1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_ktrace.h" 41 #include "opt_mac.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/eventhandler.h> 47 #include <sys/filedesc.h> 48 #include <sys/kernel.h> 49 #include <sys/kthread.h> 50 #include <sys/sysctl.h> 51 #include <sys/lock.h> 52 #include <sys/malloc.h> 53 #include <sys/mutex.h> 54 #include <sys/priv.h> 55 #include <sys/proc.h> 56 #include <sys/pioctl.h> 57 #include <sys/resourcevar.h> 58 #include <sys/sched.h> 59 #include <sys/syscall.h> 60 #include <sys/vmmeter.h> 61 #include <sys/vnode.h> 62 #include <sys/acct.h> 63 #include <sys/ktr.h> 64 #include <sys/ktrace.h> 65 #include <sys/unistd.h> 66 #include <sys/sx.h> 67 #include <sys/signalvar.h> 68 69 #include <security/audit/audit.h> 70 #include <security/mac/mac_framework.h> 71 72 #include <vm/vm.h> 73 #include <vm/pmap.h> 74 #include <vm/vm_map.h> 75 #include <vm/vm_extern.h> 76 #include <vm/uma.h> 77 78 79 #ifndef _SYS_SYSPROTO_H_ 80 struct fork_args { 81 int dummy; 82 }; 83 #endif 84 85 /* ARGSUSED */ 86 int 87 fork(td, uap) 88 struct thread *td; 89 struct fork_args *uap; 90 { 91 int error; 92 struct proc *p2; 93 94 error = fork1(td, RFFDG | RFPROC, 0, &p2); 95 if (error == 0) { 96 td->td_retval[0] = p2->p_pid; 97 td->td_retval[1] = 0; 98 } 99 return (error); 100 } 101 102 /* ARGSUSED */ 103 int 104 vfork(td, uap) 105 struct thread *td; 106 struct vfork_args *uap; 107 { 108 int error; 109 struct proc *p2; 110 111 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 112 if (error == 0) { 113 td->td_retval[0] = p2->p_pid; 114 td->td_retval[1] = 0; 115 } 116 return (error); 117 } 118 119 int 120 rfork(td, uap) 121 struct thread *td; 122 struct rfork_args *uap; 123 { 124 struct proc *p2; 125 int error; 126 127 /* Don't allow kernel-only flags. */ 128 if ((uap->flags & RFKERNELONLY) != 0) 129 return (EINVAL); 130 131 AUDIT_ARG(fflags, uap->flags); 132 error = fork1(td, uap->flags, 0, &p2); 133 if (error == 0) { 134 td->td_retval[0] = p2 ? p2->p_pid : 0; 135 td->td_retval[1] = 0; 136 } 137 return (error); 138 } 139 140 int nprocs = 1; /* process 0 */ 141 int lastpid = 0; 142 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 143 "Last used PID"); 144 145 /* 146 * Random component to lastpid generation. We mix in a random factor to make 147 * it a little harder to predict. We sanity check the modulus value to avoid 148 * doing it in critical paths. Don't let it be too small or we pointlessly 149 * waste randomness entropy, and don't let it be impossibly large. Using a 150 * modulus that is too big causes a LOT more process table scans and slows 151 * down fork processing as the pidchecked caching is defeated. 152 */ 153 static int randompid = 0; 154 155 static int 156 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 157 { 158 int error, pid; 159 160 error = sysctl_wire_old_buffer(req, sizeof(int)); 161 if (error != 0) 162 return(error); 163 sx_xlock(&allproc_lock); 164 pid = randompid; 165 error = sysctl_handle_int(oidp, &pid, 0, req); 166 if (error == 0 && req->newptr != NULL) { 167 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 168 pid = PID_MAX - 100; 169 else if (pid < 2) /* NOP */ 170 pid = 0; 171 else if (pid < 100) /* Make it reasonable */ 172 pid = 100; 173 randompid = pid; 174 } 175 sx_xunlock(&allproc_lock); 176 return (error); 177 } 178 179 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 180 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 181 182 int 183 fork1(td, flags, pages, procp) 184 struct thread *td; 185 int flags; 186 int pages; 187 struct proc **procp; 188 { 189 struct proc *p1, *p2, *pptr; 190 struct proc *newproc; 191 int ok, trypid; 192 static int curfail, pidchecked = 0; 193 static struct timeval lastfail; 194 struct filedesc *fd; 195 struct filedesc_to_leader *fdtol; 196 struct thread *td2; 197 struct sigacts *newsigacts; 198 struct vmspace *vm2; 199 int error; 200 201 /* Can't copy and clear. */ 202 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 203 return (EINVAL); 204 205 p1 = td->td_proc; 206 207 /* 208 * Here we don't create a new process, but we divorce 209 * certain parts of a process from itself. 210 */ 211 if ((flags & RFPROC) == 0) { 212 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 213 (flags & (RFCFDG | RFFDG))) { 214 PROC_LOCK(p1); 215 if (thread_single(SINGLE_BOUNDARY)) { 216 PROC_UNLOCK(p1); 217 return (ERESTART); 218 } 219 PROC_UNLOCK(p1); 220 } 221 222 error = vm_forkproc(td, NULL, NULL, NULL, flags); 223 if (error) 224 goto norfproc_fail; 225 226 /* 227 * Close all file descriptors. 228 */ 229 if (flags & RFCFDG) { 230 struct filedesc *fdtmp; 231 fdtmp = fdinit(td->td_proc->p_fd); 232 fdfree(td); 233 p1->p_fd = fdtmp; 234 } 235 236 /* 237 * Unshare file descriptors (from parent). 238 */ 239 if (flags & RFFDG) 240 fdunshare(p1, td); 241 242 norfproc_fail: 243 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 244 (flags & (RFCFDG | RFFDG))) { 245 PROC_LOCK(p1); 246 thread_single_end(); 247 PROC_UNLOCK(p1); 248 } 249 *procp = NULL; 250 return (error); 251 } 252 253 /* 254 * XXX 255 * We did have single-threading code here 256 * however it proved un-needed and caused problems 257 */ 258 259 /* Allocate new proc. */ 260 newproc = uma_zalloc(proc_zone, M_WAITOK); 261 if (TAILQ_EMPTY(&newproc->p_threads)) { 262 td2 = thread_alloc(); 263 if (td2 == NULL) { 264 error = ENOMEM; 265 goto fail1; 266 } 267 proc_linkup(newproc, td2); 268 sched_newproc(newproc, td2); 269 } else 270 td2 = FIRST_THREAD_IN_PROC(newproc); 271 272 /* Allocate and switch to an alternate kstack if specified. */ 273 if (pages != 0) { 274 if (!vm_thread_new_altkstack(td2, pages)) { 275 error = ENOMEM; 276 goto fail1; 277 } 278 } 279 if ((flags & RFMEM) == 0) { 280 vm2 = vmspace_fork(p1->p_vmspace); 281 if (vm2 == NULL) { 282 error = ENOMEM; 283 goto fail1; 284 } 285 } else 286 vm2 = NULL; 287 #ifdef MAC 288 mac_proc_init(newproc); 289 #endif 290 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 291 STAILQ_INIT(&newproc->p_ktr); 292 293 /* We have to lock the process tree while we look for a pid. */ 294 sx_slock(&proctree_lock); 295 296 /* 297 * Although process entries are dynamically created, we still keep 298 * a global limit on the maximum number we will create. Don't allow 299 * a nonprivileged user to use the last ten processes; don't let root 300 * exceed the limit. The variable nprocs is the current number of 301 * processes, maxproc is the limit. 302 */ 303 sx_xlock(&allproc_lock); 304 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, 305 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { 306 error = EAGAIN; 307 goto fail; 308 } 309 310 /* 311 * Increment the count of procs running with this uid. Don't allow 312 * a nonprivileged user to exceed their current limit. 313 * 314 * XXXRW: Can we avoid privilege here if it's not needed? 315 */ 316 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 317 if (error == 0) 318 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 319 else { 320 PROC_LOCK(p1); 321 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 322 lim_cur(p1, RLIMIT_NPROC)); 323 PROC_UNLOCK(p1); 324 } 325 if (!ok) { 326 error = EAGAIN; 327 goto fail; 328 } 329 330 /* 331 * Increment the nprocs resource before blocking can occur. There 332 * are hard-limits as to the number of processes that can run. 333 */ 334 nprocs++; 335 336 /* 337 * Find an unused process ID. We remember a range of unused IDs 338 * ready to use (from lastpid+1 through pidchecked-1). 339 * 340 * If RFHIGHPID is set (used during system boot), do not allocate 341 * low-numbered pids. 342 */ 343 trypid = lastpid + 1; 344 if (flags & RFHIGHPID) { 345 if (trypid < 10) 346 trypid = 10; 347 } else { 348 if (randompid) 349 trypid += arc4random() % randompid; 350 } 351 retry: 352 /* 353 * If the process ID prototype has wrapped around, 354 * restart somewhat above 0, as the low-numbered procs 355 * tend to include daemons that don't exit. 356 */ 357 if (trypid >= PID_MAX) { 358 trypid = trypid % PID_MAX; 359 if (trypid < 100) 360 trypid += 100; 361 pidchecked = 0; 362 } 363 if (trypid >= pidchecked) { 364 int doingzomb = 0; 365 366 pidchecked = PID_MAX; 367 /* 368 * Scan the active and zombie procs to check whether this pid 369 * is in use. Remember the lowest pid that's greater 370 * than trypid, so we can avoid checking for a while. 371 */ 372 p2 = LIST_FIRST(&allproc); 373 again: 374 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 375 while (p2->p_pid == trypid || 376 (p2->p_pgrp != NULL && 377 (p2->p_pgrp->pg_id == trypid || 378 (p2->p_session != NULL && 379 p2->p_session->s_sid == trypid)))) { 380 trypid++; 381 if (trypid >= pidchecked) 382 goto retry; 383 } 384 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 385 pidchecked = p2->p_pid; 386 if (p2->p_pgrp != NULL) { 387 if (p2->p_pgrp->pg_id > trypid && 388 pidchecked > p2->p_pgrp->pg_id) 389 pidchecked = p2->p_pgrp->pg_id; 390 if (p2->p_session != NULL && 391 p2->p_session->s_sid > trypid && 392 pidchecked > p2->p_session->s_sid) 393 pidchecked = p2->p_session->s_sid; 394 } 395 } 396 if (!doingzomb) { 397 doingzomb = 1; 398 p2 = LIST_FIRST(&zombproc); 399 goto again; 400 } 401 } 402 sx_sunlock(&proctree_lock); 403 404 /* 405 * RFHIGHPID does not mess with the lastpid counter during boot. 406 */ 407 if (flags & RFHIGHPID) 408 pidchecked = 0; 409 else 410 lastpid = trypid; 411 412 p2 = newproc; 413 p2->p_state = PRS_NEW; /* protect against others */ 414 p2->p_pid = trypid; 415 /* 416 * Allow the scheduler to initialize the child. 417 */ 418 thread_lock(td); 419 sched_fork(td, td2); 420 thread_unlock(td); 421 AUDIT_ARG(pid, p2->p_pid); 422 LIST_INSERT_HEAD(&allproc, p2, p_list); 423 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 424 425 PROC_LOCK(p2); 426 PROC_LOCK(p1); 427 428 sx_xunlock(&allproc_lock); 429 430 bcopy(&p1->p_startcopy, &p2->p_startcopy, 431 __rangeof(struct proc, p_startcopy, p_endcopy)); 432 PROC_UNLOCK(p1); 433 434 bzero(&p2->p_startzero, 435 __rangeof(struct proc, p_startzero, p_endzero)); 436 437 p2->p_ucred = crhold(td->td_ucred); 438 PROC_UNLOCK(p2); 439 440 /* 441 * Malloc things while we don't hold any locks. 442 */ 443 if (flags & RFSIGSHARE) 444 newsigacts = NULL; 445 else 446 newsigacts = sigacts_alloc(); 447 448 /* 449 * Copy filedesc. 450 */ 451 if (flags & RFCFDG) { 452 fd = fdinit(p1->p_fd); 453 fdtol = NULL; 454 } else if (flags & RFFDG) { 455 fd = fdcopy(p1->p_fd); 456 fdtol = NULL; 457 } else { 458 fd = fdshare(p1->p_fd); 459 if (p1->p_fdtol == NULL) 460 p1->p_fdtol = 461 filedesc_to_leader_alloc(NULL, 462 NULL, 463 p1->p_leader); 464 if ((flags & RFTHREAD) != 0) { 465 /* 466 * Shared file descriptor table and 467 * shared process leaders. 468 */ 469 fdtol = p1->p_fdtol; 470 FILEDESC_XLOCK(p1->p_fd); 471 fdtol->fdl_refcount++; 472 FILEDESC_XUNLOCK(p1->p_fd); 473 } else { 474 /* 475 * Shared file descriptor table, and 476 * different process leaders 477 */ 478 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 479 p1->p_fd, 480 p2); 481 } 482 } 483 /* 484 * Make a proc table entry for the new process. 485 * Start by zeroing the section of proc that is zero-initialized, 486 * then copy the section that is copied directly from the parent. 487 */ 488 489 PROC_LOCK(p2); 490 PROC_LOCK(p1); 491 492 bzero(&td2->td_startzero, 493 __rangeof(struct thread, td_startzero, td_endzero)); 494 495 bcopy(&td->td_startcopy, &td2->td_startcopy, 496 __rangeof(struct thread, td_startcopy, td_endcopy)); 497 498 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 499 td2->td_sigstk = td->td_sigstk; 500 td2->td_sigmask = td->td_sigmask; 501 td2->td_flags = TDF_INMEM; 502 503 /* 504 * Duplicate sub-structures as needed. 505 * Increase reference counts on shared objects. 506 */ 507 p2->p_flag = P_INMEM; 508 p2->p_swtick = ticks; 509 if (p1->p_flag & P_PROFIL) 510 startprofclock(p2); 511 td2->td_ucred = crhold(p2->p_ucred); 512 pargs_hold(p2->p_args); 513 514 if (flags & RFSIGSHARE) { 515 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 516 } else { 517 sigacts_copy(newsigacts, p1->p_sigacts); 518 p2->p_sigacts = newsigacts; 519 } 520 if (flags & RFLINUXTHPN) 521 p2->p_sigparent = SIGUSR1; 522 else 523 p2->p_sigparent = SIGCHLD; 524 525 p2->p_textvp = p1->p_textvp; 526 p2->p_fd = fd; 527 p2->p_fdtol = fdtol; 528 529 /* 530 * p_limit is copy-on-write. Bump its refcount. 531 */ 532 lim_fork(p1, p2); 533 534 pstats_fork(p1->p_stats, p2->p_stats); 535 536 PROC_UNLOCK(p1); 537 PROC_UNLOCK(p2); 538 539 /* Bump references to the text vnode (for procfs) */ 540 if (p2->p_textvp) 541 vref(p2->p_textvp); 542 543 /* 544 * Set up linkage for kernel based threading. 545 */ 546 if ((flags & RFTHREAD) != 0) { 547 mtx_lock(&ppeers_lock); 548 p2->p_peers = p1->p_peers; 549 p1->p_peers = p2; 550 p2->p_leader = p1->p_leader; 551 mtx_unlock(&ppeers_lock); 552 PROC_LOCK(p1->p_leader); 553 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 554 PROC_UNLOCK(p1->p_leader); 555 /* 556 * The task leader is exiting, so process p1 is 557 * going to be killed shortly. Since p1 obviously 558 * isn't dead yet, we know that the leader is either 559 * sending SIGKILL's to all the processes in this 560 * task or is sleeping waiting for all the peers to 561 * exit. We let p1 complete the fork, but we need 562 * to go ahead and kill the new process p2 since 563 * the task leader may not get a chance to send 564 * SIGKILL to it. We leave it on the list so that 565 * the task leader will wait for this new process 566 * to commit suicide. 567 */ 568 PROC_LOCK(p2); 569 psignal(p2, SIGKILL); 570 PROC_UNLOCK(p2); 571 } else 572 PROC_UNLOCK(p1->p_leader); 573 } else { 574 p2->p_peers = NULL; 575 p2->p_leader = p2; 576 } 577 578 sx_xlock(&proctree_lock); 579 PGRP_LOCK(p1->p_pgrp); 580 PROC_LOCK(p2); 581 PROC_LOCK(p1); 582 583 /* 584 * Preserve some more flags in subprocess. P_PROFIL has already 585 * been preserved. 586 */ 587 p2->p_flag |= p1->p_flag & P_SUGID; 588 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 589 SESS_LOCK(p1->p_session); 590 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 591 p2->p_flag |= P_CONTROLT; 592 SESS_UNLOCK(p1->p_session); 593 if (flags & RFPPWAIT) 594 p2->p_flag |= P_PPWAIT; 595 596 p2->p_pgrp = p1->p_pgrp; 597 LIST_INSERT_AFTER(p1, p2, p_pglist); 598 PGRP_UNLOCK(p1->p_pgrp); 599 LIST_INIT(&p2->p_children); 600 601 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 602 603 #ifdef KTRACE 604 /* 605 * Copy traceflag and tracefile if enabled. 606 */ 607 mtx_lock(&ktrace_mtx); 608 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 609 if (p1->p_traceflag & KTRFAC_INHERIT) { 610 p2->p_traceflag = p1->p_traceflag; 611 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 612 VREF(p2->p_tracevp); 613 KASSERT(p1->p_tracecred != NULL, 614 ("ktrace vnode with no cred")); 615 p2->p_tracecred = crhold(p1->p_tracecred); 616 } 617 } 618 mtx_unlock(&ktrace_mtx); 619 #endif 620 621 /* 622 * If PF_FORK is set, the child process inherits the 623 * procfs ioctl flags from its parent. 624 */ 625 if (p1->p_pfsflags & PF_FORK) { 626 p2->p_stops = p1->p_stops; 627 p2->p_pfsflags = p1->p_pfsflags; 628 } 629 630 /* 631 * This begins the section where we must prevent the parent 632 * from being swapped. 633 */ 634 _PHOLD(p1); 635 PROC_UNLOCK(p1); 636 637 /* 638 * Attach the new process to its parent. 639 * 640 * If RFNOWAIT is set, the newly created process becomes a child 641 * of init. This effectively disassociates the child from the 642 * parent. 643 */ 644 if (flags & RFNOWAIT) 645 pptr = initproc; 646 else 647 pptr = p1; 648 p2->p_pptr = pptr; 649 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 650 sx_xunlock(&proctree_lock); 651 652 /* Inform accounting that we have forked. */ 653 p2->p_acflag = AFORK; 654 PROC_UNLOCK(p2); 655 656 /* 657 * Finish creating the child process. It will return via a different 658 * execution path later. (ie: directly into user mode) 659 */ 660 vm_forkproc(td, p2, td2, vm2, flags); 661 662 if (flags == (RFFDG | RFPROC)) { 663 PCPU_INC(cnt.v_forks); 664 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 665 p2->p_vmspace->vm_ssize); 666 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 667 PCPU_INC(cnt.v_vforks); 668 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 669 p2->p_vmspace->vm_ssize); 670 } else if (p1 == &proc0) { 671 PCPU_INC(cnt.v_kthreads); 672 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 673 p2->p_vmspace->vm_ssize); 674 } else { 675 PCPU_INC(cnt.v_rforks); 676 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 677 p2->p_vmspace->vm_ssize); 678 } 679 680 /* 681 * Both processes are set up, now check if any loadable modules want 682 * to adjust anything. 683 * What if they have an error? XXX 684 */ 685 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 686 687 /* 688 * Set the child start time and mark the process as being complete. 689 */ 690 microuptime(&p2->p_stats->p_start); 691 PROC_SLOCK(p2); 692 p2->p_state = PRS_NORMAL; 693 PROC_SUNLOCK(p2); 694 695 /* 696 * If RFSTOPPED not requested, make child runnable and add to 697 * run queue. 698 */ 699 if ((flags & RFSTOPPED) == 0) { 700 thread_lock(td2); 701 TD_SET_CAN_RUN(td2); 702 sched_add(td2, SRQ_BORING); 703 thread_unlock(td2); 704 } 705 706 /* 707 * Now can be swapped. 708 */ 709 PROC_LOCK(p1); 710 _PRELE(p1); 711 712 /* 713 * Tell any interested parties about the new process. 714 */ 715 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 716 717 PROC_UNLOCK(p1); 718 719 /* 720 * Preserve synchronization semantics of vfork. If waiting for 721 * child to exec or exit, set P_PPWAIT on child, and sleep on our 722 * proc (in case of exit). 723 */ 724 PROC_LOCK(p2); 725 while (p2->p_flag & P_PPWAIT) 726 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 727 PROC_UNLOCK(p2); 728 729 /* 730 * Return child proc pointer to parent. 731 */ 732 *procp = p2; 733 return (0); 734 fail: 735 sx_sunlock(&proctree_lock); 736 if (ppsratecheck(&lastfail, &curfail, 1)) 737 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 738 td->td_ucred->cr_ruid); 739 sx_xunlock(&allproc_lock); 740 #ifdef MAC 741 mac_proc_destroy(newproc); 742 #endif 743 fail1: 744 uma_zfree(proc_zone, newproc); 745 pause("fork", hz / 2); 746 return (error); 747 } 748 749 /* 750 * Handle the return of a child process from fork1(). This function 751 * is called from the MD fork_trampoline() entry point. 752 */ 753 void 754 fork_exit(callout, arg, frame) 755 void (*callout)(void *, struct trapframe *); 756 void *arg; 757 struct trapframe *frame; 758 { 759 struct proc *p; 760 struct thread *td; 761 struct thread *dtd; 762 763 td = curthread; 764 p = td->td_proc; 765 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 766 767 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 768 td, td->td_sched, p->p_pid, td->td_name); 769 770 sched_fork_exit(td); 771 /* 772 * Processes normally resume in mi_switch() after being 773 * cpu_switch()'ed to, but when children start up they arrive here 774 * instead, so we must do much the same things as mi_switch() would. 775 */ 776 if ((dtd = PCPU_GET(deadthread))) { 777 PCPU_SET(deadthread, NULL); 778 thread_stash(dtd); 779 } 780 thread_unlock(td); 781 782 /* 783 * cpu_set_fork_handler intercepts this function call to 784 * have this call a non-return function to stay in kernel mode. 785 * initproc has its own fork handler, but it does return. 786 */ 787 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 788 callout(arg, frame); 789 790 /* 791 * Check if a kernel thread misbehaved and returned from its main 792 * function. 793 */ 794 if (p->p_flag & P_KTHREAD) { 795 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 796 td->td_name, p->p_pid); 797 kproc_exit(0); 798 } 799 mtx_assert(&Giant, MA_NOTOWNED); 800 801 EVENTHANDLER_INVOKE(schedtail, p); 802 } 803 804 /* 805 * Simplified back end of syscall(), used when returning from fork() 806 * directly into user mode. Giant is not held on entry, and must not 807 * be held on return. This function is passed in to fork_exit() as the 808 * first parameter and is called when returning to a new userland process. 809 */ 810 void 811 fork_return(td, frame) 812 struct thread *td; 813 struct trapframe *frame; 814 { 815 816 userret(td, frame); 817 #ifdef KTRACE 818 if (KTRPOINT(td, KTR_SYSRET)) 819 ktrsysret(SYS_fork, 0, 0); 820 #endif 821 mtx_assert(&Giant, MA_NOTOWNED); 822 } 823