1 /* $NetBSD: kern_fork.c,v 1.189 2012/03/13 18:40:52 elad 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 1.189 2012/03/13 18:40:52 elad 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/sched.h> 88 #include <sys/signalvar.h> 89 #include <sys/kauth.h> 90 #include <sys/atomic.h> 91 #include <sys/syscallargs.h> 92 #include <sys/uidinfo.h> 93 #include <sys/sdt.h> 94 #include <sys/ptrace.h> 95 96 #include <uvm/uvm_extern.h> 97 98 /* 99 * DTrace SDT provider definitions 100 */ 101 SDT_PROBE_DEFINE(proc,,,create, 102 "struct proc *", NULL, /* new process */ 103 "struct proc *", NULL, /* parent process */ 104 "int", NULL, /* flags */ 105 NULL, NULL, NULL, NULL); 106 107 u_int nprocs __cacheline_aligned = 1; /* process 0 */ 108 109 /* 110 * Number of ticks to sleep if fork() would fail due to process hitting 111 * limits. Exported in miliseconds to userland via sysctl. 112 */ 113 int forkfsleep = 0; 114 115 int 116 sys_fork(struct lwp *l, const void *v, register_t *retval) 117 { 118 119 return fork1(l, 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 int 127 sys_vfork(struct lwp *l, const void *v, register_t *retval) 128 { 129 130 return fork1(l, 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 int 139 sys___vfork14(struct lwp *l, const void *v, register_t *retval) 140 { 141 142 return fork1(l, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 143 NULL, NULL, retval, NULL); 144 } 145 146 /* 147 * Linux-compatible __clone(2) system call. 148 */ 149 int 150 sys___clone(struct lwp *l, const struct sys___clone_args *uap, 151 register_t *retval) 152 { 153 /* { 154 syscallarg(int) flags; 155 syscallarg(void *) stack; 156 } */ 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 /* 166 * Linux enforces CLONE_VM with CLONE_SIGHAND, do same. 167 */ 168 if (SCARG(uap, flags) & CLONE_SIGHAND 169 && (SCARG(uap, flags) & CLONE_VM) == 0) 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 /* 200 * Print the 'table full' message once per 10 seconds. 201 */ 202 static struct timeval fork_tfmrate = { 10, 0 }; 203 204 /* 205 * General fork call. Note that another LWP in the process may call exec() 206 * or exit() while we are forking. It's safe to continue here, because 207 * neither operation will complete until all LWPs have exited the process. 208 */ 209 int 210 fork1(struct lwp *l1, int flags, int exitsig, void *stack, size_t stacksize, 211 void (*func)(void *), void *arg, register_t *retval, 212 struct proc **rnewprocp) 213 { 214 struct proc *p1, *p2, *parent; 215 struct plimit *p1_lim; 216 uid_t uid; 217 struct lwp *l2; 218 int count; 219 vaddr_t uaddr; 220 int tnprocs; 221 int tracefork; 222 int error = 0; 223 224 p1 = l1->l_proc; 225 uid = kauth_cred_getuid(l1->l_cred); 226 tnprocs = atomic_inc_uint_nv(&nprocs); 227 228 /* 229 * Although process entries are dynamically created, we still keep 230 * a global limit on the maximum number we will create. 231 */ 232 if (__predict_false(tnprocs >= maxproc)) 233 error = -1; 234 else 235 error = kauth_authorize_process(l1->l_cred, 236 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL); 237 238 if (error) { 239 static struct timeval lasttfm; 240 atomic_dec_uint(&nprocs); 241 if (ratecheck(&lasttfm, &fork_tfmrate)) 242 tablefull("proc", "increase kern.maxproc or NPROC"); 243 if (forkfsleep) 244 kpause("forkmx", false, forkfsleep, NULL); 245 return EAGAIN; 246 } 247 248 /* 249 * Enforce limits. 250 */ 251 count = chgproccnt(uid, 1); 252 if (__predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 253 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT, 254 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 255 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0) { 256 (void)chgproccnt(uid, -1); 257 atomic_dec_uint(&nprocs); 258 if (forkfsleep) 259 kpause("forkulim", false, forkfsleep, NULL); 260 return EAGAIN; 261 } 262 } 263 264 /* 265 * Allocate virtual address space for the U-area now, while it 266 * is still easy to abort the fork operation if we're out of 267 * kernel virtual address space. 268 */ 269 uaddr = uvm_uarea_alloc(); 270 if (__predict_false(uaddr == 0)) { 271 (void)chgproccnt(uid, -1); 272 atomic_dec_uint(&nprocs); 273 return ENOMEM; 274 } 275 276 /* 277 * We are now committed to the fork. From here on, we may 278 * block on resources, but resource allocation may NOT fail. 279 */ 280 281 /* Allocate new proc. */ 282 p2 = proc_alloc(); 283 284 /* 285 * Make a proc table entry for the new process. 286 * Start by zeroing the section of proc that is zero-initialized, 287 * then copy the section that is copied directly from the parent. 288 */ 289 memset(&p2->p_startzero, 0, 290 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero)); 291 memcpy(&p2->p_startcopy, &p1->p_startcopy, 292 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy)); 293 294 CIRCLEQ_INIT(&p2->p_sigpend.sp_info); 295 296 LIST_INIT(&p2->p_lwps); 297 LIST_INIT(&p2->p_sigwaiters); 298 299 /* 300 * Duplicate sub-structures as needed. 301 * Increase reference counts on shared objects. 302 * Inherit flags we want to keep. The flags related to SIGCHLD 303 * handling are important in order to keep a consistent behaviour 304 * for the child after the fork. If we are a 32-bit process, the 305 * child will be too. 306 */ 307 p2->p_flag = 308 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32); 309 p2->p_emul = p1->p_emul; 310 p2->p_execsw = p1->p_execsw; 311 312 if (flags & FORK_SYSTEM) { 313 /* 314 * Mark it as a system process. Set P_NOCLDWAIT so that 315 * children are reparented to init(8) when they exit. 316 * init(8) can easily wait them out for us. 317 */ 318 p2->p_flag |= (PK_SYSTEM | PK_NOCLDWAIT); 319 } 320 321 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH); 322 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE); 323 rw_init(&p2->p_reflock); 324 cv_init(&p2->p_waitcv, "wait"); 325 cv_init(&p2->p_lwpcv, "lwpwait"); 326 327 /* 328 * Share a lock between the processes if they are to share signal 329 * state: we must synchronize access to it. 330 */ 331 if (flags & FORK_SHARESIGS) { 332 p2->p_lock = p1->p_lock; 333 mutex_obj_hold(p1->p_lock); 334 } else 335 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 336 337 kauth_proc_fork(p1, p2); 338 339 p2->p_raslist = NULL; 340 #if defined(__HAVE_RAS) 341 ras_fork(p1, p2); 342 #endif 343 344 /* bump references to the text vnode (for procfs) */ 345 p2->p_textvp = p1->p_textvp; 346 if (p2->p_textvp) 347 vref(p2->p_textvp); 348 349 if (flags & FORK_SHAREFILES) 350 fd_share(p2); 351 else if (flags & FORK_CLEANFILES) 352 p2->p_fd = fd_init(NULL); 353 else 354 p2->p_fd = fd_copy(); 355 356 /* XXX racy */ 357 p2->p_mqueue_cnt = p1->p_mqueue_cnt; 358 359 if (flags & FORK_SHARECWD) 360 cwdshare(p2); 361 else 362 p2->p_cwdi = cwdinit(); 363 364 /* 365 * Note: p_limit (rlimit stuff) is copy-on-write, so normally 366 * we just need increase pl_refcnt. 367 */ 368 p1_lim = p1->p_limit; 369 if (!p1_lim->pl_writeable) { 370 lim_addref(p1_lim); 371 p2->p_limit = p1_lim; 372 } else { 373 p2->p_limit = lim_copy(p1_lim); 374 } 375 376 p2->p_lflag = ((flags & FORK_PPWAIT) ? PL_PPWAIT : 0); 377 p2->p_sflag = 0; 378 p2->p_slflag = 0; 379 parent = (flags & FORK_NOWAIT) ? initproc : p1; 380 p2->p_pptr = parent; 381 p2->p_ppid = parent->p_pid; 382 LIST_INIT(&p2->p_children); 383 384 p2->p_aio = NULL; 385 386 #ifdef KTRACE 387 /* 388 * Copy traceflag and tracefile if enabled. 389 * If not inherited, these were zeroed above. 390 */ 391 if (p1->p_traceflag & KTRFAC_INHERIT) { 392 mutex_enter(&ktrace_lock); 393 p2->p_traceflag = p1->p_traceflag; 394 if ((p2->p_tracep = p1->p_tracep) != NULL) 395 ktradref(p2); 396 mutex_exit(&ktrace_lock); 397 } 398 #endif 399 400 /* 401 * Create signal actions for the child process. 402 */ 403 p2->p_sigacts = sigactsinit(p1, flags & FORK_SHARESIGS); 404 mutex_enter(p1->p_lock); 405 p2->p_sflag |= 406 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 407 sched_proc_fork(p1, p2); 408 mutex_exit(p1->p_lock); 409 410 p2->p_stflag = p1->p_stflag; 411 412 /* 413 * p_stats. 414 * Copy parts of p_stats, and zero out the rest. 415 */ 416 p2->p_stats = pstatscopy(p1->p_stats); 417 418 /* 419 * Set up the new process address space. 420 */ 421 uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? true : false); 422 423 /* 424 * Finish creating the child process. 425 * It will return through a different path later. 426 */ 427 lwp_create(l1, p2, uaddr, (flags & FORK_PPWAIT) ? LWP_VFORK : 0, 428 stack, stacksize, (func != NULL) ? func : child_return, arg, &l2, 429 l1->l_class); 430 431 /* 432 * Inherit l_private from the parent. 433 * Note that we cannot use lwp_setprivate() here since that 434 * also sets the CPU TLS register, which is incorrect if the 435 * process has changed that without letting the kernel know. 436 */ 437 l2->l_private = l1->l_private; 438 439 /* 440 * If emulation has a process fork hook, call it now. 441 */ 442 if (p2->p_emul->e_proc_fork) 443 (*p2->p_emul->e_proc_fork)(p2, l1, flags); 444 445 /* 446 * ...and finally, any other random fork hooks that subsystems 447 * might have registered. 448 */ 449 doforkhooks(p2, p1); 450 451 SDT_PROBE(proc,,,create, p2, p1, flags, 0, 0); 452 453 /* 454 * It's now safe for the scheduler and other processes to see the 455 * child process. 456 */ 457 mutex_enter(proc_lock); 458 459 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 460 p2->p_lflag |= PL_CONTROLT; 461 462 LIST_INSERT_HEAD(&parent->p_children, p2, p_sibling); 463 p2->p_exitsig = exitsig; /* signal for parent on exit */ 464 465 /* 466 * We don't want to tracefork vfork()ed processes because they 467 * will not receive the SIGTRAP until it is too late. 468 */ 469 tracefork = (p1->p_slflag & (PSL_TRACEFORK|PSL_TRACED)) == 470 (PSL_TRACEFORK|PSL_TRACED) && (flags && FORK_PPWAIT) == 0; 471 if (tracefork) { 472 p2->p_slflag |= PSL_TRACED; 473 p2->p_opptr = p2->p_pptr; 474 if (p2->p_pptr != p1->p_pptr) { 475 struct proc *parent1 = p2->p_pptr; 476 477 if (parent1->p_lock < p2->p_lock) { 478 if (!mutex_tryenter(parent1->p_lock)) { 479 mutex_exit(p2->p_lock); 480 mutex_enter(parent1->p_lock); 481 } 482 } else if (parent1->p_lock > p2->p_lock) { 483 mutex_enter(parent1->p_lock); 484 } 485 parent1->p_slflag |= PSL_CHTRACED; 486 proc_reparent(p2, p1->p_pptr); 487 if (parent1->p_lock != p2->p_lock) 488 mutex_exit(parent1->p_lock); 489 } 490 491 /* 492 * Set ptrace status. 493 */ 494 p1->p_fpid = p2->p_pid; 495 p2->p_fpid = p1->p_pid; 496 } 497 498 LIST_INSERT_AFTER(p1, p2, p_pglist); 499 LIST_INSERT_HEAD(&allproc, p2, p_list); 500 501 p2->p_trace_enabled = trace_is_enabled(p2); 502 #ifdef __HAVE_SYSCALL_INTERN 503 (*p2->p_emul->e_syscall_intern)(p2); 504 #endif 505 506 /* 507 * Update stats now that we know the fork was successful. 508 */ 509 uvmexp.forks++; 510 if (flags & FORK_PPWAIT) 511 uvmexp.forks_ppwait++; 512 if (flags & FORK_SHAREVM) 513 uvmexp.forks_sharevm++; 514 515 /* 516 * Pass a pointer to the new process to the caller. 517 */ 518 if (rnewprocp != NULL) 519 *rnewprocp = p2; 520 521 if (ktrpoint(KTR_EMUL)) 522 p2->p_traceflag |= KTRFAC_TRC_EMUL; 523 524 /* 525 * Notify any interested parties about the new process. 526 */ 527 if (!SLIST_EMPTY(&p1->p_klist)) { 528 mutex_exit(proc_lock); 529 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 530 mutex_enter(proc_lock); 531 } 532 533 /* 534 * Make child runnable, set start time, and add to run queue except 535 * if the parent requested the child to start in SSTOP state. 536 */ 537 mutex_enter(p2->p_lock); 538 539 /* 540 * Start profiling. 541 */ 542 if ((p2->p_stflag & PST_PROFIL) != 0) { 543 mutex_spin_enter(&p2->p_stmutex); 544 startprofclock(p2); 545 mutex_spin_exit(&p2->p_stmutex); 546 } 547 548 getmicrotime(&p2->p_stats->p_start); 549 p2->p_acflag = AFORK; 550 lwp_lock(l2); 551 KASSERT(p2->p_nrlwps == 1); 552 if (p2->p_sflag & PS_STOPFORK) { 553 struct schedstate_percpu *spc = &l2->l_cpu->ci_schedstate; 554 p2->p_nrlwps = 0; 555 p2->p_stat = SSTOP; 556 p2->p_waited = 0; 557 p1->p_nstopchild++; 558 l2->l_stat = LSSTOP; 559 KASSERT(l2->l_wchan == NULL); 560 lwp_unlock_to(l2, spc->spc_lwplock); 561 } else { 562 p2->p_nrlwps = 1; 563 p2->p_stat = SACTIVE; 564 l2->l_stat = LSRUN; 565 sched_enqueue(l2, false); 566 lwp_unlock(l2); 567 } 568 mutex_exit(p2->p_lock); 569 570 /* 571 * Preserve synchronization semantics of vfork. If waiting for 572 * child to exec or exit, set PL_PPWAIT on child, and sleep on our 573 * proc (in case of exit). 574 */ 575 while (p2->p_lflag & PL_PPWAIT) 576 cv_wait(&p1->p_waitcv, proc_lock); 577 578 /* 579 * Let the parent know that we are tracing its child. 580 */ 581 if (tracefork) { 582 ksiginfo_t ksi; 583 584 KSI_INIT_EMPTY(&ksi); 585 ksi.ksi_signo = SIGTRAP; 586 ksi.ksi_lid = l1->l_lid; 587 kpsignal(p1, &ksi, NULL); 588 } 589 590 mutex_exit(proc_lock); 591 592 /* 593 * Return child pid to parent process, 594 * marking us as parent via retval[1]. 595 */ 596 if (retval != NULL) { 597 retval[0] = p2->p_pid; 598 retval[1] = 0; 599 } 600 601 return 0; 602 } 603