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 ksegrp *kg2; 207 struct sigacts *newsigacts; 208 int error; 209 210 /* Can't copy and clear. */ 211 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 212 return (EINVAL); 213 214 p1 = td->td_proc; 215 216 /* 217 * Here we don't create a new process, but we divorce 218 * certain parts of a process from itself. 219 */ 220 if ((flags & RFPROC) == 0) { 221 vm_forkproc(td, NULL, NULL, flags); 222 223 /* 224 * Close all file descriptors. 225 */ 226 if (flags & RFCFDG) { 227 struct filedesc *fdtmp; 228 FILEDESC_LOCK(td->td_proc->p_fd); 229 fdtmp = fdinit(td->td_proc->p_fd); 230 FILEDESC_UNLOCK(td->td_proc->p_fd); 231 fdfree(td); 232 p1->p_fd = fdtmp; 233 } 234 235 /* 236 * Unshare file descriptors (from parent). 237 */ 238 if (flags & RFFDG) { 239 FILEDESC_LOCK(p1->p_fd); 240 if (p1->p_fd->fd_refcnt > 1) { 241 struct filedesc *newfd; 242 243 newfd = fdcopy(td->td_proc->p_fd); 244 FILEDESC_UNLOCK(p1->p_fd); 245 fdfree(td); 246 p1->p_fd = newfd; 247 } else 248 FILEDESC_UNLOCK(p1->p_fd); 249 } 250 *procp = NULL; 251 return (0); 252 } 253 254 /* 255 * Note 1:1 allows for forking with one thread coming out on the 256 * other side with the expectation that the process is about to 257 * exec. 258 */ 259 if (p1->p_flag & P_HADTHREADS) { 260 /* 261 * Idle the other threads for a second. 262 * Since the user space is copied, it must remain stable. 263 * In addition, all threads (from the user perspective) 264 * need to either be suspended or in the kernel, 265 * where they will try restart in the parent and will 266 * be aborted in the child. 267 */ 268 PROC_LOCK(p1); 269 if (thread_single(SINGLE_NO_EXIT)) { 270 /* Abort. Someone else is single threading before us. */ 271 PROC_UNLOCK(p1); 272 return (ERESTART); 273 } 274 PROC_UNLOCK(p1); 275 /* 276 * All other activity in this process 277 * is now suspended at the user boundary, 278 * (or other safe places if we think of any). 279 */ 280 } 281 282 /* Allocate new proc. */ 283 newproc = uma_zalloc(proc_zone, M_WAITOK); 284 #ifdef MAC 285 mac_init_proc(newproc); 286 #endif 287 knlist_init(&newproc->p_klist, &newproc->p_mtx); 288 289 /* We have to lock the process tree while we look for a pid. */ 290 sx_slock(&proctree_lock); 291 292 /* 293 * Although process entries are dynamically created, we still keep 294 * a global limit on the maximum number we will create. Don't allow 295 * a nonprivileged user to use the last ten processes; don't let root 296 * exceed the limit. The variable nprocs is the current number of 297 * processes, maxproc is the limit. 298 */ 299 sx_xlock(&allproc_lock); 300 uid = td->td_ucred->cr_ruid; 301 if ((nprocs >= maxproc - 10 && 302 suser_cred(td->td_ucred, SUSER_RUID) != 0) || 303 nprocs >= maxproc) { 304 error = EAGAIN; 305 goto fail; 306 } 307 308 /* 309 * Increment the count of procs running with this uid. Don't allow 310 * a nonprivileged user to exceed their current limit. 311 */ 312 PROC_LOCK(p1); 313 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 314 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0); 315 PROC_UNLOCK(p1); 316 if (!ok) { 317 error = EAGAIN; 318 goto fail; 319 } 320 321 /* 322 * Increment the nprocs resource before blocking can occur. There 323 * are hard-limits as to the number of processes that can run. 324 */ 325 nprocs++; 326 327 /* 328 * Find an unused process ID. We remember a range of unused IDs 329 * ready to use (from lastpid+1 through pidchecked-1). 330 * 331 * If RFHIGHPID is set (used during system boot), do not allocate 332 * low-numbered pids. 333 */ 334 trypid = lastpid + 1; 335 if (flags & RFHIGHPID) { 336 if (trypid < 10) 337 trypid = 10; 338 } else { 339 if (randompid) 340 trypid += arc4random() % randompid; 341 } 342 retry: 343 /* 344 * If the process ID prototype has wrapped around, 345 * restart somewhat above 0, as the low-numbered procs 346 * tend to include daemons that don't exit. 347 */ 348 if (trypid >= PID_MAX) { 349 trypid = trypid % PID_MAX; 350 if (trypid < 100) 351 trypid += 100; 352 pidchecked = 0; 353 } 354 if (trypid >= pidchecked) { 355 int doingzomb = 0; 356 357 pidchecked = PID_MAX; 358 /* 359 * Scan the active and zombie procs to check whether this pid 360 * is in use. Remember the lowest pid that's greater 361 * than trypid, so we can avoid checking for a while. 362 */ 363 p2 = LIST_FIRST(&allproc); 364 again: 365 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 366 PROC_LOCK(p2); 367 while (p2->p_pid == trypid || 368 (p2->p_pgrp != NULL && 369 (p2->p_pgrp->pg_id == trypid || 370 (p2->p_session != NULL && 371 p2->p_session->s_sid == trypid)))) { 372 trypid++; 373 if (trypid >= pidchecked) { 374 PROC_UNLOCK(p2); 375 goto retry; 376 } 377 } 378 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 379 pidchecked = p2->p_pid; 380 if (p2->p_pgrp != NULL) { 381 if (p2->p_pgrp->pg_id > trypid && 382 pidchecked > p2->p_pgrp->pg_id) 383 pidchecked = p2->p_pgrp->pg_id; 384 if (p2->p_session != NULL && 385 p2->p_session->s_sid > trypid && 386 pidchecked > p2->p_session->s_sid) 387 pidchecked = p2->p_session->s_sid; 388 } 389 PROC_UNLOCK(p2); 390 } 391 if (!doingzomb) { 392 doingzomb = 1; 393 p2 = LIST_FIRST(&zombproc); 394 goto again; 395 } 396 } 397 sx_sunlock(&proctree_lock); 398 399 /* 400 * RFHIGHPID does not mess with the lastpid counter during boot. 401 */ 402 if (flags & RFHIGHPID) 403 pidchecked = 0; 404 else 405 lastpid = trypid; 406 407 p2 = newproc; 408 p2->p_state = PRS_NEW; /* protect against others */ 409 p2->p_pid = trypid; 410 LIST_INSERT_HEAD(&allproc, p2, p_list); 411 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 412 sx_xunlock(&allproc_lock); 413 414 /* 415 * Malloc things while we don't hold any locks. 416 */ 417 if (flags & RFSIGSHARE) 418 newsigacts = NULL; 419 else 420 newsigacts = sigacts_alloc(); 421 422 /* 423 * Copy filedesc. 424 */ 425 if (flags & RFCFDG) { 426 FILEDESC_LOCK(td->td_proc->p_fd); 427 fd = fdinit(td->td_proc->p_fd); 428 FILEDESC_UNLOCK(td->td_proc->p_fd); 429 fdtol = NULL; 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 fdtol = NULL; 435 } else { 436 fd = fdshare(p1->p_fd); 437 if (p1->p_fdtol == NULL) 438 p1->p_fdtol = 439 filedesc_to_leader_alloc(NULL, 440 NULL, 441 p1->p_leader); 442 if ((flags & RFTHREAD) != 0) { 443 /* 444 * Shared file descriptor table and 445 * shared process leaders. 446 */ 447 fdtol = p1->p_fdtol; 448 FILEDESC_LOCK(p1->p_fd); 449 fdtol->fdl_refcount++; 450 FILEDESC_UNLOCK(p1->p_fd); 451 } else { 452 /* 453 * Shared file descriptor table, and 454 * different process leaders 455 */ 456 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 457 p1->p_fd, 458 p2); 459 } 460 } 461 /* 462 * Make a proc table entry for the new process. 463 * Start by zeroing the section of proc that is zero-initialized, 464 * then copy the section that is copied directly from the parent. 465 */ 466 td2 = FIRST_THREAD_IN_PROC(p2); 467 kg2 = FIRST_KSEGRP_IN_PROC(p2); 468 469 /* Allocate and switch to an alternate kstack if specified. */ 470 if (pages != 0) 471 vm_thread_new_altkstack(td2, pages); 472 473 PROC_LOCK(p2); 474 PROC_LOCK(p1); 475 476 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 477 478 bzero(&p2->p_startzero, 479 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 480 bzero(&td2->td_startzero, 481 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 482 bzero(&kg2->kg_startzero, 483 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 484 485 bcopy(&p1->p_startcopy, &p2->p_startcopy, 486 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 487 bcopy(&td->td_startcopy, &td2->td_startcopy, 488 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 489 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 490 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 491 #undef RANGEOF 492 493 td2->td_sigstk = td->td_sigstk; 494 495 /* 496 * Duplicate sub-structures as needed. 497 * Increase reference counts on shared objects. 498 * The p_stats substruct is set in vm_forkproc. 499 */ 500 p2->p_flag = 0; 501 if (p1->p_flag & P_PROFIL) 502 startprofclock(p2); 503 mtx_lock_spin(&sched_lock); 504 p2->p_sflag = PS_INMEM; 505 /* 506 * Allow the scheduler to adjust the priority of the child and 507 * parent while we hold the sched_lock. 508 */ 509 sched_fork(td, td2); 510 511 mtx_unlock_spin(&sched_lock); 512 p2->p_ucred = crhold(td->td_ucred); 513 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 514 515 pargs_hold(p2->p_args); 516 517 if (flags & RFSIGSHARE) { 518 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 519 } else { 520 sigacts_copy(newsigacts, p1->p_sigacts); 521 p2->p_sigacts = newsigacts; 522 } 523 if (flags & RFLINUXTHPN) 524 p2->p_sigparent = SIGUSR1; 525 else 526 p2->p_sigparent = SIGCHLD; 527 528 p2->p_textvp = p1->p_textvp; 529 p2->p_fd = fd; 530 p2->p_fdtol = fdtol; 531 532 /* 533 * p_limit is copy-on-write. Bump its refcount. 534 */ 535 p2->p_limit = lim_hold(p1->p_limit); 536 537 /* 538 * We don't need to prevent the parent from being swapped out just yet, 539 * but this is a convenient point to block to swap it in if needed. 540 */ 541 PROC_UNLOCK(p2); 542 _PHOLD(p1); 543 PROC_UNLOCK(p1); 544 545 /* Bump references to the text vnode (for procfs) */ 546 if (p2->p_textvp) 547 vref(p2->p_textvp); 548 549 /* 550 * Set up linkage for kernel based threading. 551 */ 552 if ((flags & RFTHREAD) != 0) { 553 mtx_lock(&ppeers_lock); 554 p2->p_peers = p1->p_peers; 555 p1->p_peers = p2; 556 p2->p_leader = p1->p_leader; 557 mtx_unlock(&ppeers_lock); 558 PROC_LOCK(p1->p_leader); 559 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 560 PROC_UNLOCK(p1->p_leader); 561 /* 562 * The task leader is exiting, so process p1 is 563 * going to be killed shortly. Since p1 obviously 564 * isn't dead yet, we know that the leader is either 565 * sending SIGKILL's to all the processes in this 566 * task or is sleeping waiting for all the peers to 567 * exit. We let p1 complete the fork, but we need 568 * to go ahead and kill the new process p2 since 569 * the task leader may not get a chance to send 570 * SIGKILL to it. We leave it on the list so that 571 * the task leader will wait for this new process 572 * to commit suicide. 573 */ 574 PROC_LOCK(p2); 575 psignal(p2, SIGKILL); 576 PROC_UNLOCK(p2); 577 } else 578 PROC_UNLOCK(p1->p_leader); 579 } else { 580 p2->p_peers = NULL; 581 p2->p_leader = p2; 582 } 583 584 sx_xlock(&proctree_lock); 585 PGRP_LOCK(p1->p_pgrp); 586 PROC_LOCK(p2); 587 PROC_LOCK(p1); 588 589 /* 590 * Preserve some more flags in subprocess. P_PROFIL has already 591 * been preserved. 592 */ 593 p2->p_flag |= p1->p_flag & P_SUGID; 594 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 595 SESS_LOCK(p1->p_session); 596 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 597 p2->p_flag |= P_CONTROLT; 598 SESS_UNLOCK(p1->p_session); 599 if (flags & RFPPWAIT) 600 p2->p_flag |= P_PPWAIT; 601 602 p2->p_pgrp = p1->p_pgrp; 603 LIST_INSERT_AFTER(p1, p2, p_pglist); 604 PGRP_UNLOCK(p1->p_pgrp); 605 LIST_INIT(&p2->p_children); 606 607 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 608 609 #ifdef KTRACE 610 /* 611 * Copy traceflag and tracefile if enabled. 612 */ 613 mtx_lock(&ktrace_mtx); 614 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 615 if (p1->p_traceflag & KTRFAC_INHERIT) { 616 p2->p_traceflag = p1->p_traceflag; 617 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 618 VREF(p2->p_tracevp); 619 KASSERT(p1->p_tracecred != NULL, 620 ("ktrace vnode with no cred")); 621 p2->p_tracecred = crhold(p1->p_tracecred); 622 } 623 } 624 mtx_unlock(&ktrace_mtx); 625 #endif 626 627 /* 628 * If PF_FORK is set, the child process inherits the 629 * procfs ioctl flags from its parent. 630 */ 631 if (p1->p_pfsflags & PF_FORK) { 632 p2->p_stops = p1->p_stops; 633 p2->p_pfsflags = p1->p_pfsflags; 634 } 635 636 /* 637 * This begins the section where we must prevent the parent 638 * from being swapped. We already got a hold on the parent earlier. 639 */ 640 PROC_UNLOCK(p1); 641 642 /* 643 * Attach the new process to its parent. 644 * 645 * If RFNOWAIT is set, the newly created process becomes a child 646 * of init. This effectively disassociates the child from the 647 * parent. 648 */ 649 if (flags & RFNOWAIT) 650 pptr = initproc; 651 else 652 pptr = p1; 653 p2->p_pptr = pptr; 654 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 655 sx_xunlock(&proctree_lock); 656 657 /* Inform accounting that we have forked. */ 658 p2->p_acflag = AFORK; 659 PROC_UNLOCK(p2); 660 661 /* 662 * Finish creating the child process. It will return via a different 663 * execution path later. (ie: directly into user mode) 664 */ 665 vm_forkproc(td, p2, td2, flags); 666 667 if (flags == (RFFDG | RFPROC)) { 668 atomic_add_int(&cnt.v_forks, 1); 669 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize + 670 p2->p_vmspace->vm_ssize); 671 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 672 atomic_add_int(&cnt.v_vforks, 1); 673 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 674 p2->p_vmspace->vm_ssize); 675 } else if (p1 == &proc0) { 676 atomic_add_int(&cnt.v_kthreads, 1); 677 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 678 p2->p_vmspace->vm_ssize); 679 } else { 680 atomic_add_int(&cnt.v_rforks, 1); 681 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 682 p2->p_vmspace->vm_ssize); 683 } 684 685 /* 686 * Both processes are set up, now check if any loadable modules want 687 * to adjust anything. 688 * What if they have an error? XXX 689 */ 690 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 691 692 /* 693 * Set the child start time and mark the process as being complete. 694 */ 695 microuptime(&p2->p_stats->p_start); 696 mtx_lock_spin(&sched_lock); 697 p2->p_state = PRS_NORMAL; 698 699 /* 700 * If RFSTOPPED not requested, make child runnable and add to 701 * run queue. 702 */ 703 if ((flags & RFSTOPPED) == 0) { 704 TD_SET_CAN_RUN(td2); 705 setrunqueue(td2, SRQ_BORING); 706 } 707 mtx_unlock_spin(&sched_lock); 708 709 /* 710 * Now can be swapped. 711 */ 712 PROC_LOCK(p1); 713 _PRELE(p1); 714 715 /* 716 * Tell any interested parties about the new process. 717 */ 718 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 719 720 PROC_UNLOCK(p1); 721 722 /* 723 * Preserve synchronization semantics of vfork. If waiting for 724 * child to exec or exit, set P_PPWAIT on child, and sleep on our 725 * proc (in case of exit). 726 */ 727 PROC_LOCK(p2); 728 while (p2->p_flag & P_PPWAIT) 729 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 730 PROC_UNLOCK(p2); 731 732 /* 733 * If other threads are waiting, let them continue now. 734 */ 735 if (p1->p_flag & P_HADTHREADS) { 736 PROC_LOCK(p1); 737 thread_single_end(); 738 PROC_UNLOCK(p1); 739 } 740 741 /* 742 * Return child proc pointer to parent. 743 */ 744 *procp = p2; 745 return (0); 746 fail: 747 sx_sunlock(&proctree_lock); 748 if (ppsratecheck(&lastfail, &curfail, 1)) 749 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 750 uid); 751 sx_xunlock(&allproc_lock); 752 #ifdef MAC 753 mac_destroy_proc(newproc); 754 #endif 755 uma_zfree(proc_zone, newproc); 756 if (p1->p_flag & P_HADTHREADS) { 757 PROC_LOCK(p1); 758 thread_single_end(); 759 PROC_UNLOCK(p1); 760 } 761 tsleep(&forksleep, PUSER, "fork", hz / 2); 762 return (error); 763 } 764 765 /* 766 * Handle the return of a child process from fork1(). This function 767 * is called from the MD fork_trampoline() entry point. 768 */ 769 void 770 fork_exit(callout, arg, frame) 771 void (*callout)(void *, struct trapframe *); 772 void *arg; 773 struct trapframe *frame; 774 { 775 struct proc *p; 776 struct thread *td; 777 778 /* 779 * Finish setting up thread glue so that it begins execution in a 780 * non-nested critical section with sched_lock held but not recursed. 781 */ 782 td = curthread; 783 p = td->td_proc; 784 td->td_oncpu = PCPU_GET(cpuid); 785 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 786 787 sched_lock.mtx_lock = (uintptr_t)td; 788 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 789 cpu_critical_fork_exit(); 790 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 791 td, td->td_sched, p->p_pid, p->p_comm); 792 793 /* 794 * Processes normally resume in mi_switch() after being 795 * cpu_switch()'ed to, but when children start up they arrive here 796 * instead, so we must do much the same things as mi_switch() would. 797 */ 798 799 if ((td = PCPU_GET(deadthread))) { 800 PCPU_SET(deadthread, NULL); 801 thread_stash(td); 802 } 803 td = curthread; 804 mtx_unlock_spin(&sched_lock); 805 806 /* 807 * cpu_set_fork_handler intercepts this function call to 808 * have this call a non-return function to stay in kernel mode. 809 * initproc has its own fork handler, but it does return. 810 */ 811 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 812 callout(arg, frame); 813 814 /* 815 * Check if a kernel thread misbehaved and returned from its main 816 * function. 817 */ 818 PROC_LOCK(p); 819 if (p->p_flag & P_KTHREAD) { 820 PROC_UNLOCK(p); 821 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 822 p->p_comm, p->p_pid); 823 kthread_exit(0); 824 } 825 PROC_UNLOCK(p); 826 mtx_assert(&Giant, MA_NOTOWNED); 827 } 828 829 /* 830 * Simplified back end of syscall(), used when returning from fork() 831 * directly into user mode. Giant is not held on entry, and must not 832 * be held on return. This function is passed in to fork_exit() as the 833 * first parameter and is called when returning to a new userland process. 834 */ 835 void 836 fork_return(td, frame) 837 struct thread *td; 838 struct trapframe *frame; 839 { 840 841 userret(td, frame, 0); 842 #ifdef KTRACE 843 if (KTRPOINT(td, KTR_SYSRET)) 844 ktrsysret(SYS_fork, 0, 0); 845 #endif 846 mtx_assert(&Giant, MA_NOTOWNED); 847 } 848