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 #include <sys/signalvar.h> 68 69 #include <security/audit/audit.h> 70 71 #include <vm/vm.h> 72 #include <vm/pmap.h> 73 #include <vm/vm_map.h> 74 #include <vm/vm_extern.h> 75 #include <vm/uma.h> 76 77 78 #ifndef _SYS_SYSPROTO_H_ 79 struct fork_args { 80 int dummy; 81 }; 82 #endif 83 84 static int forksleep; /* Place for fork1() to sleep on. */ 85 86 /* 87 * MPSAFE 88 */ 89 /* ARGSUSED */ 90 int 91 fork(td, uap) 92 struct thread *td; 93 struct fork_args *uap; 94 { 95 int error; 96 struct proc *p2; 97 98 error = fork1(td, RFFDG | RFPROC, 0, &p2); 99 if (error == 0) { 100 td->td_retval[0] = p2->p_pid; 101 td->td_retval[1] = 0; 102 } 103 return (error); 104 } 105 106 /* 107 * MPSAFE 108 */ 109 /* ARGSUSED */ 110 int 111 vfork(td, uap) 112 struct thread *td; 113 struct vfork_args *uap; 114 { 115 int error; 116 struct proc *p2; 117 118 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 119 if (error == 0) { 120 td->td_retval[0] = p2->p_pid; 121 td->td_retval[1] = 0; 122 } 123 return (error); 124 } 125 126 /* 127 * MPSAFE 128 */ 129 int 130 rfork(td, uap) 131 struct thread *td; 132 struct rfork_args *uap; 133 { 134 struct proc *p2; 135 int error; 136 137 /* Don't allow kernel-only flags. */ 138 if ((uap->flags & RFKERNELONLY) != 0) 139 return (EINVAL); 140 141 AUDIT_ARG(fflags, uap->flags); 142 error = fork1(td, uap->flags, 0, &p2); 143 if (error == 0) { 144 td->td_retval[0] = p2 ? p2->p_pid : 0; 145 td->td_retval[1] = 0; 146 } 147 return (error); 148 } 149 150 int nprocs = 1; /* process 0 */ 151 int lastpid = 0; 152 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 153 "Last used PID"); 154 155 /* 156 * Random component to lastpid generation. We mix in a random factor to make 157 * it a little harder to predict. We sanity check the modulus value to avoid 158 * doing it in critical paths. Don't let it be too small or we pointlessly 159 * waste randomness entropy, and don't let it be impossibly large. Using a 160 * modulus that is too big causes a LOT more process table scans and slows 161 * down fork processing as the pidchecked caching is defeated. 162 */ 163 static int randompid = 0; 164 165 static int 166 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 167 { 168 int error, pid; 169 170 error = sysctl_wire_old_buffer(req, sizeof(int)); 171 if (error != 0) 172 return(error); 173 sx_xlock(&allproc_lock); 174 pid = randompid; 175 error = sysctl_handle_int(oidp, &pid, 0, req); 176 if (error == 0 && req->newptr != NULL) { 177 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 178 pid = PID_MAX - 100; 179 else if (pid < 2) /* NOP */ 180 pid = 0; 181 else if (pid < 100) /* Make it reasonable */ 182 pid = 100; 183 randompid = pid; 184 } 185 sx_xunlock(&allproc_lock); 186 return (error); 187 } 188 189 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 190 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 191 192 int 193 fork1(td, flags, pages, procp) 194 struct thread *td; 195 int flags; 196 int pages; 197 struct proc **procp; 198 { 199 struct proc *p1, *p2, *pptr; 200 struct proc *newproc; 201 int ok, trypid; 202 static int curfail, pidchecked = 0; 203 static struct timeval lastfail; 204 struct filedesc *fd; 205 struct filedesc_to_leader *fdtol; 206 struct thread *td2; 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 if ((p1->p_flag & P_HADTHREADS) && 223 (flags & (RFCFDG | RFFDG))) { 224 PROC_LOCK(p1); 225 if (thread_single(SINGLE_BOUNDARY)) { 226 PROC_UNLOCK(p1); 227 return (ERESTART); 228 } 229 PROC_UNLOCK(p1); 230 } 231 232 vm_forkproc(td, NULL, NULL, flags); 233 234 /* 235 * Close all file descriptors. 236 */ 237 if (flags & RFCFDG) { 238 struct filedesc *fdtmp; 239 fdtmp = fdinit(td->td_proc->p_fd); 240 fdfree(td); 241 p1->p_fd = fdtmp; 242 } 243 244 /* 245 * Unshare file descriptors (from parent). 246 */ 247 if (flags & RFFDG) 248 fdunshare(p1, td); 249 250 if ((p1->p_flag & P_HADTHREADS) && 251 (flags & (RFCFDG | RFFDG))) { 252 PROC_LOCK(p1); 253 thread_single_end(); 254 PROC_UNLOCK(p1); 255 } 256 *procp = NULL; 257 return (0); 258 } 259 260 /* 261 * Note 1:1 allows for forking with one thread coming out on the 262 * other side with the expectation that the process is about to 263 * exec. 264 */ 265 if (p1->p_flag & P_HADTHREADS) { 266 /* 267 * Idle the other threads for a second. 268 * Since the user space is copied, it must remain stable. 269 * In addition, all threads (from the user perspective) 270 * need to either be suspended or in the kernel, 271 * where they will try restart in the parent and will 272 * be aborted in the child. 273 */ 274 PROC_LOCK(p1); 275 if (thread_single(SINGLE_NO_EXIT)) { 276 /* Abort. Someone else is single threading before us. */ 277 PROC_UNLOCK(p1); 278 return (ERESTART); 279 } 280 PROC_UNLOCK(p1); 281 /* 282 * All other activity in this process 283 * is now suspended at the user boundary, 284 * (or other safe places if we think of any). 285 */ 286 } 287 288 /* Allocate new proc. */ 289 newproc = uma_zalloc(proc_zone, M_WAITOK); 290 #ifdef MAC 291 mac_init_proc(newproc); 292 #endif 293 #ifdef AUDIT 294 audit_proc_alloc(newproc); 295 #endif 296 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL); 297 STAILQ_INIT(&newproc->p_ktr); 298 299 /* We have to lock the process tree while we look for a pid. */ 300 sx_slock(&proctree_lock); 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 if ((nprocs >= maxproc - 10 && 311 suser_cred(td->td_ucred, SUSER_RUID) != 0) || 312 nprocs >= maxproc) { 313 error = EAGAIN; 314 goto fail; 315 } 316 317 /* 318 * Increment the count of procs running with this uid. Don't allow 319 * a nonprivileged user to exceed their current limit. 320 */ 321 error = suser_cred(td->td_ucred, SUSER_RUID | SUSER_ALLOWJAIL); 322 if (error == 0) 323 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0); 324 else { 325 PROC_LOCK(p1); 326 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 327 lim_cur(p1, RLIMIT_NPROC)); 328 PROC_UNLOCK(p1); 329 } 330 if (!ok) { 331 error = EAGAIN; 332 goto fail; 333 } 334 335 /* 336 * Increment the nprocs resource before blocking can occur. There 337 * are hard-limits as to the number of processes that can run. 338 */ 339 nprocs++; 340 341 /* 342 * Find an unused process ID. We remember a range of unused IDs 343 * ready to use (from lastpid+1 through pidchecked-1). 344 * 345 * If RFHIGHPID is set (used during system boot), do not allocate 346 * low-numbered pids. 347 */ 348 trypid = lastpid + 1; 349 if (flags & RFHIGHPID) { 350 if (trypid < 10) 351 trypid = 10; 352 } else { 353 if (randompid) 354 trypid += arc4random() % randompid; 355 } 356 retry: 357 /* 358 * If the process ID prototype has wrapped around, 359 * restart somewhat above 0, as the low-numbered procs 360 * tend to include daemons that don't exit. 361 */ 362 if (trypid >= PID_MAX) { 363 trypid = trypid % PID_MAX; 364 if (trypid < 100) 365 trypid += 100; 366 pidchecked = 0; 367 } 368 if (trypid >= pidchecked) { 369 int doingzomb = 0; 370 371 pidchecked = PID_MAX; 372 /* 373 * Scan the active and zombie procs to check whether this pid 374 * is in use. Remember the lowest pid that's greater 375 * than trypid, so we can avoid checking for a while. 376 */ 377 p2 = LIST_FIRST(&allproc); 378 again: 379 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 380 PROC_LOCK(p2); 381 while (p2->p_pid == trypid || 382 (p2->p_pgrp != NULL && 383 (p2->p_pgrp->pg_id == trypid || 384 (p2->p_session != NULL && 385 p2->p_session->s_sid == trypid)))) { 386 trypid++; 387 if (trypid >= pidchecked) { 388 PROC_UNLOCK(p2); 389 goto retry; 390 } 391 } 392 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 393 pidchecked = p2->p_pid; 394 if (p2->p_pgrp != NULL) { 395 if (p2->p_pgrp->pg_id > trypid && 396 pidchecked > p2->p_pgrp->pg_id) 397 pidchecked = p2->p_pgrp->pg_id; 398 if (p2->p_session != NULL && 399 p2->p_session->s_sid > trypid && 400 pidchecked > p2->p_session->s_sid) 401 pidchecked = p2->p_session->s_sid; 402 } 403 PROC_UNLOCK(p2); 404 } 405 if (!doingzomb) { 406 doingzomb = 1; 407 p2 = LIST_FIRST(&zombproc); 408 goto again; 409 } 410 } 411 sx_sunlock(&proctree_lock); 412 413 /* 414 * RFHIGHPID does not mess with the lastpid counter during boot. 415 */ 416 if (flags & RFHIGHPID) 417 pidchecked = 0; 418 else 419 lastpid = trypid; 420 421 p2 = newproc; 422 p2->p_state = PRS_NEW; /* protect against others */ 423 p2->p_pid = trypid; 424 AUDIT_ARG(pid, p2->p_pid); 425 LIST_INSERT_HEAD(&allproc, p2, p_list); 426 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 427 sx_xunlock(&allproc_lock); 428 429 /* 430 * Malloc things while we don't hold any locks. 431 */ 432 if (flags & RFSIGSHARE) 433 newsigacts = NULL; 434 else 435 newsigacts = sigacts_alloc(); 436 437 /* 438 * Copy filedesc. 439 */ 440 if (flags & RFCFDG) { 441 fd = fdinit(p1->p_fd); 442 fdtol = NULL; 443 } else if (flags & RFFDG) { 444 fd = fdcopy(p1->p_fd); 445 fdtol = NULL; 446 } else { 447 fd = fdshare(p1->p_fd); 448 if (p1->p_fdtol == NULL) 449 p1->p_fdtol = 450 filedesc_to_leader_alloc(NULL, 451 NULL, 452 p1->p_leader); 453 if ((flags & RFTHREAD) != 0) { 454 /* 455 * Shared file descriptor table and 456 * shared process leaders. 457 */ 458 fdtol = p1->p_fdtol; 459 FILEDESC_LOCK_FAST(p1->p_fd); 460 fdtol->fdl_refcount++; 461 FILEDESC_UNLOCK_FAST(p1->p_fd); 462 } else { 463 /* 464 * Shared file descriptor table, and 465 * different process leaders 466 */ 467 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 468 p1->p_fd, 469 p2); 470 } 471 } 472 /* 473 * Make a proc table entry for the new process. 474 * Start by zeroing the section of proc that is zero-initialized, 475 * then copy the section that is copied directly from the parent. 476 */ 477 td2 = FIRST_THREAD_IN_PROC(p2); 478 kg2 = FIRST_KSEGRP_IN_PROC(p2); 479 480 /* Allocate and switch to an alternate kstack if specified. */ 481 if (pages != 0) 482 vm_thread_new_altkstack(td2, pages); 483 484 PROC_LOCK(p2); 485 PROC_LOCK(p1); 486 487 bzero(&p2->p_startzero, 488 __rangeof(struct proc, p_startzero, p_endzero)); 489 bzero(&td2->td_startzero, 490 __rangeof(struct thread, td_startzero, td_endzero)); 491 bzero(&kg2->kg_startzero, 492 __rangeof(struct ksegrp, kg_startzero, kg_endzero)); 493 494 bcopy(&p1->p_startcopy, &p2->p_startcopy, 495 __rangeof(struct proc, p_startcopy, p_endcopy)); 496 bcopy(&td->td_startcopy, &td2->td_startcopy, 497 __rangeof(struct thread, td_startcopy, td_endcopy)); 498 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 499 __rangeof(struct ksegrp, kg_startcopy, kg_endcopy)); 500 501 td2->td_sigstk = td->td_sigstk; 502 td2->td_sigmask = td->td_sigmask; 503 504 /* 505 * Duplicate sub-structures as needed. 506 * Increase reference counts on shared objects. 507 */ 508 p2->p_flag = 0; 509 if (p1->p_flag & P_PROFIL) 510 startprofclock(p2); 511 mtx_lock_spin(&sched_lock); 512 p2->p_sflag = PS_INMEM; 513 /* 514 * Allow the scheduler to adjust the priority of the child and 515 * parent while we hold the sched_lock. 516 */ 517 sched_fork(td, td2); 518 519 mtx_unlock_spin(&sched_lock); 520 p2->p_ucred = crhold(td->td_ucred); 521 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 522 #ifdef AUDIT 523 audit_proc_fork(p1, p2); 524 #endif 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); 546 547 pstats_fork(p1->p_stats, p2->p_stats); 548 549 PROC_UNLOCK(p1); 550 PROC_UNLOCK(p2); 551 552 /* Bump references to the text vnode (for procfs) */ 553 if (p2->p_textvp) 554 vref(p2->p_textvp); 555 556 /* 557 * Set up linkage for kernel based threading. 558 */ 559 if ((flags & RFTHREAD) != 0) { 560 mtx_lock(&ppeers_lock); 561 p2->p_peers = p1->p_peers; 562 p1->p_peers = p2; 563 p2->p_leader = p1->p_leader; 564 mtx_unlock(&ppeers_lock); 565 PROC_LOCK(p1->p_leader); 566 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 567 PROC_UNLOCK(p1->p_leader); 568 /* 569 * The task leader is exiting, so process p1 is 570 * going to be killed shortly. Since p1 obviously 571 * isn't dead yet, we know that the leader is either 572 * sending SIGKILL's to all the processes in this 573 * task or is sleeping waiting for all the peers to 574 * exit. We let p1 complete the fork, but we need 575 * to go ahead and kill the new process p2 since 576 * the task leader may not get a chance to send 577 * SIGKILL to it. We leave it on the list so that 578 * the task leader will wait for this new process 579 * to commit suicide. 580 */ 581 PROC_LOCK(p2); 582 psignal(p2, SIGKILL); 583 PROC_UNLOCK(p2); 584 } else 585 PROC_UNLOCK(p1->p_leader); 586 } else { 587 p2->p_peers = NULL; 588 p2->p_leader = p2; 589 } 590 591 sx_xlock(&proctree_lock); 592 PGRP_LOCK(p1->p_pgrp); 593 PROC_LOCK(p2); 594 PROC_LOCK(p1); 595 596 /* 597 * Preserve some more flags in subprocess. P_PROFIL has already 598 * been preserved. 599 */ 600 p2->p_flag |= p1->p_flag & P_SUGID; 601 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 602 SESS_LOCK(p1->p_session); 603 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 604 p2->p_flag |= P_CONTROLT; 605 SESS_UNLOCK(p1->p_session); 606 if (flags & RFPPWAIT) 607 p2->p_flag |= P_PPWAIT; 608 609 p2->p_pgrp = p1->p_pgrp; 610 LIST_INSERT_AFTER(p1, p2, p_pglist); 611 PGRP_UNLOCK(p1->p_pgrp); 612 LIST_INIT(&p2->p_children); 613 614 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 615 616 #ifdef KTRACE 617 /* 618 * Copy traceflag and tracefile if enabled. 619 */ 620 mtx_lock(&ktrace_mtx); 621 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 622 if (p1->p_traceflag & KTRFAC_INHERIT) { 623 p2->p_traceflag = p1->p_traceflag; 624 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 625 VREF(p2->p_tracevp); 626 KASSERT(p1->p_tracecred != NULL, 627 ("ktrace vnode with no cred")); 628 p2->p_tracecred = crhold(p1->p_tracecred); 629 } 630 } 631 mtx_unlock(&ktrace_mtx); 632 #endif 633 634 /* 635 * If PF_FORK is set, the child process inherits the 636 * procfs ioctl flags from its parent. 637 */ 638 if (p1->p_pfsflags & PF_FORK) { 639 p2->p_stops = p1->p_stops; 640 p2->p_pfsflags = p1->p_pfsflags; 641 } 642 643 /* 644 * This begins the section where we must prevent the parent 645 * from being swapped. 646 */ 647 _PHOLD(p1); 648 PROC_UNLOCK(p1); 649 650 /* 651 * Attach the new process to its parent. 652 * 653 * If RFNOWAIT is set, the newly created process becomes a child 654 * of init. This effectively disassociates the child from the 655 * parent. 656 */ 657 if (flags & RFNOWAIT) 658 pptr = initproc; 659 else 660 pptr = p1; 661 p2->p_pptr = pptr; 662 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 663 sx_xunlock(&proctree_lock); 664 665 /* Inform accounting that we have forked. */ 666 p2->p_acflag = AFORK; 667 PROC_UNLOCK(p2); 668 669 /* 670 * Finish creating the child process. It will return via a different 671 * execution path later. (ie: directly into user mode) 672 */ 673 vm_forkproc(td, p2, td2, flags); 674 675 if (flags == (RFFDG | RFPROC)) { 676 atomic_add_int(&cnt.v_forks, 1); 677 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize + 678 p2->p_vmspace->vm_ssize); 679 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 680 atomic_add_int(&cnt.v_vforks, 1); 681 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 682 p2->p_vmspace->vm_ssize); 683 } else if (p1 == &proc0) { 684 atomic_add_int(&cnt.v_kthreads, 1); 685 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 686 p2->p_vmspace->vm_ssize); 687 } else { 688 atomic_add_int(&cnt.v_rforks, 1); 689 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 690 p2->p_vmspace->vm_ssize); 691 } 692 693 /* 694 * Both processes are set up, now check if any loadable modules want 695 * to adjust anything. 696 * What if they have an error? XXX 697 */ 698 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 699 700 /* 701 * Set the child start time and mark the process as being complete. 702 */ 703 microuptime(&p2->p_stats->p_start); 704 mtx_lock_spin(&sched_lock); 705 p2->p_state = PRS_NORMAL; 706 707 /* 708 * If RFSTOPPED not requested, make child runnable and add to 709 * run queue. 710 */ 711 if ((flags & RFSTOPPED) == 0) { 712 TD_SET_CAN_RUN(td2); 713 setrunqueue(td2, SRQ_BORING); 714 } 715 mtx_unlock_spin(&sched_lock); 716 717 /* 718 * Now can be swapped. 719 */ 720 PROC_LOCK(p1); 721 _PRELE(p1); 722 723 /* 724 * Tell any interested parties about the new process. 725 */ 726 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 727 728 PROC_UNLOCK(p1); 729 730 /* 731 * Preserve synchronization semantics of vfork. If waiting for 732 * child to exec or exit, set P_PPWAIT on child, and sleep on our 733 * proc (in case of exit). 734 */ 735 PROC_LOCK(p2); 736 while (p2->p_flag & P_PPWAIT) 737 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 738 PROC_UNLOCK(p2); 739 740 /* 741 * If other threads are waiting, let them continue now. 742 */ 743 if (p1->p_flag & P_HADTHREADS) { 744 PROC_LOCK(p1); 745 thread_single_end(); 746 PROC_UNLOCK(p1); 747 } 748 749 /* 750 * Return child proc pointer to parent. 751 */ 752 *procp = p2; 753 return (0); 754 fail: 755 sx_sunlock(&proctree_lock); 756 if (ppsratecheck(&lastfail, &curfail, 1)) 757 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 758 td->td_ucred->cr_ruid); 759 sx_xunlock(&allproc_lock); 760 #ifdef MAC 761 mac_destroy_proc(newproc); 762 #endif 763 #ifdef AUDIT 764 audit_proc_free(newproc); 765 #endif 766 uma_zfree(proc_zone, newproc); 767 if (p1->p_flag & P_HADTHREADS) { 768 PROC_LOCK(p1); 769 thread_single_end(); 770 PROC_UNLOCK(p1); 771 } 772 tsleep(&forksleep, PUSER, "fork", hz / 2); 773 return (error); 774 } 775 776 /* 777 * Handle the return of a child process from fork1(). This function 778 * is called from the MD fork_trampoline() entry point. 779 */ 780 void 781 fork_exit(callout, arg, frame) 782 void (*callout)(void *, struct trapframe *); 783 void *arg; 784 struct trapframe *frame; 785 { 786 struct proc *p; 787 struct thread *td; 788 789 /* 790 * Finish setting up thread glue so that it begins execution in a 791 * non-nested critical section with sched_lock held but not recursed. 792 */ 793 td = curthread; 794 p = td->td_proc; 795 td->td_oncpu = PCPU_GET(cpuid); 796 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 797 798 sched_lock.mtx_lock = (uintptr_t)td; 799 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 800 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 801 td, td->td_sched, p->p_pid, p->p_comm); 802 803 /* 804 * Processes normally resume in mi_switch() after being 805 * cpu_switch()'ed to, but when children start up they arrive here 806 * instead, so we must do much the same things as mi_switch() would. 807 */ 808 809 if ((td = PCPU_GET(deadthread))) { 810 PCPU_SET(deadthread, NULL); 811 thread_stash(td); 812 } 813 td = curthread; 814 mtx_unlock_spin(&sched_lock); 815 816 /* 817 * cpu_set_fork_handler intercepts this function call to 818 * have this call a non-return function to stay in kernel mode. 819 * initproc has its own fork handler, but it does return. 820 */ 821 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 822 callout(arg, frame); 823 824 /* 825 * Check if a kernel thread misbehaved and returned from its main 826 * function. 827 */ 828 if (p->p_flag & P_KTHREAD) { 829 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 830 p->p_comm, p->p_pid); 831 kthread_exit(0); 832 } 833 mtx_assert(&Giant, MA_NOTOWNED); 834 } 835 836 /* 837 * Simplified back end of syscall(), used when returning from fork() 838 * directly into user mode. Giant is not held on entry, and must not 839 * be held on return. This function is passed in to fork_exit() as the 840 * first parameter and is called when returning to a new userland process. 841 */ 842 void 843 fork_return(td, frame) 844 struct thread *td; 845 struct trapframe *frame; 846 { 847 848 userret(td, frame); 849 #ifdef KTRACE 850 if (KTRPOINT(td, KTR_SYSRET)) 851 ktrsysret(SYS_fork, 0, 0); 852 #endif 853 mtx_assert(&Giant, MA_NOTOWNED); 854 } 855