1 /* $NetBSD: kern_fork.c,v 1.187 2012/02/02 02:44:06 christos Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2001, 2004, 2006, 2007, 2008 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, by Charles M. Hannum, and by Andrew Doran. 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 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1989, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * (c) UNIX System Laboratories, Inc. 37 * All or some portions of this file are derived from material licensed 38 * to the University of California by American Telephone and Telegraph 39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 40 * the permission of UNIX System Laboratories, Inc. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)kern_fork.c 8.8 (Berkeley) 2/14/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: kern_fork.c,v 1.187 2012/02/02 02:44:06 christos Exp $"); 71 72 #include "opt_ktrace.h" 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/filedesc.h> 77 #include <sys/kernel.h> 78 #include <sys/pool.h> 79 #include <sys/mount.h> 80 #include <sys/proc.h> 81 #include <sys/ras.h> 82 #include <sys/resourcevar.h> 83 #include <sys/vnode.h> 84 #include <sys/file.h> 85 #include <sys/acct.h> 86 #include <sys/ktrace.h> 87 #include <sys/vmmeter.h> 88 #include <sys/sched.h> 89 #include <sys/signalvar.h> 90 #include <sys/kauth.h> 91 #include <sys/atomic.h> 92 #include <sys/syscallargs.h> 93 #include <sys/uidinfo.h> 94 #include <sys/sdt.h> 95 #include <sys/ptrace.h> 96 97 #include <uvm/uvm_extern.h> 98 99 /* 100 * DTrace SDT provider definitions 101 */ 102 SDT_PROBE_DEFINE(proc,,,create, 103 "struct proc *", NULL, /* new process */ 104 "struct proc *", NULL, /* parent process */ 105 "int", NULL, /* flags */ 106 NULL, NULL, NULL, NULL); 107 108 u_int nprocs = 1; /* process 0 */ 109 110 /* 111 * Number of ticks to sleep if fork() would fail due to process hitting 112 * limits. Exported in miliseconds to userland via sysctl. 113 */ 114 int forkfsleep = 0; 115 116 /*ARGSUSED*/ 117 int 118 sys_fork(struct lwp *l, const void *v, register_t *retval) 119 { 120 121 return (fork1(l, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 122 } 123 124 /* 125 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 126 * Address space is not shared, but parent is blocked until child exit. 127 */ 128 /*ARGSUSED*/ 129 int 130 sys_vfork(struct lwp *l, const void *v, register_t *retval) 131 { 132 133 return (fork1(l, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 134 retval, NULL)); 135 } 136 137 /* 138 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 139 * semantics. Address space is shared, and parent is blocked until child exit. 140 */ 141 /*ARGSUSED*/ 142 int 143 sys___vfork14(struct lwp *l, const void *v, register_t *retval) 144 { 145 146 return (fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 147 NULL, NULL, retval, NULL)); 148 } 149 150 /* 151 * Linux-compatible __clone(2) system call. 152 */ 153 int 154 sys___clone(struct lwp *l, const struct sys___clone_args *uap, register_t *retval) 155 { 156 /* { 157 syscallarg(int) flags; 158 syscallarg(void *) stack; 159 } */ 160 int flags, sig; 161 162 /* 163 * We don't support the CLONE_PID or CLONE_PTRACE flags. 164 */ 165 if (SCARG(uap, flags) & (CLONE_PID|CLONE_PTRACE)) 166 return (EINVAL); 167 168 /* 169 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same. 170 */ 171 if (SCARG(uap, flags) & CLONE_SIGHAND 172 && (SCARG(uap, flags) & CLONE_VM) == 0) 173 return (EINVAL); 174 175 flags = 0; 176 177 if (SCARG(uap, flags) & CLONE_VM) 178 flags |= FORK_SHAREVM; 179 if (SCARG(uap, flags) & CLONE_FS) 180 flags |= FORK_SHARECWD; 181 if (SCARG(uap, flags) & CLONE_FILES) 182 flags |= FORK_SHAREFILES; 183 if (SCARG(uap, flags) & CLONE_SIGHAND) 184 flags |= FORK_SHARESIGS; 185 if (SCARG(uap, flags) & CLONE_VFORK) 186 flags |= FORK_PPWAIT; 187 188 sig = SCARG(uap, flags) & CLONE_CSIGNAL; 189 if (sig < 0 || sig >= _NSIG) 190 return (EINVAL); 191 192 /* 193 * Note that the Linux API does not provide a portable way of 194 * specifying the stack area; the caller must know if the stack 195 * grows up or down. So, we pass a stack size of 0, so that the 196 * code that makes this adjustment is a noop. 197 */ 198 return (fork1(l, flags, sig, SCARG(uap, stack), 0, 199 NULL, NULL, retval, NULL)); 200 } 201 202 /* print the 'table full' message once per 10 seconds */ 203 struct timeval fork_tfmrate = { 10, 0 }; 204 205 /* 206 * General fork call. Note that another LWP in the process may call exec() 207 * or exit() while we are forking. It's safe to continue here, because 208 * neither operation will complete until all LWPs have exited the process. 209 */ 210 int 211 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize, 212 void (*func)(void *), void *arg, register_t *retval, 213 struct proc **rnewprocp) 214 { 215 struct proc *p1, *p2, *parent; 216 struct plimit *p1_lim; 217 uid_t uid; 218 struct lwp *l2; 219 int count; 220 vaddr_t uaddr; 221 int tnprocs; 222 int tracefork; 223 int error = 0; 224 225 p1 = l1->l_proc; 226 uid = kauth_cred_getuid(l1->l_cred); 227 tnprocs = atomic_inc_uint_nv(&nprocs); 228 229 /* 230 * Although process entries are dynamically created, we still keep 231 * a global limit on the maximum number we will create. 232 */ 233 if (__predict_false(tnprocs >= maxproc)) 234 error = -1; 235 else 236 error = kauth_authorize_process(l1->l_cred, 237 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL); 238 239 if (error) { 240 static struct timeval lasttfm; 241 atomic_dec_uint(&nprocs); 242 if (ratecheck(&lasttfm, &fork_tfmrate)) 243 tablefull("proc", "increase kern.maxproc or NPROC"); 244 if (forkfsleep) 245 kpause("forkmx", false, forkfsleep, NULL); 246 return (EAGAIN); 247 } 248 249 /* 250 * Enforce limits. 251 */ 252 count = chgproccnt(uid, 1); 253 if (kauth_authorize_generic(l1->l_cred, KAUTH_GENERIC_ISSUSER, NULL) != 254 0 && __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 255 (void)chgproccnt(uid, -1); 256 atomic_dec_uint(&nprocs); 257 if (forkfsleep) 258 kpause("forkulim", false, forkfsleep, NULL); 259 return (EAGAIN); 260 } 261 262 /* 263 * Allocate virtual address space for the U-area now, while it 264 * is still easy to abort the fork operation if we're out of 265 * kernel virtual address space. 266 */ 267 uaddr = uvm_uarea_alloc(); 268 if (__predict_false(uaddr == 0)) { 269 (void)chgproccnt(uid, -1); 270 atomic_dec_uint(&nprocs); 271 return (ENOMEM); 272 } 273 274 /* 275 * We are now committed to the fork. From here on, we may 276 * block on resources, but resource allocation may NOT fail. 277 */ 278 279 /* Allocate new proc. */ 280 p2 = proc_alloc(); 281 282 /* 283 * Make a proc table entry for the new process. 284 * Start by zeroing the section of proc that is zero-initialized, 285 * then copy the section that is copied directly from the parent. 286 */ 287 memset(&p2->p_startzero, 0, 288 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero)); 289 memcpy(&p2->p_startcopy, &p1->p_startcopy, 290 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy)); 291 292 CIRCLEQ_INIT(&p2->p_sigpend.sp_info); 293 294 LIST_INIT(&p2->p_lwps); 295 LIST_INIT(&p2->p_sigwaiters); 296 297 /* 298 * Duplicate sub-structures as needed. 299 * Increase reference counts on shared objects. 300 * Inherit flags we want to keep. The flags related to SIGCHLD 301 * handling are important in order to keep a consistent behaviour 302 * for the child after the fork. If we are a 32-bit process, the 303 * child will be too. 304 */ 305 p2->p_flag = 306 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32); 307 p2->p_emul = p1->p_emul; 308 p2->p_execsw = p1->p_execsw; 309 310 if (flags & FORK_SYSTEM) { 311 /* 312 * Mark it as a system process. Set P_NOCLDWAIT so that 313 * children are reparented to init(8) when they exit. 314 * init(8) can easily wait them out for us. 315 */ 316 p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT); 317 } 318 319 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH); 320 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE); 321 rw_init(&p2->p_reflock); 322 cv_init(&p2->p_waitcv, "wait"); 323 cv_init(&p2->p_lwpcv, "lwpwait"); 324 325 /* 326 * Share a lock between the processes if they are to share signal 327 * state: we must synchronize access to it. 328 */ 329 if (flags & FORK_SHARESIGS) { 330 p2->p_lock = p1->p_lock; 331 mutex_obj_hold(p1->p_lock); 332 } else 333 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 334 335 kauth_proc_fork(p1, p2); 336 337 p2->p_raslist = NULL; 338 #if defined(__HAVE_RAS) 339 ras_fork(p1, p2); 340 #endif 341 342 /* bump references to the text vnode (for procfs) */ 343 p2->p_textvp = p1->p_textvp; 344 if (p2->p_textvp) 345 vref(p2->p_textvp); 346 347 if (flags & FORK_SHAREFILES) 348 fd_share(p2); 349 else if (flags & FORK_CLEANFILES) 350 p2->p_fd = fd_init(NULL); 351 else 352 p2->p_fd = fd_copy(); 353 354 /* XXX racy */ 355 p2->p_mqueue_cnt = p1->p_mqueue_cnt; 356 357 if (flags & FORK_SHARECWD) 358 cwdshare(p2); 359 else 360 p2->p_cwdi = cwdinit(); 361 362 /* 363 * Note: p_limit (rlimit stuff) is copy-on-write, so normally 364 * we just need increase pl_refcnt. 365 */ 366 p1_lim = p1->p_limit; 367 if (!p1_lim->pl_writeable) { 368 lim_addref(p1_lim); 369 p2->p_limit = p1_lim; 370 } else { 371 p2->p_limit = lim_copy(p1_lim); 372 } 373 374 p2->p_lflag = ((flags & FORK_PPWAIT) ? PL_PPWAIT : 0); 375 p2->p_sflag = 0; 376 p2->p_slflag = 0; 377 parent = (flags & FORK_NOWAIT) ? initproc : p1; 378 p2->p_pptr = parent; 379 p2->p_ppid = parent->p_pid; 380 LIST_INIT(&p2->p_children); 381 382 p2->p_aio = NULL; 383 384 #ifdef KTRACE 385 /* 386 * Copy traceflag and tracefile if enabled. 387 * If not inherited, these were zeroed above. 388 */ 389 if (p1->p_traceflag & KTRFAC_INHERIT) { 390 mutex_enter(&ktrace_lock); 391 p2->p_traceflag = p1->p_traceflag; 392 if ((p2->p_tracep = p1->p_tracep) != NULL) 393 ktradref(p2); 394 mutex_exit(&ktrace_lock); 395 } 396 #endif 397 398 /* 399 * Create signal actions for the child process. 400 */ 401 p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS); 402 mutex_enter(p1->p_lock); 403 p2->p_sflag |= 404 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 405 sched_proc_fork(p1, p2); 406 mutex_exit(p1->p_lock); 407 408 p2->p_stflag = p1->p_stflag; 409 410 /* 411 * p_stats. 412 * Copy parts of p_stats, and zero out the rest. 413 */ 414 p2->p_stats = pstatscopy(p1->p_stats); 415 416 /* 417 * Set up the new process address space. 418 */ 419 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false); 420 421 /* 422 * Finish creating the child process. 423 * It will return through a different path later. 424 */ 425 lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0, 426 stack, stacksize, (func != NULL) ? func : child_return, arg, &l2, 427 l1->l_class); 428 429 /* 430 * Inherit l_private from the parent. 431 * Note that we cannot use lwp_setprivate() here since that 432 * also sets the CPU TLS register, which is incorrect if the 433 * process has changed that without letting the kernel know. 434 */ 435 l2->l_private = l1->l_private; 436 437 /* 438 * If emulation has a process fork hook, call it now. 439 */ 440 if (p2->p_emul->e_proc_fork) 441 (*p2->p_emul->e_proc_fork)(p2, l1, flags); 442 443 /* 444 * ...and finally, any other random fork hooks that subsystems 445 * might have registered. 446 */ 447 doforkhooks(p2, p1); 448 449 SDT_PROBE(proc,,,create, p2, p1, flags, 0, 0); 450 451 /* 452 * It's now safe for the scheduler and other processes to see the 453 * child process. 454 */ 455 mutex_enter(proc_lock); 456 457 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 458 p2->p_lflag |= PL_CONTROLT; 459 460 LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling); 461 p2->p_exitsig = exitsig; /* signal for parent on exit */ 462 463 /* 464 * We don't want to tracefork vfork()ed processes because they 465 * will not receive the SIGTRAP until it is too late. 466 */ 467 tracefork = (p1->p_slflag & (PSL_TRACEFORK|PSL_TRACED)) == 468 (PSL_TRACEFORK|PSL_TRACED) && (flags && FORK_PPWAIT) == 0; 469 if (tracefork) { 470 p2->p_slflag |= PSL_TRACED; 471 p2->p_opptr = p2->p_pptr; 472 if (p2->p_pptr != p1->p_pptr) { 473 struct proc *parent1 = p2->p_pptr; 474 475 if (parent1->p_lock < p2->p_lock) { 476 if (!mutex_tryenter(parent1->p_lock)) { 477 mutex_exit(p2->p_lock); 478 mutex_enter(parent1->p_lock); 479 } 480 } else if (parent1->p_lock > p2->p_lock) { 481 mutex_enter(parent1->p_lock); 482 } 483 parent1->p_slflag |= PSL_CHTRACED; 484 proc_reparent(p2, p1->p_pptr); 485 if (parent1->p_lock != p2->p_lock) 486 mutex_exit(parent1->p_lock); 487 } 488 489 /* 490 * Set ptrace status. 491 */ 492 p1->p_fpid = p2->p_pid; 493 p2->p_fpid = p1->p_pid; 494 } 495 496 LIST_INSERT_AFTER(p1, p2, p_pglist); 497 LIST_INSERT_HEAD(&allproc, p2, p_list); 498 499 p2->p_trace_enabled = trace_is_enabled(p2); 500 #ifdef __HAVE_SYSCALL_INTERN 501 (*p2->p_emul->e_syscall_intern)(p2); 502 #endif 503 504 /* 505 * Update stats now that we know the fork was successful. 506 */ 507 uvmexp.forks++; 508 if (flags & FORK_PPWAIT) 509 uvmexp.forks_ppwait++; 510 if (flags & FORK_SHAREVM) 511 uvmexp.forks_sharevm++; 512 513 /* 514 * Pass a pointer to the new process to the caller. 515 */ 516 if (rnewprocp != NULL) 517 *rnewprocp = p2; 518 519 if (ktrpoint(KTR_EMUL)) 520 p2->p_traceflag |= KTRFAC_TRC_EMUL; 521 522 /* 523 * Notify any interested parties about the new process. 524 */ 525 if (!SLIST_EMPTY(&p1->p_klist)) { 526 mutex_exit(proc_lock); 527 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 528 mutex_enter(proc_lock); 529 } 530 531 /* 532 * Make child runnable, set start time, and add to run queue except 533 * if the parent requested the child to start in SSTOP state. 534 */ 535 mutex_enter(p2->p_lock); 536 537 /* 538 * Start profiling. 539 */ 540 if ((p2->p_stflag & PST_PROFIL) != 0) { 541 mutex_spin_enter(&p2->p_stmutex); 542 startprofclock(p2); 543 mutex_spin_exit(&p2->p_stmutex); 544 } 545 546 getmicrotime(&p2->p_stats->p_start); 547 p2->p_acflag = AFORK; 548 lwp_lock(l2); 549 KASSERT(p2->p_nrlwps == 1); 550 if (p2->p_sflag & PS_STOPFORK) { 551 struct schedstate_percpu *spc = &l2->l_cpu->ci_schedstate; 552 p2->p_nrlwps = 0; 553 p2->p_stat = SSTOP; 554 p2->p_waited = 0; 555 p1->p_nstopchild++; 556 l2->l_stat = LSSTOP; 557 KASSERT(l2->l_wchan == NULL); 558 lwp_unlock_to(l2, spc->spc_lwplock); 559 } else { 560 p2->p_nrlwps = 1; 561 p2->p_stat = SACTIVE; 562 l2->l_stat = LSRUN; 563 sched_enqueue(l2, false); 564 lwp_unlock(l2); 565 } 566 mutex_exit(p2->p_lock); 567 568 /* 569 * Preserve synchronization semantics of vfork. If waiting for 570 * child to exec or exit, set PL_PPWAIT on child, and sleep on our 571 * proc (in case of exit). 572 */ 573 while (p2->p_lflag & PL_PPWAIT) 574 cv_wait(&p1->p_waitcv, proc_lock); 575 576 /* 577 * Let the parent know that we are tracing its child. 578 */ 579 if (tracefork) { 580 ksiginfo_t ksi; 581 KSI_INIT_EMPTY(&ksi); 582 ksi.ksi_signo = SIGTRAP; 583 ksi.ksi_lid = l1->l_lid; 584 kpsignal(p1, &ksi, NULL); 585 } 586 587 mutex_exit(proc_lock); 588 589 /* 590 * Return child pid to parent process, 591 * marking us as parent via retval[1]. 592 */ 593 if (retval != NULL) { 594 retval[0] = p2->p_pid; 595 retval[1] = 0; 596 } 597 598 return (0); 599 } 600