1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94 39 * $FreeBSD$ 40 */ 41 42 #include "opt_ktrace.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/sysproto.h> 47 #include <sys/filedesc.h> 48 #include <sys/kernel.h> 49 #include <sys/sysctl.h> 50 #include <sys/lock.h> 51 #include <sys/malloc.h> 52 #include <sys/mutex.h> 53 #include <sys/proc.h> 54 #include <sys/pioctl.h> 55 #include <sys/resourcevar.h> 56 #include <sys/sched.h> 57 #include <sys/syscall.h> 58 #include <sys/vnode.h> 59 #include <sys/acct.h> 60 #include <sys/ktr.h> 61 #include <sys/ktrace.h> 62 #include <sys/kthread.h> 63 #include <sys/unistd.h> 64 #include <sys/jail.h> 65 #include <sys/sx.h> 66 67 #include <vm/vm.h> 68 #include <vm/pmap.h> 69 #include <vm/vm_map.h> 70 #include <vm/vm_extern.h> 71 #include <vm/uma.h> 72 73 #include <sys/vmmeter.h> 74 #include <sys/user.h> 75 #include <machine/critical.h> 76 77 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback"); 78 79 /* 80 * These are the stuctures used to create a callout list for things to do 81 * when forking a process 82 */ 83 struct forklist { 84 forklist_fn function; 85 TAILQ_ENTRY(forklist) next; 86 }; 87 88 static struct sx fork_list_lock; 89 90 TAILQ_HEAD(forklist_head, forklist); 91 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list); 92 93 #ifndef _SYS_SYSPROTO_H_ 94 struct fork_args { 95 int dummy; 96 }; 97 #endif 98 99 int forksleep; /* Place for fork1() to sleep on. */ 100 101 static void 102 init_fork_list(void *data __unused) 103 { 104 105 sx_init(&fork_list_lock, "fork list"); 106 } 107 SYSINIT(fork_list, SI_SUB_INTRINSIC, SI_ORDER_ANY, init_fork_list, NULL); 108 109 /* 110 * MPSAFE 111 */ 112 /* ARGSUSED */ 113 int 114 fork(td, uap) 115 struct thread *td; 116 struct fork_args *uap; 117 { 118 int error; 119 struct proc *p2; 120 121 mtx_lock(&Giant); 122 error = fork1(td, RFFDG | RFPROC, 0, &p2); 123 if (error == 0) { 124 td->td_retval[0] = p2->p_pid; 125 td->td_retval[1] = 0; 126 } 127 mtx_unlock(&Giant); 128 return error; 129 } 130 131 /* 132 * MPSAFE 133 */ 134 /* ARGSUSED */ 135 int 136 vfork(td, uap) 137 struct thread *td; 138 struct vfork_args *uap; 139 { 140 int error; 141 struct proc *p2; 142 143 mtx_lock(&Giant); 144 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2); 145 if (error == 0) { 146 td->td_retval[0] = p2->p_pid; 147 td->td_retval[1] = 0; 148 } 149 mtx_unlock(&Giant); 150 return error; 151 } 152 153 /* 154 * MPSAFE 155 */ 156 int 157 rfork(td, uap) 158 struct thread *td; 159 struct rfork_args *uap; 160 { 161 int error; 162 struct proc *p2; 163 164 /* Don't allow kernel only flags. */ 165 if ((uap->flags & RFKERNELONLY) != 0) 166 return (EINVAL); 167 mtx_lock(&Giant); 168 error = fork1(td, uap->flags, 0, &p2); 169 if (error == 0) { 170 td->td_retval[0] = p2 ? p2->p_pid : 0; 171 td->td_retval[1] = 0; 172 } 173 mtx_unlock(&Giant); 174 return error; 175 } 176 177 178 int nprocs = 1; /* process 0 */ 179 int lastpid = 0; 180 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0, 181 "Last used PID"); 182 183 /* 184 * Random component to lastpid generation. We mix in a random factor to make 185 * it a little harder to predict. We sanity check the modulus value to avoid 186 * doing it in critical paths. Don't let it be too small or we pointlessly 187 * waste randomness entropy, and don't let it be impossibly large. Using a 188 * modulus that is too big causes a LOT more process table scans and slows 189 * down fork processing as the pidchecked caching is defeated. 190 */ 191 static int randompid = 0; 192 193 static int 194 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 195 { 196 int error, pid; 197 198 sysctl_wire_old_buffer(req, sizeof(int)); 199 sx_xlock(&allproc_lock); 200 pid = randompid; 201 error = sysctl_handle_int(oidp, &pid, 0, req); 202 if (error == 0 && req->newptr != NULL) { 203 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 204 pid = PID_MAX - 100; 205 else if (pid < 2) /* NOP */ 206 pid = 0; 207 else if (pid < 100) /* Make it reasonable */ 208 pid = 100; 209 randompid = pid; 210 } 211 sx_xunlock(&allproc_lock); 212 return (error); 213 } 214 215 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 216 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 217 218 int 219 fork1(td, flags, pages, procp) 220 struct thread *td; /* parent proc */ 221 int flags; 222 int pages; 223 struct proc **procp; /* child proc */ 224 { 225 struct proc *p2, *pptr; 226 uid_t uid; 227 struct proc *newproc; 228 int trypid; 229 int ok; 230 static int pidchecked = 0; 231 struct forklist *ep; 232 struct filedesc *fd; 233 struct proc *p1 = td->td_proc; 234 struct thread *td2; 235 struct kse *ke2; 236 struct ksegrp *kg2; 237 struct sigacts *newsigacts; 238 struct procsig *newprocsig; 239 int error; 240 241 GIANT_REQUIRED; 242 243 /* Can't copy and clear */ 244 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG)) 245 return (EINVAL); 246 247 /* 248 * Here we don't create a new process, but we divorce 249 * certain parts of a process from itself. 250 */ 251 if ((flags & RFPROC) == 0) { 252 vm_forkproc(td, NULL, NULL, flags); 253 254 /* 255 * Close all file descriptors. 256 */ 257 if (flags & RFCFDG) { 258 struct filedesc *fdtmp; 259 fdtmp = fdinit(td); /* XXXKSE */ 260 fdfree(td); /* XXXKSE */ 261 p1->p_fd = fdtmp; 262 } 263 264 /* 265 * Unshare file descriptors (from parent.) 266 */ 267 if (flags & RFFDG) { 268 FILEDESC_LOCK(p1->p_fd); 269 if (p1->p_fd->fd_refcnt > 1) { 270 struct filedesc *newfd; 271 272 newfd = fdcopy(td); 273 FILEDESC_UNLOCK(p1->p_fd); 274 fdfree(td); 275 p1->p_fd = newfd; 276 } else 277 FILEDESC_UNLOCK(p1->p_fd); 278 } 279 *procp = NULL; 280 return (0); 281 } 282 283 if (p1->p_flag & P_KSES) { 284 /* 285 * Idle the other threads for a second. 286 * Since the user space is copied, it must remain stable. 287 * In addition, all threads (from the user perspective) 288 * need to either be suspended or in the kernel, 289 * where they will try restart in the parent and will 290 * be aborted in the child. 291 */ 292 PROC_LOCK(p1); 293 if (thread_single(SINGLE_NO_EXIT)) { 294 /* Abort.. someone else is single threading before us */ 295 PROC_UNLOCK(p1); 296 return (ERESTART); 297 } 298 PROC_UNLOCK(p1); 299 /* 300 * All other activity in this process 301 * is now suspended at the user boundary, 302 * (or other safe places if we think of any). 303 */ 304 } 305 306 /* Allocate new proc. */ 307 newproc = uma_zalloc(proc_zone, M_WAITOK); 308 309 /* 310 * Although process entries are dynamically created, we still keep 311 * a global limit on the maximum number we will create. Don't allow 312 * a nonprivileged user to use the last ten processes; don't let root 313 * exceed the limit. The variable nprocs is the current number of 314 * processes, maxproc is the limit. 315 */ 316 sx_xlock(&allproc_lock); 317 uid = td->td_ucred->cr_ruid; 318 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) { 319 error = EAGAIN; 320 goto fail; 321 } 322 323 /* 324 * Increment the count of procs running with this uid. Don't allow 325 * a nonprivileged user to exceed their current limit. 326 */ 327 PROC_LOCK(p1); 328 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 329 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0); 330 PROC_UNLOCK(p1); 331 if (!ok) { 332 error = EAGAIN; 333 goto fail; 334 } 335 336 /* 337 * Increment the nprocs resource before blocking can occur. There 338 * are hard-limits as to the number of processes that can run. 339 */ 340 nprocs++; 341 342 /* 343 * Find an unused process ID. We remember a range of unused IDs 344 * ready to use (from lastpid+1 through pidchecked-1). 345 * 346 * If RFHIGHPID is set (used during system boot), do not allocate 347 * low-numbered pids. 348 */ 349 trypid = lastpid + 1; 350 if (flags & RFHIGHPID) { 351 if (trypid < 10) { 352 trypid = 10; 353 } 354 } else { 355 if (randompid) 356 trypid += arc4random() % randompid; 357 } 358 retry: 359 /* 360 * If the process ID prototype has wrapped around, 361 * restart somewhat above 0, as the low-numbered procs 362 * tend to include daemons that don't exit. 363 */ 364 if (trypid >= PID_MAX) { 365 trypid = trypid % PID_MAX; 366 if (trypid < 100) 367 trypid += 100; 368 pidchecked = 0; 369 } 370 if (trypid >= pidchecked) { 371 int doingzomb = 0; 372 373 pidchecked = PID_MAX; 374 /* 375 * Scan the active and zombie procs to check whether this pid 376 * is in use. Remember the lowest pid that's greater 377 * than trypid, so we can avoid checking for a while. 378 */ 379 p2 = LIST_FIRST(&allproc); 380 again: 381 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) { 382 PROC_LOCK(p2); 383 while (p2->p_pid == trypid || 384 p2->p_pgrp->pg_id == trypid || 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->pg_id > trypid && 395 pidchecked > p2->p_pgrp->pg_id) 396 pidchecked = p2->p_pgrp->pg_id; 397 if (p2->p_session->s_sid > trypid && 398 pidchecked > p2->p_session->s_sid) 399 pidchecked = p2->p_session->s_sid; 400 PROC_UNLOCK(p2); 401 } 402 if (!doingzomb) { 403 doingzomb = 1; 404 p2 = LIST_FIRST(&zombproc); 405 goto again; 406 } 407 } 408 409 /* 410 * RFHIGHPID does not mess with the lastpid counter during boot. 411 */ 412 if (flags & RFHIGHPID) 413 pidchecked = 0; 414 else 415 lastpid = trypid; 416 417 p2 = newproc; 418 p2->p_state = PRS_NEW; /* protect against others */ 419 p2->p_pid = trypid; 420 LIST_INSERT_HEAD(&allproc, p2, p_list); 421 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 422 sx_xunlock(&allproc_lock); 423 424 /* 425 * Malloc things while we don't hold any locks. 426 */ 427 if (flags & RFSIGSHARE) { 428 MALLOC(newsigacts, struct sigacts *, 429 sizeof(struct sigacts), M_SUBPROC, M_WAITOK); 430 newprocsig = NULL; 431 } else { 432 newsigacts = NULL; 433 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig), 434 M_SUBPROC, M_WAITOK); 435 } 436 437 /* 438 * Copy filedesc. 439 * XXX: This is busted. fd*() need to not take proc 440 * arguments or something. 441 */ 442 if (flags & RFCFDG) 443 fd = fdinit(td); 444 else if (flags & RFFDG) { 445 FILEDESC_LOCK(p1->p_fd); 446 fd = fdcopy(td); 447 FILEDESC_UNLOCK(p1->p_fd); 448 } else 449 fd = fdshare(p1); 450 451 /* 452 * Make a proc table entry for the new process. 453 * Start by zeroing the section of proc that is zero-initialized, 454 * then copy the section that is copied directly from the parent. 455 */ 456 td2 = FIRST_THREAD_IN_PROC(p2); 457 kg2 = FIRST_KSEGRP_IN_PROC(p2); 458 ke2 = FIRST_KSE_IN_KSEGRP(kg2); 459 460 /* Allocate and switch to an alternate kstack if specified */ 461 if (pages != 0) 462 pmap_new_altkstack(td2, pages); 463 464 #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) 465 466 bzero(&p2->p_startzero, 467 (unsigned) RANGEOF(struct proc, p_startzero, p_endzero)); 468 bzero(&ke2->ke_startzero, 469 (unsigned) RANGEOF(struct kse, ke_startzero, ke_endzero)); 470 bzero(&td2->td_startzero, 471 (unsigned) RANGEOF(struct thread, td_startzero, td_endzero)); 472 bzero(&kg2->kg_startzero, 473 (unsigned) RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); 474 475 mtx_init(&p2->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK); 476 PROC_LOCK(p2); 477 PROC_LOCK(p1); 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 /* Set up the thread as an active thread (as if runnable). */ 488 ke2->ke_state = KES_THREAD; 489 ke2->ke_thread = td2; 490 td2->td_kse = ke2; 491 td2->td_flags &= ~TDF_UNBOUND; /* For the rest of this syscall. */ 492 493 /* 494 * Duplicate sub-structures as needed. 495 * Increase reference counts on shared objects. 496 * The p_stats and p_sigacts substructs are set in vm_forkproc. 497 */ 498 p2->p_flag = 0; 499 mtx_lock_spin(&sched_lock); 500 p2->p_sflag = PS_INMEM; 501 if (p1->p_sflag & PS_PROFIL) 502 startprofclock(p2); 503 /* 504 * Allow the scheduler to adjust the priority of the child and 505 * parent while we hold the sched_lock. 506 */ 507 sched_fork(td->td_ksegrp, kg2); 508 509 mtx_unlock_spin(&sched_lock); 510 p2->p_ucred = crhold(td->td_ucred); 511 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */ 512 513 pargs_hold(p2->p_args); 514 515 if (flags & RFSIGSHARE) { 516 p2->p_procsig = p1->p_procsig; 517 p2->p_procsig->ps_refcnt++; 518 if (p1->p_sigacts == &p1->p_uarea->u_sigacts) { 519 /* 520 * Set p_sigacts to the new shared structure. 521 * Note that this is updating p1->p_sigacts at the 522 * same time, since p_sigacts is just a pointer to 523 * the shared p_procsig->ps_sigacts. 524 */ 525 p2->p_sigacts = newsigacts; 526 newsigacts = NULL; 527 *p2->p_sigacts = p1->p_uarea->u_sigacts; 528 } 529 } else { 530 p2->p_procsig = newprocsig; 531 newprocsig = NULL; 532 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig)); 533 p2->p_procsig->ps_refcnt = 1; 534 p2->p_sigacts = NULL; /* finished in vm_forkproc() */ 535 } 536 if (flags & RFLINUXTHPN) 537 p2->p_sigparent = SIGUSR1; 538 else 539 p2->p_sigparent = SIGCHLD; 540 541 /* Bump references to the text vnode (for procfs) */ 542 p2->p_textvp = p1->p_textvp; 543 if (p2->p_textvp) 544 VREF(p2->p_textvp); 545 p2->p_fd = fd; 546 PROC_UNLOCK(p1); 547 PROC_UNLOCK(p2); 548 549 /* 550 * If p_limit is still copy-on-write, bump refcnt, 551 * otherwise get a copy that won't be modified. 552 * (If PL_SHAREMOD is clear, the structure is shared 553 * copy-on-write.) 554 */ 555 if (p1->p_limit->p_lflags & PL_SHAREMOD) 556 p2->p_limit = limcopy(p1->p_limit); 557 else { 558 p2->p_limit = p1->p_limit; 559 p2->p_limit->p_refcnt++; 560 } 561 562 /* 563 * Setup linkage for kernel based threading 564 */ 565 if((flags & RFTHREAD) != 0) { 566 mtx_lock(&ppeers_lock); 567 p2->p_peers = p1->p_peers; 568 p1->p_peers = p2; 569 p2->p_leader = p1->p_leader; 570 mtx_unlock(&ppeers_lock); 571 PROC_LOCK(p1->p_leader); 572 if ((p1->p_leader->p_flag & P_WEXIT) != 0) { 573 PROC_UNLOCK(p1->p_leader); 574 /* 575 * The task leader is exiting, so process p1 is 576 * going to be killed shortly. Since p1 obviously 577 * isn't dead yet, we know that the leader is either 578 * sending SIGKILL's to all the processes in this 579 * task or is sleeping waiting for all the peers to 580 * exit. We let p1 complete the fork, but we need 581 * to go ahead and kill the new process p2 since 582 * the task leader may not get a chance to send 583 * SIGKILL to it. We leave it on the list so that 584 * the task leader will wait for this new process 585 * to commit suicide. 586 */ 587 PROC_LOCK(p2); 588 psignal(p2, SIGKILL); 589 PROC_UNLOCK(p2); 590 } 591 } else { 592 p2->p_peers = NULL; 593 p2->p_leader = p2; 594 } 595 596 sx_xlock(&proctree_lock); 597 PGRP_LOCK(p1->p_pgrp); 598 PROC_LOCK(p2); 599 PROC_LOCK(p1); 600 601 /* 602 * Preserve some more flags in subprocess. PS_PROFIL has already 603 * been preserved. 604 */ 605 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK); 606 SESS_LOCK(p1->p_session); 607 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 608 p2->p_flag |= P_CONTROLT; 609 SESS_UNLOCK(p1->p_session); 610 if (flags & RFPPWAIT) 611 p2->p_flag |= P_PPWAIT; 612 613 LIST_INSERT_AFTER(p1, p2, p_pglist); 614 PGRP_UNLOCK(p1->p_pgrp); 615 LIST_INIT(&p2->p_children); 616 617 callout_init(&p2->p_itcallout, 0); 618 619 #ifdef KTRACE 620 /* 621 * Copy traceflag and tracefile if enabled. 622 */ 623 mtx_lock(&ktrace_mtx); 624 KASSERT(p2->p_tracep == NULL, ("new process has a ktrace vnode")); 625 if (p1->p_traceflag & KTRFAC_INHERIT) { 626 p2->p_traceflag = p1->p_traceflag; 627 if ((p2->p_tracep = p1->p_tracep) != NULL) 628 VREF(p2->p_tracep); 629 } 630 mtx_unlock(&ktrace_mtx); 631 #endif 632 633 /* 634 * If PF_FORK is set, the child process inherits the 635 * procfs ioctl flags from its parent. 636 */ 637 if (p1->p_pfsflags & PF_FORK) { 638 p2->p_stops = p1->p_stops; 639 p2->p_pfsflags = p1->p_pfsflags; 640 } 641 642 /* 643 * This begins the section where we must prevent the parent 644 * from being swapped. 645 */ 646 _PHOLD(p1); 647 PROC_UNLOCK(p1); 648 649 /* 650 * Attach the new process to its parent. 651 * 652 * If RFNOWAIT is set, the newly created process becomes a child 653 * of init. This effectively disassociates the child from the 654 * parent. 655 */ 656 if (flags & RFNOWAIT) 657 pptr = initproc; 658 else 659 pptr = p1; 660 p2->p_pptr = pptr; 661 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling); 662 PROC_UNLOCK(p2); 663 sx_xunlock(&proctree_lock); 664 665 KASSERT(newprocsig == NULL, ("unused newprocsig")); 666 if (newsigacts != NULL) 667 FREE(newsigacts, M_SUBPROC); 668 /* 669 * Finish creating the child process. It will return via a different 670 * execution path later. (ie: directly into user mode) 671 */ 672 vm_forkproc(td, p2, td2, flags); 673 674 if (flags == (RFFDG | RFPROC)) { 675 cnt.v_forks++; 676 cnt.v_forkpages += p2->p_vmspace->vm_dsize + 677 p2->p_vmspace->vm_ssize; 678 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) { 679 cnt.v_vforks++; 680 cnt.v_vforkpages += p2->p_vmspace->vm_dsize + 681 p2->p_vmspace->vm_ssize; 682 } else if (p1 == &proc0) { 683 cnt.v_kthreads++; 684 cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + 685 p2->p_vmspace->vm_ssize; 686 } else { 687 cnt.v_rforks++; 688 cnt.v_rforkpages += p2->p_vmspace->vm_dsize + 689 p2->p_vmspace->vm_ssize; 690 } 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 sx_slock(&fork_list_lock); 698 TAILQ_FOREACH(ep, &fork_list, next) { 699 (*ep->function)(p1, p2, flags); 700 } 701 sx_sunlock(&fork_list_lock); 702 703 /* 704 * If RFSTOPPED not requested, make child runnable and add to 705 * run queue. 706 */ 707 microtime(&(p2->p_stats->p_start)); 708 p2->p_acflag = AFORK; 709 if ((flags & RFSTOPPED) == 0) { 710 mtx_lock_spin(&sched_lock); 711 p2->p_state = PRS_NORMAL; 712 TD_SET_CAN_RUN(td2); 713 setrunqueue(td2); 714 mtx_unlock_spin(&sched_lock); 715 } 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(&p1->p_klist, NOTE_FORK | p2->p_pid); 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_KSES) { 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_xunlock(&allproc_lock); 755 uma_zfree(proc_zone, newproc); 756 if (p1->p_flag & P_KSES) { 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 * The next two functionms are general routines to handle adding/deleting 767 * items on the fork callout list. 768 * 769 * at_fork(): 770 * Take the arguments given and put them onto the fork callout list, 771 * However first make sure that it's not already there. 772 * Returns 0 on success or a standard error number. 773 */ 774 775 int 776 at_fork(function) 777 forklist_fn function; 778 { 779 struct forklist *ep; 780 781 #ifdef INVARIANTS 782 /* let the programmer know if he's been stupid */ 783 if (rm_at_fork(function)) 784 printf("WARNING: fork callout entry (%p) already present\n", 785 function); 786 #endif 787 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT); 788 if (ep == NULL) 789 return (ENOMEM); 790 ep->function = function; 791 sx_xlock(&fork_list_lock); 792 TAILQ_INSERT_TAIL(&fork_list, ep, next); 793 sx_xunlock(&fork_list_lock); 794 return (0); 795 } 796 797 /* 798 * Scan the exit callout list for the given item and remove it.. 799 * Returns the number of items removed (0 or 1) 800 */ 801 802 int 803 rm_at_fork(function) 804 forklist_fn function; 805 { 806 struct forklist *ep; 807 808 sx_xlock(&fork_list_lock); 809 TAILQ_FOREACH(ep, &fork_list, next) { 810 if (ep->function == function) { 811 TAILQ_REMOVE(&fork_list, ep, next); 812 sx_xunlock(&fork_list_lock); 813 free(ep, M_ATFORK); 814 return(1); 815 } 816 } 817 sx_xunlock(&fork_list_lock); 818 return (0); 819 } 820 821 /* 822 * Handle the return of a child process from fork1(). This function 823 * is called from the MD fork_trampoline() entry point. 824 */ 825 void 826 fork_exit(callout, arg, frame) 827 void (*callout)(void *, struct trapframe *); 828 void *arg; 829 struct trapframe *frame; 830 { 831 struct thread *td = curthread; 832 struct proc *p = td->td_proc; 833 834 td->td_kse->ke_oncpu = PCPU_GET(cpuid); 835 p->p_state = PRS_NORMAL; 836 /* 837 * Finish setting up thread glue. We need to initialize 838 * the thread into a td_critnest=1 state. Some platforms 839 * may have already partially or fully initialized td_critnest 840 * and/or td_md.md_savecrit (when applciable). 841 * 842 * see <arch>/<arch>/critical.c 843 */ 844 sched_lock.mtx_lock = (uintptr_t)td; 845 sched_lock.mtx_recurse = 0; 846 cpu_critical_fork_exit(); 847 CTR3(KTR_PROC, "fork_exit: new thread %p (pid %d, %s)", td, p->p_pid, 848 p->p_comm); 849 if (PCPU_GET(switchtime.sec) == 0) 850 binuptime(PCPU_PTR(switchtime)); 851 PCPU_SET(switchticks, ticks); 852 mtx_unlock_spin(&sched_lock); 853 854 /* 855 * cpu_set_fork_handler intercepts this function call to 856 * have this call a non-return function to stay in kernel mode. 857 * initproc has its own fork handler, but it does return. 858 */ 859 KASSERT(callout != NULL, ("NULL callout in fork_exit")); 860 callout(arg, frame); 861 862 /* 863 * Check if a kernel thread misbehaved and returned from its main 864 * function. 865 */ 866 PROC_LOCK(p); 867 if (p->p_flag & P_KTHREAD) { 868 PROC_UNLOCK(p); 869 mtx_lock(&Giant); 870 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", 871 p->p_comm, p->p_pid); 872 kthread_exit(0); 873 } 874 PROC_UNLOCK(p); 875 #ifdef DIAGNOSTIC 876 cred_free_thread(td); 877 #endif 878 mtx_assert(&Giant, MA_NOTOWNED); 879 } 880 881 /* 882 * Simplified back end of syscall(), used when returning from fork() 883 * directly into user mode. Giant is not held on entry, and must not 884 * be held on return. This function is passed in to fork_exit() as the 885 * first parameter and is called when returning to a new userland process. 886 */ 887 void 888 fork_return(td, frame) 889 struct thread *td; 890 struct trapframe *frame; 891 { 892 893 userret(td, frame, 0); 894 #ifdef KTRACE 895 if (KTRPOINT(td, KTR_SYSRET)) 896 ktrsysret(SYS_fork, 0, 0); 897 #endif 898 mtx_assert(&Giant, MA_NOTOWNED); 899 } 900