kern_fork.c revision 1.104
1 /* $NetBSD: kern_fork.c,v 1.104 2002/12/12 20:41:45 jdolecek 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 1.104 2002/12/12 20:41:45 jdolecek 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 /* 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 proc *p, void *v, register_t *retval) 122 { 123 124 return (fork1(p, 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 proc *p, void *v, register_t *retval) 134 { 135 136 return (fork1(p, 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 proc *p, void *v, register_t *retval) 147 { 148 149 return (fork1(p, 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 proc *p, 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 flags = 0; 172 173 if (SCARG(uap, flags) & CLONE_VM) 174 flags |= FORK_SHAREVM; 175 if (SCARG(uap, flags) & CLONE_FS) 176 flags |= FORK_SHARECWD; 177 if (SCARG(uap, flags) & CLONE_FILES) 178 flags |= FORK_SHAREFILES; 179 if (SCARG(uap, flags) & CLONE_SIGHAND) 180 flags |= FORK_SHARESIGS; 181 if (SCARG(uap, flags) & CLONE_VFORK) 182 flags |= FORK_PPWAIT; 183 184 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 185 if (sig < 0 || sig >= _NSIG) 186 return (EINVAL); 187 188 /* 189 * Note that the Linux API does not provide a portable way of 190 * specifying the stack area; the caller must know if the stack 191 * grows up or down. So, we pass a stack size of 0, so that the 192 * code that makes this adjustment is a noop. 193 */ 194 return (fork1(p, flags, sig, SCARG(uap, stack), 0, 195 NULL, NULL, retval, NULL)); 196 } 197 198 /* print the 'table full' message once per 10 seconds */ 199 struct timeval fork_tfmrate = { 10, 0 }; 200 201 int 202 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize, 203 void (*func)(void *), void *arg, register_t *retval, 204 struct proc **rnewprocp) 205 { 206 struct proc *p2, *tp; 207 uid_t uid; 208 int count, s; 209 vaddr_t uaddr; 210 boolean_t inmem; 211 static int nextpid, pidchecked; 212 213 /* 214 * Although process entries are dynamically created, we still keep 215 * a global limit on the maximum number we will create. Don't allow 216 * a nonprivileged user to use the last few processes; don't let root 217 * exceed the limit. The variable nprocs is the current number of 218 * processes, maxproc is the limit. 219 */ 220 uid = p1->p_cred->p_ruid; 221 if (__predict_false((nprocs >= maxproc - 5 && uid != 0) || 222 nprocs >= maxproc)) { 223 static struct timeval lasttfm; 224 225 if (ratecheck(&lasttfm, &fork_tfmrate)) 226 tablefull("proc", "increase kern.maxproc or NPROC"); 227 if (forkfsleep) 228 (void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep); 229 return (EAGAIN); 230 } 231 nprocs++; 232 233 /* 234 * Increment the count of procs running with this uid. Don't allow 235 * a nonprivileged user to exceed their current limit. 236 */ 237 count = chgproccnt(uid, 1); 238 if (__predict_false(uid != 0 && count > 239 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 240 (void)chgproccnt(uid, -1); 241 nprocs--; 242 if (forkfsleep) 243 (void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep); 244 return (EAGAIN); 245 } 246 247 /* 248 * Allocate virtual address space for the U-area now, while it 249 * is still easy to abort the fork operation if we're out of 250 * kernel virtual address space. The actual U-area pages will 251 * be allocated and wired in uvm_fork() if needed. 252 */ 253 254 inmem = uvm_uarea_alloc(&uaddr); 255 if (__predict_false(uaddr == 0)) { 256 (void)chgproccnt(uid, -1); 257 nprocs--; 258 return (ENOMEM); 259 } 260 261 /* 262 * We are now committed to the fork. From here on, we may 263 * block on resources, but resource allocation may NOT fail. 264 */ 265 266 /* Allocate new proc. */ 267 p2 = pool_get(&proc_pool, PR_WAITOK); 268 269 /* 270 * Make a proc table entry for the new process. 271 * Start by zeroing the section of proc that is zero-initialized, 272 * then copy the section that is copied directly from the parent. 273 */ 274 memset(&p2->p_startzero, 0, 275 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 276 memcpy(&p2->p_startcopy, &p1->p_startcopy, 277 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 278 279 #if !defined(MULTIPROCESSOR) 280 /* 281 * In the single-processor case, all processes will always run 282 * on the same CPU. So, initialize the child's CPU to the parent's 283 * now. In the multiprocessor case, the child's CPU will be 284 * initialized in the low-level context switch code when the 285 * process runs. 286 */ 287 p2->p_cpu = p1->p_cpu; 288 #else 289 /* 290 * zero child's cpu pointer so we don't get trash. 291 */ 292 p2->p_cpu = NULL; 293 #endif /* ! MULTIPROCESSOR */ 294 295 /* 296 * Duplicate sub-structures as needed. 297 * Increase reference counts on shared objects. 298 * The p_stats and p_sigacts substructs are set in uvm_fork(). 299 */ 300 p2->p_flag = (inmem ? P_INMEM : 0) | (p1->p_flag & P_SUGID); 301 p2->p_emul = p1->p_emul; 302 p2->p_execsw = p1->p_execsw; 303 304 if (p1->p_flag & P_PROFIL) 305 startprofclock(p2); 306 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK); 307 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred)); 308 p2->p_cred->p_refcnt = 1; 309 crhold(p1->p_ucred); 310 311 LIST_INIT(&p2->p_raslist); 312 p2->p_nras = 0; 313 simple_lock_init(&p2->p_raslock); 314 #if defined(__HAVE_RAS) 315 ras_fork(p1, p2); 316 #endif 317 318 /* bump references to the text vnode (for procfs) */ 319 p2->p_textvp = p1->p_textvp; 320 if (p2->p_textvp) 321 VREF(p2->p_textvp); 322 323 if (flags & FORK_SHAREFILES) 324 fdshare(p1, p2); 325 else if (flags & FORK_CLEANFILES) 326 p2->p_fd = fdinit(p1); 327 else 328 p2->p_fd = fdcopy(p1); 329 330 if (flags & FORK_SHARECWD) 331 cwdshare(p1, p2); 332 else 333 p2->p_cwdi = cwdinit(p1); 334 335 /* 336 * If p_limit is still copy-on-write, bump refcnt, 337 * otherwise get a copy that won't be modified. 338 * (If PL_SHAREMOD is clear, the structure is shared 339 * copy-on-write.) 340 */ 341 if (p1->p_limit->p_lflags & PL_SHAREMOD) 342 p2->p_limit = limcopy(p1->p_limit); 343 else { 344 p2->p_limit = p1->p_limit; 345 p2->p_limit->p_refcnt++; 346 } 347 348 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 349 p2->p_flag |= P_CONTROLT; 350 if (flags & FORK_PPWAIT) 351 p2->p_flag |= P_PPWAIT; 352 LIST_INSERT_AFTER(p1, p2, p_pglist); 353 p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1; 354 LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling); 355 LIST_INIT(&p2->p_children); 356 357 callout_init(&p2->p_realit_ch); 358 callout_init(&p2->p_tsleep_ch); 359 360 #ifdef KTRACE 361 /* 362 * Copy traceflag and tracefile if enabled. 363 * If not inherited, these were zeroed above. 364 */ 365 if (p1->p_traceflag & KTRFAC_INHERIT) { 366 p2->p_traceflag = p1->p_traceflag; 367 if ((p2->p_tracep = p1->p_tracep) != NULL) 368 ktradref(p2); 369 } 370 #endif 371 372 scheduler_fork_hook(p1, p2); 373 374 /* 375 * Create signal actions for the child process. 376 */ 377 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 378 379 /* 380 * If emulation has process fork hook, call it now. 381 */ 382 if (p2->p_emul->e_proc_fork) 383 (*p2->p_emul->e_proc_fork)(p2, p1); 384 385 /* 386 * This begins the section where we must prevent the parent 387 * from being swapped. 388 */ 389 PHOLD(p1); 390 391 /* 392 * Finish creating the child process. It will return through a 393 * different path later. 394 */ 395 p2->p_addr = (struct user *)uaddr; 396 uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE, 397 stack, stacksize, 398 (func != NULL) ? func : child_return, 399 (arg != NULL) ? arg : p2); 400 401 /* 402 * BEGIN PID ALLOCATION. 403 */ 404 s = proclist_lock_write(); 405 406 /* 407 * Find an unused process ID. We remember a range of unused IDs 408 * ready to use (from nextpid+1 through pidchecked-1). 409 */ 410 nextpid++; 411 retry: 412 /* 413 * If the process ID prototype has wrapped around, 414 * restart somewhat above 0, as the low-numbered procs 415 * tend to include daemons that don't exit. 416 */ 417 if (nextpid >= PID_MAX) { 418 nextpid = PID_SKIP; 419 pidchecked = 0; 420 } 421 if (nextpid >= pidchecked) { 422 const struct proclist_desc *pd; 423 424 pidchecked = PID_MAX; 425 /* 426 * Scan the process lists to check whether this pid 427 * is in use. Remember the lowest pid that's greater 428 * than nextpid, so we can avoid checking for a while. 429 */ 430 pd = proclists; 431 again: 432 LIST_FOREACH(tp, pd->pd_list, p_list) { 433 while (tp->p_pid == nextpid || 434 tp->p_pgrp->pg_id == nextpid || 435 tp->p_session->s_sid == nextpid) { 436 nextpid++; 437 if (nextpid >= pidchecked) 438 goto retry; 439 } 440 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 441 pidchecked = tp->p_pid; 442 443 if (tp->p_pgrp->pg_id > nextpid && 444 pidchecked > tp->p_pgrp->pg_id) 445 pidchecked = tp->p_pgrp->pg_id; 446 447 if (tp->p_session->s_sid > nextpid && 448 pidchecked > tp->p_session->s_sid) 449 pidchecked = tp->p_session->s_sid; 450 } 451 452 /* 453 * If there's another list, scan it. If we have checked 454 * them all, we've found one! 455 */ 456 pd++; 457 if (pd->pd_list != NULL) 458 goto again; 459 } 460 461 /* 462 * Put the proc on allproc before unlocking PID allocation 463 * so that waiters won't grab it as soon as we unlock. 464 */ 465 466 p2->p_stat = SIDL; /* protect against others */ 467 p2->p_pid = nextpid; 468 p2->p_exitsig = exitsig; /* signal for parent on exit */ 469 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 470 471 LIST_INSERT_HEAD(&allproc, p2, p_list); 472 473 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 474 475 /* 476 * END PID ALLOCATION. 477 */ 478 proclist_unlock_write(s); 479 480 #ifdef SYSTRACE 481 /* Tell systrace what's happening. */ 482 if (ISSET(p1->p_flag, P_SYSTRACE)) 483 systrace_sys_fork(p1, p2); 484 #endif 485 486 #ifdef __HAVE_SYSCALL_INTERN 487 (*p2->p_emul->e_syscall_intern)(p2); 488 #endif 489 490 /* 491 * Make child runnable, set start time, and add to run queue 492 * except if the parent requested the child to start in SSTOP state. 493 */ 494 SCHED_LOCK(s); 495 p2->p_stats->p_start = time; 496 p2->p_acflag = AFORK; 497 if (p1->p_flag & P_STOPFORK) { 498 p2->p_stat = SSTOP; 499 } else { 500 p2->p_stat = SRUN; 501 setrunqueue(p2); 502 } 503 SCHED_UNLOCK(s); 504 505 /* 506 * Inherit STOPFORK and STOPEXEC flags 507 */ 508 if (p1->p_flag & P_STOPFORK) 509 p2->p_flag |= P_STOPFORK; 510 if (p1->p_flag & P_STOPEXEC) 511 p2->p_flag |= P_STOPEXEC; 512 513 /* 514 * Now can be swapped. 515 */ 516 PRELE(p1); 517 518 /* 519 * Notify any interested parties about the new process. 520 */ 521 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 522 523 /* 524 * Update stats now that we know the fork was successful. 525 */ 526 uvmexp.forks++; 527 if (flags & FORK_PPWAIT) 528 uvmexp.forks_ppwait++; 529 if (flags & FORK_SHAREVM) 530 uvmexp.forks_sharevm++; 531 532 /* 533 * Pass a pointer to the new process to the caller. 534 */ 535 if (rnewprocp != NULL) 536 *rnewprocp = p2; 537 538 #ifdef KTRACE 539 if (KTRPOINT(p2, KTR_EMUL)) 540 ktremul(p2); 541 #endif 542 543 /* 544 * Preserve synchronization semantics of vfork. If waiting for 545 * child to exec or exit, set P_PPWAIT on child, and sleep on our 546 * proc (in case of exit). 547 */ 548 if (flags & FORK_PPWAIT) 549 while (p2->p_flag & P_PPWAIT) 550 tsleep(p1, PWAIT, "ppwait", 0); 551 552 /* 553 * Return child pid to parent process, 554 * marking us as parent via retval[1]. 555 */ 556 if (retval != NULL) { 557 retval[0] = p2->p_pid; 558 retval[1] = 0; 559 } 560 561 return (0); 562 } 563 564 #if defined(MULTIPROCESSOR) 565 /* 566 * XXX This is a slight hack to get newly-formed processes to 567 * XXX acquire the kernel lock as soon as they run. 568 */ 569 void 570 proc_trampoline_mp(void) 571 { 572 struct proc *p; 573 574 p = curproc; 575 576 SCHED_ASSERT_UNLOCKED(); 577 KERNEL_PROC_LOCK(p); 578 } 579 #endif 580
Indexes created Mon Sep 22 13:09:51 GMT 2025