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 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/jail.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 #include <sys/vimage.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 vm_ooffset_t mem_charged; 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 mem_charged = 0; 279 vm2 = NULL; 280 /* Allocate new proc. */ 281 newproc = uma_zalloc(proc_zone, M_WAITOK); 282 if (TAILQ_EMPTY(&newproc->p_threads)) { 283 td2 = thread_alloc(); 284 if (td2 == NULL) { 285 error = ENOMEM; 286 goto fail1; 287 } 288 proc_linkup(newproc, td2); 289 } else 290 td2 = FIRST_THREAD_IN_PROC(newproc); 291 292 /* Allocate and switch to an alternate kstack if specified. */ 293 if (pages != 0) { 294 if (!vm_thread_new_altkstack(td2, pages)) { 295 error = ENOMEM; 296 goto fail1; 297 } 298 } 299 if ((flags & RFMEM) == 0) { 300 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged); 301 if (vm2 == NULL) { 302 error = ENOMEM; 303 goto fail1; 304 } 305 if (!swap_reserve(mem_charged)) { 306 /* 307 * The swap reservation failed. The accounting 308 * from the entries of the copied vm2 will be 309 * substracted in vmspace_free(), so force the 310 * reservation there. 311 */ 312 swap_reserve_force(mem_charged); 313 error = ENOMEM; 314 goto fail1; 315 } 316 } else 317 vm2 = NULL; 318 #ifdef MAC 319 mac_proc_init(newproc); 320 #endif 321 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx); 322 STAILQ_INIT(&newproc->p_ktr); 323 324 /* We have to lock the process tree while we look for a pid. */ 325 sx_slock(&proctree_lock); 326 327 /* 328 * Although process entries are dynamically created, we still keep 329 * a global limit on the maximum number we will create. Don't allow 330 * a nonprivileged user to use the last ten processes; don't let root 331 * exceed the limit. The variable nprocs is the current number of 332 * processes, maxproc is the limit. 333 */ 334 sx_xlock(&allproc_lock); 335 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred, 336 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) { 337 error = EAGAIN; 338 goto fail; 339 } 340 341 /* 342 * Increment the count of procs running with this uid. Don't allow 343 * a nonprivileged user to exceed their current limit. 344 * 345 * XXXRW: Can we avoid privilege here if it's not needed? 346 */ 347 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0); 348 if (error == 0) 349 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 350 else { 351 PROC_LOCK(p1); 352 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 353 lim_cur(p1, RLIMIT_NPROC)); 354 PROC_UNLOCK(p1); 355 } 356 if (!ok) { 357 error = EAGAIN; 358 goto fail; 359 } 360 361 /* 362 * Increment the nprocs resource before blocking can occur. There 363 * are hard-limits as to the number of processes that can run. 364 */ 365 nprocs++; 366 #ifdef VIMAGE 367 P_TO_VPROCG(p1)->nprocs++; 368 #endif 369 370 /* 371 * Find an unused process ID. We remember a range of unused IDs 372 * ready to use (from lastpid+1 through pidchecked-1). 373 * 374 * If RFHIGHPID is set (used during system boot), do not allocate 375 * low-numbered pids. 376 */ 377 trypid = lastpid + 1; 378 if (flags & RFHIGHPID) { 379 if (trypid < 10) 380 trypid = 10; 381 } else { 382 if (randompid) 383 trypid += arc4random() % randompid; 384 } 385 retry: 386 /* 387 * If the process ID prototype has wrapped around, 388 * restart somewhat above 0, as the low-numbered procs 389 * tend to include daemons that don't exit. 390 */ 391 if (trypid >= PID_MAX) { 392 trypid = trypid % PID_MAX; 393 if (trypid < 100) 394 trypid += 100; 395 pidchecked = 0; 396 } 397 if (trypid >= pidchecked) { 398 int doingzomb = 0; 399 400 pidchecked = PID_MAX; 401 /* 402 * Scan the active and zombie procs to check whether this pid 403 * is in use. Remember the lowest pid that's greater 404 * than trypid, so we can avoid checking for a while. 405 */ 406 p2 = LIST_FIRST(&allproc); 407 again: 408 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 409 while (p2->p_pid == trypid || 410 (p2->p_pgrp != NULL && 411 (p2->p_pgrp->pg_id == trypid || 412 (p2->p_session != NULL && 413 p2->p_session->s_sid == trypid)))) { 414 trypid++; 415 if (trypid >= pidchecked) 416 goto retry; 417 } 418 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 419 pidchecked = p2->p_pid; 420 if (p2->p_pgrp != NULL) { 421 if (p2->p_pgrp->pg_id > trypid && 422 pidchecked > p2->p_pgrp->pg_id) 423 pidchecked = p2->p_pgrp->pg_id; 424 if (p2->p_session != NULL && 425 p2->p_session->s_sid > trypid && 426 pidchecked > p2->p_session->s_sid) 427 pidchecked = p2->p_session->s_sid; 428 } 429 } 430 if (!doingzomb) { 431 doingzomb = 1; 432 p2 = LIST_FIRST(&zombproc); 433 goto again; 434 } 435 } 436 sx_sunlock(&proctree_lock); 437 438 /* 439 * RFHIGHPID does not mess with the lastpid counter during boot. 440 */ 441 if (flags & RFHIGHPID) 442 pidchecked = 0; 443 else 444 lastpid = trypid; 445 446 p2 = newproc; 447 p2->p_state = PRS_NEW; /* protect against others */ 448 p2->p_pid = trypid; 449 /* 450 * Allow the scheduler to initialize the child. 451 */ 452 thread_lock(td); 453 sched_fork(td, td2); 454 thread_unlock(td); 455 AUDIT_ARG_PID(p2->p_pid); 456 LIST_INSERT_HEAD(&allproc, p2, p_list); 457 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 458 459 PROC_LOCK(p2); 460 PROC_LOCK(p1); 461 462 sx_xunlock(&allproc_lock); 463 464 bcopy(&p1->p_startcopy, &p2->p_startcopy, 465 __rangeof(struct proc, p_startcopy, p_endcopy)); 466 pargs_hold(p2->p_args); 467 PROC_UNLOCK(p1); 468 469 bzero(&p2->p_startzero, 470 __rangeof(struct proc, p_startzero, p_endzero)); 471 472 p2->p_ucred = crhold(td->td_ucred); 473 474 /* Tell the prison that we exist. */ 475 prison_proc_hold(p2->p_ucred->cr_prison); 476 477 PROC_UNLOCK(p2); 478 479 /* 480 * Malloc things while we don't hold any locks. 481 */ 482 if (flags & RFSIGSHARE) 483 newsigacts = NULL; 484 else 485 newsigacts = sigacts_alloc(); 486 487 /* 488 * Copy filedesc. 489 */ 490 if (flags & RFCFDG) { 491 fd = fdinit(p1->p_fd); 492 fdtol = NULL; 493 } else if (flags & RFFDG) { 494 fd = fdcopy(p1->p_fd); 495 fdtol = NULL; 496 } else { 497 fd = fdshare(p1->p_fd); 498 if (p1->p_fdtol == NULL) 499 p1->p_fdtol = 500 filedesc_to_leader_alloc(NULL, 501 NULL, 502 p1->p_leader); 503 if ((flags & RFTHREAD) != 0) { 504 /* 505 * Shared file descriptor table and 506 * shared process leaders. 507 */ 508 fdtol = p1->p_fdtol; 509 FILEDESC_XLOCK(p1->p_fd); 510 fdtol->fdl_refcount++; 511 FILEDESC_XUNLOCK(p1->p_fd); 512 } else { 513 /* 514 * Shared file descriptor table, and 515 * different process leaders 516 */ 517 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 518 p1->p_fd, 519 p2); 520 } 521 } 522 /* 523 * Make a proc table entry for the new process. 524 * Start by zeroing the section of proc that is zero-initialized, 525 * then copy the section that is copied directly from the parent. 526 */ 527 528 PROC_LOCK(p2); 529 PROC_LOCK(p1); 530 531 bzero(&td2->td_startzero, 532 __rangeof(struct thread, td_startzero, td_endzero)); 533 534 bcopy(&td->td_startcopy, &td2->td_startcopy, 535 __rangeof(struct thread, td_startcopy, td_endcopy)); 536 537 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name)); 538 td2->td_sigstk = td->td_sigstk; 539 td2->td_sigmask = td->td_sigmask; 540 td2->td_flags = TDF_INMEM; 541 542 #ifdef VIMAGE 543 td2->td_vnet = NULL; 544 td2->td_vnet_lpush = NULL; 545 #endif 546 547 /* 548 * Duplicate sub-structures as needed. 549 * Increase reference counts on shared objects. 550 */ 551 p2->p_flag = P_INMEM; 552 p2->p_swtick = ticks; 553 if (p1->p_flag & P_PROFIL) 554 startprofclock(p2); 555 td2->td_ucred = crhold(p2->p_ucred); 556 557 if (flags & RFSIGSHARE) { 558 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 559 } else { 560 sigacts_copy(newsigacts, p1->p_sigacts); 561 p2->p_sigacts = newsigacts; 562 } 563 if (flags & RFLINUXTHPN) 564 p2->p_sigparent = SIGUSR1; 565 else 566 p2->p_sigparent = SIGCHLD; 567 568 p2->p_textvp = p1->p_textvp; 569 p2->p_fd = fd; 570 p2->p_fdtol = fdtol; 571 572 /* 573 * p_limit is copy-on-write. Bump its refcount. 574 */ 575 lim_fork(p1, p2); 576 577 pstats_fork(p1->p_stats, p2->p_stats); 578 579 PROC_UNLOCK(p1); 580 PROC_UNLOCK(p2); 581 582 /* Bump references to the text vnode (for procfs) */ 583 if (p2->p_textvp) 584 vref(p2->p_textvp); 585 586 /* 587 * Set up linkage for kernel based threading. 588 */ 589 if ((flags & RFTHREAD) != 0) { 590 mtx_lock(&ppeers_lock); 591 p2->p_peers = p1->p_peers; 592 p1->p_peers = p2; 593 p2->p_leader = p1->p_leader; 594 mtx_unlock(&ppeers_lock); 595 PROC_LOCK(p1->p_leader); 596 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 597 PROC_UNLOCK(p1->p_leader); 598 /* 599 * The task leader is exiting, so process p1 is 600 * going to be killed shortly. Since p1 obviously 601 * isn't dead yet, we know that the leader is either 602 * sending SIGKILL's to all the processes in this 603 * task or is sleeping waiting for all the peers to 604 * exit. We let p1 complete the fork, but we need 605 * to go ahead and kill the new process p2 since 606 * the task leader may not get a chance to send 607 * SIGKILL to it. We leave it on the list so that 608 * the task leader will wait for this new process 609 * to commit suicide. 610 */ 611 PROC_LOCK(p2); 612 psignal(p2, SIGKILL); 613 PROC_UNLOCK(p2); 614 } else 615 PROC_UNLOCK(p1->p_leader); 616 } else { 617 p2->p_peers = NULL; 618 p2->p_leader = p2; 619 } 620 621 sx_xlock(&proctree_lock); 622 PGRP_LOCK(p1->p_pgrp); 623 PROC_LOCK(p2); 624 PROC_LOCK(p1); 625 626 /* 627 * Preserve some more flags in subprocess. P_PROFIL has already 628 * been preserved. 629 */ 630 p2->p_flag |= p1->p_flag & P_SUGID; 631 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 632 SESS_LOCK(p1->p_session); 633 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 634 p2->p_flag |= P_CONTROLT; 635 SESS_UNLOCK(p1->p_session); 636 if (flags & RFPPWAIT) 637 p2->p_flag |= P_PPWAIT; 638 639 p2->p_pgrp = p1->p_pgrp; 640 LIST_INSERT_AFTER(p1, p2, p_pglist); 641 PGRP_UNLOCK(p1->p_pgrp); 642 LIST_INIT(&p2->p_children); 643 644 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 645 646 #ifdef KTRACE 647 /* 648 * Copy traceflag and tracefile if enabled. 649 */ 650 mtx_lock(&ktrace_mtx); 651 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 652 if (p1->p_traceflag & KTRFAC_INHERIT) { 653 p2->p_traceflag = p1->p_traceflag; 654 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 655 VREF(p2->p_tracevp); 656 KASSERT(p1->p_tracecred != NULL, 657 ("ktrace vnode with no cred")); 658 p2->p_tracecred = crhold(p1->p_tracecred); 659 } 660 } 661 mtx_unlock(&ktrace_mtx); 662 #endif 663 664 /* 665 * If PF_FORK is set, the child process inherits the 666 * procfs ioctl flags from its parent. 667 */ 668 if (p1->p_pfsflags & PF_FORK) { 669 p2->p_stops = p1->p_stops; 670 p2->p_pfsflags = p1->p_pfsflags; 671 } 672 673 #ifdef KDTRACE_HOOKS 674 /* 675 * Tell the DTrace fasttrap provider about the new process 676 * if it has registered an interest. 677 */ 678 if (dtrace_fasttrap_fork) 679 dtrace_fasttrap_fork(p1, p2); 680 #endif 681 682 /* 683 * This begins the section where we must prevent the parent 684 * from being swapped. 685 */ 686 _PHOLD(p1); 687 PROC_UNLOCK(p1); 688 689 /* 690 * Attach the new process to its parent. 691 * 692 * If RFNOWAIT is set, the newly created process becomes a child 693 * of init. This effectively disassociates the child from the 694 * parent. 695 */ 696 if (flags & RFNOWAIT) 697 pptr = initproc; 698 else 699 pptr = p1; 700 p2->p_pptr = pptr; 701 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 702 sx_xunlock(&proctree_lock); 703 704 /* Inform accounting that we have forked. */ 705 p2->p_acflag = AFORK; 706 PROC_UNLOCK(p2); 707 708 /* 709 * Finish creating the child process. It will return via a different 710 * execution path later. (ie: directly into user mode) 711 */ 712 vm_forkproc(td, p2, td2, vm2, flags); 713 714 if (flags == (RFFDG | RFPROC)) { 715 PCPU_INC(cnt.v_forks); 716 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize + 717 p2->p_vmspace->vm_ssize); 718 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 719 PCPU_INC(cnt.v_vforks); 720 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 721 p2->p_vmspace->vm_ssize); 722 } else if (p1 == &proc0) { 723 PCPU_INC(cnt.v_kthreads); 724 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 725 p2->p_vmspace->vm_ssize); 726 } else { 727 PCPU_INC(cnt.v_rforks); 728 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 729 p2->p_vmspace->vm_ssize); 730 } 731 732 /* 733 * Both processes are set up, now check if any loadable modules want 734 * to adjust anything. 735 * What if they have an error? XXX 736 */ 737 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 738 739 /* 740 * Set the child start time and mark the process as being complete. 741 */ 742 microuptime(&p2->p_stats->p_start); 743 PROC_SLOCK(p2); 744 p2->p_state = PRS_NORMAL; 745 PROC_SUNLOCK(p2); 746 747 /* 748 * If RFSTOPPED not requested, make child runnable and add to 749 * run queue. 750 */ 751 if ((flags & RFSTOPPED) == 0) { 752 thread_lock(td2); 753 TD_SET_CAN_RUN(td2); 754 sched_add(td2, SRQ_BORING); 755 thread_unlock(td2); 756 } 757 758 /* 759 * Now can be swapped. 760 */ 761 PROC_LOCK(p1); 762 _PRELE(p1); 763 PROC_UNLOCK(p1); 764 765 /* 766 * Tell any interested parties about the new process. 767 */ 768 knote_fork(&p1->p_klist, p2->p_pid); 769 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0); 770 771 /* 772 * Preserve synchronization semantics of vfork. If waiting for 773 * child to exec or exit, set P_PPWAIT on child, and sleep on our 774 * proc (in case of exit). 775 */ 776 PROC_LOCK(p2); 777 while (p2->p_flag & P_PPWAIT) 778 cv_wait(&p2->p_pwait, &p2->p_mtx); 779 PROC_UNLOCK(p2); 780 781 /* 782 * Return child proc pointer to parent. 783 */ 784 *procp = p2; 785 return (0); 786 fail: 787 sx_sunlock(&proctree_lock); 788 if (ppsratecheck(&lastfail, &curfail, 1)) 789 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 790 td->td_ucred->cr_ruid); 791 sx_xunlock(&allproc_lock); 792 #ifdef MAC 793 mac_proc_destroy(newproc); 794 #endif 795 fail1: 796 if (vm2 != NULL) 797 vmspace_free(vm2); 798 uma_zfree(proc_zone, newproc); 799 pause("fork", hz / 2); 800 return (error); 801 } 802 803 /* 804 * Handle the return of a child process from fork1(). This function 805 * is called from the MD fork_trampoline() entry point. 806 */ 807 void 808 fork_exit(callout, arg, frame) 809 void (*callout)(void *, struct trapframe *); 810 void *arg; 811 struct trapframe *frame; 812 { 813 struct proc *p; 814 struct thread *td; 815 struct thread *dtd; 816 817 td = curthread; 818 p = td->td_proc; 819 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 820 821 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", 822 td, td->td_sched, p->p_pid, td->td_name); 823 824 sched_fork_exit(td); 825 /* 826 * Processes normally resume in mi_switch() after being 827 * cpu_switch()'ed to, but when children start up they arrive here 828 * instead, so we must do much the same things as mi_switch() would. 829 */ 830 if ((dtd = PCPU_GET(deadthread))) { 831 PCPU_SET(deadthread, NULL); 832 thread_stash(dtd); 833 } 834 thread_unlock(td); 835 836 /* 837 * cpu_set_fork_handler intercepts this function call to 838 * have this call a non-return function to stay in kernel mode. 839 * initproc has its own fork handler, but it does return. 840 */ 841 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 842 callout(arg, frame); 843 844 /* 845 * Check if a kernel thread misbehaved and returned from its main 846 * function. 847 */ 848 if (p->p_flag & P_KTHREAD) { 849 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 850 td->td_name, p->p_pid); 851 kproc_exit(0); 852 } 853 mtx_assert(&Giant, MA_NOTOWNED); 854 855 EVENTHANDLER_INVOKE(schedtail, p); 856 } 857 858 /* 859 * Simplified back end of syscall(), used when returning from fork() 860 * directly into user mode. Giant is not held on entry, and must not 861 * be held on return. This function is passed in to fork_exit() as the 862 * first parameter and is called when returning to a new userland process. 863 */ 864 void 865 fork_return(td, frame) 866 struct thread *td; 867 struct trapframe *frame; 868 { 869 870 userret(td, frame); 871 #ifdef KTRACE 872 if (KTRPOINT(td, KTR_SYSRET)) 873 ktrsysret(SYS_fork, 0, 0); 874 #endif 875 mtx_assert(&Giant, MA_NOTOWNED); 876 } 877