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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 39 */ 40 41 #include <sys/cdefs.h> 42 __FBSDID("$FreeBSD$"); 43 44 #include "opt_ktrace.h" 45 #include "opt_mac.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/sysproto.h> 50 #include <sys/eventhandler.h> 51 #include <sys/filedesc.h> 52 #include <sys/kernel.h> 53 #include <sys/kthread.h> 54 #include <sys/sysctl.h> 55 #include <sys/lock.h> 56 #include <sys/malloc.h> 57 #include <sys/mutex.h> 58 #include <sys/proc.h> 59 #include <sys/pioctl.h> 60 #include <sys/resourcevar.h> 61 #include <sys/sched.h> 62 #include <sys/syscall.h> 63 #include <sys/vmmeter.h> 64 #include <sys/vnode.h> 65 #include <sys/acct.h> 66 #include <sys/mac.h> 67 #include <sys/ktr.h> 68 #include <sys/ktrace.h> 69 #include <sys/unistd.h> 70 #include <sys/jail.h> 71 #include <sys/sx.h> 72 73 #include <vm/vm.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_map.h> 76 #include <vm/vm_extern.h> 77 #include <vm/uma.h> 78 79 #include <sys/user.h> 80 #include <machine/critical.h> 81 82 #ifndef _SYS_SYSPROTO_H_ 83 struct fork_args { 84 int dummy; 85 }; 86 #endif 87 88 static int forksleep; /* Place for fork1() to sleep on. */ 89 90 /* 91 * MPSAFE 92 */ 93 /* ARGSUSED */ 94 int 95 fork(td, uap) 96 struct thread *td; 97 struct fork_args *uap; 98 { 99 int error; 100 struct proc *p2; 101 102 error = fork1(td, RFFDG | RFPROC, 0, &p2); 103 if (error == 0) { 104 td->td_retval[0] = p2->p_pid; 105 td->td_retval[1] = 0; 106 } 107 return (error); 108 } 109 110 /* 111 * MPSAFE 112 */ 113 /* ARGSUSED */ 114 int 115 vfork(td, uap) 116 struct thread *td; 117 struct vfork_args *uap; 118 { 119 int error; 120 struct proc *p2; 121 122 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 123 if (error == 0) { 124 td->td_retval[0] = p2->p_pid; 125 td->td_retval[1] = 0; 126 } 127 return (error); 128 } 129 130 /* 131 * MPSAFE 132 */ 133 int 134 rfork(td, uap) 135 struct thread *td; 136 struct rfork_args *uap; 137 { 138 int error; 139 struct proc *p2; 140 141 /* Don't allow kernel only flags. */ 142 if ((uap->flags & RFKERNELONLY) != 0) 143 return (EINVAL); 144 error = fork1(td, uap->flags, 0, &p2); 145 if (error == 0) { 146 td->td_retval[0] = p2 ? p2->p_pid : 0; 147 td->td_retval[1] = 0; 148 } 149 return (error); 150 } 151 152 int nprocs = 1; /* process 0 */ 153 int lastpid = 0; 154 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 155 "Last used PID"); 156 157 /* 158 * Random component to lastpid generation. We mix in a random factor to make 159 * it a little harder to predict. We sanity check the modulus value to avoid 160 * doing it in critical paths. Don't let it be too small or we pointlessly 161 * waste randomness entropy, and don't let it be impossibly large. Using a 162 * modulus that is too big causes a LOT more process table scans and slows 163 * down fork processing as the pidchecked caching is defeated. 164 */ 165 static int randompid = 0; 166 167 static int 168 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 169 { 170 int error, pid; 171 172 sysctl_wire_old_buffer(req, sizeof(int)); 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 kse *ke2; 209 struct ksegrp *kg2; 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 mtx_lock(&Giant); 219 220 /* 221 * Here we don't create a new process, but we divorce 222 * certain parts of a process from itself. 223 */ 224 if ((flags & RFPROC) == 0) { 225 vm_forkproc(td, NULL, NULL, flags); 226 227 /* 228 * Close all file descriptors. 229 */ 230 if (flags & RFCFDG) { 231 struct filedesc *fdtmp; 232 fdtmp = fdinit(td->td_proc->p_fd); 233 fdfree(td); 234 p1->p_fd = fdtmp; 235 } 236 237 /* 238 * Unshare file descriptors (from parent.) 239 */ 240 if (flags & RFFDG) { 241 FILEDESC_LOCK(p1->p_fd); 242 if (p1->p_fd->fd_refcnt > 1) { 243 struct filedesc *newfd; 244 245 newfd = fdcopy(td->td_proc->p_fd); 246 FILEDESC_UNLOCK(p1->p_fd); 247 fdfree(td); 248 p1->p_fd = newfd; 249 } else 250 FILEDESC_UNLOCK(p1->p_fd); 251 } 252 mtx_unlock(&Giant); 253 *procp = NULL; 254 return (0); 255 } 256 257 /* 258 * Note 1:1 allows for forking with one thread coming out on the 259 * other side with the expectation that the process is about to 260 * exec. 261 */ 262 if (p1->p_flag & P_SA) { 263 /* 264 * Idle the other threads for a second. 265 * Since the user space is copied, it must remain stable. 266 * In addition, all threads (from the user perspective) 267 * need to either be suspended or in the kernel, 268 * where they will try restart in the parent and will 269 * be aborted in the child. 270 */ 271 PROC_LOCK(p1); 272 if (thread_single(SINGLE_NO_EXIT)) { 273 /* Abort.. someone else is single threading before us */ 274 PROC_UNLOCK(p1); 275 mtx_unlock(&Giant); 276 return (ERESTART); 277 } 278 PROC_UNLOCK(p1); 279 /* 280 * All other activity in this process 281 * is now suspended at the user boundary, 282 * (or other safe places if we think of any). 283 */ 284 } 285 286 /* Allocate new proc. */ 287 newproc = uma_zalloc(proc_zone, M_WAITOK); 288 #ifdef MAC 289 mac_init_proc(newproc); 290 #endif 291 292 /* 293 * Although process entries are dynamically created, we still keep 294 * a global limit on the maximum number we will create. Don't allow 295 * a nonprivileged user to use the last ten processes; don't let root 296 * exceed the limit. The variable nprocs is the current number of 297 * processes, maxproc is the limit. 298 */ 299 sx_xlock(&allproc_lock); 300 uid = td->td_ucred->cr_ruid; 301 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 302 error = EAGAIN; 303 goto fail; 304 } 305 306 /* 307 * Increment the count of procs running with this uid. Don't allow 308 * a nonprivileged user to exceed their current limit. 309 */ 310 PROC_LOCK(p1); 311 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 312 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 313 PROC_UNLOCK(p1); 314 if (!ok) { 315 error = EAGAIN; 316 goto fail; 317 } 318 319 /* 320 * Increment the nprocs resource before blocking can occur. There 321 * are hard-limits as to the number of processes that can run. 322 */ 323 nprocs++; 324 325 /* 326 * Find an unused process ID. We remember a range of unused IDs 327 * ready to use (from lastpid+1 through pidchecked-1). 328 * 329 * If RFHIGHPID is set (used during system boot), do not allocate 330 * low-numbered pids. 331 */ 332 trypid = lastpid + 1; 333 if (flags & RFHIGHPID) { 334 if (trypid < 10) 335 trypid = 10; 336 } else { 337 if (randompid) 338 trypid += arc4random() % randompid; 339 } 340 retry: 341 /* 342 * If the process ID prototype has wrapped around, 343 * restart somewhat above 0, as the low-numbered procs 344 * tend to include daemons that don't exit. 345 */ 346 if (trypid >= PID_MAX) { 347 trypid = trypid % PID_MAX; 348 if (trypid < 100) 349 trypid += 100; 350 pidchecked = 0; 351 } 352 if (trypid >= pidchecked) { 353 int doingzomb = 0; 354 355 pidchecked = PID_MAX; 356 /* 357 * Scan the active and zombie procs to check whether this pid 358 * is in use. Remember the lowest pid that's greater 359 * than trypid, so we can avoid checking for a while. 360 */ 361 p2 = LIST_FIRST(&allproc); 362 again: 363 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 364 PROC_LOCK(p2); 365 while (p2->p_pid == trypid || 366 p2->p_pgrp->pg_id == trypid || 367 p2->p_session->s_sid == trypid) { 368 trypid++; 369 if (trypid >= pidchecked) { 370 PROC_UNLOCK(p2); 371 goto retry; 372 } 373 } 374 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 375 pidchecked = p2->p_pid; 376 if (p2->p_pgrp->pg_id > trypid && 377 pidchecked > p2->p_pgrp->pg_id) 378 pidchecked = p2->p_pgrp->pg_id; 379 if (p2->p_session->s_sid > trypid && 380 pidchecked > p2->p_session->s_sid) 381 pidchecked = p2->p_session->s_sid; 382 PROC_UNLOCK(p2); 383 } 384 if (!doingzomb) { 385 doingzomb = 1; 386 p2 = LIST_FIRST(&zombproc); 387 goto again; 388 } 389 } 390 391 /* 392 * RFHIGHPID does not mess with the lastpid counter during boot. 393 */ 394 if (flags & RFHIGHPID) 395 pidchecked = 0; 396 else 397 lastpid = trypid; 398 399 p2 = newproc; 400 p2->p_state = PRS_NEW; /* protect against others */ 401 p2->p_pid = trypid; 402 LIST_INSERT_HEAD(&allproc, p2, p_list); 403 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 404 sx_xunlock(&allproc_lock); 405 406 /* 407 * Malloc things while we don't hold any locks. 408 */ 409 if (flags & RFSIGSHARE) 410 newsigacts = NULL; 411 else 412 newsigacts = sigacts_alloc(); 413 414 /* 415 * Copy filedesc. 416 */ 417 if (flags & RFCFDG) { 418 fd = fdinit(td->td_proc->p_fd); 419 fdtol = NULL; 420 } else if (flags & RFFDG) { 421 FILEDESC_LOCK(p1->p_fd); 422 fd = fdcopy(td->td_proc->p_fd); 423 FILEDESC_UNLOCK(p1->p_fd); 424 fdtol = NULL; 425 } else { 426 fd = fdshare(p1->p_fd); 427 if (p1->p_fdtol == NULL) 428 p1->p_fdtol = 429 filedesc_to_leader_alloc(NULL, 430 NULL, 431 p1->p_leader); 432 if ((flags & RFTHREAD) != 0) { 433 /* 434 * Shared file descriptor table and 435 * shared process leaders. 436 */ 437 fdtol = p1->p_fdtol; 438 FILEDESC_LOCK(p1->p_fd); 439 fdtol->fdl_refcount++; 440 FILEDESC_UNLOCK(p1->p_fd); 441 } else { 442 /* 443 * Shared file descriptor table, and 444 * different process leaders 445 */ 446 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 447 p1->p_fd, 448 p2); 449 } 450 } 451 /* 452 * Make a proc table entry for the new process. 453 * Start by zeroing the section of proc that is zero-initialized, 454 * then copy the section that is copied directly from the parent. 455 */ 456 td2 = FIRST_THREAD_IN_PROC(p2); 457 kg2 = FIRST_KSEGRP_IN_PROC(p2); 458 ke2 = FIRST_KSE_IN_KSEGRP(kg2); 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 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 468 469 bzero(&p2->p_startzero, 470 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 471 bzero(&ke2->ke_startzero, 472 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero)); 473 bzero(&td2->td_startzero, 474 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 475 bzero(&kg2->kg_startzero, 476 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 477 478 bcopy(&p1->p_startcopy, &p2->p_startcopy, 479 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 480 bcopy(&td->td_startcopy, &td2->td_startcopy, 481 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 482 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 483 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 484 #undef RANGEOF 485 486 td2->td_sigstk = td->td_sigstk; 487 488 /* Set up the thread as an active thread (as if runnable). */ 489 ke2->ke_state = KES_THREAD; 490 ke2->ke_thread = td2; 491 td2->td_kse = ke2; 492 493 /* 494 * Duplicate sub-structures as needed. 495 * Increase reference counts on shared objects. 496 * The p_stats substruct is set in vm_forkproc. 497 */ 498 p2->p_flag = 0; 499 if (p1->p_flag & P_PROFIL) 500 startprofclock(p2); 501 mtx_lock_spin(&sched_lock); 502 p2->p_sflag = PS_INMEM; 503 /* 504 * Allow the scheduler to adjust the priority of the child and 505 * parent while we hold the sched_lock. 506 */ 507 sched_fork(p1, p2); 508 509 mtx_unlock_spin(&sched_lock); 510 p2->p_ucred = crhold(td->td_ucred); 511 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 512 513 pargs_hold(p2->p_args); 514 515 if (flags & RFSIGSHARE) { 516 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 517 } else { 518 sigacts_copy(newsigacts, p1->p_sigacts); 519 p2->p_sigacts = newsigacts; 520 } 521 if (flags & RFLINUXTHPN) 522 p2->p_sigparent = SIGUSR1; 523 else 524 p2->p_sigparent = SIGCHLD; 525 526 /* Bump references to the text vnode (for procfs) */ 527 p2->p_textvp = p1->p_textvp; 528 if (p2->p_textvp) 529 VREF(p2->p_textvp); 530 p2->p_fd = fd; 531 p2->p_fdtol = fdtol; 532 PROC_UNLOCK(p1); 533 PROC_UNLOCK(p2); 534 535 /* 536 * p_limit is copy-on-write, bump refcnt, 537 */ 538 p2->p_limit = p1->p_limit; 539 p2->p_limit->p_refcnt++; 540 541 /* 542 * Setup linkage for kernel based threading 543 */ 544 if((flags & RFTHREAD) != 0) { 545 mtx_lock(&ppeers_lock); 546 p2->p_peers = p1->p_peers; 547 p1->p_peers = p2; 548 p2->p_leader = p1->p_leader; 549 mtx_unlock(&ppeers_lock); 550 PROC_LOCK(p1->p_leader); 551 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 552 PROC_UNLOCK(p1->p_leader); 553 /* 554 * The task leader is exiting, so process p1 is 555 * going to be killed shortly. Since p1 obviously 556 * isn't dead yet, we know that the leader is either 557 * sending SIGKILL's to all the processes in this 558 * task or is sleeping waiting for all the peers to 559 * exit. We let p1 complete the fork, but we need 560 * to go ahead and kill the new process p2 since 561 * the task leader may not get a chance to send 562 * SIGKILL to it. We leave it on the list so that 563 * the task leader will wait for this new process 564 * to commit suicide. 565 */ 566 PROC_LOCK(p2); 567 psignal(p2, SIGKILL); 568 PROC_UNLOCK(p2); 569 } else 570 PROC_UNLOCK(p1->p_leader); 571 } else { 572 p2->p_peers = NULL; 573 p2->p_leader = p2; 574 } 575 576 sx_xlock(&proctree_lock); 577 PGRP_LOCK(p1->p_pgrp); 578 PROC_LOCK(p2); 579 PROC_LOCK(p1); 580 581 /* 582 * Preserve some more flags in subprocess. P_PROFIL has already 583 * been preserved. 584 */ 585 p2->p_flag |= p1->p_flag & P_SUGID; 586 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 587 SESS_LOCK(p1->p_session); 588 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 589 p2->p_flag |= P_CONTROLT; 590 SESS_UNLOCK(p1->p_session); 591 if (flags & RFPPWAIT) 592 p2->p_flag |= P_PPWAIT; 593 594 p2->p_pgrp = p1->p_pgrp; 595 LIST_INSERT_AFTER(p1, p2, p_pglist); 596 PGRP_UNLOCK(p1->p_pgrp); 597 LIST_INIT(&p2->p_children); 598 599 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 600 601 #ifdef KTRACE 602 /* 603 * Copy traceflag and tracefile if enabled. 604 */ 605 mtx_lock(&ktrace_mtx); 606 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 607 if (p1->p_traceflag & KTRFAC_INHERIT) { 608 p2->p_traceflag = p1->p_traceflag; 609 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 610 VREF(p2->p_tracevp); 611 KASSERT(p1->p_tracecred != NULL, 612 ("ktrace vnode with no cred")); 613 p2->p_tracecred = crhold(p1->p_tracecred); 614 } 615 } 616 mtx_unlock(&ktrace_mtx); 617 #endif 618 619 /* 620 * If PF_FORK is set, the child process inherits the 621 * procfs ioctl flags from its parent. 622 */ 623 if (p1->p_pfsflags & PF_FORK) { 624 p2->p_stops = p1->p_stops; 625 p2->p_pfsflags = p1->p_pfsflags; 626 } 627 628 /* 629 * This begins the section where we must prevent the parent 630 * from being swapped. 631 */ 632 _PHOLD(p1); 633 PROC_UNLOCK(p1); 634 635 /* 636 * Attach the new process to its parent. 637 * 638 * If RFNOWAIT is set, the newly created process becomes a child 639 * of init. This effectively disassociates the child from the 640 * parent. 641 */ 642 if (flags & RFNOWAIT) 643 pptr = initproc; 644 else 645 pptr = p1; 646 p2->p_pptr = pptr; 647 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 648 sx_xunlock(&proctree_lock); 649 650 /* Inform accounting that we have forked. */ 651 p2->p_acflag = AFORK; 652 PROC_UNLOCK(p2); 653 654 /* 655 * Finish creating the child process. It will return via a different 656 * execution path later. (ie: directly into user mode) 657 */ 658 vm_forkproc(td, p2, td2, flags); 659 660 if (flags == (RFFDG | RFPROC)) { 661 cnt.v_forks++; 662 cnt.v_forkpages += p2->p_vmspace->vm_dsize + 663 p2->p_vmspace->vm_ssize; 664 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 665 cnt.v_vforks++; 666 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 667 p2->p_vmspace->vm_ssize; 668 } else if (p1 == &proc0) { 669 cnt.v_kthreads++; 670 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 671 p2->p_vmspace->vm_ssize; 672 } else { 673 cnt.v_rforks++; 674 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 675 p2->p_vmspace->vm_ssize; 676 } 677 678 /* 679 * Both processes are set up, now check if any loadable modules want 680 * to adjust anything. 681 * What if they have an error? XXX 682 */ 683 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 684 685 /* 686 * If RFSTOPPED not requested, make child runnable and add to 687 * run queue. 688 */ 689 microuptime(&p2->p_stats->p_start); 690 if ((flags & RFSTOPPED) == 0) { 691 mtx_lock_spin(&sched_lock); 692 p2->p_state = PRS_NORMAL; 693 TD_SET_CAN_RUN(td2); 694 setrunqueue(td2); 695 mtx_unlock_spin(&sched_lock); 696 } 697 698 /* 699 * Now can be swapped. 700 */ 701 PROC_LOCK(p1); 702 _PRELE(p1); 703 704 /* 705 * Tell any interested parties about the new process. 706 */ 707 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 708 709 PROC_UNLOCK(p1); 710 711 /* 712 * Preserve synchronization semantics of vfork. If waiting for 713 * child to exec or exit, set P_PPWAIT on child, and sleep on our 714 * proc (in case of exit). 715 */ 716 PROC_LOCK(p2); 717 while (p2->p_flag & P_PPWAIT) 718 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 719 PROC_UNLOCK(p2); 720 721 /* 722 * If other threads are waiting, let them continue now 723 */ 724 if (p1->p_flag & P_SA) { 725 PROC_LOCK(p1); 726 thread_single_end(); 727 PROC_UNLOCK(p1); 728 } 729 730 /* 731 * Return child proc pointer to parent. 732 */ 733 mtx_unlock(&Giant); 734 *procp = p2; 735 return (0); 736 fail: 737 if (ppsratecheck(&lastfail, &curfail, 1)) 738 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 739 uid); 740 sx_xunlock(&allproc_lock); 741 uma_zfree(proc_zone, newproc); 742 if (p1->p_flag & P_SA) { 743 PROC_LOCK(p1); 744 thread_single_end(); 745 PROC_UNLOCK(p1); 746 } 747 tsleep(&forksleep, PUSER, "fork", hz / 2); 748 mtx_unlock(&Giant); 749 return (error); 750 } 751 752 /* 753 * Handle the return of a child process from fork1(). This function 754 * is called from the MD fork_trampoline() entry point. 755 */ 756 void 757 fork_exit(callout, arg, frame) 758 void (*callout)(void *, struct trapframe *); 759 void *arg; 760 struct trapframe *frame; 761 { 762 struct proc *p; 763 struct thread *td; 764 765 /* 766 * Processes normally resume in mi_switch() after being 767 * cpu_switch()'ed to, but when children start up they arrive here 768 * instead, so we must do much the same things as mi_switch() would. 769 */ 770 771 if ((td = PCPU_GET(deadthread))) { 772 PCPU_SET(deadthread, NULL); 773 thread_stash(td); 774 } 775 td = curthread; 776 p = td->td_proc; 777 td->td_oncpu = PCPU_GET(cpuid); 778 p->p_state = PRS_NORMAL; 779 780 /* 781 * Finish setting up thread glue so that it begins execution in a 782 * non-nested critical section with sched_lock held but not recursed. 783 */ 784 sched_lock.mtx_lock = (uintptr_t)td; 785 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 786 cpu_critical_fork_exit(); 787 CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid, 788 p->p_comm); 789 mtx_unlock_spin(&sched_lock); 790 791 /* 792 * cpu_set_fork_handler intercepts this function call to 793 * have this call a non-return function to stay in kernel mode. 794 * initproc has its own fork handler, but it does return. 795 */ 796 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 797 callout(arg, frame); 798 799 /* 800 * Check if a kernel thread misbehaved and returned from its main 801 * function. 802 */ 803 PROC_LOCK(p); 804 if (p->p_flag & P_KTHREAD) { 805 PROC_UNLOCK(p); 806 mtx_lock(&Giant); 807 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 808 p->p_comm, p->p_pid); 809 kthread_exit(0); 810 } 811 PROC_UNLOCK(p); 812 #ifdef DIAGNOSTIC 813 cred_free_thread(td); 814 #endif 815 mtx_assert(&Giant, MA_NOTOWNED); 816 } 817 818 /* 819 * Simplified back end of syscall(), used when returning from fork() 820 * directly into user mode. Giant is not held on entry, and must not 821 * be held on return. This function is passed in to fork_exit() as the 822 * first parameter and is called when returning to a new userland process. 823 */ 824 void 825 fork_return(td, frame) 826 struct thread *td; 827 struct trapframe *frame; 828 { 829 830 userret(td, frame, 0); 831 #ifdef KTRACE 832 if (KTRPOINT(td, KTR_SYSRET)) 833 ktrsysret(SYS_fork, 0, 0); 834 #endif 835 mtx_assert(&Giant, MA_NOTOWNED); 836 } 837