kern_fork.c revision 1.99
1 /* $NetBSD: kern_fork.c,v 1.99 2002/11/17 08:32:44 chs 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 1.99 2002/11/17 08:32:44 chs 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/syscallargs.h> 107 108 #include <uvm/uvm_extern.h> 109 110 111 int nprocs = 1; /* process 0 */ 112 113 /*ARGSUSED*/ 114 int 115 sys_fork(struct proc *p, void *v, register_t *retval) 116 { 117 118 return (fork1(p, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 119 } 120 121 /* 122 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 123 * Address space is not shared, but parent is blocked until child exit. 124 */ 125 /*ARGSUSED*/ 126 int 127 sys_vfork(struct proc *p, void *v, register_t *retval) 128 { 129 130 return (fork1(p, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 131 retval, NULL)); 132 } 133 134 /* 135 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 136 * semantics. Address space is shared, and parent is blocked until child exit. 137 */ 138 /*ARGSUSED*/ 139 int 140 sys___vfork14(struct proc *p, void *v, register_t *retval) 141 { 142 143 return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 144 NULL, NULL, retval, NULL)); 145 } 146 147 /* 148 * Linux-compatible __clone(2) system call. 149 */ 150 int 151 sys___clone(struct proc *p, void *v, register_t *retval) 152 { 153 struct sys___clone_args /* { 154 syscallarg(int) flags; 155 syscallarg(void *) stack; 156 } */ *uap = v; 157 int flags, sig; 158 159 /* 160 * We don't support the CLONE_PID or CLONE_PTRACE flags. 161 */ 162 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE)) 163 return (EINVAL); 164 165 flags = 0; 166 167 if (SCARG(uap, flags) & CLONE_VM) 168 flags |= FORK_SHAREVM; 169 if (SCARG(uap, flags) & CLONE_FS) 170 flags |= FORK_SHARECWD; 171 if (SCARG(uap, flags) & CLONE_FILES) 172 flags |= FORK_SHAREFILES; 173 if (SCARG(uap, flags) & CLONE_SIGHAND) 174 flags |= FORK_SHARESIGS; 175 if (SCARG(uap, flags) & CLONE_VFORK) 176 flags |= FORK_PPWAIT; 177 178 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 179 if (sig < 0 || sig >= _NSIG) 180 return (EINVAL); 181 182 /* 183 * Note that the Linux API does not provide a portable way of 184 * specifying the stack area; the caller must know if the stack 185 * grows up or down. So, we pass a stack size of 0, so that the 186 * code that makes this adjustment is a noop. 187 */ 188 return (fork1(p, flags, sig, SCARG(uap, stack), 0, 189 NULL, NULL, retval, NULL)); 190 } 191 192 int 193 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize, 194 void (*func)(void *), void *arg, register_t *retval, 195 struct proc **rnewprocp) 196 { 197 struct proc *p2, *tp; 198 uid_t uid; 199 int count, s; 200 vaddr_t uaddr; 201 boolean_t inmem; 202 static int nextpid, pidchecked; 203 204 /* 205 * Although process entries are dynamically created, we still keep 206 * a global limit on the maximum number we will create. Don't allow 207 * a nonprivileged user to use the last process; don't let root 208 * exceed the limit. The variable nprocs is the current number of 209 * processes, maxproc is the limit. 210 */ 211 uid = p1->p_cred->p_ruid; 212 if (__predict_false((nprocs >= maxproc - 1 && uid != 0) || 213 nprocs >= maxproc)) { 214 tablefull("proc", "increase kern.maxproc or NPROC"); 215 return (EAGAIN); 216 } 217 nprocs++; 218 219 /* 220 * Increment the count of procs running with this uid. Don't allow 221 * a nonprivileged user to exceed their current limit. 222 */ 223 count = chgproccnt(uid, 1); 224 if (__predict_false(uid != 0 && count > 225 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 226 (void)chgproccnt(uid, -1); 227 nprocs--; 228 return (EAGAIN); 229 } 230 231 /* 232 * Allocate virtual address space for the U-area now, while it 233 * is still easy to abort the fork operation if we're out of 234 * kernel virtual address space. The actual U-area pages will 235 * be allocated and wired in uvm_fork() if needed. 236 */ 237 238 inmem = uvm_uarea_alloc(&uaddr); 239 if (__predict_false(uaddr == 0)) { 240 (void)chgproccnt(uid, -1); 241 nprocs--; 242 return (ENOMEM); 243 } 244 245 /* 246 * We are now committed to the fork. From here on, we may 247 * block on resources, but resource allocation may NOT fail. 248 */ 249 250 /* Allocate new proc. */ 251 p2 = pool_get(&proc_pool, PR_WAITOK); 252 253 /* 254 * Make a proc table entry for the new process. 255 * Start by zeroing the section of proc that is zero-initialized, 256 * then copy the section that is copied directly from the parent. 257 */ 258 memset(&p2->p_startzero, 0, 259 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 260 memcpy(&p2->p_startcopy, &p1->p_startcopy, 261 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 262 263 #if !defined(MULTIPROCESSOR) 264 /* 265 * In the single-processor case, all processes will always run 266 * on the same CPU. So, initialize the child's CPU to the parent's 267 * now. In the multiprocessor case, the child's CPU will be 268 * initialized in the low-level context switch code when the 269 * process runs. 270 */ 271 p2->p_cpu = p1->p_cpu; 272 #else 273 /* 274 * zero child's cpu pointer so we don't get trash. 275 */ 276 p2->p_cpu = NULL; 277 #endif /* ! MULTIPROCESSOR */ 278 279 /* 280 * Duplicate sub-structures as needed. 281 * Increase reference counts on shared objects. 282 * The p_stats and p_sigacts substructs are set in uvm_fork(). 283 */ 284 p2->p_flag = (inmem ? P_INMEM : 0) | (p1->p_flag & P_SUGID); 285 p2->p_emul = p1->p_emul; 286 p2->p_execsw = p1->p_execsw; 287 288 if (p1->p_flag & P_PROFIL) 289 startprofclock(p2); 290 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK); 291 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred)); 292 p2->p_cred->p_refcnt = 1; 293 crhold(p1->p_ucred); 294 295 LIST_INIT(&p2->p_raslist); 296 p2->p_nras = 0; 297 simple_lock_init(&p2->p_raslock); 298 #if defined(__HAVE_RAS) 299 ras_fork(p1, p2); 300 #endif 301 302 /* bump references to the text vnode (for procfs) */ 303 p2->p_textvp = p1->p_textvp; 304 if (p2->p_textvp) 305 VREF(p2->p_textvp); 306 307 if (flags & FORK_SHAREFILES) 308 fdshare(p1, p2); 309 else if (flags & FORK_CLEANFILES) 310 p2->p_fd = fdinit(p1); 311 else 312 p2->p_fd = fdcopy(p1); 313 314 if (flags & FORK_SHARECWD) 315 cwdshare(p1, p2); 316 else 317 p2->p_cwdi = cwdinit(p1); 318 319 /* 320 * If p_limit is still copy-on-write, bump refcnt, 321 * otherwise get a copy that won't be modified. 322 * (If PL_SHAREMOD is clear, the structure is shared 323 * copy-on-write.) 324 */ 325 if (p1->p_limit->p_lflags & PL_SHAREMOD) 326 p2->p_limit = limcopy(p1->p_limit); 327 else { 328 p2->p_limit = p1->p_limit; 329 p2->p_limit->p_refcnt++; 330 } 331 332 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 333 p2->p_flag |= P_CONTROLT; 334 if (flags & FORK_PPWAIT) 335 p2->p_flag |= P_PPWAIT; 336 LIST_INSERT_AFTER(p1, p2, p_pglist); 337 p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1; 338 LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling); 339 LIST_INIT(&p2->p_children); 340 341 callout_init(&p2->p_realit_ch); 342 callout_init(&p2->p_tsleep_ch); 343 344 #ifdef KTRACE 345 /* 346 * Copy traceflag and tracefile if enabled. 347 * If not inherited, these were zeroed above. 348 */ 349 if (p1->p_traceflag & KTRFAC_INHERIT) { 350 p2->p_traceflag = p1->p_traceflag; 351 if ((p2->p_tracep = p1->p_tracep) != NULL) 352 ktradref(p2); 353 } 354 #endif 355 356 scheduler_fork_hook(p1, p2); 357 358 /* 359 * Create signal actions for the child process. 360 */ 361 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 362 363 /* 364 * If emulation has process fork hook, call it now. 365 */ 366 if (p2->p_emul->e_proc_fork) 367 (*p2->p_emul->e_proc_fork)(p2, p1); 368 369 /* 370 * This begins the section where we must prevent the parent 371 * from being swapped. 372 */ 373 PHOLD(p1); 374 375 /* 376 * Finish creating the child process. It will return through a 377 * different path later. 378 */ 379 p2->p_addr = (struct user *)uaddr; 380 uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE, 381 stack, stacksize, 382 (func != NULL) ? func : child_return, 383 (arg != NULL) ? arg : p2); 384 385 /* 386 * BEGIN PID ALLOCATION. 387 */ 388 s = proclist_lock_write(); 389 390 /* 391 * Find an unused process ID. We remember a range of unused IDs 392 * ready to use (from nextpid+1 through pidchecked-1). 393 */ 394 nextpid++; 395 retry: 396 /* 397 * If the process ID prototype has wrapped around, 398 * restart somewhat above 0, as the low-numbered procs 399 * tend to include daemons that don't exit. 400 */ 401 if (nextpid >= PID_MAX) { 402 nextpid = 500; 403 pidchecked = 0; 404 } 405 if (nextpid >= pidchecked) { 406 const struct proclist_desc *pd; 407 408 pidchecked = PID_MAX; 409 /* 410 * Scan the process lists to check whether this pid 411 * is in use. Remember the lowest pid that's greater 412 * than nextpid, so we can avoid checking for a while. 413 */ 414 pd = proclists; 415 again: 416 LIST_FOREACH(tp, pd->pd_list, p_list) { 417 while (tp->p_pid == nextpid || 418 tp->p_pgrp->pg_id == nextpid || 419 tp->p_session->s_sid == nextpid) { 420 nextpid++; 421 if (nextpid >= pidchecked) 422 goto retry; 423 } 424 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 425 pidchecked = tp->p_pid; 426 427 if (tp->p_pgrp->pg_id > nextpid && 428 pidchecked > tp->p_pgrp->pg_id) 429 pidchecked = tp->p_pgrp->pg_id; 430 431 if (tp->p_session->s_sid > nextpid && 432 pidchecked > tp->p_session->s_sid) 433 pidchecked = tp->p_session->s_sid; 434 } 435 436 /* 437 * If there's another list, scan it. If we have checked 438 * them all, we've found one! 439 */ 440 pd++; 441 if (pd->pd_list != NULL) 442 goto again; 443 } 444 445 /* 446 * Put the proc on allproc before unlocking PID allocation 447 * so that waiters won't grab it as soon as we unlock. 448 */ 449 450 p2->p_stat = SIDL; /* protect against others */ 451 p2->p_pid = nextpid; 452 p2->p_exitsig = exitsig; /* signal for parent on exit */ 453 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 454 455 LIST_INSERT_HEAD(&allproc, p2, p_list); 456 457 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 458 459 /* 460 * END PID ALLOCATION. 461 */ 462 proclist_unlock_write(s); 463 464 #ifdef SYSTRACE 465 /* Tell systrace what's happening. */ 466 if (ISSET(p1->p_flag, P_SYSTRACE)) 467 systrace_sys_fork(p1, p2); 468 #endif 469 470 #ifdef __HAVE_SYSCALL_INTERN 471 (*p2->p_emul->e_syscall_intern)(p2); 472 #endif 473 474 /* 475 * Make child runnable, set start time, and add to run queue 476 * except if the parent requested the child to start in SSTOP state. 477 */ 478 SCHED_LOCK(s); 479 p2->p_stats->p_start = time; 480 p2->p_acflag = AFORK; 481 if (p1->p_flag & P_STOPFORK) { 482 p2->p_stat = SSTOP; 483 } else { 484 p2->p_stat = SRUN; 485 setrunqueue(p2); 486 } 487 SCHED_UNLOCK(s); 488 489 /* 490 * Inherit STOPFORK and STOPEXEC flags 491 */ 492 if (p1->p_flag & P_STOPFORK) 493 p2->p_flag |= P_STOPFORK; 494 if (p1->p_flag & P_STOPEXEC) 495 p2->p_flag |= P_STOPEXEC; 496 497 /* 498 * Now can be swapped. 499 */ 500 PRELE(p1); 501 502 /* 503 * Notify any interested parties about the new process. 504 */ 505 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 506 507 /* 508 * Update stats now that we know the fork was successful. 509 */ 510 uvmexp.forks++; 511 if (flags & FORK_PPWAIT) 512 uvmexp.forks_ppwait++; 513 if (flags & FORK_SHAREVM) 514 uvmexp.forks_sharevm++; 515 516 /* 517 * Pass a pointer to the new process to the caller. 518 */ 519 if (rnewprocp != NULL) 520 *rnewprocp = p2; 521 522 #ifdef KTRACE 523 if (KTRPOINT(p2, KTR_EMUL)) 524 ktremul(p2); 525 #endif 526 527 /* 528 * Preserve synchronization semantics of vfork. If waiting for 529 * child to exec or exit, set P_PPWAIT on child, and sleep on our 530 * proc (in case of exit). 531 */ 532 if (flags & FORK_PPWAIT) 533 while (p2->p_flag & P_PPWAIT) 534 tsleep(p1, PWAIT, "ppwait", 0); 535 536 /* 537 * Return child pid to parent process, 538 * marking us as parent via retval[1]. 539 */ 540 if (retval != NULL) { 541 retval[0] = p2->p_pid; 542 retval[1] = 0; 543 } 544 545 return (0); 546 } 547 548 #if defined(MULTIPROCESSOR) 549 /* 550 * XXX This is a slight hack to get newly-formed processes to 551 * XXX acquire the kernel lock as soon as they run. 552 */ 553 void 554 proc_trampoline_mp(void) 555 { 556 struct proc *p; 557 558 p = curproc; 559 560 SCHED_ASSERT_UNLOCKED(); 561 KERNEL_PROC_LOCK(p); 562 } 563 #endif 564
Indexes created Tue Sep 23 01:09:51 GMT 2025