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 static 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 error = fork1(td, RFFDG | RFPROC, 0, &p2); 101 if (error == 0) { 102 td->td_retval[0] = p2->p_pid; 103 td->td_retval[1] = 0; 104 } 105 return error; 106 } 107 108 /* 109 * MPSAFE 110 */ 111 /* ARGSUSED */ 112 int 113 vfork(td, uap) 114 struct thread *td; 115 struct vfork_args *uap; 116 { 117 int error; 118 struct proc *p2; 119 120 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 121 if (error == 0) { 122 td->td_retval[0] = p2->p_pid; 123 td->td_retval[1] = 0; 124 } 125 return error; 126 } 127 128 /* 129 * MPSAFE 130 */ 131 int 132 rfork(td, uap) 133 struct thread *td; 134 struct rfork_args *uap; 135 { 136 int error; 137 struct proc *p2; 138 139 /* Don't allow kernel only flags. */ 140 if ((uap->flags & RFKERNELONLY) != 0) 141 return (EINVAL); 142 /* 143 * Don't allow sharing of file descriptor table unless 144 * RFTHREAD flag is supplied 145 */ 146 if ((uap->flags & (RFPROC | RFTHREAD | RFFDG | RFCFDG)) == 147 RFPROC) 148 return(EINVAL); 149 error = fork1(td, uap->flags, 0, &p2); 150 if (error == 0) { 151 td->td_retval[0] = p2 ? p2->p_pid : 0; 152 td->td_retval[1] = 0; 153 } 154 return error; 155 } 156 157 158 int nprocs = 1; /* process 0 */ 159 int lastpid = 0; 160 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 161 "Last used PID"); 162 163 /* 164 * Random component to lastpid generation. We mix in a random factor to make 165 * it a little harder to predict. We sanity check the modulus value to avoid 166 * doing it in critical paths. Don't let it be too small or we pointlessly 167 * waste randomness entropy, and don't let it be impossibly large. Using a 168 * modulus that is too big causes a LOT more process table scans and slows 169 * down fork processing as the pidchecked caching is defeated. 170 */ 171 static int randompid = 0; 172 173 static int 174 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 175 { 176 int error, pid; 177 178 sysctl_wire_old_buffer(req, sizeof(int)); 179 sx_xlock(&allproc_lock); 180 pid = randompid; 181 error = sysctl_handle_int(oidp, &pid, 0, req); 182 if (error == 0 && req->newptr != NULL) { 183 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 184 pid = PID_MAX - 100; 185 else if (pid < 2) /* NOP */ 186 pid = 0; 187 else if (pid < 100) /* Make it reasonable */ 188 pid = 100; 189 randompid = pid; 190 } 191 sx_xunlock(&allproc_lock); 192 return (error); 193 } 194 195 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 196 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 197 198 int 199 fork1(td, flags, pages, procp) 200 struct thread *td; /* parent proc */ 201 int flags; 202 int pages; 203 struct proc **procp; /* child proc */ 204 { 205 struct proc *p2, *pptr; 206 uid_t uid; 207 struct proc *newproc; 208 int trypid; 209 int ok; 210 static int pidchecked = 0; 211 struct filedesc *fd; 212 struct proc *p1 = td->td_proc; 213 struct thread *td2; 214 struct kse *ke2; 215 struct ksegrp *kg2; 216 struct sigacts *newsigacts; 217 struct procsig *newprocsig; 218 int error; 219 220 /* Can't copy and clear */ 221 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 222 return (EINVAL); 223 224 mtx_lock(&Giant); 225 /* 226 * Here we don't create a new process, but we divorce 227 * certain parts of a process from itself. 228 */ 229 if ((flags & RFPROC) == 0) { 230 vm_forkproc(td, NULL, NULL, flags); 231 232 /* 233 * Close all file descriptors. 234 */ 235 if (flags & RFCFDG) { 236 struct filedesc *fdtmp; 237 fdtmp = fdinit(td->td_proc->p_fd); 238 fdfree(td); 239 p1->p_fd = fdtmp; 240 } 241 242 /* 243 * Unshare file descriptors (from parent.) 244 */ 245 if (flags & RFFDG) { 246 FILEDESC_LOCK(p1->p_fd); 247 if (p1->p_fd->fd_refcnt > 1) { 248 struct filedesc *newfd; 249 250 newfd = fdcopy(td->td_proc->p_fd); 251 FILEDESC_UNLOCK(p1->p_fd); 252 fdfree(td); 253 p1->p_fd = newfd; 254 } else 255 FILEDESC_UNLOCK(p1->p_fd); 256 } 257 mtx_unlock(&Giant); 258 *procp = NULL; 259 return (0); 260 } 261 262 /* 263 * Note 1:1 allows for forking with one thread coming out on the 264 * other side with the expectation that the process is about to 265 * exec. 266 */ 267 if (p1->p_flag & P_THREADED) { 268 /* 269 * Idle the other threads for a second. 270 * Since the user space is copied, it must remain stable. 271 * In addition, all threads (from the user perspective) 272 * need to either be suspended or in the kernel, 273 * where they will try restart in the parent and will 274 * be aborted in the child. 275 */ 276 PROC_LOCK(p1); 277 if (thread_single(SINGLE_NO_EXIT)) { 278 /* Abort.. someone else is single threading before us */ 279 PROC_UNLOCK(p1); 280 mtx_unlock(&Giant); 281 return (ERESTART); 282 } 283 PROC_UNLOCK(p1); 284 /* 285 * All other activity in this process 286 * is now suspended at the user boundary, 287 * (or other safe places if we think of any). 288 */ 289 } 290 291 /* Allocate new proc. */ 292 newproc = uma_zalloc(proc_zone, M_WAITOK); 293 #ifdef MAC 294 mac_init_proc(newproc); 295 #endif 296 297 /* 298 * Although process entries are dynamically created, we still keep 299 * a global limit on the maximum number we will create. Don't allow 300 * a nonprivileged user to use the last ten processes; don't let root 301 * exceed the limit. The variable nprocs is the current number of 302 * processes, maxproc is the limit. 303 */ 304 sx_xlock(&allproc_lock); 305 uid = td->td_ucred->cr_ruid; 306 if ((nprocs >= maxproc - 10 && uid != 0) || 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) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 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 } 342 } else { 343 if (randompid) 344 trypid += arc4random() % randompid; 345 } 346 retry: 347 /* 348 * If the process ID prototype has wrapped around, 349 * restart somewhat above 0, as the low-numbered procs 350 * tend to include daemons that don't exit. 351 */ 352 if (trypid >= PID_MAX) { 353 trypid = trypid % PID_MAX; 354 if (trypid < 100) 355 trypid += 100; 356 pidchecked = 0; 357 } 358 if (trypid >= pidchecked) { 359 int doingzomb = 0; 360 361 pidchecked = PID_MAX; 362 /* 363 * Scan the active and zombie procs to check whether this pid 364 * is in use. Remember the lowest pid that's greater 365 * than trypid, so we can avoid checking for a while. 366 */ 367 p2 = LIST_FIRST(&allproc); 368 again: 369 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 370 PROC_LOCK(p2); 371 while (p2->p_pid == trypid || 372 p2->p_pgrp->pg_id == trypid || 373 p2->p_session->s_sid == trypid) { 374 trypid++; 375 if (trypid >= pidchecked) { 376 PROC_UNLOCK(p2); 377 goto retry; 378 } 379 } 380 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 381 pidchecked = p2->p_pid; 382 if (p2->p_pgrp->pg_id > trypid && 383 pidchecked > p2->p_pgrp->pg_id) 384 pidchecked = p2->p_pgrp->pg_id; 385 if (p2->p_session->s_sid > trypid && 386 pidchecked > p2->p_session->s_sid) 387 pidchecked = p2->p_session->s_sid; 388 PROC_UNLOCK(p2); 389 } 390 if (!doingzomb) { 391 doingzomb = 1; 392 p2 = LIST_FIRST(&zombproc); 393 goto again; 394 } 395 } 396 397 /* 398 * RFHIGHPID does not mess with the lastpid counter during boot. 399 */ 400 if (flags & RFHIGHPID) 401 pidchecked = 0; 402 else 403 lastpid = trypid; 404 405 p2 = newproc; 406 p2->p_state = PRS_NEW; /* protect against others */ 407 p2->p_pid = trypid; 408 LIST_INSERT_HEAD(&allproc, p2, p_list); 409 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 410 sx_xunlock(&allproc_lock); 411 412 /* 413 * Malloc things while we don't hold any locks. 414 */ 415 if (flags & RFSIGSHARE) { 416 MALLOC(newsigacts, struct sigacts *, 417 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 418 newprocsig = NULL; 419 } else { 420 newsigacts = NULL; 421 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig), 422 M_SUBPROC, M_WAITOK); 423 } 424 425 /* 426 * Copy filedesc. 427 */ 428 if (flags & RFCFDG) 429 fd = fdinit(td->td_proc->p_fd); 430 else if (flags & RFFDG) { 431 FILEDESC_LOCK(p1->p_fd); 432 fd = fdcopy(td->td_proc->p_fd); 433 FILEDESC_UNLOCK(p1->p_fd); 434 } else 435 fd = fdshare(p1->p_fd); 436 437 /* 438 * Make a proc table entry for the new process. 439 * Start by zeroing the section of proc that is zero-initialized, 440 * then copy the section that is copied directly from the parent. 441 */ 442 td2 = FIRST_THREAD_IN_PROC(p2); 443 kg2 = FIRST_KSEGRP_IN_PROC(p2); 444 ke2 = FIRST_KSE_IN_KSEGRP(kg2); 445 446 /* Allocate and switch to an alternate kstack if specified */ 447 if (pages != 0) 448 pmap_new_altkstack(td2, pages); 449 450 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 451 452 bzero(&p2->p_startzero, 453 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 454 bzero(&ke2->ke_startzero, 455 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero)); 456 bzero(&td2->td_startzero, 457 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 458 bzero(&kg2->kg_startzero, 459 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 460 461 mtx_init(&p2->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK); 462 PROC_LOCK(p2); 463 PROC_LOCK(p1); 464 465 bcopy(&p1->p_startcopy, &p2->p_startcopy, 466 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 467 bcopy(&td->td_startcopy, &td2->td_startcopy, 468 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 469 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 470 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 471 #undef RANGEOF 472 473 /* Set up the thread as an active thread (as if runnable). */ 474 ke2->ke_state = KES_THREAD; 475 ke2->ke_thread = td2; 476 td2->td_kse = ke2; 477 478 /* 479 * Duplicate sub-structures as needed. 480 * Increase reference counts on shared objects. 481 * The p_stats and p_sigacts substructs are set in vm_forkproc. 482 */ 483 p2->p_flag = 0; 484 if (p1->p_flag & P_PROFIL) 485 startprofclock(p2); 486 mtx_lock_spin(&sched_lock); 487 p2->p_sflag = PS_INMEM; 488 /* 489 * Allow the scheduler to adjust the priority of the child and 490 * parent while we hold the sched_lock. 491 */ 492 sched_fork(p1, p2); 493 494 mtx_unlock_spin(&sched_lock); 495 p2->p_ucred = crhold(td->td_ucred); 496 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 497 498 pargs_hold(p2->p_args); 499 500 if (flags & RFSIGSHARE) { 501 p2->p_procsig = p1->p_procsig; 502 p2->p_procsig->ps_refcnt++; 503 if (p1->p_sigacts == &p1->p_uarea->u_sigacts) { 504 /* 505 * Set p_sigacts to the new shared structure. 506 * Note that this is updating p1->p_sigacts at the 507 * same time, since p_sigacts is just a pointer to 508 * the shared p_procsig->ps_sigacts. 509 */ 510 p2->p_sigacts = newsigacts; 511 newsigacts = NULL; 512 *p2->p_sigacts = p1->p_uarea->u_sigacts; 513 } 514 } else { 515 p2->p_procsig = newprocsig; 516 newprocsig = NULL; 517 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 518 p2->p_procsig->ps_refcnt = 1; 519 p2->p_sigacts = NULL; /* finished in vm_forkproc() */ 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 PROC_UNLOCK(p1); 532 PROC_UNLOCK(p2); 533 534 /* 535 * p_limit is copy-on-write, bump refcnt, 536 */ 537 p2->p_limit = p1->p_limit; 538 p2->p_limit->p_refcnt++; 539 540 /* 541 * Setup linkage for kernel based threading 542 */ 543 if((flags & RFTHREAD) != 0) { 544 mtx_lock(&ppeers_lock); 545 p2->p_peers = p1->p_peers; 546 p1->p_peers = p2; 547 p2->p_leader = p1->p_leader; 548 mtx_unlock(&ppeers_lock); 549 PROC_LOCK(p1->p_leader); 550 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 551 PROC_UNLOCK(p1->p_leader); 552 /* 553 * The task leader is exiting, so process p1 is 554 * going to be killed shortly. Since p1 obviously 555 * isn't dead yet, we know that the leader is either 556 * sending SIGKILL's to all the processes in this 557 * task or is sleeping waiting for all the peers to 558 * exit. We let p1 complete the fork, but we need 559 * to go ahead and kill the new process p2 since 560 * the task leader may not get a chance to send 561 * SIGKILL to it. We leave it on the list so that 562 * the task leader will wait for this new process 563 * to commit suicide. 564 */ 565 PROC_LOCK(p2); 566 psignal(p2, SIGKILL); 567 PROC_UNLOCK(p2); 568 } else 569 PROC_UNLOCK(p1->p_leader); 570 } else { 571 p2->p_peers = NULL; 572 p2->p_leader = p2; 573 } 574 575 sx_xlock(&proctree_lock); 576 PGRP_LOCK(p1->p_pgrp); 577 PROC_LOCK(p2); 578 PROC_LOCK(p1); 579 580 /* 581 * Preserve some more flags in subprocess. P_PROFIL has already 582 * been preserved. 583 */ 584 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 585 SESS_LOCK(p1->p_session); 586 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 587 p2->p_flag |= P_CONTROLT; 588 SESS_UNLOCK(p1->p_session); 589 if (flags & RFPPWAIT) 590 p2->p_flag |= P_PPWAIT; 591 592 LIST_INSERT_AFTER(p1, p2, p_pglist); 593 PGRP_UNLOCK(p1->p_pgrp); 594 LIST_INIT(&p2->p_children); 595 596 callout_init(&p2->p_itcallout, 1); 597 598 #ifdef KTRACE 599 /* 600 * Copy traceflag and tracefile if enabled. 601 */ 602 mtx_lock(&ktrace_mtx); 603 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 604 if (p1->p_traceflag & KTRFAC_INHERIT) { 605 p2->p_traceflag = p1->p_traceflag; 606 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 607 VREF(p2->p_tracevp); 608 KASSERT(p1->p_tracecred != NULL, 609 ("ktrace vnode with no cred")); 610 p2->p_tracecred = crhold(p1->p_tracecred); 611 } 612 } 613 mtx_unlock(&ktrace_mtx); 614 #endif 615 616 /* 617 * If PF_FORK is set, the child process inherits the 618 * procfs ioctl flags from its parent. 619 */ 620 if (p1->p_pfsflags & PF_FORK) { 621 p2->p_stops = p1->p_stops; 622 p2->p_pfsflags = p1->p_pfsflags; 623 } 624 625 /* 626 * This begins the section where we must prevent the parent 627 * from being swapped. 628 */ 629 _PHOLD(p1); 630 PROC_UNLOCK(p1); 631 632 /* 633 * Attach the new process to its parent. 634 * 635 * If RFNOWAIT is set, the newly created process becomes a child 636 * of init. This effectively disassociates the child from the 637 * parent. 638 */ 639 if (flags & RFNOWAIT) 640 pptr = initproc; 641 else 642 pptr = p1; 643 p2->p_pptr = pptr; 644 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 645 sx_xunlock(&proctree_lock); 646 647 /* Inform accounting that we have forked. */ 648 p2->p_acflag = AFORK; 649 PROC_UNLOCK(p2); 650 651 KASSERT(newprocsig == NULL, ("unused newprocsig")); 652 if (newsigacts != NULL) 653 FREE(newsigacts, M_SUBPROC); 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 microtime(&(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 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_THREADED) { 724 PROC_LOCK(p1); 725 thread_single_end(); 726 PROC_UNLOCK(p1); 727 } 728 729 /* 730 * Return child proc pointer to parent. 731 */ 732 mtx_unlock(&Giant); 733 *procp = p2; 734 return (0); 735 fail: 736 sx_xunlock(&allproc_lock); 737 uma_zfree(proc_zone, newproc); 738 if (p1->p_flag & P_THREADED) { 739 PROC_LOCK(p1); 740 thread_single_end(); 741 PROC_UNLOCK(p1); 742 } 743 tsleep(&forksleep, PUSER, "fork", hz / 2); 744 mtx_unlock(&Giant); 745 return (error); 746 } 747 748 /* 749 * Handle the return of a child process from fork1(). This function 750 * is called from the MD fork_trampoline() entry point. 751 */ 752 void 753 fork_exit(callout, arg, frame) 754 void (*callout)(void *, struct trapframe *); 755 void *arg; 756 struct trapframe *frame; 757 { 758 struct thread *td; 759 struct proc *p; 760 761 if ((td = PCPU_GET(deadthread))) { 762 PCPU_SET(deadthread, NULL); 763 thread_stash(td); 764 } 765 td = curthread; 766 p = td->td_proc; 767 td->td_oncpu = PCPU_GET(cpuid); 768 p->p_state = PRS_NORMAL; 769 /* 770 * Finish setting up thread glue. We need to initialize 771 * the thread into a td_critnest=1 state. Some platforms 772 * may have already partially or fully initialized td_critnest 773 * and/or td_md.md_savecrit (when applciable). 774 * 775 * see <arch>/<arch>/critical.c 776 */ 777 sched_lock.mtx_lock = (uintptr_t)td; 778 sched_lock.mtx_recurse = 0; 779 cpu_critical_fork_exit(); 780 CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid, 781 p->p_comm); 782 if (PCPU_GET(switchtime.sec) == 0) 783 binuptime(PCPU_PTR(switchtime)); 784 PCPU_SET(switchticks, ticks); 785 mtx_unlock_spin(&sched_lock); 786 787 /* 788 * cpu_set_fork_handler intercepts this function call to 789 * have this call a non-return function to stay in kernel mode. 790 * initproc has its own fork handler, but it does return. 791 */ 792 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 793 callout(arg, frame); 794 795 /* 796 * Check if a kernel thread misbehaved and returned from its main 797 * function. 798 */ 799 PROC_LOCK(p); 800 if (p->p_flag & P_KTHREAD) { 801 PROC_UNLOCK(p); 802 mtx_lock(&Giant); 803 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 804 p->p_comm, p->p_pid); 805 kthread_exit(0); 806 } 807 PROC_UNLOCK(p); 808 #ifdef DIAGNOSTIC 809 cred_free_thread(td); 810 #endif 811 mtx_assert(&Giant, MA_NOTOWNED); 812 } 813 814 /* 815 * Simplified back end of syscall(), used when returning from fork() 816 * directly into user mode. Giant is not held on entry, and must not 817 * be held on return. This function is passed in to fork_exit() as the 818 * first parameter and is called when returning to a new userland process. 819 */ 820 void 821 fork_return(td, frame) 822 struct thread *td; 823 struct trapframe *frame; 824 { 825 826 userret(td, frame, 0); 827 #ifdef KTRACE 828 if (KTRPOINT(td, KTR_SYSRET)) 829 ktrsysret(SYS_fork, 0, 0); 830 #endif 831 mtx_assert(&Giant, MA_NOTOWNED); 832 } 833