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/mac.h> 63 #include <sys/ktr.h> 64 #include <sys/ktrace.h> 65 #include <sys/unistd.h> 66 #include <sys/sx.h> 67 #include <sys/signalvar.h> 68 69 #include <security/audit/audit.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 uid_t uid; 201 struct proc *newproc; 202 int ok, trypid; 203 static int curfail, pidchecked = 0; 204 static struct timeval lastfail; 205 struct filedesc *fd; 206 struct filedesc_to_leader *fdtol; 207 struct thread *td2; 208 struct ksegrp *kg2; 209 struct sigacts *newsigacts; 210 int error; 211 212 /* Can't copy and clear. */ 213 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 214 return (EINVAL); 215 216 p1 = td->td_proc; 217 218 /* 219 * Here we don't create a new process, but we divorce 220 * certain parts of a process from itself. 221 */ 222 if ((flags & RFPROC) == 0) { 223 vm_forkproc(td, NULL, NULL, flags); 224 225 /* 226 * Close all file descriptors. 227 */ 228 if (flags & RFCFDG) { 229 struct filedesc *fdtmp; 230 fdtmp = fdinit(td->td_proc->p_fd); 231 fdfree(td); 232 p1->p_fd = fdtmp; 233 } 234 235 /* 236 * Unshare file descriptors (from parent). 237 */ 238 if (flags & RFFDG) 239 fdunshare(p1, td); 240 *procp = NULL; 241 return (0); 242 } 243 244 /* 245 * Note 1:1 allows for forking with one thread coming out on the 246 * other side with the expectation that the process is about to 247 * exec. 248 */ 249 if (p1->p_flag & P_HADTHREADS) { 250 /* 251 * Idle the other threads for a second. 252 * Since the user space is copied, it must remain stable. 253 * In addition, all threads (from the user perspective) 254 * need to either be suspended or in the kernel, 255 * where they will try restart in the parent and will 256 * be aborted in the child. 257 */ 258 PROC_LOCK(p1); 259 if (thread_single(SINGLE_NO_EXIT)) { 260 /* Abort. Someone else is single threading before us. */ 261 PROC_UNLOCK(p1); 262 return (ERESTART); 263 } 264 PROC_UNLOCK(p1); 265 /* 266 * All other activity in this process 267 * is now suspended at the user boundary, 268 * (or other safe places if we think of any). 269 */ 270 } 271 272 /* Allocate new proc. */ 273 newproc = uma_zalloc(proc_zone, M_WAITOK); 274 #ifdef MAC 275 mac_init_proc(newproc); 276 #endif 277 #ifdef AUDIT 278 audit_proc_alloc(newproc); 279 #endif 280 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 281 STAILQ_INIT(&newproc->p_ktr); 282 283 /* We have to lock the process tree while we look for a pid. */ 284 sx_slock(&proctree_lock); 285 286 /* 287 * Although process entries are dynamically created, we still keep 288 * a global limit on the maximum number we will create. Don't allow 289 * a nonprivileged user to use the last ten processes; don't let root 290 * exceed the limit. The variable nprocs is the current number of 291 * processes, maxproc is the limit. 292 */ 293 sx_xlock(&allproc_lock); 294 uid = td->td_ucred->cr_ruid; 295 if ((nprocs >= maxproc - 10 && 296 suser_cred(td->td_ucred, SUSER_RUID) != 0) || 297 nprocs >= maxproc) { 298 error = EAGAIN; 299 goto fail; 300 } 301 302 /* 303 * Increment the count of procs running with this uid. Don't allow 304 * a nonprivileged user to exceed their current limit. 305 */ 306 PROC_LOCK(p1); 307 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 308 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0); 309 PROC_UNLOCK(p1); 310 if (!ok) { 311 error = EAGAIN; 312 goto fail; 313 } 314 315 /* 316 * Increment the nprocs resource before blocking can occur. There 317 * are hard-limits as to the number of processes that can run. 318 */ 319 nprocs++; 320 321 /* 322 * Find an unused process ID. We remember a range of unused IDs 323 * ready to use (from lastpid+1 through pidchecked-1). 324 * 325 * If RFHIGHPID is set (used during system boot), do not allocate 326 * low-numbered pids. 327 */ 328 trypid = lastpid + 1; 329 if (flags & RFHIGHPID) { 330 if (trypid < 10) 331 trypid = 10; 332 } else { 333 if (randompid) 334 trypid += arc4random() % randompid; 335 } 336 retry: 337 /* 338 * If the process ID prototype has wrapped around, 339 * restart somewhat above 0, as the low-numbered procs 340 * tend to include daemons that don't exit. 341 */ 342 if (trypid >= PID_MAX) { 343 trypid = trypid % PID_MAX; 344 if (trypid < 100) 345 trypid += 100; 346 pidchecked = 0; 347 } 348 if (trypid >= pidchecked) { 349 int doingzomb = 0; 350 351 pidchecked = PID_MAX; 352 /* 353 * Scan the active and zombie procs to check whether this pid 354 * is in use. Remember the lowest pid that's greater 355 * than trypid, so we can avoid checking for a while. 356 */ 357 p2 = LIST_FIRST(&allproc); 358 again: 359 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 360 PROC_LOCK(p2); 361 while (p2->p_pid == trypid || 362 (p2->p_pgrp != NULL && 363 (p2->p_pgrp->pg_id == trypid || 364 (p2->p_session != NULL && 365 p2->p_session->s_sid == trypid)))) { 366 trypid++; 367 if (trypid >= pidchecked) { 368 PROC_UNLOCK(p2); 369 goto retry; 370 } 371 } 372 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 373 pidchecked = p2->p_pid; 374 if (p2->p_pgrp != NULL) { 375 if (p2->p_pgrp->pg_id > trypid && 376 pidchecked > p2->p_pgrp->pg_id) 377 pidchecked = p2->p_pgrp->pg_id; 378 if (p2->p_session != NULL && 379 p2->p_session->s_sid > trypid && 380 pidchecked > p2->p_session->s_sid) 381 pidchecked = p2->p_session->s_sid; 382 } 383 PROC_UNLOCK(p2); 384 } 385 if (!doingzomb) { 386 doingzomb = 1; 387 p2 = LIST_FIRST(&zombproc); 388 goto again; 389 } 390 } 391 sx_sunlock(&proctree_lock); 392 393 /* 394 * RFHIGHPID does not mess with the lastpid counter during boot. 395 */ 396 if (flags & RFHIGHPID) 397 pidchecked = 0; 398 else 399 lastpid = trypid; 400 401 p2 = newproc; 402 p2->p_state = PRS_NEW; /* protect against others */ 403 p2->p_pid = trypid; 404 AUDIT_ARG(pid, p2->p_pid); 405 LIST_INSERT_HEAD(&allproc, p2, p_list); 406 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 407 sx_xunlock(&allproc_lock); 408 409 /* 410 * Malloc things while we don't hold any locks. 411 */ 412 if (flags & RFSIGSHARE) 413 newsigacts = NULL; 414 else 415 newsigacts = sigacts_alloc(); 416 417 /* 418 * Copy filedesc. 419 */ 420 if (flags & RFCFDG) { 421 fd = fdinit(p1->p_fd); 422 fdtol = NULL; 423 } else if (flags & RFFDG) { 424 fd = fdcopy(p1->p_fd); 425 fdtol = NULL; 426 } else { 427 fd = fdshare(p1->p_fd); 428 if (p1->p_fdtol == NULL) 429 p1->p_fdtol = 430 filedesc_to_leader_alloc(NULL, 431 NULL, 432 p1->p_leader); 433 if ((flags & RFTHREAD) != 0) { 434 /* 435 * Shared file descriptor table and 436 * shared process leaders. 437 */ 438 fdtol = p1->p_fdtol; 439 FILEDESC_LOCK_FAST(p1->p_fd); 440 fdtol->fdl_refcount++; 441 FILEDESC_UNLOCK_FAST(p1->p_fd); 442 } else { 443 /* 444 * Shared file descriptor table, and 445 * different process leaders 446 */ 447 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 448 p1->p_fd, 449 p2); 450 } 451 } 452 /* 453 * Make a proc table entry for the new process. 454 * Start by zeroing the section of proc that is zero-initialized, 455 * then copy the section that is copied directly from the parent. 456 */ 457 td2 = FIRST_THREAD_IN_PROC(p2); 458 kg2 = FIRST_KSEGRP_IN_PROC(p2); 459 460 /* Allocate and switch to an alternate kstack if specified. */ 461 if (pages != 0) 462 vm_thread_new_altkstack(td2, pages); 463 464 PROC_LOCK(p2); 465 PROC_LOCK(p1); 466 467 bzero(&p2->p_startzero, 468 __rangeof(struct proc, p_startzero, p_endzero)); 469 bzero(&td2->td_startzero, 470 __rangeof(struct thread, td_startzero, td_endzero)); 471 bzero(&kg2->kg_startzero, 472 __rangeof(struct ksegrp, kg_startzero, kg_endzero)); 473 474 bcopy(&p1->p_startcopy, &p2->p_startcopy, 475 __rangeof(struct proc, p_startcopy, p_endcopy)); 476 bcopy(&td->td_startcopy, &td2->td_startcopy, 477 __rangeof(struct thread, td_startcopy, td_endcopy)); 478 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 479 __rangeof(struct ksegrp, kg_startcopy, kg_endcopy)); 480 481 td2->td_sigstk = td->td_sigstk; 482 td2->td_sigmask = td->td_sigmask; 483 484 /* 485 * Duplicate sub-structures as needed. 486 * Increase reference counts on shared objects. 487 */ 488 p2->p_flag = 0; 489 if (p1->p_flag & P_PROFIL) 490 startprofclock(p2); 491 mtx_lock_spin(&sched_lock); 492 p2->p_sflag = PS_INMEM; 493 /* 494 * Allow the scheduler to adjust the priority of the child and 495 * parent while we hold the sched_lock. 496 */ 497 sched_fork(td, td2); 498 499 mtx_unlock_spin(&sched_lock); 500 p2->p_ucred = crhold(td->td_ucred); 501 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 502 #ifdef AUDIT 503 audit_proc_fork(p1, p2); 504 #endif 505 pargs_hold(p2->p_args); 506 507 if (flags & RFSIGSHARE) { 508 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 509 } else { 510 sigacts_copy(newsigacts, p1->p_sigacts); 511 p2->p_sigacts = newsigacts; 512 } 513 if (flags & RFLINUXTHPN) 514 p2->p_sigparent = SIGUSR1; 515 else 516 p2->p_sigparent = SIGCHLD; 517 518 p2->p_textvp = p1->p_textvp; 519 p2->p_fd = fd; 520 p2->p_fdtol = fdtol; 521 522 /* 523 * p_limit is copy-on-write. Bump its refcount. 524 */ 525 p2->p_limit = lim_hold(p1->p_limit); 526 527 pstats_fork(p1->p_stats, p2->p_stats); 528 529 PROC_UNLOCK(p1); 530 PROC_UNLOCK(p2); 531 532 /* Bump references to the text vnode (for procfs) */ 533 if (p2->p_textvp) 534 vref(p2->p_textvp); 535 536 /* 537 * Set up linkage for kernel based threading. 538 */ 539 if ((flags & RFTHREAD) != 0) { 540 mtx_lock(&ppeers_lock); 541 p2->p_peers = p1->p_peers; 542 p1->p_peers = p2; 543 p2->p_leader = p1->p_leader; 544 mtx_unlock(&ppeers_lock); 545 PROC_LOCK(p1->p_leader); 546 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 547 PROC_UNLOCK(p1->p_leader); 548 /* 549 * The task leader is exiting, so process p1 is 550 * going to be killed shortly. Since p1 obviously 551 * isn't dead yet, we know that the leader is either 552 * sending SIGKILL's to all the processes in this 553 * task or is sleeping waiting for all the peers to 554 * exit. We let p1 complete the fork, but we need 555 * to go ahead and kill the new process p2 since 556 * the task leader may not get a chance to send 557 * SIGKILL to it. We leave it on the list so that 558 * the task leader will wait for this new process 559 * to commit suicide. 560 */ 561 PROC_LOCK(p2); 562 psignal(p2, SIGKILL); 563 PROC_UNLOCK(p2); 564 } else 565 PROC_UNLOCK(p1->p_leader); 566 } else { 567 p2->p_peers = NULL; 568 p2->p_leader = p2; 569 } 570 571 sx_xlock(&proctree_lock); 572 PGRP_LOCK(p1->p_pgrp); 573 PROC_LOCK(p2); 574 PROC_LOCK(p1); 575 576 /* 577 * Preserve some more flags in subprocess. P_PROFIL has already 578 * been preserved. 579 */ 580 p2->p_flag |= p1->p_flag & P_SUGID; 581 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 582 SESS_LOCK(p1->p_session); 583 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 584 p2->p_flag |= P_CONTROLT; 585 SESS_UNLOCK(p1->p_session); 586 if (flags & RFPPWAIT) 587 p2->p_flag |= P_PPWAIT; 588 589 p2->p_pgrp = p1->p_pgrp; 590 LIST_INSERT_AFTER(p1, p2, p_pglist); 591 PGRP_UNLOCK(p1->p_pgrp); 592 LIST_INIT(&p2->p_children); 593 594 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 595 596 #ifdef KTRACE 597 /* 598 * Copy traceflag and tracefile if enabled. 599 */ 600 mtx_lock(&ktrace_mtx); 601 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 602 if (p1->p_traceflag & KTRFAC_INHERIT) { 603 p2->p_traceflag = p1->p_traceflag; 604 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 605 VREF(p2->p_tracevp); 606 KASSERT(p1->p_tracecred != NULL, 607 ("ktrace vnode with no cred")); 608 p2->p_tracecred = crhold(p1->p_tracecred); 609 } 610 } 611 mtx_unlock(&ktrace_mtx); 612 #endif 613 614 /* 615 * If PF_FORK is set, the child process inherits the 616 * procfs ioctl flags from its parent. 617 */ 618 if (p1->p_pfsflags & PF_FORK) { 619 p2->p_stops = p1->p_stops; 620 p2->p_pfsflags = p1->p_pfsflags; 621 } 622 623 /* 624 * This begins the section where we must prevent the parent 625 * from being swapped. 626 */ 627 _PHOLD(p1); 628 PROC_UNLOCK(p1); 629 630 /* 631 * Attach the new process to its parent. 632 * 633 * If RFNOWAIT is set, the newly created process becomes a child 634 * of init. This effectively disassociates the child from the 635 * parent. 636 */ 637 if (flags & RFNOWAIT) 638 pptr = initproc; 639 else 640 pptr = p1; 641 p2->p_pptr = pptr; 642 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 643 sx_xunlock(&proctree_lock); 644 645 /* Inform accounting that we have forked. */ 646 p2->p_acflag = AFORK; 647 PROC_UNLOCK(p2); 648 649 /* 650 * Finish creating the child process. It will return via a different 651 * execution path later. (ie: directly into user mode) 652 */ 653 vm_forkproc(td, p2, td2, flags); 654 655 if (flags == (RFFDG | RFPROC)) { 656 atomic_add_int(&cnt.v_forks, 1); 657 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize + 658 p2->p_vmspace->vm_ssize); 659 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 660 atomic_add_int(&cnt.v_vforks, 1); 661 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 662 p2->p_vmspace->vm_ssize); 663 } else if (p1 == &proc0) { 664 atomic_add_int(&cnt.v_kthreads, 1); 665 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 666 p2->p_vmspace->vm_ssize); 667 } else { 668 atomic_add_int(&cnt.v_rforks, 1); 669 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 670 p2->p_vmspace->vm_ssize); 671 } 672 673 /* 674 * Both processes are set up, now check if any loadable modules want 675 * to adjust anything. 676 * What if they have an error? XXX 677 */ 678 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 679 680 /* 681 * Set the child start time and mark the process as being complete. 682 */ 683 microuptime(&p2->p_stats->p_start); 684 mtx_lock_spin(&sched_lock); 685 p2->p_state = PRS_NORMAL; 686 687 /* 688 * If RFSTOPPED not requested, make child runnable and add to 689 * run queue. 690 */ 691 if ((flags & RFSTOPPED) == 0) { 692 TD_SET_CAN_RUN(td2); 693 setrunqueue(td2, SRQ_BORING); 694 } 695 mtx_unlock_spin(&sched_lock); 696 697 /* 698 * Now can be swapped. 699 */ 700 PROC_LOCK(p1); 701 _PRELE(p1); 702 703 /* 704 * Tell any interested parties about the new process. 705 */ 706 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 707 708 PROC_UNLOCK(p1); 709 710 /* 711 * Preserve synchronization semantics of vfork. If waiting for 712 * child to exec or exit, set P_PPWAIT on child, and sleep on our 713 * proc (in case of exit). 714 */ 715 PROC_LOCK(p2); 716 while (p2->p_flag & P_PPWAIT) 717 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 718 PROC_UNLOCK(p2); 719 720 /* 721 * If other threads are waiting, let them continue now. 722 */ 723 if (p1->p_flag & P_HADTHREADS) { 724 PROC_LOCK(p1); 725 thread_single_end(); 726 PROC_UNLOCK(p1); 727 } 728 729 /* 730 * Return child proc pointer to parent. 731 */ 732 *procp = p2; 733 return (0); 734 fail: 735 sx_sunlock(&proctree_lock); 736 if (ppsratecheck(&lastfail, &curfail, 1)) 737 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 738 uid); 739 sx_xunlock(&allproc_lock); 740 #ifdef MAC 741 mac_destroy_proc(newproc); 742 #endif 743 #ifdef AUDIT 744 audit_proc_free(newproc); 745 #endif 746 uma_zfree(proc_zone, newproc); 747 if (p1->p_flag & P_HADTHREADS) { 748 PROC_LOCK(p1); 749 thread_single_end(); 750 PROC_UNLOCK(p1); 751 } 752 tsleep(&forksleep, PUSER, "fork", hz / 2); 753 return (error); 754 } 755 756 /* 757 * Handle the return of a child process from fork1(). This function 758 * is called from the MD fork_trampoline() entry point. 759 */ 760 void 761 fork_exit(callout, arg, frame) 762 void (*callout)(void *, struct trapframe *); 763 void *arg; 764 struct trapframe *frame; 765 { 766 struct proc *p; 767 struct thread *td; 768 769 /* 770 * Finish setting up thread glue so that it begins execution in a 771 * non-nested critical section with sched_lock held but not recursed. 772 */ 773 td = curthread; 774 p = td->td_proc; 775 td->td_oncpu = PCPU_GET(cpuid); 776 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 777 778 sched_lock.mtx_lock = (uintptr_t)td; 779 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 780 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 781 td, td->td_sched, p->p_pid, p->p_comm); 782 783 /* 784 * Processes normally resume in mi_switch() after being 785 * cpu_switch()'ed to, but when children start up they arrive here 786 * instead, so we must do much the same things as mi_switch() would. 787 */ 788 789 if ((td = PCPU_GET(deadthread))) { 790 PCPU_SET(deadthread, NULL); 791 thread_stash(td); 792 } 793 td = curthread; 794 mtx_unlock_spin(&sched_lock); 795 796 /* 797 * cpu_set_fork_handler intercepts this function call to 798 * have this call a non-return function to stay in kernel mode. 799 * initproc has its own fork handler, but it does return. 800 */ 801 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 802 callout(arg, frame); 803 804 /* 805 * Check if a kernel thread misbehaved and returned from its main 806 * function. 807 */ 808 if (p->p_flag & P_KTHREAD) { 809 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 810 p->p_comm, p->p_pid); 811 kthread_exit(0); 812 } 813 mtx_assert(&Giant, MA_NOTOWNED); 814 } 815 816 /* 817 * Simplified back end of syscall(), used when returning from fork() 818 * directly into user mode. Giant is not held on entry, and must not 819 * be held on return. This function is passed in to fork_exit() as the 820 * first parameter and is called when returning to a new userland process. 821 */ 822 void 823 fork_return(td, frame) 824 struct thread *td; 825 struct trapframe *frame; 826 { 827 828 userret(td, frame); 829 #ifdef KTRACE 830 if (KTRPOINT(td, KTR_SYSRET)) 831 ktrsysret(SYS_fork, 0, 0); 832 #endif 833 mtx_assert(&Giant, MA_NOTOWNED); 834 } 835