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 SESS_LOCK(p1->p_session);
614 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
615 p2->p_flag |= P_CONTROLT;
616 SESS_UNLOCK(p1->p_session);
617 if (fr->fr_flags & RFPPWAIT)
618 p2->p_flag |= P_PPWAIT;
619
620 p2->p_pgrp = p1->p_pgrp;
621 LIST_INSERT_AFTER(p1, p2, p_pglist);
622 PGRP_UNLOCK(p1->p_pgrp);
623 LIST_INIT(&p2->p_children);
624 LIST_INIT(&p2->p_orphans);
625
626 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
627
628 /*
629 * This begins the section where we must prevent the parent
630 * from being swapped.
631 */
632 _PHOLD(p1);
633 PROC_UNLOCK(p1);
634
635 /*
636 * Attach the new process to its parent.
637 *
638 * If RFNOWAIT is set, the newly created process becomes a child
639 * of init. This effectively disassociates the child from the
640 * parent.
641 */
642 if ((fr->fr_flags & RFNOWAIT) != 0) {
643 pptr = p1->p_reaper;
644 p2->p_reaper = pptr;
645 } else {
646 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
647 p1 : p1->p_reaper;
648 pptr = p1;
649 }
650 p2->p_pptr = pptr;
651 p2->p_oppid = pptr->p_pid;
652 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
653 LIST_INIT(&p2->p_reaplist);
654 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
655 if (p2->p_reaper == p1 && p1 != initproc) {
656 p2->p_reapsubtree = p2->p_pid;
657 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
658 }
659 sx_xunlock(&proctree_lock);
660
661 /* Inform accounting that we have forked. */
662 p2->p_acflag = AFORK;
663 PROC_UNLOCK(p2);
664
665 #ifdef KTRACE
666 ktrprocfork(p1, p2);
667 #endif
668
669 /*
670 * Finish creating the child process. It will return via a different
671 * execution path later. (ie: directly into user mode)
672 */
673 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
674
675 if (fr->fr_flags == (RFFDG | RFPROC)) {
676 VM_CNT_INC(v_forks);
677 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
678 p2->p_vmspace->vm_ssize);
679 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
680 VM_CNT_INC(v_vforks);
681 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
682 p2->p_vmspace->vm_ssize);
683 } else if (p1 == &proc0) {
684 VM_CNT_INC(v_kthreads);
685 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
686 p2->p_vmspace->vm_ssize);
687 } else {
688 VM_CNT_INC(v_rforks);
689 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
690 p2->p_vmspace->vm_ssize);
691 }
692
693 /*
694 * Associate the process descriptor with the process before anything
695 * can happen that might cause that process to need the descriptor.
696 * However, don't do this until after fork(2) can no longer fail.
697 */
698 if (fr->fr_flags & RFPROCDESC)
699 procdesc_new(p2, fr->fr_pd_flags);
700
701 /*
702 * Both processes are set up, now check if any loadable modules want
703 * to adjust anything.
704 */
705 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
706
707 /*
708 * Set the child start time and mark the process as being complete.
709 */
710 PROC_LOCK(p2);
711 PROC_LOCK(p1);
712 microuptime(&p2->p_stats->p_start);
713 PROC_SLOCK(p2);
714 p2->p_state = PRS_NORMAL;
715 PROC_SUNLOCK(p2);
716
717 #ifdef KDTRACE_HOOKS
718 /*
719 * Tell the DTrace fasttrap provider about the new process so that any
720 * tracepoints inherited from the parent can be removed. We have to do
721 * this only after p_state is PRS_NORMAL since the fasttrap module will
722 * use pfind() later on.
723 */
724 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
725 dtrace_fasttrap_fork(p1, p2);
726 #endif
727 if (fr->fr_flags & RFPPWAIT) {
728 td->td_pflags |= TDP_RFPPWAIT;
729 td->td_rfppwait_p = p2;
730 td->td_dbgflags |= TDB_VFORK;
731 }
732 PROC_UNLOCK(p2);
733
734 /*
735 * Tell any interested parties about the new process.
736 */
737 knote_fork(p1->p_klist, p2->p_pid);
738
739 /*
740 * Now can be swapped.
741 */
742 _PRELE(p1);
743 PROC_UNLOCK(p1);
744 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
745
746 if (fr->fr_flags & RFPROCDESC) {
747 procdesc_finit(p2->p_procdesc, fp_procdesc);
748 fdrop(fp_procdesc, td);
749 }
750
751 /*
752 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
753 * synced with forks in progress so it is OK if we miss it
754 * if being set atm.
755 */
756 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
757 sx_xlock(&proctree_lock);
758 PROC_LOCK(p2);
759
760 /*
761 * p1->p_ptevents & p1->p_pptr are protected by both
762 * process and proctree locks for modifications,
763 * so owning proctree_lock allows the race-free read.
764 */
765 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
766 /*
767 * Arrange for debugger to receive the fork event.
768 *
769 * We can report PL_FLAG_FORKED regardless of
770 * P_FOLLOWFORK settings, but it does not make a sense
771 * for runaway child.
772 */
773 td->td_dbgflags |= TDB_FORK;
774 td->td_dbg_forked = p2->p_pid;
775 td2->td_dbgflags |= TDB_STOPATFORK;
776 proc_set_traced(p2, true);
777 CTR2(KTR_PTRACE,
778 "do_fork: attaching to new child pid %d: oppid %d",
779 p2->p_pid, p2->p_oppid);
780 proc_reparent(p2, p1->p_pptr, false);
781 }
782 PROC_UNLOCK(p2);
783 sx_xunlock(&proctree_lock);
784 }
785
786 racct_proc_fork_done(p2);
787
788 if ((fr->fr_flags & RFSTOPPED) == 0) {
789 if (fr->fr_pidp != NULL)
790 *fr->fr_pidp = p2->p_pid;
791 /*
792 * If RFSTOPPED not requested, make child runnable and
793 * add to run queue.
794 */
795 thread_lock(td2);
796 TD_SET_CAN_RUN(td2);
797 sched_add(td2, SRQ_BORING);
798 } else {
799 *fr->fr_procp = p2;
800 }
801 }
802
803 static void
ast_vfork(struct thread * td,int tda __unused)804 ast_vfork(struct thread *td, int tda __unused)
805 {
806 struct proc *p, *p2;
807
808 MPASS(td->td_pflags & TDP_RFPPWAIT);
809
810 p = td->td_proc;
811 /*
812 * Preserve synchronization semantics of vfork. If
813 * waiting for child to exec or exit, fork set
814 * P_PPWAIT on child, and there we sleep on our proc
815 * (in case of exit).
816 *
817 * Do it after the ptracestop() above is finished, to
818 * not block our debugger until child execs or exits
819 * to finish vfork wait.
820 */
821 td->td_pflags &= ~TDP_RFPPWAIT;
822 p2 = td->td_rfppwait_p;
823 again:
824 PROC_LOCK(p2);
825 while (p2->p_flag & P_PPWAIT) {
826 PROC_LOCK(p);
827 if (thread_suspend_check_needed()) {
828 PROC_UNLOCK(p2);
829 thread_suspend_check(0);
830 PROC_UNLOCK(p);
831 goto again;
832 } else {
833 PROC_UNLOCK(p);
834 }
835 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
836 }
837 PROC_UNLOCK(p2);
838
839 if (td->td_dbgflags & TDB_VFORK) {
840 PROC_LOCK(p);
841 if (p->p_ptevents & PTRACE_VFORK)
842 ptracestop(td, SIGTRAP, NULL);
843 td->td_dbgflags &= ~TDB_VFORK;
844 PROC_UNLOCK(p);
845 }
846 }
847
848 int
fork1(struct thread * td,struct fork_req * fr)849 fork1(struct thread *td, struct fork_req *fr)
850 {
851 struct proc *p1, *newproc;
852 struct thread *td2;
853 struct vmspace *vm2;
854 struct ucred *cred;
855 struct file *fp_procdesc;
856 struct pgrp *pg;
857 vm_ooffset_t mem_charged;
858 int error, nprocs_new;
859 static int curfail;
860 static struct timeval lastfail;
861 int flags, pages;
862 bool killsx_locked, singlethreaded;
863
864 flags = fr->fr_flags;
865 pages = fr->fr_pages;
866
867 if ((flags & RFSTOPPED) != 0)
868 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
869 else
870 MPASS(fr->fr_procp == NULL);
871
872 /* Check for the undefined or unimplemented flags. */
873 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
874 return (EINVAL);
875
876 /* Signal value requires RFTSIGZMB. */
877 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
878 return (EINVAL);
879
880 /* Can't copy and clear. */
881 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
882 return (EINVAL);
883
884 /* Check the validity of the signal number. */
885 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
886 return (EINVAL);
887
888 if ((flags & RFPROCDESC) != 0) {
889 /* Can't not create a process yet get a process descriptor. */
890 if ((flags & RFPROC) == 0)
891 return (EINVAL);
892
893 /* Must provide a place to put a procdesc if creating one. */
894 if (fr->fr_pd_fd == NULL)
895 return (EINVAL);
896
897 /* Check if we are using supported flags. */
898 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
899 return (EINVAL);
900 }
901
902 p1 = td->td_proc;
903
904 /*
905 * Here we don't create a new process, but we divorce
906 * certain parts of a process from itself.
907 */
908 if ((flags & RFPROC) == 0) {
909 if (fr->fr_procp != NULL)
910 *fr->fr_procp = NULL;
911 else if (fr->fr_pidp != NULL)
912 *fr->fr_pidp = 0;
913 return (fork_norfproc(td, flags));
914 }
915
916 fp_procdesc = NULL;
917 newproc = NULL;
918 vm2 = NULL;
919 killsx_locked = false;
920 singlethreaded = false;
921
922 /*
923 * Increment the nprocs resource before allocations occur.
924 * Although process entries are dynamically created, we still
925 * keep a global limit on the maximum number we will
926 * create. There are hard-limits as to the number of processes
927 * that can run, established by the KVA and memory usage for
928 * the process data.
929 *
930 * Don't allow a nonprivileged user to use the last ten
931 * processes; don't let root exceed the limit.
932 */
933 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
934 if (nprocs_new >= maxproc - 10) {
935 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
936 nprocs_new >= maxproc) {
937 error = EAGAIN;
938 sx_xlock(&allproc_lock);
939 if (ppsratecheck(&lastfail, &curfail, 1)) {
940 printf("maxproc limit exceeded by uid %u "
941 "(pid %d); see tuning(7) and "
942 "login.conf(5)\n",
943 td->td_ucred->cr_ruid, p1->p_pid);
944 }
945 sx_xunlock(&allproc_lock);
946 goto fail2;
947 }
948 }
949
950 /*
951 * If we are possibly multi-threaded, and there is a process
952 * sending a signal to our group right now, ensure that our
953 * other threads cannot be chosen for the signal queueing.
954 * Otherwise, this might delay signal action, and make the new
955 * child escape the signaling.
956 */
957 pg = p1->p_pgrp;
958 if (p1->p_numthreads > 1) {
959 if (sx_try_slock(&pg->pg_killsx) != 0) {
960 killsx_locked = true;
961 } else {
962 PROC_LOCK(p1);
963 if (thread_single(p1, SINGLE_BOUNDARY)) {
964 PROC_UNLOCK(p1);
965 error = ERESTART;
966 goto fail2;
967 }
968 PROC_UNLOCK(p1);
969 singlethreaded = true;
970 }
971 }
972
973 /*
974 * Atomically check for signals and block processes from sending
975 * a signal to our process group until the child is visible.
976 */
977 if (!killsx_locked && sx_slock_sig(&pg->pg_killsx) != 0) {
978 error = ERESTART;
979 goto fail2;
980 }
981 if (__predict_false(p1->p_pgrp != pg || sig_intr() != 0)) {
982 /*
983 * Either the process was moved to other process
984 * group, or there is pending signal. sx_slock_sig()
985 * does not check for signals if not sleeping for the
986 * lock.
987 */
988 sx_sunlock(&pg->pg_killsx);
989 killsx_locked = false;
990 error = ERESTART;
991 goto fail2;
992 } else {
993 killsx_locked = true;
994 }
995
996 /*
997 * If required, create a process descriptor in the parent first; we
998 * will abandon it if something goes wrong. We don't finit() until
999 * later.
1000 */
1001 if (flags & RFPROCDESC) {
1002 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
1003 fr->fr_pd_flags, fr->fr_pd_fcaps);
1004 if (error != 0)
1005 goto fail2;
1006 AUDIT_ARG_FD(*fr->fr_pd_fd);
1007 }
1008
1009 mem_charged = 0;
1010 if (pages == 0)
1011 pages = kstack_pages;
1012 /* Allocate new proc. */
1013 newproc = uma_zalloc(proc_zone, M_WAITOK);
1014 td2 = FIRST_THREAD_IN_PROC(newproc);
1015 if (td2 == NULL) {
1016 td2 = thread_alloc(pages);
1017 if (td2 == NULL) {
1018 error = ENOMEM;
1019 goto fail2;
1020 }
1021 proc_linkup(newproc, td2);
1022 } else {
1023 error = thread_recycle(td2, pages);
1024 if (error != 0)
1025 goto fail2;
1026 }
1027
1028 if ((flags & RFMEM) == 0) {
1029 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
1030 if (vm2 == NULL) {
1031 error = ENOMEM;
1032 goto fail2;
1033 }
1034 if (!swap_reserve(mem_charged)) {
1035 /*
1036 * The swap reservation failed. The accounting
1037 * from the entries of the copied vm2 will be
1038 * subtracted in vmspace_free(), so force the
1039 * reservation there.
1040 */
1041 swap_reserve_force(mem_charged);
1042 error = ENOMEM;
1043 goto fail2;
1044 }
1045 } else
1046 vm2 = NULL;
1047
1048 /*
1049 * XXX: This is ugly; when we copy resource usage, we need to bump
1050 * per-cred resource counters.
1051 */
1052 newproc->p_ucred = crcowget(td->td_ucred);
1053
1054 /*
1055 * Initialize resource accounting for the child process.
1056 */
1057 error = racct_proc_fork(p1, newproc);
1058 if (error != 0) {
1059 error = EAGAIN;
1060 goto fail1;
1061 }
1062
1063 #ifdef MAC
1064 mac_proc_init(newproc);
1065 #endif
1066 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1067 STAILQ_INIT(&newproc->p_ktr);
1068
1069 /*
1070 * Increment the count of procs running with this uid. Don't allow
1071 * a nonprivileged user to exceed their current limit.
1072 */
1073 cred = td->td_ucred;
1074 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1075 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1076 goto fail0;
1077 chgproccnt(cred->cr_ruidinfo, 1, 0);
1078 }
1079
1080 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1081 error = 0;
1082 goto cleanup;
1083 fail0:
1084 error = EAGAIN;
1085 #ifdef MAC
1086 mac_proc_destroy(newproc);
1087 #endif
1088 racct_proc_exit(newproc);
1089 fail1:
1090 proc_unset_cred(newproc, false);
1091 fail2:
1092 if (vm2 != NULL)
1093 vmspace_free(vm2);
1094 uma_zfree(proc_zone, newproc);
1095 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1096 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1097 fdrop(fp_procdesc, td);
1098 }
1099 atomic_add_int(&nprocs, -1);
1100 cleanup:
1101 if (killsx_locked)
1102 sx_sunlock(&pg->pg_killsx);
1103 if (singlethreaded) {
1104 PROC_LOCK(p1);
1105 thread_single_end(p1, SINGLE_BOUNDARY);
1106 PROC_UNLOCK(p1);
1107 }
1108 if (error != 0)
1109 pause("fork", hz / 2);
1110 return (error);
1111 }
1112
1113 /*
1114 * Handle the return of a child process from fork1(). This function
1115 * is called from the MD fork_trampoline() entry point.
1116 */
1117 void
fork_exit(void (* callout)(void *,struct trapframe *),void * arg,struct trapframe * frame)1118 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1119 struct trapframe *frame)
1120 {
1121 struct proc *p;
1122 struct thread *td;
1123 struct thread *dtd;
1124
1125 kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED);
1126
1127 td = curthread;
1128 p = td->td_proc;
1129 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1130
1131 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1132 td, td_get_sched(td), p->p_pid, td->td_name);
1133
1134 sched_fork_exit(td);
1135
1136 /*
1137 * Processes normally resume in mi_switch() after being
1138 * cpu_switch()'ed to, but when children start up they arrive here
1139 * instead, so we must do much the same things as mi_switch() would.
1140 */
1141 if ((dtd = PCPU_GET(deadthread))) {
1142 PCPU_SET(deadthread, NULL);
1143 thread_stash(dtd);
1144 }
1145 thread_unlock(td);
1146
1147 /*
1148 * cpu_fork_kthread_handler intercepts this function call to
1149 * have this call a non-return function to stay in kernel mode.
1150 * initproc has its own fork handler, but it does return.
1151 */
1152 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1153 callout(arg, frame);
1154
1155 /*
1156 * Check if a kernel thread misbehaved and returned from its main
1157 * function.
1158 */
1159 if (p->p_flag & P_KPROC) {
1160 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1161 td->td_name, p->p_pid);
1162 kthread_exit();
1163 }
1164 mtx_assert(&Giant, MA_NOTOWNED);
1165
1166 /*
1167 * Now going to return to userland.
1168 */
1169
1170 if (p->p_sysent->sv_schedtail != NULL)
1171 (p->p_sysent->sv_schedtail)(td);
1172
1173 userret(td, frame);
1174 }
1175
1176 /*
1177 * Simplified back end of syscall(), used when returning from fork()
1178 * directly into user mode. This function is passed in to fork_exit()
1179 * as the first parameter and is called when returning to a new
1180 * userland process.
1181 */
1182 void
fork_return(struct thread * td,struct trapframe * frame)1183 fork_return(struct thread *td, struct trapframe *frame)
1184 {
1185 struct proc *p;
1186
1187 p = td->td_proc;
1188 if (td->td_dbgflags & TDB_STOPATFORK) {
1189 PROC_LOCK(p);
1190 if ((p->p_flag & P_TRACED) != 0) {
1191 /*
1192 * Inform the debugger if one is still present.
1193 */
1194 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1195 ptracestop(td, SIGSTOP, NULL);
1196 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1197 } else {
1198 /*
1199 * ... otherwise clear the request.
1200 */
1201 td->td_dbgflags &= ~TDB_STOPATFORK;
1202 }
1203 PROC_UNLOCK(p);
1204 } else if (p->p_flag & P_TRACED) {
1205 /*
1206 * This is the start of a new thread in a traced
1207 * process. Report a system call exit event.
1208 */
1209 PROC_LOCK(p);
1210 td->td_dbgflags |= TDB_SCX;
1211 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1212 (td->td_dbgflags & TDB_BORN) != 0)
1213 ptracestop(td, SIGTRAP, NULL);
1214 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1215 PROC_UNLOCK(p);
1216 }
1217
1218 /*
1219 * If the prison was killed mid-fork, die along with it.
1220 */
1221 if (!prison_isalive(td->td_ucred->cr_prison))
1222 exit1(td, 0, SIGKILL);
1223
1224 #ifdef KTRACE
1225 if (KTRPOINT(td, KTR_SYSRET))
1226 ktrsysret(td->td_sa.code, 0, 0);
1227 #endif
1228 }
1229
1230 static void
fork_init(void * arg __unused)1231 fork_init(void *arg __unused)
1232 {
1233 ast_register(TDA_VFORK, ASTR_ASTF_REQUIRED | ASTR_TDP, TDP_RFPPWAIT,
1234 ast_vfork);
1235 }
1236 SYSINIT(fork, SI_SUB_INTRINSIC, SI_ORDER_ANY, fork_init, NULL);
1237