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