1 /* $OpenBSD: kern_fork.c,v 1.212 2019/06/01 14:11:17 mpi 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 LIST_INIT(&pr->ps_sigiolst); 212 213 timeout_set(&pr->ps_realit_to, realitexpire, pr); 214 timeout_set(&pr->ps_rucheck_to, rucheck, pr); 215 } 216 217 218 /* 219 * Allocate and initialize a new process. 220 */ 221 struct process * 222 process_new(struct proc *p, struct process *parent, int flags) 223 { 224 struct process *pr; 225 226 pr = pool_get(&process_pool, PR_WAITOK); 227 228 /* 229 * Make a process structure for the new process. 230 * Start by zeroing the section of proc that is zero-initialized, 231 * then copy the section that is copied directly from the parent. 232 */ 233 memset(&pr->ps_startzero, 0, 234 (caddr_t)&pr->ps_endzero - (caddr_t)&pr->ps_startzero); 235 memcpy(&pr->ps_startcopy, &parent->ps_startcopy, 236 (caddr_t)&pr->ps_endcopy - (caddr_t)&pr->ps_startcopy); 237 238 process_initialize(pr, p); 239 pr->ps_pid = allocpid(); 240 241 /* post-copy fixups */ 242 pr->ps_pptr = parent; 243 pr->ps_limit->pl_refcnt++; 244 if (pr->ps_limit->pl_rlimit[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) 245 timeout_add_msec(&pr->ps_rucheck_to, RUCHECK_INTERVAL); 246 247 /* bump references to the text vnode (for sysctl) */ 248 pr->ps_textvp = parent->ps_textvp; 249 if (pr->ps_textvp) 250 vref(pr->ps_textvp); 251 252 /* copy unveil if unveil is active */ 253 unveil_copy(parent, pr); 254 255 pr->ps_flags = parent->ps_flags & 256 (PS_SUGID | PS_SUGIDEXEC | PS_PLEDGE | PS_EXECPLEDGE | PS_WXNEEDED); 257 if (parent->ps_session->s_ttyvp != NULL) 258 pr->ps_flags |= parent->ps_flags & PS_CONTROLT; 259 260 /* 261 * Duplicate sub-structures as needed. 262 * Increase reference counts on shared objects. 263 */ 264 if (flags & FORK_SHAREFILES) 265 pr->ps_fd = fdshare(parent); 266 else 267 pr->ps_fd = fdcopy(parent); 268 if (flags & FORK_SIGHAND) 269 pr->ps_sigacts = sigactsshare(parent); 270 else 271 pr->ps_sigacts = sigactsinit(parent); 272 if (flags & FORK_SHAREVM) 273 pr->ps_vmspace = uvmspace_share(parent); 274 else 275 pr->ps_vmspace = uvmspace_fork(parent); 276 277 if (parent->ps_flags & PS_PROFIL) 278 startprofclock(pr); 279 if (flags & FORK_PTRACE) 280 pr->ps_flags |= parent->ps_flags & PS_TRACED; 281 if (flags & FORK_NOZOMBIE) 282 pr->ps_flags |= PS_NOZOMBIE; 283 if (flags & FORK_SYSTEM) 284 pr->ps_flags |= PS_SYSTEM; 285 286 /* mark as embryo to protect against others */ 287 pr->ps_flags |= PS_EMBRYO; 288 289 /* Force visibility of all of the above changes */ 290 membar_producer(); 291 292 /* it's sufficiently inited to be globally visible */ 293 LIST_INSERT_HEAD(&allprocess, pr, ps_list); 294 295 return pr; 296 } 297 298 /* print the 'table full' message once per 10 seconds */ 299 struct timeval fork_tfmrate = { 10, 0 }; 300 301 int 302 fork_check_maxthread(uid_t uid) 303 { 304 /* 305 * Although process entries are dynamically created, we still keep 306 * a global limit on the maximum number we will create. We reserve 307 * the last 5 processes to root. The variable nprocesses is the 308 * current number of processes, maxprocess is the limit. Similar 309 * rules for threads (struct proc): we reserve the last 5 to root; 310 * the variable nthreads is the current number of procs, maxthread is 311 * the limit. 312 */ 313 if ((nthreads >= maxthread - 5 && uid != 0) || nthreads >= maxthread) { 314 static struct timeval lasttfm; 315 316 if (ratecheck(&lasttfm, &fork_tfmrate)) 317 tablefull("proc"); 318 return EAGAIN; 319 } 320 nthreads++; 321 322 return 0; 323 } 324 325 static inline void 326 fork_thread_start(struct proc *p, struct proc *parent, int flags) 327 { 328 int s; 329 330 SCHED_LOCK(s); 331 p->p_stat = SRUN; 332 p->p_cpu = sched_choosecpu_fork(parent, flags); 333 setrunqueue(p); 334 SCHED_UNLOCK(s); 335 } 336 337 int 338 fork1(struct proc *curp, int flags, void (*func)(void *), void *arg, 339 register_t *retval, struct proc **rnewprocp) 340 { 341 struct process *curpr = curp->p_p; 342 struct process *pr; 343 struct proc *p; 344 uid_t uid = curp->p_ucred->cr_ruid; 345 struct vmspace *vm; 346 int count; 347 vaddr_t uaddr; 348 int error; 349 struct ptrace_state *newptstat = NULL; 350 351 KASSERT((flags & ~(FORK_FORK | FORK_VFORK | FORK_PPWAIT | FORK_PTRACE 352 | FORK_IDLE | FORK_SHAREVM | FORK_SHAREFILES | FORK_NOZOMBIE 353 | FORK_SYSTEM | FORK_SIGHAND)) == 0); 354 KASSERT((flags & FORK_SIGHAND) == 0 || (flags & FORK_SHAREVM)); 355 KASSERT(func != NULL); 356 357 if ((error = fork_check_maxthread(uid))) 358 return error; 359 360 if ((nprocesses >= maxprocess - 5 && uid != 0) || 361 nprocesses >= maxprocess) { 362 static struct timeval lasttfm; 363 364 if (ratecheck(&lasttfm, &fork_tfmrate)) 365 tablefull("process"); 366 nthreads--; 367 return EAGAIN; 368 } 369 nprocesses++; 370 371 /* 372 * Increment the count of processes running with this uid. 373 * Don't allow a nonprivileged user to exceed their current limit. 374 */ 375 count = chgproccnt(uid, 1); 376 if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) { 377 (void)chgproccnt(uid, -1); 378 nprocesses--; 379 nthreads--; 380 return EAGAIN; 381 } 382 383 uaddr = uvm_uarea_alloc(); 384 if (uaddr == 0) { 385 (void)chgproccnt(uid, -1); 386 nprocesses--; 387 nthreads--; 388 return (ENOMEM); 389 } 390 391 /* 392 * From now on, we're committed to the fork and cannot fail. 393 */ 394 p = thread_new(curp, uaddr); 395 pr = process_new(p, curpr, flags); 396 397 p->p_fd = pr->ps_fd; 398 p->p_vmspace = pr->ps_vmspace; 399 if (pr->ps_flags & PS_SYSTEM) 400 atomic_setbits_int(&p->p_flag, P_SYSTEM); 401 402 if (flags & FORK_PPWAIT) { 403 atomic_setbits_int(&pr->ps_flags, PS_PPWAIT); 404 atomic_setbits_int(&curpr->ps_flags, PS_ISPWAIT); 405 } 406 407 #ifdef KTRACE 408 /* 409 * Copy traceflag and tracefile if enabled. 410 * If not inherited, these were zeroed above. 411 */ 412 if (curpr->ps_traceflag & KTRFAC_INHERIT) 413 ktrsettrace(pr, curpr->ps_traceflag, curpr->ps_tracevp, 414 curpr->ps_tracecred); 415 #endif 416 417 /* 418 * Finish creating the child thread. cpu_fork() will copy 419 * and update the pcb and make the child ready to run. If 420 * this is a normal user fork, the child will exit directly 421 * to user mode via child_return() on its first time slice 422 * and will not return here. If this is a kernel thread, 423 * the specified entry point will be executed. 424 */ 425 cpu_fork(curp, p, NULL, NULL, func, arg ? arg : p); 426 427 vm = pr->ps_vmspace; 428 429 if (flags & FORK_FORK) { 430 forkstat.cntfork++; 431 forkstat.sizfork += vm->vm_dsize + vm->vm_ssize; 432 } else if (flags & FORK_VFORK) { 433 forkstat.cntvfork++; 434 forkstat.sizvfork += vm->vm_dsize + vm->vm_ssize; 435 } else { 436 forkstat.cntkthread++; 437 } 438 439 if (pr->ps_flags & PS_TRACED && flags & FORK_FORK) 440 newptstat = malloc(sizeof(*newptstat), M_SUBPROC, M_WAITOK); 441 442 p->p_tid = alloctid(); 443 444 LIST_INSERT_HEAD(&allproc, p, p_list); 445 LIST_INSERT_HEAD(TIDHASH(p->p_tid), p, p_hash); 446 LIST_INSERT_HEAD(PIDHASH(pr->ps_pid), pr, ps_hash); 447 LIST_INSERT_AFTER(curpr, pr, ps_pglist); 448 LIST_INSERT_HEAD(&curpr->ps_children, pr, ps_sibling); 449 450 if (pr->ps_flags & PS_TRACED) { 451 pr->ps_oppid = curpr->ps_pid; 452 if (pr->ps_pptr != curpr->ps_pptr) 453 proc_reparent(pr, curpr->ps_pptr); 454 455 /* 456 * Set ptrace status. 457 */ 458 if (newptstat != NULL) { 459 pr->ps_ptstat = newptstat; 460 newptstat = NULL; 461 curpr->ps_ptstat->pe_report_event = PTRACE_FORK; 462 pr->ps_ptstat->pe_report_event = PTRACE_FORK; 463 curpr->ps_ptstat->pe_other_pid = pr->ps_pid; 464 pr->ps_ptstat->pe_other_pid = curpr->ps_pid; 465 } 466 } 467 468 /* 469 * For new processes, set accounting bits and mark as complete. 470 */ 471 getnanotime(&pr->ps_start); 472 pr->ps_acflag = AFORK; 473 atomic_clearbits_int(&pr->ps_flags, PS_EMBRYO); 474 475 if ((flags & FORK_IDLE) == 0) 476 fork_thread_start(p, curp, flags); 477 else 478 p->p_cpu = arg; 479 480 free(newptstat, M_SUBPROC, sizeof(*newptstat)); 481 482 /* 483 * Notify any interested parties about the new process. 484 */ 485 KNOTE(&curpr->ps_klist, NOTE_FORK | pr->ps_pid); 486 487 /* 488 * Update stats now that we know the fork was successful. 489 */ 490 uvmexp.forks++; 491 if (flags & FORK_PPWAIT) 492 uvmexp.forks_ppwait++; 493 if (flags & FORK_SHAREVM) 494 uvmexp.forks_sharevm++; 495 496 /* 497 * Pass a pointer to the new process to the caller. 498 */ 499 if (rnewprocp != NULL) 500 *rnewprocp = p; 501 502 /* 503 * Preserve synchronization semantics of vfork. If waiting for 504 * child to exec or exit, set PS_PPWAIT on child and PS_ISPWAIT 505 * on ourselves, and sleep on our process for the latter flag 506 * to go away. 507 * XXX Need to stop other rthreads in the parent 508 */ 509 if (flags & FORK_PPWAIT) 510 while (curpr->ps_flags & PS_ISPWAIT) 511 tsleep(curpr, PWAIT, "ppwait", 0); 512 513 /* 514 * If we're tracing the child, alert the parent too. 515 */ 516 if ((flags & FORK_PTRACE) && (curpr->ps_flags & PS_TRACED)) 517 psignal(curp, SIGTRAP); 518 519 /* 520 * Return child pid to parent process 521 */ 522 if (retval != NULL) { 523 retval[0] = pr->ps_pid; 524 retval[1] = 0; 525 } 526 return (0); 527 } 528 529 int 530 thread_fork(struct proc *curp, void *stack, void *tcb, pid_t *tidptr, 531 register_t *retval) 532 { 533 struct process *pr = curp->p_p; 534 struct proc *p; 535 pid_t tid; 536 vaddr_t uaddr; 537 int error; 538 539 if (stack == NULL) 540 return EINVAL; 541 542 if ((error = fork_check_maxthread(curp->p_ucred->cr_ruid))) 543 return error; 544 545 uaddr = uvm_uarea_alloc(); 546 if (uaddr == 0) { 547 nthreads--; 548 return ENOMEM; 549 } 550 551 /* 552 * From now on, we're committed to the fork and cannot fail. 553 */ 554 p = thread_new(curp, uaddr); 555 atomic_setbits_int(&p->p_flag, P_THREAD); 556 sigstkinit(&p->p_sigstk); 557 558 /* other links */ 559 p->p_p = pr; 560 pr->ps_refcnt++; 561 562 /* local copies */ 563 p->p_fd = pr->ps_fd; 564 p->p_vmspace = pr->ps_vmspace; 565 566 /* 567 * Finish creating the child thread. cpu_fork() will copy 568 * and update the pcb and make the child ready to run. The 569 * child will exit directly to user mode via child_return() 570 * on its first time slice and will not return here. 571 */ 572 cpu_fork(curp, p, stack, tcb, child_return, p); 573 574 p->p_tid = alloctid(); 575 576 LIST_INSERT_HEAD(&allproc, p, p_list); 577 LIST_INSERT_HEAD(TIDHASH(p->p_tid), p, p_hash); 578 TAILQ_INSERT_TAIL(&pr->ps_threads, p, p_thr_link); 579 580 /* 581 * if somebody else wants to take us to single threaded mode, 582 * count ourselves in. 583 */ 584 if (pr->ps_single) { 585 pr->ps_singlecount++; 586 atomic_setbits_int(&p->p_flag, P_SUSPSINGLE); 587 } 588 589 /* 590 * Return tid to parent thread and copy it out to userspace 591 */ 592 retval[0] = tid = p->p_tid + THREAD_PID_OFFSET; 593 retval[1] = 0; 594 if (tidptr != NULL) { 595 if (copyout(&tid, tidptr, sizeof(tid))) 596 psignal(curp, SIGSEGV); 597 } 598 599 fork_thread_start(p, curp, 0); 600 601 /* 602 * Update stats now that we know the fork was successful. 603 */ 604 forkstat.cnttfork++; 605 uvmexp.forks++; 606 uvmexp.forks_sharevm++; 607 608 return 0; 609 } 610 611 612 /* Find an unused tid */ 613 pid_t 614 alloctid(void) 615 { 616 pid_t tid; 617 618 do { 619 /* (0 .. TID_MASK+1] */ 620 tid = 1 + (arc4random() & TID_MASK); 621 } while (tfind(tid) != NULL); 622 623 return (tid); 624 } 625 626 /* 627 * Checks for current use of a pid, either as a pid or pgid. 628 */ 629 pid_t oldpids[128]; 630 int 631 ispidtaken(pid_t pid) 632 { 633 uint32_t i; 634 635 for (i = 0; i < nitems(oldpids); i++) 636 if (pid == oldpids[i]) 637 return (1); 638 639 if (prfind(pid) != NULL) 640 return (1); 641 if (pgfind(pid) != NULL) 642 return (1); 643 if (zombiefind(pid) != NULL) 644 return (1); 645 return (0); 646 } 647 648 /* Find an unused pid */ 649 pid_t 650 allocpid(void) 651 { 652 static pid_t lastpid; 653 pid_t pid; 654 655 if (!randompid) { 656 /* only used early on for system processes */ 657 pid = ++lastpid; 658 } else { 659 /* Find an unused pid satisfying lastpid < pid <= PID_MAX */ 660 do { 661 pid = arc4random_uniform(PID_MAX - lastpid) + 1 + 662 lastpid; 663 } while (ispidtaken(pid)); 664 } 665 666 return pid; 667 } 668 669 void 670 freepid(pid_t pid) 671 { 672 static uint32_t idx; 673 674 oldpids[idx++ % nitems(oldpids)] = pid; 675 } 676 677 #if defined(MULTIPROCESSOR) 678 /* 679 * XXX This is a slight hack to get newly-formed processes to 680 * XXX acquire the kernel lock as soon as they run. 681 */ 682 void 683 proc_trampoline_mp(void) 684 { 685 SCHED_ASSERT_LOCKED(); 686 __mp_unlock(&sched_lock); 687 spl0(); 688 SCHED_ASSERT_UNLOCKED(); 689 KERNEL_ASSERT_UNLOCKED(); 690 691 KERNEL_LOCK(); 692 } 693 #endif 694