1 /* $NetBSD: kern_fork.c,v 1.101 2002/12/05 16:24:46 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 1.101 2002/12/05 16:24:46 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 /*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 /* print the 'table full' message once per 10 seconds */ 193 struct timeval fork_tfmrate = { 10, 0 }; 194 195 int 196 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize, 197 void (*func)(void *), void *arg, register_t *retval, 198 struct proc **rnewprocp) 199 { 200 struct proc *p2, *tp; 201 uid_t uid; 202 int count, s; 203 vaddr_t uaddr; 204 boolean_t inmem; 205 static int nextpid, pidchecked; 206 207 /* 208 * Although process entries are dynamically created, we still keep 209 * a global limit on the maximum number we will create. Don't allow 210 * a nonprivileged user to use the last few processes; don't let root 211 * exceed the limit. The variable nprocs is the current number of 212 * processes, maxproc is the limit. 213 */ 214 uid = p1->p_cred->p_ruid; 215 if (__predict_false((nprocs >= maxproc - 5 && uid != 0) || 216 nprocs >= maxproc)) { 217 static struct timeval lasttfm; 218 219 if (ratecheck(&lasttfm, &fork_tfmrate)) 220 tablefull("proc", "increase kern.maxproc or NPROC"); 221 (void)tsleep(&nprocs, PUSER, "forkmx", hz / 2); 222 return (EAGAIN); 223 } 224 nprocs++; 225 226 /* 227 * Increment the count of procs running with this uid. Don't allow 228 * a nonprivileged user to exceed their current limit. 229 */ 230 count = chgproccnt(uid, 1); 231 if (__predict_false(uid != 0 && count > 232 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 233 (void)chgproccnt(uid, -1); 234 nprocs--; 235 (void)tsleep(&nprocs, PUSER, "forkulim", hz / 2); 236 return (EAGAIN); 237 } 238 239 /* 240 * Allocate virtual address space for the U-area now, while it 241 * is still easy to abort the fork operation if we're out of 242 * kernel virtual address space. The actual U-area pages will 243 * be allocated and wired in uvm_fork() if needed. 244 */ 245 246 inmem = uvm_uarea_alloc(&uaddr); 247 if (__predict_false(uaddr == 0)) { 248 (void)chgproccnt(uid, -1); 249 nprocs--; 250 return (ENOMEM); 251 } 252 253 /* 254 * We are now committed to the fork. From here on, we may 255 * block on resources, but resource allocation may NOT fail. 256 */ 257 258 /* Allocate new proc. */ 259 p2 = pool_get(&proc_pool, PR_WAITOK); 260 261 /* 262 * Make a proc table entry for the new process. 263 * Start by zeroing the section of proc that is zero-initialized, 264 * then copy the section that is copied directly from the parent. 265 */ 266 memset(&p2->p_startzero, 0, 267 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 268 memcpy(&p2->p_startcopy, &p1->p_startcopy, 269 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 270 271 #if !defined(MULTIPROCESSOR) 272 /* 273 * In the single-processor case, all processes will always run 274 * on the same CPU. So, initialize the child's CPU to the parent's 275 * now. In the multiprocessor case, the child's CPU will be 276 * initialized in the low-level context switch code when the 277 * process runs. 278 */ 279 p2->p_cpu = p1->p_cpu; 280 #else 281 /* 282 * zero child's cpu pointer so we don't get trash. 283 */ 284 p2->p_cpu = NULL; 285 #endif /* ! MULTIPROCESSOR */ 286 287 /* 288 * Duplicate sub-structures as needed. 289 * Increase reference counts on shared objects. 290 * The p_stats and p_sigacts substructs are set in uvm_fork(). 291 */ 292 p2->p_flag = (inmem ? P_INMEM : 0) | (p1->p_flag & P_SUGID); 293 p2->p_emul = p1->p_emul; 294 p2->p_execsw = p1->p_execsw; 295 296 if (p1->p_flag & P_PROFIL) 297 startprofclock(p2); 298 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK); 299 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred)); 300 p2->p_cred->p_refcnt = 1; 301 crhold(p1->p_ucred); 302 303 LIST_INIT(&p2->p_raslist); 304 p2->p_nras = 0; 305 simple_lock_init(&p2->p_raslock); 306 #if defined(__HAVE_RAS) 307 ras_fork(p1, p2); 308 #endif 309 310 /* bump references to the text vnode (for procfs) */ 311 p2->p_textvp = p1->p_textvp; 312 if (p2->p_textvp) 313 VREF(p2->p_textvp); 314 315 if (flags & FORK_SHAREFILES) 316 fdshare(p1, p2); 317 else if (flags & FORK_CLEANFILES) 318 p2->p_fd = fdinit(p1); 319 else 320 p2->p_fd = fdcopy(p1); 321 322 if (flags & FORK_SHARECWD) 323 cwdshare(p1, p2); 324 else 325 p2->p_cwdi = cwdinit(p1); 326 327 /* 328 * If p_limit is still copy-on-write, bump refcnt, 329 * otherwise get a copy that won't be modified. 330 * (If PL_SHAREMOD is clear, the structure is shared 331 * copy-on-write.) 332 */ 333 if (p1->p_limit->p_lflags & PL_SHAREMOD) 334 p2->p_limit = limcopy(p1->p_limit); 335 else { 336 p2->p_limit = p1->p_limit; 337 p2->p_limit->p_refcnt++; 338 } 339 340 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 341 p2->p_flag |= P_CONTROLT; 342 if (flags & FORK_PPWAIT) 343 p2->p_flag |= P_PPWAIT; 344 LIST_INSERT_AFTER(p1, p2, p_pglist); 345 p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1; 346 LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling); 347 LIST_INIT(&p2->p_children); 348 349 callout_init(&p2->p_realit_ch); 350 callout_init(&p2->p_tsleep_ch); 351 352 #ifdef KTRACE 353 /* 354 * Copy traceflag and tracefile if enabled. 355 * If not inherited, these were zeroed above. 356 */ 357 if (p1->p_traceflag & KTRFAC_INHERIT) { 358 p2->p_traceflag = p1->p_traceflag; 359 if ((p2->p_tracep = p1->p_tracep) != NULL) 360 ktradref(p2); 361 } 362 #endif 363 364 scheduler_fork_hook(p1, p2); 365 366 /* 367 * Create signal actions for the child process. 368 */ 369 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 370 371 /* 372 * If emulation has process fork hook, call it now. 373 */ 374 if (p2->p_emul->e_proc_fork) 375 (*p2->p_emul->e_proc_fork)(p2, p1); 376 377 /* 378 * This begins the section where we must prevent the parent 379 * from being swapped. 380 */ 381 PHOLD(p1); 382 383 /* 384 * Finish creating the child process. It will return through a 385 * different path later. 386 */ 387 p2->p_addr = (struct user *)uaddr; 388 uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE, 389 stack, stacksize, 390 (func != NULL) ? func : child_return, 391 (arg != NULL) ? arg : p2); 392 393 /* 394 * BEGIN PID ALLOCATION. 395 */ 396 s = proclist_lock_write(); 397 398 /* 399 * Find an unused process ID. We remember a range of unused IDs 400 * ready to use (from nextpid+1 through pidchecked-1). 401 */ 402 nextpid++; 403 retry: 404 /* 405 * If the process ID prototype has wrapped around, 406 * restart somewhat above 0, as the low-numbered procs 407 * tend to include daemons that don't exit. 408 */ 409 if (nextpid >= PID_MAX) { 410 nextpid = 500; 411 pidchecked = 0; 412 } 413 if (nextpid >= pidchecked) { 414 const struct proclist_desc *pd; 415 416 pidchecked = PID_MAX; 417 /* 418 * Scan the process lists to check whether this pid 419 * is in use. Remember the lowest pid that's greater 420 * than nextpid, so we can avoid checking for a while. 421 */ 422 pd = proclists; 423 again: 424 LIST_FOREACH(tp, pd->pd_list, p_list) { 425 while (tp->p_pid == nextpid || 426 tp->p_pgrp->pg_id == nextpid || 427 tp->p_session->s_sid == nextpid) { 428 nextpid++; 429 if (nextpid >= pidchecked) 430 goto retry; 431 } 432 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 433 pidchecked = tp->p_pid; 434 435 if (tp->p_pgrp->pg_id > nextpid && 436 pidchecked > tp->p_pgrp->pg_id) 437 pidchecked = tp->p_pgrp->pg_id; 438 439 if (tp->p_session->s_sid > nextpid && 440 pidchecked > tp->p_session->s_sid) 441 pidchecked = tp->p_session->s_sid; 442 } 443 444 /* 445 * If there's another list, scan it. If we have checked 446 * them all, we've found one! 447 */ 448 pd++; 449 if (pd->pd_list != NULL) 450 goto again; 451 } 452 453 /* 454 * Put the proc on allproc before unlocking PID allocation 455 * so that waiters won't grab it as soon as we unlock. 456 */ 457 458 p2->p_stat = SIDL; /* protect against others */ 459 p2->p_pid = nextpid; 460 p2->p_exitsig = exitsig; /* signal for parent on exit */ 461 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 462 463 LIST_INSERT_HEAD(&allproc, p2, p_list); 464 465 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 466 467 /* 468 * END PID ALLOCATION. 469 */ 470 proclist_unlock_write(s); 471 472 #ifdef SYSTRACE 473 /* Tell systrace what's happening. */ 474 if (ISSET(p1->p_flag, P_SYSTRACE)) 475 systrace_sys_fork(p1, p2); 476 #endif 477 478 #ifdef __HAVE_SYSCALL_INTERN 479 (*p2->p_emul->e_syscall_intern)(p2); 480 #endif 481 482 /* 483 * Make child runnable, set start time, and add to run queue 484 * except if the parent requested the child to start in SSTOP state. 485 */ 486 SCHED_LOCK(s); 487 p2->p_stats->p_start = time; 488 p2->p_acflag = AFORK; 489 if (p1->p_flag & P_STOPFORK) { 490 p2->p_stat = SSTOP; 491 } else { 492 p2->p_stat = SRUN; 493 setrunqueue(p2); 494 } 495 SCHED_UNLOCK(s); 496 497 /* 498 * Inherit STOPFORK and STOPEXEC flags 499 */ 500 if (p1->p_flag & P_STOPFORK) 501 p2->p_flag |= P_STOPFORK; 502 if (p1->p_flag & P_STOPEXEC) 503 p2->p_flag |= P_STOPEXEC; 504 505 /* 506 * Now can be swapped. 507 */ 508 PRELE(p1); 509 510 /* 511 * Notify any interested parties about the new process. 512 */ 513 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 514 515 /* 516 * Update stats now that we know the fork was successful. 517 */ 518 uvmexp.forks++; 519 if (flags & FORK_PPWAIT) 520 uvmexp.forks_ppwait++; 521 if (flags & FORK_SHAREVM) 522 uvmexp.forks_sharevm++; 523 524 /* 525 * Pass a pointer to the new process to the caller. 526 */ 527 if (rnewprocp != NULL) 528 *rnewprocp = p2; 529 530 #ifdef KTRACE 531 if (KTRPOINT(p2, KTR_EMUL)) 532 ktremul(p2); 533 #endif 534 535 /* 536 * Preserve synchronization semantics of vfork. If waiting for 537 * child to exec or exit, set P_PPWAIT on child, and sleep on our 538 * proc (in case of exit). 539 */ 540 if (flags & FORK_PPWAIT) 541 while (p2->p_flag & P_PPWAIT) 542 tsleep(p1, PWAIT, "ppwait", 0); 543 544 /* 545 * Return child pid to parent process, 546 * marking us as parent via retval[1]. 547 */ 548 if (retval != NULL) { 549 retval[0] = p2->p_pid; 550 retval[1] = 0; 551 } 552 553 return (0); 554 } 555 556 #if defined(MULTIPROCESSOR) 557 /* 558 * XXX This is a slight hack to get newly-formed processes to 559 * XXX acquire the kernel lock as soon as they run. 560 */ 561 void 562 proc_trampoline_mp(void) 563 { 564 struct proc *p; 565 566 p = curproc; 567 568 SCHED_ASSERT_UNLOCKED(); 569 KERNEL_PROC_LOCK(p); 570 } 571 #endif 572