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/jail.h> 58 #include <sys/pioctl.h> 59 #include <sys/resourcevar.h> 60 #include <sys/sched.h> 61 #include <sys/syscall.h> 62 #include <sys/vmmeter.h> 63 #include <sys/vnode.h> 64 #include <sys/acct.h> 65 #include <sys/ktr.h> 66 #include <sys/ktrace.h> 67 #include <sys/unistd.h> 68 #include <sys/sdt.h> 69 #include <sys/sx.h> 70 #include <sys/signalvar.h> 71 72 #include <security/audit/audit.h> 73 #include <security/mac/mac_framework.h> 74 75 #include <vm/vm.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_extern.h> 79 #include <vm/uma.h> 80 81 #ifdef KDTRACE_HOOKS 82 #include <sys/dtrace_bsd.h> 83 dtrace_fork_func_t dtrace_fasttrap_fork; 84 #endif 85 86 SDT_PROVIDER_DECLARE(proc); 87 SDT_PROBE_DEFINE(proc, kernel, , create); 88 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *"); 89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *"); 90 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int"); 91 92 #ifndef _SYS_SYSPROTO_H_ 93 struct fork_args { 94 int dummy; 95 }; 96 #endif 97 98 /* ARGSUSED */ 99 int 100 fork(td, uap) 101 struct thread *td; 102 struct fork_args *uap; 103 { 104 int error; 105 struct proc *p2; 106 107 error = fork1(td, RFFDG | RFPROC, 0, &p2); 108 if (error == 0) { 109 td->td_retval[0] = p2->p_pid; 110 td->td_retval[1] = 0; 111 } 112 return (error); 113 } 114 115 /* ARGSUSED */ 116 int 117 vfork(td, uap) 118 struct thread *td; 119 struct vfork_args *uap; 120 { 121 int error, flags; 122 struct proc *p2; 123 124 #ifdef XEN 125 flags = RFFDG | RFPROC; /* validate that this is still an issue */ 126 #else 127 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 128 #endif 129 error = fork1(td, flags, 0, &p2); 130 if (error == 0) { 131 td->td_retval[0] = p2->p_pid; 132 td->td_retval[1] = 0; 133 } 134 return (error); 135 } 136 137 int 138 rfork(td, uap) 139 struct thread *td; 140 struct rfork_args *uap; 141 { 142 struct proc *p2; 143 int error; 144 145 /* Don't allow kernel-only flags. */ 146 if ((uap->flags & RFKERNELONLY) != 0) 147 return (EINVAL); 148 149 AUDIT_ARG(fflags, uap->flags); 150 error = fork1(td, uap->flags, 0, &p2); 151 if (error == 0) { 152 td->td_retval[0] = p2 ? p2->p_pid : 0; 153 td->td_retval[1] = 0; 154 } 155 return (error); 156 } 157 158 int nprocs = 1; /* process 0 */ 159 int lastpid = 0; 160 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 161 "Last used PID"); 162 163 /* 164 * Random component to lastpid generation. We mix in a random factor to make 165 * it a little harder to predict. We sanity check the modulus value to avoid 166 * doing it in critical paths. Don't let it be too small or we pointlessly 167 * waste randomness entropy, and don't let it be impossibly large. Using a 168 * modulus that is too big causes a LOT more process table scans and slows 169 * down fork processing as the pidchecked caching is defeated. 170 */ 171 static int randompid = 0; 172 173 static int 174 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 175 { 176 int error, pid; 177 178 error = sysctl_wire_old_buffer(req, sizeof(int)); 179 if (error != 0) 180 return(error); 181 sx_xlock(&allproc_lock); 182 pid = randompid; 183 error = sysctl_handle_int(oidp, &pid, 0, req); 184 if (error == 0 && req->newptr != NULL) { 185 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 186 pid = PID_MAX - 100; 187 else if (pid < 2) /* NOP */ 188 pid = 0; 189 else if (pid < 100) /* Make it reasonable */ 190 pid = 100; 191 randompid = pid; 192 } 193 sx_xunlock(&allproc_lock); 194 return (error); 195 } 196 197 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 198 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 199 200 int 201 fork1(td, flags, pages, procp) 202 struct thread *td; 203 int flags; 204 int pages; 205 struct proc **procp; 206 { 207 struct proc *p1, *p2, *pptr; 208 struct proc *newproc; 209 int ok, trypid; 210 static int curfail, pidchecked = 0; 211 static struct timeval lastfail; 212 struct filedesc *fd; 213 struct filedesc_to_leader *fdtol; 214 struct thread *td2; 215 struct sigacts *newsigacts; 216 struct vmspace *vm2; 217 int error; 218 219 /* Can't copy and clear. */ 220 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 221 return (EINVAL); 222 223 p1 = td->td_proc; 224 225 /* 226 * Here we don't create a new process, but we divorce 227 * certain parts of a process from itself. 228 */ 229 if ((flags & RFPROC) == 0) { 230 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 231 (flags & (RFCFDG | RFFDG))) { 232 PROC_LOCK(p1); 233 if (thread_single(SINGLE_BOUNDARY)) { 234 PROC_UNLOCK(p1); 235 return (ERESTART); 236 } 237 PROC_UNLOCK(p1); 238 } 239 240 error = vm_forkproc(td, NULL, NULL, NULL, flags); 241 if (error) 242 goto norfproc_fail; 243 244 /* 245 * Close all file descriptors. 246 */ 247 if (flags & RFCFDG) { 248 struct filedesc *fdtmp; 249 fdtmp = fdinit(td->td_proc->p_fd); 250 fdfree(td); 251 p1->p_fd = fdtmp; 252 } 253 254 /* 255 * Unshare file descriptors (from parent). 256 */ 257 if (flags & RFFDG) 258 fdunshare(p1, td); 259 260 norfproc_fail: 261 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 262 (flags & (RFCFDG | RFFDG))) { 263 PROC_LOCK(p1); 264 thread_single_end(); 265 PROC_UNLOCK(p1); 266 } 267 *procp = NULL; 268 return (error); 269 } 270 271 /* 272 * XXX 273 * We did have single-threading code here 274 * however it proved un-needed and caused problems 275 */ 276 277 vm2 = NULL; 278 /* Allocate new proc. */ 279 newproc = uma_zalloc(proc_zone, M_WAITOK); 280 if (TAILQ_EMPTY(&newproc->p_threads)) { 281 td2 = thread_alloc(); 282 if (td2 == NULL) { 283 error = ENOMEM; 284 goto fail1; 285 } 286 proc_linkup(newproc, td2); 287 } else 288 td2 = FIRST_THREAD_IN_PROC(newproc); 289 290 /* Allocate and switch to an alternate kstack if specified. */ 291 if (pages != 0) { 292 if (!vm_thread_new_altkstack(td2, pages)) { 293 error = ENOMEM; 294 goto fail1; 295 } 296 } 297 if ((flags & RFMEM) == 0) { 298 vm2 = vmspace_fork(p1->p_vmspace); 299 if (vm2 == NULL) { 300 error = ENOMEM; 301 goto fail1; 302 } 303 } 304 #ifdef MAC 305 mac_proc_init(newproc); 306 #endif 307 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 308 STAILQ_INIT(&newproc->p_ktr); 309 310 /* We have to lock the process tree while we look for a pid. */ 311 sx_slock(&proctree_lock); 312 313 /* 314 * Although process entries are dynamically created, we still keep 315 * a global limit on the maximum number we will create. Don't allow 316 * a nonprivileged user to use the last ten processes; don't let root 317 * exceed the limit. The variable nprocs is the current number of 318 * processes, maxproc is the limit. 319 */ 320 sx_xlock(&allproc_lock); 321 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, 322 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { 323 error = EAGAIN; 324 goto fail; 325 } 326 327 /* 328 * Increment the count of procs running with this uid. Don't allow 329 * a nonprivileged user to exceed their current limit. 330 * 331 * XXXRW: Can we avoid privilege here if it's not needed? 332 */ 333 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 334 if (error == 0) 335 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 336 else { 337 PROC_LOCK(p1); 338 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 339 lim_cur(p1, RLIMIT_NPROC)); 340 PROC_UNLOCK(p1); 341 } 342 if (!ok) { 343 error = EAGAIN; 344 goto fail; 345 } 346 347 /* 348 * Increment the nprocs resource before blocking can occur. There 349 * are hard-limits as to the number of processes that can run. 350 */ 351 nprocs++; 352 353 /* 354 * Find an unused process ID. We remember a range of unused IDs 355 * ready to use (from lastpid+1 through pidchecked-1). 356 * 357 * If RFHIGHPID is set (used during system boot), do not allocate 358 * low-numbered pids. 359 */ 360 trypid = lastpid + 1; 361 if (flags & RFHIGHPID) { 362 if (trypid < 10) 363 trypid = 10; 364 } else { 365 if (randompid) 366 trypid += arc4random() % randompid; 367 } 368 retry: 369 /* 370 * If the process ID prototype has wrapped around, 371 * restart somewhat above 0, as the low-numbered procs 372 * tend to include daemons that don't exit. 373 */ 374 if (trypid >= PID_MAX) { 375 trypid = trypid % PID_MAX; 376 if (trypid < 100) 377 trypid += 100; 378 pidchecked = 0; 379 } 380 if (trypid >= pidchecked) { 381 int doingzomb = 0; 382 383 pidchecked = PID_MAX; 384 /* 385 * Scan the active and zombie procs to check whether this pid 386 * is in use. Remember the lowest pid that's greater 387 * than trypid, so we can avoid checking for a while. 388 */ 389 p2 = LIST_FIRST(&allproc); 390 again: 391 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 392 while (p2->p_pid == trypid || 393 (p2->p_pgrp != NULL && 394 (p2->p_pgrp->pg_id == trypid || 395 (p2->p_session != NULL && 396 p2->p_session->s_sid == trypid)))) { 397 trypid++; 398 if (trypid >= pidchecked) 399 goto retry; 400 } 401 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 402 pidchecked = p2->p_pid; 403 if (p2->p_pgrp != NULL) { 404 if (p2->p_pgrp->pg_id > trypid && 405 pidchecked > p2->p_pgrp->pg_id) 406 pidchecked = p2->p_pgrp->pg_id; 407 if (p2->p_session != NULL && 408 p2->p_session->s_sid > trypid && 409 pidchecked > p2->p_session->s_sid) 410 pidchecked = p2->p_session->s_sid; 411 } 412 } 413 if (!doingzomb) { 414 doingzomb = 1; 415 p2 = LIST_FIRST(&zombproc); 416 goto again; 417 } 418 } 419 sx_sunlock(&proctree_lock); 420 421 /* 422 * RFHIGHPID does not mess with the lastpid counter during boot. 423 */ 424 if (flags & RFHIGHPID) 425 pidchecked = 0; 426 else 427 lastpid = trypid; 428 429 p2 = newproc; 430 p2->p_state = PRS_NEW; /* protect against others */ 431 p2->p_pid = trypid; 432 /* 433 * Allow the scheduler to initialize the child. 434 */ 435 thread_lock(td); 436 sched_fork(td, td2); 437 thread_unlock(td); 438 AUDIT_ARG(pid, p2->p_pid); 439 LIST_INSERT_HEAD(&allproc, p2, p_list); 440 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 441 442 PROC_LOCK(p2); 443 PROC_LOCK(p1); 444 445 sx_xunlock(&allproc_lock); 446 447 bcopy(&p1->p_startcopy, &p2->p_startcopy, 448 __rangeof(struct proc, p_startcopy, p_endcopy)); 449 pargs_hold(p2->p_args); 450 PROC_UNLOCK(p1); 451 452 bzero(&p2->p_startzero, 453 __rangeof(struct proc, p_startzero, p_endzero)); 454 455 p2->p_ucred = crhold(td->td_ucred); 456 457 /* In case we are jailed tell the prison that we exist. */ 458 if (jailed(p2->p_ucred)) 459 prison_proc_hold(p2->p_ucred->cr_prison); 460 461 PROC_UNLOCK(p2); 462 463 /* 464 * Malloc things while we don't hold any locks. 465 */ 466 if (flags & RFSIGSHARE) 467 newsigacts = NULL; 468 else 469 newsigacts = sigacts_alloc(); 470 471 /* 472 * Copy filedesc. 473 */ 474 if (flags & RFCFDG) { 475 fd = fdinit(p1->p_fd); 476 fdtol = NULL; 477 } else if (flags & RFFDG) { 478 fd = fdcopy(p1->p_fd); 479 fdtol = NULL; 480 } else { 481 fd = fdshare(p1->p_fd); 482 if (p1->p_fdtol == NULL) 483 p1->p_fdtol = 484 filedesc_to_leader_alloc(NULL, 485 NULL, 486 p1->p_leader); 487 if ((flags & RFTHREAD) != 0) { 488 /* 489 * Shared file descriptor table and 490 * shared process leaders. 491 */ 492 fdtol = p1->p_fdtol; 493 FILEDESC_XLOCK(p1->p_fd); 494 fdtol->fdl_refcount++; 495 FILEDESC_XUNLOCK(p1->p_fd); 496 } else { 497 /* 498 * Shared file descriptor table, and 499 * different process leaders 500 */ 501 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 502 p1->p_fd, 503 p2); 504 } 505 } 506 /* 507 * Make a proc table entry for the new process. 508 * Start by zeroing the section of proc that is zero-initialized, 509 * then copy the section that is copied directly from the parent. 510 */ 511 512 PROC_LOCK(p2); 513 PROC_LOCK(p1); 514 515 bzero(&td2->td_startzero, 516 __rangeof(struct thread, td_startzero, td_endzero)); 517 518 bcopy(&td->td_startcopy, &td2->td_startcopy, 519 __rangeof(struct thread, td_startcopy, td_endcopy)); 520 521 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 522 td2->td_sigstk = td->td_sigstk; 523 td2->td_sigmask = td->td_sigmask; 524 td2->td_flags = TDF_INMEM; 525 526 /* 527 * Duplicate sub-structures as needed. 528 * Increase reference counts on shared objects. 529 */ 530 p2->p_flag = P_INMEM; 531 p2->p_swtick = ticks; 532 if (p1->p_flag & P_PROFIL) 533 startprofclock(p2); 534 td2->td_ucred = crhold(p2->p_ucred); 535 536 if (flags & RFSIGSHARE) { 537 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 538 } else { 539 sigacts_copy(newsigacts, p1->p_sigacts); 540 p2->p_sigacts = newsigacts; 541 } 542 if (flags & RFLINUXTHPN) 543 p2->p_sigparent = SIGUSR1; 544 else 545 p2->p_sigparent = SIGCHLD; 546 547 p2->p_textvp = p1->p_textvp; 548 p2->p_fd = fd; 549 p2->p_fdtol = fdtol; 550 551 /* 552 * p_limit is copy-on-write. Bump its refcount. 553 */ 554 lim_fork(p1, p2); 555 556 pstats_fork(p1->p_stats, p2->p_stats); 557 558 PROC_UNLOCK(p1); 559 PROC_UNLOCK(p2); 560 561 /* Bump references to the text vnode (for procfs) */ 562 if (p2->p_textvp) 563 vref(p2->p_textvp); 564 565 /* 566 * Set up linkage for kernel based threading. 567 */ 568 if ((flags & RFTHREAD) != 0) { 569 mtx_lock(&ppeers_lock); 570 p2->p_peers = p1->p_peers; 571 p1->p_peers = p2; 572 p2->p_leader = p1->p_leader; 573 mtx_unlock(&ppeers_lock); 574 PROC_LOCK(p1->p_leader); 575 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 576 PROC_UNLOCK(p1->p_leader); 577 /* 578 * The task leader is exiting, so process p1 is 579 * going to be killed shortly. Since p1 obviously 580 * isn't dead yet, we know that the leader is either 581 * sending SIGKILL's to all the processes in this 582 * task or is sleeping waiting for all the peers to 583 * exit. We let p1 complete the fork, but we need 584 * to go ahead and kill the new process p2 since 585 * the task leader may not get a chance to send 586 * SIGKILL to it. We leave it on the list so that 587 * the task leader will wait for this new process 588 * to commit suicide. 589 */ 590 PROC_LOCK(p2); 591 psignal(p2, SIGKILL); 592 PROC_UNLOCK(p2); 593 } else 594 PROC_UNLOCK(p1->p_leader); 595 } else { 596 p2->p_peers = NULL; 597 p2->p_leader = p2; 598 } 599 600 sx_xlock(&proctree_lock); 601 PGRP_LOCK(p1->p_pgrp); 602 PROC_LOCK(p2); 603 PROC_LOCK(p1); 604 605 /* 606 * Preserve some more flags in subprocess. P_PROFIL has already 607 * been preserved. 608 */ 609 p2->p_flag |= p1->p_flag & P_SUGID; 610 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 611 SESS_LOCK(p1->p_session); 612 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 613 p2->p_flag |= P_CONTROLT; 614 SESS_UNLOCK(p1->p_session); 615 if (flags & RFPPWAIT) 616 p2->p_flag |= P_PPWAIT; 617 618 p2->p_pgrp = p1->p_pgrp; 619 LIST_INSERT_AFTER(p1, p2, p_pglist); 620 PGRP_UNLOCK(p1->p_pgrp); 621 LIST_INIT(&p2->p_children); 622 623 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 624 625 #ifdef KTRACE 626 /* 627 * Copy traceflag and tracefile if enabled. 628 */ 629 mtx_lock(&ktrace_mtx); 630 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 631 if (p1->p_traceflag & KTRFAC_INHERIT) { 632 p2->p_traceflag = p1->p_traceflag; 633 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 634 VREF(p2->p_tracevp); 635 KASSERT(p1->p_tracecred != NULL, 636 ("ktrace vnode with no cred")); 637 p2->p_tracecred = crhold(p1->p_tracecred); 638 } 639 } 640 mtx_unlock(&ktrace_mtx); 641 #endif 642 643 /* 644 * If PF_FORK is set, the child process inherits the 645 * procfs ioctl flags from its parent. 646 */ 647 if (p1->p_pfsflags & PF_FORK) { 648 p2->p_stops = p1->p_stops; 649 p2->p_pfsflags = p1->p_pfsflags; 650 } 651 652 #ifdef KDTRACE_HOOKS 653 /* 654 * Tell the DTrace fasttrap provider about the new process 655 * if it has registered an interest. 656 */ 657 if (dtrace_fasttrap_fork) 658 dtrace_fasttrap_fork(p1, p2); 659 #endif 660 661 /* 662 * This begins the section where we must prevent the parent 663 * from being swapped. 664 */ 665 _PHOLD(p1); 666 PROC_UNLOCK(p1); 667 668 /* 669 * Attach the new process to its parent. 670 * 671 * If RFNOWAIT is set, the newly created process becomes a child 672 * of init. This effectively disassociates the child from the 673 * parent. 674 */ 675 if (flags & RFNOWAIT) 676 pptr = initproc; 677 else 678 pptr = p1; 679 p2->p_pptr = pptr; 680 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 681 sx_xunlock(&proctree_lock); 682 683 /* Inform accounting that we have forked. */ 684 p2->p_acflag = AFORK; 685 PROC_UNLOCK(p2); 686 687 /* 688 * Finish creating the child process. It will return via a different 689 * execution path later. (ie: directly into user mode) 690 */ 691 vm_forkproc(td, p2, td2, vm2, flags); 692 693 if (flags == (RFFDG | RFPROC)) { 694 PCPU_INC(cnt.v_forks); 695 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 696 p2->p_vmspace->vm_ssize); 697 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 698 PCPU_INC(cnt.v_vforks); 699 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 700 p2->p_vmspace->vm_ssize); 701 } else if (p1 == &proc0) { 702 PCPU_INC(cnt.v_kthreads); 703 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 704 p2->p_vmspace->vm_ssize); 705 } else { 706 PCPU_INC(cnt.v_rforks); 707 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 708 p2->p_vmspace->vm_ssize); 709 } 710 711 /* 712 * Both processes are set up, now check if any loadable modules want 713 * to adjust anything. 714 * What if they have an error? XXX 715 */ 716 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 717 718 /* 719 * Set the child start time and mark the process as being complete. 720 */ 721 microuptime(&p2->p_stats->p_start); 722 PROC_SLOCK(p2); 723 p2->p_state = PRS_NORMAL; 724 PROC_SUNLOCK(p2); 725 726 /* 727 * If RFSTOPPED not requested, make child runnable and add to 728 * run queue. 729 */ 730 if ((flags & RFSTOPPED) == 0) { 731 thread_lock(td2); 732 TD_SET_CAN_RUN(td2); 733 sched_add(td2, SRQ_BORING); 734 thread_unlock(td2); 735 } 736 737 /* 738 * Now can be swapped. 739 */ 740 PROC_LOCK(p1); 741 _PRELE(p1); 742 PROC_UNLOCK(p1); 743 744 /* 745 * Tell any interested parties about the new process. 746 */ 747 knote_fork(&p1->p_klist, p2->p_pid); 748 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0); 749 750 /* 751 * Preserve synchronization semantics of vfork. If waiting for 752 * child to exec or exit, set P_PPWAIT on child, and sleep on our 753 * proc (in case of exit). 754 */ 755 PROC_LOCK(p2); 756 while (p2->p_flag & P_PPWAIT) 757 cv_wait(&p2->p_pwait, &p2->p_mtx); 758 PROC_UNLOCK(p2); 759 760 /* 761 * Return child proc pointer to parent. 762 */ 763 *procp = p2; 764 return (0); 765 fail: 766 sx_sunlock(&proctree_lock); 767 if (ppsratecheck(&lastfail, &curfail, 1)) 768 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 769 td->td_ucred->cr_ruid); 770 sx_xunlock(&allproc_lock); 771 #ifdef MAC 772 mac_proc_destroy(newproc); 773 #endif 774 fail1: 775 if (vm2 != NULL) 776 vmspace_free(vm2); 777 uma_zfree(proc_zone, newproc); 778 pause("fork", hz / 2); 779 return (error); 780 } 781 782 /* 783 * Handle the return of a child process from fork1(). This function 784 * is called from the MD fork_trampoline() entry point. 785 */ 786 void 787 fork_exit(callout, arg, frame) 788 void (*callout)(void *, struct trapframe *); 789 void *arg; 790 struct trapframe *frame; 791 { 792 struct proc *p; 793 struct thread *td; 794 struct thread *dtd; 795 796 td = curthread; 797 p = td->td_proc; 798 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 799 800 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 801 td, td->td_sched, p->p_pid, td->td_name); 802 803 sched_fork_exit(td); 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 if ((dtd = PCPU_GET(deadthread))) { 810 PCPU_SET(deadthread, NULL); 811 thread_stash(dtd); 812 } 813 thread_unlock(td); 814 815 /* 816 * cpu_set_fork_handler intercepts this function call to 817 * have this call a non-return function to stay in kernel mode. 818 * initproc has its own fork handler, but it does return. 819 */ 820 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 821 callout(arg, frame); 822 823 /* 824 * Check if a kernel thread misbehaved and returned from its main 825 * function. 826 */ 827 if (p->p_flag & P_KTHREAD) { 828 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 829 td->td_name, p->p_pid); 830 kproc_exit(0); 831 } 832 mtx_assert(&Giant, MA_NOTOWNED); 833 834 EVENTHANDLER_INVOKE(schedtail, p); 835 } 836 837 /* 838 * Simplified back end of syscall(), used when returning from fork() 839 * directly into user mode. Giant is not held on entry, and must not 840 * be held on return. This function is passed in to fork_exit() as the 841 * first parameter and is called when returning to a new userland process. 842 */ 843 void 844 fork_return(td, frame) 845 struct thread *td; 846 struct trapframe *frame; 847 { 848 849 userret(td, frame); 850 #ifdef KTRACE 851 if (KTRPOINT(td, KTR_SYSRET)) 852 ktrsysret(SYS_fork, 0, 0); 853 #endif 854 mtx_assert(&Giant, MA_NOTOWNED); 855 } 856