1 /* $NetBSD: kern_fork.c,v 1.88 2001/12/08 00:35:30 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 1.88 2001/12/08 00:35:30 thorpej Exp $"); 82 83 #include "opt_ktrace.h" 84 #include "opt_multiprocessor.h" 85 86 #include <sys/param.h> 87 #include <sys/systm.h> 88 #include <sys/map.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/resourcevar.h> 96 #include <sys/vnode.h> 97 #include <sys/file.h> 98 #include <sys/acct.h> 99 #include <sys/ktrace.h> 100 #include <sys/vmmeter.h> 101 #include <sys/sched.h> 102 #include <sys/signalvar.h> 103 104 #include <sys/syscallargs.h> 105 106 #include <uvm/uvm_extern.h> 107 108 int nprocs = 1; /* process 0 */ 109 110 /*ARGSUSED*/ 111 int 112 sys_fork(struct proc *p, void *v, register_t *retval) 113 { 114 115 return (fork1(p, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 116 } 117 118 /* 119 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 120 * Address space is not shared, but parent is blocked until child exit. 121 */ 122 /*ARGSUSED*/ 123 int 124 sys_vfork(struct proc *p, void *v, register_t *retval) 125 { 126 127 return (fork1(p, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 128 retval, NULL)); 129 } 130 131 /* 132 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 133 * semantics. Address space is shared, and parent is blocked until child exit. 134 */ 135 /*ARGSUSED*/ 136 int 137 sys___vfork14(struct proc *p, void *v, register_t *retval) 138 { 139 140 return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 141 NULL, NULL, retval, NULL)); 142 } 143 144 /* 145 * Linux-compatible __clone(2) system call. 146 */ 147 int 148 sys___clone(struct proc *p, void *v, register_t *retval) 149 { 150 struct sys___clone_args /* { 151 syscallarg(int) flags; 152 syscallarg(void *) stack; 153 } */ *uap = v; 154 int flags, sig; 155 156 /* 157 * We don't support the CLONE_PID or CLONE_PTRACE flags. 158 */ 159 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE)) 160 return (EINVAL); 161 162 flags = 0; 163 164 if (SCARG(uap, flags) & CLONE_VM) 165 flags |= FORK_SHAREVM; 166 if (SCARG(uap, flags) & CLONE_FS) 167 flags |= FORK_SHARECWD; 168 if (SCARG(uap, flags) & CLONE_FILES) 169 flags |= FORK_SHAREFILES; 170 if (SCARG(uap, flags) & CLONE_SIGHAND) 171 flags |= FORK_SHARESIGS; 172 if (SCARG(uap, flags) & CLONE_VFORK) 173 flags |= FORK_PPWAIT; 174 175 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 176 if (sig < 0 || sig >= _NSIG) 177 return (EINVAL); 178 179 /* 180 * Note that the Linux API does not provide a portable way of 181 * specifying the stack area; the caller must know if the stack 182 * grows up or down. So, we pass a stack size of 0, so that the 183 * code that makes this adjustment is a noop. 184 */ 185 return (fork1(p, flags, sig, SCARG(uap, stack), 0, 186 NULL, NULL, retval, NULL)); 187 } 188 189 int 190 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize, 191 void (*func)(void *), void *arg, register_t *retval, 192 struct proc **rnewprocp) 193 { 194 struct proc *p2, *tp; 195 uid_t uid; 196 int count, s; 197 vaddr_t uaddr; 198 static int nextpid, pidchecked; 199 200 /* 201 * Although process entries are dynamically created, we still keep 202 * a global limit on the maximum number we will create. Don't allow 203 * a nonprivileged user to use the last process; don't let root 204 * exceed the limit. The variable nprocs is the current number of 205 * processes, maxproc is the limit. 206 */ 207 uid = p1->p_cred->p_ruid; 208 if (__predict_false((nprocs >= maxproc - 1 && uid != 0) || 209 nprocs >= maxproc)) { 210 tablefull("proc", "increase kern.maxproc or NPROC"); 211 return (EAGAIN); 212 } 213 nprocs++; 214 215 /* 216 * Increment the count of procs running with this uid. Don't allow 217 * a nonprivileged user to exceed their current limit. 218 */ 219 count = chgproccnt(uid, 1); 220 if (__predict_false(uid != 0 && count > 221 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 222 (void)chgproccnt(uid, -1); 223 nprocs--; 224 return (EAGAIN); 225 } 226 227 /* 228 * Allocate virtual address space for the U-area now, while it 229 * is still easy to abort the fork operation if we're out of 230 * kernel virtual address space. The actual U-area pages will 231 * be allocated and wired in vm_fork(). 232 */ 233 234 #ifndef USPACE_ALIGN 235 #define USPACE_ALIGN 0 236 #endif 237 238 uaddr = uvm_km_valloc_align(kernel_map, USPACE, USPACE_ALIGN); 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 = P_INMEM | (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 /* bump references to the text vnode (for procfs) */ 296 p2->p_textvp = p1->p_textvp; 297 if (p2->p_textvp) 298 VREF(p2->p_textvp); 299 300 if (flags & FORK_SHAREFILES) 301 fdshare(p1, p2); 302 else 303 p2->p_fd = fdcopy(p1); 304 305 if (flags & FORK_SHARECWD) 306 cwdshare(p1, p2); 307 else 308 p2->p_cwdi = cwdinit(p1); 309 310 /* 311 * If p_limit is still copy-on-write, bump refcnt, 312 * otherwise get a copy that won't be modified. 313 * (If PL_SHAREMOD is clear, the structure is shared 314 * copy-on-write.) 315 */ 316 if (p1->p_limit->p_lflags & PL_SHAREMOD) 317 p2->p_limit = limcopy(p1->p_limit); 318 else { 319 p2->p_limit = p1->p_limit; 320 p2->p_limit->p_refcnt++; 321 } 322 323 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 324 p2->p_flag |= P_CONTROLT; 325 if (flags & FORK_PPWAIT) 326 p2->p_flag |= P_PPWAIT; 327 LIST_INSERT_AFTER(p1, p2, p_pglist); 328 p2->p_pptr = p1; 329 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling); 330 LIST_INIT(&p2->p_children); 331 332 callout_init(&p2->p_realit_ch); 333 callout_init(&p2->p_tsleep_ch); 334 335 #ifdef KTRACE 336 /* 337 * Copy traceflag and tracefile if enabled. 338 * If not inherited, these were zeroed above. 339 */ 340 if (p1->p_traceflag & KTRFAC_INHERIT) { 341 p2->p_traceflag = p1->p_traceflag; 342 if ((p2->p_tracep = p1->p_tracep) != NULL) 343 ktradref(p2); 344 } 345 #endif 346 347 #ifdef __HAVE_SYSCALL_INTERN 348 (*p2->p_emul->e_syscall_intern)(p2); 349 #endif 350 351 scheduler_fork_hook(p1, p2); 352 353 /* 354 * Create signal actions for the child process. 355 */ 356 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 357 358 /* 359 * If emulation has process fork hook, call it now. 360 */ 361 if (p2->p_emul->e_proc_fork) 362 (*p2->p_emul->e_proc_fork)(p2, p1); 363 364 /* 365 * This begins the section where we must prevent the parent 366 * from being swapped. 367 */ 368 PHOLD(p1); 369 370 /* 371 * Finish creating the child process. It will return through a 372 * different path later. 373 */ 374 p2->p_addr = (struct user *)uaddr; 375 uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE, 376 stack, stacksize, 377 (func != NULL) ? func : child_return, 378 (arg != NULL) ? arg : p2); 379 380 /* 381 * BEGIN PID ALLOCATION. 382 */ 383 s = proclist_lock_write(); 384 385 /* 386 * Find an unused process ID. We remember a range of unused IDs 387 * ready to use (from nextpid+1 through pidchecked-1). 388 */ 389 nextpid++; 390 retry: 391 /* 392 * If the process ID prototype has wrapped around, 393 * restart somewhat above 0, as the low-numbered procs 394 * tend to include daemons that don't exit. 395 */ 396 if (nextpid >= PID_MAX) { 397 nextpid = 500; 398 pidchecked = 0; 399 } 400 if (nextpid >= pidchecked) { 401 const struct proclist_desc *pd; 402 403 pidchecked = PID_MAX; 404 /* 405 * Scan the process lists to check whether this pid 406 * is in use. Remember the lowest pid that's greater 407 * than nextpid, so we can avoid checking for a while. 408 */ 409 pd = proclists; 410 again: 411 LIST_FOREACH(tp, pd->pd_list, p_list) { 412 while (tp->p_pid == nextpid || 413 tp->p_pgrp->pg_id == nextpid || 414 tp->p_session->s_sid == nextpid) { 415 nextpid++; 416 if (nextpid >= pidchecked) 417 goto retry; 418 } 419 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 420 pidchecked = tp->p_pid; 421 422 if (tp->p_pgrp->pg_id > nextpid && 423 pidchecked > tp->p_pgrp->pg_id) 424 pidchecked = tp->p_pgrp->pg_id; 425 426 if (tp->p_session->s_sid > nextpid && 427 pidchecked > tp->p_session->s_sid) 428 pidchecked = tp->p_session->s_sid; 429 } 430 431 /* 432 * If there's another list, scan it. If we have checked 433 * them all, we've found one! 434 */ 435 pd++; 436 if (pd->pd_list != NULL) 437 goto again; 438 } 439 440 /* Record the pid we've allocated. */ 441 p2->p_pid = nextpid; 442 443 /* Record the signal to be delivered to the parent on exit. */ 444 p2->p_exitsig = exitsig; 445 446 /* 447 * Put the proc on allproc before unlocking PID allocation 448 * so that waiters won't grab it as soon as we unlock. 449 */ 450 451 p2->p_stat = SIDL; /* protect against others */ 452 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 453 454 LIST_INSERT_HEAD(&allproc, p2, p_list); 455 456 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 457 458 /* 459 * END PID ALLOCATION. 460 */ 461 proclist_unlock_write(s); 462 463 /* 464 * Make child runnable, set start time, and add to run queue. 465 */ 466 SCHED_LOCK(s); 467 p2->p_stats->p_start = time; 468 p2->p_acflag = AFORK; 469 p2->p_stat = SRUN; 470 setrunqueue(p2); 471 SCHED_UNLOCK(s); 472 473 /* 474 * Now can be swapped. 475 */ 476 PRELE(p1); 477 478 /* 479 * Update stats now that we know the fork was successful. 480 */ 481 uvmexp.forks++; 482 if (flags & FORK_PPWAIT) 483 uvmexp.forks_ppwait++; 484 if (flags & FORK_SHAREVM) 485 uvmexp.forks_sharevm++; 486 487 /* 488 * Pass a pointer to the new process to the caller. 489 */ 490 if (rnewprocp != NULL) 491 *rnewprocp = p2; 492 493 #ifdef KTRACE 494 if (KTRPOINT(p2, KTR_EMUL)) 495 ktremul(p2); 496 #endif 497 498 /* 499 * Preserve synchronization semantics of vfork. If waiting for 500 * child to exec or exit, set P_PPWAIT on child, and sleep on our 501 * proc (in case of exit). 502 */ 503 if (flags & FORK_PPWAIT) 504 while (p2->p_flag & P_PPWAIT) 505 tsleep(p1, PWAIT, "ppwait", 0); 506 507 /* 508 * Return child pid to parent process, 509 * marking us as parent via retval[1]. 510 */ 511 if (retval != NULL) { 512 retval[0] = p2->p_pid; 513 retval[1] = 0; 514 } 515 516 return (0); 517 } 518 519 #if defined(MULTIPROCESSOR) 520 /* 521 * XXX This is a slight hack to get newly-formed processes to 522 * XXX acquire the kernel lock as soon as they run. 523 */ 524 void 525 proc_trampoline_mp(void) 526 { 527 struct proc *p; 528 529 p = curproc; 530 531 SCHED_ASSERT_UNLOCKED(); 532 KERNEL_PROC_LOCK(p); 533 } 534 #endif 535