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 * $FreeBSD$ 40 */ 41 42 #include "opt_ktrace.h" 43 #include "opt_mac.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/sysproto.h> 48 #include <sys/eventhandler.h> 49 #include <sys/filedesc.h> 50 #include <sys/kernel.h> 51 #include <sys/sysctl.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mutex.h> 55 #include <sys/proc.h> 56 #include <sys/pioctl.h> 57 #include <sys/resourcevar.h> 58 #include <sys/sched.h> 59 #include <sys/syscall.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/kthread.h> 66 #include <sys/unistd.h> 67 #include <sys/jail.h> 68 #include <sys/sx.h> 69 70 #include <vm/vm.h> 71 #include <vm/pmap.h> 72 #include <vm/vm_map.h> 73 #include <vm/vm_extern.h> 74 #include <vm/uma.h> 75 76 #include <sys/vmmeter.h> 77 #include <sys/user.h> 78 #include <machine/critical.h> 79 80 #ifndef _SYS_SYSPROTO_H_ 81 struct fork_args { 82 int dummy; 83 }; 84 #endif 85 86 int forksleep; /* Place for fork1() to sleep on. */ 87 88 /* 89 * MPSAFE 90 */ 91 /* ARGSUSED */ 92 int 93 fork(td, uap) 94 struct thread *td; 95 struct fork_args *uap; 96 { 97 int error; 98 struct proc *p2; 99 100 mtx_lock(&Giant); 101 error = fork1(td, RFFDG | RFPROC, 0, &p2); 102 if (error == 0) { 103 td->td_retval[0] = p2->p_pid; 104 td->td_retval[1] = 0; 105 } 106 mtx_unlock(&Giant); 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 mtx_lock(&Giant); 123 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 124 if (error == 0) { 125 td->td_retval[0] = p2->p_pid; 126 td->td_retval[1] = 0; 127 } 128 mtx_unlock(&Giant); 129 return error; 130 } 131 132 /* 133 * MPSAFE 134 */ 135 int 136 rfork(td, uap) 137 struct thread *td; 138 struct rfork_args *uap; 139 { 140 int error; 141 struct proc *p2; 142 143 /* Don't allow kernel only flags. */ 144 if ((uap->flags & RFKERNELONLY) != 0) 145 return (EINVAL); 146 /* 147 * Don't allow sharing of file descriptor table unless 148 * RFTHREAD flag is supplied 149 */ 150 if ((uap->flags & (RFPROC | RFTHREAD | RFFDG | RFCFDG)) == 151 RFPROC) 152 return(EINVAL); 153 mtx_lock(&Giant); 154 error = fork1(td, uap->flags, 0, &p2); 155 if (error == 0) { 156 td->td_retval[0] = p2 ? p2->p_pid : 0; 157 td->td_retval[1] = 0; 158 } 159 mtx_unlock(&Giant); 160 return error; 161 } 162 163 164 int nprocs = 1; /* process 0 */ 165 int lastpid = 0; 166 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 167 "Last used PID"); 168 169 /* 170 * Random component to lastpid generation. We mix in a random factor to make 171 * it a little harder to predict. We sanity check the modulus value to avoid 172 * doing it in critical paths. Don't let it be too small or we pointlessly 173 * waste randomness entropy, and don't let it be impossibly large. Using a 174 * modulus that is too big causes a LOT more process table scans and slows 175 * down fork processing as the pidchecked caching is defeated. 176 */ 177 static int randompid = 0; 178 179 static int 180 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 181 { 182 int error, pid; 183 184 sysctl_wire_old_buffer(req, sizeof(int)); 185 sx_xlock(&allproc_lock); 186 pid = randompid; 187 error = sysctl_handle_int(oidp, &pid, 0, req); 188 if (error == 0 && req->newptr != NULL) { 189 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 190 pid = PID_MAX - 100; 191 else if (pid < 2) /* NOP */ 192 pid = 0; 193 else if (pid < 100) /* Make it reasonable */ 194 pid = 100; 195 randompid = pid; 196 } 197 sx_xunlock(&allproc_lock); 198 return (error); 199 } 200 201 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 202 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 203 204 int 205 fork1(td, flags, pages, procp) 206 struct thread *td; /* parent proc */ 207 int flags; 208 int pages; 209 struct proc **procp; /* child proc */ 210 { 211 struct proc *p2, *pptr; 212 uid_t uid; 213 struct proc *newproc; 214 int trypid; 215 int ok; 216 static int pidchecked = 0; 217 struct filedesc *fd; 218 struct proc *p1 = td->td_proc; 219 struct thread *td2; 220 struct kse *ke2; 221 struct ksegrp *kg2; 222 struct sigacts *newsigacts; 223 struct procsig *newprocsig; 224 int error; 225 226 GIANT_REQUIRED; 227 228 /* Can't copy and clear */ 229 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 230 return (EINVAL); 231 232 /* 233 * Here we don't create a new process, but we divorce 234 * certain parts of a process from itself. 235 */ 236 if ((flags & RFPROC) == 0) { 237 vm_forkproc(td, NULL, NULL, flags); 238 239 /* 240 * Close all file descriptors. 241 */ 242 if (flags & RFCFDG) { 243 struct filedesc *fdtmp; 244 fdtmp = fdinit(td->td_proc->p_fd); 245 fdfree(td); 246 p1->p_fd = fdtmp; 247 } 248 249 /* 250 * Unshare file descriptors (from parent.) 251 */ 252 if (flags & RFFDG) { 253 FILEDESC_LOCK(p1->p_fd); 254 if (p1->p_fd->fd_refcnt > 1) { 255 struct filedesc *newfd; 256 257 newfd = fdcopy(td->td_proc->p_fd); 258 FILEDESC_UNLOCK(p1->p_fd); 259 fdfree(td); 260 p1->p_fd = newfd; 261 } else 262 FILEDESC_UNLOCK(p1->p_fd); 263 } 264 *procp = NULL; 265 return (0); 266 } 267 268 /* 269 * Note 1:1 allows for forking with one thread coming out on the 270 * other side with the expectation that the process is about to 271 * exec. 272 */ 273 if (p1->p_flag & P_THREADED) { 274 /* 275 * Idle the other threads for a second. 276 * Since the user space is copied, it must remain stable. 277 * In addition, all threads (from the user perspective) 278 * need to either be suspended or in the kernel, 279 * where they will try restart in the parent and will 280 * be aborted in the child. 281 */ 282 PROC_LOCK(p1); 283 if (thread_single(SINGLE_NO_EXIT)) { 284 /* Abort.. someone else is single threading before us */ 285 PROC_UNLOCK(p1); 286 return (ERESTART); 287 } 288 PROC_UNLOCK(p1); 289 /* 290 * All other activity in this process 291 * is now suspended at the user boundary, 292 * (or other safe places if we think of any). 293 */ 294 } 295 296 /* Allocate new proc. */ 297 newproc = uma_zalloc(proc_zone, M_WAITOK); 298 #ifdef MAC 299 mac_init_proc(newproc); 300 #endif 301 302 /* 303 * Although process entries are dynamically created, we still keep 304 * a global limit on the maximum number we will create. Don't allow 305 * a nonprivileged user to use the last ten processes; don't let root 306 * exceed the limit. The variable nprocs is the current number of 307 * processes, maxproc is the limit. 308 */ 309 sx_xlock(&allproc_lock); 310 uid = td->td_ucred->cr_ruid; 311 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 312 error = EAGAIN; 313 goto fail; 314 } 315 316 /* 317 * Increment the count of procs running with this uid. Don't allow 318 * a nonprivileged user to exceed their current limit. 319 */ 320 PROC_LOCK(p1); 321 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 322 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 323 PROC_UNLOCK(p1); 324 if (!ok) { 325 error = EAGAIN; 326 goto fail; 327 } 328 329 /* 330 * Increment the nprocs resource before blocking can occur. There 331 * are hard-limits as to the number of processes that can run. 332 */ 333 nprocs++; 334 335 /* 336 * Find an unused process ID. We remember a range of unused IDs 337 * ready to use (from lastpid+1 through pidchecked-1). 338 * 339 * If RFHIGHPID is set (used during system boot), do not allocate 340 * low-numbered pids. 341 */ 342 trypid = lastpid + 1; 343 if (flags & RFHIGHPID) { 344 if (trypid < 10) { 345 trypid = 10; 346 } 347 } else { 348 if (randompid) 349 trypid += arc4random() % randompid; 350 } 351 retry: 352 /* 353 * If the process ID prototype has wrapped around, 354 * restart somewhat above 0, as the low-numbered procs 355 * tend to include daemons that don't exit. 356 */ 357 if (trypid >= PID_MAX) { 358 trypid = trypid % PID_MAX; 359 if (trypid < 100) 360 trypid += 100; 361 pidchecked = 0; 362 } 363 if (trypid >= pidchecked) { 364 int doingzomb = 0; 365 366 pidchecked = PID_MAX; 367 /* 368 * Scan the active and zombie procs to check whether this pid 369 * is in use. Remember the lowest pid that's greater 370 * than trypid, so we can avoid checking for a while. 371 */ 372 p2 = LIST_FIRST(&allproc); 373 again: 374 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 375 PROC_LOCK(p2); 376 while (p2->p_pid == trypid || 377 p2->p_pgrp->pg_id == trypid || 378 p2->p_session->s_sid == trypid) { 379 trypid++; 380 if (trypid >= pidchecked) { 381 PROC_UNLOCK(p2); 382 goto retry; 383 } 384 } 385 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 386 pidchecked = p2->p_pid; 387 if (p2->p_pgrp->pg_id > trypid && 388 pidchecked > p2->p_pgrp->pg_id) 389 pidchecked = p2->p_pgrp->pg_id; 390 if (p2->p_session->s_sid > trypid && 391 pidchecked > p2->p_session->s_sid) 392 pidchecked = p2->p_session->s_sid; 393 PROC_UNLOCK(p2); 394 } 395 if (!doingzomb) { 396 doingzomb = 1; 397 p2 = LIST_FIRST(&zombproc); 398 goto again; 399 } 400 } 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 MALLOC(newsigacts, struct sigacts *, 422 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 423 newprocsig = NULL; 424 } else { 425 newsigacts = NULL; 426 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig), 427 M_SUBPROC, M_WAITOK); 428 } 429 430 /* 431 * Copy filedesc. 432 * XXX: This is busted. fd*() need to not take proc 433 * arguments or something. 434 */ 435 if (flags & RFCFDG) 436 fd = fdinit(td->td_proc->p_fd); 437 else if (flags & RFFDG) { 438 FILEDESC_LOCK(p1->p_fd); 439 fd = fdcopy(td->td_proc->p_fd); 440 FILEDESC_UNLOCK(p1->p_fd); 441 } else 442 fd = fdshare(p1->p_fd); 443 444 /* 445 * Make a proc table entry for the new process. 446 * Start by zeroing the section of proc that is zero-initialized, 447 * then copy the section that is copied directly from the parent. 448 */ 449 td2 = FIRST_THREAD_IN_PROC(p2); 450 kg2 = FIRST_KSEGRP_IN_PROC(p2); 451 ke2 = FIRST_KSE_IN_KSEGRP(kg2); 452 453 /* Allocate and switch to an alternate kstack if specified */ 454 if (pages != 0) 455 pmap_new_altkstack(td2, pages); 456 457 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 458 459 bzero(&p2->p_startzero, 460 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 461 bzero(&ke2->ke_startzero, 462 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero)); 463 bzero(&td2->td_startzero, 464 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 465 bzero(&kg2->kg_startzero, 466 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 467 468 mtx_init(&p2->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK); 469 PROC_LOCK(p2); 470 PROC_LOCK(p1); 471 472 bcopy(&p1->p_startcopy, &p2->p_startcopy, 473 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 474 bcopy(&td->td_startcopy, &td2->td_startcopy, 475 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 476 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 477 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 478 #undef RANGEOF 479 480 /* Set up the thread as an active thread (as if runnable). */ 481 ke2->ke_state = KES_THREAD; 482 ke2->ke_thread = td2; 483 td2->td_kse = ke2; 484 485 /* 486 * Duplicate sub-structures as needed. 487 * Increase reference counts on shared objects. 488 * The p_stats and p_sigacts substructs are set in vm_forkproc. 489 */ 490 p2->p_flag = 0; 491 mtx_lock_spin(&sched_lock); 492 p2->p_sflag = PS_INMEM; 493 if (p1->p_sflag & PS_PROFIL) 494 startprofclock(p2); 495 /* 496 * Allow the scheduler to adjust the priority of the child and 497 * parent while we hold the sched_lock. 498 */ 499 sched_fork(td->td_ksegrp, kg2); 500 501 mtx_unlock_spin(&sched_lock); 502 p2->p_ucred = crhold(td->td_ucred); 503 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 504 505 pargs_hold(p2->p_args); 506 507 if (flags & RFSIGSHARE) { 508 p2->p_procsig = p1->p_procsig; 509 p2->p_procsig->ps_refcnt++; 510 if (p1->p_sigacts == &p1->p_uarea->u_sigacts) { 511 /* 512 * Set p_sigacts to the new shared structure. 513 * Note that this is updating p1->p_sigacts at the 514 * same time, since p_sigacts is just a pointer to 515 * the shared p_procsig->ps_sigacts. 516 */ 517 p2->p_sigacts = newsigacts; 518 newsigacts = NULL; 519 *p2->p_sigacts = p1->p_uarea->u_sigacts; 520 } 521 } else { 522 p2->p_procsig = newprocsig; 523 newprocsig = NULL; 524 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 525 p2->p_procsig->ps_refcnt = 1; 526 p2->p_sigacts = NULL; /* finished in vm_forkproc() */ 527 } 528 if (flags & RFLINUXTHPN) 529 p2->p_sigparent = SIGUSR1; 530 else 531 p2->p_sigparent = SIGCHLD; 532 533 /* Bump references to the text vnode (for procfs) */ 534 p2->p_textvp = p1->p_textvp; 535 if (p2->p_textvp) 536 VREF(p2->p_textvp); 537 p2->p_fd = fd; 538 PROC_UNLOCK(p1); 539 PROC_UNLOCK(p2); 540 541 /* 542 * p_limit is copy-on-write, bump refcnt, 543 */ 544 p2->p_limit = p1->p_limit; 545 p2->p_limit->p_refcnt++; 546 547 /* 548 * Setup linkage for kernel based threading 549 */ 550 if((flags & RFTHREAD) != 0) { 551 mtx_lock(&ppeers_lock); 552 p2->p_peers = p1->p_peers; 553 p1->p_peers = p2; 554 p2->p_leader = p1->p_leader; 555 mtx_unlock(&ppeers_lock); 556 PROC_LOCK(p1->p_leader); 557 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 558 PROC_UNLOCK(p1->p_leader); 559 /* 560 * The task leader is exiting, so process p1 is 561 * going to be killed shortly. Since p1 obviously 562 * isn't dead yet, we know that the leader is either 563 * sending SIGKILL's to all the processes in this 564 * task or is sleeping waiting for all the peers to 565 * exit. We let p1 complete the fork, but we need 566 * to go ahead and kill the new process p2 since 567 * the task leader may not get a chance to send 568 * SIGKILL to it. We leave it on the list so that 569 * the task leader will wait for this new process 570 * to commit suicide. 571 */ 572 PROC_LOCK(p2); 573 psignal(p2, SIGKILL); 574 PROC_UNLOCK(p2); 575 } else 576 PROC_UNLOCK(p1->p_leader); 577 } else { 578 p2->p_peers = NULL; 579 p2->p_leader = p2; 580 } 581 582 sx_xlock(&proctree_lock); 583 PGRP_LOCK(p1->p_pgrp); 584 PROC_LOCK(p2); 585 PROC_LOCK(p1); 586 587 /* 588 * Preserve some more flags in subprocess. PS_PROFIL has already 589 * been preserved. 590 */ 591 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 592 SESS_LOCK(p1->p_session); 593 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 594 p2->p_flag |= P_CONTROLT; 595 SESS_UNLOCK(p1->p_session); 596 if (flags & RFPPWAIT) 597 p2->p_flag |= P_PPWAIT; 598 599 LIST_INSERT_AFTER(p1, p2, p_pglist); 600 PGRP_UNLOCK(p1->p_pgrp); 601 LIST_INIT(&p2->p_children); 602 603 callout_init(&p2->p_itcallout, 0); 604 605 #ifdef KTRACE 606 /* 607 * Copy traceflag and tracefile if enabled. 608 */ 609 mtx_lock(&ktrace_mtx); 610 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 611 if (p1->p_traceflag & KTRFAC_INHERIT) { 612 p2->p_traceflag = p1->p_traceflag; 613 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 614 VREF(p2->p_tracevp); 615 KASSERT(p1->p_tracecred != NULL, 616 ("ktrace vnode with no cred")); 617 p2->p_tracecred = crhold(p1->p_tracecred); 618 } 619 } 620 mtx_unlock(&ktrace_mtx); 621 #endif 622 623 /* 624 * If PF_FORK is set, the child process inherits the 625 * procfs ioctl flags from its parent. 626 */ 627 if (p1->p_pfsflags & PF_FORK) { 628 p2->p_stops = p1->p_stops; 629 p2->p_pfsflags = p1->p_pfsflags; 630 } 631 632 /* 633 * This begins the section where we must prevent the parent 634 * from being swapped. 635 */ 636 _PHOLD(p1); 637 PROC_UNLOCK(p1); 638 639 /* 640 * Attach the new process to its parent. 641 * 642 * If RFNOWAIT is set, the newly created process becomes a child 643 * of init. This effectively disassociates the child from the 644 * parent. 645 */ 646 if (flags & RFNOWAIT) 647 pptr = initproc; 648 else 649 pptr = p1; 650 p2->p_pptr = pptr; 651 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 652 PROC_UNLOCK(p2); 653 sx_xunlock(&proctree_lock); 654 655 KASSERT(newprocsig == NULL, ("unused newprocsig")); 656 if (newsigacts != NULL) 657 FREE(newsigacts, M_SUBPROC); 658 /* 659 * Finish creating the child process. It will return via a different 660 * execution path later. (ie: directly into user mode) 661 */ 662 vm_forkproc(td, p2, td2, flags); 663 664 if (flags == (RFFDG | RFPROC)) { 665 cnt.v_forks++; 666 cnt.v_forkpages += p2->p_vmspace->vm_dsize + 667 p2->p_vmspace->vm_ssize; 668 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 669 cnt.v_vforks++; 670 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 671 p2->p_vmspace->vm_ssize; 672 } else if (p1 == &proc0) { 673 cnt.v_kthreads++; 674 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 675 p2->p_vmspace->vm_ssize; 676 } else { 677 cnt.v_rforks++; 678 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 679 p2->p_vmspace->vm_ssize; 680 } 681 682 /* 683 * Both processes are set up, now check if any loadable modules want 684 * to adjust anything. 685 * What if they have an error? XXX 686 */ 687 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 688 689 /* 690 * If RFSTOPPED not requested, make child runnable and add to 691 * run queue. 692 */ 693 microtime(&(p2->p_stats->p_start)); 694 p2->p_acflag = AFORK; 695 if ((flags & RFSTOPPED) == 0) { 696 mtx_lock_spin(&sched_lock); 697 p2->p_state = PRS_NORMAL; 698 TD_SET_CAN_RUN(td2); 699 setrunqueue(td2); 700 mtx_unlock_spin(&sched_lock); 701 } 702 703 /* 704 * Now can be swapped. 705 */ 706 PROC_LOCK(p1); 707 _PRELE(p1); 708 709 /* 710 * tell any interested parties about the new process 711 */ 712 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 713 PROC_UNLOCK(p1); 714 715 /* 716 * Preserve synchronization semantics of vfork. If waiting for 717 * child to exec or exit, set P_PPWAIT on child, and sleep on our 718 * proc (in case of exit). 719 */ 720 PROC_LOCK(p2); 721 while (p2->p_flag & P_PPWAIT) 722 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 723 PROC_UNLOCK(p2); 724 725 /* 726 * If other threads are waiting, let them continue now 727 */ 728 if (p1->p_flag & P_THREADED) { 729 PROC_LOCK(p1); 730 thread_single_end(); 731 PROC_UNLOCK(p1); 732 } 733 734 /* 735 * Return child proc pointer to parent. 736 */ 737 *procp = p2; 738 return (0); 739 fail: 740 sx_xunlock(&allproc_lock); 741 uma_zfree(proc_zone, newproc); 742 if (p1->p_flag & P_THREADED) { 743 PROC_LOCK(p1); 744 thread_single_end(); 745 PROC_UNLOCK(p1); 746 } 747 tsleep(&forksleep, PUSER, "fork", hz / 2); 748 return (error); 749 } 750 751 /* 752 * Handle the return of a child process from fork1(). This function 753 * is called from the MD fork_trampoline() entry point. 754 */ 755 void 756 fork_exit(callout, arg, frame) 757 void (*callout)(void *, struct trapframe *); 758 void *arg; 759 struct trapframe *frame; 760 { 761 struct thread *td; 762 struct proc *p; 763 764 if ((td = PCPU_GET(deadthread))) { 765 PCPU_SET(deadthread, NULL); 766 thread_stash(td); 767 } 768 td = curthread; 769 p = td->td_proc; 770 td->td_kse->ke_oncpu = PCPU_GET(cpuid); 771 p->p_state = PRS_NORMAL; 772 /* 773 * Finish setting up thread glue. We need to initialize 774 * the thread into a td_critnest=1 state. Some platforms 775 * may have already partially or fully initialized td_critnest 776 * and/or td_md.md_savecrit (when applciable). 777 * 778 * see <arch>/<arch>/critical.c 779 */ 780 sched_lock.mtx_lock = (uintptr_t)td; 781 sched_lock.mtx_recurse = 0; 782 cpu_critical_fork_exit(); 783 CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid, 784 p->p_comm); 785 if (PCPU_GET(switchtime.sec) == 0) 786 binuptime(PCPU_PTR(switchtime)); 787 PCPU_SET(switchticks, ticks); 788 mtx_unlock_spin(&sched_lock); 789 790 /* 791 * cpu_set_fork_handler intercepts this function call to 792 * have this call a non-return function to stay in kernel mode. 793 * initproc has its own fork handler, but it does return. 794 */ 795 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 796 callout(arg, frame); 797 798 /* 799 * Check if a kernel thread misbehaved and returned from its main 800 * function. 801 */ 802 PROC_LOCK(p); 803 if (p->p_flag & P_KTHREAD) { 804 PROC_UNLOCK(p); 805 mtx_lock(&Giant); 806 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 807 p->p_comm, p->p_pid); 808 kthread_exit(0); 809 } 810 PROC_UNLOCK(p); 811 #ifdef DIAGNOSTIC 812 cred_free_thread(td); 813 #endif 814 mtx_assert(&Giant, MA_NOTOWNED); 815 } 816 817 /* 818 * Simplified back end of syscall(), used when returning from fork() 819 * directly into user mode. Giant is not held on entry, and must not 820 * be held on return. This function is passed in to fork_exit() as the 821 * first parameter and is called when returning to a new userland process. 822 */ 823 void 824 fork_return(td, frame) 825 struct thread *td; 826 struct trapframe *frame; 827 { 828 829 userret(td, frame, 0); 830 #ifdef KTRACE 831 if (KTRPOINT(td, KTR_SYSRET)) 832 ktrsysret(SYS_fork, 0, 0); 833 #endif 834 mtx_assert(&Giant, MA_NOTOWNED); 835 } 836