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 "opt_ktrace.h"
38 #include "opt_kstack_pages.h"
39
40 #define EXTERR_CATEGORY EXTERR_CAT_FORK
41 #include <sys/systm.h>
42 #include <sys/acct.h>
43 #include <sys/bitstring.h>
44 #include <sys/eventhandler.h>
45 #include <sys/exterrvar.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/ktr.h>
52 #include <sys/ktrace.h>
53 #include <sys/sysctl.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/msan.h>
57 #include <sys/mutex.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procdesc.h>
61 #include <sys/ptrace.h>
62 #include <sys/racct.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sched.h>
65 #include <sys/sdt.h>
66 #include <sys/signalvar.h>
67 #include <sys/sx.h>
68 #include <sys/syscall.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/vmmeter.h>
72 #include <sys/vnode.h>
73 #include <sys/unistd.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 (EXTERROR(EINVAL, "Kernel-only flags %#jx", uap->flags));
170 /* RFSPAWN must not appear with others */
171 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
172 return (EXTERROR(EINVAL, "RFSPAWN must be the only flag %#jx",
173 uap->flags));
174
175 AUDIT_ARG_FFLAGS(uap->flags);
176 bzero(&fr, sizeof(fr));
177 if ((uap->flags & RFSPAWN) != 0) {
178 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
179 fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
180 } else {
181 fr.fr_flags = uap->flags;
182 }
183 fr.fr_pidp = &pid;
184 error = fork1(td, &fr);
185 if (error == 0) {
186 td->td_retval[0] = pid;
187 td->td_retval[1] = 0;
188 }
189 return (error);
190 }
191
192 int
sys_pdrfork(struct thread * td,struct pdrfork_args * uap)193 sys_pdrfork(struct thread *td, struct pdrfork_args *uap)
194 {
195 struct fork_req fr;
196 int error, fd, pid;
197
198 bzero(&fr, sizeof(fr));
199 fd = -1;
200
201 AUDIT_ARG_FFLAGS(uap->pdflags);
202 AUDIT_ARG_CMD(uap->rfflags);
203
204 if ((uap->rfflags & (RFSTOPPED | RFHIGHPID)) != 0)
205 return (EXTERROR(EINVAL,
206 "Kernel-only flags %#jx", uap->rfflags));
207
208 /* RFSPAWN must not appear with others */
209 if ((uap->rfflags & RFSPAWN) != 0) {
210 if (uap->rfflags != RFSPAWN)
211 return (EXTERROR(EINVAL,
212 "RFSPAWN must be the only flag %#jx",
213 uap->rfflags));
214 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPROCDESC;
215 fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
216 } else {
217 if ((uap->rfflags & (RFPROC | RFPROCDESC)) !=
218 (RFPROC | RFPROCDESC)) {
219 return (EXTERROR(EINVAL,
220 "RFPROC|RFPROCDESC required %#jx", uap->rfflags));
221 }
222 fr.fr_flags = uap->rfflags;
223 }
224
225 fr.fr_pidp = &pid;
226 fr.fr_pd_fd = &fd;
227 fr.fr_pd_flags = uap->pdflags;
228 error = fork1(td, &fr);
229 if (error == 0) {
230 td->td_retval[0] = pid;
231 td->td_retval[1] = 0;
232 if ((fr.fr_flags & (RFPROC | RFPROCDESC)) ==
233 (RFPROC | RFPROCDESC) || uap->rfflags == RFSPAWN)
234 error = copyout(&fd, uap->fdp, sizeof(fd));
235 }
236 return (error);
237 }
238
239 int __exclusive_cache_line nprocs = 1; /* process 0 */
240 int lastpid = 0;
241 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
242 "Last used PID");
243
244 /*
245 * Random component to lastpid generation. We mix in a random factor to make
246 * it a little harder to predict. We sanity check the modulus value to avoid
247 * doing it in critical paths. Don't let it be too small or we pointlessly
248 * waste randomness entropy, and don't let it be impossibly large. Using a
249 * modulus that is too big causes a LOT more process table scans and slows
250 * down fork processing as the pidchecked caching is defeated.
251 */
252 static int randompid = 0;
253
254 static int
sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)255 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
256 {
257 int error, pid;
258
259 error = sysctl_wire_old_buffer(req, sizeof(int));
260 if (error != 0)
261 return(error);
262 sx_xlock(&allproc_lock);
263 pid = randompid;
264 error = sysctl_handle_int(oidp, &pid, 0, req);
265 if (error == 0 && req->newptr != NULL) {
266 if (pid == 0)
267 randompid = 0;
268 else if (pid == 1)
269 /* generate a random PID modulus between 100 and 1123 */
270 randompid = 100 + arc4random() % 1024;
271 else if (pid < 0 || pid > pid_max - 100)
272 /* out of range */
273 randompid = pid_max - 100;
274 else if (pid < 100)
275 /* Make it reasonable */
276 randompid = 100;
277 else
278 randompid = pid;
279 }
280 sx_xunlock(&allproc_lock);
281 return (error);
282 }
283
284 SYSCTL_PROC(_kern, OID_AUTO, randompid,
285 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
286 sysctl_kern_randompid, "I",
287 "Random PID modulus. Special values: 0: disable, 1: choose random value");
288
289 extern bitstr_t proc_id_pidmap;
290 extern bitstr_t proc_id_grpidmap;
291 extern bitstr_t proc_id_sessidmap;
292 extern bitstr_t proc_id_reapmap;
293
294 /*
295 * Find an unused process ID
296 *
297 * If RFHIGHPID is set (used during system boot), do not allocate
298 * low-numbered pids.
299 */
300 static int
fork_findpid(int flags)301 fork_findpid(int flags)
302 {
303 pid_t result;
304 int trypid, random;
305
306 /*
307 * Avoid calling arc4random with procid_lock held.
308 */
309 random = 0;
310 if (__predict_false(randompid))
311 random = arc4random() % randompid;
312
313 mtx_lock(&procid_lock);
314
315 trypid = lastpid + 1;
316 if (flags & RFHIGHPID) {
317 if (trypid < 10)
318 trypid = 10;
319 } else {
320 trypid += random;
321 }
322 retry:
323 if (trypid >= pid_max)
324 trypid = 2;
325
326 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
327 if (result == -1) {
328 KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
329 trypid = 2;
330 goto retry;
331 }
332 if (bit_test(&proc_id_grpidmap, result) ||
333 bit_test(&proc_id_sessidmap, result) ||
334 bit_test(&proc_id_reapmap, result)) {
335 trypid = result + 1;
336 goto retry;
337 }
338
339 /*
340 * RFHIGHPID does not mess with the lastpid counter during boot.
341 */
342 if ((flags & RFHIGHPID) == 0)
343 lastpid = result;
344
345 bit_set(&proc_id_pidmap, result);
346 mtx_unlock(&procid_lock);
347
348 return (result);
349 }
350
351 static int
fork_norfproc(struct thread * td,int flags)352 fork_norfproc(struct thread *td, int flags)
353 {
354 struct proc *p1;
355 int error;
356
357 KASSERT((flags & RFPROC) == 0,
358 ("fork_norfproc called with RFPROC set"));
359 p1 = td->td_proc;
360
361 /*
362 * Quiesce other threads if necessary. If RFMEM is not specified we
363 * must ensure that other threads do not concurrently create a second
364 * process sharing the vmspace, see vmspace_unshare().
365 */
366 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
367 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
368 PROC_LOCK(p1);
369 if (thread_single(p1, SINGLE_BOUNDARY)) {
370 PROC_UNLOCK(p1);
371 return (ERESTART);
372 }
373 PROC_UNLOCK(p1);
374 }
375
376 error = vm_forkproc(td, NULL, NULL, NULL, flags);
377 if (error != 0)
378 goto fail;
379
380 /*
381 * Close all file descriptors.
382 */
383 if ((flags & RFCFDG) != 0) {
384 struct filedesc *fdtmp;
385 struct pwddesc *pdtmp;
386
387 pdtmp = pdinit(td->td_proc->p_pd, false);
388 fdtmp = fdinit();
389 pdescfree(td);
390 fdescfree(td);
391 p1->p_fd = fdtmp;
392 p1->p_pd = pdtmp;
393 }
394
395 /*
396 * Unshare file descriptors (from parent).
397 */
398 if ((flags & RFFDG) != 0) {
399 fdunshare(td);
400 pdunshare(td);
401 }
402
403 fail:
404 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
405 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
406 PROC_LOCK(p1);
407 thread_single_end(p1, SINGLE_BOUNDARY);
408 PROC_UNLOCK(p1);
409 }
410 return (error);
411 }
412
413 static void
do_fork(struct thread * td,struct fork_req * fr,struct proc * p2,struct thread * td2,struct vmspace * vm2,struct file * fp_procdesc)414 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
415 struct vmspace *vm2, struct file *fp_procdesc)
416 {
417 struct proc *p1, *pptr;
418 struct filedesc *fd;
419 struct filedesc_to_leader *fdtol;
420 struct pwddesc *pd;
421 struct sigacts *newsigacts;
422
423 p1 = td->td_proc;
424
425 PROC_LOCK(p1);
426 bcopy(&p1->p_startcopy, &p2->p_startcopy,
427 __rangeof(struct proc, p_startcopy, p_endcopy));
428 pargs_hold(p2->p_args);
429 PROC_UNLOCK(p1);
430
431 bzero(&p2->p_startzero,
432 __rangeof(struct proc, p_startzero, p_endzero));
433
434 /* Tell the prison that we exist. */
435 prison_proc_hold(p2->p_ucred->cr_prison);
436
437 p2->p_state = PRS_NEW; /* protect against others */
438 p2->p_pid = fork_findpid(fr->fr_flags);
439 AUDIT_ARG_PID(p2->p_pid);
440 TSFORK(p2->p_pid, p1->p_pid);
441
442 sx_xlock(&allproc_lock);
443 LIST_INSERT_HEAD(&allproc, p2, p_list);
444 allproc_gen++;
445 prison_proc_link(p2->p_ucred->cr_prison, p2);
446 sx_xunlock(&allproc_lock);
447
448 sx_xlock(PIDHASHLOCK(p2->p_pid));
449 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
450 sx_xunlock(PIDHASHLOCK(p2->p_pid));
451
452 tidhash_add(td2);
453
454 /*
455 * Malloc things while we don't hold any locks.
456 */
457 if (fr->fr_flags & RFSIGSHARE)
458 newsigacts = NULL;
459 else
460 newsigacts = sigacts_alloc();
461
462 /*
463 * Copy filedesc.
464 */
465 if (fr->fr_flags & RFCFDG) {
466 pd = pdinit(p1->p_pd, false);
467 fd = fdinit();
468 fdtol = NULL;
469 } else if (fr->fr_flags & RFFDG) {
470 if (fr->fr_flags2 & FR2_SHARE_PATHS)
471 pd = pdshare(p1->p_pd);
472 else
473 pd = pdcopy(p1->p_pd);
474 fd = fdcopy(p1->p_fd, p2);
475 fdtol = NULL;
476 } else {
477 if (fr->fr_flags2 & FR2_SHARE_PATHS)
478 pd = pdcopy(p1->p_pd);
479 else
480 pd = pdshare(p1->p_pd);
481 fd = fdshare(p1->p_fd);
482 if (p1->p_fdtol == NULL)
483 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
484 p1->p_leader);
485 if ((fr->fr_flags & RFTHREAD) != 0) {
486 /*
487 * Shared file descriptor table, and shared
488 * process leaders.
489 */
490 fdtol = filedesc_to_leader_share(p1->p_fdtol, p1->p_fd);
491 } else {
492 /*
493 * Shared file descriptor table, and different
494 * process leaders.
495 */
496 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
497 p1->p_fd, p2);
498 }
499 }
500 /*
501 * Make a proc table entry for the new process.
502 * Start by zeroing the section of proc that is zero-initialized,
503 * then copy the section that is copied directly from the parent.
504 */
505
506 PROC_LOCK(p2);
507 PROC_LOCK(p1);
508
509 bzero(&td2->td_startzero,
510 __rangeof(struct thread, td_startzero, td_endzero));
511
512 bcopy(&td->td_startcopy, &td2->td_startcopy,
513 __rangeof(struct thread, td_startcopy, td_endcopy));
514
515 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
516 td2->td_sigstk = td->td_sigstk;
517 td2->td_flags = TDF_INMEM;
518 td2->td_lend_user_pri = PRI_MAX;
519
520 #ifdef VIMAGE
521 td2->td_vnet = NULL;
522 td2->td_vnet_lpush = NULL;
523 #endif
524
525 /*
526 * Allow the scheduler to initialize the child.
527 */
528 thread_lock(td);
529 sched_fork(td, td2);
530 /*
531 * Request AST to check for TDP_RFPPWAIT. Do it here
532 * to avoid calling thread_lock() again.
533 */
534 if ((fr->fr_flags & RFPPWAIT) != 0)
535 ast_sched_locked(td, TDA_VFORK);
536 thread_unlock(td);
537
538 /*
539 * Duplicate sub-structures as needed.
540 * Increase reference counts on shared objects.
541 */
542 p2->p_flag = P_INMEM;
543 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
544 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
545 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
546 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS |
547 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC | P2_LOGSIGEXIT_CTL |
548 P2_LOGSIGEXIT_ENABLE);
549 p2->p_swtick = ticks;
550 if (p1->p_flag & P_PROFIL)
551 startprofclock(p2);
552
553 if (fr->fr_flags & RFSIGSHARE) {
554 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
555 } else {
556 sigacts_copy(newsigacts, p1->p_sigacts);
557 p2->p_sigacts = newsigacts;
558 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
559 mtx_lock(&p2->p_sigacts->ps_mtx);
560 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
561 sig_drop_caught(p2);
562 if ((fr->fr_flags2 & FR2_KPROC) != 0)
563 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
564 mtx_unlock(&p2->p_sigacts->ps_mtx);
565 }
566 }
567
568 if (fr->fr_flags & RFTSIGZMB)
569 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
570 else if (fr->fr_flags & RFLINUXTHPN)
571 p2->p_sigparent = SIGUSR1;
572 else
573 p2->p_sigparent = SIGCHLD;
574
575 if ((fr->fr_flags2 & FR2_KPROC) != 0) {
576 p2->p_flag |= P_SYSTEM | P_KPROC;
577 td2->td_pflags |= TDP_KTHREAD;
578 }
579
580 p2->p_textvp = p1->p_textvp;
581 p2->p_textdvp = p1->p_textdvp;
582 p2->p_fd = fd;
583 p2->p_fdtol = fdtol;
584 p2->p_pd = pd;
585
586 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
587 p2->p_flag |= P_PROTECTED;
588 p2->p_flag2 |= P2_INHERIT_PROTECTED;
589 }
590
591 /*
592 * p_limit is copy-on-write. Bump its refcount.
593 */
594 lim_fork(p1, p2);
595
596 thread_cow_get_proc(td2, p2);
597
598 pstats_fork(p1->p_stats, p2->p_stats);
599
600 PROC_UNLOCK(p1);
601 PROC_UNLOCK(p2);
602
603 /*
604 * Bump references to the text vnode and directory, and copy
605 * the hardlink name.
606 */
607 if (p2->p_textvp != NULL)
608 vrefact(p2->p_textvp);
609 if (p2->p_textdvp != NULL)
610 vrefact(p2->p_textdvp);
611 p2->p_binname = p1->p_binname == NULL ? NULL :
612 strdup(p1->p_binname, M_PARGS);
613
614 /*
615 * Set up linkage for kernel based threading.
616 */
617 if ((fr->fr_flags & RFTHREAD) != 0) {
618 mtx_lock(&ppeers_lock);
619 p2->p_peers = p1->p_peers;
620 p1->p_peers = p2;
621 p2->p_leader = p1->p_leader;
622 mtx_unlock(&ppeers_lock);
623 PROC_LOCK(p1->p_leader);
624 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
625 PROC_UNLOCK(p1->p_leader);
626 /*
627 * The task leader is exiting, so process p1 is
628 * going to be killed shortly. Since p1 obviously
629 * isn't dead yet, we know that the leader is either
630 * sending SIGKILL's to all the processes in this
631 * task or is sleeping waiting for all the peers to
632 * exit. We let p1 complete the fork, but we need
633 * to go ahead and kill the new process p2 since
634 * the task leader may not get a chance to send
635 * SIGKILL to it. We leave it on the list so that
636 * the task leader will wait for this new process
637 * to commit suicide.
638 */
639 PROC_LOCK(p2);
640 kern_psignal(p2, SIGKILL);
641 PROC_UNLOCK(p2);
642 } else
643 PROC_UNLOCK(p1->p_leader);
644 } else {
645 p2->p_peers = NULL;
646 p2->p_leader = p2;
647 }
648
649 sx_xlock(&proctree_lock);
650 PGRP_LOCK(p1->p_pgrp);
651 PROC_LOCK(p2);
652 PROC_LOCK(p1);
653
654 /*
655 * Preserve some more flags in subprocess. P_PROFIL has already
656 * been preserved.
657 */
658 p2->p_flag |= p1->p_flag & P_SUGID;
659 td2->td_pflags |= td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK);
660 td2->td_pflags2 |= td->td_pflags2 & TDP2_UEXTERR;
661 if (p1->p_flag & P_CONTROLT) {
662 SESS_LOCK(p1->p_session);
663 if (p1->p_session->s_ttyvp != NULL)
664 p2->p_flag |= P_CONTROLT;
665 SESS_UNLOCK(p1->p_session);
666 }
667 if (fr->fr_flags & RFPPWAIT)
668 p2->p_flag |= P_PPWAIT;
669
670 p2->p_pgrp = p1->p_pgrp;
671 LIST_INSERT_AFTER(p1, p2, p_pglist);
672 PGRP_UNLOCK(p1->p_pgrp);
673 LIST_INIT(&p2->p_children);
674 LIST_INIT(&p2->p_orphans);
675
676 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
677
678 /*
679 * This begins the section where we must prevent the parent
680 * from being swapped.
681 */
682 _PHOLD(p1);
683 PROC_UNLOCK(p1);
684
685 /*
686 * Attach the new process to its parent.
687 *
688 * If RFNOWAIT is set, the newly created process becomes a child
689 * of init. This effectively disassociates the child from the
690 * parent.
691 */
692 if ((fr->fr_flags & RFNOWAIT) != 0) {
693 pptr = p1->p_reaper;
694 p2->p_reaper = pptr;
695 } else {
696 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
697 p1 : p1->p_reaper;
698 pptr = p1;
699 }
700 p2->p_pptr = pptr;
701 p2->p_oppid = pptr->p_pid;
702 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
703 LIST_INIT(&p2->p_reaplist);
704 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
705 if (p2->p_reaper == p1 && p1 != initproc) {
706 p2->p_reapsubtree = p2->p_pid;
707 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
708 }
709 sx_xunlock(&proctree_lock);
710
711 /* Inform accounting that we have forked. */
712 p2->p_acflag = AFORK;
713 PROC_UNLOCK(p2);
714
715 #ifdef KTRACE
716 ktrprocfork(p1, p2);
717 #endif
718
719 /*
720 * Finish creating the child process. It will return via a different
721 * execution path later. (ie: directly into user mode)
722 */
723 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
724
725 if (fr->fr_flags == (RFFDG | RFPROC)) {
726 VM_CNT_INC(v_forks);
727 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
728 p2->p_vmspace->vm_ssize);
729 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
730 VM_CNT_INC(v_vforks);
731 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
732 p2->p_vmspace->vm_ssize);
733 } else if (p1 == &proc0) {
734 VM_CNT_INC(v_kthreads);
735 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
736 p2->p_vmspace->vm_ssize);
737 } else {
738 VM_CNT_INC(v_rforks);
739 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
740 p2->p_vmspace->vm_ssize);
741 }
742
743 /*
744 * Associate the process descriptor with the process before anything
745 * can happen that might cause that process to need the descriptor.
746 * However, don't do this until after fork(2) can no longer fail.
747 */
748 if (fr->fr_flags & RFPROCDESC)
749 procdesc_new(p2, fr->fr_pd_flags);
750
751 /*
752 * Both processes are set up, now check if any loadable modules want
753 * to adjust anything.
754 */
755 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
756
757 /*
758 * Set the child start time and mark the process as being complete.
759 */
760 PROC_LOCK(p2);
761 PROC_LOCK(p1);
762 microuptime(&p2->p_stats->p_start);
763 PROC_SLOCK(p2);
764 p2->p_state = PRS_NORMAL;
765 PROC_SUNLOCK(p2);
766
767 #ifdef KDTRACE_HOOKS
768 /*
769 * Tell the DTrace fasttrap provider about the new process so that any
770 * tracepoints inherited from the parent can be removed. We have to do
771 * this only after p_state is PRS_NORMAL since the fasttrap module will
772 * use pfind() later on.
773 */
774 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
775 dtrace_fasttrap_fork(p1, p2);
776 #endif
777 if (fr->fr_flags & RFPPWAIT) {
778 td->td_pflags |= TDP_RFPPWAIT;
779 td->td_rfppwait_p = p2;
780 td->td_dbgflags |= TDB_VFORK;
781 }
782 PROC_UNLOCK(p2);
783
784 /*
785 * Tell any interested parties about the new process.
786 */
787 knote_fork(p1->p_klist, p2->p_pid);
788
789 /*
790 * Now can be swapped.
791 */
792 _PRELE(p1);
793 PROC_UNLOCK(p1);
794 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
795
796 if (fr->fr_flags & RFPROCDESC) {
797 procdesc_finit(p2->p_procdesc, fp_procdesc);
798 fdrop(fp_procdesc, td);
799 }
800
801 /*
802 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
803 * synced with forks in progress so it is OK if we miss it
804 * if being set atm.
805 */
806 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
807 sx_xlock(&proctree_lock);
808 PROC_LOCK(p2);
809
810 /*
811 * p1->p_ptevents & p1->p_pptr are protected by both
812 * process and proctree locks for modifications,
813 * so owning proctree_lock allows the race-free read.
814 */
815 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
816 /*
817 * Arrange for debugger to receive the fork event.
818 *
819 * We can report PL_FLAG_FORKED regardless of
820 * P_FOLLOWFORK settings, but it does not make a sense
821 * for runaway child.
822 */
823 td->td_dbgflags |= TDB_FORK;
824 td->td_dbg_forked = p2->p_pid;
825 td2->td_dbgflags |= TDB_STOPATFORK;
826 proc_set_traced(p2, true);
827 CTR2(KTR_PTRACE,
828 "do_fork: attaching to new child pid %d: oppid %d",
829 p2->p_pid, p2->p_oppid);
830 proc_reparent(p2, p1->p_pptr, false);
831 }
832 PROC_UNLOCK(p2);
833 sx_xunlock(&proctree_lock);
834 }
835
836 racct_proc_fork_done(p2);
837
838 if ((fr->fr_flags & RFSTOPPED) == 0) {
839 if (fr->fr_pidp != NULL)
840 *fr->fr_pidp = p2->p_pid;
841 /*
842 * If RFSTOPPED not requested, make child runnable and
843 * add to run queue.
844 */
845 thread_lock(td2);
846 TD_SET_CAN_RUN(td2);
847 sched_add(td2, SRQ_BORING);
848 } else {
849 *fr->fr_procp = p2;
850 }
851 }
852
853 static void
ast_vfork(struct thread * td,int tda __unused)854 ast_vfork(struct thread *td, int tda __unused)
855 {
856 struct proc *p, *p2;
857
858 MPASS(td->td_pflags & TDP_RFPPWAIT);
859
860 p = td->td_proc;
861 /*
862 * Preserve synchronization semantics of vfork. If
863 * waiting for child to exec or exit, fork set
864 * P_PPWAIT on child, and there we sleep on our proc
865 * (in case of exit).
866 *
867 * Do it after the ptracestop() above is finished, to
868 * not block our debugger until child execs or exits
869 * to finish vfork wait.
870 */
871 td->td_pflags &= ~TDP_RFPPWAIT;
872 p2 = td->td_rfppwait_p;
873 again:
874 PROC_LOCK(p2);
875 while (p2->p_flag & P_PPWAIT) {
876 PROC_LOCK(p);
877 if (thread_suspend_check_needed()) {
878 PROC_UNLOCK(p2);
879 thread_suspend_check(0);
880 PROC_UNLOCK(p);
881 goto again;
882 } else {
883 PROC_UNLOCK(p);
884 }
885 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
886 }
887 PROC_UNLOCK(p2);
888
889 if (td->td_dbgflags & TDB_VFORK) {
890 PROC_LOCK(p);
891 if (p->p_ptevents & PTRACE_VFORK)
892 ptracestop(td, SIGTRAP, NULL);
893 td->td_dbgflags &= ~TDB_VFORK;
894 PROC_UNLOCK(p);
895 }
896 }
897
898 int
fork1(struct thread * td,struct fork_req * fr)899 fork1(struct thread *td, struct fork_req *fr)
900 {
901 struct proc *p1, *newproc;
902 struct thread *td2;
903 struct vmspace *vm2;
904 struct ucred *cred;
905 struct file *fp_procdesc;
906 struct pgrp *pg;
907 vm_ooffset_t mem_charged;
908 int error, nprocs_new;
909 static int curfail;
910 static struct timeval lastfail;
911 int flags, pages;
912 bool killsx_locked, singlethreaded;
913
914 flags = fr->fr_flags;
915 pages = fr->fr_pages;
916
917 if ((flags & RFSTOPPED) != 0)
918 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
919 else
920 MPASS(fr->fr_procp == NULL);
921
922 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
923 return (EXTERROR(EINVAL,
924 "Undef or unimplemented flags %#jx", flags));
925
926 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
927 return (EXTERROR(EINVAL,
928 "Signal value requires RFTSIGZMB", flags));
929
930 if ((flags & (RFFDG | RFCFDG)) == (RFFDG | RFCFDG))
931 return (EXTERROR(EINVAL, "Can not copy and clear"));
932
933 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
934 return (EXTERROR(EINVAL, "Invalid signal", RFTSIGNUM(flags)));
935
936 if ((flags & RFPROCDESC) != 0) {
937 if ((flags & RFPROC) == 0)
938 return (EXTERROR(EINVAL,
939 "Can not not create a process yet get a process descriptor"));
940
941 if (fr->fr_pd_fd == NULL)
942 return (EXTERROR(EINVAL,
943 "Must provide a place to put a procdesc if creating one"));
944
945 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
946 return (EXTERROR(EINVAL,
947 "Invallid pdflags at fork %#jx", fr->fr_pd_flags));
948 }
949
950 p1 = td->td_proc;
951
952 /*
953 * Here we don't create a new process, but we divorce
954 * certain parts of a process from itself.
955 */
956 if ((flags & RFPROC) == 0) {
957 if (fr->fr_procp != NULL)
958 *fr->fr_procp = NULL;
959 else if (fr->fr_pidp != NULL)
960 *fr->fr_pidp = 0;
961 return (fork_norfproc(td, flags));
962 }
963
964 fp_procdesc = NULL;
965 newproc = NULL;
966 vm2 = NULL;
967 killsx_locked = false;
968 singlethreaded = false;
969
970 /*
971 * Increment the nprocs resource before allocations occur.
972 * Although process entries are dynamically created, we still
973 * keep a global limit on the maximum number we will
974 * create. There are hard-limits as to the number of processes
975 * that can run, established by the KVA and memory usage for
976 * the process data.
977 *
978 * Don't allow a nonprivileged user to use the last ten
979 * processes; don't let root exceed the limit.
980 */
981 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
982 if (nprocs_new >= maxproc - 10) {
983 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
984 nprocs_new >= maxproc) {
985 error = EAGAIN;
986 sx_xlock(&allproc_lock);
987 if (ppsratecheck(&lastfail, &curfail, 1)) {
988 printf("maxproc limit exceeded by uid %u "
989 "(pid %d); see tuning(7) and "
990 "login.conf(5)\n",
991 td->td_ucred->cr_ruid, p1->p_pid);
992 }
993 sx_xunlock(&allproc_lock);
994 goto fail2;
995 }
996 }
997
998 /*
999 * If we are possibly multi-threaded, and there is a process
1000 * sending a signal to our group right now, ensure that our
1001 * other threads cannot be chosen for the signal queueing.
1002 * Otherwise, this might delay signal action, and make the new
1003 * child escape the signaling.
1004 */
1005 pg = p1->p_pgrp;
1006 if (p1->p_numthreads > 1) {
1007 if (sx_try_slock(&pg->pg_killsx) != 0) {
1008 killsx_locked = true;
1009 } else {
1010 PROC_LOCK(p1);
1011 if (thread_single(p1, SINGLE_BOUNDARY)) {
1012 PROC_UNLOCK(p1);
1013 error = ERESTART;
1014 goto fail2;
1015 }
1016 PROC_UNLOCK(p1);
1017 singlethreaded = true;
1018 }
1019 }
1020
1021 /*
1022 * Atomically check for signals and block processes from sending
1023 * a signal to our process group until the child is visible.
1024 */
1025 if (!killsx_locked && sx_slock_sig(&pg->pg_killsx) != 0) {
1026 error = ERESTART;
1027 goto fail2;
1028 }
1029 if (__predict_false(p1->p_pgrp != pg || sig_intr() != 0)) {
1030 /*
1031 * Either the process was moved to other process
1032 * group, or there is pending signal. sx_slock_sig()
1033 * does not check for signals if not sleeping for the
1034 * lock.
1035 */
1036 sx_sunlock(&pg->pg_killsx);
1037 killsx_locked = false;
1038 error = ERESTART;
1039 goto fail2;
1040 } else {
1041 killsx_locked = true;
1042 }
1043
1044 /*
1045 * If required, create a process descriptor in the parent first; we
1046 * will abandon it if something goes wrong. We don't finit() until
1047 * later.
1048 */
1049 if (flags & RFPROCDESC) {
1050 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
1051 fr->fr_pd_flags, fr->fr_pd_fcaps);
1052 if (error != 0)
1053 goto fail2;
1054 AUDIT_ARG_FD(*fr->fr_pd_fd);
1055 }
1056
1057 mem_charged = 0;
1058 if (pages == 0)
1059 pages = kstack_pages;
1060 /* Allocate new proc. */
1061 newproc = uma_zalloc(proc_zone, M_WAITOK);
1062 td2 = FIRST_THREAD_IN_PROC(newproc);
1063 if (td2 == NULL) {
1064 td2 = thread_alloc(pages);
1065 if (td2 == NULL) {
1066 error = ENOMEM;
1067 goto fail2;
1068 }
1069 proc_linkup(newproc, td2);
1070 } else {
1071 error = thread_recycle(td2, pages);
1072 if (error != 0)
1073 goto fail2;
1074 }
1075
1076 if ((flags & RFMEM) == 0) {
1077 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
1078 if (vm2 == NULL) {
1079 error = ENOMEM;
1080 goto fail2;
1081 }
1082 if (!swap_reserve(mem_charged)) {
1083 /*
1084 * The swap reservation failed. The accounting
1085 * from the entries of the copied vm2 will be
1086 * subtracted in vmspace_free(), so force the
1087 * reservation there.
1088 */
1089 swap_reserve_force(mem_charged);
1090 error = ENOMEM;
1091 goto fail2;
1092 }
1093 } else
1094 vm2 = NULL;
1095
1096 /*
1097 * XXX: This is ugly; when we copy resource usage, we need to bump
1098 * per-cred resource counters.
1099 */
1100 newproc->p_ucred = crcowget(td->td_ucred);
1101
1102 /*
1103 * Initialize resource accounting for the child process.
1104 */
1105 error = racct_proc_fork(p1, newproc);
1106 if (error != 0) {
1107 error = EAGAIN;
1108 goto fail1;
1109 }
1110
1111 #ifdef MAC
1112 mac_proc_init(newproc);
1113 #endif
1114
1115 /*
1116 * Increment the count of procs running with this uid. Don't allow
1117 * a nonprivileged user to exceed their current limit.
1118 */
1119 cred = td->td_ucred;
1120 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1121 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1122 goto fail0;
1123 chgproccnt(cred->cr_ruidinfo, 1, 0);
1124 }
1125
1126 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1127
1128 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1129 error = 0;
1130 goto cleanup;
1131 fail0:
1132 error = EAGAIN;
1133 #ifdef MAC
1134 mac_proc_destroy(newproc);
1135 #endif
1136 racct_proc_exit(newproc);
1137 fail1:
1138 proc_unset_cred(newproc, false);
1139 fail2:
1140 if (vm2 != NULL)
1141 vmspace_free(vm2);
1142 uma_zfree(proc_zone, newproc);
1143 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1144 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1145 fdrop(fp_procdesc, td);
1146 }
1147 atomic_add_int(&nprocs, -1);
1148 cleanup:
1149 if (killsx_locked)
1150 sx_sunlock(&pg->pg_killsx);
1151 if (singlethreaded) {
1152 PROC_LOCK(p1);
1153 thread_single_end(p1, SINGLE_BOUNDARY);
1154 PROC_UNLOCK(p1);
1155 }
1156 if (error != 0)
1157 pause("fork", hz / 2);
1158 return (error);
1159 }
1160
1161 /*
1162 * Handle the return of a child process from fork1(). This function
1163 * is called from the MD fork_trampoline() entry point.
1164 */
1165 void
fork_exit(void (* callout)(void *,struct trapframe *),void * arg,struct trapframe * frame)1166 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1167 struct trapframe *frame)
1168 {
1169 struct proc *p;
1170 struct thread *td;
1171 struct thread *dtd;
1172
1173 kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED);
1174
1175 td = curthread;
1176 p = td->td_proc;
1177 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1178
1179 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1180 td, td_get_sched(td), p->p_pid, td->td_name);
1181
1182 sched_fork_exit(td);
1183
1184 /*
1185 * Processes normally resume in mi_switch() after being
1186 * cpu_switch()'ed to, but when children start up they arrive here
1187 * instead, so we must do much the same things as mi_switch() would.
1188 */
1189 if ((dtd = PCPU_GET(deadthread))) {
1190 PCPU_SET(deadthread, NULL);
1191 thread_stash(dtd);
1192 }
1193 thread_unlock(td);
1194
1195 /*
1196 * cpu_fork_kthread_handler intercepts this function call to
1197 * have this call a non-return function to stay in kernel mode.
1198 * initproc has its own fork handler, but it does return.
1199 */
1200 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1201 callout(arg, frame);
1202
1203 /*
1204 * Check if a kernel thread misbehaved and returned from its main
1205 * function.
1206 */
1207 if (p->p_flag & P_KPROC) {
1208 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1209 td->td_name, p->p_pid);
1210 kthread_exit();
1211 }
1212 mtx_assert(&Giant, MA_NOTOWNED);
1213
1214 /*
1215 * Now going to return to userland.
1216 */
1217
1218 if (p->p_sysent->sv_schedtail != NULL)
1219 (p->p_sysent->sv_schedtail)(td);
1220
1221 userret(td, frame);
1222 }
1223
1224 /*
1225 * Simplified back end of syscall(), used when returning from fork()
1226 * directly into user mode. This function is passed in to fork_exit()
1227 * as the first parameter and is called when returning to a new
1228 * userland process.
1229 */
1230 void
fork_return(struct thread * td,struct trapframe * frame)1231 fork_return(struct thread *td, struct trapframe *frame)
1232 {
1233 struct proc *p;
1234
1235 p = td->td_proc;
1236 if (td->td_dbgflags & TDB_STOPATFORK) {
1237 PROC_LOCK(p);
1238 if ((p->p_flag & P_TRACED) != 0) {
1239 /*
1240 * Inform the debugger if one is still present.
1241 */
1242 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1243 ptracestop(td, SIGSTOP, NULL);
1244 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1245 } else {
1246 /*
1247 * ... otherwise clear the request.
1248 */
1249 td->td_dbgflags &= ~TDB_STOPATFORK;
1250 }
1251 PROC_UNLOCK(p);
1252 } else if (p->p_flag & P_TRACED) {
1253 /*
1254 * This is the start of a new thread in a traced
1255 * process. Report a system call exit event.
1256 */
1257 PROC_LOCK(p);
1258 td->td_dbgflags |= TDB_SCX;
1259 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1260 (td->td_dbgflags & TDB_BORN) != 0)
1261 ptracestop(td, SIGTRAP, NULL);
1262 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1263 PROC_UNLOCK(p);
1264 }
1265
1266 /*
1267 * If the prison was killed mid-fork, die along with it.
1268 */
1269 if (!prison_isalive(td->td_ucred->cr_prison))
1270 exit1(td, 0, SIGKILL);
1271
1272 #ifdef KTRACE
1273 if (KTRPOINT(td, KTR_SYSRET))
1274 ktrsysret(td->td_sa.code, 0, 0);
1275 #endif
1276 }
1277
1278 static void
fork_init(void * arg __unused)1279 fork_init(void *arg __unused)
1280 {
1281 ast_register(TDA_VFORK, ASTR_ASTF_REQUIRED | ASTR_TDP, TDP_RFPPWAIT,
1282 ast_vfork);
1283 }
1284 SYSINIT(fork, SI_SUB_INTRINSIC, SI_ORDER_ANY, fork_init, NULL);
1285