKS
85 #include <sys/dtrace_bsd.h>
86 dtrace_fork_func_t dtrace_fasttrap_fork;
87 #endif
88
89 SDT_PROVIDER_DECLARE(proc);
90 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
91
92 #ifndef _SYS_SYSPROTO_H_
93 struct fork_args {
94 int dummy;
95 };
96 #endif
97
98 /* ARGSUSED */
99 int
sys_fork(struct thread * td,struct fork_args * uap)100 sys_fork(struct thread *td, struct fork_args *uap)
101 {
102 struct fork_req fr;
103 int error, pid;
104
105 bzero(&fr, sizeof(fr));
106 fr.fr_flags = RFFDG | RFPROC;
107 fr.fr_pidp = &pid;
108 error = fork1(td, &fr);
109 if (error == 0) {
110 td->td_retval[0] = pid;
111 td->td_retval[1] = 0;
112 }
113 return (error);
114 }
115
116 /* ARGUSED */
117 int
sys_pdfork(struct thread * td,struct pdfork_args * uap)118 sys_pdfork(struct thread *td, struct pdfork_args *uap)
119 {
120 struct fork_req fr;
121 int error, fd, pid;
122
123 bzero(&fr, sizeof(fr));
124 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
125 fr.fr_pidp = &pid;
126 fr.fr_pd_fd = &fd;
127 fr.fr_pd_flags = uap->flags;
128 AUDIT_ARG_FFLAGS(uap->flags);
129 /*
130 * It is necessary to return fd by reference because 0 is a valid file
131 * descriptor number, and the child needs to be able to distinguish
132 * itself from the parent using the return value.
133 */
134 error = fork1(td, &fr);
135 if (error == 0) {
136 td->td_retval[0] = pid;
137 td->td_retval[1] = 0;
138 error = copyout(&fd, uap->fdp, sizeof(fd));
139 }
140 return (error);
141 }
142
143 /* ARGSUSED */
144 int
sys_vfork(struct thread * td,struct vfork_args * uap)145 sys_vfork(struct thread *td, struct vfork_args *uap)
146 {
147 struct fork_req fr;
148 int error, pid;
149
150 bzero(&fr, sizeof(fr));
151 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
152 fr.fr_pidp = &pid;
153 error = fork1(td, &fr);
154 if (error == 0) {
155 td->td_retval[0] = pid;
156 td->td_retval[1] = 0;
157 }
158 return (error);
159 }
160
161 int
sys_rfork(struct thread * td,struct rfork_args * uap)162 sys_rfork(struct thread *td, struct rfork_args *uap)
163 {
164 struct fork_req fr;
165 int error, pid;
166
167 /* Don't allow kernel-only flags. */
168 if ((uap->flags & RFKERNELONLY) != 0)
169 return (EINVAL);
170 /* RFSPAWN must not appear with others */
171 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
172 return (EINVAL);
173
174 AUDIT_ARG_FFLAGS(uap->flags);
175 bzero(&fr, sizeof(fr));
176 if ((uap->flags & RFSPAWN) != 0) {
177 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
178 fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
179 } else {
180 fr.fr_flags = uap->flags;
181 }
182 fr.fr_pidp = &pid;
183 error = fork1(td, &fr);
184 if (error == 0) {
185 td->td_retval[0] = pid;
186 td->td_retval[1] = 0;
187 }
188 return (error);
189 }
190
191 int __exclusive_cache_line nprocs = 1; /* process 0 */
192 int lastpid = 0;
193 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
194 "Last used PID");
195
196 /*
197 * Random component to lastpid generation. We mix in a random factor to make
198 * it a little harder to predict. We sanity check the modulus value to avoid
199 * doing it in critical paths. Don't let it be too small or we pointlessly
200 * waste randomness entropy, and don't let it be impossibly large. Using a
201 * modulus that is too big causes a LOT more process table scans and slows
202 * down fork processing as the pidchecked caching is defeated.
203 */
204 static int randompid = 0;
205
206 static int
sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)207 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
208 {
209 int error, pid;
210
211 error = sysctl_wire_old_buffer(req, sizeof(int));
212 if (error != 0)
213 return(error);
214 sx_xlock(&allproc_lock);
215 pid = randompid;
216 error = sysctl_handle_int(oidp, &pid, 0, req);
217 if (error == 0 && req->newptr != NULL) {
218 if (pid == 0)
219 randompid = 0;
220 else if (pid == 1)
221 /* generate a random PID modulus between 100 and 1123 */
222 randompid = 100 + arc4random() % 1024;
223 else if (pid < 0 || pid > pid_max - 100)
224 /* out of range */
225 randompid = pid_max - 100;
226 else if (pid < 100)
227 /* Make it reasonable */
228 randompid = 100;
229 else
230 randompid = pid;
231 }
232 sx_xunlock(&allproc_lock);
233 return (error);
234 }
235
236 SYSCTL_PROC(_kern, OID_AUTO, randompid,
237 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
238 sysctl_kern_randompid, "I",
239 "Random PID modulus. Special values: 0: disable, 1: choose random value");
240
241 extern bitstr_t proc_id_pidmap;
242 extern bitstr_t proc_id_grpidmap;
243 extern bitstr_t proc_id_sessidmap;
244 extern bitstr_t proc_id_reapmap;
245
246 /*
247 * Find an unused process ID
248 *
249 * If RFHIGHPID is set (used during system boot), do not allocate
250 * low-numbered pids.
251 */
252 static int
fork_findpid(int flags)253 fork_findpid(int flags)
254 {
255 pid_t result;
256 int trypid, random;
257
258 /*
259 * Avoid calling arc4random with procid_lock held.
260 */
261 random = 0;
262 if (__predict_false(randompid))
263 random = arc4random() % randompid;
264
265 mtx_lock(&procid_lock);
266
267 trypid = lastpid + 1;
268 if (flags & RFHIGHPID) {
269 if (trypid < 10)
270 trypid = 10;
271 } else {
272 trypid += random;
273 }
274 retry:
275 if (trypid >= pid_max)
276 trypid = 2;
277
278 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
279 if (result == -1) {
280 KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
281 trypid = 2;
282 goto retry;
283 }
284 if (bit_test(&proc_id_grpidmap, result) ||
285 bit_test(&proc_id_sessidmap, result) ||
286 bit_test(&proc_id_reapmap, result)) {
287 trypid = result + 1;
288 goto retry;
289 }
290
291 /*
292 * RFHIGHPID does not mess with the lastpid counter during boot.
293 */
294 if ((flags & RFHIGHPID) == 0)
295 lastpid = result;
296
297 bit_set(&proc_id_pidmap, result);
298 mtx_unlock(&procid_lock);
299
300 return (result);
301 }
302
303 static int
fork_norfproc(struct thread * td,int flags)304 fork_norfproc(struct thread *td, int flags)
305 {
306 struct proc *p1;
307 int error;
308
309 KASSERT((flags & RFPROC) == 0,
310 ("fork_norfproc called with RFPROC set"));
311 p1 = td->td_proc;
312
313 /*
314 * Quiesce other threads if necessary. If RFMEM is not specified we
315 * must ensure that other threads do not concurrently create a second
316 * process sharing the vmspace, see vmspace_unshare().
317 */
318 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
319 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
320 PROC_LOCK(p1);
321 if (thread_single(p1, SINGLE_BOUNDARY)) {
322 PROC_UNLOCK(p1);
323 return (ERESTART);
324 }
325 PROC_UNLOCK(p1);
326 }
327
328 error = vm_forkproc(td, NULL, NULL, NULL, flags);
329 if (error != 0)
330 goto fail;
331
332 /*
333 * Close all file descriptors.
334 */
335 if ((flags & RFCFDG) != 0) {
336 struct filedesc *fdtmp;
337 struct pwddesc *pdtmp;
338
339 pdtmp = pdinit(td->td_proc->p_pd, false);
340 fdtmp = fdinit();
341 pdescfree(td);
342 fdescfree(td);
343 p1->p_fd = fdtmp;
344 p1->p_pd = pdtmp;
345 }
346
347 /*
348 * Unshare file descriptors (from parent).
349 */
350 if ((flags & RFFDG) != 0) {
351 fdunshare(td);
352 pdunshare(td);
353 }
354
355 fail:
356 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
357 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
358 PROC_LOCK(p1);
359 thread_single_end(p1, SINGLE_BOUNDARY);
360 PROC_UNLOCK(p1);
361 }
362 return (error);
363 }
364
365 static void
do_fork(struct thread * td,struct fork_req * fr,struct proc * p2,struct thread * td2,struct vmspace * vm2,struct file * fp_procdesc)366 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
367 struct vmspace *vm2, struct file *fp_procdesc)
368 {
369 struct proc *p1, *pptr;
370 struct filedesc *fd;
371 struct filedesc_to_leader *fdtol;
372 struct pwddesc *pd;
373 struct sigacts *newsigacts;
374
375 p1 = td->td_proc;
376
377 PROC_LOCK(p1);
378 bcopy(&p1->p_startcopy, &p2->p_startcopy,
379 __rangeof(struct proc, p_startcopy, p_endcopy));
380 pargs_hold(p2->p_args);
381 PROC_UNLOCK(p1);
382
383 bzero(&p2->p_startzero,
384 __rangeof(struct proc, p_startzero, p_endzero));
385
386 /* Tell the prison that we exist. */
387 prison_proc_hold(p2->p_ucred->cr_prison);
388
389 p2->p_state = PRS_NEW; /* protect against others */
390 p2->p_pid = fork_findpid(fr->fr_flags);
391 AUDIT_ARG_PID(p2->p_pid);
392 TSFORK(p2->p_pid, p1->p_pid);
393
394 sx_xlock(&allproc_lock);
395 LIST_INSERT_HEAD(&allproc, p2, p_list);
396 allproc_gen++;
397 prison_proc_link(p2->p_ucred->cr_prison, p2);
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();
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 = filedesc_to_leader_share(p1->p_fdtol, p1->p_fd);
443 } else {
444 /*
445 * Shared file descriptor table, and different
446 * process leaders.
447 */
448 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
449 p1->p_fd, p2);
450 }
451 }
452 /*
453 * Make a proc table entry for the new process.
454 * Start by zeroing the section of proc that is zero-initialized,
455 * then copy the section that is copied directly from the parent.
456 */
457
458 PROC_LOCK(p2);
459 PROC_LOCK(p1);
460
461 bzero(&td2->td_startzero,
462 __rangeof(struct thread, td_startzero, td_endzero));
463
464 bcopy(&td->td_startcopy, &td2->td_startcopy,
465 __rangeof(struct thread, td_startcopy, td_endcopy));
466
467 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
468 td2->td_sigstk = td->td_sigstk;
469 td2->td_flags = TDF_INMEM;
470 td2->td_lend_user_pri = PRI_MAX;
471
472 #ifdef VIMAGE
473 td2->td_vnet = NULL;
474 td2->td_vnet_lpush = NULL;
475 #endif
476
477 /*
478 * Allow the scheduler to initialize the child.
479 */
480 thread_lock(td);
481 sched_fork(td, td2);
482 /*
483 * Request AST to check for TDP_RFPPWAIT. Do it here
484 * to avoid calling thread_lock() again.
485 */
486 if ((fr->fr_flags & RFPPWAIT) != 0)
487 ast_sched_locked(td, TDA_VFORK);
488 thread_unlock(td);
489
490 /*
491 * Duplicate sub-structures as needed.
492 * Increase reference counts on shared objects.
493 */
494 p2->p_flag = P_INMEM;
495 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
496 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
497 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
498 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS |
499 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC | P2_LOGSIGEXIT_CTL |
500 P2_LOGSIGEXIT_ENABLE);
501 p2->p_swtick = ticks;
502 if (p1->p_flag & P_PROFIL)
503 startprofclock(p2);
504
505 if (fr->fr_flags & RFSIGSHARE) {
506 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
507 } else {
508 sigacts_copy(newsigacts, p1->p_sigacts);
509 p2->p_sigacts = newsigacts;
510 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
511 mtx_lock(&p2->p_sigacts->ps_mtx);
512 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
513 sig_drop_caught(p2);
514 if ((fr->fr_flags2 & FR2_KPROC) != 0)
515 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
516 mtx_unlock(&p2->p_sigacts->ps_mtx);
517 }
518 }
519
520 if (fr->fr_flags & RFTSIGZMB)
521 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
522 else if (fr->fr_flags & RFLINUXTHPN)
523 p2->p_sigparent = SIGUSR1;
524 else
525 p2->p_sigparent = SIGCHLD;
526
527 if ((fr->fr_flags2 & FR2_KPROC) != 0) {
528 p2->p_flag |= P_SYSTEM | P_KPROC;
529 td2->td_pflags |= TDP_KTHREAD;
530 }
531
532 p2->p_textvp = p1->p_textvp;
533 p2->p_textdvp = p1->p_textdvp;
534 p2->p_fd = fd;
535 p2->p_fdtol = fdtol;
536 p2->p_pd = pd;
537
538 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
539 p2->p_flag |= P_PROTECTED;
540 p2->p_flag2 |= P2_INHERIT_PROTECTED;
541 }
542
543 /*
544 * p_limit is copy-on-write. Bump its refcount.
545 */
546 lim_fork(p1, p2);
547
548 thread_cow_get_proc(td2, p2);
549
550 pstats_fork(p1->p_stats, p2->p_stats);
551
552 PROC_UNLOCK(p1);
553 PROC_UNLOCK(p2);
554
555 /*
556 * Bump references to the text vnode and directory, and copy
557 * the hardlink name.
558 */
559 if (p2->p_textvp != NULL)
560 vrefact(p2->p_textvp);
561 if (p2->p_textdvp != NULL)
562 vrefact(p2->p_textdvp);
563 p2->p_binname = p1->p_binname == NULL ? NULL :
564 strdup(p1->p_binname, M_PARGS);
565
566 /*
567 * Set up linkage for kernel based threading.
568 */
569 if ((fr->fr_flags & RFTHREAD) != 0) {
570 mtx_lock(&ppeers_lock);
571 p2->p_peers = p1->p_peers;
572 p1->p_peers = p2;
573 p2->p_leader = p1->p_leader;
574 mtx_unlock(&ppeers_lock);
575 PROC_LOCK(p1->p_leader);
576 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
577 PROC_UNLOCK(p1->p_leader);
578 /*
579 * The task leader is exiting, so process p1 is
580 * going to be killed shortly. Since p1 obviously
581 * isn't dead yet, we know that the leader is either
582 * sending SIGKILL's to all the processes in this
583 * task or is sleeping waiting for all the peers to
584 * exit. We let p1 complete the fork, but we need
585 * to go ahead and kill the new process p2 since
586 * the task leader may not get a chance to send
587 * SIGKILL to it. We leave it on the list so that
588 * the task leader will wait for this new process
589 * to commit suicide.
590 */
591 PROC_LOCK(p2);
592 kern_psignal(p2, SIGKILL);
593 PROC_UNLOCK(p2);
594 } else
595 PROC_UNLOCK(p1->p_leader);
596 } else {
597 p2->p_peers = NULL;
598 p2->p_leader = p2;
599 }
600
601 sx_xlock(&proctree_lock);
602 PGRP_LOCK(p1->p_pgrp);
603 PROC_LOCK(p2);
604 PROC_LOCK(p1);
605
606 /*
607 * Preserve some more flags in subprocess. P_PROFIL has already
608 * been preserved.
609 */
610 p2->p_flag |= p1->p_flag & P_SUGID;
611 td2->td_pflags |= td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK);
612 td2->td_pflags2 |= td->td_pflags2 & TDP2_UEXTERR;
613 if (p1->p_flag & P_CONTROLT) {
614 SESS_LOCK(p1->p_session);
615 if (p1->p_session->s_ttyvp != NULL)
616 p2->p_flag |= P_CONTROLT;
617 SESS_UNLOCK(p1->p_session);
618 }
619 if (fr->fr_flags & RFPPWAIT)
620 p2->p_flag |= P_PPWAIT;
621
622 p2->p_pgrp = p1->p_pgrp;
623 LIST_INSERT_AFTER(p1, p2, p_pglist);
624 PGRP_UNLOCK(p1->p_pgrp);
625 LIST_INIT(&p2->p_children);
626 LIST_INIT(&p2->p_orphans);
627
628 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
629
630 /*
631 * This begins the section where we must prevent the parent
632 * from being swapped.
633 */
634 _PHOLD(p1);
635 PROC_UNLOCK(p1);
636
637 /*
638 * Attach the new process to its parent.
639 *
640 * If RFNOWAIT is set, the newly created process becomes a child
641 * of init. This effectively disassociates the child from the
642 * parent.
643 */
644 if ((fr->fr_flags & RFNOWAIT) != 0) {
645 pptr = p1->p_reaper;
646 p2->p_reaper = pptr;
647 } else {
648 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
649 p1 : p1->p_reaper;
650 pptr = p1;
651 }
652 p2->p_pptr = pptr;
653 p2->p_oppid = pptr->p_pid;
654 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
655 LIST_INIT(&p2->p_reaplist);
656 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
657 if (p2->p_reaper == p1 && p1 != initproc) {
658 p2->p_reapsubtree = p2->p_pid;
659 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
660 }
661 sx_xunlock(&proctree_lock);
662
663 /* Inform accounting that we have forked. */
664 p2->p_acflag = AFORK;
665 PROC_UNLOCK(p2);
666
667 #ifdef KTRACE
668 ktrprocfork(p1, p2);
669 #endif
670
671 /*
672 * Finish creating the child process. It will return via a different
673 * execution path later. (ie: directly into user mode)
674 */
675 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
676
677 if (fr->fr_flags == (RFFDG | RFPROC)) {
678 VM_CNT_INC(v_forks);
679 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
680 p2->p_vmspace->vm_ssize);
681 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
682 VM_CNT_INC(v_vforks);
683 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
684 p2->p_vmspace->vm_ssize);
685 } else if (p1 == &proc0) {
686 VM_CNT_INC(v_kthreads);
687 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
688 p2->p_vmspace->vm_ssize);
689 } else {
690 VM_CNT_INC(v_rforks);
691 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
692 p2->p_vmspace->vm_ssize);
693 }
694
695 /*
696 * Associate the process descriptor with the process before anything
697 * can happen that might cause that process to need the descriptor.
698 * However, don't do this until after fork(2) can no longer fail.
699 */
700 if (fr->fr_flags & RFPROCDESC)
701 procdesc_new(p2, fr->fr_pd_flags);
702
703 /*
704 * Both processes are set up, now check if any loadable modules want
705 * to adjust anything.
706 */
707 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
708
709 /*
710 * Set the child start time and mark the process as being complete.
711 */
712 PROC_LOCK(p2);
713 PROC_LOCK(p1);
714 microuptime(&p2->p_stats->p_start);
715 PROC_SLOCK(p2);
716 p2->p_state = PRS_NORMAL;
717 PROC_SUNLOCK(p2);
718
719 #ifdef KDTRACE_HOOKS
720 /*
721 * Tell the DTrace fasttrap provider about the new process so that any
722 * tracepoints inherited from the parent can be removed. We have to do
723 * this only after p_state is PRS_NORMAL since the fasttrap module will
724 * use pfind() later on.
725 */
726 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
727 dtrace_fasttrap_fork(p1, p2);
728 #endif
729 if (fr->fr_flags & RFPPWAIT) {
730 td->td_pflags |= TDP_RFPPWAIT;
731 td->td_rfppwait_p = p2;
732 td->td_dbgflags |= TDB_VFORK;
733 }
734 PROC_UNLOCK(p2);
735
736 /*
737 * Tell any interested parties about the new process.
738 */
739 knote_fork(p1->p_klist, p2->p_pid);
740
741 /*
742 * Now can be swapped.
743 */
744 _PRELE(p1);
745 PROC_UNLOCK(p1);
746 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
747
748 if (fr->fr_flags & RFPROCDESC) {
749 procdesc_finit(p2->p_procdesc, fp_procdesc);
750 fdrop(fp_procdesc, td);
751 }
752
753 /*
754 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
755 * synced with forks in progress so it is OK if we miss it
756 * if being set atm.
757 */
758 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
759 sx_xlock(&proctree_lock);
760 PROC_LOCK(p2);
761
762 /*
763 * p1->p_ptevents & p1->p_pptr are protected by both
764 * process and proctree locks for modifications,
765 * so owning proctree_lock allows the race-free read.
766 */
767 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
768 /*
769 * Arrange for debugger to receive the fork event.
770 *
771 * We can report PL_FLAG_FORKED regardless of
772 * P_FOLLOWFORK settings, but it does not make a sense
773 * for runaway child.
774 */
775 td->td_dbgflags |= TDB_FORK;
776 td->td_dbg_forked = p2->p_pid;
777 td2->td_dbgflags |= TDB_STOPATFORK;
778 proc_set_traced(p2, true);
779 CTR2(KTR_PTRACE,
780 "do_fork: attaching to new child pid %d: oppid %d",
781 p2->p_pid, p2->p_oppid);
782 proc_reparent(p2, p1->p_pptr, false);
783 }
784 PROC_UNLOCK(p2);
785 sx_xunlock(&proctree_lock);
786 }
787
788 racct_proc_fork_done(p2);
789
790 if ((fr->fr_flags & RFSTOPPED) == 0) {
791 if (fr->fr_pidp != NULL)
792 *fr->fr_pidp = p2->p_pid;
793 /*
794 * If RFSTOPPED not requested, make child runnable and
795 * add to run queue.
796 */
797 thread_lock(td2);
798 TD_SET_CAN_RUN(td2);
799 sched_add(td2, SRQ_BORING);
800 } else {
801 *fr->fr_procp = p2;
802 }
803 }
804
805 static void
ast_vfork(struct thread * td,int tda __unused)806 ast_vfork(struct thread *td, int tda __unused)
807 {
808 struct proc *p, *p2;
809
810 MPASS(td->td_pflags & TDP_RFPPWAIT);
811
812 p = td->td_proc;
813 /*
814 * Preserve synchronization semantics of vfork. If
815 * waiting for child to exec or exit, fork set
816 * P_PPWAIT on child, and there we sleep on our proc
817 * (in case of exit).
818 *
819 * Do it after the ptracestop() above is finished, to
820 * not block our debugger until child execs or exits
821 * to finish vfork wait.
822 */
823 td->td_pflags &= ~TDP_RFPPWAIT;
824 p2 = td->td_rfppwait_p;
825 again:
826 PROC_LOCK(p2);
827 while (p2->p_flag & P_PPWAIT) {
828 PROC_LOCK(p);
829 if (thread_suspend_check_needed()) {
830 PROC_UNLOCK(p2);
831 thread_suspend_check(0);
832 PROC_UNLOCK(p);
833 goto again;
834 } else {
835 PROC_UNLOCK(p);
836 }
837 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
838 }
839 PROC_UNLOCK(p2);
840
841 if (td->td_dbgflags & TDB_VFORK) {
842 PROC_LOCK(p);
843 if (p->p_ptevents & PTRACE_VFORK)
844 ptracestop(td, SIGTRAP, NULL);
845 td->td_dbgflags &= ~TDB_VFORK;
846 PROC_UNLOCK(p);
847 }
848 }
849
850 int
fork1(struct thread * td,struct fork_req * fr)851 fork1(struct thread *td, struct fork_req *fr)
852 {
853 struct proc *p1, *newproc;
854 struct thread *td2;
855 struct vmspace *vm2;
856 struct ucred *cred;
857 struct file *fp_procdesc;
858 struct pgrp *pg;
859 vm_ooffset_t mem_charged;
860 int error, nprocs_new;
861 static int curfail;
862 static struct timeval lastfail;
863 int flags, pages;
864 bool killsx_locked, singlethreaded;
865
866 flags = fr->fr_flags;
867 pages = fr->fr_pages;
868
869 if ((flags & RFSTOPPED) != 0)
870 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
871 else
872 MPASS(fr->fr_procp == NULL);
873
874 /* Check for the undefined or unimplemented flags. */
875 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
876 return (EINVAL);
877
878 /* Signal value requires RFTSIGZMB. */
879 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
880 return (EINVAL);
881
882 /* Can't copy and clear. */
883 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
884 return (EINVAL);
885
886 /* Check the validity of the signal number. */
887 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
888 return (EINVAL);
889
890 if ((flags & RFPROCDESC) != 0) {
891 /* Can't not create a process yet get a process descriptor. */
892 if ((flags & RFPROC) == 0)
893 return (EINVAL);
894
895 /* Must provide a place to put a procdesc if creating one. */
896 if (fr->fr_pd_fd == NULL)
897 return (EINVAL);
898
899 /* Check if we are using supported flags. */
900 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
901 return (EINVAL);
902 }
903
904 p1 = td->td_proc;
905
906 /*
907 * Here we don't create a new process, but we divorce
908 * certain parts of a process from itself.
909 */
910 if ((flags & RFPROC) == 0) {
911 if (fr->fr_procp != NULL)
912 *fr->fr_procp = NULL;
913 else if (fr->fr_pidp != NULL)
914 *fr->fr_pidp = 0;
915 return (fork_norfproc(td, flags));
916 }
917
918 fp_procdesc = NULL;
919 newproc = NULL;
920 vm2 = NULL;
921 killsx_locked = false;
922 singlethreaded = false;
923
924 /*
925 * Increment the nprocs resource before allocations occur.
926 * Although process entries are dynamically created, we still
927 * keep a global limit on the maximum number we will
928 * create. There are hard-limits as to the number of processes
929 * that can run, established by the KVA and memory usage for
930 * the process data.
931 *
932 * Don't allow a nonprivileged user to use the last ten
933 * processes; don't let root exceed the limit.
934 */
935 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
936 if (nprocs_new >= maxproc - 10) {
937 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
938 nprocs_new >= maxproc) {
939 error = EAGAIN;
940 sx_xlock(&allproc_lock);
941 if (ppsratecheck(&lastfail, &curfail, 1)) {
942 printf("maxproc limit exceeded by uid %u "
943 "(pid %d); see tuning(7) and "
944 "login.conf(5)\n",
945 td->td_ucred->cr_ruid, p1->p_pid);
946 }
947 sx_xunlock(&allproc_lock);
948 goto fail2;
949 }
950 }
951
952 /*
953 * If we are possibly multi-threaded, and there is a process
954 * sending a signal to our group right now, ensure that our
955 * other threads cannot be chosen for the signal queueing.
956 * Otherwise, this might delay signal action, and make the new
957 * child escape the signaling.
958 */
959 pg = p1->p_pgrp;
960 if (p1->p_numthreads > 1) {
961 if (sx_try_slock(&pg->pg_killsx) != 0) {
962 killsx_locked = true;
963 } else {
964 PROC_LOCK(p1);
965 if (thread_single(p1, SINGLE_BOUNDARY)) {
966 PROC_UNLOCK(p1);
967 error = ERESTART;
968 goto fail2;
969 }
970 PROC_UNLOCK(p1);
971 singlethreaded = true;
972 }
973 }
974
975 /*
976 * Atomically check for signals and block processes from sending
977 * a signal to our process group until the child is visible.
978 */
979 if (!killsx_locked && sx_slock_sig(&pg->pg_killsx) != 0) {
980 error = ERESTART;
981 goto fail2;
982 }
983 if (__predict_false(p1->p_pgrp != pg || sig_intr() != 0)) {
984 /*
985 * Either the process was moved to other process
986 * group, or there is pending signal. sx_slock_sig()
987 * does not check for signals if not sleeping for the
988 * lock.
989 */
990 sx_sunlock(&pg->pg_killsx);
991 killsx_locked = false;
992 error = ERESTART;
993 goto fail2;
994 } else {
995 killsx_locked = true;
996 }
997
998 /*
999 * If required, create a process descriptor in the parent first; we
1000 * will abandon it if something goes wrong. We don't finit() until
1001 * later.
1002 */
1003 if (flags & RFPROCDESC) {
1004 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
1005 fr->fr_pd_flags, fr->fr_pd_fcaps);
1006 if (error != 0)
1007 goto fail2;
1008 AUDIT_ARG_FD(*fr->fr_pd_fd);
1009 }
1010
1011 mem_charged = 0;
1012 if (pages == 0)
1013 pages = kstack_pages;
1014 /* Allocate new proc. */
1015 newproc = uma_zalloc(proc_zone, M_WAITOK);
1016 td2 = FIRST_THREAD_IN_PROC(newproc);
1017 if (td2 == NULL) {
1018 td2 = thread_alloc(pages);
1019 if (td2 == NULL) {
1020 error = ENOMEM;
1021 goto fail2;
1022 }
1023 proc_linkup(newproc, td2);
1024 } else {
1025 error = thread_recycle(td2, pages);
1026 if (error != 0)
1027 goto fail2;
1028 }
1029
1030 if ((flags & RFMEM) == 0) {
1031 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
1032 if (vm2 == NULL) {
1033 error = ENOMEM;
1034 goto fail2;
1035 }
1036 if (!swap_reserve(mem_charged)) {
1037 /*
1038 * The swap reservation failed. The accounting
1039 * from the entries of the copied vm2 will be
1040 * subtracted in vmspace_free(), so force the
1041 * reservation there.
1042 */
1043 swap_reserve_force(mem_charged);
1044 error = ENOMEM;
1045 goto fail2;
1046 }
1047 } else
1048 vm2 = NULL;
1049
1050 /*
1051 * XXX: This is ugly; when we copy resource usage, we need to bump
1052 * per-cred resource counters.
1053 */
1054 newproc->p_ucred = crcowget(td->td_ucred);
1055
1056 /*
1057 * Initialize resource accounting for the child process.
1058 */
1059 error = racct_proc_fork(p1, newproc);
1060 if (error != 0) {
1061 error = EAGAIN;
1062 goto fail1;
1063 }
1064
1065 #ifdef MAC
1066 mac_proc_init(newproc);
1067 #endif
1068 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1069 STAILQ_INIT(&newproc->p_ktr);
1070
1071 /*
1072 * Increment the count of procs running with this uid. Don't allow
1073 * a nonprivileged user to exceed their current limit.
1074 */
1075 cred = td->td_ucred;
1076 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1077 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1078 goto fail0;
1079 chgproccnt(cred->cr_ruidinfo, 1, 0);
1080 }
1081
1082 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1083 error = 0;
1084 goto cleanup;
1085 fail0:
1086 error = EAGAIN;
1087 #ifdef MAC
1088 mac_proc_destroy(newproc);
1089 #endif
1090 racct_proc_exit(newproc);
1091 fail1:
1092 proc_unset_cred(newproc, false);
1093 fail2:
1094 if (vm2 != NULL)
1095 vmspace_free(vm2);
1096 uma_zfree(proc_zone, newproc);
1097 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1098 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1099 fdrop(fp_procdesc, td);
1100 }
1101 atomic_add_int(&nprocs, -1);
1102 cleanup:
1103 if (killsx_locked)
1104 sx_sunlock(&pg->pg_killsx);
1105 if (singlethreaded) {
1106 PROC_LOCK(p1);
1107 thread_single_end(p1, SINGLE_BOUNDARY);
1108 PROC_UNLOCK(p1);
1109 }
1110 if (error != 0)
1111 pause("fork", hz / 2);
1112 return (error);
1113 }
1114
1115 /*
1116 * Handle the return of a child process from fork1(). This function
1117 * is called from the MD fork_trampoline() entry point.
1118 */
1119 void
fork_exit(void (* callout)(void *,struct trapframe *),void * arg,struct trapframe * frame)1120 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1121 struct trapframe *frame)
1122 {
1123 struct proc *p;
1124 struct thread *td;
1125 struct thread *dtd;
1126
1127 kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED);
1128
1129 td = curthread;
1130 p = td->td_proc;
1131 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1132
1133 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1134 td, td_get_sched(td), p->p_pid, td->td_name);
1135
1136 sched_fork_exit(td);
1137
1138 /*
1139 * Processes normally resume in mi_switch() after being
1140 * cpu_switch()'ed to, but when children start up they arrive here
1141 * instead, so we must do much the same things as mi_switch() would.
1142 */
1143 if ((dtd = PCPU_GET(deadthread))) {
1144 PCPU_SET(deadthread, NULL);
1145 thread_stash(dtd);
1146 }
1147 thread_unlock(td);
1148
1149 /*
1150 * cpu_fork_kthread_handler intercepts this function call to
1151 * have this call a non-return function to stay in kernel mode.
1152 * initproc has its own fork handler, but it does return.
1153 */
1154 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1155 callout(arg, frame);
1156
1157 /*
1158 * Check if a kernel thread misbehaved and returned from its main
1159 * function.
1160 */
1161 if (p->p_flag & P_KPROC) {
1162 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1163 td->td_name, p->p_pid);
1164 kthread_exit();
1165 }
1166 mtx_assert(&Giant, MA_NOTOWNED);
1167
1168 /*
1169 * Now going to return to userland.
1170 */
1171
1172 if (p->p_sysent->sv_schedtail != NULL)
1173 (p->p_sysent->sv_schedtail)(td);
1174
1175 userret(td, frame);
1176 }
1177
1178 /*
1179 * Simplified back end of syscall(), used when returning from fork()
1180 * directly into user mode. This function is passed in to fork_exit()
1181 * as the first parameter and is called when returning to a new
1182 * userland process.
1183 */
1184 void
fork_return(struct thread * td,struct trapframe * frame)1185 fork_return(struct thread *td, struct trapframe *frame)
1186 {
1187 struct proc *p;
1188
1189 p = td->td_proc;
1190 if (td->td_dbgflags & TDB_STOPATFORK) {
1191 PROC_LOCK(p);
1192 if ((p->p_flag & P_TRACED) != 0) {
1193 /*
1194 * Inform the debugger if one is still present.
1195 */
1196 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1197 ptracestop(td, SIGSTOP, NULL);
1198 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1199 } else {
1200 /*
1201 * ... otherwise clear the request.
1202 */
1203 td->td_dbgflags &= ~TDB_STOPATFORK;
1204 }
1205 PROC_UNLOCK(p);
1206 } else if (p->p_flag & P_TRACED) {
1207 /*
1208 * This is the start of a new thread in a traced
1209 * process. Report a system call exit event.
1210 */
1211 PROC_LOCK(p);
1212 td->td_dbgflags |= TDB_SCX;
1213 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1214 (td->td_dbgflags & TDB_BORN) != 0)
1215 ptracestop(td, SIGTRAP, NULL);
1216 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1217 PROC_UNLOCK(p);
1218 }
1219
1220 /*
1221 * If the prison was killed mid-fork, die along with it.
1222 */
1223 if (!prison_isalive(td->td_ucred->cr_prison))
1224 exit1(td, 0, SIGKILL);
1225
1226 #ifdef KTRACE
1227 if (KTRPOINT(td, KTR_SYSRET))
1228 ktrsysret(td->td_sa.code, 0, 0);
1229 #endif
1230 }
1231
1232 static void
fork_init(void * arg __unused)1233 fork_init(void *arg __unused)
1234 {
1235 ast_register(TDA_VFORK, ASTR_ASTF_REQUIRED | ASTR_TDP, TDP_RFPPWAIT,
1236 ast_vfork);
1237 }
1238 SYSINIT(fork, SI_SUB_INTRINSIC, SI_ORDER_ANY, fork_init, NULL);
1239