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 /* 86 * MPSAFE 87 */ 88 /* ARGSUSED */ 89 int 90 fork(td, uap) 91 struct thread *td; 92 struct fork_args *uap; 93 { 94 int error; 95 struct proc *p2; 96 97 error = fork1(td, RFFDG | RFPROC, 0, &p2); 98 if (error == 0) { 99 td->td_retval[0] = p2->p_pid; 100 td->td_retval[1] = 0; 101 } 102 return (error); 103 } 104 105 /* 106 * MPSAFE 107 */ 108 /* ARGSUSED */ 109 int 110 vfork(td, uap) 111 struct thread *td; 112 struct vfork_args *uap; 113 { 114 int error; 115 struct proc *p2; 116 117 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 118 if (error == 0) { 119 td->td_retval[0] = p2->p_pid; 120 td->td_retval[1] = 0; 121 } 122 return (error); 123 } 124 125 /* 126 * MPSAFE 127 */ 128 int 129 rfork(td, uap) 130 struct thread *td; 131 struct rfork_args *uap; 132 { 133 struct proc *p2; 134 int error; 135 136 /* Don't allow kernel-only flags. */ 137 if ((uap->flags & RFKERNELONLY) != 0) 138 return (EINVAL); 139 140 AUDIT_ARG(fflags, uap->flags); 141 error = fork1(td, uap->flags, 0, &p2); 142 if (error == 0) { 143 td->td_retval[0] = p2 ? p2->p_pid : 0; 144 td->td_retval[1] = 0; 145 } 146 return (error); 147 } 148 149 int nprocs = 1; /* process 0 */ 150 int lastpid = 0; 151 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 152 "Last used PID"); 153 154 /* 155 * Random component to lastpid generation. We mix in a random factor to make 156 * it a little harder to predict. We sanity check the modulus value to avoid 157 * doing it in critical paths. Don't let it be too small or we pointlessly 158 * waste randomness entropy, and don't let it be impossibly large. Using a 159 * modulus that is too big causes a LOT more process table scans and slows 160 * down fork processing as the pidchecked caching is defeated. 161 */ 162 static int randompid = 0; 163 164 static int 165 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 166 { 167 int error, pid; 168 169 error = sysctl_wire_old_buffer(req, sizeof(int)); 170 if (error != 0) 171 return(error); 172 sx_xlock(&allproc_lock); 173 pid = randompid; 174 error = sysctl_handle_int(oidp, &pid, 0, req); 175 if (error == 0 && req->newptr != NULL) { 176 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 177 pid = PID_MAX - 100; 178 else if (pid < 2) /* NOP */ 179 pid = 0; 180 else if (pid < 100) /* Make it reasonable */ 181 pid = 100; 182 randompid = pid; 183 } 184 sx_xunlock(&allproc_lock); 185 return (error); 186 } 187 188 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 189 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 190 191 int 192 fork1(td, flags, pages, procp) 193 struct thread *td; 194 int flags; 195 int pages; 196 struct proc **procp; 197 { 198 struct proc *p1, *p2, *pptr; 199 struct proc *newproc; 200 int ok, trypid; 201 static int curfail, pidchecked = 0; 202 static struct timeval lastfail; 203 struct filedesc *fd; 204 struct filedesc_to_leader *fdtol; 205 struct thread *td2; 206 struct sigacts *newsigacts; 207 int error; 208 209 /* Can't copy and clear. */ 210 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 211 return (EINVAL); 212 213 p1 = td->td_proc; 214 215 /* 216 * Here we don't create a new process, but we divorce 217 * certain parts of a process from itself. 218 */ 219 if ((flags & RFPROC) == 0) { 220 if ((p1->p_flag & P_HADTHREADS) && 221 (flags & (RFCFDG | RFFDG))) { 222 PROC_LOCK(p1); 223 if (thread_single(SINGLE_BOUNDARY)) { 224 PROC_UNLOCK(p1); 225 return (ERESTART); 226 } 227 PROC_UNLOCK(p1); 228 } 229 230 vm_forkproc(td, NULL, NULL, flags); 231 232 /* 233 * Close all file descriptors. 234 */ 235 if (flags & RFCFDG) { 236 struct filedesc *fdtmp; 237 fdtmp = fdinit(td->td_proc->p_fd); 238 fdfree(td); 239 p1->p_fd = fdtmp; 240 } 241 242 /* 243 * Unshare file descriptors (from parent). 244 */ 245 if (flags & RFFDG) 246 fdunshare(p1, td); 247 248 if ((p1->p_flag & P_HADTHREADS) && 249 (flags & (RFCFDG | RFFDG))) { 250 PROC_LOCK(p1); 251 thread_single_end(); 252 PROC_UNLOCK(p1); 253 } 254 *procp = NULL; 255 return (0); 256 } 257 258 /* 259 * Note 1:1 allows for forking with one thread coming out on the 260 * other side with the expectation that the process is about to 261 * exec. 262 */ 263 if (p1->p_flag & P_HADTHREADS) { 264 /* 265 * Idle the other threads for a second. 266 * Since the user space is copied, it must remain stable. 267 * In addition, all threads (from the user perspective) 268 * need to either be suspended or in the kernel, 269 * where they will try restart in the parent and will 270 * be aborted in the child. 271 */ 272 PROC_LOCK(p1); 273 if (thread_single(SINGLE_NO_EXIT)) { 274 /* Abort. Someone else is single threading before us. */ 275 PROC_UNLOCK(p1); 276 return (ERESTART); 277 } 278 PROC_UNLOCK(p1); 279 /* 280 * All other activity in this process 281 * is now suspended at the user boundary, 282 * (or other safe places if we think of any). 283 */ 284 } 285 286 /* Allocate new proc. */ 287 newproc = uma_zalloc(proc_zone, M_WAITOK); 288 #ifdef MAC 289 mac_init_proc(newproc); 290 #endif 291 #ifdef AUDIT 292 audit_proc_alloc(newproc); 293 #endif 294 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 295 STAILQ_INIT(&newproc->p_ktr); 296 297 /* We have to lock the process tree while we look for a pid. */ 298 sx_slock(&proctree_lock); 299 300 /* 301 * Although process entries are dynamically created, we still keep 302 * a global limit on the maximum number we will create. Don't allow 303 * a nonprivileged user to use the last ten processes; don't let root 304 * exceed the limit. The variable nprocs is the current number of 305 * processes, maxproc is the limit. 306 */ 307 sx_xlock(&allproc_lock); 308 if ((nprocs >= maxproc - 10 && 309 priv_check_cred(td->td_ucred, PRIV_MAXPROC, SUSER_RUID) != 0) || 310 nprocs >= maxproc) { 311 error = EAGAIN; 312 goto fail; 313 } 314 315 /* 316 * Increment the count of procs running with this uid. Don't allow 317 * a nonprivileged user to exceed their current limit. 318 * 319 * XXXRW: Can we avoid privilege here if it's not needed? 320 */ 321 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, SUSER_RUID | 322 SUSER_ALLOWJAIL); 323 if (error == 0) 324 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 325 else { 326 PROC_LOCK(p1); 327 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 328 lim_cur(p1, RLIMIT_NPROC)); 329 PROC_UNLOCK(p1); 330 } 331 if (!ok) { 332 error = EAGAIN; 333 goto fail; 334 } 335 336 /* 337 * Increment the nprocs resource before blocking can occur. There 338 * are hard-limits as to the number of processes that can run. 339 */ 340 nprocs++; 341 342 /* 343 * Find an unused process ID. We remember a range of unused IDs 344 * ready to use (from lastpid+1 through pidchecked-1). 345 * 346 * If RFHIGHPID is set (used during system boot), do not allocate 347 * low-numbered pids. 348 */ 349 trypid = lastpid + 1; 350 if (flags & RFHIGHPID) { 351 if (trypid < 10) 352 trypid = 10; 353 } else { 354 if (randompid) 355 trypid += arc4random() % randompid; 356 } 357 retry: 358 /* 359 * If the process ID prototype has wrapped around, 360 * restart somewhat above 0, as the low-numbered procs 361 * tend to include daemons that don't exit. 362 */ 363 if (trypid >= PID_MAX) { 364 trypid = trypid % PID_MAX; 365 if (trypid < 100) 366 trypid += 100; 367 pidchecked = 0; 368 } 369 if (trypid >= pidchecked) { 370 int doingzomb = 0; 371 372 pidchecked = PID_MAX; 373 /* 374 * Scan the active and zombie procs to check whether this pid 375 * is in use. Remember the lowest pid that's greater 376 * than trypid, so we can avoid checking for a while. 377 */ 378 p2 = LIST_FIRST(&allproc); 379 again: 380 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 381 while (p2->p_pid == trypid || 382 (p2->p_pgrp != NULL && 383 (p2->p_pgrp->pg_id == trypid || 384 (p2->p_session != NULL && 385 p2->p_session->s_sid == trypid)))) { 386 trypid++; 387 if (trypid >= pidchecked) 388 goto retry; 389 } 390 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 391 pidchecked = p2->p_pid; 392 if (p2->p_pgrp != NULL) { 393 if (p2->p_pgrp->pg_id > trypid && 394 pidchecked > p2->p_pgrp->pg_id) 395 pidchecked = p2->p_pgrp->pg_id; 396 if (p2->p_session != NULL && 397 p2->p_session->s_sid > trypid && 398 pidchecked > p2->p_session->s_sid) 399 pidchecked = p2->p_session->s_sid; 400 } 401 } 402 if (!doingzomb) { 403 doingzomb = 1; 404 p2 = LIST_FIRST(&zombproc); 405 goto again; 406 } 407 } 408 sx_sunlock(&proctree_lock); 409 410 /* 411 * RFHIGHPID does not mess with the lastpid counter during boot. 412 */ 413 if (flags & RFHIGHPID) 414 pidchecked = 0; 415 else 416 lastpid = trypid; 417 418 p2 = newproc; 419 p2->p_state = PRS_NEW; /* protect against others */ 420 p2->p_pid = trypid; 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_LOCK_FAST(p1->p_fd); 471 fdtol->fdl_refcount++; 472 FILEDESC_UNLOCK_FAST(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 td2 = FIRST_THREAD_IN_PROC(p2); 489 490 /* Allocate and switch to an alternate kstack if specified. */ 491 if (pages != 0) 492 vm_thread_new_altkstack(td2, pages); 493 494 PROC_LOCK(p2); 495 PROC_LOCK(p1); 496 497 bzero(&td2->td_startzero, 498 __rangeof(struct thread, td_startzero, td_endzero)); 499 500 bcopy(&td->td_startcopy, &td2->td_startcopy, 501 __rangeof(struct thread, td_startcopy, td_endcopy)); 502 503 td2->td_sigstk = td->td_sigstk; 504 td2->td_sigmask = td->td_sigmask; 505 506 /* 507 * Duplicate sub-structures as needed. 508 * Increase reference counts on shared objects. 509 */ 510 p2->p_flag = 0; 511 if (p1->p_flag & P_PROFIL) 512 startprofclock(p2); 513 mtx_lock_spin(&sched_lock); 514 p2->p_sflag = PS_INMEM; 515 /* 516 * Allow the scheduler to adjust the priority of the child and 517 * parent while we hold the sched_lock. 518 */ 519 sched_fork(td, td2); 520 521 mtx_unlock_spin(&sched_lock); 522 td2->td_ucred = crhold(p2->p_ucred); 523 #ifdef AUDIT 524 audit_proc_fork(p1, p2); 525 #endif 526 pargs_hold(p2->p_args); 527 528 if (flags & RFSIGSHARE) { 529 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 530 } else { 531 sigacts_copy(newsigacts, p1->p_sigacts); 532 p2->p_sigacts = newsigacts; 533 } 534 if (flags & RFLINUXTHPN) 535 p2->p_sigparent = SIGUSR1; 536 else 537 p2->p_sigparent = SIGCHLD; 538 539 p2->p_textvp = p1->p_textvp; 540 p2->p_fd = fd; 541 p2->p_fdtol = fdtol; 542 543 /* 544 * p_limit is copy-on-write. Bump its refcount. 545 */ 546 p2->p_limit = lim_hold(p1->p_limit); 547 548 pstats_fork(p1->p_stats, p2->p_stats); 549 550 PROC_UNLOCK(p1); 551 PROC_UNLOCK(p2); 552 553 /* Bump references to the text vnode (for procfs) */ 554 if (p2->p_textvp) 555 vref(p2->p_textvp); 556 557 /* 558 * Set up linkage for kernel based threading. 559 */ 560 if ((flags & RFTHREAD) != 0) { 561 mtx_lock(&ppeers_lock); 562 p2->p_peers = p1->p_peers; 563 p1->p_peers = p2; 564 p2->p_leader = p1->p_leader; 565 mtx_unlock(&ppeers_lock); 566 PROC_LOCK(p1->p_leader); 567 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 568 PROC_UNLOCK(p1->p_leader); 569 /* 570 * The task leader is exiting, so process p1 is 571 * going to be killed shortly. Since p1 obviously 572 * isn't dead yet, we know that the leader is either 573 * sending SIGKILL's to all the processes in this 574 * task or is sleeping waiting for all the peers to 575 * exit. We let p1 complete the fork, but we need 576 * to go ahead and kill the new process p2 since 577 * the task leader may not get a chance to send 578 * SIGKILL to it. We leave it on the list so that 579 * the task leader will wait for this new process 580 * to commit suicide. 581 */ 582 PROC_LOCK(p2); 583 psignal(p2, SIGKILL); 584 PROC_UNLOCK(p2); 585 } else 586 PROC_UNLOCK(p1->p_leader); 587 } else { 588 p2->p_peers = NULL; 589 p2->p_leader = p2; 590 } 591 592 sx_xlock(&proctree_lock); 593 PGRP_LOCK(p1->p_pgrp); 594 PROC_LOCK(p2); 595 PROC_LOCK(p1); 596 597 /* 598 * Preserve some more flags in subprocess. P_PROFIL has already 599 * been preserved. 600 */ 601 p2->p_flag |= p1->p_flag & P_SUGID; 602 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 603 SESS_LOCK(p1->p_session); 604 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 605 p2->p_flag |= P_CONTROLT; 606 SESS_UNLOCK(p1->p_session); 607 if (flags & RFPPWAIT) 608 p2->p_flag |= P_PPWAIT; 609 610 p2->p_pgrp = p1->p_pgrp; 611 LIST_INSERT_AFTER(p1, p2, p_pglist); 612 PGRP_UNLOCK(p1->p_pgrp); 613 LIST_INIT(&p2->p_children); 614 615 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 616 617 #ifdef KTRACE 618 /* 619 * Copy traceflag and tracefile if enabled. 620 */ 621 mtx_lock(&ktrace_mtx); 622 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 623 if (p1->p_traceflag & KTRFAC_INHERIT) { 624 p2->p_traceflag = p1->p_traceflag; 625 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 626 VREF(p2->p_tracevp); 627 KASSERT(p1->p_tracecred != NULL, 628 ("ktrace vnode with no cred")); 629 p2->p_tracecred = crhold(p1->p_tracecred); 630 } 631 } 632 mtx_unlock(&ktrace_mtx); 633 #endif 634 635 /* 636 * If PF_FORK is set, the child process inherits the 637 * procfs ioctl flags from its parent. 638 */ 639 if (p1->p_pfsflags & PF_FORK) { 640 p2->p_stops = p1->p_stops; 641 p2->p_pfsflags = p1->p_pfsflags; 642 } 643 644 /* 645 * This begins the section where we must prevent the parent 646 * from being swapped. 647 */ 648 _PHOLD(p1); 649 PROC_UNLOCK(p1); 650 651 /* 652 * Attach the new process to its parent. 653 * 654 * If RFNOWAIT is set, the newly created process becomes a child 655 * of init. This effectively disassociates the child from the 656 * parent. 657 */ 658 if (flags & RFNOWAIT) 659 pptr = initproc; 660 else 661 pptr = p1; 662 p2->p_pptr = pptr; 663 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 664 sx_xunlock(&proctree_lock); 665 666 /* Inform accounting that we have forked. */ 667 p2->p_acflag = AFORK; 668 PROC_UNLOCK(p2); 669 670 /* 671 * Finish creating the child process. It will return via a different 672 * execution path later. (ie: directly into user mode) 673 */ 674 vm_forkproc(td, p2, td2, flags); 675 676 if (flags == (RFFDG | RFPROC)) { 677 atomic_add_int(&cnt.v_forks, 1); 678 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize + 679 p2->p_vmspace->vm_ssize); 680 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 681 atomic_add_int(&cnt.v_vforks, 1); 682 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 683 p2->p_vmspace->vm_ssize); 684 } else if (p1 == &proc0) { 685 atomic_add_int(&cnt.v_kthreads, 1); 686 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 687 p2->p_vmspace->vm_ssize); 688 } else { 689 atomic_add_int(&cnt.v_rforks, 1); 690 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 691 p2->p_vmspace->vm_ssize); 692 } 693 694 /* 695 * Both processes are set up, now check if any loadable modules want 696 * to adjust anything. 697 * What if they have an error? XXX 698 */ 699 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 700 701 /* 702 * Set the child start time and mark the process as being complete. 703 */ 704 microuptime(&p2->p_stats->p_start); 705 mtx_lock_spin(&sched_lock); 706 p2->p_state = PRS_NORMAL; 707 708 /* 709 * If RFSTOPPED not requested, make child runnable and add to 710 * run queue. 711 */ 712 if ((flags & RFSTOPPED) == 0) { 713 TD_SET_CAN_RUN(td2); 714 sched_add(td2, SRQ_BORING); 715 } 716 mtx_unlock_spin(&sched_lock); 717 718 /* 719 * Now can be swapped. 720 */ 721 PROC_LOCK(p1); 722 _PRELE(p1); 723 724 /* 725 * Tell any interested parties about the new process. 726 */ 727 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 728 729 PROC_UNLOCK(p1); 730 731 /* 732 * Preserve synchronization semantics of vfork. If waiting for 733 * child to exec or exit, set P_PPWAIT on child, and sleep on our 734 * proc (in case of exit). 735 */ 736 PROC_LOCK(p2); 737 while (p2->p_flag & P_PPWAIT) 738 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 739 PROC_UNLOCK(p2); 740 741 /* 742 * If other threads are waiting, let them continue now. 743 */ 744 if (p1->p_flag & P_HADTHREADS) { 745 PROC_LOCK(p1); 746 thread_single_end(); 747 PROC_UNLOCK(p1); 748 } 749 750 /* 751 * Return child proc pointer to parent. 752 */ 753 *procp = p2; 754 return (0); 755 fail: 756 sx_sunlock(&proctree_lock); 757 if (ppsratecheck(&lastfail, &curfail, 1)) 758 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 759 td->td_ucred->cr_ruid); 760 sx_xunlock(&allproc_lock); 761 #ifdef MAC 762 mac_destroy_proc(newproc); 763 #endif 764 #ifdef AUDIT 765 audit_proc_free(newproc); 766 #endif 767 uma_zfree(proc_zone, newproc); 768 if (p1->p_flag & P_HADTHREADS) { 769 PROC_LOCK(p1); 770 thread_single_end(); 771 PROC_UNLOCK(p1); 772 } 773 pause("fork", hz / 2); 774 return (error); 775 } 776 777 /* 778 * Handle the return of a child process from fork1(). This function 779 * is called from the MD fork_trampoline() entry point. 780 */ 781 void 782 fork_exit(callout, arg, frame) 783 void (*callout)(void *, struct trapframe *); 784 void *arg; 785 struct trapframe *frame; 786 { 787 struct proc *p; 788 struct thread *td; 789 790 /* 791 * Finish setting up thread glue so that it begins execution in a 792 * non-nested critical section with sched_lock held but not recursed. 793 */ 794 td = curthread; 795 p = td->td_proc; 796 td->td_oncpu = PCPU_GET(cpuid); 797 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 798 799 sched_lock.mtx_lock = (uintptr_t)td; 800 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 801 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 802 td, td->td_sched, p->p_pid, p->p_comm); 803 804 /* 805 * Processes normally resume in mi_switch() after being 806 * cpu_switch()'ed to, but when children start up they arrive here 807 * instead, so we must do much the same things as mi_switch() would. 808 */ 809 810 if ((td = PCPU_GET(deadthread))) { 811 PCPU_SET(deadthread, NULL); 812 thread_stash(td); 813 } 814 td = curthread; 815 mtx_unlock_spin(&sched_lock); 816 817 /* 818 * cpu_set_fork_handler intercepts this function call to 819 * have this call a non-return function to stay in kernel mode. 820 * initproc has its own fork handler, but it does return. 821 */ 822 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 823 callout(arg, frame); 824 825 /* 826 * Check if a kernel thread misbehaved and returned from its main 827 * function. 828 */ 829 if (p->p_flag & P_KTHREAD) { 830 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 831 p->p_comm, p->p_pid); 832 kthread_exit(0); 833 } 834 mtx_assert(&Giant, MA_NOTOWNED); 835 836 EVENTHANDLER_INVOKE(schedtail, p); 837 } 838 839 /* 840 * Simplified back end of syscall(), used when returning from fork() 841 * directly into user mode. Giant is not held on entry, and must not 842 * be held on return. This function is passed in to fork_exit() as the 843 * first parameter and is called when returning to a new userland process. 844 */ 845 void 846 fork_return(td, frame) 847 struct thread *td; 848 struct trapframe *frame; 849 { 850 851 userret(td, frame); 852 #ifdef KTRACE 853 if (KTRPOINT(td, KTR_SYSRET)) 854 ktrsysret(SYS_fork, 0, 0); 855 #endif 856 mtx_assert(&Giant, MA_NOTOWNED); 857 } 858