1 /* $NetBSD: kern_fork.c,v 1.92 2002/08/28 07:16:37 gmcgarry 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 1.92 2002/08/28 07:16:37 gmcgarry 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/map.h> 90 #include <sys/filedesc.h> 91 #include <sys/kernel.h> 92 #include <sys/malloc.h> 93 #include <sys/pool.h> 94 #include <sys/mount.h> 95 #include <sys/proc.h> 96 #include <sys/ras.h> 97 #include <sys/resourcevar.h> 98 #include <sys/vnode.h> 99 #include <sys/file.h> 100 #include <sys/acct.h> 101 #include <sys/ktrace.h> 102 #include <sys/vmmeter.h> 103 #include <sys/sched.h> 104 #include <sys/signalvar.h> 105 #include <sys/systrace.h> 106 107 #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 proc *p, void *v, register_t *retval) 117 { 118 119 return (fork1(p, 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 proc *p, void *v, register_t *retval) 129 { 130 131 return (fork1(p, 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 proc *p, void *v, register_t *retval) 142 { 143 144 return (fork1(p, 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 proc *p, 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(p, flags, sig, SCARG(uap, stack), 0, 190 NULL, NULL, retval, NULL)); 191 } 192 193 int 194 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize, 195 void (*func)(void *), void *arg, register_t *retval, 196 struct proc **rnewprocp) 197 { 198 struct proc *p2, *tp; 199 uid_t uid; 200 int count, s; 201 vaddr_t uaddr; 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 vm_fork(). 236 */ 237 238 #ifndef USPACE_ALIGN 239 #define USPACE_ALIGN 0 240 #endif 241 242 uaddr = uvm_km_valloc_align(kernel_map, USPACE, USPACE_ALIGN); 243 if (__predict_false(uaddr == 0)) { 244 (void)chgproccnt(uid, -1); 245 nprocs--; 246 return (ENOMEM); 247 } 248 249 /* 250 * We are now committed to the fork. From here on, we may 251 * block on resources, but resource allocation may NOT fail. 252 */ 253 254 /* Allocate new proc. */ 255 p2 = pool_get(&proc_pool, PR_WAITOK); 256 257 /* 258 * BEGIN PID ALLOCATION. 259 */ 260 s = proclist_lock_write(); 261 262 /* 263 * Find an unused process ID. We remember a range of unused IDs 264 * ready to use (from nextpid+1 through pidchecked-1). 265 */ 266 nextpid++; 267 retry: 268 /* 269 * If the process ID prototype has wrapped around, 270 * restart somewhat above 0, as the low-numbered procs 271 * tend to include daemons that don't exit. 272 */ 273 if (nextpid >= PID_MAX) { 274 nextpid = 500; 275 pidchecked = 0; 276 } 277 if (nextpid >= pidchecked) { 278 const struct proclist_desc *pd; 279 280 pidchecked = PID_MAX; 281 /* 282 * Scan the process lists to check whether this pid 283 * is in use. Remember the lowest pid that's greater 284 * than nextpid, so we can avoid checking for a while. 285 */ 286 pd = proclists; 287 again: 288 LIST_FOREACH(tp, pd->pd_list, p_list) { 289 while (tp->p_pid == nextpid || 290 tp->p_pgrp->pg_id == nextpid || 291 tp->p_session->s_sid == nextpid) { 292 nextpid++; 293 if (nextpid >= pidchecked) 294 goto retry; 295 } 296 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 297 pidchecked = tp->p_pid; 298 299 if (tp->p_pgrp->pg_id > nextpid && 300 pidchecked > tp->p_pgrp->pg_id) 301 pidchecked = tp->p_pgrp->pg_id; 302 303 if (tp->p_session->s_sid > nextpid && 304 pidchecked > tp->p_session->s_sid) 305 pidchecked = tp->p_session->s_sid; 306 } 307 308 /* 309 * If there's another list, scan it. If we have checked 310 * them all, we've found one! 311 */ 312 pd++; 313 if (pd->pd_list != NULL) 314 goto again; 315 } 316 317 /* 318 * Put the proc on allproc before unlocking PID allocation 319 * so that waiters won't grab it as soon as we unlock. 320 */ 321 322 p2->p_stat = SIDL; /* protect against others */ 323 p2->p_pid = nextpid; 324 p2->p_exitsig = exitsig; /* signal for parent on exit */ 325 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 326 327 LIST_INSERT_HEAD(&allproc, p2, p_list); 328 329 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 330 331 /* 332 * END PID ALLOCATION. 333 */ 334 proclist_unlock_write(s); 335 336 /* 337 * Make a proc table entry for the new process. 338 * Start by zeroing the section of proc that is zero-initialized, 339 * then copy the section that is copied directly from the parent. 340 */ 341 memset(&p2->p_startzero, 0, 342 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 343 memcpy(&p2->p_startcopy, &p1->p_startcopy, 344 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 345 346 #if !defined(MULTIPROCESSOR) 347 /* 348 * In the single-processor case, all processes will always run 349 * on the same CPU. So, initialize the child's CPU to the parent's 350 * now. In the multiprocessor case, the child's CPU will be 351 * initialized in the low-level context switch code when the 352 * process runs. 353 */ 354 p2->p_cpu = p1->p_cpu; 355 #else 356 /* 357 * zero child's cpu pointer so we don't get trash. 358 */ 359 p2->p_cpu = NULL; 360 #endif /* ! MULTIPROCESSOR */ 361 362 /* 363 * Duplicate sub-structures as needed. 364 * Increase reference counts on shared objects. 365 * The p_stats and p_sigacts substructs are set in uvm_fork(). 366 */ 367 p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID); 368 p2->p_emul = p1->p_emul; 369 p2->p_execsw = p1->p_execsw; 370 371 if (p1->p_flag & P_PROFIL) 372 startprofclock(p2); 373 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK); 374 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred)); 375 p2->p_cred->p_refcnt = 1; 376 crhold(p1->p_ucred); 377 378 LIST_INIT(&p2->p_raslist); 379 p2->p_nras = 0; 380 simple_lock_init(&p2->p_raslock); 381 #if defined(__HAVE_RAS) 382 ras_fork(p1, p2); 383 #endif 384 385 /* bump references to the text vnode (for procfs) */ 386 p2->p_textvp = p1->p_textvp; 387 if (p2->p_textvp) 388 VREF(p2->p_textvp); 389 390 if (flags & FORK_SHAREFILES) 391 fdshare(p1, p2); 392 else if (flags & FORK_CLEANFILES) 393 p2->p_fd = fdinit(p1); 394 else 395 p2->p_fd = fdcopy(p1); 396 397 if (flags & FORK_SHARECWD) 398 cwdshare(p1, p2); 399 else 400 p2->p_cwdi = cwdinit(p1); 401 402 /* 403 * If p_limit is still copy-on-write, bump refcnt, 404 * otherwise get a copy that won't be modified. 405 * (If PL_SHAREMOD is clear, the structure is shared 406 * copy-on-write.) 407 */ 408 if (p1->p_limit->p_lflags & PL_SHAREMOD) 409 p2->p_limit = limcopy(p1->p_limit); 410 else { 411 p2->p_limit = p1->p_limit; 412 p2->p_limit->p_refcnt++; 413 } 414 415 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 416 p2->p_flag |= P_CONTROLT; 417 if (flags & FORK_PPWAIT) 418 p2->p_flag |= P_PPWAIT; 419 LIST_INSERT_AFTER(p1, p2, p_pglist); 420 p2->p_pptr = (flags & FORK_NOWAIT) ? initproc : p1; 421 LIST_INSERT_HEAD(&p2->p_pptr->p_children, p2, p_sibling); 422 LIST_INIT(&p2->p_children); 423 424 callout_init(&p2->p_realit_ch); 425 callout_init(&p2->p_tsleep_ch); 426 427 #ifdef KTRACE 428 /* 429 * Copy traceflag and tracefile if enabled. 430 * If not inherited, these were zeroed above. 431 */ 432 if (p1->p_traceflag & KTRFAC_INHERIT) { 433 p2->p_traceflag = p1->p_traceflag; 434 if ((p2->p_tracep = p1->p_tracep) != NULL) 435 ktradref(p2); 436 } 437 #endif 438 #ifdef SYSTRACE 439 /* Tell systrace what's happening. */ 440 if (ISSET(p1->p_flag, P_SYSTRACE)) 441 systrace_sys_fork(p1, p2); 442 #endif 443 444 445 #ifdef __HAVE_SYSCALL_INTERN 446 (*p2->p_emul->e_syscall_intern)(p2); 447 #endif 448 449 scheduler_fork_hook(p1, p2); 450 451 /* 452 * Create signal actions for the child process. 453 */ 454 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 455 456 /* 457 * If emulation has process fork hook, call it now. 458 */ 459 if (p2->p_emul->e_proc_fork) 460 (*p2->p_emul->e_proc_fork)(p2, p1); 461 462 /* 463 * This begins the section where we must prevent the parent 464 * from being swapped. 465 */ 466 PHOLD(p1); 467 468 /* 469 * Finish creating the child process. It will return through a 470 * different path later. 471 */ 472 p2->p_addr = (struct user *)uaddr; 473 uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE, 474 stack, stacksize, 475 (func != NULL) ? func : child_return, 476 (arg != NULL) ? arg : p2); 477 478 /* 479 * Make child runnable, set start time, and add to run queue. 480 */ 481 SCHED_LOCK(s); 482 p2->p_stats->p_start = time; 483 p2->p_acflag = AFORK; 484 p2->p_stat = SRUN; 485 setrunqueue(p2); 486 SCHED_UNLOCK(s); 487 488 /* 489 * Now can be swapped. 490 */ 491 PRELE(p1); 492 493 /* 494 * Update stats now that we know the fork was successful. 495 */ 496 uvmexp.forks++; 497 if (flags & FORK_PPWAIT) 498 uvmexp.forks_ppwait++; 499 if (flags & FORK_SHAREVM) 500 uvmexp.forks_sharevm++; 501 502 /* 503 * Pass a pointer to the new process to the caller. 504 */ 505 if (rnewprocp != NULL) 506 *rnewprocp = p2; 507 508 #ifdef KTRACE 509 if (KTRPOINT(p2, KTR_EMUL)) 510 ktremul(p2); 511 #endif 512 513 /* 514 * Preserve synchronization semantics of vfork. If waiting for 515 * child to exec or exit, set P_PPWAIT on child, and sleep on our 516 * proc (in case of exit). 517 */ 518 if (flags & FORK_PPWAIT) 519 while (p2->p_flag & P_PPWAIT) 520 tsleep(p1, PWAIT, "ppwait", 0); 521 522 /* 523 * Return child pid to parent process, 524 * marking us as parent via retval[1]. 525 */ 526 if (retval != NULL) { 527 retval[0] = p2->p_pid; 528 retval[1] = 0; 529 } 530 531 return (0); 532 } 533 534 #if defined(MULTIPROCESSOR) 535 /* 536 * XXX This is a slight hack to get newly-formed processes to 537 * XXX acquire the kernel lock as soon as they run. 538 */ 539 void 540 proc_trampoline_mp(void) 541 { 542 struct proc *p; 543 544 p = curproc; 545 546 SCHED_ASSERT_UNLOCKED(); 547 KERNEL_PROC_LOCK(p); 548 } 549 #endif 550