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_kdtrace.h" 41 #include "opt_ktrace.h" 42 #include "opt_mac.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/sysproto.h> 47 #include <sys/eventhandler.h> 48 #include <sys/filedesc.h> 49 #include <sys/kernel.h> 50 #include <sys/kthread.h> 51 #include <sys/sysctl.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mutex.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/pioctl.h> 58 #include <sys/resourcevar.h> 59 #include <sys/sched.h> 60 #include <sys/syscall.h> 61 #include <sys/vmmeter.h> 62 #include <sys/vnode.h> 63 #include <sys/acct.h> 64 #include <sys/ktr.h> 65 #include <sys/ktrace.h> 66 #include <sys/unistd.h> 67 #include <sys/sdt.h> 68 #include <sys/sx.h> 69 #include <sys/signalvar.h> 70 71 #include <security/audit/audit.h> 72 #include <security/mac/mac_framework.h> 73 74 #include <vm/vm.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_extern.h> 78 #include <vm/uma.h> 79 80 #ifdef KDTRACE_HOOKS 81 #include <sys/dtrace_bsd.h> 82 dtrace_fork_func_t dtrace_fasttrap_fork; 83 #endif 84 85 SDT_PROVIDER_DECLARE(proc); 86 SDT_PROBE_DEFINE(proc, kernel, , create); 87 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *"); 88 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *"); 89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int"); 90 91 #ifndef _SYS_SYSPROTO_H_ 92 struct fork_args { 93 int dummy; 94 }; 95 #endif 96 97 /* ARGSUSED */ 98 int 99 fork(td, uap) 100 struct thread *td; 101 struct fork_args *uap; 102 { 103 int error; 104 struct proc *p2; 105 106 error = fork1(td, RFFDG | RFPROC, 0, &p2); 107 if (error == 0) { 108 td->td_retval[0] = p2->p_pid; 109 td->td_retval[1] = 0; 110 } 111 return (error); 112 } 113 114 /* ARGSUSED */ 115 int 116 vfork(td, uap) 117 struct thread *td; 118 struct vfork_args *uap; 119 { 120 int error; 121 struct proc *p2; 122 123 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 124 if (error == 0) { 125 td->td_retval[0] = p2->p_pid; 126 td->td_retval[1] = 0; 127 } 128 return (error); 129 } 130 131 int 132 rfork(td, uap) 133 struct thread *td; 134 struct rfork_args *uap; 135 { 136 struct proc *p2; 137 int error; 138 139 /* Don't allow kernel-only flags. */ 140 if ((uap->flags & RFKERNELONLY) != 0) 141 return (EINVAL); 142 143 AUDIT_ARG(fflags, uap->flags); 144 error = fork1(td, uap->flags, 0, &p2); 145 if (error == 0) { 146 td->td_retval[0] = p2 ? p2->p_pid : 0; 147 td->td_retval[1] = 0; 148 } 149 return (error); 150 } 151 152 int nprocs = 1; /* process 0 */ 153 int lastpid = 0; 154 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 155 "Last used PID"); 156 157 /* 158 * Random component to lastpid generation. We mix in a random factor to make 159 * it a little harder to predict. We sanity check the modulus value to avoid 160 * doing it in critical paths. Don't let it be too small or we pointlessly 161 * waste randomness entropy, and don't let it be impossibly large. Using a 162 * modulus that is too big causes a LOT more process table scans and slows 163 * down fork processing as the pidchecked caching is defeated. 164 */ 165 static int randompid = 0; 166 167 static int 168 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 169 { 170 int error, pid; 171 172 error = sysctl_wire_old_buffer(req, sizeof(int)); 173 if (error != 0) 174 return(error); 175 sx_xlock(&allproc_lock); 176 pid = randompid; 177 error = sysctl_handle_int(oidp, &pid, 0, req); 178 if (error == 0 && req->newptr != NULL) { 179 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 180 pid = PID_MAX - 100; 181 else if (pid < 2) /* NOP */ 182 pid = 0; 183 else if (pid < 100) /* Make it reasonable */ 184 pid = 100; 185 randompid = pid; 186 } 187 sx_xunlock(&allproc_lock); 188 return (error); 189 } 190 191 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 192 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 193 194 int 195 fork1(td, flags, pages, procp) 196 struct thread *td; 197 int flags; 198 int pages; 199 struct proc **procp; 200 { 201 struct proc *p1, *p2, *pptr; 202 struct proc *newproc; 203 int ok, trypid; 204 static int curfail, pidchecked = 0; 205 static struct timeval lastfail; 206 struct filedesc *fd; 207 struct filedesc_to_leader *fdtol; 208 struct thread *td2; 209 struct sigacts *newsigacts; 210 struct vmspace *vm2; 211 int error; 212 213 /* Can't copy and clear. */ 214 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 215 return (EINVAL); 216 217 p1 = td->td_proc; 218 219 /* 220 * Here we don't create a new process, but we divorce 221 * certain parts of a process from itself. 222 */ 223 if ((flags & RFPROC) == 0) { 224 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 225 (flags & (RFCFDG | RFFDG))) { 226 PROC_LOCK(p1); 227 if (thread_single(SINGLE_BOUNDARY)) { 228 PROC_UNLOCK(p1); 229 return (ERESTART); 230 } 231 PROC_UNLOCK(p1); 232 } 233 234 error = vm_forkproc(td, NULL, NULL, NULL, flags); 235 if (error) 236 goto norfproc_fail; 237 238 /* 239 * Close all file descriptors. 240 */ 241 if (flags & RFCFDG) { 242 struct filedesc *fdtmp; 243 fdtmp = fdinit(td->td_proc->p_fd); 244 fdfree(td); 245 p1->p_fd = fdtmp; 246 } 247 248 /* 249 * Unshare file descriptors (from parent). 250 */ 251 if (flags & RFFDG) 252 fdunshare(p1, td); 253 254 norfproc_fail: 255 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 256 (flags & (RFCFDG | RFFDG))) { 257 PROC_LOCK(p1); 258 thread_single_end(); 259 PROC_UNLOCK(p1); 260 } 261 *procp = NULL; 262 return (error); 263 } 264 265 /* 266 * XXX 267 * We did have single-threading code here 268 * however it proved un-needed and caused problems 269 */ 270 271 vm2 = NULL; 272 /* Allocate new proc. */ 273 newproc = uma_zalloc(proc_zone, M_WAITOK); 274 if (TAILQ_EMPTY(&newproc->p_threads)) { 275 td2 = thread_alloc(); 276 if (td2 == NULL) { 277 error = ENOMEM; 278 goto fail1; 279 } 280 proc_linkup(newproc, td2); 281 } else 282 td2 = FIRST_THREAD_IN_PROC(newproc); 283 284 /* Allocate and switch to an alternate kstack if specified. */ 285 if (pages != 0) { 286 if (!vm_thread_new_altkstack(td2, pages)) { 287 error = ENOMEM; 288 goto fail1; 289 } 290 } 291 if ((flags & RFMEM) == 0) { 292 vm2 = vmspace_fork(p1->p_vmspace); 293 if (vm2 == NULL) { 294 error = ENOMEM; 295 goto fail1; 296 } 297 } 298 #ifdef MAC 299 mac_proc_init(newproc); 300 #endif 301 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 302 STAILQ_INIT(&newproc->p_ktr); 303 304 /* We have to lock the process tree while we look for a pid. */ 305 sx_slock(&proctree_lock); 306 307 /* 308 * Although process entries are dynamically created, we still keep 309 * a global limit on the maximum number we will create. Don't allow 310 * a nonprivileged user to use the last ten processes; don't let root 311 * exceed the limit. The variable nprocs is the current number of 312 * processes, maxproc is the limit. 313 */ 314 sx_xlock(&allproc_lock); 315 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, 316 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { 317 error = EAGAIN; 318 goto fail; 319 } 320 321 /* 322 * Increment the count of procs running with this uid. Don't allow 323 * a nonprivileged user to exceed their current limit. 324 * 325 * XXXRW: Can we avoid privilege here if it's not needed? 326 */ 327 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 328 if (error == 0) 329 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 330 else { 331 PROC_LOCK(p1); 332 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 333 lim_cur(p1, RLIMIT_NPROC)); 334 PROC_UNLOCK(p1); 335 } 336 if (!ok) { 337 error = EAGAIN; 338 goto fail; 339 } 340 341 /* 342 * Increment the nprocs resource before blocking can occur. There 343 * are hard-limits as to the number of processes that can run. 344 */ 345 nprocs++; 346 347 /* 348 * Find an unused process ID. We remember a range of unused IDs 349 * ready to use (from lastpid+1 through pidchecked-1). 350 * 351 * If RFHIGHPID is set (used during system boot), do not allocate 352 * low-numbered pids. 353 */ 354 trypid = lastpid + 1; 355 if (flags & RFHIGHPID) { 356 if (trypid < 10) 357 trypid = 10; 358 } else { 359 if (randompid) 360 trypid += arc4random() % randompid; 361 } 362 retry: 363 /* 364 * If the process ID prototype has wrapped around, 365 * restart somewhat above 0, as the low-numbered procs 366 * tend to include daemons that don't exit. 367 */ 368 if (trypid >= PID_MAX) { 369 trypid = trypid % PID_MAX; 370 if (trypid < 100) 371 trypid += 100; 372 pidchecked = 0; 373 } 374 if (trypid >= pidchecked) { 375 int doingzomb = 0; 376 377 pidchecked = PID_MAX; 378 /* 379 * Scan the active and zombie procs to check whether this pid 380 * is in use. Remember the lowest pid that's greater 381 * than trypid, so we can avoid checking for a while. 382 */ 383 p2 = LIST_FIRST(&allproc); 384 again: 385 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 386 while (p2->p_pid == trypid || 387 (p2->p_pgrp != NULL && 388 (p2->p_pgrp->pg_id == trypid || 389 (p2->p_session != NULL && 390 p2->p_session->s_sid == trypid)))) { 391 trypid++; 392 if (trypid >= pidchecked) 393 goto retry; 394 } 395 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 396 pidchecked = p2->p_pid; 397 if (p2->p_pgrp != NULL) { 398 if (p2->p_pgrp->pg_id > trypid && 399 pidchecked > p2->p_pgrp->pg_id) 400 pidchecked = p2->p_pgrp->pg_id; 401 if (p2->p_session != NULL && 402 p2->p_session->s_sid > trypid && 403 pidchecked > p2->p_session->s_sid) 404 pidchecked = p2->p_session->s_sid; 405 } 406 } 407 if (!doingzomb) { 408 doingzomb = 1; 409 p2 = LIST_FIRST(&zombproc); 410 goto again; 411 } 412 } 413 sx_sunlock(&proctree_lock); 414 415 /* 416 * RFHIGHPID does not mess with the lastpid counter during boot. 417 */ 418 if (flags & RFHIGHPID) 419 pidchecked = 0; 420 else 421 lastpid = trypid; 422 423 p2 = newproc; 424 p2->p_state = PRS_NEW; /* protect against others */ 425 p2->p_pid = trypid; 426 /* 427 * Allow the scheduler to initialize the child. 428 */ 429 thread_lock(td); 430 sched_fork(td, td2); 431 thread_unlock(td); 432 AUDIT_ARG(pid, p2->p_pid); 433 LIST_INSERT_HEAD(&allproc, p2, p_list); 434 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 435 436 PROC_LOCK(p2); 437 PROC_LOCK(p1); 438 439 sx_xunlock(&allproc_lock); 440 441 bcopy(&p1->p_startcopy, &p2->p_startcopy, 442 __rangeof(struct proc, p_startcopy, p_endcopy)); 443 PROC_UNLOCK(p1); 444 445 bzero(&p2->p_startzero, 446 __rangeof(struct proc, p_startzero, p_endzero)); 447 448 p2->p_ucred = crhold(td->td_ucred); 449 PROC_UNLOCK(p2); 450 451 /* 452 * Malloc things while we don't hold any locks. 453 */ 454 if (flags & RFSIGSHARE) 455 newsigacts = NULL; 456 else 457 newsigacts = sigacts_alloc(); 458 459 /* 460 * Copy filedesc. 461 */ 462 if (flags & RFCFDG) { 463 fd = fdinit(p1->p_fd); 464 fdtol = NULL; 465 } else if (flags & RFFDG) { 466 fd = fdcopy(p1->p_fd); 467 fdtol = NULL; 468 } else { 469 fd = fdshare(p1->p_fd); 470 if (p1->p_fdtol == NULL) 471 p1->p_fdtol = 472 filedesc_to_leader_alloc(NULL, 473 NULL, 474 p1->p_leader); 475 if ((flags & RFTHREAD) != 0) { 476 /* 477 * Shared file descriptor table and 478 * shared process leaders. 479 */ 480 fdtol = p1->p_fdtol; 481 FILEDESC_XLOCK(p1->p_fd); 482 fdtol->fdl_refcount++; 483 FILEDESC_XUNLOCK(p1->p_fd); 484 } else { 485 /* 486 * Shared file descriptor table, and 487 * different process leaders 488 */ 489 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 490 p1->p_fd, 491 p2); 492 } 493 } 494 /* 495 * Make a proc table entry for the new process. 496 * Start by zeroing the section of proc that is zero-initialized, 497 * then copy the section that is copied directly from the parent. 498 */ 499 500 PROC_LOCK(p2); 501 PROC_LOCK(p1); 502 503 bzero(&td2->td_startzero, 504 __rangeof(struct thread, td_startzero, td_endzero)); 505 506 bcopy(&td->td_startcopy, &td2->td_startcopy, 507 __rangeof(struct thread, td_startcopy, td_endcopy)); 508 509 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 510 td2->td_sigstk = td->td_sigstk; 511 td2->td_sigmask = td->td_sigmask; 512 td2->td_flags = TDF_INMEM; 513 514 /* 515 * Duplicate sub-structures as needed. 516 * Increase reference counts on shared objects. 517 */ 518 p2->p_flag = P_INMEM; 519 p2->p_swtick = ticks; 520 if (p1->p_flag & P_PROFIL) 521 startprofclock(p2); 522 td2->td_ucred = crhold(p2->p_ucred); 523 pargs_hold(p2->p_args); 524 525 if (flags & RFSIGSHARE) { 526 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 527 } else { 528 sigacts_copy(newsigacts, p1->p_sigacts); 529 p2->p_sigacts = newsigacts; 530 } 531 if (flags & RFLINUXTHPN) 532 p2->p_sigparent = SIGUSR1; 533 else 534 p2->p_sigparent = SIGCHLD; 535 536 p2->p_textvp = p1->p_textvp; 537 p2->p_fd = fd; 538 p2->p_fdtol = fdtol; 539 540 /* 541 * p_limit is copy-on-write. Bump its refcount. 542 */ 543 lim_fork(p1, p2); 544 545 pstats_fork(p1->p_stats, p2->p_stats); 546 547 PROC_UNLOCK(p1); 548 PROC_UNLOCK(p2); 549 550 /* Bump references to the text vnode (for procfs) */ 551 if (p2->p_textvp) 552 vref(p2->p_textvp); 553 554 /* 555 * Set up linkage for kernel based threading. 556 */ 557 if ((flags & RFTHREAD) != 0) { 558 mtx_lock(&ppeers_lock); 559 p2->p_peers = p1->p_peers; 560 p1->p_peers = p2; 561 p2->p_leader = p1->p_leader; 562 mtx_unlock(&ppeers_lock); 563 PROC_LOCK(p1->p_leader); 564 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 565 PROC_UNLOCK(p1->p_leader); 566 /* 567 * The task leader is exiting, so process p1 is 568 * going to be killed shortly. Since p1 obviously 569 * isn't dead yet, we know that the leader is either 570 * sending SIGKILL's to all the processes in this 571 * task or is sleeping waiting for all the peers to 572 * exit. We let p1 complete the fork, but we need 573 * to go ahead and kill the new process p2 since 574 * the task leader may not get a chance to send 575 * SIGKILL to it. We leave it on the list so that 576 * the task leader will wait for this new process 577 * to commit suicide. 578 */ 579 PROC_LOCK(p2); 580 psignal(p2, SIGKILL); 581 PROC_UNLOCK(p2); 582 } else 583 PROC_UNLOCK(p1->p_leader); 584 } else { 585 p2->p_peers = NULL; 586 p2->p_leader = p2; 587 } 588 589 sx_xlock(&proctree_lock); 590 PGRP_LOCK(p1->p_pgrp); 591 PROC_LOCK(p2); 592 PROC_LOCK(p1); 593 594 /* 595 * Preserve some more flags in subprocess. P_PROFIL has already 596 * been preserved. 597 */ 598 p2->p_flag |= p1->p_flag & P_SUGID; 599 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 600 SESS_LOCK(p1->p_session); 601 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 602 p2->p_flag |= P_CONTROLT; 603 SESS_UNLOCK(p1->p_session); 604 if (flags & RFPPWAIT) 605 p2->p_flag |= P_PPWAIT; 606 607 p2->p_pgrp = p1->p_pgrp; 608 LIST_INSERT_AFTER(p1, p2, p_pglist); 609 PGRP_UNLOCK(p1->p_pgrp); 610 LIST_INIT(&p2->p_children); 611 612 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 613 614 #ifdef KTRACE 615 /* 616 * Copy traceflag and tracefile if enabled. 617 */ 618 mtx_lock(&ktrace_mtx); 619 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 620 if (p1->p_traceflag & KTRFAC_INHERIT) { 621 p2->p_traceflag = p1->p_traceflag; 622 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 623 VREF(p2->p_tracevp); 624 KASSERT(p1->p_tracecred != NULL, 625 ("ktrace vnode with no cred")); 626 p2->p_tracecred = crhold(p1->p_tracecred); 627 } 628 } 629 mtx_unlock(&ktrace_mtx); 630 #endif 631 632 /* 633 * If PF_FORK is set, the child process inherits the 634 * procfs ioctl flags from its parent. 635 */ 636 if (p1->p_pfsflags & PF_FORK) { 637 p2->p_stops = p1->p_stops; 638 p2->p_pfsflags = p1->p_pfsflags; 639 } 640 641 #ifdef KDTRACE_HOOKS 642 /* 643 * Tell the DTrace fasttrap provider about the new process 644 * if it has registered an interest. 645 */ 646 if (dtrace_fasttrap_fork) 647 dtrace_fasttrap_fork(p1, p2); 648 #endif 649 650 /* 651 * This begins the section where we must prevent the parent 652 * from being swapped. 653 */ 654 _PHOLD(p1); 655 PROC_UNLOCK(p1); 656 657 /* 658 * Attach the new process to its parent. 659 * 660 * If RFNOWAIT is set, the newly created process becomes a child 661 * of init. This effectively disassociates the child from the 662 * parent. 663 */ 664 if (flags & RFNOWAIT) 665 pptr = initproc; 666 else 667 pptr = p1; 668 p2->p_pptr = pptr; 669 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 670 sx_xunlock(&proctree_lock); 671 672 /* Inform accounting that we have forked. */ 673 p2->p_acflag = AFORK; 674 PROC_UNLOCK(p2); 675 676 /* 677 * Finish creating the child process. It will return via a different 678 * execution path later. (ie: directly into user mode) 679 */ 680 vm_forkproc(td, p2, td2, vm2, flags); 681 682 if (flags == (RFFDG | RFPROC)) { 683 PCPU_INC(cnt.v_forks); 684 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 685 p2->p_vmspace->vm_ssize); 686 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 687 PCPU_INC(cnt.v_vforks); 688 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 689 p2->p_vmspace->vm_ssize); 690 } else if (p1 == &proc0) { 691 PCPU_INC(cnt.v_kthreads); 692 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 693 p2->p_vmspace->vm_ssize); 694 } else { 695 PCPU_INC(cnt.v_rforks); 696 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 697 p2->p_vmspace->vm_ssize); 698 } 699 700 /* 701 * Both processes are set up, now check if any loadable modules want 702 * to adjust anything. 703 * What if they have an error? XXX 704 */ 705 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 706 707 /* 708 * Set the child start time and mark the process as being complete. 709 */ 710 microuptime(&p2->p_stats->p_start); 711 PROC_SLOCK(p2); 712 p2->p_state = PRS_NORMAL; 713 PROC_SUNLOCK(p2); 714 715 /* 716 * If RFSTOPPED not requested, make child runnable and add to 717 * run queue. 718 */ 719 if ((flags & RFSTOPPED) == 0) { 720 thread_lock(td2); 721 TD_SET_CAN_RUN(td2); 722 sched_add(td2, SRQ_BORING); 723 thread_unlock(td2); 724 } 725 726 /* 727 * Now can be swapped. 728 */ 729 PROC_LOCK(p1); 730 _PRELE(p1); 731 732 /* 733 * Tell any interested parties about the new process. 734 */ 735 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 736 737 PROC_UNLOCK(p1); 738 739 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0); 740 741 /* 742 * Preserve synchronization semantics of vfork. If waiting for 743 * child to exec or exit, set P_PPWAIT on child, and sleep on our 744 * proc (in case of exit). 745 */ 746 PROC_LOCK(p2); 747 while (p2->p_flag & P_PPWAIT) 748 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 749 PROC_UNLOCK(p2); 750 751 /* 752 * Return child proc pointer to parent. 753 */ 754 *procp = p2; 755 return (0); 756 fail: 757 sx_sunlock(&proctree_lock); 758 if (ppsratecheck(&lastfail, &curfail, 1)) 759 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 760 td->td_ucred->cr_ruid); 761 sx_xunlock(&allproc_lock); 762 #ifdef MAC 763 mac_proc_destroy(newproc); 764 #endif 765 fail1: 766 if (vm2 != NULL) 767 vmspace_free(vm2); 768 uma_zfree(proc_zone, newproc); 769 pause("fork", hz / 2); 770 return (error); 771 } 772 773 /* 774 * Handle the return of a child process from fork1(). This function 775 * is called from the MD fork_trampoline() entry point. 776 */ 777 void 778 fork_exit(callout, arg, frame) 779 void (*callout)(void *, struct trapframe *); 780 void *arg; 781 struct trapframe *frame; 782 { 783 struct proc *p; 784 struct thread *td; 785 struct thread *dtd; 786 787 td = curthread; 788 p = td->td_proc; 789 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 790 791 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 792 td, td->td_sched, p->p_pid, td->td_name); 793 794 sched_fork_exit(td); 795 /* 796 * Processes normally resume in mi_switch() after being 797 * cpu_switch()'ed to, but when children start up they arrive here 798 * instead, so we must do much the same things as mi_switch() would. 799 */ 800 if ((dtd = PCPU_GET(deadthread))) { 801 PCPU_SET(deadthread, NULL); 802 thread_stash(dtd); 803 } 804 thread_unlock(td); 805 806 /* 807 * cpu_set_fork_handler intercepts this function call to 808 * have this call a non-return function to stay in kernel mode. 809 * initproc has its own fork handler, but it does return. 810 */ 811 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 812 callout(arg, frame); 813 814 /* 815 * Check if a kernel thread misbehaved and returned from its main 816 * function. 817 */ 818 if (p->p_flag & P_KTHREAD) { 819 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 820 td->td_name, p->p_pid); 821 kproc_exit(0); 822 } 823 mtx_assert(&Giant, MA_NOTOWNED); 824 825 EVENTHANDLER_INVOKE(schedtail, p); 826 } 827 828 /* 829 * Simplified back end of syscall(), used when returning from fork() 830 * directly into user mode. Giant is not held on entry, and must not 831 * be held on return. This function is passed in to fork_exit() as the 832 * first parameter and is called when returning to a new userland process. 833 */ 834 void 835 fork_return(td, frame) 836 struct thread *td; 837 struct trapframe *frame; 838 { 839 840 userret(td, frame); 841 #ifdef KTRACE 842 if (KTRPOINT(td, KTR_SYSRET)) 843 ktrsysret(SYS_fork, 0, 0); 844 #endif 845 mtx_assert(&Giant, MA_NOTOWNED); 846 } 847