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 #include <sys/vimage.h> 72 73 #include <security/audit/audit.h> 74 #include <security/mac/mac_framework.h> 75 76 #include <vm/vm.h> 77 #include <vm/pmap.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_extern.h> 80 #include <vm/uma.h> 81 82 #ifdef KDTRACE_HOOKS 83 #include <sys/dtrace_bsd.h> 84 dtrace_fork_func_t dtrace_fasttrap_fork; 85 #endif 86 87 SDT_PROVIDER_DECLARE(proc); 88 SDT_PROBE_DEFINE(proc, kernel, , create); 89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *"); 90 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *"); 91 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int"); 92 93 #ifndef _SYS_SYSPROTO_H_ 94 struct fork_args { 95 int dummy; 96 }; 97 #endif 98 99 /* ARGSUSED */ 100 int 101 fork(td, uap) 102 struct thread *td; 103 struct fork_args *uap; 104 { 105 int error; 106 struct proc *p2; 107 108 error = fork1(td, RFFDG | RFPROC, 0, &p2); 109 if (error == 0) { 110 td->td_retval[0] = p2->p_pid; 111 td->td_retval[1] = 0; 112 } 113 return (error); 114 } 115 116 /* ARGSUSED */ 117 int 118 vfork(td, uap) 119 struct thread *td; 120 struct vfork_args *uap; 121 { 122 int error, flags; 123 struct proc *p2; 124 125 #ifdef XEN 126 flags = RFFDG | RFPROC; /* validate that this is still an issue */ 127 #else 128 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM; 129 #endif 130 error = fork1(td, flags, 0, &p2); 131 if (error == 0) { 132 td->td_retval[0] = p2->p_pid; 133 td->td_retval[1] = 0; 134 } 135 return (error); 136 } 137 138 int 139 rfork(td, uap) 140 struct thread *td; 141 struct rfork_args *uap; 142 { 143 struct proc *p2; 144 int error; 145 146 /* Don't allow kernel-only flags. */ 147 if ((uap->flags & RFKERNELONLY) != 0) 148 return (EINVAL); 149 150 AUDIT_ARG(fflags, uap->flags); 151 error = fork1(td, uap->flags, 0, &p2); 152 if (error == 0) { 153 td->td_retval[0] = p2 ? p2->p_pid : 0; 154 td->td_retval[1] = 0; 155 } 156 return (error); 157 } 158 159 int nprocs = 1; /* process 0 */ 160 int lastpid = 0; 161 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 162 "Last used PID"); 163 164 /* 165 * Random component to lastpid generation. We mix in a random factor to make 166 * it a little harder to predict. We sanity check the modulus value to avoid 167 * doing it in critical paths. Don't let it be too small or we pointlessly 168 * waste randomness entropy, and don't let it be impossibly large. Using a 169 * modulus that is too big causes a LOT more process table scans and slows 170 * down fork processing as the pidchecked caching is defeated. 171 */ 172 static int randompid = 0; 173 174 static int 175 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 176 { 177 int error, pid; 178 179 error = sysctl_wire_old_buffer(req, sizeof(int)); 180 if (error != 0) 181 return(error); 182 sx_xlock(&allproc_lock); 183 pid = randompid; 184 error = sysctl_handle_int(oidp, &pid, 0, req); 185 if (error == 0 && req->newptr != NULL) { 186 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 187 pid = PID_MAX - 100; 188 else if (pid < 2) /* NOP */ 189 pid = 0; 190 else if (pid < 100) /* Make it reasonable */ 191 pid = 100; 192 randompid = pid; 193 } 194 sx_xunlock(&allproc_lock); 195 return (error); 196 } 197 198 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 199 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 200 201 int 202 fork1(td, flags, pages, procp) 203 struct thread *td; 204 int flags; 205 int pages; 206 struct proc **procp; 207 { 208 struct proc *p1, *p2, *pptr; 209 struct proc *newproc; 210 int ok, trypid; 211 static int curfail, pidchecked = 0; 212 static struct timeval lastfail; 213 struct filedesc *fd; 214 struct filedesc_to_leader *fdtol; 215 struct thread *td2; 216 struct sigacts *newsigacts; 217 struct vmspace *vm2; 218 int error; 219 220 /* Can't copy and clear. */ 221 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 222 return (EINVAL); 223 224 p1 = td->td_proc; 225 226 /* 227 * Here we don't create a new process, but we divorce 228 * certain parts of a process from itself. 229 */ 230 if ((flags & RFPROC) == 0) { 231 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 232 (flags & (RFCFDG | RFFDG))) { 233 PROC_LOCK(p1); 234 if (thread_single(SINGLE_BOUNDARY)) { 235 PROC_UNLOCK(p1); 236 return (ERESTART); 237 } 238 PROC_UNLOCK(p1); 239 } 240 241 error = vm_forkproc(td, NULL, NULL, NULL, flags); 242 if (error) 243 goto norfproc_fail; 244 245 /* 246 * Close all file descriptors. 247 */ 248 if (flags & RFCFDG) { 249 struct filedesc *fdtmp; 250 fdtmp = fdinit(td->td_proc->p_fd); 251 fdfree(td); 252 p1->p_fd = fdtmp; 253 } 254 255 /* 256 * Unshare file descriptors (from parent). 257 */ 258 if (flags & RFFDG) 259 fdunshare(p1, td); 260 261 norfproc_fail: 262 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) && 263 (flags & (RFCFDG | RFFDG))) { 264 PROC_LOCK(p1); 265 thread_single_end(); 266 PROC_UNLOCK(p1); 267 } 268 *procp = NULL; 269 return (error); 270 } 271 272 /* 273 * XXX 274 * We did have single-threading code here 275 * however it proved un-needed and caused problems 276 */ 277 278 vm2 = NULL; 279 /* Allocate new proc. */ 280 newproc = uma_zalloc(proc_zone, M_WAITOK); 281 if (TAILQ_EMPTY(&newproc->p_threads)) { 282 td2 = thread_alloc(); 283 if (td2 == NULL) { 284 error = ENOMEM; 285 goto fail1; 286 } 287 proc_linkup(newproc, td2); 288 } else 289 td2 = FIRST_THREAD_IN_PROC(newproc); 290 291 /* Allocate and switch to an alternate kstack if specified. */ 292 if (pages != 0) { 293 if (!vm_thread_new_altkstack(td2, pages)) { 294 error = ENOMEM; 295 goto fail1; 296 } 297 } 298 if ((flags & RFMEM) == 0) { 299 vm2 = vmspace_fork(p1->p_vmspace); 300 if (vm2 == NULL) { 301 error = ENOMEM; 302 goto fail1; 303 } 304 } 305 #ifdef MAC 306 mac_proc_init(newproc); 307 #endif 308 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 309 STAILQ_INIT(&newproc->p_ktr); 310 311 /* We have to lock the process tree while we look for a pid. */ 312 sx_slock(&proctree_lock); 313 314 /* 315 * Although process entries are dynamically created, we still keep 316 * a global limit on the maximum number we will create. Don't allow 317 * a nonprivileged user to use the last ten processes; don't let root 318 * exceed the limit. The variable nprocs is the current number of 319 * processes, maxproc is the limit. 320 */ 321 sx_xlock(&allproc_lock); 322 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, 323 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { 324 error = EAGAIN; 325 goto fail; 326 } 327 328 /* 329 * Increment the count of procs running with this uid. Don't allow 330 * a nonprivileged user to exceed their current limit. 331 * 332 * XXXRW: Can we avoid privilege here if it's not needed? 333 */ 334 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 335 if (error == 0) 336 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 337 else { 338 PROC_LOCK(p1); 339 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 340 lim_cur(p1, RLIMIT_NPROC)); 341 PROC_UNLOCK(p1); 342 } 343 if (!ok) { 344 error = EAGAIN; 345 goto fail; 346 } 347 348 /* 349 * Increment the nprocs resource before blocking can occur. There 350 * are hard-limits as to the number of processes that can run. 351 */ 352 nprocs++; 353 354 /* 355 * Find an unused process ID. We remember a range of unused IDs 356 * ready to use (from lastpid+1 through pidchecked-1). 357 * 358 * If RFHIGHPID is set (used during system boot), do not allocate 359 * low-numbered pids. 360 */ 361 trypid = lastpid + 1; 362 if (flags & RFHIGHPID) { 363 if (trypid < 10) 364 trypid = 10; 365 } else { 366 if (randompid) 367 trypid += arc4random() % randompid; 368 } 369 retry: 370 /* 371 * If the process ID prototype has wrapped around, 372 * restart somewhat above 0, as the low-numbered procs 373 * tend to include daemons that don't exit. 374 */ 375 if (trypid >= PID_MAX) { 376 trypid = trypid % PID_MAX; 377 if (trypid < 100) 378 trypid += 100; 379 pidchecked = 0; 380 } 381 if (trypid >= pidchecked) { 382 int doingzomb = 0; 383 384 pidchecked = PID_MAX; 385 /* 386 * Scan the active and zombie procs to check whether this pid 387 * is in use. Remember the lowest pid that's greater 388 * than trypid, so we can avoid checking for a while. 389 */ 390 p2 = LIST_FIRST(&allproc); 391 again: 392 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 393 while (p2->p_pid == trypid || 394 (p2->p_pgrp != NULL && 395 (p2->p_pgrp->pg_id == trypid || 396 (p2->p_session != NULL && 397 p2->p_session->s_sid == trypid)))) { 398 trypid++; 399 if (trypid >= pidchecked) 400 goto retry; 401 } 402 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 403 pidchecked = p2->p_pid; 404 if (p2->p_pgrp != NULL) { 405 if (p2->p_pgrp->pg_id > trypid && 406 pidchecked > p2->p_pgrp->pg_id) 407 pidchecked = p2->p_pgrp->pg_id; 408 if (p2->p_session != NULL && 409 p2->p_session->s_sid > trypid && 410 pidchecked > p2->p_session->s_sid) 411 pidchecked = p2->p_session->s_sid; 412 } 413 } 414 if (!doingzomb) { 415 doingzomb = 1; 416 p2 = LIST_FIRST(&zombproc); 417 goto again; 418 } 419 } 420 sx_sunlock(&proctree_lock); 421 422 /* 423 * RFHIGHPID does not mess with the lastpid counter during boot. 424 */ 425 if (flags & RFHIGHPID) 426 pidchecked = 0; 427 else 428 lastpid = trypid; 429 430 p2 = newproc; 431 p2->p_state = PRS_NEW; /* protect against others */ 432 p2->p_pid = trypid; 433 /* 434 * Allow the scheduler to initialize the child. 435 */ 436 thread_lock(td); 437 sched_fork(td, td2); 438 thread_unlock(td); 439 AUDIT_ARG(pid, p2->p_pid); 440 LIST_INSERT_HEAD(&allproc, p2, p_list); 441 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 442 443 PROC_LOCK(p2); 444 PROC_LOCK(p1); 445 446 sx_xunlock(&allproc_lock); 447 448 bcopy(&p1->p_startcopy, &p2->p_startcopy, 449 __rangeof(struct proc, p_startcopy, p_endcopy)); 450 pargs_hold(p2->p_args); 451 PROC_UNLOCK(p1); 452 453 bzero(&p2->p_startzero, 454 __rangeof(struct proc, p_startzero, p_endzero)); 455 456 p2->p_ucred = crhold(td->td_ucred); 457 458 /* In case we are jailed tell the prison that we exist. */ 459 if (jailed(p2->p_ucred)) 460 prison_proc_hold(p2->p_ucred->cr_prison); 461 462 PROC_UNLOCK(p2); 463 464 /* 465 * Malloc things while we don't hold any locks. 466 */ 467 if (flags & RFSIGSHARE) 468 newsigacts = NULL; 469 else 470 newsigacts = sigacts_alloc(); 471 472 /* 473 * Copy filedesc. 474 */ 475 if (flags & RFCFDG) { 476 fd = fdinit(p1->p_fd); 477 fdtol = NULL; 478 } else if (flags & RFFDG) { 479 fd = fdcopy(p1->p_fd); 480 fdtol = NULL; 481 } else { 482 fd = fdshare(p1->p_fd); 483 if (p1->p_fdtol == NULL) 484 p1->p_fdtol = 485 filedesc_to_leader_alloc(NULL, 486 NULL, 487 p1->p_leader); 488 if ((flags & RFTHREAD) != 0) { 489 /* 490 * Shared file descriptor table and 491 * shared process leaders. 492 */ 493 fdtol = p1->p_fdtol; 494 FILEDESC_XLOCK(p1->p_fd); 495 fdtol->fdl_refcount++; 496 FILEDESC_XUNLOCK(p1->p_fd); 497 } else { 498 /* 499 * Shared file descriptor table, and 500 * different process leaders 501 */ 502 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 503 p1->p_fd, 504 p2); 505 } 506 } 507 /* 508 * Make a proc table entry for the new process. 509 * Start by zeroing the section of proc that is zero-initialized, 510 * then copy the section that is copied directly from the parent. 511 */ 512 513 PROC_LOCK(p2); 514 PROC_LOCK(p1); 515 516 bzero(&td2->td_startzero, 517 __rangeof(struct thread, td_startzero, td_endzero)); 518 519 bcopy(&td->td_startcopy, &td2->td_startcopy, 520 __rangeof(struct thread, td_startcopy, td_endcopy)); 521 522 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 523 td2->td_sigstk = td->td_sigstk; 524 td2->td_sigmask = td->td_sigmask; 525 td2->td_flags = TDF_INMEM; 526 527 #ifdef VIMAGE 528 td2->td_vnet = NULL; 529 td2->td_vnet_lpush = NULL; 530 #endif 531 532 /* 533 * Duplicate sub-structures as needed. 534 * Increase reference counts on shared objects. 535 */ 536 p2->p_flag = P_INMEM; 537 p2->p_swtick = ticks; 538 if (p1->p_flag & P_PROFIL) 539 startprofclock(p2); 540 td2->td_ucred = crhold(p2->p_ucred); 541 542 if (flags & RFSIGSHARE) { 543 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 544 } else { 545 sigacts_copy(newsigacts, p1->p_sigacts); 546 p2->p_sigacts = newsigacts; 547 } 548 if (flags & RFLINUXTHPN) 549 p2->p_sigparent = SIGUSR1; 550 else 551 p2->p_sigparent = SIGCHLD; 552 553 p2->p_textvp = p1->p_textvp; 554 p2->p_fd = fd; 555 p2->p_fdtol = fdtol; 556 557 /* 558 * p_limit is copy-on-write. Bump its refcount. 559 */ 560 lim_fork(p1, p2); 561 562 pstats_fork(p1->p_stats, p2->p_stats); 563 564 PROC_UNLOCK(p1); 565 PROC_UNLOCK(p2); 566 567 /* Bump references to the text vnode (for procfs) */ 568 if (p2->p_textvp) 569 vref(p2->p_textvp); 570 571 /* 572 * Set up linkage for kernel based threading. 573 */ 574 if ((flags & RFTHREAD) != 0) { 575 mtx_lock(&ppeers_lock); 576 p2->p_peers = p1->p_peers; 577 p1->p_peers = p2; 578 p2->p_leader = p1->p_leader; 579 mtx_unlock(&ppeers_lock); 580 PROC_LOCK(p1->p_leader); 581 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 582 PROC_UNLOCK(p1->p_leader); 583 /* 584 * The task leader is exiting, so process p1 is 585 * going to be killed shortly. Since p1 obviously 586 * isn't dead yet, we know that the leader is either 587 * sending SIGKILL's to all the processes in this 588 * task or is sleeping waiting for all the peers to 589 * exit. We let p1 complete the fork, but we need 590 * to go ahead and kill the new process p2 since 591 * the task leader may not get a chance to send 592 * SIGKILL to it. We leave it on the list so that 593 * the task leader will wait for this new process 594 * to commit suicide. 595 */ 596 PROC_LOCK(p2); 597 psignal(p2, SIGKILL); 598 PROC_UNLOCK(p2); 599 } else 600 PROC_UNLOCK(p1->p_leader); 601 } else { 602 p2->p_peers = NULL; 603 p2->p_leader = p2; 604 } 605 606 sx_xlock(&proctree_lock); 607 PGRP_LOCK(p1->p_pgrp); 608 PROC_LOCK(p2); 609 PROC_LOCK(p1); 610 611 /* 612 * Preserve some more flags in subprocess. P_PROFIL has already 613 * been preserved. 614 */ 615 p2->p_flag |= p1->p_flag & P_SUGID; 616 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 617 SESS_LOCK(p1->p_session); 618 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 619 p2->p_flag |= P_CONTROLT; 620 SESS_UNLOCK(p1->p_session); 621 if (flags & RFPPWAIT) 622 p2->p_flag |= P_PPWAIT; 623 624 p2->p_pgrp = p1->p_pgrp; 625 LIST_INSERT_AFTER(p1, p2, p_pglist); 626 PGRP_UNLOCK(p1->p_pgrp); 627 LIST_INIT(&p2->p_children); 628 629 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 630 631 #ifdef KTRACE 632 /* 633 * Copy traceflag and tracefile if enabled. 634 */ 635 mtx_lock(&ktrace_mtx); 636 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 637 if (p1->p_traceflag & KTRFAC_INHERIT) { 638 p2->p_traceflag = p1->p_traceflag; 639 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 640 VREF(p2->p_tracevp); 641 KASSERT(p1->p_tracecred != NULL, 642 ("ktrace vnode with no cred")); 643 p2->p_tracecred = crhold(p1->p_tracecred); 644 } 645 } 646 mtx_unlock(&ktrace_mtx); 647 #endif 648 649 /* 650 * If PF_FORK is set, the child process inherits the 651 * procfs ioctl flags from its parent. 652 */ 653 if (p1->p_pfsflags & PF_FORK) { 654 p2->p_stops = p1->p_stops; 655 p2->p_pfsflags = p1->p_pfsflags; 656 } 657 658 #ifdef KDTRACE_HOOKS 659 /* 660 * Tell the DTrace fasttrap provider about the new process 661 * if it has registered an interest. 662 */ 663 if (dtrace_fasttrap_fork) 664 dtrace_fasttrap_fork(p1, p2); 665 #endif 666 667 /* 668 * This begins the section where we must prevent the parent 669 * from being swapped. 670 */ 671 _PHOLD(p1); 672 PROC_UNLOCK(p1); 673 674 /* 675 * Attach the new process to its parent. 676 * 677 * If RFNOWAIT is set, the newly created process becomes a child 678 * of init. This effectively disassociates the child from the 679 * parent. 680 */ 681 if (flags & RFNOWAIT) 682 pptr = initproc; 683 else 684 pptr = p1; 685 p2->p_pptr = pptr; 686 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 687 sx_xunlock(&proctree_lock); 688 689 /* Inform accounting that we have forked. */ 690 p2->p_acflag = AFORK; 691 PROC_UNLOCK(p2); 692 693 /* 694 * Finish creating the child process. It will return via a different 695 * execution path later. (ie: directly into user mode) 696 */ 697 vm_forkproc(td, p2, td2, vm2, flags); 698 699 if (flags == (RFFDG | RFPROC)) { 700 PCPU_INC(cnt.v_forks); 701 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 702 p2->p_vmspace->vm_ssize); 703 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 704 PCPU_INC(cnt.v_vforks); 705 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 706 p2->p_vmspace->vm_ssize); 707 } else if (p1 == &proc0) { 708 PCPU_INC(cnt.v_kthreads); 709 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 710 p2->p_vmspace->vm_ssize); 711 } else { 712 PCPU_INC(cnt.v_rforks); 713 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 714 p2->p_vmspace->vm_ssize); 715 } 716 717 /* 718 * Both processes are set up, now check if any loadable modules want 719 * to adjust anything. 720 * What if they have an error? XXX 721 */ 722 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 723 724 /* 725 * Set the child start time and mark the process as being complete. 726 */ 727 microuptime(&p2->p_stats->p_start); 728 PROC_SLOCK(p2); 729 p2->p_state = PRS_NORMAL; 730 PROC_SUNLOCK(p2); 731 732 /* 733 * If RFSTOPPED not requested, make child runnable and add to 734 * run queue. 735 */ 736 if ((flags & RFSTOPPED) == 0) { 737 thread_lock(td2); 738 TD_SET_CAN_RUN(td2); 739 sched_add(td2, SRQ_BORING); 740 thread_unlock(td2); 741 } 742 743 /* 744 * Now can be swapped. 745 */ 746 PROC_LOCK(p1); 747 _PRELE(p1); 748 PROC_UNLOCK(p1); 749 750 /* 751 * Tell any interested parties about the new process. 752 */ 753 knote_fork(&p1->p_klist, p2->p_pid); 754 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0); 755 756 /* 757 * Preserve synchronization semantics of vfork. If waiting for 758 * child to exec or exit, set P_PPWAIT on child, and sleep on our 759 * proc (in case of exit). 760 */ 761 PROC_LOCK(p2); 762 while (p2->p_flag & P_PPWAIT) 763 cv_wait(&p2->p_pwait, &p2->p_mtx); 764 PROC_UNLOCK(p2); 765 766 /* 767 * Return child proc pointer to parent. 768 */ 769 *procp = p2; 770 return (0); 771 fail: 772 sx_sunlock(&proctree_lock); 773 if (ppsratecheck(&lastfail, &curfail, 1)) 774 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 775 td->td_ucred->cr_ruid); 776 sx_xunlock(&allproc_lock); 777 #ifdef MAC 778 mac_proc_destroy(newproc); 779 #endif 780 fail1: 781 if (vm2 != NULL) 782 vmspace_free(vm2); 783 uma_zfree(proc_zone, newproc); 784 pause("fork", hz / 2); 785 return (error); 786 } 787 788 /* 789 * Handle the return of a child process from fork1(). This function 790 * is called from the MD fork_trampoline() entry point. 791 */ 792 void 793 fork_exit(callout, arg, frame) 794 void (*callout)(void *, struct trapframe *); 795 void *arg; 796 struct trapframe *frame; 797 { 798 struct proc *p; 799 struct thread *td; 800 struct thread *dtd; 801 802 td = curthread; 803 p = td->td_proc; 804 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 805 806 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 807 td, td->td_sched, p->p_pid, td->td_name); 808 809 sched_fork_exit(td); 810 /* 811 * Processes normally resume in mi_switch() after being 812 * cpu_switch()'ed to, but when children start up they arrive here 813 * instead, so we must do much the same things as mi_switch() would. 814 */ 815 if ((dtd = PCPU_GET(deadthread))) { 816 PCPU_SET(deadthread, NULL); 817 thread_stash(dtd); 818 } 819 thread_unlock(td); 820 821 /* 822 * cpu_set_fork_handler intercepts this function call to 823 * have this call a non-return function to stay in kernel mode. 824 * initproc has its own fork handler, but it does return. 825 */ 826 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 827 callout(arg, frame); 828 829 /* 830 * Check if a kernel thread misbehaved and returned from its main 831 * function. 832 */ 833 if (p->p_flag & P_KTHREAD) { 834 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 835 td->td_name, p->p_pid); 836 kproc_exit(0); 837 } 838 mtx_assert(&Giant, MA_NOTOWNED); 839 840 EVENTHANDLER_INVOKE(schedtail, p); 841 } 842 843 /* 844 * Simplified back end of syscall(), used when returning from fork() 845 * directly into user mode. Giant is not held on entry, and must not 846 * be held on return. This function is passed in to fork_exit() as the 847 * first parameter and is called when returning to a new userland process. 848 */ 849 void 850 fork_return(td, frame) 851 struct thread *td; 852 struct trapframe *frame; 853 { 854 855 userret(td, frame); 856 #ifdef KTRACE 857 if (KTRPOINT(td, KTR_SYSRET)) 858 ktrsysret(SYS_fork, 0, 0); 859 #endif 860 mtx_assert(&Giant, MA_NOTOWNED); 861 } 862