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 68 #include <vm/vm.h> 69 #include <vm/pmap.h> 70 #include <vm/vm_map.h> 71 #include <vm/vm_extern.h> 72 #include <vm/uma.h> 73 74 #include <sys/user.h> 75 #include <machine/critical.h> 76 77 #ifndef _SYS_SYSPROTO_H_ 78 struct fork_args { 79 int dummy; 80 }; 81 #endif 82 83 static int forksleep; /* Place for fork1() to sleep on. */ 84 85 /* 86 * MPSAFE 87 */ 88 /* ARGSUSED */ 89 int 90 fork(td, uap) 91 struct thread *td; 92 struct fork_args *uap; 93 { 94 int error; 95 struct proc *p2; 96 97 error = fork1(td, RFFDG | RFPROC, 0, &p2); 98 if (error == 0) { 99 td->td_retval[0] = p2->p_pid; 100 td->td_retval[1] = 0; 101 } 102 return (error); 103 } 104 105 /* 106 * MPSAFE 107 */ 108 /* ARGSUSED */ 109 int 110 vfork(td, uap) 111 struct thread *td; 112 struct vfork_args *uap; 113 { 114 int error; 115 struct proc *p2; 116 117 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 118 if (error == 0) { 119 td->td_retval[0] = p2->p_pid; 120 td->td_retval[1] = 0; 121 } 122 return (error); 123 } 124 125 /* 126 * MPSAFE 127 */ 128 int 129 rfork(td, uap) 130 struct thread *td; 131 struct rfork_args *uap; 132 { 133 struct proc *p2; 134 int error; 135 136 /* Don't allow kernel-only flags. */ 137 if ((uap->flags & RFKERNELONLY) != 0) 138 return (EINVAL); 139 140 error = fork1(td, uap->flags, 0, &p2); 141 if (error == 0) { 142 td->td_retval[0] = p2 ? p2->p_pid : 0; 143 td->td_retval[1] = 0; 144 } 145 return (error); 146 } 147 148 int nprocs = 1; /* process 0 */ 149 int lastpid = 0; 150 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 151 "Last used PID"); 152 153 /* 154 * Random component to lastpid generation. We mix in a random factor to make 155 * it a little harder to predict. We sanity check the modulus value to avoid 156 * doing it in critical paths. Don't let it be too small or we pointlessly 157 * waste randomness entropy, and don't let it be impossibly large. Using a 158 * modulus that is too big causes a LOT more process table scans and slows 159 * down fork processing as the pidchecked caching is defeated. 160 */ 161 static int randompid = 0; 162 163 static int 164 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 165 { 166 int error, pid; 167 168 error = sysctl_wire_old_buffer(req, sizeof(int)); 169 if (error != 0) 170 return(error); 171 sx_xlock(&allproc_lock); 172 pid = randompid; 173 error = sysctl_handle_int(oidp, &pid, 0, req); 174 if (error == 0 && req->newptr != NULL) { 175 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 176 pid = PID_MAX - 100; 177 else if (pid < 2) /* NOP */ 178 pid = 0; 179 else if (pid < 100) /* Make it reasonable */ 180 pid = 100; 181 randompid = pid; 182 } 183 sx_xunlock(&allproc_lock); 184 return (error); 185 } 186 187 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 188 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 189 190 int 191 fork1(td, flags, pages, procp) 192 struct thread *td; 193 int flags; 194 int pages; 195 struct proc **procp; 196 { 197 struct proc *p1, *p2, *pptr; 198 uid_t uid; 199 struct proc *newproc; 200 int ok, trypid; 201 static int curfail, pidchecked = 0; 202 static struct timeval lastfail; 203 struct filedesc *fd; 204 struct filedesc_to_leader *fdtol; 205 struct thread *td2; 206 struct kse *ke2; 207 struct ksegrp *kg2; 208 struct sigacts *newsigacts; 209 int error; 210 211 /* Can't copy and clear. */ 212 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 213 return (EINVAL); 214 215 p1 = td->td_proc; 216 217 /* 218 * Here we don't create a new process, but we divorce 219 * certain parts of a process from itself. 220 */ 221 if ((flags & RFPROC) == 0) { 222 mtx_lock(&Giant); 223 vm_forkproc(td, NULL, NULL, flags); 224 mtx_unlock(&Giant); 225 226 /* 227 * Close all file descriptors. 228 */ 229 if (flags & RFCFDG) { 230 struct filedesc *fdtmp; 231 FILEDESC_LOCK(td->td_proc->p_fd); 232 fdtmp = fdinit(td->td_proc->p_fd); 233 FILEDESC_UNLOCK(td->td_proc->p_fd); 234 fdfree(td); 235 p1->p_fd = fdtmp; 236 } 237 238 /* 239 * Unshare file descriptors (from parent). 240 */ 241 if (flags & RFFDG) { 242 FILEDESC_LOCK(p1->p_fd); 243 if (p1->p_fd->fd_refcnt > 1) { 244 struct filedesc *newfd; 245 246 newfd = fdcopy(td->td_proc->p_fd); 247 FILEDESC_UNLOCK(p1->p_fd); 248 fdfree(td); 249 p1->p_fd = newfd; 250 } else 251 FILEDESC_UNLOCK(p1->p_fd); 252 } 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 return (ERESTART); 276 } 277 PROC_UNLOCK(p1); 278 /* 279 * All other activity in this process 280 * is now suspended at the user boundary, 281 * (or other safe places if we think of any). 282 */ 283 } 284 285 /* Allocate new proc. */ 286 newproc = uma_zalloc(proc_zone, M_WAITOK); 287 #ifdef MAC 288 mac_init_proc(newproc); 289 #endif 290 knlist_init(&newproc->p_klist, &newproc->p_mtx); 291 292 /* We have to lock the process tree while we look for a pid. */ 293 sx_slock(&proctree_lock); 294 295 /* 296 * Although process entries are dynamically created, we still keep 297 * a global limit on the maximum number we will create. Don't allow 298 * a nonprivileged user to use the last ten processes; don't let root 299 * exceed the limit. The variable nprocs is the current number of 300 * processes, maxproc is the limit. 301 */ 302 sx_xlock(&allproc_lock); 303 uid = td->td_ucred->cr_ruid; 304 if ((nprocs >= maxproc - 10 && 305 suser_cred(td->td_ucred, SUSER_RUID) != 0) || 306 nprocs >= maxproc) { 307 error = EAGAIN; 308 goto fail; 309 } 310 311 /* 312 * Increment the count of procs running with this uid. Don't allow 313 * a nonprivileged user to exceed their current limit. 314 */ 315 PROC_LOCK(p1); 316 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 317 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0); 318 PROC_UNLOCK(p1); 319 if (!ok) { 320 error = EAGAIN; 321 goto fail; 322 } 323 324 /* 325 * Increment the nprocs resource before blocking can occur. There 326 * are hard-limits as to the number of processes that can run. 327 */ 328 nprocs++; 329 330 /* 331 * Find an unused process ID. We remember a range of unused IDs 332 * ready to use (from lastpid+1 through pidchecked-1). 333 * 334 * If RFHIGHPID is set (used during system boot), do not allocate 335 * low-numbered pids. 336 */ 337 trypid = lastpid + 1; 338 if (flags & RFHIGHPID) { 339 if (trypid < 10) 340 trypid = 10; 341 } else { 342 if (randompid) 343 trypid += arc4random() % randompid; 344 } 345 retry: 346 /* 347 * If the process ID prototype has wrapped around, 348 * restart somewhat above 0, as the low-numbered procs 349 * tend to include daemons that don't exit. 350 */ 351 if (trypid >= PID_MAX) { 352 trypid = trypid % PID_MAX; 353 if (trypid < 100) 354 trypid += 100; 355 pidchecked = 0; 356 } 357 if (trypid >= pidchecked) { 358 int doingzomb = 0; 359 360 pidchecked = PID_MAX; 361 /* 362 * Scan the active and zombie procs to check whether this pid 363 * is in use. Remember the lowest pid that's greater 364 * than trypid, so we can avoid checking for a while. 365 */ 366 p2 = LIST_FIRST(&allproc); 367 again: 368 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 369 PROC_LOCK(p2); 370 while (p2->p_pid == trypid || 371 (p2->p_pgrp != NULL && 372 (p2->p_pgrp->pg_id == trypid || 373 (p2->p_session != NULL && 374 p2->p_session->s_sid == trypid)))) { 375 trypid++; 376 if (trypid >= pidchecked) { 377 PROC_UNLOCK(p2); 378 goto retry; 379 } 380 } 381 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 382 pidchecked = p2->p_pid; 383 if (p2->p_pgrp != NULL) { 384 if (p2->p_pgrp->pg_id > trypid && 385 pidchecked > p2->p_pgrp->pg_id) 386 pidchecked = p2->p_pgrp->pg_id; 387 if (p2->p_session != NULL && 388 p2->p_session->s_sid > trypid && 389 pidchecked > p2->p_session->s_sid) 390 pidchecked = p2->p_session->s_sid; 391 } 392 PROC_UNLOCK(p2); 393 } 394 if (!doingzomb) { 395 doingzomb = 1; 396 p2 = LIST_FIRST(&zombproc); 397 goto again; 398 } 399 } 400 sx_sunlock(&proctree_lock); 401 402 /* 403 * RFHIGHPID does not mess with the lastpid counter during boot. 404 */ 405 if (flags & RFHIGHPID) 406 pidchecked = 0; 407 else 408 lastpid = trypid; 409 410 p2 = newproc; 411 p2->p_state = PRS_NEW; /* protect against others */ 412 p2->p_pid = trypid; 413 LIST_INSERT_HEAD(&allproc, p2, p_list); 414 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 415 sx_xunlock(&allproc_lock); 416 417 /* 418 * Malloc things while we don't hold any locks. 419 */ 420 if (flags & RFSIGSHARE) 421 newsigacts = NULL; 422 else 423 newsigacts = sigacts_alloc(); 424 425 /* 426 * Copy filedesc. 427 */ 428 if (flags & RFCFDG) { 429 FILEDESC_LOCK(td->td_proc->p_fd); 430 fd = fdinit(td->td_proc->p_fd); 431 FILEDESC_UNLOCK(td->td_proc->p_fd); 432 fdtol = NULL; 433 } else if (flags & RFFDG) { 434 FILEDESC_LOCK(p1->p_fd); 435 fd = fdcopy(td->td_proc->p_fd); 436 FILEDESC_UNLOCK(p1->p_fd); 437 fdtol = NULL; 438 } else { 439 fd = fdshare(p1->p_fd); 440 if (p1->p_fdtol == NULL) 441 p1->p_fdtol = 442 filedesc_to_leader_alloc(NULL, 443 NULL, 444 p1->p_leader); 445 if ((flags & RFTHREAD) != 0) { 446 /* 447 * Shared file descriptor table and 448 * shared process leaders. 449 */ 450 fdtol = p1->p_fdtol; 451 FILEDESC_LOCK(p1->p_fd); 452 fdtol->fdl_refcount++; 453 FILEDESC_UNLOCK(p1->p_fd); 454 } else { 455 /* 456 * Shared file descriptor table, and 457 * different process leaders 458 */ 459 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 460 p1->p_fd, 461 p2); 462 } 463 } 464 /* 465 * Make a proc table entry for the new process. 466 * Start by zeroing the section of proc that is zero-initialized, 467 * then copy the section that is copied directly from the parent. 468 */ 469 td2 = FIRST_THREAD_IN_PROC(p2); 470 kg2 = FIRST_KSEGRP_IN_PROC(p2); 471 ke2 = FIRST_KSE_IN_KSEGRP(kg2); 472 473 /* Allocate and switch to an alternate kstack if specified. */ 474 if (pages != 0) 475 vm_thread_new_altkstack(td2, pages); 476 477 PROC_LOCK(p2); 478 PROC_LOCK(p1); 479 480 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 481 482 bzero(&p2->p_startzero, 483 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 484 bzero(&ke2->ke_startzero, 485 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero)); 486 bzero(&td2->td_startzero, 487 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 488 bzero(&kg2->kg_startzero, 489 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 490 491 bcopy(&p1->p_startcopy, &p2->p_startcopy, 492 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 493 bcopy(&td->td_startcopy, &td2->td_startcopy, 494 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 495 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 496 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 497 #undef RANGEOF 498 499 td2->td_sigstk = td->td_sigstk; 500 501 /* Set up the thread as an active thread (as if runnable). */ 502 ke2->ke_state = KES_THREAD; 503 ke2->ke_thread = td2; 504 td2->td_kse = ke2; 505 506 /* 507 * Duplicate sub-structures as needed. 508 * Increase reference counts on shared objects. 509 * The p_stats substruct is set in vm_forkproc. 510 */ 511 p2->p_flag = 0; 512 if (p1->p_flag & P_PROFIL) 513 startprofclock(p2); 514 mtx_lock_spin(&sched_lock); 515 p2->p_sflag = PS_INMEM; 516 /* 517 * Allow the scheduler to adjust the priority of the child and 518 * parent while we hold the sched_lock. 519 */ 520 sched_fork(td, p2); 521 522 mtx_unlock_spin(&sched_lock); 523 p2->p_ucred = crhold(td->td_ucred); 524 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 525 526 pargs_hold(p2->p_args); 527 528 if (flags & RFSIGSHARE) { 529 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 530 } else { 531 sigacts_copy(newsigacts, p1->p_sigacts); 532 p2->p_sigacts = newsigacts; 533 } 534 if (flags & RFLINUXTHPN) 535 p2->p_sigparent = SIGUSR1; 536 else 537 p2->p_sigparent = SIGCHLD; 538 539 p2->p_textvp = p1->p_textvp; 540 p2->p_fd = fd; 541 p2->p_fdtol = fdtol; 542 543 /* 544 * p_limit is copy-on-write. Bump its refcount. 545 */ 546 p2->p_limit = lim_hold(p1->p_limit); 547 PROC_UNLOCK(p1); 548 PROC_UNLOCK(p2); 549 550 /* Bump references to the text vnode (for procfs) */ 551 if (p2->p_textvp) 552 vref(p2->p_textvp); 553 554 /* 555 * Set up linkage for kernel based threading. 556 */ 557 if ((flags & RFTHREAD) != 0) { 558 mtx_lock(&ppeers_lock); 559 p2->p_peers = p1->p_peers; 560 p1->p_peers = p2; 561 p2->p_leader = p1->p_leader; 562 mtx_unlock(&ppeers_lock); 563 PROC_LOCK(p1->p_leader); 564 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 565 PROC_UNLOCK(p1->p_leader); 566 /* 567 * The task leader is exiting, so process p1 is 568 * going to be killed shortly. Since p1 obviously 569 * isn't dead yet, we know that the leader is either 570 * sending SIGKILL's to all the processes in this 571 * task or is sleeping waiting for all the peers to 572 * exit. We let p1 complete the fork, but we need 573 * to go ahead and kill the new process p2 since 574 * the task leader may not get a chance to send 575 * SIGKILL to it. We leave it on the list so that 576 * the task leader will wait for this new process 577 * to commit suicide. 578 */ 579 PROC_LOCK(p2); 580 psignal(p2, SIGKILL); 581 PROC_UNLOCK(p2); 582 } else 583 PROC_UNLOCK(p1->p_leader); 584 } else { 585 p2->p_peers = NULL; 586 p2->p_leader = p2; 587 } 588 589 sx_xlock(&proctree_lock); 590 PGRP_LOCK(p1->p_pgrp); 591 PROC_LOCK(p2); 592 PROC_LOCK(p1); 593 594 /* 595 * Preserve some more flags in subprocess. P_PROFIL has already 596 * been preserved. 597 */ 598 p2->p_flag |= p1->p_flag & P_SUGID; 599 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 600 SESS_LOCK(p1->p_session); 601 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 602 p2->p_flag |= P_CONTROLT; 603 SESS_UNLOCK(p1->p_session); 604 if (flags & RFPPWAIT) 605 p2->p_flag |= P_PPWAIT; 606 607 p2->p_pgrp = p1->p_pgrp; 608 LIST_INSERT_AFTER(p1, p2, p_pglist); 609 PGRP_UNLOCK(p1->p_pgrp); 610 LIST_INIT(&p2->p_children); 611 612 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 613 614 #ifdef KTRACE 615 /* 616 * Copy traceflag and tracefile if enabled. 617 */ 618 mtx_lock(&ktrace_mtx); 619 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 620 if (p1->p_traceflag & KTRFAC_INHERIT) { 621 p2->p_traceflag = p1->p_traceflag; 622 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 623 VREF(p2->p_tracevp); 624 KASSERT(p1->p_tracecred != NULL, 625 ("ktrace vnode with no cred")); 626 p2->p_tracecred = crhold(p1->p_tracecred); 627 } 628 } 629 mtx_unlock(&ktrace_mtx); 630 #endif 631 632 /* 633 * If PF_FORK is set, the child process inherits the 634 * procfs ioctl flags from its parent. 635 */ 636 if (p1->p_pfsflags & PF_FORK) { 637 p2->p_stops = p1->p_stops; 638 p2->p_pfsflags = p1->p_pfsflags; 639 } 640 641 /* 642 * This begins the section where we must prevent the parent 643 * from being swapped. 644 */ 645 _PHOLD(p1); 646 PROC_UNLOCK(p1); 647 648 /* 649 * Attach the new process to its parent. 650 * 651 * If RFNOWAIT is set, the newly created process becomes a child 652 * of init. This effectively disassociates the child from the 653 * parent. 654 */ 655 if (flags & RFNOWAIT) 656 pptr = initproc; 657 else 658 pptr = p1; 659 p2->p_pptr = pptr; 660 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 661 sx_xunlock(&proctree_lock); 662 663 /* Inform accounting that we have forked. */ 664 p2->p_acflag = AFORK; 665 PROC_UNLOCK(p2); 666 667 /* 668 * Finish creating the child process. It will return via a different 669 * execution path later. (ie: directly into user mode) 670 */ 671 mtx_lock(&Giant); 672 vm_forkproc(td, p2, td2, flags); 673 674 if (flags == (RFFDG | RFPROC)) { 675 cnt.v_forks++; 676 cnt.v_forkpages += p2->p_vmspace->vm_dsize + 677 p2->p_vmspace->vm_ssize; 678 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 679 cnt.v_vforks++; 680 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 681 p2->p_vmspace->vm_ssize; 682 } else if (p1 == &proc0) { 683 cnt.v_kthreads++; 684 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 685 p2->p_vmspace->vm_ssize; 686 } else { 687 cnt.v_rforks++; 688 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 689 p2->p_vmspace->vm_ssize; 690 } 691 mtx_unlock(&Giant); 692 693 /* 694 * Both processes are set up, now check if any loadable modules want 695 * to adjust anything. 696 * What if they have an error? XXX 697 */ 698 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 699 700 /* 701 * Set the child start time and mark the process as being complete. 702 */ 703 microuptime(&p2->p_stats->p_start); 704 mtx_lock_spin(&sched_lock); 705 p2->p_state = PRS_NORMAL; 706 707 /* 708 * If RFSTOPPED not requested, make child runnable and add to 709 * run queue. 710 */ 711 if ((flags & RFSTOPPED) == 0) { 712 TD_SET_CAN_RUN(td2); 713 setrunqueue(td2); 714 } 715 mtx_unlock_spin(&sched_lock); 716 717 /* 718 * Now can be swapped. 719 */ 720 PROC_LOCK(p1); 721 _PRELE(p1); 722 723 /* 724 * Tell any interested parties about the new process. 725 */ 726 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 727 728 PROC_UNLOCK(p1); 729 730 /* 731 * Preserve synchronization semantics of vfork. If waiting for 732 * child to exec or exit, set P_PPWAIT on child, and sleep on our 733 * proc (in case of exit). 734 */ 735 PROC_LOCK(p2); 736 while (p2->p_flag & P_PPWAIT) 737 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 738 PROC_UNLOCK(p2); 739 740 /* 741 * If other threads are waiting, let them continue now. 742 */ 743 if (p1->p_flag & P_SA) { 744 PROC_LOCK(p1); 745 thread_single_end(); 746 PROC_UNLOCK(p1); 747 } 748 749 /* 750 * Return child proc pointer to parent. 751 */ 752 *procp = p2; 753 return (0); 754 fail: 755 sx_sunlock(&proctree_lock); 756 if (ppsratecheck(&lastfail, &curfail, 1)) 757 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 758 uid); 759 sx_xunlock(&allproc_lock); 760 #ifdef MAC 761 mac_destroy_proc(newproc); 762 #endif 763 uma_zfree(proc_zone, newproc); 764 if (p1->p_flag & P_SA) { 765 PROC_LOCK(p1); 766 thread_single_end(); 767 PROC_UNLOCK(p1); 768 } 769 tsleep(&forksleep, PUSER, "fork", hz / 2); 770 return (error); 771 } 772 773 /* 774 * Handle the return of a child process from fork1(). This function 775 * is called from the MD fork_trampoline() entry point. 776 */ 777 void 778 fork_exit(callout, arg, frame) 779 void (*callout)(void *, struct trapframe *); 780 void *arg; 781 struct trapframe *frame; 782 { 783 struct proc *p; 784 struct thread *td; 785 786 /* 787 * Finish setting up thread glue so that it begins execution in a 788 * non-nested critical section with sched_lock held but not recursed. 789 */ 790 td = curthread; 791 p = td->td_proc; 792 td->td_oncpu = PCPU_GET(cpuid); 793 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 794 795 sched_lock.mtx_lock = (uintptr_t)td; 796 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 797 cpu_critical_fork_exit(); 798 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 799 td, td->td_kse, p->p_pid, p->p_comm); 800 801 /* 802 * Processes normally resume in mi_switch() after being 803 * cpu_switch()'ed to, but when children start up they arrive here 804 * instead, so we must do much the same things as mi_switch() would. 805 */ 806 807 if ((td = PCPU_GET(deadthread))) { 808 PCPU_SET(deadthread, NULL); 809 thread_stash(td); 810 } 811 td = curthread; 812 mtx_unlock_spin(&sched_lock); 813 814 /* 815 * cpu_set_fork_handler intercepts this function call to 816 * have this call a non-return function to stay in kernel mode. 817 * initproc has its own fork handler, but it does return. 818 */ 819 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 820 callout(arg, frame); 821 822 /* 823 * Check if a kernel thread misbehaved and returned from its main 824 * function. 825 */ 826 PROC_LOCK(p); 827 if (p->p_flag & P_KTHREAD) { 828 PROC_UNLOCK(p); 829 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 830 p->p_comm, p->p_pid); 831 kthread_exit(0); 832 } 833 PROC_UNLOCK(p); 834 #ifdef DIAGNOSTIC 835 cred_free_thread(td); 836 #endif 837 mtx_assert(&Giant, MA_NOTOWNED); 838 } 839 840 /* 841 * Simplified back end of syscall(), used when returning from fork() 842 * directly into user mode. Giant is not held on entry, and must not 843 * be held on return. This function is passed in to fork_exit() as the 844 * first parameter and is called when returning to a new userland process. 845 */ 846 void 847 fork_return(td, frame) 848 struct thread *td; 849 struct trapframe *frame; 850 { 851 852 userret(td, frame, 0); 853 #ifdef KTRACE 854 if (KTRPOINT(td, KTR_SYSRET)) 855 ktrsysret(SYS_fork, 0, 0); 856 #endif 857 mtx_assert(&Giant, MA_NOTOWNED); 858 } 859