kern_fork.c revision 1.130
1 /* $NetBSD: kern_fork.c,v 1.130 2007/02/09 21:55:30 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 1.130 2007/02/09 21:55:30 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/syscallargs.h> 106 107 #include <uvm/uvm_extern.h> 108 109 110 int nprocs = 1; /* process 0 */ 111 112 /* 113 * Number of ticks to sleep if fork() would fail due to process hitting 114 * limits. Exported in miliseconds to userland via sysctl. 115 */ 116 int forkfsleep = 0; 117 118 /*ARGSUSED*/ 119 int 120 sys_fork(struct lwp *l, void *v, register_t *retval) 121 { 122 123 return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 124 } 125 126 /* 127 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 128 * Address space is not shared, but parent is blocked until child exit. 129 */ 130 /*ARGSUSED*/ 131 int 132 sys_vfork(struct lwp *l, void *v, register_t *retval) 133 { 134 135 return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 136 retval, NULL)); 137 } 138 139 /* 140 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 141 * semantics. Address space is shared, and parent is blocked until child exit. 142 */ 143 /*ARGSUSED*/ 144 int 145 sys___vfork14(struct lwp *l, void *v, register_t *retval) 146 { 147 148 return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 149 NULL, NULL, retval, NULL)); 150 } 151 152 /* 153 * Linux-compatible __clone(2) system call. 154 */ 155 int 156 sys___clone(struct lwp *l, void *v, register_t *retval) 157 { 158 struct sys___clone_args /* { 159 syscallarg(int) flags; 160 syscallarg(void *) stack; 161 } */ *uap = v; 162 int flags, sig; 163 164 /* 165 * We don't support the CLONE_PID or CLONE_PTRACE flags. 166 */ 167 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE)) 168 return (EINVAL); 169 170 /* 171 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same. 172 */ 173 if (SCARG(uap, flags) & CLONE_SIGHAND 174 && (SCARG(uap, flags) & CLONE_VM) == 0) 175 return (EINVAL); 176 177 flags = 0; 178 179 if (SCARG(uap, flags) & CLONE_VM) 180 flags |= FORK_SHAREVM; 181 if (SCARG(uap, flags) & CLONE_FS) 182 flags |= FORK_SHARECWD; 183 if (SCARG(uap, flags) & CLONE_FILES) 184 flags |= FORK_SHAREFILES; 185 if (SCARG(uap, flags) & CLONE_SIGHAND) 186 flags |= FORK_SHARESIGS; 187 if (SCARG(uap, flags) & CLONE_VFORK) 188 flags |= FORK_PPWAIT; 189 190 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 191 if (sig < 0 || sig >= _NSIG) 192 return (EINVAL); 193 194 /* 195 * Note that the Linux API does not provide a portable way of 196 * specifying the stack area; the caller must know if the stack 197 * grows up or down. So, we pass a stack size of 0, so that the 198 * code that makes this adjustment is a noop. 199 */ 200 return (fork1(l, flags, sig, SCARG(uap, stack), 0, 201 NULL, NULL, retval, NULL)); 202 } 203 204 /* print the 'table full' message once per 10 seconds */ 205 struct timeval fork_tfmrate = { 10, 0 }; 206 207 /* 208 * General fork call. Note that another LWP in the process may call exec() 209 * or exit() while we are forking. It's safe to continue here, because 210 * neither operation will complete until all LWPs have exited the process. 211 */ 212 int 213 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize, 214 void (*func)(void *), void *arg, register_t *retval, 215 struct proc **rnewprocp) 216 { 217 struct proc *p1, *p2, *parent; 218 uid_t uid; 219 struct lwp *l2; 220 int count; 221 vaddr_t uaddr; 222 boolean_t inmem; 223 224 /* 225 * Although process entries are dynamically created, we still keep 226 * a global limit on the maximum number we will create. Don't allow 227 * a nonprivileged user to use the last few processes; don't let root 228 * exceed the limit. The variable nprocs is the current number of 229 * processes, maxproc is the limit. 230 */ 231 p1 = l1->l_proc; 232 mutex_enter(&p1->p_mutex); 233 uid = kauth_cred_getuid(p1->p_cred); 234 mutex_exit(&p1->p_mutex); 235 if (__predict_false((nprocs >= maxproc - 5 && uid != 0) || 236 nprocs >= maxproc)) { 237 static struct timeval lasttfm; 238 239 if (ratecheck(&lasttfm, &fork_tfmrate)) 240 tablefull("proc", "increase kern.maxproc or NPROC"); 241 if (forkfsleep) 242 (void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep); 243 return (EAGAIN); 244 } 245 nprocs++; 246 247 /* 248 * Increment the count of procs running with this uid. Don't allow 249 * a nonprivileged user to exceed their current limit. 250 */ 251 count = chgproccnt(uid, 1); 252 if (__predict_false(uid != 0 && count > 253 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 254 (void)chgproccnt(uid, -1); 255 nprocs--; 256 if (forkfsleep) 257 (void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep); 258 return (EAGAIN); 259 } 260 261 /* 262 * Allocate virtual address space for the U-area now, while it 263 * is still easy to abort the fork operation if we're out of 264 * kernel virtual address space. The actual U-area pages will 265 * be allocated and wired in uvm_fork() if needed. 266 */ 267 268 inmem = uvm_uarea_alloc(&uaddr); 269 if (__predict_false(uaddr == 0)) { 270 (void)chgproccnt(uid, -1); 271 nprocs--; 272 return (ENOMEM); 273 } 274 275 /* 276 * We are now committed to the fork. From here on, we may 277 * block on resources, but resource allocation may NOT fail. 278 */ 279 280 /* Allocate new proc. */ 281 p2 = proc_alloc(); 282 283 /* 284 * Make a proc table entry for the new process. 285 * Start by zeroing the section of proc that is zero-initialized, 286 * then copy the section that is copied directly from the parent. 287 */ 288 memset(&p2->p_startzero, 0, 289 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 290 memcpy(&p2->p_startcopy, &p1->p_startcopy, 291 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 292 293 CIRCLEQ_INIT(&p2->p_sigpend.sp_info); 294 295 LIST_INIT(&p2->p_lwps); 296 LIST_INIT(&p2->p_sigwaiters); 297 298 /* 299 * Duplicate sub-structures as needed. 300 * Increase reference counts on shared objects. 301 * The p_stats and p_sigacts substructs are set in uvm_fork(). 302 * Inherit flags we want to keep. The flags related to SIGCHLD 303 * handling are important in order to keep a consistent behaviour 304 * for the child after the fork. 305 */ 306 p2->p_flag = p1->p_flag & (P_SUGID | P_NOCLDWAIT | P_CLDSIGIGN); 307 p2->p_emul = p1->p_emul; 308 p2->p_execsw = p1->p_execsw; 309 310 if (flags & FORK_SYSTEM) { 311 /* 312 * Mark it as a system process. Set P_NOCLDWAIT so that 313 * children are reparented to init(8) when they exit. 314 * init(8) can easily wait them out for us. 315 */ 316 p2->p_flag |= (P_SYSTEM | P_NOCLDWAIT); 317 } 318 319 /* XXX p_smutex can be IPL_VM except for audio drivers */ 320 mutex_init(&p2->p_smutex, MUTEX_SPIN, IPL_SCHED); 321 mutex_init(&p2->p_stmutex, MUTEX_SPIN, IPL_STATCLOCK); 322 mutex_init(&p2->p_rasmutex, MUTEX_SPIN, IPL_NONE); 323 mutex_init(&p2->p_mutex, MUTEX_DEFAULT, IPL_NONE); 324 cv_init(&p2->p_refcv, "drainref"); 325 cv_init(&p2->p_waitcv, "wait"); 326 cv_init(&p2->p_lwpcv, "lwpwait"); 327 328 p2->p_refcnt = 1; 329 kauth_proc_fork(p1, p2); 330 331 LIST_INIT(&p2->p_raslist); 332 #if defined(__HAVE_RAS) 333 ras_fork(p1, p2); 334 #endif 335 336 /* bump references to the text vnode (for procfs) */ 337 p2->p_textvp = p1->p_textvp; 338 if (p2->p_textvp) 339 VREF(p2->p_textvp); 340 341 if (flags & FORK_SHAREFILES) 342 fdshare(p1, p2); 343 else if (flags & FORK_CLEANFILES) 344 p2->p_fd = fdinit(p1); 345 else 346 p2->p_fd = fdcopy(p1); 347 348 if (flags & FORK_SHARECWD) 349 cwdshare(p1, p2); 350 else 351 p2->p_cwdi = cwdinit(p1); 352 353 /* 354 * If p_limit is still copy-on-write, bump refcnt, 355 * otherwise get a copy that won't be modified. 356 * (If PL_SHAREMOD is clear, the structure is shared 357 * copy-on-write.) 358 */ 359 if (p1->p_limit->p_lflags & PL_SHAREMOD) { 360 mutex_enter(&p1->p_mutex); 361 p2->p_limit = limcopy(p1); 362 mutex_exit(&p1->p_mutex); 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 mutex_enter(&ktrace_mutex); 385 p2->p_traceflag = p1->p_traceflag; 386 if ((p2->p_tracep = p1->p_tracep) != NULL) 387 ktradref(p2); 388 mutex_exit(&ktrace_mutex); 389 } 390 #endif 391 392 /* 393 * Create signal actions for the child process. 394 */ 395 mutex_enter(&p1->p_smutex); 396 p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS); 397 p2->p_sflag |= 398 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 399 scheduler_fork_hook(p1, p2); 400 mutex_exit(&p1->p_smutex); 401 402 p2->p_stflag = p1->p_stflag; 403 404 /* 405 * p_stats. 406 * Copy parts of p_stats, and zero out the rest. 407 */ 408 p2->p_stats = pstatscopy(p1->p_stats); 409 410 /* 411 * If emulation has process fork hook, call it now. 412 */ 413 if (p2->p_emul->e_proc_fork) 414 (*p2->p_emul->e_proc_fork)(p2, p1, flags); 415 416 /* 417 * ...and finally, any other random fork hooks that subsystems 418 * might have registered. 419 */ 420 doforkhooks(p2, p1); 421 422 /* 423 * This begins the section where we must prevent the parent 424 * from being swapped. 425 */ 426 PHOLD(l1); 427 428 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE); 429 430 /* 431 * Finish creating the child process. 432 * It will return through a different path later. 433 */ 434 newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize, 435 (func != NULL) ? func : child_return, 436 arg, &l2); 437 438 /* 439 * It's now safe for the scheduler and other processes to see the 440 * child process. 441 */ 442 rw_enter(&proclist_lock, RW_WRITER); 443 444 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 445 p2->p_lflag |= PL_CONTROLT; 446 447 LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling); 448 p2->p_exitsig = exitsig; /* signal for parent on exit */ 449 450 mutex_enter(&proclist_mutex); 451 LIST_INSERT_AFTER(p1, p2, p_pglist); 452 LIST_INSERT_HEAD(&allproc, p2, p_list); 453 mutex_exit(&proclist_mutex); 454 455 rw_exit(&proclist_lock); 456 457 #ifdef SYSTRACE 458 /* Tell systrace what's happening. */ 459 if (ISSET(p1->p_flag, P_SYSTRACE)) 460 systrace_sys_fork(p1, p2); 461 #endif 462 463 #ifdef __HAVE_SYSCALL_INTERN 464 (*p2->p_emul->e_syscall_intern)(p2); 465 #endif 466 467 /* 468 * Now can be swapped. 469 */ 470 PRELE(l1); 471 472 /* 473 * Notify any interested parties about the new process. 474 */ 475 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 476 477 /* 478 * Update stats now that we know the fork was successful. 479 */ 480 uvmexp.forks++; 481 if (flags & FORK_PPWAIT) 482 uvmexp.forks_ppwait++; 483 if (flags & FORK_SHAREVM) 484 uvmexp.forks_sharevm++; 485 486 /* 487 * Pass a pointer to the new process to the caller. 488 */ 489 if (rnewprocp != NULL) 490 *rnewprocp = p2; 491 492 #ifdef KTRACE 493 if (KTRPOINT(p2, KTR_EMUL)) 494 p2->p_traceflag |= KTRFAC_TRC_EMUL; 495 #endif 496 497 /* 498 * Make child runnable, set start time, and add to run queue except 499 * if the parent requested the child to start in SSTOP state. 500 */ 501 mutex_enter(&proclist_mutex); 502 mutex_enter(&p2->p_smutex); 503 504 getmicrotime(&p2->p_stats->p_start); 505 p2->p_acflag = AFORK; 506 if (p2->p_sflag & PS_STOPFORK) { 507 lwp_lock(l2); 508 p2->p_nrlwps = 0; 509 p2->p_stat = SSTOP; 510 p2->p_waited = 0; 511 p1->p_nstopchild++; 512 l2->l_stat = LSSTOP; 513 lwp_unlock(l2); 514 } else { 515 p2->p_nrlwps = 1; 516 p2->p_stat = SACTIVE; 517 lwp_lock(l2); 518 l2->l_stat = LSRUN; 519 setrunqueue(l2); 520 lwp_unlock(l2); 521 } 522 523 mutex_exit(&proclist_mutex); 524 525 /* 526 * Start profiling. 527 */ 528 if ((p2->p_stflag & PST_PROFIL) != 0) { 529 mutex_spin_enter(&p2->p_stmutex); 530 startprofclock(p2); 531 mutex_spin_exit(&p2->p_stmutex); 532 } 533 534 /* 535 * Preserve synchronization semantics of vfork. If waiting for 536 * child to exec or exit, set PS_PPWAIT on child, and sleep on our 537 * proc (in case of exit). 538 */ 539 if (flags & FORK_PPWAIT) 540 while (p2->p_sflag & PS_PPWAIT) 541 cv_wait(&p1->p_waitcv, &p2->p_smutex); 542 543 mutex_exit(&p2->p_smutex); 544 545 /* 546 * Return child pid to parent process, 547 * marking us as parent via retval[1]. 548 */ 549 if (retval != NULL) { 550 retval[0] = p2->p_pid; 551 retval[1] = 0; 552 } 553 554 return (0); 555 } 556 557 #if defined(MULTIPROCESSOR) 558 /* 559 * XXX This is a slight hack to get newly-formed processes to 560 * XXX acquire the kernel lock as soon as they run. 561 */ 562 void 563 proc_trampoline_mp(void) 564 { 565 struct lwp *l; 566 567 l = curlwp; 568 569 KERNEL_LOCK(1, l); 570 } 571 #endif 572
Indexes created Mon Sep 22 10:09:38 GMT 2025