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 fdtmp = fdinit(td->td_proc->p_fd); 229 fdfree(td); 230 p1->p_fd = fdtmp; 231 } 232 233 /* 234 * Unshare file descriptors (from parent). 235 */ 236 if (flags & RFFDG) { 237 FILEDESC_LOCK_FAST(p1->p_fd); 238 if (p1->p_fd->fd_refcnt > 1) { 239 struct filedesc *newfd; 240 241 FILEDESC_UNLOCK_FAST(p1->p_fd); 242 newfd = fdcopy(p1->p_fd); 243 fdfree(td); 244 p1->p_fd = newfd; 245 } else 246 FILEDESC_UNLOCK_FAST(p1->p_fd); 247 } 248 *procp = NULL; 249 return (0); 250 } 251 252 /* 253 * Note 1:1 allows for forking with one thread coming out on the 254 * other side with the expectation that the process is about to 255 * exec. 256 */ 257 if (p1->p_flag & P_HADTHREADS) { 258 /* 259 * Idle the other threads for a second. 260 * Since the user space is copied, it must remain stable. 261 * In addition, all threads (from the user perspective) 262 * need to either be suspended or in the kernel, 263 * where they will try restart in the parent and will 264 * be aborted in the child. 265 */ 266 PROC_LOCK(p1); 267 if (thread_single(SINGLE_NO_EXIT)) { 268 /* Abort. Someone else is single threading before us. */ 269 PROC_UNLOCK(p1); 270 return (ERESTART); 271 } 272 PROC_UNLOCK(p1); 273 /* 274 * All other activity in this process 275 * is now suspended at the user boundary, 276 * (or other safe places if we think of any). 277 */ 278 } 279 280 /* Allocate new proc. */ 281 newproc = uma_zalloc(proc_zone, M_WAITOK); 282 #ifdef MAC 283 mac_init_proc(newproc); 284 #endif 285 knlist_init(&newproc->p_klist, &newproc->p_mtx); 286 287 /* We have to lock the process tree while we look for a pid. */ 288 sx_slock(&proctree_lock); 289 290 /* 291 * Although process entries are dynamically created, we still keep 292 * a global limit on the maximum number we will create. Don't allow 293 * a nonprivileged user to use the last ten processes; don't let root 294 * exceed the limit. The variable nprocs is the current number of 295 * processes, maxproc is the limit. 296 */ 297 sx_xlock(&allproc_lock); 298 uid = td->td_ucred->cr_ruid; 299 if ((nprocs >= maxproc - 10 && 300 suser_cred(td->td_ucred, SUSER_RUID) != 0) || 301 nprocs >= maxproc) { 302 error = EAGAIN; 303 goto fail; 304 } 305 306 /* 307 * Increment the count of procs running with this uid. Don't allow 308 * a nonprivileged user to exceed their current limit. 309 */ 310 PROC_LOCK(p1); 311 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 312 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0); 313 PROC_UNLOCK(p1); 314 if (!ok) { 315 error = EAGAIN; 316 goto fail; 317 } 318 319 /* 320 * Increment the nprocs resource before blocking can occur. There 321 * are hard-limits as to the number of processes that can run. 322 */ 323 nprocs++; 324 325 /* 326 * Find an unused process ID. We remember a range of unused IDs 327 * ready to use (from lastpid+1 through pidchecked-1). 328 * 329 * If RFHIGHPID is set (used during system boot), do not allocate 330 * low-numbered pids. 331 */ 332 trypid = lastpid + 1; 333 if (flags & RFHIGHPID) { 334 if (trypid < 10) 335 trypid = 10; 336 } else { 337 if (randompid) 338 trypid += arc4random() % randompid; 339 } 340 retry: 341 /* 342 * If the process ID prototype has wrapped around, 343 * restart somewhat above 0, as the low-numbered procs 344 * tend to include daemons that don't exit. 345 */ 346 if (trypid >= PID_MAX) { 347 trypid = trypid % PID_MAX; 348 if (trypid < 100) 349 trypid += 100; 350 pidchecked = 0; 351 } 352 if (trypid >= pidchecked) { 353 int doingzomb = 0; 354 355 pidchecked = PID_MAX; 356 /* 357 * Scan the active and zombie procs to check whether this pid 358 * is in use. Remember the lowest pid that's greater 359 * than trypid, so we can avoid checking for a while. 360 */ 361 p2 = LIST_FIRST(&allproc); 362 again: 363 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 364 PROC_LOCK(p2); 365 while (p2->p_pid == trypid || 366 (p2->p_pgrp != NULL && 367 (p2->p_pgrp->pg_id == trypid || 368 (p2->p_session != NULL && 369 p2->p_session->s_sid == trypid)))) { 370 trypid++; 371 if (trypid >= pidchecked) { 372 PROC_UNLOCK(p2); 373 goto retry; 374 } 375 } 376 if (p2->p_pid > trypid && pidchecked > p2->p_pid) 377 pidchecked = p2->p_pid; 378 if (p2->p_pgrp != NULL) { 379 if (p2->p_pgrp->pg_id > trypid && 380 pidchecked > p2->p_pgrp->pg_id) 381 pidchecked = p2->p_pgrp->pg_id; 382 if (p2->p_session != NULL && 383 p2->p_session->s_sid > trypid && 384 pidchecked > p2->p_session->s_sid) 385 pidchecked = p2->p_session->s_sid; 386 } 387 PROC_UNLOCK(p2); 388 } 389 if (!doingzomb) { 390 doingzomb = 1; 391 p2 = LIST_FIRST(&zombproc); 392 goto again; 393 } 394 } 395 sx_sunlock(&proctree_lock); 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 newsigacts = NULL; 417 else 418 newsigacts = sigacts_alloc(); 419 420 /* 421 * Copy filedesc. 422 */ 423 if (flags & RFCFDG) { 424 fd = fdinit(p1->p_fd); 425 fdtol = NULL; 426 } else if (flags & RFFDG) { 427 fd = fdcopy(p1->p_fd); 428 fdtol = NULL; 429 } else { 430 fd = fdshare(p1->p_fd); 431 if (p1->p_fdtol == NULL) 432 p1->p_fdtol = 433 filedesc_to_leader_alloc(NULL, 434 NULL, 435 p1->p_leader); 436 if ((flags & RFTHREAD) != 0) { 437 /* 438 * Shared file descriptor table and 439 * shared process leaders. 440 */ 441 fdtol = p1->p_fdtol; 442 FILEDESC_LOCK_FAST(p1->p_fd); 443 fdtol->fdl_refcount++; 444 FILEDESC_UNLOCK_FAST(p1->p_fd); 445 } else { 446 /* 447 * Shared file descriptor table, and 448 * different process leaders 449 */ 450 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, 451 p1->p_fd, 452 p2); 453 } 454 } 455 /* 456 * Make a proc table entry for the new process. 457 * Start by zeroing the section of proc that is zero-initialized, 458 * then copy the section that is copied directly from the parent. 459 */ 460 td2 = FIRST_THREAD_IN_PROC(p2); 461 kg2 = FIRST_KSEGRP_IN_PROC(p2); 462 463 /* Allocate and switch to an alternate kstack if specified. */ 464 if (pages != 0) 465 vm_thread_new_altkstack(td2, pages); 466 467 PROC_LOCK(p2); 468 PROC_LOCK(p1); 469 470 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 471 472 bzero(&p2->p_startzero, 473 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 474 bzero(&td2->td_startzero, 475 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 476 bzero(&kg2->kg_startzero, 477 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 478 479 bcopy(&p1->p_startcopy, &p2->p_startcopy, 480 (unsigned) RANGEOF(struct proc, p_startcopy, p_endcopy)); 481 bcopy(&td->td_startcopy, &td2->td_startcopy, 482 (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); 483 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy, 484 (unsigned) RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); 485 #undef RANGEOF 486 487 td2->td_sigstk = td->td_sigstk; 488 489 /* 490 * Duplicate sub-structures as needed. 491 * Increase reference counts on shared objects. 492 * The p_stats substruct is set in vm_forkproc. 493 */ 494 p2->p_flag = 0; 495 if (p1->p_flag & P_PROFIL) 496 startprofclock(p2); 497 mtx_lock_spin(&sched_lock); 498 p2->p_sflag = PS_INMEM; 499 /* 500 * Allow the scheduler to adjust the priority of the child and 501 * parent while we hold the sched_lock. 502 */ 503 sched_fork(td, td2); 504 505 mtx_unlock_spin(&sched_lock); 506 p2->p_ucred = crhold(td->td_ucred); 507 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 508 509 pargs_hold(p2->p_args); 510 511 if (flags & RFSIGSHARE) { 512 p2->p_sigacts = sigacts_hold(p1->p_sigacts); 513 } else { 514 sigacts_copy(newsigacts, p1->p_sigacts); 515 p2->p_sigacts = newsigacts; 516 } 517 if (flags & RFLINUXTHPN) 518 p2->p_sigparent = SIGUSR1; 519 else 520 p2->p_sigparent = SIGCHLD; 521 522 p2->p_textvp = p1->p_textvp; 523 p2->p_fd = fd; 524 p2->p_fdtol = fdtol; 525 526 /* 527 * p_limit is copy-on-write. Bump its refcount. 528 */ 529 p2->p_limit = lim_hold(p1->p_limit); 530 PROC_UNLOCK(p1); 531 PROC_UNLOCK(p2); 532 533 /* Bump references to the text vnode (for procfs) */ 534 if (p2->p_textvp) 535 vref(p2->p_textvp); 536 537 /* 538 * Set up linkage for kernel based threading. 539 */ 540 if ((flags & RFTHREAD) != 0) { 541 mtx_lock(&ppeers_lock); 542 p2->p_peers = p1->p_peers; 543 p1->p_peers = p2; 544 p2->p_leader = p1->p_leader; 545 mtx_unlock(&ppeers_lock); 546 PROC_LOCK(p1->p_leader); 547 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 548 PROC_UNLOCK(p1->p_leader); 549 /* 550 * The task leader is exiting, so process p1 is 551 * going to be killed shortly. Since p1 obviously 552 * isn't dead yet, we know that the leader is either 553 * sending SIGKILL's to all the processes in this 554 * task or is sleeping waiting for all the peers to 555 * exit. We let p1 complete the fork, but we need 556 * to go ahead and kill the new process p2 since 557 * the task leader may not get a chance to send 558 * SIGKILL to it. We leave it on the list so that 559 * the task leader will wait for this new process 560 * to commit suicide. 561 */ 562 PROC_LOCK(p2); 563 psignal(p2, SIGKILL); 564 PROC_UNLOCK(p2); 565 } else 566 PROC_UNLOCK(p1->p_leader); 567 } else { 568 p2->p_peers = NULL; 569 p2->p_leader = p2; 570 } 571 572 sx_xlock(&proctree_lock); 573 PGRP_LOCK(p1->p_pgrp); 574 PROC_LOCK(p2); 575 PROC_LOCK(p1); 576 577 /* 578 * Preserve some more flags in subprocess. P_PROFIL has already 579 * been preserved. 580 */ 581 p2->p_flag |= p1->p_flag & P_SUGID; 582 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK; 583 SESS_LOCK(p1->p_session); 584 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 585 p2->p_flag |= P_CONTROLT; 586 SESS_UNLOCK(p1->p_session); 587 if (flags & RFPPWAIT) 588 p2->p_flag |= P_PPWAIT; 589 590 p2->p_pgrp = p1->p_pgrp; 591 LIST_INSERT_AFTER(p1, p2, p_pglist); 592 PGRP_UNLOCK(p1->p_pgrp); 593 LIST_INIT(&p2->p_children); 594 595 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE); 596 597 #ifdef KTRACE 598 /* 599 * Copy traceflag and tracefile if enabled. 600 */ 601 mtx_lock(&ktrace_mtx); 602 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 603 if (p1->p_traceflag & KTRFAC_INHERIT) { 604 p2->p_traceflag = p1->p_traceflag; 605 if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 606 VREF(p2->p_tracevp); 607 KASSERT(p1->p_tracecred != NULL, 608 ("ktrace vnode with no cred")); 609 p2->p_tracecred = crhold(p1->p_tracecred); 610 } 611 } 612 mtx_unlock(&ktrace_mtx); 613 #endif 614 615 /* 616 * If PF_FORK is set, the child process inherits the 617 * procfs ioctl flags from its parent. 618 */ 619 if (p1->p_pfsflags & PF_FORK) { 620 p2->p_stops = p1->p_stops; 621 p2->p_pfsflags = p1->p_pfsflags; 622 } 623 624 /* 625 * This begins the section where we must prevent the parent 626 * from being swapped. 627 */ 628 _PHOLD(p1); 629 PROC_UNLOCK(p1); 630 631 /* 632 * Attach the new process to its parent. 633 * 634 * If RFNOWAIT is set, the newly created process becomes a child 635 * of init. This effectively disassociates the child from the 636 * parent. 637 */ 638 if (flags & RFNOWAIT) 639 pptr = initproc; 640 else 641 pptr = p1; 642 p2->p_pptr = pptr; 643 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 644 sx_xunlock(&proctree_lock); 645 646 /* Inform accounting that we have forked. */ 647 p2->p_acflag = AFORK; 648 PROC_UNLOCK(p2); 649 650 /* 651 * Finish creating the child process. It will return via a different 652 * execution path later. (ie: directly into user mode) 653 */ 654 vm_forkproc(td, p2, td2, flags); 655 656 if (flags == (RFFDG | RFPROC)) { 657 atomic_add_int(&cnt.v_forks, 1); 658 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize + 659 p2->p_vmspace->vm_ssize); 660 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 661 atomic_add_int(&cnt.v_vforks, 1); 662 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize + 663 p2->p_vmspace->vm_ssize); 664 } else if (p1 == &proc0) { 665 atomic_add_int(&cnt.v_kthreads, 1); 666 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize + 667 p2->p_vmspace->vm_ssize); 668 } else { 669 atomic_add_int(&cnt.v_rforks, 1); 670 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize + 671 p2->p_vmspace->vm_ssize); 672 } 673 674 /* 675 * Both processes are set up, now check if any loadable modules want 676 * to adjust anything. 677 * What if they have an error? XXX 678 */ 679 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags); 680 681 /* 682 * Set the child start time and mark the process as being complete. 683 */ 684 microuptime(&p2->p_stats->p_start); 685 mtx_lock_spin(&sched_lock); 686 p2->p_state = PRS_NORMAL; 687 688 /* 689 * If RFSTOPPED not requested, make child runnable and add to 690 * run queue. 691 */ 692 if ((flags & RFSTOPPED) == 0) { 693 TD_SET_CAN_RUN(td2); 694 setrunqueue(td2, SRQ_BORING); 695 } 696 mtx_unlock_spin(&sched_lock); 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_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid); 708 709 PROC_UNLOCK(p1); 710 711 /* 712 * Preserve synchronization semantics of vfork. If waiting for 713 * child to exec or exit, set P_PPWAIT on child, and sleep on our 714 * proc (in case of exit). 715 */ 716 PROC_LOCK(p2); 717 while (p2->p_flag & P_PPWAIT) 718 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0); 719 PROC_UNLOCK(p2); 720 721 /* 722 * If other threads are waiting, let them continue now. 723 */ 724 if (p1->p_flag & P_HADTHREADS) { 725 PROC_LOCK(p1); 726 thread_single_end(); 727 PROC_UNLOCK(p1); 728 } 729 730 /* 731 * Return child proc pointer to parent. 732 */ 733 *procp = p2; 734 return (0); 735 fail: 736 sx_sunlock(&proctree_lock); 737 if (ppsratecheck(&lastfail, &curfail, 1)) 738 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n", 739 uid); 740 sx_xunlock(&allproc_lock); 741 #ifdef MAC 742 mac_destroy_proc(newproc); 743 #endif 744 uma_zfree(proc_zone, newproc); 745 if (p1->p_flag & P_HADTHREADS) { 746 PROC_LOCK(p1); 747 thread_single_end(); 748 PROC_UNLOCK(p1); 749 } 750 tsleep(&forksleep, PUSER, "fork", hz / 2); 751 return (error); 752 } 753 754 /* 755 * Handle the return of a child process from fork1(). This function 756 * is called from the MD fork_trampoline() entry point. 757 */ 758 void 759 fork_exit(callout, arg, frame) 760 void (*callout)(void *, struct trapframe *); 761 void *arg; 762 struct trapframe *frame; 763 { 764 struct proc *p; 765 struct thread *td; 766 767 /* 768 * Finish setting up thread glue so that it begins execution in a 769 * non-nested critical section with sched_lock held but not recursed. 770 */ 771 td = curthread; 772 p = td->td_proc; 773 td->td_oncpu = PCPU_GET(cpuid); 774 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); 775 776 sched_lock.mtx_lock = (uintptr_t)td; 777 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 778 cpu_critical_fork_exit(); 779 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)", 780 td, td->td_sched, p->p_pid, p->p_comm); 781 782 /* 783 * Processes normally resume in mi_switch() after being 784 * cpu_switch()'ed to, but when children start up they arrive here 785 * instead, so we must do much the same things as mi_switch() would. 786 */ 787 788 if ((td = PCPU_GET(deadthread))) { 789 PCPU_SET(deadthread, NULL); 790 thread_stash(td); 791 } 792 td = curthread; 793 mtx_unlock_spin(&sched_lock); 794 795 /* 796 * cpu_set_fork_handler intercepts this function call to 797 * have this call a non-return function to stay in kernel mode. 798 * initproc has its own fork handler, but it does return. 799 */ 800 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 801 callout(arg, frame); 802 803 /* 804 * Check if a kernel thread misbehaved and returned from its main 805 * function. 806 */ 807 PROC_LOCK(p); 808 if (p->p_flag & P_KTHREAD) { 809 PROC_UNLOCK(p); 810 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 811 p->p_comm, p->p_pid); 812 kthread_exit(0); 813 } 814 PROC_UNLOCK(p); 815 mtx_assert(&Giant, MA_NOTOWNED); 816 } 817 818 /* 819 * Simplified back end of syscall(), used when returning from fork() 820 * directly into user mode. Giant is not held on entry, and must not 821 * be held on return. This function is passed in to fork_exit() as the 822 * first parameter and is called when returning to a new userland process. 823 */ 824 void 825 fork_return(td, frame) 826 struct thread *td; 827 struct trapframe *frame; 828 { 829 830 userret(td, frame, 0); 831 #ifdef KTRACE 832 if (KTRPOINT(td, KTR_SYSRET)) 833 ktrsysret(SYS_fork, 0, 0); 834 #endif 835 mtx_assert(&Giant, MA_NOTOWNED); 836 } 837