1 /* $OpenBSD: kern_fork.c,v 1.207 2018/08/30 03:30:25 visa 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/acct.h> 53 #include <sys/ktrace.h> 54 #include <sys/sched.h> 55 #include <sys/sysctl.h> 56 #include <sys/pool.h> 57 #include <sys/mman.h> 58 #include <sys/ptrace.h> 59 #include <sys/atomic.h> 60 #include <sys/pledge.h> 61 #include <sys/unistd.h> 62 63 #include <sys/syscallargs.h> 64 65 #include <uvm/uvm.h> 66 #include <machine/tcb.h> 67 68 #include "kcov.h" 69 70 int nprocesses = 1; /* process 0 */ 71 int nthreads = 1; /* proc 0 */ 72 int randompid; /* when set to 1, pid's go random */ 73 struct forkstat forkstat; 74 75 void fork_return(void *); 76 pid_t alloctid(void); 77 pid_t allocpid(void); 78 int ispidtaken(pid_t); 79 80 void unveil_copy(struct process *parent, struct process *child); 81 82 struct proc *thread_new(struct proc *_parent, vaddr_t _uaddr); 83 struct process *process_new(struct proc *, struct process *, int); 84 int fork_check_maxthread(uid_t _uid); 85 86 void 87 fork_return(void *arg) 88 { 89 struct proc *p = (struct proc *)arg; 90 91 if (p->p_p->ps_flags & PS_TRACED) 92 psignal(p, SIGTRAP); 93 94 child_return(p); 95 } 96 97 int 98 sys_fork(struct proc *p, void *v, register_t *retval) 99 { 100 int flags; 101 102 flags = FORK_FORK; 103 if (p->p_p->ps_ptmask & PTRACE_FORK) 104 flags |= FORK_PTRACE; 105 return fork1(p, flags, fork_return, NULL, retval, NULL); 106 } 107 108 int 109 sys_vfork(struct proc *p, void *v, register_t *retval) 110 { 111 return fork1(p, FORK_VFORK|FORK_PPWAIT, child_return, NULL, 112 retval, NULL); 113 } 114 115 int 116 sys___tfork(struct proc *p, void *v, register_t *retval) 117 { 118 struct sys___tfork_args /* { 119 syscallarg(const struct __tfork) *param; 120 syscallarg(size_t) psize; 121 } */ *uap = v; 122 size_t psize = SCARG(uap, psize); 123 struct __tfork param = { 0 }; 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 #ifdef TCB_INVALID 135 if (TCB_INVALID(param.tf_tcb)) 136 return EINVAL; 137 #endif /* TCB_INVALID */ 138 139 return thread_fork(p, param.tf_stack, param.tf_tcb, param.tf_tid, 140 retval); 141 } 142 143 /* 144 * Allocate and initialize a thread (proc) structure, given the parent thread. 145 */ 146 struct proc * 147 thread_new(struct proc *parent, vaddr_t uaddr) 148 { 149 struct proc *p; 150 151 p = pool_get(&proc_pool, PR_WAITOK); 152 p->p_stat = SIDL; /* protect against others */ 153 p->p_flag = 0; 154 155 /* 156 * Make a proc table entry for the new process. 157 * Start by zeroing the section of proc that is zero-initialized, 158 * then copy the section that is copied directly from the parent. 159 */ 160 memset(&p->p_startzero, 0, 161 (caddr_t)&p->p_endzero - (caddr_t)&p->p_startzero); 162 memcpy(&p->p_startcopy, &parent->p_startcopy, 163 (caddr_t)&p->p_endcopy - (caddr_t)&p->p_startcopy); 164 crhold(p->p_ucred); 165 p->p_addr = (struct user *)uaddr; 166 167 /* 168 * Initialize the timeouts. 169 */ 170 timeout_set(&p->p_sleep_to, endtsleep, p); 171 172 /* 173 * set priority of child to be that of parent 174 * XXX should move p_estcpu into the region of struct proc which gets 175 * copied. 176 */ 177 scheduler_fork_hook(parent, p); 178 179 #ifdef WITNESS 180 p->p_sleeplocks = NULL; 181 #endif 182 183 #if NKCOV > 0 184 p->p_kd = NULL; 185 #endif 186 187 return p; 188 } 189 190 /* 191 * Initialize common bits of a process structure, given the initial thread. 192 */ 193 void 194 process_initialize(struct process *pr, struct proc *p) 195 { 196 /* initialize the thread links */ 197 pr->ps_mainproc = p; 198 TAILQ_INIT(&pr->ps_threads); 199 TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link); 200 pr->ps_refcnt = 1; 201 p->p_p = pr; 202 203 /* give the process the same creds as the initial thread */ 204 pr->ps_ucred = p->p_ucred; 205 crhold(pr->ps_ucred); 206 KASSERT(p->p_ucred->cr_ref >= 2); /* new thread and new process */ 207 208 LIST_INIT(&pr->ps_children); 209 LIST_INIT(&pr->ps_ftlist); 210 LIST_INIT(&pr->ps_kqlist); 211 212 timeout_set(&pr->ps_realit_to, realitexpire, pr); 213 } 214 215 216 /* 217 * Allocate and initialize a new process. 218 */ 219 struct process * 220 process_new(struct proc *p, struct process *parent, int flags) 221 { 222 struct process *pr; 223 224 pr = pool_get(&process_pool, PR_WAITOK); 225 226 /* 227 * Make a process structure for the new process. 228 * Start by zeroing the section of proc that is zero-initialized, 229 * then copy the section that is copied directly from the parent. 230 */ 231 memset(&pr->ps_startzero, 0, 232 (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero); 233 memcpy(&pr->ps_startcopy, &parent->ps_startcopy, 234 (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy); 235 236 process_initialize(pr, p); 237 pr->ps_pid = allocpid(); 238 239 /* post-copy fixups */ 240 pr->ps_pptr = parent; 241 pr->ps_limit->p_refcnt++; 242 243 /* bump references to the text vnode (for sysctl) */ 244 pr->ps_textvp = parent->ps_textvp; 245 if (pr->ps_textvp) 246 vref(pr->ps_textvp); 247 248 /* copy unveil if unveil is active */ 249 unveil_copy(parent, pr); 250 251 pr->ps_flags = parent->ps_flags & 252 (PS_SUGID | PS_SUGIDEXEC | PS_PLEDGE | PS_EXECPLEDGE | PS_WXNEEDED); 253 if (parent->ps_session->s_ttyvp != NULL) 254 pr->ps_flags |= parent->ps_flags & PS_CONTROLT; 255 256 /* 257 * Duplicate sub-structures as needed. 258 * Increase reference counts on shared objects. 259 */ 260 if (flags & FORK_SHAREFILES) 261 pr->ps_fd = fdshare(parent); 262 else 263 pr->ps_fd = fdcopy(parent); 264 if (flags & FORK_SIGHAND) 265 pr->ps_sigacts = sigactsshare(parent); 266 else 267 pr->ps_sigacts = sigactsinit(parent); 268 if (flags & FORK_SHAREVM) 269 pr->ps_vmspace = uvmspace_share(parent); 270 else 271 pr->ps_vmspace = uvmspace_fork(parent); 272 273 if (parent->ps_flags & PS_PROFIL) 274 startprofclock(pr); 275 if (flags & FORK_PTRACE) 276 pr->ps_flags |= parent->ps_flags & PS_TRACED; 277 if (flags & FORK_NOZOMBIE) 278 pr->ps_flags |= PS_NOZOMBIE; 279 if (flags & FORK_SYSTEM) 280 pr->ps_flags |= PS_SYSTEM; 281 282 /* mark as embryo to protect against others */ 283 pr->ps_flags |= PS_EMBRYO; 284 285 /* Force visibility of all of the above changes */ 286 membar_producer(); 287 288 /* it's sufficiently inited to be globally visible */ 289 LIST_INSERT_HEAD(&allprocess, pr, ps_list); 290 291 return pr; 292 } 293 294 /* print the 'table full' message once per 10 seconds */ 295 struct timeval fork_tfmrate = { 10, 0 }; 296 297 int 298 fork_check_maxthread(uid_t uid) 299 { 300 /* 301 * Although process entries are dynamically created, we still keep 302 * a global limit on the maximum number we will create. We reserve 303 * the last 5 processes to root. The variable nprocesses is the 304 * current number of processes, maxprocess is the limit. Similar 305 * rules for threads (struct proc): we reserve the last 5 to root; 306 * the variable nthreads is the current number of procs, maxthread is 307 * the limit. 308 */ 309 if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) { 310 static struct timeval lasttfm; 311 312 if (ratecheck(&lasttfm, &fork_tfmrate)) 313 tablefull("proc"); 314 return EAGAIN; 315 } 316 nthreads++; 317 318 return 0; 319 } 320 321 static inline void 322 fork_thread_start(struct proc *p, struct proc *parent, int flags) 323 { 324 int s; 325 326 SCHED_LOCK(s); 327 p->p_stat = SRUN; 328 p->p_cpu = sched_choosecpu_fork(parent, flags); 329 setrunqueue(p); 330 SCHED_UNLOCK(s); 331 } 332 333 int 334 fork1(struct proc *curp, int flags, void (*func)(void *), void *arg, 335 register_t *retval, struct proc **rnewprocp) 336 { 337 struct process *curpr = curp->p_p; 338 struct process *pr; 339 struct proc *p; 340 uid_t uid = curp->p_ucred->cr_ruid; 341 struct vmspace *vm; 342 int count; 343 vaddr_t uaddr; 344 int error; 345 struct ptrace_state *newptstat = NULL; 346 347 KASSERT((flags & ~(FORK_FORK | FORK_VFORK | FORK_PPWAIT | FORK_PTRACE 348 | FORK_IDLE | FORK_SHAREVM | FORK_SHAREFILES | FORK_NOZOMBIE 349 | FORK_SYSTEM | FORK_SIGHAND)) == 0); 350 KASSERT((flags & FORK_SIGHAND) == 0 || (flags & FORK_SHAREVM)); 351 KASSERT(func != NULL); 352 353 if ((error = fork_check_maxthread(uid))) 354 return error; 355 356 if ((nprocesses >= maxprocess - 5 && uid != 0) || 357 nprocesses >= maxprocess) { 358 static struct timeval lasttfm; 359 360 if (ratecheck(&lasttfm, &fork_tfmrate)) 361 tablefull("process"); 362 nthreads--; 363 return EAGAIN; 364 } 365 nprocesses++; 366 367 /* 368 * Increment the count of processes running with this uid. 369 * Don't allow a nonprivileged user to exceed their current limit. 370 */ 371 count = chgproccnt(uid, 1); 372 if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) { 373 (void)chgproccnt(uid, -1); 374 nprocesses--; 375 nthreads--; 376 return EAGAIN; 377 } 378 379 uaddr = uvm_uarea_alloc(); 380 if (uaddr == 0) { 381 (void)chgproccnt(uid, -1); 382 nprocesses--; 383 nthreads--; 384 return (ENOMEM); 385 } 386 387 /* 388 * From now on, we're committed to the fork and cannot fail. 389 */ 390 p = thread_new(curp, uaddr); 391 pr = process_new(p, curpr, flags); 392 393 p->p_fd = pr->ps_fd; 394 p->p_vmspace = pr->ps_vmspace; 395 if (pr->ps_flags & PS_SYSTEM) 396 atomic_setbits_int(&p->p_flag, P_SYSTEM); 397 398 if (flags & FORK_PPWAIT) { 399 atomic_setbits_int(&pr->ps_flags, PS_PPWAIT); 400 atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT); 401 } 402 403 #ifdef KTRACE 404 /* 405 * Copy traceflag and tracefile if enabled. 406 * If not inherited, these were zeroed above. 407 */ 408 if (curpr->ps_traceflag & KTRFAC_INHERIT) 409 ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp, 410 curpr->ps_tracecred); 411 #endif 412 413 /* 414 * Finish creating the child thread. cpu_fork() will copy 415 * and update the pcb and make the child ready to run. If 416 * this is a normal user fork, the child will exit directly 417 * to user mode via child_return() on its first time slice 418 * and will not return here. If this is a kernel thread, 419 * the specified entry point will be executed. 420 */ 421 cpu_fork(curp, p, NULL, NULL, func, arg ? arg : p); 422 423 vm = pr->ps_vmspace; 424 425 if (flags & FORK_FORK) { 426 forkstat.cntfork++; 427 forkstat.sizfork += vm->vm_dsize + vm->vm_ssize; 428 } else if (flags & FORK_VFORK) { 429 forkstat.cntvfork++; 430 forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize; 431 } else { 432 forkstat.cntkthread++; 433 } 434 435 if (pr->ps_flags & PS_TRACED && flags & FORK_FORK) 436 newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK); 437 438 p->p_tid = alloctid(); 439 440 LIST_INSERT_HEAD(&allproc, p, p_list); 441 LIST_INSERT_HEAD(TIDHASH(p->p_tid), p, p_hash); 442 LIST_INSERT_HEAD(PIDHASH(pr->ps_pid), pr, ps_hash); 443 LIST_INSERT_AFTER(curpr, pr, ps_pglist); 444 LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling); 445 446 if (pr->ps_flags & PS_TRACED) { 447 pr->ps_oppid = curpr->ps_pid; 448 if (pr->ps_pptr != curpr->ps_pptr) 449 proc_reparent(pr, curpr->ps_pptr); 450 451 /* 452 * Set ptrace status. 453 */ 454 if (newptstat != NULL) { 455 pr->ps_ptstat = newptstat; 456 newptstat = NULL; 457 curpr->ps_ptstat->pe_report_event = PTRACE_FORK; 458 pr->ps_ptstat->pe_report_event = PTRACE_FORK; 459 curpr->ps_ptstat->pe_other_pid = pr->ps_pid; 460 pr->ps_ptstat->pe_other_pid = curpr->ps_pid; 461 } 462 } 463 464 /* 465 * For new processes, set accounting bits and mark as complete. 466 */ 467 getnanotime(&pr->ps_start); 468 pr->ps_acflag = AFORK; 469 atomic_clearbits_int(&pr->ps_flags, PS_EMBRYO); 470 471 if ((flags & FORK_IDLE) == 0) 472 fork_thread_start(p, curp, flags); 473 else 474 p->p_cpu = arg; 475 476 free(newptstat, M_SUBPROC, sizeof(*newptstat)); 477 478 /* 479 * Notify any interested parties about the new process. 480 */ 481 KNOTE(&curpr->ps_klist, NOTE_FORK | pr->ps_pid); 482 483 /* 484 * Update stats now that we know the fork was successful. 485 */ 486 uvmexp.forks++; 487 if (flags & FORK_PPWAIT) 488 uvmexp.forks_ppwait++; 489 if (flags & FORK_SHAREVM) 490 uvmexp.forks_sharevm++; 491 492 /* 493 * Pass a pointer to the new process to the caller. 494 */ 495 if (rnewprocp != NULL) 496 *rnewprocp = p; 497 498 /* 499 * Preserve synchronization semantics of vfork. If waiting for 500 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT 501 * on ourselves, and sleep on our process for the latter flag 502 * to go away. 503 * XXX Need to stop other rthreads in the parent 504 */ 505 if (flags & FORK_PPWAIT) 506 while (curpr->ps_flags & PS_ISPWAIT) 507 tsleep(curpr, PWAIT, "ppwait", 0); 508 509 /* 510 * If we're tracing the child, alert the parent too. 511 */ 512 if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED)) 513 psignal(curp, SIGTRAP); 514 515 /* 516 * Return child pid to parent process 517 */ 518 if (retval != NULL) { 519 retval[0] = pr->ps_pid; 520 retval[1] = 0; 521 } 522 return (0); 523 } 524 525 int 526 thread_fork(struct proc *curp, void *stack, void *tcb, pid_t *tidptr, 527 register_t *retval) 528 { 529 struct process *pr = curp->p_p; 530 struct proc *p; 531 pid_t tid; 532 vaddr_t uaddr; 533 int error; 534 535 if (stack == NULL) 536 return EINVAL; 537 538 if ((error = fork_check_maxthread(curp->p_ucred->cr_ruid))) 539 return error; 540 541 uaddr = uvm_uarea_alloc(); 542 if (uaddr == 0) { 543 nthreads--; 544 return ENOMEM; 545 } 546 547 /* 548 * From now on, we're committed to the fork and cannot fail. 549 */ 550 p = thread_new(curp, uaddr); 551 atomic_setbits_int(&p->p_flag, P_THREAD); 552 sigstkinit(&p->p_sigstk); 553 554 /* other links */ 555 p->p_p = pr; 556 pr->ps_refcnt++; 557 558 /* local copies */ 559 p->p_fd = pr->ps_fd; 560 p->p_vmspace = pr->ps_vmspace; 561 562 /* 563 * Finish creating the child thread. cpu_fork() will copy 564 * and update the pcb and make the child ready to run. The 565 * child will exit directly to user mode via child_return() 566 * on its first time slice and will not return here. 567 */ 568 cpu_fork(curp, p, stack, tcb, child_return, p); 569 570 p->p_tid = alloctid(); 571 572 LIST_INSERT_HEAD(&allproc, p, p_list); 573 LIST_INSERT_HEAD(TIDHASH(p->p_tid), p, p_hash); 574 TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link); 575 576 /* 577 * if somebody else wants to take us to single threaded mode, 578 * count ourselves in. 579 */ 580 if (pr->ps_single) { 581 pr->ps_singlecount++; 582 atomic_setbits_int(&p->p_flag, P_SUSPSINGLE); 583 } 584 585 /* 586 * Return tid to parent thread and copy it out to userspace 587 */ 588 retval[0] = tid = p->p_tid + THREAD_PID_OFFSET; 589 retval[1] = 0; 590 if (tidptr != NULL) { 591 if (copyout(&tid, tidptr, sizeof(tid))) 592 psignal(curp, SIGSEGV); 593 } 594 595 fork_thread_start(p, curp, 0); 596 597 /* 598 * Update stats now that we know the fork was successful. 599 */ 600 forkstat.cnttfork++; 601 uvmexp.forks++; 602 uvmexp.forks_sharevm++; 603 604 return 0; 605 } 606 607 608 /* Find an unused tid */ 609 pid_t 610 alloctid(void) 611 { 612 pid_t tid; 613 614 do { 615 /* (0 .. TID_MASK+1] */ 616 tid = 1 + (arc4random() & TID_MASK); 617 } while (tfind(tid) != NULL); 618 619 return (tid); 620 } 621 622 /* 623 * Checks for current use of a pid, either as a pid or pgid. 624 */ 625 pid_t oldpids[128]; 626 int 627 ispidtaken(pid_t pid) 628 { 629 uint32_t i; 630 631 for (i = 0; i < nitems(oldpids); i++) 632 if (pid == oldpids[i]) 633 return (1); 634 635 if (prfind(pid) != NULL) 636 return (1); 637 if (pgfind(pid) != NULL) 638 return (1); 639 if (zombiefind(pid) != NULL) 640 return (1); 641 return (0); 642 } 643 644 /* Find an unused pid */ 645 pid_t 646 allocpid(void) 647 { 648 static pid_t lastpid; 649 pid_t pid; 650 651 if (!randompid) { 652 /* only used early on for system processes */ 653 pid = ++lastpid; 654 } else { 655 /* Find an unused pid satisfying lastpid < pid <= PID_MAX */ 656 do { 657 pid = arc4random_uniform(PID_MAX - lastpid) + 1 + 658 lastpid; 659 } while (ispidtaken(pid)); 660 } 661 662 return pid; 663 } 664 665 void 666 freepid(pid_t pid) 667 { 668 static uint32_t idx; 669 670 oldpids[idx++ % nitems(oldpids)] = pid; 671 } 672 673 #if defined(MULTIPROCESSOR) 674 /* 675 * XXX This is a slight hack to get newly-formed processes to 676 * XXX acquire the kernel lock as soon as they run. 677 */ 678 void 679 proc_trampoline_mp(void) 680 { 681 SCHED_ASSERT_LOCKED(); 682 __mp_unlock(&sched_lock); 683 spl0(); 684 SCHED_ASSERT_UNLOCKED(); 685 KERNEL_ASSERT_UNLOCKED(); 686 687 KERNEL_LOCK(); 688 } 689 #endif 690