FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_fork.c
1 /* $NetBSD: kern_fork.c,v 1.114.2.3 2004/08/15 14:06:29 tron Exp $ */
2
3 /*-
4 * Copyright (c) 1999, 2001 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9 * NASA Ames Research Center.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the NetBSD
22 * Foundation, Inc. and its contributors.
23 * 4. Neither the name of The NetBSD Foundation nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
38 */
39
40 /*
41 * Copyright (c) 1982, 1986, 1989, 1991, 1993
42 * The Regents of the University of California. All rights reserved.
43 * (c) UNIX System Laboratories, Inc.
44 * All or some portions of this file are derived from material licensed
45 * to the University of California by American Telephone and Telegraph
46 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47 * the permission of UNIX System Laboratories, Inc.
48 *
49 * Redistribution and use in source and binary forms, with or without
50 * modification, are permitted provided that the following conditions
51 * are met:
52 * 1. Redistributions of source code must retain the above copyright
53 * notice, this list of conditions and the following disclaimer.
54 * 2. Redistributions in binary form must reproduce the above copyright
55 * notice, this list of conditions and the following disclaimer in the
56 * documentation and/or other materials provided with the distribution.
57 * 3. Neither the name of the University nor the names of its contributors
58 * may be used to endorse or promote products derived from this software
59 * without specific prior written permission.
60 *
61 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * SUCH DAMAGE.
72 *
73 * @(#)kern_fork.c 8.8 (Berkeley) 2/14/95
74 */
75
76 #include <sys/cdefs.h>
77 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.114.2.3 2004/08/15 14:06:29 tron Exp $");
78
79 #include "opt_ktrace.h"
80 #include "opt_systrace.h"
81 #include "opt_multiprocessor.h"
82
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/filedesc.h>
86 #include <sys/kernel.h>
87 #include <sys/malloc.h>
88 #include <sys/pool.h>
89 #include <sys/mount.h>
90 #include <sys/proc.h>
91 #include <sys/ras.h>
92 #include <sys/resourcevar.h>
93 #include <sys/vnode.h>
94 #include <sys/file.h>
95 #include <sys/acct.h>
96 #include <sys/ktrace.h>
97 #include <sys/vmmeter.h>
98 #include <sys/sched.h>
99 #include <sys/signalvar.h>
100 #include <sys/systrace.h>
101
102 #include <sys/sa.h>
103 #include <sys/syscallargs.h>
104
105 #include <uvm/uvm_extern.h>
106
107
108 int nprocs = 1; /* process 0 */
109
110 /*
111 * Number of ticks to sleep if fork() would fail due to process hitting
112 * limits. Exported in miliseconds to userland via sysctl.
113 */
114 int forkfsleep = 0;
115
116 /*ARGSUSED*/
117 int
118 sys_fork(struct lwp *l, void *v, register_t *retval)
119 {
120
121 return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL));
122 }
123
124 /*
125 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
126 * Address space is not shared, but parent is blocked until child exit.
127 */
128 /*ARGSUSED*/
129 int
130 sys_vfork(struct lwp *l, void *v, register_t *retval)
131 {
132
133 return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL,
134 retval, NULL));
135 }
136
137 /*
138 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
139 * semantics. Address space is shared, and parent is blocked until child exit.
140 */
141 /*ARGSUSED*/
142 int
143 sys___vfork14(struct lwp *l, void *v, register_t *retval)
144 {
145
146 return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0,
147 NULL, NULL, retval, NULL));
148 }
149
150 /*
151 * Linux-compatible __clone(2) system call.
152 */
153 int
154 sys___clone(struct lwp *l, void *v, register_t *retval)
155 {
156 struct sys___clone_args /* {
157 syscallarg(int) flags;
158 syscallarg(void *) stack;
159 } */ *uap = v;
160 int flags, sig;
161
162 /*
163 * We don't support the CLONE_PID or CLONE_PTRACE flags.
164 */
165 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE))
166 return (EINVAL);
167
168 /*
169 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same.
170 */
171 if (SCARG(uap, flags) & CLONE_SIGHAND
172 && (SCARG(uap, flags) & CLONE_VM) == 0)
173 return (EINVAL);
174
175 flags = 0;
176
177 if (SCARG(uap, flags) & CLONE_VM)
178 flags |= FORK_SHAREVM;
179 if (SCARG(uap, flags) & CLONE_FS)
180 flags |= FORK_SHARECWD;
181 if (SCARG(uap, flags) & CLONE_FILES)
182 flags |= FORK_SHAREFILES;
183 if (SCARG(uap, flags) & CLONE_SIGHAND)
184 flags |= FORK_SHARESIGS;
185 if (SCARG(uap, flags) & CLONE_VFORK)
186 flags |= FORK_PPWAIT;
187
188 sig = SCARG(uap, flags) & CLONE_CSIGNAL;
189 if (sig < 0 || sig >= _NSIG)
190 return (EINVAL);
191
192 /*
193 * Note that the Linux API does not provide a portable way of
194 * specifying the stack area; the caller must know if the stack
195 * grows up or down. So, we pass a stack size of 0, so that the
196 * code that makes this adjustment is a noop.
197 */
198 return (fork1(l, flags, sig, SCARG(uap, stack), 0,
199 NULL, NULL, retval, NULL));
200 }
201
202 /* print the 'table full' message once per 10 seconds */
203 struct timeval fork_tfmrate = { 10, 0 };
204
205 int
206 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize,
207 void (*func)(void *), void *arg, register_t *retval,
208 struct proc **rnewprocp)
209 {
210 struct proc *p1, *p2, *parent;
211 uid_t uid;
212 struct lwp *l2;
213 int count, s;
214 vaddr_t uaddr;
215 boolean_t inmem;
216
217 /*
218 * Although process entries are dynamically created, we still keep
219 * a global limit on the maximum number we will create. Don't allow
220 * a nonprivileged user to use the last few processes; don't let root
221 * exceed the limit. The variable nprocs is the current number of
222 * processes, maxproc is the limit.
223 */
224 p1 = l1->l_proc;
225 uid = p1->p_cred->p_ruid;
226 if (__predict_false((nprocs >= maxproc - 5 && uid != 0) ||
227 nprocs >= maxproc)) {
228 static struct timeval lasttfm;
229
230 if (ratecheck(&lasttfm, &fork_tfmrate))
231 tablefull("proc", "increase kern.maxproc or NPROC");
232 if (forkfsleep)
233 (void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep);
234 return (EAGAIN);
235 }
236 nprocs++;
237
238 /*
239 * Increment the count of procs running with this uid. Don't allow
240 * a nonprivileged user to exceed their current limit.
241 */
242 count = chgproccnt(uid, 1);
243 if (__predict_false(uid != 0 && count >
244 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
245 (void)chgproccnt(uid, -1);
246 nprocs--;
247 if (forkfsleep)
248 (void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep);
249 return (EAGAIN);
250 }
251
252 /*
253 * Allocate virtual address space for the U-area now, while it
254 * is still easy to abort the fork operation if we're out of
255 * kernel virtual address space. The actual U-area pages will
256 * be allocated and wired in uvm_fork() if needed.
257 */
258
259 inmem = uvm_uarea_alloc(&uaddr);
260 if (__predict_false(uaddr == 0)) {
261 (void)chgproccnt(uid, -1);
262 nprocs--;
263 return (ENOMEM);
264 }
265
266 /*
267 * We are now committed to the fork. From here on, we may
268 * block on resources, but resource allocation may NOT fail.
269 */
270
271 /* Allocate new proc. */
272 p2 = proc_alloc();
273
274 /*
275 * Make a proc table entry for the new process.
276 * Start by zeroing the section of proc that is zero-initialized,
277 * then copy the section that is copied directly from the parent.
278 */
279 memset(&p2->p_startzero, 0,
280 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
281 memcpy(&p2->p_startcopy, &p1->p_startcopy,
282 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
283
284 simple_lock_init(&p2->p_sigctx.ps_silock);
285 CIRCLEQ_INIT(&p2->p_sigctx.ps_siginfo);
286 simple_lock_init(&p2->p_lock);
287 LIST_INIT(&p2->p_lwps);
288
289 /*
290 * Duplicate sub-structures as needed.
291 * Increase reference counts on shared objects.
292 * The p_stats and p_sigacts substructs are set in uvm_fork().
293 */
294 p2->p_flag = (p1->p_flag & P_SUGID);
295 p2->p_emul = p1->p_emul;
296 p2->p_execsw = p1->p_execsw;
297
298 if (p1->p_flag & P_PROFIL)
299 startprofclock(p2);
300 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK);
301 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
302 p2->p_cred->p_refcnt = 1;
303 crhold(p1->p_ucred);
304
305 LIST_INIT(&p2->p_raslist);
306 #if defined(__HAVE_RAS)
307 ras_fork(p1, p2);
308 #endif
309
310 /* bump references to the text vnode (for procfs) */
311 p2->p_textvp = p1->p_textvp;
312 if (p2->p_textvp)
313 VREF(p2->p_textvp);
314
315 if (flags & FORK_SHAREFILES)
316 fdshare(p1, p2);
317 else if (flags & FORK_CLEANFILES)
318 p2->p_fd = fdinit(p1);
319 else
320 p2->p_fd = fdcopy(p1);
321
322 if (flags & FORK_SHARECWD)
323 cwdshare(p1, p2);
324 else
325 p2->p_cwdi = cwdinit(p1);
326
327 /*
328 * If p_limit is still copy-on-write, bump refcnt,
329 * otherwise get a copy that won't be modified.
330 * (If PL_SHAREMOD is clear, the structure is shared
331 * copy-on-write.)
332 */
333 if (p1->p_limit->p_lflags & PL_SHAREMOD)
334 p2->p_limit = limcopy(p1->p_limit);
335 else {
336 p2->p_limit = p1->p_limit;
337 p2->p_limit->p_refcnt++;
338 }
339
340 /* Inherit STOPFORK and STOPEXEC flags */
341 p2->p_flag |= p1->p_flag & (P_STOPFORK | P_STOPEXEC);
342
343 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
344 p2->p_flag |= P_CONTROLT;
345 if (flags & FORK_PPWAIT)
346 p2->p_flag |= P_PPWAIT;
347 parent = (flags & FORK_NOWAIT) ? initproc : p1;
348 p2->p_pptr = parent;
349 LIST_INIT(&p2->p_children);
350
351 s = proclist_lock_write();
352 LIST_INSERT_AFTER(p1, p2, p_pglist);
353 LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling);
354 proclist_unlock_write(s);
355
356 #ifdef KTRACE
357 /*
358 * Copy traceflag and tracefile if enabled.
359 * If not inherited, these were zeroed above.
360 */
361 if (p1->p_traceflag & KTRFAC_INHERIT) {
362 p2->p_traceflag = p1->p_traceflag;
363 if ((p2->p_tracep = p1->p_tracep) != NULL)
364 ktradref(p2);
365 }
366 #endif
367
368 scheduler_fork_hook(p1, p2);
369
370 /*
371 * Create signal actions for the child process.
372 */
373 sigactsinit(p2, p1, flags & FORK_SHARESIGS);
374
375 /*
376 * p_stats.
377 * Copy parts of p_stats, and zero out the rest.
378 */
379 p2->p_stats = pstatscopy(p1->p_stats);
380
381 /*
382 * If emulation has process fork hook, call it now.
383 */
384 if (p2->p_emul->e_proc_fork)
385 (*p2->p_emul->e_proc_fork)(p2, p1, flags);
386
387 /*
388 * ...and finally, any other random fork hooks that subsystems
389 * might have registered.
390 */
391 doforkhooks(p2, p1);
392
393 /*
394 * This begins the section where we must prevent the parent
395 * from being swapped.
396 */
397 PHOLD(l1);
398
399 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
400
401 /*
402 * Finish creating the child process.
403 * It will return through a different path later.
404 */
405 newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize,
406 (func != NULL) ? func : child_return,
407 arg, &l2);
408
409 /* Now safe for scheduler to see child process */
410 s = proclist_lock_write();
411 p2->p_stat = SIDL; /* protect against others */
412 p2->p_exitsig = exitsig; /* signal for parent on exit */
413 LIST_INSERT_HEAD(&allproc, p2, p_list);
414 proclist_unlock_write(s);
415
416 #ifdef SYSTRACE
417 /* Tell systrace what's happening. */
418 if (ISSET(p1->p_flag, P_SYSTRACE))
419 systrace_sys_fork(p1, p2);
420 #endif
421
422 #ifdef __HAVE_SYSCALL_INTERN
423 (*p2->p_emul->e_syscall_intern)(p2);
424 #endif
425
426 /*
427 * Make child runnable, set start time, and add to run queue
428 * except if the parent requested the child to start in SSTOP state.
429 */
430 SCHED_LOCK(s);
431 p2->p_stats->p_start = time;
432 p2->p_acflag = AFORK;
433 if (p1->p_flag & P_STOPFORK) {
434 p2->p_nrlwps = 0;
435 p1->p_nstopchild++;
436 p2->p_stat = SSTOP;
437 l2->l_stat = LSSTOP;
438 } else {
439 p2->p_nrlwps = 1;
440 p2->p_stat = SACTIVE;
441 l2->l_stat = LSRUN;
442 setrunqueue(l2);
443 }
444 SCHED_UNLOCK(s);
445
446 /*
447 * Now can be swapped.
448 */
449 PRELE(l1);
450
451 /*
452 * Notify any interested parties about the new process.
453 */
454 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
455
456 /*
457 * Update stats now that we know the fork was successful.
458 */
459 uvmexp.forks++;
460 if (flags & FORK_PPWAIT)
461 uvmexp.forks_ppwait++;
462 if (flags & FORK_SHAREVM)
463 uvmexp.forks_sharevm++;
464
465 /*
466 * Pass a pointer to the new process to the caller.
467 */
468 if (rnewprocp != NULL)
469 *rnewprocp = p2;
470
471 #ifdef KTRACE
472 if (KTRPOINT(p2, KTR_EMUL))
473 p2->p_traceflag |= KTRFAC_TRC_EMUL;
474 #endif
475
476 /*
477 * Preserve synchronization semantics of vfork. If waiting for
478 * child to exec or exit, set P_PPWAIT on child, and sleep on our
479 * proc (in case of exit).
480 */
481 if (flags & FORK_PPWAIT)
482 while (p2->p_flag & P_PPWAIT)
483 tsleep(p1, PWAIT, "ppwait", 0);
484
485 /*
486 * Return child pid to parent process,
487 * marking us as parent via retval[1].
488 */
489 if (retval != NULL) {
490 retval[0] = p2->p_pid;
491 retval[1] = 0;
492 }
493
494 return (0);
495 }
496
497 #if defined(MULTIPROCESSOR)
498 /*
499 * XXX This is a slight hack to get newly-formed processes to
500 * XXX acquire the kernel lock as soon as they run.
501 */
502 void
503 proc_trampoline_mp(void)
504 {
505 struct lwp *l;
506
507 l = curlwp;
508
509 SCHED_ASSERT_UNLOCKED();
510 KERNEL_PROC_LOCK(l);
511 }
512 #endif
Cache object: d29560dd15dc5ea3739235c045ef5910
|