FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_fork.c
1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/10.4/sys/kern/kern_fork.c 321020 2017-07-15 17:25:40Z dchagin $");
39
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_kstack_pages.h"
43 #include "opt_procdesc.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/fcntl.h>
50 #include <sys/filedesc.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/kthread.h>
54 #include <sys/sysctl.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mutex.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procdesc.h>
61 #include <sys/pioctl.h>
62 #include <sys/ptrace.h>
63 #include <sys/racct.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sched.h>
66 #include <sys/syscall.h>
67 #include <sys/vmmeter.h>
68 #include <sys/vnode.h>
69 #include <sys/acct.h>
70 #include <sys/ktr.h>
71 #include <sys/ktrace.h>
72 #include <sys/unistd.h>
73 #include <sys/sdt.h>
74 #include <sys/sx.h>
75 #include <sys/sysent.h>
76 #include <sys/signalvar.h>
77
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80
81 #include <vm/vm.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_extern.h>
85 #include <vm/uma.h>
86
87 #ifdef KDTRACE_HOOKS
88 #include <sys/dtrace_bsd.h>
89 dtrace_fork_func_t dtrace_fasttrap_fork;
90 #endif
91
92 SDT_PROVIDER_DECLARE(proc);
93 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
94
95 #ifndef _SYS_SYSPROTO_H_
96 struct fork_args {
97 int dummy;
98 };
99 #endif
100
101 /* ARGSUSED */
102 int
103 sys_fork(struct thread *td, struct fork_args *uap)
104 {
105 int error;
106 struct proc *p2;
107
108 error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0);
109 if (error == 0) {
110 td->td_retval[0] = p2->p_pid;
111 td->td_retval[1] = 0;
112 }
113 return (error);
114 }
115
116 /* ARGUSED */
117 int
118 sys_pdfork(td, uap)
119 struct thread *td;
120 struct pdfork_args *uap;
121 {
122 #ifdef PROCDESC
123 int error, fd;
124 struct proc *p2;
125
126 /*
127 * It is necessary to return fd by reference because 0 is a valid file
128 * descriptor number, and the child needs to be able to distinguish
129 * itself from the parent using the return value.
130 */
131 error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2,
132 &fd, uap->flags);
133 if (error == 0) {
134 td->td_retval[0] = p2->p_pid;
135 td->td_retval[1] = 0;
136 error = copyout(&fd, uap->fdp, sizeof(fd));
137 }
138 return (error);
139 #else
140 return (ENOSYS);
141 #endif
142 }
143
144 /* ARGSUSED */
145 int
146 sys_vfork(struct thread *td, struct vfork_args *uap)
147 {
148 int error, flags;
149 struct proc *p2;
150
151 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
152 error = fork1(td, flags, 0, &p2, NULL, 0);
153 if (error == 0) {
154 td->td_retval[0] = p2->p_pid;
155 td->td_retval[1] = 0;
156 }
157 return (error);
158 }
159
160 int
161 sys_rfork(struct thread *td, struct rfork_args *uap)
162 {
163 struct proc *p2;
164 int error;
165
166 /* Don't allow kernel-only flags. */
167 if ((uap->flags & RFKERNELONLY) != 0)
168 return (EINVAL);
169
170 AUDIT_ARG_FFLAGS(uap->flags);
171 error = fork1(td, uap->flags, 0, &p2, NULL, 0);
172 if (error == 0) {
173 td->td_retval[0] = p2 ? p2->p_pid : 0;
174 td->td_retval[1] = 0;
175 }
176 return (error);
177 }
178
179 int nprocs = 1; /* process 0 */
180 int lastpid = 0;
181 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
182 "Last used PID");
183
184 /*
185 * Random component to lastpid generation. We mix in a random factor to make
186 * it a little harder to predict. We sanity check the modulus value to avoid
187 * doing it in critical paths. Don't let it be too small or we pointlessly
188 * waste randomness entropy, and don't let it be impossibly large. Using a
189 * modulus that is too big causes a LOT more process table scans and slows
190 * down fork processing as the pidchecked caching is defeated.
191 */
192 static int randompid = 0;
193
194 static int
195 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
196 {
197 int error, pid;
198
199 error = sysctl_wire_old_buffer(req, sizeof(int));
200 if (error != 0)
201 return(error);
202 sx_xlock(&allproc_lock);
203 pid = randompid;
204 error = sysctl_handle_int(oidp, &pid, 0, req);
205 if (error == 0 && req->newptr != NULL) {
206 if (pid < 0 || pid > pid_max - 100) /* out of range */
207 pid = pid_max - 100;
208 else if (pid < 2) /* NOP */
209 pid = 0;
210 else if (pid < 100) /* Make it reasonable */
211 pid = 100;
212 randompid = pid;
213 }
214 sx_xunlock(&allproc_lock);
215 return (error);
216 }
217
218 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
219 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
220
221 static int
222 fork_findpid(int flags)
223 {
224 struct proc *p;
225 int trypid;
226 static int pidchecked = 0;
227
228 /*
229 * Requires allproc_lock in order to iterate over the list
230 * of processes, and proctree_lock to access p_pgrp.
231 */
232 sx_assert(&allproc_lock, SX_LOCKED);
233 sx_assert(&proctree_lock, SX_LOCKED);
234
235 /*
236 * Find an unused process ID. We remember a range of unused IDs
237 * ready to use (from lastpid+1 through pidchecked-1).
238 *
239 * If RFHIGHPID is set (used during system boot), do not allocate
240 * low-numbered pids.
241 */
242 trypid = lastpid + 1;
243 if (flags & RFHIGHPID) {
244 if (trypid < 10)
245 trypid = 10;
246 } else {
247 if (randompid)
248 trypid += arc4random() % randompid;
249 }
250 retry:
251 /*
252 * If the process ID prototype has wrapped around,
253 * restart somewhat above 0, as the low-numbered procs
254 * tend to include daemons that don't exit.
255 */
256 if (trypid >= pid_max) {
257 trypid = trypid % pid_max;
258 if (trypid < 100)
259 trypid += 100;
260 pidchecked = 0;
261 }
262 if (trypid >= pidchecked) {
263 int doingzomb = 0;
264
265 pidchecked = PID_MAX;
266 /*
267 * Scan the active and zombie procs to check whether this pid
268 * is in use. Remember the lowest pid that's greater
269 * than trypid, so we can avoid checking for a while.
270 *
271 * Avoid reuse of the process group id, session id or
272 * the reaper subtree id. Note that for process group
273 * and sessions, the amount of reserved pids is
274 * limited by process limit. For the subtree ids, the
275 * id is kept reserved only while there is a
276 * non-reaped process in the subtree, so amount of
277 * reserved pids is limited by process limit times
278 * two.
279 */
280 p = LIST_FIRST(&allproc);
281 again:
282 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
283 while (p->p_pid == trypid ||
284 p->p_reapsubtree == trypid ||
285 (p->p_pgrp != NULL &&
286 (p->p_pgrp->pg_id == trypid ||
287 (p->p_session != NULL &&
288 p->p_session->s_sid == trypid)))) {
289 trypid++;
290 if (trypid >= pidchecked)
291 goto retry;
292 }
293 if (p->p_pid > trypid && pidchecked > p->p_pid)
294 pidchecked = p->p_pid;
295 if (p->p_pgrp != NULL) {
296 if (p->p_pgrp->pg_id > trypid &&
297 pidchecked > p->p_pgrp->pg_id)
298 pidchecked = p->p_pgrp->pg_id;
299 if (p->p_session != NULL &&
300 p->p_session->s_sid > trypid &&
301 pidchecked > p->p_session->s_sid)
302 pidchecked = p->p_session->s_sid;
303 }
304 }
305 if (!doingzomb) {
306 doingzomb = 1;
307 p = LIST_FIRST(&zombproc);
308 goto again;
309 }
310 }
311
312 /*
313 * RFHIGHPID does not mess with the lastpid counter during boot.
314 */
315 if (flags & RFHIGHPID)
316 pidchecked = 0;
317 else
318 lastpid = trypid;
319
320 return (trypid);
321 }
322
323 static int
324 fork_norfproc(struct thread *td, int flags)
325 {
326 int error;
327 struct proc *p1;
328
329 KASSERT((flags & RFPROC) == 0,
330 ("fork_norfproc called with RFPROC set"));
331 p1 = td->td_proc;
332
333 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
334 (flags & (RFCFDG | RFFDG))) {
335 PROC_LOCK(p1);
336 if (thread_single(p1, SINGLE_BOUNDARY)) {
337 PROC_UNLOCK(p1);
338 return (ERESTART);
339 }
340 PROC_UNLOCK(p1);
341 }
342
343 error = vm_forkproc(td, NULL, NULL, NULL, flags);
344 if (error)
345 goto fail;
346
347 /*
348 * Close all file descriptors.
349 */
350 if (flags & RFCFDG) {
351 struct filedesc *fdtmp;
352 fdtmp = fdinit(td->td_proc->p_fd);
353 fdescfree(td);
354 p1->p_fd = fdtmp;
355 }
356
357 /*
358 * Unshare file descriptors (from parent).
359 */
360 if (flags & RFFDG)
361 fdunshare(td);
362
363 fail:
364 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
365 (flags & (RFCFDG | RFFDG))) {
366 PROC_LOCK(p1);
367 thread_single_end(p1, SINGLE_BOUNDARY);
368 PROC_UNLOCK(p1);
369 }
370 return (error);
371 }
372
373 static void
374 do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
375 struct vmspace *vm2, int pdflags)
376 {
377 struct proc *p1, *pptr;
378 int p2_held, trypid;
379 struct filedesc *fd;
380 struct filedesc_to_leader *fdtol;
381 struct sigacts *newsigacts;
382
383 sx_assert(&proctree_lock, SX_SLOCKED);
384 sx_assert(&allproc_lock, SX_XLOCKED);
385
386 p2_held = 0;
387 p1 = td->td_proc;
388
389 trypid = fork_findpid(flags);
390
391 sx_sunlock(&proctree_lock);
392
393 p2->p_state = PRS_NEW; /* protect against others */
394 p2->p_pid = trypid;
395 AUDIT_ARG_PID(p2->p_pid);
396 LIST_INSERT_HEAD(&allproc, p2, p_list);
397 allproc_gen++;
398 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
399 tidhash_add(td2);
400 PROC_LOCK(p2);
401 PROC_LOCK(p1);
402
403 sx_xunlock(&allproc_lock);
404
405 bcopy(&p1->p_startcopy, &p2->p_startcopy,
406 __rangeof(struct proc, p_startcopy, p_endcopy));
407 p2->p_elf_machine = p1->p_elf_machine;
408 p2->p_elf_flags = p1->p_elf_flags;
409 pargs_hold(p2->p_args);
410 PROC_UNLOCK(p1);
411
412 bzero(&p2->p_startzero,
413 __rangeof(struct proc, p_startzero, p_endzero));
414 p2->p_treeflag = 0;
415 p2->p_filemon = NULL;
416 p2->p_ptevents = 0;
417
418 /* Tell the prison that we exist. */
419 prison_proc_hold(p2->p_ucred->cr_prison);
420
421 PROC_UNLOCK(p2);
422
423 /*
424 * Malloc things while we don't hold any locks.
425 */
426 if (flags & RFSIGSHARE)
427 newsigacts = NULL;
428 else
429 newsigacts = sigacts_alloc();
430
431 /*
432 * Copy filedesc.
433 */
434 if (flags & RFCFDG) {
435 fd = fdinit(p1->p_fd);
436 fdtol = NULL;
437 } else if (flags & RFFDG) {
438 fd = fdcopy(p1->p_fd);
439 fdtol = NULL;
440 } else {
441 fd = fdshare(p1->p_fd);
442 if (p1->p_fdtol == NULL)
443 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
444 p1->p_leader);
445 if ((flags & RFTHREAD) != 0) {
446 /*
447 * Shared file descriptor table, and shared
448 * process leaders.
449 */
450 fdtol = p1->p_fdtol;
451 FILEDESC_XLOCK(p1->p_fd);
452 fdtol->fdl_refcount++;
453 FILEDESC_XUNLOCK(p1->p_fd);
454 } else {
455 /*
456 * Shared file descriptor table, and different
457 * process leaders.
458 */
459 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
460 p1->p_fd, p2);
461 }
462 }
463 /*
464 * Make a proc table entry for the new process.
465 * Start by zeroing the section of proc that is zero-initialized,
466 * then copy the section that is copied directly from the parent.
467 */
468
469 PROC_LOCK(p2);
470 PROC_LOCK(p1);
471
472 bzero(&td2->td_startzero,
473 __rangeof(struct thread, td_startzero, td_endzero));
474 td2->td_su = NULL;
475 td2->td_sleeptimo = 0;
476
477 bcopy(&td->td_startcopy, &td2->td_startcopy,
478 __rangeof(struct thread, td_startcopy, td_endcopy));
479
480 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
481 td2->td_sigstk = td->td_sigstk;
482 td2->td_flags = TDF_INMEM;
483 td2->td_lend_user_pri = PRI_MAX;
484 td2->td_dbg_sc_code = td->td_dbg_sc_code;
485 td2->td_dbg_sc_narg = td->td_dbg_sc_narg;
486
487 #ifdef VIMAGE
488 td2->td_vnet = NULL;
489 td2->td_vnet_lpush = NULL;
490 #endif
491
492 /*
493 * Allow the scheduler to initialize the child.
494 */
495 thread_lock(td);
496 sched_fork(td, td2);
497 thread_unlock(td);
498
499 /*
500 * Duplicate sub-structures as needed.
501 * Increase reference counts on shared objects.
502 */
503 p2->p_flag = P_INMEM;
504 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC);
505 p2->p_swtick = ticks;
506 if (p1->p_flag & P_PROFIL)
507 startprofclock(p2);
508 td2->td_ucred = crhold(p2->p_ucred);
509
510 if (flags & RFSIGSHARE) {
511 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
512 } else {
513 sigacts_copy(newsigacts, p1->p_sigacts);
514 p2->p_sigacts = newsigacts;
515 }
516
517 if (flags & RFTSIGZMB)
518 p2->p_sigparent = RFTSIGNUM(flags);
519 else if (flags & RFLINUXTHPN)
520 p2->p_sigparent = SIGUSR1;
521 else
522 p2->p_sigparent = SIGCHLD;
523
524 p2->p_textvp = p1->p_textvp;
525 p2->p_fd = fd;
526 p2->p_fdtol = fdtol;
527
528 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
529 p2->p_flag |= P_PROTECTED;
530 p2->p_flag2 |= P2_INHERIT_PROTECTED;
531 }
532
533 /*
534 * p_limit is copy-on-write. Bump its refcount.
535 */
536 lim_fork(p1, p2);
537
538 pstats_fork(p1->p_stats, p2->p_stats);
539
540 PROC_UNLOCK(p1);
541 PROC_UNLOCK(p2);
542
543 /* Bump references to the text vnode (for procfs). */
544 if (p2->p_textvp)
545 vref(p2->p_textvp);
546
547 /*
548 * Set up linkage for kernel based threading.
549 */
550 if ((flags & RFTHREAD) != 0) {
551 mtx_lock(&ppeers_lock);
552 p2->p_peers = p1->p_peers;
553 p1->p_peers = p2;
554 p2->p_leader = p1->p_leader;
555 mtx_unlock(&ppeers_lock);
556 PROC_LOCK(p1->p_leader);
557 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
558 PROC_UNLOCK(p1->p_leader);
559 /*
560 * The task leader is exiting, so process p1 is
561 * going to be killed shortly. Since p1 obviously
562 * isn't dead yet, we know that the leader is either
563 * sending SIGKILL's to all the processes in this
564 * task or is sleeping waiting for all the peers to
565 * exit. We let p1 complete the fork, but we need
566 * to go ahead and kill the new process p2 since
567 * the task leader may not get a chance to send
568 * SIGKILL to it. We leave it on the list so that
569 * the task leader will wait for this new process
570 * to commit suicide.
571 */
572 PROC_LOCK(p2);
573 kern_psignal(p2, SIGKILL);
574 PROC_UNLOCK(p2);
575 } else
576 PROC_UNLOCK(p1->p_leader);
577 } else {
578 p2->p_peers = NULL;
579 p2->p_leader = p2;
580 }
581
582 sx_xlock(&proctree_lock);
583 PGRP_LOCK(p1->p_pgrp);
584 PROC_LOCK(p2);
585 PROC_LOCK(p1);
586
587 /*
588 * Preserve some more flags in subprocess. P_PROFIL has already
589 * been preserved.
590 */
591 p2->p_flag |= p1->p_flag & P_SUGID;
592 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
593 SESS_LOCK(p1->p_session);
594 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
595 p2->p_flag |= P_CONTROLT;
596 SESS_UNLOCK(p1->p_session);
597 if (flags & RFPPWAIT)
598 p2->p_flag |= P_PPWAIT;
599
600 p2->p_pgrp = p1->p_pgrp;
601 LIST_INSERT_AFTER(p1, p2, p_pglist);
602 PGRP_UNLOCK(p1->p_pgrp);
603 LIST_INIT(&p2->p_children);
604 LIST_INIT(&p2->p_orphans);
605
606 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
607
608 /*
609 * If PF_FORK is set, the child process inherits the
610 * procfs ioctl flags from its parent.
611 */
612 if (p1->p_pfsflags & PF_FORK) {
613 p2->p_stops = p1->p_stops;
614 p2->p_pfsflags = p1->p_pfsflags;
615 }
616
617 /*
618 * This begins the section where we must prevent the parent
619 * from being swapped.
620 */
621 _PHOLD(p1);
622 PROC_UNLOCK(p1);
623
624 /*
625 * Attach the new process to its parent.
626 *
627 * If RFNOWAIT is set, the newly created process becomes a child
628 * of init. This effectively disassociates the child from the
629 * parent.
630 */
631 if ((flags & RFNOWAIT) != 0) {
632 pptr = p1->p_reaper;
633 p2->p_reaper = pptr;
634 } else {
635 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
636 p1 : p1->p_reaper;
637 pptr = p1;
638 }
639 p2->p_pptr = pptr;
640 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
641 LIST_INIT(&p2->p_reaplist);
642 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
643 if (p2->p_reaper == p1)
644 p2->p_reapsubtree = p2->p_pid;
645 else
646 p2->p_reapsubtree = p1->p_reapsubtree;
647 sx_xunlock(&proctree_lock);
648
649 /* Inform accounting that we have forked. */
650 p2->p_acflag = AFORK;
651 PROC_UNLOCK(p2);
652
653 #ifdef KTRACE
654 ktrprocfork(p1, p2);
655 #endif
656
657 /*
658 * Finish creating the child process. It will return via a different
659 * execution path later. (ie: directly into user mode)
660 */
661 vm_forkproc(td, p2, td2, vm2, flags);
662
663 if (flags == (RFFDG | RFPROC)) {
664 PCPU_INC(cnt.v_forks);
665 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
666 p2->p_vmspace->vm_ssize);
667 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
668 PCPU_INC(cnt.v_vforks);
669 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
670 p2->p_vmspace->vm_ssize);
671 } else if (p1 == &proc0) {
672 PCPU_INC(cnt.v_kthreads);
673 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
674 p2->p_vmspace->vm_ssize);
675 } else {
676 PCPU_INC(cnt.v_rforks);
677 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
678 p2->p_vmspace->vm_ssize);
679 }
680
681 #ifdef PROCDESC
682 /*
683 * Associate the process descriptor with the process before anything
684 * can happen that might cause that process to need the descriptor.
685 * However, don't do this until after fork(2) can no longer fail.
686 */
687 if (flags & RFPROCDESC)
688 procdesc_new(p2, pdflags);
689 #endif
690
691 /*
692 * Both processes are set up, now check if any loadable modules want
693 * to adjust anything.
694 */
695 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
696
697 /*
698 * Set the child start time and mark the process as being complete.
699 */
700 PROC_LOCK(p2);
701 PROC_LOCK(p1);
702 microuptime(&p2->p_stats->p_start);
703 PROC_SLOCK(p2);
704 p2->p_state = PRS_NORMAL;
705 PROC_SUNLOCK(p2);
706
707 #ifdef KDTRACE_HOOKS
708 /*
709 * Tell the DTrace fasttrap provider about the new process so that any
710 * tracepoints inherited from the parent can be removed. We have to do
711 * this only after p_state is PRS_NORMAL since the fasttrap module will
712 * use pfind() later on.
713 */
714 if ((flags & RFMEM) == 0 && dtrace_fasttrap_fork)
715 dtrace_fasttrap_fork(p1, p2);
716 #endif
717 if (p1->p_ptevents & PTRACE_FORK) {
718 /*
719 * Arrange for debugger to receive the fork event.
720 *
721 * We can report PL_FLAG_FORKED regardless of
722 * P_FOLLOWFORK settings, but it does not make a sense
723 * for runaway child.
724 */
725 td->td_dbgflags |= TDB_FORK;
726 td->td_dbg_forked = p2->p_pid;
727 td2->td_dbgflags |= TDB_STOPATFORK;
728 _PHOLD(p2);
729 p2_held = 1;
730 }
731 if (flags & RFPPWAIT) {
732 td->td_pflags |= TDP_RFPPWAIT;
733 td->td_rfppwait_p = p2;
734 td->td_dbgflags |= TDB_VFORK;
735 }
736 PROC_UNLOCK(p2);
737 if ((flags & RFSTOPPED) == 0) {
738 /*
739 * If RFSTOPPED not requested, make child runnable and
740 * add to run queue.
741 */
742 thread_lock(td2);
743 TD_SET_CAN_RUN(td2);
744 sched_add(td2, SRQ_BORING);
745 thread_unlock(td2);
746 }
747
748 /*
749 * Now can be swapped.
750 */
751 _PRELE(p1);
752 PROC_UNLOCK(p1);
753
754 /*
755 * Tell any interested parties about the new process.
756 */
757 knote_fork(&p1->p_klist, p2->p_pid);
758 SDT_PROBE3(proc, , , create, p2, p1, flags);
759
760 /*
761 * Wait until debugger is attached to child.
762 */
763 PROC_LOCK(p2);
764 while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
765 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
766 if (p2_held)
767 _PRELE(p2);
768 PROC_UNLOCK(p2);
769 }
770
771 int
772 fork1(struct thread *td, int flags, int pages, struct proc **procp,
773 int *procdescp, int pdflags)
774 {
775 struct proc *p1, *newproc;
776 struct thread *td2;
777 struct vmspace *vm2;
778 #ifdef PROCDESC
779 struct file *fp_procdesc;
780 #endif
781 vm_ooffset_t mem_charged;
782 int error, nprocs_new, ok;
783 static int curfail;
784 static struct timeval lastfail;
785
786 /* Check for the undefined or unimplemented flags. */
787 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
788 return (EINVAL);
789
790 /* Signal value requires RFTSIGZMB. */
791 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
792 return (EINVAL);
793
794 /* Can't copy and clear. */
795 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
796 return (EINVAL);
797
798 /* Check the validity of the signal number. */
799 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
800 return (EINVAL);
801
802 #ifdef PROCDESC
803 if ((flags & RFPROCDESC) != 0) {
804 /* Can't not create a process yet get a process descriptor. */
805 if ((flags & RFPROC) == 0)
806 return (EINVAL);
807
808 /* Must provide a place to put a procdesc if creating one. */
809 if (procdescp == NULL)
810 return (EINVAL);
811 }
812 #endif
813
814 p1 = td->td_proc;
815
816 /*
817 * Here we don't create a new process, but we divorce
818 * certain parts of a process from itself.
819 */
820 if ((flags & RFPROC) == 0) {
821 *procp = NULL;
822 return (fork_norfproc(td, flags));
823 }
824
825 #ifdef PROCDESC
826 fp_procdesc = NULL;
827 #endif
828 newproc = NULL;
829 vm2 = NULL;
830
831 /*
832 * Increment the nprocs resource before allocations occur.
833 * Although process entries are dynamically created, we still
834 * keep a global limit on the maximum number we will
835 * create. There are hard-limits as to the number of processes
836 * that can run, established by the KVA and memory usage for
837 * the process data.
838 *
839 * Don't allow a nonprivileged user to use the last ten
840 * processes; don't let root exceed the limit.
841 */
842 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
843 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
844 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
845 sx_xlock(&allproc_lock);
846 if (ppsratecheck(&lastfail, &curfail, 1)) {
847 printf("maxproc limit exceeded by uid %u (pid %d); "
848 "see tuning(7) and login.conf(5)\n",
849 td->td_ucred->cr_ruid, p1->p_pid);
850 }
851 sx_xunlock(&allproc_lock);
852 error = EAGAIN;
853 goto fail1;
854 }
855
856 #ifdef PROCDESC
857 /*
858 * If required, create a process descriptor in the parent first; we
859 * will abandon it if something goes wrong. We don't finit() until
860 * later.
861 */
862 if (flags & RFPROCDESC) {
863 error = falloc(td, &fp_procdesc, procdescp, 0);
864 if (error != 0)
865 goto fail1;
866 }
867 #endif
868
869 mem_charged = 0;
870 if (pages == 0)
871 pages = KSTACK_PAGES;
872 /* Allocate new proc. */
873 newproc = uma_zalloc(proc_zone, M_WAITOK);
874 td2 = FIRST_THREAD_IN_PROC(newproc);
875 if (td2 == NULL) {
876 td2 = thread_alloc(pages);
877 if (td2 == NULL) {
878 error = ENOMEM;
879 goto fail2;
880 }
881 proc_linkup(newproc, td2);
882 } else {
883 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
884 if (td2->td_kstack != 0)
885 vm_thread_dispose(td2);
886 if (!thread_alloc_stack(td2, pages)) {
887 error = ENOMEM;
888 goto fail2;
889 }
890 }
891 }
892
893 if ((flags & RFMEM) == 0) {
894 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
895 if (vm2 == NULL) {
896 error = ENOMEM;
897 goto fail2;
898 }
899 if (!swap_reserve(mem_charged)) {
900 /*
901 * The swap reservation failed. The accounting
902 * from the entries of the copied vm2 will be
903 * subtracted in vmspace_free(), so force the
904 * reservation there.
905 */
906 swap_reserve_force(mem_charged);
907 error = ENOMEM;
908 goto fail2;
909 }
910 } else
911 vm2 = NULL;
912
913 /*
914 * XXX: This is ugly; when we copy resource usage, we need to bump
915 * per-cred resource counters.
916 */
917 proc_set_cred_init(newproc, crhold(td->td_ucred));
918
919 /*
920 * Initialize resource accounting for the child process.
921 */
922 error = racct_proc_fork(p1, newproc);
923 if (error != 0) {
924 error = EAGAIN;
925 goto fail1;
926 }
927
928 #ifdef MAC
929 mac_proc_init(newproc);
930 #endif
931 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
932 STAILQ_INIT(&newproc->p_ktr);
933
934 /* We have to lock the process tree while we look for a pid. */
935 sx_slock(&proctree_lock);
936 sx_xlock(&allproc_lock);
937
938 /*
939 * Increment the count of procs running with this uid. Don't allow
940 * a nonprivileged user to exceed their current limit.
941 *
942 * XXXRW: Can we avoid privilege here if it's not needed?
943 */
944 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
945 if (error == 0)
946 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
947 else {
948 PROC_LOCK(p1);
949 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
950 lim_cur(p1, RLIMIT_NPROC));
951 PROC_UNLOCK(p1);
952 }
953 if (ok) {
954 do_fork(td, flags, newproc, td2, vm2, pdflags);
955
956 /*
957 * Return child proc pointer to parent.
958 */
959 *procp = newproc;
960 #ifdef PROCDESC
961 if (flags & RFPROCDESC) {
962 procdesc_finit(newproc->p_procdesc, fp_procdesc);
963 fdrop(fp_procdesc, td);
964 }
965 #endif
966 racct_proc_fork_done(newproc);
967 return (0);
968 }
969
970 error = EAGAIN;
971 sx_sunlock(&proctree_lock);
972 sx_xunlock(&allproc_lock);
973 #ifdef MAC
974 mac_proc_destroy(newproc);
975 #endif
976 racct_proc_exit(newproc);
977 fail1:
978 crfree(newproc->p_ucred);
979 newproc->p_ucred = NULL;
980 fail2:
981 if (vm2 != NULL)
982 vmspace_free(vm2);
983 uma_zfree(proc_zone, newproc);
984 #ifdef PROCDESC
985 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
986 fdclose(td, fp_procdesc, *procdescp);
987 fdrop(fp_procdesc, td);
988 }
989 #endif
990 atomic_add_int(&nprocs, -1);
991 pause("fork", hz / 2);
992 return (error);
993 }
994
995 /*
996 * Handle the return of a child process from fork1(). This function
997 * is called from the MD fork_trampoline() entry point.
998 */
999 void
1000 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1001 struct trapframe *frame)
1002 {
1003 struct proc *p;
1004 struct thread *td;
1005 struct thread *dtd;
1006
1007 td = curthread;
1008 p = td->td_proc;
1009 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1010
1011 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1012 td, td->td_sched, p->p_pid, td->td_name);
1013
1014 sched_fork_exit(td);
1015 /*
1016 * Processes normally resume in mi_switch() after being
1017 * cpu_switch()'ed to, but when children start up they arrive here
1018 * instead, so we must do much the same things as mi_switch() would.
1019 */
1020 if ((dtd = PCPU_GET(deadthread))) {
1021 PCPU_SET(deadthread, NULL);
1022 thread_stash(dtd);
1023 }
1024 thread_unlock(td);
1025
1026 /*
1027 * cpu_set_fork_handler intercepts this function call to
1028 * have this call a non-return function to stay in kernel mode.
1029 * initproc has its own fork handler, but it does return.
1030 */
1031 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1032 callout(arg, frame);
1033
1034 /*
1035 * Check if a kernel thread misbehaved and returned from its main
1036 * function.
1037 */
1038 if (p->p_flag & P_KTHREAD) {
1039 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1040 td->td_name, p->p_pid);
1041 kthread_exit();
1042 }
1043 mtx_assert(&Giant, MA_NOTOWNED);
1044
1045 if (p->p_sysent->sv_schedtail != NULL)
1046 (p->p_sysent->sv_schedtail)(td);
1047 }
1048
1049 /*
1050 * Simplified back end of syscall(), used when returning from fork()
1051 * directly into user mode. This function is passed in to fork_exit()
1052 * as the first parameter and is called when returning to a new
1053 * userland process.
1054 */
1055 void
1056 fork_return(struct thread *td, struct trapframe *frame)
1057 {
1058 struct proc *p, *dbg;
1059
1060 p = td->td_proc;
1061 if (td->td_dbgflags & TDB_STOPATFORK) {
1062 sx_xlock(&proctree_lock);
1063 PROC_LOCK(p);
1064 if (p->p_pptr->p_ptevents & PTRACE_FORK) {
1065 /*
1066 * If debugger still wants auto-attach for the
1067 * parent's children, do it now.
1068 */
1069 dbg = p->p_pptr->p_pptr;
1070 proc_set_traced(p, true);
1071 CTR2(KTR_PTRACE,
1072 "fork_return: attaching to new child pid %d: oppid %d",
1073 p->p_pid, p->p_oppid);
1074 proc_reparent(p, dbg);
1075 sx_xunlock(&proctree_lock);
1076 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1077 ptracestop(td, SIGSTOP, NULL);
1078 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1079 } else {
1080 /*
1081 * ... otherwise clear the request.
1082 */
1083 sx_xunlock(&proctree_lock);
1084 td->td_dbgflags &= ~TDB_STOPATFORK;
1085 cv_broadcast(&p->p_dbgwait);
1086 }
1087 PROC_UNLOCK(p);
1088 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1089 /*
1090 * This is the start of a new thread in a traced
1091 * process. Report a system call exit event.
1092 */
1093 PROC_LOCK(p);
1094 td->td_dbgflags |= TDB_SCX;
1095 _STOPEVENT(p, S_SCX, td->td_dbg_sc_code);
1096 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1097 (td->td_dbgflags & TDB_BORN) != 0)
1098 ptracestop(td, SIGTRAP, NULL);
1099 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1100 PROC_UNLOCK(p);
1101 }
1102
1103 userret(td, frame);
1104
1105 #ifdef KTRACE
1106 if (KTRPOINT(td, KTR_SYSRET))
1107 ktrsysret(SYS_fork, 0, 0);
1108 #endif
1109 }
Cache object: e7bdef892486a526f8cb5cd0a31bffdd
|