1 /* $NetBSD: kern_fork.c,v 1.106 2003/01/24 01:42:53 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 1.106 2003/01/24 01:42:53 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 /* 115 * Number of ticks to sleep if fork() would fail due to process hitting 116 * limits. Exported in miliseconds to userland via sysctl. 117 */ 118 int forkfsleep = 0; 119 120 /*ARGSUSED*/ 121 int 122 sys_fork(struct lwp *l, void *v, register_t *retval) 123 { 124 125 return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 126 } 127 128 /* 129 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 130 * Address space is not shared, but parent is blocked until child exit. 131 */ 132 /*ARGSUSED*/ 133 int 134 sys_vfork(struct lwp *l, void *v, register_t *retval) 135 { 136 137 return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 138 retval, NULL)); 139 } 140 141 /* 142 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 143 * semantics. Address space is shared, and parent is blocked until child exit. 144 */ 145 /*ARGSUSED*/ 146 int 147 sys___vfork14(struct lwp *l, void *v, register_t *retval) 148 { 149 150 return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 151 NULL, NULL, retval, NULL)); 152 } 153 154 /* 155 * Linux-compatible __clone(2) system call. 156 */ 157 int 158 sys___clone(struct lwp *l, void *v, register_t *retval) 159 { 160 struct sys___clone_args /* { 161 syscallarg(int) flags; 162 syscallarg(void *) stack; 163 } */ *uap = v; 164 int flags, sig; 165 166 /* 167 * We don't support the CLONE_PID or CLONE_PTRACE flags. 168 */ 169 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE)) 170 return (EINVAL); 171 172 flags = 0; 173 174 if (SCARG(uap, flags) & CLONE_VM) 175 flags |= FORK_SHAREVM; 176 if (SCARG(uap, flags) & CLONE_FS) 177 flags |= FORK_SHARECWD; 178 if (SCARG(uap, flags) & CLONE_FILES) 179 flags |= FORK_SHAREFILES; 180 if (SCARG(uap, flags) & CLONE_SIGHAND) 181 flags |= FORK_SHARESIGS; 182 if (SCARG(uap, flags) & CLONE_VFORK) 183 flags |= FORK_PPWAIT; 184 185 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 186 if (sig < 0 || sig >= _NSIG) 187 return (EINVAL); 188 189 /* 190 * Note that the Linux API does not provide a portable way of 191 * specifying the stack area; the caller must know if the stack 192 * grows up or down. So, we pass a stack size of 0, so that the 193 * code that makes this adjustment is a noop. 194 */ 195 return (fork1(l, flags, sig, SCARG(uap, stack), 0, 196 NULL, NULL, retval, NULL)); 197 } 198 199 /* print the 'table full' message once per 10 seconds */ 200 struct timeval fork_tfmrate = { 10, 0 }; 201 202 int 203 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize, 204 void (*func)(void *), void *arg, register_t *retval, 205 struct proc **rnewprocp) 206 { 207 struct proc *p1, *p2, *tp; 208 uid_t uid; 209 struct lwp *l2; 210 int count, s; 211 vaddr_t uaddr; 212 boolean_t inmem; 213 static int nextpid, pidchecked; 214 215 /* 216 * Although process entries are dynamically created, we still keep 217 * a global limit on the maximum number we will create. Don't allow 218 * a nonprivileged user to use the last few processes; don't let root 219 * exceed the limit. The variable nprocs is the current number of 220 * processes, maxproc is the limit. 221 */ 222 p1 = l1->l_proc; 223 uid = p1->p_cred->p_ruid; 224 if (__predict_false((nprocs >= maxproc - 5 && uid != 0) || 225 nprocs >= maxproc)) { 226 static struct timeval lasttfm; 227 228 if (ratecheck(&lasttfm, &fork_tfmrate)) 229 tablefull("proc", "increase kern.maxproc or NPROC"); 230 if (forkfsleep) 231 (void)tsleep(&nprocs, PUSER, "forkmx", forkfsleep); 232 return (EAGAIN); 233 } 234 nprocs++; 235 236 /* 237 * Increment the count of procs running with this uid. Don't allow 238 * a nonprivileged user to exceed their current limit. 239 */ 240 count = chgproccnt(uid, 1); 241 if (__predict_false(uid != 0 && count > 242 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 243 (void)chgproccnt(uid, -1); 244 nprocs--; 245 if (forkfsleep) 246 (void)tsleep(&nprocs, PUSER, "forkulim", forkfsleep); 247 return (EAGAIN); 248 } 249 250 /* 251 * Allocate virtual address space for the U-area now, while it 252 * is still easy to abort the fork operation if we're out of 253 * kernel virtual address space. The actual U-area pages will 254 * be allocated and wired in uvm_fork() if needed. 255 */ 256 257 inmem = uvm_uarea_alloc(&uaddr); 258 if (__predict_false(uaddr == 0)) { 259 (void)chgproccnt(uid, -1); 260 nprocs--; 261 return (ENOMEM); 262 } 263 264 /* 265 * We are now committed to the fork. From here on, we may 266 * block on resources, but resource allocation may NOT fail. 267 */ 268 269 /* Allocate new proc. */ 270 p2 = pool_get(&proc_pool, PR_WAITOK); 271 272 /* 273 * Make a proc table entry for the new process. 274 * Start by zeroing the section of proc that is zero-initialized, 275 * then copy the section that is copied directly from the parent. 276 */ 277 memset(&p2->p_startzero, 0, 278 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 279 memcpy(&p2->p_startcopy, &p1->p_startcopy, 280 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 281 282 simple_lock_init(&p2->p_lwplock); 283 LIST_INIT(&p2->p_lwps); 284 285 /* 286 * Duplicate sub-structures as needed. 287 * Increase reference counts on shared objects. 288 * The p_stats and p_sigacts substructs are set in uvm_fork(). 289 */ 290 p2->p_flag = (p1->p_flag & P_SUGID); 291 p2->p_emul = p1->p_emul; 292 p2->p_execsw = p1->p_execsw; 293 294 if (p1->p_flag & P_PROFIL) 295 startprofclock(p2); 296 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK); 297 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred)); 298 p2->p_cred->p_refcnt = 1; 299 crhold(p1->p_ucred); 300 301 LIST_INIT(&p2->p_raslist); 302 p2->p_nras = 0; 303 simple_lock_init(&p2->p_raslock); 304 #if defined(__HAVE_RAS) 305 ras_fork(p1, p2); 306 #endif 307 308 /* bump references to the text vnode (for procfs) */ 309 p2->p_textvp = p1->p_textvp; 310 if (p2->p_textvp) 311 VREF(p2->p_textvp); 312 313 if (flags & FORK_SHAREFILES) 314 fdshare(p1, p2); 315 else if (flags & FORK_CLEANFILES) 316 p2->p_fd = fdinit(p1); 317 else 318 p2->p_fd = fdcopy(p1); 319 320 if (flags & FORK_SHARECWD) 321 cwdshare(p1, p2); 322 else 323 p2->p_cwdi = cwdinit(p1); 324 325 /* 326 * If p_limit is still copy-on-write, bump refcnt, 327 * otherwise get a copy that won't be modified. 328 * (If PL_SHAREMOD is clear, the structure is shared 329 * copy-on-write.) 330 */ 331 if (p1->p_limit->p_lflags & PL_SHAREMOD) 332 p2->p_limit = limcopy(p1->p_limit); 333 else { 334 p2->p_limit = p1->p_limit; 335 p2->p_limit->p_refcnt++; 336 } 337 338 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 339 p2->p_flag |= P_CONTROLT; 340 if (flags & FORK_PPWAIT) 341 p2->p_flag |= P_PPWAIT; 342 LIST_INSERT_AFTER(p1, p2, p_pglist); 343 p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1; 344 LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling); 345 LIST_INIT(&p2->p_children); 346 347 #ifdef KTRACE 348 /* 349 * Copy traceflag and tracefile if enabled. 350 * If not inherited, these were zeroed above. 351 */ 352 if (p1->p_traceflag & KTRFAC_INHERIT) { 353 p2->p_traceflag = p1->p_traceflag; 354 if ((p2->p_tracep = p1->p_tracep) != NULL) 355 ktradref(p2); 356 } 357 #endif 358 359 scheduler_fork_hook(p1, p2); 360 361 /* 362 * Create signal actions for the child process. 363 */ 364 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 365 366 /* 367 * p_stats. 368 * Copy parts of p_stats, and zero out the rest. 369 */ 370 p2->p_stats = pstatscopy(p1->p_stats); 371 372 /* 373 * If emulation has process fork hook, call it now. 374 */ 375 if (p2->p_emul->e_proc_fork) 376 (*p2->p_emul->e_proc_fork)(p2, p1); 377 378 /* 379 * ...and finally, any other random fork hooks that subsystems 380 * might have registered. 381 */ 382 doforkhooks(p2, p1); 383 384 /* 385 * This begins the section where we must prevent the parent 386 * from being swapped. 387 */ 388 PHOLD(l1); 389 390 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE); 391 392 /* 393 * Finish creating the child process. It will return through a 394 * different path later. 395 */ 396 newlwp(l1, p2, uaddr, inmem, 0, stack, stacksize, 397 (func != NULL) ? func : child_return, 398 arg, &l2); 399 400 /* 401 * BEGIN PID ALLOCATION. 402 */ 403 s = proclist_lock_write(); 404 405 /* 406 * Find an unused process ID. We remember a range of unused IDs 407 * ready to use (from nextpid+1 through pidchecked-1). 408 */ 409 nextpid++; 410 retry: 411 /* 412 * If the process ID prototype has wrapped around, 413 * restart somewhat above 0, as the low-numbered procs 414 * tend to include daemons that don't exit. 415 */ 416 if (nextpid >= PID_MAX) { 417 nextpid = PID_SKIP; 418 pidchecked = 0; 419 } 420 if (nextpid >= pidchecked) { 421 const struct proclist_desc *pd; 422 423 pidchecked = PID_MAX; 424 /* 425 * Scan the process lists to check whether this pid 426 * is in use. Remember the lowest pid that's greater 427 * than nextpid, so we can avoid checking for a while. 428 */ 429 pd = proclists; 430 again: 431 LIST_FOREACH(tp, pd->pd_list, p_list) { 432 while (tp->p_pid == nextpid || 433 tp->p_pgrp->pg_id == nextpid || 434 tp->p_session->s_sid == nextpid) { 435 nextpid++; 436 if (nextpid >= pidchecked) 437 goto retry; 438 } 439 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 440 pidchecked = tp->p_pid; 441 442 if (tp->p_pgrp->pg_id > nextpid && 443 pidchecked > tp->p_pgrp->pg_id) 444 pidchecked = tp->p_pgrp->pg_id; 445 446 if (tp->p_session->s_sid > nextpid && 447 pidchecked > tp->p_session->s_sid) 448 pidchecked = tp->p_session->s_sid; 449 } 450 451 /* 452 * If there's another list, scan it. If we have checked 453 * them all, we've found one! 454 */ 455 pd++; 456 if (pd->pd_list != NULL) 457 goto again; 458 } 459 460 /* 461 * Put the proc on allproc before unlocking PID allocation 462 * so that waiters won't grab it as soon as we unlock. 463 */ 464 465 p2->p_stat = SIDL; /* protect against others */ 466 p2->p_pid = nextpid; 467 p2->p_exitsig = exitsig; /* signal for parent on exit */ 468 469 LIST_INSERT_HEAD(&allproc, p2, p_list); 470 471 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 472 473 /* 474 * END PID ALLOCATION. 475 */ 476 proclist_unlock_write(s); 477 478 #ifdef SYSTRACE 479 /* Tell systrace what's happening. */ 480 if (ISSET(p1->p_flag, P_SYSTRACE)) 481 systrace_sys_fork(p1, p2); 482 #endif 483 484 #ifdef __HAVE_SYSCALL_INTERN 485 (*p2->p_emul->e_syscall_intern)(p2); 486 #endif 487 488 /* 489 * Make child runnable, set start time, and add to run queue 490 * except if the parent requested the child to start in SSTOP state. 491 */ 492 SCHED_LOCK(s); 493 p2->p_stats->p_start = time; 494 p2->p_acflag = AFORK; 495 p2->p_nrlwps = 1; 496 if (p1->p_flag & P_STOPFORK) { 497 p2->p_stat = SSTOP; 498 l2->l_stat = LSSTOP; 499 } else { 500 p2->p_stat = SACTIVE; 501 l2->l_stat = LSRUN; 502 setrunqueue(l2); 503 } 504 SCHED_UNLOCK(s); 505 506 /* 507 * Inherit STOPFORK and STOPEXEC flags 508 */ 509 if (p1->p_flag & P_STOPFORK) 510 p2->p_flag |= P_STOPFORK; 511 if (p1->p_flag & P_STOPEXEC) 512 p2->p_flag |= P_STOPEXEC; 513 514 /* 515 * Now can be swapped. 516 */ 517 PRELE(l1); 518 519 /* 520 * Notify any interested parties about the new process. 521 */ 522 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 523 524 /* 525 * Update stats now that we know the fork was successful. 526 */ 527 uvmexp.forks++; 528 if (flags & FORK_PPWAIT) 529 uvmexp.forks_ppwait++; 530 if (flags & FORK_SHAREVM) 531 uvmexp.forks_sharevm++; 532 533 /* 534 * Pass a pointer to the new process to the caller. 535 */ 536 if (rnewprocp != NULL) 537 *rnewprocp = p2; 538 539 #ifdef KTRACE 540 if (KTRPOINT(p2, KTR_EMUL)) 541 ktremul(p2); 542 #endif 543 544 /* 545 * Preserve synchronization semantics of vfork. If waiting for 546 * child to exec or exit, set P_PPWAIT on child, and sleep on our 547 * proc (in case of exit). 548 */ 549 if (flags & FORK_PPWAIT) 550 while (p2->p_flag & P_PPWAIT) 551 tsleep(p1, PWAIT, "ppwait", 0); 552 553 /* 554 * Return child pid to parent process, 555 * marking us as parent via retval[1]. 556 */ 557 if (retval != NULL) { 558 retval[0] = p2->p_pid; 559 retval[1] = 0; 560 } 561 562 return (0); 563 } 564 565 #if defined(MULTIPROCESSOR) 566 /* 567 * XXX This is a slight hack to get newly-formed processes to 568 * XXX acquire the kernel lock as soon as they run. 569 */ 570 void 571 proc_trampoline_mp(void) 572 { 573 struct lwp *l; 574 575 l = curlwp; 576 577 SCHED_ASSERT_UNLOCKED(); 578 KERNEL_PROC_LOCK(l); 579 } 580 #endif 581