kern_fork.c revision 1.126.4.4
1 /* $NetBSD: kern_fork.c,v 1.126.4.4 2006/11/17 16:34:36 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2001, 2004 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 * This code is derived from software contributed to The NetBSD Foundation 11 * by Charles M. Hannum. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. All advertising materials mentioning features or use of this software 22 * must display the following acknowledgement: 23 * This product includes software developed by the NetBSD 24 * Foundation, Inc. and its contributors. 25 * 4. Neither the name of The NetBSD Foundation nor the names of its 26 * contributors may be used to endorse or promote products derived 27 * from this software without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 30 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 31 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 32 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 33 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 34 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 35 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 36 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 37 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 38 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 39 * POSSIBILITY OF SUCH DAMAGE. 40 */ 41 42 /* 43 * Copyright (c) 1982, 1986, 1989, 1991, 1993 44 * The Regents of the University of California. All rights reserved. 45 * (c) UNIX System Laboratories, Inc. 46 * All or some portions of this file are derived from material licensed 47 * to the University of California by American Telephone and Telegraph 48 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 49 * the permission of UNIX System Laboratories, Inc. 50 * 51 * Redistribution and use in source and binary forms, with or without 52 * modification, are permitted provided that the following conditions 53 * are met: 54 * 1. Redistributions of source code must retain the above copyright 55 * notice, this list of conditions and the following disclaimer. 56 * 2. Redistributions in binary form must reproduce the above copyright 57 * notice, this list of conditions and the following disclaimer in the 58 * documentation and/or other materials provided with the distribution. 59 * 3. Neither the name of the University nor the names of its contributors 60 * may be used to endorse or promote products derived from this software 61 * without specific prior written permission. 62 * 63 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 64 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 65 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 66 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 67 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 68 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 69 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 70 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 71 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 72 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 73 * SUCH DAMAGE. 74 * 75 * @(#)kern_fork.c 8.8 (Berkeley) 2/14/95 76 */ 77 78 #include <sys/cdefs.h> 79 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.126.4.4 2006/11/17 16:34:36 ad Exp $"); 80 81 #include "opt_ktrace.h" 82 #include "opt_systrace.h" 83 #include "opt_multiprocessor.h" 84 85 #include <sys/param.h> 86 #include <sys/systm.h> 87 #include <sys/filedesc.h> 88 #include <sys/kernel.h> 89 #include <sys/malloc.h> 90 #include <sys/pool.h> 91 #include <sys/mount.h> 92 #include <sys/proc.h> 93 #include <sys/ras.h> 94 #include <sys/resourcevar.h> 95 #include <sys/vnode.h> 96 #include <sys/file.h> 97 #include <sys/acct.h> 98 #include <sys/ktrace.h> 99 #include <sys/vmmeter.h> 100 #include <sys/sched.h> 101 #include <sys/signalvar.h> 102 #include <sys/systrace.h> 103 #include <sys/kauth.h> 104 105 #include <sys/sa.h> 106 #include <sys/syscallargs.h> 107 108 #include <uvm/uvm_extern.h> 109 110 111 int nprocs = 1; /* process 0 */ 112 113 /* 114 * Number of ticks to sleep if fork() would fail due to process hitting 115 * limits. Exported in miliseconds to userland via sysctl. 116 */ 117 int forkfsleep = 0; 118 119 /*ARGSUSED*/ 120 int 121 sys_fork(struct lwp *l, void *v, register_t *retval) 122 { 123 124 return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 125 } 126 127 /* 128 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 129 * Address space is not shared, but parent is blocked until child exit. 130 */ 131 /*ARGSUSED*/ 132 int 133 sys_vfork(struct lwp *l, void *v, register_t *retval) 134 { 135 136 return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 137 retval, NULL)); 138 } 139 140 /* 141 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 142 * semantics. Address space is shared, and parent is blocked until child exit. 143 */ 144 /*ARGSUSED*/ 145 int 146 sys___vfork14(struct lwp *l, void *v, register_t *retval) 147 { 148 149 return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 150 NULL, NULL, retval, NULL)); 151 } 152 153 /* 154 * Linux-compatible __clone(2) system call. 155 */ 156 int 157 sys___clone(struct lwp *l, void *v, register_t *retval) 158 { 159 struct sys___clone_args /* { 160 syscallarg(int) flags; 161 syscallarg(void *) stack; 162 } */ *uap = v; 163 int flags, sig; 164 165 /* 166 * We don't support the CLONE_PID or CLONE_PTRACE flags. 167 */ 168 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE)) 169 return (EINVAL); 170 171 /* 172 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same. 173 */ 174 if (SCARG(uap, flags) & CLONE_SIGHAND 175 && (SCARG(uap, flags) & CLONE_VM) == 0) 176 return (EINVAL); 177 178 flags = 0; 179 180 if (SCARG(uap, flags) & CLONE_VM) 181 flags |= FORK_SHAREVM; 182 if (SCARG(uap, flags) & CLONE_FS) 183 flags |= FORK_SHARECWD; 184 if (SCARG(uap, flags) & CLONE_FILES) 185 flags |= FORK_SHAREFILES; 186 if (SCARG(uap, flags) & CLONE_SIGHAND) 187 flags |= FORK_SHARESIGS; 188 if (SCARG(uap, flags) & CLONE_VFORK) 189 flags |= FORK_PPWAIT; 190 191 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 192 if (sig < 0 || sig >= _NSIG) 193 return (EINVAL); 194 195 /* 196 * Note that the Linux API does not provide a portable way of 197 * specifying the stack area; the caller must know if the stack 198 * grows up or down. So, we pass a stack size of 0, so that the 199 * code that makes this adjustment is a noop. 200 */ 201 return (fork1(l, flags, sig, SCARG(uap, stack), 0, 202 NULL, NULL, retval, NULL)); 203 } 204 205 /* print the 'table full' message once per 10 seconds */ 206 struct timeval fork_tfmrate = { 10, 0 }; 207 208 /* 209 * General fork call. Note that another LWP in the process may call exec() 210 * or exit() while we are forking. It's safe to continue here, because 211 * neither operation will complete until all LWPs have exited the process. 212 */ 213 int 214 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize, 215 void (*func)(void *), void *arg, register_t *retval, 216 struct proc **rnewprocp) 217 { 218 struct proc *p1, *p2, *parent; 219 uid_t uid; 220 struct lwp *l2; 221 int count; 222 vaddr_t uaddr; 223 boolean_t inmem; 224 225 /* 226 * Although process entries are dynamically created, we still keep 227 * a global limit on the maximum number we will create. Don't allow 228 * a nonprivileged user to use the last few processes; don't let root 229 * exceed the limit. The variable nprocs is the current number of 230 * processes, maxproc is the limit. 231 */ 232 p1 = l1->l_proc; 233 mutex_enter(&p1->p_mutex); 234 uid = kauth_cred_getuid(p1->p_cred); 235 mutex_exit(&p1->p_mutex); 236 if (__predict_false((nprocs >= maxproc - 5 && uid != 0) || 237 nprocs >= maxproc)) { 238 static struct timeval lasttfm; 239 240 if (ratecheck(&lasttfm, &fork_tfmrate)) 241 tablefull("proc", "increase kern.maxproc or NPROC"); 242 if (forkfsleep) 243 (void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep); 244 return (EAGAIN); 245 } 246 nprocs++; 247 248 /* 249 * Increment the count of procs running with this uid. Don't allow 250 * a nonprivileged user to exceed their current limit. 251 */ 252 count = chgproccnt(uid, 1); 253 if (__predict_false(uid != 0 && count > 254 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 255 (void)chgproccnt(uid, -1); 256 nprocs--; 257 if (forkfsleep) 258 (void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep); 259 return (EAGAIN); 260 } 261 262 /* 263 * Allocate virtual address space for the U-area now, while it 264 * is still easy to abort the fork operation if we're out of 265 * kernel virtual address space. The actual U-area pages will 266 * be allocated and wired in uvm_fork() if needed. 267 */ 268 269 inmem = uvm_uarea_alloc(&uaddr); 270 if (__predict_false(uaddr == 0)) { 271 (void)chgproccnt(uid, -1); 272 nprocs--; 273 return (ENOMEM); 274 } 275 276 /* 277 * We are now committed to the fork. From here on, we may 278 * block on resources, but resource allocation may NOT fail. 279 */ 280 281 /* Allocate new proc. */ 282 p2 = proc_alloc(); 283 284 /* 285 * Make a proc table entry for the new process. 286 * Start by zeroing the section of proc that is zero-initialized, 287 * then copy the section that is copied directly from the parent. 288 */ 289 memset(&p2->p_startzero, 0, 290 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 291 memcpy(&p2->p_startcopy, &p1->p_startcopy, 292 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 293 294 CIRCLEQ_INIT(&p2->p_sigpend.sp_info); 295 296 LIST_INIT(&p2->p_lwps); 297 LIST_INIT(&p2->p_sigwaiters); 298 299 /* 300 * Duplicate sub-structures as needed. 301 * Increase reference counts on shared objects. 302 * The p_stats and p_sigacts substructs are set in uvm_fork(). 303 * Inherit flags we want to keep. The flags related to SIGCHLD 304 * handling are important in order to keep a consistent behaviour 305 * for the child after the fork. 306 */ 307 p2->p_flag = p1->p_flag & (P_SUGID | P_NOCLDWAIT | P_CLDSIGIGN); 308 p2->p_emul = p1->p_emul; 309 p2->p_execsw = p1->p_execsw; 310 311 if (flags & FORK_SYSTEM) { 312 /* 313 * Mark it as a system process. Set P_NOCLDWAIT so that 314 * children are reparented to init(8) when they exit. 315 * init(8) can easily wait them out for us. 316 */ 317 p2->p_flag |= (P_SYSTEM | P_NOCLDWAIT); 318 } 319 320 mutex_init(&p2->p_smutex, MUTEX_SPIN, IPL_SCHED); 321 mutex_init(&p2->p_rasmutex, MUTEX_SPIN, IPL_NONE); 322 mutex_init(&p2->p_mutex, MUTEX_DEFAULT, IPL_NONE); 323 cv_init(&p2->p_refcv, "drainref"); 324 cv_init(&p2->p_waitcv, "wait"); 325 326 p2->p_refcnt = 1; 327 328 mutex_enter(&p1->p_mutex); 329 kauth_cred_hold(p1->p_cred); 330 p2->p_cred = p1->p_cred; 331 mutex_exit(&p1->p_mutex); 332 333 LIST_INIT(&p2->p_raslist); 334 #if defined(__HAVE_RAS) 335 ras_fork(p1, p2); 336 #endif 337 338 /* bump references to the text vnode (for procfs) */ 339 p2->p_textvp = p1->p_textvp; 340 if (p2->p_textvp) 341 VREF(p2->p_textvp); 342 343 if (flags & FORK_SHAREFILES) 344 fdshare(p1, p2); 345 else if (flags & FORK_CLEANFILES) 346 p2->p_fd = fdinit(p1); 347 else 348 p2->p_fd = fdcopy(p1); 349 350 if (flags & FORK_SHARECWD) 351 cwdshare(p1, p2); 352 else 353 p2->p_cwdi = cwdinit(p1); 354 355 /* 356 * If p_limit is still copy-on-write, bump refcnt, 357 * otherwise get a copy that won't be modified. 358 * (If PL_SHAREMOD is clear, the structure is shared 359 * copy-on-write.) 360 */ 361 if (p1->p_limit->p_lflags & PL_SHAREMOD) 362 p2->p_limit = limcopy(p1->p_limit); 363 else { 364 simple_lock(&p1->p_limit->p_slock); 365 p1->p_limit->p_refcnt++; 366 simple_unlock(&p1->p_limit->p_slock); 367 p2->p_limit = p1->p_limit; 368 } 369 370 p2->p_sflag = ((flags & FORK_PPWAIT) ? PS_PPWAIT : 0); 371 p2->p_lflag = 0; 372 p2->p_slflag = 0; 373 parent = (flags & FORK_NOWAIT) ? initproc : p1; 374 p2->p_pptr = parent; 375 LIST_INIT(&p2->p_children); 376 377 378 #ifdef KTRACE 379 /* 380 * Copy traceflag and tracefile if enabled. 381 * If not inherited, these were zeroed above. 382 */ 383 if (p1->p_traceflag & KTRFAC_INHERIT) { 384 p2->p_traceflag = p1->p_traceflag; 385 if ((p2->p_tracep = p1->p_tracep) != NULL) 386 ktradref(p2); 387 } 388 #endif 389 390 /* 391 * Create signal actions for the child process. 392 */ 393 mutex_enter(&p1->p_smutex); 394 p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS); 395 p2->p_sflag |= (p1->p_sflag & 396 (PS_PROFIL | PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 397 scheduler_fork_hook(p1, p2); 398 mutex_exit(&p1->p_smutex); 399 400 /* 401 * p_stats. 402 * Copy parts of p_stats, and zero out the rest. 403 */ 404 p2->p_stats = pstatscopy(p1->p_stats); 405 406 /* 407 * If emulation has process fork hook, call it now. 408 */ 409 if (p2->p_emul->e_proc_fork) 410 (*p2->p_emul->e_proc_fork)(p2, p1, flags); 411 412 /* 413 * ...and finally, any other random fork hooks that subsystems 414 * might have registered. 415 */ 416 doforkhooks(p2, p1); 417 418 /* 419 * This begins the section where we must prevent the parent 420 * from being swapped. 421 */ 422 PHOLD(l1); 423 424 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE); 425 426 /* 427 * Finish creating the child process. 428 * It will return through a different path later. 429 */ 430 newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize, 431 (func != NULL) ? func : child_return, 432 arg, &l2); 433 434 /* 435 * It's now safe for the scheduler and other processes to see the 436 * child process. 437 */ 438 rw_enter(&proclist_lock, RW_WRITER); 439 440 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 441 p2->p_lflag |= PL_CONTROLT; 442 443 LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling); 444 p2->p_exitsig = exitsig; /* signal for parent on exit */ 445 446 mutex_enter(&proclist_mutex); 447 LIST_INSERT_AFTER(p1, p2, p_pglist); 448 LIST_INSERT_HEAD(&allproc, p2, p_list); 449 mutex_exit(&proclist_mutex); 450 451 rw_exit(&proclist_lock); 452 453 #ifdef SYSTRACE 454 /* Tell systrace what's happening. */ 455 if (ISSET(p1->p_flag, P_SYSTRACE)) 456 systrace_sys_fork(p1, p2); 457 #endif 458 459 #ifdef __HAVE_SYSCALL_INTERN 460 (*p2->p_emul->e_syscall_intern)(p2); 461 #endif 462 463 /* 464 * Now can be swapped. 465 */ 466 PRELE(l1); 467 468 /* 469 * Notify any interested parties about the new process. 470 */ 471 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 472 473 /* 474 * Update stats now that we know the fork was successful. 475 */ 476 uvmexp.forks++; 477 if (flags & FORK_PPWAIT) 478 uvmexp.forks_ppwait++; 479 if (flags & FORK_SHAREVM) 480 uvmexp.forks_sharevm++; 481 482 /* 483 * Pass a pointer to the new process to the caller. 484 */ 485 if (rnewprocp != NULL) 486 *rnewprocp = p2; 487 488 #ifdef KTRACE 489 if (KTRPOINT(p2, KTR_EMUL)) 490 p2->p_traceflag |= KTRFAC_TRC_EMUL; 491 #endif 492 493 /* 494 * Make child runnable, set start time, and add to run queue except 495 * if the parent requested the child to start in SSTOP state. 496 */ 497 mutex_enter(&p2->p_smutex); 498 499 getmicrotime(&p2->p_stats->p_start); 500 if (p2->p_sflag & PS_PROFIL) 501 startprofclock(p2); 502 p2->p_acflag = AFORK; 503 if (p2->p_sflag & PS_STOPFORK) { 504 lwp_lock(l2); 505 p2->p_nrlwps = 0; 506 p2->p_stat = SSTOP; 507 mutex_enter(&proc_stop_mutex); 508 p1->p_nstopchild++; 509 mutex_exit(&proc_stop_mutex); 510 l2->l_stat = LSSTOP; 511 lwp_unlock(l2); 512 } else { 513 p2->p_nrlwps = 1; 514 p2->p_stat = SACTIVE; 515 lwp_lock(l2); 516 lwp_relock(l2, &sched_mutex); 517 l2->l_stat = LSRUN; 518 setrunqueue(l2); 519 lwp_unlock(l2); 520 } 521 522 /* 523 * Preserve synchronization semantics of vfork. If waiting for 524 * child to exec or exit, set PS_PPWAIT on child, and sleep on our 525 * proc (in case of exit). 526 */ 527 if (flags & FORK_PPWAIT) 528 while (p2->p_sflag & PS_PPWAIT) 529 cv_wait(&p1->p_waitcv, &p2->p_smutex); 530 531 mutex_exit(&p2->p_smutex); 532 533 /* 534 * Return child pid to parent process, 535 * marking us as parent via retval[1]. 536 */ 537 if (retval != NULL) { 538 retval[0] = p2->p_pid; 539 retval[1] = 0; 540 } 541 542 return (0); 543 } 544 545 #if defined(MULTIPROCESSOR) 546 /* 547 * XXX This is a slight hack to get newly-formed processes to 548 * XXX acquire the kernel lock as soon as they run. 549 */ 550 void 551 proc_trampoline_mp(void) 552 { 553 struct lwp *l; 554 555 l = curlwp; 556 557 KERNEL_LOCK(1, l); 558 } 559 #endif 560
Indexes created Thu Sep 25 16:09:42 GMT 2025