kern_fork.c revision 1.84.2.20
1 /* $NetBSD: kern_fork.c,v 1.84.2.20 2002/12/11 06:43:03 thorpej 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. All advertising materials mentioning features or use of this software 58 * must display the following acknowledgement: 59 * This product includes software developed by the University of 60 * California, Berkeley and its contributors. 61 * 4. Neither the name of the University nor the names of its contributors 62 * may be used to endorse or promote products derived from this software 63 * without specific prior written permission. 64 * 65 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 66 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 67 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 68 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 69 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 70 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 71 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 72 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 73 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 74 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 75 * SUCH DAMAGE. 76 * 77 * @(#)kern_fork.c 8.8 (Berkeley) 2/14/95 78 */ 79 80 #include <sys/cdefs.h> 81 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.84.2.20 2002/12/11 06:43:03 thorpej Exp $"); 82 83 #include "opt_ktrace.h" 84 #include "opt_systrace.h" 85 #include "opt_multiprocessor.h" 86 87 #include <sys/param.h> 88 #include <sys/systm.h> 89 #include <sys/filedesc.h> 90 #include <sys/kernel.h> 91 #include <sys/malloc.h> 92 #include <sys/pool.h> 93 #include <sys/mount.h> 94 #include <sys/proc.h> 95 #include <sys/ras.h> 96 #include <sys/resourcevar.h> 97 #include <sys/vnode.h> 98 #include <sys/file.h> 99 #include <sys/acct.h> 100 #include <sys/ktrace.h> 101 #include <sys/vmmeter.h> 102 #include <sys/sched.h> 103 #include <sys/signalvar.h> 104 #include <sys/systrace.h> 105 106 #include <sys/sa.h> 107 #include <sys/syscallargs.h> 108 109 #include <uvm/uvm_extern.h> 110 111 112 int nprocs = 1; /* process 0 */ 113 114 /*ARGSUSED*/ 115 int 116 sys_fork(struct lwp *l, void *v, register_t *retval) 117 { 118 119 return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 120 } 121 122 /* 123 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 124 * Address space is not shared, but parent is blocked until child exit. 125 */ 126 /*ARGSUSED*/ 127 int 128 sys_vfork(struct lwp *l, void *v, register_t *retval) 129 { 130 131 return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 132 retval, NULL)); 133 } 134 135 /* 136 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 137 * semantics. Address space is shared, and parent is blocked until child exit. 138 */ 139 /*ARGSUSED*/ 140 int 141 sys___vfork14(struct lwp *l, void *v, register_t *retval) 142 { 143 144 return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 145 NULL, NULL, retval, NULL)); 146 } 147 148 /* 149 * Linux-compatible __clone(2) system call. 150 */ 151 int 152 sys___clone(struct lwp *l, void *v, register_t *retval) 153 { 154 struct sys___clone_args /* { 155 syscallarg(int) flags; 156 syscallarg(void *) stack; 157 } */ *uap = v; 158 int flags, sig; 159 160 /* 161 * We don't support the CLONE_PID or CLONE_PTRACE flags. 162 */ 163 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE)) 164 return (EINVAL); 165 166 flags = 0; 167 168 if (SCARG(uap, flags) & CLONE_VM) 169 flags |= FORK_SHAREVM; 170 if (SCARG(uap, flags) & CLONE_FS) 171 flags |= FORK_SHARECWD; 172 if (SCARG(uap, flags) & CLONE_FILES) 173 flags |= FORK_SHAREFILES; 174 if (SCARG(uap, flags) & CLONE_SIGHAND) 175 flags |= FORK_SHARESIGS; 176 if (SCARG(uap, flags) & CLONE_VFORK) 177 flags |= FORK_PPWAIT; 178 179 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 180 if (sig < 0 || sig >= _NSIG) 181 return (EINVAL); 182 183 /* 184 * Note that the Linux API does not provide a portable way of 185 * specifying the stack area; the caller must know if the stack 186 * grows up or down. So, we pass a stack size of 0, so that the 187 * code that makes this adjustment is a noop. 188 */ 189 return (fork1(l, flags, sig, SCARG(uap, stack), 0, 190 NULL, NULL, retval, NULL)); 191 } 192 193 /* print the 'table full' message once per 10 seconds */ 194 struct timeval fork_tfmrate = { 10, 0 }; 195 196 int 197 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize, 198 void (*func)(void *), void *arg, register_t *retval, 199 struct proc **rnewprocp) 200 { 201 struct proc *p1, *p2, *tp; 202 uid_t uid; 203 struct lwp *l2; 204 int count, s; 205 vaddr_t uaddr; 206 boolean_t inmem; 207 static int nextpid, pidchecked; 208 209 /* 210 * Although process entries are dynamically created, we still keep 211 * a global limit on the maximum number we will create. Don't allow 212 * a nonprivileged user to use the last few processes; don't let root 213 * exceed the limit. The variable nprocs is the current number of 214 * processes, maxproc is the limit. 215 */ 216 p1 = l1->l_proc; 217 uid = p1->p_cred->p_ruid; 218 if (__predict_false((nprocs >= maxproc - 5 && uid != 0) || 219 nprocs >= maxproc)) { 220 static struct timeval lasttfm; 221 222 if (ratecheck(&lasttfm, &fork_tfmrate)) 223 tablefull("proc", "increase kern.maxproc or NPROC"); 224 (void)tsleep(&nprocs, PUSER, "forkmx", hz / 2); 225 return (EAGAIN); 226 } 227 nprocs++; 228 229 /* 230 * Increment the count of procs running with this uid. Don't allow 231 * a nonprivileged user to exceed their current limit. 232 */ 233 count = chgproccnt(uid, 1); 234 if (__predict_false(uid != 0 && count > 235 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 236 (void)chgproccnt(uid, -1); 237 nprocs--; 238 (void)tsleep(&nprocs, PUSER, "forkulim", hz / 2); 239 return (EAGAIN); 240 } 241 242 /* 243 * Allocate virtual address space for the U-area now, while it 244 * is still easy to abort the fork operation if we're out of 245 * kernel virtual address space. The actual U-area pages will 246 * be allocated and wired in uvm_fork() if needed. 247 */ 248 249 inmem = uvm_uarea_alloc(&uaddr); 250 if (__predict_false(uaddr == 0)) { 251 (void)chgproccnt(uid, -1); 252 nprocs--; 253 return (ENOMEM); 254 } 255 256 /* 257 * We are now committed to the fork. From here on, we may 258 * block on resources, but resource allocation may NOT fail. 259 */ 260 261 /* Allocate new proc. */ 262 p2 = pool_get(&proc_pool, PR_WAITOK); 263 264 /* 265 * Make a proc table entry for the new process. 266 * Start by zeroing the section of proc that is zero-initialized, 267 * then copy the section that is copied directly from the parent. 268 */ 269 memset(&p2->p_startzero, 0, 270 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 271 memcpy(&p2->p_startcopy, &p1->p_startcopy, 272 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 273 274 simple_lock_init(&p2->p_lwplock); 275 LIST_INIT(&p2->p_lwps); 276 277 /* 278 * Duplicate sub-structures as needed. 279 * Increase reference counts on shared objects. 280 * The p_stats and p_sigacts substructs are set in uvm_fork(). 281 */ 282 p2->p_flag = (p1->p_flag & P_SUGID); 283 p2->p_emul = p1->p_emul; 284 p2->p_execsw = p1->p_execsw; 285 286 if (p1->p_flag & P_PROFIL) 287 startprofclock(p2); 288 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK); 289 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred)); 290 p2->p_cred->p_refcnt = 1; 291 crhold(p1->p_ucred); 292 293 LIST_INIT(&p2->p_raslist); 294 p2->p_nras = 0; 295 simple_lock_init(&p2->p_raslock); 296 #if defined(__HAVE_RAS) 297 ras_fork(p1, p2); 298 #endif 299 300 /* bump references to the text vnode (for procfs) */ 301 p2->p_textvp = p1->p_textvp; 302 if (p2->p_textvp) 303 VREF(p2->p_textvp); 304 305 if (flags & FORK_SHAREFILES) 306 fdshare(p1, p2); 307 else if (flags & FORK_CLEANFILES) 308 p2->p_fd = fdinit(p1); 309 else 310 p2->p_fd = fdcopy(p1); 311 312 if (flags & FORK_SHARECWD) 313 cwdshare(p1, p2); 314 else 315 p2->p_cwdi = cwdinit(p1); 316 317 /* 318 * If p_limit is still copy-on-write, bump refcnt, 319 * otherwise get a copy that won't be modified. 320 * (If PL_SHAREMOD is clear, the structure is shared 321 * copy-on-write.) 322 */ 323 if (p1->p_limit->p_lflags & PL_SHAREMOD) 324 p2->p_limit = limcopy(p1->p_limit); 325 else { 326 p2->p_limit = p1->p_limit; 327 p2->p_limit->p_refcnt++; 328 } 329 330 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 331 p2->p_flag |= P_CONTROLT; 332 if (flags & FORK_PPWAIT) 333 p2->p_flag |= P_PPWAIT; 334 LIST_INSERT_AFTER(p1, p2, p_pglist); 335 p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1; 336 LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling); 337 LIST_INIT(&p2->p_children); 338 339 #ifdef KTRACE 340 /* 341 * Copy traceflag and tracefile if enabled. 342 * If not inherited, these were zeroed above. 343 */ 344 if (p1->p_traceflag & KTRFAC_INHERIT) { 345 p2->p_traceflag = p1->p_traceflag; 346 if ((p2->p_tracep = p1->p_tracep) != NULL) 347 ktradref(p2); 348 } 349 #endif 350 351 scheduler_fork_hook(p1, p2); 352 353 /* 354 * Create signal actions for the child process. 355 */ 356 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 357 358 /* 359 * p_stats. 360 * Copy parts of p_stats, and zero out the rest. 361 */ 362 p2->p_stats = pstatscopy(p1->p_stats); 363 364 /* 365 * If emulation has process fork hook, call it now. 366 */ 367 if (p2->p_emul->e_proc_fork) 368 (*p2->p_emul->e_proc_fork)(p2, p1); 369 370 /* 371 * This begins the section where we must prevent the parent 372 * from being swapped. 373 */ 374 PHOLD(l1); 375 376 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE); 377 378 /* 379 * Finish creating the child process. It will return through a 380 * different path later. 381 */ 382 newlwp(l1, p2, uaddr, 0, stack, stacksize, 383 (func != NULL) ? func : child_return, 384 arg, &l2); 385 if (inmem) 386 l2->l_flag |= L_INMEM; 387 388 /* 389 * BEGIN PID ALLOCATION. 390 */ 391 s = proclist_lock_write(); 392 393 /* 394 * Find an unused process ID. We remember a range of unused IDs 395 * ready to use (from nextpid+1 through pidchecked-1). 396 */ 397 nextpid++; 398 retry: 399 /* 400 * If the process ID prototype has wrapped around, 401 * restart somewhat above 0, as the low-numbered procs 402 * tend to include daemons that don't exit. 403 */ 404 if (nextpid >= PID_MAX) { 405 nextpid = 500; 406 pidchecked = 0; 407 } 408 if (nextpid >= pidchecked) { 409 const struct proclist_desc *pd; 410 411 pidchecked = PID_MAX; 412 /* 413 * Scan the process lists to check whether this pid 414 * is in use. Remember the lowest pid that's greater 415 * than nextpid, so we can avoid checking for a while. 416 */ 417 pd = proclists; 418 again: 419 LIST_FOREACH(tp, pd->pd_list, p_list) { 420 while (tp->p_pid == nextpid || 421 tp->p_pgrp->pg_id == nextpid || 422 tp->p_session->s_sid == nextpid) { 423 nextpid++; 424 if (nextpid >= pidchecked) 425 goto retry; 426 } 427 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 428 pidchecked = tp->p_pid; 429 430 if (tp->p_pgrp->pg_id > nextpid && 431 pidchecked > tp->p_pgrp->pg_id) 432 pidchecked = tp->p_pgrp->pg_id; 433 434 if (tp->p_session->s_sid > nextpid && 435 pidchecked > tp->p_session->s_sid) 436 pidchecked = tp->p_session->s_sid; 437 } 438 439 /* 440 * If there's another list, scan it. If we have checked 441 * them all, we've found one! 442 */ 443 pd++; 444 if (pd->pd_list != NULL) 445 goto again; 446 } 447 448 /* 449 * Put the proc on allproc before unlocking PID allocation 450 * so that waiters won't grab it as soon as we unlock. 451 */ 452 453 p2->p_stat = SIDL; /* protect against others */ 454 p2->p_pid = nextpid; 455 p2->p_exitsig = exitsig; /* signal for parent on exit */ 456 457 LIST_INSERT_HEAD(&allproc, p2, p_list); 458 459 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 460 461 /* 462 * END PID ALLOCATION. 463 */ 464 proclist_unlock_write(s); 465 466 #ifdef SYSTRACE 467 /* Tell systrace what's happening. */ 468 if (ISSET(p1->p_flag, P_SYSTRACE)) 469 systrace_sys_fork(p1, p2); 470 #endif 471 472 #ifdef __HAVE_SYSCALL_INTERN 473 (*p2->p_emul->e_syscall_intern)(p2); 474 #endif 475 476 /* 477 * Make child runnable, set start time, and add to run queue 478 * except if the parent requested the child to start in SSTOP state. 479 */ 480 SCHED_LOCK(s); 481 p2->p_stats->p_start = time; 482 p2->p_acflag = AFORK; 483 p2->p_nrlwps = 1; 484 if (p1->p_flag & P_STOPFORK) { 485 p2->p_stat = SSTOP; 486 l2->l_stat = LSSTOP; 487 } else { 488 p2->p_stat = SACTIVE; 489 l2->l_stat = LSRUN; 490 setrunqueue(l2); 491 } 492 SCHED_UNLOCK(s); 493 494 /* 495 * Inherit STOPFORK and STOPEXEC flags 496 */ 497 if (p1->p_flag & P_STOPFORK) 498 p2->p_flag |= P_STOPFORK; 499 if (p1->p_flag & P_STOPEXEC) 500 p2->p_flag |= P_STOPEXEC; 501 502 /* 503 * Now can be swapped. 504 */ 505 PRELE(l1); 506 507 /* 508 * Notify any interested parties about the new process. 509 */ 510 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 511 512 /* 513 * Update stats now that we know the fork was successful. 514 */ 515 uvmexp.forks++; 516 if (flags & FORK_PPWAIT) 517 uvmexp.forks_ppwait++; 518 if (flags & FORK_SHAREVM) 519 uvmexp.forks_sharevm++; 520 521 /* 522 * Pass a pointer to the new process to the caller. 523 */ 524 if (rnewprocp != NULL) 525 *rnewprocp = p2; 526 527 #ifdef KTRACE 528 if (KTRPOINT(p2, KTR_EMUL)) 529 ktremul(p2); 530 #endif 531 532 /* 533 * Preserve synchronization semantics of vfork. If waiting for 534 * child to exec or exit, set P_PPWAIT on child, and sleep on our 535 * proc (in case of exit). 536 */ 537 if (flags & FORK_PPWAIT) 538 while (p2->p_flag & P_PPWAIT) 539 tsleep(p1, PWAIT, "ppwait", 0); 540 541 /* 542 * Return child pid to parent process, 543 * marking us as parent via retval[1]. 544 */ 545 if (retval != NULL) { 546 retval[0] = p2->p_pid; 547 retval[1] = 0; 548 } 549 550 return (0); 551 } 552 553 #if defined(MULTIPROCESSOR) 554 /* 555 * XXX This is a slight hack to get newly-formed processes to 556 * XXX acquire the kernel lock as soon as they run. 557 */ 558 void 559 proc_trampoline_mp(void) 560 { 561 struct lwp *l; 562 563 l = curlwp; 564 565 SCHED_ASSERT_UNLOCKED(); 566 KERNEL_PROC_LOCK(l); 567 } 568 #endif 569
Indexes created Fri Sep 26 08:10:20 GMT 2025