1 /* $OpenBSD: kern_fork.c,v 1.167 2014/05/15 03:52:25 guenther Exp $ */ 2 /* $NetBSD: kern_fork.c,v 1.29 1996/02/09 18:59:34 christos Exp $ */ 3 4 /* 5 * Copyright (c) 1982, 1986, 1989, 1991, 1993 6 * The Regents of the University of California. All rights reserved. 7 * (c) UNIX System Laboratories, Inc. 8 * All or some portions of this file are derived from material licensed 9 * to the University of California by American Telephone and Telegraph 10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 11 * the permission of UNIX System Laboratories, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 38 */ 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/filedesc.h> 43 #include <sys/kernel.h> 44 #include <sys/malloc.h> 45 #include <sys/mount.h> 46 #include <sys/proc.h> 47 #include <sys/exec.h> 48 #include <sys/resourcevar.h> 49 #include <sys/signalvar.h> 50 #include <sys/vnode.h> 51 #include <sys/vmmeter.h> 52 #include <sys/file.h> 53 #include <sys/acct.h> 54 #include <sys/ktrace.h> 55 #include <sys/sched.h> 56 #include <dev/rndvar.h> 57 #include <sys/pool.h> 58 #include <sys/mman.h> 59 #include <sys/ptrace.h> 60 61 #include <sys/syscallargs.h> 62 63 #include "systrace.h" 64 #include <dev/systrace.h> 65 66 #include <uvm/uvm.h> 67 68 #ifdef __HAVE_MD_TCB 69 # include <machine/tcb.h> 70 #endif 71 72 int nprocesses = 1; /* process 0 */ 73 int nthreads = 1; /* proc 0 */ 74 int randompid; /* when set to 1, pid's go random */ 75 struct forkstat forkstat; 76 77 void fork_return(void *); 78 void tfork_child_return(void *); 79 int pidtaken(pid_t); 80 81 void process_new(struct proc *, struct process *, int); 82 83 void 84 fork_return(void *arg) 85 { 86 struct proc *p = (struct proc *)arg; 87 88 if (p->p_p->ps_flags & PS_TRACED) 89 psignal(p, SIGTRAP); 90 91 child_return(p); 92 } 93 94 /*ARGSUSED*/ 95 int 96 sys_fork(struct proc *p, void *v, register_t *retval) 97 { 98 int flags; 99 100 flags = FORK_FORK; 101 if (p->p_p->ps_ptmask & PTRACE_FORK) 102 flags |= FORK_PTRACE; 103 return (fork1(p, flags, NULL, 0, fork_return, NULL, retval, NULL)); 104 } 105 106 /*ARGSUSED*/ 107 int 108 sys_vfork(struct proc *p, void *v, register_t *retval) 109 { 110 return (fork1(p, FORK_VFORK|FORK_PPWAIT, NULL, 0, NULL, 111 NULL, retval, NULL)); 112 } 113 114 int 115 sys___tfork(struct proc *p, void *v, register_t *retval) 116 { 117 struct sys___tfork_args /* { 118 syscallarg(const struct __tfork) *param; 119 syscallarg(size_t) psize; 120 } */ *uap = v; 121 size_t psize = SCARG(uap, psize); 122 struct __tfork param = { 0 }; 123 int flags; 124 int error; 125 126 if (psize == 0 || psize > sizeof(param)) 127 return (EINVAL); 128 if ((error = copyin(SCARG(uap, param), ¶m, psize))) 129 return (error); 130 #ifdef KTRACE 131 if (KTRPOINT(p, KTR_STRUCT)) 132 ktrstruct(p, "tfork", ¶m, sizeof(param)); 133 #endif 134 135 flags = FORK_TFORK | FORK_THREAD | FORK_SIGHAND | FORK_SHAREVM 136 | FORK_SHAREFILES; 137 138 return (fork1(p, flags, param.tf_stack, param.tf_tid, 139 tfork_child_return, param.tf_tcb, retval, NULL)); 140 } 141 142 void 143 tfork_child_return(void *arg) 144 { 145 struct proc *p = curproc; 146 147 TCB_SET(p, arg); 148 child_return(p); 149 } 150 151 /* 152 * Allocate and initialize a new process. 153 */ 154 void 155 process_new(struct proc *p, struct process *parent, int flags) 156 { 157 struct process *pr; 158 159 pr = pool_get(&process_pool, PR_WAITOK); 160 pr->ps_mainproc = p; 161 162 TAILQ_INIT(&pr->ps_threads); 163 TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link); 164 pr->ps_pptr = parent; 165 LIST_INIT(&pr->ps_children); 166 pr->ps_refcnt = 1; 167 168 /* 169 * Make a process structure for the new process. 170 * Start by zeroing the section of proc that is zero-initialized, 171 * then copy the section that is copied directly from the parent. 172 */ 173 memset(&pr->ps_startzero, 0, 174 (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero); 175 memcpy(&pr->ps_startcopy, &parent->ps_startcopy, 176 (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy); 177 178 /* post-copy fixups */ 179 pr->ps_ucred = p->p_ucred; 180 crhold(pr->ps_ucred); 181 KASSERT(p->p_ucred->cr_ref >= 3); /* fork thr, new thr, new process */ 182 pr->ps_limit->p_refcnt++; 183 184 /* bump references to the text vnode (for procfs) */ 185 pr->ps_textvp = parent->ps_textvp; 186 if (pr->ps_textvp) 187 vref(pr->ps_textvp); 188 189 timeout_set(&pr->ps_realit_to, realitexpire, pr); 190 191 pr->ps_flags = parent->ps_flags & (PS_SUGID | PS_SUGIDEXEC); 192 if (parent->ps_session->s_ttyvp != NULL && 193 parent->ps_flags & PS_CONTROLT) 194 atomic_setbits_int(&pr->ps_flags, PS_CONTROLT); 195 196 p->p_p = pr; 197 198 /* 199 * Duplicate sub-structures as needed. 200 * Increase reference counts on shared objects. 201 */ 202 if (flags & FORK_SHAREFILES) 203 pr->ps_fd = fdshare(parent); 204 else 205 pr->ps_fd = fdcopy(parent); 206 if (flags & FORK_SIGHAND) 207 pr->ps_sigacts = sigactsshare(parent); 208 else 209 pr->ps_sigacts = sigactsinit(parent); 210 if (flags & FORK_SHAREVM) 211 pr->ps_vmspace = uvmspace_share(parent); 212 else 213 pr->ps_vmspace = uvmspace_fork(parent); 214 215 if (parent->ps_flags & PS_PROFIL) 216 startprofclock(pr); 217 if ((flags & FORK_PTRACE) && (parent->ps_flags & PS_TRACED)) 218 atomic_setbits_int(&pr->ps_flags, PS_TRACED); 219 if (flags & FORK_NOZOMBIE) 220 atomic_setbits_int(&pr->ps_flags, PS_NOZOMBIE); 221 if (flags & FORK_SYSTEM) 222 atomic_setbits_int(&pr->ps_flags, PS_SYSTEM); 223 224 /* it's sufficiently inited to be globally visible */ 225 LIST_INSERT_HEAD(&allprocess, pr, ps_list); 226 } 227 228 /* print the 'table full' message once per 10 seconds */ 229 struct timeval fork_tfmrate = { 10, 0 }; 230 231 int 232 fork1(struct proc *curp, int flags, void *stack, pid_t *tidptr, 233 void (*func)(void *), void *arg, register_t *retval, 234 struct proc **rnewprocp) 235 { 236 struct process *curpr = curp->p_p; 237 struct process *pr; 238 struct proc *p; 239 uid_t uid; 240 struct vmspace *vm; 241 int count; 242 vaddr_t uaddr; 243 int s; 244 struct ptrace_state *newptstat = NULL; 245 #if NSYSTRACE > 0 246 void *newstrp = NULL; 247 #endif 248 249 /* sanity check some flag combinations */ 250 if (flags & FORK_THREAD) { 251 if ((flags & FORK_SIGHAND) == 0 || (flags & FORK_SYSTEM) != 0) 252 return (EINVAL); 253 } 254 if (flags & FORK_SIGHAND && (flags & FORK_SHAREVM) == 0) 255 return (EINVAL); 256 257 /* 258 * Although process entries are dynamically created, we still keep 259 * a global limit on the maximum number we will create. We reserve 260 * the last 5 processes to root. The variable nprocesses is the 261 * current number of processes, maxprocess is the limit. Similar 262 * rules for threads (struct proc): we reserve the last 5 to root; 263 * the variable nthreads is the current number of procs, maxthread is 264 * the limit. 265 */ 266 uid = curp->p_ucred->cr_ruid; 267 if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) { 268 static struct timeval lasttfm; 269 270 if (ratecheck(&lasttfm, &fork_tfmrate)) 271 tablefull("proc"); 272 return (EAGAIN); 273 } 274 nthreads++; 275 276 if ((flags & FORK_THREAD) == 0) { 277 if ((nprocesses >= maxprocess - 5 && uid != 0) || 278 nprocesses >= maxprocess) { 279 static struct timeval lasttfm; 280 281 if (ratecheck(&lasttfm, &fork_tfmrate)) 282 tablefull("process"); 283 nthreads--; 284 return (EAGAIN); 285 } 286 nprocesses++; 287 288 /* 289 * Increment the count of processes running with 290 * this uid. Don't allow a nonprivileged user to 291 * exceed their current limit. 292 */ 293 count = chgproccnt(uid, 1); 294 if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) { 295 (void)chgproccnt(uid, -1); 296 nprocesses--; 297 nthreads--; 298 return (EAGAIN); 299 } 300 } 301 302 uaddr = uvm_uarea_alloc(); 303 if (uaddr == 0) { 304 if ((flags & FORK_THREAD) == 0) { 305 (void)chgproccnt(uid, -1); 306 nprocesses--; 307 } 308 nthreads--; 309 return (ENOMEM); 310 } 311 312 /* 313 * From now on, we're committed to the fork and cannot fail. 314 */ 315 316 /* Allocate new proc. */ 317 p = pool_get(&proc_pool, PR_WAITOK); 318 319 p->p_stat = SIDL; /* protect against others */ 320 p->p_flag = 0; 321 322 /* 323 * Make a proc table entry for the new process. 324 * Start by zeroing the section of proc that is zero-initialized, 325 * then copy the section that is copied directly from the parent. 326 */ 327 memset(&p->p_startzero, 0, 328 (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero); 329 memcpy(&p->p_startcopy, &curp->p_startcopy, 330 (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy); 331 crhold(p->p_ucred); 332 333 /* 334 * Initialize the timeouts. 335 */ 336 timeout_set(&p->p_sleep_to, endtsleep, p); 337 338 if (flags & FORK_THREAD) { 339 atomic_setbits_int(&p->p_flag, P_THREAD); 340 p->p_p = pr = curpr; 341 pr->ps_refcnt++; 342 } else { 343 process_new(p, curpr, flags); 344 pr = p->p_p; 345 } 346 p->p_fd = pr->ps_fd; 347 p->p_vmspace = pr->ps_vmspace; 348 if (pr->ps_flags & PS_SYSTEM) 349 atomic_setbits_int(&p->p_flag, P_SYSTEM); 350 351 if (flags & FORK_PPWAIT) { 352 atomic_setbits_int(&pr->ps_flags, PS_PPWAIT); 353 atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT); 354 } 355 356 #ifdef KTRACE 357 /* 358 * Copy traceflag and tracefile if enabled. 359 * If not inherited, these were zeroed above. 360 */ 361 if ((flags & FORK_THREAD) == 0 && curpr->ps_traceflag & KTRFAC_INHERIT) 362 ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp, 363 curpr->ps_tracecred); 364 #endif 365 366 /* 367 * set priority of child to be that of parent 368 * XXX should move p_estcpu into the region of struct proc which gets 369 * copied. 370 */ 371 scheduler_fork_hook(curp, p); 372 373 if (flags & FORK_THREAD) 374 sigstkinit(&p->p_sigstk); 375 376 /* 377 * If emulation has thread fork hook, call it now. 378 */ 379 if (pr->ps_emul->e_proc_fork) 380 (*pr->ps_emul->e_proc_fork)(p, curp); 381 382 p->p_addr = (struct user *)uaddr; 383 384 /* 385 * Finish creating the child thread. cpu_fork() will copy 386 * and update the pcb and make the child ready to run. If 387 * this is a normal user fork, the child will exit directly 388 * to user mode via child_return() on its first time slice 389 * and will not return here. If this is a kernel thread, 390 * the specified entry point will be executed. 391 */ 392 cpu_fork(curp, p, stack, 0, func ? func : child_return, arg ? arg : p); 393 394 vm = pr->ps_vmspace; 395 396 if (flags & FORK_FORK) { 397 forkstat.cntfork++; 398 forkstat.sizfork += vm->vm_dsize + vm->vm_ssize; 399 } else if (flags & FORK_VFORK) { 400 forkstat.cntvfork++; 401 forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize; 402 } else if (flags & FORK_TFORK) { 403 forkstat.cnttfork++; 404 } else { 405 forkstat.cntkthread++; 406 forkstat.sizkthread += vm->vm_dsize + vm->vm_ssize; 407 } 408 409 if (pr->ps_flags & PS_TRACED && flags & FORK_FORK) 410 newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK); 411 #if NSYSTRACE > 0 412 if (ISSET(curp->p_flag, P_SYSTRACE)) 413 newstrp = systrace_getproc(); 414 #endif 415 416 p->p_pid = allocpid(); 417 418 LIST_INSERT_HEAD(&allproc, p, p_list); 419 LIST_INSERT_HEAD(PIDHASH(p->p_pid), p, p_hash); 420 if ((flags & FORK_THREAD) == 0) { 421 LIST_INSERT_AFTER(curpr, pr, ps_pglist); 422 LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling); 423 424 if (pr->ps_flags & PS_TRACED) { 425 pr->ps_oppid = curpr->ps_pid; 426 if (pr->ps_pptr != curpr->ps_pptr) 427 proc_reparent(pr, curpr->ps_pptr); 428 429 /* 430 * Set ptrace status. 431 */ 432 if (flags & FORK_FORK) { 433 pr->ps_ptstat = newptstat; 434 newptstat = NULL; 435 curpr->ps_ptstat->pe_report_event = PTRACE_FORK; 436 pr->ps_ptstat->pe_report_event = PTRACE_FORK; 437 curpr->ps_ptstat->pe_other_pid = pr->ps_pid; 438 pr->ps_ptstat->pe_other_pid = curpr->ps_pid; 439 } 440 } 441 } else { 442 TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link); 443 /* 444 * if somebody else wants to take us to single threaded mode, 445 * count ourselves in. 446 */ 447 if (pr->ps_single) { 448 curpr->ps_singlecount++; 449 atomic_setbits_int(&p->p_flag, P_SUSPSINGLE); 450 } 451 } 452 453 #if NSYSTRACE > 0 454 if (newstrp) 455 systrace_fork(curp, p, newstrp); 456 #endif 457 458 if (tidptr != NULL) { 459 pid_t pid = p->p_pid + THREAD_PID_OFFSET; 460 461 if (copyout(&pid, tidptr, sizeof(pid))) 462 psignal(curp, SIGSEGV); 463 } 464 465 /* 466 * For new processes, set accounting bits 467 */ 468 if ((flags & FORK_THREAD) == 0) { 469 getnanotime(&pr->ps_start); 470 pr->ps_acflag = AFORK; 471 } 472 473 /* 474 * Make child runnable and add to run queue. 475 */ 476 if ((flags & FORK_IDLE) == 0) { 477 SCHED_LOCK(s); 478 p->p_stat = SRUN; 479 p->p_cpu = sched_choosecpu_fork(curp, flags); 480 setrunqueue(p); 481 SCHED_UNLOCK(s); 482 } else 483 p->p_cpu = arg; 484 485 if (newptstat) 486 free(newptstat, M_SUBPROC); 487 488 /* 489 * Notify any interested parties about the new process. 490 */ 491 if ((flags & FORK_THREAD) == 0) 492 KNOTE(&curpr->ps_klist, NOTE_FORK | p->p_pid); 493 494 /* 495 * Update stats now that we know the fork was successful. 496 */ 497 uvmexp.forks++; 498 if (flags & FORK_PPWAIT) 499 uvmexp.forks_ppwait++; 500 if (flags & FORK_SHAREVM) 501 uvmexp.forks_sharevm++; 502 503 /* 504 * Pass a pointer to the new process to the caller. 505 */ 506 if (rnewprocp != NULL) 507 *rnewprocp = p; 508 509 /* 510 * Preserve synchronization semantics of vfork. If waiting for 511 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT 512 * on ourselves, and sleep on our process for the latter flag 513 * to go away. 514 * XXX Need to stop other rthreads in the parent 515 */ 516 if (flags & FORK_PPWAIT) 517 while (curpr->ps_flags & PS_ISPWAIT) 518 tsleep(curpr, PWAIT, "ppwait", 0); 519 520 /* 521 * If we're tracing the child, alert the parent too. 522 */ 523 if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED)) 524 psignal(curp, SIGTRAP); 525 526 /* 527 * Return child pid to parent process, 528 * marking us as parent via retval[1]. 529 */ 530 if (retval != NULL) { 531 retval[0] = p->p_pid + 532 (flags & FORK_THREAD ? THREAD_PID_OFFSET : 0); 533 retval[1] = 0; 534 } 535 return (0); 536 } 537 538 /* 539 * Checks for current use of a pid, either as a pid or pgid. 540 */ 541 pid_t oldpids[100]; 542 int 543 ispidtaken(pid_t pid) 544 { 545 uint32_t i; 546 struct process *pr; 547 548 for (i = 0; i < nitems(oldpids); i++) 549 if (pid == oldpids[i]) 550 return (1); 551 552 if (pfind(pid) != NULL) 553 return (1); 554 if (pgfind(pid) != NULL) 555 return (1); 556 LIST_FOREACH(pr, &zombprocess, ps_list) { 557 if (pr->ps_pid == pid || 558 (pr->ps_pgrp && pr->ps_pgrp->pg_id == pid)) 559 return (1); 560 } 561 return (0); 562 } 563 564 /* Find an unused pid satisfying 1 <= lastpid <= PID_MAX */ 565 pid_t 566 allocpid(void) 567 { 568 static pid_t lastpid; 569 pid_t pid; 570 571 if (!randompid) { 572 /* only used early on for system processes */ 573 pid = ++lastpid; 574 } else { 575 do { 576 pid = 1 + arc4random_uniform(PID_MAX); 577 } while (ispidtaken(pid)); 578 } 579 580 return pid; 581 } 582 583 void 584 freepid(pid_t pid) 585 { 586 static uint32_t idx; 587 588 oldpids[idx++ % nitems(oldpids)] = pid; 589 } 590 591 #if defined(MULTIPROCESSOR) 592 /* 593 * XXX This is a slight hack to get newly-formed processes to 594 * XXX acquire the kernel lock as soon as they run. 595 */ 596 void 597 proc_trampoline_mp(void) 598 { 599 struct proc *p; 600 601 p = curproc; 602 603 SCHED_ASSERT_LOCKED(); 604 __mp_unlock(&sched_lock); 605 spl0(); 606 SCHED_ASSERT_UNLOCKED(); 607 KASSERT(__mp_lock_held(&kernel_lock) == 0); 608 609 KERNEL_LOCK(); 610 } 611 #endif 612