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