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_ktrace.h" 41 #include "opt_mac.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/kernel.h> 49 #include <sys/kthread.h> 50 #include <sys/sysctl.h> 51 #include <sys/lock.h> 52 #include <sys/malloc.h> 53 #include <sys/mutex.h> 54 #include <sys/proc.h> 55 #include <sys/pioctl.h> 56 #include <sys/resourcevar.h> 57 #include <sys/sched.h> 58 #include <sys/syscall.h> 59 #include <sys/vmmeter.h> 60 #include <sys/vnode.h> 61 #include <sys/acct.h> 62 #include <sys/ktr.h> 63 #include <sys/ktrace.h> 64 #include <sys/unistd.h> 65 #include <sys/sx.h> 66 #include <sys/signalvar.h> 67 68 #include <security/audit/audit.h> 69 #include <security/mac/mac_framework.h> 70 71 #include <vm/vm.h> 72 #include <vm/pmap.h> 73 #include <vm/vm_map.h> 74 #include <vm/vm_extern.h> 75 #include <vm/uma.h> 76 77 78 #ifndef _SYS_SYSPROTO_H_ 79 struct fork_args { 80 int dummy; 81 }; 82 #endif 83 84 static int forksleep; /* Place for fork1() to sleep on. */ 85 86 /* 87 * MPSAFE 88 */ 89 /* ARGSUSED */ 90 int 91 fork(td, uap) 92 struct thread *td; 93 struct fork_args *uap; 94 { 95 int error; 96 struct proc *p2; 97 98 error = fork1(td, RFFDG | RFPROC, 0, &p2); 99 if (error == 0) { 100 td->td_retval[0] = p2->p_pid; 101 td->td_retval[1] = 0; 102 } 103 return (error); 104 } 105 106 /* 107 * MPSAFE 108 */ 109 /* ARGSUSED */ 110 int 111 vfork(td, uap) 112 struct thread *td; 113 struct vfork_args *uap; 114 { 115 int error; 116 struct proc *p2; 117 118 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 119 if (error == 0) { 120 td->td_retval[0] = p2->p_pid; 121 td->td_retval[1] = 0; 122 } 123 return (error); 124 } 125 126 /* 127 * MPSAFE 128 */ 129 int 130 rfork(td, uap) 131 struct thread *td; 132 struct rfork_args *uap; 133 { 134 struct proc *p2; 135 int error; 136 137 /* Don't allow kernel-only flags. */ 138 if ((uap->flags & RFKERNELONLY) != 0) 139 return (EINVAL); 140 141 AUDIT_ARG(fflags, uap->flags); 142 error = fork1(td, uap->flags, 0, &p2); 143 if (error == 0) { 144 td->td_retval[0] = p2 ? p2->p_pid : 0; 145 td->td_retval[1] = 0; 146 } 147 return (error); 148 } 149 150 int nprocs = 1; /* process 0 */ 151 int lastpid = 0; 152 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 153 "Last used PID"); 154 155 /* 156 * Random component to lastpid generation. We mix in a random factor to make 157 * it a little harder to predict. We sanity check the modulus value to avoid 158 * doing it in critical paths. Don't let it be too small or we pointlessly 159 * waste randomness entropy, and don't let it be impossibly large. Using a 160 * modulus that is too big causes a LOT more process table scans and slows 161 * down fork processing as the pidchecked caching is defeated. 162 */ 163 static int randompid = 0; 164 165 static int 166 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 167 { 168 int error, pid; 169 170 error = sysctl_wire_old_buffer(req, sizeof(int)); 171 if (error != 0) 172 return(error); 173 sx_xlock(&allproc_lock); 174 pid = randompid; 175 error = sysctl_handle_int(oidp, &pid, 0, req); 176 if (error == 0 && req->newptr != NULL) { 177 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 178 pid = PID_MAX - 100; 179 else if (pid < 2) /* NOP */ 180 pid = 0; 181 else if (pid < 100) /* Make it reasonable */ 182 pid = 100; 183 randompid = pid; 184 } 185 sx_xunlock(&allproc_lock); 186 return (error); 187 } 188 189 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 190 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 191 192 int 193 fork1(td, flags, pages, procp) 194 struct thread *td; 195 int flags; 196 int pages; 197 struct proc **procp; 198 { 199 struct proc *p1, *p2, *pptr; 200 struct proc *newproc; 201 int ok, trypid; 202 static int curfail, pidchecked = 0; 203 static struct timeval lastfail; 204 struct filedesc *fd; 205 struct filedesc_to_leader *fdtol; 206 struct thread *td2; 207 #ifdef KSE 208 struct ksegrp *kg2; 209 #endif 210 struct sigacts *newsigacts; 211 int error; 212 213 /* Can't copy and clear. */ 214 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 215 return (EINVAL); 216 217 p1 = td->td_proc; 218 219 /* 220 * Here we don't create a new process, but we divorce 221 * certain parts of a process from itself. 222 */ 223 if ((flags & RFPROC) == 0) { 224 if ((p1->p_flag & P_HADTHREADS) && 225 (flags & (RFCFDG | RFFDG))) { 226 PROC_LOCK(p1); 227 if (thread_single(SINGLE_BOUNDARY)) { 228 PROC_UNLOCK(p1); 229 return (ERESTART); 230 } 231 PROC_UNLOCK(p1); 232 } 233 234 vm_forkproc(td, NULL, NULL, flags); 235 236 /* 237 * Close all file descriptors. 238 */ 239 if (flags & RFCFDG) { 240 struct filedesc *fdtmp; 241 fdtmp = fdinit(td->td_proc->p_fd); 242 fdfree(td); 243 p1->p_fd = fdtmp; 244 } 245 246 /* 247 * Unshare file descriptors (from parent). 248 */ 249 if (flags & RFFDG) 250 fdunshare(p1, td); 251 252 if ((p1->p_flag & P_HADTHREADS) && 253 (flags & (RFCFDG | RFFDG))) { 254 PROC_LOCK(p1); 255 thread_single_end(); 256 PROC_UNLOCK(p1); 257 } 258 *procp = NULL; 259 return (0); 260 } 261 262 /* 263 * Note 1:1 allows for forking with one thread coming out on the 264 * other side with the expectation that the process is about to 265 * exec. 266 */ 267 if (p1->p_flag & P_HADTHREADS) { 268 /* 269 * Idle the other threads for a second. 270 * Since the user space is copied, it must remain stable. 271 * In addition, all threads (from the user perspective) 272 * need to either be suspended or in the kernel, 273 * where they will try restart in the parent and will 274 * be aborted in the child. 275 */ 276 PROC_LOCK(p1); 277 if (thread_single(SINGLE_NO_EXIT)) { 278 /* Abort. Someone else is single threading before us. */ 279 PROC_UNLOCK(p1); 280 return (ERESTART); 281 } 282 PROC_UNLOCK(p1); 283 /* 284 * All other activity in this process 285 * is now suspended at the user boundary, 286 * (or other safe places if we think of any). 287 */ 288 } 289 290 /* Allocate new proc. */ 291 newproc = uma_zalloc(proc_zone, M_WAITOK); 292 #ifdef MAC 293 mac_init_proc(newproc); 294 #endif 295 #ifdef AUDIT 296 audit_proc_alloc(newproc); 297 #endif 298 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 299 STAILQ_INIT(&newproc->p_ktr); 300 301 /* We have to lock the process tree while we look for a pid. */ 302 sx_slock(&proctree_lock); 303 304 /* 305 * Although process entries are dynamically created, we still keep 306 * a global limit on the maximum number we will create. Don't allow 307 * a nonprivileged user to use the last ten processes; don't let root 308 * exceed the limit. The variable nprocs is the current number of 309 * processes, maxproc is the limit. 310 */ 311 sx_xlock(&allproc_lock); 312 if ((nprocs >= maxproc - 10 && 313 suser_cred(td->td_ucred, SUSER_RUID) != 0) || 314 nprocs >= maxproc) { 315 error = EAGAIN; 316 goto fail; 317 } 318 319 /* 320 * Increment the count of procs running with this uid. Don't allow 321 * a nonprivileged user to exceed their current limit. 322 */ 323 error = suser_cred(td->td_ucred, SUSER_RUID | SUSER_ALLOWJAIL); 324 if (error == 0) 325 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 326 else { 327 PROC_LOCK(p1); 328 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 329 lim_cur(p1, RLIMIT_NPROC)); 330 PROC_UNLOCK(p1); 331 } 332 if (!ok) { 333 error = EAGAIN; 334 goto fail; 335 } 336 337 /* 338 * Increment the nprocs resource before blocking can occur. There 339 * are hard-limits as to the number of processes that can run. 340 */ 341 nprocs++; 342 343 /* 344 * Find an unused process ID. We remember a range of unused IDs 345 * ready to use (from lastpid+1 through pidchecked-1). 346 * 347 * If RFHIGHPID is set (used during system boot), do not allocate 348 * low-numbered pids. 349 */ 350 trypid = lastpid + 1; 351 if (flags & RFHIGHPID) { 352 if (trypid < 10) 353 trypid = 10; 354 } else { 355 if (randompid) 356 trypid += arc4random() % randompid; 357 } 358 retry: 359 /* 360 * If the process ID prototype has wrapped around, 361 * restart somewhat above 0, as the low-numbered procs 362 * tend to include daemons that don't exit. 363 */ 364 if (trypid >= PID_MAX) { 365 trypid = trypid % PID_MAX; 366 if (trypid < 100) 367 trypid += 100; 368 pidchecked = 0; 369 } 370 if (trypid >= pidchecked) { 371 int doingzomb = 0; 372 373 pidchecked = PID_MAX; 374 /* 375 * Scan the active and zombie procs to check whether this pid 376 * is in use. Remember the lowest pid that's greater 377 * than trypid, so we can avoid checking for a while. 378 */ 379 p2 = LIST_FIRST(&allproc); 380 again: 381 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 382 while (p2->p_pid == trypid || 383 (p2->p_pgrp != NULL && 384 (p2->p_pgrp->pg_id == trypid || 385 (p2->p_session != NULL && 386 p2->p_session->s_sid == trypid)))) { 387 trypid++; 388 if (trypid >= pidchecked) 389 goto retry; 390 } 391 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 392 pidchecked = p2->p_pid; 393 if (p2->p_pgrp != NULL) { 394 if (p2->p_pgrp->pg_id > trypid && 395 pidchecked > p2->p_pgrp->pg_id) 396 pidchecked = p2->p_pgrp->pg_id; 397 if (p2->p_session != NULL && 398 p2->p_session->s_sid > trypid && 399 pidchecked > p2->p_session->s_sid) 400 pidchecked = p2->p_session->s_sid; 401 } 402 } 403 if (!doingzomb) { 404 doingzomb = 1; 405 p2 = LIST_FIRST(&zombproc); 406 goto again; 407 } 408 } 409 sx_sunlock(&proctree_lock); 410 411 /* 412 * RFHIGHPID does not mess with the lastpid counter during boot. 413 */ 414 if (flags & RFHIGHPID) 415 pidchecked = 0; 416 else 417 lastpid = trypid; 418 419 p2 = newproc; 420 p2->p_state = PRS_NEW; /* protect against others */ 421 p2->p_pid = trypid; 422 AUDIT_ARG(pid, p2->p_pid); 423 LIST_INSERT_HEAD(&allproc, p2, p_list); 424 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 425 sx_xunlock(&allproc_lock); 426 427 /* 428 * Malloc things while we don't hold any locks. 429 */ 430 if (flags & RFSIGSHARE) 431 newsigacts = NULL; 432 else 433 newsigacts = sigacts_alloc(); 434 435 /* 436 * Copy filedesc. 437 */ 438 if (flags & RFCFDG) { 439 fd = fdinit(p1->p_fd); 440 fdtol = NULL; 441 } else if (flags & RFFDG) { 442 fd = fdcopy(p1->p_fd); 443 fdtol = NULL; 444 } else { 445 fd = fdshare(p1->p_fd); 446 if (p1->p_fdtol == NULL) 447 p1->p_fdtol = 448 filedesc_to_leader_alloc(NULL, 449 NULL, 450 p1->p_leader); 451 if ((flags & RFTHREAD) != 0) { 452 /* 453 * Shared file descriptor table and 454 * shared process leaders. 455 */ 456 fdtol = p1->p_fdtol; 457 FILEDESC_LOCK_FAST(p1->p_fd); 458 fdtol->fdl_refcount++; 459 FILEDESC_UNLOCK_FAST(p1->p_fd); 460 } else { 461 /* 462 * Shared file descriptor table, and 463 * different process leaders 464 */ 465 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 466 p1->p_fd, 467 p2); 468 } 469 } 470 /* 471 * Make a proc table entry for the new process. 472 * Start by zeroing the section of proc that is zero-initialized, 473 * then copy the section that is copied directly from the parent. 474 */ 475 td2 = FIRST_THREAD_IN_PROC(p2); 476 #ifdef KSE 477 kg2 = FIRST_KSEGRP_IN_PROC(p2); 478 #endif 479 480 /* Allocate and switch to an alternate kstack if specified. */ 481 if (pages != 0) 482 vm_thread_new_altkstack(td2, pages); 483 484 PROC_LOCK(p2); 485 PROC_LOCK(p1); 486 487 bzero(&p2->p_startzero, 488 __rangeof(struct proc, p_startzero, p_endzero)); 489 bzero(&td2->td_startzero, 490 __rangeof(struct thread, td_startzero, td_endzero)); 491 #ifdef KSE 492 bzero(&kg2->kg_startzero, 493 __rangeof(struct ksegrp, kg_startzero, kg_endzero)); 494 #endif 495 496 bcopy(&p1->p_startcopy, &p2->p_startcopy, 497 __rangeof(struct proc, p_startcopy, p_endcopy)); 498 bcopy(&td->td_startcopy, &td2->td_startcopy, 499 __rangeof(struct thread, td_startcopy, td_endcopy)); 500 #ifdef KSE 501 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 502 __rangeof(struct ksegrp, kg_startcopy, kg_endcopy)); 503 #endif 504 505 td2->td_sigstk = td->td_sigstk; 506 td2->td_sigmask = td->td_sigmask; 507 508 /* 509 * Duplicate sub-structures as needed. 510 * Increase reference counts on shared objects. 511 */ 512 p2->p_flag = 0; 513 if (p1->p_flag & P_PROFIL) 514 startprofclock(p2); 515 mtx_lock_spin(&sched_lock); 516 p2->p_sflag = PS_INMEM; 517 /* 518 * Allow the scheduler to adjust the priority of the child and 519 * parent while we hold the sched_lock. 520 */ 521 sched_fork(td, td2); 522 523 mtx_unlock_spin(&sched_lock); 524 p2->p_ucred = crhold(td->td_ucred); 525 #ifdef KSE 526 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 527 #else 528 td2->td_ucred = crhold(p2->p_ucred); 529 #endif 530 #ifdef AUDIT 531 audit_proc_fork(p1, p2); 532 #endif 533 pargs_hold(p2->p_args); 534 535 if (flags & RFSIGSHARE) { 536 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 537 } else { 538 sigacts_copy(newsigacts, p1->p_sigacts); 539 p2->p_sigacts = newsigacts; 540 } 541 if (flags & RFLINUXTHPN) 542 p2->p_sigparent = SIGUSR1; 543 else 544 p2->p_sigparent = SIGCHLD; 545 546 p2->p_textvp = p1->p_textvp; 547 p2->p_fd = fd; 548 p2->p_fdtol = fdtol; 549 550 /* 551 * p_limit is copy-on-write. Bump its refcount. 552 */ 553 p2->p_limit = lim_hold(p1->p_limit); 554 555 pstats_fork(p1->p_stats, p2->p_stats); 556 557 PROC_UNLOCK(p1); 558 PROC_UNLOCK(p2); 559 560 /* Bump references to the text vnode (for procfs) */ 561 if (p2->p_textvp) 562 vref(p2->p_textvp); 563 564 /* 565 * Set up linkage for kernel based threading. 566 */ 567 if ((flags & RFTHREAD) != 0) { 568 mtx_lock(&ppeers_lock); 569 p2->p_peers = p1->p_peers; 570 p1->p_peers = p2; 571 p2->p_leader = p1->p_leader; 572 mtx_unlock(&ppeers_lock); 573 PROC_LOCK(p1->p_leader); 574 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 575 PROC_UNLOCK(p1->p_leader); 576 /* 577 * The task leader is exiting, so process p1 is 578 * going to be killed shortly. Since p1 obviously 579 * isn't dead yet, we know that the leader is either 580 * sending SIGKILL's to all the processes in this 581 * task or is sleeping waiting for all the peers to 582 * exit. We let p1 complete the fork, but we need 583 * to go ahead and kill the new process p2 since 584 * the task leader may not get a chance to send 585 * SIGKILL to it. We leave it on the list so that 586 * the task leader will wait for this new process 587 * to commit suicide. 588 */ 589 PROC_LOCK(p2); 590 psignal(p2, SIGKILL); 591 PROC_UNLOCK(p2); 592 } else 593 PROC_UNLOCK(p1->p_leader); 594 } else { 595 p2->p_peers = NULL; 596 p2->p_leader = p2; 597 } 598 599 sx_xlock(&proctree_lock); 600 PGRP_LOCK(p1->p_pgrp); 601 PROC_LOCK(p2); 602 PROC_LOCK(p1); 603 604 /* 605 * Preserve some more flags in subprocess. P_PROFIL has already 606 * been preserved. 607 */ 608 p2->p_flag |= p1->p_flag & P_SUGID; 609 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 610 SESS_LOCK(p1->p_session); 611 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 612 p2->p_flag |= P_CONTROLT; 613 SESS_UNLOCK(p1->p_session); 614 if (flags & RFPPWAIT) 615 p2->p_flag |= P_PPWAIT; 616 617 p2->p_pgrp = p1->p_pgrp; 618 LIST_INSERT_AFTER(p1, p2, p_pglist); 619 PGRP_UNLOCK(p1->p_pgrp); 620 LIST_INIT(&p2->p_children); 621 622 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 623 624 #ifdef KTRACE 625 /* 626 * Copy traceflag and tracefile if enabled. 627 */ 628 mtx_lock(&ktrace_mtx); 629 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 630 if (p1->p_traceflag & KTRFAC_INHERIT) { 631 p2->p_traceflag = p1->p_traceflag; 632 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 633 VREF(p2->p_tracevp); 634 KASSERT(p1->p_tracecred != NULL, 635 ("ktrace vnode with no cred")); 636 p2->p_tracecred = crhold(p1->p_tracecred); 637 } 638 } 639 mtx_unlock(&ktrace_mtx); 640 #endif 641 642 /* 643 * If PF_FORK is set, the child process inherits the 644 * procfs ioctl flags from its parent. 645 */ 646 if (p1->p_pfsflags & PF_FORK) { 647 p2->p_stops = p1->p_stops; 648 p2->p_pfsflags = p1->p_pfsflags; 649 } 650 651 /* 652 * This begins the section where we must prevent the parent 653 * from being swapped. 654 */ 655 _PHOLD(p1); 656 PROC_UNLOCK(p1); 657 658 /* 659 * Attach the new process to its parent. 660 * 661 * If RFNOWAIT is set, the newly created process becomes a child 662 * of init. This effectively disassociates the child from the 663 * parent. 664 */ 665 if (flags & RFNOWAIT) 666 pptr = initproc; 667 else 668 pptr = p1; 669 p2->p_pptr = pptr; 670 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 671 sx_xunlock(&proctree_lock); 672 673 /* Inform accounting that we have forked. */ 674 p2->p_acflag = AFORK; 675 PROC_UNLOCK(p2); 676 677 /* 678 * Finish creating the child process. It will return via a different 679 * execution path later. (ie: directly into user mode) 680 */ 681 vm_forkproc(td, p2, td2, flags); 682 683 if (flags == (RFFDG | RFPROC)) { 684 atomic_add_int(&cnt.v_forks, 1); 685 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize + 686 p2->p_vmspace->vm_ssize); 687 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 688 atomic_add_int(&cnt.v_vforks, 1); 689 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 690 p2->p_vmspace->vm_ssize); 691 } else if (p1 == &proc0) { 692 atomic_add_int(&cnt.v_kthreads, 1); 693 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 694 p2->p_vmspace->vm_ssize); 695 } else { 696 atomic_add_int(&cnt.v_rforks, 1); 697 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 698 p2->p_vmspace->vm_ssize); 699 } 700 701 /* 702 * Both processes are set up, now check if any loadable modules want 703 * to adjust anything. 704 * What if they have an error? XXX 705 */ 706 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 707 708 /* 709 * Set the child start time and mark the process as being complete. 710 */ 711 microuptime(&p2->p_stats->p_start); 712 mtx_lock_spin(&sched_lock); 713 p2->p_state = PRS_NORMAL; 714 715 /* 716 * If RFSTOPPED not requested, make child runnable and add to 717 * run queue. 718 */ 719 if ((flags & RFSTOPPED) == 0) { 720 TD_SET_CAN_RUN(td2); 721 setrunqueue(td2, SRQ_BORING); 722 } 723 mtx_unlock_spin(&sched_lock); 724 725 /* 726 * Now can be swapped. 727 */ 728 PROC_LOCK(p1); 729 _PRELE(p1); 730 731 /* 732 * Tell any interested parties about the new process. 733 */ 734 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 735 736 PROC_UNLOCK(p1); 737 738 /* 739 * Preserve synchronization semantics of vfork. If waiting for 740 * child to exec or exit, set P_PPWAIT on child, and sleep on our 741 * proc (in case of exit). 742 */ 743 PROC_LOCK(p2); 744 while (p2->p_flag & P_PPWAIT) 745 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 746 PROC_UNLOCK(p2); 747 748 /* 749 * If other threads are waiting, let them continue now. 750 */ 751 if (p1->p_flag & P_HADTHREADS) { 752 PROC_LOCK(p1); 753 thread_single_end(); 754 PROC_UNLOCK(p1); 755 } 756 757 /* 758 * Return child proc pointer to parent. 759 */ 760 *procp = p2; 761 return (0); 762 fail: 763 sx_sunlock(&proctree_lock); 764 if (ppsratecheck(&lastfail, &curfail, 1)) 765 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 766 td->td_ucred->cr_ruid); 767 sx_xunlock(&allproc_lock); 768 #ifdef MAC 769 mac_destroy_proc(newproc); 770 #endif 771 #ifdef AUDIT 772 audit_proc_free(newproc); 773 #endif 774 uma_zfree(proc_zone, newproc); 775 if (p1->p_flag & P_HADTHREADS) { 776 PROC_LOCK(p1); 777 thread_single_end(); 778 PROC_UNLOCK(p1); 779 } 780 tsleep(&forksleep, PUSER, "fork", hz / 2); 781 return (error); 782 } 783 784 /* 785 * Handle the return of a child process from fork1(). This function 786 * is called from the MD fork_trampoline() entry point. 787 */ 788 void 789 fork_exit(callout, arg, frame) 790 void (*callout)(void *, struct trapframe *); 791 void *arg; 792 struct trapframe *frame; 793 { 794 struct proc *p; 795 struct thread *td; 796 797 /* 798 * Finish setting up thread glue so that it begins execution in a 799 * non-nested critical section with sched_lock held but not recursed. 800 */ 801 td = curthread; 802 p = td->td_proc; 803 td->td_oncpu = PCPU_GET(cpuid); 804 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 805 806 sched_lock.mtx_lock = (uintptr_t)td; 807 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 808 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 809 td, td->td_sched, p->p_pid, p->p_comm); 810 811 /* 812 * Processes normally resume in mi_switch() after being 813 * cpu_switch()'ed to, but when children start up they arrive here 814 * instead, so we must do much the same things as mi_switch() would. 815 */ 816 817 if ((td = PCPU_GET(deadthread))) { 818 PCPU_SET(deadthread, NULL); 819 thread_stash(td); 820 } 821 td = curthread; 822 mtx_unlock_spin(&sched_lock); 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 p->p_comm, p->p_pid); 839 kthread_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