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 290 291 /* We have to lock the process tree while we look for a pid. */ 292 sx_slock(&proctree_lock); 293 294 /* 295 * Although process entries are dynamically created, we still keep 296 * a global limit on the maximum number we will create. Don't allow 297 * a nonprivileged user to use the last ten processes; don't let root 298 * exceed the limit. The variable nprocs is the current number of 299 * processes, maxproc is the limit. 300 */ 301 sx_xlock(&allproc_lock); 302 uid = td->td_ucred->cr_ruid; 303 if ((nprocs >= maxproc - 10 && 304 suser_cred(td->td_ucred, SUSER_RUID) != 0) || 305 nprocs >= maxproc) { 306 error = EAGAIN; 307 goto fail; 308 } 309 310 /* 311 * Increment the count of procs running with this uid. Don't allow 312 * a nonprivileged user to exceed their current limit. 313 */ 314 PROC_LOCK(p1); 315 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 316 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0); 317 PROC_UNLOCK(p1); 318 if (!ok) { 319 error = EAGAIN; 320 goto fail; 321 } 322 323 /* 324 * Increment the nprocs resource before blocking can occur. There 325 * are hard-limits as to the number of processes that can run. 326 */ 327 nprocs++; 328 329 /* 330 * Find an unused process ID. We remember a range of unused IDs 331 * ready to use (from lastpid+1 through pidchecked-1). 332 * 333 * If RFHIGHPID is set (used during system boot), do not allocate 334 * low-numbered pids. 335 */ 336 trypid = lastpid + 1; 337 if (flags & RFHIGHPID) { 338 if (trypid < 10) 339 trypid = 10; 340 } else { 341 if (randompid) 342 trypid += arc4random() % randompid; 343 } 344 retry: 345 /* 346 * If the process ID prototype has wrapped around, 347 * restart somewhat above 0, as the low-numbered procs 348 * tend to include daemons that don't exit. 349 */ 350 if (trypid >= PID_MAX) { 351 trypid = trypid % PID_MAX; 352 if (trypid < 100) 353 trypid += 100; 354 pidchecked = 0; 355 } 356 if (trypid >= pidchecked) { 357 int doingzomb = 0; 358 359 pidchecked = PID_MAX; 360 /* 361 * Scan the active and zombie procs to check whether this pid 362 * is in use. Remember the lowest pid that's greater 363 * than trypid, so we can avoid checking for a while. 364 */ 365 p2 = LIST_FIRST(&allproc); 366 again: 367 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 368 PROC_LOCK(p2); 369 while (p2->p_pid == trypid || 370 (p2->p_pgrp != NULL && 371 (p2->p_pgrp->pg_id == trypid || 372 (p2->p_session != NULL && 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 != NULL) { 383 if (p2->p_pgrp->pg_id > trypid && 384 pidchecked > p2->p_pgrp->pg_id) 385 pidchecked = p2->p_pgrp->pg_id; 386 if (p2->p_session != NULL && 387 p2->p_session->s_sid > trypid && 388 pidchecked > p2->p_session->s_sid) 389 pidchecked = p2->p_session->s_sid; 390 } 391 PROC_UNLOCK(p2); 392 } 393 if (!doingzomb) { 394 doingzomb = 1; 395 p2 = LIST_FIRST(&zombproc); 396 goto again; 397 } 398 } 399 sx_sunlock(&proctree_lock); 400 401 /* 402 * RFHIGHPID does not mess with the lastpid counter during boot. 403 */ 404 if (flags & RFHIGHPID) 405 pidchecked = 0; 406 else 407 lastpid = trypid; 408 409 p2 = newproc; 410 p2->p_state = PRS_NEW; /* protect against others */ 411 p2->p_pid = trypid; 412 LIST_INSERT_HEAD(&allproc, p2, p_list); 413 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 414 sx_xunlock(&allproc_lock); 415 416 /* 417 * Malloc things while we don't hold any locks. 418 */ 419 if (flags & RFSIGSHARE) 420 newsigacts = NULL; 421 else 422 newsigacts = sigacts_alloc(); 423 424 /* 425 * Copy filedesc. 426 */ 427 if (flags & RFCFDG) { 428 FILEDESC_LOCK(td->td_proc->p_fd); 429 fd = fdinit(td->td_proc->p_fd); 430 FILEDESC_UNLOCK(td->td_proc->p_fd); 431 fdtol = NULL; 432 } else if (flags & RFFDG) { 433 FILEDESC_LOCK(p1->p_fd); 434 fd = fdcopy(td->td_proc->p_fd); 435 FILEDESC_UNLOCK(p1->p_fd); 436 fdtol = NULL; 437 } else { 438 fd = fdshare(p1->p_fd); 439 if (p1->p_fdtol == NULL) 440 p1->p_fdtol = 441 filedesc_to_leader_alloc(NULL, 442 NULL, 443 p1->p_leader); 444 if ((flags & RFTHREAD) != 0) { 445 /* 446 * Shared file descriptor table and 447 * shared process leaders. 448 */ 449 fdtol = p1->p_fdtol; 450 FILEDESC_LOCK(p1->p_fd); 451 fdtol->fdl_refcount++; 452 FILEDESC_UNLOCK(p1->p_fd); 453 } else { 454 /* 455 * Shared file descriptor table, and 456 * different process leaders 457 */ 458 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 459 p1->p_fd, 460 p2); 461 } 462 } 463 /* 464 * Make a proc table entry for the new process. 465 * Start by zeroing the section of proc that is zero-initialized, 466 * then copy the section that is copied directly from the parent. 467 */ 468 td2 = FIRST_THREAD_IN_PROC(p2); 469 kg2 = FIRST_KSEGRP_IN_PROC(p2); 470 ke2 = FIRST_KSE_IN_KSEGRP(kg2); 471 472 /* Allocate and switch to an alternate kstack if specified. */ 473 if (pages != 0) 474 vm_thread_new_altkstack(td2, pages); 475 476 PROC_LOCK(p2); 477 PROC_LOCK(p1); 478 479 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 480 481 bzero(&p2->p_startzero, 482 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 483 bzero(&ke2->ke_startzero, 484 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero)); 485 bzero(&td2->td_startzero, 486 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 487 bzero(&kg2->kg_startzero, 488 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 489 490 bcopy(&p1->p_startcopy, &p2->p_startcopy, 491 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 492 bcopy(&td->td_startcopy, &td2->td_startcopy, 493 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 494 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 495 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 496 #undef RANGEOF 497 498 td2->td_sigstk = td->td_sigstk; 499 500 /* Set up the thread as an active thread (as if runnable). */ 501 ke2->ke_state = KES_THREAD; 502 ke2->ke_thread = td2; 503 td2->td_kse = ke2; 504 505 /* 506 * Duplicate sub-structures as needed. 507 * Increase reference counts on shared objects. 508 * The p_stats substruct is set in vm_forkproc. 509 */ 510 p2->p_flag = 0; 511 if (p1->p_flag & P_PROFIL) 512 startprofclock(p2); 513 mtx_lock_spin(&sched_lock); 514 p2->p_sflag = PS_INMEM; 515 /* 516 * Allow the scheduler to adjust the priority of the child and 517 * parent while we hold the sched_lock. 518 */ 519 sched_fork(td, p2); 520 521 mtx_unlock_spin(&sched_lock); 522 p2->p_ucred = crhold(td->td_ucred); 523 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 524 525 pargs_hold(p2->p_args); 526 527 if (flags & RFSIGSHARE) { 528 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 529 } else { 530 sigacts_copy(newsigacts, p1->p_sigacts); 531 p2->p_sigacts = newsigacts; 532 } 533 if (flags & RFLINUXTHPN) 534 p2->p_sigparent = SIGUSR1; 535 else 536 p2->p_sigparent = SIGCHLD; 537 538 p2->p_textvp = p1->p_textvp; 539 p2->p_fd = fd; 540 p2->p_fdtol = fdtol; 541 542 /* 543 * p_limit is copy-on-write. Bump its refcount. 544 */ 545 p2->p_limit = lim_hold(p1->p_limit); 546 PROC_UNLOCK(p1); 547 PROC_UNLOCK(p2); 548 549 /* Bump references to the text vnode (for procfs) */ 550 if (p2->p_textvp) 551 vref(p2->p_textvp); 552 553 /* 554 * Set up linkage for kernel based threading. 555 */ 556 if ((flags & RFTHREAD) != 0) { 557 mtx_lock(&ppeers_lock); 558 p2->p_peers = p1->p_peers; 559 p1->p_peers = p2; 560 p2->p_leader = p1->p_leader; 561 mtx_unlock(&ppeers_lock); 562 PROC_LOCK(p1->p_leader); 563 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 564 PROC_UNLOCK(p1->p_leader); 565 /* 566 * The task leader is exiting, so process p1 is 567 * going to be killed shortly. Since p1 obviously 568 * isn't dead yet, we know that the leader is either 569 * sending SIGKILL's to all the processes in this 570 * task or is sleeping waiting for all the peers to 571 * exit. We let p1 complete the fork, but we need 572 * to go ahead and kill the new process p2 since 573 * the task leader may not get a chance to send 574 * SIGKILL to it. We leave it on the list so that 575 * the task leader will wait for this new process 576 * to commit suicide. 577 */ 578 PROC_LOCK(p2); 579 psignal(p2, SIGKILL); 580 PROC_UNLOCK(p2); 581 } else 582 PROC_UNLOCK(p1->p_leader); 583 } else { 584 p2->p_peers = NULL; 585 p2->p_leader = p2; 586 } 587 588 sx_xlock(&proctree_lock); 589 PGRP_LOCK(p1->p_pgrp); 590 PROC_LOCK(p2); 591 PROC_LOCK(p1); 592 593 /* 594 * Preserve some more flags in subprocess. P_PROFIL has already 595 * been preserved. 596 */ 597 p2->p_flag |= p1->p_flag & P_SUGID; 598 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 599 SESS_LOCK(p1->p_session); 600 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 601 p2->p_flag |= P_CONTROLT; 602 SESS_UNLOCK(p1->p_session); 603 if (flags & RFPPWAIT) 604 p2->p_flag |= P_PPWAIT; 605 606 p2->p_pgrp = p1->p_pgrp; 607 LIST_INSERT_AFTER(p1, p2, p_pglist); 608 PGRP_UNLOCK(p1->p_pgrp); 609 LIST_INIT(&p2->p_children); 610 611 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 612 613 #ifdef KTRACE 614 /* 615 * Copy traceflag and tracefile if enabled. 616 */ 617 mtx_lock(&ktrace_mtx); 618 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 619 if (p1->p_traceflag & KTRFAC_INHERIT) { 620 p2->p_traceflag = p1->p_traceflag; 621 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 622 VREF(p2->p_tracevp); 623 KASSERT(p1->p_tracecred != NULL, 624 ("ktrace vnode with no cred")); 625 p2->p_tracecred = crhold(p1->p_tracecred); 626 } 627 } 628 mtx_unlock(&ktrace_mtx); 629 #endif 630 631 /* 632 * If PF_FORK is set, the child process inherits the 633 * procfs ioctl flags from its parent. 634 */ 635 if (p1->p_pfsflags & PF_FORK) { 636 p2->p_stops = p1->p_stops; 637 p2->p_pfsflags = p1->p_pfsflags; 638 } 639 640 /* 641 * This begins the section where we must prevent the parent 642 * from being swapped. 643 */ 644 _PHOLD(p1); 645 PROC_UNLOCK(p1); 646 647 /* 648 * Attach the new process to its parent. 649 * 650 * If RFNOWAIT is set, the newly created process becomes a child 651 * of init. This effectively disassociates the child from the 652 * parent. 653 */ 654 if (flags & RFNOWAIT) 655 pptr = initproc; 656 else 657 pptr = p1; 658 p2->p_pptr = pptr; 659 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 660 sx_xunlock(&proctree_lock); 661 662 /* Inform accounting that we have forked. */ 663 p2->p_acflag = AFORK; 664 PROC_UNLOCK(p2); 665 666 /* 667 * Finish creating the child process. It will return via a different 668 * execution path later. (ie: directly into user mode) 669 */ 670 mtx_lock(&Giant); 671 vm_forkproc(td, p2, td2, flags); 672 673 if (flags == (RFFDG | RFPROC)) { 674 cnt.v_forks++; 675 cnt.v_forkpages += p2->p_vmspace->vm_dsize + 676 p2->p_vmspace->vm_ssize; 677 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 678 cnt.v_vforks++; 679 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 680 p2->p_vmspace->vm_ssize; 681 } else if (p1 == &proc0) { 682 cnt.v_kthreads++; 683 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 684 p2->p_vmspace->vm_ssize; 685 } else { 686 cnt.v_rforks++; 687 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 688 p2->p_vmspace->vm_ssize; 689 } 690 mtx_unlock(&Giant); 691 692 /* 693 * Both processes are set up, now check if any loadable modules want 694 * to adjust anything. 695 * What if they have an error? XXX 696 */ 697 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 698 699 /* 700 * Set the child start time and mark the process as being complete. 701 */ 702 microuptime(&p2->p_stats->p_start); 703 mtx_lock_spin(&sched_lock); 704 p2->p_state = PRS_NORMAL; 705 706 /* 707 * If RFSTOPPED not requested, make child runnable and add to 708 * run queue. 709 */ 710 if ((flags & RFSTOPPED) == 0) { 711 TD_SET_CAN_RUN(td2); 712 setrunqueue(td2); 713 } 714 mtx_unlock_spin(&sched_lock); 715 716 /* 717 * Now can be swapped. 718 */ 719 PROC_LOCK(p1); 720 _PRELE(p1); 721 722 /* 723 * Tell any interested parties about the new process. 724 */ 725 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid); 726 727 PROC_UNLOCK(p1); 728 729 /* 730 * Preserve synchronization semantics of vfork. If waiting for 731 * child to exec or exit, set P_PPWAIT on child, and sleep on our 732 * proc (in case of exit). 733 */ 734 PROC_LOCK(p2); 735 while (p2->p_flag & P_PPWAIT) 736 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 737 PROC_UNLOCK(p2); 738 739 /* 740 * If other threads are waiting, let them continue now. 741 */ 742 if (p1->p_flag & P_SA) { 743 PROC_LOCK(p1); 744 thread_single_end(); 745 PROC_UNLOCK(p1); 746 } 747 748 /* 749 * Return child proc pointer to parent. 750 */ 751 *procp = p2; 752 return (0); 753 fail: 754 sx_sunlock(&proctree_lock); 755 if (ppsratecheck(&lastfail, &curfail, 1)) 756 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 757 uid); 758 sx_xunlock(&allproc_lock); 759 #ifdef MAC 760 mac_destroy_proc(newproc); 761 #endif 762 uma_zfree(proc_zone, newproc); 763 if (p1->p_flag & P_SA) { 764 PROC_LOCK(p1); 765 thread_single_end(); 766 PROC_UNLOCK(p1); 767 } 768 tsleep(&forksleep, PUSER, "fork", hz / 2); 769 return (error); 770 } 771 772 /* 773 * Handle the return of a child process from fork1(). This function 774 * is called from the MD fork_trampoline() entry point. 775 */ 776 void 777 fork_exit(callout, arg, frame) 778 void (*callout)(void *, struct trapframe *); 779 void *arg; 780 struct trapframe *frame; 781 { 782 struct proc *p; 783 struct thread *td; 784 785 /* 786 * Finish setting up thread glue so that it begins execution in a 787 * non-nested critical section with sched_lock held but not recursed. 788 */ 789 td = curthread; 790 p = td->td_proc; 791 td->td_oncpu = PCPU_GET(cpuid); 792 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 793 794 sched_lock.mtx_lock = (uintptr_t)td; 795 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 796 cpu_critical_fork_exit(); 797 CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid, 798 p->p_comm); 799 800 /* 801 * Processes normally resume in mi_switch() after being 802 * cpu_switch()'ed to, but when children start up they arrive here 803 * instead, so we must do much the same things as mi_switch() would. 804 */ 805 806 if ((td = PCPU_GET(deadthread))) { 807 PCPU_SET(deadthread, NULL); 808 thread_stash(td); 809 } 810 td = curthread; 811 mtx_unlock_spin(&sched_lock); 812 813 /* 814 * cpu_set_fork_handler intercepts this function call to 815 * have this call a non-return function to stay in kernel mode. 816 * initproc has its own fork handler, but it does return. 817 */ 818 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 819 callout(arg, frame); 820 821 /* 822 * Check if a kernel thread misbehaved and returned from its main 823 * function. 824 */ 825 PROC_LOCK(p); 826 if (p->p_flag & P_KTHREAD) { 827 PROC_UNLOCK(p); 828 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 829 p->p_comm, p->p_pid); 830 kthread_exit(0); 831 } 832 PROC_UNLOCK(p); 833 #ifdef DIAGNOSTIC 834 cred_free_thread(td); 835 #endif 836 mtx_assert(&Giant, MA_NOTOWNED); 837 } 838 839 /* 840 * Simplified back end of syscall(), used when returning from fork() 841 * directly into user mode. Giant is not held on entry, and must not 842 * be held on return. This function is passed in to fork_exit() as the 843 * first parameter and is called when returning to a new userland process. 844 */ 845 void 846 fork_return(td, frame) 847 struct thread *td; 848 struct trapframe *frame; 849 { 850 851 userret(td, frame, 0); 852 #ifdef KTRACE 853 if (KTRPOINT(td, KTR_SYSRET)) 854 ktrsysret(SYS_fork, 0, 0); 855 #endif 856 mtx_assert(&Giant, MA_NOTOWNED); 857 } 858