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