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 int error; 199 200 /* Can't copy and clear. */ 201 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 202 return (EINVAL); 203 204 p1 = td->td_proc; 205 206 /* 207 * Here we don't create a new process, but we divorce 208 * certain parts of a process from itself. 209 */ 210 if ((flags & RFPROC) == 0) { 211 if ((p1->p_flag & P_HADTHREADS) && 212 (flags & (RFCFDG | RFFDG))) { 213 PROC_LOCK(p1); 214 if (thread_single(SINGLE_BOUNDARY)) { 215 PROC_UNLOCK(p1); 216 return (ERESTART); 217 } 218 PROC_UNLOCK(p1); 219 } 220 221 vm_forkproc(td, NULL, NULL, flags); 222 223 /* 224 * Close all file descriptors. 225 */ 226 if (flags & RFCFDG) { 227 struct filedesc *fdtmp; 228 fdtmp = fdinit(td->td_proc->p_fd); 229 fdfree(td); 230 p1->p_fd = fdtmp; 231 } 232 233 /* 234 * Unshare file descriptors (from parent). 235 */ 236 if (flags & RFFDG) 237 fdunshare(p1, td); 238 239 if ((p1->p_flag & P_HADTHREADS) && 240 (flags & (RFCFDG | RFFDG))) { 241 PROC_LOCK(p1); 242 thread_single_end(); 243 PROC_UNLOCK(p1); 244 } 245 *procp = NULL; 246 return (0); 247 } 248 249 /* 250 * Note 1:1 allows for forking with one thread coming out on the 251 * other side with the expectation that the process is about to 252 * exec. 253 */ 254 if (p1->p_flag & P_HADTHREADS) { 255 /* 256 * Idle the other threads for a second. 257 * Since the user space is copied, it must remain stable. 258 * In addition, all threads (from the user perspective) 259 * need to either be suspended or in the kernel, 260 * where they will try restart in the parent and will 261 * be aborted in the child. 262 */ 263 PROC_LOCK(p1); 264 if (thread_single(SINGLE_NO_EXIT)) { 265 /* Abort. Someone else is single threading before us. */ 266 PROC_UNLOCK(p1); 267 return (ERESTART); 268 } 269 PROC_UNLOCK(p1); 270 /* 271 * All other activity in this process 272 * is now suspended at the user boundary, 273 * (or other safe places if we think of any). 274 */ 275 } 276 277 /* Allocate new proc. */ 278 newproc = uma_zalloc(proc_zone, M_WAITOK); 279 #ifdef MAC 280 mac_init_proc(newproc); 281 #endif 282 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 283 STAILQ_INIT(&newproc->p_ktr); 284 285 /* We have to lock the process tree while we look for a pid. */ 286 sx_slock(&proctree_lock); 287 288 /* 289 * Although process entries are dynamically created, we still keep 290 * a global limit on the maximum number we will create. Don't allow 291 * a nonprivileged user to use the last ten processes; don't let root 292 * exceed the limit. The variable nprocs is the current number of 293 * processes, maxproc is the limit. 294 */ 295 sx_xlock(&allproc_lock); 296 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, 297 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { 298 error = EAGAIN; 299 goto fail; 300 } 301 302 /* 303 * Increment the count of procs running with this uid. Don't allow 304 * a nonprivileged user to exceed their current limit. 305 * 306 * XXXRW: Can we avoid privilege here if it's not needed? 307 */ 308 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 309 if (error == 0) 310 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 311 else { 312 PROC_LOCK(p1); 313 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 314 lim_cur(p1, RLIMIT_NPROC)); 315 PROC_UNLOCK(p1); 316 } 317 if (!ok) { 318 error = EAGAIN; 319 goto fail; 320 } 321 322 /* 323 * Increment the nprocs resource before blocking can occur. There 324 * are hard-limits as to the number of processes that can run. 325 */ 326 nprocs++; 327 328 /* 329 * Find an unused process ID. We remember a range of unused IDs 330 * ready to use (from lastpid+1 through pidchecked-1). 331 * 332 * If RFHIGHPID is set (used during system boot), do not allocate 333 * low-numbered pids. 334 */ 335 trypid = lastpid + 1; 336 if (flags & RFHIGHPID) { 337 if (trypid < 10) 338 trypid = 10; 339 } else { 340 if (randompid) 341 trypid += arc4random() % randompid; 342 } 343 retry: 344 /* 345 * If the process ID prototype has wrapped around, 346 * restart somewhat above 0, as the low-numbered procs 347 * tend to include daemons that don't exit. 348 */ 349 if (trypid >= PID_MAX) { 350 trypid = trypid % PID_MAX; 351 if (trypid < 100) 352 trypid += 100; 353 pidchecked = 0; 354 } 355 if (trypid >= pidchecked) { 356 int doingzomb = 0; 357 358 pidchecked = PID_MAX; 359 /* 360 * Scan the active and zombie procs to check whether this pid 361 * is in use. Remember the lowest pid that's greater 362 * than trypid, so we can avoid checking for a while. 363 */ 364 p2 = LIST_FIRST(&allproc); 365 again: 366 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 367 while (p2->p_pid == trypid || 368 (p2->p_pgrp != NULL && 369 (p2->p_pgrp->pg_id == trypid || 370 (p2->p_session != NULL && 371 p2->p_session->s_sid == trypid)))) { 372 trypid++; 373 if (trypid >= pidchecked) 374 goto retry; 375 } 376 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 377 pidchecked = p2->p_pid; 378 if (p2->p_pgrp != NULL) { 379 if (p2->p_pgrp->pg_id > trypid && 380 pidchecked > p2->p_pgrp->pg_id) 381 pidchecked = p2->p_pgrp->pg_id; 382 if (p2->p_session != NULL && 383 p2->p_session->s_sid > trypid && 384 pidchecked > p2->p_session->s_sid) 385 pidchecked = p2->p_session->s_sid; 386 } 387 } 388 if (!doingzomb) { 389 doingzomb = 1; 390 p2 = LIST_FIRST(&zombproc); 391 goto again; 392 } 393 } 394 sx_sunlock(&proctree_lock); 395 396 /* 397 * RFHIGHPID does not mess with the lastpid counter during boot. 398 */ 399 if (flags & RFHIGHPID) 400 pidchecked = 0; 401 else 402 lastpid = trypid; 403 404 p2 = newproc; 405 td2 = FIRST_THREAD_IN_PROC(newproc); 406 p2->p_state = PRS_NEW; /* protect against others */ 407 p2->p_pid = trypid; 408 /* 409 * Allow the scheduler to initialize the child. 410 */ 411 thread_lock(td); 412 sched_fork(td, td2); 413 thread_unlock(td); 414 AUDIT_ARG(pid, p2->p_pid); 415 LIST_INSERT_HEAD(&allproc, p2, p_list); 416 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 417 418 PROC_LOCK(p2); 419 PROC_LOCK(p1); 420 421 sx_xunlock(&allproc_lock); 422 423 bcopy(&p1->p_startcopy, &p2->p_startcopy, 424 __rangeof(struct proc, p_startcopy, p_endcopy)); 425 PROC_UNLOCK(p1); 426 427 bzero(&p2->p_startzero, 428 __rangeof(struct proc, p_startzero, p_endzero)); 429 430 p2->p_ucred = crhold(td->td_ucred); 431 PROC_UNLOCK(p2); 432 433 /* 434 * Malloc things while we don't hold any locks. 435 */ 436 if (flags & RFSIGSHARE) 437 newsigacts = NULL; 438 else 439 newsigacts = sigacts_alloc(); 440 441 /* 442 * Copy filedesc. 443 */ 444 if (flags & RFCFDG) { 445 fd = fdinit(p1->p_fd); 446 fdtol = NULL; 447 } else if (flags & RFFDG) { 448 fd = fdcopy(p1->p_fd); 449 fdtol = NULL; 450 } else { 451 fd = fdshare(p1->p_fd); 452 if (p1->p_fdtol == NULL) 453 p1->p_fdtol = 454 filedesc_to_leader_alloc(NULL, 455 NULL, 456 p1->p_leader); 457 if ((flags & RFTHREAD) != 0) { 458 /* 459 * Shared file descriptor table and 460 * shared process leaders. 461 */ 462 fdtol = p1->p_fdtol; 463 FILEDESC_XLOCK(p1->p_fd); 464 fdtol->fdl_refcount++; 465 FILEDESC_XUNLOCK(p1->p_fd); 466 } else { 467 /* 468 * Shared file descriptor table, and 469 * different process leaders 470 */ 471 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 472 p1->p_fd, 473 p2); 474 } 475 } 476 /* 477 * Make a proc table entry for the new process. 478 * Start by zeroing the section of proc that is zero-initialized, 479 * then copy the section that is copied directly from the parent. 480 */ 481 /* Allocate and switch to an alternate kstack if specified. */ 482 if (pages != 0) 483 vm_thread_new_altkstack(td2, pages); 484 485 PROC_LOCK(p2); 486 PROC_LOCK(p1); 487 488 bzero(&td2->td_startzero, 489 __rangeof(struct thread, td_startzero, td_endzero)); 490 491 bcopy(&td->td_startcopy, &td2->td_startcopy, 492 __rangeof(struct thread, td_startcopy, td_endcopy)); 493 494 td2->td_sigstk = td->td_sigstk; 495 td2->td_sigmask = td->td_sigmask; 496 497 /* 498 * Duplicate sub-structures as needed. 499 * Increase reference counts on shared objects. 500 */ 501 p2->p_flag = 0; 502 if (p1->p_flag & P_PROFIL) 503 startprofclock(p2); 504 PROC_SLOCK(p2); 505 p2->p_sflag = PS_INMEM; 506 PROC_SUNLOCK(p2); 507 td2->td_ucred = crhold(p2->p_ucred); 508 pargs_hold(p2->p_args); 509 510 if (flags & RFSIGSHARE) { 511 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 512 } else { 513 sigacts_copy(newsigacts, p1->p_sigacts); 514 p2->p_sigacts = newsigacts; 515 } 516 if (flags & RFLINUXTHPN) 517 p2->p_sigparent = SIGUSR1; 518 else 519 p2->p_sigparent = SIGCHLD; 520 521 p2->p_textvp = p1->p_textvp; 522 p2->p_fd = fd; 523 p2->p_fdtol = fdtol; 524 525 /* 526 * p_limit is copy-on-write. Bump its refcount. 527 */ 528 lim_fork(p1, p2); 529 530 pstats_fork(p1->p_stats, p2->p_stats); 531 532 PROC_UNLOCK(p1); 533 PROC_UNLOCK(p2); 534 535 /* Bump references to the text vnode (for procfs) */ 536 if (p2->p_textvp) 537 vref(p2->p_textvp); 538 539 /* 540 * Set up linkage for kernel based threading. 541 */ 542 if ((flags & RFTHREAD) != 0) { 543 mtx_lock(&ppeers_lock); 544 p2->p_peers = p1->p_peers; 545 p1->p_peers = p2; 546 p2->p_leader = p1->p_leader; 547 mtx_unlock(&ppeers_lock); 548 PROC_LOCK(p1->p_leader); 549 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 550 PROC_UNLOCK(p1->p_leader); 551 /* 552 * The task leader is exiting, so process p1 is 553 * going to be killed shortly. Since p1 obviously 554 * isn't dead yet, we know that the leader is either 555 * sending SIGKILL's to all the processes in this 556 * task or is sleeping waiting for all the peers to 557 * exit. We let p1 complete the fork, but we need 558 * to go ahead and kill the new process p2 since 559 * the task leader may not get a chance to send 560 * SIGKILL to it. We leave it on the list so that 561 * the task leader will wait for this new process 562 * to commit suicide. 563 */ 564 PROC_LOCK(p2); 565 psignal(p2, SIGKILL); 566 PROC_UNLOCK(p2); 567 } else 568 PROC_UNLOCK(p1->p_leader); 569 } else { 570 p2->p_peers = NULL; 571 p2->p_leader = p2; 572 } 573 574 sx_xlock(&proctree_lock); 575 PGRP_LOCK(p1->p_pgrp); 576 PROC_LOCK(p2); 577 PROC_LOCK(p1); 578 579 /* 580 * Preserve some more flags in subprocess. P_PROFIL has already 581 * been preserved. 582 */ 583 p2->p_flag |= p1->p_flag & P_SUGID; 584 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 585 SESS_LOCK(p1->p_session); 586 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 587 p2->p_flag |= P_CONTROLT; 588 SESS_UNLOCK(p1->p_session); 589 if (flags & RFPPWAIT) 590 p2->p_flag |= P_PPWAIT; 591 592 p2->p_pgrp = p1->p_pgrp; 593 LIST_INSERT_AFTER(p1, p2, p_pglist); 594 PGRP_UNLOCK(p1->p_pgrp); 595 LIST_INIT(&p2->p_children); 596 597 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 598 599 #ifdef KTRACE 600 /* 601 * Copy traceflag and tracefile if enabled. 602 */ 603 mtx_lock(&ktrace_mtx); 604 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 605 if (p1->p_traceflag & KTRFAC_INHERIT) { 606 p2->p_traceflag = p1->p_traceflag; 607 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 608 VREF(p2->p_tracevp); 609 KASSERT(p1->p_tracecred != NULL, 610 ("ktrace vnode with no cred")); 611 p2->p_tracecred = crhold(p1->p_tracecred); 612 } 613 } 614 mtx_unlock(&ktrace_mtx); 615 #endif 616 617 /* 618 * If PF_FORK is set, the child process inherits the 619 * procfs ioctl flags from its parent. 620 */ 621 if (p1->p_pfsflags & PF_FORK) { 622 p2->p_stops = p1->p_stops; 623 p2->p_pfsflags = p1->p_pfsflags; 624 } 625 626 /* 627 * This begins the section where we must prevent the parent 628 * from being swapped. 629 */ 630 _PHOLD(p1); 631 PROC_UNLOCK(p1); 632 633 /* 634 * Attach the new process to its parent. 635 * 636 * If RFNOWAIT is set, the newly created process becomes a child 637 * of init. This effectively disassociates the child from the 638 * parent. 639 */ 640 if (flags & RFNOWAIT) 641 pptr = initproc; 642 else 643 pptr = p1; 644 p2->p_pptr = pptr; 645 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 646 sx_xunlock(&proctree_lock); 647 648 /* Inform accounting that we have forked. */ 649 p2->p_acflag = AFORK; 650 PROC_UNLOCK(p2); 651 652 /* 653 * Finish creating the child process. It will return via a different 654 * execution path later. (ie: directly into user mode) 655 */ 656 vm_forkproc(td, p2, td2, flags); 657 658 if (flags == (RFFDG | RFPROC)) { 659 PCPU_INC(cnt.v_forks); 660 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 661 p2->p_vmspace->vm_ssize); 662 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 663 PCPU_INC(cnt.v_vforks); 664 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 665 p2->p_vmspace->vm_ssize); 666 } else if (p1 == &proc0) { 667 PCPU_INC(cnt.v_kthreads); 668 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 669 p2->p_vmspace->vm_ssize); 670 } else { 671 PCPU_INC(cnt.v_rforks); 672 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 673 p2->p_vmspace->vm_ssize); 674 } 675 676 /* 677 * Both processes are set up, now check if any loadable modules want 678 * to adjust anything. 679 * What if they have an error? XXX 680 */ 681 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 682 683 /* 684 * Set the child start time and mark the process as being complete. 685 */ 686 microuptime(&p2->p_stats->p_start); 687 PROC_SLOCK(p2); 688 p2->p_state = PRS_NORMAL; 689 PROC_SUNLOCK(p2); 690 691 /* 692 * If RFSTOPPED not requested, make child runnable and add to 693 * run queue. 694 */ 695 if ((flags & RFSTOPPED) == 0) { 696 thread_lock(td2); 697 TD_SET_CAN_RUN(td2); 698 sched_add(td2, SRQ_BORING); 699 thread_unlock(td2); 700 } 701 702 /* 703 * Now can be swapped. 704 */ 705 PROC_LOCK(p1); 706 _PRELE(p1); 707 708 /* 709 * Tell any interested parties about the new process. 710 */ 711 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 712 713 PROC_UNLOCK(p1); 714 715 /* 716 * Preserve synchronization semantics of vfork. If waiting for 717 * child to exec or exit, set P_PPWAIT on child, and sleep on our 718 * proc (in case of exit). 719 */ 720 PROC_LOCK(p2); 721 while (p2->p_flag & P_PPWAIT) 722 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 723 PROC_UNLOCK(p2); 724 725 /* 726 * If other threads are waiting, let them continue now. 727 */ 728 if (p1->p_flag & P_HADTHREADS) { 729 PROC_LOCK(p1); 730 thread_single_end(); 731 PROC_UNLOCK(p1); 732 } 733 734 /* 735 * Return child proc pointer to parent. 736 */ 737 *procp = p2; 738 return (0); 739 fail: 740 sx_sunlock(&proctree_lock); 741 if (ppsratecheck(&lastfail, &curfail, 1)) 742 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 743 td->td_ucred->cr_ruid); 744 sx_xunlock(&allproc_lock); 745 #ifdef MAC 746 mac_destroy_proc(newproc); 747 #endif 748 uma_zfree(proc_zone, newproc); 749 if (p1->p_flag & P_HADTHREADS) { 750 PROC_LOCK(p1); 751 thread_single_end(); 752 PROC_UNLOCK(p1); 753 } 754 pause("fork", hz / 2); 755 return (error); 756 } 757 758 /* 759 * Handle the return of a child process from fork1(). This function 760 * is called from the MD fork_trampoline() entry point. 761 */ 762 void 763 fork_exit(callout, arg, frame) 764 void (*callout)(void *, struct trapframe *); 765 void *arg; 766 struct trapframe *frame; 767 { 768 struct proc *p; 769 struct thread *td; 770 struct thread *dtd; 771 772 td = curthread; 773 p = td->td_proc; 774 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 775 776 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 777 td, td->td_sched, p->p_pid, p->p_comm); 778 779 sched_fork_exit(td); 780 /* 781 * Processes normally resume in mi_switch() after being 782 * cpu_switch()'ed to, but when children start up they arrive here 783 * instead, so we must do much the same things as mi_switch() would. 784 */ 785 if ((dtd = PCPU_GET(deadthread))) { 786 PCPU_SET(deadthread, NULL); 787 thread_stash(dtd); 788 } 789 thread_unlock(td); 790 791 /* 792 * cpu_set_fork_handler intercepts this function call to 793 * have this call a non-return function to stay in kernel mode. 794 * initproc has its own fork handler, but it does return. 795 */ 796 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 797 callout(arg, frame); 798 799 /* 800 * Check if a kernel thread misbehaved and returned from its main 801 * function. 802 */ 803 if (p->p_flag & P_KTHREAD) { 804 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 805 p->p_comm, p->p_pid); 806 kthread_exit(0); 807 } 808 mtx_assert(&Giant, MA_NOTOWNED); 809 810 EVENTHANDLER_INVOKE(schedtail, p); 811 } 812 813 /* 814 * Simplified back end of syscall(), used when returning from fork() 815 * directly into user mode. Giant is not held on entry, and must not 816 * be held on return. This function is passed in to fork_exit() as the 817 * first parameter and is called when returning to a new userland process. 818 */ 819 void 820 fork_return(td, frame) 821 struct thread *td; 822 struct trapframe *frame; 823 { 824 825 userret(td, frame); 826 #ifdef KTRACE 827 if (KTRPOINT(td, KTR_SYSRET)) 828 ktrsysret(SYS_fork, 0, 0); 829 #endif 830 mtx_assert(&Giant, MA_NOTOWNED); 831 } 832